| `corrupt-recovery-test.sh` (defaults: raid6 data + raid1c3 metadata, STOCK and STRIPE_ALLOC, 1-2 corrupt devices, corrupt + missing) | read repair (raid6 rebuild, raid1c3 mirror) still recovers, detects and repairs within-tolerance corruption | 1 | ~10 min | PASS on all four (mn lane/cand, 6.18 lane/cand): bad=0 on every case, detectors moved, scrub uncorrectable 0, post-scrub re-read clean | 2026-10-22 | ehive -fi kernels | ddd893f | harness `forensics/q-repair-20261022/*-corrupt-recovery.log` |
| W=1 `fs/btrfs/bio.o` | no new warnings | 1 | - | none on either lane, before or after the patch; bzImage links on all four kernels | 2026-10-22 | mn 2550544f9c6e / c5dc301213ee; 6.18 7626a4a10198 / c3dbf52a35a8 | - | `build-{mn,618}/w1-*.txt` |
| W=1 `make fs/btrfs/` (lane), bio.o (candidate), -fi configs | no new warnings | 1 | - | mn lane 2 pre-existing (stripe_whole_bytes unused, tree-log.c frame size), candidate bio.o 0; 6.18 lane 3 pre-existing, candidate bio.o 0; all four bzImages link | 2026-10-22 | as above (ehive) | - | harness `forensics/q-repair-20261022/build/` |
+| `read-repair-submit-fail-test.sh` RR_ARM=abortgood (**candidate, not promoted**) | after an abort, a read of a sector corrupt on one mirror returns the good copy's data; nothing is written to any member after the abort; the bad copy stays bad | P2 (P3) | ~1 min | lane FAIL as expected on mn lane 0d1010a882d3 (`7.3.0-rc7+ #1`) 3/3 and 6.18 lane 19a5204c6cda (`6.18.57+ #1`) 3/3: EIO, 17 repair lookups failed, 0 writes. Candidate PASS on mn 3e3c635885e4 (`#2`) 3/3: data correct in 0.06-0.10 s, 16 repair lookups, 0 failed, 0 writes, bad copy unchanged; 6.18 176448c38243 (`#2`) 3/3: data correct in 0.07-0.13 s, 0 writes, bad copy unchanged | 2026-10-22 | ehive, KASAN, -fi configs | 96cec4a (4c343a3) | harness `forensics/s-repair-20261022/` |
+| same, RR_ARM=shutdown (**candidate, not promoted**; misc-next only, FS_IOC_SHUTDOWN) | after an emergency shutdown (fs error set, SB_RDONLY clear), a read through an existing mapping is repaired in memory and nothing is written | P3 | ~1 min | lane FAIL as expected on #5 (568c45b64012) 1/1 and mn lane (`#1`) 3/3: 16 write requests after the shutdown, bad copy rewritten. Candidate PASS mn (`#2`) 3/3: data correct, 0 writes. 6.18: SKIP (no shutdown ioctl) | 2026-10-22 | as above | 96cec4a | as above |
+| same, RR_ARM=scrubabort (**candidate, not promoted**) | a scrub running when a log-commit abort forces RO writes no repair after the abort | P4 | ~1.5 min | lane FAIL as expected on #5 1/1 (+1 INVALID, window missed), mn lane (`#1`) 3/3 and 6.18 lane (`#1`) 3/3: 1 write request, the bad sector "fixed up" ~14 s after "forced readonly", the scrub ran to the device end. Candidate PASS mn (`#2`) 3/3: the scrub found the bad sector after the abort and wrote nothing; 6.18 (`#2`) 3/3: the same | 2026-10-22 | as above | 96cec4a | as above |
+| same, RR_ARM=dirtyabort (probe; **not fixed by the candidate**) | data pages dirtied before an abort are not written back after it | none | ~1.5 min | FAIL on every kernel: #5, mn lane, mn cand: 4 write requests after the abort; 6.18 lane: 2; 6.18 cand: 4. See Open | 2026-10-22 | as above | 0b766f2 | as above |
+| regression on the candidates: inject x3, repair x2, corrupt-recovery-test.sh (**candidate, not promoted**) | the existing patch's coverage still holds with P2-P4 | 1-P4 | ~45 min | mn cand (`#2`): inject PASS 3/3 (EIO in 1.07-1.15 s, no oops, unmounted), repair PASS 2/2, corrupt-recovery PASS. 6.18 cand (`#2`): inject PASS 3/3 (EIO in 1.61-1.71 s, no oops, unmounted), repair PASS 2/2, corrupt-recovery PASS. The abort arm is INVALID on both candidates by design (2/2 each): P2 removes its trigger (the repair lookup no longer fails); the inject arm still covers the end_io | 2026-10-22 | as above | 0b766f2 | as above |
+| W=1 `make fs/btrfs/` at every candidate commit (**candidate, not promoted**) | no new warnings | P2-P4, R56 | - | none new on either lane at any commit; only the pre-existing frame-size and stripe_whole_bytes warnings, which reappear when fs.h changes rebuild their objects; all four bzImages link | 2026-10-22 | ehive | - | harness `forensics/s-repair-20261022/build/` |
**read-repair-submit-fail-test.sh, abort arm.**
intact, detected, and repaired by scrub. It covers the normal repair path
the patch touches (repair bios that submit cleanly), not the failure path.
