2.7 KiB
| id | status | status_updated | status_note | component | severity | timestamp | last_seen | occurrences | files | tags | related | fix_commits | legacy | gitea_issue | gitea_synced_status | gitea_synced_hash | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| bug-1862 | fixed | 2026-09-25 | c29518b2 |
worker | crash | 2026-09-24T15:24:57Z | 2026-09-24T15:24:57Z | 1 |
|
|
|
true | 281 | fixed | 91d19796eef1 |
bug-1862: xcecd_virtual ASAN heap-use-after-free in find_worker_by_jobid (xcecd_worker_utils.c:702) <…
Description
xcecd_virtual ASAN heap-use-after-free in find_worker_by_jobid (xcecd_worker_utils.c:702) <- xcecd_resync_classify_job (xcecd_resync.c:33) <- xcecd_resync_build_report <- handle_jobstatequery (controller_connector.c:1662) on a vtask thread; freed by remove_worker (xcecd_worker_utils.c:283) <- cleanup_worker_list <- main loop (main.c:652). Worker aborted mid chaotic HA soak (2026-09-24, run 1 of 3), so later rounds lost worker-resync coverage.
Environment
Not recorded (migrated from buglog.json).
Reproduction
Not recorded (migrated from buglog.json).
Root cause
The worker's live job list is mutated by the main loop (add_worker / cleanup_worker_list -> remove_worker) with no lock, while the controller's OPCode_jobStateQuery is answered on a vtask thread (myreq_handler) that walks the same list. Any rejoin resync overlapping a job reap reads a freed node.
Fix
UPDATE (2026-09-25): Fix (uncommitted): xcecd_worker_utils.c gains a worker-list mutex (xcecd_worker_list_lock/unlock). add_worker takes it around linking a fully-built entry; remove_worker unlinks under it and frees after. Readers on vtask threads: handle_jobstatequery holds it from reading the list head through the whole query page (report build walks the list several times; the send is a non-blocking xce_request); the job-control handler copies job_status under it and re-finds the entry under it to set killed_signal; remote task start strdup's the five fields it sends under it and frees them after xcecli_runtask. The finished-job retention store (appended by the main loop, read by resync) gets its own internal mutex via public wrappers around *locked bodies. Main-thread-only traversals (uv_ipc.c find_worker_by_pid, cleanup/update) need no lock. Test: test_xcecd_resync.c test_build_report_safe_against_concurrent_list_mutation (mutator thread add/remove/retain vs 20000 locked report builds) -- heap-use-after-free in find_worker_by_jobid 3/3 with no-op lock stubs, clean 5/5 with the fix; worker suite 10/10.
Verification
Not recorded (migrated from buglog.json).
References
- Gitea: #281