loopdev: fix a deadlock

bd_mutex and lo_ctl_mutex can be held in different order.

Path #1:

blkdev_open
 blkdev_get
  __blkdev_get (hold bd_mutex)
   lo_open (hold lo_ctl_mutex)

Path #2:

blkdev_ioctl
 lo_ioctl (hold lo_ctl_mutex)
  lo_set_capacity (hold bd_mutex)

Lockdep does not report it, because path #2 actually holds a subclass of
lo_ctl_mutex.  This subclass seems creep into the code by mistake.  The
patch author actually just mentioned it in the changelog, see commit
f028f3b2 ("loop: fix circular locking in loop_clr_fd()"), also see:

	http://marc.info/?l=linux-kernel&m=123806169129727&w=2

Path #2 hold bd_mutex to call bd_set_size(), I've protected it
with i_mutex in a previous patch, so drop bd_mutex at this site.

Signed-off-by: Guo Chao <yan@linux.vnet.ibm.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Guo Chao <yan@linux.vnet.ibm.com>
Cc: M. Hindess <hindessm@uk.ibm.com>
Cc: Nikanth Karthikesan <knikanth@suse.de>
Cc: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

diff --git a/drivers/block/loop.c b/drivers/block/loop.c
index ae125127062487f0bdbdb2f2463dce6047a2bb04..7f8065392064232952af1d252817152156fc38bd 100644 (file)
--- a/drivers/block/loop.c
+++ b/drivers/block/loop.c
@@ -1285,11 +1285,9 @@ static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev)
        /* the width of sector_t may be narrow for bit-shift */
        sz = sec;
        sz <<= 9;
-       mutex_lock(&bdev->bd_mutex);
        bd_set_size(bdev, sz);
        /* let user-space know about the new size */
        kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
-       mutex_unlock(&bdev->bd_mutex);
 
  out:
        return err;