Loading...
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
25#define BTRFS_DELAYED_WRITEBACK 400
26#define BTRFS_DELAYED_BACKGROUND 100
27
28static struct kmem_cache *delayed_node_cache;
29
30int __init btrfs_delayed_inode_init(void)
31{
32 delayed_node_cache = kmem_cache_create("delayed_node",
33 sizeof(struct btrfs_delayed_node),
34 0,
35 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
36 NULL);
37 if (!delayed_node_cache)
38 return -ENOMEM;
39 return 0;
40}
41
42void btrfs_delayed_inode_exit(void)
43{
44 if (delayed_node_cache)
45 kmem_cache_destroy(delayed_node_cache);
46}
47
48static inline void btrfs_init_delayed_node(
49 struct btrfs_delayed_node *delayed_node,
50 struct btrfs_root *root, u64 inode_id)
51{
52 delayed_node->root = root;
53 delayed_node->inode_id = inode_id;
54 atomic_set(&delayed_node->refs, 0);
55 delayed_node->count = 0;
56 delayed_node->in_list = 0;
57 delayed_node->inode_dirty = 0;
58 delayed_node->ins_root = RB_ROOT;
59 delayed_node->del_root = RB_ROOT;
60 mutex_init(&delayed_node->mutex);
61 delayed_node->index_cnt = 0;
62 INIT_LIST_HEAD(&delayed_node->n_list);
63 INIT_LIST_HEAD(&delayed_node->p_list);
64 delayed_node->bytes_reserved = 0;
65}
66
67static inline int btrfs_is_continuous_delayed_item(
68 struct btrfs_delayed_item *item1,
69 struct btrfs_delayed_item *item2)
70{
71 if (item1->key.type == BTRFS_DIR_INDEX_KEY &&
72 item1->key.objectid == item2->key.objectid &&
73 item1->key.type == item2->key.type &&
74 item1->key.offset + 1 == item2->key.offset)
75 return 1;
76 return 0;
77}
78
79static inline struct btrfs_delayed_root *btrfs_get_delayed_root(
80 struct btrfs_root *root)
81{
82 return root->fs_info->delayed_root;
83}
84
85static struct btrfs_delayed_node *btrfs_get_delayed_node(struct inode *inode)
86{
87 struct btrfs_inode *btrfs_inode = BTRFS_I(inode);
88 struct btrfs_root *root = btrfs_inode->root;
89 u64 ino = btrfs_ino(inode);
90 struct btrfs_delayed_node *node;
91
92 node = ACCESS_ONCE(btrfs_inode->delayed_node);
93 if (node) {
94 atomic_inc(&node->refs);
95 return node;
96 }
97
98 spin_lock(&root->inode_lock);
99 node = radix_tree_lookup(&root->delayed_nodes_tree, ino);
100 if (node) {
101 if (btrfs_inode->delayed_node) {
102 atomic_inc(&node->refs); /* can be accessed */
103 BUG_ON(btrfs_inode->delayed_node != node);
104 spin_unlock(&root->inode_lock);
105 return node;
106 }
107 btrfs_inode->delayed_node = node;
108 atomic_inc(&node->refs); /* can be accessed */
109 atomic_inc(&node->refs); /* cached in the inode */
110 spin_unlock(&root->inode_lock);
111 return node;
112 }
113 spin_unlock(&root->inode_lock);
114
115 return NULL;
116}
117
118static struct btrfs_delayed_node *btrfs_get_or_create_delayed_node(
119 struct inode *inode)
120{
121 struct btrfs_delayed_node *node;
122 struct btrfs_inode *btrfs_inode = BTRFS_I(inode);
123 struct btrfs_root *root = btrfs_inode->root;
124 u64 ino = btrfs_ino(inode);
125 int ret;
126
127again:
128 node = btrfs_get_delayed_node(inode);
129 if (node)
130 return node;
131
132 node = kmem_cache_alloc(delayed_node_cache, GFP_NOFS);
133 if (!node)
134 return ERR_PTR(-ENOMEM);
135 btrfs_init_delayed_node(node, root, ino);
136
137 atomic_inc(&node->refs); /* cached in the btrfs inode */
138 atomic_inc(&node->refs); /* can be accessed */
139
140 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
141 if (ret) {
142 kmem_cache_free(delayed_node_cache, node);
143 return ERR_PTR(ret);
144 }
145
146 spin_lock(&root->inode_lock);
147 ret = radix_tree_insert(&root->delayed_nodes_tree, ino, node);
148 if (ret == -EEXIST) {
149 kmem_cache_free(delayed_node_cache, node);
150 spin_unlock(&root->inode_lock);
151 radix_tree_preload_end();
152 goto again;
153 }
154 btrfs_inode->delayed_node = node;
155 spin_unlock(&root->inode_lock);
156 radix_tree_preload_end();
157
158 return node;
159}
160
161/*
162 * Call it when holding delayed_node->mutex
163 *
164 * If mod = 1, add this node into the prepared list.
165 */
166static void btrfs_queue_delayed_node(struct btrfs_delayed_root *root,
167 struct btrfs_delayed_node *node,
168 int mod)
169{
170 spin_lock(&root->lock);
171 if (node->in_list) {
172 if (!list_empty(&node->p_list))
173 list_move_tail(&node->p_list, &root->prepare_list);
174 else if (mod)
175 list_add_tail(&node->p_list, &root->prepare_list);
176 } else {
177 list_add_tail(&node->n_list, &root->node_list);
178 list_add_tail(&node->p_list, &root->prepare_list);
179 atomic_inc(&node->refs); /* inserted into list */
180 root->nodes++;
181 node->in_list = 1;
182 }
183 spin_unlock(&root->lock);
184}
185
186/* Call it when holding delayed_node->mutex */
187static void btrfs_dequeue_delayed_node(struct btrfs_delayed_root *root,
188 struct btrfs_delayed_node *node)
189{
190 spin_lock(&root->lock);
191 if (node->in_list) {
192 root->nodes--;
193 atomic_dec(&node->refs); /* not in the list */
194 list_del_init(&node->n_list);
195 if (!list_empty(&node->p_list))
196 list_del_init(&node->p_list);
197 node->in_list = 0;
198 }
199 spin_unlock(&root->lock);
200}
201
202struct btrfs_delayed_node *btrfs_first_delayed_node(
203 struct btrfs_delayed_root *delayed_root)
204{
205 struct list_head *p;
206 struct btrfs_delayed_node *node = NULL;
207
208 spin_lock(&delayed_root->lock);
209 if (list_empty(&delayed_root->node_list))
210 goto out;
211
212 p = delayed_root->node_list.next;
213 node = list_entry(p, struct btrfs_delayed_node, n_list);
214 atomic_inc(&node->refs);
215out:
216 spin_unlock(&delayed_root->lock);
217
218 return node;
219}
220
221struct btrfs_delayed_node *btrfs_next_delayed_node(
222 struct btrfs_delayed_node *node)
223{
224 struct btrfs_delayed_root *delayed_root;
225 struct list_head *p;
226 struct btrfs_delayed_node *next = NULL;
227
228 delayed_root = node->root->fs_info->delayed_root;
229 spin_lock(&delayed_root->lock);
230 if (!node->in_list) { /* not in the list */
231 if (list_empty(&delayed_root->node_list))
232 goto out;
233 p = delayed_root->node_list.next;
234 } else if (list_is_last(&node->n_list, &delayed_root->node_list))
235 goto out;
236 else
237 p = node->n_list.next;
238
239 next = list_entry(p, struct btrfs_delayed_node, n_list);
240 atomic_inc(&next->refs);
241out:
242 spin_unlock(&delayed_root->lock);
243
244 return next;
245}
246
247static void __btrfs_release_delayed_node(
248 struct btrfs_delayed_node *delayed_node,
249 int mod)
250{
251 struct btrfs_delayed_root *delayed_root;
252
253 if (!delayed_node)
254 return;
255
256 delayed_root = delayed_node->root->fs_info->delayed_root;
257
258 mutex_lock(&delayed_node->mutex);
259 if (delayed_node->count)
260 btrfs_queue_delayed_node(delayed_root, delayed_node, mod);
261 else
262 btrfs_dequeue_delayed_node(delayed_root, delayed_node);
263 mutex_unlock(&delayed_node->mutex);
264
265 if (atomic_dec_and_test(&delayed_node->refs)) {
266 struct btrfs_root *root = delayed_node->root;
267 spin_lock(&root->inode_lock);
268 if (atomic_read(&delayed_node->refs) == 0) {
269 radix_tree_delete(&root->delayed_nodes_tree,
270 delayed_node->inode_id);
271 kmem_cache_free(delayed_node_cache, delayed_node);
272 }
273 spin_unlock(&root->inode_lock);
274 }
275}
276
277static inline void btrfs_release_delayed_node(struct btrfs_delayed_node *node)
278{
279 __btrfs_release_delayed_node(node, 0);
280}
281
282struct btrfs_delayed_node *btrfs_first_prepared_delayed_node(
283 struct btrfs_delayed_root *delayed_root)
284{
285 struct list_head *p;
286 struct btrfs_delayed_node *node = NULL;
287
288 spin_lock(&delayed_root->lock);
289 if (list_empty(&delayed_root->prepare_list))
290 goto out;
291
292 p = delayed_root->prepare_list.next;
293 list_del_init(p);
294 node = list_entry(p, struct btrfs_delayed_node, p_list);
295 atomic_inc(&node->refs);
296out:
297 spin_unlock(&delayed_root->lock);
298
299 return node;
300}
301
302static inline void btrfs_release_prepared_delayed_node(
303 struct btrfs_delayed_node *node)
304{
305 __btrfs_release_delayed_node(node, 1);
306}
307
308struct btrfs_delayed_item *btrfs_alloc_delayed_item(u32 data_len)
309{
310 struct btrfs_delayed_item *item;
311 item = kmalloc(sizeof(*item) + data_len, GFP_NOFS);
312 if (item) {
313 item->data_len = data_len;
314 item->ins_or_del = 0;
315 item->bytes_reserved = 0;
316 item->delayed_node = NULL;
317 atomic_set(&item->refs, 1);
318 }
319 return item;
320}
321
322/*
323 * __btrfs_lookup_delayed_item - look up the delayed item by key
324 * @delayed_node: pointer to the delayed node
325 * @key: the key to look up
326 * @prev: used to store the prev item if the right item isn't found
327 * @next: used to store the next item if the right item isn't found
328 *
329 * Note: if we don't find the right item, we will return the prev item and
330 * the next item.
