xref: /openbmc/linux/fs/btrfs/delayed-inode.c (revision 423997ff)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2011 Fujitsu.  All rights reserved.
4  * Written by Miao Xie <miaox@cn.fujitsu.com>
5  */
6 
7 #include <linux/slab.h>
8 #include <linux/iversion.h>
9 #include "delayed-inode.h"
10 #include "disk-io.h"
11 #include "transaction.h"
12 #include "ctree.h"
13 #include "qgroup.h"
14 
15 #define BTRFS_DELAYED_WRITEBACK		512
16 #define BTRFS_DELAYED_BACKGROUND	128
17 #define BTRFS_DELAYED_BATCH		16
18 
19 static struct kmem_cache *delayed_node_cache;
20 
21 int __init btrfs_delayed_inode_init(void)
22 {
23 	delayed_node_cache = kmem_cache_create("btrfs_delayed_node",
24 					sizeof(struct btrfs_delayed_node),
25 					0,
26 					SLAB_MEM_SPREAD,
27 					NULL);
28 	if (!delayed_node_cache)
29 		return -ENOMEM;
30 	return 0;
31 }
32 
33 void __cold btrfs_delayed_inode_exit(void)
34 {
35 	kmem_cache_destroy(delayed_node_cache);
36 }
37 
38 static inline void btrfs_init_delayed_node(
39 				struct btrfs_delayed_node *delayed_node,
40 				struct btrfs_root *root, u64 inode_id)
41 {
42 	delayed_node->root = root;
43 	delayed_node->inode_id = inode_id;
44 	refcount_set(&delayed_node->refs, 0);
45 	delayed_node->ins_root = RB_ROOT_CACHED;
46 	delayed_node->del_root = RB_ROOT_CACHED;
47 	mutex_init(&delayed_node->mutex);
48 	INIT_LIST_HEAD(&delayed_node->n_list);
49 	INIT_LIST_HEAD(&delayed_node->p_list);
50 }
51 
52 static inline int btrfs_is_continuous_delayed_item(
53 					struct btrfs_delayed_item *item1,
54 					struct btrfs_delayed_item *item2)
55 {
56 	if (item1->key.type == BTRFS_DIR_INDEX_KEY &&
57 	    item1->key.objectid == item2->key.objectid &&
58 	    item1->key.type == item2->key.type &&
59 	    item1->key.offset + 1 == item2->key.offset)
60 		return 1;
61 	return 0;
62 }
63 
64 static struct btrfs_delayed_node *btrfs_get_delayed_node(
65 		struct btrfs_inode *btrfs_inode)
66 {
67 	struct btrfs_root *root = btrfs_inode->root;
68 	u64 ino = btrfs_ino(btrfs_inode);
69 	struct btrfs_delayed_node *node;
70 
71 	node = READ_ONCE(btrfs_inode->delayed_node);
72 	if (node) {
73 		refcount_inc(&node->refs);
74 		return node;
75 	}
76 
77 	spin_lock(&root->inode_lock);
78 	node = radix_tree_lookup(&root->delayed_nodes_tree, ino);
79 
80 	if (node) {
81 		if (btrfs_inode->delayed_node) {
82 			refcount_inc(&node->refs);	/* can be accessed */
83 			BUG_ON(btrfs_inode->delayed_node != node);
84 			spin_unlock(&root->inode_lock);
85 			return node;
86 		}
87 
88 		/*
89 		 * It's possible that we're racing into the middle of removing
90 		 * this node from the radix tree.  In this case, the refcount
91 		 * was zero and it should never go back to one.  Just return
92 		 * NULL like it was never in the radix at all; our release
93 		 * function is in the process of removing it.
94 		 *
95 		 * Some implementations of refcount_inc refuse to bump the
96 		 * refcount once it has hit zero.  If we don't do this dance
97 		 * here, refcount_inc() may decide to just WARN_ONCE() instead
98 		 * of actually bumping the refcount.
99 		 *
100 		 * If this node is properly in the radix, we want to bump the
101 		 * refcount twice, once for the inode and once for this get
102 		 * operation.
103 		 */
104 		if (refcount_inc_not_zero(&node->refs)) {
105 			refcount_inc(&node->refs);
106 			btrfs_inode->delayed_node = node;
107 		} else {
108 			node = NULL;
109 		}
110 
111 		spin_unlock(&root->inode_lock);
112 		return node;
113 	}
114 	spin_unlock(&root->inode_lock);
115 
116 	return NULL;
117 }
118 
119 /* Will return either the node or PTR_ERR(-ENOMEM) */
120 static struct btrfs_delayed_node *btrfs_get_or_create_delayed_node(
121 		struct btrfs_inode *btrfs_inode)
122 {
123 	struct btrfs_delayed_node *node;
124 	struct btrfs_root *root = btrfs_inode->root;
125 	u64 ino = btrfs_ino(btrfs_inode);
126 	int ret;
127 
128 again:
129 	node = btrfs_get_delayed_node(btrfs_inode);
130 	if (node)
131 		return node;
132 
133 	node = kmem_cache_zalloc(delayed_node_cache, GFP_NOFS);
134 	if (!node)
135 		return ERR_PTR(-ENOMEM);
136 	btrfs_init_delayed_node(node, root, ino);
137 
138 	/* cached in the btrfs inode and can be accessed */
139 	refcount_set(&node->refs, 2);
140 
141 	ret = radix_tree_preload(GFP_NOFS);
142 	if (ret) {
143 		kmem_cache_free(delayed_node_cache, node);
144 		return ERR_PTR(ret);
145 	}
146 
147 	spin_lock(&root->inode_lock);
148 	ret = radix_tree_insert(&root->delayed_nodes_tree, ino, node);
149 	if (ret == -EEXIST) {
150 		spin_unlock(&root->inode_lock);
151 		kmem_cache_free(delayed_node_cache, node);
152 		radix_tree_preload_end();
153 		goto again;
154 	}
155 	btrfs_inode->delayed_node = node;
156 	spin_unlock(&root->inode_lock);
157 	radix_tree_preload_end();
158 
159 	return node;
160 }
161 
162 /*
163  * Call it when holding delayed_node->mutex
164  *
165  * If mod = 1, add this node into the prepared list.
166  */
167 static void btrfs_queue_delayed_node(struct btrfs_delayed_root *root,
168 				     struct btrfs_delayed_node *node,
169 				     int mod)
170 {
171 	spin_lock(&root->lock);
172 	if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
173 		if (!list_empty(&node->p_list))
174 			list_move_tail(&node->p_list, &root->prepare_list);
175 		else if (mod)
176 			list_add_tail(&node->p_list, &root->prepare_list);
177 	} else {
178 		list_add_tail(&node->n_list, &root->node_list);
179 		list_add_tail(&node->p_list, &root->prepare_list);
180 		refcount_inc(&node->refs);	/* inserted into list */
181 		root->nodes++;
182 		set_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
183 	}
184 	spin_unlock(&root->lock);
185 }
186 
187 /* Call it when holding delayed_node->mutex */
188 static void btrfs_dequeue_delayed_node(struct btrfs_delayed_root *root,
189 				       struct btrfs_delayed_node *node)
190 {
191 	spin_lock(&root->lock);
192 	if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
193 		root->nodes--;
194 		refcount_dec(&node->refs);	/* not in the list */
195 		list_del_init(&node->n_list);
196 		if (!list_empty(&node->p_list))
197 			list_del_init(&node->p_list);
198 		clear_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
199 	}
200 	spin_unlock(&root->lock);
201 }
202 
203 static struct btrfs_delayed_node *btrfs_first_delayed_node(
204 			struct btrfs_delayed_root *delayed_root)
205 {
206 	struct list_head *p;
207 	struct btrfs_delayed_node *node = NULL;
208 
209 	spin_lock(&delayed_root->lock);
210 	if (list_empty(&delayed_root->node_list))
211 		goto out;
212 
213 	p = delayed_root->node_list.next;
214 	node = list_entry(p, struct btrfs_delayed_node, n_list);
215 	refcount_inc(&node->refs);
216 out:
217 	spin_unlock(&delayed_root->lock);
218 
219 	return node;
220 }
221 
222 static struct btrfs_delayed_node *btrfs_next_delayed_node(
223 						struct btrfs_delayed_node *node)
224 {
225 	struct btrfs_delayed_root *delayed_root;
226 	struct list_head *p;
227 	struct btrfs_delayed_node *next = NULL;
228 
229 	delayed_root = node->root->fs_info->delayed_root;
230 	spin_lock(&delayed_root->lock);
231 	if (!test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
232 		/* not in the list */
233 		if (list_empty(&delayed_root->node_list))
234 			goto out;
235 		p = delayed_root->node_list.next;
236 	} else if (list_is_last(&node->n_list, &delayed_root->node_list))
237 		goto out;
238 	else
239 		p = node->n_list.next;
240 
241 	next = list_entry(p, struct btrfs_delayed_node, n_list);
242 	refcount_inc(&next->refs);
243 out:
244 	spin_unlock(&delayed_root->lock);
245 
246 	return next;
247 }
248 
249 static void __btrfs_release_delayed_node(
250 				struct btrfs_delayed_node *delayed_node,
251 				int mod)
252 {
253 	struct btrfs_delayed_root *delayed_root;
254 
255 	if (!delayed_node)
256 		return;
257 
258 	delayed_root = delayed_node->root->fs_info->delayed_root;
259 
260 	mutex_lock(&delayed_node->mutex);
261 	if (delayed_node->count)
262 		btrfs_queue_delayed_node(delayed_root, delayed_node, mod);
263 	else
264 		btrfs_dequeue_delayed_node(delayed_root, delayed_node);
265 	mutex_unlock(&delayed_node->mutex);
266 
267 	if (refcount_dec_and_test(&delayed_node->refs)) {
268 		struct btrfs_root *root = delayed_node->root;
269 
270 		spin_lock(&root->inode_lock);
271 		/*
272 		 * Once our refcount goes to zero, nobody is allowed to bump it
273 		 * back up.  We can delete it now.
