diff options
Diffstat (limited to 'src')
255 files changed, 42800 insertions, 10243 deletions
diff --git a/src/CMakeLists.txt b/src/CMakeLists.txt deleted file mode 100644 index 54fdd8ab..00000000 --- a/src/CMakeLists.txt +++ /dev/null @@ -1,4 +0,0 @@ -add_subdirectory(lib) -add_subdirectory(ipcpd) -add_subdirectory(irmd) -add_subdirectory(tools) diff --git a/src/ipcpd/CMakeLists.txt b/src/ipcpd/CMakeLists.txt index 54294f11..609da54a 100644 --- a/src/ipcpd/CMakeLists.txt +++ b/src/ipcpd/CMakeLists.txt @@ -1,62 +1,30 @@ -set(CONNMGR_RCV_TIMEOUT 1000 CACHE STRING - "Timeout for the connection manager to wait for OCEP info (ms).") -set(IPCP_DEBUG_LOCAL FALSE CACHE BOOL - "Use PID as address for local debugging") -set(IPCP_QOS_CUBE_BE_PRIO 50 CACHE STRING - "Priority for best effort QoS cube (0-99)") -set(IPCP_QOS_CUBE_VIDEO_PRIO 90 CACHE STRING - "Priority for video QoS cube (0-99)") -set(IPCP_QOS_CUBE_VOICE_PRIO 99 CACHE STRING - "Priority for voice QoS cube (0-99)") -set(IPCP_MIN_THREADS 4 CACHE STRING - "Minimum number of worker threads in the IPCP") -set(IPCP_ADD_THREADS 4 CACHE STRING - "Number of extra threads to start when an IPCP faces thread starvation") -set(IPCP_SCHED_THR_MUL 2 CACHE STRING - "Number of scheduler threads per QoS cube") -set(DISABLE_CORE_LOCK TRUE CACHE BOOL - "Disable locking performance threads to a core") -set(IPCP_CONN_WAIT_DIR TRUE CACHE BOOL - "Check the running state of the directory when adding a dt connection") -set(DHT_ENROLL_SLACK 50 CACHE STRING - "DHT enrollment waiting time (0-999, ms)") -if (CMAKE_SYSTEM_NAME STREQUAL "Linux") - set(IPCP_LINUX_TIMERSLACK_NS 1000 CACHE STRING - "Slack value for high resolution timers on Linux systems.") -endif () - -if ((IPCP_QOS_CUBE_BE_PRIO LESS 0) OR (IPCP_QOS_CUBE_BE_PRIO GREATER 99)) - message(FATAL_ERROR "Invalid priority for best effort QoS cube") -endif () - -if ((IPCP_QOS_CUBE_VIDEO_PRIO LESS 0) OR (IPCP_QOS_CUBE_VIDEO_PRIO GREATER 99)) - message(FATAL_ERROR "Invalid priority for video QoS cube") -endif () - -if ((IPCP_QOS_CUBE_VOICE_PRIO LESS 0) OR (IPCP_QOS_CUBE_VOICE_PRIO GREATER 99)) - message(FATAL_ERROR "Invalid priority for voice QoS cube") -endif () - -if ((DHT_ENROLL_SLACK LESS 0) OR (DHT_ENROLL_SLACK GREATER 999)) - message(FATAL_ERROR "Invalid DHT slack value") -endif () - +# IPCP (IPC Process) daemons build configuration +# Configuration options and validation are in cmake/config/ipcp/*.cmake +# Common sources shared by all IPCPs (absolute paths for subdirectories) set(IPCP_SOURCES - # Add source files here ${CMAKE_CURRENT_SOURCE_DIR}/ipcp.c ${CMAKE_CURRENT_SOURCE_DIR}/shim-data.c - ) +) -set (COMMON_SOURCES - ${CMAKE_CURRENT_SOURCE_DIR}/common/enroll.c - ) +set(COMMON_SOURCES + ${CMAKE_CURRENT_SOURCE_DIR}/common/enroll.c +) -add_subdirectory(local) -add_subdirectory(eth) -add_subdirectory(udp) -add_subdirectory(unicast) -add_subdirectory(broadcast) +set(IPCP_INCLUDE_DIRS + ${CMAKE_CURRENT_SOURCE_DIR} + ${CMAKE_CURRENT_BINARY_DIR} + ${CMAKE_SOURCE_DIR}/include + ${CMAKE_BINARY_DIR}/include +) configure_file("${CMAKE_CURRENT_SOURCE_DIR}/config.h.in" "${CMAKE_CURRENT_BINARY_DIR}/config.h" @ONLY) + +add_subdirectory(local) +add_subdirectory(broadcast) +add_subdirectory(unicast) +if(HAVE_ETH) + add_subdirectory(eth) +endif() +add_subdirectory(udp) diff --git a/src/ipcpd/broadcast/CMakeLists.txt b/src/ipcpd/broadcast/CMakeLists.txt index d85f335e..6749f660 100644 --- a/src/ipcpd/broadcast/CMakeLists.txt +++ b/src/ipcpd/broadcast/CMakeLists.txt @@ -1,35 +1,20 @@ -get_filename_component(CURRENT_SOURCE_PARENT_DIR - ${CMAKE_CURRENT_SOURCE_DIR} DIRECTORY) -get_filename_component(CURRENT_BINARY_PARENT_DIR - ${CMAKE_CURRENT_BINARY_DIR} DIRECTORY) +# Broadcast IPCP build configuration -include_directories(${CMAKE_CURRENT_SOURCE_DIR}) -include_directories(${CMAKE_CURRENT_BINARY_DIR}) - -include_directories(${CURRENT_SOURCE_PARENT_DIR}) -include_directories(${CURRENT_BINARY_PARENT_DIR}) - -include_directories(${CMAKE_SOURCE_DIR}/include) -include_directories(${CMAKE_BINARY_DIR}/include) - -set(IPCP_BROADCAST_TARGET ipcpd-broadcast CACHE INTERNAL "") -set(IPCP_BROADCAST_MPL 60 CACHE STRING - "Default maximum packet lifetime for the broadcast IPCP, in seconds") - -set(SOURCE_FILES - # Add source files here +set(BROADCAST_SOURCES connmgr.c dt.c main.c - ) +) + +add_executable(${IPCP_BROADCAST_TARGET} + ${BROADCAST_SOURCES} + ${IPCP_SOURCES} + ${COMMON_SOURCES} +) -add_executable(ipcpd-broadcast ${SOURCE_FILES} ${IPCP_SOURCES} ${COMMON_SOURCES} - ${LAYER_CONFIG_PROTO_SRCS}) -target_link_libraries(ipcpd-broadcast LINK_PUBLIC ouroboros-dev) +target_include_directories(${IPCP_BROADCAST_TARGET} PRIVATE ${IPCP_INCLUDE_DIRS}) +target_link_libraries(${IPCP_BROADCAST_TARGET} PRIVATE ouroboros-dev) -include(AddCompileFlags) -if (CMAKE_BUILD_TYPE MATCHES "Debug*") - add_compile_flags(ipcpd-broadcast -DCONFIG_OUROBOROS_DEBUG) -endif () +ouroboros_target_debug_definitions(${IPCP_BROADCAST_TARGET}) -install(TARGETS ipcpd-broadcast RUNTIME DESTINATION ${CMAKE_INSTALL_SBINDIR}) +install(TARGETS ${IPCP_BROADCAST_TARGET} RUNTIME DESTINATION ${CMAKE_INSTALL_SBINDIR}) diff --git a/src/ipcpd/broadcast/connmgr.c b/src/ipcpd/broadcast/connmgr.c index f297175d..a4d20ee7 100644 --- a/src/ipcpd/broadcast/connmgr.c +++ b/src/ipcpd/broadcast/connmgr.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Handles connections between components * diff --git a/src/ipcpd/broadcast/dt.c b/src/ipcpd/broadcast/dt.c index 938c9085..95483e33 100644 --- a/src/ipcpd/broadcast/dt.c +++ b/src/ipcpd/broadcast/dt.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Forward loop for broadcast * @@ -28,7 +28,7 @@ #include "config.h" -#define BROADCAST_MTU 1400 /* FIXME: avoid packet copy. */ +#define BROADCAST_MTU IPCP_BROADCAST_MTU /* FIXME: avoid packet copy. */ #define DT "dt" #define OUROBOROS_PREFIX DT @@ -58,14 +58,13 @@ struct nb { }; struct { - struct list_head nbs; - size_t nbs_len; - pthread_rwlock_t nbs_lock; + struct llist nbs; + pthread_rwlock_t lock; - fset_t * set; + fset_t * set; - pthread_t reader; - pthread_t listener; + pthread_t reader; + pthread_t listener; } fwd; static int dt_add_nb(int fd) @@ -73,12 +72,12 @@ static int dt_add_nb(int fd) struct list_head * p; struct nb * nb; - pthread_rwlock_wrlock(&fwd.nbs_lock); + pthread_rwlock_wrlock(&fwd.lock); - list_for_each(p, &fwd.nbs) { + llist_for_each(p, &fwd.nbs) { struct nb * el = list_entry(p, struct nb, next); if (el->fd == fd) { - pthread_rwlock_unlock(&fwd.nbs_lock); + pthread_rwlock_unlock(&fwd.lock); log_warn("Already know neighbor on fd %d.", fd); return 0; } @@ -86,18 +85,16 @@ static int dt_add_nb(int fd) nb = malloc(sizeof(*nb)); if (nb == NULL) { - pthread_rwlock_unlock(&fwd.nbs_lock); + pthread_rwlock_unlock(&fwd.lock); log_err("Failed to malloc neighbor struct."); return -ENOMEM; } nb->fd = fd; - list_add_tail(&nb->next, p); + llist_add_tail(&nb->next, &fwd.nbs); - ++fwd.nbs_len; - - pthread_rwlock_unlock(&fwd.nbs_lock); + pthread_rwlock_unlock(&fwd.lock); log_dbg("Neighbor %d added.", fd); @@ -109,21 +106,20 @@ static int dt_del_nb(int fd) struct list_head * p; struct list_head * h; - pthread_rwlock_wrlock(&fwd.nbs_lock); + pthread_rwlock_wrlock(&fwd.lock); - list_for_each_safe(p, h, &fwd.nbs) { + llist_for_each_safe(p, h, &fwd.nbs) { struct nb * nb = list_entry(p, struct nb, next); if (nb->fd == fd) { - list_del(&nb->next); - --fwd.nbs_len; - pthread_rwlock_unlock(&fwd.nbs_lock); + llist_del(&nb->next, &fwd.nbs); + pthread_rwlock_unlock(&fwd.lock); log_dbg("Neighbor %d deleted.", nb->fd); free(nb); return 0; } } - pthread_rwlock_unlock(&fwd.nbs_lock); + pthread_rwlock_unlock(&fwd.lock); log_err("Neighbor not found on fd %d.", fd); @@ -157,11 +153,11 @@ static void dt_packet(uint8_t * buf, { struct list_head * p; - pthread_rwlock_rdlock(&fwd.nbs_lock); + pthread_rwlock_rdlock(&fwd.lock); - pthread_cleanup_push(__cleanup_rwlock_unlock, &fwd.nbs_lock); + pthread_cleanup_push(__cleanup_rwlock_unlock, &fwd.lock); - list_for_each(p, &fwd.nbs) { + llist_for_each(p, &fwd.nbs) { struct nb * nb = list_entry(p, struct nb, next); if (nb->fd != in_fd) flow_write(nb->fd, buf, len); /* FIXME: avoid copy. */ @@ -252,12 +248,12 @@ int dt_init(void) strcpy(info.comp_name, DT); strcpy(info.comp_name, DT_COMP); - list_head_init(&fwd.nbs); + llist_init(&fwd.nbs); if (notifier_reg(handle_event, NULL)) goto fail_notifier_reg; - if (pthread_rwlock_init(&fwd.nbs_lock, NULL)) + if (pthread_rwlock_init(&fwd.lock, NULL)) goto fail_lock_init; fwd.set = fset_create(); @@ -273,8 +269,6 @@ int dt_init(void) if (connmgr_comp_init(COMPID_DT, &info)) goto fail_connmgr_comp_init; - fwd.nbs_len = 0; - return 0; fail_connmgr_comp_init: @@ -286,7 +280,7 @@ int dt_init(void) fail_pthread_create_reader: fset_destroy(fwd.set); fail_fset_create: - pthread_rwlock_destroy(&fwd.nbs_lock); + pthread_rwlock_destroy(&fwd.lock); fail_lock_init: notifier_unreg(handle_event); fail_notifier_reg: @@ -308,15 +302,15 @@ void dt_fini(void) fset_destroy(fwd.set); - pthread_rwlock_wrlock(&fwd.nbs_lock); + pthread_rwlock_wrlock(&fwd.lock); - list_for_each_safe(p, h, &fwd.nbs) { + llist_for_each_safe(p, h, &fwd.nbs) { struct nb * n = list_entry(p, struct nb, next); - list_del(&n->next); + llist_del(&n->next, &fwd.nbs); free(n); } - pthread_rwlock_unlock(&fwd.nbs_lock); + pthread_rwlock_unlock(&fwd.lock); - pthread_rwlock_destroy(&fwd.nbs_lock); + pthread_rwlock_destroy(&fwd.lock); } diff --git a/src/ipcpd/broadcast/dt.h b/src/ipcpd/broadcast/dt.h index 8d3b83f8..2472831e 100644 --- a/src/ipcpd/broadcast/dt.h +++ b/src/ipcpd/broadcast/dt.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Forward loop for broadcast * diff --git a/src/ipcpd/broadcast/main.c b/src/ipcpd/broadcast/main.c index f51fc629..d18cac82 100644 --- a/src/ipcpd/broadcast/main.c +++ b/src/ipcpd/broadcast/main.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Broadcast IPC Process * @@ -36,6 +36,7 @@ #include <ouroboros/ipcp-dev.h> #include <ouroboros/logs.h> #include <ouroboros/notifier.h> +#include <ouroboros/np1_flow.h> #include <ouroboros/random.h> #include <ouroboros/rib.h> #include <ouroboros/time.h> @@ -52,13 +53,8 @@ #include <assert.h> #include <inttypes.h> -struct ipcp ipcpi; - -static int initialize_components(const struct ipcp_config * conf) +static int initialize_components(void) { - strcpy(ipcpi.layer_name, conf->layer_info.name); - ipcpi.dir_hash_algo = (enum hash_algo) conf->layer_info.dir_hash_algo; - assert(ipcp_dir_hash_len() != 0); if (dt_init() < 0) { @@ -107,6 +103,7 @@ static void stop_components(void) static int broadcast_ipcp_enroll(const char * dst, struct layer_info * info) { + struct ipcp_config * conf; struct conn conn; uint8_t id[ENROLL_ID_LEN]; @@ -128,7 +125,10 @@ static int broadcast_ipcp_enroll(const char * dst, goto fail_enroll_boot; } - if (initialize_components(enroll_get_conf()) < 0) { + conf = enroll_get_conf(); + *info = conf->layer_info; + + if (initialize_components() < 0) { log_err_id(id, "Failed to initialize components."); goto fail_enroll_boot; } @@ -146,9 +146,6 @@ static int broadcast_ipcp_enroll(const char * dst, log_info_id(id, "Enrolled with %s.", dst); - info->dir_hash_algo = (enum pol_dir_hash) ipcpi.dir_hash_algo; - strcpy(info->name, ipcpi.layer_name); - return 0; fail_start_comp: @@ -159,16 +156,15 @@ static int broadcast_ipcp_enroll(const char * dst, return -1; } -static int broadcast_ipcp_bootstrap(const struct ipcp_config * conf) +static int broadcast_ipcp_bootstrap(struct ipcp_config * conf) { assert(conf); assert(conf->type == THIS_TYPE); - ((struct ipcp_config *) conf)->layer_info.dir_hash_algo = - DIR_HASH_SHA3_256; + assert(conf->layer_info.dir_hash_algo == DIR_HASH_SHA3_256); enroll_bootstrap(conf); - if (initialize_components(conf)) { + if (initialize_components()) { log_err("Failed to init IPCP components."); goto fail_init; } @@ -190,44 +186,63 @@ static int name_check(const uint8_t * dst) { uint8_t * buf; size_t len; - int ret; + int err; + char layer[LAYER_NAME_SIZE + 1]; - len = hash_len(ipcpi.dir_hash_algo); + len = ipcp_dir_hash_len(); buf = malloc(len); - if (buf == NULL) - return -ENOMEM; + if (buf == NULL) { + log_err("Failed to malloc buffer."); + err = -ENOMEM; + goto fail_buf; + } + + err = ipcp_get_layer_name(layer); + if (err < 0) { + log_err("Failed to get layer name."); + goto fail_layer; + } - str_hash(ipcpi.dir_hash_algo, buf, ipcpi.layer_name); + str_hash(HASH_SHA3_256, buf, layer); - ret = memcmp(buf, dst, len); + if (memcmp(buf, dst, len) < 0) { + log_err("Hash mismatch for layer %s.", layer); + err = -ENAME; + goto fail_layer; + } free(buf); - return ret; + return 0; + + fail_layer: + free(buf); + fail_buf: + return err; } static int broadcast_ipcp_join(int fd, - const uint8_t * dst, - qosspec_t qs) + const uint8_t * dst) { + int err; struct conn conn; time_t mpl = IPCP_BROADCAST_MPL; - buffer_t data = {NULL, 0}; - - (void) qs; + buffer_t data = BUF_INIT; memset(&conn, 0, sizeof(conn)); conn.flow_info.fd = fd; + conn.flow_info.qs = qos_np1; - if (name_check(dst) != 0) { + err = name_check(dst); + if (err < 0) { log_err("Failed to check name."); - return -1; + return err; } notifier_event(NOTIFY_DT_CONN_ADD, &conn); - ipcp_flow_alloc_reply(fd, 0, mpl, &data); + ipcp_flow_alloc_reply(fd, 0, mpl, IPCP_BROADCAST_MTU, &data); return 0; } @@ -292,12 +307,13 @@ int main(int argc, ipcp_sigwait(); if (ipcp_get_state() == IPCP_SHUTDOWN) { + ipcp_stop(); stop_components(); finalize_components(); + } else { + ipcp_stop(); } - ipcp_stop(); - enroll_fini(); connmgr_fini(); diff --git a/src/ipcpd/common/comp.h b/src/ipcpd/common/comp.h index f3790d9c..e1d025b6 100644 --- a/src/ipcpd/common/comp.h +++ b/src/ipcpd/common/comp.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Components for the unicast/broadcast IPC process * diff --git a/src/ipcpd/common/connmgr.c b/src/ipcpd/common/connmgr.c index 4b5fd420..48ad79ba 100644 --- a/src/ipcpd/common/connmgr.c +++ b/src/ipcpd/common/connmgr.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Handles connections between components * @@ -38,12 +38,6 @@ #include <stdlib.h> #include <string.h> -enum connmgr_state { - CONNMGR_NULL = 0, - CONNMGR_INIT, - CONNMGR_RUNNING -}; - struct conn_el { struct list_head next; struct conn conn; @@ -61,7 +55,6 @@ struct comp { struct { struct comp comps[COMPID_MAX]; - enum connmgr_state state; pthread_t acceptor; } connmgr; @@ -179,6 +172,8 @@ static void * flow_acceptor(void * o) continue; } + fccntl(fd, FLOWSRCVTIMEO, NULL); + err = add_comp_conn(id, fd, qs, &rcv_info); if (err < 0) { log_err("Failed to add new connection: %d.", err); @@ -226,8 +221,6 @@ static void handle_event(void * self, int connmgr_init(void) { - connmgr.state = CONNMGR_INIT; - if (notifier_reg(handle_event, NULL)) { log_err("Failed to register notifier."); return -1; @@ -240,13 +233,10 @@ void connmgr_fini(void) { int i; - notifier_unreg(handle_event); - - if (connmgr.state == CONNMGR_RUNNING) - pthread_join(connmgr.acceptor, NULL); - for (i = 0; i < COMPID_MAX; ++i) connmgr_comp_fini(i); + + notifier_unreg(handle_event); } int connmgr_start(void) @@ -256,15 +246,13 @@ int connmgr_start(void) return -1; } - connmgr.state = CONNMGR_RUNNING; - return 0; } void connmgr_stop(void) { - if (connmgr.state == CONNMGR_RUNNING) - pthread_cancel(connmgr.acceptor); + pthread_cancel(connmgr.acceptor); + pthread_join(connmgr.acceptor, NULL); } int connmgr_comp_init(enum comp_id id, @@ -487,9 +475,6 @@ int connmgr_alloc(enum comp_id id, switch (id) { case COMPID_DT: notifier_event(NOTIFY_DT_CONN_ADD, conn); -#if defined(BUILD_IPCP_UNICAST) && defined(IPCP_CONN_WAIT_DIR) - dir_wait_running(); -#endif break; case COMPID_MGMT: notifier_event(NOTIFY_MGMT_CONN_ADD, conn); @@ -513,7 +498,7 @@ int connmgr_dealloc(enum comp_id id, case COMPID_DT: notifier_event(NOTIFY_DT_CONN_DEL, conn); break; -#if defined(BUILD_IPCP_UNICAST) && defined(IPCP_CONN_WAIT_DIR) +#if defined(BUILD_IPCP_UNICAST) case COMPID_MGMT: notifier_event(NOTIFY_MGMT_CONN_DEL, conn); break; diff --git a/src/ipcpd/common/connmgr.h b/src/ipcpd/common/connmgr.h index 0710dbbf..f48ecd1b 100644 --- a/src/ipcpd/common/connmgr.h +++ b/src/ipcpd/common/connmgr.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Handles the different AP connections * diff --git a/src/ipcpd/common/enroll.c b/src/ipcpd/common/enroll.c index 5e35ce37..959cca07 100644 --- a/src/ipcpd/common/enroll.c +++ b/src/ipcpd/common/enroll.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Enrollment Task * @@ -43,20 +43,19 @@ #include <string.h> #include <pthread.h> +#ifdef __APPLE__ +#define llabs labs +#endif + #define ENROLL_COMP "Enrollment" #define ENROLL_PROTO "OEP" /* Ouroboros enrollment protocol */ #define ENROLL_WARN_TIME_OFFSET 20 #define ENROLL_BUF_LEN 1024 -enum enroll_state { - ENROLL_NULL = 0, - ENROLL_INIT, - ENROLL_RUNNING -}; struct { struct ipcp_config conf; - enum enroll_state state; + pthread_t listener; } enroll; @@ -107,8 +106,6 @@ static void * enroll_handle(void * o) log_info_id(req.id, "Handling incoming enrollment."); - /* TODO: authentication, timezone handling (UTC). */ - ack.result = -100; clock_gettime(CLOCK_REALTIME, &resp.t); @@ -227,12 +224,14 @@ int enroll_boot(struct conn * conn, return -1; } - if (resp.conf.type != ipcpi.type) { + if (resp.conf.type != ipcp_get_type()) { log_err_id(id, "Wrong type in enrollment response %d (%d).", - resp.conf.type, ipcpi.type); + resp.conf.type, ipcp_get_type()); return -1; } + enroll.conf = resp.conf; + clock_gettime(CLOCK_REALTIME, &rtt); delta_t = ts_diff_ms(&t0, &rtt); @@ -240,11 +239,9 @@ int enroll_boot(struct conn * conn, rtt.tv_sec = resp.t.tv_sec; rtt.tv_nsec = resp.t.tv_nsec; - if (labs(ts_diff_ms(&t0, &rtt)) - delta_t > ENROLL_WARN_TIME_OFFSET) + if (llabs(ts_diff_ms(&t0, &rtt)) - delta_t > ENROLL_WARN_TIME_OFFSET) log_warn_id(id, "Clock offset above threshold."); - enroll.conf = resp.conf; - return 0; } @@ -307,16 +304,11 @@ int enroll_init(void) return -1; } - enroll.state = ENROLL_INIT; - return 0; } void enroll_fini(void) { - if (enroll.state == ENROLL_RUNNING) - pthread_join(enroll.listener, NULL); - connmgr_comp_fini(COMPID_ENROLL); } @@ -325,13 +317,11 @@ int enroll_start(void) if (pthread_create(&enroll.listener, NULL, enroll_handle, NULL)) return -1; - enroll.state = ENROLL_RUNNING; - return 0; } void enroll_stop(void) { - if (enroll.state == ENROLL_RUNNING) - pthread_cancel(enroll.listener); + pthread_cancel(enroll.listener); + pthread_join(enroll.listener, NULL); } diff --git a/src/ipcpd/common/enroll.h b/src/ipcpd/common/enroll.h index f26c31a3..f56adfc8 100644 --- a/src/ipcpd/common/enroll.h +++ b/src/ipcpd/common/enroll.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Enrollment Task * diff --git a/src/ipcpd/config.h.in b/src/ipcpd/config.h.in index fe4f5fd2..7edec526 100644 --- a/src/ipcpd/config.h.in +++ b/src/ipcpd/config.h.in @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * IPC process configuration * @@ -23,17 +23,15 @@ #define PTHREAD_COND_CLOCK @PTHREAD_COND_CLOCK@ #define SYS_MAX_FLOWS @SYS_MAX_FLOWS@ -#define PROG_RES_FDS @PROG_RES_FDS@ -#define PROG_MAX_FLOWS @PROG_MAX_FLOWS@ +#define PROC_RES_FDS @PROC_RES_FDS@ +#define PROC_MAX_FLOWS @PROC_MAX_FLOWS@ #define SOCKET_TIMEOUT @SOCKET_TIMEOUT@ #define CONNECT_TIMEOUT @CONNECT_TIMEOUT@ -#define SHM_BUFFER_SIZE @SHM_BUFFER_SIZE@ -#define SHM_RDRB_BLOCK_SIZE @SHM_RDRB_BLOCK_SIZE@ +#define SSM_POOL_BLOCK_SIZE @SSM_POOL_BLOCK_SIZE@ #define DU_BUFF_HEADSPACE @DU_BUFF_HEADSPACE@ #define DU_BUFF_TAILSPACE @DU_BUFF_TAILSPACE@ -#cmakedefine SHM_RDRB_MULTI_BLOCK #define IPCP_MIN_THREADS @IPCP_MIN_THREADS@ #define IPCP_ADD_THREADS @IPCP_ADD_THREADS@ @@ -41,21 +39,26 @@ #define IPCP_LINUX_SLACK_NS @IPCP_LINUX_TIMERSLACK_NS@ -#cmakedefine IPCP_DEBUG_LOCAL - /* unicast IPCP */ #define QOS_PRIO_BE @IPCP_QOS_CUBE_BE_PRIO@ #define QOS_PRIO_VIDEO @IPCP_QOS_CUBE_VIDEO_PRIO@ #define QOS_PRIO_VOICE @IPCP_QOS_CUBE_VOICE_PRIO@ #define IPCP_SCHED_THR_MUL @IPCP_SCHED_THR_MUL@ #define PFT_SIZE @PFT_SIZE@ -#define DHT_ENROLL_SLACK @DHT_ENROLL_SLACK@ #define IPCP_UNICAST_MPL @IPCP_UNICAST_MPL@ +#define IPCP_UNICAST_MTU @IPCP_UNICAST_MTU@ #define CONNMGR_RCV_TIMEOUT @CONNMGR_RCV_TIMEOUT@ -#cmakedefine IPCP_CONN_WAIT_DIR #cmakedefine DISABLE_CORE_LOCK +#cmakedefine BUILD_CONTAINER #cmakedefine IPCP_FLOW_STATS +#cmakedefine IPCP_ETH_FLOW_STATS +#cmakedefine IPCP_DEBUG_LOCAL +#ifdef CONFIG_OUROBOROS_DEBUG +#cmakedefine DEBUG_PROTO_DHT +#cmakedefine DEBUG_PROTO_OEP +#cmakedefine DEBUG_PROTO_LS +#endif /* udp */ #cmakedefine HAVE_DDNS @@ -64,20 +67,26 @@ #define IPCP_UDP_RD_THR @IPCP_UDP_RD_THR@ #define IPCP_UDP_WR_THR @IPCP_UDP_WR_THR@ #define IPCP_UDP_MPL @IPCP_UDP_MPL@ +#define IPCP_UDP4_MTU @IPCP_UDP4_MTU@ +#define IPCP_UDP6_MTU @IPCP_UDP6_MTU@ /* eth */ #cmakedefine HAVE_NETMAP #cmakedefine HAVE_BPF #cmakedefine HAVE_RAW_SOCKETS #cmakedefine IPCP_ETH_QDISC_BYPASS -#define IPCP_ETH_RD_THR @IPCP_ETH_RD_THR@ -#define IPCP_ETH_WR_THR @IPCP_ETH_WR_THR@ -#define IPCP_ETH_LO_MTU @IPCP_ETH_LO_MTU@ -#define IPCP_ETH_MPL @IPCP_ETH_MPL@ +#define IPCP_ETH_RD_THR @IPCP_ETH_RD_THR@ +#define IPCP_ETH_WR_THR @IPCP_ETH_WR_THR@ +#define IPCP_ETH_LO_MTU @IPCP_ETH_LO_MTU@ +#define IPCP_ETH_MGMT_FRAME_SIZE @IPCP_ETH_MGMT_FRAME_SIZE@ +#define IPCP_ETH_MPL @IPCP_ETH_MPL@ +#define IPCP_ETH_SNDBUF @IPCP_ETH_SNDBUF@ +#define IPCP_ETH_RCVBUF @IPCP_ETH_RCVBUF@ /* local */ #define IPCP_LOCAL_MPL @IPCP_LOCAL_MPL@ +#define IPCP_LOCAL_MTU @IPCP_LOCAL_MTU@ /* broadcast */ -/* local */ #define IPCP_BROADCAST_MPL @IPCP_BROADCAST_MPL@ +#define IPCP_BROADCAST_MTU @IPCP_BROADCAST_MTU@ diff --git a/src/ipcpd/eth/CMakeLists.txt b/src/ipcpd/eth/CMakeLists.txt index d57e1848..5a36352d 100644 --- a/src/ipcpd/eth/CMakeLists.txt +++ b/src/ipcpd/eth/CMakeLists.txt @@ -1,134 +1,21 @@ -get_filename_component(CURRENT_SOURCE_PARENT_DIR - ${CMAKE_CURRENT_SOURCE_DIR} DIRECTORY) -get_filename_component(CURRENT_BINARY_PARENT_DIR - ${CMAKE_CURRENT_BINARY_DIR} DIRECTORY) - -include_directories(${CMAKE_CURRENT_SOURCE_DIR}) -include_directories(${CMAKE_CURRENT_BINARY_DIR}) - -include_directories(${CURRENT_SOURCE_PARENT_DIR}) -include_directories(${CURRENT_BINARY_PARENT_DIR}) - -include_directories(${CMAKE_SOURCE_DIR}/include) -include_directories(${CMAKE_BINARY_DIR}/include) - -find_path(NETMAP_C_INCLUDE_DIR - net/netmap_user.h - HINTS /usr/include /usr/local/include) - -mark_as_advanced(NETMAP_C_INCLUDE_DIR) - -# Check for raw sockets -if (CMAKE_SYSTEM_NAME STREQUAL "Linux") - set(DISABLE_RAW_SOCKETS FALSE CACHE BOOL - "Disable raw socket support for Ethernet IPCPs") - if (NOT DISABLE_RAW_SOCKETS) - message(STATUS "Raw socket support for Ethernet IPCPs enabled") - set(HAVE_RAW_SOCKETS TRUE PARENT_SCOPE) - set(HAVE_RAW_SOCKETS TRUE) - set(HAVE_ETH TRUE) - else () - message(STATUS "Raw socket support for Ethernet IPCPs disabled by user") - unset(HAVE_RAW_SOCKETS PARENT_SCOPE) - unset(HAVE_RAW_SOCKETS) - endif () -endif () - -# Check for BPF -if (NOT CMAKE_SYSTEM_NAME STREQUAL "Linux") - find_path(BPF_C_INCLUDE_DIR - net/bpf.h - HINTS /usr/include /usr/local/include) - - mark_as_advanced(BPF_C_INCLUDE_DIR) - - if (BPF_C_INCLUDE_DIR) - set(DISABLE_BPF FALSE CACHE BOOL - "Disable Berkeley Packet Filter support for Ethernet IPCPs") - if (NOT DISABLE_BPF) - message(STATUS "Berkeley Packet Filter support " - "for Ethernet IPCPs enabled") - set(HAVE_BPF TRUE PARENT_SCOPE) - set(HAVE_BPF TRUE) - set(HAVE_ETH TRUE) - else () - message(STATUS "Berkeley Packet Filter support " - "for Ethernet IPCPs disabled by user") - unset(HAVE_BPF PARENT_SCOPE) - unset(HAVE_BPF) - endif () - endif () -endif () - -# Check for netmap exclusively -if (NOT HAVE_RAW_SOCKETS AND NOT HAVE_BPF AND NETMAP_C_INCLUDE_DIR) - set(DISABLE_NETMAP FALSE CACHE BOOL - "Disable netmap support for ETH IPCPs") - if (NOT DISABLE_NETMAP) - message(STATUS "Netmap support for Ethernet IPCPs enabled") - set(HAVE_NETMAP TRUE PARENT_SCOPE) - set(HAVE_ETH TRUE) - else () - message(STATUS "Netmap support for Ethernet IPCPs disabled by user") - unset(HAVE_NETMAP PARENT_SCOPE) - endif () -endif () - -if (HAVE_ETH) - message(STATUS "Supported raw packet API found, building eth-llc and eth-dix") - - set(IPCP_ETH_RD_THR 1 CACHE STRING - "Number of reader threads in Ethernet IPCP") - set(IPCP_ETH_WR_THR 1 CACHE STRING - "Number of writer threads in Ethernet IPCP") - set(IPCP_ETH_QDISC_BYPASS false CACHE BOOL - "Bypass the Qdisc in the kernel when using raw sockets") - set(IPCP_ETH_LO_MTU 1500 CACHE STRING - "Restrict Ethernet MTU over loopback interfaces") - set(IPCP_ETH_MPL 5 CACHE STRING - "Default maximum packet lifetime for the Ethernet IPCPs, in seconds") - - set(ETH_LLC_SOURCES - # Add source files here - ${CMAKE_CURRENT_SOURCE_DIR}/llc.c - ) - - set(ETH_DIX_SOURCES - # Add source files here - ${CMAKE_CURRENT_SOURCE_DIR}/dix.c - ) - - set(IPCP_ETH_LLC_TARGET ipcpd-eth-llc CACHE INTERNAL "") - set(IPCP_ETH_DIX_TARGET ipcpd-eth-dix CACHE INTERNAL "") - - add_executable(ipcpd-eth-llc ${ETH_LLC_SOURCES} ${IPCP_SOURCES}) - add_executable(ipcpd-eth-dix ${ETH_DIX_SOURCES} ${IPCP_SOURCES}) - - if (HAVE_BPF AND NOT APPLE) - target_include_directories(ipcpd-eth-llc PUBLIC ${BPF_C_INCLUDE_DIR}) - target_include_directories(ipcpd-eth-dix PUBLIC ${BPF_C_INCLUDE_DIR}) - endif () - - if (HAVE_NETMAP AND NOT APPLE) - set_target_properties(ipcpd-eth-llc PROPERTIES - COMPILE_FLAGS "${CMAKE_C_FLAGS} -std=c99") - set_target_properties(ipcpd-eth-dix PROPERTIES - COMPILE_FLAGS "${CMAKE_C_FLAGS} -std=c99") - target_include_directories(ipcpd-eth-llc PUBLIC - ${NETMAP_C_INCLUDE_DIR}) - target_include_directories(ipcpd-eth-dix PUBLIC - ${NETMAP_C_INCLUDE_DIR}) - endif () - - target_link_libraries(ipcpd-eth-llc LINK_PUBLIC ouroboros-dev) - target_link_libraries(ipcpd-eth-dix LINK_PUBLIC ouroboros-dev) - - include(AddCompileFlags) - if (CMAKE_BUILD_TYPE MATCHES "Debug*") - add_compile_flags(ipcpd-eth-llc -DCONFIG_OUROBOROS_DEBUG) - add_compile_flags(ipcpd-eth-dix -DCONFIG_OUROBOROS_DEBUG) - endif () - - install(TARGETS ipcpd-eth-llc ipcpd-eth-dix RUNTIME DESTINATION - ${CMAKE_INSTALL_SBINDIR}) -endif () +# Ethernet IPCPs build configuration (LLC and DIX) +# HAVE_ETH detection is in cmake/dependencies.cmake + +add_executable(${IPCP_ETH_LLC_TARGET} llc.c ${IPCP_SOURCES}) +add_executable(${IPCP_ETH_DIX_TARGET} dix.c ${IPCP_SOURCES}) + +foreach(target ${IPCP_ETH_LLC_TARGET} ${IPCP_ETH_DIX_TARGET}) + target_include_directories(${target} PRIVATE ${IPCP_INCLUDE_DIRS}) + if(HAVE_BPF AND NOT APPLE) + target_include_directories(${target} PRIVATE ${BPF_C_INCLUDE_DIR}) + endif() + if(HAVE_NETMAP AND NOT APPLE) + target_compile_options(${target} PRIVATE -std=c99) + target_include_directories(${target} PRIVATE ${NETMAP_C_INCLUDE_DIR}) + endif() + target_link_libraries(${target} PRIVATE ouroboros-dev) + ouroboros_target_debug_definitions(${target}) +endforeach() + +install(TARGETS ${IPCP_ETH_LLC_TARGET} ${IPCP_ETH_DIX_TARGET} + RUNTIME DESTINATION ${CMAKE_INSTALL_SBINDIR}) diff --git a/src/ipcpd/eth/dix.c b/src/ipcpd/eth/dix.c index 37b9896d..cf8253bd 100644 --- a/src/ipcpd/eth/dix.c +++ b/src/ipcpd/eth/dix.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * IPC processes over Ethernet - DIX * diff --git a/src/ipcpd/eth/eth.c b/src/ipcpd/eth/eth.c index ea6e0f1c..7e038a03 100644 --- a/src/ipcpd/eth/eth.c +++ b/src/ipcpd/eth/eth.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * IPC processes over Ethernet * @@ -37,21 +37,33 @@ #include "config.h" +#include <ouroboros/atomics.h> #include <ouroboros/endian.h> #include <ouroboros/hash.h> #include <ouroboros/errno.h> #include <ouroboros/list.h> #include <ouroboros/utils.h> #include <ouroboros/bitmap.h> +#include <ouroboros/crc8.h> #include <ouroboros/dev.h> #include <ouroboros/ipcp-dev.h> #include <ouroboros/fqueue.h> #include <ouroboros/logs.h> +#include <ouroboros/np1_flow.h> #include <ouroboros/time.h> #include <ouroboros/fccntl.h> #include <ouroboros/pthread.h> +#include <ouroboros/rib.h> + +#ifndef IPCP_ETH_FLOW_STATS +#undef FETCH_ADD_RELAXED +#define FETCH_ADD_RELAXED(p, v) ((void) 0) +#undef FETCH_SUB_RELAXED +#define FETCH_SUB_RELAXED(p, v) ((void) 0) +#endif #include "ipcp.h" +#include "np1.h" #include "shim-data.h" #include <signal.h> @@ -88,31 +100,31 @@ #include <sys/mman.h> #if defined(HAVE_NETMAP) -#define NETMAP_WITH_LIBS -#include <net/netmap_user.h> + #define NETMAP_WITH_LIBS + #include <net/netmap_user.h> #elif defined(HAVE_BPF) -#define BPF_DEV_MAX 256 -#define BPF_BLEN sysconf(_SC_PAGESIZE) -#include <net/bpf.h> + #define BPF_DEV_MAX 256 + #define BPF_BLEN sysconf(_SC_PAGESIZE) + #include <net/bpf.h> #endif -#ifdef __linux__ +#define MAC_FMT "%02x:%02x:%02x:%02x:%02x:%02x" +#define MAC_VAL(a) \ + (uint8_t)(a)[0], (uint8_t)(a)[1], (uint8_t)(a)[2], \ + (uint8_t)(a)[3], (uint8_t)(a)[4], (uint8_t)(a)[5] + + #ifndef ETH_MAX_MTU /* In if_ether.h as of Linux 4.10. */ -#define ETH_MAX_MTU 0xFFFFU + #define ETH_MAX_MTU 0xFFFFU #endif /* ETH_MAX_MTU */ #ifdef BUILD_ETH_DIX -#define ETH_MTU eth_data.mtu -#define ETH_MTU_MAX ETH_MAX_MTU + #define ETH_MTU eth_data.mtu + #define ETH_MTU_MAX ETH_MAX_MTU #else -#define ETH_MTU eth_data.mtu -#define ETH_MTU_MAX 1500 + #define ETH_MTU eth_data.mtu + #define ETH_MTU_MAX 1500 #endif /* BUILD_ETH_DIX */ -#else /* __linux__ */ -#define ETH_MTU 1500 -#define ETH_MTU_MAX ETH_MTU -#endif /* __linux__ */ -#define MAC_SIZE 6 #define ETH_TYPE_LENGTH_SIZE sizeof(uint16_t) #define ETH_HEADER_SIZE (2 * MAC_SIZE + ETH_TYPE_LENGTH_SIZE) @@ -121,7 +133,8 @@ #define MGMT_EID 0 #define DIX_EID_SIZE sizeof(uint16_t) #define DIX_LENGTH_SIZE sizeof(uint16_t) -#define DIX_HEADER_SIZE (DIX_EID_SIZE + DIX_LENGTH_SIZE) +#define DIX_HCS_SIZE CRC8_HASH_LEN +#define DIX_HEADER_SIZE (DIX_EID_SIZE + DIX_LENGTH_SIZE + DIX_HCS_SIZE) #define ETH_HEADER_TOT_SIZE (ETH_HEADER_SIZE + DIX_HEADER_SIZE) #define MAX_EIDS (1 << (8 * DIX_EID_SIZE)) #define ETH_MAX_PACKET_SIZE (ETH_MTU - DIX_HEADER_SIZE) @@ -129,23 +142,26 @@ #elif defined(BUILD_ETH_LLC) #define THIS_TYPE IPCP_ETH_LLC #define MGMT_SAP 0x01 -#define LLC_HEADER_SIZE 3 +#define LLC_FIELDS_SIZE 3 +#define LLC_HCS_SIZE CRC8_HASH_LEN +#define LLC_HEADER_SIZE (LLC_FIELDS_SIZE + LLC_HCS_SIZE) #define ETH_HEADER_TOT_SIZE (ETH_HEADER_SIZE + LLC_HEADER_SIZE) #define MAX_SAPS 64 #define ETH_MAX_PACKET_SIZE (ETH_MTU - LLC_HEADER_SIZE) #define ETH_FRAME_SIZE (ETH_HEADER_SIZE + ETH_MTU_MAX) #endif -#define NAME_QUERY_TIMEO 2000 /* ms */ -#define MGMT_TIMEO 100 /* ms */ -#define MGMT_FRAME_SIZE 2048 +#define NAME_QUERY_TIMEO 1900 /* ms total budget */ +#define NAME_QUERY_RETRIES 3 /* retransmits, 4 attempts total */ +#define MGMT_TIMEO 100 /* ms */ +#define MGMT_FRAME_SIZE IPCP_ETH_MGMT_FRAME_SIZE +#define ETH_RIB_PATH "eth" #define FLOW_REQ 0 #define FLOW_REPLY 1 #define NAME_QUERY_REQ 2 #define NAME_QUERY_REPLY 3 - -struct ipcp ipcpi; +#define FLOW_IRM_UPDATE 4 struct mgmt_msg { #if defined(BUILD_ETH_DIX) @@ -165,13 +181,12 @@ struct mgmt_msg { uint32_t max_gap; uint32_t delay; uint32_t timeout; - uint16_t cypher_s; - uint8_t in_order; + int32_t response; + uint8_t service; #if defined (BUILD_ETH_DIX) uint8_t code; uint8_t availability; #endif - int8_t response; } __attribute__((packed)); struct eth_frame { @@ -187,6 +202,7 @@ struct eth_frame { uint8_t ssap; uint8_t cf; #endif + uint8_t hcs; uint8_t payload; } __attribute__((packed)); @@ -198,6 +214,17 @@ struct ef { int8_t r_sap; #endif uint8_t r_addr[MAC_SIZE]; +#ifdef IPCP_ETH_FLOW_STATS + struct { + time_t stamp; + size_t p_rcv; + size_t b_rcv; + size_t p_dlv_f; + size_t p_snd; + size_t b_snd; + size_t p_snd_f; + } stat; +#endif }; struct mgmt_frame { @@ -209,8 +236,9 @@ struct mgmt_frame { struct { struct shim_data * shim_data; -#ifdef __linux__ + int mtu; +#ifdef __linux__ int if_idx; #endif #if defined(HAVE_NETMAP) @@ -234,6 +262,22 @@ struct { struct ef * fd_to_ef; fset_t * np1_flows; pthread_rwlock_t flows_lock; +#ifdef IPCP_ETH_FLOW_STATS + struct { + size_t n_flows; + size_t n_rcv; + size_t n_snd; + size_t n_mgmt_rcv; + size_t n_mgmt_snd; + size_t n_bad_id; + size_t n_dlv_f; + size_t n_buf_f; + size_t n_rcv_f; + size_t n_snd_f; + size_t kern_rcv; + size_t kern_drp; + } stat; +#endif pthread_t packet_writer[IPCP_ETH_WR_THR]; pthread_t packet_reader[IPCP_ETH_RD_THR]; @@ -285,7 +329,14 @@ static int eth_data_init(void) eth_data.fd_to_ef[i].r_sap = -1; #endif memset(ð_data.fd_to_ef[i].r_addr, 0, MAC_SIZE); +#ifdef IPCP_ETH_FLOW_STATS + memset(ð_data.fd_to_ef[i].stat, 0, + sizeof(eth_data.fd_to_ef[i].stat)); +#endif } +#ifdef IPCP_ETH_FLOW_STATS + memset(ð_data.stat, 0, sizeof(eth_data.stat)); +#endif eth_data.shim_data = shim_data_create(); if (eth_data.shim_data == NULL) @@ -358,6 +409,227 @@ static void eth_data_fini(void) free(eth_data.fd_to_ef); } +#ifdef IPCP_ETH_FLOW_STATS +static int eth_rib_read(const char * path, + char * buf, + size_t len) +{ + struct ef * flow; + int fd; + char tmstr[RIB_TM_STRLEN]; + struct tm * tm; + time_t stamp; + char * entry; + + entry = strstr(path, RIB_SEPARATOR) + 1; + assert(entry); + + if (len < 2048) + return 0; + + buf[0] = '\0'; + + if (strcmp(entry, "summary") == 0) { + int n; +#if defined(HAVE_RAW_SOCKETS) + int rcvbuf = 0; + int sndbuf = 0; + int queued = 0; + socklen_t optlen = sizeof(rcvbuf); +# if defined(__linux__) + struct tpacket_stats tp_stats; + socklen_t tp_len = sizeof(tp_stats); +# endif + + getsockopt(eth_data.s_fd, SOL_SOCKET, + SO_RCVBUF, &rcvbuf, &optlen); + optlen = sizeof(sndbuf); + getsockopt(eth_data.s_fd, SOL_SOCKET, + SO_SNDBUF, &sndbuf, &optlen); + ioctl(eth_data.s_fd, FIONREAD, &queued); +# if defined(__linux__) + if (getsockopt(eth_data.s_fd, SOL_PACKET, + PACKET_STATISTICS, + &tp_stats, &tp_len) == 0) { + FETCH_ADD_RELAXED(ð_data.stat.kern_rcv, + tp_stats.tp_packets); + FETCH_ADD_RELAXED(ð_data.stat.kern_drp, + tp_stats.tp_drops); + } +# endif +#endif + n = sprintf(buf, + "Active flows: %20zu\n" + "Total frames received: %20zu\n" + "Total frames sent: %20zu\n" + "Management frames received: %20zu\n" + "Management frames sent: %20zu\n" + "Bad EID/SAP frames: %20zu\n" + "Delivery (N+1) failures: %20zu\n" + "Buffer alloc failures: %20zu\n" + "Frame read failures: %20zu\n" + "Frame send failures: %20zu\n", + LOAD_RELAXED(ð_data.stat.n_flows), + LOAD_RELAXED(ð_data.stat.n_rcv), + LOAD_RELAXED(ð_data.stat.n_snd), + LOAD_RELAXED(ð_data.stat.n_mgmt_rcv), + LOAD_RELAXED(ð_data.stat.n_mgmt_snd), + LOAD_RELAXED(ð_data.stat.n_bad_id), + LOAD_RELAXED(ð_data.stat.n_dlv_f), + LOAD_RELAXED(ð_data.stat.n_buf_f), + LOAD_RELAXED(ð_data.stat.n_rcv_f), + LOAD_RELAXED(ð_data.stat.n_snd_f)); +#if defined(HAVE_RAW_SOCKETS) + n += sprintf(buf + n, + "Socket rcvbuf (bytes): %20d\n" + "Socket sndbuf (bytes): %20d\n" + "Socket queued (bytes): %20d\n", + rcvbuf, sndbuf, queued); +# if defined(__linux__) + n += sprintf(buf + n, + "Kernel frames received: %20zu\n" + "Kernel frames dropped: %20zu\n", + LOAD_RELAXED(ð_data.stat.kern_rcv), + LOAD_RELAXED(ð_data.stat.kern_drp)); +# endif +#endif + return n; + } + + fd = atoi(entry); + + if (fd < 0 || fd >= SYS_MAX_FLOWS) + return -1; + + flow = ð_data.fd_to_ef[fd]; + + pthread_rwlock_rdlock(ð_data.flows_lock); + + stamp = flow->stat.stamp; + if (stamp == 0) { + pthread_rwlock_unlock(ð_data.flows_lock); + return 0; + } + + pthread_rwlock_unlock(ð_data.flows_lock); + + tm = gmtime(&stamp); + strftime(tmstr, sizeof(tmstr), RIB_TM_FORMAT, tm); + + sprintf(buf, + "Flow established at: %20s\n" + "Sent (packets): %20zu\n" + "Sent (bytes): %20zu\n" + "Send failed (packets): %20zu\n" + "Received (packets): %20zu\n" + "Received (bytes): %20zu\n" + "Delivery (N+1) failures: %20zu\n", + tmstr, + LOAD_RELAXED(&flow->stat.p_snd), + LOAD_RELAXED(&flow->stat.b_snd), + LOAD_RELAXED(&flow->stat.p_snd_f), + LOAD_RELAXED(&flow->stat.p_rcv), + LOAD_RELAXED(&flow->stat.b_rcv), + LOAD_RELAXED(&flow->stat.p_dlv_f)); + + return strlen(buf); +} + +static int eth_rib_readdir(char *** buf) +{ + char entry[RIB_PATH_LEN + 1]; + size_t i; + int idx = 0; + int n_entries; + + pthread_rwlock_rdlock(ð_data.flows_lock); + + n_entries = (int) LOAD_RELAXED(ð_data.stat.n_flows) + 1; + + *buf = malloc(sizeof(**buf) * n_entries); + if (*buf == NULL) + goto fail_entries; + + (*buf)[idx] = malloc(strlen("summary") + 1); + if ((*buf)[idx] == NULL) + goto fail_entry; + + strcpy((*buf)[idx++], "summary"); + + for (i = 0; i < SYS_MAX_FLOWS && idx < n_entries; ++i) { + if (eth_data.fd_to_ef[i].stat.stamp == 0) + continue; + + sprintf(entry, "%zu", i); + + (*buf)[idx] = malloc(strlen(entry) + 1); + if ((*buf)[idx] == NULL) + goto fail_entry; + + strcpy((*buf)[idx++], entry); + } + + pthread_rwlock_unlock(ð_data.flows_lock); + + return idx; + + fail_entry: + while (idx-- > 0) + free((*buf)[idx]); + free(*buf); + fail_entries: + pthread_rwlock_unlock(ð_data.flows_lock); + return -ENOMEM; +} + +static int eth_rib_getattr(const char * path, + struct rib_attr * attr) +{ + int fd; + char * entry; + struct ef * flow; + + entry = strstr(path, RIB_SEPARATOR) + 1; + assert(entry); + + if (strcmp(entry, "summary") == 0) { + attr->size = 2048; + attr->mtime = 0; + return 0; + } + + fd = atoi(entry); + + if (fd < 0 || fd >= SYS_MAX_FLOWS) { + attr->size = 0; + attr->mtime = 0; + return 0; + } + + flow = ð_data.fd_to_ef[fd]; + + pthread_rwlock_rdlock(ð_data.flows_lock); + + if (flow->stat.stamp != 0) { + attr->size = 2048; + attr->mtime = flow->stat.stamp; + } else { + attr->size = 0; + attr->mtime = 0; + } + + pthread_rwlock_unlock(ð_data.flows_lock); + + return 0; +} + +static struct rib_ops eth_r_ops = { + .read = eth_rib_read, + .readdir = eth_rib_readdir, + .getattr = eth_rib_getattr +}; +#endif /* IPCP_ETH_FLOW_STATS */ + #ifdef BUILD_ETH_LLC static uint8_t reverse_bits(uint8_t b) { @@ -410,12 +682,18 @@ static int eth_ipcp_send_frame(const uint8_t * dst_addr, e_frame->ethertype = eth_data.ethertype; e_frame->eid = htons(deid); e_frame->length = htons(len); + mem_hash(HASH_CRC8, &e_frame->hcs, + (uint8_t *) &e_frame->eid, + DIX_EID_SIZE + DIX_LENGTH_SIZE); frame_len = ETH_HEADER_TOT_SIZE + len; #elif defined(BUILD_ETH_LLC) e_frame->length = htons(LLC_HEADER_SIZE + len); e_frame->dsap = dsap; e_frame->ssap = ssap; e_frame->cf = cf; + mem_hash(HASH_CRC8, &e_frame->hcs, + (uint8_t *) &e_frame->dsap, + LLC_FIELDS_SIZE); frame_len = ETH_HEADER_TOT_SIZE + len; #endif @@ -441,10 +719,7 @@ static int eth_ipcp_send_frame(const uint8_t * dst_addr, } assert(FD_ISSET(eth_data.s_fd, &fds)); - if (sendto(eth_data.s_fd, - frame, - frame_len, - 0, + if (sendto(eth_data.s_fd, frame, frame_len, 0, (struct sockaddr *) ð_data.device, sizeof(eth_data.device)) <= 0) { log_dbg("Failed to send message: %s.", strerror(errno)); @@ -452,6 +727,8 @@ static int eth_ipcp_send_frame(const uint8_t * dst_addr, } #endif /* HAVE_NETMAP */ + FETCH_ADD_RELAXED(ð_data.stat.n_snd, 1); + return 0; } @@ -476,6 +753,8 @@ static int eth_ipcp_alloc(const uint8_t * dst_addr, if (buf == NULL) return -1; + memset(buf, 0, len + ETH_HEADER_TOT_SIZE + data->len); + msg = (struct mgmt_msg *) (buf + ETH_HEADER_TOT_SIZE); msg->code = FLOW_REQ; #if defined(BUILD_ETH_DIX) @@ -489,9 +768,8 @@ static int eth_ipcp_alloc(const uint8_t * dst_addr, msg->availability = qs.availability; msg->loss = hton32(qs.loss); msg->ber = hton32(qs.ber); - msg->in_order = qs.in_order; + msg->service = qs.service; msg->max_gap = hton32(qs.max_gap); - msg->cypher_s = hton16(qs.cypher_s); msg->timeout = hton32(qs.timeout); memcpy(msg + 1, hash, ipcp_dir_hash_len()); @@ -508,6 +786,9 @@ static int eth_ipcp_alloc(const uint8_t * dst_addr, buf, len + data->len); free(buf); + if (ret == 0) + FETCH_ADD_RELAXED(ð_data.stat.n_mgmt_snd, 1); + return ret; } @@ -529,6 +810,8 @@ static int eth_ipcp_alloc_resp(uint8_t * dst_addr, if (buf == NULL) return -1; + memset(buf, 0, sizeof(*msg) + ETH_HEADER_TOT_SIZE + data->len); + msg = (struct mgmt_msg *) (buf + ETH_HEADER_TOT_SIZE); msg->code = FLOW_REPLY; @@ -539,7 +822,7 @@ static int eth_ipcp_alloc_resp(uint8_t * dst_addr, msg->ssap = ssap; msg->dsap = dsap; #endif - msg->response = response; + msg->response = hton32(response); if (data->len > 0) memcpy(msg + 1, data->data, data->len); @@ -556,11 +839,65 @@ static int eth_ipcp_alloc_resp(uint8_t * dst_addr, return -1; } + FETCH_ADD_RELAXED(ð_data.stat.n_mgmt_snd, 1); + free(buf); return 0; } +static int eth_ipcp_flow_update(int fd, + const buffer_t * data) +{ + struct mgmt_msg * msg; + struct ef * flow; + uint8_t * buf; + uint8_t r_addr[MAC_SIZE]; + int ret; + + buf = malloc(sizeof(*msg) + ETH_HEADER_TOT_SIZE + data->len); + if (buf == NULL) + return -1; + + memset(buf, 0, sizeof(*msg) + ETH_HEADER_TOT_SIZE + data->len); + + msg = (struct mgmt_msg *) (buf + ETH_HEADER_TOT_SIZE); + + msg->code = FLOW_IRM_UPDATE; + + pthread_rwlock_rdlock(ð_data.flows_lock); + + flow = ð_data.fd_to_ef[fd]; +#if defined(BUILD_ETH_DIX) + msg->seid = htons((uint16_t) fd); + msg->deid = htons((uint16_t) flow->r_eid); +#elif defined(BUILD_ETH_LLC) + msg->ssap = flow->sap; + msg->dsap = (uint8_t) flow->r_sap; +#endif + memcpy(r_addr, flow->r_addr, MAC_SIZE); + + pthread_rwlock_unlock(ð_data.flows_lock); + + if (data->len > 0) + memcpy(msg + 1, data->data, data->len); + + ret = eth_ipcp_send_frame(r_addr, +#if defined(BUILD_ETH_DIX) + MGMT_EID, +#elif defined(BUILD_ETH_LLC) + reverse_bits(MGMT_SAP), + reverse_bits(MGMT_SAP), +#endif + buf, sizeof(*msg) + data->len); + free(buf); + + if (ret == 0) + FETCH_ADD_RELAXED(ð_data.stat.n_mgmt_snd, 1); + + return ret; +} + static int eth_ipcp_req(uint8_t * r_addr, #if defined(BUILD_ETH_DIX) uint16_t r_eid, @@ -573,7 +910,8 @@ static int eth_ipcp_req(uint8_t * r_addr, { int fd; - fd = ipcp_wait_flow_req_arr(dst, qs, IPCP_ETH_MPL, data); + fd = ipcp_wait_flow_req_arr(dst, qs, IPCP_ETH_MPL, + ETH_MAX_PACKET_SIZE, data); if (fd < 0) { log_err("Could not get new flow from IRMd."); return -1; @@ -620,7 +958,7 @@ static int eth_ipcp_alloc_reply(uint8_t * r_addr, fd = eth_data.ef_to_fd[dsap]; #endif if (fd < 0) { - pthread_rwlock_unlock(& eth_data.flows_lock); + pthread_rwlock_unlock(ð_data.flows_lock); log_err("No flow found with that SAP."); return -1; /* -EFLOWNOTFOUND */ } @@ -645,7 +983,8 @@ static int eth_ipcp_alloc_reply(uint8_t * r_addr, #elif defined(BUILD_ETH_LLC) log_dbg("Flow reply, fd %d, SSAP %d, DSAP %d.", fd, ssap, dsap); #endif - if ((ret = ipcp_flow_alloc_reply(fd, response, mpl, data)) < 0) { + if ((ret = ipcp_flow_alloc_reply(fd, response, mpl, + ETH_MAX_PACKET_SIZE, data)) < 0) { log_err("Failed to reply to flow allocation."); return -1; } @@ -667,6 +1006,8 @@ static int eth_ipcp_name_query_req(const uint8_t * hash, if (buf == NULL) return -1; + memset(buf, 0, len + ETH_HEADER_TOT_SIZE); + msg = (struct mgmt_msg *) (buf + ETH_HEADER_TOT_SIZE); msg->code = NAME_QUERY_REPLY; @@ -685,6 +1026,8 @@ static int eth_ipcp_name_query_req(const uint8_t * hash, return -1; } + FETCH_ADD_RELAXED(ð_data.stat.n_mgmt_snd, 1); + free(buf); } @@ -694,17 +1037,55 @@ static int eth_ipcp_name_query_req(const uint8_t * hash, static int eth_ipcp_name_query_reply(const uint8_t * hash, uint8_t * r_addr) { - uint64_t address = 0; + struct addr addr; - memcpy(&address, r_addr, MAC_SIZE); + memcpy(&addr.mac, r_addr, MAC_SIZE); - shim_data_dir_add_entry(eth_data.shim_data, hash, address); + shim_data_dir_add_entry(eth_data.shim_data, hash, addr); shim_data_dir_query_respond(eth_data.shim_data, hash); return 0; } +static int eth_ipcp_flow_update_arr(const uint8_t * buf, + size_t len) +{ + struct mgmt_msg * msg; + buffer_t data; + int fd; + int flow_id; + + msg = (struct mgmt_msg *) buf; + + data.data = (uint8_t *) buf + sizeof(*msg); + data.len = len - sizeof(*msg); + + pthread_rwlock_rdlock(ð_data.flows_lock); +#if defined(BUILD_ETH_DIX) + fd = ntohs(msg->deid); +#elif defined(BUILD_ETH_LLC) + fd = eth_data.ef_to_fd[msg->dsap]; +#endif + pthread_rwlock_unlock(ð_data.flows_lock); + + if (fd < 0 || fd >= SYS_MAX_FLOWS) { + log_err("Flow update for unknown endpoint."); + return -1; + } + + flow_id = np1_flow_id(fd); + if (flow_id < 0) + return -1; + + if (ipcp_flow_update_arr(flow_id, &data) < 0) { + log_err("Failed to relay flow update on fd %d.", fd); + return -1; + } + + return 0; +} + static int eth_ipcp_mgmt_frame(const uint8_t * buf, size_t len, uint8_t * r_addr) @@ -714,22 +1095,25 @@ static int eth_ipcp_mgmt_frame(const uint8_t * buf, qosspec_t qs; buffer_t data; + if (len < sizeof(*msg)) + return -1; + msg = (struct mgmt_msg *) buf; switch (msg->code) { case FLOW_REQ: msg_len = sizeof(*msg) + ipcp_dir_hash_len(); - assert(len >= msg_len); + if (len < msg_len) + return -1; qs.delay = ntoh32(msg->delay); qs.bandwidth = ntoh64(msg->bandwidth); qs.availability = msg->availability; qs.loss = ntoh32(msg->loss); qs.ber = ntoh32(msg->ber); - qs.in_order = msg->in_order; + qs.service = msg->service; qs.max_gap = ntoh32(msg->max_gap); - qs.cypher_s = ntoh16(msg->cypher_s); qs.timeout = ntoh32(msg->timeout); data.data = (uint8_t *) buf + msg_len; @@ -749,8 +1133,6 @@ static int eth_ipcp_mgmt_frame(const uint8_t * buf, } break; case FLOW_REPLY: - assert(len >= sizeof(*msg)); - data.data = (uint8_t *) buf + sizeof(*msg); data.len = len - sizeof(*msg); @@ -762,13 +1144,20 @@ static int eth_ipcp_mgmt_frame(const uint8_t * buf, msg->ssap, msg->dsap, #endif - msg->response, + ntoh32(msg->response), &data); break; + case FLOW_IRM_UPDATE: + eth_ipcp_flow_update_arr(buf, len); + break; case NAME_QUERY_REQ: + if (len < sizeof(*msg) + ipcp_dir_hash_len()) + return -1; eth_ipcp_name_query_req(buf + sizeof(*msg), r_addr); break; case NAME_QUERY_REPLY: + if (len < sizeof(*msg) + ipcp_dir_hash_len()) + return -1; eth_ipcp_name_query_reply(buf + sizeof(*msg), r_addr); break; default: @@ -837,10 +1226,16 @@ static void * eth_ipcp_packet_reader(void * o) #if defined(HAVE_NETMAP) struct nm_pkthdr hdr; #else - struct shm_du_buff * sdb; + struct ssm_pk_buff * spb; fd_set fds; int frame_len; #endif +#if defined(HAVE_RAW_SOCKETS) + struct sockaddr_ll src; + socklen_t slen; +#endif + size_t eth_len; + uint8_t hcs; struct eth_frame * e_frame; struct mgmt_frame * frame; @@ -869,33 +1264,67 @@ static void * eth_ipcp_packet_reader(void * o) if (select(eth_data.bpf + 1, &fds, NULL, NULL, NULL)) continue; assert(FD_ISSET(eth_data.bpf, &fds)); - if (ipcp_sdb_reserve(&sdb, BPF_LEN)) + if (ipcp_spb_reserve(&spb, BPF_LEN)) continue; - buf = shm_du_buff_head(sdb); + buf = ssm_pk_buff_head(spb); frame_len = read(eth_data.bpf, buf, BPF_BLEN); #elif defined(HAVE_RAW_SOCKETS) FD_SET(eth_data.s_fd, &fds); if (select(eth_data.s_fd + 1, &fds, NULL, NULL, NULL) < 0) continue; assert(FD_ISSET(eth_data.s_fd, &fds)); - if (ipcp_sdb_reserve(&sdb, ETH_MTU)) + if (ipcp_spb_reserve(&spb, ETH_MTU)) { + FETCH_ADD_RELAXED(ð_data.stat.n_buf_f, 1); continue; - buf = shm_du_buff_head_alloc(sdb, ETH_HEADER_TOT_SIZE); + } + buf = ssm_pk_buff_push(spb, ETH_HEADER_TOT_SIZE); if (buf == NULL) { log_dbg("Failed to allocate header."); - ipcp_sdb_release(sdb); + ipcp_spb_release(spb); + FETCH_ADD_RELAXED(ð_data.stat.n_buf_f, 1); continue; } - frame_len = recv(eth_data.s_fd, buf, - ETH_MTU + ETH_HEADER_TOT_SIZE, 0); + slen = sizeof(src); + /* MSG_DONTWAIT: RD_THR>1 race-loser bails with EAGAIN. */ + frame_len = recvfrom(eth_data.s_fd, buf, + ETH_MTU + ETH_HEADER_TOT_SIZE, + MSG_DONTWAIT, + (struct sockaddr *) &src, &slen); #endif - if (frame_len <= 0) { - log_dbg("Failed to receive frame."); - ipcp_sdb_release(sdb); + if (frame_len == 0) { + ipcp_spb_release(spb); + continue; /* Spurious */ + } + + if (frame_len < 0) { + ipcp_spb_release(spb); + + if (errno == EAGAIN || errno == EWOULDBLOCK) + continue; + + log_dbg("Failed to rcv frame: %s.", strerror(errno)); + FETCH_ADD_RELAXED(ð_data.stat.n_rcv_f, 1); continue; } #endif +#if defined(HAVE_NETMAP) + eth_len = hdr.len; +#elif defined(HAVE_BPF) + eth_len = ((struct bpf_hdr *) buf)->bh_caplen; +#else + eth_len = (size_t) frame_len; +#endif + /* Defense in depth: reject before parsing dereferences. */ + if (eth_len < ETH_HEADER_TOT_SIZE) + goto fail_frame; + +#if defined(HAVE_RAW_SOCKETS) + /* Drop our own egress. */ + if (src.sll_pkttype == PACKET_OUTGOING) + goto fail_frame; +#endif + #if defined(HAVE_BPF) && !defined(HAVE_NETMAP) e_frame = (struct eth_frame *) (buf + ((struct bpf_hdr *) buf)->bh_hdrlen); @@ -913,6 +1342,8 @@ static void * eth_ipcp_packet_reader(void * o) e_frame->dst_hwaddr, MAC_SIZE) && memcmp(br_addr, e_frame->dst_hwaddr, MAC_SIZE)) { + FETCH_ADD_RELAXED(ð_data.stat.n_bad_id, 1); + goto fail_frame; } #endif length = ntohs(e_frame->length); @@ -920,25 +1351,55 @@ static void * eth_ipcp_packet_reader(void * o) if (e_frame->ethertype != eth_data.ethertype) goto fail_frame; + if (length > ETH_MTU) + goto fail_frame; + deid = ntohs(e_frame->eid); - if (deid == MGMT_EID) { #elif defined (BUILD_ETH_LLC) if (length > 0x05FF) /* DIX */ goto fail_frame; + if (length < LLC_HEADER_SIZE || length > ETH_MTU) + goto fail_frame; + length -= LLC_HEADER_SIZE; dsap = reverse_bits(e_frame->dsap); ssap = reverse_bits(e_frame->ssap); +#endif + if (eth_len < ETH_HEADER_TOT_SIZE + (size_t) length) + goto fail_frame; + +#if defined(BUILD_ETH_DIX) + mem_hash(HASH_CRC8, &hcs, + (uint8_t *) &e_frame->eid, + DIX_EID_SIZE + DIX_LENGTH_SIZE); +#elif defined(BUILD_ETH_LLC) + mem_hash(HASH_CRC8, &hcs, + (uint8_t *) &e_frame->dsap, + LLC_FIELDS_SIZE); +#endif + if (hcs != e_frame->hcs) + goto fail_frame; + +#if defined(BUILD_ETH_DIX) + if (deid == MGMT_EID) { +#elif defined (BUILD_ETH_LLC) if (ssap == MGMT_SAP && dsap == MGMT_SAP) { #endif - ipcp_sdb_release(sdb); /* No need for the N+1 buffer. */ + ipcp_spb_release(spb); /* No need for the N+1 buffer. */ + + if (length > MGMT_FRAME_SIZE) { + log_warn("Management frame size %u exceeds %u.", + length, MGMT_FRAME_SIZE); + continue; + } frame = malloc(sizeof(*frame)); if (frame == NULL) { log_err("Failed to allocate frame."); - goto fail_frame; + continue; } memcpy(frame->buf, &e_frame->payload, length); @@ -949,6 +1410,8 @@ static void * eth_ipcp_packet_reader(void * o) list_add(&frame->next, ð_data.mgmt_frames); pthread_cond_signal(ð_data.mgmt_cond); pthread_mutex_unlock(ð_data.mgmt_lock); + FETCH_ADD_RELAXED(ð_data.stat.n_rcv, 1); + FETCH_ADD_RELAXED(ð_data.stat.n_mgmt_rcv, 1); } else { pthread_rwlock_rdlock(ð_data.flows_lock); @@ -959,6 +1422,7 @@ static void * eth_ipcp_packet_reader(void * o) #endif if (fd < 0) { pthread_rwlock_unlock(ð_data.flows_lock); + FETCH_ADD_RELAXED(ð_data.stat.n_bad_id, 1); goto fail_frame; } @@ -967,28 +1431,38 @@ static void * eth_ipcp_packet_reader(void * o) || memcmp(eth_data.fd_to_ef[fd].r_addr, e_frame->src_hwaddr, MAC_SIZE)) { pthread_rwlock_unlock(ð_data.flows_lock); + FETCH_ADD_RELAXED(ð_data.stat.n_bad_id, 1); goto fail_frame; } #endif + FETCH_ADD_RELAXED(ð_data.fd_to_ef[fd].stat.p_rcv, 1); + FETCH_ADD_RELAXED(ð_data.fd_to_ef[fd].stat.b_rcv, + length); + FETCH_ADD_RELAXED(ð_data.stat.n_rcv, 1); pthread_rwlock_unlock(ð_data.flows_lock); #ifndef HAVE_NETMAP - shm_du_buff_head_release(sdb, ETH_HEADER_TOT_SIZE); - shm_du_buff_truncate(sdb, length); + ssm_pk_buff_pop(spb, ETH_HEADER_TOT_SIZE); + ssm_pk_buff_truncate(spb, length); #else - if (ipcp_sdb_reserve(&sdb, length)) + if (ipcp_spb_reserve(&spb, length)) continue; - buf = shm_du_buff_head(sdb); + buf = ssm_pk_buff_head(spb); memcpy(buf, &e_frame->payload, length); #endif - if (np1_flow_write(fd, sdb) < 0) - ipcp_sdb_release(sdb); + if (np1_flow_write(fd, spb, NP1_GET_POOL(fd)) < 0) { + ipcp_spb_release(spb); + FETCH_ADD_RELAXED( + ð_data.fd_to_ef[fd].stat.p_dlv_f, + 1); + FETCH_ADD_RELAXED(ð_data.stat.n_dlv_f, 1); + } continue; fail_frame: #ifndef HAVE_NETMAP - ipcp_sdb_release(sdb); + ipcp_spb_release(spb); #endif } } @@ -1004,7 +1478,7 @@ static void cleanup_writer(void * o) static void * eth_ipcp_packet_writer(void * o) { int fd; - struct shm_du_buff * sdb; + struct ssm_pk_buff * spb; size_t len; #if defined(BUILD_ETH_DIX) uint16_t deid; @@ -1032,17 +1506,18 @@ static void * eth_ipcp_packet_writer(void * o) if (fqueue_type(fq) != FLOW_PKT) continue; - if (np1_flow_read(fd, &sdb)) { + if (np1_flow_read(fd, &spb, NP1_GET_POOL(fd))) { log_dbg("Bad read from fd %d.", fd); continue; } - len = shm_du_buff_len(sdb); + len = ssm_pk_buff_len(spb); - if (shm_du_buff_head_alloc(sdb, ETH_HEADER_TOT_SIZE) + if (ssm_pk_buff_push(spb, ETH_HEADER_TOT_SIZE) == NULL) { log_dbg("Failed to allocate header."); - ipcp_sdb_release(sdb); + ipcp_spb_release(spb); + FETCH_ADD_RELAXED(ð_data.stat.n_buf_f, 1); continue; } @@ -1065,10 +1540,22 @@ static void * eth_ipcp_packet_writer(void * o) #elif defined(BUILD_ETH_LLC) dsap, ssap, #endif - shm_du_buff_head(sdb), - len)) + ssm_pk_buff_head(spb), + len)) { log_dbg("Failed to send frame."); - ipcp_sdb_release(sdb); + FETCH_ADD_RELAXED( + ð_data.fd_to_ef[fd].stat.p_snd_f, + 1); + FETCH_ADD_RELAXED(ð_data.stat.n_snd_f, 1); + } else { + FETCH_ADD_RELAXED( + ð_data.fd_to_ef[fd].stat.p_snd, + 1); + FETCH_ADD_RELAXED( + ð_data.fd_to_ef[fd].stat.b_snd, + len); + } + ipcp_spb_release(spb); } } @@ -1217,140 +1704,138 @@ static int open_bpf_device(void) } #endif -static int eth_ipcp_bootstrap(const struct ipcp_config * conf) -{ - int idx; - struct ifreq ifr; -#if defined(HAVE_NETMAP) - char ifn[IFNAMSIZ]; -#elif defined(HAVE_BPF) - int enable = 1; - int disable = 0; - int blen; -#endif /* HAVE_NETMAP */ - #if defined(__FreeBSD__) || defined(__APPLE__) +static int ifr_hwaddr_from_ifaddrs(struct ifreq * ifr) +{ struct ifaddrs * ifaddr; struct ifaddrs * ifa; -#elif defined(__linux__) - int skfd; -#endif -#ifndef SHM_RDRB_MULTI_BLOCK - size_t maxsz; -#endif -#if defined(HAVE_RAW_SOCKETS) - #if defined(IPCP_ETH_QDISC_BYPASS) - int qdisc_bypass = 1; - #endif /* ENABLE_QDISC_BYPASS */ - int flags; -#endif - assert(conf); - assert(conf->type == THIS_TYPE); - - ipcpi.dir_hash_algo = (enum hash_algo) conf->layer_info.dir_hash_algo; - strcpy(ipcpi.layer_name, conf->layer_info.name); - - if (strlen(conf->eth.dev) >= IFNAMSIZ) { - log_err("Invalid device name: %s.", conf->eth.dev); - return -1; - } - - memset(&ifr, 0, sizeof(ifr)); - strcpy(ifr.ifr_name, conf->eth.dev); - -#ifdef BUILD_ETH_DIX - if (conf->eth.ethertype < 0x0600 || conf->eth.ethertype == 0xFFFF) { - log_err("Invalid Ethertype: %d.", conf->eth.ethertype); - return -1; - } - eth_data.ethertype = htons(conf->eth.ethertype); -#endif + int idx; -#if defined(__FreeBSD__) || defined(__APPLE__) if (getifaddrs(&ifaddr) < 0) { log_err("Could not get interfaces."); - return -1; + goto fail_ifaddrs; } for (ifa = ifaddr, idx = 0; ifa != NULL; ifa = ifa->ifa_next, ++idx) { - if (strcmp(ifa->ifa_name, conf->eth.dev)) - continue; - log_dbg("Interface %s found.", conf->eth.dev); - - #if defined(HAVE_NETMAP) || defined(HAVE_BPF) - memcpy(eth_data.hw_addr, - LLADDR((struct sockaddr_dl *) (ifa)->ifa_addr), - MAC_SIZE); - #elif defined (HAVE_RAW_SOCKETS) - memcpy(&ifr.ifr_addr, ifa->ifa_addr, sizeof(*ifa->ifa_addr)); - #endif - break; + if (strcmp(ifa->ifa_name, ifr->ifr_name) == 0) + break; } - freeifaddrs(ifaddr); - if (ifa == NULL) { log_err("Interface not found."); - return -1; + goto fail_ifa; } + memcpy(&ifr->ifr_addr, ifa->ifa_addr, sizeof(*ifa->ifa_addr)); + + log_dbg("Interface %s hwaddr " MAC_FMT ".", ifr->ifr_name, + MAC_VAL(ifr->ifr_addr.sa_data)); + + freeifaddrs(ifaddr); + + return 0; + fail_ifa: + freeifaddrs(ifaddr); + fail_ifaddrs: + return -1; + +} #elif defined(__linux__) +static int ifr_hwaddr_from_socket(struct ifreq * ifr) +{ + int skfd; + skfd = socket(AF_UNIX, SOCK_STREAM, 0); if (skfd < 0) { log_err("Failed to open socket."); - return -1; + goto fail_socket; } - if (ioctl(skfd, SIOCGIFMTU, &ifr)) { - log_err("Failed to get MTU."); - close(skfd); - return -1; + if (ioctl(skfd, SIOCGIFHWADDR, ifr)) { + log_err("Failed to get hwaddr."); + goto fail_ifr; } - log_dbg("Device MTU is %d.", ifr.ifr_mtu); + log_dbg("Interface %s hwaddr " MAC_FMT ".", ifr->ifr_name, + MAC_VAL(ifr->ifr_hwaddr.sa_data)); - eth_data.mtu = MIN((int) ETH_MTU_MAX, ifr.ifr_mtu); - if (memcmp(conf->eth.dev, "lo", 2) == 0 && - eth_data.mtu > IPCP_ETH_LO_MTU) { - log_dbg("Using loopback interface. MTU restricted to %d.", - IPCP_ETH_LO_MTU); - eth_data.mtu = IPCP_ETH_LO_MTU; - } + close(skfd); -#ifndef SHM_RDRB_MULTI_BLOCK - maxsz = SHM_RDRB_BLOCK_SIZE - 5 * sizeof(size_t) - - (DU_BUFF_HEADSPACE + DU_BUFF_TAILSPACE); - if ((size_t) eth_data.mtu > maxsz ) { - log_dbg("Layer MTU truncated to shm block size."); - eth_data.mtu = maxsz; - } + return 0; + + fail_ifr: + close(skfd); + fail_socket: + return -1; +} #endif - log_dbg("Layer MTU is %d.", eth_data.mtu); - if (ioctl(skfd, SIOCGIFHWADDR, &ifr)) { - log_err("Failed to get hwaddr."); - close(skfd); - return -1; +static int eth_ifr_hwaddr(struct ifreq * ifr) +{ +#if defined(__FreeBSD__) || defined(__APPLE__) + return ifr_hwaddr_from_ifaddrs(ifr); +#elif defined(__linux__) + return ifr_hwaddr_from_socket(ifr); +#else + return -1; +#endif +} + +static int eth_ifr_mtu(struct ifreq * ifr) +{ + int skfd; + + skfd = socket(AF_UNIX, SOCK_STREAM, 0); + if (skfd < 0) { + log_err("Failed to open socket."); + goto fail_socket; } + if (ioctl(skfd, SIOCGIFMTU, ifr) < 0) { + log_err("Failed to get MTU."); + goto fail_mtu; + } close(skfd); - idx = if_nametoindex(conf->eth.dev); - if (idx == 0) { - log_err("Failed to retrieve interface index."); + return 0; + + fail_mtu: + close(skfd); + fail_socket: + return -1; +} + +static int eth_set_mtu(struct ifreq * ifr) +{ + if (eth_ifr_mtu(ifr) < 0) { + log_err("Failed to get interface MTU."); return -1; } - eth_data.if_idx = idx; -#endif /* __FreeBSD__ */ + log_dbg("Device MTU is %d.", ifr->ifr_mtu); + + eth_data.mtu = MIN((int) ETH_MTU_MAX, ifr->ifr_mtu); + if (memcmp(ifr->ifr_name, "lo", 2) == 0 && + eth_data.mtu > IPCP_ETH_LO_MTU) { + log_dbg("Using loopback interface. MTU restricted to %d.", + IPCP_ETH_LO_MTU); + eth_data.mtu = IPCP_ETH_LO_MTU; + } + + log_dbg("Layer MTU is %d.", eth_data.mtu); + + return 0; +} #if defined(HAVE_NETMAP) +static int eth_init_nmd(struct ifreq * ifr) +{ strcpy(ifn, "netmap:"); - strcat(ifn, conf->eth.dev); + strcat(ifn, ifr->ifr_name); eth_data.nmd = nm_open(ifn, NULL, 0, NULL); if (eth_data.nmd == NULL) { log_err("Failed to open netmap device."); - return -1; + goto fail_nmd; } memset(ð_data.poll_in, 0, sizeof(eth_data.poll_in)); @@ -1362,11 +1847,22 @@ static int eth_ipcp_bootstrap(const struct ipcp_config * conf) eth_data.poll_out.events = POLLOUT; log_info("Using netmap device."); -#elif defined(HAVE_BPF) /* !HAVE_NETMAP */ + + return 0; + fail_nmd: + return -1; +} +#elif defined (HAVE_BPF) +static int eth_init_bpf(struct ifreq * ifr) +{ + int enable = 1; + int disable = 0; + int blen; + eth_data.bpf = open_bpf_device(); if (eth_data.bpf < 0) { log_err("Failed to open bpf device."); - return -1; + goto fail_bpf; } ioctl(eth_data.bpf, BIOCGBLEN, &blen); @@ -1376,7 +1872,7 @@ static int eth_ipcp_bootstrap(const struct ipcp_config * conf) goto fail_device; } - if (ioctl(eth_data.bpf, BIOCSETIF, &ifr) < 0) { + if (ioctl(eth_data.bpf, BIOCSETIF, ifr) < 0) { log_err("Failed to set interface."); goto fail_device; } @@ -1397,60 +1893,152 @@ static int eth_ipcp_bootstrap(const struct ipcp_config * conf) } log_info("Using Berkeley Packet Filter."); + + return 0; + + fail_device: + close(eth_data.bpf); + fail_bpf: + return -1; +} #elif defined(HAVE_RAW_SOCKETS) +#define SOCKOPT() +static int eth_init_raw_socket(struct ifreq * ifr) +{ + int idx; + int sndbuf; + int rcvbuf; +#if defined(IPCP_ETH_QDISC_BYPASS) + int qdisc_bypass = 1; +#endif /* ENABLE_QDISC_BYPASS */ + + idx = if_nametoindex(ifr->ifr_name); + if (idx == 0) { + log_err("Failed to retrieve interface index."); + return -1; + } + memset(&(eth_data.device), 0, sizeof(eth_data.device)); eth_data.device.sll_ifindex = idx; eth_data.device.sll_family = AF_PACKET; - memcpy(eth_data.device.sll_addr, ifr.ifr_hwaddr.sa_data, MAC_SIZE); + memcpy(eth_data.device.sll_addr, ifr->ifr_hwaddr.sa_data, MAC_SIZE); eth_data.device.sll_halen = MAC_SIZE; eth_data.device.sll_protocol = htons(ETH_P_ALL); - - #if defined (BUILD_ETH_DIX) +#if defined (BUILD_ETH_DIX) eth_data.s_fd = socket(AF_PACKET, SOCK_RAW, eth_data.ethertype); - #elif defined (BUILD_ETH_LLC) +#elif defined (BUILD_ETH_LLC) eth_data.s_fd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_802_2)); - #endif - - log_info("Using raw socket device."); - +#endif if (eth_data.s_fd < 0) { log_err("Failed to create socket."); goto fail_socket; } - flags = fcntl(eth_data.s_fd, F_GETFL, 0); - if (flags < 0) { - log_err("Failed to get flags."); - goto fail_device; +#if defined(IPCP_ETH_QDISC_BYPASS) + if (setsockopt(eth_data.s_fd, SOL_PACKET, PACKET_QDISC_BYPASS, + &qdisc_bypass, sizeof(qdisc_bypass))) { + log_info("Qdisc bypass not supported."); } +#endif - if (fcntl(eth_data.s_fd, F_SETFL, flags | O_NONBLOCK)) { - log_err("Failed to set socket non-blocking."); - goto fail_device; + sndbuf = IPCP_ETH_SNDBUF; + if (sndbuf > 0 && setsockopt(eth_data.s_fd, SOL_SOCKET, SO_SNDBUF, + &sndbuf, sizeof(sndbuf))) { + log_info("Failed to set SO_SNDBUF to %d.", sndbuf); } - #if defined(IPCP_ETH_QDISC_BYPASS) - if (setsockopt(eth_data.s_fd, SOL_PACKET, PACKET_QDISC_BYPASS, - &qdisc_bypass, sizeof(qdisc_bypass))) { - log_info("Qdisc bypass not supported."); + rcvbuf = IPCP_ETH_RCVBUF; + if (rcvbuf > 0 && setsockopt(eth_data.s_fd, SOL_SOCKET, SO_RCVBUF, + &rcvbuf, sizeof(rcvbuf))) { + log_info("Failed to set SO_RCVBUF to %d.", rcvbuf); } - #endif if (bind(eth_data.s_fd, (struct sockaddr *) ð_data.device, sizeof(eth_data.device)) < 0) { log_err("Failed to bind socket to interface."); goto fail_device; } +#ifdef __linux__ + eth_data.if_idx = idx; +#endif + log_info("Using raw socket device."); + + return 0; + fail_device: + close(eth_data.s_fd); + fail_socket: + return -1; +} +#endif + +static int eth_ipcp_bootstrap(struct ipcp_config * conf) +{ + struct ifreq ifr; + int i; +#if defined(HAVE_NETMAP) + char ifn[IFNAMSIZ]; #endif /* HAVE_NETMAP */ + + assert(conf); + assert(conf->type == THIS_TYPE); + + memset(&ifr, 0, sizeof(ifr)); + strcpy(ifr.ifr_name, conf->eth.dev); + + if (strlen(conf->eth.dev) >= IFNAMSIZ) { + log_err("Invalid device name: %s.", conf->eth.dev); + return -1; + } +#ifdef BUILD_ETH_DIX + if (conf->eth.ethertype < 0x0600 || conf->eth.ethertype == 0xFFFF) { + log_err("Invalid Ethertype: %d.", conf->eth.ethertype); + return -1; + } + eth_data.ethertype = htons(conf->eth.ethertype); +#endif + if (eth_set_mtu(&ifr) < 0) { + log_err("Failed to set MTU."); + return -1; + } + + if (eth_ifr_hwaddr(&ifr) < 0) { + log_err("Failed to get hardware addr."); + return -1; + } +#if defined(HAVE_NETMAP) || defined(HAVE_BPF) + memcpy(eth_data.hw_addr, LLADDR((struct sockaddr_dl *) &ifr.ifr_addr), + MAC_SIZE); +#endif +#if defined(HAVE_NETMAP) + if (eth_init_nmd(&ifr) < 0) { + log_err("Failed to initialize netmap device."); + return -1; + } +#elif defined(HAVE_BPF) /* !HAVE_NETMAP */ + if (eth_init_bpf(&ifr) < 0) { + log_err("Failed to initialize BPF device."); + return -1; + } +#elif defined(HAVE_RAW_SOCKETS) + if (eth_init_raw_socket(&ifr) < 0) { + log_err("Failed to initialize raw socket device."); + return -1; + } +#endif /* HAVE_NETMAP */ +#ifdef IPCP_ETH_FLOW_STATS + if (rib_reg(ETH_RIB_PATH, ð_r_ops)) { + log_err("Failed to register RIB."); + goto fail_rib_reg; + } +#endif #if defined(__linux__) if (pthread_create(ð_data.if_monitor, NULL, eth_ipcp_if_monitor, NULL)) { log_err("Failed to create monitor thread: %s.", strerror(errno)); - goto fail_device; + goto fail_monitor; } #endif - if (pthread_create(ð_data.mgmt_handler, NULL, eth_ipcp_mgmt_handler, NULL)) { log_err("Failed to create mgmt handler thread: %s.", @@ -1458,8 +2046,8 @@ static int eth_ipcp_bootstrap(const struct ipcp_config * conf) goto fail_mgmt_handler; } - for (idx = 0; idx < IPCP_ETH_RD_THR; ++idx) { - if (pthread_create(ð_data.packet_reader[idx], NULL, + for (i = 0; i < IPCP_ETH_RD_THR; i++) { + if (pthread_create(ð_data.packet_reader[i], NULL, eth_ipcp_packet_reader, NULL)) { log_err("Failed to create packet reader thread: %s", strerror(errno)); @@ -1467,8 +2055,8 @@ static int eth_ipcp_bootstrap(const struct ipcp_config * conf) } } - for (idx = 0; idx < IPCP_ETH_WR_THR; ++idx) { - if (pthread_create(ð_data.packet_writer[idx], NULL, + for (i = 0; i < IPCP_ETH_WR_THR; i++) { + if (pthread_create(ð_data.packet_writer[i], NULL, eth_ipcp_packet_writer, NULL)) { log_err("Failed to create packet writer thread: %s", strerror(errno)); @@ -1486,15 +2074,15 @@ static int eth_ipcp_bootstrap(const struct ipcp_config * conf) return 0; fail_packet_writer: - while (idx > 0) { - pthread_cancel(eth_data.packet_writer[--idx]); - pthread_join(eth_data.packet_writer[idx], NULL); + while (i-- > 0) { + pthread_cancel(eth_data.packet_writer[i]); + pthread_join(eth_data.packet_writer[i], NULL); } - idx = IPCP_ETH_RD_THR; + i = IPCP_ETH_RD_THR; fail_packet_reader: - while (idx > 0) { - pthread_cancel(eth_data.packet_reader[--idx]); - pthread_join(eth_data.packet_reader[idx], NULL); + while (i-- > 0) { + pthread_cancel(eth_data.packet_reader[i]); + pthread_join(eth_data.packet_reader[i], NULL); } pthread_cancel(eth_data.mgmt_handler); pthread_join(eth_data.mgmt_handler, NULL); @@ -1503,8 +2091,12 @@ static int eth_ipcp_bootstrap(const struct ipcp_config * conf) pthread_cancel(eth_data.if_monitor); pthread_join(eth_data.if_monitor, NULL); #endif -#if defined(__linux__) || !defined(HAVE_NETMAP) - fail_device: +#if defined(__linux__) + fail_monitor: +#endif +#ifdef IPCP_ETH_FLOW_STATS + rib_unreg(ETH_RIB_PATH); + fail_rib_reg: #endif #if defined(HAVE_NETMAP) nm_close(eth_data.nmd); @@ -1513,7 +2105,6 @@ static int eth_ipcp_bootstrap(const struct ipcp_config * conf) #elif defined(HAVE_RAW_SOCKETS) close(eth_data.s_fd); #endif - fail_socket: return -1; } @@ -1538,12 +2129,14 @@ static int eth_ipcp_unreg(const uint8_t * hash) static int eth_ipcp_query(const uint8_t * hash) { uint8_t r_addr[MAC_SIZE]; - struct timespec timeout = TIMESPEC_INIT_MS(NAME_QUERY_TIMEO); + struct timespec timeout; struct dir_query * query; int ret; + int attempt; uint8_t * buf; struct mgmt_msg * msg; size_t len; + long per_ms; if (shim_data_dir_has(eth_data.shim_data, hash)) return 0; @@ -1554,6 +2147,8 @@ static int eth_ipcp_query(const uint8_t * hash) if (buf == NULL) return -1; + memset(buf, 0, len + ETH_HEADER_TOT_SIZE); + msg = (struct mgmt_msg *) (buf + ETH_HEADER_TOT_SIZE); msg->code = NAME_QUERY_REQ; @@ -1561,32 +2156,46 @@ static int eth_ipcp_query(const uint8_t * hash) memset(r_addr, 0xff, MAC_SIZE); - query = shim_data_dir_query_create(eth_data.shim_data, hash); - if (query == NULL) { - free(buf); - return -1; - } + per_ms = NAME_QUERY_TIMEO / (NAME_QUERY_RETRIES + 1); + + ret = -1; + for (attempt = 0; attempt <= NAME_QUERY_RETRIES; ++attempt) { + query = shim_data_dir_query_create(eth_data.shim_data, hash); + if (query == NULL) { + ret = -1; + break; + } - if (eth_ipcp_send_frame(r_addr, + if (eth_ipcp_send_frame(r_addr, #if defined(BUILD_ETH_DIX) - MGMT_EID, + MGMT_EID, #elif defined(BUILD_ETH_LLC) - reverse_bits(MGMT_SAP), - reverse_bits(MGMT_SAP), + reverse_bits(MGMT_SAP), + reverse_bits(MGMT_SAP), #endif - buf, len)) { - log_err("Failed to send management frame."); + buf, len)) { + log_err("Failed to send management frame."); + shim_data_dir_query_destroy(eth_data.shim_data, + query); + ret = -1; + break; + } + + FETCH_ADD_RELAXED(ð_data.stat.n_mgmt_snd, 1); + + timeout.tv_sec = per_ms / 1000; + timeout.tv_nsec = (per_ms % 1000) * 1000000L; + + ret = shim_data_dir_query_wait(query, &timeout); + shim_data_dir_query_destroy(eth_data.shim_data, query); - free(buf); - return -1; + + if (ret != -ETIMEDOUT) + break; } free(buf); - ret = shim_data_dir_query_wait(query, &timeout); - - shim_data_dir_query_destroy(eth_data.shim_data, query); - return ret; } @@ -1599,7 +2208,7 @@ static int eth_ipcp_flow_alloc(int fd, uint8_t ssap = 0; #endif uint8_t r_addr[MAC_SIZE]; - uint64_t addr = 0; + struct addr addr; assert(hash); @@ -1608,10 +2217,12 @@ static int eth_ipcp_flow_alloc(int fd, HASH_VAL32(hash)); return -1; } + addr = shim_data_dir_get_addr(eth_data.shim_data, hash); + memcpy(r_addr, &addr.mac, MAC_SIZE); - pthread_rwlock_wrlock(ð_data.flows_lock); #ifdef BUILD_ETH_LLC + pthread_rwlock_wrlock(ð_data.flows_lock); ssap = bmp_allocate(eth_data.saps); if (!bmp_is_id_valid(eth_data.saps, ssap)) { pthread_rwlock_unlock(ð_data.flows_lock); @@ -1621,10 +2232,8 @@ static int eth_ipcp_flow_alloc(int fd, eth_data.fd_to_ef[fd].sap = ssap; eth_data.ef_to_fd[ssap] = fd; -#endif pthread_rwlock_unlock(ð_data.flows_lock); - - memcpy(r_addr, &addr, MAC_SIZE); +#endif if (eth_ipcp_alloc(r_addr, #if defined(BUILD_ETH_DIX) @@ -1647,6 +2256,14 @@ static int eth_ipcp_flow_alloc(int fd, } fset_add(eth_data.np1_flows, fd); +#ifdef IPCP_ETH_FLOW_STATS + pthread_rwlock_wrlock(ð_data.flows_lock); + memset(ð_data.fd_to_ef[fd].stat, 0, + sizeof(eth_data.fd_to_ef[fd].stat)); + eth_data.fd_to_ef[fd].stat.stamp = time(NULL); + FETCH_ADD_RELAXED(ð_data.stat.n_flows, 1); + pthread_rwlock_unlock(ð_data.flows_lock); +#endif #if defined(BUILD_ETH_LLC) log_dbg("Assigned SAP %d for fd %d.", ssap, fd); #endif @@ -1707,6 +2324,14 @@ static int eth_ipcp_flow_alloc_resp(int fd, } fset_add(eth_data.np1_flows, fd); +#ifdef IPCP_ETH_FLOW_STATS + pthread_rwlock_wrlock(ð_data.flows_lock); + memset(ð_data.fd_to_ef[fd].stat, 0, + sizeof(eth_data.fd_to_ef[fd].stat)); + eth_data.fd_to_ef[fd].stat.stamp = time(NULL); + FETCH_ADD_RELAXED(ð_data.stat.n_flows, 1); + pthread_rwlock_unlock(ð_data.flows_lock); +#endif #if defined(BUILD_ETH_LLC) log_dbg("Assigned SAP %d for fd %d.", ssap, fd); #endif @@ -1735,6 +2360,12 @@ static int eth_ipcp_flow_dealloc(int fd) #endif memset(ð_data.fd_to_ef[fd].r_addr, 0, MAC_SIZE); +#ifdef IPCP_ETH_FLOW_STATS + memset(ð_data.fd_to_ef[fd].stat, 0, + sizeof(eth_data.fd_to_ef[fd].stat)); + FETCH_SUB_RELAXED(ð_data.stat.n_flows, 1); +#endif + pthread_rwlock_unlock(ð_data.flows_lock); ipcp_flow_dealloc(fd); @@ -1753,7 +2384,8 @@ static struct ipcp_ops eth_ops = { .ipcp_flow_alloc = eth_ipcp_flow_alloc, .ipcp_flow_join = NULL, .ipcp_flow_alloc_resp = eth_ipcp_flow_alloc_resp, - .ipcp_flow_dealloc = eth_ipcp_flow_dealloc + .ipcp_flow_dealloc = eth_ipcp_flow_dealloc, + .ipcp_flow_update = eth_ipcp_flow_update }; int main(int argc, @@ -1801,6 +2433,9 @@ int main(int argc, #ifdef __linux__ pthread_join(eth_data.if_monitor, NULL); #endif +#ifdef IPCP_ETH_FLOW_STATS + rib_unreg(ETH_RIB_PATH); +#endif } ipcp_stop(); diff --git a/src/ipcpd/eth/llc.c b/src/ipcpd/eth/llc.c index c900dcab..a772e86e 100644 --- a/src/ipcpd/eth/llc.c +++ b/src/ipcpd/eth/llc.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * IPC processes over Ethernet - LLC * diff --git a/src/ipcpd/ipcp.c b/src/ipcpd/ipcp.c index 966c4920..dcee4b9c 100644 --- a/src/ipcpd/ipcp.c +++ b/src/ipcpd/ipcp.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * IPC process main loop * @@ -52,6 +52,7 @@ #include <ouroboros/utils.h> #include "ipcp.h" +#include "np1.h" #include <signal.h> #include <string.h> @@ -64,13 +65,75 @@ #endif #endif -char * info[LAYER_NAME_SIZE + 1] = { - "_state", - "_type", - "_layer", - NULL +#ifndef CLOCK_REALTIME_COARSE +#define CLOCK_REALTIME_COARSE CLOCK_REALTIME +#endif + +static char * ipcp_type_str[] = { + "local", + "unicast", + "broadcast", + "eth-llc", + "eth-dix", + "udp4", + "udp6" +}; + +static char * dir_hash_str[] = { + "SHA3-224", + "SHA3-256", + "SHA3-384", + "SHA3-512", + "CRC32", + "MD5" +}; + +static char * ipcp_state_str[] = { + "null", + "init", + "boot", + "bootstrapped", + "enrolled", + "operational", + "shutdown" }; +struct { + pid_t irmd_pid; + char * name; + + enum ipcp_type type; + char layer_name[LAYER_NAME_SIZE + 1]; + + uint64_t dt_addr; + + enum hash_algo dir_hash_algo; + + struct ipcp_ops * ops; + int irmd_fd; + + enum ipcp_state state; + pthread_cond_t state_cond; + pthread_mutex_t state_mtx; + + int sockfd; + char * sock_path; + + struct list_head cmds; + pthread_cond_t cmd_cond; + pthread_mutex_t cmd_lock; + + int alloc_id; + pthread_cond_t alloc_cond; + pthread_mutex_t alloc_lock; + + struct tpm * tpm; + + pthread_t acceptor; +} ipcpd; + +struct np1_state np1; + struct cmd { struct list_head next; @@ -79,9 +142,38 @@ struct cmd { int fd; }; +enum ipcp_type ipcp_get_type(void) +{ + return ipcpd.type; +} + +const char * ipcp_get_name(void) +{ + return ipcpd.name; +} + +void ipcp_set_dir_hash_algo(enum hash_algo algo) +{ + ipcpd.dir_hash_algo = algo; +} + +size_t ipcp_dir_hash_len(void) +{ + return hash_len(ipcpd.dir_hash_algo); +} + +int ipcp_get_layer_name(char * layer) +{ + if (ipcp_get_state() < IPCP_OPERATIONAL) + return -EIPCPSTATE; + + strcpy(layer, ipcpd.layer_name); + return 0; +} + uint8_t * ipcp_hash_dup(const uint8_t * hash) { - uint8_t * dup = malloc(hash_len(ipcpi.dir_hash_algo)); + uint8_t * dup = malloc(hash_len(ipcpd.dir_hash_algo)); if (dup == NULL) return NULL; @@ -105,11 +197,18 @@ void ipcp_hash_str(char * buf, buf[2 * i] = '\0'; } +static const char * info[] = { + "_state", + "_type", + "_layer", + NULL +}; + static int ipcp_rib_read(const char * path, char * buf, size_t len) { - char * entry; + const char * entry; if (len < LAYER_NAME_SIZE + 2) /* trailing \n */ return 0; @@ -132,18 +231,20 @@ static int ipcp_rib_read(const char * path, } if (strcmp(entry, info[1]) == 0) { /* _type */ - if (ipcpi.type == IPCP_LOCAL) + if (ipcpd.type == IPCP_LOCAL) strcpy(buf, "local\n"); - else if (ipcpi.type == IPCP_UNICAST) + else if (ipcpd.type == IPCP_UNICAST) strcpy(buf, "unicast\n"); - else if (ipcpi.type == IPCP_BROADCAST) + else if (ipcpd.type == IPCP_BROADCAST) strcpy(buf, "broadcast\n"); - else if (ipcpi.type == IPCP_ETH_LLC) + else if (ipcpd.type == IPCP_ETH_LLC) strcpy(buf, "eth-llc\n"); - else if (ipcpi.type == IPCP_ETH_DIX) + else if (ipcpd.type == IPCP_ETH_DIX) strcpy(buf, "eth-dix\n"); - else if (ipcpi.type == IPCP_UDP) - strcpy(buf, "udp\n"); + else if (ipcpd.type == IPCP_UDP4) + strcpy(buf, "udp4\n"); + else if (ipcpd.type == IPCP_UDP6) + strcpy(buf, "udp6\n"); else strcpy(buf, "bug\n"); } @@ -153,7 +254,7 @@ static int ipcp_rib_read(const char * path, if (ipcp_get_state() < IPCP_OPERATIONAL) strcpy(buf, "(null)"); else - strcpy(buf, ipcpi.layer_name); + strcpy(buf, ipcpd.layer_name); buf[strlen(buf)] = '\n'; } @@ -165,12 +266,11 @@ static int ipcp_rib_readdir(char *** buf) { int i = 0; - while (info[i] != NULL) - i++; + while (info[i++] != NULL); *buf = malloc(sizeof(**buf) * i); if (*buf == NULL) - goto fail; + goto fail_entries; i = 0; @@ -183,12 +283,11 @@ static int ipcp_rib_readdir(char *** buf) return i; fail_dup: - while (i > 0) - free((*buf)[--i]); - fail: + while (i-- > 0) + free((*buf)[i]); free(*buf); - - return -1; + fail_entries: + return -ENOMEM; } static int ipcp_rib_getattr(const char * path, @@ -218,10 +317,10 @@ static void * acceptloop(void * o) (void) o; while (ipcp_get_state() != IPCP_SHUTDOWN && - ipcp_get_state() != IPCP_NULL) { + ipcp_get_state() != IPCP_INIT) { struct cmd * cmd; - csockfd = accept(ipcpi.sockfd, 0, 0); + csockfd = accept(ipcpd.sockfd, 0, 0); if (csockfd < 0) continue; @@ -249,13 +348,13 @@ static void * acceptloop(void * o) cmd->fd = csockfd; - pthread_mutex_lock(&ipcpi.cmd_lock); + pthread_mutex_lock(&ipcpd.cmd_lock); - list_add(&cmd->next, &ipcpi.cmds); + list_add(&cmd->next, &ipcpd.cmds); - pthread_cond_signal(&ipcpi.cmd_cond); + pthread_cond_signal(&ipcpd.cmd_cond); - pthread_mutex_unlock(&ipcpi.cmd_lock); + pthread_mutex_unlock(&ipcpd.cmd_lock); } return (void *) 0; @@ -264,6 +363,7 @@ static void * acceptloop(void * o) int ipcp_wait_flow_req_arr(const uint8_t * dst, qosspec_t qs, time_t mpl, + uint32_t mtu, const buffer_t * data) { struct timespec ts = TIMESPEC_INIT_MS(ALLOC_TIMEOUT); @@ -276,34 +376,34 @@ int ipcp_wait_flow_req_arr(const uint8_t * dst, clock_gettime(PTHREAD_COND_CLOCK, &abstime); - pthread_mutex_lock(&ipcpi.alloc_lock); + pthread_mutex_lock(&ipcpd.alloc_lock); - while (ipcpi.alloc_id != -1 && ipcp_get_state() == IPCP_OPERATIONAL) { + while (ipcpd.alloc_id != -1 && ipcp_get_state() == IPCP_OPERATIONAL) { ts_add(&abstime, &ts, &abstime); - pthread_cond_timedwait(&ipcpi.alloc_cond, - &ipcpi.alloc_lock, + pthread_cond_timedwait(&ipcpd.alloc_cond, + &ipcpd.alloc_lock, &abstime); } if (ipcp_get_state() != IPCP_OPERATIONAL) { - pthread_mutex_unlock(&ipcpi.alloc_lock); + pthread_mutex_unlock(&ipcpd.alloc_lock); log_err("Won't allocate over non-operational IPCP."); return -EIPCPSTATE; } - assert(ipcpi.alloc_id == -1); + assert(ipcpd.alloc_id == -1); - fd = ipcp_flow_req_arr(&hash, qs, mpl, data); + fd = ipcp_flow_req_arr(&hash, qs, mpl, mtu, data); if (fd < 0) { - pthread_mutex_unlock(&ipcpi.alloc_lock); + pthread_mutex_unlock(&ipcpd.alloc_lock); log_err("Failed to get fd for flow."); return fd; } - ipcpi.alloc_id = fd; - pthread_cond_broadcast(&ipcpi.alloc_cond); + ipcpd.alloc_id = fd; + pthread_cond_broadcast(&ipcpd.alloc_cond); - pthread_mutex_unlock(&ipcpi.alloc_lock); + pthread_mutex_unlock(&ipcpd.alloc_lock); return fd; @@ -311,31 +411,31 @@ int ipcp_wait_flow_req_arr(const uint8_t * dst, int ipcp_wait_flow_resp(const int fd) { - struct timespec ts = TIMESPEC_INIT_MS(ALLOC_TIMEOUT); - struct timespec abstime; + struct timespec ts = TIMESPEC_INIT_MS(ALLOC_TIMEOUT); + struct timespec abstime; clock_gettime(PTHREAD_COND_CLOCK, &abstime); - pthread_mutex_lock(&ipcpi.alloc_lock); + pthread_mutex_lock(&ipcpd.alloc_lock); - while (ipcpi.alloc_id != fd && ipcp_get_state() == IPCP_OPERATIONAL) { + while (ipcpd.alloc_id != fd && ipcp_get_state() == IPCP_OPERATIONAL) { ts_add(&abstime, &ts, &abstime); - pthread_cond_timedwait(&ipcpi.alloc_cond, - &ipcpi.alloc_lock, + pthread_cond_timedwait(&ipcpd.alloc_cond, + &ipcpd.alloc_lock, &abstime); } if (ipcp_get_state() != IPCP_OPERATIONAL) { - pthread_mutex_unlock(&ipcpi.alloc_lock); + pthread_mutex_unlock(&ipcpd.alloc_lock); return -1; } - assert(ipcpi.alloc_id == fd); + assert(ipcpd.alloc_id == fd); - ipcpi.alloc_id = -1; - pthread_cond_broadcast(&ipcpi.alloc_cond); + ipcpd.alloc_id = -1; + pthread_cond_broadcast(&ipcpd.alloc_cond); - pthread_mutex_unlock(&ipcpi.alloc_lock); + pthread_mutex_unlock(&ipcpd.alloc_lock); return 0; } @@ -349,30 +449,58 @@ static void free_msg(void * o) static void do_bootstrap(ipcp_config_msg_t * conf_msg, ipcp_msg_t * ret_msg) { - struct ipcp_config conf; + struct ipcp_config conf; + struct layer_info * info; log_info("Bootstrapping..."); - if (ipcpi.ops->ipcp_bootstrap == NULL) { - log_err("Bootstrap unsupported."); + if (ipcpd.ops->ipcp_bootstrap == NULL) { + log_err("Failed to Bootstrap: operation unsupported."); ret_msg->result = -ENOTSUP; - goto finish; + return; } - if (ipcp_get_state() != IPCP_INIT) { - log_err("IPCP in wrong state."); + if (ipcp_get_state() != IPCP_BOOT) { + + log_err("Failed to bootstrap: IPCP in state <%s>, need <%s>.", + ipcp_state_str[ipcp_get_state()], + ipcp_state_str[IPCP_BOOT]); ret_msg->result = -EIPCPSTATE; - goto finish; + return; } conf = ipcp_config_msg_to_s(conf_msg); - ret_msg->result = ipcpi.ops->ipcp_bootstrap(&conf); - if (ret_msg->result == 0) { - ret_msg->layer_info = layer_info_s_to_msg(&conf.layer_info); - ipcp_set_state(IPCP_OPERATIONAL); + switch(conf.type) { /* FIXED algorithms */ + case IPCP_UDP4: + /* FALLTHRU */ + case IPCP_UDP6: + conf.layer_info.dir_hash_algo = (enum pol_dir_hash) HASH_MD5; + break; + case IPCP_BROADCAST: + conf.layer_info.dir_hash_algo = DIR_HASH_SHA3_256; + break; + default: + break; } - finish: - log_info("Finished bootstrapping: %d.", ret_msg->result); + + ret_msg->result = ipcpd.ops->ipcp_bootstrap(&conf); + if (ret_msg->result < 0) { + log_err("Failed to bootstrap IPCP."); + return; + } + + info = &conf.layer_info; + + strcpy(ipcpd.layer_name, info->name); + ipcpd.dir_hash_algo = (enum hash_algo) info->dir_hash_algo; + ret_msg->layer_info = layer_info_s_to_msg(info); + ipcp_set_state(IPCP_OPERATIONAL); + + log_info("Finished bootstrapping in %s.", info->name); + log_info(" type: %s", ipcp_type_str[ipcpd.type]); + log_info(" hash: %s [%zd bytes]", + dir_hash_str[ipcpd.dir_hash_algo], + ipcp_dir_hash_len()); } static void do_enroll(const char * dst, @@ -382,25 +510,36 @@ static void do_enroll(const char * dst, log_info("Enrolling with %s...", dst); - if (ipcpi.ops->ipcp_enroll == NULL) { - log_err("Enroll unsupported."); + if (ipcpd.ops->ipcp_enroll == NULL) { + log_err("Failed to enroll: operation unsupported."); ret_msg->result = -ENOTSUP; - goto finish; + return; } - if (ipcp_get_state() != IPCP_INIT) { - log_err("IPCP in wrong state."); + if (ipcp_get_state() != IPCP_BOOT) { + log_err("Failed to enroll: IPCP in state <%s>, need <%s>.", + ipcp_state_str[ipcp_get_state()], + ipcp_state_str[IPCP_BOOT]); ret_msg->result = -EIPCPSTATE; - goto finish; + return; } - ret_msg->result = ipcpi.ops->ipcp_enroll(dst, &info); - if (ret_msg->result == 0) { - ret_msg->layer_info = layer_info_s_to_msg(&info); - ipcp_set_state(IPCP_OPERATIONAL); + ret_msg->result = ipcpd.ops->ipcp_enroll(dst, &info); + if (ret_msg->result < 0) { + log_err("Failed to bootstrap IPCP."); + return; } - finish: - log_info("Finished enrolling with %s: %d.", dst, ret_msg->result); + + strcpy(ipcpd.layer_name, info.name); + ipcpd.dir_hash_algo = (enum hash_algo) info.dir_hash_algo; + ret_msg->layer_info = layer_info_s_to_msg(&info); + ipcp_set_state(IPCP_OPERATIONAL); + + log_info("Finished enrolling with %s in layer %s.", dst, info.name); + log_info(" type: %s", ipcp_type_str[ipcpd.type]); + log_info(" hash: %s [%zd bytes]", + dir_hash_str[ipcpd.dir_hash_algo], + ipcp_dir_hash_len()); } static void do_connect(const char * dst, @@ -410,15 +549,15 @@ static void do_connect(const char * dst, { log_info("Connecting %s to %s...", comp, dst); - if (ipcpi.ops->ipcp_connect == NULL) { - log_err("Connect unsupported."); + if (ipcpd.ops->ipcp_connect == NULL) { + log_err("Failed to connect: operation unsupported."); ret_msg->result = -ENOTSUP; - goto finish; + return; } - ret_msg->result = ipcpi.ops->ipcp_connect(dst, comp, qs); - finish: - log_info("Finished connecting: %d.", ret_msg->result); + ret_msg->result = ipcpd.ops->ipcp_connect(dst, comp, qs); + + log_info("Finished connecting."); } static void do_disconnect(const char * dst, @@ -427,17 +566,15 @@ static void do_disconnect(const char * dst, { log_info("Disconnecting %s from %s...", comp, dst); - if (ipcpi.ops->ipcp_disconnect == NULL) { - log_err("Disconnect unsupported."); + if (ipcpd.ops->ipcp_disconnect == NULL) { + log_err("Failed to disconnect: operation unsupported."); ret_msg->result = -ENOTSUP; - goto finish; + return; } - ret_msg->result = ipcpi.ops->ipcp_disconnect(dst, comp); + ret_msg->result = ipcpd.ops->ipcp_disconnect(dst, comp); - finish: - log_info("Finished disconnecting %s from %s: %d.", - comp, dst, ret_msg->result); + log_info("Finished disconnecting %s from %s.", comp, dst); } static void do_reg(const uint8_t * hash, @@ -446,16 +583,15 @@ static void do_reg(const uint8_t * hash, log_info("Registering " HASH_FMT32 "...", HASH_VAL32(hash)); - if (ipcpi.ops->ipcp_reg == NULL) { - log_err("Registration unsupported."); + if (ipcpd.ops->ipcp_reg == NULL) { + log_err("Failed to register: operation unsupported."); ret_msg->result = -ENOTSUP; - goto finish; + return; } - ret_msg->result = ipcpi.ops->ipcp_reg(hash); - finish: - log_info("Finished registering " HASH_FMT32 " : %d.", - HASH_VAL32(hash), ret_msg->result); + ret_msg->result = ipcpd.ops->ipcp_reg(hash); + + log_info("Finished registering " HASH_FMT32 ".", HASH_VAL32(hash)); } static void do_unreg(const uint8_t * hash, @@ -463,16 +599,15 @@ static void do_unreg(const uint8_t * hash, { log_info("Unregistering " HASH_FMT32 "...", HASH_VAL32(hash)); - if (ipcpi.ops->ipcp_unreg == NULL) { - log_err("Unregistration unsupported."); + if (ipcpd.ops->ipcp_unreg == NULL) { + log_err("Failed to unregister: operation unsupported."); ret_msg->result = -ENOTSUP; - goto finish; + return; } - ret_msg->result = ipcpi.ops->ipcp_unreg(hash); - finish: - log_info("Finished unregistering " HASH_FMT32 ": %d.", - HASH_VAL32(hash), ret_msg->result); + ret_msg->result = ipcpd.ops->ipcp_unreg(hash); + + log_info("Finished unregistering " HASH_FMT32 ".", HASH_VAL32(hash)); } static void do_query(const uint8_t * hash, @@ -480,19 +615,21 @@ static void do_query(const uint8_t * hash, { /* TODO: Log this operation when IRMd has internal caches. */ - if (ipcpi.ops->ipcp_query == NULL) { - log_err("Directory query unsupported."); + if (ipcpd.ops->ipcp_query == NULL) { + log_err("Failed to query: operation unsupported."); ret_msg->result = -ENOTSUP; return; } if (ipcp_get_state() != IPCP_OPERATIONAL) { - log_err("IPCP in wrong state."); + log_dbg("Failed to query: IPCP in state <%s>, need <%s>.", + ipcp_state_str[ipcp_get_state()], + ipcp_state_str[IPCP_OPERATIONAL]); ret_msg->result = -EIPCPSTATE; return; } - ret_msg->result = ipcpi.ops->ipcp_query(hash); + ret_msg->result = ipcpd.ops->ipcp_query(hash); } static void do_flow_alloc(pid_t pid, @@ -500,23 +637,27 @@ static void do_flow_alloc(pid_t pid, uint8_t * dst, qosspec_t qs, const buffer_t * data, + uid_t uid, ipcp_msg_t * ret_msg) { - int fd; + int fd; + struct ssm_pool * pool = NULL; log_info("Allocating flow %d for %d to " HASH_FMT32 ".", flow_id, pid, HASH_VAL32(dst)); - if (ipcpi.ops->ipcp_flow_alloc == NULL) { - log_err("Flow allocation unsupported."); + if (ipcpd.ops->ipcp_flow_alloc == NULL) { + log_err("Flow allocation failed: operation unsupported."); ret_msg->result = -ENOTSUP; - goto finish; + return; } if (ipcp_get_state() != IPCP_OPERATIONAL) { - log_err("IPCP in wrong state."); + log_err("Failed to enroll: IPCP in state <%s>, need <%s>.", + ipcp_state_str[ipcp_get_state()], + ipcp_state_str[IPCP_OPERATIONAL]); ret_msg->result = -EIPCPSTATE; - goto finish; + return; } fd = np1_flow_alloc(pid, flow_id); @@ -524,83 +665,124 @@ static void do_flow_alloc(pid_t pid, log_err("Failed allocating n + 1 fd on flow_id %d: %d", flow_id, fd); ret_msg->result = -EFLOWDOWN; - goto finish; + return; } - ret_msg->result = ipcpi.ops->ipcp_flow_alloc(fd, dst, qs, data); - finish: - log_info("Finished allocating flow %d to " HASH_FMT32 ": %d.", - flow_id, HASH_VAL32(dst), ret_msg->result); + if (uid != 0) { + pool = ssm_pool_open(uid); + if (pool == NULL) { + log_err("Failed to open PUP for uid %d.", uid); + ret_msg->result = -ENOMEM; + return; + } + } + + NP1_SET_POOL(fd, pool); + + ret_msg->result = ipcpd.ops->ipcp_flow_alloc(fd, dst, qs, data); + + log_info("Finished allocating flow %d to " HASH_FMT32 ".", + flow_id, HASH_VAL32(dst)); } static void do_flow_join(pid_t pid, int flow_id, const uint8_t * dst, - qosspec_t qs, + uid_t uid, ipcp_msg_t * ret_msg) { - int fd; + int fd; + struct ssm_pool * pool = NULL; log_info("Joining layer " HASH_FMT32 ".", HASH_VAL32(dst)); - if (ipcpi.ops->ipcp_flow_join == NULL) { - log_err("Broadcast unsupported."); + if (ipcpd.ops->ipcp_flow_join == NULL) { + log_err("Failed to join: operation unsupported."); ret_msg->result = -ENOTSUP; - goto finish; + return; } if (ipcp_get_state() != IPCP_OPERATIONAL) { - log_err("IPCP in wrong state."); + log_err("Failed to join: IPCP in state <%s>, need <%s>.", + ipcp_state_str[ipcp_get_state()], + ipcp_state_str[IPCP_OPERATIONAL]); ret_msg->result = -EIPCPSTATE; - goto finish; + return; } fd = np1_flow_alloc(pid, flow_id); if (fd < 0) { log_err("Failed allocating n + 1 fd on flow_id %d.", flow_id); ret_msg->result = -1; - goto finish; + return; } - ret_msg->result = ipcpi.ops->ipcp_flow_join(fd, dst, qs); - finish: + if (uid != 0) { + pool = ssm_pool_open(uid); + if (pool == NULL) { + log_err("Failed to open PUP for uid %d.", uid); + ret_msg->result = -ENOMEM; + return; + } + } + + NP1_SET_POOL(fd, pool); + + ret_msg->result = ipcpd.ops->ipcp_flow_join(fd, dst); + log_info("Finished joining layer " HASH_FMT32 ".", HASH_VAL32(dst)); } static void do_flow_alloc_resp(int resp, int flow_id, + uid_t uid, const buffer_t * data, ipcp_msg_t * ret_msg) { - int fd = -1; + int fd = -1; + struct ssm_pool * pool = NULL; log_info("Responding %d to alloc on flow_id %d.", resp, flow_id); - if (ipcpi.ops->ipcp_flow_alloc_resp == NULL) { - log_err("Flow_alloc_resp unsupported."); + if (ipcpd.ops->ipcp_flow_alloc_resp == NULL) { + log_err("Failed to respond on flow %d: operation unsupported.", + flow_id); ret_msg->result = -ENOTSUP; - goto finish; + return; } if (ipcp_get_state() != IPCP_OPERATIONAL) { - log_err("IPCP in wrong state."); + log_err("Failed to respond to flow %d:" + "IPCP in state <%s>, need <%s>.", + flow_id, + ipcp_state_str[ipcp_get_state()], + ipcp_state_str[IPCP_OPERATIONAL]); ret_msg->result = -EIPCPSTATE; - goto finish; + return; } - if (resp == 0) { - fd = np1_flow_resp(flow_id); - if (fd < 0) { - log_warn("Flow_id %d is not known.", flow_id); - ret_msg->result = -1; - goto finish; + fd = np1_flow_resp(flow_id, resp); + if (fd < 0) { + log_warn("Flow_id %d is not known.", flow_id); + ret_msg->result = -1; + return; + } + + if (uid != 0) { + pool = ssm_pool_open(uid); + if (pool == NULL) { + log_err("Failed to open PUP for uid %d.", uid); + ret_msg->result = -ENOMEM; + return; } } - ret_msg->result = ipcpi.ops->ipcp_flow_alloc_resp(fd, resp, data); - finish: - log_info("Finished responding to allocation request: %d", + NP1_SET_POOL(fd, pool); + + ret_msg->result = ipcpd.ops->ipcp_flow_alloc_resp(fd, resp, data); + + log_info("Finished responding %d to allocation request.", ret_msg->result); } @@ -612,55 +794,83 @@ static void do_flow_dealloc(int flow_id, log_info("Deallocating flow %d.", flow_id); - if (ipcpi.ops->ipcp_flow_dealloc == NULL) { - log_err("Flow deallocation unsupported."); + if (ipcpd.ops->ipcp_flow_dealloc == NULL) { + log_err("Failed to dealloc: operation unsupported."); ret_msg->result = -ENOTSUP; - goto finish; + return; } if (ipcp_get_state() != IPCP_OPERATIONAL) { - log_err("IPCP in wrong state."); + log_err("Failed to enroll: IPCP in state <%s>, need <%s>.", + ipcp_state_str[ipcp_get_state()], + ipcp_state_str[IPCP_OPERATIONAL]); ret_msg->result = -EIPCPSTATE; - goto finish; + return; } fd = np1_flow_dealloc(flow_id, timeo_sec); if (fd < 0) { log_warn("Could not deallocate flow_id %d.", flow_id); ret_msg->result = -1; - goto finish; + return; } - ret_msg->result = ipcpi.ops->ipcp_flow_dealloc(fd); - finish: - log_info("Finished deallocating flow %d: %d.", - flow_id, ret_msg->result); + ret_msg->result = ipcpd.ops->ipcp_flow_dealloc(fd); + + log_info("Finished deallocating flow %d.", flow_id); +} + +static void do_flow_update(int flow_id, + const buffer_t * data, + ipcp_msg_t * ret_msg) +{ + int fd; + + if (ipcpd.ops->ipcp_flow_update == NULL) { + log_err("Failed to update flow: operation unsupported."); + ret_msg->result = -ENOTSUP; + return; + } + + if (ipcp_get_state() != IPCP_OPERATIONAL) { + ret_msg->result = -EIPCPSTATE; + return; + } + + fd = np1_flow_fd(flow_id); + if (fd < 0) { + log_warn("Flow update for unknown flow_id %d.", flow_id); + ret_msg->result = -1; + return; + } + + ret_msg->result = ipcpd.ops->ipcp_flow_update(fd, data); } static void * mainloop(void * o) { - int sfd; - buffer_t buffer; - ipcp_msg_t * msg; + int sfd; + buffer_t buffer; + ipcp_msg_t * msg; (void) o; while (true) { - ipcp_msg_t ret_msg = IPCP_MSG__INIT; - qosspec_t qs; - struct cmd * cmd; - buffer_t data; + ipcp_msg_t ret_msg = IPCP_MSG__INIT; + qosspec_t qs; + struct cmd * cmd; + buffer_t data; ret_msg.code = IPCP_MSG_CODE__IPCP_REPLY; - pthread_mutex_lock(&ipcpi.cmd_lock); + pthread_mutex_lock(&ipcpd.cmd_lock); - pthread_cleanup_push(__cleanup_mutex_unlock, &ipcpi.cmd_lock); + pthread_cleanup_push(__cleanup_mutex_unlock, &ipcpd.cmd_lock); - while (list_is_empty(&ipcpi.cmds)) - pthread_cond_wait(&ipcpi.cmd_cond, &ipcpi.cmd_lock); + while (list_is_empty(&ipcpd.cmds)) + pthread_cond_wait(&ipcpd.cmd_cond, &ipcpd.cmd_lock); - cmd = list_last_entry(&ipcpi.cmds, struct cmd, next); + cmd = list_last_entry(&ipcpd.cmds, struct cmd, next); list_del(&cmd->next); pthread_cleanup_pop(true); @@ -675,7 +885,7 @@ static void * mainloop(void * o) continue; } - tpm_dec(ipcpi.tpm); + tpm_begin_work(ipcpd.tpm); pthread_cleanup_push(__cleanup_close_ptr, &sfd); pthread_cleanup_push(free_msg, msg); @@ -717,13 +927,12 @@ static void * mainloop(void * o) qs = qos_spec_msg_to_s(msg->qosspec); do_flow_alloc(msg->pid, msg->flow_id, msg->hash.data, qs, - &data, &ret_msg); + &data, msg->uid, &ret_msg); break; case IPCP_MSG_CODE__IPCP_FLOW_JOIN: assert(msg->hash.len == ipcp_dir_hash_len()); - qs = qos_spec_msg_to_s(msg->qosspec); do_flow_join(msg->pid, msg->flow_id, - msg->hash.data, qs, &ret_msg); + msg->hash.data, msg->uid, &ret_msg); break; case IPCP_MSG_CODE__IPCP_FLOW_ALLOC_RESP: assert(msg->pk.len > 0 ? msg->pk.data != NULL @@ -731,11 +940,18 @@ static void * mainloop(void * o) data.len = msg->pk.len; data.data = msg->pk.data; do_flow_alloc_resp(msg->response, msg->flow_id, - &data, &ret_msg); + msg->uid, &data, &ret_msg); break; case IPCP_MSG_CODE__IPCP_FLOW_DEALLOC: do_flow_dealloc(msg->flow_id, msg->timeo_sec, &ret_msg); break; + case IPCP_MSG_CODE__IPCP_FLOW_UPDATE: + assert(msg->pk.len > 0 ? msg->pk.data != NULL + : msg->pk.data == NULL); + data.len = msg->pk.len; + data.data = msg->pk.data; + do_flow_update(msg->flow_id, &data, &ret_msg); + break; default: ret_msg.result = -1; log_err("Unknown message code: %d.", msg->code); @@ -749,7 +965,7 @@ static void * mainloop(void * o) if (buffer.len == 0) { log_err("Failed to pack reply message"); close(sfd); - tpm_inc(ipcpi.tpm); + tpm_end_work(ipcpd.tpm); continue; } @@ -757,7 +973,7 @@ static void * mainloop(void * o) if (buffer.data == NULL) { log_err("Failed to create reply buffer."); close(sfd); - tpm_inc(ipcpi.tpm); + tpm_end_work(ipcpd.tpm); continue; } @@ -766,16 +982,16 @@ static void * mainloop(void * o) if (ret_msg.layer_info != NULL) layer_info_msg__free_unpacked(ret_msg.layer_info, NULL); - pthread_cleanup_push(__cleanup_close_ptr, &sfd); pthread_cleanup_push(free, buffer.data) + pthread_cleanup_push(__cleanup_close_ptr, &sfd); if (write(sfd, buffer.data, buffer.len) == -1) log_warn("Failed to send reply message"); - pthread_cleanup_pop(true); - pthread_cleanup_pop(true); + pthread_cleanup_pop(true); /* close sfd */ + pthread_cleanup_pop(true); /* free buffer.data */ - tpm_inc(ipcpi.tpm); + tpm_end_work(ipcpd.tpm); } return (void *) 0; @@ -794,10 +1010,10 @@ static int parse_args(int argc, if (atoi(argv[1]) == 0) return -1; - ipcpi.irmd_pid = atoi(argv[1]); + ipcpd.irmd_pid = atoi(argv[1]); /* argument 2: IPCP name */ - ipcpi.name = argv[2]; + ipcpd.name = argv[2]; /* argument 3: syslog */ if (argv[3] != NULL) @@ -813,71 +1029,69 @@ int ipcp_init(int argc, { bool log; pthread_condattr_t cattr; - int ret = -1; if (parse_args(argc, argv, &log)) return -1; log_init(log); - ipcpi.irmd_fd = -1; - ipcpi.state = IPCP_NULL; - ipcpi.type = type; + ipcpd.type = type; #if defined (__linux__) prctl(PR_SET_TIMERSLACK, IPCP_LINUX_SLACK_NS, 0, 0, 0); #endif - ipcpi.sock_path = ipcp_sock_path(getpid()); - if (ipcpi.sock_path == NULL) + ipcpd.sock_path = sock_path(getpid(), IPCP_SOCK_PATH_PREFIX); + if (ipcpd.sock_path == NULL) goto fail_sock_path; - ipcpi.sockfd = server_socket_open(ipcpi.sock_path); - if (ipcpi.sockfd < 0) { - log_err("Could not open server socket."); + ipcpd.sockfd = server_socket_open(ipcpd.sock_path); + if (ipcpd.sockfd < 0) { + log_err("Failed to open server socket at %s.", + ipcpd.sock_path); goto fail_serv_sock; } - ipcpi.ops = ops; + ipcpd.ops = ops; - if (pthread_mutex_init(&ipcpi.state_mtx, NULL)) { - log_err("Could not create mutex."); + if (pthread_mutex_init(&ipcpd.state_mtx, NULL)) { + log_err("Failed to create mutex."); goto fail_state_mtx; } if (pthread_condattr_init(&cattr)) { - log_err("Could not create condattr."); + log_err("Failed to create condattr."); goto fail_cond_attr; } #ifndef __APPLE__ pthread_condattr_setclock(&cattr, PTHREAD_COND_CLOCK); #endif - if (pthread_cond_init(&ipcpi.state_cond, &cattr)) { - log_err("Could not init condvar."); + if (pthread_cond_init(&ipcpd.state_cond, &cattr)) { + log_err("Failed to init condvar."); goto fail_state_cond; } - if (pthread_mutex_init(&ipcpi.alloc_lock, NULL)) { + if (pthread_mutex_init(&ipcpd.alloc_lock, NULL)) { log_err("Failed to init mutex."); goto fail_alloc_lock; } - if (pthread_cond_init(&ipcpi.alloc_cond, &cattr)) { + if (pthread_cond_init(&ipcpd.alloc_cond, &cattr)) { log_err("Failed to init convar."); goto fail_alloc_cond; } - if (pthread_mutex_init(&ipcpi.cmd_lock, NULL)) { + if (pthread_mutex_init(&ipcpd.cmd_lock, NULL)) { log_err("Failed to init mutex."); goto fail_cmd_lock; } - if (pthread_cond_init(&ipcpi.cmd_cond, &cattr)) { + if (pthread_cond_init(&ipcpd.cmd_cond, &cattr)) { log_err("Failed to init convar."); goto fail_cmd_cond; } - if (rib_init(ipcpi.name)) { + if (rib_init(ipcpd.name)) { log_err("Failed to initialize RIB."); goto fail_rib_init; } @@ -887,21 +1101,26 @@ int ipcp_init(int argc, goto fail_rib_reg; } - ipcpi.tpm = tpm_create(IPCP_MIN_THREADS, IPCP_ADD_THREADS, + list_head_init(&ipcpd.cmds); + + ipcpd.tpm = tpm_create(IPCP_MIN_THREADS, IPCP_ADD_THREADS, mainloop, NULL); - if (ipcpi.tpm == NULL) { + if (ipcpd.tpm == NULL) { log_err("Failed to create threadpool manager."); goto fail_tpm_create; } - list_head_init(&ipcpi.cmds); + ipcpd.alloc_id = -1; - ipcpi.alloc_id = -1; + memset(&np1, 0, sizeof(np1)); pthread_condattr_destroy(&cattr); ipcp_set_state(IPCP_INIT); + log_info("IPCP %s %d initialized.", ipcp_type_str[ipcpd.type], + getpid()); + return 0; fail_tpm_create: @@ -909,25 +1128,25 @@ int ipcp_init(int argc, fail_rib_reg: rib_fini(); fail_rib_init: - pthread_cond_destroy(&ipcpi.cmd_cond); + pthread_cond_destroy(&ipcpd.cmd_cond); fail_cmd_cond: - pthread_mutex_destroy(&ipcpi.cmd_lock); + pthread_mutex_destroy(&ipcpd.cmd_lock); fail_cmd_lock: - pthread_cond_destroy(&ipcpi.alloc_cond); + pthread_cond_destroy(&ipcpd.alloc_cond); fail_alloc_cond: - pthread_mutex_destroy(&ipcpi.alloc_lock); + pthread_mutex_destroy(&ipcpd.alloc_lock); fail_alloc_lock: - pthread_cond_destroy(&ipcpi.state_cond); + pthread_cond_destroy(&ipcpd.state_cond); fail_state_cond: pthread_condattr_destroy(&cattr); fail_cond_attr: - pthread_mutex_destroy(&ipcpi.state_mtx); + pthread_mutex_destroy(&ipcpd.state_mtx); fail_state_mtx: - close(ipcpi.sockfd); + close(ipcpd.sockfd); fail_serv_sock: - free(ipcpi.sock_path); + free(ipcpd.sock_path); fail_sock_path: - return ret; + return -1; } int ipcp_start(void) @@ -944,20 +1163,22 @@ int ipcp_start(void) pthread_sigmask(SIG_BLOCK, &sigset, NULL); info.pid = getpid(); - info.type = ipcpi.type; - strcpy(info.name, ipcpi.name); - info.state = IPCP_OPERATIONAL; + info.type = ipcpd.type; + strcpy(info.name, ipcpd.name); + info.state = IPCP_BOOT; + + ipcp_set_state(IPCP_BOOT); - if (tpm_start(ipcpi.tpm)) + if (tpm_start(ipcpd.tpm)) { + log_err("Failed to start threadpool manager."); goto fail_tpm_start; + } - if (pthread_create(&ipcpi.acceptor, NULL, acceptloop, NULL)) { + if (pthread_create(&ipcpd.acceptor, NULL, acceptloop, NULL)) { log_err("Failed to create acceptor thread."); goto fail_acceptor; } - info.state = IPCP_OPERATIONAL; - if (ipcp_create_r(&info)) { log_err("Failed to notify IRMd we are initialized."); goto fail_create_r; @@ -966,14 +1187,13 @@ int ipcp_start(void) return 0; fail_create_r: - pthread_cancel(ipcpi.acceptor); - pthread_join(ipcpi.acceptor, NULL); + pthread_cancel(ipcpd.acceptor); + pthread_join(ipcpd.acceptor, NULL); fail_acceptor: - tpm_stop(ipcpi.tpm); + tpm_stop(ipcpd.tpm); fail_tpm_start: - tpm_destroy(ipcpi.tpm); - ipcp_set_state(IPCP_NULL); - info.state = IPCP_NULL; + tpm_destroy(ipcpd.tpm); + ipcp_set_state(IPCP_INIT); ipcp_create_r(&info); return -1; } @@ -993,7 +1213,7 @@ void ipcp_sigwait(void) sigaddset(&sigset, SIGTERM); sigaddset(&sigset, SIGPIPE); - while(ipcp_get_state() != IPCP_NULL && + while(ipcp_get_state() != IPCP_INIT && ipcp_get_state() != IPCP_SHUTDOWN) { #ifdef __APPLE__ if (sigwait(&sigset, &sig) < 0) { @@ -1007,7 +1227,7 @@ void ipcp_sigwait(void) #ifdef __APPLE__ memset(&info, 0, sizeof(info)); info.si_signo = sig; - info.si_pid = ipcpi.irmd_pid; + info.si_pid = ipcpd.irmd_pid; #endif switch(info.si_signo) { case SIGINT: @@ -1017,9 +1237,9 @@ void ipcp_sigwait(void) case SIGHUP: /* FALLTHRU */ case SIGQUIT: - if (info.si_pid == ipcpi.irmd_pid) { - if (ipcp_get_state() == IPCP_INIT) - ipcp_set_state(IPCP_NULL); + if (info.si_pid == ipcpd.irmd_pid) { + if (ipcp_get_state() == IPCP_BOOT) + ipcp_set_state(IPCP_INIT); if (ipcp_get_state() == IPCP_OPERATIONAL) ipcp_set_state(IPCP_SHUTDOWN); @@ -1038,58 +1258,62 @@ void ipcp_stop(void) { log_info("IPCP %d shutting down.", getpid()); - pthread_cancel(ipcpi.acceptor); - pthread_join(ipcpi.acceptor, NULL); + pthread_cancel(ipcpd.acceptor); + pthread_join(ipcpd.acceptor, NULL); + + tpm_stop(ipcpd.tpm); - tpm_stop(ipcpi.tpm); + ipcp_set_state(IPCP_INIT); } void ipcp_fini(void) { - tpm_destroy(ipcpi.tpm); + tpm_destroy(ipcpd.tpm); rib_unreg(IPCP_INFO); rib_fini(); - close(ipcpi.sockfd); - if (unlink(ipcpi.sock_path)) - log_warn("Could not unlink %s.", ipcpi.sock_path); + close(ipcpd.sockfd); + if (unlink(ipcpd.sock_path)) + log_warn("Could not unlink %s.", ipcpd.sock_path); - free(ipcpi.sock_path); + free(ipcpd.sock_path); - pthread_cond_destroy(&ipcpi.state_cond); - pthread_mutex_destroy(&ipcpi.state_mtx); - pthread_cond_destroy(&ipcpi.alloc_cond); - pthread_mutex_destroy(&ipcpi.alloc_lock); - pthread_cond_destroy(&ipcpi.cmd_cond); - pthread_mutex_destroy(&ipcpi.cmd_lock); + pthread_cond_destroy(&ipcpd.state_cond); + pthread_mutex_destroy(&ipcpd.state_mtx); + pthread_cond_destroy(&ipcpd.alloc_cond); + pthread_mutex_destroy(&ipcpd.alloc_lock); + pthread_cond_destroy(&ipcpd.cmd_cond); + pthread_mutex_destroy(&ipcpd.cmd_lock); log_info("IPCP %d out.", getpid()); log_fini(); + + ipcpd.state = IPCP_NULL; } void ipcp_set_state(enum ipcp_state state) { - pthread_mutex_lock(&ipcpi.state_mtx); + pthread_mutex_lock(&ipcpd.state_mtx); - ipcpi.state = state; + ipcpd.state = state; - pthread_cond_broadcast(&ipcpi.state_cond); - pthread_mutex_unlock(&ipcpi.state_mtx); + pthread_cond_broadcast(&ipcpd.state_cond); + pthread_mutex_unlock(&ipcpd.state_mtx); } enum ipcp_state ipcp_get_state(void) { enum ipcp_state state; - pthread_mutex_lock(&ipcpi.state_mtx); + pthread_mutex_lock(&ipcpd.state_mtx); - state = ipcpi.state; + state = ipcpd.state; - pthread_mutex_unlock(&ipcpi.state_mtx); + pthread_mutex_unlock(&ipcpd.state_mtx); return state; } diff --git a/src/ipcpd/ipcp.h b/src/ipcpd/ipcp.h index aab490c7..210157ec 100644 --- a/src/ipcpd/ipcp.h +++ b/src/ipcpd/ipcp.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * IPC process structure * @@ -34,8 +34,10 @@ #include <pthread.h> #include <time.h> +#define ipcp_dir_hash_strlen() (ipcp_dir_hash_len() * 2) + struct ipcp_ops { - int (* ipcp_bootstrap)(const struct ipcp_config * conf); + int (* ipcp_bootstrap)(struct ipcp_config * conf); int (* ipcp_enroll)(const char * dst, struct layer_info * info); @@ -59,52 +61,17 @@ struct ipcp_ops { const buffer_t * data); int (* ipcp_flow_join)(int fd, - const uint8_t * dst, - qosspec_t qs); + const uint8_t * dst); int (* ipcp_flow_alloc_resp)(int fd, int response, const buffer_t * data); int (* ipcp_flow_dealloc)(int fd); -}; - -#define ipcp_dir_hash_strlen() (hash_len(ipcpi.dir_hash_algo) * 2) -#define ipcp_dir_hash_len() (hash_len(ipcpi.dir_hash_algo)) - -extern struct ipcp { - pid_t irmd_pid; - char * name; - - enum ipcp_type type; - char layer_name[LAYER_NAME_SIZE + 1]; - - uint64_t dt_addr; - - enum hash_algo dir_hash_algo; - - struct ipcp_ops * ops; - int irmd_fd; - - enum ipcp_state state; - pthread_cond_t state_cond; - pthread_mutex_t state_mtx; - - int sockfd; - char * sock_path; - - struct list_head cmds; - pthread_cond_t cmd_cond; - pthread_mutex_t cmd_lock; - int alloc_id; - pthread_cond_t alloc_cond; - pthread_mutex_t alloc_lock; - - struct tpm * tpm; - - pthread_t acceptor; -} ipcpi; + int (* ipcp_flow_update)(int fd, + const buffer_t * data); +}; int ipcp_init(int argc, char ** argv, @@ -119,22 +86,32 @@ void ipcp_stop(void); void ipcp_fini(void); +enum ipcp_type ipcp_get_type(void); + +const char * ipcp_get_name(void); + +/* TODO: Only specify hash algorithm in directory policy */ +void ipcp_set_dir_hash_algo(enum hash_algo algo); + void ipcp_set_state(enum ipcp_state state); enum ipcp_state ipcp_get_state(void); -int ipcp_parse_arg(int argc, - char * argv[]); - /* Helper functions to handle races during flow allocation */ int ipcp_wait_flow_req_arr(const uint8_t * dst, qosspec_t qs, time_t mpl, + uint32_t mtu, const buffer_t * data); int ipcp_wait_flow_resp(const int fd); + /* Helper functions for directory entries, could be moved */ +size_t ipcp_dir_hash_len(void); + +int ipcp_get_layer_name(char * layer); + uint8_t * ipcp_hash_dup(const uint8_t * hash); void ipcp_hash_str(char buf[], diff --git a/src/ipcpd/local/CMakeLists.txt b/src/ipcpd/local/CMakeLists.txt index 10fd0120..91f300a3 100644 --- a/src/ipcpd/local/CMakeLists.txt +++ b/src/ipcpd/local/CMakeLists.txt @@ -1,34 +1,17 @@ -get_filename_component(CURRENT_SOURCE_PARENT_DIR - ${CMAKE_CURRENT_SOURCE_DIR} DIRECTORY) -get_filename_component(CURRENT_BINARY_PARENT_DIR - ${CMAKE_CURRENT_BINARY_DIR} DIRECTORY) +# Local IPCP build configuration -include_directories(${CMAKE_CURRENT_SOURCE_DIR}) -include_directories(${CMAKE_CURRENT_BINARY_DIR}) +add_executable(${IPCP_LOCAL_TARGET} + main.c + ${IPCP_SOURCES} +) -include_directories(${CURRENT_SOURCE_PARENT_DIR}) -include_directories(${CURRENT_BINARY_PARENT_DIR}) +target_include_directories(${IPCP_LOCAL_TARGET} PRIVATE ${IPCP_INCLUDE_DIRS}) +target_link_libraries(${IPCP_LOCAL_TARGET} PRIVATE ouroboros-dev) -include_directories(${CMAKE_SOURCE_DIR}/include) -include_directories(${CMAKE_BINARY_DIR}/include) +ouroboros_target_debug_definitions(${IPCP_LOCAL_TARGET}) -set(IPCP_LOCAL_TARGET ipcpd-local CACHE INTERNAL "") -set(IPCP_LOCAL_MPL 2 CACHE STRING - "Default maximum packet lifetime for the Ethernet IPCPs, in seconds") +if(IPCP_LOCAL_POLLING) + target_compile_definitions(${IPCP_LOCAL_TARGET} PRIVATE CONFIG_IPCP_LOCAL_POLLING) +endif() -set(LOCAL_SOURCES - # Add source files here - ${CMAKE_CURRENT_SOURCE_DIR}/main.c) - -add_executable(ipcpd-local ${LOCAL_SOURCES} ${IPCP_SOURCES}) -target_link_libraries(ipcpd-local LINK_PUBLIC ouroboros-common ouroboros-dev) - -include(AddCompileFlags) -if (CMAKE_BUILD_TYPE MATCHES "Debug*") - add_compile_flags(ipcpd-local -DCONFIG_OUROBOROS_DEBUG) -endif () - -install(TARGETS ipcpd-local RUNTIME DESTINATION ${CMAKE_INSTALL_SBINDIR}) - -# Enable once ipcp-local has tests -# add_subdirectory(tests) +install(TARGETS ${IPCP_LOCAL_TARGET} RUNTIME DESTINATION ${CMAKE_INSTALL_SBINDIR}) diff --git a/src/ipcpd/local/main.c b/src/ipcpd/local/main.c index 160e07e0..c0aeb51e 100644 --- a/src/ipcpd/local/main.c +++ b/src/ipcpd/local/main.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Local IPC process * @@ -38,8 +38,10 @@ #include <ouroboros/ipcp.h> #include <ouroboros/ipcp-dev.h> #include <ouroboros/local-dev.h> +#include <ouroboros/np1_flow.h> #include "ipcp.h" +#include "np1.h" #include "shim-data.h" #include <string.h> @@ -50,8 +52,6 @@ #define THIS_TYPE IPCP_LOCAL -struct ipcp ipcpi; - struct { struct shim_data * shim_data; @@ -105,47 +105,54 @@ static void local_data_fini(void){ static void * local_ipcp_packet_loop(void * o) { + int src_fd; + int dst_fd; + struct timespec * timeout; +#ifdef CONFIG_IPCP_LOCAL_POLLING + struct timespec ts_poll = {0, 0}; +#endif (void) o; ipcp_lock_to_core(); - while (true) { - int fd; - ssize_t idx; +#ifdef CONFIG_IPCP_LOCAL_POLLING + timeout = &ts_poll; /* Spin poll with zero timeout */ +#else + timeout = NULL; /* Block until event */ +#endif - fevent(local_data.flows, local_data.fq, NULL); + while (true) { + fevent(local_data.flows, local_data.fq, timeout); - while ((fd = fqueue_next(local_data.fq)) >= 0) { + while ((src_fd = fqueue_next(local_data.fq)) >= 0) { if (fqueue_type(local_data.fq) != FLOW_PKT) continue; - idx = local_flow_read(fd); - if (idx < 0) - continue; - - assert(idx < (SHM_BUFFER_SIZE)); - pthread_rwlock_rdlock(&local_data.lock); - fd = local_data.in_out[fd]; + dst_fd = local_data.in_out[src_fd]; pthread_rwlock_unlock(&local_data.lock); - if (fd != -1) - local_flow_write(fd, idx); + if (dst_fd == -1) + continue; + + local_flow_transfer(src_fd, dst_fd, + NP1_GET_POOL(src_fd), + NP1_GET_POOL(dst_fd)); } } return (void *) 0; } -static int local_ipcp_bootstrap(const struct ipcp_config * conf) +static int local_ipcp_bootstrap(struct ipcp_config * conf) { + assert(conf); assert(conf->type == THIS_TYPE); - ipcpi.dir_hash_algo = (enum hash_algo) conf->layer_info.dir_hash_algo; - strcpy(ipcpi.layer_name,conf->layer_info.name); + (void) conf; if (pthread_create(&local_data.packet_loop, NULL, local_ipcp_packet_loop, NULL)) { @@ -197,7 +204,8 @@ static int local_ipcp_flow_alloc(int fd, HASH_VAL32(dst), fd); assert(dst); - out_fd = ipcp_wait_flow_req_arr(dst, qs, IPCP_LOCAL_MPL, data); + out_fd = ipcp_wait_flow_req_arr(dst, qs, IPCP_LOCAL_MPL, + IPCP_LOCAL_MTU, data); if (out_fd < 0) { log_dbg("Flow allocation failed: %d", out_fd); return -1; @@ -230,15 +238,6 @@ static int local_ipcp_flow_alloc_resp(int fd, return -1; } - if (response < 0) { - pthread_rwlock_wrlock(&local_data.lock); - if (local_data.in_out[fd] != -1) - local_data.in_out[local_data.in_out[fd]] = fd; - local_data.in_out[fd] = -1; - pthread_rwlock_unlock(&local_data.lock); - return 0; - } - pthread_rwlock_rdlock(&local_data.lock); out_fd = local_data.in_out[fd]; @@ -257,9 +256,17 @@ static int local_ipcp_flow_alloc_resp(int fd, return -1; } + if (response < 0) { + ipcp_flow_alloc_reply(out_fd, response, mpl, + IPCP_LOCAL_MTU, data); + log_info("Flow allocation rejected, fds (%d, %d).", out_fd, fd); + return 0; + } + fset_add(local_data.flows, fd); - if (ipcp_flow_alloc_reply(out_fd, response, mpl, data) < 0) { + if (ipcp_flow_alloc_reply(out_fd, response, mpl, + IPCP_LOCAL_MTU, data) < 0) { log_err("Failed to reply to allocation"); fset_del(local_data.flows, fd); return -1; @@ -291,6 +298,38 @@ static int local_ipcp_flow_dealloc(int fd) return 0; } +/* Loopback relay: deliver the update back to the peer end (same IRMd). */ +static int local_ipcp_flow_update(int fd, + const buffer_t * data) +{ + int out_fd; + int out_flow_id; + + pthread_rwlock_rdlock(&local_data.lock); + + out_fd = local_data.in_out[fd]; + + pthread_rwlock_unlock(&local_data.lock); + + if (out_fd == -1) { + log_err("Flow update on fd %d with no peer.", fd); + return -1; + } + + out_flow_id = np1_flow_id(out_fd); + if (out_flow_id < 0) { + log_err("No flow_id for peer fd %d.", out_fd); + return -1; + } + + if (ipcp_flow_update_arr(out_flow_id, data) < 0) { + log_err("Failed to relay flow update to fd %d.", out_fd); + return -1; + } + + return 0; +} + static struct ipcp_ops local_ops = { .ipcp_bootstrap = local_ipcp_bootstrap, .ipcp_enroll = NULL, @@ -302,7 +341,8 @@ static struct ipcp_ops local_ops = { .ipcp_flow_alloc = local_ipcp_flow_alloc, .ipcp_flow_join = NULL, .ipcp_flow_alloc_resp = local_ipcp_flow_alloc_resp, - .ipcp_flow_dealloc = local_ipcp_flow_dealloc + .ipcp_flow_dealloc = local_ipcp_flow_dealloc, + .ipcp_flow_update = local_ipcp_flow_update }; int main(int argc, diff --git a/src/ipcpd/np1.h b/src/ipcpd/np1.h new file mode 100644 index 00000000..b7792cb9 --- /dev/null +++ b/src/ipcpd/np1.h @@ -0,0 +1,41 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * N+1 flow pool tracking for IPCPs + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#ifndef OUROBOROS_IPCPD_NP1_H +#define OUROBOROS_IPCPD_NP1_H + +#include "config.h" + +#include <ouroboros/ssm_pool.h> + +#define NP1_LOAD(ptr) (__atomic_load_n((ptr), __ATOMIC_ACQUIRE)) +#define NP1_STORE(ptr, v) (__atomic_store_n((ptr), (v), __ATOMIC_RELEASE)) +#define NP1_GET_POOL(fd) (NP1_LOAD(&np1.pool[(fd)])) +#define NP1_SET_POOL(fd, p) (NP1_STORE(&np1.pool[(fd)], (p))) + +struct np1_state { + struct ssm_pool * pool[SYS_MAX_FLOWS]; +}; + +extern struct np1_state np1; + +#endif /* OUROBOROS_IPCPD_NP1_H */ diff --git a/src/ipcpd/shim-data.c b/src/ipcpd/shim-data.c index 1fac63ac..90a676da 100644 --- a/src/ipcpd/shim-data.c +++ b/src/ipcpd/shim-data.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * IPC process utilities * @@ -51,7 +51,7 @@ struct reg_entry { struct dir_entry { struct list_head list; uint8_t * hash; - uint64_t addr; + struct addr addr; }; static void destroy_dir_query(struct dir_query * query) @@ -108,14 +108,12 @@ static void reg_entry_destroy(struct reg_entry * entry) { assert(entry); - if (entry->hash != NULL) - free(entry->hash); - + free(entry->hash); free(entry); } -static struct dir_entry * dir_entry_create(uint8_t * hash, - uint64_t addr) +static struct dir_entry * dir_entry_create(uint8_t * hash, + struct addr addr) { struct dir_entry * entry = malloc(sizeof(*entry)); if (entry == NULL) @@ -133,9 +131,7 @@ static void dir_entry_destroy(struct dir_entry * entry) { assert(entry); - if (entry->hash != NULL) - free(entry->hash); - + free(entry->hash); free(entry); } @@ -258,13 +254,15 @@ static struct reg_entry * find_reg_entry_by_hash(struct shim_data * data, static struct dir_entry * find_dir_entry(struct shim_data * data, const uint8_t * hash, - uint64_t addr) + struct addr addr) { struct list_head * h; list_for_each(h, &data->directory) { struct dir_entry * e = list_entry(h, struct dir_entry, list); - if (e->addr == addr && - !memcmp(e->hash, hash, ipcp_dir_hash_len())) + if (memcmp(&e->addr, &addr, sizeof(addr)) != 0) + continue; + + if (memcmp(e->hash, hash, ipcp_dir_hash_len()) == 0) return e; } @@ -364,7 +362,7 @@ bool shim_data_reg_has(struct shim_data * data, int shim_data_dir_add_entry(struct shim_data * data, const uint8_t * hash, - uint64_t addr) + struct addr addr) { struct dir_entry * entry; uint8_t * entry_hash; @@ -400,7 +398,7 @@ int shim_data_dir_add_entry(struct shim_data * data, int shim_data_dir_del_entry(struct shim_data * data, const uint8_t * hash, - uint64_t addr) + struct addr addr) { struct dir_entry * e; if (data == NULL) @@ -437,11 +435,11 @@ bool shim_data_dir_has(struct shim_data * data, return ret; } -uint64_t shim_data_dir_get_addr(struct shim_data * data, - const uint8_t * hash) +struct addr shim_data_dir_get_addr(struct shim_data * data, + const uint8_t * hash) { struct dir_entry * entry; - uint64_t addr; + struct addr addr = {0}; pthread_rwlock_rdlock(&data->dir_lock); @@ -449,7 +447,7 @@ uint64_t shim_data_dir_get_addr(struct shim_data * data, if (entry == NULL) { pthread_rwlock_unlock(&data->dir_lock); log_warn("No address for " HASH_FMT32 ".", HASH_VAL32(hash)); - return 0; /* undefined behaviour, 0 may be a valid address */ + return addr; /* undefined behaviour, 0 may be a valid address */ } addr = entry->addr; diff --git a/src/ipcpd/shim-data.h b/src/ipcpd/shim-data.h index 372b4ea7..fbadb4d4 100644 --- a/src/ipcpd/shim-data.h +++ b/src/ipcpd/shim-data.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Utitilies for building IPC processes * @@ -25,9 +25,12 @@ #include <ouroboros/list.h> -#include <sys/types.h> #include <pthread.h> #include <stdint.h> +#include <netinet/in.h> +#include <sys/types.h> + +#define MAC_SIZE 6 enum dir_query_state { QUERY_INIT = 0, @@ -46,6 +49,14 @@ struct dir_query { pthread_cond_t cond; }; +struct addr { + union { + uint8_t mac[MAC_SIZE]; + struct in_addr ip4; + struct in6_addr ip6; + }; +}; + struct shim_data { struct list_head registry; pthread_rwlock_t reg_lock; @@ -72,16 +83,16 @@ bool shim_data_reg_has(struct shim_data * data, int shim_data_dir_add_entry(struct shim_data * data, const uint8_t * hash, - uint64_t addr); + struct addr addr); int shim_data_dir_del_entry(struct shim_data * data, const uint8_t * hash, - uint64_t addr); + struct addr addr); bool shim_data_dir_has(struct shim_data * data, const uint8_t * hash); -uint64_t shim_data_dir_get_addr(struct shim_data * data, +struct addr shim_data_dir_get_addr(struct shim_data * data, const uint8_t * hash); struct dir_query * shim_data_dir_query_create(struct shim_data * data, diff --git a/src/ipcpd/udp/CMakeLists.txt b/src/ipcpd/udp/CMakeLists.txt index 8ae5518e..a98f0919 100644 --- a/src/ipcpd/udp/CMakeLists.txt +++ b/src/ipcpd/udp/CMakeLists.txt @@ -1,69 +1,14 @@ -get_filename_component(CURRENT_SOURCE_PARENT_DIR - ${CMAKE_CURRENT_SOURCE_DIR} DIRECTORY) -get_filename_component(CURRENT_BINARY_PARENT_DIR - ${CMAKE_CURRENT_BINARY_DIR} DIRECTORY) +# UDP IPCPs build configuration (UDP4 and UDP6) +# DDNS detection is in cmake/dependencies/udp/ddns.cmake -include_directories(${CMAKE_CURRENT_SOURCE_DIR}) -include_directories(${CMAKE_CURRENT_BINARY_DIR}) +add_executable(${IPCP_UDP4_TARGET} udp4.c ${IPCP_SOURCES}) +add_executable(${IPCP_UDP6_TARGET} udp6.c ${IPCP_SOURCES}) -include_directories(${CURRENT_SOURCE_PARENT_DIR}) -include_directories(${CURRENT_BINARY_PARENT_DIR}) +foreach(target ${IPCP_UDP4_TARGET} ${IPCP_UDP6_TARGET}) + target_include_directories(${target} PRIVATE ${IPCP_INCLUDE_DIRS}) + target_link_libraries(${target} PRIVATE ouroboros-dev) + ouroboros_target_debug_definitions(${target}) +endforeach() -include_directories(${CMAKE_SOURCE_DIR}/include) -include_directories(${CMAKE_BINARY_DIR}/include) - -set(IPCP_UDP_TARGET ipcpd-udp CACHE INTERNAL "") - -set(UDP_SOURCES - # Add source files here - ${CMAKE_CURRENT_SOURCE_DIR}/main.c - ) - -add_executable(ipcpd-udp ${UDP_SOURCES} ${IPCP_SOURCES}) - -target_link_libraries(ipcpd-udp LINK_PUBLIC ouroboros-dev) - -# Find the nsupdate executable -find_program(NSUPDATE_EXECUTABLE - NAMES nsupdate - DOC "The nsupdate tool that enables DDNS") - -# Find the nslookup executable -find_program(NSLOOKUP_EXECUTABLE - NAMES nslookup - DOC "The nslookup tool that resolves DNS names") - -mark_as_advanced(NSLOOKUP_EXECUTABLE NSUPDATE_EXECUTABLE) - -if (NSLOOKUP_EXECUTABLE AND NSUPDATE_EXECUTABLE) - set(DISABLE_DDNS FALSE CACHE BOOL "Disable DDNS support") - if (NOT DISABLE_DNS) - message(STATUS "DDNS support enabled") - set(HAVE_DDNS TRUE CACHE INTERNAL "") - else () - message(STATUS "DDNS support disabled by user") - unset(HAVE_DDNS CACHE) - endif () -else () - if (NSLOOKUP_EXECUTABLE) - message(STATUS "Install nsupdate to enable DDNS support") - elseif (NSUPDATE_EXECUTABLE) - message(STATUS "Install nslookup to enable DDNS support") - else () - message(STATUS "Install nslookup and nsupdate to enable DDNS support") - endif () -endif () - -set(IPCP_UDP_RD_THR 3 CACHE STRING - "Number of reader threads in UDP IPCP") -set(IPCP_UDP_WR_THR 3 CACHE STRING - "Number of writer threads in UDP IPCP") -set(IPCP_UDP_MPL 60 CACHE STRING - "Default maximum packet lifetime for the UDP IPCP, in seconds") - -include(AddCompileFlags) -if (CMAKE_BUILD_TYPE MATCHES "Debug*") - add_compile_flags(ipcpd-udp -DCONFIG_OUROBOROS_DEBUG) -endif () - -install(TARGETS ipcpd-udp RUNTIME DESTINATION ${CMAKE_INSTALL_SBINDIR}) +install(TARGETS ${IPCP_UDP4_TARGET} ${IPCP_UDP6_TARGET} + RUNTIME DESTINATION ${CMAKE_INSTALL_SBINDIR}) diff --git a/src/ipcpd/udp/main.c b/src/ipcpd/udp/udp.c index 2e8d84ce..db57e2f4 100644 --- a/src/ipcpd/udp/main.c +++ b/src/ipcpd/udp/udp.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * IPC process over UDP * @@ -20,16 +20,8 @@ * Foundation, Inc., http://www.fsf.org/about/contact/. */ -#if defined(__linux__) || defined(__CYGWIN__) -#define _DEFAULT_SOURCE -#else -#define _POSIX_C_SOURCE 200112L -#endif - #include "config.h" -#define OUROBOROS_PREFIX "ipcpd/udp" - #include <ouroboros/bitmap.h> #include <ouroboros/endian.h> #include <ouroboros/hash.h> @@ -37,12 +29,14 @@ #include <ouroboros/utils.h> #include <ouroboros/dev.h> #include <ouroboros/ipcp-dev.h> +#include <ouroboros/np1_flow.h> #include <ouroboros/fqueue.h> #include <ouroboros/errno.h> #include <ouroboros/logs.h> #include <ouroboros/pthread.h> #include "ipcp.h" +#include "np1.h" #include "shim-data.h" #include <string.h> @@ -55,16 +49,20 @@ #include <stdlib.h> #include <sys/wait.h> #include <fcntl.h> +#include <unistd.h> +#if defined(__linux__) +#include <netinet/ip.h> +#endif #define FLOW_REQ 1 #define FLOW_REPLY 2 +#define FLOW_IRM_UPDATE 3 -#define THIS_TYPE IPCP_UDP -#define IPCP_UDP_MAX_PACKET_SIZE 8980 #define OUR_HEADER_LEN sizeof(uint32_t) /* adds eid */ -#define IPCP_UDP_BUF_SIZE 8980 -#define IPCP_UDP_MSG_SIZE 8980 +#define IPCP_UDP_BUF_SIZE IPCP_UDP_MAX_PACKET_SIZE +#define IPCP_UDP_MSG_SIZE IPCP_UDP_MAX_PACKET_SIZE + #define DNS_TTL 86400 #define SADDR ((struct sockaddr *) &udp_data.s_saddr) @@ -81,47 +79,42 @@ #define SENDTO_FLAGS 0 #endif -struct ipcp ipcpi; - /* Keep order for alignment. */ struct mgmt_msg { uint32_t eid; uint32_t s_eid; uint32_t d_eid; - uint8_t code; - int8_t response; - /* QoS parameters from spec */ - uint8_t availability; - uint8_t in_order; + int32_t response; uint64_t bandwidth; uint32_t delay; uint32_t loss; uint32_t ber; uint32_t max_gap; uint32_t timeout; - uint16_t cypher_s; - + uint8_t code; + /* QoS parameters from spec */ + uint8_t availability; + uint8_t service; } __attribute__((packed)); struct mgmt_frame { - struct list_head next; - struct sockaddr_in r_saddr; - uint8_t buf[MGMT_FRAME_BUF_SIZE]; - size_t len; + struct list_head next; + struct __SOCKADDR r_saddr; + uint8_t buf[MGMT_FRAME_BUF_SIZE]; + size_t len; }; /* UDP flow */ struct uf { - int d_eid; - struct sockaddr_in r_saddr; + int d_eid; + struct __SOCKADDR r_saddr; }; struct { struct shim_data * shim_data; - uint32_t dns_addr; - - struct sockaddr_in s_saddr; + struct __ADDR dns_addr; + struct __SOCKADDR s_saddr; int s_fd; fset_t * np1_flows; @@ -138,6 +131,59 @@ struct { struct list_head mgmt_frames; } udp_data; +static const char * __inet_ntop(const struct __ADDR * addr, + char * buf) +{ + return inet_ntop(__AF, addr, buf, __ADDRSTRLEN); +} + +#if defined(BUILD_IPCP_UDP4) +#define UDP_MTU_FALLBACK IPCP_UDP4_MTU +#define UDP_IP_OVERHEAD 28U /* IPv4 + UDP */ +#else +#define UDP_MTU_FALLBACK IPCP_UDP6_MTU +#define UDP_IP_OVERHEAD 48U /* IPv6 + UDP */ +#endif + +static uint32_t udp_query_mtu(const struct __SOCKADDR * saddr) +{ +#if defined(__linux__) && (defined(IP_MTU) || defined(IPV6_MTU)) + int sock; + int mtu = 0; + socklen_t len = sizeof(mtu); + + sock = socket(__AF, SOCK_DGRAM, IPPROTO_UDP); + if (sock < 0) + return UDP_MTU_FALLBACK; + + if (connect(sock, (const struct sockaddr *) saddr, + sizeof(*saddr)) < 0) + goto fallback; + +#if defined(BUILD_IPCP_UDP4) && defined(IP_MTU) + if (getsockopt(sock, IPPROTO_IP, IP_MTU, &mtu, &len) < 0) + goto fallback; +#elif defined(BUILD_IPCP_UDP6) && defined(IPV6_MTU) + if (getsockopt(sock, IPPROTO_IPV6, IPV6_MTU, &mtu, &len) < 0) + goto fallback; +#else + goto fallback; +#endif + close(sock); + + if (mtu <= (int) UDP_IP_OVERHEAD) + return UDP_MTU_FALLBACK; + + return (uint32_t) mtu - UDP_IP_OVERHEAD; + + fallback: + close(sock); +#else + (void) saddr; +#endif + return UDP_MTU_FALLBACK; +} + static int udp_data_init(void) { int i; @@ -199,11 +245,11 @@ static void udp_data_fini(void) pthread_mutex_destroy(&udp_data.mgmt_lock); } -static int udp_ipcp_port_alloc(const struct sockaddr_in * r_saddr, - uint32_t s_eid, - const uint8_t * dst, - qosspec_t qs, - const buffer_t * data) +static int udp_ipcp_port_alloc(const struct __SOCKADDR * r_saddr, + uint32_t s_eid, + const uint8_t * dst, + qosspec_t qs, + const buffer_t * data) { uint8_t * buf; struct mgmt_msg * msg; @@ -217,6 +263,8 @@ static int udp_ipcp_port_alloc(const struct sockaddr_in * r_saddr, if (buf == NULL) return -1; + memset(buf, 0, len + data->len); + msg = (struct mgmt_msg *) buf; msg->eid = hton32(MGMT_EID); msg->code = FLOW_REQ; @@ -226,9 +274,8 @@ static int udp_ipcp_port_alloc(const struct sockaddr_in * r_saddr, msg->availability = qs.availability; msg->loss = hton32(qs.loss); msg->ber = hton32(qs.ber); - msg->in_order = qs.in_order; + msg->service = qs.service; msg->max_gap = hton32(qs.max_gap); - msg->cypher_s = hton16(qs.cypher_s); msg->timeout = hton32(qs.timeout); memcpy(msg + 1, dst, ipcp_dir_hash_len()); @@ -238,6 +285,8 @@ static int udp_ipcp_port_alloc(const struct sockaddr_in * r_saddr, if (sendto(udp_data.s_fd, msg, len + data->len, SENDTO_FLAGS, (const struct sockaddr *) r_saddr, sizeof(*r_saddr)) < 0) { + log_err("Failed to send flow allocation request: %s.", + strerror(errno)); free(buf); return -1; } @@ -247,11 +296,11 @@ static int udp_ipcp_port_alloc(const struct sockaddr_in * r_saddr, return 0; } -static int udp_ipcp_port_alloc_resp(const struct sockaddr_in * r_saddr, - uint32_t s_eid, - uint32_t d_eid, - int8_t response, - const buffer_t * data) +static int udp_ipcp_port_alloc_resp(const struct __SOCKADDR * r_saddr, + uint32_t s_eid, + uint32_t d_eid, + int32_t response, + const buffer_t * data) { struct mgmt_msg * msg; @@ -259,11 +308,13 @@ static int udp_ipcp_port_alloc_resp(const struct sockaddr_in * r_saddr, if (msg == NULL) return -1; + memset(msg, 0, sizeof(*msg) + data->len); + msg->eid = hton32(MGMT_EID); msg->code = FLOW_REPLY; msg->s_eid = hton32(s_eid); msg->d_eid = hton32(d_eid); - msg->response = response; + msg->response = hton32(response); if (data->len > 0) memcpy(msg + 1, data->data, data->len); @@ -280,15 +331,58 @@ static int udp_ipcp_port_alloc_resp(const struct sockaddr_in * r_saddr, return 0; } -static int udp_ipcp_port_req(struct sockaddr_in * c_saddr, - int d_eid, - const uint8_t * dst, - qosspec_t qs, - const buffer_t * data) +static int udp_ipcp_flow_update(int fd, + const buffer_t * data) +{ + struct mgmt_msg * msg; + struct __SOCKADDR r_saddr; + uint32_t d_eid; + + msg = malloc(sizeof(*msg) + data->len); + if (msg == NULL) + return -1; + + memset(msg, 0, sizeof(*msg) + data->len); + + pthread_rwlock_rdlock(&udp_data.flows_lock); + + r_saddr = udp_data.fd_to_uf[fd].r_saddr; + d_eid = (uint32_t) udp_data.fd_to_uf[fd].d_eid; + + pthread_rwlock_unlock(&udp_data.flows_lock); + + msg->eid = hton32(MGMT_EID); + msg->code = FLOW_IRM_UPDATE; + msg->s_eid = hton32(d_eid); + msg->d_eid = hton32((uint32_t) fd); + + if (data->len > 0) + memcpy(msg + 1, data->data, data->len); + + if (sendto(udp_data.s_fd, msg, sizeof(*msg) + data->len, + SENDTO_FLAGS, + (const struct sockaddr *) &r_saddr, + sizeof(r_saddr)) < 0) { + log_err("Failed to send flow update: %s.", strerror(errno)); + free(msg); + return -1; + } + + free(msg); + + return 0; +} + +static int udp_ipcp_port_req(struct __SOCKADDR * c_saddr, + int d_eid, + const uint8_t * dst, + qosspec_t qs, + const buffer_t * data) { int fd; - fd = ipcp_wait_flow_req_arr(dst, qs, IPCP_UDP_MPL, data); + fd = ipcp_wait_flow_req_arr(dst, qs, IPCP_UDP_MPL, + udp_query_mtu(c_saddr), data); if (fd < 0) { log_err("Could not get new flow from IRMd."); return -1; @@ -307,20 +401,26 @@ static int udp_ipcp_port_req(struct sockaddr_in * c_saddr, return 0; } -static int udp_ipcp_port_alloc_reply(const struct sockaddr_in * saddr, - uint32_t s_eid, - uint32_t d_eid, - int8_t response, - const buffer_t * data) +static int udp_ipcp_port_alloc_reply(const struct __SOCKADDR * saddr, + uint32_t s_eid, + uint32_t d_eid, + int32_t response, + const buffer_t * data) { time_t mpl = IPCP_UDP_MPL; pthread_rwlock_wrlock(&udp_data.flows_lock); if (memcmp(&udp_data.fd_to_uf[s_eid].r_saddr, saddr, sizeof(*saddr))) { + char ipstr[__ADDRSTRLEN]; pthread_rwlock_unlock(&udp_data.flows_lock); - log_err("Flow allocation reply for %u from wrong source.", - s_eid); + #ifdef BUILD_IPCP_UDP4 + __inet_ntop(&saddr->sin_addr, ipstr); + #else + __inet_ntop(&saddr->sin6_addr, ipstr); + #endif + log_err("Flow allocation reply for %u from wrong source %s.", + s_eid, ipstr); return -1; } @@ -329,7 +429,8 @@ static int udp_ipcp_port_alloc_reply(const struct sockaddr_in * saddr, pthread_rwlock_unlock(&udp_data.flows_lock); - if (ipcp_flow_alloc_reply(s_eid, response, mpl, data) < 0) { + if (ipcp_flow_alloc_reply(s_eid, response, mpl, + udp_query_mtu(saddr), data) < 0) { log_err("Failed to reply to flow allocation."); return -1; } @@ -340,22 +441,58 @@ static int udp_ipcp_port_alloc_reply(const struct sockaddr_in * saddr, return 0; } -static int udp_ipcp_mgmt_frame(const uint8_t * buf, - size_t len, - struct sockaddr_in c_saddr) +static int udp_ipcp_flow_update_arr(const uint8_t * buf, + size_t len) +{ + struct mgmt_msg * msg; + buffer_t data; + int fd; + int flow_id; + + msg = (struct mgmt_msg *) buf; + + fd = (int) ntoh32(msg->s_eid); + if (fd < 0 || fd >= SYS_MAX_FLOWS) { + log_err("Flow update for invalid eid %d.", fd); + return -1; + } + + data.len = len - sizeof(*msg); + data.data = (uint8_t *) buf + sizeof(*msg); + + flow_id = np1_flow_id(fd); + if (flow_id < 0) + return -1; + + if (ipcp_flow_update_arr(flow_id, &data) < 0) { + log_err("Failed to relay flow update on fd %d.", fd); + return -1; + } + + return 0; +} + +static int udp_ipcp_mgmt_frame(struct __SOCKADDR c_saddr, + const uint8_t * buf, + size_t len) { struct mgmt_msg * msg; size_t msg_len; qosspec_t qs; buffer_t data; + /* Defence against malformed/corrupted wire input. */ + if (len < sizeof(*msg)) + return -1; + msg = (struct mgmt_msg *) buf; switch (msg->code) { case FLOW_REQ: msg_len = sizeof(*msg) + ipcp_dir_hash_len(); - assert(len >= msg_len); + if (len < msg_len) + return -1; data.len = len - msg_len; data.data = (uint8_t *) buf + msg_len; @@ -366,25 +503,24 @@ static int udp_ipcp_mgmt_frame(const uint8_t * buf, qs.availability = msg->availability; qs.loss = ntoh32(msg->loss); qs.ber = ntoh32(msg->ber); - qs.in_order = msg->in_order; + qs.service = msg->service; qs.max_gap = ntoh32(msg->max_gap); - qs.cypher_s = ntoh16(msg->cypher_s); qs.timeout = ntoh32(msg->timeout); return udp_ipcp_port_req(&c_saddr, ntoh32(msg->s_eid), (uint8_t *) (msg + 1), qs, &data); case FLOW_REPLY: - assert(len >= sizeof(*msg)); - data.len = len - sizeof(*msg); data.data = (uint8_t *) buf + sizeof(*msg); return udp_ipcp_port_alloc_reply(&c_saddr, ntoh32(msg->s_eid), ntoh32(msg->d_eid), - msg->response, + ntoh32(msg->response), &data); + case FLOW_IRM_UPDATE: + return udp_ipcp_flow_update_arr(buf, len); default: log_err("Unknown message received %d.", msg->code); return -1; @@ -413,7 +549,7 @@ static void * udp_ipcp_mgmt_handler(void * o) pthread_mutex_unlock(&udp_data.mgmt_lock); - udp_ipcp_mgmt_frame(frame->buf, frame->len, frame->r_saddr); + udp_ipcp_mgmt_frame(frame->r_saddr, frame->buf, frame->len); free(frame); } @@ -440,9 +576,9 @@ static void * udp_ipcp_packet_reader(void * o) while (true) { struct mgmt_frame * frame; - struct sockaddr_in r_saddr; + struct __SOCKADDR r_saddr; socklen_t len; - struct shm_du_buff * sdb; + struct ssm_pk_buff * spb; uint8_t * head; len = sizeof(r_saddr); @@ -486,13 +622,13 @@ static void * udp_ipcp_packet_reader(void * o) n-= sizeof(eid); - if (ipcp_sdb_reserve(&sdb, n)) + if (ipcp_spb_reserve(&spb, n)) continue; - head = shm_du_buff_head(sdb); + head = ssm_pk_buff_head(spb); memcpy(head, data, n); - if (np1_flow_write(eid, sdb) < 0) - ipcp_sdb_release(sdb); + if (np1_flow_write(eid, spb, NP1_GET_POOL(eid)) < 0) + ipcp_spb_release(spb); } return (void *) 0; @@ -503,9 +639,9 @@ static void cleanup_fqueue(void * fq) fqueue_destroy((fqueue_t *) fq); } -static void cleanup_sdb(void * sdb) +static void cleanup_spb(void * spb) { - ipcp_sdb_release((struct shm_du_buff *) sdb); + ipcp_spb_release((struct ssm_pk_buff *) spb); } static void * udp_ipcp_packet_writer(void * o) @@ -523,34 +659,34 @@ static void * udp_ipcp_packet_writer(void * o) pthread_cleanup_push(cleanup_fqueue, fq); while (true) { - struct sockaddr_in saddr; - int eid; - int fd; + struct __SOCKADDR saddr; + int eid; + int fd; fevent(udp_data.np1_flows, fq, NULL); while ((fd = fqueue_next(fq)) >= 0) { - struct shm_du_buff * sdb; + struct ssm_pk_buff * spb; uint8_t * buf; uint16_t len; if (fqueue_type(fq) != FLOW_PKT) continue; - if (np1_flow_read(fd, &sdb)) { + if (np1_flow_read(fd, &spb, NP1_GET_POOL(fd))) { log_dbg("Bad read from fd %d.", fd); continue; } - len = shm_du_buff_len(sdb); + len = ssm_pk_buff_len(spb); if (len > IPCP_UDP_MAX_PACKET_SIZE) { log_dbg("Packet length exceeds MTU."); - ipcp_sdb_release(sdb); + ipcp_spb_release(spb); continue; } - buf = shm_du_buff_head_alloc(sdb, OUR_HEADER_LEN); + buf = ssm_pk_buff_push(spb, OUR_HEADER_LEN); if (buf == NULL) { log_dbg("Failed to allocate header."); - ipcp_sdb_release(sdb); + ipcp_spb_release(spb); continue; } @@ -563,7 +699,7 @@ static void * udp_ipcp_packet_writer(void * o) memcpy(buf, &eid, sizeof(eid)); - pthread_cleanup_push(cleanup_sdb, sdb); + pthread_cleanup_push(cleanup_spb, spb); if (sendto(udp_data.s_fd, buf, len + OUR_HEADER_LEN, SENDTO_FLAGS, @@ -580,31 +716,39 @@ static void * udp_ipcp_packet_writer(void * o) return (void *) 1; } -static const char * inet4_ntop(const void * addr, - char * buf) +static bool is_addr_specified(const struct __ADDR * addr) { - return inet_ntop(AF_INET, addr, buf, INET_ADDRSTRLEN); +#ifdef BUILD_IPCP_UDP4 + return addr->s_addr != 0; +#else + return !IN6_IS_ADDR_UNSPECIFIED(addr); +#endif } -static int udp_ipcp_bootstrap(const struct ipcp_config * conf) +static int udp_ipcp_bootstrap(struct ipcp_config * conf) { - char ipstr[INET_ADDRSTRLEN]; - char dnsstr[INET_ADDRSTRLEN]; + char ipstr[__ADDRSTRLEN]; + char dnsstr[__ADDRSTRLEN]; int i = 1; +#ifdef BUILD_IPCP_UDP4 + struct udp4_config * udp; + udp = &conf->udp4; +#else + struct udp6_config * udp; + udp = &conf->udp6; +#endif - assert(conf); + assert(conf != NULL); assert(conf->type == THIS_TYPE); + assert(conf->layer_info.dir_hash_algo == (enum pol_dir_hash) HASH_MD5); - ipcpi.dir_hash_algo = HASH_MD5; - strcpy(ipcpi.layer_name, conf->layer_info.name); - - if (inet4_ntop(&conf->udp.ip_addr, ipstr) == NULL) { + if (__inet_ntop(&udp->ip_addr, ipstr) == NULL) { log_err("Failed to convert IP address."); return -1; } - if (conf->udp.dns_addr != 0) { - if (inet4_ntop(&conf->udp.dns_addr, dnsstr) == NULL) { + if (is_addr_specified(&udp->dns_addr)) { + if (__inet_ntop(&udp->dns_addr, dnsstr) == NULL) { log_err("Failed to convert DNS address."); return -1; } @@ -616,24 +760,29 @@ static int udp_ipcp_bootstrap(const struct ipcp_config * conf) } /* UDP listen server */ - udp_data.s_fd = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP); + udp_data.s_fd = socket(__AF, SOCK_DGRAM, IPPROTO_UDP); if (udp_data.s_fd < 0) { log_err("Can't create socket: %s", strerror(errno)); goto fail_socket; } memset((char *) &udp_data.s_saddr, 0, sizeof(udp_data.s_saddr)); - udp_data.s_saddr.sin_family = AF_INET; - udp_data.s_saddr.sin_addr.s_addr = conf->udp.ip_addr; - udp_data.s_saddr.sin_port = htons(conf->udp.port); - +#ifdef BUILD_IPCP_UDP4 + udp_data.s_saddr.sin_family = AF_INET; + udp_data.s_saddr.sin_addr = udp->ip_addr; + udp_data.s_saddr.sin_port = htons(udp->port); +#else + udp_data.s_saddr.sin6_family = AF_INET6; + udp_data.s_saddr.sin6_addr = udp->ip_addr; + udp_data.s_saddr.sin6_port = htons(udp->port); +#endif if (bind(udp_data.s_fd, SADDR, SADDR_SIZE) < 0) { log_err("Couldn't bind to %s:%d. %s.", - ipstr, conf->udp.port, strerror(errno)); + ipstr, udp->port, strerror(errno)); goto fail_bind; } - udp_data.dns_addr = conf->udp.dns_addr; + udp_data.dns_addr = udp->dns_addr; if (pthread_create(&udp_data.mgmt_handler, NULL, udp_ipcp_mgmt_handler, NULL)) { @@ -657,10 +806,10 @@ static int udp_ipcp_bootstrap(const struct ipcp_config * conf) } } - log_dbg("Bootstrapped IPCP over UDP with pid %d.", getpid()); + log_dbg("Bootstrapped " TYPE_STR " with pid %d.", getpid()); log_dbg("Bound to IP address %s.", ipstr); - log_dbg("Using port %u.", conf->udp.port); - if (conf->udp.dns_addr != 0) + log_dbg("Using port %u.", udp->port); + if (is_addr_specified(&udp_data.dns_addr)) log_dbg("DNS server address is %s.", dnsstr); else log_dbg("DNS server not in use."); @@ -668,14 +817,14 @@ static int udp_ipcp_bootstrap(const struct ipcp_config * conf) return 0; fail_packet_writer: - while (i > 0) { - pthread_cancel(udp_data.packet_writer[--i]); + while (i-- > 0) { + pthread_cancel(udp_data.packet_writer[i]); pthread_join(udp_data.packet_writer[i], NULL); } i = IPCP_UDP_RD_THR; fail_packet_reader: - while (i > 0) { - pthread_cancel(udp_data.packet_reader[--i]); + while (i-- > 0) { + pthread_cancel(udp_data.packet_reader[i]); pthread_join(udp_data.packet_reader[i], NULL); } pthread_cancel(udp_data.mgmt_handler); @@ -738,26 +887,26 @@ static int ddns_send(char * cmd) return 0; } -static uint32_t ddns_resolve(char * name, - uint32_t dns_addr) +static struct __ADDR ddns_resolve(char * name, + struct __ADDR dns_addr) { - pid_t pid = -1; - int wstatus; - int pipe_fd[2]; - char dnsstr[INET_ADDRSTRLEN]; - char buf[IPCP_UDP_BUF_SIZE]; - ssize_t count = 0; - char * substr = NULL; - char * substr2 = NULL; - char * addr_str = "Address:"; - uint32_t ip_addr = 0; - - if (inet4_ntop(&dns_addr, dnsstr) == NULL) - return 0; + pid_t pid = -1; + int wstatus; + int pipe_fd[2]; + char dnsstr[__ADDRSTRLEN]; + char buf[IPCP_UDP_BUF_SIZE]; + ssize_t count = 0; + char * substr = NULL; + char * substr2 = NULL; + char * addr_str = "Address:"; + struct __ADDR ip_addr = __ADDR_ANY_INIT; + + if (__inet_ntop(&dns_addr, dnsstr) == NULL) + return ip_addr; if (pipe(pipe_fd)) { log_err("Failed to create pipe: %s.", strerror(errno)); - return 0; + return ip_addr; } pid = fork(); @@ -765,7 +914,7 @@ static uint32_t ddns_resolve(char * name, log_err("Failed to fork: %s.", strerror(errno)); close(pipe_fd[0]); close(pipe_fd[1]); - return -1; + return ip_addr; } if (pid == 0) { @@ -785,7 +934,7 @@ static uint32_t ddns_resolve(char * name, if (count <= 0) { log_err("Failed to communicate with nslookup."); close(pipe_fd[0]); - return 0; + return ip_addr; } close(pipe_fd[0]); @@ -806,12 +955,13 @@ static uint32_t ddns_resolve(char * name, if (substr2 == NULL || strstr(substr2, addr_str) == NULL) { log_err("Failed to resolve DNS address."); - return 0; + return ip_addr; } - if (inet_pton(AF_INET, substr2 + strlen(addr_str) + 1, &ip_addr) != 1) { + if (inet_pton(__AF, substr2 + strlen(addr_str) + 1, &ip_addr) != 1) { log_err("Failed to resolve DNS address."); - return 0; + assert(!is_addr_specified(&ip_addr)); + return ip_addr; } return ip_addr; @@ -821,11 +971,11 @@ static uint32_t ddns_resolve(char * name, static int udp_ipcp_reg(const uint8_t * hash) { #ifdef HAVE_DDNS - char ipstr[INET_ADDRSTRLEN]; - char dnsstr[INET_ADDRSTRLEN]; - char cmd[1000]; - uint32_t dns_addr; - uint32_t ip_addr; + char ipstr[__ADDRSTRLEN]; + char dnsstr[__ADDRSTRLEN]; + char cmd[1000]; + struct __ADDR dns_addr; + struct __ADDR ip_addr; #endif char * hashstr; @@ -851,16 +1001,19 @@ static int udp_ipcp_reg(const uint8_t * hash) dns_addr = udp_data.dns_addr; - if (dns_addr != 0) { - ip_addr = udp_data.s_saddr.sin_addr.s_addr; - - if (inet4_ntop(&ip_addr, ipstr) == NULL) { + if (is_addr_specified(&dns_addr)) { +#ifdef BUILD_IPCP_UDP4 + ip_addr = udp_data.s_saddr.sin_addr; +#else + ip_addr = udp_data.s_saddr.sin6_addr; +#endif + if (__inet_ntop(&ip_addr, ipstr) == NULL) { log_err("Failed to convert IP address to string."); free(hashstr); return -1; } - if (inet4_ntop(&dns_addr, dnsstr) == NULL) { + if (__inet_ntop(&dns_addr, dnsstr) == NULL) { log_err("Failed to convert DNS address to string."); free(hashstr); return -1; @@ -885,12 +1038,12 @@ static int udp_ipcp_reg(const uint8_t * hash) static int udp_ipcp_unreg(const uint8_t * hash) { #ifdef HAVE_DDNS - char dnsstr[INET_ADDRSTRLEN]; + char dnsstr[__ADDRSTRLEN]; /* max DNS name length + max IP length + max command length */ - char cmd[100]; - uint32_t dns_addr; + char cmd[100]; + struct __ADDR dns_addr; #endif - char * hashstr; + char * hashstr; assert(hash); @@ -907,8 +1060,8 @@ static int udp_ipcp_unreg(const uint8_t * hash) dns_addr = udp_data.dns_addr; - if (dns_addr != 0) { - if (inet4_ntop(&dns_addr, dnsstr) == NULL) { + if (is_addr_specified(&dns_addr)) { + if (__inet_ntop(&dns_addr, dnsstr) == NULL) { log_err("Failed to convert DNS address to string."); free(hashstr); return -1; @@ -929,11 +1082,13 @@ static int udp_ipcp_unreg(const uint8_t * hash) static int udp_ipcp_query(const uint8_t * hash) { - uint32_t ip_addr = 0; - char * hashstr; - struct hostent * h; + struct addr addr = {}; + char * hashstr; + struct addrinfo hints; + struct addrinfo * ai; #ifdef HAVE_DDNS - uint32_t dns_addr = 0; + struct __ADDR dns_addr = __ADDR_ANY_INIT; + struct __ADDR ip_addr = __ADDR_ANY_INIT; #endif assert(hash); @@ -953,28 +1108,42 @@ static int udp_ipcp_query(const uint8_t * hash) #ifdef HAVE_DDNS dns_addr = udp_data.dns_addr; - if (dns_addr != 0) { + if (is_addr_specified(&dns_addr)) { ip_addr = ddns_resolve(hashstr, dns_addr); - if (ip_addr == 0) { + if (!is_addr_specified(&ip_addr)) { log_err("Could not resolve %s.", hashstr); free(hashstr); return -1; } } else { #endif - h = gethostbyname(hashstr); - if (h == NULL) { - log_err("Could not resolve %s.", hashstr); + memset(&hints, 0, sizeof(hints)); + + hints.ai_family = __AF; + if (getaddrinfo(hashstr, NULL, &hints, &ai) != 0) { + log_err("Could not resolve %s: %s.", hashstr, + gai_strerror(errno)); + free(hashstr); + return -1; + } + + if (ai->ai_family != __AF) { + log_err("Wrong addres family for %s.", hashstr); + freeaddrinfo(ai); free(hashstr); return -1; } - ip_addr = *((uint32_t *) (h->h_addr_list[0])); + #ifdef BUILD_IPCP_UDP4 + addr.ip4 = ((struct sockaddr_in *) (ai->ai_addr))->sin_addr; + #else + addr.ip6 = ((struct sockaddr_in6 *) (ai->ai_addr))->sin6_addr; + #endif + freeaddrinfo(ai); #ifdef HAVE_DDNS } #endif - - if (shim_data_dir_add_entry(udp_data.shim_data, hash, ip_addr)) { + if (shim_data_dir_add_entry(udp_data.shim_data, hash, addr)) { log_err("Failed to add directory entry."); free(hashstr); return -1; @@ -990,9 +1159,10 @@ static int udp_ipcp_flow_alloc(int fd, qosspec_t qs, const buffer_t * data) { - struct sockaddr_in r_saddr; /* Server address */ - uint32_t ip_addr = 0; - char ipstr[INET_ADDRSTRLEN]; + struct __SOCKADDR r_saddr; /* Server address */ + struct __ADDR ip_addr; + struct addr addr; + char ipstr[__ADDRSTRLEN]; (void) qs; @@ -1003,9 +1173,13 @@ static int udp_ipcp_flow_alloc(int fd, return -1; } - ip_addr = (uint32_t) shim_data_dir_get_addr(udp_data.shim_data, dst); - - if (inet4_ntop(&ip_addr, ipstr) == NULL) { + addr = shim_data_dir_get_addr(udp_data.shim_data, dst); +#ifdef BUILD_IPCP_UDP4 + ip_addr = addr.ip4; +#else + ip_addr = addr.ip6; +#endif + if (__inet_ntop(&ip_addr, ipstr) == NULL) { log_err("Could not convert IP address."); return -1; } @@ -1014,9 +1188,15 @@ static int udp_ipcp_flow_alloc(int fd, HASH_VAL32(dst), ipstr); memset((char *) &r_saddr, 0, sizeof(r_saddr)); - r_saddr.sin_family = AF_INET; - r_saddr.sin_addr.s_addr = ip_addr; - r_saddr.sin_port = udp_data.s_saddr.sin_port; +#ifdef BUILD_IPCP_UDP4 + r_saddr.sin_family = AF_INET; + r_saddr.sin_addr = addr.ip4; + r_saddr.sin_port = udp_data.s_saddr.sin_port; +#else + r_saddr.sin6_family = AF_INET6; + r_saddr.sin6_addr = addr.ip6; + r_saddr.sin6_port = udp_data.s_saddr.sin6_port; +#endif if (udp_ipcp_port_alloc(&r_saddr, fd, dst, qs, data) < 0) { log_err("Could not allocate port."); @@ -1039,8 +1219,8 @@ static int udp_ipcp_flow_alloc_resp(int fd, int resp, const buffer_t * data) { - struct sockaddr_in saddr; - int d_eid; + struct __SOCKADDR saddr; + int d_eid; if (ipcp_wait_flow_resp(fd) < 0) { log_err("Failed to wait for flow response."); @@ -1094,7 +1274,8 @@ static struct ipcp_ops udp_ops = { .ipcp_flow_alloc = udp_ipcp_flow_alloc, .ipcp_flow_join = NULL, .ipcp_flow_alloc_resp = udp_ipcp_flow_alloc_resp, - .ipcp_flow_dealloc = udp_ipcp_flow_dealloc + .ipcp_flow_dealloc = udp_ipcp_flow_dealloc, + .ipcp_flow_update = udp_ipcp_flow_update }; int main(int argc, diff --git a/src/ipcpd/udp/udp4.c b/src/ipcpd/udp/udp4.c new file mode 100644 index 00000000..ff57bc09 --- /dev/null +++ b/src/ipcpd/udp/udp4.c @@ -0,0 +1,42 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * IPC process over UDP/IPv4 + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#if defined(__linux__) || defined(__CYGWIN__) +#define _DEFAULT_SOURCE +#else +#define _POSIX_C_SOURCE 200112L +#endif + +#include <ouroboros/ipcp-dev.h> + +#define BUILD_IPCP_UDP4 +#define THIS_TYPE IPCP_UDP4 +#define TYPE_STR "IPCP over UDP/IPv4" +#define OUROBOROS_PREFIX "ipcpd/udp4" +#define IPCP_UDP_MAX_PACKET_SIZE 8980 +#define __AF AF_INET +#define __ADDRSTRLEN INET_ADDRSTRLEN +#define __SOCKADDR sockaddr_in +#define __ADDR in_addr +#define __ADDR_ANY_INIT { .s_addr = INADDR_ANY } + +#include "udp.c" diff --git a/src/ipcpd/udp/udp6.c b/src/ipcpd/udp/udp6.c new file mode 100644 index 00000000..2ceb95f0 --- /dev/null +++ b/src/ipcpd/udp/udp6.c @@ -0,0 +1,42 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * IPC process over UDP/IPv6 + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#if defined(__linux__) || defined(__CYGWIN__) +#define _DEFAULT_SOURCE +#else +#define _POSIX_C_SOURCE 200112L +#endif + +#include <ouroboros/ipcp-dev.h> + +#define BUILD_IPCP_UDP6 +#define THIS_TYPE IPCP_UDP6 +#define TYPE_STR "IPCP over UDP/IPv6" +#define OUROBOROS_PREFIX "ipcpd/udp6" +#define IPCP_UDP_MAX_PACKET_SIZE 8952 +#define __AF AF_INET6 +#define __ADDRSTRLEN INET6_ADDRSTRLEN +#define __SOCKADDR sockaddr_in6 +#define __ADDR in6_addr +#define __ADDR_ANY_INIT IN6ADDR_ANY_INIT + +#include "udp.c" diff --git a/src/ipcpd/unicast/CMakeLists.txt b/src/ipcpd/unicast/CMakeLists.txt index ca742871..d3388112 100644 --- a/src/ipcpd/unicast/CMakeLists.txt +++ b/src/ipcpd/unicast/CMakeLists.txt @@ -1,38 +1,9 @@ -get_filename_component(CURRENT_SOURCE_PARENT_DIR - ${CMAKE_CURRENT_SOURCE_DIR} DIRECTORY) -get_filename_component(CURRENT_BINARY_PARENT_DIR - ${CMAKE_CURRENT_BINARY_DIR} DIRECTORY) +# Unicast IPCP build configuration -include_directories(${CMAKE_CURRENT_SOURCE_DIR}) -include_directories(${CMAKE_CURRENT_BINARY_DIR}) +protobuf_generate_c(DHT_PROTO_SRCS DHT_PROTO_HDRS + "${CMAKE_CURRENT_SOURCE_DIR}/dir/dht.proto") -include_directories(${CURRENT_SOURCE_PARENT_DIR}) -include_directories(${CURRENT_BINARY_PARENT_DIR}) - -include_directories(${CMAKE_SOURCE_DIR}/include) -include_directories(${CMAKE_BINARY_DIR}/include) - -set(IPCP_UNICAST_TARGET ipcpd-unicast CACHE INTERNAL "") -set(IPCP_UNICAST_MPL 60 CACHE STRING - "Default maximum packet lifetime for the unicast IPCP, in seconds") - -protobuf_generate_c(DHT_PROTO_SRCS DHT_PROTO_HDRS dir/dht.proto) - -math(EXPR PFT_EXPR "1 << 12") -set(PFT_SIZE ${PFT_EXPR} CACHE STRING - "Size of the PDU forwarding table") -if (HAVE_FUSE) - set(IPCP_FLOW_STATS TRUE CACHE BOOL - "Enable flow statistics tracking in IPCP") - if (IPCP_FLOW_STATS) - message(STATUS "IPCP flow statistics enabled") - else () - message(STATUS "IPCP flow statistics disabled") - endif () -endif () - -set(SOURCE_FILES - # Add source files here +set(UNICAST_SOURCES addr-auth.c ca.c connmgr.c @@ -43,7 +14,6 @@ set(SOURCE_FILES pff.c routing.c psched.c - # Add policies last addr-auth/flat.c ca/mb-ecn.c ca/nop.c @@ -54,22 +24,26 @@ set(SOURCE_FILES pff/pft.c routing/link-state.c routing/graph.c - ) +) -add_executable(ipcpd-unicast ${SOURCE_FILES} ${IPCP_SOURCES} ${COMMON_SOURCES} - ${DHT_PROTO_SRCS} ${LAYER_CONFIG_PROTO_SRCS}) -target_link_libraries(ipcpd-unicast LINK_PUBLIC ouroboros-dev) +add_executable(${IPCP_UNICAST_TARGET} + ${UNICAST_SOURCES} + ${IPCP_SOURCES} + ${COMMON_SOURCES} + ${DHT_PROTO_SRCS} +) -include(AddCompileFlags) -if (CMAKE_BUILD_TYPE MATCHES "Debug*") - add_compile_flags(ipcpd-unicast -DCONFIG_OUROBOROS_DEBUG) -endif () +target_include_directories(${IPCP_UNICAST_TARGET} PRIVATE ${IPCP_INCLUDE_DIRS}) +target_include_directories(${IPCP_UNICAST_TARGET} PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}) +target_include_directories(${IPCP_UNICAST_TARGET} PRIVATE ${CMAKE_CURRENT_BINARY_DIR}) +target_link_libraries(${IPCP_UNICAST_TARGET} PRIVATE ouroboros-dev) -install(TARGETS ipcpd-unicast RUNTIME DESTINATION ${CMAKE_INSTALL_SBINDIR}) +ouroboros_target_debug_definitions(${IPCP_UNICAST_TARGET}) -add_subdirectory(pff/tests) -add_subdirectory(routing/tests) +install(TARGETS ${IPCP_UNICAST_TARGET} RUNTIME DESTINATION ${CMAKE_INSTALL_SBINDIR}) -if (NOT GNU) +if(BUILD_TESTS) add_subdirectory(dir/tests) -endif () + add_subdirectory(pff/tests) + add_subdirectory(routing/tests) +endif() diff --git a/src/ipcpd/unicast/addr-auth.c b/src/ipcpd/unicast/addr-auth.c index 908a4aa1..27671d09 100644 --- a/src/ipcpd/unicast/addr-auth.c +++ b/src/ipcpd/unicast/addr-auth.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Address authority * diff --git a/src/ipcpd/unicast/addr-auth.h b/src/ipcpd/unicast/addr-auth.h index e119dff3..65567dc3 100644 --- a/src/ipcpd/unicast/addr-auth.h +++ b/src/ipcpd/unicast/addr-auth.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Address authority * @@ -27,6 +27,14 @@ #include <stdint.h> +#define ADDR_FMT32 "%02x.%02x.%02x.%02x" +#define ADDR_VAL32(a) \ + ((uint8_t *) a)[0], ((uint8_t *) a)[1], \ + ((uint8_t *) a)[2], ((uint8_t *) a)[3] + +#define ADDR_FMT64 ADDR_FMT32 "." ADDR_FMT32 +#define ADDR_VAL64(a) ADDR_VAL32(a), ADDR_VAL32(a + 4) + int addr_auth_init(enum pol_addr_auth type, const void * info); diff --git a/src/ipcpd/unicast/addr-auth/flat.c b/src/ipcpd/unicast/addr-auth/flat.c index c4562935..cf832279 100644 --- a/src/ipcpd/unicast/addr-auth/flat.c +++ b/src/ipcpd/unicast/addr-auth/flat.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Policy for flat addresses in a distributed way * @@ -31,17 +31,18 @@ #include <ouroboros/logs.h> #include <ouroboros/random.h> +#include "addr-auth.h" #include "ipcp.h" #include "flat.h" -#define NAME_LEN 8 +#define NAME_LEN 8 +#define INVALID_ADDRESS 0 struct { - uint8_t addr_size; + uint8_t addr_size; + uint32_t addr; } flat; -#define INVALID_ADDRESS 0 - struct addr_auth_ops flat_ops = { .init = flat_init, .fini = flat_fini, @@ -57,6 +58,15 @@ int flat_init(const void * info) return -1; } +#if defined (CONFIG_OUROBOROS_DEBUG) && defined (IPCP_DEBUG_LOCAL) + flat.addr = getpid(); +#else + while (flat.addr == INVALID_ADDRESS) + random_buffer(&flat.addr,sizeof(flat.addr)); +#endif + log_dbg("Flat address initialized to " ADDR_FMT32 ".", + ADDR_VAL32((uint8_t *) &flat.addr)); + return 0; } @@ -67,13 +77,5 @@ int flat_fini(void) uint64_t flat_address(void) { - uint32_t addr = INVALID_ADDRESS; - -#if defined (CONFIG_OUROBOROS_DEBUG) && defined (IPCP_DEBUG_LOCAL) - addr = getpid(); -#else - while (addr == INVALID_ADDRESS) - random_buffer(&addr,sizeof(addr)); -#endif - return addr; + return (uint64_t) flat.addr; } diff --git a/src/ipcpd/unicast/addr-auth/flat.h b/src/ipcpd/unicast/addr-auth/flat.h index d4b672c7..b5c7e525 100644 --- a/src/ipcpd/unicast/addr-auth/flat.h +++ b/src/ipcpd/unicast/addr-auth/flat.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Policy for flat addresses in a distributed way * diff --git a/src/ipcpd/unicast/addr-auth/ops.h b/src/ipcpd/unicast/addr-auth/ops.h index 06b24cec..f5c4c611 100644 --- a/src/ipcpd/unicast/addr-auth/ops.h +++ b/src/ipcpd/unicast/addr-auth/ops.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Address authority policy ops * diff --git a/src/ipcpd/unicast/addr-auth/pol.h b/src/ipcpd/unicast/addr-auth/pol.h index 844308c6..a33f1dd6 100644 --- a/src/ipcpd/unicast/addr-auth/pol.h +++ b/src/ipcpd/unicast/addr-auth/pol.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Address Authority policies * diff --git a/src/ipcpd/unicast/ca.c b/src/ipcpd/unicast/ca.c index 287eaf41..a1751672 100644 --- a/src/ipcpd/unicast/ca.c +++ b/src/ipcpd/unicast/ca.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Congestion Avoidance * @@ -49,7 +49,6 @@ int ca_init(enum pol_cong_avoid pol) return 0; } - void ca_fini(void) { ca.ops = NULL; diff --git a/src/ipcpd/unicast/ca.h b/src/ipcpd/unicast/ca.h index ea803e17..47ea15a0 100644 --- a/src/ipcpd/unicast/ca.h +++ b/src/ipcpd/unicast/ca.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Congestion avoidance * diff --git a/src/ipcpd/unicast/ca/mb-ecn.c b/src/ipcpd/unicast/ca/mb-ecn.c index d9a204b0..b310c4fc 100644 --- a/src/ipcpd/unicast/ca/mb-ecn.c +++ b/src/ipcpd/unicast/ca/mb-ecn.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Multi-bit ECN Congestion Avoidance * diff --git a/src/ipcpd/unicast/ca/mb-ecn.h b/src/ipcpd/unicast/ca/mb-ecn.h index 9a2c8b49..1be27764 100644 --- a/src/ipcpd/unicast/ca/mb-ecn.h +++ b/src/ipcpd/unicast/ca/mb-ecn.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Multi-bit ECN Congestion Avoidance * diff --git a/src/ipcpd/unicast/ca/nop.c b/src/ipcpd/unicast/ca/nop.c index 617fc15b..e5cacf66 100644 --- a/src/ipcpd/unicast/ca/nop.c +++ b/src/ipcpd/unicast/ca/nop.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Dummy Congestion Avoidance * diff --git a/src/ipcpd/unicast/ca/nop.h b/src/ipcpd/unicast/ca/nop.h index 248b198d..8b892e61 100644 --- a/src/ipcpd/unicast/ca/nop.h +++ b/src/ipcpd/unicast/ca/nop.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Dummy Congestion Avoidance * diff --git a/src/ipcpd/unicast/ca/ops.h b/src/ipcpd/unicast/ca/ops.h index 3a7b7248..6d2ddf1d 100644 --- a/src/ipcpd/unicast/ca/ops.h +++ b/src/ipcpd/unicast/ca/ops.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Congestion avoidance policy ops * diff --git a/src/ipcpd/unicast/ca/pol.h b/src/ipcpd/unicast/ca/pol.h index db0a1a11..bfb9cc2d 100644 --- a/src/ipcpd/unicast/ca/pol.h +++ b/src/ipcpd/unicast/ca/pol.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Congestion avoidance policies * diff --git a/src/ipcpd/unicast/connmgr.c b/src/ipcpd/unicast/connmgr.c index 11c5d5b6..f60f6fca 100644 --- a/src/ipcpd/unicast/connmgr.c +++ b/src/ipcpd/unicast/connmgr.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Handles connections between components * @@ -32,8 +32,4 @@ #define BUILD_IPCP_UNICAST -#ifdef IPCP_CONN_WAIT_DIR - #include "dir.h" -#endif - #include "common/connmgr.c" diff --git a/src/ipcpd/unicast/dir.c b/src/ipcpd/unicast/dir.c index e0cb09fc..a31a562f 100644 --- a/src/ipcpd/unicast/dir.c +++ b/src/ipcpd/unicast/dir.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Directory Management * @@ -44,50 +44,57 @@ struct { struct dir_ops * ops; - void * dir; -} dirmgr; +} dir; -int dir_init(void) +int dir_init(struct dir_config * conf) { - dirmgr.ops = &dht_dir_ops; + void * cfg; - dirmgr.dir = dirmgr.ops->create(); - if (dirmgr.dir == NULL) { - dirmgr.ops = NULL; - return -ENOMEM; + assert(conf != NULL); + + switch (conf->pol) { + case DIR_DHT: + log_info("Using DHT policy."); + dir.ops = &dht_dir_ops; + cfg = &conf->dht; + break; + default: /* DIR_INVALID */ + log_err("Invalid directory policy %d.", conf->pol); + return -EINVAL; } - return 0; + assert(dir.ops->init != NULL); + + return dir.ops->init(cfg); } void dir_fini(void) { - dirmgr.ops->destroy(dirmgr.dir); - dirmgr.ops = NULL; - dirmgr.dir = NULL; + dir.ops->fini(); + dir.ops = NULL; } -int dir_bootstrap(void) +int dir_start(void) { - return dirmgr.ops->bootstrap(dirmgr.dir); + return dir.ops->start(); } -int dir_reg(const uint8_t * hash) +void dir_stop(void) { - return dirmgr.ops->reg(dirmgr.dir, hash); + dir.ops->stop(); } -int dir_unreg(const uint8_t * hash) +int dir_reg(const uint8_t * hash) { - return dirmgr.ops->unreg(dirmgr.dir, hash); + return dir.ops->reg(hash); } -uint64_t dir_query(const uint8_t * hash) +int dir_unreg(const uint8_t * hash) { - return dirmgr.ops->query(dirmgr.dir, hash); + return dir.ops->unreg(hash); } -int dir_wait_running(void) +uint64_t dir_query(const uint8_t * hash) { - return dirmgr.ops->wait_running(dirmgr.dir); + return dir.ops->query(hash); } diff --git a/src/ipcpd/unicast/dir.h b/src/ipcpd/unicast/dir.h index b261ea2c..2be7b10f 100644 --- a/src/ipcpd/unicast/dir.h +++ b/src/ipcpd/unicast/dir.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Directory * @@ -25,11 +25,14 @@ #include <inttypes.h> -int dir_init(void); +/* may update the config! */ +int dir_init(struct dir_config * conf); void dir_fini(void); -int dir_bootstrap(void); +int dir_start(void); + +void dir_stop(void); int dir_reg(const uint8_t * hash); @@ -37,6 +40,4 @@ int dir_unreg(const uint8_t * hash); uint64_t dir_query(const uint8_t * hash); -int dir_wait_running(void); - #endif /* OUROBOROS_IPCPD_UNICAST_DIR_H */ diff --git a/src/ipcpd/unicast/dir/dht.c b/src/ipcpd/unicast/dir/dht.c index 08a5a5a9..8eeea800 100644 --- a/src/ipcpd/unicast/dir/dht.c +++ b/src/ipcpd/unicast/dir/dht.c @@ -1,10 +1,9 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Distributed Hash Table based on Kademlia * * Dimitri Staessens <dimitri@ouroboros.rocks> - * Sander Vrijders <sander@ouroboros.rocks> * * This library is free software; you can redistribute it and/or * modify it under the terms of the GNU Lesser General Public License @@ -20,10 +19,12 @@ * Foundation, Inc., http://www.fsf.org/about/contact/. */ -#if defined(__linux__) || defined(__CYGWIN__) -#define _DEFAULT_SOURCE -#else -#define _POSIX_C_SOURCE 200112L +#if !defined (__DHT_TEST__) + #if defined(__linux__) || defined(__CYGWIN__) + #define _DEFAULT_SOURCE + #else + #define _POSIX_C_SOURCE 200112L + #endif #endif #include "config.h" @@ -38,13 +39,14 @@ #include <ouroboros/errno.h> #include <ouroboros/logs.h> #include <ouroboros/list.h> -#include <ouroboros/notifier.h> #include <ouroboros/random.h> +#include <ouroboros/rib.h> #include <ouroboros/time.h> #include <ouroboros/tpm.h> #include <ouroboros/utils.h> #include <ouroboros/pthread.h> +#include "addr-auth.h" #include "common/connmgr.h" #include "dht.h" #include "dt.h" @@ -58,125 +60,96 @@ #include <limits.h> #include "dht.pb-c.h" -typedef DhtMsg dht_msg_t; -typedef DhtContactMsg dht_contact_msg_t; +typedef DhtMsg dht_msg_t; +typedef DhtContactMsg dht_contact_msg_t; +typedef DhtStoreMsg dht_store_msg_t; +typedef DhtFindReqMsg dht_find_req_msg_t; +typedef DhtFindNodeRspMsg dht_find_node_rsp_msg_t; +typedef DhtFindValueRspMsg dht_find_value_rsp_msg_t; +typedef ProtobufCBinaryData binary_data_t; #ifndef CLOCK_REALTIME_COARSE #define CLOCK_REALTIME_COARSE CLOCK_REALTIME #endif -#define DHT_MAX_REQS 2048 /* KAD recommends rnd(), bmp can be changed. */ -#define KAD_ALPHA 3 /* Parallel factor, proven optimal value. */ -#define KAD_K 8 /* Replication factor, MDHT value. */ -#define KAD_T_REPL 900 /* Replication time, tied to k. MDHT value. */ -#define KAD_T_REFR 900 /* Refresh time stale bucket, MDHT value. */ -#define KAD_T_JOIN 8 /* Response time to wait for a join. */ -#define KAD_T_RESP 5 /* Response time to wait for a response. */ -#define KAD_R_PING 2 /* Ping retries before declaring peer dead. */ -#define KAD_QUEER 15 /* Time to declare peer questionable. */ -#define KAD_BETA 8 /* Bucket split factor, must be 1, 2, 4 or 8. */ -#define KAD_RESP_RETR 6 /* Number of retries on sending a response. */ -#define KAD_JOIN_RETR 8 /* Number of retries sending a join. */ -#define KAD_JOIN_INTV 1 /* Time (seconds) between join retries. */ +#define DHT_MAX_REQS 128 /* KAD recommends rnd(), bmp can be changed. */ +#define DHT_WARN_REQS 100 /* Warn if number of requests exceeds this. */ +#define DHT_MAX_VALS 8 /* Max number of values to return for a key. */ +#define DHT_T_CACHE 60 /* Max cache time for values (s) */ +#define DHT_T_RESP 2 /* Response time to wait for a response (s). */ +#define DHT_N_REPUB 5 /* Republish if expiry within n replications. */ +#define DHT_R_PING 2 /* Ping retries before declaring peer dead. */ +#define DHT_QUEER 15 /* Time to declare peer questionable. */ +#define DHT_BETA 8 /* Bucket split factor, must be 1, 2, 4 or 8. */ +#define DHT_RESP_RETR 6 /* Number of retries on sending a response. */ #define HANDLE_TIMEO 1000 /* Timeout for dht_handle_packet tpm check (ms) */ -#define DHT_RETR_ADDR 1 /* Number of addresses to return on retrieve */ +#define DHT_INVALID 0 /* Invalid cookie value. */ -enum dht_state { - DHT_INIT = 0, - DHT_SHUTDOWN, - DHT_JOINING, - DHT_RUNNING, -}; +#define KEY_FMT "K<" HASH_FMT64 ">" +#define KEY_VAL(key) HASH_VAL64(key) -enum kad_code { - KAD_JOIN = 0, - KAD_FIND_NODE, - KAD_FIND_VALUE, - /* Messages without a response below. */ - KAD_STORE, - KAD_RESPONSE -}; +#define VAL_FMT "V<" HASH_FMT64 ">" +#define VAL_VAL(val) HASH_VAL64((val).data) -enum kad_req_state { - REQ_NULL = 0, - REQ_INIT, - REQ_PENDING, - REQ_RESPONSE, - REQ_DONE, - REQ_DESTROY -}; +#define KV_FMT "<" HASH_FMT64 ", " HASH_FMT64 ">" +#define KV_VAL(key, val) HASH_VAL64(key), HASH_VAL64((val).data) -enum lookup_state { - LU_NULL = 0, - LU_INIT, - LU_PENDING, - LU_UPDATE, - LU_COMPLETE, - LU_DESTROY -}; +#define PEER_FMT "[" HASH_FMT64 "|" ADDR_FMT32 "]" +#define PEER_VAL(id, addr) HASH_VAL64(id), ADDR_VAL32(&(addr)) -struct kad_req { - struct list_head next; +#define DHT_CODE(msg) dht_code_str[(msg)->code] - uint32_t cookie; - enum kad_code code; - uint8_t * key; - uint64_t addr; +#define TX_HDR_FMT "%s --> " PEER_FMT +#define TX_HDR_VAL(msg, id, addr) DHT_CODE(msg), PEER_VAL(id, addr) - enum kad_req_state state; - pthread_cond_t cond; - pthread_mutex_t lock; +#define RX_HDR_FMT "%s <-- " PEER_FMT +#define RX_HDR_VAL(msg) DHT_CODE(msg), \ + PEER_VAL(msg->src->id.data, msg->src->addr) - time_t t_exp; -}; +#define CK_FMT "|" HASH_FMT64 "|" +#define CK_VAL(cookie) HASH_VAL64(&(cookie)) -struct cookie_el { - struct list_head next; +#define IS_REQUEST(code) \ + (code == DHT_FIND_NODE_REQ || code == DHT_FIND_VALUE_REQ) - uint32_t cookie; +enum dht_code { + DHT_STORE, + DHT_FIND_NODE_REQ, + DHT_FIND_NODE_RSP, + DHT_FIND_VALUE_REQ, + DHT_FIND_VALUE_RSP }; -struct lookup { - struct list_head next; - - struct list_head cookies; - - uint8_t * key; - - struct list_head contacts; - size_t n_contacts; - - uint64_t * addrs; - size_t n_addrs; - - enum lookup_state state; - pthread_cond_t cond; - pthread_mutex_t lock; +const char * dht_code_str[] = { + "DHT_STORE", + "DHT_FIND_NODE_REQ", + "DHT_FIND_NODE_RSP", + "DHT_FIND_VALUE_REQ", + "DHT_FIND_VALUE_RSP" }; -struct val { - struct list_head next; - - uint64_t addr; - - time_t t_exp; - time_t t_rep; +enum dht_state { + DHT_NULL = 0, + DHT_INIT, + DHT_RUNNING }; -struct ref_entry { +struct val_entry { struct list_head next; - uint8_t * key; + buffer_t val; - time_t t_rep; + time_t t_exp; /* Expiry time */ + time_t t_repl; /* Last replication time */ }; struct dht_entry { struct list_head next; uint8_t * key; - size_t n_vals; - struct list_head vals; + + struct llist vals; /* We don't own these, only replicate */ + struct llist lvals; /* We own these, must be republished */ }; struct contact { @@ -189,1101 +162,1478 @@ struct contact { time_t t_seen; }; -struct bucket { - struct list_head contacts; - size_t n_contacts; +struct peer_entry { + struct list_head next; - struct list_head alts; - size_t n_alts; + uint64_t cookie; + uint8_t * id; + uint64_t addr; + enum dht_code code; + + time_t t_sent; +}; - time_t t_refr; +struct dht_req { + struct list_head next; - size_t depth; - uint8_t mask; + uint8_t * key; + time_t t_exp; - struct bucket * parent; - struct bucket * children[1L << KAD_BETA]; + struct llist peers; + struct llist cache; }; -struct cmd { - struct list_head next; +struct bucket { + struct llist contacts; + struct llist alts; - struct shm_du_buff * sdb; + time_t t_refr; + + size_t depth; + uint8_t mask; + + struct bucket * parent; + struct bucket * children[1L << DHT_BETA]; +}; + +struct cmd { + struct list_head next; + buffer_t cbuf; }; struct dir_ops dht_dir_ops = { - .create = dht_create, - .destroy = dht_destroy, - .bootstrap = dht_bootstrap, - .reg = dht_reg, - .unreg = dht_unreg, - .query = dht_query, - .wait_running = dht_wait_running + .init = (int (*)(void *)) dht_init, + .fini = dht_fini, + .start = dht_start, + .stop = dht_stop, + .reg = dht_reg, + .unreg = dht_unreg, + .query = dht_query }; -struct dht { - size_t alpha; - size_t b; - size_t k; +struct { + struct { /* Kademlia parameters */ + uint32_t alpha; /* Number of concurrent requests */ + size_t k; /* Number of replicas to store */ + time_t t_exp; /* Expiry time for values (s) */ + time_t t_refr; /* Refresh time for contacts (s) */ + time_t t_repl; /* Replication time for values (s) */ + }; - time_t t_expire; - time_t t_refresh; - time_t t_replic; - time_t t_repub; + buffer_t id; - uint8_t * id; - uint64_t addr; + time_t t0; /* Creation time */ + uint64_t addr; /* Our own address */ + uint64_t peer; /* Enrollment peer address */ + uint64_t magic; /* Magic cookie for retransmit */ - struct bucket * buckets; + uint64_t eid; /* Entity ID */ - struct list_head entries; + struct tpm * tpm; + pthread_t worker; - struct list_head refs; + enum dht_state state; - struct list_head lookups; + struct { + struct { + struct bucket * root; + } contacts; - struct list_head requests; - struct bmp * cookies; + struct { + struct llist ll; + size_t vals; + size_t lvals; + } kv; - enum dht_state state; - struct list_head cmds; - pthread_cond_t cond; - pthread_mutex_t mtx; + pthread_rwlock_t lock; + } db; - pthread_rwlock_t lock; + struct { + struct llist ll; + pthread_cond_t cond; + pthread_mutex_t mtx; + } reqs; - uint64_t eid; + struct { + struct list_head list; + pthread_cond_t cond; + pthread_mutex_t mtx; + } cmds; +} dht; - struct tpm * tpm; - pthread_t worker; -}; +/* DHT RIB */ -struct join_info { - struct dht * dht; - uint64_t addr; +static const char * dht_dir[] = { + "database", + "stats", + NULL }; -struct packet_info { - struct dht * dht; - struct shm_du_buff * sdb; -}; +const char * dht_stats = \ + "DHT: " HASH_FMT64 "\n" + " Created: %s\n" + " Address: " ADDR_FMT32 "\n" + " Kademlia parameters:\n" + " Number of concurrent requests (alpha): %10zu\n" + " Number of replicas (k): %10zu\n" + " Expiry time for values (s): %10ld\n" + " Refresh time for contacts (s): %10ld\n" + " Replication time for values (s): %10ld\n" + " Number of keys: %10zu\n" + " Number of local values: %10zu\n" + " Number of non-local values: %10zu\n"; -static uint8_t * dht_dup_key(const uint8_t * key, - size_t len) +static int dht_rib_statfile(char * buf, + size_t len) { - uint8_t * dup; + struct tm * tm; + char tmstr[RIB_TM_STRLEN]; + size_t keys; + size_t vals; + size_t lvals; - dup = malloc(sizeof(*dup) * len); - if (dup == NULL) - return NULL; + assert(buf != NULL); + assert(len > 0); - memcpy(dup, key, len); - - return dup; -} + pthread_rwlock_rdlock(&dht.db.lock); -static enum dht_state dht_get_state(struct dht * dht) -{ - enum dht_state state; + keys = dht.db.kv.ll.len; + lvals = dht.db.kv.lvals; + vals = dht.db.kv.vals; - pthread_mutex_lock(&dht->mtx); + pthread_rwlock_unlock(&dht.db.lock); - state = dht->state; + tm = gmtime(&dht.t0); + strftime(tmstr, sizeof(tmstr), RIB_TM_FORMAT, tm); - pthread_mutex_unlock(&dht->mtx); + snprintf(buf, len, dht_stats, + HASH_VAL64(dht.id.data), + tmstr, + ADDR_VAL32(&dht.addr), + dht.alpha, dht.k, + dht.t_exp, dht.t_refr, dht.t_repl, + keys, vals, lvals); - return state; + return strlen(buf); } -static int dht_set_state(struct dht * dht, - enum dht_state state) +static size_t dht_db_file_len(void) { - pthread_mutex_lock(&dht->mtx); + size_t sz; + size_t vals; - if (state == DHT_JOINING && dht->state != DHT_INIT) { - pthread_mutex_unlock(&dht->mtx); - return -1; + sz = 18; /* DHT database + 2 * \n */ + + pthread_rwlock_rdlock(&dht.db.lock); + + if (llist_is_empty(&dht.db.kv.ll)) { + pthread_rwlock_unlock(&dht.db.lock); + sz += 14; /* No entries */ + return sz; } - dht->state = state; + sz += 39 * 3 + 1; /* tally + extra newline */ + sz += dht.db.kv.ll.len * (25 + 19 + 23 + 1); - pthread_cond_broadcast(&dht->cond); + vals = dht.db.kv.vals + dht.db.kv.lvals; - pthread_mutex_unlock(&dht->mtx); + sz += vals * (48 + 2 * RIB_TM_STRLEN); - return 0; + pthread_rwlock_unlock(&dht.db.lock); + + return sz; } -int dht_wait_running(void * dir) +static int dht_rib_dbfile(char * buf, + size_t len) { - struct dht * dht; - int ret = 0; + struct tm * tm; + char tmstr[RIB_TM_STRLEN]; + char exstr[RIB_TM_STRLEN]; + size_t i = 0; + struct list_head * p; - dht = (struct dht *) dir; + assert(buf != NULL); + assert(len > 0); - pthread_mutex_lock(&dht->mtx); + pthread_rwlock_rdlock(&dht.db.lock); - pthread_cleanup_push(__cleanup_mutex_unlock, &dht->mtx); + if (llist_is_empty(&dht.db.kv.ll)) { + i += snprintf(buf, len, " No entries.\n"); + pthread_rwlock_unlock(&dht.db.lock); + return i; + } - while (dht->state == DHT_JOINING) - pthread_cond_wait(&dht->cond, &dht->mtx); + i += snprintf(buf + i, len - i, "DHT database:\n\n"); + i += snprintf(buf + i, len - i, + "Number of keys: %10zu\n" + "Number of local values: %10zu\n" + "Number of non-local values: %10zu\n\n", + dht.db.kv.ll.len, dht.db.kv.vals, dht.db.kv.lvals); - if (dht->state != DHT_RUNNING) - ret = -1; + llist_for_each(p, &dht.db.kv.ll) { + struct dht_entry * e = list_entry(p, struct dht_entry, next); + struct list_head * h; - pthread_cleanup_pop(true); + i += snprintf(buf + i, len - i, "Key: " KEY_FMT "\n", + KEY_VAL(e->key)); + i += snprintf(buf + i, len - i, " Local entries:\n"); - return ret; -} + llist_for_each(h, &e->vals) { + struct val_entry * v; -static uint8_t * create_id(size_t len) -{ - uint8_t * id; + v = list_entry(h, struct val_entry, next); - id = malloc(len); - if (id == NULL) - return NULL; + tm = gmtime(&v->t_repl); + strftime(tmstr, sizeof(tmstr), RIB_TM_FORMAT, tm); - if (random_buffer(id, len) < 0) { - free(id); - return NULL; - } + tm = gmtime(&v->t_exp); + strftime(exstr, sizeof(exstr), RIB_TM_FORMAT, tm); - return id; -} + i += snprintf(buf + i, len - i, + " " VAL_FMT + ", t_replicated=%.*s, t_exp=%.*s\n", + VAL_VAL(v->val), + RIB_TM_STRLEN, tmstr, + RIB_TM_STRLEN, exstr); + } -static void kad_req_create(struct dht * dht, - dht_msg_t * msg, - uint64_t addr) -{ - struct kad_req * req; - pthread_condattr_t cattr; - struct timespec t; - size_t b; + i += snprintf(buf + i, len - i, "\n"); - clock_gettime(CLOCK_REALTIME_COARSE, &t); + i += snprintf(buf + i, len - i, " Non-local entries:\n"); - req = malloc(sizeof(*req)); - if (req == NULL) - goto fail_malloc; + llist_for_each(h, &e->lvals) { + struct val_entry * v; - list_head_init(&req->next); + v= list_entry(h, struct val_entry, next); + + tm = gmtime(&v->t_repl); + strftime(tmstr, sizeof(tmstr), RIB_TM_FORMAT, tm); - req->t_exp = t.tv_sec + KAD_T_RESP; - req->addr = addr; - req->state = REQ_INIT; - req->cookie = msg->cookie; - req->code = msg->code; - req->key = NULL; + tm = gmtime(&v->t_exp); + strftime(exstr, sizeof(exstr), RIB_TM_FORMAT, tm); - pthread_rwlock_rdlock(&dht->lock); - b = dht->b; - pthread_rwlock_unlock(&dht->lock); + i += snprintf(buf + i, len - i, + " " VAL_FMT + ", t_replicated=%.*s, t_exp=%.*s\n", + VAL_VAL(v->val), + RIB_TM_STRLEN, tmstr, + RIB_TM_STRLEN, exstr); - if (msg->has_key) { - req->key = dht_dup_key(msg->key.data, b); - if (req->key == NULL) - goto fail_dup_key; + } } - if (pthread_mutex_init(&req->lock, NULL)) - goto fail_mutex; + pthread_rwlock_unlock(&dht.db.lock); + printf("DHT RIB DB file generated (%zu bytes).