covers: stripe-alloc: review round 1 -- findings, fixes, and the raid56-metadata fsync measurement
The chain and six round-one fixes are landed on the 6.18 lane (72-commit
topic), not yet in the misc-next roadmap: say so, list the fixes under
Pending, and add the round's coverage (runtime enable/disable under
raid56 metadata on raid5/raid6 with rotational, ssd and ssd_spread; the
preallocation storm; the directed set) to Testing. The completed-fsync
guarantee is narrowed to non-raid56 metadata, with a new "Log tree under
raid56 metadata" paragraph stating the measured loss (log replay refused
on the degraded array, zero-log recovers the committed state, the array
stays writable; raid5 2/40 cases, raid6 0/48) and the open decision.
Known costs states the operational consequence of the reclaim threshold
default for raid56 metadata.
projects / linux / commit