xref: /openbmc/linux/fs/btrfs/delayed-inode.c (revision e7bae9bb)
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 <linux/sched/mm.h>
10 #include "misc.h"
11 #include "delayed-inode.h"
12 #include "disk-io.h"
13 #include "transaction.h"
14 #include "ctree.h"
15 #include "qgroup.h"
16 #include "locking.h"
17 
18 #define BTRFS_DELAYED_WRITEBACK		512
19 #define BTRFS_DELAYED_BACKGROUND	128
20 #define BTRFS_DELAYED_BATCH		16
21 
22 static struct kmem_cache *delayed_node_cache;
23 
24 int __init btrfs_delayed_inode_init(void)
25 {
26 	delayed_node_cache = kmem_cache_create("btrfs_delayed_node",
27 					sizeof(struct btrfs_delayed_node),
28 					0,
29 					SLAB_MEM_SPREAD,
30 					NULL);
31 	if (!delayed_node_cache)
32 		return -ENOMEM;
33 	return 0;
34 }
35 
36 void __cold btrfs_delayed_inode_exit(void)
37 {
38 	kmem_cache_destroy(delayed_node_cache);
39 }
40 
41 static inline void btrfs_init_delayed_node(
42 				struct btrfs_delayed_node *delayed_node,
43 				struct btrfs_root *root, u64 inode_id)
44 {
45 	delayed_node->root = root;
46 	delayed_node->inode_id = inode_id;
47 	refcount_set(&delayed_node->refs, 0);
48 	delayed_node->ins_root = RB_ROOT_CACHED;
49 	delayed_node->del_root = RB_ROOT_CACHED;
50 	mutex_init(&delayed_node->mutex);
51 	INIT_LIST_HEAD(&delayed_node->n_list);
52 	INIT_LIST_HEAD(&delayed_node->p_list);
53 }
54 
55 static inline int btrfs_is_continuous_delayed_item(
56 					struct btrfs_delayed_item *item1,
57 					struct btrfs_delayed_item *item2)
58 {
59 	if (item1->key.type == BTRFS_DIR_INDEX_KEY &&
60 	    item1->key.objectid == item2->key.objectid &&
61 	    item1->key.type == item2->key.type &&
62 	    item1->key.offset + 1 == item2->key.offset)
63 		return 1;
64 	return 0;
65 }
66 
67 static struct btrfs_delayed_node *btrfs_get_delayed_node(
68 		struct btrfs_inode *btrfs_inode)
69 {
70 	struct btrfs_root *root = btrfs_inode->root;
71 	u64 ino = btrfs_ino(btrfs_inode);
72 	struct btrfs_delayed_node *node;
73 
74 	node = READ_ONCE(btrfs_inode->delayed_node);
75 	if (node) {
76 		refcount_inc(&node->refs);
77 		return node;
78 	}
79 
80 	spin_lock(&root->inode_lock);
81 	node = radix_tree_lookup(&root->delayed_nodes_tree, ino);
82 
83 	if (node) {
84 		if (btrfs_inode->delayed_node) {
85 			refcount_inc(&node->refs);	/* can be accessed */
86 			BUG_ON(btrfs_inode->delayed_node != node);
87 			spin_unlock(&root->inode_lock);
88 			return node;
89 		}
90 
91 		/*
92 		 * It's possible that we're racing into the middle of removing
93 		 * this node from the radix tree.  In this case, the refcount
94 		 * was zero and it should never go back to one.  Just return
95 		 * NULL like it was never in the radix at all; our release
96 		 * function is in the process of removing it.
97 		 *
98 		 * Some implementations of refcount_inc refuse to bump the
99 		 * refcount once it has hit zero.  If we don't do this dance
100 		 * here, refcount_inc() may decide to just WARN_ONCE() instead
101 		 * of actually bumping the refcount.
102 		 *
103 		 * If this node is properly in the radix, we want to bump the
104 		 * refcount twice, once for the inode and once for this get
105 		 * operation.
106 		 */
107 		if (refcount_inc_not_zero(&node->refs)) {
108 			refcount_inc(&node->refs);
109 			btrfs_inode->delayed_node = node;
110 		} else {
111 			node = NULL;
112 		}
113 
114 		spin_unlock(&root->inode_lock);
115 		return node;
116 	}
117 	spin_unlock(&root->inode_lock);
118 
119 	return NULL;
120 }
121 
122 /* Will return either the node or PTR_ERR(-ENOMEM) */
123 static struct btrfs_delayed_node *btrfs_get_or_create_delayed_node(
124 		struct btrfs_inode *btrfs_inode)
125 {
126 	struct btrfs_delayed_node *node;
127 	struct btrfs_root *root = btrfs_inode->root;
128 	u64 ino = btrfs_ino(btrfs_inode);
129 	int ret;
130 
131 again:
132 	node = btrfs_get_delayed_node(btrfs_inode);
133 	if (node)
134 		return node;
135 
136 	node = kmem_cache_zalloc(delayed_node_cache, GFP_NOFS);
137 	if (!node)
138 		return ERR_PTR(-ENOMEM);
139 	btrfs_init_delayed_node(node, root, ino);
140 
141 	/* cached in the btrfs inode and can be accessed */
142 	refcount_set(&node->refs, 2);
143 
144 	ret = radix_tree_preload(GFP_NOFS);
145 	if (ret) {
146 		kmem_cache_free(delayed_node_cache, node);
147 		return ERR_PTR(ret);
148 	}
149 
150 	spin_lock(&root->inode_lock);
151 	ret = radix_tree_insert(&root->delayed_nodes_tree, ino, node);
152 	if (ret == -EEXIST) {
153 		spin_unlock(&root->inode_lock);
154 		kmem_cache_free(delayed_node_cache, node);
155 		radix_tree_preload_end();
156 		goto again;
157 	}
158 	btrfs_inode->delayed_node = node;
159 	spin_unlock(&root->inode_lock);
160 	radix_tree_preload_end();
161 
162 	return node;
163 }
164 
165 /*
166  * Call it when holding delayed_node->mutex
167  *
168  * If mod = 1, add this node into the prepared list.
169  */
170 static void btrfs_queue_delayed_node(struct btrfs_delayed_root *root,
171 				     struct btrfs_delayed_node *node,
172 				     int mod)
173 {
174 	spin_lock(&root->lock);
175 	if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
176 		if (!list_empty(&node->p_list))
177 			list_move_tail(&node->p_list, &root->prepare_list);
178 		else if (mod)
179 			list_add_tail(&node->p_list, &root->prepare_list);
180 	} else {
181 		list_add_tail(&node->n_list, &root->node_list);
182 		list_add_tail(&node->p_list, &root->prepare_list);
183 		refcount_inc(&node->refs);	/* inserted into list */
184 		root->nodes++;
185 		set_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
186 	}
187 	spin_unlock(&root->lock);
188 }
189 
190 /* Call it when holding delayed_node->mutex */
191 static void btrfs_dequeue_delayed_node(struct btrfs_delayed_root *root,
192 				       struct btrfs_delayed_node *node)
193 {
194 	spin_lock(&root->lock);
195 	if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
196 		root->nodes--;
197 		refcount_dec(&node->refs);	/* not in the list */
198 		list_del_init(&node->n_list);
199 		if (!list_empty(&node->p_list))
200 			list_del_init(&node->p_list);
201 		clear_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
202 	}
203 	spin_unlock(&root->lock);
204 }
205 
206 static struct btrfs_delayed_node *btrfs_first_delayed_node(
207 			struct btrfs_delayed_root *delayed_root)
208 {
209 	struct list_head *p;
210 	struct btrfs_delayed_node *node = NULL;
211 
212 	spin_lock(&delayed_root->lock);
213 	if (list_empty(&delayed_root->node_list))
214 		goto out;
215 
216 	p = delayed_root->node_list.next;
217 	node = list_entry(p, struct btrfs_delayed_node, n_list);
218 	refcount_inc(&node->refs);
219 out:
220 	spin_unlock(&delayed_root->lock);
221 
222 	return node;
223 }
224 
225 static struct btrfs_delayed_node *btrfs_next_delayed_node(
226 						struct btrfs_delayed_node *node)
227 {
228 	struct btrfs_delayed_root *delayed_root;
229 	struct list_head *p;
230 	struct btrfs_delayed_node *next = NULL;
231 
232 	delayed_root = node->root->fs_info->delayed_root;
233 	spin_lock(&delayed_root->lock);
234 	if (!test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
235 		/* not in the list */
236 		if (list_empty(&delayed_root->node_list))
237 			goto out;
238 		p = delayed_root->node_list.next;
239 	} else if (list_is_last(&node->n_list, &delayed_root->node_list))
240 		goto out;
241 	else
242 		p = node->n_list.next;
243 
244 	next = list_entry(p, struct btrfs_delayed_node, n_list);
245 	refcount_inc(&next->refs);
246 out:
247 	spin_unlock(&delayed_root->lock);
248 
249 	return next;
250 }
251 
252 static void __btrfs_release_delayed_node(
253 				struct btrfs_delayed_node *delayed_node,
254 				int mod)
255 {
256 	struct btrfs_delayed_root *delayed_root;
257 
258 	if (!delayed_node)
259 		return;
260 
261 	delayed_root = delayed_node->root->fs_info->delayed_root;
262 
263 	mutex_lock(&delayed_node->mutex);
264 	if (delayed_node->count)
265 		btrfs_queue_delayed_node(delayed_root, delayed_node, mod);
266 	else
267 		btrfs_dequeue_delayed_node(delayed_root, delayed_node);
268 	mutex_unlock(&delayed_node->mutex);
269 
270 	if (refcount_dec_and_test(&delayed_node->refs)) {
271 		struct btrfs_root *root = delayed_node->root;
272 
273 		spin_lock(&root->inode_lock);
274 		/*
275 		 * Once our refcount goes to zero, nobody is allowed to bump it
276 		 * back up.  We can delete it now.