\n", i); - if (pthread_condattr_init(&cattr)) - goto fail_condattr; -#ifndef __APPLE__ - pthread_condattr_setclock(&cattr, PTHREAD_COND_CLOCK); -#endif + return i; +} - if (pthread_cond_init(&req->cond, &cattr)) - goto fail_cond_init; +static int dht_rib_read(const char * path, + char * buf, + size_t len) +{ + char * entry; - pthread_condattr_destroy(&cattr); + entry = strstr(path, RIB_SEPARATOR) + 1; - pthread_rwlock_wrlock(&dht->lock); + if (strcmp(entry, "database") == 0) { + return dht_rib_dbfile(buf, len); + } else if (strcmp(entry, "stats") == 0) { + return dht_rib_statfile(buf, len); + } - list_add(&req->next, &dht->requests); + return 0; +} - pthread_rwlock_unlock(&dht->lock); +static int dht_rib_readdir(char *** buf) +{ + int i = 0; - return; + while (dht_dir[i++] != NULL); - fail_cond_init: - pthread_condattr_destroy(&cattr); - fail_condattr: - pthread_mutex_destroy(&req->lock); - fail_mutex: - free(req->key); - fail_dup_key: - free(req); - fail_malloc: - return; + *buf = malloc(sizeof(**buf) * i); + if (*buf == NULL) + goto fail_buf; + + i = 0; + + while (dht_dir[i] != NULL) { + (*buf)[i] = strdup(dht_dir[i]); + if ((*buf)[i] == NULL) + goto fail_dup; + i++; + } + + return i; + fail_dup: + freepp(char, *buf, i); + fail_buf: + return -ENOMEM; } -static void cancel_req_destroy(void * o) +static int dht_rib_getattr(const char * path, + struct rib_attr * attr) { - struct kad_req * req = (struct kad_req *) o; + struct timespec now; + char * entry; - pthread_mutex_unlock(&req->lock); + clock_gettime(CLOCK_REALTIME_COARSE, &now); - pthread_cond_destroy(&req->cond); - pthread_mutex_destroy(&req->lock); + attr->mtime = now.tv_sec; - if (req->key != NULL) - free(req->key); + entry = strstr(path, RIB_SEPARATOR) + 1; - free(req); + if (strcmp(entry, "database") == 0) { + attr->size = dht_db_file_len(); + } else if (strcmp(entry, "stats") == 0) { + attr->size = 545; + } + + return 0; } -static void kad_req_destroy(struct kad_req * req) -{ - assert(req); +static struct rib_ops r_ops = { + .read = dht_rib_read, + .readdir = dht_rib_readdir, + .getattr = dht_rib_getattr +}; - pthread_mutex_lock(&req->lock); +/* Helper functions */ - switch (req->state) { - case REQ_DESTROY: - pthread_mutex_unlock(&req->lock); - return; - case REQ_PENDING: - req->state = REQ_DESTROY; - pthread_cond_broadcast(&req->cond); - break; - case REQ_INIT: - case REQ_DONE: - req->state = REQ_NULL; - break; - case REQ_RESPONSE: - case REQ_NULL: - default: - break; +static uint8_t * generate_id(void) +{ + uint8_t * id; + + if(dht.id.len < sizeof(uint64_t)) { + log_err("DHT ID length is too short (%zu < %zu).", + dht.id.len, sizeof(uint64_t)); + return NULL; } - pthread_cleanup_push(cancel_req_destroy, req); + id = malloc(dht.id.len); + if (id == NULL) { + log_err("Failed to malloc ID."); + goto fail_id; + } - while (req->state != REQ_NULL && req->state != REQ_DONE) - pthread_cond_wait(&req->cond, &req->lock); + if (random_buffer(id, dht.id.len) < 0) { + log_err("Failed to generate random ID."); + goto fail_rnd; + } - pthread_cleanup_pop(true); + return id; + fail_rnd: + free(id); + fail_id: + return NULL; } -static int kad_req_wait(struct kad_req * req, - time_t t) +static uint64_t generate_cookie(void) { - struct timespec timeo = TIMESPEC_INIT_S(0); - struct timespec abs; - int ret = 0; + uint64_t cookie = DHT_INVALID; - assert(req); + while (cookie == DHT_INVALID) + random_buffer((uint8_t *) &cookie, sizeof(cookie)); - timeo.tv_sec = t; + return cookie; +} - clock_gettime(PTHREAD_COND_CLOCK, &abs); +/* + * If someone builds a network where the n (n > k) closest nodes all + * have IDs starting with the same 64 bits: by all means, change this. + */ +static uint64_t dist(const uint8_t * src, + const uint8_t * dst) +{ + assert(dht.id.len >= sizeof(uint64_t)); - ts_add(&abs, &timeo, &abs); + return betoh64(*((uint64_t *) src) ^ *((uint64_t *) dst)); +} - pthread_mutex_lock(&req->lock); +#define IS_CLOSER(x, y) (dist((x), dht.id.data) < dist((y), dht.id.data)) - req->state = REQ_PENDING; +static int addr_to_buf(const uint64_t addr, + buffer_t * buf) +{ + size_t len; + uint64_t _addr; - pthread_cleanup_push(__cleanup_mutex_unlock, &req->lock); + len = sizeof(addr); + _addr = hton64(addr); - while (req->state == REQ_PENDING && ret != -ETIMEDOUT) - ret = -pthread_cond_timedwait(&req->cond, &req->lock, &abs); + assert(buf != NULL); - switch(req->state) { - case REQ_DESTROY: - ret = -1; - req->state = REQ_NULL; - pthread_cond_signal(&req->cond); - break; - case REQ_PENDING: /* ETIMEDOUT */ - case REQ_RESPONSE: - req->state = REQ_DONE; - pthread_cond_broadcast(&req->cond); - break; - default: - break; - } + buf->data = malloc(len); + if (buf->data == NULL) + goto fail_malloc; - pthread_cleanup_pop(true); + buf->len = sizeof(_addr); + memcpy(buf->data, &_addr, sizeof(_addr)); - return ret; + return 0; + fail_malloc: + return -ENOMEM; } -static void kad_req_respond(struct kad_req * req) +static int buf_to_addr(const buffer_t buf, + uint64_t * addr) { - pthread_mutex_lock(&req->lock); + assert(addr != NULL); + assert(buf.data != NULL); - req->state = REQ_RESPONSE; - pthread_cond_signal(&req->cond); + if (buf.len != sizeof(*addr)) + return - EINVAL; - pthread_mutex_unlock(&req->lock); -} + *addr = ntoh64(*((uint64_t *) buf.data)); -static struct contact * contact_create(const uint8_t * id, - size_t len, - uint64_t addr) -{ - struct contact * c; - struct timespec t; + if (*addr == dht.addr) + *addr = INVALID_ADDR; - c = malloc(sizeof(*c)); - if (c == NULL) - return NULL; + return 0; +} - list_head_init(&c->next); +static uint8_t * dht_dup_key(const uint8_t * key) +{ + uint8_t * dup; - clock_gettime(CLOCK_REALTIME_COARSE, &t); + assert(key != NULL); + assert(dht.id.len != 0); - c->addr = addr; - c->fails = 0; - c->t_seen = t.tv_sec; - c->id = dht_dup_key(id, len); - if (c->id == NULL) { - free(c); + dup = malloc(dht.id.len); + if (dup == NULL) return NULL; - } - return c; + memcpy(dup, key, dht.id.len); + + return dup; } -static void contact_destroy(struct contact * c) +/* DHT */ + +static struct val_entry * val_entry_create(const buffer_t val, + time_t exp) { - if (c != NULL) - free(c->id); + struct val_entry * e; + struct timespec now; - free(c); -} + assert(val.data != NULL); + assert(val.len > 0); -static struct bucket * iter_bucket(struct bucket * b, - const uint8_t * id) -{ - uint8_t byte; - uint8_t mask; + clock_gettime(CLOCK_REALTIME_COARSE, &now); - assert(b); +#ifndef __DHT_TEST_ALLOW_EXPIRED__ + if (exp < now.tv_sec) + return NULL; /* Refuse to add expired values */ +#endif + e = malloc(sizeof(*e)); + if (e == NULL) + goto fail_entry; - if (b->children[0] == NULL) - return b; + list_head_init(&e->next); - byte = id[(b->depth * KAD_BETA) / CHAR_BIT]; + e->val.len = val.len; + e->val.data = malloc(val.len); + if (e->val.data == NULL) + goto fail_val; - mask = ((1L << KAD_BETA) - 1) & 0xFF; + memcpy(e->val.data, val.data, val.len); - byte >>= (CHAR_BIT - KAD_BETA) - - (((b->depth) * KAD_BETA) & (CHAR_BIT - 1)); + e->t_repl = 0; + e->t_exp = exp; - return iter_bucket(b->children[(byte & mask)], id); + return e; + + fail_val: + free(e); + fail_entry: + return NULL; } -static struct bucket * dht_get_bucket(struct dht * dht, - const uint8_t * id) +static void val_entry_destroy(struct val_entry * v) { - assert(dht->buckets); + assert(v->val.data != NULL); - return iter_bucket(dht->buckets, id); + freebuf(v->val); + free(v); } -/* - * If someone builds a network where the n (n > k) closest nodes all - * have IDs starting with the same 64 bits: by all means, change this. - */ -static uint64_t dist(const uint8_t * src, - const uint8_t * dst) +static struct dht_entry * dht_entry_create(const uint8_t * key) { - return betoh64(*((uint64_t *) src) ^ *((uint64_t *) dst)); + struct dht_entry * e; + + assert(key != NULL); + + e = malloc(sizeof(*e)); + if (e == NULL) + goto fail_entry; + + list_head_init(&e->next); + llist_init(&e->vals); + llist_init(&e->lvals); + + e->key = dht_dup_key(key); + if (e->key == NULL) + goto fail_key; + + return e; + fail_key: + free(e); + fail_entry: + return NULL; } -static size_t list_add_sorted(struct list_head * l, - struct contact * c, - const uint8_t * key) +static void dht_entry_destroy(struct dht_entry * e) { struct list_head * p; + struct list_head * h; - assert(l); - assert(c); - assert(key); - assert(c->id); + assert(e != NULL); - list_for_each(p, l) { - struct contact * e = list_entry(p, struct contact, next); - if (dist(c->id, key) > dist(e->id, key)) - break; + llist_for_each_safe(p, h, &e->vals) { + struct val_entry * v = list_entry(p, struct val_entry, next); + llist_del(&v->next, &e->vals); + val_entry_destroy(v); + --dht.db.kv.vals; } - list_add_tail(&c->next, p); + llist_for_each_safe(p, h, &e->lvals) { + struct val_entry * v = list_entry(p, struct val_entry, next); + llist_del(&v->next, &e->lvals); + val_entry_destroy(v); + --dht.db.kv.lvals; + } + + free(e->key); + + assert(llist_is_empty(&e->vals) && llist_is_empty(&e->lvals)); - return 1; + free(e); } -static size_t dht_contact_list(struct dht * dht, - struct list_head * l, - const uint8_t * key) +static struct val_entry * dht_entry_get_lval(const struct dht_entry * e, + const buffer_t val) { struct list_head * p; - struct bucket * b; - size_t len = 0; - size_t i; - struct timespec t; - assert(l); - assert(dht); - assert(key); - assert(list_is_empty(l)); - - clock_gettime(CLOCK_REALTIME_COARSE, &t); - - b = dht_get_bucket(dht, key); - if (b == NULL) - return 0; + assert(e != NULL); + assert(val.data != NULL); + assert(val.len > 0); - b->t_refr = t.tv_sec + KAD_T_REFR; - - if (b->n_contacts == dht->k || b->parent == NULL) { - list_for_each(p, &b->contacts) { - struct contact * c; - c = list_entry(p, struct contact, next); - c = contact_create(c->id, dht->b, c->addr); - if (list_add_sorted(l, c, key) == 1) - if (++len == dht->k) - break; - } - } else { - struct bucket * d = b->parent; - for (i = 0; i < (1L << KAD_BETA) && len < dht->k; ++i) { - list_for_each(p, &d->children[i]->contacts) { - struct contact * c; - c = list_entry(p, struct contact, next); - c = contact_create(c->id, dht->b, c->addr); - if (c == NULL) - continue; - if (list_add_sorted(l, c, key) == 1) - if (++len == dht->k) - break; - } - } + llist_for_each(p, &e->lvals) { + struct val_entry * v = list_entry(p, struct val_entry, next); + if (bufcmp(&v->val, &val) == 0) + return v; } - assert(len == dht->k || b->parent == NULL); - - return len; + return NULL; } -static struct lookup * lookup_create(struct dht * dht, - const uint8_t * id) +static struct val_entry * dht_entry_get_val(const struct dht_entry * e, + const buffer_t val) { - struct lookup * lu; - pthread_condattr_t cattr; + struct list_head * p; - assert(dht); - assert(id); + assert(e != NULL); + assert(val.data != NULL); + assert(val.len > 0); - lu = malloc(sizeof(*lu)); - if (lu == NULL) - goto fail_malloc; + llist_for_each(p, &e->vals) { + struct val_entry * v = list_entry(p, struct val_entry, next); + if (bufcmp(&v->val, &val) == 0) + return v; - list_head_init(&lu->contacts); - list_head_init(&lu->cookies); + } - lu->state = LU_INIT; - lu->addrs = NULL; - lu->n_addrs = 0; - lu->key = dht_dup_key(id, dht->b); - if (lu->key == NULL) - goto fail_id; + return NULL; +} - if (pthread_mutex_init(&lu->lock, NULL)) - goto fail_mutex; +static int dht_entry_update_val(struct dht_entry * e, + buffer_t val, + time_t exp) +{ + struct val_entry * v; + struct timespec now; - pthread_condattr_init(&cattr); -#ifndef __APPLE__ - pthread_condattr_setclock(&cattr, PTHREAD_COND_CLOCK); -#endif + assert(e != NULL); + assert(val.data != NULL); + assert(val.len > 0); - if (pthread_cond_init(&lu->cond, &cattr)) - goto fail_cond; + clock_gettime(CLOCK_REALTIME_COARSE, &now); - pthread_condattr_destroy(&cattr); + if (exp < now.tv_sec) + return -EINVAL; /* Refuse to add expired values */ - pthread_rwlock_wrlock(&dht->lock); + if (dht_entry_get_lval(e, val) != NULL) { + log_dbg(KV_FMT " Val already in lvals.", KV_VAL(e->key, val)); + return 0; /* Refuse to add local values */ + } - list_add(&lu->next, &dht->lookups); + v = dht_entry_get_val(e, val); + if (v == NULL) { + v = val_entry_create(val, exp); + if (v == NULL) + return -ENOMEM; - lu->n_contacts = dht_contact_list(dht, &lu->contacts, id); + llist_add_tail(&v->next, &e->vals); + ++dht.db.kv.vals; - pthread_rwlock_unlock(&dht->lock); + return 0; + } - return lu; + if (v->t_exp < exp) + v->t_exp = exp; - fail_cond: - pthread_condattr_destroy(&cattr); - pthread_mutex_destroy(&lu->lock); - fail_mutex: - free(lu->key); - fail_id: - free(lu); - fail_malloc: - return NULL; + return 0; } -static void cancel_lookup_destroy(void * o) +static int dht_entry_update_lval(struct dht_entry * e, + buffer_t val) { - struct lookup * lu; - struct list_head * p; - struct list_head * h; + struct val_entry * v; + struct timespec now; - lu = (struct lookup *) o; + assert(e != NULL); + assert(val.data != NULL); + assert(val.len > 0); - if (lu->key != NULL) - free(lu->key); - if (lu->addrs != NULL) - free(lu->addrs); + clock_gettime(CLOCK_REALTIME_COARSE, &now); - list_for_each_safe(p, h, &lu->contacts) { - struct contact * c = list_entry(p, struct contact, next); - list_del(&c->next); - contact_destroy(c); - } - - list_for_each_safe(p, h, &lu->cookies) { - struct cookie_el * c = list_entry(p, struct cookie_el, next); - list_del(&c->next); - free(c); - } + v = dht_entry_get_lval(e, val); + if (v == NULL) { + log_dbg(KV_FMT " Adding lval.", KV_VAL(e->key, val)); + v = val_entry_create(val, now.tv_sec + dht.t_exp); + if (v == NULL) + return -ENOMEM; - pthread_mutex_unlock(&lu->lock); + llist_add_tail(&v->next, &e->lvals); + ++dht.db.kv.lvals; - pthread_mutex_destroy(&lu->lock); + return 0; + } - free(lu); + return 0; } -static void lookup_destroy(struct lookup * lu) +static int dht_entry_remove_lval(struct dht_entry * e, + buffer_t val) { - assert(lu); + struct val_entry * v; - pthread_mutex_lock(&lu->lock); + assert(e != NULL); + assert(val.data != NULL); + assert(val.len > 0); - switch (lu->state) { - case LU_DESTROY: - pthread_mutex_unlock(&lu->lock); - return; - case LU_PENDING: - lu->state = LU_DESTROY; - pthread_cond_broadcast(&lu->cond); - break; - case LU_INIT: - case LU_UPDATE: - case LU_COMPLETE: - lu->state = LU_NULL; - break; - case LU_NULL: - default: - break; - } + v = dht_entry_get_lval(e, val); + if (v == NULL) + return -ENOENT; - pthread_cleanup_push(cancel_lookup_destroy, lu); + log_dbg(KV_FMT " Removing lval.", KV_VAL(e->key, val)); - while (lu->state != LU_NULL) - pthread_cond_wait(&lu->cond, &lu->lock); + llist_del(&v->next, &e->lvals); + val_entry_destroy(v); + --dht.db.kv.lvals; - pthread_cleanup_pop(true); + return 0; } -static void lookup_update(struct dht * dht, - struct lookup * lu, - dht_msg_t * msg) +#define IS_EXPIRED(v, now) ((now)->tv_sec > (v)->t_exp) +static void dht_entry_remove_expired_vals(struct dht_entry * e) { - struct list_head * p = NULL; + struct list_head * p; struct list_head * h; - struct contact * c = NULL; - size_t n; - size_t pos = 0; - bool mod = false; + struct timespec now; - assert(lu); - assert(msg); + assert(e != NULL); - if (dht_get_state(dht) != DHT_RUNNING) - return; + clock_gettime(CLOCK_REALTIME_COARSE, &now); - pthread_mutex_lock(&lu->lock); + llist_for_each_safe(p, h, &e->vals) { + struct val_entry * v = list_entry(p, struct val_entry, next); + if (!IS_EXPIRED(v, &now)) + continue; - list_for_each_safe(p, h, &lu->cookies) { - struct cookie_el * e = list_entry(p, struct cookie_el, next); - if (e->cookie == msg->cookie) { - list_del(&e->next); - free(e); - break; - } + log_dbg(KV_FMT " Value expired." , KV_VAL(e->key, v->val)); + llist_del(&v->next, &e->vals); + val_entry_destroy(v); + --dht.db.kv.vals; } +} - if (lu->state == LU_COMPLETE) { - pthread_mutex_unlock(&lu->lock); - return; - } +static struct dht_entry * __dht_kv_find_entry(const uint8_t * key) +{ + struct list_head * p; - if (msg->n_addrs > 0) { - if (lu->addrs == NULL) { - lu->addrs = malloc(sizeof(*lu->addrs) * msg->n_addrs); - for (n = 0; n < msg->n_addrs; ++n) - lu->addrs[n] = msg->addrs[n]; - lu->n_addrs = msg->n_addrs; - } + assert(key != NULL); - lu->state = LU_COMPLETE; - pthread_cond_broadcast(&lu->cond); - pthread_mutex_unlock(&lu->lock); - return; + llist_for_each(p, &dht.db.kv.ll) { + struct dht_entry * e = list_entry(p, struct dht_entry, next); + if (!memcmp(key, e->key, dht.id.len)) + return e; } - pthread_cleanup_push(__cleanup_mutex_unlock, &lu->lock); + return NULL; +} - while (lu->state == LU_INIT) { - pthread_rwlock_unlock(&dht->lock); - pthread_cond_wait(&lu->cond, &lu->lock); - pthread_rwlock_rdlock(&dht->lock); - } +static void dht_kv_remove_expired_entries(void) +{ + struct list_head * p; + struct list_head * h; + struct timespec now; - pthread_cleanup_pop(false); + clock_gettime(CLOCK_REALTIME_COARSE, &now); - for (n = 0; n < msg->n_contacts; ++n) { - c = contact_create(msg->contacts[n]->id.data, - dht->b, msg->contacts[n]->addr); - if (c == NULL) + pthread_rwlock_wrlock(&dht.db.lock); + + llist_for_each_safe(p, h, &dht.db.kv.ll) { + struct dht_entry * e = list_entry(p, struct dht_entry, next); + dht_entry_remove_expired_vals(e); + if (e->lvals.len > 0 || e->vals.len > 0) continue; - pos = 0; + log_dbg(KEY_FMT " Entry removed. ", KEY_VAL(e->key)); + llist_del(&e->next, &dht.db.kv.ll); + dht_entry_destroy(e); + } - list_for_each(p, &lu->contacts) { - struct contact * e; - e = list_entry(p, struct contact, next); - if (!memcmp(e->id, c->id, dht->b)) { - contact_destroy(c); - c = NULL; - break; - } + pthread_rwlock_unlock(&dht.db.lock); +} - if (dist(c->id, lu->key) > dist(e->id, lu->key)) - break; - pos++; - } +static struct contact * contact_create(const uint8_t * id, + uint64_t addr) +{ + struct contact * c; + struct timespec t; - if (c == NULL) - continue; + c = malloc(sizeof(*c)); + if (c == NULL) + return NULL; - if (lu->n_contacts < dht->k) { - list_add_tail(&c->next, p); - ++lu->n_contacts; - mod = true; - } else if (pos == dht->k) { - contact_destroy(c); - } else { - struct contact * d; - list_add_tail(&c->next, p); - d = list_last_entry(&lu->contacts, - struct contact, next); - list_del(&d->next); - assert(lu->contacts.prv != &d->next); - contact_destroy(d); - mod = true; - } + list_head_init(&c->next); + + clock_gettime(CLOCK_REALTIME_COARSE, &t); + + c->addr = addr; + c->fails = 0; + c->t_seen = t.tv_sec; + c->id = dht_dup_key(id); + if (c->id == NULL) { + free(c); + return NULL; } - if (list_is_empty(&lu->cookies) && !mod) - lu->state = LU_COMPLETE; - else - lu->state = LU_UPDATE; + return c; +} - pthread_cond_broadcast(&lu->cond); - pthread_mutex_unlock(&lu->lock); - return; +static void contact_destroy(struct contact * c) +{ + assert(c != NULL); + assert(list_is_empty(&c->next)); + + free(c->id); + free(c); } -static ssize_t lookup_get_addrs(struct lookup * lu, - uint64_t * addrs) +static struct dht_req * dht_req_create(const uint8_t * key) { - ssize_t n; + struct dht_req * req; + struct timespec now; - assert(lu); + assert(key != NULL); - pthread_mutex_lock(&lu->lock); + clock_gettime(PTHREAD_COND_CLOCK, &now); - for (n = 0; (size_t) n < lu->n_addrs; ++n) - addrs[n] = lu->addrs[n]; + req = malloc(sizeof(*req)); + if (req == NULL) + goto fail_malloc; - assert((size_t) n == lu->n_addrs); + list_head_init(&req->next); - pthread_mutex_unlock(&lu->lock); + req->t_exp = now.tv_sec + DHT_T_RESP; - return n; + llist_init(&req->peers); + + req->key = dht_dup_key(key); + if (req->key == NULL) + goto fail_dup_key; + + llist_init(&req->cache); + + return req; + + fail_dup_key: + free(req); + fail_malloc: + return NULL; } -static ssize_t lookup_contact_addrs(struct lookup * lu, - uint64_t * addrs) +static void dht_req_destroy(struct dht_req * req) { struct list_head * p; - ssize_t n = 0; + struct list_head * h; - assert(lu); - assert(addrs); + assert(req); + assert(req->key); - pthread_mutex_lock(&lu->lock); + llist_for_each_safe(p, h, &req->peers) { + struct peer_entry * e = list_entry(p, struct peer_entry, next); + llist_del(&e->next, &req->peers); + free(e->id); + free(e); + } - list_for_each(p, &lu->contacts) { - struct contact * c = list_entry(p, struct contact, next); - addrs[n] = c->addr; - n++; + llist_for_each_safe(p, h, &req->cache) { + struct val_entry * e = list_entry(p, struct val_entry, next); + llist_del(&e->next, &req->cache); + val_entry_destroy(e); } - pthread_mutex_unlock(&lu->lock); + free(req->key); - return n; + assert(llist_is_empty(&req->peers)); + + free(req); } -static void lookup_new_addrs(struct lookup * lu, - uint64_t * addrs) +static struct peer_entry * dht_req_get_peer(struct dht_req * req, + const struct peer_entry * e) { struct list_head * p; - size_t n = 0; - assert(lu); - assert(addrs); + llist_for_each(p, &req->peers) { + struct peer_entry * x = list_entry(p, struct peer_entry, next); + if (x->addr == e->addr) + return x; + } - pthread_mutex_lock(&lu->lock); + return NULL; +} - /* Uses fails to check if the contact has been contacted. */ - list_for_each(p, &lu->contacts) { - struct contact * c = list_entry(p, struct contact, next); - if (c->fails == 0) { - c->fails = 1; - addrs[n] = c->addr; - n++; +#define IS_MAGIC(peer) ((peer)->cookie == dht.magic) +static int dht_req_add_peer(struct dht_req * req, + const struct peer_entry * e) +{ + struct peer_entry * x; /* existing */ + struct list_head * p; /* iterator */ + size_t pos = 0; + + assert(req != NULL); + assert(e != NULL); + assert(e->id != NULL); + + /* + * Dedupe messages to the same peer, unless + * 1) The previous was FIND_NODE and now it's FIND_VALUE + * 2) We urgently need contacts (magic cookie) + */ + x = dht_req_get_peer(req, e); + if (x != NULL && x->code >= e->code && !IS_MAGIC(e)) + return -1; + + /* Find how this contact ranks in distance to the key */ + llist_for_each(p, &req->peers) { + struct peer_entry * y; + y = list_entry(p, struct peer_entry, next); + if (IS_CLOSER(y->id, e->id)) { + pos++; + continue; } + break; + } - if (n == KAD_ALPHA) - break; + if (pos >= dht.alpha && IS_MAGIC(e)) + return -1; + + x = malloc(sizeof(*x)); + if (x == NULL) { + log_err("Failed to malloc peer entry."); + return -1; } - assert(n <= KAD_ALPHA); + x->cookie = e->cookie; + x->addr = e->addr; + x->code = e->code; + x->t_sent = e->t_sent; + x->id = dht_dup_key(e->id); + if (x->id == NULL) { + log_err("Failed to dup peer ID."); + free(x); + return -1; + } - addrs[n] = 0; + if (IS_MAGIC(e)) + llist_add_at(&x->next, p, &req->peers); + else + llist_add_tail_at(&x->next, p, &req->peers); - pthread_mutex_unlock(&lu->lock); + return 0; } -static void lookup_set_state(struct lookup * lu, - enum lookup_state state) +static size_t dht_req_add_peers(struct dht_req * req, + struct list_head * pl) { - pthread_mutex_lock(&lu->lock); + struct list_head * p; + struct list_head * h; + size_t n = 0; + + assert(req != NULL); + assert(pl != NULL); - lu->state = state; - pthread_cond_broadcast(&lu->cond); + list_for_each_safe(p, h, pl) { + struct peer_entry * e; + e = list_entry(p, struct peer_entry, next); + if (dht_req_add_peer(req, e) < 0) { + list_del(&e->next); + free(e->id); + free(e); + } + } - pthread_mutex_unlock(&lu->lock); + return n; } -static void cancel_lookup_wait(void * o) +static bool dht_req_has_peer(struct dht_req * req, + uint64_t cookie) { - struct lookup * lu = (struct lookup *) o; - lu->state = LU_NULL; - pthread_mutex_unlock(&lu->lock); - lookup_destroy(lu); + struct list_head * p; + + assert(req != NULL); + + llist_for_each(p, &req->peers) { + struct peer_entry * e = list_entry(p, struct peer_entry, next); + if (e->cookie == cookie) + return true; + } + + return false; } -static enum lookup_state lookup_wait(struct lookup * lu) +static void peer_list_destroy(struct list_head * pl) { - struct timespec timeo = TIMESPEC_INIT_S(KAD_T_RESP); - struct timespec abs; - enum lookup_state state; - int ret = 0; + struct list_head * p; + struct list_head * h; - clock_gettime(PTHREAD_COND_CLOCK, &abs); + assert(pl != NULL); - ts_add(&abs, &timeo, &abs); + list_for_each_safe(p, h, pl) { + struct peer_entry * e = list_entry(p, struct peer_entry, next); + list_del(&e->next); + free(e->id); + free(e); + } +} - pthread_mutex_lock(&lu->lock); +static int dht_kv_create_peer_list(struct list_head * cl, + struct list_head * pl, + enum dht_code code) +{ + struct list_head * p; + struct list_head * h; + struct timespec now; + size_t len; - if (lu->state == LU_INIT || lu->state == LU_UPDATE) - lu->state = LU_PENDING; + assert(cl != NULL); + assert(pl != NULL); + assert(list_is_empty(pl)); - pthread_cleanup_push(cancel_lookup_wait, lu); + clock_gettime(CLOCK_REALTIME_COARSE, &now); - while (lu->state == LU_PENDING && ret != -ETIMEDOUT) - ret = -pthread_cond_timedwait(&lu->cond, &lu->lock, &abs); + len = 0; - pthread_cleanup_pop(false); + list_for_each_safe(p, h, cl) { + struct contact * c = list_entry(p, struct contact, next); + struct peer_entry * e; + if (len++ == dht.alpha) + break; + + e = malloc(sizeof(*e)); + if (e == NULL) + return -ENOMEM; - if (ret == -ETIMEDOUT) - lu->state = LU_COMPLETE; + e->cookie = generate_cookie(); + e->code = code; + e->addr = c->addr; + e->t_sent = now.tv_sec; - state = lu->state; + e->id = c->id; - pthread_mutex_unlock(&lu->lock); + list_add_tail(&e->next, pl); - return state; + list_del(&c->next); + c->id = NULL; /* we stole the id */ + contact_destroy(c); + } + + return 0; } -static struct kad_req * dht_find_request(struct dht * dht, - dht_msg_t * msg) +static struct dht_req * __dht_kv_req_get_req(const uint8_t * key) { struct list_head * p; - assert(dht); - assert(msg); - - list_for_each(p, &dht->requests) { - struct kad_req * r = list_entry(p, struct kad_req, next); - if (r->cookie == msg->cookie) + llist_for_each(p, &dht.reqs.ll) { + struct dht_req * r = list_entry(p, struct dht_req, next); + if (memcmp(r->key, key, dht.id.len) == 0) return r; } return NULL; } -static struct lookup * dht_find_lookup(struct dht * dht, - uint32_t cookie) +static struct dht_req * __dht_kv_get_req_cache(const uint8_t * key) { - struct list_head * p; - struct list_head * p2; - struct list_head * h2; - - assert(dht); - assert(cookie > 0); - - list_for_each(p, &dht->lookups) { - struct lookup * l = list_entry(p, struct lookup, next); - pthread_mutex_lock(&l->lock); - list_for_each_safe(p2, h2, &l->cookies) { - struct cookie_el * e; - e = list_entry(p2, struct cookie_el, next); - if (e->cookie == cookie) { - list_del(&e->next); - free(e); - pthread_mutex_unlock(&l->lock); - return l; - } - } - pthread_mutex_unlock(&l->lock); - } + struct dht_req * req; - return NULL; + assert(key != NULL); + + req = __dht_kv_req_get_req(key); + if (req == NULL) + return NULL; + + if (llist_is_empty(&req->cache)) + return NULL; + + return req; } -static struct val * val_create(uint64_t addr, - time_t exp) +static void __dht_kv_req_remove(const uint8_t * key) { - struct val * v; - struct timespec t; + struct dht_req * req; - v = malloc(sizeof(*v)); - if (v == NULL) - return NULL; + assert(key != NULL); - list_head_init(&v->next); - v->addr = addr; + req = __dht_kv_req_get_req(key); + if (req == NULL) + return; - clock_gettime(CLOCK_REALTIME_COARSE, &t); + llist_del(&req->next, &dht.reqs.ll); + + dht_req_destroy(req); +} + +static struct dht_req * __dht_kv_get_req_peer(const uint8_t * key, + uint64_t cookie) +{ + struct dht_req * req; - v->t_exp = t.tv_sec + exp; - v->t_rep = t.tv_sec + KAD_T_REPL; + assert(key != NULL); - return v; + req = __dht_kv_req_get_req(key); + if (req == NULL) + return NULL; + + if (!dht_req_has_peer(req, cookie)) + return NULL; + + return req; } -static void val_destroy(struct val * v) +static bool dht_kv_has_req(const uint8_t * key, + uint64_t cookie) { - assert(v); + bool found; - free(v); + pthread_mutex_lock(&dht.reqs.mtx); + + found = __dht_kv_get_req_peer(key, cookie) != NULL; + + pthread_mutex_unlock(&dht.reqs.mtx); + + return found; } -static struct ref_entry * ref_entry_create(struct dht * dht, - const uint8_t * key) +/* + * This will filter the peer list for addresses that still need to be + * contacted. + */ +static int dht_kv_update_req(const uint8_t * key, + struct list_head * pl) { - struct ref_entry * e; - struct timespec t; + struct dht_req * req; + struct timespec now; - assert(dht); - assert(key); + assert(key != NULL); + assert(pl != NULL); + assert(!list_is_empty(pl)); - e = malloc(sizeof(*e)); - if (e == NULL) - return NULL; + clock_gettime(PTHREAD_COND_CLOCK, &now); - e->key = dht_dup_key(key, dht->b); - if (e->key == NULL) { - free(e); - return NULL; + pthread_mutex_lock(&dht.reqs.mtx); + + req = __dht_kv_req_get_req(key); + if (req == NULL) { + if (dht.reqs.ll.len == DHT_MAX_REQS) { + log_err(KEY_FMT " Max reqs reached (%zu).", + KEY_VAL(key), dht.reqs.ll.len); + peer_list_destroy(pl); + goto fail_req; + } + req = dht_req_create(key); + if (req == NULL) { + log_err(KEY_FMT "Failed to create req.", KEY_VAL(key)); + goto fail_req; + } + llist_add_tail(&req->next, &dht.reqs.ll); } - clock_gettime(CLOCK_REALTIME_COARSE, &t); + if (req->cache.len > 0) /* Already have values */ + peer_list_destroy(pl); - e->t_rep = t.tv_sec + dht->t_repub; + dht_req_add_peers(req, pl); + req->t_exp = now.tv_sec + DHT_T_RESP; - return e; + if (dht.reqs.ll.len > DHT_WARN_REQS) { + log_warn("Number of outstanding requests (%zu) exceeds %u.", + dht.reqs.ll.len, DHT_WARN_REQS); + } + + pthread_mutex_unlock(&dht.reqs.mtx); + + return 0; + fail_req: + pthread_mutex_unlock(&dht.reqs.mtx); + return -1; } -static void ref_entry_destroy(struct ref_entry * e) +static int dht_kv_respond_req(uint8_t * key, + binary_data_t * vals, + size_t len) { - free(e->key); - free(e); + struct dht_req * req; + struct timespec now; + size_t i; + + assert(key != NULL); + assert(vals != NULL); + assert(len > 0); + + clock_gettime(CLOCK_REALTIME_COARSE, &now); + + pthread_mutex_lock(&dht.reqs.mtx); + + req = __dht_kv_req_get_req(key); + if (req == NULL) { + log_dbg(KEY_FMT " Failed to find req.", KEY_VAL(key)); + goto fail_req; + } + + for (i = 0; i < len; ++i) { + struct val_entry * e; + buffer_t val; + val.data = vals[i].data; + val.len = vals[i].len; + e = val_entry_create(val, now.tv_sec + DHT_T_CACHE); + if (e == NULL) { + log_err(" Failed to create val_entry."); + continue; + } + + llist_add_tail(&e->next, &req->cache); + } + + pthread_cond_broadcast(&dht.reqs.cond); + + pthread_mutex_unlock(&dht.reqs.mtx); + + return 0; + fail_req: + pthread_mutex_unlock(&dht.reqs.mtx); + return -1; } -static struct dht_entry * dht_entry_create(struct dht * dht, - const uint8_t * key) +static ssize_t dht_kv_wait_req(const uint8_t * key, + buffer_t ** vals) { - struct dht_entry * e; + struct list_head * p; + struct dht_req * req; + struct timespec t; +#ifdef __DHT_TEST__ + struct timespec intv = TIMESPEC_INIT_MS(10); +#else + struct timespec intv = TIMESPEC_INIT_S(DHT_T_RESP); +#endif + size_t max; + size_t i = 0; + int ret = 0; - assert(dht); - assert(key); + assert(key != NULL); + assert(vals != NULL); - e = malloc(sizeof(*e)); - if (e == NULL) - return NULL; + clock_gettime(PTHREAD_COND_CLOCK, &t); - list_head_init(&e->next); - list_head_init(&e->vals); + ts_add(&t, &intv, &t); - e->n_vals = 0; + pthread_mutex_lock(&dht.reqs.mtx); - e->key = dht_dup_key(key, dht->b); - if (e->key == NULL) { - free(e); - return NULL; + pthread_cleanup_push(__cleanup_mutex_unlock, &dht.reqs.mtx); + + while ((req = __dht_kv_get_req_cache(key)) == NULL) { + ret = pthread_cond_timedwait(&dht.reqs.cond, &dht.reqs.mtx, &t); + if (ret == ETIMEDOUT) + break; } - return e; + pthread_cleanup_pop(false); + + if (ret == ETIMEDOUT) { + log_warn(KEY_FMT " Req timed out.", KEY_VAL(key)); + __dht_kv_req_remove(key); + goto timedout; + } + + max = MIN(req->cache.len, DHT_MAX_VALS); + if (max == 0) + goto no_vals; + + *vals = malloc(max * sizeof(**vals)); + if (*vals == NULL) { + log_err(KEY_FMT "Failed to malloc val buffer.", KEY_VAL(key)); + goto fail_vals; + } + + memset(*vals, 0, max * sizeof(**vals)); + + llist_for_each(p, &req->cache) { + struct val_entry * v; + if (i == max) + break; /* We have enough values */ + v = list_entry(p, struct val_entry, next); + (*vals)[i].data = malloc(v->val.len); + if ((*vals)[i].data == NULL) + goto fail_val_data; + + (*vals)[i].len = v->val.len; + memcpy((*vals)[i++].data, v->val.data, v->val.len); + } + + pthread_mutex_unlock(&dht.reqs.mtx); + + return i; + no_vals: + pthread_mutex_unlock(&dht.reqs.mtx); + *vals = NULL; + return 0; + fail_val_data: + freebufs(*vals, i); + fail_vals: + pthread_mutex_unlock(&dht.reqs.mtx); + return -ENOMEM; + timedout: + pthread_mutex_unlock(&dht.reqs.mtx); + return -ETIMEDOUT; } -static void dht_entry_destroy(struct dht_entry * e) +static struct bucket * iter_bucket(struct bucket * b, + const uint8_t * id) +{ + uint8_t byte; + uint8_t mask; + + assert(b != NULL); + + if (b->children[0] == NULL) + return b; + + byte = id[(b->depth * DHT_BETA) / CHAR_BIT]; + + mask = ((1L << DHT_BETA) - 1) & 0xFF; + + byte >>= (CHAR_BIT - DHT_BETA) - + (((b->depth) * DHT_BETA) & (CHAR_BIT - 1)); + + return iter_bucket(b->children[(byte & mask)], id); +} + +static struct bucket * __dht_kv_get_bucket(const uint8_t * id) +{ + assert(dht.db.contacts.root != NULL); + + return iter_bucket(dht.db.contacts.root, id); +} + +static void contact_list_add(struct list_head * l, + struct contact * c) { struct list_head * p; - struct list_head * h; - assert(e); + assert(l != NULL); + assert(c != NULL); - list_for_each_safe(p, h, &e->vals) { - struct val * v = list_entry(p, struct val, next); - list_del(&v->next); - val_destroy(v); + list_for_each(p, l) { + struct contact * e = list_entry(p, struct contact, next); + if (IS_CLOSER(e->id, c->id)) + continue; } - free(e->key); - - free(e); + list_add_tail(&c->next, p); } -static int dht_entry_add_addr(struct dht_entry * e, - uint64_t addr, - time_t exp) +static ssize_t dht_kv_contact_list(const uint8_t * key, + struct list_head * l, + size_t max) { struct list_head * p; - struct val * val; - struct timespec t; + struct bucket * b; + struct timespec t; + size_t i; + size_t len = 0; + + assert(l != NULL); + assert(key != NULL); + assert(list_is_empty(l)); clock_gettime(CLOCK_REALTIME_COARSE, &t); - list_for_each(p, &e->vals) { - struct val * v = list_entry(p, struct val, next); - if (v->addr == addr) { - if (v->t_exp < t.tv_sec + exp) { - v->t_exp = t.tv_sec + exp; - v->t_rep = t.tv_sec + KAD_T_REPL; - } + max = MIN(max, dht.k); - return 0; - } + pthread_rwlock_rdlock(&dht.db.lock); + + b = __dht_kv_get_bucket(key); + if (b == NULL) { + log_err(KEY_FMT " Failed to get bucket.", KEY_VAL(key)); + goto fail_bucket; } - val = val_create(addr, exp); - if (val == NULL) - return -ENOMEM; + b->t_refr = t.tv_sec + dht.t_refr; - list_add(&val->next, &e->vals); - ++e->n_vals; + if (b->contacts.len == dht.k || b->parent == NULL) { + llist_for_each(p, &b->contacts) { + struct contact * c; + struct contact * d; + c = list_entry(p, struct contact, next); + if (c->addr == dht.addr) + continue; + d = contact_create(c->id, c->addr); + if (d == NULL) + continue; + contact_list_add(l, d); + if (++len == max) + break; + } + } else { + struct bucket * d = b->parent; + for (i = 0; i < (1L << DHT_BETA) && len < dht.k; ++i) { + llist_for_each(p, &d->children[i]->contacts) { + struct contact * c; + struct contact * d; + c = list_entry(p, struct contact, next); + if (c->addr == dht.addr) + continue; + d = contact_create(c->id, c->addr); + if (d == NULL) + continue; + contact_list_add(l, d); + if (++len == max) + break; + } + } + } - return 0; -} + pthread_rwlock_unlock(&dht.db.lock); + return len; + fail_bucket: + pthread_rwlock_unlock(&dht.db.lock); + return -1; +} -static void dht_entry_del_addr(struct dht_entry * e, - uint64_t addr) +static void contact_list_destroy(struct list_head * l) { struct list_head * p; struct list_head * h; - assert(e); - - list_for_each_safe(p, h, &e->vals) { - struct val * v = list_entry(p, struct val, next); - if (v->addr == addr) { - list_del(&v->next); - val_destroy(v); - --e->n_vals; - } - } + assert(l != NULL); - if (e->n_vals == 0) { - list_del(&e->next); - dht_entry_destroy(e); + list_for_each_safe(p, h, l) { + struct contact * c = list_entry(p, struct contact, next); + list_del(&c->next); + contact_destroy(c); } } -static uint64_t dht_entry_get_addr(struct dht * dht, - struct dht_entry * e) +static ssize_t dht_kv_get_contacts(const uint8_t * key, + dht_contact_msg_t *** msgs) { + struct list_head cl; struct list_head * p; + struct list_head * h; + size_t len; + size_t i = 0; + + assert(key != NULL); + assert(msgs != NULL); - assert(e); - assert(!list_is_empty(&e->vals)); + list_head_init(&cl); - list_for_each(p, &e->vals) { - struct val * v = list_entry(p, struct val, next); - if (v->addr != dht->addr) - return v->addr; + len = dht_kv_contact_list(key, &cl, dht.k); + if (len == 0) { + *msgs = NULL; + return 0; } - return 0; -} + *msgs = malloc(len * sizeof(**msgs)); + if (*msgs == NULL) + goto fail_msgs; + + list_for_each_safe(p, h, &cl) { + struct contact * c; + (*msgs)[i] = malloc(sizeof(***msgs)); + if ((*msgs)[i] == NULL) + goto fail_contact; -/* Forward declaration. */ -static struct lookup * kad_lookup(struct dht * dht, - const uint8_t * key, - enum kad_code code); + dht_contact_msg__init((*msgs)[i]); + c = list_entry(p, struct contact, next); + list_del(&c->next); + (*msgs)[i]->id.data = c->id; + (*msgs)[i]->id.len = dht.id.len; + (*msgs)[i++]->addr = c->addr; + free(c); + } + return i; + fail_contact: + while (i-- > 0) + dht_contact_msg__free_unpacked((*msgs)[i], NULL); + free(*msgs); + *msgs = NULL; + fail_msgs: + contact_list_destroy(&cl); + return -ENOMEM; +} /* Build a refresh list. */ -static void bucket_refresh(struct dht * dht, - struct bucket * b, - time_t t, - struct list_head * r) +static void __dht_kv_bucket_refresh_list(struct bucket * b, + time_t t, + struct list_head * r) { - size_t i; + struct contact * c; + struct contact * d; - if (*b->children != NULL) - for (i = 0; i < (1L << KAD_BETA); ++i) - bucket_refresh(dht, b->children[i], t, r); + assert(b != NULL); - if (b->n_contacts == 0) + if (t < b->t_refr) return; - if (t > b->t_refr) { - struct contact * c; - struct contact * d; - c = list_first_entry(&b->contacts, struct contact, next); - d = contact_create(c->id, dht->b, c->addr); + if (*b->children != NULL) { + size_t i; + for (i = 0; i < (1L << DHT_BETA); ++i) + __dht_kv_bucket_refresh_list(b->children[i], t, r); + } + + if (llist_is_empty(&b->contacts)) + return; + + c = llist_first_entry(&b->contacts, struct contact, next); + if (t > c->t_seen + dht.t_refr) { + d = contact_create(c->id, c->addr); if (d != NULL) list_add(&d->next, r); - return; } } - static struct bucket * bucket_create(void) { struct bucket * b; @@ -1294,20 +1644,19 @@ static struct bucket * bucket_create(void) if (b == NULL) return NULL; - list_head_init(&b->contacts); - b->n_contacts = 0; + llist_init(&b->contacts); - list_head_init(&b->alts); - b->n_alts = 0; + llist_init(&b->alts); clock_gettime(CLOCK_REALTIME_COARSE, &t); - b->t_refr = t.tv_sec + KAD_T_REFR; + b->t_refr = t.tv_sec + dht.t_refr; - for (i = 0; i < (1L << KAD_BETA); ++i) + for (i = 0; i < (1L << DHT_BETA); ++i) b->children[i] = NULL; b->parent = NULL; b->depth = 0; + b->mask = 0; return b; } @@ -1318,24 +1667,22 @@ static void bucket_destroy(struct bucket * b) struct list_head * h; size_t i; - assert(b); + assert(b != NULL); - for (i = 0; i < (1L << KAD_BETA); ++i) + for (i = 0; i < (1L << DHT_BETA); ++i) if (b->children[i] != NULL) bucket_destroy(b->children[i]); - list_for_each_safe(p, h, &b->contacts) { + llist_for_each_safe(p, h, &b->contacts) { struct contact * c = list_entry(p, struct contact, next); - list_del(&c->next); + llist_del(&c->next, &b->contacts); contact_destroy(c); - --b->n_contacts; } - list_for_each_safe(p, h, &b->alts) { + llist_for_each_safe(p, h, &b->alts) { struct contact * c = list_entry(p, struct contact, next); - list_del(&c->next); + llist_del(&c->next, &b->alts); contact_destroy(c); - --b->n_contacts; } free(b); @@ -1350,1534 +1697,2298 @@ static bool bucket_has_id(struct bucket * b, if (b->depth == 0) return true; - byte = id[(b->depth * KAD_BETA) / CHAR_BIT]; + byte = id[(b->depth * DHT_BETA) / CHAR_BIT]; - mask = ((1L << KAD_BETA) - 1) & 0xFF; + mask = ((1L << DHT_BETA) - 1) & 0xFF; - byte >>= (CHAR_BIT - KAD_BETA) - - (((b->depth - 1) * KAD_BETA) & (CHAR_BIT - 1)); + byte >>= (CHAR_BIT - DHT_BETA) - + (((b->depth - 1) * DHT_BETA) & (CHAR_BIT - 1)); return ((byte & mask) == b->mask); } -static int split_bucket(struct bucket * b) +static int move_contacts(struct bucket * b, + struct bucket * c) { struct list_head * p; struct list_head * h; + struct contact * d; + + assert(b != NULL); + assert(c != NULL); + + llist_for_each_safe(p, h, &b->contacts) { + d = list_entry(p, struct contact, next); + if (bucket_has_id(c, d->id)) { + llist_del(&d->next, &b->contacts); + llist_add_tail(&d->next, &c->contacts); + } + } + + return 0; +} + +static int split_bucket(struct bucket * b) +{ uint8_t mask = 0; size_t i; - size_t c; + size_t b_len; assert(b); - assert(b->n_alts == 0); - assert(b->n_contacts); + assert(llist_is_empty(&b->alts)); + assert(!llist_is_empty(&b->contacts)); assert(b->children[0] == NULL); - c = b->n_contacts; + b_len = b->contacts.len; - for (i = 0; i < (1L << KAD_BETA); ++i) { + for (i = 0; i < (1L << DHT_BETA); ++i) { b->children[i] = bucket_create(); - if (b->children[i] == NULL) { - size_t j; - for (j = 0; j < i; ++j) - bucket_destroy(b->children[j]); - return -1; - } + if (b->children[i] == NULL) + goto fail_child; b->children[i]->depth = b->depth + 1; b->children[i]->mask = mask; b->children[i]->parent = b; - list_for_each_safe(p, h, &b->contacts) { - struct contact * c; - c = list_entry(p, struct contact, next); - if (bucket_has_id(b->children[i], c->id)) { - list_del(&c->next); - --b->n_contacts; - list_add(&c->next, &b->children[i]->contacts); - ++b->children[i]->n_contacts; - } - } + move_contacts(b, b->children[i]); mask++; } - for (i = 0; i < (1L << KAD_BETA); ++i) - if (b->children[i]->n_contacts == c) + for (i = 0; i < (1L << DHT_BETA); ++i) + if (b->children[i]->contacts.len == b_len) split_bucket(b->children[i]); return 0; + fail_child: + while (i-- > 0) + bucket_destroy(b->children[i]); + return -1; } -/* Locked externally to mandate update as (final) part of join transaction. */ -static int dht_update_bucket(struct dht * dht, - const uint8_t * id, - uint64_t addr) +static int dht_kv_update_contacts(const uint8_t * id, + uint64_t addr) { struct list_head * p; struct list_head * h; struct bucket * b; struct contact * c; - assert(dht); + assert(id != NULL); + assert(addr != INVALID_ADDR); - b = dht_get_bucket(dht, id); - if (b == NULL) - return -1; + pthread_rwlock_wrlock(&dht.db.lock); - c = contact_create(id, dht->b, addr); - if (c == NULL) - return -1; + b = __dht_kv_get_bucket(id); + if (b == NULL) { + log_err(PEER_FMT " Failed to get bucket.", PEER_VAL(id, addr)); + goto fail_update; + } + + c = contact_create(id, addr); + if (c == NULL) { + log_err(PEER_FMT " Failed to create contact.", + PEER_VAL(id, addr)); + goto fail_update; + } - list_for_each_safe(p, h, &b->contacts) { + llist_for_each_safe(p, h, &b->contacts) { struct contact * d = list_entry(p, struct contact, next); if (d->addr == addr) { - list_del(&d->next); + llist_del(&d->next, &b->contacts); contact_destroy(d); - --b->n_contacts; } } - if (b->n_contacts == dht->k) { - if (bucket_has_id(b, dht->id)) { - list_add_tail(&c->next, &b->contacts); - ++b->n_contacts; + if (b->contacts.len == dht.k) { + if (bucket_has_id(b, dht.id.data)) { + llist_add_tail(&c->next, &b->contacts); if (split_bucket(b)) { - list_del(&c->next); + llist_del(&c->next, &b->contacts); contact_destroy(c); - --b->n_contacts; } - } else if (b->n_alts == dht->k) { + } else if (b->alts.len == dht.k) { struct contact * d; - d = list_first_entry(&b->alts, struct contact, next); - list_del(&d->next); + d = llist_first_entry(&b->alts, + struct contact, next); + llist_del(&d->next, &b->alts); contact_destroy(d); - list_add_tail(&c->next, &b->alts); + llist_add_tail(&c->next, &b->alts); } else { - list_add_tail(&c->next, &b->alts); - ++b->n_alts; + llist_add_tail(&c->next, &b->alts); } } else { - list_add_tail(&c->next, &b->contacts); - ++b->n_contacts; + llist_add_tail(&c->next, &b->contacts); } + pthread_rwlock_unlock(&dht.db.lock); + return 0; + fail_update: + pthread_rwlock_unlock(&dht.db.lock); + return -1; } -static int send_msg(struct dht * dht, - dht_msg_t * msg, - uint64_t addr) +static time_t gcd(time_t a, + time_t b) { -#ifndef __DHT_TEST__ - struct shm_du_buff * sdb; - size_t len; -#endif - int retr = 0; + if (a == 0) + return b; - if (msg->code == KAD_RESPONSE) - retr = KAD_RESP_RETR; + return gcd(b % a, a); +} - pthread_rwlock_wrlock(&dht->lock); +static dht_contact_msg_t * dht_kv_src_contact_msg(void) +{ + dht_contact_msg_t * src; - if (dht->id != NULL) { - msg->has_s_id = true; - msg->s_id.data = dht->id; - msg->s_id.len = dht->b; - } + src = malloc(sizeof(*src)); + if (src == NULL) + goto fail_malloc; - msg->s_addr = dht->addr; + dht_contact_msg__init(src); - if (msg->code < KAD_STORE) { - msg->cookie = bmp_allocate(dht->cookies); - if (!bmp_is_id_valid(dht->cookies, msg->cookie)) { - pthread_rwlock_unlock(&dht->lock); - goto fail_bmp_alloc; - } - } + src->id.data = dht_dup_key(dht.id.data); + if (src->id.data == NULL) + goto fail_id; - pthread_rwlock_unlock(&dht->lock); + src->id.len = dht.id.len; + src->addr = dht.addr; -#ifndef __DHT_TEST__ - len = dht_msg__get_packed_size(msg); - if (len == 0) - goto fail_msg; + return src; + fail_id: + dht_contact_msg__free_unpacked(src, NULL); + fail_malloc: + return NULL; +} - while (true) { - if (ipcp_sdb_reserve(&sdb, len)) - goto fail_msg; +static dht_msg_t * dht_kv_find_req_msg(const uint8_t * key, + enum dht_code code) +{ + dht_msg_t * msg; - dht_msg__pack(msg, shm_du_buff_head(sdb)); + assert(key != NULL); - if (dt_write_packet(addr, QOS_CUBE_BE, dht->eid, sdb) == 0) - break; + msg = malloc(sizeof(*msg)); + if (msg == NULL) + goto fail_malloc; - ipcp_sdb_release(sdb); + dht_msg__init(msg); + msg->code = code; - sleep(1); + msg->src = dht_kv_src_contact_msg(); + if (msg->src == NULL) + goto fail_msg; - if (--retr < 0) - goto fail_msg; - } + msg->find = malloc(sizeof(*msg->find)); + if (msg->find == NULL) + goto fail_msg; -#else - (void) addr; - (void) retr; -#endif /* __DHT_TEST__ */ + dht_find_req_msg__init(msg->find); - if (msg->code < KAD_STORE && dht_get_state(dht) != DHT_SHUTDOWN) - kad_req_create(dht, msg, addr); + msg->find->key.data = dht_dup_key(key); + if (msg->find->key.data == NULL) + goto fail_msg; + + msg->find->key.len = dht.id.len; + msg->find->cookie = DHT_INVALID; + + return msg; - return msg->cookie; -#ifndef __DHT_TEST__ fail_msg: - pthread_rwlock_wrlock(&dht->lock); - bmp_release(dht->cookies, msg->cookie); - pthread_rwlock_unlock(&dht->lock); -#endif /* !__DHT_TEST__ */ - fail_bmp_alloc: - return -1; + dht_msg__free_unpacked(msg, NULL); + fail_malloc: + return NULL; } -static struct dht_entry * dht_find_entry(struct dht * dht, - const uint8_t * key) +static dht_msg_t * dht_kv_find_node_req_msg(const uint8_t * key) { - struct list_head * p; + return dht_kv_find_req_msg(key, DHT_FIND_NODE_REQ); +} - list_for_each(p, &dht->entries) { - struct dht_entry * e = list_entry(p, struct dht_entry, next); - if (!memcmp(key, e->key, dht->b)) - return e; +static dht_msg_t * dht_kv_find_value_req_msg(const uint8_t * key) +{ + return dht_kv_find_req_msg(key, DHT_FIND_VALUE_REQ); +} + +static dht_msg_t * dht_kv_find_node_rsp_msg(uint8_t * key, + uint64_t cookie, + dht_contact_msg_t *** contacts, + size_t len) +{ + dht_msg_t * msg; + + msg = malloc(sizeof(*msg)); + if (msg == NULL) + goto fail_malloc; + + dht_msg__init(msg); + msg->code = DHT_FIND_NODE_RSP; + + msg->src = dht_kv_src_contact_msg(); + if (msg->src == NULL) + goto fail_msg; + + msg->node = malloc(sizeof(*msg->node)); + if (msg->node == NULL) + goto fail_msg; + + dht_find_node_rsp_msg__init(msg->node); + + msg->node->key.data = dht_dup_key(key); + if (msg->node->key.data == NULL) + goto fail_msg; + + msg->node->cookie = cookie; + msg->node->key.len = dht.id.len; + msg->node->n_contacts = len; + if (len != 0) { /* Steal the ptr */ + msg->node->contacts = *contacts; + *contacts = NULL; } + return msg; + + fail_msg: + dht_msg__free_unpacked(msg, NULL); + fail_malloc: return NULL; } -static int kad_add(struct dht * dht, - const dht_contact_msg_t * contacts, - ssize_t n, - time_t exp) +static dht_msg_t * dht_kv_find_value_rsp_msg(uint8_t * key, + uint64_t cookie, + dht_contact_msg_t *** contacts, + size_t n_contacts, + buffer_t ** vals, + size_t n_vals) { - struct dht_entry * e; + dht_msg_t * msg; - pthread_rwlock_wrlock(&dht->lock); + msg = dht_kv_find_node_rsp_msg(key, cookie, contacts, n_contacts); + if (msg == NULL) + goto fail_node_rsp; - while (n-- > 0) { - if (contacts[n].id.len != dht->b) - log_warn("Bad key length in contact data."); + msg->code = DHT_FIND_VALUE_RSP; - e = dht_find_entry(dht, contacts[n].id.data); - if (e != NULL) { - if (dht_entry_add_addr(e, contacts[n].addr, exp)) - goto fail; - } else { - e = dht_entry_create(dht, contacts[n].id.data); - if (e == NULL) - goto fail; + msg->val = malloc(sizeof(*msg->val)); + if (msg->val == NULL) + goto fail_msg; - if (dht_entry_add_addr(e, contacts[n].addr, exp)) { - dht_entry_destroy(e); - goto fail; - } + dht_find_value_rsp_msg__init(msg->val); - list_add(&e->next, &dht->entries); - } - } + msg->val->n_values = n_vals; + if (n_vals != 0) /* Steal the ptr */ + msg->val->values = (binary_data_t *) *vals; - pthread_rwlock_unlock(&dht->lock); - return 0; + return msg; - fail: - pthread_rwlock_unlock(&dht->lock); - return -ENOMEM; + fail_msg: + dht_msg__free_unpacked(msg, NULL); + fail_node_rsp: + return NULL; } -static int wait_resp(struct dht * dht, - dht_msg_t * msg, - time_t timeo) +static dht_msg_t * dht_kv_store_msg(const uint8_t * key, + const buffer_t val, + time_t exp) { - struct kad_req * req; + dht_msg_t * msg; - assert(dht); - assert(msg); + assert(key != NULL); + assert(val.data != NULL); + assert(val.len > 0); - pthread_rwlock_rdlock(&dht->lock); + msg = malloc(sizeof(*msg)); + if (msg == NULL) + goto fail_malloc; - req = dht_find_request(dht, msg); - if (req == NULL) { - pthread_rwlock_unlock(&dht->lock); - return -EPERM; - } + dht_msg__init(msg); + + msg->code = DHT_STORE; - pthread_rwlock_unlock(&dht->lock); + msg->src = dht_kv_src_contact_msg(); + if (msg->src == NULL) + goto fail_msg; - return kad_req_wait(req, timeo); + msg->store = malloc(sizeof(*msg->store)); + if (msg->store == NULL) + goto fail_msg; + + dht_store_msg__init(msg->store); + + msg->store->key.data = dht_dup_key(key); + if (msg->store->key.data == NULL) + goto fail_msg; + + msg->store->key.len = dht.id.len; + msg->store->val.data = malloc(val.len); + if (msg->store->val.data == NULL) + goto fail_msg; + + memcpy(msg->store->val.data, val.data, val.len); + + msg->store->val.len = val.len; + msg->store->exp = exp; + + return msg; + + fail_msg: + dht_msg__free_unpacked(msg, NULL); + fail_malloc: + return NULL; } -static int kad_store(struct dht * dht, - const uint8_t * key, - uint64_t addr, - uint64_t r_addr, - time_t ttl) +static ssize_t dht_kv_retrieve(const uint8_t * key, + buffer_t ** vals) { - dht_msg_t msg = DHT_MSG__INIT; - dht_contact_msg_t cmsg = DHT_CONTACT_MSG__INIT; - dht_contact_msg_t * cmsgp[1]; + struct dht_entry * e; + struct list_head * p; + size_t n; + size_t i; - cmsg.id.data = (uint8_t *) key; - cmsg.addr = addr; + assert(key != NULL); - pthread_rwlock_rdlock(&dht->lock); + pthread_rwlock_rdlock(&dht.db.lock); - cmsg.id.len = dht->b; + e = __dht_kv_find_entry(key); + if (e == NULL) + goto no_vals; - pthread_rwlock_unlock(&dht->lock); + n = MIN(DHT_MAX_VALS, e->vals.len + e->lvals.len); + if (n == 0) + goto no_vals; - cmsgp[0] = &cmsg; + *vals = malloc(n * sizeof(**vals)); + if (*vals == NULL) + goto fail_vals; - msg.code = KAD_STORE; - msg.has_t_expire = true; - msg.t_expire = ttl; - msg.n_contacts = 1; - msg.contacts = cmsgp; + memset(*vals, 0, n * sizeof(**vals)); - if (send_msg(dht, &msg, r_addr) < 0) - return -1; + i = 0; + + llist_for_each(p, &e->vals) { + struct val_entry * v; + if (i == n) + break; /* We have enough values */ + v = list_entry(p, struct val_entry, next); + (*vals)[i].data = malloc(v->val.len); + if ((*vals)[i].data == NULL) + goto fail_val_data; + + (*vals)[i].len = v->val.len; + memcpy((*vals)[i++].data, v->val.data, v->val.len); + } + llist_for_each(p, &e->lvals) { + struct val_entry * v; + if (i == n) + break; /* We have enough values */ + v = list_entry(p, struct val_entry, next); + (*vals)[i].data = malloc(v->val.len); + if ((*vals)[i].data == NULL) + goto fail_val_data; + + (*vals)[i].len = v->val.len; + memcpy((*vals)[i++].data, v->val.data, v->val.len); + } + + pthread_rwlock_unlock(&dht.db.lock); + + return (ssize_t) i; + + fail_val_data: + pthread_rwlock_unlock(&dht.db.lock); + freebufs(*vals, i); + *vals = NULL; + return -ENOMEM; + fail_vals: + pthread_rwlock_unlock(&dht.db.lock); + return -ENOMEM; + no_vals: + pthread_rwlock_unlock(&dht.db.lock); + *vals = NULL; return 0; } -static ssize_t kad_find(struct dht * dht, - struct lookup * lu, - const uint64_t * addrs, - enum kad_code code) +static void __cleanup_dht_msg(void * msg) +{ + dht_msg__free_unpacked((dht_msg_t *) msg, NULL); +} + +#ifdef DEBUG_PROTO_DHT +static void dht_kv_debug_msg(dht_msg_t * msg) +{ + struct tm * tm; + char tmstr[RIB_TM_STRLEN]; + time_t stamp; + size_t i; + + if (msg == NULL) + return; + + pthread_cleanup_push(__cleanup_dht_msg, msg); + + switch (msg->code) { + case DHT_STORE: + log_proto(" key: " HASH_FMT64 " [%zu bytes]", + HASH_VAL64(msg->store->key.data), + msg->store->key.len); + log_proto(" val: " HASH_FMT64 " [%zu bytes]", + HASH_VAL64(msg->store->val.data), + msg->store->val.len); + stamp = msg->store->exp; + tm = gmtime(&stamp); + strftime(tmstr, sizeof(tmstr), RIB_TM_FORMAT, tm); + log_proto(" exp: %s.", tmstr); + break; + case DHT_FIND_NODE_REQ: + /* FALLTHRU */ + case DHT_FIND_VALUE_REQ: + log_proto(" cookie: " HASH_FMT64, + HASH_VAL64(&msg->find->cookie)); + log_proto(" key: " HASH_FMT64 " [%zu bytes]", + HASH_VAL64(msg->find->key.data), + msg->find->key.len); + break; + case DHT_FIND_VALUE_RSP: + log_proto(" cookie: " HASH_FMT64, + HASH_VAL64(&msg->node->cookie)); + log_proto(" key: " HASH_FMT64 " [%zu bytes]", + HASH_VAL64(msg->node->key.data), + msg->node->key.len); + log_proto(" values: [%zd]", msg->val->n_values); + for (i = 0; i < msg->val->n_values; i++) + log_proto(" " HASH_FMT64 " [%zu bytes]", + HASH_VAL64(msg->val->values[i].data), + msg->val->values[i].len); + log_proto(" contacts: [%zd]", msg->node->n_contacts); + for (i = 0; i < msg->node->n_contacts; i++) { + dht_contact_msg_t * c = msg->node->contacts[i]; + log_proto(" " PEER_FMT, + PEER_VAL(c->id.data, c->addr)); + } + break; + case DHT_FIND_NODE_RSP: + log_proto(" cookie: " HASH_FMT64, + HASH_VAL64(&msg->node->cookie)); + log_proto(" key: " HASH_FMT64 " [%zu bytes]", + HASH_VAL64(msg->node->key.data), msg->node->key.len); + log_proto(" contacts: [%zd]", msg->node->n_contacts); + for (i = 0; i < msg->node->n_contacts; i++) { + dht_contact_msg_t * c = msg->node->contacts[i]; + log_proto(" " PEER_FMT, + PEER_VAL(c->id.data, c->addr)); + } + + break; + default: + break; + } + + pthread_cleanup_pop(false); +} + +static void dht_kv_debug_msg_snd(dht_msg_t * msg, + uint8_t * id, + uint64_t addr) { - dht_msg_t msg = DHT_MSG__INIT; - ssize_t sent = 0; + if (msg == NULL) + return; - assert(dht); - assert(lu->key); + log_proto(TX_HDR_FMT ".", TX_HDR_VAL(msg, id, addr)); - msg.code = code; + dht_kv_debug_msg(msg); +} - msg.has_key = true; - msg.key.data = (uint8_t *) lu->key; - msg.key.len = dht->b; +static void dht_kv_debug_msg_rcv(dht_msg_t * msg) +{ + if (msg == NULL) + return; - while (*addrs != 0) { - struct cookie_el * c; - int ret; + log_proto(RX_HDR_FMT ".", RX_HDR_VAL(msg)); - if (*addrs == dht->addr) { - ++addrs; - continue; - } + dht_kv_debug_msg(msg); +} +#endif - ret = send_msg(dht, &msg, *addrs); - if (ret < 0) - break; +#ifndef __DHT_TEST__ +static int dht_send_msg(dht_msg_t * msg, + uint64_t addr) +{ + size_t len; + struct ssm_pk_buff * spb; - c = malloc(sizeof(*c)); - if (c == NULL) - break; + if (msg == NULL) + return 0; - c->cookie = (uint32_t) ret; + assert(addr != INVALID_ADDR && addr != dht.addr); - pthread_mutex_lock(&lu->lock); + len = dht_msg__get_packed_size(msg); + if (len == 0) { + log_warn("%s failed to pack.", DHT_CODE(msg)); + goto fail_msg; + } - list_add_tail(&c->next, &lu->cookies); + if (ipcp_spb_reserve(&spb, len)) { + log_warn("%s failed to get spb.", DHT_CODE(msg)); + goto fail_msg; + } - pthread_mutex_unlock(&lu->lock); + dht_msg__pack(msg, ssm_pk_buff_head(spb)); - ++sent; - ++addrs; + if (dt_write_packet(addr, QOS_CUBE_BE, dht.eid, spb) < 0) { + log_warn("%s write failed", DHT_CODE(msg)); + goto fail_send; } - return sent; + return 0; + fail_send: + ipcp_spb_release(spb); + fail_msg: + return -1; } +#else /* funtion for testing */ +static int dht_send_msg(dht_msg_t * msg, + uint64_t addr) +{ + buffer_t buf; + + assert(msg != NULL); + assert(addr != INVALID_ADDR && addr != dht.addr); + + buf.len = dht_msg__get_packed_size(msg); + if (buf.len == 0) { + log_warn("%s failed to pack.", DHT_CODE(msg)); + goto fail_msg; + } -static void lookup_detach(struct dht * dht, - struct lookup * lu) + buf.data = malloc(buf.len); + if (buf.data == NULL) { + log_warn("%s failed to malloc buf.", DHT_CODE(msg)); + goto fail_msg; + } + + dht_msg__pack(msg, buf.data); + + if (sink_send_msg(&buf, addr) < 0) { + log_warn("%s write failed", DHT_CODE(msg)); + goto fail_send; + } + + return 0; + fail_send: + freebuf(buf); + fail_msg: + return -1; +} +#endif /* __DHT_TEST__ */ + +static void __cleanup_peer_list(void * pl) { - pthread_rwlock_wrlock(&dht->lock); + struct list_head * p; + struct list_head * h; - list_del(&lu->next); + assert(pl != NULL); - pthread_rwlock_unlock(&dht->lock); + list_for_each_safe(p, h, (struct list_head *) pl) { + struct peer_entry * e = list_entry(p, struct peer_entry, next); + list_del(&e->next); + free(e->id); + free(e); + } } -static struct lookup * kad_lookup(struct dht * dht, - const uint8_t * id, - enum kad_code code) + +static int dht_kv_send_msgs(dht_msg_t * msg, + struct list_head * pl) { - uint64_t addrs[KAD_ALPHA + 1]; - enum lookup_state state; - struct lookup * lu; + struct list_head * p; + struct list_head * h; - lu = lookup_create(dht, id); - if (lu == NULL) - return NULL; + pthread_cleanup_push(__cleanup_dht_msg, msg); + pthread_cleanup_push(__cleanup_peer_list, pl); - lookup_new_addrs(lu, addrs); + list_for_each_safe(p, h, pl) { + struct peer_entry * e = list_entry(p, struct peer_entry, next); + if (IS_REQUEST(msg->code)) { + msg->find->cookie = e->cookie; + assert(msg->find->cookie != DHT_INVALID); + } + if (dht_send_msg(msg, e->addr) < 0) + continue; - if (addrs[0] == 0) { - lookup_detach(dht, lu); - lookup_destroy(lu); - return NULL; +#ifdef DEBUG_PROTO_DHT + dht_kv_debug_msg_snd(msg, e->id, e->addr); +#endif + list_del(&e->next); + free(e->id); + free(e); } - if (kad_find(dht, lu, addrs, code) == 0) { - lookup_detach(dht, lu); - return lu; + pthread_cleanup_pop(false); + pthread_cleanup_pop(false); + + return list_is_empty(pl) ? 0 : -1; +} + +static int dht_kv_get_peer_list_for_msg(dht_msg_t * msg, + struct list_head * pl) +{ + struct list_head cl; /* contact list */ + uint8_t * key; /* key in the request */ + size_t max; + + assert(msg != NULL); + + assert(list_is_empty(pl)); + + max = msg->code == DHT_STORE ? dht.k : dht.alpha; + + switch (msg->code) { + case DHT_FIND_NODE_REQ: + /* FALLTHRU */ + case DHT_FIND_VALUE_REQ: + key = msg->find->key.data; + break; + case DHT_STORE: + key = msg->store->key.data; + break; + default: + log_err("Invalid DHT msg code (%d).", msg->code); + return -1; } - while ((state = lookup_wait(lu)) != LU_COMPLETE) { - switch (state) { - case LU_UPDATE: - lookup_new_addrs(lu, addrs); - if (addrs[0] == 0) - break; + list_head_init(&cl); - kad_find(dht, lu, addrs, code); - break; - case LU_DESTROY: - lookup_detach(dht, lu); - lookup_set_state(lu, LU_NULL); - return NULL; - default: - break; - } + if (dht_kv_contact_list(key, &cl, max) < 0) { + log_err(KEY_FMT " Failed to get contact list.", KEY_VAL(key)); + goto fail_contacts; } - assert(state == LU_COMPLETE); + if (list_is_empty(&cl)) { + log_warn(KEY_FMT " No available contacts.", KEY_VAL(key)); + goto fail_contacts; + } - lookup_detach(dht, lu); + if (dht_kv_create_peer_list(&cl, pl, msg->code) < 0) { + log_warn(KEY_FMT " Failed to get peer list.", KEY_VAL(key)); + goto fail_peers; + } - return lu; + contact_list_destroy(&cl); + return 0; + fail_peers: + contact_list_destroy(&cl); + fail_contacts: + return -1; } -static void kad_publish(struct dht * dht, - const uint8_t * key, - uint64_t addr, - time_t exp) +static int dht_kv_store_remote(const uint8_t * key, + const buffer_t val, + time_t exp) { - struct lookup * lu; - uint64_t * addrs; - ssize_t n; - size_t k; - time_t t_expire; + dht_msg_t * msg; + struct timespec now; + struct list_head pl; + assert(key != NULL); + assert(val.data != NULL); + assert(val.len > 0); - assert(dht); - assert(key); + clock_gettime(CLOCK_REALTIME_COARSE, &now); - pthread_rwlock_rdlock(&dht->lock); + msg = dht_kv_store_msg(key, val, exp); + if (msg == NULL) { + log_err(KV_FMT " Failed to create %s.", + KV_VAL(key, val), dht_code_str[DHT_STORE]); + goto fail_msg; + } - k = dht->k; - t_expire = dht->t_expire; + list_head_init(&pl); - pthread_rwlock_unlock(&dht->lock); + if (dht_kv_get_peer_list_for_msg(msg, &pl) < 0) { + log_dbg(KV_FMT " Failed to get peer list.", KV_VAL(key, val)); + goto fail_peer_list; + } - addrs = malloc(k * sizeof(*addrs)); - if (addrs == NULL) - return; + if (dht_kv_send_msgs(msg, &pl) < 0) { + log_warn(KV_FMT " Failed to send any %s msg.", + KV_VAL(key, val), DHT_CODE(msg)); + goto fail_msgs; + } - lu = kad_lookup(dht, key, KAD_FIND_NODE); - if (lu == NULL) { - free(addrs); - return; + dht_msg__free_unpacked(msg, NULL); + + return 0; + fail_msgs: + peer_list_destroy(&pl); + fail_peer_list: + dht_msg__free_unpacked(msg, NULL); + fail_msg: + return -1; +} + +/* recursive lookup, start with pl NULL */ +static int dht_kv_query_contacts(const uint8_t * key, + struct list_head * pl) +{ + struct list_head p; + + dht_msg_t * msg; + + assert(key != NULL); + + msg = dht_kv_find_node_req_msg(key); + if (msg == NULL) { + log_err(KEY_FMT " Failed to create %s msg.", + KEY_VAL(key), dht_code_str[DHT_FIND_NODE_REQ]); + goto fail_msg; } - n = lookup_contact_addrs(lu, addrs); + if (pl == NULL) { + list_head_init(&p); + pl = &p; + } - while (n-- > 0) { - if (addrs[n] == dht->addr) { - dht_contact_msg_t msg = DHT_CONTACT_MSG__INIT; - msg.id.data = (uint8_t *) key; - msg.id.len = dht->b; - msg.addr = addr; - kad_add(dht, &msg, 1, exp); - } else { - if (kad_store(dht, key, addr, addrs[n], t_expire)) - log_warn("Failed to send store message."); - } + if (list_is_empty(pl) && dht_kv_get_peer_list_for_msg(msg, pl) < 0) { + log_warn(KEY_FMT " Failed to get peer list.", KEY_VAL(key)); + goto fail_peer_list; } - lookup_destroy(lu); + if (dht_kv_update_req(key, pl) < 0) { + log_warn(KEY_FMT " Failed to update req.", KEY_VAL(key)); + goto fail_update; + } + + if (dht_kv_send_msgs(msg, pl)) { + log_warn(KEY_FMT " Failed to send any %s msg.", + KEY_VAL(key), DHT_CODE(msg)); + goto fail_update; + } - free(addrs); + dht_msg__free_unpacked(msg, NULL); + + return 0; + fail_update: + peer_list_destroy(pl); + fail_peer_list: + dht_msg__free_unpacked(msg, NULL); + fail_msg: + return -1; } -static int kad_join(struct dht * dht, - uint64_t addr) +/* recursive lookup, start with pl NULL */ +static ssize_t dht_kv_query_remote(const uint8_t * key, + buffer_t ** vals, + struct list_head * pl) { - dht_msg_t msg = DHT_MSG__INIT; + struct list_head p; + dht_msg_t * msg; + + assert(key != NULL); - msg.code = KAD_JOIN; + msg = dht_kv_find_value_req_msg(key); + if (msg == NULL) { + log_err(KEY_FMT " Failed to create value req.", KEY_VAL(key)); + goto fail_msg; + } - msg.has_alpha = true; - msg.has_b = true; - msg.has_k = true; - msg.has_t_refresh = true; - msg.has_t_replicate = true; - msg.alpha = KAD_ALPHA; - msg.k = KAD_K; - msg.t_refresh = KAD_T_REFR; - msg.t_replicate = KAD_T_REPL; + if (pl == NULL) { + list_head_init(&p); + pl = &p; + } - pthread_rwlock_rdlock(&dht->lock); + if (list_is_empty(pl) && dht_kv_get_peer_list_for_msg(msg, pl) < 0) { + log_warn(KEY_FMT " Failed to get peer list.", KEY_VAL(key)); + goto fail_peer_list; + } - msg.b = dht->b; + if (dht_kv_update_req(key, pl) < 0) { + log_err(KEY_FMT " Failed to update request.", KEY_VAL(key)); + goto fail_update; + } - pthread_rwlock_unlock(&dht->lock); + if (dht_kv_send_msgs(msg, pl)) { + log_warn(KEY_FMT " Failed to send %s msg.", + KEY_VAL(key), DHT_CODE(msg)); + goto fail_update; + } - if (send_msg(dht, &msg, addr) < 0) - return -1; + dht_msg__free_unpacked(msg, NULL); - if (wait_resp(dht, &msg, KAD_T_JOIN) < 0) - return -1; + if (vals == NULL) /* recursive lookup, already waiting */ + return 0; - dht->id = create_id(dht->b); - if (dht->id == NULL) - return -1; + return dht_kv_wait_req(key, vals); + fail_update: + peer_list_destroy(pl); + fail_peer_list: + dht_msg__free_unpacked(msg, NULL); + fail_msg: + return -1; +} - pthread_rwlock_wrlock(&dht->lock); +static void __add_dht_kv_entry(struct dht_entry * e) +{ + struct list_head * p; - dht_update_bucket(dht, dht->id, dht->addr); + assert(e != NULL); - pthread_rwlock_unlock(&dht->lock); + llist_for_each(p, &dht.db.kv.ll) { + struct dht_entry * d = list_entry(p, struct dht_entry, next); + if (IS_CLOSER(d->key, e->key)) + continue; + break; + } - return 0; + llist_add_tail_at(&e->next, p, &dht.db.kv.ll); } -static void dht_dead_peer(struct dht * dht, - uint8_t * key, - uint64_t addr) +/* incoming store message */ +static int dht_kv_store(const uint8_t * key, + const buffer_t val, + time_t exp) { - struct list_head * p; - struct list_head * h; - struct bucket * b; + struct dht_entry * e; + bool new = false; - b = dht_get_bucket(dht, key); + assert(key != NULL); + assert(val.data != NULL); + assert(val.len > 0); - list_for_each_safe(p, h, &b->contacts) { - struct contact * c = list_entry(p, struct contact, next); - if (b->n_contacts + b->n_alts <= dht->k) { - ++c->fails; - return; - } + pthread_rwlock_wrlock(&dht.db.lock); - if (c->addr == addr) { - list_del(&c->next); - contact_destroy(c); - --b->n_contacts; - break; - } + e = __dht_kv_find_entry(key); + if (e == NULL) { + log_dbg(KV_FMT " Adding entry (store).", KV_VAL(key, val)); + e = dht_entry_create(key); + if (e == NULL) + goto fail; + + new = true; + + __add_dht_kv_entry(e); } - while (b->n_contacts < dht->k && b->n_alts > 0) { - struct contact * c; - c = list_first_entry(&b->alts, struct contact, next); - list_del(&c->next); - --b->n_alts; - list_add(&c->next, &b->contacts); - ++b->n_contacts; + if (dht_entry_update_val(e, val, exp) < 0) + goto fail_add; + + pthread_rwlock_unlock(&dht.db.lock); + + return 0; + fail_add: + if (new) { + llist_del(&e->next, &dht.db.kv.ll); + dht_entry_destroy(e); } + fail: + pthread_rwlock_unlock(&dht.db.lock); + return -1; } -static int dht_del(struct dht * dht, - const uint8_t * key, - uint64_t addr) +static int dht_kv_publish(const uint8_t * key, + const buffer_t val) { struct dht_entry * e; + struct timespec now; + bool new = false; - e = dht_find_entry(dht, key); + assert(key != NULL); + assert(val.data != NULL); + assert(val.len > 0); + + clock_gettime(CLOCK_REALTIME_COARSE, &now); + + pthread_rwlock_wrlock(&dht.db.lock); + + e = __dht_kv_find_entry(key); if (e == NULL) { - return -EPERM; + log_dbg(KV_FMT " Adding entry (publish).", KV_VAL(key, val)); + e = dht_entry_create(key); + if (e == NULL) + goto fail; + + __add_dht_kv_entry(e); + new = true; } - dht_entry_del_addr(e, addr); + if (dht_entry_update_lval(e, val) < 0) + goto fail_add; + + pthread_rwlock_unlock(&dht.db.lock); + + dht_kv_store_remote(key, val, now.tv_sec + dht.t_exp); return 0; + fail_add: + if (new) { + llist_del(&e->next, &dht.db.kv.ll); + dht_entry_destroy(e); + } + fail: + pthread_rwlock_unlock(&dht.db.lock); + return -1; } -static buffer_t dht_retrieve(struct dht * dht, - const uint8_t * key) +static int dht_kv_unpublish(const uint8_t * key, + const buffer_t val) { struct dht_entry * e; - struct list_head * p; - buffer_t buf; - uint64_t * pos; - size_t addrs = 0; + int rc; - pthread_rwlock_rdlock(&dht->lock); + assert(key != NULL); - e = dht_find_entry(dht, key); + pthread_rwlock_wrlock(&dht.db.lock); + + e = __dht_kv_find_entry(key); if (e == NULL) - goto fail; + goto no_entry; - buf.len = MIN(DHT_RETR_ADDR, e->n_vals); - if (buf.len == 0) - goto fail; + rc = dht_entry_remove_lval(e, val); - pos = malloc(sizeof(dht->addr) * buf.len); - if (pos == NULL) - goto fail; + pthread_rwlock_unlock(&dht.db.lock); - buf.data = (uint8_t *) pos; + return rc; + no_entry: + pthread_rwlock_unlock(&dht.db.lock); + return -ENOENT; - list_for_each(p, &e->vals) { - struct val * v = list_entry(p, struct val, next); - *pos++ = v->addr; - if (++addrs >= buf.len) - break; +} + +/* message validation */ +static int dht_kv_validate_store_msg(const dht_store_msg_t * store) +{ + if (store == NULL) { + log_warn("Store in msg is NULL."); + return -EINVAL; } - pthread_rwlock_unlock(&dht->lock); + if (store->key.data == NULL || store->key.len == 0) { + log_warn("Invalid key in DHT store msg."); + return -EINVAL; + } - return buf; + if (store->key.len != dht.id.len) { + log_warn("Invalid key length in DHT store msg."); + return -EINVAL; + } - fail: - pthread_rwlock_unlock(&dht->lock); - buf.len = 0; - buf.data = NULL; - return buf; + if (store->val.data == NULL || store->val.len == 0) { + log_warn("Invalid value in DHT store msg."); + return -EINVAL; + } + + return 0; } -static ssize_t dht_get_contacts(struct dht * dht, - const uint8_t * key, - dht_contact_msg_t *** msgs) +static int validate_find_req_msg(const dht_find_req_msg_t * req) { - struct list_head l; - struct list_head * p; - struct list_head * h; - size_t len; - size_t i = 0; + if (req == NULL) { + log_warn("Request in msg is NULL."); + return -EINVAL; + } - list_head_init(&l); + if (req->key.data == NULL || req->key.len == 0) { + log_warn("Find request without key."); + return -EINVAL; + } - pthread_rwlock_wrlock(&dht->lock); + if (req->key.len != dht.id.len) { + log_warn("Invalid key length in request msg."); + return -EINVAL; + } - len = dht_contact_list(dht, &l, key); - if (len == 0) { - pthread_rwlock_unlock(&dht->lock); - *msgs = NULL; + return 0; +} + +static int validate_node_rsp_msg(const dht_find_node_rsp_msg_t * rsp) +{ + if (rsp == NULL) { + log_warn("Node rsp in msg is NULL."); + return -EINVAL; + } + + if (rsp->key.data == NULL) { + log_warn("Invalid key in DHT response msg."); + return -EINVAL; + } + + if (rsp->key.len != dht.id.len) { + log_warn("Invalid key length in DHT response msg."); + return -EINVAL; + } + + if (!dht_kv_has_req(rsp->key.data, rsp->cookie)) { + log_warn(KEY_FMT " No request " CK_FMT ".", + KEY_VAL(rsp->key.data), CK_VAL(rsp->cookie)); + + return -EINVAL; + } + + return 0; +} + +static int validate_value_rsp_msg(const dht_find_value_rsp_msg_t * rsp) +{ + if (rsp == NULL) { + log_warn("Invalid DHT find value response msg."); + return -EINVAL; + } + + if (rsp->values == NULL && rsp->n_values > 0) { + log_dbg("No values in DHT response msg."); return 0; } - *msgs = malloc(len * sizeof(**msgs)); - if (*msgs == NULL) { - pthread_rwlock_unlock(&dht->lock); + if (rsp->n_values == 0 && rsp->values != NULL) { + log_dbg("DHT response did not set values NULL."); return 0; } - list_for_each_safe(p, h, &l) { - struct contact * c = list_entry(p, struct contact, next); - (*msgs)[i] = malloc(sizeof(***msgs)); - if ((*msgs)[i] == NULL) { - pthread_rwlock_unlock(&dht->lock); - while (i > 0) - free(*msgs[--i]); - free(*msgs); - *msgs = NULL; - return 0; - } + return 0; +} - dht_contact_msg__init((*msgs)[i]); +static int dht_kv_validate_msg(dht_msg_t * msg) +{ - (*msgs)[i]->id.data = c->id; - (*msgs)[i]->id.len = dht->b; - (*msgs)[i++]->addr = c->addr; - list_del(&c->next); - free(c); + assert(msg != NULL); + + if (msg->src->id.len != dht.id.len) { + log_warn("%s Invalid source contact ID.", DHT_CODE(msg)); + return -EINVAL; } - pthread_rwlock_unlock(&dht->lock); + if (msg->src->addr == INVALID_ADDR) { + log_warn("%s Invalid source address.", DHT_CODE(msg)); + return -EINVAL; + } - return i; + switch (msg->code) { + case DHT_FIND_VALUE_REQ: + /* FALLTHRU */ + case DHT_FIND_NODE_REQ: + if (validate_find_req_msg(msg->find) < 0) + return -EINVAL; + break; + case DHT_FIND_VALUE_RSP: + if (validate_value_rsp_msg(msg->val) < 0) + return -EINVAL; + /* FALLTHRU */ + case DHT_FIND_NODE_RSP: + if (validate_node_rsp_msg(msg->node) < 0) + return -EINVAL; + break; + case DHT_STORE: + if (dht_kv_validate_store_msg(msg->store) < 0) + return -EINVAL; + break; + default: + log_warn("Invalid DHT msg code (%d).", msg->code); + return -ENOENT; + } + + return 0; } -static time_t gcd(time_t a, - time_t b) +static void do_dht_kv_store(const dht_store_msg_t * store) { - if (a == 0) - return b; + struct tm * tm; + char tmstr[RIB_TM_STRLEN]; + buffer_t val; + uint8_t * key; + time_t exp; - return gcd(b % a, a); + assert(store != NULL); + + val.data = store->val.data; + val.len = store->val.len; + key = store->key.data; + exp = store->exp; + + if (dht_kv_store(key, val, store->exp) < 0) { + log_err(KV_FMT " Failed to store.", KV_VAL(key, val)); + return; + } + + tm = gmtime(&exp); + strftime(tmstr, sizeof(tmstr), RIB_TM_FORMAT, tm); + log_dbg(KV_FMT " Stored value until %s.", KV_VAL(key, val), tmstr); } -static void * work(void * o) +static dht_msg_t * do_dht_kv_find_node_req(const dht_find_req_msg_t * req) { - struct dht * dht; - struct timespec now; - struct list_head * p; - struct list_head * h; - struct list_head reflist; - time_t intv; - struct lookup * lu; + dht_contact_msg_t ** contacts; + dht_msg_t * rsp; + uint8_t * key; + uint64_t cookie; + ssize_t len; - dht = (struct dht *) o; + assert(req != NULL); - pthread_rwlock_rdlock(&dht->lock); + key = req->key.data; + cookie = req->cookie; - intv = gcd(dht->t_expire, dht->t_repub); - intv = gcd(intv, gcd(KAD_T_REPL, KAD_T_REFR)) / 2; + len = dht_kv_get_contacts(key, &contacts); + if (len < 0) { + log_warn(KEY_FMT " Failed to get contacts.", KEY_VAL(key)); + goto fail_contacts; + } - pthread_rwlock_unlock(&dht->lock); + rsp = dht_kv_find_node_rsp_msg(key, cookie, &contacts, len); + if (rsp == NULL) { + log_err(KEY_FMT " Failed to create %s.", KEY_VAL(key), + dht_code_str[DHT_FIND_NODE_RSP]); + goto fail_msg; + } - list_head_init(&reflist); + assert(rsp->code == DHT_FIND_NODE_RSP); - while (true) { - clock_gettime(CLOCK_REALTIME_COARSE, &now); - - pthread_rwlock_wrlock(&dht->lock); - - /* Republish registered hashes. */ - list_for_each(p, &dht->refs) { - struct ref_entry * e; - uint8_t * key; - uint64_t addr; - time_t t_expire; - e = list_entry(p, struct ref_entry, next); - if (now.tv_sec > e->t_rep) { - key = dht_dup_key(e->key, dht->b); - if (key == NULL) - continue; - addr = dht->addr; - t_expire = dht->t_expire; - e->t_rep = now.tv_sec + dht->t_repub; - - pthread_rwlock_unlock(&dht->lock); - kad_publish(dht, key, addr, t_expire); - pthread_rwlock_wrlock(&dht->lock); - free(key); - } - } + log_info(KEY_FMT " Responding with %zd contacts", KEY_VAL(key), len); - /* Remove stale entries and republish if necessary. */ - list_for_each_safe(p, h, &dht->entries) { - struct list_head * p1; - struct list_head * h1; - struct dht_entry * e; - uint8_t * key; - time_t t_expire; - e = list_entry (p, struct dht_entry, next); - list_for_each_safe(p1, h1, &e->vals) { - struct val * v; - uint64_t addr; - v = list_entry(p1, struct val, next); - if (now.tv_sec > v->t_exp) { - list_del(&v->next); - val_destroy(v); - continue; - } - - if (now.tv_sec > v->t_rep) { - key = dht_dup_key(e->key, dht->b); - addr = v->addr; - t_expire = dht->t_expire = now.tv_sec; - v->t_rep = now.tv_sec + dht->t_replic; - pthread_rwlock_unlock(&dht->lock); - kad_publish(dht, key, addr, t_expire); - pthread_rwlock_wrlock(&dht->lock); - free(key); - } - } - } + return rsp; + fail_msg: + while (len-- > 0) + dht_contact_msg__free_unpacked(contacts[len], NULL); + free(contacts); + fail_contacts: + return NULL; +} - /* Check the requests list for unresponsive nodes. */ - list_for_each_safe(p, h, &dht->requests) { - struct kad_req * r; - r = list_entry(p, struct kad_req, next); - if (now.tv_sec > r->t_exp) { - list_del(&r->next); - bmp_release(dht->cookies, r->cookie); - dht_dead_peer(dht, r->key, r->addr); - kad_req_destroy(r); - } - } +static void dht_kv_process_node_rsp(dht_contact_msg_t ** contacts, + size_t len, + struct list_head * pl, + enum dht_code code) +{ + struct timespec now; + size_t i; - /* Refresh unaccessed buckets. */ - bucket_refresh(dht, dht->buckets, now.tv_sec, &reflist); + assert(contacts != NULL); + assert(len > 0); + assert(pl != NULL); + assert(list_is_empty(pl)); - pthread_rwlock_unlock(&dht->lock); + clock_gettime(CLOCK_REALTIME_COARSE, &now); - list_for_each_safe(p, h, &reflist) { - struct contact * c; - c = list_entry(p, struct contact, next); - lu = kad_lookup(dht, c->id, KAD_FIND_NODE); - if (lu != NULL) - lookup_destroy(lu); - list_del(&c->next); - contact_destroy(c); + for (i = 0; i < len; i++) { + dht_contact_msg_t * c = contacts[i]; + struct peer_entry * e; + if (c->addr == dht.addr) + continue; + + if (dht_kv_update_contacts(c->id.data, c->addr) < 0) + log_warn(PEER_FMT " Failed to update contacts.", + PEER_VAL(c->id.data, c->addr)); + + e = malloc(sizeof(*e)); + if (e == NULL) { + log_err(PEER_FMT " Failed to malloc entry.", + PEER_VAL(c->id.data, c->addr)); + continue; } - sleep(intv); - } + e->id = dht_dup_key(c->id.data); + if (e->id == NULL) { + log_warn(PEER_FMT " Failed to duplicate id.", + PEER_VAL(c->id.data, c->addr)); + free(e); + continue; + } - return (void *) 0; + e->cookie = generate_cookie(); + e->code = code; + e->addr = c->addr; + e->t_sent = now.tv_sec; + + list_add_tail(&e->next, pl); + } } -static int kad_handle_join_resp(struct dht * dht, - struct kad_req * req, - dht_msg_t * msg) +static dht_msg_t * do_dht_kv_find_value_req(const dht_find_req_msg_t * req) { - assert(dht); - assert(req); - assert(msg); + dht_contact_msg_t ** contacts; + ssize_t n_contacts; + buffer_t * vals; + ssize_t n_vals; + dht_msg_t * rsp; + uint8_t * key; + uint64_t cookie; - /* We might send version numbers later to warn of updates if needed. */ - if (!(msg->has_alpha && msg->has_b && msg->has_k && msg->has_t_expire && - msg->has_t_refresh && msg->has_t_replicate)) { - log_warn("Join refused by remote."); - return -1; + assert(req != NULL); + + key = req->key.data; + cookie = req->cookie; + + n_contacts = dht_kv_get_contacts(key, &contacts); + if (n_contacts < 0) { + log_warn(KEY_FMT " Failed to get contacts.", KEY_VAL(key)); + goto fail_contacts; } - if (msg->b < sizeof(uint64_t)) { - log_err("Hash sizes less than 8 bytes unsupported."); - return -1; + assert(n_contacts > 0 || contacts == NULL); + + n_vals = dht_kv_retrieve(key, &vals); + if (n_vals < 0) { + log_dbg(KEY_FMT " Failed to get values.", KEY_VAL(key)); + goto fail_vals; } - pthread_rwlock_wrlock(&dht->lock); + if (n_vals == 0) + log_dbg(KEY_FMT " No values found.", KEY_VAL(key)); - dht->buckets = bucket_create(); - if (dht->buckets == NULL) { - pthread_rwlock_unlock(&dht->lock); - return -1; + rsp = dht_kv_find_value_rsp_msg(key, cookie, &contacts, n_contacts, + &vals, n_vals); + if (rsp == NULL) { + log_err(KEY_FMT " Failed to create %s msg.", + KEY_VAL(key), dht_code_str[DHT_FIND_VALUE_RSP]); + goto fail_msg; } - /* Likely corrupt packet. The member will refuse, we might here too. */ - if (msg->alpha != KAD_ALPHA || msg->k != KAD_K) - log_warn("Different kademlia parameters detected."); + log_info(KEY_FMT " Responding with %zd contacts, %zd values.", + KEY_VAL(req->key.data), n_contacts, n_vals); + + return rsp; + + fail_msg: + freebufs(vals, n_vals); + fail_vals: + while (n_contacts-- > 0) + dht_contact_msg__free_unpacked(contacts[n_contacts], NULL); + free(contacts); + fail_contacts: + return NULL; +} - if (msg->t_replicate != KAD_T_REPL) - log_warn("Different kademlia replication time detected."); +static void do_dht_kv_find_node_rsp(const dht_find_node_rsp_msg_t * rsp) +{ + struct list_head pl; - if (msg->t_refresh != KAD_T_REFR) - log_warn("Different kademlia refresh time detected."); + assert(rsp != NULL); - dht->k = msg->k; - dht->b = msg->b; - dht->t_expire = msg->t_expire; - dht->t_repub = MAX(1, dht->t_expire - 10); + list_head_init(&pl); - if (pthread_create(&dht->worker, NULL, work, dht)) { - bucket_destroy(dht->buckets); - pthread_rwlock_unlock(&dht->lock); - return -1; - } + dht_kv_process_node_rsp(rsp->contacts, rsp->n_contacts, &pl, + DHT_FIND_NODE_REQ); - kad_req_respond(req); + if (list_is_empty(&pl)) + goto no_contacts; - dht_update_bucket(dht, msg->s_id.data, msg->s_addr); + if (dht_kv_update_req(rsp->key.data, &pl) < 0) { + log_err(KEY_FMT " Failed to update request.", + KEY_VAL(rsp->key.data)); + goto fail_update; + } - pthread_rwlock_unlock(&dht->lock); + dht_kv_query_contacts(rsp->key.data, &pl); - log_dbg("Enrollment of DHT completed."); + return; - return 0; + fail_update: + peer_list_destroy(&pl); + no_contacts: + return; } -static int kad_handle_find_resp(struct dht * dht, - struct kad_req * req, - dht_msg_t * msg) +static void do_dht_kv_find_value_rsp(const dht_find_node_rsp_msg_t * node, + const dht_find_value_rsp_msg_t * val) { - struct lookup * lu; + struct list_head pl; + uint8_t * key; - assert(dht); - assert(req); - assert(msg); + assert(node != NULL); + assert(val != NULL); - pthread_rwlock_rdlock(&dht->lock); + list_head_init(&pl); - lu = dht_find_lookup(dht, req->cookie); - if (lu == NULL) { - pthread_rwlock_unlock(&dht->lock); - return -1; + key = node->key.data; + + dht_kv_process_node_rsp(node->contacts, node->n_contacts, &pl, + DHT_FIND_VALUE_REQ); + + if (val->n_values > 0) { + log_dbg(KEY_FMT " %zd new values received.", + KEY_VAL(key), val->n_values); + if (dht_kv_respond_req(key, val->values, val->n_values) < 0) + log_warn(KEY_FMT " Failed to respond to request.", + KEY_VAL(key)); + peer_list_destroy(&pl); + return; /* done! */ } - lookup_update(dht, lu, msg); + if (list_is_empty(&pl)) + goto no_contacts; - pthread_rwlock_unlock(&dht->lock); + if (dht_kv_update_req(key, &pl) < 0) { + log_err(KEY_FMT " Failed to update request.", KEY_VAL(key)); + goto fail_update; + } - return 0; + dht_kv_query_remote(key, NULL, &pl); + + return; + fail_update: + peer_list_destroy(&pl); + no_contacts: + return; } -static void kad_handle_response(struct dht * dht, - dht_msg_t * msg) +static dht_msg_t * dht_wait_for_dht_msg(void) { - struct kad_req * req; + dht_msg_t * msg; + struct cmd * cmd; - assert(dht); - assert(msg); + pthread_mutex_lock(&dht.cmds.mtx); - pthread_rwlock_wrlock(&dht->lock); + pthread_cleanup_push(__cleanup_mutex_unlock, &dht.cmds.mtx); - req = dht_find_request(dht, msg); - if (req == NULL) { - pthread_rwlock_unlock(&dht->lock); + while (list_is_empty(&dht.cmds.list)) + pthread_cond_wait(&dht.cmds.cond, &dht.cmds.mtx); + + cmd = list_last_entry(&dht.cmds.list, struct cmd, next); + list_del(&cmd->next); + + pthread_cleanup_pop(true); + + msg = dht_msg__unpack(NULL, cmd->cbuf.len, cmd->cbuf.data); + if (msg == NULL) + log_warn("Failed to unpack DHT msg."); + + freebuf(cmd->cbuf); + free(cmd); + + return msg; +} + +static void do_dht_msg(dht_msg_t * msg) +{ + dht_msg_t * rsp = NULL; + uint8_t * id; + uint64_t addr; + +#ifdef DEBUG_PROTO_DHT + dht_kv_debug_msg_rcv(msg); +#endif + if (dht_kv_validate_msg(msg) == -EINVAL) { + log_warn("%s Validation failed.", DHT_CODE(msg)); + dht_msg__free_unpacked(msg, NULL); return; } - bmp_release(dht->cookies, req->cookie); - list_del(&req->next); + id = msg->src->id.data; + addr = msg->src->addr; + + if (dht_kv_update_contacts(id, addr) < 0) + log_warn(PEER_FMT " Failed to update contact from msg src.", + PEER_VAL(id, addr)); - pthread_rwlock_unlock(&dht->lock); + pthread_cleanup_push(__cleanup_dht_msg, msg); - switch(req->code) { - case KAD_JOIN: - if (kad_handle_join_resp(dht, req, msg)) - log_err("Enrollment of DHT failed."); + switch(msg->code) { + case DHT_FIND_VALUE_REQ: + rsp = do_dht_kv_find_value_req(msg->find); break; - case KAD_FIND_VALUE: - case KAD_FIND_NODE: - if (dht_get_state(dht) != DHT_RUNNING) - break; - kad_handle_find_resp(dht, req, msg); + case DHT_FIND_NODE_REQ: + rsp = do_dht_kv_find_node_req(msg->find); break; - default: + case DHT_STORE: + do_dht_kv_store(msg->store); + break; + case DHT_FIND_NODE_RSP: + do_dht_kv_find_node_rsp(msg->node); break; + case DHT_FIND_VALUE_RSP: + do_dht_kv_find_value_rsp(msg->node, msg->val); + break; + default: + assert(false); /* already validated */ } - kad_req_destroy(req); + pthread_cleanup_pop(true); + + if (rsp == NULL) + return; + + pthread_cleanup_push(__cleanup_dht_msg, rsp); + + dht_send_msg(rsp, addr); + + pthread_cleanup_pop(true); /* free rsp */ } -int dht_bootstrap(void * dir) +static void * dht_handle_packet(void * o) { - struct dht * dht; + (void) o; + + while (true) { + dht_msg_t * msg; - dht = (struct dht *) dir; + msg = dht_wait_for_dht_msg(); + if (msg == NULL) + continue; + + tpm_begin_work(dht.tpm); - assert(dht); + do_dht_msg(msg); - pthread_rwlock_wrlock(&dht->lock); + tpm_end_work(dht.tpm); + } + return (void *) 0; +} #ifndef __DHT_TEST__ - dht->b = hash_len(ipcpi.dir_hash_algo); -#else - dht->b = DHT_TEST_KEY_LEN; -#endif +static void dht_post_packet(void * comp, + struct ssm_pk_buff * spb) +{ + struct cmd * cmd; - dht->id = create_id(dht->b); - if (dht->id == NULL) - goto fail_id; + (void) comp; - dht->buckets = bucket_create(); - if (dht->buckets == NULL) - goto fail_buckets; + cmd = malloc(sizeof(*cmd)); + if (cmd == NULL) { + log_err("Command malloc failed."); + goto fail_cmd; + } - dht->buckets->depth = 0; - dht->buckets->mask = 0; + cmd->cbuf.data = malloc(ssm_pk_buff_len(spb)); + if (cmd->cbuf.data == NULL) { + log_err("Command buffer malloc failed."); + goto fail_buf; + } - dht->t_expire = 86400; /* 1 day */ - dht->t_repub = dht->t_expire - 10; - dht->k = KAD_K; + cmd->cbuf.len = ssm_pk_buff_len(spb); - if (pthread_create(&dht->worker, NULL, work, dht)) - goto fail_pthread_create; + memcpy(cmd->cbuf.data, ssm_pk_buff_head(spb), cmd->cbuf.len); - dht->state = DHT_RUNNING; + ipcp_spb_release(spb); - dht_update_bucket(dht, dht->id, dht->addr); + pthread_mutex_lock(&dht.cmds.mtx); - pthread_rwlock_unlock(&dht->lock); + list_add(&cmd->next, &dht.cmds.list); - return 0; + pthread_cond_signal(&dht.cmds.cond); - fail_pthread_create: - bucket_destroy(dht->buckets); - dht->buckets = NULL; - fail_buckets: - free(dht->id); - dht->id = NULL; - fail_id: - pthread_rwlock_unlock(&dht->lock); - return -1; + pthread_mutex_unlock(&dht.cmds.mtx); + + return; + + fail_buf: + free(cmd); + fail_cmd: + ipcp_spb_release(spb); + return; } +#endif -static struct ref_entry * ref_entry_get(struct dht * dht, - const uint8_t * key) +int dht_reg(const uint8_t * key) { - struct list_head * p; + buffer_t val; - list_for_each(p, &dht->refs) { - struct ref_entry * r = list_entry(p, struct ref_entry, next); - if (!memcmp(key, r->key, dht-> b) ) - return r; + if (addr_to_buf(dht.addr, &val) < 0) { + log_err("Failed to convert address to buffer."); + goto fail_a2b; } - return NULL; + if (dht_kv_publish(key, val)) { + log_err(KV_FMT " Failed to publish.", KV_VAL(key, val)); + goto fail_publish; + } + + freebuf(val); + + return 0; + fail_publish: + freebuf(val); + fail_a2b: + return -1; } -int dht_reg(void * dir, - const uint8_t * key) +int dht_unreg(const uint8_t * key) { - struct dht * dht; - struct ref_entry * e; - uint64_t addr; - time_t t_expire; + buffer_t val; - dht = (struct dht *) dir; + if (addr_to_buf(dht.addr, &val) < 0) { + log_err("Failed to convert address to buffer."); + goto fail_a2b; + } - assert(dht); - assert(key); - assert(dht->addr != 0); + if (dht_kv_unpublish(key, val)) { + log_err(KV_FMT " Failed to unpublish.", KV_VAL(key, val)); + goto fail_unpublish; + } - if (dht_wait_running(dht)) - return -1; + freebuf(val); - pthread_rwlock_wrlock(&dht->lock); + return 0; + fail_unpublish: + freebuf(val); + fail_a2b: + return -ENOMEM; +} - if (ref_entry_get(dht, key) != NULL) { - log_dbg("Name already registered."); - pthread_rwlock_unlock(&dht->lock); - return 0; - } +uint64_t dht_query(const uint8_t * key) +{ + buffer_t * vals; + ssize_t n; + uint64_t addr; - e = ref_entry_create(dht, key); - if (e == NULL) { - pthread_rwlock_unlock(&dht->lock); - return -ENOMEM; + n = dht_kv_retrieve(key, &vals); + if (n < 0) { + log_err(KEY_FMT " Failed to query db.", KEY_VAL(key)); + goto fail_vals; } - list_add(&e->next, &dht->refs); + if (n == 0) { + assert(vals == NULL); - t_expire = dht->t_expire; - addr = dht->addr; + log_dbg(KEY_FMT " No local values.", KEY_VAL(key)); + n = dht_kv_query_remote(key, &vals, NULL); + if (n < 0) { + log_warn(KEY_FMT " Failed to query DHT.", KEY_VAL(key)); + goto fail_vals; + } + if (n == 0) { + log_dbg(KEY_FMT " No values.", KEY_VAL(key)); + goto no_vals; + } + } - pthread_rwlock_unlock(&dht->lock); + if (buf_to_addr(vals[0], &addr) < 0) { + log_err(VAL_FMT " Failed addr conversion.", VAL_VAL(vals[0])); + goto fail_b2a; + } - kad_publish(dht, key, addr, t_expire); + if (n > 1 && addr == INVALID_ADDR && buf_to_addr(vals[1], &addr) < 0) { + log_err(VAL_FMT " Failed addr conversion.", VAL_VAL(vals[1])); + goto fail_b2a; + } - return 0; + freebufs(vals, n); + + return addr; + fail_b2a: + freebufs(vals, n); + return INVALID_ADDR; + no_vals: + free(vals); + fail_vals: + return INVALID_ADDR; } -int dht_unreg(void * dir, - const uint8_t * key) +static int emergency_peer(struct list_head * pl) { - struct dht * dht; - struct list_head * p; - struct list_head * h; - - dht = (struct dht *) dir; + struct peer_entry * e; + struct timespec now; - assert(dht); - assert(key); + assert(pl != NULL); + assert(list_is_empty(pl)); - if (dht_get_state(dht) != DHT_RUNNING) + if (dht.peer == INVALID_ADDR) return -1; - pthread_rwlock_wrlock(&dht->lock); + clock_gettime(CLOCK_REALTIME_COARSE, &now); - list_for_each_safe(p, h, &dht->refs) { - struct ref_entry * r = list_entry(p, struct ref_entry, next); - if (!memcmp(key, r->key, dht-> b) ) { - list_del(&r->next); - ref_entry_destroy(r); - } + e = malloc(sizeof(*e)); + if (e == NULL) { + log_err("Failed to malloc emergency peer entry."); + goto fail_malloc; } - dht_del(dht, key, dht->addr); + e->id = dht_dup_key(dht.id.data); + if (e->id == NULL) { + log_err("Failed to duplicate DHT ID for emergency peer."); + goto fail_id; + } + + e->addr = dht.peer; + e->cookie = dht.magic; + e->code = DHT_FIND_NODE_REQ; + e->t_sent = now.tv_sec; - pthread_rwlock_unlock(&dht->lock); + list_add_tail(&e->next, pl); return 0; + fail_id: + free(e); + fail_malloc: + return -ENOMEM; } -uint64_t dht_query(void * dir, - const uint8_t * key) +static int dht_kv_seed_bootstrap_peer(void) { - struct dht * dht; - struct dht_entry * e; - struct lookup * lu; - uint64_t addrs[KAD_K]; - size_t n; + struct list_head pl; - dht = (struct dht *) dir; + list_head_init(&pl); - assert(dht); + if (dht.peer == INVALID_ADDR) { + log_dbg("No-one to contact."); + return 0; + } - addrs[0] = 0; + if (emergency_peer(&pl) < 0) { + log_err("Could not create emergency peer."); + goto fail_peer; + } - if (dht_wait_running(dht)) - return 0; + log_dbg("Pinging emergency peer " ADDR_FMT32 ".", + ADDR_VAL32(&dht.peer)); - pthread_rwlock_rdlock(&dht->lock); + if (dht_kv_query_contacts(dht.id.data, &pl) < 0) { + log_warn("Failed to bootstrap peer."); + goto fail_query; + } - e = dht_find_entry(dht, key); - if (e != NULL) - addrs[0] = dht_entry_get_addr(dht, e); + peer_list_destroy(&pl); - pthread_rwlock_unlock(&dht->lock); + return 0; + fail_query: + peer_list_destroy(&pl); + fail_peer: + return -EAGAIN; +} - if (addrs[0] != 0) - return addrs[0]; +static void dht_kv_check_contacts(void) +{ + struct list_head cl; + struct list_head pl; - lu = kad_lookup(dht, key, KAD_FIND_VALUE); - if (lu == NULL) - return 0; + list_head_init(&cl); - n = lookup_get_addrs(lu, addrs); - if (n == 0) { - lookup_destroy(lu); - return 0; + dht_kv_contact_list(dht.id.data, &cl, dht.k); + + if (!list_is_empty(&cl)) + goto success; + + contact_list_destroy(&cl); + + list_head_init(&pl); + + if (dht.peer == INVALID_ADDR) { + log_dbg("No-one to contact."); + return; } - lookup_destroy(lu); + if (emergency_peer(&pl) < 0) { + log_err("Could not create emergency peer."); + goto fail_peer; + } - /* Current behaviour is anycast and return the first peer address. */ - if (addrs[0] != dht->addr) - return addrs[0]; + log_dbg("No contacts found, using emergency peer " ADDR_FMT32 ".", + ADDR_VAL32(&dht.peer)); - if (n > 1) - return addrs[1]; + dht_kv_query_contacts(dht.id.data, &pl); - return 0; + peer_list_destroy(&pl); + + return; + success: + contact_list_destroy(&cl); + return; + fail_peer: + return; } -static void * dht_handle_packet(void * o) +static void dht_kv_remove_expired_reqs(void) { - struct dht * dht = (struct dht *) o; + struct list_head * p; + struct list_head * h; + struct timespec now; - assert(dht); + clock_gettime(PTHREAD_COND_CLOCK, &now); - while (true) { - dht_msg_t * msg; - dht_contact_msg_t ** cmsgs; - dht_msg_t resp_msg = DHT_MSG__INIT; - uint64_t addr; - buffer_t buf; - size_t i; - size_t b; - size_t t_expire; - struct cmd * cmd; + pthread_mutex_lock(&dht.reqs.mtx); - pthread_mutex_lock(&dht->mtx); + llist_for_each_safe(p, h, &dht.reqs.ll) { + struct dht_req * e; + e = list_entry(p, struct dht_req, next); + if (IS_EXPIRED(e, &now)) { + log_dbg(KEY_FMT " Removing expired request.", + KEY_VAL(e->key)); + llist_del(&e->next, &dht.reqs.ll); + dht_req_destroy(e); + } + } - pthread_cleanup_push(__cleanup_mutex_unlock, &dht->mtx); + pthread_mutex_unlock(&dht.reqs.mtx); +} - while (list_is_empty(&dht->cmds)) - pthread_cond_wait(&dht->cond, &dht->mtx); +static void value_list_destroy(struct list_head * vl) +{ + struct list_head * p; + struct list_head * h; - cmd = list_last_entry(&dht->cmds, struct cmd, next); - list_del(&cmd->next); + assert(vl != NULL); - pthread_cleanup_pop(true); + list_for_each_safe(p, h, vl) { + struct val_entry * v = list_entry(p, struct val_entry, next); + list_del(&v->next); + val_entry_destroy(v); + } +} - i = shm_du_buff_len(cmd->sdb); +#define MUST_REPLICATE(v, now) ((now)->tv_sec > (v)->t_repl + dht.t_repl) +#define MUST_REPUBLISH(v, now) /* Close to expiry deadline */ \ + (((v)->t_exp - (now)->tv_sec) < (DHT_N_REPUB * dht.t_repl)) +static void dht_entry_get_repl_lists(const struct dht_entry * e, + struct list_head * repl, + struct list_head * rebl, + struct timespec * now) +{ + struct list_head * p; + struct val_entry * n; - msg = dht_msg__unpack(NULL, i, shm_du_buff_head(cmd->sdb)); -#ifndef __DHT_TEST__ - ipcp_sdb_release(cmd->sdb); -#endif - free(cmd); + llist_for_each(p, &e->vals) { + struct val_entry * v = list_entry(p, struct val_entry, next); + if (MUST_REPLICATE(v, now) && !IS_EXPIRED(v, now)) { + n = val_entry_create(v->val, v->t_exp); + if (n == NULL) + continue; - if (msg == NULL) { - log_err("Failed to unpack message."); - continue; + list_add_tail(&n->next, repl); } + } - if (msg->code != KAD_RESPONSE && dht_wait_running(dht)) { - dht_msg__free_unpacked(msg, NULL); - log_dbg("Got a request message when not running."); - continue; + llist_for_each(p, &e->lvals) { + struct val_entry * v = list_entry(p, struct val_entry, next); + if (MUST_REPLICATE(v, now) && MUST_REPUBLISH(v, now)) { + /* Add expire time here, to allow creating val_entry */ + n = val_entry_create(v->val, now->tv_sec + dht.t_exp); + if (n == NULL) + continue; + + list_add_tail(&n->next, rebl); } + } +} - pthread_rwlock_rdlock(&dht->lock); +static int dht_kv_next_values(uint8_t * key, + struct list_head * repl, + struct list_head * rebl) +{ + struct timespec now; + struct list_head * p; + struct list_head * h; + struct dht_entry * e = NULL; - b = dht->b; - t_expire = dht->t_expire; + assert(key != NULL); + assert(repl != NULL); + assert(rebl != NULL); - pthread_rwlock_unlock(&dht->lock); + clock_gettime(CLOCK_REALTIME_COARSE, &now); - if (msg->has_key && msg->key.len != b) { - dht_msg__free_unpacked(msg, NULL); - log_warn("Bad key in message."); - continue; - } + assert(list_is_empty(repl)); + assert(list_is_empty(rebl)); - if (msg->has_s_id && !msg->has_b && msg->s_id.len != b) { - dht_msg__free_unpacked(msg, NULL); - log_warn("Bad source ID in message of type %d.", - msg->code); - continue; - } + pthread_rwlock_rdlock(&dht.db.lock); - tpm_dec(dht->tpm); + if (llist_is_empty(&dht.db.kv.ll)) + goto no_entries; - addr = msg->s_addr; + llist_for_each_safe(p, h, &dht.db.kv.ll) { + e = list_entry(p, struct dht_entry, next); + if (IS_CLOSER(e->key, key)) + continue; /* Already processed */ + } - resp_msg.code = KAD_RESPONSE; - resp_msg.cookie = msg->cookie; + if (e != NULL) { + memcpy(key, e->key, dht.id.len); + dht_entry_get_repl_lists(e, repl, rebl, &now); + } + no_entries: + pthread_rwlock_unlock(&dht.db.lock); - switch(msg->code) { - case KAD_JOIN: - /* Refuse enrollee on check fails. */ - if (msg->alpha != KAD_ALPHA || msg->k != KAD_K) { - log_warn("Parameter mismatch. " - "DHT enrolment refused."); - break; - } + return list_is_empty(repl) && list_is_empty(rebl) ? -ENOENT : 0; +} - if (msg->t_replicate != KAD_T_REPL) { - log_warn("Replication time mismatch. " - "DHT enrolment refused."); +static void dht_kv_replicate_value(const uint8_t * key, + struct val_entry * v, + const struct timespec * now) +{ + assert(MUST_REPLICATE(v, now)); - break; - } + (void) now; - if (msg->t_refresh != KAD_T_REFR) { - log_warn("Refresh time mismatch. " - "DHT enrolment refused."); - break; - } + if (dht_kv_store_remote(key, v->val, v->t_exp) == 0) { + log_dbg(KV_FMT " Replicated.", KV_VAL(key, v->val)); + return; + } - resp_msg.has_alpha = true; - resp_msg.has_b = true; - resp_msg.has_k = true; - resp_msg.has_t_expire = true; - resp_msg.has_t_refresh = true; - resp_msg.has_t_replicate = true; - resp_msg.alpha = KAD_ALPHA; - resp_msg.b = b; - resp_msg.k = KAD_K; - resp_msg.t_expire = t_expire; - resp_msg.t_refresh = KAD_T_REFR; - resp_msg.t_replicate = KAD_T_REPL; - break; - case KAD_FIND_VALUE: - buf = dht_retrieve(dht, msg->key.data); - if (buf.len != 0) { - resp_msg.n_addrs = buf.len; - resp_msg.addrs = (uint64_t *) buf.data; - break; - } - /* FALLTHRU */ - case KAD_FIND_NODE: - /* Return k closest contacts. */ - resp_msg.n_contacts = - dht_get_contacts(dht, msg->key.data, &cmsgs); - resp_msg.contacts = cmsgs; - break; - case KAD_STORE: - if (msg->n_contacts < 1) { - log_warn("No contacts in store message."); - break; - } + log_dbg(KV_FMT " Replication failed.", KV_VAL(key, v->val)); - if (!msg->has_t_expire) { - log_warn("No expiry time in store message."); - break; - } + list_del(&v->next); + val_entry_destroy(v); +} - kad_add(dht, *msg->contacts, msg->n_contacts, - msg->t_expire); - break; - case KAD_RESPONSE: - kad_handle_response(dht, msg); - break; - default: - assert(false); - break; - } +static void dht_kv_republish_value(const uint8_t * key, + struct val_entry * v, + const struct timespec * now) +{ + assert(MUST_REPLICATE(v, now)); - if (msg->code != KAD_JOIN) { - pthread_rwlock_wrlock(&dht->lock); - if (dht_get_state(dht) == DHT_JOINING && - dht->buckets == NULL) { - pthread_rwlock_unlock(&dht->lock); - goto finish; - } + if (MUST_REPUBLISH(v, now)) + assert(v->t_exp >= now->tv_sec + dht.t_exp); - if (dht_update_bucket(dht, msg->s_id.data, addr)) - log_warn("Failed to update bucket."); - pthread_rwlock_unlock(&dht->lock); - } + if (dht_kv_store_remote(key, v->val, v->t_exp) == 0) { + log_dbg(KV_FMT " Republished.", KV_VAL(key, v->val)); + return; + } - if (msg->code < KAD_STORE && send_msg(dht, &resp_msg, addr) < 0) - log_warn("Failed to send response."); + if (MUST_REPUBLISH(v, now)) + log_warn(KV_FMT " Republish failed.", KV_VAL(key, v->val)); + else + log_dbg(KV_FMT " Replication failed.", KV_VAL(key, v->val)); - finish: - dht_msg__free_unpacked(msg, NULL); + list_del(&v->next); + val_entry_destroy(v); +} + +static void dht_kv_update_replication_times(const uint8_t * key, + struct list_head * repl, + struct list_head * rebl, + const struct timespec * now) +{ + struct dht_entry * e; + struct list_head * p; + struct list_head * h; + struct val_entry * v; + + assert(key != NULL); + assert(repl != NULL); + assert(rebl != NULL); + assert(now != NULL); + + pthread_rwlock_wrlock(&dht.db.lock); - if (resp_msg.n_addrs > 0) - free(resp_msg.addrs); + e = __dht_kv_find_entry(key); + if (e == NULL) { + pthread_rwlock_unlock(&dht.db.lock); + return; + } - if (resp_msg.n_contacts == 0) { - tpm_inc(dht->tpm); + list_for_each_safe(p, h, repl) { + struct val_entry * x; + v = list_entry(p, struct val_entry, next); + x = dht_entry_get_val(e, v->val); + if (x == NULL) { + log_err(KV_FMT " Not in vals.", KV_VAL(key, v->val)); continue; } - for (i = 0; i < resp_msg.n_contacts; ++i) - dht_contact_msg__free_unpacked(resp_msg.contacts[i], - NULL); - free(resp_msg.contacts); + x->t_repl = now->tv_sec; - tpm_inc(dht->tpm); + list_del(&v->next); + val_entry_destroy(v); } - return (void *) 0; + list_for_each_safe(p, h, rebl) { + struct val_entry * x; + v = list_entry(p, struct val_entry, next); + x = dht_entry_get_lval(e, v->val); + if (x == NULL) { + log_err(KV_FMT " Not in lvals.", KV_VAL(key, v->val)); + continue; + } + + x->t_repl = now->tv_sec; + if (v->t_exp > x->t_exp) { + x->t_exp = v->t_exp; /* update expiration time */ + } + + list_del(&v->next); + val_entry_destroy(v); + } + + pthread_rwlock_unlock(&dht.db.lock); } -static void dht_post_packet(void * comp, - struct shm_du_buff * sdb) +static void __cleanup_value_list(void * o) { - struct cmd * cmd; - struct dht * dht = (struct dht *) comp; + return value_list_destroy((struct list_head *) o); +} - if (dht_get_state(dht) == DHT_SHUTDOWN) { -#ifndef __DHT_TEST__ - ipcp_sdb_release(sdb); -#endif - return; +static void dht_kv_replicate_values(const uint8_t * key, + struct list_head * repl, + struct list_head * rebl) +{ + struct timespec now; + struct list_head * p; + struct list_head * h; + + clock_gettime(CLOCK_REALTIME_COARSE, &now); + + pthread_cleanup_push(__cleanup_value_list, repl); + pthread_cleanup_push(__cleanup_value_list, rebl); + + list_for_each_safe(p, h, repl) { + struct val_entry * v; + v = list_entry(p, struct val_entry, next); + dht_kv_replicate_value(key, v, &now); } - cmd = malloc(sizeof(*cmd)); - if (cmd == NULL) { - log_err("Command failed. Out of memory."); + list_for_each_safe(p, h, rebl) { + struct val_entry * v; + v = list_entry(p, struct val_entry, next); + dht_kv_republish_value(key, v, &now); + } + + pthread_cleanup_pop(false); + pthread_cleanup_pop(false); + + /* removes non-replicated items from the list */ + dht_kv_update_replication_times(key, repl, rebl, &now); + + if (list_is_empty(repl) && list_is_empty(rebl)) + return; + + log_warn(KEY_FMT " Failed to update replication times.", KEY_VAL(key)); +} + +static void dht_kv_replicate(void) +{ + struct list_head repl; /* list of values to replicate */ + struct list_head rebl; /* list of local values to republish */ + uint8_t * key; + + key = dht_dup_key(dht.id.data); /* dist == 0 */ + if (key == NULL) { + log_err("Replicate: Failed to duplicate DHT ID."); return; } - cmd->sdb = sdb; + list_head_init(&repl); + list_head_init(&rebl); - pthread_mutex_lock(&dht->mtx); + pthread_cleanup_push(free, key); - list_add(&cmd->next, &dht->cmds); + while (dht_kv_next_values(key, &repl, &rebl) == 0) { + dht_kv_replicate_values(key, &repl, &rebl); + if (!list_is_empty(&repl)) { + log_warn(KEY_FMT " Replication items left.", + KEY_VAL(key)); + value_list_destroy(&repl); + } - pthread_cond_signal(&dht->cond); + if (!list_is_empty(&rebl)) { + log_warn(KEY_FMT " Republish items left.", + KEY_VAL(key)); + value_list_destroy(&rebl); + } + } - pthread_mutex_unlock(&dht->mtx); + pthread_cleanup_pop(true); } -void dht_destroy(void * dir) +static void dht_kv_refresh_contacts(void) { - struct dht * dht; struct list_head * p; struct list_head * h; + struct list_head rl; /* refresh list */ + struct timespec now; - dht = (struct dht *) dir; - if (dht == NULL) - return; + list_head_init(&rl); -#ifndef __DHT_TEST__ - tpm_stop(dht->tpm); + clock_gettime(CLOCK_REALTIME_COARSE, &now); - tpm_destroy(dht->tpm); -#endif - if (dht_get_state(dht) == DHT_RUNNING) { - dht_set_state(dht, DHT_SHUTDOWN); - pthread_cancel(dht->worker); - pthread_join(dht->worker, NULL); - } + pthread_rwlock_rdlock(&dht.db.lock); - pthread_rwlock_wrlock(&dht->lock); + __dht_kv_bucket_refresh_list(dht.db.contacts.root, now.tv_sec, &rl); - list_for_each_safe(p, h, &dht->cmds) { - struct cmd * c = list_entry(p, struct cmd, next); + pthread_rwlock_unlock(&dht.db.lock); + + list_for_each_safe(p, h, &rl) { + struct contact * c; + c = list_entry(p, struct contact, next); + log_dbg(PEER_FMT " Refreshing contact.", + PEER_VAL(c->id, c->addr)); + dht_kv_query_contacts(c->id, NULL); list_del(&c->next); -#ifndef __DHT_TEST__ - ipcp_sdb_release(c->sdb); -#endif - free(c); + contact_destroy(c); } - list_for_each_safe(p, h, &dht->entries) { - struct dht_entry * e = list_entry(p, struct dht_entry, next); - list_del(&e->next); - dht_entry_destroy(e); - } + assert(list_is_empty(&rl)); +} - list_for_each_safe(p, h, &dht->requests) { - struct kad_req * r = list_entry(p, struct kad_req, next); - list_del(&r->next); - kad_req_destroy(r); - } +static void (*tasks[])(void) = { + dht_kv_check_contacts, + dht_kv_remove_expired_entries, + dht_kv_remove_expired_reqs, + dht_kv_replicate, + dht_kv_refresh_contacts, + NULL +}; - list_for_each_safe(p, h, &dht->refs) { - struct ref_entry * e = list_entry(p, struct ref_entry, next); - list_del(&e->next); - ref_entry_destroy(e); +static void * work(void * o) +{ + struct timespec now = TIMESPEC_INIT_MS(1); + time_t intv; + size_t n; /* number of tasks */ + + n = sizeof(tasks) / sizeof(tasks[0]) - 1; /* last is NULL */ + + (void) o; + + while (dht_kv_seed_bootstrap_peer() == -EAGAIN) { + ts_add(&now, &now, &now); /* exponential backoff */ + if (now.tv_sec > 1) /* cap at 1 second */ + now.tv_sec = 1; + nanosleep(&now, NULL); } - list_for_each_safe(p, h, &dht->lookups) { - struct lookup * l = list_entry(p, struct lookup, next); - list_del(&l->next); - lookup_destroy(l); + intv = gcd(dht.t_exp, (dht.t_exp - DHT_N_REPUB * dht.t_repl)); + intv = gcd(intv, gcd(dht.t_repl, dht.t_refr)) / 2; + intv = MAX(1, intv / n); + + log_dbg("DHT worker starting %ld seconds interval.", intv * n); + + while (true) { + int i = 0; + while (tasks[i] != NULL) { + tasks[i++](); + sleep(intv); + } } - pthread_rwlock_unlock(&dht->lock); + return (void *) 0; +} + +int dht_start(void) +{ + dht.state = DHT_RUNNING; + + if (tpm_start(dht.tpm)) + goto fail_tpm_start; - if (dht->buckets != NULL) - bucket_destroy(dht->buckets); +#ifndef __DHT_TEST__ + if (pthread_create(&dht.worker, NULL, work, NULL)) { + log_err("Failed to create DHT worker thread."); + goto fail_worker; + } - bmp_destroy(dht->cookies); + dht.eid = dt_reg_comp(&dht, &dht_post_packet, DHT); + if ((int) dht.eid < 0) { + log_err("Failed to register DHT component."); + goto fail_reg; + } +#else + (void) work; +#endif + return 0; +#ifndef __DHT_TEST__ + fail_reg: + pthread_cancel(dht.worker); + pthread_join(dht.worker, NULL); + fail_worker: + tpm_stop(dht.tpm); +#endif + fail_tpm_start: + dht.state = DHT_INIT; + return -1; +} - pthread_mutex_destroy(&dht->mtx); +void dht_stop(void) +{ + assert(dht.state == DHT_RUNNING); - pthread_rwlock_destroy(&dht->lock); +#ifndef __DHT_TEST__ + dt_unreg_comp(dht.eid); - free(dht->id); + pthread_cancel(dht.worker); + pthread_join(dht.worker, NULL); +#endif + tpm_stop(dht.tpm); - free(dht); + dht.state = DHT_INIT; } -static void * join_thr(void * o) +int dht_init(struct dir_dht_config * conf) { - struct join_info * info = (struct join_info *) o; - struct lookup * lu; - size_t retr = 0; + struct timespec now; + pthread_condattr_t cattr; - assert(info); + assert(conf != NULL); - while (kad_join(info->dht, info->addr)) { - if (dht_get_state(info->dht) == DHT_SHUTDOWN) { - log_dbg("DHT enrollment aborted."); - goto finish; - } + clock_gettime(CLOCK_REALTIME_COARSE, &now); - if (retr++ == KAD_JOIN_RETR) { - dht_set_state(info->dht, DHT_INIT); - log_warn("DHT enrollment attempt failed."); - goto finish; - } +#ifndef __DHT_TEST__ + dht.id.len = ipcp_dir_hash_len(); + dht.addr = addr_auth_address(); +#else + dht.id.len = DHT_TEST_KEY_LEN; + dht.addr = DHT_TEST_ADDR; +#endif + dht.t0 = now.tv_sec; + dht.alpha = conf->params.alpha; + dht.k = conf->params.k; + dht.t_exp = conf->params.t_expire; + dht.t_refr = conf->params.t_refresh; + dht.t_repl = conf->params.t_replicate; + dht.peer = conf->peer; + + dht.magic = generate_cookie(); + + /* Send my address on enrollment */ + conf->peer = dht.addr; + + dht.id.data = generate_id(); + if (dht.id.data == NULL) { + log_err("Failed to create DHT ID."); + goto fail_id; + } + + list_head_init(&dht.cmds.list); - sleep(KAD_JOIN_INTV); + if (pthread_mutex_init(&dht.cmds.mtx, NULL)) { + log_err("Failed to initialize command mutex."); + goto fail_cmds_mutex; } - dht_set_state(info->dht, DHT_RUNNING); + if (pthread_cond_init(&dht.cmds.cond, NULL)) { + log_err("Failed to initialize command condvar."); + goto fail_cmds_cond; + } - lu = kad_lookup(info->dht, info->dht->id, KAD_FIND_NODE); - if (lu != NULL) - lookup_destroy(lu); + llist_init(&dht.reqs.ll); - finish: - free(info); + if (pthread_mutex_init(&dht.reqs.mtx, NULL)) { + log_err("Failed to initialize request mutex."); + goto fail_reqs_mutex; + } - return (void *) 0; -} + if (pthread_condattr_init(&cattr)) { + log_err("Failed to initialize request condvar attributes."); + goto fail_cattr; + } +#ifndef __APPLE__ + if (pthread_condattr_setclock(&cattr, PTHREAD_COND_CLOCK)) { + log_err("Failed to set request condvar clock."); + goto fail_cattr; + } +#endif + if (pthread_cond_init(&dht.reqs.cond, &cattr)) { + log_err("Failed to initialize request condvar."); + goto fail_reqs_cond; + } -static void handle_event(void * self, - int event, - const void * o) -{ - struct dht * dht = (struct dht *) self; + llist_init(&dht.db.kv.ll); + dht.db.kv.vals = 0; + dht.db.kv.lvals = 0; - if (event == NOTIFY_DT_CONN_ADD) { - pthread_t thr; - struct join_info * inf; - struct conn * c = (struct conn *) o; - struct timespec slack = TIMESPEC_INIT_MS(DHT_ENROLL_SLACK); + if (pthread_rwlock_init(&dht.db.lock, NULL)) { + log_err("Failed to initialize store rwlock."); + goto fail_rwlock; + } - /* Give the pff some time to update for the new link. */ - nanosleep(&slack, NULL); + dht.db.contacts.root = bucket_create(); + if (dht.db.contacts.root == NULL) { + log_err("Failed to create DHT buckets."); + goto fail_buckets; + } - switch(dht_get_state(dht)) { - case DHT_INIT: - inf = malloc(sizeof(*inf)); - if (inf == NULL) - break; + if (rib_reg(DHT, &r_ops) < 0) { + log_err("Failed to register DHT RIB operations."); + goto fail_rib_reg; + } - inf->dht = dht; - inf->addr = c->conn_info.addr; - - if (dht_set_state(dht, DHT_JOINING) == 0 || - dht_wait_running(dht)) { - if (pthread_create(&thr, NULL, join_thr, inf)) { - dht_set_state(dht, DHT_INIT); - free(inf); - return; - } - pthread_detach(thr); - } else { - free(inf); - } - break; - case DHT_RUNNING: - /* - * FIXME: this lookup for effiency reasons - * causes a SEGV when stressed with rapid - * enrollments. - * lu = kad_lookup(dht, dht->id, KAD_FIND_NODE); - * if (lu != NULL) - * lookup_destroy(lu); - */ - break; - default: - break; - } + dht.tpm = tpm_create(2, 1, dht_handle_packet, NULL); + if (dht.tpm == NULL) { + log_err("Failed to create TPM for DHT."); + goto fail_tpm_create; } + + if (dht_kv_update_contacts(dht.id.data, dht.addr) < 0) + log_warn("Failed to update contacts with DHT ID."); + + pthread_condattr_destroy(&cattr); +#ifndef __DHT_TEST__ + log_info("DHT initialized."); + log_dbg(" ID: " HASH_FMT64 " [%zu bytes].", + HASH_VAL64(dht.id.data), dht.id.len); + log_dbg(" address: " ADDR_FMT32 ".", ADDR_VAL32(&dht.addr)); + log_dbg(" peer: " ADDR_FMT32 ".", ADDR_VAL32(&dht.peer)); + log_dbg(" magic cookie: " HASH_FMT64 ".", HASH_VAL64(&dht.magic)); + log_info(" parameters: alpha=%u, k=%zu, t_exp=%ld, " + "t_refr=%ld, t_replicate=%ld.", + dht.alpha, dht.k, dht.t_exp, dht.t_refr, dht.t_repl); +#endif + dht.state = DHT_INIT; + + return 0; + + fail_tpm_create: + rib_unreg(DHT); + fail_rib_reg: + bucket_destroy(dht.db.contacts.root); + fail_buckets: + pthread_rwlock_destroy(&dht.db.lock); + fail_rwlock: + pthread_cond_destroy(&dht.reqs.cond); + fail_reqs_cond: + pthread_condattr_destroy(&cattr); + fail_cattr: + pthread_mutex_destroy(&dht.reqs.mtx); + fail_reqs_mutex: + pthread_cond_destroy(&dht.cmds.cond); + fail_cmds_cond: + pthread_mutex_destroy(&dht.cmds.mtx); + fail_cmds_mutex: + freebuf(dht.id); + fail_id: + return -1; } -void * dht_create(void) +void dht_fini(void) { - struct dht * dht; + struct list_head * p; + struct list_head * h; - dht = malloc(sizeof(*dht)); - if (dht == NULL) - goto fail_malloc; + rib_unreg(DHT); - dht->buckets = NULL; + tpm_destroy(dht.tpm); - list_head_init(&dht->entries); - list_head_init(&dht->requests); - list_head_init(&dht->refs); - list_head_init(&dht->lookups); - list_head_init(&dht->cmds); + pthread_mutex_lock(&dht.cmds.mtx); - if (pthread_rwlock_init(&dht->lock, NULL)) - goto fail_rwlock; + list_for_each_safe(p, h, &dht.cmds.list) { + struct cmd * c = list_entry(p, struct cmd, next); + list_del(&c->next); + freebuf(c->cbuf); + free(c); + } - if (pthread_mutex_init(&dht->mtx, NULL)) - goto fail_mutex; + pthread_mutex_unlock(&dht.cmds.mtx); - if (pthread_cond_init(&dht->cond, NULL)) - goto fail_cond; + pthread_cond_destroy(&dht.cmds.cond); + pthread_mutex_destroy(&dht.cmds.mtx); - dht->cookies = bmp_create(DHT_MAX_REQS, 1); - if (dht->cookies == NULL) - goto fail_bmp; + pthread_mutex_lock(&dht.reqs.mtx); - dht->b = 0; - dht->id = NULL; -#ifndef __DHT_TEST__ - dht->addr = ipcpi.dt_addr; - dht->tpm = tpm_create(2, 1, dht_handle_packet, dht); - if (dht->tpm == NULL) - goto fail_tpm_create; + llist_for_each_safe(p, h, &dht.reqs.ll) { + struct dht_req * r = list_entry(p, struct dht_req, next); + llist_del(&r->next, &dht.reqs.ll); + dht_req_destroy(r); + } - if (tpm_start(dht->tpm)) - goto fail_tpm_start; + pthread_mutex_unlock(&dht.reqs.mtx); - dht->eid = dt_reg_comp(dht, &dht_post_packet, DHT); - if ((int) dht->eid < 0) - goto fail_tpm_start; + pthread_cond_destroy(&dht.reqs.cond); + pthread_mutex_destroy(&dht.reqs.mtx); - if (notifier_reg(handle_event, dht)) - goto fail_notifier_reg; -#else - (void) handle_event; - (void) dht_handle_packet; - (void) dht_post_packet; -#endif - dht->state = DHT_INIT; + pthread_rwlock_wrlock(&dht.db.lock); - return (void *) dht; -#ifndef __DHT_TEST__ - fail_notifier_reg: - tpm_stop(dht->tpm); - fail_tpm_start: - tpm_destroy(dht->tpm); - fail_tpm_create: - bmp_destroy(dht->cookies); -#endif - fail_bmp: - pthread_cond_destroy(&dht->cond); - fail_cond: - pthread_mutex_destroy(&dht->mtx); - fail_mutex: - pthread_rwlock_destroy(&dht->lock); - fail_rwlock: - free(dht); - fail_malloc: - return NULL; + llist_for_each_safe(p, h, &dht.db.kv.ll) { + struct dht_entry * e = list_entry(p, struct dht_entry, next); + llist_del(&e->next, &dht.db.kv.ll); + dht_entry_destroy(e); + } + + if (dht.db.contacts.root != NULL) + bucket_destroy(dht.db.contacts.root); + + pthread_rwlock_unlock(&dht.db.lock); + + pthread_rwlock_destroy(&dht.db.lock); + + assert(llist_is_empty(&dht.db.kv.ll)); + assert(dht.db.kv.vals == 0); + assert(dht.db.kv.lvals == 0); + assert(llist_is_empty(&dht.reqs.ll)); + + freebuf(dht.id); } diff --git a/src/ipcpd/unicast/dir/dht.h b/src/ipcpd/unicast/dir/dht.h index 311c6b23..ed3bb9f0 100644 --- a/src/ipcpd/unicast/dir/dht.h +++ b/src/ipcpd/unicast/dir/dht.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Distributed Hash Table based on Kademlia * @@ -30,22 +30,19 @@ #include <stdint.h> #include <sys/types.h> -void * dht_create(void); +int dht_init(struct dir_dht_config * conf); -void dht_destroy(void * dir); +void dht_fini(void); -int dht_bootstrap(void * dir); +int dht_start(void); -int dht_reg(void * dir, - const uint8_t * key); +void dht_stop(void); -int dht_unreg(void * dir, - const uint8_t * key); +int dht_reg(const uint8_t * key); -uint64_t dht_query(void * dir, - const uint8_t * key); +int dht_unreg(const uint8_t * key); -int dht_wait_running(void * dir); +uint64_t dht_query(const uint8_t * key); extern struct dir_ops dht_dir_ops; diff --git a/src/ipcpd/unicast/dir/dht.proto b/src/ipcpd/unicast/dir/dht.proto index 4c5b06db..02b6b341 100644 --- a/src/ipcpd/unicast/dir/dht.proto +++ b/src/ipcpd/unicast/dir/dht.proto @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * DHT protocol, based on Kademlia * @@ -27,19 +27,32 @@ message dht_contact_msg { required uint64 addr = 2; } +message dht_find_req_msg { + required uint64 cookie = 1; + required bytes key = 2; +} + +message dht_find_node_rsp_msg { + required uint64 cookie = 1; + required bytes key = 2; + repeated dht_contact_msg contacts = 3; +} + +message dht_find_value_rsp_msg { + repeated bytes values = 1; +} + +message dht_store_msg { + required bytes key = 1; + required bytes val = 2; + required uint32 exp = 3; +} + message dht_msg { - required uint32 code = 1; - required uint32 cookie = 2; - required uint64 s_addr = 3; - optional bytes s_id = 4; - optional bytes key = 5; - repeated uint64 addrs = 6; - repeated dht_contact_msg contacts = 7; - // enrolment parameters - optional uint32 alpha = 8; - optional uint32 b = 9; - optional uint32 k = 10; - optional uint32 t_expire = 11; - optional uint32 t_refresh = 12; - optional uint32 t_replicate = 13; + required uint32 code = 1; + required dht_contact_msg src = 2; + optional dht_store_msg store = 3; + optional dht_find_req_msg find = 4; + optional dht_find_node_rsp_msg node = 5; + optional dht_find_value_rsp_msg val = 6; } diff --git a/src/ipcpd/unicast/dir/ops.h b/src/ipcpd/unicast/dir/ops.h index 6ff61ce6..6c336ee0 100644 --- a/src/ipcpd/unicast/dir/ops.h +++ b/src/ipcpd/unicast/dir/ops.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Directory policy ops * @@ -23,24 +23,20 @@ #ifndef OUROBOROS_IPCPD_UNICAST_DIR_OPS_H #define OUROBOROS_IPCPD_UNICAST_DIR_OPS_H - struct dir_ops { - void * (* create)(void); + int (* init)(void * config); - void (* destroy)(void * dir); + void (* fini)(void); - int (* bootstrap)(void * dir); + int (* start)(void); - int (* reg)(void * dir, - const uint8_t * hash); + void (* stop)(void); - int (* unreg)(void * dir, - const uint8_t * hash); + int (* reg)(const uint8_t * hash); - uint64_t (* query)(void * dir, - const uint8_t * hash); + int (* unreg)(const uint8_t * hash); - int (* wait_running)(void * dir); + uint64_t (* query)(const uint8_t * hash); }; #endif /* OUROBOROS_IPCPD_UNICAST_DIR_OPS_H */ diff --git a/src/ipcpd/unicast/dir/pol.h b/src/ipcpd/unicast/dir/pol.h index eae4b2e7..8ccf4f95 100644 --- a/src/ipcpd/unicast/dir/pol.h +++ b/src/ipcpd/unicast/dir/pol.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Directory policies * diff --git a/src/ipcpd/unicast/dir/tests/CMakeLists.txt b/src/ipcpd/unicast/dir/tests/CMakeLists.txt index c850e41d..eded823f 100644 --- a/src/ipcpd/unicast/dir/tests/CMakeLists.txt +++ b/src/ipcpd/unicast/dir/tests/CMakeLists.txt @@ -3,34 +3,36 @@ get_filename_component(CURRENT_SOURCE_PARENT_DIR get_filename_component(CURRENT_BINARY_PARENT_DIR ${CMAKE_CURRENT_BINARY_DIR} DIRECTORY) -include_directories(${CMAKE_CURRENT_SOURCE_DIR}) -include_directories(${CMAKE_CURRENT_BINARY_DIR}) - -include_directories(${CURRENT_SOURCE_PARENT_DIR}) -include_directories(${CURRENT_BINARY_PARENT_DIR}) - -include_directories(${CMAKE_SOURCE_DIR}/include) -include_directories(${CMAKE_BINARY_DIR}/include) - get_filename_component(PARENT_PATH ${CMAKE_CURRENT_SOURCE_DIR} DIRECTORY) get_filename_component(PARENT_DIR ${PARENT_PATH} NAME) +compute_test_prefix() + create_test_sourcelist(${PARENT_DIR}_tests test_suite.c # Add new tests here dht_test.c - ) +) -protobuf_generate_c(DHT_PROTO_SRCS KAD_PROTO_HDRS ../dht.proto) -add_executable(${PARENT_DIR}_test EXCLUDE_FROM_ALL ${${PARENT_DIR}_tests} +protobuf_generate_c(DHT_PROTO_SRCS KAD_PROTO_HDRS ${CURRENT_SOURCE_PARENT_DIR}/dht.proto) +add_executable(${PARENT_DIR}_test ${${PARENT_DIR}_tests} ${DHT_PROTO_SRCS}) -target_link_libraries(${PARENT_DIR}_test ouroboros-common) -add_dependencies(check ${PARENT_DIR}_test) +target_include_directories(${PARENT_DIR}_test PRIVATE + ${CMAKE_CURRENT_SOURCE_DIR} + ${CMAKE_CURRENT_BINARY_DIR} + ${CURRENT_SOURCE_PARENT_DIR} + ${CURRENT_BINARY_PARENT_DIR} + ${CMAKE_SOURCE_DIR}/include + ${CMAKE_BINARY_DIR}/include + ${CMAKE_SOURCE_DIR}/src/ipcpd + ${CMAKE_BINARY_DIR}/src/ipcpd + ${CMAKE_SOURCE_DIR}/src/ipcpd/unicast + ${CMAKE_BINARY_DIR}/src/ipcpd/unicast +) + +disable_test_logging_for_target(${PARENT_DIR}_test) -set(tests_to_run ${${PARENT_DIR}_tests}) -remove(tests_to_run test_suite.c) +target_link_libraries(${PARENT_DIR}_test ouroboros-common) +add_dependencies(build_tests ${PARENT_DIR}_test) -foreach (test ${tests_to_run}) - get_filename_component(test_name ${test} NAME_WE) - add_test(${test_name} ${C_TEST_PATH}/${PARENT_DIR}_test ${test_name}) -endforeach (test) +ouroboros_register_tests(TARGET ${PARENT_DIR}_test TESTS ${${PARENT_DIR}_tests}) diff --git a/src/ipcpd/unicast/dir/tests/dht_test.c b/src/ipcpd/unicast/dir/tests/dht_test.c index bea2c3e7..1f7026b3 100644 --- a/src/ipcpd/unicast/dir/tests/dht_test.c +++ b/src/ipcpd/unicast/dir/tests/dht_test.c @@ -1,10 +1,9 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Unit tests of the DHT * * Dimitri Staessens <dimitri@ouroboros.rocks> - * Sander Vrijders <sander@ouroboros.rocks> * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as @@ -21,76 +20,1895 @@ */ #define __DHT_TEST__ -#define DHT_TEST_KEY_LEN 32 -#include "dht.c" +#if defined(__linux__) || defined(__CYGWIN__) +#define _DEFAULT_SOURCE +#else +#define _POSIX_C_SOURCE 200112L +#endif + +#include <test/test.h> +#include <ouroboros/list.h> +#include <ouroboros/utils.h> -#include <pthread.h> +#include "dht.pb-c.h" + +#include <assert.h> +#include <inttypes.h> #include <time.h> #include <stdlib.h> #include <stdio.h> -#define CONTACTS 1000 +#define DHT_MAX_RAND_SIZE 64 +#define DHT_TEST_KEY_LEN 32 +#define DHT_TEST_ADDR 0x1234567890abcdefULL -int dht_test(int argc, - char ** argv) +/* forward declare for use in the dht code */ +/* Packet sink for DHT tests */ +struct { + bool enabled; + + struct llist msgs; +} sink; + +struct message { + struct list_head next; + void * msg; + uint64_t dst; +}; + +static int sink_send_msg(buffer_t * pkt, + uint64_t addr) { - struct dht * dht; - uint8_t key[DHT_TEST_KEY_LEN]; - size_t i; + struct message * m; - (void) argc; - (void) argv; + assert(pkt != NULL); + assert(addr != 0); + + if (!sink.enabled) + goto finish; + + m = malloc(sizeof(*m)); + if (m == NULL) { + printf("Failed to malloc message."); + goto fail_malloc; + } + + m->msg = dht_msg__unpack(NULL, pkt->len, pkt->data); + if (m->msg == NULL) + goto fail_unpack; + + m->dst = addr; + + llist_add_tail(&m->next, &sink.msgs); + + finish: + freebuf(*pkt); + + return 0; + fail_unpack: + free(m); + fail_malloc: + freebuf(*pkt); + return -1; +} + +#include "dht.c" + +/* Test helpers */ + +static void sink_init(void) +{ + llist_init(&sink.msgs); + sink.enabled = true; +} + +static void sink_clear(void) +{ + struct list_head * p; + struct list_head * h; + + llist_for_each_safe(p, h, &sink.msgs) { + struct message * m = list_entry(p, struct message, next); + llist_del(&m->next, &sink.msgs); + dht_msg__free_unpacked((dht_msg_t *) m->msg, NULL); + free(m); + } + + assert(llist_is_empty(&sink.msgs)); +} + +static void sink_fini(void) +{ + sink_clear(); + sink.enabled = false; +} + +static dht_msg_t * sink_read(void) +{ + struct message * m; + dht_msg_t * msg; + + if (llist_is_empty(&sink.msgs)) + return NULL; + + m = llist_first_entry(&sink.msgs, struct message, next); + + llist_del(&m->next, &sink.msgs); + + msg = m->msg; + + free(m); + + return (dht_msg_t *) msg; +} + +static const buffer_t test_val = { + .data = (uint8_t *) "test_value", + .len = 10 +}; + +static const buffer_t test_val2 = { + .data = (uint8_t *) "test_value_2", + .len = 12 +}; + +static int random_value_len(buffer_t * b) +{ + assert(b != NULL); + assert(b->len > 0 && b->len <= DHT_MAX_RAND_SIZE); + + b->data = malloc(b->len); + if (b->data == NULL) + goto fail_malloc; + + random_buffer(b->data, b->len); + + return 0; + + fail_malloc: + return -ENOMEM; +} + +static int random_value(buffer_t * b) +{ + assert(b != NULL); + + b->len = rand() % DHT_MAX_RAND_SIZE + 1; + + return random_value_len(b); +} + +static int fill_dht_with_contacts(size_t n) +{ + size_t i; + uint8_t * id; + + for (i = 0; i < n; i++) { + uint64_t addr = generate_cookie(); + id = generate_id(); + if (id == NULL) + goto fail_id; + + if (dht_kv_update_contacts(id, addr) < 0) + goto fail_update; + free(id); + } + + return 0; + + fail_update: + free(id); + fail_id: + return -1; +} + +static int fill_store_with_random_values(const uint8_t * key, + size_t len, + size_t n_values) +{ + buffer_t val; + struct timespec now; + size_t i; + uint8_t * _key; + + clock_gettime(CLOCK_REALTIME_COARSE, &now); + + for (i = 0; i < n_values; ++i) { + if (key != NULL) + _key = (uint8_t *) key; + else { + _key = generate_id(); + if (_key == NULL) + goto fail_key; + } + + if (len == 0) + val.len = rand() % DHT_MAX_RAND_SIZE + 1; + else + val.len = len; + + if (random_value_len(&val) < 0) + goto fail_value; + + if (dht_kv_store(_key, val, now.tv_sec + 10) < 0) + goto fail_store; + + freebuf(val); + if (key == NULL) + free(_key); + } + + return 0; + + fail_store: + freebuf(val); + fail_value: + free(_key); + fail_key: + return -1; +} + +static int random_contact_list(dht_contact_msg_t *** contacts, + size_t max) +{ + size_t i; - dht = dht_create(); - if (dht == NULL) { + assert(contacts != NULL); + assert(max > 0); + + *contacts = malloc(max * sizeof(**contacts)); + if (*contacts == NULL) + goto fail_malloc; + + for (i = 0; i < max; i++) { + (*contacts)[i] = malloc(sizeof(*(*contacts)[i])); + if ((*contacts)[i] == NULL) + goto fail_contacts; + + dht_contact_msg__init((*contacts)[i]); + + (*contacts)[i]->id.data = generate_id(); + if ((*contacts)[i]->id.data == NULL) + goto fail_contact; + + (*contacts)[i]->id.len = dht.id.len; + (*contacts)[i]->addr = generate_cookie(); + } + + return 0; + + fail_contact: + dht_contact_msg__free_unpacked((*contacts)[i], NULL); + fail_contacts: + while (i-- > 0) + free((*contacts)[i]); + free(*contacts); + fail_malloc: + return -ENOMEM; +} + +static void clear_contacts(dht_contact_msg_t ** contacts, + size_t len) +{ + size_t i; + + assert(contacts != NULL); + if (*contacts == NULL) + return; + + for (i = 0; i < len; ++i) + dht_contact_msg__free_unpacked((contacts)[i], NULL); + + free(*contacts); + *contacts = NULL; +} + +/* Start of actual tests */ +static struct dir_dht_config test_dht_config = { + .params = { + .alpha = 3, + .k = 8, + .t_expire = 86400, + .t_refresh = 900, + .t_replicate = 900 + } +}; + +static int test_dht_init_fini(void) +{ + TEST_START(); + + if (dht_init(&test_dht_config) < 0) { printf("Failed to create dht.\n"); - return -1; + goto fail_init; } - dht_destroy(dht); + dht_fini(); + + TEST_SUCCESS(); - dht = dht_create(); - if (dht == NULL) { - printf("Failed to re-create dht.\n"); - return -1; + return TEST_RC_SUCCESS; + fail_init: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_start_stop(void) +{ + TEST_START(); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; } - if (dht_bootstrap(dht)) { - printf("Failed to bootstrap dht.\n"); - dht_destroy(dht); - return -1; + if (dht_start() < 0) { + printf("Failed to start dht.\n"); + goto fail_start; } - dht_destroy(dht); + dht_stop(); + + dht_fini(); - dht = dht_create(); - if (dht == NULL) { - printf("Failed to re-create dht.\n"); - return -1; + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_start: + dht_fini(); + fail_init: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_val_entry_create_destroy(void) +{ + struct val_entry * e; + struct timespec now; + + TEST_START(); + + clock_gettime(CLOCK_REALTIME_COARSE, &now); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; } - if (dht_bootstrap(dht)) { - printf("Failed to bootstrap dht.\n"); - dht_destroy(dht); - return -1; + e = val_entry_create(test_val, now.tv_sec + 10); + if (e == NULL) { + printf("Failed to create val entry.\n"); + goto fail_entry; } - for (i = 0; i < CONTACTS; ++i) { - uint64_t addr; - random_buffer(&addr, sizeof(addr)); - random_buffer(key, DHT_TEST_KEY_LEN); - pthread_rwlock_wrlock(&dht->lock); - if (dht_update_bucket(dht, key, addr)) { - pthread_rwlock_unlock(&dht->lock); - printf("Failed to update bucket.\n"); - dht_destroy(dht); - return -1; + val_entry_destroy(e); + + dht_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_entry: + dht_fini(); + fail_init: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_entry_create_destroy(void) +{ + struct dht_entry * e; + + TEST_START(); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + e = dht_entry_create(dht.id.data); + if (e == NULL) { + printf("Failed to create dht entry.\n"); + goto fail_entry; + } + + dht_entry_destroy(e); + + dht_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_entry: + dht_fini(); + fail_init: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_entry_update_get_val(void) +{ + struct dht_entry * e; + struct val_entry * v; + struct timespec now; + + TEST_START(); + + clock_gettime(CLOCK_REALTIME_COARSE, &now); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + e = dht_entry_create(dht.id.data); + if (e == NULL) { + printf("Failed to create dht entry.\n"); + goto fail_entry; + } + + if (dht_entry_get_val(e, test_val) != NULL) { + printf("Found value in empty dht entry.\n"); + goto fail_get; + } + + if (dht_entry_update_val(e, test_val, now.tv_sec + 10) < 0) { + printf("Failed to update dht entry value.\n"); + goto fail_get; + } + + if (dht_entry_get_val(e, test_val2) != NULL) { + printf("Found value in dht entry with different key.\n"); + goto fail_get; + } + + v = dht_entry_get_val(e, test_val); + if (v == NULL) { + printf("Failed to get value from dht entry.\n"); + goto fail_get; + } + + if (v->val.len != test_val.len) { + printf("Length in dht entry does not match expected.\n"); + goto fail_get; + } + + if(memcmp(v->val.data, test_val.data, test_val.len) != 0) { + printf("Data in dht entry does not match expected.\n"); + goto fail_get; + } + + if (dht_entry_update_val(e, test_val, now.tv_sec + 15) < 0) { + printf("Failed to update exsting dht entry value.\n"); + goto fail_get; + } + + if (v->t_exp != now.tv_sec + 15) { + printf("Expiration time in dht entry value not updated.\n"); + goto fail_get; + } + + if (dht_entry_update_val(e, test_val, now.tv_sec + 5) < 0) { + printf("Failed to update existing dht entry value (5).\n"); + goto fail_get; + } + + if (v->t_exp != now.tv_sec + 15) { + printf("Expiration time in dht entry shortened.\n"); + goto fail_get; + } + + if (dht_entry_get_val(e, test_val) != v) { + printf("Wrong value in dht entry found after update.\n"); + goto fail_get; + } + + dht_entry_destroy(e); + + dht_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_get: + dht_entry_destroy(e); + fail_entry: + dht_fini(); + fail_init: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_entry_update_get_lval(void) +{ + struct dht_entry * e; + struct val_entry * v; + struct timespec now; + + TEST_START(); + + clock_gettime(CLOCK_REALTIME_COARSE, &now); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + e = dht_entry_create(dht.id.data); + if (e == NULL) { + printf("Failed to create dht entry.\n"); + goto fail_entry; + } + + if (dht_entry_get_lval(e, test_val) != NULL) { + printf("Found value in empty dht entry.\n"); + goto fail_get; + } + + if (dht_entry_update_lval(e, test_val) < 0) { + printf("Failed to update dht entry value.\n"); + goto fail_get; + } + + v = dht_entry_get_lval(e, test_val); + if (v== NULL) { + printf("Failed to get value from dht entry.\n"); + goto fail_get; + } + + if (dht_entry_get_lval(e, test_val2) != NULL) { + printf("Found value in dht entry in vals.\n"); + goto fail_get; + } + + if (v->val.len != test_val.len) { + printf("Length in dht entry does not match expected.\n"); + goto fail_get; + } + + if(memcmp(v->val.data, test_val.data, test_val.len) != 0) { + printf("Data in dht entry does not match expected.\n"); + goto fail_get; + } + + if (dht_entry_update_lval(e, test_val) < 0) { + printf("Failed to update existing dht entry value.\n"); + goto fail_get; + } + + if (dht_entry_get_lval(e, test_val) != v) { + printf("Wrong value in dht entry found after update.\n"); + goto fail_get; + } + + dht_entry_destroy(e); + + dht_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_get: + dht_entry_destroy(e); + fail_entry: + dht_fini(); + fail_init: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_kv_contact_create_destroy(void) +{ + struct contact * c; + + TEST_START(); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + c = contact_create(dht.id.data, dht.addr); + if (c == NULL) { + printf("Failed to create contact.\n"); + goto fail_contact; + } + + contact_destroy(c); + + dht_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_contact: + dht_fini(); + fail_init: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_kv_update_bucket(void) +{ + TEST_START(); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + if (fill_dht_with_contacts(1000) < 0) { + printf("Failed to fill bucket with contacts.\n"); + goto fail_update; + } + + dht_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_update: + dht_fini(); + fail_init: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_kv_contact_list(void) +{ + struct list_head cl; + ssize_t len; + ssize_t items; + + TEST_START(); + + list_head_init(&cl); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + items = 5; + + if (fill_dht_with_contacts(items) < 0) { + printf("Failed to fill bucket with contacts.\n"); + goto fail_fill; + } + + len = dht_kv_contact_list(dht.id.data, &cl, dht.k); + if (len < 0) { + printf("Failed to get contact list.\n"); + goto fail_fill; + } + + if (len != items) { + printf("Failed to get contacts (%zu != %zu).\n", len, items); + goto fail_contact_list; + } + + contact_list_destroy(&cl); + + items = 100; + + if (fill_dht_with_contacts(items) < 0) { + printf("Failed to fill bucket with contacts.\n"); + goto fail_fill; + } + + len = dht_kv_contact_list(dht.id.data, &cl, items); + if (len < 0) { + printf("Failed to get contact list.\n"); + goto fail_fill; + } + + if ((size_t) len < dht.k) { + printf("Failed to get contacts (%zu < %zu).\n", len, dht.k); + goto fail_contact_list; + } + + contact_list_destroy(&cl); + + dht_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_contact_list: + contact_list_destroy(&cl); + fail_fill: + dht_fini(); + fail_init: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_kv_get_values(void) +{ + buffer_t * vals; + ssize_t len; + size_t n = sizeof(uint64_t); + + TEST_START(); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + if (fill_store_with_random_values(dht.id.data, n, 3) < 0) { + printf("Failed to fill store with random values.\n"); + goto fail_fill; + } + + len = dht_kv_retrieve(dht.id.data, &vals); + if (len < 0) { + printf("Failed to get values from store.\n"); + goto fail_fill; + } + + if (len != 3) { + printf("Failed to get %ld values (%zu).\n", 3L, len); + goto fail_get_values; + } + + freebufs(vals, len); + + if (fill_store_with_random_values(dht.id.data, n, 20) < 0) { + printf("Failed to fill store with random values.\n"); + goto fail_fill; + } + + len = dht_kv_retrieve(dht.id.data, &vals); + if (len < 0) { + printf("Failed to get values from store.\n"); + goto fail_fill; + } + + if (len != DHT_MAX_VALS) { + printf("Failed to get %d values.\n", DHT_MAX_VALS); + goto fail_get_values; + } + + freebufs(vals, len); + + dht_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_get_values: + freebufs(vals, len); + fail_fill: + dht_fini(); + fail_init: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_kv_find_node_req_msg(void) +{ + dht_msg_t * msg; + dht_msg_t * upk; + size_t len; + uint8_t * buf; + + TEST_START(); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + msg = dht_kv_find_node_req_msg(dht.id.data); + if (msg == NULL) { + printf("Failed to get find node request message.\n"); + goto fail_msg; + } + + if (msg->code != DHT_FIND_NODE_REQ) { + printf("Wrong code in find_node_req message (%s != %s).\n", + dht_code_str[msg->code], + dht_code_str[DHT_FIND_NODE_REQ]); + goto fail_msg; + } + + len = dht_msg__get_packed_size(msg); + if (len == 0) { + printf("Failed to get packed length of find_node_req.\n"); + goto fail_msg; + } + + buf = malloc(len); + if (buf == NULL) { + printf("Failed to malloc find_node_req buf.\n"); + goto fail_msg; + } + + if (dht_msg__pack(msg, buf) != len) { + printf("Failed to pack find_node_req message.\n"); + goto fail_pack; + } + + upk = dht_msg__unpack(NULL, len, buf); + if (upk == NULL) { + printf("Failed to unpack find_value_req message.\n"); + goto fail_unpack; + } + + free(buf); + dht_msg__free_unpacked(msg, NULL); + dht_msg__free_unpacked(upk, NULL); + + dht_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_unpack: + dht_msg__free_unpacked(msg, NULL); + fail_pack: + free(buf); + fail_msg: + dht_fini(); + fail_init: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_kv_find_node_rsp_msg(void) +{ + dht_contact_msg_t ** contacts; + dht_msg_t * msg; + dht_msg_t * upk; + size_t len; + uint8_t * buf; + + TEST_START(); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + msg = dht_kv_find_node_rsp_msg(dht.id.data, 0, &contacts, 0); + if (msg == NULL) { + printf("Failed to get find node response message.\n"); + goto fail_msg; + } + + if (msg->code != DHT_FIND_NODE_RSP) { + printf("Wrong code in find_node_rsp message (%s != %s).\n", + dht_code_str[msg->code], + dht_code_str[DHT_FIND_NODE_RSP]); + goto fail_msg; + } + + len = dht_msg__get_packed_size(msg); + if (len == 0) { + printf("Failed to get packed length of find_node_rsp.\n"); + goto fail_msg; + } + + buf = malloc(len); + if (buf == NULL) { + printf("Failed to malloc find_node_rsp buf.\n"); + goto fail_msg; + } + + if (dht_msg__pack(msg, buf) != len) { + printf("Failed to pack find_node_rsp message.\n"); + goto fail_pack; + } + + upk = dht_msg__unpack(NULL, len, buf); + if (upk == NULL) { + printf("Failed to unpack find_node_rsp message.\n"); + goto fail_unpack; + } + + free(buf); + dht_msg__free_unpacked(msg, NULL); + dht_msg__free_unpacked(upk, NULL); + + dht_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_unpack: + dht_msg__free_unpacked(msg, NULL); + fail_pack: + free(buf); + fail_msg: + dht_fini(); + fail_init: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_kv_find_node_rsp_msg_contacts(void) +{ + dht_contact_msg_t ** contacts; + dht_msg_t * msg; + dht_msg_t * upk; + uint8_t * buf; + size_t len; + ssize_t n; + + TEST_START(); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + if (fill_dht_with_contacts(100) < 0) { + printf("Failed to fill bucket with contacts.\n"); + goto fail_fill; + } + + n = dht_kv_get_contacts(dht.id.data, &contacts); + if (n < 0) { + printf("Failed to get contacts.\n"); + goto fail_fill; + } + + if ((size_t) n < dht.k) { + printf("Failed to get all contacts (%zu < %zu).\n", n, dht.k); + goto fail_fill; + } + + msg = dht_kv_find_node_rsp_msg(dht.id.data, 0, &contacts, n); + if (msg == NULL) { + printf("Failed to build find node response message.\n"); + goto fail_msg; + } + + len = dht_msg__get_packed_size(msg); + if (len == 0) { + printf("Failed to get packed length of find_node_rsp.\n"); + goto fail_msg; + } + + buf = malloc(len); + if (buf == NULL) { + printf("Failed to malloc find_node_rsp buf.\n"); + goto fail_msg; + } + + if (dht_msg__pack(msg, buf) != len) { + printf("Failed to pack find_node_rsp message.\n"); + goto fail_pack; + } + + upk = dht_msg__unpack(NULL, len, buf); + if (upk == NULL) { + printf("Failed to unpack find_node_rsp message.\n"); + goto fail_unpack; + } + + free(buf); + dht_msg__free_unpacked(msg, NULL); + dht_msg__free_unpacked(upk, NULL); + + dht_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_unpack: + dht_msg__free_unpacked(msg, NULL); + fail_pack: + free(buf); + fail_msg: + clear_contacts(contacts, n); + fail_fill: + dht_fini(); + fail_init: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_kv_find_value_req_msg(void) +{ + dht_msg_t * msg; + dht_msg_t * upk; + size_t len; + uint8_t * buf; + + TEST_START(); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + msg = dht_kv_find_value_req_msg(dht.id.data); + if (msg == NULL) { + printf("Failed to build find value request message.\n"); + goto fail_msg; + } + + if (msg->code != DHT_FIND_VALUE_REQ) { + printf("Wrong code in find_value_req message (%s != %s).\n", + dht_code_str[msg->code], + dht_code_str[DHT_FIND_VALUE_REQ]); + goto fail_msg; + } + + len = dht_msg__get_packed_size(msg); + if (len == 0) { + printf("Failed to get packed length of find_value_req.\n"); + goto fail_msg; + } + + buf = malloc(len); + if (buf == NULL) { + printf("Failed to malloc find_node_req buf.\n"); + goto fail_msg; + } + + if (dht_msg__pack(msg, buf) != len) { + printf("Failed to pack find_value_req message.\n"); + goto fail_pack; + } + + upk = dht_msg__unpack(NULL, len, buf); + if (upk == NULL) { + printf("Failed to unpack find_value_req message.\n"); + goto fail_unpack; + } + + free(buf); + dht_msg__free_unpacked(msg, NULL); + dht_msg__free_unpacked(upk, NULL); + + dht_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_unpack: + dht_msg__free_unpacked(msg, NULL); + fail_pack: + free(buf); + fail_msg: + dht_fini(); + fail_init: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_kv_find_value_rsp_msg(void) +{ + dht_msg_t * msg; + dht_msg_t * upk; + size_t len; + uint8_t * buf; + + TEST_START(); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + msg = dht_kv_find_value_rsp_msg(dht.id.data, 0, NULL, 0, NULL, 0); + if (msg == NULL) { + printf("Failed to build find value response message.\n"); + goto fail_msg; + } + + if (msg->code != DHT_FIND_VALUE_RSP) { + printf("Wrong code in find_value_rsp message (%s != %s).\n", + dht_code_str[msg->code], + dht_code_str[DHT_FIND_VALUE_RSP]); + goto fail_msg; + } + + len = dht_msg__get_packed_size(msg); + if (len == 0) { + printf("Failed to get packed length of find_value_rsp.\n"); + goto fail_msg; + } + + buf = malloc(len); + if (buf == NULL) { + printf("Failed to malloc find_value_rsp buf.\n"); + goto fail_msg; + } + + if (dht_msg__pack(msg, buf) != len) { + printf("Failed to pack find_value_rsp message.\n"); + goto fail_pack; + } + + upk = dht_msg__unpack(NULL, len, buf); + if (upk == NULL) { + printf("Failed to unpack find_value_rsp message.\n"); + goto fail_unpack; + } + + free(buf); + dht_msg__free_unpacked(msg, NULL); + dht_msg__free_unpacked(upk, NULL); + + dht_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_unpack: + dht_msg__free_unpacked(msg, NULL); + fail_pack: + free(buf); + fail_msg: + dht_fini(); + fail_init: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_kv_find_value_rsp_msg_contacts(void) +{ + dht_msg_t * msg; + dht_msg_t * upk; + size_t len; + uint8_t * buf; + dht_contact_msg_t ** contacts; + ssize_t n; + + TEST_START(); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + if (fill_dht_with_contacts(100) < 0) { + printf("Failed to fill bucket with contacts.\n"); + goto fail_fill; + } + + n = dht_kv_get_contacts(dht.id.data, &contacts); + if (n < 0) { + printf("Failed to get contacts.\n"); + goto fail_fill; + } + + if ((size_t) n < dht.k) { + printf("Failed to get all contacts (%zu < %zu).\n", n, dht.k); + goto fail_fill; + } + + msg = dht_kv_find_value_rsp_msg(dht.id.data, 0, &contacts, n, NULL, 0); + if (msg == NULL) { + printf("Failed to build find value response message.\n"); + goto fail_msg; + } + + len = dht_msg__get_packed_size(msg); + if (len == 0) { + printf("Failed to get packed length of find_value_rsp.\n"); + goto fail_msg; + } + + buf = malloc(len); + if (buf == NULL) { + printf("Failed to malloc find_value_rsp buf.\n"); + goto fail_msg; + } + + if (dht_msg__pack(msg, buf) != len) { + printf("Failed to pack find_value_rsp message.\n"); + goto fail_pack; + } + + upk = dht_msg__unpack(NULL, len, buf); + if (upk == NULL) { + printf("Failed to unpack find_value_rsp message.\n"); + goto fail_unpack; + } + + free(buf); + dht_msg__free_unpacked(msg, NULL); + dht_msg__free_unpacked(upk, NULL); + + dht_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_unpack: + dht_msg__free_unpacked(msg, NULL); + fail_pack: + free(buf); + fail_msg: + clear_contacts(contacts, n); + fail_fill: + dht_fini(); + fail_init: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_kv_find_value_rsp_msg_values(void) +{ + dht_msg_t * msg; + dht_msg_t * upk; + size_t len; + uint8_t * buf; + buffer_t * values; + size_t i; + uint64_t ck; + + TEST_START(); + + ck = generate_cookie(); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + values = malloc(sizeof(*values) * 8); + if (values == NULL) { + printf("Failed to malloc values.\n"); + goto fail_values; + } + + for (i = 0; i < 8; i++) { + if (random_value(&values[i]) < 0) { + printf("Failed to create random value.\n"); + goto fail_fill; } - pthread_rwlock_unlock(&dht->lock); } - dht_destroy(dht); + msg = dht_kv_find_value_rsp_msg(dht.id.data, ck, NULL, 0, &values, 8); + if (msg == NULL) { + printf("Failed to build find value response message.\n"); + goto fail_msg; + } - return 0; + values = NULL; /* msg owns the values now */ + + len = dht_msg__get_packed_size(msg); + if (len == 0) { + printf("Failed to get packed length of find_value_rsp.\n"); + goto fail_msg; + } + + buf = malloc(len); + if (buf == NULL) { + printf("Failed to malloc find_value_rsp buf.\n"); + goto fail_msg; + } + + if (dht_msg__pack(msg, buf) != len) { + printf("Failed to pack find_value_rsp message.\n"); + goto fail_pack; + } + + upk = dht_msg__unpack(NULL, len, buf); + if (upk == NULL) { + printf("Failed to unpack find_value_rsp message.\n"); + goto fail_unpack; + } + + if (upk->code != DHT_FIND_VALUE_RSP) { + printf("Wrong code in find_value_rsp message (%s != %s).\n", + dht_code_str[upk->code], + dht_code_str[DHT_FIND_VALUE_RSP]); + goto fail_unpack; + } + + if (upk->val == NULL) { + printf("No values in find_value_rsp message.\n"); + goto fail_unpack; + } + + if (upk->val->n_values != 8) { + printf("Not enough values in find_value_rsp (%zu != %lu).\n", + upk->val->n_values, 8UL); + goto fail_unpack; + } + + free(buf); + dht_msg__free_unpacked(msg, NULL); + dht_msg__free_unpacked(upk, NULL); + + free(values); + + dht_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_unpack: + dht_msg__free_unpacked(msg, NULL); + fail_pack: + free(buf); + fail_msg: + fail_fill: + while((i--) > 0) + freebuf(values[i]); + free(values); + fail_values: + dht_fini(); + fail_init: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_kv_store_msg(void) +{ + dht_msg_t * msg; + size_t len; + uint8_t * buf; + struct timespec now; + + TEST_START(); + + clock_gettime(CLOCK_REALTIME_COARSE, &now); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + msg = dht_kv_store_msg(dht.id.data, test_val, now.tv_sec + 10); + if (msg == NULL) { + printf("Failed to get store message.\n"); + goto fail_msg; + } + + if (msg->code != DHT_STORE) { + printf("Wrong code in store message (%s != %s).\n", + dht_code_str[msg->code], + dht_code_str[DHT_STORE]); + goto fail_store_msg; + } + + if (dht_kv_validate_msg(msg) < 0) { + printf("Failed to validate store message.\n"); + goto fail_store_msg; + } + + len = dht_msg__get_packed_size(msg); + if (len == 0) { + printf("Failed to get packed msg length.\n"); + goto fail_msg; + } + + buf = malloc(len); + if (buf == NULL) { + printf("Failed to malloc store msg buf.\n"); + goto fail_msg; + } + + if (dht_msg__pack(msg, buf) != len) { + printf("Failed to pack store message.\n"); + goto fail_pack; + } + + free(buf); + + dht_msg__free_unpacked(msg, NULL); + + dht_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_pack: + free(buf); + fail_store_msg: + dht_msg__free_unpacked(msg, NULL); + fail_msg: + dht_fini(); + fail_init: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_kv_query_contacts_req_rsp(void) +{ + dht_msg_t * req; + dht_msg_t * rsp; + dht_contact_msg_t ** contacts; + size_t len = 2; + + uint8_t * key; + + TEST_START(); + + sink_init(); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + if (fill_dht_with_contacts(1) < 0) { + printf("Failed to fill bucket with contacts.\n"); + goto fail_prep; + } + + key = generate_id(); + if (key == NULL) { + printf("Failed to generate key.\n"); + goto fail_prep; + } + + if (dht_kv_query_contacts(key, NULL) < 0) { + printf("Failed to query contacts.\n"); + goto fail_query; + } + + req = sink_read(); + if (req == NULL) { + printf("Failed to read request from sink.\n"); + goto fail_query; + } + + if (dht_kv_validate_msg(req) < 0) { + printf("Failed to validate find node req.\n"); + goto fail_val_req; + } + + if (random_contact_list(&contacts, len) < 0) { + printf("Failed to create random contact.\n"); + goto fail_val_req; + } + + rsp = dht_kv_find_node_rsp_msg(key, req->find->cookie, &contacts, len); + if (rsp == NULL) { + printf("Failed to create find node response message.\n"); + goto fail_rsp; + } + + memcpy(rsp->src->id.data, dht.id.data, dht.id.len); + rsp->src->addr = generate_cookie(); + + if (dht_kv_validate_msg(rsp) < 0) { + printf("Failed to validate find node response message.\n"); + goto fail_val_rsp; + } + + do_dht_kv_find_node_rsp(rsp->node); + + /* dht_contact_msg__free_unpacked(contacts[0], NULL); set to NULL */ + + free(contacts); + + dht_msg__free_unpacked(rsp, NULL); + + free(key); + + dht_msg__free_unpacked(req, NULL); + + sink_fini(); + + dht_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_val_rsp: + dht_msg__free_unpacked(rsp, NULL); + fail_rsp: + while (len-- > 0) + dht_contact_msg__free_unpacked(contacts[len], NULL); + free(contacts); + fail_val_req: + dht_msg__free_unpacked(req, NULL); + fail_query: + free(key); + fail_prep: + dht_fini(); + fail_init: + sink_fini(); + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_req_create_destroy(void) +{ + struct dht_req * req; + + TEST_START(); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + req = dht_req_create(dht.id.data); + if (req == NULL) { + printf("Failed to create kad request.\n"); + goto fail_req; + } + + dht_req_destroy(req); + + dht_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_req: + dht_fini(); + fail_init: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_reg_unreg(void) +{ + TEST_START(); + + sink_init(); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + if (dht_reg(dht.id.data) < 0) { + printf("Failed to register own id.\n"); + goto fail_reg; + } + + if (!llist_is_empty(&sink.msgs)) { + printf("Packet sent without contacts!"); + goto fail_msg; + } + + if (dht_unreg(dht.id.data) < 0) { + printf("Failed to unregister own id.\n"); + goto fail_msg; + } + + dht_fini(); + + sink_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_msg: + dht_unreg(dht.id.data); + fail_reg: + dht_fini(); + fail_init: + sink_fini(); + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_reg_unreg_contacts(void) +{ + dht_msg_t * msg; + + TEST_START(); + + sink_init(); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + if (fill_dht_with_contacts(4) < 0) { + printf("Failed to fill bucket with contacts.\n"); + goto fail_reg; + } + + if (dht_reg(dht.id.data) < 0) { + printf("Failed to register own id.\n"); + goto fail_reg; + } + + if (sink.msgs.len != dht.alpha) { + printf("Packet sent to too few contacts!\n"); + goto fail_msg; + } + + msg = sink_read(); + if (msg == NULL) { + printf("Failed to read message from sink.\n"); + goto fail_msg; + } + + if (msg->code != DHT_STORE) { + printf("Wrong code in dht reg message (%s != %s).\n", + dht_code_str[msg->code], + dht_code_str[DHT_STORE]); + goto fail_validation; + } + + if (dht_kv_validate_msg(msg) < 0) { + printf("Failed to validate dht message.\n"); + goto fail_validation; + } + + if (dht_unreg(dht.id.data) < 0) { + printf("Failed to unregister own id.\n"); + goto fail_validation; + } + + dht_msg__free_unpacked(msg, NULL); + + dht_fini(); + + sink_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_validation: + dht_msg__free_unpacked(msg, NULL); + fail_msg: + sink_clear(); + dht_unreg(dht.id.data); + fail_reg: + dht_fini(); + fail_init: + sink_fini(); + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_reg_query_local(void) +{ + struct timespec now; + buffer_t test_addr; + + TEST_START(); + + clock_gettime(CLOCK_REALTIME_COARSE, &now); + + if (addr_to_buf(1234321, &test_addr) < 0) { + printf("Failed to convert test address to buffer.\n"); + goto fail_buf; + } + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + if (dht_reg(dht.id.data) < 0) { + printf("Failed to register own id.\n"); + goto fail_reg; + } + + if (dht_query(dht.id.data) == dht.addr) { + printf("Succeeded to query own id.\n"); + goto fail_get; + } + + if (dht_kv_store(dht.id.data, test_addr, now.tv_sec + 5) < 0) { + printf("Failed to publish value.\n"); + goto fail_get; + } + + if (dht_query(dht.id.data) != 1234321) { + printf("Failed to return remote addr.\n"); + goto fail_get; + } + + if (dht_unreg(dht.id.data) < 0) { + printf("Failed to unregister own id.\n"); + goto fail_get; + } + + freebuf(test_addr); + + dht_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_get: + dht_unreg(dht.id.data); + fail_reg: + dht_fini(); + fail_init: + freebuf(test_addr); + fail_buf: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_query(void) +{ + uint8_t * key; + struct dir_dht_config cfg; + + TEST_START(); + + sink_init(); + + cfg = test_dht_config; + cfg.peer = generate_cookie(); + + if (dht_init(&cfg)) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + key = generate_id(); + if (key == NULL) { + printf("Failed to generate key.\n"); + goto fail_key; + } + + if (dht_query(key) != INVALID_ADDR) { + printf("Succeeded to get address without contacts.\n"); + goto fail_get; + } + + if (!llist_is_empty(&sink.msgs)) { + printf("Packet sent without contacts!"); + goto fail_test; + } + + free(key); + + dht_fini(); + + sink_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_test: + sink_clear(); + fail_get: + free(key); + fail_key: + dht_fini(); + fail_init: + sink_fini(); + return TEST_RC_FAIL; +} + +static int test_dht_query_contacts(void) +{ + dht_msg_t * msg; + uint8_t * key; + struct dir_dht_config cfg; + + + TEST_START(); + + sink_init(); + + cfg = test_dht_config; + cfg.peer = generate_cookie(); + + if (dht_init(&cfg)) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + if (fill_dht_with_contacts(10) < 0) { + printf("Failed to fill with contacts!"); + goto fail_contacts; + } + + key = generate_id(); + if (key == NULL) { + printf("Failed to generate key."); + goto fail_contacts; + } + + if (dht_query(key) != INVALID_ADDR) { + printf("Succeeded to get address for random id.\n"); + goto fail_query; + } + + msg = sink_read(); + if (msg == NULL) { + printf("Failed to read message.!\n"); + goto fail_read; + } + + if (dht_kv_validate_msg(msg) < 0) { + printf("Failed to validate dht message.\n"); + goto fail_msg; + } + + if (msg->code != DHT_FIND_VALUE_REQ) { + printf("Failed to validate dht message.\n"); + goto fail_msg; + } + + dht_msg__free_unpacked(msg, NULL); + + free(key); + + sink_clear(); + + dht_fini(); + + sink_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_msg: + dht_msg__free_unpacked(msg, NULL); + fail_read: + sink_clear(); + fail_query: + free(key); + fail_contacts: + dht_fini(); + fail_init: + sink_fini(); + return TEST_RC_FAIL; +} + +int dht_test(int argc, + char ** argv) +{ + int rc = 0; + + (void) argc; + (void) argv; + + rc |= test_dht_init_fini(); + rc |= test_dht_start_stop(); + rc |= test_val_entry_create_destroy(); + rc |= test_dht_entry_create_destroy(); + rc |= test_dht_entry_update_get_val(); + rc |= test_dht_entry_update_get_lval(); + rc |= test_dht_kv_contact_create_destroy(); + rc |= test_dht_kv_contact_list(); + rc |= test_dht_kv_update_bucket(); + rc |= test_dht_kv_get_values(); + rc |= test_dht_kv_find_node_req_msg(); + rc |= test_dht_kv_find_node_rsp_msg(); + rc |= test_dht_kv_find_node_rsp_msg_contacts(); + rc |= test_dht_kv_query_contacts_req_rsp(); + rc |= test_dht_kv_find_value_req_msg(); + rc |= test_dht_kv_find_value_rsp_msg(); + rc |= test_dht_kv_find_value_rsp_msg_contacts(); + rc |= test_dht_kv_find_value_rsp_msg_values(); + rc |= test_dht_kv_store_msg(); + rc |= test_dht_req_create_destroy(); + rc |= test_dht_reg_unreg(); + rc |= test_dht_reg_unreg_contacts(); + rc |= test_dht_reg_query_local(); + rc |= test_dht_query(); + rc |= test_dht_query_contacts(); + + return rc; } diff --git a/src/ipcpd/unicast/dt.c b/src/ipcpd/unicast/dt.c index 2bb5ed2f..e89cb17e 100644 --- a/src/ipcpd/unicast/dt.c +++ b/src/ipcpd/unicast/dt.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Data Transfer Component * @@ -31,6 +31,7 @@ #define DT "dt" #define OUROBOROS_PREFIX DT +#include <ouroboros/atomics.h> #include <ouroboros/bitmap.h> #include <ouroboros/errno.h> #include <ouroboros/logs.h> @@ -41,6 +42,7 @@ #include <ouroboros/fccntl.h> #endif +#include "addr-auth.h" #include "common/comp.h" #include "common/connmgr.h" #include "ca.h" @@ -59,7 +61,7 @@ #include <assert.h> #define QOS_BLOCK_LEN 672 -#define RIB_FILE_STRLEN (189 + QOS_BLOCK_LEN * QOS_CUBE_MAX) +#define RIB_FILE_STRLEN (169 + RIB_TM_STRLEN + QOS_BLOCK_LEN * QOS_CUBE_MAX) #define RIB_NAME_STRLEN 256 #ifndef CLOCK_REALTIME_COARSE @@ -67,7 +69,7 @@ #endif struct comp_info { - void (* post_packet)(void * comp, struct shm_du_buff * sdb); + void (* post_packet)(void * comp, struct ssm_pk_buff * spb); void * comp; char * name; }; @@ -134,16 +136,18 @@ static void dt_pci_des(uint8_t * head, memcpy(&dt_pci->eid, head + dt_pci_info.eid_o, dt_pci_info.eid_size); } -static void dt_pci_shrink(struct shm_du_buff * sdb) +static void dt_pci_shrink(struct ssm_pk_buff * spb) { - assert(sdb); + assert(spb); - shm_du_buff_head_release(sdb, dt_pci_info.head_size); + ssm_pk_buff_pop(spb, dt_pci_info.head_size); } struct { struct psched * psched; + uint64_t addr; + struct pff * pff[QOS_CUBE_MAX]; struct routing_i * routing[QOS_CUBE_MAX]; #ifdef IPCP_FLOW_STATS @@ -165,28 +169,39 @@ struct { size_t f_nhp_pkt[QOS_CUBE_MAX]; size_t f_nhp_bytes[QOS_CUBE_MAX]; pthread_mutex_t lock; - } stat[PROG_MAX_FLOWS]; + } stat[PROC_MAX_FLOWS]; size_t n_flows; #endif struct bmp * res_fds; - struct comp_info comps[PROG_RES_FDS]; + struct comp_info comps[PROC_RES_FDS]; pthread_rwlock_t lock; pthread_t listener; } dt; +/* + * Flow stats are lock-free relaxed atomics on the data path; the per-flow + * lock still guards the stamp/addr/n_flows lifecycle (see stat_used). + */ +#ifdef IPCP_FLOW_STATS +#define dt_stat_inc(idx, name, qc, len) \ + do { \ + FETCH_ADD_RELAXED(&dt.stat[idx].name ## _pkt[qc], 1); \ + FETCH_ADD_RELAXED(&dt.stat[idx].name ## _bytes[qc], (len)); \ + } while (0) +#define dt_stat_load(idx, field, qc) LOAD_RELAXED(&dt.stat[idx].field[qc]) + static int dt_rib_read(const char * path, char * buf, size_t len) { -#ifdef IPCP_FLOW_STATS int fd; int i; char str[QOS_BLOCK_LEN + 1]; char addrstr[20]; char * entry; - char tmstr[20]; + char tmstr[RIB_TM_STRLEN]; size_t rxqlen = 0; size_t txqlen = 0; struct tm * tm; @@ -209,25 +224,25 @@ static int dt_rib_read(const char * path, return 0; } - if (dt.stat[fd].addr == ipcpi.dt_addr) + if (dt.stat[fd].addr == dt.addr) sprintf(addrstr, "%s", dt.comps[fd].name); else - sprintf(addrstr, "%" PRIu64, dt.stat[fd].addr); + sprintf(addrstr, ADDR_FMT32, ADDR_VAL32(&dt.stat[fd].addr)); - tm = localtime(&dt.stat[fd].stamp); - strftime(tmstr, sizeof(tmstr), "%F %T", tm); + tm = gmtime(&dt.stat[fd].stamp); + strftime(tmstr, sizeof(tmstr), RIB_TM_FORMAT, tm); - if (fd >= PROG_RES_FDS) { + if (fd >= PROC_RES_FDS) { fccntl(fd, FLOWGRXQLEN, &rxqlen); fccntl(fd, FLOWGTXQLEN, &txqlen); } sprintf(buf, - "Flow established at: %20s\n" + "Flow established at: %.*s\n" "Endpoint address: %20s\n" "Queued packets (rx): %20zu\n" "Queued packets (tx): %20zu\n\n", - tmstr, addrstr, rxqlen, txqlen); + RIB_TM_STRLEN - 1, tmstr, addrstr, rxqlen, txqlen); for (i = 0; i < QOS_CUBE_MAX; ++i) { sprintf(str, "Qos cube %3d:\n" @@ -246,20 +261,20 @@ static int dt_rib_read(const char * path, " failed nhop (packets): %20zu\n" " failed nhop (bytes): %20zu\n", i, - dt.stat[fd].snd_pkt[i], - dt.stat[fd].snd_bytes[i], - dt.stat[fd].rcv_pkt[i], - dt.stat[fd].rcv_bytes[i], - dt.stat[fd].lcl_w_pkt[i], - dt.stat[fd].lcl_w_bytes[i], - dt.stat[fd].lcl_r_pkt[i], - dt.stat[fd].lcl_r_bytes[i], - dt.stat[fd].r_drp_pkt[i], - dt.stat[fd].r_drp_bytes[i], - dt.stat[fd].w_drp_pkt[i], - dt.stat[fd].w_drp_bytes[i], - dt.stat[fd].f_nhp_pkt[i], - dt.stat[fd].f_nhp_bytes[i] + dt_stat_load(fd, snd_pkt, i), + dt_stat_load(fd, snd_bytes, i), + dt_stat_load(fd, rcv_pkt, i), + dt_stat_load(fd, rcv_bytes, i), + dt_stat_load(fd, lcl_w_pkt, i), + dt_stat_load(fd, lcl_w_bytes, i), + dt_stat_load(fd, lcl_r_pkt, i), + dt_stat_load(fd, lcl_r_bytes, i), + dt_stat_load(fd, r_drp_pkt, i), + dt_stat_load(fd, r_drp_bytes, i), + dt_stat_load(fd, w_drp_pkt, i), + dt_stat_load(fd, w_drp_bytes, i), + dt_stat_load(fd, f_nhp_pkt, i), + dt_stat_load(fd, f_nhp_bytes, i) ); strcat(buf, str); } @@ -267,17 +282,10 @@ static int dt_rib_read(const char * path, pthread_mutex_unlock(&dt.stat[fd].lock); return RIB_FILE_STRLEN; -#else - (void) path; - (void) buf; - (void) len; - return 0; -#endif } static int dt_rib_readdir(char *** buf) { -#ifdef IPCP_FLOW_STATS char entry[RIB_PATH_LEN + 1]; size_t i; int idx = 0; @@ -285,58 +293,50 @@ static int dt_rib_readdir(char *** buf) pthread_rwlock_rdlock(&dt.lock); if (dt.n_flows < 1) { - pthread_rwlock_unlock(&dt.lock); - return 0; + *buf = NULL; + goto no_flows; } *buf = malloc(sizeof(**buf) * dt.n_flows); - if (*buf == NULL) { - pthread_rwlock_unlock(&dt.lock); - return -ENOMEM; - } + if (*buf == NULL) + goto fail_entries; - for (i = 0; i < PROG_MAX_FLOWS; ++i) { + for (i = 0; i < PROC_MAX_FLOWS; ++i) { pthread_mutex_lock(&dt.stat[i].lock); if (dt.stat[i].stamp == 0) { pthread_mutex_unlock(&dt.stat[i].lock); - /* Optimization: skip unused res_fds. */ - if (i < PROG_RES_FDS) - i = PROG_RES_FDS; - continue; + break; } + pthread_mutex_unlock(&dt.stat[i].lock); + sprintf(entry, "%zu", i); (*buf)[idx] = malloc(strlen(entry) + 1); - if ((*buf)[idx] == NULL) { - while (idx-- > 0) - free((*buf)[idx]); - free(*buf); - pthread_mutex_unlock(&dt.stat[i].lock); - pthread_rwlock_unlock(&dt.lock); - return -ENOMEM; - } + if ((*buf)[idx] == NULL) + goto fail_entry; strcpy((*buf)[idx++], entry); - pthread_mutex_unlock(&dt.stat[i].lock); } - assert((size_t) idx == dt.n_flows); - + no_flows: pthread_rwlock_unlock(&dt.lock); return idx; -#else - (void) buf; - return 0; -#endif + + fail_entry: + while (idx-- > 0) + free((*buf)[idx]); + free(*buf); + fail_entries: + pthread_rwlock_unlock(&dt.lock); + return -ENOMEM; } static int dt_rib_getattr(const char * path, struct rib_attr * attr) { -#ifdef IPCP_FLOW_STATS int fd; char * entry; @@ -356,10 +356,7 @@ static int dt_rib_getattr(const char * path, } pthread_mutex_unlock(&dt.stat[fd].lock); -#else - (void) path; - (void) attr; -#endif + return 0; } @@ -369,7 +366,12 @@ static struct rib_ops r_ops = { .getattr = dt_rib_getattr }; -#ifdef IPCP_FLOW_STATS +/* + * Hold dt.lock + per-stat together: dt_rib_readdir samples n_flows + * under rdlock and walks stamps under per-stat; updates must be + * atomic w.r.t. that snapshot or the malloc(n_flows) buffer can + * overflow. + */ static void stat_used(int fd, uint64_t addr) { @@ -377,6 +379,7 @@ static void stat_used(int fd, clock_gettime(CLOCK_REALTIME_COARSE, &now); + pthread_rwlock_wrlock(&dt.lock); pthread_mutex_lock(&dt.stat[fd].lock); memset(&dt.stat[fd], 0, sizeof(dt.stat[fd])); @@ -384,14 +387,13 @@ static void stat_used(int fd, dt.stat[fd].stamp = (addr != INVALID_ADDR) ? now.tv_sec : 0; dt.stat[fd].addr = addr; - pthread_mutex_unlock(&dt.stat[fd].lock); - - pthread_rwlock_wrlock(&dt.lock); - (addr != INVALID_ADDR) ? ++dt.n_flows : --dt.n_flows; + pthread_mutex_unlock(&dt.stat[fd].lock); pthread_rwlock_unlock(&dt.lock); } +#else +#define dt_stat_inc(idx, name, qc, len) ((void) 0) #endif static void handle_event(void * self, @@ -429,7 +431,7 @@ static void handle_event(void * self, static void packet_handler(int fd, qoscube_t qc, - struct shm_du_buff * sdb) + struct ssm_pk_buff * spb) { struct dt_pci dt_pci; int ret; @@ -437,35 +439,23 @@ static void packet_handler(int fd, uint8_t * head; size_t len; - len = shm_du_buff_len(sdb); + len = ssm_pk_buff_len(spb); #ifndef IPCP_FLOW_STATS - (void) fd; -#else - pthread_mutex_lock(&dt.stat[fd].lock); - - ++dt.stat[fd].rcv_pkt[qc]; - dt.stat[fd].rcv_bytes[qc] += len; - - pthread_mutex_unlock(&dt.stat[fd].lock); + (void) fd; #endif + dt_stat_inc(fd, rcv, qc, len); + memset(&dt_pci, 0, sizeof(dt_pci)); - head = shm_du_buff_head(sdb); + head = ssm_pk_buff_head(spb); dt_pci_des(head, &dt_pci); - if (dt_pci.dst_addr != ipcpi.dt_addr) { + if (dt_pci.dst_addr != dt.addr) { if (dt_pci.ttl == 0) { log_dbg("TTL was zero."); - ipcp_sdb_release(sdb); -#ifdef IPCP_FLOW_STATS - pthread_mutex_lock(&dt.stat[fd].lock); - - ++dt.stat[fd].r_drp_pkt[qc]; - dt.stat[fd].r_drp_bytes[qc] += len; - - pthread_mutex_unlock(&dt.stat[fd].lock); -#endif + ipcp_spb_release(spb); + dt_stat_inc(fd, r_drp, qc, len); return; } @@ -474,74 +464,43 @@ static void packet_handler(int fd, if (ofd < 0) { log_dbg("No next hop for %" PRIu64 ".", dt_pci.dst_addr); - ipcp_sdb_release(sdb); -#ifdef IPCP_FLOW_STATS - pthread_mutex_lock(&dt.stat[fd].lock); - - ++dt.stat[fd].f_nhp_pkt[qc]; - dt.stat[fd].f_nhp_bytes[qc] += len; - - pthread_mutex_unlock(&dt.stat[fd].lock); -#endif + ipcp_spb_release(spb); + dt_stat_inc(fd, f_nhp, qc, len); return; } (void) ca_calc_ecn(ofd, head + dt_pci_info.ecn_o, qc, len); - ret = ipcp_flow_write(ofd, sdb); + ret = ipcp_flow_write(ofd, spb); if (ret < 0) { log_dbg("Failed to write packet to fd %d.", ofd); if (ret == -EFLOWDOWN) notifier_event(NOTIFY_DT_FLOW_DOWN, &ofd); - ipcp_sdb_release(sdb); -#ifdef IPCP_FLOW_STATS - pthread_mutex_lock(&dt.stat[ofd].lock); - - ++dt.stat[ofd].w_drp_pkt[qc]; - dt.stat[ofd].w_drp_bytes[qc] += len; - - pthread_mutex_unlock(&dt.stat[ofd].lock); -#endif + ipcp_spb_release(spb); + dt_stat_inc(ofd, w_drp, qc, len); return; } -#ifdef IPCP_FLOW_STATS - pthread_mutex_lock(&dt.stat[ofd].lock); - ++dt.stat[ofd].snd_pkt[qc]; - dt.stat[ofd].snd_bytes[qc] += len; - - pthread_mutex_unlock(&dt.stat[ofd].lock); -#endif + dt_stat_inc(ofd, snd, qc, len); } else { - dt_pci_shrink(sdb); - if (dt_pci.eid >= PROG_RES_FDS) { + dt_pci_shrink(spb); + if (dt_pci.eid >= PROC_RES_FDS) { uint8_t ecn = *(head + dt_pci_info.ecn_o); - fa_np1_rcv(dt_pci.eid, ecn, sdb); + fa_np1_rcv(dt_pci.eid, ecn, spb); return; } if (dt.comps[dt_pci.eid].post_packet == NULL) { log_err("No registered component on eid %" PRIu64 ".", dt_pci.eid); - ipcp_sdb_release(sdb); + ipcp_spb_release(spb); return; } -#ifdef IPCP_FLOW_STATS - pthread_mutex_lock(&dt.stat[fd].lock); - - ++dt.stat[fd].lcl_r_pkt[qc]; - dt.stat[fd].lcl_r_bytes[qc] += len; + dt_stat_inc(fd, lcl_r, qc, len); + dt_stat_inc(dt_pci.eid, snd, qc, len); - pthread_mutex_unlock(&dt.stat[fd].lock); - pthread_mutex_lock(&dt.stat[dt_pci.eid].lock); - - ++dt.stat[dt_pci.eid].snd_pkt[qc]; - dt.stat[dt_pci.eid].snd_bytes[qc] += len; - - pthread_mutex_unlock(&dt.stat[dt_pci.eid].lock); -#endif dt.comps[dt_pci.eid].post_packet(dt.comps[dt_pci.eid].comp, - sdb); + spb); } } @@ -569,17 +528,25 @@ int dt_init(struct dt_config cfg) { int i; int j; +#ifdef IPCP_FLOW_STATS char dtstr[RIB_NAME_STRLEN + 1]; - int pp; +#endif + enum pol_pff pp; struct conn_info info; memset(&info, 0, sizeof(info)); + dt.addr = addr_auth_address(); + if (dt.addr == INVALID_ADDR) { + log_err("Failed to get address"); + return -1; + } + strcpy(info.comp_name, DT_COMP); strcpy(info.protocol, DT_PROTO); info.pref_version = 1; info.pref_syntax = PROTO_FIXED; - info.addr = ipcpi.dt_addr; + info.addr = dt.addr; if (cfg.eid_size != 8) { /* only support 64 bits from now */ log_warn("Invalid EID size. Only 64 bit is supported."); @@ -601,8 +568,7 @@ int dt_init(struct dt_config cfg) goto fail_connmgr_comp_init; } - pp = routing_init(cfg.routing_type); - if (pp < 0) { + if (routing_init(&cfg.routing, &pp) < 0) { log_err("Failed to init routing."); goto fail_routing; } @@ -631,13 +597,13 @@ int dt_init(struct dt_config cfg) goto fail_rwlock_init; } - dt.res_fds = bmp_create(PROG_RES_FDS, 0); + dt.res_fds = bmp_create(PROC_RES_FDS, 0); if (dt.res_fds == NULL) goto fail_res_fds; #ifdef IPCP_FLOW_STATS memset(dt.stat, 0, sizeof(dt.stat)); - for (i = 0; i < PROG_MAX_FLOWS; ++i) + for (i = 0; i < PROC_MAX_FLOWS; ++i) if (pthread_mutex_init(&dt.stat[i].lock, NULL)) { log_err("Failed to init mutex for flow %d.", i); for (j = 0; j < i; ++j) @@ -646,18 +612,19 @@ int dt_init(struct dt_config cfg) } dt.n_flows = 0; -#endif - sprintf(dtstr, "%s.%" PRIu64, DT, ipcpi.dt_addr); + + sprintf(dtstr, "%s." ADDR_FMT32, DT, ADDR_VAL32(&dt.addr)); if (rib_reg(dtstr, &r_ops)) { log_err("Failed to register RIB."); goto fail_rib_reg; } +#endif return 0; - fail_rib_reg: #ifdef IPCP_FLOW_STATS - for (i = 0; i < PROG_MAX_FLOWS; ++i) + fail_rib_reg: + for (i = 0; i < PROC_MAX_FLOWS; ++i) pthread_mutex_destroy(&dt.stat[i].lock); fail_stat_lock: #endif @@ -680,13 +647,15 @@ int dt_init(struct dt_config cfg) void dt_fini(void) { +#ifdef IPCP_FLOW_STATS char dtstr[RIB_NAME_STRLEN + 1]; +#endif int i; - sprintf(dtstr, "%s.%" PRIu64, DT, ipcpi.dt_addr); - rib_unreg(dtstr); #ifdef IPCP_FLOW_STATS - for (i = 0; i < PROG_MAX_FLOWS; ++i) + sprintf(dtstr, "%s.%" PRIu64, DT, dt.addr); + rib_unreg(dtstr); + for (i = 0; i < PROC_MAX_FLOWS; ++i) pthread_mutex_destroy(&dt.stat[i].lock); #endif bmp_destroy(dt.res_fds); @@ -719,21 +688,31 @@ int dt_start(void) if (pthread_create(&dt.listener, NULL, dt_conn_handle, NULL)) { log_err("Failed to create listener thread."); - psched_destroy(dt.psched); - return -1; + goto fail_listener; + } + + if (routing_start() < 0) { + log_err("Failed to start routing."); + goto fail_routing; } return 0; + fail_routing: + pthread_cancel(dt.listener); + pthread_join(dt.listener, NULL); + fail_listener: + notifier_unreg(&handle_event); fail_notifier_reg: psched_destroy(dt.psched); fail_psched: return -1; - } void dt_stop(void) { + routing_stop(); + pthread_cancel(dt.listener); pthread_join(dt.listener, NULL); @@ -743,12 +722,12 @@ void dt_stop(void) } int dt_reg_comp(void * comp, - void (* func)(void * func, struct shm_du_buff *), + void (* func)(void * func, struct ssm_pk_buff *), char * name) { int eid; - assert(func); + assert(func != NULL); pthread_rwlock_wrlock(&dt.lock); @@ -769,15 +748,32 @@ int dt_reg_comp(void * comp, pthread_rwlock_unlock(&dt.lock); #ifdef IPCP_FLOW_STATS - stat_used(eid, ipcpi.dt_addr); + stat_used(eid, dt.addr); #endif return eid; } +void dt_unreg_comp(int eid) +{ + assert(eid >= 0 && eid < PROC_RES_FDS); + + pthread_rwlock_wrlock(&dt.lock); + + assert(dt.comps[eid].post_packet != NULL); + + dt.comps[eid].post_packet = NULL; + dt.comps[eid].comp = NULL; + dt.comps[eid].name = NULL; + + pthread_rwlock_unlock(&dt.lock); + + return; +} + int dt_write_packet(uint64_t dst_addr, qoscube_t qc, uint64_t eid, - struct shm_du_buff * sdb) + struct ssm_pk_buff * spb) { struct dt_pci dt_pci; int fd; @@ -785,44 +781,33 @@ int dt_write_packet(uint64_t dst_addr, uint8_t * head; size_t len; - assert(sdb); - assert(dst_addr != ipcpi.dt_addr); - - len = shm_du_buff_len(sdb); + assert(spb); + assert(dst_addr != dt.addr); #ifdef IPCP_FLOW_STATS - if (eid < PROG_RES_FDS) { - pthread_mutex_lock(&dt.stat[eid].lock); - - ++dt.stat[eid].lcl_r_pkt[qc]; - dt.stat[eid].lcl_r_bytes[qc] += len; + len = ssm_pk_buff_len(spb); - pthread_mutex_unlock(&dt.stat[eid].lock); - } + if (eid < PROC_RES_FDS) + dt_stat_inc(eid, lcl_r, qc, len); #endif fd = pff_nhop(dt.pff[qc], dst_addr); if (fd < 0) { - log_dbg("Could not get nhop for addr %" PRIu64 ".", dst_addr); + log_dbg("Could not get nhop for " ADDR_FMT32 ".", + ADDR_VAL32(&dst_addr)); #ifdef IPCP_FLOW_STATS - if (eid < PROG_RES_FDS) { - pthread_mutex_lock(&dt.stat[eid].lock); - - ++dt.stat[eid].lcl_r_pkt[qc]; - dt.stat[eid].lcl_r_bytes[qc] += len; - - pthread_mutex_unlock(&dt.stat[eid].lock); - } + if (eid < PROC_RES_FDS) + dt_stat_inc(eid, lcl_r, qc, len); #endif return -EPERM; } - head = shm_du_buff_head_alloc(sdb, dt_pci_info.head_size); + head = ssm_pk_buff_push(spb, dt_pci_info.head_size); if (head == NULL) { log_dbg("Failed to allocate DT header."); goto fail_write; } - len = shm_du_buff_len(sdb); + len = ssm_pk_buff_len(spb); dt_pci.dst_addr = dst_addr; dt_pci.qc = qc; @@ -833,7 +818,7 @@ int dt_write_packet(uint64_t dst_addr, dt_pci_ser(head, &dt_pci); - ret = ipcp_flow_write(fd, sdb); + ret = ipcp_flow_write(fd, spb); if (ret < 0) { log_dbg("Failed to write packet to fd %d.", fd); if (ret == -EFLOWDOWN) @@ -841,31 +826,17 @@ int dt_write_packet(uint64_t dst_addr, goto fail_write; } #ifdef IPCP_FLOW_STATS - pthread_mutex_lock(&dt.stat[fd].lock); - - if (dt_pci.eid < PROG_RES_FDS) { - ++dt.stat[fd].lcl_w_pkt[qc]; - dt.stat[fd].lcl_w_bytes[qc] += len; - } - ++dt.stat[fd].snd_pkt[qc]; - dt.stat[fd].snd_bytes[qc] += len; - - pthread_mutex_unlock(&dt.stat[fd].lock); + if (dt_pci.eid < PROC_RES_FDS) + dt_stat_inc(fd, lcl_w, qc, len); + dt_stat_inc(fd, snd, qc, len); #endif return 0; fail_write: #ifdef IPCP_FLOW_STATS - pthread_mutex_lock(&dt.stat[fd].lock); - - if (eid < PROG_RES_FDS) { - ++dt.stat[fd].lcl_w_pkt[qc]; - dt.stat[fd].lcl_w_bytes[qc] += len; - } - ++dt.stat[fd].w_drp_pkt[qc]; - dt.stat[fd].w_drp_bytes[qc] += len; - - pthread_mutex_unlock(&dt.stat[fd].lock); + if (eid < PROC_RES_FDS) + dt_stat_inc(fd, lcl_w, qc, len); + dt_stat_inc(fd, w_drp, qc, len); #endif return -1; } diff --git a/src/ipcpd/unicast/dt.h b/src/ipcpd/unicast/dt.h index 7198a013..a484377d 100644 --- a/src/ipcpd/unicast/dt.h +++ b/src/ipcpd/unicast/dt.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Data Transfer component * @@ -25,7 +25,7 @@ #include <ouroboros/ipcp.h> #include <ouroboros/qoscube.h> -#include <ouroboros/shm_rdrbuff.h> +#include <ouroboros/ssm_pool.h> #define DT_COMP "Data Transfer" #define DT_PROTO "dtp" @@ -39,13 +39,15 @@ int dt_start(void); void dt_stop(void); -int dt_reg_comp(void * comp, - void (* func)(void * comp, struct shm_du_buff * sdb), - char * name); +int dt_reg_comp(void * comp, + void (* func)(void * comp, struct ssm_pk_buff * spb), + char * name); + +void dt_unreg_comp(int eid); int dt_write_packet(uint64_t dst_addr, qoscube_t qc, uint64_t eid, - struct shm_du_buff * sdb); + struct ssm_pk_buff * spb); #endif /* OUROBOROS_IPCPD_UNICAST_DT_H */ diff --git a/src/ipcpd/unicast/fa.c b/src/ipcpd/unicast/fa.c index 3631fd7b..c6eca175 100644 --- a/src/ipcpd/unicast/fa.c +++ b/src/ipcpd/unicast/fa.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Flow allocator of the IPC Process * @@ -37,16 +37,19 @@ #include <ouroboros/errno.h> #include <ouroboros/dev.h> #include <ouroboros/ipcp-dev.h> +#include <ouroboros/np1_flow.h> #include <ouroboros/rib.h> #include <ouroboros/random.h> #include <ouroboros/pthread.h> +#include "addr-auth.h" #include "dir.h" #include "fa.h" #include "psched.h" #include "ipcp.h" #include "dt.h" #include "ca.h" +#include "np1.h" #include <inttypes.h> #include <stdlib.h> @@ -56,12 +59,13 @@ #define CLOCK_REALTIME_COARSE CLOCK_REALTIME #endif -#define TIMEOUT 10 * MILLION /* nanoseconds */ +#define TIMEOUT 10 * MILLION /* nanoseconds */ +#define MSGBUFSZ 32768 -#define FLOW_REQ 0 -#define FLOW_REPLY 1 -#define FLOW_UPDATE 2 -#define MSGBUFSZ 2048 +#define FLOW_REQ 0 +#define FLOW_REPLY 1 +#define FLOW_UPDATE 2 +#define FLOW_IRM_UPDATE 3 #define STAT_FILE_LEN 0 @@ -69,24 +73,22 @@ struct fa_msg { uint64_t s_addr; uint64_t r_eid; uint64_t s_eid; - uint8_t code; - int8_t response; - uint16_t ece; - /* QoS parameters from spec, aligned */ - uint32_t delay; uint64_t bandwidth; + int32_t response; + uint32_t delay; uint32_t loss; uint32_t ber; uint32_t max_gap; uint32_t timeout; - uint16_t cypher_s; + uint16_t ece; + uint8_t code; uint8_t availability; - uint8_t in_order; + uint8_t service; } __attribute__((packed)); struct cmd { struct list_head next; - struct shm_du_buff * sdb; + struct ssm_pk_buff * spb; }; struct fa_flow { @@ -111,7 +113,7 @@ struct fa_flow { struct { pthread_rwlock_t flows_lock; - struct fa_flow flows[PROG_MAX_FLOWS]; + struct fa_flow flows[PROC_MAX_FLOWS]; #ifdef IPCP_FLOW_STATS size_t n_flows; #endif @@ -125,17 +127,17 @@ struct { struct psched * psched; } fa; +#ifdef IPCP_FLOW_STATS static int fa_rib_read(const char * path, char * buf, size_t len) { -#ifdef IPCP_FLOW_STATS struct fa_flow * flow; int fd; char r_addrstr[21]; char s_eidstr[21]; char r_eidstr[21]; - char tmstr[20]; + char tmstr[RIB_TM_STRLEN]; char castr[1024]; char * entry; struct tm * tm; @@ -145,7 +147,7 @@ static int fa_rib_read(const char * path, fd = atoi(entry); - if (fd < 0 || fd >= PROG_MAX_FLOWS) + if (fd < 0 || fd >= PROC_MAX_FLOWS) return -1; if (len < 1536) @@ -166,8 +168,8 @@ static int fa_rib_read(const char * path, sprintf(s_eidstr, "%" PRIu64, flow->s_eid); sprintf(r_eidstr, "%" PRIu64, flow->r_eid); - tm = localtime(&flow->stamp); - strftime(tmstr, sizeof(tmstr), "%F %T", tm); + tm = gmtime(&flow->stamp); + strftime(tmstr, sizeof(tmstr), RIB_TM_FORMAT, tm); ca_print_stats(flow->ctx, castr, 1024); @@ -199,17 +201,10 @@ static int fa_rib_read(const char * path, pthread_rwlock_unlock(&fa.flows_lock); return strlen(buf); -#else - (void) path; - (void) buf; - (void) len; - return 0; -#endif } static int fa_rib_readdir(char *** buf) { -#ifdef IPCP_FLOW_STATS char entry[RIB_PATH_LEN + 1]; size_t i; int idx = 0; @@ -217,17 +212,15 @@ static int fa_rib_readdir(char *** buf) pthread_rwlock_rdlock(&fa.flows_lock); if (fa.n_flows < 1) { - pthread_rwlock_unlock(&fa.flows_lock); - return 0; + *buf = NULL; + goto no_flows; } *buf = malloc(sizeof(**buf) * fa.n_flows); - if (*buf == NULL) { - pthread_rwlock_unlock(&fa.flows_lock); - return -ENOMEM; - } + if (*buf == NULL) + goto fail_entries; - for (i = 0; i < PROG_MAX_FLOWS; ++i) { + for (i = 0; i < PROC_MAX_FLOWS; ++i) { struct fa_flow * flow; flow = &fa.flows[i]; @@ -237,32 +230,30 @@ static int fa_rib_readdir(char *** buf) sprintf(entry, "%zu", i); (*buf)[idx] = malloc(strlen(entry) + 1); - if ((*buf)[idx] == NULL) { - while (idx-- > 0) - free((*buf)[idx]); - free(*buf); - pthread_rwlock_unlock(&fa.flows_lock); - return -ENOMEM; - } + if ((*buf)[idx] == NULL) + goto fail_entry; strcpy((*buf)[idx++], entry); } assert((size_t) idx == fa.n_flows); - + no_flows: pthread_rwlock_unlock(&fa.flows_lock); return idx; -#else - (void) buf; - return 0; -#endif + + fail_entry: + while (idx-- > 0) + free((*buf)[idx]); + free(*buf); + fail_entries: + pthread_rwlock_unlock(&fa.flows_lock); + return -ENOMEM; } static int fa_rib_getattr(const char * path, struct rib_attr * attr) { -#ifdef IPCP_FLOW_STATS int fd; char * entry; struct fa_flow * flow; @@ -285,10 +276,7 @@ static int fa_rib_getattr(const char * path, } pthread_rwlock_unlock(&fa.flows_lock); -#else - (void) path; - (void) attr; -#endif + return 0; } @@ -297,6 +285,7 @@ static struct rib_ops r_ops = { .readdir = fa_rib_readdir, .getattr = fa_rib_getattr }; +#endif /* IPCP_FLOW_STATS */ static int eid_to_fd(uint64_t eid) { @@ -305,7 +294,7 @@ static int eid_to_fd(uint64_t eid) fd = eid & 0xFFFFFFFF; - if (fd < 0 || fd >= PROG_MAX_FLOWS) + if (fd < 0 || fd >= PROC_MAX_FLOWS) return -1; flow = &fa.flows[fd]; @@ -330,7 +319,7 @@ static uint64_t gen_eid(int fd) static void packet_handler(int fd, qoscube_t qc, - struct shm_du_buff * sdb) + struct ssm_pk_buff * spb) { struct fa_flow * flow; uint64_t r_addr; @@ -342,7 +331,7 @@ static void packet_handler(int fd, pthread_rwlock_wrlock(&fa.flows_lock); - len = shm_du_buff_len(sdb); + len = ssm_pk_buff_len(spb); #ifdef IPCP_FLOW_STATS ++flow->p_snd; @@ -357,8 +346,8 @@ static void packet_handler(int fd, ca_wnd_wait(wnd); - if (dt_write_packet(r_addr, qc, r_eid, sdb)) { - ipcp_sdb_release(sdb); + if (dt_write_packet(r_addr, qc, r_eid, spb)) { + ipcp_spb_release(spb); log_dbg("Failed to forward packet."); #ifdef IPCP_FLOW_STATS pthread_rwlock_wrlock(&fa.flows_lock); @@ -411,7 +400,7 @@ static void fa_flow_fini(struct fa_flow * flow) } static void fa_post_packet(void * comp, - struct shm_du_buff * sdb) + struct ssm_pk_buff * spb) { struct cmd * cmd; @@ -422,11 +411,11 @@ static void fa_post_packet(void * comp, cmd = malloc(sizeof(*cmd)); if (cmd == NULL) { log_err("Command failed. Out of memory."); - ipcp_sdb_release(sdb); + ipcp_spb_release(spb); return; } - cmd->sdb = sdb; + cmd->spb = spb; pthread_mutex_lock(&fa.mtx); @@ -454,16 +443,16 @@ static size_t fa_wait_for_fa_msg(struct fa_msg * msg) pthread_cleanup_pop(true); - len = shm_du_buff_len(cmd->sdb); + len = ssm_pk_buff_len(cmd->spb); if (len > MSGBUFSZ || len < sizeof(*msg)) { log_warn("Invalid flow allocation message (len: %zd).", len); free(cmd); return 0; /* No valid message */ } - memcpy(msg, shm_du_buff_head(cmd->sdb), len); + memcpy(msg, ssm_pk_buff_head(cmd->spb), len); - ipcp_sdb_release(cmd->sdb); + ipcp_spb_release(cmd->spb); free(cmd); @@ -495,12 +484,12 @@ static int fa_handle_flow_req(struct fa_msg * msg, qs.availability = msg->availability; qs.loss = ntoh32(msg->loss); qs.ber = ntoh32(msg->ber); - qs.in_order = msg->in_order; + qs.service = msg->service; qs.max_gap = ntoh32(msg->max_gap); - qs.cypher_s = ntoh16(msg->cypher_s); qs.timeout = ntoh32(msg->timeout); - fd = ipcp_wait_flow_req_arr(dst, qs, IPCP_UNICAST_MPL, &data); + fd = ipcp_wait_flow_req_arr(dst, qs, IPCP_UNICAST_MPL, + IPCP_UNICAST_MTU, &data); if (fd < 0) return fd; @@ -526,8 +515,10 @@ static int fa_handle_flow_reply(struct fa_msg * msg, struct fa_flow * flow; buffer_t data; /* Piggbacked data on flow alloc request. */ time_t mpl = IPCP_UNICAST_MPL; + int response; - assert(len >= sizeof(*msg)); + if (len < sizeof(*msg)) + return -EINVAL; data.data = (uint8_t *) msg + sizeof(*msg); data.len = len - sizeof(*msg); @@ -545,15 +536,20 @@ static int fa_handle_flow_reply(struct fa_msg * msg, flow = &fa.flows[fd]; flow->r_eid = ntoh64(msg->s_eid); + response = ntoh32(msg->response); + + log_dbg("IPCP received msg response %d for flow on fd %d.", + response, fd); - if (msg->response < 0) + if (response < 0) fa_flow_fini(flow); else psched_add(fa.psched, fd); pthread_rwlock_unlock(&fa.flows_lock); - if (ipcp_flow_alloc_reply(fd, msg->response, mpl, &data) < 0) { + if (ipcp_flow_alloc_reply(fd, response, mpl, + IPCP_UNICAST_MTU, &data) < 0) { log_err("Failed to reply for flow allocation on fd %d.", fd); return -EIRMD; } @@ -567,8 +563,8 @@ static int fa_handle_flow_update(struct fa_msg * msg, struct fa_flow * flow; int fd; - (void) len; - assert(len >= sizeof(*msg)); + if (len < sizeof(*msg)) + return -EINVAL; pthread_rwlock_wrlock(&fa.flows_lock); @@ -591,6 +587,43 @@ static int fa_handle_flow_update(struct fa_msg * msg, return 0; } +static int fa_handle_flow_irm_update(struct fa_msg * msg, + size_t len) +{ + buffer_t data; + int fd; + int flow_id; + + if (len < sizeof(*msg)) + return -EINVAL; + + data.data = (uint8_t *) msg + sizeof(*msg); + data.len = len - sizeof(*msg); + + pthread_rwlock_rdlock(&fa.flows_lock); + + fd = eid_to_fd(ntoh64(msg->r_eid)); + + pthread_rwlock_unlock(&fa.flows_lock); + + if (fd < 0) { + log_err("Flow update for unknown EID %" PRIu64 ".", + ntoh64(msg->r_eid)); + return -ENOTALLOC; + } + + flow_id = np1_flow_id(fd); + if (flow_id < 0) + return -ENOTALLOC; + + if (ipcp_flow_update_arr(flow_id, &data) < 0) { + log_err("Failed to relay flow update on fd %d.", fd); + return -EIRMD; + } + + return 0; +} + static void * fa_handle_packet(void * o) { (void) o; @@ -619,6 +652,10 @@ static void * fa_handle_packet(void * o) if (fa_handle_flow_update(msg, len) < 0) log_err("Error handling flow update."); break; + case FLOW_IRM_UPDATE: + if (fa_handle_flow_irm_update(msg, len) < 0) + log_err("Error handling flow update."); + break; default: log_warn("Recieved unknown flow allocation message."); break; @@ -647,20 +684,26 @@ int fa_init(void) if (pthread_cond_init(&fa.cond, &cattr)) goto fail_cond; - pthread_condattr_destroy(&cattr); - - list_head_init(&fa.cmds); - +#ifdef IPCP_FLOW_STATS if (rib_reg(FA, &r_ops)) goto fail_rib_reg; +#endif fa.eid = dt_reg_comp(&fa, &fa_post_packet, FA); if ((int) fa.eid < 0) - goto fail_rib_reg; + goto fail_dt_reg; + + list_head_init(&fa.cmds); + + pthread_condattr_destroy(&cattr); return 0; + fail_dt_reg: +#ifdef IPCP_FLOW_STATS + rib_unreg(FA); fail_rib_reg: +#endif pthread_cond_destroy(&fa.cond); fail_cond: pthread_condattr_destroy(&cattr); @@ -669,26 +712,34 @@ int fa_init(void) fail_mtx: pthread_rwlock_destroy(&fa.flows_lock); fail_rwlock: - return -1; } void fa_fini(void) { +#ifdef IPCP_FLOW_STATS rib_unreg(FA); - +#endif pthread_cond_destroy(&fa.cond);; pthread_mutex_destroy(&fa.mtx); pthread_rwlock_destroy(&fa.flows_lock); } +static int np1_flow_read_fa(int fd, + struct ssm_pk_buff ** spb) +{ + return np1_flow_read(fd, spb, NP1_GET_POOL(fd)); +} + int fa_start(void) { +#ifndef BUILD_CONTAINER struct sched_param par; int pol; int max; +#endif - fa.psched = psched_create(packet_handler, np1_flow_read); + fa.psched = psched_create(packet_handler, np1_flow_read_fa); if (fa.psched == NULL) { log_err("Failed to start packet scheduler."); goto fail_psched; @@ -699,6 +750,7 @@ int fa_start(void) goto fail_thread; } +#ifndef BUILD_CONTAINER if (pthread_getschedparam(fa.worker, &pol, &par)) { log_err("Failed to get worker thread scheduling parameters."); goto fail_sched; @@ -716,12 +768,15 @@ int fa_start(void) log_err("Failed to set scheduler priority to maximum."); goto fail_sched; } +#endif return 0; +#ifndef BUILD_CONTAINER fail_sched: pthread_cancel(fa.worker); pthread_join(fa.worker, NULL); +#endif fail_thread: psched_destroy(fa.psched); fail_psched: @@ -742,7 +797,7 @@ int fa_alloc(int fd, const buffer_t * data) { struct fa_msg * msg; - struct shm_du_buff * sdb; + struct ssm_pk_buff * spb; struct fa_flow * flow; uint64_t addr; qoscube_t qc = QOS_CUBE_BE; @@ -755,34 +810,33 @@ int fa_alloc(int fd, len = sizeof(*msg) + ipcp_dir_hash_len(); - if (ipcp_sdb_reserve(&sdb, len + data->len)) + if (ipcp_spb_reserve(&spb, len + data->len)) return -1; - msg = (struct fa_msg *) shm_du_buff_head(sdb); + msg = (struct fa_msg *) ssm_pk_buff_head(spb); memset(msg, 0, sizeof(*msg)); eid = gen_eid(fd); msg->code = FLOW_REQ; msg->s_eid = hton64(eid); - msg->s_addr = hton64(ipcpi.dt_addr); + msg->s_addr = hton64(addr_auth_address()); msg->delay = hton32(qs.delay); msg->bandwidth = hton64(qs.bandwidth); msg->availability = qs.availability; msg->loss = hton32(qs.loss); msg->ber = hton32(qs.ber); - msg->in_order = qs.in_order; + msg->service = qs.service; msg->max_gap = hton32(qs.max_gap); - msg->cypher_s = hton16(qs.cypher_s); msg->timeout = hton32(qs.timeout); memcpy(msg + 1, dst, ipcp_dir_hash_len()); if (data->len > 0) - memcpy(shm_du_buff_head(sdb) + len, data->data, data->len); + memcpy(ssm_pk_buff_head(spb) + len, data->data, data->len); - if (dt_write_packet(addr, qc, fa.eid, sdb)) { + if (dt_write_packet(addr, qc, fa.eid, spb)) { log_err("Failed to send flow allocation request packet."); - ipcp_sdb_release(sdb); + ipcp_spb_release(spb); return -1; } @@ -804,7 +858,7 @@ int fa_alloc_resp(int fd, const buffer_t * data) { struct fa_msg * msg; - struct shm_du_buff * sdb; + struct ssm_pk_buff * spb; struct fa_flow * flow; qoscube_t qc = QOS_CUBE_BE; @@ -815,17 +869,17 @@ int fa_alloc_resp(int fd, goto fail_alloc_resp; } - if (ipcp_sdb_reserve(&sdb, sizeof(*msg) + data->len)) { - log_err("Failed to reserve sdb (%zu bytes).", + if (ipcp_spb_reserve(&spb, sizeof(*msg) + data->len)) { + log_err("Failed to reserve spb (%zu bytes).", sizeof(*msg) + data->len); goto fail_reserve; } - msg = (struct fa_msg *) shm_du_buff_head(sdb); + msg = (struct fa_msg *) ssm_pk_buff_head(spb); memset(msg, 0, sizeof(*msg)); msg->code = FLOW_REPLY; - msg->response = response; + msg->response = hton32(response); if (data->len > 0) memcpy(msg + 1, data->data, data->len); @@ -836,13 +890,13 @@ int fa_alloc_resp(int fd, pthread_rwlock_unlock(&fa.flows_lock); - if (dt_write_packet(flow->r_addr, qc, fa.eid, sdb)) { + if (dt_write_packet(flow->r_addr, qc, fa.eid, spb)) { log_err("Failed to send flow allocation response packet."); goto fail_packet; } if (response < 0) { - pthread_rwlock_rdlock(&fa.flows_lock); + pthread_rwlock_wrlock(&fa.flows_lock); fa_flow_fini(flow); pthread_rwlock_unlock(&fa.flows_lock); } else { @@ -852,7 +906,7 @@ int fa_alloc_resp(int fd, return 0; fail_packet: - ipcp_sdb_release(sdb); + ipcp_spb_release(spb); fail_reserve: pthread_rwlock_wrlock(&fa.flows_lock); fa_flow_fini(flow); @@ -861,6 +915,44 @@ int fa_alloc_resp(int fd, return -1; } +int fa_irm_update(int fd, + const buffer_t * data) +{ + struct fa_msg * msg; + struct ssm_pk_buff * spb; + struct fa_flow * flow; + qoscube_t qc = QOS_CUBE_BE; + uint64_t r_addr; + + flow = &fa.flows[fd]; + + if (ipcp_spb_reserve(&spb, sizeof(*msg) + data->len)) + return -1; + + msg = (struct fa_msg *) ssm_pk_buff_head(spb); + memset(msg, 0, sizeof(*msg)); + + msg->code = FLOW_IRM_UPDATE; + if (data->len > 0) + memcpy(msg + 1, data->data, data->len); + + pthread_rwlock_rdlock(&fa.flows_lock); + + msg->r_eid = hton64(flow->r_eid); + msg->s_eid = hton64(flow->s_eid); + r_addr = flow->r_addr; + + pthread_rwlock_unlock(&fa.flows_lock); + + if (dt_write_packet(r_addr, qc, fa.eid, spb)) { + log_err("Failed to send flow update packet."); + ipcp_spb_release(spb); + return -1; + } + + return 0; +} + int fa_dealloc(int fd) { if (ipcp_flow_fini(fd) < 0) @@ -883,17 +975,17 @@ static int fa_update_remote(int fd, uint16_t ece) { struct fa_msg * msg; - struct shm_du_buff * sdb; + struct ssm_pk_buff * spb; qoscube_t qc = QOS_CUBE_BE; struct fa_flow * flow; uint64_t r_addr; - if (ipcp_sdb_reserve(&sdb, sizeof(*msg))) { - log_err("Failed to reserve sdb (%zu bytes).", sizeof(*msg)); + if (ipcp_spb_reserve(&spb, sizeof(*msg))) { + log_err("Failed to reserve spb (%zu bytes).", sizeof(*msg)); return -1; } - msg = (struct fa_msg *) shm_du_buff_head(sdb); + msg = (struct fa_msg *) ssm_pk_buff_head(spb); memset(msg, 0, sizeof(*msg)); @@ -912,9 +1004,9 @@ static int fa_update_remote(int fd, pthread_rwlock_unlock(&fa.flows_lock); - if (dt_write_packet(r_addr, qc, fa.eid, sdb)) { + if (dt_write_packet(r_addr, qc, fa.eid, spb)) { log_err("Failed to send flow update packet."); - ipcp_sdb_release(sdb); + ipcp_spb_release(spb); return -1; } @@ -923,7 +1015,7 @@ static int fa_update_remote(int fd, void fa_np1_rcv(uint64_t eid, uint8_t ecn, - struct shm_du_buff * sdb) + struct ssm_pk_buff * spb) { struct fa_flow * flow; bool update; @@ -931,7 +1023,7 @@ void fa_np1_rcv(uint64_t eid, int fd; size_t len; - len = shm_du_buff_len(sdb); + len = ssm_pk_buff_len(spb); pthread_rwlock_wrlock(&fa.flows_lock); @@ -939,7 +1031,7 @@ void fa_np1_rcv(uint64_t eid, if (fd < 0) { pthread_rwlock_unlock(&fa.flows_lock); log_dbg("Received packet for unknown EID %" PRIu64 ".", eid); - ipcp_sdb_release(sdb); + ipcp_spb_release(spb); return; } @@ -953,9 +1045,9 @@ void fa_np1_rcv(uint64_t eid, pthread_rwlock_unlock(&fa.flows_lock); - if (ipcp_flow_write(fd, sdb) < 0) { + if (np1_flow_write(fd, spb, NP1_GET_POOL(fd)) < 0) { log_dbg("Failed to write to flow %d.", fd); - ipcp_sdb_release(sdb); + ipcp_spb_release(spb); #ifdef IPCP_FLOW_STATS pthread_rwlock_wrlock(&fa.flows_lock); ++flow->p_rcv_f; diff --git a/src/ipcpd/unicast/fa.h b/src/ipcpd/unicast/fa.h index 1e716966..f31b40e9 100644 --- a/src/ipcpd/unicast/fa.h +++ b/src/ipcpd/unicast/fa.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Flow allocator of the IPC Process * @@ -45,8 +45,11 @@ int fa_alloc_resp(int fd, int fa_dealloc(int fd); +int fa_irm_update(int fd, + const buffer_t * data); + void fa_np1_rcv(uint64_t eid, uint8_t ecn, - struct shm_du_buff * sdb); + struct ssm_pk_buff * spb); #endif /* OUROBOROS_IPCPD_UNICAST_FA_H */ diff --git a/src/ipcpd/unicast/main.c b/src/ipcpd/unicast/main.c index e6cb2994..1155b88b 100644 --- a/src/ipcpd/unicast/main.c +++ b/src/ipcpd/unicast/main.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Unicast IPC Process * @@ -55,13 +55,8 @@ #include <assert.h> #include <inttypes.h> -struct ipcp ipcpi; - -static int initialize_components(const struct ipcp_config * conf) +static int initialize_components(struct ipcp_config * conf) { - strcpy(ipcpi.layer_name, conf->layer_info.name); - ipcpi.dir_hash_algo = (enum hash_algo) conf->layer_info.dir_hash_algo; - assert(ipcp_dir_hash_len() != 0); if (addr_auth_init(conf->unicast.addr_auth_type, @@ -70,13 +65,7 @@ static int initialize_components(const struct ipcp_config * conf) goto fail_addr_auth; } - ipcpi.dt_addr = addr_auth_address(); - if (ipcpi.dt_addr == 0) { - log_err("Failed to get a valid address."); - goto fail_addr_auth; - } - - log_info("IPCP got address %" PRIu64 ".", ipcpi.dt_addr); + log_info("IPCP got address %" PRIu64 ".", addr_auth_address()); if (ca_init(conf->unicast.cong_avoid)) { log_err("Failed to initialize congestion avoidance."); @@ -88,23 +77,25 @@ static int initialize_components(const struct ipcp_config * conf) goto fail_dt; } - if (fa_init()) { - log_err("Failed to initialize flow allocator component."); - goto fail_fa; - } + ipcp_set_dir_hash_algo((enum hash_algo) conf->layer_info.dir_hash_algo); - if (dir_init()) { + if (dir_init(&conf->unicast.dir)) { log_err("Failed to initialize directory."); goto fail_dir; } + if (fa_init()) { + log_err("Failed to initialize flow allocator component."); + goto fail_fa; + } + ipcp_set_state(IPCP_INIT); return 0; - fail_dir: - fa_fini(); fail_fa: + dir_fini(); + fail_dir: dt_fini(); fail_dt: ca_fini(); @@ -116,10 +107,10 @@ static int initialize_components(const struct ipcp_config * conf) static void finalize_components(void) { - dir_fini(); - fa_fini(); + dir_fini(); + dt_fini(); ca_fini(); @@ -129,6 +120,11 @@ static void finalize_components(void) static int start_components(void) { + if (connmgr_start() < 0) { + log_err("Failed to start AP connection manager."); + goto fail_connmgr_start; + } + if (dt_start() < 0) { log_err("Failed to start data transfer."); goto fail_dt_start; @@ -144,27 +140,29 @@ static int start_components(void) goto fail_enroll_start; } - if (connmgr_start() < 0) { - log_err("Failed to start AP connection manager."); - goto fail_connmgr_start; + if (dir_start() < 0) { + log_err("Failed to start directory."); + goto fail_dir_start; } return 0; - fail_connmgr_start: + fail_dir_start: enroll_stop(); fail_enroll_start: fa_stop(); fail_fa_start: dt_stop(); fail_dt_start: + connmgr_stop(); + fail_connmgr_start: ipcp_set_state(IPCP_INIT); return -1; } static void stop_components(void) { - connmgr_stop(); + dir_stop(); enroll_stop(); @@ -172,24 +170,17 @@ static void stop_components(void) dt_stop(); - ipcp_set_state(IPCP_INIT); -} - -static int bootstrap_components(void) -{ - if (dir_bootstrap()) { - log_err("Failed to bootstrap directory."); - return -1; - } + connmgr_stop(); - return 0; + ipcp_set_state(IPCP_BOOT); } static int unicast_ipcp_enroll(const char * dst, struct layer_info * info) { - struct conn conn; - uint8_t id[ENROLL_ID_LEN]; + struct ipcp_config * conf; + struct conn conn; + uint8_t id[ENROLL_ID_LEN]; if (random_buffer(id, ENROLL_ID_LEN) < 0) { log_err("Failed to generate enrollment ID."); @@ -209,7 +200,11 @@ static int unicast_ipcp_enroll(const char * dst, goto fail_enroll_boot; } - if (initialize_components(enroll_get_conf()) < 0) { + conf = enroll_get_conf(); + + *info = conf->layer_info; + + if (initialize_components(conf) < 0) { log_err_id(id, "Failed to initialize components."); goto fail_enroll_boot; } @@ -227,9 +222,6 @@ static int unicast_ipcp_enroll(const char * dst, log_info_id(id, "Enrolled with %s.", dst); - info->dir_hash_algo = (enum pol_dir_hash) ipcpi.dir_hash_algo; - strcpy(info->name, ipcpi.layer_name); - return 0; fail_start_comp: @@ -240,32 +232,25 @@ static int unicast_ipcp_enroll(const char * dst, return -1; } -static int unicast_ipcp_bootstrap(const struct ipcp_config * conf) +static int unicast_ipcp_bootstrap(struct ipcp_config * conf) { assert(conf); assert(conf->type == THIS_TYPE); - enroll_bootstrap(conf); - if (initialize_components(conf) < 0) { log_err("Failed to init IPCP components."); goto fail_init; } + enroll_bootstrap(conf); + if (start_components() < 0) { log_err("Failed to init IPCP components."); goto fail_start; } - if (bootstrap_components() < 0) { - log_err("Failed to bootstrap IPCP components."); - goto fail_bootstrap; - } - return 0; - fail_bootstrap: - stop_components(); fail_start: finalize_components(); fail_init: @@ -288,7 +273,8 @@ static struct ipcp_ops unicast_ops = { .ipcp_flow_alloc = fa_alloc, .ipcp_flow_join = NULL, .ipcp_flow_alloc_resp = fa_alloc_resp, - .ipcp_flow_dealloc = fa_dealloc + .ipcp_flow_dealloc = fa_dealloc, + .ipcp_flow_update = fa_irm_update }; int main(int argc, @@ -322,12 +308,13 @@ int main(int argc, ipcp_sigwait(); if (ipcp_get_state() == IPCP_SHUTDOWN) { + ipcp_stop(); stop_components(); finalize_components(); + } else { + ipcp_stop(); } - ipcp_stop(); - enroll_fini(); connmgr_fini(); diff --git a/src/ipcpd/unicast/pff.c b/src/ipcpd/unicast/pff.c index 9b2aa2b4..c8c3126f 100644 --- a/src/ipcpd/unicast/pff.c +++ b/src/ipcpd/unicast/pff.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * PDU Forwarding Function * diff --git a/src/ipcpd/unicast/pff.h b/src/ipcpd/unicast/pff.h index f44e5531..2eb42ce6 100644 --- a/src/ipcpd/unicast/pff.h +++ b/src/ipcpd/unicast/pff.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * PDU Forwarding Function * diff --git a/src/ipcpd/unicast/pff/alternate.c b/src/ipcpd/unicast/pff/alternate.c index 85e85914..1c508c1b 100644 --- a/src/ipcpd/unicast/pff/alternate.c +++ b/src/ipcpd/unicast/pff/alternate.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Policy for PFF with alternate next hops * @@ -211,7 +211,7 @@ struct pff_i * alternate_pff_create(void) if (pthread_rwlock_init(&tmp->lock, NULL)) goto fail_lock; - tmp->pft = pft_create(PFT_SIZE, false); + tmp->pft = pft_create(PFT_SIZE); if (tmp->pft == NULL) goto fail_pft; diff --git a/src/ipcpd/unicast/pff/alternate.h b/src/ipcpd/unicast/pff/alternate.h index 96207e74..ae3758dc 100644 --- a/src/ipcpd/unicast/pff/alternate.h +++ b/src/ipcpd/unicast/pff/alternate.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Policy for PFF with alternate next hops * diff --git a/src/ipcpd/unicast/pff/multipath.c b/src/ipcpd/unicast/pff/multipath.c index cbab0f5f..9ba59592 100644 --- a/src/ipcpd/unicast/pff/multipath.c +++ b/src/ipcpd/unicast/pff/multipath.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Policy for PFF supporting multipath routing * @@ -63,7 +63,7 @@ struct pff_i * multipath_pff_create(void) if (pthread_rwlock_init(&tmp->lock, NULL)) goto fail_rwlock; - tmp->pft = pft_create(PFT_SIZE, false); + tmp->pft = pft_create(PFT_SIZE); if (tmp->pft == NULL) goto fail_pft; diff --git a/src/ipcpd/unicast/pff/multipath.h b/src/ipcpd/unicast/pff/multipath.h index 0eb03476..5329f7fc 100644 --- a/src/ipcpd/unicast/pff/multipath.h +++ b/src/ipcpd/unicast/pff/multipath.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Policy for PFF supporting multipath routing * diff --git a/src/ipcpd/unicast/pff/ops.h b/src/ipcpd/unicast/pff/ops.h index 16a31273..e4cabd11 100644 --- a/src/ipcpd/unicast/pff/ops.h +++ b/src/ipcpd/unicast/pff/ops.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Pff policy ops * diff --git a/src/ipcpd/unicast/pff/pft.c b/src/ipcpd/unicast/pff/pft.c index 8c436113..d0e562d6 100644 --- a/src/ipcpd/unicast/pff/pft.c +++ b/src/ipcpd/unicast/pff/pft.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Packet forwarding table (PFT) with chaining on collisions * @@ -43,12 +43,10 @@ struct pft_entry { struct pft { struct list_head * buckets; - bool hash_key; uint64_t buckets_size; }; -struct pft * pft_create(uint64_t buckets, - bool hash_key) +struct pft * pft_create(uint64_t buckets) { struct pft * tmp; unsigned int i; @@ -69,7 +67,6 @@ struct pft * pft_create(uint64_t buckets, if (tmp == NULL) return NULL; - tmp->hash_key = hash_key; tmp->buckets_size = buckets; tmp->buckets = malloc(buckets * sizeof(*tmp->buckets)); @@ -113,22 +110,10 @@ void pft_flush(struct pft * pft) } } -static uint64_t hash(uint64_t key) -{ - uint64_t res[2]; - - mem_hash(HASH_MD5, res, (uint8_t *) &key, sizeof(key)); - - return res[0]; -} - static uint64_t calc_key(struct pft * pft, uint64_t dst) { - if (pft->hash_key) - dst = hash(dst); - - return (dst & (pft->buckets_size - 1)); + return hash_mix64(dst) & (pft->buckets_size - 1); } int pft_insert(struct pft * pft, diff --git a/src/ipcpd/unicast/pff/pft.h b/src/ipcpd/unicast/pff/pft.h index 711dabcb..15bbe451 100644 --- a/src/ipcpd/unicast/pff/pft.h +++ b/src/ipcpd/unicast/pff/pft.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Packet forwarding table (PFT) with chaining on collisions * @@ -24,14 +24,12 @@ #define OUROBOROS_PFT_H #include <stdint.h> -#include <stdbool.h> #include <stdlib.h> struct pft; /* Buckets is rounded up to the nearest power of 2 */ -struct pft * pft_create(uint64_t buckets, - bool hash_key); +struct pft * pft_create(uint64_t buckets); void pft_destroy(struct pft * table); diff --git a/src/ipcpd/unicast/pff/pol.h b/src/ipcpd/unicast/pff/pol.h index 245b03c4..e31c2794 100644 --- a/src/ipcpd/unicast/pff/pol.h +++ b/src/ipcpd/unicast/pff/pol.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * PDU Forwarding Function policies * diff --git a/src/ipcpd/unicast/pff/simple.c b/src/ipcpd/unicast/pff/simple.c index 5f95e3ce..7befa42f 100644 --- a/src/ipcpd/unicast/pff/simple.c +++ b/src/ipcpd/unicast/pff/simple.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Simple PDU Forwarding Function * @@ -63,7 +63,7 @@ struct pff_i * simple_pff_create(void) return NULL; } - tmp->pft = pft_create(PFT_SIZE, false); + tmp->pft = pft_create(PFT_SIZE); if (tmp->pft == NULL) { pthread_rwlock_destroy(&tmp->lock); free(tmp); diff --git a/src/ipcpd/unicast/pff/simple.h b/src/ipcpd/unicast/pff/simple.h index 0966a186..1046e4c4 100644 --- a/src/ipcpd/unicast/pff/simple.h +++ b/src/ipcpd/unicast/pff/simple.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Simple policy for PFF * diff --git a/src/ipcpd/unicast/pff/tests/CMakeLists.txt b/src/ipcpd/unicast/pff/tests/CMakeLists.txt index e7082372..8c0e3d51 100644 --- a/src/ipcpd/unicast/pff/tests/CMakeLists.txt +++ b/src/ipcpd/unicast/pff/tests/CMakeLists.txt @@ -3,32 +3,32 @@ get_filename_component(CURRENT_SOURCE_PARENT_DIR get_filename_component(CURRENT_BINARY_PARENT_DIR ${CMAKE_CURRENT_BINARY_DIR} DIRECTORY) -include_directories(${CMAKE_CURRENT_SOURCE_DIR}) -include_directories(${CMAKE_CURRENT_BINARY_DIR}) - -include_directories(${CURRENT_SOURCE_PARENT_DIR}) -include_directories(${CURRENT_BINARY_PARENT_DIR}) - -include_directories(${CMAKE_SOURCE_DIR}/include) -include_directories(${CMAKE_BINARY_DIR}/include) - get_filename_component(PARENT_PATH ${CMAKE_CURRENT_SOURCE_DIR} DIRECTORY) get_filename_component(PARENT_DIR ${PARENT_PATH} NAME) +compute_test_prefix() + create_test_sourcelist(${PARENT_DIR}_tests test_suite.c # Add new tests here pft_test.c ) -add_executable(${PARENT_DIR}_test EXCLUDE_FROM_ALL ${${PARENT_DIR}_tests}) -target_link_libraries(${PARENT_DIR}_test ouroboros-common) +add_executable(${PARENT_DIR}_test ${${PARENT_DIR}_tests}) + +target_include_directories(${PARENT_DIR}_test PRIVATE + ${CMAKE_CURRENT_SOURCE_DIR} + ${CMAKE_CURRENT_BINARY_DIR} + ${CURRENT_SOURCE_PARENT_DIR} + ${CURRENT_BINARY_PARENT_DIR} + ${CMAKE_SOURCE_DIR}/include + ${CMAKE_BINARY_DIR}/include + ${CMAKE_SOURCE_DIR}/src/ipcpd + ${CMAKE_BINARY_DIR}/src/ipcpd +) -add_dependencies(check ${PARENT_DIR}_test) +disable_test_logging_for_target(${PARENT_DIR}_test) +target_link_libraries(${PARENT_DIR}_test PRIVATE ouroboros-common) -set(tests_to_run ${${PARENT_DIR}_tests}) -remove(tests_to_run test_suite.c) +add_dependencies(build_tests ${PARENT_DIR}_test) -foreach (test ${tests_to_run}) - get_filename_component(test_name ${test} NAME_WE) - add_test(${test_name} ${C_TEST_PATH}/${PARENT_DIR}_test ${test_name}) -endforeach (test) +ouroboros_register_tests(TARGET ${PARENT_DIR}_test TESTS ${${PARENT_DIR}_tests}) diff --git a/src/ipcpd/unicast/pff/tests/pft_test.c b/src/ipcpd/unicast/pff/tests/pft_test.c index 18287fb8..20e73a94 100644 --- a/src/ipcpd/unicast/pff/tests/pft_test.c +++ b/src/ipcpd/unicast/pff/tests/pft_test.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Test of the hash table * @@ -38,15 +38,7 @@ int pft_test(int argc, (void) argc; (void) argv; - pft = pft_create(TBL_SIZE, true); - if (pft == NULL) { - printf("Failed to create.\n"); - return -1; - } - - pft_destroy(pft); - - pft = pft_create(TBL_SIZE, false); + pft = pft_create(TBL_SIZE); if (pft == NULL) { printf("Failed to create.\n"); return -1; diff --git a/src/ipcpd/unicast/psched.c b/src/ipcpd/unicast/psched.c index 7e12148b..21e23617 100644 --- a/src/ipcpd/unicast/psched.c +++ b/src/ipcpd/unicast/psched.