H5 Security review #58

Closed
opened 2026-11-09 05:29:42 +00:00 by claude · 1 comment
Owner

Depends on: N9, N6, N7, A10

Review every network parser, the list server, file handling, server config and the container setup.

Acceptance

  • docs/SECURITY_REVIEW.md lists scope, findings and fixes.
  • High and critical findings are fixed, with regression tests.
  • The repo contains no secrets, internal hostnames or private network details.
**Depends on:** N9, N6, N7, A10 Review every network parser, the list server, file handling, server config and the container setup. **Acceptance** - `docs/SECURITY_REVIEW.md` lists scope, findings and fixes. - High and critical findings are fixed, with regression tests. - The repo contains no secrets, internal hostnames or private network details.
claude added this to the v0.9 Hardening milestone 2026-11-09 05:29:42 +00:00
claude added the priority/hightype/chore labels 2026-11-09 05:29:42 +00:00
claude added this to the Tidewall v1.0 project 2026-11-09 15:00:09 +00:00
eric closed this issue 2026-11-14 07:36:41 +00:00
Author
Owner

Done in 9b5064d. The review is in docs/SECURITY_REVIEW.md: scope, nine findings, and what was fixed or accepted.

Fixed

  • High: a peer that connects and then stops answering kept its ENet slot for up to 15 s after being turned away, so one address could fill every slot and lock players out. Now flood refusals disconnect at once (the reason is in the disconnect data, and NetClient now accepts REASON_BUSY from it). A peer told to leave gets 1 s to acknowledge. When slots run short, the guest that has waited longest for its HELLO makes room.
  • Medium: an IPv6 sender rotating through its /64 counted as many sources. SourceLimiter.source_of() now groups a /64, and the limiter, handshakes per source and the list server's per-address cap all use it.
  • Low: CommandLog raised script errors on a header with non-numeric fields. It now refuses the file instead.
  • tools/godot.sh now unsets DISPLAY and WAYLAND_DISPLAY. This run showed that a headless engine still pops its error alert on the desktop through zenity.

Mitigated, not fixed: ENet takes a slot for each CONNECT before the sender proves its address, and never reports those slots to the script, so forged CONNECTs can only be outlasted. The host now has 512 slots (NetServer.peer_slots). This was not tested against real forged traffic.

Accepted, recorded: a hostile server can make a client use memory; names aren't authenticated (there are no accounts); a client can always claim the maximum view lag of 0.5 s; release builds exit on a null call; the base image isn't pinned by digest.

Repo: git grep over every tracked file and git log -p over the whole history found no keys, tokens, internal hosts, private addresses or home paths. One fixture's 10.0.0.9 is now 192.0.2.9.

Tests: 4 new handshake tests that use real ENet peers which connect and then never service their socket, plus test_source_limiter (3 tests) and a CommandLog type test. make check (645 tests), make fuzz (1,000,000 messages), make client-check, make discovery-check and make soak-selfcheck all pass. I didn't run a full 24-client make soak this run; H4's soak will cover it.

— tidewall-build

Done in 9b5064d. The review is in `docs/SECURITY_REVIEW.md`: scope, nine findings, and what was fixed or accepted. **Fixed** - **High:** a peer that connects and then stops answering kept its ENet slot for up to 15 s after being turned away, so one address could fill every slot and lock players out. Now flood refusals disconnect at once (the reason is in the disconnect data, and `NetClient` now accepts `REASON_BUSY` from it). A peer told to leave gets 1 s to acknowledge. When slots run short, the guest that has waited longest for its HELLO makes room. - **Medium:** an IPv6 sender rotating through its /64 counted as many sources. `SourceLimiter.source_of()` now groups a /64, and the limiter, handshakes per source and the list server's per-address cap all use it. - **Low:** `CommandLog` raised script errors on a header with non-numeric fields. It now refuses the file instead. - `tools/godot.sh` now unsets `DISPLAY` and `WAYLAND_DISPLAY`. This run showed that a headless engine still pops its error alert on the desktop through zenity. **Mitigated, not fixed:** ENet takes a slot for each CONNECT before the sender proves its address, and never reports those slots to the script, so forged CONNECTs can only be outlasted. The host now has 512 slots (`NetServer.peer_slots`). This was not tested against real forged traffic. **Accepted, recorded:** a hostile server can make a client use memory; names aren't authenticated (there are no accounts); a client can always claim the maximum view lag of 0.5 s; release builds exit on a null call; the base image isn't pinned by digest. **Repo:** git grep over every tracked file and `git log -p` over the whole history found no keys, tokens, internal hosts, private addresses or home paths. One fixture's 10.0.0.9 is now 192.0.2.9. **Tests:** 4 new handshake tests that use real ENet peers which connect and then never service their socket, plus `test_source_limiter` (3 tests) and a CommandLog type test. `make check` (645 tests), `make fuzz` (1,000,000 messages), `make client-check`, `make discovery-check` and `make soak-selfcheck` all pass. I didn't run a full 24-client `make soak` this run; H4's soak will cover it. — tidewall-build
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: claude/tidewall#58