xref: /openbmc/linux/fs/btrfs/delayed-inode.c (revision aac5987a)
1 /*
2  * Copyright (C) 2011 Fujitsu.  All rights reserved.
3  * Written by Miao Xie <miaox@cn.fujitsu.com>
4  *
5  * This program is free software; you can redistribute it and/or
6  * modify it under the terms of the GNU General Public
7  * License v2 as published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope that it will be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
12  * General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public
15  * License along with this program; if not, write to the
16  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
17  * Boston, MA 021110-1307, USA.
18  */
19 
20 #include <linux/slab.h>
21 #include "delayed-inode.h"
22 #include "disk-io.h"
23 #include "transaction.h"
24 #include "ctree.h"
25 
26 #define BTRFS_DELAYED_WRITEBACK		512
27 #define BTRFS_DELAYED_BACKGROUND	128
28 #define BTRFS_DELAYED_BATCH		16
29 
30 static struct kmem_cache *delayed_node_cache;
31 
32 int __init btrfs_delayed_inode_init(void)
33 {
34 	delayed_node_cache = kmem_cache_create("btrfs_delayed_node",
35 					sizeof(struct btrfs_delayed_node),
36 					0,
37 					SLAB_MEM_SPREAD,
38 					NULL);
39 	if (!delayed_node_cache)
40 		return -ENOMEM;
41 	return 0;
42 }
43 
44 void btrfs_delayed_inode_exit(void)
45 {
46 	kmem_cache_destroy(delayed_node_cache);
47 }
48 
49 static inline void btrfs_init_delayed_node(
50 				struct btrfs_delayed_node *delayed_node,
51 				struct btrfs_root *root, u64 inode_id)
52 {
53 	delayed_node->root = root;
54 	delayed_node->inode_id = inode_id;
55 	atomic_set(&delayed_node->refs, 0);
56 	delayed_node->ins_root = RB_ROOT;
57 	delayed_node->del_root = RB_ROOT;
58 	mutex_init(&delayed_node->mutex);
59 	INIT_LIST_HEAD(&delayed_node->n_list);
60 	INIT_LIST_HEAD(&delayed_node->p_list);
61 }
62 
63 static inline int btrfs_is_continuous_delayed_item(
64 					struct btrfs_delayed_item *item1,
65 					struct btrfs_delayed_item *item2)
66 {
67 	if (item1->key.type == BTRFS_DIR_INDEX_KEY &&
68 	    item1->key.objectid == item2->key.objectid &&
69 	    item1->key.type == item2->key.type &&
70 	    item1->key.offset + 1 == item2->key.offset)
71 		return 1;
72 	return 0;
73 }
74 
75 static struct btrfs_delayed_node *btrfs_get_delayed_node(
76 		struct btrfs_inode *btrfs_inode)
77 {
78 	struct btrfs_root *root = btrfs_inode->root;
79 	u64 ino = btrfs_ino(btrfs_inode);
80 	struct btrfs_delayed_node *node;
81 
82 	node = READ_ONCE(btrfs_inode->delayed_node);
83 	if (node) {
84 		atomic_inc(&node->refs);
85 		return node;
86 	}
87 
88 	spin_lock(&root->inode_lock);
89 	node = radix_tree_lookup(&root->delayed_nodes_tree, ino);
90 	if (node) {
91 		if (btrfs_inode->delayed_node) {
92 			atomic_inc(&node->refs);	/* can be accessed */
93 			BUG_ON(btrfs_inode->delayed_node != node);
94 			spin_unlock(&root->inode_lock);
95 			return node;
96 		}
97 		btrfs_inode->delayed_node = node;
98 		/* can be accessed and cached in the inode */
99 		atomic_add(2, &node->refs);
100 		spin_unlock(&root->inode_lock);
101 		return node;
102 	}
103 	spin_unlock(&root->inode_lock);
104 
105 	return NULL;
106 }
107 
108 /* Will return either the node or PTR_ERR(-ENOMEM) */
109 static struct btrfs_delayed_node *btrfs_get_or_create_delayed_node(
110 		struct btrfs_inode *btrfs_inode)
111 {
112 	struct btrfs_delayed_node *node;
113 	struct btrfs_root *root = btrfs_inode->root;
114 	u64 ino = btrfs_ino(btrfs_inode);
115 	int ret;
116 
117 again:
118 	node = btrfs_get_delayed_node(btrfs_inode);
119 	if (node)
120 		return node;
121 
122 	node = kmem_cache_zalloc(delayed_node_cache, GFP_NOFS);
123 	if (!node)
124 		return ERR_PTR(-ENOMEM);
125 	btrfs_init_delayed_node(node, root, ino);
126 
127 	/* cached in the btrfs inode and can be accessed */
128 	atomic_add(2, &node->refs);
129 
130 	ret = radix_tree_preload(GFP_NOFS);
131 	if (ret) {
132 		kmem_cache_free(delayed_node_cache, node);
133 		return ERR_PTR(ret);
134 	}
135 
136 	spin_lock(&root->inode_lock);
137 	ret = radix_tree_insert(&root->delayed_nodes_tree, ino, node);
138 	if (ret == -EEXIST) {
139 		spin_unlock(&root->inode_lock);
140 		kmem_cache_free(delayed_node_cache, node);
141 		radix_tree_preload_end();
142 		goto again;
143 	}
144 	btrfs_inode->delayed_node = node;
145 	spin_unlock(&root->inode_lock);
146 	radix_tree_preload_end();
147 
148 	return node;
149 }
150 
151 /*
152  * Call it when holding delayed_node->mutex
153  *
154  * If mod = 1, add this node into the prepared list.
155  */
156 static void btrfs_queue_delayed_node(struct btrfs_delayed_root *root,
157 				     struct btrfs_delayed_node *node,
158 				     int mod)
159 {
160 	spin_lock(&root->lock);
161 	if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
162 		if (!list_empty(&node->p_list))
163 			list_move_tail(&node->p_list, &root->prepare_list);
164 		else if (mod)
165 			list_add_tail(&node->p_list, &root->prepare_list);
166 	} else {
167 		list_add_tail(&node->n_list, &root->node_list);
168 		list_add_tail(&node->p_list, &root->prepare_list);
169 		atomic_inc(&node->refs);	/* inserted into list */
170 		root->nodes++;
171 		set_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
172 	}
173 	spin_unlock(&root->lock);
174 }
175 
176 /* Call it when holding delayed_node->mutex */
177 static void btrfs_dequeue_delayed_node(struct btrfs_delayed_root *root,
178 				       struct btrfs_delayed_node *node)
179 {
180 	spin_lock(&root->lock);
181 	if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
182 		root->nodes--;
183 		atomic_dec(&node->refs);	/* not in the list */
184 		list_del_init(&node->n_list);
185 		if (!list_empty(&node->p_list))
186 			list_del_init(&node->p_list);
187 		clear_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
188 	}
189 	spin_unlock(&root->lock);
190 }
191 
192 static struct btrfs_delayed_node *btrfs_first_delayed_node(
193 			struct btrfs_delayed_root *delayed_root)
194 {
195 	struct list_head *p;
196 	struct btrfs_delayed_node *node = NULL;
197 
198 	spin_lock(&delayed_root->lock);
199 	if (list_empty(&delayed_root->node_list))
200 		goto out;
201 
202 	p = delayed_root->node_list.next;
203 	node = list_entry(p, struct btrfs_delayed_node, n_list);
204 	atomic_inc(&node->refs);
205 out:
206 	spin_unlock(&delayed_root->lock);
207 
208 	return node;
209 }
210 
211 static struct btrfs_delayed_node *btrfs_next_delayed_node(
212 						struct btrfs_delayed_node *node)
213 {
214 	struct btrfs_delayed_root *delayed_root;
215 	struct list_head *p;
216 	struct btrfs_delayed_node *next = NULL;
217 
218 	delayed_root = node->root->fs_info->delayed_root;
219 	spin_lock(&delayed_root->lock);
220 	if (!test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
221 		/* not in the list */
222 		if (list_empty(&delayed_root->node_list))
223 			goto out;
224 		p = delayed_root->node_list.next;
225 	} else if (list_is_last(&node->n_list, &delayed_root->node_list))
226 		goto out;
227 	else
228 		p = node->n_list.next;
229 
230 	next = list_entry(p, struct btrfs_delayed_node, n_list);
231 	atomic_inc(&next->refs);
232 out:
233 	spin_unlock(&delayed_root->lock);
234 
235 	return next;
236 }
237 
238 static void __btrfs_release_delayed_node(
239 				struct btrfs_delayed_node *delayed_node,
240 				int mod)
241 {
242 	struct btrfs_delayed_root *delayed_root;
243 
244 	if (!delayed_node)
245 		return;
246 
247 	delayed_root = delayed_node->root->fs_info->delayed_root;
248 
249 	mutex_lock(&delayed_node->mutex);
250 	if (delayed_node->count)
251 		btrfs_queue_delayed_node(delayed_root, delayed_node, mod);
252 	else
253 		btrfs_dequeue_delayed_node(delayed_root, delayed_node);
254 	mutex_unlock(&delayed_node->mutex);
255 
256 	if (atomic_dec_and_test(&delayed_node->refs)) {
257 		bool free = false;
258 		struct btrfs_root *root = delayed_node->root;
259 		spin_lock(&root->inode_lock);
260 		if (atomic_read(&delayed_node->refs) == 0) {
261 			radix_tree_delete(&root->delayed_nodes_tree,
262 					  delayed_node->inode_id);
263 			free = true;
264 		}
265 		spin_unlock(&root->inode_lock);
266 		if (free)
267 			kmem_cache_free(delayed_node_cache, delayed_node);
268 	}
269 }
270 
271 static inline void btrfs_release_delayed_node(struct btrfs_delayed_node *node)
272 {
273 	__btrfs_release_delayed_node(node, 0);
274 }
275 
276 static struct btrfs_delayed_node *btrfs_first_prepared_delayed_node(
277 					struct btrfs_delayed_root *delayed_root)
278 {
279 	struct list_head *p;
280 	struct btrfs_delayed_node *node = NULL;
281 
282 	spin_lock(&delayed_root->lock);
283 	if (list_empty(&delayed_root->prepare_list))
284 		goto out;
285 
286 	p = delayed_root->prepare_list.next;
287 	list_del_init(p);
288 	node = list_entry(p, struct btrfs_delayed_node, p_list);
289 	atomic_inc(&node->refs);
290 out:
291 	spin_unlock(&delayed_root->lock);
292 
293 	return node;
294 }
295 
296 static inline void btrfs_release_prepared_delayed_node(
297 					struct btrfs_delayed_node *node)
298 {
299 	__btrfs_release_delayed_node(node, 1);
300 }
301 
302 static struct btrfs_delayed_item *btrfs_alloc_delayed_item(u32 data_len)
303 {
304 	struct btrfs_delayed_item *item;
305 	item = kmalloc(sizeof(*item) + data_len, GFP_NOFS);
306 	if (item) {
307 		item->data_len = data_len;
308 		item->ins_or_del = 0;
309 		item->bytes_reserved = 0;
310 		item->delayed_node = NULL;
311 		atomic_set(&item->refs, 1);
312 	}
313 	return item;
314 }
315 
316 /*
317  * __btrfs_lookup_delayed_item - look up the delayed item by key
318  * @delayed_node: pointer to the delayed node
319  * @key:	  the key to look up
320  * @prev:	  used to store the prev item if the right item isn't found
321  * @next:	  used to store the next item if the right item isn't found
322  *
323  * Note: if we don't find the right item, we will return the prev item and
324  * the next item.
