md/raid5: Fix bio retry on interrupted reshape
PATCH 1/1 md/raid5: Fix bio retry on interrupted reshape
When a bio encounters LOC_INSIDE_RESHAPE during a reshape that is interrupted (stopped or unable to progress), the code sets bi->bi_status = BLK_STS_RESOURCE to signal the block layer for retry. However, bio_endio() is never called, so the block layer never receives the completion notification and the retry never happens.
This causes I/O to hang when a filesystem is layered over RAID5 and reshape gets stuck.
Fix this by calling bio_endio(bi) before md_free_cloned_bio(bi) so the block layer is properly notified of the BLK_STS_RESOURCE status and can retry the request.
Tested stripes and stripe size conversions under load comparing files multiple times during each conversion (i.e. MD reshape) on ext4 after dropping caches degrading the RaidLV each time and no data corruption.
---
drivers/md/raid5.c | 1 +
1 file changed, 1 insertion(+)
drivers/md/raid5.c +0 -0
@@ -6217,6 +6217,7 @@ static bool raid5_make_request(struct mddev *mddev, struct bio * bi)
mempool_free(ctx, conf->ctx_pool);
if (res == STRIPE_WAIT_RESHAPE) {
+ bio_endio(bi);
md_free_cloned_bio(bi);
return false;
}
Signature · 1 line
2.47.3
Dear Nigel,
Thank you for your patch.
Am 29.04.26 um 13:10 schrieb Nigel Croxon:
Quoted 9 lines
When a bio encounters
LOC_INSIDE_RESHAPEduring a reshape that is interrupted (stopped or unable to progress), the code setsbi->bi_status=BLK_STS_RESOURCEto signal the block layer for retry. However,bio_endio()is never called, so the block layer never receives the completion notification and the retry never happens.This causes I/O to hang when a filesystem is layered over RAID5 and reshape gets stuck.
Quoted 7 lines
Fix this by calling
bio_endio(bi)beforemd_free_cloned_bio(bi)so the block layer is properly notified of theBLK_STS_RESOURCEstatus and can retry the request.Tested stripes and stripe size conversions under load comparing files multiple times during each conversion (i.e. MD reshape) on ext4 after dropping caches degrading the RaidLV each time and
I thought RaidLV misspelled Raid V (Raid 5), so should you resend, maybe write it as RAID LV.
Quoted 3 lines
no data corruption.
Which paragraph/comment exactly?
Quoted 17 lines
Signed-off-by Nigel Croxon <[email protected]>--- drivers/md/raid5.c | 1 + 1 file changed, 1 insertion(+)drivers/md/raid5.c +1
-0@@ -6217,6 +6217,7 @@ static bool raid5_make_request(struct mddev *mddev, struct bio * bi) mempool_free(ctx, conf->ctx_pool); if (res == STRIPE_WAIT_RESHAPE) { + bio_endio(bi); md_free_cloned_bio(bi); return false;
Quoted 1 line
}
Kind regards,
Paul
Looks like this was just fixed by Benjamin Marzinski's
commit 418b3e64e4459 md/raid5: Fix UAF on IO across the reshape position
So my patch is not needed anymore.
-Nigel
Quoted 57 lines · Paul Menzel
Dear Nigel,
Thank you for your patch.
Am 29.04.26 um 13:10 schrieb Nigel Croxon:
Quoted 9 lines
When a bio encounters
LOC_INSIDE_RESHAPEduring a reshape that is interrupted (stopped or unable to progress), the code setsbi->bi_status=BLK_STS_RESOURCEto signal the block layer for retry. However,bio_endio()is never called, so the block layer never receives the completion notification and the retry never happens.This causes I/O to hang when a filesystem is layered over RAID5 and reshape gets stuck.
Quoted 7 lines
Fix this by calling
bio_endio(bi)beforemd_free_cloned_bio(bi)so the block layer is properly notified of theBLK_STS_RESOURCEstatus and can retry the request.Tested stripes and stripe size conversions under load comparing files multiple times during each conversion (i.e. MD reshape) on ext4 after dropping caches degrading the RaidLV each time and
I thought RaidLV misspelled Raid V (Raid 5), so should you resend, maybe write it as RAID LV.
Quoted 3 lines
no data corruption.
Which paragraph/comment exactly?
Quoted 17 lines
Signed-off-by Nigel Croxon <[email protected]>---
drivers/md/raid5.c | 1 +1 file changed, 1 insertion(+)drivers/md/raid5.c +1
-0@@ -6217,6 +6217,7 @@ static bool raid5_make_request(struct mddev *mddev, struct bio * bi) mempool_free(ctx, conf->ctx_pool); if (res == STRIPE_WAIT_RESHAPE) { + bio_endio(bi); md_free_cloned_bio(bi); return false;Quoted 1 line
}
Reviewed-by Paul Menzel <[email protected]>Kind regards,
Paul