Avoid dereferencing a 'request_queue' after last close.
authorNeilBrown <neilb@suse.de>
Sat, 10 Sep 2011 07:20:21 +0000 (17:20 +1000)
committerNeilBrown <neilb@suse.de>
Sat, 10 Sep 2011 07:20:21 +0000 (17:20 +1000)
On the last close of an 'md' device which as been stopped, the device
is destroyed and in particular the request_queue is freed.  The free
is done in a separate thread so it might happen a short time later.

__blkdev_put calls bdev_inode_switch_bdi *after* ->release has been
called.

Since commit f758eeabeb96f878c860e8f110f94ec8820822a9
bdev_inode_switch_bdi will dereference the 'old' bdi, which lives
inside a request_queue, to get a spin lock.  This causes the last
close on an md device to sometime take a spin_lock which lives in
freed memory - which results in an oops.

So move the called to bdev_inode_switch_bdi before the call to
->release.

Cc: Christoph Hellwig <hch@lst.de>
Cc: Hugh Dickins <hughd@google.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Acked-by: Wu Fengguang <fengguang.wu@intel.com>
Cc: stable@kernel.org
Signed-off-by: NeilBrown <neilb@suse.de>
fs/block_dev.c

index ff77262e887cb66819f31d781e679eccb2d1975a..95f786ec7f088f1c85b7e82dfd01fe9bb7246fe3 100644 (file)
@@ -1429,6 +1429,11 @@ static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
                WARN_ON_ONCE(bdev->bd_holders);
                sync_blockdev(bdev);
                kill_bdev(bdev);
+               /* ->release can cause the old bdi to disappear,
+                * so must switch it out first
+                */
+               bdev_inode_switch_bdi(bdev->bd_inode,
+                                       &default_backing_dev_info);
        }
        if (bdev->bd_contains == bdev) {
                if (disk->fops->release)
@@ -1442,8 +1447,6 @@ static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
                disk_put_part(bdev->bd_part);
                bdev->bd_part = NULL;
                bdev->bd_disk = NULL;
-               bdev_inode_switch_bdi(bdev->bd_inode,
-                                       &default_backing_dev_info);
                if (bdev != bdev->bd_contains)
                        victim = bdev->bd_contains;
                bdev->bd_contains = NULL;