?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 4.8-rc8
[linux/fpc-iii.git] / fs / btrfs / delayed-inode.c
blob3eeb9cd8cfa57edc81bdeba91b1d37a3011d7bf9
1 /*
2 * Copyright (C) 2011 Fujitsu. All rights reserved.
3 * Written by Miao Xie <miaox@cn.fujitsu.com>
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public
7 * License v2 as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12 * General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public
15 * License along with this program; if not, write to the
16 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
17 * Boston, MA 021110-1307, USA.
18 */
19
20 #include <linux/slab.h>
21 #include "delayed-inode.h"
22 #include "disk-io.h"
23 #include "transaction.h"
24 #include "ctree.h"
25
26 #define BTRFS_DELAYED_WRITEBACK 512
27 #define BTRFS_DELAYED_BACKGROUND 128
28 #define BTRFS_DELAYED_BATCH 16
29
30 static struct kmem_cache *delayed_node_cache;
31
32 int __init btrfs_delayed_inode_init(void)
33 {
34 delayed_node_cache = kmem_cache_create("btrfs_delayed_node",
35 sizeof(struct btrfs_delayed_node),
36 0,
37 SLAB_MEM_SPREAD,
38 NULL);
39 if (!delayed_node_cache)
40 return -ENOMEM;
41 return 0;
42 }
43
44 void btrfs_delayed_inode_exit(void)
45 {
46 kmem_cache_destroy(delayed_node_cache);
47 }
48
49 static inline void btrfs_init_delayed_node(
50 struct btrfs_delayed_node *delayed_node,
51 struct btrfs_root *root, u64 inode_id)
52 {
53 delayed_node->root = root;
54 delayed_node->inode_id = inode_id;
55 atomic_set(&delayed_node->refs, 0);
56 delayed_node->ins_root = RB_ROOT;
57 delayed_node->del_root = RB_ROOT;
58 mutex_init(&delayed_node->mutex);
59 INIT_LIST_HEAD(&delayed_node->n_list);
60 INIT_LIST_HEAD(&delayed_node->p_list);
61 }
62
63 static inline int btrfs_is_continuous_delayed_item(
64 struct btrfs_delayed_item *item1,
65 struct btrfs_delayed_item *item2)
66 {
67 if (item1->key.type == BTRFS_DIR_INDEX_KEY &&
68 item1->key.objectid == item2->key.objectid &&
69 item1->key.type == item2->key.type &&
70 item1->key.offset + 1 == item2->key.offset)
71 return 1;
72 return 0;
73 }
74
75 static 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);
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, 1);
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 * If our block_rsv is the delalloc block reserve then check and see if
602 * we have our extra reservation for updating the inode. If not fall
603 * through and try to reserve space quickly.
604 *
605 * We used to try and steal from the delalloc block rsv or the global
606 * reserve, but we'd steal a full reservation, which isn't kind. We are
607 * here through delalloc which means we've likely just cowed down close
608 * to the leaf that contains the inode, so we would steal less just
609 * doing the fallback inode update, so if we do end up having to steal
610 * from the global block rsv we hopefully only steal one or two blocks
611 * worth which is less likely to hurt us.
612 */
613 if (src_rsv && src_rsv->type == BTRFS_BLOCK_RSV_DELALLOC) {
614 spin_lock(&BTRFS_I(inode)->lock);
615 if (test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
616 &BTRFS_I(inode)->runtime_flags))
617 release = true;
618 else
619 src_rsv = NULL;
620 spin_unlock(&BTRFS_I(inode)->lock);
621 }
622
623 /*
624 * btrfs_dirty_inode will update the inode under btrfs_join_transaction
625 * which doesn't reserve space for speed. This is a problem since we
626 * still need to reserve space for this update, so try to reserve the
627 * space.
628 *
629 * Now if src_rsv == delalloc_block_rsv we'll let it just steal since
630 * we're accounted for.
631 */
632 if (!src_rsv || (!trans->bytes_reserved &&
633 src_rsv->type != BTRFS_BLOCK_RSV_DELALLOC)) {
634 ret = btrfs_block_rsv_add(root, dst_rsv, num_bytes,
635 BTRFS_RESERVE_NO_FLUSH);
636 /*
637 * Since we're under a transaction reserve_metadata_bytes could
638 * try to commit the transaction which will make it return
639 * EAGAIN to make us stop the transaction we have, so return
640 * ENOSPC instead so that btrfs_dirty_inode knows what to do.
