blk-mq: sync the update nr_hw_queues with blk_mq_queue_tag_busy_iter

commit f5bbbbe4d63577026f908a809f22f5fd5a90ea1f upstream.

For blk-mq, part_in_flight/rw will invoke blk_mq_in_flight/rw to
account the inflight requests. It will access the queue_hw_ctx and
nr_hw_queues w/o any protection. When updating nr_hw_queues and
blk_mq_in_flight/rw occur concurrently, panic comes up.

Before update nr_hw_queues, the q will be frozen. So we could use
q_usage_counter to avoid the race. percpu_ref_is_zero is used here
so that we will not miss any in-flight request. The access to
nr_hw_queues and queue_hw_ctx in blk_mq_queue_tag_busy_iter are
under rcu critical section, __blk_mq_update_nr_hw_queues could use
synchronize_rcu to ensure the zeroed q_usage_counter to be globally
visible.

--------------
NOTE: Back-ported to 4.4.y.

The upstream commit was intended to prevent concurrent manipulation of
nr_hw_queues and iteration over queues. The former doesn't happen in
this in 4.4.7 (as __blk_mq_update_nr_hw_queues doesn't exist). The
extra locking is also buggy in this commit but fixed in a follow-up.

It may protect against other concurrent accesses such as queue removal
by synchronising RCU locking around q_usage_counter.
--------------

Signed-off-by: Jianchao Wang <jianchao.w.wang@oracle.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Giuliano Procida <gprocida@google.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
1 file changed