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/*
2 * Copyright (C) 2011 Fujitsu. All rights reserved.
3 * Written by Miao Xie <miaox@cn.fujitsu.com>
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public
7 * License v2 as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12 * General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public
15 * License along with this program; if not, write to the
16 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
17 * Boston, MA 021110-1307, USA.
18 */
19
20#include <linux/slab.h>
21#include "delayed-inode.h"
22#include "disk-io.h"
23#include "transaction.h"
24#include "ctree.h"
25
26#define BTRFS_DELAYED_WRITEBACK 512
27#define BTRFS_DELAYED_BACKGROUND 128
28#define BTRFS_DELAYED_BATCH 16
29
30static struct kmem_cache *delayed_node_cache;
31
32int __init btrfs_delayed_inode_init(void)
33{
34 delayed_node_cache = kmem_cache_create("btrfs_delayed_node",
35 sizeof(struct btrfs_delayed_node),
36 0,
37 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
38 NULL);
39 if (!delayed_node_cache)
40 return -ENOMEM;
41 return 0;
42}
43
44void btrfs_delayed_inode_exit(void)
45{
46 if (delayed_node_cache)
47 kmem_cache_destroy(delayed_node_cache);
48}
49
50static inline void btrfs_init_delayed_node(
51 struct btrfs_delayed_node *delayed_node,
52 struct btrfs_root *root, u64 inode_id)
53{
54 delayed_node->root = root;
55 delayed_node->inode_id = inode_id;
56 atomic_set(&delayed_node->refs, 0);
57 delayed_node->count = 0;
58 delayed_node->flags = 0;
59 delayed_node->ins_root = RB_ROOT;
60 delayed_node->del_root = RB_ROOT;
61 mutex_init(&delayed_node->mutex);
62 delayed_node->index_cnt = 0;
63 INIT_LIST_HEAD(&delayed_node->n_list);
64 INIT_LIST_HEAD(&delayed_node->p_list);
65 delayed_node->bytes_reserved = 0;
66 memset(&delayed_node->inode_item, 0, sizeof(delayed_node->inode_item));
67}
68
69static inline int btrfs_is_continuous_delayed_item(
70 struct btrfs_delayed_item *item1,
71 struct btrfs_delayed_item *item2)
72{
73 if (item1->key.type == BTRFS_DIR_INDEX_KEY &&
74 item1->key.objectid == item2->key.objectid &&
75 item1->key.type == item2->key.type &&
76 item1->key.offset + 1 == item2->key.offset)
77 return 1;
78 return 0;
79}
80
81static inline struct btrfs_delayed_root *btrfs_get_delayed_root(
82 struct btrfs_root *root)
83{
84 return root->fs_info->delayed_root;
85}
86
87static struct btrfs_delayed_node *btrfs_get_delayed_node(struct inode *inode)
88{
89 struct btrfs_inode *btrfs_inode = BTRFS_I(inode);
90 struct btrfs_root *root = btrfs_inode->root;
91 u64 ino = btrfs_ino(inode);
92 struct btrfs_delayed_node *node;
93
94 node = ACCESS_ONCE(btrfs_inode->delayed_node);
95 if (node) {
96 atomic_inc(&node->refs);
97 return node;
98 }
99
100 spin_lock(&root->inode_lock);
101 node = radix_tree_lookup(&root->delayed_nodes_tree, ino);
102 if (node) {
103 if (btrfs_inode->delayed_node) {
104 atomic_inc(&node->refs); /* can be accessed */
105 BUG_ON(btrfs_inode->delayed_node != node);
106 spin_unlock(&root->inode_lock);
107 return node;
108 }
109 btrfs_inode->delayed_node = node;
110 /* can be accessed and cached in the inode */
111 atomic_add(2, &node->refs);
112 spin_unlock(&root->inode_lock);
113 return node;
114 }
115 spin_unlock(&root->inode_lock);
116
117 return NULL;
118}
119
120/* Will return either the node or PTR_ERR(-ENOMEM) */
121static struct btrfs_delayed_node *btrfs_get_or_create_delayed_node(
122 struct inode *inode)
123{
124 struct btrfs_delayed_node *node;
125 struct btrfs_inode *btrfs_inode = BTRFS_I(inode);
126 struct btrfs_root *root = btrfs_inode->root;
127 u64 ino = btrfs_ino(inode);
128 int ret;
129
130again:
131 node = btrfs_get_delayed_node(inode);
132 if (node)
133 return node;
134
135 node = kmem_cache_alloc(delayed_node_cache, GFP_NOFS);
136 if (!node)
137 return ERR_PTR(-ENOMEM);
138 btrfs_init_delayed_node(node, root, ino);
139
140 /* cached in the btrfs inode and can be accessed */
141 atomic_add(2, &node->refs);
142
143 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
144 if (ret) {
145 kmem_cache_free(delayed_node_cache, node);
146 return ERR_PTR(ret);
147 }
148
149 spin_lock(&root->inode_lock);
150 ret = radix_tree_insert(&root->delayed_nodes_tree, ino, node);
151 if (ret == -EEXIST) {
152 kmem_cache_free(delayed_node_cache, node);
153 spin_unlock(&root->inode_lock);
154 radix_tree_preload_end();
155 goto again;
156 }
157 btrfs_inode->delayed_node = node;
158 spin_unlock(&root->inode_lock);
159 radix_tree_preload_end();
160
161 return node;
162}
163
164/*
165 * Call it when holding delayed_node->mutex
166 *
167 * If mod = 1, add this node into the prepared list.
168 */
169static void btrfs_queue_delayed_node(struct btrfs_delayed_root *root,
170 struct btrfs_delayed_node *node,
171 int mod)
172{
173 spin_lock(&root->lock);
174 if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
175 if (!list_empty(&node->p_list))
176 list_move_tail(&node->p_list, &root->prepare_list);
177 else if (mod)
178 list_add_tail(&node->p_list, &root->prepare_list);
179 } else {
180 list_add_tail(&node->n_list, &root->node_list);
181 list_add_tail(&node->p_list, &root->prepare_list);
182 atomic_inc(&node->refs); /* inserted into list */
183 root->nodes++;
184 set_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
185 }
186 spin_unlock(&root->lock);
187}
188
189/* Call it when holding delayed_node->mutex */
190static void btrfs_dequeue_delayed_node(struct btrfs_delayed_root *root,
191 struct btrfs_delayed_node *node)
192{
193 spin_lock(&root->lock);
194 if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
195 root->nodes--;
196 atomic_dec(&node->refs); /* not in the list */
197 list_del_init(&node->n_list);
198 if (!list_empty(&node->p_list))
199 list_del_init(&node->p_list);
200 clear_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
201 }
202 spin_unlock(&root->lock);
203}
204
205static struct btrfs_delayed_node *btrfs_first_delayed_node(
206 struct btrfs_delayed_root *delayed_root)
207{
208 struct list_head *p;
209 struct btrfs_delayed_node *node = NULL;
210
211 spin_lock(&delayed_root->lock);
212 if (list_empty(&delayed_root->node_list))
213 goto out;
214
215 p = delayed_root->node_list.next;
216 node = list_entry(p, struct btrfs_delayed_node, n_list);
217 atomic_inc(&node->refs);
218out:
219 spin_unlock(&delayed_root->lock);
220
221 return node;
222}
223
224static struct btrfs_delayed_node *btrfs_next_delayed_node(
225 struct btrfs_delayed_node *node)
226{
227 struct btrfs_delayed_root *delayed_root;
228 struct list_head *p;
229 struct btrfs_delayed_node *next = NULL;
230
231 delayed_root = node->root->fs_info->delayed_root;
232 spin_lock(&delayed_root->lock);
233 if (!test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
234 /* not in the list */
235 if (list_empty(&delayed_root->node_list))
236 goto out;
237 p = delayed_root->node_list.next;
238 } else if (list_is_last(&node->n_list, &delayed_root->node_list))
239 goto out;
240 else
241 p = node->n_list.next;
242
243 next = list_entry(p, struct btrfs_delayed_node, n_list);
244 atomic_inc(&next->refs);
245out:
246 spin_unlock(&delayed_root->lock);
247
248 return next;
249}
250
251static void __btrfs_release_delayed_node(
252 struct btrfs_delayed_node *delayed_node,
253 int mod)
254{
255 struct btrfs_delayed_root *delayed_root;
256
257 if (!delayed_node)
258 return;
259
260 delayed_root = delayed_node->root->fs_info->delayed_root;
