- u_async.c: process immediate_queue BEFORE epoll_wait in uasync_poll
- stcp_link.c: defensive NULL etcp_conn/etcp_link in stcp_link_close
- utun_instance.c: move stcp_server_list_destroy_all to Phase J (before ETCP),
add deferred drain in both Phase J and L
- stcp.h/c: STCP_HS_ENC_CLIENT 6→38, STCP_HS_ENC_SERVER 7→39,
exchange ed25519_pubkey in both handshake directions
- etcp_connections.c: copy peer_ed25519 from STCP to ETCP_CONN in
etcp_link_enter_ready_tcp; guard link->etcp in tcp_link_close_cb
- chat_sync.c: fix lk->conn NULL dereference for TCP links
- test_stcp.c: update stcp_server_create/stcp_client_connect callsites
- tools/chat_tcp_test: change transport udp→tcp
- Added CFG_ALLOWED_KEY struct with binary key storage (64 bytes) for fast memcmp comparison
- Implemented parsing of [allowed_keys] section with allow_all flag and per-key entries
- Added key validation in etcp_connections.c server handler (errorcode=8 for rejected keys)
- Default behavior: deny all if [allowed_keys] section is missing or allow_all=0 with no keys
- Added proper memory cleanup and debug logging for allowed_keys configuration
- Updated all test configs to include [allowed_keys] allow_all=1 for backward compatibility
- Add Windows implementation of default_CSPRNG() using CryptGenRandom
- Add fallback stub for platforms without CSPRNG support
- Fix build.sh to properly detect build failures (use PIPESTATUS)
- Show proper error message instead of "Build completed successfully" on failure
- Create test_utils.h with cross-platform utility functions:
* test_mkdtemp() - Windows uses GetTempPath + _mkdir, Linux uses mkdtemp
* test_unlink() - Windows _unlink, Linux unlink
* test_rmdir() - Windows _rmdir, Linux rmdir
- Update all test files to use test_utils.h
- Replace arpa/inet.h and netinet/in.h with platform_compat.h
- Replace unistd.h with conditional includes
- Replace mkdtemp(), unlink(), rmdir() with test_* versions
- Tests now build on both Linux and Windows
- Note: Some integration tests may fail on Linux due to test environment
Changed from global bindings to per-instance bindings:
1. Added struct ETCP_BINDINGS in etcp_api.h:
- Contains array of callbacks[ETCP_MAX_BINDINGS]
- NULL means not bound
2. Updated UTUN_INSTANCE (utun_instance.h):
- Added field: struct ETCP_BINDINGS api_bindings
- Initialized to NULL by calloc in utun_instance_create
3. Updated API functions (etcp_api.h/c):
- etcp_bind(inst, id, callback) - per-instance binding
- etcp_unbind(inst, id) - per-instance unbinding
- Removed etcp_api_init/etcp_api_deinit (not needed)
- etcp_int_recv now uses conn->instance->api_bindings.callbacks[id]
4. Updated test_etcp_api.c:
- Separate callbacks for server and client
- Register via etcp_bind(server_instance, ...) and etcp_bind(client_instance, ...)
- Removed global etcp_api_init/etcp_api_deinit calls
Test passes: All 100 packets transmitted in each direction.
etcp_bind uses global table - multiple binds with same ID overwrite.
Fixed by using single recv_callback that checks conn pointer:
- conn == server_conn: forward direction (client->server)
- conn == client_conn: backward direction (server->client)
Test now passes: All 100 packets transmitted in each direction.
Based on test_pkt_normalizer_etcp.c but uses high-level API:
- etcp_send() for sending packets via ETCP normalizer
- etcp_bind() for registering receive callbacks
- etcp_api_init()/etcp_api_deinit() for API lifecycle
NOTE: Test has current limitation - pkt_normalizer packer aggregates
packets into ~1536 byte chunks, exceeding ETCP max payload of 1480 bytes.
This causes encrypt_send to fail. The test demonstrates correct API usage
pattern but requires packer fix for full functionality.
evgeny