| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Congestion avoidance is a property of the layer, orthogonal to ARQ
and to flow control: FRCP retransmits and lets the peer pace the
sender, per flow, end-to-end; the IPCP paces path aggregates. Each
signal means one thing: a loss triggers a retransmission, a mark
means congestion, the peer window means a slow receiver. Every flow
is paced by the same rate law whatever its QoS, so a greedy raw
sender shares a bottleneck fairly with a reliable stream.
The unit of control is the (destination address, QoS cube)
aggregate: all flows toward that destination share one controller
and one rate; a start-time fair-queuing pacer divides the rate
across them by deadline instead of blocking the send path, and a
new flow rides the aggregate's estimates at its current rate, with
no probing of its own. Slow start runs once per aggregate.
The congestion signal is a multi-bit magnitude: forwarders mark
packets with their standing queue depth, MAX-combined across hops,
so a packet carries the deepest queue on its path. The receiver
feeds back a time-integral mean over a window that adapts to the
flow's byte rate, measuring a slow flow with the same fidelity as a
fast one. The sender runs AIMD scaled by elapsed wall-clock time,
which makes the steady-state allocation RTT-independent.
The PCI gains one byte: the path capacity as a quarter-log2 code.
Forwarders estimate their egress rate from busy-period drain and
MIN-stamp the byte, the receiver returns the window minimum with
its feedback, and the sender scales its rate floor and additive
slope to the bottleneck (C / 32). A deep cut implies a backlogged
bottleneck and a backlogged bottleneck advertises its capacity, so
the scaled floor is live exactly when recovery needs it: the probe
heals a halving in seconds at any link rate, and the floor bounds
the deepest hole to a factor 32 below the bottleneck.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
| |
|
|
|
|
|
|
|
| |
Cancelling a thread that was blocked in ssm_rbuff_read_b/write_b
didn't clean up the n_users reference. If it was in flow_rx_one it
didn't unlock the process lock.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Re-key each encrypted flow's batch root periodically so a long-lived
flow never exhausts or over-uses a single root. The IRMd re-runs the
OAP exchange with the peer IRMd over the flow-update relay. The
per-flow re-keying state is tracked in the registry (reg_flow).
A re-key delivers one root seed from the OAP exchange. keyrot
immediately HKDF-expands it into 128 node keys (KR_NODES_SZ = 128 × 32
B) and wipes the root. Then each of the 128 node keys is itself a root
→ HKDF-expanded into 64 (2^KEY_NODE_BITS) leaf keys, forked per
direction; each leaf key is the actual AEAD key, good for 2^20 packets
(the low counter bits are its nonce/seq). If the number of keys runs
low, a re-key will be triggered (KEY_REKEY_WATERMARK).
The rekey is signalled out of band to the application. The rbuff ACL
is generalized into a flags word, so an RB_REKEY bit rides alongside
the access RB_RD/RB_WR and FLOWDOWN/FLOWPEER bits. The RD and WR bits
are revised ditching the fcntl historical weirdness. The seed is
pulled via flow_read/flow_write, installed with crypt_rekey(). TX
holds the old epoch until the peer is observed on the new one (or a
grace deadline elapses), promoted from both the read and write paths
so a recv-mostly flow still advances.
Also fix the FLOW_ACCEPT and FLOW_ALLOC handlers, which on a key-buffer
allocation failure returned from inside the cleanup-push region: that
leaked the reply message and skipped both the stack-key scrub and the
cleanup pop.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
ssm_rbuff_close used to unmap the SHM page immediately, leaving any
in-flight peer-process thread that was inside pthread_mutex_lock or
pthread_cond_wait on the SHM-resident sync primitives reading freed
memory. Adds an n_users counter, bumped on entry and dropped on exit
of every function that touches the mutex / cond vars (write, write_b,
read, read_b, fini), and have ssm_rbuff_close poll-spin until the
counter drains before tearing down.
ssm_rbuff_read now re-checks IS_EMPTY after taking the mutex, plugging
a TOCTOU where two readers could both pass a lock-free fast path and
the loser would read a stale TAIL.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
| |
|
|
|
|
|
|
|
|
| |
The shared memory pool is now offset based instead of block
index-based like the old shm_rdrbuff allocator. This renames the API
more consistently. Also changes variables names to off instead of idx
for consistency.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
| |
|
|
|
| |
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
This is a first step towards the Secure Shared Memory (SSM)
infrastructure for Ouroboros, which will allow proper resource
separation for non-privileged processes.
This replaces the rdrbuff (random-deletion ring buffer) PoC allocator
with a sharded slab allocator for the packet buffer pool to avoid the
head-of-line blocking behaviour of the rdrb and reduce lock contention
in multi-process scenarios. Each size class contains multiple
independent shards, allowing parallel allocations without blocking.
- Configurable shard count per size class (default: 4, set via
SSM_POOL_SHARDS in CMake). The configured number of blocks are
spread over the number of shards. As an example:
SSM_POOL_512_BLOCKS = 768 blocks total
These 768 blocks are shared among 4 shards
(not 768 × 4 = 3072 blocks)
- Lazy block distribution: all blocks initially reside in shard 0
and naturally migrate to process-local shards upon first
allocation and subsequent free operations
- Fallback with work stealing: processes attempt allocation from
their local shard (pid % SSM_POOL_SHARDS) first, then steal
from other shards if local is exhausted, eliminating
fragmentation while maintaining low contention
- Round-robin condvar signaling: blocking allocations cycle
through all shard condition variables to ensure fairness
- Blocks freed to allocator's shard: uses allocator_pid to
determine target shard, enabling natural load balancing as
process allocation patterns stabilize over time
Maintains existing robust mutex semantics including EOWNERDEAD
handling for dead process recovery. Internal structures exposed in
ssm.h for testing purposes. Adds some tests (pool_test,
pool_sharding_test.c. etc) verifying lazy distribution, migration,
fallback stealing, and multiprocess behavior.
Updates the ring buffer (rbuff) to use relaxed/acquire/release
ordering on atomic indices. The ring buffer requires the (robust)
mutex to ensure cross-structure synchronization between pool buffer
writes and ring buffer index publication.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