261
262 mutex_lock(&delayed_node->mutex);
263 if (delayed_node->count)
264 btrfs_queue_delayed_node(delayed_root, delayed_node, mod);
265 else
266 btrfs_dequeue_delayed_node(delayed_root, delayed_node);
267 mutex_unlock(&delayed_node->mutex);
268
269 if (atomic_dec_and_test(&delayed_node->refs)) {
270 struct btrfs_root *root = delayed_node->root;
271 spin_lock(&root->inode_lock);
272 if (atomic_read(&delayed_node->refs) == 0) {
273 radix_tree_delete(&root->delayed_nodes_tree,
274 delayed_node->inode_id);
275 kmem_cache_free(delayed_node_cache, delayed_node);
276 }
277 spin_unlock(&root->inode_lock);
278 }
279}
280
281static inline void btrfs_release_delayed_node(struct btrfs_delayed_node *node)
282{
283 __btrfs_release_delayed_node(node, 0);
284}
285
286static struct btrfs_delayed_node *btrfs_first_prepared_delayed_node(
287 struct btrfs_delayed_root *delayed_root)
288{
289 struct list_head *p;
290 struct btrfs_delayed_node *node = NULL;
291
292 spin_lock(&delayed_root->lock);
293 if (list_empty(&delayed_root->prepare_list))
294 goto out;
295
296 p = delayed_root->prepare_list.next;
297 list_del_init(p);
298 node = list_entry(p, struct btrfs_delayed_node, p_list);
299 atomic_inc(&node->refs);
300out:
301 spin_unlock(&delayed_root->lock);
302
303 return node;
304}
305
306static inline void btrfs_release_prepared_delayed_node(
307 struct btrfs_delayed_node *node)
308{
309 __btrfs_release_delayed_node(node, 1);
310}
311
312static struct btrfs_delayed_item *btrfs_alloc_delayed_item(u32 data_len)
313{
314 struct btrfs_delayed_item *item;
315 item = kmalloc(sizeof(*item) + data_len, GFP_NOFS);
316 if (item) {
317 item->data_len = data_len;
318 item->ins_or_del = 0;
319 item->bytes_reserved = 0;
320 item->delayed_node = NULL;
321 atomic_set(&item->refs, 1);
322 }
323 return item;
324}
325
326/*
327 * __btrfs_lookup_delayed_item - look up the delayed item by key
328 * @delayed_node: pointer to the delayed node
329 * @key: the key to look up
330 * @prev: used to store the prev item if the right item isn't found
331 * @next: used to store the next item if the right item isn't found
332 *
333 * Note: if we don't find the right item, we will return the prev item and
334 * the next item.
335 */
336static struct btrfs_delayed_item *__btrfs_lookup_delayed_item(
337 struct rb_root *root,
338 struct btrfs_key *key,
339 struct btrfs_delayed_item **prev,
340 struct btrfs_delayed_item **next)
341{
342 struct rb_node *node, *prev_node = NULL;
343 struct btrfs_delayed_item *delayed_item = NULL;
344 int ret = 0;
345
346 node = root->rb_node;
347
348 while (node) {
349 delayed_item = rb_entry(node, struct btrfs_delayed_item,
350 rb_node);
351 prev_node = node;
352 ret = btrfs_comp_cpu_keys(&delayed_item->key, key);
353 if (ret < 0)
354 node = node->rb_right;
355 else if (ret > 0)
356 node = node->rb_left;
357 else
358 return delayed_item;
359 }
360
361 if (prev) {
362 if (!prev_node)
363 *prev = NULL;
364 else if (ret < 0)
365 *prev = delayed_item;
366 else if ((node = rb_prev(prev_node)) != NULL) {
367 *prev = rb_entry(node, struct btrfs_delayed_item,
368 rb_node);
369 } else
370 *prev = NULL;
371 }
372
373 if (next) {
374 if (!prev_node)
375 *next = NULL;
376 else if (ret > 0)
377 *next = delayed_item;
378 else if ((node = rb_next(prev_node)) != NULL) {
379 *next = rb_entry(node, struct btrfs_delayed_item,
380 rb_node);
381 } else
382 *next = NULL;
383 }
384 return NULL;
385}
386
387static struct btrfs_delayed_item *__btrfs_lookup_delayed_insertion_item(
388 struct btrfs_delayed_node *delayed_node,
389 struct btrfs_key *key)
390{
391 struct btrfs_delayed_item *item;
392
393 item = __btrfs_lookup_delayed_item(&delayed_node->ins_root, key,
394 NULL, NULL);
395 return item;
396}
397
398static int __btrfs_add_delayed_item(struct btrfs_delayed_node *delayed_node,
399 struct btrfs_delayed_item *ins,
400 int action)
401{
402 struct rb_node **p, *node;
403 struct rb_node *parent_node = NULL;
404 struct rb_root *root;
405 struct btrfs_delayed_item *item;
406 int cmp;
407
408 if (action == BTRFS_DELAYED_INSERTION_ITEM)
409 root = &delayed_node->ins_root;
410 else if (action == BTRFS_DELAYED_DELETION_ITEM)
411 root = &delayed_node->del_root;
412 else
413 BUG();
414 p = &root->rb_node;
415 node = &ins->rb_node;
416
417 while (*p) {
418 parent_node = *p;
419 item = rb_entry(parent_node, struct btrfs_delayed_item,
420 rb_node);
421
422 cmp = btrfs_comp_cpu_keys(&item->key, &ins->key);
423 if (cmp < 0)
424 p = &(*p)->rb_right;
425 else if (cmp > 0)
426 p = &(*p)->rb_left;
427 else
428 return -EEXIST;
429 }
430
431 rb_link_node(node, parent_node, p);
432 rb_insert_color(node, root);
433 ins->delayed_node = delayed_node;
434 ins->ins_or_del = action;
435
436 if (ins->key.type == BTRFS_DIR_INDEX_KEY &&
437 action == BTRFS_DELAYED_INSERTION_ITEM &&
438 ins->key.offset >= delayed_node->index_cnt)
439 delayed_node->index_cnt = ins->key.offset + 1;
440
441 delayed_node->count++;
442 atomic_inc(&delayed_node->root->fs_info->delayed_root->items);
443 return 0;
444}
445
446static int __btrfs_add_delayed_insertion_item(struct btrfs_delayed_node *node,
447 struct btrfs_delayed_item *item)
448{
449 return __btrfs_add_delayed_item(node, item,
450 BTRFS_DELAYED_INSERTION_ITEM);
451}
452
453static int __btrfs_add_delayed_deletion_item(struct btrfs_delayed_node *node,
454 struct btrfs_delayed_item *item)
455{
456 return __btrfs_add_delayed_item(node, item,
457 BTRFS_DELAYED_DELETION_ITEM);
458}
459
460static void finish_one_item(struct btrfs_delayed_root *delayed_root)
461{
462 int seq = atomic_inc_return(&delayed_root->items_seq);
463 if ((atomic_dec_return(&delayed_root->items) <
464 BTRFS_DELAYED_BACKGROUND || seq % BTRFS_DELAYED_BATCH == 0) &&
465 waitqueue_active(&delayed_root->wait))
466 wake_up(&delayed_root->wait);
467}
468
469static void __btrfs_remove_delayed_item(struct btrfs_delayed_item *delayed_item)
470{
471 struct rb_root *root;
472 struct btrfs_delayed_root *delayed_root;
473
474 delayed_root = delayed_item->delayed_node->root->fs_info->delayed_root;
475
476 BUG_ON(!delayed_root);
477 BUG_ON(delayed_item->ins_or_del != BTRFS_DELAYED_DELETION_ITEM &&
478 delayed_item->ins_or_del != BTRFS_DELAYED_INSERTION_ITEM);
479
480 if (delayed_item->ins_or_del == BTRFS_DELAYED_INSERTION_ITEM)
481 root = &delayed_item->delayed_node->ins_root;
482 else
483 root = &delayed_item->delayed_node->del_root;
484
485 rb_erase(&delayed_item->rb_node, root);
486 delayed_item->delayed_node->count--;
487
488 finish_one_item(delayed_root);
489}
490
491static void btrfs_release_delayed_item(struct btrfs_delayed_item *item)
492{
493 if (item) {
494 __btrfs_remove_delayed_item(item);
495 if (atomic_dec_and_test(&item->refs))
496 kfree(item);
497 }
498}
499
500static struct btrfs_delayed_item *__btrfs_first_delayed_insertion_item(
501 struct btrfs_delayed_node *delayed_node)
502{
503 struct rb_node *p;
504 struct btrfs_delayed_item *item = NULL;
505
506 p = rb_first(&delayed_node->ins_root);
507 if (p)
508 item = rb_entry(p, struct btrfs_delayed_item, rb_node);
509
510 return item;
511}
512
513static struct btrfs_delayed_item *__btrfs_first_delayed_deletion_item(
514 struct btrfs_delayed_node *delayed_node)
515{
516 struct rb_node *p;
517 struct btrfs_delayed_item *item = NULL;
518
519 p = rb_first(&delayed_node->del_root);
520 if (p)
521 item = rb_entry(p, struct btrfs_delayed_item, rb_node);
522
523 return item;
