Btrfs: fix lock leak when resuming snapshot deletion
Josef Bacik [Mon, 15 Jul 2013 16:41:42 +0000 (12:41 -0400)]
commit fec386ac1428f9c0e672df952cbca5cebd4e4e2f upstream.

We aren't setting path->locks[level] when we resume a snapshot deletion which
means we won't unlock the buffer when we free the path.  This causes deadlocks
if we happen to re-allocate the block before we've evicted the extent buffer
from cache.  Thanks,

Reported-by: Alex Lyakas <alex.btrfs@zadarastorage.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

fs/btrfs/extent-tree.c

index fef1f21..d7c363d 100644 (file)
@@ -6903,6 +6903,7 @@ int btrfs_drop_snapshot(struct btrfs_root *root,
                while (1) {
                        btrfs_tree_lock(path->nodes[level]);
                        btrfs_set_lock_blocking(path->nodes[level]);
+                       path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
 
                        ret = btrfs_lookup_extent_info(trans, root,
                                                path->nodes[level]->start,
@@ -6919,6 +6920,7 @@ int btrfs_drop_snapshot(struct btrfs_root *root,
                                break;
 
                        btrfs_tree_unlock(path->nodes[level]);
+                       path->locks[level] = 0;
                        WARN_ON(wc->refs[level] != 1);
                        level--;
                }