Mooncake/mooncake-transfer-engine/include
Jiangtian Feng 9153043b86
[TE] Harden TCP transport: validate remote addresses, fix idle cleanup, add TCP_NODELAY (#2314)
* [TE] Harden TCP transport: validate remote addresses, fix idle cleanup, add TCP_NODELAY

Fixes #2313

1. Security: ServerSession now validates remote-supplied memory addresses
   against registered local buffers before use, preventing arbitrary
   memory read/write from malicious peers.

2. Performance: Set TCP_NODELAY on all server-accept and client-connect
   paths to eliminate Nagle-induced latency on small control messages.

3. Correctness: Call io_context.restart() after exception in worker loop
   to prevent busy-spin when io_context enters stopped state.

4. Correctness: Fix cleanupIdleConnections to scan the full deque instead
   of only the back, so idle connections anywhere in the pool get cleaned.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* [TE] Fix clang-format violations in TCP transport

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-06-08 10:29:09 +08:00
..
common [TE]: Add HIP transport for AMD GPUs support (#1208) 2025-12-16 19:27:47 +08:00
gpu_vendor [TE] add device API support (#2333) 2026-06-05 16:00:07 +08:00
transport [TE] Harden TCP transport: validate remote addresses, fix idle cleanup, add TCP_NODELAY (#2314) 2026-06-08 10:29:09 +08:00
CMakeLists.txt [TE] add device API support (#2333) 2026-06-05 16:00:07 +08:00
ascend_allocator.h refactor ascend direct transport (#1720) 2026-03-24 21:51:58 +08:00
common.h [TE] Fix error-path safety: freeaddrinfo leak, null deref, OOB read (#2343) 2026-06-08 10:00:45 +08:00
config.h [TE] fix: improve auto gid selection and retry (#2269) 2026-06-03 20:52:24 +08:00
cuda_alike.h [] feat: add Hygon DCU/DTK and Iluvatar CoreX platform support (#2118) 2026-05-18 10:26:56 +08:00
error.h [TransferEngine] Refactor code to hide transport logics from user APIs (#51) 2025-01-02 16:38:24 +08:00
memory_location.h [Store] fix: restrict NUMA-segmented allocation to RDMA protocol 2026-03-12 18:26:11 +08:00
multi_transport.h [Store] auto-enable MC_STORE_MEMCPY in TCP-only environments (#1936) 2026-04-21 15:01:46 +08:00
topology.h [TransferEngine] Mitigating performance overhead from large cluster and large bulks (#999) 2025-11-03 11:39:36 +08:00
transfer_engine.h [TE] add device API support (#2333) 2026-06-05 16:00:07 +08:00
transfer_engine_c.h [TE] PTE-aware auto-split large MR registration for EFA transport (#1912) 2026-04-20 15:04:49 +08:00
transfer_engine_impl.h [TE] Fix lock leak, missing transport_, and empty entries UB (#2349) 2026-06-08 10:06:20 +08:00
transfer_metadata.h Fix P2PHANDSHAKE in dual-NIC container setups via MC_RDMA_BIND_ADDRESS (#2280) 2026-06-01 16:49:58 +08:00
transfer_metadata_plugin.h [PG][TE][TENT] Add dedicated peer liveness probe for recovery and enable elastic GPU test (#1808) 2026-04-08 09:56:58 +08:00
ub_allocator.h [Store] enables the Ubtransport for Mooncake Store And optimize UrmaEndpoint (#2196) 2026-06-01 10:29:39 +08:00