524}
525
526static struct btrfs_delayed_item *__btrfs_next_delayed_item(
527 struct btrfs_delayed_item *item)
528{
529 struct rb_node *p;
530 struct btrfs_delayed_item *next = NULL;
531
532 p = rb_next(&item->rb_node);
533 if (p)
534 next = rb_entry(p, struct btrfs_delayed_item, rb_node);
535
536 return next;
537}
538
539static int btrfs_delayed_item_reserve_metadata(struct btrfs_trans_handle *trans,
540 struct btrfs_root *root,
541 struct btrfs_delayed_item *item)
542{
543 struct btrfs_block_rsv *src_rsv;
544 struct btrfs_block_rsv *dst_rsv;
545 u64 num_bytes;
546 int ret;
547
548 if (!trans->bytes_reserved)
549 return 0;
550
551 src_rsv = trans->block_rsv;
552 dst_rsv = &root->fs_info->delayed_block_rsv;
553
554 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
555 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes);
556 if (!ret) {
557 trace_btrfs_space_reservation(root->fs_info, "delayed_item",
558 item->key.objectid,
559 num_bytes, 1);
560 item->bytes_reserved = num_bytes;
561 }
562
563 return ret;
564}
565
566static void btrfs_delayed_item_release_metadata(struct btrfs_root *root,
567 struct btrfs_delayed_item *item)
568{
569 struct btrfs_block_rsv *rsv;
570
571 if (!item->bytes_reserved)
572 return;
573
574 rsv = &root->fs_info->delayed_block_rsv;
575 trace_btrfs_space_reservation(root->fs_info, "delayed_item",
576 item->key.objectid, item->bytes_reserved,
577 0);
578 btrfs_block_rsv_release(root, rsv,
579 item->bytes_reserved);
580}
581
582static int btrfs_delayed_inode_reserve_metadata(
583 struct btrfs_trans_handle *trans,
584 struct btrfs_root *root,
585 struct inode *inode,
586 struct btrfs_delayed_node *node)
587{
588 struct btrfs_block_rsv *src_rsv;
589 struct btrfs_block_rsv *dst_rsv;
590 u64 num_bytes;
591 int ret;
592 bool release = false;
593
594 src_rsv = trans->block_rsv;
595 dst_rsv = &root->fs_info->delayed_block_rsv;
596
597 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
598
599 /*
600 * btrfs_dirty_inode will update the inode under btrfs_join_transaction
601 * which doesn't reserve space for speed. This is a problem since we
602 * still need to reserve space for this update, so try to reserve the
603 * space.
604 *
605 * Now if src_rsv == delalloc_block_rsv we'll let it just steal since
606 * we're accounted for.
607 */
608 if (!src_rsv || (!trans->bytes_reserved &&
609 src_rsv->type != BTRFS_BLOCK_RSV_DELALLOC)) {
610 ret = btrfs_block_rsv_add(root, dst_rsv, num_bytes,
611 BTRFS_RESERVE_NO_FLUSH);
612 /*
613 * Since we're under a transaction reserve_metadata_bytes could
614 * try to commit the transaction which will make it return
615 * EAGAIN to make us stop the transaction we have, so return
616 * ENOSPC instead so that btrfs_dirty_inode knows what to do.
617 */
618 if (ret == -EAGAIN)
619 ret = -ENOSPC;
620 if (!ret) {
621 node->bytes_reserved = num_bytes;
622 trace_btrfs_space_reservation(root->fs_info,
623 "delayed_inode",
624 btrfs_ino(inode),
625 num_bytes, 1);
626 }
627 return ret;
628 } else if (src_rsv->type == BTRFS_BLOCK_RSV_DELALLOC) {
629 spin_lock(&BTRFS_I(inode)->lock);
630 if (test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
631 &BTRFS_I(inode)->runtime_flags)) {
632 spin_unlock(&BTRFS_I(inode)->lock);
633 release = true;
634 goto migrate;
635 }
636 spin_unlock(&BTRFS_I(inode)->lock);
637
638 /* Ok we didn't have space pre-reserved. This shouldn't happen
639 * too often but it can happen if we do delalloc to an existing
640 * inode which gets dirtied because of the time update, and then
641 * isn't touched again until after the transaction commits and
642 * then we try to write out the data. First try to be nice and
643 * reserve something strictly for us. If not be a pain and try
644 * to steal from the delalloc block rsv.
645 */
646 ret = btrfs_block_rsv_add(root, dst_rsv, num_bytes,
647 BTRFS_RESERVE_NO_FLUSH);
648 if (!ret)
649 goto out;
650
651 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes);
652 if (!WARN_ON(ret))
653 goto out;
654
655 /*
656 * Ok this is a problem, let's just steal from the global rsv
657 * since this really shouldn't happen that often.
658 */
659 ret = btrfs_block_rsv_migrate(&root->fs_info->global_block_rsv,
660 dst_rsv, num_bytes);
661 goto out;
662 }
663
664migrate:
665 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes);
666
667out:
668 /*
669 * Migrate only takes a reservation, it doesn't touch the size of the
670 * block_rsv. This is to simplify people who don't normally have things
671 * migrated from their block rsv. If they go to release their
672 * reservation, that will decrease the size as well, so if migrate
673 * reduced size we'd end up with a negative size. But for the
674 * delalloc_meta_reserved stuff we will only know to drop 1 reservation,
675 * but we could in fact do this reserve/migrate dance several times
676 * between the time we did the original reservation and we'd clean it
677 * up. So to take care of this, release the space for the meta
678 * reservation here. I think it may be time for a documentation page on
679 * how block rsvs. work.
680 */
681 if (!ret) {
682 trace_btrfs_space_reservation(root->fs_info, "delayed_inode",
683 btrfs_ino(inode), num_bytes, 1);
684 node->bytes_reserved = num_bytes;
685 }
686
687 if (release) {
688 trace_btrfs_space_reservation(root->fs_info, "delalloc",
689 btrfs_ino(inode), num_bytes, 0);
690 btrfs_block_rsv_release(root, src_rsv, num_bytes);
691 }
692
693 return ret;
694}
695
696static void btrfs_delayed_inode_release_metadata(struct btrfs_root *root,
697 struct btrfs_delayed_node *node)
698{
699 struct btrfs_block_rsv *rsv;
700
701 if (!node->bytes_reserved)
702 return;
703
704 rsv = &root->fs_info->delayed_block_rsv;
705 trace_btrfs_space_reservation(root->fs_info, "delayed_inode",
706 node->inode_id, node->bytes_reserved, 0);
707 btrfs_block_rsv_release(root, rsv,
708 node->bytes_reserved);
709 node->bytes_reserved = 0;
710}
711
712/*
713 * This helper will insert some continuous items into the same leaf according
714 * to the free space of the leaf.
715 */
716static int btrfs_batch_insert_items(struct btrfs_root *root,
717 struct btrfs_path *path,
718 struct btrfs_delayed_item *item)
719{
720 struct btrfs_delayed_item *curr, *next;
721 int free_space;
722 int total_data_size = 0, total_size = 0;
723 struct extent_buffer *leaf;
724 char *data_ptr;
725 struct btrfs_key *keys;
726 u32 *data_size;
727 struct list_head head;
728 int slot;
729 int nitems;
730 int i;
731 int ret = 0;
732
733 BUG_ON(!path->nodes[0]);
734
735 leaf = path->nodes[0];
736 free_space = btrfs_leaf_free_space(root, leaf);
737 INIT_LIST_HEAD(&head);
738
739 next = item;
740 nitems = 0;
741
742 /*
743 * count the number of the continuous items that we can insert in batch
744 */
745 while (total_size + next->data_len + sizeof(struct btrfs_item) <=
746 free_space) {
747 total_data_size += next->data_len;
748 total_size += next->data_len + sizeof(struct btrfs_item);
749 list_add_tail(&next->tree_list, &head);
750 nitems++;
751
752 curr = next;
753 next = __btrfs_next_delayed_item(curr);
754 if (!next)
755 break;
756
757 if (!btrfs_is_continuous_delayed_item(curr, next))
758 break;
759 }
760
761 if (!nitems) {
762 ret = 0;
763 goto out;
764 }
765
766 /*
767 * we need allocate some memory space, but it might cause the task
768 * to sleep, so we set all locked nodes in the path to blocking locks
769 * first.
