blk-mq: call commit_rqs while list empty but error happen
authoryangerkun <yangerkun@huawei.com>
Sat, 5 Sep 2020 11:25:56 +0000 (19:25 +0800)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 7 Oct 2020 06:02:53 +0000 (08:02 +0200)
commitc53cd18774060d53babe8bd754f5f9dfdc022609
tree66d7f3efcc80ac4583c7880398e9ae4b703a76d2
parenta6141f191d830f5437eff08816187ae077e377d7
blk-mq: call commit_rqs while list empty but error happen

[ Upstream commit 632bfb6323799c087fcb4108dfe59518609667a7 ]

Blk-mq should call commit_rqs once 'bd.last != true' and no more
request will come(so virtscsi can kick the virtqueue, e.g.). We already
do that in 'blk_mq_dispatch_rq_list/blk_mq_try_issue_list_directly' while
list not empty and 'queued > 0'. However, we can seen the same scene
once the last request in list call queue_rq and return error like
BLK_STS_IOERR which will not requeue the request, and lead that list
empty but need call commit_rqs too(Or the request for virtscsi will stay
timeout until other request kick virtqueue).

We found this problem by do fsstress test with offline/online virtscsi
device repeat quickly.

Fixes: d666ba98f849 ("blk-mq: add mq_ops->commit_rqs()")
Reported-by: zhangyi (F) <yi.zhang@huawei.com>
Signed-off-by: yangerkun <yangerkun@huawei.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Sasha Levin <sashal@kernel.org>
block/blk-mq.c