641 */
642 if (ret == -EAGAIN)
643 ret = -ENOSPC;
644 if (!ret) {
645 node->bytes_reserved = num_bytes;
646 trace_btrfs_space_reservation(root->fs_info,
647 "delayed_inode",
648 btrfs_ino(inode),
649 num_bytes, 1);
650 }
651 return ret;
652 }
653
654 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1);
655
656 /*
657 * Migrate only takes a reservation, it doesn't touch the size of the
658 * block_rsv. This is to simplify people who don't normally have things
659 * migrated from their block rsv. If they go to release their
660 * reservation, that will decrease the size as well, so if migrate
661 * reduced size we'd end up with a negative size. But for the
662 * delalloc_meta_reserved stuff we will only know to drop 1 reservation,
663 * but we could in fact do this reserve/migrate dance several times
664 * between the time we did the original reservation and we'd clean it
665 * up. So to take care of this, release the space for the meta
666 * reservation here. I think it may be time for a documentation page on
667 * how block rsvs. work.
668 */
669 if (!ret) {
670 trace_btrfs_space_reservation(root->fs_info, "delayed_inode",
671 btrfs_ino(inode), num_bytes, 1);
672 node->bytes_reserved = num_bytes;
673 }
674
675 if (release) {
676 trace_btrfs_space_reservation(root->fs_info, "delalloc",
677 btrfs_ino(inode), num_bytes, 0);
678 btrfs_block_rsv_release(root, src_rsv, num_bytes);
679 }
680
681 return ret;
682 }
683
684 static void btrfs_delayed_inode_release_metadata(struct btrfs_root *root,
685 struct btrfs_delayed_node *node)
686 {
687 struct btrfs_block_rsv *rsv;
688
689 if (!node->bytes_reserved)
690 return;
691
692 rsv = &root->fs_info->delayed_block_rsv;
693 trace_btrfs_space_reservation(root->fs_info, "delayed_inode",
694 node->inode_id, node->bytes_reserved, 0);
695 btrfs_block_rsv_release(root, rsv,
696 node->bytes_reserved);
697 node->bytes_reserved = 0;
698 }
699
700 /*
701 * This helper will insert some continuous items into the same leaf according
702 * to the free space of the leaf.
703 */
704 static int btrfs_batch_insert_items(struct btrfs_root *root,
705 struct btrfs_path *path,
706 struct btrfs_delayed_item *item)
707 {
708 struct btrfs_delayed_item *curr, *next;
709 int free_space;
710 int total_data_size = 0, total_size = 0;
711 struct extent_buffer *leaf;
712 char *data_ptr;
713 struct btrfs_key *keys;
714 u32 *data_size;
715 struct list_head head;
716 int slot;
717 int nitems;
718 int i;
719 int ret = 0;
720
721 BUG_ON(!path->nodes[0]);
722
723 leaf = path->nodes[0];
724 free_space = btrfs_leaf_free_space(root, leaf);
725 INIT_LIST_HEAD(&head);
726
727 next = item;
728 nitems = 0;
729
730 /*
731 * count the number of the continuous items that we can insert in batch
732 */
733 while (total_size + next->data_len + sizeof(struct btrfs_item) <=
734 free_space) {
735 total_data_size += next->data_len;
736 total_size += next->data_len + sizeof(struct btrfs_item);
737 list_add_tail(&next->tree_list, &head);
738 nitems++;
739
740 curr = next;
741 next = __btrfs_next_delayed_item(curr);
742 if (!next)
743 break;
744
745 if (!btrfs_is_continuous_delayed_item(curr, next))
746 break;
747 }
748
749 if (!nitems) {
750 ret = 0;
751 goto out;
752 }
753
754 /*
755 * we need allocate some memory space, but it might cause the task
756 * to sleep, so we set all locked nodes in the path to blocking locks
757 * first.
