Skip to contentSkip to conversation

block: clear zone write plugging flag before failing rejected BIOs

PATCH In review Jackie Liu 2 participants 3 messages Jun 7 – Jun 9

Subsystems (2)
lore
Share & tools
lore

Jackie Liu

author
#0 lore raw

PATCH 1/1 block: clear zone write plugging flag before failing rejected BIOs

From: Jackie Liu <[email protected]>

Commit fe0418eb9bd6 ("block: Prevent potential deadlocks in zone write plug error recovery") changed blk_zone_wplug_handle_write() to fail BIOs directly when blk_zone_wplug_prepare_bio() rejects them, for example because the write is not aligned to the cached write pointer or the plug needs a write pointer update. However, the BIO is already marked with BIO_ZONE_WRITE_PLUGGING at that point even though it is not issued.

Completing such a BIO with bio_io_error() makes bio_endio() call blk_zone_write_plug_bio_endio(), which treats the completion as a failed device write and may poison the cached zone write pointer state by setting BLK_ZONE_WPLUG_NEED_WP_UPDATE.

Clear BIO_ZONE_WRITE_PLUGGING and drop the zone write plug reference before failing the rejected BIO.

Fixes fe0418eb9bd6 ("block: Prevent potential deadlocks in zone write plug error recovery")
---
 block/blk-zoned.c | 2 ++
 1 file changed, 2 insertions(+)
block/blk-zoned.c +2 -0
index 6a221c180889..855767d8bfc1 100644
@@ -1502,7 +1502,9 @@ static bool blk_zone_wplug_handle_write(struct bio *bio, unsigned int nr_segs)
 		goto queue_bio;
 
 	if (!blk_zone_wplug_prepare_bio(zwplug, bio)) {
+		bio_clear_flag(bio, BIO_ZONE_WRITE_PLUGGING);
 		spin_unlock_irqrestore(&zwplug->lock, flags);
+		disk_put_zone_wplug(zwplug);
 		bio_io_error(bio);
 		return true;
 	}
Signature · 1 line
2.54.0
maintainer
#1 lore raw
Quoted 9 lines · Jackie Liu

From: Jackie Liu <[email protected]>

Commit fe0418eb9bd6 ("block: Prevent potential deadlocks in zone write plug error recovery") changed blk_zone_wplug_handle_write() to fail BIOs directly when blk_zone_wplug_prepare_bio() rejects them, for example because the write is not aligned to the cached write pointer or the plug needs a write pointer update. However, the BIO is already marked with BIO_ZONE_WRITE_PLUGGING at that point even though it is not issued.

Quoted 4 lines

Completing such a BIO with bio_io_error() makes bio_endio() call blk_zone_write_plug_bio_endio(), which treats the completion as a failed device write and may poison the cached zone write pointer state by setting BLK_ZONE_WPLUG_NEED_WP_UPDATE.

Yes, true. But you did not explain clearly why that is a problem. After all, if we hit this case, the user issued an unaligned BIO, and so forcing it to do a report zones to get everything in sync and the correct write pointer is not a bad thing.

If fe0418eb9bd6 change is actually causing you problems, please describe that problem clearly. But ideally, I do not want to special case some error completions over others and prefer to have a single error path that result in the same state for the zone write plugs, regardless of a write error root cause.

Quoted 25 lines

Clear BIO_ZONE_WRITE_PLUGGING and drop the zone write plug reference before failing the rejected BIO.

Fixes fe0418eb9bd6 ("block: Prevent potential deadlocks in zone write plug error recovery")
---
 block/blk-zoned.c | 2 ++
 1 file changed, 2 insertions(+)
block/blk-zoned.c +2 -0
index 6a221c180889..855767d8bfc1 100644
@@ -1502,7 +1502,9 @@ static bool blk_zone_wplug_handle_write(struct bio *bio, unsigned int nr_segs)
 		goto queue_bio;
 
 	if (!blk_zone_wplug_prepare_bio(zwplug, bio)) {
+		bio_clear_flag(bio, BIO_ZONE_WRITE_PLUGGING);
 		spin_unlock_irqrestore(&zwplug->lock, flags);
+		disk_put_zone_wplug(zwplug);
 		bio_io_error(bio);
 		return true;
 	}
Signature · 2 lines
Damien Le Moal
Western Digital Research
author
#2 lore raw

2026年6月8日 19:42, "Damien Le Moal" <[email protected] mailto:[email protected]?to=%22Damien%20Le%20Moal%22%20%3Cdlemoal%40kernel.org%3E > 写到:

Quoted 23 lines · Jackie Liu
Quoted 11 lines

From: Jackie Liu <[email protected]>

Commit fe0418eb9bd6 ("block: Prevent potential deadlocks in zone write plug error recovery") changed blk_zone_wplug_handle_write() to fail BIOs directly when blk_zone_wplug_prepare_bio() rejects them, for example because the write is not aligned to the cached write pointer or the plug needs a write pointer update. However, the BIO is already marked with BIO_ZONE_WRITE_PLUGGING at that point even though it is not issued.

Quoted 5 lines

Completing such a BIO with bio_io_error() makes bio_endio() call blk_zone_write_plug_bio_endio(), which treats the completion as a failed device write and may poison the cached zone write pointer state by setting BLK_ZONE_WPLUG_NEED_WP_UPDATE.

Yes, true. But you did not explain clearly why that is a problem. After all, if we hit this case, the user issued an unaligned BIO, and so forcing it to do a report zones to get everything in sync and the correct write pointer is not a bad thing.

Quoted 4 lines

If fe0418eb9bd6 change is actually causing you problems, please describe that problem clearly. But ideally, I do not want to special case some error completions over others and prefer to have a single error path that result in the same state for the zone write plugs, regardless of a write error root cause.

Thanks for the review. I agree that the changelog did not describe a concrete user-visible problem clearly enough.

I was treating NEED_WP_UPDATE on a BIO rejected before submission as stale state poisoning, because no device write was actually issued. But as you pointed out, for an invalid/non-sequential write, forcing the user to resynchronize the write pointer through report zones is consistent with the current conservative recovery model.

I do not have a concrete regression from fe0418eb9bd6 beyond that extra recovery requirement, so please drop this patch for now.

Thanks.

Jackie

Quoted 31 lines
Quoted 27 lines

Clear BIO_ZONE_WRITE_PLUGGING and drop the zone write plug reference before
failing the rejected BIO.

Fixes fe0418eb9bd6 ("block: Prevent potential deadlocks in zone write plug error recovery")

---

 block/blk-zoned.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/block/blk-zoned.c b/block/blk-zoned.c
index 6a221c180889..855767d8bfc1 100644
--- a/block/blk-zoned.c
+++ b/block/blk-zoned.c
@@ -1502,7 +1502,9 @@ static bool blk_zone_wplug_handle_write(struct bio *bio, unsigned int nr_segs) goto queue_bio;

if (!blk_zone_wplug_prepare_bio(zwplug, bio)) {
+ bio_clear_flag(bio, BIO_ZONE_WRITE_PLUGGING);
spin_unlock_irqrestore(&zwplug->lock, flags);
+ disk_put_zone_wplug(zwplug);
bio_io_error(bio);
return true;
}

Signature · 2 lines
Damien Le Moal
Western Digital Research