770 */
771 btrfs_set_path_blocking(path);
772
773 keys = kmalloc_array(nitems, sizeof(struct btrfs_key), GFP_NOFS);
774 if (!keys) {
775 ret = -ENOMEM;
776 goto out;
777 }
778
779 data_size = kmalloc_array(nitems, sizeof(u32), GFP_NOFS);
780 if (!data_size) {
781 ret = -ENOMEM;
782 goto error;
783 }
784
785 /* get keys of all the delayed items */
786 i = 0;
787 list_for_each_entry(next, &head, tree_list) {
788 keys[i] = next->key;
789 data_size[i] = next->data_len;
790 i++;
791 }
792
793 /* reset all the locked nodes in the patch to spinning locks. */
794 btrfs_clear_path_blocking(path, NULL, 0);
795
796 /* insert the keys of the items */
797 setup_items_for_insert(root, path, keys, data_size,
798 total_data_size, total_size, nitems);
799
800 /* insert the dir index items */
801 slot = path->slots[0];
802 list_for_each_entry_safe(curr, next, &head, tree_list) {
803 data_ptr = btrfs_item_ptr(leaf, slot, char);
804 write_extent_buffer(leaf, &curr->data,
805 (unsigned long)data_ptr,
806 curr->data_len);
807 slot++;
808
809 btrfs_delayed_item_release_metadata(root, curr);
810
811 list_del(&curr->tree_list);
812 btrfs_release_delayed_item(curr);
813 }
814
815error:
816 kfree(data_size);
817 kfree(keys);
818out:
819 return ret;
820}
821
822/*
823 * This helper can just do simple insertion that needn't extend item for new
824 * data, such as directory name index insertion, inode insertion.
825 */
826static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans,
827 struct btrfs_root *root,
828 struct btrfs_path *path,
829 struct btrfs_delayed_item *delayed_item)
830{
831 struct extent_buffer *leaf;
832 char *ptr;
833 int ret;
834
835 ret = btrfs_insert_empty_item(trans, root, path, &delayed_item->key,
836 delayed_item->data_len);
837 if (ret < 0 && ret != -EEXIST)
838 return ret;
839
840 leaf = path->nodes[0];
841
842 ptr = btrfs_item_ptr(leaf, path->slots[0], char);
843
844 write_extent_buffer(leaf, delayed_item->data, (unsigned long)ptr,
845 delayed_item->data_len);
846 btrfs_mark_buffer_dirty(leaf);
847
848 btrfs_delayed_item_release_metadata(root, delayed_item);
849 return 0;
850}
851
852/*
853 * we insert an item first, then if there are some continuous items, we try
854 * to insert those items into the same leaf.
855 */
856static int btrfs_insert_delayed_items(struct btrfs_trans_handle *trans,
857 struct btrfs_path *path,
858 struct btrfs_root *root,
859 struct btrfs_delayed_node *node)
860{
861 struct btrfs_delayed_item *curr, *prev;
862 int ret = 0;
863
864do_again:
865 mutex_lock(&node->mutex);
866 curr = __btrfs_first_delayed_insertion_item(node);
867 if (!curr)
868 goto insert_end;
869
870 ret = btrfs_insert_delayed_item(trans, root, path, curr);
871 if (ret < 0) {
872 btrfs_release_path(path);
873 goto insert_end;
874 }
875
876 prev = curr;
877 curr = __btrfs_next_delayed_item(prev);
878 if (curr && btrfs_is_continuous_delayed_item(prev, curr)) {
879 /* insert the continuous items into the same leaf */
880 path->slots[0]++;
881 btrfs_batch_insert_items(root, path, curr);
882 }
883 btrfs_release_delayed_item(prev);
884 btrfs_mark_buffer_dirty(path->nodes[0]);
885
886 btrfs_release_path(path);
887 mutex_unlock(&node->mutex);
888 goto do_again;
889
890insert_end:
891 mutex_unlock(&node->mutex);
892 return ret;
893}
894
895static int btrfs_batch_delete_items(struct btrfs_trans_handle *trans,
896 struct btrfs_root *root,
897 struct btrfs_path *path,
898 struct btrfs_delayed_item *item)
899{
900 struct btrfs_delayed_item *curr, *next;
901 struct extent_buffer *leaf;
902 struct btrfs_key key;
903 struct list_head head;
904 int nitems, i, last_item;
905 int ret = 0;
906
907 BUG_ON(!path->nodes[0]);
908
909 leaf = path->nodes[0];
910
911 i = path->slots[0];
912 last_item = btrfs_header_nritems(leaf) - 1;
913 if (i > last_item)
914 return -ENOENT; /* FIXME: Is errno suitable? */
915
916 next = item;
917 INIT_LIST_HEAD(&head);
918 btrfs_item_key_to_cpu(leaf, &key, i);
919 nitems = 0;
920 /*
921 * count the number of the dir index items that we can delete in batch
922 */
923 while (btrfs_comp_cpu_keys(&next->key, &key) == 0) {
924 list_add_tail(&next->tree_list, &head);
925 nitems++;
926
927 curr = next;
928 next = __btrfs_next_delayed_item(curr);
929 if (!next)
930 break;
931
932 if (!btrfs_is_continuous_delayed_item(curr, next))
933 break;
934
935 i++;
936 if (i > last_item)
937 break;
938 btrfs_item_key_to_cpu(leaf, &key, i);
939 }
940
941 if (!nitems)
942 return 0;
943
944 ret = btrfs_del_items(trans, root, path, path->slots[0], nitems);
945 if (ret)
946 goto out;
947
948 list_for_each_entry_safe(curr, next, &head, tree_list) {
949 btrfs_delayed_item_release_metadata(root, curr);
950 list_del(&curr->tree_list);
951 btrfs_release_delayed_item(curr);
952 }
953
954out:
955 return ret;
956}
957
958static int btrfs_delete_delayed_items(struct btrfs_trans_handle *trans,
959 struct btrfs_path *path,
960 struct btrfs_root *root,
961 struct btrfs_delayed_node *node)
962{
963 struct btrfs_delayed_item *curr, *prev;
964 int ret = 0;
965
966do_again:
967 mutex_lock(&node->mutex);
968 curr = __btrfs_first_delayed_deletion_item(node);
969 if (!curr)
970 goto delete_fail;
971
972 ret = btrfs_search_slot(trans, root, &curr->key, path, -1, 1);
973 if (ret < 0)
974 goto delete_fail;
975 else if (ret > 0) {
976 /*
977 * can't find the item which the node points to, so this node
978 * is invalid, just drop it.