758 */
759 btrfs_set_path_blocking(path);
760
761 keys = kmalloc_array(nitems, sizeof(struct btrfs_key), GFP_NOFS);
762 if (!keys) {
763 ret = -ENOMEM;
764 goto out;
765 }
766
767 data_size = kmalloc_array(nitems, sizeof(u32), GFP_NOFS);
768 if (!data_size) {
769 ret = -ENOMEM;
770 goto error;
771 }
772
773 /* get keys of all the delayed items */
774 i = 0;
775 list_for_each_entry(next, &head, tree_list) {
776 keys[i] = next->key;
777 data_size[i] = next->data_len;
778 i++;
779 }
780
781 /* reset all the locked nodes in the patch to spinning locks. */
782 btrfs_clear_path_blocking(path, NULL, 0);
783
784 /* insert the keys of the items */
785 setup_items_for_insert(root, path, keys, data_size,
786 total_data_size, total_size, nitems);
787
788 /* insert the dir index items */
789 slot = path->slots[0];
790 list_for_each_entry_safe(curr, next, &head, tree_list) {
791 data_ptr = btrfs_item_ptr(leaf, slot, char);
792 write_extent_buffer(leaf, &curr->data,
793 (unsigned long)data_ptr,
794 curr->data_len);
795 slot++;
796
797 btrfs_delayed_item_release_metadata(root, curr);
798
799 list_del(&curr->tree_list);
800 btrfs_release_delayed_item(curr);
801 }
802
803 error:
804 kfree(data_size);
805 kfree(keys);
806 out:
807 return ret;
808 }
809
810 /*
811 * This helper can just do simple insertion that needn't extend item for new
812 * data, such as directory name index insertion, inode insertion.
813 */
814 static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans,
815 struct btrfs_root *root,
816 struct btrfs_path *path,
817 struct btrfs_delayed_item *delayed_item)
818 {
819 struct extent_buffer *leaf;
820 char *ptr;
821 int ret;
822
823 ret = btrfs_insert_empty_item(trans, root, path, &delayed_item->key,
824 delayed_item->data_len);
825 if (ret < 0 && ret != -EEXIST)
826 return ret;
827
828 leaf = path->nodes[0];
829
830 ptr = btrfs_item_ptr(leaf, path->slots[0], char);
831
832 write_extent_buffer(leaf, delayed_item->data, (unsigned long)ptr,
833 delayed_item->data_len);
834 btrfs_mark_buffer_dirty(leaf);
835
836 btrfs_delayed_item_release_metadata(root, delayed_item);
837 return 0;
838 }
839
840 /*
841 * we insert an item first, then if there are some continuous items, we try
842 * to insert those items into the same leaf.
843 */
844 static int btrfs_insert_delayed_items(struct btrfs_trans_handle *trans,
845 struct btrfs_path *path,
846 struct btrfs_root *root,
847 struct btrfs_delayed_node *node)
848 {
849 struct btrfs_delayed_item *curr, *prev;
850 int ret = 0;
851
852 do_again:
853 mutex_lock(&node->mutex);
854 curr = __btrfs_first_delayed_insertion_item(node);
855 if (!curr)
856 goto insert_end;
857
858 ret = btrfs_insert_delayed_item(trans, root, path, curr);
859 if (ret < 0) {
860 btrfs_release_path(path);
861 goto insert_end;
862 }
863
864 prev = curr;
865 curr = __btrfs_next_delayed_item(prev);
866 if (curr && btrfs_is_continuous_delayed_item(prev, curr)) {
867 /* insert the continuous items into the same leaf */
868 path->slots[0]++;
869 btrfs_batch_insert_items(root, path, curr);
870 }
871 btrfs_release_delayed_item(prev);
872 btrfs_mark_buffer_dirty(path->nodes[0]);
873
874 btrfs_release_path(path);
875 mutex_unlock(&node->mutex);
876 goto do_again;
877
878 insert_end:
879 mutex_unlock(&node->mutex);
880 return ret;
881 }
882
883 static int btrfs_batch_delete_items(struct btrfs_trans_handle *trans,
884 struct btrfs_root *root,
885 struct btrfs_path *path,
886 struct btrfs_delayed_item *item)
887 {
888 struct btrfs_delayed_item *curr, *next;
889 struct extent_buffer *leaf;
890 struct btrfs_key key;
891 struct list_head head;
892 int nitems, i, last_item;
893 int ret = 0;
894
895 BUG_ON(!path->nodes[0]);
896
897 leaf = path->nodes[0];
898
899 i = path->slots[0];
900 last_item = btrfs_header_nritems(leaf) - 1;
901 if (i > last_item)
902 return -ENOENT; /* FIXME: Is errno suitable? */
903
904 next = item;
905 INIT_LIST_HEAD(&head);
906 btrfs_item_key_to_cpu(leaf, &key, i);
907 nitems = 0;
908 /*
909 * count the number of the dir index items that we can delete in batch
910 */
911 while (btrfs_comp_cpu_keys(&next->key, &key) == 0) {
912 list_add_tail(&next->tree_list, &head);
913 nitems++;
914