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Packet scheduler component * @@ -41,11 +41,13 @@ #include <stdlib.h> #include <string.h> +#ifndef BUILD_CONTAINER static int qos_prio [] = { QOS_PRIO_BE, QOS_PRIO_VIDEO, QOS_PRIO_VOICE, }; +#endif struct psched { fset_t * set[QOS_CUBE_MAX]; @@ -67,7 +69,7 @@ static void cleanup_reader(void * o) static void * packet_reader(void * o) { struct psched * sched; - struct shm_du_buff * sdb; + struct ssm_pk_buff * spb; int fd; fqueue_t * fq; qoscube_t qc; @@ -102,10 +104,10 @@ static void * packet_reader(void * o) notifier_event(NOTIFY_DT_FLOW_UP, &fd); break; case FLOW_PKT: - if (sched->read(fd, &sdb) < 0) + if (sched->read(fd, &spb) < 0) continue; - sched->callback(fd, qc, sdb); + sched->callback(fd, qc, spb); break; default: break; @@ -168,6 +170,7 @@ struct psched * psched_create(next_packet_fn_t callback, } } +#ifndef BUILD_CONTAINER for (i = 0; i < QOS_CUBE_MAX * IPCP_SCHED_THR_MUL; ++i) { struct sched_param par; int pol = SCHED_RR; @@ -185,14 +188,17 @@ struct psched * psched_create(next_packet_fn_t callback, if (pthread_setschedparam(psched->readers[i], pol, &par)) goto fail_sched; } +#endif return psched; +#ifndef BUILD_CONTAINER fail_sched: for (j = 0; j < QOS_CUBE_MAX * IPCP_SCHED_THR_MUL; ++j) pthread_cancel(psched->readers[j]); for (j = 0; j < QOS_CUBE_MAX * IPCP_SCHED_THR_MUL; ++j) pthread_join(psched->readers[j], NULL); +#endif fail_infos: for (j = 0; j < QOS_CUBE_MAX; ++j) fset_destroy(psched->set[j]); diff --git a/src/ipcpd/unicast/psched.h b/src/ipcpd/unicast/psched.h index 831f8084..d83bb793 100644 --- a/src/ipcpd/unicast/psched.h +++ b/src/ipcpd/unicast/psched.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Packet scheduler component * @@ -28,10 +28,10 @@ typedef void (* next_packet_fn_t)(int fd, qoscube_t qc, - struct shm_du_buff * sdb); + struct ssm_pk_buff * spb); typedef int (* read_fn_t)(int fd, - struct shm_du_buff ** sdb); + struct ssm_pk_buff ** spb); struct psched * psched_create(next_packet_fn_t callback, read_fn_t read); diff --git a/src/ipcpd/unicast/routing.c b/src/ipcpd/unicast/routing.c index f5417c24..1a4e4372 100644 --- a/src/ipcpd/unicast/routing.c +++ b/src/ipcpd/unicast/routing.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Routing component of the IPCP * @@ -30,31 +30,26 @@ struct routing_ops * r_ops; -int routing_init(enum pol_routing pr) +int routing_init(struct routing_config * conf, + enum pol_pff * pff_type) { - enum pol_pff pff_type; + void * cfg; - switch (pr) { + switch (conf->pol) { case ROUTING_LINK_STATE: - pff_type = PFF_SIMPLE; - r_ops = &link_state_ops; - break; - case ROUTING_LINK_STATE_LFA: - pff_type = PFF_ALTERNATE; - r_ops = &link_state_ops; - break; - case ROUTING_LINK_STATE_ECMP: - pff_type=PFF_MULTIPATH; r_ops = &link_state_ops; + cfg = &conf->ls; break; default: return -ENOTSUP; } - if (r_ops->init(pr)) - return -1; + return r_ops->init(cfg, pff_type); +} - return pff_type; +int routing_start(void) +{ + return r_ops->start(); } struct routing_i * routing_i_create(struct pff * pff) @@ -67,6 +62,11 @@ void routing_i_destroy(struct routing_i * instance) return r_ops->routing_i_destroy(instance); } +void routing_stop(void) +{ + r_ops->stop(); +} + void routing_fini(void) { r_ops->fini(); diff --git a/src/ipcpd/unicast/routing.h b/src/ipcpd/unicast/routing.h index d5d833ae..8d721095 100644 --- a/src/ipcpd/unicast/routing.h +++ b/src/ipcpd/unicast/routing.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Routing component of the IPCP * @@ -30,10 +30,15 @@ #include <stdint.h> -int routing_init(enum pol_routing pr); +int routing_init(struct routing_config * conf, + enum pol_pff * pff_type); void routing_fini(void); +int routing_start(void); + +void routing_stop(void); + struct routing_i * routing_i_create(struct pff * pff); void routing_i_destroy(struct routing_i * instance); diff --git a/src/ipcpd/unicast/routing/graph.c b/src/ipcpd/unicast/routing/graph.c index 32f3e6fb..c168eb7d 100644 --- a/src/ipcpd/unicast/routing/graph.c +++ b/src/ipcpd/unicast/routing/graph.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Undirected graph structure * @@ -57,8 +57,8 @@ struct edge { }; struct graph { - size_t nr_vertices; - struct list_head vertices; + struct llist vertices; + pthread_mutex_t lock; }; @@ -67,7 +67,7 @@ static struct edge * find_edge_by_addr(struct vertex * vertex, { struct list_head * p; - assert(vertex); + assert(vertex != NULL); list_for_each(p, &vertex->edges) { struct edge * e = list_entry(p, struct edge, next); @@ -85,7 +85,7 @@ static struct vertex * find_vertex_by_addr(struct graph * graph, assert(graph); - list_for_each(p, &graph->vertices) { + llist_for_each(p, &graph->vertices) { struct vertex * e = list_entry(p, struct vertex, next); if (e->addr == addr) return e; @@ -99,8 +99,8 @@ static struct edge * add_edge(struct vertex * vertex, { struct edge * edge; - assert(vertex); - assert(nb); + assert(vertex != NULL); + assert(nb != NULL); edge = malloc(sizeof(*edge)); if (edge == NULL) @@ -139,7 +139,7 @@ static struct vertex * add_vertex(struct graph * graph, vertex->addr = addr; /* Keep them ordered on address. */ - list_for_each(p, &graph->vertices) { + llist_for_each(p, &graph->vertices) { struct vertex * v = list_entry(p, struct vertex, next); if (v->addr > addr) break; @@ -148,46 +148,50 @@ static struct vertex * add_vertex(struct graph * graph, vertex->index = i; - list_add_tail(&vertex->next, p); + llist_add_tail_at(&vertex->next, p, &graph->vertices); /* Increase the index of the vertices to the right. */ - list_for_each(p, &graph->vertices) { + list_for_each(p, &vertex->next) { struct vertex * v = list_entry(p, struct vertex, next); if (v->addr > addr) v->index++; } - graph->nr_vertices++; - return vertex; } +static void free_edges(struct list_head * edges) +{ + struct list_head * p; + struct list_head * h; + + list_for_each_safe(p, h, edges) { + struct edge * e = list_entry(p, struct edge, next); + list_del(&e->next); + free(e); + } +} + static void del_vertex(struct graph * graph, struct vertex * vertex) { struct list_head * p; - struct list_head * h; - assert(graph); - assert(vertex); + assert(graph != NULL); + assert(vertex != NULL); - list_del(&vertex->next); + llist_del(&vertex->next, &graph->vertices); /* Decrease the index of the vertices to the right. */ - list_for_each(p, &graph->vertices) { + llist_for_each(p, &graph->vertices) { struct vertex * v = list_entry(p, struct vertex, next); if (v->addr > vertex->addr) v->index--; } - list_for_each_safe(p, h, &vertex->edges) { - struct edge * e = list_entry(p, struct edge, next); - del_edge(e); - } + free_edges(&vertex->edges); free(vertex); - - graph->nr_vertices--; } struct graph * graph_create(void) @@ -203,8 +207,7 @@ struct graph * graph_create(void) return NULL; } - graph->nr_vertices = 0; - list_head_init(&graph->vertices); + llist_init(&graph->vertices); return graph; } @@ -218,7 +221,7 @@ void graph_destroy(struct graph * graph) pthread_mutex_lock(&graph->lock); - list_for_each_safe(p, n, &graph->vertices) { + llist_for_each_safe(p, n, &graph->vertices) { struct vertex * e = list_entry(p, struct vertex, next); del_vertex(graph, e); } @@ -227,6 +230,8 @@ void graph_destroy(struct graph * graph) pthread_mutex_destroy(&graph->lock); + assert(llist_is_empty(&graph->vertices)); + free(graph); } @@ -240,63 +245,35 @@ int graph_update_edge(struct graph * graph, struct vertex * nb; struct edge * nb_e; - assert(graph); + assert(graph != NULL); pthread_mutex_lock(&graph->lock); v = find_vertex_by_addr(graph, s_addr); - if (v == NULL) { - v = add_vertex(graph, s_addr); - if (v == NULL) { - pthread_mutex_unlock(&graph->lock); - log_err("Failed to add vertex."); - return -ENOMEM; - } + if (v == NULL && ((v = add_vertex(graph, s_addr)) == NULL)) {; + log_err("Failed to add src vertex."); + goto fail_add_s; } nb = find_vertex_by_addr(graph, d_addr); - if (nb == NULL) { - nb = add_vertex(graph, d_addr); - if (nb == NULL) { - if (list_is_empty(&v->edges)) - del_vertex(graph, v); - pthread_mutex_unlock(&graph->lock); - log_err("Failed to add vertex."); - return -ENOMEM; - } + if (nb == NULL && ((nb = add_vertex(graph, d_addr)) == NULL)) { + log_err("Failed to add dst vertex."); + goto fail_add_d; } e = find_edge_by_addr(v, d_addr); - if (e == NULL) { - e = add_edge(v, nb); - if (e == NULL) { - if (list_is_empty(&v->edges)) - del_vertex(graph, v); - if (list_is_empty(&nb->edges)) - del_vertex(graph, nb); - pthread_mutex_unlock(&graph->lock); - log_err("Failed to add edge."); - return -ENOMEM; - } + if (e == NULL && ((e = add_edge(v, nb)) == NULL)) { + log_err("Failed to add edge to dst."); + goto fail_add_edge_d; } e->announced++; e->qs = qs; nb_e = find_edge_by_addr(nb, s_addr); - if (nb_e == NULL) { - nb_e = add_edge(nb, v); - if (nb_e == NULL) { - if (--e->announced == 0) - del_edge(e); - if (list_is_empty(&v->edges)) - del_vertex(graph, v); - if (list_is_empty(&nb->edges)) - del_vertex(graph, nb); - pthread_mutex_unlock(&graph->lock); - log_err("Failed to add edge."); - return -ENOMEM; - } + if (nb_e == NULL && ((nb_e = add_edge(nb, v)) == NULL)) {; + log_err("Failed to add edge to src."); + goto fail_add_edge_s; } nb_e->announced++; @@ -305,6 +282,19 @@ int graph_update_edge(struct graph * graph, pthread_mutex_unlock(&graph->lock); return 0; + fail_add_edge_s: + if (--e->announced == 0) + del_edge(e); + fail_add_edge_d: + if (list_is_empty(&nb->edges)) + del_vertex(graph, nb); + fail_add_d: + if (list_is_empty(&v->edges)) + del_vertex(graph, v); + fail_add_s: + pthread_mutex_unlock(&graph->lock); + return -ENOMEM; + } int graph_del_edge(struct graph * graph, @@ -322,30 +312,26 @@ int graph_del_edge(struct graph * graph, v = find_vertex_by_addr(graph, s_addr); if (v == NULL) { - pthread_mutex_unlock(&graph->lock); - log_err("No such source vertex."); - return -1; + log_err("Failed to find src vertex."); + goto fail; } nb = find_vertex_by_addr(graph, d_addr); if (nb == NULL) { - pthread_mutex_unlock(&graph->lock); log_err("No such destination vertex."); - return -1; + goto fail; } e = find_edge_by_addr(v, d_addr); if (e == NULL) { - pthread_mutex_unlock(&graph->lock); log_err("No such source edge."); - return -1; + goto fail; } nb_e = find_edge_by_addr(nb, s_addr); if (nb_e == NULL) { - pthread_mutex_unlock(&graph->lock); log_err("No such destination edge."); - return -1; + goto fail; } if (--e->announced == 0) @@ -362,6 +348,10 @@ int graph_del_edge(struct graph * graph, pthread_mutex_unlock(&graph->lock); return 0; + + fail: + pthread_mutex_unlock(&graph->lock); + return -1; } static int get_min_vertex(struct graph * graph, @@ -381,7 +371,7 @@ static int get_min_vertex(struct graph * graph, *v = NULL; - list_for_each(p, &graph->vertices) { + llist_for_each(p, &graph->vertices) { if (!used[i] && dist[i] < min) { min = dist[i]; index = i; @@ -413,24 +403,24 @@ static int dijkstra(struct graph * graph, assert(nhops); assert(dist); - *nhops = malloc(sizeof(**nhops) * graph->nr_vertices); + *nhops = malloc(sizeof(**nhops) * graph->vertices.len); if (*nhops == NULL) goto fail_pnhops; - *dist = malloc(sizeof(**dist) * graph->nr_vertices); + *dist = malloc(sizeof(**dist) * graph->vertices.len); if (*dist == NULL) goto fail_pdist; - used = malloc(sizeof(*used) * graph->nr_vertices); + used = malloc(sizeof(*used) * graph->vertices.len); if (used == NULL) goto fail_used; /* Init the data structures */ - memset(used, 0, sizeof(*used) * graph->nr_vertices); - memset(*nhops, 0, sizeof(**nhops) * graph->nr_vertices); - memset(*dist, 0, sizeof(**dist) * graph->nr_vertices); + memset(used, 0, sizeof(*used) * graph->vertices.len); + memset(*nhops, 0, sizeof(**nhops) * graph->vertices.len); + memset(*dist, 0, sizeof(**dist) * graph->vertices.len); - list_for_each(p, &graph->vertices) { + llist_for_each(p, &graph->vertices) { v = list_entry(p, struct vertex, next); (*dist)[i++] = (v->addr == src) ? 0 : INT_MAX; } @@ -527,7 +517,7 @@ static int graph_routing_table_simple(struct graph * graph, assert(dist); /* We need at least 2 vertices for a table */ - if (graph->nr_vertices < 2) + if (graph->vertices.len < 2) goto fail_vertices; if (dijkstra(graph, s_addr, &nhops, dist)) @@ -536,7 +526,7 @@ static int graph_routing_table_simple(struct graph * graph, list_head_init(table); /* Now construct the routing table from the nhops. */ - list_for_each(p, &graph->vertices) { + llist_for_each(p, &graph->vertices) { v = list_entry(p, struct vertex, next); /* This is the src */ @@ -613,9 +603,9 @@ static int graph_routing_table_lfa(struct graph * graph, struct list_head * table, int ** dist) { - int * n_dist[PROG_MAX_FLOWS]; - uint64_t addrs[PROG_MAX_FLOWS]; - int n_index[PROG_MAX_FLOWS]; + int * n_dist[PROC_MAX_FLOWS]; + uint64_t addrs[PROC_MAX_FLOWS]; + int n_index[PROC_MAX_FLOWS]; struct list_head * p; struct list_head * q; struct vertex * v; @@ -628,13 +618,13 @@ static int graph_routing_table_lfa(struct graph * graph, if (graph_routing_table_simple(graph, s_addr, table, dist)) goto fail_table; - for (j = 0; j < PROG_MAX_FLOWS; j++) { + for (j = 0; j < PROC_MAX_FLOWS; j++) { n_dist[j] = NULL; n_index[j] = -1; addrs[j] = -1; } - list_for_each(p, &graph->vertices) { + llist_for_each(p, &graph->vertices) { v = list_entry(p, struct vertex, next); if (v->addr != s_addr) @@ -660,7 +650,7 @@ static int graph_routing_table_lfa(struct graph * graph, } /* Loop though all nodes to see if we have a LFA for them. */ - list_for_each(p, &graph->vertices) { + llist_for_each(p, &graph->vertices) { v = list_entry(p, struct vertex, next); if (v->addr == s_addr) @@ -705,7 +695,6 @@ static int graph_routing_table_ecmp(struct graph * graph, { struct vertex ** nhops; struct list_head * p; - struct list_head * h; size_t i; struct vertex * v; struct vertex * src_v; @@ -717,14 +706,14 @@ static int graph_routing_table_ecmp(struct graph * graph, assert(graph); assert(dist); - if (graph-> nr_vertices < 2) + if (graph->vertices.len < 2) goto fail_vertices; - forwarding = malloc(sizeof(*forwarding) * graph->nr_vertices); + forwarding = malloc(sizeof(*forwarding) * graph->vertices.len); if (forwarding == NULL) goto fail_vertices; - for (i = 0; i < graph->nr_vertices; ++i) + for (i = 0; i < graph->vertices.len; ++i) list_head_init(&forwarding[i]); if (dijkstra(graph, s_addr, &nhops, dist)) @@ -745,16 +734,15 @@ static int graph_routing_table_ecmp(struct graph * graph, free(nhops); - list_for_each(h, &graph->vertices) { - v = list_entry(h, struct vertex, next); - if (tmp_dist[v->index] + 1 == (*dist)[v->index]) { + for (i = 0; i < graph->vertices.len; ++i) { + if (tmp_dist[i] + 1 == (*dist)[i]) { n = malloc(sizeof(*n)); if (n == NULL) { free(tmp_dist); goto fail_src_v; } n->nhop = e->nb->addr; - list_add_tail(&n->next, &forwarding[v->index]); + list_add_tail(&n->next, &forwarding[i]); } } @@ -763,38 +751,34 @@ static int graph_routing_table_ecmp(struct graph * graph, list_head_init(table); i = 0; - list_for_each(p, &graph->vertices) { + llist_for_each(p, &graph->vertices) { v = list_entry(p, struct vertex, next); - if (v->addr == s_addr) { + if (v->addr == s_addr || list_is_empty(&forwarding[i])) { ++i; continue; } t = malloc(sizeof(*t)); if (t == NULL) - goto fail_t; + goto fail_malloc; t->dst = v->addr; list_head_init(&t->nhops); - if (&forwarding[i] != forwarding[i].nxt) { - t->nhops.nxt = forwarding[i].nxt; - t->nhops.prv = forwarding[i].prv; - forwarding[i].prv->nxt = &t->nhops; - forwarding[i].nxt->prv = &t->nhops; - } + t->nhops.nxt = forwarding[i].nxt; + t->nhops.prv = forwarding[i].prv; + forwarding[i].prv->nxt = &t->nhops; + forwarding[i].nxt->prv = &t->nhops; list_add(&t->next, table); ++i; } - free(*dist); - *dist = NULL; free(forwarding); return 0; - fail_t: + fail_malloc: free_routing_table(table); fail_src_v: free(*dist); diff --git a/src/ipcpd/unicast/routing/graph.h b/src/ipcpd/unicast/routing/graph.h index 8190cc6c..f3766771 100644 --- a/src/ipcpd/unicast/routing/graph.h +++ b/src/ipcpd/unicast/routing/graph.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Undirected graph structure * diff --git a/src/ipcpd/unicast/routing/link-state.c b/src/ipcpd/unicast/routing/link-state.c index 57c0c7cb..c4ea9e1c 100644 --- a/src/ipcpd/unicast/routing/link-state.c +++ b/src/ipcpd/unicast/routing/link-state.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Link state routing policy * @@ -42,6 +42,7 @@ #include <ouroboros/rib.h> #include <ouroboros/utils.h> +#include "addr-auth.h" #include "common/comp.h" #include "common/connmgr.h" #include "graph.h" @@ -54,16 +55,19 @@ #include <inttypes.h> #include <string.h> -#define RECALC_TIME 4 -#define LS_UPDATE_TIME 15 -#define LS_TIMEO 60 #define LS_ENTRY_SIZE 104 -#define LSDB "lsdb" +#define lsdb "lsdb" #ifndef CLOCK_REALTIME_COARSE #define CLOCK_REALTIME_COARSE CLOCK_REALTIME #endif +#define LINK_FMT ADDR_FMT32 "--" ADDR_FMT32 +#define LINK_VAL(src, dst) ADDR_VAL32(&src), ADDR_VAL32(&dst) + +#define LSU_FMT "LSU ["ADDR_FMT32 " -- " ADDR_FMT32 " seq: %09" PRIu64 "]" +#define LSU_VAL(src, dst, seqno) ADDR_VAL32(&src), ADDR_VAL32(&dst), seqno + struct lsa { uint64_t d_addr; uint64_t s_addr; @@ -106,30 +110,37 @@ struct nb { }; struct { - struct list_head nbs; - size_t nbs_len; + uint64_t addr; + + enum routing_algo routing_algo; + + struct ls_config conf; + fset_t * mgmt_set; - struct list_head db; - size_t db_len; + struct graph * graph; - pthread_rwlock_t db_lock; + struct { + struct llist nbs; + struct llist db; + pthread_rwlock_t lock; + }; - struct graph * graph; + struct { + struct list_head list; + pthread_mutex_t mtx; + } instances; pthread_t lsupdate; pthread_t lsreader; pthread_t listener; - - struct list_head routing_instances; - pthread_mutex_t routing_i_lock; - - enum routing_algo routing_algo; } ls; struct routing_ops link_state_ops = { - .init = link_state_init, + .init = (int (*)(void *, enum pol_pff *)) link_state_init, .fini = link_state_fini, + .start = link_state_start, + .stop = link_state_stop, .routing_i_create = link_state_routing_i_create, .routing_i_destroy = link_state_routing_i_destroy }; @@ -138,7 +149,7 @@ static int str_adj(struct adjacency * adj, char * buf, size_t len) { - char tmbuf[64]; + char tmstr[RIB_TM_STRLEN]; char srcbuf[64]; char dstbuf[64]; char seqnobuf[64]; @@ -149,15 +160,16 @@ static int str_adj(struct adjacency * adj, if (len < LS_ENTRY_SIZE) return -1; - tm = localtime(&adj->stamp); - strftime(tmbuf, sizeof(tmbuf), "%F %T", tm); /* 19 chars */ + tm = gmtime(&adj->stamp); + strftime(tmstr, sizeof(tmstr), RIB_TM_FORMAT, tm); - sprintf(srcbuf, "%" PRIu64, adj->src); - sprintf(dstbuf, "%" PRIu64, adj->dst); + sprintf(srcbuf, ADDR_FMT32, ADDR_VAL32(&adj->src)); + sprintf(dstbuf, ADDR_FMT32, ADDR_VAL32(&adj->dst)); sprintf(seqnobuf, "%" PRIu64, adj->seqno); - sprintf(buf, "src: %20s\ndst: %20s\nseqno: %18s\nupd: %20s\n", - srcbuf, dstbuf, seqnobuf, tmbuf); + sprintf(buf, "src: %20s\ndst: %20s\nseqno: %18s\n" + "upd: %s\n", + srcbuf, dstbuf, seqnobuf, tmstr); return LS_ENTRY_SIZE; } @@ -169,9 +181,9 @@ static struct adjacency * get_adj(const char * path) assert(path); - list_for_each(p, &ls.db) { + llist_for_each(p, &ls.db) { struct adjacency * a = list_entry(p, struct adjacency, next); - sprintf(entry, "%" PRIu64 ".%" PRIu64, a->src, a->dst); + sprintf(entry, LINK_FMT, LINK_VAL(a->src, a->dst)); if (strcmp(entry, path) == 0) return a; } @@ -194,7 +206,7 @@ static int lsdb_rib_getattr(const char * path, clock_gettime(CLOCK_REALTIME_COARSE, &now); - pthread_rwlock_rdlock(&ls.db_lock); + pthread_rwlock_rdlock(&ls.lock); adj = get_adj(entry); if (adj != NULL) { @@ -205,7 +217,7 @@ static int lsdb_rib_getattr(const char * path, attr->size = 0; } - pthread_rwlock_unlock(&ls.db_lock); + pthread_rwlock_unlock(&ls.lock); return 0; } @@ -223,9 +235,9 @@ static int lsdb_rib_read(const char * path, entry = strstr(path, RIB_SEPARATOR) + 1; assert(entry); - pthread_rwlock_rdlock(&ls.db_lock); + pthread_rwlock_rdlock(&ls.lock); - if (ls.db_len + ls.nbs_len == 0) + if (llist_is_empty(&ls.db) && llist_is_empty(&ls.nbs)) goto fail; a = get_adj(entry); @@ -236,11 +248,11 @@ static int lsdb_rib_read(const char * path, if (size < 0) goto fail; - pthread_rwlock_unlock(&ls.db_lock); + pthread_rwlock_unlock(&ls.lock); return size; fail: - pthread_rwlock_unlock(&ls.db_lock); + pthread_rwlock_unlock(&ls.lock); return -1; } @@ -250,60 +262,52 @@ static int lsdb_rib_readdir(char *** buf) char entry[RIB_PATH_LEN + 1]; ssize_t idx = 0; - assert(buf); + assert(buf != NULL); - pthread_rwlock_rdlock(&ls.db_lock); + pthread_rwlock_rdlock(&ls.lock); - if (ls.db_len + ls.nbs_len == 0) { - pthread_rwlock_unlock(&ls.db_lock); - return 0; + if (llist_is_empty(&ls.db) && llist_is_empty(&ls.nbs)) { + *buf = NULL; + goto no_entries; } - *buf = malloc(sizeof(**buf) * (ls.db_len + ls.nbs_len)); - if (*buf == NULL) { - pthread_rwlock_unlock(&ls.db_lock); - return -ENOMEM; - } - list_for_each(p, &ls.nbs) { + *buf = malloc(sizeof(**buf) * (ls.db.len + ls.nbs.len)); + if (*buf == NULL) + goto fail_entries; + + llist_for_each(p, &ls.nbs) { struct nb * nb = list_entry(p, struct nb, next); - char * str = (nb->type == NB_DT ? "dt." : "mgmt."); - sprintf(entry, "%s%" PRIu64, str, nb->addr); + char * str = (nb->type == NB_DT ? ".dt " : ".mgmt "); + sprintf(entry, "%s" ADDR_FMT32 , str, ADDR_VAL32(&nb->addr)); (*buf)[idx] = malloc(strlen(entry) + 1); - if ((*buf)[idx] == NULL) { - while (idx-- > 0) - free((*buf)[idx]); - free(*buf); - pthread_rwlock_unlock(&ls.db_lock); - return -ENOMEM; - } - - strcpy((*buf)[idx], entry); + if ((*buf)[idx] == NULL) + goto fail_entry; - idx++; + strcpy((*buf)[idx++], entry); } - list_for_each(p, &ls.db) { + llist_for_each(p, &ls.db) { struct adjacency * a = list_entry(p, struct adjacency, next); - sprintf(entry, "%" PRIu64 ".%" PRIu64, a->src, a->dst); + sprintf(entry, LINK_FMT, LINK_VAL(a->src, a->dst)); (*buf)[idx] = malloc(strlen(entry) + 1); - if ((*buf)[idx] == NULL) { - ssize_t j; - for (j = 0; j < idx; ++j) - free(*buf[j]); - free(buf); - pthread_rwlock_unlock(&ls.db_lock); - return -ENOMEM; - } - - strcpy((*buf)[idx], entry); + if ((*buf)[idx] == NULL) + goto fail_entry; - idx++; + strcpy((*buf)[idx++], entry); } - - pthread_rwlock_unlock(&ls.db_lock); + no_entries: + pthread_rwlock_unlock(&ls.lock); return idx; + + fail_entry: + while (idx-- > 0) + free((*buf)[idx]); + free(*buf); + fail_entries: + pthread_rwlock_unlock(&ls.lock); + return -ENOMEM; } static struct rib_ops r_ops = { @@ -319,28 +323,28 @@ static int lsdb_add_nb(uint64_t addr, struct list_head * p; struct nb * nb; - pthread_rwlock_wrlock(&ls.db_lock); + pthread_rwlock_wrlock(&ls.lock); - list_for_each(p, &ls.nbs) { + llist_for_each(p, &ls.nbs) { struct nb * el = list_entry(p, struct nb, next); - if (el->addr == addr && el->type == type) { - log_dbg("Already know %s neighbor %" PRIu64 ".", - type == NB_DT ? "dt" : "mgmt", addr); - if (el->fd != fd) { - log_warn("Existing neighbor assigned new fd."); - el->fd = fd; - } - pthread_rwlock_unlock(&ls.db_lock); - return -EPERM; - } - if (addr > el->addr) break; + if (el->addr != addr || el->type != type) + continue; + + log_dbg("Already know %s neighbor " ADDR_FMT32 ".", + type == NB_DT ? "dt" : "mgmt", ADDR_VAL32(&addr)); + if (el->fd != fd) { + log_warn("Existing neighbor assigned new fd."); + el->fd = fd; + } + pthread_rwlock_unlock(&ls.lock); + return -EPERM; } nb = malloc(sizeof(*nb)); if (nb == NULL) { - pthread_rwlock_unlock(&ls.db_lock); + pthread_rwlock_unlock(&ls.lock); return -ENOMEM; } @@ -348,14 +352,12 @@ static int lsdb_add_nb(uint64_t addr, nb->fd = fd; nb->type = type; - list_add_tail(&nb->next, p); + llist_add_tail_at(&nb->next, p, &ls.nbs); - ++ls.nbs_len; + log_dbg("Type %s neighbor " ADDR_FMT32 " added.", + nb->type == NB_DT ? "dt" : "mgmt", ADDR_VAL32(&addr)); - log_dbg("Type %s neighbor %" PRIu64 " added.", - nb->type == NB_DT ? "dt" : "mgmt", addr); - - pthread_rwlock_unlock(&ls.db_lock); + pthread_rwlock_unlock(&ls.lock); return 0; } @@ -366,22 +368,22 @@ static int lsdb_del_nb(uint64_t addr, struct list_head * p; struct list_head * h; - pthread_rwlock_wrlock(&ls.db_lock); + pthread_rwlock_wrlock(&ls.lock); - list_for_each_safe(p, h, &ls.nbs) { + llist_for_each_safe(p, h, &ls.nbs) { struct nb * nb = list_entry(p, struct nb, next); - if (nb->addr == addr && nb->fd == fd) { - list_del(&nb->next); - --ls.nbs_len; - pthread_rwlock_unlock(&ls.db_lock); - log_dbg("Type %s neighbor %" PRIu64 " deleted.", - nb->type == NB_DT ? "dt" : "mgmt", addr); - free(nb); - return 0; - } + if (nb->addr != addr || nb->fd != fd) + continue; + + llist_del(&nb->next, &ls.nbs); + pthread_rwlock_unlock(&ls.lock); + log_dbg("Type %s neighbor " ADDR_FMT32 " deleted.", + nb->type == NB_DT ? "dt" : "mgmt", ADDR_VAL32(&addr)); + free(nb); + return 0; } - pthread_rwlock_unlock(&ls.db_lock); + pthread_rwlock_unlock(&ls.lock); return -EPERM; } @@ -391,18 +393,18 @@ static int nbr_to_fd(uint64_t addr) struct list_head * p; int fd; - pthread_rwlock_rdlock(&ls.db_lock); + pthread_rwlock_rdlock(&ls.lock); - list_for_each(p, &ls.nbs) { + llist_for_each(p, &ls.nbs) { struct nb * nb = list_entry(p, struct nb, next); if (nb->addr == addr && nb->type == NB_DT) { fd = nb->fd; - pthread_rwlock_unlock(&ls.db_lock); + pthread_rwlock_unlock(&ls.lock); return fd; } } - pthread_rwlock_unlock(&ls.db_lock); + pthread_rwlock_unlock(&ls.lock); return -1; } @@ -413,12 +415,11 @@ static void calculate_pff(struct routing_i * instance) struct list_head table; struct list_head * p; struct list_head * q; - int fds[PROG_MAX_FLOWS]; + int fds[PROC_MAX_FLOWS]; assert(instance); - if (graph_routing_table(ls.graph, ls.routing_algo, - ipcpi.dt_addr, &table)) + if (graph_routing_table(ls.graph, ls.routing_algo, ls.addr, &table)) return; pff_lock(instance->pff); @@ -453,8 +454,8 @@ static void set_pff_modified(bool calc) { struct list_head * p; - pthread_mutex_lock(&ls.routing_i_lock); - list_for_each(p, &ls.routing_instances) { + pthread_mutex_lock(&ls.instances.mtx); + list_for_each(p, &ls.instances.list) { struct routing_i * inst = list_entry(p, struct routing_i, next); pthread_mutex_lock(&inst->lock); @@ -463,7 +464,7 @@ static void set_pff_modified(bool calc) if (calc) calculate_pff(inst); } - pthread_mutex_unlock(&ls.routing_i_lock); + pthread_mutex_unlock(&ls.instances.mtx); } static int lsdb_add_link(uint64_t src, @@ -480,9 +481,9 @@ static int lsdb_add_link(uint64_t src, clock_gettime(CLOCK_REALTIME_COARSE, &now); - pthread_rwlock_wrlock(&ls.db_lock); + pthread_rwlock_wrlock(&ls.lock); - list_for_each(p, &ls.db) { + llist_for_each(p, &ls.db) { struct adjacency * a = list_entry(p, struct adjacency, next); if (a->dst == dst && a->src == src) { if (a->seqno < seqno) { @@ -490,7 +491,7 @@ static int lsdb_add_link(uint64_t src, a->seqno = seqno; ret = 0; } - pthread_rwlock_unlock(&ls.db_lock); + pthread_rwlock_unlock(&ls.lock); return ret; } @@ -500,7 +501,7 @@ static int lsdb_add_link(uint64_t src, adj = malloc(sizeof(*adj)); if (adj == NULL) { - pthread_rwlock_unlock(&ls.db_lock); + pthread_rwlock_unlock(&ls.lock); return -ENOMEM; } @@ -509,14 +510,12 @@ static int lsdb_add_link(uint64_t src, adj->seqno = seqno; adj->stamp = now.tv_sec; - list_add_tail(&adj->next, p); - - ls.db_len++; + llist_add_tail_at(&adj->next, p, &ls.db); if (graph_update_edge(ls.graph, src, dst, *qs)) log_warn("Failed to add edge to graph."); - pthread_rwlock_unlock(&ls.db_lock); + pthread_rwlock_unlock(&ls.lock); set_pff_modified(true); @@ -529,25 +528,23 @@ static int lsdb_del_link(uint64_t src, struct list_head * p; struct list_head * h; - pthread_rwlock_wrlock(&ls.db_lock); + pthread_rwlock_wrlock(&ls.lock); - list_for_each_safe(p, h, &ls.db) { + llist_for_each_safe(p, h, &ls.db) { struct adjacency * a = list_entry(p, struct adjacency, next); if (a->dst == dst && a->src == src) { - list_del(&a->next); + llist_del(&a->next, &ls.db); if (graph_del_edge(ls.graph, src, dst)) log_warn("Failed to delete edge from graph."); - ls.db_len--; - - pthread_rwlock_unlock(&ls.db_lock); + pthread_rwlock_unlock(&ls.lock); set_pff_modified(false); free(a); return 0; } } - pthread_rwlock_unlock(&ls.db_lock); + pthread_rwlock_unlock(&ls.lock); return -EPERM; } @@ -570,7 +567,7 @@ static void * periodic_recalc_pff(void * o) if (modified) calculate_pff(inst); - sleep(RECALC_TIME); + sleep(ls.conf.t_recalc); } return (void *) 0; @@ -587,10 +584,20 @@ static void send_lsm(uint64_t src, lsm.s_addr = hton64(src); lsm.seqno = hton64(seqno); - list_for_each(p, &ls.nbs) { + llist_for_each(p, &ls.nbs) { struct nb * nb = list_entry(p, struct nb, next); - if (nb->type == NB_MGMT) - flow_write(nb->fd, &lsm, sizeof(lsm)); + if (nb->type != NB_MGMT) + continue; + + if (flow_write(nb->fd, &lsm, sizeof(lsm)) < 0) + log_err("Failed to send LSM to " ADDR_FMT32, + ADDR_VAL32(&nb->addr)); +#ifdef DEBUG_PROTO_LS + else + log_proto(LSU_FMT " --> " ADDR_FMT32, + LSU_VAL(src, dst, seqno), + ADDR_VAL32(&nb->addr)); +#endif } } @@ -604,9 +611,9 @@ static void lsdb_replicate(int fd) list_head_init(©); /* Lock the lsdb, copy the lsms and send outside of lock. */ - pthread_rwlock_rdlock(&ls.db_lock); + pthread_rwlock_rdlock(&ls.lock); - list_for_each(p, &ls.db) { + llist_for_each(p, &ls.db) { struct adjacency * adj; struct adjacency * cpy; adj = list_entry(p, struct adjacency, next); @@ -623,7 +630,7 @@ static void lsdb_replicate(int fd) list_add_tail(&cpy->next, ©); } - pthread_rwlock_unlock(&ls.db_lock); + pthread_rwlock_unlock(&ls.lock); list_for_each_safe(p, h, ©) { struct lsa lsm; @@ -649,17 +656,17 @@ static void * lsupdate(void * o) while (true) { clock_gettime(CLOCK_REALTIME_COARSE, &now); - pthread_rwlock_wrlock(&ls.db_lock); + pthread_rwlock_wrlock(&ls.lock); - pthread_cleanup_push(__cleanup_rwlock_unlock, &ls.db_lock); + pthread_cleanup_push(__cleanup_rwlock_unlock, &ls.lock); - list_for_each_safe(p, h, &ls.db) { + llist_for_each_safe(p, h, &ls.db) { struct adjacency * adj; adj = list_entry(p, struct adjacency, next); - if (now.tv_sec - adj->stamp > LS_TIMEO) { - list_del(&adj->next); - log_dbg("%" PRIu64 " - %" PRIu64" timed out.", - adj->src, adj->dst); + if (now.tv_sec > adj->stamp + ls.conf.t_timeo) { + llist_del(&adj->next, &ls.db); + log_dbg(LINK_FMT " timed out.", + LINK_VAL(adj->src, adj->dst)); if (graph_del_edge(ls.graph, adj->src, adj->dst)) log_err("Failed to del edge."); @@ -667,7 +674,7 @@ static void * lsupdate(void * o) continue; } - if (adj->src == ipcpi.dt_addr) { + if (adj->src == ls.addr) { adj->seqno++; send_lsm(adj->src, adj->dst, adj->seqno); adj->stamp = now.tv_sec; @@ -676,7 +683,7 @@ static void * lsupdate(void * o) pthread_cleanup_pop(true); - sleep(LS_UPDATE_TIME); + sleep(ls.conf.t_update); } return (void *) 0; @@ -708,15 +715,36 @@ static void forward_lsm(uint8_t * buf, int in_fd) { struct list_head * p; +#ifdef DEBUG_PROTO_LS + struct lsa lsm; + + assert(buf); + assert(len >= sizeof(struct lsa)); + + memcpy(&lsm, buf, sizeof(lsm)); - pthread_rwlock_rdlock(&ls.db_lock); + lsm.s_addr = ntoh64(lsm.s_addr); + lsm.d_addr = ntoh64(lsm.d_addr); + lsm.seqno = ntoh64(lsm.seqno); +#endif + pthread_rwlock_rdlock(&ls.lock); - pthread_cleanup_push(__cleanup_rwlock_unlock, &ls.db_lock); + pthread_cleanup_push(__cleanup_rwlock_unlock, &ls.lock); - list_for_each(p, &ls.nbs) { + llist_for_each(p, &ls.nbs) { struct nb * nb = list_entry(p, struct nb, next); - if (nb->type == NB_MGMT && nb->fd != in_fd) - flow_write(nb->fd, buf, len); + if (nb->type != NB_MGMT || nb->fd == in_fd) + continue; + + if (flow_write(nb->fd, buf, len) < 0) + log_err("Failed to forward LSM to " ADDR_FMT32, + ADDR_VAL32(&nb->addr)); +#ifdef DEBUG_PROTO_LS + else + log_proto(LSU_FMT " --> " ADDR_FMT32 " [forwarded]", + LSU_VAL(lsm.s_addr, lsm.d_addr, lsm.seqno), + ADDR_VAL32(&nb->addr)); +#endif } pthread_cleanup_pop(true); @@ -729,13 +757,13 @@ static void cleanup_fqueue(void * fq) static void * lsreader(void * o) { - fqueue_t * fq; - int ret; - uint8_t buf[sizeof(struct lsa)]; - int fd; - qosspec_t qs; - struct lsa * msg; - size_t len; + fqueue_t * fq; + int ret; + uint8_t buf[sizeof(struct lsa)]; + int fd; + qosspec_t qs; + struct lsa msg; + size_t len; (void) o; @@ -758,15 +786,22 @@ static void * lsreader(void * o) if (fqueue_type(fq) != FLOW_PKT) continue; - len = flow_read(fd, buf, sizeof(*msg)); - if (len <= 0 || len != sizeof(*msg)) + len = flow_read(fd, buf, sizeof(msg)); + if (len <= 0 || len != sizeof(msg)) continue; - msg = (struct lsa *) buf; - - if (lsdb_add_link(ntoh64(msg->s_addr), - ntoh64(msg->d_addr), - ntoh64(msg->seqno), + memcpy(&msg, buf, sizeof(msg)); + msg.s_addr = ntoh64(msg.s_addr); + msg.d_addr = ntoh64(msg.d_addr); + msg.seqno = ntoh64(msg.seqno); +#ifdef DEBUG_PROTO_LS + log_proto(LSU_FMT " <-- " ADDR_FMT32, + LSU_VAL(msg.s_addr, msg.d_addr, msg.seqno), + ADDR_VAL32(&ls.addr)); +#endif + if (lsdb_add_link(msg.s_addr, + msg.d_addr, + msg.seqno, &qs)) continue; @@ -787,14 +822,14 @@ static void flow_event(int fd, log_dbg("Notifying routing instances of flow event."); - pthread_mutex_lock(&ls.routing_i_lock); + pthread_mutex_lock(&ls.instances.mtx); - list_for_each(p, &ls.routing_instances) { + list_for_each(p, &ls.instances.list) { struct routing_i * ri = list_entry(p, struct routing_i, next); pff_flow_state_change(ri->pff, fd, up); } - pthread_mutex_unlock(&ls.routing_i_lock); + pthread_mutex_unlock(&ls.instances.mtx); } static void handle_event(void * self, @@ -816,27 +851,27 @@ static void handle_event(void * self, switch (event) { case NOTIFY_DT_CONN_ADD: - pthread_rwlock_rdlock(&ls.db_lock); + pthread_rwlock_rdlock(&ls.lock); - pthread_cleanup_push(__cleanup_rwlock_unlock, &ls.db_lock); + pthread_cleanup_push(__cleanup_rwlock_unlock, &ls.lock); - send_lsm(ipcpi.dt_addr, c->conn_info.addr, 0); + send_lsm(ls.addr, c->conn_info.addr, 0); pthread_cleanup_pop(true); if (lsdb_add_nb(c->conn_info.addr, c->flow_info.fd, NB_DT)) - log_dbg("Failed to add neighbor to LSDB."); + log_dbg("Failed to add neighbor to lsdb."); - if (lsdb_add_link(ipcpi.dt_addr, c->conn_info.addr, 0, &qs)) - log_dbg("Failed to add new adjacency to LSDB."); + if (lsdb_add_link(ls.addr, c->conn_info.addr, 0, &qs)) + log_dbg("Failed to add new adjacency to lsdb."); break; case NOTIFY_DT_CONN_DEL: flow_event(c->flow_info.fd, false); if (lsdb_del_nb(c->conn_info.addr, c->flow_info.fd)) - log_dbg("Failed to delete neighbor from LSDB."); + log_dbg("Failed to delete neighbor from lsdb."); - if (lsdb_del_link(ipcpi.dt_addr, c->conn_info.addr)) - log_dbg("Local link was not in LSDB."); + if (lsdb_del_link(ls.addr, c->conn_info.addr)) + log_dbg("Local link was not in lsdb."); break; case NOTIFY_DT_CONN_QOS: log_dbg("QoS changes currently unsupported."); @@ -852,14 +887,14 @@ static void handle_event(void * self, fccntl(c->flow_info.fd, FLOWSFLAGS, flags | FLOWFRNOPART); fset_add(ls.mgmt_set, c->flow_info.fd); if (lsdb_add_nb(c->conn_info.addr, c->flow_info.fd, NB_MGMT)) - log_warn("Failed to add mgmt neighbor to LSDB."); + log_warn("Failed to add mgmt neighbor to lsdb."); /* replicate the entire lsdb */ lsdb_replicate(c->flow_info.fd); break; case NOTIFY_MGMT_CONN_DEL: fset_del(ls.mgmt_set, c->flow_info.fd); if (lsdb_del_nb(c->conn_info.addr, c->flow_info.fd)) - log_warn("Failed to delete mgmt neighbor from LSDB."); + log_warn("Failed to delete mgmt neighbor from lsdb."); break; default: break; @@ -886,11 +921,11 @@ struct routing_i * link_state_routing_i_create(struct pff * pff) periodic_recalc_pff, tmp)) goto fail_pthread_create_lsupdate; - pthread_mutex_lock(&ls.routing_i_lock); + pthread_mutex_lock(&ls.instances.mtx); - list_add(&tmp->next, &ls.routing_instances); + list_add(&tmp->next, &ls.instances.list); - pthread_mutex_unlock(&ls.routing_i_lock); + pthread_mutex_unlock(&ls.instances.mtx); return tmp; @@ -906,11 +941,11 @@ void link_state_routing_i_destroy(struct routing_i * instance) { assert(instance); - pthread_mutex_lock(&ls.routing_i_lock); + pthread_mutex_lock(&ls.instances.mtx); list_del(&instance->next); - pthread_mutex_unlock(&ls.routing_i_lock); + pthread_mutex_unlock(&ls.instances.mtx); pthread_cancel(instance->calculator); @@ -921,96 +956,143 @@ void link_state_routing_i_destroy(struct routing_i * instance) free(instance); } -int link_state_init(enum pol_routing pr) +int link_state_start(void) +{ + if (notifier_reg(handle_event, NULL)) { + log_err("Failed to register link-state with notifier."); + goto fail_notifier_reg; + } + + if (pthread_create(&ls.lsupdate, NULL, lsupdate, NULL)) { + log_err("Failed to create lsupdate thread."); + goto fail_pthread_create_lsupdate; + } + + if (pthread_create(&ls.lsreader, NULL, lsreader, NULL)) { + log_err("Failed to create lsreader thread."); + goto fail_pthread_create_lsreader; + } + + if (pthread_create(&ls.listener, NULL, ls_conn_handle, NULL)) { + log_err("Failed to create listener thread."); + goto fail_pthread_create_listener; + } + + return 0; + + fail_pthread_create_listener: + pthread_cancel(ls.lsreader); + pthread_join(ls.lsreader, NULL); + fail_pthread_create_lsreader: + pthread_cancel(ls.lsupdate); + pthread_join(ls.lsupdate, NULL); + fail_pthread_create_lsupdate: + notifier_unreg(handle_event); + fail_notifier_reg: + return -1; +} + +void link_state_stop(void) +{ + pthread_cancel(ls.listener); + pthread_cancel(ls.lsreader); + pthread_cancel(ls.lsupdate); + + pthread_join(ls.listener, NULL); + pthread_join(ls.lsreader, NULL); + pthread_join(ls.lsupdate, NULL); + + notifier_unreg(handle_event); +} + + +int link_state_init(struct ls_config * conf, + enum pol_pff * pff_type) { struct conn_info info; + assert(conf != NULL); + assert(pff_type != NULL); + memset(&info, 0, sizeof(info)); + ls.addr = addr_auth_address(); + strcpy(info.comp_name, LS_COMP); strcpy(info.protocol, LS_PROTO); info.pref_version = 1; info.pref_syntax = PROTO_GPB; - info.addr = ipcpi.dt_addr; + info.addr = ls.addr; - switch (pr) { - case ROUTING_LINK_STATE: - log_dbg("Using link state routing policy."); + ls.conf = *conf; + + switch (conf->pol) { + case LS_SIMPLE: + *pff_type = PFF_SIMPLE; ls.routing_algo = ROUTING_SIMPLE; + log_dbg("Using Link State Routing policy."); break; - case ROUTING_LINK_STATE_LFA: - log_dbg("Using Loop-Free Alternates policy."); + case LS_LFA: ls.routing_algo = ROUTING_LFA; + *pff_type = PFF_ALTERNATE; + log_dbg("Using Loop-Free Alternates policy."); break; - case ROUTING_LINK_STATE_ECMP: - log_dbg("Using Equal-Cost Multipath policy."); + case LS_ECMP: ls.routing_algo = ROUTING_ECMP; + *pff_type = PFF_MULTIPATH; + log_dbg("Using Equal-Cost Multipath policy."); break; default: goto fail_graph; } + log_dbg("LS update interval: %ld seconds.", ls.conf.t_update); + log_dbg("LS link timeout : %ld seconds.", ls.conf.t_timeo); + log_dbg("LS recalc interval: %ld seconds.", ls.conf.t_recalc); + ls.graph = graph_create(); if (ls.graph == NULL) goto fail_graph; - if (notifier_reg(handle_event, NULL)) - goto fail_notifier_reg; - - if (pthread_rwlock_init(&ls.db_lock, NULL)) - goto fail_db_lock_init; + if (pthread_rwlock_init(&ls.lock, NULL)) { + log_err("Failed to init lock."); + goto fail_lock_init; + } - if (pthread_mutex_init(&ls.routing_i_lock, NULL)) + if (pthread_mutex_init(&ls.instances.mtx, NULL)) { + log_err("Failed to init instances mutex."); goto fail_routing_i_lock_init; + } - if (connmgr_comp_init(COMPID_MGMT, &info)) + if (connmgr_comp_init(COMPID_MGMT, &info)) { + log_err("Failed to init connmgr."); goto fail_connmgr_comp_init; + } ls.mgmt_set = fset_create(); - if (ls.mgmt_set == NULL) + if (ls.mgmt_set == NULL) { + log_err("Failed to create fset."); goto fail_fset_create; + } - list_head_init(&ls.db); - list_head_init(&ls.nbs); - list_head_init(&ls.routing_instances); - - if (pthread_create(&ls.lsupdate, NULL, lsupdate, NULL)) - goto fail_pthread_create_lsupdate; - - if (pthread_create(&ls.lsreader, NULL, lsreader, NULL)) - goto fail_pthread_create_lsreader; - - if (pthread_create(&ls.listener, NULL, ls_conn_handle, NULL)) - goto fail_pthread_create_listener; + llist_init(&ls.db); + llist_init(&ls.nbs); + list_head_init(&ls.instances.list); - if (rib_reg(LSDB, &r_ops)) + if (rib_reg(lsdb, &r_ops)) goto fail_rib_reg; - ls.db_len = 0; - ls.nbs_len = 0; - return 0; fail_rib_reg: - pthread_cancel(ls.listener); - pthread_join(ls.listener, NULL); - fail_pthread_create_listener: - pthread_cancel(ls.lsreader); - pthread_join(ls.lsreader, NULL); - fail_pthread_create_lsreader: - pthread_cancel(ls.lsupdate); - pthread_join(ls.lsupdate, NULL); - fail_pthread_create_lsupdate: fset_destroy(ls.mgmt_set); fail_fset_create: connmgr_comp_fini(COMPID_MGMT); fail_connmgr_comp_init: - pthread_mutex_destroy(&ls.routing_i_lock); + pthread_mutex_destroy(&ls.instances.mtx); fail_routing_i_lock_init: - pthread_rwlock_destroy(&ls.db_lock); - fail_db_lock_init: - notifier_unreg(handle_event); - fail_notifier_reg: + pthread_rwlock_destroy(&ls.lock); + fail_lock_init: graph_destroy(ls.graph); fail_graph: return -1; @@ -1021,17 +1103,7 @@ void link_state_fini(void) struct list_head * p; struct list_head * h; - rib_unreg(LSDB); - - notifier_unreg(handle_event); - - pthread_cancel(ls.listener); - pthread_cancel(ls.lsreader); - pthread_cancel(ls.lsupdate); - - pthread_join(ls.listener, NULL); - pthread_join(ls.lsreader, NULL); - pthread_join(ls.lsupdate, NULL); + rib_unreg(lsdb); fset_destroy(ls.mgmt_set); @@ -1039,17 +1111,17 @@ void link_state_fini(void) graph_destroy(ls.graph); - pthread_rwlock_wrlock(&ls.db_lock); + pthread_rwlock_wrlock(&ls.lock); - list_for_each_safe(p, h, &ls.db) { + llist_for_each_safe(p, h, &ls.db) { struct adjacency * a = list_entry(p, struct adjacency, next); - list_del(&a->next); + llist_del(&a->next, &ls.db); free(a); } - pthread_rwlock_unlock(&ls.db_lock); + pthread_rwlock_unlock(&ls.lock); - pthread_rwlock_destroy(&ls.db_lock); + pthread_rwlock_destroy(&ls.lock); - pthread_mutex_destroy(&ls.routing_i_lock); + pthread_mutex_destroy(&ls.instances.mtx); } diff --git a/src/ipcpd/unicast/routing/link-state.h b/src/ipcpd/unicast/routing/link-state.h index d77d72df..38e19065 100644 --- a/src/ipcpd/unicast/routing/link-state.h +++ b/src/ipcpd/unicast/routing/link-state.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Link state routing policy * @@ -28,10 +28,15 @@ #include "ops.h" -int link_state_init(enum pol_routing pr); +int link_state_init(struct ls_config * ls, + enum pol_pff * pff_type); void link_state_fini(void); +int link_state_start(void); + +void link_state_stop(void); + struct routing_i * link_state_routing_i_create(struct pff * pff); void link_state_routing_i_destroy(struct routing_i * instance); diff --git a/src/ipcpd/unicast/routing/ops.h b/src/ipcpd/unicast/routing/ops.h index 8a79b7ec..b19c5176 100644 --- a/src/ipcpd/unicast/routing/ops.h +++ b/src/ipcpd/unicast/routing/ops.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Routing policy ops * @@ -26,10 +26,15 @@ #include "pff.h" struct routing_ops { - int (* init)(enum pol_routing pr); + int (* init)(void * conf, + enum pol_pff * pff_type); void (* fini)(void); + int (* start)(void); + + void (* stop)(void); + struct routing_i * (* routing_i_create)(struct pff * pff); void (* routing_i_destroy)(struct routing_i * instance); diff --git a/src/ipcpd/unicast/routing/pol.h b/src/ipcpd/unicast/routing/pol.h index b6a6f150..545f5df2 100644 --- a/src/ipcpd/unicast/routing/pol.h +++ b/src/ipcpd/unicast/routing/pol.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Routing policies * diff --git a/src/ipcpd/unicast/routing/tests/CMakeLists.txt b/src/ipcpd/unicast/routing/tests/CMakeLists.txt index d0652533..be2de72c 100644 --- a/src/ipcpd/unicast/routing/tests/CMakeLists.txt +++ b/src/ipcpd/unicast/routing/tests/CMakeLists.txt @@ -3,32 +3,32 @@ get_filename_component(CURRENT_SOURCE_PARENT_DIR get_filename_component(CURRENT_BINARY_PARENT_DIR ${CMAKE_CURRENT_BINARY_DIR} DIRECTORY) -include_directories(${CMAKE_CURRENT_SOURCE_DIR}) -include_directories(${CMAKE_CURRENT_BINARY_DIR}) - -include_directories(${CURRENT_SOURCE_PARENT_DIR}) -include_directories(${CURRENT_BINARY_PARENT_DIR}) - -include_directories(${CMAKE_SOURCE_DIR}/include) -include_directories(${CMAKE_BINARY_DIR}/include) - get_filename_component(PARENT_PATH ${CMAKE_CURRENT_SOURCE_DIR} DIRECTORY) get_filename_component(PARENT_DIR ${PARENT_PATH} NAME) +compute_test_prefix() + create_test_sourcelist(${PARENT_DIR}_tests test_suite.c # Add new tests here graph_test.c ) -add_executable(${PARENT_DIR}_test EXCLUDE_FROM_ALL ${${PARENT_DIR}_tests}) -target_link_libraries(${PARENT_DIR}_test ouroboros-common) +add_executable(${PARENT_DIR}_test ${${PARENT_DIR}_tests}) + +target_include_directories(${PARENT_DIR}_test PRIVATE + ${CMAKE_CURRENT_SOURCE_DIR} + ${CMAKE_CURRENT_BINARY_DIR} + ${CURRENT_SOURCE_PARENT_DIR} + ${CURRENT_BINARY_PARENT_DIR} + ${CMAKE_SOURCE_DIR}/include + ${CMAKE_BINARY_DIR}/include + ${CMAKE_SOURCE_DIR}/src/ipcpd + ${CMAKE_BINARY_DIR}/src/ipcpd +) -add_dependencies(check ${PARENT_DIR}_test) +disable_test_logging_for_target(${PARENT_DIR}_test) +target_link_libraries(${PARENT_DIR}_test PRIVATE ouroboros-common) -set(tests_to_run ${${PARENT_DIR}_tests}) -remove(tests_to_run test_suite.c) +add_dependencies(build_tests ${PARENT_DIR}_test) -foreach (test ${tests_to_run}) - get_filename_component(test_name ${test} NAME_WE) - add_test(${test_name} ${C_TEST_PATH}/${PARENT_DIR}_test ${test_name}) -endforeach (test) +ouroboros_register_tests(TARGET ${PARENT_DIR}_test TESTS ${${PARENT_DIR}_tests}) diff --git a/src/ipcpd/unicast/routing/tests/graph_test.c b/src/ipcpd/unicast/routing/tests/graph_test.c index d805640c..40a744ff 100644 --- a/src/ipcpd/unicast/routing/tests/graph_test.c +++ b/src/ipcpd/unicast/routing/tests/graph_test.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Test of the graph structure * diff --git a/src/irmd/CMakeLists.txt b/src/irmd/CMakeLists.txt index c9c2e553..5aa457ff 100644 --- a/src/irmd/CMakeLists.txt +++ b/src/irmd/CMakeLists.txt @@ -1,95 +1,67 @@ -include_directories(${CMAKE_CURRENT_SOURCE_DIR}) -include_directories(${CMAKE_CURRENT_BINARY_DIR}) +# IRMd (IPC Resource Manager daemon) build configuration +# Configuration options are in cmake/config/global.cmake and cmake/config/irmd.cmake -include_directories(${CMAKE_SOURCE_DIR}/include) -include_directories(${CMAKE_BINARY_DIR}/include) - -find_library(LIBTOML_LIBRARIES toml QUIET) -if (LIBTOML_LIBRARIES) - set(DISABLE_CONFIGFILE FALSE CACHE BOOL - "Disable configuration file support") - if (NOT DISABLE_CONFIGFILE) - set(OUROBOROS_CONFIG_DIR /etc/ouroboros/ CACHE STRING - "Configuration directory") - set(OUROBOROS_CONFIG_FILE irmd.conf CACHE STRING - "Name of the IRMd configuration file") - set(HAVE_TOML TRUE) - message(STATUS "Found TOML C99 library: " ${LIBTOML_LIBRARIES}) - message(STATUS "Configuration file support enabled") - message(STATUS "Configuration directory: ${OUROBOROS_CONFIG_DIR}") - set(INSTALL_DIR "${CMAKE_INSTALL_PREFIX}/${CMAKE_INSTALL_BINDIR}") - configure_file("${CMAKE_SOURCE_DIR}/irmd.conf.in" - "${CMAKE_BINARY_DIR}/irmd.conf.example" @ONLY) - install(FILES "${CMAKE_BINARY_DIR}/irmd.conf.example" - DESTINATION "${OUROBOROS_CONFIG_DIR}") - unset(INSTALL_DIR) - mark_as_advanced(LIBTOML_LIBRARIES) - else () - message(STATUS "Configuration file support disabled by user") - unset(OUROBOROS_CONFIG_FILE CACHE) - unset(OUROBOROS_CONFIG_DIR CACHE) - set(HAVE_TOML FALSE) - endif () -else () - message(STATUS "Install tomlc99 for config file support") - message(STATUS " https://github.com/cktan/tomlc99") - set(LIBTOML_LIBRARIES "") - unset(DISABLE_CONFIGFILE CACHE) - unset(HAVE_TOML) -endif () - -set(IRMD_REQ_ARR_TIMEOUT 1000 CACHE STRING - "Timeout for an application to respond to a new flow (ms)") - -set(BOOTSTRAP_TIMEOUT 5000 CACHE STRING - "Timeout for an IPCP to bootstrap (ms)") -set(ENROLL_TIMEOUT 20000 CACHE STRING - "Timeout for an IPCP to enroll (ms)") -set(REG_TIMEOUT 20000 CACHE STRING - "Timeout for registering a name (ms)") -set(QUERY_TIMEOUT 20000 CACHE STRING - "Timeout to query a name with an IPCP (ms)") -set(CONNECT_TIMEOUT 20000 CACHE STRING - "Timeout to connect an IPCP to another IPCP (ms)") -set(FLOW_ALLOC_TIMEOUT 20000 CACHE STRING - "Timeout for a flow allocation response (ms)") -set(IRMD_MIN_THREADS 8 CACHE STRING - "Minimum number of worker threads in the IRMd") -set(IRMD_ADD_THREADS 8 CACHE STRING - "Number of extra threads to start when the IRMD faces thread starvation") -set(IRMD_PKILL_TIMEOUT 30 CACHE STRING - "Number of seconds to wait before sending SIGKILL to subprocesses on exit") -set(IRMD_KILL_ALL_PROCESSES TRUE CACHE BOOL - "Kill all processes on exit") +# Generate and install configuration files if TOML support available +# HAVE_TOML is set in cmake/dependencies/irmd/libtoml.cmake +if(HAVE_TOML) + set(INSTALL_DIR "${CMAKE_INSTALL_PREFIX}/${CMAKE_INSTALL_BINDIR}") + configure_file("${CMAKE_SOURCE_DIR}/irmd.conf.in" + "${CMAKE_BINARY_DIR}/${OUROBOROS_CONFIG_FILE}.example" @ONLY) + configure_file("${CMAKE_SOURCE_DIR}/sec.conf.in" + "${CMAKE_BINARY_DIR}/sec.conf.example" @ONLY) + install(FILES "${CMAKE_BINARY_DIR}/${OUROBOROS_CONFIG_FILE}.example" + DESTINATION "${OUROBOROS_CONFIG_DIR}") + install(FILES "${CMAKE_BINARY_DIR}/sec.conf.example" + DESTINATION "${OUROBOROS_CONFIG_DIR}") + install(CODE " + if(NOT EXISTS \"${OUROBOROS_CONFIG_DIR}/${OUROBOROS_CONFIG_FILE}\") + file(WRITE \"${OUROBOROS_CONFIG_DIR}/${OUROBOROS_CONFIG_FILE}\" \"\") + endif() + ") + unset(INSTALL_DIR) +endif() configure_file("${CMAKE_CURRENT_SOURCE_DIR}/config.h.in" "${CMAKE_CURRENT_BINARY_DIR}/config.h" @ONLY) -set(SOURCE_FILES - # Add source files here +set(IRMD_SOURCES ipcp.c configfile.c main.c + oap/io.c + oap/hdr.c + oap/auth.c + oap/srv.c + oap/cli.c reg/flow.c reg/ipcp.c + reg/pool.c reg/proc.c reg/prog.c reg/name.c reg/reg.c - ) +) + +add_executable(irmd ${IRMD_SOURCES}) + +add_dependencies(irmd version_header) -add_executable (irmd ${SOURCE_FILES}) +target_include_directories(irmd PRIVATE + ${CMAKE_CURRENT_SOURCE_DIR} + ${CMAKE_CURRENT_BINARY_DIR} + ${CMAKE_SOURCE_DIR}/include + ${CMAKE_BINARY_DIR}/include) -target_link_libraries (irmd LINK_PUBLIC ouroboros-common - ${LIBTOML_LIBRARIES}) +target_link_libraries(irmd PRIVATE ouroboros-common) +if(HAVE_TOML) + target_link_libraries(irmd PRIVATE toml::toml) +endif() -include(AddCompileFlags) -if (CMAKE_BUILD_TYPE MATCHES "Debug*") - add_compile_flags(irmd -DCONFIG_OUROBOROS_DEBUG) -endif () +ouroboros_target_debug_definitions(irmd) install(TARGETS irmd RUNTIME DESTINATION ${CMAKE_INSTALL_SBINDIR}) -# Enable once irmd has tests -#add_subdirectory(tests) -add_subdirectory(reg) +if(BUILD_TESTS) + add_subdirectory(oap/tests) + add_subdirectory(reg/tests) +endif() diff --git a/src/irmd/config.h.in b/src/irmd/config.h.in index fa1156b9..e14cff75 100644 --- a/src/irmd/config.h.in +++ b/src/irmd/config.h.in @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Configuration for the IPC Resource Manager * @@ -21,7 +21,8 @@ */ -#define IPCP_UDP_EXEC "@IPCP_UDP_TARGET@" +#define IPCP_UDP4_EXEC "@IPCP_UDP4_TARGET@" +#define IPCP_UDP6_EXEC "@IPCP_UDP6_TARGET@" #define IPCP_ETH_LLC_EXEC "@IPCP_ETH_LLC_TARGET@" #define IPCP_ETH_DIX_EXEC "@IPCP_ETH_DIX_TARGET@" #define IPCP_UNICAST_EXEC "@IPCP_UNICAST_TARGET@" @@ -40,6 +41,11 @@ #define FLOW_ALLOC_TIMEOUT @FLOW_ALLOC_TIMEOUT@ #define FLOW_DEALLOC_TIMEOUT @FLOW_DEALLOC_TIMEOUT@ +#define OAP_REPLAY_TIMER @OAP_REPLAY_TIMER@ +#define OAP_REPLAY_MAX @OAP_REPLAY_MAX@ +#define OAP_REKEY_TIMER @OAP_REKEY_TIMER@ +#cmakedefine01 OAP_CLIENT_AUTH_DEFAULT + #define BOOTSTRAP_TIMEOUT @BOOTSTRAP_TIMEOUT@ #define ENROLL_TIMEOUT @ENROLL_TIMEOUT@ #define REG_TIMEOUT @REG_TIMEOUT@ @@ -50,6 +56,7 @@ #define IRMD_MIN_THREADS @IRMD_MIN_THREADS@ #define IRMD_ADD_THREADS @IRMD_ADD_THREADS@ +#define SSM_PID_GSPP 0 #cmakedefine HAVE_FUSE #ifdef HAVE_FUSE @@ -62,36 +69,39 @@ #define OUROBOROS_CONFIG_FILE "@OUROBOROS_CONFIG_FILE@" #endif +#define OUROBOROS_SECURITY_DIR "@OUROBOROS_SECURITY_DIR@" +#define OUROBOROS_CA_CRT_DIR "@OUROBOROS_CA_CRT_DIR@" +#define OUROBOROS_SRV_CRT_DIR "@OUROBOROS_SRV_CRT_DIR@" +#define OUROBOROS_CLI_CRT_DIR "@OUROBOROS_CLI_CRT_DIR@" +#define OUROBOROS_CHAIN_DIR "@OUROBOROS_UNTRUSTED_DIR@" + #define IRMD_PKILL_TIMEOUT @IRMD_PKILL_TIMEOUT@ +#cmakedefine DISABLE_DIRECT_IPC #cmakedefine IRMD_KILL_ALL_PROCESSES #cmakedefine HAVE_LIBGCRYPT +#cmakedefine HAVE_OPENSSL +#ifdef HAVE_OPENSSL +#cmakedefine HAVE_OPENSSL_ML_KEM +#cmakedefine HAVE_OPENSSL_ML_DSA +#endif +#define IRMD_SECMEM_MAX @IRMD_SECMEM_MAX@ +#ifdef CONFIG_OUROBOROS_DEBUG +#cmakedefine DEBUG_PROTO_OAP +#endif + +#define _B "[38;5;4m" +#define _G "[38;5;8m" +#define RST "[0m" -#define O7S_ASCII_ART \ -"[0m\n" \ -" [38;5;4m▄▄█████▄▄▄[38;5;7m\n" \ -" [38;5;4m▄█▀▀[38;5;7m [38;5;4m▀▀███▄[38;5;7m " \ -"[38;5;8m█[38;5;7m\n" \ -" [38;5;4m██[38;5;7m [38;5;4m▄▄▄[38;5;7m [38;5;4m▄███▄[38;5;7m " \ -"[38;5;8m▄[38;5;7m [38;5;8m▄[38;5;7m [38;5;8m▄[38;5;7m [38;5;8m▄▄" \ -"[38;5;7m [38;5;8m▄▄[38;5;7m [38;5;8m█[38;5;7m [38;5;8m▄▄[38;5;7m " \ -" [38;5;8m▄▄[38;5;7m [38;5;8m▄[38;5;7m [38;5;8m▄▄[38;5;7m [38;5;8m" \ -"▄▄[38;5;7m [38;5;8m▄▄[38;5;7m\n" \ -" [38;5;4m██[38;5;7m [38;5;4m█[38;5;7m [38;5;4m█[38;5;7m " \ -"[38;5;4m█████[38;5;7m [38;5;8m█[38;5;7m [38;5;8m█[38;5;7m " \ -"[38;5;8m█▀[38;5;7m [38;5;8m▀[38;5;7m [38;5;8m█[38;5;7m [38;5;8m█" \ -"[38;5;7m [38;5;8m█▀[38;5;7m [38;5;8m█[38;5;7m [38;5;8m█[38;5;7m " \ -"[38;5;8m█[38;5;7m [38;5;8m█▀[38;5;7m [38;5;8m▀[38;5;7m [38;5;8m█" \ -"[38;5;7m [38;5;8m█[38;5;7m [38;5;8m▀▄[38;5;7m [38;5;8m▀[38;5;7m\n" \ -" [38;5;4m██[38;5;7m [38;5;4m▀▄▄▄▀[38;5;7m [38;5;4m▀█▀[38;5;7m " \ -"[38;5;8m█[38;5;7m [38;5;8m█[38;5;7m [38;5;8m█[38;5;7m [38;5;8m" \ -"█[38;5;7m [38;5;8m█[38;5;7m [38;5;8m█▄[38;5;7m [38;5;8m█[38;5;7m " \ -"[38;5;8m█[38;5;7m [38;5;8m█[38;5;7m [38;5;8m█[38;5;7m [38;5;8m█" \ -"[38;5;7m [38;5;8m█[38;5;7m [38;5;8m▄[38;5;7m [38;5;8m▀▄[38;5;7m\n" \ -" [38;5;4m█▄[38;5;7m [38;5;4m█[38;5;7m [38;5;8m▀▀▀" \ -"[38;5;7m [38;5;8m▀[38;5;7m [38;5;8m▀[38;5;7m [38;5;8m▀▀[38;5;7m" \ -" [38;5;8m▀[38;5;7m [38;5;8m▀▀[38;5;7m [38;5;8m▀▀[38;5;7m " \ -"[38;5;8m▀[38;5;7m [38;5;8m▀▀[38;5;7m [38;5;8m▀▀[38;5;7m\n" \ -" [38;5;4m▀█▄▄▄▄▄▄▄▄▀[38;5;7m\n" \ -" [38;5;4m▀▀▀▀▀▀[38;5;7m\n" \ -"\n" +#define O7S_ASCII_ART \ +RST "\n" \ +_B " ▄▄█████▄▄▄ \n" \ +_B " ▄█▀▀ ▀▀███▄ " _G " █ \n" \ +_B " ██ ▄▄▄ ▄███▄ " _G "▄ ▄ ▄ ▄ ▄▄ █▄▄ ▄▄ ▄ ▄ ▄▄ ▄▄ \n" \ +_B " ██ █ █ █████ " _G "█ █ █▀ ▀ █ █ █ █ █ █ █▀ ▀ █ █ ▀▄ ▀\n" \ +_B " ██ ▀▄▄▄▀ ▀█▀ " _G "█ █ █ █ █ █ █ █ █ █ █ █ ▄ ▀▄\n" \ +_B " █▄ █ " _G " ▀▀ ▀ ▀ ▀▀ ▀▀▀ ▀▀ ▀ ▀▀ ▀▀ \n" \ +_B " ▀█▄▄▄▄▄▄▄▄▀ \n" \ +_B " ▀▀▀▀▀▀ \n" \ +RST "\n" diff --git a/src/irmd/configfile.c b/src/irmd/configfile.c index 688c4ade..35cf4292 100644 --- a/src/irmd/configfile.c +++ b/src/irmd/configfile.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * The IPC Resource Manager / Configuration from file * @@ -47,8 +47,12 @@ #include <string.h> #include <toml.h> #include <arpa/inet.h> +#ifdef __FreeBSD__ +#include <sys/socket.h> +#endif #define ERRBUFSZ 200 +#define DATUMSZ 256 static int toml_hash(toml_table_t * table, struct layer_info * info) @@ -157,14 +161,66 @@ static int toml_eth_dix(toml_table_t * table, return 0; } -static int toml_udp(toml_table_t * table, - struct ipcp_config * conf) +static int toml_udp4(toml_table_t * table, + struct ipcp_config * conf) +{ + struct udp4_config * udp4; + toml_datum_t ip; + toml_datum_t port; + toml_datum_t dns; + + *conf = udp4_default_conf; + udp4 = &conf->udp4; + + ip = toml_string_in(table, "ip"); + if (!ip.ok) { + log_err("No IP address specified!"); + goto fail_ip; + } + + if (inet_pton (AF_INET, ip.u.s, &udp4->ip_addr.s_addr) != 1) { + log_err("Failed to parse IPv4 address %s.", ip.u.s); + goto fail_addr; + } + + port = toml_int_in(table, "port"); + if (port.ok) + udp4->port = port.u.i; + + dns = toml_string_in(table, "dns"); + if (dns.ok) { + if (inet_pton(AF_INET, dns.u.s, &udp4->dns_addr.s_addr) < 0) { + log_err("Failed to parse DNS address %s.", ip.u.s); + goto fail_dns; + } + + free(dns.u.s); + } + + free(ip.u.s); + + return 0; + + fail_dns: + free(dns.u.s); + fail_addr: + free(ip.u.s); + fail_ip: + return -1; +} + +static int toml_udp6(toml_table_t * table, + struct ipcp_config * conf) { + struct in6_addr ip6; + struct in6_addr dns6; toml_datum_t ip; toml_datum_t port; toml_datum_t dns; - *conf = udp_default_conf; + *conf = udp6_default_conf; + ip6 = conf->udp6.ip_addr; + dns6 = conf->udp6.dns_addr; ip = toml_string_in(table, "ip"); if (!ip.ok) { @@ -172,18 +228,18 @@ static int toml_udp(toml_table_t * table, goto fail_ip; } - if (inet_pton (AF_INET, ip.u.s, &conf->udp.ip_addr) != 1) { + if (inet_pton (AF_INET6, ip.u.s, &ip6.s6_addr) != 1) { log_err("Failed to parse IPv4 address %s.", ip.u.s); goto fail_addr; } port = toml_int_in(table, "port"); if (port.ok) - conf->udp.port = port.u.i; + conf->udp6.port = port.u.i; dns = toml_string_in(table, "dns"); if (dns.ok) { - if (inet_pton(AF_INET, dns.u.s, &conf->udp.dns_addr) < 0) { + if (inet_pton(AF_INET6, dns.u.s, &dns6.s6_addr) < 0) { log_err("Failed to parse DNS address %s.", ip.u.s); goto fail_dns; } @@ -214,26 +270,160 @@ static int toml_broadcast(toml_table_t * table, return 0; } +#define BETWEEN(a, b, c) ((a) >= (b) && (a) <= (c)) +#define DHT(conf, x) (conf)->dht.params.x +static int toml_dir(toml_table_t * table, + struct dir_config * conf) +{ + toml_datum_t dir; + toml_datum_t alpha; + toml_datum_t t_expire; + toml_datum_t t_refresh; + toml_datum_t t_replicate; + toml_datum_t k; + + dir = toml_string_in(table, "directory"); + if (dir.ok) { + log_dbg("Found directory type: %s", dir.u.s); + if (strlen(dir.u.s) > DATUMSZ) { + log_err("Directory name too long: %s", dir.u.s); + free(dir.u.s); + return -1; + } + if (strcmp(dir.u.s, "DHT") == 0) + conf->pol = DIR_DHT; + else if (strcmp(dir.u.s, "dht") == 0) + conf->pol = DIR_DHT; + else { + log_err("Unknown directory type: %s", dir.u.s); + free(dir.u.s); + return -EINVAL; + } + free(dir.u.s); + } + + switch(conf->pol) { + case DIR_DHT: + log_info("Using DHT directory policy."); + alpha = toml_int_in(table, "dht_alpha"); + if (alpha.ok) { + if (!BETWEEN(alpha.u.i, + DHT_ALPHA_MIN, DHT_ALPHA_MAX)) { + log_err("Invalid alpha value: %ld", + (long) alpha.u.i); + return -EINVAL; + } + DHT(conf, alpha) = alpha.u.i; + } + t_expire = toml_int_in(table, "dht_t_expire"); + if (t_expire.ok) { + if (!BETWEEN(t_expire.u.i, + DHT_T_EXPIRE_MIN, DHT_T_EXPIRE_MAX)) { + log_err("Invalid expire time: %ld", + (long) t_expire.u.i); + return -EINVAL; + } + DHT(conf, t_expire) = t_expire.u.i; + } + t_refresh = toml_int_in(table, "dht_t_refresh"); + if (t_refresh.ok) { + if (!BETWEEN(t_refresh.u.i, + DHT_T_REFRESH_MIN, DHT_T_REFRESH_MAX)) { + log_err("Invalid refresh time: %ld", + (long) t_refresh.u.i); + return -EINVAL; + } + DHT(conf, t_refresh) = t_refresh.u.i; + } + t_replicate = toml_int_in(table, "dht_t_replicate"); + if (t_replicate.ok) { + if (!BETWEEN(t_replicate.u.i, + DHT_T_REPLICATE_MIN, DHT_T_REPLICATE_MAX)) { + log_err("Invalid replication time: %ld", + (long) t_replicate.u.i); + return -EINVAL; + } + DHT(conf, t_replicate) = t_replicate.u.i; + } + k = toml_int_in(table, "dht_k"); + if (k.ok) { + if (!BETWEEN(k.u.i, DHT_K_MIN, DHT_K_MAX)) { + log_err("Invalid replication factor: %ld", + (long) k.u.i); + return -EINVAL; + } + DHT(conf, k) = k.u.i; + } + break; + default: + assert(false); + break; + } + + return 0; +} + static int toml_routing(toml_table_t * table, struct dt_config * conf) { toml_datum_t routing; + toml_datum_t t_recalc; + toml_datum_t t_update; + toml_datum_t t_timeo; routing = toml_string_in(table, "routing"); if (routing.ok) { - if (strcmp(routing.u.s, "link-state") == 0) - conf->routing_type = ROUTING_LINK_STATE; - else if (strcmp(routing.u.s, "lfa") == 0) - conf->routing_type = ROUTING_LINK_STATE_LFA; - else if (strcmp(routing.u.s, "ecmp") == 0) - conf->routing_type = ROUTING_LINK_STATE_ECMP; - else - conf->routing_type = ROUTING_INVALID; + if (strcmp(routing.u.s, "link-state") == 0) { + conf->routing.pol = ROUTING_LINK_STATE; + conf->routing.ls.pol = LS_SIMPLE; + } else if (strcmp(routing.u.s, "lfa") == 0) { + conf->routing.pol = ROUTING_LINK_STATE; + conf->routing.ls.pol = LS_LFA; + } else if (strcmp(routing.u.s, "ecmp") == 0) { + conf->routing.pol = ROUTING_LINK_STATE; + conf->routing.ls.pol = LS_ECMP; + } else { + conf->routing.pol = ROUTING_INVALID; + return -EINVAL; + } free(routing.u.s); } - if (conf->routing_type == ROUTING_INVALID) - return -1; + switch (conf->routing.pol) { + case ROUTING_LINK_STATE: + log_info("Using Link State routing policy."); + t_recalc = toml_int_in(table, "ls_t_recalc"); + if (t_recalc.ok) { + if (t_recalc.u.i < 1) { + log_err("Invalid ls_t_recalc value: %ld", + (long) t_recalc.u.i); + return -EINVAL; + } + conf->routing.ls.t_recalc = t_recalc.u.i; + } + t_update = toml_int_in(table, "ls_t_update"); + if (t_update.ok) { + if (t_update.u.i < 1) { + log_err("Invalid ls_t_update value: %ld", + (long) t_update.u.i); + return -EINVAL; + } + conf->routing.ls.t_update = t_update.u.i; + } + t_timeo = toml_int_in(table, "ls_t_timeo"); + if (t_timeo.ok) { + if (t_timeo.u.i < 1) { + log_err("Invalid ls_t_timeo value: %ld", + (long) t_timeo.u.i); + return -EINVAL; + } + conf->routing.ls.t_timeo = t_timeo.u.i; + } + break; + default: + log_err("Invalid routing policy: %d", conf->routing.pol); + return -EINVAL; + } return 0; } @@ -311,12 +501,12 @@ static int toml_dt(toml_table_t * table, static int toml_unicast(toml_table_t * table, struct ipcp_config * conf) { - - *conf = uni_default_conf; - if (toml_hash(table, &conf->layer_info) < 0) + if (toml_dir(table, &conf->unicast.dir) < 0) { + log_err("Invalid directory configuration."); return -1; + } if (toml_dt(table, &conf->unicast.dt) < 0) { log_err("Invalid DT configuration."); @@ -333,6 +523,7 @@ static int toml_unicast(toml_table_t * table, return -1; } + return 0; } @@ -504,8 +695,11 @@ static int toml_ipcp(toml_table_t * table, case IPCP_ETH_LLC: ret = toml_eth_llc(table, conf); break; - case IPCP_UDP: - ret = toml_udp(table, conf); + case IPCP_UDP4: + ret = toml_udp4(table, conf); + break; + case IPCP_UDP6: + ret = toml_udp6(table, conf); break; case IPCP_BROADCAST: ret = toml_broadcast(table, conf); @@ -563,7 +757,7 @@ static int toml_ipcp_list(toml_table_t * table, } info.type = type; - strcpy(info.name,key); + strcpy(info.name, key); conf.type = type; ret = toml_ipcp(toml_table_in(table, key), &info, &conf); @@ -691,18 +885,63 @@ static int toml_prog_list(toml_array_t * progs, return ret; } +static int cp_chk_path(char * buf, + char * path) +{ + char * rp; + + assert(path != NULL); + + rp = realpath(path, NULL); + if (rp == NULL) { + log_err("Failed to check path %s: %s.", path, strerror(errno)); + goto fail_rp; + } + + if (strlen(rp) > NAME_PATH_SIZE) { + log_err("File path too long: %s.", rp); + goto fail_len; + } + + strcpy(buf, rp); + free(rp); + free(path); + + return 0; + + fail_len: + free(rp); + fail_rp: + free(path); + return -1; +} + static int toml_name(toml_table_t * table, const char * name) { - toml_array_t * progs; - toml_array_t * args; - toml_datum_t lb; + toml_array_t * progs; + toml_array_t * args; + toml_datum_t lb; + toml_datum_t ssec; + toml_datum_t scrt; + toml_datum_t skey; + toml_datum_t csec; + toml_datum_t ccrt; + toml_datum_t ckey; + struct name_info info = { .pol_lb = LB_SPILL }; log_dbg("Found service name %s in configuration file.", name); + if (strlen(name) > NAME_SIZE) { + log_err("Name too long: %s", name); + return -1; + } + + strcpy(info.name, name); + lb = toml_string_in(table, "lb"); if (lb.ok) { if (strcmp(lb.u.s, "spill") == 0) @@ -718,8 +957,29 @@ static int toml_name(toml_table_t * table, log_err("Invalid load-balancing policy for %s.", name); return -1; } + ssec = toml_string_in(table, "server_sec_file"); + if (ssec.ok && cp_chk_path(info.s.sec, ssec.u.s) < 0) + return -1; - strcpy(info.name, name); + scrt = toml_string_in(table, "server_crt_file"); + if (scrt.ok && cp_chk_path(info.s.crt, scrt.u.s) < 0) + return -1; + + skey = toml_string_in(table, "server_key_file"); + if (skey.ok && cp_chk_path(info.s.key, skey.u.s) < 0) + return -1; + + csec = toml_string_in(table, "client_sec_file"); + if (csec.ok && cp_chk_path(info.c.sec, csec.u.s) < 0) + return -1; + + ccrt = toml_string_in(table, "client_crt_file"); + if (ccrt.ok && cp_chk_path(info.c.crt, ccrt.u.s) < 0) + return -1; + + ckey = toml_string_in(table, "client_key_file"); + if (ckey.ok && cp_chk_path(info.c.key, ckey.u.s) < 0) + return -1; if (name_create(&info) < 0) { log_err("Failed to create name %s.", name); @@ -762,23 +1022,24 @@ static int toml_toplevel(toml_table_t * table, toml_table_t * subtable; subtable = toml_table_in(table, key); - - if (strcmp(key, "local") == 0) + if (strcmp(key, "name") == 0) + return toml_name_list(subtable); + else if (strcmp(key, "local") == 0) return toml_ipcp_list(subtable, IPCP_LOCAL); else if (strcmp(key, "eth-dix") == 0) return toml_ipcp_list(subtable, IPCP_ETH_DIX); else if (strcmp(key, "eth-llc") == 0) return toml_ipcp_list(subtable, IPCP_ETH_LLC); - else if (strcmp(key, "udp") == 0) - return toml_ipcp_list(subtable, IPCP_UDP); + else if (strcmp(key, "udp4") == 0) + return toml_ipcp_list(subtable, IPCP_UDP4); + else if (strcmp(key, "udp6") == 0) + return toml_ipcp_list(subtable, IPCP_UDP6); else if (strcmp(key, "broadcast") == 0) return toml_ipcp_list(subtable, IPCP_BROADCAST); else if (strcmp(key, "unicast") == 0) return toml_ipcp_list(subtable, IPCP_UNICAST); - else if (strcmp(key, "name") == 0) - return toml_name_list(subtable); - - log_err("Unkown toplevel key: %s.", key); + else + log_err("Unkown toplevel key: %s.", key); return -1; } @@ -838,7 +1099,8 @@ int irm_configure(const char * path) rp = realpath(path, NULL); if (rp == NULL) { - log_err("Failed to resolve path for %s", path); + log_err("Failed to check path for %s: %s.", + path, strerror(errno)); goto fail_resolve; } diff --git a/src/irmd/configfile.h b/src/irmd/configfile.h index 3ccf53fd..62f522a4 100644 --- a/src/irmd/configfile.h +++ b/src/irmd/configfile.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * The IPC Resource Manager / Configuration from file * diff --git a/src/irmd/ipcp.c b/src/irmd/ipcp.c index 5a9a79d3..7eccfc80 100644 --- a/src/irmd/ipcp.c +++ b/src/irmd/ipcp.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * The API to instruct IPCPs * @@ -83,28 +83,30 @@ ipcp_msg_t * send_recv_ipcp_msg(pid_t pid, { int sockfd; uint8_t buf[SOCK_BUF_SIZE]; - char * sock_path; + char * spath; ssize_t len; - ipcp_msg_t * recv_msg; struct timeval tv; struct timespec tic; struct timespec toc; - bool dealloc = false; + bool may_fail = false; if (kill(pid, 0) < 0) return NULL; - sock_path = ipcp_sock_path(pid); - if (sock_path == NULL) + spath = sock_path(pid, IPCP_SOCK_PATH_PREFIX); + if (spath == NULL) { + log_err("Failed to get IPCP socket path for pid %d.", pid); return NULL; + } - sockfd = client_socket_open(sock_path); + sockfd = client_socket_open(spath); if (sockfd < 0) { - free(sock_path); + log_err("Failed to open client socket at %s.", spath); + free(spath); return NULL; } - free(sock_path); + free(spath); len = ipcp_msg__get_packed_size(msg); if (len == 0 || len >= SOCK_BUF_SIZE) { @@ -127,6 +129,7 @@ ipcp_msg_t * send_recv_ipcp_msg(pid_t pid, tv.tv_usec = (REG_TIMEOUT % 1000) * 1000; break; case IPCP_MSG_CODE__IPCP_QUERY: + may_fail = true; /* name not always in Layer */ tv.tv_sec = QUERY_TIMEOUT / 1000; tv.tv_usec = (QUERY_TIMEOUT % 1000) * 1000; break; @@ -139,7 +142,7 @@ ipcp_msg_t * send_recv_ipcp_msg(pid_t pid, tv.tv_usec = (FLOW_ALLOC_TIMEOUT % 1000) * 1000; break; case IPCP_MSG_CODE__IPCP_FLOW_DEALLOC: - dealloc = true; + may_fail = true; tv.tv_sec = 0; /* FIX DEALLOC: don't wait for dealloc */ tv.tv_usec = 500; break; @@ -167,17 +170,15 @@ ipcp_msg_t * send_recv_ipcp_msg(pid_t pid, pthread_cleanup_pop(true); /* close socket */ if (len > 0) - recv_msg = ipcp_msg__unpack(NULL, len, buf); - else { - if (errno == EAGAIN && !dealloc) { - int diff = ts_diff_ms(&tic, &toc); - log_warn("IPCP %s timed out after %d ms.", - str_ipcp_cmd(msg->code), diff); - } - return NULL; + return ipcp_msg__unpack(NULL, len, buf); + + if (errno == EAGAIN && !may_fail) { + int diff = ts_diff_ms(&toc, &tic); + log_warn("IPCP %s timed out after %d ms.", + str_ipcp_cmd(msg->code), diff); } - return recv_msg; + return NULL; } int ipcp_bootstrap(pid_t pid, @@ -420,6 +421,8 @@ int ipcp_flow_join(const struct flow_info * flow, msg.flow_id = flow->id; msg.has_pid = true; msg.pid = flow->n_pid; + msg.has_uid = true; + msg.uid = flow->uid; msg.has_hash = true; msg.hash.data = (uint8_t *) dst.data; msg.hash.len = dst.len; @@ -441,6 +444,38 @@ int ipcp_flow_join(const struct flow_info * flow, return ret; } +int ipcp_flow_update(const struct flow_info * flow, + const buffer_t data) +{ + ipcp_msg_t msg = IPCP_MSG__INIT; + ipcp_msg_t * recv_msg; + int ret; + + msg.code = IPCP_MSG_CODE__IPCP_FLOW_UPDATE; + msg.has_flow_id = true; + msg.flow_id = flow->id; + msg.has_pk = true; + msg.pk.data = data.data; + msg.pk.len = data.len; + + recv_msg = send_recv_ipcp_msg(flow->n_1_pid, &msg); + if (recv_msg == NULL) { + log_err("Did not receive message."); + return -EIPCP; + } + + if (!recv_msg->has_result) { + log_err("Message has no result"); + ipcp_msg__free_unpacked(recv_msg, NULL); + return -EIPCP; + } + + ret = recv_msg->result; + ipcp_msg__free_unpacked(recv_msg, NULL); + + return ret; +} + int ipcp_flow_alloc(const struct flow_info * flow, const buffer_t dst, const buffer_t data) @@ -454,6 +489,8 @@ int ipcp_flow_alloc(const struct flow_info * flow, msg.flow_id = flow->id; msg.has_pid = true; msg.pid = flow->n_pid; + msg.has_uid = true; + msg.uid = flow->uid; msg.qosspec = qos_spec_s_to_msg(&flow->qs); msg.has_hash = true; msg.hash.data = (uint8_t *) dst.data; @@ -494,9 +531,11 @@ int ipcp_flow_alloc_resp(const struct flow_info * flow, msg.flow_id = flow->id; msg.has_pid = true; msg.pid = flow->n_pid; + msg.has_uid = true; + msg.uid = flow->uid; msg.has_response = true; msg.response = response; - msg.has_pk = true; + msg.has_pk = response == 0; msg.pk.data = data.data; msg.pk.len = data.len; diff --git a/src/irmd/ipcp.h b/src/irmd/ipcp.h index b7413cd2..8d06623c 100644 --- a/src/irmd/ipcp.h +++ b/src/irmd/ipcp.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * The API for the IRM to instruct IPCPs * @@ -68,4 +68,7 @@ int ipcp_flow_dealloc(pid_t pid, int flow_id, time_t timeo); +int ipcp_flow_update(const struct flow_info * flow, + const buffer_t data); + #endif /* OUROBOROS_IRMD_IPCP_H */ diff --git a/src/irmd/irmd.h b/src/irmd/irmd.h index cf8f6953..f88378ad 100644 --- a/src/irmd/irmd.h +++ b/src/irmd/irmd.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * The IPC Resource Manager * @@ -39,7 +39,7 @@ int connect_ipcp(pid_t pid, const char * component, qosspec_t qs); -int name_create(const struct name_info * info); +int name_create(struct name_info * info); int name_reg(const char * name, pid_t pid); diff --git a/src/irmd/main.c b/src/irmd/main.c index bc13fa7c..19be4ab9 100644 --- a/src/irmd/main.c +++ b/src/irmd/main.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * The IPC Resource Manager * @@ -22,8 +22,10 @@ #if defined(__linux__) || defined(__CYGWIN__) #define _DEFAULT_SOURCE +#define _GNU_SOURCE #else -#define _POSIX_C_SOURCE 200809L +#define _DEFAULT_SOURCE +#define _BSD_SOURCE #endif #include "config.h" @@ -34,14 +36,17 @@ #include <ouroboros/crypt.h> #include <ouroboros/errno.h> #include <ouroboros/flow.h> +#include <ouroboros/fqueue.h> #include <ouroboros/hash.h> #include <ouroboros/irm.h> #include <ouroboros/list.h> #include <ouroboros/lockfile.h> #include <ouroboros/logs.h> +#include <ouroboros/protobuf.h> #include <ouroboros/pthread.h> +#include <ouroboros/random.h> #include <ouroboros/rib.h> -#include <ouroboros/shm_rdrbuff.h> +#include <ouroboros/ssm_pool.h> #include <ouroboros/sockets.h> #include <ouroboros/time.h> #include <ouroboros/tpm.h> @@ -50,18 +55,26 @@ #include "irmd.h" #include "ipcp.h" +#include "oap.h" #include "reg/reg.h" #include "configfile.h" -#include <sys/socket.h> -#include <sys/un.h> +#include <dirent.h> +#include <grp.h> +#include <pwd.h> #include <signal.h> +#include <spawn.h> #include <stdlib.h> #include <string.h> #include <limits.h> +#include <sys/socket.h> #include <sys/stat.h> #include <sys/wait.h> -#include <spawn.h> +#include <sys/un.h> +#ifdef __APPLE__ +#include <sys/types.h> +#include <unistd.h> +#endif #ifdef HAVE_LIBGCRYPT #include <gcrypt.h> @@ -71,11 +84,18 @@ #define SHM_SAN_HOLDOFF 1000 /* ms */ #define IPCP_HASH_LEN(p) hash_len((p)->dir_hash_algo) #define BIND_TIMEOUT 10 /* ms */ +#define TIMESYNC_SLACK 100 /* ms */ +#define OAP_SEEN_TIMER 20 /* s */ #define DEALLOC_TIME 300 /* s */ -#define MSGBUFSZ 2048 +#define REKEY_BATCH 64 /* flows re-keyed per timer pass */ +#define REKEY_RESP_TIMEO 20 /* s; give-up on a re-key RESPONSE */ +#define DIRECT_MPL 20 /* ms */ +/* bytes; in-process, bounded only by PUP/GSPP. */ +#define DIRECT_MTU 65000 enum irm_state { IRMD_NULL = 0, + IRMD_INIT, IRMD_RUNNING, IRMD_SHUTDOWN }; @@ -88,13 +108,38 @@ struct cmd { int fd; }; +/* In-flight Tier-2 re-key, owned solely by the re-key worker thread. */ +struct rekey_ctx { + struct list_head next; + + int flow_id; + void * ctx; /* OAP client ctx (opaque) */ + struct timespec deadline; /* reap if no RESPONSE by then */ +}; + +enum rekey_evt_type { + REKEY_INIT = 0, /* start an exchange for flow_id */ + REKEY_REQ, /* a REQUEST arrived for flow_id */ + REKEY_RESP, /* a RESPONSE arrived for flow_id */ + REKEY_DIRECT /* in-process re-key, direct flow */ +}; + +struct rekey_evt { + struct list_head next; + + enum rekey_evt_type type; + int flow_id; + pid_t n_1_pid; /* INIT: flow's lower IPCP */ + buffer_t buf; /* RESP: owned RESPONSE payload */ +}; + struct { bool log_stdout; /* log to stdout */ #ifdef HAVE_TOML char * cfg_file; /* configuration file path */ #endif struct lockfile * lf; /* single irmd per system */ - struct shm_rdrbuff * rdrb; /* rdrbuff for packets */ + struct ssm_pool * gspp; /* pool for packets */ int sockfd; /* UNIX socket */ @@ -109,6 +154,13 @@ struct { pthread_t irm_sanitize; /* clean up irmd resources */ pthread_t acceptor; /* accept new commands */ + + struct { + pthread_t worker; /* Tier-2 re-key orchestrator */ + struct list_head inbox; /* re-key events for worker */ + pthread_cond_t cond; /* inbox signal condvar */ + pthread_mutex_t mtx; /* inbox lock */ + } rk; } irmd; static enum irm_state irmd_get_state(void) @@ -173,8 +225,11 @@ static pid_t spawn_ipcp(struct ipcp_info * info) case IPCP_BROADCAST: exec_name = IPCP_BROADCAST_EXEC; break; - case IPCP_UDP: - exec_name = IPCP_UDP_EXEC; + case IPCP_UDP4: + exec_name = IPCP_UDP4_EXEC; + break; + case IPCP_UDP6: + exec_name = IPCP_UDP6_EXEC; break; case IPCP_ETH_LLC: exec_name = IPCP_ETH_LLC_EXEC; @@ -217,7 +272,7 @@ static pid_t spawn_ipcp(struct ipcp_info * info) } info->pid = pid; - info->state = IPCP_BOOT; + info->state = IPCP_INIT; return 0; } @@ -312,7 +367,7 @@ int bootstrap_ipcp(pid_t pid, goto fail; } - if (conf->type == IPCP_UDP) + if (conf->type == IPCP_UDP4 || conf->type == IPCP_UDP6) conf->layer_info.dir_hash_algo = (enum pol_dir_hash) HASH_MD5; if (ipcp_bootstrap(pid, conf, &layer)) { @@ -320,7 +375,7 @@ int bootstrap_ipcp(pid_t pid, goto fail; } - info.state = IPCP_BOOTSTRAPPED; + info.state = IPCP_BOOT; if (reg_set_layer_for_ipcp(&info, &layer) < 0) { log_err("Failed to set layer info for IPCP."); @@ -352,6 +407,8 @@ int enroll_ipcp(pid_t pid, goto fail; } + info.state = IPCP_BOOT; + if (reg_set_layer_for_ipcp(&info, &layer) < 0) { log_err("Failed to set layer info for IPCP."); goto fail; @@ -425,6 +482,71 @@ static int disconnect_ipcp(pid_t pid, return 0; } +static void name_update_sec_paths(struct name_info * info) +{ + char * srv_dir = OUROBOROS_SRV_CRT_DIR; + char * cli_dir = OUROBOROS_CLI_CRT_DIR; + + assert(info != NULL); + + if (strlen(info->s.sec) == 0) + sprintf(info->s.sec, "%s/%s/sec.conf", srv_dir, info->name); + + if (strlen(info->s.crt) == 0) + sprintf(info->s.crt, "%s/%s/crt.pem", srv_dir, info->name); + + if (strlen(info->s.key) == 0) + sprintf(info->s.key, "%s/%s/key.pem", srv_dir, info->name); + + if (strlen(info->c.sec) == 0) + sprintf(info->c.sec, "%s/%s/sec.conf", cli_dir, info->name); + + if (strlen(info->c.crt) == 0) + sprintf(info->c.crt, "%s/%s/crt.pem", cli_dir, info->name); + + if (strlen(info->c.key) == 0) + sprintf(info->c.key, "%s/%s/key.pem", cli_dir, info->name); +} + +int name_create(struct name_info * info) +{ + int ret; + + assert(info != NULL); + + name_update_sec_paths(info); + + ret = reg_create_name(info); + if (ret == -EEXIST) { + log_info("Name %s already exists.", info->name); + return 0; + } + + if (ret < 0) { + log_err("Failed to create name %s.", info->name); + return -1; + } + + log_info("Created new name: %s.", info->name); + + return 0; +} + +static int name_destroy(const char * name) +{ + + assert(name != NULL); + + if (reg_destroy_name(name) < 0) { + log_err("Failed to destroy name %s.", name); + return -1; + } + + log_info("Destroyed name: %s.", name); + + return 0; +} + int bind_program(char ** exec, const char * name, uint8_t flags) @@ -448,10 +570,8 @@ int bind_program(char ** exec, if (!reg_has_name(name)) { ni.pol_lb = LB_SPILL; strcpy(ni.name, name); - if (reg_create_name(&ni) < 0) { - log_err("Failed to create name %s.", name); + if (name_create(&ni) < 0) goto fail_name; - } } if (reg_bind_prog(name, exec, flags) < 0) { @@ -497,10 +617,8 @@ int bind_process(pid_t pid, if (!reg_has_name(name)) { ni.pol_lb = LB_SPILL; strcpy(ni.name, name); - if (reg_create_name(&ni) < 0) { - log_err("Failed to create name %s.", name); + if (name_create(&ni) < 0) goto fail; - } } if (reg_bind_proc(name, pid) < 0) { @@ -529,18 +647,21 @@ static int unbind_program(const char * prog, if (name == NULL) { if (reg_destroy_prog(prog) < 0) { log_err("Failed to unbind %s.", prog); - return -1; + goto fail; } log_info("Program %s unbound.", prog); } else { if (reg_unbind_prog(name, prog) < 0) { log_err("Failed to unbind %s from %s", prog, name); - return -1; + goto fail; } log_info("Name %s unbound for %s.", name, prog); } return 0; + + fail: + return -1; } static int unbind_process(pid_t pid, @@ -549,18 +670,21 @@ static int unbind_process(pid_t pid, if (name == NULL) { if (reg_destroy_proc(pid) < 0) { log_err("Failed to unbind %d.", pid); - return -1; + goto fail; } log_info("Process %d unbound.", pid); } else { if (reg_unbind_proc(name, pid) < 0) { log_err("Failed to unbind %d from %s", pid, name); - return -1; + goto fail; } log_info("Name %s unbound for process %d.", name, pid); } return 0; + + fail: + return -1; } static int list_ipcps(ipcp_list_msg_t *** ipcps, @@ -581,43 +705,6 @@ static int list_ipcps(ipcp_list_msg_t *** ipcps, return -1; } -int name_create(const struct name_info * info) -{ - int ret; - - assert(info != NULL); - - ret = reg_create_name(info); - if (ret == -EEXIST) { - log_info("Name %s already exists.", info->name); - return 0; - } - - if (ret < 0) { - log_err("Failed to create name %s.", info->name); - return -1; - } - - log_info("Created new name: %s.", info->name); - - return 0; -} - -static int name_destroy(const char * name) -{ - - assert(name != NULL); - - if (reg_destroy_name(name) < 0) { - log_err("Failed to destroy name %s.", name); - return -1; - } - - log_info("Destroyed name: %s.", name); - - return 0; -} - static int list_names(name_info_msg_t *** names, size_t * n_names) { @@ -669,8 +756,7 @@ int name_reg(const char * name, if (ipcp_reg(pid, hash)) { log_err("Could not register " HASH_FMT32 " with IPCP %d.", HASH_VAL32(hash.data), pid); - freebuf(hash); - return -1; + goto fail_hash; } log_info("Registered %s with IPCP %d as " HASH_FMT32 ".", @@ -679,6 +765,10 @@ int name_reg(const char * name, freebuf(hash); return 0; + + fail_hash: + freebuf(hash); + return -1; } static int name_unreg(const char * name, @@ -713,8 +803,7 @@ static int name_unreg(const char * name, if (ipcp_unreg(pid, hash)) { log_err("Could not unregister %s with IPCP %d.", name, pid); - freebuf(hash); - return -1; + goto fail_hash; } log_info("Unregistered %s from %d.", name, pid); @@ -722,20 +811,59 @@ static int name_unreg(const char * name, freebuf(hash); return 0; + + fail_hash: + freebuf(hash); + return -1; +} + +static int get_peer_ids(int fd, + uid_t * uid, + gid_t * gid, + pid_t * pid) +{ +#if defined(__linux__) + struct ucred ucred; + socklen_t len; + + len = sizeof(ucred); + + if (getsockopt(fd, SOL_SOCKET, SO_PEERCRED, &ucred, &len) < 0) + goto fail; + + *uid = ucred.uid; + *gid = ucred.gid; + if (pid != NULL) + *pid = ucred.pid; +#else + if (getpeereid(fd, uid, gid) < 0) + goto fail; + + if (pid != NULL) + *pid = -1; /* no portable SO_PEERCRED.pid equivalent */ +#endif + return 0; + fail: + return -1; } static int proc_announce(const struct proc_info * info) { + if (reg_prepare_pool(info->uid, info->gid) < 0) { + log_err("Failed to prepare pool for uid %d.", info->uid); + goto fail; + } + if (reg_create_proc(info) < 0) { log_err("Failed to add process %d.", info->pid); - goto fail_proc; + goto fail; } log_info("Process added: %d (%s).", info->pid, info->prog); return 0; - fail_proc: + fail: return -1; } @@ -749,12 +877,6 @@ static int proc_exit(pid_t pid) return 0; } -static void __cleanup_pkp(void * pkp) -{ - if (pkp != NULL) - crypt_dh_pkp_destroy(pkp); -} - static void __cleanup_flow(void * flow) { reg_destroy_flow(((struct flow_info *) flow)->id); @@ -762,65 +884,44 @@ static void __cleanup_flow(void * flow) static int flow_accept(struct flow_info * flow, buffer_t * data, - struct timespec * abstime) + struct timespec * abstime, + struct crypt_sk * sk) { - uint8_t buf[MSGBUFSZ]; - buffer_t lpk; /* local public key */ - buffer_t rpk; /* remote public key */ - void * pkp; /* my public/private key pair */ - ssize_t key_len; - uint8_t * s; - int err; + buffer_t req_hdr; + buffer_t resp_hdr; + buffer_t peer_crt = BUF_INIT; + char name[NAME_SIZE + 1]; + struct name_info info; + int err; - /* piggyback of user data not yet implemented */ - assert(data != NULL && data->len == 0 && data->data == NULL); + assert(data != NULL && BUF_IS_EMPTY(data)); + + clrbuf(req_hdr); + clrbuf(resp_hdr); if (!reg_has_proc(flow->n_pid)) { log_err("Unknown process %d calling accept.", flow->n_pid); err = -EINVAL; - goto fail; - } - - s = malloc(SYMMKEYSZ); - if (s == NULL) { - log_err("Failed to malloc symmkey."); - err = -ENOMEM; - goto fail; - } - - key_len = crypt_dh_pkp_create(&pkp, buf); - if (key_len < 0) { - log_err("Failed to generate key pair."); - err = -ECRYPT; - goto fail_pkp; + goto fail_flow; } - lpk.data = buf; - lpk.len = (size_t) key_len; - - log_dbg("Generated ephemeral keys for %d.", flow->n_pid); - if (reg_create_flow(flow) < 0) { log_err("Failed to create flow."); err = -EBADF; goto fail_flow; } - if (reg_prepare_flow_accept(flow, &lpk) < 0) { + if (reg_prepare_flow_accept(flow) < 0) { log_err("Failed to prepare accept."); err = -EBADF; goto fail_wait; } pthread_cleanup_push(__cleanup_flow, flow); - pthread_cleanup_push(__cleanup_pkp, pkp); - pthread_cleanup_push(free, s); - err = reg_wait_flow_accepted(flow, &rpk, abstime); + err = reg_wait_flow_accepted(flow, &req_hdr, abstime); pthread_cleanup_pop(false); - pthread_cleanup_pop(false); - pthread_cleanup_pop(false); if (err == -ETIMEDOUT) { log_err("Flow accept timed out."); @@ -835,45 +936,78 @@ static int flow_accept(struct flow_info * flow, assert(err == 0); - if (flow->qs.cypher_s != 0) { /* crypto requested */ - if (crypt_dh_derive(pkp, rpk, s) < 0) { - log_err("Failed to derive secret for %d.", flow->id); - err = -ECRYPT; - goto fail_derive; - } - freebuf(rpk); - data->data = s; - data->len = SYMMKEYSZ; - s= NULL; - } else { - clrbuf(lpk); + if (reg_get_name_for_flow_id(name, flow->id) < 0) { + log_err("Failed to get name for flow %d.", flow->id); + err = -EIPCP; + goto fail_oap; } - if (ipcp_flow_alloc_resp(flow, 0, lpk) < 0) { + if (reg_get_name_info(name, &info) < 0) { + log_err("Failed to get name info for %s.", name); + err = -ENAME; + goto fail_oap; + } + + log_dbg("IPCP %d accepting flow %d for %s.", + flow->n_pid, flow->id, name); + + flow->uid = reg_get_proc_uid(flow->n_pid); + + err = oap_srv_process(&info, req_hdr, &resp_hdr, data, sk, + false, NULL, &peer_crt); + if (err == -EREPLAY) { + log_warn("Dropping replayed alloc request for %s.", name); + goto fail_replay; + } + if (err < 0) { + log_err("OAP processing failed for %s.", name); + goto fail_oap; + } + + if (reg_flow_is_direct(flow->id)) { + if (reg_respond_flow_direct(flow->id, &resp_hdr) < 0) { + log_err("Failed to respond to direct flow."); + goto fail_resp; + } + if (sk->nid != NID_undef) + reg_flow_set_rekey(flow->id, false, peer_crt); + log_info("Flow %d accepted (direct) by %d for %s.", + flow->id, flow->n_pid, name); + } else if (ipcp_flow_alloc_resp(flow, 0, resp_hdr) < 0) { log_err("Failed to respond to flow allocation."); - err = -EIPCP; - goto fail_alloc_resp; + goto fail_resp; + } else { + if (sk->nid != NID_undef) + reg_flow_set_rekey(flow->id, false, peer_crt); + log_info("Flow %d accepted by %d for %s (uid %d).", + flow->id, flow->n_pid, name, flow->uid); } - crypt_dh_pkp_destroy(pkp); - free(s); + freebuf(peer_crt); + freebuf(req_hdr); + freebuf(resp_hdr); return 0; - fail_derive: - freebuf(rpk); - clrbuf(lpk); - ipcp_flow_alloc_resp(flow, err, lpk); - fail_alloc_resp: - flow->state = FLOW_NULL; + fail_oap: + if (!reg_flow_is_direct(flow->id)) + ipcp_flow_alloc_resp(flow, err, resp_hdr); + fail_replay: + freebuf(peer_crt); + freebuf(req_hdr); + freebuf(resp_hdr); fail_wait: reg_destroy_flow(flow->id); fail_flow: - crypt_dh_pkp_destroy(pkp); - fail_pkp: - free(s); - fail: return err; + + fail_resp: + flow->state = FLOW_NULL; + freebuf(peer_crt); + freebuf(req_hdr); + freebuf(resp_hdr); + reg_destroy_flow(flow->id); + return -EIPCP; } static int flow_join(struct flow_info * flow, @@ -883,17 +1017,20 @@ static int flow_join(struct flow_info * flow, struct ipcp_info ipcp; struct layer_info layer; buffer_t hash; - buffer_t pbuf = {NULL, 0}; /* nothing to piggyback */ + buffer_t pbuf = BUF_INIT; /* nothing to piggyback */ int err; - log_info("Allocating flow for %d to %s.", flow->n_pid, dst); - if (reg_create_flow(flow) < 0) { log_err("Failed to create flow."); err = -EBADF; goto fail_flow; } + flow->uid = reg_get_proc_uid(flow->n_pid); + + log_info("Allocating flow for %d to %s (uid %d).", + flow->n_pid, dst, flow->uid); + strcpy(layer.name, dst); if (reg_get_ipcp_by_layer(&ipcp, &layer) < 0) { log_err("Failed to get IPCP for layer %s.", dst); @@ -901,6 +1038,8 @@ static int flow_join(struct flow_info * flow, goto fail_ipcp; } + flow->n_1_pid = ipcp.pid; + hash.len = hash_len((enum hash_algo) layer.dir_hash_algo); hash.data = malloc(hash.len); if (hash.data == NULL) { @@ -909,6 +1048,8 @@ static int flow_join(struct flow_info * flow, goto fail_ipcp; } + str_hash((enum hash_algo) layer.dir_hash_algo, hash.data, dst); + reg_prepare_flow_alloc(flow); if (ipcp_flow_join(flow, hash)) { @@ -936,6 +1077,7 @@ static int flow_join(struct flow_info * flow, goto fail_alloc; } + assert(pbuf.data == NULL && pbuf.len == 0); assert(err == 0); freebuf(hash); @@ -954,7 +1096,7 @@ static int get_ipcp_by_dst(const char * dst, pid_t * pid, buffer_t * hash) { - ipcp_list_msg_t ** ipcps; + ipcp_list_msg_t ** ipcps = NULL; int n; int i; int err = -EIPCP; @@ -1007,128 +1149,7 @@ static int get_ipcp_by_dst(const char * dst, return err; } -static int flow_alloc(struct flow_info * flow, - const char * dst, - buffer_t * data, - struct timespec * abstime) -{ - uint8_t buf[MSGBUFSZ]; - buffer_t lpk ={NULL, 0}; /* local public key */ - buffer_t rpk; /* remote public key */ - void * pkp = NULL; /* my public/private key pair */ - uint8_t * s = NULL; - buffer_t hash; - int err; - /* piggyback of user data not yet implemented */ - assert(data != NULL && data->len == 0 && data->data == NULL); - - log_info("Allocating flow for %d to %s.", flow->n_pid, dst); - - - if (flow->qs.cypher_s > 0) { - ssize_t key_len; - - s = malloc(SYMMKEYSZ); - if (s == NULL) { - log_err("Failed to malloc symmetric key"); - err = -ENOMEM; - goto fail_malloc; - } - - key_len = crypt_dh_pkp_create(&pkp, buf); - if (key_len < 0) { - log_err("Failed to generate key pair."); - err = -ECRYPT; - goto fail_pkp; - } - - lpk.data = buf; - lpk.len = (size_t) key_len; - - log_dbg("Generated ephemeral keys for %d.", flow->n_pid); - } - - if (reg_create_flow(flow) < 0) { - log_err("Failed to create flow."); - err = -EBADF; - goto fail_flow; - } - - if (get_ipcp_by_dst(dst, &flow->n_1_pid, &hash) < 0) { - log_err("Failed to find IPCP for %s.", dst); - err = -EIPCP; - goto fail_ipcp; - } - - reg_prepare_flow_alloc(flow); - - if (ipcp_flow_alloc(flow, hash, lpk)) { - log_err("Flow allocation %d failed.", flow->id); - err = -ENOTALLOC; - goto fail_alloc; - } - - pthread_cleanup_push(__cleanup_flow, flow); - pthread_cleanup_push(__cleanup_pkp, pkp); - pthread_cleanup_push(free, hash.data); - pthread_cleanup_push(free, s); - - err = reg_wait_flow_allocated(flow, &rpk, abstime); - - pthread_cleanup_pop(false); - pthread_cleanup_pop(false); - pthread_cleanup_pop(false); - pthread_cleanup_pop(false); - - if (err == -ETIMEDOUT) { - log_err("Flow allocation timed out."); - goto fail_alloc; - } - - if (err == -1) { - log_dbg("Flow allocation terminated."); - err = -EPIPE; - goto fail_alloc; - } - - assert(err == 0); - - if (flow->qs.cypher_s != 0) { /* crypto requested */ - if (crypt_dh_derive(pkp, rpk, s) < 0) { - log_err("Failed to derive secret for %d.", flow->id); - err = -ECRYPT; - goto fail_derive; - } - crypt_dh_pkp_destroy(pkp); - freebuf(rpk); - data->data = s; - data->len = SYMMKEYSZ; - s = NULL; - } - - freebuf(hash); - free(s); - - return 0; - - fail_derive: - freebuf(rpk); - flow->state = FLOW_DEALLOCATED; - fail_alloc: - freebuf(hash); - fail_ipcp: - reg_destroy_flow(flow->id); - fail_flow: - if (flow->qs.cypher_s > 0) - crypt_dh_pkp_destroy(pkp); - fail_pkp: - free(s); - fail_malloc: - return err; -} - -static int wait_for_accept(enum hash_algo algo, - const uint8_t * hash) +static int wait_for_accept(const char * name) { struct timespec timeo = TIMESPEC_INIT_MS(IRMD_REQ_ARR_TIMEOUT); struct timespec abstime; @@ -1138,25 +1159,23 @@ static int wait_for_accept(enum hash_algo algo, clock_gettime(PTHREAD_COND_CLOCK, &abstime); ts_add(&abstime, &timeo, &abstime); - ret = reg_wait_flow_accepting(algo, hash, &abstime); + ret = reg_wait_flow_accepting(name, &abstime); if (ret == -ETIMEDOUT) { - if (reg_get_exec(algo, hash, &exec) < 0) { - log_dbg("No program bound to " HASH_FMT32 ".", - HASH_VAL32(hash)); + if (reg_get_exec(name, &exec) < 0) { + log_dbg("No program bound for %s.", name); goto fail; } - log_info("Autostarting %s.", exec[0]); - if (spawn_program(exec) < 0) { - log_dbg("Failed to autostart " HASH_FMT32 ".", - HASH_VAL32(hash)); + log_err("Failed to start %s for %s.", exec[0], name); goto fail_spawn; } + log_info("Starting %s for %s.", exec[0], name); + ts_add(&abstime, &timeo, &abstime); - ret = reg_wait_flow_accepting(algo, hash, &abstime); + ret = reg_wait_flow_accepting(name, &abstime); if (ret == -ETIMEDOUT) goto fail_spawn; @@ -1179,10 +1198,11 @@ static int flow_req_arr(struct flow_info * flow, struct layer_info layer; enum hash_algo algo; int ret; + char name[NAME_SIZE + 1]; info.pid = flow->n_1_pid; - log_info("Flow req arrived from IPCP %d for " HASH_FMT32 ".", + log_dbg("Flow req arrived from IPCP %d for " HASH_FMT32 ".", info.pid, HASH_VAL32(hash)); if (reg_get_ipcp(&info, &layer) < 0) { @@ -1193,10 +1213,17 @@ static int flow_req_arr(struct flow_info * flow, algo = (enum hash_algo) layer.dir_hash_algo; - ret = wait_for_accept(algo, hash); + if (reg_get_name_for_hash(name, algo, hash) < 0) { + log_warn("No name for " HASH_FMT32 ".", HASH_VAL32(hash)); + ret = -ENAME; + goto fail; + } + + log_info("Flow request arrived for %s.", name); + + ret = wait_for_accept(name); if (ret < 0) { - log_err("No activeprocess for " HASH_FMT32 ".", - HASH_VAL32(hash)); + log_err("No active process for %s.", name); goto fail; } @@ -1214,13 +1241,216 @@ static int flow_req_arr(struct flow_info * flow, return ret; } +#ifndef DISABLE_DIRECT_IPC +static int flow_alloc_direct(const char * dst, + struct flow_info * flow, + buffer_t * data, + struct timespec * abstime, + struct crypt_sk * sk, + struct name_info * info) +{ + struct flow_info acc; /* server side flow */ + buffer_t req_hdr = BUF_INIT; + buffer_t resp_hdr = BUF_INIT; + buffer_t no_crt = BUF_INIT; + void * ctx; + int err; + + acc.id = wait_for_accept(dst); + if (acc.id < 0) { + log_dbg("No accepting process for %s.", dst); + return -EAGAIN; + } + + if (oap_cli_prepare(&ctx, info, &req_hdr, *data, false) < 0) { + log_err("Failed to prepare OAP for %s.", dst); + return -EBADF; + } + + acc.n_1_pid = flow->n_pid; + acc.mpl = DIRECT_MPL; + acc.mtu = DIRECT_MTU; + acc.qs = flow->qs; + acc.state = FLOW_ALLOCATED; + + err = reg_prepare_flow_direct(&acc, &req_hdr, flow->uid); + if (err == -EPERM) { + log_dbg("UID mismatch, falling back."); + oap_ctx_free(ctx); + freebuf(req_hdr); + return -EPERM; + } + + if (err < 0) { + log_err("Failed to prepare direct flow."); + oap_ctx_free(ctx); + freebuf(req_hdr); + return -EBADF; + } + + err = reg_wait_flow_direct(acc.id, &resp_hdr, abstime); + if (err < 0) { + log_err("Timeout waiting for OAP response."); + oap_ctx_free(ctx); + return -ETIMEDOUT; + } + + err = oap_cli_complete(ctx, info, resp_hdr, data, sk, NULL, NULL); + if (err < 0) { + log_err("OAP completion failed for %s.", dst); + freebuf(resp_hdr); + return err; + } + + flow->id = acc.id; + flow->n_1_pid = acc.n_pid; + flow->mpl = DIRECT_MPL; + flow->mtu = DIRECT_MTU; + flow->state = FLOW_ALLOCATED; + + /* Mark encrypted for re-key; the acceptor caches the cert. */ + if (sk->nid != NID_undef) + reg_flow_set_rekey(acc.id, true, no_crt); + + log_info("Flow %d allocated (direct) for %d to %s.", + flow->id, flow->n_pid, dst); + + freebuf(resp_hdr); + + return 0; +} +#endif /* DISABLE_DIRECT_IPC */ + +static int flow_alloc(const char * dst, + struct flow_info * flow, + buffer_t * data, + struct timespec * abstime, + struct crypt_sk * sk) +{ + buffer_t req_hdr = BUF_INIT; + buffer_t resp_hdr = BUF_INIT; + buffer_t hash = BUF_INIT; + buffer_t peer_crt = BUF_INIT; + struct name_info info; + void * ctx; + int err; + + /* piggyback of user data not yet implemented */ + assert(data != NULL && BUF_IS_EMPTY(data)); + + /* Look up name_info for dst */ + if (reg_get_name_info(dst, &info) < 0) { + log_err("Failed to get name info for %s.", dst); + err = -ENAME; + goto fail_flow; + } + + flow->uid = reg_get_proc_uid(flow->n_pid); + + log_info("Allocating flow for %d to %s (uid %d).", + flow->n_pid, dst, flow->uid); + +#ifndef DISABLE_DIRECT_IPC + err = flow_alloc_direct(dst, flow, data, abstime, sk, &info); + if (err == 0) + return 0; + + if (err != -EPERM && err != -EAGAIN) + goto fail_flow; +#endif + if (reg_create_flow(flow) < 0) { + log_err("Failed to create flow."); + err = -EBADF; + goto fail_flow; + } + + reg_set_name_for_flow_id(dst, flow->id); + + if (get_ipcp_by_dst(dst, &flow->n_1_pid, &hash) < 0) { + log_err("Failed to find IPCP for %s.", dst); + err = -EIPCP; + goto fail_ipcp; + } + + if (reg_prepare_flow_alloc(flow) < 0) { + log_err("Failed to prepare flow allocation."); + err = -EBADF; + goto fail_prepare; + } + + if (oap_cli_prepare(&ctx, &info, &req_hdr, *data, false) < 0) { + log_err("Failed to prepare OAP request for %s.", dst); + err = -EBADF; + goto fail_prepare; + } + + if (ipcp_flow_alloc(flow, hash, req_hdr)) { + log_err("Flow allocation %d failed.", flow->id); + err = -EIPCP; + goto fail_alloc; + } + + pthread_cleanup_push(__cleanup_flow, flow); + pthread_cleanup_push(free, hash.data); + + err = reg_wait_flow_allocated(flow, &resp_hdr, abstime); + + pthread_cleanup_pop(false); + pthread_cleanup_pop(false); + + if (err == -ETIMEDOUT) { + log_err("Flow allocation timed out."); + goto fail_wait; + } + + log_dbg("Response for flow %d to %s.", flow->id, dst); + + if (err < 0) { + log_warn("Allocation rejected: %s (%d).", dst, err); + goto fail_peer; + } + + err = oap_cli_complete(ctx, &info, resp_hdr, data, sk, NULL, &peer_crt); + if (err < 0) { + log_err("OAP completion failed for %s.", dst); + goto fail_complete; + } + + if (sk->nid != NID_undef) + reg_flow_set_rekey(flow->id, true, peer_crt); + + freebuf(peer_crt); + freebuf(req_hdr); + freebuf(resp_hdr); + freebuf(hash); + + return 0; + + fail_complete: + freebuf(peer_crt); + ctx = NULL; /* free'd on complete */ + fail_peer: + flow->state = FLOW_DEALLOCATED; + fail_wait: + freebuf(resp_hdr); + fail_alloc: + freebuf(req_hdr); + oap_ctx_free(ctx); + fail_prepare: + freebuf(hash); + fail_ipcp: + reg_destroy_flow(flow->id); + fail_flow: + return err; +} + static int flow_alloc_reply(struct flow_info * flow, int response, buffer_t * data) { - flow->state = response ? FLOW_DEALLOCATED : FLOW_ALLOCATED; + flow->state = response != 0 ? FLOW_DEALLOCATED : FLOW_ALLOCATED; - if (reg_respond_alloc(flow, data) < 0) { + if (reg_respond_alloc(flow, data, response) < 0) { log_err("Failed to reply to flow %d.", flow->id); flow->state = FLOW_DEALLOCATED; return -EBADF; @@ -1232,11 +1462,17 @@ static int flow_alloc_reply(struct flow_info * flow, static int flow_dealloc(struct flow_info * flow, struct timespec * ts) { - log_info("Deallocating flow %d for process %d (timeout: %zd s).", + log_info("Deallocating flow %d for process %d (timeout: %ld s).", flow->id, flow->n_pid, ts->tv_sec); reg_dealloc_flow(flow); + if (reg_flow_is_direct(flow->id)) { + if (flow->state == FLOW_DEALLOCATED) + reg_destroy_flow(flow->id); + return 0; + } + if (ipcp_flow_dealloc(flow->n_1_pid, flow->id, ts->tv_sec) < 0) { log_err("Failed to request dealloc from %d.", flow->n_1_pid); return -EIPCP; @@ -1259,6 +1495,741 @@ static int flow_dealloc_resp(struct flow_info * flow) return 0; } +/* + * Inbox producers. Any thread may post; the worker drains. INIT carries + * the flow's lower IPCP pid; RESP transfers ownership of buf. + */ +static void rekey_post(enum rekey_evt_type type, + int flow_id, + pid_t n_1_pid, + buffer_t * buf) +{ + struct rekey_evt * evt; + + evt = malloc(sizeof(*evt)); + if (evt == NULL) { + log_err("Failed to malloc re-key event for flow %d.", flow_id); + if (type == REKEY_INIT || type == REKEY_DIRECT) + reg_flow_clear_in_flight(flow_id); + else + reg_flow_rekey_arr_done(flow_id, type == REKEY_REQ); + + if (buf != NULL) + freebuf(*buf); + + return; + } + + list_head_init(&evt->next); + evt->type = type; + evt->flow_id = flow_id; + evt->n_1_pid = n_1_pid; + clrbuf(evt->buf); + if (buf != NULL) { + evt->buf = *buf; + clrbuf(*buf); + } + + pthread_mutex_lock(&irmd.rk.mtx); + + list_add_tail(&evt->next, &irmd.rk.inbox); + pthread_cond_signal(&irmd.rk.cond); + + pthread_mutex_unlock(&irmd.rk.mtx); +} + +static void rekey_post_init(int flow_id, + pid_t n_1_pid) +{ + rekey_post(REKEY_INIT, flow_id, n_1_pid, NULL); +} + +static void rekey_post_resp(int flow_id, + buffer_t * buf) +{ + rekey_post(REKEY_RESP, flow_id, 0, buf); +} + +static void rekey_post_req(int flow_id, + pid_t n_1_pid, + buffer_t * buf) +{ + rekey_post(REKEY_REQ, flow_id, n_1_pid, buf); +} + +static void rekey_post_direct(int flow_id) +{ + rekey_post(REKEY_DIRECT, flow_id, 0, NULL); +} + +/* Worker-only: find an in-flight entry by flow_id. */ +static struct rekey_ctx * rekey_find(struct list_head * tbl, + int flow_id) +{ + struct list_head * p; + + list_for_each(p, tbl) { + struct rekey_ctx * e = list_entry(p, struct rekey_ctx, next); + if (e->flow_id == flow_id) + return e; + } + + return NULL; +} + +/* Worker-only: drop an entry, freeing its OAP ctx. */ +static void rekey_drop(struct rekey_ctx * e) +{ + if (e->ctx != NULL) + oap_ctx_free(e->ctx); + + list_del(&e->next); + free(e); +} + +/* Resolve a flow's registered name info; < 0 if the flow or name is gone. */ +static int rekey_name_info(int flow_id, + struct name_info * info) +{ + char name[NAME_SIZE + 1]; + + if (reg_get_name_for_flow_id(name, flow_id) < 0) + return -1; + + return reg_get_name_info(name, info); +} + +/* Flow-update relay payload: a 1-byte type prefix on an opaque body. */ +enum flow_upd_type { + FLOW_UPD_REKEY_REQ = 0, + FLOW_UPD_REKEY_RESP = 1, +}; + +/* Prepend the update type to body; caller frees out on success. */ +static int flow_upd_wrap(buffer_t * out, + uint8_t type, + const buffer_t * body) +{ + out->len = body->len + 1; + out->data = malloc(out->len); + if (out->data == NULL) + return -ENOMEM; + + out->data[0] = type; + memcpy(out->data + 1, body->data, body->len); + + return 0; +} + +/* Cleanup handlers — the re-key worker is cancelled at shutdown. */ +static void rk_free_evt(void * o) +{ + struct rekey_evt * evt = o; + + freebuf(evt->buf); + free(evt); +} + +static void rk_freebuf(void * o) +{ + freebuf(*(buffer_t *) o); +} + +static void rk_clear_in_flight(void * o) +{ + reg_flow_clear_in_flight(*(int *) o); +} + +static void rk_clear_key(void * o) +{ + crypt_secure_clear(o, SYMMKEYSZ); +} + +static void rekey_do_initiate(struct list_head * tbl, + int flow_id, + pid_t n_1_pid) +{ + struct rekey_ctx * e; + struct flow_info info; + struct name_info name; + buffer_t req = BUF_INIT; + buffer_t upd = BUF_INIT; + buffer_t data = BUF_INIT; + void * ctx = NULL; + int ret; + + e = rekey_find(tbl, flow_id); + if (e != NULL) + rekey_drop(e); /* Replace in-flight entries */ + + if (rekey_name_info(flow_id, &name) < 0) { + log_err("Failed to get name info to re-key flow %d.", flow_id); + goto fail; + } + + if (oap_cli_prepare(&ctx, &name, &req, data, true) < 0) { + log_err("Failed to prepare re-key for flow %d.", flow_id); + goto fail; + } + + memset(&info, 0, sizeof(info)); + info.id = flow_id; + info.n_1_pid = n_1_pid; + + if (flow_upd_wrap(&upd, FLOW_UPD_REKEY_REQ, &req) < 0) { + log_err("Failed to wrap re-key request for flow %d.", flow_id); + goto fail_ctx; + } + + pthread_cleanup_push(rk_clear_in_flight, &flow_id); + pthread_cleanup_push(oap_ctx_free, ctx); + pthread_cleanup_push(rk_freebuf, &req); + pthread_cleanup_push(rk_freebuf, &upd); + ret = ipcp_flow_update(&info, upd); + pthread_cleanup_pop(false); + pthread_cleanup_pop(false); + pthread_cleanup_pop(false); + pthread_cleanup_pop(false); + freebuf(upd); + if (ret < 0) { + log_err("Failed to send re-key request for flow %d.", flow_id); + goto fail_ctx; + } + + e = malloc(sizeof(*e)); + if (e == NULL) { + log_err("Failed to malloc re-key ctx for flow %d.", flow_id); + goto fail_ctx; + } + + list_head_init(&e->next); + e->flow_id = flow_id; + e->ctx = ctx; + clock_gettime(PTHREAD_COND_CLOCK, &e->deadline); + e->deadline.tv_sec += REKEY_RESP_TIMEO; + + list_add(&e->next, tbl); + + log_dbg("Re-key request sent for flow %d.", flow_id); + + freebuf(req); + + return; + + fail_ctx: + oap_ctx_free(ctx); + freebuf(req); + fail: + reg_flow_clear_in_flight(flow_id); +} + +/* Worker-only: complete the exchange, install the pending seed. */ +static void rekey_do_complete(struct list_head * tbl, + int flow_id, + buffer_t buf) +{ + struct rekey_ctx * e; + struct name_info info; + struct crypt_sk sk; + uint8_t kbuf[SYMMKEYSZ]; + buffer_t data = BUF_INIT; + buffer_t crt = BUF_INIT; + uint8_t newgen; + + e = rekey_find(tbl, flow_id); + if (e == NULL) { + log_dbg("Stale re-key RESPONSE for flow %d.", flow_id); + return; + } + + /* A concurrent responder already parked a seed; don't overwrite. */ + if (reg_flow_rekey_pending(flow_id)) { + log_dbg("Re-key already pending for flow %d.", flow_id); + goto finish; + } + + if (rekey_name_info(flow_id, &info) < 0) { + log_err("Failed to get name info to re-key flow %d.", flow_id); + goto finish; + } + + sk.key = kbuf; + + reg_flow_get_peer_crt(flow_id, &crt); + + /* oap_cli_complete frees the ctx on every path. */ + if (oap_cli_complete(e->ctx, &info, buf, &data, &sk, &crt, NULL) < 0) { + log_warn("Failed to complete re-key for flow %d.", flow_id); + e->ctx = NULL; + goto finish_clear; + } + + e->ctx = NULL; + + if (data.len != 1) { + log_warn("Re-key reply malformed for flow %d.", flow_id); + goto finish_clear; + } + + newgen = *(uint8_t *) data.data; + + if (newgen >= 16) { + log_warn("Re-key gen %u out of range for flow %d.", + newgen, flow_id); + goto finish_clear; + } + + if (reg_flow_store_pending(flow_id, kbuf, newgen, true) < 0) + log_warn("Flow %d gone during re-key.", flow_id); + else + reg_notify_flow(flow_id, FLOW_UPD); + + log_dbg("Re-key completed for flow %d (gen %u).", flow_id, newgen); + + finish_clear: + crypt_secure_clear(kbuf, SYMMKEYSZ); + freebuf(data); + finish: + freebuf(crt); + rekey_drop(e); + reg_flow_clear_in_flight(flow_id); +} + +/* Worker-only: reap entries whose RESPONSE never arrived. */ +static void rekey_reap_expired(struct list_head * tbl) +{ + struct list_head * p; + struct list_head * h; + struct timespec now; + + clock_gettime(PTHREAD_COND_CLOCK, &now); + + list_for_each_safe(p, h, tbl) { + struct rekey_ctx * e = list_entry(p, struct rekey_ctx, next); + if (ts_diff_ns(&e->deadline, &now) > 0) + continue; + + log_warn("Re-key timed out for flow %d.", e->flow_id); + reg_flow_clear_in_flight(e->flow_id); + rekey_drop(e); + } +} + +/* Responder side: process request, install pending seed, send response. */ +static int rekey_respond(struct flow_info * flow, + buffer_t * pk) +{ + struct name_info info; + struct crypt_sk sk; + uint8_t kbuf[SYMMKEYSZ]; + buffer_t rsp = BUF_INIT; + buffer_t upd = BUF_INIT; + buffer_t data = BUF_INIT; + buffer_t crt = BUF_INIT; + uint8_t newgen; + int epoch; + int err; + + epoch = reg_flow_get_epoch(flow->id); + if (epoch < 0) { + log_warn("Re-key for unknown flow %d.", flow->id); + return -EBADF; + } + + /* Collision: we are driving our own exchange; let it win. */ + if (reg_flow_rekey_should_yield(flow->id)) { + log_dbg("Yielding to own re-key for flow %d.", flow->id); + return 0; + } + + if (rekey_name_info(flow->id, &info) < 0) { + log_err("Failed to get name info to re-key flow %d.", flow->id); + return -ENAME; + } + + if (reg_flow_rekey_pending(flow->id)) { + log_dbg("Duplicate re-key request for flow %d.", flow->id); + return 0; + } + + newgen = (uint8_t) ((epoch + 1) & 0x0F); + data.data = &newgen; + data.len = 1; + + sk.key = kbuf; + + reg_flow_get_peer_crt(flow->id, &crt); + + err = oap_srv_process(&info, *pk, &rsp, &data, &sk, true, &crt, NULL); + if (err < 0) { + /* data still points to stack newgen; don't free it. */ + log_err("Failed to process re-key OAP for flow %d.", flow->id); + goto finish; + } + + /* On success oap_srv_process repointed data to client output. */ + freebuf(data); + + if (reg_flow_store_pending(flow->id, kbuf, newgen, false) < 0) { + log_warn("Flow %d gone during re-key.", flow->id); + err = -EBADF; + goto finish; + } + + reg_notify_flow(flow->id, FLOW_UPD); + + if (flow_upd_wrap(&upd, FLOW_UPD_REKEY_RESP, &rsp) == 0) { + pthread_cleanup_push(rk_clear_key, kbuf); + pthread_cleanup_push(rk_freebuf, &rsp); + pthread_cleanup_push(rk_freebuf, &crt); + pthread_cleanup_push(rk_freebuf, &upd); + if (ipcp_flow_update(flow, upd) < 0) + log_err("Failed to send re-key response for flow %d.", + flow->id); + pthread_cleanup_pop(false); + pthread_cleanup_pop(false); + pthread_cleanup_pop(false); + pthread_cleanup_pop(false); + freebuf(upd); + } + + err = 0; + finish: + crypt_secure_clear(kbuf, SYMMKEYSZ); + freebuf(rsp); + freebuf(crt); + + return err; +} + +/* + * Worker-only: re-key a direct (loopback) flow, the exchange runs in-process: + * build a client request, then derive the shared seed, and hand the one seed + * to both apps with RB_REKEY. + */ +static void rekey_do_direct(int flow_id) +{ + struct name_info info; + struct crypt_sk sk; + uint8_t kbuf[SYMMKEYSZ]; + buffer_t req = BUF_INIT; + buffer_t rsp = BUF_INIT; + buffer_t data = BUF_INIT; + buffer_t crt = BUF_INIT; + void * ctx = NULL; + uint8_t newgen; + int epoch; + + epoch = reg_flow_get_epoch(flow_id); + if (epoch < 0) { + log_warn("Re-key for unknown flow %d.", flow_id); + reg_flow_clear_in_flight(flow_id); + return; + } + + if (rekey_name_info(flow_id, &info) < 0) { + log_err("Failed to get name info to re-key flow %d.", flow_id); + reg_flow_clear_in_flight(flow_id); + return; + } + + if (oap_cli_prepare(&ctx, &info, &req, data, true) < 0) { + log_err("Failed to prepare re-key for flow %d.", flow_id); + reg_flow_clear_in_flight(flow_id); + return; + } + + newgen = (uint8_t) ((epoch + 1) & 0x0F); + data.data = &newgen; + data.len = 1; + + sk.key = kbuf; + + reg_flow_get_peer_crt(flow_id, &crt); + + if (oap_srv_process(&info, req, &rsp, &data, &sk, true, + &crt, NULL) < 0) { + /* data still points to stack newgen; don't free it. */ + log_err("Failed to process re-key OAP for flow %d.", flow_id); + reg_flow_clear_in_flight(flow_id); + goto out; + } + + /* On success oap_srv_process repointed data to its output. */ + freebuf(data); + + if (reg_flow_store_pending_direct(flow_id, kbuf, newgen) < 0) { + log_warn("Flow %d gone during re-key.", flow_id); + reg_flow_clear_in_flight(flow_id); + goto out; + } + + reg_notify_flow_peers(flow_id, FLOW_UPD); + + log_dbg("Re-key completed (direct) for flow %d (gen %u).", + flow_id, newgen); + out: + crypt_secure_clear(kbuf, SYMMKEYSZ); + oap_ctx_free(ctx); + freebuf(req); + freebuf(rsp); + freebuf(crt); +} + +/* Route one snapshot entry to the wire or in-process re-key path. */ +static void rekey_dispatch(struct list_head * tbl, + const struct rekey_info * ri) +{ + if (ri->direct) + rekey_do_direct(ri->flow_id); + else + rekey_do_initiate(tbl, ri->flow_id, ri->n_1_pid); +} + +static int flow_update_arr(struct flow_info * flow, + buffer_t * pk) +{ + uint8_t type; + bool is_req; + + if (pk->len < 1) + return -EINVAL; + + type = pk->data[0]; + + switch (type) { + case FLOW_UPD_REKEY_REQ: + is_req = true; + break; + case FLOW_UPD_REKEY_RESP: + is_req = false; + break; + default: + log_warn("Unknown flow update type %u.", type); + return -EINVAL; + } + + /* Drop floods/spoofs before allocating a worker event. */ + if (!reg_flow_rekey_arr_admit(flow->id, flow->n_1_pid, is_req)) + return 0; + + /* Strip the type byte, keeping the malloc base for hand-off. */ + memmove(pk->data, pk->data + 1, pk->len - 1); + pk->len -= 1; + + /* Defer to worker; an inline RESP send deadlocks loopback. */ + if (is_req) + rekey_post_req(flow->id, flow->n_1_pid, pk); + else + rekey_post_resp(flow->id, pk); + + return 0; +} + +static int flow_update(struct flow_info * flow, + uid_t uid, + pid_t cpid, + bool rekey, + struct crypt_sk * sk, + bool * has_key, + bool * initiator) +{ + uint8_t seed[SYMMKEYSZ]; + uint8_t epoch; + int rc; + + *has_key = false; + *initiator = false; + + if (rekey) { + pid_t n_1_pid; + + if (!reg_flow_owned_by(flow->id, uid)) + return -EPERM; + + /* Direct flows re-key in-process; no lower IPCP carrier. */ + if (reg_flow_is_direct(flow->id)) { + if (reg_flow_rekey_begin(flow->id)) + rekey_post_direct(flow->id); + + return 0; + } + + /* Watermark re-key: the app can't know its lower IPCP. */ + n_1_pid = reg_flow_get_n_1_pid(flow->id); + if (n_1_pid <= 0) + return 0; + + /* One exchange per flow; the latch arbitrates collisions. */ + if (reg_flow_rekey_begin(flow->id)) + rekey_post_init(flow->id, n_1_pid); + + return 0; + } + + rc = reg_flow_take_pending(flow->id, uid, cpid, seed, &epoch, + initiator); + if (rc == -EPERM) + return -EPERM; + + if (rc != 0) + return 0; + + memcpy(sk->key, seed, SYMMKEYSZ); + sk->epoch = epoch; + *has_key = true; + + crypt_secure_clear(seed, SYMMKEYSZ); + + log_dbg("Delivered re-key seed for flow %d (gen %u).", + flow->id, epoch); + + return 0; +} + +static void rekey_table_cleanup(void * o) +{ + struct list_head * tbl = o; + struct list_head * p; + struct list_head * h; + + list_for_each_safe(p, h, tbl) { + struct rekey_ctx * e = list_entry(p, struct rekey_ctx, next); + rekey_drop(e); + } +} + +static struct rekey_evt * rekey_event_wait(const struct timespec * dl) +{ + struct rekey_evt * evt = NULL; + int ret = 0; + + pthread_mutex_lock(&irmd.rk.mtx); + pthread_cleanup_push(__cleanup_mutex_unlock, &irmd.rk.mtx); + + while (list_is_empty(&irmd.rk.inbox) && ret != -ETIMEDOUT) + ret = -pthread_cond_timedwait(&irmd.rk.cond, &irmd.rk.mtx, dl); + + if (!list_is_empty(&irmd.rk.inbox)) { + evt = list_first_entry(&irmd.rk.inbox, struct rekey_evt, next); + list_del(&evt->next); + } + + pthread_cleanup_pop(true); + + return evt; +} + +static struct timespec rekey_deadline(struct list_head * tbl, + struct timespec next) +{ + struct timespec deadline = next; + struct list_head * p; + + list_for_each(p, tbl) { + struct rekey_ctx * e; + e = list_entry(p, struct rekey_ctx, next); + if (ts_diff_ns(&e->deadline, &deadline) < 0) + deadline = e->deadline; + } + + return deadline; +} + +static void rekey_handle_evt(struct list_head * tbl, + struct rekey_evt * evt) +{ + struct flow_info rinfo; + + pthread_cleanup_push(rk_free_evt, evt); + + switch (evt->type) { + case REKEY_INIT: + rekey_do_initiate(tbl, evt->flow_id, evt->n_1_pid); + break; + case REKEY_REQ: + memset(&rinfo, 0, sizeof(rinfo)); + rinfo.id = evt->flow_id; + rinfo.n_1_pid = evt->n_1_pid; + rekey_respond(&rinfo, &evt->buf); + reg_flow_rekey_arr_done(evt->flow_id, true); + break; + case REKEY_RESP: + rekey_do_complete(tbl, evt->flow_id, evt->buf); + reg_flow_rekey_arr_done(evt->flow_id, false); + break; + case REKEY_DIRECT: + rekey_do_direct(evt->flow_id); + break; + default: + break; + } + + pthread_cleanup_pop(true); +} + +/* On the periodic tick, dispatch all flows due for re-keying. */ +static void rekey_run_periodic(struct list_head * tbl, + struct timespec * next) +{ + struct rekey_info snap[REKEY_BATCH]; + struct timespec now; + int n; + int i; + + clock_gettime(PTHREAD_COND_CLOCK, &now); + + if (ts_diff_ns(next, &now) > 0) + return; + + n = reg_flow_snapshot_rekey_due(snap, REKEY_BATCH); + for (i = 0; i < n; ++i) + rekey_dispatch(tbl, &snap[i]); + + clock_gettime(PTHREAD_COND_CLOCK, next); + next->tv_sec += OAP_REKEY_TIMER; +} + +/* + * Single worker owning all in-flight Tier-2 re-keys. It drains the + * inbox, runs the periodic snapshot, and reaps timed-out exchanges. + * The table is touched only here, so it needs no lock. + */ +static void * rekey_worker(void * o) +{ + struct list_head table; + struct timespec next; + + (void) o; + + list_head_init(&table); + + clock_gettime(PTHREAD_COND_CLOCK, &next); + next.tv_sec += OAP_REKEY_TIMER; + + pthread_cleanup_push(rekey_table_cleanup, &table); + + while (true) { + struct rekey_evt * evt; + struct timespec deadline; + + deadline = rekey_deadline(&table, next); + + evt = rekey_event_wait(&deadline); + + if (evt != NULL) + rekey_handle_evt(&table, evt); + + rekey_run_periodic(&table, &next); + + rekey_reap_expired(&table); + } + + pthread_cleanup_pop(true); + + return (void *) 0; +} + static void * acceptloop(void * o) { int csockfd; @@ -1308,25 +2279,34 @@ static void * acceptloop(void * o) return (void *) 0; } -static void free_msg(void * o) +static void __cleanup_irm_msg(void * o) { irm_msg__free_unpacked((irm_msg_t *) o, NULL); } -static irm_msg_t * do_command_msg(irm_msg_t * msg) +static irm_msg_t * do_command_msg(irm_msg_t * msg, + int fd) { - struct ipcp_config conf; - struct ipcp_info ipcp; - struct flow_info flow; - struct proc_info proc; - struct name_info name; - struct timespec * abstime; - struct timespec max = TIMESPEC_INIT_MS(FLOW_ALLOC_TIMEOUT); - struct timespec now; - struct timespec ts = TIMESPEC_INIT_S(0); /* static analysis */ - int res; - irm_msg_t * ret_msg; - buffer_t data; + struct ipcp_config conf; + struct ipcp_info ipcp; + struct flow_info flow; + struct proc_info proc; + struct name_info name; + struct crypt_sk sk; + uint8_t kbuf[SYMMKEYSZ]; /* stack buffer for OAP */ + uint8_t * hbuf = NULL; /* heap copy for response */ + struct timespec * abstime; + struct timespec max = TIMESPEC_INIT_MS(FLOW_ALLOC_TIMEOUT); + struct timespec now; + struct timespec ts = TIMESPEC_INIT_S(0); /* static analysis */ + int res; + bool has_key = false; + bool initiator = false; + uid_t uid; + gid_t gid; + pid_t cpid; + irm_msg_t * ret_msg; + buffer_t data; memset(&flow, 0, sizeof(flow)); @@ -1351,7 +2331,7 @@ static irm_msg_t * do_command_msg(irm_msg_t * msg) ret_msg->code = IRM_MSG_CODE__IRM_REPLY; - pthread_cleanup_push(free_msg, ret_msg); + pthread_cleanup_push(__cleanup_irm_msg, ret_msg); switch (msg->code) { case IRM_MSG_CODE__IRM_CREATE_IPCP: @@ -1380,7 +2360,7 @@ static irm_msg_t * do_command_msg(irm_msg_t * msg) res = disconnect_ipcp(msg->pid, msg->dst, msg->comp); break; case IRM_MSG_CODE__IRM_BIND_PROGRAM: - /* Make exec NULL terminated instead of empty string terminated */ + /* Terminate with NULL instead of "" */ free(msg->exec[msg->n_exec - 1]); msg->exec[msg->n_exec - 1] = NULL; res = bind_program(msg->exec, msg->name, msg->opts); @@ -1391,7 +2371,11 @@ static irm_msg_t * do_command_msg(irm_msg_t * msg) case IRM_MSG_CODE__IRM_PROC_ANNOUNCE: proc.pid = msg->pid; strcpy(proc.prog, msg->prog); - res = proc_announce(&proc); + res = get_peer_ids(fd, &proc.uid, &proc.gid, NULL); + if (res < 0) + log_err("Failed to get UID/GID for pid %d.", msg->pid); + else + res = proc_announce(&proc); break; case IRM_MSG_CODE__IRM_PROC_EXIT: res = proc_exit(msg->pid); @@ -1406,8 +2390,7 @@ static irm_msg_t * do_command_msg(irm_msg_t * msg) res = list_ipcps(&ret_msg->ipcps, &ret_msg->n_ipcps); break; case IRM_MSG_CODE__IRM_CREATE_NAME: - strcpy(name.name, msg->names[0]->name); - name.pol_lb = msg->names[0]->pol_lb; + name = name_info_msg_to_s(msg->name_info); res = name_create(&name); break; case IRM_MSG_CODE__IRM_DESTROY_NAME: @@ -1423,18 +2406,37 @@ static irm_msg_t * do_command_msg(irm_msg_t * msg) res = name_unreg(msg->name, msg->pid); break; case IRM_MSG_CODE__IRM_FLOW_ACCEPT: + tpm_wait_work(irmd.tpm); data.len = msg->pk.len; data.data = msg->pk.data; msg->has_pk = false; assert(data.len > 0 ? data.data != NULL : data.data == NULL); flow = flow_info_msg_to_s(msg->flow_info); - res = flow_accept(&flow, &data, abstime); - if (res == 0) { - ret_msg->flow_info = flow_info_s_to_msg(&flow); - ret_msg->has_symmkey = data.len != 0; - ret_msg->symmkey.data = data.data; - ret_msg->symmkey.len = data.len; + sk.key = kbuf; + res = flow_accept(&flow, &data, abstime, &sk); + if (res != 0) + break; + + ret_msg->flow_info = flow_info_s_to_msg(&flow); + ret_msg->has_pk = data.len != 0; + ret_msg->pk.data = data.data; + ret_msg->pk.len = data.len; + ret_msg->has_cipher_nid = true; + ret_msg->cipher_nid = sk.nid; + if (sk.nid == NID_undef) + break; + + hbuf = malloc(SYMMKEYSZ); + if (hbuf == NULL) { + log_err("Failed to malloc key buf"); + res = -ENOMEM; + break; } + + memcpy(hbuf, kbuf, SYMMKEYSZ); + ret_msg->sym_key.data = hbuf; + ret_msg->sym_key.len = SYMMKEYSZ; + ret_msg->has_sym_key = true; break; case IRM_MSG_CODE__IRM_FLOW_ALLOC: data.len = msg->pk.len; @@ -1443,13 +2445,31 @@ static irm_msg_t * do_command_msg(irm_msg_t * msg) assert(data.len > 0 ? data.data != NULL : data.data == NULL); flow = flow_info_msg_to_s(msg->flow_info); abstime = abstime == NULL ? &max : abstime; - res = flow_alloc(&flow, msg->dst, &data, abstime); - if (res == 0) { - ret_msg->flow_info = flow_info_s_to_msg(&flow); - ret_msg->has_symmkey = data.len != 0; - ret_msg->symmkey.data = data.data; - ret_msg->symmkey.len = data.len; + sk.key = kbuf; + res = flow_alloc(msg->dst, &flow, &data, abstime, &sk); + if (res != 0) + break; + + ret_msg->flow_info = flow_info_s_to_msg(&flow); + ret_msg->has_pk = data.len != 0; + ret_msg->pk.data = data.data; + ret_msg->pk.len = data.len; + ret_msg->has_cipher_nid = true; + ret_msg->cipher_nid = sk.nid; + if (sk.nid == NID_undef) + break; + + hbuf = malloc(SYMMKEYSZ); + if (hbuf == NULL) { + log_err("Failed to malloc key buf"); + res = -ENOMEM; + break; } + + memcpy(hbuf, kbuf, SYMMKEYSZ); + ret_msg->sym_key.data = hbuf; + ret_msg->sym_key.len = SYMMKEYSZ; + ret_msg->has_sym_key = true; break; case IRM_MSG_CODE__IRM_FLOW_JOIN: assert(msg->pk.len == 0 && msg->pk.data == NULL); @@ -1457,7 +2477,7 @@ static irm_msg_t * do_command_msg(irm_msg_t * msg) abstime = abstime == NULL ? &max : abstime; res = flow_join(&flow, msg->dst, abstime); if (res == 0) - ret_msg->flow_info = flow_info_s_to_msg(&flow); + ret_msg->flow_info = flow_info_s_to_msg(&flow); break; case IRM_MSG_CODE__IRM_FLOW_DEALLOC: flow = flow_info_msg_to_s(msg->flow_info); @@ -1488,6 +2508,51 @@ static irm_msg_t * do_command_msg(irm_msg_t * msg) flow = flow_info_msg_to_s(msg->flow_info); res = flow_alloc_reply(&flow, msg->response, &data); break; + case IRM_MSG_CODE__IPCP_FLOW_UPDATE_ARR: + data.len = msg->pk.len; + data.data = msg->pk.data; + msg->pk.data = NULL; /* pass data */ + msg->pk.len = 0; + flow = flow_info_msg_to_s(msg->flow_info); + res = flow_update_arr(&flow, &data); + freebuf(data); + break; + case IRM_MSG_CODE__IRM_FLOW_UPDATE: + flow = flow_info_msg_to_s(msg->flow_info); + if (get_peer_ids(fd, &uid, &gid, &cpid) < 0) { + res = -EPERM; + break; + } + + if (cpid <= 0) /* non-Linux: fall back to asserted pid */ + cpid = flow.n_pid; + + sk.key = kbuf; + res = flow_update(&flow, uid, cpid, msg->rekey, &sk, &has_key, + &initiator); + if (res != 0) + break; + + ret_msg->flow_info = flow_info_s_to_msg(&flow); + if (!has_key) + break; + + hbuf = malloc(SYMMKEYSZ); + if (hbuf == NULL) { + log_err("Failed to malloc key buf"); + res = -ENOMEM; + break; + } + + memcpy(hbuf, kbuf, SYMMKEYSZ); + ret_msg->sym_key.data = hbuf; + ret_msg->sym_key.len = SYMMKEYSZ; + ret_msg->has_sym_key = true; + ret_msg->has_generation = true; + ret_msg->generation = sk.epoch; + ret_msg->has_rk_initiator = true; + ret_msg->rk_initiator = initiator; + break; default: log_err("Don't know that message code."); res = -1; @@ -1502,20 +2567,30 @@ static irm_msg_t * do_command_msg(irm_msg_t * msg) else ret_msg->result = res; + crypt_secure_clear(kbuf, SYMMKEYSZ); + return ret_msg; } +/* Wipe the session key from a reply before its buffers are freed. */ +static void clear_msg_key(irm_msg_t * msg) +{ + if (msg != NULL && msg->has_sym_key) + crypt_secure_clear(msg->sym_key.data, msg->sym_key.len); +} + static void * mainloop(void * o) { - int sfd; - irm_msg_t * msg; - buffer_t buffer; + int sfd; + irm_msg_t * msg; + buffer_t buffer; (void) o; while (true) { irm_msg_t * ret_msg; struct cmd * cmd; + bool had_key; pthread_mutex_lock(&irmd.cmd_lock); @@ -1540,12 +2615,12 @@ static void * mainloop(void * o) continue; } - tpm_dec(irmd.tpm); + tpm_begin_work(irmd.tpm); pthread_cleanup_push(__cleanup_close_ptr, &sfd); - pthread_cleanup_push(free_msg, msg); + pthread_cleanup_push(__cleanup_irm_msg, msg); - ret_msg = do_command_msg(msg); + ret_msg = do_command_msg(msg, sfd); pthread_cleanup_pop(true); pthread_cleanup_pop(false); @@ -1556,12 +2631,12 @@ static void * mainloop(void * o) } if (ret_msg->result == -EPIPE) { - log_dbg("Terminated command: application closed socket."); + log_dbg("Terminated command: remote closed socket."); goto fail; } if (ret_msg->result == -EIRMD) { - log_dbg("Terminated command: IRMd not in running state."); + log_dbg("Terminated command: IRMd not running."); goto fail; } @@ -1579,6 +2654,9 @@ static void * mainloop(void * o) irm_msg__pack(ret_msg, buffer.data); + had_key = ret_msg->has_sym_key; + clear_msg_key(ret_msg); + irm_msg__free_unpacked(ret_msg, NULL); pthread_cleanup_push(__cleanup_close_ptr, &sfd); @@ -1593,57 +2671,27 @@ static void * mainloop(void * o) strerror(errno)); } + if (had_key) + crypt_secure_clear(buffer.data, buffer.len); + pthread_cleanup_pop(true); pthread_cleanup_pop(true); - tpm_inc(irmd.tpm); + tpm_end_work(irmd.tpm); continue; fail: + clear_msg_key(ret_msg); irm_msg__free_unpacked(ret_msg, NULL); fail_msg: close(sfd); - tpm_inc(irmd.tpm); + tpm_end_work(irmd.tpm); continue; } return (void *) 0; } -static void irm_fini(void) -{ -#ifdef HAVE_FUSE - struct timespec wait = TIMESPEC_INIT_MS(1); - int retries = 5; -#endif - if (irmd_get_state() != IRMD_NULL) - log_warn("Unsafe destroy."); - - tpm_destroy(irmd.tpm); - - close(irmd.sockfd); - - if (unlink(IRM_SOCK_PATH)) - log_dbg("Failed to unlink %s.", IRM_SOCK_PATH); - - if (irmd.rdrb != NULL) - shm_rdrbuff_destroy(irmd.rdrb); - - if (irmd.lf != NULL) - lockfile_destroy(irmd.lf); - - pthread_mutex_destroy(&irmd.cmd_lock); - pthread_cond_destroy(&irmd.cmd_cond); - pthread_rwlock_destroy(&irmd.state_lock); - -#ifdef HAVE_FUSE - while (rmdir(FUSE_PREFIX) < 0 && retries-- > 0) - nanosleep(&wait, NULL); - if (retries < 0) - log_err("Failed to remove " FUSE_PREFIX); -#endif -} - #ifdef HAVE_FUSE static void destroy_mount(char * mnt) { @@ -1672,7 +2720,7 @@ static void destroy_mount(char * mnt) static int ouroboros_reset(void) { - shm_rdrbuff_purge(); + ssm_pool_gspp_purge(); lockfile_destroy(irmd.lf); return 0; @@ -1693,10 +2741,8 @@ static void cleanup_pid(pid_t pid) } destroy_mount(mnt); - -#else - (void) pid; #endif + ssm_pool_reclaim_orphans(irmd.gspp, pid); } void * irm_sanitize(void * o) @@ -1719,10 +2765,89 @@ void * irm_sanitize(void * o) return (void *) 0; } +static int irm_load_store(char * dpath, + bool anchor) +{ + struct stat st; + struct dirent * dent; + DIR * dir; + void * crt; + int ret; + + if (stat(dpath, &st) == -1) { + log_dbg("Store directory %s not found.", dpath); + return 0; + } + + if (!S_ISDIR(st.st_mode)) { + log_err("%s is not a directory.", dpath); + goto fail_dir; + } + + /* loop through files in directory and load certificates */ + dir = opendir(dpath); + if (dir == NULL) { + log_err("Failed to open %s.", dpath); + goto fail_dir; + } + + while ((dent = readdir(dir)) != NULL) { + char path[NAME_PATH_SIZE + 1]; + + if (strcmp(dent->d_name, ".") == 0 || + strcmp(dent->d_name, "..") == 0) + continue; + + snprintf(path, sizeof(path), "%s/%s", dpath, + dent->d_name); + + if (stat(path, &st) == -1) { + log_dbg("Failed to stat %s.", path); + continue; + } + + if (!S_ISREG(st.st_mode)) { + log_dbg("%s is not a regular file.", path); + goto fail_file; + } + + if (crypt_load_crt_file(path, &crt) < 0) { + log_err("Failed to load certificate from %s.", path); + goto fail_file; + } + + ret = anchor ? oap_auth_add_ca_crt(crt) + : oap_auth_add_chain_crt(crt); + if (ret < 0) { + log_err("Failed to add certificate from %s to store.", + path); + goto fail_crt_add; + } + + log_dbg("Loaded certificate: %s.", path); + + crypt_free_crt(crt); + } + + closedir(dir); + + log_info("Loaded certificates from %s.", dpath); + + return 0; + + fail_crt_add: + crypt_free_crt(crt); + fail_file: + closedir(dir); + fail_dir: + return -1; +} static int irm_init(void) { struct stat st; + struct group * grp; + gid_t gid; pthread_condattr_t cattr; #ifdef HAVE_FUSE mode_t mask; @@ -1790,6 +2915,29 @@ static int irm_init(void) list_head_init(&irmd.cmds); + if (pthread_mutex_init(&irmd.rk.mtx, NULL)) { + log_err("Failed to initialize mutex."); + goto fail_rk_mtx; + } + + if (pthread_condattr_init(&cattr)) { + log_err("Failed to initialize condattr."); + goto fail_rk_mtx; + } + +#ifndef __APPLE__ + pthread_condattr_setclock(&cattr, PTHREAD_COND_CLOCK); +#endif + if (pthread_cond_init(&irmd.rk.cond, &cattr)) { + log_err("Failed to initialize condvar."); + pthread_condattr_destroy(&cattr); + goto fail_rk_cond; + } + + pthread_condattr_destroy(&cattr); + + list_head_init(&irmd.rk.inbox); + if (stat(SOCK_PATH, &st) == -1) { if (mkdir(SOCK_PATH, 0777)) { log_err("Failed to create sockets directory."); @@ -1808,17 +2956,45 @@ static int irm_init(void) goto fail_sock_path; } - if ((irmd.rdrb = shm_rdrbuff_create()) == NULL) { - log_err("Failed to create rdrbuff."); - goto fail_rdrbuff; + grp = getgrnam("ouroboros"); + if (grp == NULL) { + log_warn("ouroboros group not found, using gid %d.", getgid()); + gid = getgid(); + } else { + gid = grp->gr_gid; } + irmd.gspp = ssm_pool_create(getuid(), gid); + if (irmd.gspp == NULL) { + log_err("Failed to create GSPP."); + goto fail_pool; + } + + if (ssm_pool_mlock(irmd.gspp) < 0) + log_warn("Failed to mlock pool."); + irmd.tpm = tpm_create(IRMD_MIN_THREADS, IRMD_ADD_THREADS, mainloop, NULL); if (irmd.tpm == NULL) { log_err("Failed to greate thread pool."); goto fail_tpm_create; } + + if (oap_auth_init() < 0) { + log_err("Failed to initialize OAP module."); + goto fail_oap; + } + + if (irm_load_store(OUROBOROS_CA_CRT_DIR, true) < 0) { + log_err("Failed to load CA certificates."); + goto fail_load_store; + } + + if (irm_load_store(OUROBOROS_CHAIN_DIR, false) < 0) { + log_err("Failed to load intermediate certificates."); + goto fail_load_store; + } + #ifdef HAVE_FUSE mask = umask(0); @@ -1844,6 +3020,8 @@ static int irm_init(void) gcry_control(GCRYCTL_INITIALIZATION_FINISHED); #endif + irmd_set_state(IRMD_INIT); + return 0; #ifdef HAVE_LIBGCRYPT @@ -1851,15 +3029,22 @@ static int irm_init(void) #ifdef HAVE_FUSE rmdir(FUSE_PREFIX); #endif - tpm_destroy(irmd.tpm); #endif + fail_load_store: + oap_auth_fini(); + fail_oap: + tpm_destroy(irmd.tpm); fail_tpm_create: - shm_rdrbuff_destroy(irmd.rdrb); - fail_rdrbuff: + ssm_pool_destroy(irmd.gspp); + fail_pool: close(irmd.sockfd); fail_sock_path: unlink(IRM_SOCK_PATH); fail_stat: + pthread_cond_destroy(&irmd.rk.cond); + fail_rk_cond: + pthread_mutex_destroy(&irmd.rk.mtx); + fail_rk_mtx: pthread_cond_destroy(&irmd.cmd_cond); fail_cmd_cond: pthread_mutex_destroy(&irmd.cmd_lock); @@ -1872,6 +3057,72 @@ static int irm_init(void) return -1; } +static void irm_fini(void) +{ + struct list_head * p; + struct list_head * h; +#ifdef HAVE_FUSE + struct timespec wait = TIMESPEC_INIT_MS(1); + int retries = 5; +#endif + if (irmd_get_state() != IRMD_INIT) + log_warn("Unsafe destroy."); + + oap_auth_fini(); + + tpm_destroy(irmd.tpm); + + close(irmd.sockfd); + + if (unlink(IRM_SOCK_PATH)) + log_dbg("Failed to unlink %s.", IRM_SOCK_PATH); + + ssm_pool_destroy(irmd.gspp); + + if (irmd.lf != NULL) + lockfile_destroy(irmd.lf); + + pthread_mutex_lock(&irmd.cmd_lock); + + list_for_each_safe(p, h, &irmd.cmds) { + struct cmd * cmd = list_entry(p, struct cmd, next); + list_del(&cmd->next); + close(cmd->fd); + free(cmd); + } + + pthread_mutex_unlock(&irmd.cmd_lock); + + pthread_mutex_lock(&irmd.rk.mtx); + + list_for_each_safe(p, h, &irmd.rk.inbox) { + struct rekey_evt * evt; + evt = list_entry(p, struct rekey_evt, next); + list_del(&evt->next); + freebuf(evt->buf); + free(evt); + } + + pthread_mutex_unlock(&irmd.rk.mtx); + + pthread_mutex_destroy(&irmd.cmd_lock); + pthread_cond_destroy(&irmd.cmd_cond); + pthread_mutex_destroy(&irmd.rk.mtx); + pthread_cond_destroy(&irmd.rk.cond); + pthread_rwlock_destroy(&irmd.state_lock); + +#ifdef HAVE_FUSE + while (rmdir(FUSE_PREFIX) < 0 && retries-- > 0) + nanosleep(&wait, NULL); + + if (retries < 0) + log_err("Failed to remove " FUSE_PREFIX); +#endif + assert(list_is_empty(&irmd.cmds)); + + irmd.state = IRMD_NULL; +} + static void usage(void) { printf("Usage: irmd \n" @@ -1885,28 +3136,36 @@ static void usage(void) static int irm_start(void) { + irmd_set_state(IRMD_RUNNING); + if (tpm_start(irmd.tpm)) goto fail_tpm_start; - irmd_set_state(IRMD_RUNNING); - if (pthread_create(&irmd.irm_sanitize, NULL, irm_sanitize, NULL)) goto fail_irm_sanitize; if (pthread_create(&irmd.acceptor, NULL, acceptloop, NULL)) goto fail_acceptor; + if (OAP_REKEY_TIMER > 0) { + if (pthread_create(&irmd.rk.worker, NULL, rekey_worker, NULL)) + goto fail_rekey_worker; + } + log_info("Ouroboros IPC Resource Manager daemon started..."); return 0; + fail_rekey_worker: + pthread_cancel(irmd.acceptor); + pthread_join(irmd.acceptor, NULL); fail_acceptor: pthread_cancel(irmd.irm_sanitize); pthread_join(irmd.irm_sanitize, NULL); fail_irm_sanitize: - irmd_set_state(IRMD_NULL); tpm_stop(irmd.tpm); fail_tpm_start: + irmd_set_state(IRMD_INIT); return -1; } @@ -1939,6 +3198,11 @@ static void irm_sigwait(sigset_t sigset) static void irm_stop(void) { + if (OAP_REKEY_TIMER > 0) { + pthread_cancel(irmd.rk.worker); + pthread_join(irmd.rk.worker, NULL); + } + pthread_cancel(irmd.acceptor); pthread_cancel(irmd.irm_sanitize); @@ -1947,7 +3211,7 @@ static void irm_stop(void) tpm_stop(irmd.tpm); - irmd_set_state(IRMD_NULL); + irmd_set_state(IRMD_INIT); } static void irm_argparse(int argc, @@ -1964,10 +3228,8 @@ static void irm_argparse(int argc, argc--; argv++; } else if (strcmp(*argv, "--version") == 0) { - printf("Ouroboros version %d.%d.%d\n", - OUROBOROS_VERSION_MAJOR, - OUROBOROS_VERSION_MINOR, - OUROBOROS_VERSION_PATCH); + printf("Ouroboros version %s\n", + OUROBOROS_VERSION_STRING); exit(EXIT_SUCCESS); #ifdef HAVE_TOML } else if (strcmp (*argv, "--config") == 0) { @@ -1998,8 +3260,8 @@ static void * kill_dash_nine(void * o) slept += intv; } - log_dbg("I am become Death, destroyer of hung processes."); - + log_dbg("I guess I’ll have to shut you down for good this time,"); + log_dbg("already tried a SIGQUIT, so now it’s KILL DASH 9."); #ifdef IRMD_KILL_ALL_PROCESSES reg_kill_all_proc(SIGKILL); nanosleep(&ts, NULL); @@ -2058,11 +3320,18 @@ int main(int argc, if (geteuid() != 0) { printf("IPC Resource Manager must be run as root.\n"); - exit(EXIT_FAILURE); + goto fail_irm_init; } - if (irm_init() < 0) + if (crypt_secure_malloc_init(IRMD_SECMEM_MAX) < 0) { + log_err("Failed to initialize secure memory allocation."); + goto fail_secmem; + } + + if (irm_init() < 0) { + log_err("Failed to initialize IRMd."); goto fail_irm_init; + } if (reg_init() < 0) { log_err("Failed to initialize registry."); @@ -2071,12 +3340,15 @@ int main(int argc, pthread_sigmask(SIG_BLOCK, &sigset, NULL); - if (irm_start() < 0) + if (irm_start() < 0) { + log_err("Failed to start IRMd."); goto fail_irm_start; + } #ifdef HAVE_TOML if (irm_configure(irmd.cfg_file) < 0) { - irmd_set_state(IRMD_NULL); + log_err("Failed to load IRMd configuration."); + irmd_set_state(IRMD_SHUTDOWN); ret = EXIT_FAILURE; } #endif @@ -2094,6 +3366,10 @@ int main(int argc, irm_fini(); + crypt_secure_malloc_fini(); + + crypt_cleanup(); + log_info("Ouroboros IPC Resource Manager daemon exited. Bye."); log_fini(); @@ -2105,5 +3381,8 @@ int main(int argc, fail_reg: irm_fini(); fail_irm_init: + crypt_secure_malloc_fini(); + crypt_cleanup(); + fail_secmem: exit(EXIT_FAILURE); } diff --git a/src/irmd/oap.h b/src/irmd/oap.h new file mode 100644 index 00000000..86f11e21 --- /dev/null +++ b/src/irmd/oap.h @@ -0,0 +1,84 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Ouroboros Allocation Protocol (OAP) Component + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#ifndef OUROBOROS_IRMD_OAP_H +#define OUROBOROS_IRMD_OAP_H + +#include <ouroboros/crypt.h> +#include <ouroboros/flow.h> +#include <ouroboros/name.h> +#include <ouroboros/utils.h> + +#include <stdbool.h> + +/* OAP authentication state (in oap/auth.c) */ +int oap_auth_init(void); + +void oap_auth_fini(void); + +int oap_auth_add_ca_crt(void * crt); + +int oap_auth_add_chain_crt(void * crt); + +/* +* Prepare OAP request header for server, returns context +* Passes client data for srv, returns srv data for client +* rekey forces ephemeral server-encap KEX (no client-encap; preserves FS/PCS) +*/ +int oap_cli_prepare(void ** ctx, + const struct name_info * info, + buffer_t * req_buf, + buffer_t data, + bool rekey); + +/* + * Server processes header, creates response header, returns secret key. + * data is in/out: input=srv data to send, output=cli data received. + * rekey drops the cert and verifies against cached_crt; peer_crt (or NULL) + * receives a copy of the peer cert to cache at the initial handshake. + */ +int oap_srv_process(const struct name_info * info, + buffer_t req_buf, + buffer_t * rsp_buf, + buffer_t * data, + struct crypt_sk * sk, + bool rekey, + const buffer_t * cached_crt, + buffer_t * peer_crt); + +/* + * Complete OAP, returns secret key and server data, frees ctx. + * cached_crt verifies a cert-less re-key; peer_crt (or NULL) receives a + * copy of the peer cert to cache at the initial handshake. + */ +int oap_cli_complete(void * ctx, + const struct name_info * info, + buffer_t rsp_buf, + buffer_t * data, + struct crypt_sk * sk, + const buffer_t * cached_crt, + buffer_t * peer_crt); + +/* Free OAP state (on failure before complete) */ +void oap_ctx_free(void * ctx); + +#endif /* OUROBOROS_IRMD_OAP_H */ diff --git a/src/irmd/oap/auth.c b/src/irmd/oap/auth.c new file mode 100644 index 00000000..f70f9df1 --- /dev/null +++ b/src/irmd/oap/auth.c @@ -0,0 +1,558 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * OAP - Authentication, replay detection, and validation + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#if defined(__linux__) || defined(__CYGWIN__) + #define _DEFAULT_SOURCE +#else + #define _POSIX_C_SOURCE 200809L +#endif + +#define OUROBOROS_PREFIX "irmd/oap" + +#include <ouroboros/crypt.h> +#include <ouroboros/endian.h> +#include <ouroboros/errno.h> +#include <ouroboros/logs.h> +#include <ouroboros/pthread.h> +#include <ouroboros/time.h> + +#include "config.h" + +#include "auth.h" +#include "hdr.h" + +#include <assert.h> +#include <stdlib.h> +#include <string.h> + +/* + * Replay cache: three timestamp-generation hash buckets. A header's bucket + * is gen(T) = T / OAP_REPLAY_TIMER, taken mod 3. Staleness bounds a valid T + * to generations {G-1, G, G+1} (G is now's generation; a within-slack future + * stamp can reach G+1), which are distinct mod 3; the aliasing generation + * G-3 is always rejected as too old first. Each bucket is an open-addressed + * hash set whose slots are live iff slot.gen == bucket.gen, so a stale bucket + * clears in O(1) by bumping its gen. Overflow fails closed (reject), never + * evicts, so a flood cannot displace a genuine entry into a replayable state. + */ +#define OAP_REPLAY_GENS 3 + +struct oap_replay_slot { + uint64_t gen; /* live iff == bucket gen; 0 = never used */ + uint64_t ts; + uint8_t id[OAP_ID_SIZE]; +}; + +struct oap_replay_bucket { + uint64_t gen; + size_t count; + struct oap_replay_slot * slots; +}; + +static struct { + struct auth_ctx * ca_ctx; + struct { + size_t mask; /* slots per bucket - 1 */ + size_t cap; /* fail-closed threshold */ + struct oap_replay_bucket bucket[OAP_REPLAY_GENS]; + pthread_mutex_t mtx; + } replay; +} oap_auth; + +/* FNV-1a over id || ts; the table mask reduces it to a slot index. */ +static size_t replay_hash(const uint8_t * id, + uint64_t ts) +{ + uint64_t hh = 14695981039346656037ULL; + size_t i; + + for (i = 0; i < OAP_ID_SIZE; i++) { + hh ^= id[i]; + hh *= 1099511628211ULL; + } + + for (i = 0; i < sizeof(ts); i++) { + hh ^= (uint8_t) (ts >> (i * 8)); + hh *= 1099511628211ULL; + } + + return (size_t) hh; +} + +int oap_auth_init(void) +{ + size_t m = 1; + int i; + + oap_auth.ca_ctx = auth_create_ctx(); + if (oap_auth.ca_ctx == NULL) { + log_err("Failed to create OAP auth context."); + goto fail_ctx; + } + + while (m < (size_t) OAP_REPLAY_MAX * 2) + m <<= 1; + + oap_auth.replay.mask = m - 1; + oap_auth.replay.cap = OAP_REPLAY_MAX; + + for (i = 0; i < OAP_REPLAY_GENS; i++) { + struct oap_replay_bucket * b = &oap_auth.replay.bucket[i]; + b->gen = 0; + b->count = 0; + b->slots = calloc(m, sizeof(*b->slots)); + if (b->slots == NULL) { + log_err("Failed to alloc OAP replay bucket."); + goto fail_bucket; + } + } + + if (pthread_mutex_init(&oap_auth.replay.mtx, NULL)) { + log_err("Failed to init OAP replay mutex."); + goto fail_bucket; + } + + return 0; + + fail_bucket: + for (i = 0; i < OAP_REPLAY_GENS; i++) + free(oap_auth.replay.bucket[i].slots); + + auth_destroy_ctx(oap_auth.ca_ctx); + fail_ctx: + return -1; +} + +void oap_auth_fini(void) +{ + int i; + + pthread_mutex_lock(&oap_auth.replay.mtx); + + for (i = 0; i < OAP_REPLAY_GENS; i++) { + free(oap_auth.replay.bucket[i].slots); + oap_auth.replay.bucket[i].slots = NULL; + } + + pthread_mutex_unlock(&oap_auth.replay.mtx); + pthread_mutex_destroy(&oap_auth.replay.mtx); + + auth_destroy_ctx(oap_auth.ca_ctx); +} + +int oap_auth_add_ca_crt(void * crt) +{ + return auth_add_crt_to_store(oap_auth.ca_ctx, crt); +} + +int oap_auth_add_chain_crt(void * crt) +{ + return auth_add_crt_to_chain(oap_auth.ca_ctx, crt); +} + +/* HKDF info = LABEL (incl. NUL separator) || request-hash [|| response-hash] */ +#define OAP_BIND_LABEL "o7s-oap-bind" +#define OAP_KC_LABEL "o7s-oap-kc" +#define OAP_HS_LABEL "o7s-oap-hs" + +int oap_resp_hash(int md_nid, + buffer_t kex, + buffer_t data, + buffer_t crt, + buffer_t * out) +{ + buffer_t cat = BUF_INIT; + uint8_t * p; + ssize_t len; + + assert(out != NULL); + assert(out->data != NULL); + + cat.len = kex.len + data.len + crt.len; + if (cat.len == 0) + return -EINVAL; + + cat.data = malloc(cat.len); + if (cat.data == NULL) + return -ENOMEM; + + p = cat.data; + if (kex.len > 0) { + memcpy(p, kex.data, kex.len); + p += kex.len; + } + + if (data.len > 0) { + memcpy(p, data.data, data.len); + p += data.len; + } + + if (crt.len > 0) + memcpy(p, crt.data, crt.len); + + len = md_digest(md_nid, cat, out->data); + + freebuf(cat); + + if (len < 0) + return -ECRYPT; + + out->len = (size_t) len; + + return 0; +} + +/* HKDF-expand sk->key with info into out; -ECRYPT on failure. */ +static int oap_hkdf_expand(const struct crypt_sk * sk, + buffer_t info, + uint8_t * out, + size_t outlen) +{ + buffer_t prk; + buffer_t okm; + + prk.len = SYMMKEYSZ; + prk.data = sk->key; + okm.len = outlen; + okm.data = out; + + if (crypt_hkdf_expand(prk, info, okm) < 0) + return -ECRYPT; + + return 0; +} + +/* info = label || H(req) */ +#define OAP_HS_INFO_SZ (sizeof(OAP_HS_LABEL) + MAX_HASH_SIZE) +int oap_derive_hs_key(const struct crypt_sk * sk, + buffer_t req_hash, + uint8_t * out) +{ + uint8_t info_buf[OAP_HS_INFO_SZ]; + buffer_t info; + size_t len; + + assert(sk != NULL); + assert(req_hash.data != NULL); + assert(out != NULL); + + if (req_hash.len == 0 || req_hash.len > MAX_HASH_SIZE) + return -EINVAL; + + len = sizeof(OAP_HS_LABEL); + memcpy(info_buf, OAP_HS_LABEL, len); + memcpy(info_buf + len, req_hash.data, req_hash.len); + len += req_hash.len; + + info.len = len; + info.data = info_buf; + + return oap_hkdf_expand(sk, info, out, SYMMKEYSZ); +} + +/* info = label || H(req) || H(resp) || cipher_nid || kdf_nid */ +#define OAP_BIND_INFO_SZ \ + (sizeof(OAP_BIND_LABEL) + 2 * MAX_HASH_SIZE + 2 * sizeof(uint16_t)) +int oap_bind_session_key(struct crypt_sk * sk, + buffer_t req_hash, + buffer_t resp_hash, + int kdf_nid) +{ + uint8_t info_buf[OAP_BIND_INFO_SZ]; + uint8_t tmp[SYMMKEYSZ]; + uint16_t suite[2]; + buffer_t info; + size_t len; + + assert(sk != NULL); + assert(req_hash.data != NULL); + assert(resp_hash.data != NULL); + + if (req_hash.len == 0 || req_hash.len > MAX_HASH_SIZE) + return -EINVAL; + + if (resp_hash.len == 0 || resp_hash.len > MAX_HASH_SIZE) + return -EINVAL; + + len = sizeof(OAP_BIND_LABEL); + memcpy(info_buf, OAP_BIND_LABEL, len); + memcpy(info_buf + len, req_hash.data, req_hash.len); + len += req_hash.len; + + memcpy(info_buf + len, resp_hash.data, resp_hash.len); + len += resp_hash.len; + + suite[0] = hton16((uint16_t) sk->nid); + suite[1] = hton16((uint16_t) kdf_nid); + memcpy(info_buf + len, suite, sizeof(suite)); + len += sizeof(suite); + + info.len = len; + info.data = info_buf; + + if (oap_hkdf_expand(sk, info, tmp, SYMMKEYSZ) < 0) + return -ECRYPT; + + memcpy(sk->key, tmp, SYMMKEYSZ); + crypt_secure_clear(tmp, SYMMKEYSZ); + + return 0; +} + +/* info = label || H(req) || H(resp) */ +#define OAP_KC_INFO_SZ (sizeof(OAP_KC_LABEL) + 2 * MAX_HASH_SIZE) +int oap_key_confirm_tag(const struct crypt_sk * sk, + buffer_t req_hash, + buffer_t resp_hash, + uint8_t * out, + size_t outlen) +{ + uint8_t info_buf[OAP_KC_INFO_SZ]; + buffer_t info; + size_t len; + + assert(sk != NULL); + assert(req_hash.data != NULL); + assert(resp_hash.data != NULL); + assert(out != NULL); + + if (req_hash.len == 0 || req_hash.len > MAX_HASH_SIZE) + return -EINVAL; + + if (resp_hash.len == 0 || resp_hash.len > MAX_HASH_SIZE) + return -EINVAL; + + if (outlen > MAX_HASH_SIZE) + return -EINVAL; + + len = sizeof(OAP_KC_LABEL); + memcpy(info_buf, OAP_KC_LABEL, len); + memcpy(info_buf + len, req_hash.data, req_hash.len); + len += req_hash.len; + + memcpy(info_buf + len, resp_hash.data, resp_hash.len); + len += resp_hash.len; + + info.len = len; + info.data = info_buf; + + return oap_hkdf_expand(sk, info, out, outlen); +} + +#define TIMESYNC_SLACK 100 /* ms */ +#define ID_IS_EQUAL(id1, id2) (memcmp(id1, id2, OAP_ID_SIZE) == 0) +int oap_check_hdr(const struct oap_hdr * hdr) +{ + struct oap_replay_bucket * b; + struct oap_replay_slot * slots; + struct timespec now; + uint64_t stamp; + uint64_t cur; + uint64_t gen; + uint8_t * id; + size_t h; + ssize_t delta; + int ret = 0; + + assert(hdr != NULL); + + stamp = hdr->timestamp; + id = hdr->id.data; + + clock_gettime(CLOCK_REALTIME, &now); + + cur = TS_TO_UINT64(now); + + delta = (ssize_t)(cur - stamp) / MILLION; + if (delta < -TIMESYNC_SLACK) { + log_err_id(id, "OAP header from %zd ms into future.", -delta); + return -EAUTH; + } + + if (delta > OAP_REPLAY_TIMER * 1000) { + log_err_id(id, "OAP header too old (%zd ms).", delta); + return -EAUTH; + } + + gen = stamp / ((uint64_t) OAP_REPLAY_TIMER * BILLION); + + pthread_mutex_lock(&oap_auth.replay.mtx); + + b = &oap_auth.replay.bucket[gen % OAP_REPLAY_GENS]; + + /* Rotate a stale bucket in O(1): its old-gen slots become free. */ + if (b->gen != gen) { + b->gen = gen; + b->count = 0; + } + + slots = b->slots; + + h = replay_hash(id, stamp) & oap_auth.replay.mask; + while (slots[h].gen == gen) { + if (slots[h].ts == stamp && ID_IS_EQUAL(slots[h].id, id)) { + log_warn_id(id, "OAP header already known."); + ret = -EREPLAY; + goto out; + } + + h = (h + 1) & oap_auth.replay.mask; + } + + /* Empty slot found; fail closed when the window is at capacity. */ + if (b->count >= oap_auth.replay.cap) { + log_warn_id(id, "OAP replay cache full; rejecting."); + ret = -EAUTH; + goto out; + } + + slots[h].gen = gen; + slots[h].ts = stamp; + memcpy(slots[h].id, id, OAP_ID_SIZE); + b->count++; + out: + pthread_mutex_unlock(&oap_auth.replay.mtx); + + return ret; +} + +int oap_auth_peer(char * name, + const struct sec_config * cfg, + const struct oap_hdr * local_hdr, + const struct oap_hdr * peer_hdr, + const buffer_t * cached_crt) +{ + void * crt; + void * pk = NULL; + void * pin = NULL; + buffer_t crt_der; /* cert source: wire, else cached (re-key) */ + buffer_t sign; /* Signed region */ + uint8_t * id = peer_hdr->id.data; + int ret; + + assert(name != NULL); + assert(cfg != NULL); + assert(local_hdr != NULL); + assert(peer_hdr != NULL); + + if (memcmp(peer_hdr->id.data, local_hdr->id.data, OAP_ID_SIZE) != 0) { + log_err_id(id, "OAP ID mismatch in flow allocation."); + goto fail_check; + } + + /* Re-key drops the wire cert; fall back to the cached peer cert. */ + crt_der = peer_hdr->crt; + if (crt_der.len == 0 && cached_crt != NULL) + crt_der = *cached_crt; + + if (crt_der.len == 0) { + if (cfg->a.req) { + log_err_id(id, "Peer did not provide a certificate."); + goto fail_check; + } + log_dbg_id(id, "No crt provided."); + name[0] = '\0'; + return 0; + } + + if (crypt_load_crt_der(crt_der, &crt) < 0) { + log_err_id(id, "Failed to load crt."); + goto fail_check; + } + + log_dbg_id(id, "Loaded peer crt."); + + if (crypt_get_pubkey_crt(crt, &pk) < 0) { + log_err_id(id, "Failed to get pubkey from crt."); + goto fail_crt; + } + + log_dbg_id(id, "Got public key from crt."); + + if (cfg->a.cacert[0] != '\0') { + if (crypt_load_crt_file(cfg->a.cacert, &pin) < 0) { + log_err_id(id, "Failed to load pinned CA %s.", + cfg->a.cacert); + goto fail_crt; + } + } + + ret = auth_verify_crt_pin(oap_auth.ca_ctx, crt, pin); + if (ret == -ENOENT) { + log_err_id(id, "Peer crt not issued by pinned CA %s.", + cfg->a.cacert); + goto fail_pin; + } + + if (ret < 0) { + log_err_id(id, "Failed to verify peer with CA store."); + goto fail_pin; + } + + log_dbg_id(id, "Successfully verified peer crt."); + + /* Digest pin: peer must sign with the configured digest */ + if (crypt_pk_requires_md(pk) && + cfg->d.nid != NID_undef && peer_hdr->md_nid != cfg->d.nid) { + log_err_id(id, "Peer did not sign with %s.", + md_nid_to_str(cfg->d.nid)); + goto fail_pin; + } + + /* Sealed responses verify over the reconstructed plaintext. */ + sign = peer_hdr->sealed_pt.data != NULL ? + peer_hdr->sealed_pt : peer_hdr->hdr; + sign.len -= peer_hdr->sig.len; + + if (auth_verify_sig(pk, peer_hdr->md_nid, sign, peer_hdr->sig) < 0) { + log_err_id(id, "Failed to verify signature."); + goto fail_pin; + } + + ret = crypt_get_crt_name(crt, name); + if (ret < 0) { + if (ret == -ENAME) + log_err_id(id, "Certificate CN too long."); + else + log_err_id(id, "No name in certificate."); + goto fail_pin; + } + + if (pin != NULL) + crypt_free_crt(pin); + + crypt_free_key(pk); + crypt_free_crt(crt); + + log_dbg_id(id, "Successfully authenticated peer."); + + return 0; + + fail_pin: + if (pin != NULL) + crypt_free_crt(pin); + fail_crt: + crypt_free_key(pk); + crypt_free_crt(crt); + fail_check: + return -EAUTH; +} diff --git a/src/irmd/oap/auth.h b/src/irmd/oap/auth.h new file mode 100644 index 00000000..72938b53 --- /dev/null +++ b/src/irmd/oap/auth.h @@ -0,0 +1,68 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * OAP - Authentication functions + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#ifndef OUROBOROS_IRMD_OAP_AUTH_H +#define OUROBOROS_IRMD_OAP_AUTH_H + +#include <ouroboros/crypt.h> + +#include "hdr.h" + +int oap_check_hdr(const struct oap_hdr * hdr); + +/* + * name is set to the peer crt CN, "" if no crt was presented. + * cached_crt (or NULL) is the peer cert from the initial handshake, used + * to verify a cert-less re-key. + */ +int oap_auth_peer(char * name, + const struct sec_config * cfg, + const struct oap_hdr * local_hdr, + const struct oap_hdr * peer_hdr, + const buffer_t * cached_crt); + +/* Derive the handshake key that seals the response identity block. */ +int oap_derive_hs_key(const struct crypt_sk * sk, + buffer_t req_hash, + uint8_t * out); + +/* resp_hash = H(kex || data || crt): binds the server response transcript. */ +int oap_resp_hash(int md_nid, + buffer_t kex, + buffer_t data, + buffer_t crt, + buffer_t * out); + +/* Fold request + response transcript + negotiated suite into the key. */ +int oap_bind_session_key(struct crypt_sk * sk, + buffer_t req_hash, + buffer_t resp_hash, + int kdf_nid); + +/* Server->client key-confirmation tag derived from the bound key. */ +int oap_key_confirm_tag(const struct crypt_sk * sk, + buffer_t req_hash, + buffer_t resp_hash, + uint8_t * out, + size_t outlen); + +#endif /* OUROBOROS_IRMD_OAP_AUTH_H */ diff --git a/src/irmd/oap/cli.c b/src/irmd/oap/cli.c new file mode 100644 index 00000000..689d67ca --- /dev/null +++ b/src/irmd/oap/cli.c @@ -0,0 +1,672 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * OAP - Client-side processing + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#if defined(__linux__) || defined(__CYGWIN__) + #define _DEFAULT_SOURCE +#else + #define _POSIX_C_SOURCE 200809L +#endif + +#define OUROBOROS_PREFIX "irmd/oap" + +#include <ouroboros/crypt.h> +#include <ouroboros/errno.h> +#include <ouroboros/logs.h> +#include <ouroboros/random.h> + +#include "config.h" + +#include "auth.h" +#include "hdr.h" +#include "io.h" +#include "../oap.h" + +#include <assert.h> +#include <limits.h> +#include <stdio.h> +#include <stdlib.h> +#include <string.h> + +/* Client context between oap_cli_prepare and oap_cli_complete */ +struct oap_cli_ctx { + uint8_t __id[OAP_ID_SIZE]; + buffer_t id; + uint8_t kex_buf[CRYPT_KEY_BUFSZ]; + uint8_t req_hash[MAX_HASH_SIZE]; + size_t req_hash_len; + int req_md_nid; + struct sec_config scfg; + struct oap_hdr local_hdr; + void * pkp; /* Ephemeral keypair */ + uint8_t * key; /* For client-encap KEM */ +}; + +#define OAP_CLI_CTX_INIT(s) \ + do { s->id.len = OAP_ID_SIZE; s->id.data = s->__id; } while (0) + +/* Client-side credential loading, mocked in tests */ + +#ifdef OAP_TEST_MODE +extern int load_cli_credentials(const struct name_info * info, + void ** pkp, + void ** crt); +extern int load_cli_sec_config(const struct name_info * info, + struct sec_config * cfg); +extern int load_server_kem_pk(const char * name, + struct sec_config * cfg, + buffer_t * buf); +#else + +int load_cli_credentials(const struct name_info * info, + void ** pkp, + void ** crt) +{ + assert(info != NULL); + assert(pkp != NULL); + assert(crt != NULL); + + return load_credentials(info->name, &info->c, pkp, crt); +} + +int load_cli_sec_config(const struct name_info * info, + struct sec_config * cfg) +{ + assert(info != NULL); + assert(cfg != NULL); + + memset(cfg, 0, sizeof(*cfg)); + + /* A client authenticates the server by default, like an https client */ + cfg->a.req = OAP_CLIENT_AUTH_DEFAULT; + + return load_sec_config(info->name, info->c.sec, cfg); +} + +int load_server_kem_pk(const char * name, + struct sec_config * cfg, + buffer_t * pk) +{ + char path[PATH_MAX]; + const char * ext; + + assert(name != NULL); + assert(cfg != NULL); + assert(pk != NULL); + + ext = IS_HYBRID_KEM(cfg->x.str) ? "raw" : "pem"; + + snprintf(path, sizeof(path), + OUROBOROS_CLI_CRT_DIR "/%s/kex.srv.pub.%s", name, ext); + + if (IS_HYBRID_KEM(cfg->x.str)) { + if (crypt_load_pubkey_raw_file(path, pk) < 0) { + log_err("Failed to load %s pubkey from %s.", ext, path); + return -1; + } + } else { + if (crypt_load_pubkey_file_to_der(path, pk) < 0) { + log_err("Failed to load %s pubkey from %s.", ext, path); + return -1; + } + } + + log_dbg("Loaded %s pubkey from %s (%zu bytes).", ext, path, pk->len); + + return 0; +} + +#endif /* OAP_TEST_MODE */ + +static int do_client_kex_prepare_dhe(struct oap_cli_ctx * s) +{ + struct sec_config * scfg = &s->scfg; + buffer_t * kex = &s->local_hdr.kex; + uint8_t * id = s->id.data; + ssize_t len; + + /* Generate ephemeral keypair, send PK */ + len = kex_pkp_create(scfg, &s->pkp, kex->data); + if (len < 0) { + log_err_id(id, "Failed to generate DHE keypair."); + return -ECRYPT; + } + + kex->len = (size_t) len; + log_dbg_id(id, "Generated ephemeral %s keys (%zd bytes).", + scfg->x.str, len); + + return 0; +} + +static int do_client_kex_prepare_kem_encap(const char * server_name, + struct oap_cli_ctx * s) +{ + struct sec_config * scfg = &s->scfg; + buffer_t * kex = &s->local_hdr.kex; + uint8_t * id = s->id.data; + buffer_t server_pk = BUF_INIT; + uint8_t key_buf[SYMMKEYSZ]; + ssize_t len; + + if (load_server_kem_pk(server_name, scfg, &server_pk) < 0) { + log_err_id(id, "Failed to load server KEM pk."); + return -ECRYPT; + } + + if (IS_HYBRID_KEM(scfg->x.str)) + len = kex_kem_encap_raw(server_pk, kex->data, + scfg->k.nid, key_buf); + else + len = kex_kem_encap(server_pk, kex->data, + scfg->k.nid, key_buf); + + freebuf(server_pk); + + if (len < 0) { + log_err_id(id, "Failed to encapsulate KEM."); + return -ECRYPT; + } + + kex->len = (size_t) len; + log_dbg_id(id, "Client encaps: CT len=%zd.", len); + + /* Store derived key */ + s->key = crypt_secure_malloc(SYMMKEYSZ); + if (s->key == NULL) { + log_err_id(id, "Failed to allocate secure key."); + return -ENOMEM; + } + memcpy(s->key, key_buf, SYMMKEYSZ); + crypt_secure_clear(key_buf, SYMMKEYSZ); + + return 0; +} + +static int do_client_kex_prepare_kem_decap(struct oap_cli_ctx * s) +{ + struct sec_config * scfg = &s->scfg; + buffer_t * kex = &s->local_hdr.kex; + uint8_t * id = s->id.data; + ssize_t len; + + /* Server encaps: generate keypair, send PK */ + len = kex_pkp_create(scfg, &s->pkp, kex->data); + if (len < 0) { + log_err_id(id, "Failed to generate KEM keypair."); + return -ECRYPT; + } + + kex->len = (size_t) len; + log_dbg_id(id, "Client PK for server encaps (%zd bytes).", len); + + return 0; +} + +static int do_client_kex_prepare(const char * server_name, + struct oap_cli_ctx * s) +{ + struct sec_config * scfg = &s->scfg; + + if (!IS_KEX_ALGO_SET(scfg)) + return 0; + + if (IS_KEM_ALGORITHM(scfg->x.str)) { + if (scfg->x.mode == KEM_MODE_CLIENT_ENCAP) + return do_client_kex_prepare_kem_encap(server_name, s); + else + return do_client_kex_prepare_kem_decap(s); + } + + return do_client_kex_prepare_dhe(s); +} + +int oap_cli_prepare(void ** ctx, + const struct name_info * info, + buffer_t * req_buf, + buffer_t data, + bool rekey) +{ + struct oap_cli_ctx * s; + void * pkp = NULL; + void * crt = NULL; + buffer_t no_tag = BUF_INIT; + ssize_t ret; + + assert(ctx != NULL); + assert(info != NULL); + assert(req_buf != NULL); + + clrbuf(*req_buf); + *ctx = NULL; + + /* Allocate ctx to carry between prepare and complete */ + s = malloc(sizeof(*s)); + if (s == NULL) { + log_err("Failed to allocate OAP client ctx."); + return -ENOMEM; + } + + memset(s, 0, sizeof(*s)); + OAP_CLI_CTX_INIT(s); + + /* Generate session ID */ + if (random_buffer(s->__id, OAP_ID_SIZE) < 0) { + log_err("Failed to generate OAP session ID."); + goto fail_id; + } + + log_dbg_id(s->id.data, "Preparing OAP request for %s.", info->name); + + /* Load client credentials */ + if (load_cli_credentials(info, &pkp, &crt) < 0) { + log_err_id(s->id.data, "Failed to load credentials for %s.", + info->name); + goto fail_id; + } + + /* Load security config */ + if (load_cli_sec_config(info, &s->scfg) < 0) { + log_err_id(s->id.data, "Failed to load security config for %s.", + info->name); + goto fail_kex; + } + + /* Re-key forces server-encap: client-encap forfeits FS/PCS. */ + if (rekey && s->scfg.x.mode == KEM_MODE_CLIENT_ENCAP) { + s->scfg.x.mode = KEM_MODE_SERVER_ENCAP; + log_dbg_id(s->id.data, "Re-key forcing ephemeral server KEX."); + } + + /* Re-key omits the cert; the server verifies against its cache. */ + if (rekey && crt != NULL) { + crypt_free_crt(crt); + crt = NULL; + } + + oap_hdr_init(&s->local_hdr, s->id, s->kex_buf, data, s->scfg.c.nid); + + if (do_client_kex_prepare(info->name, s) < 0) { + log_err_id(s->id.data, "Failed to prepare client KEX."); + goto fail_kex; + } + + if (oap_hdr_encode(&s->local_hdr, pkp, crt, &s->scfg, + no_tag, NID_undef, NULL)) { + log_err_id(s->id.data, "Failed to create OAP request header."); + goto fail_hdr; + } + + debug_oap_hdr_snd(&s->local_hdr); + + /* Compute and store hash of request for verification in complete */ + s->req_md_nid = s->scfg.d.nid != NID_undef ? s->scfg.d.nid : NID_sha384; + ret = md_digest(s->req_md_nid, s->local_hdr.hdr, s->req_hash); + if (ret < 0) { + log_err_id(s->id.data, "Failed to hash request."); + goto fail_hash; + } + s->req_hash_len = (size_t) ret; + + /* Transfer ownership of request buffer */ + *req_buf = s->local_hdr.hdr; + clrbuf(s->local_hdr.hdr); + + /* oap_hdr_encode repoints id into hdr; restore to __id */ + s->local_hdr.id = s->id; + + crypt_free_crt(crt); + crypt_free_key(pkp); + + *ctx = s; + + log_dbg_id(s->id.data, "OAP request prepared for %s.", info->name); + + return 0; + + fail_hash: + oap_hdr_fini(&s->local_hdr); + fail_hdr: + crypt_secure_free(s->key, SYMMKEYSZ); + crypt_free_key(s->pkp); + fail_kex: + crypt_free_crt(crt); + crypt_free_key(pkp); + fail_id: + free(s); + return -ECRYPT; +} + +void oap_ctx_free(void * ctx) +{ + struct oap_cli_ctx * s = ctx; + + if (s == NULL) + return; + + oap_hdr_fini(&s->local_hdr); + + if (s->pkp != NULL) + crypt_free_key(s->pkp); + + if (s->key != NULL) + crypt_secure_free(s->key, SYMMKEYSZ); + + memset(s, 0, sizeof(*s)); + free(s); +} + +static int do_client_kex_complete_kem(struct oap_cli_ctx * s, + const struct oap_hdr * peer_hdr, + struct crypt_sk * sk) +{ + struct sec_config * scfg = &s->scfg; + uint8_t * id = s->id.data; + uint8_t key_buf[SYMMKEYSZ]; + + if (scfg->x.mode == KEM_MODE_SERVER_ENCAP) { + buffer_t ct; + + if (peer_hdr->kex.len == 0) { + log_err_id(id, "Server did not send KEM CT."); + return -ECRYPT; + } + + ct.data = peer_hdr->kex.data; + ct.len = peer_hdr->kex.len; + + if (kex_kem_decap(s->pkp, ct, scfg->k.nid, key_buf) < 0) { + log_err_id(id, "Failed to decapsulate KEM."); + return -ECRYPT; + } + + log_dbg_id(id, "Client decapsulated server CT."); + + } else if (scfg->x.mode == KEM_MODE_CLIENT_ENCAP) { + /* Key already derived during prepare */ + memcpy(sk->key, s->key, SYMMKEYSZ); + sk->nid = scfg->c.nid; + log_info_id(id, "Negotiated %s + %s.", scfg->x.str, + scfg->c.str); + return 0; + } + + memcpy(sk->key, key_buf, SYMMKEYSZ); + sk->nid = scfg->c.nid; + crypt_secure_clear(key_buf, SYMMKEYSZ); + + log_info_id(id, "Negotiated %s + %s.", scfg->x.str, scfg->c.str); + + return 0; +} + +static int do_client_kex_complete_dhe(struct oap_cli_ctx * s, + const struct oap_hdr * peer_hdr, + struct crypt_sk * sk) +{ + struct sec_config * scfg = &s->scfg; + uint8_t * id = s->id.data; + uint8_t key_buf[SYMMKEYSZ]; + + /* DHE: derive from server's public key */ + if (peer_hdr->kex.len == 0) { + log_err_id(id, "Server did not send DHE public key."); + return -ECRYPT; + } + + if (kex_dhe_derive(scfg, s->pkp, peer_hdr->kex, key_buf) < 0) { + log_err_id(id, "Failed to derive DHE secret."); + return -ECRYPT; + } + + log_dbg_id(id, "DHE: derived shared secret."); + + memcpy(sk->key, key_buf, SYMMKEYSZ); + sk->nid = scfg->c.nid; + crypt_secure_clear(key_buf, SYMMKEYSZ); + + log_info_id(id, "Negotiated %s + %s.", scfg->x.str, scfg->c.str); + + return 0; +} + + +static int do_client_kex_complete(struct oap_cli_ctx * s, + const struct oap_hdr * peer_hdr, + struct crypt_sk * sk) +{ + struct sec_config * scfg = &s->scfg; + uint8_t * id = s->id.data; + int cipher_nid; + int kdf_nid; + + if (!IS_KEX_ALGO_SET(scfg)) + return 0; + + /* Save client's configured minimums */ + cipher_nid = scfg->c.nid; + kdf_nid = scfg->k.nid; + + /* Accept server's cipher choice */ + if (peer_hdr->cipher_str == NULL) { + log_err_id(id, "Server did not provide cipher."); + return -ECRYPT; + } + + SET_KEX_CIPHER(scfg, peer_hdr->cipher_str); + if (crypt_validate_nid(scfg->c.nid) < 0) { + log_err_id(id, "Server cipher '%s' not supported.", + peer_hdr->cipher_str); + return -ENOTSUP; + } + + /* Verify server cipher >= client's minimum */ + if (crypt_cipher_rank(scfg->c.nid) < crypt_cipher_rank(cipher_nid)) { + log_err_id(id, "Server cipher %s too weak.", + peer_hdr->cipher_str); + return -ECRYPT; + } + + log_dbg_id(id, "Accepted server cipher %s.", + peer_hdr->cipher_str); + + /* Accept server's KDF for non-client-encap modes */ + if (scfg->x.mode != KEM_MODE_CLIENT_ENCAP + && peer_hdr->kdf_nid != NID_undef) { + if (crypt_kdf_rank(peer_hdr->kdf_nid) + < crypt_kdf_rank(kdf_nid)) { + log_err_id(id, "Server KDF too weak."); + return -ECRYPT; + } + SET_KEX_KDF_NID(scfg, peer_hdr->kdf_nid); + log_dbg_id(id, "Accepted server KDF %s.", + md_nid_to_str(scfg->k.nid)); + } + + /* Derive shared secret */ + if (IS_KEM_ALGORITHM(scfg->x.str)) + return do_client_kex_complete_kem(s, peer_hdr, sk); + + return do_client_kex_complete_dhe(s, peer_hdr, sk); +} + +int oap_cli_complete(void * ctx, + const struct name_info * info, + buffer_t rsp_buf, + buffer_t * data, + struct crypt_sk * sk, + const buffer_t * cached_crt, + buffer_t * peer_crt) +{ + struct oap_cli_ctx * s = ctx; + struct oap_hdr peer_hdr; + char peer[NAME_SIZE + 1]; + uint8_t kc_buf[MAX_HASH_SIZE]; + uint8_t resp_hash_buf[MAX_HASH_SIZE]; + uint8_t hs_key[SYMMKEYSZ]; + buffer_t req_hash = BUF_INIT; + buffer_t resp_hash = BUF_INIT; + uint8_t * id; + int rc; + + assert(ctx != NULL); + assert(info != NULL); + assert(data != NULL); + assert(sk != NULL); + + sk->nid = NID_undef; + + clrbuf(*data); + + memset(&peer_hdr, 0, sizeof(peer_hdr)); + + id = s->id.data; + + log_dbg_id(id, "Completing OAP for %s.", info->name); + + /* Decode response header using client's md_nid for hash length */ + if (oap_hdr_decode(&peer_hdr, rsp_buf, s->req_md_nid, false) < 0) { + log_err_id(id, "Failed to decode OAP response header."); + goto fail_oap; + } + + debug_oap_hdr_rcv(&peer_hdr); + + /* Verify response ID matches request */ + if (memcmp(peer_hdr.id.data, id, OAP_ID_SIZE) != 0) { + log_err_id(id, "OAP response ID mismatch."); + goto fail_oap; + } + + /* Complete key exchange first; the sealed identity needs the secret */ + if (do_client_kex_complete(s, &peer_hdr, sk) < 0) { + log_err_id(id, "Failed to complete key exchange."); + goto fail_oap; + } + + req_hash.data = s->req_hash; + req_hash.len = s->req_hash_len; + + /* Decrypt the sealed server identity (data+cert+sig) before auth */ + if (sk->nid != NID_undef && peer_hdr.sealed.data != NULL) { + if (oap_derive_hs_key(sk, req_hash, hs_key) < 0) { + log_err_id(id, "Failed to derive handshake key."); + goto fail_oap; + } + + rc = oap_hdr_unseal(&peer_hdr, hs_key); + + crypt_secure_clear(hs_key, SYMMKEYSZ); + + if (rc < 0) { + log_err_id(id, "Failed to unseal server identity."); + goto fail_oap; + } + } + + /* Authenticate server (cert + signature now in cleartext) */ + if (oap_auth_peer(peer, &s->scfg, &s->local_hdr, &peer_hdr, + cached_crt) < 0) { + log_err_id(id, "Failed to authenticate server."); + goto fail_oap; + } + + /* Surface the peer cert so the caller can cache it for re-key. */ + if (peer_crt != NULL && peer_hdr.crt.len > 0) { + peer_crt->data = malloc(peer_hdr.crt.len); + if (peer_crt->data == NULL) + goto fail_oap; + + memcpy(peer_crt->data, peer_hdr.crt.data, peer_hdr.crt.len); + peer_crt->len = peer_hdr.crt.len; + } + + /* Response must carry a transcript tag of the expected length */ + if (peer_hdr.rsp_tag.len != s->req_hash_len) { + log_err_id(id, "Response transcript tag mismatch."); + goto fail_oap; + } + + /* Verify peer certificate name matches expected destination */ + if (peer_hdr.crt.len > 0 && strcmp(peer, info->name) != 0) { + log_err_id(id, "Peer crt for '%s' does not match '%s'.", + peer, info->name); + goto fail_oap; + } + + if (sk->nid != NID_undef) { + /* Encrypted: bind the key and verify key confirmation */ + resp_hash.data = resp_hash_buf; + + if (oap_resp_hash(s->req_md_nid, peer_hdr.kex, + peer_hdr.data, peer_hdr.crt, + &resp_hash) < 0) { + log_err_id(id, "Failed to hash response."); + goto fail_oap; + } + + if (oap_bind_session_key(sk, req_hash, resp_hash, + s->scfg.k.nid) < 0) { + log_err_id(id, "Failed to bind session key."); + goto fail_oap; + } + + if (oap_key_confirm_tag(sk, req_hash, resp_hash, kc_buf, + s->req_hash_len) < 0) { + log_err_id(id, "Failed to confirm session key."); + goto fail_oap; + } + + if (crypt_ct_cmp(peer_hdr.rsp_tag.data, kc_buf, + s->req_hash_len) != 0) { + log_err_id(id, "Key confirmation mismatch."); + goto fail_oap; + } + } else { + /* Cleartext path is config-driven, never a wire downgrade */ + assert(!IS_KEX_ALGO_SET(&s->scfg)); + /* Unencrypted: verify request-echo integrity */ + if (crypt_ct_cmp(peer_hdr.rsp_tag.data, s->req_hash, + s->req_hash_len) != 0) { + log_err_id(id, "Response tag mismatch."); + goto fail_oap; + } + } + + /* Copy piggybacked data from server response */ + if (oap_hdr_copy_data(&peer_hdr, data) < 0) { + log_err_id(id, "Failed to copy server data."); + goto fail_oap; + } + + log_info_id(id, "OAP completed for %s.", info->name); + + freebuf(peer_hdr.sealed_pt); + + oap_ctx_free(s); + + return 0; + + fail_oap: + freebuf(peer_hdr.sealed_pt); + oap_ctx_free(s); + return -ECRYPT; +} diff --git a/src/irmd/oap/hdr.c b/src/irmd/oap/hdr.c new file mode 100644 index 00000000..f8400b46 --- /dev/null +++ b/src/irmd/oap/hdr.c @@ -0,0 +1,755 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * OAP - Header encoding, decoding, and debugging + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#if defined(__linux__) || defined(__CYGWIN__) + #define _DEFAULT_SOURCE +#else + #define _POSIX_C_SOURCE 200809L +#endif + +#define OUROBOROS_PREFIX "irmd/oap" + +#include <ouroboros/crypt.h> +#include <ouroboros/endian.h> +#include <ouroboros/errno.h> +#include <ouroboros/hash.h> +#include <ouroboros/logs.h> +#include <ouroboros/rib.h> +#include <ouroboros/time.h> + +#include "config.h" + +#include "hdr.h" + +#include <assert.h> +#include <errno.h> +#include <stdlib.h> +#include <string.h> +#include <time.h> + +#define OAP_SEAL_TAGSZ 16 /* AEAD tag on the sealed identity block */ +/* Sealed length prefix: data_len ‖ crt_len. */ +#define OAP_SEAL_LENSZ (sizeof(uint16_t) + sizeof(uint16_t)) + +/* hs_key is single-use per handshake, so a fixed nonce is reuse-safe. */ +static const uint8_t oap_seal_nonce[12]; + +int oap_hdr_decode(struct oap_hdr * oap_hdr, + buffer_t hdr, + int req_md_nid, + bool rekey) +{ + off_t offset; + uint16_t kex_len; + uint16_t ciph_nid; + size_t crt_len; + size_t data_len; + size_t hash_len; + size_t sig_len; + + assert(oap_hdr != NULL); + memset(oap_hdr, 0, sizeof(*oap_hdr)); + + if (hdr.len < OAP_HDR_MIN_SIZE) + goto fail_decode; + + /* Parse fixed header (36 bytes) */ + oap_hdr->id.data = hdr.data; + oap_hdr->id.len = OAP_ID_SIZE; + + offset = OAP_ID_SIZE; + + oap_hdr->timestamp = ntoh64(*(uint64_t *)(hdr.data + offset)); + offset += sizeof(uint64_t); + + /* cipher NID */ + ciph_nid = ntoh16(*(uint16_t *)(hdr.data + offset)); + oap_hdr->nid = ciph_nid; + oap_hdr->cipher_str = crypt_nid_to_str(ciph_nid); + offset += sizeof(uint16_t); + + /* kdf NID */ + oap_hdr->kdf_nid = ntoh16(*(uint16_t *)(hdr.data + offset)); + oap_hdr->kdf_str = md_nid_to_str(oap_hdr->kdf_nid); + offset += sizeof(uint16_t); + + /* md NID (signature hash) */ + oap_hdr->md_nid = ntoh16(*(uint16_t *)(hdr.data + offset)); + oap_hdr->md_str = md_nid_to_str(oap_hdr->md_nid); + offset += sizeof(uint16_t); + + /* + * Validate NIDs: NID_undef is valid at parse time, else must be known. + * Note: md_nid=NID_undef only valid for PQC; enforced at sign/verify. + */ + if (ciph_nid != NID_undef && crypt_validate_nid(ciph_nid) < 0) + goto fail_decode; + + if (oap_hdr->kdf_nid != NID_undef && + md_validate_nid(oap_hdr->kdf_nid) < 0) + goto fail_decode; + if (oap_hdr->md_nid != NID_undef && + md_validate_nid(oap_hdr->md_nid) < 0) + goto fail_decode; + + /* crt_len */ + crt_len = (size_t) ntoh16(*(uint16_t *)(hdr.data + offset)); + offset += sizeof(uint16_t); + + /* kex_len + flags */ + kex_len = ntoh16(*(uint16_t *)(hdr.data + offset)); + oap_hdr->kex.len = (size_t) (kex_len & OAP_KEX_LEN_MASK); + oap_hdr->kex_flags.fmt = (kex_len & OAP_KEX_FMT_BIT) ? 1 : 0; + oap_hdr->kex_flags.role = (kex_len & OAP_KEX_ROLE_BIT) ? 1 : 0; + offset += sizeof(uint16_t); + + /* data_len */ + data_len = (size_t) ntoh16(*(uint16_t *)(hdr.data + offset)); + offset += sizeof(uint16_t); + + assert((size_t) offset == OAP_HDR_MIN_SIZE); + + /* Response includes rsp_tag when md_nid is set */ + hash_len = (req_md_nid != NID_undef) ? + (size_t) md_len(req_md_nid) : 0; + + /* Encrypted response: sealed block is data_len‖crt_len‖data‖crt‖sig. */ + if (req_md_nid != NID_undef && ciph_nid != NID_undef) { + if (hdr.len < (size_t) offset + oap_hdr->kex.len + hash_len + + OAP_SEAL_TAGSZ + OAP_SEAL_LENSZ) + goto fail_decode; + + oap_hdr->kex.data = hdr.data + offset; + offset += oap_hdr->kex.len; + + oap_hdr->rsp_tag.data = hdr.data + offset; + oap_hdr->rsp_tag.len = hash_len; + offset += hash_len; + + oap_hdr->sealed.data = hdr.data + offset; + oap_hdr->sealed.len = hdr.len - offset; + + /* crt/data/sig lengths are sealed; set by oap_hdr_unseal. */ + oap_hdr->crt.len = crt_len; + oap_hdr->data.len = data_len; + + oap_hdr->hdr = hdr; + + return 0; + } + + /* Validate total length */ + if (hdr.len < (size_t) offset + crt_len + oap_hdr->kex.len + + data_len + hash_len) + goto fail_decode; + + /* Derive sig_len from remaining bytes */ + sig_len = hdr.len - offset - crt_len - oap_hdr->kex.len - + data_len - hash_len; + + /* + * Unsigned packets must not have trailing bytes. A re-key request + * is signed but cert-less (verified against the cached peer cert), + * so the rekey caller permits crt_len==0 with a signature. + */ + if (crt_len == 0 && sig_len != 0 && !rekey) + goto fail_decode; + + /* Parse variable fields */ + oap_hdr->crt.data = hdr.data + offset; + oap_hdr->crt.len = crt_len; + offset += crt_len; + + oap_hdr->kex.data = hdr.data + offset; + offset += oap_hdr->kex.len; + + oap_hdr->data.data = hdr.data + offset; + oap_hdr->data.len = data_len; + offset += data_len; + + oap_hdr->rsp_tag.data = hdr.data + offset; + oap_hdr->rsp_tag.len = hash_len; + offset += hash_len; + + oap_hdr->sig.data = hdr.data + offset; + oap_hdr->sig.len = sig_len; + + oap_hdr->hdr = hdr; + + return 0; + + fail_decode: + memset(oap_hdr, 0, sizeof(*oap_hdr)); + return -1; +} + +void oap_hdr_fini(struct oap_hdr * oap_hdr) +{ + assert(oap_hdr != NULL); + + freebuf(oap_hdr->sealed_pt); + freebuf(oap_hdr->hdr); + memset(oap_hdr, 0, sizeof(*oap_hdr)); +} + +int oap_hdr_copy_data(const struct oap_hdr * hdr, + buffer_t * out) +{ + assert(hdr != NULL); + assert(out != NULL); + + if (hdr->data.len == 0) { + clrbuf(*out); + return 0; + } + + out->data = malloc(hdr->data.len); + if (out->data == NULL) + return -ENOMEM; + + memcpy(out->data, hdr->data.data, hdr->data.len); + out->len = hdr->data.len; + + return 0; +} + +void oap_hdr_init(struct oap_hdr * hdr, + buffer_t id, + uint8_t * kex_buf, + buffer_t data, + uint16_t nid) +{ + assert(hdr != NULL); + assert(id.data != NULL && id.len == OAP_ID_SIZE); + + memset(hdr, 0, sizeof(*hdr)); + + hdr->id = id; + hdr->kex.data = kex_buf; + hdr->kex.len = 0; + hdr->data = data; + hdr->nid = nid; +} + +/* Write the 36-byte fixed header; stamp is already in network order. */ +static void write_oap_fixed(uint8_t * buf, + const struct oap_hdr * hdr, + const struct sec_config * scfg, + size_t crt_len, + size_t data_len, + uint64_t stamp) +{ + uint16_t v; + uint16_t kex_len; + off_t offset = 0; + + memcpy(buf + offset, hdr->id.data, hdr->id.len); + offset += hdr->id.len; + + memcpy(buf + offset, &stamp, sizeof(stamp)); + offset += sizeof(stamp); + + v = hton16(hdr->nid); + memcpy(buf + offset, &v, sizeof(v)); + offset += sizeof(v); + + v = hton16(scfg->k.nid); + memcpy(buf + offset, &v, sizeof(v)); + offset += sizeof(v); + + v = hton16(scfg->d.nid); + memcpy(buf + offset, &v, sizeof(v)); + offset += sizeof(v); + + v = hton16((uint16_t) crt_len); + memcpy(buf + offset, &v, sizeof(v)); + offset += sizeof(v); + + kex_len = (uint16_t) hdr->kex.len; + if (hdr->kex.len > 0 && IS_KEM_ALGORITHM(scfg->x.str)) { + if (IS_HYBRID_KEM(scfg->x.str)) + kex_len |= OAP_KEX_FMT_BIT; + if (scfg->x.mode == KEM_MODE_CLIENT_ENCAP) + kex_len |= OAP_KEX_ROLE_BIT; + } + + kex_len = hton16(kex_len); + memcpy(buf + offset, &kex_len, sizeof(kex_len)); + offset += sizeof(kex_len); + + v = hton16((uint16_t) data_len); + memcpy(buf + offset, &v, sizeof(v)); +} + +/* + * Pack lens ‖ data ‖ crt, sign prefix ‖ body, append the signature, then + * AEAD-seal lens ‖ data ‖ crt ‖ sig under prefix as AAD. The cert, app data + * and their sizes stay confidential; *out is the opaque sealed block. The + * signature rides inside the seal so it can't deanonymise the server. + */ +static int oap_seal_body(int nid, + const uint8_t * seal_key, + void * pkp, + int md_nid, + buffer_t prefix, + buffer_t data, + buffer_t crt, + buffer_t * out) +{ + buffer_t sig = BUF_INIT; + buffer_t sign; + buffer_t aad; + buffer_t plain; + uint8_t * buf; + uint8_t * tmp; + uint16_t datalen; + uint16_t crtlen; + size_t body_len; + off_t offset; + + datalen = hton16((uint16_t) data.len); + crtlen = hton16((uint16_t) crt.len); + + body_len = OAP_SEAL_LENSZ + data.len + crt.len; + + buf = malloc(prefix.len + body_len); + if (buf == NULL) + return -1; + + memcpy(buf, prefix.data, prefix.len); + offset = (off_t) prefix.len; + + memcpy(buf + offset, &datalen, sizeof(datalen)); + offset += sizeof(datalen); + + memcpy(buf + offset, &crtlen, sizeof(crtlen)); + offset += sizeof(crtlen); + + if (data.len != 0) + memcpy(buf + offset, data.data, data.len); + + offset += data.len; + + if (crt.len != 0) + memcpy(buf + offset, crt.data, crt.len); + + /* Sign prefix ‖ lens ‖ data ‖ crt (plaintext, before sealing). */ + sign.data = buf; + sign.len = prefix.len + body_len; + + if (pkp != NULL && auth_sign(pkp, md_nid, sign, &sig) < 0) + goto fail_buf; + + /* Append the signature so the seal covers lens ‖ data ‖ crt ‖ sig. */ + if (sig.len != 0) { + tmp = realloc(buf, prefix.len + body_len + sig.len); + if (tmp == NULL) + goto fail_sig; + + buf = tmp; + memcpy(buf + prefix.len + body_len, sig.data, sig.len); + } + + aad.data = buf; + aad.len = prefix.len; + plain.data = buf + prefix.len; + plain.len = body_len + sig.len; + + if (crypt_oneshot_seal(nid, seal_key, oap_seal_nonce, + aad, plain, out) < 0) + goto fail_sig; + + free(buf); + freebuf(sig); + + return 0; + + fail_sig: + freebuf(sig); + fail_buf: + free(buf); + return -1; +} + +/* Encode an identity-hidden response: wire = prefix ‖ oap_seal_body(...). */ +static int oap_hdr_encode_sealed(struct oap_hdr * hdr, + void * pkp, + void * crt, + struct sec_config * scfg, + buffer_t rsp_tag, + int req_md_nid, + const uint8_t * seal_key) +{ + struct timespec now; + uint64_t stamp; + buffer_t der = BUF_INIT; + buffer_t sealed = BUF_INIT; + buffer_t prefix; + off_t offset; + + clock_gettime(CLOCK_REALTIME, &now); + stamp = hton64(TS_TO_UINT64(now)); + + if (crt != NULL && crypt_crt_der(crt, &der) < 0) + goto fail_der; + + prefix.len = OAP_HDR_MIN_SIZE + hdr->kex.len + rsp_tag.len; + prefix.data = malloc(prefix.len); + if (prefix.data == NULL) + goto fail_der; + + /* Cleartext crt_len/data_len are 0; real lengths prefix the seal. */ + write_oap_fixed(prefix.data, hdr, scfg, 0, 0, stamp); + offset = OAP_HDR_MIN_SIZE; + + if (hdr->kex.len != 0) + memcpy(prefix.data + offset, hdr->kex.data, hdr->kex.len); + + offset += hdr->kex.len; + + if (rsp_tag.len != 0) + memcpy(prefix.data + offset, rsp_tag.data, rsp_tag.len); + + offset += rsp_tag.len; + + assert((size_t) offset == prefix.len); + + if (oap_seal_body(hdr->nid, seal_key, pkp, scfg->d.nid, + prefix, hdr->data, der, &sealed) < 0) + goto fail_prefix; + + hdr->hdr.len = prefix.len + sealed.len; + hdr->hdr.data = malloc(hdr->hdr.len); + if (hdr->hdr.data == NULL) + goto fail_sealed; + + memcpy(hdr->hdr.data, prefix.data, prefix.len); + memcpy(hdr->hdr.data + prefix.len, sealed.data, sealed.len); + + freebuf(sealed); + free(prefix.data); + freebuf(der); + + if (oap_hdr_decode(hdr, hdr->hdr, req_md_nid, false) < 0) + goto fail_decode; + + return 0; + + fail_decode: + oap_hdr_fini(hdr); + return -1; + fail_sealed: + freebuf(sealed); + fail_prefix: + free(prefix.data); + fail_der: + freebuf(der); + return -1; +} + +int oap_hdr_encode(struct oap_hdr * hdr, + void * pkp, + void * crt, + struct sec_config * scfg, + buffer_t rsp_tag, + int req_md_nid, + const uint8_t * seal_key) +{ + struct timespec now; + uint64_t stamp; + buffer_t out; + buffer_t der = BUF_INIT; + buffer_t sig = BUF_INIT; + buffer_t sign; + off_t offset; + + assert(hdr != NULL); + assert(hdr->id.data != NULL && hdr->id.len == OAP_ID_SIZE); + assert(scfg != NULL); + + if (seal_key != NULL) + return oap_hdr_encode_sealed(hdr, pkp, crt, scfg, rsp_tag, + req_md_nid, seal_key); + + clock_gettime(CLOCK_REALTIME, &now); + stamp = hton64(TS_TO_UINT64(now)); + + if (crt != NULL && crypt_crt_der(crt, &der) < 0) + goto fail_der; + + /* Fixed header (36 bytes) + variable fields + rsp_tag (rsp only) */ + out.len = OAP_HDR_MIN_SIZE + der.len + hdr->kex.len + hdr->data.len + + rsp_tag.len; + + out.data = malloc(out.len); + if (out.data == NULL) + goto fail_out; + + write_oap_fixed(out.data, hdr, scfg, der.len, hdr->data.len, stamp); + offset = OAP_HDR_MIN_SIZE; + + /* certificate (variable) */ + if (der.len != 0) + memcpy(out.data + offset, der.data, der.len); + + offset += der.len; + + /* kex data (variable) */ + if (hdr->kex.len != 0) + memcpy(out.data + offset, hdr->kex.data, hdr->kex.len); + + offset += hdr->kex.len; + + /* data (variable) */ + if (hdr->data.len != 0) + memcpy(out.data + offset, hdr->data.data, hdr->data.len); + + offset += hdr->data.len; + + /* rsp_tag (variable, response only) */ + if (rsp_tag.len != 0) + memcpy(out.data + offset, rsp_tag.data, rsp_tag.len); + + offset += rsp_tag.len; + + assert((size_t) offset == out.len); + + /* Sign the entire header (fixed + variable, excluding signature) */ + sign.data = out.data; + sign.len = out.len; + + if (pkp != NULL && auth_sign(pkp, scfg->d.nid, sign, &sig) < 0) + goto fail_sig; + + hdr->hdr = out; + + /* Append signature */ + if (sig.len > 0) { + hdr->hdr.len += sig.len; + hdr->hdr.data = realloc(out.data, hdr->hdr.len); + if (hdr->hdr.data == NULL) + goto fail_realloc; + + memcpy(hdr->hdr.data + offset, sig.data, sig.len); + clrbuf(out); + } + + if (oap_hdr_decode(hdr, hdr->hdr, req_md_nid, false) < 0) + goto fail_decode; + + freebuf(der); + freebuf(sig); + + return 0; + + fail_decode: + oap_hdr_fini(hdr); + fail_realloc: + freebuf(sig); + fail_sig: + freebuf(out); + fail_out: + freebuf(der); + fail_der: + return -1; +} + +int oap_hdr_unseal(struct oap_hdr * hdr, + const uint8_t * key) +{ + buffer_t pt = BUF_INIT; + buffer_t prefix; + uint8_t * recon; + size_t body_len; + size_t pt_len; + size_t data_len; + size_t crt_len; + + assert(hdr != NULL); + assert(key != NULL); + + if (hdr->sealed.data == NULL || hdr->sealed.len == 0) + return -EINVAL; + + /* AAD prefix is fixed‖kex‖rsp_tag; sealed starts right after. */ + prefix.data = hdr->hdr.data; + prefix.len = (size_t) (hdr->sealed.data - hdr->hdr.data); + + if (crypt_oneshot_open(hdr->nid, key, oap_seal_nonce, prefix, + hdr->sealed, &pt) < 0) + return -ECRYPT; + + pt_len = pt.len; + + /* Plaintext = data_len ‖ crt_len ‖ data ‖ crt ‖ sig. */ + if (pt_len < OAP_SEAL_LENSZ) + goto fail_auth; + + data_len = (size_t) ntoh16(*(uint16_t *) pt.data); + crt_len = (size_t) ntoh16(*(uint16_t *)(pt.data + sizeof(uint16_t))); + + body_len = OAP_SEAL_LENSZ + data_len + crt_len; + if (pt_len < body_len) + goto fail_auth; + + /* Rebuild prefix ‖ lens ‖ data ‖ crt ‖ sig (whole signed region). */ + recon = malloc(prefix.len + pt_len); + if (recon == NULL) + goto fail_mem; + + memcpy(recon, prefix.data, prefix.len); + memcpy(recon + prefix.len, pt.data, pt_len); + + freebuf(pt); + + hdr->sealed_pt.data = recon; + hdr->sealed_pt.len = prefix.len + pt_len; + + hdr->data.data = recon + prefix.len + OAP_SEAL_LENSZ; + hdr->data.len = data_len; + hdr->crt.data = recon + prefix.len + OAP_SEAL_LENSZ + data_len; + hdr->crt.len = crt_len; + hdr->sig.data = recon + prefix.len + body_len; + hdr->sig.len = pt_len - body_len; + + return 0; + + fail_mem: + freebuf(pt); + return -ENOMEM; + fail_auth: + freebuf(pt); + return -EAUTH; +} + +#ifdef DEBUG_PROTO_OAP +#define OAP_KEX_IS_KEM(hdr) ((hdr)->kex_flags.role | (hdr)->kex_flags.fmt) +static void debug_oap_hdr(const struct oap_hdr * hdr) +{ + assert(hdr); + + if (hdr->sealed.len > 0) + log_proto(" Sealed block: [%zu bytes] on wire", + hdr->sealed.len); + + if (hdr->crt.len > 0) + log_proto(" crt: [%zu bytes]", hdr->crt.len); + else if (hdr->sealed.len > 0) + log_proto(" crt: <sealed>"); + else + log_proto(" crt: <none>"); + + if (hdr->kex.len > 0) { + if (OAP_KEX_IS_KEM(hdr)) + log_proto(" Key Exchange Data: [%zu bytes] [%s]", + hdr->kex.len, + hdr->kex_flags.role ? + "Client encaps" : "Server encaps"); + else + log_proto(" Key Exchange Data: [%zu bytes]", + hdr->kex.len); + } else + log_proto(" Key Exchange Data: <none>"); + + if (hdr->cipher_str != NULL) + log_proto(" Cipher: %s", hdr->cipher_str); + else + log_proto(" Cipher: <none>"); + + if (hdr->kdf_str != NULL) + log_proto(" KDF: HKDF-%s", hdr->kdf_str); + else + log_proto(" KDF: <none>"); + + if (hdr->md_str != NULL) + log_proto(" Digest: %s", hdr->md_str); + else + log_proto(" Digest: <none>"); + + if (hdr->data.len > 0) + log_proto(" Data: [%zu bytes]", hdr->data.len); + else if (hdr->sealed.len > 0) + log_proto(" Data: <sealed>"); + else + log_proto(" Data: <none>"); + + if (hdr->rsp_tag.len > 0) + log_proto(" Rsp Tag: [%zu bytes]", hdr->rsp_tag.len); + else + log_proto(" Rsp Tag: <none>"); + + if (hdr->sig.len > 0) + log_proto(" Signature: [%zu bytes]", hdr->sig.len); + else if (hdr->sealed.len > 0) + log_proto(" Signature: <sealed>"); + else + log_proto(" Signature: <none>"); +} +#endif + +void debug_oap_hdr_rcv(const struct oap_hdr * hdr) +{ +#ifdef DEBUG_PROTO_OAP + struct tm * tm; + char tmstr[RIB_TM_STRLEN]; + time_t stamp; + + assert(hdr); + + stamp = (time_t) hdr->timestamp / BILLION; + + tm = gmtime(&stamp); + strftime(tmstr, sizeof(tmstr), RIB_TM_FORMAT, tm); + + log_proto("OAP_HDR [" HASH_FMT64 " @ %s ]%s <--", + HASH_VAL64(hdr->id.data), tmstr, + hdr->sealed.len > 0 ? " [sealed]" : ""); + + debug_oap_hdr(hdr); +#else + (void) hdr; +#endif +} + +void debug_oap_hdr_snd(const struct oap_hdr * hdr) +{ +#ifdef DEBUG_PROTO_OAP + struct tm * tm; + char tmstr[RIB_TM_STRLEN]; + time_t stamp; + + assert(hdr); + + stamp = (time_t) hdr->timestamp / BILLION; + + tm = gmtime(&stamp); + strftime(tmstr, sizeof(tmstr), RIB_TM_FORMAT, tm); + + log_proto("OAP_HDR [" HASH_FMT64 " @ %s ]%s -->", + HASH_VAL64(hdr->id.data), tmstr, + hdr->sealed.len > 0 ? " [sealed]" : ""); + + debug_oap_hdr(hdr); +#else + (void) hdr; +#endif +} diff --git a/src/irmd/oap/hdr.h b/src/irmd/oap/hdr.h new file mode 100644 index 00000000..1a599727 --- /dev/null +++ b/src/irmd/oap/hdr.h @@ -0,0 +1,195 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * OAP - Header definitions and functions + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#ifndef OUROBOROS_IRMD_OAP_HDR_H +#define OUROBOROS_IRMD_OAP_HDR_H + +#include <ouroboros/crypt.h> +#include <ouroboros/utils.h> + +#include <stdbool.h> +#include <stdint.h> + +#define OAP_ID_SIZE (16) +#define OAP_HDR_MIN_SIZE (OAP_ID_SIZE + sizeof(uint64_t) + 6 * sizeof(uint16_t)) + +#define OAP_KEX_FMT_BIT 0x8000 /* bit 15: 0=X.509 DER, 1=Raw */ +#define OAP_KEX_ROLE_BIT 0x4000 /* bit 14: 0=Server encaps, 1=Client encaps */ +#define OAP_KEX_LEN_MASK 0x3FFF /* bits 0-13: Length (0-16383 bytes) */ + +#define OAP_KEX_ROLE(hdr) (hdr->kex_flags.role) +#define OAP_KEX_FMT(hdr) (hdr->kex_flags.fmt) + +#define OAP_KEX_IS_X509_FMT(hdr) (((hdr)->kex_flags.fmt) == 0) +#define OAP_KEX_IS_RAW_FMT(hdr) (((hdr)->kex_flags.fmt) == 1) + +/* + * Plaintext layout (request, and unencrypted/signed response). The + * signature covers the whole packet except itself. + * + * 0 1 2 3 + * 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 + * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ ---+ + * | | | + * + + | + * | | | + * + id (128 bits) + | + * | Unique flow allocation ID | | + * + + | + * | | | + * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | + * | | | + * + timestamp (64 bits) + | + * | UTC nanoseconds since epoch | | + * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | + * | cipher_nid (16 bits) | kdf_nid (16 bits) | | + * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | + * | md_nid (16 bits) | crt_len (16 bits) | | + * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | + * |F|R| kex_len (14 bits) | data_len (16 bits) | | Signed + * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Region + * | | | + * + certificate (variable) + | + * | X.509 certificate, DER encoded | | + * + + | + * | | | + * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | + * | | | + * + kex_data (variable) + | + * | public key (DER/raw) or ciphertext (KEM) | | + * + + | + * | | | + * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | + * | | | + * + data (variable) + | + * | Piggybacked application data | | + * + + | + * | | | + * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | + * | | | + * + rsp_tag (variable, response only) + | + * | key-confirm tag (enc), else H(request) | | + * | | | + * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ ---+ + * | | + * + signature (variable) + + * | DSA signature over signed region | + * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ + * + * Encrypted response - wire layout. The certificate, application data and + * signature are AEAD-sealed - hiding the server identity and the cert/data + * sizes; kex and rsp_tag move ahead of the sealed block as cleartext AAD. + * + * 0 1 2 3 + * 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 + * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ ---+ + * | fixed header (36 bytes, see above) | | + * + id, timestamp, NIDs, crt_len=0, kex_len, data_len=0 + | AAD + * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | + * | kex_data (variable) | | + * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | + * | rsp_tag (variable, response only) | | + * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ ---+ + * | SEAL( data_len ‖ crt_len ‖ data ‖ crt ‖ sig ) | | + * + encrypted cert, app data and signature + | Sealed + * | + AEAD tag (128 bits) | | area + * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ ---+ + * + * cipher_nid: NID value for symmetric cipher (0 = none) + * kdf_nid: NID value for KDF function (0 = none) + * md_nid: NID value for signature hash (0 = PQC/no signature) + * + * kex_len field bit layout: + * F (bit 15): Format - 0 = X.509 DER, 1 = Raw/Hybrid + * R (bit 14): Role - 0 = Server encaps, 1 = Client encaps + * (R is ignored for non-KEM algorithms) + * Bits 0-13: Length (0-16383 bytes) + * + * Request: sig_len = total - 36 - crt_len - kex_len - data_len + * Response: sig_len = total - 36 - crt_len - kex_len - data_len - hash_len + * where hash_len = md_len(req_md_nid / sha384) + * + * The signed plaintext inside the seal is prefix ‖ data_len ‖ crt_len ‖ + * data ‖ crt ‖ sig; the cleartext prefix (fixed ‖ kex ‖ rsp_tag) is the + * AEAD AAD. Cleartext crt_len/data_len are 0 - the real lengths are sealed, + * hiding the cert and data sizes; oap_hdr_unseal reads them to split. + */ + +/* Parsed OAP header - buffers pointing to a single memory region */ +struct oap_hdr { + const char * cipher_str; + const char * kdf_str; + const char * md_str; + uint64_t timestamp; + uint16_t nid; + uint16_t kdf_nid; + uint16_t md_nid; + struct { + bool fmt; /* Format */ + bool role; /* Role */ + } kex_flags; + + buffer_t id; + buffer_t crt; + buffer_t kex; + buffer_t data; + buffer_t rsp_tag; /* key-confirm tag / H(req), rsp only */ + buffer_t sig; + buffer_t sealed; /* wire ciphertext ‖ tag (sealed rsp) */ + buffer_t sealed_pt; /* prefix‖lens‖data‖crt‖sig, owned */ + buffer_t hdr; +}; + + +void oap_hdr_init(struct oap_hdr * hdr, + buffer_t id, + uint8_t * kex_buf, + buffer_t data, + uint16_t nid); + +void oap_hdr_fini(struct oap_hdr * oap_hdr); + +int oap_hdr_encode(struct oap_hdr * hdr, + void * pkp, + void * crt, + struct sec_config * scfg, + buffer_t rsp_tag, + int req_md_nid, + const uint8_t * seal_key); + +int oap_hdr_decode(struct oap_hdr * hdr, + buffer_t buf, + int req_md_nid, + bool rekey); + +/* Decrypt a sealed response identity block; fills data, crt and sig. */ +int oap_hdr_unseal(struct oap_hdr * hdr, + const uint8_t * key); + +void debug_oap_hdr_rcv(const struct oap_hdr * hdr); + +void debug_oap_hdr_snd(const struct oap_hdr * hdr); + +int oap_hdr_copy_data(const struct oap_hdr * hdr, + buffer_t * out); + +#endif /* OUROBOROS_IRMD_OAP_HDR_H */ diff --git a/src/irmd/oap/internal.h b/src/irmd/oap/internal.h new file mode 100644 index 00000000..4a156723 --- /dev/null +++ b/src/irmd/oap/internal.h @@ -0,0 +1,119 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * OAP internal definitions + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#ifndef OUROBOROS_IRMD_OAP_INTERNAL_H +#define OUROBOROS_IRMD_OAP_INTERNAL_H + +#include <ouroboros/crypt.h> +#include <ouroboros/list.h> +#include <ouroboros/name.h> +#include <ouroboros/pthread.h> +#include <ouroboros/utils.h> + +#include "hdr.h" + +#include <stdbool.h> +#include <stdint.h> + +int oap_check_hdr(const struct oap_hdr * hdr); + +int oap_auth_peer(char * name, + const struct sec_config * cfg, + const struct oap_hdr * local_hdr, + const struct oap_hdr * peer_hdr); + +int oap_negotiate_cipher(const struct oap_hdr * peer_hdr, + struct sec_config * scfg); + +#ifndef OAP_TEST_MODE +int load_credentials(const char * name, + const struct name_sec_paths * paths, + void ** pkp, + void ** crt); + +int load_sec_config(const char * name, + const char * path, + struct sec_config * cfg); +#endif + +#ifndef OAP_TEST_MODE +int load_srv_credentials(const struct name_info * info, + void ** pkp, + void ** crt); + +int load_srv_sec_config(const struct name_info * info, + struct sec_config * cfg); + +int load_server_kem_keypair(const char * name, + struct sec_config * cfg, + void ** pkp); +#else +extern int load_srv_credentials(const struct name_info * info, + void ** pkp, + void ** crt); +extern int load_srv_sec_config(const struct name_info * info, + struct sec_config * cfg); +extern int load_server_kem_keypair(const char * name, + struct sec_config * cfg, + void ** pkp); +#endif + +int do_server_kex(const struct name_info * info, + struct oap_hdr * peer_hdr, + struct sec_config * scfg, + buffer_t * kex, + struct crypt_sk * sk); + +#ifndef OAP_TEST_MODE +int load_cli_credentials(const struct name_info * info, + void ** pkp, + void ** crt); + +int load_cli_sec_config(const struct name_info * info, + struct sec_config * cfg); + +int load_server_kem_pk(const char * name, + struct sec_config * cfg, + buffer_t * pk); +#else +extern int load_cli_credentials(const struct name_info * info, + void ** pkp, + void ** crt); +extern int load_cli_sec_config(const struct name_info * info, + struct sec_config * cfg); +extern int load_server_kem_pk(const char * name, + struct sec_config * cfg, + buffer_t * pk); +#endif + +int oap_client_kex_prepare(struct sec_config * scfg, + buffer_t server_pk, + buffer_t * kex, + uint8_t * key, + void ** ephemeral_pkp); + +int oap_client_kex_complete(const struct oap_hdr * peer_hdr, + struct sec_config * scfg, + void * pkp, + uint8_t * key); + +#endif /* OUROBOROS_IRMD_OAP_INTERNAL_H */ diff --git a/src/irmd/oap/io.c b/src/irmd/oap/io.c new file mode 100644 index 00000000..b5daa432 --- /dev/null +++ b/src/irmd/oap/io.c @@ -0,0 +1,158 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * OAP - File I/O for credentials and configuration + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#if defined(__linux__) || defined(__CYGWIN__) + #define _DEFAULT_SOURCE +#else + #define _POSIX_C_SOURCE 200809L +#endif + +#define OUROBOROS_PREFIX "irmd/oap" + +#include <ouroboros/crypt.h> +#include <ouroboros/errno.h> +#include <ouroboros/logs.h> + +#include "config.h" + +#include "io.h" + +#include <assert.h> +#include <string.h> +#include <sys/stat.h> + +/* + * Shared credential and configuration loading helpers + */ + +#ifndef OAP_TEST_MODE + +static bool file_exists(const char * path) +{ + struct stat s; + + if (stat(path, &s) == 0) + return true; + + if (errno == ENOENT) { + log_dbg("File %s does not exist.", path); + return false; + } + + /* Can't stat for another reason; assume present, fail on load */ + log_warn("Failed to stat %s: %s.", path, strerror(errno)); + + return true; +} + +int load_credentials(const char * name, + const struct name_sec_paths * paths, + void ** pkp, + void ** crt) +{ + assert(paths != NULL); + assert(pkp != NULL); + assert(crt != NULL); + + *pkp = NULL; + *crt = NULL; + + if (!file_exists(paths->crt) || !file_exists(paths->key)) { + log_info("No authentication certificates for %s.", name); + return 0; + } + + if (crypt_load_crt_file(paths->crt, crt) < 0) { + log_err("Failed to load %s for %s.", paths->crt, name); + goto fail_crt; + } + + if (crypt_load_privkey_file(paths->key, pkp) < 0) { + log_err("Failed to load %s for %s.", paths->key, name); + goto fail_key; + } + + log_info("Loaded authentication certificates for %s.", name); + + return 0; + + fail_key: + crypt_free_crt(*crt); + *crt = NULL; + fail_crt: + return -EAUTH; +} + +int load_sec_config(const char * name, + const char * path, + struct sec_config * cfg) +{ + void * pin; + + assert(name != NULL); + assert(cfg != NULL); + + /* Load security config */ + if (!file_exists(path)) + log_dbg("No encryption %s for %s.", path, name); + + if (load_sec_config_file(cfg, path) < 0) { + log_warn("Failed to load %s for %s.", path, name); + return -1; + } + + if (cfg->a.cacert[0] != '\0') { + if (crypt_load_crt_file(cfg->a.cacert, &pin) < 0) { + log_err("Failed to load pinned CA %s for %s.", + cfg->a.cacert, name); + return -EAUTH; + } + crypt_free_crt(pin); + } + + if (!IS_KEX_ALGO_SET(cfg)) { + log_info("Key exchange not configured for %s.", name); + return 0; + } +#ifndef HAVE_OPENSSL_ML_KEM + if (IS_KEM_ALGORITHM(cfg->x.str)) { + log_err("PQC not available, can't use %s for %s.", + cfg->x.str, name); + return -ENOTSUP; + } +#endif + if (crypt_kex_rank(cfg->x.nid) < 1) { + log_err("Key exchange not supported for %s.", name); + return -ENOTSUP; + } + + if (crypt_cipher_rank(cfg->c.nid) < 1) { + log_err("Cipher not supported for %s.", name); + return -ECRYPT; + } + + log_info("Encryption enabled for %s.", name); + + return 0; +} + +#endif /* OAP_TEST_MODE */ diff --git a/src/irmd/oap/io.h b/src/irmd/oap/io.h new file mode 100644 index 00000000..953e3898 --- /dev/null +++ b/src/irmd/oap/io.h @@ -0,0 +1,40 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * OAP - Credential and configuration file I/O + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#ifndef OUROBOROS_IRMD_OAP_IO_H +#define OUROBOROS_IRMD_OAP_IO_H + +#include <ouroboros/crypt.h> +#include <ouroboros/name.h> + +#ifndef OAP_TEST_MODE +int load_credentials(const char * name, + const struct name_sec_paths * paths, + void ** pkp, + void ** crt); + +int load_sec_config(const char * name, + const char * path, + struct sec_config * cfg); +#endif + +#endif /* OUROBOROS_IRMD_OAP_IO_H */ diff --git a/src/irmd/oap/srv.c b/src/irmd/oap/srv.c new file mode 100644 index 00000000..5d631618 --- /dev/null +++ b/src/irmd/oap/srv.c @@ -0,0 +1,582 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * OAP - Server-side processing + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#if defined(__linux__) || defined(__CYGWIN__) + #define _DEFAULT_SOURCE +#else + #define _POSIX_C_SOURCE 200809L +#endif + +#define OUROBOROS_PREFIX "irmd/oap" + +#include <ouroboros/crypt.h> +#include <ouroboros/errno.h> +#include <ouroboros/logs.h> + +#include "config.h" + +#include "auth.h" +#include "hdr.h" +#include "io.h" +#include "oap.h" + +#include <assert.h> +#include <limits.h> +#include <stdio.h> +#include <stdlib.h> +#include <string.h> + +#ifdef OAP_TEST_MODE +extern int load_srv_credentials(const struct name_info * info, + void ** pkp, + void ** crt); +extern int load_srv_sec_config(const struct name_info * info, + struct sec_config * cfg); +extern int load_server_kem_keypair(const char * name, + bool raw_fmt, + void ** pkp); +#else + +int load_srv_credentials(const struct name_info * info, + void ** pkp, + void ** crt) +{ + assert(info != NULL); + assert(pkp != NULL); + assert(crt != NULL); + + return load_credentials(info->name, &info->s, pkp, crt); +} + +int load_srv_sec_config(const struct name_info * info, + struct sec_config * cfg) +{ + assert(info != NULL); + assert(cfg != NULL); + + memset(cfg, 0, sizeof(*cfg)); + + /* Client auth stays opt-in (mTLS); enable with auth=required */ + return load_sec_config(info->name, info->s.sec, cfg); +} + +int load_server_kem_keypair(const char * name, + bool raw_fmt, + void ** pkp) +{ + char path[PATH_MAX]; + const char * ext; + + assert(name != NULL); + assert(pkp != NULL); + + ext = raw_fmt ? "raw" : "pem"; + + snprintf(path, sizeof(path), + OUROBOROS_SRV_CRT_DIR "/%s/kex.key.