277 		 */
278 		ASSERT(refcount_read(&delayed_node->refs) == 0);
279 		radix_tree_delete(&root->delayed_nodes_tree,
280 				  delayed_node->inode_id);
281 		spin_unlock(&root->inode_lock);
282 		kmem_cache_free(delayed_node_cache, delayed_node);
283 	}
284 }
285 
286 static inline void btrfs_release_delayed_node(struct btrfs_delayed_node *node)
287 {
288 	__btrfs_release_delayed_node(node, 0);
289 }
290 
291 static struct btrfs_delayed_node *btrfs_first_prepared_delayed_node(
292 					struct btrfs_delayed_root *delayed_root)
293 {
294 	struct list_head *p;
295 	struct btrfs_delayed_node *node = NULL;
296 
297 	spin_lock(&delayed_root->lock);
298 	if (list_empty(&delayed_root->prepare_list))
299 		goto out;
300 
301 	p = delayed_root->prepare_list.next;
302 	list_del_init(p);
303 	node = list_entry(p, struct btrfs_delayed_node, p_list);
304 	refcount_inc(&node->refs);
305 out:
306 	spin_unlock(&delayed_root->lock);
307 
308 	return node;
309 }
310 
311 static inline void btrfs_release_prepared_delayed_node(
312 					struct btrfs_delayed_node *node)
313 {
314 	__btrfs_release_delayed_node(node, 1);
315 }
316 
317 static struct btrfs_delayed_item *btrfs_alloc_delayed_item(u32 data_len)
318 {
319 	struct btrfs_delayed_item *item;
320 	item = kmalloc(sizeof(*item) + data_len, GFP_NOFS);
321 	if (item) {
322 		item->data_len = data_len;
323 		item->ins_or_del = 0;
324 		item->bytes_reserved = 0;
325 		item->delayed_node = NULL;
326 		refcount_set(&item->refs, 1);
327 	}
328 	return item;
329 }
330 
331 /*
332  * __btrfs_lookup_delayed_item - look up the delayed item by key
333  * @delayed_node: pointer to the delayed node
334  * @key:	  the key to look up
335  * @prev:	  used to store the prev item if the right item isn't found
336  * @next:	  used to store the next item if the right item isn't found
337  *
338  * Note: if we don't find the right item, we will return the prev item and
339  * the next item.
340  */
341 static struct btrfs_delayed_item *__btrfs_lookup_delayed_item(
342 				struct rb_root *root,
343 				struct btrfs_key *key,
344 				struct btrfs_delayed_item **prev,
345 				struct btrfs_delayed_item **next)
346 {
347 	struct rb_node *node, *prev_node = NULL;
348 	struct btrfs_delayed_item *delayed_item = NULL;
349 	int ret = 0;
350 
351 	node = root->rb_node;
352 
353 	while (node) {
354 		delayed_item = rb_entry(node, struct btrfs_delayed_item,
355 					rb_node);
356 		prev_node = node;
357 		ret = btrfs_comp_cpu_keys(&delayed_item->key, key);
358 		if (ret < 0)
359 			node = node->rb_right;
360 		else if (ret > 0)
361 			node = node->rb_left;
362 		else
363 			return delayed_item;
364 	}
365 
366 	if (prev) {
367 		if (!prev_node)
368 			*prev = NULL;
369 		else if (ret < 0)
370 			*prev = delayed_item;
371 		else if ((node = rb_prev(prev_node)) != NULL) {
372 			*prev = rb_entry(node, struct btrfs_delayed_item,
373 					 rb_node);
374 		} else
375 			*prev = NULL;
376 	}
377 
378 	if (next) {
379 		if (!prev_node)
380 			*next = NULL;
381 		else if (ret > 0)
382 			*next = delayed_item;
383 		else if ((node = rb_next(prev_node)) != NULL) {
384 			*next = rb_entry(node, struct btrfs_delayed_item,
385 					 rb_node);
386 		} else
387 			*next = NULL;
388 	}
389 	return NULL;
390 }
391 
392 static struct btrfs_delayed_item *__btrfs_lookup_delayed_insertion_item(
393 					struct btrfs_delayed_node *delayed_node,
394 					struct btrfs_key *key)
395 {
396 	return __btrfs_lookup_delayed_item(&delayed_node->ins_root.rb_root, key,
397 					   NULL, NULL);
398 }
399 
400 static int __btrfs_add_delayed_item(struct btrfs_delayed_node *delayed_node,
401 				    struct btrfs_delayed_item *ins,
402 				    int action)
403 {
404 	struct rb_node **p, *node;
405 	struct rb_node *parent_node = NULL;
406 	struct rb_root_cached *root;
407 	struct btrfs_delayed_item *item;
408 	int cmp;
409 	bool leftmost = true;
410 
411 	if (action == BTRFS_DELAYED_INSERTION_ITEM)
412 		root = &delayed_node->ins_root;
413 	else if (action == BTRFS_DELAYED_DELETION_ITEM)
414 		root = &delayed_node->del_root;
415 	else
416 		BUG();
417 	p = &root->rb_root.rb_node;
418 	node = &ins->rb_node;
419 
420 	while (*p) {
421 		parent_node = *p;
422 		item = rb_entry(parent_node, struct btrfs_delayed_item,
423 				 rb_node);
424 
425 		cmp = btrfs_comp_cpu_keys(&item->key, &ins->key);
426 		if (cmp < 0) {
427 			p = &(*p)->rb_right;
428 			leftmost = false;
429 		} else if (cmp > 0) {
430 			p = &(*p)->rb_left;
431 		} else {
432 			return -EEXIST;
433 		}
434 	}
435 
436 	rb_link_node(node, parent_node, p);
437 	rb_insert_color_cached(node, root, leftmost);
438 	ins->delayed_node = delayed_node;
439 	ins->ins_or_del = action;
440 
441 	if (ins->key.type == BTRFS_DIR_INDEX_KEY &&
442 	    action == BTRFS_DELAYED_INSERTION_ITEM &&
443 	    ins->key.offset >= delayed_node->index_cnt)
444 			delayed_node->index_cnt = ins->key.offset + 1;
445 
446 	delayed_node->count++;
447 	atomic_inc(&delayed_node->root->fs_info->delayed_root->items);
448 	return 0;
449 }
450 
451 static int __btrfs_add_delayed_insertion_item(struct btrfs_delayed_node *node,
452 					      struct btrfs_delayed_item *item)
453 {
454 	return __btrfs_add_delayed_item(node, item,
455 					BTRFS_DELAYED_INSERTION_ITEM);
456 }
457 
458 static int __btrfs_add_delayed_deletion_item(struct btrfs_delayed_node *node,
459 					     struct btrfs_delayed_item *item)
460 {
461 	return __btrfs_add_delayed_item(node, item,
462 					BTRFS_DELAYED_DELETION_ITEM);
463 }
464 
465 static void finish_one_item(struct btrfs_delayed_root *delayed_root)
466 {
467 	int seq = atomic_inc_return(&delayed_root->items_seq);
468 
469 	/* atomic_dec_return implies a barrier */
470 	if ((atomic_dec_return(&delayed_root->items) <
471 	    BTRFS_DELAYED_BACKGROUND || seq % BTRFS_DELAYED_BATCH == 0))
472 		cond_wake_up_nomb(&delayed_root->wait);
473 }
474 
475 static void __btrfs_remove_delayed_item(struct btrfs_delayed_item *delayed_item)
476 {
477 	struct rb_root_cached *root;
478 	struct btrfs_delayed_root *delayed_root;
479 
480 	/* Not associated with any delayed_node */
481 	if (!delayed_item->delayed_node)
482 		return;
483 	delayed_root = delayed_item->delayed_node->root->fs_info->delayed_root;
484 
485 	BUG_ON(!delayed_root);
486 	BUG_ON(delayed_item->ins_or_del != BTRFS_DELAYED_DELETION_ITEM &&
487 	       delayed_item->ins_or_del != BTRFS_DELAYED_INSERTION_ITEM);
488 
489 	if (delayed_item->ins_or_del == BTRFS_DELAYED_INSERTION_ITEM)
490 		root = &delayed_item->delayed_node->ins_root;
491 	else
492 		root = &delayed_item->delayed_node->del_root;
493 
494 	rb_erase_cached(&delayed_item->rb_node, root);
495 	delayed_item->delayed_node->count--;
496 
497 	finish_one_item(delayed_root);
498 }
499 
500 static void btrfs_release_delayed_item(struct btrfs_delayed_item *item)
501 {
502 	if (item) {
503 		__btrfs_remove_delayed_item(item);
504 		if (refcount_dec_and_test(&item->refs))
505 			kfree(item);
506 	}
507 }
508 
509 static struct btrfs_delayed_item *__btrfs_first_delayed_insertion_item(
510 					struct btrfs_delayed_node *delayed_node)
511 {
512 	struct rb_node *p;
