Rongsong Shen
d26ff8870d
The buglog was one 27K-line JSON file whose fields covered only symptom,
root cause and fix, as free text: 2% of entries named the code version a
bug was found on, 12% the build profile, 36% a way to reproduce, and 58% of
the file was OpenWolf hook auto-detections ("Missing guard clause",
"Function not marked async" on .md files).
- .claude/skills/buglog/bug-record.md: the record standard. Eight
sections (Summary as the title, Description, Environment, Reproduction;
Root cause, Fix, Verification, References once fixed), each with the
content it must contain -- e.g. Environment needs the OS, the libc the
failing binary loads (from its ELF interpreter: glibc 2.42 in the nix
shell vs 2.40 on the host), toolchain, build profile and commit.
- .wolf/bugs/<id>.md: 500 records migrated from buglog.json; every field
verified to round-trip. They carry legacy: true and an explicit "not
recorded" where the old log had no such field; component derived from
file paths, severity from keywords (else unknown), fix_commits from
hashes in status_note/fix that git resolves (91 of 463 fixed).
- .wolf/buglog-archive.json: the 710 hook auto-detections, verbatim.
- .wolf/buglog.json is now an index generated from the records, keeping
the fields OpenWolf's hooks and `openwolf bug search` read.
- scripts/buglog.py rewritten: new (template + Environment collected from
the machine and --binary), check, status (refuses a status the record
doesn't support; fixed needs Root cause, Fix, Verification and
--commit hashes git resolves), note, recur, set, show, list, triage,
env, index. 29 unittest cases in scripts/tests/test_buglog.py.
- scripts/git-hooks/pre-commit runs `check --staged`: incomplete staged
records, a legacy record hand-changed to fixed without the fix-time
sections, a new record claiming legacy: true, or a stale index refuse
the commit.
- scripts/gitea-sync.py reads the records: issue body = the record's
sections plus linked fix commits (origin remote, GITEA_COMMIT_REPO to
override), and a body hash so an issue's text is refreshed when the
record changes.
- bug-1866 rewritten to the full standard as the worked example.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
5 KiB
| id | status | status_updated | status_note | component | severity | timestamp | last_seen | occurrences | files | tags | related | legacy | ||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| bug-1795 | fixed | 2026-09-24 | controller | crash | 2026-09-09T20:10:00Z | 2026-09-19T09:00:00Z | 1 |
|
|
|
true |
bug-1795: leave_ctx() (test.c, called by cmd_up/cmd_quit) could free a connection's api_ctx_t…
Description
leave_ctx() (test.c, called by cmd_up/cmd_quit) could free a connection's api_ctx_t (xcecli_stop()+xcecli_ctx_fini(), src/lib/client.c) while an abandoned worker thread from a prior Ctrl-C-interrupted RPC on that SAME connection was still running and still reading/writing it -- a genuine use-after-free, found by kiro-review during xce-test-cli-ux task 3.2's own round-1 review (not reproduced live/under ASAN on that run; identified structurally by tracing the code paths).
Environment
Not recorded (migrated from buglog.json).
Reproduction
Not recorded (migrated from buglog.json).
Root cause
Task 3.2's own interruptible-call adapter (tasks 1.4/1.5, interruptcall.c) runs every blocking xce_* RPC on a detached worker thread; when the operator interrupts (Ctrl-C) a blocked call, xcetest_run_interruptible() returns -EINTR to the REPL immediately, but the worker thread it spawned keeps running -- possibly for up to its own ~10s cbuf_timedget deadline -- reading/writing whatever connection state its ctx closes over. Before this task, the REPL was fully synchronous, so leave_ctx() (test.c, shared by cmd_up's '..' and cmd_quit) could never run while an RPC was outstanding on that connection; task 3.2's adapter broke that invariant by letting control return to the prompt WHILE a worker is still abandoned-but-running. leave_ctx() already called xcecli_wakeup(active_ctx->ctx, -EINTR) before tearing the connection down, but that is only a best-effort nudge (it wakes the underlying condvar wait sooner than the worker's own deadline) -- nothing forced leave_ctx() to actually WAIT for that woken worker to finish before proceeding to xcecli_stop() (tears down the live socket + failover lock) and xcecli_ctx_fini() (unconditionally free()s the whole api_ctx_t). If the operator's very next action after abandoning a hung command is 'up' or 'quit' -- an entirely ordinary thing to do -- the still-running worker can be mid-read/write on memory that just got freed out from under it. This directly contradicts the Interruptible Call Adapter's own documented invariant ('at most one worker thread -- running or abandoned-but-still-running -- is ever active per connection at a time', interruptcall.h/design.md) -- that invariant was enforced only BETWEEN xcetest_run_interruptible() calls (via the per-connection semaphore registry, tasks 1.4/1.5), never against connection TEARDOWN, since leave_ctx() never itself goes through xcetest_run_interruptible().
Fix
FIXED (2026-09-09, xce-test-cli-ux task 3.2 remediation round 1). Added xcetest_wait_connection_idle(void* conn_key) to interruptcall.c/.h, reusing the EXACT SAME per-connection semaphore registry lookup (xcetest_conn_lock_for()) that xcetest_run_interruptible() already builds and maintains -- not a new/parallel mechanism. It does a plain BLOCKING sem_wait() (deliberately not the interruptible sem_trywait()-poll variant used elsewhere -- this is a narrow, safety-critical wait where correctness outranks Ctrl-C responsiveness, and it resolves quickly in practice since xcecli_wakeup() already nudges the worker) then immediately sem_post()s the semaphore back rather than holding it -- the semantic is 'prove no worker is active on this connection right now', not 'hold the connection locked'. leave_ctx() now calls xcetest_wait_connection_idle(ctx->ctx) right after the existing xcecli_wakeup() call and right before xcecli_stop()/xcecli_ctx_fini() -- since both cmd_up and cmd_quit call this SAME shared leave_ctx(), both paths are covered by a single call site. New black-box regression test test_interrupt_then_immediate_teardown_no_uaf (test_test_cli.c, reuses the existing clitest_start_blackhole_listener() harness helper) interrupts a blocked hello RPC against a black-hole TCP peer and sends 'up' IMMEDIATELY afterward -- with no settling wait for the interrupted prompt to redisplay first, unlike the sibling interrupt test -- to race leave_ctx()'s teardown against the abandoned worker on purpose; asserts the session exits only via a normal (non-signal) status and that stderr never contains an 'AddressSanitizer' diagnostic anywhere across the whole session. Verified under the asan build profile.
Verification
Not recorded (migrated from buglog.json).
References
- Related: bug-1794