274 		 */
275 		ASSERT(refcount_read(&delayed_node->refs) == 0);
276 		radix_tree_delete(&root->delayed_nodes_tree,
277 				  delayed_node->inode_id);
278 		spin_unlock(&root->inode_lock);
279 		kmem_cache_free(delayed_node_cache, delayed_node);
280 	}
281 }
282 
283 static inline void btrfs_release_delayed_node(struct btrfs_delayed_node *node)
284 {
285 	__btrfs_release_delayed_node(node, 0);
286 }
287 
288 static struct btrfs_delayed_node *btrfs_first_prepared_delayed_node(
289 					struct btrfs_delayed_root *delayed_root)
290 {
291 	struct list_head *p;
292 	struct btrfs_delayed_node *node = NULL;
293 
294 	spin_lock(&delayed_root->lock);
295 	if (list_empty(&delayed_root->prepare_list))
296 		goto out;
297 
298 	p = delayed_root->prepare_list.next;
299 	list_del_init(p);
300 	node = list_entry(p, struct btrfs_delayed_node, p_list);
301 	refcount_inc(&node->refs);
302 out:
303 	spin_unlock(&delayed_root->lock);
304 
305 	return node;
306 }
307 
308 static inline void btrfs_release_prepared_delayed_node(
309 					struct btrfs_delayed_node *node)
310 {
311 	__btrfs_release_delayed_node(node, 1);
312 }
313 
314 static struct btrfs_delayed_item *btrfs_alloc_delayed_item(u32 data_len)
315 {
316 	struct btrfs_delayed_item *item;
317 	item = kmalloc(sizeof(*item) + data_len, GFP_NOFS);
318 	if (item) {
319 		item->data_len = data_len;
320 		item->ins_or_del = 0;
321 		item->bytes_reserved = 0;
322 		item->delayed_node = NULL;
323 		refcount_set(&item->refs, 1);
324 	}
325 	return item;
326 }
327 
328 /*
329  * __btrfs_lookup_delayed_item - look up the delayed item by key
330  * @delayed_node: pointer to the delayed node
331  * @key:	  the key to look up
332  * @prev:	  used to store the prev item if the right item isn't found
333  * @next:	  used to store the next item if the right item isn't found
334  *
335  * Note: if we don't find the right item, we will return the prev item and
336  * the next item.
337  */
338 static struct btrfs_delayed_item *__btrfs_lookup_delayed_item(
339 				struct rb_root *root,
340 				struct btrfs_key *key,
341 				struct btrfs_delayed_item **prev,
342 				struct btrfs_delayed_item **next)
343 {
344 	struct rb_node *node, *prev_node = NULL;
345 	struct btrfs_delayed_item *delayed_item = NULL;
346 	int ret = 0;
347 
348 	node = root->rb_node;
349 
350 	while (node) {
351 		delayed_item = rb_entry(node, struct btrfs_delayed_item,
352 					rb_node);
353 		prev_node = node;
354 		ret = btrfs_comp_cpu_keys(&delayed_item->key, key);
355 		if (ret < 0)
356 			node = node->rb_right;
357 		else if (ret > 0)
358 			node = node->rb_left;
359 		else
360 			return delayed_item;
361 	}
362 
363 	if (prev) {
364 		if (!prev_node)
365 			*prev = NULL;
366 		else if (ret < 0)
367 			*prev = delayed_item;
368 		else if ((node = rb_prev(prev_node)) != NULL) {
369 			*prev = rb_entry(node, struct btrfs_delayed_item,
370 					 rb_node);
371 		} else
372 			*prev = NULL;
373 	}
374 
375 	if (next) {
376 		if (!prev_node)
377 			*next = NULL;
378 		else if (ret > 0)
379 			*next = delayed_item;
380 		else if ((node = rb_next(prev_node)) != NULL) {
381 			*next = rb_entry(node, struct btrfs_delayed_item,
382 					 rb_node);
383 		} else
384 			*next = NULL;
385 	}
386 	return NULL;
387 }
388 
389 static struct btrfs_delayed_item *__btrfs_lookup_delayed_insertion_item(
390 					struct btrfs_delayed_node *delayed_node,
391 					struct btrfs_key *key)
392 {
393 	return __btrfs_lookup_delayed_item(&delayed_node->ins_root.rb_root, key,
394 					   NULL, NULL);
395 }
396 
397 static int __btrfs_add_delayed_item(struct btrfs_delayed_node *delayed_node,
398 				    struct btrfs_delayed_item *ins,
399 				    int action)
400 {
401 	struct rb_node **p, *node;
402 	struct rb_node *parent_node = NULL;
403 	struct rb_root_cached *root;
404 	struct btrfs_delayed_item *item;
405 	int cmp;
406 	bool leftmost = true;
407 
408 	if (action == BTRFS_DELAYED_INSERTION_ITEM)
409 		root = &delayed_node->ins_root;
410 	else if (action == BTRFS_DELAYED_DELETION_ITEM)
411 		root = &delayed_node->del_root;
412 	else
413 		BUG();
414 	p = &root->rb_root.rb_node;
415 	node = &ins->rb_node;
416 
417 	while (*p) {
418 		parent_node = *p;
419 		item = rb_entry(parent_node, struct btrfs_delayed_item,
420 				 rb_node);
421 
422 		cmp = btrfs_comp_cpu_keys(&item->key, &ins->key);
423 		if (cmp < 0) {
424 			p = &(*p)->rb_right;
425 			leftmost = false;
426 		} else if (cmp > 0) {
427 			p = &(*p)->rb_left;
428 		} else {
429 			return -EEXIST;
430 		}
431 	}
432 
433 	rb_link_node(node, parent_node, p);
434 	rb_insert_color_cached(node, root, leftmost);
435 	ins->delayed_node = delayed_node;
436 	ins->ins_or_del = action;
437 
438 	if (ins->key.type == BTRFS_DIR_INDEX_KEY &&
439 	    action == BTRFS_DELAYED_INSERTION_ITEM &&
440 	    ins->key.offset >= delayed_node->index_cnt)
441 			delayed_node->index_cnt = ins->key.offset + 1;
442 
443 	delayed_node->count++;
444 	atomic_inc(&delayed_node->root->fs_info->delayed_root->items);
445 	return 0;
446 }
447 
448 static int __btrfs_add_delayed_insertion_item(struct btrfs_delayed_node *node,
449 					      struct btrfs_delayed_item *item)
450 {
451 	return __btrfs_add_delayed_item(node, item,
452 					BTRFS_DELAYED_INSERTION_ITEM);
453 }
454 
455 static int __btrfs_add_delayed_deletion_item(struct btrfs_delayed_node *node,
456 					     struct btrfs_delayed_item *item)
457 {
458 	return __btrfs_add_delayed_item(node, item,
459 					BTRFS_DELAYED_DELETION_ITEM);
460 }
461 
462 static void finish_one_item(struct btrfs_delayed_root *delayed_root)
463 {
464 	int seq = atomic_inc_return(&delayed_root->items_seq);
465 
466 	/* atomic_dec_return implies a barrier */
467 	if ((atomic_dec_return(&delayed_root->items) <
468 	    BTRFS_DELAYED_BACKGROUND || seq % BTRFS_DELAYED_BATCH == 0))
469 		cond_wake_up_nomb(&delayed_root->wait);
470 }
471 
472 static void __btrfs_remove_delayed_item(struct btrfs_delayed_item *delayed_item)
473 {
474 	struct rb_root_cached *root;
475 	struct btrfs_delayed_root *delayed_root;
476 
477 	delayed_root = delayed_item->delayed_node->root->fs_info->delayed_root;
478 
479 	BUG_ON(!delayed_root);
480 	BUG_ON(delayed_item->ins_or_del != BTRFS_DELAYED_DELETION_ITEM &&
481 	       delayed_item->ins_or_del != BTRFS_DELAYED_INSERTION_ITEM);
482 
483 	if (delayed_item->ins_or_del == BTRFS_DELAYED_INSERTION_ITEM)
484 		root = &delayed_item->delayed_node->ins_root;
485 	else
486 		root = &delayed_item->delayed_node->del_root;
487 
488 	rb_erase_cached(&delayed_item->rb_node, root);
489 	delayed_item->delayed_node->count--;
490 
491 	finish_one_item(delayed_root);
492 }
493 
494 static void btrfs_release_delayed_item(struct btrfs_delayed_item *item)
495 {
496 	if (item) {
497 		__btrfs_remove_delayed_item(item);
498 		if (refcount_dec_and_test(&item->refs))
499 			kfree(item);
500 	}
501 }
502 
503 static struct btrfs_delayed_item *__btrfs_first_delayed_insertion_item(
504 					struct btrfs_delayed_node *delayed_node)
505 {
506 	struct rb_node *p;
507 	struct btrfs_delayed_item *item = NULL;
508 
509 	p = rb_first_cached(&delayed_node->ins_root);
510 	if (p)
511 		item = rb_entry(p, struct btrfs_delayed_item, rb_node);
512 
513 	return item;
514 }
515 
516 static struct btrfs_delayed_item *__btrfs_first_delayed_deletion_item(
517 					struct btrfs_delayed_node *delayed_node)
518 {
519 	struct rb_node *p;
520 	struct btrfs_delayed_item *item = NULL;
521 
522 	p = rb_first_cached(&delayed_node->del_root);
523 	if (p)
524 		item = rb_entry(p, struct btrfs_delayed_item, rb_node);
525 
526 	return item;
527 }
528 
529 static struct btrfs_delayed_item *__btrfs_next_delayed_item(
530 						struct btrfs_delayed_item *item)
531 {
532 	struct rb_node *p;
533 	struct btrfs_delayed_item *next = NULL;
534 
535 	p = rb_next(&item->rb_node);
536 	if (p)
537 		next = rb_entry(p, struct btrfs_delayed_item, rb_node);
538 
539 	return next;
540 }
541 
542 static int btrfs_delayed_item_reserve_metadata(struct btrfs_trans_handle *trans,
543 					       struct btrfs_root *root,
544 					       struct btrfs_delayed_item *item)
545 {
546 	struct btrfs_block_rsv *src_rsv;
547 	struct btrfs_block_rsv *dst_rsv;
548 	struct btrfs_fs_info *fs_info = root->fs_info;
549 	u64 num_bytes;
550 	int ret;
551 
552 	if (!trans->bytes_reserved)
553 		return 0;
554 
555 	src_rsv = trans->block_rsv;
556 	dst_rsv = &fs_info->delayed_block_rsv;
557 
558 	num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
559 
560 	/*
561 	 * Here we migrate space rsv from transaction rsv, since have already
562 	 * reserved space when starting a transaction.  So no need to reserve
563 	 * qgroup space here.
