[PATCH V2 0/3] block/loop: handle discard/zeroout error

[PATCH V2 0/3] block/loop: handle discard/zeroout error

[Shaohua Li]

From: Shaohua Li <shli@fb.com>

Fix some problems when setting up loop device with a block device as back file
and create/mount ext4 in the loop device.

BTW: blkdev_issue_zeroout retries if we immediately find the device doesn't
support zeroout, but it doesn't retry if submit_bio_wait returns -EOPNOTSUPP.
Is this correct behavior?

Thanks,
Shaohua

Shaohua Li (3):
  block/loop: don't hijack error number
  block/loop: use FALLOC_FL_ZERO_RANGE for REQ_OP_WRITE_ZEROES
  block/loop: suppress discard IO error message

 drivers/block/loop.c | 12 +++++++++---
 1 file changed, 9 insertions(+), 3 deletions(-)

-- 
2.9.5

[PATCH V2 1/3] block/loop: don't hijack error number

[Shaohua Li]

From: Shaohua Li <shli@fb.com>

If the bio returns -EOPNOTSUPP, we shouldn't hijack it and return -EIO

Signed-off-by: Shaohua Li <shli@fb.com>
---
 drivers/block/loop.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/drivers/block/loop.c b/drivers/block/loop.c
index 85de673..715b762 100644
--- a/drivers/block/loop.c
+++ b/drivers/block/loop.c
@@ -460,7 +460,7 @@ static void lo_complete_rq(struct request *rq)
 		zero_fill_bio(bio);
 	}
 
-	blk_mq_end_request(rq, cmd->ret < 0 ? BLK_STS_IOERR : BLK_STS_OK);
+	blk_mq_end_request(rq, errno_to_blk_status(cmd->ret));
 }
 
 static void lo_rw_aio_do_completion(struct loop_cmd *cmd)
@@ -476,7 +476,7 @@ static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2)
 {
 	struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
 
-	cmd->ret = ret;
+	cmd->ret = ret > 0 ? 0 : ret;
 	lo_rw_aio_do_completion(cmd);
 }
 
@@ -1706,7 +1706,7 @@ static void loop_handle_cmd(struct loop_cmd *cmd)
  failed:
 	/* complete non-aio request */
 	if (!cmd->use_aio || ret) {
-		cmd->ret = ret ? -EIO : 0;
+		cmd->ret = ret;
 		blk_mq_complete_request(cmd->rq);
 	}
 }
-- 
2.9.5

[PATCH V2 2/3] block/loop: use FALLOC_FL_ZERO_RANGE for REQ_OP_WRITE_ZEROES

[Shaohua Li]

From: Shaohua Li <shli@fb.com>

REQ_OP_WRITE_ZEROES really means zero the data. And in blkdev_fallocate,
FALLOC_FL_ZERO_RANGE will retry but FALLOC_FL_PUNCH_HOLE not, even loop
request doesn't have BLKDEV_ZERO_NOFALLBACK set.

Signed-off-by: Shaohua Li <shli@fb.com>
---
 drivers/block/loop.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/drivers/block/loop.c b/drivers/block/loop.c
index 715b762..8934e25 100644
--- a/drivers/block/loop.c
+++ b/drivers/block/loop.c
@@ -426,6 +426,9 @@ static int lo_discard(struct loop_device *lo, struct request *rq, loff_t pos)
 	int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE;
 	int ret;
 
+	if (req_op(rq) == REQ_OP_WRITE_ZEROES)
+		mode = FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE;
+
 	if ((!file->f_op->fallocate) || lo->lo_encrypt_key_size) {
 		ret = -EOPNOTSUPP;
 		goto out;
-- 
2.9.5

[PATCH V2 3/3] block/loop: suppress discard IO error message

[Shaohua Li]

From: Shaohua Li <shli@fb.com>

We don't know if fallocate really supports FALLOC_FL_PUNCH_HOLE till
fallocate is called. If it doesn't support, loop will return -EOPNOTSUPP
and we see a lot of error message printed by blk_update_request. Failure
for discard IO isn't a big problem, so we just return 0 in loop which
will suppress the IO error message.