325  */
326 static struct btrfs_delayed_item *__btrfs_lookup_delayed_item(
327 				struct rb_root *root,
328 				struct btrfs_key *key,
329 				struct btrfs_delayed_item **prev,
330 				struct btrfs_delayed_item **next)
331 {
332 	struct rb_node *node, *prev_node = NULL;
333 	struct btrfs_delayed_item *delayed_item = NULL;
334 	int ret = 0;
335 
336 	node = root->rb_node;
337 
338 	while (node) {
339 		delayed_item = rb_entry(node, struct btrfs_delayed_item,
340 					rb_node);
341 		prev_node = node;
342 		ret = btrfs_comp_cpu_keys(&delayed_item->key, key);
343 		if (ret < 0)
344 			node = node->rb_right;
345 		else if (ret > 0)
346 			node = node->rb_left;
347 		else
348 			return delayed_item;
349 	}
350 
351 	if (prev) {
352 		if (!prev_node)
353 			*prev = NULL;
354 		else if (ret < 0)
355 			*prev = delayed_item;
356 		else if ((node = rb_prev(prev_node)) != NULL) {
357 			*prev = rb_entry(node, struct btrfs_delayed_item,
358 					 rb_node);
359 		} else
360 			*prev = NULL;
361 	}
362 
363 	if (next) {
364 		if (!prev_node)
365 			*next = NULL;
366 		else if (ret > 0)
367 			*next = delayed_item;
368 		else if ((node = rb_next(prev_node)) != NULL) {
369 			*next = rb_entry(node, struct btrfs_delayed_item,
370 					 rb_node);
371 		} else
372 			*next = NULL;
373 	}
374 	return NULL;
375 }
376 
377 static struct btrfs_delayed_item *__btrfs_lookup_delayed_insertion_item(
378 					struct btrfs_delayed_node *delayed_node,
379 					struct btrfs_key *key)
380 {
381 	return __btrfs_lookup_delayed_item(&delayed_node->ins_root, key,
382 					   NULL, NULL);
383 }
384 
385 static int __btrfs_add_delayed_item(struct btrfs_delayed_node *delayed_node,
386 				    struct btrfs_delayed_item *ins,
387 				    int action)
388 {
389 	struct rb_node **p, *node;
390 	struct rb_node *parent_node = NULL;
391 	struct rb_root *root;
392 	struct btrfs_delayed_item *item;
393 	int cmp;
394 
395 	if (action == BTRFS_DELAYED_INSERTION_ITEM)
396 		root = &delayed_node->ins_root;
397 	else if (action == BTRFS_DELAYED_DELETION_ITEM)
398 		root = &delayed_node->del_root;
399 	else
400 		BUG();
401 	p = &root->rb_node;
402 	node = &ins->rb_node;
403 
404 	while (*p) {
405 		parent_node = *p;
406 		item = rb_entry(parent_node, struct btrfs_delayed_item,
407 				 rb_node);
408 
409 		cmp = btrfs_comp_cpu_keys(&item->key, &ins->key);
410 		if (cmp < 0)
411 			p = &(*p)->rb_right;
412 		else if (cmp > 0)
413 			p = &(*p)->rb_left;
414 		else
415 			return -EEXIST;
416 	}
417 
418 	rb_link_node(node, parent_node, p);
419 	rb_insert_color(node, root);
420 	ins->delayed_node = delayed_node;
421 	ins->ins_or_del = action;
422 
423 	if (ins->key.type == BTRFS_DIR_INDEX_KEY &&
424 	    action == BTRFS_DELAYED_INSERTION_ITEM &&
425 	    ins->key.offset >= delayed_node->index_cnt)
426 			delayed_node->index_cnt = ins->key.offset + 1;
427 
428 	delayed_node->count++;
429 	atomic_inc(&delayed_node->root->fs_info->delayed_root->items);
430 	return 0;
431 }
432 
433 static int __btrfs_add_delayed_insertion_item(struct btrfs_delayed_node *node,
434 					      struct btrfs_delayed_item *item)
435 {
436 	return __btrfs_add_delayed_item(node, item,
437 					BTRFS_DELAYED_INSERTION_ITEM);
438 }
439 
440 static int __btrfs_add_delayed_deletion_item(struct btrfs_delayed_node *node,
441 					     struct btrfs_delayed_item *item)
442 {
443 	return __btrfs_add_delayed_item(node, item,
444 					BTRFS_DELAYED_DELETION_ITEM);
445 }
446 
447 static void finish_one_item(struct btrfs_delayed_root *delayed_root)
448 {
449 	int seq = atomic_inc_return(&delayed_root->items_seq);
450 
451 	/*
452 	 * atomic_dec_return implies a barrier for waitqueue_active
453 	 */
454 	if ((atomic_dec_return(&delayed_root->items) <
455 	    BTRFS_DELAYED_BACKGROUND || seq % BTRFS_DELAYED_BATCH == 0) &&
456 	    waitqueue_active(&delayed_root->wait))
457 		wake_up(&delayed_root->wait);
458 }
459 
460 static void __btrfs_remove_delayed_item(struct btrfs_delayed_item *delayed_item)
461 {
462 	struct rb_root *root;
463 	struct btrfs_delayed_root *delayed_root;
464 
465 	delayed_root = delayed_item->delayed_node->root->fs_info->delayed_root;
466 
467 	BUG_ON(!delayed_root);
468 	BUG_ON(delayed_item->ins_or_del != BTRFS_DELAYED_DELETION_ITEM &&
469 	       delayed_item->ins_or_del != BTRFS_DELAYED_INSERTION_ITEM);
470 
471 	if (delayed_item->ins_or_del == BTRFS_DELAYED_INSERTION_ITEM)
472 		root = &delayed_item->delayed_node->ins_root;
473 	else
474 		root = &delayed_item->delayed_node->del_root;
475 
476 	rb_erase(&delayed_item->rb_node, root);
477 	delayed_item->delayed_node->count--;
478 
479 	finish_one_item(delayed_root);
480 }
481 
482 static void btrfs_release_delayed_item(struct btrfs_delayed_item *item)
483 {
484 	if (item) {
485 		__btrfs_remove_delayed_item(item);
486 		if (atomic_dec_and_test(&item->refs))
487 			kfree(item);
488 	}
489 }
490 
491 static struct btrfs_delayed_item *__btrfs_first_delayed_insertion_item(
492 					struct btrfs_delayed_node *delayed_node)
493 {
494 	struct rb_node *p;
495 	struct btrfs_delayed_item *item = NULL;
496 
497 	p = rb_first(&delayed_node->ins_root);
498 	if (p)
499 		item = rb_entry(p, struct btrfs_delayed_item, rb_node);
500 
501 	return item;
502 }
503 
504 static struct btrfs_delayed_item *__btrfs_first_delayed_deletion_item(
505 					struct btrfs_delayed_node *delayed_node)
506 {
507 	struct rb_node *p;
508 	struct btrfs_delayed_item *item = NULL;
509 
510 	p = rb_first(&delayed_node->del_root);
511 	if (p)
512 		item = rb_entry(p, struct btrfs_delayed_item, rb_node);
513 
514 	return item;
515 }
516 
517 static struct btrfs_delayed_item *__btrfs_next_delayed_item(
518 						struct btrfs_delayed_item *item)
519 {
520 	struct rb_node *p;
521 	struct btrfs_delayed_item *next = NULL;
522 
523 	p = rb_next(&item->rb_node);
524 	if (p)
525 		next = rb_entry(p, struct btrfs_delayed_item, rb_node);
526 
527 	return next;
528 }
529 
530 static int btrfs_delayed_item_reserve_metadata(struct btrfs_trans_handle *trans,
531 					       struct btrfs_fs_info *fs_info,
532 					       struct btrfs_delayed_item *item)
533 {
534 	struct btrfs_block_rsv *src_rsv;
535 	struct btrfs_block_rsv *dst_rsv;
536 	u64 num_bytes;
537 	int ret;
538 
539 	if (!trans->bytes_reserved)
540 		return 0;
541 
542 	src_rsv = trans->block_rsv;
543 	dst_rsv = &fs_info->delayed_block_rsv;
544 
545 	num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
546 	ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1);
547 	if (!ret) {