564 	 */
565 	ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, true);
566 	if (!ret) {
567 		trace_btrfs_space_reservation(fs_info, "delayed_item",
568 					      item->key.objectid,
569 					      num_bytes, 1);
570 		item->bytes_reserved = num_bytes;
571 	}
572 
573 	return ret;
574 }
575 
576 static void btrfs_delayed_item_release_metadata(struct btrfs_root *root,
577 						struct btrfs_delayed_item *item)
578 {
579 	struct btrfs_block_rsv *rsv;
580 	struct btrfs_fs_info *fs_info = root->fs_info;
581 
582 	if (!item->bytes_reserved)
583 		return;
584 
585 	rsv = &fs_info->delayed_block_rsv;
586 	/*
587 	 * Check btrfs_delayed_item_reserve_metadata() to see why we don't need
588 	 * to release/reserve qgroup space.
589 	 */
590 	trace_btrfs_space_reservation(fs_info, "delayed_item",
591 				      item->key.objectid, item->bytes_reserved,
592 				      0);
593 	btrfs_block_rsv_release(fs_info, rsv,
594 				item->bytes_reserved);
595 }
596 
597 static int btrfs_delayed_inode_reserve_metadata(
598 					struct btrfs_trans_handle *trans,
599 					struct btrfs_root *root,
600 					struct btrfs_inode *inode,
601 					struct btrfs_delayed_node *node)
602 {
603 	struct btrfs_fs_info *fs_info = root->fs_info;
604 	struct btrfs_block_rsv *src_rsv;
605 	struct btrfs_block_rsv *dst_rsv;
606 	u64 num_bytes;
607 	int ret;
608 
609 	src_rsv = trans->block_rsv;
610 	dst_rsv = &fs_info->delayed_block_rsv;
611 
612 	num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
613 
614 	/*
615 	 * btrfs_dirty_inode will update the inode under btrfs_join_transaction
616 	 * which doesn't reserve space for speed.  This is a problem since we
617 	 * still need to reserve space for this update, so try to reserve the
618 	 * space.
619 	 *
620 	 * Now if src_rsv == delalloc_block_rsv we'll let it just steal since
621 	 * we always reserve enough to update the inode item.
622 	 */
623 	if (!src_rsv || (!trans->bytes_reserved &&
624 			 src_rsv->type != BTRFS_BLOCK_RSV_DELALLOC)) {
625 		ret = btrfs_qgroup_reserve_meta_prealloc(root,
626 				fs_info->nodesize, true);
627 		if (ret < 0)
628 			return ret;
629 		ret = btrfs_block_rsv_add(root, dst_rsv, num_bytes,
630 					  BTRFS_RESERVE_NO_FLUSH);
631 		/*
632 		 * Since we're under a transaction reserve_metadata_bytes could
633 		 * try to commit the transaction which will make it return
634 		 * EAGAIN to make us stop the transaction we have, so return
635 		 * ENOSPC instead so that btrfs_dirty_inode knows what to do.
636 		 */
637 		if (ret == -EAGAIN) {
638 			ret = -ENOSPC;
639 			btrfs_qgroup_free_meta_prealloc(root, num_bytes);
640 		}
641 		if (!ret) {
642 			node->bytes_reserved = num_bytes;
643 			trace_btrfs_space_reservation(fs_info,
644 						      "delayed_inode",
645 						      btrfs_ino(inode),
646 						      num_bytes, 1);
647 		} else {
648 			btrfs_qgroup_free_meta_prealloc(root, fs_info->nodesize);
649 		}
650 		return ret;
651 	}
652 
653 	ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, true);
654 	if (!ret) {
655 		trace_btrfs_space_reservation(fs_info, "delayed_inode",
656 					      btrfs_ino(inode), num_bytes, 1);
657 		node->bytes_reserved = num_bytes;
658 	}
659 
660 	return ret;
661 }
662 
663 static void btrfs_delayed_inode_release_metadata(struct btrfs_fs_info *fs_info,
664 						struct btrfs_delayed_node *node,
665 						bool qgroup_free)
666 {
667 	struct btrfs_block_rsv *rsv;
668 
669 	if (!node->bytes_reserved)
670 		return;
671 
672 	rsv = &fs_info->delayed_block_rsv;
673 	trace_btrfs_space_reservation(fs_info, "delayed_inode",
674 				      node->inode_id, node->bytes_reserved, 0);
675 	btrfs_block_rsv_release(fs_info, rsv,
676 				node->bytes_reserved);
677 	if (qgroup_free)
678 		btrfs_qgroup_free_meta_prealloc(node->root,
679 				node->bytes_reserved);
680 	else
681 		btrfs_qgroup_convert_reserved_meta(node->root,
682 				node->bytes_reserved);
683 	node->bytes_reserved = 0;
684 }
685 
686 /*
687  * This helper will insert some continuous items into the same leaf according
688  * to the free space of the leaf.
689  */
690 static int btrfs_batch_insert_items(struct btrfs_root *root,
691 				    struct btrfs_path *path,
692 				    struct btrfs_delayed_item *item)
693 {
694 	struct btrfs_delayed_item *curr, *next;
695 	int free_space;
696 	int total_data_size = 0, total_size = 0;
697 	struct extent_buffer *leaf;
698 	char *data_ptr;
699 	struct btrfs_key *keys;
700 	u32 *data_size;
701 	struct list_head head;
702 	int slot;
703 	int nitems;
704 	int i;
705 	int ret = 0;
706 
707 	BUG_ON(!path->nodes[0]);
708 
709 	leaf = path->nodes[0];
710 	free_space = btrfs_leaf_free_space(leaf);
711 	INIT_LIST_HEAD(&head);
712 
713 	next = item;
714 	nitems = 0;
715 
716 	/*
717 	 * count the number of the continuous items that we can insert in batch
718 	 */
719 	while (total_size + next->data_len + sizeof(struct btrfs_item) <=
720 	       free_space) {
721 		total_data_size += next->data_len;
722 		total_size += next->data_len + sizeof(struct btrfs_item);
723 		list_add_tail(&next->tree_list, &head);
724 		nitems++;
725 
726 		curr = next;
727 		next = __btrfs_next_delayed_item(curr);
728 		if (!next)
729 			break;
730 
731 		if (!btrfs_is_continuous_delayed_item(curr, next))
732 			break;
733 	}
734 
735 	if (!nitems) {
736 		ret = 0;
737 		goto out;
738 	}
739 
740 	/*
741 	 * we need allocate some memory space, but it might cause the task
742 	 * to sleep, so we set all locked nodes in the path to blocking locks
743 	 * first.
