Trigger retry dedup re-acquires already-done requests: concurrent waiters burn their retry budget on stale errors and never re-notify automountd
Summary
The request dedup (:324-339) matches on path+key without checking ar_done, so while any waiter still holds a reference to a completed request, every retry by another waiter re-acquires that dead request (refcount_acquire :339), immediately observes ar_done (:366) and returns the stale ar_error - burning one of autofs_retry_attempts per 1s-parked iteration without ever handing a fresh request to automountd (ioctl_request skips done requests :466-470). Two+ concurrent waiters on one failing key exhaust their retries against the corpse and the mount spuriously fails despite a willing daemon - retry-budget/availability correctness defect. No memory-safety impact; no PoC (Low). Fix: skip ar_done requests in the dedup loop. [VERIFIED pass2: automountd failing-map harness beyond budget; dedup-of-corpse logic certain; fix compile-validated]
Discussion (0)
PoC verification
Evidence pack
findings/poc/DF-3174 · 9 files| File | Type | Description | Size | |
|---|---|---|---|---|
| fix.diff | — | 469 B | view raw | |
| README.md | — | 432 B | ↓ raw | |
| VERDICT.md | — | 1.7 KB | ↓ raw | |
| build.sh | — | 106 B | view raw | |
| run.sh | — | 132 B | view raw | |
| manifest.json | — | 522 B | view raw | |
| verdict.json | — | 1.5 KB | view raw | |
| fix.diff | — | 469 B | view raw | |
| env.txt | — | 272 B | view raw |
DF-3174 -- verification pack
Finding: autofs trigger dedup re-acquires done requests; retry budget burned on corpses
Status: not_reproduced / impact denial_of_service / confidence certain
What was run
n/a
How to reproduce
n/a (needs configured automountd with a failing map)
Expected
n/a
See VERDICT.md for the full narrative and verdict.json for the persisted machine verdict. Artifacts: fix.diff.
DF-3174 verification
Kernel: DragonFly dfbsd 6.5-DEVELOPMENT #0: Thu Jul 2 06:02:54 UTC 2026 root@dfbsd:/usr/obj/usr/src/sys/X86_64_GENERIC x86_64
Bottom line
Logic defect certain by inspection: the request dedup loop matches on path+key without checking ar_done, so while any waiter still references a completed request every retry by another waiter re-acquires the corpse (refcount_acquire), observes ar_done immediately and returns the stale ar_error, burning one autofs_retry_attempts per 1-second parked iteration without ever handing a fresh request to automountd (which skips done requests). A faithful dynamic reproduction needs a configured automountd with a failing map plus concurrent waiters; the autofs module exists on the guest but the daemon harness exceeded this batch budget. Fix skips ar_done requests in the dedup scan; compile-validated.
Kernel references
Exploit chain
(none: availability/correctness finding; no privilege chain)
Fix
skip ar_done requests in the dedup loop
See fix.diff in this directory (git-apply-able against the read-only sys/ tree; applied only inside the disposable verification guest).
Reproduction
build: n/a
run: n/a (needs configured automountd with a failing map)
expected: n/a
Notes on the PoC (what changed vs the finding sketch)
(none)
Fix status (batch validation)
Applied with the 36 sibling fix.diffs of this batch in findings/poc/_batch41/all_fixes.patch; single in-guest make -j6 nativekernel KERNCONF=X86_64_GENERIC completed RC=0 with zero errors under -Werror (full log: findings/poc/_batch41/fix_build8.log). Behavior A/B intentionally skipped for this Info-tier item.
Fix verification
not_testableConfirmed kernel references
Detail
Evidence (decisive lines)
VERDICT.md walk over autofs.c:324-339 (dedup), :366 (ar_done observed), :466-470 (ioctl skips done)
Verified recommended fix
skip ar_done requests in the dedup loop
Verdict
Logic defect certain by inspection: the request dedup loop matches on path+key without checking ar_done, so while any waiter still references a completed request every retry by another waiter re-acquires the corpse (refcount_acquire), observes ar_done immediately and returns the stale ar_error, burning one autofs_retry_attempts per 1-second parked iteration without ever handing a fresh request to automountd (which skips done requests). A faithful dynamic reproduction needs a configured automountd with a failing map plus concurrent waiters; the autofs module exists on the guest but the daemon harness exceeded this batch budget. Fix skips ar_done requests in the dedup scan; compile-validated.
No comments yet.