513 	struct btrfs_delayed_item *item = NULL;
514 
515 	p = rb_first_cached(&delayed_node->ins_root);
516 	if (p)
517 		item = rb_entry(p, struct btrfs_delayed_item, rb_node);
518 
519 	return item;
520 }
521 
522 static struct btrfs_delayed_item *__btrfs_first_delayed_deletion_item(
523 					struct btrfs_delayed_node *delayed_node)
524 {
525 	struct rb_node *p;
526 	struct btrfs_delayed_item *item = NULL;
527 
528 	p = rb_first_cached(&delayed_node->del_root);
529 	if (p)
530 		item = rb_entry(p, struct btrfs_delayed_item, rb_node);
531 
532 	return item;
533 }
534 
535 static struct btrfs_delayed_item *__btrfs_next_delayed_item(
536 						struct btrfs_delayed_item *item)
537 {
538 	struct rb_node *p;
539 	struct btrfs_delayed_item *next = NULL;
540 
541 	p = rb_next(&item->rb_node);
542 	if (p)
543 		next = rb_entry(p, struct btrfs_delayed_item, rb_node);
544 
545 	return next;
546 }
547 
548 static int btrfs_delayed_item_reserve_metadata(struct btrfs_trans_handle *trans,
549 					       struct btrfs_root *root,
550 					       struct btrfs_delayed_item *item)
551 {
552 	struct btrfs_block_rsv *src_rsv;
553 	struct btrfs_block_rsv *dst_rsv;
554 	struct btrfs_fs_info *fs_info = root->fs_info;
555 	u64 num_bytes;
556 	int ret;
557 
558 	if (!trans->bytes_reserved)
559 		return 0;
560 
561 	src_rsv = trans->block_rsv;
562 	dst_rsv = &fs_info->delayed_block_rsv;
563 
564 	num_bytes = btrfs_calc_insert_metadata_size(fs_info, 1);
565 
566 	/*
567 	 * Here we migrate space rsv from transaction rsv, since have already
568 	 * reserved space when starting a transaction.  So no need to reserve
569 	 * qgroup space here.
570 	 */
571 	ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, true);
572 	if (!ret) {
573 		trace_btrfs_space_reservation(fs_info, "delayed_item",
574 					      item->key.objectid,
575 					      num_bytes, 1);
576 		item->bytes_reserved = num_bytes;
577 	}
578 
579 	return ret;
580 }
581 
582 static void btrfs_delayed_item_release_metadata(struct btrfs_root *root,
583 						struct btrfs_delayed_item *item)
584 {
585 	struct btrfs_block_rsv *rsv;
586 	struct btrfs_fs_info *fs_info = root->fs_info;
587 
588 	if (!item->bytes_reserved)
589 		return;
590 
591 	rsv = &fs_info->delayed_block_rsv;
592 	/*
593 	 * Check btrfs_delayed_item_reserve_metadata() to see why we don't need
594 	 * to release/reserve qgroup space.
595 	 */
596 	trace_btrfs_space_reservation(fs_info, "delayed_item",
597 				      item->key.objectid, item->bytes_reserved,
598 				      0);
599 	btrfs_block_rsv_release(fs_info, rsv, item->bytes_reserved, NULL);
600 }
601 
602 static int btrfs_delayed_inode_reserve_metadata(
603 					struct btrfs_trans_handle *trans,
604 					struct btrfs_root *root,
605 					struct btrfs_inode *inode,
606 					struct btrfs_delayed_node *node)
607 {
608 	struct btrfs_fs_info *fs_info = root->fs_info;
609 	struct btrfs_block_rsv *src_rsv;
610 	struct btrfs_block_rsv *dst_rsv;
611 	u64 num_bytes;
612 	int ret;
613 
614 	src_rsv = trans->block_rsv;
615 	dst_rsv = &fs_info->delayed_block_rsv;
616 
617 	num_bytes = btrfs_calc_metadata_size(fs_info, 1);
618 
619 	/*
620 	 * btrfs_dirty_inode will update the inode under btrfs_join_transaction
621 	 * which doesn't reserve space for speed.  This is a problem since we
622 	 * still need to reserve space for this update, so try to reserve the
623 	 * space.
624 	 *
625 	 * Now if src_rsv == delalloc_block_rsv we'll let it just steal since
626 	 * we always reserve enough to update the inode item.
627 	 */
628 	if (!src_rsv || (!trans->bytes_reserved &&
629 			 src_rsv->type != BTRFS_BLOCK_RSV_DELALLOC)) {
630 		ret = btrfs_qgroup_reserve_meta_prealloc(root,
631 				fs_info->nodesize, true);
632 		if (ret < 0)
633 			return ret;
634 		ret = btrfs_block_rsv_add(root, dst_rsv, num_bytes,
635 					  BTRFS_RESERVE_NO_FLUSH);
636 		/*
637 		 * Since we're under a transaction reserve_metadata_bytes could
638 		 * try to commit the transaction which will make it return
639 		 * EAGAIN to make us stop the transaction we have, so return
640 		 * ENOSPC instead so that btrfs_dirty_inode knows what to do.
641 		 */
642 		if (ret == -EAGAIN) {
643 			ret = -ENOSPC;
644 			btrfs_qgroup_free_meta_prealloc(root, num_bytes);
645 		}
646 		if (!ret) {
647 			node->bytes_reserved = num_bytes;
648 			trace_btrfs_space_reservation(fs_info,
649 						      "delayed_inode",
650 						      btrfs_ino(inode),
651 						      num_bytes, 1);
652 		} else {
653 			btrfs_qgroup_free_meta_prealloc(root, fs_info->nodesize);
654 		}
655 		return ret;
656 	}
657 
658 	ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, true);
659 	if (!ret) {
660 		trace_btrfs_space_reservation(fs_info, "delayed_inode",
661 					      btrfs_ino(inode), num_bytes, 1);
662 		node->bytes_reserved = num_bytes;
663 	}
664 
665 	return ret;
666 }
667 
668 static void btrfs_delayed_inode_release_metadata(struct btrfs_fs_info *fs_info,
669 						struct btrfs_delayed_node *node,
670 						bool qgroup_free)
671 {
672 	struct btrfs_block_rsv *rsv;
673 
674 	if (!node->bytes_reserved)
675 		return;
676 
677 	rsv = &fs_info->delayed_block_rsv;
678 	trace_btrfs_space_reservation(fs_info, "delayed_inode",
679 				      node->inode_id, node->bytes_reserved, 0);
680 	btrfs_block_rsv_release(fs_info, rsv, node->bytes_reserved, NULL);
681 	if (qgroup_free)
682 		btrfs_qgroup_free_meta_prealloc(node->root,
683 				node->bytes_reserved);
684 	else
685 		btrfs_qgroup_convert_reserved_meta(node->root,
686 				node->bytes_reserved);
687 	node->bytes_reserved = 0;
688 }
689 
690 /*
691  * This helper will insert some continuous items into the same leaf according
692  * to the free space of the leaf.
693  */
694 static int btrfs_batch_insert_items(struct btrfs_root *root,
695 				    struct btrfs_path *path,
696 				    struct btrfs_delayed_item *item)
697 {
698 	struct btrfs_delayed_item *curr, *next;
699 	int free_space;
700 	int total_data_size = 0, total_size = 0;
701 	struct extent_buffer *leaf;
702 	char *data_ptr;
703 	struct btrfs_key *keys;
704 	u32 *data_size;
705 	struct list_head head;
706 	int slot;
707 	int nitems;
708 	int i;
709 	int ret = 0;
710 
711 	BUG_ON(!path->nodes[0]);
712 
713 	leaf = path->nodes[0];
714 	free_space = btrfs_leaf_free_space(leaf);
715 	INIT_LIST_HEAD(&head);
716 
717 	next = item;
718 	nitems = 0;
719 
720 	/*
721 	 * count the number of the continuous items that we can insert in batch
722 	 */
723 	while (total_size + next->data_len + sizeof(struct btrfs_item) <=
724 	       free_space) {
725 		total_data_size += next->data_len;
726 		total_size += next->data_len + sizeof(struct btrfs_item);
727 		list_add_tail(&next->tree_list, &head);
728 		nitems++;
729 
730 		curr = next;
731 		next = __btrfs_next_delayed_item(curr);
732 		if (!next)
733 			break;
734 
735 		if (!btrfs_is_continuous_delayed_item(curr, next))
736 			break;
737 	}
738 
739 	if (!nitems) {
740 		ret = 0;
741 		goto out;
742 	}
743 
744 	/*
745 	 * we need allocate some memory space, but it might cause the task
746 	 * to sleep, so we set all locked nodes in the path to blocking locks
747 	 * first.