744 	 */
745 	btrfs_set_path_blocking(path);
746 
747 	keys = kmalloc_array(nitems, sizeof(struct btrfs_key), GFP_NOFS);
748 	if (!keys) {
749 		ret = -ENOMEM;
750 		goto out;
751 	}
752 
753 	data_size = kmalloc_array(nitems, sizeof(u32), GFP_NOFS);
754 	if (!data_size) {
755 		ret = -ENOMEM;
756 		goto error;
757 	}
758 
759 	/* get keys of all the delayed items */
760 	i = 0;
761 	list_for_each_entry(next, &head, tree_list) {
762 		keys[i] = next->key;
763 		data_size[i] = next->data_len;
764 		i++;
765 	}
766 
767 	/* insert the keys of the items */
768 	setup_items_for_insert(root, path, keys, data_size,
769 			       total_data_size, total_size, nitems);
770 
771 	/* insert the dir index items */
772 	slot = path->slots[0];
773 	list_for_each_entry_safe(curr, next, &head, tree_list) {
774 		data_ptr = btrfs_item_ptr(leaf, slot, char);
775 		write_extent_buffer(leaf, &curr->data,
776 				    (unsigned long)data_ptr,
777 				    curr->data_len);
778 		slot++;
779 
780 		btrfs_delayed_item_release_metadata(root, curr);
781 
782 		list_del(&curr->tree_list);
783 		btrfs_release_delayed_item(curr);
784 	}
785 
786 error:
787 	kfree(data_size);
788 	kfree(keys);
789 out:
790 	return ret;
791 }
792 
793 /*
794  * This helper can just do simple insertion that needn't extend item for new
795  * data, such as directory name index insertion, inode insertion.
796  */
797 static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans,
798 				     struct btrfs_root *root,
799 				     struct btrfs_path *path,
800 				     struct btrfs_delayed_item *delayed_item)
801 {
802 	struct extent_buffer *leaf;
803 	char *ptr;
804 	int ret;
805 
806 	ret = btrfs_insert_empty_item(trans, root, path, &delayed_item->key,
807 				      delayed_item->data_len);
808 	if (ret < 0 && ret != -EEXIST)
809 		return ret;
810 
811 	leaf = path->nodes[0];
812 
813 	ptr = btrfs_item_ptr(leaf, path->slots[0], char);
814 
815 	write_extent_buffer(leaf, delayed_item->data, (unsigned long)ptr,
816 			    delayed_item->data_len);
817 	btrfs_mark_buffer_dirty(leaf);
818 
819 	btrfs_delayed_item_release_metadata(root, delayed_item);
820 	return 0;
821 }
822 
823 /*
824  * we insert an item first, then if there are some continuous items, we try
825  * to insert those items into the same leaf.
826  */
827 static int btrfs_insert_delayed_items(struct btrfs_trans_handle *trans,
828 				      struct btrfs_path *path,
829 				      struct btrfs_root *root,
830 				      struct btrfs_delayed_node *node)
831 {
832 	struct btrfs_delayed_item *curr, *prev;
833 	int ret = 0;
834 
835 do_again:
836 	mutex_lock(&node->mutex);
837 	curr = __btrfs_first_delayed_insertion_item(node);
838 	if (!curr)
839 		goto insert_end;
840 
841 	ret = btrfs_insert_delayed_item(trans, root, path, curr);
842 	if (ret < 0) {
843 		btrfs_release_path(path);
844 		goto insert_end;
845 	}
846 
847 	prev = curr;
848 	curr = __btrfs_next_delayed_item(prev);
849 	if (curr && btrfs_is_continuous_delayed_item(prev, curr)) {
850 		/* insert the continuous items into the same leaf */
851 		path->slots[0]++;
852 		btrfs_batch_insert_items(root, path, curr);
853 	}
854 	btrfs_release_delayed_item(prev);
855 	btrfs_mark_buffer_dirty(path->nodes[0]);
856 
857 	btrfs_release_path(path);
858 	mutex_unlock(&node->mutex);
859 	goto do_again;
860 
861 insert_end:
862 	mutex_unlock(&node->mutex);
863 	return ret;
864 }
865 
866 static int btrfs_batch_delete_items(struct btrfs_trans_handle *trans,
867 				    struct btrfs_root *root,
868 				    struct btrfs_path *path,
869 				    struct btrfs_delayed_item *item)
870 {
871 	struct btrfs_delayed_item *curr, *next;
872 	struct extent_buffer *leaf;
873 	struct btrfs_key key;
874 	struct list_head head;
875 	int nitems, i, last_item;
876 	int ret = 0;
877 
878 	BUG_ON(!path->nodes[0]);
879 
880 	leaf = path->nodes[0];
881 
882 	i = path->slots[0];
883 	last_item = btrfs_header_nritems(leaf) - 1;
884 	if (i > last_item)
885 		return -ENOENT;	/* FIXME: Is errno suitable? */
886 
887 	next = item;
888 	INIT_LIST_HEAD(&head);
889 	btrfs_item_key_to_cpu(leaf, &key, i);
890 	nitems = 0;
891 	/*
892 	 * count the number of the dir index items that we can delete in batch
893 	 */
894 	while (btrfs_comp_cpu_keys(&next->key, &key) == 0) {
895 		list_add_tail(&next->tree_list, &head);
896 		nitems++;
897 
898 		curr = next;
899 		next = __btrfs_next_delayed_item(curr);
900 		if (!next)
901 			break;
902 
903 		if (!btrfs_is_continuous_delayed_item(curr, next))
904 			break;
905 
906 		i++;
907 		if (i > last_item)
908 			break;
909 		btrfs_item_key_to_cpu(leaf, &key, i);
910 	}
911 
912 	if (!nitems)
913 		return 0;
914 
915 	ret = btrfs_del_items(trans, root, path, path->slots[0], nitems);
916 	if (ret)
917 		goto out;
918 
919 	list_for_each_entry_safe(curr, next, &head, tree_list) {
920 		btrfs_delayed_item_release_metadata(root, curr);
921 		list_del(&curr->tree_list);
922 		btrfs_release_delayed_item(curr);
923 	}
924 
925 out:
926 	return ret;
927 }
928 
929 static int btrfs_delete_delayed_items(struct btrfs_trans_handle *trans,
930 				      struct btrfs_path *path,
931 				      struct btrfs_root *root,
932 				      struct btrfs_delayed_node *node)
933 {
934 	struct btrfs_delayed_item *curr, *prev;
935 	int ret = 0;
936 
937 do_again:
938 	mutex_lock(&node->mutex);
939 	curr = __btrfs_first_delayed_deletion_item(node);
940 	if (!curr)
941 		goto delete_fail;
942 
943 	ret = btrfs_search_slot(trans, root, &curr->key, path, -1, 1);
944 	if (ret < 0)
945 		goto delete_fail;
946 	else if (ret > 0) {
947 		/*
948 		 * can't find the item which the node points to, so this node
949 		 * is invalid, just drop it.
950 		 */
951 		prev = curr;
952 		curr = __btrfs_next_delayed_item(prev);
953 		btrfs_release_delayed_item(prev);
954 		ret = 0;
955 		btrfs_release_path(path);
956 		if (curr) {
957 			mutex_unlock(&node->mutex);
958 			goto do_again;
959 		} else
960 			goto delete_fail;
961 	}
962 
963 	btrfs_batch_delete_items(trans, root, path, curr);
964 	btrfs_release_path(path);
965 	mutex_unlock(&node->mutex);
966 	goto do_again;
967 
968 delete_fail:
969 	btrfs_release_path(path);
970 	mutex_unlock(&node->mutex);
971 	return ret;
972 }
973 
974 static void btrfs_release_delayed_inode(struct btrfs_delayed_node *delayed_node)
975 {
976 	struct btrfs_delayed_root *delayed_root;
977 
978 	if (delayed_node &&
979 	    test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
980 		BUG_ON(!delayed_node->root);
981 		clear_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
982 		delayed_node->count--;
983 
984 		delayed_root = delayed_node->root->fs_info->delayed_root;
985 		finish_one_item(delayed_root);
986 	}
987 }
988 
989 static void btrfs_release_delayed_iref(struct btrfs_delayed_node *delayed_node)
990 {
991 	struct btrfs_delayed_root *delayed_root;
992 
993 	ASSERT(delayed_node->root);
994 	clear_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
995 	delayed_node->count--;
996 
997 	delayed_root = delayed_node->root->fs_info->delayed_root;
998 	finish_one_item(delayed_root);
999 }
1000 
1001 static int __btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1002 					struct btrfs_root *root,
1003 					struct btrfs_path *path,
1004 					struct btrfs_delayed_node *node)
1005 {
1006 	struct btrfs_fs_info *fs_info = root->fs_info;
1007 	struct btrfs_key key;
1008 	struct btrfs_inode_item *inode_item;
1009 	struct extent_buffer *leaf;
1010 	int mod;
1011 	int ret;
1012 
1013 	key.objectid = node->inode_id;
1014 	key.type = BTRFS_INODE_ITEM_KEY;
1015 	key.offset = 0;
1016 
1017 	if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1018 		mod = -1;
1019 	else
1020 		mod = 1;
1021 
1022 	ret = btrfs_lookup_inode(trans, root, path, &key, mod);
1023 	if (ret > 0) {
1024 		btrfs_release_path(path);
1025 		return -ENOENT;
1026 	} else if (ret < 0) {
1027 		return ret;
1028 	}
1029 
1030 	leaf = path->nodes[0];
1031 	inode_item = btrfs_item_ptr(leaf, path->slots[0],
1032 				    struct btrfs_inode_item);
1033 	write_extent_buffer(leaf, &node->inode_item, (unsigned long)inode_item,
1034 			    sizeof(struct btrfs_inode_item));
1035 	btrfs_mark_buffer_dirty(leaf);
1036 
1037 	if (!test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1038 		goto no_iref;
1039 
1040 	path->slots[0]++;
1041 	if (path->slots[0] >= btrfs_header_nritems(leaf))
1042 		goto search;
1043 again:
1044 	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1045 	if (key.objectid != node->inode_id)
1046 		goto out;
1047 
1048 	if (key.type != BTRFS_INODE_REF_KEY &&
1049 	    key.type != BTRFS_INODE_EXTREF_KEY)
1050 		goto out;
1051 
1052 	/*
1053 	 * Delayed iref deletion is for the inode who has only one link,
1054 	 * so there is only one iref. The case that several irefs are
1055 	 * in the same item doesn't exist.
