?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
ext4: fix oops on corrupted filesystem
[linux/fpc-iii.git] / fs / btrfs / delayed-inode.c
blob6cef0062f9292933435f8b6df6ff875e93a816aa
1 /*
2 * Copyright (C) 2011 Fujitsu. All rights reserved.
3 * Written by Miao Xie <miaox@cn.fujitsu.com>
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public
7 * License v2 as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12 * General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public
15 * License along with this program; if not, write to the
16 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
17 * Boston, MA 021110-1307, USA.
18 */
19
20 #include <linux/slab.h>
21 #include "delayed-inode.h"
22 #include "disk-io.h"
23 #include "transaction.h"
24 #include "ctree.h"
25
26 #define BTRFS_DELAYED_WRITEBACK 512
27 #define BTRFS_DELAYED_BACKGROUND 128
28 #define BTRFS_DELAYED_BATCH 16
29
30 static struct kmem_cache *delayed_node_cache;
31
32 int __init btrfs_delayed_inode_init(void)
33 {
34 delayed_node_cache = kmem_cache_create("btrfs_delayed_node",
35 sizeof(struct btrfs_delayed_node),
36 0,
37 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
38 NULL);
39 if (!delayed_node_cache)
40 return -ENOMEM;
41 return 0;
42 }
43
44 void btrfs_delayed_inode_exit(void)
45 {
46 kmem_cache_destroy(delayed_node_cache);
47 }
48
49 static inline void btrfs_init_delayed_node(
50 struct btrfs_delayed_node *delayed_node,
51 struct btrfs_root *root, u64 inode_id)
52 {
53 delayed_node->root = root;
54 delayed_node->inode_id = inode_id;
55 atomic_set(&delayed_node->refs, 0);
56 delayed_node->ins_root = RB_ROOT;
57 delayed_node->del_root = RB_ROOT;
58 mutex_init(&delayed_node->mutex);
59 INIT_LIST_HEAD(&delayed_node->n_list);
60 INIT_LIST_HEAD(&delayed_node->p_list);
61 }
62
63 static inline int btrfs_is_continuous_delayed_item(
64 struct btrfs_delayed_item *item1,
65 struct btrfs_delayed_item *item2)
66 {
67 if (item1->key.type == BTRFS_DIR_INDEX_KEY &&
68 item1->key.objectid == item2->key.objectid &&
69 item1->key.type == item2->key.type &&
70 item1->key.offset + 1 == item2->key.offset)
71 return 1;
72 return 0;
73 }
74
75 static inline struct btrfs_delayed_root *btrfs_get_delayed_root(
76 struct btrfs_root *root)
77 {
78 return root->fs_info->delayed_root;
79 }
80
81 static struct btrfs_delayed_node *btrfs_get_delayed_node(struct inode *inode)
82 {
83 struct btrfs_inode *btrfs_inode = BTRFS_I(inode);
84 struct btrfs_root *root = btrfs_inode->root;
85 u64 ino = btrfs_ino(inode);
86 struct btrfs_delayed_node *node;
87
88 node = ACCESS_ONCE(btrfs_inode->delayed_node);
89 if (node) {
90 atomic_inc(&node->refs);
91 return node;
92 }
93
94 spin_lock(&root->inode_lock);
95 node = radix_tree_lookup(&root->delayed_nodes_tree, ino);
96 if (node) {
97 if (btrfs_inode->delayed_node) {
98 atomic_inc(&node->refs); /* can be accessed */
99 BUG_ON(btrfs_inode->delayed_node != node);
100 spin_unlock(&root->inode_lock);
101 return node;
102 }
103 btrfs_inode->delayed_node = node;
104 /* can be accessed and cached in the inode */
105 atomic_add(2, &node->refs);
106 spin_unlock(&root->inode_lock);
107 return node;
108 }
109 spin_unlock(&root->inode_lock);
110
111 return NULL;
112 }
113
114 /* Will return either the node or PTR_ERR(-ENOMEM) */
115 static struct btrfs_delayed_node *btrfs_get_or_create_delayed_node(
116 struct inode *inode)
117 {
118 struct btrfs_delayed_node *node;
119 struct btrfs_inode *btrfs_inode = BTRFS_I(inode);
120 struct btrfs_root *root = btrfs_inode->root;
121 u64 ino = btrfs_ino(inode);
122 int ret;
123
124 again:
125 node = btrfs_get_delayed_node(inode);
126 if (node)
127 return node;
128
129 node = kmem_cache_zalloc(delayed_node_cache, GFP_NOFS);
130 if (!node)
131 return ERR_PTR(-ENOMEM);
132 btrfs_init_delayed_node(node, root, ino);
133
134 /* cached in the btrfs inode and can be accessed */
135 atomic_add(2, &node->refs);
136
137 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
138 if (ret) {
139 kmem_cache_free(delayed_node_cache, node);
140 return ERR_PTR(ret);
141 }
142
143 spin_lock(&root->inode_lock);
144 ret = radix_tree_insert(&root->delayed_nodes_tree, ino, node);
145 if (ret == -EEXIST) {
146 spin_unlock(&root->inode_lock);
147 kmem_cache_free(delayed_node_cache, node);
148 radix_tree_preload_end();
149 goto again;
150 }
151 btrfs_inode->delayed_node = node;
152 spin_unlock(&root->inode_lock);
153 radix_tree_preload_end();
154
155 return node;
156 }
157
158 /*
159 * Call it when holding delayed_node->mutex
160 *
161 * If mod = 1, add this node into the prepared list.
162 */
163 static void btrfs_queue_delayed_node(struct btrfs_delayed_root *root,
164 struct btrfs_delayed_node *node,
165 int mod)
166 {
167 spin_lock(&root->lock);
168 if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
169 if (!list_empty(&node->p_list))
170 list_move_tail(&node->p_list, &root->prepare_list);
171 else if (mod)
172 list_add_tail(&node->p_list, &root->prepare_list);
173 } else {
174 list_add_tail(&node->n_list, &root->node_list);
175 list_add_tail(&node->p_list, &root->prepare_list);
176 atomic_inc(&node->refs); /* inserted into list */
177 root->nodes++;
178 set_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
179 }
180 spin_unlock(&root->lock);
181 }
182
183 /* Call it when holding delayed_node->mutex */
184 static void btrfs_dequeue_delayed_node(struct btrfs_delayed_root *root,
185 struct btrfs_delayed_node *node)
186 {
187 spin_lock(&root->lock);
188 if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
189 root->nodes--;
190 atomic_dec(&node->refs); /* not in the list */
191 list_del_init(&node->n_list);
192 if (!list_empty(&node->p_list))
193 list_del_init(&node->p_list);
194 clear_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
195 }
196 spin_unlock(&root->lock);
197 }
198
199 static struct btrfs_delayed_node *btrfs_first_delayed_node(
200 struct btrfs_delayed_root *delayed_root)
201 {
202 struct list_head *p;
203 struct btrfs_delayed_node *node = NULL;
204
205 spin_lock(&delayed_root->lock);
206 if (list_empty(&delayed_root->node_list))
207 goto out;
208
209 p = delayed_root->node_list.next;
210 node = list_entry(p, struct btrfs_delayed_node, n_list);
211 atomic_inc(&node->refs);
212 out:
213 spin_unlock(&delayed_root->lock);
214
215 return node;
216 }
217
218 static struct btrfs_delayed_node *btrfs_next_delayed_node(
219 struct btrfs_delayed_node *node)
220 {
221 struct btrfs_delayed_root *delayed_root;
222 struct list_head *p;
223 struct btrfs_delayed_node *next = NULL;
224
225 delayed_root = node->root->fs_info->delayed_root;
226 spin_lock(&delayed_root->lock);
227 if (!test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
228 /* not in the list */
229 if (list_empty(&delayed_root->node_list))
230 goto out;
231 p = delayed_root->node_list.next;
232 } else if (list_is_last(&node->n_list, &delayed_root->node_list))
233 goto out;
234 else
235 p = node->n_list.next;
236
237 next = list_entry(p, struct btrfs_delayed_node, n_list);
238 atomic_inc(&next->refs);
239 out:
240 spin_unlock(&delayed_root->lock);
241
242 return next;
243 }
244
245 static void __btrfs_release_delayed_node(
246 struct btrfs_delayed_node *delayed_node,
247 int mod)
248 {
249 struct btrfs_delayed_root *delayed_root;
250
251 if (!delayed_node)
252 return;
253
254 delayed_root = delayed_node->root->fs_info->delayed_root;
255
256 mutex_lock(&delayed_node->mutex);
257 if (delayed_node->count)
258 btrfs_queue_delayed_node(delayed_root, delayed_node, mod);
259 else
260 btrfs_dequeue_delayed_node(delayed_root, delayed_node);
261 mutex_unlock(&delayed_node->mutex);
262
263 if (atomic_dec_and_test(&delayed_node->refs)) {
264 bool free = false;
265 struct btrfs_root *root = delayed_node->root;
266 spin_lock(&root->inode_lock);
267 if (atomic_read(&delayed_node->refs) == 0) {
268 radix_tree_delete(&root->delayed_nodes_tree,
269 delayed_node->inode_id);
270 free = true;
271 }
272 spin_unlock(&root->inode_lock);
273 if (free)
274 kmem_cache_free(delayed_node_cache, delayed_node);
275 }
276 }
277
278 static inline void btrfs_release_delayed_node(struct btrfs_delayed_node *node)
279 {
280 __btrfs_release_delayed_node(node, 0);
281 }
282
283 static struct btrfs_delayed_node *btrfs_first_prepared_delayed_node(
284 struct btrfs_delayed_root *delayed_root)
285 {
286 struct list_head *p;
287 struct btrfs_delayed_node *node = NULL;
288
289 spin_lock(&delayed_root->lock);
290 if (list_empty(&delayed_root->prepare_list))
291 goto out;
292
293 p = delayed_root->prepare_list.next;
294 list_del_init(p);
295 node = list_entry(p, struct btrfs_delayed_node, p_list);
296 atomic_inc(&node->refs);
297 out:
298 spin_unlock(&delayed_root->lock);
299
300 return node;
301 }
302
303 static inline void btrfs_release_prepared_delayed_node(
304 struct btrfs_delayed_node *node)
305 {
306 __btrfs_release_delayed_node(node, 1);
307 }
308
309 static struct btrfs_delayed_item *btrfs_alloc_delayed_item(u32 data_len)
310 {
311 struct btrfs_delayed_item *item;
312 item = kmalloc(sizeof(*item) + data_len, GFP_NOFS);
313 if (item) {
314 item->data_len = data_len;
315 item->ins_or_del = 0;
316 item->bytes_reserved = 0;
317 item->delayed_node = NULL;
318 atomic_set(&item->refs, 1);
319 }
320 return item;
321 }
322
323 /*
324 * __btrfs_lookup_delayed_item - look up the delayed item by key
325 * @delayed_node: pointer to the delayed node
326 * @key: the key to look up
327 * @prev: used to store the prev item if the right item isn't found
328 * @next: used to store the next item if the right item isn't found
329 *
330 * Note: if we don't find the right item, we will return the prev item and
331 * the next item.