331 */
332static struct btrfs_delayed_item *__btrfs_lookup_delayed_item(
333 struct rb_root *root,
334 struct btrfs_key *key,
335 struct btrfs_delayed_item **prev,
336 struct btrfs_delayed_item **next)
337{
338 struct rb_node *node, *prev_node = NULL;
339 struct btrfs_delayed_item *delayed_item = NULL;
340 int ret = 0;
341
342 node = root->rb_node;
343
344 while (node) {
345 delayed_item = rb_entry(node, struct btrfs_delayed_item,
346 rb_node);
347 prev_node = node;
348 ret = btrfs_comp_cpu_keys(&delayed_item->key, key);
349 if (ret < 0)
350 node = node->rb_right;
351 else if (ret > 0)
352 node = node->rb_left;
353 else
354 return delayed_item;
355 }
356
357 if (prev) {
358 if (!prev_node)
359 *prev = NULL;
360 else if (ret < 0)
361 *prev = delayed_item;
362 else if ((node = rb_prev(prev_node)) != NULL) {
363 *prev = rb_entry(node, struct btrfs_delayed_item,
364 rb_node);
365 } else
366 *prev = NULL;
367 }
368
369 if (next) {
370 if (!prev_node)
371 *next = NULL;
372 else if (ret > 0)
373 *next = delayed_item;
374 else if ((node = rb_next(prev_node)) != NULL) {
375 *next = rb_entry(node, struct btrfs_delayed_item,
376 rb_node);
377 } else
378 *next = NULL;
379 }
380 return NULL;
381}
382
383struct btrfs_delayed_item *__btrfs_lookup_delayed_insertion_item(
384 struct btrfs_delayed_node *delayed_node,
385 struct btrfs_key *key)
386{
387 struct btrfs_delayed_item *item;
388
389 item = __btrfs_lookup_delayed_item(&delayed_node->ins_root, key,
390 NULL, NULL);
391 return item;
392}
393
394struct btrfs_delayed_item *__btrfs_lookup_delayed_deletion_item(
395 struct btrfs_delayed_node *delayed_node,
396 struct btrfs_key *key)
397{
398 struct btrfs_delayed_item *item;
399
400 item = __btrfs_lookup_delayed_item(&delayed_node->del_root, key,
401 NULL, NULL);
402 return item;
403}
404
405struct btrfs_delayed_item *__btrfs_search_delayed_insertion_item(
406 struct btrfs_delayed_node *delayed_node,
407 struct btrfs_key *key)
408{
409 struct btrfs_delayed_item *item, *next;
410
411 item = __btrfs_lookup_delayed_item(&delayed_node->ins_root, key,
412 NULL, &next);
413 if (!item)
414 item = next;
415
416 return item;
417}
418
419struct btrfs_delayed_item *__btrfs_search_delayed_deletion_item(
420 struct btrfs_delayed_node *delayed_node,
421 struct btrfs_key *key)
422{
423 struct btrfs_delayed_item *item, *next;
424
425 item = __btrfs_lookup_delayed_item(&delayed_node->del_root, key,
426 NULL, &next);
427 if (!item)
428 item = next;
429
430 return item;
431}
432
433static int __btrfs_add_delayed_item(struct btrfs_delayed_node *delayed_node,
434 struct btrfs_delayed_item *ins,
435 int action)
436{
437 struct rb_node **p, *node;
438 struct rb_node *parent_node = NULL;
439 struct rb_root *root;
440 struct btrfs_delayed_item *item;
441 int cmp;
442
443 if (action == BTRFS_DELAYED_INSERTION_ITEM)
444 root = &delayed_node->ins_root;
445 else if (action == BTRFS_DELAYED_DELETION_ITEM)
446 root = &delayed_node->del_root;
447 else
448 BUG();
449 p = &root->rb_node;
450 node = &ins->rb_node;
451
452 while (*p) {
453 parent_node = *p;
454 item = rb_entry(parent_node, struct btrfs_delayed_item,
455 rb_node);
456
457 cmp = btrfs_comp_cpu_keys(&item->key, &ins->key);
458 if (cmp < 0)
459 p = &(*p)->rb_right;
460 else if (cmp > 0)
461 p = &(*p)->rb_left;
462 else
463 return -EEXIST;
464 }
465
466 rb_link_node(node, parent_node, p);
467 rb_insert_color(node, root);
468 ins->delayed_node = delayed_node;
469 ins->ins_or_del = action;
470
471 if (ins->key.type == BTRFS_DIR_INDEX_KEY &&
472 action == BTRFS_DELAYED_INSERTION_ITEM &&
473 ins->key.offset >= delayed_node->index_cnt)
474 delayed_node->index_cnt = ins->key.offset + 1;
475
476 delayed_node->count++;
477 atomic_inc(&delayed_node->root->fs_info->delayed_root->items);
478 return 0;
479}
480
481static int __btrfs_add_delayed_insertion_item(struct btrfs_delayed_node *node,
482 struct btrfs_delayed_item *item)
483{
484 return __btrfs_add_delayed_item(node, item,
485 BTRFS_DELAYED_INSERTION_ITEM);
486}
487
488static int __btrfs_add_delayed_deletion_item(struct btrfs_delayed_node *node,
489 struct btrfs_delayed_item *item)
490{
491 return __btrfs_add_delayed_item(node, item,
492 BTRFS_DELAYED_DELETION_ITEM);
493}
494
495static void __btrfs_remove_delayed_item(struct btrfs_delayed_item *delayed_item)
496{
497 struct rb_root *root;
498 struct btrfs_delayed_root *delayed_root;
499
500 delayed_root = delayed_item->delayed_node->root->fs_info->delayed_root;
501
502 BUG_ON(!delayed_root);
503 BUG_ON(delayed_item->ins_or_del != BTRFS_DELAYED_DELETION_ITEM &&
504 delayed_item->ins_or_del != BTRFS_DELAYED_INSERTION_ITEM);
505
506 if (delayed_item->ins_or_del == BTRFS_DELAYED_INSERTION_ITEM)
507 root = &delayed_item->delayed_node->ins_root;
508 else
509 root = &delayed_item->delayed_node->del_root;
510
511 rb_erase(&delayed_item->rb_node, root);
512 delayed_item->delayed_node->count--;
513 atomic_dec(&delayed_root->items);
514 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND &&
515 waitqueue_active(&delayed_root->wait))
516 wake_up(&delayed_root->wait);
517}
518
519static void btrfs_release_delayed_item(struct btrfs_delayed_item *item)
520{
521 if (item) {
522 __btrfs_remove_delayed_item(item);
523 if (atomic_dec_and_test(&item->refs))
524 kfree(item);
525 }
526}
527
528struct btrfs_delayed_item *__btrfs_first_delayed_insertion_item(
529 struct btrfs_delayed_node *delayed_node)
530{
531 struct rb_node *p;
532 struct btrfs_delayed_item *item = NULL;
533
534 p = rb_first(&delayed_node->ins_root);
535 if (p)
536 item = rb_entry(p, struct btrfs_delayed_item, rb_node);
537
538 return item;
539}
540
541struct btrfs_delayed_item *__btrfs_first_delayed_deletion_item(
542 struct btrfs_delayed_node *delayed_node)
543{
544 struct rb_node *p;
545 struct btrfs_delayed_item *item = NULL;
546
547 p = rb_first(&delayed_node->del_root);
548 if (p)
549 item = rb_entry(p, struct btrfs_delayed_item, rb_node);
550
551 return item;
552}
553
554struct btrfs_delayed_item *__btrfs_next_delayed_item(
555 struct btrfs_delayed_item *item)
556{
557 struct rb_node *p;
558 struct btrfs_delayed_item *next = NULL;
559
560 p = rb_next(&item->rb_node);
561 if (p)
562 next = rb_entry(p, struct btrfs_delayed_item, rb_node);
563
564 return next;
565}
566
567static inline struct btrfs_root *btrfs_get_fs_root(struct btrfs_root *root,
568 u64 root_id)
569{
570 struct btrfs_key root_key;
571
572 if (root->objectid == root_id)
573 return root;
574
575 root_key.objectid = root_id;
576 root_key.type = BTRFS_ROOT_ITEM_KEY;
577 root_key.offset = (u64)-1;
578 return btrfs_read_fs_root_no_name(root->fs_info, &root_key);
579}
580
581static int btrfs_delayed_item_reserve_metadata(struct btrfs_trans_handle *trans,
582 struct btrfs_root *root,
583 struct btrfs_delayed_item *item)
584{
585 struct btrfs_block_rsv *src_rsv;
586 struct btrfs_block_rsv *dst_rsv;
587 u64 num_bytes;
588 int ret;
589
590 if (!trans->bytes_reserved)
591 return 0;
592
593 src_rsv = trans->block_rsv;
594 dst_rsv = &root->fs_info->global_block_rsv;
595
596 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
597 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes);
598 if (!ret)
599 item->bytes_reserved = num_bytes;
600
601 return ret;
602}
603
604static void btrfs_delayed_item_release_metadata(struct btrfs_root *root,
605 struct btrfs_delayed_item *item)
606{
607 struct btrfs_block_rsv *rsv;
608
609 if (!item->bytes_reserved)
610 return;
611
612 rsv = &root->fs_info->global_block_rsv;
613 btrfs_block_rsv_release(root, rsv,
614 item->bytes_reserved);
615}
616
617static int btrfs_delayed_inode_reserve_metadata(
618 struct btrfs_trans_handle *trans,
619 struct btrfs_root *root,
620 struct btrfs_delayed_node *node)
621{
622 struct btrfs_block_rsv *src_rsv;
623 struct btrfs_block_rsv *dst_rsv;
624 u64 num_bytes;
625 int ret;
626
627 if (!trans->bytes_reserved)
628 return 0;
629
630 src_rsv = trans->block_rsv;
631 dst_rsv = &root->fs_info->global_block_rsv;
632
633 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
634 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes);
635 if (!ret)
636 node->bytes_reserved = num_bytes;
637
638 return ret;
639}
640
641static void btrfs_delayed_inode_release_metadata(struct btrfs_root *root,
642 struct btrfs_delayed_node *node)
643{
644 struct btrfs_block_rsv *rsv;
645
646 if (!node->bytes_reserved)
647 return;
648
649 rsv = &root->fs_info->global_block_rsv;
650 btrfs_block_rsv_release(root, rsv,
651 node->bytes_reserved);
652 node->bytes_reserved = 0;
653}
654
655/*
656 * This helper will insert some continuous items into the same leaf according
657 * to the free space of the leaf.
658 */
659static int btrfs_batch_insert_items(struct btrfs_trans_handle *trans,
660 struct btrfs_root *root,
661 struct btrfs_path *path,
662 struct btrfs_delayed_item *item)
663{
664 struct btrfs_delayed_item *curr, *next;
665 int free_space;
666 int total_data_size = 0, total_size = 0;
667 struct extent_buffer *leaf;
668 char *data_ptr;
669 struct btrfs_key *keys;
670 u32 *data_size;
671 struct list_head head;
672 int slot;
673 int nitems;
674 int i;
675 int ret = 0;
676
677 BUG_ON(!path->nodes[0]);
678
679 leaf = path->nodes[0];
680 free_space = btrfs_leaf_free_space(root, leaf);
681 INIT_LIST_HEAD(&head);
682
683 next = item;
684 nitems = 0;
685
686 /*
687 * count the number of the continuous items that we can insert in batch
688 */
689 while (total_size + next->data_len + sizeof(struct btrfs_item) <=
690 free_space) {
691 total_data_size += next->data_len;
692 total_size += next->data_len + sizeof(struct btrfs_item);
693 list_add_tail(&next->tree_list, &head);
694 nitems++;
695
696 curr = next;
697 next = __btrfs_next_delayed_item(curr);
698 if (!next)
699 break;
700
701 if (!btrfs_is_continuous_delayed_item(curr, next))
702 break;
703 }
704
705 if (!nitems) {
706 ret = 0;
707 goto out;
708 }
709
710 /*
711 * we need allocate some memory space, but it might cause the task
712 * to sleep, so we set all locked nodes in the path to blocking locks
713 * first.
714 */
715 btrfs_set_path_blocking(path);
716
717 keys = kmalloc(sizeof(struct btrfs_key) * nitems, GFP_NOFS);
718 if (!keys) {
719 ret = -ENOMEM;
720 goto out;
721 }
722
723 data_size = kmalloc(sizeof(u32) * nitems, GFP_NOFS);
724 if (!data_size) {
725 ret = -ENOMEM;
726 goto error;
727 }
728
729 /* get keys of all the delayed items */
730 i = 0;
731 list_for_each_entry(next, &head, tree_list) {
732 keys[i] = next->key;
733 data_size[i] = next->data_len;
734 i++;
735 }
736
737 /* reset all the locked nodes in the patch to spinning locks. */
738 btrfs_clear_path_blocking(path, NULL, 0);
739
740 /* insert the keys of the items */
741 ret = setup_items_for_insert(trans, root, path, keys, data_size,
742 total_data_size, total_size, nitems);
743 if (ret)
744 goto error;
745
746 /* insert the dir index items */
747 slot = path->slots[0];
748 list_for_each_entry_safe(curr, next, &head, tree_list) {
749 data_ptr = btrfs_item_ptr(leaf, slot, char);
750 write_extent_buffer(leaf, &curr->data,
751 (unsigned long)data_ptr,
752 curr->data_len);
753 slot++;
754
755 btrfs_delayed_item_release_metadata(root, curr);
756
757 list_del(&curr->tree_list);
758 btrfs_release_delayed_item(curr);
759 }
760
761error:
762 kfree(data_size);
763 kfree(keys);
764out:
765 return ret;
766}
767
768/*
769 * This helper can just do simple insertion that needn't extend item for new
770 * data, such as directory name index insertion, inode insertion.
771 */
772static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans,
773 struct btrfs_root *root,
774 struct btrfs_path *path,
775 struct btrfs_delayed_item *delayed_item)
776{
777 struct extent_buffer *leaf;
778 struct btrfs_item *item;
779 char *ptr;
780 int ret;
781
782 ret = btrfs_insert_empty_item(trans, root, path, &delayed_item->key,
783 delayed_item->data_len);
784 if (ret < 0 && ret != -EEXIST)
785 return ret;
786
787 leaf = path->nodes[0];
788
789 item = btrfs_item_nr(leaf, path->slots[0]);
790 ptr = btrfs_item_ptr(leaf, path->slots[0], char);
791
792 write_extent_buffer(leaf, delayed_item->data, (unsigned long)ptr,
793 delayed_item->data_len);
794 btrfs_mark_buffer_dirty(leaf);
795
796 btrfs_delayed_item_release_metadata(root, delayed_item);
797 return 0;
798}
799
800/*
801 * we insert an item first, then if there are some continuous items, we try
802 * to insert those items into the same leaf.
803 */
804static int btrfs_insert_delayed_items(struct btrfs_trans_handle *trans,
805 struct btrfs_path *path,
806 struct btrfs_root *root,
807 struct btrfs_delayed_node *node)
808{
809 struct btrfs_delayed_item *curr, *prev;
810 int ret = 0;
811
812do_again:
813 mutex_lock(&node->mutex);
814 curr = __btrfs_first_delayed_insertion_item(node);
815 if (!curr)
816 goto insert_end;
817
818 ret = btrfs_insert_delayed_item(trans, root, path, curr);
819 if (ret < 0) {
820 btrfs_release_path(path);
821 goto insert_end;
822 }
823
824 prev = curr;
825 curr = __btrfs_next_delayed_item(prev);
826 if (curr && btrfs_is_continuous_delayed_item(prev, curr)) {
827 /* insert the continuous items into the same leaf */
828 path->slots[0]++;
829 btrfs_batch_insert_items(trans, root, path, curr);
830 }
831 btrfs_release_delayed_item(prev);
832 btrfs_mark_buffer_dirty(path->nodes[0]);
833
834 btrfs_release_path(path);
835 mutex_unlock(&node->mutex);
836 goto do_again;
837
838insert_end:
839 mutex_unlock(&node->mutex);
840 return ret;
841}
842
843static int btrfs_batch_delete_items(struct btrfs_trans_handle *trans,
844 struct btrfs_root *root,
845 struct btrfs_path *path,
846 struct btrfs_delayed_item *item)
847{
848 struct btrfs_delayed_item *curr, *next;
849 struct extent_buffer *leaf;
850 struct btrfs_key key;
851 struct list_head head;
852 int nitems, i, last_item;
853 int ret = 0;
854
855 BUG_ON(!path->nodes[0]);
856
857 leaf = path->nodes[0];
858
859 i = path->slots[0];
860 last_item = btrfs_header_nritems(leaf) - 1;
861 if (i > last_item)
862 return -ENOENT; /* FIXME: Is errno suitable? */
863
864 next = item;
865 INIT_LIST_HEAD(&head);
866 btrfs_item_key_to_cpu(leaf, &key, i);
867 nitems = 0;
868 /*
869 * count the number of the dir index items that we can delete in batch
870 */
871 while (btrfs_comp_cpu_keys(&next->key, &key) == 0) {
872 list_add_tail(&next->tree_list, &head);
873 nitems++;
874
875 curr = next;
876 next = __btrfs_next_delayed_item(curr);
877 if (!next)
878 break;
879
880 if (!btrfs_is_continuous_delayed_item(curr, next))
881 break;
882
883 i++;
884 if (i > last_item)
885 break;
886 btrfs_item_key_to_cpu(leaf, &key, i);
887 }
888
889 if (!nitems)
890 return 0;
891
892 ret = btrfs_del_items(trans, root, path, path->slots[0], nitems);
893 if (ret)
894 goto out;
895
896 list_for_each_entry_safe(curr, next, &head, tree_list) {
897 btrfs_delayed_item_release_metadata(root, curr);
898 list_del(&curr->tree_list);
899 btrfs_release_delayed_item(curr);
900 }
901
902out:
903 return ret;
904}
905
906static int btrfs_delete_delayed_items(struct btrfs_trans_handle *trans,
907 struct btrfs_path *path,
908 struct btrfs_root *root,
909 struct btrfs_delayed_node *node)
910{
911 struct btrfs_delayed_item *curr, *prev;
912 int ret = 0;
913
914do_again:
915 mutex_lock(&node->mutex);
916 curr = __btrfs_first_delayed_deletion_item(node);
917 if (!curr)
918 goto delete_fail;
919
920 ret = btrfs_search_slot(trans, root, &curr->key, path, -1, 1);
921 if (ret < 0)
922 goto delete_fail;
923 else if (ret > 0) {
924 /*
925 * can't find the item which the node points to, so this node
926 * is invalid, just drop it.