+## Candidate follow-up (s-repair, candidate, not promoted)
+
+Zygo's rule (2026-10-22): no writes after an abort, repair writes
+included; a read may use the commit root to repair in memory but must not
+write the repaired block back. Three patches after the topic patch, and
+one on raid56-fixes:
+- P2 "btrfs: look up a repair read's csums in the same root as the failed
+ read": the repair bbio inherits `csum_search_commit_root`. Fixes
+ f07b855c56b1, Cc 6.18+.
+- P3 "btrfs: do not write back a read repair after a transaction abort":
+ `btrfs_can_submit_repair_write()` (not SB_RDONLY, no fs error, no
+ TRANS_ABORTED) in the repair write; covers data and tree block repair.
+ The abort sets TRANS_ABORTED, then the error, and only then SB_RDONLY;
+ an emergency shutdown (misc-next) never sets SB_RDONLY. Fixes
+ 8ceaad6cd6e7 (and 9b2839451dfc on misc-next), Cc 6.18+.
+- P4 "btrfs: scrub: do not write back repairs after a transaction abort":
+ scrub repair and RAID56 parity scrub never checked the fs state, and a
+ scrub keeps going through every block group after a non-commit abort.
+ Fixes 54765392a1b9, 5a6ac9eacb49, Cc 6.6+.
+- R56 "btrfs: raid56: no RMW write-back of repaired sectors after an
+ abort", on the raid56-fixes topic (refs/cand/s-repair/{mn,618}/r56-topic
+ d575213c1316 / 67a43df4db89) and on top of each zb64. No dedicated test.
+
+Refs: refs/cand/s-repair/mn/topic f2d670d70ed4 (on a0d12f2d99a3),
+mn/zb64 af0f30422e76 (on 0d1010a882d3), 618/topic b53d20733b1a (on
+5943f40e7a76), 618/zb64 e02320b57626 (on 19a5204c6cda). The tested
+kernels were built from 3e3c635885e4 and 176448c38243; the P4 message was
+reworded afterwards, and the trees are identical (927c7ec85a88,
+e9bc363f56fc).
+
+Write detector for the new arms: write, discard and flush requests on the
+loop devices under both members (/sys/block/loopN/stat), from the abort to
+after the unmount, with a negative control in every run (each loop's last
+4 KiB rewritten with its own contents inside the window must be counted;
+2 requests each run, subtracted) and a content oracle (mirror 1's corrupt
+range read back after the unmount).
+
+Paths checked: data read repair (window, P2+P3); tree block repair (csum in
+the header, no root mismatch; same write function, P3); scrub repair and
+parity scrub (real, P4); raid56 RMW write-back (R56); direct IO and encoded
+reads (never use the commit root, so read and repair agree; the original
+lookup can itself fail after an abort); scrub csum lookups (commit root,
+consistent); dev-replace copy (continues on purpose, untouched).
+
## Open
-- Possible follow-up: the repair bbio could inherit `csum_search_commit_root`
- from the failed bbio, so read repair keeps working after an abort as the
- original read does. Not done; the fix above is needed either way, for
- the other submit failures.
+- Data writeback after an abort (Zygo's call): pages dirtied before an
+ abort are still written back after it (dirtyabort: 2-4 write requests on
+ every kernel, lane and candidate). An emergency shutdown blocks
+ that writeback; an abort does not.
+- Dev-replace keeps copying after an abort (deliberate, comment in
+ `__btrfs_handle_fs_error()`). Under the "no writes after abort" rule it
+ is a write; left as is pending Zygo.
+- R56 (raid56 RMW write-back guard) has no test of its own.
+- 6.12 and 6.6 stable lack even the SB_RDONLY check in the repair write
+ (only an ASSERT); P3's Cc says 6.18+ because 8ceaad6cd6e7 is its
+ prerequisite there.
- Not btrfs, found while building the inject arm: a trace kretprobe on
btrfs_search_slot next to fail_function's kprobe on the same function hit
a kprobes core use-after-free when the two were removed (WARNING at