1 files changed, 445 insertions, 365 deletions

diff --git a/fs/btrfs/delayed-inode.c b/fs/btrfs/delayed-inode.c
index 0095c6e4c3d1..ce6e9f8812e0 100644
--- a/fs/btrfs/delayed-inode.c
+++ b/fs/btrfs/delayed-inode.c
@@ -28,11 +28,7 @@ static struct kmem_cache *delayed_node_cache;
 
 int __init btrfs_delayed_inode_init(void)
 {
-	delayed_node_cache = kmem_cache_create("btrfs_delayed_node",
-					sizeof(struct btrfs_delayed_node),
-					0,
-					SLAB_MEM_SPREAD,
-					NULL);
+	delayed_node_cache = KMEM_CACHE(btrfs_delayed_node, 0);
 	if (!delayed_node_cache)
 		return -ENOMEM;
 	return 0;
@@ -43,6 +39,17 @@ void __cold btrfs_delayed_inode_exit(void)
 	kmem_cache_destroy(delayed_node_cache);
 }
 
+void btrfs_init_delayed_root(struct btrfs_delayed_root *delayed_root)
+{
+	atomic_set(&delayed_root->items, 0);
+	atomic_set(&delayed_root->items_seq, 0);
+	delayed_root->nodes = 0;
+	spin_lock_init(&delayed_root->lock);
+	init_waitqueue_head(&delayed_root->wait);
+	INIT_LIST_HEAD(&delayed_root->node_list);
+	INIT_LIST_HEAD(&delayed_root->prepare_list);
+}
+
 static inline void btrfs_init_delayed_node(
 				struct btrfs_delayed_node *delayed_node,
 				struct btrfs_root *root, u64 inode_id)
@@ -50,6 +57,7 @@ static inline void btrfs_init_delayed_node(
 	delayed_node->root = root;
 	delayed_node->inode_id = inode_id;
 	refcount_set(&delayed_node->refs, 0);
+	btrfs_delayed_node_ref_tracker_dir_init(delayed_node);
 	delayed_node->ins_root = RB_ROOT_CACHED;
 	delayed_node->del_root = RB_ROOT_CACHED;
 	mutex_init(&delayed_node->mutex);
@@ -58,7 +66,8 @@ static inline void btrfs_init_delayed_node(
 }
 
 static struct btrfs_delayed_node *btrfs_get_delayed_node(
-		struct btrfs_inode *btrfs_inode)
+		struct btrfs_inode *btrfs_inode,
+		struct btrfs_ref_tracker *tracker)
 {
 	struct btrfs_root *root = btrfs_inode->root;
 	u64 ino = btrfs_ino(btrfs_inode);
@@ -67,25 +76,27 @@ static struct btrfs_delayed_node *btrfs_get_delayed_node(
 	node = READ_ONCE(btrfs_inode->delayed_node);
 	if (node) {
 		refcount_inc(&node->refs);
+		btrfs_delayed_node_ref_tracker_alloc(node, tracker, GFP_NOFS);
 		return node;
 	}
 
-	spin_lock(&root->inode_lock);
-	node = radix_tree_lookup(&root->delayed_nodes_tree, ino);
+	xa_lock(&root->delayed_nodes);
+	node = xa_load(&root->delayed_nodes, ino);
 
 	if (node) {
 		if (btrfs_inode->delayed_node) {
 			refcount_inc(&node->refs);	/* can be accessed */
+			btrfs_delayed_node_ref_tracker_alloc(node, tracker, GFP_ATOMIC);
 			BUG_ON(btrfs_inode->delayed_node != node);
-			spin_unlock(&root->inode_lock);
+			xa_unlock(&root->delayed_nodes);
 			return node;
 		}
 
 		/*
 		 * It's possible that we're racing into the middle of removing
-		 * this node from the radix tree.  In this case, the refcount
+		 * this node from the xarray.  In this case, the refcount
 		 * was zero and it should never go back to one.  Just return
-		 * NULL like it was never in the radix at all; our release
+		 * NULL like it was never in the xarray at all; our release
 		 * function is in the process of removing it.
 		 *
 		 * Some implementations of refcount_inc refuse to bump the
@@ -93,36 +104,46 @@ static struct btrfs_delayed_node *btrfs_get_delayed_node(
 		 * here, refcount_inc() may decide to just WARN_ONCE() instead
 		 * of actually bumping the refcount.
 		 *
-		 * If this node is properly in the radix, we want to bump the
+		 * If this node is properly in the xarray, we want to bump the
 		 * refcount twice, once for the inode and once for this get
 		 * operation.
 		 */
 		if (refcount_inc_not_zero(&node->refs)) {
 			refcount_inc(&node->refs);
+			btrfs_delayed_node_ref_tracker_alloc(node, tracker, GFP_ATOMIC);
+			btrfs_delayed_node_ref_tracker_alloc(node, &node->inode_cache_tracker,
+							     GFP_ATOMIC);
 			btrfs_inode->delayed_node = node;
 		} else {
 			node = NULL;
 		}
 
-		spin_unlock(&root->inode_lock);
+		xa_unlock(&root->delayed_nodes);
 		return node;
 	}
-	spin_unlock(&root->inode_lock);
+	xa_unlock(&root->delayed_nodes);
 
 	return NULL;
 }
 
-/* Will return either the node or PTR_ERR(-ENOMEM) */
+/*
+ * Look up an existing delayed node associated with @btrfs_inode or create a new
+ * one and insert it to the delayed nodes of the root.
+ *
+ * Return the delayed node, or error pointer on failure.
+ */
 static struct btrfs_delayed_node *btrfs_get_or_create_delayed_node(
-		struct btrfs_inode *btrfs_inode)
+		struct btrfs_inode *btrfs_inode,
+		struct btrfs_ref_tracker *tracker)
 {
 	struct btrfs_delayed_node *node;
 	struct btrfs_root *root = btrfs_inode->root;
 	u64 ino = btrfs_ino(btrfs_inode);
 	int ret;
+	void *ptr;
 
 again:
-	node = btrfs_get_delayed_node(btrfs_inode);
+	node = btrfs_get_delayed_node(btrfs_inode, tracker);
 	if (node)
 		return node;
 
@@ -131,26 +152,35 @@ again:
 		return ERR_PTR(-ENOMEM);
 	btrfs_init_delayed_node(node, root, ino);
 
-	/* cached in the btrfs inode and can be accessed */
-	refcount_set(&node->refs, 2);
-
-	ret = radix_tree_preload(GFP_NOFS);
-	if (ret) {
+	/* Allocate and reserve the slot, from now it can return a NULL from xa_load(). */
+	ret = xa_reserve(&root->delayed_nodes, ino, GFP_NOFS);
+	if (ret == -ENOMEM) {
+		btrfs_delayed_node_ref_tracker_dir_exit(node);
 		kmem_cache_free(delayed_node_cache, node);
-		return ERR_PTR(ret);
+		return ERR_PTR(-ENOMEM);
 	}
-
-	spin_lock(&root->inode_lock);
-	ret = radix_tree_insert(&root->delayed_nodes_tree, ino, node);
-	if (ret == -EEXIST) {
-		spin_unlock(&root->inode_lock);
+	xa_lock(&root->delayed_nodes);
+	ptr = xa_load(&root->delayed_nodes, ino);
+	if (ptr) {
+		/* Somebody inserted it, go back and read it. */
+		xa_unlock(&root->delayed_nodes);
+		btrfs_delayed_node_ref_tracker_dir_exit(node);
 		kmem_cache_free(delayed_node_cache, node);
-		radix_tree_preload_end();
+		node = NULL;
 		goto again;
 	}
+	ptr = __xa_store(&root->delayed_nodes, ino, node, GFP_ATOMIC);
+	ASSERT(xa_err(ptr) != -EINVAL);
+	ASSERT(xa_err(ptr) != -ENOMEM);
+	ASSERT(ptr == NULL);
+
+	/* Cached in the inode and can be accessed. */
+	refcount_set(&node->refs, 2);
+	btrfs_delayed_node_ref_tracker_alloc(node, tracker, GFP_ATOMIC);
+	btrfs_delayed_node_ref_tracker_alloc(node, &node->inode_cache_tracker, GFP_ATOMIC);
+
 	btrfs_inode->delayed_node = node;
-	spin_unlock(&root->inode_lock);
-	radix_tree_preload_end();
+	xa_unlock(&root->delayed_nodes);
 