332 */
333 static struct btrfs_delayed_item *__btrfs_lookup_delayed_item(
334 struct rb_root *root,
335 struct btrfs_key *key,
336 struct btrfs_delayed_item **prev,
337 struct btrfs_delayed_item **next)
338 {
339 struct rb_node *node, *prev_node = NULL;
340 struct btrfs_delayed_item *delayed_item = NULL;
341 int ret = 0;
342
343 node = root->rb_node;
344
345 while (node) {
346 delayed_item = rb_entry(node, struct btrfs_delayed_item,
347 rb_node);
348 prev_node = node;
349 ret = btrfs_comp_cpu_keys(&delayed_item->key, key);
350 if (ret < 0)
351 node = node->rb_right;
352 else if (ret > 0)
353 node = node->rb_left;
354 else
355 return delayed_item;
356 }
357
358 if (prev) {
359 if (!prev_node)
360 *prev = NULL;
361 else if (ret < 0)
362 *prev = delayed_item;
363 else if ((node = rb_prev(prev_node)) != NULL) {
364 *prev = rb_entry(node, struct btrfs_delayed_item,
365 rb_node);
366 } else
367 *prev = NULL;
368 }
369
370 if (next) {
371 if (!prev_node)
372 *next = NULL;
373 else if (ret > 0)
374 *next = delayed_item;
375 else if ((node = rb_next(prev_node)) != NULL) {
376 *next = rb_entry(node, struct btrfs_delayed_item,
377 rb_node);
378 } else
379 *next = NULL;
380 }
381 return NULL;
382 }
383
384 static struct btrfs_delayed_item *__btrfs_lookup_delayed_insertion_item(
385 struct btrfs_delayed_node *delayed_node,
386 struct btrfs_key *key)
387 {
388 struct btrfs_delayed_item *item;
389
390 item = __btrfs_lookup_delayed_item(&delayed_node->ins_root, key,
391 NULL, NULL);
392 return item;
393 }
394
395 static int __btrfs_add_delayed_item(struct btrfs_delayed_node *delayed_node,
396 struct btrfs_delayed_item *ins,
397 int action)
398 {
399 struct rb_node **p, *node;
400 struct rb_node *parent_node = NULL;
401 struct rb_root *root;
402 struct btrfs_delayed_item *item;
403 int cmp;
404
405 if (action == BTRFS_DELAYED_INSERTION_ITEM)
406 root = &delayed_node->ins_root;
407 else if (action == BTRFS_DELAYED_DELETION_ITEM)
408 root = &delayed_node->del_root;
409 else
410 BUG();
411 p = &root->rb_node;
412 node = &ins->rb_node;
413
414 while (*p) {
415 parent_node = *p;
416 item = rb_entry(parent_node, struct btrfs_delayed_item,
417 rb_node);
418
419 cmp = btrfs_comp_cpu_keys(&item->key, &ins->key);
420 if (cmp < 0)
421 p = &(*p)->rb_right;
422 else if (cmp > 0)
423 p = &(*p)->rb_left;
424 else
425 return -EEXIST;
426 }
427
428 rb_link_node(node, parent_node, p);
429 rb_insert_color(node, root);
430 ins->delayed_node = delayed_node;
431 ins->ins_or_del = action;
432
433 if (ins->key.type == BTRFS_DIR_INDEX_KEY &&
434 action == BTRFS_DELAYED_INSERTION_ITEM &&
435 ins->key.offset >= delayed_node->index_cnt)
436 delayed_node->index_cnt = ins->key.offset + 1;
437
438 delayed_node->count++;
439 atomic_inc(&delayed_node->root->fs_info->delayed_root->items);
440 return 0;
441 }
442
443 static int __btrfs_add_delayed_insertion_item(struct btrfs_delayed_node *node,
444 struct btrfs_delayed_item *item)
445 {
446 return __btrfs_add_delayed_item(node, item,
447 BTRFS_DELAYED_INSERTION_ITEM);
448 }
449
450 static int __btrfs_add_delayed_deletion_item(struct btrfs_delayed_node *node,
451 struct btrfs_delayed_item *item)
452 {
453 return __btrfs_add_delayed_item(node, item,
454 BTRFS_DELAYED_DELETION_ITEM);
455 }
456
457 static void finish_one_item(struct btrfs_delayed_root *delayed_root)
458 {
459 int seq = atomic_inc_return(&delayed_root->items_seq);
460
461 /*
462 * atomic_dec_return implies a barrier for waitqueue_active
463 */
464 if ((atomic_dec_return(&delayed_root->items) <
465 BTRFS_DELAYED_BACKGROUND || seq % BTRFS_DELAYED_BATCH == 0) &&
466 waitqueue_active(&delayed_root->wait))
467 wake_up(&delayed_root->wait);
468 }
469
470 static void __btrfs_remove_delayed_item(struct btrfs_delayed_item *delayed_item)
471 {
472 struct rb_root *root;
473 struct btrfs_delayed_root *delayed_root;
474
475 delayed_root = delayed_item->delayed_node->root->fs_info->delayed_root;
476
477 BUG_ON(!delayed_root);
478 BUG_ON(delayed_item->ins_or_del != BTRFS_DELAYED_DELETION_ITEM &&
479 delayed_item->ins_or_del != BTRFS_DELAYED_INSERTION_ITEM);
480
481 if (delayed_item->ins_or_del == BTRFS_DELAYED_INSERTION_ITEM)
482 root = &delayed_item->delayed_node->ins_root;
483 else
484 root = &delayed_item->delayed_node->del_root;
485
486 rb_erase(&delayed_item->rb_node, root);
487 delayed_item->delayed_node->count--;
488
489 finish_one_item(delayed_root);
490 }
491
492 static void btrfs_release_delayed_item(struct btrfs_delayed_item *item)
493 {
494 if (item) {
495 __btrfs_remove_delayed_item(item);
496 if (atomic_dec_and_test(&item->refs))
497 kfree(item);
498 }
499 }
500
501 static struct btrfs_delayed_item *__btrfs_first_delayed_insertion_item(
502 struct btrfs_delayed_node *delayed_node)
503 {
504 struct rb_node *p;
505 struct btrfs_delayed_item *item = NULL;
506
507 p = rb_first(&delayed_node->ins_root);
508 if (p)
509 item = rb_entry(p, struct btrfs_delayed_item, rb_node);
510
511 return item;
512 }
513
514 static struct btrfs_delayed_item *__btrfs_first_delayed_deletion_item(
515 struct btrfs_delayed_node *delayed_node)
516 {
517 struct rb_node *p;
518 struct btrfs_delayed_item *item = NULL;
519
520 p = rb_first(&delayed_node->del_root);
521 if (p)
522 item = rb_entry(p, struct btrfs_delayed_item, rb_node);
523
524 return item;
525 }
526
527 static struct btrfs_delayed_item *__btrfs_next_delayed_item(
528 struct btrfs_delayed_item *item)
529 {
530 struct rb_node *p;
531 struct btrfs_delayed_item *next = NULL;
532
533 p = rb_next(&item->rb_node);
534 if (p)
535 next = rb_entry(p, struct btrfs_delayed_item, rb_node);
536
537 return next;
538 }
539
540 static int btrfs_delayed_item_reserve_metadata(struct btrfs_trans_handle *trans,
541 struct btrfs_root *root,
542 struct btrfs_delayed_item *item)
543 {
544 struct btrfs_block_rsv *src_rsv;
545 struct btrfs_block_rsv *dst_rsv;
546 u64 num_bytes;
547 int ret;
548
549 if (!trans->bytes_reserved)
550 return 0;
551
552 src_rsv = trans->block_rsv;
553 dst_rsv = &root->fs_info->delayed_block_rsv;
554
555 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
556 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes);
557 if (!ret) {
558 trace_btrfs_space_reservation(root->fs_info, "delayed_item",
559 item->key.objectid,
560 num_bytes, 1);
561 item->bytes_reserved = num_bytes;
562 }
563
564 return ret;
565 }
566
567 static void btrfs_delayed_item_release_metadata(struct btrfs_root *root,
