1 files changed, 72 insertions, 34 deletions

diff --git a/fs/btrfs/delayed-inode.c b/fs/btrfs/delayed-inode.c
index 53c1211dd60b..90aaedce1548 100644
--- a/fs/btrfs/delayed-inode.c
+++ b/fs/btrfs/delayed-inode.c
@@ -313,7 +313,7 @@ static struct btrfs_delayed_item *btrfs_alloc_delayed_item(u16 data_len,
 {
 	struct btrfs_delayed_item *item;
 
-	item = kmalloc(sizeof(*item) + data_len, GFP_NOFS);
+	item = kmalloc(struct_size(item, data, data_len), GFP_NOFS);
 	if (item) {
 		item->data_len = data_len;
 		item->type = type;
@@ -412,6 +412,7 @@ static void finish_one_item(struct btrfs_delayed_root *delayed_root)
 
 static void __btrfs_remove_delayed_item(struct btrfs_delayed_item *delayed_item)
 {
+	struct btrfs_delayed_node *delayed_node = delayed_item->delayed_node;
 	struct rb_root_cached *root;
 	struct btrfs_delayed_root *delayed_root;
 
@@ -419,18 +420,21 @@ static void __btrfs_remove_delayed_item(struct btrfs_delayed_item *delayed_item)
 	if (RB_EMPTY_NODE(&delayed_item->rb_node))
 		return;
 
-	delayed_root = delayed_item->delayed_node->root->fs_info->delayed_root;
+	/* If it's in a rbtree, then we need to have delayed node locked. */
+	lockdep_assert_held(&delayed_node->mutex);
+
+	delayed_root = delayed_node->root->fs_info->delayed_root;
 
 	BUG_ON(!delayed_root);
 
 	if (delayed_item->type == BTRFS_DELAYED_INSERTION_ITEM)
-		root = &delayed_item->delayed_node->ins_root;
+		root = &delayed_node->ins_root;
 	else
-		root = &delayed_item->delayed_node->del_root;
+		root = &delayed_node->del_root;
 
 	rb_erase_cached(&delayed_item->rb_node, root);
 	RB_CLEAR_NODE(&delayed_item->rb_node);
-	delayed_item->delayed_node->count--;
+	delayed_node->count--;
 
 	finish_one_item(delayed_root);
 }
@@ -1153,20 +1157,33 @@ static int __btrfs_run_delayed_items(struct btrfs_trans_handle *trans, int nr)
 		ret = __btrfs_commit_inode_delayed_items(trans, path,
 							 curr_node);
 		if (ret) {
-			btrfs_release_delayed_node(curr_node);
-			curr_node = NULL;
 			btrfs_abort_transaction(trans, ret);
 			break;
 		}
 
 		prev_node = curr_node;
 		curr_node = btrfs_next_delayed_node(curr_node);
+		/*
+		 * See the comment below about releasing path before releasing
+		 * node. If the commit of delayed items was successful the path
+		 * should always be released, but in case of an error, it may
+		 * point to locked extent buffers (a leaf at the very least).
+		 */
+		ASSERT(path->nodes[0] == NULL);
 		btrfs_release_delayed_node(prev_node);
 	}
 
+	/*
+	 * Release the path to avoid a potential deadlock and lockdep splat when
+	 * releasing the delayed node, as that requires taking the delayed node's
+	 * mutex. If another task starts running delayed items before we take
+	 * the mutex, it will first lock the mutex and then it may try to lock
+	 * the same btree path (leaf).
+	 */
+	btrfs_free_path(path);
+
 	if (curr_node)
 		btrfs_release_delayed_node(curr_node);
-	btrfs_free_path(path);
 	trans->block_rsv = block_rsv;
 
 	return ret;
@@ -1413,7 +1430,29 @@ void btrfs_balance_delayed_items(struct btrfs_fs_info *fs_info)
 	btrfs_wq_run_delayed_node(delayed_root, fs_info, BTRFS_DELAYED_BATCH);
 }
 