915 curr = next;
916 next = __btrfs_next_delayed_item(curr);
917 if (!next)
918 break;
919
920 if (!btrfs_is_continuous_delayed_item(curr, next))
921 break;
922
923 i++;
924 if (i > last_item)
925 break;
926 btrfs_item_key_to_cpu(leaf, &key, i);
927 }
928
929 if (!nitems)
930 return 0;
931
932 ret = btrfs_del_items(trans, root, path, path->slots[0], nitems);
933 if (ret)
934 goto out;
935
936 list_for_each_entry_safe(curr, next, &head, tree_list) {
937 btrfs_delayed_item_release_metadata(root, curr);
938 list_del(&curr->tree_list);
939 btrfs_release_delayed_item(curr);
940 }
941
942 out:
943 return ret;
944 }
945
946 static int btrfs_delete_delayed_items(struct btrfs_trans_handle *trans,
947 struct btrfs_path *path,
948 struct btrfs_root *root,
949 struct btrfs_delayed_node *node)
950 {
951 struct btrfs_delayed_item *curr, *prev;
952 int ret = 0;
953
954 do_again:
955 mutex_lock(&node->mutex);
956 curr = __btrfs_first_delayed_deletion_item(node);
957 if (!curr)
958 goto delete_fail;
959
960 ret = btrfs_search_slot(trans, root, &curr->key, path, -1, 1);
961 if (ret < 0)
962 goto delete_fail;
963 else if (ret > 0) {
964 /*
965 * can't find the item which the node points to, so this node
966 * is invalid, just drop it.
967 */
968 prev = curr;
969 curr = __btrfs_next_delayed_item(prev);
970 btrfs_release_delayed_item(prev);
971 ret = 0;
972 btrfs_release_path(path);
973 if (curr) {
974 mutex_unlock(&node->mutex);
975 goto do_again;
976 } else
977 goto delete_fail;
978 }
979
980 btrfs_batch_delete_items(trans, root, path, curr);
981 btrfs_release_path(path);
982 mutex_unlock(&node->mutex);
983 goto do_again;
984
985 delete_fail:
986 btrfs_release_path(path);
987 mutex_unlock(&node->mutex);
988 return ret;
989 }
990
991 static void btrfs_release_delayed_inode(struct btrfs_delayed_node *delayed_node)
992 {
993 struct btrfs_delayed_root *delayed_root;
994
995 if (delayed_node &&
996 test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
997 BUG_ON(!delayed_node->root);
998 clear_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
999 delayed_node->count--;
1000
1001 delayed_root = delayed_node->root->fs_info->delayed_root;
1002 finish_one_item(delayed_root);
1003 }
1004 }
1005
1006 static void btrfs_release_delayed_iref(struct btrfs_delayed_node *delayed_node)
1007 {
1008 struct btrfs_delayed_root *delayed_root;
1009
1010 ASSERT(delayed_node->root);
1011 clear_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
1012 delayed_node->count--;
1013
1014 delayed_root = delayed_node->root->fs_info->delayed_root;
1015 finish_one_item(delayed_root);
1016 }
1017
1018 static int __btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1019 struct btrfs_root *root,
1020 struct btrfs_path *path,
1021 struct btrfs_delayed_node *node)
1022 {
1023 struct btrfs_key key;
1024 struct btrfs_inode_item *inode_item;
1025 struct extent_buffer *leaf;
1026 int mod;
1027 int ret;
1028
1029 key.objectid = node->inode_id;
1030 key.type = BTRFS_INODE_ITEM_KEY;
1031 key.offset = 0;
1032
1033 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1034 mod = -1;
1035 else
1036 mod = 1;
1037
1038 ret = btrfs_lookup_inode(trans, root, path, &key, mod);
1039 if (ret > 0) {
1040 btrfs_release_path(path);
1041 return -ENOENT;
1042 } else if (ret < 0) {
1043 return ret;
1044 }
1045
1046 leaf = path->nodes[0];
1047 inode_item = btrfs_item_ptr(leaf, path->slots[0],
1048 struct btrfs_inode_item);
1049 write_extent_buffer(leaf, &node->inode_item, (unsigned long)inode_item,
1050 sizeof(struct btrfs_inode_item));
1051 btrfs_mark_buffer_dirty(leaf);
1052
1053 if (!test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1054 goto no_iref;
1055
1056 path->slots[0]++;
1057 if (path->slots[0] >= btrfs_header_nritems(leaf))
1058 goto search;
1059 again:
1060 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1061 if (key.objectid != node->inode_id)
1062 goto out;
1063
1064 if (key.type != BTRFS_INODE_REF_KEY &&
1065 key.type != BTRFS_INODE_EXTREF_KEY)
1066 goto out;
1067
1068 /*
1069 * Delayed iref deletion is for the inode who has only one link,
1070 * so there is only one iref. The case that several irefs are
1071 * in the same item doesn't exist.