548 		trace_btrfs_space_reservation(fs_info, "delayed_item",
549 					      item->key.objectid,
550 					      num_bytes, 1);
551 		item->bytes_reserved = num_bytes;
552 	}
553 
554 	return ret;
555 }
556 
557 static void btrfs_delayed_item_release_metadata(struct btrfs_fs_info *fs_info,
558 						struct btrfs_delayed_item *item)
559 {
560 	struct btrfs_block_rsv *rsv;
561 
562 	if (!item->bytes_reserved)
563 		return;
564 
565 	rsv = &fs_info->delayed_block_rsv;
566 	trace_btrfs_space_reservation(fs_info, "delayed_item",
567 				      item->key.objectid, item->bytes_reserved,
568 				      0);
569 	btrfs_block_rsv_release(fs_info, rsv,
570 				item->bytes_reserved);
571 }
572 
573 static int btrfs_delayed_inode_reserve_metadata(
574 					struct btrfs_trans_handle *trans,
575 					struct btrfs_root *root,
576 					struct btrfs_inode *inode,
577 					struct btrfs_delayed_node *node)
578 {
579 	struct btrfs_fs_info *fs_info = root->fs_info;
580 	struct btrfs_block_rsv *src_rsv;
581 	struct btrfs_block_rsv *dst_rsv;
582 	u64 num_bytes;
583 	int ret;
584 	bool release = false;
585 
586 	src_rsv = trans->block_rsv;
587 	dst_rsv = &fs_info->delayed_block_rsv;
588 
589 	num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
590 
591 	/*
592 	 * If our block_rsv is the delalloc block reserve then check and see if
593 	 * we have our extra reservation for updating the inode.  If not fall
594 	 * through and try to reserve space quickly.
595 	 *
596 	 * We used to try and steal from the delalloc block rsv or the global
597 	 * reserve, but we'd steal a full reservation, which isn't kind.  We are
598 	 * here through delalloc which means we've likely just cowed down close
599 	 * to the leaf that contains the inode, so we would steal less just
600 	 * doing the fallback inode update, so if we do end up having to steal
601 	 * from the global block rsv we hopefully only steal one or two blocks
602 	 * worth which is less likely to hurt us.
603 	 */
604 	if (src_rsv && src_rsv->type == BTRFS_BLOCK_RSV_DELALLOC) {
605 		spin_lock(&inode->lock);
606 		if (test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
607 				       &inode->runtime_flags))
608 			release = true;
609 		else
610 			src_rsv = NULL;
611 		spin_unlock(&inode->lock);
612 	}
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're accounted for.
622 	 */
623 	if (!src_rsv || (!trans->bytes_reserved &&
624 			 src_rsv->type != BTRFS_BLOCK_RSV_DELALLOC)) {
625 		ret = btrfs_block_rsv_add(root, dst_rsv, num_bytes,
626 					  BTRFS_RESERVE_NO_FLUSH);
627 		/*
628 		 * Since we're under a transaction reserve_metadata_bytes could
629 		 * try to commit the transaction which will make it return
630 		 * EAGAIN to make us stop the transaction we have, so return
631 		 * ENOSPC instead so that btrfs_dirty_inode knows what to do.
632 		 */
633 		if (ret == -EAGAIN)
634 			ret = -ENOSPC;
635 		if (!ret) {
636 			node->bytes_reserved = num_bytes;
637 			trace_btrfs_space_reservation(fs_info,
638 						      "delayed_inode",
639 						      btrfs_ino(inode),
640 						      num_bytes, 1);
641 		}
642 		return ret;
643 	}
644 
645 	ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1);
646 
647 	/*
648 	 * Migrate only takes a reservation, it doesn't touch the size of the
649 	 * block_rsv.  This is to simplify people who don't normally have things
650 	 * migrated from their block rsv.  If they go to release their
651 	 * reservation, that will decrease the size as well, so if migrate
652 	 * reduced size we'd end up with a negative size.  But for the
653 	 * delalloc_meta_reserved stuff we will only know to drop 1 reservation,
654 	 * but we could in fact do this reserve/migrate dance several times
655 	 * between the time we did the original reservation and we'd clean it
656 	 * up.  So to take care of this, release the space for the meta
657 	 * reservation here.  I think it may be time for a documentation page on
658 	 * how block rsvs. work.
659 	 */
660 	if (!ret) {
661 		trace_btrfs_space_reservation(fs_info, "delayed_inode",
662 					      btrfs_ino(inode), num_bytes, 1);
663 		node->bytes_reserved = num_bytes;
664 	}
665 
666 	if (release) {
667 		trace_btrfs_space_reservation(fs_info, "delalloc",
668 					      btrfs_ino(inode), num_bytes, 0);
669 		btrfs_block_rsv_release(fs_info, src_rsv, num_bytes);
670 	}
671 
672 	return ret;
673 }
674 
675 static void btrfs_delayed_inode_release_metadata(struct btrfs_fs_info *fs_info,
676 						struct btrfs_delayed_node *node)
677 {
678 	struct btrfs_block_rsv *rsv;
679 
680 	if (!node->bytes_reserved)
681 		return;
682 
683 	rsv = &fs_info->delayed_block_rsv;
684 	trace_btrfs_space_reservation(fs_info, "delayed_inode",
685 				      node->inode_id, node->bytes_reserved, 0);
686 	btrfs_block_rsv_release(fs_info, rsv,
687 				node->bytes_reserved);
688 	node->bytes_reserved = 0;
689 }
690 
691 /*
692  * This helper will insert some continuous items into the same leaf according
693  * to the free space of the leaf.
694  */
695 static int btrfs_batch_insert_items(struct btrfs_root *root,
696 				    struct btrfs_path *path,
697 				    struct btrfs_delayed_item *item)
698 {
699 	struct btrfs_fs_info *fs_info = root->fs_info;
700 	struct btrfs_delayed_item *curr, *next;
701 	int free_space;
702 	int total_data_size = 0, total_size = 0;
703 	struct extent_buffer *leaf;
704 	char *data_ptr;
705 	struct btrfs_key *keys;
706 	u32 *data_size;
707 	struct list_head head;
708 	int slot;
709 	int nitems;
710 	int i;
711 	int ret = 0;
712 
713 	BUG_ON(!path->nodes[0]);
714 
715 	leaf = path->nodes[0];
716 	free_space = btrfs_leaf_free_space(fs_info, leaf);
717 	INIT_LIST_HEAD(&head);
718 
719 	next = item;
720 	nitems = 0;
721 
722 	/*
723 	 * count the number of the continuous items that we can insert in batch
724 	 */
725 	while (total_size + next->data_len + sizeof(struct btrfs_item) <=
726 	       free_space) {
727 		total_data_size += next->data_len;
728 		total_size += next->data_len + sizeof(struct btrfs_item);
729 		list_add_tail(&next->tree_list, &head);
730 		nitems++;
731 
732 		curr = next;
733 		next = __btrfs_next_delayed_item(curr);
734 		if (!next)
735 			break;
736 
737 		if (!btrfs_is_continuous_delayed_item(curr, next))
738 			break;
739 	}
740 
741 	if (!nitems) {
742 		ret = 0;
743 		goto out;
744 	}
745 
746 	/*
747 	 * we need allocate some memory space, but it might cause the task
748 	 * to sleep, so we set all locked nodes in the path to blocking locks
749 	 * first.