748 	 */
749 	btrfs_set_path_blocking(path);
750 
751 	keys = kmalloc_array(nitems, sizeof(struct btrfs_key), GFP_NOFS);
752 	if (!keys) {
753 		ret = -ENOMEM;
754 		goto out;
755 	}
756 
757 	data_size = kmalloc_array(nitems, sizeof(u32), GFP_NOFS);
758 	if (!data_size) {
759 		ret = -ENOMEM;
760 		goto error;
761 	}
762 
763 	/* get keys of all the delayed items */
764 	i = 0;
765 	list_for_each_entry(next, &head, tree_list) {
766 		keys[i] = next->key;
767 		data_size[i] = next->data_len;
768 		i++;
769 	}
770 
771 	/* insert the keys of the items */
772 	setup_items_for_insert(root, path, keys, data_size,
773 			       total_data_size, total_size, nitems);
774 
775 	/* insert the dir index items */
776 	slot = path->slots[0];
777 	list_for_each_entry_safe(curr, next, &head, tree_list) {
778 		data_ptr = btrfs_item_ptr(leaf, slot, char);
779 		write_extent_buffer(leaf, &curr->data,
780 				    (unsigned long)data_ptr,
781 				    curr->data_len);
782 		slot++;
783 
784 		btrfs_delayed_item_release_metadata(root, curr);
785 
786 		list_del(&curr->tree_list);
787 		btrfs_release_delayed_item(curr);
788 	}
789 
790 error:
791 	kfree(data_size);
792 	kfree(keys);
793 out:
794 	return ret;
795 }
796 
797 /*
798  * This helper can just do simple insertion that needn't extend item for new
799  * data, such as directory name index insertion, inode insertion.
800  */
801 static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans,
802 				     struct btrfs_root *root,
803 				     struct btrfs_path *path,
804 				     struct btrfs_delayed_item *delayed_item)
805 {
806 	struct extent_buffer *leaf;
807 	unsigned int nofs_flag;
808 	char *ptr;
809 	int ret;
810 
811 	nofs_flag = memalloc_nofs_save();
812 	ret = btrfs_insert_empty_item(trans, root, path, &delayed_item->key,
813 				      delayed_item->data_len);
814 	memalloc_nofs_restore(nofs_flag);
815 	if (ret < 0 && ret != -EEXIST)
816 		return ret;
817 
818 	leaf = path->nodes[0];
819 
820 	ptr = btrfs_item_ptr(leaf, path->slots[0], char);
821 
822 	write_extent_buffer(leaf, delayed_item->data, (unsigned long)ptr,
823 			    delayed_item->data_len);
824 	btrfs_mark_buffer_dirty(leaf);
825 
826 	btrfs_delayed_item_release_metadata(root, delayed_item);
827 	return 0;
828 }
829 
830 /*
831  * we insert an item first, then if there are some continuous items, we try
832  * to insert those items into the same leaf.
833  */
834 static int btrfs_insert_delayed_items(struct btrfs_trans_handle *trans,
835 				      struct btrfs_path *path,
836 				      struct btrfs_root *root,
837 				      struct btrfs_delayed_node *node)
838 {
839 	struct btrfs_delayed_item *curr, *prev;
840 	int ret = 0;
841 
842 do_again:
843 	mutex_lock(&node->mutex);
844 	curr = __btrfs_first_delayed_insertion_item(node);
845 	if (!curr)
846 		goto insert_end;
847 
848 	ret = btrfs_insert_delayed_item(trans, root, path, curr);
849 	if (ret < 0) {
850 		btrfs_release_path(path);
851 		goto insert_end;
852 	}
853 
854 	prev = curr;
855 	curr = __btrfs_next_delayed_item(prev);
856 	if (curr && btrfs_is_continuous_delayed_item(prev, curr)) {
857 		/* insert the continuous items into the same leaf */
858 		path->slots[0]++;
859 		btrfs_batch_insert_items(root, path, curr);
860 	}
861 	btrfs_release_delayed_item(prev);
862 	btrfs_mark_buffer_dirty(path->nodes[0]);
863 
864 	btrfs_release_path(path);
865 	mutex_unlock(&node->mutex);
866 	goto do_again;
867 
868 insert_end:
869 	mutex_unlock(&node->mutex);
870 	return ret;
871 }
872 
873 static int btrfs_batch_delete_items(struct btrfs_trans_handle *trans,
874 				    struct btrfs_root *root,
875 				    struct btrfs_path *path,
876 				    struct btrfs_delayed_item *item)
877 {
878 	struct btrfs_delayed_item *curr, *next;
879 	struct extent_buffer *leaf;
880 	struct btrfs_key key;
881 	struct list_head head;
882 	int nitems, i, last_item;
883 	int ret = 0;
884 
885 	BUG_ON(!path->nodes[0]);
886 
887 	leaf = path->nodes[0];
888 
889 	i = path->slots[0];
890 	last_item = btrfs_header_nritems(leaf) - 1;
891 	if (i > last_item)
892 		return -ENOENT;	/* FIXME: Is errno suitable? */
893 
894 	next = item;
895 	INIT_LIST_HEAD(&head);
896 	btrfs_item_key_to_cpu(leaf, &key, i);
897 	nitems = 0;
898 	/*
899 	 * count the number of the dir index items that we can delete in batch
900 	 */
901 	while (btrfs_comp_cpu_keys(&next->key, &key) == 0) {
902 		list_add_tail(&next->tree_list, &head);
903 		nitems++;
904 
905 		curr = next;
906 		next = __btrfs_next_delayed_item(curr);
907 		if (!next)
908 			break;
909 
910 		if (!btrfs_is_continuous_delayed_item(curr, next))
911 			break;
912 
913 		i++;
914 		if (i > last_item)
915 			break;
916 		btrfs_item_key_to_cpu(leaf, &key, i);
917 	}
918 
919 	if (!nitems)
920 		return 0;
921 
922 	ret = btrfs_del_items(trans, root, path, path->slots[0], nitems);
923 	if (ret)
924 		goto out;
925 
926 	list_for_each_entry_safe(curr, next, &head, tree_list) {
927 		btrfs_delayed_item_release_metadata(root, curr);
928 		list_del(&curr->tree_list);
929 		btrfs_release_delayed_item(curr);
930 	}
931 
932 out:
933 	return ret;
934 }
935 
936 static int btrfs_delete_delayed_items(struct btrfs_trans_handle *trans,
937 				      struct btrfs_path *path,
938 				      struct btrfs_root *root,
939 				      struct btrfs_delayed_node *node)
940 {
941 	struct btrfs_delayed_item *curr, *prev;
942 	unsigned int nofs_flag;
943 	int ret = 0;
944 
945 do_again:
946 	mutex_lock(&node->mutex);
947 	curr = __btrfs_first_delayed_deletion_item(node);
948 	if (!curr)
949 		goto delete_fail;
950 
951 	nofs_flag = memalloc_nofs_save();
952 	ret = btrfs_search_slot(trans, root, &curr->key, path, -1, 1);
953 	memalloc_nofs_restore(nofs_flag);
954 	if (ret < 0)
955 		goto delete_fail;
956 	else if (ret > 0) {
957 		/*
958 		 * can't find the item which the node points to, so this node
959 		 * is invalid, just drop it.