1072 */
1073 btrfs_del_item(trans, root, path);
1074 out:
1075 btrfs_release_delayed_iref(node);
1076 no_iref:
1077 btrfs_release_path(path);
1078 err_out:
1079 btrfs_delayed_inode_release_metadata(root, node);
1080 btrfs_release_delayed_inode(node);
1081
1082 return ret;
1083
1084 search:
1085 btrfs_release_path(path);
1086
1087 key.type = BTRFS_INODE_EXTREF_KEY;
1088 key.offset = -1;
1089 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1090 if (ret < 0)
1091 goto err_out;
1092 ASSERT(ret);
1093
1094 ret = 0;
1095 leaf = path->nodes[0];
1096 path->slots[0]--;
1097 goto again;
1098 }
1099
1100 static inline int btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1101 struct btrfs_root *root,
1102 struct btrfs_path *path,
1103 struct btrfs_delayed_node *node)
1104 {
1105 int ret;
1106
1107 mutex_lock(&node->mutex);
1108 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &node->flags)) {
1109 mutex_unlock(&node->mutex);
1110 return 0;
1111 }
1112
1113 ret = __btrfs_update_delayed_inode(trans, root, path, node);
1114 mutex_unlock(&node->mutex);
1115 return ret;
1116 }
1117
1118 static inline int
1119 __btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1120 struct btrfs_path *path,
1121 struct btrfs_delayed_node *node)
1122 {
1123 int ret;
1124
1125 ret = btrfs_insert_delayed_items(trans, path, node->root, node);
1126 if (ret)
1127 return ret;
1128
1129 ret = btrfs_delete_delayed_items(trans, path, node->root, node);
1130 if (ret)
1131 return ret;
1132
1133 ret = btrfs_update_delayed_inode(trans, node->root, path, node);
1134 return ret;
1135 }
1136
1137 /*
1138 * Called when committing the transaction.
1139 * Returns 0 on success.
1140 * Returns < 0 on error and returns with an aborted transaction with any
1141 * outstanding delayed items cleaned up.