927 */
928 prev = curr;
929 curr = __btrfs_next_delayed_item(prev);
930 btrfs_release_delayed_item(prev);
931 ret = 0;
932 btrfs_release_path(path);
933 if (curr)
934 goto do_again;
935 else
936 goto delete_fail;
937 }
938
939 btrfs_batch_delete_items(trans, root, path, curr);
940 btrfs_release_path(path);
941 mutex_unlock(&node->mutex);
942 goto do_again;
943
944delete_fail:
945 btrfs_release_path(path);
946 mutex_unlock(&node->mutex);
947 return ret;
948}
949
950static void btrfs_release_delayed_inode(struct btrfs_delayed_node *delayed_node)
951{
952 struct btrfs_delayed_root *delayed_root;
953
954 if (delayed_node && delayed_node->inode_dirty) {
955 BUG_ON(!delayed_node->root);
956 delayed_node->inode_dirty = 0;
957 delayed_node->count--;
958
959 delayed_root = delayed_node->root->fs_info->delayed_root;
960 atomic_dec(&delayed_root->items);
961 if (atomic_read(&delayed_root->items) <
962 BTRFS_DELAYED_BACKGROUND &&
963 waitqueue_active(&delayed_root->wait))
964 wake_up(&delayed_root->wait);
965 }
966}
967
968static int btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
969 struct btrfs_root *root,
970 struct btrfs_path *path,
971 struct btrfs_delayed_node *node)
972{
973 struct btrfs_key key;
974 struct btrfs_inode_item *inode_item;
975 struct extent_buffer *leaf;
976 int ret;
977
978 mutex_lock(&node->mutex);
979 if (!node->inode_dirty) {
980 mutex_unlock(&node->mutex);
981 return 0;
982 }
983
984 key.objectid = node->inode_id;
985 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
986 key.offset = 0;
987 ret = btrfs_lookup_inode(trans, root, path, &key, 1);
988 if (ret > 0) {
989 btrfs_release_path(path);
990 mutex_unlock(&node->mutex);
991 return -ENOENT;
992 } else if (ret < 0) {
993 mutex_unlock(&node->mutex);
994 return ret;
995 }
996
997 btrfs_unlock_up_safe(path, 1);
998 leaf = path->nodes[0];
999 inode_item = btrfs_item_ptr(leaf, path->slots[0],
1000 struct btrfs_inode_item);
1001 write_extent_buffer(leaf, &node->inode_item, (unsigned long)inode_item,
1002 sizeof(struct btrfs_inode_item));
1003 btrfs_mark_buffer_dirty(leaf);
1004 btrfs_release_path(path);
1005
1006 btrfs_delayed_inode_release_metadata(root, node);
1007 btrfs_release_delayed_inode(node);
1008 mutex_unlock(&node->mutex);
1009
1010 return 0;
1011}
1012
1013/* Called when committing the transaction. */
1014int btrfs_run_delayed_items(struct btrfs_trans_handle *trans,
1015 struct btrfs_root *root)
1016{
1017 struct btrfs_delayed_root *delayed_root;
1018 struct btrfs_delayed_node *curr_node, *prev_node;
1019 struct btrfs_path *path;
1020 struct btrfs_block_rsv *block_rsv;
1021 int ret = 0;
1022
1023 path = btrfs_alloc_path();
1024 if (!path)
1025 return -ENOMEM;
1026 path->leave_spinning = 1;
1027
1028 block_rsv = trans->block_rsv;
1029 trans->block_rsv = &root->fs_info->global_block_rsv;
1030
1031 delayed_root = btrfs_get_delayed_root(root);
1032
1033 curr_node = btrfs_first_delayed_node(delayed_root);
1034 while (curr_node) {
1035 root = curr_node->root;
1036 ret = btrfs_insert_delayed_items(trans, path, root,
1037 curr_node);
1038 if (!ret)
1039 ret = btrfs_delete_delayed_items(trans, path, root,
1040 curr_node);
1041 if (!ret)
1042 ret = btrfs_update_delayed_inode(trans, root, path,
1043 curr_node);
1044 if (ret) {
1045 btrfs_release_delayed_node(curr_node);
1046 break;
1047 }
1048
1049 prev_node = curr_node;
1050 curr_node = btrfs_next_delayed_node(curr_node);
1051 btrfs_release_delayed_node(prev_node);
1052 }
1053
1054 btrfs_free_path(path);
1055 trans->block_rsv = block_rsv;
1056 return ret;
1057}
1058
1059static int __btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1060 struct btrfs_delayed_node *node)
1061{
1062 struct btrfs_path *path;
1063 struct btrfs_block_rsv *block_rsv;
1064 int ret;
1065
1066 path = btrfs_alloc_path();
1067 if (!path)
1068 return -ENOMEM;
1069 path->leave_spinning = 1;
1070
1071 block_rsv = trans->block_rsv;
1072 trans->block_rsv = &node->root->fs_info->global_block_rsv;
1073
1074 ret = btrfs_insert_delayed_items(trans, path, node->root, node);
1075 if (!ret)
1076 ret = btrfs_delete_delayed_items(trans, path, node->root, node);
1077 if (!ret)
1078 ret = btrfs_update_delayed_inode(trans, node->root, path, node);
1079 btrfs_free_path(path);
1080
1081 trans->block_rsv = block_rsv;
1082 return ret;
1083}
1084
1085int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1086 struct inode *inode)
1087{
1088 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1089 int ret;
1090
1091 if (!delayed_node)
1092 return 0;
1093
1094 mutex_lock(&delayed_node->mutex);
1095 if (!delayed_node->count) {
1096 mutex_unlock(&delayed_node->mutex);
1097 btrfs_release_delayed_node(delayed_node);
1098 return 0;
1099 }
1100 mutex_unlock(&delayed_node->mutex);
1101
1102 ret = __btrfs_commit_inode_delayed_items(trans, delayed_node);
1103 btrfs_release_delayed_node(delayed_node);
1104 return ret;
1105}
1106
1107void btrfs_remove_delayed_node(struct inode *inode)
1108{
1109 struct btrfs_delayed_node *delayed_node;
1110
1111 delayed_node = ACCESS_ONCE(BTRFS_I(inode)->delayed_node);
1112 if (!delayed_node)
1113 return;
1114
1115 BTRFS_I(inode)->delayed_node = NULL;
1116 btrfs_release_delayed_node(delayed_node);
1117}
1118
1119struct btrfs_async_delayed_node {
1120 struct btrfs_root *root;
1121 struct btrfs_delayed_node *delayed_node;
1122 struct btrfs_work work;
1123};
1124
1125static void btrfs_async_run_delayed_node_done(struct btrfs_work *work)
1126{
1127 struct btrfs_async_delayed_node *async_node;
1128 struct btrfs_trans_handle *trans;
1129 struct btrfs_path *path;
1130 struct btrfs_delayed_node *delayed_node = NULL;
1131 struct btrfs_root *root;
1132 struct btrfs_block_rsv *block_rsv;
1133 unsigned long nr = 0;
1134 int need_requeue = 0;
1135 int ret;
1136
1137 async_node = container_of(work, struct btrfs_async_delayed_node, work);
1138
1139 path = btrfs_alloc_path();
1140 if (!path)
1141 goto out;
1142 path->leave_spinning = 1;
1143
1144 delayed_node = async_node->delayed_node;
1145 root = delayed_node->root;
1146
1147 trans = btrfs_join_transaction(root);
1148 if (IS_ERR(trans))
1149 goto free_path;
1150
1151 block_rsv = trans->block_rsv;
1152 trans->block_rsv = &root->fs_info->global_block_rsv;
1153
1154 ret = btrfs_insert_delayed_items(trans, path, root, delayed_node);
1155 if (!ret)
1156 ret = btrfs_delete_delayed_items(trans, path, root,
1157 delayed_node);
1158
1159 if (!ret)
1160 btrfs_update_delayed_inode(trans, root, path, delayed_node);
1161
1162 /*
1163 * Maybe new delayed items have been inserted, so we need requeue
1164 * the work. Besides that, we must dequeue the empty delayed nodes
1165 * to avoid the race between delayed items balance and the worker.
1166 * The race like this:
1167 * Task1 Worker thread
1168 * count == 0, needn't requeue
1169 * also needn't insert the
1170 * delayed node into prepare
1171 * list again.
1172 * add lots of delayed items
1173 * queue the delayed node
1174 * already in the list,
1175 * and not in the prepare
1176 * list, it means the delayed
1177 * node is being dealt with
1178 * by the worker.
1179 * do delayed items balance
1180 * the delayed node is being
1181 * dealt with by the worker
1182 * now, just wait.
1183 * the worker goto idle.
1184 * Task1 will sleep until the transaction is commited.
