diff options
Diffstat (limited to 'src/ipcpd')
-rw-r--r-- | src/ipcpd/raptor/CMakeLists.txt | 60 | ||||
-rw-r--r-- | src/ipcpd/raptor/main.c | 713 | ||||
-rw-r--r-- | src/ipcpd/raptor/raptor_messages.proto | 39 |
3 files changed, 377 insertions, 435 deletions
diff --git a/src/ipcpd/raptor/CMakeLists.txt b/src/ipcpd/raptor/CMakeLists.txt index 2cb86a76..03411941 100644 --- a/src/ipcpd/raptor/CMakeLists.txt +++ b/src/ipcpd/raptor/CMakeLists.txt @@ -12,26 +12,40 @@ include_directories(${CURRENT_BINARY_PARENT_DIR}) include_directories(${CMAKE_SOURCE_DIR}/include) include_directories(${CMAKE_BINARY_DIR}/include) -protobuf_generate_c(RAPTOR_PROTO_SRCS RAPTOR_PROTO_HDRS - raptor_messages.proto) - -set(IPCP_RAPTOR_TARGET ipcpd-raptor - CACHE STRING "IPCP_RAPTOR_TARGET") - -set(RAPTOR_SOURCES - # Add source files here - ${CMAKE_CURRENT_SOURCE_DIR}/main.c) - -add_executable(ipcpd-raptor ${RAPTOR_SOURCES} ${IPCP_SOURCES} - ${RAPTOR_PROTO_SRCS}) -target_link_libraries(ipcpd-raptor LINK_PUBLIC ouroboros - ${PROTOBUF_C_LIBRARY}) - -test_and_set_c_compiler_flag_global(-std=c99) - -include(AddCompileFlags) -if (CMAKE_BUILD_TYPE MATCHES Debug) - add_compile_flags(ipcpd-raptor -DCONFIG_OUROBOROS_DEBUG) -endif (CMAKE_BUILD_TYPE MATCHES Debug) - -install(TARGETS ipcpd-raptor RUNTIME DESTINATION sbin) +if (CMAKE_SYSTEM_NAME STREQUAL "Linux") + find_path(RAPTOR_KERNEL_MODULE + NAMES + raptor.ko + raptor.ko.gz + HINTS + /lib/modules/${CMAKE_SYSTEM_VERSION}/ + /lib/modules/${CMAKE_SYSTEM_VERSION}/extramodules/ + ) + + mark_as_advanced(RAPTOR_KERNEL_MODULE) + + if (RAPTOR_KERNEL_MODULE) + set(DISABLE_RAPTOR FALSE CACHE BOOL + "Disable support for raptor devices") + if (NOT DISABLE_RAPTOR) + message(STATUS "Kernel module for raptor found. Building raptor.") + set(IPCP_RAPTOR_TARGET ipcpd-raptor CACHE INTERNAL "") + + set(RAPTOR_SOURCES + # Add source files here + ${CMAKE_CURRENT_SOURCE_DIR}/main.c) + + add_executable(ipcpd-raptor ${RAPTOR_SOURCES} ${IPCP_SOURCES}) + target_link_libraries(ipcpd-raptor LINK_PUBLIC ouroboros-dev) + + include(AddCompileFlags) + if (CMAKE_BUILD_TYPE MATCHES "Debug*") + add_compile_flags(ipcpd-raptor -DCONFIG_OUROBOROS_DEBUG) + endif () + + install(TARGETS ipcpd-raptor RUNTIME DESTINATION sbin) + else () + message(STATUS "Raptor support disabled by user") + endif () + endif () +endif () diff --git a/src/ipcpd/raptor/main.c b/src/ipcpd/raptor/main.c index be66bd9a..d0790321 100644 --- a/src/ipcpd/raptor/main.c +++ b/src/ipcpd/raptor/main.c @@ -23,9 +23,10 @@ #define _DEFAULT_SOURCE +#include "config.h" + #define OUROBOROS_PREFIX "ipcpd/raptor" -#include <ouroboros/config.h> #include <ouroboros/hash.h> #include <ouroboros/errno.h> #include <ouroboros/list.h> @@ -39,7 +40,7 @@ #include <ouroboros/time_utils.h> #include "ipcp.h" -#include "raptor_messages.pb-c.h" +#include "shim-data.h" #include <net/if.h> #include <signal.h> @@ -62,49 +63,58 @@ #include <poll.h> #include <sys/mman.h> -typedef RaptorMsg raptor_msg_t; - -#define THIS_TYPE IPCP_RAPTOR -#define MGMT_SAP 0x01 -#define MAC_SIZE 6 -#define MAX_SAPS 64 +#define THIS_TYPE IPCP_RAPTOR +#define MGMT_SAP 0x01 +#define MAC_SIZE 6 +#define MAX_SAPS 64 -#define EVENT_WAIT_TIMEOUT 100 /* us */ +#define EVENT_WAIT_TIMEOUT 100 /* us */ #define NAME_QUERY_TIMEOUT 2000 /* ms */ -#define MGMT_TIMEOUT 100 /* ms */ +#define MGMT_TIMEOUT 100 /* ms */ -#define IOCTL_SEND 0xAD420000 -#define IOCTL_RECV 0xAD430000 -#define IOCTL_SEND_DONE 0xAD440000 -#define IOCTL_RECV_DONE 0xAD450000 -#define IOCTL_RECV_NEED 0xAD460000 +#define IOCTL_SEND 0xAD420000 +#define IOCTL_RECV 0xAD430000 +#define IOCTL_SEND_DONE 0xAD440000 +#define IOCTL_RECV_DONE 0xAD450000 +#define IOCTL_RECV_NEED 0xAD460000 -#define RAPTOR_PAGE ((1 << 12) - 200) /* 4kB - 200 */ +#define RAPTOR_PAGE ((1 << 12) - 200) /* 4kB - 200 */ +#define RAPTOR_PAGE_MASK (~0xFFF) -#define RAPTOR_PAGE_MASK (~0xFFF) +#define RAPTOR_BATCH 100 +#define RAPTOR_HEADER 3 -#define RAPTOR_BATCH 100 +#define FLOW_REQ 0 +#define FLOW_REPLY 1 +#define NAME_QUERY_REQ 2 +#define NAME_QUERY_REPLY 3 -#define RAPTOR_HEADER 3 +struct mgmt_msg { + uint8_t