1056 	 */
1057 	btrfs_del_item(trans, root, path);
1058 out:
1059 	btrfs_release_delayed_iref(node);
1060 no_iref:
1061 	btrfs_release_path(path);
1062 err_out:
1063 	btrfs_delayed_inode_release_metadata(fs_info, node, (ret < 0));
1064 	btrfs_release_delayed_inode(node);
1065 
1066 	return ret;
1067 
1068 search:
1069 	btrfs_release_path(path);
1070 
1071 	key.type = BTRFS_INODE_EXTREF_KEY;
1072 	key.offset = -1;
1073 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1074 	if (ret < 0)
1075 		goto err_out;
1076 	ASSERT(ret);
1077 
1078 	ret = 0;
1079 	leaf = path->nodes[0];
1080 	path->slots[0]--;
1081 	goto again;
1082 }
1083 
1084 static inline int btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1085 					     struct btrfs_root *root,
1086 					     struct btrfs_path *path,
1087 					     struct btrfs_delayed_node *node)
1088 {
1089 	int ret;
1090 
1091 	mutex_lock(&node->mutex);
1092 	if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &node->flags)) {
1093 		mutex_unlock(&node->mutex);
1094 		return 0;
1095 	}
1096 
1097 	ret = __btrfs_update_delayed_inode(trans, root, path, node);
1098 	mutex_unlock(&node->mutex);
1099 	return ret;
1100 }
1101 
1102 static inline int
1103 __btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1104 				   struct btrfs_path *path,
1105 				   struct btrfs_delayed_node *node)
1106 {
1107 	int ret;
1108 
1109 	ret = btrfs_insert_delayed_items(trans, path, node->root, node);
1110 	if (ret)
1111 		return ret;
1112 
1113 	ret = btrfs_delete_delayed_items(trans, path, node->root, node);
1114 	if (ret)
1115 		return ret;
1116 
1117 	ret = btrfs_update_delayed_inode(trans, node->root, path, node);
1118 	return ret;
1119 }
1120 
1121 /*
1122  * Called when committing the transaction.
1123  * Returns 0 on success.
1124  * Returns < 0 on error and returns with an aborted transaction with any
1125  * outstanding delayed items cleaned up.
1126  */
1127 static int __btrfs_run_delayed_items(struct btrfs_trans_handle *trans, int nr)
1128 {
1129 	struct btrfs_fs_info *fs_info = trans->fs_info;
1130 	struct btrfs_delayed_root *delayed_root;
1131 	struct btrfs_delayed_node *curr_node, *prev_node;
1132 	struct btrfs_path *path;
1133 	struct btrfs_block_rsv *block_rsv;
1134 	int ret = 0;
1135 	bool count = (nr > 0);
1136 
1137 	if (trans->aborted)
1138 		return -EIO;
1139 
1140 	path = btrfs_alloc_path();
1141 	if (!path)
1142 		return -ENOMEM;
1143 	path->leave_spinning = 1;
1144 
1145 	block_rsv = trans->block_rsv;
1146 	trans->block_rsv = &fs_info->delayed_block_rsv;
1147 
1148 	delayed_root = fs_info->delayed_root;
1149 
1150 	curr_node = btrfs_first_delayed_node(delayed_root);
1151 	while (curr_node && (!count || (count && nr--))) {
1152 		ret = __btrfs_commit_inode_delayed_items(trans, path,
1153 							 curr_node);
1154 		if (ret) {
1155 			btrfs_release_delayed_node(curr_node);
1156 			curr_node = NULL;
1157 			btrfs_abort_transaction(trans, ret);
1158 			break;
1159 		}
1160 
1161 		prev_node = curr_node;
1162 		curr_node = btrfs_next_delayed_node(curr_node);
1163 		btrfs_release_delayed_node(prev_node);
1164 	}
1165 
1166 	if (curr_node)
1167 		btrfs_release_delayed_node(curr_node);
1168 	btrfs_free_path(path);
1169 	trans->block_rsv = block_rsv;
1170 
1171 	return ret;
1172 }
1173 
1174 int btrfs_run_delayed_items(struct btrfs_trans_handle *trans)
1175 {
1176 	return __btrfs_run_delayed_items(trans, -1);
1177 }
1178 
1179 int btrfs_run_delayed_items_nr(struct btrfs_trans_handle *trans, int nr)
1180 {
1181 	return __btrfs_run_delayed_items(trans, nr);
1182 }
1183 
1184 int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1185 				     struct btrfs_inode *inode)
1186 {
1187 	struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1188 	struct btrfs_path *path;
1189 	struct btrfs_block_rsv *block_rsv;
1190 	int ret;
1191 
1192 	if (!delayed_node)
1193 		return 0;
1194 
1195 	mutex_lock(&delayed_node->mutex);
1196 	if (!delayed_node->count) {
1197 		mutex_unlock(&delayed_node->mutex);
1198 		btrfs_release_delayed_node(delayed_node);
1199 		return 0;
1200 	}
1201 	mutex_unlock(&delayed_node->mutex);
1202 
1203 	path = btrfs_alloc_path();
1204 	if (!path) {
1205 		btrfs_release_delayed_node(delayed_node);
1206 		return -ENOMEM;
1207 	}
1208 	path->leave_spinning = 1;
1209 
1210 	block_rsv = trans->block_rsv;
1211 	trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv;
1212 
1213 	ret = __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1214 
1215 	btrfs_release_delayed_node(delayed_node);
1216 	btrfs_free_path(path);
1217 	trans->block_rsv = block_rsv;
1218 
1219 	return ret;
1220 }
1221 
1222 int btrfs_commit_inode_delayed_inode(struct btrfs_inode *inode)
1223 {
1224 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
1225 	struct btrfs_trans_handle *trans;
1226 	struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1227 	struct btrfs_path *path;
1228 	struct btrfs_block_rsv *block_rsv;
1229 	int ret;
1230 
1231 	if (!delayed_node)
1232 		return 0;
1233 
1234 	mutex_lock(&delayed_node->mutex);
1235 	if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1236 		mutex_unlock(&delayed_node->mutex);
1237 		btrfs_release_delayed_node(delayed_node);
1238 		return 0;
1239 	}
1240 	mutex_unlock(&delayed_node->mutex);
1241 
1242 	trans = btrfs_join_transaction(delayed_node->root);
1243 	if (IS_ERR(trans)) {
1244 		ret = PTR_ERR(trans);
1245 		goto out;
1246 	}
1247 
1248 	path = btrfs_alloc_path();
1249 	if (!path) {
1250 		ret = -ENOMEM;
1251 		goto trans_out;
1252 	}
1253 	path->leave_spinning = 1;
1254 
1255 	block_rsv = trans->block_rsv;
1256 	trans->block_rsv = &fs_info->delayed_block_rsv;
1257 
1258 	mutex_lock(&delayed_node->mutex);
1259 	if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags))
1260 		ret = __btrfs_update_delayed_inode(trans, delayed_node->root,
1261 						   path, delayed_node);
1262 	else
1263 		ret = 0;
1264 	mutex_unlock(&delayed_node->mutex);
1265 
1266 	btrfs_free_path(path);
1267 	trans->block_rsv = block_rsv;
1268 trans_out:
1269 	btrfs_end_transaction(trans);
1270 	btrfs_btree_balance_dirty(fs_info);
1271 out:
1272 	btrfs_release_delayed_node(delayed_node);
1273 
1274 	return ret;
1275 }
1276 
1277 void btrfs_remove_delayed_node(struct btrfs_inode *inode)
1278 {
1279 	struct btrfs_delayed_node *delayed_node;
1280 
1281 	delayed_node = READ_ONCE(inode->delayed_node);
1282 	if (!delayed_node)
1283 		return;
1284 
1285 	inode->delayed_node = NULL;
1286 	btrfs_release_delayed_node(delayed_node);
1287 }
1288 
1289 struct btrfs_async_delayed_work {
1290 	struct btrfs_delayed_root *delayed_root;
1291 	int nr;
1292 	struct btrfs_work work;
1293 };
1294 
1295 static void btrfs_async_run_delayed_root(struct btrfs_work *work)
1296 {
1297 	struct btrfs_async_delayed_work *async_work;
1298 	struct btrfs_delayed_root *delayed_root;
1299 	struct btrfs_trans_handle *trans;
1300 	struct btrfs_path *path;
1301 	struct btrfs_delayed_node *delayed_node = NULL;
1302 	struct btrfs_root *root;