979 */
980 prev = curr;
981 curr = __btrfs_next_delayed_item(prev);
982 btrfs_release_delayed_item(prev);
983 ret = 0;
984 btrfs_release_path(path);
985 if (curr) {
986 mutex_unlock(&node->mutex);
987 goto do_again;
988 } else
989 goto delete_fail;
990 }
991
992 btrfs_batch_delete_items(trans, root, path, curr);
993 btrfs_release_path(path);
994 mutex_unlock(&node->mutex);
995 goto do_again;
996
997delete_fail:
998 btrfs_release_path(path);
999 mutex_unlock(&node->mutex);
1000 return ret;
1001}
1002
1003static void btrfs_release_delayed_inode(struct btrfs_delayed_node *delayed_node)
1004{
1005 struct btrfs_delayed_root *delayed_root;
1006
1007 if (delayed_node &&
1008 test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1009 BUG_ON(!delayed_node->root);
1010 clear_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
1011 delayed_node->count--;
1012
1013 delayed_root = delayed_node->root->fs_info->delayed_root;
1014 finish_one_item(delayed_root);
1015 }
1016}
1017
1018static void btrfs_release_delayed_iref(struct btrfs_delayed_node *delayed_node)
1019{
1020 struct btrfs_delayed_root *delayed_root;
1021
1022 ASSERT(delayed_node->root);
1023 clear_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
1024 delayed_node->count--;
1025
1026 delayed_root = delayed_node->root->fs_info->delayed_root;
1027 finish_one_item(delayed_root);
1028}
1029
1030static int __btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1031 struct btrfs_root *root,
1032 struct btrfs_path *path,
1033 struct btrfs_delayed_node *node)
1034{
1035 struct btrfs_key key;
1036 struct btrfs_inode_item *inode_item;
1037 struct extent_buffer *leaf;
1038 int mod;
1039 int ret;
1040
1041 key.objectid = node->inode_id;
1042 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
1043 key.offset = 0;
1044
1045 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1046 mod = -1;
1047 else
1048 mod = 1;
1049
1050 ret = btrfs_lookup_inode(trans, root, path, &key, mod);
1051 if (ret > 0) {
1052 btrfs_release_path(path);
1053 return -ENOENT;
1054 } else if (ret < 0) {
1055 return ret;
1056 }
1057
1058 leaf = path->nodes[0];
1059 inode_item = btrfs_item_ptr(leaf, path->slots[0],
1060 struct btrfs_inode_item);
1061 write_extent_buffer(leaf, &node->inode_item, (unsigned long)inode_item,
1062 sizeof(struct btrfs_inode_item));
1063 btrfs_mark_buffer_dirty(leaf);
1064
1065 if (!test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1066 goto no_iref;
1067
1068 path->slots[0]++;
1069 if (path->slots[0] >= btrfs_header_nritems(leaf))
1070 goto search;
1071again:
1072 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1073 if (key.objectid != node->inode_id)
1074 goto out;
1075
1076 if (key.type != BTRFS_INODE_REF_KEY &&
1077 key.type != BTRFS_INODE_EXTREF_KEY)
1078 goto out;
1079
1080 /*
1081 * Delayed iref deletion is for the inode who has only one link,
1082 * so there is only one iref. The case that several irefs are
1083 * in the same item doesn't exist.
1084 */
1085 btrfs_del_item(trans, root, path);
1086out:
1087 btrfs_release_delayed_iref(node);
1088no_iref:
1089 btrfs_release_path(path);
1090err_out:
1091 btrfs_delayed_inode_release_metadata(root, node);
1092 btrfs_release_delayed_inode(node);
1093
1094 return ret;
1095
1096search:
1097 btrfs_release_path(path);
1098
1099 btrfs_set_key_type(&key, BTRFS_INODE_EXTREF_KEY);
1100 key.offset = -1;
1101 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1102 if (ret < 0)
1103 goto err_out;
1104 ASSERT(ret);
1105
1106 ret = 0;
1107 leaf = path->nodes[0];
1108 path->slots[0]--;
1109 goto again;
1110}
1111
1112static inline int btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1113 struct btrfs_root *root,
1114 struct btrfs_path *path,
1115 struct btrfs_delayed_node *node)
1116{
1117 int ret;
1118
1119 mutex_lock(&node->mutex);
1120 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &node->flags)) {
1121 mutex_unlock(&node->mutex);
1122 return 0;
1123 }
1124
1125 ret = __btrfs_update_delayed_inode(trans, root, path, node);
1126 mutex_unlock(&node->mutex);
1127 return ret;
1128}
1129
1130static inline int
1131__btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1132 struct btrfs_path *path,
1133 struct btrfs_delayed_node *node)
1134{
1135 int ret;
1136
1137 ret = btrfs_insert_delayed_items(trans, path, node->root, node);
1138 if (ret)
1139 return ret;
1140
1141 ret = btrfs_delete_delayed_items(trans, path, node->root, node);
1142 if (ret)
1143 return ret;
1144
1145 ret = btrfs_update_delayed_inode(trans, node->root, path, node);
1146 return ret;
1147}
1148
1149/*
1150 * Called when committing the transaction.
1151 * Returns 0 on success.
1152 * Returns < 0 on error and returns with an aborted transaction with any
1153 * outstanding delayed items cleaned up.
1154 */
1155static int __btrfs_run_delayed_items(struct btrfs_trans_handle *trans,
1156 struct btrfs_root *root, int nr)
1157{
1158 struct btrfs_delayed_root *delayed_root;
1159 struct btrfs_delayed_node *curr_node, *prev_node;
1160 struct btrfs_path *path;
1161 struct btrfs_block_rsv *block_rsv;
1162 int ret = 0;
1163 bool count = (nr > 0);
1164
1165 if (trans->aborted)
1166 return -EIO;
1167
1168 path = btrfs_alloc_path();
1169 if (!path)
1170 return -ENOMEM;
1171 path->leave_spinning = 1;
1172
1173 block_rsv = trans->block_rsv;
1174 trans->block_rsv = &root->fs_info->delayed_block_rsv;
1175
1176 delayed_root = btrfs_get_delayed_root(root);
1177
1178 curr_node = btrfs_first_delayed_node(delayed_root);
1179 while (curr_node && (!count || (count && nr--))) {
1180 ret = __btrfs_commit_inode_delayed_items(trans, path,
1181 curr_node);
1182 if (ret) {
1183 btrfs_release_delayed_node(curr_node);
1184 curr_node = NULL;
1185 btrfs_abort_transaction(trans, root, ret);
1186 break;
1187 }
1188
1189 prev_node = curr_node;
1190 curr_node = btrfs_next_delayed_node(curr_node);
1191 btrfs_release_delayed_node(prev_node);
1192 }
1193
1194 if (curr_node)
1195 btrfs_release_delayed_node(curr_node);
1196 btrfs_free_path(path);
1197 trans->block_rsv = block_rsv;
1198
1199 return ret;
1200}
1201
1202int btrfs_run_delayed_items(struct btrfs_trans_handle *trans,
1203 struct btrfs_root *root)
1204{
1205 return __btrfs_run_delayed_items(trans, root, -1);
1206}
1207
1208int btrfs_run_delayed_items_nr(struct btrfs_trans_handle *trans,
1209 struct btrfs_root *root, int nr)
1210{
1211 return __btrfs_run_delayed_items(trans, root, nr);
1212}
1213
1214int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1215 struct inode *inode)
1216{
1217 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1218 struct btrfs_path *path;
1219 struct btrfs_block_rsv *block_rsv;
1220 int ret;
1221
1222 if (!delayed_node)
1223 return 0;
1224
1225 mutex_lock(&delayed_node->mutex);
1226 if (!delayed_node->count) {
1227 mutex_unlock(&delayed_node->mutex);
1228 btrfs_release_delayed_node(delayed_node);
1229 return 0;
1230 }
1231 mutex_unlock(&delayed_node->mutex);
1232
1233 path = btrfs_alloc_path();
1234 if (!path) {
1235 btrfs_release_delayed_node(delayed_node);
1236 return -ENOMEM;
1237 }
1238 path->leave_spinning = 1;
1239
1240 block_rsv = trans->block_rsv;
1241 trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv;
1242
1243 ret = __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1244
1245 btrfs_release_delayed_node(delayed_node);
1246 btrfs_free_path(path);
1247 trans->block_rsv = block_rsv;
1248
1249 return ret;
1250}
1251
1252int btrfs_commit_inode_delayed_inode(struct inode *inode)
1253{
1254 struct btrfs_trans_handle *trans;
1255 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1256 struct btrfs_path *path;
1257 struct btrfs_block_rsv *block_rsv;
1258 int ret;
1259
1260 if (!delayed_node)
1261 return 0;
1262
1263 mutex_lock(&delayed_node->mutex);
1264 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1265 mutex_unlock(&delayed_node->mutex);
1266 btrfs_release_delayed_node(delayed_node);
1267 return 0;
1268 }
1269 mutex_unlock(&delayed_node->mutex);
1270
1271 trans = btrfs_join_transaction(delayed_node->root);
1272 if (IS_ERR(trans)) {
1273 ret = PTR_ERR(trans);
1274 goto out;
1275 }
1276
1277 path = btrfs_alloc_path();
1278 if (!path) {