960 		 */
961 		prev = curr;
962 		curr = __btrfs_next_delayed_item(prev);
963 		btrfs_release_delayed_item(prev);
964 		ret = 0;
965 		btrfs_release_path(path);
966 		if (curr) {
967 			mutex_unlock(&node->mutex);
968 			goto do_again;
969 		} else
970 			goto delete_fail;
971 	}
972 
973 	btrfs_batch_delete_items(trans, root, path, curr);
974 	btrfs_release_path(path);
975 	mutex_unlock(&node->mutex);
976 	goto do_again;
977 
978 delete_fail:
979 	btrfs_release_path(path);
980 	mutex_unlock(&node->mutex);
981 	return ret;
982 }
983 
984 static void btrfs_release_delayed_inode(struct btrfs_delayed_node *delayed_node)
985 {
986 	struct btrfs_delayed_root *delayed_root;
987 
988 	if (delayed_node &&
989 	    test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
990 		BUG_ON(!delayed_node->root);
991 		clear_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
992 		delayed_node->count--;
993 
994 		delayed_root = delayed_node->root->fs_info->delayed_root;
995 		finish_one_item(delayed_root);
996 	}
997 }
998 
999 static void btrfs_release_delayed_iref(struct btrfs_delayed_node *delayed_node)
1000 {
1001 	struct btrfs_delayed_root *delayed_root;
1002 
1003 	ASSERT(delayed_node->root);
1004 	clear_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
1005 	delayed_node->count--;
1006 
1007 	delayed_root = delayed_node->root->fs_info->delayed_root;
1008 	finish_one_item(delayed_root);
1009 }
1010 
1011 static int __btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1012 					struct btrfs_root *root,
1013 					struct btrfs_path *path,
1014 					struct btrfs_delayed_node *node)
1015 {
1016 	struct btrfs_fs_info *fs_info = root->fs_info;
1017 	struct btrfs_key key;
1018 	struct btrfs_inode_item *inode_item;
1019 	struct extent_buffer *leaf;
1020 	unsigned int nofs_flag;
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 	nofs_flag = memalloc_nofs_save();
1034 	ret = btrfs_lookup_inode(trans, root, path, &key, mod);
1035 	memalloc_nofs_restore(nofs_flag);
1036 	if (ret > 0) {
1037 		btrfs_release_path(path);
1038 		return -ENOENT;
1039 	} else if (ret < 0) {
1040 		return ret;
1041 	}
1042 
1043 	leaf = path->nodes[0];
1044 	inode_item = btrfs_item_ptr(leaf, path->slots[0],
1045 				    struct btrfs_inode_item);
1046 	write_extent_buffer(leaf, &node->inode_item, (unsigned long)inode_item,
1047 			    sizeof(struct btrfs_inode_item));
1048 	btrfs_mark_buffer_dirty(leaf);
1049 
1050 	if (!test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1051 		goto no_iref;
1052 
1053 	path->slots[0]++;
1054 	if (path->slots[0] >= btrfs_header_nritems(leaf))
1055 		goto search;
1056 again:
1057 	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1058 	if (key.objectid != node->inode_id)
1059 		goto out;
1060 
1061 	if (key.type != BTRFS_INODE_REF_KEY &&
1062 	    key.type != BTRFS_INODE_EXTREF_KEY)
1063 		goto out;
1064 
1065 	/*
1066 	 * Delayed iref deletion is for the inode who has only one link,
1067 	 * so there is only one iref. The case that several irefs are
1068 	 * in the same item doesn't exist.
1069 	 */
1070 	btrfs_del_item(trans, root, path);
1071 out:
1072 	btrfs_release_delayed_iref(node);
1073 no_iref:
1074 	btrfs_release_path(path);
1075 err_out:
1076 	btrfs_delayed_inode_release_metadata(fs_info, node, (ret < 0));
1077 	btrfs_release_delayed_inode(node);
1078 
1079 	return ret;
1080 
1081 search:
1082 	btrfs_release_path(path);
1083 
1084 	key.type = BTRFS_INODE_EXTREF_KEY;
1085 	key.offset = -1;
1086 
1087 	nofs_flag = memalloc_nofs_save();
1088 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1089 	memalloc_nofs_restore(nofs_flag);
1090 	if (ret < 0)
1091 		goto err_out;
1092 	ASSERT(ret);
1093 
1094 	ret = 0;
1095 	leaf = path->nodes[0];
1096 	path->slots[0]--;
1097 	goto again;
1098 }
1099 
1100 static inline int btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1101 					     struct btrfs_root *root,
1102 					     struct btrfs_path *path,
1103 					     struct btrfs_delayed_node *node)
1104 {
1105 	int ret;
1106 
1107 	mutex_lock(&node->mutex);
1108 	if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &node->flags)) {
1109 		mutex_unlock(&node->mutex);
1110 		return 0;
1111 	}
1112 
1113 	ret = __btrfs_update_delayed_inode(trans, root, path, node);
1114 	mutex_unlock(&node->mutex);
1115 	return ret;
1116 }
1117 
1118 static inline int
1119 __btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1120 				   struct btrfs_path *path,
1121 				   struct btrfs_delayed_node *node)
1122 {
1123 	int ret;
1124 
1125 	ret = btrfs_insert_delayed_items(trans, path, node->root, node);
1126 	if (ret)
1127 		return ret;
1128 
1129 	ret = btrfs_delete_delayed_items(trans, path, node->root, node);
1130 	if (ret)
1131 		return ret;
1132 
1133 	ret = btrfs_update_delayed_inode(trans, node->root, path, node);
1134 	return ret;
1135 }
1136 
1137 /*
1138  * Called when committing the transaction.
1139  * Returns 0 on success.
1140  * Returns < 0 on error and returns with an aborted transaction with any
1141  * outstanding delayed items cleaned up.
1142  */
1143 static int __btrfs_run_delayed_items(struct btrfs_trans_handle *trans, int nr)
1144 {
1145 	struct btrfs_fs_info *fs_info = trans->fs_info;
1146 	struct btrfs_delayed_root *delayed_root;
1147 	struct btrfs_delayed_node *curr_node, *prev_node;
1148 	struct btrfs_path *path;
1149 	struct btrfs_block_rsv *block_rsv;
1150 	int ret = 0;
1151 	bool count = (nr > 0);
1152 
1153 	if (TRANS_ABORTED(trans))
1154 		return -EIO;
1155 
1156 	path = btrfs_alloc_path();
1157 	if (!path)
1158 		return -ENOMEM;
1159 	path->leave_spinning = 1;
1160 
1161 	block_rsv = trans->block_rsv;
1162 	trans->block_rsv = &fs_info->delayed_block_rsv;
1163 
1164 	delayed_root = fs_info->delayed_root;
1165 
1166 	curr_node = btrfs_first_delayed_node(delayed_root);
1167 	while (curr_node && (!count || (count && nr--))) {
1168 		ret = __btrfs_commit_inode_delayed_items(trans, path,
1169 							 curr_node);
1170 		if (ret) {
1171 			btrfs_release_delayed_node(curr_node);
1172 			curr_node = NULL;
1173 			btrfs_abort_transaction(trans, ret);
1174 			break;
1175 		}
1176 
1177 		prev_node = curr_node;
1178 		curr_node = btrfs_next_delayed_node(curr_node);
1179 		btrfs_release_delayed_node(prev_node);
1180 	}
1181 
1182 	if (curr_node)
1183 		btrfs_release_delayed_node(curr_node);
1184 	btrfs_free_path(path);
1185 	trans->block_rsv = block_rsv;
1186 
1187 	return ret;
1188 }
1189 
1190 int btrfs_run_delayed_items(struct btrfs_trans_handle *trans)
1191 {
1192 	return __btrfs_run_delayed_items(trans, -1);
1193 }
1194 
1195 int btrfs_run_delayed_items_nr(struct btrfs_trans_handle *trans, int nr)
1196 {
1197 	return __btrfs_run_delayed_items(trans, nr);
1198 }
1199 
1200 int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1201 				     struct btrfs_inode *inode)
1202 {
1203 	struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1204 	struct btrfs_path *path;
1205 	struct btrfs_block_rsv *block_rsv;
1206 	int ret;
1207 
1208 	if (!delayed_node)
1209 		return 0;
1210 
1211 	mutex_lock(&delayed_node->mutex);
1212 	if (!delayed_node->count) {
1213 		mutex_unlock(&delayed_node->mutex);
1214 		btrfs_release_delayed_node(delayed_node);
1215 		return 0;
1216 	}
1217 	mutex_unlock(&delayed_node->mutex);
1218 
1219 	path = btrfs_alloc_path();
1220 	if (!path) {
1221 		btrfs_release_delayed_node(delayed_node);
1222 		return -ENOMEM;
1223 	}
1224 	path->leave_spinning = 1;
1225 
1226 	block_rsv = trans->block_rsv;
1227 	trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv;
1228 
1229 	ret = __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1230 
1231 	btrfs_release_delayed_node(delayed_node);
1232 	btrfs_free_path(path);
1233 	trans->block_rsv = block_rsv;
1234 
1235 	return ret;
1236 }
1237 
1238 int btrfs_commit_inode_delayed_inode(struct btrfs_inode *inode)
1239 {
1240 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
1241 	struct btrfs_trans_handle *trans;