Signed-off-by: Shaohua Li <shli@fb.com>
---
 drivers/block/loop.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/drivers/block/loop.c b/drivers/block/loop.c
index 8934e25..9d4545f 100644
--- a/drivers/block/loop.c
+++ b/drivers/block/loop.c
@@ -437,6 +437,9 @@ static int lo_discard(struct loop_device *lo, struct request *rq, loff_t pos)
 	ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
 	if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
 		ret = -EIO;
+
+	if (req_op(rq) == REQ_OP_DISCARD && ret == -EOPNOTSUPP)
+		ret = 0;
  out:
 	return ret;
 }
-- 
2.9.5

Re: [PATCH V2 2/3] block/loop: use FALLOC_FL_ZERO_RANGE for REQ_OP_WRITE_ZEROES

[Ming Lei]

On Thu, Sep 7, 2017 at 8:13 AM, Shaohua Li <shli@kernel.org> wrote:
> From: Shaohua Li <shli@fb.com>
>
> REQ_OP_WRITE_ZEROES really means zero the data. And in blkdev_fallocate,
> FALLOC_FL_ZERO_RANGE will retry but FALLOC_FL_PUNCH_HOLE not, even loop
> request doesn't have BLKDEV_ZERO_NOFALLBACK set.
>
> Signed-off-by: Shaohua Li <shli@fb.com>
> ---
>  drivers/block/loop.c | 3 +++
>  1 file changed, 3 insertions(+)
>
> diff --git a/drivers/block/loop.c b/drivers/block/loop.c
> index 715b762..8934e25 100644
> --- a/drivers/block/loop.c
> +++ b/drivers/block/loop.c
> @@ -426,6 +426,9 @@ static int lo_discard(struct loop_device *lo, struct request *rq, loff_t pos)
>         int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE;
>         int ret;
>
> +       if (req_op(rq) == REQ_OP_WRITE_ZEROES)
> +               mode = FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE;
> +
>         if ((!file->f_op->fallocate) || lo->lo_encrypt_key_size) {
>                 ret = -EOPNOTSUPP;
>                 goto out;
> --
> 2.9.5
>

Looks fine,

      Reviewed-by: Ming Lei <ming.lei@redhat.com>


-- 
Ming Lei

Re: [PATCH V2 3/3] block/loop: suppress discard IO error message

[Ming Lei]

On Thu, Sep 7, 2017 at 8:13 AM, Shaohua Li <shli@kernel.org> wrote:
> From: Shaohua Li <shli@fb.com>
>
> We don't know if fallocate really supports FALLOC_FL_PUNCH_HOLE till
> fallocate is called. If it doesn't support, loop will return -EOPNOTSUPP
> and we see a lot of error message printed by blk_update_request. Failure
> for discard IO isn't a big problem, so we just return 0 in loop which
> will suppress the IO error message.

Setting RQF_QUIET for discard IO should be more clean for suppressing error,
and upper layer can get the failure notification too.

-- 
Ming Lei

Re: [PATCH V2 1/3] block/loop: don't hijack error number

[Ming Lei]