code; + uint8_t ssap; + uint8_t dsap; + uint8_t qoscube; + int8_t response; +} __attribute__((packed)); struct ef { int8_t sap; int8_t r_sap; - uint8_t r_addr[MAC_SIZE]; }; struct mgmt_frame { struct list_head next; - uint8_t r_addr[MAC_SIZE]; uint8_t buf[RAPTOR_PAGE]; size_t len; }; struct { + struct shim_data * shim_data; + int ioctl_fd; struct bmp * saps; - flow_set_t * np1_flows; + fset_t * np1_flows; fqueue_t * fq; int * ef_to_fd; struct ef * fd_to_ef; @@ -126,119 +136,113 @@ struct { static int raptor_data_init(void) { int i; - int ret = -1; + int ret = -ENOMEM; pthread_condattr_t cattr; - log_info("%s", __FUNCTION__); - - raptor_data.fd_to_ef = malloc(sizeof(struct ef) * IRMD_MAX_FLOWS); + raptor_data.fd_to_ef = + malloc(sizeof(*raptor_data.fd_to_ef) * SYS_MAX_FLOWS); if (raptor_data.fd_to_ef == NULL) - return -ENOMEM; + goto fail_fd_to_ef; - raptor_data.ef_to_fd = malloc(sizeof(struct ef) * MAX_SAPS); - if (raptor_data.ef_to_fd == NULL) { - ret = -ENOMEM; - goto free_fd_to_ef; - } + raptor_data.ef_to_fd = + malloc(sizeof(*raptor_data.ef_to_fd) * MAX_SAPS); + if (raptor_data.ef_to_fd == NULL) + goto fail_ef_to_fd; raptor_data.saps = bmp_create(MAX_SAPS, 2); - if (raptor_data.saps == NULL) { - ret = -ENOMEM; - goto free_ef_to_fd; - } + if (raptor_data.saps == NULL) + goto fail_saps; - raptor_data.np1_flows = flow_set_create(); - if (raptor_data.np1_flows == NULL) { - ret = -ENOMEM; - goto bmp_destroy; - } + raptor_data.np1_flows = fset_create(); + if (raptor_data.np1_flows == NULL) + goto fail_np1_flows; raptor_data.fq = fqueue_create(); - if (raptor_data.fq == NULL) { - ret = -ENOMEM; - goto flow_set_destroy; - } + if (raptor_data.fq == NULL) + goto fail_fq; for (i = 0; i < MAX_SAPS; ++i) raptor_data.ef_to_fd[i] = -1; - for (i = 0; i < IRMD_MAX_FLOWS; ++i) { + for (i = 0; i < SYS_MAX_FLOWS; ++i) { raptor_data.fd_to_ef[i].sap = -1; raptor_data.fd_to_ef[i].r_sap = -1; - memset(&raptor_data.fd_to_ef[i].r_addr, 0, MAC_SIZE); } + raptor_data.shim_data = shim_data_create(); + if (raptor_data.shim_data == NULL) + goto fail_shim_data; + + ret = -1; + if (pthread_rwlock_init(&raptor_data.flows_lock, NULL)) - goto fqueue_destroy; + goto fail_flows_lock; if (pthread_mutex_init(&raptor_data.mgmt_lock, NULL)) - goto flow_lock_destroy; + goto fail_mgmt_lock; if (pthread_condattr_init(&cattr)) - goto mgmt_lock_destroy;; + goto fail_condattr; #ifndef __APPLE__ pthread_condattr_setclock(&cattr, PTHREAD_COND_CLOCK); #endif if (pthread_cond_init(&raptor_data.mgmt_cond, &cattr)) - goto mgmt_lock_destroy; + goto fail_mgmt_cond; + + pthread_condattr_destroy(&cattr); list_head_init(&raptor_data.mgmt_frames); return 0; - mgmt_lock_destroy: + fail_mgmt_cond: + pthread_condattr_destroy(&cattr); + fail_condattr: pthread_mutex_destroy(&raptor_data.mgmt_lock); - flow_lock_destroy: + fail_mgmt_lock: pthread_rwlock_destroy(&raptor_data.flows_lock); - fqueue_destroy: + fail_flows_lock: + shim_data_destroy(raptor_data.shim_data); + fail_shim_data: fqueue_destroy(raptor_data.fq); - flow_set_destroy: - flow_set_destroy(raptor_data.np1_flows); - bmp_destroy: + fail_fq: + fset_destroy(raptor_data.np1_flows); + fail_np1_flows: bmp_destroy(raptor_data.saps); - free_ef_to_fd: + fail_saps: free(raptor_data.ef_to_fd); - free_fd_to_ef: + fail_ef_to_fd: free(raptor_data.fd_to_ef); - + fail_fd_to_ef: return ret; } -void raptor_data_fini(void) +static void raptor_data_fini(void) { - log_info("%s", __FUNCTION__); - close(raptor_data.ioctl_fd); pthread_cond_destroy(&raptor_data.mgmt_cond); pthread_mutex_destroy(&raptor_data.mgmt_lock); pthread_rwlock_destroy(&raptor_data.flows_lock); fqueue_destroy(raptor_data.fq); - flow_set_destroy(raptor_data.np1_flows); + fset_destroy(raptor_data.np1_flows); bmp_destroy(raptor_data.saps); free(raptor_data.fd_to_ef); free(raptor_data.ef_to_fd); } -static int raptor_ipcp_send_frame(struct shm_du_buff * sdb, - uint8_t dsap) +static int raptor_send_frame(struct shm_du_buff * sdb, + uint8_t dsap) { - /* length (16b) + dsap (8b) */ - uint8_t * frame; - size_t frame_len; + size_t frame_len; uint8_t * payload; - size_t len; + size_t