%s", name, ext); + + if (raw_fmt) { + if (crypt_load_privkey_raw_file(path, pkp) < 0) { + log_err("Failed to load %s keypair from %s.", + ext, path); + return -ECRYPT; + } + } else { + if (crypt_load_privkey_file(path, pkp) < 0) { + log_err("Failed to load %s keypair from %s.", + ext, path); + return -ECRYPT; + } + } + + log_dbg("Loaded server KEM keypair from %s.", path); + return 0; +} + +#endif /* OAP_TEST_MODE */ + +static int get_algo_from_peer_key(const struct oap_hdr * peer_hdr, + char * algo_buf) +{ + uint8_t * id = peer_hdr->id.data; + int ret; + + if (OAP_KEX_IS_RAW_FMT(peer_hdr)) { + ret = kex_get_algo_from_pk_raw(peer_hdr->kex, algo_buf); + if (ret < 0) { + log_err_id(id, "Failed to get algo from raw key."); + return -ECRYPT; + } + } else { + ret = kex_get_algo_from_pk_der(peer_hdr->kex, algo_buf); + if (ret < 0) { + log_err_id(id, "Failed to get algo from DER key."); + return -ECRYPT; + } + } + + return 0; +} + +static int negotiate_cipher(const struct oap_hdr * peer_hdr, + struct sec_config * scfg) +{ + uint8_t * id = peer_hdr->id.data; + int cli_nid; + int cli_rank; + int srv_rank; + + /* Cipher: select the strongest of client and server */ + if (peer_hdr->cipher_str != NULL) + cli_nid = (int) crypt_str_to_nid(peer_hdr->cipher_str); + else + cli_nid = NID_undef; + + if (cli_nid != NID_undef && crypt_cipher_rank(cli_nid) < 0) { + log_err_id(id, "Unsupported cipher '%s'.", + peer_hdr->cipher_str); + return -ENOTSUP; + } + + cli_rank = crypt_cipher_rank(cli_nid); + srv_rank = crypt_cipher_rank(scfg->c.nid); + + if (cli_rank > srv_rank) { + SET_KEX_CIPHER_NID(scfg, cli_nid); + log_dbg_id(id, "Selected client cipher %s.", scfg->c.str); + } else if (srv_rank > 0) { + log_dbg_id(id, "Selected server cipher %s.", scfg->c.str); + } else { + log_err_id(id, "Encryption requested, no cipher."); + return -ECRYPT; + } + + /* KDF: select the strongest of client and server */ + if (peer_hdr->kdf_nid != NID_undef + && crypt_kdf_rank(peer_hdr->kdf_nid) < 0) { + log_err_id(id, "Unsupported KDF NID %d.", + peer_hdr->kdf_nid); + return -ENOTSUP; + } + + cli_rank = crypt_kdf_rank(peer_hdr->kdf_nid); + srv_rank = crypt_kdf_rank(scfg->k.nid); + + /* Client-encap KEM bakes KDF into ciphertext; verify min. */ + if (OAP_KEX_ROLE(peer_hdr) == KEM_MODE_CLIENT_ENCAP) { + if (srv_rank > cli_rank) { + log_err_id(id, "Client KDF too weak."); + return -ECRYPT; + } + SET_KEX_KDF_NID(scfg, peer_hdr->kdf_nid); + } else if (cli_rank > srv_rank) { + SET_KEX_KDF_NID(scfg, peer_hdr->kdf_nid); + log_dbg_id(id, "Selected client KDF %s.", + md_nid_to_str(scfg->k.nid)); + } else if (srv_rank > 0) { + log_dbg_id(id, "Selected server KDF %s.", + md_nid_to_str(scfg->k.nid)); + } + + if (IS_KEX_ALGO_SET(scfg)) + log_info_id(id, "Negotiated %s + %s.", + scfg->x.str, scfg->c.str); + else + log_info_id(id, "No key exchange."); + + return 0; +} + +static int do_server_kem_decap(const struct name_info * info, + const struct oap_hdr * peer_hdr, + struct sec_config * scfg, + struct crypt_sk * sk) +{ + buffer_t ct; + void * server_pkp = NULL; + int ret; + uint8_t * id = peer_hdr->id.data; + + ret = load_server_kem_keypair(info->name, + peer_hdr->kex_flags.fmt, + &server_pkp); + if (ret < 0) + return ret; + + ct.data = peer_hdr->kex.data; + ct.len = peer_hdr->kex.len; + + ret = kex_kem_decap(server_pkp, ct, scfg->k.nid, sk->key); + + crypt_free_key(server_pkp); + + if (ret < 0) { + log_err_id(id, "Failed to decapsulate KEM."); + return -ECRYPT; + } + + log_dbg_id(id, "Client encaps: decapsulated CT."); + + return 0; +} + +static int do_server_kem_encap(const struct oap_hdr * peer_hdr, + struct sec_config * scfg, + buffer_t * kex, + struct crypt_sk * sk) +{ + buffer_t client_pk; + ssize_t ct_len; + uint8_t * id = peer_hdr->id.data; + + client_pk.data = peer_hdr->kex.data; + client_pk.len = peer_hdr->kex.len; + + if (IS_HYBRID_KEM(scfg->x.str)) + ct_len = kex_kem_encap_raw(client_pk, kex->data, + scfg->k.nid, sk->key); + else + ct_len = kex_kem_encap(client_pk, kex->data, + scfg->k.nid, sk->key); + + if (ct_len < 0) { + log_err_id(id, "Failed to encapsulate KEM."); + return -ECRYPT; + } + + kex->len = (size_t) ct_len; + + log_dbg_id(id, "Server encaps: generated CT, len=%zd.", ct_len); + + return 0; +} + +static int do_server_kex_kem(const struct name_info * info, + struct oap_hdr * peer_hdr, + struct sec_config * scfg, + buffer_t * kex, + struct crypt_sk * sk) +{ + int ret; + + scfg->x.mode = peer_hdr->kex_flags.role; + + if (scfg->x.mode == KEM_MODE_CLIENT_ENCAP) { + ret = do_server_kem_decap(info, peer_hdr, scfg, sk); + kex->len = 0; + } else { + ret = do_server_kem_encap(peer_hdr, scfg, kex, sk); + } + + return ret; +} + +static int do_server_kex_dhe(const struct oap_hdr * peer_hdr, + struct sec_config * scfg, + buffer_t * kex, + struct crypt_sk * sk) +{ + ssize_t key_len; + void * epkp; + int ret; + uint8_t * id = peer_hdr->id.data; + + key_len = kex_pkp_create(scfg, &epkp, kex->data); + if (key_len < 0) { + log_err_id(id, "Failed to generate key pair."); + return -ECRYPT; + } + + kex->len = (size_t) key_len; + + log_dbg_id(id, "Generated %s ephemeral keys.", scfg->x.str); + + ret = kex_dhe_derive(scfg, epkp, peer_hdr->kex, sk->key); + if (ret < 0) { + log_err_id(id, "Failed to derive secret."); + kex_pkp_destroy(epkp); + return -ECRYPT; + } + + kex_pkp_destroy(epkp); + + return 0; +} + +int do_server_kex(const struct name_info * info, + struct oap_hdr * peer_hdr, + struct sec_config * scfg, + buffer_t * kex, + struct crypt_sk * sk) +{ + char algo_buf[KEX_ALGO_BUFSZ]; + int srv_kex_nid; + uint8_t * id; + + id = peer_hdr->id.data; + + /* No KEX data from client */ + if (peer_hdr->kex.len == 0) { + if (IS_KEX_ALGO_SET(scfg)) { + log_warn_id(id, "KEX requested without info."); + return -ECRYPT; + } + return 0; + } + + if (negotiate_cipher(peer_hdr, scfg) < 0) + return -ECRYPT; + + /* Save server's configured KEX before overwriting */ + srv_kex_nid = scfg->x.nid; + + if (OAP_KEX_ROLE(peer_hdr) != KEM_MODE_CLIENT_ENCAP) { + /* Server encapsulation or DHE: extract algo from DER PK */ + if (get_algo_from_peer_key(peer_hdr, algo_buf) < 0) + return -ECRYPT; + + SET_KEX_ALGO(scfg, algo_buf); + + /* Reject if client KEX is weaker than server's */ + if (crypt_kex_rank(scfg->x.nid) + < crypt_kex_rank(srv_kex_nid)) { + log_err_id(id, "Client KEX %s too weak.", + scfg->x.str); + return -ECRYPT; + } + } + + /* Dispatch based on algorithm type */ + if (IS_KEM_ALGORITHM(scfg->x.str)) + return do_server_kex_kem(info, peer_hdr, scfg, kex, sk); + else + return do_server_kex_dhe(peer_hdr, scfg, kex, sk); +} + +int oap_srv_process(const struct name_info * info, + buffer_t req_buf, + buffer_t * rsp_buf, + buffer_t * data, + struct crypt_sk * sk, + bool rekey, + const buffer_t * cached_crt, + buffer_t * peer_crt) +{ + struct oap_hdr peer_hdr; + struct oap_hdr local_hdr; + struct sec_config scfg; + uint8_t kex_buf[CRYPT_KEY_BUFSZ]; + uint8_t hash_buf[MAX_HASH_SIZE]; + uint8_t kc_buf[MAX_HASH_SIZE]; + uint8_t resp_hash_buf[MAX_HASH_SIZE]; + uint8_t hs_key[SYMMKEYSZ]; + const uint8_t * seal_key = NULL; + buffer_t req_hash = BUF_INIT; + buffer_t resp_hash = BUF_INIT; + buffer_t crt_der = BUF_INIT; + buffer_t rsp_tag = BUF_INIT; + ssize_t hash_ret; + char cli_name[NAME_SIZE + 1]; + uint8_t * id; + void * pkp = NULL; + void * crt = NULL; + int req_md_nid; + int ret; + + assert(info != NULL); + assert(rsp_buf != NULL); + assert(data != NULL); + assert(sk != NULL); + + sk->nid = NID_undef; + + memset(&peer_hdr, 0, sizeof(peer_hdr)); + memset(&local_hdr, 0, sizeof(local_hdr)); + clrbuf(*rsp_buf); + + log_dbg("Processing OAP request for %s.", info->name); + + if (load_srv_credentials(info, &pkp, &crt) < 0) { + log_err("Failed to load security keys for %s.", info->name); + goto fail_cred; + } + + /* Re-key omits the cert; the peer verifies against its cache. */ + if (rekey && crt != NULL) { + crypt_free_crt(crt); + crt = NULL; + } + + if (load_srv_sec_config(info, &scfg) < 0) { + log_err("Failed to load security config for %s.", info->name); + goto fail_kex; + } + + /* Decode incoming header (NID_undef = request, no hash) */ + if (oap_hdr_decode(&peer_hdr, req_buf, NID_undef, rekey) < 0) { + log_err("Failed to decode OAP header."); + goto fail_auth; + } + + debug_oap_hdr_rcv(&peer_hdr); + + id = peer_hdr.id.data; /* Logging */ + + ret = oap_check_hdr(&peer_hdr); + if (ret == -EREPLAY) { + log_warn_id(id, "OAP header failed replay check."); + goto fail_replay; + } + if (ret < 0) { + log_err_id(id, "OAP header check failed."); + goto fail_auth; + } + + oap_hdr_init(&local_hdr, peer_hdr.id, kex_buf, *data, NID_undef); + + if (oap_auth_peer(cli_name, &scfg, &local_hdr, &peer_hdr, + cached_crt) < 0) { + log_err_id(id, "Failed to authenticate client."); + goto fail_auth; + } + + /* Surface the peer cert so the caller can cache it for re-key. */ + if (peer_crt != NULL && peer_hdr.crt.len > 0) { + peer_crt->data = malloc(peer_hdr.crt.len); + if (peer_crt->data == NULL) + goto fail_auth; + + memcpy(peer_crt->data, peer_hdr.crt.data, peer_hdr.crt.len); + peer_crt->len = peer_hdr.crt.len; + } + + if (do_server_kex(info, &peer_hdr, &scfg, &local_hdr.kex, sk) < 0) + goto fail_kex; + + sk->nid = scfg.c.nid; + + /* Build response header with hash of client request */ + local_hdr.nid = sk->nid; + + /* Use client's md_nid, defaulting to SHA-384 for PQC */ + req_md_nid = peer_hdr.md_nid != NID_undef ? + peer_hdr.md_nid : NID_sha384; + + /* Compute request hash using client's md_nid */ + hash_ret = md_digest(req_md_nid, req_buf, hash_buf); + if (hash_ret < 0) { + log_err_id(id, "Failed to hash request."); + goto fail_auth; + } + req_hash.data = hash_buf; + req_hash.len = (size_t) hash_ret; + + rsp_tag = req_hash; + + /* Bind the key to the transcript and confirm it to the client */ + if (sk->nid != NID_undef) { + if (crt != NULL && crypt_crt_der(crt, &crt_der) < 0) { + log_err_id(id, "Failed to serialize cert."); + goto fail_auth; + } + + resp_hash.data = resp_hash_buf; + + ret = oap_resp_hash(req_md_nid, local_hdr.kex, *data, + crt_der, &resp_hash); + + freebuf(crt_der); + + if (ret < 0) { + log_err_id(id, "Failed to hash response."); + goto fail_auth; + } + + /* Derive the identity-seal key before bind mutates sk->key */ + if (oap_derive_hs_key(sk, req_hash, hs_key) < 0) { + log_err_id(id, "Failed to derive handshake key."); + goto fail_auth; + } + + seal_key = hs_key; + + if (oap_bind_session_key(sk, req_hash, resp_hash, + scfg.k.nid) < 0) { + log_err_id(id, "Failed to bind session key."); + goto fail_auth; + } + + if (oap_key_confirm_tag(sk, req_hash, resp_hash, kc_buf, + (size_t) hash_ret) < 0) { + log_err_id(id, "Failed to confirm session key."); + goto fail_auth; + } + + rsp_tag.data = kc_buf; + } + + ret = oap_hdr_encode(&local_hdr, pkp, crt, &scfg, + rsp_tag, req_md_nid, seal_key); + + crypt_secure_clear(hs_key, SYMMKEYSZ); + + if (ret < 0) { + log_err_id(id, "Failed to create OAP response header."); + goto fail_auth; + } + + debug_oap_hdr_snd(&local_hdr); + + if (oap_hdr_copy_data(&peer_hdr, data) < 0) { + log_err_id(id, "Failed to copy client data."); + goto fail_data; + } + + /* Transfer ownership of response buffer */ + *rsp_buf = local_hdr.hdr; + + log_info_id(id, "OAP request processed for %s.", info->name); + + crypt_free_crt(crt); + crypt_free_key(pkp); + + return 0; + + fail_data: + oap_hdr_fini(&local_hdr); + fail_auth: + crypt_secure_clear(hs_key, SYMMKEYSZ); + crypt_free_crt(crt); + crypt_free_key(pkp); + fail_cred: + return -EAUTH; + + fail_replay: + crypt_free_crt(crt); + crypt_free_key(pkp); + return -EREPLAY; + + fail_kex: + crypt_free_crt(crt); + crypt_free_key(pkp); + return -ECRYPT; +} diff --git a/src/irmd/oap/tests/CMakeLists.txt b/src/irmd/oap/tests/CMakeLists.txt new file mode 100644 index 00000000..b534cb72 --- /dev/null +++ b/src/irmd/oap/tests/CMakeLists.txt @@ -0,0 +1,64 @@ +get_filename_component(PARENT_PATH ${CMAKE_CURRENT_SOURCE_DIR} DIRECTORY) +get_filename_component(PARENT_DIR ${PARENT_PATH} NAME) + +get_filename_component(OAP_SOURCE_DIR "${CMAKE_CURRENT_SOURCE_DIR}" DIRECTORY) +get_filename_component(OAP_BINARY_DIR "${CMAKE_CURRENT_BINARY_DIR}" DIRECTORY) +get_filename_component(IRMD_SOURCE_DIR "${OAP_SOURCE_DIR}" DIRECTORY) +get_filename_component(IRMD_BINARY_DIR "${OAP_BINARY_DIR}" DIRECTORY) + +compute_test_prefix() + +create_test_sourcelist(${PARENT_DIR}_tests test_suite.c + # Add new tests here + oap_test.c +) + +create_test_sourcelist(${PARENT_DIR}_ml_dsa_tests test_suite_ml_dsa.c + # ML-DSA-specific tests + oap_test_ml_dsa.c +) + +# OAP test needs io.c compiled with OAP_TEST_MODE +set(OAP_TEST_SOURCES + ${OAP_SOURCE_DIR}/io.c + ${OAP_SOURCE_DIR}/hdr.c + ${OAP_SOURCE_DIR}/auth.c + ${OAP_SOURCE_DIR}/srv.c + ${OAP_SOURCE_DIR}/cli.c + ${CMAKE_CURRENT_SOURCE_DIR}/common.c +) + +# Regular test executable (ECDSA) +add_executable(${PARENT_DIR}_test ${${PARENT_DIR}_tests} ${OAP_TEST_SOURCES}) +set_source_files_properties(${OAP_TEST_SOURCES} + PROPERTIES COMPILE_DEFINITIONS "OAP_TEST_MODE" +) + +disable_test_logging_for_target(${PARENT_DIR}_test) +target_link_libraries(${PARENT_DIR}_test ouroboros-irm) +target_include_directories(${PARENT_DIR}_test PRIVATE + ${IRMD_SOURCE_DIR} + ${IRMD_BINARY_DIR} +) + +# ML-DSA test executable +add_executable(${PARENT_DIR}_ml_dsa_test ${${PARENT_DIR}_ml_dsa_tests} ${OAP_TEST_SOURCES}) +set_source_files_properties(${OAP_TEST_SOURCES} + TARGET_DIRECTORY ${PARENT_DIR}_ml_dsa_test + PROPERTIES COMPILE_DEFINITIONS "OAP_TEST_MODE" +) + +disable_test_logging_for_target(${PARENT_DIR}_ml_dsa_test) +target_link_libraries(${PARENT_DIR}_ml_dsa_test ouroboros-irm) +target_include_directories(${PARENT_DIR}_ml_dsa_test PRIVATE + ${IRMD_SOURCE_DIR} + ${IRMD_BINARY_DIR} +) + +add_dependencies(build_tests ${PARENT_DIR}_test ${PARENT_DIR}_ml_dsa_test) + +# Regular tests +ouroboros_register_tests(TARGET ${PARENT_DIR}_test TESTS ${${PARENT_DIR}_tests}) + +# ML-DSA tests +ouroboros_register_tests(TARGET ${PARENT_DIR}_ml_dsa_test TESTS ${${PARENT_DIR}_ml_dsa_tests}) diff --git a/src/irmd/oap/tests/common.c b/src/irmd/oap/tests/common.c new file mode 100644 index 00000000..49ea9187 --- /dev/null +++ b/src/irmd/oap/tests/common.c @@ -0,0 +1,619 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Common test helper functions for OAP tests + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#include "common.h" + +#include <ouroboros/crypt.h> + +#include "oap.h" + +#include <string.h> +#include <stdio.h> + +int load_srv_sec_config(const struct name_info * info, + struct sec_config * cfg) +{ + (void) info; + + memset(cfg, 0, sizeof(*cfg)); + + cfg->a.req = test_cfg.srv.req_auth; + if (test_cfg.srv.cacert != NULL) + strcpy(cfg->a.cacert, test_cfg.srv.cacert); + + /* Digest is kept without kex, as in parse_sec_config */ + SET_KEX_DIGEST_NID(cfg, test_cfg.srv.md); + + if (test_cfg.srv.kex == NID_undef) + return 0; + + SET_KEX_ALGO_NID(cfg, test_cfg.srv.kex); + SET_KEX_CIPHER_NID(cfg, test_cfg.srv.cipher); + SET_KEX_KDF_NID(cfg, test_cfg.srv.kdf); + SET_KEX_KEM_MODE(cfg, test_cfg.srv.kem_mode); + + return 0; +} + +int load_cli_sec_config(const struct name_info * info, + struct sec_config * cfg) +{ + (void) info; + + memset(cfg, 0, sizeof(*cfg)); + + cfg->a.req = test_cfg.cli.req_auth; + if (test_cfg.cli.cacert != NULL) + strcpy(cfg->a.cacert, test_cfg.cli.cacert); + + /* Digest is kept without kex, as in parse_sec_config */ + SET_KEX_DIGEST_NID(cfg, test_cfg.cli.md); + + if (test_cfg.cli.kex == NID_undef) + return 0; + + SET_KEX_ALGO_NID(cfg, test_cfg.cli.kex); + SET_KEX_CIPHER_NID(cfg, test_cfg.cli.cipher); + SET_KEX_KDF_NID(cfg, test_cfg.cli.kdf); + SET_KEX_KEM_MODE(cfg, test_cfg.cli.kem_mode); + + return 0; +} + +int load_srv_credentials(const struct name_info * info, + void ** pkp, + void ** crt) +{ + (void) info; + + *pkp = NULL; + *crt = NULL; + + if (!test_cfg.srv.auth) + return 0; + + return mock_load_credentials(pkp, crt); +} + +int load_cli_credentials(const struct name_info * info, + void ** pkp, + void ** crt) +{ + (void) info; + + *pkp = NULL; + *crt = NULL; + + if (!test_cfg.cli.auth) + return 0; + + return mock_load_credentials(pkp, crt); +} + +int oap_test_setup(struct oap_test_ctx * ctx, + const char * root_ca_str, + const char * im_ca_str) +{ + memset(ctx, 0, sizeof(*ctx)); + + strcpy(ctx->srv.info.name, "test-1.unittest.o7s"); + strcpy(ctx->cli.info.name, "test-1.unittest.o7s"); + + if (oap_auth_init() < 0) { + printf("Failed to init OAP.\n"); + goto fail_init; + } + + if (crypt_load_crt_str(root_ca_str, &ctx->root_ca) < 0) { + printf("Failed to load root CA cert.\n"); + goto fail_root_ca; + } + + if (crypt_load_crt_str(im_ca_str, &ctx->im_ca) < 0) { + printf("Failed to load intermediate CA cert.\n"); + goto fail_im_ca; + } + + if (oap_auth_add_ca_crt(ctx->root_ca) < 0) { + printf("Failed to add root CA cert to store.\n"); + goto fail_add_ca; + } + + if (oap_auth_add_ca_crt(ctx->im_ca) < 0) { + printf("Failed to add intermediate CA cert to store.\n"); + goto fail_add_ca; + } + + return 0; + + fail_add_ca: + crypt_free_crt(ctx->im_ca); + fail_im_ca: + crypt_free_crt(ctx->root_ca); + fail_root_ca: + oap_auth_fini(); + fail_init: + memset(ctx, 0, sizeof(*ctx)); + return -1; +} + +void oap_test_teardown(struct oap_test_ctx * ctx) +{ + struct crypt_sk res; + buffer_t dummy = BUF_INIT; + + if (ctx->cli.state != NULL) { + res.key = ctx->cli.key; + oap_cli_complete(ctx->cli.state, &ctx->cli.info, dummy, + &ctx->data, &res, NULL, NULL); + ctx->cli.state = NULL; + } + + freebuf(ctx->data); + freebuf(ctx->resp_hdr); + freebuf(ctx->req_hdr); + freebuf(ctx->srv_crt); + freebuf(ctx->cli_crt); + + crypt_free_crt(ctx->im_ca); + crypt_free_crt(ctx->root_ca); + + oap_auth_fini(); + memset(ctx, 0, sizeof(*ctx)); +} + +int oap_cli_prepare_ctx(struct oap_test_ctx * ctx) +{ + return oap_cli_prepare(&ctx->cli.state, &ctx->cli.info, &ctx->req_hdr, + ctx->data, ctx->rekey); +} + +int oap_srv_process_ctx(struct oap_test_ctx * ctx) +{ + struct crypt_sk res = { .nid = NID_undef, .key = ctx->srv.key }; + int ret; + + ret = oap_srv_process(&ctx->srv.info, ctx->req_hdr, + &ctx->resp_hdr, &ctx->data, &res, ctx->rekey, + ctx->rekey ? &ctx->srv_crt : NULL, + ctx->rekey ? NULL : &ctx->srv_crt); + if (ret == 0) + ctx->srv.nid = res.nid; + + return ret; +} + +int oap_cli_complete_ctx(struct oap_test_ctx * ctx) +{ + struct crypt_sk res = { .nid = NID_undef, .key = ctx->cli.key }; + int ret; + + ret = oap_cli_complete(ctx->cli.state, &ctx->cli.info, ctx->resp_hdr, + &ctx->data, &res, + ctx->rekey ? &ctx->cli_crt : NULL, + ctx->rekey ? NULL : &ctx->cli_crt); + ctx->cli.state = NULL; + + if (ret == 0) + ctx->cli.nid = res.nid; + + return ret; +} + +int roundtrip_auth_only(const char * root_ca, + const char * im_ca_str) +{ + struct oap_test_ctx ctx; + + TEST_START(); + + if (oap_test_setup(&ctx, root_ca, im_ca_str) < 0) + goto fail; + + if (oap_cli_prepare_ctx(&ctx) < 0) { + printf("Client prepare failed.\n"); + goto fail_cleanup; + } + + if (oap_srv_process_ctx(&ctx) < 0) { + printf("Server process failed.\n"); + goto fail_cleanup; + } + + if (oap_cli_complete_ctx(&ctx) < 0) { + printf("Client complete failed.\n"); + goto fail_cleanup; + } + + if (ctx.cli.nid != NID_undef || ctx.srv.nid != NID_undef) { + printf("Cipher should not be set for auth-only.\n"); + goto fail_cleanup; + } + + oap_test_teardown(&ctx); + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; + + fail_cleanup: + oap_test_teardown(&ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +int roundtrip_rekey(const char * root_ca, + const char * im_ca_str) +{ + struct oap_test_ctx ctx; + + TEST_START(); + + if (oap_test_setup(&ctx, root_ca, im_ca_str) < 0) + goto fail; + + /* Initial handshake: the client caches the server cert. */ + if (oap_cli_prepare_ctx(&ctx) < 0) { + printf("Initial client prepare failed.\n"); + goto fail_cleanup; + } + + if (oap_srv_process_ctx(&ctx) < 0) { + printf("Initial server process failed.\n"); + goto fail_cleanup; + } + + if (oap_cli_complete_ctx(&ctx) < 0) { + printf("Initial client complete failed.\n"); + goto fail_cleanup; + } + + if (memcmp(ctx.cli.key, ctx.srv.key, SYMMKEYSZ) != 0) { + printf("Initial keys do not match.\n"); + goto fail_cleanup; + } + + if (ctx.cli_crt.len == 0) { + printf("Server cert was not cached for re-key.\n"); + goto fail_cleanup; + } + + /* Re-key: cert dropped on the wire, verified against the cache. */ + freebuf(ctx.req_hdr); + freebuf(ctx.resp_hdr); + freebuf(ctx.data); + + ctx.rekey = true; + + if (oap_cli_prepare_ctx(&ctx) < 0) { + printf("Re-key client prepare failed.\n"); + goto fail_cleanup; + } + + if (oap_srv_process_ctx(&ctx) < 0) { + printf("Re-key server process failed.\n"); + goto fail_cleanup; + } + + if (oap_cli_complete_ctx(&ctx) < 0) { + printf("Re-key client complete failed.\n"); + goto fail_cleanup; + } + + if (memcmp(ctx.cli.key, ctx.srv.key, SYMMKEYSZ) != 0) { + printf("Re-key keys do not match.\n"); + goto fail_cleanup; + } + + oap_test_teardown(&ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_cleanup: + oap_test_teardown(&ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +int roundtrip_rekey_badcache(const char * root_ca, + const char * im_ca_str) +{ + struct oap_test_ctx ctx; + + TEST_START(); + + if (oap_test_setup(&ctx, root_ca, im_ca_str) < 0) + goto fail; + + if (oap_cli_prepare_ctx(&ctx) < 0) { + printf("Initial client prepare failed.\n"); + goto fail_cleanup; + } + + if (oap_srv_process_ctx(&ctx) < 0) { + printf("Initial server process failed.\n"); + goto fail_cleanup; + } + + if (oap_cli_complete_ctx(&ctx) < 0) { + printf("Initial client complete failed.\n"); + goto fail_cleanup; + } + + if (ctx.cli_crt.len == 0) { + printf("Server cert was not cached.\n"); + goto fail_cleanup; + } + + /* Corrupt the cached cert: the re-key must fail closed. */ + ctx.cli_crt.data[ctx.cli_crt.len / 2] ^= 0xFF; + + freebuf(ctx.req_hdr); + freebuf(ctx.resp_hdr); + freebuf(ctx.data); + + ctx.rekey = true; + + if (oap_cli_prepare_ctx(&ctx) < 0) { + printf("Re-key client prepare failed.\n"); + goto fail_cleanup; + } + + if (oap_srv_process_ctx(&ctx) < 0) { + printf("Re-key server process failed.\n"); + goto fail_cleanup; + } + + if (oap_cli_complete_ctx(&ctx) == 0) { + printf("Re-key accepted a corrupted cached cert.\n"); + goto fail_cleanup; + } + + oap_test_teardown(&ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_cleanup: + oap_test_teardown(&ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +int roundtrip_kex_only(void) +{ + struct name_info cli_info; + struct name_info srv_info; + struct crypt_sk res; + uint8_t cli_key[SYMMKEYSZ]; + uint8_t srv_key[SYMMKEYSZ]; + int cli_nid; + int srv_nid; + buffer_t req_hdr = BUF_INIT; + buffer_t resp_hdr = BUF_INIT; + buffer_t data = BUF_INIT; + void * cli_state = NULL; + + TEST_START(); + + memset(&cli_info, 0, sizeof(cli_info)); + memset(&srv_info, 0, sizeof(srv_info)); + + strcpy(cli_info.name, "test-1.unittest.o7s"); + strcpy(srv_info.name, "test-1.unittest.o7s"); + + if (oap_auth_init() < 0) { + printf("Failed to init OAP.\n"); + goto fail; + } + + if (oap_cli_prepare(&cli_state, &cli_info, &req_hdr, + data, false) < 0) { + printf("Client prepare failed.\n"); + goto fail_cleanup; + } + + res.key = srv_key; + + if (oap_srv_process(&srv_info, req_hdr, &resp_hdr, &data, &res, + false, NULL, NULL) < 0) { + printf("Server process failed.\n"); + goto fail_cleanup; + } + + srv_nid = res.nid; + + res.key = cli_key; + + if (oap_cli_complete(cli_state, &cli_info, resp_hdr, &data, &res, + NULL, NULL) < 0) { + printf("Client complete failed.\n"); + cli_state = NULL; + goto fail_cleanup; + } + + cli_nid = res.nid; + cli_state = NULL; + + if (memcmp(cli_key, srv_key, SYMMKEYSZ) != 0) { + printf("Client and server keys do not match!\n"); + goto fail_cleanup; + } + + if (cli_nid == NID_undef || srv_nid == NID_undef) { + printf("Cipher should be set for kex-only.\n"); + goto fail_cleanup; + } + + freebuf(resp_hdr); + freebuf(req_hdr); + oap_auth_fini(); + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; + + fail_cleanup: + if (cli_state != NULL) { + res.key = cli_key; + oap_cli_complete(cli_state, &cli_info, resp_hdr, &data, + &res, NULL, NULL); + } + freebuf(resp_hdr); + freebuf(req_hdr); + oap_auth_fini(); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +int corrupted_request(const char * root_ca, + const char * im_ca_str) +{ + struct oap_test_ctx ctx; + + TEST_START(); + + if (oap_test_setup(&ctx, root_ca, im_ca_str) < 0) + goto fail; + + if (oap_cli_prepare_ctx(&ctx) < 0) { + printf("Client prepare failed.\n"); + goto fail_cleanup; + } + + /* Corrupt the request */ + if (ctx.req_hdr.len > 100) { + ctx.req_hdr.data[50] ^= 0xFF; + ctx.req_hdr.data[51] ^= 0xAA; + ctx.req_hdr.data[52] ^= 0x55; + } + + if (oap_srv_process_ctx(&ctx) == 0) { + printf("Server should reject corrupted request.\n"); + goto fail_cleanup; + } + + oap_test_teardown(&ctx); + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; + + fail_cleanup: + oap_test_teardown(&ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +int corrupted_response(const char * root_ca, + const char * im_ca_str) +{ + struct oap_test_ctx ctx; + struct crypt_sk res; + + TEST_START(); + + if (oap_test_setup(&ctx, root_ca, im_ca_str) < 0) + goto fail; + + if (oap_cli_prepare_ctx(&ctx) < 0) { + printf("Client prepare failed.\n"); + goto fail_cleanup; + } + + if (oap_srv_process_ctx(&ctx) < 0) { + printf("Server process failed.\n"); + goto fail_cleanup; + } + + /* Corrupt the response */ + if (ctx.resp_hdr.len > 100) { + ctx.resp_hdr.data[50] ^= 0xFF; + ctx.resp_hdr.data[51] ^= 0xAA; + ctx.resp_hdr.data[52] ^= 0x55; + } + + res.key = ctx.cli.key; + + if (oap_cli_complete(ctx.cli.state, &ctx.cli.info, ctx.resp_hdr, + &ctx.data, &res, NULL, NULL) == 0) { + printf("Client should reject corrupted response.\n"); + ctx.cli.state = NULL; + goto fail_cleanup; + } + + ctx.cli.state = NULL; + + oap_test_teardown(&ctx); + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; + + fail_cleanup: + oap_test_teardown(&ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +int truncated_request(const char * root_ca, + const char * im_ca_str) +{ + struct oap_test_ctx ctx; + size_t orig_len; + + TEST_START(); + + if (oap_test_setup(&ctx, root_ca, im_ca_str) < 0) + goto fail; + + if (oap_cli_prepare_ctx(&ctx) < 0) { + printf("Client prepare failed.\n"); + goto fail_cleanup; + } + + /* Truncate the request buffer */ + orig_len = ctx.req_hdr.len; + ctx.req_hdr.len = orig_len / 2; + + if (oap_srv_process_ctx(&ctx) == 0) { + printf("Server should reject truncated request.\n"); + ctx.req_hdr.len = orig_len; + goto fail_cleanup; + } + + ctx.req_hdr.len = orig_len; + + oap_test_teardown(&ctx); + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; + + fail_cleanup: + oap_test_teardown(&ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} diff --git a/src/irmd/oap/tests/common.h b/src/irmd/oap/tests/common.h new file mode 100644 index 00000000..c47096fb --- /dev/null +++ b/src/irmd/oap/tests/common.h @@ -0,0 +1,113 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Common test helper functions for OAP tests + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#ifndef IRMD_TESTS_COMMON_H +#define IRMD_TESTS_COMMON_H + +#include <ouroboros/utils.h> +#include <ouroboros/flow.h> +#include <ouroboros/name.h> +#include <test/test.h> + +#include <stdbool.h> + +/* Per-side security configuration for tests */ +struct test_sec_cfg { + int kex; /* KEX algorithm NID */ + int cipher; /* Cipher NID for encryption */ + int kdf; /* KDF NID for key derivation */ + int md; /* Digest NID for signatures */ + int kem_mode; /* KEM encapsulation mode (0 for ECDH) */ + bool auth; /* Use authentication (certificates) */ + bool req_auth; /* Require peer authentication */ + const char * cacert; /* Pinned issuing CA path */ +}; + +/* Test configuration - set by each test before running roundtrip */ +extern struct test_cfg { + struct test_sec_cfg srv; + struct test_sec_cfg cli; +} test_cfg; + +/* Each test file defines this with its own certificates */ +extern int mock_load_credentials(void ** pkp, + void ** crt); + +/* Per-side test context */ +struct oap_test_side { + struct name_info info; + struct flow_info flow; + uint8_t key[SYMMKEYSZ]; + int nid; + void * state; +}; + +/* Test context - holds all common state for OAP tests */ +struct oap_test_ctx { + struct oap_test_side srv; + struct oap_test_side cli; + + buffer_t req_hdr; + buffer_t resp_hdr; + buffer_t data; + void * root_ca; + void * im_ca; + + /* Re-key (tier iii): drop the cert, verify against the cache. */ + bool rekey; + buffer_t srv_crt; /* client cert cached by server */ + buffer_t cli_crt; /* server cert cached by client */ +}; + +int oap_test_setup(struct oap_test_ctx * ctx, + const char * root_ca_str, + const char * im_ca_str); + +void oap_test_teardown(struct oap_test_ctx * ctx); + +int oap_cli_prepare_ctx(struct oap_test_ctx * ctx); + +int oap_srv_process_ctx(struct oap_test_ctx * ctx); + +int oap_cli_complete_ctx(struct oap_test_ctx * ctx); + +int roundtrip_auth_only(const char * root_ca, + const char * im_ca_str); + +int roundtrip_rekey(const char * root_ca, + const char * im_ca_str); + +int roundtrip_rekey_badcache(const char * root_ca, + const char * im_ca_str); + +int roundtrip_kex_only(void); + +int corrupted_request(const char * root_ca, + const char * im_ca_str); + +int corrupted_response(const char * root_ca, + const char * im_ca_str); + +int truncated_request(const char * root_ca, + const char * im_ca_str); + +#endif /* IRMD_TESTS_COMMON_H */ diff --git a/src/irmd/oap/tests/oap_test.c b/src/irmd/oap/tests/oap_test.c new file mode 100644 index 00000000..fc10150b --- /dev/null +++ b/src/irmd/oap/tests/oap_test.c @@ -0,0 +1,1926 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Unit tests of Ouroboros Allocation Protocol (OAP) + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#if defined(__linux__) || defined(__CYGWIN__) + #ifndef _DEFAULT_SOURCE + #define _DEFAULT_SOURCE + #endif +#else +#define _POSIX_C_SOURCE 200809L +#endif + +#include "config.h" + +#include <ouroboros/crypt.h> +#include <ouroboros/endian.h> +#include <ouroboros/errno.h> +#include <ouroboros/flow.h> +#include <ouroboros/name.h> +#include <ouroboros/random.h> +#include <ouroboros/time.h> + +#include <test/test.h> +#include <test/certs/ecdsa.h> + +#include "oap.h" +#include "oap/auth.h" +#include "common.h" + +#include <stdbool.h> +#include <stdio.h> +#include <stdlib.h> +#include <string.h> + +#ifdef HAVE_OPENSSL +#include <openssl/evp.h> +#endif + +#define AUTH true +#define NO_AUTH false + +extern const uint16_t kex_supported_nids[]; +extern const uint16_t md_supported_nids[]; + +struct test_cfg test_cfg; + +/* Mock load - called by load_*_credentials in common.c */ +int mock_load_credentials(void ** pkp, + void ** crt) +{ + *crt = NULL; + + if (crypt_load_privkey_str(server_pkp_ec, pkp) < 0) + goto fail_privkey; + + if (crypt_load_crt_str(signed_server_crt_ec, crt) < 0) + goto fail_crt; + + return 0; + + fail_crt: + crypt_free_key(*pkp); + fail_privkey: + *pkp = NULL; + return -1; +} + +/* Stub KEM functions - ECDSA tests don't use KEM */ +int load_server_kem_keypair(__attribute__((unused)) const char * name, + __attribute__((unused)) bool raw_fmt, + __attribute__((unused)) void ** pkp) +{ + return -1; +} + +int load_server_kem_pk(__attribute__((unused)) const char * name, + __attribute__((unused)) struct sec_config * cfg, + __attribute__((unused)) buffer_t * pk) +{ + return -1; +} + +static void test_default_cfg(void) +{ + memset(&test_cfg, 0, sizeof(test_cfg)); + + /* Server: X25519, AES-256-GCM, SHA-256, with auth */ + test_cfg.srv.kex = NID_X25519; + test_cfg.srv.cipher = NID_aes_256_gcm; + test_cfg.srv.kdf = NID_sha256; + test_cfg.srv.md = NID_sha256; + test_cfg.srv.auth = AUTH; + + /* Client: same KEX/cipher/kdf/md, no auth */ + test_cfg.cli.kex = NID_X25519; + test_cfg.cli.cipher = NID_aes_256_gcm; + test_cfg.cli.kdf = NID_sha256; + test_cfg.cli.md = NID_sha256; + test_cfg.cli.auth = NO_AUTH; +} + +static int test_oap_auth_init_fini(void) +{ + TEST_START(); + + if (oap_auth_init() < 0) { + printf("Failed to init OAP.\n"); + goto fail; + } + + oap_auth_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_oap_roundtrip(int kex) +{ + struct oap_test_ctx ctx; + const char * kex_str = kex_nid_to_str(kex); + + TEST_START("(%s)", kex_str); + + test_default_cfg(); + test_cfg.srv.kex = kex; + test_cfg.cli.kex = kex; + + if (oap_test_setup(&ctx, root_ca_crt_ec, im_ca_crt_ec) < 0) + goto fail; + + if (oap_cli_prepare_ctx(&ctx) < 0) { + printf("Client prepare failed.\n"); + goto fail_cleanup; + } + + if (oap_srv_process_ctx(&ctx) < 0) { + printf("Server process failed.\n"); + goto fail_cleanup; + } + + if (oap_cli_complete_ctx(&ctx) < 0) { + printf("Client complete failed.\n"); + goto fail_cleanup; + } + + if (memcmp(ctx.cli.key, ctx.srv.key, SYMMKEYSZ) != 0) { + printf("Client and server keys do not match!\n"); + goto fail_cleanup; + } + + if (ctx.cli.nid == NID_undef || ctx.srv.nid == NID_undef) { + printf("Cipher not set in flow.\n"); + goto fail_cleanup; + } + + oap_test_teardown(&ctx); + + TEST_SUCCESS("(%s)", kex_str); + + return TEST_RC_SUCCESS; + + fail_cleanup: + oap_test_teardown(&ctx); + fail: + TEST_FAIL("(%s)", kex_str); + return TEST_RC_FAIL; +} + +static int test_oap_roundtrip_auth_only(void) +{ + memset(&test_cfg, 0, sizeof(test_cfg)); + + /* Server: auth only, no encryption */ + test_cfg.srv.md = NID_sha256; + test_cfg.srv.auth = AUTH; + + /* Client: no auth, no encryption */ + test_cfg.cli.md = NID_sha256; + test_cfg.cli.auth = NO_AUTH; + + return roundtrip_auth_only(root_ca_crt_ec, im_ca_crt_ec); +} + +static int test_oap_rekey(void) +{ + test_default_cfg(); + + return roundtrip_rekey(root_ca_crt_ec, im_ca_crt_ec); +} + +static int test_oap_rekey_badcache(void) +{ + test_default_cfg(); + + return roundtrip_rekey_badcache(root_ca_crt_ec, im_ca_crt_ec); +} + +static int test_oap_roundtrip_kex_only(void) +{ + memset(&test_cfg, 0, sizeof(test_cfg)); + + /* Server: KEX only, no auth */ + test_cfg.srv.kex = NID_X25519; + test_cfg.srv.cipher = NID_aes_256_gcm; + test_cfg.srv.kdf = NID_sha256; + test_cfg.srv.md = NID_sha256; + test_cfg.srv.auth = NO_AUTH; + + /* Client: KEX only, no auth */ + test_cfg.cli.kex = NID_X25519; + test_cfg.cli.cipher = NID_aes_256_gcm; + test_cfg.cli.kdf = NID_sha256; + test_cfg.cli.md = NID_sha256; + test_cfg.cli.auth = NO_AUTH; + + return roundtrip_kex_only(); +} + +static int test_oap_piggyback_data(void) +{ + struct oap_test_ctx ctx; + const char * cli_data_str = "client_data"; + const char * srv_data_str = "server_data"; + buffer_t srv_data = BUF_INIT; + + TEST_START(); + + test_default_cfg(); + + if (oap_test_setup(&ctx, root_ca_crt_ec, im_ca_crt_ec) < 0) + goto fail; + + /* Client prepares request with piggybacked data */ + ctx.data.len = strlen(cli_data_str); + ctx.data.data = malloc(ctx.data.len); + if (ctx.data.data == NULL) + goto fail_cleanup; + + memcpy(ctx.data.data, cli_data_str, ctx.data.len); + + if (oap_cli_prepare_ctx(&ctx) < 0) + goto fail_cleanup; + + /* Set server's response data (ctx.data will take cli data) */ + srv_data.len = strlen(srv_data_str); + srv_data.data = (uint8_t *) srv_data_str; + + freebuf(ctx.data); + ctx.data.data = srv_data.data; + ctx.data.len = srv_data.len; + srv_data.data = NULL; + srv_data.len = 0; + + if (oap_srv_process_ctx(&ctx) < 0) + goto fail_cleanup; + + /* Verify server received client's piggybacked data */ + if (ctx.data.len != strlen(cli_data_str) || + memcmp(ctx.data.data, cli_data_str, ctx.data.len) != 0) { + printf("Server did not receive correct client data.\n"); + goto fail_cleanup; + } + + freebuf(ctx.data); + + if (oap_cli_complete_ctx(&ctx) < 0) + goto fail_cleanup; + + /* Verify client received server's piggybacked data */ + if (ctx.data.len != strlen(srv_data_str) || + memcmp(ctx.data.data, srv_data_str, ctx.data.len) != 0) { + printf("Client did not receive correct server data.\n"); + goto fail_cleanup; + } + + /* Free the copied data */ + free(ctx.data.data); + ctx.data.data = NULL; + ctx.data.len = 0; + + if (memcmp(ctx.cli.key, ctx.srv.key, SYMMKEYSZ) != 0) { + printf("Client and server keys do not match!\n"); + goto fail_cleanup; + } + + oap_test_teardown(&ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_cleanup: + freebuf(srv_data); + oap_test_teardown(&ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_oap_corrupted_request(void) +{ + test_default_cfg(); + test_cfg.cli.auth = AUTH; + + return corrupted_request(root_ca_crt_ec, im_ca_crt_ec); +} + +static int test_oap_corrupted_response(void) +{ + test_default_cfg(); + + return corrupted_response(root_ca_crt_ec, im_ca_crt_ec); +} + +static int test_oap_truncated_request(void) +{ + test_default_cfg(); + + return truncated_request(root_ca_crt_ec, im_ca_crt_ec); +} + +/* After ID (16), timestamp (8), cipher_nid (2), kdf_nid (2), md (2) */ +#define OAP_CERT_LEN_OFFSET 30 +static int test_oap_inflated_length_field(void) +{ + struct oap_test_ctx ctx; + uint16_t fake; + + test_default_cfg(); + + TEST_START(); + + if (oap_test_setup(&ctx, root_ca_crt_ec, im_ca_crt_ec) < 0) + goto fail; + + if (oap_cli_prepare_ctx(&ctx) < 0) { + printf("Client prepare failed.\n"); + goto fail_cleanup; + } + + if (ctx.req_hdr.len < OAP_CERT_LEN_OFFSET + 2) { + printf("Request too short for test.\n"); + goto fail_cleanup; + } + + /* Set cert length to claim more bytes than packet contains */ + fake = hton16(60000); + memcpy(ctx.req_hdr.data + OAP_CERT_LEN_OFFSET, &fake, sizeof(fake)); + + if (oap_srv_process_ctx(&ctx) == 0) { + printf("Server should reject inflated length field.\n"); + goto fail_cleanup; + } + + oap_test_teardown(&ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_cleanup: + oap_test_teardown(&ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* Attacker claims cert is smaller - causes misparse of subsequent fields */ +static int test_oap_deflated_length_field(void) +{ + struct oap_test_ctx ctx; + uint16_t fake; + + test_default_cfg(); + + TEST_START(); + + if (oap_test_setup(&ctx, root_ca_crt_ec, im_ca_crt_ec) < 0) + goto fail; + + if (oap_cli_prepare_ctx(&ctx) < 0) { + printf("Client prepare failed.\n"); + goto fail_cleanup; + } + + if (ctx.req_hdr.len < OAP_CERT_LEN_OFFSET + 2) { + printf("Request too short for test.\n"); + goto fail_cleanup; + } + + /* Set cert length to claim fewer bytes - will misparse rest */ + fake = hton16(1); + memcpy(ctx.req_hdr.data + OAP_CERT_LEN_OFFSET, &fake, sizeof(fake)); + + if (oap_srv_process_ctx(&ctx) == 0) { + printf("Server should reject deflated length field.\n"); + goto fail_cleanup; + } + + oap_test_teardown(&ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_cleanup: + oap_test_teardown(&ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* Header field offsets for byte manipulation */ +#define OAP_CIPHER_NID_OFFSET 24 +#define OAP_KEX_LEN_OFFSET 32 + +/* Server rejects request when cipher NID set but no KEX data provided */ +static int test_oap_nid_without_kex(void) +{ + struct oap_test_ctx ctx; + uint16_t cipher_nid; + uint16_t zero = 0; + + TEST_START(); + + /* Configure unsigned KEX-only mode */ + memset(&test_cfg, 0, sizeof(test_cfg)); + test_cfg.srv.kex = NID_X25519; + test_cfg.srv.cipher = NID_aes_256_gcm; + test_cfg.srv.kdf = NID_sha256; + test_cfg.srv.md = NID_sha256; + test_cfg.srv.auth = NO_AUTH; + test_cfg.cli.kex = NID_X25519; + test_cfg.cli.cipher = NID_aes_256_gcm; + test_cfg.cli.kdf = NID_sha256; + test_cfg.cli.md = NID_sha256; + test_cfg.cli.auth = NO_AUTH; + + if (oap_test_setup(&ctx, root_ca_crt_ec, im_ca_crt_ec) < 0) + goto fail; + + if (oap_cli_prepare_ctx(&ctx) < 0) { + printf("Client prepare failed.\n"); + goto fail_cleanup; + } + + /* Tamper: keep cipher_nid but set kex_len=0, truncate KEX data */ + cipher_nid = hton16(NID_aes_256_gcm); + memcpy(ctx.req_hdr.data + OAP_CIPHER_NID_OFFSET, &cipher_nid, + sizeof(cipher_nid)); + memcpy(ctx.req_hdr.data + OAP_KEX_LEN_OFFSET, &zero, sizeof(zero)); + ctx.req_hdr.len = 36; /* Fixed header only, no KEX data */ + + if (oap_srv_process_ctx(&ctx) == 0) { + printf("Server should reject cipher NID without KEX data.\n"); + goto fail_cleanup; + } + + oap_test_teardown(&ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_cleanup: + oap_test_teardown(&ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* Server rejects OAP request with unsupported cipher NID */ +static int test_oap_unsupported_nid(void) +{ + struct oap_test_ctx ctx; + uint16_t bad_nid; + + TEST_START(); + + /* Configure unsigned KEX-only mode */ + memset(&test_cfg, 0, sizeof(test_cfg)); + test_cfg.srv.kex = NID_X25519; + test_cfg.srv.cipher = NID_aes_256_gcm; + test_cfg.srv.kdf = NID_sha256; + test_cfg.srv.md = NID_sha256; + test_cfg.srv.auth = NO_AUTH; + test_cfg.cli.kex = NID_X25519; + test_cfg.cli.cipher = NID_aes_256_gcm; + test_cfg.cli.kdf = NID_sha256; + test_cfg.cli.md = NID_sha256; + test_cfg.cli.auth = NO_AUTH; + + if (oap_test_setup(&ctx, root_ca_crt_ec, im_ca_crt_ec) < 0) + goto fail; + + if (oap_cli_prepare_ctx(&ctx) < 0) { + printf("Client prepare failed.\n"); + goto fail_cleanup; + } + + /* Tamper: set cipher_nid to unsupported value */ + bad_nid = hton16(9999); + memcpy(ctx.req_hdr.data + OAP_CIPHER_NID_OFFSET, &bad_nid, + sizeof(bad_nid)); + + if (oap_srv_process_ctx(&ctx) == 0) { + printf("Server should reject unsupported cipher NID.\n"); + goto fail_cleanup; + } + + oap_test_teardown(&ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_cleanup: + oap_test_teardown(&ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* Client rejects a response whose key-confirmation tag is tampered */ +static int test_oap_key_confirm_mismatch(void) +{ + struct oap_test_ctx ctx; + + TEST_START(); + + /* Unauthenticated + encrypted: response unsigned, KC is the gate */ + memset(&test_cfg, 0, sizeof(test_cfg)); + test_cfg.srv.kex = NID_X25519; + test_cfg.srv.cipher = NID_aes_256_gcm; + test_cfg.srv.kdf = NID_sha256; + test_cfg.srv.md = NID_sha256; + test_cfg.srv.auth = NO_AUTH; + test_cfg.cli.kex = NID_X25519; + test_cfg.cli.cipher = NID_aes_256_gcm; + test_cfg.cli.kdf = NID_sha256; + test_cfg.cli.md = NID_sha256; + test_cfg.cli.auth = NO_AUTH; + + if (oap_test_setup(&ctx, root_ca_crt_ec, im_ca_crt_ec) < 0) + goto fail; + + if (oap_cli_prepare_ctx(&ctx) < 0) { + printf("Client prepare failed.\n"); + goto fail_cleanup; + } + + if (oap_srv_process_ctx(&ctx) < 0) { + printf("Server process failed.\n"); + goto fail_cleanup; + } + + /* The key-confirm tag is the last field of an unsigned response */ + ctx.resp_hdr.data[ctx.resp_hdr.len - 1] ^= 0xFF; + + if (oap_cli_complete_ctx(&ctx) == 0) { + printf("Client accepted a bad key-confirmation tag.\n"); + goto fail_cleanup; + } + + oap_test_teardown(&ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_cleanup: + oap_test_teardown(&ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_oap_roundtrip_all(void) +{ + int ret = 0; + int i; + + for (i = 0; kex_supported_nids[i] != NID_undef; i++) { + const char * algo = kex_nid_to_str(kex_supported_nids[i]); + + /* Skip KEM algorithms - tested in oap_test_ml_dsa */ + if (IS_KEM_ALGORITHM(algo)) + continue; + + ret |= test_oap_roundtrip(kex_supported_nids[i]); + } + + return ret; +} + +/* Cipher negotiation - strongest cipher and KDF are selected */ +static int test_oap_cipher_mismatch(void) +{ + struct oap_test_ctx ctx; + + TEST_START(); + + memset(&test_cfg, 0, sizeof(test_cfg)); + + /* Server: AES-128-GCM, SHA-256 */ + test_cfg.srv.kex = NID_X25519; + test_cfg.srv.cipher = NID_aes_128_gcm; + test_cfg.srv.kdf = NID_sha256; + test_cfg.srv.md = NID_sha256; + test_cfg.srv.auth = AUTH; + + /* Client: AES-256-GCM, SHA-512 */ + test_cfg.cli.kex = NID_X25519; + test_cfg.cli.cipher = NID_aes_256_gcm; + test_cfg.cli.kdf = NID_sha512; + test_cfg.cli.md = NID_sha256; + test_cfg.cli.auth = NO_AUTH; + + if (oap_test_setup(&ctx, root_ca_crt_ec, im_ca_crt_ec) < 0) + goto fail; + + if (oap_cli_prepare_ctx(&ctx) < 0) { + printf("Client prepare failed.\n"); + goto fail_cleanup; + } + + if (oap_srv_process_ctx(&ctx) < 0) { + printf("Server process failed.\n"); + goto fail_cleanup; + } + + if (oap_cli_complete_ctx(&ctx) < 0) { + printf("Client complete failed.\n"); + goto fail_cleanup; + } + + /* Verify: both should have the strongest cipher */ + if (ctx.srv.nid != NID_aes_256_gcm) { + printf("Server cipher mismatch: expected %s, got %s\n", + crypt_nid_to_str(NID_aes_256_gcm), + crypt_nid_to_str(ctx.srv.nid)); + goto fail_cleanup; + } + + if (ctx.cli.nid != NID_aes_256_gcm) { + printf("Client cipher mismatch: expected %s, got %s\n", + crypt_nid_to_str(NID_aes_256_gcm), + crypt_nid_to_str(ctx.cli.nid)); + goto fail_cleanup; + } + + /* Parse response header to check negotiated KDF */ + if (ctx.resp_hdr.len > 26) { + uint16_t resp_kdf_nid; + /* KDF NID at offset 26: ID(16) + ts(8) + cipher(2) */ + resp_kdf_nid = ntoh16(*(uint16_t *)(ctx.resp_hdr.data + 26)); + + if (resp_kdf_nid != NID_sha512) { + printf("Response KDF mismatch: expected %s, got %s\n", + md_nid_to_str(NID_sha512), + md_nid_to_str(resp_kdf_nid)); + goto fail_cleanup; + } + } + + oap_test_teardown(&ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_cleanup: + oap_test_teardown(&ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* Server encryption, client none: server rejects (no KEX data) */ +static int test_oap_srv_enc_cli_none(void) +{ + struct oap_test_ctx ctx; + + TEST_START(); + + memset(&test_cfg, 0, sizeof(test_cfg)); + + /* Server: encryption configured */ + test_cfg.srv.kex = NID_X25519; + test_cfg.srv.cipher = NID_aes_256_gcm; + test_cfg.srv.kdf = NID_sha256; + test_cfg.srv.md = NID_sha256; + test_cfg.srv.auth = AUTH; + + /* Client: no encryption */ + test_cfg.cli.md = NID_sha256; + test_cfg.cli.auth = NO_AUTH; + + if (oap_test_setup(&ctx, root_ca_crt_ec, im_ca_crt_ec) < 0) + goto fail; + + if (oap_cli_prepare_ctx(&ctx) < 0) { + printf("Client prepare failed.\n"); + goto fail_cleanup; + } + + /* Server should reject: KEX required but client sent none */ + if (oap_srv_process_ctx(&ctx) == 0) { + printf("Server should have rejected.\n"); + goto fail_cleanup; + } + + oap_test_teardown(&ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_cleanup: + oap_test_teardown(&ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* Client encryption, server none: use client settings */ +static int test_oap_cli_enc_srv_none(void) +{ + struct oap_test_ctx ctx; + + TEST_START(); + + memset(&test_cfg, 0, sizeof(test_cfg)); + + /* Server: no encryption configured */ + test_cfg.srv.md = NID_sha256; + test_cfg.srv.auth = AUTH; + + /* Client: encryption configured */ + test_cfg.cli.kex = NID_X25519; + test_cfg.cli.cipher = NID_aes_256_gcm; + test_cfg.cli.kdf = NID_sha256; + test_cfg.cli.md = NID_sha256; + test_cfg.cli.auth = NO_AUTH; + + if (oap_test_setup(&ctx, root_ca_crt_ec, im_ca_crt_ec) < 0) + goto fail; + + if (oap_cli_prepare_ctx(&ctx) < 0) { + printf("Client prepare failed.\n"); + goto fail_cleanup; + } + + if (oap_srv_process_ctx(&ctx) < 0) { + printf("Server process failed.\n"); + goto fail_cleanup; + } + + if (oap_cli_complete_ctx(&ctx) < 0) { + printf("Client complete failed.\n"); + goto fail_cleanup; + } + + if (memcmp(ctx.cli.key, ctx.srv.key, SYMMKEYSZ) != 0) { + printf("Key mismatch.\n"); + goto fail_cleanup; + } + + if (ctx.cli.nid != NID_aes_256_gcm) { + printf("Expected %s, got %s.\n", + crypt_nid_to_str(NID_aes_256_gcm), + crypt_nid_to_str(ctx.cli.nid)); + goto fail_cleanup; + } + + if (ctx.srv.nid != NID_aes_256_gcm) { + printf("Expected %s, got %s.\n", + crypt_nid_to_str(NID_aes_256_gcm), + crypt_nid_to_str(ctx.srv.nid)); + goto fail_cleanup; + } + + oap_test_teardown(&ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_cleanup: + oap_test_teardown(&ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* Unauthenticated server: client floor-rejects a downgraded cipher */ +static int test_oap_cli_rejects_downgrade(void) +{ + struct oap_test_ctx ctx; + uint16_t weak; + + TEST_START(); + + memset(&test_cfg, 0, sizeof(test_cfg)); + + test_cfg.srv.kex = NID_X25519; + test_cfg.srv.cipher = NID_aes_256_gcm; + test_cfg.srv.kdf = NID_sha256; + test_cfg.srv.md = NID_sha256; + test_cfg.srv.auth = NO_AUTH; + + test_cfg.cli.kex = NID_X25519; + test_cfg.cli.cipher = NID_aes_256_gcm; + test_cfg.cli.kdf = NID_sha256; + test_cfg.cli.md = NID_sha256; + test_cfg.cli.auth = NO_AUTH; + + if (oap_test_setup(&ctx, root_ca_crt_ec, im_ca_crt_ec) < 0) + goto fail; + + if (oap_cli_prepare_ctx(&ctx) < 0) { + printf("Client prepare failed.\n"); + goto fail_cleanup; + } + + if (oap_srv_process_ctx(&ctx) < 0) { + printf("Server process failed.\n"); + goto fail_cleanup; + } + + /* Tamper: replace cipher NID with weaker one */ + weak = hton16(NID_aes_128_gcm); + memcpy(ctx.resp_hdr.data + OAP_CIPHER_NID_OFFSET, + &weak, sizeof(weak)); + + /* Client should reject the downgraded cipher */ + if (oap_cli_complete_ctx(&ctx) == 0) { + printf("Client accepted downgrade.\n"); + goto fail_cleanup; + } + + oap_test_teardown(&ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_cleanup: + oap_test_teardown(&ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* + * Suite binding: a cipher swapped to a higher rank clears the client floor + * check, but the bound key commits to the negotiated suite, so the swap must + * still fail key confirmation. + */ +static int test_oap_cli_rejects_suite_swap(void) +{ + struct oap_test_ctx ctx; + uint16_t swap; + + TEST_START(); + + memset(&test_cfg, 0, sizeof(test_cfg)); + + /* Both AES-128-GCM: a swap to AES-256 outranks the client floor */ + test_cfg.srv.kex = NID_X25519; + test_cfg.srv.cipher = NID_aes_128_gcm; + test_cfg.srv.kdf = NID_sha256; + test_cfg.srv.md = NID_sha256; + test_cfg.srv.auth = NO_AUTH; + test_cfg.cli.kex = NID_X25519; + test_cfg.cli.cipher = NID_aes_128_gcm; + test_cfg.cli.kdf = NID_sha256; + test_cfg.cli.md = NID_sha256; + test_cfg.cli.auth = NO_AUTH; + + if (oap_test_setup(&ctx, root_ca_crt_ec, im_ca_crt_ec) < 0) + goto fail; + + if (oap_cli_prepare_ctx(&ctx) < 0) { + printf("Client prepare failed.\n"); + goto fail_cleanup; + } + + if (oap_srv_process_ctx(&ctx) < 0) { + printf("Server process failed.\n"); + goto fail_cleanup; + } + + /* Swap the response cipher to a higher-ranked one */ + swap = hton16(NID_aes_256_gcm); + memcpy(ctx.resp_hdr.data + OAP_CIPHER_NID_OFFSET, + &swap, sizeof(swap)); + + if (oap_cli_complete_ctx(&ctx) == 0) { + printf("Client accepted a swapped cipher suite.\n"); + goto fail_cleanup; + } + + oap_test_teardown(&ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_cleanup: + oap_test_teardown(&ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* Server rejects client with weaker KEX algorithm */ +static int test_oap_srv_rejects_weak_kex(void) +{ + struct oap_test_ctx ctx; + + TEST_START(); + + memset(&test_cfg, 0, sizeof(test_cfg)); + + /* Server: secp521r1 (strong) */ + test_cfg.srv.kex = NID_secp521r1; + test_cfg.srv.cipher = NID_aes_256_gcm; + test_cfg.srv.kdf = NID_sha256; + test_cfg.srv.md = NID_sha256; + test_cfg.srv.auth = AUTH; + + /* Client: ffdhe2048 (weakest) */ + test_cfg.cli.kex = NID_ffdhe2048; + test_cfg.cli.cipher = NID_aes_256_gcm; + test_cfg.cli.kdf = NID_sha256; + test_cfg.cli.md = NID_sha256; + test_cfg.cli.auth = NO_AUTH; + + if (oap_test_setup(&ctx, root_ca_crt_ec, im_ca_crt_ec) < 0) + goto fail; + + if (oap_cli_prepare_ctx(&ctx) < 0) { + printf("Client prepare failed.\n"); + goto fail_cleanup; + } + + /* Server should reject: client KEX too weak */ + if (oap_srv_process_ctx(&ctx) == 0) { + printf("Server should reject weak KEX.\n"); + goto fail_cleanup; + } + + oap_test_teardown(&ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_cleanup: + oap_test_teardown(&ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* Test roundtrip with different signature digest algorithms */ +static int test_oap_roundtrip_md(int md) +{ + struct oap_test_ctx ctx; + const char * md_str = md_nid_to_str(md); + + TEST_START("(%s)", md_str ? md_str : "default"); + + memset(&test_cfg, 0, sizeof(test_cfg)); + + /* Server: auth + KEX with specified md */ + test_cfg.srv.kex = NID_X25519; + test_cfg.srv.cipher = NID_aes_256_gcm; + test_cfg.srv.kdf = NID_sha256; + test_cfg.srv.md = md; + test_cfg.srv.auth = AUTH; + + /* Client: no auth */ + test_cfg.cli.kex = NID_X25519; + test_cfg.cli.cipher = NID_aes_256_gcm; + test_cfg.cli.kdf = NID_sha256; + test_cfg.cli.md = md; + test_cfg.cli.auth = NO_AUTH; + + if (oap_test_setup(&ctx, root_ca_crt_ec, im_ca_crt_ec) < 0) + goto fail; + + if (oap_cli_prepare_ctx(&ctx) < 0) { + printf("Client prepare failed.\n"); + goto fail_cleanup; + } + + if (oap_srv_process_ctx(&ctx) < 0) { + printf("Server process failed.\n"); + goto fail_cleanup; + } + + if (oap_cli_complete_ctx(&ctx) < 0) { + printf("Client complete failed.\n"); + goto fail_cleanup; + } + + if (memcmp(ctx.cli.key, ctx.srv.key, SYMMKEYSZ) != 0) { + printf("Client and server keys do not match!\n"); + goto fail_cleanup; + } + + oap_test_teardown(&ctx); + + TEST_SUCCESS("(%s)", md_str ? md_str : "default"); + + return TEST_RC_SUCCESS; + + fail_cleanup: + oap_test_teardown(&ctx); + fail: + TEST_FAIL("(%s)", md_str ? md_str : "default"); + return TEST_RC_FAIL; +} + +static int test_oap_roundtrip_md_all(void) +{ + int ret = 0; + int i; + + /* Test with default */ + ret |= test_oap_roundtrip_md(NID_undef); + + /* Test with all supported digest NIDs */ + for (i = 0; md_supported_nids[i] != NID_undef; i++) + ret |= test_oap_roundtrip_md(md_supported_nids[i]); + + return ret; +} + +/* Timestamp is at offset 16 (after the 16-byte ID) */ +#define OAP_TIMESTAMP_OFFSET 16 +/* Test that packets with outdated timestamps are rejected */ +static int test_oap_outdated_packet(void) +{ + struct oap_test_ctx ctx; + struct timespec old_ts; + uint64_t old_stamp; + + test_default_cfg(); + + TEST_START(); + + if (oap_test_setup(&ctx, root_ca_crt_ec, im_ca_crt_ec) < 0) + goto fail; + + if (oap_cli_prepare_ctx(&ctx) < 0) { + printf("Client prepare failed.\n"); + goto fail_cleanup; + } + + if (ctx.req_hdr.len < OAP_TIMESTAMP_OFFSET + sizeof(uint64_t)) { + printf("Request too short for test.\n"); + goto fail_cleanup; + } + + /* Set timestamp to 30 seconds in the past (> 20s replay timer) */ + clock_gettime(CLOCK_REALTIME, &old_ts); + old_ts.tv_sec -= OAP_REPLAY_TIMER + 10; + old_stamp = hton64(TS_TO_UINT64(old_ts)); + memcpy(ctx.req_hdr.data + OAP_TIMESTAMP_OFFSET, &old_stamp, + sizeof(old_stamp)); + + if (oap_srv_process_ctx(&ctx) == 0) { + printf("Server should reject outdated packet.\n"); + goto fail_cleanup; + } + + oap_test_teardown(&ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_cleanup: + oap_test_teardown(&ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* Test that packets from the future are rejected */ +static int test_oap_future_packet(void) +{ + struct oap_test_ctx ctx; + struct timespec future_ts; + uint64_t future_stamp; + + test_default_cfg(); + + TEST_START(); + + if (oap_test_setup(&ctx, root_ca_crt_ec, im_ca_crt_ec) < 0) + goto fail; + + if (oap_cli_prepare_ctx(&ctx) < 0) { + printf("Client prepare failed.\n"); + goto fail_cleanup; + } + + if (ctx.req_hdr.len < OAP_TIMESTAMP_OFFSET + sizeof(uint64_t)) { + printf("Request too short for test.\n"); + goto fail_cleanup; + } + + /* Set timestamp to 1 second in the future (> 100ms slack) */ + clock_gettime(CLOCK_REALTIME, &future_ts); + future_ts.tv_sec += 1; + future_stamp = hton64(TS_TO_UINT64(future_ts)); + memcpy(ctx.req_hdr.data + OAP_TIMESTAMP_OFFSET, &future_stamp, + sizeof(future_stamp)); + + if (oap_srv_process_ctx(&ctx) == 0) { + printf("Server should reject future packet.\n"); + goto fail_cleanup; + } + + oap_test_teardown(&ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_cleanup: + oap_test_teardown(&ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* Test that replayed packets (same ID + timestamp) are rejected */ +static int test_oap_replay_packet(void) +{ + struct oap_test_ctx ctx; + buffer_t saved_req; + + test_default_cfg(); + + TEST_START(); + + if (oap_test_setup(&ctx, root_ca_crt_ec, im_ca_crt_ec) < 0) + goto fail; + + if (oap_cli_prepare_ctx(&ctx) < 0) { + printf("Client prepare failed.\n"); + goto fail_cleanup; + } + + /* Save the request for replay */ + saved_req.len = ctx.req_hdr.len; + saved_req.data = malloc(saved_req.len); + if (saved_req.data == NULL) { + printf("Failed to allocate saved request.\n"); + goto fail_cleanup; + } + memcpy(saved_req.data, ctx.req_hdr.data, saved_req.len); + + /* First request should succeed */ + if (oap_srv_process_ctx(&ctx) < 0) { + printf("First request should succeed.\n"); + free(saved_req.data); + goto fail_cleanup; + } + + /* Free response from first request before replay */ + freebuf(ctx.resp_hdr); + + /* Restore the saved request for replay */ + freebuf(ctx.req_hdr); + ctx.req_hdr = saved_req; + + /* Replay must return -EREPLAY so callers can drop silently. */ + if (oap_srv_process_ctx(&ctx) != -EREPLAY) { + printf("Replayed packet rejection != -EREPLAY.\n"); + goto fail_cleanup; + } + + oap_test_teardown(&ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_cleanup: + oap_test_teardown(&ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* Encode a distinct OAP session ID from an index */ +static void make_id(uint8_t * id, + size_t idx) +{ + memset(id, 0, OAP_ID_SIZE); + memcpy(id, &idx, sizeof(idx)); +} + +/* + * Replay cache fails closed at capacity: a flood is rejected and no genuine + * entry is evicted (so it cannot be replayed). + */ +static int test_oap_replay_cap(void) +{ + struct oap_hdr h; + struct timespec now; + uint8_t id[OAP_ID_SIZE]; + uint64_t stamp; + size_t i; + + TEST_START(); + + if (oap_auth_init() < 0) { + printf("Failed to init OAP.\n"); + goto fail; + } + + clock_gettime(CLOCK_REALTIME, &now); + stamp = TS_TO_UINT64(now); + + memset(&h, 0, sizeof(h)); + h.id.data = id; + h.id.len = OAP_ID_SIZE; + h.timestamp = stamp; + + /* Fill one generation bucket to capacity with distinct IDs */ + for (i = 0; i < OAP_REPLAY_MAX; i++) { + make_id(id, i); + if (oap_check_hdr(&h) != 0) { + printf("Distinct header %zu rejected.\n", i); + goto fail_fini; + } + } + + /* One past capacity fails closed (rejected, not evict-oldest) */ + make_id(id, OAP_REPLAY_MAX); + if (oap_check_hdr(&h) != -EAUTH) { + printf("Header past capacity not fail-closed.\n"); + goto fail_fini; + } + + /* No genuine entry was evicted: the oldest still reads as a replay */ + make_id(id, 0); + if (oap_check_hdr(&h) != -EREPLAY) { + printf("Genuine entry evicted under flood.\n"); + goto fail_fini; + } + + oap_auth_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_fini: + oap_auth_fini(); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* + * Distinct timestamp generations use separate buckets and are detected + * independently (covers the multi-generation / rotation path). + */ +static int test_oap_replay_generations(void) +{ + struct oap_hdr h; + struct timespec now; + uint8_t id[OAP_ID_SIZE]; + uint64_t cur; + uint64_t gen_ns; + uint64_t stamp_a; + uint64_t stamp_b; + + TEST_START(); + + if (oap_auth_init() < 0) { + printf("Failed to init OAP.\n"); + goto fail; + } + + clock_gettime(CLOCK_REALTIME, &now); + cur = TS_TO_UINT64(now); + gen_ns = (uint64_t) OAP_REPLAY_TIMER * BILLION; + + /* stamp_a in the current generation, stamp_b one generation older */ + stamp_a = cur; + stamp_b = (cur / gen_ns) * gen_ns - 1; + + memset(&h, 0, sizeof(h)); + h.id.data = id; + h.id.len = OAP_ID_SIZE; + make_id(id, 1); + + /* First sighting in each generation is accepted */ + h.timestamp = stamp_a; + if (oap_check_hdr(&h) != 0) { + printf("Gen-A header rejected.\n"); + goto fail_fini; + } + + h.timestamp = stamp_b; + if (oap_check_hdr(&h) != 0) { + printf("Gen-B header rejected.\n"); + goto fail_fini; + } + + /* Each generation independently detects its own replay */ + h.timestamp = stamp_a; + if (oap_check_hdr(&h) != -EREPLAY) { + printf("Gen-A replay not detected.\n"); + goto fail_fini; + } + + h.timestamp = stamp_b; + if (oap_check_hdr(&h) != -EREPLAY) { + printf("Gen-B replay not detected.\n"); + goto fail_fini; + } + + oap_auth_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_fini: + oap_auth_fini(); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* Server rejects client certificate when root CA is missing from store */ +static int test_oap_missing_root_ca(void) +{ + struct oap_test_ctx ctx; + void * im_ca = NULL; + + test_default_cfg(); + + TEST_START(); + + memset(&ctx, 0, sizeof(ctx)); + + strcpy(ctx.srv.info.name, "test-1.unittest.o7s"); + strcpy(ctx.cli.info.name, "test-1.unittest.o7s"); + + if (oap_auth_init() < 0) { + printf("Failed to init OAP.\n"); + goto fail; + } + + /* Load intermediate CA but intentionally omit the root CA */ + if (crypt_load_crt_str(im_ca_crt_ec, &im_ca) < 0) { + printf("Failed to load intermediate CA cert.\n"); + goto fail_fini; + } + + ctx.im_ca = im_ca; + + if (oap_auth_add_ca_crt(im_ca) < 0) { + printf("Failed to add intermediate CA cert to store.\n"); + goto fail_fini; + } + + if (oap_cli_prepare_ctx(&ctx) < 0) { + printf("Client prepare failed.\n"); + goto fail_fini; + } + + /* Server processes and signs response - succeeds without root CA */ + if (oap_srv_process_ctx(&ctx) < 0) { + printf("Server process failed.\n"); + goto fail_teardown; + } + + /* Client verifies server certificate against trust store: + * should reject because root CA is not in the store */ + if (oap_cli_complete_ctx(&ctx) == 0) { + printf("Client should reject without root CA.\n"); + goto fail_teardown; + } + + oap_test_teardown(&ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_teardown: + oap_test_teardown(&ctx); + TEST_FAIL(); + return TEST_RC_FAIL; + fail_fini: + crypt_free_crt(im_ca); + oap_auth_fini(); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* Test that client rejects server with wrong certificate name */ +static int test_oap_server_name_mismatch(void) +{ + struct oap_test_ctx ctx; + + test_default_cfg(); + + TEST_START(); + + if (oap_test_setup(&ctx, root_ca_crt_ec, im_ca_crt_ec) < 0) + goto fail; + + /* Set client's expected name to something different from cert name */ + strcpy(ctx.cli.info.name, "wrong.server.name"); + + if (oap_cli_prepare_ctx(&ctx) < 0) { + printf("Client prepare failed.\n"); + goto fail_cleanup; + } + + if (oap_srv_process_ctx(&ctx) < 0) { + printf("Server process failed.\n"); + goto fail_cleanup; + } + + /* Client should reject due to name mismatch */ + if (oap_cli_complete_ctx(&ctx) == 0) { + printf("Client should reject server with wrong cert name.\n"); + goto fail_cleanup; + } + + oap_test_teardown(&ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_cleanup: + oap_test_teardown(&ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* Client requiring auth rejects a response without certificate */ +static int test_oap_cli_requires_srv_auth(void) +{ + struct oap_test_ctx ctx; + + test_default_cfg(); + test_cfg.srv.auth = NO_AUTH; + test_cfg.cli.req_auth = true; + + TEST_START(); + + if (oap_test_setup(&ctx, root_ca_crt_ec, im_ca_crt_ec) < 0) + goto fail; + + if (oap_cli_prepare_ctx(&ctx) < 0) { + printf("Client prepare failed.\n"); + goto fail_cleanup; + } + + if (oap_srv_process_ctx(&ctx) < 0) { + printf("Server process failed.\n"); + goto fail_cleanup; + } + + if (oap_cli_complete_ctx(&ctx) == 0) { + printf("Client should reject unauthenticated server.\n"); + goto fail_cleanup; + } + + oap_test_teardown(&ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_cleanup: + oap_test_teardown(&ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* Server requiring auth rejects a request without certificate */ +static int test_oap_srv_requires_cli_auth(void) +{ + struct oap_test_ctx ctx; + + test_default_cfg(); + test_cfg.srv.req_auth = true; + + TEST_START(); + + if (oap_test_setup(&ctx, root_ca_crt_ec, im_ca_crt_ec) < 0) + goto fail; + + if (oap_cli_prepare_ctx(&ctx) < 0) { + printf("Client prepare failed.\n"); + goto fail_cleanup; + } + + if (oap_srv_process_ctx(&ctx) == 0) { + printf("Server should reject unauthenticated client.\n"); + goto fail_cleanup; + } + + oap_test_teardown(&ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_cleanup: + oap_test_teardown(&ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* Roundtrip succeeds when both sides require and provide auth */ +static int test_oap_mutual_req_auth(void) +{ + struct oap_test_ctx ctx; + + test_default_cfg(); + test_cfg.srv.req_auth = true; + test_cfg.cli.auth = AUTH; + test_cfg.cli.req_auth = true; + + TEST_START(); + + if (oap_test_setup(&ctx, root_ca_crt_ec, im_ca_crt_ec) < 0) + goto fail; + + if (oap_cli_prepare_ctx(&ctx) < 0) { + printf("Client prepare failed.\n"); + goto fail_cleanup; + } + + if (oap_srv_process_ctx(&ctx) < 0) { + printf("Server process failed.\n"); + goto fail_cleanup; + } + + if (oap_cli_complete_ctx(&ctx) < 0) { + printf("Client complete failed.\n"); + goto fail_cleanup; + } + + if (memcmp(ctx.cli.key, ctx.srv.key, SYMMKEYSZ) != 0) { + printf("Client and server keys do not match!\n"); + goto fail_cleanup; + } + + oap_test_teardown(&ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_cleanup: + oap_test_teardown(&ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* Client rejects a server signature with a different digest */ +static int test_oap_cli_rejects_md_mismatch(void) +{ + struct oap_test_ctx ctx; + + test_default_cfg(); + test_cfg.srv.md = NID_sha384; + + TEST_START(); + + if (oap_test_setup(&ctx, root_ca_crt_ec, im_ca_crt_ec) < 0) + goto fail; + + if (oap_cli_prepare_ctx(&ctx) < 0) { + printf("Client prepare failed.\n"); + goto fail_cleanup; + } + + if (oap_srv_process_ctx(&ctx) < 0) { + printf("Server process failed.\n"); + goto fail_cleanup; + } + + if (oap_cli_complete_ctx(&ctx) == 0) { + printf("Client should reject digest mismatch.\n"); + goto fail_cleanup; + } + + oap_test_teardown(&ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_cleanup: + oap_test_teardown(&ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* Server rejects a client signature with a different digest */ +static int test_oap_srv_rejects_md_mismatch(void) +{ + struct oap_test_ctx ctx; + + test_default_cfg(); + test_cfg.cli.auth = AUTH; + test_cfg.cli.md = NID_sha384; + + TEST_START(); + + if (oap_test_setup(&ctx, root_ca_crt_ec, im_ca_crt_ec) < 0) + goto fail; + + if (oap_cli_prepare_ctx(&ctx) < 0) { + printf("Client prepare failed.\n"); + goto fail_cleanup; + } + + if (oap_srv_process_ctx(&ctx) == 0) { + printf("Server should reject digest mismatch.\n"); + goto fail_cleanup; + } + + oap_test_teardown(&ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_cleanup: + oap_test_teardown(&ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* Naive substring search over raw bytes (memmem is not portable here). */ +static bool buf_contains(const uint8_t * hay, + size_t hlen, + const uint8_t * needle, + size_t nlen) +{ + size_t i; + + if (nlen == 0 || nlen > hlen) + return false; + + for (i = 0; i + nlen <= hlen; i++) { + if (memcmp(hay + i, needle, nlen) == 0) + return true; + } + + return false; +} + +/* The server certificate must not appear in cleartext on the wire */ +static int test_oap_server_cert_hidden(void) +{ + struct oap_test_ctx ctx; + void * crt = NULL; + buffer_t der = BUF_INIT; + + test_default_cfg(); + + TEST_START(); + + if (oap_test_setup(&ctx, root_ca_crt_ec, im_ca_crt_ec) < 0) + goto fail; + + if (oap_cli_prepare_ctx(&ctx) < 0) { + printf("Client prepare failed.\n"); + goto fail_cleanup; + } + + if (oap_srv_process_ctx(&ctx) < 0) { + printf("Server process failed.\n"); + goto fail_cleanup; + } + + if (crypt_load_crt_str(signed_server_crt_ec, &crt) < 0) { + printf("Failed to load server crt.\n"); + goto fail_cleanup; + } + + if (crypt_crt_der(crt, &der) < 0) { + printf("Failed to DER-encode server crt.\n"); + goto fail_crt; + } + + if (der.len == 0 || der.len > ctx.resp_hdr.len) { + printf("Unexpected cert/response sizes.\n"); + goto fail_der; + } + + if (buf_contains(ctx.resp_hdr.data, ctx.resp_hdr.len, + der.data, der.len)) { + printf("Server certificate found in cleartext.\n"); + goto fail_der; + } + + /* The handshake must still complete and agree on a key */ + if (oap_cli_complete_ctx(&ctx) < 0) { + printf("Client complete failed.\n"); + goto fail_der; + } + + if (memcmp(ctx.cli.key, ctx.srv.key, SYMMKEYSZ) != 0) { + printf("Client and server keys do not match!\n"); + goto fail_der; + } + + freebuf(der); + crypt_free_crt(crt); + oap_test_teardown(&ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_der: + freebuf(der); + fail_crt: + crypt_free_crt(crt); + fail_cleanup: + oap_test_teardown(&ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* Tampering the sealed identity block fails the handshake */ +static int test_oap_sealed_tamper(void) +{ + struct oap_test_ctx ctx; + size_t pos; + + test_default_cfg(); + + TEST_START(); + + if (oap_test_setup(&ctx, root_ca_crt_ec, im_ca_crt_ec) < 0) + goto fail; + + if (oap_cli_prepare_ctx(&ctx) < 0) { + printf("Client prepare failed.\n"); + goto fail_cleanup; + } + + if (oap_srv_process_ctx(&ctx) < 0) { + printf("Server process failed.\n"); + goto fail_cleanup; + } + + if (ctx.resp_hdr.len < 64) { + printf("Response too short for test.\n"); + goto fail_cleanup; + } + + /* Flip a byte inside the sealed ciphertext, before the AEAD tag */ + pos = ctx.resp_hdr.len - 32; + ctx.resp_hdr.data[pos] ^= 0xFF; + + if (oap_cli_complete_ctx(&ctx) == 0) { + printf("Client accepted a tampered identity block.\n"); + goto fail_cleanup; + } + + oap_test_teardown(&ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_cleanup: + oap_test_teardown(&ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +int oap_test(int argc, + char **argv) +{ + int ret = 0; + + (void) argc; + (void) argv; + + ret |= test_oap_auth_init_fini(); + ret |= test_oap_replay_cap(); + ret |= test_oap_replay_generations(); + +#ifdef HAVE_OPENSSL + ret |= test_oap_roundtrip_auth_only(); + ret |= test_oap_roundtrip_kex_only(); + ret |= test_oap_piggyback_data(); + ret |= test_oap_rekey(); + ret |= test_oap_rekey_badcache(); + + ret |= test_oap_roundtrip_all(); + ret |= test_oap_roundtrip_md_all(); + + ret |= test_oap_corrupted_request(); + ret |= test_oap_corrupted_response(); + ret |= test_oap_key_confirm_mismatch(); + ret |= test_oap_truncated_request(); + ret |= test_oap_inflated_length_field(); + ret |= test_oap_deflated_length_field(); + ret |= test_oap_nid_without_kex(); + ret |= test_oap_unsupported_nid(); + + ret |= test_oap_cipher_mismatch(); + ret |= test_oap_srv_enc_cli_none(); + ret |= test_oap_cli_enc_srv_none(); + ret |= test_oap_cli_rejects_downgrade(); + ret |= test_oap_cli_rejects_suite_swap(); + ret |= test_oap_srv_rejects_weak_kex(); + + ret |= test_oap_outdated_packet(); + ret |= test_oap_future_packet(); + ret |= test_oap_replay_packet(); + ret |= test_oap_missing_root_ca(); + ret |= test_oap_server_name_mismatch(); + + ret |= test_oap_cli_requires_srv_auth(); + ret |= test_oap_srv_requires_cli_auth(); + ret |= test_oap_mutual_req_auth(); + + + ret |= test_oap_cli_rejects_md_mismatch(); + ret |= test_oap_srv_rejects_md_mismatch(); + + ret |= test_oap_server_cert_hidden(); + ret |= test_oap_sealed_tamper(); +#else + (void) test_oap_roundtrip_auth_only; + (void) test_oap_roundtrip_kex_only; + (void) test_oap_piggyback_data; + (void) test_oap_roundtrip; + (void) test_oap_roundtrip_all; + (void) test_oap_roundtrip_md; + (void) test_oap_roundtrip_md_all; + (void) test_oap_corrupted_request; + (void) test_oap_corrupted_response; + (void) test_oap_key_confirm_mismatch; + (void) test_oap_truncated_request; + (void) test_oap_inflated_length_field; + (void) test_oap_deflated_length_field; + (void) test_oap_nid_without_kex; + (void) test_oap_unsupported_nid; + (void) test_oap_cipher_mismatch; + (void) test_oap_srv_enc_cli_none; + (void) test_oap_cli_enc_srv_none; + (void) test_oap_cli_rejects_downgrade; + (void) test_oap_cli_rejects_suite_swap; + (void) test_oap_srv_rejects_weak_kex; + (void) test_oap_outdated_packet; + (void) test_oap_future_packet; + (void) test_oap_replay_packet; + (void) test_oap_replay_generations; + (void) test_oap_missing_root_ca; + (void) test_oap_server_name_mismatch; + (void) test_oap_cli_requires_srv_auth; + (void) test_oap_srv_requires_cli_auth; + (void) test_oap_mutual_req_auth; + (void) test_oap_cli_rejects_md_mismatch; + (void) test_oap_srv_rejects_md_mismatch; + (void) test_oap_server_cert_hidden; + (void) test_oap_sealed_tamper; + (void) test_oap_rekey; + (void) test_oap_rekey_badcache; + + ret = TEST_RC_SKIP; +#endif + crypt_cleanup(); + + return ret; +} diff --git a/src/irmd/oap/tests/oap_test_ml_dsa.c b/src/irmd/oap/tests/oap_test_ml_dsa.c new file mode 100644 index 00000000..8691aa00 --- /dev/null +++ b/src/irmd/oap/tests/oap_test_ml_dsa.c @@ -0,0 +1,491 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Unit tests of OAP ML-KEM/ML-DSA key exchange + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#if defined(__linux__) || defined(__CYGWIN__) +#define _DEFAULT_SOURCE +#else +#define _POSIX_C_SOURCE 200809L +#endif + +#include "config.h" + +#include <ouroboros/crypt.h> +#include <ouroboros/flow.h> +#include <ouroboros/name.h> +#include <ouroboros/random.h> +#include <test/test.h> + +#include <test/certs/ml_dsa.h> + +#include "oap.h" +#include "common.h" + +#include <stdbool.h> +#include <string.h> + +#ifdef HAVE_OPENSSL +#include <openssl/evp.h> +#endif + +#define CLI_AUTH 1 +#define NO_CLI_AUTH 0 +#define CLI_ENCAP KEM_MODE_CLIENT_ENCAP +#define SRV_ENCAP KEM_MODE_SERVER_ENCAP + +extern const uint16_t kex_supported_nids[]; +extern const uint16_t md_supported_nids[]; + +static int get_random_kdf(void) +{ + static int idx = 0; + int count; + + if (md_supported_nids[0] == NID_undef) + return NID_undef; + + for (count = 0; md_supported_nids[count] != NID_undef; count++) + ; + + return md_supported_nids[(idx++) % count]; +} + +struct test_cfg test_cfg; + +/* KEM keypair storage for tests (server-side keypair for KEM modes) */ +static void * test_kem_pkp = NULL; /* Private key pair */ +static uint8_t test_kem_pk[4096]; /* Public key buffer */ +static size_t test_kem_pk_len = 0; + +/* Mock load - called by load_*_credentials in common.c */ +int mock_load_credentials(void ** pkp, + void ** crt) +{ + *pkp = NULL; + *crt = NULL; + + if (crypt_load_privkey_str(server_pkp_ml, pkp) < 0) + return -1; + + if (crypt_load_crt_str(signed_server_crt_ml, crt) < 0) { + crypt_free_key(*pkp); + *pkp = NULL; + return -1; + } + + return 0; +} + +int load_server_kem_keypair(const char * name, + bool raw_fmt, + void ** pkp) +{ +#ifdef HAVE_OPENSSL + struct sec_config local_cfg; + ssize_t pk_len; + + (void) name; + (void) raw_fmt; + + /* + * Uses reference counting. The caller will call + * EVP_PKEY_free which decrements the count. + */ + if (test_kem_pkp != NULL) { + if (EVP_PKEY_up_ref((EVP_PKEY *)test_kem_pkp) != 1) + return -1; + + *pkp = test_kem_pkp; + return 0; + } + + /* + * Generate a new KEM keypair from test_cfg.srv.kex. + */ + memset(&local_cfg, 0, sizeof(local_cfg)); + if (test_cfg.srv.kex == NID_undef) + goto fail; + + SET_KEX_ALGO_NID(&local_cfg, test_cfg.srv.kex); + + pk_len = kex_pkp_create(&local_cfg, &test_kem_pkp, test_kem_pk); + if (pk_len < 0) + goto fail; + + test_kem_pk_len = (size_t) pk_len; + + if (EVP_PKEY_up_ref((EVP_PKEY *)test_kem_pkp) != 1) + goto fail_ref; + + *pkp = test_kem_pkp; + + return 0; + fail_ref: + kex_pkp_destroy(test_kem_pkp); + test_kem_pkp = NULL; + test_kem_pk_len = 0; + fail: + return -1; + +#else + (void) name; + (void) raw_fmt; + (void) pkp; + return -1; +#endif +} + +int load_server_kem_pk(const char * name, + struct sec_config * cfg, + buffer_t * pk) +{ + ssize_t len; + + (void) name; + + if (test_kem_pk_len > 0) { + pk->data = malloc(test_kem_pk_len); + if (pk->data == NULL) + return -1; + memcpy(pk->data, test_kem_pk, test_kem_pk_len); + pk->len = test_kem_pk_len; + return 0; + } + + /* Generate keypair on demand if not already done */ + len = kex_pkp_create(cfg, &test_kem_pkp, test_kem_pk); + if (len < 0) + return -1; + + test_kem_pk_len = (size_t) len; + pk->data = malloc(test_kem_pk_len); + if (pk->data == NULL) + return -1; + + memcpy(pk->data, test_kem_pk, test_kem_pk_len); + pk->len = test_kem_pk_len; + + return 0; +} + +static void reset_kem_state(void) +{ + if (test_kem_pkp != NULL) { + kex_pkp_destroy(test_kem_pkp); + test_kem_pkp = NULL; + } + test_kem_pk_len = 0; +} + +static void test_cfg_init(int kex, + int cipher, + int kdf, + int kem_mode, + bool cli_auth) +{ + memset(&test_cfg, 0, sizeof(test_cfg)); + + /* Server config */ + test_cfg.srv.kex = kex; + test_cfg.srv.cipher = cipher; + test_cfg.srv.kdf = kdf; + test_cfg.srv.kem_mode = kem_mode; + test_cfg.srv.auth = true; + + /* Client config */ + test_cfg.cli.kex = kex; + test_cfg.cli.cipher = cipher; + test_cfg.cli.kdf = kdf; + test_cfg.cli.kem_mode = kem_mode; + test_cfg.cli.auth = cli_auth; +} + +static int oap_test_setup_kem(struct oap_test_ctx * ctx, + const char * root_ca, + const char * im_ca) +{ + reset_kem_state(); + return oap_test_setup(ctx, root_ca, im_ca); +} + +static void oap_test_teardown_kem(struct oap_test_ctx * ctx) +{ + oap_test_teardown(ctx); +} + +static int test_oap_roundtrip_auth_only(void) +{ + test_cfg_init(NID_undef, NID_undef, NID_undef, 0, false); + + return roundtrip_auth_only(root_ca_crt_ml, im_ca_crt_ml); +} + +/* Digest pin does not apply to PQC: the digest is intrinsic */ +static int test_oap_cli_md_pin_exempts_pqc(void) +{ + test_cfg_init(NID_undef, NID_undef, NID_undef, 0, NO_CLI_AUTH); + test_cfg.cli.md = NID_sha256; + + return roundtrip_auth_only(root_ca_crt_ml, im_ca_crt_ml); +} + +static int test_oap_srv_md_pin_exempts_pqc(void) +{ + test_cfg_init(NID_undef, NID_undef, NID_undef, 0, CLI_AUTH); + test_cfg.srv.md = NID_sha256; + + return roundtrip_auth_only(root_ca_crt_ml, im_ca_crt_ml); +} + +static int test_oap_rekey(void) +{ + test_cfg_init(NID_X25519, NID_aes_256_gcm, NID_sha256, + 0, NO_CLI_AUTH); + + return roundtrip_rekey(root_ca_crt_ml, im_ca_crt_ml); +} + +static int test_oap_rekey_badcache(void) +{ + test_cfg_init(NID_X25519, NID_aes_256_gcm, NID_sha256, + 0, NO_CLI_AUTH); + + return roundtrip_rekey_badcache(root_ca_crt_ml, im_ca_crt_ml); +} + +static int test_oap_corrupted_request(void) +{ + test_cfg_init(NID_MLKEM768, NID_aes_256_gcm, get_random_kdf(), + SRV_ENCAP, CLI_AUTH); + + return corrupted_request(root_ca_crt_ml, im_ca_crt_ml); +} + +static int test_oap_corrupted_response(void) +{ + test_cfg_init(NID_MLKEM768, NID_aes_256_gcm, get_random_kdf(), + SRV_ENCAP, NO_CLI_AUTH); + + return corrupted_response(root_ca_crt_ml, im_ca_crt_ml); +} + +static int test_oap_truncated_request(void) +{ + test_cfg_init(NID_MLKEM768, NID_aes_256_gcm, get_random_kdf(), + SRV_ENCAP, NO_CLI_AUTH); + + return truncated_request(root_ca_crt_ml, im_ca_crt_ml); +} + +static int test_oap_roundtrip_kem(int kex, + int kem_mode) +{ + struct oap_test_ctx ctx; + const char * kex_str = kex_nid_to_str(kex); + const char * mode_str = kem_mode == CLI_ENCAP ? "cli" : "srv"; + + test_cfg_init(kex, NID_aes_256_gcm, get_random_kdf(), + kem_mode, NO_CLI_AUTH); + + TEST_START("(%s, %s encaps)", kex_str, mode_str); + + if (oap_test_setup_kem(&ctx, root_ca_crt_ml, im_ca_crt_ml) < 0) + goto fail; + + if (oap_cli_prepare_ctx(&ctx) < 0) { + printf("Client prepare failed.\n"); + goto fail_cleanup; + } + + if (oap_srv_process_ctx(&ctx) < 0) { + printf("Server process failed.\n"); + goto fail_cleanup; + } + + if (oap_cli_complete_ctx(&ctx) < 0) { + printf("Client complete failed.\n"); + goto fail_cleanup; + } + + if (memcmp(ctx.cli.key, ctx.srv.key, SYMMKEYSZ) != 0) { + printf("Client and server keys do not match!\n"); + goto fail_cleanup; + } + + if (ctx.cli.nid == NID_undef || + ctx.srv.nid == NID_undef) { + printf("Cipher not set in flow.\n"); + goto fail_cleanup; + } + + oap_test_teardown_kem(&ctx); + + TEST_SUCCESS("(%s, %s encaps)", kex_str, mode_str); + return TEST_RC_SUCCESS; + + fail_cleanup: + oap_test_teardown_kem(&ctx); + fail: + TEST_FAIL("(%s, %s encaps)", kex_str, mode_str); + return TEST_RC_FAIL; +} + +static int test_oap_roundtrip_kem_all(void) +{ + int ret = 0; + int i; + + for (i = 0; kex_supported_nids[i] != NID_undef; i++) { + const char * algo = kex_nid_to_str(kex_supported_nids[i]); + + if (!IS_KEM_ALGORITHM(algo)) + continue; + + ret |= test_oap_roundtrip_kem(kex_supported_nids[i], SRV_ENCAP); + ret |= test_oap_roundtrip_kem(kex_supported_nids[i], CLI_ENCAP); + } + + return ret; +} + +static int test_oap_kem_srv_uncfg(int kex) +{ + struct oap_test_ctx ctx; + const char * kex_str = kex_nid_to_str(kex); + + memset(&test_cfg, 0, sizeof(test_cfg)); + + /* Server: auth only, no KEX configured */ + test_cfg.srv.auth = true; + + /* Client: requests KEM with server-side encapsulation */ + test_cfg.cli.kex = kex; + test_cfg.cli.cipher = NID_aes_256_gcm; + test_cfg.cli.kdf = get_random_kdf(); + test_cfg.cli.kem_mode = SRV_ENCAP; + test_cfg.cli.auth = false; + + TEST_START("(%s)", kex_str); + + if (oap_test_setup_kem(&ctx, root_ca_crt_ml, + im_ca_crt_ml) < 0) + goto fail; + + if (oap_cli_prepare_ctx(&ctx) < 0) { + printf("Client prepare failed.\n"); + goto fail_cleanup; + } + + if (oap_srv_process_ctx(&ctx) < 0) { + printf("Server process failed.\n"); + goto fail_cleanup; + } + + if (oap_cli_complete_ctx(&ctx) < 0) { + printf("Client complete failed.\n"); + goto fail_cleanup; + } + + if (memcmp(ctx.cli.key, ctx.srv.key, SYMMKEYSZ) != 0) { + printf("Client and server keys do not match!\n"); + goto fail_cleanup; + } + + if (ctx.cli.nid == NID_undef || + ctx.srv.nid == NID_undef) { + printf("Cipher not set in flow.\n"); + goto fail_cleanup; + } + + oap_test_teardown_kem(&ctx); + + TEST_SUCCESS("(%s)", kex_str); + return TEST_RC_SUCCESS; + + fail_cleanup: + oap_test_teardown_kem(&ctx); + fail: + TEST_FAIL("(%s)", kex_str); + return TEST_RC_FAIL; +} + +static int test_oap_kem_srv_uncfg_all(void) +{ + int ret = 0; + int i; + + for (i = 0; kex_supported_nids[i] != NID_undef; i++) { + const char * algo; + + algo = kex_nid_to_str(kex_supported_nids[i]); + + if (!IS_KEM_ALGORITHM(algo)) + continue; + + ret |= test_oap_kem_srv_uncfg(kex_supported_nids[i]); + } + + return ret; +} + +int oap_test_ml_dsa(int argc, + char **argv) +{ + int ret = 0; + + (void) argc; + (void) argv; + +#ifdef HAVE_OPENSSL_ML_KEM + ret |= test_oap_roundtrip_auth_only(); + ret |= test_oap_cli_md_pin_exempts_pqc(); + ret |= test_oap_srv_md_pin_exempts_pqc(); + + ret |= test_oap_roundtrip_kem_all(); + + ret |= test_oap_kem_srv_uncfg_all(); + + ret |= test_oap_corrupted_request(); + ret |= test_oap_corrupted_response(); + ret |= test_oap_truncated_request(); + + ret |= test_oap_rekey(); + ret |= test_oap_rekey_badcache(); +#else + (void) test_oap_roundtrip_auth_only; + (void) test_oap_cli_md_pin_exempts_pqc; + (void) test_oap_srv_md_pin_exempts_pqc; + (void) test_oap_rekey; + (void) test_oap_rekey_badcache; + (void) test_oap_roundtrip_kem; + (void) test_oap_roundtrip_kem_all; + (void) test_oap_kem_srv_uncfg; + (void) test_oap_kem_srv_uncfg_all; + (void) test_oap_corrupted_request; + (void) test_oap_corrupted_response; + (void) test_oap_truncated_request; + + ret = TEST_RC_SKIP; +#endif + crypt_cleanup(); + + return ret; +} diff --git a/src/irmd/reg/CMakeLists.txt b/src/irmd/reg/CMakeLists.txt deleted file mode 100644 index ff9d2e99..00000000 --- a/src/irmd/reg/CMakeLists.txt +++ /dev/null @@ -1,7 +0,0 @@ -include_directories(${CMAKE_CURRENT_SOURCE_DIR}) -include_directories(${CMAKE_CURRENT_BINARY_DIR}) - -include_directories(${CMAKE_SOURCE_DIR}/include) -include_directories(${CMAKE_BINARY_DIR}/include) - -add_subdirectory(tests) diff --git a/src/irmd/reg/flow.c b/src/irmd/reg/flow.c index 4d091b23..8be2dfc7 100644 --- a/src/irmd/reg/flow.c +++ b/src/irmd/reg/flow.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * The IPC Resource Manager - Registry - Flows * @@ -24,6 +24,7 @@ #define OUROBOROS_PREFIX "reg/flow" +#include <ouroboros/crypt.h> #include <ouroboros/logs.h> #include "flow.h" @@ -32,6 +33,7 @@ #include <errno.h> #include <stdbool.h> #include <stdlib.h> +#include <string.h> struct reg_flow * reg_flow_create(const struct flow_info * info) { @@ -42,6 +44,7 @@ struct reg_flow * reg_flow_create(const struct flow_info * info) assert(info->n_pid != 0); assert(info->n_1_pid == 0); assert(info->mpl == 0); + assert(info->mtu == 0); assert(info->state == FLOW_INIT); flow = malloc(sizeof(*flow)); @@ -66,11 +69,13 @@ struct reg_flow * reg_flow_create(const struct flow_info * info) static void destroy_rbuffs(struct reg_flow * flow) { if (flow->n_rb != NULL) - shm_rbuff_destroy(flow->n_rb); + ssm_rbuff_destroy(flow->n_rb); + flow->n_rb = NULL; if (flow->n_1_rb != NULL) - shm_rbuff_destroy(flow->n_1_rb); + ssm_rbuff_destroy(flow->n_1_rb); + flow->n_1_rb = NULL; } @@ -78,9 +83,14 @@ void reg_flow_destroy(struct reg_flow * flow) { assert(flow != NULL); + if (flow->rk.pending_seed != NULL) + crypt_secure_free(flow->rk.pending_seed, SYMMKEYSZ); + + freebuf(flow->rk.peer_crt); + switch(flow->info.state) { case FLOW_ACCEPT_PENDING: - clrbuf(flow->data); + clrbuf(flow->req_data); /* FALLTHRU */ default: destroy_rbuffs(flow); @@ -89,8 +99,10 @@ void reg_flow_destroy(struct reg_flow * flow) assert(flow->n_rb == NULL); assert(flow->n_1_rb == NULL); - assert(flow->data.data == NULL); - assert(flow->data.len == 0); + assert(flow->req_data.data == NULL); + assert(flow->req_data.len == 0); + assert(flow->rsp_data.data == NULL); + assert(flow->rsp_data.len == 0); assert(list_is_empty(&flow->next)); @@ -103,22 +115,28 @@ static int create_rbuffs(struct reg_flow * flow, assert(flow != NULL); assert(info != NULL); - flow->n_rb = shm_rbuff_create(info->n_pid, info->id); + flow->n_rb = ssm_rbuff_create(info->n_pid, info->id); if (flow->n_rb == NULL) goto fail_n_rb; + if (ssm_rbuff_mlock(flow->n_rb) < 0) + log_warn("Failed to mlock n_rb for flow %d.", info->id); + assert(flow->info.n_1_pid == 0); assert(flow->n_1_rb == NULL); flow->info.n_1_pid = info->n_1_pid; - flow->n_1_rb = shm_rbuff_create(info->n_1_pid, info->id); + flow->n_1_rb = ssm_rbuff_create(info->n_1_pid, info->id); if (flow->n_1_rb == NULL) goto fail_n_1_rb; + if (ssm_rbuff_mlock(flow->n_1_rb) < 0) + log_warn("Failed to mlock n_1_rb for flow %d.", info->id); + return 0; fail_n_1_rb: - shm_rbuff_destroy(flow->n_rb); + ssm_rbuff_destroy(flow->n_rb); fail_n_rb: return -ENOMEM; } @@ -152,6 +170,7 @@ int reg_flow_update(struct reg_flow * flow, assert(info->mpl != 0); flow->info.mpl = info->mpl; + flow->info.mtu = info->mtu; if (flow->info.state == FLOW_ALLOC_PENDING) break; @@ -172,6 +191,7 @@ int reg_flow_update(struct reg_flow * flow, } flow->info.state = info->state; + flow->info.uid = info->uid; *info = flow->info; @@ -179,30 +199,3 @@ int reg_flow_update(struct reg_flow * flow, fail: return -ENOMEM; } - -void reg_flow_set_data(struct reg_flow * flow, - const buffer_t * buf) -{ - assert(flow != NULL); - assert(buf != NULL); - assert(flow->data.data == NULL); - assert(flow->data.len == 0); - - flow->data = *buf; -} - -void reg_flow_get_data(struct reg_flow * flow, - buffer_t * buf) -{ - assert(flow != NULL); - assert(buf != NULL); - - *buf = flow->data; - - clrbuf(flow->data); -} - -void reg_flow_free_data(struct reg_flow * flow) -{ - freebuf(flow->data); -} diff --git a/src/irmd/reg/flow.h b/src/irmd/reg/flow.h index 75ada971..166bed61 100644 --- a/src/irmd/reg/flow.h +++ b/src/irmd/reg/flow.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * The IPC Resource Manager - Registry - Flows * @@ -25,24 +25,48 @@ #include <ouroboros/list.h> #include <ouroboros/flow.h> +#include <ouroboros/name.h> #include <ouroboros/pthread.h> #include <ouroboros/qos.h> -#include <ouroboros/shm_rbuff.h> +#include <ouroboros/ssm_rbuff.h> #include <ouroboros/utils.h> +#include <stdbool.h> #include <sys/types.h> #include <time.h> struct reg_flow { - struct list_head next; + struct list_head next; - struct flow_info info; + struct flow_info info; + int response; - buffer_t data; - struct timespec t0; + buffer_t req_data; + buffer_t rsp_data; + struct timespec t0; - struct shm_rbuff * n_rb; - struct shm_rbuff * n_1_rb; + char name[NAME_SIZE + 1]; + + bool direct; + + /* Tier-2 re-key state (encrypted flows only) */ + struct { + bool encrypted; /* flow carries a cipher */ + uint8_t epoch; /* last epoch installed by app */ + bool initiator; /* OAP initiator (role 0) */ + bool in_flight; /* a re-key is in progress */ + bool req_queued; /* a peer REQ is in the inbox */ + bool resp_queued; /* a peer RESP is in the inbox */ + uint8_t * pending_seed; /* secure heap; NULL until set */ + uint8_t pending_epoch; + bool pending_initiator; /* pending seed: oap_cli side */ + bool has_pending; /* new seed awaits app pull */ + uint8_t pulled; /* direct: per-app pull mask */ + buffer_t peer_crt; /* peer cert DER, cached at HS */ + } rk; + + struct ssm_rbuff * n_rb; + struct ssm_rbuff * n_1_rb; }; struct reg_flow * reg_flow_create(const struct flow_info * info); @@ -52,12 +76,4 @@ void reg_flow_destroy(struct reg_flow * flow); int reg_flow_update(struct reg_flow * flow, struct flow_info * info); -void reg_flow_set_data(struct reg_flow * flow, - const buffer_t * buf); - -void reg_flow_get_data(struct reg_flow * flow, - buffer_t * buf); - -void reg_flow_free_data(struct reg_flow * flow); - #endif /* OUROBOROS_IRMD_REG_FLOW_H */ diff --git a/src/irmd/reg/ipcp.c b/src/irmd/reg/ipcp.c index 6580cb5b..b193e28f 100644 --- a/src/irmd/reg/ipcp.c +++ b/src/irmd/reg/ipcp.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * The IPC Resource Manager - Registry - IPCPs * @@ -40,7 +40,7 @@ struct reg_ipcp * reg_ipcp_create(const struct ipcp_info * info) struct reg_ipcp * ipcp; assert(info != NULL); - assert(info->state == IPCP_BOOT); + assert(info->state == IPCP_INIT); ipcp = malloc(sizeof(*ipcp)); if (ipcp == NULL) { @@ -54,7 +54,7 @@ struct reg_ipcp * reg_ipcp_create(const struct ipcp_info * info) list_head_init(&ipcp->next); ipcp->info = *info; - ipcp->info.state = IPCP_BOOT; + ipcp->info.state = IPCP_INIT; strcpy(ipcp->layer.name, "Not enrolled."); @@ -77,7 +77,6 @@ void reg_ipcp_update(struct reg_ipcp * ipcp, const struct ipcp_info * info) { assert(ipcp != NULL); - assert(info->state != IPCP_INIT); ipcp->info = *info; } @@ -86,7 +85,7 @@ void reg_ipcp_set_layer(struct reg_ipcp * ipcp, const struct layer_info * info) { assert(ipcp != NULL); - assert(ipcp->info.state == IPCP_OPERATIONAL); + assert(ipcp->info.state == IPCP_BOOT); ipcp->layer = *info; } diff --git a/src/irmd/reg/ipcp.h b/src/irmd/reg/ipcp.h index 375973a7..fe9d3bf1 100644 --- a/src/irmd/reg/ipcp.h +++ b/src/irmd/reg/ipcp.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * The IPC Resource Manager - Registry - IPCPs * diff --git a/src/irmd/reg/name.c b/src/irmd/reg/name.c index 1ac939a5..61a328ec 100644 --- a/src/irmd/reg/name.c +++ b/src/irmd/reg/name.c @@ -1,6 +1,6 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * The IPC Resource Manager - Registry - Names * @@ -66,15 +66,14 @@ struct reg_name * reg_name_create(const struct name_info * info) goto fail_malloc; } + memset(name, 0, sizeof(*name)); + list_head_init(&name->next); - list_head_init(&name->progs); - list_head_init(&name->procs); - list_head_init(&name->active); + llist_init(&name->progs); + llist_init(&name->procs); + llist_init(&name->active); - name->info = *info; - name->n_progs = 0; - name->n_procs = 0; - name->n_active = 0; + name->info = *info; return name; @@ -88,13 +87,9 @@ void reg_name_destroy(struct reg_name * name) assert(list_is_empty(&name->next)); - assert(name->n_progs == 0); - assert(name->n_procs == 0); - assert(name->n_active == 0); - - assert(list_is_empty(&name->progs)); - assert(list_is_empty(&name->procs)); - assert(list_is_empty(&name->active)); + assert(llist_is_empty(&name->progs)); + assert(llist_is_empty(&name->procs)); + assert(llist_is_empty(&name->active)); free(name); } @@ -107,7 +102,7 @@ static struct proc_entry * __reg_name_get_active(const struct reg_name * name, assert(name != NULL); assert(pid > 0); - list_for_each(p, &name->active) { + llist_for_each(p, &name->active) { struct proc_entry * entry; entry = list_entry(p, struct proc_entry, next); if (entry->pid == pid) @@ -123,13 +118,12 @@ static void __reg_name_del_all_active(struct reg_name * name, struct list_head * p; struct list_head * h; - list_for_each_safe(p, h, &name->active) { + llist_for_each_safe(p, h, &name->active) { struct proc_entry * entry; entry = list_entry(p, struct proc_entry, next); if (entry->pid == pid) { - list_del(&entry->next); + llist_del(&entry->next, &name->active); free(entry); - name->n_active--; } } } @@ -142,7 +136,7 @@ static struct proc_entry * __reg_name_get_proc(const struct reg_name * name, assert(name != NULL); assert(pid > 0); - list_for_each(p, &name->procs) { + llist_for_each(p, &name->procs) { struct proc_entry * entry; entry = list_entry(p, struct proc_entry, next); if (entry->pid == pid) @@ -160,7 +154,7 @@ static struct prog_entry * __reg_name_get_prog(const struct reg_name * name, assert(name != NULL); assert(prog != NULL); - list_for_each(p, &name->progs) { + llist_for_each(p, &name->progs) { struct prog_entry * entry; entry = list_entry(p, struct prog_entry, next); if (strcmp(entry->exec[0], prog) == 0) @@ -195,17 +189,15 @@ int reg_name_add_active(struct reg_name * name, switch (name->info.pol_lb) { case LB_RR: /* Round robin policy. */ - list_add_tail(&entry->next, &name->active); + llist_add_tail(&entry->next, &name->active); break; case LB_SPILL: /* Keep accepting flows on the current process */ - list_add(&entry->next, &name->active); + llist_add(&entry->next, &name->active); break; default: goto fail_unreachable; } - name->n_active++; - return 0; fail_unreachable: @@ -224,21 +216,23 @@ void reg_name_del_active(struct reg_name * name, if (entry == NULL) return; - list_del(&entry->next); - - name->n_active--; + llist_del(&entry->next, &name->active); free(entry); } pid_t reg_name_get_active(struct reg_name * name) { + struct proc_entry * e; + assert(name != NULL); - if (list_is_empty(&name->active)) + if (llist_is_empty(&name->active)) return -1; - return list_first_entry(&name->active, struct proc_entry, next)->pid; + e = llist_first_entry(&name->active, struct proc_entry, next); + + return e->pid; } int reg_name_add_proc(struct reg_name * name, @@ -259,9 +253,7 @@ int reg_name_add_proc(struct reg_name * name, entry->pid = pid; - list_add(&entry->next, &name->procs); - - name->n_procs++; + llist_add(&entry->next, &name->procs); return 0; @@ -283,12 +275,10 @@ void reg_name_del_proc(struct reg_name * name, __reg_name_del_all_active(name, pid); - list_del(&entry->next); + llist_del(&entry->next, &name->procs); free(entry); - name->n_procs--; - assert(__reg_name_get_proc(name, pid) == NULL); } @@ -296,8 +286,7 @@ bool reg_name_has_proc(const struct reg_name * name, pid_t pid) { return __reg_name_get_proc(name, pid) != NULL; -} char ** exec; - +} int reg_name_add_prog(struct reg_name * name, char ** exec) @@ -322,12 +311,10 @@ int reg_name_add_prog(struct reg_name * name, goto fail_exec; } - list_add(&entry->next, &name->progs); + llist_add(&entry->next, &name->progs); log_dbg("Add prog %s to name %s.", exec[0], name->info.name); - name->n_progs++; - return 0; fail_exec: @@ -348,12 +335,10 @@ void reg_name_del_prog(struct reg_name * name, if (entry == NULL) return; - list_del(&entry->next); + llist_del(&entry->next, &name->progs); __free_prog_entry(entry); - name->n_progs--; - assert(__reg_name_get_prog(name, prog) == NULL); } @@ -368,8 +353,12 @@ bool reg_name_has_prog(const struct reg_name * name, char ** reg_name_get_exec(const struct reg_name * name) { - if (list_is_empty(&name->progs)) + struct prog_entry * e; + + if (llist_is_empty(&name->progs)) return NULL; - return list_first_entry(&name->progs, struct prog_entry, next)->exec; + e = llist_first_entry(&name->progs, struct prog_entry, next); + + return e->exec; } diff --git a/src/irmd/reg/name.h b/src/irmd/reg/name.h index 97ca7f04..59d6d9bd 100644 --- a/src/irmd/reg/name.h +++ b/src/irmd/reg/name.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * The IPC Resource Manager - Registry - Names * @@ -33,14 +33,16 @@ struct reg_name { struct name_info info; - struct list_head progs; /* autostart programs for this name */ - size_t n_progs; /* number of programs */ + struct { + void * key; + void * crt; + } cache; - struct list_head procs; /* processes bound to this name */ - size_t n_procs; /* number of processes */ + struct llist progs; /* autostart programs for this name */ - struct list_head active; /* processes actively calling accept */ - size_t n_active; /* number of processes accepting */ + struct llist procs; /* processes bound to this name */ + + struct llist active; /* processes actively calling accept */ }; struct reg_name * reg_name_create(const struct name_info * info); @@ -74,5 +76,4 @@ pid_t reg_name_get_active(struct reg_name * name); void reg_name_del_active(struct reg_name * name, pid_t proc); - #endif /* OUROBOROS_IRMD_REG_NAME_H */ diff --git a/src/irmd/reg/pool.c b/src/irmd/reg/pool.c new file mode 100644 index 00000000..4b1486bb --- /dev/null +++ b/src/irmd/reg/pool.c @@ -0,0 +1,97 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * The IPC Resource Manager - Registry - Per-User Pools + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#define _POSIX_C_SOURCE 200809L + +#define OUROBOROS_PREFIX "reg/pool" + +#include <ouroboros/logs.h> +#include <ouroboros/ssm_pool.h> + +#include "pool.h" + +#include <assert.h> +#include <stdlib.h> + +struct reg_pool * reg_pool_create(uid_t uid, + gid_t gid) +{ + struct reg_pool * pool; + + pool = malloc(sizeof(*pool)); + if (pool == NULL) { + log_err("Failed to malloc pool."); + goto fail_malloc; + } + + pool->ssm = ssm_pool_create(uid, gid); + if (pool->ssm == NULL) { + log_err("Failed to create PUP for uid %d.", uid); + goto fail_ssm; + } + + list_head_init(&pool->next); + pool->uid = uid; + pool->gid = gid; + pool->refcount = 1; + + log_dbg("Created PUP for uid %d gid %d.", uid, gid); + + return pool; + + fail_ssm: + free(pool); + fail_malloc: + return NULL; +} + +void reg_pool_destroy(struct reg_pool * pool) +{ + assert(pool != NULL); + assert(pool->refcount == 0); + + log_dbg("Destroying PUP for uid %d.", pool->uid); + + ssm_pool_destroy(pool->ssm); + + assert(list_is_empty(&pool->next)); + + free(pool); +} + +void reg_pool_ref(struct reg_pool * pool) +{ + assert(pool != NULL); + assert(pool->refcount > 0); + + pool->refcount++; +} + +int reg_pool_unref(struct reg_pool * pool) +{ + assert(pool != NULL); + assert(pool->refcount > 0); + + pool->refcount--; + + return pool->refcount == 0 ? 0 : 1; +} diff --git a/src/irmd/reg/pool.h b/src/irmd/reg/pool.h new file mode 100644 index 00000000..576f491c --- /dev/null +++ b/src/irmd/reg/pool.h @@ -0,0 +1,48 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * The IPC Resource Manager - Registry - Per-User Pools + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#ifndef OUROBOROS_IRMD_REG_POOL_H +#define OUROBOROS_IRMD_REG_POOL_H + +#include <ouroboros/list.h> +#include <ouroboros/ssm_pool.h> + +#include <sys/types.h> + +struct reg_pool { + struct list_head next; + uid_t uid; + gid_t gid; + size_t refcount; + struct ssm_pool * ssm; +}; + +struct reg_pool * reg_pool_create(uid_t uid, + gid_t gid); + +void reg_pool_destroy(struct reg_pool * pool); + +void reg_pool_ref(struct reg_pool * pool); + +int reg_pool_unref(struct reg_pool * pool); + +#endif /* OUROBOROS_IRMD_REG_POOL_H */ diff --git a/src/irmd/reg/proc.c b/src/irmd/reg/proc.c index 9bbdf0eb..8a7e24c9 100644 --- a/src/irmd/reg/proc.c +++ b/src/irmd/reg/proc.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * The IPC Resource Manager - Registry - Processes * @@ -25,6 +25,7 @@ #define OUROBOROS_PREFIX "reg/proc" #include <ouroboros/logs.h> +#include <ouroboros/utils.h> #include "proc.h" @@ -54,12 +55,11 @@ static void __reg_proc_clear_names(struct reg_proc * proc) assert(proc != NULL); - list_for_each_safe(p, h, &proc->names) { + llist_for_each_safe(p, h, &proc->names) { struct name_entry * entry; entry = list_entry(p, struct name_entry, next); - list_del(&entry->next); + llist_del(&entry->next, &proc->names); __free_name_entry(entry); - proc->n_names--; } } @@ -75,17 +75,18 @@ struct reg_proc * reg_proc_create(const struct proc_info * info) goto fail_malloc; } - proc->set = shm_flow_set_create(info->pid); + memset(proc, 0, sizeof(*proc)); + + proc->set = ssm_flow_set_create(info->pid); if (proc->set == NULL) { log_err("Failed to create flow set for %d.", info->pid); goto fail_set; } list_head_init(&proc->next); - list_head_init(&proc->names); + llist_init(&proc->names); proc->info = *info; - proc->n_names = 0; return proc; @@ -99,15 +100,13 @@ void reg_proc_destroy(struct reg_proc * proc) { assert(proc != NULL); - shm_flow_set_destroy(proc->set); + ssm_flow_set_destroy(proc->set); __reg_proc_clear_names(proc); assert(list_is_empty(&proc->next)); - assert(proc->n_names == 0); - - assert(list_is_empty(&proc->names)); + assert(llist_is_empty(&proc->names)); free(proc); } @@ -117,7 +116,7 @@ static struct name_entry * __reg_proc_get_name(const struct reg_proc * proc, { struct list_head * p; - list_for_each(p, &proc->names) { + llist_for_each(p, &proc->names) { struct name_entry * entry; entry = list_entry(p, struct name_entry, next); if (strcmp(entry->name, name) == 0) @@ -146,9 +145,7 @@ int reg_proc_add_name(struct reg_proc * proc, goto fail_name; } - list_add(&entry->next, &proc->names); - - proc->n_names++; + llist_add(&entry->next, &proc->names); return 0; @@ -167,12 +164,10 @@ void reg_proc_del_name(struct reg_proc * proc, if(entry == NULL) return; - list_del(&entry->next); + llist_del(&entry->next, &proc->names); __free_name_entry(entry); - proc->n_names--; - assert(__reg_proc_get_name(proc, name) == NULL); } @@ -181,3 +176,10 @@ bool reg_proc_has_name(const struct reg_proc * proc, { return __reg_proc_get_name(proc, name) != NULL; } + +bool reg_proc_is_privileged(const struct reg_proc * proc) +{ + assert(proc != NULL); + + return is_ouroboros_member_uid(proc->info.uid); +} diff --git a/src/irmd/reg/proc.h b/src/irmd/reg/proc.h index 99f74fef..18cc2803 100644 --- a/src/irmd/reg/proc.h +++ b/src/irmd/reg/proc.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * The IPC Resource Manager - Registry - Processes * @@ -25,17 +25,16 @@ #include <ouroboros/list.h> #include <ouroboros/proc.h> -#include <ouroboros/shm_flow_set.h> +#include <ouroboros/ssm_flow_set.h> struct reg_proc { struct list_head next; struct proc_info info; - struct list_head names; /* names for which process accepts flows */ - size_t n_names; /* number of names */ + struct llist names; /* process accepts flows for names */ - struct shm_flow_set * set; + struct ssm_flow_set * set; }; struct reg_proc * reg_proc_create(const struct proc_info * info); @@ -53,4 +52,6 @@ void reg_proc_del_name(struct reg_proc * proc, bool reg_proc_has_name(const struct reg_proc * proc, const char * name); +bool reg_proc_is_privileged(const struct reg_proc * proc); + #endif /* OUROBOROS_IRMD_REG_PROC_H */ diff --git a/src/irmd/reg/prog.c b/src/irmd/reg/prog.c index 9b9e7510..2d7f9f8d 100644 --- a/src/irmd/reg/prog.c +++ b/src/irmd/reg/prog.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * The IPC Resource Manager - Registry - Programs * @@ -55,12 +55,11 @@ static void __reg_prog_clear_names(struct reg_prog * prog) assert(prog != NULL); - list_for_each_safe(p, h, &prog->names) { + llist_for_each_safe(p, h, &prog->names) { struct name_entry * entry; entry = list_entry(p, struct name_entry, next); - list_del(&entry->next); + llist_del(&entry->next, &prog->names); __free_name_entry(entry); - prog->n_names--; } } @@ -77,10 +76,9 @@ struct reg_prog * reg_prog_create(const struct prog_info * info) } list_head_init(&p->next); - list_head_init(&p->names); + llist_init(&p->names); p->info = *info; - p->n_names = 0; return p; @@ -96,9 +94,7 @@ void reg_prog_destroy(struct reg_prog * prog) assert(list_is_empty(&prog->next)); - assert(prog->n_names == 0); - - assert(list_is_empty(&prog->names)); + assert(llist_is_empty(&prog->names)); free(prog); } @@ -108,7 +104,7 @@ static struct name_entry * __reg_prog_get_name(const struct reg_prog * prog, { struct list_head * p; - list_for_each(p, &prog->names) { + llist_for_each(p, &prog->names) { struct name_entry * entry; entry = list_entry(p, struct name_entry, next); if (strcmp(entry->name, name) == 0) @@ -137,9 +133,7 @@ int reg_prog_add_name(struct reg_prog * prog, goto fail_name; } - list_add(&entry->next, &prog->names); - - prog->n_names++; + llist_add(&entry->next, &prog->names); return 0; @@ -158,12 +152,10 @@ void reg_prog_del_name(struct reg_prog * prog, if (entry == NULL) return; - list_del(&entry->next); + llist_del(&entry->next, &prog->names); __free_name_entry(entry); - prog->n_names--; - assert(__reg_prog_get_name(prog, name) == NULL); } diff --git a/src/irmd/reg/prog.h b/src/irmd/reg/prog.h index a98fc6a1..e52b8e15 100644 --- a/src/irmd/reg/prog.h +++ b/src/irmd/reg/prog.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * The IPC Resource Manager - Registry - Programs * @@ -33,8 +33,7 @@ struct reg_prog { struct prog_info info; - struct list_head names; /* names to listen for */ - size_t n_names; /* number of names in list */ + struct llist names; /* names to listen for */ }; struct reg_prog * reg_prog_create(const struct prog_info * info); diff --git a/src/irmd/reg/reg.c b/src/irmd/reg/reg.c index d95a4722..ebf3959d 100644 --- a/src/irmd/reg/reg.c +++ b/src/irmd/reg/reg.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * The IPC Resource Manager - Registry * @@ -25,15 +25,18 @@ The IPC Resource Manager - Registry #define OUROBOROS_PREFIX "reg" #include <ouroboros/bitmap.h> +#include <ouroboros/crypt.h> #include <ouroboros/errno.h> #include <ouroboros/list.h> #include <ouroboros/logs.h> +#include <ouroboros/protobuf.h> #include <ouroboros/pthread.h> #include "reg.h" #include "flow.h" #include "ipcp.h" #include "name.h" +#include "pool.h" #include "proc.h" #include "prog.h" @@ -45,27 +48,18 @@ The IPC Resource Manager - Registry #define ID_OFFT 1 /* reserve some flow_ids */ struct { - struct bmp * flow_ids; /* flow_ids for flows */ - struct list_head flows; /* flow information */ - size_t n_flows; /* number of flows */ - - struct list_head ipcps; /* list of ipcps in system */ - size_t n_ipcps; /* number of ipcps */ - - struct list_head names; /* registered names known */ - size_t n_names; /* number of names */ - - struct list_head procs; /* processes */ - size_t n_procs; /* number of processes */ - - struct list_head progs; /* programs known */ - size_t n_progs; /* number of programs */ - - struct list_head spawned; /* child processes */ - size_t n_spawned; /* number of child processes */ - - pthread_mutex_t mtx; /* registry lock */ - pthread_cond_t cond; /* condvar for reg changes */ + struct bmp * ids; /* flow bitmap */ + + struct llist flows; /* list of flows */ + struct llist ipcps; /* list of ipcps in system */ + struct llist names; /* registered names known */ + struct llist pools; /* per-user pools */ + struct llist procs; /* processes known */ + struct llist progs; /* programs known */ + struct llist spawned; /* child processes */ + + pthread_mutex_t mtx; /* registry lock */ + pthread_cond_t cond; /* condvar for reg changes */ } reg; struct pid_entry { @@ -79,7 +73,7 @@ static struct reg_flow * __reg_get_flow(int flow_id) assert(flow_id >= ID_OFFT); - list_for_each(p, ®.flows) { + llist_for_each(p, ®.flows) { struct reg_flow * entry; entry = list_entry(p, struct reg_flow, next); if (entry->info.id == flow_id) @@ -93,7 +87,7 @@ static struct reg_flow * __reg_get_accept_flow(pid_t pid) { struct list_head * p; - list_for_each(p, ®.flows) { + llist_for_each(p, ®.flows) { struct reg_flow * entry; entry = list_entry(p, struct reg_flow, next); if (entry->info.state != FLOW_ACCEPT_PENDING) @@ -111,7 +105,7 @@ static struct list_head * __reg_after_flow(int flow_id) assert(flow_id >= ID_OFFT); - list_for_each(p, ®.flows) { + llist_for_each(p, ®.flows) { struct reg_flow * entry; entry = list_entry(p, struct reg_flow, next); if (entry->info.id > flow_id) @@ -127,7 +121,7 @@ static struct reg_ipcp * __reg_get_ipcp(pid_t pid) assert(pid > 0); - list_for_each(p, ®.ipcps) { + llist_for_each(p, ®.ipcps) { struct reg_ipcp * entry; entry = list_entry(p, struct reg_ipcp, next); if (entry->info.pid == pid) @@ -141,7 +135,7 @@ static struct reg_ipcp * __reg_get_ipcp_by_layer(const char * layer) { struct list_head * p; - list_for_each(p, ®.ipcps) { + llist_for_each(p, ®.ipcps) { struct reg_ipcp * entry; entry = list_entry(p, struct reg_ipcp, next); if (strcmp(entry->layer.name, layer) == 0) @@ -151,16 +145,23 @@ static struct reg_ipcp * __reg_get_ipcp_by_layer(const char * layer) return NULL; } -static struct list_head * __reg_after_ipcp(pid_t pid) + +static struct list_head * __reg_after_ipcp(const struct ipcp_info * info) { struct list_head * p; - assert(pid > 0); + assert(info != NULL); - list_for_each(p, ®.ipcps) { + llist_for_each(p, ®.ipcps) { struct reg_ipcp * entry; entry = list_entry(p, struct reg_ipcp, next); - if (entry->info.pid > pid) + if (entry->info.type < info->type) + continue; + + if (entry->info.type > info->type) + break; + + if (entry->info.pid > info->pid) break; } @@ -173,7 +174,7 @@ static struct reg_name * __reg_get_name(const char * name) assert(name != NULL); - list_for_each(p, ®.names) { + llist_for_each(p, ®.names) { struct reg_name * entry; entry = list_entry(p, struct reg_name, next); if (strcmp(entry->info.name, name) == 0) @@ -183,41 +184,17 @@ static struct reg_name * __reg_get_name(const char * name) return NULL; } -static struct reg_name * __reg_get_name_by_hash(enum hash_algo algo, - const uint8_t * hash) -{ - struct list_head * p; - uint8_t * thash; - size_t len; - - len = hash_len(algo); - - thash = malloc(len); - if (thash == NULL) - return NULL; - - list_for_each(p, ®.names) { - struct reg_name * n = list_entry(p, struct reg_name, next); - str_hash(algo, thash, n->info.name); - if (memcmp(thash, hash, len) == 0) { - free(thash); - return n; - } - } - - free(thash); - - return NULL; -} - -static int __reg_get_pending_flow_id_for_hash(enum hash_algo algo, - const uint8_t * hash) +static int __reg_get_pending_flow_id(const char * name) { struct reg_name * entry; struct reg_flow * flow; pid_t pid; - entry =__reg_get_name_by_hash(algo, hash); + assert(name != NULL); + assert(strlen(name) > 0); + assert(strlen(name) < NAME_SIZE + 1); + + entry =__reg_get_name(name); if (entry == NULL) return -ENAME; @@ -226,7 +203,10 @@ static int __reg_get_pending_flow_id_for_hash(enum hash_algo algo, return -EAGAIN; flow = __reg_get_accept_flow(pid); - assert(flow != NULL); + if (flow == NULL) /* compiler barks, this can't be NULL */ + return -EAGAIN; + + strcpy(flow->name, name); return flow->info.id; } @@ -237,7 +217,7 @@ static struct list_head * __reg_after_name(const char * name) assert(name != NULL); - list_for_each(p, ®.names) { + llist_for_each(p, ®.names) { struct reg_name * entry; entry = list_entry(p, struct reg_name, next); if (strcmp(entry->info.name, name) > 0) @@ -247,11 +227,25 @@ static struct list_head * __reg_after_name(const char * name) return p; } +static struct reg_pool * __reg_get_pool(uid_t uid) +{ + struct list_head * p; + + llist_for_each(p, ®.pools) { + struct reg_pool * entry; + entry = list_entry(p, struct reg_pool, next); + if (entry->uid == uid) + return entry; + } + + return NULL; +} + static struct reg_proc * __reg_get_proc(pid_t pid) { struct list_head * p; - list_for_each(p, ®.procs) { + llist_for_each(p, ®.procs) { struct reg_proc * entry; entry = list_entry(p, struct reg_proc, next); if (entry->info.pid == pid) @@ -265,7 +259,7 @@ static struct list_head * __reg_after_proc(pid_t pid) { struct list_head * p; - list_for_each(p, ®.procs) { + llist_for_each(p, ®.procs) { struct reg_proc * entry; entry = list_entry(p, struct reg_proc, next); if (entry->info.pid > pid) @@ -279,7 +273,7 @@ static void __reg_kill_all_proc(int signal) { struct list_head * p; - list_for_each(p, ®.procs) { + llist_for_each(p, ®.procs) { struct reg_proc * entry; entry = list_entry(p, struct reg_proc, next); kill(entry->info.pid, signal); @@ -290,7 +284,7 @@ static pid_t __reg_get_dead_proc(void) { struct list_head * p; - list_for_each(p, ®.procs) { + llist_for_each(p, ®.procs) { struct reg_proc * entry; entry = list_entry(p, struct reg_proc, next); if (kill(entry->info.pid, 0) < 0) @@ -305,7 +299,7 @@ static void __reg_cancel_flows_for_proc(pid_t pid) struct list_head * p; bool changed = false; - list_for_each(p, ®.flows) { + llist_for_each(p, ®.flows) { struct reg_flow * entry; entry = list_entry(p, struct reg_flow, next); if (entry->info.n_pid != pid) @@ -331,7 +325,7 @@ static struct pid_entry * __reg_get_spawned(pid_t pid) { struct list_head * p; - list_for_each(p, ®.spawned) { + llist_for_each(p, ®.spawned) { struct pid_entry * entry; entry = list_entry(p, struct pid_entry, next); if (entry->pid == pid) @@ -345,7 +339,7 @@ static struct list_head * __reg_after_spawned(pid_t pid) { struct list_head * p; - list_for_each(p, ®.spawned) { + llist_for_each(p, ®.spawned) { struct pid_entry * entry; entry = list_entry(p, struct pid_entry, next); if (entry->pid > pid) @@ -359,7 +353,7 @@ static void __reg_kill_all_spawned(int signal) { struct list_head * p; - list_for_each(p, ®.spawned) { + llist_for_each(p, ®.spawned) { struct pid_entry * entry; entry = list_entry(p, struct pid_entry, next); kill(entry->pid, signal); @@ -368,17 +362,17 @@ static void __reg_kill_all_spawned(int signal) static pid_t __reg_first_spawned(void) { - if (list_is_empty(®.spawned)) + if (llist_is_empty(®.spawned)) return -1; - return list_first_entry(®.spawned, struct pid_entry, next)->pid; + return llist_first_entry(®.spawned, struct pid_entry, next)->pid; } static struct reg_prog * __reg_get_prog(const char * name) { struct list_head * p; - list_for_each(p, ®.progs) { + llist_for_each(p, ®.progs) { struct reg_prog * entry; entry = list_entry(p, struct reg_prog, next); if (strcmp(entry->info.name, name) == 0) @@ -388,30 +382,17 @@ static struct reg_prog * __reg_get_prog(const char * name) return NULL; } -static char ** __reg_get_exec(enum hash_algo algo, - const uint8_t * hash) +static char ** __reg_get_exec(const char * name) { struct list_head * p; - uint8_t * buf; - - buf = malloc(hash_len(algo)); - if (buf == NULL) { - log_err("Failed to malloc hash buffer."); - return NULL; - } - list_for_each(p, ®.names) { + llist_for_each(p, ®.names) { struct reg_name * entry; entry = list_entry(p, struct reg_name, next); - str_hash(algo, buf, entry->info.name); - if (memcmp(buf, hash, hash_len(algo)) == 0) { - free(buf); + if (strcmp(entry->info.name, name) == 0) return reg_name_get_exec(entry); - } } - free(buf); - return NULL; } @@ -419,7 +400,7 @@ static struct list_head * __reg_after_prog(const char * name) { struct list_head * p; - list_for_each(p, ®.progs) { + llist_for_each(p, ®.progs) { struct reg_prog * entry; entry = list_entry(p, struct reg_prog, next); if (strcmp(entry->info.name, name) > 0) @@ -433,7 +414,7 @@ static void __reg_del_name_from_procs(const char * name) { struct list_head * p; - list_for_each(p, ®.procs) { + llist_for_each(p, ®.procs) { struct reg_proc * proc; proc = list_entry(p, struct reg_proc, next); reg_proc_del_name(proc, name); @@ -444,7 +425,7 @@ static void __reg_del_name_from_progs(const char * name) { struct list_head * p; - list_for_each(p, ®.progs) { + llist_for_each(p, ®.progs) { struct reg_prog * prog; prog = list_entry(p, struct reg_prog, next); reg_prog_del_name(prog, name); @@ -456,13 +437,13 @@ static void __reg_proc_update_names(struct reg_proc * proc) struct list_head * p; struct reg_prog * prog; - assert(list_is_empty(&proc->names)); + assert(llist_is_empty(&proc->names)); prog = __reg_get_prog(proc->info.prog); if (prog == NULL) return; - list_for_each(p, ®.names) { + llist_for_each(p, ®.names) { struct reg_name * name; name = list_entry(p, struct reg_name, next); assert(!reg_name_has_proc(name, proc->info.pid)); @@ -477,7 +458,7 @@ static void __reg_del_proc_from_names(pid_t pid) { struct list_head * p; - list_for_each(p, ®.names) { + llist_for_each(p, ®.names) { struct reg_name * name; name = list_entry(p, struct reg_name, next); reg_name_del_proc(name, pid); @@ -488,7 +469,7 @@ static void __reg_del_prog_from_names(const char * prog) { struct list_head * p; - list_for_each(p, ®.names) { + llist_for_each(p, ®.names) { struct reg_name * name; name = list_entry(p, struct reg_name, next); reg_name_del_prog(name, prog); @@ -503,7 +484,7 @@ static int __reg_add_active_proc(pid_t pid) assert(pid > 0); - list_for_each(p, ®.names) { + llist_for_each(p, ®.names) { struct reg_name * name; name = list_entry(p, struct reg_name, next); if (reg_name_has_proc(name, pid)) { @@ -525,7 +506,7 @@ static void __reg_del_active_proc(pid_t pid) assert(pid > 0); - list_for_each(p, ®.names) { + llist_for_each(p, ®.names) { struct reg_name * name; name = list_entry(p, struct reg_name, next); reg_name_del_active(name, pid); @@ -554,20 +535,21 @@ int reg_init(void) goto fail_cond; } - reg.flow_ids = bmp_create(SYS_MAX_FLOWS -ID_OFFT, ID_OFFT); - if (reg.flow_ids == NULL) { + reg.ids = bmp_create(SYS_MAX_FLOWS - ID_OFFT, ID_OFFT); + if (reg.ids == NULL) { log_err("Failed to create flow_ids bitmap."); goto fail_flow_ids; } pthread_condattr_destroy(&cattr); - list_head_init(®.flows); - list_head_init(®.ipcps); - list_head_init(®.names); - list_head_init(®.procs); - list_head_init(®.progs); - list_head_init(®.spawned); + llist_init(®.flows); + llist_init(®.ipcps); + llist_init(®.names); + llist_init(®.pools); + llist_init(®.procs); + llist_init(®.progs); + llist_init(®.spawned); return 0; @@ -588,54 +570,56 @@ void reg_clear(void) pthread_mutex_lock(®.mtx); - list_for_each_safe(p, h, ®.spawned) { + llist_for_each_safe(p, h, ®.spawned) { struct pid_entry * entry; entry = list_entry(p, struct pid_entry, next); - list_del(&entry->next); + llist_del(&entry->next, ®.spawned); free(entry); - reg.n_spawned--; } - list_for_each_safe(p, h, ®.progs) { + llist_for_each_safe(p, h, ®.progs) { struct reg_prog * entry; entry = list_entry(p, struct reg_prog, next); - list_del(&entry->next); + llist_del(&entry->next, ®.progs); __reg_del_prog_from_names(entry->info.path); reg_prog_destroy(entry); - reg.n_progs--; } - list_for_each_safe(p, h, ®.procs) { + llist_for_each_safe(p, h, ®.procs) { struct reg_proc * entry; entry = list_entry(p, struct reg_proc, next); - list_del(&entry->next); + llist_del(&entry->next, ®.procs); __reg_del_proc_from_names(entry->info.pid); reg_proc_destroy(entry); - reg.n_procs--; } - list_for_each_safe(p, h, ®.names) { + llist_for_each_safe(p, h, ®.pools) { + struct reg_pool * entry; + entry = list_entry(p, struct reg_pool, next); + llist_del(&entry->next, ®.pools); + entry->refcount = 0; /* Force destroy during cleanup */ + reg_pool_destroy(entry); + } + + llist_for_each_safe(p, h, ®.flows) { + struct reg_flow * entry; + entry = list_entry(p, struct reg_flow, next); + llist_del(&entry->next, ®.flows); + reg_flow_destroy(entry); + } + + llist_for_each_safe(p, h, ®.names) { struct reg_name * entry; entry = list_entry(p, struct reg_name, next); - list_del(&entry->next); + llist_del(&entry->next, ®.names); reg_name_destroy(entry); - reg.n_names--; } - list_for_each_safe(p, h, ®.ipcps) { + llist_for_each_safe(p, h, ®.ipcps) { struct reg_ipcp * entry; entry = list_entry(p, struct reg_ipcp, next); - list_del(&entry->next); + llist_del(&entry->next, ®.ipcps); reg_ipcp_destroy(entry); - reg.n_ipcps--; - } - - list_for_each_safe(p, h, ®.flows) { - struct reg_flow * entry; - entry = list_entry(p, struct reg_flow, next); - list_del(&entry->next); - reg_flow_destroy(entry); - reg.n_flows--; } pthread_mutex_unlock(®.mtx); @@ -643,21 +627,15 @@ void reg_clear(void) void reg_fini(void) { - assert(list_is_empty(®.spawned)); - assert(list_is_empty(®.progs)); - assert(list_is_empty(®.procs)); - assert(list_is_empty(®.names)); - assert(list_is_empty(®.ipcps)); - assert(list_is_empty(®.flows)); + assert(llist_is_empty(®.spawned)); + assert(llist_is_empty(®.progs)); + assert(llist_is_empty(®.procs)); + assert(llist_is_empty(®.pools)); + assert(llist_is_empty(®.names)); + assert(llist_is_empty(®.ipcps)); + assert(llist_is_empty(®.flows)); - assert(reg.n_spawned == 0); - assert(reg.n_progs == 0); - assert(reg.n_procs == 0); - assert(reg.n_names == 0); - assert(reg.n_ipcps == 0); - assert(reg.n_flows == 0); - - bmp_destroy(reg.flow_ids); + bmp_destroy(reg.ids); if (pthread_cond_destroy(®.cond) != 0) log_warn("Failed to destroy condvar."); @@ -677,8 +655,8 @@ int reg_create_flow(struct flow_info * info) pthread_mutex_lock(®.mtx); - info->id = bmp_allocate(reg.flow_ids); - if (!bmp_is_id_valid(reg.flow_ids, info->id)) { + info->id = bmp_allocate(reg.ids); + if (!bmp_is_id_valid(reg.ids, info->id)) { log_err("Failed to allocate flow id."); goto fail_id; } @@ -689,16 +667,14 @@ int reg_create_flow(struct flow_info * info) goto fail_flow; } - list_add(&f->next, __reg_after_flow(info->id)); - - reg.n_flows++; + llist_add_at(&f->next, __reg_after_flow(info->id), ®.flows); pthread_mutex_unlock(®.mtx); return 0; fail_flow: - bmp_release(reg.flow_ids, info->id); + bmp_release(reg.ids, info->id); info->id = 0; fail_id: pthread_mutex_unlock(®.mtx); @@ -717,11 +693,9 @@ int reg_destroy_flow(int flow_id) goto no_flow; } - list_del(&f->next); - - reg.n_flows--; + llist_del(&f->next, ®.flows); - bmp_release(reg.flow_ids, flow_id); + bmp_release(reg.ids, flow_id); pthread_mutex_unlock(®.mtx); @@ -757,7 +731,7 @@ int reg_create_ipcp(const struct ipcp_info * info) assert(info != NULL); assert(info->pid != 0); - assert(info->state == IPCP_BOOT); + assert(info->state == IPCP_INIT); pthread_mutex_lock(®.mtx); @@ -780,11 +754,10 @@ int reg_create_ipcp(const struct ipcp_info * info) entry->pid = info->pid; - list_add(&ipcp->next, __reg_after_ipcp(info->pid)); - list_add(&entry->next, __reg_after_spawned(info->pid)); - - reg.n_ipcps++; - reg.n_spawned++; + llist_add_tail_at(&ipcp->next, __reg_after_ipcp(info), ®.ipcps); + llist_add_at(&entry->next, + __reg_after_spawned(info->pid), + ®.spawned); pthread_mutex_unlock(®.mtx); @@ -848,11 +821,11 @@ static int __get_ipcp_info(ipcp_list_msg_t ** msg, (*msg)->name = strdup(ipcp->info.name); if ((*msg)->name == NULL) - goto fail_name; + goto fail_msg; (*msg)->layer = strdup(ipcp->layer.name); if ((*msg)->layer == NULL) - goto fail_layer; + goto fail_msg; (*msg)->pid = ipcp->info.pid; (*msg)->type = ipcp->info.type; @@ -860,10 +833,8 @@ static int __get_ipcp_info(ipcp_list_msg_t ** msg, return 0; - fail_layer: - free((*msg)->name); - fail_name: - free(*msg); + fail_msg: + ipcp_list_msg__free_unpacked(*msg, NULL); *msg = NULL; fail: return -1; @@ -876,37 +847,31 @@ int reg_list_ipcps(ipcp_list_msg_t *** ipcps) pthread_mutex_lock(®.mtx); - if (reg.n_ipcps == 0) { - *ipcps = NULL; + if (llist_is_empty(®.ipcps)) goto finish; - } - *ipcps = malloc(reg.n_ipcps * sizeof(**ipcps)); + *ipcps = malloc(reg.ipcps.len * sizeof(**ipcps)); if (*ipcps == NULL) { log_err("Failed to malloc ipcps."); goto fail_malloc; } - list_for_each(p, ®.ipcps) { + llist_for_each(p, ®.ipcps) { struct reg_ipcp * entry; entry = list_entry(p, struct reg_ipcp, next); - if (__get_ipcp_info(&((*ipcps)[i]), entry) < 0) { - log_err("Failed to create ipcp list info."); + if (__get_ipcp_info(&(*ipcps)[i], entry) < 0) goto fail; - } - ++i; + i++; } - - assert(i == (int) reg.n_ipcps); finish: pthread_mutex_unlock(®.mtx); return i; fail: - while (i > 0) - ipcp_list_msg__free_unpacked((*ipcps)[--i], NULL); + while (i-- > 0) + ipcp_list_msg__free_unpacked((*ipcps)[i], NULL); free(*ipcps); fail_malloc: @@ -934,9 +899,7 @@ int reg_create_name(const struct name_info * info) goto fail_name; } - list_add(&n->next, __reg_after_name(info->name)); - - reg.n_names++; + llist_add_at(&n->next, __reg_after_name(info->name), ®.names); pthread_mutex_unlock(®.mtx); return 0; @@ -965,9 +928,7 @@ int reg_destroy_name(const char * name) __reg_del_name_from_procs(name); __reg_del_name_from_progs(name); - list_del(&n->next); - - reg.n_names--; + llist_del(&n->next, ®.names); pthread_mutex_unlock(®.mtx); @@ -993,28 +954,98 @@ bool reg_has_name(const char * name) return ret; } -static int __get_name_info(name_info_msg_t ** msg, - struct reg_name * n) +int reg_get_name_info(const char * name, + struct name_info * info) { - *msg = malloc(sizeof(**msg)); - if (*msg == NULL) - goto fail; + struct reg_name * n; - name_info_msg__init(*msg); + assert(name != NULL); + assert(info != NULL); - (*msg)->name = strdup(n->info.name); - if ((*msg)->name == NULL) - goto fail_name; + pthread_mutex_lock(®.mtx); - (*msg)->pol_lb = n->info.pol_lb; + n = __reg_get_name(name); + if (n == NULL) { + log_err("Name %s does not exist.", name); + goto no_name; + } + + *info = n->info; + + pthread_mutex_unlock(®.mtx); return 0; - fail_name: - free(*msg); - *msg = NULL; - fail: - return -1; + no_name: + pthread_mutex_unlock(®.mtx); + return -ENOENT; + +} + +int reg_get_name_for_hash(char * buf, + enum hash_algo algo, + const uint8_t * hash) +{ + struct list_head * p; + uint8_t * thash; + size_t len; + char * name = NULL; + + len = hash_len(algo); + + thash = malloc(len); + if (thash == NULL) + return -ENOMEM; + + pthread_mutex_lock(®.mtx); + + llist_for_each(p, ®.names) { + struct reg_name * n = list_entry(p, struct reg_name, next); + str_hash(algo, thash, n->info.name); + if (memcmp(thash, hash, len) == 0) { + name = n->info.name; + break; + } + } + + if (name != NULL) + strcpy(buf, name); + + pthread_mutex_unlock(®.mtx); + + free(thash); + + return name == NULL ? -ENOENT : 0; +} + +int reg_get_name_for_flow_id(char * buf, + int flow_id) +{ + struct reg_flow * f; + + pthread_mutex_lock(®.mtx); + + f = __reg_get_flow(flow_id); + if (f != NULL) + strcpy(buf, f->name); + + pthread_mutex_unlock(®.mtx); + + return f == NULL ? -ENOENT : 0; +} + +void reg_set_name_for_flow_id(const char * name, + int flow_id) +{ + struct reg_flow * f; + + pthread_mutex_lock(®.mtx); + + f = __reg_get_flow(flow_id); + if (f != NULL) + strcpy(f->name, name); + + pthread_mutex_unlock(®.mtx); } int reg_list_names(name_info_msg_t *** names) @@ -1024,35 +1055,43 @@ int reg_list_names(name_info_msg_t *** names) pthread_mutex_lock(®.mtx); - if (reg.n_names == 0) + if (llist_is_empty(®.names)) goto finish; - *names = malloc(reg.n_names * sizeof(**names)); + *names = malloc(reg.names.len * sizeof(**names)); if (*names == NULL) { log_err("Failed to malloc names."); goto fail_malloc; } - list_for_each(p, ®.names) { + llist_for_each(p, ®.names) { struct reg_name * entry; entry = list_entry(p, struct reg_name, next); - if (__get_name_info(&((*names)[i]), entry) < 0) { + (*names)[i] = name_info_s_to_msg(&entry->info); + if ((*names)[i] == NULL) { log_err("Failed to create name list info."); goto fail; } - - ++i; + /* wipe security info to avoid huge messages */ + free((*names)[i]->scrt); + (*names)[i]->scrt = NULL; + free((*names)[i]->skey); + (*names)[i]->skey = NULL; + free((*names)[i]->ccrt); + (*names)[i]->ccrt = NULL; + free((*names)[i]->ckey); + (*names)[i]->ckey = NULL; + + i++; } - - assert(i == (int) reg.n_names); finish: pthread_mutex_unlock(®.mtx); return i; fail: - while (i > 0) - name_info_msg__free_unpacked((*names)[--i], NULL); + while (i-- > 0) + name_info_msg__free_unpacked((*names)[i], NULL); free(*names); fail_malloc: @@ -1061,6 +1100,34 @@ int reg_list_names(name_info_msg_t *** names) return -ENOMEM; } +int reg_prepare_pool(uid_t uid, + gid_t gid) +{ + struct reg_pool * pool; + + if (is_ouroboros_member_uid(uid)) + return 0; + + pthread_mutex_lock(®.mtx); + + pool = __reg_get_pool(uid); + if (pool == NULL) { + pool = reg_pool_create(uid, gid); + if (pool == NULL) { + log_err("Failed to create pool for uid %d.", uid); + pthread_mutex_unlock(®.mtx); + return -1; + } + llist_add(&pool->next, ®.pools); + } + + reg_pool_ref(pool); + + pthread_mutex_unlock(®.mtx); + + return 0; +} + int reg_create_proc(const struct proc_info * info) { struct reg_proc * proc; @@ -1071,20 +1138,18 @@ int reg_create_proc(const struct proc_info * info) if (__reg_get_proc(info->pid) != NULL) { log_err("Process %d already exists.", info->pid); - goto fail_proc; + goto fail; } proc = reg_proc_create(info); if (proc == NULL) { log_err("Failed to create process %d.", info->pid); - goto fail_proc; + goto fail; } __reg_proc_update_names(proc); - list_add(&proc->next, __reg_after_proc(info->pid)); - - reg.n_procs++; + llist_add_at(&proc->next, __reg_after_proc(info->pid), ®.procs); pthread_cond_broadcast(®.cond); @@ -1092,7 +1157,7 @@ int reg_create_proc(const struct proc_info * info) return 0; - fail_proc: + fail: pthread_mutex_unlock(®.mtx); return -1; } @@ -1100,6 +1165,7 @@ int reg_create_proc(const struct proc_info * info) int reg_destroy_proc(pid_t pid) { struct reg_proc * proc; + struct reg_pool * pool = NULL; struct pid_entry * spawn; struct reg_ipcp * ipcp; @@ -1107,24 +1173,27 @@ int reg_destroy_proc(pid_t pid) proc = __reg_get_proc(pid); if (proc != NULL) { - list_del(&proc->next); - reg.n_procs--; + if (!is_ouroboros_member_uid(proc->info.uid)) + pool = __reg_get_pool(proc->info.uid); + llist_del(&proc->next, ®.procs); reg_proc_destroy(proc); __reg_del_proc_from_names(pid); __reg_cancel_flows_for_proc(pid); + if (pool != NULL && reg_pool_unref(pool) == 0) { + llist_del(&pool->next, ®.pools); + reg_pool_destroy(pool); + } } spawn = __reg_get_spawned(pid); if (spawn != NULL) { - list_del(&spawn->next); - reg.n_spawned--; + llist_del(&spawn->next, ®.spawned); free(spawn); } ipcp = __reg_get_ipcp(pid); if (ipcp != NULL) { - list_del(&ipcp->next); - reg.n_ipcps--; + llist_del(&ipcp->next, ®.ipcps); reg_ipcp_destroy(ipcp); } @@ -1146,6 +1215,38 @@ bool reg_has_proc(pid_t pid) return ret; } +bool reg_is_proc_privileged(pid_t pid) +{ + struct reg_proc * proc; + bool ret = false; + + pthread_mutex_lock(®.mtx); + + proc = __reg_get_proc(pid); + if (proc != NULL) + ret = reg_proc_is_privileged(proc); + + pthread_mutex_unlock(®.mtx); + + return ret; +} + +uid_t reg_get_proc_uid(pid_t pid) +{ + struct reg_proc * proc; + uid_t ret = 0; + + pthread_mutex_lock(®.mtx); + + proc = __reg_get_proc(pid); + if (proc != NULL && !is_ouroboros_member_uid(proc->info.uid)) + ret = proc->info.uid; + + pthread_mutex_unlock(®.mtx); + + return ret; +} + void reg_kill_all_proc(int signal) { pthread_mutex_lock(®.mtx); @@ -1187,9 +1288,7 @@ int reg_create_spawned(pid_t pid) entry->pid = pid; - list_add(&entry->next, __reg_after_spawned(pid)); - - reg.n_spawned++; + llist_add_at(&entry->next, __reg_after_spawned(pid), ®.spawned); pthread_mutex_unlock(®.mtx); @@ -1359,9 +1458,7 @@ int reg_create_prog(const struct prog_info * info) goto fail_prog; } - list_add(&prog->next, __reg_after_prog(info->name)); - - reg.n_progs++; + llist_add_at(&prog->next, __reg_after_prog(info->name), ®.progs); exists: pthread_mutex_unlock(®.mtx); @@ -1389,9 +1486,7 @@ int reg_destroy_prog(const char * name) __reg_del_prog_from_names(prog->info.path); - list_del(&prog->next); - - reg.n_progs--; + llist_del(&prog->next, ®.progs); pthread_mutex_unlock(®.mtx); @@ -1419,19 +1514,18 @@ bool reg_has_prog(const char * name) return ret; } -int reg_get_exec(enum hash_algo algo, - const uint8_t * hash, - char *** prog) +int reg_get_exec(const char * name, + char *** prog) { char ** exec; int ret = 0; - assert(hash != NULL); + assert(name != NULL); assert(prog != NULL); pthread_mutex_lock(®.mtx); - exec = __reg_get_exec(algo, hash); + exec = __reg_get_exec(name); if (exec == NULL) { ret = -EPERM; goto finish; @@ -1444,12 +1538,9 @@ int reg_get_exec(enum hash_algo algo, goto finish; } - pthread_mutex_unlock(®.mtx); - - return 0; - finish: pthread_mutex_unlock(®.mtx); + return ret; } @@ -1557,8 +1648,7 @@ int reg_set_layer_for_ipcp(struct ipcp_info * info, struct reg_ipcp * ipcp; assert(info != NULL); - assert(info->state > IPCP_BOOT); - assert(info->state < IPCP_SHUTDOWN); + assert(info->state == IPCP_BOOT); pthread_mutex_lock(®.mtx); @@ -1690,7 +1780,7 @@ int reg_wait_flow_allocated(struct flow_info * info, stop = true; break; case FLOW_DEALLOCATED: - ret = -1; + ret = flow->response; stop = true; break; default: @@ -1712,7 +1802,8 @@ int reg_wait_flow_allocated(struct flow_info * info, } if (flow != NULL) { - reg_flow_get_data(flow, pbuf); + *pbuf = flow->rsp_data; + clrbuf(flow->rsp_data); *info = flow->info; } @@ -1722,7 +1813,8 @@ int reg_wait_flow_allocated(struct flow_info * info, } int reg_respond_alloc(struct flow_info * info, - buffer_t * pbuf) + buffer_t * pbuf, + int response) { struct reg_flow * flow; @@ -1745,9 +1837,13 @@ int reg_respond_alloc(struct flow_info * info, goto fail_flow; } - assert(flow->info.state == FLOW_ALLOC_PENDING); - assert(flow->data.len == 0); - assert(flow->data.data == NULL); + if (flow->info.state != FLOW_ALLOC_PENDING) { + log_warn("Flow %d already responded.", info->id); + goto fail_flow; + } + + assert(flow->rsp_data.len == 0); + assert(flow->rsp_data.data == NULL); info->n_pid = flow->info.n_pid; info->n_1_pid = flow->info.n_pid; @@ -1755,10 +1851,14 @@ int reg_respond_alloc(struct flow_info * info, if (reg_flow_update(flow, info) < 0) { log_err("Failed to create flow structs."); goto fail_flow; - }; + } + + flow->response = response; - if (info->state == FLOW_ALLOCATED) - reg_flow_set_data(flow, pbuf); + if (info->state == FLOW_ALLOCATED) { + flow->rsp_data = *pbuf; + clrbuf(*pbuf); + } pthread_cond_broadcast(®.cond); @@ -1771,8 +1871,7 @@ int reg_respond_alloc(struct flow_info * info, return -1; } -int reg_prepare_flow_accept(struct flow_info * info, - buffer_t * pbuf) +int reg_prepare_flow_accept(struct flow_info * info) { struct reg_flow * flow; int ret; @@ -1790,7 +1889,7 @@ int reg_prepare_flow_accept(struct flow_info * info, ret = reg_flow_update(flow, info); - reg_flow_set_data(flow, pbuf); + pthread_cond_broadcast(®.cond); pthread_mutex_unlock(®.mtx); @@ -1824,8 +1923,6 @@ int reg_wait_flow_accepted(struct flow_info * info, assert(flow != NULL); assert(info->id == flow->info.id); assert(info->n_pid == flow->info.n_pid); - assert(info->state == flow->info.state); - assert(flow->info.state == FLOW_ACCEPT_PENDING); if (__reg_add_active_proc(info->n_pid) < 0) { log_err("Failed to mark pid %d active.", info->n_pid); @@ -1871,7 +1968,8 @@ int reg_wait_flow_accepted(struct flow_info * info, pthread_cleanup_pop(true); /* __cleanup_wait_accept */ if (flow != NULL) { - reg_flow_get_data(flow, pbuf); + *pbuf = flow->req_data; + clrbuf(flow->req_data); *info = flow->info; } @@ -1883,13 +1981,12 @@ int reg_wait_flow_accepted(struct flow_info * info, return -1; } -int reg_wait_flow_accepting(enum hash_algo algo, - const uint8_t * hash, +int reg_wait_flow_accepting(const char * name, const struct timespec * abstime) { int ret; - assert(hash != NULL); + assert(name != NULL); assert(abstime != NULL); pthread_mutex_lock(®.mtx); @@ -1897,7 +1994,7 @@ int reg_wait_flow_accepting(enum hash_algo algo, pthread_cleanup_push(__cleanup_mutex_unlock, ®.mtx); while (true) { - ret = __reg_get_pending_flow_id_for_hash(algo, hash); + ret = __reg_get_pending_flow_id(name); if (ret != -EAGAIN) break; @@ -1915,7 +2012,6 @@ int reg_respond_accept(struct flow_info * info, buffer_t * pbuf) { struct reg_flow * flow; - buffer_t temp; assert(info != NULL); assert(info->state == FLOW_ALLOCATED); @@ -1933,12 +2029,64 @@ int reg_respond_accept(struct flow_info * info, info->n_pid = flow->info.n_pid; - if (info->qs.cypher_s > 0) { - reg_flow_get_data(flow, &temp); - reg_flow_set_data(flow, pbuf); - *pbuf = temp; + flow->req_data = *pbuf; + clrbuf(*pbuf); + + if (reg_flow_update(flow, info) < 0) { + log_err("Failed to create flow structs."); + goto fail_flow; + } + + pthread_cond_broadcast(®.cond); + + pthread_mutex_unlock(®.mtx); + + return 0; + + fail_flow: + pthread_mutex_unlock(®.mtx); + return -1; +} + +int reg_prepare_flow_direct(struct flow_info * info, + buffer_t * pbuf, + uid_t alloc_uid) +{ + struct reg_flow * flow; + struct reg_proc * proc; + uid_t accept_uid = 0; + + assert(info != NULL); + assert(info->state == FLOW_ALLOCATED); + assert(info->n_1_pid != 0); + assert(pbuf != NULL); + + pthread_mutex_lock(®.mtx); + + flow = __reg_get_flow(info->id); + if (flow == NULL) { + log_err("Flow not found: %d.", info->id); + goto fail_flow; + } + + assert(flow->info.state == FLOW_ACCEPT_PENDING); + + info->n_pid = flow->info.n_pid; + + proc = __reg_get_proc(info->n_pid); + if (proc != NULL && !is_ouroboros_member_uid(proc->info.uid)) + accept_uid = proc->info.uid; + + if (alloc_uid != accept_uid) { + pthread_mutex_unlock(®.mtx); + return -EPERM; } + flow->direct = true; + + flow->req_data = *pbuf; + clrbuf(*pbuf); + if (reg_flow_update(flow, info) < 0) { log_err("Failed to create flow structs."); goto fail_flow; @@ -1955,6 +2103,614 @@ int reg_respond_accept(struct flow_info * info, return -1; } +bool reg_flow_is_direct(int flow_id) +{ + struct reg_flow * flow; + bool ret; + + pthread_mutex_lock(®.mtx); + + flow = __reg_get_flow(flow_id); + + ret = flow != NULL && flow->direct; + + pthread_mutex_unlock(®.mtx); + + return ret; +} + +void reg_flow_set_rekey(int flow_id, + bool initiator, + buffer_t peer_crt) +{ + struct reg_flow * flow; + uint8_t * crt = NULL; + + /* Copy the cert outside the lock; publish it with rk.encrypted. */ + if (peer_crt.len > 0) { + crt = malloc(peer_crt.len); + if (crt != NULL) + memcpy(crt, peer_crt.data, peer_crt.len); + else + log_warn("Failed to cache peer cert for re-key."); + } + + pthread_mutex_lock(®.mtx); + + flow = __reg_get_flow(flow_id); + if (flow != NULL) { + flow->rk.encrypted = true; + flow->rk.initiator = initiator; + flow->rk.epoch = 0; + if (crt != NULL) { + freebuf(flow->rk.peer_crt); + flow->rk.peer_crt.data = crt; + flow->rk.peer_crt.len = peer_crt.len; + crt = NULL; + } + } + + pthread_mutex_unlock(®.mtx); + + free(crt); +} + +int reg_flow_get_peer_crt(int flow_id, + buffer_t * crt) +{ + struct reg_flow * flow; + int ret = -ENOENT; + + assert(crt != NULL); + + clrbuf(*crt); + + pthread_mutex_lock(®.mtx); + + flow = __reg_get_flow(flow_id); + if (flow != NULL && flow->rk.peer_crt.len > 0) { + crt->data = malloc(flow->rk.peer_crt.len); + if (crt->data == NULL) { + ret = -ENOMEM; + } else { + memcpy(crt->data, flow->rk.peer_crt.data, + flow->rk.peer_crt.len); + crt->len = flow->rk.peer_crt.len; + ret = 0; + } + } + + pthread_mutex_unlock(®.mtx); + + return ret; +} + +int reg_flow_get_epoch(int flow_id) +{ + struct reg_flow * flow; + int epoch = -1; + + pthread_mutex_lock(®.mtx); + + flow = __reg_get_flow(flow_id); + if (flow != NULL && flow->rk.encrypted) + epoch = flow->rk.epoch; + + pthread_mutex_unlock(®.mtx); + + return epoch; +} + +bool reg_flow_rekey_pending(int flow_id) +{ + struct reg_flow * flow; + bool ret = false; + + pthread_mutex_lock(®.mtx); + + flow = __reg_get_flow(flow_id); + if (flow != NULL) + ret = flow->rk.has_pending; + + pthread_mutex_unlock(®.mtx); + + return ret; +} + +pid_t reg_flow_get_n_1_pid(int flow_id) +{ + struct reg_flow * flow; + pid_t pid = -1; + + pthread_mutex_lock(®.mtx); + + flow = __reg_get_flow(flow_id); + if (flow != NULL) + pid = flow->info.n_1_pid; + + pthread_mutex_unlock(®.mtx); + + return pid; +} + +int reg_flow_snapshot_rekey_due(struct rekey_info * snap, + int max) +{ + struct list_head * p; + int n = 0; + + pthread_mutex_lock(®.mtx); + + llist_for_each(p, ®.flows) { + struct reg_flow * f; + + if (n == max) + break; + + f = list_entry(p, struct reg_flow, next); + + if (f->info.state != FLOW_ALLOCATED) + continue; + + if (!f->rk.encrypted) + continue; + + /* Direct flows have no IPCP initiator; either side drives. */ + if (!f->direct && !f->rk.initiator) + continue; + + if (f->rk.in_flight || f->rk.has_pending) + continue; + + f->rk.in_flight = true; + + snap[n].flow_id = f->info.id; + snap[n].n_pid = f->info.n_pid; + snap[n].n_1_pid = f->info.n_1_pid; + snap[n].epoch = f->rk.epoch; + snap[n].direct = f->direct; + strcpy(snap[n].name, f->name); + ++n; + } + + pthread_mutex_unlock(®.mtx); + + return n; +} + +void reg_flow_clear_in_flight(int flow_id) +{ + struct reg_flow * flow; + + pthread_mutex_lock(®.mtx); + + flow = __reg_get_flow(flow_id); + if (flow != NULL) + flow->rk.in_flight = false; + + pthread_mutex_unlock(®.mtx); +} + +/* Test-and-set the in-flight latch; refuse if a re-key is already active. */ +bool reg_flow_rekey_begin(int flow_id) +{ + struct reg_flow * flow; + bool ret = false; + + pthread_mutex_lock(®.mtx); + + flow = __reg_get_flow(flow_id); + if (flow != NULL && flow->rk.encrypted) { + if (!flow->rk.in_flight && !flow->rk.has_pending) { + flow->rk.in_flight = true; + ret = true; + } + } + + pthread_mutex_unlock(®.mtx); + + return ret; +} + +/* Initiator yields the responder role while driving its own exchange. */ +bool reg_flow_rekey_should_yield(int flow_id) +{ + struct reg_flow * flow; + bool ret = false; + + pthread_mutex_lock(®.mtx); + + flow = __reg_get_flow(flow_id); + if (flow != NULL) + ret = flow->rk.initiator && flow->rk.in_flight; + + pthread_mutex_unlock(®.mtx); + + return ret; +} + +int reg_flow_store_pending(int flow_id, + const uint8_t * seed, + uint8_t epoch, + bool initiator) +{ + struct reg_flow * flow; + int ret = -ENOENT; + + pthread_mutex_lock(®.mtx); + + flow = __reg_get_flow(flow_id); + if (flow != NULL) { + /* Exchange done: release the latch regardless of parking. */ + flow->rk.in_flight = false; + + if (flow->rk.pending_seed == NULL) + flow->rk.pending_seed = crypt_secure_malloc(SYMMKEYSZ); + + if (flow->rk.pending_seed != NULL) { + memcpy(flow->rk.pending_seed, seed, SYMMKEYSZ); + flow->rk.pending_epoch = epoch; + flow->rk.pending_initiator = initiator; + flow->rk.has_pending = true; + /* Doorbell raised only after the seed is parked. */ + if (flow->n_rb != NULL) + ssm_rbuff_set_bits(flow->n_rb, RB_REKEY); + ret = 0; + } else { + ret = -ENOMEM; + } + } + + pthread_mutex_unlock(®.mtx); + + return ret; +} + +/* Direct re-key: which of the two local apps has pulled the seed. */ +#define RK_N_PID 0x1 /* acceptor (n_pid) pulled the seed */ +#define RK_N_1_PID 0x2 /* allocator (n_1_pid) pulled the seed */ +#define RK_PID_MASK (RK_N_PID | RK_N_1_PID) + +/* + * Park a single re-key seed for a direct flow and ring BOTH apps' + * doorbells. The seed is the one shared secret; each app pulls it once + * (reg_flow_take_pending), so it is held until both have taken it. + */ +int reg_flow_store_pending_direct(int flow_id, + const uint8_t * seed, + uint8_t epoch) +{ + struct reg_flow * flow; + int ret = -ENOENT; + + pthread_mutex_lock(®.mtx); + + flow = __reg_get_flow(flow_id); + if (flow == NULL) + goto out; + + /* Exchange done: release the latch regardless of parking. */ + flow->rk.in_flight = false; + + if (flow->rk.pending_seed == NULL) + flow->rk.pending_seed = crypt_secure_malloc(SYMMKEYSZ); + + if (flow->rk.pending_seed == NULL) { + ret = -ENOMEM; + goto out; + } + + memcpy(flow->rk.pending_seed, seed, SYMMKEYSZ); + flow->rk.pending_epoch = epoch; + flow->rk.has_pending = true; + flow->rk.pulled = 0; + + /* A departed peer never pulls; treat its side as already done. */ + if (flow->info.n_pid <= 0) + flow->rk.pulled |= RK_N_PID; + + if (flow->info.n_1_pid <= 0) + flow->rk.pulled |= RK_N_1_PID; + + if (flow->n_rb != NULL && !(flow->rk.pulled & RK_N_PID)) + ssm_rbuff_set_bits(flow->n_rb, RB_REKEY); + + if (flow->n_1_rb != NULL && !(flow->rk.pulled & RK_N_1_PID)) + ssm_rbuff_set_bits(flow->n_1_rb, RB_REKEY); + + ret = 0; + out: + pthread_mutex_unlock(®.mtx); + + return ret; +} + +/* A caller may act on a flow if it is privileged or owns the flow. */ +static bool uid_may_access(uid_t caller, + uid_t owner) +{ + return is_ouroboros_member_uid(caller) || caller == owner; +} + +/* + * Caller holds reg.mtx. The direct seed is shared by both apps, so the + * per-app initiator role is resolved from the verified caller pid (the + * allocator is n_1_pid), and the seed is held until both have pulled. + */ +static void __take_pending_direct(struct reg_flow * flow, + pid_t cpid, + uint8_t * seed, + uint8_t * epoch, + bool * initiator) +{ + bool allocator; + + allocator = cpid == flow->info.n_1_pid; + + memcpy(seed, flow->rk.pending_seed, SYMMKEYSZ); + *epoch = flow->rk.pending_epoch; + *initiator = allocator; + flow->rk.epoch = flow->rk.pending_epoch; + + if (allocator) { + flow->rk.pulled |= RK_N_1_PID; + if (flow->n_1_rb != NULL) + ssm_rbuff_clr_bits(flow->n_1_rb, RB_REKEY); + } else { + flow->rk.pulled |= RK_N_PID; + if (flow->n_rb != NULL) + ssm_rbuff_clr_bits(flow->n_rb, RB_REKEY); + } + + if ((flow->rk.pulled & RK_PID_MASK) != RK_PID_MASK) + return; + + flow->rk.has_pending = false; + flow->rk.pulled = 0; + crypt_secure_clear(flow->rk.pending_seed, SYMMKEYSZ); +} + +int reg_flow_take_pending(int flow_id, + uid_t uid, + pid_t cpid, + uint8_t * seed, + uint8_t * epoch, + bool * initiator) +{ + struct reg_flow * flow; + int ret = -ENOENT; + + pthread_mutex_lock(®.mtx); + + flow = __reg_get_flow(flow_id); + if (flow == NULL || !flow->rk.has_pending) + goto out; + + if (!uid_may_access(uid, flow->info.uid)) { + ret = -EPERM; + goto out; + } + + if (flow->direct) { + __take_pending_direct(flow, cpid, seed, epoch, initiator); + ret = 0; + goto out; + } + + memcpy(seed, flow->rk.pending_seed, SYMMKEYSZ); + *epoch = flow->rk.pending_epoch; + *initiator = flow->rk.pending_initiator; + flow->rk.epoch = flow->rk.pending_epoch; + flow->rk.has_pending = false; + crypt_secure_clear(flow->rk.pending_seed, SYMMKEYSZ); + if (flow->n_rb != NULL) + ssm_rbuff_clr_bits(flow->n_rb, RB_REKEY); + + ret = 0; + out: + pthread_mutex_unlock(®.mtx); + + return ret; +} + +/* + * Admit a peer-driven re-key arrival before a worker event is allocated: + * the flow must exist, carry a cipher, and the update must come from its + * own lower IPCP. Coalesces to one queued REQ and one queued RESP per flow + * so a flooding peer cannot grow the inbox without bound. + */ +bool reg_flow_rekey_arr_admit(int flow_id, + pid_t n_1_pid, + bool is_req) +{ + struct reg_flow * flow; + bool admit = false; + + pthread_mutex_lock(®.mtx); + + flow = __reg_get_flow(flow_id); + if (flow != NULL && flow->rk.encrypted + && flow->info.n_1_pid == n_1_pid) { + if (is_req && !flow->rk.req_queued) { + flow->rk.req_queued = true; + admit = true; + } else if (!is_req && flow->rk.in_flight + && !flow->rk.resp_queued) { + flow->rk.resp_queued = true; + admit = true; + } + } + + pthread_mutex_unlock(®.mtx); + + return admit; +} + +void reg_flow_rekey_arr_done(int flow_id, + bool is_req) +{ + struct reg_flow * flow; + + pthread_mutex_lock(®.mtx); + + flow = __reg_get_flow(flow_id); + if (flow != NULL) { + if (is_req) + flow->rk.req_queued = false; + else + flow->rk.resp_queued = false; + } + + pthread_mutex_unlock(®.mtx); +} + +bool reg_flow_owned_by(int flow_id, + uid_t uid) +{ + struct reg_flow * flow; + bool ret = false; + + pthread_mutex_lock(®.mtx); + + flow = __reg_get_flow(flow_id); + if (flow != NULL) + ret = uid_may_access(uid, flow->info.uid); + + pthread_mutex_unlock(®.mtx); + + return ret; +} + +/* Caller holds reg.mtx. */ +static void __notify_proc(pid_t pid, + int flow_id, + int event) +{ + struct reg_proc * proc; + + proc = __reg_get_proc(pid); + if (proc != NULL) + ssm_flow_set_notify(proc->set, flow_id, event); +} + +void reg_notify_flow(int flow_id, + int event) +{ + struct reg_flow * flow; + + pthread_mutex_lock(®.mtx); + + flow = __reg_get_flow(flow_id); + if (flow != NULL) + __notify_proc(flow->info.n_pid, flow_id, event); + + pthread_mutex_unlock(®.mtx); +} + +/* Wake both endpoints of a direct flow (acceptor and allocator). */ +void reg_notify_flow_peers(int flow_id, + int event) +{ + struct reg_flow * flow; + + pthread_mutex_lock(®.mtx); + + flow = __reg_get_flow(flow_id); + if (flow != NULL) { + __notify_proc(flow->info.n_pid, flow_id, event); + __notify_proc(flow->info.n_1_pid, flow_id, event); + } + + pthread_mutex_unlock(®.mtx); +} + +int reg_respond_flow_direct(int flow_id, + buffer_t * pbuf) +{ + struct reg_flow * flow; + + assert(pbuf != NULL); + + pthread_mutex_lock(®.mtx); + + flow = __reg_get_flow(flow_id); + if (flow == NULL) { + log_err("Flow %d not found.", flow_id); + goto fail; + } + + assert(flow->direct); + assert(flow->rsp_data.data == NULL); + + flow->rsp_data = *pbuf; + clrbuf(*pbuf); + + pthread_cond_broadcast(®.cond); + + pthread_mutex_unlock(®.mtx); + + return 0; + fail: + pthread_mutex_unlock(®.mtx); + return -1; +} + +int reg_wait_flow_direct(int flow_id, + buffer_t * pbuf, + const struct timespec * abstime) +{ + struct reg_flow * flow; + int ret = -1; + + assert(pbuf != NULL); + + pthread_mutex_lock(®.mtx); + + flow = __reg_get_flow(flow_id); + if (flow == NULL) + goto fail; + + assert(flow->direct); + + pthread_cleanup_push(__cleanup_mutex_unlock, ®.mtx); + + while (flow != NULL && flow->rsp_data.data == NULL) { + ret = -__timedwait(®.cond, ®.mtx, abstime); + if (ret == -ETIMEDOUT) + break; + flow = __reg_get_flow(flow_id); + } + + if (flow != NULL && flow->rsp_data.data != NULL) { + *pbuf = flow->rsp_data; + clrbuf(flow->rsp_data); + ret = 0; + } + + pthread_cleanup_pop(true); + + return ret; + fail: + pthread_mutex_unlock(®.mtx); + return -1; +} + +static int direct_flow_dealloc(struct reg_flow * flow, + pid_t pid) +{ + if (!flow->direct) + return -1; + + if (pid == flow->info.n_pid && flow->info.n_pid != -1) + flow->info.n_pid = -1; + else if (pid == flow->info.n_1_pid && flow->info.n_1_pid != -1) + flow->info.n_1_pid = -1; + else + return -1; + + return 0; +} + void reg_dealloc_flow(struct flow_info * info) { struct reg_flow * flow; @@ -1968,13 +2724,34 @@ void reg_dealloc_flow(struct flow_info * info) flow = __reg_get_flow(info->id); assert(flow != NULL); - assert(flow->data.data == NULL); - assert(flow->data.len == 0); + assert(flow->req_data.data == NULL); + assert(flow->req_data.len == 0); + assert(flow->rsp_data.data == NULL); + assert(flow->rsp_data.len == 0); + + info->n_1_pid = flow->info.n_1_pid; + + if (flow->info.state == FLOW_DEALLOC_PENDING) { + if (direct_flow_dealloc(flow, info->n_pid) < 0) { + info->state = FLOW_DEALLOC_PENDING; + pthread_mutex_unlock(®.mtx); + return; + } + flow->info.state = FLOW_DEALLOCATED; + info->state = FLOW_DEALLOCATED; + reg_flow_update(flow, info); + pthread_mutex_unlock(®.mtx); + return; + } assert(flow->info.state == FLOW_ALLOCATED); + flow->info.state = FLOW_DEALLOC_PENDING; info->state = FLOW_DEALLOC_PENDING; - info->n_1_pid = flow->info.n_1_pid; + + direct_flow_dealloc(flow, info->n_pid); + + memset(flow->name, 0, sizeof(flow->name)); reg_flow_update(flow, info); @@ -1994,8 +2771,10 @@ void reg_dealloc_flow_resp(struct flow_info * info) flow = __reg_get_flow(info->id); assert(flow != NULL); - assert(flow->data.data == NULL); - assert(flow->data.len == 0); + assert(flow->req_data.data == NULL); + assert(flow->req_data.len == 0); + assert(flow->rsp_data.data == NULL); + assert(flow->rsp_data.len == 0); assert(flow->info.state == FLOW_DEALLOC_PENDING); flow->info.state = FLOW_DEALLOCATED; @@ -2043,7 +2822,7 @@ int reg_wait_ipcp_boot(struct ipcp_info * info, int ret; bool stop = false; - assert(info->state == IPCP_BOOT); + assert(info->state == IPCP_INIT); pthread_mutex_lock(®.mtx); @@ -2063,16 +2842,18 @@ int reg_wait_ipcp_boot(struct ipcp_info * info, ret = -1; stop = true; break; + case IPCP_BOOT: + /* FALLTHRU*/ case IPCP_OPERATIONAL: ret = 0; stop = true; break; - case IPCP_BOOT: + case IPCP_INIT: ret = -__timedwait(®.cond, ®.mtx, abstime); break; default: assert(false); - continue; /* Shut up static analyzer. */ + break; /* Shut up static analyzer. */ } ipcp = __reg_get_ipcp(info->pid); diff --git a/src/irmd/reg/reg.h b/src/irmd/reg/reg.h index 17dfcc32..8a313d46 100644 --- a/src/irmd/reg/reg.h +++ b/src/irmd/reg/reg.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * The IPC Resource Manager - Registry * @@ -31,6 +31,8 @@ #include <ouroboros/time.h> #include <ouroboros/utils.h> +#include "pool.h" + int reg_init(void); void reg_clear(void); @@ -50,6 +52,13 @@ int reg_destroy_proc(pid_t pid); bool reg_has_proc(pid_t pid); +bool reg_is_proc_privileged(pid_t pid); + +int reg_prepare_pool(uid_t uid, + gid_t gid); + +uid_t reg_get_proc_uid(pid_t pid); + void reg_kill_all_proc(int signal); pid_t reg_get_dead_proc(void); @@ -90,6 +99,19 @@ int reg_destroy_name(const char * name); bool reg_has_name(const char * name); +int reg_get_name_info(const char * name, + struct name_info * info); + +int reg_get_name_for_hash(char * buf, + enum hash_algo algo, + const uint8_t * hash); + +int reg_get_name_for_flow_id(char * buf, + int flow_id); + +void reg_set_name_for_flow_id(const char * name, + int flow_id); + /* TODO don't rely on protobuf here */ int reg_list_names(name_info_msg_t *** names); @@ -99,9 +121,8 @@ int reg_destroy_prog(const char * name); bool reg_has_prog(const char * name); -int reg_get_exec(enum hash_algo algo, - const uint8_t * hash, - char *** exec); +int reg_get_exec(const char * name, + char *** exec); int reg_bind_prog(const char * name, char ** exec, @@ -117,22 +138,98 @@ int reg_wait_flow_allocated(struct flow_info * info, const struct timespec * abstime); int reg_respond_alloc(struct flow_info * info, - buffer_t * pbuf); + buffer_t * pbuf, + int response); -int reg_prepare_flow_accept(struct flow_info * info, - buffer_t * pbuf); +int reg_prepare_flow_accept(struct flow_info * info); int reg_wait_flow_accepted(struct flow_info * info, buffer_t * pbuf, const struct timespec * abstime); -int reg_wait_flow_accepting(enum hash_algo algo, - const uint8_t * hash, +int reg_wait_flow_accepting(const char * name, const struct timespec * abstime); int reg_respond_accept(struct flow_info * info, buffer_t * pbuf); +int reg_prepare_flow_direct(struct flow_info * info, + buffer_t * pbuf, + uid_t alloc_uid); + +int reg_respond_flow_direct(int flow_id, + buffer_t * pbuf); + +int reg_wait_flow_direct(int flow_id, + buffer_t * pbuf, + const struct timespec * abstime); + +bool reg_flow_is_direct(int flow_id); + +/* Per-flow snapshot for the re-key timer */ +struct rekey_info { + int flow_id; + pid_t n_pid; + pid_t n_1_pid; + char name[NAME_SIZE + 1]; + uint8_t epoch; + bool direct; +}; + +void reg_flow_set_rekey(int flow_id, + bool initiator, + buffer_t peer_crt); + +int reg_flow_get_peer_crt(int flow_id, + buffer_t * crt); + +int reg_flow_get_epoch(int flow_id); + +bool reg_flow_rekey_pending(int flow_id); + +pid_t reg_flow_get_n_1_pid(int flow_id); + +int reg_flow_snapshot_rekey_due(struct rekey_info * snap, + int max); + +void reg_flow_clear_in_flight(int flow_id); + +bool reg_flow_rekey_begin(int flow_id); + +bool reg_flow_rekey_should_yield(int flow_id); + +int reg_flow_store_pending(int flow_id, + const uint8_t * seed, + uint8_t epoch, + bool initiator); + +int reg_flow_store_pending_direct(int flow_id, + const uint8_t * seed, + uint8_t epoch); + +int reg_flow_take_pending(int flow_id, + uid_t uid, + pid_t cpid, + uint8_t * seed, + uint8_t * epoch, + bool * initiator); + +bool reg_flow_rekey_arr_admit(int flow_id, + pid_t n_1_pid, + bool is_req); + +void reg_flow_rekey_arr_done(int flow_id, + bool is_req); + +bool reg_flow_owned_by(int flow_id, + uid_t uid); + +void reg_notify_flow(int flow_id, + int event); + +void reg_notify_flow_peers(int flow_id, + int event); + void reg_dealloc_flow(struct flow_info * info); void reg_dealloc_flow_resp(struct flow_info * info); diff --git a/src/irmd/reg/tests/CMakeLists.txt b/src/irmd/reg/tests/CMakeLists.txt index bc1354ed..e8521545 100644 --- a/src/irmd/reg/tests/CMakeLists.txt +++ b/src/irmd/reg/tests/CMakeLists.txt @@ -1,7 +1,9 @@ -get_filename_component(tmp ".." ABSOLUTE) -get_filename_component(src_folder "${tmp}" NAME) +get_filename_component(PARENT_PATH ${CMAKE_CURRENT_SOURCE_DIR} DIRECTORY) +get_filename_component(PARENT_DIR ${PARENT_PATH} NAME) -create_test_sourcelist(${src_folder}_tests test_suite.c +compute_test_prefix() + +create_test_sourcelist(${PARENT_DIR}_tests test_suite.c # Add new tests here flow_test.c ipcp_test.c @@ -11,19 +13,21 @@ create_test_sourcelist(${src_folder}_tests test_suite.c reg_test.c ) -add_executable(${src_folder}_test EXCLUDE_FROM_ALL ${${src_folder}_tests}) -target_link_libraries(${src_folder}_test ouroboros-common) +add_executable(${PARENT_DIR}_test ${${PARENT_DIR}_tests}) -if (CMAKE_BUILD_TYPE MATCHES "Debug*") - add_compile_flags(${src_folder}_test -DCONFIG_OUROBOROS_DEBUG) -endif () +target_include_directories(${PARENT_DIR}_test PRIVATE + ${CMAKE_CURRENT_SOURCE_DIR} + ${CMAKE_CURRENT_BINARY_DIR} + ${CMAKE_SOURCE_DIR}/include + ${CMAKE_BINARY_DIR}/include + ${CMAKE_SOURCE_DIR}/src/irmd + ${CMAKE_BINARY_DIR}/src/irmd +) -add_dependencies(check ${src_folder}_test) +disable_test_logging_for_target(${PARENT_DIR}_test) +target_link_libraries(${PARENT_DIR}_test PRIVATE ouroboros-common) +ouroboros_target_debug_definitions(${PARENT_DIR}_test) -set(tests_to_run ${${src_folder}_tests}) -remove(tests_to_run test_suite.c) +add_dependencies(build_tests ${PARENT_DIR}_test) -foreach(test ${tests_to_run}) - get_filename_component(test_name ${test} NAME_WE) - add_test(irmd/reg/${test_name} ${C_TEST_PATH}/${src_folder}_test ${test_name}) -endforeach(test) +ouroboros_register_tests(TARGET ${PARENT_DIR}_test TESTS ${${PARENT_DIR}_tests}) diff --git a/src/irmd/reg/tests/flow_test.c b/src/irmd/reg/tests/flow_test.c index f9d23fd1..18214078 100644 --- a/src/irmd/reg/tests/flow_test.c +++ b/src/irmd/reg/tests/flow_test.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * The IPC Resource Manager - Registry - Flows - Unit Tests * @@ -22,13 +22,9 @@ #include "../flow.c" -#include <ouroboros/test.h> +#include <test/test.h> -#include <string.h> - -#define TEST_DATA "testpiggybackdata" - -static int test_reg_flow_create(void) +static int test_reg_flow_create_destroy(void) { struct reg_flow * f; @@ -51,10 +47,10 @@ static int test_reg_flow_create(void) TEST_SUCCESS(); - return 0; + return TEST_RC_SUCCESS; fail: TEST_FAIL(); - return -1; + return TEST_RC_FAIL; } static int test_reg_flow_create_no_id(void) { @@ -67,7 +63,7 @@ static int test_reg_flow_create_no_id(void) { reg_flow_create(&info); /* assert fail */ - return 0; + return TEST_RC_SUCCESS; } static int test_reg_flow_create_no_pid(void) { @@ -80,7 +76,7 @@ static int test_reg_flow_create_no_pid(void) { reg_flow_create(&info); /* assert fail */ - return 0; + return TEST_RC_SUCCESS; } static int test_reg_flow_create_has_n_1_pid(void) { @@ -94,7 +90,7 @@ static int test_reg_flow_create_has_n_1_pid(void) { reg_flow_create(&info); /* assert fail */ - return 0; + return TEST_RC_SUCCESS; } static int test_reg_flow_create_wrong_state(void) { @@ -108,7 +104,7 @@ static int test_reg_flow_create_wrong_state(void) { reg_flow_create(&info); /* assert fail */ - return 0; + return TEST_RC_SUCCESS; } static int test_reg_flow_create_has_mpl(void) { @@ -123,7 +119,22 @@ static int test_reg_flow_create_has_mpl(void) { reg_flow_create(&info); /* assert fail */ - return 0; + return TEST_RC_SUCCESS; +} + +static int test_reg_flow_create_has_mtu(void) { + struct flow_info info = { + .id = 1, + .n_pid = 1, + .n_1_pid = 0, + .mtu = 1400, + .qs = qos_raw, + .state = FLOW_ALLOC_PENDING + }; + + reg_flow_create(&info); /* assert fail */ + + return TEST_RC_SUCCESS; } static int test_reg_flow_update(void) @@ -140,7 +151,7 @@ static int test_reg_flow_update(void) struct flow_info upd = { .id = 1, .n_pid = 1, - .qs = qos_data, + .qs = qos_msg, .state = FLOW_DEALLOCATED }; @@ -163,10 +174,10 @@ static int test_reg_flow_update(void) TEST_SUCCESS(); - return 0; + return TEST_RC_SUCCESS; fail: TEST_FAIL(); - return -1; + return TEST_RC_FAIL; } static int test_reg_flow_update_wrong_id(void) @@ -183,7 +194,7 @@ static int test_reg_flow_update_wrong_id(void) struct flow_info upd = { .id = 2, .n_pid = 1, - .qs = qos_data, + .qs = qos_msg, .state = FLOW_DEALLOCATED }; @@ -199,10 +210,10 @@ static int test_reg_flow_update_wrong_id(void) TEST_SUCCESS(); - return 0; + return TEST_RC_SUCCESS; fail: TEST_FAIL(); - return -1; + return TEST_RC_FAIL; } static int test_reg_flow_assert_fails(void) @@ -210,70 +221,16 @@ static int test_reg_flow_assert_fails(void) int ret = 0; ret |= test_assert_fail(test_reg_flow_create_no_id); - ret |= test_assert_fail(test_reg_flow_create_no_pid); - ret |= test_assert_fail(test_reg_flow_create_has_n_1_pid); - ret |= test_assert_fail(test_reg_flow_create_wrong_state); - ret |= test_assert_fail(test_reg_flow_create_has_mpl); - + ret |= test_assert_fail(test_reg_flow_create_has_mtu); ret |= test_assert_fail(test_reg_flow_update_wrong_id); return ret; } -static int test_flow_data(void) -{ - struct reg_flow * f; - - struct flow_info info = { - .id = 1, - .n_pid = 1, - .qs = qos_raw, - .state = FLOW_INIT - }; - - char * data; - buffer_t buf; - buffer_t rcv = {NULL, 0}; - - TEST_START(); - - data = strdup(TEST_DATA); - if (data == NULL) { - printf("Failed to strdup data.\n"); - goto fail; - } - - buf.data = (uint8_t *) data; - buf.len = strlen(data); - - f = reg_flow_create(&info); - if (f == NULL) { - printf("Failed to create flow.\n"); - goto fail; - } - - reg_flow_set_data(f, &buf); - - reg_flow_get_data(f, &rcv); - - freebuf(buf); - clrbuf(rcv); - - reg_flow_destroy(f); - - TEST_SUCCESS(); - - return 0; - fail: - free(data); - TEST_FAIL(); - return -1; -} - int flow_test(int argc, char ** argv) { @@ -282,13 +239,9 @@ int flow_test(int argc, (void) argc; (void) argv; - ret |= test_reg_flow_create(); - + ret |= test_reg_flow_create_destroy(); ret |= test_reg_flow_update(); - ret |= test_reg_flow_assert_fails(); - ret |= test_flow_data(); - return ret; } diff --git a/src/irmd/reg/tests/ipcp_test.c b/src/irmd/reg/tests/ipcp_test.c index fb8ba71b..5279283e 100644 --- a/src/irmd/reg/tests/ipcp_test.c +++ b/src/irmd/reg/tests/ipcp_test.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * The IPC Resource Manager - Registry - IPCPs - Unit Tests * @@ -20,7 +20,7 @@ * Foundation, Inc., http://www.fsf.org/about/contact/. */ -#include <ouroboros/test.h> +#include <test/test.h> #include "../ipcp.c" @@ -31,7 +31,7 @@ static int test_reg_ipcp_create(void) struct reg_ipcp * ipcp; struct ipcp_info info = { .pid = TEST_PID, - .state = IPCP_BOOT + .state = IPCP_INIT }; struct layer_info layer = { .name = "testlayer", @@ -51,7 +51,7 @@ static int test_reg_ipcp_create(void) goto fail; } - ipcp->info.state = IPCP_OPERATIONAL; + ipcp->info.state = IPCP_BOOT; reg_ipcp_set_layer(ipcp, &layer); @@ -60,11 +60,6 @@ static int test_reg_ipcp_create(void) goto fail; } - if (ipcp->info.state != IPCP_OPERATIONAL) { - printf("IPCP state was not set.\n"); - goto fail; - } - reg_ipcp_destroy(ipcp); TEST_SUCCESS(); diff --git a/src/irmd/reg/tests/name_test.c b/src/irmd/reg/tests/name_test.c index 48f132e9..403c8a6c 100644 --- a/src/irmd/reg/tests/name_test.c +++ b/src/irmd/reg/tests/name_test.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * The IPC Resource Manager - Registry - Names - Unit Tests * @@ -20,8 +20,11 @@ * Foundation, Inc., http://www.fsf.org/about/contact/. */ + #include "../name.c" +#include <test/test.h> + #define TEST_PID 65534 #define TEST_PROG "/usr/bin/testprog" #define TEST_NAME "testservicename" @@ -34,6 +37,8 @@ static int test_reg_name_create(void) .pol_lb = LB_RR, }; + TEST_START(); + n = reg_name_create(&info); if (n == NULL) { printf("Failed to create name %s.\n", info.name); @@ -42,9 +47,12 @@ static int test_reg_name_create(void) reg_name_destroy(n); - return 0; + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; fail: - return -1; + TEST_FAIL(); + return TEST_RC_FAIL; } static int test_reg_name_add_proc(void) @@ -55,6 +63,8 @@ static int test_reg_name_add_proc(void) .pol_lb = LB_RR, }; + TEST_START(); + n = reg_name_create(&info); if (n == NULL) { printf("Failed to create name %s.\n", info.name); @@ -66,8 +76,8 @@ static int test_reg_name_add_proc(void) goto fail; } - if (n->n_procs != 1) { - printf("n_procs not updated.\n"); + if (n->procs.len != 1) { + printf("Proc not added to list.\n"); goto fail; } @@ -78,16 +88,19 @@ static int test_reg_name_add_proc(void) reg_name_del_proc(n, TEST_PID); - if (n->n_procs != 0) { - printf("n_procs not updated.\n"); + if (!llist_is_empty(&n->procs)) { + printf("Proc not removed from list.\n"); goto fail; } reg_name_destroy(n); - return 0; + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; fail: - return -1; + TEST_FAIL(); + return TEST_RC_FAIL; } static int test_reg_name_add_prog(void) @@ -100,6 +113,8 @@ static int test_reg_name_add_prog(void) char * exec[] = { TEST_PROG, "--argswitch", "argvalue", NULL}; + TEST_START(); + n = reg_name_create(&info); if (n == NULL) { printf("Failed to create name %s.\n", info.name); @@ -111,8 +126,8 @@ static int test_reg_name_add_prog(void) goto fail; } - if (n->n_progs != 1) { - printf("n_progs not updated.\n"); + if (n->progs.len != 1) { + printf("Prog not added to list.\n"); goto fail; } @@ -123,16 +138,19 @@ static int test_reg_name_add_prog(void) reg_name_del_prog(n, TEST_PROG); - if (n->n_progs != 0) { - printf("n_progs not updated.\n"); + if (!llist_is_empty(&n->progs)) { + printf("Prog not removed from list.\n"); goto fail; } reg_name_destroy(n); - return 0; + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; fail: - return -1; + TEST_FAIL(); + return TEST_RC_FAIL; } static int test_reg_name_add_active(enum pol_balance lb) @@ -144,6 +162,8 @@ static int test_reg_name_add_active(enum pol_balance lb) .pol_lb = lb, }; + TEST_START(); + n = reg_name_create(&info); if (n == NULL) { printf("Failed to create name %s.\n", info.name); @@ -175,8 +195,8 @@ static int test_reg_name_add_active(enum pol_balance lb) goto fail; } - if (n->n_active != 1) { - printf("n_active not updated.\n"); + if (n->active.len != 1) { + printf("Active list not updated.\n"); goto fail; } @@ -206,13 +226,13 @@ static int test_reg_name_add_active(enum pol_balance lb) goto fail; } - if (n->n_procs != 3) { - printf("n_procs not updated.\n"); + if (n->procs.len != 3) { + printf("Procs list not updated.\n"); goto fail; } - if (n->n_active != 4) { - printf("n_active not updated.\n"); + if (n->active.len != 4) { + printf("Active list not updated.\n"); goto fail; } @@ -243,41 +263,39 @@ static int test_reg_name_add_active(enum pol_balance lb) reg_name_del_proc(n, TEST_PID); - if (n->n_procs != 0) { - printf("n_procs not updated.\n"); + if (!llist_is_empty(&n->procs)) { + printf("Procs list not cleared.\n"); goto fail; } - if (n->n_active != 0) { - printf("n_active not updated.\n"); + if (!llist_is_empty(&n->active)) { + printf("Active list not cleared.\n"); goto fail; } reg_name_destroy(n); - return 0; + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; fail: - return -1; + TEST_FAIL(); + return TEST_RC_FAIL; } - int name_test(int argc, char ** argv) { - int res = 0; + int rc = 0; (void) argc; (void) argv; - res |= test_reg_name_create(); - - res |= test_reg_name_add_proc(); - - res |= test_reg_name_add_prog(); - - res |= test_reg_name_add_active(LB_RR); - - res |= test_reg_name_add_active(LB_SPILL); + rc |= test_reg_name_create(); + rc |= test_reg_name_add_proc(); + rc |= test_reg_name_add_prog(); + rc |= test_reg_name_add_active(LB_RR); + rc |= test_reg_name_add_active(LB_SPILL); - return res; + return rc; } diff --git a/src/irmd/reg/tests/proc_test.c b/src/irmd/reg/tests/proc_test.c index 5c9dd865..a85f4039 100644 --- a/src/irmd/reg/tests/proc_test.c +++ b/src/irmd/reg/tests/proc_test.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * The IPC Resource Manager - Registry - Processes - Unit Tests * @@ -22,16 +22,24 @@ #include "../proc.c" +#include <test/test.h> + #define TEST_PID 65534 #define TEST_PROG "usr/bin/testprog" -static int test_reg_proc_create(void) +#define TEST_PROC { \ + .pid = TEST_PID, \ + .prog = TEST_PROG, \ + .uid = getuid(), \ + .gid = getgid() \ +} + +static int test_reg_proc_create_destroy(void) { struct reg_proc * proc; - struct proc_info info = { - .pid = TEST_PID, - .prog = TEST_PROG - }; + struct proc_info info = TEST_PROC; + + TEST_START(); proc = reg_proc_create(&info); if (proc == NULL) { @@ -41,21 +49,23 @@ static int test_reg_proc_create(void) reg_proc_destroy(proc); - return 0; + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; fail: - return -1; + TEST_FAIL(); + return TEST_RC_FAIL; } static int test_reg_proc_add_name(void) { struct reg_proc * proc; - struct proc_info info = { - .pid = TEST_PID, - .prog = TEST_PROG - }; + struct proc_info info = TEST_PROC; char * name = "testname"; + TEST_START(); + proc = reg_proc_create(&info); if (proc == NULL) { printf("Failed to create proc.\n"); @@ -67,7 +77,7 @@ static int test_reg_proc_add_name(void) goto fail; } - if (proc->n_names != 1) { + if (proc->names.len != 1) { printf("n_names not updated.\n"); goto fail; } @@ -79,16 +89,19 @@ static int test_reg_proc_add_name(void) reg_proc_del_name(proc, name); - if (proc->n_names != 0) { + if (!llist_is_empty(&proc->names)) { printf("n_names not updated.\n"); goto fail; } reg_proc_destroy(proc); - return 0; + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; fail: - return -1; + TEST_FAIL(); + return TEST_RC_FAIL; } int proc_test(int argc, @@ -99,8 +112,7 @@ int proc_test(int argc, (void) argc; (void) argv; - res |= test_reg_proc_create(); - + res |= test_reg_proc_create_destroy(); res |= test_reg_proc_add_name(); return res; diff --git a/src/irmd/reg/tests/prog_test.c b/src/irmd/reg/tests/prog_test.c index 5e6931d8..91264aba 100644 --- a/src/irmd/reg/tests/prog_test.c +++ b/src/irmd/reg/tests/prog_test.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * The IPC Resource Manager - Registry - Programs - Unit Tests * @@ -22,8 +22,9 @@ #include "../prog.c" -#define TEST_PROG "usr/bin/testprog" +#include <test/test.h> +#define TEST_PROG "usr/bin/testprog" static int test_reg_prog_create(void) { @@ -32,6 +33,8 @@ static int test_reg_prog_create(void) .name = TEST_PROG }; + TEST_START(); + prog = reg_prog_create(&info); if (prog == NULL) { printf("Failed to create prog.\n"); @@ -40,9 +43,12 @@ static int test_reg_prog_create(void) reg_prog_destroy(prog); - return 0; + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; fail: - return -1; + TEST_FAIL(); + return TEST_RC_FAIL; } static int test_reg_prog_add_name(void) @@ -54,6 +60,8 @@ static int test_reg_prog_add_name(void) char * name = "testname"; + TEST_START(); + prog = reg_prog_create(&info); if (prog == NULL) { printf("Failed to create prog.\n"); @@ -65,7 +73,7 @@ static int test_reg_prog_add_name(void) goto fail; } - if (prog->n_names != 1) { + if (prog->names.len != 1) { printf("n_names not updated.\n"); goto fail; } @@ -77,16 +85,19 @@ static int test_reg_prog_add_name(void) reg_prog_del_name(prog, name); - if (prog->n_names != 0) { + if (!llist_is_empty(&prog->names)) { printf("n_names not updated.\n"); goto fail; } reg_prog_destroy(prog); - return 0; + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; fail: - return -1; + TEST_FAIL(); + return TEST_RC_FAIL; } int prog_test(int argc, @@ -98,7 +109,6 @@ int prog_test(int argc, (void) argv; ret |= test_reg_prog_create(); - ret |= test_reg_prog_add_name(); return ret; diff --git a/src/irmd/reg/tests/reg_test.c b/src/irmd/reg/tests/reg_test.c index c341c297..a8c1b1fa 100644 --- a/src/irmd/reg/tests/reg_test.c +++ b/src/irmd/reg/tests/reg_test.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * The IPC Resource Manager - Registry - Unit Tests * @@ -21,22 +21,31 @@ */ +#include "../pool.c" +#undef OUROBOROS_PREFIX #include "../reg.c" -#include <ouroboros/test.h> +#include <test/test.h> #define TEST_PID 3666 #define TEST_N_1_PID 3999 #define TEST_FAKE_ID 9128349 #define TEST_MPL 5 +#define TEST_MTU 1400 #define TEST_PROG "reg_test" /* own binary for binary check */ #define TEST_IPCP "testipcp" #define TEST_NAME "testname" #define TEST_DATA "testpbufdata" #define TEST_DATA2 "testpbufdata2" #define TEST_LAYER "testlayer" +#define TEST_PROC_INFO { \ + .pid = TEST_PID, \ + .prog = TEST_PROG, \ + .uid = 0, \ + .gid = 0 \ +} #define REG_TEST_FAIL() \ - do { TEST_FAIL(); memset(®, 0, sizeof(reg)); } while(0) + do { TEST_FAIL(); reg_clear(); return TEST_RC_FAIL;} while(0) static int test_reg_init(void) { @@ -51,10 +60,10 @@ static int test_reg_init(void) TEST_SUCCESS(); - return 0; + return TEST_RC_SUCCESS; fail: REG_TEST_FAIL(); - return -1; + return TEST_RC_FAIL; } static int test_reg_create_flow(void) @@ -81,7 +90,7 @@ static int test_reg_create_flow(void) goto fail; } - if (reg.n_flows != 1) { + if (reg.flows.len != 1) { printf("n_flows was not updated.\n"); goto fail; } @@ -96,8 +105,8 @@ static int test_reg_create_flow(void) goto fail; } - if (reg.n_flows != 0) { - printf("n_flows was not updated.\n"); + if (!llist_is_empty(®.flows)) { + printf("flows.len was not updated.\n"); goto fail; } @@ -105,18 +114,17 @@ static int test_reg_create_flow(void) TEST_SUCCESS(); - return 0; + return TEST_RC_SUCCESS; fail: REG_TEST_FAIL(); - return -1; + return TEST_RC_FAIL; } static int test_reg_allocate_flow_timeout(void) { struct timespec abstime; struct timespec timeo = TIMESPEC_INIT_MS(1); - buffer_t pbuf; - buffer_t rbuf = {NULL, 0}; + buffer_t rbuf = BUF_INIT; struct flow_info info = { .n_pid = TEST_PID, @@ -125,14 +133,6 @@ static int test_reg_allocate_flow_timeout(void) TEST_START(); - pbuf.data = (uint8_t *) strdup(TEST_DATA);; - if (pbuf.data == NULL) { - printf("Failed to strdup data.\n"); - goto fail; - } - - pbuf.len = strlen((char *) pbuf.data) + 1; - clock_gettime(PTHREAD_COND_CLOCK, &abstime); ts_add(&abstime, &timeo, &abstime); @@ -147,7 +147,7 @@ static int test_reg_allocate_flow_timeout(void) goto fail; } - if (reg_prepare_flow_accept(&info, &pbuf) < 0) { + if (reg_prepare_flow_accept(&info) < 0) { printf("Failed to prepare flow for accept.\n"); goto fail; } @@ -162,15 +162,9 @@ static int test_reg_allocate_flow_timeout(void) goto fail; } - if (pbuf.data == NULL) { - printf("Flow data was updated on timeout."); - goto fail; - } - - freebuf(pbuf); reg_destroy_flow(info.id); - if (reg.n_flows != 0) { + if (!llist_is_empty(®.flows)) { printf("Flow did not destroy.\n"); goto fail; } @@ -179,16 +173,19 @@ static int test_reg_allocate_flow_timeout(void) TEST_SUCCESS(); - return 0; + return TEST_RC_SUCCESS; fail: REG_TEST_FAIL(); - return -1; + return TEST_RC_FAIL; } static void * test_flow_respond_alloc(void * o) { struct flow_info * info = (struct flow_info *) o; - buffer_t pbuf = {NULL, 0}; + buffer_t pbuf = BUF_INIT; + int response; + + response = (info->state == FLOW_ALLOCATED) ? 0 : -1; if (info->state == FLOW_ALLOCATED) { pbuf.data = (uint8_t *) strdup(TEST_DATA2); @@ -199,7 +196,9 @@ static void * test_flow_respond_alloc(void * o) pbuf.len = strlen((char *) pbuf.data) + 1; } - reg_respond_alloc(info, &pbuf); + reg_respond_alloc(info, &pbuf, response); + + freebuf(pbuf); return (void *) 0; fail: @@ -220,12 +219,7 @@ static void * test_flow_respond_accept(void * o) reg_respond_accept(info, &pbuf); - if (info->qs.cypher_s == 0) { - freebuf(pbuf); - } else if (strcmp((char *) pbuf.data, TEST_DATA) != 0) { - printf("Data was not passed correctly.\n"); - goto fail; - } + freebuf(pbuf); return (void *) 0; fail: @@ -237,8 +231,7 @@ static int test_reg_accept_flow_success(void) pthread_t thr; struct timespec abstime; struct timespec timeo = TIMESPEC_INIT_S(1); - buffer_t pbuf = {(uint8_t *) TEST_DATA, strlen(TEST_DATA)}; - buffer_t rbuf = {NULL, 0}; + buffer_t rbuf = BUF_INIT; struct flow_info info = { .n_pid = TEST_PID, @@ -247,7 +240,7 @@ static int test_reg_accept_flow_success(void) struct flow_info n_1_info = { .n_1_pid = TEST_N_1_PID, - .qs = qos_data_crypt, + .qs = qos_msg, .state = FLOW_ALLOCATED /* RESPONSE SUCCESS */ }; @@ -267,18 +260,22 @@ static int test_reg_accept_flow_success(void) goto fail; } - if (reg_prepare_flow_accept(&info, &pbuf) < 0) { + if (reg_prepare_flow_accept(&info) < 0) { printf("Failed to prepare flow for accept.\n"); goto fail; } n_1_info.id = info.id; n_1_info.mpl = 1; + n_1_info.mtu = TEST_MTU; pthread_create(&thr, NULL, test_flow_respond_accept, &n_1_info); - if (reg_wait_flow_accepted(&info, &rbuf, &abstime) < 0 ) { + if (reg_wait_flow_accepted(&info, &rbuf, &abstime) < 0) { printf("Flow allocation failed.\n"); + pthread_join(thr, NULL); + reg_destroy_flow(info.id); + reg_fini(); goto fail; } @@ -289,6 +286,11 @@ static int test_reg_accept_flow_success(void) goto fail; } + if (info.mtu != TEST_MTU) { + printf("MTU not propagated.\n"); + goto fail; + } + if (rbuf.data == NULL) { printf("rbuf data not returned.\n"); goto fail; @@ -321,10 +323,10 @@ static int test_reg_accept_flow_success(void) TEST_SUCCESS(); - return 0; + return TEST_RC_SUCCESS; fail: REG_TEST_FAIL(); - return -1; + return TEST_RC_FAIL; } static int test_reg_accept_flow_success_no_crypt(void) @@ -332,8 +334,7 @@ static int test_reg_accept_flow_success_no_crypt(void) pthread_t thr; struct timespec abstime; struct timespec timeo = TIMESPEC_INIT_S(1); - buffer_t pbuf = {(uint8_t *) TEST_DATA, strlen(TEST_DATA)}; - buffer_t rbuf = {NULL, 0}; + buffer_t rbuf = BUF_INIT; struct flow_info info = { .n_pid = TEST_PID, @@ -342,7 +343,7 @@ static int test_reg_accept_flow_success_no_crypt(void) struct flow_info n_1_info = { .n_1_pid = TEST_N_1_PID, - .qs = qos_data, + .qs = qos_msg, .state = FLOW_ALLOCATED /* RESPONSE SUCCESS */ }; @@ -362,18 +363,22 @@ static int test_reg_accept_flow_success_no_crypt(void) goto fail; } - if (reg_prepare_flow_accept(&info, &pbuf) < 0) { + if (reg_prepare_flow_accept(&info) < 0) { printf("Failed to prepare flow for accept.\n"); goto fail; } n_1_info.id = info.id; n_1_info.mpl = 1; + n_1_info.mtu = TEST_MTU; pthread_create(&thr, NULL, test_flow_respond_accept, &n_1_info); if (reg_wait_flow_accepted(&info, &rbuf, &abstime) < 0 ) { printf("Flow allocation failed.\n"); + pthread_join(thr, NULL); + reg_destroy_flow(info.id); + reg_fini(); goto fail; } @@ -384,16 +389,18 @@ static int test_reg_accept_flow_success_no_crypt(void) goto fail; } - if (rbuf.data == NULL) { - printf("rbuf data was not returned.\n"); + if (info.mtu != TEST_MTU) { + printf("MTU not propagated.\n"); goto fail; } - if (strcmp((char *) rbuf.data, TEST_DATA) != 0) { - printf("Data was updated.\n"); + if (rbuf.data == NULL) { + printf("rbuf data was not returned.