btrfs_drop_extents: handle BTRFS_INODE_REF_KEY types

It's possible "key.type == BTRFS_INODE_REF_KEY" and "key.offset >= end".
Signed-off-by: NChris Mason <chris.mason@oracle.com>

fs/btrfs/file.c

@@ -507,11 +507,12 @@ int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 		slot = path->slots[0];
 		ret = 0;
 		btrfs_item_key_to_cpu(leaf, &key, slot);
-
-		if (key.offset >= end || key.objectid != inode->i_ino) {
+		if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY &&
+		    key.offset >= end) {
 			goto out;
 		}
-		if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY) {
+		if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
+		    key.objectid != inode->i_ino) {
 			goto out;
 		}
 		if (recow) {