568 struct btrfs_delayed_item *item)
569 {
570 struct btrfs_block_rsv *rsv;
571
572 if (!item->bytes_reserved)
573 return;
574
575 rsv = &root->fs_info->delayed_block_rsv;
576 trace_btrfs_space_reservation(root->fs_info, "delayed_item",
577 item->key.objectid, item->bytes_reserved,
578 0);
579 btrfs_block_rsv_release(root, rsv,
580 item->bytes_reserved);
581 }
582
583 static int btrfs_delayed_inode_reserve_metadata(
584 struct btrfs_trans_handle *trans,
585 struct btrfs_root *root,
586 struct inode *inode,
587 struct btrfs_delayed_node *node)
588 {
589 struct btrfs_block_rsv *src_rsv;
590 struct btrfs_block_rsv *dst_rsv;
591 u64 num_bytes;
592 int ret;
593 bool release = false;
594
595 src_rsv = trans->block_rsv;
596 dst_rsv = &root->fs_info->delayed_block_rsv;
597
598 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
599
600 /*
601 * btrfs_dirty_inode will update the inode under btrfs_join_transaction
602 * which doesn't reserve space for speed. This is a problem since we
603 * still need to reserve space for this update, so try to reserve the
604 * space.
605 *
606 * Now if src_rsv == delalloc_block_rsv we'll let it just steal since
607 * we're accounted for.
608 */
609 if (!src_rsv || (!trans->bytes_reserved &&
610 src_rsv->type != BTRFS_BLOCK_RSV_DELALLOC)) {
611 ret = btrfs_block_rsv_add(root, dst_rsv, num_bytes,
612 BTRFS_RESERVE_NO_FLUSH);
613 /*
614 * Since we're under a transaction reserve_metadata_bytes could
615 * try to commit the transaction which will make it return
616 * EAGAIN to make us stop the transaction we have, so return
617 * ENOSPC instead so that btrfs_dirty_inode knows what to do.
618 */
619 if (ret == -EAGAIN)
620 ret = -ENOSPC;
621 if (!ret) {
622 node->bytes_reserved = num_bytes;
623 trace_btrfs_space_reservation(root->fs_info,
624 "delayed_inode",
625 btrfs_ino(inode),
626 num_bytes, 1);
627 }
628 return ret;
629 } else if (src_rsv->type == BTRFS_BLOCK_RSV_DELALLOC) {
630 spin_lock(&BTRFS_I(inode)->lock);
631 if (test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
632 &BTRFS_I(inode)->runtime_flags)) {
633 spin_unlock(&BTRFS_I(inode)->lock);
634 release = true;
635 goto migrate;
636 }
637 spin_unlock(&BTRFS_I(inode)->lock);
638
639 /* Ok we didn't have space pre-reserved. This shouldn't happen
640 * too often but it can happen if we do delalloc to an existing
641 * inode which gets dirtied because of the time update, and then
642 * isn't touched again until after the transaction commits and
643 * then we try to write out the data. First try to be nice and
644 * reserve something strictly for us. If not be a pain and try
645 * to steal from the delalloc block rsv.
646 */
647 ret = btrfs_block_rsv_add(root, dst_rsv, num_bytes,
648 BTRFS_RESERVE_NO_FLUSH);
649 if (!ret)
650 goto out;
651
652 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes);
653 if (!ret)
654 goto out;
655
656 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
657 btrfs_debug(root->fs_info,
658 "block rsv migrate returned %d", ret);
659 WARN_ON(1);
660 }
661 /*
662 * Ok this is a problem, let's just steal from the global rsv
663 * since this really shouldn't happen that often.
664 */
665 ret = btrfs_block_rsv_migrate(&root->fs_info->global_block_rsv,
666 dst_rsv, num_bytes);
667 goto out;
668 }
669
670 migrate:
671 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes);
672
673 out:
674 /*
675 * Migrate only takes a reservation, it doesn't touch the size of the
676 * block_rsv. This is to simplify people who don't normally have things
677 * migrated from their block rsv. If they go to release their
678 * reservation, that will decrease the size as well, so if migrate
679 * reduced size we'd end up with a negative size. But for the
680 * delalloc_meta_reserved stuff we will only know to drop 1 reservation,
681 * but we could in fact do this reserve/migrate dance several times
682 * between the time we did the original reservation and we'd clean it
683 * up. So to take care of this, release the space for the meta
684 * reservation here. I think it may be time for a documentation page on
685 * how block rsvs. work.
686 */
687 if (!ret) {
688 trace_btrfs_space_reservation(root->fs_info, "delayed_inode",
689 btrfs_ino(inode), num_bytes, 1);
690 node->bytes_reserved = num_bytes;
691 }
692
693 if (release) {
694 trace_btrfs_space_reservation(root->fs_info, "delalloc",
695 btrfs_ino(inode), num_bytes, 0);
696 btrfs_block_rsv_release(root, src_rsv, num_bytes);
697 }
698
699 return ret;
700 }
701
702 static void btrfs_delayed_inode_release_metadata(struct btrfs_root *root,
703 struct btrfs_delayed_node *node)
704 {
705 struct btrfs_block_rsv *rsv;
706
707 if (!node->bytes_reserved)
708 return;
709
710 rsv = &root->fs_info->delayed_block_rsv;
711 trace_btrfs_space_reservation(root->fs_info, "delayed_inode",
712 node->inode_id, node->bytes_reserved, 0);
713 btrfs_block_rsv_release(root, rsv,
714 node->bytes_reserved);
715 node->bytes_reserved = 0;
716 }
717
718 /*
719 * This helper will insert some continuous items into the same leaf according
720 * to the free space of the leaf.
721 */
722 static int btrfs_batch_insert_items(struct btrfs_root *root,
723 struct btrfs_path *path,
724 struct btrfs_delayed_item *item)
725 {
726 struct btrfs_delayed_item *curr, *next;
727 int free_space;
728 int total_data_size = 0, total_size = 0;
729 struct extent_buffer *leaf;
730 char *data_ptr;
731 struct btrfs_key *keys;
732 u32 *data_size;
733 struct list_head head;
734 int slot;
735 int nitems;
736 int i;
737 int ret = 0;
738
739 BUG_ON(!path->nodes[0]);
740
741 leaf = path->nodes[0];
742 free_space = btrfs_leaf_free_space(root, leaf);
743 INIT_LIST_HEAD(&head);
744
745 next = item;
746 nitems = 0;
747
748 /*
749 * count the number of the continuous items that we can insert in batch
750 */
751 while (total_size + next->data_len + sizeof(struct btrfs_item) <=
752 free_space) {
753 total_data_size += next->data_len;
754 total_size += next->data_len + sizeof(struct btrfs_item);
755 list_add_tail(&next->tree_list, &head);
756 nitems++;
757
758 curr = next;
759 next = __btrfs_next_delayed_item(curr);
760 if (!next)
761 break;
762
763 if (!btrfs_is_continuous_delayed_item(curr, next))
764 break;
765 }
766
767 if (!nitems) {
768 ret = 0;
769 goto out;
770 }
771
772 /*
773 * we need allocate some memory space, but it might cause the task
774 * to sleep, so we set all locked nodes in the path to blocking locks
775 * first.