 	return node;
 }
@@ -174,6 +204,8 @@ static void btrfs_queue_delayed_node(struct btrfs_delayed_root *root,
 		list_add_tail(&node->n_list, &root->node_list);
 		list_add_tail(&node->p_list, &root->prepare_list);
 		refcount_inc(&node->refs);	/* inserted into list */
+		btrfs_delayed_node_ref_tracker_alloc(node, &node->node_list_tracker,
+						     GFP_ATOMIC);
 		root->nodes++;
 		set_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
 	}
@@ -187,6 +219,7 @@ static void btrfs_dequeue_delayed_node(struct btrfs_delayed_root *root,
 	spin_lock(&root->lock);
 	if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
 		root->nodes--;
+		btrfs_delayed_node_ref_tracker_free(node, &node->node_list_tracker);
 		refcount_dec(&node->refs);	/* not in the list */
 		list_del_init(&node->n_list);
 		if (!list_empty(&node->p_list))
@@ -197,26 +230,26 @@ static void btrfs_dequeue_delayed_node(struct btrfs_delayed_root *root,
 }
 
 static struct btrfs_delayed_node *btrfs_first_delayed_node(
-			struct btrfs_delayed_root *delayed_root)
+			struct btrfs_delayed_root *delayed_root,
+			struct btrfs_ref_tracker *tracker)
 {
-	struct list_head *p;
-	struct btrfs_delayed_node *node = NULL;
+	struct btrfs_delayed_node *node;
 
 	spin_lock(&delayed_root->lock);
-	if (list_empty(&delayed_root->node_list))
-		goto out;
-
-	p = delayed_root->node_list.next;
-	node = list_entry(p, struct btrfs_delayed_node, n_list);
-	refcount_inc(&node->refs);
-out:
+	node = list_first_entry_or_null(&delayed_root->node_list,
+					struct btrfs_delayed_node, n_list);
+	if (node) {
+		refcount_inc(&node->refs);
+		btrfs_delayed_node_ref_tracker_alloc(node, tracker, GFP_ATOMIC);
+	}
 	spin_unlock(&delayed_root->lock);
 
 	return node;
 }
 
 static struct btrfs_delayed_node *btrfs_next_delayed_node(
-						struct btrfs_delayed_node *node)
+						struct btrfs_delayed_node *node,
+						struct btrfs_ref_tracker *tracker)
 {
 	struct btrfs_delayed_root *delayed_root;
 	struct list_head *p;
@@ -236,6 +269,7 @@ static struct btrfs_delayed_node *btrfs_next_delayed_node(
 
 	next = list_entry(p, struct btrfs_delayed_node, n_list);
 	refcount_inc(&next->refs);
+	btrfs_delayed_node_ref_tracker_alloc(next, tracker, GFP_ATOMIC);
 out:
 	spin_unlock(&delayed_root->lock);
 
@@ -244,7 +278,7 @@ out:
 
 static void __btrfs_release_delayed_node(
 				struct btrfs_delayed_node *delayed_node,
-				int mod)
+				int mod, struct btrfs_ref_tracker *tracker)
 {
 	struct btrfs_delayed_root *delayed_root;
 
@@ -260,51 +294,51 @@ static void __btrfs_release_delayed_node(
 		btrfs_dequeue_delayed_node(delayed_root, delayed_node);
 	mutex_unlock(&delayed_node->mutex);
 
+	btrfs_delayed_node_ref_tracker_free(delayed_node, tracker);
 	if (refcount_dec_and_test(&delayed_node->refs)) {
 		struct btrfs_root *root = delayed_node->root;
 
-		spin_lock(&root->inode_lock);
+		xa_erase(&root->delayed_nodes, delayed_node->inode_id);
 		/*
 		 * Once our refcount goes to zero, nobody is allowed to bump it
 		 * back up.  We can delete it now.
 		 */
 		ASSERT(refcount_read(&delayed_node->refs) == 0);
-		radix_tree_delete(&root->delayed_nodes_tree,
-				  delayed_node->inode_id);
-		spin_unlock(&root->inode_lock);
+		btrfs_delayed_node_ref_tracker_dir_exit(delayed_node);
 		kmem_cache_free(delayed_node_cache, delayed_node);
 	}
 }
 
-static inline void btrfs_release_delayed_node(struct btrfs_delayed_node *node)
+static inline void btrfs_release_delayed_node(struct btrfs_delayed_node *node,
+					      struct btrfs_ref_tracker *tracker)
 {
-	__btrfs_release_delayed_node(node, 0);
+	__btrfs_release_delayed_node(node, 0, tracker);
 }
 
 static struct btrfs_delayed_node *btrfs_first_prepared_delayed_node(
-					struct btrfs_delayed_root *delayed_root)
+					struct btrfs_delayed_root *delayed_root,
+					struct btrfs_ref_tracker *tracker)
 {
-	struct list_head *p;
-	struct btrfs_delayed_node *node = NULL;
+	struct btrfs_delayed_node *node;
 
 	spin_lock(&delayed_root->lock);
-	if (list_empty(&delayed_root->prepare_list))
-		goto out;
-
-	p = delayed_root->prepare_list.next;
-	list_del_init(p);
-	node = list_entry(p, struct btrfs_delayed_node, p_list);
-	refcount_inc(&node->refs);
-out:
+	node = list_first_entry_or_null(&delayed_root->prepare_list,
+					struct btrfs_delayed_node, p_list);
+	if (node) {
+		list_del_init(&node->p_list);
+		refcount_inc(&node->refs);
+		btrfs_delayed_node_ref_tracker_alloc(node, tracker, GFP_ATOMIC);
+	}
 	spin_unlock(&delayed_root->lock);
 
 	return node;
 }
 
 static inline void btrfs_release_prepared_delayed_node(
-					struct btrfs_delayed_node *node)
+					struct btrfs_delayed_node *node,
+					struct btrfs_ref_tracker *tracker)
 {
-	__btrfs_release_delayed_node(node, 1);
+	__btrfs_release_delayed_node(node, 1, tracker);
 }
 
 static struct btrfs_delayed_item *btrfs_alloc_delayed_item(u16 data_len,
@@ -313,7 +347,7 @@ 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;
@@ -327,8 +361,23 @@ static struct btrfs_delayed_item *btrfs_alloc_delayed_item(u16 data_len,
 	return item;
 }
 
+static int delayed_item_index_cmp(const void *key, const struct rb_node *node)
+{
+	const u64 *index = key;
+	const struct btrfs_delayed_item *delayed_item = rb_entry(node,
+						 struct btrfs_delayed_item, rb_node);
+
+	if (delayed_item->index < *index)
+		return 1;
+	else if (delayed_item->index > *index)
+		return -1;
+
+	return 0;
+}
+
 /*
- * __btrfs_lookup_delayed_item - look up the delayed item by key
+ * Look up the delayed item by key.
+ *
  * @delayed_node: pointer to the delayed node
  * @index:	  the dir index value to lookup (offset of a dir index key)
  *
@@ -339,57 +388,35 @@ static struct btrfs_delayed_item *__btrfs_lookup_delayed_item(
 				struct rb_root *root,
 				u64 index)
 {
-	struct rb_node *node = root->rb_node;
-	struct btrfs_delayed_item *delayed_item = NULL;
+	struct rb_node *node;
 