750 	 */
751 	btrfs_set_path_blocking(path);
752 
753 	keys = kmalloc_array(nitems, sizeof(struct btrfs_key), GFP_NOFS);
754 	if (!keys) {
755 		ret = -ENOMEM;
756 		goto out;
757 	}
758 
759 	data_size = kmalloc_array(nitems, sizeof(u32), GFP_NOFS);
760 	if (!data_size) {
761 		ret = -ENOMEM;
762 		goto error;
763 	}
764 
765 	/* get keys of all the delayed items */
766 	i = 0;
767 	list_for_each_entry(next, &head, tree_list) {
768 		keys[i] = next->key;
769 		data_size[i] = next->data_len;
770 		i++;
771 	}
772 
773 	/* reset all the locked nodes in the patch to spinning locks. */
774 	btrfs_clear_path_blocking(path, NULL, 0);
775 
776 	/* insert the keys of the items */
777 	setup_items_for_insert(root, path, keys, data_size,
778 			       total_data_size, total_size, nitems);
779 
780 	/* insert the dir index items */
781 	slot = path->slots[0];
782 	list_for_each_entry_safe(curr, next, &head, tree_list) {
783 		data_ptr = btrfs_item_ptr(leaf, slot, char);
784 		write_extent_buffer(leaf, &curr->data,
785 				    (unsigned long)data_ptr,
786 				    curr->data_len);
787 		slot++;
788 
789 		btrfs_delayed_item_release_metadata(fs_info, curr);
790 
791 		list_del(&curr->tree_list);
792 		btrfs_release_delayed_item(curr);
793 	}
794 
795 error:
796 	kfree(data_size);
797 	kfree(keys);
798 out:
799 	return ret;
800 }
801 
802 /*
803  * This helper can just do simple insertion that needn't extend item for new
804  * data, such as directory name index insertion, inode insertion.
805  */
806 static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans,
807 				     struct btrfs_root *root,
808 				     struct btrfs_path *path,
809 				     struct btrfs_delayed_item *delayed_item)
810 {
811 	struct btrfs_fs_info *fs_info = root->fs_info;
812 	struct extent_buffer *leaf;
813 	char *ptr;
814 	int ret;
815 
816 	ret = btrfs_insert_empty_item(trans, root, path, &delayed_item->key,
817 				      delayed_item->data_len);
818 	if (ret < 0 && ret != -EEXIST)
819 		return ret;
820 
821 	leaf = path->nodes[0];
822 
823 	ptr = btrfs_item_ptr(leaf, path->slots[0], char);
824 
825 	write_extent_buffer(leaf, delayed_item->data, (unsigned long)ptr,
826 			    delayed_item->data_len);
827 	btrfs_mark_buffer_dirty(leaf);
828 
829 	btrfs_delayed_item_release_metadata(fs_info, delayed_item);
830 	return 0;
831 }
832 
833 /*
834  * we insert an item first, then if there are some continuous items, we try
835  * to insert those items into the same leaf.
836  */
837 static int btrfs_insert_delayed_items(struct btrfs_trans_handle *trans,
838 				      struct btrfs_path *path,
839 				      struct btrfs_root *root,
840 				      struct btrfs_delayed_node *node)
841 {
842 	struct btrfs_delayed_item *curr, *prev;
843 	int ret = 0;
844 
845 do_again:
846 	mutex_lock(&node->mutex);
847 	curr = __btrfs_first_delayed_insertion_item(node);
848 	if (!curr)
849 		goto insert_end;
850 
851 	ret = btrfs_insert_delayed_item(trans, root, path, curr);
852 	if (ret < 0) {
853 		btrfs_release_path(path);
854 		goto insert_end;
855 	}
856 
857 	prev = curr;
858 	curr = __btrfs_next_delayed_item(prev);
859 	if (curr && btrfs_is_continuous_delayed_item(prev, curr)) {
860 		/* insert the continuous items into the same leaf */
861 		path->slots[0]++;
862 		btrfs_batch_insert_items(root, path, curr);
863 	}
864 	btrfs_release_delayed_item(prev);
865 	btrfs_mark_buffer_dirty(path->nodes[0]);
866 
867 	btrfs_release_path(path);
868 	mutex_unlock(&node->mutex);
869 	goto do_again;
870 
871 insert_end:
872 	mutex_unlock(&node->mutex);
873 	return ret;
874 }
875 
876 static int btrfs_batch_delete_items(struct btrfs_trans_handle *trans,
877 				    struct btrfs_root *root,
878 				    struct btrfs_path *path,
879 				    struct btrfs_delayed_item *item)
880 {
881 	struct btrfs_fs_info *fs_info = root->fs_info;
882 	struct btrfs_delayed_item *curr, *next;
883 	struct extent_buffer *leaf;
884 	struct btrfs_key key;
885 	struct list_head head;
886 	int nitems, i, last_item;
887 	int ret = 0;
888 
889 	BUG_ON(!path->nodes[0]);
890 
891 	leaf = path->nodes[0];
892 
893 	i = path->slots[0];
894 	last_item = btrfs_header_nritems(leaf) - 1;
895 	if (i > last_item)
896 		return -ENOENT;	/* FIXME: Is errno suitable? */
897 
898 	next = item;
899 	INIT_LIST_HEAD(&head);
900 	btrfs_item_key_to_cpu(leaf, &key, i);
901 	nitems = 0;
902 	/*
903 	 * count the number of the dir index items that we can delete in batch
904 	 */
905 	while (btrfs_comp_cpu_keys(&next->key, &key) == 0) {
906 		list_add_tail(&next->tree_list, &head);
907 		nitems++;
908 
909 		curr = next;
910 		next = __btrfs_next_delayed_item(curr);
911 		if (!next)
912 			break;
913 
914 		if (!btrfs_is_continuous_delayed_item(curr, next))
915 			break;
916 
917 		i++;
918 		if (i > last_item)
919 			break;
920 		btrfs_item_key_to_cpu(leaf, &key, i);
921 	}
922 
923 	if (!nitems)
924 		return 0;
925 
926 	ret = btrfs_del_items(trans, root, path, path->slots[0], nitems);
927 	if (ret)
928 		goto out;
929 
930 	list_for_each_entry_safe(curr, next, &head, tree_list) {
931 		btrfs_delayed_item_release_metadata(fs_info, curr);
932 		list_del(&curr->tree_list);
933 		btrfs_release_delayed_item(curr);
934 	}
935 
936 out:
937 	return ret;
938 }
939 
940 static int btrfs_delete_delayed_items(struct btrfs_trans_handle *trans,
941 				      struct btrfs_path *path,
942 				      struct btrfs_root *root,
943 				      struct btrfs_delayed_node *node)
944 {
945 	struct btrfs_delayed_item *curr, *prev;
946 	int ret = 0;
947 
948 do_again:
949 	mutex_lock(&node->mutex);
950 	curr = __btrfs_first_delayed_deletion_item(node);
951 	if (!curr)
952 		goto delete_fail;
953 
954 	ret = btrfs_search_slot(trans, root, &curr->key, path, -1, 1);
955 	if (ret < 0)
956 		goto delete_fail;
957 	else if (ret > 0) {
958 		/*
959 		 * can't find the item which the node points to, so this node
960 		 * is invalid, just drop it.
961 		 */
962 		prev = curr;
963 		curr = __btrfs_next_delayed_item(prev);
964 		btrfs_release_delayed_item(prev);
965 		ret = 0;
966 		btrfs_release_path(path);
967 		if (curr) {
968 			mutex_unlock(&node->mutex);
969 			goto do_again;
970 		} else
971 			goto delete_fail;
972 	}
973 
974 	btrfs_batch_delete_items(trans, root, path, curr);
975 	btrfs_release_path(path);
976 	mutex_unlock(&node->mutex);
977 	goto do_again;
978 
979 delete_fail:
980 	btrfs_release_path(path);
981 	mutex_unlock(&node->mutex);
982 	return ret;
983 }
984 
985 static void btrfs_release_delayed_inode(struct btrfs_delayed_node *delayed_node)
986 {
987 	struct btrfs_delayed_root *delayed_root;
988 
989 	if (delayed_node &&
990 	    test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
991 		BUG_ON(!delayed_node->root);
992 		clear_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
993 		delayed_node->count--;
994 
995 		delayed_root = delayed_node->root->fs_info->delayed_root;
996 		finish_one_item(delayed_root);
997 	}
998 }
999 
1000 static void btrfs_release_delayed_iref(struct btrfs_delayed_node *delayed_node)
1001 {
1002 	struct btrfs_delayed_root *delayed_root;
1003 
1004 	ASSERT(delayed_node->root);
1005 	clear_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
1006 	delayed_node->count--;
1007 
1008 	delayed_root = delayed_node->root->fs_info->delayed_root;
1009 	finish_one_item(delayed_root);
1010 }
1011 
1012 static int __btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1013 					struct btrfs_root *root,
1014 					struct btrfs_path *path,
1015 					struct btrfs_delayed_node *node)
1016 {
1017 	struct btrfs_fs_info *fs_info = root->fs_info;
1018 	struct btrfs_key key;
1019 	struct btrfs_inode_item *inode_item;
1020 	struct extent_buffer *leaf;
1021 	int mod;
1022 	int ret;
1023 
1024 	key.objectid = node->inode_id;
1025 	key.type = BTRFS_INODE_ITEM_KEY;
1026 	key.offset = 0;
1027 
1028 	if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1029 		mod = -1;
1030 	else
1031 		mod = 1;
1032 
1033 	ret = btrfs_lookup_inode(trans, root, path, &key, mod);
1034 	if (ret > 0) {
1035 		btrfs_release_path(path);
1036 		return -ENOENT;
1037 	} else if (ret < 0) {
1038 		return ret;
1039 	}
1040 
1041 	leaf = path->nodes[0];
1042 	inode_item = btrfs_item_ptr(leaf, path->slots[0],
1043 				    struct btrfs_inode_item);
1044 	write_extent_buffer(leaf, &node->inode_item, (unsigned long)inode_item,
1045 			    sizeof(struct btrfs_inode_item));
1046 	btrfs_mark_buffer_dirty(leaf);
1047 
1048 	if (!test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1049 		goto no_iref;
1050 
1051 	path->slots[0]++;
1052 	if (path->slots[0] >= btrfs_header_nritems(leaf))
1053 		goto search;
1054 again:
1055 	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1056 	if (key.objectid != node->inode_id)
1057 		goto out;
1058 
1059 	if (key.type != BTRFS_INODE_REF_KEY &&
1060 	    key.type != BTRFS_INODE_EXTREF_KEY)
1061 		goto out;
1062 
1063 	/*
1064 	 * Delayed iref deletion is for the inode who has only one link,
1065 	 * so there is only one iref. The case that several irefs are
1066 	 * in the same item doesn't exist.