On Thu, Sep 7, 2017 at 8:13 AM, Shaohua Li <shli@kernel.org> wrote:
> From: Shaohua Li <shli@fb.com>
>
> If the bio returns -EOPNOTSUPP, we shouldn't hijack it and return -EIO
>
> Signed-off-by: Shaohua Li <shli@fb.com>
> ---
>  drivers/block/loop.c | 6 +++---
>  1 file changed, 3 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/block/loop.c b/drivers/block/loop.c
> index 85de673..715b762 100644
> --- a/drivers/block/loop.c
> +++ b/drivers/block/loop.c
> @@ -460,7 +460,7 @@ static void lo_complete_rq(struct request *rq)
>                 zero_fill_bio(bio);
>         }
>
> -       blk_mq_end_request(rq, cmd->ret < 0 ? BLK_STS_IOERR : BLK_STS_OK);
> +       blk_mq_end_request(rq, errno_to_blk_status(cmd->ret));
>  }
>
>  static void lo_rw_aio_do_completion(struct loop_cmd *cmd)
> @@ -476,7 +476,7 @@ static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2)
>  {
>         struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
>
> -       cmd->ret = ret;
> +       cmd->ret = ret > 0 ? 0 : ret;
>         lo_rw_aio_do_completion(cmd);
>  }
>
> @@ -1706,7 +1706,7 @@ static void loop_handle_cmd(struct loop_cmd *cmd)
>   failed:
>         /* complete non-aio request */
>         if (!cmd->use_aio || ret) {
> -               cmd->ret = ret ? -EIO : 0;
> +               cmd->ret = ret;
>                 blk_mq_complete_request(cmd->rq);
>         }
>  }
> --
> 2.9.5
>

Looks fine:

      Reviewed-by: Ming Lei <ming.lei@redhat.com>


-- 
Ming Lei

Re: [PATCH V2 0/3] block/loop: handle discard/zeroout error

[Ilya Dryomov]

Hi Shaohua,

You wrote:
> BTW: blkdev_issue_zeroout retries if we immediately find the device doesn't
> support zeroout, but it doesn't retry if submit_bio_wait returns -EOPNOTSUPP.
> Is this correct behavior?

I sent a patch for that yesterday, see "[PATCH] block: cope with WRITE
SAME failing in blkdev_issue_zeroout()" on linux-block.  It checks for
-EREMOTEIO, because that's what I hit, but I wonder if it should check
for -EOPNOTSUPP from submit_bio_wait() as well.  Can you think of
a case where, given bdev_write_zeroes_sectors() != 0, submit_bio_wait()
would fail with -EOPNOTSUPP and we would want to do explicit zeroing?

Please excuse broken threading.

Thanks,

                Ilya

Re: [PATCH V2 3/3] block/loop: suppress discard IO error message

[Shaohua Li]

On Thu, Sep 07, 2017 at 05:16:21PM +0800, Ming Lei wrote:
> On Thu, Sep 7, 2017 at 8:13 AM, Shaohua Li <shli@kernel.org> wrote:
> > From: Shaohua Li <shli@fb.com>
> >
> > We don't know if fallocate really supports FALLOC_FL_PUNCH_HOLE till
> > fallocate is called. If it doesn't support, loop will return -EOPNOTSUPP
> > and we see a lot of error message printed by blk_update_request. Failure
> > for discard IO isn't a big problem, so we just return 0 in loop which
> > will suppress the IO error message.
> 
> Setting RQF_QUIET for discard IO should be more clean for suppressing error,
> and upper layer can get the failure notification too.

Hmm, forgot why I didn't do this, did consider it before. Probably because
nobody check the return value of discard request. Probably we should skip the
error message for all discard IO in block layer.

I'll repost a patch to fix this issue.

Thanks,
Shaohua

Re: [PATCH V2 0/3] block/loop: handle discard/zeroout error

[Shaohua Li]

On Thu, Sep 07, 2017 at 03:20:01PM +0200, Ilya Dryomov wrote:
> Hi Shaohua,
> 
> You wrote:
> > BTW: blkdev_issue_zeroout retries if we immediately find the device doesn't
> > support zeroout, but it doesn't retry if submit_bio_wait returns -EOPNOTSUPP.
> > Is this correct behavior?
> 
> I sent a patch for that yesterday, see "[PATCH] block: cope with WRITE
> SAME failing in blkdev_issue_zeroout()" on linux-block.  It checks for
> -EREMOTEIO, because that's what I hit, but I wonder if it should check
> for -EOPNOTSUPP from submit_bio_wait() as well.  Can you think of
> a case where, given bdev_write_zeroes_sectors() != 0, submit_bio_wait()
> would fail with -EOPNOTSUPP and we would want to do explicit zeroing?

loop block device returns -EOPNOTSUPP before for zeroout. With the second
patch, it doesn't any more though.