1303 	struct btrfs_block_rsv *block_rsv;
1304 	int total_done = 0;
1305 
1306 	async_work = container_of(work, struct btrfs_async_delayed_work, work);
1307 	delayed_root = async_work->delayed_root;
1308 
1309 	path = btrfs_alloc_path();
1310 	if (!path)
1311 		goto out;
1312 
1313 	do {
1314 		if (atomic_read(&delayed_root->items) <
1315 		    BTRFS_DELAYED_BACKGROUND / 2)
1316 			break;
1317 
1318 		delayed_node = btrfs_first_prepared_delayed_node(delayed_root);
1319 		if (!delayed_node)
1320 			break;
1321 
1322 		path->leave_spinning = 1;
1323 		root = delayed_node->root;
1324 
1325 		trans = btrfs_join_transaction(root);
1326 		if (IS_ERR(trans)) {
1327 			btrfs_release_path(path);
1328 			btrfs_release_prepared_delayed_node(delayed_node);
1329 			total_done++;
1330 			continue;
1331 		}
1332 
1333 		block_rsv = trans->block_rsv;
1334 		trans->block_rsv = &root->fs_info->delayed_block_rsv;
1335 
1336 		__btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1337 
1338 		trans->block_rsv = block_rsv;
1339 		btrfs_end_transaction(trans);
1340 		btrfs_btree_balance_dirty_nodelay(root->fs_info);
1341 
1342 		btrfs_release_path(path);
1343 		btrfs_release_prepared_delayed_node(delayed_node);
1344 		total_done++;
1345 
1346 	} while ((async_work->nr == 0 && total_done < BTRFS_DELAYED_WRITEBACK)
1347 		 || total_done < async_work->nr);
1348 
1349 	btrfs_free_path(path);
1350 out:
1351 	wake_up(&delayed_root->wait);
1352 	kfree(async_work);
1353 }
1354 
1355 
1356 static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root,
1357 				     struct btrfs_fs_info *fs_info, int nr)
1358 {
1359 	struct btrfs_async_delayed_work *async_work;
1360 
1361 	async_work = kmalloc(sizeof(*async_work), GFP_NOFS);
1362 	if (!async_work)
1363 		return -ENOMEM;
1364 
1365 	async_work->delayed_root = delayed_root;
1366 	btrfs_init_work(&async_work->work, btrfs_delayed_meta_helper,
1367 			btrfs_async_run_delayed_root, NULL, NULL);
1368 	async_work->nr = nr;
1369 
1370 	btrfs_queue_work(fs_info->delayed_workers, &async_work->work);
1371 	return 0;
1372 }
1373 
1374 void btrfs_assert_delayed_root_empty(struct btrfs_fs_info *fs_info)
1375 {
1376 	WARN_ON(btrfs_first_delayed_node(fs_info->delayed_root));
1377 }
1378 
1379 static int could_end_wait(struct btrfs_delayed_root *delayed_root, int seq)
1380 {
1381 	int val = atomic_read(&delayed_root->items_seq);
1382 
1383 	if (val < seq || val >= seq + BTRFS_DELAYED_BATCH)
1384 		return 1;
1385 
1386 	if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1387 		return 1;
1388 
1389 	return 0;
1390 }
1391 
1392 void btrfs_balance_delayed_items(struct btrfs_fs_info *fs_info)
1393 {
1394 	struct btrfs_delayed_root *delayed_root = fs_info->delayed_root;
1395 
1396 	if ((atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND) ||
1397 		btrfs_workqueue_normal_congested(fs_info->delayed_workers))
1398 		return;
1399 
1400 	if (atomic_read(&delayed_root->items) >= BTRFS_DELAYED_WRITEBACK) {
1401 		int seq;
1402 		int ret;
1403 
1404 		seq = atomic_read(&delayed_root->items_seq);
1405 
1406 		ret = btrfs_wq_run_delayed_node(delayed_root, fs_info, 0);
1407 		if (ret)
1408 			return;
1409 
1410 		wait_event_interruptible(delayed_root->wait,
1411 					 could_end_wait(delayed_root, seq));
1412 		return;
1413 	}
1414 
1415 	btrfs_wq_run_delayed_node(delayed_root, fs_info, BTRFS_DELAYED_BATCH);
1416 }
1417 
1418 /* Will return 0 or -ENOMEM */
1419 int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans,
1420 				   const char *name, int name_len,
1421 				   struct btrfs_inode *dir,
1422 				   struct btrfs_disk_key *disk_key, u8 type,
1423 				   u64 index)
1424 {
1425 	struct btrfs_delayed_node *delayed_node;
1426 	struct btrfs_delayed_item *delayed_item;
1427 	struct btrfs_dir_item *dir_item;
1428 	int ret;
1429 
1430 	delayed_node = btrfs_get_or_create_delayed_node(dir);
1431 	if (IS_ERR(delayed_node))
1432 		return PTR_ERR(delayed_node);
1433 
1434 	delayed_item = btrfs_alloc_delayed_item(sizeof(*dir_item) + name_len);
1435 	if (!delayed_item) {
1436 		ret = -ENOMEM;
1437 		goto release_node;
1438 	}
1439 
1440 	delayed_item->key.objectid = btrfs_ino(dir);
1441 	delayed_item->key.type = BTRFS_DIR_INDEX_KEY;
1442 	delayed_item->key.offset = index;
1443 
1444 	dir_item = (struct btrfs_dir_item *)delayed_item->data;
1445 	dir_item->location = *disk_key;
1446 	btrfs_set_stack_dir_transid(dir_item, trans->transid);
1447 	btrfs_set_stack_dir_data_len(dir_item, 0);
1448 	btrfs_set_stack_dir_name_len(dir_item, name_len);
1449 	btrfs_set_stack_dir_type(dir_item, type);
1450 	memcpy((char *)(dir_item + 1), name, name_len);
1451 
1452 	ret = btrfs_delayed_item_reserve_metadata(trans, dir->root, delayed_item);
1453 	/*
1454 	 * we have reserved enough space when we start a new transaction,
1455 	 * so reserving metadata failure is impossible
1456 	 */
1457 	BUG_ON(ret);
1458 
1459 	mutex_lock(&delayed_node->mutex);
1460 	ret = __btrfs_add_delayed_insertion_item(delayed_node, delayed_item);
1461 	if (unlikely(ret)) {
1462 		btrfs_err(trans->fs_info,
1463 			  "err add delayed dir index item(name: %.*s) into the insertion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
1464 			  name_len, name, delayed_node->root->root_key.objectid,
1465 			  delayed_node->inode_id, ret);
1466 		BUG();
1467 	}
1468 	mutex_unlock(&delayed_node->mutex);
1469 
1470 release_node:
1471 	btrfs_release_delayed_node(delayed_node);
1472 	return ret;
1473 }
1474 
1475 static int btrfs_delete_delayed_insertion_item(struct btrfs_fs_info *fs_info,
1476 					       struct btrfs_delayed_node *node,
1477 					       struct btrfs_key *key)
1478 {
1479 	struct btrfs_delayed_item *item;
1480 
1481 	mutex_lock(&node->mutex);
1482 	item = __btrfs_lookup_delayed_insertion_item(node, key);
1483 	if (!item) {
1484 		mutex_unlock(&node->mutex);
1485 		return 1;
1486 	}
1487 
1488 	btrfs_delayed_item_release_metadata(node->root, item);
1489 	btrfs_release_delayed_item(item);
1490 	mutex_unlock(&node->mutex);
1491 	return 0;
1492 }
1493 
1494 int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans,
1495 				   struct btrfs_inode *dir, u64 index)
1496 {
1497 	struct btrfs_delayed_node *node;
1498 	struct btrfs_delayed_item *item;
1499 	struct btrfs_key item_key;
1500 	int ret;
1501 
1502 	node = btrfs_get_or_create_delayed_node(dir);
1503 	if (IS_ERR(node))
1504 		return PTR_ERR(node);
1505 
1506 	item_key.objectid = btrfs_ino(dir);
1507 	item_key.type = BTRFS_DIR_INDEX_KEY;
1508 	item_key.offset = index;
1509 
1510 	ret = btrfs_delete_delayed_insertion_item(trans->fs_info, node,
1511 						  &item_key);
1512 	if (!ret)
1513 		goto end;
1514 
1515 	item = btrfs_alloc_delayed_item(0);
1516 	if (!item) {
1517 		ret = -ENOMEM;
1518 		goto end;
1519 	}
1520 
1521 	item->key = item_key;
1522 
1523 	ret = btrfs_delayed_item_reserve_metadata(trans, dir->root, item);
1524 	/*
1525 	 * we have reserved enough space when we start a new transaction,
1526 	 * so reserving metadata failure is impossible.