1185 */
1186 mutex_lock(&delayed_node->mutex);
1187 if (delayed_node->count)
1188 need_requeue = 1;
1189 else
1190 btrfs_dequeue_delayed_node(root->fs_info->delayed_root,
1191 delayed_node);
1192 mutex_unlock(&delayed_node->mutex);
1193
1194 nr = trans->blocks_used;
1195
1196 trans->block_rsv = block_rsv;
1197 btrfs_end_transaction_dmeta(trans, root);
1198 __btrfs_btree_balance_dirty(root, nr);
1199free_path:
1200 btrfs_free_path(path);
1201out:
1202 if (need_requeue)
1203 btrfs_requeue_work(&async_node->work);
1204 else {
1205 btrfs_release_prepared_delayed_node(delayed_node);
1206 kfree(async_node);
1207 }
1208}
1209
1210static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root,
1211 struct btrfs_root *root, int all)
1212{
1213 struct btrfs_async_delayed_node *async_node;
1214 struct btrfs_delayed_node *curr;
1215 int count = 0;
1216
1217again:
1218 curr = btrfs_first_prepared_delayed_node(delayed_root);
1219 if (!curr)
1220 return 0;
1221
1222 async_node = kmalloc(sizeof(*async_node), GFP_NOFS);
1223 if (!async_node) {
1224 btrfs_release_prepared_delayed_node(curr);
1225 return -ENOMEM;
1226 }
1227
1228 async_node->root = root;
1229 async_node->delayed_node = curr;
1230
1231 async_node->work.func = btrfs_async_run_delayed_node_done;
1232 async_node->work.flags = 0;
1233
1234 btrfs_queue_worker(&root->fs_info->delayed_workers, &async_node->work);
1235 count++;
1236
1237 if (all || count < 4)
1238 goto again;
1239
1240 return 0;
1241}
1242
1243void btrfs_assert_delayed_root_empty(struct btrfs_root *root)
1244{
1245 struct btrfs_delayed_root *delayed_root;
1246 delayed_root = btrfs_get_delayed_root(root);
1247 WARN_ON(btrfs_first_delayed_node(delayed_root));
1248}
1249
1250void btrfs_balance_delayed_items(struct btrfs_root *root)
1251{
1252 struct btrfs_delayed_root *delayed_root;
1253
1254 delayed_root = btrfs_get_delayed_root(root);
1255
1256 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1257 return;
1258
1259 if (atomic_read(&delayed_root->items) >= BTRFS_DELAYED_WRITEBACK) {
1260 int ret;
1261 ret = btrfs_wq_run_delayed_node(delayed_root, root, 1);
1262 if (ret)
1263 return;
1264
1265 wait_event_interruptible_timeout(
1266 delayed_root->wait,
1267 (atomic_read(&delayed_root->items) <
1268 BTRFS_DELAYED_BACKGROUND),
1269 HZ);
1270 return;
1271 }
1272
1273 btrfs_wq_run_delayed_node(delayed_root, root, 0);
1274}
1275
1276int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans,
1277 struct btrfs_root *root, const char *name,
1278 int name_len, struct inode *dir,
1279 struct btrfs_disk_key *disk_key, u8 type,
1280 u64 index)
1281{
1282 struct btrfs_delayed_node *delayed_node;
1283 struct btrfs_delayed_item *delayed_item;
1284 struct btrfs_dir_item *dir_item;
1285 int ret;
1286
1287 delayed_node = btrfs_get_or_create_delayed_node(dir);
1288 if (IS_ERR(delayed_node))
1289 return PTR_ERR(delayed_node);
1290
1291 delayed_item = btrfs_alloc_delayed_item(sizeof(*dir_item) + name_len);
1292 if (!delayed_item) {
1293 ret = -ENOMEM;
1294 goto release_node;
1295 }
1296
1297 ret = btrfs_delayed_item_reserve_metadata(trans, root, delayed_item);
1298 /*
1299 * we have reserved enough space when we start a new transaction,
1300 * so reserving metadata failure is impossible
1301 */
1302 BUG_ON(ret);
1303
1304 delayed_item->key.objectid = btrfs_ino(dir);
1305 btrfs_set_key_type(&delayed_item->key, BTRFS_DIR_INDEX_KEY);
1306 delayed_item->key.offset = index;
1307
1308 dir_item = (struct btrfs_dir_item *)delayed_item->data;
1309 dir_item->location = *disk_key;
1310 dir_item->transid = cpu_to_le64(trans->transid);
1311 dir_item->data_len = 0;
1312 dir_item->name_len = cpu_to_le16(name_len);
1313 dir_item->type = type;
1314 memcpy((char *)(dir_item + 1), name, name_len);
1315
1316 mutex_lock(&delayed_node->mutex);
1317 ret = __btrfs_add_delayed_insertion_item(delayed_node, delayed_item);
1318 if (unlikely(ret)) {
1319 printk(KERN_ERR "err add delayed dir index item(name: %s) into "
1320 "the insertion tree of the delayed node"
1321 "(root id: %llu, inode id: %llu, errno: %d)\n",
1322 name,
1323 (unsigned long long)delayed_node->root->objectid,
1324 (unsigned long long)delayed_node->inode_id,
1325 ret);
1326 BUG();
1327 }
1328 mutex_unlock(&delayed_node->mutex);
1329
1330release_node:
1331 btrfs_release_delayed_node(delayed_node);
1332 return ret;
1333}
1334
1335static int btrfs_delete_delayed_insertion_item(struct btrfs_root *root,
1336 struct btrfs_delayed_node *node,
1337 struct btrfs_key *key)
1338{
1339 struct btrfs_delayed_item *item;
1340
1341 mutex_lock(&node->mutex);
1342 item = __btrfs_lookup_delayed_insertion_item(node, key);
1343 if (!item) {
1344 mutex_unlock(&node->mutex);
1345 return 1;
1346 }
1347
1348 btrfs_delayed_item_release_metadata(root, item);
1349 btrfs_release_delayed_item(item);
1350 mutex_unlock(&node->mutex);
1351 return 0;
1352}
1353
1354int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans,
1355 struct btrfs_root *root, struct inode *dir,
1356 u64 index)
1357{
1358 struct btrfs_delayed_node *node;
1359 struct btrfs_delayed_item *item;
1360 struct btrfs_key item_key;
1361 int ret;
1362
1363 node = btrfs_get_or_create_delayed_node(dir);
1364 if (IS_ERR(node))
1365 return PTR_ERR(node);
1366
1367 item_key.objectid = btrfs_ino(dir);
1368 btrfs_set_key_type(&item_key, BTRFS_DIR_INDEX_KEY);
1369 item_key.offset = index;
1370
1371 ret = btrfs_delete_delayed_insertion_item(root, node, &item_key);
1372 if (!ret)
1373 goto end;
1374
1375 item = btrfs_alloc_delayed_item(0);
1376 if (!item) {
1377 ret = -ENOMEM;
1378 goto end;
1379 }
1380
1381 item->key = item_key;
1382
1383 ret = btrfs_delayed_item_reserve_metadata(trans, root, item);
1384 /*
1385 * we have reserved enough space when we start a new transaction,
1386 * so reserving metadata failure is impossible.
1387 */
1388 BUG_ON(ret);
1389
1390 mutex_lock(&node->mutex);
1391 ret = __btrfs_add_delayed_deletion_item(node, item);
1392 if (unlikely(ret)) {
1393 printk(KERN_ERR "err add delayed dir index item(index: %llu) "
1394 "into the deletion tree of the delayed node"
1395 "(root id: %llu, inode id: %llu, errno: %d)\n",
1396 (unsigned long long)index,
1397 (unsigned long long)node->root->objectid,
1398 (unsigned long long)node->inode_id,
1399 ret);
1400 BUG();
1401 }
1402 mutex_unlock(&node->mutex);
1403end:
1404 btrfs_release_delayed_node(node);
1405 return ret;
1406}
1407
1408int btrfs_inode_delayed_dir_index_count(struct inode *inode)
1409{
1410 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1411
1412 if (!delayed_node)
1413 return -ENOENT;
1414
1415 /*
1416 * Since we have held i_mutex of this directory, it is impossible that
1417 * a new directory index is added into the delayed node and index_cnt
1418 * is updated now. So we needn't lock the delayed node.
1419 */
1420 if (!delayed_node->index_cnt) {
1421 btrfs_release_delayed_node(delayed_node);
1422 return -EINVAL;
1423 }
1424
1425 BTRFS_I(inode)->index_cnt = delayed_node->index_cnt;
1426 btrfs_release_delayed_node(delayed_node);
1427 return 0;
1428}
1429
1430void btrfs_get_delayed_items(struct inode *inode, struct list_head *ins_list,
1431 struct list_head *del_list)
1432{
1433 struct btrfs_delayed_node *delayed_node;
1434 struct btrfs_delayed_item *item;
1435
1436 delayed_node = btrfs_get_delayed_node(inode);
1437 if (!delayed_node)
1438 return;
1439
1440 mutex_lock(&delayed_node->mutex);
1441 item = __btrfs_first_delayed_insertion_item(delayed_node);
1442 while (item) {
1443 atomic_inc(&item->refs);
1444 list_add_tail(&item->readdir_list, ins_list);
1445 item = __btrfs_next_delayed_item(item);
1446 }
1447
1448 item = __btrfs_first_delayed_deletion_item(delayed_node);
1449 while (item) {
1450 atomic_inc(&item->refs);
1451 list_add_tail(&item->readdir_list, del_list);
1452 item = __btrfs_next_delayed_item(item);
1453 }
1454 mutex_unlock(&delayed_node->mutex);
1455 /*
1456 * This delayed node is still cached in the btrfs inode, so refs
1457 * must be > 1 now, and we needn't check it is going to be freed
1458 * or not.
1459 *
1460 * Besides that, this function is used to read dir, we do not
1461 * insert/delete delayed items in this period. So we also needn't
1462 * requeue or dequeue this delayed node.
1463 */
1464 atomic_dec(&delayed_node->refs);
1465}
1466
1467void btrfs_put_delayed_items(struct list_head *ins_list,
1468 struct list_head *del_list)
1469{
1470 struct btrfs_delayed_item *curr, *next;
1471
1472 list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1473 list_del(&curr->readdir_list);
1474 if (atomic_dec_and_test(&curr->refs))
1475 kfree(curr);
1476 }
1477
1478 list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1479 list_del(&curr->readdir_list);
1480 if (atomic_dec_and_test(&curr->refs))
1481 kfree(curr);
1482 }
1483}
1484
1485int btrfs_should_delete_dir_index(struct list_head *del_list,
1486 u64 index)
1487{
1488 struct btrfs_delayed_item *curr, *next;
1489 int ret;
1490
1491 if (list_empty(del_list))
1492 return 0;
1493
1494 list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1495 if (curr->key.offset > index)
1496 break;
1497
1498 list_del(&curr->readdir_list);
1499 ret = (curr->key.offset == index);
1500
1501 if (atomic_dec_and_test(&curr->refs))
1502 kfree(curr);
1503
1504 if (ret)
1505 return 1;
1506 else
1507 continue;
1508 }
1509 return 0;
1510}
1511
1512/*
1513 * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree
1514 *
1515 */
1516int btrfs_readdir_delayed_dir_index(struct file *filp, void *dirent,
1517 filldir_t filldir,
1518 struct list_head *ins_list)
1519{
1520 struct btrfs_dir_item *di;
1521 struct btrfs_delayed_item *curr, *next;
1522 struct btrfs_key location;
1523 char *name;
1524 int name_len;
1525 int over = 0;
1526 unsigned char d_type;
1527
1528 if (list_empty(ins_list))
1529 return 0;
1530
1531 /*
1532 * Changing the data of the delayed item is impossible. So
1533 * we needn't lock them. And we have held i_mutex of the
1534 * directory, nobody can delete any directory indexes now.
1535 */
1536 list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1537 list_del(&curr->readdir_list);
1538
1539 if (curr->key.offset < filp->f_pos) {
1540 if (atomic_dec_and_test(&curr->refs))
1541 kfree(curr);
1542 continue;
1543 }
1544
1545 filp->f_pos = curr->key.offset;
1546
1547 di = (struct btrfs_dir_item *)curr->data;
1548 name = (char *)(di + 1);
1549 name_len = le16_to_cpu(di->name_len);
1550
1551 d_type = btrfs_filetype_table[di->type];
1552 btrfs_disk_key_to_cpu(&location, &di->location);
1553
1554 over = filldir(dirent, name, name_len, curr->key.offset,
1555 location.objectid, d_type);
1556
1557 if (atomic_dec_and_test(&curr->refs))
1558 kfree(curr);
1559
1560 if (over)
1561 return 1;
1562 }
1563 return 0;
1564}
1565
1566BTRFS_SETGET_STACK_FUNCS(stack_inode_generation, struct btrfs_inode_item,
1567 generation, 64);
1568BTRFS_SETGET_STACK_FUNCS(stack_inode_sequence, struct btrfs_inode_item,
1569 sequence, 64);
1570BTRFS_SETGET_STACK_FUNCS(stack_inode_transid, struct btrfs_inode_item,
1571 transid, 64);
1572BTRFS_SETGET_STACK_FUNCS(stack_inode_size, struct btrfs_inode_item, size, 64);
1573BTRFS_SETGET_STACK_FUNCS(stack_inode_nbytes, struct btrfs_inode_item,
1574 nbytes, 64);
1575BTRFS_SETGET_STACK_FUNCS(stack_inode_block_group, struct btrfs_inode_item,
1576 block_group, 64);
1577BTRFS_SETGET_STACK_FUNCS(stack_inode_nlink, struct btrfs_inode_item, nlink, 32);
1578BTRFS_SETGET_STACK_FUNCS(stack_inode_uid, struct btrfs_inode_item, uid, 32);
1579BTRFS_SETGET_STACK_FUNCS(stack_inode_gid, struct btrfs_inode_item, gid, 32);
1580BTRFS_SETGET_STACK_FUNCS(stack_inode_mode, struct btrfs_inode_item, mode, 32);
1581BTRFS_SETGET_STACK_FUNCS(stack_inode_rdev, struct btrfs_inode_item, rdev, 64);
1582BTRFS_SETGET_STACK_FUNCS(stack_inode_flags, struct btrfs_inode_item, flags, 64);
1583
1584BTRFS_SETGET_STACK_FUNCS(stack_timespec_sec, struct btrfs_timespec, sec, 64);
1585BTRFS_SETGET_STACK_FUNCS(stack_timespec_nsec, struct btrfs_timespec, nsec, 32);
1586
1587static void fill_stack_inode_item(struct btrfs_trans_handle *trans,
1588 struct btrfs_inode_item *inode_item,
1589 struct inode *inode)
1590{
1591 btrfs_set_stack_inode_uid(inode_item, inode->i_uid);
1592 btrfs_set_stack_inode_gid(inode_item, inode->i_gid);
1593 btrfs_set_stack_inode_size(inode_item, BTRFS_I(inode)->disk_i_size);
1594 btrfs_set_stack_inode_mode(inode_item, inode->i_mode);
1595 btrfs_set_stack_inode_nlink(inode_item, inode->i_nlink);
1596 btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(inode));
1597 btrfs_set_stack_inode_generation(inode_item,
1598 BTRFS_I(inode)->generation);
1599 btrfs_set_stack_inode_sequence(inode_item, BTRFS_I(inode)->sequence);
1600 btrfs_set_stack_inode_transid(inode_item, trans->transid);
1601 btrfs_set_stack_inode_rdev(inode_item, inode->i_rdev);
1602 btrfs_set_stack_inode_flags(inode_item, BTRFS_I(inode)->flags);
1603 btrfs_set_stack_inode_block_group(inode_item, 0);
1604
1605 btrfs_set_stack_timespec_sec(btrfs_inode_atime(inode_item),
1606 inode->i_atime.tv_sec);
1607 btrfs_set_stack_timespec_nsec(btrfs_inode_atime(inode_item),
1608 inode->i_atime.tv_nsec);
1609
1610 btrfs_set_stack_timespec_sec(btrfs_inode_mtime(inode_item),
1611 inode->i_mtime.tv_sec);
1612 btrfs_set_stack_timespec_nsec(btrfs_inode_mtime(inode_item),
1613 inode->i_mtime.tv_nsec);
1614