len; payload = shm_du_buff_head(sdb); len = shm_du_buff_tail(sdb) - shm_du_buff_head(sdb); - if (payload == NULL) { - log_err("Payload was NULL."); - return -1; - } - frame_len = RAPTOR_HEADER + len; if (frame_len >= RAPTOR_PAGE) { @@ -247,7 +251,6 @@ static int raptor_ipcp_send_frame(struct shm_du_buff * sdb, } frame = memalign(1 << 12, 1 << 12); - if (frame == NULL) { log_err("frame == NULL"); return -1; @@ -264,99 +267,82 @@ static int raptor_ipcp_send_frame(struct shm_du_buff * sdb, memcpy(&frame[RAPTOR_HEADER], payload, len); - ipcp_sdb_release(sdb); - if (ioctl(raptor_data.ioctl_fd, IOCTL_SEND | 1, &frame) != 1) { log_err("Ioctl send failed."); free(frame); return -1; } + return 0; } -static int raptor_ipcp_send_mgmt_frame(raptor_msg_t * msg, - const uint8_t * dst_addr) +static int raptor_sap_alloc(uint8_t ssap, + const uint8_t * hash, + qoscube_t cube) { - size_t len; - uint8_t * buf; + struct mgmt_msg * msg; struct shm_du_buff * sdb; - (void)dst_addr; - - log_info("%s", __FUNCTION__); - - len = raptor_msg__get_packed_size(msg); - if (len == 0) - return -1; - - if (ipcp_sdb_reserve(&sdb, len) < 0) { + if (ipcp_sdb_reserve(&sdb, sizeof(*msg) + ipcp_dir_hash_len()) < 0) { log_err("failed to reserve sdb for management frame."); return -1; } - buf = shm_du_buff_head(sdb); + msg = (struct mgmt_msg *) shm_du_buff_head(sdb); + msg->code = FLOW_REQ; + msg->ssap = ssap; + msg->qoscube = cube; - raptor_msg__pack(msg, buf); + memcpy(msg + 1, hash, ipcp_dir_hash_len()); - if (raptor_ipcp_send_frame(sdb, MGMT_SAP)) { + if (raptor_send_frame(sdb, MGMT_SAP)) { log_err("Failed to send management frame."); + ipcp_sdb_release(sdb); return -1; } + + ipcp_sdb_release(sdb); + return 0; } -static int raptor_ipcp_sap_alloc(const uint8_t * dst_addr, - uint8_t ssap, - const uint8_t * hash, - qoscube_t cube) +static int raptor_sap_alloc_resp(uint8_t ssap, + uint8_t dsap, + int response) { - raptor_msg_t msg = RAPTOR_MSG__INIT; - - log_info("%s", __FUNCTION__); - - msg.code = RAPTOR_MSG_CODE__FLOW_REQ; - msg.has_ssap = true; - msg.ssap = ssap; - msg.has_hash = true; - msg.hash.len = ipcp_dir_hash_len(); - msg.hash.data = (uint8_t *) hash; - msg.has_qoscube = true; - msg.qoscube = cube; + struct mgmt_msg * msg; + struct shm_du_buff * sdb; - return raptor_ipcp_send_mgmt_frame(&msg, dst_addr); -} + if (ipcp_sdb_reserve(&sdb, sizeof(*msg)) < 0) { + log_err("failed to reserve sdb for management frame."); + return -1; + } -static int raptor_ipcp_sap_alloc_resp(uint8_t * dst_addr, - uint8_t ssap, - uint8_t dsap, - int response) -{ - raptor_msg_t msg = RAPTOR_MSG__INIT; + msg = (struct mgmt_msg *) shm_du_buff_head(sdb); + msg->code = FLOW_REPLY; + msg->ssap = ssap; + msg->dsap = dsap; + msg->response = response; - log_info("%s", __FUNCTION__); + if (raptor_send_frame(sdb, MGMT_SAP)) { + log_err("Failed to send management frame."); + ipcp_sdb_release(sdb); + return -1; + } - msg.code = RAPTOR_MSG_CODE__FLOW_REPLY; - msg.has_ssap = true; - msg.ssap = ssap; - msg.has_dsap = true; - msg.dsap = dsap; - msg.has_response = true; - msg.response = response; + ipcp_sdb_release(sdb); - return raptor_ipcp_send_mgmt_frame(&msg, dst_addr); + return 0; } -static int raptor_ipcp_sap_req(uint8_t r_sap, - uint8_t * r_addr, - const uint8_t * dst, - qoscube_t cube) +static int raptor_sap_req(uint8_t r_sap, + const uint8_t * dst, + qoscube_t cube) { struct timespec ts = {0, EVENT_WAIT_TIMEOUT * 1000}; struct timespec abstime; int fd; - log_info("%s", __FUNCTION__); - clock_gettime(PTHREAD_COND_CLOCK, &abstime); pthread_mutex_lock(&ipcpi.alloc_lock); @@ -385,7 +371,6 @@ static int raptor_ipcp_sap_req(uint8_t r_sap, pthread_rwlock_wrlock(&raptor_data.flows_lock); raptor_data.fd_to_ef[fd].r_sap = r_sap; - memcpy(raptor_data.fd_to_ef[fd].r_addr, r_addr, MAC_SIZE); ipcpi.alloc_id = fd; pthread_cond_broadcast(&ipcpi.alloc_cond); @@ -398,16 +383,13 @@ static int raptor_ipcp_sap_req(uint8_t r_sap, return 0; } -static int raptor_ipcp_sap_alloc_reply(uint8_t ssap, - uint8_t * r_addr, - int dsap, - int response) +static int raptor_sap_alloc_reply(uint8_t ssap, + int dsap, + int