[PATCH] block: always set partition number to '0' in blk_partition_remap()

[PATCH] block: always set partition number to '0' in blk_partition_remap()

[Hannes Reinecke]

blk_partition_remap() will only clear bi_partno if an actual remapping
has happened. But flush request et al don't have an actual size, so
the remapping doesn't happen and bi_partno is never cleared.
So for stacked devices blk_partition_remap() will be called on each level.
If (as is the case for native nvme multipathing) one of the lower-level
devices do _not_support partitioning a spurious I/O error is generated.

Signed-off-by: Hannes Reinecke <hare at suse.com>
---
 block/blk-core.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/block/blk-core.c b/block/blk-core.c
index cee03cad99f2..8a2c3a474234 100644
--- a/block/blk-core.c
+++ b/block/blk-core.c
@@ -2225,7 +2225,10 @@ static inline int blk_partition_remap(struct bio *bio)
 		bio->bi_partno = 0;
 		trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p),
 				      bio->bi_iter.bi_sector - p->start_sect);
-	}
+	} else
+		/* Set partition number to '0' to avoid repetitive calls */
+		bio->bi_partno = 0;
+
 	ret = 0;
 out:
 	rcu_read_unlock();
-- 
2.12.3

[PATCH] block: always set partition number to '0' in blk_partition_remap()

[Jens Axboe]

On 6/6/18 8:22 AM, Hannes Reinecke wrote:
> blk_partition_remap() will only clear bi_partno if an actual remapping
> has happened. But flush request et al don't have an actual size, so
> the remapping doesn't happen and bi_partno is never cleared.
> So for stacked devices blk_partition_remap() will be called on each level.
> If (as is the case for native nvme multipathing) one of the lower-level
> devices do _not_support partitioning a spurious I/O error is generated.

Just move it down, we're now clearing it for both cases.

diff --git a/block/blk-core.c b/block/blk-core.c
index 3f56be15f17e..cf0ee764b908 100644
--- a/block/blk-core.c
+++ b/block/blk-core.c
@@ -2220,10 +2220,10 @@ static inline int blk_partition_remap(struct bio *bio)
 		if (bio_check_eod(bio, part_nr_sects_read(p)))
 			goto out;
 		bio->bi_iter.bi_sector += p->start_sect;
-		bio->bi_partno = 0;
 		trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p),
 				      bio->bi_iter.bi_sector - p->start_sect);
 	}
+	bio->bi_partno = 0;
 	ret = 0;
 out:
 	rcu_read_unlock();

-- 
Jens Axboe

[PATCH] block: always set partition number to '0' in blk_partition_remap()

[Hannes Reinecke]

On Wed, 6 Jun 2018 08:26:56 -0600
Jens Axboe <axboe@kernel.dk> wrote:

> On 6/6/18 8:22 AM, Hannes Reinecke wrote:
> > blk_partition_remap() will only clear bi_partno if an actual
> > remapping has happened. But flush request et al don't have an
> > actual size, so the remapping doesn't happen and bi_partno is never
> > cleared. So for stacked devices blk_partition_remap() will be
> > called on each level. If (as is the case for native nvme
> > multipathing) one of the lower-level devices do _not_support
> > partitioning a spurious I/O error is generated.  
> 
> Just move it down, we're now clearing it for both cases.
> 
> diff --git a/block/blk-core.c b/block/blk-core.c
> index 3f56be15f17e..cf0ee764b908 100644
> --- a/block/blk-core.c
> +++ b/block/blk-core.c
> @@ -2220,10 +2220,10 @@ static inline int blk_partition_remap(struct
> bio *bio) if (bio_check_eod(bio, part_nr_sects_read(p)))
>  			goto out;
>  		bio->bi_iter.bi_sector += p->start_sect;
> -		bio->bi_partno = 0;
>  		trace_block_bio_remap(bio->bi_disk->queue, bio,
> part_devt(p), bio->bi_iter.bi_sector - p->start_sect);
>  	}
> +	bio->bi_partno = 0;
>  	ret = 0;
>  out:
>  	rcu_read_unlock();
> 

Okay, will be resending.

Cheers,

Hannes

[PATCH] block: always set partition number to '0' in blk_partition_remap()

[Christoph Hellwig]

On Wed, Jun 06, 2018@08:26:56AM -0600, Jens Axboe wrote:
> On 6/6/18 8:22 AM, Hannes Reinecke wrote:
> > blk_partition_remap() will only clear bi_partno if an actual remapping
> > has happened. But flush request et al don't have an actual size, so
> > the remapping doesn't happen and bi_partno is never cleared.
> > So for stacked devices blk_partition_remap() will be called on each level.
> > If (as is the case for native nvme multipathing) one of the lower-level
> > devices do _not_support partitioning a spurious I/O error is generated.
> 
> Just move it down, we're now clearing it for both cases.

Agreed.