-	while (node) {
-		delayed_item = rb_entry(node, struct btrfs_delayed_item,
-					rb_node);
-		if (delayed_item->index < index)
-			node = node->rb_right;
-		else if (delayed_item->index > index)
-			node = node->rb_left;
-		else
-			return delayed_item;
-	}
+	node = rb_find(&index, root, delayed_item_index_cmp);
+	return rb_entry_safe(node, struct btrfs_delayed_item, rb_node);
+}
 
-	return NULL;
+static int btrfs_delayed_item_cmp(const struct rb_node *new,
+				  const struct rb_node *exist)
+{
+	const struct btrfs_delayed_item *new_item =
+		rb_entry(new, struct btrfs_delayed_item, rb_node);
+
+	return delayed_item_index_cmp(&new_item->index, exist);
 }
 
 static int __btrfs_add_delayed_item(struct btrfs_delayed_node *delayed_node,
 				    struct btrfs_delayed_item *ins)
 {
-	struct rb_node **p, *node;
-	struct rb_node *parent_node = NULL;
 	struct rb_root_cached *root;
-	struct btrfs_delayed_item *item;
-	bool leftmost = true;
+	struct rb_node *exist;
 
 	if (ins->type == BTRFS_DELAYED_INSERTION_ITEM)
 		root = &delayed_node->ins_root;
 	else
 		root = &delayed_node->del_root;
 
-	p = &root->rb_root.rb_node;
-	node = &ins->rb_node;
-
-	while (*p) {
-		parent_node = *p;
-		item = rb_entry(parent_node, struct btrfs_delayed_item,
-				 rb_node);
-
-		if (item->index < ins->index) {
-			p = &(*p)->rb_right;
-			leftmost = false;
-		} else if (item->index > ins->index) {
-			p = &(*p)->rb_left;
-		} else {
-			return -EEXIST;
-		}
-	}
-
-	rb_link_node(node, parent_node, p);
-	rb_insert_color_cached(node, root, leftmost);
+	exist = rb_find_add_cached(&ins->rb_node, root, btrfs_delayed_item_cmp);
+	if (exist)
+		return -EEXIST;
 
 	if (ins->type == BTRFS_DELAYED_INSERTION_ITEM &&
 	    ins->index >= delayed_node->index_cnt)
@@ -412,6 +439,7 @@ 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 +447,19 @@ 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);
 
-	BUG_ON(!delayed_root);
+	delayed_root = delayed_node->root->fs_info->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);
 }
@@ -447,40 +476,25 @@ static void btrfs_release_delayed_item(struct btrfs_delayed_item *item)
 static struct btrfs_delayed_item *__btrfs_first_delayed_insertion_item(
 					struct btrfs_delayed_node *delayed_node)
 {
-	struct rb_node *p;
-	struct btrfs_delayed_item *item = NULL;
-
-	p = rb_first_cached(&delayed_node->ins_root);
-	if (p)
-		item = rb_entry(p, struct btrfs_delayed_item, rb_node);
+	struct rb_node *p = rb_first_cached(&delayed_node->ins_root);
 
-	return item;
+	return rb_entry_safe(p, struct btrfs_delayed_item, rb_node);
 }
 
 static struct btrfs_delayed_item *__btrfs_first_delayed_deletion_item(
 					struct btrfs_delayed_node *delayed_node)
 {
-	struct rb_node *p;
-	struct btrfs_delayed_item *item = NULL;
-
-	p = rb_first_cached(&delayed_node->del_root);
-	if (p)
-		item = rb_entry(p, struct btrfs_delayed_item, rb_node);
+	struct rb_node *p = rb_first_cached(&delayed_node->del_root);
 
-	return item;
+	return rb_entry_safe(p, struct btrfs_delayed_item, rb_node);
 }
 
 static struct btrfs_delayed_item *__btrfs_next_delayed_item(
 						struct btrfs_delayed_item *item)
 {
-	struct rb_node *p;
-	struct btrfs_delayed_item *next = NULL;
+	struct rb_node *p = rb_next(&item->rb_node);
 
-	p = rb_next(&item->rb_node);
-	if (p)
-		next = rb_entry(p, struct btrfs_delayed_item, rb_node);
-
-	return next;
+	return rb_entry_safe(p, struct btrfs_delayed_item, rb_node);
 }
 
 static int btrfs_delayed_item_reserve_metadata(struct btrfs_trans_handle *trans,
@@ -513,7 +527,7 @@ static int btrfs_delayed_item_reserve_metadata(struct btrfs_trans_handle *trans,
 		/*
 		 * For insertions we track reserved metadata space by accounting
 		 * for the number of leaves that will be used, based on the delayed
-		 * node's index_items_size field.
+		 * node's curr_index_batch_size and index_item_leaves fields.
 		 */
 		if (item->type == BTRFS_DELAYED_DELETION_ITEM)
 			item->bytes_reserved = num_bytes;
@@ -654,7 +668,7 @@ static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans,
 	struct btrfs_key first_key;
 	const u32 first_data_size = first_item->data_len;
 	int total_size;
-	char *ins_data = NULL;
+	char AUTO_KFREE(ins_data);
 	int ret;
 	bool continuous_keys_only = false;
 
@@ -724,12 +738,10 @@ static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans,
 		u32 *ins_sizes;
 		int i = 0;
 
-		ins_data = kmalloc(batch.nr * sizeof(u32) +
-				   batch.nr * sizeof(struct btrfs_key), GFP_NOFS);
-		if (!ins_data) {
-			ret = -ENOMEM;
-			goto out;
-		}
+		ins_data = kmalloc_array(batch.nr,
+					 sizeof(u32) + sizeof(struct btrfs_key), GFP_NOFS);
+		if (!ins_data)
+			return -ENOMEM;
 		ins_sizes = (u32 *)ins_data;
 		ins_keys = (struct btrfs_key *)(ins_data + batch.nr * sizeof(u32));
 		batch.keys = ins_keys;
@@ -745,7 +757,7 @@ static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans,
 
 	ret = btrfs_insert_empty_items(trans, root, path, &batch);
 	if (ret)
-		goto out;
+		return ret;
 
 	list_for_each_entry(curr, &item_list, tree_list) {
 		char *data_ptr;
@@ -800,9 +812,8 @@ static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans,
 		list_del(&curr->tree_list);
 		btrfs_release_delayed_item(curr);
 	}
-out:
-	kfree(ins_data);
-	return ret;
+
+	return 0;
 }
 
 static int btrfs_insert_delayed_items(struct btrfs_trans_handle *trans,
@@ -971,7 +982,7 @@ static void btrfs_release_delayed_inode(struct btrfs_delayed_node *delayed_node)
 
 	if (delayed_node &&
 	    test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
-		BUG_ON(!delayed_node->root);
+		ASSERT(delayed_node->root);
 		clear_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
 		delayed_node->count--;
 
@@ -1018,27 +1029,55 @@ static int __btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
 	ret = btrfs_lookup_inode(trans, root, path, &key, mod);
 	if (ret > 0)
 		ret = -ENOENT;
-	if (ret < 0)
+	if (ret < 0) {
+		/*
+		 * If we fail to update the delayed inode we need to abort the
+		 * transaction, because we could leave the inode with the
+		 * improper counts behind.
+		 */
+		if (unlikely(ret != -ENOENT))
+			btrfs_abort_transaction(trans, ret);
 		goto out;
+	}
 
 	leaf = path->nodes[0];
 	inode_item = btrfs_item_ptr(leaf, path->slots[0],
 				    struct btrfs_inode_item);
 	write_extent_buffer(leaf, &node->inode_item, (unsigned long)inode_item,
 			    sizeof(struct btrfs_inode_item));
-	btrfs_mark_buffer_dirty(leaf);
 
 	if (!test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
 		goto out;
 