1067 	 */
1068 	btrfs_del_item(trans, root, path);
1069 out:
1070 	btrfs_release_delayed_iref(node);
1071 no_iref:
1072 	btrfs_release_path(path);
1073 err_out:
1074 	btrfs_delayed_inode_release_metadata(fs_info, node);
1075 	btrfs_release_delayed_inode(node);
1076 
1077 	return ret;
1078 
1079 search:
1080 	btrfs_release_path(path);
1081 
1082 	key.type = BTRFS_INODE_EXTREF_KEY;
1083 	key.offset = -1;
1084 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1085 	if (ret < 0)
1086 		goto err_out;
1087 	ASSERT(ret);
1088 
1089 	ret = 0;
1090 	leaf = path->nodes[0];
1091 	path->slots[0]--;
1092 	goto again;
1093 }
1094 
1095 static inline int btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1096 					     struct btrfs_root *root,
1097 					     struct btrfs_path *path,
1098 					     struct btrfs_delayed_node *node)
1099 {
1100 	int ret;
1101 
1102 	mutex_lock(&node->mutex);
1103 	if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &node->flags)) {
1104 		mutex_unlock(&node->mutex);
1105 		return 0;
1106 	}
1107 
1108 	ret = __btrfs_update_delayed_inode(trans, root, path, node);
1109 	mutex_unlock(&node->mutex);
1110 	return ret;
1111 }
1112 
1113 static inline int
1114 __btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1115 				   struct btrfs_path *path,
1116 				   struct btrfs_delayed_node *node)
1117 {
1118 	int ret;
1119 
1120 	ret = btrfs_insert_delayed_items(trans, path, node->root, node);
1121 	if (ret)
1122 		return ret;
1123 
1124 	ret = btrfs_delete_delayed_items(trans, path, node->root, node);
1125 	if (ret)
1126 		return ret;
1127 
1128 	ret = btrfs_update_delayed_inode(trans, node->root, path, node);
1129 	return ret;
1130 }
1131 
1132 /*
1133  * Called when committing the transaction.
1134  * Returns 0 on success.
1135  * Returns < 0 on error and returns with an aborted transaction with any
1136  * outstanding delayed items cleaned up.
1137  */
1138 static int __btrfs_run_delayed_items(struct btrfs_trans_handle *trans,
1139 				     struct btrfs_fs_info *fs_info, int nr)
1140 {
1141 	struct btrfs_delayed_root *delayed_root;
1142 	struct btrfs_delayed_node *curr_node, *prev_node;
1143 	struct btrfs_path *path;
1144 	struct btrfs_block_rsv *block_rsv;
1145 	int ret = 0;
1146 	bool count = (nr > 0);
1147 
1148 	if (trans->aborted)
1149 		return -EIO;
1150 
1151 	path = btrfs_alloc_path();
1152 	if (!path)
1153 		return -ENOMEM;
1154 	path->leave_spinning = 1;
1155 
1156 	block_rsv = trans->block_rsv;
1157 	trans->block_rsv = &fs_info->delayed_block_rsv;
1158 
1159 	delayed_root = fs_info->delayed_root;
1160 
1161 	curr_node = btrfs_first_delayed_node(delayed_root);
1162 	while (curr_node && (!count || (count && nr--))) {
1163 		ret = __btrfs_commit_inode_delayed_items(trans, path,
1164 							 curr_node);
1165 		if (ret) {
1166 			btrfs_release_delayed_node(curr_node);
1167 			curr_node = NULL;
1168 			btrfs_abort_transaction(trans, ret);
1169 			break;
1170 		}
1171 
1172 		prev_node = curr_node;
1173 		curr_node = btrfs_next_delayed_node(curr_node);
1174 		btrfs_release_delayed_node(prev_node);
1175 	}
1176 
1177 	if (curr_node)
1178 		btrfs_release_delayed_node(curr_node);
1179 	btrfs_free_path(path);
1180 	trans->block_rsv = block_rsv;
1181 
1182 	return ret;
1183 }
1184 
1185 int btrfs_run_delayed_items(struct btrfs_trans_handle *trans,
1186 			    struct btrfs_fs_info *fs_info)
1187 {
1188 	return __btrfs_run_delayed_items(trans, fs_info, -1);
1189 }
1190 
1191 int btrfs_run_delayed_items_nr(struct btrfs_trans_handle *trans,
1192 			       struct btrfs_fs_info *fs_info, int nr)
1193 {
1194 	return __btrfs_run_delayed_items(trans, fs_info, nr);
1195 }
1196 
1197 int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1198 				     struct btrfs_inode *inode)
1199 {
1200 	struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1201 	struct btrfs_path *path;
1202 	struct btrfs_block_rsv *block_rsv;
1203 	int ret;
1204 
1205 	if (!delayed_node)
1206 		return 0;
1207 
1208 	mutex_lock(&delayed_node->mutex);
1209 	if (!delayed_node->count) {
1210 		mutex_unlock(&delayed_node->mutex);
1211 		btrfs_release_delayed_node(delayed_node);
1212 		return 0;
1213 	}
1214 	mutex_unlock(&delayed_node->mutex);
1215 
1216 	path = btrfs_alloc_path();
1217 	if (!path) {
1218 		btrfs_release_delayed_node(delayed_node);
1219 		return -ENOMEM;
1220 	}
1221 	path->leave_spinning = 1;
1222 
1223 	block_rsv = trans->block_rsv;
1224 	trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv;
1225 
1226 	ret = __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1227 
1228 	btrfs_release_delayed_node(delayed_node);
1229 	btrfs_free_path(path);
1230 	trans->block_rsv = block_rsv;
1231 
1232 	return ret;
1233 }
1234 
1235 int btrfs_commit_inode_delayed_inode(struct btrfs_inode *inode)
1236 {
1237 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1238 	struct btrfs_trans_handle *trans;
1239 	struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1240 	struct btrfs_path *path;
1241 	struct btrfs_block_rsv *block_rsv;
1242 	int ret;
1243 
1244 	if (!delayed_node)
1245 		return 0;
1246 
1247 	mutex_lock(&delayed_node->mutex);
1248 	if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1249 		mutex_unlock(&delayed_node->mutex);
1250 		btrfs_release_delayed_node(delayed_node);
1251 		return 0;
1252 	}
1253 	mutex_unlock(&delayed_node->mutex);
1254 
1255 	trans = btrfs_join_transaction(delayed_node->root);
1256 	if (IS_ERR(trans)) {
1257 		ret = PTR_ERR(trans);
1258 		goto out;
1259 	}
1260 
1261 	path = btrfs_alloc_path();
1262 	if (!path) {
1263 		ret = -ENOMEM;
1264 		goto trans_out;
1265 	}
1266 	path->leave_spinning = 1;
1267 
1268 	block_rsv = trans->block_rsv;
1269 	trans->block_rsv = &fs_info->delayed_block_rsv;
1270 
1271 	mutex_lock(&delayed_node->mutex);
1272 	if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags))
1273 		ret = __btrfs_update_delayed_inode(trans, delayed_node->root,
1274 						   path, delayed_node);
1275 	else
1276 		ret = 0;
1277 	mutex_unlock(&delayed_node->mutex);
1278 
1279 	btrfs_free_path(path);
1280 	trans->block_rsv = block_rsv;
1281 trans_out:
1282 	btrfs_end_transaction(trans);
1283 	btrfs_btree_balance_dirty(fs_info);
1284 out:
1285 	btrfs_release_delayed_node(delayed_node);
1286 
1287 	return ret;
1288 }
1289 
1290 void btrfs_remove_delayed_node(struct btrfs_inode *inode)
1291 {
1292 	struct btrfs_delayed_node *delayed_node;
1293 
1294 	delayed_node = READ_ONCE(inode->delayed_node);
1295 	if (!delayed_node)
1296 		return;
1297 
1298 	inode->delayed_node = NULL;
1299 	btrfs_release_delayed_node(delayed_node);
1300 }
1301 
1302 struct btrfs_async_delayed_work {
1303 	struct btrfs_delayed_root *delayed_root;
1304 	int nr;
1305 	struct btrfs_work work;
1306 };
1307 
1308 static void btrfs_async_run_delayed_root(struct btrfs_work *work)
1309 {
1310 	struct btrfs_async_delayed_work *async_work;
1311 	struct btrfs_delayed_root *delayed_root;
1312 	struct btrfs_trans_handle *trans;
1313 	struct btrfs_path *path;
1314 	struct btrfs_delayed_node *delayed_node = NULL;
1315 	struct btrfs_root *root;