1615 btrfs_set_stack_timespec_sec(btrfs_inode_ctime(inode_item),
1616 inode->i_ctime.tv_sec);
1617 btrfs_set_stack_timespec_nsec(btrfs_inode_ctime(inode_item),
1618 inode->i_ctime.tv_nsec);
1619}
1620
1621int btrfs_fill_inode(struct inode *inode, u32 *rdev)
1622{
1623 struct btrfs_delayed_node *delayed_node;
1624 struct btrfs_inode_item *inode_item;
1625 struct btrfs_timespec *tspec;
1626
1627 delayed_node = btrfs_get_delayed_node(inode);
1628 if (!delayed_node)
1629 return -ENOENT;
1630
1631 mutex_lock(&delayed_node->mutex);
1632 if (!delayed_node->inode_dirty) {
1633 mutex_unlock(&delayed_node->mutex);
1634 btrfs_release_delayed_node(delayed_node);
1635 return -ENOENT;
1636 }
1637
1638 inode_item = &delayed_node->inode_item;
1639
1640 inode->i_uid = btrfs_stack_inode_uid(inode_item);
1641 inode->i_gid = btrfs_stack_inode_gid(inode_item);
1642 btrfs_i_size_write(inode, btrfs_stack_inode_size(inode_item));
1643 inode->i_mode = btrfs_stack_inode_mode(inode_item);
1644 inode->i_nlink = btrfs_stack_inode_nlink(inode_item);
1645 inode_set_bytes(inode, btrfs_stack_inode_nbytes(inode_item));
1646 BTRFS_I(inode)->generation = btrfs_stack_inode_generation(inode_item);
1647 BTRFS_I(inode)->sequence = btrfs_stack_inode_sequence(inode_item);
1648 inode->i_rdev = 0;
1649 *rdev = btrfs_stack_inode_rdev(inode_item);
1650 BTRFS_I(inode)->flags = btrfs_stack_inode_flags(inode_item);
1651
1652 tspec = btrfs_inode_atime(inode_item);
1653 inode->i_atime.tv_sec = btrfs_stack_timespec_sec(tspec);
1654 inode->i_atime.tv_nsec = btrfs_stack_timespec_nsec(tspec);
1655
1656 tspec = btrfs_inode_mtime(inode_item);
1657 inode->i_mtime.tv_sec = btrfs_stack_timespec_sec(tspec);
1658 inode->i_mtime.tv_nsec = btrfs_stack_timespec_nsec(tspec);
1659
1660 tspec = btrfs_inode_ctime(inode_item);
1661 inode->i_ctime.tv_sec = btrfs_stack_timespec_sec(tspec);
1662 inode->i_ctime.tv_nsec = btrfs_stack_timespec_nsec(tspec);
1663
1664 inode->i_generation = BTRFS_I(inode)->generation;
1665 BTRFS_I(inode)->index_cnt = (u64)-1;
1666
1667 mutex_unlock(&delayed_node->mutex);
1668 btrfs_release_delayed_node(delayed_node);
1669 return 0;
1670}
1671
1672int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans,
1673 struct btrfs_root *root, struct inode *inode)
1674{
1675 struct btrfs_delayed_node *delayed_node;
1676 int ret = 0;
1677
1678 delayed_node = btrfs_get_or_create_delayed_node(inode);
1679 if (IS_ERR(delayed_node))
1680 return PTR_ERR(delayed_node);
1681
1682 mutex_lock(&delayed_node->mutex);
1683 if (delayed_node->inode_dirty) {
1684 fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1685 goto release_node;
1686 }
1687
1688 ret = btrfs_delayed_inode_reserve_metadata(trans, root, delayed_node);
1689 /*
1690 * we must reserve enough space when we start a new transaction,
1691 * so reserving metadata failure is impossible
1692 */
1693 BUG_ON(ret);
1694
1695 fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1696 delayed_node->inode_dirty = 1;
1697 delayed_node->count++;
1698 atomic_inc(&root->fs_info->delayed_root->items);
1699release_node:
1700 mutex_unlock(&delayed_node->mutex);
1701 btrfs_release_delayed_node(delayed_node);
1702 return ret;
1703}
1704
1705static void __btrfs_kill_delayed_node(struct btrfs_delayed_node *delayed_node)
1706{
1707 struct btrfs_root *root = delayed_node->root;
1708 struct btrfs_delayed_item *curr_item, *prev_item;
1709
1710 mutex_lock(&delayed_node->mutex);
1711 curr_item = __btrfs_first_delayed_insertion_item(delayed_node);
1712 while (curr_item) {
1713 btrfs_delayed_item_release_metadata(root, curr_item);
1714 prev_item = curr_item;
1715 curr_item = __btrfs_next_delayed_item(prev_item);
1716 btrfs_release_delayed_item(prev_item);
1717 }
1718
1719 curr_item = __btrfs_first_delayed_deletion_item(delayed_node);
1720 while (curr_item) {
1721 btrfs_delayed_item_release_metadata(root, curr_item);
1722 prev_item = curr_item;
1723 curr_item = __btrfs_next_delayed_item(prev_item);
1724 btrfs_release_delayed_item(prev_item);
1725 }
1726
1727 if (delayed_node->inode_dirty) {
1728 btrfs_delayed_inode_release_metadata(root, delayed_node);
1729 btrfs_release_delayed_inode(delayed_node);
1730 }
1731 mutex_unlock(&delayed_node->mutex);
1732}
1733
1734void btrfs_kill_delayed_inode_items(struct inode *inode)
1735{
1736 struct btrfs_delayed_node *delayed_node;
1737
1738 delayed_node = btrfs_get_delayed_node(inode);
1739 if (!delayed_node)
1740 return;
1741
1742 __btrfs_kill_delayed_node(delayed_node);
1743 btrfs_release_delayed_node(delayed_node);
1744}
1745
1746void btrfs_kill_all_delayed_nodes(struct btrfs_root *root)
1747{
1748 u64 inode_id = 0;
1749 struct btrfs_delayed_node *delayed_nodes[8];
1750 int i, n;
1751
1752 while (1) {
1753 spin_lock(&root->inode_lock);
1754 n = radix_tree_gang_lookup(&root->delayed_nodes_tree,
1755 (void **)delayed_nodes, inode_id,
1756 ARRAY_SIZE(delayed_nodes));
1757 if (!n) {
1758 spin_unlock(&root->inode_lock);
1759 break;
1760 }
1761
1762 inode_id = delayed_nodes[n - 1]->inode_id + 1;
1763
1764 for (i = 0; i < n; i++)
1765 atomic_inc(&delayed_nodes[i]->refs);
1766 spin_unlock(&root->inode_lock);
1767
1768 for (i = 0; i < n; i++) {
1769 __btrfs_kill_delayed_node(delayed_nodes[i]);
1770 btrfs_release_delayed_node(delayed_nodes[i]);
1771 }
1772 }
1773}
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
22static struct kmem_cache *delayed_node_cache;
23
24int __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
36void __cold btrfs_delayed_inode_exit(void)
37{
38 kmem_cache_destroy(delayed_node_cache);
39}
40
41static 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
55static 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
67static 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) */
123static 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
131again:
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 */
170static 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 */
191static 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
206static 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);
219out:
220 spin_unlock(&delayed_root->lock);
221
222 return node;
223}
224
225static 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);
246out:
247 spin_unlock(&delayed_root->lock);
248
249 return next;
250}
251
252static 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
286static inline void btrfs_release_delayed_node(struct btrfs_delayed_node *node)
287{
288 __btrfs_release_delayed_node(node, 0);
289}
290
291static 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);
305out:
306 spin_unlock(&delayed_root->lock);
307
308 return node;
309}
310
311static inline void btrfs_release_prepared_delayed_node(
312 struct btrfs_delayed_node *node)
313{
314 __btrfs_release_delayed_node(node, 1);
315}
316
317static 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 */
341static 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
392static 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
400static 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
451static 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
458static 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
465static 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
475static 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
500static 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
509static 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
522static 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
535static 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
548static 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
582static 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
602static int btrfs_delayed_inode_reserve_metadata(
603 struct btrfs_trans_handle *trans,
604 struct btrfs_root *root,
605 struct btrfs_delayed_node *node)
606{
607 struct btrfs_fs_info *fs_info = root->fs_info;
608 struct btrfs_block_rsv *src_rsv;
609 struct btrfs_block_rsv *dst_rsv;
610 u64 num_bytes;
611 int ret;
612
613 src_rsv = trans->block_rsv;
614 dst_rsv = &fs_info->delayed_block_rsv;
615
616 num_bytes = btrfs_calc_metadata_size(fs_info, 1);
617
618 /*
619 * btrfs_dirty_inode will update the inode under btrfs_join_transaction
620 * which doesn't reserve space for speed. This is a problem since we
621 * still need to reserve space for this update, so try to reserve the
622 * space.
623 *
624 * Now if src_rsv == delalloc_block_rsv we'll let it just steal since
625 * we always reserve enough to update the inode item.
626 */
627 if (!src_rsv || (!trans->bytes_reserved &&
628 src_rsv->type != BTRFS_BLOCK_RSV_DELALLOC)) {
629 ret = btrfs_qgroup_reserve_meta(root, num_bytes,
630 BTRFS_QGROUP_RSV_META_PREALLOC, true);
631 if (ret < 0)
632 return ret;
633 ret = btrfs_block_rsv_add(root, dst_rsv, num_bytes,
634 BTRFS_RESERVE_NO_FLUSH);
635 /* NO_FLUSH could only fail with -ENOSPC */
636 ASSERT(ret == 0 || ret == -ENOSPC);
637 if (ret)
638 btrfs_qgroup_free_meta_prealloc(root, num_bytes);
639 } else {
640 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, true);
641 }
642
643 if (!ret) {
644 trace_btrfs_space_reservation(fs_info, "delayed_inode",
645 node->inode_id, num_bytes, 1);
646 node->bytes_reserved = num_bytes;
647 }
648
649 return ret;
650}
651
652static void btrfs_delayed_inode_release_metadata(struct btrfs_fs_info *fs_info,
653 struct btrfs_delayed_node *node,
654 bool qgroup_free)
655{
656 struct btrfs_block_rsv *rsv;
657
658 if (!node->bytes_reserved)
659 return;
660
661 rsv = &fs_info->delayed_block_rsv;
662 trace_btrfs_space_reservation(fs_info, "delayed_inode",
663 node->inode_id, node->bytes_reserved, 0);
664 btrfs_block_rsv_release(fs_info, rsv, node->bytes_reserved, NULL);
665 if (qgroup_free)
666 btrfs_qgroup_free_meta_prealloc(node->root,
667 node->bytes_reserved);
668 else
669 btrfs_qgroup_convert_reserved_meta(node->root,
670 node->bytes_reserved);
671 node->bytes_reserved = 0;
672}
673
674/*
675 * This helper will insert some continuous items into the same leaf according
676 * to the free space of the leaf.
677 */
678static int btrfs_batch_insert_items(struct btrfs_root *root,
679 struct btrfs_path *path,
680 struct btrfs_delayed_item *item)
681{
682 struct btrfs_delayed_item *curr, *next;
683 int free_space;
684 int total_size = 0;
685 struct extent_buffer *leaf;
686 char *data_ptr;
687 struct btrfs_key *keys;
688 u32 *data_size;
689 struct list_head head;
690 int slot;
691 int nitems;
692 int i;
693 int ret = 0;
694
695 BUG_ON(!path->nodes[0]);
696
697 leaf = path->nodes[0];
698 free_space = btrfs_leaf_free_space(leaf);
699 INIT_LIST_HEAD(&head);
700
701 next = item;
702 nitems = 0;
703
704 /*
705 * count the number of the continuous items that we can insert in batch
706 */
707 while (total_size + next->data_len + sizeof(struct btrfs_item) <=
708 free_space) {
709 total_size += next->data_len + sizeof(struct btrfs_item);
710 list_add_tail(&next->tree_list, &head);
711 nitems++;
712
713 curr = next;
714 next = __btrfs_next_delayed_item(curr);
715 if (!next)
716 break;
717
718 if (!btrfs_is_continuous_delayed_item(curr, next))
719 break;
720 }
721
722 if (!nitems) {
723 ret = 0;
724 goto out;
725 }
726
727 keys = kmalloc_array(nitems, sizeof(struct btrfs_key), GFP_NOFS);
728 if (!keys) {
729 ret = -ENOMEM;
730 goto out;
731 }
732
733 data_size = kmalloc_array(nitems, sizeof(u32), GFP_NOFS);
734 if (!data_size) {
735 ret = -ENOMEM;
736 goto error;
737 }
738
739 /* get keys of all the delayed items */
740 i = 0;
741 list_for_each_entry(next, &head, tree_list) {
742 keys[i] = next->key;
743 data_size[i] = next->data_len;
744 i++;
745 }
746
747 /* insert the keys of the items */
748 setup_items_for_insert(root, path, keys, data_size, nitems);
749
750 /* insert the dir index items */
751 slot = path->slots[0];
752 list_for_each_entry_safe(curr, next, &head, tree_list) {
753 data_ptr = btrfs_item_ptr(leaf, slot, char);
754 write_extent_buffer(leaf, &curr->data,
755 (unsigned long)data_ptr,
756 curr->data_len);
757 slot++;
758
759 btrfs_delayed_item_release_metadata(root, curr);
760
761 list_del(&curr->tree_list);
762 btrfs_release_delayed_item(curr);
763 }
764
765error:
766 kfree(data_size);
767 kfree(keys);
768out:
769 return ret;
770}
771
772/*
773 * This helper can just do simple insertion that needn't extend item for new
774 * data, such as directory name index insertion, inode insertion.
775 */
776static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans,
777 struct btrfs_root *root,
778 struct btrfs_path *path,
779 struct btrfs_delayed_item *delayed_item)
780{
781 struct extent_buffer *leaf;
782 unsigned int nofs_flag;
783 char *ptr;
784 int ret;
785
786 nofs_flag = memalloc_nofs_save();
787 ret = btrfs_insert_empty_item(trans, root, path, &delayed_item->key,
788 delayed_item->data_len);
789 memalloc_nofs_restore(nofs_flag);
790 if (ret < 0 && ret != -EEXIST)
791 return ret;
792
793 leaf = path->nodes[0];
794
795 ptr = btrfs_item_ptr(leaf, path->slots[0], char);
796
797 write_extent_buffer(leaf, delayed_item->data, (unsigned long)ptr,
798 delayed_item->data_len);
799 btrfs_mark_buffer_dirty(leaf);
800
801 btrfs_delayed_item_release_metadata(root, delayed_item);
802 return 0;
803}
804
805/*
806 * we insert an item first, then if there are some continuous items, we try
807 * to insert those items into the same leaf.
