block: don't call blk_drain_queue() if elevator is not up
authorTejun Heo <tj@kernel.org>
Thu, 3 Nov 2011 17:52:11 +0000 (18:52 +0100)
committerJens Axboe <axboe@kernel.dk>
Thu, 3 Nov 2011 17:52:11 +0000 (18:52 +0100)
blk_cleanup_queue() may be called before elevator is set up on a
queue which triggers the following oops.

 BUG: unable to handle kernel NULL pointer dereference at           (null)
 IP: [<ffffffff8125a69c>] elv_drain_elevator+0x1c/0x70
 ...
 Pid: 830, comm: kworker/0:2 Not tainted 3.1.0-next-20111025_64+ #1590
 Bochs Bochs
 RIP: 0010:[<ffffffff8125a69c>]  [<ffffffff8125a69c>] elv_drain_elevator+0x1c/0x70
 ...
 Call Trace:
  [<ffffffff8125da92>] blk_drain_queue+0x42/0x70
  [<ffffffff8125db90>] blk_cleanup_queue+0xd0/0x1c0
  [<ffffffff81469640>] md_free+0x50/0x70
  [<ffffffff8126f43b>] kobject_release+0x8b/0x1d0
  [<ffffffff81270d56>] kref_put+0x36/0xa0
  [<ffffffff8126f2b7>] kobject_put+0x27/0x60
  [<ffffffff814693af>] mddev_delayed_delete+0x2f/0x40
  [<ffffffff81083450>] process_one_work+0x100/0x3b0
  [<ffffffff8108527f>] worker_thread+0x15f/0x3a0
  [<ffffffff81089937>] kthread+0x87/0x90
  [<ffffffff81621834>] kernel_thread_helper+0x4/0x10

Fix it by making blk_cleanup_queue() check whether q->elevator is set
up before invoking blk_drain_queue.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-and-tested-by: Jiri Slaby <jslaby@suse.cz>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
block/blk-core.c

index f658711..f43c8a5 100644 (file)
@@ -407,8 +407,13 @@ void blk_cleanup_queue(struct request_queue *q)
        spin_unlock_irq(lock);
        mutex_unlock(&q->sysfs_lock);
 
-       /* drain all requests queued before DEAD marking */
-       blk_drain_queue(q, true);
+       /*
+        * Drain all requests queued before DEAD marking.  The caller might
+        * be trying to tear down @q before its elevator is initialized, in
+        * which case we don't want to call into draining.
+        */
+       if (q->elevator)
+               blk_drain_queue(q, true);
 
        /* @q won't process any more request, flush async actions */
        del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);