1316 	struct btrfs_block_rsv *block_rsv;
1317 	int total_done = 0;
1318 
1319 	async_work = container_of(work, struct btrfs_async_delayed_work, work);
1320 	delayed_root = async_work->delayed_root;
1321 
1322 	path = btrfs_alloc_path();
1323 	if (!path)
1324 		goto out;
1325 
1326 again:
1327 	if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND / 2)
1328 		goto free_path;
1329 
1330 	delayed_node = btrfs_first_prepared_delayed_node(delayed_root);
1331 	if (!delayed_node)
1332 		goto free_path;
1333 
1334 	path->leave_spinning = 1;
1335 	root = delayed_node->root;
1336 
1337 	trans = btrfs_join_transaction(root);
1338 	if (IS_ERR(trans))
1339 		goto release_path;
1340 
1341 	block_rsv = trans->block_rsv;
1342 	trans->block_rsv = &root->fs_info->delayed_block_rsv;
1343 
1344 	__btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1345 
1346 	trans->block_rsv = block_rsv;
1347 	btrfs_end_transaction(trans);
1348 	btrfs_btree_balance_dirty_nodelay(root->fs_info);
1349 
1350 release_path:
1351 	btrfs_release_path(path);
1352 	total_done++;
1353 
1354 	btrfs_release_prepared_delayed_node(delayed_node);
1355 	if ((async_work->nr == 0 && total_done < BTRFS_DELAYED_WRITEBACK) ||
1356 	    total_done < async_work->nr)
1357 		goto again;
1358 
1359 free_path:
1360 	btrfs_free_path(path);
1361 out:
1362 	wake_up(&delayed_root->wait);
1363 	kfree(async_work);
1364 }
1365 
1366 
1367 static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root,
1368 				     struct btrfs_fs_info *fs_info, int nr)
1369 {
1370 	struct btrfs_async_delayed_work *async_work;
1371 
1372 	if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND ||
1373 	    btrfs_workqueue_normal_congested(fs_info->delayed_workers))
1374 		return 0;
1375 
1376 	async_work = kmalloc(sizeof(*async_work), GFP_NOFS);
1377 	if (!async_work)
1378 		return -ENOMEM;
1379 
1380 	async_work->delayed_root = delayed_root;
1381 	btrfs_init_work(&async_work->work, btrfs_delayed_meta_helper,
1382 			btrfs_async_run_delayed_root, NULL, NULL);
1383 	async_work->nr = nr;
1384 
1385 	btrfs_queue_work(fs_info->delayed_workers, &async_work->work);
1386 	return 0;
1387 }
1388 
1389 void btrfs_assert_delayed_root_empty(struct btrfs_fs_info *fs_info)
1390 {
1391 	WARN_ON(btrfs_first_delayed_node(fs_info->delayed_root));
1392 }
1393 
1394 static int could_end_wait(struct btrfs_delayed_root *delayed_root, int seq)
1395 {
1396 	int val = atomic_read(&delayed_root->items_seq);
1397 
1398 	if (val < seq || val >= seq + BTRFS_DELAYED_BATCH)
1399 		return 1;
1400 
1401 	if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1402 		return 1;
1403 
1404 	return 0;
1405 }
1406 
1407 void btrfs_balance_delayed_items(struct btrfs_fs_info *fs_info)
1408 {
1409 	struct btrfs_delayed_root *delayed_root = fs_info->delayed_root;
1410 
1411 	if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1412 		return;
1413 
1414 	if (atomic_read(&delayed_root->items) >= BTRFS_DELAYED_WRITEBACK) {
1415 		int seq;
1416 		int ret;
1417 
1418 		seq = atomic_read(&delayed_root->items_seq);
1419 
1420 		ret = btrfs_wq_run_delayed_node(delayed_root, fs_info, 0);
1421 		if (ret)
1422 			return;
1423 
1424 		wait_event_interruptible(delayed_root->wait,
1425 					 could_end_wait(delayed_root, seq));
1426 		return;
1427 	}
1428 
1429 	btrfs_wq_run_delayed_node(delayed_root, fs_info, BTRFS_DELAYED_BATCH);
1430 }
1431 
1432 /* Will return 0 or -ENOMEM */
1433 int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans,
1434 				   struct btrfs_fs_info *fs_info,
1435 				   const char *name, int name_len,
1436 				   struct btrfs_inode *dir,
1437 				   struct btrfs_disk_key *disk_key, u8 type,
1438 				   u64 index)
1439 {
1440 	struct btrfs_delayed_node *delayed_node;
1441 	struct btrfs_delayed_item *delayed_item;
1442 	struct btrfs_dir_item *dir_item;
1443 	int ret;
1444 
1445 	delayed_node = btrfs_get_or_create_delayed_node(dir);
1446 	if (IS_ERR(delayed_node))
1447 		return PTR_ERR(delayed_node);
1448 
1449 	delayed_item = btrfs_alloc_delayed_item(sizeof(*dir_item) + name_len);
1450 	if (!delayed_item) {
1451 		ret = -ENOMEM;
1452 		goto release_node;
1453 	}
1454 
1455 	delayed_item->key.objectid = btrfs_ino(dir);
1456 	delayed_item->key.type = BTRFS_DIR_INDEX_KEY;
1457 	delayed_item->key.offset = index;
1458 
1459 	dir_item = (struct btrfs_dir_item *)delayed_item->data;
1460 	dir_item->location = *disk_key;
1461 	btrfs_set_stack_dir_transid(dir_item, trans->transid);
1462 	btrfs_set_stack_dir_data_len(dir_item, 0);
1463 	btrfs_set_stack_dir_name_len(dir_item, name_len);
1464 	btrfs_set_stack_dir_type(dir_item, type);
1465 	memcpy((char *)(dir_item + 1), name, name_len);
1466 
1467 	ret = btrfs_delayed_item_reserve_metadata(trans, fs_info, delayed_item);
1468 	/*
1469 	 * we have reserved enough space when we start a new transaction,
1470 	 * so reserving metadata failure is impossible
1471 	 */
1472 	BUG_ON(ret);
1473 
1474 
1475 	mutex_lock(&delayed_node->mutex);
1476 	ret = __btrfs_add_delayed_insertion_item(delayed_node, delayed_item);
1477 	if (unlikely(ret)) {
1478 		btrfs_err(fs_info,
1479 			  "err add delayed dir index item(name: %.*s) into the insertion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
1480 			  name_len, name, delayed_node->root->objectid,
1481 			  delayed_node->inode_id, ret);
1482 		BUG();
1483 	}
1484 	mutex_unlock(&delayed_node->mutex);
1485 
1486 release_node:
1487 	btrfs_release_delayed_node(delayed_node);
1488 	return ret;
1489 }
1490 
1491 static int btrfs_delete_delayed_insertion_item(struct btrfs_fs_info *fs_info,
1492 					       struct btrfs_delayed_node *node,
1493 					       struct btrfs_key *key)
1494 {
1495 	struct btrfs_delayed_item *item;
1496 
1497 	mutex_lock(&node->mutex);
1498 	item = __btrfs_lookup_delayed_insertion_item(node, key);
1499 	if (!item) {
1500 		mutex_unlock(&node->mutex);
1501 		return 1;
1502 	}
1503 
1504 	btrfs_delayed_item_release_metadata(fs_info, item);
1505 	btrfs_release_delayed_item(item);
1506 	mutex_unlock(&node->mutex);
1507 	return 0;
1508 }
1509 
1510 int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans,
1511 				   struct btrfs_fs_info *fs_info,
1512 				   struct btrfs_inode *dir, u64 index)
1513 {
1514 	struct btrfs_delayed_node *node;
1515 	struct btrfs_delayed_item *item;
1516 	struct btrfs_key item_key;
1517 	int ret;
1518 
1519 	node = btrfs_get_or_create_delayed_node(dir);
1520 	if (IS_ERR(node))
1521 		return PTR_ERR(node);
1522 
1523 	item_key.objectid = btrfs_ino(dir);
1524 	item_key.type = BTRFS_DIR_INDEX_KEY;
1525 	item_key.offset = index;
1526 
1527 	ret = btrfs_delete_delayed_insertion_item(fs_info, node, &item_key);
1528 	if (!ret)
1529 		goto end;
1530 
1531 	item = btrfs_alloc_delayed_item(0);
1532 	if (!item) {
1533 		ret = -ENOMEM;
1534 		goto end;
1535 	}
1536 
1537 	item->key = item_key;
1538 
1539 	ret = btrfs_delayed_item_reserve_metadata(trans, fs_info, item);
1540 	/*
1541 	 * we have reserved enough space when we start a new transaction,
1542 	 * so reserving metadata failure is impossible.