776 */
777 btrfs_set_path_blocking(path);
778
779 keys = kmalloc_array(nitems, sizeof(struct btrfs_key), GFP_NOFS);
780 if (!keys) {
781 ret = -ENOMEM;
782 goto out;
783 }
784
785 data_size = kmalloc_array(nitems, sizeof(u32), GFP_NOFS);
786 if (!data_size) {
787 ret = -ENOMEM;
788 goto error;
789 }
790
791 /* get keys of all the delayed items */
792 i = 0;
793 list_for_each_entry(next, &head, tree_list) {
794 keys[i] = next->key;
795 data_size[i] = next->data_len;
796 i++;
797 }
798
799 /* reset all the locked nodes in the patch to spinning locks. */
800 btrfs_clear_path_blocking(path, NULL, 0);
801
802 /* insert the keys of the items */
803 setup_items_for_insert(root, path, keys, data_size,
804 total_data_size, total_size, nitems);
805
806 /* insert the dir index items */
807 slot = path->slots[0];
808 list_for_each_entry_safe(curr, next, &head, tree_list) {
809 data_ptr = btrfs_item_ptr(leaf, slot, char);
810 write_extent_buffer(leaf, &curr->data,
811 (unsigned long)data_ptr,
812 curr->data_len);
813 slot++;
814
815 btrfs_delayed_item_release_metadata(root, curr);
816
817 list_del(&curr->tree_list);
818 btrfs_release_delayed_item(curr);
819 }
820
821 error:
822 kfree(data_size);
823 kfree(keys);
824 out:
825 return ret;
826 }
827
828 /*
829 * This helper can just do simple insertion that needn't extend item for new
830 * data, such as directory name index insertion, inode insertion.
831 */
832 static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans,
833 struct btrfs_root *root,
834 struct btrfs_path *path,
835 struct btrfs_delayed_item *delayed_item)
836 {
837 struct extent_buffer *leaf;
838 char *ptr;
839 int ret;
840
841 ret = btrfs_insert_empty_item(trans, root, path, &delayed_item->key,
842 delayed_item->data_len);
843 if (ret < 0 && ret != -EEXIST)
844 return ret;
845
846 leaf = path->nodes[0];
847
848 ptr = btrfs_item_ptr(leaf, path->slots[0], char);
849
850 write_extent_buffer(leaf, delayed_item->data, (unsigned long)ptr,
851 delayed_item->data_len);
852 btrfs_mark_buffer_dirty(leaf);
853
854 btrfs_delayed_item_release_metadata(root, delayed_item);
855 return 0;
856 }
857
858 /*
859 * we insert an item first, then if there are some continuous items, we try
860 * to insert those items into the same leaf.
861 */
862 static int btrfs_insert_delayed_items(struct btrfs_trans_handle *trans,
863 struct btrfs_path *path,
864 struct btrfs_root *root,
865 struct btrfs_delayed_node *node)
866 {
867 struct btrfs_delayed_item *curr, *prev;
868 int ret = 0;
869
870 do_again:
871 mutex_lock(&node->mutex);
872 curr = __btrfs_first_delayed_insertion_item(node);
873 if (!curr)
874 goto insert_end;
875
876 ret = btrfs_insert_delayed_item(trans, root, path, curr);
877 if (ret < 0) {
878 btrfs_release_path(path);
879 goto insert_end;
880 }
881
882 prev = curr;
883 curr = __btrfs_next_delayed_item(prev);
884 if (curr && btrfs_is_continuous_delayed_item(prev, curr)) {
885 /* insert the continuous items into the same leaf */
886 path->slots[0]++;
887 btrfs_batch_insert_items(root, path, curr);
888 }
889 btrfs_release_delayed_item(prev);
890 btrfs_mark_buffer_dirty(path->nodes[0]);
891
892 btrfs_release_path(path);
893 mutex_unlock(&node->mutex);
894 goto do_again;
895
896 insert_end:
897 mutex_unlock(&node->mutex);
898 return ret;
899 }
900
901 static int btrfs_batch_delete_items(struct btrfs_trans_handle *trans,
902 struct btrfs_root *root,
903 struct btrfs_path *path,
904 struct btrfs_delayed_item *item)
905 {
906 struct btrfs_delayed_item *curr, *next;
907 struct extent_buffer *leaf;
908 struct btrfs_key key;
909 struct list_head head;
910 int nitems, i, last_item;
911 int ret = 0;
912
913 BUG_ON(!path->nodes[0]);
914
915 leaf = path->nodes[0];
916
917 i = path->slots[0];
918 last_item = btrfs_header_nritems(leaf) - 1;
919 if (i > last_item)
920 return -ENOENT; /* FIXME: Is errno suitable? */
921
922 next = item;
923 INIT_LIST_HEAD(&head);
924 btrfs_item_key_to_cpu(leaf, &key, i);
925 nitems = 0;
926 /*
927 * count the number of the dir index items that we can delete in batch
928 */
929 while (btrfs_comp_cpu_keys(&next->key, &key) == 0) {
930 list_add_tail(&next->tree_list, &head);
931 nitems++;
932
933 curr = next;
934 next = __btrfs_next_delayed_item(curr);
935 if (!next)
936 break;
937
938 if (!btrfs_is_continuous_delayed_item(curr, next))
939 break;
940
941 i++;
942 if (i > last_item)
943 break;
944 btrfs_item_key_to_cpu(leaf, &key, i);
945 }
946
947 if (!nitems)
948 return 0;
949
950 ret = btrfs_del_items(trans, root, path, path->slots[0], nitems);
951 if (ret)
952 goto out;
953
954 list_for_each_entry_safe(curr, next, &head, tree_list) {
955 btrfs_delayed_item_release_metadata(root, curr);
956 list_del(&curr->tree_list);
957 btrfs_release_delayed_item(curr);
958 }
959
960 out:
961 return ret;
962 }
963
964 static int btrfs_delete_delayed_items(struct btrfs_trans_handle *trans,
965 struct btrfs_path *path,
966 struct btrfs_root *root,
967 struct btrfs_delayed_node *node)
968 {
969 struct btrfs_delayed_item *curr, *prev;
970 int ret = 0;
971
972 do_again:
973 mutex_lock(&node->mutex);
974 curr = __btrfs_first_delayed_deletion_item(node);
975 if (!curr)
976 goto delete_fail;
977
978 ret = btrfs_search_slot(trans, root, &curr->key, path, -1, 1);
979 if (ret < 0)
980 goto delete_fail;
981 else if (ret > 0) {
982 /*
983 * can't find the item which the node points to, so this node
984 * is invalid, just drop it.
