Skip to contentSkip to conversation

mq-deadline: Make sure to always unlock zones

PATCH Applied Bart Van Assche 3 participants 3 messages Feb 28, 2018 – Mar 1, 2018 Block layer

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 mq-deadline: Make sure to always unlock zones

From: Damien Le Moal <[email protected]>

In case of a failed write request (all retries failed) and when using libata, the SCSI error handler calls scsi_finish_command(). In the case of blk-mq this means that scsi_mq_done() does not get called, that blk_mq_complete_request() does not get called and also that the mq-deadline .completed_request() method is not called. This results in the target zone of the failed write request being left in a locked state, preventing that any new write requests are issued to the same zone.

Fix this by replacing the .completed_request() method with the .finish_request() method as this method is always called whether or not a request completes successfully. Since the .finish_request() method is only called by the blk-mq core if a .prepare_request() method exists, add a dummy .prepare_request() method.

Fixes 5700f69178e9 ("mq-deadline: Introduce zone locking support")
[ bvanassche: edited patch description ]
Cc Hannes Reinecke <[email protected]>
Cc Ming Lei <[email protected]>
---
 block/mq-deadline.c | 16 +++++++++++++---
 1 file changed, 13 insertions(+), 3 deletions(-)
block/mq-deadline.c +13 -3
index c56f211c8440..8ec0ba9f5386 100644
@@ -535,13 +535,22 @@ static void dd_insert_requests(struct blk_mq_hw_ctx *hctx,
 	spin_unlock(&dd->lock);
 }
 
+/*
+ * Nothing to do here. This is defined only to ensure that .finish_request
+ * method is called upon request completion.
+ */
+static void dd_prepare_request(struct request *rq, struct bio *bio)
+{
+}
+
 /*
  * For zoned block devices, write unlock the target zone of
  * completed write requests. Do this while holding the zone lock
  * spinlock so that the zone is never unlocked while deadline_fifo_request()
- * while deadline_next_request() are executing.
+ * or deadline_next_request() are executing. This function is called for
+ * all requests, whether or not these requests complete successfully.
  */
-static void dd_completed_request(struct request *rq)
+static void dd_finish_request(struct request *rq)
 {
 	struct request_queue *q = rq->q;
 
@@ -756,7 +765,8 @@ static struct elevator_type mq_deadline = {
 	.ops.mq = {
 		.insert_requests	= dd_insert_requests,
 		.dispatch_request	= dd_dispatch_request,
-		.completed_request	= dd_completed_request,
+		.prepare_request	= dd_prepare_request,
+		.finish_request		= dd_finish_request,
 		.next_request		= elv_rb_latter_request,
 		.former_request		= elv_rb_former_request,
 		.bio_merge		= dd_bio_merge,
Signature · 1 line
2.16.2
Reviewed-by on v1 1/1
#1 lore raw
Quoted 69 lines · Bart Van Assche

From: Damien Le Moal <[email protected]>

In case of a failed write request (all retries failed) and when using libata, the SCSI error handler calls scsi_finish_command(). In the case of blk-mq this means that scsi_mq_done() does not get called, that blk_mq_complete_request() does not get called and also that the mq-deadline .completed_request() method is not called. This results in the target zone of the failed write request being left in a locked state, preventing that any new write requests are issued to the same zone.

Fix this by replacing the .completed_request() method with the .finish_request() method as this method is always called whether or not a request completes successfully. Since the .finish_request() method is only called by the blk-mq core if a .prepare_request() method exists, add a dummy .prepare_request() method.

Fixes 5700f69178e9 ("mq-deadline: Introduce zone locking support")
[ bvanassche: edited patch description ]
Cc Hannes Reinecke <[email protected]>
Cc Ming Lei <[email protected]>
---
 block/mq-deadline.c | 16 +++++++++++++---
 1 file changed, 13 insertions(+), 3 deletions(-)
block/mq-deadline.c +13 -3
index c56f211c8440..8ec0ba9f5386 100644
@@ -535,13 +535,22 @@ static void dd_insert_requests(struct blk_mq_hw_ctx *hctx,
 	spin_unlock(&dd->lock);
 }
 
+/*
+ * Nothing to do here. This is defined only to ensure that .finish_request
+ * method is called upon request completion.
+ */
+static void dd_prepare_request(struct request *rq, struct bio *bio)
+{
+}
+
 /*
  * For zoned block devices, write unlock the target zone of
  * completed write requests. Do this while holding the zone lock
  * spinlock so that the zone is never unlocked while deadline_fifo_request()
- * while deadline_next_request() are executing.
+ * or deadline_next_request() are executing. This function is called for
+ * all requests, whether or not these requests complete successfully.
  */
-static void dd_completed_request(struct request *rq)
+static void dd_finish_request(struct request *rq)
 {
 	struct request_queue *q = rq->q;
 
@@ -756,7 +765,8 @@ static struct elevator_type mq_deadline = {
 	.ops.mq = {
 		.insert_requests	= dd_insert_requests,
 		.dispatch_request	= dd_dispatch_request,
-		.completed_request	= dd_completed_request,
+		.prepare_request	= dd_prepare_request,
+		.finish_request		= dd_finish_request,
 		.next_request		= elv_rb_latter_request,
 		.former_request		= elv_rb_former_request,
 		.bio_merge		= dd_bio_merge,
Signature · 1 line
2.16.2

Looks fine:

Thanks,
Ming

maintainer Applied
#2 lore raw
In reply to Bart Van Assche
Quoted 11 lines · Bart Van Assche

From: Damien Le Moal <[email protected]>

In case of a failed write request (all retries failed) and when using libata, the SCSI error handler calls scsi_finish_command(). In the case of blk-mq this means that scsi_mq_done() does not get called, that blk_mq_complete_request() does not get called and also that the mq-deadline .completed_request() method is not called. This results in the target zone of the failed write request being left in a locked state, preventing that any new write requests are issued to the same zone.

Quoted 5 lines

Fix this by replacing the .completed_request() method with the .finish_request() method as this method is always called whether or not a request completes successfully. Since the .finish_request() method is only called by the blk-mq core if a .prepare_request() method exists, add a dummy .prepare_request() method.

Thanks, applied for 4.16.

Signature · 1 line
Jens Axboe