1242 	struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1243 	struct btrfs_path *path;
1244 	struct btrfs_block_rsv *block_rsv;
1245 	int ret;
1246 
1247 	if (!delayed_node)
1248 		return 0;
1249 
1250 	mutex_lock(&delayed_node->mutex);
1251 	if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1252 		mutex_unlock(&delayed_node->mutex);
1253 		btrfs_release_delayed_node(delayed_node);
1254 		return 0;
1255 	}
1256 	mutex_unlock(&delayed_node->mutex);
1257 
1258 	trans = btrfs_join_transaction(delayed_node->root);
1259 	if (IS_ERR(trans)) {
1260 		ret = PTR_ERR(trans);
1261 		goto out;
1262 	}
1263 
1264 	path = btrfs_alloc_path();
1265 	if (!path) {
1266 		ret = -ENOMEM;
1267 		goto trans_out;
1268 	}
1269 	path->leave_spinning = 1;
1270 
1271 	block_rsv = trans->block_rsv;
1272 	trans->block_rsv = &fs_info->delayed_block_rsv;
1273 
1274 	mutex_lock(&delayed_node->mutex);
1275 	if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags))
1276 		ret = __btrfs_update_delayed_inode(trans, delayed_node->root,
1277 						   path, delayed_node);
1278 	else
1279 		ret = 0;
1280 	mutex_unlock(&delayed_node->mutex);
1281 
1282 	btrfs_free_path(path);
1283 	trans->block_rsv = block_rsv;
1284 trans_out:
1285 	btrfs_end_transaction(trans);
1286 	btrfs_btree_balance_dirty(fs_info);
1287 out:
1288 	btrfs_release_delayed_node(delayed_node);
1289 
1290 	return ret;
1291 }
1292 
1293 void btrfs_remove_delayed_node(struct btrfs_inode *inode)
1294 {
1295 	struct btrfs_delayed_node *delayed_node;
1296 
1297 	delayed_node = READ_ONCE(inode->delayed_node);
1298 	if (!delayed_node)
1299 		return;
1300 
1301 	inode->delayed_node = NULL;
1302 	btrfs_release_delayed_node(delayed_node);
1303 }
1304 
1305 struct btrfs_async_delayed_work {
1306 	struct btrfs_delayed_root *delayed_root;
1307 	int nr;
1308 	struct btrfs_work work;
1309 };
1310 
1311 static void btrfs_async_run_delayed_root(struct btrfs_work *work)
1312 {
1313 	struct btrfs_async_delayed_work *async_work;
1314 	struct btrfs_delayed_root *delayed_root;
1315 	struct btrfs_trans_handle *trans;
1316 	struct btrfs_path *path;
1317 	struct btrfs_delayed_node *delayed_node = NULL;
1318 	struct btrfs_root *root;
1319 	struct btrfs_block_rsv *block_rsv;
1320 	int total_done = 0;
1321 
1322 	async_work = container_of(work, struct btrfs_async_delayed_work, work);
1323 	delayed_root = async_work->delayed_root;
1324 
1325 	path = btrfs_alloc_path();
1326 	if (!path)
1327 		goto out;
1328 
1329 	do {
1330 		if (atomic_read(&delayed_root->items) <
1331 		    BTRFS_DELAYED_BACKGROUND / 2)
1332 			break;
1333 
1334 		delayed_node = btrfs_first_prepared_delayed_node(delayed_root);
1335 		if (!delayed_node)
1336 			break;
1337 
1338 		path->leave_spinning = 1;
1339 		root = delayed_node->root;
1340 
1341 		trans = btrfs_join_transaction(root);
1342 		if (IS_ERR(trans)) {
1343 			btrfs_release_path(path);
1344 			btrfs_release_prepared_delayed_node(delayed_node);
1345 			total_done++;
1346 			continue;
1347 		}
1348 
1349 		block_rsv = trans->block_rsv;
1350 		trans->block_rsv = &root->fs_info->delayed_block_rsv;
1351 
1352 		__btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1353 
1354 		trans->block_rsv = block_rsv;
1355 		btrfs_end_transaction(trans);
1356 		btrfs_btree_balance_dirty_nodelay(root->fs_info);
1357 
1358 		btrfs_release_path(path);
1359 		btrfs_release_prepared_delayed_node(delayed_node);
1360 		total_done++;
1361 
1362 	} while ((async_work->nr == 0 && total_done < BTRFS_DELAYED_WRITEBACK)
1363 		 || total_done < async_work->nr);
1364 
1365 	btrfs_free_path(path);
1366 out:
1367 	wake_up(&delayed_root->wait);
1368 	kfree(async_work);
1369 }
1370 
1371 
1372 static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root,
1373 				     struct btrfs_fs_info *fs_info, int nr)
1374 {
1375 	struct btrfs_async_delayed_work *async_work;
1376 
1377 	async_work = kmalloc(sizeof(*async_work), GFP_NOFS);
1378 	if (!async_work)
1379 		return -ENOMEM;
1380 
1381 	async_work->delayed_root = delayed_root;
1382 	btrfs_init_work(&async_work->work, btrfs_async_run_delayed_root, NULL,
1383 			NULL);
1384 	async_work->nr = nr;
1385 
1386 	btrfs_queue_work(fs_info->delayed_workers, &async_work->work);
1387 	return 0;
1388 }
1389 
1390 void btrfs_assert_delayed_root_empty(struct btrfs_fs_info *fs_info)
1391 {
1392 	WARN_ON(btrfs_first_delayed_node(fs_info->delayed_root));
1393 }
1394 
1395 static int could_end_wait(struct btrfs_delayed_root *delayed_root, int seq)
1396 {
1397 	int val = atomic_read(&delayed_root->items_seq);
1398 
1399 	if (val < seq || val >= seq + BTRFS_DELAYED_BATCH)
1400 		return 1;
1401 
1402 	if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1403 		return 1;
1404 
1405 	return 0;
1406 }
1407 
1408 void btrfs_balance_delayed_items(struct btrfs_fs_info *fs_info)
1409 {
1410 	struct btrfs_delayed_root *delayed_root = fs_info->delayed_root;
1411 
1412 	if ((atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND) ||
1413 		btrfs_workqueue_normal_congested(fs_info->delayed_workers))
1414 		return;
1415 
1416 	if (atomic_read(&delayed_root->items) >= BTRFS_DELAYED_WRITEBACK) {
1417 		int seq;
1418 		int ret;
1419 
1420 		seq = atomic_read(&delayed_root->items_seq);
1421 
1422 		ret = btrfs_wq_run_delayed_node(delayed_root, fs_info, 0);
1423 		if (ret)
1424 			return;
1425 
1426 		wait_event_interruptible(delayed_root->wait,
1427 					 could_end_wait(delayed_root, seq));
1428 		return;
1429 	}
1430 
1431 	btrfs_wq_run_delayed_node(delayed_root, fs_info, BTRFS_DELAYED_BATCH);
1432 }
1433 
1434 /* Will return 0 or -ENOMEM */
1435 int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans,
1436 				   const char *name, int name_len,
1437 				   struct btrfs_inode *dir,
1438 				   struct btrfs_disk_key *disk_key, u8 type,
1439 				   u64 index)
1440 {
1441 	struct btrfs_delayed_node *delayed_node;
1442 	struct btrfs_delayed_item *delayed_item;
1443 	struct btrfs_dir_item *dir_item;
1444 	int ret;
1445 
1446 	delayed_node = btrfs_get_or_create_delayed_node(dir);
1447 	if (IS_ERR(delayed_node))
1448 		return PTR_ERR(delayed_node);
1449 
1450 	delayed_item = btrfs_alloc_delayed_item(sizeof(*dir_item) + name_len);
1451 	if (!delayed_item) {
1452 		ret = -ENOMEM;
1453 		goto release_node;
1454 	}
1455 
1456 	delayed_item->key.objectid = btrfs_ino(dir);
1457 	delayed_item->key.type = BTRFS_DIR_INDEX_KEY;
1458 	delayed_item->key.offset = index;
1459 
1460 	dir_item = (struct btrfs_dir_item *)delayed_item->data;
1461 	dir_item->location = *disk_key;
1462 	btrfs_set_stack_dir_transid(dir_item, trans->transid);
1463 	btrfs_set_stack_dir_data_len(dir_item, 0);
1464 	btrfs_set_stack_dir_name_len(dir_item, name_len);
1465 	btrfs_set_stack_dir_type(dir_item, type);
1466 	memcpy((char *)(dir_item + 1), name, name_len);
1467 
1468 	ret = btrfs_delayed_item_reserve_metadata(trans, dir->root, delayed_item);
1469 	/*
1470 	 * we have reserved enough space when we start a new transaction,
1471 	 * so reserving metadata failure is impossible
1472 	 */
1473 	BUG_ON(ret);
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(trans->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->root_key.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(node->root, 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_inode *dir, u64 index)
1512 {
1513 	struct btrfs_delayed_node *node;
1514 	struct btrfs_delayed_item *item;
1515 	struct btrfs_key item_key;
1516 	int ret;
1517 
1518 	node = btrfs_get_or_create_delayed_node(dir);
1519 	if (IS_ERR(node))
1520 		return PTR_ERR(node);
1521 
1522 	item_key.objectid = btrfs_ino(dir);
1523 	item_key.type = BTRFS_DIR_INDEX_KEY;
1524 	item_key.offset = index;
1525 
1526 	ret = btrfs_delete_delayed_insertion_item(trans->fs_info, node,
1527 						  &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, dir->root, item);
1540 	/*
1541 	 * we have reserved enough space when we start a new transaction,
1542 	 * so reserving metadata failure is impossible.
