btrfs: do not call synchronize_srcu() in inode_tree_del
authorJosef Bacik <josef@toxicpanda.com>
Tue, 19 Nov 2019 18:59:35 +0000 (13:59 -0500)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Tue, 31 Dec 2019 15:41:53 +0000 (16:41 +0100)
commite3fb5bb7eb584be6dedb8b22d4d2f0495374b071
treeadb1a9269fc41169c057e35847b175ce8961c57a
parent831140602a555d5fa8d27f706c4ad3f432c1b2b1
btrfs: do not call synchronize_srcu() in inode_tree_del

commit f72ff01df9cf5db25c76674cac16605992d15467 upstream.

Testing with the new fsstress uncovered a pretty nasty deadlock with
lookup and snapshot deletion.

Process A
unlink
 -> final iput
   -> inode_tree_del
     -> synchronize_srcu(subvol_srcu)

Process B
btrfs_lookup  <- srcu_read_lock() acquired here
  -> btrfs_iget
    -> find inode that has I_FREEING set
      -> __wait_on_freeing_inode()

We're holding the srcu_read_lock() while doing the iget in order to make
sure our fs root doesn't go away, and then we are waiting for the inode
to finish freeing.  However because the free'ing process is doing a
synchronize_srcu() we deadlock.

Fix this by dropping the synchronize_srcu() in inode_tree_del().  We
don't need people to stop accessing the fs root at this point, we're
only adding our empty root to the dead roots list.

A larger much more invasive fix is forthcoming to address how we deal
with fs roots, but this fixes the immediate problem.

Fixes: 76dda93c6ae2 ("Btrfs: add snapshot/subvolume destroy ioctl")
CC: stable@vger.kernel.org # 4.4+
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
fs/btrfs/inode.c