blk-mq: consider non-idle request as "inflight" in blk_mq_rq_inflight()
authorMing Lei <ming.lei@redhat.com>
Tue, 7 Jul 2020 15:04:33 +0000 (11:04 -0400)
committerJens Axboe <axboe@kernel.dk>
Tue, 7 Jul 2020 15:06:25 +0000 (09:06 -0600)
commit05a4fed69ff00a8bd83538684cb602a4636b07a7
tree17db1d5a04efd784ca07e743b205b2012cd13232
parent65752aef0a407e1ef17ec78a7fc31ba4e0b360f9
blk-mq: consider non-idle request as "inflight" in blk_mq_rq_inflight()

dm-multipath is the only user of blk_mq_queue_inflight().  When
dm-multipath calls blk_mq_queue_inflight() to check if it has
outstanding IO it can get a false negative.  The reason for this is
blk_mq_rq_inflight() doesn't consider requests that are no longer
MQ_RQ_IN_FLIGHT but that are now MQ_RQ_COMPLETE (->complete isn't
called or finished yet) as "inflight".

This causes request-based dm-multipath's dm_wait_for_completion() to
return before all outstanding dm-multipath requests have actually
completed.  This breaks DM multipath's suspend functionality because
blk-mq requests complete after DM's suspend has finished -- which
shouldn't happen.

Fix this by considering any request not in the MQ_RQ_IDLE state
(so either MQ_RQ_COMPLETE or MQ_RQ_IN_FLIGHT) as "inflight" in
blk_mq_rq_inflight().

Fixes: 3c94d83cb3526 ("blk-mq: change blk_mq_queue_busy() to blk_mq_queue_inflight()")
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Cc: stable@vger.kernel.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
block/blk-mq.c