985 */
986 prev = curr;
987 curr = __btrfs_next_delayed_item(prev);
988 btrfs_release_delayed_item(prev);
989 ret = 0;
990 btrfs_release_path(path);
991 if (curr) {
992 mutex_unlock(&node->mutex);
993 goto do_again;
994 } else
995 goto delete_fail;
996 }
997
998 btrfs_batch_delete_items(trans, root, path, curr);
999 btrfs_release_path(path);
1000 mutex_unlock(&node->mutex);
1001 goto do_again;
1002
1003 delete_fail:
1004 btrfs_release_path(path);
1005 mutex_unlock(&node->mutex);
1006 return ret;
1007 }
1008
1009 static void btrfs_release_delayed_inode(struct btrfs_delayed_node *delayed_node)
1010 {
1011 struct btrfs_delayed_root *delayed_root;
1012
1013 if (delayed_node &&
1014 test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1015 BUG_ON(!delayed_node->root);
1016 clear_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
1017 delayed_node->count--;
1018
1019 delayed_root = delayed_node->root->fs_info->delayed_root;
1020 finish_one_item(delayed_root);
1021 }
1022 }
1023
1024 static void btrfs_release_delayed_iref(struct btrfs_delayed_node *delayed_node)
1025 {
1026 struct btrfs_delayed_root *delayed_root;
1027
1028 ASSERT(delayed_node->root);
1029 clear_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
1030 delayed_node->count--;
1031
1032 delayed_root = delayed_node->root->fs_info->delayed_root;
1033 finish_one_item(delayed_root);
1034 }
1035
1036 static int __btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1037 struct btrfs_root *root,
1038 struct btrfs_path *path,
1039 struct btrfs_delayed_node *node)
1040 {
1041 struct btrfs_key key;
1042 struct btrfs_inode_item *inode_item;
1043 struct extent_buffer *leaf;
1044 int mod;
1045 int ret;
1046
1047 key.objectid = node->inode_id;
1048 key.type = BTRFS_INODE_ITEM_KEY;
1049 key.offset = 0;
1050
1051 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1052 mod = -1;
1053 else
1054 mod = 1;
1055
1056 ret = btrfs_lookup_inode(trans, root, path, &key, mod);
1057 if (ret > 0) {
1058 btrfs_release_path(path);
1059 return -ENOENT;
1060 } else if (ret < 0) {
1061 return ret;
1062 }
1063
1064 leaf = path->nodes[0];
1065 inode_item = btrfs_item_ptr(leaf, path->slots[0],
1066 struct btrfs_inode_item);
1067 write_extent_buffer(leaf, &node->inode_item, (unsigned long)inode_item,
1068 sizeof(struct btrfs_inode_item));
1069 btrfs_mark_buffer_dirty(leaf);
1070
1071 if (!test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1072 goto no_iref;
1073
1074 path->slots[0]++;
1075 if (path->slots[0] >= btrfs_header_nritems(leaf))
1076 goto search;
1077 again:
1078 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1079 if (key.objectid != node->inode_id)
1080 goto out;
1081
1082 if (key.type != BTRFS_INODE_REF_KEY &&
1083 key.type != BTRFS_INODE_EXTREF_KEY)
1084 goto out;
1085
1086 /*
1087 * Delayed iref deletion is for the inode who has only one link,
1088 * so there is only one iref. The case that several irefs are
1089 * in the same item doesn't exist.
1090 */
1091 btrfs_del_item(trans, root, path);
1092 out:
1093 btrfs_release_delayed_iref(node);
1094 no_iref:
1095 btrfs_release_path(path);
1096 err_out:
1097 btrfs_delayed_inode_release_metadata(root, node);
1098 btrfs_release_delayed_inode(node);
1099
1100 return ret;
1101
1102 search:
1103 btrfs_release_path(path);
1104
1105 key.type = BTRFS_INODE_EXTREF_KEY;
1106 key.offset = -1;
1107 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1108 if (ret < 0)
1109 goto err_out;
1110 ASSERT(ret);
1111
1112 ret = 0;
1113 leaf = path->nodes[0];
1114 path->slots[0]--;
1115 goto again;
1116 }
1117
1118 static inline int btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1119 struct btrfs_root *root,
1120 struct btrfs_path *path,
1121 struct btrfs_delayed_node *node)
1122 {
1123 int ret;
1124
1125 mutex_lock(&node->mutex);
1126 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &node->flags)) {
1127 mutex_unlock(&node->mutex);
1128 return 0;
1129 }
1130
1131 ret = __btrfs_update_delayed_inode(trans, root, path, node);
1132 mutex_unlock(&node->mutex);
1133 return ret;
1134 }
1135
1136 static inline int
1137 __btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1138 struct btrfs_path *path,
1139 struct btrfs_delayed_node *node)
1140 {
1141 int ret;
1142
1143 ret = btrfs_insert_delayed_items(trans, path, node->root, node);
1144 if (ret)
1145 return ret;
1146
1147 ret = btrfs_delete_delayed_items(trans, path, node->root, node);
1148 if (ret)
1149 return ret;
1150
1151 ret = btrfs_update_delayed_inode(trans, node->root, path, node);
1152 return ret;
1153 }
1154
1155 /*
1156 * Called when committing the transaction.
1157 * Returns 0 on success.
1158 * Returns < 0 on error and returns with an aborted transaction with any
1159 * outstanding delayed items cleaned up.
1160 */
1161 static int __btrfs_run_delayed_items(struct btrfs_trans_handle *trans,
1162 struct btrfs_root *root, int nr)
1163 {
1164 struct btrfs_delayed_root *delayed_root;
1165 struct btrfs_delayed_node *curr_node, *prev_node;
1166 struct btrfs_path *path;
1167 struct btrfs_block_rsv *block_rsv;
1168 int ret = 0;
1169 bool count = (nr > 0);
1170
1171 if (trans->aborted)
1172 return -EIO;
1173
1174 path = btrfs_alloc_path();
1175 if (!path)
1176 return -ENOMEM;
1177 path->leave_spinning = 1;
1178
1179 block_rsv = trans->block_rsv;
1180 trans->block_rsv = &root->fs_info->delayed_block_rsv;
1181
1182 delayed_root = btrfs_get_delayed_root(root);
1183
1184 curr_node = btrfs_first_delayed_node(delayed_root);
1185 while (curr_node && (!count || (count && nr--))) {
1186 ret = __btrfs_commit_inode_delayed_items(trans, path,
1187 curr_node);
1188 if (ret) {
1189 btrfs_release_delayed_node(curr_node);
1190 curr_node = NULL;
1191 btrfs_abort_transaction(trans, root, ret);
1192 break;
1193 }
1194
1195 prev_node = curr_node;
1196 curr_node = btrfs_next_delayed_node(curr_node);
1197 btrfs_release_delayed_node(prev_node);
1198 }
1199
1200 if (curr_node)
1201 btrfs_release_delayed_node(curr_node);
1202 btrfs_free_path(path);
1203 trans->block_rsv = block_rsv;
1204
1205 return ret;
1206 }
1207
1208 int btrfs_run_delayed_items(struct btrfs_trans_handle *trans,
1209 struct btrfs_root *root)
1210 {
1211 return __btrfs_run_delayed_items(trans, root, -1);
1212 }
1213
1214 int btrfs_run_delayed_items_nr(struct btrfs_trans_handle *trans,
1215 struct btrfs_root *root, int nr)
1216 {
1217 return __btrfs_run_delayed_items(trans, root, nr);
1218 }
1219
1220 int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1221 struct inode *inode)
1222 {
1223 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1224 struct btrfs_path *path;
1225 struct btrfs_block_rsv *block_rsv;
1226 int ret;
1227
1228 if (!delayed_node)
1229 return 0;
1230
1231 mutex_lock(&delayed_node->mutex);
1232 if (!delayed_node->count) {
1233 mutex_unlock(&delayed_node->mutex);
1234 btrfs_release_delayed_node(delayed_node);
1235 return 0;
1236 }
1237 mutex_unlock(&delayed_node->mutex);
1238
1239 path = btrfs_alloc_path();
1240 if (!path) {
1241 btrfs_release_delayed_node(delayed_node);
1242 return -ENOMEM;
1243 }
1244 path->leave_spinning = 1;
1245
1246 block_rsv = trans->block_rsv;
1247 trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv;
1248
1249 ret = __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1250
1251 btrfs_release_delayed_node(delayed_node);
1252 btrfs_free_path(path);
1253 trans->block_rsv = block_rsv;
1254
1255 return ret;
1256 }
1257
1258 int btrfs_commit_inode_delayed_inode(struct inode *inode)
1259 {
1260 struct btrfs_trans_handle *trans;
1261 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1262 struct btrfs_path *path;
1263 struct btrfs_block_rsv *block_rsv;
1264 int ret;
1265
1266 if (!delayed_node)
1267 return 0;
1268
1269 mutex_lock(&delayed_node->mutex);
1270 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1271 mutex_unlock(&delayed_node->mutex);
1272 btrfs_release_delayed_node(delayed_node);
1273 return 0;
1274 }
1275 mutex_unlock(&delayed_node->mutex);
1276
1277 trans = btrfs_join_transaction(delayed_node->root);
1278 if (IS_ERR(trans)) {
1279 ret = PTR_ERR(trans);
1280 goto out;
1281 }
1282
1283 path = btrfs_alloc_path();
1284 if (!path) {
1285 ret = -ENOMEM;
1286 goto trans_out;
1287 }