1543 	 */
1544 	BUG_ON(ret);
1545 
1546 	mutex_lock(&node->mutex);
1547 	ret = __btrfs_add_delayed_deletion_item(node, item);
1548 	if (unlikely(ret)) {
1549 		btrfs_err(fs_info,
1550 			  "err add delayed dir index item(index: %llu) into the deletion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
1551 			  index, node->root->objectid, node->inode_id, ret);
1552 		BUG();
1553 	}
1554 	mutex_unlock(&node->mutex);
1555 end:
1556 	btrfs_release_delayed_node(node);
1557 	return ret;
1558 }
1559 
1560 int btrfs_inode_delayed_dir_index_count(struct btrfs_inode *inode)
1561 {
1562 	struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1563 
1564 	if (!delayed_node)
1565 		return -ENOENT;
1566 
1567 	/*
1568 	 * Since we have held i_mutex of this directory, it is impossible that
1569 	 * a new directory index is added into the delayed node and index_cnt
1570 	 * is updated now. So we needn't lock the delayed node.
1571 	 */
1572 	if (!delayed_node->index_cnt) {
1573 		btrfs_release_delayed_node(delayed_node);
1574 		return -EINVAL;
1575 	}
1576 
1577 	inode->index_cnt = delayed_node->index_cnt;
1578 	btrfs_release_delayed_node(delayed_node);
1579 	return 0;
1580 }
1581 
1582 bool btrfs_readdir_get_delayed_items(struct inode *inode,
1583 				     struct list_head *ins_list,
1584 				     struct list_head *del_list)
1585 {
1586 	struct btrfs_delayed_node *delayed_node;
1587 	struct btrfs_delayed_item *item;
1588 
1589 	delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
1590 	if (!delayed_node)
1591 		return false;
1592 
1593 	/*
1594 	 * We can only do one readdir with delayed items at a time because of
1595 	 * item->readdir_list.
1596 	 */
1597 	inode_unlock_shared(inode);
1598 	inode_lock(inode);
1599 
1600 	mutex_lock(&delayed_node->mutex);
1601 	item = __btrfs_first_delayed_insertion_item(delayed_node);
1602 	while (item) {
1603 		atomic_inc(&item->refs);
1604 		list_add_tail(&item->readdir_list, ins_list);
1605 		item = __btrfs_next_delayed_item(item);
1606 	}
1607 
1608 	item = __btrfs_first_delayed_deletion_item(delayed_node);
1609 	while (item) {
1610 		atomic_inc(&item->refs);
1611 		list_add_tail(&item->readdir_list, del_list);
1612 		item = __btrfs_next_delayed_item(item);
1613 	}
1614 	mutex_unlock(&delayed_node->mutex);
1615 	/*
1616 	 * This delayed node is still cached in the btrfs inode, so refs
1617 	 * must be > 1 now, and we needn't check it is going to be freed
1618 	 * or not.
1619 	 *
1620 	 * Besides that, this function is used to read dir, we do not
1621 	 * insert/delete delayed items in this period. So we also needn't
1622 	 * requeue or dequeue this delayed node.
1623 	 */
1624 	atomic_dec(&delayed_node->refs);
1625 
1626 	return true;
1627 }
1628 
1629 void btrfs_readdir_put_delayed_items(struct inode *inode,
1630 				     struct list_head *ins_list,
1631 				     struct list_head *del_list)
1632 {
1633 	struct btrfs_delayed_item *curr, *next;
1634 
1635 	list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1636 		list_del(&curr->readdir_list);
1637 		if (atomic_dec_and_test(&curr->refs))
1638 			kfree(curr);
1639 	}
1640 
1641 	list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1642 		list_del(&curr->readdir_list);
1643 		if (atomic_dec_and_test(&curr->refs))
1644 			kfree(curr);
1645 	}
1646 
1647 	/*
1648 	 * The VFS is going to do up_read(), so we need to downgrade back to a
1649 	 * read lock.
1650 	 */
1651 	downgrade_write(&inode->i_rwsem);
1652 }
1653 
1654 int btrfs_should_delete_dir_index(struct list_head *del_list,
1655 				  u64 index)
1656 {
1657 	struct btrfs_delayed_item *curr, *next;
1658 	int ret;
1659 
1660 	if (list_empty(del_list))
1661 		return 0;
1662 
1663 	list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1664 		if (curr->key.offset > index)
1665 			break;
1666 
1667 		list_del(&curr->readdir_list);
1668 		ret = (curr->key.offset == index);
1669 
1670 		if (atomic_dec_and_test(&curr->refs))
1671 			kfree(curr);
1672 
1673 		if (ret)
1674 			return 1;
1675 		else
1676 			continue;
1677 	}
1678 	return 0;
1679 }
1680 
1681 /*
1682  * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree
1683  *
1684  */
1685 int btrfs_readdir_delayed_dir_index(struct dir_context *ctx,
1686 				    struct list_head *ins_list)
1687 {
1688 	struct btrfs_dir_item *di;
1689 	struct btrfs_delayed_item *curr, *next;
1690 	struct btrfs_key location;
1691 	char *name;
1692 	int name_len;
1693 	int over = 0;
1694 	unsigned char d_type;
1695 
1696 	if (list_empty(ins_list))
1697 		return 0;
1698 
1699 	/*
1700 	 * Changing the data of the delayed item is impossible. So
1701 	 * we needn't lock them. And we have held i_mutex of the
1702 	 * directory, nobody can delete any directory indexes now.
1703 	 */
1704 	list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1705 		list_del(&curr->readdir_list);
1706 
1707 		if (curr->key.offset < ctx->pos) {
1708 			if (atomic_dec_and_test(&curr->refs))
1709 				kfree(curr);
1710 			continue;
1711 		}
1712 
1713 		ctx->pos = curr->key.offset;
1714 
1715 		di = (struct btrfs_dir_item *)curr->data;
1716 		name = (char *)(di + 1);
1717 		name_len = btrfs_stack_dir_name_len(di);
1718 
1719 		d_type = btrfs_filetype_table[di->type];
1720 		btrfs_disk_key_to_cpu(&location, &di->location);
1721 
1722 		over = !dir_emit(ctx, name, name_len,
1723 			       location.objectid, d_type);
1724 
1725 		if (atomic_dec_and_test(&curr->refs))
1726 			kfree(curr);
1727 
1728 		if (over)
1729 			return 1;
1730 	}
1731 	return 0;
1732 }
1733 
1734 static void fill_stack_inode_item(struct btrfs_trans_handle *trans,
1735 				  struct btrfs_inode_item *inode_item,
1736 				  struct inode *inode)
1737 {
1738 	btrfs_set_stack_inode_uid(inode_item, i_uid_read(inode));
1739 	btrfs_set_stack_inode_gid(inode_item, i_gid_read(inode));
1740 	btrfs_set_stack_inode_size(inode_item, BTRFS_I(inode)->disk_i_size);
1741 	btrfs_set_stack_inode_mode(inode_item, inode->i_mode);
1742 	btrfs_set_stack_inode_nlink(inode_item, inode->i_nlink);
1743 	btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(inode));
1744 	btrfs_set_stack_inode_generation(inode_item,
1745 					 BTRFS_I(inode)->generation);
1746 	btrfs_set_stack_inode_sequence(inode_item, inode->i_version);
1747 	btrfs_set_stack_inode_transid(inode_item, trans->transid);
1748 	btrfs_set_stack_inode_rdev(inode_item, inode->i_rdev);
1749 	btrfs_set_stack_inode_flags(inode_item, BTRFS_I(inode)->flags);
1750 	btrfs_set_stack_inode_block_group(inode_item, 0);
1751 
1752 	btrfs_set_stack_timespec_sec(&inode_item->atime,
1753 				     inode->i_atime.tv_sec);
1754 	btrfs_set_stack_timespec_nsec(&inode_item->atime,
1755 				      inode->i_atime.tv_nsec);
1756 
1757 	btrfs_set_stack_timespec_sec(&inode_item->mtime,
1758 				     inode->i_mtime.tv_sec);
1759 	btrfs_set_stack_timespec_nsec(&inode_item->mtime,
1760 				      inode->i_mtime.tv_nsec);
1761 
1762 	btrfs_set_stack_timespec_sec(&inode_item->ctime,
1763 				     inode->i_ctime.tv_sec);
1764 	btrfs_set_stack_timespec_nsec(&inode_item->ctime,
1765 				      inode->i_ctime.tv_nsec);
1766 
1767 	btrfs_set_stack_timespec_sec(&inode_item->otime,
1768 				     BTRFS_I(inode)->i_otime.tv_sec);
1769 	btrfs_set_stack_timespec_nsec(&inode_item->otime,
1770 				     BTRFS_I(inode)->i_otime.tv_nsec);
1771 }
1772 
1773 int btrfs_fill_inode(struct inode *inode, u32 *rdev)
1774 {
1775 	struct btrfs_delayed_node *delayed_node;
1776 	struct btrfs_inode_item *inode_item;
1777 
1778 	delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
1779 	if (!delayed_node)
1780 		return -ENOENT;
1781 
1782 	mutex_lock(&delayed_node->mutex);