-/* Will return 0 or -ENOMEM */
+static void btrfs_release_dir_index_item_space(struct btrfs_trans_handle *trans)
+{
+	struct btrfs_fs_info *fs_info = trans->fs_info;
+	const u64 bytes = btrfs_calc_insert_metadata_size(fs_info, 1);
+
+	if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
+		return;
+
+	/*
+	 * Adding the new dir index item does not require touching another
+	 * leaf, so we can release 1 unit of metadata that was previously
+	 * reserved when starting the transaction. This applies only to
+	 * the case where we had a transaction start and excludes the
+	 * transaction join case (when replaying log trees).
+	 */
+	trace_btrfs_space_reservation(fs_info, "transaction",
+				      trans->transid, bytes, 0);
+	btrfs_block_rsv_release(fs_info, trans->block_rsv, bytes, NULL);
+	ASSERT(trans->bytes_reserved >= bytes);
+	trans->bytes_reserved -= bytes;
+}
+
+/* Will return 0, -ENOMEM or -EEXIST (index number collision, unexpected). */
 int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans,
 				   const char *name, int name_len,
 				   struct btrfs_inode *dir,
@@ -1455,6 +1494,27 @@ int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans,
 
 	mutex_lock(&delayed_node->mutex);
 
+	/*
+	 * First attempt to insert the delayed item. This is to make the error
+	 * handling path simpler in case we fail (-EEXIST). There's no risk of
+	 * any other task coming in and running the delayed item before we do
+	 * the metadata space reservation below, because we are holding the
+	 * delayed node's mutex and that mutex must also be locked before the
+	 * node's delayed items can be run.
+	 */
+	ret = __btrfs_add_delayed_item(delayed_node, delayed_item);
+	if (unlikely(ret)) {
+		btrfs_err(trans->fs_info,
+"error adding delayed dir index item, name: %.*s, index: %llu, root: %llu, dir: %llu, dir->index_cnt: %llu, delayed_node->index_cnt: %llu, error: %d",
+			  name_len, name, index, btrfs_root_id(delayed_node->root),
+			  delayed_node->inode_id, dir->index_cnt,
+			  delayed_node->index_cnt, ret);
+		btrfs_release_delayed_item(delayed_item);
+		btrfs_release_dir_index_item_space(trans);
+		mutex_unlock(&delayed_node->mutex);
+		goto release_node;
+	}
+
 	if (delayed_node->index_item_leaves == 0 ||
 	    delayed_node->curr_index_batch_size + data_len > leaf_data_size) {
 		delayed_node->curr_index_batch_size = data_len;
@@ -1472,36 +1532,14 @@ int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans,
 		 * impossible.
 		 */
 		if (WARN_ON(ret)) {
-			mutex_unlock(&delayed_node->mutex);
 			btrfs_release_delayed_item(delayed_item);
+			mutex_unlock(&delayed_node->mutex);
 			goto release_node;
 		}
 
 		delayed_node->index_item_leaves++;
-	} else if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
-		const u64 bytes = btrfs_calc_insert_metadata_size(fs_info, 1);
-
-		/*
-		 * Adding the new dir index item does not require touching another
-		 * leaf, so we can release 1 unit of metadata that was previously
-		 * reserved when starting the transaction. This applies only to
-		 * the case where we had a transaction start and excludes the
-		 * transaction join case (when replaying log trees).
-		 */
-		trace_btrfs_space_reservation(fs_info, "transaction",
-					      trans->transid, bytes, 0);
-		btrfs_block_rsv_release(fs_info, trans->block_rsv, bytes, NULL);
-		ASSERT(trans->bytes_reserved >= bytes);
-		trans->bytes_reserved -= bytes;
-	}
-
-	ret = __btrfs_add_delayed_item(delayed_node, delayed_item);
-	if (unlikely(ret)) {
-		btrfs_err(trans->fs_info,
-			  "err add delayed dir index item(name: %.*s) into the insertion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
-			  name_len, name, delayed_node->root->root_key.objectid,
-			  delayed_node->inode_id, ret);
-		BUG();
+	} else {
+		btrfs_release_dir_index_item_space(trans);
 	}
 	mutex_unlock(&delayed_node->mutex);