response) { int ret = 0; int fd = -1; - log_info("%s", __FUNCTION__); - pthread_rwlock_wrlock(&raptor_data.flows_lock); fd = raptor_data.ef_to_fd[dsap]; @@ -417,12 +399,10 @@ static int raptor_ipcp_sap_alloc_reply(uint8_t ssap, return -1; /* -EFLOWNOTFOUND */ } - if (response) { + if (response) bmp_release(raptor_data.saps, raptor_data.fd_to_ef[fd].sap); - } else { + else raptor_data.fd_to_ef[fd].r_sap = ssap; - memcpy(raptor_data.fd_to_ef[fd].r_addr, r_addr, MAC_SIZE); - } pthread_rwlock_unlock(&raptor_data.flows_lock); @@ -435,96 +415,93 @@ static int raptor_ipcp_sap_alloc_reply(uint8_t ssap, } -static int raptor_ipcp_name_query_req(const uint8_t * hash, - uint8_t * r_addr) +static int raptor_name_query_req(const uint8_t * hash) { - raptor_msg_t msg = RAPTOR_MSG__INIT; + struct mgmt_msg * msg; + struct shm_du_buff * sdb; + + if (!shim_data_reg_has(raptor_data.shim_data, hash)) + return 0; + + if (ipcp_sdb_reserve(&sdb, sizeof(*msg) + ipcp_dir_hash_len()) < 0) { + log_err("failed to reserve sdb for management frame."); + return -1; + } - log_info("%s", __FUNCTION__); + msg = (struct mgmt_msg *) shm_du_buff_head(sdb); + msg->code = NAME_QUERY_REPLY; - if (shim_data_reg_has(ipcpi.shim_data, hash)) { - msg.code = RAPTOR_MSG_CODE__NAME_QUERY_REPLY; - msg.has_hash = true; - msg.hash.len = ipcp_dir_hash_len(); - msg.hash.data = (uint8_t *) hash; + memcpy(msg + 1, hash, ipcp_dir_hash_len()); - raptor_ipcp_send_mgmt_frame(&msg, r_addr); + if (raptor_send_frame(sdb, MGMT_SAP)) { + log_err("Failed to send management frame."); + ipcp_sdb_release(sdb); + return -1; } + ipcp_sdb_release(sdb); + return 0; } -static int raptor_ipcp_name_query_reply(const uint8_t * hash, - uint8_t * r_addr) +static int raptor_name_query_reply(const uint8_t * hash) { - uint64_t address = 0; - struct list_head * pos; - - log_info("%s", __FUNCTION__); - - memcpy(&address, r_addr, MAC_SIZE); + shim_data_dir_add_entry(raptor_data.shim_data, hash, 0); - shim_data_dir_add_entry(ipcpi.shim_data, hash, address); - - pthread_mutex_lock(&ipcpi.shim_data->dir_queries_lock); - list_for_each(pos, &ipcpi.shim_data->dir_queries) { - struct dir_query * e = - list_entry(pos, struct dir_query, next); - if (memcmp(e->hash, hash, ipcp_dir_hash_len()) == 0) { - shim_data_dir_query_respond(e); - } - } - pthread_mutex_unlock(&ipcpi.shim_data->dir_queries_lock); + shim_data_dir_query_respond(raptor_data.shim_data, hash); return 0; } -static int raptor_ipcp_mgmt_frame(const uint8_t * buf, - size_t len, - uint8_t * r_addr) +static int raptor_mgmt_frame(const uint8_t * buf, + size_t len) { - raptor_msg_t * msg; - - log_info("%s", __FUNCTION__); - - msg = raptor_msg__unpack(NULL, len, buf); - if (msg == NULL) { - log_err("Failed to unpack."); - return -1; - } + struct mgmt_msg * msg = (struct mgmt_msg *) buf; + uint8_t * hash = (uint8_t *) (msg + 1); switch (msg->code) { - case RAPTOR_MSG_CODE__FLOW_REQ: - if (shim_data_reg_has(ipcpi.shim_data, msg->hash.data)) { - raptor_ipcp_sap_req(msg->ssap, - r_addr, - msg->hash.data, - msg->qoscube); + case FLOW_REQ: + if (len != sizeof(*msg) + ipcp_dir_hash_len()) { + log_err("Corrupt message received."); + return -1; } + + if (shim_data_reg_has(raptor_data.shim_data, hash)) + raptor_sap_req(msg->ssap, hash, msg->qoscube); break; - case RAPTOR_MSG_CODE__FLOW_REPLY: - raptor_ipcp_sap_alloc_reply(msg->ssap, - r_addr, - msg->dsap, - msg->response); + case FLOW_REPLY: + if (len != sizeof(*msg)) { + log_err("Corrupt message received."); + return -1; + } + + raptor_sap_alloc_reply(msg->ssap, msg->dsap, msg->response); break; - case RAPTOR_MSG_CODE__NAME_QUERY_REQ: - raptor_ipcp_name_query_req(msg->hash.data, r_addr); + case NAME_QUERY_REQ: + if (len != sizeof(*msg) + ipcp_dir_hash_len()) { + log_err("Corrupt message received."); + return -1; + } + + raptor_name_query_req(hash); break; - case RAPTOR_MSG_CODE__NAME_QUERY_REPLY: - raptor_ipcp_name_query_reply(msg->hash.data, r_addr); + case NAME_QUERY_REPLY: + if (len != sizeof(*msg) + ipcp_dir_hash_len()) { + log_err("Corrupt message received."); + return -1; + } + + raptor_name_query_reply(hash); break; default: log_err("Unknown message received %d.", msg->code); - raptor_msg__free_unpacked(msg, NULL); return -1; } - raptor_msg__free_unpacked(msg, NULL); return 0; } -static void * raptor_ipcp_mgmt_handler(void * o) +static void * raptor_mgmt_handler(void * o) { int ret; struct timespec timeout = {(MGMT_TIMEOUT / 1000), @@ -534,10 +511,6 @@ static void * raptor_ipcp_mgmt_handler(void * o) (void) o; - log_info("%s", __FUNCTION__); - - log_info("Mgmt handler started."); - while (true) { ret = 0; @@ -570,44 +543,43 @@ static void * raptor_ipcp_mgmt_handler(void * o) list_del(&frame->next); pthread_mutex_unlock(&raptor_data.mgmt_lock); - raptor_ipcp_mgmt_frame(frame->buf, frame->len, frame->r_addr); + raptor_mgmt_frame(frame->buf, frame->len); free(frame); } - log_info("Mgmt handler stopped."); return NULL; } -static void raptor_ipcp_recv_frame(uint8_t* frame) +static void raptor_recv_frame(uint8_t * frame) { - uint8_t dsap; - uint8_t* payload; - size_t frame_len; - size_t length; - int fd; - struct mgmt_frame * mgmt_frame; + uint8_t dsap; + uint8_t * payload; + size_t frame_len; + size_t length; + int fd; + struct mgmt_frame * mgmt_frame; struct shm_du_buff * sdb; - size_t idx; + size_t idx; - sdb = (struct shm_du_buff*)((uint64_t)frame & RAPTOR_PAGE_MASK); + sdb = (struct shm_du_buff *)((uint64_t) frame & RAPTOR_PAGE_MASK); idx = shm_du_buff_get_idx(sdb); frame_len = frame[0] | (frame[1] << 8); - if (frame_len < RAPTOR_HEADER) { log_err("Received packet smaller than header alone."); ipcp_sdb_release(sdb); return; } + if (frame_len >= RAPTOR_PAGE) { log_err("Received packet too large."); ipcp_sdb_release(sdb); return; } - dsap = frame[2]; + dsap = frame[2]; payload = &frame[RAPTOR_HEADER]; - length = frame_len - RAPTOR_HEADER; + length = frame_len - RAPTOR_HEADER; shm_du_buff_head_release(sdb, RAPTOR_HEADER); shm_du_buff_tail_release(sdb, RAPTOR_PAGE - frame_len); @@ -623,7 +595,6 @@ static void raptor_ipcp_recv_frame(uint8_t* frame) } memcpy(mgmt_frame->buf, payload, length); - memset(mgmt_frame->r_addr, 0, MAC_SIZE); mgmt_frame->len = length; list_add(&mgmt_frame->next, &raptor_data.mgmt_frames); pthread_cond_signal(&raptor_data.mgmt_cond); @@ -646,34 +617,32 @@ static void raptor_ipcp_recv_frame(uint8_t* frame) } } -static void * raptor_ipcp_recv_done_thread(void * o) +static void * raptor_recv_done_thread(void * o) { uint8_t * frames[RAPTOR_BATCH]; - int count; - int i; + int count; + int i; - (void)o; - - log_info("Recv_done thread started."); + (void) o; while (true) { if (ipcp_get_state() != IPCP_OPERATIONAL) break; - count = ioctl(raptor_data.ioctl_fd, IOCTL_RECV_DONE | RAPTOR_BATCH, frames); + count = ioctl(raptor_data.ioctl_fd, + IOCTL_RECV_DONE | RAPTOR_BATCH, frames); if (count <= 0) continue; for (i = 0; i < count; i++) - raptor_ipcp_recv_frame(frames[i]); + raptor_recv_frame(frames[i]); } - log_info("Recv_done thread stopped."); return NULL; } -static void * raptor_ipcp_send_thread(void * o) +static void * raptor_send_thread(void * o) { struct timespec timeout = {0, EVENT_WAIT_TIMEOUT * 1000}; int fd; @@ -682,12 +651,7 @@ static void * raptor_ipcp_send_thread(void * o) (void) o; - log_info("Send thread started."); - - while (flow_event_wait(raptor_data.np1_flows, - raptor_data.fq, - &timeout)) { - + while (fevent(raptor_data.np1_flows, raptor_data.fq, &timeout)) { if (ipcp_get_state() != IPCP_OPERATIONAL) break; @@ -700,80 +664,75 @@ static void * raptor_ipcp_send_thread(void * o) dsap = raptor_data.fd_to_ef[fd].r_sap; - raptor_ipcp_send_frame(sdb, dsap); + raptor_send_frame(sdb, dsap); } pthread_rwlock_unlock(&raptor_data.flows_lock); } - log_info("Send thread stopped."); return NULL; } -static void * raptor_ipcp_send_done_thread(void * o) +static void * raptor_send_done_thread(void * o) { uint8_t * frames[RAPTOR_BATCH]; - int count; - int i; + int count; + int i; - (void)o; - - log_info("Send_done thread started."); + (void) o; while (true) { if (ipcp_get_state() != IPCP_OPERATIONAL) break; - count = ioctl(raptor_data.ioctl_fd, IOCTL_SEND_DONE | RAPTOR_BATCH, frames); + count = ioctl(raptor_data.ioctl_fd, + IOCTL_SEND_DONE | RAPTOR_BATCH, frames); if (count <= 0) continue; - for (i = 0; i < count; i++) { + for (i = 0; i < count; i++) free(frames[i]); - } } - log_info("Send_done thread stopped."); return NULL; } -static void * raptor_ipcp_recv_thread(void * o) +static void * raptor_recv_thread(void * o) { struct shm_du_buff * sdb; - uint8_t * frames[RAPTOR_BATCH]; - uint8_t ** head; - int needed = 0; - int count; - int i; - - (void)o; + uint8_t * frames[RAPTOR_BATCH]; + uint8_t ** head; + int needed = 0; + int count; + int i; - log_info("Recv thread started."); + (void) o; while (true) { if (ipcp_get_state() != IPCP_OPERATIONAL) break; - needed = ioctl(raptor_data.ioctl_fd, IOCTL_RECV_NEED | RAPTOR_BATCH, NULL); + needed = ioctl(raptor_data.ioctl_fd, + IOCTL_RECV_NEED | RAPTOR_BATCH, NULL); if (needed <= 0) continue; for (i = 0; i < needed; i++) { if (ipcp_sdb_reserve(&sdb, RAPTOR_PAGE) < 0) { - log_err("Recv thread: reserve sdb failed. Stopping thread."); + log_err("Recv thread: reserve sdb failed."); return NULL; } if ((uint64_t)sdb & (~RAPTOR_PAGE_MASK)) { - log_err("Recv thread: sdb not at offset 0 in page. Stopping thread."); + log_err("Recv thread: sdb not at offset 0."); return NULL; } frames[i] = shm_du_buff_head(sdb); if ((uint64_t)frames[i] & 0x7) { - log_err("Recv thread: frame not 64bit aligned. Stopping thread."); + log_err("Recv thread: frame not aligned."); return NULL; } } @@ -781,8 +740,8 @@ static void * raptor_ipcp_recv_thread(void * o) head = frames; do { - count = ioctl(raptor_data.ioctl_fd, IOCTL_RECV | needed, head); - + count = ioctl(raptor_data.ioctl_fd, + IOCTL_RECV | needed, head); if (count <= 0) continue; @@ -794,69 +753,93 @@ static void * raptor_ipcp_recv_thread(void * o) } while (needed > 0 && ipcp_get_state() == IPCP_OPERATIONAL); } - log_info("Recv thread stopped."); return NULL; } -static int raptor_ipcp_bootstrap(const struct ipcp_config * conf) +static int raptor_bootstrap(const struct ipcp_config * conf) { assert(conf); assert(conf->type == THIS_TYPE); - log_info("%s", __FUNCTION__); + (void) conf; raptor_data.ioctl_fd = open("/dev/raptor", 0); - if (raptor_data.ioctl_fd < 0) { log_err("Failed to open /dev/raptor."); - return -1; + goto fail_ioctl; } ipcp_set_state(IPCP_OPERATIONAL); - pthread_create(&raptor_data.mgmt_handler, - NULL, - raptor_ipcp_mgmt_handler, - NULL); + if (pthread_create(&raptor_data.mgmt_handler, + NULL, + raptor_mgmt_handler, + NULL)) { + ipcp_set_state(IPCP_INIT); + goto fail_mgmt_handler; + } - pthread_create(&raptor_data.send_thread, - NULL, - raptor_ipcp_send_thread, - NULL); + if (pthread_create(&raptor_data.send_thread, + NULL, + raptor_send_thread, + NULL)) { + ipcp_set_state(IPCP_INIT); + goto fail_send_thread; + } - pthread_create(&raptor_data.recv_thread, - NULL, - raptor_ipcp_recv_thread, - NULL); + if (pthread_create(&raptor_data.recv_thread, + NULL, + raptor_recv_thread, + NULL)) { + ipcp_set_state(IPCP_INIT); + goto fail_recv_thread; + } - pthread_create(&raptor_data.send_done_thread, - NULL, - raptor_ipcp_send_done_thread, - NULL); + if (pthread_create(&raptor_data.send_done_thread, + NULL, + raptor_send_done_thread, + NULL)) { + ipcp_set_state(IPCP_INIT); + goto fail_send_done_thread; + } - pthread_create(&raptor_data.recv_done_thread, - NULL, - raptor_ipcp_recv_done_thread, - NULL); + if (pthread_create(&raptor_data.recv_done_thread, + NULL, + raptor_recv_done_thread, + NULL)) { + ipcp_set_state(IPCP_INIT); + goto fail_recv_done_thread; + } log_dbg("Bootstrapped raptor IPCP with api %d.", getpid()); return 0; + + fail_recv_done_thread: + pthread_join(raptor_data.send_done_thread, NULL); + fail_send_done_thread: + pthread_join(raptor_data.recv_thread, NULL); + fail_recv_thread: + pthread_join(raptor_data.send_thread, NULL); + fail_send_thread: + pthread_join(raptor_data.mgmt_handler, NULL); + fail_mgmt_handler: + close(raptor_data.ioctl_fd); + fail_ioctl: + return -1; } -static int raptor_ipcp_reg(const uint8_t * hash) +static int raptor_reg(const uint8_t * hash) { uint8_t * hash_dup; - log_info("%s", __FUNCTION__); - hash_dup = ipcp_hash_dup(hash); if (hash_dup == NULL) { log_err("Failed to duplicate hash."); return -ENOMEM; } - if (shim_data_reg_add_entry(ipcpi.shim_data, hash_dup)) { + if (shim_data_reg_add_entry(raptor_data.shim_data, hash_dup)) { log_err("Failed to add " HASH_FMT " to local registry.", HASH_VAL(hash)); free(hash_dup); @@ -868,65 +851,59 @@ static int raptor_ipcp_reg(const uint8_t * hash) return 0; } -static int raptor_ipcp_unreg(const uint8_t * hash) +static int raptor_unreg(const uint8_t * hash) { - log_info("%s", __FUNCTION__); - - shim_data_reg_del_entry(ipcpi.shim_data, hash); + shim_data_reg_del_entry(raptor_data.shim_data, hash); return 0; } -static int raptor_ipcp_query(const uint8_t * hash) +static int raptor_query(const uint8_t * hash) { - uint8_t r_addr[MAC_SIZE]; - struct timespec timeout = {(NAME_QUERY_TIMEOUT / 1000), - (NAME_QUERY_TIMEOUT % 1000) * MILLION}; - raptor_msg_t msg = RAPTOR_MSG__INIT; - struct dir_query * query; - int ret; - - log_info("%s", __FUNCTION__); + struct timespec timeout = {(NAME_QUERY_TIMEOUT / 1000), + (NAME_QUERY_TIMEOUT % 1000) * MILLION}; + struct mgmt_msg * msg; + struct dir_query * query; + int ret; + struct shm_du_buff * sdb; - if (shim_data_dir_has(ipcpi.shim_data, hash)) + if (shim_data_dir_has(raptor_data.shim_data, hash)) return 0; - msg.code = RAPTOR_MSG_CODE__NAME_QUERY_REQ; - msg.has_hash = true; - msg.hash.len = ipcp_dir_hash_len(); - msg.hash.data = (uint8_t *) hash; + if (ipcp_sdb_reserve(&sdb, sizeof(*msg) + ipcp_dir_hash_len()) < 0) { + log_err("failed to reserve sdb for management frame."); + return -1; + } + + msg = (struct mgmt_msg *) shm_du_buff_head(sdb); + msg->code = NAME_QUERY_REQ; - memset(r_addr, 0xff, MAC_SIZE); + memcpy(msg + 1, hash, ipcp_dir_hash_len()); - query = shim_data_dir_query_create(hash); - if (query == NULL) + query = shim_data_dir_query_create(raptor_data.shim_data, hash); + if (query == NULL) { + ipcp_sdb_release(sdb); return -1; + } - pthread_mutex_lock(&ipcpi.shim_data->dir_queries_lock); - list_add(&query->next, &ipcpi.shim_data->dir_queries); - pthread_mutex_unlock(&ipcpi.shim_data->dir_queries_lock); - - raptor_ipcp_send_mgmt_frame(&msg, r_addr); + if (raptor_send_frame(sdb, MGMT_SAP)) { + log_err("Failed to send management frame."); + ipcp_sdb_release(sdb); + return -1; + } ret = shim_data_dir_query_wait(query, &timeout); - pthread_mutex_lock(&ipcpi.shim_data->dir_queries_lock); - list_del(&query->next); - shim_data_dir_query_destroy(query); - pthread_mutex_unlock(&ipcpi.shim_data->dir_queries_lock); + shim_data_dir_query_destroy(raptor_data.shim_data, query); return ret; } -static int raptor_ipcp_flow_alloc(int fd, - const uint8_t * hash, - qoscube_t cube) +static int raptor_flow_alloc(int fd, + const uint8_t * hash, + qoscube_t cube) { uint8_t ssap = 0; - uint8_t r_addr[MAC_SIZE]; - uint64_t addr = 0; - - log_info("%s", __FUNCTION__); log_dbg("Allocating flow to " HASH_FMT ".", HASH_VAL(hash)); @@ -937,11 +914,10 @@ static int raptor_ipcp_flow_alloc(int fd, return -1; } - if (!shim_data_dir_has(ipcpi.shim_data, hash)) { + if (!shim_data_dir_has(raptor_data.shim_data, hash)) { log_err("Destination unreachable."); return -1; } - addr = shim_data_dir_get_addr(ipcpi.shim_data, hash); pthread_rwlock_wrlock(&raptor_data.flows_lock); @@ -956,9 +932,7 @@ static int raptor_ipcp_flow_alloc(int fd, pthread_rwlock_unlock(&raptor_data.flows_lock); - memcpy(r_addr, &addr, MAC_SIZE); - - if (raptor_ipcp_sap_alloc(r_addr, ssap, hash, cube) < 0) { + if (raptor_sap_alloc(ssap, hash, cube) < 0) { pthread_rwlock_wrlock(&raptor_data.flows_lock); bmp_release(raptor_data.saps, raptor_data.fd_to_ef[fd].sap); raptor_data.fd_to_ef[fd].sap = -1; @@ -967,23 +941,20 @@ static int raptor_ipcp_flow_alloc(int fd, return -1; } - flow_set_add(raptor_data.np1_flows, fd); + fset_add(raptor_data.np1_flows, fd); log_dbg("Pending flow with fd %d on SAP %d.", fd, ssap); return 0; } -static int raptor_ipcp_flow_alloc_resp(int fd, - int response) +static int raptor_flow_alloc_resp(int fd, + int response) { struct timespec ts = {0, EVENT_WAIT_TIMEOUT * 1000}; struct timespec abstime; uint8_t ssap = 0; uint8_t r_sap = 0; - uint8_t r_addr[MAC_SIZE]; - - log_info("%s", __FUNCTION__); clock_gettime(PTHREAD_COND_CLOCK, &abstime); @@ -1015,45 +986,39 @@ static int raptor_ipcp_flow_alloc_resp(int fd, } raptor_data.fd_to_ef[fd].sap = ssap; - memcpy(r_addr, raptor_data.fd_to_ef[fd].r_addr, MAC_SIZE); r_sap = raptor_data.fd_to_ef[fd].r_sap; raptor_data.ef_to_fd[ssap] = fd; pthread_rwlock_unlock(&raptor_data.flows_lock); - if (raptor_ipcp_sap_alloc_resp(r_addr, ssap, r_sap, response) < 0) { + if (raptor_sap_alloc_resp(ssap, r_sap, response) < 0) { pthread_rwlock_wrlock(&raptor_data.flows_lock); bmp_release(raptor_data.saps, raptor_data.fd_to_ef[fd].sap); pthread_rwlock_unlock(&raptor_data.flows_lock); return -1; } - flow_set_add(raptor_data.np1_flows, fd); + fset_add(raptor_data.np1_flows, fd); log_dbg("Accepted flow, fd %d, SAP %d.", fd, (uint8_t)ssap); return 0; } -static int raptor_ipcp_flow_dealloc(int fd) +static int raptor_flow_dealloc(int fd) { uint8_t sap; - uint8_t addr[MAC_SIZE]; - - log_info("%s", __FUNCTION__); ipcp_flow_fini(fd); pthread_rwlock_wrlock(&raptor_data.flows_lock); - flow_set_del(raptor_data.np1_flows, fd); + fset_del(raptor_data.np1_flows, fd); sap = raptor_data.fd_to_ef[fd].sap; - memcpy(addr, raptor_data.fd_to_ef[fd].r_addr, MAC_SIZE); bmp_release(raptor_data.saps, sap); raptor_data.fd_to_ef[fd].sap = -1; raptor_data.fd_to_ef[fd].r_sap = -1; - memset(&raptor_data.fd_to_ef[fd].r_addr, 0, MAC_SIZE); raptor_data.ef_to_fd[sap] = -1; @@ -1067,46 +1032,38 @@ static int raptor_ipcp_flow_dealloc(int fd) } static struct ipcp_ops raptor_ops = { - .ipcp_bootstrap = raptor_ipcp_bootstrap, + .ipcp_bootstrap = raptor_bootstrap, .ipcp_enroll = NULL, - .ipcp_reg = raptor_ipcp_reg, - .ipcp_unreg = raptor_ipcp_unreg, - .ipcp_query = raptor_ipcp_query, - .ipcp_flow_alloc = raptor_ipcp_flow_alloc, - .ipcp_flow_alloc_resp = raptor_ipcp_flow_alloc_resp, - .ipcp_flow_dealloc = raptor_ipcp_flow_dealloc + .ipcp_reg = raptor_reg, + .ipcp_unreg = raptor_unreg, + .ipcp_query = raptor_query, + .ipcp_flow_alloc = raptor_flow_alloc, + .ipcp_flow_alloc_resp = raptor_flow_alloc_resp, + .ipcp_flow_dealloc = raptor_flow_dealloc }; int main(int argc, char * argv[]) { - if (ipcp_init(argc, argv, THIS_TYPE, &raptor_ops) < 0) { - ipcp_create_r(getpid(), -1); - exit(EXIT_FAILURE); + if (ipcp_init(argc, argv, &raptor_ops) < 0) { + log_err("Failed to init IPCP."); + goto fail_init; } if (raptor_data_init() < 0) { log_err("Failed to init shim-eth-llc data."); - ipcp_create_r(getpid(), -1); - ipcp_fini(); - exit(EXIT_FAILURE); + goto fail_data_init; } if (ipcp_boot() < 0) { log_err("Failed to boot IPCP."); - ipcp_create_r(getpid(), -1); - raptor_data_fini(); - ipcp_fini(); - exit(EXIT_FAILURE); + goto fail_boot; } if (ipcp_create_r(getpid(), 0)) { log_err("Failed to notify IRMd we are initialized."); ipcp_set_state(IPCP_NULL); - ipcp_shutdown(); - raptor_data_fini(); - ipcp_fini(); - exit(EXIT_FAILURE); + goto fail_create_r; } log_info("Raptor created."); @@ -1126,4 +1083,14 @@ int main(int argc, ipcp_fini(); exit(EXIT_SUCCESS); + + fail_create_r: + ipcp_shutdown(); + fail_boot: + raptor_data_fini(); + fail_data_init: + ipcp_fini(); + fail_init: + ipcp_create_r(getpid(), -1); + exit(EXIT_FAILURE); } diff --git a/src/ipcpd/raptor/raptor_messages.proto b/src/ipcpd/raptor/raptor_messages.proto deleted file mode 100644 index 6891b154..00000000 --- a/src/ipcpd/raptor/raptor_messages.proto +++ /dev/null @@ -1,39 +0,0 @@ -/* - * Ouroboros - Copyright (C) 2016 - 2017 - * - * Raptor messages - * - * Dimitri Staessens <dimitri.staessens@intec.ugent.be> - * Sander Vrijders <sander.vrijders@intec.ugent.be> - * - * 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., 675 Mass Ave, Cambridge, MA 02139, USA. - */ - -syntax = "proto2"; - -enum raptor_msg_code { - FLOW_REQ = 1; - FLOW_REPLY = 2; - NAME_QUERY_REQ = 3; - NAME_QUERY_REPLY = 4; -}; - -message raptor_msg { - required raptor_msg_code code = 1; - optional bytes hash = 2; - optional uint32 ssap = 3; - optional uint32 dsap = 4; - optional uint32 qoscube = 5; - optional sint32 response = 6; -}; |