bdev: Fixup error handling in blkdev_get()
authorJan Kara <jack@suse.cz>
Wed, 7 Aug 2019 09:36:47 +0000 (11:36 +0200)
committerJens Axboe <axboe@kernel.dk>
Thu, 8 Aug 2019 13:37:03 +0000 (07:37 -0600)
Commit 89e524c04fa9 ("loop: Fix mount(2) failure due to race with
LOOP_SET_FD") converted blkdev_get() to use the new helpers for
finishing claiming of a block device. However the conversion botched the
error handling in blkdev_get() and thus the bdev has been marked as held
even in case __blkdev_get() returned error. This led to occasional
warnings with block/001 test from blktests like:

kernel: WARNING: CPU: 5 PID: 907 at fs/block_dev.c:1899 __blkdev_put+0x396/0x3a0

Correct the error handling.

CC: stable@vger.kernel.org
Fixes: 89e524c04fa9 ("loop: Fix mount(2) failure due to race with LOOP_SET_FD")
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
fs/block_dev.c

index 131e2e0582a69f5ab60d35ae0850718b73ccece4..eb657ab9406062cdd23136e66fc265e472523a80 100644 (file)
@@ -1750,7 +1750,10 @@ int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
 
                /* finish claiming */
                mutex_lock(&bdev->bd_mutex);
-               bd_finish_claiming(bdev, whole, holder);
+               if (!res)
+                       bd_finish_claiming(bdev, whole, holder);
+               else
+                       bd_abort_claiming(bdev, whole, holder);
                /*
                 * Block event polling for write claims if requested.  Any
                 * write holder makes the write_holder state stick until