null_blk: fix spurious IO errors after failed past-wp access

PATCH Applied Alexey Dobriyan 4 participants 4 messages Feb 12, 2020 – Mar 12, 2020 Block layer

lore

Share & tools

lore

Alexey Dobriyan

author

2020-02-12 20:23 UTC #0 lore raw

PATCH 1/1 null_blk: fix spurious IO errors after failed past-wp access

Steps to reproduce:

	BLKRESETZONE zone 0

	// force EIO
	pwrite(fd, buf, 4096, 4096);

	[issue more IO including zone ioctls]

It will start failing randomly including IO to unrelated zones because of ->error "reuse". Trigger can be partition detection as well if test is not run immediately which is even more entertaining.

The fix is of course to clear ->error where necessary.

Signed-off-by Alexey Dobriyan (SK hynix) <[email protected]>

---

 drivers/block/null_blk_main.c |    2 ++
 1 file changed, 2 insertions(+)

drivers/block/null_blk_main.c +2 -0

@@ -605,6 +605,7 @@ static struct nullb_cmd *__alloc_cmd(struct nullb_queue *nq)
 	if (tag != -1U) {
 		cmd = &nq->cmds[tag];
 		cmd->tag = tag;
+		cmd->error = BLK_STS_OK;
 		cmd->nq = nq;
 		if (nq->dev->irqmode == NULL_IRQ_TIMER) {
 			hrtimer_init(&cmd->timer, CLOCK_MONOTONIC,

@@ -1385,6 +1386,7 @@ static blk_status_t null_queue_rq(struct blk_mq_hw_ctx *hctx,
 		cmd->timer.function = null_cmd_timer_expired;
 	}
 	cmd->rq = bd->rq;
+	cmd->error = BLK_STS_OK;
 	cmd->nq = nq;
 
 	blk_mq_start_request(bd->rq);

Chaitanya Kulkarni

2020-02-15 03:20 UTC #1 lore raw

Alexey, thanks for the patch however the description is not simple to understand.

I just sent a patch with a description and the test result.

Quoted 19 lines · Alexey Dobriyan

Steps to reproduce:
	BLKRESETZONE zone 0

	// force EIO
	pwrite(fd, buf, 4096, 4096);

	[issue more IO including zone ioctls]
It will start failing randomly including IO to unrelated zones because of ->error "reuse". Trigger can be partition detection as well if test is not run immediately which is even more entertaining.
The fix is of course to clear ->error where necessary.
Signed-off-by Alexey Dobriyan (SK hynix)<[email protected]>
---
  drivers/block/null_blk_main.c |    2 ++

Christoph Hellwig

Reviewed-by on v1 1/1

2020-02-19 16:35 UTC #2 lore raw

In reply to Alexey Dobriyan

Quoted 24 lines · Alexey Dobriyan

Steps to reproduce:
	BLKRESETZONE zone 0

	// force EIO
	pwrite(fd, buf, 4096, 4096);

	[issue more IO including zone ioctls]
It will start failing randomly including IO to unrelated zones because of ->error "reuse". Trigger can be partition detection as well if test is not run immediately which is even more entertaining.
The fix is of course to clear ->error where necessary.
Signed-off-by Alexey Dobriyan (SK hynix) <[email protected]>
---
 drivers/block/null_blk_main.c |    2 ++
 1 file changed, 2 insertions(+)

drivers/block/null_blk_main.c static struct nullb_cmd *__alloc_cmd(struct nullb_queue *nq) · line 608

Quoted 4 lines

drivers/block/null_blk_main.c +1 -0

@@ -605,6 +605,7 @@ static struct nullb_cmd *__alloc_cmd(struct nullb_queue *nq)
 	if (tag != -1U) {
 		cmd = &nq->cmds[tag];
 		cmd->tag = tag;
+		cmd->error = BLK_STS_OK;

I'd place this line in null_queue_bio to match the blk-mq patch more closely.

Otherwise this looks fine:

Reviewed-by Christoph Hellwig <[email protected]>

Can you add your testcase to blktests?

Jens Axboe

maintainer Applied

2020-03-12 15:10 UTC #3 lore raw

In reply to Alexey Dobriyan

Quoted 9 lines · Alexey Dobriyan

Steps to reproduce:

	BLKRESETZONE zone 0

	// force EIO
	pwrite(fd, buf, 4096, 4096);

	[issue more IO including zone ioctls]

Quoted 5 lines

It will start failing randomly including IO to unrelated zones because of ->error "reuse". Trigger can be partition detection as well if test is not run immediately which is even more entertaining.
The fix is of course to clear ->error where necessary.

Applied, thanks.

Signature · 1 line

Jens Axboe