block: add proper state guards to __elv_next_request
James Bottomley [Wed, 18 May 2011 14:20:10 +0000 (16:20 +0200)]
blk_cleanup_queue() calls elevator_exit() and after this, we can't
touch the elevator without oopsing.  __elv_next_request() must check
for this state because in the refcounted queue model, we can still
call it after blk_cleanup_queue() has been called.

This was reported as causing an oops attributable to scsi.

Signed-off-by: James Bottomley <James.Bottomley@suse.de>
Cc: stable@kernel.org
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>

block/blk.h

index 6126346..4df474d 100644 (file)
@@ -62,7 +62,8 @@ static inline struct request *__elv_next_request(struct request_queue *q)
                        return rq;
                }
 
-               if (!q->elevator->ops->elevator_dispatch_fn(q, 0))
+               if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags) ||
+                   !q->elevator->ops->elevator_dispatch_fn(q, 0))
                        return NULL;
        }
 }