btrfs: use btrfs_block_group_cache_done in update_block_group

When free'ing extents in a block group we check to see if the block
group is not cached, and then cache it if we need to.  However we'll
just carry on as long as we're loading the cache.  This is problematic
because we are dirtying the block group here.  If we are fast enough we
could do a transaction commit and clear the free space cache while we're
still loading the space cache in another thread.  This truncates the
free space inode, which will keep it from loading the space cache.

Fix this by using the btrfs_block_group_cache_done helper so that we try
to load the space cache unconditionally here, which will result in the
caller waiting for the fast caching to complete and keep us from
truncating the free space inode.

CC: stable@vger.kernel.org # 4.4+
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>

diff --git a/fs/btrfs/block-group.c b/fs/btrfs/block-group.c
index 384659dc7818142300d7d65c59d8226d9ee1db0c..540a7a63601e7a61ea346251ea126ae33419110a 100644 (file)
--- a/fs/btrfs/block-group.c
+++ b/fs/btrfs/block-group.c
@@ -2661,7 +2661,7 @@ int btrfs_update_block_group(struct btrfs_trans_handle *trans,
                 * is because we need the unpinning stage to actually add the
                 * space back to the block group, otherwise we will leak space.
                 */
-               if (!alloc && cache->cached == BTRFS_CACHE_NO)
+               if (!alloc && !btrfs_block_group_cache_done(cache))
                        btrfs_cache_block_group(cache, 1);
 
                byte_in_group = bytenr - cache->key.objectid;