1527 	 */
1528 	BUG_ON(ret);
1529 
1530 	mutex_lock(&node->mutex);
1531 	ret = __btrfs_add_delayed_deletion_item(node, item);
1532 	if (unlikely(ret)) {
1533 		btrfs_err(trans->fs_info,
1534 			  "err add delayed dir index item(index: %llu) into the deletion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
1535 			  index, node->root->root_key.objectid,
1536 			  node->inode_id, ret);
1537 		BUG();
1538 	}
1539 	mutex_unlock(&node->mutex);
1540 end:
1541 	btrfs_release_delayed_node(node);
1542 	return ret;
1543 }
1544 
1545 int btrfs_inode_delayed_dir_index_count(struct btrfs_inode *inode)
1546 {
1547 	struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1548 
1549 	if (!delayed_node)
1550 		return -ENOENT;
1551 
1552 	/*
1553 	 * Since we have held i_mutex of this directory, it is impossible that
1554 	 * a new directory index is added into the delayed node and index_cnt
1555 	 * is updated now. So we needn't lock the delayed node.
1556 	 */
1557 	if (!delayed_node->index_cnt) {
1558 		btrfs_release_delayed_node(delayed_node);
1559 		return -EINVAL;
1560 	}
1561 
1562 	inode->index_cnt = delayed_node->index_cnt;
1563 	btrfs_release_delayed_node(delayed_node);
1564 	return 0;
1565 }
1566 
1567 bool btrfs_readdir_get_delayed_items(struct inode *inode,
1568 				     struct list_head *ins_list,
1569 				     struct list_head *del_list)
1570 {
1571 	struct btrfs_delayed_node *delayed_node;
1572 	struct btrfs_delayed_item *item;
1573 
1574 	delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
1575 	if (!delayed_node)
1576 		return false;
1577 
1578 	/*
1579 	 * We can only do one readdir with delayed items at a time because of
1580 	 * item->readdir_list.
1581 	 */
1582 	inode_unlock_shared(inode);
1583 	inode_lock(inode);
1584 
1585 	mutex_lock(&delayed_node->mutex);
1586 	item = __btrfs_first_delayed_insertion_item(delayed_node);
1587 	while (item) {
1588 		refcount_inc(&item->refs);
1589 		list_add_tail(&item->readdir_list, ins_list);
1590 		item = __btrfs_next_delayed_item(item);
1591 	}
1592 
1593 	item = __btrfs_first_delayed_deletion_item(delayed_node);
1594 	while (item) {
1595 		refcount_inc(&item->refs);
1596 		list_add_tail(&item->readdir_list, del_list);
1597 		item = __btrfs_next_delayed_item(item);
1598 	}
1599 	mutex_unlock(&delayed_node->mutex);
1600 	/*
1601 	 * This delayed node is still cached in the btrfs inode, so refs
1602 	 * must be > 1 now, and we needn't check it is going to be freed
1603 	 * or not.
1604 	 *
1605 	 * Besides that, this function is used to read dir, we do not
1606 	 * insert/delete delayed items in this period. So we also needn't
1607 	 * requeue or dequeue this delayed node.
1608 	 */
1609 	refcount_dec(&delayed_node->refs);
1610 
1611 	return true;
1612 }
1613 
1614 void btrfs_readdir_put_delayed_items(struct inode *inode,
1615 				     struct list_head *ins_list,
1616 				     struct list_head *del_list)
1617 {
1618 	struct btrfs_delayed_item *curr, *next;
1619 
1620 	list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1621 		list_del(&curr->readdir_list);
1622 		if (refcount_dec_and_test(&curr->refs))
1623 			kfree(curr);
1624 	}
1625 
1626 	list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1627 		list_del(&curr->readdir_list);
1628 		if (refcount_dec_and_test(&curr->refs))
1629 			kfree(curr);
1630 	}
1631 
1632 	/*
1633 	 * The VFS is going to do up_read(), so we need to downgrade back to a
1634 	 * read lock.
1635 	 */
1636 	downgrade_write(&inode->i_rwsem);
1637 }
1638 
1639 int btrfs_should_delete_dir_index(struct list_head *del_list,
1640 				  u64 index)
1641 {
1642 	struct btrfs_delayed_item *curr;
1643 	int ret = 0;
1644 
1645 	list_for_each_entry(curr, del_list, readdir_list) {
1646 		if (curr->key.offset > index)
1647 			break;
1648 		if (curr->key.offset == index) {
1649 			ret = 1;
1650 			break;
1651 		}
1652 	}
1653 	return ret;
1654 }
1655 
1656 /*
1657  * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree
1658  *
1659  */
1660 int btrfs_readdir_delayed_dir_index(struct dir_context *ctx,
1661 				    struct list_head *ins_list)
1662 {
1663 	struct btrfs_dir_item *di;
1664 	struct btrfs_delayed_item *curr, *next;
1665 	struct btrfs_key location;
1666 	char *name;
1667 	int name_len;
1668 	int over = 0;
1669 	unsigned char d_type;
1670 
1671 	if (list_empty(ins_list))
1672 		return 0;
1673 
1674 	/*
1675 	 * Changing the data of the delayed item is impossible. So
1676 	 * we needn't lock them. And we have held i_mutex of the
1677 	 * directory, nobody can delete any directory indexes now.
1678 	 */
1679 	list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1680 		list_del(&curr->readdir_list);
1681 
1682 		if (curr->key.offset < ctx->pos) {
1683 			if (refcount_dec_and_test(&curr->refs))
1684 				kfree(curr);
1685 			continue;
1686 		}
1687 
1688 		ctx->pos = curr->key.offset;
1689 
1690 		di = (struct btrfs_dir_item *)curr->data;
1691 		name = (char *)(di + 1);
1692 		name_len = btrfs_stack_dir_name_len(di);
1693 
1694 		d_type = fs_ftype_to_dtype(di->type);
1695 		btrfs_disk_key_to_cpu(&location, &di->location);
1696 
1697 		over = !dir_emit(ctx, name, name_len,
1698 			       location.objectid, d_type);
1699 
1700 		if (refcount_dec_and_test(&curr->refs))
1701 			kfree(curr);
1702 
1703 		if (over)
1704 			return 1;
1705 		ctx->pos++;
1706 	}
1707 	return 0;
1708 }
1709 
1710 static void fill_stack_inode_item(struct btrfs_trans_handle *trans,
1711 				  struct btrfs_inode_item *inode_item,
1712 				  struct inode *inode)
1713 {
1714 	btrfs_set_stack_inode_uid(inode_item, i_uid_read(inode));
1715 	btrfs_set_stack_inode_gid(inode_item, i_gid_read(inode));
1716 	btrfs_set_stack_inode_size(inode_item, BTRFS_I(inode)->disk_i_size);
1717 	btrfs_set_stack_inode_mode(inode_item, inode->i_mode);
1718 	btrfs_set_stack_inode_nlink(inode_item, inode->i_nlink);
1719 	btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(inode));
1720 	btrfs_set_stack_inode_generation(inode_item,
1721 					 BTRFS_I(inode)->generation);
1722 	btrfs_set_stack_inode_sequence(inode_item,
1723 				       inode_peek_iversion(inode));
1724 	btrfs_set_stack_inode_transid(inode_item, trans->transid);
1725 	btrfs_set_stack_inode_rdev(inode_item, inode->i_rdev);
1726 	btrfs_set_stack_inode_flags(inode_item, BTRFS_I(inode)->flags);
1727 	btrfs_set_stack_inode_block_group(inode_item, 0);
1728 
1729 	btrfs_set_stack_timespec_sec(&inode_item->atime,
1730 				     inode->i_atime.tv_sec);
1731 	btrfs_set_stack_timespec_nsec(&inode_item->atime,
1732 				      inode->i_atime.tv_nsec);
1733 
1734 	btrfs_set_stack_timespec_sec(&inode_item->mtime,
1735 				     inode->i_mtime.tv_sec);
1736 	btrfs_set_stack_timespec_nsec(&inode_item->mtime,
1737 				      inode->i_mtime.tv_nsec);
1738 
1739 	btrfs_set_stack_timespec_sec(&inode_item->ctime,
1740 				     inode->i_ctime.tv_sec);
1741 	btrfs_set_stack_timespec_nsec(&inode_item->ctime,
1742 				      inode->i_ctime.tv_nsec);
1743 
1744 	btrfs_set_stack_timespec_sec(&inode_item->otime,
1745 				     BTRFS_I(inode)->i_otime.tv_sec);
1746 	btrfs_set_stack_timespec_nsec(&inode_item->otime,
1747 				     BTRFS_I(inode)->i_otime.tv_nsec);
1748 }
1749 
1750 int btrfs_fill_inode(struct inode *inode, u32 *rdev)
1751 {
1752 	struct btrfs_delayed_node *delayed_node;
1753 	struct btrfs_inode_item *inode_item;
1754 
1755 	delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
1756 	if (!delayed_node)
1757 		return -ENOENT;
1758 
1759 	mutex_lock(&delayed_node->mutex);
1760 	if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1761 		mutex_unlock(&delayed_node->mutex);