-	path->slots[0]++;
-	if (path->slots[0] >= btrfs_header_nritems(leaf))
-		goto search;
-again:
+	/*
+	 * Now we're going to delete the INODE_REF/EXTREF, which should be the
+	 * only one ref left.  Check if the next item is an INODE_REF/EXTREF.
+	 *
+	 * But if we're the last item already, release and search for the last
+	 * INODE_REF/EXTREF.
+	 */
+	if (path->slots[0] + 1 >= btrfs_header_nritems(leaf)) {
+		key.objectid = node->inode_id;
+		key.type = BTRFS_INODE_EXTREF_KEY;
+		key.offset = (u64)-1;
+
+		btrfs_release_path(path);
+		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
+		if (unlikely(ret < 0)) {
+			btrfs_abort_transaction(trans, ret);
+			goto err_out;
+		}
+		ASSERT(ret > 0);
+		ASSERT(path->slots[0] > 0);
+		ret = 0;
+		path->slots[0]--;
+		leaf = path->nodes[0];
+	} else {
+		path->slots[0]++;
+	}
 	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 	if (key.objectid != node->inode_id)
 		goto out;
-
 	if (key.type != BTRFS_INODE_REF_KEY &&
 	    key.type != BTRFS_INODE_EXTREF_KEY)
 		goto out;
@@ -1048,39 +1087,16 @@ again:
 	 * so there is only one iref. The case that several irefs are
 	 * in the same item doesn't exist.
 	 */
-	btrfs_del_item(trans, root, path);
+	ret = btrfs_del_item(trans, root, path);
+	if (ret < 0)
+		btrfs_abort_transaction(trans, ret);
 out:
 	btrfs_release_delayed_iref(node);
 	btrfs_release_path(path);
 err_out:
 	btrfs_delayed_inode_release_metadata(fs_info, node, (ret < 0));
 	btrfs_release_delayed_inode(node);
-
-	/*
-	 * If we fail to update the delayed inode we need to abort the
-	 * transaction, because we could leave the inode with the improper
-	 * counts behind.
-	 */
-	if (ret && ret != -ENOENT)
-		btrfs_abort_transaction(trans, ret);
-
 	return ret;
-
-search:
-	btrfs_release_path(path);
-
-	key.type = BTRFS_INODE_EXTREF_KEY;
-	key.offset = -1;
-
-	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
-	if (ret < 0)
-		goto err_out;
-	ASSERT(ret);
-
-	ret = 0;
-	leaf = path->nodes[0];
-	path->slots[0]--;
-	goto again;
 }
 
 static inline int btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
@@ -1116,6 +1132,9 @@ __btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
 	if (ret)
 		return ret;
 
+	ret = btrfs_record_root_in_trans(trans, node->root);
+	if (ret)
+		return ret;
 	ret = btrfs_update_delayed_inode(trans, node->root, path, node);
 	return ret;
 }
@@ -1131,6 +1150,7 @@ static int __btrfs_run_delayed_items(struct btrfs_trans_handle *trans, int nr)
 	struct btrfs_fs_info *fs_info = trans->fs_info;
 	struct btrfs_delayed_root *delayed_root;
 	struct btrfs_delayed_node *curr_node, *prev_node;
+	struct btrfs_ref_tracker curr_delayed_node_tracker, prev_delayed_node_tracker;
 	struct btrfs_path *path;
 	struct btrfs_block_rsv *block_rsv;
 	int ret = 0;
@@ -1148,25 +1168,39 @@ static int __btrfs_run_delayed_items(struct btrfs_trans_handle *trans, int nr)
 
 	delayed_root = fs_info->delayed_root;
 
-	curr_node = btrfs_first_delayed_node(delayed_root);
+	curr_node = btrfs_first_delayed_node(delayed_root, &curr_delayed_node_tracker);
 	while (curr_node && (!count || nr--)) {
 		ret = __btrfs_commit_inode_delayed_items(trans, path,
 							 curr_node);
-		if (ret) {
-			btrfs_release_delayed_node(curr_node);
-			curr_node = NULL;
+		if (unlikely(ret)) {
 			btrfs_abort_transaction(trans, ret);
 			break;
 		}
 
 		prev_node = curr_node;
-		curr_node = btrfs_next_delayed_node(curr_node);
-		btrfs_release_delayed_node(prev_node);
+		prev_delayed_node_tracker = curr_delayed_node_tracker;
+		curr_node = btrfs_next_delayed_node(curr_node, &curr_delayed_node_tracker);
+		/*
+		 * 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, &prev_delayed_node_tracker);
 	}
 
-	if (curr_node)
-		btrfs_release_delayed_node(curr_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, &curr_delayed_node_tracker);
 	trans->block_rsv = block_rsv;
 
 	return ret;
@@ -1185,8 +1219,10 @@ int btrfs_run_delayed_items_nr(struct btrfs_trans_handle *trans, int nr)
 int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
 				     struct btrfs_inode *inode)
 {
-	struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
-	struct btrfs_path *path;
+	struct btrfs_ref_tracker delayed_node_tracker;
+	struct btrfs_delayed_node *delayed_node =
+		btrfs_get_delayed_node(inode, &delayed_node_tracker);
+	BTRFS_PATH_AUTO_FREE(path);
 	struct btrfs_block_rsv *block_rsv;
 	int ret;
 
@@ -1196,14 +1232,14 @@ int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
 	mutex_lock(&delayed_node->mutex);
 	if (!delayed_node->count) {
 		mutex_unlock(&delayed_node->mutex);
-		btrfs_release_delayed_node(delayed_node);
+		btrfs_release_delayed_node(delayed_node, &delayed_node_tracker);
 		return 0;
 	}
 	mutex_unlock(&delayed_node->mutex);
 
 	path = btrfs_alloc_path();
 	if (!path) {
-		btrfs_release_delayed_node(delayed_node);
+		btrfs_release_delayed_node(delayed_node, &delayed_node_tracker);
 		return -ENOMEM;
 	}
 
@@ -1212,8 +1248,7 @@ int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
 
 	ret = __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
 
-	btrfs_release_delayed_node(delayed_node);
-	btrfs_free_path(path);
+	btrfs_release_delayed_node(delayed_node, &delayed_node_tracker);
 	trans->block_rsv = block_rsv;
 
 	return ret;
@@ -1223,18 +1258,20 @@ int btrfs_commit_inode_delayed_inode(struct btrfs_inode *inode)
 {
 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
 	struct btrfs_trans_handle *trans;
-	struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
+	struct btrfs_ref_tracker delayed_node_tracker;
+	struct btrfs_delayed_node *delayed_node;
 	struct btrfs_path *path;
 	struct btrfs_block_rsv *block_rsv;
 	int ret;
 
+	delayed_node = btrfs_get_delayed_node(inode, &delayed_node_tracker);
 	if (!delayed_node)
 		return 0;
 
 	mutex_lock(&delayed_node->mutex);
 	if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
 		mutex_unlock(&delayed_node->mutex);
-		btrfs_release_delayed_node(delayed_node);
+		btrfs_release_delayed_node(delayed_node, &delayed_node_tracker);
 		return 0;
 	}
 	mutex_unlock(&delayed_node->mutex);
@@ -1268,7 +1305,7 @@ trans_out:
 	btrfs_end_transaction(trans);
 	btrfs_btree_balance_dirty(fs_info);
 out:
-	btrfs_release_delayed_node(delayed_node);
+	btrfs_release_delayed_node(delayed_node, &delayed_node_tracker);
 
 	return ret;
 }
@@ -1282,7 +1319,8 @@ void btrfs_remove_delayed_node(struct btrfs_inode *inode)
 		return;
 
 	inode->delayed_node = NULL;
-	btrfs_release_delayed_node(delayed_node);
+
+	btrfs_release_delayed_node(delayed_node, &delayed_node->inode_cache_tracker);
 }
 
 struct btrfs_async_delayed_work {
@@ -1298,6 +1336,7 @@ static void btrfs_async_run_delayed_root(struct btrfs_work *work)
 	struct btrfs_trans_handle *trans;
 	struct btrfs_path *path;
 	struct btrfs_delayed_node *delayed_node = NULL;
+	struct btrfs_ref_tracker delayed_node_tracker;
 	struct btrfs_root *root;
 	struct btrfs_block_rsv *block_rsv;
 	int total_done = 0;
@@ -1314,7 +1353,8 @@ static void btrfs_async_run_delayed_root(struct btrfs_work *work)
 		    BTRFS_DELAYED_BACKGROUND / 2)
 			break;
 