1279 ret = -ENOMEM;
1280 goto trans_out;
1281 }
1282 path->leave_spinning = 1;
1283
1284 block_rsv = trans->block_rsv;
1285 trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv;
1286
1287 mutex_lock(&delayed_node->mutex);
1288 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags))
1289 ret = __btrfs_update_delayed_inode(trans, delayed_node->root,
1290 path, delayed_node);
1291 else
1292 ret = 0;
1293 mutex_unlock(&delayed_node->mutex);
1294
1295 btrfs_free_path(path);
1296 trans->block_rsv = block_rsv;
1297trans_out:
1298 btrfs_end_transaction(trans, delayed_node->root);
1299 btrfs_btree_balance_dirty(delayed_node->root);
1300out:
1301 btrfs_release_delayed_node(delayed_node);
1302
1303 return ret;
1304}
1305
1306void btrfs_remove_delayed_node(struct inode *inode)
1307{
1308 struct btrfs_delayed_node *delayed_node;
1309
1310 delayed_node = ACCESS_ONCE(BTRFS_I(inode)->delayed_node);
1311 if (!delayed_node)
1312 return;
1313
1314 BTRFS_I(inode)->delayed_node = NULL;
1315 btrfs_release_delayed_node(delayed_node);
1316}
1317
1318struct btrfs_async_delayed_work {
1319 struct btrfs_delayed_root *delayed_root;
1320 int nr;
1321 struct btrfs_work work;
1322};
1323
1324static void btrfs_async_run_delayed_root(struct btrfs_work *work)
1325{
1326 struct btrfs_async_delayed_work *async_work;
1327 struct btrfs_delayed_root *delayed_root;
1328 struct btrfs_trans_handle *trans;
1329 struct btrfs_path *path;
1330 struct btrfs_delayed_node *delayed_node = NULL;
1331 struct btrfs_root *root;
1332 struct btrfs_block_rsv *block_rsv;
1333 int total_done = 0;
1334
1335 async_work = container_of(work, struct btrfs_async_delayed_work, work);
1336 delayed_root = async_work->delayed_root;
1337
1338 path = btrfs_alloc_path();
1339 if (!path)
1340 goto out;
1341
1342again:
1343 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND / 2)
1344 goto free_path;
1345
1346 delayed_node = btrfs_first_prepared_delayed_node(delayed_root);
1347 if (!delayed_node)
1348 goto free_path;
1349
1350 path->leave_spinning = 1;
1351 root = delayed_node->root;
1352
1353 trans = btrfs_join_transaction(root);
1354 if (IS_ERR(trans))
1355 goto release_path;
1356
1357 block_rsv = trans->block_rsv;
1358 trans->block_rsv = &root->fs_info->delayed_block_rsv;
1359
1360 __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1361
1362 trans->block_rsv = block_rsv;
1363 btrfs_end_transaction(trans, root);
1364 btrfs_btree_balance_dirty_nodelay(root);
1365
1366release_path:
1367 btrfs_release_path(path);
1368 total_done++;
1369
1370 btrfs_release_prepared_delayed_node(delayed_node);
1371 if (async_work->nr == 0 || total_done < async_work->nr)
1372 goto again;
1373
1374free_path:
1375 btrfs_free_path(path);
1376out:
1377 wake_up(&delayed_root->wait);
1378 kfree(async_work);
1379}
1380
1381
1382static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root,
1383 struct btrfs_root *root, int nr)
1384{
1385 struct btrfs_async_delayed_work *async_work;
1386
1387 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1388 return 0;
1389
1390 async_work = kmalloc(sizeof(*async_work), GFP_NOFS);
1391 if (!async_work)
1392 return -ENOMEM;
1393
1394 async_work->delayed_root = delayed_root;
1395 btrfs_init_work(&async_work->work, btrfs_async_run_delayed_root,
1396 NULL, NULL);
1397 async_work->nr = nr;
1398
1399 btrfs_queue_work(root->fs_info->delayed_workers, &async_work->work);
1400 return 0;
1401}
1402
1403void btrfs_assert_delayed_root_empty(struct btrfs_root *root)
1404{
1405 struct btrfs_delayed_root *delayed_root;
1406 delayed_root = btrfs_get_delayed_root(root);
1407 WARN_ON(btrfs_first_delayed_node(delayed_root));
1408}
1409
1410static int could_end_wait(struct btrfs_delayed_root *delayed_root, int seq)
1411{
1412 int val = atomic_read(&delayed_root->items_seq);
1413
1414 if (val < seq || val >= seq + BTRFS_DELAYED_BATCH)
1415 return 1;
1416
1417 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1418 return 1;
1419
1420 return 0;
1421}
1422
1423void btrfs_balance_delayed_items(struct btrfs_root *root)
1424{
1425 struct btrfs_delayed_root *delayed_root;
1426
1427 delayed_root = btrfs_get_delayed_root(root);
1428
1429 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1430 return;
1431
1432 if (atomic_read(&delayed_root->items) >= BTRFS_DELAYED_WRITEBACK) {
1433 int seq;
1434 int ret;
1435
1436 seq = atomic_read(&delayed_root->items_seq);
1437
1438 ret = btrfs_wq_run_delayed_node(delayed_root, root, 0);
1439 if (ret)
1440 return;
1441
1442 wait_event_interruptible(delayed_root->wait,
1443 could_end_wait(delayed_root, seq));
1444 return;
1445 }
1446
1447 btrfs_wq_run_delayed_node(delayed_root, root, BTRFS_DELAYED_BATCH);
1448}
1449
1450/* Will return 0 or -ENOMEM */
1451int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans,
1452 struct btrfs_root *root, const char *name,
1453 int name_len, struct inode *dir,
1454 struct btrfs_disk_key *disk_key, u8 type,
1455 u64 index)
1456{
1457 struct btrfs_delayed_node *delayed_node;
1458 struct btrfs_delayed_item *delayed_item;
1459 struct btrfs_dir_item *dir_item;
1460 int ret;
1461
1462 delayed_node = btrfs_get_or_create_delayed_node(dir);
1463 if (IS_ERR(delayed_node))
1464 return PTR_ERR(delayed_node);
1465
1466 delayed_item = btrfs_alloc_delayed_item(sizeof(*dir_item) + name_len);
1467 if (!delayed_item) {
1468 ret = -ENOMEM;
1469 goto release_node;
1470 }
1471
1472 delayed_item->key.objectid = btrfs_ino(dir);
1473 btrfs_set_key_type(&delayed_item->key, BTRFS_DIR_INDEX_KEY);
1474 delayed_item->key.offset = index;
1475
1476 dir_item = (struct btrfs_dir_item *)delayed_item->data;
1477 dir_item->location = *disk_key;
1478 btrfs_set_stack_dir_transid(dir_item, trans->transid);
1479 btrfs_set_stack_dir_data_len(dir_item, 0);
1480 btrfs_set_stack_dir_name_len(dir_item, name_len);
1481 btrfs_set_stack_dir_type(dir_item, type);
1482 memcpy((char *)(dir_item + 1), name, name_len);
1483
1484 ret = btrfs_delayed_item_reserve_metadata(trans, root, delayed_item);
1485 /*
1486 * we have reserved enough space when we start a new transaction,
1487 * so reserving metadata failure is impossible
1488 */
1489 BUG_ON(ret);
1490
1491
1492 mutex_lock(&delayed_node->mutex);
1493 ret = __btrfs_add_delayed_insertion_item(delayed_node, delayed_item);
1494 if (unlikely(ret)) {
1495 btrfs_err(root->fs_info, "err add delayed dir index item(name: %.*s) "
1496 "into the insertion tree of the delayed node"
1497 "(root id: %llu, inode id: %llu, errno: %d)",
1498 name_len, name, delayed_node->root->objectid,
1499 delayed_node->inode_id, ret);
1500 BUG();
1501 }
1502 mutex_unlock(&delayed_node->mutex);
1503
1504release_node:
1505 btrfs_release_delayed_node(delayed_node);
1506 return ret;
1507}
1508
1509static int btrfs_delete_delayed_insertion_item(struct btrfs_root *root,
1510 struct btrfs_delayed_node *node,
1511 struct btrfs_key *key)
1512{
1513 struct btrfs_delayed_item *item;
1514
1515 mutex_lock(&node->mutex);
1516 item = __btrfs_lookup_delayed_insertion_item(node, key);
1517 if (!item) {
1518 mutex_unlock(&node->mutex);
1519 return 1;
1520 }
1521
1522 btrfs_delayed_item_release_metadata(root, item);
1523 btrfs_release_delayed_item(item);
1524 mutex_unlock(&node->mutex);
1525 return 0;
1526}
1527
1528int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans,
1529 struct btrfs_root *root, struct inode *dir,
1530 u64 index)
1531{
1532 struct btrfs_delayed_node *node;
1533 struct btrfs_delayed_item *item;
1534 struct btrfs_key item_key;
1535 int ret;
1536
1537 node = btrfs_get_or_create_delayed_node(dir);
1538 if (IS_ERR(node))
1539 return PTR_ERR(node);
1540
1541 item_key.objectid = btrfs_ino(dir);
1542 btrfs_set_key_type(&item_key, BTRFS_DIR_INDEX_KEY);
1543 item_key.offset = index;
1544
1545 ret = btrfs_delete_delayed_insertion_item(root, node, &item_key);
1546 if (!ret)
1547 goto end;
1548
1549 item = btrfs_alloc_delayed_item(0);
1550 if (!item) {
1551 ret = -ENOMEM;
1552 goto end;
1553 }
1554
1555 item->key = item_key;
1556
1557 ret = btrfs_delayed_item_reserve_metadata(trans, root, item);
1558 /*
1559 * we have reserved enough space when we start a new transaction,
1560 * so reserving metadata failure is impossible.