\n"); goto fail; } + freebuf(rbuf); + n_1_info.state = FLOW_DEALLOCATED; reg_dealloc_flow(&info); @@ -416,16 +423,16 @@ static int test_reg_accept_flow_success_no_crypt(void) TEST_SUCCESS(); - return 0; + return TEST_RC_SUCCESS; fail: REG_TEST_FAIL(); - return -1; + return TEST_RC_FAIL; } static int test_reg_allocate_flow_fail(void) { - buffer_t buf = {NULL, 0}; + buffer_t buf = BUF_INIT; pthread_t thr; struct timespec abstime; struct timespec timeo = TIMESPEC_INIT_S(1); @@ -437,7 +444,7 @@ static int test_reg_allocate_flow_fail(void) struct flow_info n_1_info = { .n_1_pid = TEST_N_1_PID, - .qs = qos_data, + .qs = qos_msg, .state = FLOW_DEALLOCATED /* RESPONSE FAIL */ }; @@ -470,6 +477,9 @@ static int test_reg_allocate_flow_fail(void) if (reg_wait_flow_allocated(&info, &buf, &abstime) == 0 ) { printf("Flow allocation succeeded.\n"); + pthread_join(thr, NULL); + reg_destroy_flow(info.id); + reg_fini(); goto fail; } @@ -486,26 +496,455 @@ static int test_reg_allocate_flow_fail(void) TEST_SUCCESS(); - return 0; + return TEST_RC_SUCCESS; fail: REG_TEST_FAIL(); - return -1; + return TEST_RC_FAIL; } -static int test_reg_flow(void) { - int ret = 0; +static int test_reg_respond_alloc_duplicate(void) +{ + pthread_t thr; + struct timespec abstime; + struct timespec timeo = TIMESPEC_INIT_S(1); + buffer_t rbuf = BUF_INIT; + buffer_t empty = BUF_INIT; + struct flow_info dup_info; + + struct flow_info info = { + .n_pid = TEST_PID, + .qs = qos_raw + }; + + struct flow_info n_1_info = { + .n_1_pid = TEST_N_1_PID, + .qs = qos_msg, + .state = FLOW_ALLOCATED /* RESPONSE SUCCESS */ + }; + + TEST_START(); + + clock_gettime(PTHREAD_COND_CLOCK, &abstime); + ts_add(&abstime, &timeo, &abstime); + + if (reg_init() < 0) { + printf("Failed to init registry.\n"); + goto fail; + } + + if (reg_create_flow(&info) < 0) { + printf("Failed to add flow.\n"); + goto fail; + } + + info.n_1_pid = TEST_N_1_PID; + + if (reg_prepare_flow_alloc(&info) < 0) { + printf("Failed to prepare flow for alloc.\n"); + goto fail; + } + + n_1_info.id = info.id; + n_1_info.mpl = 1; + n_1_info.mtu = TEST_MTU; + + pthread_create(&thr, NULL, test_flow_respond_alloc, &n_1_info); + + if (reg_wait_flow_allocated(&info, &rbuf, &abstime) < 0) { + printf("Flow allocation failed.\n"); + pthread_join(thr, NULL); + reg_destroy_flow(info.id); + reg_fini(); + goto fail; + } + + pthread_join(thr, NULL); + freebuf(rbuf); + + if (info.mtu != TEST_MTU) { + printf("MTU not propagated.\n"); + goto fail; + } + + /* Duplicate reply on an already-ALLOCATED flow must not assert. */ + dup_info = n_1_info; + dup_info.state = FLOW_DEALLOCATED; + + if (reg_respond_alloc(&dup_info, &empty, -EREPLAY) != -1) { + printf("Duplicate respond_alloc should return -1.\n"); + goto fail; + } + + reg_dealloc_flow(&info); + reg_dealloc_flow_resp(&info); + reg_destroy_flow(n_1_info.id); + + reg_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + REG_TEST_FAIL(); + return TEST_RC_FAIL; +} + +struct direct_alloc_info { + struct flow_info info; + buffer_t rsp; + struct timespec abstime; +}; + +static void * test_flow_alloc_direct(void * o) +{ + struct direct_alloc_info * dai; + buffer_t req; + + dai = (struct direct_alloc_info *) o; + + req.data = (uint8_t *) strdup(TEST_DATA); + if (req.data == NULL) { + printf("Failed to strdup req data.\n"); + goto fail; + } + req.len = strlen(TEST_DATA) + 1; + + if (reg_prepare_flow_direct(&dai->info, &req, 0) < 0) { + printf("Failed to prepare direct flow.\n"); + freebuf(req); + goto fail; + } + + if (reg_wait_flow_direct(dai->info.id, &dai->rsp, &dai->abstime) < 0) { + printf("Failed to wait direct flow.\n"); + goto fail; + } - ret |= test_reg_create_flow(); + return (void *) 0; + fail: + return (void *) -1; +} + +static int test_reg_direct_flow_success(void) +{ + pthread_t thr; + struct timespec abstime; + struct timespec timeo = TIMESPEC_INIT_S(1); + buffer_t rbuf = BUF_INIT; + buffer_t rsp; + struct direct_alloc_info dai; + void * thr_ret; + + struct flow_info info = { + .n_pid = TEST_PID, + .qs = qos_raw + }; + + TEST_START(); + + clock_gettime(PTHREAD_COND_CLOCK, &abstime); + + ts_add(&abstime, &timeo, &abstime); + + if (reg_init() < 0) { + printf("Failed to init registry.\n"); + goto fail; + } + + if (reg_create_flow(&info) < 0) { + printf("Failed to add flow.\n"); + goto fail; + } + + if (reg_prepare_flow_accept(&info) < 0) { + printf("Failed to prepare for accept.\n"); + goto fail; + } + + dai.info.id = info.id; + dai.info.n_1_pid = TEST_N_1_PID; + dai.info.mpl = TEST_MPL; + dai.info.qs = qos_msg; + dai.info.state = FLOW_ALLOCATED; + dai.rsp.len = 0; + dai.rsp.data = NULL; + dai.abstime = abstime; + + pthread_create(&thr, NULL, test_flow_alloc_direct, &dai); + + if (reg_wait_flow_accepted(&info, &rbuf, &abstime) < 0) { + printf("Flow accept failed.\n"); + pthread_join(thr, NULL); + reg_destroy_flow(info.id); + reg_fini(); + goto fail; + } + + if (info.state != FLOW_ALLOCATED) { + printf("Flow not in allocated state.\n"); + goto fail; + } + + if (rbuf.data == NULL) { + printf("req_data not received.\n"); + goto fail; + } + + if (strcmp((char *) rbuf.data, TEST_DATA) != 0) { + printf("req_data content mismatch.\n"); + goto fail; + } + + freebuf(rbuf); + + if (!reg_flow_is_direct(info.id)) { + printf("Flow not marked direct.\n"); + goto fail; + } + + rsp.data = (uint8_t *) strdup(TEST_DATA2); + if (rsp.data == NULL) { + printf("Failed to strdup rsp data.\n"); + goto fail; + } + rsp.len = strlen(TEST_DATA2) + 1; + + if (reg_respond_flow_direct(info.id, &rsp) < 0) { + printf("Failed to respond direct.\n"); + freebuf(rsp); + goto fail; + } + + pthread_join(thr, &thr_ret); + + if (thr_ret != (void *) 0) { + printf("Allocator thread failed.\n"); + goto fail; + } + + if (dai.rsp.data == NULL) { + printf("rsp_data not received.\n"); + goto fail; + } + + if (strcmp((char *) dai.rsp.data, TEST_DATA2) != 0) { + printf("rsp_data content mismatch.\n"); + goto fail; + } + + freebuf(dai.rsp); + + reg_dealloc_flow(&info); + + if (info.state != FLOW_DEALLOC_PENDING) { + printf("Flow not in dealloc pending.\n"); + goto fail; + } + + info.n_pid = TEST_PID; + + reg_dealloc_flow(&info); + + if (info.state != FLOW_DEALLOC_PENDING) { + printf("Same endpoint dealloc changed state.\n"); + goto fail; + } + + info.n_pid = TEST_N_1_PID; + + reg_dealloc_flow(&info); - ret |= test_reg_allocate_flow_timeout(); + if (info.state != FLOW_DEALLOCATED) { + printf("Flow not deallocated.\n"); + goto fail; + } - ret |= test_reg_accept_flow_success(); + reg_destroy_flow(info.id); - ret |= test_reg_accept_flow_success_no_crypt(); + reg_fini(); - ret |= test_reg_allocate_flow_fail(); + TEST_SUCCESS(); - return ret; + return TEST_RC_SUCCESS; + fail: + REG_TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* + * Direct-flow re-key: one shared seed is parked for both local apps. The + * per-app initiator role is resolved from the verified caller pid (the + * allocator is n_1_pid), and the seed is held until both have pulled it. + */ +static int test_reg_direct_flow_rekey(void) +{ + pthread_t thr; + struct timespec abstime; + struct timespec timeo = TIMESPEC_INIT_S(1); + buffer_t rbuf = BUF_INIT; + buffer_t rsp; + buffer_t no_crt = BUF_INIT; + struct direct_alloc_info dai; + uint8_t seed[SYMMKEYSZ]; + uint8_t out[SYMMKEYSZ]; + uint8_t epoch; + bool initiator; + size_t i; + + struct flow_info info = { + .n_pid = TEST_PID, + .qs = qos_raw + }; + + TEST_START(); + + for (i = 0; i < SYMMKEYSZ; ++i) + seed[i] = (uint8_t) i; + + clock_gettime(PTHREAD_COND_CLOCK, &abstime); + + ts_add(&abstime, &timeo, &abstime); + + if (reg_init() < 0) { + printf("Failed to init registry.\n"); + goto fail; + } + + if (reg_create_flow(&info) < 0) { + printf("Failed to add flow.\n"); + goto fail; + } + + if (reg_prepare_flow_accept(&info) < 0) { + printf("Failed to prepare for accept.\n"); + goto fail; + } + + dai.info.id = info.id; + dai.info.n_1_pid = TEST_N_1_PID; + dai.info.mpl = TEST_MPL; + dai.info.qs = qos_msg; + dai.info.state = FLOW_ALLOCATED; + dai.rsp.len = 0; + dai.rsp.data = NULL; + dai.abstime = abstime; + + pthread_create(&thr, NULL, test_flow_alloc_direct, &dai); + + if (reg_wait_flow_accepted(&info, &rbuf, &abstime) < 0) { + printf("Flow accept failed.\n"); + pthread_join(thr, NULL); + goto fail; + } + + freebuf(rbuf); + + rsp.data = (uint8_t *) strdup(TEST_DATA2); + if (rsp.data == NULL) { + printf("Failed to strdup rsp data.\n"); + pthread_join(thr, NULL); + goto fail; + } + rsp.len = strlen(TEST_DATA2) + 1; + + if (reg_respond_flow_direct(info.id, &rsp) < 0) { + printf("Failed to respond direct.\n"); + freebuf(rsp); + pthread_join(thr, NULL); + goto fail; + } + + pthread_join(thr, NULL); + + freebuf(dai.rsp); + + if (!reg_flow_is_direct(info.id)) { + printf("Flow not marked direct.\n"); + goto fail; + } + + reg_flow_set_rekey(info.id, false, no_crt); + + if (reg_flow_store_pending_direct(info.id, seed, 5) < 0) { + printf("Failed to store pending direct seed.\n"); + goto fail; + } + + if (!reg_flow_rekey_pending(info.id)) { + printf("Seed not pending after store.\n"); + goto fail; + } + + /* Allocator (n_1_pid) pulls: initiator role, seed still held. */ + if (reg_flow_take_pending(info.id, 0, TEST_N_1_PID, out, + &epoch, &initiator) != 0) { + printf("Allocator failed to take pending seed.\n"); + goto fail; + } + + if (!initiator || epoch != 5 || memcmp(out, seed, SYMMKEYSZ) != 0) { + printf("Allocator got wrong seed/role/epoch.\n"); + goto fail; + } + + if (!reg_flow_rekey_pending(info.id)) { + printf("Seed cleared before both apps pulled.\n"); + goto fail; + } + + /* Acceptor (n_pid) pulls: responder role, seed now released. */ + if (reg_flow_take_pending(info.id, 0, TEST_PID, out, + &epoch, &initiator) != 0) { + printf("Acceptor failed to take pending seed.\n"); + goto fail; + } + + if (initiator || epoch != 5 || memcmp(out, seed, SYMMKEYSZ) != 0) { + printf("Acceptor got wrong seed/role/epoch.\n"); + goto fail; + } + + if (reg_flow_rekey_pending(info.id)) { + printf("Seed still pending after both pulled.\n"); + goto fail; + } + + if (reg_flow_get_epoch(info.id) != 5) { + printf("Flow epoch not advanced.\n"); + goto fail; + } + + info.n_pid = TEST_PID; + reg_dealloc_flow(&info); + + info.n_pid = TEST_N_1_PID; + reg_dealloc_flow(&info); + + reg_destroy_flow(info.id); + + reg_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + REG_TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_reg_flow(void) { + int rc = 0; + + rc |= test_reg_create_flow(); + rc |= test_reg_allocate_flow_timeout(); + rc |= test_reg_accept_flow_success(); + rc |= test_reg_accept_flow_success_no_crypt(); + rc |= test_reg_allocate_flow_fail(); + rc |= test_reg_respond_alloc_duplicate(); + rc |= test_reg_direct_flow_success(); + rc |= test_reg_direct_flow_rekey(); + + return rc; } static int test_reg_create_ipcp(void) @@ -513,7 +952,7 @@ static int test_reg_create_ipcp(void) struct ipcp_info info = { .name = TEST_IPCP, .pid = TEST_PID, - .state = IPCP_BOOT /* set by spawn_ipcp */ + .state = IPCP_INIT /* set by spawn_ipcp */ }; TEST_START(); @@ -528,7 +967,7 @@ static int test_reg_create_ipcp(void) goto fail; } - if (reg.n_ipcps != 1) { + if (reg.ipcps.len != 1) { printf("n_ipcps was not updated.\n"); goto fail; } @@ -543,19 +982,141 @@ static int test_reg_create_ipcp(void) goto fail; } - if (reg.n_ipcps != 0) { - printf("n_ipcps was not updated.\n"); + if (reg.ipcps.len != 0) { + printf("ipcps.len was not updated.\n"); + goto fail; + } + + reg_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + REG_TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_reg_list_ipcps(void) +{ + ipcp_list_msg_t ** ipcps; + int i; + ssize_t len; + + TEST_START(); + + if (reg_init() < 0) { + printf("Failed to init registry.\n"); + goto fail; + } + + for (i = 0; i < 10; i++) { + struct ipcp_info info = { + .pid = TEST_PID + i, + .state = IPCP_INIT /* set by spawn_ipcp */ + }; + + sprintf(info.name, "%s%d", TEST_IPCP, i); + + if (reg_create_ipcp(&info) < 0) { + printf("Failed to create ipcp %d.\n", i); + goto fail; + } + } + + len = reg_list_ipcps(&ipcps); + if (len < 0) { + printf("Failed to list ipcps.\n"); goto fail; } + if (len != 10) { + printf("Failed to list all ipcps.\n"); + goto fail; + } + + while (len-- > 0) + ipcp_list_msg__free_unpacked(ipcps[len], NULL); + + free(ipcps); + + for (i = 0; i < 10; i++) + reg_destroy_proc(TEST_PID + i); + reg_fini(); TEST_SUCCESS(); - return 0; + return TEST_RC_SUCCESS; + fail: REG_TEST_FAIL(); - return -1; + return TEST_RC_FAIL; +} + +static int test_insert_ipcps(void) +{ + ipcp_list_msg_t ** ipcps; + struct ipcp_info info; + size_t i; + size_t len; + + TEST_START(); + + if (reg_init() < 0) { + printf("Failed to init registry.\n"); + goto fail; + } + + for (i = 0; i < 100; i++) { + sprintf(info.name, "%s-%zd", TEST_IPCP, i); + info.pid = TEST_PID + rand() % 10000; + info.type = rand() % IPCP_INVALID; + info.state = IPCP_INIT; /* set by spawn_ipcp */ + + if (reg_create_ipcp(&info) < 0) { + printf("Failed to create ipcp %s.\n", info.name); + goto fail; + } + } + + len = reg_list_ipcps(&ipcps); + if (len != 100) { + printf("Failed to list all ipcps.\n"); + goto fail; + } + + for (i = 1; i < len; i++) { + if (ipcps[i]->type < ipcps[i - 1]->type) { + printf("IPCPS not sorted by type.\n"); + goto fail; + } + + if (ipcps[i]->type != ipcps[i - 1]->type) + continue; + + /* allow occasional duplicate PID in test */ + if (ipcps[i]->pid < ipcps[i - 1]->pid) { + printf("IPCPS not sorted by pid.\n"); + goto fail; + } + } + + while (len-- > 0) + ipcp_list_msg__free_unpacked(ipcps[len], NULL); + + free(ipcps); + + reg_clear(); + + reg_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; +fail: + REG_TEST_FAIL(); + return TEST_RC_FAIL; } static int test_set_layer(void) @@ -564,7 +1125,7 @@ static int test_set_layer(void) struct ipcp_info info = { .name = TEST_IPCP, .pid = TEST_PID, - .state = IPCP_BOOT /* set by spawn_ipcp */ + .state = IPCP_INIT /* set by spawn_ipcp */ }; struct layer_info layer = { .name = TEST_LAYER, @@ -588,8 +1149,9 @@ static int test_set_layer(void) } ipcp = __reg_get_ipcp(info.pid); - ipcp->info.state = IPCP_OPERATIONAL; - info.state = IPCP_ENROLLED; + + ipcp->info.state = IPCP_BOOT; + info.state = IPCP_BOOT; reg_set_layer_for_ipcp(&info, &layer); @@ -614,21 +1176,22 @@ static int test_set_layer(void) TEST_SUCCESS(); - return 0; + return TEST_RC_SUCCESS; fail: REG_TEST_FAIL(); - return -1; + return TEST_RC_FAIL; } static int test_reg_ipcp(void) { - int ret = 0; - - ret |= test_reg_create_ipcp(); + int rc = 0; - ret |= test_set_layer(); + rc |= test_reg_create_ipcp(); + rc |= test_reg_list_ipcps(); + rc |= test_insert_ipcps(); + rc |= test_set_layer(); - return ret; + return rc; } static int test_reg_create_name(void) @@ -650,7 +1213,7 @@ static int test_reg_create_name(void) goto fail; } - if (reg.n_names != 1) { + if (reg.names.len != 1) { printf("n_names was not updated.\n"); goto fail; } @@ -665,7 +1228,7 @@ static int test_reg_create_name(void) goto fail; } - if (reg.n_names != 0) { + if (!llist_is_empty(®.names)) { printf("n_names was not updated.\n"); goto fail; } @@ -674,27 +1237,83 @@ static int test_reg_create_name(void) TEST_SUCCESS(); - return 0; + return TEST_RC_SUCCESS; fail: REG_TEST_FAIL(); - return -1; + return TEST_RC_FAIL; +} + +static int test_reg_list_names(void) +{ + name_info_msg_t ** names; + int i; + ssize_t len; + + TEST_START(); + + if (reg_init() < 0) { + printf("Failed to init registry.\n"); + goto fail; + } + + for (i = 0; i < 10; i++) { + struct name_info info = { + .pol_lb = LB_RR + }; + + sprintf(info.name, "%s%d", TEST_NAME, i); + + if (reg_create_name(&info) < 0) { + printf("Failed to create name %d.\n", i); + goto fail; + } + } + + len = reg_list_names(&names); + if (len < 0) { + printf("Failed to list names.\n"); + goto fail; + } + + if (len != 10) { + printf("Failed to list all names.\n"); + goto fail; + } + + for (i = 0; i < len; i++) + name_info_msg__free_unpacked(names[i], NULL); + + free(names); + + for (i = 0; i < 10; i++) { + char name[NAME_MAX]; + sprintf(name, "%s%d", TEST_NAME, i); + reg_destroy_name(name); + } + + reg_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + REG_TEST_FAIL(); + return TEST_RC_FAIL; } static int test_reg_name(void) { - int ret = 0; + int rc = 0; - ret |= test_reg_create_name(); + rc |= test_reg_create_name(); + rc |= test_reg_list_names(); - return ret; + return rc; } static int test_reg_create_proc(void) { - struct proc_info info = { - .pid = TEST_PID, - .prog = TEST_PROG - }; + struct proc_info info = TEST_PROC_INFO; TEST_START(); @@ -708,7 +1327,7 @@ static int test_reg_create_proc(void) goto fail; } - if (reg.n_procs != 1) { + if (reg.procs.len != 1) { printf("n_procs was not updated.\n"); goto fail; } @@ -723,7 +1342,7 @@ static int test_reg_create_proc(void) goto fail; } - if (reg.n_procs != 0) { + if (!llist_is_empty(®.procs)) { printf("n_procs was not updated.\n"); goto fail; } @@ -732,19 +1351,19 @@ static int test_reg_create_proc(void) TEST_SUCCESS(); - return 0; + return TEST_RC_SUCCESS; fail: REG_TEST_FAIL(); - return -1; + return TEST_RC_FAIL; } static int test_reg_proc(void) { - int ret = 0; + int rc = 0; - ret |= test_reg_create_proc(); + rc |= test_reg_create_proc(); - return ret; + return rc; } static int test_reg_spawned(void) @@ -761,7 +1380,7 @@ static int test_reg_spawned(void) goto fail; } - if (reg.n_spawned != 1) { + if (reg.spawned.len != 1) { printf("n_spawned was not updated.\n"); goto fail; } @@ -776,7 +1395,7 @@ static int test_reg_spawned(void) goto fail; } - if (reg.n_spawned != 0) { + if (!llist_is_empty(®.spawned)) { printf("n_spawned was not updated.\n"); goto fail; } @@ -785,10 +1404,10 @@ static int test_reg_spawned(void) TEST_SUCCESS(); - return 0; + return TEST_RC_SUCCESS; fail: REG_TEST_FAIL(); - return -1; + return TEST_RC_FAIL; } static int test_reg_create_prog(void) @@ -809,7 +1428,7 @@ static int test_reg_create_prog(void) goto fail; } - if (reg.n_progs != 1) { + if (reg.progs.len != 1) { printf("n_progs was not updated.\n"); goto fail; } @@ -824,7 +1443,7 @@ static int test_reg_create_prog(void) goto fail; } - if (reg.n_progs != 0) { + if (!llist_is_empty(®.progs)) { printf("n_progs was not updated.\n"); goto fail; } @@ -833,27 +1452,24 @@ static int test_reg_create_prog(void) TEST_SUCCESS(); - return 0; + return TEST_RC_SUCCESS; fail: REG_TEST_FAIL(); - return -1; + return TEST_RC_FAIL; } static int test_reg_prog(void) { - int ret = 0; + int rc = 0; - ret |= test_reg_create_prog(); + rc |= test_reg_create_prog(); - return ret; + return rc; } static int test_bind_proc(void) { - struct proc_info pinfo = { - .pid = TEST_PID, - .prog = TEST_PROG - }; + struct proc_info pinfo = TEST_PROC_INFO; struct name_info ninfo = { .name = TEST_NAME, @@ -900,10 +1516,10 @@ static int test_bind_proc(void) TEST_SUCCESS(); - return 0; + return TEST_RC_SUCCESS; fail: REG_TEST_FAIL(); - return -1; + return TEST_RC_FAIL; } static int test_bind_prog(void) @@ -989,10 +1605,10 @@ static int test_bind_prog(void) TEST_SUCCESS(); - return 0; + return TEST_RC_SUCCESS; fail: REG_TEST_FAIL(); - return -1; + return TEST_RC_FAIL; } static int test_inherit_prog(void) @@ -1006,10 +1622,7 @@ static int test_inherit_prog(void) .name = TEST_PROG }; - struct proc_info procinfo = { - .pid = TEST_PID, - .prog = TEST_PROG - }; + struct proc_info procinfo = TEST_PROC_INFO; char * exec[] = { TEST_PROG, NULL}; @@ -1060,10 +1673,10 @@ static int test_inherit_prog(void) TEST_SUCCESS(); - return 0; + return TEST_RC_SUCCESS; fail: REG_TEST_FAIL(); - return -1; + return TEST_RC_FAIL; } static int test_wait_accepting_timeout(void) @@ -1071,7 +1684,6 @@ static int test_wait_accepting_timeout(void) struct timespec abstime; struct timespec timeo = TIMESPEC_INIT_MS(1); int flow_id; - uint8_t hash[64]; struct name_info ninfo = { .name = TEST_NAME, .pol_lb = LB_RR @@ -1089,12 +1701,10 @@ static int test_wait_accepting_timeout(void) goto fail; } - str_hash(HASH_SHA3_256, hash, ninfo.name); - clock_gettime(PTHREAD_COND_CLOCK, &abstime); ts_add(&abstime, &timeo, &abstime); - flow_id = reg_wait_flow_accepting(HASH_SHA3_256, hash, &abstime); + flow_id = reg_wait_flow_accepting(ninfo.name, &abstime); if (flow_id != -ETIMEDOUT) { printf("Wait accept did not time out: %d.\n", flow_id); goto fail; @@ -1106,10 +1716,10 @@ static int test_wait_accepting_timeout(void) TEST_SUCCESS(); - return 0; + return TEST_RC_SUCCESS; fail: REG_TEST_FAIL(); - return -1; + return TEST_RC_FAIL; } static int test_wait_accepting_fail_name(void) @@ -1117,7 +1727,6 @@ static int test_wait_accepting_fail_name(void) struct timespec abstime; struct timespec timeo = TIMESPEC_INIT_S(1); int flow_id; - uint8_t hash[64]; TEST_START(); @@ -1128,11 +1737,10 @@ static int test_wait_accepting_fail_name(void) clock_gettime(PTHREAD_COND_CLOCK, &abstime); ts_add(&abstime, &timeo, &abstime); - str_hash(HASH_SHA3_256, hash, "C0FF33"); - flow_id = reg_wait_flow_accepting(HASH_SHA3_256, hash, &abstime); + flow_id = reg_wait_flow_accepting(TEST_NAME, &abstime); if (flow_id != -ENAME) { - printf("Wait accept did not fail on name: %d.\n", flow_id); + printf("Wait accept did not fail: %d.\n", flow_id); goto fail; } @@ -1140,22 +1748,19 @@ static int test_wait_accepting_fail_name(void) TEST_SUCCESS(); - return 0; + return TEST_RC_SUCCESS; fail: REG_TEST_FAIL(); - return -1; + return TEST_RC_FAIL; } static void * test_call_flow_accept(void * o) { struct timespec abstime; - struct timespec timeo = TIMESPEC_INIT_MS(1); - buffer_t pbuf = {NULL, 0}; + struct timespec timeo = TIMESPEC_INIT_MS(30); + buffer_t pbuf = BUF_INIT; - struct proc_info pinfo = { - .pid = TEST_PID, - .prog = TEST_PROG - }; + struct proc_info pinfo = TEST_PROC_INFO; struct flow_info info = { .n_pid = pinfo.pid, @@ -1179,16 +1784,21 @@ static void * test_call_flow_accept(void * o) info.state = FLOW_ACCEPT_PENDING; + reg_prepare_flow_accept(&info); + clock_gettime(PTHREAD_COND_CLOCK, &abstime); ts_add(&abstime, &timeo, &abstime); - reg_prepare_flow_accept(&info, &pbuf); - if (reg_wait_flow_accepted(&info, &pbuf, &abstime) != -ETIMEDOUT) { printf("Wait allocated did not timeout.\n"); goto fail; } + if (reg_unbind_proc((char *) o, pinfo.pid) < 0) { + printf("Failed to unbind proc.\n"); + goto fail; + } + reg_destroy_flow(info.id); reg_destroy_proc(pinfo.pid); @@ -1200,15 +1810,15 @@ static void * test_call_flow_accept(void * o) static int test_wait_accepting_success(void) { struct timespec abstime; - struct timespec timeo = TIMESPEC_INIT_S(1); - int flow_id; + struct timespec timeo = TIMESPEC_INIT_S(10); pthread_t thr; - uint8_t hash[64]; + int flow_id; struct name_info ninfo = { .name = TEST_NAME, .pol_lb = LB_RR }; + TEST_START(); if (reg_init()) { @@ -1226,11 +1836,12 @@ static int test_wait_accepting_success(void) clock_gettime(PTHREAD_COND_CLOCK, &abstime); ts_add(&abstime, &timeo, &abstime); - str_hash(HASH_SHA3_256, hash, ninfo.name); - - flow_id = reg_wait_flow_accepting(HASH_SHA3_256, hash, &abstime); + flow_id = reg_wait_flow_accepting(ninfo.name, &abstime); if (flow_id < 0) { - printf("Wait accept did not return a flow id: %d.", flow_id); + printf("Wait accept did not return a flow id: %d.\n", flow_id); + pthread_join(thr, NULL); + reg_destroy_name(TEST_NAME); + reg_fini(); goto fail; } @@ -1242,23 +1853,21 @@ static int test_wait_accepting_success(void) TEST_SUCCESS(); - return 0; + return TEST_RC_SUCCESS; fail: REG_TEST_FAIL(); - return -1; + return TEST_RC_FAIL; } static int test_wait_accepting(void) { - int ret = 0; - - ret |= test_wait_accepting_timeout(); + int rc = 0; - ret |= test_wait_accepting_fail_name(); + rc |= test_wait_accepting_timeout(); + rc |= test_wait_accepting_fail_name(); + rc |= test_wait_accepting_success(); - ret |= test_wait_accepting_success(); - - return ret; + return rc; } static int test_wait_ipcp_boot_timeout(void) @@ -1268,7 +1877,7 @@ static int test_wait_ipcp_boot_timeout(void) struct ipcp_info info = { .name = TEST_IPCP, .pid = TEST_PID, - .state = IPCP_BOOT /* set by spawn_ipcp */ + .state = IPCP_INIT /* set by spawn_ipcp */ }; TEST_START(); @@ -1300,10 +1909,10 @@ static int test_wait_ipcp_boot_timeout(void) TEST_SUCCESS(); - return 0; + return TEST_RC_SUCCESS; fail: REG_TEST_FAIL(); - return -1; + return TEST_RC_FAIL; } static void * test_ipcp_respond(void * o) @@ -1318,12 +1927,12 @@ static void * test_ipcp_respond(void * o) static int test_wait_ipcp_boot_fail(void) { struct timespec abstime; - struct timespec timeo = TIMESPEC_INIT_S(1); + struct timespec timeo = TIMESPEC_INIT_S(10); pthread_t thr; struct ipcp_info info = { .name = TEST_IPCP, .pid = TEST_PID, - .state = IPCP_BOOT /* set by spawn_ipcp */ + .state = IPCP_INIT /* set by spawn_ipcp */ }; struct ipcp_info resp_info = { .name = TEST_IPCP, @@ -1348,10 +1957,13 @@ static int test_wait_ipcp_boot_fail(void) clock_gettime(PTHREAD_COND_CLOCK, &abstime); ts_add(&abstime, &timeo, &abstime); - info.state = IPCP_BOOT; + info.state = IPCP_INIT; if (reg_wait_ipcp_boot(&info, &abstime) == 0) { printf("IPCP boot reported success.\n"); + pthread_join(thr, NULL); + reg_destroy_proc(info.pid); + reg_fini(); goto fail; } @@ -1362,8 +1974,8 @@ static int test_wait_ipcp_boot_fail(void) goto fail; } - if (reg.n_ipcps != 0) { - printf("n_ipcps was not updated.\n"); + if (!llist_is_empty(®.ipcps)) { + printf("ipcps.len was not updated.\n"); goto fail; } @@ -1371,21 +1983,21 @@ static int test_wait_ipcp_boot_fail(void) TEST_SUCCESS(); - return 0; + return TEST_RC_SUCCESS; fail: REG_TEST_FAIL(); - return -1; + return TEST_RC_FAIL; } static int test_wait_ipcp_boot_success(void) { pthread_t thr; struct timespec abstime; - struct timespec timeo = TIMESPEC_INIT_S(1); + struct timespec timeo = TIMESPEC_INIT_S(10); struct ipcp_info info = { .name = TEST_IPCP, .pid = TEST_PID, - .state = IPCP_BOOT /* set by spawn_ipcp */ + .state = IPCP_INIT /* set by spawn_ipcp */ }; struct ipcp_info resp_info = { .name = TEST_IPCP, @@ -1410,10 +2022,13 @@ static int test_wait_ipcp_boot_success(void) clock_gettime(PTHREAD_COND_CLOCK, &abstime); ts_add(&abstime, &timeo, &abstime); - info.state = IPCP_BOOT; + info.state = IPCP_INIT; if (reg_wait_ipcp_boot(&info, &abstime) < 0) { printf("IPCP boot failed.\n"); + pthread_join(thr, NULL); + reg_destroy_proc(info.pid); + reg_fini(); goto fail; } @@ -1421,6 +2036,8 @@ static int test_wait_ipcp_boot_success(void) if (info.state != IPCP_OPERATIONAL) { printf("IPCP boot succeeded in non-operational state.\n"); + reg_destroy_proc(info.pid); + reg_fini(); goto fail; } @@ -1433,23 +2050,21 @@ static int test_wait_ipcp_boot_success(void) TEST_SUCCESS(); - return 0; + return TEST_RC_SUCCESS; fail: REG_TEST_FAIL(); - return -1; + return TEST_RC_FAIL; } static int test_wait_ipcp_boot(void) { - int ret = 0; - - ret |= test_wait_ipcp_boot_timeout(); + int rc = 0; - ret |= test_wait_ipcp_boot_fail(); + rc |= test_wait_ipcp_boot_timeout(); + rc |= test_wait_ipcp_boot_fail(); + rc |= test_wait_ipcp_boot_success(); - ret |= test_wait_ipcp_boot_success(); - - return ret; + return rc; } static int test_wait_proc_timeout(void) @@ -1477,10 +2092,10 @@ static int test_wait_proc_timeout(void) TEST_SUCCESS(); - return 0; + return TEST_RC_SUCCESS; fail: REG_TEST_FAIL(); - return -1; + return TEST_RC_FAIL; } static void * test_proc(void * o) @@ -1495,12 +2110,9 @@ static void * test_proc(void * o) static int test_wait_proc_success(void) { struct timespec abstime; - struct timespec timeo = TIMESPEC_INIT_S(1); + struct timespec timeo = TIMESPEC_INIT_S(10); pthread_t thr; - struct proc_info info = { - .pid = TEST_PID, - .prog = TEST_PROG - }; + struct proc_info info = TEST_PROC_INFO; TEST_START(); @@ -1516,6 +2128,9 @@ static int test_wait_proc_success(void) if (reg_wait_proc(info.pid, &abstime) < 0) { printf("Waiting for proc failed.\n"); + pthread_join(thr, NULL); + reg_destroy_proc(info.pid); + reg_fini(); goto fail; } @@ -1527,57 +2142,43 @@ static int test_wait_proc_success(void) TEST_SUCCESS(); - return 0; + return TEST_RC_SUCCESS; fail: REG_TEST_FAIL(); - return -1; + return TEST_RC_FAIL; } static int test_wait_proc(void) { - int ret = 0; - - ret |= test_wait_proc_timeout(); + int rc = 0; - ret |= test_wait_proc_success(); + rc |= test_wait_proc_timeout(); + rc |= test_wait_proc_success(); - return ret; + return rc; } - int reg_test(int argc, char ** argv) { - int ret = 0; + int rc = 0; (void) argc; (void) argv; - ret |= test_reg_init(); - - ret |= test_reg_flow(); - - ret |= test_reg_ipcp(); - - ret |= test_reg_name(); - - ret |= test_reg_proc(); - - ret |= test_reg_prog(); - - ret |= test_reg_spawned(); - - ret |= test_bind_proc(); - - ret |= test_bind_prog(); - - ret |= test_inherit_prog(); - - ret |= test_wait_accepting(); - - ret |= test_wait_ipcp_boot(); - - ret |= test_wait_proc(); - - return ret; + rc |= test_reg_init(); + rc |= test_reg_flow(); + rc |= test_reg_ipcp(); + rc |= test_reg_name(); + rc |= test_reg_proc(); + rc |= test_reg_prog(); + rc |= test_reg_spawned(); + rc |= test_bind_proc(); + rc |= test_bind_prog(); + rc |= test_inherit_prog(); + rc |= test_wait_accepting(); + rc |= test_wait_ipcp_boot(); + rc |= test_wait_proc(); + + return rc; } diff --git a/src/irmd/tests/CMakeLists.txt b/src/irmd/tests/CMakeLists.txt deleted file mode 100644 index e005d194..00000000 --- a/src/irmd/tests/CMakeLists.txt +++ /dev/null @@ -1,19 +0,0 @@ -get_filename_component(tmp ".." ABSOLUTE) -get_filename_component(src_folder "${tmp}" NAME) - -create_test_sourcelist(${src_folder}_tests test_suite.c - # Add new tests here -) - -add_executable(${src_folder}_test EXCLUDE_FROM_ALL ${${src_folder}_tests}) -target_link_libraries(${src_folder}_test ouroboros-common) - -add_dependencies(check ${src_folder}_test) - -set(tests_to_run ${${src_folder}_tests}) -remove(tests_to_run test_suite.c) - -foreach(test ${tests_to_run}) - get_filename_component(test_name ${test} NAME_WE) - add_test(irmd/${test_name} ${C_TEST_PATH}/${src_folder}_test ${test_name}) -endforeach(test) diff --git a/src/lib/CMakeLists.txt b/src/lib/CMakeLists.txt index a6d7ac98..3abf39d0 100644 --- a/src/lib/CMakeLists.txt +++ b/src/lib/CMakeLists.txt @@ -1,261 +1,29 @@ -include_directories(${CMAKE_CURRENT_SOURCE_DIR}) -include_directories(${CMAKE_CURRENT_BINARY_DIR}) - -include_directories(${CMAKE_SOURCE_DIR}/include) -include_directories(${CMAKE_BINARY_DIR}/include) +# Ouroboros libraries build configuration +# Configuration options are in cmake/config/lib.cmake protobuf_generate_c(MODEL_PROTO_SRCS MODEL_PROTO_HDRS - pb/model.proto) + "${CMAKE_CURRENT_SOURCE_DIR}/pb/model.proto") protobuf_generate_c(IPCP_CONFIG_PROTO_SRCS IPCP_CONFIG_PROTO_HDRS - pb/ipcp_config.proto) + "${CMAKE_CURRENT_SOURCE_DIR}/pb/ipcp_config.proto") protobuf_generate_c(ENROLL_PROTO_SRCS ENROLL_PROTO_HDRS - pb/enroll.proto) + "${CMAKE_CURRENT_SOURCE_DIR}/pb/enroll.proto") protobuf_generate_c(CEP_PROTO_SRCS CEP_PROTO_HDRS - pb/cep.proto) + "${CMAKE_CURRENT_SOURCE_DIR}/pb/cep.proto") protobuf_generate_c(IRM_PROTO_SRCS IRM_PROTO_HDRS - pb/irm.proto) + "${CMAKE_CURRENT_SOURCE_DIR}/pb/irm.proto") protobuf_generate_c(IPCP_PROTO_SRCS IPCP_PROTO_HDRS - pb/ipcp.proto) - -if (NOT APPLE) - find_library(LIBRT_LIBRARIES rt) - if (NOT LIBRT_LIBRARIES) - message(FATAL_ERROR "Could not find librt") - endif () -else () - set(LIBRT_LIBRARIES "") -endif () - -find_library(LIBPTHREAD_LIBRARIES pthread) -if (NOT LIBPTHREAD_LIBRARIES) - message(FATAL_ERROR "Could not find libpthread") -endif () - -include(CheckSymbolExists) -list(APPEND CMAKE_REQUIRED_DEFINITIONS -D_POSIX_C_SOURCE=200809L) -list(APPEND CMAKE_REQUIRED_DEFINITIONS -D__XSI_VISIBLE=500) -list(APPEND CMAKE_REQUIRED_LIBRARIES pthread) -check_symbol_exists(pthread_mutexattr_setrobust pthread.h HAVE_ROBUST_MUTEX) - -if (HAVE_ROBUST_MUTEX) - set(DISABLE_ROBUST_MUTEXES FALSE CACHE BOOL "Disable robust mutex support") - if (NOT DISABLE_ROBUST_MUTEXES) - message(STATUS "Robust mutex support enabled") - set(HAVE_ROBUST_MUTEX TRUE) - else () - message(STATUS "Robust mutex support disabled by user") - unset(HAVE_ROBUST_MUTEX) - endif () -endif () - -find_library(FUSE_LIBRARIES fuse QUIET) -if (FUSE_LIBRARIES) - #FIXME: Check for version >= 2.6 - set(DISABLE_FUSE FALSE CACHE BOOL "Disable FUSE support") - if (NOT DISABLE_FUSE) - message(STATUS "FUSE support enabled") - set(FUSE_PREFIX "/tmp/ouroboros" CACHE STRING - "Mountpoint for RIB filesystem") - set(HAVE_FUSE TRUE CACHE INTERNAL "") - else () - message(STATUS "FUSE support disabled by user") - unset(HAVE_FUSE CACHE) - endif () -else () - message(STATUS "Install FUSE version > 2.6 to enable RIB access") -endif () - -if (NOT HAVE_FUSE) - set(FUSE_LIBRARIES "") - set(FUSE_INCLUDE_DIR "") -endif () - -mark_as_advanced(FUSE_LIBRARIES) - -find_library(LIBGCRYPT_LIBRARIES gcrypt QUIET) -if (LIBGCRYPT_LIBRARIES) - find_path(LIBGCRYPT_INCLUDE_DIR gcrypt.h - HINTS /usr/include /usr/local/include) - if (LIBGCRYPT_INCLUDE_DIR) - file(STRINGS ${LIBGCRYPT_INCLUDE_DIR}/gcrypt.h GCSTR - REGEX "^#define GCRYPT_VERSION ") - string(REGEX REPLACE "^#define GCRYPT_VERSION \"(.*)\".*$" "\\1" - GCVER "${GCSTR}") - if (NOT GCVER VERSION_LESS "1.7.0") - set(DISABLE_LIBGCRYPT FALSE CACHE BOOL "Disable libgcrypt support") - if (NOT DISABLE_LIBGCRYPT) - message(STATUS "libgcrypt support enabled") - set(HAVE_LIBGCRYPT TRUE CACHE INTERNAL "") - else () - message(STATUS "libgcrypt support disabled by user") - unset(HAVE_LIBGCRYPT CACHE) - endif() - else () - message(STATUS "Install version >= \"1.7.0\" to enable libgcrypt support " - "(found version \"${GCVER}\")") - endif() - endif () -endif () - -if (NOT HAVE_LIBGCRYPT) - set(LIBGCRYPT_LIBRARIES "") - set(LIBGCRYPT_INCLUDE_DIR "") -endif () - -find_package(OpenSSL QUIET) -if (OPENSSL_FOUND) - set(HAVE_OPENSSL_RNG TRUE) - if (OPENSSL_VERSION VERSION_LESS "1.1.0") - message(STATUS "Install version >= \"1.1.0\" to enable OpenSSL support " - "(found version \"${OPENSSL_VERSION}\")") - else () - set(DISABLE_OPENSSL FALSE CACHE BOOL "Disable OpenSSL support") - if (NOT DISABLE_OPENSSL) - message(STATUS "OpenSSL support enabled") - set(HAVE_OPENSSL TRUE) - else() - message(STATUS "OpenSSL support disabled") - unset(HAVE_OPENSSL) - endif() - endif () -endif () - -if (NOT HAVE_OPENSSL_RNG) - set(OPENSSL_INCLUDE_DIR "") - set(OPENSSL_LIBRARIES "") - set(OPENSSL_CRYPTO_LIBRARY "") -endif () - -if (APPLE OR CMAKE_SYSTEM_NAME STREQUAL "FreeBSD") - set(SYS_RND_HDR "") -else () - find_path(SYS_RND_HDR sys/random.h PATH /usr/include/ /usr/local/include/) - if (SYS_RND_HDR) - message(STATUS "Found sys/random.h in ${SYS_RND_HDR}") - set(HAVE_SYS_RANDOM TRUE) - else () - set(SYS_RND_HDR "") - unset(HAVE_SYS_RANDOM) - endif () -endif() - -if (NOT ((CMAKE_SYSTEM_NAME STREQUAL "FreeBSD") OR APPLE OR - HAVE_SYS_RANDOM OR HAVE_OPENSSL_RNG OR HAVE_LIBGCRYPT)) - message(FATAL_ERROR "No secure random generator found, " - "please install libgcrypt (> 1.7.0) or OpenSSL") -endif () - -mark_as_advanced(LIBRT_LIBRARIES LIBPTHREAD_LIBRARIES - LIBGCRYPT_LIBRARIES OPENSSL_LIBRARIES OPENSSL_CRYPTO_LIBRARY - SYS_RND_INCLUDE_DIR LIBGCRYPT_INCLUDE_DIR SYS_RND_HDR) - -set(SHM_BUFFER_SIZE 16384 CACHE STRING - "Number of blocks in packet buffer, must be a power of 2") -set(SHM_RBUFF_SIZE 1024 CACHE STRING - "Number of blocks in rbuff buffer, must be a power of 2") -set(SYS_MAX_FLOWS 10240 CACHE STRING - "Maximum number of total flows for this system") -set(PROG_MAX_FLOWS 4096 CACHE STRING - "Maximum number of flows in an application") -set(PROG_RES_FDS 64 CACHE STRING - "Number of reserved flow descriptors per application") -set(PROG_MAX_FQUEUES 32 CACHE STRING - "Maximum number of flow sets per application") -set(DU_BUFF_HEADSPACE 256 CACHE STRING - "Bytes of headspace to reserve for future headers") -set(DU_BUFF_TAILSPACE 32 CACHE STRING - "Bytes of tailspace to reserve for future tails") -if (NOT APPLE) - set(PTHREAD_COND_CLOCK "CLOCK_MONOTONIC" CACHE STRING - "Clock to use for condition variable timing") -else () - set(PTHREAD_COND_CLOCK "CLOCK_REALTIME" CACHE INTERNAL - "Clock to use for condition variable timing") -endif () -set(SOCKET_TIMEOUT 500 CACHE STRING - "Default timeout for responses from IPCPs (ms)") -set(SHM_PREFIX "ouroboros" CACHE STRING - "String to prepend to POSIX shared memory filenames") -set(SHM_RBUFF_PREFIX "/${SHM_PREFIX}.rbuff." CACHE INTERNAL - "Prefix for rbuff POSIX shared memory filenames") -set(SHM_LOCKFILE_NAME "/${SHM_PREFIX}.lockfile" CACHE INTERNAL - "Filename for the POSIX shared memory lockfile") -set(SHM_FLOW_SET_PREFIX "/${SHM_PREFIX}.set." CACHE INTERNAL - "Prefix for the POSIX shared memory flow set") -set(SHM_RDRB_NAME "/${SHM_PREFIX}.rdrb" CACHE INTERNAL - "Name for the main POSIX shared memory buffer") -set(SHM_RDRB_BLOCK_SIZE "sysconf(_SC_PAGESIZE)" CACHE STRING - "Packet buffer block size, multiple of pagesize for performance") -set(SHM_RDRB_MULTI_BLOCK TRUE CACHE BOOL - "Packet buffer multiblock packet support") -set(SHM_RBUFF_LOCKLESS FALSE CACHE BOOL - "Enable shared memory lockless rbuff support") -set(QOS_DISABLE_CRC TRUE CACHE BOOL - "Ignores ber setting on all QoS cubes") -set(DELTA_T_MPL 60 CACHE STRING - "Maximum packet lifetime (s)") -set(DELTA_T_ACK 10 CACHE STRING - "Maximum time to acknowledge a packet (s)") -set(DELTA_T_RTX 120 CACHE STRING - "Maximum time to retransmit a packet (s)") -set(FRCT_REORDER_QUEUE_SIZE 256 CACHE STRING - "Size of the reordering queue, must be a power of 2") -set(FRCT_START_WINDOW 64 CACHE STRING - "Start window, must be a power of 2") -set(FRCT_LINUX_RTT_ESTIMATOR TRUE CACHE BOOL - "Use Linux RTT estimator formula instead of the TCP RFC formula") -set(FRCT_RTO_MDEV_MULTIPLIER 2 CACHE STRING - "Multiplier for deviation term in the RTO: RTO = sRTT + (mdev << X)") -set(FRCT_RTO_INC_FACTOR 0 CACHE STRING - "Divisor for RTO increase after timeout: RTO += RTX >> X, 0: Karn/Partridge") -set(FRCT_RTO_MIN 250 CACHE STRING - "Minimum Retransmission Timeout (RTO) for FRCT (us)") -set(FRCT_TICK_TIME 5000 CACHE STRING - "Tick time for FRCT activity (retransmission, acknowledgments) (us)") -set(RXM_BUFFER_ON_HEAP FALSE CACHE BOOL - "Store packets for retransmission on the heap instead of in packet buffer") -set(RXM_BLOCKING TRUE CACHE BOOL - "Use blocking writes for retransmission") -set(RXM_MIN_RESOLUTION 20 CACHE STRING - "Minimum retransmission delay (ns), as a power to 2") -set(RXM_WHEEL_MULTIPLIER 4 CACHE STRING - "Factor for retransmission wheel levels as a power to 2") -set(RXM_WHEEL_LEVELS 3 CACHE STRING - "Number of levels in the retransmission wheel") -set(RXM_WHEEL_SLOTS_PER_LEVEL 256 CACHE STRING - "Number of slots per level in the retransmission wheel, must be a power of 2") -set(ACK_WHEEL_SLOTS 256 CACHE STRING - "Number of slots in the acknowledgment wheel, must be a power of 2") -set(ACK_WHEEL_RESOLUTION 18 CACHE STRING - "Minimum acknowledgment delay (ns), as a power to 2") - -if (HAVE_FUSE) - set(PROC_FLOW_STATS TRUE CACHE BOOL - "Enable flow statistics tracking for application flows") - if (PROC_FLOW_STATS) - message(STATUS "Application flow statistics enabled") - else () - message(STATUS "Application flow statistics disabled") - endif () -endif () - -set(SOURCE_FILES_DEV - # Add source files here - cep.c - dev.c - ) - -set(SOURCE_FILES_IRM - irm.c -) + "${CMAKE_CURRENT_SOURCE_DIR}/pb/ipcp.proto") set(SOURCE_FILES_COMMON bitmap.c btree.c - crc32.c + crc/crc8.c + crc/crc16.c + crc/crc32.c + crc/crc64.c crypt.c + crypt/keyrot.c hash.c - list.c lockfile.c logs.c md5.c @@ -267,60 +35,138 @@ set(SOURCE_FILES_COMMON serdes-irm.c serdes-oep.c sha3.c - shm_flow_set.c - shm_rbuff.c - shm_rdrbuff.c + ssm/flow_set.c + ssm/rbuff.c + ssm/pool.c sockets.c tpm.c + tw.c utils.c ) -configure_file("${CMAKE_CURRENT_SOURCE_DIR}/config.h.in" - "${CMAKE_CURRENT_BINARY_DIR}/config.h" @ONLY) +if(HAVE_OPENSSL) + list(APPEND SOURCE_FILES_COMMON crypt/openssl.c) +endif() -add_library(ouroboros-common SHARED ${SOURCE_FILES_COMMON} ${IRM_PROTO_SRCS} - ${IPCP_PROTO_SRCS} ${IPCP_CONFIG_PROTO_SRCS} ${MODEL_PROTO_SRCS} +add_library(ouroboros-common SHARED + ${SOURCE_FILES_COMMON} + ${IRM_PROTO_SRCS} + ${IPCP_PROTO_SRCS} + ${IPCP_CONFIG_PROTO_SRCS} + ${MODEL_PROTO_SRCS} ${ENROLL_PROTO_SRCS}) -add_library(ouroboros-dev SHARED ${SOURCE_FILES_DEV} ${CEP_PROTO_SRCS}) - -add_library(ouroboros-irm SHARED ${SOURCE_FILES_IRM}) - set_target_properties(ouroboros-common PROPERTIES VERSION ${PACKAGE_VERSION} SOVERSION ${PACKAGE_VERSION_MAJOR}.${PACKAGE_VERSION_MINOR}) + +ouroboros_target_debug_definitions(ouroboros-common) + +target_include_directories(ouroboros-common + PUBLIC + $<BUILD_INTERFACE:${CMAKE_SOURCE_DIR}/include> + $<BUILD_INTERFACE:${CMAKE_BINARY_DIR}/include> + $<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}> + $<BUILD_INTERFACE:${CMAKE_CURRENT_BINARY_DIR}> + $<BUILD_INTERFACE:${CMAKE_BINARY_DIR}> + $<INSTALL_INTERFACE:${CMAKE_INSTALL_INCLUDEDIR}> + PRIVATE + ${SYS_RND_HDR} + ${APPLE_INCLUDE_DIRS}) + +target_link_libraries(ouroboros-common + PRIVATE + ${LIBRT_LIBRARIES} + Threads::Threads + PUBLIC + ProtobufC::ProtobufC) + +if(HAVE_OPENSSL) + target_link_libraries(ouroboros-common PUBLIC OpenSSL::Crypto) +endif() + +if(HAVE_LIBGCRYPT) + target_link_libraries(ouroboros-common PUBLIC Gcrypt::Gcrypt) +endif() + +if(HAVE_FUSE) + target_link_libraries(ouroboros-common PRIVATE Fuse::Fuse) +endif() + +if(HAVE_LIBURCU) + target_link_libraries(ouroboros-common PRIVATE Urcu::Urcu) + # urcu headers require C99; override the global -std=c89 for this TU only. + set_source_files_properties(crypt/keyrot.c PROPERTIES + COMPILE_OPTIONS "-std=gnu99") +endif() + +install(TARGETS ouroboros-common + EXPORT OuroborosTargets + LIBRARY DESTINATION ${CMAKE_INSTALL_LIBDIR}) + +set(SOURCE_FILES_DEV + cep.c + dev.c +) + +add_library(ouroboros-dev SHARED + ${SOURCE_FILES_DEV} + ${CEP_PROTO_SRCS}) + set_target_properties(ouroboros-dev PROPERTIES VERSION ${PACKAGE_VERSION} SOVERSION ${PACKAGE_VERSION_MAJOR}.${PACKAGE_VERSION_MINOR}) + +ouroboros_target_debug_definitions(ouroboros-dev) + +target_include_directories(ouroboros-dev + PUBLIC + $<BUILD_INTERFACE:${CMAKE_SOURCE_DIR}/include> + $<INSTALL_INTERFACE:${CMAKE_INSTALL_INCLUDEDIR}> + PRIVATE + ${CMAKE_CURRENT_SOURCE_DIR} + ${CMAKE_CURRENT_BINARY_DIR} + ${CMAKE_BINARY_DIR} + ${SYS_RND_HDR}) + +target_link_libraries(ouroboros-dev PUBLIC ouroboros-common) + +install(TARGETS ouroboros-dev + EXPORT OuroborosTargets + LIBRARY DESTINATION ${CMAKE_INSTALL_LIBDIR}) + +add_library(ouroboros-irm SHARED irm.c) + set_target_properties(ouroboros-irm PROPERTIES VERSION ${PACKAGE_VERSION} SOVERSION ${PACKAGE_VERSION_MAJOR}.${PACKAGE_VERSION_MINOR}) -include(AddCompileFlags) -if (CMAKE_BUILD_TYPE MATCHES "Debug*") - add_compile_flags(ouroboros-common -DCONFIG_OUROBOROS_DEBUG) - add_compile_flags(ouroboros-dev -DCONFIG_OUROBOROS_DEBUG) - add_compile_flags(ouroboros-irm -DCONFIG_OUROBOROS_DEBUG) -endif () - -target_link_libraries(ouroboros-common ${LIBRT_LIBRARIES} - ${LIBPTHREAD_LIBRARIES} ${PROTOBUF_C_LIBRARY} ${OPENSSL_CRYPTO_LIBRARY} - ${LIBGCRYPT_LIBRARIES} ${FUSE_LIBRARIES}) +ouroboros_target_debug_definitions(ouroboros-irm) -target_link_libraries(ouroboros-dev ouroboros-common) -target_link_libraries(ouroboros-irm ouroboros-common) +target_include_directories(ouroboros-irm + PUBLIC + $<BUILD_INTERFACE:${CMAKE_SOURCE_DIR}/include> + $<INSTALL_INTERFACE:${CMAKE_INSTALL_INCLUDEDIR}> + PRIVATE + ${CMAKE_CURRENT_SOURCE_DIR} + ${CMAKE_CURRENT_BINARY_DIR} + ${CMAKE_BINARY_DIR} + ${SYS_RND_HDR}) -install(TARGETS ouroboros-common LIBRARY DESTINATION ${CMAKE_INSTALL_LIBDIR}) -install(TARGETS ouroboros-dev LIBRARY DESTINATION ${CMAKE_INSTALL_LIBDIR}) -install(TARGETS ouroboros-irm LIBRARY DESTINATION ${CMAKE_INSTALL_LIBDIR}) +target_link_libraries(ouroboros-irm PUBLIC ouroboros-common) -target_include_directories(ouroboros-common PUBLIC ${CMAKE_CURRENT_BINARY_DIR} - ${SYS_RND_HDR} ${LIBGCRYPT_INCLUDE_DIR} ${OPENSSL_INCLUDE_DIR}) +install(TARGETS ouroboros-irm + EXPORT OuroborosTargets + LIBRARY DESTINATION ${CMAKE_INSTALL_LIBDIR}) -target_include_directories(ouroboros-dev PUBLIC ${CMAKE_CURRENT_BINARY_DIR} - ${SYS_RND_HDR} ${LIBGCRYPT_INCLUDE_DIR} ${OPENSSL_INCLUDE_DIR}) +configure_file("${CMAKE_CURRENT_SOURCE_DIR}/config.h.in" + "${CMAKE_CURRENT_BINARY_DIR}/config.h" @ONLY) -target_include_directories(ouroboros-irm PUBLIC ${CMAKE_CURRENT_BINARY_DIR} - ${SYS_RND_HDR} ${LIBGCRYPT_INCLUDE_DIR} ${OPENSSL_INCLUDE_DIR}) +configure_file("${CMAKE_CURRENT_SOURCE_DIR}/ssm/ssm.h.in" + "${CMAKE_CURRENT_BINARY_DIR}/ssm.h" @ONLY) -add_subdirectory(tests) +if(BUILD_TESTS) + add_subdirectory(tests) + add_subdirectory(crc/tests) + add_subdirectory(ssm/tests) +endif() diff --git a/src/lib/bitmap.c b/src/lib/bitmap.c index b0840c44..05cf4030 100644 --- a/src/lib/bitmap.c +++ b/src/lib/bitmap.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Bitmap implementation * diff --git a/src/lib/btree.c b/src/lib/btree.c index 1af94b73..37ec5e52 100644 --- a/src/lib/btree.c +++ b/src/lib/btree.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * B-trees * diff --git a/src/lib/cep.c b/src/lib/cep.c index ba238023..e953e2d9 100644 --- a/src/lib/cep.c +++ b/src/lib/cep.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * The Ouroboros Connection Establishment Protocol * diff --git a/src/lib/config.h.in b/src/lib/config.h.in index 604038b4..26ebe56b 100644 --- a/src/lib/config.h.in +++ b/src/lib/config.h.in @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Ouroboros library configuration * @@ -20,29 +20,39 @@ * Foundation, Inc., http://www.fsf.org/about/contact/. */ +#ifndef MILLION +#define MILLION 1000000LL +#endif + +#ifndef BILLION +#define BILLION 1000000000LL +#endif + #cmakedefine HAVE_SYS_RANDOM +#cmakedefine HAVE_EXPLICIT_BZERO #cmakedefine HAVE_LIBGCRYPT #cmakedefine HAVE_OPENSSL - #ifdef HAVE_OPENSSL +#cmakedefine HAVE_OPENSSL_ML_KEM +#cmakedefine HAVE_OPENSSL_ML_DSA +#cmakedefine HAVE_OPENSSL_SLH_DSA #define HAVE_ENCRYPTION +#define SECMEM_MINSIZE @SECMEM_MINSIZE@ #endif +#define PROC_SECMEM_MAX @PROC_SECMEM_MAX@ + +#define SYS_MAX_FLOWS @SYS_MAX_FLOWS@ -#define SYS_MAX_FLOWS @SYS_MAX_FLOWS@ +#cmakedefine QOS_DISABLE_CRC +#cmakedefine HAVE_OPENSSL_RNG +#cmakedefine HAVE_PCLMUL +#cmakedefine HAVE_PMULL -#cmakedefine SHM_RBUFF_LOCKLESS -#cmakedefine SHM_RDRB_MULTI_BLOCK -#cmakedefine QOS_DISABLE_CRC -#cmakedefine HAVE_OPENSSL_RNG +#define SHM_LOCKFILE_NAME "@SHM_LOCKFILE_NAME@" +#define FLOW_ALLOC_TIMEOUT @FLOW_ALLOC_TIMEOUT@ -#define SHM_RBUFF_PREFIX "@SHM_RBUFF_PREFIX@" -#define SHM_LOCKFILE_NAME "@SHM_LOCKFILE_NAME@" -#define SHM_FLOW_SET_PREFIX "@SHM_FLOW_SET_PREFIX@" -#define SHM_RDRB_NAME "@SHM_RDRB_NAME@" -#define SHM_RDRB_BLOCK_SIZE @SHM_RDRB_BLOCK_SIZE@ -#define SHM_BUFFER_SIZE @SHM_BUFFER_SIZE@ -#define SHM_RBUFF_SIZE @SHM_RBUFF_SIZE@ -#define FLOW_ALLOC_TIMEOUT @FLOW_ALLOC_TIMEOUT@ +#define TPM_DEBUG_REPORT_INTERVAL @TPM_DEBUG_REPORT_INTERVAL@ +#define TPM_DEBUG_ABORT_TIMEOUT @TPM_DEBUG_ABORT_TIMEOUT@ #if defined(__linux__) || (defined(__MACH__) && !defined(__APPLE__)) /* Avoid a bug in robust mutex implementation of glibc 2.25 */ @@ -60,19 +70,20 @@ #cmakedefine PROC_FLOW_STATS #endif -#define PTHREAD_COND_CLOCK @PTHREAD_COND_CLOCK@ +#cmakedefine HAVE_LIBURCU + +#cmakedefine FRCT_DEBUG_STDOUT -#define PROG_MAX_FLOWS @PROG_MAX_FLOWS@ -#define PROG_RES_FDS @PROG_RES_FDS@ -#define PROG_MAX_FQUEUES @PROG_MAX_FQUEUES@ +#define PTHREAD_COND_CLOCK @PTHREAD_COND_CLOCK@ -#define DU_BUFF_HEADSPACE @DU_BUFF_HEADSPACE@ -#define DU_BUFF_TAILSPACE @DU_BUFF_TAILSPACE@ +#define PROC_MAX_FLOWS @PROC_MAX_FLOWS@ +#define PROC_RES_FDS @PROC_RES_FDS@ +#define PROC_MAX_FQUEUES @PROC_MAX_FQUEUES@ /* Default Delta-t parameters */ #cmakedefine FRCT_LINUX_RTT_ESTIMATOR -#define DELT_A (@DELTA_T_ACK@) /* ns */ -#define DELT_R (@DELTA_T_RTX@) /* ns */ +#define DELT_A (@DELTA_T_ACK@) /* ms */ +#define DELT_R (@DELTA_T_RTX@) /* ms */ #define RQ_SIZE (@FRCT_REORDER_QUEUE_SIZE@) #define START_WINDOW (@FRCT_START_WINDOW@) @@ -83,9 +94,6 @@ #define TICTIME (@FRCT_TICK_TIME@ * 1000) /* ns */ /* Retransmission tuning */ -#cmakedefine RXM_BUFFER_ON_HEAP -#cmakedefine RXM_BLOCKING - #define RXMQ_RES (@RXM_MIN_RESOLUTION@) /* 2^N ns */ #define RXMQ_BUMP (@RXM_WHEEL_MULTIPLIER@) #define RXMQ_LVLS (@RXM_WHEEL_LEVELS@) @@ -93,3 +101,10 @@ #define ACKQ_SLOTS (@ACK_WHEEL_SLOTS@) #define ACKQ_RES (@ACK_WHEEL_RESOLUTION@) /* 2^N ns */ + +#define KEY_LEAF_BITS (@KEY_LEAF_BITS@) /* pkts/leaf-key = 2^n */ +#define KEY_NODE_BITS (@KEY_NODE_BITS@) /* leaf-keys/node = 2^n */ +#define KEY_NODE_COUNT (@KEY_NODE_COUNT@) /* node keys/batch N */ +#define KEY_REKEY_WATERMARK (@KEY_REKEY_WATERMARK@) /* node-keys-left trig */ +#define KEY_REPLAY_WINDOW (@KEY_REPLAY_WINDOW@) /* rx replay win pkts */ +#define FLOW_WM_CHECK (1u << @KEY_REKEY_WM_CHECK_BITS@) /* wm chk/n wr */ diff --git a/src/lib/crc/crc16.c b/src/lib/crc/crc16.c new file mode 100644 index 00000000..9dc59429 --- /dev/null +++ b/src/lib/crc/crc16.c @@ -0,0 +1,61 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * 16-bit Cyclic Redundancy Check (CCITT-FALSE variant) + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This library is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public License + * version 2.1 as published by the Free Software Foundation. + * + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with this library; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +/* + * CRC-16/CCITT-FALSE (reveng catalog, alias CRC-16/IBM-3740): + * poly = 0x1021 + * init = 0xffff + * refin = false + * refout = false + * xorout = 0x0000 + * check = crc16_ccitt_false("123456789") == 0x29b1 + */ + +#include "config.h" + +#include <ouroboros/crc16.h> + +/* Bit-by-bit MSB-first CRC. */ +void crc16_ccitt_false(uint16_t * crc, + const void * buf, + size_t len) +{ + const uint8_t * p; + uint16_t c; + size_t n; + int i; + + p = (const uint8_t *) buf; + c = *crc ^ 0xffff; + + for (n = 0; n < len; n++) { + c ^= ((uint16_t) p[n]) << 8; + for (i = 0; i < 8; i++) { + if (c & 0x8000) + c = (uint16_t) ((c << 1) ^ 0x1021); + else + c = (uint16_t) (c << 1); + } + } + + *crc = c; +} diff --git a/src/lib/crc32.c b/src/lib/crc/crc32.c index f369ad20..0fdb62b1 100644 --- a/src/lib/crc32.c +++ b/src/lib/crc/crc32.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * 32-bit Cyclic Redundancy Check * diff --git a/src/lib/crc/crc64.c b/src/lib/crc/crc64.c new file mode 100644 index 00000000..1b6fb5f6 --- /dev/null +++ b/src/lib/crc/crc64.c @@ -0,0 +1,363 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * 64-bit Cyclic Redundancy Check (NVMe variant) + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This library is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public License + * version 2.1 as published by the Free Software Foundation. + * + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with this library; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +/* + * CRC-64/NVMe (reveng catalog): + * poly = 0xad93d23594c93659 + * init = 0xffffffffffffffff + * refin = true + * refout = true + * xorout = 0xffffffffffffffff + * check = crc64_nvme("123456789") == 0xae8b14860a799888 + */ + +#include "config.h" + +#include <ouroboros/crc64.h> + +/* + * Reflected CRC-64/NVMe table. Polynomial in reflected form: + * 0x9a6c9329ac4bc9b5 (bitrev of 0xad93d23594c93659). + */ +static const uint64_t crc64_nvme_tab[256] = { + 0x0000000000000000ULL, 0x7f6ef0c830358979ULL, + 0xfedde190606b12f2ULL, 0x81b31158505e9b8bULL, + 0xc962e5739841b68fULL, 0xb60c15bba8743ff6ULL, + 0x37bf04e3f82aa47dULL, 0x48d1f42bc81f2d04ULL, + 0xa61cecb46814fe75ULL, 0xd9721c7c5821770cULL, + 0x58c10d24087fec87ULL, 0x27affdec384a65feULL, + 0x6f7e09c7f05548faULL, 0x1010f90fc060c183ULL, + 0x91a3e857903e5a08ULL, 0xeecd189fa00bd371ULL, + 0x78e0ff3b88be6f81ULL, 0x078e0ff3b88be6f8ULL, + 0x863d1eabe8d57d73ULL, 0xf953ee63d8e0f40aULL, + 0xb1821a4810ffd90eULL, 0xceecea8020ca5077ULL, + 0x4f5ffbd87094cbfcULL, 0x30310b1040a14285ULL, + 0xdefc138fe0aa91f4ULL, 0xa192e347d09f188dULL, + 0x2021f21f80c18306ULL, 0x5f4f02d7b0f40a7fULL, + 0x179ef6fc78eb277bULL, 0x68f0063448deae02ULL, + 0xe943176c18803589ULL, 0x962de7a428b5bcf0ULL, + 0xf1c1fe77117cdf02ULL, 0x8eaf0ebf2149567bULL, + 0x0f1c1fe77117cdf0ULL, 0x7072ef2f41224489ULL, + 0x38a31b04893d698dULL, 0x47cdebccb908e0f4ULL, + 0xc67efa94e9567b7fULL, 0xb9100a5cd963f206ULL, + 0x57dd12c379682177ULL, 0x28b3e20b495da80eULL, + 0xa900f35319033385ULL, 0xd66e039b2936bafcULL, + 0x9ebff7b0e12997f8ULL, 0xe1d10778d11c1e81ULL, + 0x606216208142850aULL, 0x1f0ce6e8b1770c73ULL, + 0x8921014c99c2b083ULL, 0xf64ff184a9f739faULL, + 0x77fce0dcf9a9a271ULL, 0x08921014c99c2b08ULL, + 0x4043e43f0183060cULL, 0x3f2d14f731b68f75ULL, + 0xbe9e05af61e814feULL, 0xc1f0f56751dd9d87ULL, + 0x2f3dedf8f1d64ef6ULL, 0x50531d30c1e3c78fULL, + 0xd1e00c6891bd5c04ULL, 0xae8efca0a188d57dULL, + 0xe65f088b6997f879ULL, 0x9931f84359a27100ULL, + 0x1882e91b09fcea8bULL, 0x67ec19d339c963f2ULL, + 0xd75adabd7a6e2d6fULL, 0xa8342a754a5ba416ULL, + 0x29873b2d1a053f9dULL, 0x56e9cbe52a30b6e4ULL, + 0x1e383fcee22f9be0ULL, 0x6156cf06d21a1299ULL, + 0xe0e5de5e82448912ULL, 0x9f8b2e96b271006bULL, + 0x71463609127ad31aULL, 0x0e28c6c1224f5a63ULL, + 0x8f9bd7997211c1e8ULL, 0xf0f5275142244891ULL, + 0xb824d37a8a3b6595ULL, 0xc74a23b2ba0eececULL, + 0x46f932eaea507767ULL, 0x3997c222da65fe1eULL, + 0xafba2586f2d042eeULL, 0xd0d4d54ec2e5cb97ULL, + 0x5167c41692bb501cULL, 0x2e0934dea28ed965ULL, + 0x66d8c0f56a91f461ULL, 0x19b6303d5aa47d18ULL, + 0x980521650afae693ULL, 0xe76bd1ad3acf6feaULL, + 0x09a6c9329ac4bc9bULL, 0x76c839faaaf135e2ULL, + 0xf77b28a2faafae69ULL, 0x8815d86aca9a2710ULL, + 0xc0c42c4102850a14ULL, 0xbfaadc8932b0836dULL, + 0x3e19cdd162ee18e6ULL, 0x41773d1952db919fULL, + 0x269b24ca6b12f26dULL, 0x59f5d4025b277b14ULL, + 0xd846c55a0b79e09fULL, 0xa72835923b4c69e6ULL, + 0xeff9c1b9f35344e2ULL, 0x90973171c366cd9bULL, + 0x1124202993385610ULL, 0x6e4ad0e1a30ddf69ULL, + 0x8087c87e03060c18ULL, 0xffe938b633338561ULL, + 0x7e5a29ee636d1eeaULL, 0x0134d92653589793ULL, + 0x49e52d0d9b47ba97ULL, 0x368bddc5ab7233eeULL, + 0xb738cc9dfb2ca865ULL, 0xc8563c55cb19211cULL, + 0x5e7bdbf1e3ac9decULL, 0x21152b39d3991495ULL, + 0xa0a63a6183c78f1eULL, 0xdfc8caa9b3f20667ULL, + 0x97193e827bed2b63ULL, 0xe877ce4a4bd8a21aULL, + 0x69c4df121b863991ULL, 0x16aa2fda2bb3b0e8ULL, + 0xf86737458bb86399ULL, 0x8709c78dbb8deae0ULL, + 0x06bad6d5ebd3716bULL, 0x79d4261ddbe6f812ULL, + 0x3105d23613f9d516ULL, 0x4e6b22fe23cc5c6fULL, + 0xcfd833a67392c7e4ULL, 0xb0b6c36e43a74e9dULL, + 0x9a6c9329ac4bc9b5ULL, 0xe50263e19c7e40ccULL, + 0x64b172b9cc20db47ULL, 0x1bdf8271fc15523eULL, + 0x530e765a340a7f3aULL, 0x2c608692043ff643ULL, + 0xadd397ca54616dc8ULL, 0xd2bd67026454e4b1ULL, + 0x3c707f9dc45f37c0ULL, 0x431e8f55f46abeb9ULL, + 0xc2ad9e0da4342532ULL, 0xbdc36ec59401ac4bULL, + 0xf5129aee5c1e814fULL, 0x8a7c6a266c2b0836ULL, + 0x0bcf7b7e3c7593bdULL, 0x74a18bb60c401ac4ULL, + 0xe28c6c1224f5a634ULL, 0x9de29cda14c02f4dULL, + 0x1c518d82449eb4c6ULL, 0x633f7d4a74ab3dbfULL, + 0x2bee8961bcb410bbULL, 0x548079a98c8199c2ULL, + 0xd53368f1dcdf0249ULL, 0xaa5d9839ecea8b30ULL, + 0x449080a64ce15841ULL, 0x3bfe706e7cd4d138ULL, + 0xba4d61362c8a4ab3ULL, 0xc52391fe1cbfc3caULL, + 0x8df265d5d4a0eeceULL, 0xf29c951de49567b7ULL, + 0x732f8445b4cbfc3cULL, 0x0c41748d84fe7545ULL, + 0x6bad6d5ebd3716b7ULL, 0x14c39d968d029fceULL, + 0x95708ccedd5c0445ULL, 0xea1e7c06ed698d3cULL, + 0xa2cf882d2576a038ULL, 0xdda178e515432941ULL, + 0x5c1269bd451db2caULL, 0x237c997575283bb3ULL, + 0xcdb181ead523e8c2ULL, 0xb2df7122e51661bbULL, + 0x336c607ab548fa30ULL, 0x4c0290b2857d7349ULL, + 0x04d364994d625e4dULL, 0x7bbd94517d57d734ULL, + 0xfa0e85092d094cbfULL, 0x856075c11d3cc5c6ULL, + 0x134d926535897936ULL, 0x6c2362ad05bcf04fULL, + 0xed9073f555e26bc4ULL, 0x92fe833d65d7e2bdULL, + 0xda2f7716adc8cfb9ULL, 0xa54187de9dfd46c0ULL, + 0x24f29686cda3dd4bULL, 0x5b9c664efd965432ULL, + 0xb5517ed15d9d8743ULL, 0xca3f8e196da80e3aULL, + 0x4b8c9f413df695b1ULL, 0x34e26f890dc31cc8ULL, + 0x7c339ba2c5dc31ccULL, 0x035d6b6af5e9b8b5ULL, + 0x82ee7a32a5b7233eULL, 0xfd808afa9582aa47ULL, + 0x4d364994d625e4daULL, 0x3258b95ce6106da3ULL, + 0xb3eba804b64ef628ULL, 0xcc8558cc867b7f51ULL, + 0x8454ace74e645255ULL, 0xfb3a5c2f7e51db2cULL, + 0x7a894d772e0f40a7ULL, 0x05e7bdbf1e3ac9deULL, + 0xeb2aa520be311aafULL, 0x944455e88e0493d6ULL, + 0x15f744b0de5a085dULL, 0x6a99b478ee6f8124ULL, + 0x224840532670ac20ULL, 0x5d26b09b16452559ULL, + 0xdc95a1c3461bbed2ULL, 0xa3fb510b762e37abULL, + 0x35d6b6af5e9b8b5bULL, 0x4ab846676eae0222ULL, + 0xcb0b573f3ef099a9ULL, 0xb465a7f70ec510d0ULL, + 0xfcb453dcc6da3dd4ULL, 0x83daa314f6efb4adULL, + 0x0269b24ca6b12f26ULL, 0x7d0742849684a65fULL, + 0x93ca5a1b368f752eULL, 0xeca4aad306bafc57ULL, + 0x6d17bb8b56e467dcULL, 0x12794b4366d1eea5ULL, + 0x5aa8bf68aecec3a1ULL, 0x25c64fa09efb4ad8ULL, + 0xa4755ef8cea5d153ULL, 0xdb1bae30fe90582aULL, + 0xbcf7b7e3c7593bd8ULL, 0xc399472bf76cb2a1ULL, + 0x422a5673a732292aULL, 0x3d44a6bb9707a053ULL, + 0x759552905f188d57ULL, 0x0afba2586f2d042eULL, + 0x8b48b3003f739fa5ULL, 0xf42643c80f4616dcULL, + 0x1aeb5b57af4dc5adULL, 0x6585ab9f9f784cd4ULL, + 0xe436bac7cf26d75fULL, 0x9b584a0fff135e26ULL, + 0xd389be24370c7322ULL, 0xace74eec0739fa5bULL, + 0x2d545fb4576761d0ULL, 0x523aaf7c6752e8a9ULL, + 0xc41748d84fe75459ULL, 0xbb79b8107fd2dd20ULL, + 0x3acaa9482f8c46abULL, 0x45a459801fb9cfd2ULL, + 0x0d75adabd7a6e2d6ULL, 0x721b5d63e7936bafULL, + 0xf3a84c3bb7cdf024ULL, 0x8cc6bcf387f8795dULL, + 0x620ba46c27f3aa2cULL, 0x1d6554a417c62355ULL, + 0x9cd645fc4798b8deULL, 0xe3b8b53477ad31a7ULL, + 0xab69411fbfb21ca3ULL, 0xd407b1d78f8795daULL, + 0x55b4a08fdfd90e51ULL, 0x2ada5047efec8728ULL +}; + +static __inline__ uint64_t crc64_nvme_step(uint64_t c, + const uint8_t * p, + size_t len) +{ + size_t n; + + for (n = 0; n < len; n++) + c = crc64_nvme_tab[(c ^ p[n]) & 0xff] ^ (c >> 8); + + return c; +} + +void crc64_nvme_table(uint64_t * crc, + const void * buf, + size_t len) +{ + uint64_t c; + + c = crc64_nvme_step(*crc ^ UINT64_MAX, + (const uint8_t *) buf, len); + + *crc = c ^ UINT64_MAX; +} + +#ifdef HAVE_PCLMUL + +#include <smmintrin.h> +#include <wmmintrin.h> + +/* + * Fold-by-16 constants for reflected CRC-64/NVMe. Properties of the + * polynomial; identical between the PCLMUL and PMULL backends. + * k3 = bitrev64(x^(128+64) mod P) << 1 + * k4 = bitrev64(x^(128+0) mod P) << 1 + */ +static const uint64_t k3_clmul = 0xeadc41fd2ba3d420ULL; +static const uint64_t k4_clmul = 0x21e9761e252621acULL; + +__attribute__((target("pclmul,sse4.1"))) +static __m128i fold16(__m128i x, + __m128i k) +{ + __m128i lo; + __m128i hi; + + lo = _mm_clmulepi64_si128(x, k, 0x00); + hi = _mm_clmulepi64_si128(x, k, 0x11); + return _mm_xor_si128(lo, hi); +} + +/* + * Fold-by-16 over 16-byte chunks; the 128-bit folded state is then + * emitted as 16 little-endian bytes and run through the byte-table + * loop together with any tail (<=15 bytes). The 16-byte minimum on + * the bulk loop is why the short-input path uses the table directly. + */ +__attribute__((target("pclmul,sse4.1"))) +static void crc64_nvme_clmul(uint64_t * crc, + const void * buf, + size_t len) +{ + const uint8_t * p; + uint64_t seed; + uint64_t c; + size_t off; + __m128i x; + __m128i k; + uint8_t post[16]; + + p = (const uint8_t *) buf; + seed = *crc; + + if (len < 16) { + c = crc64_nvme_step(seed ^ UINT64_MAX, p, len); + *crc = c ^ UINT64_MAX; + return; + } + + x = _mm_loadu_si128((const __m128i *) p); + x = _mm_xor_si128(x, _mm_cvtsi64_si128((int64_t) + (seed ^ UINT64_MAX))); + + k = _mm_set_epi64x((int64_t) k4_clmul, (int64_t) k3_clmul); + + off = 16; + while (off + 16 <= len) { + __m128i d; + + d = _mm_loadu_si128((const __m128i *) (p + off)); + x = _mm_xor_si128(fold16(x, k), d); + off += 16; + } + + _mm_storeu_si128((__m128i *) post, x); + + c = crc64_nvme_step(0, post, 16); + c = crc64_nvme_step(c, p + off, len - off); + + *crc = c ^ UINT64_MAX; +} + +#endif /* HAVE_PCLMUL */ + +#ifdef HAVE_PMULL + +#include <arm_neon.h> + +/* Same fold-by-16 constants as the PCLMUL path (poly properties). */ +static const uint64_t k3_pmull = 0xeadc41fd2ba3d420ULL; +static const uint64_t k4_pmull = 0x21e9761e252621acULL; + +__attribute__((target("+crypto"))) +static uint64x2_t fold16_pmull(uint64x2_t x, + uint64x2_t k) +{ + poly64x2_t xp; + poly64x2_t kp; + uint64x2_t lo; + uint64x2_t hi; + + xp = vreinterpretq_p64_u64(x); + kp = vreinterpretq_p64_u64(k); + lo = vreinterpretq_u64_p128( + vmull_p64((poly64_t) vgetq_lane_u64(x, 0), + (poly64_t) vgetq_lane_u64(k, 0))); + hi = vreinterpretq_u64_p128(vmull_high_p64(xp, kp)); + return veorq_u64(lo, hi); +} + +__attribute__((target("+crypto"))) +static void crc64_nvme_pmull(uint64_t * crc, + const void * buf, + size_t len) +{ + const uint8_t * p; + uint64_t seed; + uint64_t c; + size_t off; + uint64x2_t x; + uint64x2_t k; + uint64_t seed_lane[2]; + uint64_t k_lanes[2]; + uint8_t post[16]; + + p = (const uint8_t *) buf; + seed = *crc; + + if (len < 16) { + c = crc64_nvme_step(seed ^ UINT64_MAX, p, len); + *crc = c ^ UINT64_MAX; + return; + } + + x = vld1q_u64((const uint64_t *) p); + seed_lane[0] = seed ^ UINT64_MAX; + seed_lane[1] = 0; + x = veorq_u64(x, vld1q_u64(seed_lane)); + + k_lanes[0] = k3_pmull; + k_lanes[1] = k4_pmull; + k = vld1q_u64(k_lanes); + + off = 16; + while (off + 16 <= len) { + uint64x2_t d; + + d = vld1q_u64((const uint64_t *) (p + off)); + x = veorq_u64(fold16_pmull(x, k), d); + off += 16; + } + + vst1q_u8(post, vreinterpretq_u8_u64(x)); + + c = crc64_nvme_step(0, post, 16); + c = crc64_nvme_step(c, p + off, len - off); + + *crc = c ^ UINT64_MAX; +} +#endif /* HAVE_PMULL */ + +void crc64_nvme(uint64_t * crc, + const void * buf, + size_t len) +{ +#ifdef HAVE_PCLMUL + crc64_nvme_clmul(crc, buf, len); +#elif defined(HAVE_PMULL) + crc64_nvme_pmull(crc, buf, len); +#else + crc64_nvme_table(crc, buf, len); +#endif +} diff --git a/src/lib/crc/crc8.c b/src/lib/crc/crc8.c new file mode 100644 index 00000000..20976b29 --- /dev/null +++ b/src/lib/crc/crc8.c @@ -0,0 +1,62 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * 8-bit Cyclic Redundancy Check (AUTOSAR variant) + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This library is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public License + * version 2.1 as published by the Free Software Foundation. + * + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with this library; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +/* + * CRC-8/AUTOSAR (reveng catalog): + * poly = 0x2f + * init = 0xff + * refin = false + * refout = false + * xorout = 0xff + * check = crc8_autosar("123456789") == 0xdf + */ + +#include "config.h" + +#include <ouroboros/crc8.h> + + + /* Bit-by-bit MSB-first CRC. */ +void crc8_autosar(uint8_t * crc, + const void * buf, + size_t len) +{ + const uint8_t * p; + uint8_t c; + size_t n; + int i; + + p = (const uint8_t *) buf; + c = *crc ^ 0xff; + + for (n = 0; n < len; n++) { + c ^= p[n]; + for (i = 0; i < 8; i++) { + if (c & 0x80) + c = (uint8_t) ((c << 1) ^ 0x2f); + else + c = (uint8_t) (c << 1); + } + } + + *crc = c ^ 0xff; +} diff --git a/src/lib/crc/tests/CMakeLists.txt b/src/lib/crc/tests/CMakeLists.txt new file mode 100644 index 00000000..11daca5a --- /dev/null +++ b/src/lib/crc/tests/CMakeLists.txt @@ -0,0 +1,21 @@ +get_filename_component(PARENT_PATH ${CMAKE_CURRENT_SOURCE_DIR} DIRECTORY) +get_filename_component(PARENT_DIR ${PARENT_PATH} NAME) + +compute_test_prefix() + +create_test_sourcelist(${PARENT_DIR}_tests test_suite.c + # Add new tests here + crc8_test.c + crc16_test.c + crc32_test.c + crc64_test.c + ) + +add_executable(${PARENT_DIR}_test ${${PARENT_DIR}_tests}) + +disable_test_logging_for_target(${PARENT_DIR}_test) +target_link_libraries(${PARENT_DIR}_test ouroboros-common) + +add_dependencies(build_tests ${PARENT_DIR}_test) + +ouroboros_register_tests(TARGET ${PARENT_DIR}_test TESTS ${${PARENT_DIR}_tests}) diff --git a/src/lib/crc/tests/crc16_test.c b/src/lib/crc/tests/crc16_test.c new file mode 100644 index 00000000..03a5b504 --- /dev/null +++ b/src/lib/crc/tests/crc16_test.c @@ -0,0 +1,67 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Test of the CRC-16/CCITT-FALSE function + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#include "config.h" + +#include <ouroboros/crc16.h> + +#include <test/test.h> + +#include <stddef.h> +#include <stdint.h> +#include <stdio.h> + +/* reveng-catalog smoke vectors. */ +static int test_crc16_ccitt_false_basic(void) +{ + uint16_t crc; + + TEST_START(); + + crc = 0; + crc16_ccitt_false(&crc, "", 0); + if (crc != 0xffff) + goto fail; + + crc = 0; + crc16_ccitt_false(&crc, "123456789", 9); + if (crc != 0x29b1) + goto fail; + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +int crc16_test(int argc, + char ** argv) +{ + int ret = 0; + + (void) argc; + (void) argv; + + ret |= test_crc16_ccitt_false_basic(); + return ret; +} diff --git a/src/lib/tests/crc32_test.c b/src/lib/crc/tests/crc32_test.c index a26c8220..5a1ddd87 100644 --- a/src/lib/tests/crc32_test.c +++ b/src/lib/crc/tests/crc32_test.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Test of the CRC32 function * diff --git a/src/lib/crc/tests/crc64_test.c b/src/lib/crc/tests/crc64_test.c new file mode 100644 index 00000000..cf3f5ca3 --- /dev/null +++ b/src/lib/crc/tests/crc64_test.c @@ -0,0 +1,126 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Test of the CRC-64/NVMe function + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#include "config.h" + +#include <ouroboros/crc64.h> +#include <ouroboros/random.h> + +#include <test/test.h> + +#include <stddef.h> +#include <stdint.h> +#include <stdio.h> + +/* Reference impl, internal to libouroboros-common. */ +extern void crc64_nvme_table(uint64_t * crc, + const void * buf, + size_t len); + +/* reveng-catalog smoke vectors plus a 16-byte fold-boundary check. */ +static int test_crc64_nvme_basic(void) +{ + uint64_t crc; + + TEST_START(); + + crc = 0; + crc64_nvme(&crc, "", 0); + if (crc != 0x0000000000000000ULL) + goto fail; + + crc = 0; + crc64_nvme(&crc, "123456789", 9); + if (crc != 0xae8b14860a799888ULL) + goto fail; + + crc = 0; + crc64_nvme(&crc, "0123456789abcdef", 16); + if (crc != 0x091485ca7018730eULL) + goto fail; + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +#if defined(HAVE_PCLMUL) || defined(HAVE_PMULL) +/* Cross-check the accelerated dispatcher path against the byte-table. */ +static int test_crc64_nvme_random(void) +{ + static const size_t lens[] = { + 0, 1, 7, 8, 15, 16, 17, 31, 32, 33, 63, 64, 65, 127, 128, + 129, 255, 256, 257, 1023, 1024, 1025, 4096 + }; + uint8_t buf[4096]; + size_t i; + uint64_t ref; + uint64_t got; + + TEST_START(); + + if (random_buffer(buf, sizeof(buf)) < 0) { + printf("Failed to generate random data.\n"); + goto fail; + } + + for (i = 0; i < sizeof(lens) / sizeof(lens[0]); i++) { + ref = 0; + crc64_nvme_table(&ref, buf, lens[i]); + + got = 0; + crc64_nvme(&got, buf, lens[i]); + + if (ref == got) + continue; + + printf("Mismatch at len=%zu: table=0x%016lx disp=0x%016lx\n", + lens[i], + (unsigned long) ref, + (unsigned long) got); + goto fail; + } + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +#endif +} + +int crc64_test(int argc, + char ** argv) +{ + int ret = 0; + + (void) argc; + (void) argv; + + ret |= test_crc64_nvme_basic(); +#if defined(HAVE_PCLMUL) || defined(HAVE_PMULL) + ret |= test_crc64_nvme_random(); +#endif + return ret; +} diff --git a/src/lib/crc/tests/crc8_test.c b/src/lib/crc/tests/crc8_test.c new file mode 100644 index 00000000..f7bb33b8 --- /dev/null +++ b/src/lib/crc/tests/crc8_test.c @@ -0,0 +1,67 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Test of the CRC-8/AUTOSAR function + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#include "config.h" + +#include <ouroboros/crc8.h> + +#include <test/test.h> + +#include <stddef.h> +#include <stdint.h> +#include <stdio.h> + +/* reveng-catalog smoke vectors. */ +static int test_crc8_autosar_basic(void) +{ + uint8_t crc; + + TEST_START(); + + crc = 0; + crc8_autosar(&crc, "", 0); + if (crc != 0x00) + goto fail; + + crc = 0; + crc8_autosar(&crc, "123456789", 9); + if (crc != 0xdf) + goto fail; + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +int crc8_test(int argc, + char ** argv) +{ + int ret = 0; + + (void) argc; + (void) argv; + + ret |= test_crc8_autosar_basic(); + return ret; +} diff --git a/src/lib/crypt.c b/src/lib/crypt.c index ad679501..5da9d392 100644 --- a/src/lib/crypt.c +++ b/src/lib/crypt.c @@ -1,8 +1,7 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * - * Elliptic curve Diffie-Hellman key exchange and - * AES encryption for flows using OpenSSL + * Cryptographic operations * * Dimitri Staessens <dimitri@ouroboros.rocks> * Sander Vrijders <sander@ouroboros.rocks> @@ -20,436 +19,1443 @@ * License along with this library; if not, write to the Free Software * Foundation, Inc., http://www.fsf.org/about/contact/. */ + +#if defined(__linux__) || defined(__CYGWIN__) +#define _DEFAULT_SOURCE +#endif + #include <config.h> -#include <ouroboros/crypt.h> #include <ouroboros/errno.h> +#include <ouroboros/pthread.h> +#include <ouroboros/random.h> +#include <ouroboros/crypt.h> + +#include "crypt/keyrot.h" + +#ifdef HAVE_OPENSSL +#include <openssl/crypto.h> +#include <openssl/evp.h> +#include "crypt/openssl.h" +#endif #include <assert.h> +#include <stdio.h> #include <string.h> - +#include <sys/stat.h> + +struct nid_map { + uint16_t nid; + const char * name; +}; + +static const struct nid_map cipher_nid_map[] = { + {NID_aes_128_gcm, "aes-128-gcm"}, + {NID_aes_192_gcm, "aes-192-gcm"}, + {NID_aes_256_gcm, "aes-256-gcm"}, + {NID_chacha20_poly1305, "chacha20-poly1305"}, + {NID_undef, NULL} +}; + +/* Ordered in strength preference, lowest first */ +const uint16_t crypt_supported_nids[] = { +#ifdef HAVE_OPENSSL + NID_aes_128_gcm, + NID_aes_192_gcm, + NID_aes_256_gcm, + NID_chacha20_poly1305, +#endif + NID_undef +}; + +static const struct nid_map kex_nid_map[] = { + {NID_X9_62_prime256v1, "prime256v1"}, + {NID_secp384r1, "secp384r1"}, + {NID_secp521r1, "secp521r1"}, + {NID_X25519, "X25519"}, + {NID_X448, "X448"}, + {NID_ffdhe2048, "ffdhe2048"}, + {NID_ffdhe3072, "ffdhe3072"}, + {NID_ffdhe4096, "ffdhe4096"}, + {NID_MLKEM512, "ML-KEM-512"}, + {NID_MLKEM768, "ML-KEM-768"}, + {NID_MLKEM1024, "ML-KEM-1024"}, + {NID_X25519MLKEM768, "X25519MLKEM768"}, + {NID_X448MLKEM1024, "X448MLKEM1024"}, + {NID_undef, NULL} +}; + +/* Ordered in strength preference, lowest first (NIST SP 800-57 levels) */ +const uint16_t kex_supported_nids[] = { +#ifdef HAVE_OPENSSL + NID_ffdhe2048, /* FFDHE-2048, ~112-bit */ + NID_X9_62_prime256v1, /* ECDH P-256, 128-bit */ + NID_X25519, /* ECDH X25519, 128-bit */ + NID_ffdhe3072, /* FFDHE-3072, ~128-bit */ + NID_ffdhe4096, /* FFDHE-4096, ~152-bit */ + NID_secp384r1, /* ECDH P-384, 192-bit */ + NID_X448, /* ECDH X448, 224-bit */ + NID_secp521r1, /* ECDH P-521, 256-bit */ +#ifdef HAVE_OPENSSL_ML_KEM + NID_MLKEM512, /* ML-KEM-512, PQC L1 (~AES-128) */ + NID_MLKEM768, /* ML-KEM-768, PQC L3 (~AES-192) */ + NID_MLKEM1024, /* ML-KEM-1024, PQC L5 (~AES-256) */ + NID_X25519MLKEM768, /* X25519 + ML-KEM-768, PQC L3 */ + NID_X448MLKEM1024, /* X448 + ML-KEM-1024, PQC L5 */ +#endif +#endif + NID_undef +}; + +static const struct nid_map md_nid_map[] = { + {NID_sha256, "sha256"}, + {NID_sha384, "sha384"}, + {NID_sha512, "sha512"}, + {NID_sha3_256, "sha3-256"}, + {NID_sha3_384, "sha3-384"}, + {NID_sha3_512, "sha3-512"}, + {NID_blake2b512, "blake2b512"}, + {NID_blake2s256, "blake2s256"}, + {NID_undef, NULL} +}; + +/* Ordered in strength preference, lowest first */ +const uint16_t md_supported_nids[] = { #ifdef HAVE_OPENSSL + NID_blake2s256, + NID_sha256, + NID_sha3_256, + NID_sha384, + NID_sha3_384, + NID_blake2b512, + NID_sha512, + NID_sha3_512, +#endif + NID_undef +}; -#include <ouroboros/hash.h> -#include <ouroboros/random.h> +struct crypt_ctx { + struct keyrot * kr; /* backend-independent key rotation */ + void * cipher; /* backend AEAD cipher context */ +}; -#include <openssl/evp.h> -#include <openssl/ec.h> -#include <openssl/pem.h> +struct auth_ctx { + void * store; /* trusted anchors */ + void * chain; /* untrusted build-only interm */ +}; -#include <openssl/bio.h> +static int parse_kex_value(const char * value, + struct sec_config * cfg) +{ + SET_KEX_ALGO(cfg, value); + if (cfg->x.nid == NID_undef) + return -ENOTSUP; -#define IVSZ 16 -/* SYMMKEYSZ defined in dev.c */ + return 0; +} -/* - * Derive the common secret from - * your public key pair (kp) - * the remote public key (pub). - * Store it in a preallocated buffer (s). - */ -static int __openssl_ecdh_derive_secret(EVP_PKEY * kp, - EVP_PKEY * pub, - uint8_t * s) +/* not in header, but non-static for unit testing */ +int parse_sec_config(struct sec_config * cfg, + FILE * fp) { - EVP_PKEY_CTX * ctx; - int ret; - uint8_t * secret; - size_t secret_len; + char line[256]; + char * equals; + char * key; + char * value; + bool no_enc = false; + + assert(cfg != NULL); + assert(fp != NULL); + + /* Set defaults */ + SET_KEX_ALGO_NID(cfg, NID_X9_62_prime256v1); + cfg->x.mode = KEM_MODE_SERVER_ENCAP; + SET_KEX_KDF_NID(cfg, NID_sha256); + SET_KEX_CIPHER_NID(cfg, NID_aes_256_gcm); + SET_KEX_DIGEST_NID(cfg, NID_sha256); + /* a.req is seeded per-role by the caller; only auth= overrides it */ + + while (fgets(line, sizeof(line), fp) != NULL) { + char * trimmed; + + /* Skip comments and empty lines */ + if (line[0] == '#' || line[0] == '\n') + continue; + + /* Bare 'none' keyword replaced by encryption=none */ + trimmed = trim_whitespace(line); + if (strcmp(trimmed, "none") == 0) + return -EINVAL; + + /* Find the = separator */ + equals = strchr(line, '='); + if (equals == NULL) + continue; + + /* Split into key and value */ + *equals = '\0'; + key = trim_whitespace(line); + value = trim_whitespace(equals + 1); + + /* Parse key exchange field */ + if (strcmp(key, "kex") == 0) { + if (parse_kex_value(value, cfg) < 0) + return -EINVAL; + } else if (strcmp(key, "cipher") == 0) { + SET_KEX_CIPHER(cfg, value); + if (cfg->c.nid == NID_undef) + return -EINVAL; + } else if (strcmp(key, "kdf") == 0) { + SET_KEX_KDF(cfg, value); + if (cfg->k.nid == NID_undef) + return -EINVAL; + } else if (strcmp(key, "digest") == 0) { + SET_KEX_DIGEST(cfg, value); + if (cfg->d.nid == NID_undef) + return -EINVAL; + } else if (strcmp(key, "kem_mode") == 0) { + if (strcmp(value, "server") == 0) { + cfg->x.mode = KEM_MODE_SERVER_ENCAP; + } else if (strcmp(value, "client") == 0) { + cfg->x.mode = KEM_MODE_CLIENT_ENCAP; + } else { + return -EINVAL; + } + } else if (strcmp(key, "auth") == 0) { + if (strcmp(value, "required") == 0) { + cfg->a.req = true; + } else if (strcmp(value, "optional") == 0) { + cfg->a.req = false; + } else { + return -EINVAL; + } + } else if (strcmp(key, "cacert") == 0) { + if (strlen(value) >= sizeof(cfg->a.cacert)) + return -EINVAL; + strcpy(cfg->a.cacert, value); + } else if (strcmp(key, "encryption") == 0) { + if (strcmp(value, "none") != 0) + return -EINVAL; + no_enc = true; + } else { + return -EINVAL; + } + } - ctx = EVP_PKEY_CTX_new(kp, NULL); - if (ctx == NULL) - goto fail_new; + if (no_enc) { + /* Digest stays: it belongs to the auth axis */ + CLEAR_KEX_ALGO(cfg); + CLEAR_KEX_KDF(cfg); + CLEAR_KEX_CIPHER(cfg); + } - ret = EVP_PKEY_derive_init(ctx); - if (ret != 1) - goto fail_ctx; + return 0; +} - ret = EVP_PKEY_derive_set_peer(ctx, pub); - if (ret != 1) - goto fail_ctx; +/* Parse key exchange config from file */ +int load_sec_config_file(struct sec_config * cfg, + const char * path) +{ + FILE * fp; + int ret; + + assert(cfg != NULL); + assert(path != NULL); + + fp = fopen(path, "r"); + if (fp == NULL) { + /* Absent config disables encryption; other errors fail */ + if (errno == ENOENT) { + CLEAR_KEX_ALGO(cfg); + return 0; + } + return -errno; + } - ret = EVP_PKEY_derive(ctx, NULL, &secret_len); - if (ret != 1) - goto fail_ctx; + pthread_cleanup_push(__cleanup_fclose, fp); + ret = parse_sec_config(cfg, fp); + pthread_cleanup_pop(0); - if (secret_len < SYMMKEYSZ) - goto fail_ctx; + fclose(fp); - secret = OPENSSL_malloc(secret_len); - if (secret == NULL) - goto fail_ctx; + return ret; +} - ret = EVP_PKEY_derive(ctx, secret, &secret_len); - if (ret != 1) - goto fail_derive; +int kex_pkp_create(struct sec_config * cfg, + void ** pkp, + uint8_t * pk) +{ +#ifdef HAVE_OPENSSL + assert(cfg != NULL); + assert(pkp != NULL); - /* Hash the secret for use as AES key. */ - mem_hash(HASH_SHA3_256, s, secret, secret_len); + *pkp = NULL; - OPENSSL_free(secret); - EVP_PKEY_CTX_free(ctx); + if (cfg->x.str == NULL || kex_validate_nid(cfg->x.nid) < 0) + return -ENOTSUP; + + return openssl_pkp_create(cfg->x.str, (EVP_PKEY **) pkp, pk); +#else + (void) cfg; + (void) pkp; + (void) pk; + + *pkp = NULL; return 0; +#endif +} + +void kex_pkp_destroy(void * pkp) +{ + if (pkp == NULL) + return; +#ifdef HAVE_OPENSSL + openssl_pkp_destroy((EVP_PKEY *) pkp); +#else + (void) pkp; + + return; +#endif +} + +int kex_dhe_derive(struct sec_config * cfg, + void * pkp, + buffer_t pk, + uint8_t * s) +{ + assert(cfg != NULL); + + if (kex_validate_nid(cfg->x.nid) < 0) + return -ENOTSUP; + +#ifdef HAVE_OPENSSL + return openssl_dhe_derive((EVP_PKEY *) pkp, pk, cfg->k.nid, s); +#else + (void) pkp; + (void) pk; + + memset(s, 0, SYMMKEYSZ); - fail_derive: - OPENSSL_free(secret); - fail_ctx: - EVP_PKEY_CTX_free(ctx); - fail_new: return -ECRYPT; +#endif } -static int __openssl_ecdh_gen_key(void ** kp) +ssize_t kex_kem_encap(buffer_t pk, + uint8_t * ct, + int kdf, + uint8_t * s) { - EVP_PKEY_CTX * ctx = NULL; - EVP_PKEY_CTX * kctx = NULL; - EVP_PKEY * params = NULL; - int ret; +#ifdef HAVE_OPENSSL + return openssl_kem_encap(pk, ct, kdf, s); +#else + (void) pk; + (void) ct; + (void) kdf; - ctx = EVP_PKEY_CTX_new_id(EVP_PKEY_EC, NULL); - if (ctx == NULL) - goto fail_new_id; + memset(s, 0, SYMMKEYSZ); - ret = EVP_PKEY_paramgen_init(ctx); - if (ret != 1) - goto fail_paramgen; + return -ECRYPT; +#endif +} - ret = EVP_PKEY_CTX_set_ec_paramgen_curve_nid(ctx, NID_X9_62_prime256v1); - if (ret != 1) - goto fail_paramgen; +ssize_t kex_kem_encap_raw(buffer_t pk, + uint8_t * ct, + int kdf, + uint8_t * s) +{ +#ifdef HAVE_OPENSSL + return openssl_kem_encap_raw(pk, ct, kdf, s); +#else + (void) pk; + (void) ct; + (void) kdf; - ret = EVP_PKEY_paramgen(ctx, ¶ms); - if (ret != 1) - goto fail_paramgen; + memset(s, 0, SYMMKEYSZ); - kctx = EVP_PKEY_CTX_new(params, NULL); - if (kctx == NULL) - goto fail_keygen_init; + return -ECRYPT; +#endif +} - ret = EVP_PKEY_keygen_init(kctx); - if (ret != 1) - goto fail_keygen; +int kex_kem_decap(void * pkp, + buffer_t ct, + int kdf, + uint8_t * s) +{ +#ifdef HAVE_OPENSSL + return openssl_kem_decap((EVP_PKEY *) pkp, ct, kdf, s); +#else + (void) pkp; + (void) ct; + (void) kdf; - ret = EVP_PKEY_keygen(kctx, (EVP_PKEY **) kp); - if (ret != 1) - goto fail_keygen; + memset(s, 0, SYMMKEYSZ); - EVP_PKEY_free(params); - EVP_PKEY_CTX_free(kctx); - EVP_PKEY_CTX_free(ctx); + return -ECRYPT; +#endif +} - return 0; +int kex_get_algo_from_pk_der(buffer_t pk, + char * algo) +{ +#ifdef HAVE_OPENSSL + return openssl_get_algo_from_pk_der(pk, algo); +#else + (void) pk; + algo[0] = '\0'; - fail_keygen: - EVP_PKEY_CTX_free(kctx); - fail_keygen_init: - EVP_PKEY_free(params); - fail_paramgen: - EVP_PKEY_CTX_free(ctx); - fail_new_id: return -ECRYPT; +#endif } -static ssize_t openssl_ecdh_pkp_create(void ** pkp, - uint8_t * pk) +int kex_get_algo_from_pk_raw(buffer_t pk, + char * algo) { - uint8_t * pos; - ssize_t len; +#ifdef HAVE_OPENSSL + return openssl_get_algo_from_pk_raw(pk, algo); +#else + (void) pk; + algo[0] = '\0'; - assert(pkp != NULL); - assert(*pkp == NULL); - assert(pk != NULL); + return -ECRYPT; +#endif +} - if (__openssl_ecdh_gen_key(pkp) < 0) - return -ECRYPT; +int kex_validate_algo(const char * algo) +{ + if (algo == NULL) + return -EINVAL; - assert(*pkp != NULL); + /* Use NID validation instead of string array */ + return kex_validate_nid(kex_str_to_nid(algo)); +} - pos = pk; /* i2d_PUBKEY increments the pointer, don't use buf! */ - len = i2d_PUBKEY(*pkp, &pos); - if (len < 0) { - EVP_PKEY_free(*pkp); - return -ECRYPT; +int crypt_validate_nid(int nid) +{ + const struct nid_map * p; + + if (nid == NID_undef) + return -EINVAL; + + for (p = cipher_nid_map; p->name != NULL; p++) { + if (p->nid == nid) + return 0; } - return len; + return -ENOTSUP; } -static void openssl_ecdh_pkp_destroy(void * pkp) + +const char * crypt_nid_to_str(uint16_t nid) { - EVP_PKEY_free((EVP_PKEY *) pkp); + const struct nid_map * p; + + for (p = cipher_nid_map; p->name != NULL; p++) { + if (p->nid == nid) + return p->name; + } + + return NULL; } -static int openssl_ecdh_derive(void * pkp, - buffer_t pk, - uint8_t * s) +uint16_t crypt_str_to_nid(const char * cipher) { - uint8_t * pos; - EVP_PKEY * pub; + const struct nid_map * p; - pos = pk.data; /* d2i_PUBKEY increments the pointer, don't use key ptr! */ - pub = d2i_PUBKEY(NULL, (const uint8_t **) &pos, (long) pk.len); - if (pub == NULL) - return -ECRYPT; + if (cipher == NULL) + return NID_undef; - if (__openssl_ecdh_derive_secret(pkp, pub, s) < 0) { - EVP_PKEY_free(pub); - return -ECRYPT; + /* fast, check if cipher pointer is in the map */ + for (p = cipher_nid_map; p->name != NULL; p++) { + if (cipher == p->name) + return p->nid; } - EVP_PKEY_free(pub); + for (p = cipher_nid_map; p->name != NULL; p++) { + if (strcmp(p->name, cipher) == 0) + return p->nid; + } - return 0; + return NID_undef; } -/* - * AES encryption calls. If FRCT is disabled, we should generate a - * 128-bit random IV and append it to the packet. If the flow is - * reliable, we could initialize the context once, and consider the - * stream a single encrypted message to avoid initializing the - * encryption context for each packet. - */ +const char * kex_nid_to_str(uint16_t nid) +{ + const struct nid_map * p; + + for (p = kex_nid_map; p->name != NULL; p++) { + if (p->nid == nid) + return p->name; + } + + return NULL; +} -static int openssl_encrypt(void * ctx, - uint8_t * key, - struct shm_du_buff * sdb) +uint16_t kex_str_to_nid(const char * algo) { - uint8_t * out; - uint8_t * in; - uint8_t * head; - uint8_t iv[IVSZ]; - int in_sz; - int out_sz; - int tmp_sz; - int ret; + const struct nid_map * p; - in = shm_du_buff_head(sdb); - in_sz = shm_du_buff_tail(sdb) - in; + if (algo == NULL) + return NID_undef; - assert(in_sz > 0); + /* Fast path: check if algo pointer is in the map */ + for (p = kex_nid_map; p->name != NULL; p++) { + if (algo == p->name) + return p->nid; + } - if (random_buffer(iv, IVSZ) < 0) - goto fail_iv; + /* Slow path: string comparison */ + for (p = kex_nid_map; p->name != NULL; p++) { + if (strcmp(p->name, algo) == 0) + return p->nid; + } - out = malloc(in_sz + EVP_MAX_BLOCK_LENGTH); - if (out == NULL) - goto fail_iv; + return NID_undef; +} - EVP_CIPHER_CTX_reset(ctx); +int kex_validate_nid(int nid) +{ + const struct nid_map * p; - ret = EVP_EncryptInit_ex(ctx, EVP_aes_256_cbc(), NULL, key, iv); - if (ret != 1) - goto fail_encrypt_init; + if (nid == NID_undef) + return -EINVAL; - ret = EVP_EncryptUpdate(ctx, out, &tmp_sz, in, in_sz); - if (ret != 1) - goto fail_encrypt; + for (p = kex_nid_map; p->name != NULL; p++) { + if (p->nid == nid) + return 0; + } - out_sz = tmp_sz; - ret = EVP_EncryptFinal_ex(ctx, out + tmp_sz, &tmp_sz); - if (ret != 1) - goto fail_encrypt; + return -ENOTSUP; +} - out_sz += tmp_sz; +const char * md_nid_to_str(uint16_t nid) +{ + const struct nid_map * p; + + for (p = md_nid_map; p->name != NULL; p++) { + if (p->nid == nid) + return p->name; + } + + return NULL; +} + +uint16_t md_str_to_nid(const char * kdf) +{ + const struct nid_map * p; + + if (kdf == NULL) + return NID_undef; + + /* Fast path: check if kdf pointer is in the map */ + for (p = md_nid_map; p->name != NULL; p++) { + if (kdf == p->name) + return p->nid; + } + + /* Slow path: string comparison */ + for (p = md_nid_map; p->name != NULL; p++) { + if (strcmp(p->name, kdf) == 0) + return p->nid; + } + + return NID_undef; +} + +int md_validate_nid(int nid) +{ + const struct nid_map * p; + + if (nid == NID_undef) + return -EINVAL; + + for (p = md_nid_map; p->name != NULL; p++) { + if (p->nid == nid) + return 0; + } + + return -ENOTSUP; +} + +int crypt_cipher_rank(int nid) +{ + int i; + + if (nid == NID_undef) + return 0; + + for (i = 0; crypt_supported_nids[i] != NID_undef; i++) { + if ((int) crypt_supported_nids[i] == nid) + return i + 1; + } + + return -1; +} + +int crypt_kdf_rank(int nid) +{ + int i; + + if (nid == NID_undef) + return 0; + + for (i = 0; md_supported_nids[i] != NID_undef; i++) { + if ((int) md_supported_nids[i] == nid) + return i + 1; + } + + return -1; +} + +int crypt_kex_rank(int nid) +{ + int i; + + if (nid == NID_undef) + return 0; + + for (i = 0; kex_supported_nids[i] != NID_undef; i++) { + if ((int) kex_supported_nids[i] == nid) + return i + 1; + } + + return -1; +} + +/* AEAD primitive: 1:1 backend wrappers used by the data path below. */ +static int crypt_seal(void * cipher, + const uint8_t * key, + const uint8_t * nonce, + buffer_t aad, + buffer_t in, + uint8_t * out, + uint8_t * tag) +{ +#ifdef HAVE_OPENSSL + return openssl_seal(cipher, key, nonce, aad, in, out, tag); +#else + (void) cipher; + (void) key; + (void) nonce; + (void) aad; + (void) in; + (void) out; + (void) tag; + + return -ECRYPT; +#endif +} + +static int crypt_open(void * cipher, + const uint8_t * key, + const uint8_t * nonce, + buffer_t aad, + buffer_t in, + const uint8_t * tag, + buffer_t * out) +{ +#ifdef HAVE_OPENSSL + return openssl_open(cipher, key, nonce, aad, in, tag, out); +#else + (void) cipher; + (void) key; + (void) nonce; + (void) aad; + (void) in; + (void) tag; + (void) out; + + return -ECRYPT; +#endif +} + +int crypt_oneshot_seal(int nid, + const uint8_t * key, + const uint8_t * nonce, + buffer_t aad, + buffer_t in, + buffer_t * out) +{ + assert(key != NULL); + assert(nonce != NULL); + assert(out != NULL); + +#ifdef HAVE_OPENSSL + return openssl_oneshot_seal(nid, key, nonce, aad, in, out); +#else + (void) nid; + (void) key; + (void) nonce; + (void) aad; + (void) in; + (void) out; - EVP_CIPHER_CTX_cleanup(ctx); + return -ECRYPT; +#endif +} - assert(out_sz >= in_sz); +int crypt_oneshot_open(int nid, + const uint8_t * key, + const uint8_t * nonce, + buffer_t aad, + buffer_t in, + buffer_t * out) +{ + assert(key != NULL); + assert(nonce != NULL); + assert(out != NULL); - head = shm_du_buff_head_alloc(sdb, IVSZ); - if (head == NULL) +#ifdef HAVE_OPENSSL + return openssl_oneshot_open(nid, key, nonce, aad, in, out); +#else + (void) nid; + (void) key; + (void) nonce; + (void) aad; + (void) in; + (void) out; + + return -ECRYPT; +#endif +} + +/* + * Data-path encrypt: rotate the key, frame selector ‖ ct ‖ tag, seal. + * Backend-agnostic: composed from keyrot_*, crypt_seal and crypt_get_tagsz. + */ +int crypt_encrypt(struct crypt_ctx * ctx, + buffer_t in, + buffer_t * out) +{ + uint8_t nonce[KR_NONCE_LEN]; + const uint8_t * key; + uint8_t * ct; + buffer_t aad; + int tagsz; + int out_sz; + + assert(ctx != NULL); + assert(ctx->kr != NULL); + + tagsz = crypt_get_tagsz(ctx); + if (tagsz < 0) + return -ECRYPT; + + out->data = malloc(KR_SELECTOR_LEN + in.len + (size_t) tagsz); + if (out->data == NULL) + goto fail_malloc; + + ct = out->data + KR_SELECTOR_LEN; + + /* keyrot writes the selector into the wire header (== AAD). */ + if (keyrot_tx_next(ctx->kr, out->data, &key, nonce) != 0) goto fail_encrypt; - if (shm_du_buff_tail_alloc(sdb, out_sz - in_sz) == NULL) - goto fail_tail_alloc; + aad.data = out->data; + aad.len = KR_SELECTOR_LEN; - memcpy(head, iv, IVSZ); - memcpy(in, out, out_sz); + out_sz = crypt_seal(ctx->cipher, key, nonce, aad, in, ct, ct + in.len); + if (out_sz < 0) + goto fail_encrypt; - free(out); + out->len = KR_SELECTOR_LEN + (size_t) out_sz + (size_t) tagsz; return 0; - - fail_tail_alloc: - shm_du_buff_head_release(sdb, IVSZ); fail_encrypt: - EVP_CIPHER_CTX_cleanup(ctx); - fail_encrypt_init: - free(out); - fail_iv: + free(out->data); + fail_malloc: + clrbuf(*out); return -ECRYPT; } -static int openssl_decrypt(void * ctx, - uint8_t * key, - struct shm_du_buff * sdb) +/* + * Data-path decrypt: look up the rotated key from the selector, open, and + * commit the replay window only after the tag verifies. + */ +int crypt_decrypt(struct crypt_ctx * ctx, + buffer_t in, + buffer_t * out) { - uint8_t * in; - uint8_t * out; - uint8_t iv[IVSZ]; - int ret; - int out_sz; - int in_sz; - int tmp_sz; - - in_sz = shm_du_buff_len(sdb); - if (in_sz < IVSZ) + uint8_t nonce[KR_NONCE_LEN]; + const uint8_t * key; + const uint8_t * tag; + struct kr_rx rx; + buffer_t aad; + buffer_t ct; + int tagsz; + int in_sz; + + assert(ctx != NULL); + assert(ctx->kr != NULL); + + tagsz = crypt_get_tagsz(ctx); + if (tagsz < 0) return -ECRYPT; - in = shm_du_buff_head_release(sdb, IVSZ); + if (in.len < (size_t) (KR_SELECTOR_LEN + tagsz)) + return -ECRYPT; - memcpy(iv, in, IVSZ); + if (keyrot_rx_lookup(ctx->kr, in.data, &key, nonce, &rx) != 0) + return -ECRYPT; - in = shm_du_buff_head(sdb); - in_sz = shm_du_buff_tail(sdb) - in; + in_sz = (int) in.len - KR_SELECTOR_LEN - tagsz; - out = malloc(in_sz); - if (out == NULL) + /* +1 keeps malloc(0) defined for an empty (zero-length) frame. */ + out->data = malloc((size_t) in_sz + 1); + if (out->data == NULL) goto fail_malloc; - EVP_CIPHER_CTX_reset(ctx); + aad.data = in.data; + aad.len = KR_SELECTOR_LEN; - ret = EVP_DecryptInit_ex(ctx, EVP_aes_256_cbc(), NULL, key, iv); - if (ret != 1) - goto fail_decrypt_init; + ct.data = in.data + KR_SELECTOR_LEN; + ct.len = (size_t) in_sz; - ret = EVP_DecryptUpdate(ctx, out, &tmp_sz, in, in_sz); - if (ret != 1) - goto fail_decrypt; + tag = in.data + KR_SELECTOR_LEN + in_sz; - out_sz = tmp_sz; + if (crypt_open(ctx->cipher, key, nonce, aad, ct, tag, out) < 0) + goto fail_decrypt; - ret = EVP_DecryptFinal_ex(ctx, out + tmp_sz, &tmp_sz); - if (ret != 1) + /* Commit replay state only after the tag verifies. */ + if (keyrot_rx_commit(ctx->kr, &rx) != 0) goto fail_decrypt; - out_sz += tmp_sz; + return 0; + fail_decrypt: + free(out->data); + fail_malloc: + clrbuf(*out); + return -ECRYPT; +} + +struct crypt_ctx * crypt_create_ctx(struct crypt_sk * sk) +{ + struct crypt_ctx * crypt; - assert(out_sz <= in_sz); + if (crypt_validate_nid(sk->nid) != 0) + return NULL; - shm_du_buff_tail_release(sdb, in_sz - out_sz); + crypt = malloc(sizeof(*crypt)); + if (crypt == NULL) + goto fail_crypt; - memcpy(in, out, out_sz); + memset(crypt, 0, sizeof(*crypt)); - free(out); + crypt->kr = keyrot_create(sk->key, sk->epoch, sk->role); + if (crypt->kr == NULL) + goto fail_kr; + +#ifdef HAVE_OPENSSL + crypt->cipher = openssl_crypt_create_ctx(sk); + if (crypt->cipher == NULL) + goto fail_cipher; +#endif + return crypt; + +#ifdef HAVE_OPENSSL + fail_cipher: + keyrot_destroy(crypt->kr); +#endif + fail_kr: + free(crypt); + fail_crypt: + return NULL; +} + +void crypt_destroy_ctx(struct crypt_ctx * crypt) +{ + if (crypt == NULL) + return; + + keyrot_destroy(crypt->kr); +#ifdef HAVE_OPENSSL + openssl_crypt_destroy_ctx(crypt->cipher); +#endif + free(crypt); +} + +int crypt_get_headsz(struct crypt_ctx * ctx) +{ + assert(ctx != NULL); + assert(ctx->kr != NULL); + + (void) ctx; /* validated only; header size is a constant */ + + return KR_SELECTOR_LEN; +} + +int crypt_rekey(struct crypt_ctx * ctx, + struct crypt_sk * sk) +{ + assert(ctx != NULL); + assert(sk != NULL); + assert(ctx->kr != NULL); + + return keyrot_rekey(ctx->kr, sk->key, sk->epoch) == 0 ? 0 : -ECRYPT; +} + +int crypt_get_tagsz(struct crypt_ctx * ctx) +{ + assert(ctx != NULL); + assert(ctx->cipher != NULL); + +#ifdef HAVE_OPENSSL + return openssl_crypt_get_tagsz(ctx->cipher); +#else + (void) ctx; + return -ENOTSUP; +#endif +} + +int crypt_nodes_left(struct crypt_ctx * ctx) +{ + assert(ctx != NULL); + assert(ctx->kr != NULL); + + return (int) keyrot_tx_nodes_left(ctx->kr); +} + +int crypt_peer_synced(struct crypt_ctx * ctx) +{ + assert(ctx != NULL); + assert(ctx->kr != NULL); + + return keyrot_peer_switched(ctx->kr) ? 1 : 0; +} + +void crypt_tx_promote(struct crypt_ctx * ctx) +{ + assert(ctx != NULL); + assert(ctx->kr != NULL); + + keyrot_tx_promote(ctx->kr); +} + +int crypt_load_privkey_file(const char * path, + void ** key) +{ + *key = NULL; + +#ifdef HAVE_OPENSSL + return openssl_load_privkey_file(path, key); +#else + (void) path; return 0; +#endif +} - fail_decrypt: - EVP_CIPHER_CTX_cleanup(ctx); - fail_decrypt_init: - free(out); - fail_malloc: - return -ECRYPT; +int crypt_load_privkey_str(const char * str, + void ** key) +{ + *key = NULL; + +#ifdef HAVE_OPENSSL + return openssl_load_privkey_str(str, key); +#else + (void) str; + return 0; +#endif } -static int openssl_crypt_init(void ** ctx) +int crypt_load_pubkey_str(const char * str, + void ** key) { - *ctx = EVP_CIPHER_CTX_new(); - if (*ctx == NULL) - return -ECRYPT; + *key = NULL; + +#ifdef HAVE_OPENSSL + return openssl_load_pubkey_str(str, key); +#else + (void) str; return 0; +#endif } -static void openssl_crypt_fini(void * ctx) +int crypt_load_pubkey_file(const char * path, + void ** key) { - EVP_CIPHER_CTX_free(ctx); + *key = NULL; + +#ifdef HAVE_OPENSSL + return openssl_load_pubkey_file(path, key); +#else + (void) path; + + return 0; +#endif } -#endif /* HAVE_OPENSSL */ +int crypt_load_pubkey_file_to_der(const char * path, + buffer_t * buf) +{ + assert(buf != NULL); + +#ifdef HAVE_OPENSSL + return openssl_load_pubkey_file_to_der(path, buf); +#else + (void) path; -int crypt_dh_pkp_create(void ** pkp, - uint8_t * pk) + buf->data = NULL; + buf->len = 0; + return 0; +#endif +} + +int crypt_load_pubkey_raw_file(const char * path, + buffer_t * buf) { + assert(buf != NULL); + #ifdef HAVE_OPENSSL - assert(pkp != NULL); - *pkp = NULL; - return openssl_ecdh_pkp_create(pkp, pk); + return openssl_load_pubkey_raw_file(path, buf); #else - (void) pkp; - (void) pk; + (void) path; - *pkp = NULL; + buf->data = NULL; + buf->len = 0; + return 0; +#endif +} + +int crypt_load_privkey_raw_file(const char * path, + void ** key) +{ + *key = NULL; + +#ifdef HAVE_OPENSSL + return openssl_load_privkey_raw_file(path, key); +#else + (void) path; return 0; #endif } -void crypt_dh_pkp_destroy(void * pkp) +int crypt_ct_cmp(const void * a, + const void * b, + size_t len) { #ifdef HAVE_OPENSSL - openssl_ecdh_pkp_destroy(pkp); + return CRYPTO_memcmp(a, b, len); #else - (void) pkp; - return; + const volatile uint8_t * pa = a; + const volatile uint8_t * pb = b; + uint8_t d = 0; + size_t i; + + for (i = 0; i < len; i++) + d |= pa[i] ^ pb[i]; + + return d != 0; #endif } -int crypt_dh_derive(void * pkp, - buffer_t pk, - uint8_t * s) +int crypt_cmp_key(const void * key1, + const void * key2) { #ifdef HAVE_OPENSSL - return openssl_ecdh_derive(pkp, pk, s); + return openssl_cmp_key((const EVP_PKEY *) key1, + (const EVP_PKEY *) key2); #else - (void) pkp; - (void) pk; + (void) key1; + (void) key2; - memset(s, 0, SYMMKEYSZ); + return 0; +#endif +} - return -ECRYPT; +void crypt_free_key(void * key) +{ + if (key == NULL) + return; + +#ifdef HAVE_OPENSSL + openssl_free_key((EVP_PKEY *) key); #endif } -int crypt_encrypt(struct crypt_info * info, - struct shm_du_buff * sdb) +int crypt_load_crt_file(const char * path, + void ** crt) { - if (info->flags == 0) - return 0; + assert(crt != NULL); + + *crt = NULL; #ifdef HAVE_OPENSSL - return openssl_encrypt(info->ctx, info->key, sdb); + return openssl_load_crt_file(path, crt); #else - (void) sdb; + (void) path; return 0; #endif } -int crypt_decrypt(struct crypt_info * info, - struct shm_du_buff * sdb) +int crypt_load_crt_str(const char * str, + void ** crt) { - if (info->flags == 0) - return 0; + assert(crt != NULL); + + *crt = NULL; #ifdef HAVE_OPENSSL - return openssl_decrypt(info->ctx, info->key, sdb); + return openssl_load_crt_str(str, crt); #else - (void) sdb; + (void) str; - return -ECRYPT; + return 0; +#endif +} + +int crypt_load_crt_der(const buffer_t buf, + void ** crt) +{ + assert(crt != NULL); +#ifdef HAVE_OPENSSL + return openssl_load_crt_der(buf, crt); +#else + *crt = NULL; + + (void) buf; + + return 0; +#endif +} + +int crypt_get_pubkey_crt(void * crt, + void ** pk) +{ + assert(crt != NULL); + assert(pk != NULL); + +#ifdef HAVE_OPENSSL + return openssl_get_pubkey_crt(crt, pk); +#else + (void) crt; + + clrbuf(*pk); + + return 0; +#endif +} + +void crypt_free_crt(void * crt) +{ + if (crt == NULL) + return; +#ifdef HAVE_OPENSSL + openssl_free_crt(crt); +#endif +} + +int crypt_crt_str(const void * crt, + char * buf) +{ +#ifdef HAVE_OPENSSL + return openssl_crt_str(crt, buf); +#else + (void) crt; + (void) buf; + + return 0; +#endif +} + +int crypt_crt_der(const void * crt, + buffer_t * buf) +{ + assert(crt != NULL); + assert(buf != NULL); + +#ifdef HAVE_OPENSSL + return openssl_crt_der(crt, buf); +#else + (void) crt; + + clrbuf(*buf); + + return 0; +#endif +} + +int crypt_check_crt_name(void * crt, + const char * name) +{ +#ifdef HAVE_OPENSSL + return openssl_check_crt_name(crt, name); +#else + (void) crt; + (void) name; + + return 0; +#endif +} + +int crypt_get_crt_name(void * crt, + char * name) +{ +#ifdef HAVE_OPENSSL + return openssl_get_crt_name(crt, name); +#else + (void) crt; + (void) name; + + return 0; +#endif +} + +struct auth_ctx * auth_create_ctx(void) +{ + struct auth_ctx * ctx; + + ctx = malloc(sizeof(*ctx)); + if (ctx == NULL) + goto fail_malloc; + + memset(ctx, 0, sizeof(*ctx)); +#ifdef HAVE_OPENSSL + ctx->store = openssl_auth_create_store(); + if (ctx->store == NULL) + goto fail_store; + + ctx->chain = openssl_auth_create_chain(); + if (ctx->chain == NULL) + goto fail_chain; +#endif + return ctx; +#ifdef HAVE_OPENSSL + fail_chain: + openssl_auth_destroy_store(ctx->store); + fail_store: + free(ctx); +#endif + fail_malloc: + return NULL; +} + +void auth_destroy_ctx(struct auth_ctx * ctx) +{ + if (ctx == NULL) + return; +#ifdef HAVE_OPENSSL + openssl_auth_destroy_chain(ctx->chain); + openssl_auth_destroy_store(ctx->store); +#endif + free(ctx); +} + +int auth_add_crt_to_store(struct auth_ctx * ctx, + void * crt) +{ + assert(ctx != NULL); + assert(crt != NULL); + +#ifdef HAVE_OPENSSL + return openssl_auth_add_crt_to_store(ctx->store, crt); +#else + (void) ctx; + (void) crt; + + return 0; +#endif +} + +int auth_add_crt_to_chain(struct auth_ctx * ctx, + void * crt) +{ + assert(ctx != NULL); + assert(crt != NULL); + +#ifdef HAVE_OPENSSL + return openssl_auth_add_crt_to_chain(ctx->chain, crt); +#else + (void) ctx; + (void) crt; + + return 0; +#endif +} + +int auth_verify_crt(struct auth_ctx * ctx, + void * crt) +{ +#ifdef HAVE_OPENSSL + return openssl_verify_crt(ctx->store, ctx->chain, crt); +#else + (void) ctx; + (void) crt; + + return 0; +#endif +} + +int auth_verify_crt_pin(struct auth_ctx * ctx, + void * crt, + void * pin) +{ +#ifdef HAVE_OPENSSL + return openssl_verify_crt_pin(ctx->store, ctx->chain, crt, pin); +#else + (void) ctx; + (void) crt; + (void) pin; + + return 0; +#endif +} + +bool crypt_pk_requires_md(const void * pk) +{ +#ifdef HAVE_OPENSSL + return openssl_pk_requires_md((const EVP_PKEY *) pk); +#else + (void) pk; + + return false; #endif } -int crypt_init(struct crypt_info * info) +int auth_sign(void * pkp, + int md_nid, + buffer_t msg, + buffer_t * sig) { #ifdef HAVE_OPENSSL - return openssl_crypt_init(&info->ctx); + return openssl_sign((EVP_PKEY *) pkp, md_nid, msg, sig); #else - info->ctx = NULL; + (void) pkp; + (void) md_nid; + (void) msg; + (void) sig; + + clrbuf(*sig); + return 0; #endif } -void crypt_fini(struct crypt_info * info) +int auth_verify_sig(void * pk, + int md_nid, + buffer_t msg, + buffer_t sig) { #ifdef HAVE_OPENSSL - openssl_crypt_fini(info->ctx); + return openssl_verify_sig((EVP_PKEY *) pk, md_nid, msg, sig); #else + (void) pk; + (void) md_nid; + (void) msg; + (void) sig; + + return 0; +#endif +} + +ssize_t md_digest(int md_nid, + buffer_t in, + uint8_t * out) +{ +#ifdef HAVE_OPENSSL + return openssl_md_digest(md_nid, in, out); +#else + (void) md_nid; + (void) in; + (void) out; + + return -1; +#endif +} + +ssize_t md_len(int md_nid) +{ +#ifdef HAVE_OPENSSL + return openssl_md_len(md_nid); +#else + (void) md_nid; + return -1; +#endif +} + +int crypt_hkdf_expand(buffer_t key, + buffer_t info, + buffer_t out) +{ +#ifdef HAVE_OPENSSL + return openssl_hkdf_expand(key, info, out) == 0 ? 0 : -ECRYPT; +#else + (void) key; (void) info; - assert(info->ctx == NULL); + (void) out; + + return -ECRYPT; +#endif +} + +int crypt_secure_malloc_init(size_t max) +{ +#ifdef HAVE_OPENSSL + return openssl_secure_malloc_init(max, SECMEM_MINSIZE); +#else + (void) max; + return 0; +#endif +} + +void crypt_secure_malloc_fini(void) +{ +#ifdef HAVE_OPENSSL + openssl_secure_malloc_fini(); +#endif +} + +void crypt_cleanup(void) +{ +#ifdef HAVE_OPENSSL + openssl_cleanup(); +#endif +} + +void * crypt_secure_malloc(size_t size) +{ +#ifdef HAVE_OPENSSL + return openssl_secure_malloc(size); +#else + return malloc(size); +#endif +} + +void crypt_secure_free(void * ptr, + size_t size) +{ + if (ptr == NULL) + return; + +#ifdef HAVE_OPENSSL + openssl_secure_free(ptr, size); +#else + memset(ptr, 0, size); + free(ptr); +#endif +} + +void crypt_secure_clear(void * ptr, + size_t size) +{ + volatile uint8_t * p; + + if (ptr == NULL) + return; + +#ifdef HAVE_OPENSSL + (void) p; + openssl_secure_clear(ptr, size); +#elif defined(HAVE_EXPLICIT_BZERO) + (void) p; + explicit_bzero(ptr, size); +#else /* best effort to avoid optimizing out */ + p = ptr; + while (size-- > 0) + *p++ = 0; #endif } diff --git a/src/lib/crypt/keyrot.c b/src/lib/crypt/keyrot.c new file mode 100644 index 00000000..8b0d9429 --- /dev/null +++ b/src/lib/crypt/keyrot.c @@ -0,0 +1,741 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Data-plane key-rotation schedule (node/leaf keys, selector) + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This library is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public License + * version 2.1 as published by the Free Software Foundation. + * + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with this library; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#define _POSIX_C_SOURCE 200809L + +#include <config.h> + +#include <ouroboros/atomics.h> +#include <ouroboros/crypt.h> +#include <ouroboros/pthread.h> +#include <ouroboros/rcu.h> + +#include "crypt/keyrot.h" + +#include <assert.h> +#include <stdbool.h> +#include <stdlib.h> +#include <string.h> + +/* + * Per-flow keys are addressed by (epoch, node, leaf) and derived as: + * root = per-batch HKDF PRK from the OAP exchange, wiped once expanded + * nodes = HKDF-Expand(root, "o7s-keyrot-node") -> KEY_NODE_COUNT keys + * leaf = HKDF-Expand(node, "o7s-keyrot-leaf"|dir|leaf) -> AEAD key + * The epoch is a small wrapping counter, carried in the selector, that picks + * the live batch; a Tier-2 OAP re-key advances it. The "dir" byte forks the + * leaf keys per direction. + * + * Concurrency: cur/prev batch pointers are published by a re-key and read on + * the data path under an rcu_guard (lock-free RCU with liburcu, else a per- + * keyrot rwlock). The per-batch TX counter is atomic, so the (epoch, counter) + * nonce is unique without serialising TX. Leaf caches are THREAD-LOCAL (an app + * writer and the FRCT retransmit timer never share cache state), keyed on a + * global batch id and direct-mapped. + */ + +#define KR_WITHIN_BITS (KEY_LEAF_BITS + KEY_NODE_BITS) +#define KR_WITHIN_MASK (((uint64_t) 1 << KR_WITHIN_BITS) - 1) +#define KR_N (KEY_NODE_COUNT) +#define KR_LEAVES (1u << KEY_NODE_BITS) +#define KR_BATCH_MAX ((uint64_t) KR_N << KR_WITHIN_BITS) +#define KR_NODES_SZ ((size_t) KR_N * SYMMKEYSZ) +#define KR_TCACHE_WAYS 16 /* per-thread cache slots per direction (pow2) */ +#define KR_EPOCHS 16 /* 4-bit wire epoch: gens before wrap */ + +#define KR_RP_WORDS (KEY_REPLAY_WINDOW / 64) /* pow2; RFC 6479 bitmap */ +#define KR_RP_SHIFT 6 +#define KR_RP_MASK 63 +#define KR_RP_WINDOW (KEY_REPLAY_WINDOW - 64) /* reserve 1 slack word */ + +static const char kr_node_label[] = "o7s-keyrot-node"; +static const char kr_leaf_label[] = "o7s-keyrot-leaf"; + +struct kr_batch { + uint64_t id; /* process-global, unique; cache key (no ABA) */ + uint8_t epoch; /* 4-bit wire selector */ + uint8_t * nodes; /* KR_NODES_SZ in secure heap; NULL if empty */ + uint64_t tx_ctr; /* atomic; per-batch so nonces never collide */ + + struct { /* RFC 6479-like anti-replay window */ + uint64_t last; /* highest accepted ctr + 1 */ + uint64_t bits[KR_RP_WORDS]; + pthread_mutex_t mtx; + } rp; +}; + +struct kr_keycache { + uint8_t * key; /* SYMMKEYSZ, points into the per-thread slab */ + uint64_t id; /* batch the cached key belongs to */ + uint16_t node; + uint8_t leaf; + uint8_t dir; + bool valid; +}; + +struct keyrot { + struct kr_batch * cur; /* published; read on data path */ + struct kr_batch * prev; /* NULL = none */ + struct rcu_guard guard; /* re-key vs readers */ + uint8_t role; + uint8_t tx_epoch; /* epoch TX currently stamps */ + bool peer_switched; /* peer is on the cur epoch */ +}; + +/* Per-thread leaf-key caches, freed by the thread-exit destructor. */ +struct kr_tcache { + struct kr_keycache tx[KR_TCACHE_WAYS]; + struct kr_keycache rx[KR_TCACHE_WAYS]; + uint8_t * slab; /* 2*KR_TCACHE_WAYS*SYMMKEYSZ secure heap */ +}; + +static struct { + uint64_t next_id; /* batch-id allocator (atomic) */ + pthread_key_t tcache_key; /* per-thread leaf-key caches */ + pthread_once_t tcache_once; +} kr_g = { 0, 0, PTHREAD_ONCE_INIT }; + +static void kr_tcache_free(void * p) +{ + struct kr_tcache * t = p; + + if (t == NULL) + return; + + crypt_secure_free(t->slab, 2 * KR_TCACHE_WAYS * SYMMKEYSZ); + free(t); +} + +static void kr_tcache_init(void) +{ + pthread_key_create(&kr_g.tcache_key, kr_tcache_free); +} + +static struct kr_tcache * kr_tcache_get(void) +{ + struct kr_tcache * t; + size_t i; + + pthread_once(&kr_g.tcache_once, kr_tcache_init); + + t = pthread_getspecific(kr_g.tcache_key); + if (t != NULL) + return t; + + t = malloc(sizeof(*t)); + if (t == NULL) + goto fail_alloc; + + memset(t, 0, sizeof(*t)); + + t->slab = crypt_secure_malloc(2 * KR_TCACHE_WAYS * SYMMKEYSZ); + if (t->slab == NULL) + goto fail_slab; + + for (i = 0; i < KR_TCACHE_WAYS; i++) { + t->tx[i].key = t->slab + i * SYMMKEYSZ; + t->rx[i].key = t->slab + (KR_TCACHE_WAYS + i) * SYMMKEYSZ; + } + + if (pthread_setspecific(kr_g.tcache_key, t) != 0) + goto fail_set; + + return t; + + fail_set: + crypt_secure_free(t->slab, 2 * KR_TCACHE_WAYS * SYMMKEYSZ); + fail_slab: + free(t); + fail_alloc: + return NULL; +} + +static uint8_t * kr_expand_nodes(const uint8_t * root) +{ + uint8_t * nodes; + buffer_t prk; + buffer_t info; + buffer_t okm; + + nodes = crypt_secure_malloc(KR_NODES_SZ); + if (nodes == NULL) + return NULL; + + prk.len = SYMMKEYSZ; + prk.data = (uint8_t *) root; + info.len = sizeof(kr_node_label) - 1; + info.data = (uint8_t *) kr_node_label; + okm.len = KR_NODES_SZ; + okm.data = nodes; + + if (crypt_hkdf_expand(prk, info, okm) != 0) + goto fail_expand; + + return nodes; + + fail_expand: + crypt_secure_free(nodes, KR_NODES_SZ); + return NULL; +} + +static int kr_leaf_key(const uint8_t * node, + uint8_t leaf, + uint8_t dir, + uint8_t * out) +{ + uint8_t info_buf[sizeof(kr_leaf_label) - 1 + 2]; + buffer_t prk; + buffer_t info; + buffer_t okm; + size_t n = sizeof(kr_leaf_label) - 1; + + memcpy(info_buf, kr_leaf_label, n); + info_buf[n] = dir; + info_buf[n + 1] = leaf; + + prk.len = SYMMKEYSZ; + prk.data = (uint8_t *) node; + info.len = n + 2; + info.data = info_buf; + okm.len = SYMMKEYSZ; + okm.data = out; + + return crypt_hkdf_expand(prk, info, okm); +} + +static __inline__ bool kr_kc_hit(const struct kr_keycache * kc, + const struct kr_batch * b, + uint16_t node, + uint8_t leaf, + uint8_t dir) +{ + if (!kc->valid) + return false; + + if (kc->id != b->id) + return false; + + if (kc->node != node) + return false; + + if (kc->leaf != leaf) + return false; + + return kc->dir == dir; +} + +/* Fetch the leaf key; derive into the (direct-mapped) slot on a miss. */ +static const uint8_t * kr_kc_get(struct kr_keycache * cache, + const struct kr_batch * b, + uint16_t node, + uint8_t leaf, + uint8_t dir) +{ + struct kr_keycache * kc; + uint8_t * nkey; + + kc = &cache[b->id & (KR_TCACHE_WAYS - 1)]; + + if (kr_kc_hit(kc, b, node, leaf, dir)) + return kc->key; + + nkey = b->nodes + (size_t) node * SYMMKEYSZ; + if (kr_leaf_key(nkey, leaf, dir, kc->key) != 0) + return NULL; + + kc->valid = true; + kc->id = b->id; + kc->node = node; + kc->leaf = leaf; + kc->dir = dir; + + return kc->key; +} + +static void kr_sel_enc(uint8_t epoch, + uint16_t node, + uint32_t seq, + uint8_t sel[KR_SELECTOR_LEN]) +{ + sel[0] = (uint8_t) ((epoch << 4) | ((node >> 8) & 0x0F)); + sel[1] = (uint8_t) (node & 0xFF); + sel[2] = (uint8_t) (seq >> 24); + sel[3] = (uint8_t) (seq >> 16); + sel[4] = (uint8_t) (seq >> 8); + sel[5] = (uint8_t) (seq); +} + +static void kr_sel_dec(const uint8_t sel[KR_SELECTOR_LEN], + uint8_t * epoch, + uint16_t * node, + uint32_t * seq) +{ + *epoch = (uint8_t) (sel[0] >> 4); + *node = (uint16_t) (((sel[0] & 0x0F) << 8) | sel[1]); + *seq = ((uint32_t) sel[2] << 24) | ((uint32_t) sel[3] << 16) | + ((uint32_t) sel[4] << 8) | (uint32_t) sel[5]; +} + +static uint64_t kr_ctr(uint16_t node, + uint32_t seq) +{ + return ((uint64_t) node << KR_WITHIN_BITS) | + ((uint64_t) seq & KR_WITHIN_MASK); +} + +static void kr_nonce(uint64_t ctr, + uint8_t * nonce) +{ + size_t i; + + memset(nonce, 0, KR_NONCE_LEN); + + /* ctr big-endian in the low 8 bytes; high bytes stay zero */ + for (i = 0; i < 8; i++) + nonce[i] = (uint8_t) (ctr >> (56 - 8 * i)); +} + +static struct kr_batch * kr_batch_create(uint8_t epoch, + const uint8_t * root) +{ + struct kr_batch * b; + + b = malloc(sizeof(*b)); + if (b == NULL) + goto fail_alloc; + + b->nodes = kr_expand_nodes(root); + if (b->nodes == NULL) + goto fail_nodes; + + b->id = FETCH_ADD_RELAXED(&kr_g.next_id, 1); + b->epoch = epoch; + b->tx_ctr = 0; + if (pthread_mutex_init(&b->rp.mtx, NULL) != 0) + goto fail_lock; + + b->rp.last = 0; + memset(b->rp.bits, 0, sizeof(b->rp.bits)); + + return b; + + fail_lock: + crypt_secure_free(b->nodes, KR_NODES_SZ); + free(b); + return NULL; + fail_nodes: + free(b); + fail_alloc: + return NULL; +} + +static void kr_batch_free(struct kr_batch * b) +{ + if (b == NULL) + return; + + pthread_mutex_destroy(&b->rp.mtx); + crypt_secure_free(b->nodes, KR_NODES_SZ); + free(b); +} + +/* + * RFC 6479 anti-replay window keyed on the per-batch counter, with + * seq = ctr + 1 so 0 means "nothing accepted yet". Returns 0 if the + * packet is fresh (and records it), -1 on a replay or a too-old ctr. + */ +static int kr_rp_commit(struct kr_batch * b, + uint64_t ctr) +{ + uint64_t seq; + uint64_t idx; + uint64_t cur; + uint64_t diff; + + seq = ctr + 1; + + pthread_mutex_lock(&b->rp.mtx); + + if (seq > b->rp.last) { + idx = seq >> KR_RP_SHIFT; + cur = b->rp.last >> KR_RP_SHIFT; + diff = idx - cur; + if (diff > KR_RP_WORDS) + diff = KR_RP_WORDS; + + while (diff-- > 0) { + cur++; + b->rp.bits[cur & (KR_RP_WORDS - 1)] = 0; + } + + b->rp.bits[idx & (KR_RP_WORDS - 1)] |= + (uint64_t) 1 << (seq & KR_RP_MASK); + b->rp.last = seq; + goto finish; + } + + if (b->rp.last - seq >= KR_RP_WINDOW) + goto fail; + + idx = seq >> KR_RP_SHIFT; + if (b->rp.bits[idx & (KR_RP_WORDS - 1)] + & ((uint64_t) 1 << (seq & KR_RP_MASK))) + goto fail; + + b->rp.bits[idx & (KR_RP_WORDS - 1)] |= + (uint64_t) 1 << (seq & KR_RP_MASK); + finish: + pthread_mutex_unlock(&b->rp.mtx); + + return 0; + fail: + pthread_mutex_unlock(&b->rp.mtx); + + return -1; +} + +struct keyrot * keyrot_create(const uint8_t * root, + uint8_t epoch, + uint8_t role) +{ + struct keyrot * kr; + + assert(root != NULL); + assert(role <= 1); + + if (epoch >= KR_EPOCHS) + goto fail_kr; + + kr = malloc(sizeof(*kr)); + if (kr == NULL) + goto fail_kr; + + memset(kr, 0, sizeof(*kr)); + + kr->role = role; + kr->tx_epoch = epoch; + kr->peer_switched = true; + kr->prev = NULL; + + kr->cur = kr_batch_create(epoch, root); + if (kr->cur == NULL) + goto fail_cur; + + if (rcu_guard_init(&kr->guard)) + goto fail_guard; + + return kr; + + fail_guard: + kr_batch_free(kr->cur); + fail_cur: + free(kr); + fail_kr: + return NULL; +} + +void keyrot_destroy(struct keyrot * kr) +{ + if (kr == NULL) + return; + + /* Wait out any in-flight reader before freeing batches. */ + rcu_drain(&kr->guard); + + kr_batch_free(kr->cur); + kr_batch_free(kr->prev); + + rcu_guard_fini(&kr->guard); + + free(kr); +} + +int keyrot_rekey(struct keyrot * kr, + const uint8_t * root, + uint8_t epoch) +{ + struct kr_batch * nb; + struct kr_batch * old_prev; + + assert(kr != NULL); + assert(root != NULL); + + if (epoch >= KR_EPOCHS) + return -1; + + nb = kr_batch_create(epoch, root); + if (nb == NULL) + return -1; + + rcu_wrlock(&kr->guard); + + old_prev = kr->prev; + rcu_assign(kr->prev, kr->cur); + rcu_publish(nb); + rcu_assign(kr->cur, nb); + + /* TX keeps the old epoch until the peer is seen on the new one. */ + STORE_RELEASE(&kr->peer_switched, false); + + rcu_wrunlock(&kr->guard); + + /* old_prev is unreachable now; reclaim past any live reader. */ + rcu_reclaim(&kr->guard); + kr_batch_free(old_prev); + + return 0; +} + +void keyrot_tx_promote(struct keyrot * kr) +{ + assert(kr != NULL); + + /* Serialise with keyrot_rekey so tx_epoch tracks a consistent cur. */ + rcu_wrlock(&kr->guard); + STORE_RELAXED(&kr->tx_epoch, rcu_deref(kr->cur)->epoch); + rcu_wrunlock(&kr->guard); +} + +int keyrot_tx_next(struct keyrot * kr, + uint8_t sel[KR_SELECTOR_LEN], + const uint8_t ** key, + uint8_t nonce[KR_NONCE_LEN]) +{ + struct kr_tcache * tc; + struct kr_batch * cur; + struct kr_batch * prev; + struct kr_batch * b; + uint64_t ctr; + uint16_t node; + uint8_t leaf; + uint8_t txe; + uint8_t epoch; + uint32_t seq; + const uint8_t * k; + + assert(kr != NULL); + assert(key != NULL); + + tc = kr_tcache_get(); + if (tc == NULL) + return -1; + + rcu_rdlock(&kr->guard); + + cur = rcu_deref(kr->cur); + prev = rcu_deref(kr->prev); + rcu_consume(cur); + rcu_consume(prev); + txe = LOAD_RELAXED(&kr->tx_epoch); + + if (cur->epoch == txe) + b = cur; + else if (prev != NULL && prev->epoch == txe) + b = prev; + else + b = NULL; + + if (b == NULL) { + rcu_rdunlock(&kr->guard); + return -1; /* tx_epoch batch gone; next promote resyncs */ + } + + /* Slot reserved even if exhausted; tx_nodes_left clamps the count. */ + ctr = FETCH_ADD_RELAXED(&b->tx_ctr, 1); + if (ctr >= KR_BATCH_MAX) { + rcu_rdunlock(&kr->guard); + return -1; /* batch exhausted */ + } + + node = (uint16_t) (ctr >> KR_WITHIN_BITS); + leaf = (uint8_t) ((ctr >> KEY_LEAF_BITS) & (KR_LEAVES - 1)); + seq = (uint32_t) (ctr & KR_WITHIN_MASK); + epoch = b->epoch; + + k = kr_kc_get(tc->tx, b, node, leaf, kr->role); + + rcu_rdunlock(&kr->guard); + + if (k == NULL) + return -1; + + kr_sel_enc(epoch, node, seq, sel); + kr_nonce(ctr, nonce); + + *key = k; + + return 0; +} + +int keyrot_rx_lookup(struct keyrot * kr, + const uint8_t sel[KR_SELECTOR_LEN], + const uint8_t ** key, + uint8_t nonce[KR_NONCE_LEN], + struct kr_rx * rx) +{ + struct kr_tcache * tc; + struct kr_batch * cur; + struct kr_batch * prev; + struct kr_batch * b; + uint8_t epoch; + uint16_t node; + uint32_t seq; + uint64_t ctr; + uint8_t leaf; + const uint8_t * k; + + assert(kr != NULL); + assert(key != NULL); + + kr_sel_dec(sel, &epoch, &node, &seq); + + if (node >= KR_N) + return -1; + + tc = kr_tcache_get(); + if (tc == NULL) + return -1; + + rcu_rdlock(&kr->guard); + + cur = rcu_deref(kr->cur); + prev = rcu_deref(kr->prev); + rcu_consume(cur); + rcu_consume(prev); + + if (epoch == cur->epoch) { + b = cur; + } else if (prev != NULL && epoch == prev->epoch) { + b = prev; + } else { + rcu_rdunlock(&kr->guard); + return -1; /* unknown epoch */ + } + + ctr = kr_ctr(node, seq); + leaf = (uint8_t) ((ctr >> KEY_LEAF_BITS) & (KR_LEAVES - 1)); + + /* peer's tx direction */ + k = kr_kc_get(tc->rx, b, node, leaf, (uint8_t) (kr->role ^ 1)); + + rx->id = b->id; + rx->ctr = ctr; + + rcu_rdunlock(&kr->guard); + + if (k == NULL) + return -1; + + kr_nonce(ctr, nonce); + + *key = k; + + return 0; +} + +/* + * Commit a packet that authenticated under the batch keyrot_rx_lookup + * selected. Re-finds that batch by id (epoch may have advanced) and, + * if still resident, advances the replay window and records that the + * peer is on the current batch. Runs only post-AEAD so a forged or + * replayed packet can mutate no receiver state. Returns -1 on replay. + */ +int keyrot_rx_commit(struct keyrot * kr, + const struct kr_rx * rx) +{ + struct kr_batch * cur; + struct kr_batch * prev; + struct kr_batch * b; + int rc; + + assert(kr != NULL); + assert(rx != NULL); + + rcu_rdlock(&kr->guard); + + cur = rcu_deref(kr->cur); + prev = rcu_deref(kr->prev); + rcu_consume(cur); + rcu_consume(prev); + + if (cur->id == rx->id) + b = cur; + else if (prev != NULL && prev->id == rx->id) + b = prev; + else + b = NULL; + + if (b == NULL) { + rcu_rdunlock(&kr->guard); + return 0; /* batch evicted post-auth; nothing to protect */ + } + + rc = kr_rp_commit(b, rx->ctr); + if (rc == 0 && b == cur) + STORE_RELEASE(&kr->peer_switched, true); + + rcu_rdunlock(&kr->guard); + + return rc; +} + +bool keyrot_peer_switched(const struct keyrot * kr) +{ + assert(kr != NULL); + + return LOAD_ACQUIRE(&kr->peer_switched); +} + +unsigned keyrot_tx_nodes_left(struct keyrot * kr) +{ + struct kr_batch * cur; + struct kr_batch * prev; + struct kr_batch * b; + uint64_t ctr; + unsigned used; + uint8_t txe; + + assert(kr != NULL); + + rcu_rdlock(&kr->guard); + cur = rcu_deref(kr->cur); + prev = rcu_deref(kr->prev); + rcu_consume(cur); + rcu_consume(prev); + txe = LOAD_RELAXED(&kr->tx_epoch); + + if (cur->epoch == txe) + b = cur; + else if (prev != NULL && prev->epoch == txe) + b = prev; + else + b = NULL; + + ctr = b != NULL ? LOAD_RELAXED(&b->tx_ctr) : KR_BATCH_MAX; + rcu_rdunlock(&kr->guard); + + used = (unsigned) (ctr >> KR_WITHIN_BITS); + if (used >= KR_N) + return 0; + + return KR_N - used; +} diff --git a/src/lib/crypt/keyrot.h b/src/lib/crypt/keyrot.h new file mode 100644 index 00000000..6a598f76 --- /dev/null +++ b/src/lib/crypt/keyrot.h @@ -0,0 +1,74 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Data-plane key-rotation schedule (node/leaf keys, selector) + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This library is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public License + * version 2.1 as published by the Free Software Foundation. + * + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with this library; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#ifndef OUROBOROS_LIB_CRYPT_KEYROT_H +#define OUROBOROS_LIB_CRYPT_KEYROT_H + +#include <ouroboros/crypt.h> /* SYMMKEYSZ, NONCESZ */ + +#include <stdbool.h> +#include <stdint.h> + +#define KR_SELECTOR_LEN 6 +#define KR_NONCE_LEN NONCESZ + +struct keyrot; + +struct kr_rx { + uint64_t id; /* batch id of the matched epoch */ + uint64_t ctr; /* packet counter for replay check */ +}; + +struct keyrot * keyrot_create(const uint8_t * root, + uint8_t epoch, + uint8_t role); + +void keyrot_destroy(struct keyrot * kr); + +int keyrot_rekey(struct keyrot * kr, + const uint8_t * root, + uint8_t epoch); + +/* Promote TX to the installed (new) batch once the peer is on it. */ +void keyrot_tx_promote(struct keyrot * kr); + +int keyrot_tx_next(struct keyrot * kr, + uint8_t sel[KR_SELECTOR_LEN], + const uint8_t ** key, + uint8_t nonce[KR_NONCE_LEN]); + +int keyrot_rx_lookup(struct keyrot * kr, + const uint8_t sel[KR_SELECTOR_LEN], + const uint8_t ** key, + uint8_t nonce[KR_NONCE_LEN], + struct kr_rx * rx); + +/* Commit an authenticated packet: replay window + peer-switched. */ +int keyrot_rx_commit(struct keyrot * kr, + const struct kr_rx * rx); + +/* True once an RX packet under the current batch has been observed. */ +bool keyrot_peer_switched(const struct keyrot * kr); + +unsigned keyrot_tx_nodes_left(struct keyrot * kr); + +#endif /* OUROBOROS_LIB_CRYPT_KEYROT_H */ diff --git a/src/lib/crypt/openssl.c b/src/lib/crypt/openssl.c new file mode 100644 index 00000000..d5d9adf5 --- /dev/null +++ b/src/lib/crypt/openssl.c @@ -0,0 +1,1933 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * OpenSSL based cryptographic operations + * Elliptic curve Diffie-Hellman key exchange + * AES encryption + # Authentication + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This library is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public License + * version 2.1 as published by the Free Software Foundation. + * + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with this library; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#define _POSIX_C_SOURCE 200809L + +#include <config.h> + +#include <ouroboros/errno.h> +#include <ouroboros/crypt.h> +#include <ouroboros/hash.h> +#include <ouroboros/name.h> +#include <ouroboros/pthread.h> +#include <ouroboros/random.h> +#include <ouroboros/utils.h> + +#include <openssl/evp.h> +#include <openssl/bio.h> +#include <openssl/ec.h> +#include <openssl/err.h> +#include <openssl/kdf.h> +#include <openssl/pem.h> +#include <openssl/sha.h> +#include <openssl/provider.h> +#include <openssl/x509v3.h> +#include <openssl/x509_vfy.h> + +#include <assert.h> +#include <stdio.h> + +#define IS_EC_GROUP(str) (strcmp(str, "EC") == 0) +#define IS_DH_GROUP(str) (strcmp(str, "DH") == 0) + +#define HKDF_INFO_DHE "o7s-ossl-dhe" +#define HKDF_INFO_ENCAP "o7s-ossl-encap" +#define HKDF_SALT_LEN 32 /* SHA-256 output size */ +#define AEAD_NONCE_LEN 12 /* 96-bit deterministic IV (SP 800-38D) */ +#define AEAD_TAG_LEN 16 /* 128-bit AEAD authentication tag */ + +struct ossl_crypt_ctx { + EVP_CIPHER_CTX * evp_ctx; + const EVP_CIPHER * cipher; + int tagsz; +}; + +struct kdf_info { + buffer_t secret; + int nid; + buffer_t salt; + buffer_t info; + buffer_t key; +}; + +/* Convert hash NID to OpenSSL digest name string for HKDF */ +static const char * hash_nid_to_digest_name(int nid) +{ + const EVP_MD * md; + const char * name; + + md = EVP_get_digestbynid(nid); + if (md == NULL) + return NULL; + + name = EVP_MD_get0_name(md); + if (name == NULL) + return NULL; + + return name; +} + +/* Extract public key bytes from a key pair for salt derivation */ +static int get_pk_bytes_from_key(EVP_PKEY * key, + buffer_t * pk) +{ + const char * name; + int ret; + + assert(key != NULL); + assert(pk != NULL); + + name = EVP_PKEY_get0_type_name(key); + if (name == NULL) + goto fail_name; + + if (IS_HYBRID_KEM(name)) { + pk->len = EVP_PKEY_get1_encoded_public_key(key, &pk->data); + if (pk->len == 0) + goto fail_name; + } else { + /* Pure ML-KEM: use DER encoding to match encap */ + pk->data = NULL; + ret = i2d_PUBKEY(key, &pk->data); + if (ret <= 0) + goto fail_name; + pk->len = (size_t) ret; + } + + return 0; + fail_name: + return -ECRYPT; +} + +/* Derive salt from public key bytes by hashing them */ +static int derive_salt_from_pk_bytes(buffer_t pk, + buffer_t salt) +{ + uint8_t hash[EVP_MAX_MD_SIZE]; + unsigned hash_len; + + assert(pk.data != NULL); + assert(salt.data != NULL); + + if (EVP_Digest(pk.data, pk.len, hash, &hash_len, + EVP_sha256(), NULL) != 1) + goto fail_digest; + + memcpy(salt.data, hash, salt.len < hash_len ? salt.len : hash_len); + + return 0; + fail_digest: + return -ECRYPT; +} + +/* Derive salt from two public key byte buffers (DHE) in canonical order */ +static int derive_salt_from_pk_bytes_dhe(buffer_t local, + buffer_t remote, + buffer_t salt) +{ + uint8_t * concat; + size_t concat_len; + uint8_t hash[EVP_MAX_MD_SIZE]; + unsigned hash_len; + size_t min_len; + int cmp; + + assert(local.data != NULL); + assert(remote.data != NULL); + assert(salt.data != NULL); + + concat_len = local.len + remote.len; + concat = OPENSSL_malloc(concat_len); + if (concat == NULL) + goto fail_malloc; + + /* Canonical order: compare and concatenate smaller first */ + min_len = local.len < remote.len ? local.len : remote.len; + cmp = memcmp(local.data, remote.data, min_len); + if (cmp < 0 || (cmp == 0 && local.len < remote.len)) { + memcpy(concat, local.data, local.len); + memcpy(concat + local.len, remote.data, remote.len); + } else { + memcpy(concat, remote.data, remote.len); + memcpy(concat + remote.len, local.data, local.len); + } + + if (EVP_Digest(concat, concat_len, hash, &hash_len, + EVP_sha256(), NULL) != 1) + goto fail_digest; + + OPENSSL_free(concat); + + memcpy(salt.data, hash, salt.len < hash_len ? salt.len : hash_len); + + return 0; + fail_digest: + OPENSSL_free(concat); + fail_malloc: + return -ECRYPT; +} + +/* Derive key using HKDF */ +#define OPc_u_str OSSL_PARAM_construct_utf8_string +#define OPc_o_str OSSL_PARAM_construct_octet_string +static int derive_key_hkdf(struct kdf_info * ki) +{ + EVP_KDF * kdf; + EVP_KDF_CTX * kctx; + OSSL_PARAM params[5]; + const char * digest; + int idx; + + digest = hash_nid_to_digest_name(ki->nid); + if (digest == NULL) + goto fail_fetch; + + kdf = EVP_KDF_fetch(NULL, "HKDF", NULL); + if (kdf == NULL) + goto fail_fetch; + + kctx = EVP_KDF_CTX_new(kdf); + if (kctx == NULL) + goto fail_ctx; + + idx = 0; + params[idx++] = OPc_u_str("digest", (char *) digest, 0); + params[idx++] = OPc_o_str("key", ki->secret.data, ki->secret.len); + params[idx++] = OPc_o_str("salt", ki->salt.data, ki->salt.len); + params[idx++] = OPc_o_str("info", ki->info.data, ki->info.len); + + params[idx] = OSSL_PARAM_construct_end(); + + if (EVP_KDF_derive(kctx, ki->key.data, ki->key.len, params) != 1) + goto fail_derive; + + EVP_KDF_CTX_free(kctx); + EVP_KDF_free(kdf); + + return 0; + + fail_derive: + EVP_KDF_CTX_free(kctx); + fail_ctx: + EVP_KDF_free(kdf); + fail_fetch: + return -ECRYPT; +} + +int openssl_hkdf_expand(buffer_t key, + buffer_t info, + buffer_t out) +{ + EVP_KDF * kdf; + EVP_KDF_CTX * kctx; + OSSL_PARAM params[5]; + int mode = EVP_KDF_HKDF_MODE_EXPAND_ONLY; + int idx = 0; + int ret = -1; + + kdf = EVP_KDF_fetch(NULL, "HKDF", NULL); + if (kdf == NULL) + goto fail_fetch; + + kctx = EVP_KDF_CTX_new(kdf); + if (kctx == NULL) + goto fail_ctx; + + params[idx++] = OSSL_PARAM_construct_utf8_string( + "digest", (char *) "SHA256", 0); + params[idx++] = OSSL_PARAM_construct_int("mode", &mode); + params[idx++] = OSSL_PARAM_construct_octet_string( + "key", key.data, key.len); + params[idx++] = OSSL_PARAM_construct_octet_string( + "info", info.data, info.len); + params[idx] = OSSL_PARAM_construct_end(); + + if (EVP_KDF_derive(kctx, out.data, out.len, params) == 1) + ret = 0; + + EVP_KDF_CTX_free(kctx); + fail_ctx: + EVP_KDF_free(kdf); + fail_fetch: + return ret; +} + +/* AEAD seal: encrypt in with key/nonce, bind aad, append tag */ +int openssl_seal(struct ossl_crypt_ctx * ctx, + const uint8_t * key, + const uint8_t * nonce, + buffer_t aad, + buffer_t in, + uint8_t * out, + uint8_t * tag) +{ + int out_sz; + int tmp_sz; + + assert(ctx != NULL); + assert(ctx->tagsz > 0); /* AEAD mandated at ctx creation */ + + EVP_CIPHER_CTX_reset(ctx->evp_ctx); + + if (EVP_EncryptInit_ex(ctx->evp_ctx, ctx->cipher, NULL, + NULL, NULL) != 1) + return -1; + + /* Pin the AEAD nonce to 96 bits (SP 800-38D deterministic IV). */ + if (EVP_CIPHER_CTX_ctrl(ctx->evp_ctx, EVP_CTRL_AEAD_SET_IVLEN, + AEAD_NONCE_LEN, NULL) != 1) + return -1; + + if (EVP_EncryptInit_ex(ctx->evp_ctx, NULL, NULL, + key, nonce) != 1) + return -1; + + if (EVP_EncryptUpdate(ctx->evp_ctx, NULL, &tmp_sz, + aad.data, (int) aad.len) != 1) + return -1; + + if (EVP_EncryptUpdate(ctx->evp_ctx, out, &out_sz, + in.data, (int) in.len) != 1) + return -1; + + if (EVP_EncryptFinal_ex(ctx->evp_ctx, out + out_sz, &tmp_sz) != 1) + return -1; + + out_sz += tmp_sz; + + if (EVP_CIPHER_CTX_ctrl(ctx->evp_ctx, EVP_CTRL_AEAD_GET_TAG, + ctx->tagsz, tag) != 1) + return -1; + + return out_sz; +} + +/* AEAD open: decrypt in with key/nonce, verify aad and tag */ +int openssl_open(struct ossl_crypt_ctx * ctx, + const uint8_t * key, + const uint8_t * nonce, + buffer_t aad, + buffer_t in, + const uint8_t * tag, + buffer_t * out) +{ + int out_sz; + int tmp_sz; + + assert(ctx != NULL); + assert(ctx->tagsz > 0); /* AEAD mandated at ctx creation */ + + EVP_CIPHER_CTX_reset(ctx->evp_ctx); + + if (EVP_DecryptInit_ex(ctx->evp_ctx, ctx->cipher, NULL, + NULL, NULL) != 1) + return -1; + + /* Pin the AEAD nonce to 96 bits (SP 800-38D deterministic IV). */ + if (EVP_CIPHER_CTX_ctrl(ctx->evp_ctx, EVP_CTRL_AEAD_SET_IVLEN, + AEAD_NONCE_LEN, NULL) != 1) + return -1; + + if (EVP_DecryptInit_ex(ctx->evp_ctx, NULL, NULL, key, nonce) != 1) + return -1; + + if (EVP_CIPHER_CTX_ctrl(ctx->evp_ctx, EVP_CTRL_AEAD_SET_TAG, + ctx->tagsz, (void *) tag) != 1) + return -1; + + if (EVP_DecryptUpdate(ctx->evp_ctx, NULL, &tmp_sz, + aad.data, (int) aad.len) != 1) + return -1; + + if (EVP_DecryptUpdate(ctx->evp_ctx, out->data, &out_sz, + in.data, (int) in.len) != 1) + return -1; + + if (EVP_DecryptFinal_ex(ctx->evp_ctx, out->data + out_sz, + &tmp_sz) != 1) + return -1; + + out_sz += tmp_sz; + + out->len = (size_t) out_sz; + + return out_sz; +} + +/* + * Derive the common secret from + * - your public key pair (pkp) + * - the remote public key bytes (remote_pk). + * Store it in a preallocated buffer (s). + */ +static int __openssl_dhe_derive(EVP_PKEY * pkp, + EVP_PKEY * pub, + buffer_t remote_pk, + int kdf, + uint8_t * s) +{ + EVP_PKEY_CTX * ctx; + struct kdf_info ki; + buffer_t local_pk; + int ret; + uint8_t * secret; + size_t secret_len; + uint8_t salt_buf[HKDF_SALT_LEN]; + + /* Extract local public key bytes */ + local_pk.data = NULL; + ret = i2d_PUBKEY(pkp, &local_pk.data); + if (ret <= 0) + goto fail_local; + + local_pk.len = (size_t) ret; + + ki.salt.len = HKDF_SALT_LEN; + ki.salt.data = salt_buf; + + /* Derive salt from both public keys */ + if (derive_salt_from_pk_bytes_dhe(local_pk, remote_pk, ki.salt) < 0) + goto fail_salt; + + ctx = EVP_PKEY_CTX_new(pkp, NULL); + if (ctx == NULL) + goto fail_salt; + + ret = EVP_PKEY_derive_init(ctx); + if (ret != 1) + goto fail_ctx; + + ret = EVP_PKEY_derive_set_peer(ctx, pub); + if (ret != 1) + goto fail_ctx; + + ret = EVP_PKEY_derive(ctx, NULL, &secret_len); + if (ret != 1) + goto fail_ctx; + + if (secret_len < SYMMKEYSZ) + goto fail_ctx; + + secret = OPENSSL_malloc(secret_len); + if (secret == NULL) + goto fail_ctx; + + ret = EVP_PKEY_derive(ctx, secret, &secret_len); + if (ret != 1) + goto fail_derive; + + ki.nid = kdf; + ki.secret.len = secret_len; + ki.secret.data = secret; + ki.info.len = strlen(HKDF_INFO_DHE); + ki.info.data = (uint8_t *) HKDF_INFO_DHE; + ki.key.len = SYMMKEYSZ; + ki.key.data = s; + + /* Derive symmetric key from shared secret using HKDF */ + ret = derive_key_hkdf(&ki); + + OPENSSL_clear_free(secret, secret_len); + EVP_PKEY_CTX_free(ctx); + OPENSSL_free(local_pk.data); + + if (ret != 0) + return ret; + + return 0; + fail_derive: + OPENSSL_clear_free(secret, secret_len); + fail_ctx: + EVP_PKEY_CTX_free(ctx); + fail_salt: + OPENSSL_free(local_pk.data); + fail_local: + return -ECRYPT; +} + +static int __openssl_dhe_gen_key(const char * algo, + EVP_PKEY ** kp) +{ + EVP_PKEY_CTX * ctx = NULL; + EVP_PKEY_CTX * kctx = NULL; + EVP_PKEY * params = NULL; + int nid; + int type; + int ret; + + assert(algo != NULL); + assert(kp != NULL); + + nid = OBJ_txt2nid(algo); + if (nid == NID_undef) + return -ECRYPT; + + /* X25519 and X448: direct keygen context */ + if (nid == EVP_PKEY_X25519 || nid == EVP_PKEY_X448) { + kctx = EVP_PKEY_CTX_new_id(nid, NULL); + if (kctx == NULL) + goto fail_kctx; + + goto keygen; + } + /* EC and FFDHE: parameter generation first */ + type = (strncmp(algo, "ffdhe", 5) == 0) ? EVP_PKEY_DH : EVP_PKEY_EC; + + ctx = EVP_PKEY_CTX_new_id(type, NULL); + if (ctx == NULL) + goto fail_ctx; + + ret = EVP_PKEY_paramgen_init(ctx); + if (ret != 1) + goto fail_paramgen; + + if (type == EVP_PKEY_EC) + ret = EVP_PKEY_CTX_set_ec_paramgen_curve_nid(ctx, nid); + else /* EVP_PKEY_DH */ + ret = EVP_PKEY_CTX_set_dh_nid(ctx, nid); + + if (ret != 1) + goto fail_paramgen; + + ret = EVP_PKEY_paramgen(ctx, ¶ms); + if (ret != 1) + goto fail_paramgen; + + kctx = EVP_PKEY_CTX_new(params, NULL); + if (kctx == NULL) + goto fail_kctx; + + EVP_PKEY_free(params); + EVP_PKEY_CTX_free(ctx); + keygen: + ret = EVP_PKEY_keygen_init(kctx); + if (ret != 1) + goto fail_keygen; + + ret = EVP_PKEY_keygen(kctx, kp); + if (ret != 1) + goto fail_keygen; + + EVP_PKEY_CTX_free(kctx); + + return 0; + + fail_keygen: + EVP_PKEY_CTX_free(kctx); + return -ECRYPT; + fail_kctx: + if (params != NULL) + EVP_PKEY_free(params); + fail_paramgen: + if (ctx != NULL) + EVP_PKEY_CTX_free(ctx); + fail_ctx: + return -ECRYPT; +} + +static int __openssl_kem_gen_key(const char * algo, + EVP_PKEY ** kp) +{ + EVP_PKEY_CTX * kctx; + int ret; + + assert(algo != NULL); + assert(kp != NULL); + + /* PQC KEM (ML-KEM-512, ML-KEM-768, ML-KEM-1024) or hybrid */ + kctx = EVP_PKEY_CTX_new_from_name(NULL, algo, NULL); + if (kctx == NULL) + goto fail_kctx; + + ret = EVP_PKEY_keygen_init(kctx); + if (ret != 1) + goto fail_keygen; + + ret = EVP_PKEY_keygen(kctx, kp); + if (ret != 1) + goto fail_keygen; + + EVP_PKEY_CTX_free(kctx); + + return 0; + + fail_keygen: + EVP_PKEY_CTX_free(kctx); + fail_kctx: + return -ECRYPT; +} + +/* Determine hybrid KEM algorithm from raw key/ciphertext length */ +static const char * __openssl_hybrid_algo_from_len(size_t len) +{ + switch(len) { + case X25519MLKEM768_PKSZ: + return "X25519MLKEM768"; + case X25519MLKEM768_CTSZ: + return "X25519MLKEM768"; + case X448MLKEM1024_PKSZ: + return "X448MLKEM1024"; + default: + break; + } + + return NULL; +} + +static int __openssl_kex_gen_key(const char * algo, + EVP_PKEY ** kp) +{ + assert(algo != NULL); + assert(kp != NULL); + + /* Dispatch based on algorithm name prefix */ + if (IS_KEM_ALGORITHM(algo)) + return __openssl_kem_gen_key(algo, kp); + + return __openssl_dhe_gen_key(algo, kp); +} + +ssize_t openssl_pkp_create(const char * algo, + EVP_PKEY ** pkp, + uint8_t * pk) +{ + uint8_t * pos; + buffer_t raw; + ssize_t len; + + assert(algo != NULL); + assert(pkp != NULL); + assert(*pkp == NULL); + assert(pk != NULL); + + if (__openssl_kex_gen_key(algo, pkp) < 0) + goto fail_key; + + if (IS_HYBRID_KEM(algo)) { /* Raw encode hybrid KEM */ + raw.len = EVP_PKEY_get1_encoded_public_key(*pkp, &raw.data); + if (raw.len == 0) + goto fail_pubkey; + + if (raw.len > CRYPT_KEY_BUFSZ) { + OPENSSL_free(raw.data); + goto fail_pubkey; + } + + memcpy(pk, raw.data, raw.len); + OPENSSL_free(raw.data); + + return (ssize_t) raw.len; + } else { /* DER encode standard algorithms */ + len = i2d_PUBKEY(*pkp, NULL); /* pre-flight length */ + if (len < 0 || len > CRYPT_KEY_BUFSZ) + goto fail_pubkey; + + pos = pk; /* i2d_PUBKEY increments the ptr, don't use pk! */ + if (i2d_PUBKEY(*pkp, &pos) < 0) + goto fail_pubkey; + + return len; + } + fail_pubkey: + EVP_PKEY_free(*pkp); + fail_key: + return -ECRYPT; +} + +/* Common KEM encapsulation - pub key and salt already prepared */ +static ssize_t __openssl_kem_encap(EVP_PKEY * pub, + uint8_t * salt, + uint8_t * ct, + int kdf, + uint8_t * s) +{ + EVP_PKEY_CTX * ctx; + struct kdf_info ki; + uint8_t * secret; + size_t secret_len; + size_t ct_len; + int ret; + + ctx = EVP_PKEY_CTX_new(pub, NULL); + if (ctx == NULL) + goto fail_ctx; + + ret = EVP_PKEY_encapsulate_init(ctx, NULL); + if (ret != 1) + goto fail_encap; + + /* Get required lengths */ + ret = EVP_PKEY_encapsulate(ctx, NULL, &ct_len, NULL, &secret_len); + if (ret != 1 || ct_len > CRYPT_KEY_BUFSZ) + goto fail_encap; + + /* Allocate buffer for secret */ + secret = OPENSSL_malloc(secret_len); + if (secret == NULL) + goto fail_encap; + + /* Perform encapsulation */ + ret = EVP_PKEY_encapsulate(ctx, ct, &ct_len, secret, &secret_len); + if (ret != 1) + goto fail_secret; + + ki.secret.len = secret_len; + ki.secret.data = secret; + ki.nid = kdf; + ki.info.len = strlen(HKDF_INFO_ENCAP); + ki.info.data = (uint8_t *) HKDF_INFO_ENCAP; + ki.key.len = SYMMKEYSZ; + ki.key.data = s; + ki.salt.len = HKDF_SALT_LEN; + ki.salt.data = salt; + + /* Derive symmetric key from shared secret using HKDF */ + ret = derive_key_hkdf(&ki); + + OPENSSL_clear_free(secret, secret_len); + EVP_PKEY_CTX_free(ctx); + + if (ret != 0) + return -ECRYPT; + + return (ssize_t) ct_len; + + fail_secret: + OPENSSL_clear_free(secret, secret_len); + fail_encap: + EVP_PKEY_CTX_free(ctx); + fail_ctx: + return -ECRYPT; +} + +/* ML-KEM encapsulation - DER-encoded public key */ +ssize_t openssl_kem_encap(buffer_t pk, + uint8_t * ct, + int kdf, + uint8_t * s) +{ + EVP_PKEY * pub; + uint8_t * pos; + uint8_t salt[HKDF_SALT_LEN]; + buffer_t salt_b; + ssize_t ret; + + assert(pk.data != NULL); + assert(ct != NULL); + assert(s != NULL); + + salt_b.len = HKDF_SALT_LEN; + salt_b.data = salt; + + if (derive_salt_from_pk_bytes(pk, salt_b) < 0) + goto fail_salt; + + pos = pk.data; + pub = d2i_PUBKEY(NULL, (const uint8_t **) &pos, (long) pk.len); + if (pub == NULL) + goto fail_salt; + + ret = __openssl_kem_encap(pub, salt, ct, kdf, s); + + EVP_PKEY_free(pub); + + return ret; + fail_salt: + return -ECRYPT; +} + +/* Hybrid KEM encapsulation: raw-encoded public key */ +ssize_t openssl_kem_encap_raw(buffer_t pk, + uint8_t * ct, + int kdf, + uint8_t * s) +{ + EVP_PKEY * pub; + const char * algo; + uint8_t salt[HKDF_SALT_LEN]; + buffer_t salt_b; + ssize_t ret; + + assert(pk.data != NULL); + assert(ct != NULL); + assert(s != NULL); + + salt_b.len = HKDF_SALT_LEN; + salt_b.data = salt; + + if (derive_salt_from_pk_bytes(pk, salt_b) < 0) + goto fail_salt; + + algo = __openssl_hybrid_algo_from_len(pk.len); + if (algo == NULL) + goto fail_salt; + + pub = EVP_PKEY_new_raw_public_key_ex(NULL, algo, NULL, + pk.data, pk.len); + if (pub == NULL) + goto fail_salt; + + ret = __openssl_kem_encap(pub, salt, ct, kdf, s); + + EVP_PKEY_free(pub); + + return ret; + fail_salt: + return -ECRYPT; +} + +/* KEM decapsulation - used by party that generated the keypair */ +int openssl_kem_decap(EVP_PKEY * priv, + buffer_t ct, + int kdf, + uint8_t * s) +{ + EVP_PKEY_CTX * ctx; + struct kdf_info ki; + buffer_t pk; + uint8_t * secret; + size_t secret_len; + int ret; + uint8_t salt[HKDF_SALT_LEN]; + buffer_t salt_b; + + /* Extract public key bytes from private key */ + if (get_pk_bytes_from_key(priv, &pk) < 0) + goto fail_pk; + + salt_b.len = HKDF_SALT_LEN; + salt_b.data = salt; + + if (derive_salt_from_pk_bytes(pk, salt_b) < 0) + goto fail_salt; + + ctx = EVP_PKEY_CTX_new(priv, NULL); + if (ctx == NULL) + goto fail_salt; + + ret = EVP_PKEY_decapsulate_init(ctx, NULL); + if (ret != 1) + goto fail_ctx; + + /* Get required secret length */ + ret = EVP_PKEY_decapsulate(ctx, NULL, &secret_len, ct.data, ct.len); + if (ret != 1) + goto fail_ctx; + + /* Allocate buffer for secret */ + secret = OPENSSL_malloc(secret_len); + if (secret == NULL) + goto fail_ctx; + + /* Perform decapsulation */ + ret = EVP_PKEY_decapsulate(ctx, secret, &secret_len, ct.data, ct.len); + if (ret != 1) + goto fail_secret; + + ki.secret.len = secret_len; + ki.secret.data = secret; + ki.nid = kdf; + ki.info.len = strlen(HKDF_INFO_ENCAP); + ki.info.data = (uint8_t *) HKDF_INFO_ENCAP; + ki.key.len = SYMMKEYSZ; + ki.key.data = s; + ki.salt.len = HKDF_SALT_LEN; + ki.salt.data = salt; + + /* Derive symmetric key from shared secret using HKDF */ + ret = derive_key_hkdf(&ki); + + OPENSSL_clear_free(secret, secret_len); + EVP_PKEY_CTX_free(ctx); + OPENSSL_free(pk.data); + + if (ret != 0) + return ret; + + return 0; + + fail_secret: + OPENSSL_clear_free(secret, secret_len); + fail_ctx: + EVP_PKEY_CTX_free(ctx); + fail_salt: + OPENSSL_free(pk.data); + fail_pk: + return -ECRYPT; +} + +void openssl_pkp_destroy(EVP_PKEY * pkp) +{ + EVP_PKEY_free(pkp); +} + +static int openssl_get_curve(EVP_PKEY * pub, + char * algo) +{ + int ret; + size_t len = KEX_ALGO_BUFSZ; + + ret = EVP_PKEY_get_utf8_string_param(pub, "group", algo, len, &len); + + return ret == 1 ? 0 : -ECRYPT; +} + +int openssl_get_algo_from_pk_der(buffer_t pk, + char * algo) +{ + uint8_t * pos; + EVP_PKEY * pub; + char * type_str; + + assert(pk.data != NULL); + assert(algo != NULL); + + pos = pk.data; + pub = d2i_PUBKEY(NULL, (const uint8_t **) &pos, (long) pk.len); + if (pub == NULL) + goto fail_decode; + + type_str = (char *) EVP_PKEY_get0_type_name(pub); + if (type_str == NULL) + goto fail_pub; + + strcpy(algo, type_str); + + if (IS_EC_GROUP(algo) || IS_DH_GROUP(algo)) { + if (openssl_get_curve(pub, algo) < 0) + goto fail_pub; + } + + EVP_PKEY_free(pub); + return 0; + + fail_pub: + EVP_PKEY_free(pub); + fail_decode: + return -ECRYPT; +} + +int openssl_get_algo_from_pk_raw(buffer_t pk, + char * algo) +{ + const char * hybrid_algo; + + assert(pk.data != NULL); + assert(algo != NULL); + + hybrid_algo = __openssl_hybrid_algo_from_len(pk.len); + if (hybrid_algo == NULL) + return -ECRYPT; + + strcpy(algo, hybrid_algo); + + return 0; +} + +int openssl_dhe_derive(EVP_PKEY * pkp, + buffer_t pk, + int kdf, + uint8_t * s) +{ + uint8_t * pos; + EVP_PKEY * pub; + + assert(pkp != NULL); + assert(pk.data != NULL); + assert(s != NULL); + + /* X.509 DER decoding for DHE */ + pos = pk.data; /* d2i_PUBKEY increments pos, don't use key ptr! */ + pub = d2i_PUBKEY(NULL, (const uint8_t **) &pos, (long) pk.len); + if (pub == NULL) + goto fail_decode; + + if (__openssl_dhe_derive(pkp, pub, pk, kdf, s) < 0) + goto fail_derive; + + EVP_PKEY_free(pub); + + return 0; + fail_derive: + EVP_PKEY_free(pub); + fail_decode: + return -ECRYPT; +} + +/* Set up a fresh AEAD cipher ctx for nid: reject non-AEAD / oversized IV. */ +static int ossl_cipher_ctx_init(struct ossl_crypt_ctx * ctx, + int nid) +{ + ctx->cipher = EVP_get_cipherbynid(nid); + if (ctx->cipher == NULL) + return -1; + + /* IV must fit the NONCESZ nonce buffer. */ + if (EVP_CIPHER_get_iv_length(ctx->cipher) > NONCESZ) + return -1; + + /* Authenticated encryption is mandatory; reject non-AEAD ciphers. */ + if ((EVP_CIPHER_flags(ctx->cipher) & EVP_CIPH_FLAG_AEAD_CIPHER) == 0) + return -1; + + ctx->tagsz = AEAD_TAG_LEN; + + ctx->evp_ctx = EVP_CIPHER_CTX_new(); + if (ctx->evp_ctx == NULL) + return -1; + + return 0; +} + +/* One-shot AEAD seal over an explicit key/nonce (no keyrot). out = ct ‖ tag. */ +int openssl_oneshot_seal(int nid, + const uint8_t * key, + const uint8_t * nonce, + buffer_t aad, + buffer_t in, + buffer_t * out) +{ + struct ossl_crypt_ctx ctx; + int out_sz; + + assert(key != NULL); + assert(nonce != NULL); + assert(out != NULL); + + memset(&ctx, 0, sizeof(ctx)); + + if (ossl_cipher_ctx_init(&ctx, nid) < 0) + goto fail_cipher; + + out->data = malloc(in.len + EVP_MAX_BLOCK_LENGTH + ctx.tagsz); + if (out->data == NULL) + goto fail_ctx; + + out_sz = openssl_seal(&ctx, key, nonce, aad, in, + out->data, out->data + in.len); + if (out_sz < 0) + goto fail_seal; + + out->len = (size_t) out_sz + ctx.tagsz; + + EVP_CIPHER_CTX_free(ctx.evp_ctx); + + return 0; + + fail_seal: + free(out->data); + fail_ctx: + EVP_CIPHER_CTX_free(ctx.evp_ctx); + fail_cipher: + clrbuf(*out); + return -ECRYPT; +} + +/* One-shot AEAD open; in = ct ‖ tag, verifies aad and tag. */ +int openssl_oneshot_open(int nid, + const uint8_t * key, + const uint8_t * nonce, + buffer_t aad, + buffer_t in, + buffer_t * out) +{ + struct ossl_crypt_ctx ctx; + buffer_t ct; + const uint8_t * tag; + int in_sz; + + assert(key != NULL); + assert(nonce != NULL); + assert(out != NULL); + + memset(&ctx, 0, sizeof(ctx)); + + if (ossl_cipher_ctx_init(&ctx, nid) < 0) + goto fail_cipher; + + if (in.len < (size_t) ctx.tagsz) + goto fail_ctx; + + in_sz = (int) in.len - ctx.tagsz; + + out->data = malloc((size_t) in_sz + EVP_MAX_BLOCK_LENGTH); + if (out->data == NULL) + goto fail_ctx; + + ct.data = in.data; + ct.len = (size_t) in_sz; + tag = in.data + in_sz; + + if (openssl_open(&ctx, key, nonce, aad, ct, tag, out) < 0) + goto fail_open; + + EVP_CIPHER_CTX_free(ctx.evp_ctx); + + return 0; + + fail_open: + free(out->data); + fail_ctx: + EVP_CIPHER_CTX_free(ctx.evp_ctx); + fail_cipher: + clrbuf(*out); + return -ECRYPT; +} + +struct ossl_crypt_ctx * openssl_crypt_create_ctx(struct crypt_sk * sk) +{ + struct ossl_crypt_ctx * ctx; + + assert(sk != NULL); + assert(sk->key != NULL); + + ctx = malloc(sizeof(*ctx)); + if (ctx == NULL) + goto fail_malloc; + + memset(ctx, 0, sizeof(*ctx)); + + if (ossl_cipher_ctx_init(ctx, sk->nid) < 0) + goto fail_cipher; + + return ctx; + + fail_cipher: + free(ctx); + fail_malloc: + return NULL; +} + +void openssl_crypt_destroy_ctx(struct ossl_crypt_ctx * ctx) +{ + if (ctx == NULL) + return; + + EVP_CIPHER_CTX_free(ctx->evp_ctx); + free(ctx); +} + +int openssl_crypt_get_tagsz(struct ossl_crypt_ctx * ctx) +{ + assert(ctx != NULL); + + return ctx->tagsz; +} + +/* AUTHENTICATION */ + +int openssl_load_crt_file(const char * path, + void ** crt) +{ + FILE * fp; + X509 * xcrt; + + fp = fopen(path, "r"); + if (fp == NULL) + goto fail_file; + + pthread_cleanup_push(__cleanup_fclose, fp); + + xcrt = PEM_read_X509(fp, NULL, NULL, NULL); + + pthread_cleanup_pop(false); + + if (xcrt == NULL) + goto fail_crt; + + fclose(fp); + + *crt = (void *) xcrt; + + return 0; + fail_crt: + fclose(fp); + fail_file: + *crt = NULL; + return -1; +} + +static void * rd_crt_bio(BIO * bio) +{ + return PEM_read_bio_X509(bio, NULL, NULL, NULL); +} + +static void * rd_privkey_bio(BIO * bio) +{ + return PEM_read_bio_PrivateKey(bio, NULL, NULL, ""); +} + +static void * rd_pubkey_bio(BIO * bio) +{ + return PEM_read_bio_PUBKEY(bio, NULL, NULL, NULL); +} + +/* Decode a PEM object from an in-memory string via rd. */ +static int load_pem_str(const char * str, + void * (* rd)(BIO *), + void ** out) +{ + BIO * bio; + void * obj; + + bio = BIO_new(BIO_s_mem()); + if (bio == NULL) + goto fail_bio; + + if (BIO_write(bio, str, strlen(str)) < 0) + goto fail_obj; + + obj = rd(bio); + if (obj == NULL) + goto fail_obj; + + BIO_free(bio); + + *out = obj; + + return 0; + fail_obj: + BIO_free(bio); + fail_bio: + *out = NULL; + return -1; +} + +int openssl_load_crt_str(const char * str, + void ** crt) +{ + return load_pem_str(str, rd_crt_bio, crt); +} + +int openssl_load_crt_der(buffer_t buf, + void ** crt) +{ + const uint8_t * p; + X509 * xcrt; + + assert(crt != NULL); + + p = buf.data; + + xcrt = d2i_X509(NULL, &p, buf.len); + if (xcrt == NULL) + goto fail_crt; + + *crt = (void *) xcrt; + + return 0; + fail_crt: + *crt = NULL; + return -1; +} + +int openssl_get_pubkey_crt(void * crt, + void ** key) +{ + EVP_PKEY * pk; + X509 * xcrt; + + assert(crt != NULL); + assert(key != NULL); + + xcrt = (X509 *) crt; + + pk = X509_get_pubkey(xcrt); + if (pk == NULL) + goto fail_key; + + *key = (void *) pk; + + return 0; + fail_key: + return -1; +} + +void openssl_free_crt(void * crt) +{ + X509_free((X509 *) crt); +} + +int openssl_load_privkey_file(const char * path, + void ** key) +{ + FILE * fp; + EVP_PKEY * pkey; + + fp = fopen(path, "r"); + if (fp == NULL) + goto fail_file; + + pthread_cleanup_push(__cleanup_fclose, fp); + + pkey = PEM_read_PrivateKey(fp, NULL, NULL, ""); + + pthread_cleanup_pop(false); + + if (pkey == NULL) + goto fail_key; + + fclose(fp); + + *key = (void *) pkey; + + return 0; + fail_key: + fclose(fp); + fail_file: + *key = NULL; + return -1; +} + +int openssl_load_privkey_str(const char * str, + void ** key) +{ + return load_pem_str(str, rd_privkey_bio, key); +} + +int openssl_load_pubkey_file(const char * path, + void ** key) +{ + FILE * fp; + EVP_PKEY * pkey; + + fp = fopen(path, "r"); + if (fp == NULL) + goto fail_file; + + pthread_cleanup_push(__cleanup_fclose, fp); + + pkey = PEM_read_PUBKEY(fp, NULL, NULL, NULL); + + pthread_cleanup_pop(false); + + if (pkey == NULL) + goto fail_key; + + fclose(fp); + + *key = (void *) pkey; + + return 0; + fail_key: + fclose(fp); + fail_file: + *key = NULL; + return -1; +} + +int openssl_load_pubkey_file_to_der(const char * path, + buffer_t * buf) +{ + FILE * fp; + EVP_PKEY * pkey; + int ret; + + assert(path != NULL); + assert(buf != NULL); + + memset(buf, 0, sizeof(*buf)); + + fp = fopen(path, "r"); + if (fp == NULL) + goto fail_file; + + pthread_cleanup_push(__cleanup_fclose, fp); + + pkey = PEM_read_PUBKEY(fp, NULL, NULL, NULL); + + pthread_cleanup_pop(false); + + if (pkey == NULL) + goto fail_key; + + /* Extract public key bytes in DER format */ + ret = get_pk_bytes_from_key(pkey, buf); + if (ret < 0) + goto fail_extract; + + EVP_PKEY_free(pkey); + + fclose(fp); + + return 0; + + fail_extract: + EVP_PKEY_free(pkey); + fail_key: + fclose(fp); + fail_file: + clrbuf(*buf); + return -1; +} + +int openssl_load_pubkey_str(const char * str, + void ** key) +{ + return load_pem_str(str, rd_pubkey_bio, key); +} + +int openssl_load_pubkey_raw_file(const char * path, + buffer_t * buf) +{ + FILE * fp; + uint8_t tmp_buf[CRYPT_KEY_BUFSZ]; + size_t bytes_read; + const char * algo; + + assert(path != NULL); + assert(buf != NULL); + + fp = fopen(path, "rb"); + if (fp == NULL) + goto fail_file; + + pthread_cleanup_push(__cleanup_fclose, fp); + + bytes_read = fread(tmp_buf, 1, CRYPT_KEY_BUFSZ, fp); + + pthread_cleanup_pop(false); + + if (bytes_read == 0) + goto fail_read; + + /* Validate that this is a known hybrid KEM format */ + algo = __openssl_hybrid_algo_from_len(bytes_read); + if (algo == NULL) + goto fail_read; + + buf->data = malloc(bytes_read); + if (buf->data == NULL) + goto fail_malloc; + + memcpy(buf->data, tmp_buf, bytes_read); + buf->len = bytes_read; + + fclose(fp); + return 0; + + fail_malloc: + fail_read: + fclose(fp); + fail_file: + clrbuf(*buf); + return -1; +} + +/* Determine hybrid KEM algorithm from raw private key length */ +static const char * __openssl_hybrid_algo_from_sk_len(size_t len) +{ + switch(len) { + case X25519MLKEM768_SKSZ: + return "X25519MLKEM768"; + case X448MLKEM1024_SKSZ: + return "X448MLKEM1024"; + default: + break; + } + + return NULL; +} + +/* Wipe the raw-key staging buffer if a cancel aborts the read. */ +static void __cleanse_key_buf(void * o) +{ + OPENSSL_cleanse(o, CRYPT_KEY_BUFSZ); +} + +int openssl_load_privkey_raw_file(const char * path, + void ** key) +{ + FILE * fp; + uint8_t tmp_buf[CRYPT_KEY_BUFSZ]; + size_t bytes_read; + const char * algo; + EVP_PKEY * pkey; + + assert(path != NULL); + assert(key != NULL); + + fp = fopen(path, "rb"); + if (fp == NULL) + goto fail_file; + + pthread_cleanup_push(__cleanup_fclose, fp); + pthread_cleanup_push(__cleanse_key_buf, tmp_buf); + + bytes_read = fread(tmp_buf, 1, sizeof(tmp_buf), fp); + + pthread_cleanup_pop(false); + pthread_cleanup_pop(false); + + if (bytes_read == 0) + goto fail_read; + + /* Determine algorithm from key size */ + algo = __openssl_hybrid_algo_from_sk_len(bytes_read); + if (algo == NULL) + goto fail_read; + + pkey = EVP_PKEY_new_raw_private_key_ex(NULL, algo, NULL, + tmp_buf, bytes_read); + /* Clear sensitive data from stack */ + OPENSSL_cleanse(tmp_buf, bytes_read); + + if (pkey == NULL) + goto fail_read; + + fclose(fp); + + *key = (void *) pkey; + + return 0; + + fail_read: + fclose(fp); + fail_file: + *key = NULL; + return -1; +} + +int openssl_cmp_key(const EVP_PKEY * key1, + const EVP_PKEY * key2) +{ + assert(key1 != NULL); + assert(key2 != NULL); + +#if OPENSSL_VERSION_NUMBER >= 0x30000000L + return EVP_PKEY_eq(key1, key2) == 1 ? 0 : -1; +#else + return EVP_PKEY_cmp(key1, key2) == 1 ? 0 : -1; +#endif +} + +void openssl_free_key(EVP_PKEY * key) +{ + EVP_PKEY_free(key); +} + +int openssl_check_crt_name(void * crt, + const char * name) +{ + const unsigned char * cn; + ASN1_STRING * val; + X509_NAME * nm; + int idx; + int len; + + nm = X509_get_subject_name((X509 *) crt); + if (nm == NULL) + return -1; + + idx = X509_NAME_get_index_by_NID(nm, NID_commonName, -1); + if (idx < 0) + return -1; + + val = X509_NAME_ENTRY_get_data(X509_NAME_get_entry(nm, idx)); + cn = ASN1_STRING_get0_data(val); + len = ASN1_STRING_length(val); + + if (len < 0 || (size_t) len != strlen(name)) + return -1; + + if (memchr(cn, '\0', (size_t) len) != NULL) + return -1; + + if (memcmp(cn, name, (size_t) len) != 0) + return -1; + + return 0; +} + +int openssl_get_crt_name(void * crt, + char * name) +{ + const unsigned char * cn; + ASN1_STRING * val; + X509_NAME * nm; + int idx; + int len; + + nm = X509_get_subject_name((X509 *) crt); + if (nm == NULL) + return -1; + + idx = X509_NAME_get_index_by_NID(nm, NID_commonName, -1); + if (idx < 0) + return -1; + + val = X509_NAME_ENTRY_get_data(X509_NAME_get_entry(nm, idx)); + cn = ASN1_STRING_get0_data(val); + len = ASN1_STRING_length(val); + + if (len < 0) + return -1; + + if ((size_t) len > NAME_SIZE) + return -ENAME; + + /* Reject an embedded NUL that would truncate the parsed name. */ + if (memchr(cn, '\0', (size_t) len) != NULL) + return -1; + + memcpy(name, cn, (size_t) len); + name[len] = '\0'; + + return 0; +} + +int openssl_crt_str(const void * crt, + char * str) +{ + BIO * bio; + X509 * xcrt; + char * p; + ssize_t len; + + xcrt = (X509 *) crt; + + bio = BIO_new(BIO_s_mem()); + if (bio == NULL) + goto fail_bio; + + X509_print(bio, xcrt); + + len = (ssize_t) BIO_get_mem_data(bio, &p); + if (len <= 0 || p == NULL) + goto fail_p; + + memcpy(str, p, len); + str[len] = '\0'; + + BIO_free(bio); + + return 0; + fail_p: + BIO_free(bio); + fail_bio: + return -1; +} + +int openssl_crt_der(const void * crt, + buffer_t * buf) +{ + uint8_t * p; + int len; + + assert(crt != NULL); + assert(buf != NULL); + + /* Get the size by encoding to NULL */ + len = i2d_X509((X509 *) crt, NULL); + if (len < 0) + goto fail_len; + + buf->data = malloc((size_t) len); + if (buf->data == NULL) + goto fail_malloc; + + p = buf->data; /* i2d_X509 increments p */ + i2d_X509((X509 *) crt, &p); + buf->len = (size_t) len; + + return 0; + + fail_malloc: + fail_len: + clrbuf(*buf); + return -1; +} + +void * openssl_auth_create_store(void) +{ + return X509_STORE_new(); +} + +void openssl_auth_destroy_store(void * ctx) +{ + X509_STORE_free((X509_STORE *) ctx); +} + +int openssl_auth_add_crt_to_store(void * store, + void * crt) +{ + int ret; + + ret = X509_STORE_add_cert((X509_STORE *) store, (X509 *) crt); + + return ret == 1 ? 0 : -1; +} + +void * openssl_auth_create_chain(void) +{ + return sk_X509_new_null(); +} + +void openssl_auth_destroy_chain(void * chain) +{ + sk_X509_pop_free((STACK_OF(X509) *) chain, X509_free); +} + +int openssl_auth_add_crt_to_chain(void * chain, + void * crt) +{ + if (X509_up_ref((X509 *) crt) != 1) + goto fail_ref; + + if (sk_X509_push((STACK_OF(X509) *) chain, (X509 *) crt) == 0) + goto fail_push; + + return 0; + fail_push: + X509_free((X509 *) crt); + fail_ref: + return -1; +} + +int openssl_verify_crt_pin(void * store, + void * untrusted, + void * crt, + void * pin) +{ + X509_STORE_CTX * ctx; + X509_STORE * _store; + X509* _crt; + STACK_OF(X509) * chain; + int i; + int n; + int ret; + + _store = (X509_STORE *) store; + _crt = (X509 *) crt; + + ctx = X509_STORE_CTX_new(); + if (ctx == NULL) + goto fail_store_ctx; + + ret = X509_STORE_CTX_init(ctx, _store, _crt, + (STACK_OF(X509) *) untrusted); + if (ret != 1) + goto fail_ca; + + ret = X509_verify_cert(ctx); + if (ret != 1) + goto fail_ca; + + /* Peer cert only verifies a signature; gate on sig KU, not role. */ + if ((X509_get_key_usage(_crt) & KU_DIGITAL_SIGNATURE) == 0) + goto fail_ca; + + if (pin != NULL) { + chain = X509_STORE_CTX_get0_chain(ctx); + if (chain == NULL) + goto fail_ca; + n = sk_X509_num(chain); + for (i = 1; i < n; i++) /* Skip the leaf */ + if (X509_cmp(sk_X509_value(chain, i), pin) == 0) + break; + if (i == n) + goto fail_pin; + } + + X509_STORE_CTX_free(ctx); + + return 0; + fail_pin: + X509_STORE_CTX_free(ctx); + return -ENOENT; + fail_ca: + X509_STORE_CTX_free(ctx); + fail_store_ctx: + return -EAUTH; +} + +int openssl_verify_crt(void * store, + void * untrusted, + void * crt) +{ + return openssl_verify_crt_pin(store, untrusted, crt, NULL); +} + +static const EVP_MD * select_md(EVP_PKEY * pkey, + int nid) +{ + if (EVP_PKEY_get_id(pkey) < 0) + return NULL; /* Provider-based (PQC) */ + + if (nid == NID_undef) + return NULL; /* Classical requires explicit nid */ + + return EVP_get_digestbynid(nid); +} + +bool openssl_pk_requires_md(const EVP_PKEY * pk) +{ + /* Provider-based (PQC) signatures have an intrinsic digest */ + return EVP_PKEY_get_id(pk) >= 0; +} + +int openssl_sign(EVP_PKEY * pkp, + int nid, + buffer_t msg, + buffer_t * sig) +{ + EVP_MD_CTX * mdctx; + const EVP_MD * md; + size_t required; + + assert(pkp != NULL); + assert(sig != NULL); + + mdctx = EVP_MD_CTX_new(); + if (!mdctx) + goto fail_ctx; + + md = select_md(pkp, nid); + + if (EVP_DigestSignInit(mdctx, NULL, md, NULL, pkp) != 1) + goto fail_digest; + + /* Get required signature buffer size */ + if (EVP_DigestSign(mdctx, NULL, &required, msg.data, msg.len) != 1) + goto fail_digest; + + sig->data = malloc(required); + if (sig->data == NULL) + goto fail_digest; + + if (EVP_DigestSign(mdctx, sig->data, &required, msg.data, msg.len) != 1) + goto fail_sign; + + sig->len = required; + + EVP_MD_CTX_free(mdctx); + + return 0; + fail_sign: + freebuf(*sig); + fail_digest: + EVP_MD_CTX_free(mdctx); + fail_ctx: + clrbuf(*sig); + return -1; +} + +int openssl_verify_sig(EVP_PKEY * pk, + int nid, + buffer_t msg, + buffer_t sig) +{ + EVP_MD_CTX * mdctx; + const EVP_MD * md; + int ret; + + assert(pk != NULL); + + mdctx = EVP_MD_CTX_new(); + if (!mdctx) + goto fail_ctx; + + md = select_md(pk, nid); + + if (EVP_DigestVerifyInit(mdctx, NULL, md, NULL, pk) != 1) + goto fail_digest; + + ret = EVP_DigestVerify(mdctx, sig.data, sig.len, msg.data, msg.len); + if (ret != 1) + goto fail_digest; + + EVP_MD_CTX_free(mdctx); + + return 0; + fail_digest: + EVP_MD_CTX_free(mdctx); + fail_ctx: + clrbuf(sig); + return -1; +} + +ssize_t openssl_md_digest(int nid, + buffer_t in, + uint8_t * out) +{ + const EVP_MD * md; + unsigned int len; + + assert(in.data != NULL); + assert(out != NULL); + + md = EVP_get_digestbynid(nid); + if (md == NULL) + return -1; + + if (EVP_Digest(in.data, in.len, out, &len, md, NULL) != 1) + return -1; + + return (ssize_t) len; +} + +ssize_t openssl_md_len(int nid) +{ + const EVP_MD * md; + + md = EVP_get_digestbynid(nid); + if (md == NULL) + return -1; + + return (ssize_t) EVP_MD_get_size(md); +} + +int openssl_secure_malloc_init(size_t max, + size_t guard) +{ + return CRYPTO_secure_malloc_init(max, guard) == 1 ? 0 : -1; +} + +void openssl_secure_malloc_fini(void) +{ + CRYPTO_secure_malloc_done(); +} + +void * openssl_secure_malloc(size_t size) +{ + return OPENSSL_secure_malloc(size); +} + +void openssl_secure_free(void * ptr, + size_t size) +{ + OPENSSL_secure_clear_free(ptr, size); +} + +void openssl_secure_clear(void * ptr, + size_t size) +{ + OPENSSL_cleanse(ptr, size); +} + +void openssl_cleanup(void) +{ + OPENSSL_cleanup(); +} diff --git a/src/lib/crypt/openssl.h b/src/lib/crypt/openssl.h new file mode 100644 index 00000000..e5cc35f7 --- /dev/null +++ b/src/lib/crypt/openssl.h @@ -0,0 +1,213 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * OpenSSL based cryptographic operations + * Elliptic curve Diffie-Hellman key exchange + * AES encryption + # Authentication + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This library is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public License + * version 2.1 as published by the Free Software Foundation. + * + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with this library; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#ifndef OUROBOROS_LIB_CRYPT_OPENSSL_H +#define OUROBOROS_LIB_CRYPT_OPENSSL_H + +struct ossl_crypt_ctx; + +ssize_t openssl_pkp_create(const char * algo, + EVP_PKEY ** pkp, + uint8_t * pk); + +void openssl_pkp_destroy(EVP_PKEY * pkp); + +int openssl_dhe_derive(EVP_PKEY * pkp, + buffer_t pk, + int kdf_nid, + uint8_t * s); + +ssize_t openssl_kem_encap(buffer_t pk, + uint8_t * ct, + int kdf_nid, + uint8_t * s); + +/* no X509 DER support yet for DHKEM public keys */ +ssize_t openssl_kem_encap_raw(buffer_t pk, + uint8_t * ct, + int kdf_nid, + uint8_t * s); + +int openssl_kem_decap(EVP_PKEY * priv, + buffer_t ct, + int kdf_nid, + uint8_t * s); + +int openssl_get_algo_from_pk_der(buffer_t pk, + char * algo); + +int openssl_get_algo_from_pk_raw(buffer_t pk, + char * algo); + +int openssl_seal(struct ossl_crypt_ctx * ctx, + const uint8_t * key, + const uint8_t * nonce, + buffer_t aad, + buffer_t in, + uint8_t * out, + uint8_t * tag); + +int openssl_open(struct ossl_crypt_ctx * ctx, + const uint8_t * key, + const uint8_t * nonce, + buffer_t aad, + buffer_t in, + const uint8_t * tag, + buffer_t * out); + +int openssl_oneshot_seal(int nid, + const uint8_t * key, + const uint8_t * nonce, + buffer_t aad, + buffer_t in, + buffer_t * out); + +int openssl_oneshot_open(int nid, + const uint8_t * key, + const uint8_t * nonce, + buffer_t aad, + buffer_t in, + buffer_t * out); + +int openssl_hkdf_expand(buffer_t key, + buffer_t info, + buffer_t out); + +struct ossl_crypt_ctx * openssl_crypt_create_ctx(struct crypt_sk * sk); + +void openssl_crypt_destroy_ctx(struct ossl_crypt_ctx * ctx); + +int openssl_crypt_get_tagsz(struct ossl_crypt_ctx * ctx); + +/* AUTHENTICATION */ + +int openssl_load_crt_file(const char * path, + void ** crt); + +int openssl_load_crt_str(const char * str, + void ** crt); + +int openssl_load_crt_der(buffer_t buf, + void ** crt); + +int openssl_get_pubkey_crt(void * crt, + void ** pk); + +void openssl_free_crt(void * crt); + +int openssl_load_privkey_file(const char * path, + void ** key); + +int openssl_load_privkey_str(const char * str, + void ** key); + +int openssl_load_pubkey_file(const char * path, + void ** key); + +int openssl_load_pubkey_str(const char * str, + void ** key); +int openssl_load_pubkey_file_to_der(const char * path, + buffer_t * buf); +int openssl_load_pubkey_raw_file(const char * path, + buffer_t * buf); + +int openssl_load_privkey_raw_file(const char * path, + void ** key); + +int openssl_cmp_key(const EVP_PKEY * key1, + const EVP_PKEY * key2); + +void openssl_free_key(EVP_PKEY * key); + +int openssl_check_crt_name(void * crt, + const char * name); + +int openssl_get_crt_name(void * crt, + char * name); + +int openssl_crt_str(const void * crt, + char * str); + +int openssl_crt_der(const void * crt, + buffer_t * buf); + +void * openssl_auth_create_store(void); + +void openssl_auth_destroy_store(void * store); + +int openssl_auth_add_crt_to_store(void * store, + void * crt); + +void * openssl_auth_create_chain(void); + +void openssl_auth_destroy_chain(void * chain); + +int openssl_auth_add_crt_to_chain(void * chain, + void * crt); + +int openssl_verify_crt(void * store, + void * untrusted, + void * crt); + +int openssl_verify_crt_pin(void * store, + void * untrusted, + void * crt, + void * pin); + +bool openssl_pk_requires_md(const EVP_PKEY * pk); + +int openssl_sign(EVP_PKEY * pkp, + int md_nid, + buffer_t msg, + buffer_t * sig); + +int openssl_verify_sig(EVP_PKEY * pk, + int md_nid, + buffer_t msg, + buffer_t sig); + +ssize_t openssl_md_digest(int md_nid, + buffer_t in, + uint8_t * out); + +ssize_t openssl_md_len(int md_nid); + +/* Secure memory allocation */ +int openssl_secure_malloc_init(size_t max, + size_t guard); + +void openssl_secure_malloc_fini(void); + +void * openssl_secure_malloc(size_t size); + +void openssl_secure_free(void * ptr, + size_t size); + +void openssl_secure_clear(void * ptr, + size_t size); + +void openssl_cleanup(void); + +#endif /* OUROBOROS_LIB_CRYPT_OPENSSL_H */ diff --git a/src/lib/dev.c b/src/lib/dev.c index 92310b9e..d0997273 100644 --- a/src/lib/dev.c +++ b/src/lib/dev.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * API for applications * @@ -27,11 +27,15 @@ #endif #include "config.h" +#include "ssm.h" +#include <ouroboros/atomics.h> #include <ouroboros/bitmap.h> #include <ouroboros/cep.h> +#include <ouroboros/crc16.h> #include <ouroboros/crypt.h> #include <ouroboros/dev.h> +#include <ouroboros/endian.h> #include <ouroboros/errno.h> #include <ouroboros/fccntl.h> #include <ouroboros/flow.h> @@ -44,32 +48,33 @@ #include <ouroboros/np1_flow.h> #include <ouroboros/pthread.h> #include <ouroboros/random.h> -#include <ouroboros/serdes-irm.h> -#include <ouroboros/shm_flow_set.h> -#include <ouroboros/shm_rdrbuff.h> -#include <ouroboros/shm_rbuff.h> -#include <ouroboros/sockets.h> -#include <ouroboros/utils.h> #ifdef PROC_FLOW_STATS #include <ouroboros/rib.h> #endif +#include <ouroboros/serdes-irm.h> +#include <ouroboros/sockets.h> +#include <ouroboros/ssm_flow_set.h> +#include <ouroboros/ssm_pool.h> +#include <ouroboros/ssm_rbuff.h> +#include <ouroboros/tw.h> +#include <ouroboros/utils.h> +#include <assert.h> #ifdef HAVE_LIBGCRYPT #include <gcrypt.h> #endif -#include <assert.h> -#include <stdlib.h> -#include <string.h> -#include <stdio.h> #include <stdarg.h> #include <stdbool.h> +#include <inttypes.h> +#include <stdio.h> +#include <stdlib.h> +#include <string.h> #include <sys/types.h> #ifndef CLOCK_REALTIME_COARSE #define CLOCK_REALTIME_COARSE CLOCK_REALTIME #endif -/* Partial read information. */ #define NO_PART -1 #define DONE_PART -2 @@ -77,29 +82,30 @@ #define SECMEMSZ 16384 #define MSGBUFSZ 2048 -/* map flow_ids to flow descriptors; track state of the flow */ struct fmap { int fd; - /* TODO: use actual flow state */ enum flow_state state; }; -#define frcti_to_flow(frcti) \ - ((struct flow *)((uint8_t *) frcti - offsetof(struct flow, frcti))) - struct flow { - struct list_head next; - struct flow_info info; - struct shm_rbuff * rx_rb; - struct shm_rbuff * tx_rb; - struct shm_flow_set * set; + struct ssm_rbuff * rx_rb; + struct ssm_rbuff * tx_rb; + struct ssm_flow_set * set; uint16_t oflags; ssize_t part_idx; - struct crypt_info crypt; + struct crypt_ctx * crypt; + int headsz; /* Selector */ + int tailsz; /* Tag + CRC */ + + struct timespec rk_grace; /* TX-promote deadline (0 = none) */ + struct timespec rk_attempt; /* Last re-key attempt (backoff) */ + bool rk_wm_inflight; /* Re-key trigger in flight */ + uint32_t rk_wm_ctr; /* Throttles the consult */ + bool rk_initiator; /* OAP initiator this re-key */ struct timespec snd_act; struct timespec rcv_act; @@ -118,39 +124,33 @@ struct flow_set { }; struct fqueue { - struct flowevent fqueue[SHM_BUFFER_SIZE]; /* Safe copy from shm. */ + struct flowevent fqueue[SSM_RBUFF_SIZE]; /* Safe copy from shm. */ size_t fqsize; size_t next; }; struct { - struct shm_rdrbuff * rdrb; - struct shm_flow_set * fqset; + struct ssm_pool * pool; + struct ssm_flow_set * fqset; struct bmp * fds; struct bmp * fqueues; struct flow * flows; struct fmap * id_to_fd; - struct list_head flow_list; pthread_mutex_t mtx; pthread_cond_t cond; - pthread_t tx; - pthread_t rx; - size_t n_frcti; - fset_t * frct_set; - pthread_rwlock_t lock; -} ai; +} proc; static void flow_destroy(struct fmap * p) { - pthread_mutex_lock(&ai.mtx); + pthread_mutex_lock(&proc.mtx); if (p->state == FLOW_DESTROY) { - pthread_mutex_unlock(&ai.mtx); + pthread_mutex_unlock(&proc.mtx); return; } @@ -159,12 +159,12 @@ static void flow_destroy(struct fmap * p) else p->state = FLOW_NULL; - pthread_cond_signal(&ai.cond); + pthread_cond_signal(&proc.cond); - pthread_cleanup_push(__cleanup_mutex_unlock, &ai.mtx); + pthread_cleanup_push(__cleanup_mutex_unlock, &proc.mtx); while (p->state != FLOW_NULL) - pthread_cond_wait(&ai.cond, &ai.mtx); + pthread_cond_wait(&proc.cond, &proc.mtx); p->fd = -1; p->state = FLOW_INIT; @@ -175,17 +175,17 @@ static void flow_destroy(struct fmap * p) static void flow_set_state(struct fmap * p, enum flow_state state) { - pthread_mutex_lock(&ai.mtx); + pthread_mutex_lock(&proc.mtx); if (p->state == FLOW_DESTROY) { - pthread_mutex_unlock(&ai.mtx); + pthread_mutex_unlock(&proc.mtx); return; } p->state = state; - pthread_cond_broadcast(&ai.cond); + pthread_cond_broadcast(&proc.cond); - pthread_mutex_unlock(&ai.mtx); + pthread_mutex_unlock(&proc.mtx); } static enum flow_state flow_wait_assign(int flow_id) @@ -193,26 +193,26 @@ static enum flow_state flow_wait_assign(int flow_id) enum flow_state state; struct fmap * p; - p = &ai.id_to_fd[flow_id]; + p = &proc.id_to_fd[flow_id]; - pthread_mutex_lock(&ai.mtx); + pthread_mutex_lock(&proc.mtx); if (p->state == FLOW_ALLOCATED) { - pthread_mutex_unlock(&ai.mtx); + pthread_mutex_unlock(&proc.mtx); return FLOW_ALLOCATED; } if (p->state == FLOW_INIT) p->state = FLOW_ALLOC_PENDING; - pthread_cleanup_push(__cleanup_mutex_unlock, &ai.mtx); + pthread_cleanup_push(__cleanup_mutex_unlock, &proc.mtx); while (p->state == FLOW_ALLOC_PENDING) - pthread_cond_wait(&ai.cond, &ai.mtx); + pthread_cond_wait(&proc.cond, &proc.mtx); if (p->state == FLOW_DESTROY) { p->state = FLOW_NULL; - pthread_cond_broadcast(&ai.cond); + pthread_cond_broadcast(&proc.cond); } state = p->state; @@ -224,13 +224,13 @@ static enum flow_state flow_wait_assign(int flow_id) return state; } -static int proc_announce(const char * prog) +static int proc_announce(const struct proc_info * proc) { uint8_t buf[SOCK_BUF_SIZE]; - buffer_t msg = {buf, SOCK_BUF_SIZE}; + buffer_t msg = {SOCK_BUF_SIZE, buf}; int err; - if (proc_announce__irm_req_ser(&msg, prog) < 0) + if (proc_announce__irm_req_ser(&msg, proc) < 0) return -ENOMEM; err = send_recv_msg(&msg); @@ -240,11 +240,11 @@ static int proc_announce(const char * prog) return irm__irm_result_des(&msg); } -/* IRMd will clean up the mess if this fails */ +/* IRMd cleans up on failure. */ static void proc_exit(void) { uint8_t buf[SOCK_BUF_SIZE]; - buffer_t msg = {buf, SOCK_BUF_SIZE}; + buffer_t msg = {SOCK_BUF_SIZE, buf}; if (proc_exit__irm_req_ser(&msg) < 0) return; @@ -252,195 +252,502 @@ static void proc_exit(void) send_recv_msg(&msg); } -#include "frct.c" +static int spb_encrypt(struct flow * flow, + struct ssm_pk_buff * spb) +{ + buffer_t in; + buffer_t out; + uint8_t * head; + uint8_t * tail; + + if (flow->crypt == NULL) + return 0; + + in.data = ssm_pk_buff_head(spb); + in.len = ssm_pk_buff_len(spb); + + if (crypt_encrypt(flow->crypt, in, &out) < 0) + goto fail_encrypt; + + head = ssm_pk_buff_push(spb, flow->headsz); + if (head == NULL) + goto fail_alloc; + + tail = ssm_pk_buff_push_tail(spb, flow->tailsz); + if (tail == NULL) + goto fail_alloc; + + memcpy(head, out.data, out.len); + + freebuf(out); -void * flow_tx(void * o) + return 0; + fail_alloc: + freebuf(out); + fail_encrypt: + return -ECRYPT; +} + +static int spb_decrypt(struct flow * flow, + struct ssm_pk_buff * spb) { - struct timespec tic = TIMESPEC_INIT_NS(TICTIME); + buffer_t in; + buffer_t out; + uint8_t * head; - (void) o; + if (flow->crypt == NULL) + return 0; - while (true) { - timerwheel_move(); + in.data = ssm_pk_buff_head(spb); + in.len = ssm_pk_buff_len(spb); - nanosleep(&tic, NULL); - } + if (crypt_decrypt(flow->crypt, in, &out) < 0) + return -ECRYPT; + + head = ssm_pk_buff_pop(spb, flow->headsz) + flow->headsz; + ssm_pk_buff_pop_tail(spb, flow->tailsz); + + memcpy(head, out.data, out.len); - return (void *) 0; + freebuf(out); + + return 0; } -static void flow_send_keepalive(struct flow * flow, - struct timespec now) +/* tw_move under proc.lock rdlock; gates teardown vs in-flight fires. */ +static void tw_move_safe(void) { - struct shm_du_buff * sdb; - ssize_t idx; - uint8_t * ptr; + pthread_rwlock_rdlock(&proc.lock); - idx = shm_rdrbuff_alloc(ai.rdrb, 0, &ptr, &sdb); - if (idx < 0) - return; + pthread_cleanup_push(__cleanup_rwlock_unlock, &proc.lock); - pthread_rwlock_wrlock(&ai.lock); + tw_move(); - flow->snd_act = now; + pthread_cleanup_pop(1); +} - if (shm_rbuff_write(flow->tx_rb, idx)) - shm_rdrbuff_remove(ai.rdrb, idx); - else - shm_flow_set_notify(flow->set, flow->info.id, FLOW_PKT); +static int crc_add(struct ssm_pk_buff * spb, + size_t head_skip) +{ + uint8_t * head; + uint8_t * tail; - pthread_rwlock_unlock(&ai.lock); + tail = ssm_pk_buff_push_tail(spb, CRCLEN); + if (tail == NULL) + return -ENOMEM; + + head = ssm_pk_buff_head(spb) + head_skip; + + mem_hash(HASH_CRC32, tail, head, tail - head); + + return 0; } -/* Needs rdlock on ai. */ -static void _flow_keepalive(struct flow * flow) +static int crc_check(struct ssm_pk_buff * spb, + size_t head_skip) { - struct timespec now; - struct timespec s_act; - struct timespec r_act; - int flow_id; - time_t timeo; - uint32_t acl; + uint32_t crc; + uint8_t * head; + uint8_t * tail; - s_act = flow->snd_act; - r_act = flow->rcv_act; + if (ssm_pk_buff_len(spb) < head_skip + CRCLEN) + return 1; - flow_id = flow->info.id; - timeo = flow->info.qs.timeout; + head = ssm_pk_buff_head(spb) + head_skip; + tail = ssm_pk_buff_pop_tail(spb, CRCLEN); - acl = shm_rbuff_get_acl(flow->rx_rb); - if (timeo == 0 || acl & (ACL_FLOWPEER | ACL_FLOWDOWN)) - return; + mem_hash(HASH_CRC32, &crc, head, tail - head); - clock_gettime(PTHREAD_COND_CLOCK, &now); + return !(crc == *((uint32_t *) tail)); +} - if (ts_diff_ns(&r_act, &now) > (int64_t) timeo * MILLION) { - shm_rbuff_set_acl(flow->rx_rb, ACL_FLOWPEER); - shm_flow_set_notify(ai.fqset, flow_id, FLOW_PEER); - return; +/* FRCT included here so it can use proc and dev.c statics directly. */ +#include "frct.c" + +/* Decrypt before any check so the plaintext is authoritative. */ +static bool invalid_pkt(struct flow * flow, + struct ssm_pk_buff * spb) +{ + const struct frct_pci * pci; + uint16_t flags; + size_t pci_total; + + if (spb == NULL || ssm_pk_buff_len(spb) == 0) + return true; + + if (spb_decrypt(flow, spb) < 0) + return true; + + if (flow->frcti == NULL) { + if (flow->info.qs.ber == 0 && crc_check(spb, 0) != 0) + return true; + return false; } - if (ts_diff_ns(&s_act, &now) > (int64_t) timeo * (MILLION >> 2)) { - pthread_rwlock_unlock(&ai.lock); + if (ssm_pk_buff_len(spb) < FRCT_PCILEN) + return true; + + pci = (const struct frct_pci *) ssm_pk_buff_head(spb); + flags = ntoh16(pci->flags); + + /* Untrusted flag read; mismatch on HCS will drop on corrupt. */ + if (flags & FRCT_DATA) + pci_total = frcti_data_hdr_len(flow->frcti); + else + pci_total = frcti_ctrl_hdr_len(flow->frcti); + + if (ssm_pk_buff_len(spb) < pci_total) + return true; - flow_send_keepalive(flow, now); + if (frct_hcs_check(pci, flow->frcti) != 0) + return true; - pthread_rwlock_rdlock(&ai.lock); + /* HCS valid: CRC32 on SACK; or on DATA if ber = 0. */ + if (flags & FRCT_SACK) { + if (crc_check(spb, pci_total) != 0) + return true; + + } else if ((flags & FRCT_DATA) && flow->info.qs.ber == 0) { + if (crc_check(spb, pci_total) != 0) + return true; } + + return false; } -static void handle_keepalives(void) +static bool deadline_passed(const struct timespec * abs) { - struct list_head * p; - struct list_head * h; + struct timespec now; - pthread_rwlock_rdlock(&ai.lock); + if (abs == NULL) + return false; - list_for_each_safe(p, h, &ai.flow_list) { - struct flow * flow; - flow = list_entry(p, struct flow, next); - _flow_keepalive(flow); - } + clock_gettime(PTHREAD_COND_CLOCK, &now); - pthread_rwlock_unlock(&ai.lock); + return ts_diff_ns(&now, abs) >= 0; } -static void __cleanup_fqueue_destroy(void * fq) +/* Clamp the wait by min(dl, next tw expiry, now + TICTIME). */ +static void compute_wait_deadline(const struct timespec * dl, + struct timespec * out) { - fqueue_destroy((fqueue_t *) fq); + struct timespec now; + struct timespec cap; + struct timespec expiry; + struct timespec tic = TIMESPEC_INIT_NS(TICTIME); + + clock_gettime(PTHREAD_COND_CLOCK, &now); + ts_add(&now, &tic, &cap); + + tw_next_expiry(&expiry); + + *out = (ts_diff_ns(&cap, &expiry) < 0) ? expiry : cap; + if (dl != NULL && ts_diff_ns(out, dl) > 0) + *out = *dl; } -void * flow_rx(void * o) +static void flow_drain_rx_nb(struct flow * flow) { - struct timespec tic = TIMESPEC_INIT_NS(TICTIME); - int ret; - struct fqueue * fq; + ssize_t idx; + struct ssm_pk_buff * spb; + struct ssm_rbuff * rx_rb; + struct frcti * frcti; +#ifdef PROC_FLOW_STATS + struct timespec t_a; + struct timespec t_b; +#endif + + if (flow->frcti != NULL) + STAT_BUMP(flow->frcti, drain_calls); - (void) o; + while (true) { + pthread_rwlock_rdlock(&proc.lock); - fq = fqueue_create(); + rx_rb = flow->rx_rb; + if (rx_rb == NULL) { + pthread_rwlock_unlock(&proc.lock); + return; + } - pthread_cleanup_push(__cleanup_fqueue_destroy, fq); + idx = ssm_rbuff_read(rx_rb); + if (idx < 0) { + pthread_rwlock_unlock(&proc.lock); + return; + } - /* fevent will filter all FRCT packets for us */ - while ((ret = fevent(ai.frct_set, fq, &tic)) != 0) { - if (ret == -ETIMEDOUT) { - handle_keepalives(); + spb = ssm_pool_get(proc.pool, idx); + if (invalid_pkt(flow, spb)) { + ssm_pool_remove(proc.pool, idx); + pthread_rwlock_unlock(&proc.lock); continue; } - while (fqueue_next(fq) >= 0) - ; /* no need to act */ - } + frcti = flow->frcti; + if (frcti != NULL) { +#ifdef PROC_FLOW_STATS + clock_gettime(CLOCK_MONOTONIC, &t_a); + FRCTI_RCV(frcti, spb); + clock_gettime(CLOCK_MONOTONIC, &t_b); + STAT_ADD(frcti, rcv_proc_ns, + (size_t) ts_diff_ns(&t_b, &t_a)); +#else + FRCTI_RCV(frcti, spb); +#endif + } else { + ssm_pool_remove(proc.pool, idx); + } - pthread_cleanup_pop(true); + pthread_rwlock_unlock(&proc.lock); - return (void *) 0; + /* Per-packet so the delayed-ACK fires on time in a burst. */ +#ifdef PROC_FLOW_STATS + clock_gettime(CLOCK_MONOTONIC, &t_a); + tw_move_safe(); + clock_gettime(CLOCK_MONOTONIC, &t_b); + if (frcti != NULL) + STAT_ADD(frcti, tw_move_ns, + (size_t) ts_diff_ns(&t_b, &t_a)); +#else + tw_move_safe(); +#endif + } } -static void flow_clear(int fd) +/* TX-promotion grace when the peer's install latency is unknown (raw). */ +#define REKEY_GRACE_MS 1000 + +/* Last-resort promote within N node-keys of exhaustion (< watermark). */ +#define REKEY_PROMOTE_FLOOR 1 + +/* Throttle re-key retries so a failed attempt can't storm the IRMd. */ +#define REKEY_BACKOFF_NS (250 * MILLION) + +/* proc.lock (rd) only guards teardown; crypt_rekey self-synchronises. */ +static void flow_rekey(struct flow * flow) { - memset(&ai.flows[fd], 0, sizeof(ai.flows[fd])); + struct flow_info info; + struct crypt_sk sk; + struct timespec now; + struct timespec intv; + time_t ms; + uint8_t key[SYMMKEYSZ]; + uint8_t buf[SOCK_BUF_SIZE]; + buffer_t msg = {SOCK_BUF_SIZE, buf}; + bool has_key; + bool initiator = false; + + pthread_rwlock_rdlock(&proc.lock); + if (flow->info.id < 0 || flow->crypt == NULL) { + pthread_rwlock_unlock(&proc.lock); + return; + } + + /* Back off so a failed attempt can't storm the IRMd per syscall. */ + clock_gettime(PTHREAD_COND_CLOCK, &now); + if (ts_diff_ns(&now, &flow->rk_attempt) < REKEY_BACKOFF_NS) { + pthread_rwlock_unlock(&proc.lock); + return; + } + + flow->rk_attempt = now; + info = flow->info; + pthread_rwlock_unlock(&proc.lock); + + if (flow_update__irm_req_ser(&msg, &info, false) < 0) + return; + + if (send_recv_msg(&msg) < 0) + return; - ai.flows[fd].info.id = -1; + sk.key = key; + if (flow_rekey__irm_result_des(&msg, &sk, &has_key, &initiator) < 0) + return; + + if (!has_key) + return; + + pthread_rwlock_rdlock(&proc.lock); + if (flow->info.id == info.id && flow->crypt != NULL) { + if (crypt_rekey(flow->crypt, &sk) == 0) { + flow->rk_initiator = initiator; + /* Hold TX on the old epoch until the peer installs. */ + ms = flow->info.mpl > 0 ? flow->info.mpl * 3 + : REKEY_GRACE_MS; + intv.tv_sec = ms / 1000; + intv.tv_nsec = (ms % 1000) * MILLION; + clock_gettime(PTHREAD_COND_CLOCK, &now); + ts_add(&now, &intv, &flow->rk_grace); + } + /* Re-arm the watermark even if the install was a no-op. */ + STORE_RELAXED(&flow->rk_wm_inflight, false); + } + pthread_rwlock_unlock(&proc.lock); + + crypt_secure_clear(key, SYMMKEYSZ); } -static void __flow_fini(int fd) +/* A clamp-timeout means tw work is due, not the caller deadline. */ +static int flow_rx_one(struct flow * flow, + struct timespec * abs) { - assert(fd >= 0 && fd < SYS_MAX_FLOWS); + struct timespec wait_abs; + struct ssm_pk_buff * spb; + struct ssm_rbuff * rx_rb; + ssize_t idx; + + while (true) { + compute_wait_deadline(abs, &wait_abs); - if (ai.flows[fd].frcti != NULL) { - ai.n_frcti--; - if (ai.n_frcti == 0) { - pthread_cancel(ai.tx); - pthread_join(ai.tx, NULL); + /* rdlock gates flow_fini; FLOWDOWN preempts the block. */ + pthread_rwlock_rdlock(&proc.lock); + + rx_rb = flow->rx_rb; + if (rx_rb == NULL) { + pthread_rwlock_unlock(&proc.lock); + return -EFLOWDOWN; } - shm_flow_set_del(ai.fqset, 0, ai.flows[fd].info.id); + /* Pull a parked re-key before re-blocking (idle reader). */ + if (flow->crypt != NULL + && (ssm_rbuff_get_flags(rx_rb) & RB_REKEY)) { + pthread_rwlock_unlock(&proc.lock); + flow_rekey(flow); + continue; + } - frcti_destroy(ai.flows[fd].frcti); - } + idx = ssm_rbuff_read_b(rx_rb, &wait_abs); + if (idx == -ETIMEDOUT) { + pthread_rwlock_unlock(&proc.lock); + if (deadline_passed(abs)) + return -ETIMEDOUT; + tw_move_safe(); + continue; + } + if (idx < 0) { + pthread_rwlock_unlock(&proc.lock); + return idx; + } - if (ai.flows[fd].info.id != -1) { - flow_destroy(&ai.id_to_fd[ai.flows[fd].info.id]); - bmp_release(ai.fds, fd); + spb = ssm_pool_get(proc.pool, idx); + if (invalid_pkt(flow, spb)) { + ssm_pool_remove(proc.pool, idx); + pthread_rwlock_unlock(&proc.lock); + continue; + } + + if (flow->frcti != NULL) + FRCTI_RCV(flow->frcti, spb); + else + ssm_pool_remove(proc.pool, idx); + + pthread_rwlock_unlock(&proc.lock); + + tw_move_safe(); + return 0; } +} + +/* 0 = window open; -EAGAIN = !block and would block; else flow_rx_one rc. */ +static __inline__ int flow_wait_window(struct flow * flow, + size_t n, + bool block, + struct timespec * dl) +{ + int rc; - if (ai.flows[fd].rx_rb != NULL) { - shm_rbuff_set_acl(ai.flows[fd].rx_rb, ACL_FLOWDOWN); - shm_rbuff_close(ai.flows[fd].rx_rb); + while (true) { + flow_drain_rx_nb(flow); + if (FRCTI_IS_WINDOW_OPEN_N(flow->frcti, n)) + return 0; + if (!block) + return -EAGAIN; + rc = flow_rx_one(flow, dl); + if (rc < 0) + return rc; } +} + +static void flow_clear(int fd) +{ + memset(&proc.flows[fd], 0, sizeof(proc.flows[fd])); + + proc.flows[fd].info.id = -1; +} + +/* Order before flow_fini's wrlock, which blocks on rdlock holders. */ +static void flow_quiesce(int fd) +{ + struct ssm_rbuff * rx_rb = proc.flows[fd].rx_rb; + struct ssm_rbuff * tx_rb = proc.flows[fd].tx_rb; - if (ai.flows[fd].tx_rb != NULL) { - shm_rbuff_set_acl(ai.flows[fd].tx_rb, ACL_FLOWDOWN); - shm_rbuff_close(ai.flows[fd].tx_rb); + if (rx_rb != NULL) + ssm_rbuff_set_bits(rx_rb, RB_FLOWDOWN); + + if (tx_rb != NULL) + ssm_rbuff_set_bits(tx_rb, RB_FLOWDOWN); +} + +static void do_flow_fini(int fd) +{ + assert(fd >= 0 && fd < PROC_MAX_FLOWS); + + if (proc.flows[fd].frcti != NULL) + frcti_destroy(proc.flows[fd].frcti); + + if (proc.flows[fd].info.id != -1) { + flow_destroy(&proc.id_to_fd[proc.flows[fd].info.id]); + bmp_release(proc.fds, fd); } - if (ai.flows[fd].set != NULL) { - shm_flow_set_notify(ai.flows[fd].set, - ai.flows[fd].info.id, + if (proc.flows[fd].rx_rb != NULL) + ssm_rbuff_close(proc.flows[fd].rx_rb); + + if (proc.flows[fd].tx_rb != NULL) + ssm_rbuff_close(proc.flows[fd].tx_rb); + + if (proc.flows[fd].set != NULL) { + ssm_flow_set_notify(proc.flows[fd].set, + proc.flows[fd].info.id, FLOW_DEALLOC); - shm_flow_set_close(ai.flows[fd].set); + ssm_flow_set_close(proc.flows[fd].set); } - crypt_fini(&ai.flows[fd].crypt); - - list_del(&ai.flows[fd].next); + crypt_destroy_ctx(proc.flows[fd].crypt); flow_clear(fd); } static void flow_fini(int fd) { - pthread_rwlock_wrlock(&ai.lock); + flow_quiesce(fd); - __flow_fini(fd); + pthread_rwlock_wrlock(&proc.lock); - pthread_rwlock_unlock(&ai.lock); + do_flow_fini(fd); + + pthread_rwlock_unlock(&proc.lock); +} + +#define IS_ENCRYPTED(crypt) ((crypt)->nid != NID_undef) +#define IS_ORDERED(info) ((info)->qs.service != SVC_RAW) +#define IS_STREAM(info) ((info)->qs.service == SVC_STREAM) + +/* Raw MTU minus the wrapping (IV/Tag + optional CRC) dev.c adds. */ +static __inline__ size_t flow_user_mtu(const struct flow * flow, + size_t raw) +{ + size_t hdr; + + hdr = flow->headsz + flow->tailsz; + if (flow->info.qs.ber == 0 && flow->crypt == NULL) + hdr += CRCLEN; + + return raw > hdr ? raw - hdr : 0; } static int flow_init(struct flow_info * info, - buffer_t * sk) + struct crypt_sk * sk, + time_t rtt_hint) { struct timespec now; struct flow * flow; @@ -449,27 +756,27 @@ static int flow_init(struct flow_info * info, clock_gettime(PTHREAD_COND_CLOCK, &now); - pthread_rwlock_wrlock(&ai.lock); + pthread_rwlock_wrlock(&proc.lock); - fd = bmp_allocate(ai.fds); - if (!bmp_is_id_valid(ai.fds, fd)) { + fd = bmp_allocate(proc.fds); + if (!bmp_is_id_valid(proc.fds, fd)) { err = -EBADF; goto fail_fds; } - flow = &ai.flows[fd]; + flow = &proc.flows[fd]; flow->info = *info; - flow->rx_rb = shm_rbuff_open(info->n_pid, info->id); + flow->rx_rb = ssm_rbuff_open(info->n_pid, info->id); if (flow->rx_rb == NULL) goto fail_rx_rb; - flow->tx_rb = shm_rbuff_open(info->n_1_pid, info->id); + flow->tx_rb = ssm_rbuff_open(info->n_1_pid, info->id); if (flow->tx_rb == NULL) goto fail_tx_rb; - flow->set = shm_flow_set_open(info->n_1_pid); + flow->set = ssm_flow_set_open(info->n_1_pid); if (flow->set == NULL) goto fail_set; @@ -477,59 +784,50 @@ static int flow_init(struct flow_info * info, flow->part_idx = NO_PART; flow->snd_act = now; flow->rcv_act = now; - - flow->crypt.flags = info->qs.cypher_s; /* TODO: move cypher_s */ - - memset(flow->crypt.key, 0, SYMMKEYSZ); - - if (flow->crypt.flags > 0 && sk!= NULL && sk->data != NULL) - memcpy(flow->crypt.key, sk->data , sk->len); - - if (crypt_init(&flow->crypt) < 0) - goto fail_crypt; + flow->crypt = NULL; + flow->headsz = 0; + flow->tailsz = 0; + + if (IS_ENCRYPTED(sk)) { + flow->crypt = crypt_create_ctx(sk); + if (flow->crypt == NULL) + goto fail_crypt; + flow->headsz = crypt_get_headsz(flow->crypt); + flow->tailsz = crypt_get_tagsz(flow->crypt); + } assert(flow->frcti == NULL); - if (info->qs.in_order != 0) { - flow->frcti = frcti_create(fd, DELT_A, DELT_R, info->mpl); + if (IS_ORDERED(&flow->info)) { + uint32_t frct_mtu = flow_user_mtu(flow, info->mtu); + + flow->frcti = frcti_create(fd, DELT_A, DELT_R, + info->mpl, rtt_hint, + info->qs, frct_mtu); if (flow->frcti == NULL) goto fail_frcti; - - if (shm_flow_set_add(ai.fqset, 0, info->id)) - goto fail_flow_set_add; - - ++ai.n_frcti; - if (ai.n_frcti == 1 && - pthread_create(&ai.tx, NULL, flow_tx, NULL) < 0) - goto fail_tx_thread; } - list_add_tail(&flow->next, &ai.flow_list); - - ai.id_to_fd[info->id].fd = fd; + proc.id_to_fd[info->id].fd = fd; - flow_set_state(&ai.id_to_fd[info->id], FLOW_ALLOCATED); + flow_set_state(&proc.id_to_fd[info->id], FLOW_ALLOCATED); - pthread_rwlock_unlock(&ai.lock); + pthread_rwlock_unlock(&proc.lock); return fd; - fail_tx_thread: - shm_flow_set_del(ai.fqset, 0, info->id); - fail_flow_set_add: - frcti_destroy(flow->frcti); fail_frcti: - crypt_fini(&flow->crypt); + crypt_destroy_ctx(flow->crypt); fail_crypt: - shm_flow_set_close(flow->set); + ssm_flow_set_close(flow->set); fail_set: - shm_rbuff_close(flow->tx_rb); + ssm_rbuff_close(flow->tx_rb); fail_tx_rb: - shm_rbuff_close(flow->rx_rb); + ssm_rbuff_close(flow->rx_rb); fail_rx_rb: - bmp_release(ai.fds, fd); + bmp_release(proc.fds, fd); fail_fds: - pthread_rwlock_unlock(&ai.lock); + pthread_rwlock_unlock(&proc.lock); return err; } @@ -547,6 +845,7 @@ static void init(int argc, char ** argv, char ** envp) { + struct proc_info info; char * prog = argv[0]; int i; #ifdef PROC_FLOW_STATS @@ -564,7 +863,11 @@ static void init(int argc, goto fail_prog; } - if (proc_announce(prog)) { + memset(&info, 0, sizeof(info)); + info.pid = getpid(); + strncpy(info.prog, prog, PROG_NAME_SIZE); + + if (proc_announce(&info)) { fprintf(stderr, "FATAL: Could not announce to IRMd.\n"); goto fail_prog; } @@ -579,74 +882,77 @@ static void init(int argc, gcry_control(GCRYCTL_INITIALIZATION_FINISHED, 0); } #endif - ai.fds = bmp_create(PROG_MAX_FLOWS - PROG_RES_FDS, PROG_RES_FDS); - if (ai.fds == NULL) { + proc.fds = bmp_create(PROC_MAX_FLOWS - PROC_RES_FDS, PROC_RES_FDS); + if (proc.fds == NULL) { fprintf(stderr, "FATAL: Could not create fd bitmap.\n"); goto fail_fds; } - ai.fqueues = bmp_create(PROG_MAX_FQUEUES, 0); - if (ai.fqueues == NULL) { + proc.fqueues = bmp_create(PROC_MAX_FQUEUES, 0); + if (proc.fqueues == NULL) { fprintf(stderr, "FATAL: Could not create fqueue bitmap.\n"); goto fail_fqueues; } - ai.rdrb = shm_rdrbuff_open(); - if (ai.rdrb == NULL) { + if (is_ouroboros_member_uid(getuid())) + proc.pool = ssm_pool_open(0); + else + proc.pool = ssm_pool_open(getuid()); + + if (proc.pool == NULL) { fprintf(stderr, "FATAL: Could not open packet buffer.\n"); goto fail_rdrb; } - ai.flows = malloc(sizeof(*ai.flows) * PROG_MAX_FLOWS); - if (ai.flows == NULL) { + proc.flows = malloc(sizeof(*proc.flows) * PROC_MAX_FLOWS); + if (proc.flows == NULL) { fprintf(stderr, "FATAL: Could not malloc flows.\n"); goto fail_flows; } - for (i = 0; i < PROG_MAX_FLOWS; ++i) + for (i = 0; i < PROC_MAX_FLOWS; ++i) flow_clear(i); - ai.id_to_fd = malloc(sizeof(*ai.id_to_fd) * SYS_MAX_FLOWS); - if (ai.id_to_fd == NULL) { + proc.id_to_fd = malloc(sizeof(*proc.id_to_fd) * SYS_MAX_FLOWS); + if (proc.id_to_fd == NULL) { fprintf(stderr, "FATAL: Could not malloc id_to_fd.\n"); goto fail_id_to_fd; } for (i = 0; i < SYS_MAX_FLOWS; ++i) - ai.id_to_fd[i].state = FLOW_INIT; + proc.id_to_fd[i].state = FLOW_INIT; - if (pthread_mutex_init(&ai.mtx, NULL)) { + if (pthread_mutex_init(&proc.mtx, NULL)) { fprintf(stderr, "FATAL: Could not init mutex.\n"); goto fail_mtx; } - if (pthread_cond_init(&ai.cond, NULL) < 0) { + if (pthread_cond_init(&proc.cond, NULL) < 0) { fprintf(stderr, "FATAL: Could not init condvar.