-		delayed_node = btrfs_first_prepared_delayed_node(delayed_root);
+		delayed_node = btrfs_first_prepared_delayed_node(delayed_root,
+								 &delayed_node_tracker);
 		if (!delayed_node)
 			break;
 
@@ -1323,7 +1363,8 @@ static void btrfs_async_run_delayed_root(struct btrfs_work *work)
 		trans = btrfs_join_transaction(root);
 		if (IS_ERR(trans)) {
 			btrfs_release_path(path);
-			btrfs_release_prepared_delayed_node(delayed_node);
+			btrfs_release_prepared_delayed_node(delayed_node,
+							    &delayed_node_tracker);
 			total_done++;
 			continue;
 		}
@@ -1338,7 +1379,8 @@ static void btrfs_async_run_delayed_root(struct btrfs_work *work)
 		btrfs_btree_balance_dirty_nodelay(root->fs_info);
 
 		btrfs_release_path(path);
-		btrfs_release_prepared_delayed_node(delayed_node);
+		btrfs_release_prepared_delayed_node(delayed_node,
+						    &delayed_node_tracker);
 		total_done++;
 
 	} while ((async_work->nr == 0 && total_done < BTRFS_DELAYED_WRITEBACK)
@@ -1361,8 +1403,7 @@ static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root,
 		return -ENOMEM;
 
 	async_work->delayed_root = delayed_root;
-	btrfs_init_work(&async_work->work, btrfs_async_run_delayed_root, NULL,
-			NULL);
+	btrfs_init_work(&async_work->work, btrfs_async_run_delayed_root, NULL);
 	async_work->nr = nr;
 
 	btrfs_queue_work(fs_info->delayed_workers, &async_work->work);
@@ -1371,20 +1412,28 @@ static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root,
 
 void btrfs_assert_delayed_root_empty(struct btrfs_fs_info *fs_info)
 {
-	WARN_ON(btrfs_first_delayed_node(fs_info->delayed_root));
+	struct btrfs_ref_tracker delayed_node_tracker;
+	struct btrfs_delayed_node *node;
+
+	node = btrfs_first_delayed_node( fs_info->delayed_root, &delayed_node_tracker);
+	if (WARN_ON(node)) {
+		btrfs_delayed_node_ref_tracker_free(node,
+						    &delayed_node_tracker);
+		refcount_dec(&node->refs);
+	}
 }
 
-static int could_end_wait(struct btrfs_delayed_root *delayed_root, int seq)
+static bool could_end_wait(struct btrfs_delayed_root *delayed_root, int seq)
 {
 	int val = atomic_read(&delayed_root->items_seq);
 
 	if (val < seq || val >= seq + BTRFS_DELAYED_BATCH)
-		return 1;
+		return true;
 
 	if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
-		return 1;
+		return true;
 
-	return 0;
+	return false;
 }
 
 void btrfs_balance_delayed_items(struct btrfs_fs_info *fs_info)
@@ -1413,23 +1462,46 @@ 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,
-				   struct btrfs_disk_key *disk_key, u8 flags,
+				   const struct btrfs_disk_key *disk_key, u8 flags,
 				   u64 index)
 {
 	struct btrfs_fs_info *fs_info = trans->fs_info;
 	const unsigned int leaf_data_size = BTRFS_LEAF_DATA_SIZE(fs_info);
 	struct btrfs_delayed_node *delayed_node;
+	struct btrfs_ref_tracker delayed_node_tracker;
 	struct btrfs_delayed_item *delayed_item;
 	struct btrfs_dir_item *dir_item;
 	bool reserve_leaf_space;
 	u32 data_len;
 	int ret;
 
-	delayed_node = btrfs_get_or_create_delayed_node(dir);
+	delayed_node = btrfs_get_or_create_delayed_node(dir, &delayed_node_tracker);
 	if (IS_ERR(delayed_node))
 		return PTR_ERR(delayed_node);
 
@@ -1455,6 +1527,27 @@ 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;
@@ -1472,47 +1565,24 @@ 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);
 
 release_node:
-	btrfs_release_delayed_node(delayed_node);
+	btrfs_release_delayed_node(delayed_node, &delayed_node_tracker);
 	return ret;
 }
 
-static int btrfs_delete_delayed_insertion_item(struct btrfs_fs_info *fs_info,
-					       struct btrfs_delayed_node *node,
-					       u64 index)
+static bool btrfs_delete_delayed_insertion_item(struct btrfs_delayed_node *node,
+						u64 index)
 {
 	struct btrfs_delayed_item *item;
 
@@ -1520,7 +1590,7 @@ static int btrfs_delete_delayed_insertion_item(struct btrfs_fs_info *fs_info,
 	item = __btrfs_lookup_delayed_item(&node->ins_root.rb_root, index);
 	if (!item) {
 		mutex_unlock(&node->mutex);
-		return 1;
+		return false;
 	}
 
 	/*
@@ -1555,23 +1625,25 @@ static int btrfs_delete_delayed_insertion_item(struct btrfs_fs_info *fs_info,
 	}
 
 	mutex_unlock(&node->mutex);
-	return 0;
+	return true;
 }
 
 int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans,
 				   struct btrfs_inode *dir, u64 index)
 {
 	struct btrfs_delayed_node *node;
+	struct btrfs_ref_tracker delayed_node_tracker;
 	struct btrfs_delayed_item *item;
 	int ret;
 
-	node = btrfs_get_or_create_delayed_node(dir);
+	node = btrfs_get_or_create_delayed_node(dir, &delayed_node_tracker);
 	if (IS_ERR(node))
 		return PTR_ERR(node);
 
-	ret = btrfs_delete_delayed_insertion_item(trans->fs_info, node, index);
-	if (!ret)
+	if (btrfs_delete_delayed_insertion_item(node, index)) {
+		ret = 0;
 		goto end;
+	}
 
 	item = btrfs_alloc_delayed_item(0, node, BTRFS_DELAYED_DELETION_ITEM);
 	if (!item) {
@@ -1588,7 +1660,8 @@ int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans,
 	 */
 	if (ret < 0) {
 		btrfs_err(trans->fs_info,
-"metadata reservation failed for delayed dir item deltiona, should have been reserved");
+"metadata reservation failed for delayed dir item deletion, index: %llu, root: %llu, inode: %llu, error: %d",
+			  index, btrfs_root_id(node->root), node->inode_id, ret);
 		btrfs_release_delayed_item(item);
 		goto end;
 	}
@@ -1597,22 +1670,23 @@ int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans,
 	ret = __btrfs_add_delayed_item(node, item);
 	if (unlikely(ret)) {
 		btrfs_err(trans->fs_info,
-			  "err add delayed dir index item(index: %llu) into the deletion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
-			  index, node->root->root_key.objectid,
-			  node->inode_id, ret);
+"failed to add delayed dir index item, root: %llu, inode: %llu, index: %llu, error: %d",
+			  index, btrfs_root_id(node->root), node->inode_id, ret);
 		btrfs_delayed_item_release_metadata(dir->root, item);
 		btrfs_release_delayed_item(item);
 	}
 	mutex_unlock(&node->mutex);
 end:
-	btrfs_release_delayed_node(node);
+	btrfs_release_delayed_node(node, &delayed_node_tracker);
 	return ret;
 }
 
 int btrfs_inode_delayed_dir_index_count(struct btrfs_inode *inode)
 {
-	struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
+	struct btrfs_ref_tracker delayed_node_tracker;
+	struct btrfs_delayed_node *delayed_node;
 