808 */
809static int btrfs_insert_delayed_items(struct btrfs_trans_handle *trans,
810 struct btrfs_path *path,
811 struct btrfs_root *root,
812 struct btrfs_delayed_node *node)
813{
814 struct btrfs_delayed_item *curr, *prev;
815 int ret = 0;
816
817do_again:
818 mutex_lock(&node->mutex);
819 curr = __btrfs_first_delayed_insertion_item(node);
820 if (!curr)
821 goto insert_end;
822
823 ret = btrfs_insert_delayed_item(trans, root, path, curr);
824 if (ret < 0) {
825 btrfs_release_path(path);
826 goto insert_end;
827 }
828
829 prev = curr;
830 curr = __btrfs_next_delayed_item(prev);
831 if (curr && btrfs_is_continuous_delayed_item(prev, curr)) {
832 /* insert the continuous items into the same leaf */
833 path->slots[0]++;
834 btrfs_batch_insert_items(root, path, curr);
835 }
836 btrfs_release_delayed_item(prev);
837 btrfs_mark_buffer_dirty(path->nodes[0]);
838
839 btrfs_release_path(path);
840 mutex_unlock(&node->mutex);
841 goto do_again;
842
843insert_end:
844 mutex_unlock(&node->mutex);
845 return ret;
846}
847
848static int btrfs_batch_delete_items(struct btrfs_trans_handle *trans,
849 struct btrfs_root *root,
850 struct btrfs_path *path,
851 struct btrfs_delayed_item *item)
852{
853 struct btrfs_delayed_item *curr, *next;
854 struct extent_buffer *leaf;
855 struct btrfs_key key;
856 struct list_head head;
857 int nitems, i, last_item;
858 int ret = 0;
859
860 BUG_ON(!path->nodes[0]);
861
862 leaf = path->nodes[0];
863
864 i = path->slots[0];
865 last_item = btrfs_header_nritems(leaf) - 1;
866 if (i > last_item)
867 return -ENOENT; /* FIXME: Is errno suitable? */
868
869 next = item;
870 INIT_LIST_HEAD(&head);
871 btrfs_item_key_to_cpu(leaf, &key, i);
872 nitems = 0;
873 /*
874 * count the number of the dir index items that we can delete in batch
875 */
876 while (btrfs_comp_cpu_keys(&next->key, &key) == 0) {
877 list_add_tail(&next->tree_list, &head);
878 nitems++;
879
880 curr = next;
881 next = __btrfs_next_delayed_item(curr);
882 if (!next)
883 break;
884
885 if (!btrfs_is_continuous_delayed_item(curr, next))
886 break;
887
888 i++;
889 if (i > last_item)
890 break;
891 btrfs_item_key_to_cpu(leaf, &key, i);
892 }
893
894 if (!nitems)
895 return 0;
896
897 ret = btrfs_del_items(trans, root, path, path->slots[0], nitems);
898 if (ret)
899 goto out;
900
901 list_for_each_entry_safe(curr, next, &head, tree_list) {
902 btrfs_delayed_item_release_metadata(root, curr);
903 list_del(&curr->tree_list);
904 btrfs_release_delayed_item(curr);
905 }
906
907out:
908 return ret;
909}
910
911static int btrfs_delete_delayed_items(struct btrfs_trans_handle *trans,
912 struct btrfs_path *path,
913 struct btrfs_root *root,
914 struct btrfs_delayed_node *node)
915{
916 struct btrfs_delayed_item *curr, *prev;
917 unsigned int nofs_flag;
918 int ret = 0;
919
920do_again:
921 mutex_lock(&node->mutex);
922 curr = __btrfs_first_delayed_deletion_item(node);
923 if (!curr)
924 goto delete_fail;
925
926 nofs_flag = memalloc_nofs_save();
927 ret = btrfs_search_slot(trans, root, &curr->key, path, -1, 1);
928 memalloc_nofs_restore(nofs_flag);
929 if (ret < 0)
930 goto delete_fail;
931 else if (ret > 0) {
932 /*
933 * can't find the item which the node points to, so this node
934 * is invalid, just drop it.
935 */
936 prev = curr;
937 curr = __btrfs_next_delayed_item(prev);
938 btrfs_release_delayed_item(prev);
939 ret = 0;
940 btrfs_release_path(path);
941 if (curr) {
942 mutex_unlock(&node->mutex);
943 goto do_again;
944 } else
945 goto delete_fail;
946 }
947
948 btrfs_batch_delete_items(trans, root, path, curr);
949 btrfs_release_path(path);
950 mutex_unlock(&node->mutex);
951 goto do_again;
952
953delete_fail:
954 btrfs_release_path(path);
955 mutex_unlock(&node->mutex);
956 return ret;
957}
958
959static void btrfs_release_delayed_inode(struct btrfs_delayed_node *delayed_node)
960{
961 struct btrfs_delayed_root *delayed_root;
962
963 if (delayed_node &&
964 test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
965 BUG_ON(!delayed_node->root);
966 clear_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
967 delayed_node->count--;
968
969 delayed_root = delayed_node->root->fs_info->delayed_root;
970 finish_one_item(delayed_root);
971 }
972}
973
974static void btrfs_release_delayed_iref(struct btrfs_delayed_node *delayed_node)
975{
976
977 if (test_and_clear_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags)) {
978 struct btrfs_delayed_root *delayed_root;
979
980 ASSERT(delayed_node->root);
981 delayed_node->count--;
982
983 delayed_root = delayed_node->root->fs_info->delayed_root;
984 finish_one_item(delayed_root);
985 }
986}
987
988static int __btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
989 struct btrfs_root *root,
990 struct btrfs_path *path,
991 struct btrfs_delayed_node *node)
992{
993 struct btrfs_fs_info *fs_info = root->fs_info;
994 struct btrfs_key key;
995 struct btrfs_inode_item *inode_item;
996 struct extent_buffer *leaf;
997 unsigned int nofs_flag;
998 int mod;
999 int ret;
1000
1001 key.objectid = node->inode_id;
1002 key.type = BTRFS_INODE_ITEM_KEY;
1003 key.offset = 0;
1004
1005 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1006 mod = -1;
1007 else
1008 mod = 1;
1009
1010 nofs_flag = memalloc_nofs_save();
1011 ret = btrfs_lookup_inode(trans, root, path, &key, mod);
1012 memalloc_nofs_restore(nofs_flag);
1013 if (ret > 0)
1014 ret = -ENOENT;
1015 if (ret < 0)
1016 goto out;
1017
1018 leaf = path->nodes[0];
1019 inode_item = btrfs_item_ptr(leaf, path->slots[0],
1020 struct btrfs_inode_item);
1021 write_extent_buffer(leaf, &node->inode_item, (unsigned long)inode_item,
1022 sizeof(struct btrfs_inode_item));
1023 btrfs_mark_buffer_dirty(leaf);
1024
1025 if (!test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1026 goto out;
1027
1028 path->slots[0]++;
1029 if (path->slots[0] >= btrfs_header_nritems(leaf))
1030 goto search;
1031again:
1032 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1033 if (key.objectid != node->inode_id)
1034 goto out;
1035
1036 if (key.type != BTRFS_INODE_REF_KEY &&
1037 key.type != BTRFS_INODE_EXTREF_KEY)
1038 goto out;
1039
1040 /*
1041 * Delayed iref deletion is for the inode who has only one link,
1042 * so there is only one iref. The case that several irefs are
1043 * in the same item doesn't exist.
1044 */
1045 btrfs_del_item(trans, root, path);
1046out:
1047 btrfs_release_delayed_iref(node);
1048 btrfs_release_path(path);
1049err_out:
1050 btrfs_delayed_inode_release_metadata(fs_info, node, (ret < 0));
1051 btrfs_release_delayed_inode(node);
1052
1053 /*
1054 * If we fail to update the delayed inode we need to abort the
1055 * transaction, because we could leave the inode with the improper
1056 * counts behind.
1057 */
1058 if (ret && ret != -ENOENT)
1059 btrfs_abort_transaction(trans, ret);
1060
1061 return ret;
1062
1063search:
1064 btrfs_release_path(path);
1065
1066 key.type = BTRFS_INODE_EXTREF_KEY;
1067 key.offset = -1;
1068
1069 nofs_flag = memalloc_nofs_save();
1070 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1071 memalloc_nofs_restore(nofs_flag);
1072 if (ret < 0)
1073 goto err_out;
1074 ASSERT(ret);
1075
1076 ret = 0;
1077 leaf = path->nodes[0];
1078 path->slots[0]--;
1079 goto again;
1080}
1081
1082static inline int btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1083 struct btrfs_root *root,
1084 struct btrfs_path *path,
1085 struct btrfs_delayed_node *node)
1086{
1087 int ret;
1088
1089 mutex_lock(&node->mutex);
1090 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &node->flags)) {
1091 mutex_unlock(&node->mutex);
1092 return 0;
1093 }
1094
1095 ret = __btrfs_update_delayed_inode(trans, root, path, node);
1096 mutex_unlock(&node->mutex);
1097 return ret;
1098}
1099
1100static inline int
1101__btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1102 struct btrfs_path *path,
1103 struct btrfs_delayed_node *node)
1104{
1105 int ret;
1106
1107 ret = btrfs_insert_delayed_items(trans, path, node->root, node);
1108 if (ret)
1109 return ret;
1110
1111 ret = btrfs_delete_delayed_items(trans, path, node->root, node);
1112 if (ret)
1113 return ret;
1114
1115 ret = btrfs_update_delayed_inode(trans, node->root, path, node);
1116 return ret;
1117}
1118
1119/*
1120 * Called when committing the transaction.
1121 * Returns 0 on success.
1122 * Returns < 0 on error and returns with an aborted transaction with any
1123 * outstanding delayed items cleaned up.
