bb52f0ff2398; misc-next stripe-alloc e77e4c9c2371, testing 3b5b4d424790,
zb64 dcf19c743b68.
+## km1107 (2026-11-07): Mark Korenberg's "btrfs: raid56: pass REQ_META to per-device bios"
+
+Zygo, 2026-11-07: carry Mark's v2 2/2 (posted separately, not re-sent by us)
+as the first commit of the stripe_meta section, since it matters only while
+raid56 metadata is kept alive. If stripe_meta is dropped, the patch is
+dropped with it. The rbio ORs REQ_META from the bios it indexes into the
+per-device bios it builds: data-strip writes, P/Q writes, RMW reads and
+rebuild reads. Scrub and replace rbios stay untagged. A review of our
+parking, pin and cache-steal paths found no path that misses the tag, so
+there is no follow-up patch.
+
+Checks (raid56-wh-harness `forensics/km1107-{6.18,misc-next}/`):
+- New test raid56-req-meta-test.sh (raid5/raid6, stripe_alloc off/on; work,
+ scrub, degraded and replace phases). It classifies every block-layer bio
+ by the chunk type of its dev extent and matches the bios each rbio built.
+ It PASSes on misc-next #70 and 6.18 #320. Without the patch (#69, #319),
+ it FAILs exactly the metadata rbio and rebuild-read checks in all four
+ arms: 0 of 175-241 and 0 of 234-340 bios tagged.
+- Per-commit buildcheck of stripe-alloc, low-space and test-knobs on both
+ lanes: failed=0, no new warnings, tip links rc 0.
+- KASAN + lockdep + ASSERT kernels: stripe_meta tests (meta-ro-commit,
+ meta-crash-replay, meta-reloc-drain, sys-chunk-tear, stripe-meta-toggle)
+ PASS. rb1028 regression set minus qu-p1 PASS (retire-wait-abort SKIP as
+ before), with DATA LOSS 0. Standing counters match the fl1106 baseline.
+
+Not promoted (candidates refs/cand/km1107{,s}/<lane>/*): the 6.18 lane
+must be rebased onto v6.18.62 in the same round (0 conflicts), and that
+rebase moves a branch that is checked out in another worktree.
+
## Not covered / open
Patches without a test of their own (covered at most incidentally by the