1762 		btrfs_release_delayed_node(delayed_node);
1763 		return -ENOENT;
1764 	}
1765 
1766 	inode_item = &delayed_node->inode_item;
1767 
1768 	i_uid_write(inode, btrfs_stack_inode_uid(inode_item));
1769 	i_gid_write(inode, btrfs_stack_inode_gid(inode_item));
1770 	btrfs_i_size_write(BTRFS_I(inode), btrfs_stack_inode_size(inode_item));
1771 	inode->i_mode = btrfs_stack_inode_mode(inode_item);
1772 	set_nlink(inode, btrfs_stack_inode_nlink(inode_item));
1773 	inode_set_bytes(inode, btrfs_stack_inode_nbytes(inode_item));
1774 	BTRFS_I(inode)->generation = btrfs_stack_inode_generation(inode_item);
1775         BTRFS_I(inode)->last_trans = btrfs_stack_inode_transid(inode_item);
1776 
1777 	inode_set_iversion_queried(inode,
1778 				   btrfs_stack_inode_sequence(inode_item));
1779 	inode->i_rdev = 0;
1780 	*rdev = btrfs_stack_inode_rdev(inode_item);
1781 	BTRFS_I(inode)->flags = btrfs_stack_inode_flags(inode_item);
1782 
1783 	inode->i_atime.tv_sec = btrfs_stack_timespec_sec(&inode_item->atime);
1784 	inode->i_atime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->atime);
1785 
1786 	inode->i_mtime.tv_sec = btrfs_stack_timespec_sec(&inode_item->mtime);
1787 	inode->i_mtime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->mtime);
1788 
1789 	inode->i_ctime.tv_sec = btrfs_stack_timespec_sec(&inode_item->ctime);
1790 	inode->i_ctime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->ctime);
1791 
1792 	BTRFS_I(inode)->i_otime.tv_sec =
1793 		btrfs_stack_timespec_sec(&inode_item->otime);
1794 	BTRFS_I(inode)->i_otime.tv_nsec =
1795 		btrfs_stack_timespec_nsec(&inode_item->otime);
1796 
1797 	inode->i_generation = BTRFS_I(inode)->generation;
1798 	BTRFS_I(inode)->index_cnt = (u64)-1;
1799 
1800 	mutex_unlock(&delayed_node->mutex);
1801 	btrfs_release_delayed_node(delayed_node);
1802 	return 0;
1803 }
1804 
1805 int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans,
1806 			       struct btrfs_root *root, struct inode *inode)
1807 {
1808 	struct btrfs_delayed_node *delayed_node;
1809 	int ret = 0;
1810 
1811 	delayed_node = btrfs_get_or_create_delayed_node(BTRFS_I(inode));
1812 	if (IS_ERR(delayed_node))
1813 		return PTR_ERR(delayed_node);
1814 
1815 	mutex_lock(&delayed_node->mutex);
1816 	if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1817 		fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1818 		goto release_node;
1819 	}
1820 
1821 	ret = btrfs_delayed_inode_reserve_metadata(trans, root, BTRFS_I(inode),
1822 						   delayed_node);
1823 	if (ret)
1824 		goto release_node;
1825 
1826 	fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1827 	set_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
1828 	delayed_node->count++;
1829 	atomic_inc(&root->fs_info->delayed_root->items);
1830 release_node:
1831 	mutex_unlock(&delayed_node->mutex);
1832 	btrfs_release_delayed_node(delayed_node);
1833 	return ret;
1834 }
1835 
1836 int btrfs_delayed_delete_inode_ref(struct btrfs_inode *inode)
1837 {
1838 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
1839 	struct btrfs_delayed_node *delayed_node;
1840 
1841 	/*
1842 	 * we don't do delayed inode updates during log recovery because it
1843 	 * leads to enospc problems.  This means we also can't do
1844 	 * delayed inode refs
1845 	 */
1846 	if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
1847 		return -EAGAIN;
1848 
1849 	delayed_node = btrfs_get_or_create_delayed_node(inode);
1850 	if (IS_ERR(delayed_node))
1851 		return PTR_ERR(delayed_node);
1852 
1853 	/*
1854 	 * We don't reserve space for inode ref deletion is because:
1855 	 * - We ONLY do async inode ref deletion for the inode who has only
1856 	 *   one link(i_nlink == 1), it means there is only one inode ref.
1857 	 *   And in most case, the inode ref and the inode item are in the
1858 	 *   same leaf, and we will deal with them at the same time.
1859 	 *   Since we are sure we will reserve the space for the inode item,
1860 	 *   it is unnecessary to reserve space for inode ref deletion.
1861 	 * - If the inode ref and the inode item are not in the same leaf,
1862 	 *   We also needn't worry about enospc problem, because we reserve
1863 	 *   much more space for the inode update than it needs.
1864 	 * - At the worst, we can steal some space from the global reservation.
1865 	 *   It is very rare.
1866 	 */
1867 	mutex_lock(&delayed_node->mutex);
1868 	if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1869 		goto release_node;
1870 
1871 	set_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
1872 	delayed_node->count++;
1873 	atomic_inc(&fs_info->delayed_root->items);
1874 release_node:
1875 	mutex_unlock(&delayed_node->mutex);
1876 	btrfs_release_delayed_node(delayed_node);
1877 	return 0;
1878 }
1879 
1880 static void __btrfs_kill_delayed_node(struct btrfs_delayed_node *delayed_node)
1881 {
1882 	struct btrfs_root *root = delayed_node->root;
1883 	struct btrfs_fs_info *fs_info = root->fs_info;
1884 	struct btrfs_delayed_item *curr_item, *prev_item;
1885 
1886 	mutex_lock(&delayed_node->mutex);
1887 	curr_item = __btrfs_first_delayed_insertion_item(delayed_node);
1888 	while (curr_item) {
1889 		btrfs_delayed_item_release_metadata(root, curr_item);
1890 		prev_item = curr_item;
1891 		curr_item = __btrfs_next_delayed_item(prev_item);
1892 		btrfs_release_delayed_item(prev_item);
1893 	}
1894 
1895 	curr_item = __btrfs_first_delayed_deletion_item(delayed_node);
1896 	while (curr_item) {
1897 		btrfs_delayed_item_release_metadata(root, curr_item);
1898 		prev_item = curr_item;
1899 		curr_item = __btrfs_next_delayed_item(prev_item);
1900 		btrfs_release_delayed_item(prev_item);
1901 	}
1902 
1903 	if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1904 		btrfs_release_delayed_iref(delayed_node);
1905 
1906 	if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1907 		btrfs_delayed_inode_release_metadata(fs_info, delayed_node, false);
1908 		btrfs_release_delayed_inode(delayed_node);
1909 	}
1910 	mutex_unlock(&delayed_node->mutex);
1911 }
1912 
1913 void btrfs_kill_delayed_inode_items(struct btrfs_inode *inode)
1914 {
1915 	struct btrfs_delayed_node *delayed_node;
1916 
1917 	delayed_node = btrfs_get_delayed_node(inode);
1918 	if (!delayed_node)
1919 		return;
1920 
1921 	__btrfs_kill_delayed_node(delayed_node);
1922 	btrfs_release_delayed_node(delayed_node);
1923 }
1924 
1925 void btrfs_kill_all_delayed_nodes(struct btrfs_root *root)
1926 {
1927 	u64 inode_id = 0;
1928 	struct btrfs_delayed_node *delayed_nodes[8];
1929 	int i, n;
1930 
1931 	while (1) {
1932 		spin_lock(&root->inode_lock);
1933 		n = radix_tree_gang_lookup(&root->delayed_nodes_tree,
1934 					   (void **)delayed_nodes, inode_id,
1935 					   ARRAY_SIZE(delayed_nodes));
1936 		if (!n) {
1937 			spin_unlock(&root->inode_lock);
1938 			break;
1939 		}
1940 
1941 		inode_id = delayed_nodes[n - 1]->inode_id + 1;
1942 
1943 		for (i = 0; i < n; i++)
1944 			refcount_inc(&delayed_nodes[i]->refs);
1945 		spin_unlock(&root->inode_lock);
1946 
1947 		for (i = 0; i < n; i++) {
1948 			__btrfs_kill_delayed_node(delayed_nodes[i]);
1949 			btrfs_release_delayed_node(delayed_nodes[i]);
1950 		}
1951 	}
1952 }
1953 
1954 void btrfs_destroy_delayed_inodes(struct btrfs_fs_info *fs_info)
1955 {
1956 	struct btrfs_delayed_node *curr_node, *prev_node;
1957 
1958 	curr_node = btrfs_first_delayed_node(fs_info->delayed_root);
1959 	while (curr_node) {
1960 		__btrfs_kill_delayed_node(curr_node);
1961 
1962 		prev_node = curr_node;
1963 		curr_node = btrfs_next_delayed_node(curr_node);
1964 		btrfs_release_delayed_node(prev_node);
1965 	}
1966 }
1967 
1968