1124 */
1125static int __btrfs_run_delayed_items(struct btrfs_trans_handle *trans, int nr)
1126{
1127 struct btrfs_fs_info *fs_info = trans->fs_info;
1128 struct btrfs_delayed_root *delayed_root;
1129 struct btrfs_delayed_node *curr_node, *prev_node;
1130 struct btrfs_path *path;
1131 struct btrfs_block_rsv *block_rsv;
1132 int ret = 0;
1133 bool count = (nr > 0);
1134
1135 if (TRANS_ABORTED(trans))
1136 return -EIO;
1137
1138 path = btrfs_alloc_path();
1139 if (!path)
1140 return -ENOMEM;
1141
1142 block_rsv = trans->block_rsv;
1143 trans->block_rsv = &fs_info->delayed_block_rsv;
1144
1145 delayed_root = fs_info->delayed_root;
1146
1147 curr_node = btrfs_first_delayed_node(delayed_root);
1148 while (curr_node && (!count || nr--)) {
1149 ret = __btrfs_commit_inode_delayed_items(trans, path,
1150 curr_node);
1151 if (ret) {
1152 btrfs_release_delayed_node(curr_node);
1153 curr_node = NULL;
1154 btrfs_abort_transaction(trans, ret);
1155 break;
1156 }
1157
1158 prev_node = curr_node;
1159 curr_node = btrfs_next_delayed_node(curr_node);
1160 btrfs_release_delayed_node(prev_node);
1161 }
1162
1163 if (curr_node)
1164 btrfs_release_delayed_node(curr_node);
1165 btrfs_free_path(path);
1166 trans->block_rsv = block_rsv;
1167
1168 return ret;
1169}
1170
1171int btrfs_run_delayed_items(struct btrfs_trans_handle *trans)
1172{
1173 return __btrfs_run_delayed_items(trans, -1);
1174}
1175
1176int btrfs_run_delayed_items_nr(struct btrfs_trans_handle *trans, int nr)
1177{
1178 return __btrfs_run_delayed_items(trans, nr);
1179}
1180
1181int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1182 struct btrfs_inode *inode)
1183{
1184 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1185 struct btrfs_path *path;
1186 struct btrfs_block_rsv *block_rsv;
1187 int ret;
1188
1189 if (!delayed_node)
1190 return 0;
1191
1192 mutex_lock(&delayed_node->mutex);
1193 if (!delayed_node->count) {
1194 mutex_unlock(&delayed_node->mutex);
1195 btrfs_release_delayed_node(delayed_node);
1196 return 0;
1197 }
1198 mutex_unlock(&delayed_node->mutex);
1199
1200 path = btrfs_alloc_path();
1201 if (!path) {
1202 btrfs_release_delayed_node(delayed_node);
1203 return -ENOMEM;
1204 }
1205
1206 block_rsv = trans->block_rsv;
1207 trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv;
1208
1209 ret = __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1210
1211 btrfs_release_delayed_node(delayed_node);
1212 btrfs_free_path(path);
1213 trans->block_rsv = block_rsv;
1214
1215 return ret;
1216}
1217
1218int btrfs_commit_inode_delayed_inode(struct btrfs_inode *inode)
1219{
1220 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1221 struct btrfs_trans_handle *trans;
1222 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1223 struct btrfs_path *path;
1224 struct btrfs_block_rsv *block_rsv;
1225 int ret;
1226
1227 if (!delayed_node)
1228 return 0;
1229
1230 mutex_lock(&delayed_node->mutex);
1231 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1232 mutex_unlock(&delayed_node->mutex);
1233 btrfs_release_delayed_node(delayed_node);
1234 return 0;
1235 }
1236 mutex_unlock(&delayed_node->mutex);
1237
1238 trans = btrfs_join_transaction(delayed_node->root);
1239 if (IS_ERR(trans)) {
1240 ret = PTR_ERR(trans);
1241 goto out;
1242 }
1243
1244 path = btrfs_alloc_path();
1245 if (!path) {
1246 ret = -ENOMEM;
1247 goto trans_out;
1248 }
1249
1250 block_rsv = trans->block_rsv;
1251 trans->block_rsv = &fs_info->delayed_block_rsv;
1252
1253 mutex_lock(&delayed_node->mutex);
1254 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags))
1255 ret = __btrfs_update_delayed_inode(trans, delayed_node->root,
1256 path, delayed_node);
1257 else
1258 ret = 0;
1259 mutex_unlock(&delayed_node->mutex);
1260
1261 btrfs_free_path(path);
1262 trans->block_rsv = block_rsv;
1263trans_out:
1264 btrfs_end_transaction(trans);
1265 btrfs_btree_balance_dirty(fs_info);
1266out:
1267 btrfs_release_delayed_node(delayed_node);
1268
1269 return ret;
1270}
1271
1272void btrfs_remove_delayed_node(struct btrfs_inode *inode)
1273{
1274 struct btrfs_delayed_node *delayed_node;
1275
1276 delayed_node = READ_ONCE(inode->delayed_node);
1277 if (!delayed_node)
1278 return;
1279
1280 inode->delayed_node = NULL;
1281 btrfs_release_delayed_node(delayed_node);
1282}
1283
1284struct btrfs_async_delayed_work {
1285 struct btrfs_delayed_root *delayed_root;
1286 int nr;
1287 struct btrfs_work work;
1288};
1289
1290static void btrfs_async_run_delayed_root(struct btrfs_work *work)
1291{
1292 struct btrfs_async_delayed_work *async_work;
1293 struct btrfs_delayed_root *delayed_root;
1294 struct btrfs_trans_handle *trans;
1295 struct btrfs_path *path;
1296 struct btrfs_delayed_node *delayed_node = NULL;
1297 struct btrfs_root *root;
1298 struct btrfs_block_rsv *block_rsv;
1299 int total_done = 0;
1300
1301 async_work = container_of(work, struct btrfs_async_delayed_work, work);
1302 delayed_root = async_work->delayed_root;
1303
1304 path = btrfs_alloc_path();
1305 if (!path)
1306 goto out;
1307
1308 do {
1309 if (atomic_read(&delayed_root->items) <
1310 BTRFS_DELAYED_BACKGROUND / 2)
1311 break;
1312
1313 delayed_node = btrfs_first_prepared_delayed_node(delayed_root);
1314 if (!delayed_node)
1315 break;
1316
1317 root = delayed_node->root;
1318
1319 trans = btrfs_join_transaction(root);
1320 if (IS_ERR(trans)) {
1321 btrfs_release_path(path);
1322 btrfs_release_prepared_delayed_node(delayed_node);
1323 total_done++;
1324 continue;
1325 }
1326
1327 block_rsv = trans->block_rsv;
1328 trans->block_rsv = &root->fs_info->delayed_block_rsv;
1329
1330 __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1331
1332 trans->block_rsv = block_rsv;
1333 btrfs_end_transaction(trans);
1334 btrfs_btree_balance_dirty_nodelay(root->fs_info);
1335
1336 btrfs_release_path(path);
1337 btrfs_release_prepared_delayed_node(delayed_node);
1338 total_done++;
1339
1340 } while ((async_work->nr == 0 && total_done < BTRFS_DELAYED_WRITEBACK)
1341 || total_done < async_work->nr);
1342
1343 btrfs_free_path(path);
1344out:
1345 wake_up(&delayed_root->wait);
1346 kfree(async_work);
1347}
1348
1349
1350static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root,
1351 struct btrfs_fs_info *fs_info, int nr)
1352{
1353 struct btrfs_async_delayed_work *async_work;
1354
1355 async_work = kmalloc(sizeof(*async_work), GFP_NOFS);
1356 if (!async_work)
1357 return -ENOMEM;
1358
1359 async_work->delayed_root = delayed_root;
1360 btrfs_init_work(&async_work->work, btrfs_async_run_delayed_root, NULL,
1361 NULL);
1362 async_work->nr = nr;
1363
1364 btrfs_queue_work(fs_info->delayed_workers, &async_work->work);
1365 return 0;
1366}
1367
1368void btrfs_assert_delayed_root_empty(struct btrfs_fs_info *fs_info)
1369{
1370 WARN_ON(btrfs_first_delayed_node(fs_info->delayed_root));
1371}
1372
1373static int could_end_wait(struct btrfs_delayed_root *delayed_root, int seq)
1374{
1375 int val = atomic_read(&delayed_root->items_seq);
1376
1377 if (val < seq || val >= seq + BTRFS_DELAYED_BATCH)
1378 return 1;
1379
1380 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1381 return 1;
1382
1383 return 0;
1384}
1385
1386void btrfs_balance_delayed_items(struct btrfs_fs_info *fs_info)
1387{
1388 struct btrfs_delayed_root *delayed_root = fs_info->delayed_root;
1389
1390 if ((atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND) ||
1391 btrfs_workqueue_normal_congested(fs_info->delayed_workers))
1392 return;
1393
1394 if (atomic_read(&delayed_root->items) >= BTRFS_DELAYED_WRITEBACK) {
1395 int seq;
1396 int ret;
1397
1398 seq = atomic_read(&delayed_root->items_seq);
1399
1400 ret = btrfs_wq_run_delayed_node(delayed_root, fs_info, 0);
1401 if (ret)
1402 return;
1403
1404 wait_event_interruptible(delayed_root->wait,
1405 could_end_wait(delayed_root, seq));
1406 return;
1407 }
1408
1409 btrfs_wq_run_delayed_node(delayed_root, fs_info, BTRFS_DELAYED_BATCH);
1410}
1411
1412/* Will return 0 or -ENOMEM */
1413int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans,
1414 const char *name, int name_len,
1415 struct btrfs_inode *dir,
1416 struct btrfs_disk_key *disk_key, u8 type,
1417 u64 index)
1418{
1419 struct btrfs_delayed_node *delayed_node;
1420 struct btrfs_delayed_item *delayed_item;
1421 struct btrfs_dir_item *dir_item;
1422 int ret;
1423
1424 delayed_node = btrfs_get_or_create_delayed_node(dir);
1425 if (IS_ERR(delayed_node))
1426 return PTR_ERR(delayed_node);
1427
1428 delayed_item = btrfs_alloc_delayed_item(sizeof(*dir_item) + name_len);
1429 if (!delayed_item) {
1430 ret = -ENOMEM;
1431 goto release_node;
1432 }
1433
1434 delayed_item->key.objectid = btrfs_ino(dir);
1435 delayed_item->key.type = BTRFS_DIR_INDEX_KEY;
1436 delayed_item->key.offset = index;
1437
1438 dir_item = (struct btrfs_dir_item *)delayed_item->data;
1439 dir_item->location = *disk_key;
1440 btrfs_set_stack_dir_transid(dir_item, trans->transid);
1441 btrfs_set_stack_dir_data_len(dir_item, 0);
1442 btrfs_set_stack_dir_name_len(dir_item, name_len);
1443 btrfs_set_stack_dir_type(dir_item, type);
1444 memcpy((char *)(dir_item + 1), name, name_len);
1445
1446 ret = btrfs_delayed_item_reserve_metadata(trans, dir->root, delayed_item);
1447 /*
1448 * we have reserved enough space when we start a new transaction,
1449 * so reserving metadata failure is impossible
1450 */
1451 BUG_ON(ret);
1452
1453 mutex_lock(&delayed_node->mutex);
1454 ret = __btrfs_add_delayed_insertion_item(delayed_node, delayed_item);
1455 if (unlikely(ret)) {
1456 btrfs_err(trans->fs_info,
1457 "err add delayed dir index item(name: %.*s) into the insertion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
1458 name_len, name, delayed_node->root->root_key.objectid,
1459 delayed_node->inode_id, ret);
1460 BUG();
1461 }
1462 mutex_unlock(&delayed_node->mutex);
1463
1464release_node:
1465 btrfs_release_delayed_node(delayed_node);
1466 return ret;
1467}
1468
1469static int btrfs_delete_delayed_insertion_item(struct btrfs_fs_info *fs_info,
1470 struct btrfs_delayed_node *node,
1471 struct btrfs_key *key)
1472{
1473 struct btrfs_delayed_item *item;
1474
1475 mutex_lock(&node->mutex);
1476 item = __btrfs_lookup_delayed_insertion_item(node, key);
1477 if (!item) {
1478 mutex_unlock(&node->mutex);
1479 return 1;
1480 }
1481
1482 btrfs_delayed_item_release_metadata(node->root, item);
1483 btrfs_release_delayed_item(item);
1484 mutex_unlock(&node->mutex);
1485 return 0;
1486}
1487
1488int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans,
1489 struct btrfs_inode *dir, u64 index)
1490{
1491 struct btrfs_delayed_node *node;
1492 struct btrfs_delayed_item *item;
1493 struct btrfs_key item_key;
1494 int ret;
1495
1496 node = btrfs_get_or_create_delayed_node(dir);
1497 if (IS_ERR(node))
1498 return PTR_ERR(node);
1499
1500 item_key.objectid = btrfs_ino(dir);
1501 item_key.type = BTRFS_DIR_INDEX_KEY;
1502 item_key.offset = index;
1503
1504 ret = btrfs_delete_delayed_insertion_item(trans->fs_info, node,
1505 &item_key);
1506 if (!ret)
1507 goto end;
1508
1509 item = btrfs_alloc_delayed_item(0);
1510 if (!item) {
1511 ret = -ENOMEM;
1512 goto end;
1513 }
1514
1515 item->key = item_key;
1516
1517 ret = btrfs_delayed_item_reserve_metadata(trans, dir->root, item);
1518 /*
1519 * we have reserved enough space when we start a new transaction,
1520 * so reserving metadata failure is impossible.
1521 */
1522 if (ret < 0) {
1523 btrfs_err(trans->fs_info,
1524"metadata reservation failed for delayed dir item deltiona, should have been reserved");
1525 btrfs_release_delayed_item(item);
1526 goto end;
1527 }
1528
1529 mutex_lock(&node->mutex);
1530 ret = __btrfs_add_delayed_deletion_item(node, item);
1531 if (unlikely(ret)) {
1532 btrfs_err(trans->fs_info,
1533 "err add delayed dir index item(index: %llu) into the deletion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
1534 index, node->root->root_key.objectid,
1535 node->inode_id, ret);
1536 btrfs_delayed_item_release_metadata(dir->root, item);
1537 btrfs_release_delayed_item(item);
1538 }
1539 mutex_unlock(&node->mutex);
1540end:
1541 btrfs_release_delayed_node(node);
1542 return ret;
1543}
1544
1545int btrfs_inode_delayed_dir_index_count(struct btrfs_inode *inode)
1546{
1547 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1548
1549 if (!delayed_node)
1550 return -ENOENT;
1551
1552 /*
1553 * Since we have held i_mutex of this directory, it is impossible that
1554 * a new directory index is added into the delayed node and index_cnt
1555 * is updated now. So we needn't lock the delayed node.
1556 */
1557 if (!delayed_node->index_cnt) {
1558 btrfs_release_delayed_node(delayed_node);
1559 return -EINVAL;
1560 }
1561
1562 inode->index_cnt = delayed_node->index_cnt;
1563 btrfs_release_delayed_node(delayed_node);
1564 return 0;
1565}
1566
1567bool btrfs_readdir_get_delayed_items(struct inode *inode,
1568 struct list_head *ins_list,
1569 struct list_head *del_list)
1570{
1571 struct btrfs_delayed_node *delayed_node;
1572 struct btrfs_delayed_item *item;
1573
1574 delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
1575 if (!delayed_node)
1576 return false;
1577
1578 /*
1579 * We can only do one readdir with delayed items at a time because of
1580 * item->readdir_list.
1581 */
1582 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
1583 btrfs_inode_lock(inode, 0);
1584
1585 mutex_lock(&delayed_node->mutex);
1586 item = __btrfs_first_delayed_insertion_item(delayed_node);
1587 while (item) {
1588 refcount_inc(&item->refs);
1589 list_add_tail(&item->readdir_list, ins_list);
1590 item = __btrfs_next_delayed_item(item);
1591 }
1592
1593 item = __btrfs_first_delayed_deletion_item(delayed_node);
1594 while (item) {
1595 refcount_inc(&item->refs);
1596 list_add_tail(&item->readdir_list, del_list);
1597 item = __btrfs_next_delayed_item(item);
1598 }
1599 mutex_unlock(&delayed_node->mutex);
1600 /*
1601 * This delayed node is still cached in the btrfs inode, so refs
1602 * must be > 1 now, and we needn't check it is going to be freed
1603 * or not.
1604 *
1605 * Besides that, this function is used to read dir, we do not
1606 * insert/delete delayed items in this period. So we also needn't
1607 * requeue or dequeue this delayed node.
1608 */
1609 refcount_dec(&delayed_node->refs);
1610
1611 return true;
1612}
1613
1614void btrfs_readdir_put_delayed_items(struct inode *inode,
1615 struct list_head *ins_list,
1616 struct list_head *del_list)
1617{
1618 struct btrfs_delayed_item *curr, *next;
1619
1620 list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1621 list_del(&curr->readdir_list);
1622 if (refcount_dec_and_test(&curr->refs))
1623 kfree(curr);
1624 }
1625
1626 list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1627 list_del(&curr->readdir_list);
1628 if (refcount_dec_and_test(&curr->refs))
1629 kfree(curr);
1630 }
1631
1632 /*
1633 * The VFS is going to do up_read(), so we need to downgrade back to a
1634 * read lock.