1561 */
1562 BUG_ON(ret);
1563
1564 mutex_lock(&node->mutex);
1565 ret = __btrfs_add_delayed_deletion_item(node, item);
1566 if (unlikely(ret)) {
1567 btrfs_err(root->fs_info, "err add delayed dir index item(index: %llu) "
1568 "into the deletion tree of the delayed node"
1569 "(root id: %llu, inode id: %llu, errno: %d)",
1570 index, node->root->objectid, node->inode_id,
1571 ret);
1572 BUG();
1573 }
1574 mutex_unlock(&node->mutex);
1575end:
1576 btrfs_release_delayed_node(node);
1577 return ret;
1578}
1579
1580int btrfs_inode_delayed_dir_index_count(struct inode *inode)
1581{
1582 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1583
1584 if (!delayed_node)
1585 return -ENOENT;
1586
1587 /*
1588 * Since we have held i_mutex of this directory, it is impossible that
1589 * a new directory index is added into the delayed node and index_cnt
1590 * is updated now. So we needn't lock the delayed node.
1591 */
1592 if (!delayed_node->index_cnt) {
1593 btrfs_release_delayed_node(delayed_node);
1594 return -EINVAL;
1595 }
1596
1597 BTRFS_I(inode)->index_cnt = delayed_node->index_cnt;
1598 btrfs_release_delayed_node(delayed_node);
1599 return 0;
1600}
1601
1602void btrfs_get_delayed_items(struct inode *inode, struct list_head *ins_list,
1603 struct list_head *del_list)
1604{
1605 struct btrfs_delayed_node *delayed_node;
1606 struct btrfs_delayed_item *item;
1607
1608 delayed_node = btrfs_get_delayed_node(inode);
1609 if (!delayed_node)
1610 return;
1611
1612 mutex_lock(&delayed_node->mutex);
1613 item = __btrfs_first_delayed_insertion_item(delayed_node);
1614 while (item) {
1615 atomic_inc(&item->refs);
1616 list_add_tail(&item->readdir_list, ins_list);
1617 item = __btrfs_next_delayed_item(item);
1618 }
1619
1620 item = __btrfs_first_delayed_deletion_item(delayed_node);
1621 while (item) {
1622 atomic_inc(&item->refs);
1623 list_add_tail(&item->readdir_list, del_list);
1624 item = __btrfs_next_delayed_item(item);
1625 }
1626 mutex_unlock(&delayed_node->mutex);
1627 /*
1628 * This delayed node is still cached in the btrfs inode, so refs
1629 * must be > 1 now, and we needn't check it is going to be freed
1630 * or not.
1631 *
1632 * Besides that, this function is used to read dir, we do not
1633 * insert/delete delayed items in this period. So we also needn't
1634 * requeue or dequeue this delayed node.
1635 */
1636 atomic_dec(&delayed_node->refs);
1637}
1638
1639void btrfs_put_delayed_items(struct list_head *ins_list,
1640 struct list_head *del_list)
1641{
1642 struct btrfs_delayed_item *curr, *next;
1643
1644 list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1645 list_del(&curr->readdir_list);
1646 if (atomic_dec_and_test(&curr->refs))
1647 kfree(curr);
1648 }
1649
1650 list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1651 list_del(&curr->readdir_list);
1652 if (atomic_dec_and_test(&curr->refs))
1653 kfree(curr);
1654 }
1655}
1656
1657int btrfs_should_delete_dir_index(struct list_head *del_list,
1658 u64 index)
1659{
1660 struct btrfs_delayed_item *curr, *next;
1661 int ret;
1662
1663 if (list_empty(del_list))
1664 return 0;
1665
1666 list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1667 if (curr->key.offset > index)
1668 break;
1669
1670 list_del(&curr->readdir_list);
1671 ret = (curr->key.offset == index);
1672
1673 if (atomic_dec_and_test(&curr->refs))
1674 kfree(curr);
1675
1676 if (ret)
1677 return 1;
1678 else
1679 continue;
1680 }
1681 return 0;
1682}
1683
1684/*
1685 * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree
1686 *
1687 */
1688int btrfs_readdir_delayed_dir_index(struct dir_context *ctx,
1689 struct list_head *ins_list)
1690{
1691 struct btrfs_dir_item *di;
1692 struct btrfs_delayed_item *curr, *next;
1693 struct btrfs_key location;
1694 char *name;
1695 int name_len;
1696 int over = 0;
1697 unsigned char d_type;
1698
1699 if (list_empty(ins_list))
1700 return 0;
1701
1702 /*
1703 * Changing the data of the delayed item is impossible. So
1704 * we needn't lock them. And we have held i_mutex of the
1705 * directory, nobody can delete any directory indexes now.
1706 */
1707 list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1708 list_del(&curr->readdir_list);
1709
1710 if (curr->key.offset < ctx->pos) {
1711 if (atomic_dec_and_test(&curr->refs))
1712 kfree(curr);
1713 continue;
1714 }
1715
1716 ctx->pos = curr->key.offset;
1717
1718 di = (struct btrfs_dir_item *)curr->data;
1719 name = (char *)(di + 1);
1720 name_len = btrfs_stack_dir_name_len(di);
1721
1722 d_type = btrfs_filetype_table[di->type];
1723 btrfs_disk_key_to_cpu(&location, &di->location);
1724
1725 over = !dir_emit(ctx, name, name_len,
1726 location.objectid, d_type);
1727
1728 if (atomic_dec_and_test(&curr->refs))
1729 kfree(curr);
1730
1731 if (over)
1732 return 1;
1733 }
1734 return 0;
1735}
1736
1737static void fill_stack_inode_item(struct btrfs_trans_handle *trans,
1738 struct btrfs_inode_item *inode_item,
1739 struct inode *inode)
1740{
1741 btrfs_set_stack_inode_uid(inode_item, i_uid_read(inode));
1742 btrfs_set_stack_inode_gid(inode_item, i_gid_read(inode));
1743 btrfs_set_stack_inode_size(inode_item, BTRFS_I(inode)->disk_i_size);
1744 btrfs_set_stack_inode_mode(inode_item, inode->i_mode);
1745 btrfs_set_stack_inode_nlink(inode_item, inode->i_nlink);
1746 btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(inode));
1747 btrfs_set_stack_inode_generation(inode_item,
1748 BTRFS_I(inode)->generation);
1749 btrfs_set_stack_inode_sequence(inode_item, inode->i_version);
1750 btrfs_set_stack_inode_transid(inode_item, trans->transid);
1751 btrfs_set_stack_inode_rdev(inode_item, inode->i_rdev);
1752 btrfs_set_stack_inode_flags(inode_item, BTRFS_I(inode)->flags);
1753 btrfs_set_stack_inode_block_group(inode_item, 0);
1754
1755 btrfs_set_stack_timespec_sec(btrfs_inode_atime(inode_item),
1756 inode->i_atime.tv_sec);
1757 btrfs_set_stack_timespec_nsec(btrfs_inode_atime(inode_item),
1758 inode->i_atime.tv_nsec);
1759
1760 btrfs_set_stack_timespec_sec(btrfs_inode_mtime(inode_item),
1761 inode->i_mtime.tv_sec);
1762 btrfs_set_stack_timespec_nsec(btrfs_inode_mtime(inode_item),
1763 inode->i_mtime.tv_nsec);
1764
1765 btrfs_set_stack_timespec_sec(btrfs_inode_ctime(inode_item),
1766 inode->i_ctime.tv_sec);
1767 btrfs_set_stack_timespec_nsec(btrfs_inode_ctime(inode_item),
1768 inode->i_ctime.tv_nsec);
1769}
1770
1771int btrfs_fill_inode(struct inode *inode, u32 *rdev)
1772{
1773 struct btrfs_delayed_node *delayed_node;
1774 struct btrfs_inode_item *inode_item;
1775 struct btrfs_timespec *tspec;
1776
1777 delayed_node = btrfs_get_delayed_node(inode);
1778 if (!delayed_node)
1779 return -ENOENT;
1780