+	delayed_node = btrfs_get_delayed_node(inode, &delayed_node_tracker);
 	if (!delayed_node)
 		return -ENOENT;
 
@@ -1622,23 +1696,25 @@ int btrfs_inode_delayed_dir_index_count(struct btrfs_inode *inode)
 	 * is updated now. So we needn't lock the delayed node.
 	 */
 	if (!delayed_node->index_cnt) {
-		btrfs_release_delayed_node(delayed_node);
+		btrfs_release_delayed_node(delayed_node, &delayed_node_tracker);
 		return -EINVAL;
 	}
 
 	inode->index_cnt = delayed_node->index_cnt;
-	btrfs_release_delayed_node(delayed_node);
+	btrfs_release_delayed_node(delayed_node, &delayed_node_tracker);
 	return 0;
 }
 
-bool btrfs_readdir_get_delayed_items(struct inode *inode,
+bool btrfs_readdir_get_delayed_items(struct btrfs_inode *inode,
+				     u64 last_index,
 				     struct list_head *ins_list,
 				     struct list_head *del_list)
 {
 	struct btrfs_delayed_node *delayed_node;
 	struct btrfs_delayed_item *item;
+	struct btrfs_ref_tracker delayed_node_tracker;
 
-	delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
+	delayed_node = btrfs_get_delayed_node(inode, &delayed_node_tracker);
 	if (!delayed_node)
 		return false;
 
@@ -1646,19 +1722,19 @@ bool btrfs_readdir_get_delayed_items(struct inode *inode,
 	 * We can only do one readdir with delayed items at a time because of
 	 * item->readdir_list.
 	 */
-	btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_SHARED);
-	btrfs_inode_lock(BTRFS_I(inode), 0);
+	btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
+	btrfs_inode_lock(inode, 0);
 
 	mutex_lock(&delayed_node->mutex);
 	item = __btrfs_first_delayed_insertion_item(delayed_node);
-	while (item) {
+	while (item && item->index <= last_index) {
 		refcount_inc(&item->refs);
 		list_add_tail(&item->readdir_list, ins_list);
 		item = __btrfs_next_delayed_item(item);
 	}
 
 	item = __btrfs_first_delayed_deletion_item(delayed_node);
-	while (item) {
+	while (item && item->index <= last_index) {
 		refcount_inc(&item->refs);
 		list_add_tail(&item->readdir_list, del_list);
 		item = __btrfs_next_delayed_item(item);
@@ -1673,12 +1749,13 @@ bool btrfs_readdir_get_delayed_items(struct inode *inode,
 	 * insert/delete delayed items in this period. So we also needn't
 	 * requeue or dequeue this delayed node.
 	 */
+	btrfs_delayed_node_ref_tracker_free(delayed_node, &delayed_node_tracker);
 	refcount_dec(&delayed_node->refs);
 
 	return true;
 }
 
-void btrfs_readdir_put_delayed_items(struct inode *inode,
+void btrfs_readdir_put_delayed_items(struct btrfs_inode *inode,
 				     struct list_head *ins_list,
 				     struct list_head *del_list)
 {
@@ -1700,20 +1777,19 @@ void btrfs_readdir_put_delayed_items(struct inode *inode,
 	 * The VFS is going to do up_read(), so we need to downgrade back to a
 	 * read lock.
 	 */
-	downgrade_write(&inode->i_rwsem);
+	downgrade_write(&inode->vfs_inode.i_rwsem);
 }
 
-int btrfs_should_delete_dir_index(struct list_head *del_list,
-				  u64 index)
+bool btrfs_should_delete_dir_index(const struct list_head *del_list, u64 index)
 {
 	struct btrfs_delayed_item *curr;
-	int ret = 0;
+	bool ret = false;
 
 	list_for_each_entry(curr, del_list, readdir_list) {
 		if (curr->index > index)
 			break;
 		if (curr->index == index) {
-			ret = 1;
+			ret = true;
 			break;
 		}
 	}
@@ -1721,29 +1797,26 @@ int btrfs_should_delete_dir_index(struct list_head *del_list,
 }
 
 /*
- * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree
- *
+ * Read dir info stored in the delayed tree.
  */
-int btrfs_readdir_delayed_dir_index(struct dir_context *ctx,
-				    struct list_head *ins_list)
+bool btrfs_readdir_delayed_dir_index(struct dir_context *ctx,
+				     const struct list_head *ins_list)
 {
 	struct btrfs_dir_item *di;
 	struct btrfs_delayed_item *curr, *next;
 	struct btrfs_key location;
 	char *name;
 	int name_len;
-	int over = 0;
 	unsigned char d_type;
 
-	if (list_empty(ins_list))
-		return 0;
-
 	/*
 	 * Changing the data of the delayed item is impossible. So
 	 * we needn't lock them. And we have held i_mutex of the
 	 * directory, nobody can delete any directory indexes now.
 	 */
 	list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
+		bool over;
+
 		list_del(&curr->readdir_list);
 
 		if (curr->index < ctx->pos) {
@@ -1761,137 +1834,130 @@ int btrfs_readdir_delayed_dir_index(struct dir_context *ctx,
 		d_type = fs_ftype_to_dtype(btrfs_dir_flags_to_ftype(di->type));
 		btrfs_disk_key_to_cpu(&location, &di->location);
 
-		over = !dir_emit(ctx, name, name_len,
-			       location.objectid, d_type);
+		over = !dir_emit(ctx, name, name_len, location.objectid, d_type);
 
 		if (refcount_dec_and_test(&curr->refs))
 			kfree(curr);
 
 		if (over)
-			return 1;
+			return true;
 		ctx->pos++;
 	}
-	return 0;
+	return false;
 }
 
 static void fill_stack_inode_item(struct btrfs_trans_handle *trans,
 				  struct btrfs_inode_item *inode_item,
-				  struct inode *inode)
+				  struct btrfs_inode *inode)
 {
+	struct inode *vfs_inode = &inode->vfs_inode;
 	u64 flags;
 
-	btrfs_set_stack_inode_uid(inode_item, i_uid_read(inode));
-	btrfs_set_stack_inode_gid(inode_item, i_gid_read(inode));
-	btrfs_set_stack_inode_size(inode_item, BTRFS_I(inode)->disk_i_size);
-	btrfs_set_stack_inode_mode(inode_item, inode->i_mode);
-	btrfs_set_stack_inode_nlink(inode_item, inode->i_nlink);
-	btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(inode));
-	btrfs_set_stack_inode_generation(inode_item,
-					 BTRFS_I(inode)->generation);
+	btrfs_set_stack_inode_uid(inode_item, i_uid_read(vfs_inode));
+	btrfs_set_stack_inode_gid(inode_item, i_gid_read(vfs_inode));
+	btrfs_set_stack_inode_size(inode_item, inode->disk_i_size);
+	btrfs_set_stack_inode_mode(inode_item, vfs_inode->i_mode);
+	btrfs_set_stack_inode_nlink(inode_item, vfs_inode->i_nlink);
+	btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(vfs_inode));
+	btrfs_set_stack_inode_generation(inode_item, inode->generation);
 	btrfs_set_stack_inode_sequence(inode_item,
-				       inode_peek_iversion(inode));
+				       inode_peek_iversion(vfs_inode));
 	btrfs_set_stack_inode_transid(inode_item, trans->transid);
-	btrfs_set_stack_inode_rdev(inode_item, inode->i_rdev);
-	flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
-					  BTRFS_I(inode)->ro_flags);
+	btrfs_set_stack_inode_rdev(inode_item, vfs_inode->i_rdev);
+	flags = btrfs_inode_combine_flags(inode->flags, inode->ro_flags);
 	btrfs_set_stack_inode_flags(inode_item, flags);
 	btrfs_set_stack_inode_block_group(inode_item, 0);
 
