Skip to contentSkip to conversation

block: pop cached rq before potentially blocking rq_qos_throttle()

PATCH In review Jens Axboe 3 participants 5 messages Jun 21, 2022 – Jun 23, 2022 Block layer

lore
Share & tools
lore

Jens Axboe

author maintainer
#0 lore raw

PATCH 1/1 block: pop cached rq before potentially blocking rq_qos_throttle()

If rq_qos_throttle() ends up blocking, then we will have invalidated and flushed our current plug. Since blk_mq_get_cached_request() hasn't popped the cached request off the plug list just yet, we end holding a pointer to a request that is no longer valid. This insta-crashes with rq->mq_hctx being NULL in the validity checks just after.

Pop the request off the cached list before doing rq_qos_throttle() to avoid using a potentially stale request.

Fixes 0a5aa8d161d1 ("block: fix blk_mq_attempt_bio_merge and rq_qos_throttle protection")

---

block/blk-mq.c +8 -3
index 33145ba52c96..93d9d60980fb 100644
@@ -2765,15 +2765,20 @@ static inline struct request *blk_mq_get_cached_request(struct request_queue *q,
 		return NULL;
 	}
 
-	rq_qos_throttle(q, *bio);
-
 	if (blk_mq_get_hctx_type((*bio)->bi_opf) != rq->mq_hctx->type)
 		return NULL;
 	if (op_is_flush(rq->cmd_flags) != op_is_flush((*bio)->bi_opf))
 		return NULL;
 
-	rq->cmd_flags = (*bio)->bi_opf;
+	/*
+	 * If any qos ->throttle() end up blocking, we will have flushed the
+	 * plug and hence killed the cached_rq list as well. Pop this entry
+	 * before we throttle.
+	 */
 	plug->cached_rq = rq_list_next(rq);
+	rq_qos_throttle(q, *bio);
+
+	rq->cmd_flags = (*bio)->bi_opf;
 	INIT_LIST_HEAD(&rq->queuelist);
 	return rq;
 }
Signature · 1 line
Jens Axboe
Tested-by on v1 1/1
#1 lore raw
Quoted 43 lines · Jens Axboe

If rq_qos_throttle() ends up blocking, then we will have invalidated and
flushed our current plug. Since blk_mq_get_cached_request() hasn't popped the cached request off the plug list just yet, we end holding a
pointer to a request that is no longer valid. This insta-crashes with rq->mq_hctx being NULL in the validity checks just after.

Pop the request off the cached list before doing rq_qos_throttle() to avoid using a potentially stale request.

Fixes 0a5aa8d161d1 ("block: fix blk_mq_attempt_bio_merge and

rq_qos_throttle protection")

---

block/blk-mq.c +0 -0
index 33145ba52c96..93d9d60980fb 100644
@@ -2765,15 +2765,20 @@ static inline struct request
 *blk_mq_get_cached_request(struct request_queue *q,
                 return NULL;
         }


-       rq_qos_throttle(q, *bio);
-
        if (blk_mq_get_hctx_type((*bio)->bi_opf) != rq->mq_hctx-

Quoted 1 line

type)

                return NULL;
        if (op_is_flush(rq->cmd_flags) != op_is_flush((*bio)-

Quoted 1 line

bi_opf))

                return NULL;

-       rq->cmd_flags = (*bio)->bi_opf;
+       /*
+        * If any qos ->throttle() end up blocking, we will have flushed the
+        * plug and hence killed the cached_rq list as well. Pop this entry
+        * before we throttle.

Quoted 8 lines

+        */
        plug->cached_rq = rq_list_next(rq);
+       rq_qos_throttle(q, *bio);
+
+       rq->cmd_flags = (*bio)->bi_opf;
        INIT_LIST_HEAD(&rq->queuelist);
        return rq;
 }

This fixes the problems I was seeing

Tested-by on v1 1/1
#2 lore raw
In reply to Jens Axboe
Quoted 6 lines · Jens Axboe

If rq_qos_throttle() ends up blocking, then we will have invalidated and flushed our current plug. Since blk_mq_get_cached_request() hasn't popped the cached request off the plug list just yet, we end holding a pointer to a request that is no longer valid. This insta-crashes with rq->mq_hctx being NULL in the validity checks just after.

Quoted 6 lines

Pop the request off the cached list before doing rq_qos_throttle() to avoid using a potentially stale request.

Fixes 0a5aa8d161d1 ("block: fix blk_mq_attempt_bio_merge and rq_qos_throttle protection")

Thank you for the fix and sorry for the trouble. The patch passed my test set:

BTW, may I know the workload or script which triggers the failure? I'm interested in if we can add a blktests test case to exercises qos throttle and prevent similar bugs in the future.

Signature · 1 line
Shin'ichiro Kawasaki
author maintainer
#3 lore raw
Quoted 14 lines · Shinichiro Kawasaki
Quoted 6 lines · Jens Axboe

If rq_qos_throttle() ends up blocking, then we will have invalidated and flushed our current plug. Since blk_mq_get_cached_request() hasn't popped the cached request off the plug list just yet, we end holding a pointer to a request that is no longer valid. This insta-crashes with rq->mq_hctx being NULL in the validity checks just after.

Quoted 6 lines

Pop the request off the cached list before doing rq_qos_throttle() to avoid using a potentially stale request.

Fixes 0a5aa8d161d1 ("block: fix blk_mq_attempt_bio_merge and rq_qos_throttle protection")
Quoted 8 lines

Thank you for the fix and sorry for the trouble. The patch passed my test set:

BTW, may I know the workload or script which triggers the failure? I'm interested in if we can add a blktests test case to exercises qos throttle and prevent similar bugs in the future.

Not sure if there are others, but specifically blk-iocost will do an explicit schedule() for some conditions. See the bottom of ioc_rqos_throttle().

Signature · 1 line
Jens Axboe
#4 lore raw
Quoted 20 lines · Jens Axboe
Quoted 19 lines · Shinichiro Kawasaki
Quoted 6 lines · Jens Axboe

If rq_qos_throttle() ends up blocking, then we will have invalidated and flushed our current plug. Since blk_mq_get_cached_request() hasn't popped the cached request off the plug list just yet, we end holding a pointer to a request that is no longer valid. This insta-crashes with rq->mq_hctx being NULL in the validity checks just after.

Quoted 6 lines

Pop the request off the cached list before doing rq_qos_throttle() to avoid using a potentially stale request.

Fixes 0a5aa8d161d1 ("block: fix blk_mq_attempt_bio_merge and rq_qos_throttle protection")

Thank you for the fix and sorry for the trouble. The patch passed my test set:

Quoted 7 lines
Quoted 3 lines

BTW, may I know the workload or script which triggers the failure? I'm interested in if we can add a blktests test case to exercises qos throttle and prevent similar bugs in the future.

Not sure if there are others, but specifically blk-iocost will do an explicit schedule() for some conditions. See the bottom of ioc_rqos_throttle().

Thanks. Will try to create a script which triggers the schedule() in ioc_rqos_throttle().

Signature · 1 line
Shin'ichiro Kawasaki