1543 	 */
1544 	if (ret < 0) {
1545 		btrfs_err(trans->fs_info,
1546 "metadata reservation failed for delayed dir item deltiona, should have been reserved");
1547 		btrfs_release_delayed_item(item);
1548 		goto end;
1549 	}
1550 
1551 	mutex_lock(&node->mutex);
1552 	ret = __btrfs_add_delayed_deletion_item(node, item);
1553 	if (unlikely(ret)) {
1554 		btrfs_err(trans->fs_info,
1555 			  "err add delayed dir index item(index: %llu) into the deletion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
1556 			  index, node->root->root_key.objectid,
1557 			  node->inode_id, ret);
1558 		btrfs_delayed_item_release_metadata(dir->root, item);
1559 		btrfs_release_delayed_item(item);
1560 	}
1561 	mutex_unlock(&node->mutex);
1562 end:
1563 	btrfs_release_delayed_node(node);
1564 	return ret;
1565 }
1566 
1567 int btrfs_inode_delayed_dir_index_count(struct btrfs_inode *inode)
1568 {
1569 	struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1570 
1571 	if (!delayed_node)
1572 		return -ENOENT;
1573 
1574 	/*
1575 	 * Since we have held i_mutex of this directory, it is impossible that
1576 	 * a new directory index is added into the delayed node and index_cnt
1577 	 * is updated now. So we needn't lock the delayed node.
1578 	 */
1579 	if (!delayed_node->index_cnt) {
1580 		btrfs_release_delayed_node(delayed_node);
1581 		return -EINVAL;
1582 	}
1583 
1584 	inode->index_cnt = delayed_node->index_cnt;
1585 	btrfs_release_delayed_node(delayed_node);
1586 	return 0;
1587 }
1588 
1589 bool btrfs_readdir_get_delayed_items(struct inode *inode,
1590 				     struct list_head *ins_list,
1591 				     struct list_head *del_list)
1592 {
1593 	struct btrfs_delayed_node *delayed_node;
1594 	struct btrfs_delayed_item *item;
1595 
1596 	delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
1597 	if (!delayed_node)
1598 		return false;
1599 
1600 	/*
1601 	 * We can only do one readdir with delayed items at a time because of
1602 	 * item->readdir_list.
1603 	 */
1604 	inode_unlock_shared(inode);
1605 	inode_lock(inode);
1606 
1607 	mutex_lock(&delayed_node->mutex);
1608 	item = __btrfs_first_delayed_insertion_item(delayed_node);
1609 	while (item) {
1610 		refcount_inc(&item->refs);
1611 		list_add_tail(&item->readdir_list, ins_list);
1612 		item = __btrfs_next_delayed_item(item);
1613 	}
1614 
1615 	item = __btrfs_first_delayed_deletion_item(delayed_node);
1616 	while (item) {
1617 		refcount_inc(&item->refs);
1618 		list_add_tail(&item->readdir_list, del_list);
1619 		item = __btrfs_next_delayed_item(item);
1620 	}
1621 	mutex_unlock(&delayed_node->mutex);
1622 	/*
1623 	 * This delayed node is still cached in the btrfs inode, so refs
1624 	 * must be > 1 now, and we needn't check it is going to be freed
1625 	 * or not.
1626 	 *
1627 	 * Besides that, this function is used to read dir, we do not
1628 	 * insert/delete delayed items in this period. So we also needn't
1629 	 * requeue or dequeue this delayed node.
1630 	 */
1631 	refcount_dec(&delayed_node->refs);
1632 
1633 	return true;
1634 }
1635 
1636 void btrfs_readdir_put_delayed_items(struct inode *inode,
1637 				     struct list_head *ins_list,
1638 				     struct list_head *del_list)
1639 {
1640 	struct btrfs_delayed_item *curr, *next;
1641 
1642 	list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1643 		list_del(&curr->readdir_list);
1644 		if (refcount_dec_and_test(&curr->refs))
1645 			kfree(curr);
1646 	}
1647 
1648 	list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1649 		list_del(&curr->readdir_list);
1650 		if (refcount_dec_and_test(&curr->refs))
1651 			kfree(curr);
1652 	}
1653 
1654 	/*
1655 	 * The VFS is going to do up_read(), so we need to downgrade back to a
1656 	 * read lock.
1657 	 */
1658 	downgrade_write(&inode->i_rwsem);
1659 }
1660 
1661 int btrfs_should_delete_dir_index(struct list_head *del_list,
1662 				  u64 index)
1663 {
1664 	struct btrfs_delayed_item *curr;
1665 	int ret = 0;
1666 
1667 	list_for_each_entry(curr, del_list, readdir_list) {
1668 		if (curr->key.offset > index)
1669 			break;
1670 		if (curr->key.offset == index) {
1671 			ret = 1;
1672 			break;
1673 		}
1674 	}
1675 	return ret;
1676 }
1677 
1678 /*
1679  * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree
1680  *
1681  */
1682 int btrfs_readdir_delayed_dir_index(struct dir_context *ctx,
1683 				    struct list_head *ins_list)
1684 {
1685 	struct btrfs_dir_item *di;
1686 	struct btrfs_delayed_item *curr, *next;
1687 	struct btrfs_key location;
1688 	char *name;
1689 	int name_len;
1690 	int over = 0;
1691 	unsigned char d_type;
1692 
1693 	if (list_empty(ins_list))
1694 		return 0;
1695 
1696 	/*
1697 	 * Changing the data of the delayed item is impossible. So
1698 	 * we needn't lock them. And we have held i_mutex of the
1699 	 * directory, nobody can delete any directory indexes now.
1700 	 */
1701 	list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1702 		list_del(&curr->readdir_list);
1703 
1704 		if (curr->key.offset < ctx->pos) {
1705 			if (refcount_dec_and_test(&curr->refs))
1706 				kfree(curr);
1707 			continue;
1708 		}
1709 
1710 		ctx->pos = curr->key.offset;
1711 
1712 		di = (struct btrfs_dir_item *)curr->data;
1713 		name = (char *)(di + 1);
1714 		name_len = btrfs_stack_dir_name_len(di);
1715 
1716 		d_type = fs_ftype_to_dtype(di->type);
1717 		btrfs_disk_key_to_cpu(&location, &di->location);
1718 
1719 		over = !dir_emit(ctx, name, name_len,
1720 			       location.objectid, d_type);
1721 
1722 		if (refcount_dec_and_test(&curr->refs))
1723 			kfree(curr);
1724 
1725 		if (over)
1726 			return 1;
1727 		ctx->pos++;
1728 	}
1729 	return 0;
1730 }
1731 
1732 static void fill_stack_inode_item(struct btrfs_trans_handle *trans,
1733 				  struct btrfs_inode_item *inode_item,
1734 				  struct inode *inode)
1735 {
1736 	btrfs_set_stack_inode_uid(inode_item, i_uid_read(inode));
1737 	btrfs_set_stack_inode_gid(inode_item, i_gid_read(inode));
1738 	btrfs_set_stack_inode_size(inode_item, BTRFS_I(inode)->disk_i_size);
1739 	btrfs_set_stack_inode_mode(inode_item, inode->i_mode);
1740 	btrfs_set_stack_inode_nlink(inode_item, inode->i_nlink);
1741 	btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(inode));
1742 	btrfs_set_stack_inode_generation(inode_item,
1743 					 BTRFS_I(inode)->generation);
1744 	btrfs_set_stack_inode_sequence(inode_item,
1745 				       inode_peek_iversion(inode));
1746 	btrfs_set_stack_inode_transid(inode_item, trans->transid);
1747 	btrfs_set_stack_inode_rdev(inode_item, inode->i_rdev);
1748 	btrfs_set_stack_inode_flags(inode_item, BTRFS_I(inode)->flags);
1749 	btrfs_set_stack_inode_block_group(inode_item, 0);
1750 
1751 	btrfs_set_stack_timespec_sec(&inode_item->atime,
1752 				     inode->i_atime.tv_sec);
1753 	btrfs_set_stack_timespec_nsec(&inode_item->atime,
1754 				      inode->i_atime.tv_nsec);
1755 
1756 	btrfs_set_stack_timespec_sec(&inode_item->mtime,
1757 				     inode->i_mtime.tv_sec);
1758 	btrfs_set_stack_timespec_nsec(&inode_item->mtime,
1759 				      inode->i_mtime.tv_nsec);
1760 
1761 	btrfs_set_stack_timespec_sec(&inode_item->ctime,
1762 				     inode->i_ctime.tv_sec);
1763 	btrfs_set_stack_timespec_nsec(&inode_item->ctime,
1764 				      inode->i_ctime.tv_nsec);
1765 
1766 	btrfs_set_stack_timespec_sec(&inode_item->otime,
1767 				     BTRFS_I(inode)->i_otime.tv_sec);
1768 	btrfs_set_stack_timespec_nsec(&inode_item->otime,
1769 				     BTRFS_I(inode)->i_otime.tv_nsec);
1770 }
1771 
1772 int btrfs_fill_inode(struct inode *inode, u32 *rdev)
1773 {
1774 	struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