1288 path->leave_spinning = 1;
1289
1290 block_rsv = trans->block_rsv;
1291 trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv;
1292
1293 mutex_lock(&delayed_node->mutex);
1294 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags))
1295 ret = __btrfs_update_delayed_inode(trans, delayed_node->root,
1296 path, delayed_node);
1297 else
1298 ret = 0;
1299 mutex_unlock(&delayed_node->mutex);
1300
1301 btrfs_free_path(path);
1302 trans->block_rsv = block_rsv;
1303 trans_out:
1304 btrfs_end_transaction(trans, delayed_node->root);
1305 btrfs_btree_balance_dirty(delayed_node->root);
1306 out:
1307 btrfs_release_delayed_node(delayed_node);
1308
1309 return ret;
1310 }
1311
1312 void btrfs_remove_delayed_node(struct inode *inode)
1313 {
1314 struct btrfs_delayed_node *delayed_node;
1315
1316 delayed_node = ACCESS_ONCE(BTRFS_I(inode)->delayed_node);
1317 if (!delayed_node)
1318 return;
1319
1320 BTRFS_I(inode)->delayed_node = NULL;
1321 btrfs_release_delayed_node(delayed_node);
1322 }
1323
1324 struct btrfs_async_delayed_work {
1325 struct btrfs_delayed_root *delayed_root;
1326 int nr;
1327 struct btrfs_work work;
1328 };
1329
1330 static void btrfs_async_run_delayed_root(struct btrfs_work *work)
1331 {
1332 struct btrfs_async_delayed_work *async_work;
1333 struct btrfs_delayed_root *delayed_root;
1334 struct btrfs_trans_handle *trans;
1335 struct btrfs_path *path;
1336 struct btrfs_delayed_node *delayed_node = NULL;
1337 struct btrfs_root *root;
1338 struct btrfs_block_rsv *block_rsv;
1339 int total_done = 0;
1340
1341 async_work = container_of(work, struct btrfs_async_delayed_work, work);
1342 delayed_root = async_work->delayed_root;
1343
1344 path = btrfs_alloc_path();
1345 if (!path)
1346 goto out;
1347
1348 again:
1349 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND / 2)
1350 goto free_path;
1351
1352 delayed_node = btrfs_first_prepared_delayed_node(delayed_root);
1353 if (!delayed_node)
1354 goto free_path;
1355
1356 path->leave_spinning = 1;
1357 root = delayed_node->root;
1358
1359 trans = btrfs_join_transaction(root);
1360 if (IS_ERR(trans))
1361 goto release_path;
1362
1363 block_rsv = trans->block_rsv;
1364 trans->block_rsv = &root->fs_info->delayed_block_rsv;
1365
1366 __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1367
1368 trans->block_rsv = block_rsv;
1369 btrfs_end_transaction(trans, root);
1370 btrfs_btree_balance_dirty_nodelay(root);
1371
1372 release_path:
1373 btrfs_release_path(path);
1374 total_done++;
1375
1376 btrfs_release_prepared_delayed_node(delayed_node);
1377 if (async_work->nr == 0 || total_done < async_work->nr)
1378 goto again;
1379
1380 free_path:
1381 btrfs_free_path(path);
1382 out:
1383 wake_up(&delayed_root->wait);
1384 kfree(async_work);
1385 }
1386
1387
1388 static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root,
1389 struct btrfs_fs_info *fs_info, int nr)
1390 {
1391 struct btrfs_async_delayed_work *async_work;
1392
1393 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1394 return 0;
1395
1396 async_work = kmalloc(sizeof(*async_work), GFP_NOFS);
1397 if (!async_work)
1398 return -ENOMEM;
1399
1400 async_work->delayed_root = delayed_root;
1401 btrfs_init_work(&async_work->work, btrfs_delayed_meta_helper,
1402 btrfs_async_run_delayed_root, NULL, NULL);
1403 async_work->nr = nr;
1404
1405 btrfs_queue_work(fs_info->delayed_workers, &async_work->work);
1406 return 0;
1407 }
1408
1409 void btrfs_assert_delayed_root_empty(struct btrfs_root *root)
1410 {
1411 struct btrfs_delayed_root *delayed_root;
1412 delayed_root = btrfs_get_delayed_root(root);
1413 WARN_ON(btrfs_first_delayed_node(delayed_root));
1414 }
1415
1416 static int could_end_wait(struct btrfs_delayed_root *delayed_root, int seq)
1417 {
1418 int val = atomic_read(&delayed_root->items_seq);
1419
1420 if (val < seq || val >= seq + BTRFS_DELAYED_BATCH)
1421 return 1;
1422
1423 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1424 return 1;
1425
1426 return 0;
1427 }
1428
1429 void btrfs_balance_delayed_items(struct btrfs_root *root)
1430 {
1431 struct btrfs_delayed_root *delayed_root;
1432 struct btrfs_fs_info *fs_info = root->fs_info;
1433
1434 delayed_root = btrfs_get_delayed_root(root);
1435
1436 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1437 return;
1438
1439 if (atomic_read(&delayed_root->items) >= BTRFS_DELAYED_WRITEBACK) {
1440 int seq;
1441 int ret;
1442
1443 seq = atomic_read(&delayed_root->items_seq);
1444
1445 ret = btrfs_wq_run_delayed_node(delayed_root, fs_info, 0);
1446 if (ret)
1447 return;
1448
1449 wait_event_interruptible(delayed_root->wait,
1450 could_end_wait(delayed_root, seq));
1451 return;
1452 }
1453
1454 btrfs_wq_run_delayed_node(delayed_root, fs_info, BTRFS_DELAYED_BATCH);
1455 }
1456
1457 /* Will return 0 or -ENOMEM */
1458 int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans,
1459 struct btrfs_root *root, const char *name,
1460 int name_len, struct inode *dir,
1461 struct btrfs_disk_key *disk_key, u8 type,
1462 u64 index)
1463 {
1464 struct btrfs_delayed_node *delayed_node;
1465 struct btrfs_delayed_item *delayed_item;
1466 struct btrfs_dir_item *dir_item;
1467 int ret;
1468
1469 delayed_node = btrfs_get_or_create_delayed_node(dir);
1470 if (IS_ERR(delayed_node))
1471 return PTR_ERR(delayed_node);
1472
1473 delayed_item = btrfs_alloc_delayed_item(sizeof(*dir_item) + name_len);
1474 if (!delayed_item) {
1475 ret = -ENOMEM;
1476 goto release_node;
1477 }
1478
1479 delayed_item->key.objectid = btrfs_ino(dir);
1480 delayed_item->key.type = BTRFS_DIR_INDEX_KEY;
1481 delayed_item->key.offset = index;
1482
1483 dir_item = (struct btrfs_dir_item *)delayed_item->data;
1484 dir_item->location = *disk_key;
1485 btrfs_set_stack_dir_transid(dir_item, trans->transid);
1486 btrfs_set_stack_dir_data_len(dir_item, 0);
1487 btrfs_set_stack_dir_name_len(dir_item, name_len);
1488 btrfs_set_stack_dir_type(dir_item, type);
1489 memcpy((char *)(dir_item + 1), name, name_len);
1490
1491 ret = btrfs_delayed_item_reserve_metadata(trans, root, delayed_item);
1492 /*
1493 * we have reserved enough space when we start a new transaction,
1494 * so reserving metadata failure is impossible
1495 */
1496 BUG_ON(ret);
1497
1498
1499 mutex_lock(&delayed_node->mutex);
1500 ret = __btrfs_add_delayed_insertion_item(delayed_node, delayed_item);
1501 if (unlikely(ret)) {
1502 btrfs_err(root->fs_info, "err add delayed dir index item(name: %.*s) "
1503 "into the insertion tree of the delayed node"
1504 "(root id: %llu, inode id: %llu, errno: %d)",
1505 name_len, name, delayed_node->root->objectid,
1506 delayed_node->inode_id, ret);
1507 BUG();
1508 }
1509 mutex_unlock(&delayed_node->mutex);
1510
1511 release_node:
1512 btrfs_release_delayed_node(delayed_node);
1513 return ret;
1514 }
1515
1516 static int btrfs_delete_delayed_insertion_item(struct btrfs_root *root,
1517 struct btrfs_delayed_node *node,
1518 struct btrfs_key *key)
1519 {
1520 struct btrfs_delayed_item *item;
1521
1522 mutex_lock(&node->mutex);
1523 item = __btrfs_lookup_delayed_insertion_item(node, key);
1524 if (!item) {
1525 mutex_unlock(&node->mutex);
1526 return 1;
1527 }
1528
1529 btrfs_delayed_item_release_metadata(root, item);
1530 btrfs_release_delayed_item(item);
1531 mutex_unlock(&node->mutex);
1532 return 0;
1533 }
1534
1535 int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans,
1536 struct btrfs_root *root, struct inode *dir,
1537 u64 index)
1538 {
1539 struct btrfs_delayed_node *node;
1540 struct btrfs_delayed_item *item;
1541 struct btrfs_key item_key;
1542 int ret;
1543
1544 node = btrfs_get_or_create_delayed_node(dir);
1545 if (IS_ERR(node))
1546 return PTR_ERR(node);
1547
1548 item_key.objectid = btrfs_ino(dir);
1549 item_key.type = BTRFS_DIR_INDEX_KEY;
1550 item_key.offset = index;
1551
1552 ret = btrfs_delete_delayed_insertion_item(root, node, &item_key);
1553 if (!ret)
1554 goto end;
1555
1556 item = btrfs_alloc_delayed_item(0);
1557 if (!item) {
1558 ret = -ENOMEM;
1559 goto end;
1560 }
1561
1562 item->key = item_key;
1563
1564 ret = btrfs_delayed_item_reserve_metadata(trans, root, item);
1565 /*
1566 * we have reserved enough space when we start a new transaction,
1567 * so reserving metadata failure is impossible.