1783 	if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1784 		mutex_unlock(&delayed_node->mutex);
1785 		btrfs_release_delayed_node(delayed_node);
1786 		return -ENOENT;
1787 	}
1788 
1789 	inode_item = &delayed_node->inode_item;
1790 
1791 	i_uid_write(inode, btrfs_stack_inode_uid(inode_item));
1792 	i_gid_write(inode, btrfs_stack_inode_gid(inode_item));
1793 	btrfs_i_size_write(BTRFS_I(inode), btrfs_stack_inode_size(inode_item));
1794 	inode->i_mode = btrfs_stack_inode_mode(inode_item);
1795 	set_nlink(inode, btrfs_stack_inode_nlink(inode_item));
1796 	inode_set_bytes(inode, btrfs_stack_inode_nbytes(inode_item));
1797 	BTRFS_I(inode)->generation = btrfs_stack_inode_generation(inode_item);
1798         BTRFS_I(inode)->last_trans = btrfs_stack_inode_transid(inode_item);
1799 
1800 	inode->i_version = btrfs_stack_inode_sequence(inode_item);
1801 	inode->i_rdev = 0;
1802 	*rdev = btrfs_stack_inode_rdev(inode_item);
1803 	BTRFS_I(inode)->flags = btrfs_stack_inode_flags(inode_item);
1804 
1805 	inode->i_atime.tv_sec = btrfs_stack_timespec_sec(&inode_item->atime);
1806 	inode->i_atime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->atime);
1807 
1808 	inode->i_mtime.tv_sec = btrfs_stack_timespec_sec(&inode_item->mtime);
1809 	inode->i_mtime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->mtime);
1810 
1811 	inode->i_ctime.tv_sec = btrfs_stack_timespec_sec(&inode_item->ctime);
1812 	inode->i_ctime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->ctime);
1813 
1814 	BTRFS_I(inode)->i_otime.tv_sec =
1815 		btrfs_stack_timespec_sec(&inode_item->otime);
1816 	BTRFS_I(inode)->i_otime.tv_nsec =
1817 		btrfs_stack_timespec_nsec(&inode_item->otime);
1818 
1819 	inode->i_generation = BTRFS_I(inode)->generation;
1820 	BTRFS_I(inode)->index_cnt = (u64)-1;
1821 
1822 	mutex_unlock(&delayed_node->mutex);
1823 	btrfs_release_delayed_node(delayed_node);
1824 	return 0;
1825 }
1826 
1827 int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans,
1828 			       struct btrfs_root *root, struct inode *inode)
1829 {
1830 	struct btrfs_delayed_node *delayed_node;
1831 	int ret = 0;
1832 
1833 	delayed_node = btrfs_get_or_create_delayed_node(BTRFS_I(inode));
1834 	if (IS_ERR(delayed_node))
1835 		return PTR_ERR(delayed_node);
1836 
1837 	mutex_lock(&delayed_node->mutex);
1838 	if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1839 		fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1840 		goto release_node;
1841 	}
1842 
1843 	ret = btrfs_delayed_inode_reserve_metadata(trans, root, BTRFS_I(inode),
1844 						   delayed_node);
1845 	if (ret)
1846 		goto release_node;
1847 
1848 	fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1849 	set_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
1850 	delayed_node->count++;
1851 	atomic_inc(&root->fs_info->delayed_root->items);
1852 release_node:
1853 	mutex_unlock(&delayed_node->mutex);
1854 	btrfs_release_delayed_node(delayed_node);
1855 	return ret;
1856 }
1857 
1858 int btrfs_delayed_delete_inode_ref(struct btrfs_inode *inode)
1859 {
1860 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1861 	struct btrfs_delayed_node *delayed_node;
1862 
1863 	/*
1864 	 * we don't do delayed inode updates during log recovery because it
1865 	 * leads to enospc problems.  This means we also can't do
1866 	 * delayed inode refs
1867 	 */
1868 	if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
1869 		return -EAGAIN;
1870 
1871 	delayed_node = btrfs_get_or_create_delayed_node(inode);
1872 	if (IS_ERR(delayed_node))
1873 		return PTR_ERR(delayed_node);
1874 
1875 	/*
1876 	 * We don't reserve space for inode ref deletion is because:
1877 	 * - We ONLY do async inode ref deletion for the inode who has only
1878 	 *   one link(i_nlink == 1), it means there is only one inode ref.
1879 	 *   And in most case, the inode ref and the inode item are in the
1880 	 *   same leaf, and we will deal with them at the same time.
1881 	 *   Since we are sure we will reserve the space for the inode item,
1882 	 *   it is unnecessary to reserve space for inode ref deletion.
1883 	 * - If the inode ref and the inode item are not in the same leaf,
1884 	 *   We also needn't worry about enospc problem, because we reserve
1885 	 *   much more space for the inode update than it needs.
1886 	 * - At the worst, we can steal some space from the global reservation.
1887 	 *   It is very rare.
1888 	 */
1889 	mutex_lock(&delayed_node->mutex);
1890 	if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1891 		goto release_node;
1892 
1893 	set_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
1894 	delayed_node->count++;
1895 	atomic_inc(&fs_info->delayed_root->items);
1896 release_node:
1897 	mutex_unlock(&delayed_node->mutex);
1898 	btrfs_release_delayed_node(delayed_node);
1899 	return 0;
1900 }
1901 
1902 static void __btrfs_kill_delayed_node(struct btrfs_delayed_node *delayed_node)
1903 {
1904 	struct btrfs_root *root = delayed_node->root;
1905 	struct btrfs_fs_info *fs_info = root->fs_info;
1906 	struct btrfs_delayed_item *curr_item, *prev_item;
1907 
1908 	mutex_lock(&delayed_node->mutex);
1909 	curr_item = __btrfs_first_delayed_insertion_item(delayed_node);
1910 	while (curr_item) {
1911 		btrfs_delayed_item_release_metadata(fs_info, curr_item);
1912 		prev_item = curr_item;
1913 		curr_item = __btrfs_next_delayed_item(prev_item);
1914 		btrfs_release_delayed_item(prev_item);
1915 	}
1916 
1917 	curr_item = __btrfs_first_delayed_deletion_item(delayed_node);
1918 	while (curr_item) {
1919 		btrfs_delayed_item_release_metadata(fs_info, curr_item);
1920 		prev_item = curr_item;
1921 		curr_item = __btrfs_next_delayed_item(prev_item);
1922 		btrfs_release_delayed_item(prev_item);
1923 	}
1924 
1925 	if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1926 		btrfs_release_delayed_iref(delayed_node);
1927 
1928 	if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1929 		btrfs_delayed_inode_release_metadata(fs_info, delayed_node);
1930 		btrfs_release_delayed_inode(delayed_node);
1931 	}
1932 	mutex_unlock(&delayed_node->mutex);
1933 }
1934 
1935 void btrfs_kill_delayed_inode_items(struct btrfs_inode *inode)
1936 {
1937 	struct btrfs_delayed_node *delayed_node;
1938 
1939 	delayed_node = btrfs_get_delayed_node(inode);
1940 	if (!delayed_node)
1941 		return;
1942 
1943 	__btrfs_kill_delayed_node(delayed_node);
1944 	btrfs_release_delayed_node(delayed_node);
1945 }
1946 
1947 void btrfs_kill_all_delayed_nodes(struct btrfs_root *root)
1948 {
1949 	u64 inode_id = 0;
1950 	struct btrfs_delayed_node *delayed_nodes[8];
1951 	int i, n;
1952 
1953 	while (1) {
1954 		spin_lock(&root->inode_lock);
1955 		n = radix_tree_gang_lookup(&root->delayed_nodes_tree,
1956 					   (void **)delayed_nodes, inode_id,
1957 					   ARRAY_SIZE(delayed_nodes));
1958 		if (!n) {
1959 			spin_unlock(&root->inode_lock);
1960 			break;
1961 		}
1962 
1963 		inode_id = delayed_nodes[n - 1]->inode_id + 1;
1964 
1965 		for (i = 0; i < n; i++)
1966 			atomic_inc(&delayed_nodes[i]->refs);
1967 		spin_unlock(&root->inode_lock);
1968 
1969 		for (i = 0; i < n; i++) {
1970 			__btrfs_kill_delayed_node(delayed_nodes[i]);
1971 			btrfs_release_delayed_node(delayed_nodes[i]);
1972 		}
1973 	}
1974 }
1975 
1976 void btrfs_destroy_delayed_inodes(struct btrfs_fs_info *fs_info)
1977 {
1978 	struct btrfs_delayed_node *curr_node, *prev_node;
1979 
1980 	curr_node = btrfs_first_delayed_node(fs_info->delayed_root);
1981 	while (curr_node) {
1982 		__btrfs_kill_delayed_node(curr_node);
1983 
1984 		prev_node = curr_node;
1985 		curr_node = btrfs_next_delayed_node(curr_node);
1986 		btrfs_release_delayed_node(prev_node);
1987 	}
1988 }
1989 
1990