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 	btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
1795 			round_up(i_size_read(inode), fs_info->sectorsize));
1796 	inode->i_mode = btrfs_stack_inode_mode(inode_item);
1797 	set_nlink(inode, btrfs_stack_inode_nlink(inode_item));
1798 	inode_set_bytes(inode, btrfs_stack_inode_nbytes(inode_item));
1799 	BTRFS_I(inode)->generation = btrfs_stack_inode_generation(inode_item);
1800         BTRFS_I(inode)->last_trans = btrfs_stack_inode_transid(inode_item);
1801 
1802 	inode_set_iversion_queried(inode,
1803 				   btrfs_stack_inode_sequence(inode_item));
1804 	inode->i_rdev = 0;
1805 	*rdev = btrfs_stack_inode_rdev(inode_item);
1806 	BTRFS_I(inode)->flags = btrfs_stack_inode_flags(inode_item);
1807 
1808 	inode->i_atime.tv_sec = btrfs_stack_timespec_sec(&inode_item->atime);
1809 	inode->i_atime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->atime);
1810 
1811 	inode->i_mtime.tv_sec = btrfs_stack_timespec_sec(&inode_item->mtime);
1812 	inode->i_mtime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->mtime);
1813 
1814 	inode->i_ctime.tv_sec = btrfs_stack_timespec_sec(&inode_item->ctime);
1815 	inode->i_ctime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->ctime);
1816 
1817 	BTRFS_I(inode)->i_otime.tv_sec =
1818 		btrfs_stack_timespec_sec(&inode_item->otime);
1819 	BTRFS_I(inode)->i_otime.tv_nsec =
1820 		btrfs_stack_timespec_nsec(&inode_item->otime);
1821 
1822 	inode->i_generation = BTRFS_I(inode)->generation;
1823 	BTRFS_I(inode)->index_cnt = (u64)-1;
1824 
1825 	mutex_unlock(&delayed_node->mutex);
1826 	btrfs_release_delayed_node(delayed_node);
1827 	return 0;
1828 }
1829 
1830 int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans,
1831 			       struct btrfs_root *root, struct inode *inode)
1832 {
1833 	struct btrfs_delayed_node *delayed_node;
1834 	int ret = 0;
1835 
1836 	delayed_node = btrfs_get_or_create_delayed_node(BTRFS_I(inode));
1837 	if (IS_ERR(delayed_node))
1838 		return PTR_ERR(delayed_node);
1839 
1840 	mutex_lock(&delayed_node->mutex);
1841 	if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1842 		fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1843 		goto release_node;
1844 	}
1845 
1846 	ret = btrfs_delayed_inode_reserve_metadata(trans, root, BTRFS_I(inode),
1847 						   delayed_node);
1848 	if (ret)
1849 		goto release_node;
1850 
1851 	fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1852 	set_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
1853 	delayed_node->count++;
1854 	atomic_inc(&root->fs_info->delayed_root->items);
1855 release_node:
1856 	mutex_unlock(&delayed_node->mutex);
1857 	btrfs_release_delayed_node(delayed_node);
1858 	return ret;
1859 }
1860 
1861 int btrfs_delayed_delete_inode_ref(struct btrfs_inode *inode)
1862 {
1863 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
1864 	struct btrfs_delayed_node *delayed_node;
1865 
1866 	/*
1867 	 * we don't do delayed inode updates during log recovery because it
1868 	 * leads to enospc problems.  This means we also can't do
1869 	 * delayed inode refs
1870 	 */
1871 	if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
1872 		return -EAGAIN;
1873 
1874 	delayed_node = btrfs_get_or_create_delayed_node(inode);
1875 	if (IS_ERR(delayed_node))
1876 		return PTR_ERR(delayed_node);
1877 
1878 	/*
1879 	 * We don't reserve space for inode ref deletion is because:
1880 	 * - We ONLY do async inode ref deletion for the inode who has only
1881 	 *   one link(i_nlink == 1), it means there is only one inode ref.
1882 	 *   And in most case, the inode ref and the inode item are in the
1883 	 *   same leaf, and we will deal with them at the same time.
1884 	 *   Since we are sure we will reserve the space for the inode item,
1885 	 *   it is unnecessary to reserve space for inode ref deletion.
1886 	 * - If the inode ref and the inode item are not in the same leaf,
1887 	 *   We also needn't worry about enospc problem, because we reserve
1888 	 *   much more space for the inode update than it needs.
1889 	 * - At the worst, we can steal some space from the global reservation.
1890 	 *   It is very rare.
1891 	 */
1892 	mutex_lock(&delayed_node->mutex);
1893 	if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1894 		goto release_node;
1895 
1896 	set_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
1897 	delayed_node->count++;
1898 	atomic_inc(&fs_info->delayed_root->items);
1899 release_node:
1900 	mutex_unlock(&delayed_node->mutex);
1901 	btrfs_release_delayed_node(delayed_node);
1902 	return 0;
1903 }
1904 
1905 static void __btrfs_kill_delayed_node(struct btrfs_delayed_node *delayed_node)
1906 {
1907 	struct btrfs_root *root = delayed_node->root;
1908 	struct btrfs_fs_info *fs_info = root->fs_info;
1909 	struct btrfs_delayed_item *curr_item, *prev_item;
1910 
1911 	mutex_lock(&delayed_node->mutex);
1912 	curr_item = __btrfs_first_delayed_insertion_item(delayed_node);
1913 	while (curr_item) {
1914 		btrfs_delayed_item_release_metadata(root, curr_item);
1915 		prev_item = curr_item;
1916 		curr_item = __btrfs_next_delayed_item(prev_item);
1917 		btrfs_release_delayed_item(prev_item);
1918 	}
1919 
1920 	curr_item = __btrfs_first_delayed_deletion_item(delayed_node);
1921 	while (curr_item) {
1922 		btrfs_delayed_item_release_metadata(root, curr_item);
1923 		prev_item = curr_item;
1924 		curr_item = __btrfs_next_delayed_item(prev_item);
1925 		btrfs_release_delayed_item(prev_item);
1926 	}
1927 
1928 	if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1929 		btrfs_release_delayed_iref(delayed_node);
1930 
1931 	if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1932 		btrfs_delayed_inode_release_metadata(fs_info, delayed_node, false);
1933 		btrfs_release_delayed_inode(delayed_node);
1934 	}
1935 	mutex_unlock(&delayed_node->mutex);
1936 }
1937 
1938 void btrfs_kill_delayed_inode_items(struct btrfs_inode *inode)
1939 {
1940 	struct btrfs_delayed_node *delayed_node;
1941 
1942 	delayed_node = btrfs_get_delayed_node(inode);
1943 	if (!delayed_node)
1944 		return;
1945 
1946 	__btrfs_kill_delayed_node(delayed_node);
1947 	btrfs_release_delayed_node(delayed_node);
1948 }
1949 
1950 void btrfs_kill_all_delayed_nodes(struct btrfs_root *root)
1951 {
1952 	u64 inode_id = 0;
1953 	struct btrfs_delayed_node *delayed_nodes[8];
1954 	int i, n;
1955 
1956 	while (1) {
1957 		spin_lock(&root->inode_lock);
1958 		n = radix_tree_gang_lookup(&root->delayed_nodes_tree,
1959 					   (void **)delayed_nodes, inode_id,
1960 					   ARRAY_SIZE(delayed_nodes));
1961 		if (!n) {
1962 			spin_unlock(&root->inode_lock);
1963 			break;
1964 		}
1965 
1966 		inode_id = delayed_nodes[n - 1]->inode_id + 1;
1967 		for (i = 0; i < n; i++) {
1968 			/*
1969 			 * Don't increase refs in case the node is dead and
1970 			 * about to be removed from the tree in the loop below
1971 			 */
1972 			if (!refcount_inc_not_zero(&delayed_nodes[i]->refs))
1973 				delayed_nodes[i] = NULL;
1974 		}
1975 		spin_unlock(&root->inode_lock);
1976 
1977 		for (i = 0; i < n; i++) {
1978 			if (!delayed_nodes[i])
1979 				continue;
1980 			__btrfs_kill_delayed_node(delayed_nodes[i]);
1981 			btrfs_release_delayed_node(delayed_nodes[i]);
1982 		}
1983 	}
1984 }
1985 
1986 void btrfs_destroy_delayed_inodes(struct btrfs_fs_info *fs_info)
1987 {
1988 	struct btrfs_delayed_node *curr_node, *prev_node;
1989 
1990 	curr_node = btrfs_first_delayed_node(fs_info->delayed_root);
1991 	while (curr_node) {
1992 		__btrfs_kill_delayed_node(curr_node);
1993 
1994 		prev_node = curr_node;
1995 		curr_node = btrfs_next_delayed_node(curr_node);
1996 		btrfs_release_delayed_node(prev_node);
1997 	}
1998 }
1999 
2000