 	btrfs_set_stack_timespec_sec(&inode_item->atime,
-				     inode->i_atime.tv_sec);
+				     inode_get_atime_sec(vfs_inode));
 	btrfs_set_stack_timespec_nsec(&inode_item->atime,
-				      inode->i_atime.tv_nsec);
+				      inode_get_atime_nsec(vfs_inode));
 
 	btrfs_set_stack_timespec_sec(&inode_item->mtime,
-				     inode->i_mtime.tv_sec);
+				     inode_get_mtime_sec(vfs_inode));
 	btrfs_set_stack_timespec_nsec(&inode_item->mtime,
-				      inode->i_mtime.tv_nsec);
+				      inode_get_mtime_nsec(vfs_inode));
 
 	btrfs_set_stack_timespec_sec(&inode_item->ctime,
-				     inode->i_ctime.tv_sec);
+				     inode_get_ctime_sec(vfs_inode));
 	btrfs_set_stack_timespec_nsec(&inode_item->ctime,
-				      inode->i_ctime.tv_nsec);
+				      inode_get_ctime_nsec(vfs_inode));
 
-	btrfs_set_stack_timespec_sec(&inode_item->otime,
-				     BTRFS_I(inode)->i_otime.tv_sec);
-	btrfs_set_stack_timespec_nsec(&inode_item->otime,
-				     BTRFS_I(inode)->i_otime.tv_nsec);
+	btrfs_set_stack_timespec_sec(&inode_item->otime, inode->i_otime_sec);
+	btrfs_set_stack_timespec_nsec(&inode_item->otime, inode->i_otime_nsec);
 }
 
-int btrfs_fill_inode(struct inode *inode, u32 *rdev)
+int btrfs_fill_inode(struct btrfs_inode *inode, u32 *rdev)
 {
-	struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
 	struct btrfs_delayed_node *delayed_node;
+	struct btrfs_ref_tracker delayed_node_tracker;
 	struct btrfs_inode_item *inode_item;
+	struct inode *vfs_inode = &inode->vfs_inode;
 
-	delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
+	delayed_node = btrfs_get_delayed_node(inode, &delayed_node_tracker);
 	if (!delayed_node)
 		return -ENOENT;
 
 	mutex_lock(&delayed_node->mutex);
 	if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
 		mutex_unlock(&delayed_node->mutex);
-		btrfs_release_delayed_node(delayed_node);
+		btrfs_release_delayed_node(delayed_node, &delayed_node_tracker);
 		return -ENOENT;
 	}
 
 	inode_item = &delayed_node->inode_item;
 
-	i_uid_write(inode, btrfs_stack_inode_uid(inode_item));
-	i_gid_write(inode, btrfs_stack_inode_gid(inode_item));
-	btrfs_i_size_write(BTRFS_I(inode), btrfs_stack_inode_size(inode_item));
-	btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
-			round_up(i_size_read(inode), fs_info->sectorsize));
-	inode->i_mode = btrfs_stack_inode_mode(inode_item);
-	set_nlink(inode, btrfs_stack_inode_nlink(inode_item));
-	inode_set_bytes(inode, btrfs_stack_inode_nbytes(inode_item));
-	BTRFS_I(inode)->generation = btrfs_stack_inode_generation(inode_item);
-        BTRFS_I(inode)->last_trans = btrfs_stack_inode_transid(inode_item);
-
-	inode_set_iversion_queried(inode,
-				   btrfs_stack_inode_sequence(inode_item));
-	inode->i_rdev = 0;
+	i_uid_write(vfs_inode, btrfs_stack_inode_uid(inode_item));
+	i_gid_write(vfs_inode, btrfs_stack_inode_gid(inode_item));
+	btrfs_i_size_write(inode, btrfs_stack_inode_size(inode_item));
+	vfs_inode->i_mode = btrfs_stack_inode_mode(inode_item);
+	set_nlink(vfs_inode, btrfs_stack_inode_nlink(inode_item));
+	inode_set_bytes(vfs_inode, btrfs_stack_inode_nbytes(inode_item));
+	inode->generation = btrfs_stack_inode_generation(inode_item);
+	inode->last_trans = btrfs_stack_inode_transid(inode_item);
+
+	inode_set_iversion_queried(vfs_inode, btrfs_stack_inode_sequence(inode_item));
+	vfs_inode->i_rdev = 0;
 	*rdev = btrfs_stack_inode_rdev(inode_item);
 	btrfs_inode_split_flags(btrfs_stack_inode_flags(inode_item),
-				&BTRFS_I(inode)->flags, &BTRFS_I(inode)->ro_flags);
+				&inode->flags, &inode->ro_flags);
 
-	inode->i_atime.tv_sec = btrfs_stack_timespec_sec(&inode_item->atime);
-	inode->i_atime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->atime);
+	inode_set_atime(vfs_inode, btrfs_stack_timespec_sec(&inode_item->atime),
+			btrfs_stack_timespec_nsec(&inode_item->atime));
 
-	inode->i_mtime.tv_sec = btrfs_stack_timespec_sec(&inode_item->mtime);
-	inode->i_mtime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->mtime);
+	inode_set_mtime(vfs_inode, btrfs_stack_timespec_sec(&inode_item->mtime),
+			btrfs_stack_timespec_nsec(&inode_item->mtime));
 
-	inode->i_ctime.tv_sec = btrfs_stack_timespec_sec(&inode_item->ctime);
-	inode->i_ctime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->ctime);
+	inode_set_ctime(vfs_inode, btrfs_stack_timespec_sec(&inode_item->ctime),
+			btrfs_stack_timespec_nsec(&inode_item->ctime));
 
-	BTRFS_I(inode)->i_otime.tv_sec =
-		btrfs_stack_timespec_sec(&inode_item->otime);
-	BTRFS_I(inode)->i_otime.tv_nsec =
-		btrfs_stack_timespec_nsec(&inode_item->otime);
+	inode->i_otime_sec = btrfs_stack_timespec_sec(&inode_item->otime);
+	inode->i_otime_nsec = btrfs_stack_timespec_nsec(&inode_item->otime);
 
-	inode->i_generation = BTRFS_I(inode)->generation;
-	BTRFS_I(inode)->index_cnt = (u64)-1;
+	vfs_inode->i_generation = inode->generation;
+	if (S_ISDIR(vfs_inode->i_mode))
+		inode->index_cnt = (u64)-1;
 
 	mutex_unlock(&delayed_node->mutex);
-	btrfs_release_delayed_node(delayed_node);
+	btrfs_release_delayed_node(delayed_node, &delayed_node_tracker);
 	return 0;
 }
 
 int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans,
-			       struct btrfs_root *root,
 			       struct btrfs_inode *inode)
 {
+	struct btrfs_root *root = inode->root;
 	struct btrfs_delayed_node *delayed_node;
+	struct btrfs_ref_tracker delayed_node_tracker;
 	int ret = 0;
 
-	delayed_node = btrfs_get_or_create_delayed_node(inode);
+	delayed_node = btrfs_get_or_create_delayed_node(inode, &delayed_node_tracker);
 	if (IS_ERR(delayed_node))
 		return PTR_ERR(delayed_node);
 
 	mutex_lock(&delayed_node->mutex);
 	if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
-		fill_stack_inode_item(trans, &delayed_node->inode_item,
-				      &inode->vfs_inode);
+		fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
 		goto release_node;
 	}
 
@@ -1899,13 +1965,13 @@ int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans,
 	if (ret)
 		goto release_node;
 
