1 files changed, 25 insertions, 5 deletions

diff --git a/fs/btrfs/tree-defrag.c b/fs/btrfs/tree-defrag.c
index a4b9c8b2d35a..cb65089127cc 100644
--- a/fs/btrfs/tree-defrag.c
+++ b/fs/btrfs/tree-defrag.c
@@ -89,6 +89,12 @@ int btrfs_defrag_leaves(struct btrfs_trans_handle *trans,
 		goto out;
 	}
 	btrfs_release_path(path);
+	/*
+	 * We don't need a lock on a leaf. btrfs_realloc_node() will lock all
+	 * leafs from path->nodes[1], so set lowest_level to 1 to avoid later
+	 * a deadlock (attempting to write lock an already write locked leaf).
+	 */
+	path->lowest_level = 1;
 	wret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 
 	if (wret < 0) {
@@ -99,9 +105,12 @@ int btrfs_defrag_leaves(struct btrfs_trans_handle *trans,
 		ret = 0;
 		goto out;
 	}
-	path->slots[1] = btrfs_header_nritems(path->nodes[1]);
-	next_key_ret = btrfs_find_next_key(root, path, &key, 1,
-					   min_trans);
+	/*
+	 * The node at level 1 must always be locked when our path has
+	 * keep_locks set and lowest_level is 1, regardless of the value of
+	 * path->slots[1].
+	 */
+	BUG_ON(path->locks[1] == 0);
 	ret = btrfs_realloc_node(trans, root,
 				 path->nodes[1], 0,
 				 &last_ret,
@@ -110,13 +119,24 @@ int btrfs_defrag_leaves(struct btrfs_trans_handle *trans,
 		WARN_ON(ret == -EAGAIN);
 		goto out;
 	}
+	/*
+	 * Now that we reallocated the node we can find the next key. Note that
+	 * btrfs_find_next_key() can release our path and do another search
+	 * without COWing, this is because even with path->keep_locks = 1,
+	 * btrfs_search_slot() / ctree.c:unlock_up() does not keeps a lock on a
+	 * node when path->slots[node_level - 1] does not point to the last
+	 * item or a slot beyond the last item (ctree.c:unlock_up()). Therefore
+	 * we search for the next key after reallocating our node.
+	 */
+	path->slots[1] = btrfs_header_nritems(path->nodes[1]);
+	next_key_ret = btrfs_find_next_key(root, path, &key, 1,
+					   min_trans);
 	if (next_key_ret == 0) {
 		memcpy(&root->defrag_progress, &key, sizeof(key));
 		ret = -EAGAIN;
 	}
 out:
-	if (path)
-		btrfs_free_path(path);
+	btrfs_free_path(path);
 	if (ret == -EAGAIN) {
 		if (root->defrag_max.objectid > root->defrag_progress.objectid)
 			goto done;