1781 mutex_lock(&delayed_node->mutex);
1782 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1783 mutex_unlock(&delayed_node->mutex);
1784 btrfs_release_delayed_node(delayed_node);
1785 return -ENOENT;
1786 }
1787
1788 inode_item = &delayed_node->inode_item;
1789
1790 i_uid_write(inode, btrfs_stack_inode_uid(inode_item));
1791 i_gid_write(inode, btrfs_stack_inode_gid(inode_item));
1792 btrfs_i_size_write(inode, btrfs_stack_inode_size(inode_item));
1793 inode->i_mode = btrfs_stack_inode_mode(inode_item);
1794 set_nlink(inode, btrfs_stack_inode_nlink(inode_item));
1795 inode_set_bytes(inode, btrfs_stack_inode_nbytes(inode_item));
1796 BTRFS_I(inode)->generation = btrfs_stack_inode_generation(inode_item);
1797 inode->i_version = btrfs_stack_inode_sequence(inode_item);
1798 inode->i_rdev = 0;
1799 *rdev = btrfs_stack_inode_rdev(inode_item);
1800 BTRFS_I(inode)->flags = btrfs_stack_inode_flags(inode_item);
1801
1802 tspec = btrfs_inode_atime(inode_item);
1803 inode->i_atime.tv_sec = btrfs_stack_timespec_sec(tspec);
1804 inode->i_atime.tv_nsec = btrfs_stack_timespec_nsec(tspec);
1805
1806 tspec = btrfs_inode_mtime(inode_item);
1807 inode->i_mtime.tv_sec = btrfs_stack_timespec_sec(tspec);
1808 inode->i_mtime.tv_nsec = btrfs_stack_timespec_nsec(tspec);
1809
1810 tspec = btrfs_inode_ctime(inode_item);
1811 inode->i_ctime.tv_sec = btrfs_stack_timespec_sec(tspec);
1812 inode->i_ctime.tv_nsec = btrfs_stack_timespec_nsec(tspec);
1813
1814 inode->i_generation = BTRFS_I(inode)->generation;
1815 BTRFS_I(inode)->index_cnt = (u64)-1;
1816
1817 mutex_unlock(&delayed_node->mutex);
1818 btrfs_release_delayed_node(delayed_node);
1819 return 0;
1820}
1821
1822int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans,
1823 struct btrfs_root *root, struct inode *inode)
1824{
1825 struct btrfs_delayed_node *delayed_node;
1826 int ret = 0;
1827
1828 delayed_node = btrfs_get_or_create_delayed_node(inode);
1829 if (IS_ERR(delayed_node))
1830 return PTR_ERR(delayed_node);
1831
1832 mutex_lock(&delayed_node->mutex);
1833 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1834 fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1835 goto release_node;
1836 }
1837
1838 ret = btrfs_delayed_inode_reserve_metadata(trans, root, inode,
1839 delayed_node);
1840 if (ret)
1841 goto release_node;
1842
1843 fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1844 set_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
1845 delayed_node->count++;
1846 atomic_inc(&root->fs_info->delayed_root->items);
1847release_node:
1848 mutex_unlock(&delayed_node->mutex);
1849 btrfs_release_delayed_node(delayed_node);
1850 return ret;
1851}
1852
1853int btrfs_delayed_delete_inode_ref(struct inode *inode)
1854{
1855 struct btrfs_delayed_node *delayed_node;
1856
1857 delayed_node = btrfs_get_or_create_delayed_node(inode);
1858 if (IS_ERR(delayed_node))
1859 return PTR_ERR(delayed_node);
1860
1861 /*
1862 * We don't reserve space for inode ref deletion is because:
1863 * - We ONLY do async inode ref deletion for the inode who has only
1864 * one link(i_nlink == 1), it means there is only one inode ref.
1865 * And in most case, the inode ref and the inode item are in the
1866 * same leaf, and we will deal with them at the same time.
1867 * Since we are sure we will reserve the space for the inode item,
1868 * it is unnecessary to reserve space for inode ref deletion.
1869 * - If the inode ref and the inode item are not in the same leaf,
1870 * We also needn't worry about enospc problem, because we reserve
1871 * much more space for the inode update than it needs.
1872 * - At the worst, we can steal some space from the global reservation.
1873 * It is very rare.
1874 */
1875 mutex_lock(&delayed_node->mutex);
1876 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1877 goto release_node;
1878
1879 set_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
1880 delayed_node->count++;
1881 atomic_inc(&BTRFS_I(inode)->root->fs_info->delayed_root->items);
1882release_node:
1883 mutex_unlock(&delayed_node->mutex);
1884 btrfs_release_delayed_node(delayed_node);
1885 return 0;
1886}
1887
1888static void __btrfs_kill_delayed_node(struct btrfs_delayed_node *delayed_node)
1889{
1890 struct btrfs_root *root = delayed_node->root;
1891 struct btrfs_delayed_item *curr_item, *prev_item;
1892
1893 mutex_lock(&delayed_node->mutex);
1894 curr_item = __btrfs_first_delayed_insertion_item(delayed_node);
1895 while (curr_item) {
1896 btrfs_delayed_item_release_metadata(root, curr_item);
1897 prev_item = curr_item;
1898 curr_item = __btrfs_next_delayed_item(prev_item);
1899 btrfs_release_delayed_item(prev_item);
1900 }
1901
1902 curr_item = __btrfs_first_delayed_deletion_item(delayed_node);
1903 while (curr_item) {
1904 btrfs_delayed_item_release_metadata(root, curr_item);
1905 prev_item = curr_item;
1906 curr_item = __btrfs_next_delayed_item(prev_item);
1907 btrfs_release_delayed_item(prev_item);
1908 }
1909
1910 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1911 btrfs_release_delayed_iref(delayed_node);
1912
1913 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1914 btrfs_delayed_inode_release_metadata(root, delayed_node);
1915 btrfs_release_delayed_inode(delayed_node);
1916 }
1917 mutex_unlock(&delayed_node->mutex);
1918}
1919
1920void btrfs_kill_delayed_inode_items(struct inode *inode)
1921{
1922 struct btrfs_delayed_node *delayed_node;
1923
1924 delayed_node = btrfs_get_delayed_node(inode);
1925 if (!delayed_node)
1926 return;
1927
1928 __btrfs_kill_delayed_node(delayed_node);
1929 btrfs_release_delayed_node(delayed_node);
1930}
1931
1932void btrfs_kill_all_delayed_nodes(struct btrfs_root *root)
1933{
1934 u64 inode_id = 0;
1935 struct btrfs_delayed_node *delayed_nodes[8];
1936 int i, n;
1937
1938 while (1) {
1939 spin_lock(&root->inode_lock);
1940 n = radix_tree_gang_lookup(&root->delayed_nodes_tree,
1941 (void **)delayed_nodes, inode_id,
1942 ARRAY_SIZE(delayed_nodes));
1943 if (!n) {
1944 spin_unlock(&root->inode_lock);
1945 break;
1946 }
1947
1948 inode_id = delayed_nodes[n - 1]->inode_id + 1;
1949
1950 for (i = 0; i < n; i++)
1951 atomic_inc(&delayed_nodes[i]->refs);
1952 spin_unlock(&root->inode_lock);
1953
1954 for (i = 0; i < n; i++) {
1955 __btrfs_kill_delayed_node(delayed_nodes[i]);
1956 btrfs_release_delayed_node(delayed_nodes[i]);
1957 }
1958 }
1959}
1960
1961void btrfs_destroy_delayed_inodes(struct btrfs_root *root)
1962{
1963 struct btrfs_delayed_root *delayed_root;
1964 struct btrfs_delayed_node *curr_node, *prev_node;
1965
1966 delayed_root = btrfs_get_delayed_root(root);
1967
1968 curr_node = btrfs_first_delayed_node(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