1635 */
1636 downgrade_write(&inode->i_rwsem);
1637}
1638
1639int btrfs_should_delete_dir_index(struct list_head *del_list,
1640 u64 index)
1641{
1642 struct btrfs_delayed_item *curr;
1643 int ret = 0;
1644
1645 list_for_each_entry(curr, del_list, readdir_list) {
1646 if (curr->key.offset > index)
1647 break;
1648 if (curr->key.offset == index) {
1649 ret = 1;
1650 break;
1651 }
1652 }
1653 return ret;
1654}
1655
1656/*
1657 * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree
1658 *
1659 */
1660int btrfs_readdir_delayed_dir_index(struct dir_context *ctx,
1661 struct list_head *ins_list)
1662{
1663 struct btrfs_dir_item *di;
1664 struct btrfs_delayed_item *curr, *next;
1665 struct btrfs_key location;
1666 char *name;
1667 int name_len;
1668 int over = 0;
1669 unsigned char d_type;
1670
1671 if (list_empty(ins_list))
1672 return 0;
1673
1674 /*
1675 * Changing the data of the delayed item is impossible. So
1676 * we needn't lock them. And we have held i_mutex of the
1677 * directory, nobody can delete any directory indexes now.
1678 */
1679 list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1680 list_del(&curr->readdir_list);
1681
1682 if (curr->key.offset < ctx->pos) {
1683 if (refcount_dec_and_test(&curr->refs))
1684 kfree(curr);
1685 continue;
1686 }
1687
1688 ctx->pos = curr->key.offset;
1689
1690 di = (struct btrfs_dir_item *)curr->data;
1691 name = (char *)(di + 1);
1692 name_len = btrfs_stack_dir_name_len(di);
1693
1694 d_type = fs_ftype_to_dtype(di->type);
1695 btrfs_disk_key_to_cpu(&location, &di->location);
1696
1697 over = !dir_emit(ctx, name, name_len,
1698 location.objectid, d_type);
1699
1700 if (refcount_dec_and_test(&curr->refs))
1701 kfree(curr);
1702
1703 if (over)
1704 return 1;
1705 ctx->pos++;
1706 }
1707 return 0;
1708}
1709
1710static void fill_stack_inode_item(struct btrfs_trans_handle *trans,
1711 struct btrfs_inode_item *inode_item,
1712 struct inode *inode)
1713{
1714 btrfs_set_stack_inode_uid(inode_item, i_uid_read(inode));
1715 btrfs_set_stack_inode_gid(inode_item, i_gid_read(inode));
1716 btrfs_set_stack_inode_size(inode_item, BTRFS_I(inode)->disk_i_size);
1717 btrfs_set_stack_inode_mode(inode_item, inode->i_mode);
1718 btrfs_set_stack_inode_nlink(inode_item, inode->i_nlink);
1719 btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(inode));
1720 btrfs_set_stack_inode_generation(inode_item,
1721 BTRFS_I(inode)->generation);
1722 btrfs_set_stack_inode_sequence(inode_item,
1723 inode_peek_iversion(inode));
1724 btrfs_set_stack_inode_transid(inode_item, trans->transid);
1725 btrfs_set_stack_inode_rdev(inode_item, inode->i_rdev);
1726 btrfs_set_stack_inode_flags(inode_item, BTRFS_I(inode)->flags);
1727 btrfs_set_stack_inode_block_group(inode_item, 0);
1728
1729 btrfs_set_stack_timespec_sec(&inode_item->atime,
1730 inode->i_atime.tv_sec);
1731 btrfs_set_stack_timespec_nsec(&inode_item->atime,
1732 inode->i_atime.tv_nsec);
1733
1734 btrfs_set_stack_timespec_sec(&inode_item->mtime,
1735 inode->i_mtime.tv_sec);
1736 btrfs_set_stack_timespec_nsec(&inode_item->mtime,
1737 inode->i_mtime.tv_nsec);
1738
1739 btrfs_set_stack_timespec_sec(&inode_item->ctime,
1740 inode->i_ctime.tv_sec);
1741 btrfs_set_stack_timespec_nsec(&inode_item->ctime,
1742 inode->i_ctime.tv_nsec);
1743
1744 btrfs_set_stack_timespec_sec(&inode_item->otime,
1745 BTRFS_I(inode)->i_otime.tv_sec);
1746 btrfs_set_stack_timespec_nsec(&inode_item->otime,
1747 BTRFS_I(inode)->i_otime.tv_nsec);
1748}
1749
1750int btrfs_fill_inode(struct inode *inode, u32 *rdev)
1751{
1752 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1753 struct btrfs_delayed_node *delayed_node;
1754 struct btrfs_inode_item *inode_item;
1755
1756 delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
1757 if (!delayed_node)
1758 return -ENOENT;
1759
1760 mutex_lock(&delayed_node->mutex);
1761 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1762 mutex_unlock(&delayed_node->mutex);
1763 btrfs_release_delayed_node(delayed_node);
1764 return -ENOENT;
1765 }
1766
1767 inode_item = &delayed_node->inode_item;
1768
1769 i_uid_write(inode, btrfs_stack_inode_uid(inode_item));
1770 i_gid_write(inode, btrfs_stack_inode_gid(inode_item));
1771 btrfs_i_size_write(BTRFS_I(inode), btrfs_stack_inode_size(inode_item));
1772 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
1773 round_up(i_size_read(inode), fs_info->sectorsize));
1774 inode->i_mode = btrfs_stack_inode_mode(inode_item);
1775 set_nlink(inode, btrfs_stack_inode_nlink(inode_item));
1776 inode_set_bytes(inode, btrfs_stack_inode_nbytes(inode_item));
1777 BTRFS_I(inode)->generation = btrfs_stack_inode_generation(inode_item);
1778 BTRFS_I(inode)->last_trans = btrfs_stack_inode_transid(inode_item);
1779
1780 inode_set_iversion_queried(inode,
1781 btrfs_stack_inode_sequence(inode_item));
1782 inode->i_rdev = 0;
1783 *rdev = btrfs_stack_inode_rdev(inode_item);
1784 BTRFS_I(inode)->flags = btrfs_stack_inode_flags(inode_item);
1785
1786 inode->i_atime.tv_sec = btrfs_stack_timespec_sec(&inode_item->atime);
1787 inode->i_atime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->atime);
1788
1789 inode->i_mtime.tv_sec = btrfs_stack_timespec_sec(&inode_item->mtime);
1790 inode->i_mtime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->mtime);
1791
1792 inode->i_ctime.tv_sec = btrfs_stack_timespec_sec(&inode_item->ctime);
1793 inode->i_ctime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->ctime);
1794
1795 BTRFS_I(inode)->i_otime.tv_sec =
1796 btrfs_stack_timespec_sec(&inode_item->otime);
1797 BTRFS_I(inode)->i_otime.tv_nsec =
1798 btrfs_stack_timespec_nsec(&inode_item->otime);
1799
1800 inode->i_generation = BTRFS_I(inode)->generation;
1801 BTRFS_I(inode)->index_cnt = (u64)-1;
1802
1803 mutex_unlock(&delayed_node->mutex);
1804 btrfs_release_delayed_node(delayed_node);
1805 return 0;
1806}
1807
1808int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans,
1809 struct btrfs_root *root,
1810 struct btrfs_inode *inode)
1811{
1812 struct btrfs_delayed_node *delayed_node;
1813 int ret = 0;
1814
1815 delayed_node = btrfs_get_or_create_delayed_node(inode);
1816 if (IS_ERR(delayed_node))
1817 return PTR_ERR(delayed_node);
1818
1819 mutex_lock(&delayed_node->mutex);
1820 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1821 fill_stack_inode_item(trans, &delayed_node->inode_item,
1822 &inode->vfs_inode);
1823 goto release_node;
1824 }
1825
1826 ret = btrfs_delayed_inode_reserve_metadata(trans, root, delayed_node);
1827 if (ret)
1828 goto release_node;
1829
1830 fill_stack_inode_item(trans, &delayed_node->inode_item, &inode->vfs_inode);
1831 set_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
1832 delayed_node->count++;
1833 atomic_inc(&root->fs_info->delayed_root->items);
1834release_node:
1835 mutex_unlock(&delayed_node->mutex);
1836 btrfs_release_delayed_node(delayed_node);
1837 return ret;
1838}
1839
1840int btrfs_delayed_delete_inode_ref(struct btrfs_inode *inode)
1841{
1842 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1843 struct btrfs_delayed_node *delayed_node;
1844
1845 /*
1846 * we don't do delayed inode updates during log recovery because it
1847 * leads to enospc problems. This means we also can't do
1848 * delayed inode refs
1849 */
1850 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
1851 return -EAGAIN;
1852
1853 delayed_node = btrfs_get_or_create_delayed_node(inode);
1854 if (IS_ERR(delayed_node))
1855 return PTR_ERR(delayed_node);
1856
1857 /*
1858 * We don't reserve space for inode ref deletion is because:
1859 * - We ONLY do async inode ref deletion for the inode who has only
1860 * one link(i_nlink == 1), it means there is only one inode ref.
1861 * And in most case, the inode ref and the inode item are in the
1862 * same leaf, and we will deal with them at the same time.
1863 * Since we are sure we will reserve the space for the inode item,
1864 * it is unnecessary to reserve space for inode ref deletion.
1865 * - If the inode ref and the inode item are not in the same leaf,
1866 * We also needn't worry about enospc problem, because we reserve
1867 * much more space for the inode update than it needs.
1868 * - At the worst, we can steal some space from the global reservation.
1869 * It is very rare.
1870 */
1871 mutex_lock(&delayed_node->mutex);
1872 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1873 goto release_node;
1874
1875 set_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
1876 delayed_node->count++;
1877 atomic_inc(&fs_info->delayed_root->items);
1878release_node:
1879 mutex_unlock(&delayed_node->mutex);
1880 btrfs_release_delayed_node(delayed_node);
1881 return 0;
1882}
1883
1884static void __btrfs_kill_delayed_node(struct btrfs_delayed_node *delayed_node)
1885{
1886 struct btrfs_root *root = delayed_node->root;
1887 struct btrfs_fs_info *fs_info = root->fs_info;
1888 struct btrfs_delayed_item *curr_item, *prev_item;
1889
1890 mutex_lock(&delayed_node->mutex);
1891 curr_item = __btrfs_first_delayed_insertion_item(delayed_node);
1892 while (curr_item) {
1893 btrfs_delayed_item_release_metadata(root, curr_item);
1894 prev_item = curr_item;
1895 curr_item = __btrfs_next_delayed_item(prev_item);
1896 btrfs_release_delayed_item(prev_item);
1897 }
1898
1899 curr_item = __btrfs_first_delayed_deletion_item(delayed_node);
1900 while (curr_item) {
1901 btrfs_delayed_item_release_metadata(root, curr_item);
1902 prev_item = curr_item;
1903 curr_item = __btrfs_next_delayed_item(prev_item);
1904 btrfs_release_delayed_item(prev_item);
1905 }
1906
1907 btrfs_release_delayed_iref(delayed_node);
1908
1909 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1910 btrfs_delayed_inode_release_metadata(fs_info, delayed_node, false);
1911 btrfs_release_delayed_inode(delayed_node);
1912 }
1913 mutex_unlock(&delayed_node->mutex);
1914}
1915
1916void btrfs_kill_delayed_inode_items(struct btrfs_inode *inode)
1917{
1918 struct btrfs_delayed_node *delayed_node;
1919
1920 delayed_node = btrfs_get_delayed_node(inode);
1921 if (!delayed_node)
1922 return;
1923
1924 __btrfs_kill_delayed_node(delayed_node);
1925 btrfs_release_delayed_node(delayed_node);
1926}
1927
1928void btrfs_kill_all_delayed_nodes(struct btrfs_root *root)
1929{
1930 u64 inode_id = 0;
1931 struct btrfs_delayed_node *delayed_nodes[8];
1932 int i, n;
1933
1934 while (1) {
1935 spin_lock(&root->inode_lock);
1936 n = radix_tree_gang_lookup(&root->delayed_nodes_tree,
1937 (void **)delayed_nodes, inode_id,
1938 ARRAY_SIZE(delayed_nodes));
1939 if (!n) {
1940 spin_unlock(&root->inode_lock);
1941 break;
1942 }
1943
1944 inode_id = delayed_nodes[n - 1]->inode_id + 1;
1945 for (i = 0; i < n; i++) {
1946 /*
1947 * Don't increase refs in case the node is dead and
1948 * about to be removed from the tree in the loop below
1949 */
1950 if (!refcount_inc_not_zero(&delayed_nodes[i]->refs))
1951 delayed_nodes[i] = NULL;
1952 }
1953 spin_unlock(&root->inode_lock);
1954
1955 for (i = 0; i < n; i++) {
1956 if (!delayed_nodes[i])
1957 continue;
1958 __btrfs_kill_delayed_node(delayed_nodes[i]);
1959 btrfs_release_delayed_node(delayed_nodes[i]);
1960 }
1961 }
1962}
1963
1964void btrfs_destroy_delayed_inodes(struct btrfs_fs_info *fs_info)
1965{
1966 struct btrfs_delayed_node *curr_node, *prev_node;
1967
1968 curr_node = btrfs_first_delayed_node(fs_info->delayed_root);
1969 while (curr_node) {
1970 __btrfs_kill_delayed_node(curr_node);
1971
1972 prev_node = curr_node;
1973 curr_node = btrfs_next_delayed_node(curr_node);
1974 btrfs_release_delayed_node(prev_node);
1975 }
1976}
1977