-	fill_stack_inode_item(trans, &delayed_node->inode_item, &inode->vfs_inode);
+	fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
 	set_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
 	delayed_node->count++;
 	atomic_inc(&root->fs_info->delayed_root->items);
 release_node:
 	mutex_unlock(&delayed_node->mutex);
-	btrfs_release_delayed_node(delayed_node);
+	btrfs_release_delayed_node(delayed_node, &delayed_node_tracker);
 	return ret;
 }
 
@@ -1913,6 +1979,7 @@ int btrfs_delayed_delete_inode_ref(struct btrfs_inode *inode)
 {
 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
 	struct btrfs_delayed_node *delayed_node;
+	struct btrfs_ref_tracker delayed_node_tracker;
 
 	/*
 	 * we don't do delayed inode updates during log recovery because it
@@ -1922,7 +1989,7 @@ int btrfs_delayed_delete_inode_ref(struct btrfs_inode *inode)
 	if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
 		return -EAGAIN;
 
-	delayed_node = btrfs_get_or_create_delayed_node(inode);
+	delayed_node = btrfs_get_or_create_delayed_node(inode, &delayed_node_tracker);
 	if (IS_ERR(delayed_node))
 		return PTR_ERR(delayed_node);
 
@@ -1941,15 +2008,12 @@ int btrfs_delayed_delete_inode_ref(struct btrfs_inode *inode)
 	 *   It is very rare.
 	 */
 	mutex_lock(&delayed_node->mutex);
-	if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
-		goto release_node;
-
-	set_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
-	delayed_node->count++;
-	atomic_inc(&fs_info->delayed_root->items);
-release_node:
+	if (!test_and_set_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags)) {
+		delayed_node->count++;
+		atomic_inc(&fs_info->delayed_root->items);
+	}
 	mutex_unlock(&delayed_node->mutex);
-	btrfs_release_delayed_node(delayed_node);
+	btrfs_release_delayed_node(delayed_node, &delayed_node_tracker);
 	return 0;
 }
 
@@ -1993,47 +2057,56 @@ static void __btrfs_kill_delayed_node(struct btrfs_delayed_node *delayed_node)
 void btrfs_kill_delayed_inode_items(struct btrfs_inode *inode)
 {
 	struct btrfs_delayed_node *delayed_node;
+	struct btrfs_ref_tracker delayed_node_tracker;
 
-	delayed_node = btrfs_get_delayed_node(inode);
+	delayed_node = btrfs_get_delayed_node(inode, &delayed_node_tracker);
 	if (!delayed_node)
 		return;
 
 	__btrfs_kill_delayed_node(delayed_node);
-	btrfs_release_delayed_node(delayed_node);
+	btrfs_release_delayed_node(delayed_node, &delayed_node_tracker);
 }
 
 void btrfs_kill_all_delayed_nodes(struct btrfs_root *root)
 {
-	u64 inode_id = 0;
+	unsigned long index = 0;
 	struct btrfs_delayed_node *delayed_nodes[8];
-	int i, n;
+	struct btrfs_ref_tracker delayed_node_trackers[8];
 
 	while (1) {
-		spin_lock(&root->inode_lock);
-		n = radix_tree_gang_lookup(&root->delayed_nodes_tree,
-					   (void **)delayed_nodes, inode_id,
-					   ARRAY_SIZE(delayed_nodes));
-		if (!n) {
-			spin_unlock(&root->inode_lock);
-			break;
+		struct btrfs_delayed_node *node;
+		int count;
+
+		xa_lock(&root->delayed_nodes);
+		if (xa_empty(&root->delayed_nodes)) {
+			xa_unlock(&root->delayed_nodes);
+			return;
 		}
 
-		inode_id = delayed_nodes[n - 1]->inode_id + 1;
-		for (i = 0; i < n; i++) {
+		count = 0;
+		xa_for_each_start(&root->delayed_nodes, index, node, index) {
 			/*
 			 * Don't increase refs in case the node is dead and
 			 * about to be removed from the tree in the loop below
 			 */
-			if (!refcount_inc_not_zero(&delayed_nodes[i]->refs))
-				delayed_nodes[i] = NULL;
+			if (refcount_inc_not_zero(&node->refs)) {
+				btrfs_delayed_node_ref_tracker_alloc(node,
+						     &delayed_node_trackers[count],
+						     GFP_ATOMIC);
+				delayed_nodes[count] = node;
+				count++;
+			}
+			if (count >= ARRAY_SIZE(delayed_nodes))
+				break;
 		}
-		spin_unlock(&root->inode_lock);
+		xa_unlock(&root->delayed_nodes);
+		index++;
 
-		for (i = 0; i < n; i++) {
-			if (!delayed_nodes[i])
-				continue;
+		for (int i = 0; i < count; i++) {
 			__btrfs_kill_delayed_node(delayed_nodes[i]);
-			btrfs_release_delayed_node(delayed_nodes[i]);
+			btrfs_delayed_node_ref_tracker_dir_print(delayed_nodes[i]);
+			btrfs_release_delayed_node(delayed_nodes[i],
+						   &delayed_node_trackers[i]);
 		}
 	}
 }
@@ -2041,14 +2114,17 @@ void btrfs_kill_all_delayed_nodes(struct btrfs_root *root)
 void btrfs_destroy_delayed_inodes(struct btrfs_fs_info *fs_info)
 {
 	struct btrfs_delayed_node *curr_node, *prev_node;
+	struct btrfs_ref_tracker curr_delayed_node_tracker, prev_delayed_node_tracker;
 
-	curr_node = btrfs_first_delayed_node(fs_info->delayed_root);
+	curr_node = btrfs_first_delayed_node(fs_info->delayed_root,
+					     &curr_delayed_node_tracker);
 	while (curr_node) {
 		__btrfs_kill_delayed_node(curr_node);
 
 		prev_node = curr_node;
-		curr_node = btrfs_next_delayed_node(curr_node);
-		btrfs_release_delayed_node(prev_node);
+		prev_delayed_node_tracker = curr_delayed_node_tracker;
+		curr_node = btrfs_next_delayed_node(curr_node, &curr_delayed_node_tracker);
+		btrfs_release_delayed_node(prev_node, &prev_delayed_node_tracker);
 	}
 }
 
@@ -2058,8 +2134,9 @@ void btrfs_log_get_delayed_items(struct btrfs_inode *inode,
 {
 	struct btrfs_delayed_node *node;
 	struct btrfs_delayed_item *item;
+	struct btrfs_ref_tracker delayed_node_tracker;
 
-	node = btrfs_get_delayed_node(inode);
+	node = btrfs_get_delayed_node(inode, &delayed_node_tracker);
 	if (!node)
 		return;
 
@@ -2117,6 +2194,7 @@ void btrfs_log_get_delayed_items(struct btrfs_inode *inode,
 	 * delete delayed items.
 	 */
 	ASSERT(refcount_read(&node->refs) > 1);
+	btrfs_delayed_node_ref_tracker_free(node, &delayed_node_tracker);
 	refcount_dec(&node->refs);
 }
 
@@ -2127,8 +2205,9 @@ void btrfs_log_put_delayed_items(struct btrfs_inode *inode,
 	struct btrfs_delayed_node *node;
 	struct btrfs_delayed_item *item;
 	struct btrfs_delayed_item *next;
+	struct btrfs_ref_tracker delayed_node_tracker;
 
-	node = btrfs_get_delayed_node(inode);
+	node = btrfs_get_delayed_node(inode, &delayed_node_tracker);
 	if (!node)
 		return;
 
@@ -2160,5 +2239,6 @@ void btrfs_log_put_delayed_items(struct btrfs_inode *inode,
 	 * delete delayed items.
 	 */
 	ASSERT(refcount_read(&node->refs) > 1);
+	btrfs_delayed_node_ref_tracker_free(node, &delayed_node_tracker);
 	refcount_dec(&node->refs);
 }