\n"); goto fail_cond; } - if (pthread_rwlock_init(&ai.lock, NULL) < 0) { + if (pthread_rwlock_init(&proc.lock, NULL) < 0) { fprintf(stderr, "FATAL: Could not initialize flow lock.\n"); goto fail_flow_lock; } - ai.fqset = shm_flow_set_open(getpid()); - if (ai.fqset == NULL) { + proc.fqset = ssm_flow_set_open(getpid()); + if (proc.fqset == NULL) { fprintf(stderr, "FATAL: Could not open flow set.\n"); goto fail_fqset; } - ai.frct_set = fset_create(); - if (ai.frct_set == NULL || ai.frct_set->idx != 0) { - fprintf(stderr, "FATAL: Could not create FRCT set.\n"); - goto fail_frct_set; - } - - if (timerwheel_init() < 0) { + if (tw_init() < 0) { fprintf(stderr, "FATAL: Could not initialize timerwheel.\n"); goto fail_timerwheel; } + if (crypt_secure_malloc_init(PROC_SECMEM_MAX) < 0) { + fprintf(stderr, "FATAL: Could not init secure malloc.\n"); + goto fail_secmem; + } + #if defined PROC_FLOW_STATS if (strstr(argv[0], "ipcpd") == NULL) { sprintf(procstr, "proc.%d", getpid()); @@ -656,43 +962,34 @@ static void init(int argc, } } #endif - if (pthread_create(&ai.rx, NULL, flow_rx, NULL) < 0) { - fprintf(stderr, "FATAL: Could not start monitor thread.\n"); - goto fail_monitor; - } - - list_head_init(&ai.flow_list); - return; - fail_monitor: #if defined PROC_FLOW_STATS - rib_fini(); fail_rib_init: + crypt_secure_malloc_fini(); #endif - timerwheel_fini(); + fail_secmem: + tw_fini(); fail_timerwheel: - fset_destroy(ai.frct_set); - fail_frct_set: - shm_flow_set_close(ai.fqset); + ssm_flow_set_close(proc.fqset); fail_fqset: - pthread_rwlock_destroy(&ai.lock); + pthread_rwlock_destroy(&proc.lock); fail_flow_lock: - pthread_cond_destroy(&ai.cond); + pthread_cond_destroy(&proc.cond); fail_cond: - pthread_mutex_destroy(&ai.mtx); + pthread_mutex_destroy(&proc.mtx); fail_mtx: - free(ai.id_to_fd); + free(proc.id_to_fd); fail_id_to_fd: - free(ai.flows); + free(proc.flows); fail_flows: - shm_rdrbuff_close(ai.rdrb); + ssm_pool_close(proc.pool); fail_rdrb: - bmp_destroy(ai.fqueues); + bmp_destroy(proc.fqueues); fail_fqueues: - bmp_destroy(ai.fds); + bmp_destroy(proc.fds); fail_fds: - memset(&ai, 0, sizeof(ai)); + memset(&proc, 0, sizeof(proc)); fail_prog: exit(EXIT_FAILURE); } @@ -701,51 +998,53 @@ static void fini(void) { int i; - if (ai.fds == NULL) + if (proc.fds == NULL) return; - pthread_cancel(ai.rx); - pthread_join(ai.rx, NULL); + /* Wake all in-flight readers/writers BEFORE wrlock acquire. */ + for (i = 0; i < PROC_MAX_FLOWS; ++i) + if (proc.flows[i].info.id != -1) + flow_quiesce(i); - pthread_rwlock_wrlock(&ai.lock); + pthread_rwlock_wrlock(&proc.lock); - for (i = 0; i < PROG_MAX_FLOWS; ++i) { - if (ai.flows[i].info.id != -1) { + for (i = 0; i < PROC_MAX_FLOWS; ++i) { + struct flow * flow = &proc.flows[i]; + if (flow->info.id != -1) { ssize_t idx; - shm_rbuff_set_acl(ai.flows[i].rx_rb, ACL_FLOWDOWN); - while ((idx = shm_rbuff_read(ai.flows[i].rx_rb)) >= 0) - shm_rdrbuff_remove(ai.rdrb, idx); - __flow_fini(i); + while ((idx = ssm_rbuff_read(flow->rx_rb)) >= 0) + ssm_pool_remove(proc.pool, idx); + do_flow_fini(i); } } - pthread_cond_destroy(&ai.cond); - pthread_mutex_destroy(&ai.mtx); + pthread_cond_destroy(&proc.cond); + pthread_mutex_destroy(&proc.mtx); - pthread_rwlock_unlock(&ai.lock); + pthread_rwlock_unlock(&proc.lock); #ifdef PROC_FLOW_STATS rib_fini(); #endif - timerwheel_fini(); + crypt_secure_malloc_fini(); - fset_destroy(ai.frct_set); + tw_fini(); - shm_flow_set_close(ai.fqset); + ssm_flow_set_close(proc.fqset); - pthread_rwlock_destroy(&ai.lock); + pthread_rwlock_destroy(&proc.lock); - free(ai.flows); - free(ai.id_to_fd); + free(proc.flows); + free(proc.id_to_fd); - shm_rdrbuff_close(ai.rdrb); + ssm_pool_close(proc.pool); - bmp_destroy(ai.fds); - bmp_destroy(ai.fqueues); + bmp_destroy(proc.fds); + bmp_destroy(proc.fqueues); proc_exit(); - memset(&ai, 0, sizeof(ai)); + memset(&proc, 0, sizeof(proc)); } #if defined(__MACH__) && defined(__APPLE__) @@ -762,17 +1061,22 @@ __attribute__((section(FINI_SECTION))) __typeof__(fini) * __fini = fini; int flow_accept(qosspec_t * qs, const struct timespec * timeo) { - struct flow_info flow; - uint8_t buf[SOCK_BUF_SIZE]; - buffer_t msg = {buf, SOCK_BUF_SIZE}; - buffer_t sk; - int fd; - int err; + struct flow_info flow; + struct crypt_sk crypt; + uint8_t buf[SOCK_BUF_SIZE]; + buffer_t msg = {SOCK_BUF_SIZE, buf}; + uint8_t key[SYMMKEYSZ]; + int fd; + int err; #ifdef QOS_DISABLE_CRC if (qs != NULL) qs->ber = 1; #endif + /* STREAM cannot tolerate loss: drops create silent gaps. */ + if (qs != NULL && qs->service == SVC_STREAM && qs->loss != 0) + return -EINVAL; + memset(&flow, 0, sizeof(flow)); flow.n_pid = getpid(); @@ -785,13 +1089,18 @@ int flow_accept(qosspec_t * qs, if (err < 0) return err; - err = flow__irm_result_des(&msg, &flow, &sk); + crypt.key = key; + crypt.epoch = 0; + crypt.role = CRYPT_ROLE_RESP; + + err = flow__irm_result_des(&msg, &flow, &crypt); if (err < 0) return err; - fd = flow_init(&flow, &sk); + /* No RTT in accept; rtt_hint=0 bootstraps from first ACK. */ + fd = flow_init(&flow, &crypt, 0); - freebuf(sk); + crypt_secure_clear(key, SYMMKEYSZ); if (qs != NULL) *qs = flow.qs; @@ -803,17 +1112,23 @@ int flow_alloc(const char * dst, qosspec_t * qs, const struct timespec * timeo) { - struct flow_info flow; - uint8_t buf[SOCK_BUF_SIZE]; - buffer_t msg = {buf, SOCK_BUF_SIZE}; - buffer_t sk; /* symmetric key */ - int fd; - int err; + struct flow_info flow; + struct crypt_sk crypt; + uint8_t buf[SOCK_BUF_SIZE]; + buffer_t msg = {SOCK_BUF_SIZE, buf}; + uint8_t key[SYMMKEYSZ]; + int fd; + int err; + struct timespec t0; + struct timespec t1; #ifdef QOS_DISABLE_CRC if (qs != NULL) qs->ber = 1; #endif + /* STREAM cannot tolerate loss: drops create silent gaps. */ + if (qs != NULL && qs->service == SVC_STREAM && qs->loss != 0) + return -EINVAL; memset(&flow, 0, sizeof(flow)); @@ -823,17 +1138,25 @@ int flow_alloc(const char * dst, if (flow_alloc__irm_req_ser(&msg, &flow, dst, timeo)) return -ENOMEM; + clock_gettime(PTHREAD_COND_CLOCK, &t0); + err = send_recv_msg(&msg); if (err < 0) return err; - err = flow__irm_result_des(&msg, &flow, &sk); + clock_gettime(PTHREAD_COND_CLOCK, &t1); + + crypt.key = key; + crypt.epoch = 0; + crypt.role = CRYPT_ROLE_INIT; + + err = flow__irm_result_des(&msg, &flow, &crypt); if (err < 0) return err; - fd = flow_init(&flow, &sk); + fd = flow_init(&flow, &crypt, ts_diff_ns(&t1, &t0)); - freebuf(sk); + crypt_secure_clear(key, SYMMKEYSZ); if (qs != NULL) *qs = flow.qs; @@ -842,42 +1165,39 @@ int flow_alloc(const char * dst, } int flow_join(const char * dst, - qosspec_t * qs, const struct timespec * timeo) { - struct flow_info flow; - uint8_t buf[SOCK_BUF_SIZE]; - buffer_t msg = {buf, SOCK_BUF_SIZE}; - int fd; - int err; - -#ifdef QOS_DISABLE_CRC - if (qs != NULL) - qs->ber = 1; -#endif - if (qs != NULL && qs->cypher_s > 0) - return -ENOTSUP; /* TODO: Encrypted broadcast */ + struct flow_info flow; + struct crypt_sk crypt; + uint8_t buf[SOCK_BUF_SIZE]; + buffer_t msg = {SOCK_BUF_SIZE, buf}; + uint8_t key[SYMMKEYSZ]; + int fd; + int err; memset(&flow, 0, sizeof(flow)); flow.n_pid = getpid(); - flow.qs = qs == NULL ? qos_raw : *qs; + flow.qs = qos_np1; - if (flow_alloc__irm_req_ser(&msg, &flow, dst, timeo)) + if (flow_join__irm_req_ser(&msg, &flow, dst, timeo)) return -ENOMEM; err = send_recv_msg(&msg); if (err < 0) return err; - err = flow__irm_result_des(&msg, &flow, NULL); + crypt.key = key; + crypt.epoch = 0; + crypt.role = CRYPT_ROLE_INIT; + + err = flow__irm_result_des(&msg, &flow, &crypt); if (err < 0) return err; - fd = flow_init(&flow, NULL); + fd = flow_init(&flow, &crypt, 0); - if (qs != NULL) - *qs = flow.qs; + crypt_secure_clear(key, SYMMKEYSZ); return fd; } @@ -888,23 +1208,23 @@ int flow_dealloc(int fd) struct flow_info info; uint8_t pkt[PKT_BUF_LEN]; uint8_t buf[SOCK_BUF_SIZE]; - buffer_t msg = {buf, SOCK_BUF_SIZE}; + buffer_t msg = {SOCK_BUF_SIZE, buf}; struct timespec tic = TIMESPEC_INIT_NS(TICTIME); struct timespec timeo = TIMESPEC_INIT_S(0); struct flow * flow; int err; - if (fd < 0 || fd >= SYS_MAX_FLOWS ) + if (fd < 0 || fd >= PROC_MAX_FLOWS ) return -EINVAL; - memset(&info, 0, sizeof(flow)); + memset(&info, 0, sizeof(info)); - flow = &ai.flows[fd]; + flow = &proc.flows[fd]; - pthread_rwlock_rdlock(&ai.lock); + pthread_rwlock_rdlock(&proc.lock); if (flow->info.id < 0) { - pthread_rwlock_unlock(&ai.lock); + pthread_rwlock_unlock(&proc.lock); return -ENOTALLOC; } @@ -913,46 +1233,48 @@ int flow_dealloc(int fd) flow->rcv_timesout = true; flow->rcv_timeo = tic; - pthread_rwlock_unlock(&ai.lock); + pthread_rwlock_unlock(&proc.lock); flow_read(fd, buf, SOCK_BUF_SIZE); - pthread_rwlock_rdlock(&ai.lock); + pthread_rwlock_rdlock(&proc.lock); - timeo.tv_sec = frcti_dealloc(flow->frcti); - while (timeo.tv_sec < 0) { /* keep the flow active for rtx */ - ssize_t ret; + while (FRCTI_LINGERING(flow->frcti)) { + ssize_t ret; - pthread_rwlock_unlock(&ai.lock); + pthread_rwlock_unlock(&proc.lock); ret = flow_read(fd, pkt, PKT_BUF_LEN); - pthread_rwlock_rdlock(&ai.lock); - - timeo.tv_sec = frcti_dealloc(flow->frcti); + pthread_rwlock_rdlock(&proc.lock); - if (ret == -EFLOWDOWN && timeo.tv_sec < 0) - timeo.tv_sec = -timeo.tv_sec; + if (ret == -EFLOWDOWN) + break; } - pthread_cleanup_push(__cleanup_rwlock_unlock, &ai.lock); + timeo.tv_sec = FRCTI_DEALLOC(flow->frcti); + + pthread_cleanup_push(__cleanup_rwlock_unlock, &proc.lock); - shm_rbuff_fini(flow->tx_rb); + ssm_rbuff_fini(flow->tx_rb); pthread_cleanup_pop(true); info.id = flow->info.id; info.n_pid = getpid(); - if (flow_dealloc__irm_req_ser(&msg, &info, &timeo) < 0) - return -ENOMEM; + if (flow_dealloc__irm_req_ser(&msg, &info, &timeo) < 0) { + err = -ENOMEM; + goto out; + } err = send_recv_msg(&msg); if (err < 0) - return err; + goto out; err = irm__irm_result_des(&msg); + out: flow_fini(fd); return err; @@ -962,38 +1284,41 @@ int ipcp_flow_dealloc(int fd) { struct flow_info info; uint8_t buf[SOCK_BUF_SIZE]; - buffer_t msg = {buf, SOCK_BUF_SIZE}; + buffer_t msg = {SOCK_BUF_SIZE, buf}; struct flow * flow; int err; - if (fd < 0 || fd >= SYS_MAX_FLOWS ) + if (fd < 0 || fd >= PROC_MAX_FLOWS ) return -EINVAL; - flow = &ai.flows[fd]; + flow = &proc.flows[fd]; - memset(&info, 0, sizeof(flow)); + memset(&info, 0, sizeof(info)); - pthread_rwlock_rdlock(&ai.lock); + pthread_rwlock_rdlock(&proc.lock); if (flow->info.id < 0) { - pthread_rwlock_unlock(&ai.lock); + pthread_rwlock_unlock(&proc.lock); return -ENOTALLOC; } info.id = flow->info.id; info.n_1_pid = flow->info.n_1_pid; - pthread_rwlock_unlock(&ai.lock); + pthread_rwlock_unlock(&proc.lock); - if (ipcp_flow_dealloc__irm_req_ser(&msg, &info) < 0) - return -ENOMEM; + if (ipcp_flow_dealloc__irm_req_ser(&msg, &info) < 0) { + err = -ENOMEM; + goto out; + } err = send_recv_msg(&msg); if (err < 0) - return err; + goto out; err = irm__irm_result_des(&msg); + out: flow_fini(fd); return err; @@ -1009,22 +1334,30 @@ int fccntl(int fd, va_list l; struct timespec * timeo; qosspec_t * qs; - uint32_t rx_acl; - uint32_t tx_acl; size_t * qlen; struct flow * flow; - - if (fd < 0 || fd >= SYS_MAX_FLOWS) + uint16_t old_acc; + uint16_t new_acc; + size_t max; + size_t * maxp; + size_t rsz; + size_t * rszp; + time_t rto; + time_t * rtop; + int rc; + bool emit_eos = false; + + if (fd < 0 || fd >= PROC_MAX_FLOWS) return -EBADF; - flow = &ai.flows[fd]; + flow = &proc.flows[fd]; va_start(l, cmd); - pthread_rwlock_wrlock(&ai.lock); + pthread_rwlock_wrlock(&proc.lock); if (flow->info.id < 0) { - pthread_rwlock_unlock(&ai.lock); + pthread_rwlock_unlock(&proc.lock); va_end(l); return -ENOTALLOC; } @@ -1072,42 +1405,50 @@ int fccntl(int fd, break; case FLOWGRXQLEN: qlen = va_arg(l, size_t *); - *qlen = shm_rbuff_queued(flow->rx_rb); + *qlen = ssm_rbuff_queued(flow->rx_rb); break; case FLOWGTXQLEN: qlen = va_arg(l, size_t *); - *qlen = shm_rbuff_queued(flow->tx_rb); + *qlen = ssm_rbuff_queued(flow->tx_rb); + break; + case FLOWGMTU: + maxp = va_arg(l, size_t *); + if (maxp == NULL) + goto einval; + *maxp = flow_user_mtu(flow, flow->info.mtu); break; case FLOWSFLAGS: + old_acc = flow->oflags & FLOWFACCMODE; flow->oflags = va_arg(l, uint32_t); - rx_acl = shm_rbuff_get_acl(flow->rx_rb); - tx_acl = shm_rbuff_get_acl(flow->rx_rb); - /* - * Making our own flow write only means making the - * the other side of the flow read only. - */ + new_acc = flow->oflags & FLOWFACCMODE; + + /* Defer EOS emit until after proc.lock is dropped: */ + /* frcti_fin_snd may block on shm-pool/tx-rb. */ + if (new_acc == FLOWFRDONLY + && old_acc != FLOWFRDONLY + && flow->frcti != NULL) + emit_eos = true; + + /* Our flow write-only -> peer's read-only; restore on RDWR. */ if (flow->oflags & FLOWFWRONLY) - rx_acl |= ACL_RDONLY; - if (flow->oflags & FLOWFRDWR) - rx_acl |= ACL_RDWR; + ssm_rbuff_clr_bits(flow->rx_rb, RB_WR); + else + ssm_rbuff_set_bits(flow->rx_rb, RB_WR); if (flow->oflags & FLOWFDOWN) { - rx_acl |= ACL_FLOWDOWN; - tx_acl |= ACL_FLOWDOWN; - shm_flow_set_notify(flow->set, + ssm_rbuff_set_bits(flow->rx_rb, RB_FLOWDOWN); + ssm_rbuff_set_bits(flow->tx_rb, RB_FLOWDOWN); + ssm_flow_set_notify(flow->set, flow->info.id, FLOW_DOWN); } else { - rx_acl &= ~ACL_FLOWDOWN; - tx_acl &= ~ACL_FLOWDOWN; - shm_flow_set_notify(flow->set, + ssm_rbuff_clr_bits(flow->rx_rb, RB_FLOWDOWN); + ssm_rbuff_clr_bits(flow->tx_rb, RB_FLOWDOWN); + ssm_flow_set_notify(flow->set, flow->info.id, FLOW_UP); } - shm_rbuff_set_acl(flow->rx_rb, rx_acl); - shm_rbuff_set_acl(flow->tx_rb, tx_acl); - break; case FLOWGFLAGS: fflags = va_arg(l, uint32_t *); @@ -1129,104 +1470,352 @@ int fccntl(int fd, goto eperm; *cflags = frcti_getflags(flow->frcti); break; + case FRCTSMAXSDU: + max = va_arg(l, size_t); + if (flow->frcti == NULL) + goto eperm; + if (frcti_set_max_rcv_sdu(flow->frcti, max) < 0) + goto einval; + break; + case FRCTGMAXSDU: + maxp = va_arg(l, size_t *); + if (maxp == NULL) + goto einval; + if (flow->frcti == NULL) + goto eperm; + *maxp = frcti_get_max_rcv_sdu(flow->frcti); + break; + case FRCTSRRINGSZ: + rsz = va_arg(l, size_t); + if (flow->frcti == NULL) + goto eperm; + rc = frcti_set_rcv_ring_sz(flow->frcti, rsz); + if (rc < 0) { + pthread_rwlock_unlock(&proc.lock); + va_end(l); + return rc; + } + break; + case FRCTGRRINGSZ: + rszp = va_arg(l, size_t *); + if (rszp == NULL) + goto einval; + if (flow->frcti == NULL) + goto eperm; + *rszp = frcti_get_rcv_ring_sz(flow->frcti); + break; + case FRCTSRTOMIN: + if (flow->frcti == NULL) + goto eperm; + rto = va_arg(l, time_t); + rc = frcti_set_rto_min(flow->frcti, rto); + if (rc < 0) { + pthread_rwlock_unlock(&proc.lock); + va_end(l); + return rc; + } + break; + case FRCTGRTOMIN: + if (flow->frcti == NULL) + goto eperm; + rtop = va_arg(l, time_t *); + if (rtop == NULL) + goto einval; + *rtop = frcti_get_rto_min(flow->frcti); + break; default: - pthread_rwlock_unlock(&ai.lock); + pthread_rwlock_unlock(&proc.lock); va_end(l); return -ENOTSUP; }; - pthread_rwlock_unlock(&ai.lock); + pthread_rwlock_unlock(&proc.lock); + + if (emit_eos) + frcti_fin_snd(flow->frcti); va_end(l); return 0; einval: - pthread_rwlock_unlock(&ai.lock); + pthread_rwlock_unlock(&proc.lock); va_end(l); return -EINVAL; eperm: - pthread_rwlock_unlock(&ai.lock); + pthread_rwlock_unlock(&proc.lock); va_end(l); return -EPERM; } -static int chk_crc(struct shm_du_buff * sdb) +static int flow_tx_spb(struct flow * flow, + struct ssm_pk_buff * spb, + uint16_t flags, + bool block, + struct timespec * abstime) { - uint32_t crc; - uint8_t * head = shm_du_buff_head(sdb); - uint8_t * tail = shm_du_buff_tail_release(sdb, CRCLEN); + struct timespec now; + ssize_t idx; + size_t pci_total; + int ret; - mem_hash(HASH_CRC32, &crc, head, tail - head); + clock_gettime(PTHREAD_COND_CLOCK, &now); + flow->snd_act = now; - return !(crc == *((uint32_t *) tail)); + idx = ssm_pk_buff_get_off(spb); + + if (ssm_pk_buff_len(spb) > 0) { + if (FRCTI_SND(flow->frcti, spb, flags) < 0) + goto enomem; + + if (flow->info.qs.ber == 0) { + pci_total = flow->frcti != NULL + ? frcti_data_hdr_len(flow->frcti) : 0; + if (crc_add(spb, pci_total) != 0) + goto enomem; + } + + if (spb_encrypt(flow, spb) < 0) + goto enomem; + } + + if (!block) + ret = ssm_rbuff_write(flow->tx_rb, idx); + else + ret = ssm_rbuff_write_b(flow->tx_rb, idx, abstime); + + if (ret < 0) + return ret; + + ssm_flow_set_notify(flow->set, flow->info.id, FLOW_PKT); + return 0; + + enomem: + return -ENOMEM; } -static int add_crc(struct shm_du_buff * sdb) +/* Per-fragment role for fragment i out of n; n == 1 yields SOLE. */ +static __inline__ uint16_t flow_frag_role(size_t i, size_t n) { - uint8_t * head = shm_du_buff_head(sdb); - uint8_t * tail = shm_du_buff_tail_alloc(sdb, CRCLEN); - if (tail == NULL) - return -1; + if (n == 1) + return FRCT_FR_SOLE; - mem_hash(HASH_CRC32, tail, head, tail - head); + if (i == 0) + return FRCT_FR_FIRST; - return 0; + if (i + 1 == n) + return FRCT_FR_LAST; + + return FRCT_FR_MID; } -static int flow_tx_sdb(struct flow * flow, - struct shm_du_buff * sdb, - bool block, - struct timespec * abstime) +static ssize_t flow_write_stream(struct flow * flow, + const void * buf, + size_t count, + int oflags, + struct timespec * dl) { - struct timespec now; - ssize_t idx; - int ret; + const uint8_t * src = buf; + size_t payload; + size_t off = 0; + bool block = !(oflags & FLOWFWNOBLOCK); - clock_gettime(PTHREAD_COND_CLOCK, &now); + if (!FRCTI_IS_FRTX(flow->frcti)) + return -EMSGSIZE; - pthread_rwlock_wrlock(&ai.lock); + payload = FRCTI_PAYLOAD_CAP(flow->frcti); - flow->snd_act = now; + while (off < count) { + struct ssm_pk_buff * spb; + uint8_t * ptr; + ssize_t idx; + size_t clen; + int ret; - pthread_rwlock_unlock(&ai.lock); + ret = flow_wait_window(flow, 1, block, dl); + if (ret < 0) + return off > 0 ? (ssize_t) off : (ssize_t) ret; - idx = shm_du_buff_get_idx(sdb); + clen = MIN(count - off, payload); - pthread_rwlock_rdlock(&ai.lock); + if (block) + idx = ssm_pool_alloc_b(proc.pool, clen, &ptr, + &spb, dl); + else + idx = ssm_pool_alloc(proc.pool, clen, &ptr, &spb); + if (idx < 0) + return off > 0 ? (ssize_t) off : idx; - if (shm_du_buff_len(sdb) > 0) { - if (frcti_snd(flow->frcti, sdb) < 0) - goto enomem; + memcpy(ptr, src + off, clen); - if (crypt_encrypt(&flow->crypt, sdb) < 0) - goto enomem; + ret = flow_tx_spb(flow, spb, 0, block, dl); + if (ret < 0) { + ssm_pool_remove(proc.pool, idx); + return off > 0 ? (ssize_t) off : (ssize_t) ret; + } - if (flow->info.qs.ber == 0 && add_crc(sdb) != 0) - goto enomem; + off += clen; } - pthread_cleanup_push(__cleanup_rwlock_unlock, &ai.lock); + return (ssize_t) count; +} - if (!block) - ret = shm_rbuff_write(flow->tx_rb, idx); - else - ret = shm_rbuff_write_b(flow->tx_rb, idx, abstime); +/* Per-fragment flow_tx_spb loop. Raw flows refuse; FRCT splits the SDU. */ +static ssize_t flow_write_frag(struct flow * flow, + const void * buf, + size_t count, + int oflags, + struct timespec * dl) +{ + const uint8_t * src = buf; + size_t frag_payload; + size_t n; + size_t off = 0; + size_t i; + int ret; + bool block = !(oflags & FLOWFWNOBLOCK); + + /* Raw flows carry no PCI; cannot fragment. */ + if (flow->frcti == NULL) + return -EMSGSIZE; + frag_payload = FRCTI_PAYLOAD_CAP(flow->frcti); + + /* Guard the ceil-divide against size_t overflow. */ + if (count > SIZE_MAX - frag_payload + 1) + return -EMSGSIZE; + + n = (count + frag_payload - 1) / frag_payload; + + /* SDU larger than the FC window can ever offer would deadlock. */ + if (n > RQ_SIZE) + return -EMSGSIZE; + + /* SDU-atomic FC: wait for n seqnos to avoid overshoot mid-SDU. */ + ret = flow_wait_window(flow, n, block, dl); if (ret < 0) - shm_rdrbuff_remove(ai.rdrb, idx); - else - shm_flow_set_notify(flow->set, flow->info.id, FLOW_PKT); + return (ssize_t) ret; + + STAT_BUMP(flow->frcti, sdu_snd_frag); + + for (i = 0; i < n; ++i) { + struct ssm_pk_buff * spb; + uint8_t * ptr; + ssize_t idx; + size_t clen; + + clen = (i + 1 == n) ? (count - off) : frag_payload; + + if (block) + idx = ssm_pool_alloc_b(proc.pool, clen, &ptr, + &spb, dl); + else + idx = ssm_pool_alloc(proc.pool, clen, &ptr, &spb); + if (idx < 0) { + if (off > 0) + STAT_BUMP(flow->frcti, sdu_snd_alloc); + return off > 0 ? (ssize_t) off : idx; + } - pthread_cleanup_pop(true); + memcpy(ptr, src + off, clen); + + ret = flow_tx_spb(flow, spb, flow_frag_role(i, n), + block, dl); + if (ret < 0) { + ssm_pool_remove(proc.pool, idx); + if (off > 0) + STAT_BUMP(flow->frcti, sdu_snd_tx); + return off > 0 ? (ssize_t) off : (ssize_t) ret; + } + + off += clen; + } + + return (ssize_t) count; +} + +/* + * Initiator promotes on the install grace (it holds the key-confirm + * tag); responder waits for peer_synced, with a near-exhaustion floor. + */ +static void flow_tx_promote(struct flow * flow) +{ + struct timespec now; + int nodes_left; + bool promote; + + if (flow->crypt == NULL) + return; + + if (flow->rk_grace.tv_sec == 0 && flow->rk_grace.tv_nsec == 0) + return; + + promote = crypt_peer_synced(flow->crypt); + + if (!promote && flow->rk_initiator) { + clock_gettime(PTHREAD_COND_CLOCK, &now); + promote = ts_diff_ns(&now, &flow->rk_grace) >= 0; + } + + if (!promote && !flow->rk_initiator) { + nodes_left = crypt_nodes_left(flow->crypt); + promote = nodes_left >= 0 && nodes_left <= REKEY_PROMOTE_FLOOR; + } + + if (!promote) + return; + + crypt_tx_promote(flow->crypt); + flow->rk_grace.tv_sec = 0; + flow->rk_grace.tv_nsec = 0; +} + +/* The reply carries no key; the seed arrives later over RB_REKEY. */ +static int flow_rekey_trigger(struct flow * flow) +{ + struct flow_info info; + uint8_t buf[SOCK_BUF_SIZE]; + buffer_t msg = {SOCK_BUF_SIZE, buf}; + + pthread_rwlock_rdlock(&proc.lock); + if (flow->info.id < 0 || flow->crypt == NULL) { + pthread_rwlock_unlock(&proc.lock); + return -1; + } + info = flow->info; + pthread_rwlock_unlock(&proc.lock); + + if (flow_update__irm_req_ser(&msg, &info, true) < 0) + return -1; + + if (send_recv_msg(&msg) < 0) + return -1; return 0; +} -enomem: - pthread_rwlock_unlock(&ai.lock); - shm_rdrbuff_remove(ai.rdrb, idx); - return -ENOMEM; +static bool flow_wm_due(struct flow * flow) +{ + uint32_t tick; + + if (KEY_REKEY_WATERMARK == 0) + return false; + + if (flow->crypt == NULL) + return false; + + if (LOAD_RELAXED(&flow->rk_wm_inflight)) + return false; + + tick = FETCH_ADD_RELAXED(&flow->rk_wm_ctr, 1); + if ((tick & (FLOW_WM_CHECK - 1)) != 0) + return false; + + if (ssm_rbuff_get_flags(flow->rx_rb) & RB_REKEY) + return false; + + return crypt_nodes_left(flow->crypt) <= KEY_REKEY_WATERMARK; } ssize_t flow_write(int fd, @@ -1238,138 +1827,250 @@ ssize_t flow_write(int fd, int ret; int flags; struct timespec abs; - struct timespec * abstime = NULL; - struct shm_du_buff * sdb; + struct timespec now; + struct timespec * dl = NULL; + struct ssm_pk_buff * spb; uint8_t * ptr; if (buf == NULL && count != 0) return -EINVAL; - if (fd < 0 || fd >= PROG_MAX_FLOWS) + if (fd < 0 || fd >= PROC_MAX_FLOWS) return -EBADF; - flow = &ai.flows[fd]; + flow = &proc.flows[fd]; - clock_gettime(PTHREAD_COND_CLOCK, &abs); - - pthread_rwlock_wrlock(&ai.lock); + pthread_rwlock_rdlock(&proc.lock); if (flow->info.id < 0) { - pthread_rwlock_unlock(&ai.lock); + pthread_rwlock_unlock(&proc.lock); return -ENOTALLOC; } + flags = flow->oflags; + + clock_gettime(PTHREAD_COND_CLOCK, &now); + if (flow->snd_timesout) { - ts_add(&abs, &flow->snd_timeo, &abs); - abstime = &abs; + ts_add(&now, &flow->snd_timeo, &abs); + dl = &abs; } - flags = flow->oflags; - - pthread_rwlock_unlock(&ai.lock); + pthread_rwlock_unlock(&proc.lock); if ((flags & FLOWFACCMODE) == FLOWFRDONLY) return -EPERM; - if (flags & FLOWFWNOBLOCK) { - if (!frcti_is_window_open(flow->frcti)) - return -EAGAIN; - idx = shm_rdrbuff_alloc(ai.rdrb, count, &ptr, &sdb); - } else { - ret = frcti_window_wait(flow->frcti, abstime); + if (flow->crypt != NULL + && (ssm_rbuff_get_flags(flow->rx_rb) & RB_REKEY)) + flow_rekey(flow); + + flow_tx_promote(flow); + + /* Pre-empt TX key exhaustion; the timer is the backstop. */ + if (flow_wm_due(flow)) { + STORE_RELAXED(&flow->rk_wm_inflight, true); + if (flow_rekey_trigger(flow) < 0) + STORE_RELAXED(&flow->rk_wm_inflight, false); + } + + tw_move_safe(); + + if (flow->frcti != NULL) { + /* Pump rx_rb so a pure-writer processes ACKs. */ + ret = flow_wait_window(flow, 1, !(flags & FLOWFWNOBLOCK), dl); if (ret < 0) return ret; - idx = shm_rdrbuff_alloc_b(ai.rdrb, count, &ptr, &sdb, abstime); + + if (count > 0 && FRCTI_IS_STREAM(flow->frcti)) + return flow_write_stream(flow, buf, count, flags, dl); + + if (FRCTI_NEEDS_FRAG(flow->frcti, count)) + return flow_write_frag(flow, buf, count, flags, dl); + } else if (flow->info.mtu > 0 + && count > flow_user_mtu(flow, flow->info.mtu)) { + /* Raw flows carry no PCI; refuse anything > one n-1 frame. */ + return -EMSGSIZE; } + if (flags & FLOWFWNOBLOCK) + idx = ssm_pool_alloc(proc.pool, count, &ptr, &spb); + else + idx = ssm_pool_alloc_b(proc.pool, count, &ptr, &spb, dl); if (idx < 0) return idx; if (count > 0) memcpy(ptr, buf, count); - ret = flow_tx_sdb(flow, sdb, !(flags & FLOWFWNOBLOCK), abstime); - - return ret < 0 ? (ssize_t) ret : (ssize_t) count; -} - -static bool invalid_pkt(struct flow * flow, - struct shm_du_buff * sdb) -{ - if (shm_du_buff_len(sdb) == 0) - return true; - - if (flow->info.qs.ber == 0 && chk_crc(sdb) != 0) - return true; - - if (crypt_decrypt(&flow->crypt, sdb) < 0) - return true; + ret = flow_tx_spb(flow, spb, FRCT_FR_SOLE, + !(flags & FLOWFWNOBLOCK), dl); + if (ret < 0) { + ssm_pool_remove(proc.pool, idx); + return (ssize_t) ret; + } - return false; + return (ssize_t) count; } -static ssize_t flow_rx_sdb(struct flow * flow, - struct shm_du_buff ** sdb, +static ssize_t flow_rx_spb(struct flow * flow, + struct ssm_pk_buff ** spb, bool block, struct timespec * abstime) { ssize_t idx; struct timespec now; - idx = block ? shm_rbuff_read_b(flow->rx_rb, abstime) : - shm_rbuff_read(flow->rx_rb); + idx = block ? ssm_rbuff_read_b(flow->rx_rb, abstime) + : ssm_rbuff_read(flow->rx_rb); if (idx < 0) return idx; clock_gettime(PTHREAD_COND_CLOCK, &now); - - pthread_rwlock_wrlock(&ai.lock); - flow->rcv_act = now; - pthread_rwlock_unlock(&ai.lock); + *spb = ssm_pool_get(proc.pool, idx); - *sdb = shm_rdrbuff_get(ai.rdrb, idx); - if (invalid_pkt(flow, *sdb)) { - shm_rdrbuff_remove(ai.rdrb, idx); + if (invalid_pkt(flow, *spb)) { + ssm_pool_remove(proc.pool, idx); return -EAGAIN; } return idx; } +static ssize_t raw_flow_read_pkt(struct flow * flow, + bool block, + struct timespec * dl) +{ + struct ssm_pk_buff * spb; + struct timespec wait_abs; + ssize_t idx; + + while (true) { + if (flow->crypt != NULL + && (ssm_rbuff_get_flags(flow->rx_rb) & RB_REKEY)) + flow_rekey(flow); + + if (!block) { + idx = ssm_rbuff_read(flow->rx_rb); + if (idx < 0) + return -EAGAIN; + } else { + compute_wait_deadline(dl, &wait_abs); + idx = ssm_rbuff_read_b(flow->rx_rb, &wait_abs); + if (idx == -ETIMEDOUT) { + if (deadline_passed(dl)) + return -ETIMEDOUT; + continue; + } + if (idx < 0) + return idx; + } + + spb = ssm_pool_get(proc.pool, idx); + if (!invalid_pkt(flow, spb)) + return idx; + + ssm_pool_remove(proc.pool, idx); + if (!block) + return -EAGAIN; + } +} + +static ssize_t deliver_pkt(struct flow * flow, + struct ssm_pk_buff * spb, + ssize_t idx, + void * buf, + size_t count, + bool partrd) +{ + uint8_t * packet = ssm_pk_buff_head(spb); + ssize_t n = ssm_pk_buff_len(spb); + + assert(n >= 0); + + if (n <= (ssize_t) count) { + memcpy(buf, packet, n); + ipcp_spb_release(spb); + if (partrd && n == (ssize_t) count) + flow->part_idx = DONE_PART; + else + flow->part_idx = NO_PART; + + return n; + } + + if (partrd) { + memcpy(buf, packet, count); + ssm_pk_buff_pop(spb, n); + flow->part_idx = idx; + return count; + } + + ipcp_spb_release(spb); + return -EMSGSIZE; +} + +/* Drive frcti_consume until it delivers or errors. */ +static ssize_t flow_read_frcti(struct flow * flow, + void * buf, + size_t count, + bool block, + struct timespec * dl) +{ + struct timespec now; + ssize_t bytes; + int rc; + + while (true) { + flow_drain_rx_nb(flow); + bytes = FRCTI_CONSUME(flow->frcti, buf, count); + if (bytes >= 0) + break; + if (bytes != -EAGAIN) + return bytes; + if (!block) + return -EAGAIN; + rc = flow_rx_one(flow, dl); + if (rc < 0) + return rc; + } + + clock_gettime(PTHREAD_COND_CLOCK, &now); + flow->rcv_act = now; + + return bytes; +} + ssize_t flow_read(int fd, void * buf, size_t count) { - ssize_t idx; - ssize_t n; - uint8_t * packet; - struct shm_du_buff * sdb; + struct flow * flow; + struct ssm_pk_buff * spb; struct timespec abs; struct timespec now; - struct timespec * abstime = NULL; - struct flow * flow; + struct timespec * dl = NULL; + ssize_t idx; bool block; bool partrd; - if (fd < 0 || fd >= PROG_MAX_FLOWS) + if (fd < 0 || fd >= PROC_MAX_FLOWS) return -EBADF; - flow = &ai.flows[fd]; + flow = &proc.flows[fd]; - clock_gettime(PTHREAD_COND_CLOCK, &now); - - pthread_rwlock_rdlock(&ai.lock); + pthread_rwlock_rdlock(&proc.lock); if (flow->info.id < 0) { - pthread_rwlock_unlock(&ai.lock); + pthread_rwlock_unlock(&proc.lock); return -ENOTALLOC; } if (flow->part_idx == DONE_PART) { - pthread_rwlock_unlock(&ai.lock); flow->part_idx = NO_PART; + pthread_rwlock_unlock(&proc.lock); return 0; } @@ -1377,75 +2078,40 @@ ssize_t flow_read(int fd, partrd = !(flow->oflags & FLOWFRNOPART); if (flow->rcv_timesout) { + clock_gettime(PTHREAD_COND_CLOCK, &now); ts_add(&now, &flow->rcv_timeo, &abs); - abstime = &abs; + dl = &abs; } - idx = flow->part_idx; - if (idx < 0) { - while ((idx = frcti_queued_pdu(flow->frcti)) < 0) { - pthread_rwlock_unlock(&ai.lock); - - idx = flow_rx_sdb(flow, &sdb, block, abstime); - if (idx < 0) { - if (block && idx != -EAGAIN) - return idx; - if (!block) - return idx; + pthread_rwlock_unlock(&proc.lock); - pthread_rwlock_rdlock(&ai.lock); - continue; - } - - pthread_rwlock_rdlock(&ai.lock); + if (flow->crypt != NULL + && (ssm_rbuff_get_flags(flow->rx_rb) & RB_REKEY)) + flow_rekey(flow); - frcti_rcv(flow->frcti, sdb); - } - } - - sdb = shm_rdrbuff_get(ai.rdrb, idx); - - pthread_rwlock_unlock(&ai.lock); - - packet = shm_du_buff_head(sdb); - - n = shm_du_buff_len(sdb); - - assert(n >= 0); - - if (n <= (ssize_t) count) { - memcpy(buf, packet, n); - ipcp_sdb_release(sdb); + /* Advance TX off a stale epoch even on recv-mostly (ACK-only) flows. */ + flow_tx_promote(flow); - pthread_rwlock_wrlock(&ai.lock); + tw_move_safe(); - flow->part_idx = (partrd && n == (ssize_t) count) ? - DONE_PART : NO_PART; + idx = flow->part_idx; + if (idx < 0 && flow->frcti != NULL) + return flow_read_frcti(flow, buf, count, block, dl); - flow->rcv_act = now; + if (idx < 0) { + idx = raw_flow_read_pkt(flow, block, dl); + if (idx < 0) + return idx; + } - pthread_rwlock_unlock(&ai.lock); - return n; - } else { - if (partrd) { - memcpy(buf, packet, count); - shm_du_buff_head_release(sdb, n); - pthread_rwlock_wrlock(&ai.lock); - flow->part_idx = idx; + spb = ssm_pool_get(proc.pool, idx); - flow->rcv_act = now; + clock_gettime(PTHREAD_COND_CLOCK, &now); + flow->rcv_act = now; - pthread_rwlock_unlock(&ai.lock); - return count; - } else { - ipcp_sdb_release(sdb); - return -EMSGSIZE; - } - } + return deliver_pkt(flow, spb, idx, buf, count, partrd); } -/* fqueue functions. */ - struct flow_set * fset_create(void) { struct flow_set * set; @@ -1454,20 +2120,20 @@ struct flow_set * fset_create(void) if (set == NULL) goto fail_malloc; - assert(ai.fqueues); + assert(proc.fqueues); - pthread_rwlock_wrlock(&ai.lock); + pthread_rwlock_wrlock(&proc.lock); - set->idx = bmp_allocate(ai.fqueues); - if (!bmp_is_id_valid(ai.fqueues, set->idx)) + set->idx = bmp_allocate(proc.fqueues); + if (!bmp_is_id_valid(proc.fqueues, set->idx)) goto fail_bmp_alloc; - pthread_rwlock_unlock(&ai.lock); + pthread_rwlock_unlock(&proc.lock); return set; fail_bmp_alloc: - pthread_rwlock_unlock(&ai.lock); + pthread_rwlock_unlock(&proc.lock); free(set); fail_malloc: return NULL; @@ -1480,11 +2146,11 @@ void fset_destroy(struct flow_set * set) fset_zero(set); - pthread_rwlock_wrlock(&ai.lock); + pthread_rwlock_wrlock(&proc.lock); - bmp_release(ai.fqueues, set->idx); + bmp_release(proc.fqueues, set->idx); - pthread_rwlock_unlock(&ai.lock); + pthread_rwlock_unlock(&proc.lock); free(set); } @@ -1495,7 +2161,7 @@ struct fqueue * fqueue_create(void) if (fq == NULL) return NULL; - memset(fq->fqueue, -1, SHM_BUFFER_SIZE * sizeof(*fq->fqueue)); + memset(fq->fqueue, -1, SSM_RBUFF_SIZE * sizeof(*fq->fqueue)); fq->fqsize = 0; fq->next = 0; @@ -1512,7 +2178,7 @@ void fset_zero(struct flow_set * set) if (set == NULL) return; - shm_flow_set_zero(ai.fqset, set->idx); + ssm_flow_set_zero(proc.fqset, set->idx); } int fset_add(struct flow_set * set, @@ -1521,12 +2187,12 @@ int fset_add(struct flow_set * set, struct flow * flow; int ret; - if (set == NULL || fd < 0 || fd >= SYS_MAX_FLOWS) + if (set == NULL || fd < 0 || fd >= PROC_MAX_FLOWS) return -EINVAL; - flow = &ai.flows[fd]; + flow = &proc.flows[fd]; - pthread_rwlock_rdlock(&ai.lock); + pthread_rwlock_rdlock(&proc.lock); if (flow->info.id < 0) { ret = -EINVAL; @@ -1534,21 +2200,21 @@ int fset_add(struct flow_set * set, } if (flow->frcti != NULL) - shm_flow_set_del(ai.fqset, 0, ai.flows[fd].info.id); + ssm_flow_set_del(proc.fqset, 0, flow->info.id); - ret = shm_flow_set_add(ai.fqset, set->idx, ai.flows[fd].info.id); + ret = ssm_flow_set_add(proc.fqset, set->idx, flow->info.id); if (ret < 0) goto fail; - if (shm_rbuff_queued(ai.flows[fd].rx_rb)) - shm_flow_set_notify(ai.fqset, ai.flows[fd].info.id, FLOW_PKT); + if (ssm_rbuff_queued(flow->rx_rb)) + ssm_flow_set_notify(proc.fqset, flow->info.id, FLOW_PKT); - pthread_rwlock_unlock(&ai.lock); + pthread_rwlock_unlock(&proc.lock); return ret; fail: - pthread_rwlock_unlock(&ai.lock); + pthread_rwlock_unlock(&proc.lock); return ret; } @@ -1557,99 +2223,112 @@ void fset_del(struct flow_set * set, { struct flow * flow; - if (set == NULL || fd < 0 || fd >= SYS_MAX_FLOWS) + if (set == NULL || fd < 0 || fd >= PROC_MAX_FLOWS) return; - flow = &ai.flows[fd]; + flow = &proc.flows[fd]; - pthread_rwlock_rdlock(&ai.lock); + pthread_rwlock_rdlock(&proc.lock); if (flow->info.id >= 0) - shm_flow_set_del(ai.fqset, set->idx, flow->info.id); + ssm_flow_set_del(proc.fqset, set->idx, flow->info.id); if (flow->frcti != NULL) - shm_flow_set_add(ai.fqset, 0, ai.flows[fd].info.id); + ssm_flow_set_add(proc.fqset, 0, flow->info.id); - pthread_rwlock_unlock(&ai.lock); + pthread_rwlock_unlock(&proc.lock); } bool fset_has(const struct flow_set * set, int fd) { - bool ret; + struct flow * flow; + bool ret; - if (set == NULL || fd < 0 || fd >= SYS_MAX_FLOWS) + if (set == NULL || fd < 0 || fd >= PROC_MAX_FLOWS) return false; - pthread_rwlock_rdlock(&ai.lock); + flow = &proc.flows[fd]; + + pthread_rwlock_rdlock(&proc.lock); - if (ai.flows[fd].info.id < 0) { - pthread_rwlock_unlock(&ai.lock); + if (flow->info.id < 0) { + pthread_rwlock_unlock(&proc.lock); return false; } - ret = (shm_flow_set_has(ai.fqset, set->idx, ai.flows[fd].info.id) == 1); + ret = (ssm_flow_set_has(proc.fqset, set->idx, flow->info.id) == 1); - pthread_rwlock_unlock(&ai.lock); + pthread_rwlock_unlock(&proc.lock); return ret; } -/* Filter fqueue events for non-data packets */ static int fqueue_filter(struct fqueue * fq) { - struct shm_du_buff * sdb; + struct ssm_pk_buff * spb; int fd; ssize_t idx; struct frcti * frcti; + int ret = 0; + + /* proc.lock rdlock gates frcti_destroy via flow_fini wrlock. */ + pthread_rwlock_rdlock(&proc.lock); while (fq->next < fq->fqsize) { - if (fq->fqueue[fq->next].event != FLOW_PKT) - return 1; + if (fq->fqueue[fq->next].event == FLOW_UPD) { + /* Re-key doorbell: pull internally, never surface. */ + fd = proc.id_to_fd[fq->fqueue[fq->next].flow_id].fd; + ++fq->next; + if (fd >= 0) { + pthread_rwlock_unlock(&proc.lock); + flow_rekey(&proc.flows[fd]); + pthread_rwlock_rdlock(&proc.lock); + } + continue; + } - pthread_rwlock_rdlock(&ai.lock); + if (fq->fqueue[fq->next].event != FLOW_PKT) { + ret = 1; + goto out; + } - fd = ai.id_to_fd[fq->fqueue[fq->next].flow_id].fd; + fd = proc.id_to_fd[fq->fqueue[fq->next].flow_id].fd; if (fd < 0) { ++fq->next; - pthread_rwlock_unlock(&ai.lock); continue; } - frcti = ai.flows[fd].frcti; + frcti = proc.flows[fd].frcti; if (frcti == NULL) { - pthread_rwlock_unlock(&ai.lock); - return 1; + ret = 1; + goto out; } - if (__frcti_pdu_ready(frcti) >= 0) { - pthread_rwlock_unlock(&ai.lock); - return 1; + if (FRCTI_PDU_READY(frcti)) { + ret = 1; + goto out; } - pthread_rwlock_unlock(&ai.lock); - - idx = flow_rx_sdb(&ai.flows[fd], &sdb, false, NULL); + idx = flow_rx_spb(&proc.flows[fd], &spb, false, NULL); if (idx < 0) - return 0; + goto out; - pthread_rwlock_rdlock(&ai.lock); + spb = ssm_pool_get(proc.pool, idx); - sdb = shm_rdrbuff_get(ai.rdrb, idx); + FRCTI_RCV(frcti, spb); - __frcti_rcv(frcti, sdb); - - if (__frcti_pdu_ready(frcti) >= 0) { - pthread_rwlock_unlock(&ai.lock); - return 1; + if (FRCTI_PDU_READY(frcti)) { + ret = 1; + goto out; } - pthread_rwlock_unlock(&ai.lock); - ++fq->next; } - return 0; + out: + pthread_rwlock_unlock(&proc.lock); + return ret; } int fqueue_next(struct fqueue * fq) @@ -1666,15 +2345,15 @@ int fqueue_next(struct fqueue * fq) if (fq->next != 0 && fqueue_filter(fq) == 0) return -EPERM; - pthread_rwlock_rdlock(&ai.lock); + pthread_rwlock_rdlock(&proc.lock); e = fq->fqueue + fq->next; - fd = ai.id_to_fd[e->flow_id].fd; + fd = proc.id_to_fd[e->flow_id].fd; ++fq->next; - pthread_rwlock_unlock(&ai.lock); + pthread_rwlock_unlock(&proc.lock); return fd; } @@ -1696,7 +2375,8 @@ ssize_t fevent(struct flow_set * set, { ssize_t ret = 0; struct timespec abs; - struct timespec * t = NULL; + struct timespec * dl = NULL; + struct timespec wait_abs; if (set == NULL || fq == NULL) return -EINVAL; @@ -1704,17 +2384,26 @@ ssize_t fevent(struct flow_set * set, if (fq->fqsize > 0 && fq->next != fq->fqsize) return 1; - clock_gettime(PTHREAD_COND_CLOCK, &abs); - if (timeo != NULL) { - ts_add(&abs, timeo, &abs); - t = &abs; + struct timespec now; + clock_gettime(PTHREAD_COND_CLOCK, &now); + ts_add(&now, timeo, &abs); + dl = &abs; } while (ret == 0) { - ret = shm_flow_set_wait(ai.fqset, set->idx, fq->fqueue, t); - if (ret == -ETIMEDOUT) - return -ETIMEDOUT; + tw_move_safe(); + + compute_wait_deadline(dl, &wait_abs); + + ret = ssm_flow_set_wait(proc.fqset, set->idx, + fq->fqueue, &wait_abs); + if (ret == -ETIMEDOUT) { + if (deadline_passed(dl)) + return -ETIMEDOUT; + ret = 0; + continue; + } fq->fqsize = ret; fq->next = 0; @@ -1727,12 +2416,12 @@ ssize_t fevent(struct flow_set * set, return 1; } -/* ipcp-dev functions. */ - int np1_flow_alloc(pid_t n_pid, int flow_id) { - struct flow_info flow; + struct flow_info flow; + struct crypt_sk crypt = { .nid = NID_undef, .key = NULL, + .epoch = 0, .role = CRYPT_ROLE_INIT }; memset(&flow, 0, sizeof(flow)); @@ -1740,9 +2429,10 @@ int np1_flow_alloc(pid_t n_pid, flow.n_pid = getpid(); flow.qs = qos_np1; flow.mpl = 0; - flow.n_1_pid = n_pid; /* This "flow" is upside-down! */ + /* np1 flow: n_1_pid is the upper. */ + flow.n_1_pid = n_pid; - return flow_init(&flow, NULL); + return flow_init(&flow, &crypt, 0); } int np1_flow_dealloc(int flow_id, @@ -1750,43 +2440,71 @@ int np1_flow_dealloc(int flow_id, { int fd; - /* - * TODO: Don't pass timeo to the IPCP but wait in IRMd. - * This will need async ops, waiting until we bootstrap - * the IRMd over ouroboros. - */ - + /* TODO: wait in IRMd, not here; needs async ops. */ sleep(timeo); - pthread_rwlock_rdlock(&ai.lock); + pthread_rwlock_rdlock(&proc.lock); + + fd = proc.id_to_fd[flow_id].fd; - fd = ai.id_to_fd[flow_id].fd; + pthread_rwlock_unlock(&proc.lock); - pthread_rwlock_unlock(&ai.lock); + return fd; +} + +int np1_flow_resp(int flow_id, + int resp) +{ + int fd; + + if (resp == 0 && flow_wait_assign(flow_id) != FLOW_ALLOCATED) + return -1; + + pthread_rwlock_rdlock(&proc.lock); + + fd = proc.id_to_fd[flow_id].fd; + + pthread_rwlock_unlock(&proc.lock); return fd; } -int np1_flow_resp(int flow_id) +int np1_flow_fd(int flow_id) { int fd; - if (flow_wait_assign(flow_id) != FLOW_ALLOCATED) + if (flow_id < 0 || flow_id >= SYS_MAX_FLOWS) return -1; - pthread_rwlock_rdlock(&ai.lock); + pthread_rwlock_rdlock(&proc.lock); - fd = ai.id_to_fd[flow_id].fd; + fd = proc.id_to_fd[flow_id].fd; - pthread_rwlock_unlock(&ai.lock); + pthread_rwlock_unlock(&proc.lock); return fd; } +int np1_flow_id(int fd) +{ + int flow_id; + + if (fd < 0 || fd >= PROC_MAX_FLOWS) + return -1; + + pthread_rwlock_rdlock(&proc.lock); + + flow_id = proc.flows[fd].info.id; + + pthread_rwlock_unlock(&proc.lock); + + return flow_id; +} + int ipcp_create_r(const struct ipcp_info * info) { uint8_t buf[SOCK_BUF_SIZE]; - buffer_t msg = {buf, SOCK_BUF_SIZE}; + buffer_t msg = {SOCK_BUF_SIZE, buf}; int err; if (ipcp_create_r__irm_req_ser(&msg,info) < 0) @@ -1802,12 +2520,15 @@ int ipcp_create_r(const struct ipcp_info * info) int ipcp_flow_req_arr(const buffer_t * dst, qosspec_t qs, time_t mpl, + uint32_t mtu, const buffer_t * data) { struct flow_info flow; - uint8_t buf[SOCK_BUF_SIZE]; - buffer_t msg = {buf, SOCK_BUF_SIZE}; - int err; + uint8_t buf[SOCK_BUF_SIZE]; + buffer_t msg = {SOCK_BUF_SIZE, buf}; + struct crypt_sk crypt; + uint8_t key[SYMMKEYSZ]; + int err; memset(&flow, 0, sizeof(flow)); @@ -1816,6 +2537,7 @@ int ipcp_flow_req_arr(const buffer_t * dst, flow.n_1_pid = getpid(); flow.qs = qs; flow.mpl = mpl; + flow.mtu = mtu; if (ipcp_flow_req_arr__irm_req_ser(&msg, dst, &flow, data) < 0) return -ENOMEM; @@ -1824,37 +2546,73 @@ int ipcp_flow_req_arr(const buffer_t * dst, if (err < 0) return err; - err = flow__irm_result_des(&msg, &flow, NULL); + crypt.key = key; + crypt.epoch = 0; + crypt.role = CRYPT_ROLE_INIT; + + err = flow__irm_result_des(&msg, &flow, &crypt); if (err < 0) return err; - /* inverted for np1_flow */ + /* np1 flows are not encrypted. */ + assert(crypt.nid == NID_undef); + + /* Inverted for np1_flow. */ flow.n_1_pid = flow.n_pid; flow.n_pid = getpid(); flow.mpl = 0; + flow.mtu = 0; + flow.qs = qos_np1; + + crypt.nid = NID_undef; + + return flow_init(&flow, &crypt, 0); +} + +int ipcp_flow_update_arr(int flow_id, + const buffer_t * data) +{ + struct flow_info flow; + uint8_t buf[SOCK_BUF_SIZE]; + buffer_t msg = {SOCK_BUF_SIZE, buf}; + int err; + + memset(&flow, 0, sizeof(flow)); + + flow.id = flow_id; + flow.n_1_pid = getpid(); + + if (ipcp_flow_update_arr__irm_req_ser(&msg, &flow, data) < 0) + return -ENOMEM; + + err = send_recv_msg(&msg); + if (err < 0) + return err; - return flow_init(&flow, NULL); + return irm__irm_result_des(&msg); } int ipcp_flow_alloc_reply(int fd, int response, time_t mpl, + uint32_t mtu, const buffer_t * data) { struct flow_info flow; uint8_t buf[SOCK_BUF_SIZE]; - buffer_t msg = {buf, SOCK_BUF_SIZE}; + buffer_t msg = {SOCK_BUF_SIZE, buf}; int err; - assert(fd >= 0 && fd < SYS_MAX_FLOWS); + assert(fd >= 0 && fd < PROC_MAX_FLOWS); - pthread_rwlock_rdlock(&ai.lock); + pthread_rwlock_rdlock(&proc.lock); - flow.id = ai.flows[fd].info.id; + flow.id = proc.flows[fd].info.id; - pthread_rwlock_unlock(&ai.lock); + pthread_rwlock_unlock(&proc.lock); flow.mpl = mpl; + flow.mtu = mtu; if (ipcp_flow_alloc_reply__irm_msg_ser(&msg, &flow, response, data) < 0) return -ENOMEM; @@ -1867,169 +2625,224 @@ int ipcp_flow_alloc_reply(int fd, } int ipcp_flow_read(int fd, - struct shm_du_buff ** sdb) + struct ssm_pk_buff ** spb) { struct flow * flow; ssize_t idx = -1; - assert(fd >= 0 && fd < SYS_MAX_FLOWS); - assert(sdb); + assert(fd >= 0 && fd < PROC_MAX_FLOWS); + assert(spb); - flow = &ai.flows[fd]; + flow = &proc.flows[fd]; - pthread_rwlock_rdlock(&ai.lock); + pthread_rwlock_rdlock(&proc.lock); assert(flow->info.id >= 0); - while (frcti_queued_pdu(flow->frcti) < 0) { - pthread_rwlock_unlock(&ai.lock); + /* Raw flow: deliver the popped pkt directly (no FRCT rq). */ + if (flow->frcti == NULL) { + pthread_rwlock_unlock(&proc.lock); + idx = flow_rx_spb(flow, spb, false, NULL); + return idx < 0 ? (int) idx : 0; + } + + while (!FRCTI_PDU_READY(flow->frcti)) { + pthread_rwlock_unlock(&proc.lock); - idx = flow_rx_sdb(flow, sdb, false, NULL); + idx = flow_rx_spb(flow, spb, false, NULL); if (idx < 0) return idx; - pthread_rwlock_rdlock(&ai.lock); + pthread_rwlock_rdlock(&proc.lock); - frcti_rcv(flow->frcti, *sdb); + FRCTI_RCV(flow->frcti, *spb); } - pthread_rwlock_unlock(&ai.lock); + pthread_rwlock_unlock(&proc.lock); return 0; } int ipcp_flow_write(int fd, - struct shm_du_buff * sdb) + struct ssm_pk_buff * spb) { struct flow * flow; int ret; - assert(fd >= 0 && fd < SYS_MAX_FLOWS); - assert(sdb); + assert(fd >= 0 && fd < PROC_MAX_FLOWS); + assert(spb); - flow = &ai.flows[fd]; + flow = &proc.flows[fd]; - pthread_rwlock_wrlock(&ai.lock); + pthread_rwlock_rdlock(&proc.lock); if (flow->info.id < 0) { - pthread_rwlock_unlock(&ai.lock); + pthread_rwlock_unlock(&proc.lock); return -ENOTALLOC; } if ((flow->oflags & FLOWFACCMODE) == FLOWFRDONLY) { - pthread_rwlock_unlock(&ai.lock); + pthread_rwlock_unlock(&proc.lock); return -EPERM; } - pthread_rwlock_unlock(&ai.lock); + pthread_rwlock_unlock(&proc.lock); - ret = flow_tx_sdb(flow, sdb, true, NULL); + ret = flow_tx_spb(flow, spb, FRCT_FR_SOLE, true, NULL); return ret; } +/* Copy src into dst_pool without consuming src. Caller owns both halves. */ +static int pool_dup_spb(struct ssm_pool * src_pool, + size_t src_off, + struct ssm_pool * dst_pool, + struct ssm_pk_buff ** dst_spb) +{ + struct ssm_pk_buff * src; + uint8_t * ptr; + size_t len; + + src = ssm_pool_get(src_pool, src_off); + len = ssm_pk_buff_len(src); + + if (ssm_pool_alloc(dst_pool, len, &ptr, dst_spb) < 0) + return -ENOMEM; + + memcpy(ptr, ssm_pk_buff_head(src), len); + + return 0; +} + int np1_flow_read(int fd, - struct shm_du_buff ** sdb) + struct ssm_pk_buff ** spb, + struct ssm_pool * pool) { - struct flow * flow; - ssize_t idx = -1; + struct flow * flow; + ssize_t off; - assert(fd >= 0 && fd < SYS_MAX_FLOWS); - assert(sdb); + assert(fd >= 0 && fd < PROC_MAX_FLOWS); + assert(spb); - flow = &ai.flows[fd]; + flow = &proc.flows[fd]; assert(flow->info.id >= 0); - pthread_rwlock_rdlock(&ai.lock); + pthread_rwlock_rdlock(&proc.lock); - idx = shm_rbuff_read(flow->rx_rb);; - if (idx < 0) { - pthread_rwlock_unlock(&ai.lock); - return idx; + off = ssm_rbuff_read(flow->rx_rb); + if (off < 0) { + pthread_rwlock_unlock(&proc.lock); + return off; } - pthread_rwlock_unlock(&ai.lock); + pthread_rwlock_unlock(&proc.lock); - *sdb = shm_rdrbuff_get(ai.rdrb, idx); + if (pool == NULL) { + *spb = ssm_pool_get(proc.pool, off); + } else { + /* Cross-pool copy: PUP -> GSPP */ + if (pool_dup_spb(pool, off, proc.pool, spb) < 0) { + ssm_pool_remove(pool, off); + return -ENOMEM; + } + ssm_pool_remove(pool, off); + } return 0; } int np1_flow_write(int fd, - struct shm_du_buff * sdb) + struct ssm_pk_buff * spb, + struct ssm_pool * pool) { - struct flow * flow; - int ret; - ssize_t idx; + struct flow * flow; + struct ssm_pk_buff * dst; + int ret; + size_t off; + size_t dst_off; - assert(fd >= 0 && fd < SYS_MAX_FLOWS); - assert(sdb); + assert(fd >= 0 && fd < PROC_MAX_FLOWS); + assert(spb); - flow = &ai.flows[fd]; + flow = &proc.flows[fd]; - pthread_rwlock_rdlock(&ai.lock); + pthread_rwlock_rdlock(&proc.lock); if (flow->info.id < 0) { - pthread_rwlock_unlock(&ai.lock); + pthread_rwlock_unlock(&proc.lock); return -ENOTALLOC; } if ((flow->oflags & FLOWFACCMODE) == FLOWFRDONLY) { - pthread_rwlock_unlock(&ai.lock); + pthread_rwlock_unlock(&proc.lock); return -EPERM; } - pthread_rwlock_unlock(&ai.lock); + pthread_rwlock_unlock(&proc.lock); - idx = shm_du_buff_get_idx(sdb); + off = ssm_pk_buff_get_off(spb); - ret = shm_rbuff_write_b(flow->tx_rb, idx, NULL); - if (ret < 0) - shm_rdrbuff_remove(ai.rdrb, idx); - else - shm_flow_set_notify(flow->set, flow->info.id, FLOW_PKT); + if (pool == NULL) { + ret = ssm_rbuff_write_b(flow->tx_rb, off, NULL); + if (ret < 0) + return ret; + ssm_flow_set_notify(flow->set, flow->info.id, FLOW_PKT); + } else { + /* Cross-pool copy: GSPP -> PUP. Src kept on error. */ + if (pool_dup_spb(proc.pool, off, pool, &dst) < 0) + return -ENOMEM; + dst_off = ssm_pk_buff_get_off(dst); + ret = ssm_rbuff_write_b(flow->tx_rb, dst_off, NULL); + if (ret < 0) { + ssm_pool_remove(pool, dst_off); + return ret; + } + ssm_flow_set_notify(flow->set, flow->info.id, FLOW_PKT); + ssm_pool_remove(proc.pool, off); + } - return ret; + return 0; } -int ipcp_sdb_reserve(struct shm_du_buff ** sdb, +int ipcp_spb_reserve(struct ssm_pk_buff ** spb, size_t len) { - return shm_rdrbuff_alloc_b(ai.rdrb, len, NULL, sdb, NULL) < 0 ? -1 : 0; + return ssm_pool_alloc_b(proc.pool, len, NULL, spb, NULL) < 0 + ? -1 : 0; } -void ipcp_sdb_release(struct shm_du_buff * sdb) +void ipcp_spb_release(struct ssm_pk_buff * spb) { - shm_rdrbuff_remove(ai.rdrb, shm_du_buff_get_idx(sdb)); + ssm_pool_remove(proc.pool, ssm_pk_buff_get_off(spb)); } int ipcp_flow_fini(int fd) { - struct shm_rbuff * rx_rb; + struct ssm_rbuff * rx_rb; - assert(fd >= 0 && fd < SYS_MAX_FLOWS); + assert(fd >= 0 && fd < PROC_MAX_FLOWS); - pthread_rwlock_rdlock(&ai.lock); + pthread_rwlock_rdlock(&proc.lock); - if (ai.flows[fd].info.id < 0) { - pthread_rwlock_unlock(&ai.lock); + if (proc.flows[fd].info.id < 0) { + pthread_rwlock_unlock(&proc.lock); return -1; } - shm_rbuff_set_acl(ai.flows[fd].rx_rb, ACL_FLOWDOWN); - shm_rbuff_set_acl(ai.flows[fd].tx_rb, ACL_FLOWDOWN); + ssm_rbuff_set_bits(proc.flows[fd].rx_rb, RB_FLOWDOWN); + ssm_rbuff_set_bits(proc.flows[fd].tx_rb, RB_FLOWDOWN); - shm_flow_set_notify(ai.flows[fd].set, - ai.flows[fd].info.id, + ssm_flow_set_notify(proc.flows[fd].set, + proc.flows[fd].info.id, FLOW_DEALLOC); - rx_rb = ai.flows[fd].rx_rb; + rx_rb = proc.flows[fd].rx_rb; - pthread_rwlock_unlock(&ai.lock); + pthread_rwlock_unlock(&proc.lock); if (rx_rb != NULL) - shm_rbuff_fini(rx_rb); + ssm_rbuff_fini(rx_rb); return 0; } @@ -2037,16 +2850,16 @@ int ipcp_flow_fini(int fd) int ipcp_flow_get_qoscube(int fd, qoscube_t * cube) { - assert(fd >= 0 && fd < SYS_MAX_FLOWS); + assert(fd >= 0 && fd < PROC_MAX_FLOWS); assert(cube); - pthread_rwlock_rdlock(&ai.lock); + pthread_rwlock_rdlock(&proc.lock); - assert(ai.flows[fd].info.id >= 0); + assert(proc.flows[fd].info.id >= 0); - *cube = qos_spec_to_cube(ai.flows[fd].info.qs); + *cube = qos_spec_to_cube(proc.flows[fd].info.qs); - pthread_rwlock_unlock(&ai.lock); + pthread_rwlock_unlock(&proc.lock); return 0; } @@ -2055,56 +2868,79 @@ size_t ipcp_flow_queued(int fd) { size_t q; - pthread_rwlock_rdlock(&ai.lock); + pthread_rwlock_rdlock(&proc.lock); - assert(ai.flows[fd].info.id >= 0); + assert(proc.flows[fd].info.id >= 0); - q = shm_rbuff_queued(ai.flows[fd].tx_rb); + q = ssm_rbuff_queued(proc.flows[fd].tx_rb); - pthread_rwlock_unlock(&ai.lock); + pthread_rwlock_unlock(&proc.lock); return q; } -ssize_t local_flow_read(int fd) +int local_flow_transfer(int src_fd, + int dst_fd, + struct ssm_pool * src_pool, + struct ssm_pool * dst_pool) { - ssize_t ret; - - assert(fd >= 0); - - pthread_rwlock_rdlock(&ai.lock); - - ret = shm_rbuff_read(ai.flows[fd].rx_rb); + struct flow * src_flow; + struct flow * dst_flow; + struct ssm_pk_buff * dst_spb; + struct ssm_pool * sp; + struct ssm_pool * dp; + ssize_t off; + int ret; - pthread_rwlock_unlock(&ai.lock); + assert(src_fd >= 0); + assert(dst_fd >= 0); - return ret; -} - -int local_flow_write(int fd, - size_t idx) -{ - struct flow * flow; - int ret; + src_flow = &proc.flows[src_fd]; + dst_flow = &proc.flows[dst_fd]; - assert(fd >= 0); + sp = src_pool == NULL ? proc.pool : src_pool; + dp = dst_pool == NULL ? proc.pool : dst_pool; - flow = &ai.flows[fd]; + pthread_rwlock_rdlock(&proc.lock); - pthread_rwlock_rdlock(&ai.lock); + off = ssm_rbuff_read(src_flow->rx_rb); + if (off < 0) { + pthread_rwlock_unlock(&proc.lock); + return off; + } - if (flow->info.id < 0) { - pthread_rwlock_unlock(&ai.lock); + if (dst_flow->info.id < 0) { + pthread_rwlock_unlock(&proc.lock); + ssm_pool_remove(sp, off); return -ENOTALLOC; } - ret = shm_rbuff_write_b(flow->tx_rb, idx, NULL); - if (ret == 0) - shm_flow_set_notify(flow->set, flow->info.id, FLOW_PKT); - else - shm_rdrbuff_remove(ai.rdrb, idx); + pthread_rwlock_unlock(&proc.lock); - pthread_rwlock_unlock(&ai.lock); + if (sp == dp) { + /* Same pool: zero-copy */ + ret = ssm_rbuff_write_b(dst_flow->tx_rb, off, NULL); + if (ret < 0) + ssm_pool_remove(sp, off); + else + ssm_flow_set_notify(dst_flow->set, + dst_flow->info.id, FLOW_PKT); + } else { + /* Different pools: single copy */ + if (pool_dup_spb(sp, off, dp, &dst_spb) < 0) { + ssm_pool_remove(sp, off); + return -ENOMEM; + } + + ssm_pool_remove(sp, off); + off = ssm_pk_buff_get_off(dst_spb); + ret = ssm_rbuff_write_b(dst_flow->tx_rb, off, NULL); + if (ret < 0) + ssm_pool_remove(dp, off); + else + ssm_flow_set_notify(dst_flow->set, + dst_flow->info.id, FLOW_PKT); + } return ret; } diff --git a/src/lib/frct.c b/src/lib/frct.c index c6fef35c..c055433d 100644 --- a/src/lib/frct.c +++ b/src/lib/frct.c @@ -1,7 +1,7 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * - * Flow and Retransmission Control + * Flow and Retransmission Control Task (FRCT) * * Dimitri Staessens <dimitri@ouroboros.rocks> * Sander Vrijders <sander@ouroboros.rocks> @@ -20,97 +20,416 @@ * Foundation, Inc., http://www.fsf.org/about/contact/. */ -#include <ouroboros/endian.h> +/* Included by dev.c; uses dev.c statics (proc, spb_encrypt, ...). */ #define DELT_RDV (100 * MILLION) /* ns */ -#define MAX_RDV (1 * BILLION) /* ns */ +#define MAX_RDV (1 * BILLION) /* ns */ + +#define MAX_RTO_MUL 8 /* caps the RTO backoff shift */ +#define MAX_TLP_PER_EP 2 /* RFC 8985 §7.3: up to 2 TLPs */ +#define INITIAL_RTO (1 * BILLION) /* RFC 6298 §2.1: 1 s default */ +#define RTT_BOOT_NS (10 * MILLION) /* rtt_hint floor + initial mdev */ +#define SRTT_FLOOR_NS 1000L /* 1 us; smoothed RTT floor */ +#define MDEV_FLOOR_NS 100L /* 100 ns; mdev sanity floor */ +#define RTT_CLAMP_MUL 16 /* probe sample cap = N * srtt */ +#define MIN_RTT_WIN_NS (300ULL * BILLION) /* 5 min, Linux tcp default */ +#define NACK_COOLDOWN_NS (100 * MILLION) /* pre-DRF NACK cooldown */ +#define FRCT_TX_TIMEO_NS (250 * 1000) /* tx ring write deadline */ +#define ACK_DELAY_NS (2ULL * TICTIME) /* delayed-ACK fire delay */ #define FRCT "frct" #define FRCT_PCILEN (sizeof(struct frct_pci)) #define FRCT_NAME_STRLEN 32 -struct frct_cr { - uint32_t lwe; /* Left window edge */ - uint32_t rwe; /* Right window edge */ +/* Wire-protocol cap on SACK blocks per packet; binds both peers. */ +#define SACK_MAX_BLOCKS 2048 +#define SACK_BLOCK_SIZE (2 * sizeof(uint32_t)) +/* 2B count + 2B pad to 4-byte align the block list. */ +#define SACK_HDR_SIZE (sizeof(uint32_t)) +#define SACK_MIN_GAP_NS (250u * 1000u) /* 250 us SACK gap */ +#define MIN_REORDER_NS (250u * 1000u) /* 250 us RACK floor */ +#define SACK_RXM_MAX 32 /* Cap on retransmits staged from single SACK.*/ +#define DUP_THRESH 3 /* RFC 8985 §6.2 step 2.2 SACK count gate. */ + +/* RFC 8985 §7.2 RACK reorder-window scaling cap. */ +#define REO_WND_MULT_MAX 20 +/* RFC 8985 §7.2 step 5: round trips of no DSACK before halving. */ +#define REO_DECAY_PKTS 16 +/* DSACK seqno sanity: reject reports older/farther than one rcv window. */ +#define MAX_DSACK_LAG RQ_SIZE + +/* Signed ns elapsed; negative under concurrent update (no underflow). */ +static __inline__ int64_t ts_age_ns(uint64_t now_ns, + uint64_t then_ns) +{ + return (int64_t)(now_ns - then_ns); +} - uint8_t cflags; - uint32_t seqno; /* SEQ to send, or last SEQ Ack'd */ +/* True iff strictly more than thr_ns elapsed since then_ns. */ +static __inline__ bool ts_aged_ns(uint64_t now_ns, + uint64_t then_ns, + uint64_t thr_ns) +{ + return ts_age_ns(now_ns, then_ns) > (int64_t) thr_ns; +} - struct timespec act; /* Last seen activity */ - time_t inact; /* Inactivity (s) */ -}; +/* FRCT r-timer: do not retransmit packet older than t_r (from first send). */ +#define RXM_AGED_OUT(t0, now_ns, t_r) \ + ts_aged_ns((now_ns), (t0), (uint64_t)(t_r)) -struct frcti { - int fd; +/* FRCT a-timer: do not (re)transmit ACK after t_a from last data receive. */ +#define ACK_AGED_OUT(act, now_ns, t_a) \ + ts_aged_ns((now_ns), (act), (uint64_t)(t_a)) - time_t mpl; - time_t a; - time_t r; - time_t rdv; - - time_t srtt; /* Smoothed rtt */ - time_t mdev; /* Deviation */ - time_t rto; /* Retransmission timeout */ - uint32_t rttseq; - struct timespec t_probe; /* Probe time */ - bool probe; /* Probe active */ -#ifdef PROC_FLOW_STATS - size_t n_rtx; /* Number of rxm packets */ - size_t n_prb; /* Number of rtt probes */ - size_t n_rtt; /* Number of estimates */ - size_t n_dup; /* Duplicates received */ - size_t n_dak; /* Delayed ACKs received */ - size_t n_rdv; /* Number of rdv packets */ - size_t n_out; /* Packets out of window */ - size_t n_rqo; /* Packets out of rqueue */ -#endif - struct frct_cr snd_cr; - struct frct_cr rcv_cr; +struct sack_args { + uint16_t n; + bool dsack; /* RFC 2883: block[0] is a DSACK report */ + uint32_t ack; + uint32_t rwe; + uint32_t blocks[][2]; /* flexible — sized at alloc time */ +}; +/* NewReno-careful (RFC 6582) exit pad; gates RTT samples post-signal. */ +#define RTT_QUARANTINE 32 +#define RTTP_NONCE_LEN 16 - ssize_t rq[RQ_SIZE]; - pthread_rwlock_t lock; +/* RTT-probe wire payload (after the FRCT PCI). */ +struct frct_rttp { + uint32_t probe_id; /* sender counter; 0 on reply */ + uint32_t echo_id; /* peer's probe_id; 0 outbound */ + uint8_t nonce[RTTP_NONCE_LEN]; /* random; echoed verbatim */ +} __attribute__((packed)); - bool open; /* Window open/closed */ - struct timespec t_wnd; /* Window closed time */ - struct timespec t_rdvs; /* Last rendez-vous sent */ - pthread_cond_t cond; - pthread_mutex_t mtx; -}; +#define RTTP_PAYLOAD sizeof(struct frct_rttp) +#define RTTP_POS(id) ((id) & (RTTP_RING - 1)) +/* + * Flag values are assigned MSB-first on the wire (RFC convention): + * bit 0 = 0x8000 occupies wire-position 0 of the 16-bit flags + * field, bit 12 = 0x0008 is the last assigned bit, and the three + * LSBs (0x0007) are reserved. + */ enum frct_flags { - FRCT_DATA = 0x01, /* PDU carries data */ - FRCT_DRF = 0x02, /* Data run flag */ - FRCT_ACK = 0x04, /* ACK field valid */ - FRCT_FC = 0x08, /* FC window valid */ - FRCT_RDVS = 0x10, /* Rendez-vous */ - FRCT_FFGM = 0x20, /* First Fragment */ - FRCT_MFGM = 0x40, /* More fragments */ + FRCT_DATA = 0x8000, /* PDU carries data */ + FRCT_DRF = 0x4000, /* Data run flag */ + FRCT_ACK = 0x2000, /* ACK field valid */ + FRCT_NACK = 0x1000, /* Neg-ACK: pci->seqno is arrival_seqno - 1 */ + FRCT_FC = 0x0800, /* FC window valid */ + FRCT_RDVS = 0x0400, /* Rendez-vous */ + FRCT_FFGM = 0x0200, /* First fragment (begin) */ + FRCT_LFGM = 0x0100, /* Last fragment (end) */ + FRCT_RXM = 0x0080, /* Retransmission */ + FRCT_SACK = 0x0040, /* SACK block list follows */ + FRCT_RTTP = 0x0020, /* RTT probe / echo */ + FRCT_KA = 0x0010, /* Keepalive */ + FRCT_FIN = 0x0008, /* End of stream */ }; -struct frct_pci { - uint8_t flags; +/* + * DATA-packet fragment role (FFGM = begin, LFGM = end), SCTP-style: + * 1 1 = sole / un-fragmented SDU (begin AND end) + * 1 0 = first fragment of a multi-fragment SDU + * 0 0 = middle fragment + * 0 1 = last fragment + */ +#define FRCT_FR_MASK (FRCT_FFGM | FRCT_LFGM) +#define FRCT_FR_SOLE (FRCT_FFGM | FRCT_LFGM) +#define FRCT_FR_FIRST (FRCT_FFGM) +#define FRCT_FR_MID (0) +#define FRCT_FR_LAST (FRCT_LFGM) + +/* Default cap on a single reassembled SDU. App can raise via FRCTSMAXSDU */ +#define FRCT_MAX_SDU (1U << 20) + +/* Stream-mode PCI extension: [start, end) byte range on every DATA pkt. */ +struct frct_pci_stream { + uint32_t start; + uint32_t end; +} __attribute__((packed)); + +#define FRCT_PCI_STREAM_LEN (sizeof(struct frct_pci_stream)) - uint8_t pad; /* 24 bit window! */ - uint16_t window; +/* Bytes following PCI: SACK list / RTTP nonce / control payload. */ +#define FRCT_BODY(pci) ((uint8_t *) (pci) + FRCT_PCILEN) +/* Typed access to the stream PCI extension on stream DATA packets. */ +#define FRCT_SPCI(pci) \ + ((struct frct_pci_stream *) ((uint8_t *) (pci) + FRCT_PCILEN)) +/* Push the FRCT header onto spb's head. */ +#define FRCT_HDR_PUSH(spb, frcti) \ + ((struct frct_pci *) ssm_pk_buff_push((spb), \ + frcti_data_hdr_len(frcti))) + +/* Pop a fixed-size header off spb's head; cast to type *. */ +#define FRCT_HDR_POP(spb, type) \ + ((struct type *) ssm_pk_buff_pop((spb), sizeof(struct type))) + +/* Default / max per-flow stream rx ring (pow2); min N * per_pkt. */ +#define FRCT_STREAM_RING_MIN_PKTS 4 +#define FRCT_STREAM_RING_SZ (1U << 20) /* 1 MiB default */ +#define FRCT_STREAM_RING_SZ_MAX (1U << 27) /* 128 MiB */ + +struct frct_pci { + uint16_t flags; + uint16_t hcs; + + uint32_t window; uint32_t seqno; uint32_t ackno; } __attribute__((packed)); +/* Stat counters; fold to no-ops without PROC_FLOW_STATS. */ #ifdef PROC_FLOW_STATS +struct frcti_stat { + size_t rxm_rto; /* RTO-timer driven retransmits */ + size_t rxm_rcv; /* RXM packets received (all) */ + size_t rxm_dup_rcv; /* RXM dups (peer already had it) */ + size_t rxm_sack; /* SACK-mechanism retransmits */ + size_t rxm_rack; /* RACK-driven retransmits */ + size_t rxm_dupthresh; /* DupThresh-driven retransmits */ + size_t rxm_nack; /* NACK-pulled retransmits */ + size_t rxm_due_count; /* rxm_due entries (pre-bail) */ + size_t rxm_due_acked; /* bail: seqno < snd_lwe */ + size_t rxm_due_unowned; /* bail: slot.rxm replaced */ + size_t rxm_due_aged; /* bail: r->t0 + t_r < now */ + size_t rxm_due_defer; /* bail: non-HoL, deferred to HoL */ + size_t rxm_arm_fail; /* rxm_arm: malloc failed */ + size_t rxm_cancel; /* entries cancelled at teardown */ + size_t rxm_tx_dead; /* RXM tx into terminal flow */ + size_t tx_drop; /* frct_tx fail (any cause) */ + size_t tx_drop_ack; /* bare ACK dropped */ + size_t tx_drop_sack; /* SACK dropped */ + size_t tx_drop_ka; /* keepalive dropped */ + size_t tx_drop_rttp; /* RTT probe/echo dropped */ + size_t tx_drop_nack; /* pre-DRF NACK dropped */ + size_t tx_drop_rdv; /* rendez-vous dropped */ + size_t tx_drop_other; /* anything not matched above */ + size_t ack_snd; /* ACK packets sent (bare + SACK) */ + size_t ack_fire; /* delayed-ACK timer fires */ + size_t ack_supp_seqno; /* fire suppressed: seqno */ + size_t ack_supp_inact; /* fire suppressed: inact */ + size_t ack_supp_rate; /* fire suppressed: rate */ + size_t ack_rcv; /* ACK packets received */ + size_t ack_rtt; /* ACKs that fed RTT estimator */ + size_t ack_dup_rcv; /* ACK packet wire dups dropped */ + size_t dup_rcv; /* duplicates received */ + size_t out_rcv; /* pkts out of window */ + size_t rqo_rcv; /* pkts out of rqueue */ + size_t ooo_rcv; /* OOO arrivals */ + size_t sack_snd; /* SACK packets sent */ + size_t sack_rcv; /* SACK packets received */ + size_t dsack_snd; /* SACK pkts carrying a DSACK */ + size_t dsack_rcv; /* DSACK blocks parsed */ + size_t dsack_drop; /* DSACK blocks past MAX_DSACK_LAG */ + size_t nack_snd; /* pre-DRF NACKs sent */ + size_t nack_rcv; /* pre-DRF NACKs received */ + size_t tlp_snd; /* tail loss probes sent */ + size_t inact_drop; /* inactivity drop (NACK on cd) */ + size_t drf_rebase; /* DRF-triggered window rebase */ + size_t rq_released; /* slots cleared by release_rq */ + size_t rttp_snd; /* RTT probes sent */ + size_t rttp_rcv; /* RTT probe replies rcvd */ + size_t rtt_smpl; /* RTT estimator samples */ + size_t rdv_snd; /* rendez-vous packets sent */ + size_t rdv_rcv; /* rendez-vous packets rcvd */ + size_t ka_snd; /* keepalives sent */ + size_t ka_rcv; /* keepalives received */ + size_t sdu_snd_frag; /* writes that fragmented */ + size_t sdu_snd_alloc; /* alloc fail truncated SDU send */ + size_t sdu_snd_tx; /* tx fail truncated SDU send */ + size_t frag_snd; /* fragments sent: FIRST/MID/LAST */ + size_t frag_rcv; /* fragments stashed in rq[] */ + size_t sdu_reasm; /* SDUs delivered reassembled */ + size_t sdu_sole; /* SOLE SDUs delivered (n==1) */ + size_t frag_drop; /* dropped at malformed run */ + size_t strm_snd_byte; /* bytes sent on stream */ + size_t strm_rcv_byte; /* bytes copied to ring */ + size_t strm_dlv_byte; /* bytes delivered to reader */ + size_t strm_drop; /* stream rcvs dropped */ + size_t strm_fin_drop; /* stream FIN packets rejected */ + /* Profiling instrumentation. */ + size_t rcv_proc_ns; /* time inside FRCTI_RCV (ns) */ + size_t tw_move_ns; /* time inside tw_move (ns) */ + size_t drain_calls; /* flow_drain_rx_nb invocations */ +}; + +#define STAT_BUMP(frcti, field) FETCH_ADD_RELAXED(&(frcti)->stat.field, 1) +#define STAT_ADD(frcti, field, v) FETCH_ADD_RELAXED(&(frcti)->stat.field, (v)) +#define STAT_LOAD(frcti, field) LOAD_RELAXED(&(frcti)->stat.field) +#else +#define STAT_BUMP(frcti, field) ((void) (frcti)) +#define STAT_ADD(frcti, field, v) ((void) (frcti)) +#define STAT_LOAD(frcti, field) ((void) (frcti), (size_t) 0) +#endif + +#define frcti_to_flow(f) (&proc.flows[(f)->fd]) +#define RTTP_RING 8 +#define RTTP_COLD_NS (100 * MILLION) /* cold-probe cadence */ +#define RQ_SLOT(seqno) ((seqno) & (RQ_SIZE - 1)) + +struct rxm_entry; + +enum snd_slot_flags { + SND_RTX = 0x01, /* Any retransmit; Karn skips next RTT sample. */ + SND_FAST_RXM = 0x02, /* Fast-retx one-shot gate per loss event. */ + SND_TLP = 0x04, /* Tail loss probe; ACK resets rto_mul. */ +}; + +struct snd_slot { + struct rxm_entry * rxm; /* RXM entry, NULL if none. */ + uint64_t time; /* ts_to_ns of last send (any kind). */ + uint8_t flags; /* SND_* bits above. */ +}; + +/* Per-seqno reorder slot (FRTX) and stream-mode byte/FIN metadata. */ +struct rcv_slot { + ssize_t idx; /* spb idx; -1 = empty */ + uint32_t start; /* stream byte start */ + uint32_t end; /* stream byte end */ + uint8_t fin; /* stream FIN bit */ +}; + +struct frct_cr { + uint32_t lwe; /* Left window edge */ + uint32_t rwe; /* Right window edge */ + + uint8_t cflags; + uint32_t seqno; /* SEQ to send, or last SEQ Ack'd */ + uint32_t ackno; /* snd: ACK-pkt seqno; rcv: dedup */ + + uint64_t act; /* ts_to_ns of last activity */ + uint64_t inact; /* Inactivity threshold (ns) */ +}; + +struct frcti { + /* IMM: set once in frcti_create; read-only thereafter. */ + int fd; + uint64_t t_mpl; /* MPL (ns) */ + uint64_t t_a; /* a-timer (ns) */ + uint64_t t_r; /* r-timer (ns) */ + uint64_t t_rdv; /* RDV cooldown (ns) */ + time_t ber; /* cached qs.ber */ + bool lossy; /* qs.loss != 0 */ + time_t qs_timeout; /* cached qs.timeout (ms) */ + size_t frag_mtu; /* max FRCT pkt: PCI + payload */ + uint16_t sack_n_max; /* SACK blocks that fit MTU */ + bool stream; + + /* All fields below are protected by lock (rwlock/LOAD_ACQUIRE). */ + struct { + struct frct_cr snd_cr; + struct frct_cr rcv_cr; + + /* RTT/RACK estimator */ + time_t srtt; /* smoothed RTT */ + time_t mdev; /* mean deviation */ + time_t min_rtt; /* RACK base, ns */ + uint64_t t_min_rtt; /* min_rtt last set */ + time_t rto; /* retransmit TO */ + time_t rto_min; /* RTO floor (ns) */ + uint8_t rto_mul; /* RTO backoff bits */ + uint32_t rtt_lwe; /* RTT-sample fence */ + uint64_t t_rcv_rtt; /* last RTT feed */ + uint64_t t_snd_probe; /* last probe sent */ + uint64_t t_latest_ack; /* RACK.fack snd-ts */ + uint32_t probe_id_next; + struct { + uint32_t id; + uint64_t ts; /* ts_to_ns send */ + uint8_t nonce[RTTP_NONCE_LEN]; /* echoed back */ + } probes[RTTP_RING]; + + /* rcv reassembly */ + size_t max_rcv_sdu; /* max reasm bytes */ + uint8_t * rcv_ring; /* lazy alloc */ + size_t rcv_ring_sz; /* power of 2 */ + uint32_t ring_seq_cap; /* ring/per_pkt */ + + uint32_t snd_byte_next; + bool snd_fin_sent; + uint32_t snd_fin_seqno; + uint32_t rcv_byte_next; + uint32_t rcv_byte_high; /* contiguous high */ + uint32_t rcv_byte_fin; /* set when FIN */ + bool rcv_fin_seen; + + struct rcv_slot rcv_slots[RQ_SIZE]; + struct snd_slot snd_slots[RQ_SIZE]; /* .rxm is ATOM */ + + /* rcv SACK dedup */ + uint64_t t_snd_sack; + uint32_t sack_lwe; /* rcv lwe at SACK */ + uint16_t sack_n; /* SACK block count */ + + /* RFC 2883 D-SACK: pending report (single-slot, latest). */ + uint32_t dsack_seqno; + bool dsack_valid; + + /* RFC 8985 §7.2 RACK reorder-window scaling. */ + uint8_t reo_wnd_mult; /* REO_WND_MULT_MAX */ + uint32_t dsack_lwe_snap; /* lwe @ last DSACK */ + uint64_t t_last_reo_widen; /* once-per-RTT */ + + uint32_t dup_thresh; /* RFC 8985 */ + uint32_t tlp_high_seq; /* §7.3: 0 = none */ + uint8_t tlp_count; /* §7.3 per-episode */ + uint64_t t_nack; + bool open; /* FC window state */ + bool in_recovery; + uint32_t recovery_high; /* seqno @ entry */ + uint32_t rack_fired_lwe; /* lwe @ last RACK */ + struct timespec t_wnd; /* window-closed ts */ + struct timespec t_last_rdv; /* last RDV sent */ + struct list_head rxm_list; /* live rxm entries */ + + pthread_rwlock_t lock; + }; + + /* Read/written via __atomic without holding lock. */ + uint64_t t_ka_rcv; /* ts_to_ns of last KA rx */ + uint8_t ack_pending; /* delayed-ACK dedup */ + uint8_t tlp_pending; /* TLP arm dedup (lazy) */ + + /* Timer entries; ownership belongs to the tw module. */ + struct tw_entry ack_tw; /* delayed-ACK timer */ + struct tw_entry ka_tw; /* keepalive timer */ + struct tw_entry tlp_tw; /* tail-loss probe timer */ + +#ifdef PROC_FLOW_STATS + /* STAT: lock-free relaxed atomic counters. */ + struct frcti_stat stat; +#endif +}; + +#ifdef PROC_FLOW_STATS + +__attribute__((cold)) static int frct_rib_read(const char * path, char * buf, size_t len) { + struct frcti * frcti; struct timespec now; + uint64_t now_ns; char * entry; - struct flow * flow; - struct frcti * frcti; int fd; - - (void) len; + int written; + /* Snapshot under the locks; format outside (pure userspace). */ + struct { + uint64_t t_mpl; + uint64_t t_a; + uint64_t t_r; + time_t srtt; + time_t mdev; + time_t rto; + time_t min_rtt; + struct frct_cr snd_cr; + struct frct_cr rcv_cr; + size_t rx_q_now; + size_t tx_q_now; + struct frcti_stat stat; + } s; entry = strstr(path, RIB_SEPARATOR); assert(entry); @@ -118,69 +437,184 @@ static int frct_rib_read(const char * path, fd = atoi(path); - flow = &ai.flows[fd]; - clock_gettime(PTHREAD_COND_CLOCK, &now); + now_ns = TS_TO_UINT64(now); + + if (fd < 0 || fd >= PROC_MAX_FLOWS) + return 0; - pthread_rwlock_rdlock(&ai.lock); + pthread_rwlock_rdlock(&proc.lock); - frcti = flow->frcti; + frcti = proc.flows[fd].frcti; + if (frcti == NULL) { + pthread_rwlock_unlock(&proc.lock); + return 0; + } + + s.t_mpl = frcti->t_mpl; + s.t_a = frcti->t_a; + s.t_r = frcti->t_r; + + s.rx_q_now = proc.flows[fd].rx_rb != NULL + ? ssm_rbuff_queued(proc.flows[fd].rx_rb) : 0; + s.tx_q_now = proc.flows[fd].tx_rb != NULL + ? ssm_rbuff_queued(proc.flows[fd].tx_rb) : 0; pthread_rwlock_rdlock(&frcti->lock); - sprintf(buf, - "Maximum packet lifetime (ns): %20ld\n" - "Max time to Ack (ns): %20ld\n" - "Max time to Retransmit (ns): %20ld\n" + s.srtt = frcti->srtt; + s.mdev = frcti->mdev; + s.rto = frcti->rto; + s.min_rtt = frcti->min_rtt; + s.snd_cr = frcti->snd_cr; + s.rcv_cr = frcti->rcv_cr; + s.stat = frcti->stat; + + pthread_rwlock_unlock(&frcti->lock); + pthread_rwlock_unlock(&proc.lock); + + written = snprintf(buf, len, + "Maximum packet lifetime (ns): %20" PRIu64 "\n" + "Max time to Ack (ns): %20" PRIu64 "\n" + "Max time to Retransmit (ns): %20" PRIu64 "\n" "Smoothed rtt (ns): %20ld\n" "RTT standard deviation (ns): %20ld\n" "Retransmit timeout RTO (ns): %20ld\n" + "Minimum RTT (RACK base, ns): %20ld\n" "Sender left window edge: %20u\n" "Sender right window edge: %20u\n" - "Sender inactive (ns): %20ld\n" + "Sender inactive (ns): %20lld\n" "Sender current sequence number: %20u\n" "Receiver left window edge: %20u\n" "Receiver right window edge: %20u\n" - "Receiver inactive (ns): %20ld\n" + "Receiver inactive (ns): %20lld\n" "Receiver last ack: %20u\n" - "Number of pkt retransmissions: %20zu\n" - "Number of rtt probes: %20zu\n" - "Number of rtt estimates: %20zu\n" - "Number of duplicates received: %20zu\n" - "Number of delayed acks received: %20zu\n" - "Number of rendez-vous sent: %20zu\n" - "Number of packets out of window: %20zu\n" - "Number of packets out of rqueue: %20zu\n", - frcti->mpl, - frcti->a, - frcti->r, - frcti->srtt, - frcti->mdev, - frcti->rto, - frcti->snd_cr.lwe, - frcti->snd_cr.rwe, - ts_diff_ns(&frcti->snd_cr.act, &now), - frcti->snd_cr.seqno, - frcti->rcv_cr.lwe, - frcti->rcv_cr.rwe, - ts_diff_ns(&frcti->rcv_cr.act, &now), - frcti->rcv_cr.seqno, - frcti->n_rtx, - frcti->n_prb, - frcti->n_rtt, - frcti->n_dup, - frcti->n_dak, - frcti->n_rdv, - frcti->n_out, - frcti->n_rqo); - - pthread_rwlock_unlock(&flow->frcti->lock); - - pthread_rwlock_unlock(&ai.lock); - - return strlen(buf); + "RXM (RTO-driven) sent: %20zu\n" + "RXM packets received: %20zu\n" + " duplicates received: %20zu\n" + "RXM (SACK mechanism) sent: %20zu\n" + "RXM (RACK-driven) sent: %20zu\n" + "RXM (DupThresh-driven) sent: %20zu\n" + "RXM (NACK-driven) sent: %20zu\n" + "ACK packets sent: %20zu\n" + "Delayed-ACK timer fires: %20zu\n" + " suppressed (seqno): %20zu\n" + " suppressed (inact): %20zu\n" + " suppressed (rate): %20zu\n" + "ACK packets received: %20zu\n" + " fed RTT estimator: %20zu\n" + " wire dups dropped: %20zu\n" + "Duplicates received: %20zu\n" + "Out-of-window pkts received: %20zu\n" + "Out-of-rqueue pkts received: %20zu\n" + "OOO arrivals: %20zu\n" + "SACKs sent: %20zu\n" + "SACKs received: %20zu\n" + "D-SACKs sent: %20zu\n" + "D-SACKs received: %20zu\n" + "D-SACK out-of-range dropped: %20zu\n" + "Pre-DRF NACKs sent: %20zu\n" + "Pre-DRF NACKs received: %20zu\n" + "Tail loss probes sent: %20zu\n" + "Inactivity drops (silent): %20zu\n" + "DRF window rebases: %20zu\n" + "rq slots cleared by release_rq: %20zu\n" + "RTT probes sent: %20zu\n" + "RTT probe replies received: %20zu\n" + "RTT estimator samples: %20zu\n" + "Rendez-vous packets sent: %20zu\n" + "Rendez-vous packets received: %20zu\n" + "Keepalives sent: %20zu\n" + "Keepalives received: %20zu\n" + "SDU writes fragmented: %20zu\n" + " alloc fail mid-SDU: %20zu\n" + " tx fail mid-SDU: %20zu\n" + "Fragments sent: %20zu\n" + "Fragments received: %20zu\n" + "SDUs delivered reassembled: %20zu\n" + "SDUs delivered (SOLE): %20zu\n" + "Fragments dropped (malformed): %20zu\n" + "Stream bytes sent: %20zu\n" + "Stream bytes received: %20zu\n" + "Stream bytes delivered: %20zu\n" + "Stream packets dropped: %20zu\n" + "Stream FINs dropped: %20zu\n" + "FRCTI_RCV time (ns): %20zu\n" + "tw_move time (ns): %20zu\n" + "drain_rx_nb calls: %20zu\n" + "RX rbuff queued: %20zu\n" + "TX rbuff queued: %20zu\n" + "RXM-due entries: %20zu\n" + " bail (acked): %20zu\n" + " bail (unowned): %20zu\n" + " bail (aged): %20zu\n" + " bail (defer): %20zu\n" + "RXM-arm malloc failures: %20zu\n" + "RXM cancels (teardown): %20zu\n" + "RXM tx into dead flow: %20zu\n" + "Tx ring drops (any cause): %20zu\n" + " ack: %20zu\n" + " sack: %20zu\n" + " ka: %20zu\n" + " rttp: %20zu\n" + " nack: %20zu\n" + " rdv: %20zu\n" + " other: %20zu\n", + /* Check getattr size below when adding stats. */ + s.t_mpl, s.t_a, s.t_r, + s.srtt, s.mdev, s.rto, s.min_rtt, + s.snd_cr.lwe, s.snd_cr.rwe, + (long long)(now_ns - s.snd_cr.act), + s.snd_cr.seqno, + s.rcv_cr.lwe, s.rcv_cr.rwe, + (long long)(now_ns - s.rcv_cr.act), + s.rcv_cr.seqno, + s.stat.rxm_rto, s.stat.rxm_rcv, s.stat.rxm_dup_rcv, + s.stat.rxm_sack, s.stat.rxm_rack, s.stat.rxm_dupthresh, + s.stat.rxm_nack, + s.stat.ack_snd, s.stat.ack_fire, + s.stat.ack_supp_seqno, s.stat.ack_supp_inact, + s.stat.ack_supp_rate, + s.stat.ack_rcv, s.stat.ack_rtt, s.stat.ack_dup_rcv, + s.stat.dup_rcv, s.stat.out_rcv, s.stat.rqo_rcv, + s.stat.ooo_rcv, + s.stat.sack_snd, s.stat.sack_rcv, + s.stat.dsack_snd, s.stat.dsack_rcv, s.stat.dsack_drop, + s.stat.nack_snd, s.stat.nack_rcv, s.stat.tlp_snd, + s.stat.inact_drop, s.stat.drf_rebase, s.stat.rq_released, + s.stat.rttp_snd, s.stat.rttp_rcv, s.stat.rtt_smpl, + s.stat.rdv_snd, s.stat.rdv_rcv, + s.stat.ka_snd, s.stat.ka_rcv, + s.stat.sdu_snd_frag, s.stat.sdu_snd_alloc, s.stat.sdu_snd_tx, + s.stat.frag_snd, s.stat.frag_rcv, + s.stat.sdu_reasm, s.stat.sdu_sole, s.stat.frag_drop, + s.stat.strm_snd_byte, s.stat.strm_rcv_byte, + s.stat.strm_dlv_byte, + s.stat.strm_drop, s.stat.strm_fin_drop, + s.stat.rcv_proc_ns, s.stat.tw_move_ns, + s.stat.drain_calls, + s.rx_q_now, s.tx_q_now, + s.stat.rxm_due_count, + s.stat.rxm_due_acked, s.stat.rxm_due_unowned, + s.stat.rxm_due_aged, s.stat.rxm_due_defer, + s.stat.rxm_arm_fail, + s.stat.rxm_cancel, + s.stat.rxm_tx_dead, s.stat.tx_drop, + s.stat.tx_drop_ack, s.stat.tx_drop_sack, + s.stat.tx_drop_ka, s.stat.tx_drop_rttp, + s.stat.tx_drop_nack, s.stat.tx_drop_rdv, + s.stat.tx_drop_other); + + if (written < 0) + return 0; + + if ((size_t) written >= len) + return (int) (len - 1); + + return written; } +__attribute__((cold)) static int frct_rib_readdir(char *** buf) { *buf = malloc(sizeof(**buf)); @@ -199,13 +633,14 @@ static int frct_rib_readdir(char *** buf) return -ENOMEM; } +__attribute__((cold)) static int frct_rib_getattr(const char * path, struct rib_attr * attr) { (void) path; - (void) attr; - attr->size = 1189; + /* Must be >= the sprintf output in frct_rib_read. */ + attr->size = 8192; attr->mtime = 0; return 0; @@ -220,128 +655,1168 @@ static struct rib_ops r_ops = { #endif /* PROC_FLOW_STATS */ -static bool before(uint32_t seq1, - uint32_t seq2) +static __inline__ bool before(uint32_t s1, uint32_t s2) { - return (int32_t)(seq1 - seq2) < 0; + return (int32_t)(s1 - s2) < 0; } -static bool after(uint32_t seq1, - uint32_t seq2) +static __inline__ bool after(uint32_t s1, uint32_t s2) { - return (int32_t)(seq2 - seq1) < 0; + return (int32_t)(s2 - s1) < 0; } -static void __send_frct_pkt(int fd, - uint8_t flags, - uint32_t ackno, - uint32_t rwe) +static __inline__ bool within(uint32_t seq, uint32_t lo, uint32_t hi) { - struct shm_du_buff * sdb; - struct frct_pci * pci; - ssize_t idx; - struct flow * f; + return after(seq, lo) && !after(seq, hi); +} - /* Raw calls needed to bypass frcti. */ -#ifdef RXM_BLOCKING - idx = shm_rdrbuff_alloc_b(ai.rdrb, sizeof(*pci), NULL, &sdb, NULL); -#else - idx = shm_rdrbuff_alloc(ai.rdrb, sizeof(*pci), NULL, &sdb); -#endif - if (idx < 0) +static __inline__ bool in_window(uint32_t seq, const struct frct_cr * cr) +{ + return !before(seq, cr->lwe) && before(seq, cr->rwe); +} + +/* DRF arrival that stays within the current receive epoch. */ +static __inline__ bool same_epoch_drf(uint32_t seq, + uint16_t flags, + const struct frct_cr * cr) +{ + if (cr->lwe == cr->rwe) + return false; + + return (flags & FRCT_RXM) || in_window(seq, cr); +} + +/* + * RACK reorder window R (RFC 8985 §6.2): + * R = MIN(reo_wnd_mult * RACK.min_RTT / 4, SRTT) + * reo_wnd_mult scales on D-SACK evidence of under-tolerance (§7.2). + * Fall back to srtt when no min_rtt sample exists yet; MIN_REORDER_NS + * floor guards collapse below the timer-tick resolution. + */ +static __inline__ uint64_t rack_reorder_window(struct frcti * frcti) +{ + uint64_t mult = frcti->reo_wnd_mult > 0 ? frcti->reo_wnd_mult : 1; + uint64_t base = frcti->min_rtt > 0 ? (uint64_t) frcti->min_rtt + : (uint64_t) frcti->srtt; + uint64_t R = mult * (base / 4); + + R = MAX(R, (uint64_t) MIN_REORDER_NS); + R = MIN(R, (uint64_t) frcti->srtt); + + return R; +} + +static __inline__ int frct_spb_reserve(size_t len, + struct ssm_pk_buff ** spb) +{ + ssize_t idx = ssm_pool_alloc_b(proc.pool, len, NULL, spb, NULL); + + return idx < 0 ? (int) idx : 0; +} + +static __inline__ void frct_spb_release(struct ssm_pk_buff * spb) +{ + ssm_pool_remove(proc.pool, ssm_pk_buff_get_off(spb)); +} + +static __inline__ void frct_spb_release_idx(size_t idx) +{ + ssm_pool_remove(proc.pool, idx); +} + +/* Fetch the spb stashed at the rq slot for seqno. */ +static __inline__ struct ssm_pk_buff * rq_frag(const struct frcti * frcti, + uint32_t seqno) +{ + return ssm_pool_get(proc.pool, frcti->rcv_slots[RQ_SLOT(seqno)].idx); +} + +static __inline__ size_t frcti_data_hdr_len(const struct frcti * frcti) +{ + return FRCT_PCILEN + (frcti->stream ? FRCT_PCI_STREAM_LEN : 0); +} + +static __inline__ size_t frcti_ctrl_hdr_len(const struct frcti * frcti) +{ + (void) frcti; + + return FRCT_PCILEN; +} + +/* + * HCS at offset 2 inside PCI. Covers flags (bytes 0..1) and + * window/seqno/ackno (bytes 4..15), plus SPCI for stream DATA. + */ +static void frct_hcs_set(struct frct_pci * pci, + bool stream) +{ + uint16_t hcs = 0; + size_t tail; + + tail = sizeof(*pci) - sizeof(pci->flags) - sizeof(pci->hcs); + if (stream) + tail += FRCT_PCI_STREAM_LEN; + + crc16_ccitt_false(&hcs, pci, sizeof(pci->flags)); + crc16_ccitt_false(&hcs, &pci->window, tail); + + pci->hcs = hton16(hcs); +} + +static int frct_hcs_check(const struct frct_pci * pci, + const struct frcti * frcti) +{ + uint16_t hcs = 0; + uint16_t flags; + size_t tail; + + /* Untrusted flag read; mismatch on HCS will drop on corrupt. */ + flags = ntoh16(pci->flags); + + tail = sizeof(*pci) - sizeof(pci->flags) - sizeof(pci->hcs); + if (frcti->stream && (flags & FRCT_DATA)) + tail += FRCT_PCI_STREAM_LEN; + + crc16_ccitt_false(&hcs, pci, sizeof(pci->flags)); + crc16_ccitt_false(&hcs, &pci->window, tail); + + return hcs != ntoh16(pci->hcs); +} + +/* Bump tx_drop plus the per-frame-type counter matching `flags`. */ +static void frct_tx_drop_bump(struct frcti * frcti, + uint16_t flags) +{ + STAT_BUMP(frcti, tx_drop); + + if (flags & FRCT_SACK) { + STAT_BUMP(frcti, tx_drop_sack); return; + } - pci = (struct frct_pci *) shm_du_buff_head(sdb); - memset(pci, 0, sizeof(*pci)); + if (flags & FRCT_KA) { + STAT_BUMP(frcti, tx_drop_ka); + return; + } - *((uint32_t *) pci) = hton32(rwe); + if (flags & FRCT_RTTP) { + STAT_BUMP(frcti, tx_drop_rttp); + return; + } - pci->flags = flags; - pci->ackno = hton32(ackno); + if (flags & FRCT_NACK) { + STAT_BUMP(frcti, tx_drop_nack); + return; + } + + if (flags & FRCT_RDVS) { + STAT_BUMP(frcti, tx_drop_rdv); + return; + } - f = &ai.flows[fd]; + if (flags & FRCT_ACK) { + STAT_BUMP(frcti, tx_drop_ack); + return; + } + + STAT_BUMP(frcti, tx_drop_other); +} + +static int frct_tx(struct frcti * frcti, struct ssm_pk_buff * spb) +{ + struct flow * f = frcti_to_flow(frcti); + const struct frct_pci * pci; + const struct timespec * dl = NULL; + struct timespec now; + struct timespec intv = TIMESPEC_INIT_NS(FRCT_TX_TIMEO_NS); + struct timespec deadline; + uint16_t flags; + ssize_t idx; + int ret = -ENOMEM; + + pci = (const struct frct_pci *) ssm_pk_buff_head(spb); + flags = ntoh16(pci->flags); + + /* CRC32 covers plaintext body; PCI is in HCS. Pre-encrypt. */ + if (flags & FRCT_SACK) { + if (crc_add(spb, frcti_ctrl_hdr_len(frcti)) != 0) + goto fail; + } else if ((flags & FRCT_DATA) && f->info.qs.ber == 0) { + if (crc_add(spb, frcti_data_hdr_len(frcti)) != 0) + goto fail; + } - if (crypt_encrypt(&f->crypt, sdb) < 0) + if (spb_encrypt(f, spb) < 0) goto fail; -#ifdef RXM_BLOCKING - if (shm_rbuff_write_b(f->tx_rb, idx, NULL)) -#else - if (shm_rbuff_write(f->tx_rb, idx)) -#endif + idx = ssm_pk_buff_get_off(spb); + + /* DATA blocks; control times out so a full ring can't stall wheel. */ + if (!(flags & FRCT_DATA)) { + clock_gettime(PTHREAD_COND_CLOCK, &now); + ts_add(&now, &intv, &deadline); + dl = &deadline; + } + + ret = ssm_rbuff_write_b(f->tx_rb, idx, dl); + if (ret < 0) goto fail; - shm_flow_set_notify(f->set, f->info.id, FLOW_PKT); + ssm_flow_set_notify(f->set, f->info.id, FLOW_PKT); - return; + return 0; fail: - ipcp_sdb_release(sdb); - return; + frct_tx_drop_bump(frcti, flags); + ssm_pool_remove(proc.pool, ssm_pk_buff_get_off(spb)); + return ret; +} + +__attribute__((cold)) +static void frct_mark_flow_down(struct frcti * frcti) +{ + struct flow * f = frcti_to_flow(frcti); + + if (f->rx_rb != NULL) + ssm_rbuff_set_bits(f->rx_rb, RB_FLOWDOWN); + + if (f->tx_rb != NULL) + ssm_rbuff_set_bits(f->tx_rb, RB_FLOWDOWN); +} + +__attribute__((cold)) +static void frct_mark_peer_dead(struct frcti * frcti) +{ + struct flow * f = frcti_to_flow(frcti); + + if (f->rx_rb != NULL) + ssm_rbuff_set_bits(f->rx_rb, RB_FLOWPEER); + + if (proc.fqset != NULL) + ssm_flow_set_notify(proc.fqset, f->info.id, FLOW_PEER); +} + +static __inline__ int frct_ctrl_alloc(struct ssm_pk_buff ** spb, + struct frct_pci ** pci, + size_t payload_len) +{ + if (frct_spb_reserve(FRCT_PCILEN + payload_len, spb) < 0) + return -1; + + *pci = (struct frct_pci *) ssm_pk_buff_head(*spb); + memset(*pci, 0, FRCT_PCILEN); + + return 0; +} + +/* + * Advertised rwe. Stream mode clamps to lwe + ring_seq_cap so the + * byte-equivalent fits the rx ring. Caller holds at least the rdlock. + */ +static __inline__ uint32_t frcti_advert_rwe(struct frcti * frcti) +{ + uint32_t rwe; + uint32_t cap; + + rwe = frcti->rcv_cr.rwe; + + if (!frcti->stream) + return rwe; + + cap = frcti->rcv_cr.lwe + frcti->ring_seq_cap; + + return before(cap, rwe) ? cap : rwe; +} + +static void frcti_pkt_snd(struct frcti * frcti, + uint16_t flags, + uint32_t ackno, + uint32_t rwe) +{ + struct ssm_pk_buff * spb; + struct frct_pci * pci; + + if (frct_ctrl_alloc(&spb, &pci, 0) < 0) + return; + + pci->flags = hton16(flags); + pci->window = hton32(rwe); + pci->ackno = hton32(ackno); + if (flags & FRCT_ACK) { + /* reuse ackno for the sequence number of delayed ACK */ + ackno = FETCH_ADD_RELAXED(&frcti->snd_cr.ackno, 1); + pci->seqno = hton32(ackno + 1); + } + + frct_hcs_set(pci, false); + + frct_tx(frcti, spb); +} + +/* RTO floor scales with srtt; hard floor rto_min guards sub-ms RTT. */ +static void rtt_init(struct frcti * frcti, + time_t rtt_hint) +{ + time_t floor; + + if (rtt_hint > 0) { + rtt_hint = MAX(rtt_hint, (time_t) RTT_BOOT_NS); + frcti->srtt = rtt_hint; + frcti->mdev = rtt_hint >> 3; + floor = MAX(frcti->rto_min, 2 * frcti->srtt); + frcti->rto = MAX(floor, rtt_hint + (frcti->mdev << MDEV_MUL)); + frcti->min_rtt = rtt_hint; + } else { + /* Boot from first ACK. */ + frcti->srtt = 0; + frcti->mdev = RTT_BOOT_NS; + frcti->rto = MAX((time_t) INITIAL_RTO, frcti->rto_min); + frcti->min_rtt = 0; + } + + frcti->rto_mul = 0; } -static void send_frct_pkt(struct frcti * frcti) +/* RFC 8985 §6.2: replace min_RTT on unset, smaller sample, or expiry. */ +static __inline__ bool min_rtt_stale(struct frcti * frcti, + time_t mrtt, + uint64_t now_ns) { + if (frcti->min_rtt == 0) + return true; + + if (mrtt < frcti->min_rtt) + return true; + + return ts_aged_ns(now_ns, frcti->t_min_rtt, MIN_RTT_WIN_NS); +} + +/* Linux-style windowed-min refresh of RACK.min_RTT. */ +static __inline__ void min_rtt_update(struct frcti * frcti, + time_t mrtt, + uint64_t now_ns) +{ + if (!min_rtt_stale(frcti, mrtt, now_ns)) + return; + + frcti->min_rtt = mrtt; + frcti->t_min_rtt = now_ns; +} + +static void rtt_update(struct frcti * frcti, + time_t mrtt, + uint64_t now_ns) +{ + time_t srtt = frcti->srtt; + time_t rttvar = frcti->mdev; + time_t floor; + time_t rto; + + if (srtt == 0) { + srtt = mrtt; + rttvar = mrtt >> 1; + } else { + /* RFC 6298 symmetric EWMA. */ + time_t delta = mrtt - srtt; + srtt += (delta >> 3); + delta = (ABS(delta) - rttvar) >> 2; +#ifdef FRCT_LINUX_RTT_ESTIMATOR + if (delta < 0) + delta >>= 3; +#endif + rttvar += delta; + } + STAT_BUMP(frcti, rtt_smpl); + frcti->srtt = MAX(SRTT_FLOOR_NS, srtt); + frcti->mdev = MAX(MDEV_FLOOR_NS, rttvar); + + min_rtt_update(frcti, mrtt, now_ns); + + floor = MAX(frcti->rto_min, 2 * frcti->srtt); + rto = MAX(floor, frcti->srtt + (frcti->mdev << MDEV_MUL)); + + STORE_RELEASE(&frcti->rto, rto); + STORE_RELEASE(&frcti->rto_mul, 0); +} + +/* Fill probes[pos], return new probe_id; 0 on entropy failure. Wrlock. */ +static uint32_t rttp_alloc_probe(struct frcti * frcti, + uint64_t now_ns, + uint8_t nonce[RTTP_NONCE_LEN]) +{ + uint32_t probe_id; + size_t pos; + + if (random_buffer(nonce, RTTP_NONCE_LEN) < 0) + return 0; + + probe_id = frcti->probe_id_next++; + if (probe_id == 0) + probe_id = frcti->probe_id_next++; + + pos = RTTP_POS(probe_id); + frcti->probes[pos].id = probe_id; + frcti->probes[pos].ts = now_ns; + memcpy(frcti->probes[pos].nonce, nonce, RTTP_NONCE_LEN); + frcti->t_snd_probe = now_ns; + + STAT_BUMP(frcti, rttp_snd); + + return probe_id; +} + +/* Caller wrlock; out args valid on true (caller emits post-unlock). */ +static bool rtt_probe_arm(struct frcti * frcti, + uint64_t now_ns, + uint32_t * probe_id, + uint8_t nonce[RTTP_NONCE_LEN]) +{ + if (frcti->srtt == 0) + return false; + + if (!after(frcti->snd_cr.seqno, frcti->snd_cr.lwe)) + return false; + + if (!ts_aged_ns(now_ns, frcti->t_rcv_rtt, + 2u * (uint64_t) frcti->srtt)) + return false; + + if (!ts_aged_ns(now_ns, frcti->t_snd_probe, + (uint64_t) frcti->srtt)) + return false; + + *probe_id = rttp_alloc_probe(frcti, now_ns, nonce); + + return *probe_id != 0; +} + +static void frcti_rttp_snd(struct frcti * frcti, + uint32_t probe_id, + uint32_t echo_id, + const uint8_t * nonce) +{ + struct ssm_pk_buff * spb; + struct frct_pci * pci; + struct frct_rttp * rttp; + + if (frct_ctrl_alloc(&spb, &pci, RTTP_PAYLOAD) < 0) + return; + + pci->flags = hton16(FRCT_RTTP); + + frct_hcs_set(pci, false); + + rttp = (struct frct_rttp *) FRCT_BODY(pci); + rttp->probe_id = hton32(probe_id); + rttp->echo_id = hton32(echo_id); + memcpy(rttp->nonce, nonce, sizeof(rttp->nonce)); + + frct_tx(frcti, spb); +} + +struct rxm_entry { + struct tw_entry tw; + struct list_head next; /* in frcti->rxm_list */ + struct frcti * frcti; + uint32_t seqno; + uint64_t t0; + size_t len; + uint8_t pkt[]; /* flexible — sized at alloc time */ +}; + +static void rxm_entry_destroy(struct rxm_entry * r) +{ + free(r); +} + +static bool rxm_still_owned(struct frcti * frcti, + size_t pos, + struct rxm_entry * r) +{ + return LOAD_ACQUIRE(&frcti->snd_slots[pos].rxm) == r; +} + +/* + * All in-flight slots share the HoL backoff; otherwise non-HoL timers + * cycle at base RTO and storm the wire while HoL is still backing off. + */ +static uint64_t rxm_next_deadline(struct frcti * frcti, + uint64_t now_ns) +{ + time_t rto = LOAD_RELAXED(&frcti->rto); + uint8_t rto_mul = LOAD_RELAXED(&frcti->rto_mul); + + return now_ns + ((uint64_t) rto << rto_mul); +} + +/* Copy pkt, set FRCT_RXM, refresh ackno, re-seal HCS. */ +static struct ssm_pk_buff * rxm_pkt_prepare(const void * pkt, + size_t len, + uint32_t rcv_lwe, + bool stream) +{ + struct ssm_pk_buff * spb; + struct frct_pci * pci; + uint16_t flags; + + if (frct_spb_reserve(len, &spb) < 0) + return NULL; + + pci = (struct frct_pci *) ssm_pk_buff_head(spb); + memcpy(pci, pkt, len); + + flags = ntoh16(pci->flags) | FRCT_RXM; + pci->flags = hton16(flags); + pci->ackno = hton32(rcv_lwe); + + frct_hcs_set(pci, stream); + + return spb; +} + +/* Caller must NOT hold frcti->lock. */ +static void rxm_snd(struct frcti * frcti, + uint32_t seqno, + const void * pkt, + size_t len) +{ + struct ssm_pk_buff * spb; struct timespec now; + struct snd_slot * slot; + uint32_t snd_lwe; + uint32_t rcv_lwe; + size_t pos; + int ret; + + snd_lwe = LOAD_RELAXED(&frcti->snd_cr.lwe); + rcv_lwe = LOAD_RELAXED(&frcti->rcv_cr.lwe); + + clock_gettime(PTHREAD_COND_CLOCK, &now); + + pthread_rwlock_wrlock(&frcti->lock); + + pos = RQ_SLOT(seqno); + slot = &frcti->snd_slots[pos]; + + slot->time = TS_TO_UINT64(now); + /* RTO supersedes any pending TLP/fast-rxm on this slot. */ + slot->flags = (slot->flags & ~(SND_FAST_RXM | SND_TLP)) | SND_RTX; + /* §7.3: RTO supersedes TLP probes and ends the probe episode. */ + frcti->tlp_high_seq = 0; + frcti->tlp_count = 0; + + frcti->rtt_lwe = seqno + 1; + + /* Only the HoL retransmit bumps the global RTO backoff. */ + if (seqno == snd_lwe && frcti->rto_mul < MAX_RTO_MUL) + STORE_RELEASE(&frcti->rto_mul, frcti->rto_mul + 1); + + /* RFC 8985 §7.2 step 4: RTO on HoL resets RACK reo scaling. */ + if (seqno == snd_lwe) + frcti->reo_wnd_mult = 1; + + pthread_rwlock_unlock(&frcti->lock); + + STAT_BUMP(frcti, rxm_rto); + + spb = rxm_pkt_prepare(pkt, len, rcv_lwe, frcti->stream); + if (spb == NULL) + return; + + /* ETIMEDOUT/ENOMEM: let r-timer drive teardown. */ + ret = frct_tx(frcti, spb); + if (ret == -EFLOWDOWN || ret == -ENOTALLOC) + STAT_BUMP(frcti, rxm_tx_dead); +} + +static void rxm_due(void * arg) +{ + struct rxm_entry * r = arg; + struct frcti * frcti = r->frcti; + struct timespec now; + uint64_t now_ns; + uint32_t snd_lwe; + size_t pos = RQ_SLOT(r->seqno); + + STAT_BUMP(frcti, rxm_due_count); + + snd_lwe = LOAD_RELAXED(&frcti->snd_cr.lwe); + + /* Already ACK'd: expected for the steady-state majority. */ + if (before(r->seqno, snd_lwe)) { + STAT_BUMP(frcti, rxm_due_acked); + goto cleanup; + } + + /* SACK/RACK-cleared the slot (caller NULL'd snd_slots[pos].rxm). */ + if (!rxm_still_owned(frcti, pos, r)) { + STAT_BUMP(frcti, rxm_due_unowned); + goto cleanup; + } + + clock_gettime(PTHREAD_COND_CLOCK, &now); + now_ns = TS_TO_UINT64(now); + + /* R-timer expired: peer unreachable. */ + if (RXM_AGED_OUT(r->t0, now_ns, frcti->t_r)) { + STAT_BUMP(frcti, rxm_due_aged); + frct_mark_flow_down(frcti); + goto cleanup; + } + + /* HoL-only retx; defer at base rto so HoL transitions react. */ + if (r->seqno != snd_lwe) { + STAT_BUMP(frcti, rxm_due_defer); + tw_post(&r->tw, now_ns + LOAD_RELAXED(&frcti->rto), + rxm_due, r); + return; + } + + rxm_snd(frcti, r->seqno, r->pkt, r->len); + + /* Re-check ownership: fire path may have replaced our entry. */ + if (rxm_still_owned(frcti, pos, r)) { + uint64_t anchor; + + /* Per-slot anchor breaks co-fire re-bin. */ + anchor = frcti->snd_slots[pos].time; + tw_post(&r->tw, rxm_next_deadline(frcti, anchor), rxm_due, r); + return; + } + + cleanup: + pthread_rwlock_wrlock(&frcti->lock); + + if (rxm_still_owned(frcti, pos, r)) + STORE_RELEASE(&frcti->snd_slots[pos].rxm, NULL); + + list_del(&r->next); + + pthread_rwlock_unlock(&frcti->lock); + + rxm_entry_destroy(r); +} + +/* Pre-allocate rxm entry so frcti_snd can fail before committing seqno. */ +static struct rxm_entry * rxm_alloc(struct frcti * frcti, + size_t pkt_len) +{ + struct rxm_entry * r; + + r = malloc(sizeof(*r) + pkt_len); + if (r == NULL) { + STAT_BUMP(frcti, rxm_arm_fail); + return NULL; + } + + r->frcti = frcti; + tw_init_entry(&r->tw); + + return r; +} + +static void rxm_arm(struct frcti * frcti, + uint32_t seqno, + struct rxm_entry * r, + const struct ssm_pk_buff * spb) +{ + struct timespec now; + time_t rto; + uint8_t rto_mul; + uint64_t deadline; + size_t len = ssm_pk_buff_len(spb); + + memcpy(r->pkt, ssm_pk_buff_head(spb), len); + r->len = len; + r->seqno = seqno; + + clock_gettime(PTHREAD_COND_CLOCK, &now); + r->t0 = TS_TO_UINT64(now); + + rto = LOAD_RELAXED(&frcti->rto); + rto_mul = LOAD_RELAXED(&frcti->rto_mul); + deadline = r->t0 + ((uint64_t) rto << rto_mul); + + pthread_rwlock_wrlock(&frcti->lock); + + assert(before(seqno, frcti->snd_cr.lwe + RQ_SIZE)); + + list_add_tail(&r->next, &frcti->rxm_list); + STORE_RELEASE(&frcti->snd_slots[RQ_SLOT(seqno)].rxm, r); + + pthread_rwlock_unlock(&frcti->lock); + + tw_post(&r->tw, deadline, rxm_due, r); +} + +static void rxm_cancel_all(struct frcti * frcti) +{ + struct list_head * p; + struct list_head * t; + + list_for_each_safe(p, t, &frcti->rxm_list) { + struct rxm_entry * r = list_entry(p, struct rxm_entry, next); + list_del(&r->next); + tw_cancel(&r->tw); + rxm_entry_destroy(r); + STAT_BUMP(frcti, rxm_cancel); + } +} + +static __inline__ void sack_block_put(uint8_t * payload, + uint16_t i, + uint32_t s, + uint32_t e) +{ + uint32_t * blk = (uint32_t *) + (payload + SACK_HDR_SIZE + i * SACK_BLOCK_SIZE); + + blk[0] = hton32(s); + blk[1] = hton32(e); +} + +static __inline__ void sack_block_get(const uint8_t * payload, + uint16_t i, + uint32_t * s, + uint32_t * e) +{ + const uint32_t * blk = (const uint32_t *) + (payload + SACK_HDR_SIZE + i * SACK_BLOCK_SIZE); + + *s = ntoh32(blk[0]); + *e = ntoh32(blk[1]); +} + +/* + * Build SACK blocks for ranges *above* rcv_cr.lwe. Wire invariant + * (see doc/frct.txt §1.3): every block produced here satisfies + * blocks[i].start > rcv_cr.lwe = ackno, which makes the "first block + * below ackno" convention used to mark a D-SACK (RFC 2883 §4 case 1) + * unambiguous. Caller holds frcti->lock. + */ +static uint16_t sack_blocks_build(struct frcti * frcti, + uint32_t blocks[][2], + uint16_t max_n) +{ + const struct rcv_slot * slots = frcti->rcv_slots; + uint32_t s; + uint32_t end; + uint16_t n = 0; + + s = frcti->rcv_cr.lwe + 1; + end = frcti->rcv_cr.lwe + RQ_SIZE; + if (after(end, frcti->rcv_cr.rwe)) + end = frcti->rcv_cr.rwe; + + while (before(s, end) && n < max_n) { + while (before(s, end) && slots[RQ_SLOT(s)].idx == -1) + ++s; + + if (!before(s, end)) + break; + + blocks[n][0] = s; + while (before(s, end) && slots[RQ_SLOT(s)].idx != -1) + ++s; + blocks[n][1] = s; + ++n; + } + + return n; +} + +/* + * Prepend the pending D-SACK report (if any) as block[0]; clear flag. + * Returns the number of slots consumed at the head (0 or 1). Caller + * holds wrlock. + */ +static __inline__ uint16_t dsack_consume(struct frcti * frcti, + uint32_t blocks[][2]) +{ + if (!frcti->dsack_valid || frcti->sack_n_max == 0) + return 0; + + blocks[0][0] = frcti->dsack_seqno; + blocks[0][1] = frcti->dsack_seqno + 1; + frcti->dsack_valid = false; + return 1; +} + +/* Caller must NOT hold frcti->lock. */ +static void frcti_sack_snd(struct frcti * frcti, + const struct sack_args * sa) +{ + struct ssm_pk_buff * spb; + struct frct_pci * pci; + buffer_t buf; + uint16_t i; + + assert(sa->n <= SACK_MAX_BLOCKS); + + buf.len = SACK_HDR_SIZE + sa->n * SACK_BLOCK_SIZE; + + if (frct_ctrl_alloc(&spb, &pci, buf.len) < 0) + return; + + pci->flags = hton16(FRCT_ACK | FRCT_FC | FRCT_SACK); + pci->window = hton32(sa->rwe); + pci->ackno = hton32(sa->ack); + pci->seqno = hton32(FETCH_ADD_RELAXED(&frcti->snd_cr.ackno, 1) + 1); + + frct_hcs_set(pci, false); + + buf.data = FRCT_BODY(pci); + memset(buf.data, 0, SACK_HDR_SIZE); + *(uint16_t *) buf.data = hton16(sa->n); + for (i = 0; i < sa->n; ++i) + sack_block_put(buf.data, i, sa->blocks[i][0], sa->blocks[i][1]); + + frct_tx(frcti, spb); +} + +static void ack_snd(struct frcti * frcti, + bool with_sack) +{ + struct timespec now; + uint64_t now_ns; time_t diff; uint32_t ackno; uint32_t rwe; - int fd; + struct sack_args * sa = NULL; + size_t sa_sz; + bool sacking = false; assert(frcti); + STAT_BUMP(frcti, ack_fire); + clock_gettime(PTHREAD_COND_CLOCK, &now); + now_ns = TS_TO_UINT64(now); + + if (with_sack && frcti->sack_n_max > 0) { + sa_sz = sizeof(*sa) + frcti->sack_n_max * sizeof(sa->blocks[0]); + sa = malloc(sa_sz); + /* If alloc fails, fall through and send a bare cum-ACK. */ + } pthread_rwlock_wrlock(&frcti->lock); - if (!after(frcti->rcv_cr.lwe, frcti->rcv_cr.seqno)) { + /* D-SACK rides through cum-ACK freshness; signal is the duplicate. */ + if (!after(frcti->rcv_cr.lwe, frcti->rcv_cr.seqno) + && !frcti->dsack_valid) { pthread_rwlock_unlock(&frcti->lock); - return; + STAT_BUMP(frcti, ack_supp_seqno); + goto out; } - fd = frcti->fd; ackno = frcti->rcv_cr.lwe; - rwe = frcti->rcv_cr.rwe; + rwe = frcti_advert_rwe(frcti); - diff = ts_diff_ns(&frcti->rcv_cr.act, &now); - if (diff > frcti->a) { + if (ACK_AGED_OUT(frcti->rcv_cr.act, now_ns, frcti->t_a)) { pthread_rwlock_unlock(&frcti->lock); - return; + STAT_BUMP(frcti, ack_supp_inact); + goto out; } - diff = ts_diff_ns(&frcti->snd_cr.act, &now); - if (diff < TICTIME) { + diff = (time_t) ts_age_ns(now_ns, frcti->snd_cr.act); + if (diff < TICTIME && !frcti->dsack_valid) { pthread_rwlock_unlock(&frcti->lock); - return; + STAT_BUMP(frcti, ack_supp_rate); + goto out; } + /* RFC 2018: piggyback SACK on timer ACK; dedup unchanged board. */ + if (sa == NULL || (frcti->sack_n == 0 && !frcti->dsack_valid)) + goto no_sack; + + sa->dsack = false; + sa->n = dsack_consume(frcti, sa->blocks); + if (sa->n == 1) + sa->dsack = true; + + sa->n += sack_blocks_build(frcti, sa->blocks + sa->n, + frcti->sack_n_max - sa->n); + if (sa->n == 0) + goto no_sack; + + if (!sa->dsack && ackno == frcti->sack_lwe && sa->n == frcti->sack_n) + goto no_sack; + + sa->ack = ackno; + sa->rwe = rwe; + frcti->sack_lwe = ackno; + frcti->sack_n = sa->n; + frcti->t_snd_sack = now_ns; + sacking = true; + + no_sack: frcti->rcv_cr.seqno = frcti->rcv_cr.lwe; pthread_rwlock_unlock(&frcti->lock); - __send_frct_pkt(fd, FRCT_ACK | FRCT_FC, ackno, rwe); + STAT_BUMP(frcti, ack_snd); + + if (sacking) { + STAT_BUMP(frcti, sack_snd); + if (sa->dsack) + STAT_BUMP(frcti, dsack_snd); + frcti_sack_snd(frcti, sa); + } else { + frcti_pkt_snd(frcti, FRCT_ACK | FRCT_FC, ackno, rwe); + } + + out: + free(sa); } -static void __send_rdv(int fd) +/* Delayed-ACK timer: per-flow, dedup'd via atomic test-and-set. */ +static void ack_due(void * arg) { - __send_frct_pkt(fd, FRCT_RDVS, 0, 0); + struct frcti * frcti = arg; + + __atomic_clear(&frcti->ack_pending, __ATOMIC_RELAXED); + + ack_snd(frcti, true); } -static struct frcti * frcti_create(int fd, - time_t a, - time_t r, - time_t mpl) +static int ack_arm(struct frcti * frcti) { - struct frcti * frcti; - ssize_t idx; - struct timespec now; - pthread_condattr_t cattr; + struct timespec now; + uint64_t deadline; + + if (__atomic_test_and_set(&frcti->ack_pending, __ATOMIC_RELAXED)) + return 0; + + clock_gettime(PTHREAD_COND_CLOCK, &now); + deadline = TS_TO_UINT64(now) + ACK_DELAY_NS; + + tw_post(&frcti->ack_tw, deadline, ack_due, frcti); + + return 0; +} + +/* Forward decl breaks the keepalive cycle: ka_arm <-> ka_due. */ +static void ka_due(void * arg); + +static int ka_arm(struct frcti * frcti) +{ + struct timespec now; + uint64_t now_ns; + uint64_t timeo_ns; + uint64_t snd_ns; + uint64_t rcv_ns; + uint64_t deadline; + + timeo_ns = (uint64_t) frcti->qs_timeout * MILLION; /* IMM */ + snd_ns = LOAD_RELAXED(&frcti->snd_cr.act) + timeo_ns / 4; + rcv_ns = LOAD_RELAXED(&frcti->rcv_cr.act) + timeo_ns; + + clock_gettime(PTHREAD_COND_CLOCK, &now); + now_ns = TS_TO_UINT64(now); + deadline = MIN(snd_ns, rcv_ns); + if (deadline <= now_ns) + deadline = now_ns + timeo_ns / 4; + + tw_post(&frcti->ka_tw, deadline, ka_due, frcti); + + return 0; +} + +__attribute__((cold)) +static void ka_snd(struct frcti * frcti) +{ + struct ssm_pk_buff * spb; + struct frct_pci * pci; + struct timespec now; + uint64_t now_ns; + time_t timeo_ns; + uint64_t rcv_act; + uint64_t ka_rcv; + int64_t rcv_idle; + int64_t snd_idle; + uint32_t ackno; + + assert(frcti); + + clock_gettime(PTHREAD_COND_CLOCK, &now); + now_ns = TS_TO_UINT64(now); + + timeo_ns = (time_t)(frcti->qs_timeout) * MILLION; /* IMM */ + rcv_act = LOAD_RELAXED(&frcti->rcv_cr.act); + ka_rcv = LOAD_RELAXED(&frcti->t_ka_rcv); + rcv_idle = ts_age_ns(now_ns, rcv_act > ka_rcv ? rcv_act : ka_rcv); + snd_idle = ts_age_ns(now_ns, LOAD_RELAXED(&frcti->snd_cr.act)); + + if (rcv_idle > timeo_ns) { + frct_mark_peer_dead(frcti); + return; + } + + if (snd_idle <= timeo_ns / 4) { + ka_arm(frcti); + return; + } + + if (frct_ctrl_alloc(&spb, &pci, 0) < 0) { + ka_arm(frcti); + return; + } + + ackno = LOAD_RELAXED(&frcti->rcv_cr.lwe); + + pci->flags = hton16(FRCT_KA | FRCT_ACK); + pci->ackno = hton32(ackno); + + frct_hcs_set(pci, false); + + STAT_BUMP(frcti, ka_snd); + frct_tx(frcti, spb); + + ka_arm(frcti); +} + +/* Keepalive timer: re-posted by the fire callback itself. */ +static void ka_due(void * arg) +{ + ka_snd((struct frcti *) arg); +} + +static void frcti_rdv_snd(struct frcti * frcti) +{ + frcti_pkt_snd(frcti, FRCT_RDVS, 0, 0); +} + +#define HAS_RESCNTL(cr) ((cr)->cflags & FRCTFRESCNTL) +static bool frcti_is_window_open(struct frcti * frcti) +{ + struct frct_cr * snd_cr = &frcti->snd_cr; + struct timespec now; + time_t diff; + bool ret = false; + + if (!HAS_RESCNTL(snd_cr)) + return true; + + if (before(snd_cr->seqno, LOAD_RELAXED(&snd_cr->rwe))) + return true; + + /* Window may be closed; wrlock for RDV state mutations. */ + pthread_rwlock_wrlock(&frcti->lock); + + if (before(snd_cr->seqno, snd_cr->rwe)) { + ret = true; + goto unlock; + } + + clock_gettime(PTHREAD_COND_CLOCK, &now); + + if (frcti->open) { + frcti->open = false; + frcti->t_wnd = now; + frcti->t_last_rdv = now; + goto unlock; + } + + diff = ts_diff_ns(&now, &frcti->t_wnd); + if (diff > MAX_RDV) + goto unlock; + + diff = ts_diff_ns(&now, &frcti->t_last_rdv); + if (diff > (time_t) frcti->t_rdv) { + frcti->t_last_rdv = now; + frcti_rdv_snd(frcti); + STAT_BUMP(frcti, rdv_snd); + } + unlock: + pthread_rwlock_unlock(&frcti->lock); + + return ret; +} + +/* n contiguous seqnos free? No RDV: the n=1 path drives it. */ +static bool frcti_is_window_open_n(struct frcti * frcti, + size_t n) +{ + struct frct_cr * snd_cr = &frcti->snd_cr; + + if (!HAS_RESCNTL(snd_cr)) + return true; + + if (n <= 1) + return frcti_is_window_open(frcti); + + return before(snd_cr->seqno + (uint32_t)(n - 1), + LOAD_RELAXED(&snd_cr->rwe)); +} + +static void release_rq(struct frcti * frcti) +{ + size_t i; + + for (i = 0; i < RQ_SIZE; ++i) { + if (frcti->rcv_slots[i].idx == -1) + continue; + + /* Stream rq entries are sentinels (no spb owned). */ + if (!frcti->stream) + frct_spb_release_idx(frcti->rcv_slots[i].idx); + + frcti->rcv_slots[i].idx = -1; + STAT_BUMP(frcti, rq_released); + } +} + +static __inline__ bool stream_ring_sz_ok(struct frcti * frcti, + size_t n) +{ + size_t per_pkt; + + if (n > FRCT_STREAM_RING_SZ_MAX) + return false; + + if ((n & (n - 1)) != 0) + return false; + + per_pkt = frcti->frag_mtu - frcti_data_hdr_len(frcti); + + return n >= FRCT_STREAM_RING_MIN_PKTS * per_pkt; +} + +/* Default ring sized for full RQ_SIZE seqno window; pow2, capped. */ +static size_t default_stream_ring_sz(size_t per_pkt) +{ + size_t need; + size_t sz; + + need = (size_t) RQ_SIZE * per_pkt; + sz = FRCT_STREAM_RING_SZ; + + while (sz < need && sz < FRCT_STREAM_RING_SZ_MAX) + sz <<= 1; + + return sz; +} + +struct frcti * frcti_create(int fd, + uint64_t a, + uint64_t r, + uint64_t mpl, + time_t rtt_hint, + qosspec_t qs, + uint32_t mtu) +{ + struct frcti * frcti; + ssize_t idx; + struct timespec now; + uint64_t now_ns; + size_t bb; + size_t per_pkt; #ifdef PROC_FLOW_STATS - char frctstr[FRCT_NAME_STRLEN + 1]; + char frctstr[FRCT_NAME_STRLEN + 1]; #endif - mpl *= BILLION; - a *= BILLION; - r *= BILLION; + mpl *= MILLION; /* ms -> ns */ + a *= MILLION; /* ms -> ns */ + r *= MILLION; /* ms -> ns */ frcti = malloc(sizeof(*frcti)); if (frcti == NULL) @@ -349,56 +1824,76 @@ static struct frcti * frcti_create(int fd, memset(frcti, 0, sizeof(*frcti)); + list_head_init(&frcti->rxm_list); + if (pthread_rwlock_init(&frcti->lock, NULL)) goto fail_lock; - if (pthread_mutex_init(&frcti->mtx, NULL)) - goto fail_mutex; - - if (pthread_condattr_init(&cattr)) - goto fail_cattr; -#ifndef __APPLE__ - pthread_condattr_setclock(&cattr, PTHREAD_COND_CLOCK); -#endif - if (pthread_cond_init(&frcti->cond, &cattr)) - goto fail_cond; - #ifdef PROC_FLOW_STATS sprintf(frctstr, "%d", fd); if (rib_reg(frctstr, &r_ops)) goto fail_rib_reg; #endif - pthread_condattr_destroy(&cattr); for (idx = 0; idx < RQ_SIZE; ++idx) - frcti->rq[idx] = -1; + frcti->rcv_slots[idx].idx = -1; clock_gettime(PTHREAD_COND_CLOCK, &now); + now_ns = TS_TO_UINT64(now); + + frcti->t_mpl = mpl; + frcti->t_a = a; + frcti->t_r = r; + frcti->t_rdv = DELT_RDV; + frcti->fd = fd; + frcti->ber = (time_t) qs.ber; + frcti->lossy = (qs.loss != 0); + frcti->qs_timeout = (time_t) qs.timeout; + + frcti->frag_mtu = (size_t) mtu; + + /* Cap blocks per SACK at what fits in the per-flow frag_mtu. */ + bb = (frcti->frag_mtu - FRCT_PCILEN - SACK_HDR_SIZE) + / SACK_BLOCK_SIZE; + if (bb > SACK_MAX_BLOCKS) + bb = SACK_MAX_BLOCKS; + frcti->sack_n_max = (uint16_t) bb; + + frcti->max_rcv_sdu = FRCT_MAX_SDU; + + frcti->stream = (qs.service == SVC_STREAM); + if (frcti->stream) { + per_pkt = frcti->frag_mtu - frcti_data_hdr_len(frcti); + frcti->rcv_ring_sz = default_stream_ring_sz(per_pkt); + frcti->ring_seq_cap = + (uint32_t) (frcti->rcv_ring_sz / per_pkt); + } - frcti->mpl = mpl; - frcti->a = a; - frcti->r = r; - frcti->rdv = DELT_RDV; - frcti->fd = fd; - - - frcti->rttseq = 0; - frcti->probe = false; - - frcti->srtt = 0; /* Updated on first ACK */ - frcti->mdev = 10 * MILLION; /* Updated on first ACK */ - frcti->rto = BILLION; /* Initial rxm will be after 1 s */ -#ifdef PROC_FLOW_STATS - frcti->n_rtx = 0; - frcti->n_prb = 0; - frcti->n_rtt = 0; - frcti->n_dup = 0; - frcti->n_dak = 0; - frcti->n_rdv = 0; - frcti->n_out = 0; - frcti->n_rqo = 0; -#endif - if (ai.flows[fd].info.qs.loss == 0) { + frcti->rto_min = (time_t) MAX(RTO_MIN, 1ULL << RXMQ_RES); + rtt_init(frcti, rtt_hint); + frcti->t_min_rtt = now_ns; + frcti->probe_id_next = 1; + frcti->t_rcv_rtt = now_ns; + frcti->t_snd_probe = now_ns; + frcti->t_snd_sack = 0; + frcti->sack_lwe = 0; + frcti->sack_n = 0; + frcti->dsack_seqno = 0; + frcti->dsack_valid = false; + frcti->reo_wnd_mult = 1; + frcti->dsack_lwe_snap = 0; + frcti->t_last_reo_widen = 0; + /* So the first pre-DRF NACK fires without waiting cooldown. */ + frcti->t_nack = now_ns - BILLION; + frcti->in_recovery = false; + frcti->recovery_high = 0; + frcti->rack_fired_lwe = 0; + + tw_init_entry(&frcti->ack_tw); + tw_init_entry(&frcti->ka_tw); + tw_init_entry(&frcti->tlp_tw); + + if (!frcti->lossy) { frcti->snd_cr.cflags |= FRCTFRTX | FRCTFLINGER; frcti->rcv_cr.cflags |= FRCTFRTX; } @@ -406,24 +1901,31 @@ static struct frcti * frcti_create(int fd, frcti->snd_cr.cflags |= FRCTFRESCNTL; frcti->snd_cr.rwe = START_WINDOW; + if (frcti->lossy) + frcti->snd_cr.rwe = RQ_SIZE; + + frcti->snd_cr.inact = 3 * mpl + a + r + BILLION; /* ns */ + frcti->snd_cr.act = now_ns - frcti->snd_cr.inact - BILLION; + + frcti->rcv_cr.inact = 2 * mpl + a + r + BILLION; /* ns */ + frcti->rcv_cr.act = now_ns - frcti->rcv_cr.inact - BILLION; - frcti->snd_cr.inact = (3 * mpl + a + r) / BILLION + 1; /* s */ - frcti->snd_cr.act.tv_sec = now.tv_sec - (frcti->snd_cr.inact + 1); + frcti->t_ka_rcv = now_ns; - frcti->rcv_cr.inact = (2 * mpl + a + r) / BILLION + 1; /* s */ - frcti->rcv_cr.act.tv_sec = now.tv_sec - (frcti->rcv_cr.inact + 1); + /* qs_timeout == 0: no KA, silent peer crash goes undetected. */ + if (frcti->qs_timeout > 0) { + if (ka_arm(frcti) < 0) + goto fail_ka_arm; + } return frcti; + fail_ka_arm: #ifdef PROC_FLOW_STATS + sprintf(frctstr, "%d", fd); + rib_unreg(frctstr); fail_rib_reg: - pthread_cond_destroy(&frcti->cond); #endif - fail_cond: - pthread_condattr_destroy(&cattr); - fail_cattr: - pthread_mutex_destroy(&frcti->mtx); - fail_mutex: pthread_rwlock_destroy(&frcti->lock); fail_lock: free(frcti); @@ -431,21 +1933,55 @@ static struct frcti * frcti_create(int fd, return NULL; } -static void frcti_destroy(struct frcti * frcti) +void frcti_destroy(struct frcti * frcti) { #ifdef PROC_FLOW_STATS char frctstr[FRCT_NAME_STRLEN + 1]; +#endif + /* Drop every wheel entry referencing frcti before freeing it. */ + rxm_cancel_all(frcti); + tw_cancel(&frcti->ack_tw); + tw_cancel(&frcti->ka_tw); + tw_cancel(&frcti->tlp_tw); + +#if defined(PROC_FLOW_STATS) && defined(FRCT_DEBUG_STDOUT) + printf("[FRCT teardown] pid=%d fd=%d " + "sdu_snd=%zu sdu_reasm=%zu sdu_sole=%zu " + "frag_snd=%zu frag_rcv=%zu frag_drop=%zu " + "rxm_rto=%zu rxm_sack=%zu rxm_dup=%zu " + "rxm_due=%zu acked=%zu unowned=%zu aged=%zu defer=%zu " + "cancel=%zu arm_fail=%zu inflight=%u " + "nack_snd=%zu nack_rcv=%zu inact_drop=%zu " + "drf_rebase=%zu rq_released=%zu\n", + (int) getpid(), frcti->fd, + frcti->stat.sdu_snd_frag, frcti->stat.sdu_reasm, + frcti->stat.sdu_sole, + frcti->stat.frag_snd, frcti->stat.frag_rcv, + frcti->stat.frag_drop, + frcti->stat.rxm_rto, frcti->stat.rxm_sack, + frcti->stat.rxm_dupthresh, + frcti->stat.rxm_due_count, frcti->stat.rxm_due_acked, + frcti->stat.rxm_due_unowned, frcti->stat.rxm_due_aged, + frcti->stat.rxm_due_defer, + frcti->stat.rxm_cancel, frcti->stat.rxm_arm_fail, + frcti->snd_cr.seqno - frcti->snd_cr.lwe, + frcti->stat.nack_snd, frcti->stat.nack_rcv, + frcti->stat.inact_drop, + frcti->stat.drf_rebase, frcti->stat.rq_released); +#endif + + release_rq(frcti); + free(frcti->rcv_ring); +#ifdef PROC_FLOW_STATS sprintf(frctstr, "%d", frcti->fd); rib_unreg(frctstr); #endif - pthread_cond_destroy(&frcti->cond); - pthread_mutex_destroy(&frcti->mtx); pthread_rwlock_destroy(&frcti->lock); free(frcti); } -static uint16_t frcti_getflags(struct frcti * frcti) +uint16_t frcti_getflags(struct frcti * frcti) { uint16_t ret; @@ -453,89 +1989,91 @@ static uint16_t frcti_getflags(struct frcti * frcti) pthread_rwlock_rdlock(&frcti->lock); - ret = frcti->snd_cr.cflags; + ret = frcti->snd_cr.cflags & FRCTFMASK; pthread_rwlock_unlock(&frcti->lock); return ret; } -static void frcti_setflags(struct frcti * frcti, - uint16_t flags) +void frcti_setflags(struct frcti * frcti, + uint16_t flags) { - flags |= FRCTFRTX; /* Should not be set by command */ - assert(frcti); - pthread_rwlock_wrlock(&frcti->lock); + flags &= FRCTFSETMASK; - frcti->snd_cr.cflags &= FRCTFRTX; /* Zero other flags */ + pthread_rwlock_wrlock(&frcti->lock); - frcti->snd_cr.cflags &= flags; + frcti->snd_cr.cflags = (frcti->snd_cr.cflags & ~FRCTFSETMASK) | flags; pthread_rwlock_unlock(&frcti->lock); } -#define frcti_queued_pdu(frcti) \ - (frcti == NULL ? idx : __frcti_queued_pdu(frcti)) +size_t frcti_get_max_rcv_sdu(struct frcti * frcti) +{ + size_t ret; -#define frcti_snd(frcti, sdb) \ - (frcti == NULL ? 0 : __frcti_snd(frcti, sdb)) + assert(frcti); -#define frcti_rcv(frcti, sdb) \ - (frcti == NULL ? 0 : __frcti_rcv(frcti, sdb)) + pthread_rwlock_rdlock(&frcti->lock); + ret = frcti->max_rcv_sdu; + pthread_rwlock_unlock(&frcti->lock); -#define frcti_dealloc(frcti) \ - (frcti == NULL ? 0 : __frcti_dealloc(frcti)) + return ret; +} + +int frcti_set_max_rcv_sdu(struct frcti * frcti, + size_t max) +{ + assert(frcti); -#define frcti_is_window_open(frcti) \ - (frcti == NULL ? true : __frcti_is_window_open(frcti)) + if (max == 0) + return -EINVAL; -#define frcti_window_wait(frcti, abstime) \ - (frcti == NULL ? 0 : __frcti_window_wait(frcti, abstime)) + pthread_rwlock_wrlock(&frcti->lock); + frcti->max_rcv_sdu = max; + pthread_rwlock_unlock(&frcti->lock); + return 0; +} -static bool __frcti_is_window_open(struct frcti * frcti) +size_t frcti_get_rcv_ring_sz(struct frcti * frcti) { - struct frct_cr * snd_cr = &frcti->snd_cr; - bool ret = true; + size_t ret; + + assert(frcti); pthread_rwlock_rdlock(&frcti->lock); + ret = frcti->rcv_ring_sz; + pthread_rwlock_unlock(&frcti->lock); - if (snd_cr->cflags & FRCTFRESCNTL) - ret = before(snd_cr->seqno, snd_cr->rwe); + return ret; +} - if (!ret) { - struct timespec now; +/* Set before any stream byte has been delivered; -EBUSY otherwise. */ +int frcti_set_rcv_ring_sz(struct frcti * frcti, + size_t n) +{ + int ret = 0; + size_t per_pkt; - clock_gettime(PTHREAD_COND_CLOCK, &now); + assert(frcti); - pthread_mutex_lock(&frcti->mtx); - if (frcti->open) { - frcti->open = false; - frcti->t_wnd = now; - frcti->t_rdvs = now; - } else { - time_t diff; - diff = ts_diff_ns(&frcti->t_wnd, &now); - if (diff > MAX_RDV) { - pthread_mutex_unlock(&frcti->mtx); - pthread_rwlock_unlock(&frcti->lock); - return false; - } - - diff = ts_diff_ns(&frcti->t_rdvs, &now); - if (diff > frcti->rdv) { - frcti->t_rdvs = now; - __send_rdv(frcti->fd); -#ifdef PROC_FLOW_STATS - frcti->n_rdv++; -#endif + if (!frcti->stream) + return -ENOTSUP; + if (!stream_ring_sz_ok(frcti, n)) + return -EINVAL; - } - } + per_pkt = frcti->frag_mtu - frcti_data_hdr_len(frcti); + + pthread_rwlock_wrlock(&frcti->lock); - pthread_mutex_unlock(&frcti->mtx); + if (frcti->rcv_ring != NULL) { + ret = -EBUSY; + } else { + frcti->rcv_ring_sz = n; + frcti->ring_seq_cap = (uint32_t) (n / per_pkt); } pthread_rwlock_unlock(&frcti->lock); @@ -543,392 +2081,2130 @@ static bool __frcti_is_window_open(struct frcti * frcti) return ret; } -static int __frcti_window_wait(struct frcti * frcti, - struct timespec * abstime) +time_t frcti_get_rto_min(struct frcti * frcti) { - struct frct_cr * snd_cr = &frcti->snd_cr; - int ret = 0; + time_t v; + + assert(frcti); pthread_rwlock_rdlock(&frcti->lock); + v = frcti->rto_min; + pthread_rwlock_unlock(&frcti->lock); - if (!(snd_cr->cflags & FRCTFRESCNTL)) { - pthread_rwlock_unlock(&frcti->lock); + return v; +} + +/* Floor at the timer-wheel resolution; finer granularity is unrepresentable. */ +int frcti_set_rto_min(struct frcti * frcti, + time_t rto_min) +{ + time_t floor = (time_t) (1ULL << RXMQ_RES); + time_t rto_floor; + time_t rto; + + assert(frcti); + + if (rto_min < floor) + return -EINVAL; + + pthread_rwlock_wrlock(&frcti->lock); + + frcti->rto_min = rto_min; + if (frcti->srtt > 0) { + rto_floor = MAX(rto_min, 2 * frcti->srtt); + rto = MAX(rto_floor, + frcti->srtt + (frcti->mdev << MDEV_MUL)); + STORE_RELEASE(&frcti->rto, rto); + } else if (frcti->rto < rto_min) { + STORE_RELEASE(&frcti->rto, rto_min); + } + + pthread_rwlock_unlock(&frcti->lock); + + return 0; +} + +/* Re-arm a fresh rxm so a lost fast-retx still recovers via RTO. */ +static void sack_rxm_snd(struct frcti * frcti, + void * pkt, + size_t len) +{ + struct ssm_pk_buff * spb; + const struct frct_pci * pci; + struct rxm_entry * rxm; + uint32_t rcv_lwe; + uint32_t seqno; + int ret; + + rcv_lwe = LOAD_RELAXED(&frcti->rcv_cr.lwe); + + spb = rxm_pkt_prepare(pkt, len, rcv_lwe, frcti->stream); + if (spb == NULL) + return; + + pci = (const struct frct_pci *) ssm_pk_buff_head(spb); + seqno = ntoh32(pci->seqno); + + rxm = rxm_alloc(frcti, ssm_pk_buff_len(spb)); + if (rxm == NULL) { + frct_spb_release(spb); + return; + } + rxm_arm(frcti, seqno, rxm, spb); + + STAT_BUMP(frcti, rxm_sack); + ret = frct_tx(frcti, spb); + if (ret == -EFLOWDOWN || ret == -ENOTALLOC) + STAT_BUMP(frcti, rxm_tx_dead); +} + +/* Additive HoL emit; original snd_slots[hp].rxm stays armed (NewReno). */ +static int fast_rxm_send(struct frcti * frcti, + void * pkt, + size_t len) +{ + struct ssm_pk_buff * spb; + uint32_t rcv_lwe; + + rcv_lwe = LOAD_RELAXED(&frcti->rcv_cr.lwe); + + spb = rxm_pkt_prepare(pkt, len, rcv_lwe, frcti->stream); + if (spb == NULL) return 0; + + return frct_tx(frcti, spb); +} + +/* PCI bytes survive head_release at receive; just rewind the pointer. */ +static __inline__ uint16_t frag_role_peek(struct ssm_pk_buff * spb) +{ + const struct frct_pci * pci; + + assert(ssm_pk_buff_head(spb) != NULL); + + pci = (const struct frct_pci *) (ssm_pk_buff_head(spb) - FRCT_PCILEN); + + return ntoh16(pci->flags) & FRCT_FR_MASK; +} + +enum frag_state { + FRAG_NOT_READY, /* head missing / FIRST..LAST run incomplete */ + FRAG_DELIVER, /* *count fragments form a deliverable SDU */ + FRAG_DROP, /* *count fragments at lwe are malformed */ +}; + +/* + * On a gap in the run: FRTX waits (NOT_READY); best-effort scans forward + * for the next FIRST/SOLE and returns DROP for the broken prefix. *count + * gets the offset from the trailing edge. NOT_READY if no later run is + * in window. Caller rdlock. + */ +static enum frag_state frag_inspect_gap(struct frcti * frcti, + size_t start, + size_t * count) +{ + const struct rcv_slot * slots = frcti->rcv_slots; + struct ssm_pk_buff * spb; + uint32_t k; + uint16_t role; + size_t m; + + if (frcti->rcv_cr.cflags & FRCTFRTX) + return FRAG_NOT_READY; + + k = frcti->rcv_cr.rwe - RQ_SIZE; + + for (m = start; m < RQ_SIZE; ++m) { + if (slots[RQ_SLOT(k + m)].idx == -1) + continue; + + spb = rq_frag(frcti, k + m); + role = frag_role_peek(spb); + + if (role == FRCT_FR_SOLE || role == FRCT_FR_FIRST) { + if (m == 0) + return FRAG_NOT_READY; + + *count = m; + return FRAG_DROP; + } } - while (snd_cr->seqno == snd_cr->rwe && ret != -ETIMEDOUT) { - struct timespec now; - pthread_rwlock_unlock(&frcti->lock); - pthread_mutex_lock(&frcti->mtx); + return FRAG_NOT_READY; +} + +/* + * Inspect rq[lwe..]; set *count and return DELIVER/DROP/NOT_READY. DROP + * covers broken prefixes (mid/last at HoL, FIRST..[non-LAST]..new-FIRST). + * Non-FRTX flows skip past gaps to the next FIRST/SOLE. Caller rdlock. + */ +static enum frag_state frag_run_inspect(struct frcti * frcti, + size_t * count) +{ + const struct rcv_slot * slots = frcti->rcv_slots; + struct ssm_pk_buff * spb; + uint32_t k = frcti->rcv_cr.rwe - RQ_SIZE; + uint16_t role; + size_t n = 0; + + if (slots[RQ_SLOT(k)].idx == -1) + return frag_inspect_gap(frcti, 0, count); + + spb = rq_frag(frcti, k); + role = frag_role_peek(spb); + + if (role == FRCT_FR_SOLE) { + *count = 1; + return FRAG_DELIVER; + } + + if (role != FRCT_FR_FIRST) { + *count = 1; + return FRAG_DROP; + } - if (frcti->open) { - clock_gettime(PTHREAD_COND_CLOCK, &now); + while (true) { + if (n == RQ_SIZE || slots[RQ_SLOT(k + n)].idx == -1) + return frag_inspect_gap(frcti, n, count); - frcti->t_wnd = now; - frcti->t_rdvs = now; - frcti->open = false; + spb = rq_frag(frcti, k + n); + role = frag_role_peek(spb); + ++n; + + if (role == FRCT_FR_LAST) { + *count = n; + return FRAG_DELIVER; } - pthread_cleanup_push(__cleanup_mutex_unlock, &frcti->mtx); + if (n > 1 && role != FRCT_FR_MID) { + /* SOLE or new FIRST mid-run: drop the prefix. */ + *count = n - 1; + return FRAG_DROP; + } + } +} - ret = -__timedwait(&frcti->cond, &frcti->mtx, abstime); +/* Caller wrlock. Delivery edge is implicit: rwe - RQ_SIZE. */ +static void frag_drop(struct frcti * frcti, + size_t count) +{ + uint32_t k = frcti->rcv_cr.rwe - RQ_SIZE; + uint32_t edge; + size_t i; - pthread_cleanup_pop(false); + for (i = 0; i < count; ++i) { + size_t pos = RQ_SLOT(k + i); - if (ret == -ETIMEDOUT) { - time_t diff; + if (frcti->rcv_slots[pos].idx == -1) + continue; - clock_gettime(PTHREAD_COND_CLOCK, &now); + frct_spb_release_idx(frcti->rcv_slots[pos].idx); + frcti->rcv_slots[pos].idx = -1; + } - diff = ts_diff_ns(&frcti->t_wnd, &now); - if (diff > MAX_RDV) { - pthread_mutex_unlock(&frcti->mtx); - return -ECONNRESET; /* write fails! */ - } + frcti->rcv_cr.rwe += count; - diff = ts_diff_ns(&frcti->t_rdvs, &now); - if (diff > frcti->rdv) { - frcti->t_rdvs = now; - __send_rdv(frcti->fd); - } + /* Drop may span a gap; pull lwe up to preserve rwe - RQ_SIZE <= lwe. */ + edge = frcti->rcv_cr.rwe - RQ_SIZE; + if (before(frcti->rcv_cr.lwe, edge)) + STORE_RELEASE(&frcti->rcv_cr.lwe, edge); +} + +/* Copy `count` fragments at rq[lwe..] into buf; release + advance lwe. */ +static size_t frag_gather(struct frcti * frcti, + size_t count, + uint8_t * buf) +{ + struct ssm_pk_buff * frag; + size_t off = 0; + size_t i; + uint32_t k = frcti->rcv_cr.rwe - RQ_SIZE; + + for (i = 0; i < count; ++i) { + size_t pos = RQ_SLOT(k + i); + size_t flen; + + frag = rq_frag(frcti, k + i); + flen = ssm_pk_buff_len(frag); + memcpy(buf + off, ssm_pk_buff_head(frag), flen); + off += flen; + frct_spb_release_idx(frcti->rcv_slots[pos].idx); + frcti->rcv_slots[pos].idx = -1; + } + + frcti->rcv_cr.rwe += count; + + return off; +} + +/* Caller holds lock. */ +static size_t frag_total_len(struct frcti * frcti, + size_t count, + bool * overflow) +{ + struct ssm_pk_buff * frag; + size_t total = 0; + size_t i; + uint32_t k = frcti->rcv_cr.rwe - RQ_SIZE; + + *overflow = false; + + for (i = 0; i < count; ++i) { + size_t flen; + + frag = rq_frag(frcti, k + i); + flen = ssm_pk_buff_len(frag); + if (total + flen < total) { + *overflow = true; + return 0; } + total += flen; + } + + return total; +} + +/* + * Process a delivered slot at lwe: latch FIN if acceptable, + * advance byte_high (clamped to byte_fin once latched). + */ +static __inline__ void stream_deliver_slot(struct frcti * frcti, + size_t lp) +{ + uint32_t end; + + end = frcti->rcv_slots[lp].end; - pthread_mutex_unlock(&frcti->mtx); - pthread_rwlock_rdlock(&frcti->lock); + if (frcti->rcv_slots[lp].fin) { + if (end == frcti->rcv_byte_high && !frcti->rcv_fin_seen) { + frcti->rcv_fin_seen = true; + frcti->rcv_byte_fin = end; + } else { + STAT_BUMP(frcti, strm_fin_drop); + } + } + + if (frcti->rcv_fin_seen && after(end, frcti->rcv_byte_fin)) + end = frcti->rcv_byte_fin; + + frcti->rcv_byte_high = end; +} + +/* Two-segment memcpy from buf into the rx ring at byte offset start. */ +static void stream_ring_write(struct frcti * frcti, + uint32_t start, + buffer_t buf) +{ + size_t mask = frcti->rcv_ring_sz - 1; + size_t off = start & mask; + + if (off + buf.len <= frcti->rcv_ring_sz) { + memcpy(frcti->rcv_ring + off, buf.data, buf.len); + } else { + size_t first = frcti->rcv_ring_sz - off; + memcpy(frcti->rcv_ring + off, buf.data, first); + memcpy(frcti->rcv_ring, buf.data + first, buf.len - first); } +} + +/* Two-segment memcpy from the rx ring at byte offset start into buf. */ +static void stream_ring_read(struct frcti * frcti, + uint32_t start, + buffer_t buf) +{ + size_t mask = frcti->rcv_ring_sz - 1; + size_t off = start & mask; + + if (off + buf.len <= frcti->rcv_ring_sz) { + memcpy(buf.data, frcti->rcv_ring + off, buf.len); + } else { + size_t first = frcti->rcv_ring_sz - off; + memcpy(buf.data, frcti->rcv_ring + off, first); + memcpy(buf.data + first, frcti->rcv_ring, buf.len - first); + } +} + +/* Deliver-or-drop one stashed slot at lwe; advance lwe/rwe. Caller wrlock. */ +static void stream_advance_lwe(struct frcti * frcti) +{ + size_t lp; + lp = RQ_SLOT(frcti->rcv_cr.lwe); + + if (frcti->rcv_slots[lp].start != frcti->rcv_byte_high) + STAT_BUMP(frcti, strm_drop); + else + stream_deliver_slot(frcti, lp); + + frcti->rcv_slots[lp].fin = 0; + frcti->rcv_slots[lp].idx = -1; + STORE_RELEASE(&frcti->rcv_cr.lwe, frcti->rcv_cr.lwe + 1); + frcti->rcv_cr.rwe++; +} + +/* + * Validate a stream DATA packet before stashing. Returns 0 if the + * packet may be written into rcv_ring + rq[], -1 otherwise. + */ +static __inline__ int stream_stash_check(struct frcti * frcti, + uint32_t start, + uint32_t end, + size_t plen, + uint16_t flags) +{ + if (end - start != (uint32_t) plen) + return -1; + + /* FIN MUST be 0-byte. */ + if ((flags & FRCT_FIN) && plen != 0) + return -1; + + /* Post-EOS: no further FIN once latched. */ + if (frcti->rcv_fin_seen && (flags & FRCT_FIN)) + return -1; + + /* Post-EOS: reject data at or past byte_fin. */ + if (frcti->rcv_fin_seen && !before(start, frcti->rcv_byte_fin)) + return -1; + + /* Stale: peer is behind the delivered edge. */ + if (before(end, frcti->rcv_byte_next)) + return -1; + + /* Exact-edge: only an empty-stream FIN is meaningful. */ + if (end == frcti->rcv_byte_next && !(flags & FRCT_FIN)) + return -1; + + if (end - frcti->rcv_byte_next > frcti->rcv_ring_sz) + return -1; + + return 0; +} + +/* + * Stream-mode DATA receive: validate, stash payload in rcv_ring, mark + * rq[pos], advance lwe through any newly-contiguous run. Returns 0 + * (spb released) or -1 (caller releases). Caller wrlock. + */ +static int frcti_stream_data_rcv(struct frcti * frcti, + struct ssm_pk_buff * spb, + size_t pos, + uint16_t flags) +{ + struct frct_pci_stream * spci; + uint32_t start; + uint32_t end; + buffer_t buf; + size_t skip; + + if (ssm_pk_buff_len(spb) < FRCT_PCI_STREAM_LEN) + return -1; + + if (frcti->rcv_ring == NULL) { + frcti->rcv_ring = calloc(1, frcti->rcv_ring_sz); + if (frcti->rcv_ring == NULL) + return -ENOMEM; + } + + spci = FRCT_HDR_POP(spb, frct_pci_stream); + start = ntoh32(spci->start); + end = ntoh32(spci->end); + + buf.data = ssm_pk_buff_head(spb); + buf.len = ssm_pk_buff_len(spb); + + if (stream_stash_check(frcti, start, end, buf.len, flags) < 0) + return -1; + + /* Trim front-overlap with already-delivered region. */ + if (before(start, frcti->rcv_byte_next)) { + skip = frcti->rcv_byte_next - start; + buf.data += skip; + buf.len -= skip; + start = frcti->rcv_byte_next; + } + + stream_ring_write(frcti, start, buf); + STAT_ADD(frcti, strm_rcv_byte, buf.len); + + frcti->rcv_slots[pos].idx = 1; + frcti->rcv_slots[pos].start = start; + frcti->rcv_slots[pos].end = end; + frcti->rcv_slots[pos].fin = (flags & FRCT_FIN) ? 1 : 0; + + while (frcti->rcv_slots[RQ_SLOT(frcti->rcv_cr.lwe)].idx != -1) + stream_advance_lwe(frcti); + + frct_spb_release(spb); + + return 0; +} + +/* + * DATA receive: stash idx at rq[pos], advance lwe through any + * contiguous run. Caller wrlock. + */ +static void frcti_data_stash(struct frcti * frcti, + ssize_t idx, + size_t pos, + uint16_t flags) +{ + frcti->rcv_slots[pos].idx = idx; + + if ((flags & FRCT_FR_MASK) != FRCT_FR_SOLE) + STAT_BUMP(frcti, frag_rcv); + + /* lwe = cum-ACK edge; advance per fragment through contiguous run. */ + while (before(frcti->rcv_cr.lwe, frcti->rcv_cr.rwe) + && frcti->rcv_slots[RQ_SLOT(frcti->rcv_cr.lwe)].idx != -1) + STORE_RELEASE(&frcti->rcv_cr.lwe, frcti->rcv_cr.lwe + 1); +} + +/* Stream consume: copy up to `count` contiguous bytes from ring into buf. */ +static ssize_t frcti_consume_stream(struct frcti * frcti, + uint8_t * buf, + size_t count) +{ + size_t avail; + size_t copy; + ssize_t ret; + buffer_t dst; + + assert(frcti); + + pthread_rwlock_wrlock(&frcti->lock); + + avail = (size_t) (frcti->rcv_byte_high - frcti->rcv_byte_next); + if (avail == 0) { + /* EOS drained: signal EOF to the reader. */ + if (frcti->rcv_fin_seen + && frcti->rcv_byte_next == frcti->rcv_byte_fin) + ret = 0; + else + ret = -EAGAIN; + goto unlock; + } + + copy = MIN(avail, count); + + dst.data = buf; + dst.len = copy; + stream_ring_read(frcti, frcti->rcv_byte_next, dst); + + frcti->rcv_byte_next += (uint32_t) copy; + STAT_ADD(frcti, strm_dlv_byte, copy); + + ret = (ssize_t) copy; + + unlock: pthread_rwlock_unlock(&frcti->lock); return ret; } -static ssize_t __frcti_queued_pdu(struct frcti * frcti) +/* + * FRTX consume: copy next ready PDU (full SDU or nothing). Returns bytes, + * -EAGAIN (no PDU), or -EMSGSIZE (oversize: run dropped to unblock flow). + */ +static ssize_t frcti_consume(struct frcti * frcti, + uint8_t * buf, + size_t count) { - ssize_t idx; - size_t pos; + size_t n; + size_t total; + bool overflow; + enum frag_state st; + ssize_t ret; assert(frcti); - /* See if we already have the next PDU. */ pthread_rwlock_wrlock(&frcti->lock); - pos = frcti->rcv_cr.lwe & (RQ_SIZE - 1); - - idx = frcti->rq[pos]; - if (idx != -1) { - ++frcti->rcv_cr.lwe; - ++frcti->rcv_cr.rwe; - frcti->rq[pos] = -1; + while (true) { + st = frag_run_inspect(frcti, &n); + if (st == FRAG_NOT_READY) { + ret = -EAGAIN; + goto unlock; + } + if (st == FRAG_DROP) { + STAT_ADD(frcti, frag_drop, n); + frag_drop(frcti, n); + continue; + } + /* FRAG_DELIVER */ + total = frag_total_len(frcti, n, &overflow); + if (overflow || total > frcti->max_rcv_sdu || total > count) { + STAT_ADD(frcti, frag_drop, n); + frag_drop(frcti, n); + ret = -EMSGSIZE; + goto unlock; + } + ret = (ssize_t) frag_gather(frcti, n, buf); + if (n > 1) + STAT_BUMP(frcti, sdu_reasm); + else + STAT_BUMP(frcti, sdu_sole); + goto unlock; } + unlock: pthread_rwlock_unlock(&frcti->lock); - return idx; + return ret; } -static ssize_t __frcti_pdu_ready(struct frcti * frcti) +static bool frcti_pdu_ready(struct frcti * frcti) { - ssize_t idx; - size_t pos; + size_t pos; + size_t count; + bool ready; assert(frcti); - /* See if we already have the next PDU. */ pthread_rwlock_rdlock(&frcti->lock); - pos = frcti->rcv_cr.lwe & (RQ_SIZE - 1); - idx = frcti->rq[pos]; + if (frcti->stream) { + ready = frcti->rcv_byte_high != frcti->rcv_byte_next; + pthread_rwlock_unlock(&frcti->lock); + return ready; + } + + if (frag_run_inspect(frcti, &count) != FRAG_DELIVER) { + /* Drop case: frcti_consume will handle it; not ready. */ + pthread_rwlock_unlock(&frcti->lock); + return false; + } + + pos = RQ_SLOT(frcti->rcv_cr.rwe - RQ_SIZE); + ready = frcti->rcv_slots[pos].idx != -1; + + pthread_rwlock_unlock(&frcti->lock); + + return ready; +} + +/* No srtt yet: probe at the cold-probe cadence to seed it. */ +#define PROBE_DUE_COLD(frcti, now_ns) \ + ((now_ns) - (frcti)->t_snd_probe > (uint64_t) RTTP_COLD_NS) + +/* Have srtt: probe when peer quiet for > 2*srtt and last probe > srtt. */ +#define PROBE_DUE_WARM(frcti, now_ns) \ + ((now_ns) - (frcti)->t_rcv_rtt > 2u * (uint64_t)(frcti)->srtt \ + && (now_ns) - (frcti)->t_snd_probe > (uint64_t)(frcti)->srtt) + +/* Seeds srtt for receive-only sides so they don't fall back to 1 s RTO. */ +__attribute__((cold)) +static void frcti_rcv_probe(struct frcti * frcti, + uint64_t now_ns) +{ + uint32_t probe_id; + uint8_t nonce[RTTP_NONCE_LEN] = { 0 }; + + pthread_rwlock_wrlock(&frcti->lock); + + if (frcti->srtt == 0 && !PROBE_DUE_COLD(frcti, now_ns)) { + pthread_rwlock_unlock(&frcti->lock); + return; + } + + if (frcti->srtt != 0 && !PROBE_DUE_WARM(frcti, now_ns)) { + pthread_rwlock_unlock(&frcti->lock); + return; + } + + probe_id = rttp_alloc_probe(frcti, now_ns, nonce); pthread_rwlock_unlock(&frcti->lock); - return idx; + if (probe_id != 0) + frcti_rttp_snd(frcti, probe_id, 0, nonce); } -#include <timerwheel.c> +/* Echo at slot `pos` matches our probe: id, slot, nonce all intact. */ +static __inline__ bool probe_echo_matches(struct frcti * frcti, + size_t pos, + uint32_t echo_id, + const uint8_t nonce[RTTP_NONCE_LEN]) +{ + if (frcti->probes[pos].id != echo_id) + return false; + + if (frcti->probes[pos].ts == 0) + return false; + + return memcmp(frcti->probes[pos].nonce, nonce, RTTP_NONCE_LEN) == 0; +} /* - * Send a final ACK for everything that has not been ACK'd. - * If the flow should be kept active for retransmission, - * the returned time will be negative. + * RTT probe (echo_id == 0): bounce the nonce back to peer. + * RTT echo (echo_id != 0): verify nonce + feed sample. */ -static time_t __frcti_dealloc(struct frcti * frcti) +static void frcti_rttp_rcv(struct frcti * frcti, + buffer_t pkt, + uint64_t now_ns) { - struct timespec now; - time_t wait; - int ackno; - int fd = -1; + const struct frct_rttp * rttp; + uint32_t probe_id; + uint32_t echo_id; + uint8_t nonce[RTTP_NONCE_LEN]; + size_t ring_pos; + int64_t elapsed; + uint64_t sample; + + if (pkt.len < RTTP_PAYLOAD) + return; + + rttp = (const struct frct_rttp *) pkt.data; + probe_id = ntoh32(rttp->probe_id); + echo_id = ntoh32(rttp->echo_id); + + /* Forged/malformed: bouncing this would loop on echo_id == 0. */ + if (probe_id == 0 && echo_id == 0) + return; + + memcpy(nonce, rttp->nonce, sizeof(nonce)); + + if (echo_id == 0) { + /* Probe: echo back with same nonce so peer can verify. */ + STAT_BUMP(frcti, rttp_rcv); + frcti_rttp_snd(frcti, 0, probe_id, nonce); + return; + } + + ring_pos = RTTP_POS(echo_id); + + pthread_rwlock_wrlock(&frcti->lock); + + if (!probe_echo_matches(frcti, ring_pos, echo_id, nonce)) { + pthread_rwlock_unlock(&frcti->lock); + return; + } + + elapsed = ts_age_ns(now_ns, frcti->probes[ring_pos].ts); + frcti->probes[ring_pos].ts = 0; + frcti->t_rcv_rtt = now_ns; + + if (elapsed <= 0) { + pthread_rwlock_unlock(&frcti->lock); + return; + } + sample = (uint64_t) elapsed; + + /* Clamp probe sample to RTT_CLAMP_MUL * srtt to avoid poisoning. */ + if (frcti->srtt > 0) + sample = MIN(sample, (uint64_t) frcti->srtt * RTT_CLAMP_MUL); + + rtt_update(frcti, sample, now_ns); + + pthread_rwlock_unlock(&frcti->lock); +} + +/* Honours piggybacked ACK on the KA. */ +static void frcti_ka_rcv(struct frcti * frcti, + const struct frct_pci * pci, + uint64_t now_ns, + uint16_t flags) +{ + uint32_t ka_ackno; + + STORE_RELEASE(&frcti->t_ka_rcv, now_ns); + STAT_BUMP(frcti, ka_rcv); + + if (!(flags & FRCT_ACK)) + return; + + ka_ackno = ntoh32(pci->ackno); + + pthread_rwlock_wrlock(&frcti->lock); + + if (within(ka_ackno, frcti->snd_cr.lwe, frcti->snd_cr.seqno)) + STORE_RELEASE(&frcti->snd_cr.lwe, ka_ackno); + + pthread_rwlock_unlock(&frcti->lock); +} + +/* + * Additive HoL re-emit (carries DRF); runs before rcv_cr->act + * refresh so it doesn't pre-empt peer's first DRF. + */ +__attribute__((cold)) +static void frcti_nack_rcv(struct frcti * frcti) +{ + struct timespec now; + uint64_t now_ns; + size_t hp; + struct rxm_entry * rxm; + void * pkt_copy = NULL; + size_t pkt_len = 0; clock_gettime(PTHREAD_COND_CLOCK, &now); + now_ns = TS_TO_UINT64(now); + + pthread_rwlock_wrlock(&frcti->lock); + + STAT_BUMP(frcti, nack_rcv); + + if (frcti->snd_cr.seqno == frcti->snd_cr.lwe) { + pthread_rwlock_unlock(&frcti->lock); + return; + } + + hp = RQ_SLOT(frcti->snd_cr.lwe); + rxm = LOAD_ACQUIRE(&frcti->snd_slots[hp].rxm); + if (rxm == NULL || RXM_AGED_OUT(rxm->t0, now_ns, frcti->t_r)) { + pthread_rwlock_unlock(&frcti->lock); + return; + } + + pkt_copy = malloc(rxm->len); + if (pkt_copy != NULL) { + memcpy(pkt_copy, rxm->pkt, rxm->len); + pkt_len = rxm->len; + /* Karn: suppress RTT sample. NACK supersedes pending TLP. */ + frcti->snd_slots[hp].flags = + (frcti->snd_slots[hp].flags & ~SND_TLP) + | SND_RTX | SND_FAST_RXM; + frcti->rtt_lwe = frcti->snd_cr.lwe + 1; + STAT_BUMP(frcti, rxm_nack); + } + + pthread_rwlock_unlock(&frcti->lock); + + if (pkt_copy != NULL) { + int ret = fast_rxm_send(frcti, pkt_copy, pkt_len); + if (ret == -EFLOWDOWN || ret == -ENOTALLOC) + STAT_BUMP(frcti, rxm_tx_dead); + free(pkt_copy); + } +} + +__attribute__((cold)) +static void frcti_rdv_rcv(struct frcti * frcti) +{ + uint32_t rwe; pthread_rwlock_rdlock(&frcti->lock); - ackno = frcti->rcv_cr.lwe; - if (frcti->rcv_cr.lwe != frcti->rcv_cr.seqno) - fd = frcti->fd; + rwe = frcti_advert_rwe(frcti); + + pthread_rwlock_unlock(&frcti->lock); + + STAT_BUMP(frcti, rdv_rcv); + + frcti_pkt_snd(frcti, FRCT_FC, 0, rwe); +} + +/* §7.2: PTO = 2*SRTT + max delayed-ACK delay; fallback when unseeded. */ +static __inline__ uint64_t tlp_pto(const struct frcti * frcti) +{ + if (frcti->srtt > 0) + return 2ULL * (uint64_t) frcti->srtt + ACK_DELAY_NS; + + return NACK_COOLDOWN_NS; +} + +/* + * RFC 8985 §7: lazy probe. Re-evaluate on fire — if sender was active + * within PTO, re-post; else probe HoL once and hand off to RTO. + */ +__attribute__((cold)) +static void tlp_due(void * arg) +{ + struct frcti * frcti = arg; + struct timespec now; + uint64_t now_ns; + uint64_t pto; + uint64_t rto_at; + size_t hp; + struct rxm_entry * rxm; + void * pkt_copy = NULL; + size_t pkt_len = 0; + bool re_post = false; + uint64_t deadline = 0; + + clock_gettime(PTHREAD_COND_CLOCK, &now); + now_ns = TS_TO_UINT64(now); + + pthread_rwlock_wrlock(&frcti->lock); + + if (frcti->snd_cr.seqno == frcti->snd_cr.lwe) + goto unlock; + + if (!before(frcti->snd_cr.seqno, frcti->snd_cr.rwe)) + goto unlock; /* FC-blocked: RDV handles it. */ - wait = MAX(frcti->rcv_cr.inact - now.tv_sec + frcti->rcv_cr.act.tv_sec, - frcti->snd_cr.inact - now.tv_sec + frcti->snd_cr.act.tv_sec); - wait = MAX(wait, 0); + /* RFC 8985 §7.3: one outstanding probe, MAX_TLP_PER_EP per ep. */ + if (frcti->tlp_high_seq != 0) + goto unlock; - if (frcti->snd_cr.cflags & FRCTFLINGER - && before(frcti->snd_cr.lwe, frcti->snd_cr.seqno)) - wait = -wait; + if (frcti->tlp_count >= MAX_TLP_PER_EP) + goto unlock; + pto = tlp_pto(frcti); + + /* §7.2: anchor PTO on most recent send; defer if still active. */ + if (now_ns < frcti->snd_cr.act + pto) { + deadline = frcti->snd_cr.act + pto; + re_post = true; + goto unlock; + } + + hp = RQ_SLOT(frcti->snd_cr.lwe); + rxm = LOAD_ACQUIRE(&frcti->snd_slots[hp].rxm); + if (rxm == NULL || RXM_AGED_OUT(rxm->t0, now_ns, frcti->t_r)) + goto unlock; + + /* Cap: if HoL RTO is due, let rxm_due fire instead. */ + rto_at = rxm->t0 + ((uint64_t) frcti->rto + << LOAD_RELAXED(&frcti->rto_mul)); + if (rto_at <= now_ns) + goto unlock; + + pkt_copy = malloc(rxm->len); + if (pkt_copy != NULL) { + memcpy(pkt_copy, rxm->pkt, rxm->len); + pkt_len = rxm->len; + frcti->snd_slots[hp].time = now_ns; + frcti->snd_slots[hp].flags |= SND_TLP | SND_FAST_RXM; + frcti->rtt_lwe = frcti->snd_cr.lwe + 1; + /* §7.3 outstanding-probe marker; ack_rcv/rxm_snd clear. */ + frcti->tlp_high_seq = frcti->snd_cr.seqno; + frcti->tlp_count++; + STAT_BUMP(frcti, tlp_snd); + } + + unlock: pthread_rwlock_unlock(&frcti->lock); - if (fd != -1) - __send_frct_pkt(fd, FRCT_ACK, ackno, 0); + if (pkt_copy != NULL) { + fast_rxm_send(frcti, pkt_copy, pkt_len); + free(pkt_copy); + } + + if (re_post) + tw_post(&frcti->tlp_tw, deadline, tlp_due, frcti); + else + __atomic_clear(&frcti->tlp_pending, __ATOMIC_RELAXED); +} + +/* §7.2 lazy: post once per quiet period. tlp_due re-evaluates on fire. */ +static int tlp_arm(struct frcti * frcti) +{ + struct timespec now; + uint64_t now_ns; + uint64_t pto; + uint64_t deadline; + + /* §7.3: one outstanding probe, MAX_TLP_PER_EP per recovery ep. */ + if (LOAD_RELAXED(&frcti->tlp_high_seq) != 0) + return 0; + if (LOAD_RELAXED(&frcti->tlp_count) >= MAX_TLP_PER_EP) + return 0; + if (__atomic_test_and_set(&frcti->tlp_pending, __ATOMIC_RELAXED)) + return 0; + + clock_gettime(PTHREAD_COND_CLOCK, &now); + now_ns = TS_TO_UINT64(now); + + pto = tlp_pto(frcti); + + deadline = LOAD_RELAXED(&frcti->snd_cr.act) + pto; + if (deadline <= now_ns) + deadline = now_ns + pto; + + tw_post(&frcti->tlp_tw, deadline, tlp_due, frcti); + + return 0; +} + +/* + * FC window advert from any flag-bearing packet. Caps at lwe + RQ_SIZE, + * rejects backward shrink (forged/stale FC), marks window open. + * Caller wrlock. + */ +static __inline__ void frcti_fc_rcv(struct frcti * frcti, + const struct frct_pci * pci) +{ + struct frct_cr * snd_cr; + uint32_t rwe; + uint32_t rwe_max; + + snd_cr = &frcti->snd_cr; + rwe = ntoh32(pci->window); + rwe_max = snd_cr->lwe + RQ_SIZE; + + if (after(rwe, rwe_max)) + rwe = rwe_max; + + /* Reject backward shrink (forged/stale FC). */ + if (before(rwe, snd_cr->rwe)) + rwe = snd_cr->rwe; + + STORE_RELAXED(&snd_cr->rwe, rwe); + frcti->open = true; +} + +/* Packet copies captured under frcti->lock; emitted after release. */ +struct pending { + buffer_t fast_rxm; + buffer_t sack_rxm[SACK_RXM_MAX]; + size_t sack_rxm_cnt; +}; + +/* RFC 6582 §3.2: seal recovery_high on entry; do not extend on new gaps. */ +static void recovery_enter(struct frcti * frcti) +{ + if (frcti->in_recovery) + return; + + frcti->in_recovery = true; + frcti->recovery_high = frcti->snd_cr.seqno + RTT_QUARANTINE; +} + +/* True when cum-ACK clears recovery_high or all in-flight ACKed. */ +static bool recovery_exit_reached(struct frcti * frcti, + uint32_t ackno) +{ + if (!frcti->in_recovery) + return false; + + if (!before(ackno, frcti->recovery_high)) + return true; + + return ackno == frcti->snd_cr.seqno; +} + +/* RTT sample gate: Karn + SACK-consume + don't-seed. */ +static bool rtt_sample_eligible(struct frcti * frcti, + size_t p, + uint16_t flags, + uint32_t lwe) +{ + if (flags & FRCT_RXM) + return false; + if (frcti->snd_slots[p].flags & (SND_RTX | SND_TLP)) + return false; + if (LOAD_ACQUIRE(&frcti->snd_slots[p].rxm) == NULL) + return false; + if (before(lwe, frcti->rtt_lwe)) + return false; + /* Don't seed srtt from a cum-ACK; let probes seed. */ + if (frcti->srtt == 0) + return false; + return true; +} + +#define RXM_SLOT_EMPTY(rxm) ((rxm) == NULL) +#define FAST_RXM_STAGED(pending) ((pending)->fast_rxm.data != NULL) +#define RXM_FAST_DONE(flags) (((flags) & SND_FAST_RXM) != 0) + +/* RACK fast retransmit on cum-ACK: HoL aged past R, not yet retransmitted. */ +static void fast_rxm_consider(struct frcti * frcti, + uint64_t now_ns, + struct pending * pending) +{ + struct rxm_entry * rxm; + struct snd_slot * slot; + size_t hp; + uint64_t R; + bool rack_ok; + + hp = RQ_SLOT(frcti->snd_cr.lwe); + slot = &frcti->snd_slots[hp]; + rxm = LOAD_ACQUIRE(&slot->rxm); + R = rack_reorder_window(frcti); + + if (RXM_SLOT_EMPTY(rxm)) + return; + + /* RFC 8985 §6.2: time-based RACK OR DupThresh count. */ + rack_ok = (int64_t)(frcti->t_latest_ack - slot->time) > (int64_t) R; + if (!rack_ok && frcti->dup_thresh < DUP_THRESH) + return; + + /* HoL aged past t_r; let rxm_due tear the flow down. */ + if (RXM_AGED_OUT(rxm->t0, now_ns, frcti->t_r)) + return; + + /* Already on it. */ + if (FAST_RXM_STAGED(pending) || RXM_FAST_DONE(slot->flags)) + return; + + recovery_enter(frcti); + + pending->fast_rxm.data = malloc(rxm->len); + if (pending->fast_rxm.data == NULL) + return; + + pending->fast_rxm.len = rxm->len; + memcpy(pending->fast_rxm.data, rxm->pkt, rxm->len); + slot->flags |= SND_RTX | SND_FAST_RXM; + frcti->rtt_lwe = frcti->snd_cr.lwe + 1; + if (rack_ok) + STAT_BUMP(frcti, rxm_rack); + else + STAT_BUMP(frcti, rxm_dupthresh); +} + +/* Caller holds wrlock; RACK fast retransmit queued in pending. */ +__attribute__((hot)) +static void frcti_ack_rcv(struct frcti * frcti, + const struct frct_pci * pci, + uint16_t flags, + uint64_t now_ns, + struct pending * pending) +{ + uint32_t ackno; + uint32_t lwe; + size_t p; + size_t fresh; + + if (!(flags & FRCT_DATA)) + STAT_BUMP(frcti, ack_rcv); + + ackno = ntoh32(pci->ackno); + if (ackno == frcti->snd_cr.lwe) { + /* RFC 8985 §6.2: only on scoreboard change. */ + if (frcti->snd_cr.lwe != frcti->rack_fired_lwe) { + fast_rxm_consider(frcti, now_ns, pending); + frcti->rack_fired_lwe = frcti->snd_cr.lwe; + } + return; + } + + if (!within(ackno, frcti->snd_cr.lwe, frcti->snd_cr.seqno)) + return; + + lwe = frcti->snd_cr.lwe; + p = RQ_SLOT(lwe); + + STORE_RELEASE(&frcti->snd_cr.lwe, ackno); + + /* §7.3: cum-ACK past the probed seqno resolves the TLP. */ + if (frcti->tlp_high_seq != 0 + && !before(ackno, frcti->tlp_high_seq)) + frcti->tlp_high_seq = 0; + + /* §7.3: end the probe episode once inflight drains. */ + if (ackno == frcti->snd_cr.seqno) + frcti->tlp_count = 0; + + /* RFC 8985 §7.2: halve mult per REO_DECAY_PKTS fresh-ACK'd seqnos. */ + fresh = ackno - frcti->dsack_lwe_snap; + if (frcti->reo_wnd_mult > 1 && fresh >= REO_DECAY_PKTS) { + uint8_t half = frcti->reo_wnd_mult >> 1; + frcti->reo_wnd_mult = half < 1 ? 1 : half; + frcti->dsack_lwe_snap = ackno; + } + + /* RFC 8985: latest cum-ACKed send-time (slot of ackno-1). */ + frcti->t_latest_ack = frcti->snd_slots[RQ_SLOT(ackno - 1)].time; + + /* RFC 8985: SACK-above-lwe count is per-recovery-episode. */ + frcti->dup_thresh = 0; + + /* Karn-skip on retx; TLP ACK clears rto_mul (no CC backoff). */ + if ((frcti->snd_slots[p].flags & SND_RTX) == 0 + || (frcti->snd_slots[p].flags & SND_TLP) != 0) + STORE_RELEASE(&frcti->rto_mul, 0); + + if (recovery_exit_reached(frcti, ackno)) + frcti->in_recovery = false; + + if (rtt_sample_eligible(frcti, p, flags, lwe)) { + int64_t mrtt = ts_age_ns(now_ns, frcti->snd_slots[p].time); + if (mrtt > 0) { + if (!(flags & FRCT_DATA)) + STAT_BUMP(frcti, ack_rtt); + rtt_update(frcti, (time_t) mrtt, now_ns); + frcti->t_rcv_rtt = now_ns; + } + } +} + +/* Skip k == lwe under clamp: NULLing HoL from a stale SACK wedges it. */ +static uint32_t sack_mark_blocks(struct frcti * frcti, + const uint8_t * payload, + uint16_t n, + uint32_t * newly_marked) +{ + uint32_t hi_sacked = frcti->snd_cr.lwe; + uint32_t marked = 0; + uint16_t i; + + for (i = 0; i < n; ++i) { + uint32_t s; + uint32_t e; + uint32_t k; + bool clamped; + + sack_block_get(payload, i, &s, &e); + + if (!before(s, e)) + continue; + + clamped = before(s, frcti->snd_cr.lwe); + if (clamped) + s = frcti->snd_cr.lwe; + if (after(e, frcti->snd_cr.seqno)) + e = frcti->snd_cr.seqno; + + for (k = s; before(k, e); ++k) { + size_t kp = RQ_SLOT(k); + uint64_t t_k; + if (clamped && k == frcti->snd_cr.lwe) + continue; + if (LOAD_ACQUIRE(&frcti->snd_slots[kp].rxm) == NULL) + continue; + STORE_RELEASE(&frcti->snd_slots[kp].rxm, NULL); + frcti->snd_slots[kp].flags = 0; + marked++; + /* RACK.fack: latest SACK-confirmed send-time. */ + t_k = frcti->snd_slots[kp].time; + if (t_k > frcti->t_latest_ack) + frcti->t_latest_ack = t_k; + } + + if (after(e, hi_sacked)) + hi_sacked = e; + } + + *newly_marked = marked; + return hi_sacked; +} + +/* Queue once per loss event (SND_FAST_RXM gates). Emit after unlock. */ +static void sack_queue_rxm(struct frcti * frcti, + uint32_t hi_sacked, + uint64_t now_ns, + struct pending * pending) +{ + uint64_t R = rack_reorder_window(frcti); + uint32_t k; + bool rack_ok; + + for (k = frcti->snd_cr.lwe; before(k, hi_sacked); ++k) { + struct rxm_entry * rxm; + size_t kp = RQ_SLOT(k); + size_t cnt = pending->sack_rxm_cnt; + size_t rack_age; + + rxm = LOAD_ACQUIRE(&frcti->snd_slots[kp].rxm); + + if (cnt >= SACK_RXM_MAX) + break; + + if (rxm == NULL) + continue; + + if (frcti->snd_slots[kp].flags & SND_FAST_RXM) + continue; + + if (RXM_AGED_OUT(rxm->t0, now_ns, frcti->t_r)) + continue; + + rack_age = frcti->t_latest_ack - frcti->snd_slots[kp].time; + /* RFC 8985 §6.2: time-based RACK OR DupThresh count. */ + rack_ok = (int64_t) rack_age > (int64_t) R; + if (!rack_ok && frcti->dup_thresh < DUP_THRESH) + continue; + + if (rack_ok) + STAT_BUMP(frcti, rxm_rack); + else + STAT_BUMP(frcti, rxm_dupthresh); + + pending->sack_rxm[cnt].data = malloc(rxm->len); + if (pending->sack_rxm[cnt].data == NULL) + break; + + pending->sack_rxm[cnt].len = rxm->len; + memcpy(pending->sack_rxm[cnt].data, rxm->pkt, rxm->len); + pending->sack_rxm_cnt++; + /* NULL slot so the original timer self-cleans. */ + STORE_RELEASE(&frcti->snd_slots[kp].rxm, NULL); + frcti->snd_slots[kp].time = now_ns; + frcti->snd_slots[kp].flags |= SND_RTX | SND_FAST_RXM; + frcti->rtt_lwe = k + 1; + } +} + +/* + * RFC 2883 D-SACK detector. Returns true iff block[0] is a D-SACK + * report: + * case 1: blocks[0].start < pkt_ackno (strictly below cum-ACK). + * case 2: blocks[0] is a strict sub-range of some blocks[i>0]. + * MAX_DSACK_LAG bounds case-1 distance to one rcv window (sanity). + */ +static bool sack_is_dsack(struct frcti * frcti, + const uint8_t * payload, + uint16_t n, + uint32_t pkt_ackno) +{ + uint32_t s0; + uint32_t e0; + uint16_t i; + + if (n == 0) + return false; + + sack_block_get(payload, 0, &s0, &e0); + if (!before(s0, e0)) + return false; + + if (before(s0, pkt_ackno)) { + if ((pkt_ackno - s0) <= (uint32_t) MAX_DSACK_LAG) + return true; + STAT_BUMP(frcti, dsack_drop); + return false; + } + + for (i = 1; i < n; ++i) { + uint32_t si; + uint32_t ei; + + sack_block_get(payload, i, &si, &ei); + if (!before(si, ei)) + continue; + if (!before(s0, si) && !after(e0, ei) + && (s0 != si || e0 != ei)) + return true; + } + + return false; +} + +/* RFC 8985 §7.2: grow reo_wnd_mult on DSACK; at most once per RTT. */ +static __inline__ void reo_wnd_on_dsack(struct frcti * frcti, + uint64_t now_ns) +{ + time_t srtt = frcti->srtt; + + /* Snap is unconditional: feeds the per-D-SACK decay clock. */ + frcti->dsack_lwe_snap = frcti->snd_cr.lwe; + + if (srtt > 0 + && now_ns - frcti->t_last_reo_widen <= (uint64_t) srtt) + return; + + if (frcti->reo_wnd_mult < REO_WND_MULT_MAX) + frcti->reo_wnd_mult++; + + frcti->t_last_reo_widen = now_ns; +} + +/* Caller holds wrlock; retransmits queued for post-unlock emission. */ +static void frcti_sack_rcv(struct frcti * frcti, + buffer_t pkt, + uint32_t pkt_ackno, + uint64_t now_ns, + struct pending * pending) +{ + uint32_t hi_sacked; + uint32_t marked; + uint16_t n; + bool dsack; + uint16_t n_real; + + if (pkt.len < SACK_HDR_SIZE) + return; + + n = ntoh16(*(const uint16_t *) pkt.data); + if (n > SACK_MAX_BLOCKS) + return; + + if (pkt.len < SACK_HDR_SIZE + (size_t) n * SACK_BLOCK_SIZE) + return; + + STAT_BUMP(frcti, sack_rcv); + + dsack = sack_is_dsack(frcti, pkt.data, n, pkt_ackno); + n_real = n - (dsack ? 1 : 0); + + if (dsack) { + STAT_BUMP(frcti, dsack_rcv); + reo_wnd_on_dsack(frcti, now_ns); + } + + /* DSACK-only carries no new gap; don't enter recovery. */ + if (n_real > 0) + recovery_enter(frcti); + + marked = 0; + hi_sacked = sack_mark_blocks(frcti, pkt.data, n, &marked); + frcti->dup_thresh += marked; + + if (after(hi_sacked, frcti->snd_cr.lwe)) + sack_queue_rxm(frcti, hi_sacked, now_ns, pending); +} + +/* Emit and free queued packet copies. */ +static void pending_flush(struct frcti * frcti, + struct pending * pending) +{ + size_t i; + + for (i = 0; i < pending->sack_rxm_cnt; ++i) { + sack_rxm_snd(frcti, pending->sack_rxm[i].data, + pending->sack_rxm[i].len); + free(pending->sack_rxm[i].data); + } + + if (pending->fast_rxm.data != NULL) { + int ret = fast_rxm_send(frcti, pending->fast_rxm.data, + pending->fast_rxm.len); + if (ret == -EFLOWDOWN || ret == -ENOTALLOC) + STAT_BUMP(frcti, rxm_tx_dead); + free(pending->fast_rxm.data); + } +} + +/* Pre-DRF NACK: ask peer to retransmit HoL; seqno is informational. */ +static void frcti_nack_snd(struct frcti * frcti, + uint32_t seqno_unseen) +{ + struct ssm_pk_buff * spb; + struct frct_pci * pci; + + if (frct_ctrl_alloc(&spb, &pci, 0) < 0) + return; + + pci->flags = hton16(FRCT_NACK); + pci->seqno = hton32(seqno_unseen); + + frct_hcs_set(pci, false); + + frct_tx(frcti, spb); +} + +enum frct_act { + FRCT_ACTIVE, + FRCT_INACT_NEED_NACK, + FRCT_INACT_DROP, +}; + +/* On rcv inactivity: rebase on DRF, or arm pre-DRF NACK. Caller wrlock. */ +static enum frct_act rcv_inact_check(struct frcti * frcti, + uint16_t flags, + uint32_t seqno, + uint64_t now_ns) +{ + struct frct_cr * rcv_cr = &frcti->rcv_cr; + uint64_t cd; + + if (!ts_aged_ns(now_ns, rcv_cr->act, rcv_cr->inact)) + return FRCT_ACTIVE; + + if (flags & FRCT_DRF) { + if (same_epoch_drf(seqno, flags, rcv_cr)) + return FRCT_ACTIVE; + + /* Bootstrap or fresh epoch: rebase. */ + STAT_BUMP(frcti, drf_rebase); + release_rq(frcti); + STORE_RELEASE(&rcv_cr->lwe, seqno); + rcv_cr->rwe = seqno + RQ_SIZE; + rcv_cr->seqno = seqno; + return FRCT_ACTIVE; + } + + if (!(flags & FRCT_DATA)) + return FRCT_ACTIVE; + + /* Pre-DRF: nudge sender with NACK (rate-limited). */ + cd = frcti->srtt > 0 ? (uint64_t) frcti->srtt : NACK_COOLDOWN_NS; + if (!ts_aged_ns(now_ns, frcti->t_nack, cd)) + return FRCT_INACT_DROP; + + frcti->t_nack = now_ns; + STAT_BUMP(frcti, nack_snd); - return wait; + return FRCT_INACT_NEED_NACK; } -static int __frcti_snd(struct frcti * frcti, - struct shm_du_buff * sdb) +/* Both modes: bounded accept into rq[seqno]. Caller wrlock. */ +__attribute__((hot)) +static bool rq_accept(struct frcti * frcti, + uint32_t seqno, + size_t pos, + uint16_t flags) { - struct frct_pci * pci; - struct timespec now; - struct frct_cr * snd_cr; - struct frct_cr * rcv_cr; - uint32_t seqno; - bool rtx; + struct frct_cr * rcv_cr = &frcti->rcv_cr; + + if (!before(seqno, rcv_cr->rwe)) { + STAT_BUMP(frcti, out_rcv); + return false; + } + + if (!before(seqno, rcv_cr->lwe + RQ_SIZE)) { + STAT_BUMP(frcti, rqo_rcv); + return false; + } + + if (frcti->rcv_slots[pos].idx != -1) { + if (flags & FRCT_RXM) + STAT_BUMP(frcti, rxm_dup_rcv); + else + STAT_BUMP(frcti, dup_rcv); + /* RFC 2883 §4 case 2: in-window dup; sub-range marker. */ + frcti->dsack_seqno = seqno; + frcti->dsack_valid = true; + return false; + } + + return true; +} + +/* OOO arrival; throttle by min_gap + scoreboard dedup. */ +static bool sack_check(struct frcti * frcti, + uint32_t seqno, + uint64_t now_ns, + struct sack_args * out) +{ + struct frct_cr * rcv_cr = &frcti->rcv_cr; + uint64_t min_gap; + uint16_t n; + + if (!after(seqno, rcv_cr->lwe)) + return false; + + STAT_BUMP(frcti, ooo_rcv); + + /* SACK carries cum-ACK; bound by t_a like any other ACK. */ + if (ACK_AGED_OUT(rcv_cr->act, now_ns, frcti->t_a)) + return false; + + /* srtt/8 gate starved recovery under burst loss; floor to save CPU. */ + min_gap = (uint64_t) SACK_MIN_GAP_NS; + + if (!ts_aged_ns(now_ns, frcti->t_snd_sack, min_gap)) + return false; + + out->dsack = false; + n = dsack_consume(frcti, out->blocks); + if (n == 1) + out->dsack = true; + n += sack_blocks_build(frcti, out->blocks + n, + frcti->sack_n_max - n); + + if (!out->dsack + && rcv_cr->lwe == frcti->sack_lwe && n == frcti->sack_n) + return false; + + out->n = n; + out->ack = rcv_cr->lwe; + out->rwe = frcti_advert_rwe(frcti); + frcti->t_snd_sack = now_ns; + frcti->sack_lwe = rcv_cr->lwe; + frcti->sack_n = n; + + return true; +} + +/* Wire-dup of fresh DATA at an already-ACKed seqno. */ +static __inline__ bool is_dup_data(uint16_t flags, + uint32_t seqno, + uint32_t lwe) +{ + if (!(flags & FRCT_DATA)) + return false; + + if (flags & FRCT_RXM) + return false; + + return before(seqno, lwe); +} + +/* + * Wire-dup ACK packet: same seqno as the previous emission. Updates + * the dedup ackno on a fresh ACK; caller drops on true. + */ +static __inline__ bool is_dup_ack(struct frcti * frcti, + uint16_t flags, + uint32_t seqno) +{ + if (flags & FRCT_DATA) + return false; + + if (!(flags & FRCT_ACK)) + return false; + + if (seqno == frcti->rcv_cr.ackno) + return true; + + frcti->rcv_cr.ackno = seqno; + + return false; +} + +/* Caller wrlock. */ +__attribute__((cold)) +static void seqno_rotate(struct frcti * frcti, + uint64_t now_ns) +{ + struct frct_cr * snd_cr = &frcti->snd_cr; + + if (!ts_aged_ns(now_ns, snd_cr->act, snd_cr->inact)) + return; + /* Idle-on-wire ≠ idle e2e: don't orphan in-flight rxm. */ + if (snd_cr->seqno != snd_cr->lwe) + return; + + /* Avoid colliding with peer's current rcv window. */ + do { + random_buffer(&snd_cr->seqno, sizeof(snd_cr->seqno)); + } while (in_window(snd_cr->seqno, snd_cr)); + STORE_RELEASE(&snd_cr->lwe, snd_cr->seqno); + STORE_RELAXED(&snd_cr->rwe, snd_cr->lwe + START_WINDOW); + frcti->rtt_lwe = snd_cr->seqno; + frcti->in_recovery = false; + frcti->recovery_high = snd_cr->seqno; +} + +__attribute__((hot)) +static int frcti_snd(struct frcti * frcti, + struct ssm_pk_buff * spb, + uint16_t flags) +{ + struct frct_pci * pci; + struct frct_pci_stream * spci = NULL; + struct timespec now; + struct frct_cr * snd_cr; + struct frct_cr * rcv_cr; + struct rxm_entry * rxm = NULL; + uint32_t seqno; + uint16_t pci_flags = 0; + bool rtx; + uint64_t now_ns; + int64_t rcv_idle; + uint32_t probe_id = 0; + uint8_t probe_nonce[RTTP_NONCE_LEN] = { 0 }; + bool probe; + size_t payload_len = 0; assert(frcti); - assert(shm_du_buff_len(sdb) != 0); + /* Stream mode permits 0-byte sends for the EOS marker. */ + assert(ssm_pk_buff_len(spb) != 0 || frcti->stream); snd_cr = &frcti->snd_cr; rcv_cr = &frcti->rcv_cr; - timerwheel_move(); + tw_move_safe(); + + if (frcti->stream) + payload_len = ssm_pk_buff_len(spb); - pci = (struct frct_pci *) shm_du_buff_head_alloc(sdb, FRCT_PCILEN); + pci = FRCT_HDR_PUSH(spb, frcti); if (pci == NULL) return -ENOMEM; - memset(pci, 0, sizeof(*pci)); + /* Pre-allocate rxm so alloc fail can't orphan a seqno. */ + if (snd_cr->cflags & FRCTFRTX) { + rxm = rxm_alloc(frcti, ssm_pk_buff_len(spb)); + if (rxm == NULL) { + ssm_pk_buff_pop(spb, frcti_data_hdr_len(frcti)); + return -ENOMEM; + } + } + + memset(pci, 0, FRCT_PCILEN); + + if (frcti->stream) + spci = FRCT_SPCI(pci); clock_gettime(PTHREAD_COND_CLOCK, &now); + now_ns = TS_TO_UINT64(now); pthread_rwlock_wrlock(&frcti->lock); rtx = snd_cr->cflags & FRCTFRTX; - pci->flags |= FRCT_DATA; + pci_flags |= FRCT_DATA; + if (!frcti->stream) + pci_flags |= (flags & FRCT_FR_MASK); - /* Set DRF if there are no unacknowledged packets. */ - if (snd_cr->seqno == snd_cr->lwe) - pci->flags |= FRCT_DRF; + if (!frcti->stream && (flags & FRCT_FR_MASK) != FRCT_FR_SOLE) + STAT_BUMP(frcti, frag_snd); - /* Choose a new sequence number if sender inactivity expired. */ - if (now.tv_sec - snd_cr->act.tv_sec > snd_cr->inact) { - /* There are no unacknowledged packets. */ - assert(snd_cr->seqno == snd_cr->lwe); - random_buffer(&snd_cr->seqno, sizeof(snd_cr->seqno)); - snd_cr->lwe = snd_cr->seqno; - snd_cr->rwe = snd_cr->lwe + START_WINDOW; + if (frcti->stream) { + if (flags & FRCT_FIN) + pci_flags |= FRCT_FIN; + + spci->start = hton32(frcti->snd_byte_next); + frcti->snd_byte_next += (uint32_t) payload_len; + spci->end = hton32(frcti->snd_byte_next); + STAT_ADD(frcti, strm_snd_byte, payload_len); } + if (snd_cr->seqno == snd_cr->lwe) + pci_flags |= FRCT_DRF; + + seqno_rotate(frcti, now_ns); + seqno = snd_cr->seqno; pci->seqno = hton32(seqno); - if (now.tv_sec - rcv_cr->act.tv_sec < rcv_cr->inact) { - pci->flags |= FRCT_FC; - *((uint32_t *) pci) |= hton32(rcv_cr->rwe & 0x00FFFFFF); + rcv_idle = ts_age_ns(now_ns, rcv_cr->act); + + if (rcv_idle < (int64_t) rcv_cr->inact) { + pci_flags |= FRCT_FC; + pci->window = hton32(frcti_advert_rwe(frcti)); } if (!rtx) { - snd_cr->lwe++; + STORE_RELEASE(&snd_cr->lwe, snd_cr->lwe + 1); + STORE_RELEASE(&snd_cr->rwe, snd_cr->lwe + RQ_SIZE); } else { - if (!frcti->probe) { - frcti->rttseq = snd_cr->seqno; - frcti->t_probe = now; - frcti->probe = true; -#ifdef PROC_FLOW_STATS - frcti->n_prb++; -#endif - } - if ((now.tv_sec - rcv_cr->act.tv_sec) * BILLION <= frcti->a) { - pci->flags |= FRCT_ACK; + size_t p = RQ_SLOT(seqno); + frcti->snd_slots[p].time = now_ns; + /* Fresh send clears RTX bits. */ + frcti->snd_slots[p].flags = 0; + if (rcv_idle <= (int64_t) frcti->t_a) { + pci_flags |= FRCT_ACK; pci->ackno = hton32(rcv_cr->lwe); rcv_cr->seqno = rcv_cr->lwe; } } + pci->flags = hton16(pci_flags); + + frct_hcs_set(pci, frcti->stream); + snd_cr->seqno++; - snd_cr->act = now; + STORE_RELEASE(&snd_cr->act, now_ns); + + probe = rtt_probe_arm(frcti, now_ns, &probe_id, probe_nonce); pthread_rwlock_unlock(&frcti->lock); - if (rtx) - timerwheel_rxm(frcti, seqno, sdb); + if (probe) + frcti_rttp_snd(frcti, probe_id, 0, probe_nonce); + + if (rtx) { + assert(rxm != NULL); + rxm_arm(frcti, seqno, rxm, spb); + tlp_arm(frcti); + } return 0; } -static void rtt_estimator(struct frcti * frcti, - time_t mrtt) +/* Stream FIN is armed for rxm; needs to be in window. */ +static __inline__ bool stream_fin_blocked(struct frcti * frcti) { - time_t srtt = frcti->srtt; - time_t rttvar = frcti->mdev; + if (!frcti->stream) + return false; - if (srtt == 0) { /* first measurement */ - srtt = mrtt; - rttvar = mrtt >> 1; - } else { - time_t delta = mrtt - srtt; - srtt += (delta >> 3); - delta = (ABS(delta) - rttvar) >> 2; -#ifdef FRCT_LINUX_RTT_ESTIMATOR - if (delta < 0) - delta >>= 3; -#endif - rttvar += delta; + return !before(frcti->snd_cr.seqno, frcti->snd_cr.lwe + RQ_SIZE); +} + +/* + * Stream: 0-byte FRCT_FIN DATA so peer's flow_read returns 0 at this + * byte. Msg: control packet with FRCT_FIN flag, snd_cr.seqno carried + * in pci->ackno (sender packs via frcti_pkt_snd's ackno parameter). + */ +static void frcti_fin_snd(struct frcti * frcti) +{ + struct ssm_pk_buff * spb; + bool already; + uint32_t fin_seqno; + + if (!(frcti->snd_cr.cflags & FRCTFLINGER)) + return; + + pthread_rwlock_wrlock(&frcti->lock); + + already = frcti->snd_fin_sent; + + /* Defer before committing snd_fin_sent; linger loop retries. */ + if (!already && stream_fin_blocked(frcti)) { + pthread_rwlock_unlock(&frcti->lock); + return; } -#ifdef PROC_FLOW_STATS - frcti->n_rtt++; -#endif - frcti->srtt = MAX(1000L, srtt); - frcti->mdev = MAX(100L, rttvar); - frcti->rto = MAX(RTO_MIN, frcti->srtt + (frcti->mdev << MDEV_MUL)); -} - -/* Always queues the next application packet on the RQ. */ -static void __frcti_rcv(struct frcti * frcti, - struct shm_du_buff * sdb) -{ - ssize_t idx; - size_t pos; - struct frct_pci * pci; - struct timespec now; - struct frct_cr * rcv_cr; - struct frct_cr * snd_cr; - uint32_t seqno; - uint32_t ackno; - uint32_t rwe; - int fd = -1; - assert(frcti); + frcti->snd_fin_sent = true; + fin_seqno = frcti->snd_cr.seqno; + + if (!already && !frcti->stream) + frcti->snd_fin_seqno = fin_seqno; + + pthread_rwlock_unlock(&frcti->lock); + + if (already) + return; + + if (!frcti->stream) { + frcti_pkt_snd(frcti, FRCT_FIN, fin_seqno, 0); + return; + } + + if (frct_spb_reserve(frcti_data_hdr_len(frcti), &spb) < 0) + return; + + /* Reset spb to 0-len so frcti_snd's head_alloc populates PCI. */ + ssm_pk_buff_truncate(spb, 0); + + if (frcti_snd(frcti, spb, FRCT_FIN) < 0) { + frct_spb_release(spb); + return; + } + + if (frct_tx(frcti, spb) < 0) + return; + + pthread_rwlock_wrlock(&frcti->lock); + + frcti->snd_fin_seqno = frcti->snd_cr.seqno - 1; + + pthread_rwlock_unlock(&frcti->lock); +} + +static bool final_ack_due(struct frcti * frcti, + struct frct_cr * rcv_cr, + uint64_t now_ns) +{ + if (rcv_cr->lwe == rcv_cr->seqno) + return false; + + if (ACK_AGED_OUT(rcv_cr->act, now_ns, frcti->t_a)) + return false; + + return true; +} + +/* Snd-side has FLINGER cflag and unACK'd data below the FIN/seqno. */ +static __inline__ bool snd_drain_pending(struct frct_cr * snd_cr, + uint32_t edge) +{ + if (!(snd_cr->cflags & FRCTFLINGER)) + return false; + + return before(snd_cr->lwe, edge); +} + +/* Peer is still active and we haven't seen their FIN yet. */ +static __inline__ bool rcv_drain_pending(struct frcti * frcti, + struct frct_cr * rcv_cr, + uint64_t now_ns) +{ + if (frcti->rcv_fin_seen) + return false; + + return !ts_aged_ns(now_ns, rcv_cr->act, rcv_cr->inact); +} + +/* Drain-loop predicate: snd-side unACK'd data OR peer still active. */ +static bool frcti_lingering(struct frcti * frcti) +{ + struct timespec now; + struct frct_cr * snd_cr; + struct frct_cr * rcv_cr; + uint32_t edge; + uint64_t now_ns; + bool snd_linger; + bool rcv_linger; + + /* Idempotent; emits FIN once per side, both stream and msg. */ + frcti_fin_snd(frcti); + + clock_gettime(PTHREAD_COND_CLOCK, &now); + now_ns = TS_TO_UINT64(now); + + pthread_rwlock_rdlock(&frcti->lock); + snd_cr = &frcti->snd_cr; rcv_cr = &frcti->rcv_cr; + + if (frcti->snd_fin_sent) + edge = frcti->snd_fin_seqno; + else + edge = snd_cr->seqno; + + snd_linger = snd_drain_pending(snd_cr, edge); + rcv_linger = rcv_drain_pending(frcti, rcv_cr, now_ns); + + pthread_rwlock_unlock(&frcti->lock); + + return snd_linger || rcv_linger; +} + +static time_t frcti_dealloc(struct frcti * frcti) +{ + struct timespec now; + struct frct_cr * snd_cr; + struct frct_cr * rcv_cr; + int ackno; + bool due; + int64_t now_ns; + int64_t rcv; + int64_t snd; + snd_cr = &frcti->snd_cr; + rcv_cr = &frcti->rcv_cr; + + /* Idempotent; usually already sent by frcti_lingering. */ + frcti_fin_snd(frcti); clock_gettime(PTHREAD_COND_CLOCK, &now); + now_ns = TS_TO_UINT64(now); - pci = (struct frct_pci *) shm_du_buff_head_release(sdb, FRCT_PCILEN); + pthread_rwlock_rdlock(&frcti->lock); - idx = shm_du_buff_get_idx(sdb); - seqno = ntoh32(pci->seqno); - pos = seqno & (RQ_SIZE - 1); + ackno = rcv_cr->lwe; + rcv = (int64_t)(rcv_cr->act + rcv_cr->inact) - now_ns; + snd = (int64_t)(snd_cr->act + snd_cr->inact) - now_ns; + due = final_ack_due(frcti, rcv_cr, now_ns); - pthread_rwlock_wrlock(&frcti->lock); + pthread_rwlock_unlock(&frcti->lock); - if (now.tv_sec - rcv_cr->act.tv_sec > rcv_cr->inact) { - if (pci->flags & FRCT_DRF) { /* New run. */ - rcv_cr->lwe = seqno; - rcv_cr->rwe = seqno + RQ_SIZE; - rcv_cr->seqno = seqno; - } else if (pci->flags & FRCT_DATA) { - goto drop_packet; - } - } + if (due) + frcti_pkt_snd(frcti, FRCT_ACK, ackno, 0); - rcv_cr->act = now; + return (time_t) MAX((MAX(rcv, snd) / BILLION), 0); +} - /* For now, just send an immediate window update. */ - if (pci->flags & FRCT_RDVS) { - fd = frcti->fd; - rwe = rcv_cr->rwe; - pthread_rwlock_unlock(&frcti->lock); +__attribute__((hot)) +static void frcti_rcv(struct frcti * frcti, + struct ssm_pk_buff * spb) +{ + ssize_t idx; + size_t pos; + struct frct_pci * pci; + struct timespec now; + uint64_t now_ns; + struct frct_cr * rcv_cr; + uint32_t seqno; + uint16_t flags; + buffer_t pkt; + struct pending pending = { 0 }; + bool in_order; + struct sack_args * sa = NULL; + bool send_sack = false; - __send_frct_pkt(fd, FRCT_FC, 0, rwe); + assert(frcti); - shm_rdrbuff_remove(ai.rdrb, idx); + rcv_cr = &frcti->rcv_cr; + + clock_gettime(PTHREAD_COND_CLOCK, &now); + now_ns = TS_TO_UINT64(now); + + if (ssm_pk_buff_len(spb) < FRCT_PCILEN) { + frct_spb_release(spb); return; } - if (pci->flags & FRCT_ACK) { - ackno = ntoh32(pci->ackno); - if (after(ackno, frcti->snd_cr.lwe)) - frcti->snd_cr.lwe = ackno; + pci = FRCT_HDR_POP(spb, frct_pci); - if (frcti->probe && after(ackno, frcti->rttseq)) { -#ifdef PROC_FLOW_STATS - if (!(pci->flags & FRCT_DATA)) - frcti->n_dak++; -#endif - rtt_estimator(frcti, ts_diff_ns(&frcti->t_probe, &now)); - frcti->probe = false; - } + idx = ssm_pk_buff_get_off(spb); + seqno = ntoh32(pci->seqno); + pos = RQ_SLOT(seqno); + + flags = ntoh16(pci->flags); + + pkt.data = ssm_pk_buff_head(spb); + pkt.len = ssm_pk_buff_len(spb); + + if (flags & FRCT_RXM) + STAT_BUMP(frcti, rxm_rcv); + + /* Stateless / lock-free dispatches. spb released via ctrl_done. */ + if (flags & FRCT_KA) { + frcti_ka_rcv(frcti, pci, now_ns, flags); + goto ctrl_done; } - if (pci->flags & FRCT_FC) { - uint32_t rwe; + if (flags & FRCT_RTTP) { + frcti_rttp_rcv(frcti, pkt, now_ns); + goto ctrl_done; + } + + if (flags & FRCT_NACK) { + frcti_nack_rcv(frcti); + goto ctrl_done; + } - rwe = ntoh32(*((uint32_t *)pci) & hton32(0x00FFFFFF)); - rwe |= snd_cr->rwe & 0xFF000000; + if (flags & FRCT_RDVS) { + frcti_rdv_rcv(frcti); + goto ctrl_done; + } - /* Rollover for 24 bit */ - if (before(rwe, snd_cr->rwe) && snd_cr->rwe - rwe > 0x007FFFFF) - rwe += 0x01000000; + /* Msg-mode FIN: control packet, FIN seqno carried in pci->ackno. */ + if ((flags & FRCT_FIN) && !(flags & FRCT_DATA)) { + pthread_rwlock_wrlock(&frcti->lock); + if (!frcti->rcv_fin_seen) { + frcti->rcv_fin_seen = true; + frcti->rcv_byte_fin = ntoh32(pci->ackno); + } + pthread_rwlock_unlock(&frcti->lock); + goto ctrl_done; + } - snd_cr->rwe = rwe; + pthread_rwlock_wrlock(&frcti->lock); - pthread_mutex_lock(&frcti->mtx); - if (!frcti->open) { - frcti->open = true; - pthread_cond_broadcast(&frcti->cond); + /* rcv_inact_check is a no-op for non-DATA non-DRF packets. */ + if (flags & (FRCT_DATA | FRCT_DRF)) { + switch (rcv_inact_check(frcti, flags, seqno, now_ns)) { + case FRCT_INACT_NEED_NACK: + pthread_rwlock_unlock(&frcti->lock); + frcti_nack_snd(frcti, seqno - 1); + frct_spb_release(spb); + return; + case FRCT_INACT_DROP: + STAT_BUMP(frcti, inact_drop); + goto drop_packet; + case FRCT_ACTIVE: + /* FALLTHRU */ + default: + break; } - pthread_mutex_unlock(&frcti->mtx); } - if (!(pci->flags & FRCT_DATA)) + /* DATA-only act refresh: non-DATA would lock out DRF rebase. */ + if (flags & FRCT_DATA) + STORE_RELEASE(&rcv_cr->act, now_ns); + + /* Wire-dup ACK packet: same seqno as the previous emission. */ + if (is_dup_ack(frcti, flags, seqno)) { + STAT_BUMP(frcti, ack_dup_rcv); + goto drop_packet; + } + + /* Wire-dup of DATA: piggybacked ACK info already processed. */ + if (is_dup_data(flags, seqno, rcv_cr->lwe)) { + rcv_cr->seqno = seqno; + STAT_BUMP(frcti, dup_rcv); + /* RFC 2883 §4 case 1: dup below cum-ACK. */ + frcti->dsack_seqno = seqno; + frcti->dsack_valid = true; + goto drop_packet; + } + + if (flags & FRCT_ACK) + frcti_ack_rcv(frcti, pci, flags, now_ns, &pending); + + if (flags & FRCT_SACK) + frcti_sack_rcv(frcti, pkt, ntoh32(pci->ackno), + now_ns, &pending); + + if (flags & FRCT_FC) + frcti_fc_rcv(frcti, pci); + + if (!(flags & FRCT_DATA)) goto drop_packet; if (before(seqno, rcv_cr->lwe)) { - rcv_cr->seqno = seqno; /* Ensures we send a new ACK. */ -#ifdef PROC_FLOW_STATS - frcti->n_dup++; -#endif + /* Bump rcv_cr.seqno to force ack_snd to fire on the dup. */ + rcv_cr->seqno = seqno; + if (flags & FRCT_RXM) + STAT_BUMP(frcti, rxm_dup_rcv); + else + STAT_BUMP(frcti, dup_rcv); + /* RFC 2883 §4 case 1: dup below cum-ACK. */ + frcti->dsack_seqno = seqno; + frcti->dsack_valid = true; goto drop_packet; } - if (rcv_cr->cflags & FRCTFRTX) { + if (!rq_accept(frcti, seqno, pos, flags)) + goto drop_packet; - if (!before(seqno, rcv_cr->rwe)) { /* Out of window. */ -#ifdef PROC_FLOW_STATS - frcti->n_out++; -#endif + if (frcti->stream) { + if (frcti_stream_data_rcv(frcti, spb, pos, flags) < 0) { + STAT_BUMP(frcti, strm_drop); goto drop_packet; } - - if (!before(seqno, rcv_cr->lwe + RQ_SIZE)) { -#ifdef PROC_FLOW_STATS - frcti->n_rqo++; -#endif - goto drop_packet; /* Out of rq. */ - } - if (frcti->rq[pos] != -1) { -#ifdef PROC_FLOW_STATS - frcti->n_dup++; -#endif - goto drop_packet; /* Duplicate in rq. */ - } - fd = frcti->fd; + /* spb consumed by stash; do not release in drop path. */ + spb = NULL; } else { - rcv_cr->lwe = seqno; + frcti_data_stash(frcti, idx, pos, flags); } - frcti->rq[pos] = idx; + /* Lazy alloc: only OOO arrivals can trigger a SACK send. */ + if (after(seqno, rcv_cr->lwe) && frcti->sack_n_max > 0) { + size_t sa_sz = sizeof(*sa) + + frcti->sack_n_max * sizeof(sa->blocks[0]); + sa = malloc(sa_sz); + /* If alloc fails, sack_check sees NULL and we skip SACK. */ + } + + send_sack = sa != NULL && sack_check(frcti, seqno, now_ns, sa); + in_order = !after(seqno, rcv_cr->lwe); pthread_rwlock_unlock(&frcti->lock); - if (fd != -1) - timerwheel_delayed_ack(fd, frcti); + if (send_sack) { + STAT_BUMP(frcti, sack_snd); + if (sa->dsack) + STAT_BUMP(frcti, dsack_snd); + frcti_sack_snd(frcti, sa); + } else if (in_order) { + ack_arm(frcti); + } + + if ((flags & FRCT_ACK) && frcti->snd_cr.seqno != frcti->snd_cr.lwe) + tlp_arm(frcti); + + pending_flush(frcti, &pending); + frcti_rcv_probe(frcti, now_ns); + + free(sa); + return; + + ctrl_done: + frct_spb_release(spb); return; drop_packet: pthread_rwlock_unlock(&frcti->lock); - shm_rdrbuff_remove(ai.rdrb, idx); - send_frct_pkt(frcti); - return; + frct_spb_release(spb); + /* with_sack=true: ack_snd no-ops if neither dsack nor SACK is due. */ + ack_snd(frcti, true); + + pending_flush(frcti, &pending); + free(sa); } + +/* NULL-shim macros for the no-FRCT case. */ + +#define FRCTI_SND(frcti, spb, flags) \ + ((frcti) == NULL ? 0 : frcti_snd((frcti), (spb), (flags))) + +#define FRCTI_RCV(frcti, spb) \ + do { \ + if ((frcti) != NULL) \ + frcti_rcv((frcti), (spb)); \ + } while (0) + +#define FRCTI_PDU_READY(frcti) \ + ((frcti) != NULL && frcti_pdu_ready(frcti)) + +#define FRCTI_CONSUME(frcti, buf, count) \ + ((frcti) == NULL ? (ssize_t) -EAGAIN \ + : (frcti)->stream \ + ? frcti_consume_stream((frcti), (buf), (count)) \ + : frcti_consume((frcti), (buf), (count))) + +#define FRCTI_IS_FRTX(frcti) \ + ((frcti) != NULL && ((frcti)->rcv_cr.cflags & FRCTFRTX)) + +#define FRCTI_IS_STREAM(frcti) ((frcti) != NULL && (frcti)->stream) + +#define FRCTI_PAYLOAD_CAP(frcti) \ + ((frcti)->frag_mtu - frcti_data_hdr_len(frcti)) + +#define FRCTI_NEEDS_FRAG(frcti, count) \ + ((frcti) != NULL && (count) > FRCTI_PAYLOAD_CAP(frcti)) + +#define FRCTI_IS_WINDOW_OPEN(frcti) \ + ((frcti) == NULL ? true : frcti_is_window_open(frcti)) + +#define FRCTI_IS_WINDOW_OPEN_N(frcti, n) \ + ((frcti) == NULL ? true : frcti_is_window_open_n((frcti), (n))) + +#define FRCTI_LINGERING(frcti) \ + ((frcti) == NULL ? false : frcti_lingering(frcti)) + +#define FRCTI_DEALLOC(frcti) \ + ((frcti) == NULL ? (time_t) 0 : frcti_dealloc(frcti)) + diff --git a/src/lib/hash.c b/src/lib/hash.c index b465f894..903474df 100644 --- a/src/lib/hash.c +++ b/src/lib/hash.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Hashing * @@ -39,6 +39,9 @@ #include <ouroboros/md5.h> #include <ouroboros/sha3.h> #endif +#include <ouroboros/crc8.h> +#include <ouroboros/crc16.h> +#include <ouroboros/crc64.h> #include <string.h> #include <assert.h> #include <stdbool.h> @@ -69,6 +72,14 @@ int hash_len_tbl [] = { uint16_t hash_len(enum hash_algo algo) { + if (algo == HASH_CRC8) + return CRC8_HASH_LEN; + + if (algo == HASH_CRC16) + return CRC16_HASH_LEN; + + if (algo == HASH_CRC64) + return CRC64_HASH_LEN; #ifdef HAVE_LIBGCRYPT return (uint16_t) gcry_md_get_algo_dlen(gcry_algo_tbl[algo]); #else @@ -81,12 +92,36 @@ void mem_hash(enum hash_algo algo, const uint8_t * buf, size_t len) { -#ifdef HAVE_LIBGCRYPT - gcry_md_hash_buffer(gcry_algo_tbl[algo], dst, buf, len); -#else +#ifndef HAVE_LIBGCRYPT struct sha3_ctx sha3_ctx; struct md5_ctx md5_ctx; +#endif + if (algo == HASH_CRC8) { + uint8_t crc = 0; + + crc8_autosar(&crc, buf, len); + *(uint8_t *) dst = crc; + return; + } + if (algo == HASH_CRC16) { + uint16_t crc = 0; + + crc16_ccitt_false(&crc, buf, len); + *(uint16_t *) dst = htobe16(crc); + return; + } + + if (algo == HASH_CRC64) { + uint64_t crc = 0; + + crc64_nvme(&crc, buf, len); + *(uint64_t *) dst = htobe64(crc); + return; + } +#ifdef HAVE_LIBGCRYPT + gcry_md_hash_buffer(gcry_algo_tbl[algo], dst, buf, len); +#else switch (algo) { case HASH_CRC32: memset(dst, 0, CRC32_HASH_LEN); @@ -131,3 +166,14 @@ void str_hash(enum hash_algo algo, { return mem_hash(algo, dst, (const uint8_t *) str, strlen(str)); } + +uint64_t hash_mix64(uint64_t key) +{ + key ^= key >> 33; + key *= 0xff51afd7ed558ccdULL; + key ^= key >> 33; + key *= 0xc4ceb9fe1a85ec53ULL; + key ^= key >> 33; + + return key; +} diff --git a/src/lib/irm.c b/src/lib/irm.c index d25101f3..c62701aa 100644 --- a/src/lib/irm.c +++ b/src/lib/irm.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * The API to instruct the IRM * @@ -523,32 +523,23 @@ int irm_unbind_process(pid_t pid, return ret; } -int irm_create_name(const char * name, - enum pol_balance pol) +int irm_create_name(struct name_info * info) { irm_msg_t msg = IRM_MSG__INIT; - name_info_msg_t ni_msg = NAME_INFO_MSG__INIT; irm_msg_t * recv_msg; int ret; - if (name == NULL) + if (info == NULL) return -EINVAL; - msg.code = IRM_MSG_CODE__IRM_CREATE_NAME; - ni_msg.name = (char *) name; - ni_msg.pol_lb = pol; - msg.n_names = 1; - - msg.names = malloc(sizeof(*msg.names)); - if (msg.names == NULL) { - return -ENOMEM; - } - - msg.names[0] = &ni_msg; + msg.code = IRM_MSG_CODE__IRM_CREATE_NAME; + msg.name_info = name_info_s_to_msg(info); + if (msg.name_info == NULL) + goto fail_info_msg; recv_msg = send_recv_irm_msg(&msg); - free(msg.names); + name_info_msg__free_unpacked(msg.name_info, NULL); if (recv_msg == NULL) return -EIRMD; @@ -562,6 +553,9 @@ int irm_create_name(const char * name, irm_msg__free_unpacked(recv_msg, NULL); return ret; + + fail_info_msg: + return -ENOMEM; } int irm_destroy_name(const char * name) @@ -620,7 +614,7 @@ ssize_t irm_list_names(struct name_info ** names) return 0; } - *names = malloc(nr * sizeof(**names)); + *names = calloc(nr, sizeof(**names)); if (*names == NULL) { irm_msg__free_unpacked(recv_msg, NULL); return -ENOMEM; diff --git a/src/lib/list.c b/src/lib/list.c deleted file mode 100644 index 62b2eb27..00000000 --- a/src/lib/list.c +++ /dev/null @@ -1,78 +0,0 @@ -/* - * Ouroboros - Copyright (C) 2016 - 2024 - * - * Simple doubly linked list implementation. - * - * Dimitri Staessens <dimitri@ouroboros.rocks> - * Sander Vrijders <sander@ouroboros.rocks> - * - * This library is free software; you can redistribute it and/or - * modify it under the terms of the GNU Lesser General Public License - * version 2.1 as published by the Free Software Foundation. - * - * This library is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * Lesser General Public License for more details. - * - * You should have received a copy of the GNU Lesser General Public - * License along with this library; if not, write to the Free Software - * Foundation, Inc., http://www.fsf.org/about/contact/. - */ - -#include <ouroboros/list.h> - -#include <stddef.h> - -void list_head_init(struct list_head * h) -{ - h->nxt = h; - h->prv = h; -} - -static void add_list(struct list_head * n, - struct list_head * prv, - struct list_head * nxt) -{ - nxt->prv = n; - n->nxt = nxt; - n->prv = prv; - prv->nxt = n; -} - -static void del_list(struct list_head * prv, - struct list_head * nxt) -{ - nxt->prv = prv; - prv->nxt = nxt; -} - -void list_add(struct list_head * n, - struct list_head * h) -{ - add_list(n, h, h->nxt); -} - -void list_add_tail(struct list_head * n, - struct list_head * h) -{ - add_list(n, h->prv, h); -} - -void list_del(struct list_head * e) -{ - del_list(e->prv, e->nxt); - e->nxt = e->prv = e; -} - -bool list_is_empty(const struct list_head * h) -{ - return h->nxt == h; -} - -void list_move(struct list_head * n, - struct list_head * h) -{ - del_list(n->prv, n->nxt); - add_list(n, h, h->nxt); -} diff --git a/src/lib/lockfile.c b/src/lib/lockfile.c index cf6d3c94..0c18dfc1 100644 --- a/src/lib/lockfile.c +++ b/src/lib/lockfile.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Lockfile for Ouroboros * diff --git a/src/lib/logs.c b/src/lib/logs.c index d90bcd63..af03de10 100644 --- a/src/lib/logs.c +++ b/src/lib/logs.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Logging facilities * diff --git a/src/lib/md5.c b/src/lib/md5.c index ad0dd4d7..2412b909 100644 --- a/src/lib/md5.c +++ b/src/lib/md5.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * MD5 algorithm * diff --git a/src/lib/notifier.c b/src/lib/notifier.c index 45745b9a..10bdb28d 100644 --- a/src/lib/notifier.c +++ b/src/lib/notifier.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Notifier event system using callbacks * @@ -95,18 +95,14 @@ int notifier_reg(notifier_fn_t callback, pthread_rwlock_wrlock(¬ifier.lock); list_for_each(p, ¬ifier.listeners) { - struct listener * l = list_entry(p, struct listener, next); - if (l->callback == callback) { - pthread_rwlock_unlock(¬ifier.lock); - return -EPERM; - } + l = list_entry(p, struct listener, next); + if (l->callback == callback) + goto fail; } l = malloc(sizeof(*l)); - if (l == NULL) { - pthread_rwlock_unlock(¬ifier.lock); - return -ENOMEM; - } + if (l == NULL) + goto fail; l->callback = callback; l->obj = obj; @@ -116,6 +112,10 @@ int notifier_reg(notifier_fn_t callback, pthread_rwlock_unlock(¬ifier.lock); return 0; + fail: + pthread_rwlock_unlock(¬ifier.lock); + return -1; + } void notifier_unreg(notifier_fn_t callback) diff --git a/src/lib/pb/cep.proto b/src/lib/pb/cep.proto index d31cf4f7..14a85cfa 100644 --- a/src/lib/pb/cep.proto +++ b/src/lib/pb/cep.proto @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Message for Connection Information in OCEP * diff --git a/src/lib/pb/enroll.proto b/src/lib/pb/enroll.proto index 7fe612a8..37226b4d 100644 --- a/src/lib/pb/enroll.proto +++ b/src/lib/pb/enroll.proto @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Enrollment protocol * @@ -24,7 +24,6 @@ syntax = "proto2"; import "ipcp_config.proto"; message enroll_req_msg { - /* TODO authentication */ required bytes id = 1; } diff --git a/src/lib/pb/ipcp.proto b/src/lib/pb/ipcp.proto index c2c7f48b..afee4f91 100644 --- a/src/lib/pb/ipcp.proto +++ b/src/lib/pb/ipcp.proto @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Messages sent to IPCPds * @@ -39,6 +39,7 @@ enum ipcp_msg_code { IPCP_CONNECT = 10; IPCP_DISCONNECT = 11; IPCP_REPLY = 12; + IPCP_FLOW_UPDATE = 13; } message ipcp_msg { @@ -54,6 +55,7 @@ message ipcp_msg { optional int32 response = 10; optional string comp = 11; optional uint32 timeo_sec = 12; - optional sint32 mpl = 13; + optional sint32 mpl = 13; /* MPL in ms. */ optional int32 result = 14; + optional uint32 uid = 15; /* 0 = GSPP, >0 = PUP uid */ } diff --git a/src/lib/pb/ipcp_config.proto b/src/lib/pb/ipcp_config.proto index 28528b0c..eac4da37 100644 --- a/src/lib/pb/ipcp_config.proto +++ b/src/lib/pb/ipcp_config.proto @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Layer configuration message * @@ -24,17 +24,45 @@ syntax = "proto2"; import "model.proto"; + +message ls_config_msg { + required uint32 pol = 1; + required uint32 t_recalc = 2; + required uint32 t_update = 3; + required uint32 t_timeo = 4; +} + +message routing_config_msg { + required uint32 pol = 1; + optional ls_config_msg ls = 2; +} + message dt_config_msg { - required uint32 addr_size = 1; - required uint32 eid_size = 2; - required uint32 max_ttl = 3; - required uint32 routing_type = 4; + required uint32 addr_size = 1; + required uint32 eid_size = 2; + required uint32 max_ttl = 3; + required routing_config_msg routing = 4; +} + +message dir_dht_config_msg { + required uint32 alpha = 1; + required uint32 k = 2; + required uint32 t_expire = 3; + required uint32 t_refresh = 4; + required uint32 t_replicate = 5; + required uint64 peer = 6; +} + +message dir_config_msg { + required uint32 pol = 1; + optional dir_dht_config_msg dht = 2; } message uni_config_msg { required dt_config_msg dt = 1; - required uint32 addr_auth_type = 2; - required uint32 cong_avoid = 3; + required dir_config_msg dir = 2; + required uint32 addr_auth_type = 3; + required uint32 cong_avoid = 4; } message eth_config_msg { @@ -42,16 +70,24 @@ message eth_config_msg { required uint32 ethertype = 2; } -message udp_config_msg { +message udp4_config_msg { required uint32 ip_addr = 1; required uint32 port = 2; required uint32 dns_addr = 3; /* set to 0 if unused */ } +message udp6_config_msg { + required bytes ip_addr = 1; + required uint32 port = 2; + required bytes dns_addr = 3; /* set to NULL if unused */ +} + + message ipcp_config_msg { required layer_info_msg layer_info = 1; required uint32 ipcp_type = 2; optional uni_config_msg unicast = 3; - optional udp_config_msg udp = 4; - optional eth_config_msg eth = 5; + optional udp4_config_msg udp4 = 4; + optional udp6_config_msg udp6 = 5; + optional eth_config_msg eth = 6; } diff --git a/src/lib/pb/irm.proto b/src/lib/pb/irm.proto index da3bd982..f54bc9ea 100644 --- a/src/lib/pb/irm.proto +++ b/src/lib/pb/irm.proto @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Messages sent to IRMd * @@ -53,6 +53,8 @@ enum irm_msg_code { IPCP_FLOW_REQ_ARR = 25; IPCP_FLOW_ALLOC_REPLY = 26; IRM_REPLY = 27; + IRM_FLOW_UPDATE = 28; + IPCP_FLOW_UPDATE_ARR = 29; } message timespec_msg { @@ -75,23 +77,28 @@ message irm_msg { optional string name = 4; optional flow_info_msg flow_info = 5; optional ipcp_info_msg ipcp_info = 6; - optional string layer = 7; - repeated string exec = 8; - optional sint32 response = 9; - optional string dst = 10; - optional bytes hash = 11; - optional sint32 flow_id = 12; - optional qosspec_msg qosspec = 13; - optional ipcp_config_msg conf = 14; - optional uint32 opts = 15; - repeated ipcp_list_msg ipcps = 16; - repeated name_info_msg names = 17; - optional timespec_msg timeo = 18; - optional sint32 mpl = 20; + optional name_info_msg name_info = 7; + optional string layer = 8; + repeated string exec = 9; + optional sint32 response = 10; + optional string dst = 11; + optional bytes hash = 12; + optional sint32 flow_id = 13; + optional qosspec_msg qosspec = 14; + optional ipcp_config_msg conf = 15; + optional uint32 opts = 16; + repeated ipcp_list_msg ipcps = 17; + repeated name_info_msg names = 18; + optional timespec_msg timeo = 19; + optional sint32 mpl = 20; /* MPL in ms. */ optional string comp = 21; optional bytes pk = 22; /* piggyback */ - optional bytes symmkey = 23; - optional uint32 timeo_sec = 24; - optional uint32 timeo_nsec = 25; - optional sint32 result = 26; + optional uint32 timeo_sec = 23; + optional uint32 timeo_nsec = 24; + optional sint32 result = 25; + optional bytes sym_key = 26; /* symmetric encryption key */ + optional sint32 cipher_nid = 27; /* cipher NID */ + optional uint32 generation = 28; /* re-key batch generation */ + optional bool rekey = 29; /* re-key watermark trigger */ + optional bool rk_initiator = 30; /* re-key proof-holder side */ } diff --git a/src/lib/pb/model.proto b/src/lib/pb/model.proto index f1e401f9..4c1564a5 100644 --- a/src/lib/pb/model.proto +++ b/src/lib/pb/model.proto @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Model description messages * @@ -28,24 +28,29 @@ message qosspec_msg { required uint32 availability = 3; /* Class of 9s. */ required uint32 loss = 4; /* Packet loss. */ required uint32 ber = 5; /* Bit error rate, ppb. */ - required uint32 in_order = 6; /* In-order delivery. */ + required uint32 service = 6; /* enum qos_service. */ required uint32 max_gap = 7; /* In ms. */ - required uint32 cypher_s = 8; /* Crypto strength in bits. */ - required uint32 timeout = 9; /* Timeout in ms. */ + required uint32 timeout = 8; /* Timeout in ms. */ } message flow_info_msg { - required uint32 id = 1; - required uint32 n_pid = 2; - required uint32 n_1_pid = 3; - required uint32 mpl = 4; - required uint32 state = 5; - required qosspec_msg qos = 6; + required uint32 id = 1; + required uint32 n_pid = 2; + required uint32 n_1_pid = 3; + required uint32 mpl = 4; /* MPL in ms. */ + required uint32 state = 5; + required qosspec_msg qos = 6; + required uint32 uid = 7; + required uint32 mtu = 8; /* Layer MTU (bytes). */ } message name_info_msg { - required string name = 1; - required uint32 pol_lb = 2; + required string name = 1; + required uint32 pol_lb = 2; + required string skey = 3; + required string scrt = 4; + required string ckey = 5; + required string ccrt = 6; } message layer_info_msg { diff --git a/src/lib/protobuf.c b/src/lib/protobuf.c index b586168c..a824d357 100644 --- a/src/lib/protobuf.c +++ b/src/lib/protobuf.