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>
39 lines
3.4 KiB
Markdown
39 lines
3.4 KiB
Markdown
---
|
|
id: "bug-1600"
|
|
status: "fixed"
|
|
status_updated: "2026-09-24"
|
|
status_note: "ddee884c"
|
|
component: "lib"
|
|
severity: "crash"
|
|
timestamp: "2026-08-19T21:05:00Z"
|
|
last_seen: "2026-09-19T09:00:00Z"
|
|
occurrences: 8
|
|
files: ["src/lib/group.c"]
|
|
tags: ["scheduler", "crash", "sigabrt", "heap-corruption", "bad-free", "allocator-mismatch", "rtti", "makeString", "group.c", "config-push", "failover", "production"]
|
|
related: ["bug-1503"]
|
|
fix_commits: ["ddee884c"]
|
|
legacy: true
|
|
---
|
|
|
|
# bug-1600: xce_scheduler crashed with free(): invalid pointer / ASAN bad-free during controller failover…
|
|
|
|
## Description
|
|
xce_scheduler crashed with free(): invalid pointer / ASAN bad-free during controller failover retry (3x config-push retry loop); 6 crash-loop occurrences on 2026-08-17, 2 more on 2026-08-19 (core dumps confirmed via gdb bt full and coredumpctl, crash site consistently makeGroup() at src/lib/group.c:234)
|
|
|
|
## Environment
|
|
_Not recorded (migrated from buglog.json)._
|
|
|
|
## Reproduction
|
|
_Not recorded (migrated from buglog.json)._
|
|
|
|
## Root cause
|
|
makeGroup() existing-group update path (the branch taken when a config-push re-applies a group name that already exists, e.g. every retried/duplicate config push) freed xce_g_member_t.u.name via FREEUP() (raw free()). u.name is allocated by makeString() (make_g_member(), group.c:77), an rtti-managed pointer returned by makeInstance()/objAlloc() that is NOT the malloc()-returned base address (it is an offset into a combined header+data block) and is refcounted, not malloc-owned. Calling free() on it is undefined behavior -> heap corruption, surfacing later as glibc/ASAN bad-free aborts on unrelated allocations. group_killer() (same file) already released the identical field correctly via dereference(). The buggy FREEUP() path is hit on every config re-push for a group with existing members (e.g. controller failover retrying a config push), which explains the crash correlating with the 3-attempt xcecli_retry_next() failover loop even though no thread concurrency is required to trigger it.
|
|
|
|
## Fix
|
|
Changed both FREEUP(m->u.name) call sites inside makeGroup()'s existing-group branch (group.c, the g->members loop and the g->ex_members loop) to dereference(m->u.name), matching group_killer()'s established correct pattern for releasing rtti makeString() pointers. Added regression test test_reconfigure_same_group_members (src/lib/testsuite/test_group.c) that calls makeGroup() twice with the same group name/members, reproducing the exact ASAN bad-free at group.c:234 pre-fix (confirmed RED), and passing clean post-fix (confirmed GREEN, full lib asan suite Ok:1 Fail:0, no regressions). NOTE: a separate, related latent issue was also found but NOT fixed here (out of scope, no concurrency needed to explain the observed crash) - the entire existing-group update branch in makeGroup() only takes RTTI_LOCK_READ (a genuine shared pthread_rwlock_rdlock) while mutating shared group state, and sched_init_config()/plugin_config_update() (the config-push handler chain) hold no serializing lock at all, so two vtask worker threads (scheduler runs a 4-vcpu pool) could in principle race on the same group concurrently -- most plausibly triggered by xce_ctx_reconnect() reusing the same endpoint struct and endpoint_init() resetting ep->busy=0 before an already-dispatched handler finishes. This should be tracked as a follow-up.
|
|
|
|
## Verification
|
|
_Not recorded (migrated from buglog.json)._
|
|
|
|
## References
|
|
- Related: bug-1503
|