1568 */
1569 BUG_ON(ret);
1570
1571 mutex_lock(&node->mutex);
1572 ret = __btrfs_add_delayed_deletion_item(node, item);
1573 if (unlikely(ret)) {
1574 btrfs_err(root->fs_info, "err add delayed dir index item(index: %llu) "
1575 "into the deletion tree of the delayed node"
1576 "(root id: %llu, inode id: %llu, errno: %d)",
1577 index, node->root->objectid, node->inode_id,
1578 ret);
1579 BUG();
1580 }
1581 mutex_unlock(&node->mutex);
1582 end:
1583 btrfs_release_delayed_node(node);
1584 return ret;
1585 }
1586
1587 int btrfs_inode_delayed_dir_index_count(struct inode *inode)
1588 {
1589 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1590
1591 if (!delayed_node)
1592 return -ENOENT;
1593
1594 /*
1595 * Since we have held i_mutex of this directory, it is impossible that
1596 * a new directory index is added into the delayed node and index_cnt
1597 * is updated now. So we needn't lock the delayed node.
1598 */
1599 if (!delayed_node->index_cnt) {
1600 btrfs_release_delayed_node(delayed_node);
1601 return -EINVAL;
1602 }
1603
1604 BTRFS_I(inode)->index_cnt = delayed_node->index_cnt;
1605 btrfs_release_delayed_node(delayed_node);
1606 return 0;
1607 }
1608
1609 void btrfs_get_delayed_items(struct inode *inode, struct list_head *ins_list,
1610 struct list_head *del_list)
1611 {
1612 struct btrfs_delayed_node *delayed_node;
1613 struct btrfs_delayed_item *item;
1614
1615 delayed_node = btrfs_get_delayed_node(inode);
1616 if (!delayed_node)
1617 return;
1618
1619 mutex_lock(&delayed_node->mutex);
1620 item = __btrfs_first_delayed_insertion_item(delayed_node);
1621 while (item) {
1622 atomic_inc(&item->refs);
1623 list_add_tail(&item->readdir_list, ins_list);
1624 item = __btrfs_next_delayed_item(item);
1625 }
1626
1627 item = __btrfs_first_delayed_deletion_item(delayed_node);
1628 while (item) {
1629 atomic_inc(&item->refs);
1630 list_add_tail(&item->readdir_list, del_list);
1631 item = __btrfs_next_delayed_item(item);
1632 }
1633 mutex_unlock(&delayed_node->mutex);
1634 /*
1635 * This delayed node is still cached in the btrfs inode, so refs
1636 * must be > 1 now, and we needn't check it is going to be freed
1637 * or not.
1638 *
1639 * Besides that, this function is used to read dir, we do not
1640 * insert/delete delayed items in this period. So we also needn't
1641 * requeue or dequeue this delayed node.
1642 */
1643 atomic_dec(&delayed_node->refs);
1644 }
1645
1646 void btrfs_put_delayed_items(struct list_head *ins_list,
1647 struct list_head *del_list)
1648 {
1649 struct btrfs_delayed_item *curr, *next;
1650
1651 list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1652 list_del(&curr->readdir_list);
1653 if (atomic_dec_and_test(&curr->refs))
1654 kfree(curr);
1655 }
1656
1657 list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1658 list_del(&curr->readdir_list);
1659 if (atomic_dec_and_test(&curr->refs))
1660 kfree(curr);
1661 }
1662 }
1663
1664 int btrfs_should_delete_dir_index(struct list_head *del_list,
1665 u64 index)
1666 {
1667 struct btrfs_delayed_item *curr, *next;
1668 int ret;
1669
1670 if (list_empty(del_list))
1671 return 0;
1672
1673 list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1674 if (curr->key.offset > index)
1675 break;
1676
1677 list_del(&curr->readdir_list);
1678 ret = (curr->key.offset == index);
1679
1680 if (atomic_dec_and_test(&curr->refs))
1681 kfree(curr);
1682
1683 if (ret)
1684 return 1;
1685 else
1686 continue;
1687 }
1688 return 0;
1689 }
1690
1691 /*
1692 * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree
1693 *
1694 */
1695 int btrfs_readdir_delayed_dir_index(struct dir_context *ctx,
1696 struct list_head *ins_list, bool *emitted)
1697 {
1698 struct btrfs_dir_item *di;
1699 struct btrfs_delayed_item *curr, *next;
1700 struct btrfs_key location;
1701 char *name;
1702 int name_len;
1703 int over = 0;
1704 unsigned char d_type;
1705
1706 if (list_empty(ins_list))
1707 return 0;
1708
1709 /*
1710 * Changing the data of the delayed item is impossible. So
1711 * we needn't lock them. And we have held i_mutex of the
1712 * directory, nobody can delete any directory indexes now.
1713 */
1714 list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1715 list_del(&curr->readdir_list);
1716
1717 if (curr->key.offset < ctx->pos) {
1718 if (atomic_dec_and_test(&curr->refs))
1719 kfree(curr);
1720 continue;
1721 }
1722
1723 ctx->pos = curr->key.offset;
1724
1725 di = (struct btrfs_dir_item *)curr->data;
1726 name = (char *)(di + 1);
1727 name_len = btrfs_stack_dir_name_len(di);
1728
1729 d_type = btrfs_filetype_table[di->type];
1730 btrfs_disk_key_to_cpu(&location, &di->location);
1731
1732 over = !dir_emit(ctx, name, name_len,
1733 location.objectid, d_type);
1734
1735 if (atomic_dec_and_test(&curr->refs))
1736 kfree(curr);
1737
1738 if (over)
1739 return 1;
1740 *emitted = true;
1741 }
1742 return 0;
1743 }
1744
1745 static void fill_stack_inode_item(struct btrfs_trans_handle *trans,
1746 struct btrfs_inode_item *inode_item,
1747 struct inode *inode)
1748 {
1749 btrfs_set_stack_inode_uid(inode_item, i_uid_read(inode));
1750 btrfs_set_stack_inode_gid(inode_item, i_gid_read(inode));
1751 btrfs_set_stack_inode_size(inode_item, BTRFS_I(inode)->disk_i_size);
1752 btrfs_set_stack_inode_mode(inode_item, inode->i_mode);
1753 btrfs_set_stack_inode_nlink(inode_item, inode->i_nlink);
1754 btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(inode));
1755 btrfs_set_stack_inode_generation(inode_item,
1756 BTRFS_I(inode)->generation);
1757 btrfs_set_stack_inode_sequence(inode_item, inode->i_version);
1758 btrfs_set_stack_inode_transid(inode_item, trans->transid);
1759 btrfs_set_stack_inode_rdev(inode_item, inode->i_rdev);
1760 btrfs_set_stack_inode_flags(inode_item, BTRFS_I(inode)->flags);
1761 btrfs_set_stack_inode_block_group(inode_item, 0);
1762
1763 btrfs_set_stack_timespec_sec(&inode_item->atime,
1764 inode->i_atime.tv_sec);
1765 btrfs_set_stack_timespec_nsec(&inode_item->atime,
1766 inode->i_atime.tv_nsec);
1767
1768 btrfs_set_stack_timespec_sec(&inode_item->mtime,
1769 inode->i_mtime.tv_sec);
1770 btrfs_set_stack_timespec_nsec(&inode_item->mtime,
1771 inode->i_mtime.tv_nsec);
1772
1773 btrfs_set_stack_timespec_sec(&inode_item->ctime,
1774 inode->i_ctime.tv_sec);
1775 btrfs_set_stack_timespec_nsec(&inode_item->ctime,
1776 inode->i_ctime.tv_nsec);
1777
1778 btrfs_set_stack_timespec_sec(&inode_item->otime,
1779 BTRFS_I(inode)->i_otime.tv_sec);
1780 btrfs_set_stack_timespec_nsec(&inode_item->otime,
1781 BTRFS_I(inode)->i_otime.tv_nsec);
1782 }
1783
1784 int btrfs_fill_inode(struct inode *inode, u32 *rdev)
1785 {
1786 struct btrfs_delayed_node *delayed_node;
1787 struct btrfs_inode_item *inode_item;
1788
1789 delayed_node = btrfs_get_delayed_node(inode);
1790 if (!delayed_node)
1791 return -ENOENT;
1792
1793 mutex_lock(&delayed_node->mutex);
1794 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1795 mutex_unlock(&delayed_node->mutex);
1796 btrfs_release_delayed_node(delayed_node);
1797 return -ENOENT;
1798 }
1799
1800 inode_item = &delayed_node->inode_item;
1801
1802 i_uid_write(inode, btrfs_stack_inode_uid(inode_item));
