[PATCH] block/null_blk: Fix double blk_mq_start_request() warning

[PATCH] block/null_blk: Fix double blk_mq_start_request() warning

[chengming.zhou]

From: Chengming Zhou <zhouchengming@bytedance.com>

When CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION is enabled, null_queue_rq()
would return BLK_STS_RESOURCE or BLK_STS_DEV_RESOURCE for the request,
which has been marked as MQ_RQ_IN_FLIGHT by blk_mq_start_request().

Then null_queue_rqs() put these requests in the rqlist, return back to
the block layer core, which would try to queue them individually again,
so the warning in blk_mq_start_request() triggered.

Fix it by splitting the null_queue_rq() into two parts: the first is the
preparation of request, the second is the handling of request. We put
the blk_mq_start_request() after the preparation part, which may fail
and return back to the block layer core.

The throttling also belongs to the preparation part, so move it before
blk_mq_start_request(). And change the return type of null_handle_cmd()
to void, since it always return BLK_STS_OK now.

Reported-by: syzbot+fcc47ba2476570cbbeb0@syzkaller.appspotmail.com
Closes: https://lore.kernel.org/all/0000000000000e6aac06098aee0c@google.com/
Fixes: d78bfa1346ab ("block/null_blk: add queue_rqs() support")
Suggested-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com>
---
 drivers/block/null_blk/main.c | 25 +++++++++++++------------
 1 file changed, 13 insertions(+), 12 deletions(-)

diff --git a/drivers/block/null_blk/main.c b/drivers/block/null_blk/main.c
index 22a3cf7f32e2..3021d58ca51c 100644
--- a/drivers/block/null_blk/main.c
+++ b/drivers/block/null_blk/main.c
@@ -1464,19 +1464,13 @@ blk_status_t null_process_cmd(struct nullb_cmd *cmd, enum req_op op,
 	return BLK_STS_OK;
 }
 
-static blk_status_t null_handle_cmd(struct nullb_cmd *cmd, sector_t sector,
-				    sector_t nr_sectors, enum req_op op)
+static void null_handle_cmd(struct nullb_cmd *cmd, sector_t sector,
+			    sector_t nr_sectors, enum req_op op)
 {
 	struct nullb_device *dev = cmd->nq->dev;
 	struct nullb *nullb = dev->nullb;
 	blk_status_t sts;
 
-	if (test_bit(NULLB_DEV_FL_THROTTLED, &dev->flags)) {
-		sts = null_handle_throttled(cmd);
-		if (sts != BLK_STS_OK)
-			return sts;
-	}
-
 	if (op == REQ_OP_FLUSH) {
 		cmd->error = errno_to_blk_status(null_handle_flush(nullb));
 		goto out;
@@ -1493,7 +1487,6 @@ static blk_status_t null_handle_cmd(struct nullb_cmd *cmd, sector_t sector,
 
 out:
 	nullb_complete_cmd(cmd);
-	return BLK_STS_OK;
 }
 
 static enum hrtimer_restart nullb_bwtimer_fn(struct hrtimer *timer)
@@ -1724,8 +1717,6 @@ static blk_status_t null_queue_rq(struct blk_mq_hw_ctx *hctx,
 	cmd->fake_timeout = should_timeout_request(rq) ||
 		blk_should_fake_timeout(rq->q);
 
-	blk_mq_start_request(rq);
-
 	if (should_requeue_request(rq)) {
 		/*
 		 * Alternate between hitting the core BUSY path, and the
@@ -1738,6 +1729,15 @@ static blk_status_t null_queue_rq(struct blk_mq_hw_ctx *hctx,
 		return BLK_STS_OK;
 	}
 
+	if (test_bit(NULLB_DEV_FL_THROTTLED, &nq->dev->flags)) {
+		blk_status_t sts = null_handle_throttled(cmd);
+
+		if (sts != BLK_STS_OK)
+			return sts;
+	}
+
+	blk_mq_start_request(rq);
+
 	if (is_poll) {
 		spin_lock(&nq->poll_lock);
 		list_add_tail(&rq->queuelist, &nq->poll_list);
@@ -1747,7 +1747,8 @@ static blk_status_t null_queue_rq(struct blk_mq_hw_ctx *hctx,
 	if (cmd->fake_timeout)
 		return BLK_STS_OK;
 
-	return null_handle_cmd(cmd, sector, nr_sectors, req_op(rq));
+	null_handle_cmd(cmd, sector, nr_sectors, req_op(rq));
+	return BLK_STS_OK;
 }
 
 static void null_queue_rqs(struct request **rqlist)
-- 
2.40.1

Re: [PATCH] block/null_blk: Fix double blk_mq_start_request() warning

[Jens Axboe]

On Mon, 20 Nov 2023 03:25:21 +0000, chengming.zhou@linux.dev wrote:
> When CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION is enabled, null_queue_rq()
> would return BLK_STS_RESOURCE or BLK_STS_DEV_RESOURCE for the request,
> which has been marked as MQ_RQ_IN_FLIGHT by blk_mq_start_request().
> 
> Then null_queue_rqs() put these requests in the rqlist, return back to
> the block layer core, which would try to queue them individually again,
> so the warning in blk_mq_start_request() triggered.
> 
> [...]

Applied, thanks!

[1/1] block/null_blk: Fix double blk_mq_start_request() warning
      commit: 53f2bca2609237f910531f2c1a7779b16ce7952d

Best regards,
-- 
Jens Axboe