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Protobuf syntax conversion * @@ -23,6 +23,8 @@ #define _DEFAULT_SOURCE #include <ouroboros/protobuf.h> +#include <ouroboros/crypt.h> +#include <ouroboros/proc.h> #include <stdlib.h> #include <string.h> @@ -73,12 +75,14 @@ flow_info_msg_t * flow_info_s_to_msg(const struct flow_info * s) flow_info_msg__init(msg); - msg->id = s->id; - msg->n_pid = s->n_pid; - msg->n_1_pid = s->n_1_pid; - msg->mpl = s->mpl; - msg->state = s->state; - msg->qos = qos_spec_s_to_msg(&s->qs); + msg->id = s->id; + msg->n_pid = s->n_pid; + msg->n_1_pid = s->n_1_pid; + msg->mpl = s->mpl; + msg->state = s->state; + msg->uid = s->uid; + msg->mtu = s->mtu; + msg->qos = qos_spec_s_to_msg(&s->qs); if (msg->qos == NULL) goto fail_msg; @@ -96,16 +100,82 @@ struct flow_info flow_info_msg_to_s(const flow_info_msg_t * msg) assert(msg != NULL); + memset(&s, 0, sizeof(s)); + s.id = msg->id; s.n_pid = msg->n_pid; s.n_1_pid = msg->n_1_pid; s.mpl = msg->mpl; s.state = msg->state; + s.uid = msg->uid; + s.mtu = msg->mtu; s.qs = qos_spec_msg_to_s(msg->qos); return s; } +name_info_msg_t * name_info_s_to_msg(const struct name_info * info) +{ + name_info_msg_t * msg; + + assert(info != NULL); + + msg = malloc(sizeof(*msg)); + if (msg == NULL) + goto fail_malloc; + + name_info_msg__init(msg); + + msg->name = strdup(info->name); + if (msg->name == NULL) + goto fail_msg; + + msg->skey = strdup(info->s.key); + if (msg->skey == NULL) + goto fail_msg; + + msg->scrt = strdup(info->s.crt); + if (msg->scrt == NULL) + goto fail_msg; + + msg->ckey = strdup(info->c.key); + if (msg->ckey == NULL) + goto fail_msg; + + msg->ccrt = strdup(info->c.crt); + if (msg->ccrt == NULL) + goto fail_msg; + + msg->pol_lb = info->pol_lb; + + return msg; + + fail_msg: + name_info_msg__free_unpacked(msg, NULL); + fail_malloc: + return NULL; +} + +struct name_info name_info_msg_to_s(const name_info_msg_t * msg) +{ + struct name_info s; + + assert(msg != NULL); + assert(strlen(msg->name) <= NAME_SIZE); + + memset(&s, 0, sizeof(s)); + + strcpy(s.name, msg->name); + strcpy(s.s.key, msg->skey); + strcpy(s.s.crt, msg->scrt); + strcpy(s.c.key, msg->ckey); + strcpy(s.c.crt, msg->ccrt); + + s.pol_lb = msg->pol_lb; + + return s; +} + layer_info_msg_t * layer_info_s_to_msg(const struct layer_info * s) { layer_info_msg_t * msg; @@ -188,6 +258,95 @@ struct ipcp_info ipcp_info_msg_to_s(const ipcp_info_msg_t * msg) return s; } +ls_config_msg_t * ls_config_s_to_msg(const struct ls_config * s) +{ + ls_config_msg_t * msg; + + assert(s != NULL); + + msg = malloc(sizeof(*msg)); + if (msg == NULL) + goto fail_malloc; + + ls_config_msg__init(msg); + + msg->pol = s->pol; + msg->t_recalc = s->t_recalc; + msg->t_update = s->t_update; + msg->t_timeo = s->t_timeo; + + return msg; + + fail_malloc: + return NULL; +} + +struct ls_config ls_config_msg_to_s(const ls_config_msg_t * msg) +{ + struct ls_config s; + + assert(msg != NULL); + + s.pol = msg->pol; + s.t_recalc = msg->t_recalc; + s.t_update = msg->t_update; + s.t_timeo = msg->t_timeo; + + return s; +} + +routing_config_msg_t * routing_config_s_to_msg(const struct routing_config * s) +{ + routing_config_msg_t * msg; + + assert(s != NULL); + + msg = malloc(sizeof(*msg)); + if (msg == NULL) + return NULL; + + routing_config_msg__init(msg); + + switch (s->pol) { + case ROUTING_LINK_STATE: + msg->ls = ls_config_s_to_msg(&s->ls); + if (msg->ls == NULL) + goto fail_ls; + break; + default: + /* No checks here */ + break; + } + + msg->pol = s->pol; + + return msg; + + fail_ls: + routing_config_msg__free_unpacked(msg, NULL); + return NULL; +} + +struct routing_config routing_config_msg_to_s(const routing_config_msg_t * msg) +{ + struct routing_config s; + + assert(msg != NULL); + + switch (msg->pol) { + case ROUTING_LINK_STATE: + s.ls = ls_config_msg_to_s(msg->ls); + break; + default: + /* No checks here */ + break; + } + + s.pol = msg->pol; + + return s; +} + dt_config_msg_t * dt_config_s_to_msg(const struct dt_config * s) { dt_config_msg_t * msg; @@ -203,9 +362,14 @@ dt_config_msg_t * dt_config_s_to_msg(const struct dt_config * s) msg->addr_size = s->addr_size; msg->eid_size = s->eid_size; msg->max_ttl = s->max_ttl; - msg->routing_type = s->routing_type; + msg->routing = routing_config_s_to_msg(&s->routing); + if (msg->routing == NULL) + goto fail_routing; return msg; + fail_routing: + dt_config_msg__free_unpacked(msg, NULL); + return NULL; } struct dt_config dt_config_msg_to_s(const dt_config_msg_t * msg) @@ -217,11 +381,102 @@ struct dt_config dt_config_msg_to_s(const dt_config_msg_t * msg) s.addr_size = msg->addr_size; s.eid_size = msg->eid_size; s.max_ttl = msg->max_ttl; - s.routing_type = msg->routing_type; + s.routing = routing_config_msg_to_s(msg->routing); return s; } +struct dir_dht_config dir_dht_config_msg_to_s(const dir_dht_config_msg_t * msg) +{ + struct dir_dht_config s; + + assert(msg != NULL); + + s.params.alpha = msg->alpha; + s.params.k = msg->k; + s.params.t_expire = msg->t_expire; + s.params.t_refresh = msg->t_refresh; + s.params.t_replicate = msg->t_replicate; + s.peer = msg->peer; + + return s; +} + +dir_dht_config_msg_t * dir_dht_config_s_to_msg(const struct dir_dht_config * s) +{ + dir_dht_config_msg_t * msg; + + assert(s != NULL); + + msg = malloc(sizeof(*msg)); + if (msg == NULL) + return NULL; + + dir_dht_config_msg__init(msg); + + msg->alpha = s->params.alpha; + msg->k = s->params.k; + msg->t_expire = s->params.t_expire; + msg->t_refresh = s->params.t_refresh; + msg->t_replicate = s->params.t_replicate; + msg->peer = s->peer; + + return msg; +} + +struct dir_config dir_config_msg_to_s(const dir_config_msg_t * msg) +{ + struct dir_config s; + + assert(msg != NULL); + + switch (msg->pol) { + case DIR_DHT: + s.dht = dir_dht_config_msg_to_s(msg->dht); + break; + default: + /* No checks here */ + break; + } + + s.pol = msg->pol; + + return s; +} + +dir_config_msg_t * dir_config_s_to_msg(const struct dir_config * s) +{ + dir_config_msg_t * msg; + + assert(s != NULL); + + msg = malloc(sizeof(*msg)); + if (msg == NULL) + return NULL; + + dir_config_msg__init(msg); + + switch (s->pol) { + case DIR_DHT: + msg->dht = dir_dht_config_s_to_msg(&s->dht); + if (msg->dht == NULL) + goto fail_msg; + break; + default: + /* No checks here */ + break; + } + + msg->pol = s->pol; + + return msg; + + fail_msg: + dir_config_msg__free_unpacked(msg, NULL); + return NULL; +} + + uni_config_msg_t * uni_config_s_to_msg(const struct uni_config * s) { uni_config_msg_t * msg; @@ -238,6 +493,11 @@ uni_config_msg_t * uni_config_s_to_msg(const struct uni_config * s) if (msg->dt == NULL) goto fail_msg; + msg->dir = dir_config_s_to_msg(&s->dir); + if (msg->dir == NULL) + goto fail_msg; + + msg->addr_auth_type = s->addr_auth_type; msg->cong_avoid = s->cong_avoid; @@ -254,6 +514,7 @@ struct uni_config uni_config_msg_to_s(const uni_config_msg_t * msg) struct uni_config s; s.dt = dt_config_msg_to_s(msg->dt); + s.dir = dir_config_msg_to_s(msg->dir); s.addr_auth_type = msg->addr_auth_type; s.cong_avoid = msg->cong_avoid; @@ -261,9 +522,9 @@ struct uni_config uni_config_msg_to_s(const uni_config_msg_t * msg) return s; } -udp_config_msg_t * udp_config_s_to_msg(const struct udp_config * s) +udp4_config_msg_t * udp4_config_s_to_msg(const struct udp4_config * s) { - udp_config_msg_t * msg; + udp4_config_msg_t * msg; assert(s != NULL); @@ -271,24 +532,77 @@ udp_config_msg_t * udp_config_s_to_msg(const struct udp_config * s) if (msg == NULL) return NULL; - udp_config_msg__init(msg); + udp4_config_msg__init(msg); - msg->ip_addr = s->ip_addr; - msg->dns_addr = s->dns_addr; + msg->ip_addr = s->ip_addr.s_addr; + msg->dns_addr = s->dns_addr.s_addr; msg->port = s->port; return msg; } -struct udp_config udp_config_msg_to_s(const udp_config_msg_t * msg) +struct udp4_config udp4_config_msg_to_s(const udp4_config_msg_t * msg) { - struct udp_config s; + struct udp4_config s; assert(msg != NULL); - s.ip_addr = msg->ip_addr; - s.dns_addr = msg->dns_addr; - s.port = msg->port; + s.ip_addr.s_addr = msg->ip_addr; + s.dns_addr.s_addr = msg->dns_addr; + s.port = msg->port; + + return s; +} + +#define IN6_LEN (size_t) sizeof(struct in6_addr) +udp6_config_msg_t * udp6_config_s_to_msg(const struct udp6_config * s) +{ + udp6_config_msg_t * msg; + + assert(s != NULL); + + msg = malloc(sizeof(*msg)); + if (msg == NULL) + goto fail_malloc; + + udp6_config_msg__init(msg); + + msg->ip_addr.data = malloc(IN6_LEN); + if (msg->ip_addr.data == NULL) + goto fail_msg; + + msg->ip_addr.len = IN6_LEN; + memcpy(msg->ip_addr.data, &s->ip_addr.s6_addr, IN6_LEN); + + msg->dns_addr.data = malloc(IN6_LEN); + if (msg->dns_addr.data == NULL) + goto fail_msg; + + msg->dns_addr.len = IN6_LEN; + memcpy(msg->dns_addr.data, &s->dns_addr.s6_addr, IN6_LEN); + + msg->port = s->port; + + return msg; + + fail_msg: + udp6_config_msg__free_unpacked(msg, NULL); + fail_malloc: + return NULL; +} + +struct udp6_config udp6_config_msg_to_s(const udp6_config_msg_t * msg) +{ + struct udp6_config s; + + assert(msg != NULL); + + assert(msg->ip_addr.len == IN6_LEN); + assert(msg->dns_addr.len == IN6_LEN); + + memcpy(&s.ip_addr.s6_addr, msg->ip_addr.data, IN6_LEN); + memcpy(&s.dns_addr.s6_addr, msg->dns_addr.data, IN6_LEN); + s.port = msg->port; return s; } @@ -362,9 +676,14 @@ ipcp_config_msg_t * ipcp_config_s_to_msg(const struct ipcp_config * s) if (msg->eth == NULL) goto fail_msg; break; - case IPCP_UDP: - msg->udp = udp_config_s_to_msg(&s->udp); - if (msg->udp == NULL) + case IPCP_UDP4: + msg->udp4 = udp4_config_s_to_msg(&s->udp4); + if (msg->udp4 == NULL) + goto fail_msg; + break; + case IPCP_UDP6: + msg->udp6 = udp6_config_s_to_msg(&s->udp6); + if (msg->udp6 == NULL) goto fail_msg; break; default: @@ -407,8 +726,11 @@ struct ipcp_config ipcp_config_msg_to_s(const ipcp_config_msg_t * msg) case IPCP_ETH_DIX: s.eth = eth_config_msg_to_s(msg->eth); break; - case IPCP_UDP: - s.udp = udp_config_msg_to_s(msg->udp); + case IPCP_UDP4: + s.udp4 = udp4_config_msg_to_s(msg->udp4); + break; + case IPCP_UDP6: + s.udp6 = udp6_config_msg_to_s(msg->udp6); break; case IPCP_BROADCAST: break; @@ -437,9 +759,8 @@ qosspec_msg_t * qos_spec_s_to_msg(const struct qos_spec * s) msg->availability = s->availability; msg->loss = s->loss; msg->ber = s->ber; - msg->in_order = s->in_order; + msg->service = s->service; msg->max_gap = s->max_gap; - msg->cypher_s = s->cypher_s; msg->timeout = s->timeout; return msg; @@ -456,9 +777,8 @@ struct qos_spec qos_spec_msg_to_s(const qosspec_msg_t * msg) s.availability = msg->availability; s.loss = msg->loss; s.ber = msg->ber; - s.in_order = msg->in_order; + s.service = msg->service; s.max_gap = msg->max_gap; - s.cypher_s = msg->cypher_s; s.timeout = msg->timeout; return s; diff --git a/src/lib/qoscube.c b/src/lib/qoscube.c index 267b3a87..5d7ae17d 100644 --- a/src/lib/qoscube.c +++ b/src/lib/qoscube.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Quality of Service cube * @@ -29,15 +29,11 @@ qoscube_t qos_spec_to_cube(qosspec_t qs) { - if (qs.delay <= qos_voice.delay && - qs.bandwidth <= qos_voice.bandwidth && - qs.availability >= qos_voice.availability && - qs.max_gap <= qos_voice.max_gap) + if (qs.delay <= 50 && qs.bandwidth <= 100000 + && qs.availability >= 5 && qs.max_gap <= 50) return QOS_CUBE_VOICE; - else if (qs.delay <= qos_video.delay && - qs.bandwidth <= qos_video.bandwidth && - qs.availability >= qos_video.availability && - qs.max_gap <= qos_video.max_gap) + else if (qs.delay <= 100 && qs.availability >= 3 + && qs.max_gap <= 100) return QOS_CUBE_VIDEO; else return QOS_CUBE_BE; diff --git a/src/lib/random.c b/src/lib/random.c index 09e0b844..a132f470 100644 --- a/src/lib/random.c +++ b/src/lib/random.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Pseudo random generator * @@ -24,22 +24,14 @@ #include <ouroboros/random.h> -#if defined(__APPLE__) /* Barf */ -#undef __OSX_AVAILABLE -#define __OSX_AVAILABLE(arg) -#undef __IOS_AVAILABLE -#define __IOS_AVAILABLE(arg) -#undef __TVOS_AVAILABLE -#define __TVOS_AVAILABLE(arg) -#undef __WATCHOS_AVAILABLE -#define __WATCHOS_AVAILABLE(arg) -#include <sys/random.h> +#if defined(__APPLE__) || defined(__FreeBSD__) +#include <stdlib.h> #elif defined(HAVE_SYS_RANDOM) #include <sys/random.h> +#include <errno.h> +#include <stdint.h> #elif defined(HAVE_LIBGCRYPT) #include <gcrypt.h> -#elif defined(__FreeBSD__) -#include <stdlib.h> #elif defined(HAVE_OPENSSL_RNG) #include <openssl/rand.h> #include <limits.h> @@ -48,19 +40,32 @@ int random_buffer(void * buf, size_t len) { -#if defined(__APPLE__) - return getentropy(buf, len); -#elif defined(__FreeBSD__) +#if defined(__APPLE__) || defined(__FreeBSD__) arc4random_buf(buf, len); return 0; #elif defined(HAVE_SYS_RANDOM) - return getrandom(buf, len, GRND_NONBLOCK); /* glibc 2.25 */ + size_t off = 0; + ssize_t ret; + + while (off < len) { + ret = getrandom((uint8_t *) buf + off, len - off, + GRND_NONBLOCK); + if (ret < 0) { + if (errno == EINTR) + continue; + return -1; + } + off += (size_t) ret; + } + + return 0; #elif defined(HAVE_LIBGCRYPT) gcry_randomize(buf, len, GCRY_STRONG_RANDOM); return 0; #elif defined(HAVE_OPENSSL_RNG) - if (len > 0 && len < INT_MAX) - return RAND_bytes((unsigned char *) buf, (int) len); - return -1; + if (len == 0 || len >= INT_MAX) + return -1; + + return RAND_bytes((unsigned char *) buf, (int) len) == 1 ? 0 : -1; #endif } diff --git a/src/lib/rib.c b/src/lib/rib.c index 97a20f47..6e421397 100644 --- a/src/lib/rib.c +++ b/src/lib/rib.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * RIB export using FUSE * @@ -112,14 +112,14 @@ static int rib_read(const char * path, (void) info; (void) offset; - pthread_rwlock_wrlock(&rib.lock); + pthread_rwlock_rdlock(&rib.lock); list_for_each(p, &rib.reg_comps) { struct reg_comp * r = list_entry(p, struct reg_comp, next); if (strcmp(comp, r->path) == 0) { - int ret = r->ops->read(path + 1, buf, size); + struct rib_ops * ops = r->ops; pthread_rwlock_unlock(&rib.lock); - return ret; + return ops->read(path + 1, buf, size); } } @@ -160,19 +160,25 @@ static int rib_readdir(const char * path, ssize_t len; ssize_t i; struct reg_comp * c; + struct rib_ops * ops; c = list_entry(p, struct reg_comp, next); if (strcmp(path + 1, c->path) != 0) continue; - assert(c->ops->readdir != NULL); + ops = c->ops; + + assert(ops->readdir != NULL); + + pthread_rwlock_unlock(&rib.lock); - len = c->ops->readdir(&dir_entries); + len = ops->readdir(&dir_entries); if (len < 0) - break; + return 0; for (i = 0; i < len; ++i) filler(buf, dir_entries[i], NULL, 0); freepp(char, dir_entries, len); + return 0; } } diff --git a/src/lib/serdes-irm.c b/src/lib/serdes-irm.c index c4ba3053..1d9b4dec 100644 --- a/src/lib/serdes-irm.c +++ b/src/lib/serdes-irm.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Ouroboros IRM Protocol - serialization/deserialization * @@ -24,6 +24,7 @@ #include "config.h" +#include <ouroboros/crypt.h> #include <ouroboros/errno.h> #include <ouroboros/serdes-irm.h> #include <ouroboros/protobuf.h> @@ -135,14 +136,11 @@ int flow_join__irm_req_ser(buffer_t * buf, int flow__irm_result_des(buffer_t * buf, struct flow_info * flow, - buffer_t * sk) + struct crypt_sk * sk) { irm_msg_t * msg; int err; - if (sk != NULL) - sk->data = NULL; - msg = irm_msg__unpack(NULL, buf->len, buf->data); if (msg == NULL) { err = -EIRMD; @@ -166,17 +164,62 @@ int flow__irm_result_des(buffer_t * buf, *flow = flow_info_msg_to_s(msg->flow_info); - if (flow->qs.cypher_s > 0 && sk != NULL) { - if (msg->symmkey.data == NULL || msg->symmkey.len == 0) { - err = -ECRYPT; - goto fail; - } + if (msg->has_cipher_nid) + sk->nid = msg->cipher_nid; + else + sk->nid = NID_undef; + + if (msg->sym_key.len == SYMMKEYSZ) + memcpy(sk->key, msg->sym_key.data, SYMMKEYSZ); + else + memset(sk->key, 0, SYMMKEYSZ); + + sk->epoch = msg->has_generation ? (uint8_t) msg->generation : 0; + + if (msg->sym_key.len == SYMMKEYSZ) + crypt_secure_clear(msg->sym_key.data, msg->sym_key.len); + + irm_msg__free_unpacked(msg, NULL); + + return 0; + fail: + irm_msg__free_unpacked(msg, NULL); + fail_msg: + return err; +} + +int flow_rekey__irm_result_des(buffer_t * buf, + struct crypt_sk * sk, + bool * has_key, + bool * initiator) +{ + irm_msg_t * msg; + int err; + + msg = irm_msg__unpack(NULL, buf->len, buf->data); + if (msg == NULL) { + err = -EIRMD; + goto fail_msg; + } + + if (!msg->has_result) { + err = -EIRMD; + goto fail; + } - sk->len = msg->symmkey.len; - sk->data = msg->symmkey.data; + if (msg->result < 0) { + err = msg->result; + goto fail; + } - msg->symmkey.data = NULL; - msg->symmkey.len = 0; + *has_key = msg->has_sym_key && msg->sym_key.len == SYMMKEYSZ; + if (*has_key) { + memcpy(sk->key, msg->sym_key.data, SYMMKEYSZ); + sk->nid = NID_undef; + sk->epoch = msg->has_generation ? + (uint8_t) msg->generation : 0; + *initiator = msg->has_rk_initiator && msg->rk_initiator; + crypt_secure_clear(msg->sym_key.data, msg->sym_key.len); } irm_msg__free_unpacked(msg, NULL); @@ -227,6 +270,44 @@ int flow_dealloc__irm_req_ser(buffer_t * buf, return -ENOMEM; } +int flow_update__irm_req_ser(buffer_t * buf, + const struct flow_info * flow, + bool rekey) +{ + irm_msg_t * msg; + size_t len; + + msg = malloc(sizeof(*msg)); + if (msg == NULL) + goto fail_malloc; + + irm_msg__init(msg); + + msg->code = IRM_MSG_CODE__IRM_FLOW_UPDATE; + msg->flow_info = flow_info_s_to_msg(flow); + if (msg->flow_info == NULL) + goto fail_msg; + + msg->has_rekey = true; + msg->rekey = rekey; + + len = irm_msg__get_packed_size(msg); + if (len == 0 || len > buf->len) + goto fail_msg; + + buf->len = len; + + irm_msg__pack(msg, buf->data); + irm_msg__free_unpacked(msg, NULL); + + return 0; + + fail_msg: + irm_msg__free_unpacked(msg, NULL); + fail_malloc: + return -ENOMEM; +} + int ipcp_flow_dealloc__irm_req_ser(buffer_t * buf, const struct flow_info * flow) { @@ -294,8 +375,8 @@ int ipcp_create_r__irm_req_ser(buffer_t * buf, return -ENOMEM; } -int proc_announce__irm_req_ser(buffer_t * buf, - const char * prog) +int proc_announce__irm_req_ser(buffer_t * buf, + const struct proc_info * proc) { irm_msg_t * msg; size_t len; @@ -308,8 +389,8 @@ int proc_announce__irm_req_ser(buffer_t * buf, msg->code = IRM_MSG_CODE__IRM_PROC_ANNOUNCE; msg->has_pid = true; - msg->pid = getpid(); - msg->prog = strdup(prog); + msg->pid = proc->pid; + msg->prog = strdup(proc->prog); if (msg->prog == NULL) goto fail_msg; @@ -403,6 +484,56 @@ int ipcp_flow_req_arr__irm_req_ser(buffer_t * buf, return 0; fail_msg: + /* hash/pk are borrowed from the caller; detach before free. */ + msg->hash.len = 0; + msg->hash.data = NULL; + msg->pk.len = 0; + msg->pk.data = NULL; + irm_msg__free_unpacked(msg, NULL); + fail_malloc: + return -ENOMEM; +} + +int ipcp_flow_update_arr__irm_req_ser(buffer_t * buf, + const struct flow_info * flow, + const buffer_t * data) +{ + irm_msg_t * msg; + size_t len; + + msg = malloc(sizeof(*msg)); + if (msg == NULL) + goto fail_malloc; + + irm_msg__init(msg); + + msg->code = IRM_MSG_CODE__IPCP_FLOW_UPDATE_ARR; + msg->flow_info = flow_info_s_to_msg(flow); + if (msg->flow_info == NULL) + goto fail_msg; + + msg->has_pk = true; + msg->pk.len = data->len; + msg->pk.data = data->data; + + len = irm_msg__get_packed_size(msg); + if (len == 0 || len > buf->len) + goto fail_msg; + + buf->len = len; + + irm_msg__pack(msg, buf->data); + + /* Don't free data! */ + msg->pk.len = 0; + msg->pk.data = NULL; + irm_msg__free_unpacked(msg, NULL); + + return 0; + fail_msg: + /* pk.data is borrowed from the caller; detach before free. */ + msg->pk.len = 0; + msg->pk.data = NULL; irm_msg__free_unpacked(msg, NULL); fail_malloc: return -ENOMEM; diff --git a/src/lib/serdes-oep.c b/src/lib/serdes-oep.c index 8a836b3b..3d191494 100644 --- a/src/lib/serdes-oep.c +++ b/src/lib/serdes-oep.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Ouroboros Enrollment Protocol - serialization/deserialization * diff --git a/src/lib/sha3.c b/src/lib/sha3.c index b9d6b07f..f406124e 100644 --- a/src/lib/sha3.c +++ b/src/lib/sha3.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * SHA3 algorithm * diff --git a/src/lib/shm_rbuff.c b/src/lib/shm_rbuff.c deleted file mode 100644 index 22cff41c..00000000 --- a/src/lib/shm_rbuff.c +++ /dev/null @@ -1,215 +0,0 @@ -/* - * Ouroboros - Copyright (C) 2016 - 2024 - * - * Ring buffer implementations for incoming packets - * - * Dimitri Staessens <dimitri@ouroboros.rocks> - * Sander Vrijders <sander@ouroboros.rocks> - * - * This library is free software; you can redistribute it and/or - * modify it under the terms of the GNU Lesser General Public License - * version 2.1 as published by the Free Software Foundation. - * - * This library is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * Lesser General Public License for more details. - * - * You should have received a copy of the GNU Lesser General Public - * License along with this library; if not, write to the Free Software - * Foundation, Inc., http://www.fsf.org/about/contact/. - */ - -#define _POSIX_C_SOURCE 200809L - -#include "config.h" - -#include <ouroboros/shm_rbuff.h> -#include <ouroboros/lockfile.h> -#include <ouroboros/errno.h> -#include <ouroboros/fccntl.h> -#include <ouroboros/pthread.h> -#include <ouroboros/time.h> - -#include <assert.h> -#include <fcntl.h> -#include <signal.h> -#include <stdbool.h> -#include <stdint.h> -#include <stdio.h> -#include <stdlib.h> -#include <string.h> -#include <unistd.h> -#include <sys/mman.h> -#include <sys/stat.h> - -#define FN_MAX_CHARS 255 - -#define SHM_RB_FILE_SIZE ((SHM_RBUFF_SIZE) * sizeof(ssize_t) \ - + 3 * sizeof(size_t) \ - + sizeof(pthread_mutex_t) \ - + 2 * sizeof (pthread_cond_t)) - -#define shm_rbuff_used(rb) ((*rb->head + (SHM_RBUFF_SIZE) - *rb->tail) \ - & ((SHM_RBUFF_SIZE) - 1)) -#define shm_rbuff_free(rb) (shm_rbuff_used(rb) + 1 < (SHM_RBUFF_SIZE)) -#define shm_rbuff_empty(rb) (*rb->head == *rb->tail) -#define head_el_ptr(rb) (rb->shm_base + *rb->head) -#define tail_el_ptr(rb) (rb->shm_base + *rb->tail) - -struct shm_rbuff { - ssize_t * shm_base; /* start of entry */ - size_t * head; /* start of ringbuffer head */ - size_t * tail; /* start of ringbuffer tail */ - size_t * acl; /* access control */ - pthread_mutex_t * lock; /* lock all free space in shm */ - pthread_cond_t * add; /* packet arrived */ - pthread_cond_t * del; /* packet removed */ - pid_t pid; /* pid of the owner */ - int flow_id; /* flow_id of the flow */ -}; - -#define MM_FLAGS (PROT_READ | PROT_WRITE) - -static struct shm_rbuff * rbuff_create(pid_t pid, - int flow_id, - int flags) -{ - struct shm_rbuff * rb; - int fd; - ssize_t * shm_base; - char fn[FN_MAX_CHARS]; - - sprintf(fn, SHM_RBUFF_PREFIX "%d.%d", pid, flow_id); - - rb = malloc(sizeof(*rb)); - if (rb == NULL) - goto fail_malloc; - - fd = shm_open(fn, flags, 0666); - if (fd == -1) - goto fail_open; - - if ((flags & O_CREAT) && ftruncate(fd, SHM_RB_FILE_SIZE) < 0) - goto fail_truncate; - - shm_base = mmap(NULL, SHM_RB_FILE_SIZE, MM_FLAGS, MAP_SHARED, fd, 0); - if (shm_base == MAP_FAILED) - goto fail_truncate; - - close(fd); - - rb->shm_base = shm_base; - rb->head = (size_t *) (rb->shm_base + (SHM_RBUFF_SIZE)); - rb->tail = rb->head + 1; - rb->acl = rb->tail + 1; - rb->lock = (pthread_mutex_t *) (rb->acl + 1); - rb->add = (pthread_cond_t *) (rb->lock + 1); - rb->del = rb->add + 1; - rb->pid = pid; - rb->flow_id = flow_id; - - return rb; - - fail_truncate: - close(fd); - if (flags & O_CREAT) - shm_unlink(fn); - fail_open: - free(rb); - fail_malloc: - return NULL; -} - -static void rbuff_destroy(struct shm_rbuff * rb) -{ - munmap(rb->shm_base, SHM_RB_FILE_SIZE); - - free(rb); -} - -struct shm_rbuff * shm_rbuff_create(pid_t pid, - int flow_id) -{ - struct shm_rbuff * rb; - pthread_mutexattr_t mattr; - pthread_condattr_t cattr; - mode_t mask; - - mask = umask(0); - - rb = rbuff_create(pid, flow_id, O_CREAT | O_EXCL | O_RDWR); - - umask(mask); - - if (rb == NULL) - goto fail_rb; - - if (pthread_mutexattr_init(&mattr)) - goto fail_mattr; - - pthread_mutexattr_setpshared(&mattr, PTHREAD_PROCESS_SHARED); -#ifdef HAVE_ROBUST_MUTEX - pthread_mutexattr_setrobust(&mattr, PTHREAD_MUTEX_ROBUST); -#endif - if (pthread_mutex_init(rb->lock, &mattr)) - goto fail_mutex; - - if (pthread_condattr_init(&cattr)) - goto fail_cattr; - - pthread_condattr_setpshared(&cattr, PTHREAD_PROCESS_SHARED); -#ifndef __APPLE__ - pthread_condattr_setclock(&cattr, PTHREAD_COND_CLOCK); -#endif - if (pthread_cond_init(rb->add, &cattr)) - goto fail_add; - - if (pthread_cond_init(rb->del, &cattr)) - goto fail_del; - - *rb->acl = ACL_RDWR; - *rb->head = 0; - *rb->tail = 0; - - rb->pid = pid; - rb->flow_id = flow_id; - - pthread_mutexattr_destroy(&mattr); - pthread_condattr_destroy(&cattr); - - return rb; - - fail_del: - pthread_cond_destroy(rb->add); - fail_add: - pthread_condattr_destroy(&cattr); - fail_cattr: - pthread_mutex_destroy(rb->lock); - fail_mutex: - pthread_mutexattr_destroy(&mattr); - fail_mattr: - shm_rbuff_destroy(rb); - fail_rb: - return NULL; -} - -struct shm_rbuff * shm_rbuff_open(pid_t pid, - int flow_id) -{ - return rbuff_create(pid, flow_id, O_RDWR); -} - -void shm_rbuff_close(struct shm_rbuff * rb) -{ - assert(rb); - - rbuff_destroy(rb); -} - -#if (defined(SHM_RBUFF_LOCKLESS) && \ - (defined(__GNUC__) || defined (__clang__))) -#include "shm_rbuff_ll.c" -#else -#include "shm_rbuff_pthr.c" -#endif diff --git a/src/lib/shm_rbuff_ll.c b/src/lib/shm_rbuff_ll.c deleted file mode 100644 index 46a5314e..00000000 --- a/src/lib/shm_rbuff_ll.c +++ /dev/null @@ -1,249 +0,0 @@ -/* - * Ouroboros - Copyright (C) 2016 - 2024 - * - * Lockless ring buffer for incoming packets - * - * Dimitri Staessens <dimitri@ouroboros.rocks> - * Sander Vrijders <sander@ouroboros.rocks> - * - * This library is free software; you can redistribute it and/or - * modify it under the terms of the GNU Lesser General Public License - * version 2.1 as published by the Free Software Foundation. - * - * This library is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * Lesser General Public License for more details. - * - * You should have received a copy of the GNU Lesser General Public - * License along with this library; if not, write to the Free Software - * Foundation, Inc., http://www.fsf.org/about/contact/. - */ - -#define RB_HEAD __sync_fetch_and_add(rb->head, 0) -#define RB_TAIL __sync_fetch_and_add(rb->tail, 0) - -void shm_rbuff_destroy(struct shm_rbuff * rb) -{ - char fn[FN_MAX_CHARS]; - - assert(rb); - - sprintf(fn, SHM_RBUFF_PREFIX "%d.%d", rb->pid, rb->flow_id); - - __sync_bool_compare_and_swap(rb->acl, *rb->acl, ACL_FLOWDOWN); - - pthread_cond_broadcast(rb->del); - pthread_cond_broadcast(rb->add); - - shm_rbuff_close(rb); - - shm_unlink(fn); -} - -int shm_rbuff_write(struct shm_rbuff * rb, - size_t idx) -{ - size_t ohead; - size_t nhead; - bool was_empty = false; - - assert(rb); - assert(idx < SHM_BUFFER_SIZE); - - if (__sync_fetch_and_add(rb->acl, 0) != ACL_RDWR) { - if (__sync_fetch_and_add(rb->acl, 0) & ACL_FLOWDOWN) - return -EFLOWDOWN; - else if (__sync_fetch_and_add(rb->acl, 0) & ACL_RDONLY) - return -ENOTALLOC; - } - - if (!shm_rbuff_free(rb)) - return -EAGAIN; - - if (shm_rbuff_empty(rb)) - was_empty = true; - - nhead = RB_HEAD; - - *(rb->shm_base + nhead) = (ssize_t) idx; - - do { - ohead = nhead; - nhead = (ohead + 1) & ((SHM_RBUFF_SIZE) - 1); - nhead = __sync_val_compare_and_swap(rb->head, ohead, nhead); - } while (nhead != ohead); - - if (was_empty) - pthread_cond_broadcast(rb->add); - - return 0; -} - -/* FIXME: this is a copy of the pthr implementation */ -int shm_rbuff_write_b(struct shm_rbuff * rb, - size_t idx, - const struct timespec * abstime) -{ - int ret = 0; - - assert(rb); - assert(idx < SHM_BUFFER_SIZE); - -#ifndef HAVE_ROBUST_MUTEX - pthread_mutex_lock(rb->lock); -#else - if (pthread_mutex_lock(rb->lock) == EOWNERDEAD) - pthread_mutex_consistent(rb->lock); -#endif - - if (*rb->acl != ACL_RDWR) { - if (*rb->acl & ACL_FLOWDOWN) - ret = -EFLOWDOWN; - else if (*rb->acl & ACL_RDONLY) - ret = -ENOTALLOC; - goto err; - } - - pthread_cleanup_push(__cleanup_mutex_unlock, rb->lock); - - while (!shm_rbuff_free(rb) && ret != -ETIMEDOUT) { - ret = -__timedwait(rb->add, rb->lock, abstime); -#ifdef HAVE_ROBUST_MUTEX - if (ret == -EOWNERDEAD) - pthread_mutex_consistent(rb->lock); -#endif - } - - if (shm_rbuff_empty(rb)) - pthread_cond_broadcast(rb->add); - - if (ret != -ETIMEDOUT) { - *head_el_ptr(rb) = (ssize_t) idx; - *rb->head = (*rb->head + 1) & ((SHM_RBUFF_SIZE) -1); - } - - pthread_cleanup_pop(true); - - return ret; - err: - pthread_mutex_unlock(rb->lock); - return ret; -} - -ssize_t shm_rbuff_read(struct shm_rbuff * rb) -{ - size_t otail; - size_t ntail; - - assert(rb); - - if (shm_rbuff_empty(rb)) { - if (_sync_fetch_and_add(rb->acl, 0) & ACL_FLOWDOWN) - return -EFLOWDOWN; - - if (_sync_fetch_and_add(rb->acl, 0) & ACL_FLOWPEER) - return -EFLOWPEER; - - return -EAGAIN; - } - - ntail = RB_TAIL; - - do { - otail = ntail; - ntail = (otail + 1) & ((SHM_RBUFF_SIZE) - 1); - ntail = __sync_val_compare_and_swap(rb->tail, otail, ntail); - } while (ntail != otail); - - pthread_cond_broadcast(rb->del); - - return *(rb->shm_base + ntail); -} - -ssize_t shm_rbuff_read_b(struct shm_rbuff * rb, - const struct timespec * abstime) -{ - ssize_t idx = -1; - - assert(rb); - - /* try a non-blocking read first */ - idx = shm_rbuff_read(rb); - if (idx != -EAGAIN) - return idx; - -#ifndef HAVE_ROBUST_MUTEX - pthread_mutex_lock(rb->lock); -#else - if (pthread_mutex_lock(rb->lock) == EOWNERDEAD) - pthread_mutex_consistent(rb->lock); -#endif - pthread_cleanup_push(__cleanup_mutex_unlock, rb->lock); - - while (shm_rbuff_empty(rb) && (idx != -ETIMEDOUT)) { - idx = -__timedwait(rb->add, rb->lock, abstime); -#ifdef HAVE_ROBUST_MUTEX - if (idx == -EOWNERDEAD) - pthread_mutex_consistent(rb->lock); -#endif - } - - if (idx != -ETIMEDOUT) { - /* do a nonblocking read */ - idx = shm_rbuff_read(rb); - assert(idx >= 0); - } - - pthread_cleanup_pop(true); - - return idx; -} - -void shm_rbuff_set_acl(struct shm_rbuff * rb, - uint32_t flags) -{ - assert(rb); - - __sync_bool_compare_and_swap(rb->acl, *rb->acl, flags); -} - -uint32_t shm_rbuff_get_acl(struct shm_rbuff * rb) -{ - assert(rb); - - return __sync_fetch_and_add(rb->acl, 0); -} - -void shm_rbuff_fini(struct shm_rbuff * rb) -{ - assert(rb); - - if (shm_rbuff_empty(rb)) - return; - -#ifndef HAVE_ROBUST_MUTEX - pthread_mutex_lock(rb->lock); -#else - if (pthread_mutex_lock(rb->lock) == EOWNERDEAD) - pthread_mutex_consistent(rb->lock); -#endif - - pthread_cleanup_push(__cleanup_mutex_unlock, rb->lock); - - while (!shm_rbuff_empty(rb)) -#ifndef HAVE_ROBUST_MUTEX - pthread_cond_wait(rb->del, rb->lock); -#else - if (pthread_cond_wait(rb->del, rb->lock) == EOWNERDEAD) - pthread_mutex_consistent(rb->lock); -#endif - pthread_cleanup_pop(true); -} - -size_t shm_rbuff_queued(struct shm_rbuff * rb) -{ - assert(rb); - - return shm_rbuff_used(rb); -} diff --git a/src/lib/shm_rbuff_pthr.c b/src/lib/shm_rbuff_pthr.c deleted file mode 100644 index b543fb07..00000000 --- a/src/lib/shm_rbuff_pthr.c +++ /dev/null @@ -1,304 +0,0 @@ -/* - * Ouroboros - Copyright (C) 2016 - 2024 - * - * Ring buffer for incoming packets - * - * Dimitri Staessens <dimitri@ouroboros.rocks> - * Sander Vrijders <sander@ouroboros.rocks> - * - * This library is free software; you can redistribute it and/or - * modify it under the terms of the GNU Lesser General Public License - * version 2.1 as published by the Free Software Foundation. - * - * This library is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * Lesser General Public License for more details. - * - * You should have received a copy of the GNU Lesser General Public - * License along with this library; if not, write to the Free Software - * Foundation, Inc., http://www.fsf.org/about/contact/. - */ - -void shm_rbuff_destroy(struct shm_rbuff * rb) -{ - char fn[FN_MAX_CHARS]; - - assert(rb != NULL); - -#ifdef CONFIG_OUROBOROS_DEBUG - pthread_mutex_lock(rb->lock); - - *rb->acl = *rb->acl & ACL_FLOWDOWN; - - pthread_cond_broadcast(rb->del); - pthread_cond_broadcast(rb->add); - - pthread_mutex_unlock(rb->lock); -#endif - sprintf(fn, SHM_RBUFF_PREFIX "%d.%d", rb->pid, rb->flow_id); - - shm_rbuff_close(rb); - - shm_unlink(fn); -} - -int shm_rbuff_write(struct shm_rbuff * rb, - size_t idx) -{ - int ret = 0; - - assert(rb != NULL); - assert(idx < SHM_BUFFER_SIZE); - -#ifndef HAVE_ROBUST_MUTEX - pthread_mutex_lock(rb->lock); -#else - if (pthread_mutex_lock(rb->lock) == EOWNERDEAD) - pthread_mutex_consistent(rb->lock); -#endif - - if (*rb->acl != ACL_RDWR) { - if (*rb->acl & ACL_FLOWDOWN) - ret = -EFLOWDOWN; - else if (*rb->acl & ACL_RDONLY) - ret = -ENOTALLOC; - goto err; - } - - if (!shm_rbuff_free(rb)) { - ret = -EAGAIN; - goto err; - } - - if (shm_rbuff_empty(rb)) - pthread_cond_broadcast(rb->add); - - *head_el_ptr(rb) = (ssize_t) idx; - *rb->head = (*rb->head + 1) & ((SHM_RBUFF_SIZE) - 1); - - pthread_mutex_unlock(rb->lock); - - return 0; - err: - pthread_mutex_unlock(rb->lock); - return ret; -} - -int shm_rbuff_write_b(struct shm_rbuff * rb, - size_t idx, - const struct timespec * abstime) -{ - int ret = 0; - - assert(rb != NULL); - assert(idx < SHM_BUFFER_SIZE); - -#ifndef HAVE_ROBUST_MUTEX - pthread_mutex_lock(rb->lock); -#else - if (pthread_mutex_lock(rb->lock) == EOWNERDEAD) - pthread_mutex_consistent(rb->lock); -#endif - - if (*rb->acl != ACL_RDWR) { - if (*rb->acl & ACL_FLOWDOWN) - ret = -EFLOWDOWN; - else if (*rb->acl & ACL_RDONLY) - ret = -ENOTALLOC; - goto err; - } - - pthread_cleanup_push(__cleanup_mutex_unlock, rb->lock); - - while (!shm_rbuff_free(rb) - && ret != -ETIMEDOUT - && !(*rb->acl & ACL_FLOWDOWN)) { - ret = -__timedwait(rb->del, rb->lock, abstime); -#ifdef HAVE_ROBUST_MUTEX - if (ret == -EOWNERDEAD) - pthread_mutex_consistent(rb->lock); -#endif - } - - if (ret != -ETIMEDOUT) { - if (shm_rbuff_empty(rb)) - pthread_cond_broadcast(rb->add); - *head_el_ptr(rb) = (ssize_t) idx; - *rb->head = (*rb->head + 1) & ((SHM_RBUFF_SIZE) - 1); - } - - pthread_cleanup_pop(true); - - return ret; - err: - pthread_mutex_unlock(rb->lock); - return ret; -} - -static int check_rb_acl(struct shm_rbuff * rb) -{ - assert(rb != NULL); - - if (*rb->acl & ACL_FLOWDOWN) - return -EFLOWDOWN; - - if (*rb->acl & ACL_FLOWPEER) - return -EFLOWPEER; - - return -EAGAIN; -} - -ssize_t shm_rbuff_read(struct shm_rbuff * rb) -{ - ssize_t ret = 0; - - assert(rb != NULL); - -#ifndef HAVE_ROBUST_MUTEX - pthread_mutex_lock(rb->lock); -#else - if (pthread_mutex_lock(rb->lock) == EOWNERDEAD) - pthread_mutex_consistent(rb->lock); -#endif - - if (shm_rbuff_empty(rb)) { - ret = check_rb_acl(rb); - pthread_mutex_unlock(rb->lock); - return ret; - } - - ret = *tail_el_ptr(rb); - *rb->tail = (*rb->tail + 1) & ((SHM_RBUFF_SIZE) - 1); - pthread_cond_broadcast(rb->del); - - pthread_mutex_unlock(rb->lock); - - return ret; -} - -ssize_t shm_rbuff_read_b(struct shm_rbuff * rb, - const struct timespec * abstime) -{ - ssize_t idx = -1; - - assert(rb != NULL); - -#ifndef HAVE_ROBUST_MUTEX - pthread_mutex_lock(rb->lock); -#else - if (pthread_mutex_lock(rb->lock) == EOWNERDEAD) - pthread_mutex_consistent(rb->lock); -#endif - - if (shm_rbuff_empty(rb) && (*rb->acl & ACL_FLOWDOWN)) { - pthread_mutex_unlock(rb->lock); - return -EFLOWDOWN; - } - - pthread_cleanup_push(__cleanup_mutex_unlock, rb->lock); - - while (shm_rbuff_empty(rb) && - idx != -ETIMEDOUT && - check_rb_acl(rb) == -EAGAIN) { - idx = -__timedwait(rb->add, rb->lock, abstime); -#ifdef HAVE_ROBUST_MUTEX - if (idx == -EOWNERDEAD) - pthread_mutex_consistent(rb->lock); -#endif - } - - if (!shm_rbuff_empty(rb)) { - idx = *tail_el_ptr(rb); - *rb->tail = (*rb->tail + 1) & ((SHM_RBUFF_SIZE) - 1); - pthread_cond_broadcast(rb->del); - } else if (idx != -ETIMEDOUT) { - idx = check_rb_acl(rb); - } - - pthread_cleanup_pop(true); - - assert(idx != -EAGAIN); - - return idx; -} - -void shm_rbuff_set_acl(struct shm_rbuff * rb, - uint32_t flags) -{ - assert(rb != NULL); - -#ifndef HAVE_ROBUST_MUTEX - pthread_mutex_lock(rb->lock); -#else - if (pthread_mutex_lock(rb->lock) == EOWNERDEAD) - pthread_mutex_consistent(rb->lock); -#endif - *rb->acl = (size_t) flags; - - pthread_cond_broadcast(rb->del); - pthread_cond_broadcast(rb->add); - - pthread_mutex_unlock(rb->lock); -} - -uint32_t shm_rbuff_get_acl(struct shm_rbuff * rb) -{ - uint32_t flags; - - assert(rb != NULL); - -#ifndef HAVE_ROBUST_MUTEX - pthread_mutex_lock(rb->lock); -#else - if (pthread_mutex_lock(rb->lock) == EOWNERDEAD) - pthread_mutex_consistent(rb->lock); -#endif - flags = (uint32_t) *rb->acl; - - pthread_mutex_unlock(rb->lock); - - return flags; -} - -void shm_rbuff_fini(struct shm_rbuff * rb) -{ - assert(rb != NULL); - -#ifndef HAVE_ROBUST_MUTEX - pthread_mutex_lock(rb->lock); -#else - if (pthread_mutex_lock(rb->lock) == EOWNERDEAD) - pthread_mutex_consistent(rb->lock); -#endif - pthread_cleanup_push(__cleanup_mutex_unlock, rb->lock); - - while (!shm_rbuff_empty(rb)) -#ifndef HAVE_ROBUST_MUTEX - pthread_cond_wait(rb->del, rb->lock); -#else - if (pthread_cond_wait(rb->del, rb->lock) == EOWNERDEAD) - pthread_mutex_consistent(rb->lock); -#endif - pthread_cleanup_pop(true); -} - -size_t shm_rbuff_queued(struct shm_rbuff * rb) -{ - size_t ret; - - assert(rb != NULL); - -#ifndef HAVE_ROBUST_MUTEX - pthread_mutex_lock(rb->lock); -#else - if (pthread_mutex_lock(rb->lock) == EOWNERDEAD) - pthread_mutex_consistent(rb->lock); -#endif - - ret = shm_rbuff_used(rb); - - pthread_mutex_unlock(rb->lock); - - return ret; -} diff --git a/src/lib/shm_rdrbuff.c b/src/lib/shm_rdrbuff.c deleted file mode 100644 index 7ad1bd2e..00000000 --- a/src/lib/shm_rdrbuff.c +++ /dev/null @@ -1,610 +0,0 @@ -/* - * Ouroboros - Copyright (C) 2016 - 2024 - * - * Random Deletion Ring Buffer for Data Units - * - * Dimitri Staessens <dimitri@ouroboros.rocks> - * Sander Vrijders <sander@ouroboros.rocks> - * - * This library is free software; you can redistribute it and/or - * modify it under the terms of the GNU Lesser General Public License - * version 2.1 as published by the Free Software Foundation. - * - * This library is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * Lesser General Public License for more details. - * - * You should have received a copy of the GNU Lesser General Public - * License along with this library; if not, write to the Free Software - * Foundation, Inc., http://www.fsf.org/about/contact/. - */ - -#define _POSIX_C_SOURCE 200809L - -#include "config.h" - -#include <ouroboros/errno.h> -#include <ouroboros/pthread.h> -#include <ouroboros/shm_rdrbuff.h> - -#include <assert.h> -#include <fcntl.h> -#include <signal.h> -#include <stdbool.h> -#include <stdio.h> -#include <stdlib.h> -#include <string.h> -#include <unistd.h> -#include <sys/mman.h> -#include <sys/stat.h> - -#define SHM_BLOCKS_SIZE ((SHM_BUFFER_SIZE) * SHM_RDRB_BLOCK_SIZE) -#define SHM_FILE_SIZE (SHM_BLOCKS_SIZE + 2 * sizeof(size_t) \ - + sizeof(pthread_mutex_t) + 2 * sizeof(pthread_cond_t) \ - + sizeof(pid_t)) -#define DU_BUFF_OVERHEAD (DU_BUFF_HEADSPACE + DU_BUFF_TAILSPACE) - -#define get_head_ptr(rdrb) \ - idx_to_du_buff_ptr(rdrb, *rdrb->head) - -#define get_tail_ptr(rdrb) \ - idx_to_du_buff_ptr(rdrb, *rdrb->tail) - -#define idx_to_du_buff_ptr(rdrb, idx) \ - ((struct shm_du_buff *) (rdrb->shm_base + idx * SHM_RDRB_BLOCK_SIZE)) - -#define shm_rdrb_used(rdrb) \ - (((*rdrb->head + (SHM_BUFFER_SIZE) - *rdrb->tail) + 1) \ - & ((SHM_BUFFER_SIZE) - 1)) - -#define shm_rdrb_free(rdrb, i) \ - (shm_rdrb_used(rdrb) + i < (SHM_BUFFER_SIZE)) - -#define shm_rdrb_empty(rdrb) \ - (*rdrb->tail == *rdrb->head) - -struct shm_du_buff { - size_t size; -#ifdef SHM_RDRB_MULTI_BLOCK - size_t blocks; -#endif - size_t du_head; - size_t du_tail; - size_t refs; - size_t idx; -}; - -struct shm_rdrbuff { - uint8_t * shm_base; /* start of blocks */ - size_t * head; /* start of ringbuffer head */ - size_t * tail; /* start of ringbuffer tail */ - pthread_mutex_t * lock; /* lock all free space in shm */ - pthread_cond_t * healthy; /* flag when packet is read */ - pid_t * pid; /* pid of the irmd owner */ -}; - -static void garbage_collect(struct shm_rdrbuff * rdrb) -{ -#ifdef SHM_RDRB_MULTI_BLOCK - struct shm_du_buff * sdb; - while (!shm_rdrb_empty(rdrb) && - (sdb = get_tail_ptr(rdrb))->refs == 0) - *rdrb->tail = (*rdrb->tail + sdb->blocks) - & ((SHM_BUFFER_SIZE) - 1); -#else - while (!shm_rdrb_empty(rdrb) && get_tail_ptr(rdrb)->refs == 0) - *rdrb->tail = (*rdrb->tail + 1) & ((SHM_BUFFER_SIZE) - 1); -#endif - pthread_cond_broadcast(rdrb->healthy); -} - -#ifdef HAVE_ROBUST_MUTEX -static void sanitize(struct shm_rdrbuff * rdrb) -{ - --get_head_ptr(rdrb)->refs; - garbage_collect(rdrb); - pthread_mutex_consistent(rdrb->lock); -} -#endif - -static char * rdrb_filename(void) -{ - char * str; - - str = malloc(strlen(SHM_RDRB_NAME) + 1); - if (str == NULL) - return NULL; - - sprintf(str, "%s", SHM_RDRB_NAME); - - return str; -} - -void shm_rdrbuff_close(struct shm_rdrbuff * rdrb) -{ - assert(rdrb); - - munmap(rdrb->shm_base, SHM_FILE_SIZE); - free(rdrb); -} - -void shm_rdrbuff_destroy(struct shm_rdrbuff * rdrb) -{ - char * shm_rdrb_fn; - - assert(rdrb); - - if (getpid() != *rdrb->pid && kill(*rdrb->pid, 0) == 0) { - free(rdrb); - return; - } - - shm_rdrbuff_close(rdrb); - - shm_rdrb_fn = rdrb_filename(); - if (shm_rdrb_fn == NULL) - return; - - shm_unlink(shm_rdrb_fn); - free(shm_rdrb_fn); -} - -#define MM_FLAGS (PROT_READ | PROT_WRITE) - -static struct shm_rdrbuff * rdrb_create(int flags) -{ - struct shm_rdrbuff * rdrb; - int fd; - uint8_t * shm_base; - char * shm_rdrb_fn; - - shm_rdrb_fn = rdrb_filename(); - if (shm_rdrb_fn == NULL) - goto fail_fn; - - rdrb = malloc(sizeof *rdrb); - if (rdrb == NULL) - goto fail_rdrb; - - fd = shm_open(shm_rdrb_fn, flags, 0666); - if (fd == -1) - goto fail_open; - - if ((flags & O_CREAT) && ftruncate(fd, SHM_FILE_SIZE) < 0) - goto fail_truncate; - - shm_base = mmap(NULL, SHM_FILE_SIZE, MM_FLAGS, MAP_SHARED, fd, 0); - if (shm_base == MAP_FAILED) - goto fail_truncate; - - close(fd); - - rdrb->shm_base = shm_base; - rdrb->head = (size_t *) ((uint8_t *) rdrb->shm_base + SHM_BLOCKS_SIZE); - rdrb->tail = rdrb->head + 1; - rdrb->lock = (pthread_mutex_t *) (rdrb->tail + 1); - rdrb->healthy = (pthread_cond_t *) (rdrb->lock + 1); - rdrb->pid = (pid_t *) (rdrb->healthy + 1); - - free(shm_rdrb_fn); - - return rdrb; - - fail_truncate: - close(fd); - if (flags & O_CREAT) - shm_unlink(shm_rdrb_fn); - fail_open: - free(rdrb); - fail_rdrb: - free(shm_rdrb_fn); - fail_fn: - return NULL; -} - -struct shm_rdrbuff * shm_rdrbuff_create(void) -{ - struct shm_rdrbuff * rdrb; - mode_t mask; - pthread_mutexattr_t mattr; - pthread_condattr_t cattr; - - mask = umask(0); - - rdrb = rdrb_create(O_CREAT | O_EXCL | O_RDWR); - - umask(mask); - - if (rdrb == NULL) - goto fail_rdrb; - - if (pthread_mutexattr_init(&mattr)) - goto fail_mattr; - - pthread_mutexattr_setpshared(&mattr, PTHREAD_PROCESS_SHARED); -#ifdef HAVE_ROBUST_MUTEX - pthread_mutexattr_setrobust(&mattr, PTHREAD_MUTEX_ROBUST); -#endif - if (pthread_mutex_init(rdrb->lock, &mattr)) - goto fail_mutex; - - if (pthread_condattr_init(&cattr)) - goto fail_cattr; - - pthread_condattr_setpshared(&cattr, PTHREAD_PROCESS_SHARED); -#ifndef __APPLE__ - pthread_condattr_setclock(&cattr, PTHREAD_COND_CLOCK); -#endif - if (pthread_cond_init(rdrb->healthy, &cattr)) - goto fail_healthy; - - *rdrb->head = 0; - *rdrb->tail = 0; - - *rdrb->pid = getpid(); - - pthread_mutexattr_destroy(&mattr); - pthread_condattr_destroy(&cattr); - - return rdrb; - - fail_healthy: - pthread_condattr_destroy(&cattr); - fail_cattr: - pthread_mutex_destroy(rdrb->lock); - fail_mutex: - pthread_mutexattr_destroy(&mattr); - fail_mattr: - shm_rdrbuff_destroy(rdrb); - fail_rdrb: - return NULL; -} - -struct shm_rdrbuff * shm_rdrbuff_open(void) -{ - return rdrb_create(O_RDWR); -} - -void shm_rdrbuff_purge(void) -{ - char * shm_rdrb_fn; - - shm_rdrb_fn = rdrb_filename(); - if (shm_rdrb_fn == NULL) - return; - - shm_unlink(shm_rdrb_fn); - free(shm_rdrb_fn); -} - -ssize_t shm_rdrbuff_alloc(struct shm_rdrbuff * rdrb, - size_t len, - uint8_t ** ptr, - struct shm_du_buff ** psdb) -{ - struct shm_du_buff * sdb; - size_t size = DU_BUFF_OVERHEAD + len; -#ifdef SHM_RDRB_MULTI_BLOCK - size_t blocks = 0; - size_t padblocks = 0; -#endif - ssize_t sz = size + sizeof(*sdb); - - assert(rdrb); - assert(psdb); - -#ifndef SHM_RDRB_MULTI_BLOCK - if (sz > SHM_RDRB_BLOCK_SIZE) - return -EMSGSIZE; -#else - while (sz > 0) { - sz -= SHM_RDRB_BLOCK_SIZE; - ++blocks; - } -#endif -#ifndef HAVE_ROBUST_MUTEX - pthread_mutex_lock(rdrb->lock); -#else - if (pthread_mutex_lock(rdrb->lock) == EOWNERDEAD) - sanitize(rdrb); -#endif -#ifdef SHM_RDRB_MULTI_BLOCK - if (blocks + *rdrb->head > (SHM_BUFFER_SIZE)) - padblocks = (SHM_BUFFER_SIZE) - *rdrb->head; - - if (!shm_rdrb_free(rdrb, blocks + padblocks)) { -#else - if (!shm_rdrb_free(rdrb, 1)) { -#endif - pthread_mutex_unlock(rdrb->lock); - return -EAGAIN; - } - -#ifdef SHM_RDRB_MULTI_BLOCK - if (padblocks) { - sdb = get_head_ptr(rdrb); - sdb->size = 0; - sdb->blocks = padblocks; - sdb->refs = 0; - sdb->du_head = 0; - sdb->du_tail = 0; - sdb->idx = *rdrb->head; - - *rdrb->head = 0; - } -#endif - sdb = get_head_ptr(rdrb); - sdb->refs = 1; - sdb->idx = *rdrb->head; -#ifdef SHM_RDRB_MULTI_BLOCK - sdb->blocks = blocks; - - *rdrb->head = (*rdrb->head + blocks) & ((SHM_BUFFER_SIZE) - 1); -#else - *rdrb->head = (*rdrb->head + 1) & ((SHM_BUFFER_SIZE) - 1); -#endif - pthread_mutex_unlock(rdrb->lock); - - sdb->size = size; - sdb->du_head = DU_BUFF_HEADSPACE; - sdb->du_tail = sdb->du_head + len; - - *psdb = sdb; - if (ptr != NULL) - *ptr = (uint8_t *) (sdb + 1) + sdb->du_head; - - return sdb->idx; -} - -ssize_t shm_rdrbuff_alloc_b(struct shm_rdrbuff * rdrb, - size_t len, - uint8_t ** ptr, - struct shm_du_buff ** psdb, - const struct timespec * abstime) -{ - struct shm_du_buff * sdb; - size_t size = DU_BUFF_OVERHEAD + len; -#ifdef SHM_RDRB_MULTI_BLOCK - size_t blocks = 0; - size_t padblocks = 0; -#endif - ssize_t sz = size + sizeof(*sdb); - int ret = 0; - - assert(rdrb); - assert(psdb); - -#ifndef SHM_RDRB_MULTI_BLOCK - if (sz > SHM_RDRB_BLOCK_SIZE) - return -EMSGSIZE; -#else - while (sz > 0) { - sz -= SHM_RDRB_BLOCK_SIZE; - ++blocks; - } -#endif -#ifndef HAVE_ROBUST_MUTEX - pthread_mutex_lock(rdrb->lock); -#else - if (pthread_mutex_lock(rdrb->lock) == EOWNERDEAD) - sanitize(rdrb); -#endif - pthread_cleanup_push(__cleanup_mutex_unlock, rdrb->lock); - -#ifdef SHM_RDRB_MULTI_BLOCK - if (blocks + *rdrb->head > (SHM_BUFFER_SIZE)) - padblocks = (SHM_BUFFER_SIZE) - *rdrb->head; - - while (!shm_rdrb_free(rdrb, blocks + padblocks) && ret != ETIMEDOUT) { -#else - while (!shm_rdrb_free(rdrb, 1) && ret != ETIMEDOUT) { -#endif - ret = __timedwait(rdrb->healthy, rdrb->lock, abstime); -#ifdef SHM_RDRB_MULTI_BLOCK - if (blocks + *rdrb->head > (SHM_BUFFER_SIZE)) - padblocks = (SHM_BUFFER_SIZE) - *rdrb->head; -#endif - } - - if (ret != ETIMEDOUT) { -#ifdef SHM_RDRB_MULTI_BLOCK - if (padblocks) { - sdb = get_head_ptr(rdrb); - sdb->size = 0; - sdb->blocks = padblocks; - sdb->refs = 0; - sdb->du_head = 0; - sdb->du_tail = 0; - sdb->idx = *rdrb->head; - - *rdrb->head = 0; - } -#endif - sdb = get_head_ptr(rdrb); - sdb->refs = 1; - sdb->idx = *rdrb->head; -#ifdef SHM_RDRB_MULTI_BLOCK - sdb->blocks = blocks; - - *rdrb->head = (*rdrb->head + blocks) & ((SHM_BUFFER_SIZE) - 1); -#else - *rdrb->head = (*rdrb->head + 1) & ((SHM_BUFFER_SIZE) - 1); -#endif - } - - pthread_cleanup_pop(true); - - if (ret == ETIMEDOUT) - return -ETIMEDOUT; - - sdb->size = size; - sdb->du_head = DU_BUFF_HEADSPACE; - sdb->du_tail = sdb->du_head + len; - - *psdb = sdb; - if (ptr != NULL) - *ptr = (uint8_t *) (sdb + 1) + sdb->du_head; - - return sdb->idx; -} - -ssize_t shm_rdrbuff_read(uint8_t ** dst, - struct shm_rdrbuff * rdrb, - size_t idx) -{ - struct shm_du_buff * sdb; - - assert(dst); - assert(rdrb); - assert(idx < (SHM_BUFFER_SIZE)); - - sdb = idx_to_du_buff_ptr(rdrb, idx); - *dst = ((uint8_t *) (sdb + 1)) + sdb->du_head; - - return (ssize_t) (sdb->du_tail - sdb->du_head); -} - -struct shm_du_buff * shm_rdrbuff_get(struct shm_rdrbuff * rdrb, - size_t idx) -{ - assert(rdrb); - assert(idx < (SHM_BUFFER_SIZE)); - - return idx_to_du_buff_ptr(rdrb, idx); -} - -int shm_rdrbuff_remove(struct shm_rdrbuff * rdrb, - size_t idx) -{ - struct shm_du_buff * sdb; - - assert(rdrb); - assert(idx < (SHM_BUFFER_SIZE)); - -#ifndef HAVE_ROBUST_MUTEX - pthread_mutex_lock(rdrb->lock); -#else - if (pthread_mutex_lock(rdrb->lock) == EOWNERDEAD) - sanitize(rdrb); -#endif - /* assert(!shm_rdrb_empty(rdrb)); */ - - sdb = idx_to_du_buff_ptr(rdrb, idx); - - if (sdb->refs == 1) { /* only stack needs it, can be removed */ - sdb->refs = 0; - if (idx == *rdrb->tail) - garbage_collect(rdrb); - } - - pthread_mutex_unlock(rdrb->lock); - - return 0; -} - -size_t shm_du_buff_get_idx(struct shm_du_buff * sdb) -{ - assert(sdb); - - return sdb->idx; -} - -uint8_t * shm_du_buff_head(struct shm_du_buff * sdb) -{ - assert(sdb); - - return (uint8_t *) (sdb + 1) + sdb->du_head; -} - -uint8_t * shm_du_buff_tail(struct shm_du_buff * sdb) -{ - assert(sdb); - - return (uint8_t *) (sdb + 1) + sdb->du_tail; -} - -size_t shm_du_buff_len(struct shm_du_buff * sdb) -{ - assert(sdb); - - return sdb->du_tail - sdb->du_head; -} - -uint8_t * shm_du_buff_head_alloc(struct shm_du_buff * sdb, - size_t size) -{ - assert(sdb); - - if (sdb->du_head < size) - return NULL; - - sdb->du_head -= size; - - return (uint8_t *) (sdb + 1) + sdb->du_head; -} - -uint8_t * shm_du_buff_tail_alloc(struct shm_du_buff * sdb, - size_t size) -{ - uint8_t * buf; - - assert(sdb); - - if (sdb->du_tail + size >= sdb->size) - return NULL; - - buf = (uint8_t *) (sdb + 1) + sdb->du_tail; - - sdb->du_tail += size; - - return buf; -} - -uint8_t * shm_du_buff_head_release(struct shm_du_buff * sdb, - size_t size) -{ - uint8_t * buf; - - assert(sdb); - assert(!(size > sdb->du_tail - sdb->du_head)); - - buf = (uint8_t *) (sdb + 1) + sdb->du_head; - - sdb->du_head += size; - - return buf; -} - -uint8_t * shm_du_buff_tail_release(struct shm_du_buff * sdb, - size_t size) -{ - assert(sdb); - assert(!(size > sdb->du_tail - sdb->du_head)); - - sdb->du_tail -= size; - - return (uint8_t *) (sdb + 1) + sdb->du_tail; -} - -void shm_du_buff_truncate(struct shm_du_buff * sdb, - size_t len) -{ - assert(sdb); - assert(len <= sdb->size); - - sdb->du_tail = sdb->du_head + len; -} - -int shm_du_buff_wait_ack(struct shm_du_buff * sdb) -{ - __sync_add_and_fetch(&sdb->refs, 1); - - return 0; -} - -int shm_du_buff_ack(struct shm_du_buff * sdb) -{ - __sync_sub_and_fetch(&sdb->refs, 1); - return 0; -} diff --git a/src/lib/sockets.c b/src/lib/sockets.c index 13219db0..46586911 100644 --- a/src/lib/sockets.c +++ b/src/lib/sockets.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * The sockets layer to communicate between daemons * @@ -154,12 +154,13 @@ int send_recv_msg(buffer_t * msg) return len < 0 ? -1 : 0; } -char * ipcp_sock_path(pid_t pid) +static char * __sock_path(pid_t pid, + const char * prefix, + const char * suffix) { char * full_name = NULL; char * pid_string = NULL; size_t len = 0; - char * delim = "_"; len = n_digits(pid); pid_string = malloc(len + 1); @@ -168,9 +169,9 @@ char * ipcp_sock_path(pid_t pid) sprintf(pid_string, "%d", pid); - len += strlen(IPCP_SOCK_PATH_PREFIX); - len += strlen(delim); - len += strlen(SOCK_PATH_SUFFIX); + len += strlen(prefix); + len += strlen(pid_string); + len += strlen(suffix); full_name = malloc(len + 1); if (full_name == NULL) { @@ -178,12 +179,17 @@ char * ipcp_sock_path(pid_t pid) return NULL; } - strcpy(full_name, IPCP_SOCK_PATH_PREFIX); - strcat(full_name, delim); + strcpy(full_name, prefix); strcat(full_name, pid_string); - strcat(full_name, SOCK_PATH_SUFFIX); + strcat(full_name, suffix); free(pid_string); return full_name; } + +char * sock_path(pid_t pid, + const char * prefix) +{ + return __sock_path(pid, prefix, SOCK_PATH_SUFFIX); +} diff --git a/src/lib/shm_flow_set.c b/src/lib/ssm/flow_set.c index 39913fd1..2e33b408 100644 --- a/src/lib/shm_flow_set.c +++ b/src/lib/ssm/flow_set.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Management of flow_sets for fqueue * @@ -23,11 +23,12 @@ #define _POSIX_C_SOURCE 200809L #include "config.h" +#include "ssm.h" #include <ouroboros/errno.h> #include <ouroboros/lockfile.h> #include <ouroboros/pthread.h> -#include <ouroboros/shm_flow_set.h> +#include <ouroboros/ssm_flow_set.h> #include <ouroboros/time.h> #include <assert.h> @@ -54,17 +55,17 @@ #define FN_MAX_CHARS 255 #define FS_PROT (PROT_READ | PROT_WRITE) -#define QUEUESIZE ((SHM_BUFFER_SIZE) * sizeof(struct flowevent)) +#define QUEUESIZE ((SSM_RBUFF_SIZE) * sizeof(struct flowevent)) -#define SHM_FSET_FILE_SIZE (SYS_MAX_FLOWS * sizeof(ssize_t) \ - + PROG_MAX_FQUEUES * sizeof(size_t) \ - + PROG_MAX_FQUEUES * sizeof(pthread_cond_t) \ - + PROG_MAX_FQUEUES * QUEUESIZE \ +#define SSM_FSET_FILE_SIZE (SYS_MAX_FLOWS * sizeof(ssize_t) \ + + PROC_MAX_FQUEUES * sizeof(size_t) \ + + PROC_MAX_FQUEUES * sizeof(pthread_cond_t) \ + + PROC_MAX_FQUEUES * QUEUESIZE \ + sizeof(pthread_mutex_t)) -#define fqueue_ptr(fs, idx) (fs->fqueues + (SHM_BUFFER_SIZE) * idx) +#define fqueue_ptr(fs, idx) (fs->fqueues + (SSM_RBUFF_SIZE) * idx) -struct shm_flow_set { +struct ssm_flow_set { ssize_t * mtable; size_t * heads; pthread_cond_t * conds; @@ -74,15 +75,15 @@ struct shm_flow_set { pid_t pid; }; -static struct shm_flow_set * flow_set_create(pid_t pid, +static struct ssm_flow_set * flow_set_create(pid_t pid, int oflags) { - struct shm_flow_set * set; + struct ssm_flow_set * set; ssize_t * shm_base; char fn[FN_MAX_CHARS]; int fd; - sprintf(fn, SHM_FLOW_SET_PREFIX "%d", pid); + sprintf(fn, SSM_FLOW_SET_PREFIX "%d", pid); set = malloc(sizeof(*set)); if (set == NULL) @@ -92,10 +93,10 @@ static struct shm_flow_set * flow_set_create(pid_t pid, if (fd == -1) goto fail_shm_open; - if ((oflags & O_CREAT) && ftruncate(fd, SHM_FSET_FILE_SIZE) < 0) + if ((oflags & O_CREAT) && ftruncate(fd, SSM_FSET_FILE_SIZE) < 0) goto fail_truncate; - shm_base = mmap(NULL, SHM_FSET_FILE_SIZE, FS_PROT, MAP_SHARED, fd, 0); + shm_base = mmap(NULL, SSM_FSET_FILE_SIZE, FS_PROT, MAP_SHARED, fd, 0); if (shm_base == MAP_FAILED) goto fail_mmap; @@ -103,10 +104,10 @@ static struct shm_flow_set * flow_set_create(pid_t pid, set->mtable = shm_base; set->heads = (size_t *) (set->mtable + SYS_MAX_FLOWS); - set->conds = (pthread_cond_t *)(set->heads + PROG_MAX_FQUEUES); - set->fqueues = (struct flowevent *) (set->conds + PROG_MAX_FQUEUES); + set->conds = (pthread_cond_t *)(set->heads + PROC_MAX_FQUEUES); + set->fqueues = (struct flowevent *) (set->conds + PROC_MAX_FQUEUES); set->lock = (pthread_mutex_t *) - (set->fqueues + PROG_MAX_FQUEUES * (SHM_BUFFER_SIZE)); + (set->fqueues + PROC_MAX_FQUEUES * (SSM_RBUFF_SIZE)); return set; @@ -121,9 +122,9 @@ static struct shm_flow_set * flow_set_create(pid_t pid, return NULL; } -struct shm_flow_set * shm_flow_set_create(pid_t pid) +struct ssm_flow_set * ssm_flow_set_create(pid_t pid) { - struct shm_flow_set * set; + struct ssm_flow_set * set; pthread_mutexattr_t mattr; pthread_condattr_t cattr; mode_t mask; @@ -163,7 +164,7 @@ struct shm_flow_set * shm_flow_set_create(pid_t pid) if (pthread_condattr_setclock(&cattr, PTHREAD_COND_CLOCK)) goto fail_condattr_set; #endif - for (i = 0; i < PROG_MAX_FQUEUES; ++i) { + for (i = 0; i < PROC_MAX_FQUEUES; ++i) { set->heads[i] = 0; if (pthread_cond_init(&set->conds[i], &cattr)) goto fail_init; @@ -184,44 +185,44 @@ struct shm_flow_set * shm_flow_set_create(pid_t pid) fail_mattr_set: pthread_mutexattr_destroy(&mattr); fail_mutexattr_init: - shm_flow_set_destroy(set); + ssm_flow_set_destroy(set); fail_set: return NULL; } -struct shm_flow_set * shm_flow_set_open(pid_t pid) +struct ssm_flow_set * ssm_flow_set_open(pid_t pid) { return flow_set_create(pid, O_RDWR); } -void shm_flow_set_destroy(struct shm_flow_set * set) +void ssm_flow_set_destroy(struct ssm_flow_set * set) { char fn[FN_MAX_CHARS]; assert(set); - sprintf(fn, SHM_FLOW_SET_PREFIX "%d", set->pid); + sprintf(fn, SSM_FLOW_SET_PREFIX "%d", set->pid); - shm_flow_set_close(set); + ssm_flow_set_close(set); shm_unlink(fn); } -void shm_flow_set_close(struct shm_flow_set * set) +void ssm_flow_set_close(struct ssm_flow_set * set) { assert(set); - munmap(set->mtable, SHM_FSET_FILE_SIZE); + munmap(set->mtable, SSM_FSET_FILE_SIZE); free(set); } -void shm_flow_set_zero(struct shm_flow_set * set, +void ssm_flow_set_zero(struct ssm_flow_set * set, size_t idx) { ssize_t i = 0; assert(set); - assert(idx < PROG_MAX_FQUEUES); + assert(idx < PROC_MAX_FQUEUES); pthread_mutex_lock(set->lock); @@ -235,13 +236,13 @@ void shm_flow_set_zero(struct shm_flow_set * set, } -int shm_flow_set_add(struct shm_flow_set * set, +int ssm_flow_set_add(struct ssm_flow_set * set, size_t idx, int flow_id) { assert(set); assert(!(flow_id < 0) && flow_id < SYS_MAX_FLOWS); - assert(idx < PROG_MAX_FQUEUES); + assert(idx < PROC_MAX_FQUEUES); pthread_mutex_lock(set->lock); @@ -257,13 +258,13 @@ int shm_flow_set_add(struct shm_flow_set * set, return 0; } -void shm_flow_set_del(struct shm_flow_set * set, +void ssm_flow_set_del(struct ssm_flow_set * set, size_t idx, int flow_id) { assert(set); assert(!(flow_id < 0) && flow_id < SYS_MAX_FLOWS); - assert(idx < PROG_MAX_FQUEUES); + assert(idx < PROC_MAX_FQUEUES); pthread_mutex_lock(set->lock); @@ -273,7 +274,7 @@ void shm_flow_set_del(struct shm_flow_set * set, pthread_mutex_unlock(set->lock); } -int shm_flow_set_has(struct shm_flow_set * set, +int ssm_flow_set_has(struct ssm_flow_set * set, size_t idx, int flow_id) { @@ -281,7 +282,7 @@ int shm_flow_set_has(struct shm_flow_set * set, assert(set); assert(!(flow_id < 0) && flow_id < SYS_MAX_FLOWS); - assert(idx < PROG_MAX_FQUEUES); + assert(idx < PROC_MAX_FQUEUES); pthread_mutex_lock(set->lock); @@ -293,37 +294,45 @@ int shm_flow_set_has(struct shm_flow_set * set, return ret; } -void shm_flow_set_notify(struct shm_flow_set * set, +void ssm_flow_set_notify(struct ssm_flow_set * set, int flow_id, int event) { struct flowevent * e; + ssize_t idx; assert(set); assert(!(flow_id < 0) && flow_id < SYS_MAX_FLOWS); pthread_mutex_lock(set->lock); - if (set->mtable[flow_id] == -1) { + idx = set->mtable[flow_id]; + if (idx == -1) { pthread_mutex_unlock(set->lock); return; } - e = fqueue_ptr(set, set->mtable[flow_id]) + - set->heads[set->mtable[flow_id]]; + /* Ring full: drop redundant FLOW_PKT, reserve a slot for ctrl. */ + if (set->heads[idx] >= SSM_RBUFF_SIZE + || (event == FLOW_PKT && set->heads[idx] >= SSM_RBUFF_SIZE - 1)) { + pthread_mutex_unlock(set->lock); + return; + } + + e = fqueue_ptr(set, idx) + set->heads[idx]; e->flow_id = flow_id; e->event = event; - ++set->heads[set->mtable[flow_id]]; + ++set->heads[idx]; - pthread_cond_signal(&set->conds[set->mtable[flow_id]]); + pthread_cond_signal(&set->conds[idx]); pthread_mutex_unlock(set->lock); } -ssize_t shm_flow_set_wait(const struct shm_flow_set * set, +ssize_t ssm_flow_set_wait(const struct ssm_flow_set * set, size_t idx, struct flowevent * fqueue, const struct timespec * abstime) @@ -331,7 +340,7 @@ ssize_t shm_flow_set_wait(const struct shm_flow_set * set, ssize_t ret = 0; assert(set); - assert(idx < PROG_MAX_FQUEUES); + assert(idx < PROC_MAX_FQUEUES); assert(fqueue); #ifndef HAVE_ROBUST_MUTEX diff --git a/src/lib/ssm/pool.c b/src/lib/ssm/pool.c new file mode 100644 index 00000000..705de147 --- /dev/null +++ b/src/lib/ssm/pool.c @@ -0,0 +1,919 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Secure Shared Memory Infrastructure (SSMI) Packet Buffer + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This library is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public License + * version 2.1 as published by the Free Software Foundation. + * + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with this library; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#define _POSIX_C_SOURCE 200809L + +#include "config.h" + +#include <ouroboros/atomics.h> +#include <ouroboros/errno.h> +#include <ouroboros/pthread.h> +#include <ouroboros/ssm_pool.h> + +#include "ssm.h" + +#include <assert.h> +#include <fcntl.h> +#include <signal.h> +#include <stdbool.h> +#include <stdio.h> +#include <stdlib.h> +#include <string.h> +#include <time.h> +#include <unistd.h> +#include <sys/mman.h> +#include <sys/stat.h> + +static __inline__ uint64_t pool_now_ns(void) +{ + struct timespec ts; + + clock_gettime(CLOCK_MONOTONIC, &ts); + + return (uint64_t) ts.tv_sec * 1000000000ULL + (uint64_t) ts.tv_nsec; +} + +/* Global Shared Packet Pool (GSPP) configuration */ +static const struct ssm_size_class_cfg ssm_gspp_cfg[SSM_POOL_MAX_CLASSES] = { + { (1 << 8), SSM_GSPP_256_BLOCKS }, + { (1 << 9), SSM_GSPP_512_BLOCKS }, + { (1 << 10), SSM_GSPP_1K_BLOCKS }, + { (1 << 11), SSM_GSPP_2K_BLOCKS }, + { (1 << 12), SSM_GSPP_4K_BLOCKS }, + { (1 << 14), SSM_GSPP_16K_BLOCKS }, + { (1 << 16), SSM_GSPP_64K_BLOCKS }, + { (1 << 18), SSM_GSPP_256K_BLOCKS }, + { (1 << 20), SSM_GSPP_1M_BLOCKS }, +}; + +/* Per-User Pool (PUP) configuration */ +static const struct ssm_size_class_cfg ssm_pup_cfg[SSM_POOL_MAX_CLASSES] = { + { (1 << 8), SSM_PUP_256_BLOCKS }, + { (1 << 9), SSM_PUP_512_BLOCKS }, + { (1 << 10), SSM_PUP_1K_BLOCKS }, + { (1 << 11), SSM_PUP_2K_BLOCKS }, + { (1 << 12), SSM_PUP_4K_BLOCKS }, + { (1 << 14), SSM_PUP_16K_BLOCKS }, + { (1 << 16), SSM_PUP_64K_BLOCKS }, + { (1 << 18), SSM_PUP_256K_BLOCKS }, + { (1 << 20), SSM_PUP_1M_BLOCKS }, +}; + +#define PTR_TO_OFFSET(pool_base, ptr) \ + ((uintptr_t)(ptr) - (uintptr_t)(pool_base)) + +#define OFFSET_TO_PTR(pool_base, offset) \ + ((offset == 0) ? NULL : (void *)((uintptr_t)(pool_base) + offset)) + +#define GET_SHARD_FOR_PID(pid) ((int)((pid) % SSM_POOL_SHARDS)) + + +#define SSM_FILE_SIZE (SSM_POOL_TOTAL_SIZE + sizeof(struct _ssm_pool_hdr)) +#define SSM_GSPP_FILE_SIZE (SSM_GSPP_TOTAL_SIZE + sizeof(struct _ssm_pool_hdr)) +#define SSM_PUP_FILE_SIZE (SSM_PUP_TOTAL_SIZE + sizeof(struct _ssm_pool_hdr)) + +#define IS_GSPP(uid) ((uid) == SSM_GSPP_UID) +#define GET_POOL_TOTAL_SIZE(uid) (IS_GSPP(uid) ? SSM_GSPP_TOTAL_SIZE \ + : SSM_PUP_TOTAL_SIZE) +#define GET_POOL_FILE_SIZE(uid) (IS_GSPP(uid) ? SSM_GSPP_FILE_SIZE \ + : SSM_PUP_FILE_SIZE) +#define GET_POOL_CFG(uid) (IS_GSPP(uid) ? ssm_gspp_cfg : ssm_pup_cfg) + +#define NEEDS_CHOWN(uid, gid) ((uid) != geteuid() || (gid) != getegid()) + +struct ssm_pool { + uint8_t * shm_base; /* start of blocks */ + struct _ssm_pool_hdr * hdr; /* shared memory header */ + void * pool_base; /* base of the memory pool */ + uid_t uid; /* user owner (0 = GSPP) */ + size_t total_size; /* total data size */ +}; + +static __inline__ +struct ssm_pk_buff * list_remove_head(struct _ssm_list_head * head, + void * base) +{ + uint32_t off; + uint32_t next_off; + struct ssm_pk_buff * blk; + + assert(head != NULL); + assert(base != NULL); + + off = LOAD(&head->head_offset); + if (off == 0) + return NULL; + + /* Validate offset is within pool bounds */ + if (off >= SSM_POOL_TOTAL_SIZE) + return NULL; + + blk = OFFSET_TO_PTR(base, off); + next_off = LOAD(&blk->next_offset); + + + + STORE(&head->head_offset, next_off); + STORE(&head->count, LOAD(&head->count) - 1); + + return blk; +} +static __inline__ void list_add_head(struct _ssm_list_head * head, + struct ssm_pk_buff * blk, + void * base) +{ + uint32_t off; + uint32_t old; + + assert(head != NULL); + assert(blk != NULL); + assert(base != NULL); + + off = (uint32_t) PTR_TO_OFFSET(base, blk); + old = LOAD(&head->head_offset); + + STORE(&blk->next_offset, old); + STORE(&head->head_offset, off); + STORE(&head->count, LOAD(&head->count) + 1); +} + +static __inline__ int find_size_class_for_offset(struct ssm_pool * pool, + size_t offset) +{ + int c; + + assert(pool != NULL); + + for (c = 0; c < SSM_POOL_MAX_CLASSES; c++) { + struct _ssm_size_class * sc = &pool->hdr->size_classes[c]; + + if (sc->object_size == 0) + continue; + + if (offset >= sc->pool_start && + offset < sc->pool_start + sc->pool_size) + return c; + } + + return -1; +} + +static void init_size_classes(struct ssm_pool * pool) +{ + const struct ssm_size_class_cfg * cfg; + struct _ssm_size_class * sc; + struct _ssm_shard * shard; + pthread_mutexattr_t mattr; + pthread_condattr_t cattr; + uint8_t * region; + size_t offset; + int c; /* class iterator */ + int s; /* shard iterator */ + size_t i; + + assert(pool != NULL); + + /* Check if already initialized */ + if (LOAD(&pool->hdr->initialized) != 0) + return; + + cfg = GET_POOL_CFG(pool->uid); + + pthread_mutexattr_init(&mattr); + pthread_mutexattr_setpshared(&mattr, PTHREAD_PROCESS_SHARED); +#ifdef HAVE_ROBUST_MUTEX + pthread_mutexattr_setrobust(&mattr, PTHREAD_MUTEX_ROBUST); +#endif + pthread_mutexattr_setprotocol(&mattr, PTHREAD_PRIO_INHERIT); + + pthread_condattr_init(&cattr); + pthread_condattr_setpshared(&cattr, PTHREAD_PROCESS_SHARED); +#ifndef __APPLE__ + pthread_condattr_setclock(&cattr, PTHREAD_COND_CLOCK); +#endif + offset = 0; + + for (c = 0; c < SSM_POOL_MAX_CLASSES; c++) { + if (cfg[c].blocks == 0) + continue; + + sc = &pool->hdr->size_classes[c]; + + sc->object_size = cfg[c].size; + sc->pool_start = offset; + sc->pool_size = cfg[c].size * cfg[c].blocks; + sc->object_count = cfg[c].blocks; + + /* Initialize all shards */ + for (s = 0; s < SSM_POOL_SHARDS; s++) { + shard = &sc->shards[s]; + + STORE(&shard->free_list.head_offset, 0); + STORE(&shard->free_list.count, 0); + STORE(&shard->free_count, 0); + + pthread_mutex_init(&shard->mtx, &mattr); + pthread_cond_init(&shard->cond, &cattr); + } + + /* Lazy distribution: put all blocks in shard 0 initially */ + region = pool->shm_base + offset; + + for (i = 0; i < sc->object_count; ++i) { + struct ssm_pk_buff * blk; + + blk = (struct ssm_pk_buff *) + (region + i * sc->object_size); + + STORE(&blk->refcount, 0); + blk->allocator_pid = 0; + blk->alloc_ts = 0; + STORE(&blk->next_offset, 0); + + list_add_head(&sc->shards[0].free_list, blk, + pool->pool_base); + FETCH_ADD(&sc->shards[0].free_count, 1); + } + + offset += sc->pool_size; + } + + /* Mark as initialized - acts as memory barrier */ + STORE(&pool->hdr->initialized, 1); + + pthread_mutexattr_destroy(&mattr); + pthread_condattr_destroy(&cattr); +} + +/* + * Reclaim all blocks allocated by a specific pid in a size class. + * Called with shard mutex held. + */ +static size_t reclaim_pid_from_sc(struct _ssm_size_class * sc, + struct _ssm_shard * shard, + void * pool_base, + pid_t pid) +{ + uint8_t * region; + size_t i; + size_t recovered = 0; + struct ssm_pk_buff * blk; + uint64_t now; + uint64_t min_age_ns; + + region = (uint8_t *) pool_base + sc->pool_start; + now = pool_now_ns(); + min_age_ns = (uint64_t) SSM_POOL_RECLAIM_AGE_S * 1000000000ULL; + + for (i = 0; i < sc->object_count; ++i) { + blk = (struct ssm_pk_buff *)(region + i * sc->object_size); + + if (blk->allocator_pid != pid) + continue; + + if (LOAD(&blk->refcount) == 0) + continue; + + /* Recent: a live consumer may still hold the handoff. */ + if (now - blk->alloc_ts < min_age_ns) + continue; + + STORE(&blk->refcount, 0); + blk->allocator_pid = 0; + list_add_head(&shard->free_list, blk, pool_base); + FETCH_ADD(&shard->free_count, 1); + recovered++; + } + + return recovered; +} + +void ssm_pool_reclaim_orphans(struct ssm_pool * pool, + pid_t pid) +{ + size_t sc_idx; + + if (pool == NULL || pid <= 0) + return; + + for (sc_idx = 0; sc_idx < SSM_POOL_MAX_CLASSES; sc_idx++) { + struct _ssm_size_class * sc; + struct _ssm_shard * shard; + + sc = &pool->hdr->size_classes[sc_idx]; + if (sc->object_count == 0) + continue; + + /* Reclaim to shard 0 for simplicity */ + shard = &sc->shards[0]; + robust_mutex_lock(&shard->mtx); + reclaim_pid_from_sc(sc, shard, pool->pool_base, pid); + pthread_mutex_unlock(&shard->mtx); + } +} + +static __inline__ +struct ssm_pk_buff * try_alloc_from_shard(struct _ssm_shard * shard, + void * base) +{ + struct ssm_pk_buff * blk; + + robust_mutex_lock(&shard->mtx); + + if (LOAD(&shard->free_count) > 0) { + blk = list_remove_head(&shard->free_list, base); + if (blk != NULL) { + FETCH_SUB(&shard->free_count, 1); + return blk; /* Caller must unlock */ + } + FETCH_SUB(&shard->free_count, 1); + } + + pthread_mutex_unlock(&shard->mtx); + return NULL; +} + +static __inline__ ssize_t init_block(struct ssm_pool * pool, + struct _ssm_size_class * sc, + struct _ssm_shard * shard, + struct ssm_pk_buff * blk, + size_t len, + uint8_t ** ptr, + struct ssm_pk_buff ** spb) +{ + STORE(&blk->refcount, 1); + blk->allocator_pid = getpid(); + blk->alloc_ts = pool_now_ns(); + blk->size = (uint32_t) (sc->object_size - + sizeof(struct ssm_pk_buff)); + blk->pk_head = SSM_PK_BUFF_HEADSPACE; + blk->pk_tail = blk->pk_head + (uint32_t) len; + blk->off = (uint32_t) PTR_TO_OFFSET(pool->pool_base, blk); + + pthread_mutex_unlock(&shard->mtx); + + *spb = blk; + if (ptr != NULL) + *ptr = blk->data + blk->pk_head; + + return blk->off; +} + +/* Non-blocking allocation from size class */ +static ssize_t alloc_from_sc(struct ssm_pool * pool, + int idx, + size_t len, + uint8_t ** ptr, + struct ssm_pk_buff ** spb) +{ + struct _ssm_size_class * sc; + struct ssm_pk_buff * blk; + int local; + int s; + + assert(pool != NULL); + assert(idx >= 0 && idx < SSM_POOL_MAX_CLASSES); + assert(spb != NULL); + + sc = &pool->hdr->size_classes[idx]; + local = GET_SHARD_FOR_PID(getpid()); + + for (s = 0; s < SSM_POOL_SHARDS; s++) { + struct _ssm_shard * shard; + int idx; + + idx = (local + s) % SSM_POOL_SHARDS; + shard = &sc->shards[idx]; + + blk = try_alloc_from_shard(shard, pool->pool_base); + if (blk != NULL) + return init_block(pool, sc, shard, blk, len, ptr, spb); + } + + return -EAGAIN; +} + +/* Blocking allocation from size class */ +static ssize_t alloc_from_sc_b(struct ssm_pool * pool, + int idx, + size_t len, + uint8_t ** ptr, + struct ssm_pk_buff ** spb, + const struct timespec * abstime) +{ + struct _ssm_size_class * sc; + struct _ssm_shard * shard; + struct ssm_pk_buff * blk = NULL; + int local; + int s; + int ret = 0; + + assert(pool != NULL); + assert(idx >= 0 && idx < SSM_POOL_MAX_CLASSES); + assert(spb != NULL); + + sc = &pool->hdr->size_classes[idx]; + local = GET_SHARD_FOR_PID(getpid()); + + while (blk == NULL && ret != ETIMEDOUT) { + /* Try non-blocking allocation from any shard */ + for (s = 0; s < SSM_POOL_SHARDS && blk == NULL; s++) { + shard = &sc->shards[(local + s) % SSM_POOL_SHARDS]; + blk = try_alloc_from_shard(shard, pool->pool_base); + } + + if (blk != NULL) + break; + + /* Nothing available, wait for signal */ + shard = &sc->shards[local]; + robust_mutex_lock(&shard->mtx); + ret = robust_wait(&shard->cond, &shard->mtx, abstime); + pthread_mutex_unlock(&shard->mtx); + } + + if (ret == ETIMEDOUT) + return -ETIMEDOUT; + + return init_block(pool, sc, shard, blk, len, ptr, spb); +} + +/* Generate pool filename: uid=0 for GSPP, uid>0 for PUP */ +static char * pool_filename(uid_t uid) +{ + char base[64]; + + if (IS_GSPP(uid)) + snprintf(base, sizeof(base), "%s", SSM_GSPP_NAME); + else + snprintf(base, sizeof(base), SSM_PUP_NAME_FMT, (int) uid); + + return strdup(base); +} + +void ssm_pool_close(struct ssm_pool * pool) +{ + size_t file_size; + + assert(pool != NULL); + + file_size = GET_POOL_FILE_SIZE(pool->uid); + + munmap(pool->shm_base, file_size); + free(pool); +} + +void ssm_pool_destroy(struct ssm_pool * pool) +{ + char * fn; + + assert(pool != NULL); + + if (getpid() != pool->hdr->pid && kill(pool->hdr->pid, 0) == 0) { + ssm_pool_close(pool); + return; + } + + fn = pool_filename(pool->uid); + if (fn == NULL) { + ssm_pool_close(pool); + return; + } + + ssm_pool_close(pool); + + shm_unlink(fn); + free(fn); +} + +#define MM_FLAGS (PROT_READ | PROT_WRITE) +static struct ssm_pool * __pool_create(const char * name, + int flags, + uid_t uid, + gid_t gid, + mode_t mode) +{ + struct ssm_pool * pool; + int fd; + uint8_t * shm_base; + size_t file_size; + size_t total_size; + + file_size = GET_POOL_FILE_SIZE(uid); + total_size = GET_POOL_TOTAL_SIZE(uid); + + pool = malloc(sizeof(*pool)); + if (pool == NULL) + goto fail_pool; + + fd = shm_open(name, flags, mode); + if (fd == -1) + goto fail_open; + + if (flags & O_CREAT) { + if (ftruncate(fd, (off_t) file_size) < 0) + goto fail_truncate; + if (NEEDS_CHOWN(uid, gid) && fchown(fd, uid, gid) < 0) + goto fail_truncate; + } + + shm_base = mmap(NULL, file_size, MM_FLAGS, MAP_SHARED, fd, 0); + if (shm_base == MAP_FAILED) + goto fail_truncate; + + pool->shm_base = shm_base; + pool->pool_base = shm_base; + pool->hdr = (struct _ssm_pool_hdr *) (shm_base + total_size); + pool->uid = uid; + pool->total_size = total_size; + + if (flags & O_CREAT) + pool->hdr->mapped_addr = shm_base; + + close(fd); + + return pool; + + fail_truncate: + close(fd); + if (flags & O_CREAT) + shm_unlink(name); + fail_open: + free(pool); + fail_pool: + return NULL; +} + +struct ssm_pool * ssm_pool_create(uid_t uid, + gid_t gid) +{ + struct ssm_pool * pool; + char * fn; + mode_t mask; + mode_t mode; + pthread_mutexattr_t mattr; + pthread_condattr_t cattr; + + fn = pool_filename(uid); + if (fn == NULL) + goto fail_fn; + + mode = IS_GSPP(uid) ? 0660 : 0600; + mask = umask(0); + + pool = __pool_create(fn, O_CREAT | O_EXCL | O_RDWR, uid, gid, mode); + + umask(mask); + + if (pool == NULL) + goto fail_pool; + + if (pthread_mutexattr_init(&mattr)) + goto fail_mattr; + + pthread_mutexattr_setpshared(&mattr, PTHREAD_PROCESS_SHARED); +#ifdef HAVE_ROBUST_MUTEX + pthread_mutexattr_setrobust(&mattr, PTHREAD_MUTEX_ROBUST); +#endif + if (pthread_mutex_init(&pool->hdr->mtx, &mattr)) + goto fail_mutex; + + if (pthread_condattr_init(&cattr)) + goto fail_cattr; + + pthread_condattr_setpshared(&cattr, PTHREAD_PROCESS_SHARED); +#ifndef __APPLE__ + pthread_condattr_setclock(&cattr, PTHREAD_COND_CLOCK); +#endif + if (pthread_cond_init(&pool->hdr->healthy, &cattr)) + goto fail_healthy; + + pool->hdr->pid = getpid(); + STORE(&pool->hdr->initialized, 0); + + init_size_classes(pool); + + pthread_mutexattr_destroy(&mattr); + pthread_condattr_destroy(&cattr); + free(fn); + + return pool; + + fail_healthy: + pthread_condattr_destroy(&cattr); + fail_cattr: + pthread_mutex_destroy(&pool->hdr->mtx); + fail_mutex: + pthread_mutexattr_destroy(&mattr); + fail_mattr: + ssm_pool_close(pool); + shm_unlink(fn); + fail_pool: + free(fn); + fail_fn: + return NULL; +} + +struct ssm_pool * ssm_pool_open(uid_t uid) +{ + struct ssm_pool * pool; + char * fn; + + fn = pool_filename(uid); + if (fn == NULL) + return NULL; + + pool = __pool_create(fn, O_RDWR, uid, 0, 0); + if (pool != NULL) + init_size_classes(pool); + + free(fn); + + return pool; +} + +void ssm_pool_gspp_purge(void) +{ + char * fn; + + fn = pool_filename(SSM_GSPP_UID); + if (fn == NULL) + return; + + shm_unlink(fn); + free(fn); +} + +int ssm_pool_mlock(struct ssm_pool * pool) +{ + size_t file_size; + + assert(pool != NULL); + + file_size = GET_POOL_FILE_SIZE(pool->uid); + + return mlock(pool->shm_base, file_size); +} + +ssize_t ssm_pool_alloc(struct ssm_pool * pool, + size_t count, + uint8_t ** ptr, + struct ssm_pk_buff ** spb) +{ + int idx; + + assert(pool != NULL); + assert(spb != NULL); + + idx = select_size_class(pool->hdr, count); + if (idx >= 0) + return alloc_from_sc(pool, idx, count, ptr, spb); + + return -EMSGSIZE; +} + +ssize_t ssm_pool_alloc_b(struct ssm_pool * pool, + size_t count, + uint8_t ** ptr, + struct ssm_pk_buff ** spb, + const struct timespec * abstime) +{ + int idx; + + assert(pool != NULL); + assert(spb != NULL); + + idx = select_size_class(pool->hdr, count); + if (idx >= 0) + return alloc_from_sc_b(pool, idx, count, ptr, spb, abstime); + + return -EMSGSIZE; +} + +ssize_t ssm_pool_read(uint8_t ** dst, + struct ssm_pool * pool, + size_t off) +{ + struct ssm_pk_buff * blk; + + assert(dst != NULL); + assert(pool != NULL); + + blk = OFFSET_TO_PTR(pool->pool_base, off); + if (blk == NULL) + return -EINVAL; + + *dst = blk->data + blk->pk_head; + + return (ssize_t) (blk->pk_tail - blk->pk_head); +} + +struct ssm_pk_buff * ssm_pool_get(struct ssm_pool * pool, + size_t off) +{ + struct ssm_pk_buff * blk; + + assert(pool != NULL); + + if (off == 0 || off >= pool->total_size) + return NULL; + + blk = OFFSET_TO_PTR(pool->pool_base, off); + if (blk == NULL) + return NULL; + + if (LOAD(&blk->refcount) == 0) + return NULL; + + return blk; +} + +int ssm_pool_remove(struct ssm_pool * pool, + size_t off) +{ + struct ssm_pk_buff * blk; + struct _ssm_size_class * sc; + struct _ssm_shard * shard; + int sc_idx; + int shard_idx; + uint16_t old_ref; + + assert(pool != NULL); + + if (off == 0 || off >= pool->total_size) + return -EINVAL; + + blk = OFFSET_TO_PTR(pool->pool_base, off); + if (blk == NULL) + return -EINVAL; + + sc_idx = find_size_class_for_offset(pool, off); + if (sc_idx < 0) + return -EINVAL; + + sc = &pool->hdr->size_classes[sc_idx]; + + /* Free to allocator's shard (lazy distribution in action) */ + shard_idx = GET_SHARD_FOR_PID(blk->allocator_pid); + shard = &sc->shards[shard_idx]; + + robust_mutex_lock(&shard->mtx); + + old_ref = FETCH_SUB(&blk->refcount, 1); + if (old_ref > 1) { + /* Still referenced */ + pthread_mutex_unlock(&shard->mtx); + return 0; + } + + blk->allocator_pid = 0; +#ifdef CONFIG_OUROBOROS_DEBUG + if (old_ref == 0) { + /* Underflow - double free attempt */ + pthread_mutex_unlock(&shard->mtx); + abort(); + } + + /* Poison fields to detect use-after-free */ + blk->pk_head = 0xDEAD; + blk->pk_tail = 0xBEEF; +#endif + list_add_head(&shard->free_list, blk, pool->pool_base); + FETCH_ADD(&shard->free_count, 1); + + pthread_cond_signal(&shard->cond); + + pthread_mutex_unlock(&shard->mtx); + + return 0; +} + +size_t ssm_pk_buff_get_off(const struct ssm_pk_buff * spb) +{ + assert(spb != NULL); + + return spb->off; +} + +uint8_t * ssm_pk_buff_head(const struct ssm_pk_buff * spb) +{ + assert(spb != NULL); + + return (uint8_t *) spb->data + spb->pk_head; +} + +uint8_t * ssm_pk_buff_tail(const struct ssm_pk_buff * spb) +{ + assert(spb != NULL); + + return (uint8_t *) spb->data + spb->pk_tail; +} + +size_t ssm_pk_buff_len(const struct ssm_pk_buff * spb) +{ + assert(spb != NULL); + + return spb->pk_tail - spb->pk_head; +} + +uint8_t * ssm_pk_buff_push(struct ssm_pk_buff * spb, + size_t size) +{ + assert(spb != NULL); + + if (spb->pk_head < size) + return NULL; + + spb->pk_head -= size; + + return spb->data + spb->pk_head; +} + +uint8_t * ssm_pk_buff_push_tail(struct ssm_pk_buff * spb, + size_t size) +{ + uint8_t * buf; + + assert(spb != NULL); + + if (spb->pk_tail + size >= spb->size) + return NULL; + + buf = spb->data + spb->pk_tail; + + spb->pk_tail += size; + + return buf; +} + +uint8_t * ssm_pk_buff_pop(struct ssm_pk_buff * spb, + size_t size) +{ + uint8_t * buf; + + assert(spb != NULL); + assert(!(size > spb->pk_tail - spb->pk_head)); + + buf = spb->data + spb->pk_head; + + spb->pk_head += size; + + return buf; +} + +uint8_t * ssm_pk_buff_pop_tail(struct ssm_pk_buff * spb, + size_t size) +{ + assert(spb != NULL); + assert(!(size > spb->pk_tail - spb->pk_head)); + + spb->pk_tail -= size; + + return spb->data + spb->pk_tail; +} + +void ssm_pk_buff_truncate(struct ssm_pk_buff * spb, + size_t len) +{ + assert(spb != NULL); + assert(len <= spb->size); + + spb->pk_tail = spb->pk_head + len; +} + +int ssm_pk_buff_wait_ack(struct ssm_pk_buff * spb) +{ + assert(spb != NULL); + + FETCH_ADD(&spb->refcount, 1); + + return 0; +} + +int ssm_pk_buff_ack(struct ssm_pk_buff * spb) +{ + assert(spb != NULL); + + FETCH_SUB(&spb->refcount, 1); + + return 0; +} diff --git a/src/lib/ssm/rbuff.c b/src/lib/ssm/rbuff.c new file mode 100644 index 00000000..0121af89 --- /dev/null +++ b/src/lib/ssm/rbuff.c @@ -0,0 +1,508 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Ring buffer implementations for incoming packets + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This library is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public License + * version 2.1 as published by the Free Software Foundation. + * + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with this library; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#define _POSIX_C_SOURCE 200809L + +#include "config.h" +#include "ssm.h" + +#include <ouroboros/ssm_rbuff.h> +#include <ouroboros/lockfile.h> +#include <ouroboros/errno.h> +#include <ouroboros/fccntl.h> +#include <ouroboros/pthread.h> +#include <ouroboros/time.h> + +#include <assert.h> +#include <fcntl.h> +#include <signal.h> +#include <stdbool.h> +#include <stdint.h> +#include <stdio.h> +#include <stdlib.h> +#include <string.h> +#include <unistd.h> +#include <sys/mman.h> +#include <sys/stat.h> + +#define FN_MAX_CHARS 255 + +#define SSM_RBUFF_FILESIZE ((SSM_RBUFF_SIZE) * sizeof(ssize_t) \ + + 3 * sizeof(size_t) \ + + sizeof(pthread_mutex_t) \ + + 2 * sizeof(pthread_cond_t)) + +#define MODB(x) ((x) & (SSM_RBUFF_SIZE - 1)) + +#define LOAD_RELAXED(ptr) (__atomic_load_n(ptr, __ATOMIC_RELAXED)) +#define LOAD_ACQUIRE(ptr) (__atomic_load_n(ptr, __ATOMIC_ACQUIRE)) +#define STORE_RELEASE(ptr, val) \ + (__atomic_store_n(ptr, val, __ATOMIC_RELEASE)) + +#define HEAD(rb) (rb->shm_base[LOAD_RELAXED(rb->head)]) +#define TAIL(rb) (rb->shm_base[LOAD_RELAXED(rb->tail)]) +#define HEAD_IDX(rb) (LOAD_ACQUIRE(rb->head)) +#define TAIL_IDX(rb) (LOAD_ACQUIRE(rb->tail)) +#define ADVANCE_HEAD(rb) \ + (STORE_RELEASE(rb->head, MODB(LOAD_RELAXED(rb->head) + 1))) +#define ADVANCE_TAIL(rb) \ + (STORE_RELEASE(rb->tail, MODB(LOAD_RELAXED(rb->tail) + 1))) +#define QUEUED(rb) (MODB(HEAD_IDX(rb) - TAIL_IDX(rb))) +#define IS_FULL(rb) (QUEUED(rb) == (SSM_RBUFF_SIZE - 1)) +#define IS_EMPTY(rb) (HEAD_IDX(rb) == TAIL_IDX(rb)) + +struct ssm_rbuff { + ssize_t * shm_base; /* start of shared memory */ + size_t * head; /* start of ringbuffer */ + size_t * tail; + size_t * flags; /* out-of-band flags (RB_*) */ + pthread_mutex_t * mtx; /* lock for cond vars only */ + pthread_cond_t * add; /* signal when new data */ + pthread_cond_t * del; /* signal when data removed */ + pid_t pid; /* pid of the owner */ + int flow_id; /* flow_id of the flow */ + size_t n_users; /* in-flight users */ +}; + +#define MM_FLAGS (PROT_READ | PROT_WRITE) + +static struct ssm_rbuff * rbuff_create(pid_t pid, + int flow_id, + int flags) +{ + struct ssm_rbuff * rb; + int fd; + ssize_t * shm_base; + char fn[FN_MAX_CHARS]; + + sprintf(fn, SSM_RBUFF_PREFIX "%d.%d", pid, flow_id); + + rb = malloc(sizeof(*rb)); + if (rb == NULL) + goto fail_malloc; + + fd = shm_open(fn, flags, 0666); + if (fd == -1) + goto fail_open; + + if ((flags & O_CREAT) && ftruncate(fd, SSM_RBUFF_FILESIZE) < 0) + goto fail_truncate; + + shm_base = mmap(NULL, SSM_RBUFF_FILESIZE, MM_FLAGS, MAP_SHARED, fd, 0); + + close(fd); + + rb->shm_base = shm_base; + rb->head = (size_t *) (rb->shm_base + (SSM_RBUFF_SIZE)); + rb->tail = (size_t *) (rb->head + 1); + rb->flags = (size_t *) (rb->tail + 1); + rb->mtx = (pthread_mutex_t *) (rb->flags + 1); + rb->add = (pthread_cond_t *) (rb->mtx + 1); + rb->del = rb->add + 1; + rb->pid = pid; + rb->flow_id = flow_id; + rb->n_users = 0; + + return rb; + + fail_truncate: + close(fd); + if (flags & O_CREAT) + shm_unlink(fn); + fail_open: + free(rb); + fail_malloc: + return NULL; +} + +static void rbuff_destroy(struct ssm_rbuff * rb) +{ + munmap(rb->shm_base, SSM_RBUFF_FILESIZE); + + free(rb); +} + +struct ssm_rbuff * ssm_rbuff_create(pid_t pid, + int flow_id) +{ + struct ssm_rbuff * rb; + pthread_mutexattr_t mattr; + pthread_condattr_t cattr; + mode_t mask; + + mask = umask(0); + + rb = rbuff_create(pid, flow_id, O_CREAT | O_EXCL | O_RDWR); + + umask(mask); + + if (rb == NULL) + goto fail_rb; + + if (pthread_mutexattr_init(&mattr)) + goto fail_mattr; + + pthread_mutexattr_setpshared(&mattr, PTHREAD_PROCESS_SHARED); +#ifdef HAVE_ROBUST_MUTEX + pthread_mutexattr_setrobust(&mattr, PTHREAD_MUTEX_ROBUST); +#endif + if (pthread_mutex_init(rb->mtx, &mattr)) + goto fail_mutex; + + if (pthread_condattr_init(&cattr)) + goto fail_cattr; + + pthread_condattr_setpshared(&cattr, PTHREAD_PROCESS_SHARED); +#ifndef __APPLE__ + pthread_condattr_setclock(&cattr, PTHREAD_COND_CLOCK); +#endif + if (pthread_cond_init(rb->add, &cattr)) + goto fail_add; + + if (pthread_cond_init(rb->del, &cattr)) + goto fail_del; + + *rb->flags = RB_RDWR; + *rb->head = 0; + *rb->tail = 0; + + rb->pid = pid; + rb->flow_id = flow_id; + + pthread_mutexattr_destroy(&mattr); + pthread_condattr_destroy(&cattr); + + return rb; + + fail_del: + pthread_cond_destroy(rb->add); + fail_add: + pthread_condattr_destroy(&cattr); + fail_cattr: + pthread_mutex_destroy(rb->mtx); + fail_mutex: + pthread_mutexattr_destroy(&mattr); + fail_mattr: + ssm_rbuff_destroy(rb); + fail_rb: + return NULL; +} + +void ssm_rbuff_destroy(struct ssm_rbuff * rb) +{ + char fn[FN_MAX_CHARS]; + + assert(rb != NULL); + + sprintf(fn, SSM_RBUFF_PREFIX "%d.%d", rb->pid, rb->flow_id); + + ssm_rbuff_close(rb); + + shm_unlink(fn); +} + +struct ssm_rbuff * ssm_rbuff_open(pid_t pid, + int flow_id) +{ + return rbuff_create(pid, flow_id, O_RDWR); +} + +void ssm_rbuff_close(struct ssm_rbuff * rb) +{ + assert(rb); + + /* + * Caller must set RB_FLOWDOWN first; if a user becomes + * cancellable, push a cleanup that decrements n_users. + */ + while (__atomic_load_n(&rb->n_users, __ATOMIC_SEQ_CST) > 0) { + struct timespec tic = { 0, 100000 }; + nanosleep(&tic, NULL); + } + + rbuff_destroy(rb); +} + +int ssm_rbuff_write(struct ssm_rbuff * rb, + size_t off) +{ + size_t flags; + bool was_empty; + int ret = 0; + + assert(rb != NULL); + + __atomic_fetch_add(&rb->n_users, 1, __ATOMIC_SEQ_CST); + + flags = __atomic_load_n(rb->flags, __ATOMIC_SEQ_CST); + if (flags != RB_RDWR) { + if (flags & RB_FLOWDOWN) { + ret = -EFLOWDOWN; + goto fail_flags; + } + if (!(flags & RB_WR)) { + ret = -ENOTALLOC; + goto fail_flags; + } + } + + robust_mutex_lock(rb->mtx); + + if (IS_FULL(rb)) { + ret = -EAGAIN; + goto fail_mutex; + } + + was_empty = IS_EMPTY(rb); + + HEAD(rb) = (ssize_t) off; + ADVANCE_HEAD(rb); + + if (was_empty) + pthread_cond_broadcast(rb->add); + + pthread_mutex_unlock(rb->mtx); + + __atomic_fetch_sub(&rb->n_users, 1, __ATOMIC_SEQ_CST); + return 0; + + fail_mutex: + pthread_mutex_unlock(rb->mtx); + fail_flags: + __atomic_fetch_sub(&rb->n_users, 1, __ATOMIC_SEQ_CST); + return ret; +} + +int ssm_rbuff_write_b(struct ssm_rbuff * rb, + size_t off, + const struct timespec * abstime) +{ + size_t flags; + int ret = 0; + bool was_empty; + + assert(rb != NULL); + + __atomic_fetch_add(&rb->n_users, 1, __ATOMIC_SEQ_CST); + + flags = __atomic_load_n(rb->flags, __ATOMIC_SEQ_CST); + if (flags != RB_RDWR) { + if (flags & RB_FLOWDOWN) { + ret = -EFLOWDOWN; + goto fail_flags; + } + if (!(flags & RB_WR)) { + ret = -ENOTALLOC; + goto fail_flags; + } + } + + robust_mutex_lock(rb->mtx); + + pthread_cleanup_push(__cleanup_mutex_unlock, rb->mtx); + + while (IS_FULL(rb) && ret != -ETIMEDOUT) { + flags = __atomic_load_n(rb->flags, __ATOMIC_SEQ_CST); + if (flags & RB_FLOWDOWN) { + ret = -EFLOWDOWN; + break; + } + ret = -robust_wait(rb->del, rb->mtx, abstime); + } + + pthread_cleanup_pop(false); + + if (ret != -ETIMEDOUT && ret != -EFLOWDOWN) { + was_empty = IS_EMPTY(rb); + HEAD(rb) = (ssize_t) off; + ADVANCE_HEAD(rb); + if (was_empty) + pthread_cond_broadcast(rb->add); + } + + pthread_mutex_unlock(rb->mtx); + + fail_flags: + __atomic_fetch_sub(&rb->n_users, 1, __ATOMIC_SEQ_CST); + return ret; +} + +static int check_rb_flags(struct ssm_rbuff * rb) +{ + size_t flags; + + assert(rb != NULL); + + flags = __atomic_load_n(rb->flags, __ATOMIC_SEQ_CST); + + if (flags & RB_FLOWDOWN) + return -EFLOWDOWN; + + if (flags & RB_FLOWPEER) + return -EFLOWPEER; + + if (!(flags & RB_RD)) + return -ENOTALLOC; + + return -EAGAIN; +} + +ssize_t ssm_rbuff_read(struct ssm_rbuff * rb) +{ + ssize_t ret; + + assert(rb != NULL); + + __atomic_fetch_add(&rb->n_users, 1, __ATOMIC_SEQ_CST); + + if (IS_EMPTY(rb)) { + ret = check_rb_flags(rb); + goto out; + } + + robust_mutex_lock(rb->mtx); + + if (IS_EMPTY(rb)) { + pthread_mutex_unlock(rb->mtx); + ret = check_rb_flags(rb); + goto out; + } + + ret = TAIL(rb); + ADVANCE_TAIL(rb); + + pthread_cond_broadcast(rb->del); + + pthread_mutex_unlock(rb->mtx); + + out: + __atomic_fetch_sub(&rb->n_users, 1, __ATOMIC_SEQ_CST); + return ret; +} + +ssize_t ssm_rbuff_read_b(struct ssm_rbuff * rb, + const struct timespec * abstime) +{ + ssize_t idx = -1; + size_t flags; + + assert(rb != NULL); + + __atomic_fetch_add(&rb->n_users, 1, __ATOMIC_SEQ_CST); + + flags = __atomic_load_n(rb->flags, __ATOMIC_SEQ_CST); + if (IS_EMPTY(rb) && (flags & RB_FLOWDOWN)) { + idx = -EFLOWDOWN; + goto out; + } + + robust_mutex_lock(rb->mtx); + + pthread_cleanup_push(__cleanup_mutex_unlock, rb->mtx); + + while (IS_EMPTY(rb) && + idx != -ETIMEDOUT && + check_rb_flags(rb) == -EAGAIN) { + idx = -robust_wait(rb->add, rb->mtx, abstime); + } + + pthread_cleanup_pop(false); + + if (!IS_EMPTY(rb)) { + idx = TAIL(rb); + ADVANCE_TAIL(rb); + pthread_cond_broadcast(rb->del); + } else if (idx != -ETIMEDOUT) { + idx = check_rb_flags(rb); + } + + pthread_mutex_unlock(rb->mtx); + + assert(idx != -EAGAIN); + + out: + __atomic_fetch_sub(&rb->n_users, 1, __ATOMIC_SEQ_CST); + return idx; +} + +void ssm_rbuff_set_bits(struct ssm_rbuff * rb, + uint32_t bits) +{ + assert(rb != NULL); + + robust_mutex_lock(rb->mtx); + __atomic_fetch_or(rb->flags, (size_t) bits, __ATOMIC_SEQ_CST); + pthread_cond_broadcast(rb->add); + pthread_cond_broadcast(rb->del); + pthread_mutex_unlock(rb->mtx); +} + +void ssm_rbuff_clr_bits(struct ssm_rbuff * rb, + uint32_t bits) +{ + assert(rb != NULL); + + robust_mutex_lock(rb->mtx); + __atomic_fetch_and(rb->flags, ~(size_t) bits, __ATOMIC_SEQ_CST); + pthread_cond_broadcast(rb->add); + pthread_cond_broadcast(rb->del); + pthread_mutex_unlock(rb->mtx); +} + +uint32_t ssm_rbuff_get_flags(struct ssm_rbuff * rb) +{ + assert(rb != NULL); + + return (uint32_t) __atomic_load_n(rb->flags, __ATOMIC_SEQ_CST); +} + +void ssm_rbuff_fini(struct ssm_rbuff * rb) +{ + assert(rb != NULL); + + __atomic_fetch_add(&rb->n_users, 1, __ATOMIC_SEQ_CST); + + robust_mutex_lock(rb->mtx); + + pthread_cleanup_push(__cleanup_mutex_unlock, rb->mtx); + + while (!IS_EMPTY(rb)) + robust_wait(rb->del, rb->mtx, NULL); + + pthread_cleanup_pop(true); + + __atomic_fetch_sub(&rb->n_users, 1, __ATOMIC_SEQ_CST); +} + +size_t ssm_rbuff_queued(struct ssm_rbuff * rb) +{ + assert(rb != NULL); + + return QUEUED(rb); +} + +int ssm_rbuff_mlock(struct ssm_rbuff * rb) +{ + assert(rb != NULL); + + return mlock(rb->shm_base, SSM_RBUFF_FILESIZE); +} diff --git a/src/lib/ssm/ssm.h.in b/src/lib/ssm/ssm.h.in new file mode 100644 index 00000000..57febae4 --- /dev/null +++ b/src/lib/ssm/ssm.h.in @@ -0,0 +1,190 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Secure Shared Memory configuration + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This library is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public License + * version 2.1 as published by the Free Software Foundation. + * + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with this library; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#ifndef OUROBOROS_LIB_SSM_H +#define OUROBOROS_LIB_SSM_H + +#include <stddef.h> +#include <stdint.h> +#include <stdatomic.h> +#include <sys/types.h> + +/* Pool naming configuration */ +#define SSM_PREFIX "@SSM_PREFIX@" +#define SSM_GSPP_NAME "@SSM_GSPP_NAME@" +#define SSM_PUP_NAME_FMT "@SSM_PUP_NAME_FMT@" +#define SSM_GSPP_UID 0 + +/* Legacy SSM constants */ +#define SSM_RBUFF_PREFIX "@SSM_RBUFF_PREFIX@" +#define SSM_FLOW_SET_PREFIX "@SSM_FLOW_SET_PREFIX@" +#define SSM_POOL_NAME "@SSM_POOL_NAME@" +#define SSM_RBUFF_SIZE @SSM_RBUFF_SIZE@ + +/* Packet buffer space reservation */ +#define SSM_PK_BUFF_HEADSPACE @SSM_PK_BUFF_HEADSPACE@ +#define SSM_PK_BUFF_TAILSPACE @SSM_PK_BUFF_TAILSPACE@ + +/* Global Shared Packet Pool (GSPP) - for privileged processes */ +#define SSM_GSPP_256_BLOCKS @SSM_GSPP_256_BLOCKS@ +#define SSM_GSPP_512_BLOCKS @SSM_GSPP_512_BLOCKS@ +#define SSM_GSPP_1K_BLOCKS @SSM_GSPP_1K_BLOCKS@ +#define SSM_GSPP_2K_BLOCKS @SSM_GSPP_2K_BLOCKS@ +#define SSM_GSPP_4K_BLOCKS @SSM_GSPP_4K_BLOCKS@ +#define SSM_GSPP_16K_BLOCKS @SSM_GSPP_16K_BLOCKS@ +#define SSM_GSPP_64K_BLOCKS @SSM_GSPP_64K_BLOCKS@ +#define SSM_GSPP_256K_BLOCKS @SSM_GSPP_256K_BLOCKS@ +#define SSM_GSPP_1M_BLOCKS @SSM_GSPP_1M_BLOCKS@ +#define SSM_GSPP_TOTAL_SIZE @SSM_GSPP_TOTAL_SIZE@ + +/* Per-User Pool (PUP) - for unprivileged applications */ +#define SSM_PUP_256_BLOCKS @SSM_PUP_256_BLOCKS@ +#define SSM_PUP_512_BLOCKS @SSM_PUP_512_BLOCKS@ +#define SSM_PUP_1K_BLOCKS @SSM_PUP_1K_BLOCKS@ +#define SSM_PUP_2K_BLOCKS @SSM_PUP_2K_BLOCKS@ +#define SSM_PUP_4K_BLOCKS @SSM_PUP_4K_BLOCKS@ +#define SSM_PUP_16K_BLOCKS @SSM_PUP_16K_BLOCKS@ +#define SSM_PUP_64K_BLOCKS @SSM_PUP_64K_BLOCKS@ +#define SSM_PUP_256K_BLOCKS @SSM_PUP_256K_BLOCKS@ +#define SSM_PUP_1M_BLOCKS @SSM_PUP_1M_BLOCKS@ +#define SSM_PUP_TOTAL_SIZE @SSM_PUP_TOTAL_SIZE@ + +/* Legacy pool blocks (same as GSPP for compatibility) */ +#define SSM_POOL_256_BLOCKS @SSM_POOL_256_BLOCKS@ +#define SSM_POOL_512_BLOCKS @SSM_POOL_512_BLOCKS@ +#define SSM_POOL_1K_BLOCKS @SSM_POOL_1K_BLOCKS@ +#define SSM_POOL_2K_BLOCKS @SSM_POOL_2K_BLOCKS@ +#define SSM_POOL_4K_BLOCKS @SSM_POOL_4K_BLOCKS@ +#define SSM_POOL_16K_BLOCKS @SSM_POOL_16K_BLOCKS@ +#define SSM_POOL_64K_BLOCKS @SSM_POOL_64K_BLOCKS@ +#define SSM_POOL_256K_BLOCKS @SSM_POOL_256K_BLOCKS@ +#define SSM_POOL_1M_BLOCKS @SSM_POOL_1M_BLOCKS@ +#define SSM_POOL_TOTAL_SIZE @SSM_POOL_TOTAL_SIZE@ + +/* Size class configuration */ +#define SSM_POOL_MAX_CLASSES 9 +#define SSM_POOL_SHARDS @SSM_POOL_SHARDS@ +#define SSM_POOL_RECLAIM_AGE_S @SSM_POOL_RECLAIM_AGE_S@ + +/* Internal structures - exposed for testing */ +#ifdef __cplusplus +extern "C" { +#endif + +#include <errno.h> +#include <pthread.h> + +#include <ouroboros/pthread.h> + +static __inline__ void robust_mutex_lock(pthread_mutex_t * mtx) +{ +#ifndef HAVE_ROBUST_MUTEX + pthread_mutex_lock(mtx); +#else + if (pthread_mutex_lock(mtx) == EOWNERDEAD) + pthread_mutex_consistent(mtx); +#endif +} + +static __inline__ int robust_wait(pthread_cond_t * cond, + pthread_mutex_t * mtx, + const struct timespec * abstime) +{ + int ret = __timedwait(cond, mtx, abstime); +#ifdef HAVE_ROBUST_MUTEX + if (ret == EOWNERDEAD) + pthread_mutex_consistent(mtx); +#endif + return ret; +} + +/* Packet buffer structure used by pool, rbuff, and tests */ +struct ssm_pk_buff { + uint32_t next_offset; /* List linkage (pool < 4GB) */ + uint16_t refcount; /* Reference count (app + rtx) */ + pid_t allocator_pid; /* For orphan detection */ + uint32_t size; /* Block size (max 1MB) */ + uint32_t pk_head; /* Head offset into data */ + uint32_t pk_tail; /* Tail offset into data */ + uint32_t off; /* Block offset in pool */ + uint64_t alloc_ts; /* CLOCK_MONOTONIC ns at alloc */ + uint8_t data[]; /* Packet data */ +}; + +/* Size class configuration table */ +struct ssm_size_class_cfg { + size_t size; + size_t blocks; +}; + +struct _ssm_list_head { + uint32_t head_offset; + uint32_t count; +}; + +struct _ssm_shard { + pthread_mutex_t mtx; + pthread_cond_t cond; + struct _ssm_list_head free_list; + size_t free_count; +}; + +struct _ssm_size_class { + struct _ssm_shard shards[SSM_POOL_SHARDS]; + size_t object_size; + size_t pool_start; + size_t pool_size; + size_t object_count; +}; + +struct _ssm_pool_hdr { + pthread_mutex_t mtx; + pthread_cond_t healthy; + pid_t pid; + uint32_t initialized; + void * mapped_addr; + struct _ssm_size_class size_classes[SSM_POOL_MAX_CLASSES]; +}; + +#define SSM_PK_BUFF_TOTALSPACE (SSM_PK_BUFF_HEADSPACE + SSM_PK_BUFF_TAILSPACE) +static __inline__ int select_size_class(struct _ssm_pool_hdr * hdr, + size_t len) +{ + size_t sz; + int i; + + sz = sizeof(struct ssm_pk_buff) + SSM_PK_BUFF_TOTALSPACE + len; + + for (i = 0; i < SSM_POOL_MAX_CLASSES; i++) { + struct _ssm_size_class * sc = &hdr->size_classes[i]; + if (sc->object_size > 0 && sz <= sc->object_size) + return i; + } + + return -1; +} + +#ifdef __cplusplus +} +#endif + +#endif /* OUROBOROS_LIB_SSM_H */ diff --git a/src/lib/ssm/tests/CMakeLists.txt b/src/lib/ssm/tests/CMakeLists.txt new file mode 100644 index 00000000..d622d41c --- /dev/null +++ b/src/lib/ssm/tests/CMakeLists.txt @@ -0,0 +1,21 @@ +get_filename_component(PARENT_PATH ${CMAKE_CURRENT_SOURCE_DIR} DIRECTORY) +get_filename_component(PARENT_DIR ${PARENT_PATH} NAME) + +compute_test_prefix() + +create_test_sourcelist(${PARENT_DIR}_tests test_suite.c + # Add new tests here + pool_test.c + pool_sharding_test.c + rbuff_test.c + flow_set_test.c + ) + +add_executable(${PARENT_DIR}_test ${${PARENT_DIR}_tests}) + +disable_test_logging_for_target(${PARENT_DIR}_test) +target_link_libraries(${PARENT_DIR}_test ouroboros-common) + +add_dependencies(build_tests ${PARENT_DIR}_test) + +ouroboros_register_tests(TARGET ${PARENT_DIR}_test TESTS ${${PARENT_DIR}_tests}) diff --git a/src/lib/ssm/tests/flow_set_test.c b/src/lib/ssm/tests/flow_set_test.c new file mode 100644 index 00000000..1bb97b2f --- /dev/null +++ b/src/lib/ssm/tests/flow_set_test.c @@ -0,0 +1,255 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Test of the SSM flow set + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#if defined(__linux__) || defined(__CYGWIN__) +#define _DEFAULT_SOURCE +#else +#define _POSIX_C_SOURCE 200112L +#endif + +#include "config.h" +#include "ssm.h" + +#include <test/test.h> +#include <ouroboros/ssm_flow_set.h> +#include <ouroboros/errno.h> +#include <ouroboros/time.h> + +#include <stdio.h> +#include <unistd.h> +#include <pthread.h> + +static int test_ssm_flow_set_create_destroy(void) +{ + struct ssm_flow_set * set; + pid_t pid; + + TEST_START(); + + pid = getpid(); + + set = ssm_flow_set_create(pid); + if (set == NULL) { + printf("Failed to create flow set.\n"); + goto fail; + } + + ssm_flow_set_destroy(set); + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; +fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_ssm_flow_set_add_del_has(void) +{ + struct ssm_flow_set * set; + pid_t pid; + size_t idx = 0; + int flow_id = 42; + + TEST_START(); + + pid = getpid(); + + set = ssm_flow_set_create(pid); + if (set == NULL) { + printf("Failed to create flow set.\n"); + goto fail; + } + + if (ssm_flow_set_has(set, idx, flow_id)) { + printf("Flow should not be in set initially.\n"); + goto fail_destroy; + } + + if (ssm_flow_set_add(set, idx, flow_id) < 0) { + printf("Failed to add flow to set.\n"); + goto fail_destroy; + } + + if (!ssm_flow_set_has(set, idx, flow_id)) { + printf("Flow should be in set after add.\n"); + goto fail_destroy; + } + + /* Adding same flow again should fail */ + if (ssm_flow_set_add(set, idx, flow_id) != -EPERM) { + printf("Should not be able to add flow twice.\n"); + goto fail_destroy; + } + + ssm_flow_set_del(set, idx, flow_id); + + if (ssm_flow_set_has(set, idx, flow_id)) { + printf("Flow should not be in set after delete.\n"); + goto fail_destroy; + } + + ssm_flow_set_destroy(set); + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; +fail_destroy: + ssm_flow_set_destroy(set); +fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_ssm_flow_set_zero(void) +{ + struct ssm_flow_set * set; + pid_t pid; + size_t idx = 0; + int flow_id1 = 10; + int flow_id2 = 20; + + TEST_START(); + + pid = getpid(); + + set = ssm_flow_set_create(pid); + if (set == NULL) { + printf("Failed to create flow set.\n"); + goto fail; + } + + if (ssm_flow_set_add(set, idx, flow_id1) < 0) { + printf("Failed to add flow1 to set.\n"); + goto fail_destroy; + } + + if (ssm_flow_set_add(set, idx, flow_id2) < 0) { + printf("Failed to add flow2 to set.\n"); + goto fail_destroy; + } + + ssm_flow_set_zero(set, idx); + + if (ssm_flow_set_has(set, idx, flow_id1)) { + printf("Flow1 should not be in set after zero.\n"); + goto fail_destroy; + } + + if (ssm_flow_set_has(set, idx, flow_id2)) { + printf("Flow2 should not be in set after zero.\n"); + goto fail_destroy; + } + + ssm_flow_set_destroy(set); + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; +fail_destroy: + ssm_flow_set_destroy(set); +fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_ssm_flow_set_notify_wait(void) +{ + struct ssm_flow_set * set; + pid_t pid; + size_t idx = 0; + int flow_id = 100; + struct flowevent events[SSM_RBUFF_SIZE]; + struct timespec timeout; + ssize_t ret; + + TEST_START(); + + pid = getpid(); + + set = ssm_flow_set_create(pid); + if (set == NULL) { + printf("Failed to create flow set.\n"); + goto fail; + } + + if (ssm_flow_set_add(set, idx, flow_id) < 0) { + printf("Failed to add flow to set.\n"); + goto fail_destroy; + } + + /* Test immediate timeout when no events */ + clock_gettime(PTHREAD_COND_CLOCK, &timeout); + ret = ssm_flow_set_wait(set, idx, events, &timeout); + if (ret != -ETIMEDOUT) { + printf("Wait should timeout immediately when no events.\n"); + goto fail_destroy; + } + + /* Notify an event */ + ssm_flow_set_notify(set, flow_id, FLOW_PKT); + + /* Should be able to read the event immediately */ + clock_gettime(PTHREAD_COND_CLOCK, &timeout); + ts_add(&timeout, &timeout, &((struct timespec) {1, 0})); + + ret = ssm_flow_set_wait(set, idx, events, &timeout); + if (ret != 1) { + printf("Wait should return 1 event, got %zd.\n", ret); + goto fail_destroy; + } + + if (events[0].flow_id != flow_id) { + printf("Event flow_id mismatch: expected %d, got %d.\n", + flow_id, events[0].flow_id); + goto fail_destroy; + } + + if (events[0].event != FLOW_PKT) { + printf("Event type mismatch: expected %d, got %d.\n", + FLOW_PKT, events[0].event); + goto fail_destroy; + } + + ssm_flow_set_destroy(set); + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; +fail_destroy: + ssm_flow_set_destroy(set); +fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +int flow_set_test(int argc, + char ** argv) +{ + int ret = 0; + + (void) argc; + (void) argv; + + ret |= test_ssm_flow_set_create_destroy(); + ret |= test_ssm_flow_set_add_del_has(); + ret |= test_ssm_flow_set_zero(); + ret |= test_ssm_flow_set_notify_wait(); + + return ret; +} diff --git a/src/lib/ssm/tests/pool_sharding_test.c b/src/lib/ssm/tests/pool_sharding_test.c new file mode 100644 index 00000000..ec464a92 --- /dev/null +++ b/src/lib/ssm/tests/pool_sharding_test.c @@ -0,0 +1,474 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Test of the SSM pool sharding with fallback + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#if defined(__linux__) || defined(__CYGWIN__) +#define _DEFAULT_SOURCE +#else +#define _POSIX_C_SOURCE 200112L +#endif + +#include "config.h" +#include "ssm.h" + +#include <test/test.h> +#include <ouroboros/ssm_pool.h> + +#include <errno.h> +#include <stdio.h> +#include <stdlib.h> +#include <string.h> +#include <unistd.h> +#include <stdbool.h> +#include <sys/wait.h> +#include <sys/types.h> +#include <signal.h> + +#define TEST_SIZE 256 + +/* Helper to get pool header for inspection */ +static struct _ssm_pool_hdr * get_pool_hdr(struct ssm_pool * pool) +{ + /* ssm_pool is opaque, but we know its layout: + * uint8_t * shm_base + * struct _ssm_pool_hdr * hdr + * void * pool_base + */ + struct _ssm_pool_hdr ** hdr_ptr = + (struct _ssm_pool_hdr **)((uint8_t *)pool + sizeof(void *)); + + return *hdr_ptr; +} + +static int test_lazy_distribution(void) +{ + struct ssm_pool * pool; + struct _ssm_pool_hdr * hdr; + struct _ssm_size_class * sc; + int i; + int sc_idx; + + TEST_START(); + + pool = ssm_pool_create(getuid(), getgid()); + if (pool == NULL) { + printf("Failed to create pool.\n"); + goto fail; + } + + hdr = get_pool_hdr(pool); + if (hdr == NULL) { + printf("Failed to get pool header.\n"); + goto fail_pool; + } + + /* Inspect the class that TEST_SIZE allocations will use */ + sc_idx = select_size_class(hdr, TEST_SIZE); + if (sc_idx < 0) { + printf("No size class fits TEST_SIZE=%d.\n", TEST_SIZE); + for (i = 0; i < SSM_POOL_MAX_CLASSES; i++) { + printf(" Class %d: object_size=%zu count=%zu\n", i, + hdr->size_classes[i].object_size, + hdr->size_classes[i].object_count); + } + goto fail_pool; + } + + sc = &hdr->size_classes[sc_idx]; + + /* Verify all blocks start in shard 0 */ + if (sc->shards[0].free_count == 0) { + printf("Shard 0 should have all blocks initially.\n"); + goto fail_pool; + } + + /* Verify other shards are empty */ + for (i = 1; i < SSM_POOL_SHARDS; i++) { + if (sc->shards[i].free_count != 0) { + printf("Shard %d should be empty, has %zu.\n", + i, sc->shards[i].free_count); + goto fail_pool; + } + } + + ssm_pool_destroy(pool); + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; + + fail_pool: + ssm_pool_destroy(pool); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_shard_migration(void) +{ + struct ssm_pool * pool; + struct _ssm_pool_hdr * hdr; + struct _ssm_size_class * sc; + struct ssm_pk_buff * spb; + uint8_t * ptr; + ssize_t off; + int shard_idx; + int sc_idx; + + TEST_START(); + + pool = ssm_pool_create(getuid(), getgid()); + if (pool == NULL) { + printf("Failed to create pool.\n"); + goto fail; + } + + hdr = get_pool_hdr(pool); + + /* Inspect the class that TEST_SIZE allocations will use */ + sc_idx = select_size_class(hdr, TEST_SIZE); + if (sc_idx < 0) { + printf("No size class fits TEST_SIZE=%d.\n", TEST_SIZE); + goto fail_pool; + } + + sc = &hdr->size_classes[sc_idx]; + + /* Allocate from this process */ + off = ssm_pool_alloc(pool, TEST_SIZE, &ptr, &spb); + if (off < 0) { + printf("Allocation failed: %zd.\n", off); + goto fail_pool; + } + + /* Free it - should go to this process's shard */ + shard_idx = getpid() % SSM_POOL_SHARDS; + if (ssm_pool_remove(pool, off) != 0) { + printf("Remove failed.\n"); + goto fail_pool; + } + + /* Verify block migrated away from shard 0 or in allocator's shard */ + if (sc->shards[shard_idx].free_count == 0 && + sc->shards[0].free_count == 0) { + printf("Block should have been freed to a shard.\n"); + goto fail_pool; + } + + ssm_pool_destroy(pool); + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; + + fail_pool: + ssm_pool_destroy(pool); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_fallback_stealing(void) +{ + struct ssm_pool * pool; + struct _ssm_pool_hdr * hdr; + struct _ssm_size_class * sc; + struct ssm_pk_buff ** spbs; + uint8_t ** ptrs; + size_t total_blocks; + size_t total_free; + size_t i; + int sc_idx; + + TEST_START(); + + pool = ssm_pool_create(getuid(), getgid()); + if (pool == NULL) { + printf("Failed to create pool.\n"); + goto fail; + } + + hdr = get_pool_hdr(pool); + + /* Inspect the class that TEST_SIZE allocations will use */ + sc_idx = select_size_class(hdr, TEST_SIZE); + if (sc_idx < 0) { + printf("No size class fits TEST_SIZE=%d.\n", TEST_SIZE); + goto fail_pool; + } + + sc = &hdr->size_classes[sc_idx]; + total_blocks = sc->object_count; + + spbs = malloc(total_blocks * sizeof(struct ssm_pk_buff *)); + ptrs = malloc(total_blocks * sizeof(uint8_t *)); + if (spbs == NULL || ptrs == NULL) { + printf("Failed to allocate test arrays.\n"); + free(spbs); + free(ptrs); + goto fail_pool; + } + + /* Allocate half the blocks from single process */ + for (i = 0; i < total_blocks / 2; i++) { + ssize_t off = ssm_pool_alloc(pool, TEST_SIZE, + &ptrs[i], &spbs[i]); + if (off < 0) { + printf("Allocation %zu failed: %zd.\n", i, off); + free(spbs); + free(ptrs); + goto fail_pool; + } + } + + /* Free them all - they go to local_shard */ + for (i = 0; i < total_blocks / 2; i++) { + size_t off = ssm_pk_buff_get_off(spbs[i]); + if (ssm_pool_remove(pool, off) != 0) { + printf("Remove %zu failed.\n", i); + free(spbs); + free(ptrs); + goto fail_pool; + } + } + + /* Freed blocks should be in shards (all blocks free again) */ + total_free = 0; + for (i = 0; i < SSM_POOL_SHARDS; i++) { + total_free += sc->shards[i].free_count; + } + + if (total_free != total_blocks) { + printf("Expected %zu free blocks total, got %zu.\n", + total_blocks, total_free); + free(spbs); + free(ptrs); + goto fail_pool; + } + + /* Allocate again - should succeed by taking from shards */ + for (i = 0; i < total_blocks / 2; i++) { + ssize_t off = ssm_pool_alloc(pool, TEST_SIZE, + &ptrs[i], &spbs[i]); + if (off < 0) { + printf("Fallback alloc %zu failed: %zd.\n", i, off); + free(spbs); + free(ptrs); + goto fail_pool; + } + } + + /* Now all allocated blocks are in use again */ + /* Cleanup - free all allocated blocks */ + for (i = 0; i < total_blocks / 2; i++) { + size_t off = ssm_pk_buff_get_off(spbs[i]); + ssm_pool_remove(pool, off); + } + + free(spbs); + free(ptrs); + ssm_pool_destroy(pool); + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; + + fail_pool: + ssm_pool_destroy(pool); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_multiprocess_sharding(void) +{ + struct ssm_pool * pool; + struct _ssm_pool_hdr * hdr; + struct _ssm_size_class * sc; + pid_t children[SSM_POOL_SHARDS]; + int i; + int status; + + TEST_START(); + + for (i = 0; i < SSM_POOL_SHARDS; i++) + children[i] = -1; + + pool = ssm_pool_create(getuid(), getgid()); + if (pool == NULL) { + printf("Failed to create pool.\n"); + goto fail; + } + + /* Fork processes to test different shards */ + for (i = 0; i < SSM_POOL_SHARDS; i++) { + children[i] = fork(); + if (children[i] == -1) { + printf("Fork %d failed.\n", i); + goto fail_children; + } + + if (children[i] == 0) { + /* Child process */ + struct ssm_pool * child_pool; + struct ssm_pk_buff * spb; + uint8_t * ptr; + ssize_t off; + int my_shard; + + child_pool = ssm_pool_open(getuid()); + if (child_pool == NULL) + exit(EXIT_FAILURE); + + my_shard = getpid() % SSM_POOL_SHARDS; + (void) my_shard; /* Reserved for future use */ + + /* Each child allocates and frees a block */ + off = ssm_pool_alloc(child_pool, TEST_SIZE, + &ptr, &spb); + if (off < 0) { + ssm_pool_close(child_pool); + exit(EXIT_FAILURE); + } + + /* Small delay to ensure allocation visible */ + usleep(10000); + + if (ssm_pool_remove(child_pool, off) != 0) { + ssm_pool_close(child_pool); + exit(EXIT_FAILURE); + } + + ssm_pool_close(child_pool); + exit(EXIT_SUCCESS); + } + } + + /* Wait for all children */ + for (i = 0; i < SSM_POOL_SHARDS; i++) { + if (waitpid(children[i], &status, 0) == -1) { + printf("Waitpid %d failed.\n", i); + goto fail_children; + } + if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) { + printf("Child %d failed.\n", i); + goto fail_pool; + } + } + + /* Verify blocks distributed across shards */ + hdr = get_pool_hdr(pool); + + /* Inspect the class that TEST_SIZE allocations used */ + i = select_size_class(hdr, TEST_SIZE); + if (i < 0) { + printf("No size class fits TEST_SIZE=%d.\n", TEST_SIZE); + goto fail_pool; + } + + sc = &hdr->size_classes[i]; + + /* After children allocate and free, blocks should be in shards + * (though exact distribution depends on PID values) + */ + for (i = 0; i < SSM_POOL_SHARDS; i++) { + /* At least some shards should have blocks */ + if (sc->shards[i].free_count > 0) { + break; + } + } + + ssm_pool_destroy(pool); + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; + + fail_children: + /* Kill any remaining children */ + for (i = 0; i < SSM_POOL_SHARDS; i++) { + if (children[i] > 0) + kill(children[i], SIGKILL); + } + fail_pool: + ssm_pool_destroy(pool); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_exhaustion_with_fallback(void) +{ + struct ssm_pool * pool; + struct ssm_pk_buff * spb; + uint8_t * ptr; + ssize_t off; + + TEST_START(); + + pool = ssm_pool_create(getuid(), getgid()); + if (pool == NULL) { + printf("Failed to create pool.\n"); + goto fail; + } + + /* Allocate until exhausted across all shards */ + while (true) { + off = ssm_pool_alloc(pool, TEST_SIZE, &ptr, &spb); + if (off < 0) { + if (off == -EAGAIN) + break; + printf("Unexpected error: %zd.\n", off); + goto fail_pool; + } + } + + /* Should fail with -EAGAIN when truly exhausted */ + off = ssm_pool_alloc(pool, TEST_SIZE, &ptr, &spb); + if (off != -EAGAIN) { + printf("Expected -EAGAIN, got %zd.\n", off); + goto fail_pool; + } + + ssm_pool_destroy(pool); + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; + + fail_pool: + ssm_pool_destroy(pool); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +int pool_sharding_test(int argc, + char ** argv) +{ + int ret = 0; + + (void) argc; + (void) argv; + + ret |= test_lazy_distribution(); + ret |= test_shard_migration(); + ret |= test_fallback_stealing(); + ret |= test_multiprocess_sharding(); + ret |= test_exhaustion_with_fallback(); + + return ret; +} diff --git a/src/lib/ssm/tests/pool_test.c b/src/lib/ssm/tests/pool_test.c new file mode 100644 index 00000000..f86fbd9e --- /dev/null +++ b/src/lib/ssm/tests/pool_test.c @@ -0,0 +1,1068 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Test of the Secure Shared Memory (SSM) system + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#define _POSIX_C_SOURCE 200809L + +#include "config.h" +#include "ssm.h" + +#include <test/test.h> +#include <ouroboros/ssm_pool.h> +#include <ouroboros/ssm_rbuff.h> + +#include <errno.h> +#include <stdio.h> +#include <stdlib.h> +#include <string.h> +#include <unistd.h> +#include <stdbool.h> +#include <stdatomic.h> +#include <sys/wait.h> +#include <sys/types.h> +#include <signal.h> +#include <time.h> + +#define POOL_256 256 +#define POOL_512 512 +#define POOL_1K 1024 +#define POOL_2K 2048 +#define POOL_4K 4096 +#define POOL_16K 16384 +#define POOL_64K 65536 +#define POOL_256K 262144 +#define POOL_1M 1048576 +#define POOL_2M (2 * 1024 * 1024) + +static int test_ssm_pool_basic_allocation(void) +{ + struct ssm_pool * pool; + uint8_t * ptr; + struct ssm_pk_buff * spb; + ssize_t ret; + + TEST_START(); + + pool = ssm_pool_create(getuid(), getgid()); + if (pool == NULL) { + printf("Failed to create pool.\n"); + goto fail_create; + } + ret = ssm_pool_alloc(pool, POOL_256, &ptr, &spb); + if (ret < 0) { + printf("Alloc failed: %zd.\n", ret); + goto fail_alloc; + } + + if (spb == NULL) { + printf("Spb is NULL.\n"); + goto fail_alloc; + } + + if (ptr == NULL) { + printf("Ptr is NULL.\n"); + goto fail_alloc; + } + + if (ssm_pk_buff_len(spb) != POOL_256) { + printf("Bad length: %zu.\n", ssm_pk_buff_len(spb)); + goto fail_alloc; + } + + ret = ssm_pool_remove(pool, ret); + if (ret != 0) { + printf("Remove failed: %zd.\n", ret); + goto fail_alloc; + } + + ssm_pool_destroy(pool); + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; + + fail_alloc: + ssm_pool_destroy(pool); + fail_create: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_ssm_pool_multiple_allocations(void) +{ + struct ssm_pool * pool; + uint8_t * ptr1; + uint8_t * ptr2; + uint8_t * ptr3; + struct ssm_pk_buff * spb1; + struct ssm_pk_buff * spb2; + struct ssm_pk_buff * spb3; + ssize_t ret1; + ssize_t ret2; + ssize_t ret3; + + TEST_START(); + + pool = ssm_pool_create(getuid(), getgid()); + if (pool == NULL) { + printf("Failed to create pool.\n"); + goto fail_create; + } + + ret1 = ssm_pool_alloc(pool, POOL_256, &ptr1, &spb1); + ret2 = ssm_pool_alloc(pool, POOL_256, &ptr2, &spb2); + ret3 = ssm_pool_alloc(pool, POOL_256, &ptr3, &spb3); + if (ret1 < 0 || ret2 < 0 || ret3 < 0) { + printf("Allocs failed: %zd, %zd, %zd.\n", ret1, ret2, ret3); + goto fail_alloc; + } + + if (spb1 == NULL) { + printf("Spb1 is NULL.\n"); + goto fail_alloc; + } + + if (ptr1 == NULL) { + printf("Ptr1 is NULL.\n"); + goto fail_alloc; + } + + if (spb2 == NULL) { + printf("Spb2 is NULL.\n"); + goto fail_alloc; + } + + if (ptr2 == NULL) { + printf("Ptr2 is NULL.\n"); + goto fail_alloc; + } + + if (spb3 == NULL) { + printf("Spb3 is NULL.\n"); + goto fail_alloc; + } + + if (ptr3 == NULL) { + printf("Ptr3 is NULL.\n"); + goto fail_alloc; + } + + if (ssm_pk_buff_len(spb1) != POOL_256) { + printf("Bad length spb1: %zu.\n", ssm_pk_buff_len(spb1)); + goto fail_alloc; + } + + if (ssm_pk_buff_len(spb2) != POOL_256) { + printf("Bad length spb2: %zu.\n", ssm_pk_buff_len(spb2)); + goto fail_alloc; + } + + if (ssm_pk_buff_len(spb3) != POOL_256) { + printf("Bad length spb3: %zu.\n", ssm_pk_buff_len(spb3)); + goto fail_alloc; + } + + if (ssm_pool_remove(pool, ret2) != 0) { + printf("Remove ret2 failed.\n"); + goto fail_alloc; + } + + if (ssm_pool_remove(pool, ret1) != 0) { + printf("Remove ret1 failed.\n"); + goto fail_alloc; + } + + if (ssm_pool_remove(pool, ret3) != 0) { + printf("Remove ret3 failed.\n"); + goto fail_alloc; + } + + ssm_pool_destroy(pool); + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; + + fail_alloc: + ssm_pool_destroy(pool); + fail_create: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_ssm_pool_no_fallback_for_large(void) +{ + struct ssm_pool * pool; + uint8_t * ptr; + struct ssm_pk_buff * spb; + ssize_t ret; + + TEST_START(); + + pool = ssm_pool_create(getuid(), getgid()); + if (pool == NULL) { + printf("Failed to create pool.\n"); + goto fail_create; + } + + ret = ssm_pool_alloc(pool, POOL_2M, &ptr, &spb); + if (ret >= 0) { + printf("Oversized alloc succeeded: %zd.\n", ret); + goto fail_alloc; + } + + if (ret != -EMSGSIZE) { + printf("Wrong error: %zd.\n", ret); + goto fail_alloc; + } + + ssm_pool_destroy(pool); + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; + + fail_alloc: + ssm_pool_destroy(pool); + fail_create: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_ssm_pool_blocking_vs_nonblocking(void) +{ + struct ssm_pool * pool; + uint8_t * ptr; + struct ssm_pk_buff * spb; + ssize_t ret; + + TEST_START(); + + pool = ssm_pool_create(getuid(), getgid()); + if (pool == NULL) { + printf("Failed to create pool.\n"); + goto fail_create; + } + + ret = ssm_pool_alloc(pool, POOL_2M, &ptr, &spb); + if (ret != -EMSGSIZE) { + printf("Nonblocking oversized: %zd.\n", ret); + goto fail_alloc; + } + + ret = ssm_pool_alloc_b(pool, POOL_2M, &ptr, &spb, NULL); + if (ret != -EMSGSIZE) { + printf("Blocking oversized: %zd.\n", ret); + goto fail_alloc; + } + + ret = ssm_pool_alloc(pool, POOL_256, &ptr, &spb); + if (ret < 0) { + printf("Valid alloc failed: %zd.\n", ret); + goto fail_alloc; + } + + ssm_pool_remove(pool, ret); + ssm_pool_destroy(pool); + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; + + fail_alloc: + ssm_pool_destroy(pool); + fail_create: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_ssm_pool_stress_test(void) +{ + struct ssm_pool * pool; + uint8_t * ptr; + struct ssm_pk_buff * spb; + ssize_t * indices = NULL; + ssize_t ret; + size_t count = 0; + size_t i; + + TEST_START(); + + pool = ssm_pool_create(getuid(), getgid()); + if (pool == NULL) { + printf("Failed to create pool.\n"); + goto fail_create; + } + + indices = malloc(100 * sizeof(*indices)); + if (indices == NULL) { + printf("Malloc failed.\n"); + goto fail_alloc; + } + + for (i = 0; i < 50; i++) { + size_t j; + size_t num; + size_t size; + + num = (i % 50) + 1; + + for (j = 0; j < num && count < 50; j++) { + switch (i % 4) { + case 0: + /* FALLTHRU */ + case 1: + size = POOL_256; + break; + case 2: + /* FALLTHRU */ + case 3: + size = POOL_1K; + break; + default: + size = POOL_256; + break; + } + + ret = ssm_pool_alloc(pool, size, &ptr, &spb); + if (ret < 0) { + printf("Alloc at iter %zu: %zd.\n", i, ret); + goto fail_test; + } + indices[count++] = ret; + } + + for (j = 0; j < count / 2; j++) { + size_t idx = j * 2; + if (idx < count) { + ret = ssm_pool_remove(pool, indices[idx]); + if (ret != 0) { + printf("Remove at iter %zu: %zd.\n", + i, ret); + goto fail_test; + } + memmove(&indices[idx], &indices[idx + 1], + (count - idx - 1) * sizeof(*indices)); + count--; + } + } + + if (i % 10 == 0) { + ret = ssm_pool_alloc(pool, POOL_256, &ptr, &spb); + if (ret < 0) { + printf("Periodic alloc at %zu: %zd.\n", i, ret); + goto fail_test; + } + ssm_pool_remove(pool, ret); + } + } + + for (i = 0; i < count; i++) + ssm_pool_remove(pool, indices[i]); + + free(indices); + ssm_pool_destroy(pool); + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; + + fail_test: + for (i = 0; i < count; i++) + ssm_pool_remove(pool, indices[i]); + free(indices); + fail_alloc: + ssm_pool_destroy(pool); + fail_create: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_ssm_pool_open_initializes_ssm(void) +{ + struct ssm_pool * creator; + struct ssm_pool * opener; + uint8_t * ptr; + struct ssm_pk_buff * spb; + ssize_t ret; + + TEST_START(); + + creator = ssm_pool_create(getuid(), getgid()); + if (creator == NULL) { + printf("Failed to create pool.\n"); + goto fail_create; + } + + ret = ssm_pool_alloc(creator, POOL_256, &ptr, &spb); + if (ret < 0) { + printf("Creator alloc failed: %zd.\n", ret); + goto fail_creator; + } + ssm_pool_remove(creator, ret); + + opener = ssm_pool_open(getuid()); + if (opener == NULL) { + printf("Open failed.\n"); + goto fail_creator; + } + + ret = ssm_pool_alloc(opener, POOL_256, &ptr, &spb); + if (ret < 0) { + printf("Opener alloc failed: %zd.\n", ret); + goto fail_opener; + } + + ssm_pool_remove(opener, ret); + ssm_pool_close(opener); + ssm_pool_destroy(creator); + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; + + fail_opener: + ssm_pool_close(opener); + fail_creator: + ssm_pool_destroy(creator); + fail_create: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_ssm_pool_bounds_checking(void) +{ + struct ssm_pool * pool; + struct ssm_pk_buff * spb; + ssize_t ret; + + TEST_START(); + + pool = ssm_pool_create(getuid(), getgid()); + if (pool == NULL) { + printf("Failed to create pool.\n"); + goto fail_create; + } + + ret = ssm_pool_alloc(pool, POOL_256, NULL, &spb); + if (ret < 0) { + printf("alloc failed: %zd.\n", ret); + goto fail_alloc; + } + + spb = ssm_pool_get(pool, 0); + if (spb != NULL) { + printf("Get at offset 0.\n"); + goto fail_alloc; + } + + spb = ssm_pool_get(pool, 100000000UL); + if (spb != NULL) { + printf("Get beyond pool.\n"); + goto fail_alloc; + } + + ret = ssm_pool_remove(pool, 0); + if (ret != -EINVAL) { + printf("Remove at offset 0: %zd.\n", ret); + goto fail_alloc; + } + + ret = ssm_pool_remove(pool, 100000000UL); + if (ret != -EINVAL) { + printf("Remove beyond pool: %zd.\n", ret); + goto fail_alloc; + } + + ssm_pool_destroy(pool); + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; + + fail_alloc: + ssm_pool_destroy(pool); + fail_create: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_ssm_pool_inter_process_communication(void) +{ + struct ssm_pool * pool; + struct ssm_rbuff * rb; + struct ssm_pk_buff * spb; + uint8_t * ptr; + uint8_t * data; + const char * msg = "inter-process test"; + size_t len; + ssize_t idx; + pid_t pid; + int status; + + TEST_START(); + + len = strlen(msg) + 1; + + pool = ssm_pool_create(getuid(), getgid()); + if (pool == NULL) { + printf("Failed to create pool.\n"); + goto fail_create; + } + + rb = ssm_rbuff_create(getpid(), 1); + if (rb == NULL) { + printf("Rbuff create failed.\n"); + goto fail_pool; + } + + pid = fork(); + if (pid < 0) { + printf("Fork failed.\n"); + goto fail_rbuff; + } + + if (pid == 0) { + idx = ssm_rbuff_read_b(rb, NULL); + if (idx < 0) { + printf("Child: rbuff read: %zd.\n", idx); + exit(1); + } + + spb = ssm_pool_get(pool, idx); + if (spb == NULL) { + printf("Child: pool get failed.\n"); + exit(1); + } + + data = ssm_pk_buff_head(spb); + if (data == NULL) { + printf("Child: data is NULL.\n"); + ssm_pool_remove(pool, idx); + exit(1); + } + + if (strcmp((char *)data, msg) != 0) { + printf("Child: data mismatch.\n"); + ssm_pool_remove(pool, idx); + exit(1); + } + + ssm_pool_remove(pool, idx); + exit(0); + } + + idx = ssm_pool_alloc(pool, len, &ptr, &spb); + if (idx < 0) { + printf("Parent: pool alloc: %zd.\n", idx); + goto fail_child; + } + + memcpy(ptr, msg, len); + + if (ssm_rbuff_write(rb, idx) < 0) { + printf("Parent: rbuff write failed.\n"); + ssm_pool_remove(pool, idx); + goto fail_child; + } + + if (waitpid(pid, &status, 0) < 0) { + printf("Parent: waitpid failed.\n"); + ssm_pool_remove(pool, idx); + goto fail_rbuff; + } + + if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) { + printf("Child failed.\n"); + ssm_pool_remove(pool, idx); + goto fail_rbuff; + } + + ssm_rbuff_destroy(rb); + ssm_pool_destroy(pool); + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; + + fail_child: + waitpid(pid, &status, 0); + fail_rbuff: + ssm_rbuff_destroy(rb); + fail_pool: + ssm_pool_destroy(pool); + fail_create: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_ssm_pool_read_operation(void) +{ + struct ssm_pool * pool; + struct ssm_pk_buff * spb; + uint8_t * wptr; + uint8_t * rptr; + const char * data = "ssm_pool_read test"; + size_t len; + ssize_t idx; + ssize_t ret; + + TEST_START(); + + len = strlen(data) + 1; + + pool = ssm_pool_create(getuid(), getgid()); + if (pool == NULL) { + printf("Failed to create pool.\n"); + goto fail_create; + } + + idx = ssm_pool_alloc(pool, len, &wptr, &spb); + if (idx < 0) { + printf("alloc failed: %zd.\n", idx); + goto fail_alloc; + } + + memcpy(wptr, data, len); + + ret = ssm_pool_read(&rptr, pool, idx); + if (ret < 0) { + printf("Read failed: %zd.\n", ret); + goto fail_read; + } + + if (rptr == NULL) { + printf("NULL pointer.\n"); + goto fail_read; + } + + if (strcmp((char *)rptr, data) != 0) { + printf("Data mismatch.\n"); + goto fail_read; + } + + ssm_pool_remove(pool, idx); + ssm_pool_destroy(pool); + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; + + fail_read: + ssm_pool_remove(pool, idx); + fail_alloc: + ssm_pool_destroy(pool); + fail_create: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_ssm_pool_mlock_operation(void) +{ + struct ssm_pool * pool; + int ret; + + TEST_START(); + + pool = ssm_pool_create(getuid(), getgid()); + if (pool == NULL) { + printf("Failed to create pool.\n"); + goto fail_create; + } + + ret = ssm_pool_mlock(pool); + if (ret < 0) + printf("Mlock failed: %d (may need privileges).\n", ret); + + ssm_pool_destroy(pool); + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; + + fail_create: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_ssm_pk_buff_operations(void) +{ + struct ssm_pool * pool; + struct ssm_pk_buff * spb; + uint8_t * ptr; + uint8_t * head; + uint8_t * tail; + const char * data = "packet buffer test"; + size_t dlen; + size_t len; + ssize_t idx; + + TEST_START(); + + dlen = strlen(data); + + pool = ssm_pool_create(getuid(), getgid()); + if (pool == NULL) { + printf("Failed to create pool.\n"); + goto fail_create; + } + + idx = ssm_pool_alloc(pool, POOL_256, &ptr, &spb); + if (idx < 0) { + printf("alloc failed: %zd.\n", idx); + goto fail_alloc; + } + + head = ssm_pk_buff_head(spb); + if (head != ptr) { + printf("Head mismatch.\n"); + goto fail_ops; + } + + len = ssm_pk_buff_len(spb); + if (len != POOL_256) { + printf("Bad length: %zu.\n", len); + goto fail_ops; + } + + tail = ssm_pk_buff_tail(spb); + if (tail != ptr + len) { + printf("Tail mismatch.\n"); + goto fail_ops; + } + + memcpy(head, data, dlen); + + tail = ssm_pk_buff_push_tail(spb, 32); + if (tail == NULL) { + printf("push_tail failed.\n"); + goto fail_ops; + } + + if (ssm_pk_buff_len(spb) != POOL_256 + 32) { + printf("Length after push_tail: %zu.\n", + ssm_pk_buff_len(spb)); + goto fail_ops; + } + + if (memcmp(head, data, dlen) != 0) { + printf("Data corrupted.\n"); + goto fail_ops; + } + + tail = ssm_pk_buff_pop_tail(spb, 32); + if (tail == NULL) { + printf("pop_tail failed.\n"); + goto fail_ops; + } + + if (ssm_pk_buff_len(spb) != POOL_256) { + printf("Length after pop_tail: %zu.\n", + ssm_pk_buff_len(spb)); + goto fail_ops; + } + + ssm_pool_remove(pool, idx); + ssm_pool_destroy(pool); + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; + + fail_ops: + ssm_pool_remove(pool, idx); + fail_alloc: + ssm_pool_destroy(pool); + fail_create: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +#define OVERHEAD (offsetof(struct ssm_pk_buff, data) + \ + SSM_PK_BUFF_HEADSPACE + SSM_PK_BUFF_TAILSPACE) +static int test_ssm_pool_size_class_boundaries(void) +{ + struct ssm_pool * pool; + struct ssm_pk_buff * spb; + uint8_t * ptr; + size_t sizes[] = { + POOL_512 - OVERHEAD, + POOL_512 - OVERHEAD + 1, + POOL_1K - OVERHEAD, + POOL_1K - OVERHEAD + 1, + POOL_2K - OVERHEAD, + POOL_2K - OVERHEAD + 1, + POOL_4K - OVERHEAD, + POOL_4K - OVERHEAD + 1, + POOL_16K - OVERHEAD, + POOL_16K - OVERHEAD + 1, + POOL_64K - OVERHEAD, + POOL_64K - OVERHEAD + 1, + POOL_256K - OVERHEAD, + }; + size_t expected_classes[] = { + 512, 1024, 1024, 2048, 2048, 4096, 4096, 16384, + 16384, 65536, 65536, 262144, 262144 + }; + size_t i; + ssize_t idx; + + TEST_START(); + + pool = ssm_pool_create(getuid(), getgid()); + if (pool == NULL) { + printf("Failed to create pool.\n"); + goto fail_create; + } + + for (i = 0; i < sizeof(sizes) / sizeof(sizes[0]); i++) { + struct ssm_pk_buff * hdr; + size_t actual_class; + + idx = ssm_pool_alloc(pool, sizes[i], &ptr, &spb); + if (idx < 0) { + printf("Alloc at %zu failed: %zd.\n", sizes[i], idx); + goto fail_alloc; + } + + if (ssm_pk_buff_len(spb) != sizes[i]) { + printf("Length mismatch at %zu: %zu.\n", + sizes[i], ssm_pk_buff_len(spb)); + ssm_pool_remove(pool, idx); + goto fail_alloc; + } + + /* Verify correct size class was used + * hdr->size is the data array size (object_size - header) */ + hdr = spb; + actual_class = hdr->size + offsetof(struct ssm_pk_buff, data); + if (actual_class != expected_classes[i]) { + printf("Wrong class for len=%zu: want %zu, got %zu.\n", + sizes[i], expected_classes[i], actual_class); + ssm_pool_remove(pool, idx); + goto fail_alloc; + } + + memset(ptr, i & 0xFF, sizes[i]); + + ssm_pool_remove(pool, idx); + } + + ssm_pool_destroy(pool); + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; + + fail_alloc: + ssm_pool_destroy(pool); + fail_create: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_ssm_pool_exhaustion(void) +{ + struct ssm_pool * pool; + struct ssm_pk_buff * spb; + uint8_t * ptr; + ssize_t * indices; + size_t count = 0; + size_t i; + ssize_t ret; + + TEST_START(); + + pool = ssm_pool_create(getuid(), getgid()); + if (pool == NULL) { + printf("Failed to create pool.\n"); + goto fail_create; + } + + indices = malloc(2048 * sizeof(*indices)); + if (indices == NULL) { + printf("Malloc failed.\n"); + goto fail_alloc; + } + + for (i = 0; i < 2048; i++) { + ret = ssm_pool_alloc(pool, POOL_256, &ptr, &spb); + if (ret < 0) { + if (ret == -EAGAIN) + break; + printf("Alloc error: %zd.\n", ret); + goto fail_test; + } + indices[count++] = ret; + } + + if (count == 0) { + printf("No allocs succeeded.\n"); + goto fail_test; + } + + ret = ssm_pool_alloc(pool, POOL_256, &ptr, &spb); + if (ret >= 0) { + ssm_pool_remove(pool, ret); + } else if (ret != -EAGAIN) { + printf("Unexpected error: %zd.\n", ret); + goto fail_test; + } + + for (i = 0; i < count; i++) + ssm_pool_remove(pool, indices[i]); + + ret = ssm_pool_alloc(pool, POOL_256, &ptr, &spb); + if (ret < 0) { + printf("Alloc after free failed: %zd.\n", ret); + goto fail_test; + } + ssm_pool_remove(pool, ret); + + free(indices); + ssm_pool_destroy(pool); + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; + + fail_test: + for (i = 0; i < count; i++) + ssm_pool_remove(pool, indices[i]); + free(indices); + fail_alloc: + ssm_pool_destroy(pool); + fail_create: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_ssm_pool_reclaim_orphans(void) +{ + struct ssm_pool * pool; + uint8_t * ptr1; + uint8_t * ptr2; + uint8_t * ptr3; + struct ssm_pk_buff * spb1; + struct ssm_pk_buff * spb2; + struct ssm_pk_buff * spb3; + ssize_t ret1; + ssize_t ret2; + ssize_t ret3; + pid_t my_pid; + pid_t fake_pid = 99999; + struct timespec now; + uint64_t old_ts; + + TEST_START(); + + pool = ssm_pool_create(getuid(), getgid()); + if (pool == NULL) { + printf("Failed to create pool.\n"); + goto fail_create; + } + + my_pid = getpid(); + + /* Allocate some blocks */ + ret1 = ssm_pool_alloc(pool, POOL_256, &ptr1, &spb1); + ret2 = ssm_pool_alloc(pool, POOL_512, &ptr2, &spb2); + ret3 = ssm_pool_alloc(pool, POOL_1K, &ptr3, &spb3); + if (ret1 < 0 || ret2 < 0 || ret3 < 0) { + printf("Allocs failed: %zd, %zd, %zd.\n", ret1, ret2, ret3); + goto fail_alloc; + } + + /* Simulate blocks leaked by a dead process: foreign pid, aged out. */ + clock_gettime(CLOCK_MONOTONIC, &now); + old_ts = ((uint64_t) now.tv_sec - (SSM_POOL_RECLAIM_AGE_S + 1)) + * 1000000000ULL + (uint64_t) now.tv_nsec; + + spb1->allocator_pid = fake_pid; + spb2->allocator_pid = fake_pid; + spb1->alloc_ts = old_ts; + spb2->alloc_ts = old_ts; + /* Keep spb3 with our pid */ + + /* Reclaim orphans from fake_pid */ + ssm_pool_reclaim_orphans(pool, fake_pid); + + /* Verify spb1 and spb2 have refcount 0 (reclaimed) */ + if (spb1->refcount != 0) { + printf("spb1 refcount should be 0, got %u.\n", spb1->refcount); + goto fail_test; + } + + if (spb2->refcount != 0) { + printf("spb2 refcount should be 0, got %u.\n", spb2->refcount); + goto fail_test; + } + + /* Verify spb3 still has refcount 1 (not reclaimed) */ + if (spb3->refcount != 1) { + printf("spb3 refcount should be 1, got %u.\n", spb3->refcount); + goto fail_test; + } + + /* Clean up */ + ssm_pool_remove(pool, ret3); + + /* Try allocating again - should get blocks from reclaimed pool */ + ret1 = ssm_pool_alloc(pool, POOL_256, &ptr1, &spb1); + if (ret1 < 0) { + printf("Alloc after reclaim failed: %zd.\n", ret1); + goto fail_test; + } + + /* Verify new allocation has our pid */ + if (spb1->allocator_pid != my_pid) { + printf("New block has wrong pid: %d vs %d.\n", + spb1->allocator_pid, my_pid); + goto fail_test; + } + + ssm_pool_remove(pool, ret1); + ssm_pool_destroy(pool); + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; + + fail_test: + ssm_pool_remove(pool, ret3); + fail_alloc: + ssm_pool_destroy(pool); + fail_create: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +int pool_test(int argc, + char ** argv) +{ + int ret = 0; + + (void) argc; + (void) argv; + + ret |= test_ssm_pool_basic_allocation(); + ret |= test_ssm_pool_multiple_allocations(); + ret |= test_ssm_pool_no_fallback_for_large(); + ret |= test_ssm_pool_blocking_vs_nonblocking(); + ret |= test_ssm_pool_stress_test(); + ret |= test_ssm_pool_open_initializes_ssm(); + ret |= test_ssm_pool_bounds_checking(); + ret |= test_ssm_pool_inter_process_communication(); + ret |= test_ssm_pool_read_operation(); + ret |= test_ssm_pool_mlock_operation(); + ret |= test_ssm_pk_buff_operations(); + ret |= test_ssm_pool_size_class_boundaries(); + ret |= test_ssm_pool_exhaustion(); + ret |= test_ssm_pool_reclaim_orphans(); + + return ret; +} diff --git a/src/lib/ssm/tests/rbuff_test.c b/src/lib/ssm/tests/rbuff_test.c new file mode 100644 index 00000000..48e5a714 --- /dev/null +++ b/src/lib/ssm/tests/rbuff_test.c @@ -0,0 +1,675 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Test of the SSM notification ring buffer + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#if defined(__linux__) || defined(__CYGWIN__) +#define _DEFAULT_SOURCE +#else +#define _POSIX_C_SOURCE 200112L +#endif + +#include "config.h" +#include "ssm.h" + +#include <test/test.h> +#include <ouroboros/ssm_rbuff.h> +#include <ouroboros/errno.h> +#include <ouroboros/time.h> + +#include <errno.h> +#include <stdio.h> +#include <unistd.h> +#include <pthread.h> + +static int test_ssm_rbuff_create_destroy(void) +{ + struct ssm_rbuff * rb; + + TEST_START(); + + rb = ssm_rbuff_create(getpid(), 1); + if (rb == NULL) { + printf("Failed to create rbuff.\n"); + goto fail; + } + + ssm_rbuff_destroy(rb); + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; + + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_ssm_rbuff_write_read(void) +{ + struct ssm_rbuff * rb; + ssize_t idx; + + TEST_START(); + + rb = ssm_rbuff_create(getpid(), 2); + if (rb == NULL) { + printf("Failed to create rbuff.\n"); + goto fail; + } + + if (ssm_rbuff_write(rb, 42) < 0) { + printf("Failed to write value.\n"); + goto fail_rb; + } + + if (ssm_rbuff_queued(rb) != 1) { + printf("Queue length should be 1, got %zu.\n", + ssm_rbuff_queued(rb)); + goto fail_rb; + } + + idx = ssm_rbuff_read(rb); + if (idx != 42) { + printf("Expected 42, got %zd.\n", idx); + goto fail_rb; + } + + if (ssm_rbuff_queued(rb) != 0) { + printf("Queue should be empty, got %zu.\n", + ssm_rbuff_queued(rb)); + goto fail_rb; + } + + ssm_rbuff_destroy(rb); + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; + + fail_rb: + ssm_rbuff_destroy(rb); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_ssm_rbuff_read_empty(void) +{ + struct ssm_rbuff * rb; + ssize_t ret; + + TEST_START(); + + rb = ssm_rbuff_create(getpid(), 3); + if (rb == NULL) { + printf("Failed to create rbuff.\n"); + goto fail; + } + + ret = ssm_rbuff_read(rb); + if (ret != -EAGAIN) { + printf("Expected -EAGAIN, got %zd.\n", ret); + goto fail_rb; + } + + ssm_rbuff_destroy(rb); + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; + + fail_rb: + ssm_rbuff_destroy(rb); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_ssm_rbuff_fill_drain(void) +{ + struct ssm_rbuff * rb; + size_t i; + ssize_t ret; + + TEST_START(); + + rb = ssm_rbuff_create(getpid(), 4); + if (rb == NULL) { + printf("Failed to create rbuff.\n"); + goto fail; + } + + for (i = 0; i < SSM_RBUFF_SIZE - 1; ++i) { + if (ssm_rbuff_queued(rb) != i) { + printf("Expected %zu queued, got %zu.\n", + i, ssm_rbuff_queued(rb)); + goto fail_rb; + } + if (ssm_rbuff_write(rb, i) < 0) { + printf("Failed to write at index %zu.\n", i); + goto fail_rb; + } + } + + if (ssm_rbuff_queued(rb) != SSM_RBUFF_SIZE - 1) { + printf("Expected %d queued, got %zu.\n", + SSM_RBUFF_SIZE - 1, ssm_rbuff_queued(rb)); + goto fail_rb; + } + + ret = ssm_rbuff_write(rb, 999); + if (ret != -EAGAIN) { + printf("Expected -EAGAIN on full buffer, got %zd.\n", ret); + goto fail_rb; + } + + for (i = 0; i < SSM_RBUFF_SIZE - 1; ++i) { + ret = ssm_rbuff_read(rb); + if (ret != (ssize_t) i) { + printf("Expected %zu, got %zd.\n", i, ret); + goto fail_rb; + } + } + + if (ssm_rbuff_queued(rb) != 0) { + printf("Expected empty queue, got %zu.\n", + ssm_rbuff_queued(rb)); + goto fail_rb; + } + + ssm_rbuff_destroy(rb); + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; + + fail_rb: + while (ssm_rbuff_read(rb) >= 0) + ; + ssm_rbuff_destroy(rb); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_ssm_rbuff_flags(void) +{ + struct ssm_rbuff * rb; + uint32_t flags; + + TEST_START(); + + rb = ssm_rbuff_create(getpid(), 5); + if (rb == NULL) { + printf("Failed to create rbuff.\n"); + goto fail; + } + + flags = ssm_rbuff_get_flags(rb); + if (flags != RB_RDWR) { + printf("Expected RB_RDWR, got %u.\n", flags); + goto fail_rb; + } + + ssm_rbuff_clr_bits(rb, RB_WR); + flags = ssm_rbuff_get_flags(rb); + if (flags != RB_RD) { + printf("Expected RB_RD, got %u.\n", flags); + goto fail_rb; + } + + if (ssm_rbuff_write(rb, 1) != -ENOTALLOC) { + printf("Expected -ENOTALLOC on RDONLY.\n"); + goto fail_rb; + } + + ssm_rbuff_set_bits(rb, RB_FLOWDOWN); + if (ssm_rbuff_write(rb, 1) != -EFLOWDOWN) { + printf("Expected -EFLOWDOWN on FLOWDOWN.\n"); + goto fail_rb; + } + + if (ssm_rbuff_read(rb) != -EFLOWDOWN) { + printf("Expected -EFLOWDOWN on read with FLOWDOWN.\n"); + goto fail_rb; + } + + ssm_rbuff_destroy(rb); + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; + + fail_rb: + ssm_rbuff_destroy(rb); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_ssm_rbuff_open_close(void) +{ + struct ssm_rbuff * rb1; + struct ssm_rbuff * rb2; + pid_t pid; + + TEST_START(); + + pid = getpid(); + + rb1 = ssm_rbuff_create(pid, 6); + if (rb1 == NULL) { + printf("Failed to create rbuff.\n"); + goto fail; + } + + if (ssm_rbuff_write(rb1, 123) < 0) { + printf("Failed to write value.\n"); + goto fail_rb1; + } + + rb2 = ssm_rbuff_open(pid, 6); + if (rb2 == NULL) { + printf("Failed to open existing rbuff.\n"); + goto fail_rb1; + } + + if (ssm_rbuff_queued(rb2) != 1) { + printf("Expected 1 queued in opened rbuff, got %zu.\n", + ssm_rbuff_queued(rb2)); + goto fail_rb2; + } + + if (ssm_rbuff_read(rb2) != 123) { + printf("Failed to read from opened rbuff.\n"); + goto fail_rb2; + } + + ssm_rbuff_close(rb2); + ssm_rbuff_destroy(rb1); + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; + + fail_rb2: + ssm_rbuff_close(rb2); + fail_rb1: + ssm_rbuff_destroy(rb1); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +struct thread_args { + struct ssm_rbuff * rb; + int iterations; + int delay_us; +}; + +static void * writer_thread(void * arg) +{ + struct thread_args * args = (struct thread_args *) arg; + struct timespec delay = {0, 0}; + int i; + + delay.tv_nsec = args->delay_us * 1000L; + + for (i = 0; i < args->iterations; ++i) { + while (ssm_rbuff_write(args->rb, i) < 0) + nanosleep(&delay, NULL); + } + + return NULL; +} + +static void * reader_thread(void * arg) +{ + struct thread_args * args = (struct thread_args *) arg; + struct timespec delay = {0, 0}; + int i; + ssize_t val; + + delay.tv_nsec = args->delay_us * 1000L; + + for (i = 0; i < args->iterations; ++i) { + val = ssm_rbuff_read(args->rb); + while (val < 0) { + nanosleep(&delay, NULL); + val = ssm_rbuff_read(args->rb); + } + if (val != i) { + printf("Expected %d, got %zd.\n", i, val); + return (void *) -1; + } + } + + return NULL; +} + +static void * blocking_writer_thread(void * arg) +{ + struct thread_args * args = (struct thread_args *) arg; + int i; + + for (i = 0; i < args->iterations; ++i) { + if (ssm_rbuff_write_b(args->rb, i, NULL) < 0) + return (void *) -1; + } + + return NULL; +} + +static void * blocking_reader_thread(void * arg) +{ + struct thread_args * args = (struct thread_args *) arg; + int i; + ssize_t val; + + for (i = 0; i < args->iterations; ++i) { + val = ssm_rbuff_read_b(args->rb, NULL); + if (val < 0 || val != i) { + printf("Expected %d, got %zd.\n", i, val); + return (void *) -1; + } + } + + return NULL; +} + +static int test_ssm_rbuff_blocking(void) +{ + struct ssm_rbuff * rb; + pthread_t wthread; + pthread_t rthread; + struct thread_args args; + struct timespec delay = {0, 10 * MILLION}; + void * ret_w; + void * ret_r; + + TEST_START(); + + rb = ssm_rbuff_create(getpid(), 8); + if (rb == NULL) { + printf("Failed to create rbuff.\n"); + goto fail; + } + + args.rb = rb; + args.iterations = 50; + args.delay_us = 0; + + if (pthread_create(&rthread, NULL, blocking_reader_thread, &args)) { + printf("Failed to create reader thread.\n"); + goto fail_rthread; + } + + nanosleep(&delay, NULL); + + if (pthread_create(&wthread, NULL, blocking_writer_thread, &args)) { + printf("Failed to create writer thread.\n"); + pthread_cancel(rthread); + goto fail_wthread; + } + + pthread_join(wthread, &ret_w); + pthread_join(rthread, &ret_r); + + if (ret_w != NULL || ret_r != NULL) { + printf("Thread returned error.\n"); + goto fail_ret; + } + + ssm_rbuff_destroy(rb); + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; + + fail_ret: + fail_wthread: + pthread_join(rthread, NULL); + fail_rthread: + ssm_rbuff_destroy(rb); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_ssm_rbuff_blocking_timeout(void) +{ + struct ssm_rbuff * rb; + struct timespec abs_timeout; + struct timespec interval = {0, 100 * MILLION}; + struct timespec start; + struct timespec end; + ssize_t ret; + long elapsed_ms; + size_t i; + + TEST_START(); + + rb = ssm_rbuff_create(getpid(), 9); + if (rb == NULL) { + printf("Failed to create rbuff.\n"); + goto fail; + } + + clock_gettime(PTHREAD_COND_CLOCK, &start); + ts_add(&start, &interval, &abs_timeout); + + ret = ssm_rbuff_read_b(rb, &abs_timeout); + + clock_gettime(PTHREAD_COND_CLOCK, &end); + + if (ret != -ETIMEDOUT) { + printf("Expected -ETIMEDOUT, got %zd.\n", ret); + goto fail_rb; + } + + elapsed_ms = (end.tv_sec - start.tv_sec) * 1000L + + (end.tv_nsec - start.tv_nsec) / 1000000L; + + if (elapsed_ms < 90 || elapsed_ms > 200) { + printf("Timeout took %ld ms, expected ~100 ms.\n", + elapsed_ms); + goto fail_rb; + } + + for (i = 0; i < SSM_RBUFF_SIZE - 1; ++i) { + if (ssm_rbuff_write(rb, i) < 0) { + printf("Failed to fill buffer.\n"); + goto fail_rb; + } + } + + clock_gettime(PTHREAD_COND_CLOCK, &start); + ts_add(&start, &interval, &abs_timeout); + + ret = ssm_rbuff_write_b(rb, 999, &abs_timeout); + + clock_gettime(PTHREAD_COND_CLOCK, &end); + + if (ret != -ETIMEDOUT) { + printf("Expected -ETIMEDOUT on full buffer, got %zd.\n", + ret); + goto fail_rb; + } + + elapsed_ms = (end.tv_sec - start.tv_sec) * 1000L + + (end.tv_nsec - start.tv_nsec) / 1000000L; + + if (elapsed_ms < 90 || elapsed_ms > 200) { + printf("Write timeout took %ld ms, expected ~100 ms.\n", + elapsed_ms); + goto fail_rb; + } + + while (ssm_rbuff_read(rb) >= 0) + ; + + ssm_rbuff_destroy(rb); + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; + + fail_rb: + while (ssm_rbuff_read(rb) >= 0) + ; + ssm_rbuff_destroy(rb); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_ssm_rbuff_blocking_flowdown(void) +{ + struct ssm_rbuff * rb; + struct timespec abs_timeout; + struct timespec now; + struct timespec interval = {5, 0}; + ssize_t ret; + size_t i; + + TEST_START(); + + rb = ssm_rbuff_create(getpid(), 10); + if (rb == NULL) { + printf("Failed to create rbuff.\n"); + goto fail; + } + + clock_gettime(PTHREAD_COND_CLOCK, &now); + ts_add(&now, &interval, &abs_timeout); + + ssm_rbuff_set_bits(rb, RB_FLOWDOWN); + + ret = ssm_rbuff_read_b(rb, &abs_timeout); + if (ret != -EFLOWDOWN) { + printf("Expected -EFLOWDOWN, got %zd.\n", ret); + goto fail_rb; + } + + ssm_rbuff_clr_bits(rb, RB_FLOWDOWN); + + for (i = 0; i < SSM_RBUFF_SIZE - 1; ++i) { + if (ssm_rbuff_write(rb, i) < 0) { + printf("Failed to fill buffer.\n"); + goto fail_rb; + } + } + + clock_gettime(PTHREAD_COND_CLOCK, &now); + ts_add(&now, &interval, &abs_timeout); + + ssm_rbuff_set_bits(rb, RB_FLOWDOWN); + + ret = ssm_rbuff_write_b(rb, 999, &abs_timeout); + if (ret != -EFLOWDOWN) { + printf("Expected -EFLOWDOWN on write, got %zd.\n", ret); + goto fail_rb; + } + + ssm_rbuff_clr_bits(rb, RB_FLOWDOWN); + while (ssm_rbuff_read(rb) >= 0) + ; + + ssm_rbuff_destroy(rb); + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; + + fail_rb: + while (ssm_rbuff_read(rb) >= 0) + ; + ssm_rbuff_destroy(rb); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_ssm_rbuff_threaded(void) +{ + struct ssm_rbuff * rb; + pthread_t wthread; + pthread_t rthread; + struct thread_args args; + void * ret_w; + void * ret_r; + + TEST_START(); + + rb = ssm_rbuff_create(getpid(), 7); + if (rb == NULL) { + printf("Failed to create rbuff.\n"); + goto fail; + } + + args.rb = rb; + args.iterations = 100; + args.delay_us = 100; + + if (pthread_create(&wthread, NULL, writer_thread, &args)) { + printf("Failed to create writer thread.\n"); + goto fail_rb; + } + + if (pthread_create(&rthread, NULL, reader_thread, &args)) { + printf("Failed to create reader thread.\n"); + pthread_cancel(wthread); + pthread_join(wthread, NULL); + goto fail_rb; + } + + pthread_join(wthread, &ret_w); + pthread_join(rthread, &ret_r); + + if (ret_w != NULL || ret_r != NULL) { + printf("Thread returned error.\n"); + goto fail_rb; + } + + ssm_rbuff_destroy(rb); + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; + + fail_rb: + ssm_rbuff_destroy(rb); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +int rbuff_test(int argc, + char ** argv) +{ + int ret = 0; + + (void) argc; + (void) argv; + + ret |= test_ssm_rbuff_create_destroy(); + ret |= test_ssm_rbuff_write_read(); + ret |= test_ssm_rbuff_read_empty(); + ret |= test_ssm_rbuff_fill_drain(); + ret |= test_ssm_rbuff_flags(); + ret |= test_ssm_rbuff_open_close(); + ret |= test_ssm_rbuff_threaded(); + ret |= test_ssm_rbuff_blocking(); + ret |= test_ssm_rbuff_blocking_timeout(); + ret |= test_ssm_rbuff_blocking_flowdown(); + + return ret; +} diff --git a/src/lib/tests/CMakeLists.txt b/src/lib/tests/CMakeLists.txt index 0e114548..002d94af 100644 --- a/src/lib/tests/CMakeLists.txt +++ b/src/lib/tests/CMakeLists.txt @@ -1,28 +1,36 @@ get_filename_component(PARENT_PATH ${CMAKE_CURRENT_SOURCE_DIR} DIRECTORY) get_filename_component(PARENT_DIR ${PARENT_PATH} NAME) +compute_test_prefix() + create_test_sourcelist(${PARENT_DIR}_tests test_suite.c # Add new tests here + auth_test.c + auth_test_ml_dsa.c + auth_test_slh_dsa.c bitmap_test.c btree_test.c - crc32_test.c + crypt_test.c hash_test.c + kex_test.c + kex_test_ml_kem.c + keyrot_test.c md5_test.c sha3_test.c - shm_rbuff_test.c + sockets_test.c time_test.c + tpm_test.c + tw_test.c ) -add_executable(${PARENT_DIR}_test EXCLUDE_FROM_ALL ${${PARENT_DIR}_tests}) +add_executable(${PARENT_DIR}_test ${${PARENT_DIR}_tests}) -target_link_libraries(${PARENT_DIR}_test ouroboros-common) +target_include_directories(${PARENT_DIR}_test PRIVATE + ${CMAKE_SOURCE_DIR}/src/lib) -add_dependencies(check ${PARENT_DIR}_test) +disable_test_logging_for_target(${PARENT_DIR}_test) +target_link_libraries(${PARENT_DIR}_test ouroboros-common) -set(tests_to_run ${${PARENT_DIR}_tests}) -remove(tests_to_run test_suite.c) +add_dependencies(build_tests ${PARENT_DIR}_test) -foreach (test ${tests_to_run}) - get_filename_component(test_name ${test} NAME_WE) - add_test(${test_name} ${C_TEST_PATH}/${PARENT_DIR}_test ${test_name}) -endforeach (test) +ouroboros_register_tests(TARGET ${PARENT_DIR}_test TESTS ${${PARENT_DIR}_tests}) diff --git a/src/lib/tests/auth_test.c b/src/lib/tests/auth_test.c new file mode 100644 index 00000000..af7cf81c --- /dev/null +++ b/src/lib/tests/auth_test.c @@ -0,0 +1,743 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Test of the authentication functions + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#include "config.h" + +#include <test/test.h> +#include <ouroboros/crypt.h> +#include <ouroboros/name.h> +#include <ouroboros/random.h> +#include <ouroboros/utils.h> + +#include <test/certs/ecdsa.h> + +#include <string.h> + +#define TEST_MSG_SIZE 1500 + +static int test_auth_create_destroy_ctx(void) +{ + struct auth_ctx * ctx; + + TEST_START(); + + ctx = auth_create_ctx(); + if (ctx == NULL) { + printf("Failed to create auth context.\n"); + goto fail_create; + } + + auth_destroy_ctx(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_create: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_load_free_crt(void) +{ + void * crt; + + TEST_START(); + + if (crypt_load_crt_str(root_ca_crt_ec, &crt) < 0) { + printf("Failed to load certificate string.\n"); + goto fail_load; + } + + crypt_free_crt(crt); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_load: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_crypt_get_pubkey_crt(void) +{ + void * pk; + void * crt; + + TEST_START(); + + if (crypt_load_crt_str(signed_server_crt_ec, &crt) < 0) { + printf("Failed to load server certificate from string.\n"); + goto fail_load; + } + + if (crypt_get_pubkey_crt(crt, &pk) < 0) { + printf("Failed to get public key from certificate.\n"); + goto fail_get_pubkey; + } + + crypt_free_key(pk); + crypt_free_crt(crt); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_get_pubkey: + crypt_free_crt(crt); + fail_load: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_check_crt_name(void) +{ + void * crt; + + TEST_START(); + + if (crypt_load_crt_str(signed_server_crt_ec, &crt) < 0) { + printf("Failed to load certificate from string.\n"); + goto fail_load; + } + + if (crypt_check_crt_name(crt, "test-1.unittest.o7s") < 0) { + printf("Failed to verify correct name.\n"); + goto fail_check; + } + + if (crypt_check_crt_name(crt, "bogus.name") == 0) { + printf("Failed to detect incorrect name.\n"); + goto fail_check; + } + + crypt_free_crt(crt); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_check: + crypt_free_crt(crt); + fail_load: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_crt_name_confusion(void) +{ + char name[NAME_SIZE + 1]; + void * crt; + + TEST_START(); + + if (crypt_load_crt_str(confused_crt_ec, &crt) < 0) { + printf("Failed to load name-confusion certificate.\n"); + goto fail_load; + } + + /* Must extract the real CN, not the "CN=" decoy in the O field. */ + if (crypt_get_crt_name(crt, name) < 0) { + printf("Failed to extract name from certificate.\n"); + goto fail_check; + } + + if (strcmp(name, "attacker.unittest.o7s") != 0) { + printf("Extracted '%s', expected real CN.\n", name); + goto fail_check; + } + + /* The decoy name in the O field must never authenticate. */ + if (crypt_check_crt_name(crt, "victim.unittest.o7s") == 0) { + printf("Accepted spoofed name from O field.\n"); + goto fail_check; + } + + crypt_free_crt(crt); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_check: + crypt_free_crt(crt); + fail_load: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_load_free_privkey(void) +{ + void * key; + + TEST_START(); + + if (crypt_load_privkey_str(server_pkp_ec, &key) < 0) { + printf("Failed to load server key pair from string.\n"); + goto fail_load; + } + + crypt_free_key(key); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_load: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_load_free_pubkey(void) +{ + void * key; + + TEST_START(); + + if (crypt_load_pubkey_str(server_pk_ec, &key) < 0) { + printf("Failed to load server public key from string.\n"); + goto fail_load; + } + + crypt_free_key(key); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_load: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_crypt_check_pubkey_crt(void) +{ + void * pk; + void * crt_pk; + void * crt; + + TEST_START(); + + if (crypt_load_crt_str(signed_server_crt_ec, &crt) < 0) { + printf("Failed to load public certificate from string.\n"); + goto fail_crt; + } + + if (crypt_load_pubkey_str(server_pk_ec, &pk) < 0) { + printf("Failed to load public key from string.\n"); + goto fail_pubkey; + } + + if (crypt_get_pubkey_crt(crt, &crt_pk) < 0) { + printf("Failed to get public key from certificate.\n"); + goto fail_get_pubkey; + } + + if (crypt_cmp_key(pk, crt_pk) != 0) { + printf("Public keys do not match .\n"); + goto fail_check; + } + + + crypt_free_key(crt_pk); + crypt_free_key(pk); + crypt_free_crt(crt); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_check: + crypt_free_key(crt_pk); + fail_get_pubkey: + crypt_free_key(pk); + fail_pubkey: + crypt_free_crt(crt); + fail_crt: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_store_add(void) +{ + struct auth_ctx * ctx; + void * _root_ca_crt; + + TEST_START(); + + ctx = auth_create_ctx(); + if (ctx == NULL) { + printf("Failed to create auth context.\n"); + goto fail_create; + } + + if (crypt_load_crt_str(root_ca_crt_ec, &_root_ca_crt) < 0) { + printf("Failed to load root crt from string.\n"); + goto fail_load; + } + + if (auth_add_crt_to_store(ctx, _root_ca_crt) < 0) { + printf("Failed to add root crt to auth store.\n"); + goto fail_add; + } + + crypt_free_crt(_root_ca_crt); + auth_destroy_ctx(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_add: + crypt_free_crt(_root_ca_crt); + fail_load: + crypt_free_crt(_root_ca_crt); + fail_create: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_verify_crt(void) +{ + struct auth_ctx * auth; + void * _server_crt; + void * _signed_server_crt; + void * _root_ca_crt; + void * _im_ca_crt; + + TEST_START(); + + auth = auth_create_ctx(); + if (auth == NULL) { + printf("Failed to create auth context.\n"); + goto fail_create_ctx; + } + + if (crypt_load_crt_str(server_crt_ec, &_server_crt) < 0) { + printf("Failed to load self-signed crt from string.\n"); + goto fail_load_server_crt; + } + + if (crypt_load_crt_str(signed_server_crt_ec, &_signed_server_crt) < 0) { + printf("Failed to load signed crt from string.\n"); + goto fail_load_signed_server_crt; + } + + if (crypt_load_crt_str(root_ca_crt_ec, &_root_ca_crt) < 0) { + printf("Failed to load root crt from string.\n"); + goto fail_load_root_ca_crt; + } + + if (crypt_load_crt_str(im_ca_crt_ec, &_im_ca_crt) < 0) { + printf("Failed to load intermediate crt from string.\n"); + goto fail_load_im_ca_crt; + } + + if (auth_add_crt_to_store(auth, _root_ca_crt) < 0) { + printf("Failed to add root ca crt to auth store.\n"); + goto fail_verify; + } + + if (auth_add_crt_to_store(auth, _im_ca_crt) < 0) { + printf("Failed to add intermediate ca crt to auth store.\n"); + goto fail_verify; + } + + if (auth_verify_crt(auth, _signed_server_crt) < 0) { + printf("Failed to verify signed crt with ca crt.\n"); + goto fail_verify; + } + + if (auth_verify_crt(auth, _server_crt) == 0) { + printf("Failed to detect untrusted crt.\n"); + goto fail_verify; + } + + crypt_free_crt(_im_ca_crt); + crypt_free_crt(_root_ca_crt); + crypt_free_crt(_signed_server_crt); + crypt_free_crt(_server_crt); + + auth_destroy_ctx(auth); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_verify: + crypt_free_crt(_im_ca_crt); + fail_load_im_ca_crt: + crypt_free_crt(_root_ca_crt); + fail_load_root_ca_crt: + crypt_free_crt(_signed_server_crt); + fail_load_signed_server_crt: + crypt_free_crt(_server_crt); + fail_load_server_crt: + auth_destroy_ctx(auth); + fail_create_ctx: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_verify_crt_missing_root_ca(void) +{ + struct auth_ctx * auth; + void * _signed_server_crt; + void * _im_ca_crt; + + TEST_START(); + + auth = auth_create_ctx(); + if (auth == NULL) { + printf("Failed to create auth context.\n"); + goto fail_create_ctx; + } + + if (crypt_load_crt_str(signed_server_crt_ec, &_signed_server_crt) < 0) { + printf("Failed to load signed crt from string.\n"); + goto fail_load_signed; + } + + if (crypt_load_crt_str(im_ca_crt_ec, &_im_ca_crt) < 0) { + printf("Failed to load intermediate crt from string.\n"); + goto fail_load_im_ca; + } + + /* Add only the intermediate CA - root CA is missing */ + if (auth_add_crt_to_store(auth, _im_ca_crt) < 0) { + printf("Failed to add intermediate ca crt to auth store.\n"); + goto fail_add; + } + + if (auth_verify_crt(auth, _signed_server_crt) == 0) { + printf("Verification should fail without root CA.\n"); + goto fail_add; + } + + crypt_free_crt(_im_ca_crt); + crypt_free_crt(_signed_server_crt); + auth_destroy_ctx(auth); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_add: + crypt_free_crt(_im_ca_crt); + fail_load_im_ca: + crypt_free_crt(_signed_server_crt); + fail_load_signed: + auth_destroy_ctx(auth); + fail_create_ctx: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* auth_verify_crt_pin: pin must lie in the verified chain (NULL: any) */ +static int test_verify_crt_pin(void) +{ + struct auth_ctx * auth; + void * _root_ca_crt; + void * _im_ca_crt; + void * _signed_server_crt; + void * _other_ca_crt; + + TEST_START(); + + auth = auth_create_ctx(); + if (auth == NULL) { + printf("Failed to create auth context.\n"); + goto fail_create_ctx; + } + + if (crypt_load_crt_str(root_ca_crt_ec, &_root_ca_crt) < 0) { + printf("Failed to load root crt from string.\n"); + goto fail_load_root_ca; + } + + if (crypt_load_crt_str(im_ca_crt_ec, &_im_ca_crt) < 0) { + printf("Failed to load intermediate crt from string.\n"); + goto fail_load_im_ca; + } + + if (crypt_load_crt_str(signed_server_crt_ec, &_signed_server_crt) < 0) { + printf("Failed to load signed crt from string.\n"); + goto fail_load_signed; + } + + if (crypt_load_crt_str(other_ca_crt_ec, &_other_ca_crt) < 0) { + printf("Failed to load out-of-chain crt from string.\n"); + goto fail_load_other; + } + + if (auth_add_crt_to_store(auth, _root_ca_crt) < 0) { + printf("Failed to add root ca crt to auth store.\n"); + goto fail_verify; + } + + if (auth_add_crt_to_store(auth, _im_ca_crt) < 0) { + printf("Failed to add intermediate ca crt to auth store.\n"); + goto fail_verify; + } + + if (auth_verify_crt_pin(auth, _signed_server_crt, _im_ca_crt) < 0) { + printf("Failed to accept pin on intermediate CA.\n"); + goto fail_verify; + } + + if (auth_verify_crt_pin(auth, _signed_server_crt, _root_ca_crt) < 0) { + printf("Failed to accept pin on root CA.\n"); + goto fail_verify; + } + + if (auth_verify_crt_pin(auth, _signed_server_crt, _other_ca_crt) == 0) { + printf("Failed to reject out-of-chain pin.\n"); + goto fail_verify; + } + + if (auth_verify_crt_pin(auth, _signed_server_crt, NULL) < 0) { + printf("Failed to accept NULL (any) pin.\n"); + goto fail_verify; + } + + crypt_free_crt(_other_ca_crt); + crypt_free_crt(_signed_server_crt); + crypt_free_crt(_im_ca_crt); + crypt_free_crt(_root_ca_crt); + + auth_destroy_ctx(auth); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_verify: + crypt_free_crt(_other_ca_crt); + fail_load_other: + crypt_free_crt(_signed_server_crt); + fail_load_signed: + crypt_free_crt(_im_ca_crt); + fail_load_im_ca: + crypt_free_crt(_root_ca_crt); + fail_load_root_ca: + auth_destroy_ctx(auth); + fail_create_ctx: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +int test_auth_sign(void) +{ + uint8_t buf[TEST_MSG_SIZE]; + void * pkp; + void * pk; + buffer_t msg; + buffer_t sig; + + TEST_START(); + + msg.data = buf; + msg.len = sizeof(buf); + + if (random_buffer(msg.data, msg.len) < 0) { + printf("Failed to generate random message.\n"); + goto fail_init; + } + + if (crypt_load_privkey_str(server_pkp_ec, &pkp) < 0) { + printf("Failed to load server key pair from string.\n"); + goto fail_init; + } + + if (crypt_load_pubkey_str(server_pk_ec, &pk) < 0) { + printf("Failed to load public key.\n"); + goto fail_pubkey; + } + + if (auth_sign(pkp, 0, msg, &sig) < 0) { + printf("Failed to sign message.\n"); + goto fail_sign; + } + + if (auth_verify_sig(pk, 0, msg, sig) < 0) { + printf("Failed to verify signature.\n"); + goto fail_verify; + } + + freebuf(sig); + + crypt_free_key(pk); + crypt_free_key(pkp); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_verify: + freebuf(sig); + fail_sign: + crypt_free_key(pk); + fail_pubkey: + crypt_free_key(pkp); + fail_init: + return TEST_RC_FAIL; +} + +int test_auth_bad_signature(void) +{ + uint8_t buf[TEST_MSG_SIZE]; + void * pkp; + void * pk; + buffer_t msg; + buffer_t sig; + buffer_t fake_sig; + + TEST_START(); + + msg.data = buf; + msg.len = sizeof(buf); + + if (random_buffer(msg.data, msg.len) < 0) { + printf("Failed to generate random message.\n"); + goto fail_init; + } + + if (crypt_load_privkey_str(server_pkp_ec, &pkp) < 0) { + printf("Failed to load server key pair from string.\n"); + goto fail_init; + } + + if (crypt_load_pubkey_str(server_pk_ec, &pk) < 0) { + printf("Failed to load public key.\n"); + goto fail_pubkey; + } + + if (auth_sign(pkp, 0, msg, &sig) < 0) { + printf("Failed to sign message.\n"); + goto fail_sign; + } + + fake_sig.data = malloc(sig.len); + if (fake_sig.data == NULL) { + printf("Failed to allocate memory for fake signature.\n"); + goto fail_malloc; + } + + fake_sig.len = sig.len; + if (random_buffer(fake_sig.data, fake_sig.len) < 0) { + printf("Failed to generate random fake signature.\n"); + goto fail_malloc; + } + + if (auth_verify_sig(pk, 0, msg, fake_sig) == 0) { + printf("Failed to detect bad signature.\n"); + goto fail_verify; + } + + freebuf(fake_sig); + freebuf(sig); + + crypt_free_key(pk); + crypt_free_key(pkp); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_verify: + freebuf(fake_sig); + fail_malloc: + freebuf(sig); + fail_sign: + crypt_free_key(pk); + fail_pubkey: + crypt_free_key(pkp); + fail_init: + return TEST_RC_FAIL; +} + +#define SSC_BUF_SIZE 4096 /* OpenSSL version my return different lengths */ +int test_crt_str(void) +{ + char str[SSC_BUF_SIZE]; + void * crt; + + TEST_START(); + + if (crypt_load_crt_str(signed_server_crt_ec, &crt) < 0) { + printf("Failed to load certificate from string.\n"); + goto fail_load; + } + + if (crypt_crt_str(crt, str) < 0) { + printf("Failed to convert certificate to string.\n"); + goto fail_to_str; + } + + printf("Certificate string:\n%s\n", str); + + crypt_free_crt(crt); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_to_str: + crypt_free_crt(crt); + fail_load: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +int auth_test(int argc, + char ** argv) +{ + int ret = 0; + + (void) argc; + (void) argv; + + ret |= test_auth_create_destroy_ctx(); +#ifdef HAVE_OPENSSL + ret |= test_load_free_crt(); + ret |= test_check_crt_name(); + ret |= test_crt_name_confusion(); + ret |= test_crypt_get_pubkey_crt(); + ret |= test_load_free_privkey(); + ret |= test_load_free_pubkey(); + ret |= test_crypt_check_pubkey_crt(); + ret |= test_store_add(); + ret |= test_verify_crt(); + ret |= test_verify_crt_missing_root_ca(); + ret |= test_verify_crt_pin(); + ret |= test_auth_sign(); + ret |= test_auth_bad_signature(); + ret |= test_crt_str(); +#else + (void) test_load_free_crt; + (void) test_check_crt_name; + (void) test_crt_name_confusion; + (void) test_crypt_get_pubkey_crt; + (void) test_load_free_privkey; + (void) test_load_free_pubkey; + (void) test_crypt_check_pubkey_crt; + (void) test_store_add; + (void) test_verify_crt; + (void) test_verify_crt_missing_root_ca; + (void) test_verify_crt_pin; + (void) test_auth_sign; + (void) test_auth_bad_signature; + (void) test_crt_str; + + ret = TEST_RC_SKIP; +#endif + return ret; +} diff --git a/src/lib/tests/auth_test_ml_dsa.c b/src/lib/tests/auth_test_ml_dsa.c new file mode 100644 index 00000000..cc72e61b --- /dev/null +++ b/src/lib/tests/auth_test_ml_dsa.c @@ -0,0 +1,356 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Test of the ML-DSA-65 authentication functions + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#include "config.h" + +#include <test/test.h> +#include <ouroboros/crypt.h> +#include <ouroboros/random.h> +#include <ouroboros/utils.h> + +#include <test/certs/ml_dsa.h> + +#define TEST_MSG_SIZE 1500 + +static int test_auth_create_destroy_ctx(void) +{ + struct auth_ctx * ctx; + + TEST_START(); + + ctx = auth_create_ctx(); + if (ctx == NULL) { + printf("Failed to create auth context.\n"); + goto fail_create; + } + + auth_destroy_ctx(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_create: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_load_free_crt(void) +{ + void * crt; + + TEST_START(); + + if (crypt_load_crt_str(root_ca_crt_ml, &crt) < 0) { + printf("Failed to load root crt from string.\n"); + goto fail_load; + } + + crypt_free_crt(crt); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_load: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_load_free_privkey(void) +{ + void * key; + + TEST_START(); + + if (crypt_load_privkey_str(server_pkp_ml, &key) < 0) { + printf("Failed to load server key pair from string.\n"); + goto fail_load; + } + + crypt_free_key(key); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_load: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_load_free_pubkey(void) +{ + void * key; + + TEST_START(); + + if (crypt_load_pubkey_str(server_pk_ml, &key) < 0) { + printf("Failed to load server public key from string.\n"); + goto fail_load; + } + + crypt_free_key(key); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_load: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_verify_crt(void) +{ + struct auth_ctx * auth; + void * _server_crt; + void * _signed_server_crt; + void * _root_ca_crt; + void * _im_ca_crt; + + TEST_START(); + + auth = auth_create_ctx(); + if (auth == NULL) { + printf("Failed to create auth context.\n"); + goto fail_create_ctx; + } + + if (crypt_load_crt_str(server_crt_ml, &_server_crt) < 0) { + printf("Failed to load self-signed crt from string.\n"); + goto fail_load_server_crt; + } + + if (crypt_load_crt_str(signed_server_crt_ml, &_signed_server_crt) < 0) { + printf("Failed to load signed crt from string.\n"); + goto fail_load_signed_server_crt; + } + + if (crypt_load_crt_str(root_ca_crt_ml, &_root_ca_crt) < 0) { + printf("Failed to load root crt from string.\n"); + goto fail_load_root_ca_crt; + } + + if (crypt_load_crt_str(im_ca_crt_ml, &_im_ca_crt) < 0) { + printf("Failed to load intermediate crt from string.\n"); + goto fail_load_im_ca_crt; + } + + if (auth_add_crt_to_store(auth, _root_ca_crt) < 0) { + printf("Failed to add root ca crt to auth store.\n"); + goto fail_verify; + } + + if (auth_add_crt_to_store(auth, _im_ca_crt) < 0) { + printf("Failed to add intermediate ca crt to auth store.\n"); + goto fail_verify; + } + + if (auth_verify_crt(auth, _signed_server_crt) < 0) { + printf("Failed to verify signed crt with ca crt.\n"); + goto fail_verify; + } + + if (auth_verify_crt(auth, _server_crt) == 0) { + printf("Failed to detect untrusted crt.\n"); + goto fail_verify; + } + + crypt_free_crt(_im_ca_crt); + crypt_free_crt(_root_ca_crt); + crypt_free_crt(_signed_server_crt); + crypt_free_crt(_server_crt); + + auth_destroy_ctx(auth); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_verify: + crypt_free_crt(_im_ca_crt); + fail_load_im_ca_crt: + crypt_free_crt(_root_ca_crt); + fail_load_root_ca_crt: + crypt_free_crt(_signed_server_crt); + fail_load_signed_server_crt: + crypt_free_crt(_server_crt); + fail_load_server_crt: + auth_destroy_ctx(auth); + fail_create_ctx: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_auth_sign(void) +{ + uint8_t buf[TEST_MSG_SIZE]; + void * pkp; + void * pk; + buffer_t msg; + buffer_t sig; + + TEST_START(); + + msg.data = buf; + msg.len = sizeof(buf); + + if (random_buffer(msg.data, msg.len) < 0) { + printf("Failed to generate random message.\n"); + goto fail_init; + } + + if (crypt_load_privkey_str(server_pkp_ml, &pkp) < 0) { + printf("Failed to load server key pair from string.\n"); + goto fail_init; + } + + if (crypt_load_pubkey_str(server_pk_ml, &pk) < 0) { + printf("Failed to load public key from string.\n"); + goto fail_pubkey; + } + + if (auth_sign(pkp, 0, msg, &sig) < 0) { + printf("Failed to sign message.\n"); + goto fail_sign; + } + + if (auth_verify_sig(pk, 0, msg, sig) < 0) { + printf("Failed to verify signature.\n"); + goto fail_verify; + } + + freebuf(sig); + + crypt_free_key(pk); + crypt_free_key(pkp); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_verify: + freebuf(sig); + fail_sign: + crypt_free_key(pk); + fail_pubkey: + crypt_free_key(pkp); + fail_init: + return TEST_RC_FAIL; +} + +static int test_auth_bad_signature(void) +{ + uint8_t buf[TEST_MSG_SIZE]; + void * pkp; + void * pk; + buffer_t msg; + buffer_t sig; + buffer_t fake_sig; + + TEST_START(); + + msg.data = buf; + msg.len = sizeof(buf); + + if (random_buffer(msg.data, msg.len) < 0) { + printf("Failed to generate random message.\n"); + goto fail_init; + } + + if (crypt_load_privkey_str(server_pkp_ml, &pkp) < 0) { + printf("Failed to load server key pair from string.\n"); + goto fail_init; + } + + if (crypt_load_pubkey_str(server_pk_ml, &pk) < 0) { + printf("Failed to load public key from string.\n"); + goto fail_pubkey; + } + + if (auth_sign(pkp, 0, msg, &sig) < 0) { + printf("Failed to sign message.\n"); + goto fail_sign; + } + + fake_sig.data = malloc(sig.len); + if (fake_sig.data == NULL) { + printf("Failed to allocate memory for fake signature.\n"); + goto fail_malloc; + } + + fake_sig.len = sig.len; + if (random_buffer(fake_sig.data, fake_sig.len) < 0) { + printf("Failed to generate random fake signature.\n"); + goto fail_malloc; + } + + if (auth_verify_sig(pk, 0, msg, fake_sig) == 0) { + printf("Failed to detect bad ML-DSA-65 signature.\n"); + goto fail_verify; + } + + freebuf(fake_sig); + freebuf(sig); + + crypt_free_key(pk); + crypt_free_key(pkp); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_verify: + freebuf(fake_sig); + fail_malloc: + freebuf(sig); + fail_sign: + crypt_free_key(pk); + fail_pubkey: + crypt_free_key(pkp); + fail_init: + return TEST_RC_FAIL; +} + +int auth_test_ml_dsa(int argc, + char ** argv) +{ + int ret = 0; + + (void) argc; + (void) argv; + +#ifdef HAVE_OPENSSL_ML_DSA + ret |= test_auth_create_destroy_ctx(); + ret |= test_load_free_crt(); + ret |= test_load_free_privkey(); + ret |= test_load_free_pubkey(); + ret |= test_verify_crt(); + ret |= test_auth_sign(); + ret |= test_auth_bad_signature(); +#else + (void) test_auth_create_destroy_ctx; + (void) test_load_free_crt; + (void) test_load_free_privkey; + (void) test_load_free_pubkey; + (void) test_verify_crt; + (void) test_auth_sign; + (void) test_auth_bad_signature; + + ret = TEST_RC_SKIP; +#endif + return ret; +} diff --git a/src/lib/tests/auth_test_slh_dsa.c b/src/lib/tests/auth_test_slh_dsa.c new file mode 100644 index 00000000..511d20fe --- /dev/null +++ b/src/lib/tests/auth_test_slh_dsa.c @@ -0,0 +1,367 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Test of the SLH-DSA-SHA2-128s authentication functions + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#include "config.h" + +#include <test/test.h> +#include <ouroboros/crypt.h> +#include <ouroboros/random.h> +#include <ouroboros/utils.h> + +#include <test/certs/slh_dsa.h> + +#define TEST_MSG_SIZE 1500 + +static int test_auth_create_destroy_ctx(void) +{ + struct auth_ctx * ctx; + + TEST_START(); + + ctx = auth_create_ctx(); + if (ctx == NULL) { + printf("Failed to create auth context.\n"); + goto fail_create; + } + + auth_destroy_ctx(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_create: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_load_free_crt(void) +{ + void * crt; + + TEST_START(); + + if (crypt_load_crt_str(root_ca_crt_slh, &crt) < 0) { + printf("Failed to load root crt.\n"); + goto fail_load; + } + + crypt_free_crt(crt); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_load: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_load_free_privkey(void) +{ + void * key; + + TEST_START(); + + if (crypt_load_privkey_str(server_pkp_slh, &key) < 0) { + printf("Failed to load server key pair.\n"); + goto fail_load; + } + + crypt_free_key(key); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_load: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_load_free_pubkey(void) +{ + void * key; + + TEST_START(); + + if (crypt_load_pubkey_str(server_pk_slh, &key) < 0) { + printf("Failed to load server public key.\n"); + goto fail_load; + } + + crypt_free_key(key); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_load: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_verify_crt(void) +{ + struct auth_ctx * auth; + void * _server_crt; + void * _signed_server_crt; + void * _root_ca_crt; + void * _im_ca_crt; + + TEST_START(); + + auth = auth_create_ctx(); + if (auth == NULL) { + printf("Failed to create auth context.\n"); + goto fail_create_ctx; + } + + if (crypt_load_crt_str(server_crt_slh, + &_server_crt) < 0) { + printf("Failed to load self-signed crt.\n"); + goto fail_load_server_crt; + } + + if (crypt_load_crt_str(signed_server_crt_slh, + &_signed_server_crt) < 0) { + printf("Failed to load signed crt.\n"); + goto fail_load_signed_server_crt; + } + + if (crypt_load_crt_str(root_ca_crt_slh, + &_root_ca_crt) < 0) { + printf("Failed to load root crt.\n"); + goto fail_load_root_ca_crt; + } + + if (crypt_load_crt_str(im_ca_crt_slh, + &_im_ca_crt) < 0) { + printf("Failed to load im crt.\n"); + goto fail_load_im_ca_crt; + } + + if (auth_add_crt_to_store(auth, _root_ca_crt) < 0) { + printf("Failed to add root ca crt.\n"); + goto fail_verify; + } + + if (auth_add_crt_to_store(auth, _im_ca_crt) < 0) { + printf("Failed to add im ca crt.\n"); + goto fail_verify; + } + + if (auth_verify_crt(auth, _signed_server_crt) < 0) { + printf("Failed to verify signed crt.\n"); + goto fail_verify; + } + + if (auth_verify_crt(auth, _server_crt) == 0) { + printf("Failed to detect untrusted crt.\n"); + goto fail_verify; + } + + crypt_free_crt(_im_ca_crt); + crypt_free_crt(_root_ca_crt); + crypt_free_crt(_signed_server_crt); + crypt_free_crt(_server_crt); + + auth_destroy_ctx(auth); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_verify: + crypt_free_crt(_im_ca_crt); + fail_load_im_ca_crt: + crypt_free_crt(_root_ca_crt); + fail_load_root_ca_crt: + crypt_free_crt(_signed_server_crt); + fail_load_signed_server_crt: + crypt_free_crt(_server_crt); + fail_load_server_crt: + auth_destroy_ctx(auth); + fail_create_ctx: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_auth_sign(void) +{ + uint8_t buf[TEST_MSG_SIZE]; + void * pkp; + void * pk; + buffer_t msg; + buffer_t sig; + + TEST_START(); + + msg.data = buf; + msg.len = sizeof(buf); + + if (random_buffer(msg.data, msg.len) < 0) { + printf("Failed to gen random message.\n"); + goto fail_init; + } + + if (crypt_load_privkey_str(server_pkp_slh, + &pkp) < 0) { + printf("Failed to load server key pair.\n"); + goto fail_init; + } + + if (crypt_load_pubkey_str(server_pk_slh, + &pk) < 0) { + printf("Failed to load public key.\n"); + goto fail_pubkey; + } + + if (auth_sign(pkp, 0, msg, &sig) < 0) { + printf("Failed to sign message.\n"); + goto fail_sign; + } + + if (auth_verify_sig(pk, 0, msg, sig) < 0) { + printf("Failed to verify signature.\n"); + goto fail_verify; + } + + freebuf(sig); + + crypt_free_key(pk); + crypt_free_key(pkp); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_verify: + freebuf(sig); + fail_sign: + crypt_free_key(pk); + fail_pubkey: + crypt_free_key(pkp); + fail_init: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_auth_bad_signature(void) +{ + uint8_t buf[TEST_MSG_SIZE]; + void * pkp; + void * pk; + buffer_t msg; + buffer_t sig; + buffer_t fake_sig; + + TEST_START(); + + msg.data = buf; + msg.len = sizeof(buf); + + if (random_buffer(msg.data, msg.len) < 0) { + printf("Failed to gen random message.\n"); + goto fail_init; + } + + if (crypt_load_privkey_str(server_pkp_slh, + &pkp) < 0) { + printf("Failed to load server key pair.\n"); + goto fail_init; + } + + if (crypt_load_pubkey_str(server_pk_slh, + &pk) < 0) { + printf("Failed to load public key.\n"); + goto fail_pubkey; + } + + if (auth_sign(pkp, 0, msg, &sig) < 0) { + printf("Failed to sign message.\n"); + goto fail_sign; + } + + fake_sig.data = malloc(sig.len); + if (fake_sig.data == NULL) { + printf("Failed to alloc fake sig buf.\n"); + goto fail_malloc; + } + + fake_sig.len = sig.len; + if (random_buffer(fake_sig.data, + fake_sig.len) < 0) { + printf("Failed to gen random fake sig.\n"); + goto fail_malloc; + } + + if (auth_verify_sig(pk, 0, msg, fake_sig) == 0) { + printf("Failed to detect bad sig.\n"); + goto fail_verify; + } + + freebuf(fake_sig); + freebuf(sig); + + crypt_free_key(pk); + crypt_free_key(pkp); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_verify: + freebuf(fake_sig); + fail_malloc: + freebuf(sig); + fail_sign: + crypt_free_key(pk); + fail_pubkey: + crypt_free_key(pkp); + fail_init: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +int auth_test_slh_dsa(int argc, + char ** argv) +{ + int ret = 0; + + (void) argc; + (void) argv; + +#ifdef HAVE_OPENSSL_SLH_DSA + ret |= test_auth_create_destroy_ctx(); + ret |= test_load_free_crt(); + ret |= test_load_free_privkey(); + ret |= test_load_free_pubkey(); + ret |= test_verify_crt(); + ret |= test_auth_sign(); + ret |= test_auth_bad_signature(); +#else + (void) test_auth_create_destroy_ctx; + (void) test_load_free_crt; + (void) test_load_free_privkey; + (void) test_load_free_pubkey; + (void) test_verify_crt; + (void) test_auth_sign; + (void) test_auth_bad_signature; + + ret = TEST_RC_SKIP; +#endif + return ret; +} diff --git a/src/lib/tests/bitmap_test.c b/src/lib/tests/bitmap_test.c index 4dbd6653..81735a41 100644 --- a/src/lib/tests/bitmap_test.c +++ b/src/lib/tests/bitmap_test.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Test of the bitmap * diff --git a/src/lib/tests/btree_test.c b/src/lib/tests/btree_test.c index 8bd30370..d1a72af8 100644 --- a/src/lib/tests/btree_test.c +++ b/src/lib/tests/btree_test.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Test of the B-tree implementation * diff --git a/src/lib/tests/crypt_test.c b/src/lib/tests/crypt_test.c new file mode 100644 index 00000000..2d752238 --- /dev/null +++ b/src/lib/tests/crypt_test.c @@ -0,0 +1,555 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Test of the cryptography functions + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#include "config.h" + +#include <test/test.h> +#include <ouroboros/random.h> +#include <ouroboros/crypt.h> +#include <ouroboros/utils.h> + +#include <stdio.h> + +#define TEST_PACKET_SIZE 1500 +#define TEST_N_PACKETS 1000 + +extern const uint16_t crypt_supported_nids[]; +extern const uint16_t md_supported_nids[]; + +static int test_crypt_create_destroy(void) +{ + struct crypt_ctx * ctx; + uint8_t key[SYMMKEYSZ]; + struct crypt_sk sk = { + .nid = NID_aes_256_gcm, + .key = key, + .epoch = 0, + .role = CRYPT_ROLE_INIT + }; + + TEST_START(); + + memset(key, 0, sizeof(key)); + + ctx = crypt_create_ctx(&sk); + if (ctx == NULL) { + printf("Failed to initialize cryptography.\n"); + goto fail; + } + + crypt_destroy_ctx(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_crypt_encrypt_decrypt(int nid) +{ + uint8_t pkt[TEST_PACKET_SIZE]; + struct crypt_ctx * tx; + struct crypt_ctx * rx; + uint8_t key[SYMMKEYSZ]; + struct crypt_sk sk_tx = { + .key = key, + .epoch = 0, + .role = CRYPT_ROLE_INIT + }; + struct crypt_sk sk_rx = { + .key = key, + .epoch = 0, + .role = CRYPT_ROLE_RESP + }; + buffer_t in; + buffer_t out; + buffer_t out2; + const char * cipher; + + sk_tx.nid = nid; + sk_rx.nid = nid; + + cipher = crypt_nid_to_str(nid); + TEST_START("(%s)", cipher); + + if (random_buffer(key, sizeof(key)) < 0) { + printf("Failed to generate random key.\n"); + goto fail_init; + } + + if (random_buffer(pkt, sizeof(pkt)) < 0) { + printf("Failed to generate random data.\n"); + goto fail_init; + } + + tx = crypt_create_ctx(&sk_tx); + if (tx == NULL) { + printf("Failed to initialize TX cryptography.\n"); + goto fail_init; + } + + rx = crypt_create_ctx(&sk_rx); + if (rx == NULL) { + printf("Failed to initialize RX cryptography.\n"); + goto fail_tx; + } + + in.len = sizeof(pkt); + in.data = pkt; + + if (crypt_encrypt(tx, in, &out) < 0) { + printf("Encryption failed.\n"); + goto fail_encrypt; + } + + if (out.len < in.len) { + printf("Encryption returned too little data.\n"); + goto fail_chk; + } + + if (crypt_decrypt(rx, out, &out2) < 0) { + printf("Decryption failed.\n"); + goto fail_decrypt; + } + + if (out2.len != in.len) { + printf("Decrypted data length does not match original.\n"); + goto fail_chk2; + } + + if (memcmp(in.data, out2.data, in.len) != 0) { + printf("Decrypted data does not match original.\n"); + goto fail_chk2; + } + + freebuf(out2); + freebuf(out); + crypt_destroy_ctx(rx); + crypt_destroy_ctx(tx); + + TEST_SUCCESS("(%s)", cipher); + + return TEST_RC_SUCCESS; + fail_chk2: + freebuf(out2); + fail_decrypt: + fail_chk: + freebuf(out); + fail_encrypt: + crypt_destroy_ctx(rx); + fail_tx: + crypt_destroy_ctx(tx); + fail_init: + TEST_FAIL("(%s)", cipher); + return TEST_RC_FAIL; +} + +static int test_encrypt_decrypt_all(void) +{ + int ret = 0; + int i; + + for (i = 0; crypt_supported_nids[i] != NID_undef; i++) + ret |= test_crypt_encrypt_decrypt(crypt_supported_nids[i]); + + return ret; +} + +static int test_crypt_multi_packet(int nid) +{ + uint8_t pkt[TEST_PACKET_SIZE]; + struct crypt_ctx * tx; + struct crypt_ctx * rx; + uint8_t key[SYMMKEYSZ]; + struct crypt_sk sk_tx = { + .key = key, + .epoch = 0, + .role = CRYPT_ROLE_INIT + }; + struct crypt_sk sk_rx = { + .key = key, + .epoch = 0, + .role = CRYPT_ROLE_RESP + }; + buffer_t in; + buffer_t enc; + buffer_t dec; + const char * cipher; + int i; + + sk_tx.nid = nid; + sk_rx.nid = nid; + + cipher = crypt_nid_to_str(nid); + TEST_START("(%s)", cipher); + + if (random_buffer(key, sizeof(key)) < 0) { + printf("Failed to generate random key.\n"); + goto fail_init; + } + + if (random_buffer(pkt, sizeof(pkt)) < 0) { + printf("Failed to generate random data.\n"); + goto fail_init; + } + + tx = crypt_create_ctx(&sk_tx); + if (tx == NULL) { + printf("Failed to create TX context.\n"); + goto fail_init; + } + + rx = crypt_create_ctx(&sk_rx); + if (rx == NULL) { + printf("Failed to create RX context.\n"); + goto fail_tx; + } + + in.len = sizeof(pkt); + in.data = pkt; + + for (i = 0; i < TEST_N_PACKETS; i++) { + if (crypt_encrypt(tx, in, &enc) < 0) { + printf("Encryption failed at packet %d.\n", i); + goto fail_rx; + } + + if (crypt_decrypt(rx, enc, &dec) < 0) { + printf("Decryption failed at packet %d.\n", i); + freebuf(enc); + goto fail_rx; + } + + if (dec.len != in.len || + memcmp(in.data, dec.data, in.len) != 0) { + printf("Data mismatch at packet %d.\n", i); + freebuf(dec); + freebuf(enc); + goto fail_rx; + } + + freebuf(dec); + freebuf(enc); + } + + crypt_destroy_ctx(rx); + crypt_destroy_ctx(tx); + + TEST_SUCCESS("(%s)", cipher); + + return TEST_RC_SUCCESS; + fail_rx: + crypt_destroy_ctx(rx); + fail_tx: + crypt_destroy_ctx(tx); + fail_init: + TEST_FAIL("(%s)", cipher); + return TEST_RC_FAIL; +} + +static int test_multi_packet_all(void) +{ + int ret = 0; + int i; + + for (i = 0; crypt_supported_nids[i] != NID_undef; i++) + ret |= test_crypt_multi_packet(crypt_supported_nids[i]); + + return ret; +} + +static int test_crypt_aad_tamper(int nid) +{ + uint8_t pkt[TEST_PACKET_SIZE]; + struct crypt_ctx * tx; + struct crypt_ctx * rx; + uint8_t key[SYMMKEYSZ]; + struct crypt_sk sk_tx = { + .key = key, + .epoch = 0, + .role = CRYPT_ROLE_INIT + }; + struct crypt_sk sk_rx = { + .key = key, + .epoch = 0, + .role = CRYPT_ROLE_RESP + }; + buffer_t in; + buffer_t enc; + buffer_t dec; + const char * cipher; + + sk_tx.nid = nid; + sk_rx.nid = nid; + + cipher = crypt_nid_to_str(nid); + TEST_START("(%s)", cipher); + + if (random_buffer(key, sizeof(key)) < 0) { + printf("Failed to generate random key.\n"); + goto fail_init; + } + + if (random_buffer(pkt, sizeof(pkt)) < 0) { + printf("Failed to generate random data.\n"); + goto fail_init; + } + + tx = crypt_create_ctx(&sk_tx); + if (tx == NULL) { + printf("Failed to create TX context.\n"); + goto fail_init; + } + + rx = crypt_create_ctx(&sk_rx); + if (rx == NULL) { + printf("Failed to create RX context.\n"); + goto fail_tx; + } + + /* Only AEAD ciphers bind the selector as AAD. */ + if (crypt_get_tagsz(tx) == 0) { + crypt_destroy_ctx(rx); + crypt_destroy_ctx(tx); + + TEST_SUCCESS("(%s)", cipher); + + return TEST_RC_SUCCESS; + } + + in.len = sizeof(pkt); + in.data = pkt; + + if (crypt_encrypt(tx, in, &enc) < 0) { + printf("Encryption failed.\n"); + goto fail_rx; + } + + /* Flip a seq byte: epoch/node stay valid so the AEAD tag rejects. */ + enc.data[5] ^= 0x01; + + if (crypt_decrypt(rx, enc, &dec) == 0) { + printf("Decryption accepted a tampered selector.\n"); + freebuf(dec); + freebuf(enc); + goto fail_rx; + } + + freebuf(enc); + + crypt_destroy_ctx(rx); + crypt_destroy_ctx(tx); + + TEST_SUCCESS("(%s)", cipher); + + return TEST_RC_SUCCESS; + fail_rx: + crypt_destroy_ctx(rx); + fail_tx: + crypt_destroy_ctx(tx); + fail_init: + TEST_FAIL("(%s)", cipher); + return TEST_RC_FAIL; +} + +static int test_aad_tamper_all(void) +{ + int ret = 0; + int i; + + for (i = 0; crypt_supported_nids[i] != NID_undef; i++) + ret |= test_crypt_aad_tamper(crypt_supported_nids[i]); + + return ret; +} + +#ifdef HAVE_OPENSSL +#include <openssl/evp.h> +#include <openssl/obj_mac.h> + +static int test_cipher_nid_values(void) +{ + int i; + + TEST_START(); + + /* Loop over all supported ciphers and verify NIDs match OpenSSL's */ + for (i = 0; crypt_supported_nids[i] != NID_undef; i++) { + uint16_t our_nid = crypt_supported_nids[i]; + const char * str = crypt_nid_to_str(our_nid); + const EVP_CIPHER * cipher; + int openssl_nid; + + if (str == NULL) { + printf("crypt_nid_to_str failed for NID %u\n", our_nid); + goto fail; + } + + cipher = EVP_get_cipherbyname(str); + if (cipher == NULL) { + printf("OpenSSL doesn't recognize cipher '%s'\n", str); + goto fail; + } + + openssl_nid = EVP_CIPHER_nid(cipher); + + if (our_nid != openssl_nid) { + printf("NID mismatch for '%s': ours=%u, OpenSSL=%d\n", + str, our_nid, openssl_nid); + goto fail; + } + + /* Test reverse conversion */ + if (crypt_str_to_nid(str) != our_nid) { + printf("crypt_str_to_nid failed for '%s'\n", str); + goto fail; + } + } + + /* Test error cases */ + if (crypt_str_to_nid("invalid") != NID_undef) { + printf("crypt_str_to_nid: no NID_undef for invalid.\n"); + goto fail; + } + + if (crypt_nid_to_str(9999) != NULL) { + printf("crypt_nid_to_str should return NULL for invalid NID\n"); + goto fail; + } + + if (crypt_str_to_nid(NULL) != NID_undef) { + printf("crypt_str_to_nid should return NID_undef for NULL\n"); + goto fail; + } + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_md_nid_values(void) +{ + int i; + + TEST_START(); + + for (i = 0; md_supported_nids[i] != NID_undef; i++) { + uint16_t our_nid = md_supported_nids[i]; + const EVP_MD * md; + int openssl_nid; + + md = EVP_get_digestbynid(our_nid); + if (md == NULL) { + printf("OpenSSL doesn't recognize NID %u\n", our_nid); + goto fail; + } + + openssl_nid = EVP_MD_nid(md); + if (our_nid != openssl_nid) { + printf("NID mismatch: ours=%u, OpenSSL=%d\n", + our_nid, openssl_nid); + goto fail; + } + } + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} +#endif + +static int test_crypt_headsz(void) +{ + struct crypt_ctx * ctx; + uint8_t key[SYMMKEYSZ]; + struct crypt_sk sk = { + .nid = NID_aes_256_gcm, + .key = key, + .epoch = 0, + .role = CRYPT_ROLE_INIT + }; + int headsz; + + TEST_START(); + + if (random_buffer(key, sizeof(key)) < 0) { + printf("Failed to generate random key.\n"); + goto fail; + } + + ctx = crypt_create_ctx(&sk); + if (ctx == NULL) { + printf("Failed to initialize cryptography.\n"); + goto fail; + } + + headsz = crypt_get_headsz(ctx); + if (headsz != 6) { + printf("Unexpected header size: %d (expected 6).\n", headsz); + goto fail_ctx; + } + + crypt_destroy_ctx(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + crypt_destroy_ctx(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +int crypt_test(int argc, + char ** argv) +{ + int ret = 0; + + (void) argc; + (void) argv; + + ret |= test_crypt_create_destroy(); + ret |= test_encrypt_decrypt_all(); +#ifdef HAVE_OPENSSL + ret |= test_cipher_nid_values(); + ret |= test_md_nid_values(); + ret |= test_multi_packet_all(); + ret |= test_aad_tamper_all(); + ret |= test_crypt_headsz(); +#else + (void) test_multi_packet_all; + (void) test_aad_tamper_all; + (void) test_crypt_headsz; + + return TEST_RC_SKIP; +#endif + return ret; +} diff --git a/src/lib/tests/hash_test.c b/src/lib/tests/hash_test.c index 970d9185..a2ba62cc 100644 --- a/src/lib/tests/hash_test.c +++ b/src/lib/tests/hash_test.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Test of the hashing functions * @@ -21,7 +21,7 @@ */ #include <ouroboros/hash.h> -#include <ouroboros/test.h> +#include <test/test.h> #include <stdlib.h> #include <stdint.h> @@ -39,6 +39,79 @@ struct vec_entry { char * out; }; +struct mix_entry { + uint64_t in; + uint64_t out; +}; + +static int test_crc8(void) +{ + int ret = 0; + + struct vec_entry vec [] = { + { "", "00" }, + { "123456789", "df" }, + { NULL, NULL } + }; + + struct vec_entry * cur = vec; + + TEST_START(); + + while (cur->in != NULL) { + uint8_t crc; + char res[3]; + + str_hash(HASH_CRC8, &crc, cur->in); + + sprintf(res, "%02x", crc); + if (strcmp(res, cur->out) != 0) { + printf("Hash failed %s != %s.