1803 i_gid_write(inode, btrfs_stack_inode_gid(inode_item));
1804 btrfs_i_size_write(inode, btrfs_stack_inode_size(inode_item));
1805 inode->i_mode = btrfs_stack_inode_mode(inode_item);
1806 set_nlink(inode, btrfs_stack_inode_nlink(inode_item));
1807 inode_set_bytes(inode, btrfs_stack_inode_nbytes(inode_item));
1808 BTRFS_I(inode)->generation = btrfs_stack_inode_generation(inode_item);
1809 BTRFS_I(inode)->last_trans = btrfs_stack_inode_transid(inode_item);
1810
1811 inode->i_version = btrfs_stack_inode_sequence(inode_item);
1812 inode->i_rdev = 0;
1813 *rdev = btrfs_stack_inode_rdev(inode_item);
1814 BTRFS_I(inode)->flags = btrfs_stack_inode_flags(inode_item);
1815
1816 inode->i_atime.tv_sec = btrfs_stack_timespec_sec(&inode_item->atime);
1817 inode->i_atime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->atime);
1818
1819 inode->i_mtime.tv_sec = btrfs_stack_timespec_sec(&inode_item->mtime);
1820 inode->i_mtime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->mtime);
1821
1822 inode->i_ctime.tv_sec = btrfs_stack_timespec_sec(&inode_item->ctime);
1823 inode->i_ctime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->ctime);
1824
1825 BTRFS_I(inode)->i_otime.tv_sec =
1826 btrfs_stack_timespec_sec(&inode_item->otime);
1827 BTRFS_I(inode)->i_otime.tv_nsec =
1828 btrfs_stack_timespec_nsec(&inode_item->otime);
1829
1830 inode->i_generation = BTRFS_I(inode)->generation;
1831 BTRFS_I(inode)->index_cnt = (u64)-1;
1832
1833 mutex_unlock(&delayed_node->mutex);
1834 btrfs_release_delayed_node(delayed_node);
1835 return 0;
1836 }
1837
1838 int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans,
1839 struct btrfs_root *root, struct inode *inode)
1840 {
1841 struct btrfs_delayed_node *delayed_node;
1842 int ret = 0;
1843
1844 delayed_node = btrfs_get_or_create_delayed_node(inode);
1845 if (IS_ERR(delayed_node))
1846 return PTR_ERR(delayed_node);
1847
1848 mutex_lock(&delayed_node->mutex);
1849 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1850 fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1851 goto release_node;
1852 }
1853
1854 ret = btrfs_delayed_inode_reserve_metadata(trans, root, inode,
1855 delayed_node);
1856 if (ret)
1857 goto release_node;
1858
1859 fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1860 set_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
1861 delayed_node->count++;
1862 atomic_inc(&root->fs_info->delayed_root->items);
1863 release_node:
1864 mutex_unlock(&delayed_node->mutex);
1865 btrfs_release_delayed_node(delayed_node);
1866 return ret;
1867 }
1868
1869 int btrfs_delayed_delete_inode_ref(struct inode *inode)
1870 {
1871 struct btrfs_delayed_node *delayed_node;
1872
1873 /*
1874 * we don't do delayed inode updates during log recovery because it
1875 * leads to enospc problems. This means we also can't do
1876 * delayed inode refs
1877 */
1878 if (BTRFS_I(inode)->root->fs_info->log_root_recovering)
1879 return -EAGAIN;
1880
1881 delayed_node = btrfs_get_or_create_delayed_node(inode);
1882 if (IS_ERR(delayed_node))
1883 return PTR_ERR(delayed_node);
1884
1885 /*
1886 * We don't reserve space for inode ref deletion is because:
1887 * - We ONLY do async inode ref deletion for the inode who has only
1888 * one link(i_nlink == 1), it means there is only one inode ref.
1889 * And in most case, the inode ref and the inode item are in the
1890 * same leaf, and we will deal with them at the same time.
1891 * Since we are sure we will reserve the space for the inode item,
1892 * it is unnecessary to reserve space for inode ref deletion.
1893 * - If the inode ref and the inode item are not in the same leaf,
1894 * We also needn't worry about enospc problem, because we reserve
1895 * much more space for the inode update than it needs.
1896 * - At the worst, we can steal some space from the global reservation.
1897 * It is very rare.
1898 */
1899 mutex_lock(&delayed_node->mutex);
1900 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1901 goto release_node;
1902
1903 set_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
1904 delayed_node->count++;
1905 atomic_inc(&BTRFS_I(inode)->root->fs_info->delayed_root->items);
1906 release_node:
1907 mutex_unlock(&delayed_node->mutex);
1908 btrfs_release_delayed_node(delayed_node);
1909 return 0;
1910 }
1911
1912 static void __btrfs_kill_delayed_node(struct btrfs_delayed_node *delayed_node)
1913 {
1914 struct btrfs_root *root = delayed_node->root;
1915 struct btrfs_delayed_item *curr_item, *prev_item;
1916
1917 mutex_lock(&delayed_node->mutex);
1918 curr_item = __btrfs_first_delayed_insertion_item(delayed_node);
1919 while (curr_item) {
1920 btrfs_delayed_item_release_metadata(root, curr_item);
1921 prev_item = curr_item;
1922 curr_item = __btrfs_next_delayed_item(prev_item);
1923 btrfs_release_delayed_item(prev_item);
1924 }
1925
1926 curr_item = __btrfs_first_delayed_deletion_item(delayed_node);
1927 while (curr_item) {
1928 btrfs_delayed_item_release_metadata(root, curr_item);
1929 prev_item = curr_item;
1930 curr_item = __btrfs_next_delayed_item(prev_item);
1931 btrfs_release_delayed_item(prev_item);
1932 }
1933
1934 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1935 btrfs_release_delayed_iref(delayed_node);
1936
1937 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1938 btrfs_delayed_inode_release_metadata(root, delayed_node);
1939 btrfs_release_delayed_inode(delayed_node);
1940 }
1941 mutex_unlock(&delayed_node->mutex);
1942 }
1943
1944 void btrfs_kill_delayed_inode_items(struct inode *inode)
1945 {
1946 struct btrfs_delayed_node *delayed_node;
1947
1948 delayed_node = btrfs_get_delayed_node(inode);
1949 if (!delayed_node)
1950 return;
1951
1952 __btrfs_kill_delayed_node(delayed_node);
1953 btrfs_release_delayed_node(delayed_node);
1954 }
1955
1956 void btrfs_kill_all_delayed_nodes(struct btrfs_root *root)
1957 {
1958 u64 inode_id = 0;
1959 struct btrfs_delayed_node *delayed_nodes[8];
1960 int i, n;
1961
1962 while (1) {
1963 spin_lock(&root->inode_lock);
1964 n = radix_tree_gang_lookup(&root->delayed_nodes_tree,
1965 (void **)delayed_nodes, inode_id,
1966 ARRAY_SIZE(delayed_nodes));
1967 if (!n) {
1968 spin_unlock(&root->inode_lock);
1969 break;
1970 }
1971
1972 inode_id = delayed_nodes[n - 1]->inode_id + 1;
1973
1974 for (i = 0; i < n; i++)
1975 atomic_inc(&delayed_nodes[i]->refs);
1976 spin_unlock(&root->inode_lock);
1977
1978 for (i = 0; i < n; i++) {
1979 __btrfs_kill_delayed_node(delayed_nodes[i]);
1980 btrfs_release_delayed_node(delayed_nodes[i]);
1981 }
1982 }
1983 }
1984
1985 void btrfs_destroy_delayed_inodes(struct btrfs_root *root)
1986 {
1987 struct btrfs_delayed_root *delayed_root;
1988 struct btrfs_delayed_node *curr_node, *prev_node;
1989
1990 delayed_root = btrfs_get_delayed_root(root);
1991
1992 curr_node = btrfs_first_delayed_node(delayed_root);
1993 while (curr_node) {
1994 __btrfs_kill_delayed_node(curr_node);
1995
1996 prev_node = curr_node;
1997 curr_node = btrfs_next_delayed_node(curr_node);
1998 btrfs_release_delayed_node(prev_node);
1999 }
2000 }
2001
Linux with added FPC-III support