[PATCH v2] blk-mq: enforce op-specific segment limits in blk_insert_cloned_request

[PATCH v2] blk-mq: enforce op-specific segment limits in blk_insert_cloned_request

[Uday Shankar]

The block layer might merge together discard requests up until the
max_discard_segments limit is hit, but blk_insert_cloned_request checks
the segment count against max_segments regardless of the req op. This
can result in errors like the following when discards are issued through
a DM device and max_discard_segments exceeds max_segments for the queue
of the chosen underlying device.

blk_insert_cloned_request: over max segments limit. (256 > 129)

Fix this by looking at the req_op and enforcing the appropriate segment
limit - max_discard_segments for REQ_OP_DISCARDs and max_segments for
everything else.

Signed-off-by: Uday Shankar <ushankar@purestorage.com>
---
Changes since v1:
- Fixed format specifier type mismatch
Reported-by: kernel test robot <lkp@intel.com>
Link: https://lore.kernel.org/oe-kbuild-all/202302162040.FaI25ul2-lkp@intel.com/

 block/blk-merge.c | 4 +---
 block/blk-mq.c    | 7 ++++---
 block/blk.h       | 8 ++++++++
 3 files changed, 13 insertions(+), 6 deletions(-)

diff --git a/block/blk-merge.c b/block/blk-merge.c
index b7c193d67..7f663c2d3 100644
--- a/block/blk-merge.c
+++ b/block/blk-merge.c
@@ -588,9 +588,7 @@ EXPORT_SYMBOL(__blk_rq_map_sg);
 
 static inline unsigned int blk_rq_get_max_segments(struct request *rq)
 {
-	if (req_op(rq) == REQ_OP_DISCARD)
-		return queue_max_discard_segments(rq->q);
-	return queue_max_segments(rq->q);
+	return blk_queue_get_max_segments(rq->q, req_op(rq));
 }
 
 static inline unsigned int blk_rq_get_max_sectors(struct request *rq,
diff --git a/block/blk-mq.c b/block/blk-mq.c
index d3494a796..b053a9255 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -3000,6 +3000,7 @@ blk_status_t blk_insert_cloned_request(struct request *rq)
 {
 	struct request_queue *q = rq->q;
 	unsigned int max_sectors = blk_queue_get_max_sectors(q, req_op(rq));
+	unsigned int max_segments = blk_queue_get_max_segments(q, req_op(rq));
 	blk_status_t ret;
 
 	if (blk_rq_sectors(rq) > max_sectors) {
@@ -3026,9 +3027,9 @@ blk_status_t blk_insert_cloned_request(struct request *rq)
 	 * original queue.
 	 */
 	rq->nr_phys_segments = blk_recalc_rq_segments(rq);
-	if (rq->nr_phys_segments > queue_max_segments(q)) {
-		printk(KERN_ERR "%s: over max segments limit. (%hu > %hu)\n",
-			__func__, rq->nr_phys_segments, queue_max_segments(q));
+	if (rq->nr_phys_segments > max_segments) {
+		printk(KERN_ERR "%s: over max segments limit. (%u > %u)\n",
+			__func__, rq->nr_phys_segments, max_segments);
 		return BLK_STS_IOERR;
 	}
 
diff --git a/block/blk.h b/block/blk.h
index f02381405..8d705c13a 100644
--- a/block/blk.h
+++ b/block/blk.h
@@ -169,6 +169,14 @@ static inline unsigned int blk_queue_get_max_sectors(struct request_queue *q,
 	return q->limits.max_sectors;
 }
 
+static inline unsigned int blk_queue_get_max_segments(struct request_queue *q,
+						      enum req_op op)
+{
+	if (op == REQ_OP_DISCARD)
+		return queue_max_discard_segments(q);
+	return queue_max_segments(q);
+}
+
 #ifdef CONFIG_BLK_DEV_INTEGRITY
 void blk_flush_integrity(void);
 bool __bio_integrity_endio(struct bio *);

base-commit: 6bea9ac7c6481c09eb2b61d7cd844fc64a526e3e
-- 
2.25.1

Re: [PATCH v2] blk-mq: enforce op-specific segment limits in blk_insert_cloned_request

[Keith Busch]

On Wed, Feb 22, 2023 at 11:52:25AM -0700, Uday Shankar wrote:
>  static inline unsigned int blk_rq_get_max_segments(struct request *rq)
>  {
> -	if (req_op(rq) == REQ_OP_DISCARD)
> -		return queue_max_discard_segments(rq->q);
> -	return queue_max_segments(rq->q);
> +	return blk_queue_get_max_segments(rq->q, req_op(rq));
>  }

I think you should just move this function to blk.h instead of
introducing a new one.

Re: [PATCH v2] blk-mq: enforce op-specific segment limits in blk_insert_cloned_request

[Uday Shankar]

On Wed, Feb 22, 2023 at 12:16:02PM -0700, Keith Busch wrote:
> On Wed, Feb 22, 2023 at 11:52:25AM -0700, Uday Shankar wrote:
> >  static inline unsigned int blk_rq_get_max_segments(struct request *rq)
> >  {
> > -	if (req_op(rq) == REQ_OP_DISCARD)
> > -		return queue_max_discard_segments(rq->q);
> > -	return queue_max_segments(rq->q);
> > +	return blk_queue_get_max_segments(rq->q, req_op(rq));
> >  }
> 
> I think you should just move this function to blk.h instead of
> introducing a new one.

I chose to add blk_queue_get_max_segments as a public function because
it parallels blk_queue_get_max_sectors. If you don't want two functions,
I could manually inline the (2) uses of blk_rq_get_max_segments(rq),
converting them to blk_queue_get_max_segments(rq->q, req_op(rq)).

Re: [PATCH v2] blk-mq: enforce op-specific segment limits in blk_insert_cloned_request

[Christoph Hellwig]

On Thu, Feb 23, 2023 at 12:34:46PM -0700, Uday Shankar wrote:
> I chose to add blk_queue_get_max_segments as a public function because
> it parallels blk_queue_get_max_sectors. If you don't want two functions,
> I could manually inline the (2) uses of blk_rq_get_max_segments(rq),
> converting them to blk_queue_get_max_segments(rq->q, req_op(rq)).

I'd be much happier with a single function that takes a request instead
of two decoded arguments.  This should not be a public API in any form.