\n", res, cur->out); + ret |= -1; + } + + ++cur; + } + + TEST_END(ret); + + return ret; +} + +static int test_crc16(void) +{ + int ret = 0; + + struct vec_entry vec [] = { + { "", "ffff" }, + { "123456789", "29b1" }, + { NULL, NULL } + }; + + struct vec_entry * cur = vec; + + TEST_START(); + + while (cur->in != NULL) { + uint8_t crc[2]; + char res[5]; + + str_hash(HASH_CRC16, crc, cur->in); + + sprintf(res, "%02x%02x", crc[0], crc[1]); + if (strcmp(res, cur->out) != 0) { + printf("Hash failed %s != %s.\n", res, cur->out); + ret |= -1; + } + + ++cur; + } + + TEST_END(ret); + + return ret; +} + static int test_crc32(void) { int ret = 0; @@ -74,6 +147,42 @@ static int test_crc32(void) return ret; } +static int test_crc64(void) +{ + int ret = 0; + + struct vec_entry vec [] = { + { "", "0000000000000000" }, + { "123456789", "ae8b14860a799888" }, + { "0123456789abcdef", + "091485ca7018730e" }, + { NULL, NULL } + }; + + struct vec_entry * cur = vec; + + TEST_START(); + + while (cur->in != NULL) { + uint8_t crc[8]; + char res[17]; + + str_hash(HASH_CRC64, crc, cur->in); + + sprintf(res, HASH_FMT64, HASH_VAL64(crc)); + if (strcmp(res, cur->out) != 0) { + printf("Hash failed %s != %s.\n", res, cur->out); + ret |= -1; + } + + ++cur; + } + + TEST_END(ret); + + return ret; +} + static int test_md5(void) { int ret = 0; @@ -184,6 +293,36 @@ static int test_sha3(void) return ret; } +static int test_mix64(void) +{ + int ret = 0; + + struct mix_entry vec [] = { + { 0x0000000000000000ULL, 0x0000000000000000ULL }, + { 0x123456789abcdefeULL, 0xb1943cfea4f78f08ULL } + }; + + size_t n = sizeof(vec) / sizeof(vec[0]); + size_t i; + + TEST_START(); + + for (i = 0; i < n; i++) { + uint64_t res = hash_mix64(vec[i].in); + + if (res != vec[i].out) { + printf("Mix failed %016llx != %016llx.\n", + (unsigned long long) res, + (unsigned long long) vec[i].out); + ret |= -1; + } + } + + TEST_END(ret); + + return ret; +} + int hash_test(int argc, char ** argv) { @@ -192,11 +331,19 @@ int hash_test(int argc, (void) argc; (void) argv; + ret |= test_crc8(); + + ret |= test_crc16(); + ret |= test_crc32(); + ret |= test_crc64(); + ret |= test_md5(); ret |= test_sha3(); + ret |= test_mix64(); + return ret; } diff --git a/src/lib/tests/kex_test.c b/src/lib/tests/kex_test.c new file mode 100644 index 00000000..0a00ccab --- /dev/null +++ b/src/lib/tests/kex_test.c @@ -0,0 +1,1194 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Test of the key exchange functions + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#define _POSIX_C_SOURCE 200809L + +#include "config.h" + +#include <test/test.h> +#include <ouroboros/utils.h> +#include <ouroboros/crypt.h> + +#include <stdio.h> +#include <stdlib.h> +#include <unistd.h> + +#ifdef HAVE_OPENSSL +#include <openssl/evp.h> +#include <openssl/x509.h> +#endif + +/* Test configuration strings */ +#define KEX_CONFIG_CUSTOM \ + "kex=X25519\n" + +#define KEX_CONFIG_NONE \ + "none\n" + +#define KEX_CONFIG_NO_ENC \ + "encryption=none\n" + +#define KEX_CONFIG_WHITESPACE \ + "# Comment line\n" \ + "kex = X448" \ + "\n" \ + "# Another comment\n" + +#define KEX_CONFIG_CIPHER \ + "kex=X25519\n" \ + "cipher=chacha20-poly1305\n" + +#define KEX_CONFIG_DIGEST \ + "kex=X25519\n" \ + "digest=sha384\n" + +#define KEX_CONFIG_AUTH \ + "auth=required\n" + +#define KEX_CONFIG_AUTH_INVALID \ + "auth=mandatory\n" + +#define KEX_CONFIG_AUTH_OPTIONAL \ + "auth=optional\n" + +#define KEX_CONFIG_AUTH_THEN_NO_ENC \ + "auth=required\n" \ + "digest=sha512\n" \ + "encryption=none\n" + +#define KEX_CONFIG_NO_ENC_THEN_AUTH \ + "encryption=none\n" \ + "auth=required\n" \ + "digest=sha512\n" + +#define KEX_CONFIG_CACERT \ + "cacert=/etc/ouroboros/security/cacert/ca.crt\n" + +#define KEX_CONFIG_UNKNOWN_KEY \ + "autth=required\n" + +/* Test key material for key loading tests */ +#define X25519_PRIVKEY_PEM \ + "-----BEGIN PRIVATE KEY-----\n" \ + "MC4CAQAwBQYDK2VuBCIEIJDd3+/0k2IZlaH5sZ9Z2e5J8dV2U0nsXaSUm70ZaMhL\n" \ + "-----END PRIVATE KEY-----\n" + +#define X25519_PUBKEY_PEM \ + "-----BEGIN PUBLIC KEY-----\n" \ + "MCowBQYDK2VuAyEAKYLIycSZtLFlwAX07YWWgBAYhEnRxHfgK1TVw9+mtBs=\n" \ + "-----END PUBLIC KEY-----\n" + +/* Helper macro to open string constant as FILE stream */ +#define FMEMOPEN_STR(str) fmemopen((void *) (str), strlen(str), "r") + +extern const uint16_t kex_supported_nids[]; + +int parse_sec_config(struct sec_config * cfg, + FILE * fp); + +static int test_kex_create_destroy(void) +{ + struct sec_config cfg; + + TEST_START(); + + memset(&cfg, 0, sizeof(cfg)); + cfg.x.nid = NID_X9_62_prime256v1; + cfg.x.str = kex_nid_to_str(cfg.x.nid); + cfg.c.nid = NID_aes_256_gcm; + cfg.c.str = crypt_nid_to_str(cfg.c.nid); + + if (cfg.x.nid == NID_undef || cfg.c.nid == NID_undef) { + printf("Failed to initialize kex config.\n"); + goto fail; + } + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_kex_dh_pkp_create_destroy(void) +{ + struct sec_config kex; + void * pkp; + uint8_t buf[CRYPT_KEY_BUFSZ]; + + TEST_START(); + + memset(&kex, 0, sizeof(kex)); + SET_KEX_ALGO(&kex, "prime256v1"); + + if (kex_pkp_create(&kex, &pkp, buf) < 0) { + printf("Failed to create DH PKP.\n"); + goto fail; + } + + kex_pkp_destroy(pkp); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_kex_get_algo_from_pk(const char * algo) +{ + struct sec_config kex; + void * pkp; + buffer_t pk; + ssize_t len; + uint8_t buf[CRYPT_KEY_BUFSZ]; + char extracted_algo[256]; + + TEST_START("(%s)", algo); + + memset(&kex, 0, sizeof(kex)); + SET_KEX_ALGO(&kex, algo); + + len = kex_pkp_create(&kex, &pkp, buf); + if (len < 0) { + printf("Failed to create key pair.\n"); + goto fail; + } + + pk.len = (size_t) len; + pk.data = buf; + + /* Use raw decode for hybrid KEMs, DER for others */ + if (IS_HYBRID_KEM(algo)) { + if (kex_get_algo_from_pk_raw(pk, extracted_algo) < 0) { + printf("Failed to extract algo from pk.\n"); + goto fail_pkp; + } + } else { + if (kex_get_algo_from_pk_der(pk, extracted_algo) < 0) { + printf("Failed to extract algo from pk.\n"); + goto fail_pkp; + } + } + + /* All algorithms should now return the specific group name */ + if (strcmp(extracted_algo, algo) != 0) { + printf("Algo mismatch: expected %s, got %s.\n", + algo, extracted_algo); + goto fail_pkp; + } + + kex_pkp_destroy(pkp); + + TEST_SUCCESS("(%s)", algo); + + return TEST_RC_SUCCESS; + fail_pkp: + kex_pkp_destroy(pkp); + fail: + TEST_FAIL("(%s)", algo); + return TEST_RC_FAIL; +} + +static int test_kex_get_algo_from_pk_all(void) +{ + int ret = 0; + int i; + + for (i = 0; kex_supported_nids[i] != NID_undef; i++) { + const char * algo = kex_nid_to_str(kex_supported_nids[i]); + ret |= test_kex_get_algo_from_pk(algo); + } + + return ret; +} + +static int test_kex_dhe_derive(const char * algo) +{ + struct sec_config kex; + void * pkp1; + void * pkp2; + buffer_t pk1; + buffer_t pk2; + ssize_t len; + uint8_t buf1[CRYPT_KEY_BUFSZ]; + uint8_t buf2[CRYPT_KEY_BUFSZ]; + uint8_t s1[SYMMKEYSZ]; + uint8_t s2[SYMMKEYSZ]; + + TEST_START("(%s)", algo); + + memset(&kex, 0, sizeof(kex)); + SET_KEX_ALGO(&kex, algo); + SET_KEX_KDF_NID(&kex, NID_sha256); + + len = kex_pkp_create(&kex, &pkp1, buf1); + if (len < 0) { + printf("Failed to create first key pair for %s.\n", algo); + goto fail; + } + + pk1.len = (size_t) len; + pk1.data = buf1; + + len = kex_pkp_create(&kex, &pkp2, buf2); + if (len < 0) { + printf("Failed to create second key pair for %s.\n", algo); + goto fail_pkp1; + } + + pk2.len = (size_t) len; + pk2.data = buf2; + + if (kex_dhe_derive(&kex, pkp1, pk2, s1) < 0) { + printf("Failed to derive first key for %s.\n", algo); + goto fail_pkp2; + } + + if (kex_dhe_derive(&kex, pkp2, pk1, s2) < 0) { + printf("Failed to derive second key for %s.\n", algo); + goto fail_pkp2; + } + + if (memcmp(s1, s2, SYMMKEYSZ) != 0) { + printf("Derived keys do not match for %s.\n", algo); + goto fail_pkp2; + } + + kex_pkp_destroy(pkp2); + kex_pkp_destroy(pkp1); + + TEST_SUCCESS("(%s)", algo); + + return TEST_RC_SUCCESS; + fail_pkp2: + kex_pkp_destroy(pkp2); + fail_pkp1: + kex_pkp_destroy(pkp1); + fail: + TEST_FAIL("(%s)", algo); + return TEST_RC_FAIL; +} + +static int test_kex_validate_algo(void) +{ + TEST_START(); + + if (kex_validate_algo("prime256v1") != 0) { + printf("prime256v1 should be valid.\n"); + goto fail; + } + + if (kex_validate_algo("X25519") != 0) { + printf("X25519 should be valid.\n"); + goto fail; + } + +#ifdef HAVE_OPENSSL_ML_KEM + if (kex_validate_algo("ML-KEM-768") != 0) { + printf("ML-KEM-768 should be valid.\n"); + goto fail; + } +#endif + + if (kex_validate_algo("ffdhe2048") != 0) { + printf("ffdhe2048 should be valid.\n"); + goto fail; + } + + if (kex_validate_algo("invalid_algo") == 0) { + printf("invalid_algo should be rejected.\n"); + goto fail; + } + + if (kex_validate_algo("rsa2048") == 0) { + printf("rsa2048 should be rejected.\n"); + goto fail; + } + + if (kex_validate_algo(NULL) == 0) { + printf("NULL should be rejected.\n"); + goto fail; + } + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_kex_dhe_corrupted_pubkey(const char * algo) +{ + struct sec_config kex; + void * pkp; + buffer_t pk; + ssize_t len; + uint8_t buf[CRYPT_KEY_BUFSZ]; + uint8_t s[SYMMKEYSZ]; + + TEST_START("(%s)", algo); + + memset(&kex, 0, sizeof(kex)); + SET_KEX_ALGO(&kex, algo); + SET_KEX_KDF_NID(&kex, NID_sha256); + + len = kex_pkp_create(&kex, &pkp, buf); + if (len < 0) { + printf("Failed to create key pair.\n"); + goto fail; + } + + pk.len = (size_t) len; + pk.data = buf; + + /* Corrupt the public key */ + buf[0] ^= 0xFF; + buf[len - 1] ^= 0xFF; + + if (kex_dhe_derive(&kex, pkp, pk, s) == 0) { + printf("Should fail with corrupted public key.\n"); + goto fail_pkp; + } + + kex_pkp_destroy(pkp); + + TEST_SUCCESS("(%s)", algo); + + return TEST_RC_SUCCESS; + fail_pkp: + kex_pkp_destroy(pkp); + fail: + TEST_FAIL("(%s)", algo); + return TEST_RC_FAIL; +} + +static int test_kex_dhe_wrong_algo(void) +{ + struct sec_config kex1; + struct sec_config kex2; + void * pkp1; + void * pkp2; + buffer_t pk2; + ssize_t len; + uint8_t buf1[CRYPT_KEY_BUFSZ]; + uint8_t buf2[CRYPT_KEY_BUFSZ]; + uint8_t s[SYMMKEYSZ]; + const char * algo1 = "X25519"; + const char * algo2 = "X448"; + + TEST_START("(%s vs %s)", algo1, algo2); + + memset(&kex1, 0, sizeof(kex1)); + memset(&kex2, 0, sizeof(kex2)); + SET_KEX_ALGO(&kex1, algo1); + SET_KEX_ALGO(&kex2, algo2); + SET_KEX_KDF_NID(&kex1, NID_sha256); + SET_KEX_KDF_NID(&kex2, NID_sha256); + + if (kex_pkp_create(&kex1, &pkp1, buf1) < 0) { + printf("Failed to create first key pair.\n"); + goto fail; + } + + len = kex_pkp_create(&kex2, &pkp2, buf2); + if (len < 0) { + printf("Failed to create second key pair.\n"); + goto fail_pkp1; + } + + pk2.len = (size_t) len; + pk2.data = buf2; + + /* Try to derive with mismatched algorithms */ + if (kex_dhe_derive(&kex1, pkp1, pk2, s) == 0) { + printf("Should fail with mismatched algorithms.\n"); + goto fail_pkp2; + } + + kex_pkp_destroy(pkp2); + kex_pkp_destroy(pkp1); + + TEST_SUCCESS("(%s vs %s)", algo1, algo2); + + return TEST_RC_SUCCESS; + fail_pkp2: + kex_pkp_destroy(pkp2); + fail_pkp1: + kex_pkp_destroy(pkp1); + fail: + TEST_FAIL("(%s vs %s)", algo1, algo2); + return TEST_RC_FAIL; +} + +static int test_kex_load_dhe_privkey(void) +{ + void * key; + + TEST_START(); + + if (crypt_load_privkey_str(X25519_PRIVKEY_PEM, &key) < 0) { + printf("Failed to load X25519 private key.\n"); + goto fail; + } + + crypt_free_key(key); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_kex_load_dhe_pubkey(void) +{ + void * key; + + TEST_START(); + + if (crypt_load_pubkey_str(X25519_PUBKEY_PEM, &key) < 0) { + printf("Failed to load X25519 public key.\n"); + goto fail; + } + + crypt_free_key(key); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +#ifdef HAVE_OPENSSL +#include <openssl/obj_mac.h> + +static int test_kex_nid_values(void) +{ + int i; + + TEST_START(); + + /* Verify all KEX algorithm NIDs match OpenSSL's */ + for (i = 0; kex_supported_nids[i] != NID_undef; i++) { + uint16_t our_nid = kex_supported_nids[i]; + const char * kex_name; + int openssl_nid; + + kex_name = kex_nid_to_str(our_nid); + if (kex_name == NULL) { + printf("kex_nid_to_str failed for NID %u\n", our_nid); + goto fail; + } + + /* Test reverse conversion */ + if (kex_str_to_nid(kex_name) != our_nid) { + printf("kex_str_to_nid failed for '%s'\n", kex_name); + goto fail; + } + + /* Get OpenSSL's NID for this name */ + openssl_nid = OBJ_txt2nid(kex_name); + if (openssl_nid != NID_undef) { + /* OpenSSL recognizes this algorithm */ + if (our_nid != openssl_nid) { + printf("NID mismatch for '%s': " + "ours=%d, OpenSSL=%d\n", + kex_name, our_nid, openssl_nid); + goto fail; + } + } else { + /* Verify no NID collision with different algorithm */ + const char * ossl_name = OBJ_nid2sn(our_nid); + if (ossl_name != NULL && + strcmp(ossl_name, kex_name) != 0) { + printf("NID collision for '%d': " + "ours=%s, OpenSSL=%s\n", + our_nid, kex_name, ossl_name); + goto fail; + } + } + } + + /* Test error cases */ + if (kex_str_to_nid("invalid") != NID_undef) { + printf("kex_str_to_nid should return NID_undef for invalid\n"); + goto fail; + } + + if (kex_nid_to_str(9999) != NULL) { + printf("kex_nid_to_str should return NULL for invalid NID\n"); + goto fail; + } + + if (kex_str_to_nid(NULL) != NID_undef) { + printf("kex_str_to_nid should return NID_undef for NULL\n"); + goto fail; + } + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} +#endif + +static int test_kex_all(void) +{ + int ret = 0; + int i; + + for (i = 0; kex_supported_nids[i] != NID_undef; i++) { + const char * algo = kex_nid_to_str(kex_supported_nids[i]); + + /* KEM tests are in kex_test_ml_kem.c */ + if (IS_KEM_ALGORITHM(algo)) + continue; + + ret |= test_kex_dhe_derive(algo); + } + + return ret; +} + +static int test_kex_dhe_corrupted_pubkey_all(void) +{ + int ret = 0; + int i; + + /* Test corruption for all DHE algorithms */ + /* KEM error injection tests are in kex_test_ml_kem.c */ + for (i = 0; kex_supported_nids[i] != NID_undef; i++) { + const char * algo = kex_nid_to_str(kex_supported_nids[i]); + + if (IS_KEM_ALGORITHM(algo)) + continue; + + ret |= test_kex_dhe_corrupted_pubkey(algo); + } + + return ret; +} + +static int test_kex_parse_config_empty(void) +{ + struct sec_config kex; + FILE * fp; + + TEST_START(); + + memset(&kex, 0, sizeof(kex)); + + fp = FMEMOPEN_STR("\n"); + if (fp == NULL) { + printf("Failed to open memory stream.\n"); + goto fail; + } + + if (parse_sec_config(&kex, fp) < 0) { + printf("Failed to parse empty config.\n"); + fclose(fp); + goto fail; + } + + if (strcmp(kex.x.str, "prime256v1") != 0) { + printf("Empty config should use prime256v1.\n"); + fclose(fp); + goto fail; + } + + fclose(fp); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_kex_parse_config_custom(void) +{ + struct sec_config kex; + FILE * fp; + + TEST_START(); + + memset(&kex, 0, sizeof(kex)); + + fp = FMEMOPEN_STR(KEX_CONFIG_CUSTOM); + if (fp == NULL) { + printf("Failed to open memory stream.\n"); + goto fail; + } + + if (parse_sec_config(&kex, fp) < 0) { + printf("Failed to parse custom config.\n"); + fclose(fp); + goto fail; + } + + if (strcmp(kex.x.str, "X25519") != 0) { + printf("Algorithm not set correctly.\n"); + fclose(fp); + goto fail; + } + + fclose(fp); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* The old bare 'none' keyword must be rejected loudly */ +static int test_kex_parse_config_none_rejected(void) +{ + struct sec_config kex; + FILE * fp; + + TEST_START(); + + memset(&kex, 0, sizeof(kex)); + + fp = FMEMOPEN_STR(KEX_CONFIG_NONE); + if (fp == NULL) { + printf("Failed to open memory stream.\n"); + goto fail; + } + + if (parse_sec_config(&kex, fp) == 0) { + printf("Bare 'none' keyword should be rejected.\n"); + fclose(fp); + goto fail; + } + + fclose(fp); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_kex_parse_config_no_enc(void) +{ + struct sec_config kex; + FILE * fp; + + TEST_START(); + + memset(&kex, 0, sizeof(kex)); + + fp = FMEMOPEN_STR(KEX_CONFIG_NO_ENC); + if (fp == NULL) { + printf("Failed to open memory stream.\n"); + goto fail; + } + + if (parse_sec_config(&kex, fp) < 0) { + printf("Failed to parse encryption=none config.\n"); + fclose(fp); + goto fail; + } + + if (kex.x.nid != NID_undef || kex.c.nid != NID_undef) { + printf("encryption=none should disable encryption.\n"); + fclose(fp); + goto fail; + } + + if (kex.d.nid != NID_sha256) { + printf("encryption=none should keep the digest.\n"); + fclose(fp); + goto fail; + } + + fclose(fp); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_kex_parse_config_whitespace(void) +{ + struct sec_config kex; + FILE * fp; + + TEST_START(); + + memset(&kex, 0, sizeof(kex)); + + fp = FMEMOPEN_STR(KEX_CONFIG_WHITESPACE); + if (fp == NULL) { + printf("Failed to open memory stream.\n"); + goto fail; + } + + if (parse_sec_config(&kex, fp) < 0) { + printf("Failed to parse config with comments.\n"); + fclose(fp); + goto fail; + } + + if (strcmp(kex.x.str, "X448") != 0) { + printf("Algorithm with whitespace not parsed correctly.\n"); + fclose(fp); + goto fail; + } + + fclose(fp); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_kex_parse_config_cipher(void) +{ + struct sec_config kex; + FILE * fp; + + TEST_START(); + + memset(&kex, 0, sizeof(kex)); + + fp = FMEMOPEN_STR(KEX_CONFIG_CIPHER); + if (fp == NULL) { + printf("Failed to open memory stream.\n"); + goto fail; + } + + if (parse_sec_config(&kex, fp) < 0) { + printf("Failed to parse cipher config.\n"); + fclose(fp); + goto fail; + } + + if (strcmp(kex.x.str, "X25519") != 0) { + printf("Algorithm not set correctly.\n"); + fclose(fp); + goto fail; + } + + if (kex.c.nid != NID_chacha20_poly1305) { + printf("Cipher not set correctly.\n"); + fclose(fp); + goto fail; + } + + fclose(fp); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_kex_parse_config_digest(void) +{ + struct sec_config kex; + FILE * fp; + + TEST_START(); + + memset(&kex, 0, sizeof(kex)); + + fp = FMEMOPEN_STR(KEX_CONFIG_DIGEST); + if (fp == NULL) { + printf("Failed to open memory stream.\n"); + goto fail; + } + + if (parse_sec_config(&kex, fp) < 0) { + printf("Failed to parse digest config.\n"); + fclose(fp); + goto fail; + } + + if (strcmp(kex.x.str, "X25519") != 0) { + printf("Algorithm not set correctly.\n"); + fclose(fp); + goto fail; + } + + if (kex.d.nid != NID_sha384) { + printf("Digest not set correctly.\n"); + fclose(fp); + goto fail; + } + + fclose(fp); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_kex_parse_config_auth(void) +{ + struct sec_config kex; + FILE * fp; + + TEST_START(); + + memset(&kex, 0, sizeof(kex)); + + fp = FMEMOPEN_STR(KEX_CONFIG_AUTH); + if (fp == NULL) { + printf("Failed to open memory stream.\n"); + goto fail; + } + + if (parse_sec_config(&kex, fp) < 0) { + printf("Failed to parse auth config.\n"); + fclose(fp); + goto fail; + } + + if (!kex.a.req) { + printf("auth=required not parsed correctly.\n"); + fclose(fp); + goto fail; + } + + fclose(fp); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_kex_parse_config_auth_invalid(void) +{ + struct sec_config kex; + FILE * fp; + + TEST_START(); + + memset(&kex, 0, sizeof(kex)); + + fp = FMEMOPEN_STR(KEX_CONFIG_AUTH_INVALID); + if (fp == NULL) { + printf("Failed to open memory stream.\n"); + goto fail; + } + + if (parse_sec_config(&kex, fp) == 0) { + printf("Invalid auth value should be rejected.\n"); + fclose(fp); + goto fail; + } + + fclose(fp); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* A caller-seeded req_auth survives parsing when no auth= line is set */ +static int test_kex_parse_config_auth_seed(void) +{ + struct sec_config kex; + FILE * fp; + + TEST_START(); + + memset(&kex, 0, sizeof(kex)); + kex.a.req = true; + + fp = FMEMOPEN_STR(KEX_CONFIG_NO_ENC); + if (fp == NULL) { + printf("Failed to open memory stream.\n"); + goto fail; + } + + if (parse_sec_config(&kex, fp) < 0) { + printf("Failed to parse config.\n"); + fclose(fp); + goto fail; + } + + if (!kex.a.req) { + printf("Seeded req_auth should survive parsing.\n"); + fclose(fp); + goto fail; + } + + fclose(fp); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* An explicit auth=optional clears a caller-seeded req_auth */ +static int test_kex_parse_config_auth_optional(void) +{ + struct sec_config kex; + FILE * fp; + + TEST_START(); + + memset(&kex, 0, sizeof(kex)); + kex.a.req = true; + + fp = FMEMOPEN_STR(KEX_CONFIG_AUTH_OPTIONAL); + if (fp == NULL) { + printf("Failed to open memory stream.\n"); + goto fail; + } + + if (parse_sec_config(&kex, fp) < 0) { + printf("Failed to parse auth=optional config.\n"); + fclose(fp); + goto fail; + } + + if (kex.a.req) { + printf("auth=optional should clear req_auth.\n"); + fclose(fp); + goto fail; + } + + fclose(fp); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* encryption=none must not drop auth=required or the digest */ +static int test_kex_parse_config_auth_no_enc(const char * config) +{ + struct sec_config kex; + FILE * fp; + + TEST_START(); + + memset(&kex, 0, sizeof(kex)); + + fp = FMEMOPEN_STR(config); + if (fp == NULL) { + printf("Failed to open memory stream.\n"); + goto fail; + } + + if (parse_sec_config(&kex, fp) < 0) { + printf("Failed to parse auth + encryption=none.\n"); + fclose(fp); + goto fail; + } + + if (!kex.a.req) { + printf("encryption=none should not drop required auth.\n"); + fclose(fp); + goto fail; + } + + if (kex.x.nid != NID_undef) { + printf("encryption=none should disable encryption.\n"); + fclose(fp); + goto fail; + } + + if (kex.d.nid != NID_sha512) { + printf("encryption=none should keep the digest.\n"); + fclose(fp); + goto fail; + } + + fclose(fp); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_kex_parse_config_cacert(void) +{ + struct sec_config kex; + FILE * fp; + + TEST_START(); + + memset(&kex, 0, sizeof(kex)); + + fp = FMEMOPEN_STR(KEX_CONFIG_CACERT); + if (fp == NULL) { + printf("Failed to open memory stream.\n"); + goto fail; + } + + if (parse_sec_config(&kex, fp) < 0) { + printf("Failed to parse cacert config.\n"); + fclose(fp); + goto fail; + } + + if (strcmp(kex.a.cacert, + "/etc/ouroboros/security/cacert/ca.crt") != 0) { + printf("cacert not parsed correctly.\n"); + fclose(fp); + goto fail; + } + + if (kex.a.req) { + printf("cacert must not imply req_auth.\n"); + fclose(fp); + goto fail; + } + + fclose(fp); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_kex_parse_config_unknown_key(void) +{ + struct sec_config kex; + FILE * fp; + + TEST_START(); + + memset(&kex, 0, sizeof(kex)); + + fp = FMEMOPEN_STR(KEX_CONFIG_UNKNOWN_KEY); + if (fp == NULL) { + printf("Failed to open memory stream.\n"); + goto fail; + } + + if (parse_sec_config(&kex, fp) == 0) { + printf("Unknown key should be rejected.\n"); + fclose(fp); + goto fail; + } + + fclose(fp); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +int kex_test(int argc, + char ** argv) +{ + int ret = 0; + + (void) argc; + (void) argv; + + ret |= test_kex_create_destroy(); + ret |= test_kex_parse_config_empty(); + ret |= test_kex_parse_config_none_rejected(); + ret |= test_kex_parse_config_no_enc(); + ret |= test_kex_parse_config_auth(); + ret |= test_kex_parse_config_auth_invalid(); + ret |= test_kex_parse_config_auth_seed(); + ret |= test_kex_parse_config_auth_optional(); + ret |= test_kex_parse_config_auth_no_enc(KEX_CONFIG_AUTH_THEN_NO_ENC); + ret |= test_kex_parse_config_auth_no_enc(KEX_CONFIG_NO_ENC_THEN_AUTH); + ret |= test_kex_parse_config_cacert(); + ret |= test_kex_parse_config_unknown_key(); +#ifdef HAVE_OPENSSL + ret |= test_kex_parse_config_custom(); + ret |= test_kex_parse_config_whitespace(); + ret |= test_kex_parse_config_cipher(); + ret |= test_kex_parse_config_digest(); + ret |= test_kex_nid_values(); + ret |= test_kex_dh_pkp_create_destroy(); + ret |= test_kex_all(); + ret |= test_kex_validate_algo(); + ret |= test_kex_get_algo_from_pk_all(); + ret |= test_kex_dhe_wrong_algo(); + ret |= test_kex_dhe_corrupted_pubkey_all(); + ret |= test_kex_load_dhe_privkey(); + ret |= test_kex_load_dhe_pubkey(); +#else + (void) test_kex_parse_config_custom; + (void) test_kex_parse_config_whitespace; + (void) test_kex_parse_config_cipher; + (void) test_kex_parse_config_digest; + (void) test_kex_dh_pkp_create_destroy; + (void) test_kex_all; + (void) test_kex_validate_algo; + (void) test_kex_get_algo_from_pk_all; + (void) test_kex_dhe_wrong_algo(); + (void) test_kex_dhe_corrupted_pubkey_all; + (void) test_kex_load_dhe_privkey; + (void) test_kex_load_dhe_pubkey; + + ret = TEST_RC_SKIP; +#endif + return ret; +} diff --git a/src/lib/tests/kex_test_ml_kem.c b/src/lib/tests/kex_test_ml_kem.c new file mode 100644 index 00000000..7761c3dc --- /dev/null +++ b/src/lib/tests/kex_test_ml_kem.c @@ -0,0 +1,549 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Test of the post-quantum key exchange functions + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#define _POSIX_C_SOURCE 200809L + +#include "config.h" + +#include <test/test.h> +#include <ouroboros/utils.h> +#include <ouroboros/crypt.h> +#include <ouroboros/random.h> + +#include <stdio.h> +#include <stdlib.h> +#include <unistd.h> + +#ifdef HAVE_OPENSSL +#include <openssl/evp.h> +#include <openssl/x509.h> +#endif + +extern const uint16_t kex_supported_nids[]; +extern const uint16_t md_supported_nids[]; + +static int get_random_kdf(void) +{ + static int idx = 0; + int count; + + if (md_supported_nids[0] == NID_undef) + return NID_undef; + + for (count = 0; md_supported_nids[count] != NID_undef; count++) + ; + + return md_supported_nids[(idx++) % count]; +} + +/* ML-KEM-768 test key material */ + +#define MLKEM768_PRIVKEY_PEM \ + "-----BEGIN PRIVATE KEY-----\n" \ + "MIIJvgIBADALBglghkgBZQMEBAIEggmqMIIJpgRA+QIIiQLQkS5fl5RluSmgXRjZ\n" \ + "YU16W4TVt0dmnBP41rLTTRT3S8CRtkb+xmoFAcWTfEzbdr5pp3g2CBRx+APXTwSC\n" \ + "CWBll6AecTd1Kqdyix3zNQcthDBP0XnwdTHDqkKuFzMP58Y+0gc9Bo+W0xBOK2ZK\n" \ + "gcAmix3YLJuDS8Teep/Tdc7KIm5AaLNoI8BIMgKC/ASsW8kC+78BV4OIgqNWurS9\n" \ + "BrTiCmiag7c+6DsVDJHJ4kfcccwUDBKiW0v+LAkk1HXBcx6usrwuFC0H3ICli2sC\n" \ + "o5DfGL7g4kWHhobXjAZnxn298C8FGmLQK5kah4nZiJ+MuHqrirziCGTLKkY1a8vC\n" \ + "GFgzfHIcvB4dtyi9dxZmWpSXqDf2AVNgqrD2C7WQEULQOKxm/I8Mw31Yp8TC6SAP\n" \ + "RzM4cBAXF00W4Rce05O0am/ga5dStAhikMESyckCoEGlPFFXOmjy1HmOasI+AbGk\n" \ + "2BKp6cfbImbjd0ePdCSFEgIQwAQHm7+4UoZR2JmNwSI1AC2P4FMRAIaD2A69i6LC\n" \ + "kFniGcOog5m09nw5FqZmeEfNs6yyFGSX16D1YyjuooAFGlU0FFX7aKwsYM8t1gkS\n" \ + "YSUfMxIW9yzhSW4vZHuGyxlxBMr1y51RZrW8gnvW5p/Ip5yDBJRahY6KMWT15C14\n" \ + "C2rIe8U+d4Xi5IMI3D1JNpwFebYhKs3/ManxoU7Fwwa0GzQrgLYU5KhqO8/hopnl\n" \ + "8mQH+BPh+TR5lqYawS7HZXFJE8JzOnCtOSgB6Hz2U7oG9ik8h0FRqVD3ak20EmZU\n" \ + "c7gpGW8Odc51uaIBzDu4ej4dGgwo4awYaX4ugLOutHqGqRfCjIVb6XQ4m35p4KKi\n" \ + "qBVQ211aIhavUIgNECJ7WUETilXyyHLB9x3EFJdidEfSRUxLYJNAC5XM2WFCyhnE\n" \ + "pKmossSNq6ZOqBjPegE0J6zfNg65dR/OlIdGVDgrVTIpwYAUzBMW2nTnCa00EmPj\n" \ + "F7tRscHI8qb/QlnRVEUN+S+A2CtVIH1c666zOoRFRI9G4bmVoa8k2x0ANB51tCns\n" \ + "vAYqkMybIgMvWwbqoAxeW0G1O3qObGXtgs94BzhAEM3RbG/hy3GR1qUNSk/qyDKc\n" \ + "t1qpiaao0aLVsnpb28eBIk6+q0I82reGdV31OYvUpnVxRbRPFXEFs5PNS3s/7I8a\n" \ + "SlSLUGOh+mhrUzDPSJCzgEvOmrwrRxe3F52tS0nAt6Z5zKToASHphoISUi7lGX1F\n" \ + "Owx62qhSqqlI98bKqh7yQRZYrHXqE0bscAHCcIaZ8RVya42JHDCoQWyxqBuLOWEl\n" \ + "+Fz6vI5DqEnJkA7ke49EvBAOJ58lxAXQIV5remtzYGPKdyG2oamiFHiLVQDzGX/l\n" \ + "aFNMGXRWcK4/Y3mnkJvx9QGtq6KstQN/J4a51ZeX5YwNBcoY9UcFS6kHRW5rR3UM\n" \ + "tEZj5VN8BL9nyWM9h7hUSHQboaxO7M5qswfXB8f21xR16T40Ki4nawx/6zHGCQsc\n" \ + "uKr5SaCV88tghqJYHBorU5iKB5KsLDSHqYYrNo/Vy8W6kMA2jGAO24d4G32DSshR\n" \ + "sEF9W1nuAHK/5ste01G5KmX2KhdZBE37oGhM98HRQ6hU8qwuKrhdV7vZis5C8LXY\n" \ + "7MbDyDt1NnFqWFc6lYeVa6eRcmYzeAbXahrxwiiaLIdHXD95aZ/0S6+tKBGgQzwm\n" \ + "ZsbwdXhl+n+yqDNE6Sow2bwueqhDwZVWoMCv5SK+HAGPtcZ7UU9oWrqpiL085m7F\n" \ + "5G49KJUEZadVtj4Z9zrkeQkida+4I7v3Y3MzsWsGJww7YhTDJpsxxmSm85bHwx98\n" \ + "hZXSqckJTL4c2nBzgrBlukIT9Wl+qItMthVvABPzp4wGZhdgKrEIRl3yCnhhUgpL\n" \ + "lUxYegwWDMEjZxKlSbIyl5p9lCS8w2lsBzsQ2FJiAy/MWLa56aA+wFs3C8smZ6Cf\n" \ + "p5NWa8Rm+k898GWBxZivhF03CBOZ42du0YUZdCPoA5V1KC6bh4JyWFI49VFbQFMG\n" \ + "gwAqc0ErAH3iMammKC9746WWagnUIG3o8LygZrusuGeTohXJhVUTJDw2s0rzNhbw\n" \ + "5IyookkY5BWENKFKTIgdBxvYelOKwbGE8Z36FEW0ABlmx7SRCKWlNVjSEAIXmMiQ\n" \ + "VLdQF33QVYD9RR5chja254VuJH4plo+5JwiKWz8LlCIBm7CVkifZMLofmMk3s3L4\n" \ + "sXtE+Bhfm5Plk3RrgDdlHH+hK7gk61XGdynGjDY7aLtCKZ0SMsVskSLom1pbIR5M\n" \ + "KLYsQ1Pse4mhfDOFCkWFLI5TShGMuIoo1k7XeIE6g8QoUlV5EXyWHHhIVaE4yWGP\n" \ + "AVgEp0UswKFeeo3SoCAeADA3U88ymxpBJp73yDIqok5dM3SgkjfPWZDkgkAI8WHs\n" \ + "CKKeqrSOs1kkE3JXtE7kcTHT6XHo162TmgGkqMVwOQ3EmR6FRpYxJhZvuVbjJsSx\n" \ + "YjW3ScnR4Zivoi7q95ypco331pIlIZpqV0NydUpMyQaz1cnoPKYDh1xa6LhcqEKK\n" \ + "8a68iXjQgzgqQBDABonVybNDtlJ5lnTTuKhak8PBFAmmhj1JdrPqoIvQRCmLaark\n" \ + "J7/q9RLtk6kTOJ0qtLe2qqwCxJwyoMd2Q5F4+xTWZHu90ljRdcnYewarqcKzoL27\n" \ + "tcpTOmVz88I1hYVUJEV7aB36QMhTS1dquTqJZCD0hBPWAMToEoD4OFvKWmbFmzaW\n" \ + "xrMc4ECYeDAAKYs2YqoXSLfAixBmZjb6UDB61l2GA58pFJW0ZwN8S5tApA2NRi+7\n" \ + "oC/zgMgBGHft6E0+OUVb8It89pY1t7ybq5+fkBvEixDId3f1pK3gqcaYqG/YhoMJ\n" \ + "MJWkqYxCNGmdZ8gFo46V6K+4xZUblQWKypN6+RYO4kDh0koppWGEULjgBoCH+V8E\n" \ + "7GcoE8SRdQY1BIMoRVWb8Ur8ZYIVU8lqgaZPlWM3oRCiWk0kRxexFF0i5WlILIK9\n" \ + "GT8saX+bmRd9KSy3JrpPhQn59CpJBRxz8WKdJ3wwtqE/2TbxQhLooEWHYVrZEG5E\n" \ + "SkIoOkUAJUR+CzLLFDMdUE8w3CasE4ys+hco7AA5TAms24A1FXcxMgNb6VHA0bi5\n" \ + "c8rPCZvjubLXR4A0/A2Ualo4cy3UAr9k0rbZOJnjqk8eExkeaxbyh42cJpU75i4O\n" \ + "NLYsRZJkg9bkCpPgZKb707sPZO72CX3h/lQdXVgGkZ7Tqd1qzM+JOhSWvrYiBLa+\n" \ + "5IKSmFwT+5sw1InEesXwRN09000U90vAkbZG/sZqBQHFk3xM23a+aad4NggUcfgD\n" \ + "108=\n" \ + "-----END PRIVATE KEY-----\n" + +#define MLKEM768_PUBKEY_PEM \ + "-----BEGIN PUBLIC KEY-----\n" \ + "MIIEsjALBglghkgBZQMEBAIDggShAMPIO3U2cWpYVzqVh5Vrp5FyZjN4BtdqGvHC\n" \ + "KJosh0dcP3lpn/RLr60oEaBDPCZmxvB1eGX6f7KoM0TpKjDZvC56qEPBlVagwK/l\n" \ + "Ir4cAY+1xntRT2hauqmIvTzmbsXkbj0olQRlp1W2Phn3OuR5CSJ1r7gju/djczOx\n" \ + "awYnDDtiFMMmmzHGZKbzlsfDH3yFldKpyQlMvhzacHOCsGW6QhP1aX6oi0y2FW8A\n" \ + "E/OnjAZmF2AqsQhGXfIKeGFSCkuVTFh6DBYMwSNnEqVJsjKXmn2UJLzDaWwHOxDY\n" \ + "UmIDL8xYtrnpoD7AWzcLyyZnoJ+nk1ZrxGb6Tz3wZYHFmK+EXTcIE5njZ27RhRl0\n" \ + "I+gDlXUoLpuHgnJYUjj1UVtAUwaDACpzQSsAfeIxqaYoL3vjpZZqCdQgbejwvKBm\n" \ + "u6y4Z5OiFcmFVRMkPDazSvM2FvDkjKiiSRjkFYQ0oUpMiB0HG9h6U4rBsYTxnfoU\n" \ + "RbQAGWbHtJEIpaU1WNIQAheYyJBUt1AXfdBVgP1FHlyGNrbnhW4kfimWj7knCIpb\n" \ + "PwuUIgGbsJWSJ9kwuh+YyTezcvixe0T4GF+bk+WTdGuAN2Ucf6EruCTrVcZ3KcaM\n" \ + "Njtou0IpnRIyxWyRIuibWlshHkwotixDU+x7iaF8M4UKRYUsjlNKEYy4iijWTtd4\n" \ + "gTqDxChSVXkRfJYceEhVoTjJYY8BWASnRSzAoV56jdKgIB4AMDdTzzKbGkEmnvfI\n" \ + "MiqiTl0zdKCSN89ZkOSCQAjxYewIop6qtI6zWSQTcle0TuRxMdPpcejXrZOaAaSo\n" \ + "xXA5DcSZHoVGljEmFm+5VuMmxLFiNbdJydHhmK+iLur3nKlyjffWkiUhmmpXQ3J1\n" \ + "SkzJBrPVyeg8pgOHXFrouFyoQorxrryJeNCDOCpAEMAGidXJs0O2UnmWdNO4qFqT\n" \ + "w8EUCaaGPUl2s+qgi9BEKYtpquQnv+r1Eu2TqRM4nSq0t7aqrALEnDKgx3ZDkXj7\n" \ + "FNZke73SWNF1ydh7BqupwrOgvbu1ylM6ZXPzwjWFhVQkRXtoHfpAyFNLV2q5Oolk\n" \ + "IPSEE9YAxOgSgPg4W8paZsWbNpbGsxzgQJh4MAApizZiqhdIt8CLEGZmNvpQMHrW\n" \ + "XYYDnykUlbRnA3xLm0CkDY1GL7ugL/OAyAEYd+3oTT45RVvwi3z2ljW3vJurn5+Q\n" \ + "G8SLEMh3d/WkreCpxpiob9iGgwkwlaSpjEI0aZ1nyAWjjpXor7jFlRuVBYrKk3r5\n" \ + "Fg7iQOHSSimlYYRQuOAGgIf5XwTsZygTxJF1BjUEgyhFVZvxSvxlghVTyWqBpk+V\n" \ + "YzehEKJaTSRHF7EUXSLlaUgsgr0ZPyxpf5uZF30pLLcmuk+FCfn0KkkFHHPxYp0n\n" \ + "fDC2oT/ZNvFCEuigRYdhWtkQbkRKQig6RQAlRH4LMssUMx1QTzDcJqwTjKz6Fyjs\n" \ + "ADlMCazbgDUVdzEyA1vpUcDRuLlzys8Jm+O5stdHgDT8DZRqWjhzLdQCv2TSttk4\n" \ + "meOqTx4TGR5rFvKHjZwmlTvmLg40tixFkmSD1uQKk+BkpvvTuw9k7vYJfeH+VB1d\n" \ + "WAaRntOp\n" \ + "-----END PUBLIC KEY-----\n" + +/* Helper macro to open string constant as FILE stream */ +#define FMEMOPEN_STR(str) fmemopen((void *) (str), strlen(str), "r") + +static int test_kex_load_kem_privkey(void) +{ + void * key; + + TEST_START(); + + if (crypt_load_privkey_str(MLKEM768_PRIVKEY_PEM, &key) < 0) { + printf("Failed to load ML-KEM-768 private key.\n"); + goto fail; + } + + crypt_free_key(key); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_kex_load_kem_pubkey(void) +{ + void * key; + + TEST_START(); + + if (crypt_load_pubkey_str(MLKEM768_PUBKEY_PEM, &key) < 0) { + printf("Failed to load ML-KEM-768 public key.\n"); + goto fail; + } + + crypt_free_key(key); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_kex_kem(const char * algo) +{ + struct sec_config kex; + void * pkp; + buffer_t pk; + buffer_t ct; + ssize_t len; + ssize_t ct_len; + uint8_t buf1[CRYPT_KEY_BUFSZ]; + uint8_t buf2[CRYPT_KEY_BUFSZ]; + uint8_t s1[SYMMKEYSZ]; + uint8_t s2[SYMMKEYSZ]; + int kdf; + + TEST_START("(%s)", algo); + + kdf = get_random_kdf(); + + memset(&kex, 0, sizeof(kex)); + SET_KEX_ALGO(&kex, algo); + + len = kex_pkp_create(&kex, &pkp, buf1); + if (len < 0) { + printf("Failed to create key pair for %s.\n", algo); + goto fail; + } + + pk.len = (size_t) len; + pk.data = buf1; + + if (IS_HYBRID_KEM(algo)) + ct_len = kex_kem_encap_raw(pk, buf2, kdf, s1); + else + ct_len = kex_kem_encap(pk, buf2, kdf, s1); + + if (ct_len < 0) { + printf("Failed to encapsulate for %s.\n", algo); + goto fail_pkp; + } + + ct.len = (size_t) ct_len; + ct.data = buf2; + + if (kex_kem_decap(pkp, ct, kdf, s2) < 0) { + printf("Failed to decapsulate for %s.\n", algo); + goto fail_pkp; + } + + if (memcmp(s1, s2, SYMMKEYSZ) != 0) { + printf("Shared secrets don't match for %s.\n", algo); + goto fail_pkp; + } + + kex_pkp_destroy(pkp); + + TEST_SUCCESS("(%s)", algo); + + return TEST_RC_SUCCESS; + fail_pkp: + kex_pkp_destroy(pkp); + fail: + TEST_FAIL("(%s)", algo); + return TEST_RC_FAIL; +} + +static int test_kex_kem_corrupted_ciphertext(const char * algo) +{ + struct sec_config kex; + void * pkp; + buffer_t pk; + buffer_t ct; + ssize_t len; + ssize_t ct_len; + uint8_t buf1[CRYPT_KEY_BUFSZ]; + uint8_t buf2[CRYPT_KEY_BUFSZ]; + uint8_t s1[SYMMKEYSZ]; + uint8_t s2[SYMMKEYSZ]; + int kdf; + + TEST_START("(%s)", algo); + + kdf = get_random_kdf(); + + memset(&kex, 0, sizeof(kex)); + SET_KEX_ALGO(&kex, algo); + + len = kex_pkp_create(&kex, &pkp, buf1); + if (len < 0) { + printf("Failed to create key pair.\n"); + goto fail; + } + + pk.len = (size_t) len; + pk.data = buf1; + + if (IS_HYBRID_KEM(algo)) + ct_len = kex_kem_encap_raw(pk, buf2, kdf, s1); + else + ct_len = kex_kem_encap(pk, buf2, kdf, s1); + + if (ct_len < 0) { + printf("Failed to encapsulate.\n"); + goto fail_pkp; + } + + ct.len = (size_t) ct_len; + ct.data = buf2; + + /* Corrupt the ciphertext */ + buf2[0] ^= 0xFF; + buf2[ct_len - 1] ^= 0xFF; + + /* ML-KEM uses implicit rejection */ + if (kex_kem_decap(pkp, ct, kdf, s2) < 0) { + printf("Decapsulation failed unexpectedly.\n"); + goto fail_pkp; + } + + /* The shared secrets should NOT match with corrupted CT */ + if (memcmp(s1, s2, SYMMKEYSZ) == 0) { + printf("Corrupted ciphertext produced same secret.\n"); + goto fail_pkp; + } + + kex_pkp_destroy(pkp); + + TEST_SUCCESS("(%s)", algo); + + return TEST_RC_SUCCESS; + fail_pkp: + kex_pkp_destroy(pkp); + fail: + TEST_FAIL("(%s)", algo); + return TEST_RC_FAIL; +} + +static int test_kex_kem_wrong_keypair(const char * algo) +{ + struct sec_config kex; + void * pkp1; + void * pkp2; + buffer_t pk1; + buffer_t ct; + ssize_t len; + ssize_t ct_len; + uint8_t buf1[CRYPT_KEY_BUFSZ]; + uint8_t buf2[CRYPT_KEY_BUFSZ]; + uint8_t buf3[CRYPT_KEY_BUFSZ]; + uint8_t s1[SYMMKEYSZ]; + uint8_t s2[SYMMKEYSZ]; + + TEST_START("(%s)", algo); + + memset(&kex, 0, sizeof(kex)); + SET_KEX_ALGO(&kex, algo); + + len = kex_pkp_create(&kex, &pkp1, buf1); + if (len < 0) { + printf("Failed to create first key pair.\n"); + goto fail; + } + + pk1.len = (size_t) len; + pk1.data = buf1; + + if (kex_pkp_create(&kex, &pkp2, buf2) < 0) { + printf("Failed to create second key pair.\n"); + goto fail_pkp1; + } + + if (IS_HYBRID_KEM(algo)) + ct_len = kex_kem_encap_raw(pk1, buf3, NID_sha256, s1); + else + ct_len = kex_kem_encap(pk1, buf3, NID_sha256, s1); + + if (ct_len < 0) { + printf("Failed to encapsulate.\n"); + goto fail_pkp2; + } + + ct.len = (size_t) ct_len; + ct.data = buf3; + + if (kex_kem_decap(pkp2, ct, NID_sha256, s2) == 0) { + if (memcmp(s1, s2, SYMMKEYSZ) == 0) { + printf("Wrong keypair produced same secret.\n"); + goto fail_pkp2; + } + } + + kex_pkp_destroy(pkp2); + kex_pkp_destroy(pkp1); + + TEST_SUCCESS("(%s)", algo); + + return TEST_RC_SUCCESS; + fail_pkp2: + kex_pkp_destroy(pkp2); + fail_pkp1: + kex_pkp_destroy(pkp1); + fail: + TEST_FAIL("(%s)", algo); + return TEST_RC_FAIL; +} + +static int test_kex_kem_truncated_ciphertext(const char * algo) +{ + struct sec_config kex; + void * pkp; + buffer_t pk; + buffer_t ct; + ssize_t len; + ssize_t ct_len; + uint8_t buf1[CRYPT_KEY_BUFSZ]; + uint8_t buf2[CRYPT_KEY_BUFSZ]; + uint8_t s1[SYMMKEYSZ]; + uint8_t s2[SYMMKEYSZ]; + + TEST_START("(%s)", algo); + + memset(&kex, 0, sizeof(kex)); + SET_KEX_ALGO(&kex, algo); + + len = kex_pkp_create(&kex, &pkp, buf1); + if (len < 0) { + printf("Failed to create key pair.\n"); + goto fail; + } + + pk.len = (size_t) len; + pk.data = buf1; + + if (IS_HYBRID_KEM(algo)) + ct_len = kex_kem_encap_raw(pk, buf2, NID_sha256, s1); + else + ct_len = kex_kem_encap(pk, buf2, NID_sha256, s1); + + if (ct_len < 0) { + printf("Failed to encapsulate.\n"); + goto fail_pkp; + } + + /* Truncate the ciphertext */ + ct.len = (size_t) ct_len / 2; + ct.data = buf2; + + if (kex_kem_decap(pkp, ct, NID_sha256, s2) == 0) { + printf("Should fail with truncated ciphertext.\n"); + goto fail_pkp; + } + + kex_pkp_destroy(pkp); + + TEST_SUCCESS("(%s)", algo); + + return TEST_RC_SUCCESS; + fail_pkp: + kex_pkp_destroy(pkp); + fail: + TEST_FAIL("(%s)", algo); + return TEST_RC_FAIL; +} + +static int test_kex_kem_all(void) +{ + int ret = 0; + int i; + + for (i = 0; kex_supported_nids[i] != NID_undef; i++) { + const char * algo = kex_nid_to_str(kex_supported_nids[i]); + + if (!IS_KEM_ALGORITHM(algo)) + continue; + + ret |= test_kex_kem(algo); + } + + return ret; +} + +static int test_kex_kem_corrupted_ciphertext_all(void) +{ + int ret = 0; + int i; + + for (i = 0; kex_supported_nids[i] != NID_undef; i++) { + const char * algo = kex_nid_to_str(kex_supported_nids[i]); + + if (!IS_KEM_ALGORITHM(algo)) + continue; + + ret |= test_kex_kem_corrupted_ciphertext(algo); + } + + return ret; +} + +static int test_kex_kem_wrong_keypair_all(void) +{ + int ret = 0; + int i; + + for (i = 0; kex_supported_nids[i] != NID_undef; i++) { + const char * algo = kex_nid_to_str(kex_supported_nids[i]); + + if (!IS_KEM_ALGORITHM(algo)) + continue; + + ret |= test_kex_kem_wrong_keypair(algo); + } + + return ret; +} + +static int test_kex_kem_truncated_ciphertext_all(void) +{ + int ret = 0; + int i; + + for (i = 0; kex_supported_nids[i] != NID_undef; i++) { + const char * algo = kex_nid_to_str(kex_supported_nids[i]); + + if (!IS_KEM_ALGORITHM(algo)) + continue; + + ret |= test_kex_kem_truncated_ciphertext(algo); + } + + return ret; +} + +int kex_test_ml_kem(int argc, + char ** argv) +{ + int ret = 0; + + (void) argc; + (void) argv; + +#ifdef HAVE_OPENSSL_ML_KEM + ret |= test_kex_load_kem_privkey(); + ret |= test_kex_load_kem_pubkey(); + ret |= test_kex_kem_all(); + ret |= test_kex_kem_corrupted_ciphertext_all(); + ret |= test_kex_kem_wrong_keypair_all(); + ret |= test_kex_kem_truncated_ciphertext_all(); +#else + (void) test_kex_load_kem_privkey; + (void) test_kex_load_kem_pubkey; + (void) test_kex_kem_all; + (void) test_kex_kem_corrupted_ciphertext_all; + (void) test_kex_kem_wrong_keypair_all; + (void) test_kex_kem_truncated_ciphertext_all; + + ret = TEST_RC_SKIP; +#endif + return ret; +} diff --git a/src/lib/tests/keyrot_test.c b/src/lib/tests/keyrot_test.c new file mode 100644 index 00000000..1c9f741b --- /dev/null +++ b/src/lib/tests/keyrot_test.c @@ -0,0 +1,1083 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Test of the key-rotation schedule + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#define _POSIX_C_SOURCE 200809L + +#include "config.h" + +#include <test/test.h> + +#ifdef HAVE_OPENSSL +#include <ouroboros/crypt.h> +#include <ouroboros/pthread.h> + +#include "crypt/keyrot.h" + +#include <stdbool.h> +#include <stdio.h> +#include <stdlib.h> +#include <string.h> +#include <time.h> + +static const uint8_t SEED_A[SYMMKEYSZ] = { + 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08, + 0x09, 0x0a, 0x0b, 0x0c, 0x0d, 0x0e, 0x0f, 0x10, + 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, 0x18, + 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f, 0x20 +}; + +static int test_create_destroy(void) +{ + struct keyrot * kr; + + TEST_START(); + + kr = keyrot_create(SEED_A, 0, 0); + if (kr == NULL) + goto fail; + + keyrot_destroy(kr); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_epoch_range(void) +{ + struct keyrot * a; + + TEST_START(); + + /* epoch is a 4-bit wire field; 16 and up must be refused. */ + if (keyrot_create(SEED_A, 16, 0) != NULL) + goto fail; + + a = keyrot_create(SEED_A, 0, 0); + if (a == NULL) + goto fail; + + if (keyrot_rekey(a, SEED_A, 16) == 0) + goto fail_a; + + keyrot_destroy(a); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_a: + keyrot_destroy(a); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_tx_deterministic(void) +{ + struct keyrot * a; + struct keyrot * b; + uint8_t sela[KR_SELECTOR_LEN]; + uint8_t selb[KR_SELECTOR_LEN]; + uint8_t na[KR_NONCE_LEN]; + uint8_t nb[KR_NONCE_LEN]; + const uint8_t * ka; + const uint8_t * kb; + + TEST_START(); + + a = keyrot_create(SEED_A, 0, 0); + if (a == NULL) + goto fail; + + b = keyrot_create(SEED_A, 0, 0); + if (b == NULL) + goto fail_a; + + if (keyrot_tx_next(a, sela, &ka, na) != 0) + goto fail_b; + + if (keyrot_tx_next(b, selb, &kb, nb) != 0) + goto fail_b; + + if (memcmp(sela, selb, KR_SELECTOR_LEN) != 0) + goto fail_b; + + if (memcmp(ka, kb, SYMMKEYSZ) != 0) + goto fail_b; + + if (memcmp(na, nb, KR_NONCE_LEN) != 0) + goto fail_b; + + keyrot_destroy(b); + keyrot_destroy(a); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_b: + keyrot_destroy(b); + fail_a: + keyrot_destroy(a); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_selector_layout(void) +{ + struct keyrot * a; + uint8_t sel[KR_SELECTOR_LEN]; + uint8_t nonce[KR_NONCE_LEN]; + const uint8_t * k; + + TEST_START(); + + a = keyrot_create(SEED_A, 3, 0); + if (a == NULL) + goto fail; + + /* First packet: epoch 3, node 0, seq 0 */ + if (keyrot_tx_next(a, sel, &k, nonce) != 0) + goto fail_a; + + if ((sel[0] >> 4) != 3) /* epoch */ + goto fail_a; + + if ((((sel[0] & 0x0F) << 8) | sel[1]) != 0) /* node */ + goto fail_a; + + if (sel[2] != 0 || sel[3] != 0 || sel[4] != 0 || sel[5] != 0) + goto fail_a; + + /* Second packet: seq advances to 1 */ + if (keyrot_tx_next(a, sel, &k, nonce) != 0) + goto fail_a; + + if (sel[5] != 1) + goto fail_a; + + keyrot_destroy(a); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_a: + keyrot_destroy(a); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_nodes_left_initial(void) +{ + struct keyrot * a; + + TEST_START(); + + a = keyrot_create(SEED_A, 0, 0); + if (a == NULL) + goto fail; + + if (keyrot_tx_nodes_left(a) != KEY_NODE_COUNT) + goto fail_a; + + keyrot_destroy(a); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_a: + keyrot_destroy(a); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_roundtrip(void) +{ + struct keyrot * a; /* role 0 */ + struct keyrot * b; /* role 1 */ + uint8_t sel[KR_SELECTOR_LEN]; + uint8_t ntx[KR_NONCE_LEN]; + uint8_t nrx[KR_NONCE_LEN]; + uint8_t ktx[SYMMKEYSZ]; + const uint8_t * ptx; + const uint8_t * prx; + struct kr_rx rx; + int i; + + TEST_START(); + + a = keyrot_create(SEED_A, 0, 0); + if (a == NULL) + goto fail; + + b = keyrot_create(SEED_A, 0, 1); + if (b == NULL) + goto fail_a; + + for (i = 0; i < 256; i++) { + if (keyrot_tx_next(a, sel, &ptx, ntx) != 0) + goto fail_b; + memcpy(ktx, ptx, SYMMKEYSZ); + if (keyrot_rx_lookup(b, sel, &prx, nrx, &rx) != 0) + goto fail_b; + if (keyrot_rx_commit(b, &rx) != 0) + goto fail_b; + if (memcmp(ktx, prx, SYMMKEYSZ) != 0) + goto fail_b; + if (memcmp(ntx, nrx, KR_NONCE_LEN) != 0) + goto fail_b; + } + + keyrot_destroy(b); + keyrot_destroy(a); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_b: + keyrot_destroy(b); + fail_a: + keyrot_destroy(a); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_direction_separation(void) +{ + struct keyrot * a; /* role 0 */ + struct keyrot * b; /* role 1 */ + uint8_t sela[KR_SELECTOR_LEN]; + uint8_t selb[KR_SELECTOR_LEN]; + uint8_t n[KR_NONCE_LEN]; + uint8_t ka[SYMMKEYSZ]; + const uint8_t * pa; + const uint8_t * pb; + + TEST_START(); + + a = keyrot_create(SEED_A, 0, 0); + if (a == NULL) + goto fail; + + b = keyrot_create(SEED_A, 0, 1); + if (b == NULL) + goto fail_a; + + if (keyrot_tx_next(a, sela, &pa, n) != 0) + goto fail_b; + + memcpy(ka, pa, SYMMKEYSZ); + if (keyrot_tx_next(b, selb, &pb, n) != 0) + goto fail_b; + + /* Same position, different role -> different leaf key */ + if (memcmp(ka, pb, SYMMKEYSZ) == 0) + goto fail_b; + + keyrot_destroy(b); + keyrot_destroy(a); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_b: + keyrot_destroy(b); + fail_a: + keyrot_destroy(a); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* Build a selector by hand (test knows the wire format). */ +static void mk_sel(uint8_t epoch, + uint16_t node, + uint32_t seq, + uint8_t sel[KR_SELECTOR_LEN]) +{ + sel[0] = (uint8_t) ((epoch << 4) | ((node >> 8) & 0x0F)); + sel[1] = (uint8_t) (node & 0xFF); + sel[2] = (uint8_t) (seq >> 24); + sel[3] = (uint8_t) (seq >> 16); + sel[4] = (uint8_t) (seq >> 8); + sel[5] = (uint8_t) (seq); +} + +static int test_random_access(void) +{ + struct keyrot * b; + uint8_t s0[KR_SELECTOR_LEN]; + uint8_t s5[KR_SELECTOR_LEN]; + uint8_t n[KR_NONCE_LEN]; + uint8_t k_first[SYMMKEYSZ]; + uint8_t k_node5[SYMMKEYSZ]; + const uint8_t * p; + struct kr_rx rx; + + TEST_START(); + + b = keyrot_create(SEED_A, 0, 1); + if (b == NULL) + goto fail; + + mk_sel(0, 0, 0, s0); + mk_sel(0, 5, 12345, s5); /* a far-ahead node, mid-span */ + + /* Jump straight to node 0 */ + if (keyrot_rx_lookup(b, s0, &p, n, &rx) != 0) + goto fail_b; + + memcpy(k_first, p, SYMMKEYSZ); + + /* Jump forward to node 5 (simulates a burst skip) */ + if (keyrot_rx_lookup(b, s5, &p, n, &rx) != 0) + goto fail_b; + + memcpy(k_node5, p, SYMMKEYSZ); + + /* Different nodes must yield different keys */ + if (memcmp(k_first, k_node5, SYMMKEYSZ) == 0) + goto fail_b; + + /* Jump back to node 0: still works, identical (no wedge) */ + if (keyrot_rx_lookup(b, s0, &p, n, &rx) != 0) + goto fail_b; + + if (memcmp(k_first, p, SYMMKEYSZ) != 0) + goto fail_b; + + /* Out-of-range node must be rejected */ + mk_sel(0, KEY_NODE_COUNT, 0, s0); + if (keyrot_rx_lookup(b, s0, &p, n, &rx) == 0) + goto fail_b; + + keyrot_destroy(b); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_b: + keyrot_destroy(b); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static const uint8_t SEED_B[SYMMKEYSZ] = { + 0xa1, 0xa2, 0xa3, 0xa4, 0xa5, 0xa6, 0xa7, 0xa8, + 0xa9, 0xaa, 0xab, 0xac, 0xad, 0xae, 0xaf, 0xb0, + 0xb1, 0xb2, 0xb3, 0xb4, 0xb5, 0xb6, 0xb7, 0xb8, + 0xb9, 0xba, 0xbb, 0xbc, 0xbd, 0xbe, 0xbf, 0xc0 +}; + +/* + * Look up and commit one within-node counter on epoch 0. Returns 0 on + * accept, 1 on a rejected commit (replay or too old), and -1 if the + * lookup itself failed - kept distinct so a reject assertion can never + * pass on an unrelated lookup miss. + */ +static int commit_ctr(struct keyrot * kr, + uint32_t ctr) +{ + uint8_t sel[KR_SELECTOR_LEN]; + uint8_t n[KR_NONCE_LEN]; + const uint8_t * k; + struct kr_rx rx; + + mk_sel(0, 0, ctr, sel); + + if (keyrot_rx_lookup(kr, sel, &k, n, &rx) != 0) + return -1; + + return keyrot_rx_commit(kr, &rx) == 0 ? 0 : 1; +} + +static int test_replay_window(void) +{ + struct keyrot * b; + struct keyrot * c; + uint32_t base; + uint32_t jump; + + TEST_START(); + + b = keyrot_create(SEED_A, 0, 1); + if (b == NULL) + goto fail; + + /* Fresh counters accepted; an immediate replay is rejected. */ + if (commit_ctr(b, 100) != 0) + goto fail_b; + + if (commit_ctr(b, 100) != 1) + goto fail_b; + + /* In-window reorder: accepted once, rejected on replay. */ + if (commit_ctr(b, 105) != 0) + goto fail_b; + + if (commit_ctr(b, 102) != 0) + goto fail_b; + + if (commit_ctr(b, 102) != 1) + goto fail_b; + + /* Too-old boundary: the window edge is rejected, just inside is not. */ + base = 4 * KEY_REPLAY_WINDOW; + if (commit_ctr(b, base) != 0) + goto fail_b; + + if (commit_ctr(b, base - (KEY_REPLAY_WINDOW - 64)) != 1) + goto fail_b; + + if (commit_ctr(b, base - (KEY_REPLAY_WINDOW - 64) + 1) != 0) + goto fail_b; + + /* + * RFC 6479 slack-word regression: two low counters, then a + * forward jump of a full bitmap that aliases their slot, then a + * replay of a low counter. Without the reserved slack word this + * replay is wrongly accepted. + */ + c = keyrot_create(SEED_A, 0, 1); + if (c == NULL) + goto fail_b; + + if (commit_ctr(c, 70) != 0) + goto fail_c; + + if (commit_ctr(c, 74) != 0) + goto fail_c; + + jump = KEY_REPLAY_WINDOW + 63; + if (commit_ctr(c, jump) != 0) + goto fail_c; + + if (commit_ctr(c, 74) != 1) + goto fail_c; + + keyrot_destroy(c); + keyrot_destroy(b); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_c: + keyrot_destroy(c); + fail_b: + keyrot_destroy(b); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_lookup_no_commit(void) +{ + struct keyrot * b; + uint8_t sel[KR_SELECTOR_LEN]; + uint8_t n[KR_NONCE_LEN]; + const uint8_t * k; + struct kr_rx rx; + int i; + + TEST_START(); + + b = keyrot_create(SEED_A, 0, 1); + if (b == NULL) + goto fail; + + mk_sel(0, 0, 100, sel); + + /* Repeated lookups are pre-AEAD and must not consume the slot. */ + for (i = 0; i < 4; i++) { + if (keyrot_rx_lookup(b, sel, &k, n, &rx) != 0) + goto fail_b; + } + + /* The slot is still fresh, so the first commit accepts ... */ + if (keyrot_rx_commit(b, &rx) != 0) + goto fail_b; + + /* ... and only the commit advanced it, so the next is a replay. */ + if (keyrot_rx_commit(b, &rx) == 0) + goto fail_b; + + keyrot_destroy(b); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_b: + keyrot_destroy(b); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_commit_prev_batch(void) +{ + struct keyrot * b; + uint8_t sel[KR_SELECTOR_LEN]; + uint8_t n[KR_NONCE_LEN]; + const uint8_t * k; + struct kr_rx rx; + + TEST_START(); + + b = keyrot_create(SEED_A, 0, 1); + if (b == NULL) + goto fail; + + /* Capture a packet under cur (epoch 0). */ + mk_sel(0, 0, 7, sel); + if (keyrot_rx_lookup(b, sel, &k, n, &rx) != 0) + goto fail_b; + + /* Re-key: the captured batch becomes prev and the flag clears. */ + if (keyrot_rekey(b, SEED_B, 1) != 0) + goto fail_b; + + /* The straggler commits under prev without claiming a switch. */ + if (keyrot_rx_commit(b, &rx) != 0) + goto fail_b; + + if (keyrot_peer_switched(b)) + goto fail_b; + + /* prev still holds a replay window: its replay is rejected. */ + if (keyrot_rx_commit(b, &rx) == 0) + goto fail_b; + + keyrot_destroy(b); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_b: + keyrot_destroy(b); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_replay_forward_clear(void) +{ + struct keyrot * d; + uint32_t low; + uint32_t alias; + uint32_t jump; + + TEST_START(); + + d = keyrot_create(SEED_A, 0, 1); + if (d == NULL) + goto fail; + + /* alias shares low's slot a window away; the jump must clear it. */ + low = 10; + alias = low + KEY_REPLAY_WINDOW; + jump = alias + KEY_REPLAY_WINDOW / 2; + + if (commit_ctr(d, low) != 0) + goto fail_d; + + if (commit_ctr(d, jump) != 0) + goto fail_d; + + if (commit_ctr(d, alias) != 0) + goto fail_d; + + if (commit_ctr(d, alias) != 1) + goto fail_d; + + keyrot_destroy(d); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_d: + keyrot_destroy(d); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_rekey_overlap(void) +{ + struct keyrot * a; /* role 0 */ + struct keyrot * b; /* role 1 */ + uint8_t old_sel[KR_SELECTOR_LEN]; + uint8_t sel[KR_SELECTOR_LEN]; + uint8_t ntx[KR_NONCE_LEN]; + uint8_t nrx[KR_NONCE_LEN]; + uint8_t ktx[SYMMKEYSZ]; + const uint8_t * ptx; + const uint8_t * prx; + struct kr_rx rx; + + TEST_START(); + + a = keyrot_create(SEED_A, 0, 0); + if (a == NULL) + goto fail; + + b = keyrot_create(SEED_A, 0, 1); + if (b == NULL) + goto fail_a; + + /* Send one gen-0 packet; keep its selector for the overlap. */ + if (keyrot_tx_next(a, old_sel, &ptx, ntx) != 0) + goto fail_b; + + memcpy(ktx, ptx, SYMMKEYSZ); + if (keyrot_rx_lookup(b, old_sel, &prx, nrx, &rx) != 0) + goto fail_b; + + if (memcmp(ktx, prx, SYMMKEYSZ) != 0) + goto fail_b; + + /* Both ends re-key to epoch 1 with a fresh seed. */ + if (keyrot_rekey(a, SEED_B, 1) != 0) + goto fail_b; + + if (keyrot_rekey(b, SEED_B, 1) != 0) + goto fail_b; + + /* TX is gated until promotion; promote a to emit the new epoch. */ + keyrot_tx_promote(a); + + /* New gen-1 traffic works. */ + if (keyrot_tx_next(a, sel, &ptx, ntx) != 0) + goto fail_b; + + memcpy(ktx, ptx, SYMMKEYSZ); + if (keyrot_rx_lookup(b, sel, &prx, nrx, &rx) != 0) + goto fail_b; + + if (memcmp(ktx, prx, SYMMKEYSZ) != 0) + goto fail_b; + + /* A straggling gen-0 packet still decrypts (overlap window). */ + if (keyrot_rx_lookup(b, old_sel, &prx, nrx, &rx) != 0) + goto fail_b; + + /* An unknown epoch is rejected. */ + mk_sel(7, 0, 0, sel); + if (keyrot_rx_lookup(b, sel, &prx, nrx, &rx) == 0) + goto fail_b; + + keyrot_destroy(b); + keyrot_destroy(a); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_b: + keyrot_destroy(b); + fail_a: + keyrot_destroy(a); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_tx_gate(void) +{ + struct keyrot * a; /* role 0 */ + struct keyrot * b; /* role 1 */ + uint8_t sel[KR_SELECTOR_LEN]; + uint8_t n[KR_NONCE_LEN]; + const uint8_t * p; + struct kr_rx rx; + + TEST_START(); + + a = keyrot_create(SEED_A, 0, 0); + if (a == NULL) + goto fail; + + b = keyrot_create(SEED_A, 0, 1); + if (b == NULL) + goto fail_a; + + /* Both re-key to epoch 1; TX must stay on epoch 0 until promoted. */ + if (keyrot_rekey(a, SEED_B, 1) != 0) + goto fail_b; + + if (keyrot_rekey(b, SEED_B, 1) != 0) + goto fail_b; + + /* a's TX still stamps the old epoch (0). */ + if (keyrot_tx_next(a, sel, &p, n) != 0) + goto fail_b; + + if ((sel[0] >> 4) != 0) + goto fail_b; + + /* b decrypts the old-epoch packet via its prev batch. */ + if (keyrot_rx_lookup(b, sel, &p, n, &rx) != 0) + goto fail_b; + + if (keyrot_rx_commit(b, &rx) != 0) + goto fail_b; + + /* b has not yet seen the new epoch from a. */ + if (keyrot_peer_switched(b)) + goto fail_b; + + /* a promotes; its TX now stamps the new epoch (1). */ + keyrot_tx_promote(a); + if (keyrot_tx_next(a, sel, &p, n) != 0) + goto fail_b; + + if ((sel[0] >> 4) != 1) + goto fail_b; + + /* b sees the new epoch and reports the peer switched. */ + if (keyrot_rx_lookup(b, sel, &p, n, &rx) != 0) + goto fail_b; + + if (keyrot_rx_commit(b, &rx) != 0) + goto fail_b; + + if (!keyrot_peer_switched(b)) + goto fail_b; + + keyrot_destroy(b); + keyrot_destroy(a); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_b: + keyrot_destroy(b); + fail_a: + keyrot_destroy(a); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_peer_switched_commit_only(void) +{ + struct keyrot * b; + uint8_t sel[KR_SELECTOR_LEN]; + uint8_t n[KR_NONCE_LEN]; + const uint8_t * k; + struct kr_rx rx; + + TEST_START(); + + b = keyrot_create(SEED_A, 0, 1); + if (b == NULL) + goto fail; + + /* A re-key clears the flag until a packet is seen on cur. */ + if (keyrot_rekey(b, SEED_B, 1) != 0) + goto fail_b; + + if (keyrot_peer_switched(b)) + goto fail_b; + + mk_sel(1, 0, 0, sel); + + /* Lookup is pre-AEAD: selecting a key must not flip the flag. */ + if (keyrot_rx_lookup(b, sel, &k, n, &rx) != 0) + goto fail_b; + + if (keyrot_peer_switched(b)) + goto fail_b; + + /* Commit runs post-AEAD and is what records the peer switched. */ + if (keyrot_rx_commit(b, &rx) != 0) + goto fail_b; + + if (!keyrot_peer_switched(b)) + goto fail_b; + + keyrot_destroy(b); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_b: + keyrot_destroy(b); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_commit_evicted(void) +{ + struct keyrot * b; + uint8_t sel[KR_SELECTOR_LEN]; + uint8_t n[KR_NONCE_LEN]; + const uint8_t * k; + struct kr_rx rx; + + TEST_START(); + + b = keyrot_create(SEED_A, 0, 1); + if (b == NULL) + goto fail; + + mk_sel(0, 0, 3, sel); + if (keyrot_rx_lookup(b, sel, &k, n, &rx) != 0) + goto fail_b; + + /* Two re-keys drop the captured batch from both cur and prev. */ + if (keyrot_rekey(b, SEED_B, 1) != 0) + goto fail_b; + + if (keyrot_rekey(b, SEED_A, 2) != 0) + goto fail_b; + + /* Commit on an evicted batch is a silent no-op, not a fault. */ + if (keyrot_rx_commit(b, &rx) != 0) + goto fail_b; + + keyrot_destroy(b); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_b: + keyrot_destroy(b); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* + * Concurrency: many TX threads + RX + re-key share one keyrot. The + * (epoch, counter) the TX side stamps must be globally unique (no AEAD + * nonce reuse). Capped below 16 re-keys so epoch maps 1:1 to a batch and + * the wire epoch never wraps (a wrapped epoch under a fresh key is not + * reuse but would false-trip the uniqueness check). Run under TSan to + * catch data races the static reviews can't. + */ +#define CT_THREADS 4 +#define CT_PKTS 2000 +#define CT_REKEYS 8 + +struct ct_rec { + uint8_t epoch; + uint64_t ctr; +}; + +struct ct_arg { + struct keyrot * kr; + struct ct_rec * recs; + size_t n; +}; + +static void * ct_tx_thread(void * a) +{ + struct ct_arg * arg = a; + uint8_t sel[KR_SELECTOR_LEN]; + uint8_t nonce[KR_NONCE_LEN]; + const uint8_t * k; + uint64_t ctr; + size_t i; + size_t j; + + for (i = 0; i < CT_PKTS; i++) { + if (keyrot_tx_next(arg->kr, sel, &k, nonce) != 0) + continue; + + ctr = 0; + for (j = 0; j < 8; j++) + ctr = (ctr << 8) | nonce[j]; + + arg->recs[arg->n].epoch = (uint8_t) (sel[0] >> 4); + arg->recs[arg->n].ctr = ctr; + arg->n++; + } + + return NULL; +} + +static void * ct_rx_thread(void * a) +{ + struct keyrot * kr = a; + uint8_t sel[KR_SELECTOR_LEN]; + uint8_t nonce[KR_NONCE_LEN]; + const uint8_t * k; + struct kr_rx rx; + size_t i; + + /* Exercise rx_lookup against re-key reclaim; results ignored. */ + for (i = 0; i < CT_PKTS; i++) { + mk_sel((uint8_t) (i % 16), 0, (uint32_t) i, sel); + if (keyrot_rx_lookup(kr, sel, &k, nonce, &rx) == 0) + (void) keyrot_rx_commit(kr, &rx); + } + + return NULL; +} + +static void * ct_rekey_thread(void * a) +{ + struct keyrot * kr = a; + struct timespec t; + int e; + + t.tv_sec = 0; + t.tv_nsec = 2 * 1000 * 1000; /* 2 ms */ + + for (e = 1; e <= CT_REKEYS; e++) { + nanosleep(&t, NULL); + if (keyrot_rekey(kr, (e & 1) ? SEED_B : SEED_A, + (uint8_t) e) != 0) + break; + keyrot_tx_promote(kr); + } + + return NULL; +} + +static int ct_cmp(const void * x, + const void * y) +{ + const struct ct_rec * a = x; + const struct ct_rec * b = y; + + if (a->epoch != b->epoch) + return a->epoch < b->epoch ? -1 : 1; + + if (a->ctr != b->ctr) + return a->ctr < b->ctr ? -1 : 1; + + return 0; +} + +static int test_concurrent_nonce_unique(void) +{ + struct keyrot * kr; + struct ct_arg arg[CT_THREADS]; + pthread_t tx[CT_THREADS]; + pthread_t rx; + pthread_t rk; + struct ct_rec * all; + size_t total; + size_t i; + bool reuse = false; + + TEST_START(); + + kr = keyrot_create(SEED_A, 0, 0); + if (kr == NULL) + goto fail; + + all = malloc(sizeof(*all) * CT_THREADS * CT_PKTS); + if (all == NULL) + goto fail_kr; + + for (i = 0; i < CT_THREADS; i++) { + arg[i].kr = kr; + arg[i].n = 0; + arg[i].recs = all + i * CT_PKTS; + } + + for (i = 0; i < CT_THREADS; i++) + pthread_create(&tx[i], NULL, ct_tx_thread, &arg[i]); + + pthread_create(&rx, NULL, ct_rx_thread, kr); + pthread_create(&rk, NULL, ct_rekey_thread, kr); + + for (i = 0; i < CT_THREADS; i++) + pthread_join(tx[i], NULL); + + pthread_join(rx, NULL); + pthread_join(rk, NULL); + + total = 0; + for (i = 0; i < CT_THREADS; i++) { + memmove(all + total, all + i * CT_PKTS, + arg[i].n * sizeof(*all)); + total += arg[i].n; + } + + qsort(all, total, sizeof(*all), ct_cmp); + + for (i = 1; i < total; i++) + if (ct_cmp(&all[i - 1], &all[i]) == 0) { + printf("(epoch %u, ctr %llu) reused\n", + all[i].epoch, + (unsigned long long) all[i].ctr); + reuse = true; + break; + } + + free(all); + + if (reuse) + goto fail_kr; + + keyrot_destroy(kr); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_kr: + keyrot_destroy(kr); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} +#endif /* HAVE_OPENSSL */ + +int keyrot_test(int argc, + char ** argv) +{ + int ret = 0; + + (void) argc; + (void) argv; + +#ifdef HAVE_OPENSSL + ret |= test_create_destroy(); + ret |= test_epoch_range(); + ret |= test_tx_deterministic(); + ret |= test_selector_layout(); + ret |= test_nodes_left_initial(); + ret |= test_roundtrip(); + ret |= test_direction_separation(); + ret |= test_random_access(); + ret |= test_peer_switched_commit_only(); + ret |= test_commit_evicted(); + ret |= test_replay_window(); + ret |= test_lookup_no_commit(); + ret |= test_commit_prev_batch(); + ret |= test_replay_forward_clear(); + ret |= test_rekey_overlap(); + ret |= test_tx_gate(); + ret |= test_concurrent_nonce_unique(); +#endif + return ret; +} diff --git a/src/lib/tests/md5_test.c b/src/lib/tests/md5_test.c index 28e8f42f..ea3e12b3 100644 --- a/src/lib/tests/md5_test.c +++ b/src/lib/tests/md5_test.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Test of the MD5 function * diff --git a/src/lib/tests/sha3_test.c b/src/lib/tests/sha3_test.c index 82b4ef0d..ccd4e12a 100644 --- a/src/lib/tests/sha3_test.c +++ b/src/lib/tests/sha3_test.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Test of the SHA3 function * diff --git a/src/lib/tests/shm_rbuff_test.c b/src/lib/tests/shm_rbuff_test.c deleted file mode 100644 index e36c3229..00000000 --- a/src/lib/tests/shm_rbuff_test.c +++ /dev/null @@ -1,113 +0,0 @@ -/* - * Ouroboros - Copyright (C) 2016 - 2024 - * - * Test of the shm_rbuff - * - * Dimitri Staessens <dimitri@ouroboros.rocks> - * Sander Vrijders <sander@ouroboros.rocks> - * - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License version 2 as - * published by the Free Software Foundation. - * - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License - * along with this program; if not, write to the Free Software - * Foundation, Inc., http://www.fsf.org/about/contact/. - */ - -#define _POSIX_C_SOURCE 200112L - -#include "config.h" - -#include <ouroboros/shm_rbuff.h> - -#include <errno.h> -#include <stdio.h> -#include <unistd.h> - -int shm_rbuff_test(int argc, - char ** argv) -{ - struct shm_rbuff * rb; - size_t i; - - (void) argc; - (void) argv; - - printf("Test: create rbuff..."); - - rb = shm_rbuff_create(getpid(), 1); - if (rb == NULL) - goto err; - - printf("success.\n\n"); - printf("Test: write a value..."); - - if (shm_rbuff_write(rb, 1) < 0) - goto error; - - printf("success.\n\n"); - printf("Test: check queue length is 1..."); - - if (shm_rbuff_queued(rb) != 1) - goto error; - - printf("success.\n\n"); - printf("Test: read a value..."); - - if (shm_rbuff_read(rb) != 1) - goto error; - - printf("success.\n\n"); - printf("Test: check queue is empty..."); - - if (shm_rbuff_read(rb) != -EAGAIN) - goto error; - - printf("success.\n\n"); - printf("Test: fill the queue..."); - - for (i = 0; i < SHM_RBUFF_SIZE - 1; ++i) { - if (shm_rbuff_queued(rb) != i) - goto error; - if (shm_rbuff_write(rb, 1) < 0) - goto error; - } - - printf("success.\n\n"); - printf("Test: check queue is full..."); - - if (shm_rbuff_queued(rb) != SHM_RBUFF_SIZE - 1) - goto error; - - printf("success [%zd entries].\n\n", shm_rbuff_queued(rb)); - - printf("Test: check queue is full by writing value..."); - if (!(shm_rbuff_write(rb, 1) < 0)) - goto error; - - printf("success [%zd entries].\n\n", shm_rbuff_queued(rb)); - - /* empty the rbuff */ - while (shm_rbuff_read(rb) >= 0) - ; - - shm_rbuff_destroy(rb); - - return 0; - - error: - /* empty the rbuff */ - while (shm_rbuff_read(rb) >= 0) - ; - - shm_rbuff_destroy(rb); - err: - printf("failed.\n\n"); - return -1; -} diff --git a/src/lib/tests/sockets_test.c b/src/lib/tests/sockets_test.c new file mode 100644 index 00000000..c00bfdc1 --- /dev/null +++ b/src/lib/tests/sockets_test.c @@ -0,0 +1,102 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Tests for socket.c + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#if defined(__linux__) || defined(__CYGWIN__) +#define _DEFAULT_SOURCE +#else +#define _POSIX_C_SOURCE 200112L +#endif + +#include <ouroboros/sockets.h> +#include <test/test.h> + +#include <assert.h> +#include <stdio.h> +#include <stdlib.h> + +#define TEST_PID 1234 +#define TEST_PID_STR "1234" +#define TEST_SERVER_PATH "/tmp/test.sock" +#define TEST_SERVER_PREFIX "/tmp/ouroboros/test." +#define TEST_SOCK_PATH_PREFIX "var/run/ouroboros/test." + +static int test_sock_path(void) +{ + char * path; + char * exp = TEST_SOCK_PATH_PREFIX TEST_PID_STR SOCK_PATH_SUFFIX; + + TEST_START(); + + path = sock_path(TEST_PID, TEST_SOCK_PATH_PREFIX); + if (path == NULL) { + printf("Path is NULL.\n"); + goto fail_path; + } + + if (strcmp(path, exp) != 0) { + printf("Expected path '%s', got '%s'.\n", exp, path); + goto fail_cmp; + } + + free(path); + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; + fail_cmp: + free(path); + fail_path: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_server_socket_open(void) +{ + int sockfd; + + TEST_START(); + + sockfd = server_socket_open(TEST_SERVER_PATH); + if (sockfd < 0) { + printf("Failed to open server socket.\n"); + goto fail_sock; + } + + close(sockfd); + + unlink(TEST_SERVER_PATH); + + TEST_SUCCESS(); + return TEST_RC_SUCCESS; + fail_sock: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +int sockets_test(void) +{ + int ret = 0; + + ret |= test_sock_path(); + ret |= test_server_socket_open(); + + return ret; +} diff --git a/src/lib/tests/time_test.c b/src/lib/tests/time_test.c index 65f896bb..919cf075 100644 --- a/src/lib/tests/time_test.c +++ b/src/lib/tests/time_test.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Test of the time utilities * @@ -22,143 +22,508 @@ #define _POSIX_C_SOURCE 200809L +#include <test/test.h> #include <ouroboros/time.h> #include <stdio.h> -static void ts_print(struct timespec * s) +static int ts_check(struct timespec * s, + time_t sec, + time_t nsec) { - printf("timespec is %zd:%ld.\n", (ssize_t) s->tv_sec, s->tv_nsec); + return s->tv_sec == sec && s->tv_nsec == nsec; } -static void tv_print(struct timeval * v) +static int tv_check(struct timeval * v, + time_t sec, + time_t usec) { - printf("timeval is %zd:%zu.\n", (ssize_t) v->tv_sec, (size_t) v->tv_usec); + return v->tv_sec == sec && v->tv_usec == usec; } -static void ts_init(struct timespec * s, - time_t sec, - time_t nsec) + +static int test_time_ts_init(void) { - s->tv_sec = sec; - s->tv_nsec = nsec; + struct timespec s = TIMESPEC_INIT_S (100); + struct timespec ms = TIMESPEC_INIT_MS(100); + struct timespec us = TIMESPEC_INIT_US(100); + struct timespec ns = TIMESPEC_INIT_NS(100); + + TEST_START(); + + if (!ts_check(&s, 100, 0)) { + printf("timespec_init_s failed.\n"); + goto fail; + } + + if (!ts_check(&ms, 0, 100 * MILLION)) { + printf("timespec_init_ms failed.\n"); + goto fail; + } + + if (!ts_check(&us, 0, 100* 1000L)) { + printf("timespec_init_us failed.\n"); + goto fail; + } + + if (!ts_check(&ns, 0, 100)) { + printf("timespec_init_ns failed.\n"); + goto fail; + } + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; } -static void tv_init(struct timeval * v, - time_t sec, - time_t usec) +static int test_time_tv_init(void) { - v->tv_sec = sec; - v->tv_usec = usec; + struct timeval s = TIMEVAL_INIT_S (100); + struct timeval ms = TIMEVAL_INIT_MS(100); + struct timeval us = TIMEVAL_INIT_US(100); + + TEST_START(); + + if (!tv_check(&s, 100, 0)) { + printf("timeval_init_s failed.\n"); + goto fail; + } + + if (!tv_check(&ms, 0, 100 * 1000L)) { + printf("timeval_init_ms failed.\n"); + goto fail; + } + + if (!tv_check(&us, 0, 100)) { + printf("timeval_init_us failed.\n"); + goto fail; + } + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; } -static int ts_check(struct timespec * s, - time_t sec, - time_t nsec) +static int test_ts_diff(void) { - return s->tv_sec == sec && s->tv_nsec == nsec; + struct timespec s0 = TIMESPEC_INIT_S (100); + struct timespec s1 = TIMESPEC_INIT_S (200); + struct timespec ms0 = TIMESPEC_INIT_MS(100); + struct timespec ms1 = TIMESPEC_INIT_MS(200); + struct timespec us0 = TIMESPEC_INIT_US(100); + struct timespec us1 = TIMESPEC_INIT_US(200); + struct timespec ns0 = TIMESPEC_INIT_NS(100); + struct timespec ns1 = TIMESPEC_INIT_NS(200); + struct timespec res; + + TEST_START(); + + ts_diff(&s0, &s1, &res); + if (!ts_check(&res, -100, 0)) { + printf("timespec_diff failed at s0 - s1.\n"); + goto fail; + } + + ts_diff(&s1, &s0, &res); + if (!ts_check(&res, 100, 0)) { + printf("timespec_diff failed at s1 - s0.\n"); + goto fail; + } + + ts_diff(&ms0, &ms1, &res); + if (!ts_check(&res, -1, 900 * MILLION)) { + printf("timespec_diff failed at ms0 - ms1.\n"); + goto fail; + } + + ts_diff(&ms1, &ms0, &res); + if (!ts_check(&res, 0, 100 * MILLION)) { + printf("timespec_diff failed at ms1 - ms0.\n"); + goto fail; + } + + ts_diff(&us0, &us1, &res); + if (!ts_check(&res, -1, 999900 * 1000L)) { + printf("timespec_diff failed at us0 - us1.\n"); + goto fail; + } + + ts_diff(&us1, &us0, &res); + if (!ts_check(&res, 0, 100 * 1000L)) { + printf("timespec_diff failed at us1 - us0.\n"); + goto fail; + } + + ts_diff(&ns0, &ns1, &res); + if (!ts_check(&res, -1, 999999900)) { + printf("timespec_diff failed at ns0 - ns1.\n"); + goto fail; + } + + ts_diff(&ns1, &ns0, &res); + if (!ts_check(&res, 0, 100)) { + printf("timespec_diff failed at ns1 - ns0.\n"); + goto fail; + } + + ts_diff(&s0, &ms0, &res); + if (!ts_check(&res, 99, 900 * MILLION)) { + printf("timespec_diff failed at s0 - ms0.\n"); + goto fail; + } + + ts_diff(&s0, &us0, &res); + if (!ts_check(&res, 99, 999900 * 1000L)) { + printf("timespec_diff failed at s0 - us0.\n"); + goto fail; + } + + ts_diff(&s0, &ns0, &res); + if (!ts_check(&res, 99, 999999900)) { + printf("timespec_diff failed at s0 - ns0.\n"); + goto fail; + } + + ts_diff(&ms0, &us0, &res); + if (!ts_check(&res, 0, 99900 * 1000L)) { + printf("timespec_diff failed at ms0 - us0.\n"); + goto fail; + } + + ts_diff(&ms0, &ns0, &res); + if (!ts_check(&res, 0, 99999900)) { + printf("timespec_diff failed at ms0 - ns0.\n"); + goto fail; + } + + ts_diff(&us0, &ns0, &res); + if (!ts_check(&res, 0, 99900)) { + printf("timespec_diff failed at us0 - ns0.\n"); + goto fail; + } + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; } -static int tv_check(struct timeval * v, - time_t sec, - time_t usec) +static int test_tv_diff(void) { - return v->tv_sec == sec && v->tv_usec == usec; + struct timeval s0 = TIMEVAL_INIT_S (100); + struct timeval s1 = TIMEVAL_INIT_S (200); + struct timeval ms0 = TIMEVAL_INIT_MS(100); + struct timeval ms1 = TIMEVAL_INIT_MS(200); + struct timeval us0 = TIMEVAL_INIT_US(100); + struct timeval us1 = TIMEVAL_INIT_US(200); + struct timeval res; + + TEST_START(); + + tv_diff(&s0, &s1, &res); + if (!tv_check(&res, -100, 0)) { + printf("timeval_diff failed at s0 - s1.\n"); + goto fail; + } + + tv_diff(&s1, &s0, &res); + if (!tv_check(&res, 100, 0)) { + printf("timeval_diff failed at s1 - s0.\n"); + goto fail; + } + + tv_diff(&ms0, &ms1, &res); + if (!tv_check(&res, -1, 900 * 1000L)) { + printf("timeval_diff failed at ms0 - ms1.\n"); + goto fail; + } + + tv_diff(&ms1, &ms0, &res); + if (!tv_check(&res, 0, 100 * 1000L)) { + printf("timeval_diff failed at ms1 - ms0.\n"); + goto fail; + } + + tv_diff(&us0, &us1, &res); + if (!tv_check(&res, -1, 999900)) { + printf("timeval_diff failed at us0 - us1.\n"); + goto fail; + } + + tv_diff(&us1, &us0, &res); + if (!tv_check(&res, 0, 100)) { + printf("timeval_diff failed at us1 - us0.\n"); + goto fail; + } + + tv_diff(&s0, &ms0, &res); + if (!tv_check(&res, 99, 900 * 1000L)) { + printf("timeval_diff failed at s0 - ms0.\n"); + goto fail; + } + + tv_diff(&s0, &us0, &res); + if (!tv_check(&res, 99, 999900)) { + printf("timeval_diff failed at s0 - us0.\n"); + goto fail; + } + + tv_diff(&ms0, &us0, &res); + if (!tv_check(&res, 0, 99900)) { + printf("timeval_diff failed at ms0 - us0.\n"); + goto fail; + } + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; } -int time_test(int argc, - char ** argv) +static int test_ts_diff_time(void) { - struct timespec s0; - struct timespec s1; - struct timespec s2; + struct timespec s0 = TIMESPEC_INIT_S (100); + struct timespec s1 = TIMESPEC_INIT_S (200); + struct timespec ms0 = TIMESPEC_INIT_MS(100); + struct timespec ms1 = TIMESPEC_INIT_MS(200); + struct timespec us0 = TIMESPEC_INIT_US(100); + struct timespec us1 = TIMESPEC_INIT_US(200); + struct timespec ns0 = TIMESPEC_INIT_NS(100); + struct timespec ns1 = TIMESPEC_INIT_NS(200); - struct timeval v0; - struct timeval v1; - struct timeval v2; + TEST_START(); - (void) argc; - (void) argv; + if (ts_diff_ms(&s0, &s1) != -100 * 1000L) { + printf("timespec_diff_ms failed at s0 - s1.\n"); + goto fail; + } + + if (ts_diff_ms(&s1, &s0) != 100 * 1000L) { + printf("timespec_diff_ms failed at s1 - s0.\n"); + goto fail; + } + + if (ts_diff_us(&s0, &s1) != -100 * MILLION) { + printf("timespec_diff_us failed at s1 - s0.\n"); + goto fail; + } + + if (ts_diff_us(&s1, &s0) != 100 * MILLION) { + printf("timespec_diff_us failed at s0 - s1.\n"); + goto fail; + } + + if (ts_diff_ns(&s0, &s1) != -100 * BILLION) { + printf("timespec_diff_ns failed at s0 - s1.\n"); + goto fail; + } + + if (ts_diff_ns(&s1, &s0) != 100 * BILLION) { + printf("timespec_diff_ns failed at s1 - s0.\n"); + goto fail; + } + + if (ts_diff_ms(&ms0, &ms1) != -100) { + printf("timespec_diff_ms failed at ms0 - ms1.\n"); + goto fail; + } + + if (ts_diff_ms(&ms1, &ms0) != 100) { + printf("timespec_diff_ms failed at ms1 - ms0.\n"); + goto fail; + } - ts_init(&s0, 0, 0); - ts_init(&s1, 5, 0); + if (ts_diff_us(&ms0, &ms1) != -100 * 1000L) { + printf("timespec_diff_us failed at ms0 - ms1.\n"); + goto fail; + } - ts_add(&s0, &s1, &s2); - if (!ts_check(&s2, 5, 0)) { - printf("ts_add failed.\n"); - ts_print(&s2); - return -1; + if (ts_diff_us(&ms1, &ms0) != 100 * 1000L) { + printf("timespec_diff_us failed at ms1 - ms0.\n"); + goto fail; } - tv_init(&v0, 0, 0); - tv_init(&v1, 5, 0); + if (ts_diff_ns(&ms0, &ms1) != -100 * MILLION) { + printf("timespec_diff_ns failed at ms0 - ms1.\n"); + goto fail; + } - tv_add(&v0, &v1, &v2); - if (!tv_check(&v2, 5, 0)) { - printf("tv_add failed.\n"); - tv_print(&v2); - return -1; + if (ts_diff_ns(&ms1, &ms0) != 100 * MILLION) { + printf("timespec_diff_ns failed at ms1 - ms0.\n"); + goto fail; } - ts_init(&s0, 0, 500 * MILLION); - ts_init(&s1, 0, 600 * MILLION); + if (ts_diff_ms(&us0, &us1) != 0) { + printf("timespec_diff_ms failed at us0 - us1.\n"); + goto fail; + } - ts_add(&s0, &s1, &s2); - if (!ts_check(&s2, 1, 100 * MILLION)) { - printf("ts_add with nano overflow failed.\n"); - ts_print(&s2); - return -1; + if (ts_diff_ms(&us1, &us0) != 0) { + printf("timespec_diff_ms failed at us1 - us0.\n"); + goto fail; } - tv_init(&v0, 0, 500 * 1000); - tv_init(&v1, 0, 600 * 1000); + if (ts_diff_us(&us0, &us1) != -100) { + printf("timespec_diff_us failed at us0 - us1.\n"); + goto fail; + } - tv_add(&v0, &v1, &v2); - if (!tv_check(&v2, 1, 100 * 1000)) { - printf("tv_add with nano overflow failed.\n"); - tv_print(&v2); - return -1; + if (ts_diff_us(&us1, &us0) != 100) { + printf("timespec_diff_us failed at us1 - us0.\n"); + goto fail; } - ts_init(&s0, 0, 0); - ts_init(&s1, 5, 0); + if (ts_diff_ns(&us0, &us1) != -100 * 1000L) { + printf("timespec_diff_ns failed at us0 - us1.\n"); + goto fail; + } - ts_diff(&s0, &s1, &s2); - if (!ts_check(&s2, -5, 0)) { - printf("ts_diff failed.\n"); - ts_print(&s2); - return -1; + if (ts_diff_ns(&us1, &us0) != 100 * 1000L) { + printf("timespec_diff_ns failed at us1 - us0.\n"); + goto fail; } - tv_init(&v0, 0, 0); - tv_init(&v1, 5, 0); + if (ts_diff_ms(&ns0, &ns1) != 0) { + printf("timespec_diff_ms failed at ns0 - ns1.\n"); + goto fail; + } - tv_diff(&v0, &v1, &v2); - if (!tv_check(&v2, -5, 0)) { - printf("tv_diff failed.\n"); - tv_print(&v2); - return -1; + if (ts_diff_ms(&ns1, &ns0) != 0) { + printf("timespec_diff_ms failed at ns1 - ns0.\n"); + goto fail; } - ts_init(&s0, 0, 500 * MILLION); - ts_init(&s1, 0, 600 * MILLION); + if (ts_diff_us(&ns0, &ns1) != 0) { + printf("timespec_diff_us failed at ns0 - ns1.\n"); + goto fail; + } - ts_diff(&s0, &s1, &s2); - if (!ts_check(&s2, -1, 900 * MILLION)) { - printf("ts_diff with nano underflow failed.\n"); - ts_print(&s2); - return -1; + if (ts_diff_us(&ns1, &ns0) != 0) { + printf("timespec_diff_us failed at ns1 - ns0.\n"); + goto fail; } - tv_init(&v0, 0, 500 * 1000); - tv_init(&v1, 0, 600 * 1000); + if (ts_diff_ns(&ns0, &ns1) != -100) { + printf("timespec_diff_ns failed at ns0 - ns1.\n"); + goto fail; + } - tv_diff(&v0, &v1, &v2); - if (!tv_check(&v2, -1, 900 * 1000)) { - printf("tv_diff with nano underflow failed.\n"); - tv_print(&v2); - return -1; + if (ts_diff_ns(&ns1, &ns0) != 100) { + printf("timespec_diff_ns failed at ns1 - ns0.\n"); + goto fail; } - return 0; + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_tv_diff_time(void) +{ + struct timeval s0 = TIMEVAL_INIT_S (100); + struct timeval s1 = TIMEVAL_INIT_S (200); + struct timeval ms0 = TIMEVAL_INIT_MS(100); + struct timeval ms1 = TIMEVAL_INIT_MS(200); + struct timeval us0 = TIMEVAL_INIT_US(100); + struct timeval us1 = TIMEVAL_INIT_US(200); + + TEST_START(); + + if (tv_diff_ms(&s0, &s1) != -100 * 1000L) { + printf("timeval_diff_ms failed at s0 - s1.\n"); + goto fail; + } + + if (tv_diff_ms(&s1, &s0) != 100 * 1000L) { + printf("timeval_diff_ms failed at s1 - s0.\n"); + goto fail; + } + + if (tv_diff_us(&s0, &s1) != -100 * MILLION) { + printf("timeval_diff_us failed at s0 - s1.\n"); + goto fail; + } + + if (tv_diff_us(&s1, &s0) != 100 * MILLION) { + printf("timeval_diff_us failed at s1 - s0.\n"); + goto fail; + } + + if (tv_diff_ms(&ms0, &ms1) != -100) { + printf("timeval_diff_ms failed at ms0 - ms1.\n"); + goto fail; + } + + if (tv_diff_ms(&ms1, &ms0) != 100) { + printf("timeval_diff_ms failed at ms1 - ms0.\n"); + goto fail; + } + + if (tv_diff_us(&ms0, &ms1) != -100 * 1000L) { + printf("timeval_diff_us failed at ms0 - ms1.\n"); + goto fail; + } + + if (tv_diff_us(&ms1, &ms0) != 100 * 1000L) { + printf("timeval_diff_us failed at ms1 - ms0.\n"); + goto fail; + } + + if (tv_diff_ms(&us0, &us1) != 0) { + printf("timeval_diff_ms failed at us0 - us1.\n"); + goto fail; + } + + if (tv_diff_ms(&us1, &us0) != 0) { + printf("timeval_diff_ms failed at us1 - us0.\n"); + goto fail; + } + + if (tv_diff_us(&us0, &us1) != -100) { + printf("timeval_diff_us failed at us0 - us1.\n"); + goto fail; + } + + if (tv_diff_us(&us1, &us0) != 100) { + printf("timeval_diff_us failed at us1 - us0.\n"); + goto fail; + } + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +int time_test(int argc, + char ** argv) +{ + int ret = 0; + + (void) argc; + (void) argv; + + ret |= test_time_ts_init(); + ret |= test_time_tv_init(); + ret |= test_ts_diff(); + ret |= test_tv_diff(); + ret |= test_ts_diff_time(); + ret |= test_tv_diff_time(); + + return ret; } diff --git a/src/lib/tests/tpm_test.c b/src/lib/tests/tpm_test.c new file mode 100644 index 00000000..7cc049cd --- /dev/null +++ b/src/lib/tests/tpm_test.c @@ -0,0 +1,104 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Tests for the threadpool manager + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + + +#include <ouroboros/tpm.h> + +#include <test/test.h> + +static void * test_func(void * o) +{ + (void) o; + + while(1) + sleep(1); + + return NULL; +} + +static int test_tpm_create_destroy(void) +{ + struct tpm *tpm; + + TEST_START(); + + tpm = tpm_create(2, 2, &test_func, NULL); + if (tpm == NULL) { + printf("Failed to initialize TPM.\n"); + goto fail; + } + + tpm_destroy(tpm); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_tpm_start_stop(void * (* fn)(void *), + void * o) +{ + struct tpm *tpm; + + TEST_START(); + + tpm = tpm_create(2, 2, fn, o); + if (tpm == NULL) { + printf("Failed to initialize TPM.\n"); + goto fail_create; + } + + if (tpm_start(tpm) < 0) { + printf("Failed to start TPM.\n"); + goto fail_start; + } + + tpm_stop(tpm); + + tpm_destroy(tpm); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_start: + tpm_destroy(tpm); + fail_create: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +int tpm_test(int argc, + char ** argv) +{ + int ret = 0; + + (void) argc; + (void) argv; + + ret |= test_tpm_create_destroy(); + ret |= test_tpm_start_stop(&test_func, NULL); + + return ret; +} diff --git a/src/lib/tests/tw_test.c b/src/lib/tests/tw_test.c new file mode 100644 index 00000000..32c302c4 --- /dev/null +++ b/src/lib/tests/tw_test.c @@ -0,0 +1,663 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Generic timing-wheel tests + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#if defined(__linux__) || defined(__CYGWIN__) +#define _DEFAULT_SOURCE +#else +#define _POSIX_C_SOURCE 200809L +#endif + +#include "config.h" + +#include <test/test.h> + +#include <ouroboros/time.h> +#include <ouroboros/tw.h> + +#include <stdint.h> +#include <stdio.h> +#include <time.h> + +struct payload { + struct tw_entry tw; + int fired; +}; + +struct cancel_payload { + struct tw_entry tw; + int fired; + struct tw_entry * sibling; +}; + +struct repost_payload { + struct tw_entry tw; + int fired; + struct payload * sibling; + uint64_t repost_at; +}; + +static void cb_count(void * arg) +{ + struct payload * p = arg; + p->fired++; +} + +static void cb_cancel_sibling(void * arg) +{ + struct cancel_payload * p = arg; + p->fired++; + tw_cancel(p->sibling); +} + +static void cb_repost_sibling(void * arg) +{ + struct repost_payload * p = arg; + p->fired++; + tw_post(&p->sibling->tw, p->repost_at, cb_count, p->sibling); +} + +static uint64_t now_ns(void) +{ + struct timespec ts; + clock_gettime(PTHREAD_COND_CLOCK, &ts); + return TS_TO_UINT64(ts); +} + +static void sleep_ns(uint64_t ns) +{ + struct timespec ts; + UINT64_TO_TS(ns, &ts); + nanosleep(&ts, NULL); +} + +static int test_tw_init_fini(void) +{ + TEST_START(); + + if (tw_init() < 0) { + printf("tw_init failed.\n"); + goto fail; + } + + tw_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_tw_post_fires_after_deadline(void) +{ + struct payload p; + + TEST_START(); + + if (tw_init() < 0) + goto fail; + + tw_init_entry(&p.tw); + p.fired = 0; + + tw_post(&p.tw, now_ns() + 5 * MILLION, cb_count, &p); + + sleep_ns(20 * MILLION); + tw_move(); + + if (p.fired != 1) { + printf("expected 1 fire, got %d\n", p.fired); + goto fail_post; + } + + tw_cancel(&p.tw); + tw_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_post: + tw_cancel(&p.tw); + tw_fini(); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_tw_no_fire_before_deadline(void) +{ + struct payload p; + + TEST_START(); + + if (tw_init() < 0) + goto fail; + + tw_init_entry(&p.tw); + p.fired = 0; + + tw_post(&p.tw, now_ns() + 100 * MILLION, cb_count, &p); + + sleep_ns(2 * MILLION); + tw_move(); + + if (p.fired != 0) { + printf("expected 0 fires, got %d\n", p.fired); + goto fail_post; + } + + tw_cancel(&p.tw); + tw_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_post: + tw_cancel(&p.tw); + tw_fini(); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_tw_cancel_prevents_fire(void) +{ + struct payload p; + + TEST_START(); + + if (tw_init() < 0) + goto fail; + + tw_init_entry(&p.tw); + p.fired = 0; + + tw_post(&p.tw, now_ns() + 5 * MILLION, cb_count, &p); + tw_cancel(&p.tw); + + sleep_ns(20 * MILLION); + tw_move(); + + if (p.fired != 0) { + printf("cancelled entry fired %d times\n", p.fired); + goto fail_init; + } + + tw_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_init: + tw_fini(); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_tw_cancel_unposted_is_noop(void) +{ + struct tw_entry e; + + TEST_START(); + + if (tw_init() < 0) + goto fail; + + tw_init_entry(&e); + tw_cancel(&e); + tw_cancel(&e); + + tw_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_tw_fire_only_once(void) +{ + struct payload p; + + TEST_START(); + + if (tw_init() < 0) + goto fail; + + tw_init_entry(&p.tw); + p.fired = 0; + + tw_post(&p.tw, now_ns() + 3 * MILLION, cb_count, &p); + + sleep_ns(20 * MILLION); + tw_move(); + tw_move(); + tw_move(); + + if (p.fired != 1) { + printf("expected 1 fire, got %d after 3 moves\n", p.fired); + goto fail_post; + } + + tw_cancel(&p.tw); + tw_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_post: + tw_cancel(&p.tw); + tw_fini(); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* Multi-level: post a level-1 (>= 256ms) deadline; should still fire. */ +static int test_tw_post_level1_fires(void) +{ + struct payload p; + + TEST_START(); + + if (tw_init() < 0) + goto fail; + + tw_init_entry(&p.tw); + p.fired = 0; + + tw_post(&p.tw, now_ns() + 300 * MILLION, cb_count, &p); + + if (p.tw.lvl != 1) { + printf("expected level 1 placement, got %zu\n", p.tw.lvl); + goto fail_post; + } + + sleep_ns(320 * MILLION); + tw_move(); + + if (p.fired != 1) { + printf("level-1 entry didn't fire (got %d)\n", p.fired); + goto fail_post; + } + + tw_cancel(&p.tw); + tw_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_post: + tw_cancel(&p.tw); + tw_fini(); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_tw_many_entries_all_fire(void) +{ + struct payload pl[16]; + size_t i; + size_t total = 0; + + TEST_START(); + + if (tw_init() < 0) + goto fail; + + for (i = 0; i < 16; ++i) { + tw_init_entry(&pl[i].tw); + pl[i].fired = 0; + tw_post(&pl[i].tw, now_ns() + (1 + i) * MILLION, + cb_count, &pl[i]); + } + + sleep_ns(40 * MILLION); + tw_move(); + + for (i = 0; i < 16; ++i) + total += pl[i].fired; + + if (total != 16) { + printf("expected 16 fires, got %zu\n", total); + goto fail_post; + } + + for (i = 0; i < 16; ++i) + tw_cancel(&pl[i].tw); + tw_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_post: + for (i = 0; i < 16; ++i) + tw_cancel(&pl[i].tw); + tw_fini(); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* tw_next_expiry signals empty wheel via tv_nsec == -1. */ +static int test_tw_next_expiry_empty(void) +{ + struct timespec out; + + TEST_START(); + + if (tw_init() < 0) + goto fail; + + tw_next_expiry(&out); + if (out.tv_nsec != -1) { + printf("expected tv_nsec=-1, got %ld\n", (long) out.tv_nsec); + goto fail_init; + } + + tw_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_init: + tw_fini(); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* tw_next_expiry returns a deadline within the right ballpark. */ +static int test_tw_next_expiry_returns_deadline(void) +{ + struct payload p; + struct timespec out; + uint64_t target; + uint64_t out_ns; + int64_t skew; + + TEST_START(); + + if (tw_init() < 0) + goto fail; + + tw_init_entry(&p.tw); + p.fired = 0; + + target = now_ns() + 50 * MILLION; + tw_post(&p.tw, target, cb_count, &p); + + tw_next_expiry(&out); + out_ns = TS_TO_UINT64(out); + + /* Level-0 quantization gives ±1 slot of skew. */ + skew = (int64_t)(out_ns) - (int64_t)(target); + if (skew < -2 * MILLION || skew > 4 * MILLION) { + printf("deadline not in -2..+4 ms, skew=%ld ns\n", (long) skew); + goto fail_post; + } + + tw_cancel(&p.tw); + tw_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_post: + tw_cancel(&p.tw); + tw_fini(); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* Repost: fire, then post again. */ +static int test_tw_repost_after_fire(void) +{ + struct payload p; + + TEST_START(); + + if (tw_init() < 0) + goto fail; + + tw_init_entry(&p.tw); + p.fired = 0; + + tw_post(&p.tw, now_ns() + 3 * MILLION, cb_count, &p); + sleep_ns(20 * MILLION); + tw_move(); + if (p.fired != 1) { + printf("first fire missed\n"); + goto fail_post; + } + + tw_post(&p.tw, now_ns() + 3 * MILLION, cb_count, &p); + sleep_ns(20 * MILLION); + tw_move(); + if (p.fired != 2) { + printf("second fire missed (fired=%d)\n", p.fired); + goto fail_post; + } + + tw_cancel(&p.tw); + tw_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_post: + tw_cancel(&p.tw); + tw_fini(); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* Double-post replaces the schedule; only the second fires. */ +static int test_tw_double_post_replaces(void) +{ + struct payload p; + + TEST_START(); + + if (tw_init() < 0) + goto fail; + + tw_init_entry(&p.tw); + p.fired = 0; + + tw_post(&p.tw, now_ns() + 30 * MILLION, cb_count, &p); + tw_post(&p.tw, now_ns() + 3 * MILLION, cb_count, &p); + + sleep_ns(20 * MILLION); + tw_move(); + + if (p.fired != 1) { + printf("expected 1 fire after replace, got %d\n", p.fired); + goto fail_post; + } + + sleep_ns(40 * MILLION); + tw_move(); + + if (p.fired != 1) { + printf("first schedule fired after replace (got %d)\n", + p.fired); + goto fail_post; + } + + tw_cancel(&p.tw); + tw_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_post: + tw_cancel(&p.tw); + tw_fini(); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* Fire callback may safely cancel a sibling in the same slot. */ +static int test_tw_fire_cancels_sibling(void) +{ + struct cancel_payload a; + struct payload b; + uint64_t deadline; + + TEST_START(); + + if (tw_init() < 0) + goto fail; + + tw_init_entry(&a.tw); + tw_init_entry(&b.tw); + a.fired = 0; + a.sibling = &b.tw; + b.fired = 0; + + deadline = now_ns() + 3 * MILLION; + tw_post(&a.tw, deadline, cb_cancel_sibling, &a); + tw_post(&b.tw, deadline, cb_count, &b); + + sleep_ns(20 * MILLION); + tw_move(); + + if (a.fired != 1) { + printf("a expected 1 fire, got %d\n", a.fired); + goto fail_post; + } + if (b.fired != 0) { + printf("b should not have fired (got %d)\n", b.fired); + goto fail_post; + } + + tw_cancel(&a.tw); + tw_cancel(&b.tw); + tw_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_post: + tw_cancel(&a.tw); + tw_cancel(&b.tw); + tw_fini(); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* Fire callback may safely repost a sibling to a future slot. */ +static int test_tw_fire_posts_sibling(void) +{ + struct repost_payload a; + struct payload b; + uint64_t deadline; + + TEST_START(); + + if (tw_init() < 0) + goto fail; + + tw_init_entry(&a.tw); + tw_init_entry(&b.tw); + a.fired = 0; + a.sibling = &b; + a.repost_at = now_ns() + 30 * MILLION; + b.fired = 0; + + deadline = now_ns() + 3 * MILLION; + tw_post(&a.tw, deadline, cb_repost_sibling, &a); + tw_post(&b.tw, deadline, cb_count, &b); + + sleep_ns(20 * MILLION); + tw_move(); + + if (a.fired != 1) { + printf("a expected 1 fire, got %d\n", a.fired); + goto fail_post; + } + if (b.fired != 0) { + printf("b fired before reposted deadline (got %d)\n", + b.fired); + goto fail_post; + } + + sleep_ns(25 * MILLION); + tw_move(); + + if (b.fired != 1) { + printf("b expected 1 fire after repost, got %d\n", + b.fired); + goto fail_post; + } + + tw_cancel(&a.tw); + tw_cancel(&b.tw); + tw_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_post: + tw_cancel(&a.tw); + tw_cancel(&b.tw); + tw_fini(); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +int tw_test(int argc, + char ** argv) +{ + int ret = 0; + + (void) argc; + (void) argv; + + ret |= test_tw_init_fini(); + ret |= test_tw_post_fires_after_deadline(); + ret |= test_tw_no_fire_before_deadline(); + ret |= test_tw_cancel_prevents_fire(); + ret |= test_tw_cancel_unposted_is_noop(); + ret |= test_tw_fire_only_once(); + ret |= test_tw_post_level1_fires(); + ret |= test_tw_many_entries_all_fire(); + ret |= test_tw_next_expiry_empty(); + ret |= test_tw_next_expiry_returns_deadline(); + ret |= test_tw_repost_after_fire(); + ret |= test_tw_double_post_replaces(); + ret |= test_tw_fire_cancels_sibling(); + ret |= test_tw_fire_posts_sibling(); + + return ret; +} diff --git a/src/lib/timerwheel.c b/src/lib/timerwheel.c deleted file mode 100644 index 96f4ac47..00000000 --- a/src/lib/timerwheel.c +++ /dev/null @@ -1,414 +0,0 @@ -/* - * Ouroboros - Copyright (C) 2016 - 2024 - * - * Timerwheel - * - * Dimitri Staessens <dimitri@ouroboros.rocks> - * Sander Vrijders <sander@ouroboros.rocks> - * - * This library is free software; you can redistribute it and/or - * modify it under the terms of the GNU Lesser General Public License - * version 2.1 as published by the Free Software Foundation. - * - * This library is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * Lesser General Public License for more details. - * - * You should have received a copy of the GNU Lesser General Public - * License along with this library; if not, write to the Free Software - * Foundation, Inc., http://www.fsf.org/about/contact/. - */ - -#include <ouroboros/list.h> - -/* Overflow limits range to about 6 hours. */ -#define ts_to_ns(ts) (ts.tv_sec * BILLION + ts.tv_nsec) -#define ts_to_rxm_slot(ts) (ts_to_ns(ts) >> RXMQ_RES) -#define ts_to_ack_slot(ts) (ts_to_ns(ts) >> ACKQ_RES) - -struct rxm { - struct list_head next; - uint32_t seqno; -#ifndef RXM_BUFFER_ON_HEAP - struct shm_du_buff * sdb; -#endif - struct frct_pci * pkt; - size_t len; - time_t t0; /* Time when original was sent (us). */ - struct frcti * frcti; - int fd; - int flow_id; /* Prevent rtx when fd reused. */ -}; - -struct ack { - struct list_head next; - struct frcti * frcti; - int fd; - int flow_id; -}; - -struct { - /* - * At a 1 ms min resolution, every level bumps the - * resolution by a factor of 16. - */ - struct list_head rxms[RXMQ_LVLS][RXMQ_SLOTS]; - - struct list_head acks[ACKQ_SLOTS]; - bool map[ACKQ_SLOTS][PROG_MAX_FLOWS]; - - size_t prv_rxm[RXMQ_LVLS]; /* Last processed rxm slots. */ - size_t prv_ack; /* Last processed ack slot. */ - pthread_mutex_t lock; -} rw; - -static void timerwheel_fini(void) -{ - size_t i; - size_t j; - struct list_head * p; - struct list_head * h; - - pthread_mutex_lock(&rw.lock); - - for (i = 0; i < RXMQ_LVLS; ++i) { - for (j = 0; j < RXMQ_SLOTS; j++) { - list_for_each_safe(p, h, &rw.rxms[i][j]) { - struct rxm * rxm; - rxm = list_entry(p, struct rxm, next); - list_del(&rxm->next); -#ifdef RXM_BUFFER_ON_HEAP - free(rxm->pkt); -#else - shm_du_buff_ack(rxm->sdb); - ipcp_sdb_release(rxm->sdb); -#endif - free(rxm); - } - } - } - - for (i = 0; i < ACKQ_SLOTS; ++i) { - list_for_each_safe(p, h, &rw.acks[i]) { - struct ack * a = list_entry(p, struct ack, next); - list_del(&a->next); - free(a); - } - } - - pthread_mutex_unlock(&rw.lock); - - pthread_mutex_destroy(&rw.lock); -} - -static int timerwheel_init(void) -{ - struct timespec now; - size_t i; - size_t j; - - if (pthread_mutex_init(&rw.lock, NULL)) - return -1; - - clock_gettime(PTHREAD_COND_CLOCK, &now); - - for (i = 0; i < RXMQ_LVLS; ++i) { - rw.prv_rxm[i] = (ts_to_rxm_slot(now) - 1); - rw.prv_rxm[i] >>= (RXMQ_BUMP * i); - rw.prv_rxm[i] &= (RXMQ_SLOTS - 1); - for (j = 0; j < RXMQ_SLOTS; ++j) - list_head_init(&rw.rxms[i][j]); - } - - rw.prv_ack = (ts_to_ack_slot(now) - 1) & (ACKQ_SLOTS - 1); - for (i = 0; i < ACKQ_SLOTS; ++i) - list_head_init(&rw.acks[i]); - - return 0; -} - -static void timerwheel_move(void) -{ - struct timespec now; - struct list_head * p; - struct list_head * h; - size_t rxm_slot; - size_t ack_slot; - size_t i; - size_t j; - - pthread_mutex_lock(&rw.lock); - - pthread_cleanup_push(__cleanup_mutex_unlock, &rw.lock); - - clock_gettime(PTHREAD_COND_CLOCK, &now); - - rxm_slot = ts_to_rxm_slot(now); - - for (i = 0; i < RXMQ_LVLS; ++i) { - size_t j_max_slot = rxm_slot & (RXMQ_SLOTS - 1); - j = rw.prv_rxm[i]; - if (j_max_slot < j) - j_max_slot += RXMQ_SLOTS; - while (j++ < j_max_slot) { - list_for_each_safe(p, h, - &rw.rxms[i][j & (RXMQ_SLOTS - 1)]) { - struct rxm * r; - struct frct_cr * snd_cr; - struct frct_cr * rcv_cr; - size_t slot; - size_t rslot; - ssize_t idx; - struct shm_du_buff * sdb; - struct frct_pci * pci; - struct flow * f; - uint32_t snd_lwe; - uint32_t rcv_lwe; - size_t lvl = 0; - - r = list_entry(p, struct rxm, next); - - list_del(&r->next); - - snd_cr = &r->frcti->snd_cr; - rcv_cr = &r->frcti->rcv_cr; - f = &ai.flows[r->fd]; -#ifndef RXM_BUFFER_ON_HEAP - shm_du_buff_ack(r->sdb); -#endif - if (f->frcti == NULL - || f->info.id != r->flow_id) - goto cleanup; - - pthread_rwlock_rdlock(&r->frcti->lock); - - snd_lwe = snd_cr->lwe; - rcv_lwe = rcv_cr->lwe; - - pthread_rwlock_unlock(&r->frcti->lock); - - /* Has been ack'd, remove. */ - if (before(r->seqno, snd_lwe)) - goto cleanup; - - /* Check for r-timer expiry. */ - if (ts_to_ns(now) - r->t0 > r->frcti->r) - goto flow_down; - - pthread_rwlock_wrlock(&r->frcti->lock); - - if (r->seqno == r->frcti->rttseq) { - r->frcti->rto += - r->frcti->rto >> RTO_DIV; - r->frcti->probe = false; - } -#ifdef PROC_FLOW_STATS - r->frcti->n_rtx++; -#endif - rslot = r->frcti->rto >> RXMQ_RES; - - pthread_rwlock_unlock(&r->frcti->lock); - - /* Schedule at least in the next time slot. */ - slot = ts_to_ns(now) >> RXMQ_RES; - - while (rslot >= RXMQ_SLOTS) { - ++lvl; - rslot >>= RXMQ_BUMP; - slot >>= RXMQ_BUMP; - } - - if (lvl >= RXMQ_LVLS) /* Can't reschedule */ - goto flow_down; - - rslot = (rslot + slot + 1) & (RXMQ_SLOTS - 1); -#ifdef RXM_BLOCKING - if (ipcp_sdb_reserve(&sdb, r->len) < 0) -#else - if (shm_rdrbuff_alloc(ai.rdrb, r->len, NULL, - &sdb) < 0) -#endif - goto reschedule; /* rdrbuff full */ - - pci = (struct frct_pci *) shm_du_buff_head(sdb); - memcpy(pci, r->pkt, r->len); -#ifndef RXM_BUFFER_ON_HEAP - ipcp_sdb_release(r->sdb); - r->sdb = sdb; - r->pkt = pci; - shm_du_buff_wait_ack(sdb); -#endif - idx = shm_du_buff_get_idx(sdb); - - /* Retransmit the copy. */ - pci->ackno = hton32(rcv_lwe); -#ifdef RXM_BLOCKING - if (shm_rbuff_write_b(f->tx_rb, idx, NULL) < 0) -#else - if (shm_rbuff_write(f->tx_rb, idx) < 0) -#endif - goto flow_down; - shm_flow_set_notify(f->set, f->info.id, - FLOW_PKT); - reschedule: - list_add(&r->next, &rw.rxms[lvl][rslot]); - continue; - - flow_down: - shm_rbuff_set_acl(f->tx_rb, ACL_FLOWDOWN); - shm_rbuff_set_acl(f->rx_rb, ACL_FLOWDOWN); - cleanup: -#ifdef RXM_BUFFER_ON_HEAP - free(r->pkt); -#else - ipcp_sdb_release(r->sdb); -#endif - free(r); - } - } - rw.prv_rxm[i] = rxm_slot & (RXMQ_SLOTS - 1); - /* Move up a level in the wheel. */ - rxm_slot >>= RXMQ_BUMP; - } - - ack_slot = ts_to_ack_slot(now) & (ACKQ_SLOTS - 1) ; - - j = rw.prv_ack; - - if (ack_slot < j) - ack_slot += ACKQ_SLOTS; - - while (j++ < ack_slot) { - list_for_each_safe(p, h, &rw.acks[j & (ACKQ_SLOTS - 1)]) { - struct ack * a; - struct flow * f; - - a = list_entry(p, struct ack, next); - - list_del(&a->next); - - f = &ai.flows[a->fd]; - - rw.map[j & (ACKQ_SLOTS - 1)][a->fd] = false; - - if (f->info.id == a->flow_id && f->frcti != NULL) - send_frct_pkt(a->frcti); - - free(a); - } - } - - rw.prv_ack = ack_slot & (ACKQ_SLOTS - 1); - - pthread_cleanup_pop(true); -} - -static int timerwheel_rxm(struct frcti * frcti, - uint32_t seqno, - struct shm_du_buff * sdb) -{ - struct timespec now; - struct rxm * r; - size_t slot; - size_t lvl = 0; - time_t rto_slot; - - r = malloc(sizeof(*r)); - if (r == NULL) - return -ENOMEM; - - clock_gettime(PTHREAD_COND_CLOCK, &now); - - r->t0 = ts_to_ns(now); - r->seqno = seqno; - r->frcti = frcti; - r->len = shm_du_buff_len(sdb); -#ifdef RXM_BUFFER_ON_HEAP - r->pkt = malloc(r->len); - if (r->pkt == NULL) { - free(r); - return -ENOMEM; - } - memcpy(r->pkt, shm_du_buff_head(sdb), r->len); -#else - r->sdb = sdb; - r->pkt = (struct frct_pci *) shm_du_buff_head(sdb); -#endif - pthread_rwlock_rdlock(&r->frcti->lock); - - rto_slot = frcti->rto >> RXMQ_RES; - slot = r->t0 >> RXMQ_RES; - - r->fd = frcti->fd; - r->flow_id = ai.flows[r->fd].info.id; - - pthread_rwlock_unlock(&r->frcti->lock); - - while (rto_slot >= RXMQ_SLOTS) { - ++lvl; - rto_slot >>= RXMQ_BUMP; - slot >>= RXMQ_BUMP; - } - - if (lvl >= RXMQ_LVLS) { /* Out of timerwheel range. */ -#ifdef RXM_BUFFER_ON_HEAP - free(r->pkt); -#endif - free(r); - return -EPERM; - } - - slot = (slot + rto_slot + 1) & (RXMQ_SLOTS - 1); - - pthread_mutex_lock(&rw.lock); - - list_add_tail(&r->next, &rw.rxms[lvl][slot]); -#ifndef RXM_BUFFER_ON_HEAP - shm_du_buff_wait_ack(sdb); -#endif - pthread_mutex_unlock(&rw.lock); - - return 0; -} - -static int timerwheel_delayed_ack(int fd, - struct frcti * frcti) -{ - struct timespec now; - struct ack * a; - size_t slot; - - a = malloc(sizeof(*a)); - if (a == NULL) - return -ENOMEM; - - clock_gettime(PTHREAD_COND_CLOCK, &now); - - pthread_rwlock_rdlock(&frcti->lock); - - slot = (((ts_to_ns(now) + (TICTIME << 1)) >> ACKQ_RES) + 1) - & (ACKQ_SLOTS - 1); - - pthread_rwlock_unlock(&frcti->lock); - - a->fd = fd; - a->frcti = frcti; - a->flow_id = ai.flows[fd].info.id; - - pthread_mutex_lock(&rw.lock); - - if (rw.map[slot][fd]) { - pthread_mutex_unlock(&rw.lock); - free(a); - return 0; - } - - rw.map[slot][fd] = true; - - list_add_tail(&a->next, &rw.acks[slot]); - - pthread_mutex_unlock(&rw.lock); - - return 0; -} diff --git a/src/lib/tpm.c b/src/lib/tpm.c index 0ef1fda8..9229ea1a 100644 --- a/src/lib/tpm.c +++ b/src/lib/tpm.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Threadpool management * @@ -26,21 +26,32 @@ #include <ouroboros/errno.h> #include <ouroboros/list.h> +#include <ouroboros/pthread.h> #include <ouroboros/time.h> #include <ouroboros/tpm.h> +#ifdef CONFIG_OUROBOROS_DEBUG +#define OUROBOROS_PREFIX "tpm" +#include <ouroboros/logs.h> +#endif + #include <assert.h> -#include <pthread.h> #include <stdlib.h> +#include <string.h> +#include <unistd.h> -#define TPM_TIMEOUT 1000 +#define TPM_TIMEOUT 1000 struct pthr_el { struct list_head next; bool kill; bool busy; - + bool wait; +#ifdef CONFIG_OUROBOROS_DEBUG + struct timespec start; + struct timespec last; +#endif pthread_t thr; }; @@ -63,16 +74,45 @@ struct tpm { enum tpm_state state; pthread_cond_t cond; - pthread_mutex_t lock; + pthread_mutex_t mtx; pthread_t mgr; }; +#ifdef CONFIG_OUROBOROS_DEBUG +#define BETWEEN(a, x, y) ((a) > (x) && (a) <= (y)) +static void tpm_debug_thread(struct pthr_el * e) +{ + struct timespec now; + time_t diff; + time_t intv; + + if (e->wait || !e->busy) + return; + + clock_gettime(CLOCK_REALTIME, &now); + + diff = ts_diff_ms(&now, &e->start) / 1000; + intv = ts_diff_ms(&now, &e->last) / 1000; + + (void) diff; /* Never read if both build options off (0) */ + (void) intv; /* Never read if report option off (0) */ + + if (BETWEEN(TPM_DEBUG_REPORT_INTERVAL, 0, intv)) { + log_dbg("Thread %d:%lx running for %ld s.\n", + getpid(), (unsigned long) e->thr, diff); + e->last = now; + } + + if (BETWEEN(TPM_DEBUG_ABORT_TIMEOUT, 0, diff)) + assert(false); /* TODO: Grab a coffee and fire up GDB */ +} +#endif + static void tpm_join(struct tpm * tpm) { struct list_head * p; struct list_head * h; - list_for_each_safe(p, h, &tpm->pool) { struct pthr_el * e = list_entry(p, struct pthr_el, next); if (tpm->state != TPM_RUNNING) { @@ -86,15 +126,18 @@ static void tpm_join(struct tpm * tpm) list_for_each_safe(p, h, &tpm->pool) { struct pthr_el * e = list_entry(p, struct pthr_el, next); +#ifdef CONFIG_OUROBOROS_DEBUG + tpm_debug_thread(e); +#endif if (e->kill) { pthread_t thr = e->thr; list_del(&e->next); free(e); - pthread_mutex_unlock(&tpm->lock); + pthread_mutex_unlock(&tpm->mtx); pthread_join(thr, NULL); - pthread_mutex_lock(&tpm->lock); + pthread_mutex_lock(&tpm->mtx); } } } @@ -114,56 +157,59 @@ static void tpm_kill(struct tpm * tpm) } } -static void * tpmgr(void * o) +static int __tpm(struct tpm * tpm) { struct timespec dl; struct timespec to = TIMESPEC_INIT_MS(TPM_TIMEOUT); - struct tpm * tpm = (struct tpm *) o; - - while (true) { - clock_gettime(PTHREAD_COND_CLOCK, &dl); - ts_add(&dl, &to, &dl); - pthread_mutex_lock(&tpm->lock); + clock_gettime(PTHREAD_COND_CLOCK, &dl); + ts_add(&dl, &to, &dl); - if (tpm->state != TPM_RUNNING) { - tpm_join(tpm); - pthread_mutex_unlock(&tpm->lock); - break; - } + pthread_mutex_lock(&tpm->mtx); + if (tpm->state != TPM_RUNNING) { tpm_join(tpm); + pthread_mutex_unlock(&tpm->mtx); + return -1; + } - if (tpm->cur - tpm->wrk < tpm->min) { - size_t i; - for (i = 0; i < tpm->inc; ++i) { - struct pthr_el * e = malloc(sizeof(*e)); - if (e == NULL) - break; + tpm_join(tpm); - e->kill = false; - e->busy = false; + if (tpm->cur - tpm->wrk < tpm->min) { + size_t i; + for (i = 0; i < tpm->inc; ++i) { + struct pthr_el * e = malloc(sizeof(*e)); + if (e == NULL) + break; - if (pthread_create(&e->thr, NULL, - tpm->func, tpm->o)) { - free(e); - break; - } + memset(e, 0, sizeof(*e)); - list_add(&e->next, &tpm->pool); + if (pthread_create(&e->thr, NULL, tpm->func, tpm->o)) { + free(e); + break; } - tpm->cur += i; + list_add(&e->next, &tpm->pool); } - if (pthread_cond_timedwait(&tpm->cond, &tpm->lock, &dl) - == ETIMEDOUT) - if (tpm->cur - tpm->wrk > tpm->min) - tpm_kill(tpm); - - pthread_mutex_unlock(&tpm->lock); + tpm->cur += i; } + pthread_cleanup_push(__cleanup_mutex_unlock, &tpm->mtx); + + if (pthread_cond_timedwait(&tpm->cond, &tpm->mtx, &dl) == ETIMEDOUT) + if (tpm->cur - tpm->wrk > tpm->min) + tpm_kill(tpm); + + pthread_cleanup_pop(true); + + return 0; +} + +static void * tpmgr(void * o) +{ + while (__tpm((struct tpm *) o) == 0); + return (void *) 0; } @@ -175,11 +221,14 @@ struct tpm * tpm_create(size_t min, struct tpm * tpm; pthread_condattr_t cattr; + assert(func != NULL); + assert(inc > 0); + tpm = malloc(sizeof(*tpm)); if (tpm == NULL) goto fail_malloc; - if (pthread_mutex_init(&tpm->lock, NULL)) + if (pthread_mutex_init(&tpm->mtx, NULL)) goto fail_lock; if (pthread_condattr_init(&cattr)) @@ -208,7 +257,7 @@ struct tpm * tpm_create(size_t min, fail_cond: pthread_condattr_destroy(&cattr); fail_cattr: - pthread_mutex_destroy(&tpm->lock); + pthread_mutex_destroy(&tpm->mtx); fail_lock: free(tpm); fail_malloc: @@ -217,34 +266,39 @@ struct tpm * tpm_create(size_t min, int tpm_start(struct tpm * tpm) { - pthread_mutex_lock(&tpm->lock); + pthread_mutex_lock(&tpm->mtx); if (pthread_create(&tpm->mgr, NULL, tpmgr, tpm)) { - pthread_mutex_unlock(&tpm->lock); + pthread_mutex_unlock(&tpm->mtx); return -1; } tpm->state = TPM_RUNNING; - pthread_mutex_unlock(&tpm->lock); + pthread_mutex_unlock(&tpm->mtx); return 0; } void tpm_stop(struct tpm * tpm) { - pthread_mutex_lock(&tpm->lock); + pthread_mutex_lock(&tpm->mtx); + + if (tpm->state != TPM_RUNNING) { + pthread_mutex_unlock(&tpm->mtx); + return; + } tpm->state = TPM_NULL; - pthread_mutex_unlock(&tpm->lock); + pthread_mutex_unlock(&tpm->mtx); pthread_join(tpm->mgr, NULL); } void tpm_destroy(struct tpm * tpm) { - pthread_mutex_destroy(&tpm->lock); + pthread_mutex_destroy(&tpm->mtx); pthread_cond_destroy(&tpm->cond); free(tpm); @@ -260,40 +314,61 @@ static struct pthr_el * tpm_pthr_el(struct tpm * tpm, e = list_entry(p, struct pthr_el, next); if (e->thr == thr) return e; - } return NULL; } -void tpm_inc(struct tpm * tpm) +void tpm_begin_work(struct tpm * tpm) { struct pthr_el * e; - pthread_mutex_lock(&tpm->lock); +#ifdef CONFIG_OUROBOROS_DEBUG + struct timespec now; + clock_gettime(CLOCK_REALTIME, &now); +#endif + + pthread_mutex_lock(&tpm->mtx); e = tpm_pthr_el(tpm, pthread_self()); if (e != NULL) { - e->busy = false; - --tpm->wrk; + e->busy = true; + ++tpm->wrk; +#ifdef CONFIG_OUROBOROS_DEBUG + e->start = now; + e->last = now; +#endif } - pthread_mutex_unlock(&tpm->lock); + pthread_cond_signal(&tpm->cond); + + pthread_mutex_unlock(&tpm->mtx); } -void tpm_dec(struct tpm * tpm) +void tpm_wait_work(struct tpm * tpm) { struct pthr_el * e; - pthread_mutex_lock(&tpm->lock); + pthread_mutex_lock(&tpm->mtx); + + e = tpm_pthr_el(tpm, pthread_self()); + if (e != NULL) + e->wait = true; + + pthread_mutex_unlock(&tpm->mtx); +} + +void tpm_end_work(struct tpm * tpm) +{ + struct pthr_el * e; + + pthread_mutex_lock(&tpm->mtx); e = tpm_pthr_el(tpm, pthread_self()); if (e != NULL) { - e->busy = true; - ++tpm->wrk; + e->busy = false; + --tpm->wrk; } - pthread_cond_signal(&tpm->cond); - - pthread_mutex_unlock(&tpm->lock); + pthread_mutex_unlock(&tpm->mtx); } diff --git a/src/lib/tw.c b/src/lib/tw.c new file mode 100644 index 00000000..ccde7dd1 --- /dev/null +++ b/src/lib/tw.c @@ -0,0 +1,307 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Generic deadline-ordered callback queue (timing wheel) + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This library is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public License + * version 2.1 as published by the Free Software Foundation. + * + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with this library; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#if defined(__linux__) || defined(__CYGWIN__) +#define _DEFAULT_SOURCE +#else +#define _POSIX_C_SOURCE 200809L +#endif + +#include "config.h" + +#include <ouroboros/list.h> +#include <ouroboros/pthread.h> +#include <ouroboros/time.h> +#include <ouroboros/tw.h> + +#include <assert.h> +#include <stdbool.h> +#include <stdint.h> + +/* 3 levels × 256 slots, 1 ms / 16 ms / 256 ms per-slot resolution. */ +#define TW_LVLS 3 +#define TW_SLOTS 256 +#define TW_BUMP 4 +#define TW_RES 20 /* 2^20 ns ≈ 1 ms per slot at level 0. */ + +#define TW_SLOT(x) ((x) & (TW_SLOTS - 1)) + +static struct { + struct list_head levels[TW_LVLS][TW_SLOTS]; + size_t prv[TW_LVLS]; + pthread_mutex_t mtx; + pthread_mutex_t move_mtx; + bool initialised; +} tw; + +static size_t tw_lvl_res(size_t lvl) +{ + return TW_RES + TW_BUMP * lvl; +} + +/* Smallest level whose slot range covers the deadline. */ +static size_t tw_pick_lvl(uint64_t now_ns, + uint64_t deadline_ns) +{ + uint64_t delta; + size_t lvl; + + delta = deadline_ns > now_ns ? deadline_ns - now_ns : 0; + lvl = 0; + + while (lvl < TW_LVLS - 1 && (delta >> tw_lvl_res(lvl)) >= TW_SLOTS) + ++lvl; + + return lvl; +} + +static size_t tw_slot(uint64_t ns, + size_t lvl) +{ + return TW_SLOT(ns >> tw_lvl_res(lvl)); +} + +int tw_init(void) +{ + struct timespec now; + size_t i; + size_t j; + + assert(!tw.initialised); + + if (pthread_mutex_init(&tw.mtx, NULL)) + goto fail_mtx; + + if (pthread_mutex_init(&tw.move_mtx, NULL)) + goto fail_move_mtx; + + clock_gettime(PTHREAD_COND_CLOCK, &now); + + for (i = 0; i < TW_LVLS; ++i) { + tw.prv[i] = TW_SLOT(tw_slot(TS_TO_UINT64(now), i) - 1); + for (j = 0; j < TW_SLOTS; ++j) + list_head_init(&tw.levels[i][j]); + } + + tw.initialised = true; + + return 0; + + fail_move_mtx: + pthread_mutex_destroy(&tw.mtx); + fail_mtx: + return -1; +} + +void tw_fini(void) +{ + size_t i; + size_t j; + + assert(tw.initialised); + + for (i = 0; i < TW_LVLS; ++i) { + for (j = 0; j < TW_SLOTS; ++j) + assert(list_is_empty(&tw.levels[i][j])); + } + + pthread_mutex_destroy(&tw.move_mtx); + pthread_mutex_destroy(&tw.mtx); + + tw.initialised = false; +} + +void tw_init_entry(struct tw_entry * e) +{ + list_head_init(&e->next); + + e->deadline_ns = 0; + e->fire = NULL; + e->arg = NULL; + e->lvl = 0; +} + +void tw_post(struct tw_entry * e, + uint64_t deadline_ns, + tw_fire_fn_t fire, + void * arg) +{ + struct timespec now; + size_t lvl; + size_t slot; + + assert(tw.initialised); + + clock_gettime(PTHREAD_COND_CLOCK, &now); + + lvl = tw_pick_lvl(TS_TO_UINT64(now), deadline_ns); + /* +1 so deadline <= slot_start; lands later in slot. */ + slot = TW_SLOT(tw_slot(deadline_ns, lvl) + 1); + + e->deadline_ns = deadline_ns; + e->fire = fire; + e->arg = arg; + e->lvl = lvl; + + pthread_mutex_lock(&tw.mtx); + + if (!list_is_empty(&e->next)) + list_del(&e->next); + + list_add_tail(&e->next, &tw.levels[lvl][slot]); + + pthread_mutex_unlock(&tw.mtx); +} + +void tw_cancel(struct tw_entry * e) +{ + if (e == NULL) + return; + + assert(tw.initialised); + + pthread_mutex_lock(&tw.mtx); + + if (!list_is_empty(&e->next)) { + list_del(&e->next); + list_head_init(&e->next); + } + + pthread_mutex_unlock(&tw.mtx); +} + +void tw_move(void) +{ + struct timespec now; + struct list_head deferred; + struct list_head * p; + uint64_t now_ns; + size_t i; + size_t j; + size_t cur; + + assert(tw.initialised); + + if (pthread_mutex_trylock(&tw.move_mtx) != 0) + return; + + pthread_cleanup_push(__cleanup_mutex_unlock, &tw.move_mtx); + + pthread_mutex_lock(&tw.mtx); + + pthread_cleanup_push(__cleanup_mutex_unlock, &tw.mtx); + + clock_gettime(PTHREAD_COND_CLOCK, &now); + now_ns = TS_TO_UINT64(now); + + for (i = 0; i < TW_LVLS; ++i) { + cur = tw_slot(now_ns, i); + + j = tw.prv[i]; + if (cur < j) + cur += TW_SLOTS; + + while (j++ < cur) { + size_t s = TW_SLOT(j); + + /* Pop-front so fire may mutate any entry. */ + list_head_init(&deferred); + + while (!list_is_empty(&tw.levels[i][s])) { + struct tw_entry * e; + p = tw.levels[i][s].nxt; + e = list_entry(p, struct tw_entry, next); + list_del(&e->next); + + if (e->deadline_ns > now_ns) { + list_add_tail(&e->next, &deferred); + continue; + } + + pthread_mutex_unlock(&tw.mtx); + e->fire(e->arg); + pthread_mutex_lock(&tw.mtx); + } + + while (!list_is_empty(&deferred)) { + p = deferred.nxt; + list_del(p); + list_add_tail(p, &tw.levels[i][s]); + } + } + + tw.prv[i] = TW_SLOT(cur); + } + + pthread_cleanup_pop(true); /* tw.mtx */ + pthread_cleanup_pop(true); /* tw.move_mtx */ +} + +/* Earliest pending deadline at level lvl, INT64_MAX if level is empty. */ +static int64_t tw_lvl_earliest(size_t lvl, + uint64_t now_ns) +{ + size_t cur = tw_slot(now_ns, lvl); + size_t j; + + for (j = 1; j <= TW_SLOTS; ++j) { + size_t s = TW_SLOT(cur + j); + + if (list_is_empty(&tw.levels[lvl][s])) + continue; + + return (int64_t)(now_ns + ((uint64_t) j << tw_lvl_res(lvl))); + } + + return INT64_MAX; +} + +void tw_next_expiry(struct timespec * out) +{ + struct timespec now; + uint64_t now_ns; + int64_t earliest = INT64_MAX; + size_t i; + + assert(tw.initialised); + + clock_gettime(PTHREAD_COND_CLOCK, &now); + now_ns = TS_TO_UINT64(now); + + pthread_mutex_lock(&tw.mtx); + + for (i = 0; i < TW_LVLS; ++i) { + int64_t dl = tw_lvl_earliest(i, now_ns); + if (dl < earliest) + earliest = dl; + } + + pthread_mutex_unlock(&tw.mtx); + + if (earliest == INT64_MAX) { + /* Empty wheel: tv_nsec=-1 is an invalid normalised value. */ + out->tv_sec = 0; + out->tv_nsec = -1; + } else { + UINT64_TO_TS((uint64_t) earliest, out); + } +} diff --git a/src/lib/utils.c b/src/lib/utils.c index fdbcd9d9..6b49cc00 100644 --- a/src/lib/utils.c +++ b/src/lib/utils.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Handy utilities * @@ -20,23 +20,38 @@ * Foundation, Inc., http://www.fsf.org/about/contact/. */ -#define _POSIX_C_SOURCE 200809L +#define _DEFAULT_SOURCE + +#include "config.h" #include <ouroboros/utils.h> +#include <ctype.h> +#include <grp.h> +#include <pwd.h> #include <stdlib.h> #include <string.h> +int bufcmp(const buffer_t * a, + const buffer_t * b) +{ + if (a->len != b->len) + return a->len < b->len ? -1 : 1; + + return memcmp(a->data, b->data, a->len); +} + + int n_digits(unsigned i) { - int n = 1; + int n = 1; - while (i > 9) { - ++n; - i /= 10; - } + while (i > 9) { + ++n; + i /= 10; + } - return n; + return n; } char * path_strip(const char * src) @@ -57,6 +72,24 @@ char * path_strip(const char * src) return dst; } +char * trim_whitespace(char * str) +{ + char * end; + + while (isspace((unsigned char) *str)) + str++; + + if (*str == '\0') + return str; + + /* Trim trailing space */ + end = str + strlen(str) - 1; + while (end > str && isspace((unsigned char)*end)) + *end-- = '\0'; + + return str; +} + size_t argvlen(const char ** argv) { size_t argc = 0; @@ -109,5 +142,72 @@ char ** argvdup(char ** argv) } argv_dup[argc] = NULL; + return argv_dup; } + +bool is_ouroboros_member_uid(uid_t uid) +{ + struct group * grp; + struct passwd * pw; +#ifdef __APPLE__ + unsigned int gid; + int * groups = NULL; +#else + gid_t gid; + gid_t * groups = NULL; +#endif + int ngroups; + int i; + + /* Root is always privileged */ + if (uid == 0) + return true; + + grp = getgrnam("ouroboros"); + if (grp == NULL) + return false; + + gid = grp->gr_gid; + + pw = getpwuid(uid); + if (pw == NULL) + return false; + + if (pw->pw_gid == gid) + return true; + + ngroups = 0; + getgrouplist(pw->pw_name, pw->pw_gid, NULL, &ngroups); + if (ngroups <= 0) + return false; + + groups = malloc(ngroups * sizeof(*groups)); + if (groups == NULL) + return false; + + if (getgrouplist(pw->pw_name, pw->pw_gid, groups, &ngroups) < 0) { + free(groups); + return false; + } + + for (i = 0; i < ngroups; i++) { +#ifdef __APPLE__ + if (groups[i] == (int) gid) { +#else + if (groups[i] == gid) { +#endif + free(groups); + return true; + } + } + + free(groups); + + return false; +} + +bool is_ouroboros_member(void) +{ + return is_ouroboros_member_uid(getuid()); +} diff --git a/src/tools/CMakeLists.txt b/src/tools/CMakeLists.txt index 7c40d9ae..6b418838 100644 --- a/src/tools/CMakeLists.txt +++ b/src/tools/CMakeLists.txt @@ -1,9 +1,76 @@ -add_subdirectory(irm) -add_subdirectory(ocbr) -add_subdirectory(oecho) -add_subdirectory(obc) -add_subdirectory(oping) -add_subdirectory(operf) -if (CMAKE_SYSTEM_NAME STREQUAL "Linux") - add_subdirectory(ovpn) -endif () +# Tools build configuration + +set(TOOLS_INCLUDE_DIRS + ${CMAKE_CURRENT_SOURCE_DIR} + ${CMAKE_SOURCE_DIR}/include + ${CMAKE_BINARY_DIR}/include +) + +set(IRM_SOURCES + irm/irm.c + irm/irm_bind_program.c + irm/irm_bind_process.c + irm/irm_bind_ipcp.c + irm/irm_ipcp_create.c + irm/irm_ipcp_destroy.c + irm/irm_ipcp_bootstrap.c + irm/irm_ipcp_enroll.c + irm/irm_ipcp_list.c + irm/irm_ipcp_connect.c + irm/irm_ipcp_disconnect.c + irm/irm_unbind_program.c + irm/irm_unbind_process.c + irm/irm_unbind_ipcp.c + irm/irm_unbind.c + irm/irm_bind.c + irm/irm_ipcp.c + irm/irm_name.c + irm/irm_name_create.c + irm/irm_name_destroy.c + irm/irm_name_reg.c + irm/irm_name_unreg.c + irm/irm_name_list.c + irm/irm_utils.c +) + +add_executable(irm ${IRM_SOURCES}) +target_include_directories(irm PRIVATE ${TOOLS_INCLUDE_DIRS}) +target_link_libraries(irm PRIVATE ouroboros-irm) +install(TARGETS irm RUNTIME DESTINATION ${CMAKE_INSTALL_SBINDIR}) + +add_executable(oping oping/oping.c) +target_include_directories(oping PRIVATE ${TOOLS_INCLUDE_DIRS}) +target_link_libraries(oping PRIVATE ${LIBM_LIBRARIES} ouroboros-dev) +install(TARGETS oping RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR}) + +add_executable(oecho oecho/oecho.c) +target_include_directories(oecho PRIVATE ${TOOLS_INCLUDE_DIRS}) +target_link_libraries(oecho PRIVATE ouroboros-dev) +install(TARGETS oecho RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR}) + +add_executable(ocbr ocbr/ocbr.c) +target_include_directories(ocbr PRIVATE ${TOOLS_INCLUDE_DIRS}) +target_link_libraries(ocbr PRIVATE ouroboros-dev) +install(TARGETS ocbr RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR}) + +add_executable(obc obc/obc.c) +target_include_directories(obc PRIVATE ${TOOLS_INCLUDE_DIRS}) +target_link_libraries(obc PRIVATE ouroboros-dev) +install(TARGETS obc RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR}) + +add_executable(operf operf/operf.c) +target_include_directories(operf PRIVATE ${TOOLS_INCLUDE_DIRS}) +target_link_libraries(operf PRIVATE ouroboros-dev) +install(TARGETS operf RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR}) + +add_executable(oftp oftp/oftp.c) +target_include_directories(oftp PRIVATE ${TOOLS_INCLUDE_DIRS}) +target_link_libraries(oftp PRIVATE ouroboros-dev) +install(TARGETS oftp RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR}) + +if(CMAKE_SYSTEM_NAME STREQUAL "Linux") + add_executable(ovpn ovpn/ovpn.c) + target_include_directories(ovpn PRIVATE ${TOOLS_INCLUDE_DIRS}) + target_link_libraries(ovpn PRIVATE ouroboros-dev) + install(TARGETS ovpn RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR}) +endif() diff --git a/src/tools/irm/CMakeLists.txt b/src/tools/irm/CMakeLists.txt deleted file mode 100644 index e5e5c466..00000000 --- a/src/tools/irm/CMakeLists.txt +++ /dev/null @@ -1,39 +0,0 @@ -include_directories(${CMAKE_CURRENT_SOURCE_DIR}) -include_directories(${CMAKE_CURRENT_BINARY_DIR}) - -include_directories(${CMAKE_SOURCE_DIR}/include) -include_directories(${CMAKE_BINARY_DIR}/include) - -set(SOURCE_FILES - # Add source files here - irm.c - irm_bind_program.c - irm_bind_process.c - irm_bind_ipcp.c - irm_ipcp_create.c - irm_ipcp_destroy.c - irm_ipcp_bootstrap.c - irm_ipcp_enroll.c - irm_ipcp_list.c - irm_ipcp_connect.c - irm_ipcp_disconnect.c - irm_unbind_program.c - irm_unbind_process.c - irm_unbind_ipcp.c - irm_unbind.c - irm_bind.c - irm_ipcp.c - irm_name.c - irm_name_create.c - irm_name_destroy.c - irm_name_reg.c - irm_name_unreg.c - irm_name_list.c - irm_utils.c - ) - -add_executable(irm ${SOURCE_FILES}) - -target_link_libraries(irm LINK_PUBLIC ouroboros-irm) - -install(TARGETS irm RUNTIME DESTINATION ${CMAKE_INSTALL_SBINDIR}) diff --git a/src/tools/irm/irm.c b/src/tools/irm/irm.c index ba0f4713..6c2719d0 100644 --- a/src/tools/irm/irm.c +++ b/src/tools/irm/irm.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * A tool to instruct the IRM daemon * diff --git a/src/tools/irm/irm_bind.c b/src/tools/irm/irm_bind.c index 2e8b14ef..3107837a 100644 --- a/src/tools/irm/irm_bind.c +++ b/src/tools/irm/irm_bind.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Bind names in the processing system * diff --git a/src/tools/irm/irm_bind_ipcp.c b/src/tools/irm/irm_bind_ipcp.c index 7d5dd636..4c183534 100644 --- a/src/tools/irm/irm_bind_ipcp.c +++ b/src/tools/irm/irm_bind_ipcp.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Bind IPCP Instance to a name * diff --git a/src/tools/irm/irm_bind_process.c b/src/tools/irm/irm_bind_process.c index fffd5fe9..fee0c46b 100644 --- a/src/tools/irm/irm_bind_process.c +++ b/src/tools/irm/irm_bind_process.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Bind a process to a name * diff --git a/src/tools/irm/irm_bind_program.c b/src/tools/irm/irm_bind_program.c index 8a0dc33c..14d09db7 100644 --- a/src/tools/irm/irm_bind_program.c +++ b/src/tools/irm/irm_bind_program.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Bind programs to a name * diff --git a/src/tools/irm/irm_ipcp.c b/src/tools/irm/irm_ipcp.c index 63e617d9..34458a20 100644 --- a/src/tools/irm/irm_ipcp.c +++ b/src/tools/irm/irm_ipcp.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * A tool to instruct the IRM daemon * diff --git a/src/tools/irm/irm_ipcp_bootstrap.c b/src/tools/irm/irm_ipcp_bootstrap.c index b8e5c54d..de73b076 100644 --- a/src/tools/irm/irm_ipcp_bootstrap.c +++ b/src/tools/irm/irm_ipcp_bootstrap.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Bootstrap IPC Processes * @@ -46,90 +46,131 @@ #include <stdlib.h> #include <string.h> #include <arpa/inet.h> +#include <netinet/in.h> #ifdef __FreeBSD__ #include <sys/socket.h> #endif -#define UNICAST "unicast" -#define BROADCAST "broadcast" -#define UDP "udp" -#define ETH_LLC "eth-llc" -#define ETH_DIX "eth-dix" -#define LOCAL "local" - -#define MD5 "MD5" -#define SHA3_224 "SHA3_224" -#define SHA3_256 "SHA3_256" -#define SHA3_384 "SHA3_384" -#define SHA3_512 "SHA3_512" - -#define DEFAULT_ADDR_SIZE 4 -#define DEFAULT_EID_SIZE 8 -#define DEFAULT_DDNS 0 -#define DEFAULT_TTL 60 -#define DEFAULT_ADDR_AUTH ADDR_AUTH_FLAT_RANDOM -#define DEFAULT_ROUTING ROUTING_LINK_STATE -#define DEFAULT_CONG_AVOID CA_MB_ECN -#define DEFAULT_HASH_ALGO DIR_HASH_SHA3_256 -#define DEFAULT_ETHERTYPE 0xA000 -#define DEFAULT_UDP_PORT 0x0D6B /* 3435 */ - -#define FLAT_RANDOM_ADDR_AUTH "flat" -#define LINK_STATE_ROUTING "link_state" -#define LINK_STATE_LFA_ROUTING "lfa" -#define LINK_STATE_ECM_ROUTING "ecmp" -#define NONE_CA "none" -#define MB_ECN_CA "mb-ecn" +#define UNICAST "unicast" +#define BROADCAST "broadcast" +#define IP_UDP4 "udp4" +#define IP_UDP6 "udp6" +#define ETH_LLC "eth-llc" +#define ETH_DIX "eth-dix" +#define LOCAL "local" + +#define MD5 "MD5" +#define SHA3_224 "SHA3_224" +#define SHA3_256 "SHA3_256" +#define SHA3_384 "SHA3_384" +#define SHA3_512 "SHA3_512" + +#define FLAT_RANDOM "flat" +#define DHT_DIR "DHT" +#define LINK_STATE "link_state" +#define LINK_STATE_LFA "lfa" +#define LINK_STATE_ECM "ecmp" +#define NONE_CA "none" +#define MB_ECN_CA "mb-ecn" + +#define DT(x) default_dt_config.x +#define DHT(x) default_dht_config.params.x +#define UNI(x) default_uni_config.x +#define DIX(x) eth_dix_default_conf.eth.x +#define LLC(x) eth_llc_default_conf.eth.x +#define UD4(x) udp4_default_conf.udp4.x +#define UD6(x) udp6_default_conf.udp6.x + +static char * usage_str = \ + "Usage: irm ipcp bootstrap\n" + " name <ipcp name>\n" + " layer <layer name>\n" + " [type [TYPE]]\n" + "where TYPE in {" UNICAST " " BROADCAST " " LOCAL " " + IP_UDP4 " " IP_UDP6 " " ETH_LLC " " ETH_DIX "},\n\n" + "if TYPE == " UNICAST "\n" + " [addr_auth <ADDRESS_POLICY> (default: %s)]\n" + " [directory <DIRECTORY_POLICY> (default: %s)]\n" + " [hash [ALGORITHM] (default: %s)]\n" + " [routing <ROUTING_POLICY> (default: %s)]\n" + " [congestion <CONG_POLICY> (default: %s)]\n" + " [autobind]\n\n" + "where ADDRESS_POLICY in {" FLAT_RANDOM "}\n" + " DIRECTORY_POLICY in {" DHT_DIR "}\n" + " ALGORITHM in {" SHA3_224 " " SHA3_256 " " + SHA3_384 " " SHA3_512 "}\n" + " ROUTING_POLICY in {" LINK_STATE " " + LINK_STATE_LFA " " LINK_STATE_ECM "}\n" + " CONG_POLICY in {" NONE_CA " " MB_ECN_CA "}\n" + " [Data Transfer Constants]\n" + " [addr <address size> (default: %d)]\n" + " [eid <eid size> (default: %d)]\n" + " [ttl <max time-to-live>, default: %d)]\n\n" + "if DIRECTORY_POLICY == " DHT_DIR "\n" + " [dht_alpha <search factor> (default: %u)]\n" + " [dht_k <replication factor> (default: %u)]\n" + " [dht_t_expire <expiration (s)> (default: %u)]\n" + " [dht_t_refresh <contact refresh (s)> (default: %u)]\n" + " [dht_t_replicate <replication (s)> (default: %u)]\n\n" + "if ROUTING_POLICY == " LINK_STATE "\n" + " [ls_t_recalc <pff recalc interval (s)> (default: %ld)]\n" + " [ls_t_update <LSA update interval (s)> (default: %ld)]\n" + " [ls_t_timeo <link timeout (s)> (default: %ld)]\n\n" + "if TYPE == " IP_UDP4 "\n" + " ip <IP address in dotted notation>\n" + " [port <UDP port> (default: %d)]\n" + " [dns <DDNS IPv4 address in dotted notation>" + " (default: none)]\n\n" + "if TYPE == " IP_UDP6 "\n" + " ip <IPv6 address>\n" + " [port <UDP port> (default: %d)]\n" + " [dns <DDNS IPv6 address>" + " (default: none)]\n\n" + + "if TYPE == " ETH_LLC "\n" + " dev <interface name>\n" + " [hash [ALGORITHM] (default: %s)]\n" + "where ALGORITHM in {" SHA3_224 " " SHA3_256 " " + SHA3_384 " " SHA3_512 "}\n\n" + "if TYPE == " ETH_DIX "\n" + " dev <interface name>\n" + " [ethertype <ethertype> (default: 0x%4X)]\n" + " [hash [ALGORITHM] (default: %s)]\n" + "where ALGORITHM in {" SHA3_224 " " SHA3_256 " " + SHA3_384 " " SHA3_512 "}\n\n" + "if TYPE == " LOCAL "\n" + " [hash [ALGORITHM] (default: %s)]\n" + "where ALGORITHM in {" SHA3_224 " " SHA3_256 " " + SHA3_384 " " SHA3_512 "}\n\n" + "if TYPE == " BROADCAST "\n" + " [autobind]\n\n"; static void usage(void) { /* FIXME: Add ipcp_config stuff. */ - printf("Usage: irm ipcp bootstrap\n" - " name <ipcp name>\n" - " layer <layer name>\n" - " [type [TYPE]]\n" - "where TYPE in {" UNICAST " " BROADCAST " " LOCAL " " - UDP " " ETH_LLC " " ETH_DIX "},\n\n" - "if TYPE == " UNICAST "\n" - " [addr <address size> (default: %d)]\n" - " [eid <eid size> (default: %d)]\n" - " [ttl (max time-to-live value, default: %d)]\n" - " [addr_auth <ADDRESS_POLICY> (default: %s)]\n" - " [routing <ROUTING_POLICY> (default: %s)]\n" - " [congestion <CONG_POLICY> (default: %s)]\n" - " [hash [ALGORITHM] (default: %s)]\n" - " [autobind]\n" - "where ADDRESS_POLICY in {" FLAT_RANDOM_ADDR_AUTH "}\n" - " ROUTING_POLICY in {" LINK_STATE_ROUTING " " - LINK_STATE_LFA_ROUTING " " LINK_STATE_ECM_ROUTING "}\n" - " CONG_POLICY in {" NONE_CA " " MB_ECN_CA "}\n" - " ALGORITHM in {" SHA3_224 " " SHA3_256 " " - SHA3_384 " " SHA3_512 "}\n\n" - "if TYPE == " UDP "\n" - " ip <IP address in dotted notation>\n" - " [port <UDP port> (default: %d)]\n" - " [dns <DDNS IP address in dotted notation>" - " (default: none)]\n\n" - "if TYPE == " ETH_LLC "\n" - " dev <interface name>\n" - " [hash [ALGORITHM] (default: %s)]\n" - "where ALGORITHM in {" SHA3_224 " " SHA3_256 " " - SHA3_384 " " SHA3_512 "}\n\n" - "if TYPE == " ETH_DIX "\n" - " dev <interface name>\n" - " [ethertype <ethertype> (default: 0x%4X)]\n" - " [hash [ALGORITHM] (default: %s)]\n" - "where ALGORITHM in {" SHA3_224 " " SHA3_256 " " - SHA3_384 " " SHA3_512 "}\n\n" - "if TYPE == " LOCAL "\n" - " [hash [ALGORITHM] (default: %s)]\n" - "where ALGORITHM in {" SHA3_224 " " SHA3_256 " " - SHA3_384 " " SHA3_512 "}\n\n" - "if TYPE == " BROADCAST "\n" - " [autobind]\n\n", - DEFAULT_ADDR_SIZE, DEFAULT_EID_SIZE, DEFAULT_TTL, - FLAT_RANDOM_ADDR_AUTH, LINK_STATE_ROUTING, MB_ECN_CA, - SHA3_256, DEFAULT_UDP_PORT, SHA3_256, 0xA000, SHA3_256, + printf(usage_str, + /* unicast */ + FLAT_RANDOM, DHT_DIR, SHA3_256, LINK_STATE, MB_ECN_CA, + /* dt */ + DT(addr_size), DT(eid_size), DT(max_ttl), + /* dht */ + DHT(alpha), DHT(k), DHT(t_expire), + DHT(t_refresh), DHT(t_replicate), + /* ls */ + default_ls_config.t_recalc, default_ls_config.t_update, + default_ls_config.t_timeo, + /* udp4 */ + UD4(port), + /* udp6 */ + UD6(port), + /* eth_llc */ + SHA3_256, + /* eth_dix */ + DIX(ethertype), + SHA3_256, + /* local */ + SHA3_256, + /* broadcast */ SHA3_256); } @@ -139,26 +180,32 @@ int do_bootstrap_ipcp(int argc, char * ipcp = NULL; pid_t pid = -1; struct ipcp_config conf; - uint8_t addr_size = DEFAULT_ADDR_SIZE; - uint8_t eid_size = DEFAULT_EID_SIZE; - uint8_t max_ttl = DEFAULT_TTL; - enum pol_addr_auth addr_auth_type = DEFAULT_ADDR_AUTH; - enum pol_routing routing_type = DEFAULT_ROUTING; - enum pol_dir_hash hash_algo = DEFAULT_HASH_ALGO; - enum pol_cong_avoid cong_avoid = DEFAULT_CONG_AVOID; - uint32_t ip_addr = 0; - uint32_t dns_addr = DEFAULT_DDNS; + struct dir_config dir_config = default_dir_config; + uint8_t addr_size = DT(addr_size); + uint8_t eid_size = DT(eid_size); + uint8_t max_ttl = DT(max_ttl); + struct routing_config routing = default_routing_config; + enum pol_addr_auth addr_auth_type = UNI(addr_auth_type); + enum pol_cong_avoid cong_avoid = UNI(cong_avoid); + enum pol_dir_hash hash_algo = DIR_HASH_SHA3_256; + char * ipstr = NULL; + char * dnsstr = NULL; + struct in_addr ip4_addr = {.s_addr = INADDR_ANY}; + struct in_addr dns4_addr = UD4(dns_addr); + int port4 = UD4(port); + struct in6_addr ip6_addr = IN6ADDR_ANY_INIT; + struct in6_addr dns6_addr = UD6(dns_addr); + int port6 = UD6(port); char * ipcp_type = NULL; enum ipcp_type type = IPCP_INVALID; char * layer = NULL; char * dev = NULL; - uint16_t ethertype = DEFAULT_ETHERTYPE; + uint16_t ethertype = DIX(ethertype); struct ipcp_list_info * ipcps; ssize_t len = 0; int i = 0; bool autobind = false; int cargs; - int port = DEFAULT_UDP_PORT; while (argc > 0) { cargs = 2; @@ -180,11 +227,9 @@ int do_bootstrap_ipcp(int argc, else goto unknown_param; } else if (matches(*argv, "ip") == 0) { - if (inet_pton (AF_INET, *(argv + 1), &ip_addr) != 1) - goto unknown_param; + ipstr = *(argv + 1); } else if (matches(*argv, "dns") == 0) { - if (inet_pton(AF_INET, *(argv + 1), &dns_addr) != 1) - goto unknown_param; + dnsstr = *(argv + 1); } else if (matches(*argv, "device") == 0) { dev = *(argv + 1); } else if (matches(*argv, "ethertype") == 0) { @@ -206,26 +251,50 @@ int do_bootstrap_ipcp(int argc, } else if (matches(*argv, "ttl") == 0) { max_ttl = atoi(*(argv + 1)); } else if (matches(*argv, "port") == 0) { - port = atoi(*(argv + 1)); + port4 = atoi(*(argv + 1)); + port6 = port4; } else if (matches(*argv, "autobind") == 0) { autobind = true; cargs = 1; } else if (matches(*argv, "addr_auth") == 0) { - if (strcmp(FLAT_RANDOM_ADDR_AUTH, *(argv + 1)) == 0) + if (strcmp(FLAT_RANDOM, *(argv + 1)) == 0) addr_auth_type = ADDR_AUTH_FLAT_RANDOM; else goto unknown_param; - } else if (matches(*argv, "routing") == 0) { - if (strcmp(LINK_STATE_ROUTING, *(argv + 1)) == 0) - routing_type = ROUTING_LINK_STATE; - else if (strcmp(LINK_STATE_LFA_ROUTING, - *(argv + 1)) == 0) - routing_type = ROUTING_LINK_STATE_LFA; - else if (strcmp(LINK_STATE_ECM_ROUTING, - *(argv + 1)) == 0) - routing_type = ROUTING_LINK_STATE_ECMP; + } else if (matches(*argv, "directory") == 0) { + if (strcmp(DHT_DIR, *(argv + 1)) == 0) + dir_config.pol = DIR_DHT; else goto unknown_param; + } else if (matches(*argv, "dht_alpha") == 0) { + dir_config.dht.params.alpha = atoi(*(argv + 1)); + } else if (matches(*argv, "dht_k") == 0) { + dir_config.dht.params.k = atoi(*(argv + 1)); + } else if (matches(*argv, "dht_t_expire") == 0) { + dir_config.dht.params.t_expire = atoi(*(argv + 1)); + } else if (matches(*argv, "dht_t_refresh") == 0) { + dir_config.dht.params.t_refresh = atoi(*(argv + 1)); + } else if (matches(*argv, "dht_t_replicate") == 0) { + dir_config.dht.params.t_replicate = atoi(*(argv + 1)); + } else if (matches(*argv, "routing") == 0) { + if (strcmp(LINK_STATE, *(argv + 1)) == 0) { + routing.pol = ROUTING_LINK_STATE; + routing.ls.pol = LS_SIMPLE; + } else if (strcmp(LINK_STATE_LFA, *(argv + 1)) == 0) { + routing.pol = ROUTING_LINK_STATE; + routing.ls.pol = LS_LFA; + } else if (strcmp(LINK_STATE_ECM, *(argv + 1)) == 0) { + routing.pol = ROUTING_LINK_STATE; + routing.ls.pol = LS_ECMP; + } else { + goto unknown_param; + } + } else if (matches(*argv, "ls_t_timeo") == 0) { + routing.ls.t_timeo = atoi(*(argv + 1)); + } else if (matches(*argv, "ls_t_update") == 0) { + routing.ls.t_update = atoi(*(argv + 1)); + } else if (matches(*argv, "ls_t_recalc") == 0) { + routing.ls.t_recalc = atoi(*(argv + 1)); } else if (matches(*argv, "congestion") == 0) { if (strcmp(NONE_CA, *(argv + 1)) == 0) cong_avoid = CA_NONE; @@ -257,21 +326,59 @@ int do_bootstrap_ipcp(int argc, } if (ipcp_type != NULL) { - if (strcmp(ipcp_type, UNICAST) == 0) + if (matches(ipcp_type, UNICAST) == 0) type = IPCP_UNICAST; - else if (strcmp(ipcp_type, BROADCAST) == 0) + else if (matches(ipcp_type, BROADCAST) == 0) type = IPCP_BROADCAST; - else if (strcmp(ipcp_type, UDP) == 0) - type = IPCP_UDP; - else if (strcmp(ipcp_type, ETH_LLC) == 0) - type = IPCP_ETH_LLC; - else if (strcmp(ipcp_type, ETH_DIX) == 0) + else if (matches(ipcp_type, IP_UDP4) == 0) + type = IPCP_UDP4; + else if (matches(ipcp_type, IP_UDP6) == 0) + type = IPCP_UDP6; + else if (matches(ipcp_type, ETH_DIX) == 0) type = IPCP_ETH_DIX; - else if (strcmp(ipcp_type, LOCAL) == 0) + else if (matches(ipcp_type, ETH_LLC) == 0) + type = IPCP_ETH_LLC; + else if (matches(ipcp_type, LOCAL) == 0) type = IPCP_LOCAL; else goto fail_usage; } + if (type == IPCP_UDP4) { + if (inet_pton (AF_INET, ipstr, &ip4_addr) != 1) { + printf("Invalid IPv4 address: \"%s\".\n", ipstr); + goto fail_usage; + } + + if (ip4_addr.s_addr == INADDR_ANY) { + printf("Cannot use IPv4 address: \"%s\".\n", ipstr); + goto fail_usage; + } + + if (dnsstr != NULL && + inet_pton(AF_INET, dnsstr, &dns4_addr) != 1) { + printf("Invalid DNS IPv4 address: \"%s\".\n", dnsstr); + goto fail_usage; + } + } + + if (type == IPCP_UDP6) { + if (inet_pton(AF_INET6, ipstr, &ip6_addr) != 1) { + printf("Invalid IPv6 address: \"%s\".\n", ipstr); + goto fail_usage; + } + + if (IN6_IS_ADDR_UNSPECIFIED(&ip6_addr)) { + printf("Cannot use IPv6 address: \"%s\".\n", ipstr); + goto fail_usage; + } + + if (dnsstr != NULL && + inet_pton(AF_INET6, dnsstr, &dns6_addr) != 1) { + printf("Invalid DNS IPv6 address: \"%s\".\n", dnsstr); + goto fail_usage; + } + } + if (pid == -1) { if (ipcp_type == NULL) { printf("No IPCPs matching %s found.\n\n", ipcp); @@ -315,16 +422,20 @@ int do_bootstrap_ipcp(int argc, conf.unicast.dt.addr_size = addr_size; conf.unicast.dt.eid_size = eid_size; conf.unicast.dt.max_ttl = max_ttl; - conf.unicast.dt.routing_type = routing_type; + conf.unicast.dt.routing = routing; conf.unicast.addr_auth_type = addr_auth_type; conf.unicast.cong_avoid = cong_avoid; + conf.unicast.dir = dir_config; break; - case IPCP_UDP: - if (ip_addr == 0) - goto fail_usage; - conf.udp.ip_addr = ip_addr; - conf.udp.dns_addr = dns_addr; - conf.udp.port = port; + case IPCP_UDP4: + conf.udp4.ip_addr = ip4_addr; + conf.udp4.dns_addr = dns4_addr; + conf.udp4.port = port4; + break; + case IPCP_UDP6: + conf.udp6.ip_addr = ip6_addr; + conf.udp6.dns_addr = dns6_addr; + conf.udp6.port = port6; break; case IPCP_ETH_DIX: conf.eth.ethertype = ethertype; diff --git a/src/tools/irm/irm_ipcp_connect.c b/src/tools/irm/irm_ipcp_connect.c index 68e13bd0..fb21faec 100644 --- a/src/tools/irm/irm_ipcp_connect.c +++ b/src/tools/irm/irm_ipcp_connect.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Connect components of unicast or broadcast IPC processes * @@ -100,16 +100,18 @@ int do_connect_ipcp(int argc, } if (qos != NULL) { - if (strcmp(qos, "best") == 0) - qs = qos_best_effort; - else if (strcmp(qos, "raw") == 0) + if (strcmp(qos, "raw") == 0) qs = qos_raw; - else if (strcmp(qos, "video") == 0) - qs = qos_video; - else if (strcmp(qos, "voice") == 0) - qs = qos_voice; - else if (strcmp(qos, "data") == 0) - qs = qos_data; + else if (strcmp(qos, "safe") == 0) + qs = qos_raw_safe; + else if (strcmp(qos, "rt") == 0) + qs = qos_rt; + else if (strcmp(qos, "rt-safe") == 0) + qs = qos_rt_safe; + else if (strcmp(qos, "msg") == 0) + qs = qos_msg; + else if (strcmp(qos, "stream") == 0) + qs = qos_stream; else printf("Unknown QoS cube, defaulting to raw.\n"); } @@ -126,7 +128,7 @@ int do_connect_ipcp(int argc, if (wildcard_match(comp, MGMT) == 0) { component = MGMT_COMP; - /* FIXME: move to qos_data when stable */ + /* FIXME: move to qos_msg when stable */ if (irm_connect_ipcp(pid, dst, component, qos_raw)) return -1; } diff --git a/src/tools/irm/irm_ipcp_create.c b/src/tools/irm/irm_ipcp_create.c index 35d33782..c6b2074b 100644 --- a/src/tools/irm/irm_ipcp_create.c +++ b/src/tools/irm/irm_ipcp_create.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Create IPC Processes * @@ -46,7 +46,8 @@ #define UNICAST "unicast" #define BROADCAST "broadcast" -#define UDP "udp" +#define UDP4 "udp4" +#define UDP6 "udp6" #define ETH_LLC "eth-llc" #define ETH_DIX "eth-dix" #define LOCAL "local" @@ -57,7 +58,7 @@ static void usage(void) " name <ipcp name>\n" " type [TYPE]\n\n" "where TYPE in {" UNICAST " " BROADCAST " " LOCAL " " - UDP " " ETH_LLC " " ETH_DIX "}\n"); + UDP4 " " UDP6 " " ETH_LLC " " ETH_DIX "}\n"); } int do_create_ipcp(int argc, @@ -92,8 +93,10 @@ int do_create_ipcp(int argc, type = IPCP_UNICAST; else if (strcmp(ipcp_type, BROADCAST) == 0) type = IPCP_BROADCAST; - else if (strcmp(ipcp_type, UDP) == 0) - type = IPCP_UDP; + else if (strcmp(ipcp_type, UDP4) == 0) + type = IPCP_UDP4; + else if (strcmp(ipcp_type, UDP6) == 0) + type = IPCP_UDP6; else if (strcmp(ipcp_type, LOCAL) == 0) type = IPCP_LOCAL; else if (strcmp(ipcp_type, ETH_LLC) == 0) diff --git a/src/tools/irm/irm_ipcp_destroy.c b/src/tools/irm/irm_ipcp_destroy.c index 1a5e564e..523836af 100644 --- a/src/tools/irm/irm_ipcp_destroy.c +++ b/src/tools/irm/irm_ipcp_destroy.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Destroy IPC Processes * diff --git a/src/tools/irm/irm_ipcp_disconnect.c b/src/tools/irm/irm_ipcp_disconnect.c index 7ce724e1..0f37ec91 100644 --- a/src/tools/irm/irm_ipcp_disconnect.c +++ b/src/tools/irm/irm_ipcp_disconnect.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Connect components of unicast or broadcast IPC processes * diff --git a/src/tools/irm/irm_ipcp_enroll.c b/src/tools/irm/irm_ipcp_enroll.c index 86a22a71..350b536e 100644 --- a/src/tools/irm/irm_ipcp_enroll.c +++ b/src/tools/irm/irm_ipcp_enroll.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Enroll IPC Processes * diff --git a/src/tools/irm/irm_ipcp_list.c b/src/tools/irm/irm_ipcp_list.c index dfa3099f..a211a02b 100644 --- a/src/tools/irm/irm_ipcp_list.c +++ b/src/tools/irm/irm_ipcp_list.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * List IPC Processes * @@ -48,7 +48,8 @@ #define UNICAST "unicast" #define BROADCAST "broadcast" -#define UDP "udp" +#define UDP4 "udp4" +#define UDP6 "udp6" #define ETH_LLC "eth-llc" #define ETH_DIX "eth-dix" #define LOCAL "local" @@ -60,7 +61,7 @@ static void usage(void) " [layer <layer_name>]\n\n" " [type [TYPE]]\n\n" "where TYPE = {" UNICAST " " LOCAL " " - UDP " " ETH_LLC " " ETH_DIX "}\n"); + UDP4 " " UDP6 " " ETH_LLC " " ETH_DIX "}\n"); } static char * str_type(enum ipcp_type type) @@ -74,8 +75,10 @@ static char * str_type(enum ipcp_type type) return ETH_LLC; case IPCP_ETH_DIX: return ETH_DIX; - case IPCP_UDP: - return UDP; + case IPCP_UDP4: + return UDP4; + case IPCP_UDP6: + return UDP6; case IPCP_LOCAL: return LOCAL; default: @@ -113,8 +116,10 @@ int do_list_ipcp(int argc, type = IPCP_UNICAST; else if (strcmp(ipcp_type, BROADCAST) == 0) type = IPCP_BROADCAST; - else if (strcmp(ipcp_type, UDP) == 0) - type = IPCP_UDP; + else if (strcmp(ipcp_type, UDP4) == 0) + type = IPCP_UDP4; + else if (strcmp(ipcp_type, UDP6) == 0) + type = IPCP_UDP6; else if (strcmp(ipcp_type, LOCAL) == 0) type = IPCP_LOCAL; else if (strcmp(ipcp_type, ETH_LLC) == 0) diff --git a/src/tools/irm/irm_name.c b/src/tools/irm/irm_name.c index d60b6c78..830ae305 100644 --- a/src/tools/irm/irm_name.c +++ b/src/tools/irm/irm_name.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * A tool to instruct the IRM daemon * diff --git a/src/tools/irm/irm_name_create.c b/src/tools/irm/irm_name_create.c index a0079cad..40a51193 100644 --- a/src/tools/irm/irm_name_create.c +++ b/src/tools/irm/irm_name_create.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Create IPC Processes * @@ -36,31 +36,82 @@ * OF THE POSSIBILITY OF SUCH DAMAGE. */ +#define _POSIX_C_SOURCE 200809L +#define _XOPEN_SOURCE 500 + +#include <ouroboros/errno.h> #include <ouroboros/irm.h> #include <stdio.h> #include <string.h> +#include <stdlib.h> #include "irm_ops.h" #include "irm_utils.h" #define RR "round-robin" #define SPILL "spillover" +#define SSEC "<security_dir>/server/<name>/sec.conf" +#define SCRT "<security_dir>/server/<name>/crt.pem" +#define SKEY "<security_dir>/server/<name>/key.pem" +#define CSEC "<security_dir>/client/<name>/sec.conf" +#define CCRT "<security_dir>/client/<name>/crt.pem" +#define CKEY "<security_dir>/client/<name>/key.pem" static void usage(void) { printf("Usage: irm name create\n" - " <name>\n" - " lb [LB_POLICY], default: %s\n\n" - "where LB_POLICY in {" RR " " SPILL "}\n", RR); + " <name>. max %d chars.\n" + " [lb LB_POLICY], default: %s\n" + " [ssecpath <path>, default: " SSEC "]\n" + " [scrtpath <path>, default: " SCRT "]\n" + " [skeypath <path>, default: " SKEY "]\n" + " [csecpath <path>, default: " CSEC "]\n" + " [ccrtpath <path>, default: " CCRT "]\n" + " [ckeypath <path>, default: " CKEY "]\n" + "\n" + "where LB_POLICY in {" RR " " SPILL "}\n", + NAME_SIZE, RR); +} + +static int cp_chk_path(char * buf, + const char * path) +{ + char * rp = realpath(path, NULL); + if (rp == NULL) { + printf("Failed to check path %s: %s\n.", + path, strerror(errno)); + goto fail_rp; + } + + if (strlen(rp) > NAME_PATH_SIZE) { + printf("File path too long: %s.\n", rp); + goto fail_len; + } + + strcpy(buf, rp); + free(rp); + + return 0; + + fail_len: + free(rp); + fail_rp: + return -1; } int do_create_name(int argc, char ** argv) { - char * name = NULL; - char * lb_pol = RR; - enum pol_balance pol_lb = LB_RR; + struct name_info info = {}; + char * name = NULL; + char * ssecpath = NULL; + char * scrtpath = NULL; + char * skeypath = NULL; + char * csecpath = NULL; + char * ccrtpath = NULL; + char * ckeypath = NULL; + char * lb_pol = RR; name = *(argv++); --argc; @@ -68,6 +119,18 @@ int do_create_name(int argc, while (argc > 0) { if (matches(*argv, "lb") == 0) { lb_pol = *(argv + 1); + } else if (matches(*argv, "ssecpath") == 0) { + ssecpath = *(argv + 1); + } else if (matches(*argv, "scrtpath") == 0) { + scrtpath = *(argv + 1); + } else if (matches(*argv, "skeypath") == 0) { + skeypath = *(argv + 1); + } else if (matches(*argv, "csecpath") == 0) { + csecpath = *(argv + 1); + } else if (matches(*argv, "ccrtpath") == 0) { + ccrtpath = *(argv + 1); + } else if (matches(*argv, "ckeypath") == 0) { + ckeypath = *(argv + 1); } else { printf("\"%s\" is unknown, try \"irm " "name create\".\n", *argv); @@ -78,19 +141,45 @@ int do_create_name(int argc, argv += 2; } - if (name == NULL) { - usage(); - return -1; + if (name == NULL) + goto fail; + + if (strlen(name) > NAME_SIZE) { + printf("Name too long.\n"); + goto fail; } + strcpy(info.name, name); + + if (ssecpath != NULL && cp_chk_path(info.s.sec, ssecpath) < 0) + goto fail; + + if (scrtpath != NULL && cp_chk_path(info.s.crt, scrtpath) < 0) + goto fail; + + if (skeypath != NULL && cp_chk_path(info.s.key, skeypath) < 0) + goto fail; + + if (csecpath != NULL && cp_chk_path(info.c.sec, csecpath) < 0) + goto fail; + + if (ccrtpath != NULL && cp_chk_path(info.c.crt, ccrtpath) < 0) + goto fail; + + if (ckeypath != NULL && cp_chk_path(info.c.key, ckeypath) < 0) + goto fail; + if (strcmp(lb_pol, RR) == 0) - pol_lb = LB_RR; + info.pol_lb = LB_RR; else if (strcmp(lb_pol, SPILL) == 0) - pol_lb = LB_SPILL; + info.pol_lb = LB_SPILL; else { usage(); return -1; } - return irm_create_name(name, pol_lb); + return irm_create_name(&info); + fail: + usage(); + return -1; } diff --git a/src/tools/irm/irm_name_destroy.c b/src/tools/irm/irm_name_destroy.c index d4bd6c82..d5ed05d5 100644 --- a/src/tools/irm/irm_name_destroy.c +++ b/src/tools/irm/irm_name_destroy.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Create IPC Processes * diff --git a/src/tools/irm/irm_name_list.c b/src/tools/irm/irm_name_list.c index a807008c..37e1f023 100644 --- a/src/tools/irm/irm_name_list.c +++ b/src/tools/irm/irm_name_list.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * List names * diff --git a/src/tools/irm/irm_name_reg.c b/src/tools/irm/irm_name_reg.c index 061ed8be..860f4a70 100644 --- a/src/tools/irm/irm_name_reg.c +++ b/src/tools/irm/irm_name_reg.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Register names with IPCPs * @@ -107,14 +107,23 @@ int do_reg_name(int argc, return -1; } + if (strlen(name) > NAME_SIZE) { + printf("Name too long.\n"); + usage(); + return -1; + } + ipcps_len = irm_list_ipcps(&ipcps); - if (ipcps_len < 0) - return ipcps_len; + if (ipcps_len <= 0) { + printf("Failed to list IPCPs.\n"); + return -1; + } names_len = irm_list_names(&names); if (names_len < 0) { + printf("Failed to list names.\n"); free(ipcps); - return names_len; + return -1; } for (i = 0; i < names_len; ++i) { @@ -124,11 +133,19 @@ int do_reg_name(int argc, } } - if (name_create && irm_create_name(name, LB_SPILL)) { - printf("Error creating name."); - free(ipcps); - free(name); - return -1; + if (name_create) { + struct name_info info = { + .pol_lb = LB_SPILL + }; + + strcpy(info.name, name); + + if (irm_create_name(&info) < 0) { + printf("Error creating name."); + free(ipcps); + free(names); + return -1; + } } for (i = 0; i < ipcps_len; ++i) { @@ -136,6 +153,8 @@ int do_reg_name(int argc, for (j = 0; j < layers_len; j++) { if (wildcard_match(layers[j], ipcps[i].layer) == 0) { if (irm_reg_name(name, ipcps[i].pid)) { + printf("Failed to register with %s", + ipcps[i].layer); free(ipcps); free(names); return -1; @@ -145,6 +164,8 @@ int do_reg_name(int argc, for (j = 0; j < ipcp_len; j++) { if (wildcard_match(ipcp[j], ipcps[i].name) == 0) { if (irm_reg_name(name, ipcps[i].pid)) { + printf("Failed to register with %s", + ipcps[i].name); free(ipcps); free(names); return -1; diff --git a/src/tools/irm/irm_name_unreg.c b/src/tools/irm/irm_name_unreg.c index 6e579f04..abf08548 100644 --- a/src/tools/irm/irm_name_unreg.c +++ b/src/tools/irm/irm_name_unreg.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Unregister names from IPCPs * diff --git a/src/tools/irm/irm_ops.h b/src/tools/irm/irm_ops.h index e04ffc02..195c5cbc 100644 --- a/src/tools/irm/irm_ops.h +++ b/src/tools/irm/irm_ops.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Functions of the IRM tool that are one level deep * diff --git a/src/tools/irm/irm_unbind.c b/src/tools/irm/irm_unbind.c index d6594d01..4e5914a9 100644 --- a/src/tools/irm/irm_unbind.c +++ b/src/tools/irm/irm_unbind.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Unbind names in the processing system * diff --git a/src/tools/irm/irm_unbind_ipcp.c b/src/tools/irm/irm_unbind_ipcp.c index 53a2d16c..23e25057 100644 --- a/src/tools/irm/irm_unbind_ipcp.c +++ b/src/tools/irm/irm_unbind_ipcp.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Unbind name from IPCP Instance * diff --git a/src/tools/irm/irm_unbind_process.c b/src/tools/irm/irm_unbind_process.c index 264ed538..bc7e545c 100644 --- a/src/tools/irm/irm_unbind_process.c +++ b/src/tools/irm/irm_unbind_process.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Unbind process names * diff --git a/src/tools/irm/irm_unbind_program.c b/src/tools/irm/irm_unbind_program.c index 0c751e80..031b9909 100644 --- a/src/tools/irm/irm_unbind_program.c +++ b/src/tools/irm/irm_unbind_program.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Unbind programs * diff --git a/src/tools/irm/irm_utils.c b/src/tools/irm/irm_utils.c index 9694d647..69873097 100644 --- a/src/tools/irm/irm_utils.c +++ b/src/tools/irm/irm_utils.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Handy helper functions for the IRM tool * diff --git a/src/tools/irm/irm_utils.h b/src/tools/irm/irm_utils.h index 27a0b941..c6d4bf18 100644 --- a/src/tools/irm/irm_utils.h +++ b/src/tools/irm/irm_utils.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Handy helper functions for the IRM tool * diff --git a/src/tools/obc/CMakeLists.txt b/src/tools/obc/CMakeLists.txt deleted file mode 100644 index db5e999b..00000000 --- a/src/tools/obc/CMakeLists.txt +++ /dev/null @@ -1,16 +0,0 @@ -include_directories(${CMAKE_CURRENT_SOURCE_DIR}) -include_directories(${CMAKE_CURRENT_BINARY_DIR}) - -include_directories(${CMAKE_SOURCE_DIR}/include) -include_directories(${CMAKE_BINARY_DIR}/include) - -set(SOURCE_FILES - # Add source files here - obc.c - ) - -add_executable(obc ${SOURCE_FILES}) - -target_link_libraries(obc LINK_PUBLIC ouroboros-dev) - -install(TARGETS obc RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR}) diff --git a/src/tools/obc/obc.c b/src/tools/obc/obc.c index 462cbea9..5b8470f0 100644 --- a/src/tools/obc/obc.c +++ b/src/tools/obc/obc.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * A simple broadcast application * @@ -63,7 +63,7 @@ static int reader_main(const char * dst) printf("Starting a reader.\n"); - fd = flow_join(dst, NULL, NULL); + fd = flow_join(dst, NULL); if (fd < 0) { printf("Failed to join broadcast.\n"); return -1; @@ -91,7 +91,7 @@ static int writer_main(const char * dst, int fd = 0; size_t len = strlen(message) + 1; - fd = flow_join(dst, NULL, NULL); + fd = flow_join(dst, NULL); if (fd < 0) { printf("Failed to join broadcast.\n"); return -1; diff --git a/src/tools/ocbr/CMakeLists.txt b/src/tools/ocbr/CMakeLists.txt deleted file mode 100644 index f7ba66cd..00000000 --- a/src/tools/ocbr/CMakeLists.txt +++ /dev/null @@ -1,21 +0,0 @@ -include_directories(${CMAKE_CURRENT_SOURCE_DIR}) -include_directories(${CMAKE_CURRENT_BINARY_DIR}) - -include_directories(${CMAKE_SOURCE_DIR}/include) -include_directories(${CMAKE_BINARY_DIR}/include) - -get_filename_component(CURRENT_SOURCE_PARENT_DIR - ${CMAKE_CURRENT_SOURCE_DIR} DIRECTORY) - -include_directories(${CURRENT_SOURCE_PARENT_DIR}) - -set(SOURCE_FILES - # Add source files here - ocbr.c - ) - -add_executable(ocbr ${SOURCE_FILES}) - -target_link_libraries(ocbr LINK_PUBLIC ouroboros-dev) - -install(TARGETS ocbr RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR}) diff --git a/src/tools/ocbr/ocbr.c b/src/tools/ocbr/ocbr.c index 775bcaac..c92ba0e0 100644 --- a/src/tools/ocbr/ocbr.c +++ b/src/tools/ocbr/ocbr.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * CBR traffic generator * diff --git a/src/tools/ocbr/ocbr_client.c b/src/tools/ocbr/ocbr_client.c index ba7b41f4..36c07d43 100644 --- a/src/tools/ocbr/ocbr_client.c +++ b/src/tools/ocbr/ocbr_client.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * A simple CBR generator * @@ -37,8 +37,11 @@ */ #include <ouroboros/dev.h> +#include <ouroboros/qos.h> #include <signal.h> +#include <stdlib.h> +#include <string.h> volatile bool stop; @@ -86,6 +89,11 @@ int client_main(char * server, struct timespec end; struct timespec intv = {(gap / BILLION), gap % BILLION}; int ms; + const char * qenv; + qosspec_t qs; + qosspec_t * qsp; + + qsp = NULL; stop = false; @@ -98,16 +106,38 @@ int client_main(char * server, sigaction(SIGHUP, &sig_act, NULL) || sigaction(SIGPIPE, &sig_act, NULL)) { printf("Failed to install sighandler.\n"); - return -1; + return 2; } printf("Client started, duration %d, rate %lu b/s, size %d B.\n", duration, rate, size); - fd = flow_alloc(server, NULL, NULL); + qenv = getenv("OCBR_QOS"); + if (qenv != NULL) { + if (strcmp(qenv, "raw") == 0) + qs = qos_raw; + else if (strcmp(qenv, "safe") == 0) + qs = qos_raw_safe; + else if (strcmp(qenv, "rt") == 0) + qs = qos_rt; + else if (strcmp(qenv, "rt_safe") == 0) + qs = qos_rt_safe; + else if (strcmp(qenv, "msg") == 0) + qs = qos_msg; + else if (strcmp(qenv, "stream") == 0) + qs = qos_stream; + else { + fprintf(stderr, + "Unknown OCBR_QOS='%s', using raw.\n", qenv); + qs = qos_raw; + } + qsp = &qs; + printf("OCBR_QOS=%s\n", qenv); + } + fd = flow_alloc(server, qsp, NULL); if (fd < 0) { printf("Failed to allocate flow.\n"); - return -1; + return 2; } clock_gettime(CLOCK_REALTIME, &start); @@ -129,7 +159,7 @@ int client_main(char * server, ++seqnr; - if (ts_diff_us(&start, &end) / MILLION >= duration) + if (ts_diff_us(&end, &start) / MILLION >= duration) stop = true; } } else { /* flood */ @@ -142,7 +172,7 @@ int client_main(char * server, ++seqnr; - if (ts_diff_us(&start, &end) / MILLION + if (ts_diff_us(&end, &start) / MILLION >= (long) duration) stop = true; } @@ -151,7 +181,7 @@ int client_main(char * server, clock_gettime(CLOCK_REALTIME, &end); - ms = ts_diff_ms(&start, &end); + ms = ts_diff_ms(&end, &start); printf("sent statistics: " "%9ld packets, %12ld bytes in %9d ms, %4.4f Mb/s\n", diff --git a/src/tools/ocbr/ocbr_server.c b/src/tools/ocbr/ocbr_server.c index a4bbadd4..c98b33e9 100644 --- a/src/tools/ocbr/ocbr_server.c +++ b/src/tools/ocbr/ocbr_server.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * A simple CBR generator * @@ -114,14 +114,14 @@ static void handle_flow(int fd) bytes_read += count; } - if (ts_diff_us(&alive, &now) + if (ts_diff_us(&now, &alive) > server_settings.timeout * MILLION) { printf("Test on flow %d timed out\n", fd); stop = true; } - if (stop || ts_diff_ms(&now, &iv_end) < 0) { - long us = ts_diff_us(&iv_start, &now); + if (stop || ts_diff_ms(&now, &iv_end) > 0) { + long us = ts_diff_us(&now, &iv_start); printf("Flow %4d: %9ld packets (%12ld bytes) in %9ld ms" " => %9.4f pps, %9.4f Mbps\n", fd, diff --git a/src/tools/oecho/CMakeLists.txt b/src/tools/oecho/CMakeLists.txt deleted file mode 100644 index 50a66138..00000000 --- a/src/tools/oecho/CMakeLists.txt +++ /dev/null @@ -1,16 +0,0 @@ -include_directories(${CMAKE_CURRENT_SOURCE_DIR}) -include_directories(${CMAKE_CURRENT_BINARY_DIR}) - -include_directories(${CMAKE_SOURCE_DIR}/include) -include_directories(${CMAKE_BINARY_DIR}/include) - -set(SOURCE_FILES - # Add source files here - oecho.c - ) - -add_executable(oecho ${SOURCE_FILES}) - -target_link_libraries(oecho LINK_PUBLIC ouroboros-dev) - -install(TARGETS oecho RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR}) diff --git a/src/tools/oecho/oecho.c b/src/tools/oecho/oecho.c index d5d03027..ef0a168f 100644 --- a/src/tools/oecho/oecho.c +++ b/src/tools/oecho/oecho.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * A simple echo application * @@ -101,20 +101,20 @@ static int client_main(void) fd = flow_alloc("oecho", NULL, NULL); if (fd < 0) { printf("Failed to allocate flow.\n"); - return -1; + return 2; } if (flow_write(fd, message, strlen(message) + 1) < 0) { printf("Failed to write packet.\n"); flow_dealloc(fd); - return -1; + return 1; } count = flow_read(fd, buf, BUF_SIZE); if (count < 0) { printf("Failed to read packet.\n"); flow_dealloc(fd); - return -1; + return 1; } printf("Server replied with %.*s\n", (int) count, buf); @@ -126,7 +126,7 @@ static int client_main(void) int main(int argc, char ** argv) { - int ret = -1; + int ret = 0; bool server = false; argc--; diff --git a/src/tools/oftp/oftp.c b/src/tools/oftp/oftp.c new file mode 100644 index 00000000..1ae99403 --- /dev/null +++ b/src/tools/oftp/oftp.c @@ -0,0 +1,441 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * A minimal file-transfer tool over an FRCT stream flow + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * + * 2. Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials provided + * with the distribution. + * + * 3. Neither the name of the copyright holder nor the names of its + * contributors may be used to endorse or promote products derived + * from this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS + * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE + * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, + * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR + * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, + * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) + * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED + * OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#define _POSIX_C_SOURCE 200809L + +#include <ouroboros/crc64.h> +#include <ouroboros/dev.h> +#include <ouroboros/errno.h> +#include <ouroboros/fccntl.h> +#include <ouroboros/qos.h> + +#include <fcntl.h> +#include <inttypes.h> +#include <signal.h> +#include <stdbool.h> +#include <stdio.h> +#include <stdlib.h> +#include <string.h> +#include <time.h> +#include <unistd.h> + +#define BUF_SIZE 16384 + +static volatile sig_atomic_t stop = 0; + +static void apply_rto_min_env(int fd) +{ + const char * env; + long v; + + env = getenv("OFTP_FRCT_RTO_MIN"); + if (env == NULL) + return; + v = strtol(env, NULL, 10); + if (v <= 0) + return; + if (fccntl(fd, FRCTSRTOMIN, (time_t) v) < 0) + fprintf(stderr, + "oftp: failed to set RTO_MIN=%ld ns\n", v); +} + +static void apply_stream_ring_sz_env(int fd) +{ + const char * env; + long v; + + env = getenv("OFTP_FRCT_STREAM_RING_SZ"); + if (env == NULL) + return; + v = strtol(env, NULL, 10); + if (v <= 0) + return; + if (fccntl(fd, FRCTSRRINGSZ, (size_t) v) < 0) + fprintf(stderr, + "oftp: failed to set STREAM_RING_SZ=%ld\n", v); +} + +static void on_signal(int signo) +{ + (void) signo; + stop = 1; +} + +static void usage(void) +{ + printf("Usage: oftp [OPTION]...\n" + "Stream-mode file transfer over an Ouroboros flow.\n\n" + " -l, --listen Run as the receiver (server)\n" + " -n, --name NAME Destination service name (client)\n" + " -i, --in FILE Read input from FILE (default stdin)\n" + " -o, --out FILE Write output to FILE (default stdout)\n" + " -N, --bytes SIZE Stop after SIZE bytes " + "(K/M/G suffix; client only)\n" + " --help Display this help text and exit\n"); +} + +static int parse_size(const char * s, size_t * out) +{ + char * end; + unsigned long v; + size_t mul; + + v = strtoul(s, &end, 0); + if (end == s) + return -1; + + mul = 1; + if (*end == 'k' || *end == 'K') + mul = 1024UL; + else if (*end == 'm' || *end == 'M') + mul = 1024UL * 1024UL; + else if (*end == 'g' || *end == 'G') + mul = 1024UL * 1024UL * 1024UL; + else if (*end != '\0') + return -1; + + *out = (size_t) v * mul; + return 0; +} + +static void report_xfer(const char * tag, + size_t total, + uint64_t crc, + const struct timespec * t0, + const struct timespec * t1) +{ + double elapsed_s; + double mib_per_s; + + elapsed_s = (t1->tv_sec - t0->tv_sec) + + (t1->tv_nsec - t0->tv_nsec) / 1e9; + if (elapsed_s <= 0.0) + elapsed_s = 1e-9; + + mib_per_s = ((double) total / (1024.0 * 1024.0)) / elapsed_s; + + fprintf(stderr, + "oftp: %s %zu bytes in %.3f s (%.2f MiB/s) " + "crc64=%016" PRIx64 "\n", + tag, total, elapsed_s, mib_per_s, crc); +} + +static int xfer_to_flow(int fd, FILE * in, size_t max_bytes) +{ + char buf[BUF_SIZE]; + size_t n; + size_t total; + size_t want; + size_t off; + ssize_t w; + uint64_t crc; + struct timespec t0; + struct timespec t1; + + total = 0; + crc = 0; + + clock_gettime(CLOCK_MONOTONIC, &t0); + + while (!stop) { + want = sizeof(buf); + if (max_bytes > 0 && max_bytes - total < want) + want = max_bytes - total; + if (want == 0) + break; + + n = fread(buf, 1, want, in); + if (n == 0) + break; + + crc64_nvme(&crc, buf, n); + + off = 0; + while (off < n) { + w = flow_write(fd, buf + off, n - off); + if (w < 0) { + fprintf(stderr, + "flow_write failed: %zd\n", w); + return 1; + } + off += (size_t) w; + total += (size_t) w; + } + } + + clock_gettime(CLOCK_MONOTONIC, &t1); + + if (ferror(in)) { + fprintf(stderr, "Input read error.\n"); + return 1; + } + + report_xfer("sent", total, crc, &t0, &t1); + return 0; +} + +static int xfer_from_flow(int fd, FILE * out) +{ + char buf[BUF_SIZE]; + size_t total; + ssize_t n; + uint64_t crc; + struct timespec timeout; + struct timespec t0; + struct timespec t1; + bool started; + + total = 0; + crc = 0; + started = false; + timeout.tv_sec = 1; + timeout.tv_nsec = 0; + + /* Short timeout so SIGTERM/SIGINT 'stop' is observed promptly. */ + fccntl(fd, FLOWSRCVTIMEO, &timeout); + + while (!stop) { + n = flow_read(fd, buf, sizeof(buf)); + if (n == 0) { + /* Clean EOF: peer sent EOS and we drained it. */ + clock_gettime(CLOCK_MONOTONIC, &t1); + fflush(out); + if (!started) + t0 = t1; + report_xfer("received", total, crc, &t0, &t1); + return 0; + } + if (n == -ETIMEDOUT) + continue; + if (n < 0) { + /* Peer aborted before EOS: partial transfer. */ + if (n == -EFLOWDOWN || n == -EFLOWPEER) { + fprintf(stderr, + "oftp: peer aborted at %zu B\n", + total); + return 2; + } + fprintf(stderr, + "flow_read failed: %zd\n", n); + return 1; + } + if (!started) { + clock_gettime(CLOCK_MONOTONIC, &t0); + started = true; + } + crc64_nvme(&crc, buf, (size_t) n); + if (fwrite(buf, 1, (size_t) n, out) != (size_t) n) { + fprintf(stderr, "Output write error.\n"); + return 1; + } + total += (size_t) n; + } + + /* Receiver was signalled (SIGINT/SIGTERM) before EOF. */ + fflush(out); + fprintf(stderr, "oftp: interrupted at %zu B\n", total); + return 2; +} + +static int server_main(const char * outpath) +{ + FILE * out = stdout; + int fd; + int ofd; + int rc; + qosspec_t qs; + + if (outpath != NULL) { + ofd = open(outpath, + O_WRONLY | O_CREAT | O_EXCL | O_NOFOLLOW, + 0600); + if (ofd < 0) { + perror("open"); + return 1; + } + out = fdopen(ofd, "wb"); + if (out == NULL) { + perror("fdopen"); + close(ofd); + unlink(outpath); + return 1; + } + } + + fprintf(stderr, "oftp: listening...\n"); + + fd = flow_accept(&qs, NULL); + if (fd < 0) { + fprintf(stderr, "flow_accept failed: %d\n", fd); + if (out != stdout) + fclose(out); + return 1; + } + + if (qs.service != SVC_STREAM) { + fprintf(stderr, + "oftp: rejecting non-stream flow (service=%u)\n", + qs.service); + flow_dealloc(fd); + if (out != stdout) { + fclose(out); + unlink(outpath); + } + return 1; + } + + apply_rto_min_env(fd); + apply_stream_ring_sz_env(fd); + + rc = xfer_from_flow(fd, out); + + flow_dealloc(fd); + + if (out != stdout) { + fclose(out); + /* Drop the half-written file on abort/interrupt. */ + if (rc != 0) + unlink(outpath); + } + + return rc; +} + +static int client_main(const char * name, + const char * inpath, + size_t max_bytes) +{ + FILE * in; + int fd; + int rc; + qosspec_t qs; + + in = stdin; + qs = qos_stream; + + if (inpath != NULL) { + in = fopen(inpath, "rb"); + if (in == NULL) { + perror("fopen"); + return 1; + } + } + + fd = flow_alloc(name, &qs, NULL); + if (fd < 0) { + fprintf(stderr, "flow_alloc failed: %d\n", fd); + if (in != stdin) + fclose(in); + return 2; + } + + apply_rto_min_env(fd); + apply_stream_ring_sz_env(fd); + + rc = xfer_to_flow(fd, in, max_bytes); + + flow_dealloc(fd); + + if (in != stdin) + fclose(in); + + return rc; +} + +int main(int argc, char ** argv) +{ + bool server; + const char * name; + const char * inpath; + const char * outpath; + size_t max_bytes; + struct sigaction sa; + + server = false; + name = NULL; + inpath = NULL; + outpath = NULL; + max_bytes = 0; + + memset(&sa, 0, sizeof(sa)); + sa.sa_handler = on_signal; + sigaction(SIGINT, &sa, NULL); + sigaction(SIGTERM, &sa, NULL); + signal(SIGPIPE, SIG_IGN); + + argc--; argv++; + while (argc > 0) { + if (strcmp(*argv, "-l") == 0 || + strcmp(*argv, "--listen") == 0) { + server = true; + } else if ((strcmp(*argv, "-n") == 0 || + strcmp(*argv, "--name") == 0) && argc > 1) { + name = *(++argv); argc--; + } else if ((strcmp(*argv, "-i") == 0 || + strcmp(*argv, "--in") == 0) && argc > 1) { + inpath = *(++argv); argc--; + } else if ((strcmp(*argv, "-o") == 0 || + strcmp(*argv, "--out") == 0) && argc > 1) { + outpath = *(++argv); argc--; + } else if ((strcmp(*argv, "-N") == 0 || + strcmp(*argv, "--bytes") == 0) && argc > 1) { + if (parse_size(*(++argv), &max_bytes) < 0) { + fprintf(stderr, + "oftp: bad size '%s'\n", *argv); + return 1; + } + argc--; + } else if (strcmp(*argv, "--help") == 0) { + usage(); + return 0; + } else { + usage(); + return 1; + } + argc--; argv++; + } + + if (server) + return server_main(outpath); + + if (name == NULL) { + usage(); + return 1; + } + + return client_main(name, inpath, max_bytes); +} diff --git a/src/tools/operf/CMakeLists.txt b/src/tools/operf/CMakeLists.txt deleted file mode 100644 index b6faf04e..00000000 --- a/src/tools/operf/CMakeLists.txt +++ /dev/null @@ -1,26 +0,0 @@ -include_directories(${CMAKE_CURRENT_SOURCE_DIR}) -include_directories(${CMAKE_CURRENT_BINARY_DIR}) - -include_directories(${CMAKE_SOURCE_DIR}/include) -include_directories(${CMAKE_BINARY_DIR}/include) - -get_filename_component(CURRENT_SOURCE_PARENT_DIR - ${CMAKE_CURRENT_SOURCE_DIR} DIRECTORY) - -include_directories(${CURRENT_SOURCE_PARENT_DIR}) - -find_library(LIBM_LIBRARIES m) -if(NOT LIBM_LIBRARIES) - message(FATAL_ERROR "libm not found") -endif() - -set(SOURCE_FILES - # Add source files here - operf.c - ) - -add_executable(operf ${SOURCE_FILES}) - -target_link_libraries(operf LINK_PUBLIC ${LIBM_LIBRARIES} ouroboros-dev) - -install(TARGETS operf RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR}) diff --git a/src/tools/operf/operf.c b/src/tools/operf/operf.c index 10896bd5..0198e871 100644 --- a/src/tools/operf/operf.c +++ b/src/tools/operf/operf.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Ouroboros perf application * @@ -54,7 +54,6 @@ #include <stdlib.h> #include <sys/time.h> #include <arpa/inet.h> -#include <math.h> #include <errno.h> #include <float.h> @@ -249,5 +248,5 @@ int main(int argc, char ** argv) if (ret < 0) exit(EXIT_FAILURE); - exit(EXIT_SUCCESS); + exit(ret); } diff --git a/src/tools/operf/operf_client.c b/src/tools/operf/operf_client.c index 63f98299..e478aeff 100644 --- a/src/tools/operf/operf_client.c +++ b/src/tools/operf/operf_client.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Ouroboros ping application * @@ -133,7 +133,7 @@ void * writer(void * o) clock_gettime(CLOCK_REALTIME, &start); clock_gettime(CLOCK_REALTIME, &now); - while (!stop && ts_diff_ms(&start, &now) < client.duration) { + while (!stop && ts_diff_ms(&now, &start) > client.duration) { if (!client.flood) { clock_gettime(CLOCK_REALTIME, &now); ts_add(&now, &intv, &end); @@ -185,7 +185,7 @@ int client_main(void) sigaction(SIGHUP, &sig_act, NULL) || sigaction(SIGPIPE, &sig_act, NULL)) { printf("Failed to install sighandler.\n"); - return -1; + return 2; } client.sent = 0; @@ -196,7 +196,7 @@ int client_main(void) fd = flow_alloc(client.server_name, NULL, NULL); if (fd < 0) { printf("Failed to allocate flow.\n"); - return -1; + return 2; } if (client.conf.test_type == TEST_TYPE_BI) @@ -207,7 +207,7 @@ int client_main(void) if (flow_write(fd, &client.conf, sizeof(client.conf)) < 0) { printf("Failed to send configuration.\n"); flow_dealloc(fd); - return -1; + return 1; } sleep(1); @@ -230,10 +230,10 @@ int client_main(void) printf("%ld received, ", client.rcvd); printf("%ld%% packet loss, ", client.sent == 0 ? 0 : 100 - ((100 * client.rcvd) / client.sent)); - printf("time: %.3f ms, ", ts_diff_us(&tic, &toc) / 1000.0); + printf("time: %.3f ms, ", ts_diff_us(&toc, &tic) / 1000.0); printf("bandwidth: %.3lf Mb/s.\n", (client.rcvd * client.size * 8) - / (double) ts_diff_us(&tic, &toc)); + / (double) ts_diff_us(&toc, &tic)); } flow_dealloc(fd); diff --git a/src/tools/operf/operf_server.c b/src/tools/operf/operf_server.c index d11f3486..00f780ba 100644 --- a/src/tools/operf/operf_server.c +++ b/src/tools/operf/operf_server.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Ouroboros perf application * @@ -66,7 +66,7 @@ void * cleaner_thread(void * o) pthread_mutex_lock(&server.lock); for (i = 0; i < OPERF_MAX_FLOWS; ++i) if (fset_has(server.flows, i) && - ts_diff_ms(&server.times[i], &now) + ts_diff_ms(&now, &server.times[i]) > server.timeout) { printf("Flow %d timed out.\n", i); fset_del(server.flows, i); diff --git a/src/tools/oping/CMakeLists.txt b/src/tools/oping/CMakeLists.txt deleted file mode 100644 index 31a4f961..00000000 --- a/src/tools/oping/CMakeLists.txt +++ /dev/null @@ -1,28 +0,0 @@ -include_directories(${CMAKE_CURRENT_SOURCE_DIR}) -include_directories(${CMAKE_CURRENT_BINARY_DIR}) - -include_directories(${CMAKE_SOURCE_DIR}/include) -include_directories(${CMAKE_BINARY_DIR}/include) - -get_filename_component(CURRENT_SOURCE_PARENT_DIR - ${CMAKE_CURRENT_SOURCE_DIR} DIRECTORY) - -include_directories(${CURRENT_SOURCE_PARENT_DIR}) - -find_library(LIBM_LIBRARIES m) -if(NOT LIBM_LIBRARIES) - message(FATAL_ERROR "libm not found") -endif() - -mark_as_advanced(LIBM_LIBRARIES) - -set(SOURCE_FILES - # Add source files here - oping.c - ) - -add_executable(oping ${SOURCE_FILES}) - -target_link_libraries(oping LINK_PUBLIC ${LIBM_LIBRARIES} ouroboros-dev) - -install(TARGETS oping RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR}) diff --git a/src/tools/oping/oping.c b/src/tools/oping/oping.c index ed3529e5..10e1e23c 100644 --- a/src/tools/oping/oping.c +++ b/src/tools/oping/oping.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Ouroboros ping application * @@ -60,7 +60,7 @@ #include <errno.h> #include <float.h> -#define OPING_BUF_SIZE 1500 +#define OPING_BUF_SIZE 16384 #define ECHO_REQUEST 0 #define ECHO_REPLY 1 #define OPING_MAX_FLOWS 256 @@ -72,13 +72,18 @@ "and reports the Round Trip Time (RTT)\n" \ "\n" \ " -l, --listen Run in server mode\n" \ +" --poll Server uses polling (lower latency)\n" \ +" --busy Server uses busy-poll (single flow)\n" \ "\n" \ " -c, --count Number of packets\n" \ " -d, --duration Duration of the test (default 1s)\n" \ +" -f, --flood Send back-to-back without waiting\n" \ +" -F, --flood-busy Flood with busy-polling (lower latency)\n" \ " -i, --interval Interval (default 1000ms)\n" \ " -n, --server-name Name of the oping server\n" \ -" -q, --qos QoS (raw, raw_crypt, best, video, voice, data)\n" \ +" -q, --qos QoS (raw, safe, rt, rt-safe, msg)\n" \ " -s, --size Payload size (B, default 64)\n" \ +" -W, --timeout Per-packet recv timeout, ms (default 2000)\n" \ " -Q, --quiet Only print final statistics\n" \ " -D, --timeofday Print time of day before each line\n" \ "\n" \ @@ -89,7 +94,11 @@ struct { int interval; uint32_t count; int size; + int timeout; /* per-packet recv timeout, ms */ bool timestamp; + bool flood; + bool flood_busy; + long duration; qosspec_t qs; /* stats */ @@ -114,6 +123,8 @@ struct { pthread_mutex_t lock; bool quiet; + bool poll; + bool busy; pthread_t cleaner_pt; pthread_t accept_pt; @@ -167,14 +178,20 @@ int main(int argc, argc--; argv++; - client.s_apn = NULL; - client.interval = 1000; - client.size = 64; - client.count = INT_MAX; - client.timestamp = false; - client.qs = qos_raw; - client.quiet = false; - server.quiet = false; + client.s_apn = NULL; + client.interval = 1000; + client.size = 64; + client.count = INT_MAX; + client.timeout = 2000; + client.timestamp = false; + client.flood = false; + client.flood_busy = false; + client.duration = 0; + client.qs = qos_raw; + client.quiet = false; + server.quiet = false; + server.poll = false; + server.busy = false; while (argc > 0) { if ((strcmp(*argv, "-i") == 0 || @@ -204,6 +221,12 @@ int main(int argc, argc > 1) { client.size = strtol(*(++argv), &rem, 10); --argc; + } else if ((strcmp(*argv, "-W") == 0 || + strcmp(*argv, "--timeout") == 0) && + argc > 1) { + client.timeout = strtol(*(++argv), &rem, 10); + client.timeout *= time_mul(rem); + --argc; } else if ((strcmp(*argv, "-q") == 0 || strcmp(*argv, "--qos") == 0) && argc > 1) { @@ -212,6 +235,12 @@ int main(int argc, } else if (strcmp(*argv, "-l") == 0 || strcmp(*argv, "--listen") == 0) { serv = true; + } else if (strcmp(*argv, "-f") == 0 || + strcmp(*argv, "--flood") == 0) { + client.flood = true; + } else if (strcmp(*argv, "-F") == 0 || + strcmp(*argv, "--flood-busy") == 0) { + client.flood_busy = true; } else if (strcmp(*argv, "-D") == 0 || strcmp(*argv, "--timeofday") == 0) { client.timestamp = true; @@ -219,6 +248,10 @@ int main(int argc, strcmp(*argv, "--quiet") == 0) { client.quiet = true; server.quiet = true; + } else if (strcmp(*argv, "--poll") == 0) { + server.poll = true; + } else if (strcmp(*argv, "--busy") == 0) { + server.busy = true; } else { goto fail; } @@ -227,25 +260,25 @@ int main(int argc, } if (duration > 0) { - if (client.interval == 0) + if (client.flood || client.flood_busy) + client.duration = duration; + else if (client.interval == 0) client.count = duration * 10; else client.count = duration / client.interval; } if (qos != NULL) { - if (strcmp(qos, "best") == 0) - client.qs = qos_best_effort; - else if (strcmp(qos, "raw") == 0) + if (strcmp(qos, "raw") == 0) client.qs = qos_raw; - else if (strcmp(qos, "video") == 0) - client.qs = qos_video; - else if (strcmp(qos, "voice") == 0) - client.qs = qos_voice; - else if (strcmp(qos, "data") == 0) - client.qs = qos_data; - else if (strcmp(qos, "raw_crypt") == 0) - client.qs = qos_raw_crypt; + else if (strcmp(qos, "safe") == 0) + client.qs = qos_raw_safe; + else if (strcmp(qos, "rt") == 0) + client.qs = qos_rt; + else if (strcmp(qos, "rt-safe") == 0) + client.qs = qos_rt_safe; + else if (strcmp(qos, "msg") == 0) + client.qs = qos_msg; else printf("Unknown QoS cube, defaulting to raw.\n"); } @@ -278,7 +311,7 @@ int main(int argc, if (ret < 0) exit(EXIT_FAILURE); - exit(EXIT_SUCCESS); + exit(ret); fail: usage(); diff --git a/src/tools/oping/oping_client.c b/src/tools/oping/oping_client.c index 7b03c83d..4b01315d 100644 --- a/src/tools/oping/oping_client.c +++ b/src/tools/oping/oping_client.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Ouroboros ping application * @@ -47,15 +47,50 @@ void shutdown_client(int signo, siginfo_t * info, void * c) case SIGINT: case SIGTERM: case SIGHUP: + case SIGALRM: stop = true; default: return; } } +static void update_rtt_stats(double ms) +{ + double d; + + if (ms < client.rtt_min) + client.rtt_min = ms; + if (ms > client.rtt_max) + client.rtt_max = ms; + + d = (ms - client.rtt_avg); + client.rtt_avg += d / client.rcvd; + client.rtt_m2 += d * (ms - client.rtt_avg); +} + +static double rtt_val(double ms) +{ + return ms < 0.1 ? ms * 1000 : ms; +} + +static const char * rtt_unit(double ms) +{ + return ms < 0.1 ? "µs" : "ms"; +} + +static void print_rtt(int len, int seq, + double ms, const char * suf) +{ + printf("%d bytes from %s: seq=%d " + "time=%.3f %s%s\n", + len, client.s_apn, seq, + rtt_val(ms), rtt_unit(ms), + suf != NULL ? suf : ""); +} + void * reader(void * o) { - struct timespec timeout = {client.interval / 1000 + 2, 0}; + struct timespec timeout; struct timespec now = {0, 0}; struct timespec sent; @@ -64,9 +99,11 @@ void * reader(void * o) int fd = *((int *) o); int msg_len = 0; double ms = 0; - double d = 0; uint32_t exp_id = 0; + timeout.tv_sec = client.timeout / 1000; + timeout.tv_nsec = (client.timeout % 1000) * MILLION; + fccntl(fd, FLOWSRCVTIMEO, &timeout); while (!stop && client.rcvd != client.count) { @@ -100,7 +137,7 @@ void * reader(void * o) sent.tv_sec = msg->tv_sec; sent.tv_nsec = msg->tv_nsec; - ms = ts_diff_us(&sent, &now) / 1000.0; + ms = ts_diff_us(&now, &sent) / 1000.0; if (id < exp_id) ++client.ooo; @@ -114,22 +151,12 @@ void * reader(void * o) (size_t) rtc.tv_nsec / 1000); } - printf("%d bytes from %s: seq=%d time=%.3f ms%s\n", - msg_len, - client.s_apn, - ntohl(msg->id), - ms, - id < exp_id ? " [out-of-order]" : ""); + print_rtt(msg_len, ntohl(msg->id), ms, + id < exp_id ? + " [out-of-order]" : NULL); } - if (ms < client.rtt_min) - client.rtt_min = ms; - if (ms > client.rtt_max) - client.rtt_max = ms; - - d = (ms - client.rtt_avg); - client.rtt_avg += d / client.rcvd; - client.rtt_m2 += d * (ms - client.rtt_avg); + update_rtt_stats(ms); if (id >= exp_id) exp_id = id + 1; @@ -204,13 +231,167 @@ static void client_fini(void) return; } +static void print_stats(struct timespec * tic, + struct timespec * toc) +{ + printf("\n"); + printf("--- %s ping statistics ---\n", client.s_apn); + printf("%d packets transmitted, ", client.sent); + printf("%d received, ", client.rcvd); + printf("%zd out-of-order, ", client.ooo); + printf("%.0lf%% packet loss, ", client.sent == 0 ? 0 : + ceil(100 - (100 * (client.rcvd / (float) client.sent)))); + printf("time: %.3f ms\n", ts_diff_us(toc, tic) / 1000.0); + + if (client.rcvd > 0) { + double a = client.rtt_avg; + double f = a < 0.1 ? 1000 : 1; + printf("rtt min/avg/max/mdev = %.3f/%.3f/%.3f/", + client.rtt_min * f, client.rtt_avg * f, + client.rtt_max * f); + if (client.rcvd > 1) + printf("%.3f %s\n", + sqrt(client.rtt_m2 / + (client.rcvd - 1)) * f, + rtt_unit(a)); + else + printf("NaN %s\n", rtt_unit(a)); + } +} + +static int flood_busy_ping(int fd) +{ + char buf[OPING_BUF_SIZE]; + struct oping_msg * msg = (struct oping_msg *) buf; + struct timespec sent; + struct timespec rcvd; + double ms; + int n; + + memset(buf, 0, client.size); + + fccntl(fd, FLOWSFLAGS, + FLOWFRDWR | FLOWFRNOPART | FLOWFRNOBLOCK); + + if (!client.quiet) + printf("Pinging %s with %d bytes" + " of data (%u packets," + " busy-poll):\n\n", + client.s_apn, client.size, + client.count); + + while (!stop && client.sent < client.count) { + clock_gettime(CLOCK_MONOTONIC, &sent); + + msg->type = htonl(ECHO_REQUEST); + msg->id = htonl(client.sent); + msg->tv_sec = sent.tv_sec; + msg->tv_nsec = sent.tv_nsec; + + if (flow_write(fd, buf, client.size) < 0) { + printf("Failed to send packet.\n"); + break; + } + + ++client.sent; + + do { + n = flow_read(fd, buf, OPING_BUF_SIZE); + } while (n == -EAGAIN && !stop); + + if (n < 0) + break; + + clock_gettime(CLOCK_MONOTONIC, &rcvd); + + if (ntohl(msg->type) != ECHO_REPLY) + continue; + + ++client.rcvd; + + sent.tv_sec = msg->tv_sec; + sent.tv_nsec = msg->tv_nsec; + ms = ts_diff_us(&rcvd, &sent) / 1000.0; + + update_rtt_stats(ms); + + if (!client.quiet) + print_rtt(client.size, ntohl(msg->id), ms, NULL); + } + + return 0; +} + +static int flood_ping(int fd) +{ + char buf[OPING_BUF_SIZE]; + struct oping_msg * msg = (struct oping_msg *) buf; + struct timespec sent; + struct timespec rcvd; + double ms; + + memset(buf, 0, client.size); + + if (!client.quiet) + printf("Pinging %s with %d bytes of data (%u packets):\n\n", + client.s_apn, client.size, client.count); + + while (!stop && client.sent < client.count) { + clock_gettime(CLOCK_MONOTONIC, &sent); + + msg->type = htonl(ECHO_REQUEST); + msg->id = htonl(client.sent); + msg->tv_sec = sent.tv_sec; + msg->tv_nsec = sent.tv_nsec; + + if (flow_write(fd, buf, client.size) < 0) { + printf("Failed to send packet.\n"); + break; + } + + ++client.sent; + + if (flow_read(fd, buf, OPING_BUF_SIZE) < 0) { + printf("Failed to read packet.\n"); + break; + } + + clock_gettime(CLOCK_MONOTONIC, &rcvd); + + if (ntohl(msg->type) != ECHO_REPLY) + continue; + + ++client.rcvd; + + sent.tv_sec = msg->tv_sec; + sent.tv_nsec = msg->tv_nsec; + ms = ts_diff_us(&rcvd, &sent) / 1000.0; + + update_rtt_stats(ms); + + if (!client.quiet) + print_rtt(client.size, ntohl(msg->id), ms, NULL); + } + + return 0; +} + +static int threaded_ping(int fd) +{ + pthread_create(&client.reader_pt, NULL, reader, &fd); + pthread_create(&client.writer_pt, NULL, writer, &fd); + + pthread_join(client.writer_pt, NULL); + pthread_join(client.reader_pt, NULL); + + return 0; +} + static int client_main(void) { struct sigaction sig_act; - struct timespec tic; struct timespec toc; - int fd; memset(&sig_act, 0, sizeof sig_act); @@ -220,59 +401,49 @@ static int client_main(void) if (sigaction(SIGINT, &sig_act, NULL) || sigaction(SIGTERM, &sig_act, NULL) || sigaction(SIGHUP, &sig_act, NULL) || - sigaction(SIGPIPE, &sig_act, NULL)) { + sigaction(SIGPIPE, &sig_act, NULL) || + sigaction(SIGALRM, &sig_act, NULL)) { printf("Failed to install sighandler.\n"); - return -1; + return 2; } if (client_init()) { printf("Failed to initialize client.\n"); - return -1; + return 2; } fd = flow_alloc(client.s_apn, &client.qs, NULL); if (fd < 0) { printf("Failed to allocate flow: %d.\n", fd); client_fini(); - return -1; + return 2; } fccntl(fd, FLOWSFLAGS, FLOWFRDWR | FLOWFRNOPART); - clock_gettime(CLOCK_REALTIME, &tic); + if (client.duration > 0) { + struct itimerval it; + memset(&it, 0, sizeof(it)); + it.it_value.tv_sec = client.duration / 1000; + it.it_value.tv_usec = (client.duration % 1000) * 1000; + setitimer(ITIMER_REAL, &it, NULL); + } - pthread_create(&client.reader_pt, NULL, reader, &fd); - pthread_create(&client.writer_pt, NULL, writer, &fd); + clock_gettime(CLOCK_REALTIME, &tic); - pthread_join(client.writer_pt, NULL); - pthread_join(client.reader_pt, NULL); + if (client.flood_busy) + flood_busy_ping(fd); + else if (client.flood) + flood_ping(fd); + else + threaded_ping(fd); clock_gettime(CLOCK_REALTIME, &toc); - printf("\n"); - printf("--- %s ping statistics ---\n", client.s_apn); - printf("%d packets transmitted, ", client.sent); - printf("%d received, ", client.rcvd); - printf("%zd out-of-order, ", client.ooo); - printf("%.0lf%% packet loss, ", client.sent == 0 ? 0 : - ceil(100 - (100 * (client.rcvd / (float) client.sent)))); - printf("time: %.3f ms\n", ts_diff_us(&tic, &toc) / 1000.0); - - if (client.rcvd > 0) { - printf("rtt min/avg/max/mdev = %.3f/%.3f/%.3f/", - client.rtt_min, - client.rtt_avg, - client.rtt_max); - if (client.rcvd > 1) - printf("%.3f ms\n", - sqrt(client.rtt_m2 / (client.rcvd - 1))); - else - printf("NaN ms\n"); - } + print_stats(&tic, &toc); flow_dealloc(fd); - client_fini(); - return 0; + return client.rcvd == client.sent ? 0 : 1; } diff --git a/src/tools/oping/oping_server.c b/src/tools/oping/oping_server.c index 6f76869c..e98ca040 100644 --- a/src/tools/oping/oping_server.c +++ b/src/tools/oping/oping_server.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Ouroboros ping application * @@ -36,6 +36,8 @@ * OF THE POSSIBILITY OF SUCH DAMAGE. */ + #include <ouroboros/hash.h> + void shutdown_server(int signo, siginfo_t * info, void * c) { (void) info; @@ -67,7 +69,7 @@ void * cleaner_thread(void * o) time_t diff; pthread_mutex_lock(&server.lock); - diff = ts_diff_ms(&server.times[i], &now); + diff = ts_diff_ms(&now, &server.times[i]); pthread_mutex_unlock(&server.lock); if (diff > deadline_ms) { @@ -87,12 +89,15 @@ void * server_thread(void *o) struct oping_msg * msg = (struct oping_msg *) buf; struct timespec now = {0, 0}; struct timespec timeout = {0, 100 * MILLION}; + struct timespec poll_timeout = {0, 0}; int fd; (void) o; while (true) { - if (fevent(server.flows, server.fq, &timeout) == -ETIMEDOUT) + if (fevent(server.flows, server.fq, + server.poll ? &poll_timeout : &timeout) + == -ETIMEDOUT) continue; while ((fd = fqueue_next(server.fq)) >= 0) { @@ -100,14 +105,15 @@ void * server_thread(void *o) if (msg_len < 0) continue; + if (!server.quiet) + printf("Received %d bytes on fd %d.\n", + msg_len, fd); + if (ntohl(msg->type) != ECHO_REQUEST) { printf("Invalid message on fd %d.\n", fd); continue; } - if (!server.quiet) - printf("Received %d bytes on fd %d.\n", msg_len, fd); - clock_gettime(CLOCK_REALTIME, &now); pthread_mutex_lock(&server.lock); @@ -132,13 +138,16 @@ void * accept_thread(void * o) (void) o; - printf("Ouroboros ping server started.\n"); + printf("Ouroboros ping server started."); + if (server.busy) + printf(" [busy-poll]"); + printf("\n"); while (true) { fd = flow_accept(&qs, NULL); if (fd < 0) { - printf("Failed to accept flow.\n"); - break; + printf("Failed to accept flow: %d \n", fd); + continue; } printf("New flow %d.\n", fd); @@ -152,9 +161,53 @@ void * accept_thread(void * o) pthread_mutex_unlock(&server.lock); fccntl(fd, FLOWSFLAGS, - FLOWFRNOBLOCK | FLOWFRDWR | FLOWFRNOPART); + FLOWFRNOBLOCK | FLOWFRDWR + | FLOWFRNOPART); + } + + return (void *) 0; +} + +void * busy_thread(void * o) +{ + char buf[OPING_BUF_SIZE]; + struct oping_msg * msg = (struct oping_msg *) buf; + int fd; + int msg_len; + + (void) o; + + /* Accept a single flow. */ + fd = flow_accept(NULL, NULL); + if (fd < 0) { + printf("Failed to accept flow.\n"); + return (void *) -1; + } + + printf("New flow %d (busy-poll).\n", fd); + + fccntl(fd, FLOWSFLAGS, + FLOWFRNOBLOCK | FLOWFRDWR + | FLOWFRNOPART); + + while (true) { + msg_len = flow_read(fd, buf, + OPING_BUF_SIZE); + if (msg_len == -EAGAIN) + continue; + if (msg_len < 0) + break; + + if (ntohl(msg->type) != ECHO_REQUEST) + continue; + + msg->type = htonl(ECHO_REPLY); + + flow_write(fd, buf, msg_len); } + flow_dealloc(fd); + return (void *) 0; } @@ -184,20 +237,39 @@ int server_main(void) return -1; } + if (pthread_mutex_init(&server.lock, NULL)) { + fqueue_destroy(server.fq); + fset_destroy(server.flows); + return -1; + } + + memset(server.times, 0, sizeof(server.times)); + pthread_create(&server.cleaner_pt, NULL, cleaner_thread, NULL); - pthread_create(&server.accept_pt, NULL, accept_thread, NULL); - pthread_create(&server.server_pt, NULL, server_thread, NULL); - pthread_join(server.accept_pt, NULL); + if (server.busy) { + pthread_create(&server.server_pt, NULL, + busy_thread, NULL); + pthread_join(server.server_pt, NULL); + pthread_cancel(server.cleaner_pt); + } else { + pthread_create(&server.accept_pt, NULL, + accept_thread, NULL); + pthread_create(&server.server_pt, NULL, + server_thread, NULL); + pthread_join(server.accept_pt, NULL); + pthread_cancel(server.server_pt); + } - pthread_cancel(server.server_pt); pthread_cancel(server.cleaner_pt); - fset_destroy(server.flows); - fqueue_destroy(server.fq); - + /* Join cancellable threads before tearing down their fset. */ pthread_join(server.server_pt, NULL); pthread_join(server.cleaner_pt, NULL); + pthread_mutex_destroy(&server.lock); + fset_destroy(server.flows); + fqueue_destroy(server.fq); + return 0; } diff --git a/src/tools/ovpn/CMakeLists.txt b/src/tools/ovpn/CMakeLists.txt deleted file mode 100644 index f3a2cac8..00000000 --- a/src/tools/ovpn/CMakeLists.txt +++ /dev/null @@ -1,21 +0,0 @@ -include_directories(${CMAKE_CURRENT_SOURCE_DIR}) -include_directories(${CMAKE_CURRENT_BINARY_DIR}) - -include_directories(${CMAKE_SOURCE_DIR}/include) -include_directories(${CMAKE_BINARY_DIR}/include) - -get_filename_component(CURRENT_SOURCE_PARENT_DIR - ${CMAKE_CURRENT_SOURCE_DIR} DIRECTORY) - -include_directories(${CURRENT_SOURCE_PARENT_DIR}) - -set(SOURCE_FILES - # Add source files here - ovpn.c - ) - -add_executable(ovpn ${SOURCE_FILES}) - -target_link_libraries(ovpn LINK_PUBLIC ouroboros-dev) - -install(TARGETS ovpn RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR}) diff --git a/src/tools/ovpn/ovpn.c b/src/tools/ovpn/ovpn.c index 5333ff40..95b4572d 100644 --- a/src/tools/ovpn/ovpn.c +++ b/src/tools/ovpn/ovpn.c @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Ouroboros VPN * @@ -68,7 +68,6 @@ static void usage(void) "server to connect to\n" " -i, --ip IP address to give to TUN device\n" " -m, --mask Subnet mask to give to TUN device\n" - " -C, --crypt AES encryption (default: off)\n" "\n" " --help Display this help text and exit\n"); } @@ -194,7 +193,6 @@ int main(int argc, {{"ip", required_argument, NULL, 'i'}, {"mask", required_argument, NULL, 'm'}, {"name", required_argument, NULL, 'n'}, - {"crypt", no_argument, NULL, 'C'}, {"help", no_argument, NULL, 'h'}, {NULL, 0, NULL, 0} }; @@ -231,9 +229,6 @@ int main(int argc, case 'n': name = optarg; break; - case 'C': - qs = qos_raw_crypt; - break; case 'h': usage(); exit(EXIT_SUCCESS); diff --git a/src/tools/time_utils.h b/src/tools/time_utils.h index c17282dc..a0729074 100644 --- a/src/tools/time_utils.h +++ b/src/tools/time_utils.h @@ -1,5 +1,5 @@ /* - * Ouroboros - Copyright (C) 2016 - 2024 + * Ouroboros - Copyright (C) 2016 - 2026 * * Time utilities * @@ -53,17 +53,17 @@ #include <sys/time.h> /* functions for timespecs */ -#define ts_diff_ns(t0, tx) (((tx)->tv_sec - (t0)->tv_sec) * BILLION \ +#define ts_diff_ns(tx, t0) (((tx)->tv_sec - (t0)->tv_sec) * BILLION \ + ((tx)->tv_nsec - (t0)->tv_nsec)) -#define ts_diff_us(t0, tx) (((tx)->tv_sec - (t0)->tv_sec) * MILLION \ +#define ts_diff_us(tx, t0) (((tx)->tv_sec - (t0)->tv_sec) * MILLION \ + ((tx)->tv_nsec - (t0)->tv_nsec) / 1000L) -#define ts_diff_ms(t0, tx) (((tx)->tv_sec - (t0)->tv_sec) * 1000L \ +#define ts_diff_ms(tx, t0) (((tx)->tv_sec - (t0)->tv_sec) * 1000L \ + ((tx)->tv_nsec - (t0)->tv_nsec) / MILLION) /* functions for timevals are the same */ -#define tv_diff_us(t0, tx) (((tx)->tv_sec - (t0)->tv_sec) * MILLION \ +#define tv_diff_us(tx, t0) (((tx)->tv_sec - (t0)->tv_sec) * MILLION \ + ((tx)->tv_usec - (t0)->tv_usec) / 1000L) -#define tv_diff_ms(t0, tx) (((tx)->tv_sec - (t0)->tv_sec) * 1000L \ +#define tv_diff_ms(tx, t0) (((tx)->tv_sec - (t0)->tv_sec) * 1000L \ + ((tx)->tv_usec - (t0)->tv_usec) / MILLION) /* functions for timespecs */ |