1142 */
1143 static int __btrfs_run_delayed_items(struct btrfs_trans_handle *trans,
1144 struct btrfs_root *root, int nr)
1145 {
1146 struct btrfs_delayed_root *delayed_root;
1147 struct btrfs_delayed_node *curr_node, *prev_node;
1148 struct btrfs_path *path;
1149 struct btrfs_block_rsv *block_rsv;
1150 int ret = 0;
1151 bool count = (nr > 0);
1152
1153 if (trans->aborted)
1154 return -EIO;
1155
1156 path = btrfs_alloc_path();
1157 if (!path)
1158 return -ENOMEM;
1159 path->leave_spinning = 1;
1160
1161 block_rsv = trans->block_rsv;
1162 trans->block_rsv = &root->fs_info->delayed_block_rsv;
1163
1164 delayed_root = btrfs_get_delayed_root(root);
1165
1166 curr_node = btrfs_first_delayed_node(delayed_root);
1167 while (curr_node && (!count || (count && nr--))) {
1168 ret = __btrfs_commit_inode_delayed_items(trans, path,
1169 curr_node);
1170 if (ret) {
1171 btrfs_release_delayed_node(curr_node);
1172 curr_node = NULL;
1173 btrfs_abort_transaction(trans, ret);
1174 break;
1175 }
1176
1177 prev_node = curr_node;
1178 curr_node = btrfs_next_delayed_node(curr_node);
1179 btrfs_release_delayed_node(prev_node);
1180 }
1181
1182 if (curr_node)
1183 btrfs_release_delayed_node(curr_node);
1184 btrfs_free_path(path);
1185 trans->block_rsv = block_rsv;
1186
1187 return ret;
1188 }
1189
1190 int btrfs_run_delayed_items(struct btrfs_trans_handle *trans,
1191 struct btrfs_root *root)
1192 {
1193 return __btrfs_run_delayed_items(trans, root, -1);
1194 }
1195
1196 int btrfs_run_delayed_items_nr(struct btrfs_trans_handle *trans,
1197 struct btrfs_root *root, int nr)
1198 {
1199 return __btrfs_run_delayed_items(trans, root, nr);
1200 }
1201
1202 int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1203 struct inode *inode)
1204 {
1205 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1206 struct btrfs_path *path;
1207 struct btrfs_block_rsv *block_rsv;
1208 int ret;
1209
1210 if (!delayed_node)
1211 return 0;
1212
1213 mutex_lock(&delayed_node->mutex);
1214 if (!delayed_node->count) {
1215 mutex_unlock(&delayed_node->mutex);
1216 btrfs_release_delayed_node(delayed_node);
1217 return 0;
1218 }
1219 mutex_unlock(&delayed_node->mutex);
1220
1221 path = btrfs_alloc_path();
1222 if (!path) {
1223 btrfs_release_delayed_node(delayed_node);
1224 return -ENOMEM;
1225 }
1226 path->leave_spinning = 1;
1227
1228 block_rsv = trans->block_rsv;
1229 trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv;
1230
1231 ret = __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1232
1233 btrfs_release_delayed_node(delayed_node);
1234 btrfs_free_path(path);
1235 trans->block_rsv = block_rsv;
1236
1237 return ret;
1238 }
1239
1240 int btrfs_commit_inode_delayed_inode(struct inode *inode)
1241 {
1242 struct btrfs_trans_handle *trans;
1243 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1244 struct btrfs_path *path;
1245 struct btrfs_block_rsv *block_rsv;
1246 int ret;
1247
1248 if (!delayed_node)
1249 return 0;
1250
1251 mutex_lock(&delayed_node->mutex);
1252 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1253 mutex_unlock(&delayed_node->mutex);
1254 btrfs_release_delayed_node(delayed_node);
1255 return 0;
1256 }
1257 mutex_unlock(&delayed_node->mutex);
1258
1259 trans = btrfs_join_transaction(delayed_node->root);
1260 if (IS_ERR(trans)) {
1261 ret = PTR_ERR(trans);
1262 goto out;
1263 }
1264
1265 path = btrfs_alloc_path();
1266 if (!path) {
1267 ret = -ENOMEM;
1268 goto trans_out;
1269 }
1270 path->leave_spinning = 1;
1271
1272 block_rsv = trans->block_rsv;
1273 trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv;
1274
1275 mutex_lock(&delayed_node->mutex);
1276 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags))
1277 ret = __btrfs_update_delayed_inode(trans, delayed_node->root,
1278 path, delayed_node);
1279 else
1280 ret = 0;
1281 mutex_unlock(&delayed_node->mutex);
1282
1283 btrfs_free_path(path);
1284 trans->block_rsv = block_rsv;
1285 trans_out:
1286 btrfs_end_transaction(trans, delayed_node->root);
1287 btrfs_btree_balance_dirty(delayed_node->root);
1288 out:
1289 btrfs_release_delayed_node(delayed_node);
1290
1291 return ret;
1292 }
1293
1294 void btrfs_remove_delayed_node(struct inode *inode)
1295 {
1296 struct btrfs_delayed_node *delayed_node;
1297
1298 delayed_node = ACCESS_ONCE(BTRFS_I(inode)->delayed_node);
1299 if (!delayed_node)
1300 return;
1301
1302 BTRFS_I(inode)->delayed_node = NULL;
1303 btrfs_release_delayed_node(delayed_node);
1304 }
1305
1306 struct btrfs_async_delayed_work {
1307 struct btrfs_delayed_root *delayed_root;
1308 int nr;
1309 struct btrfs_work work;
1310 };
1311
1312 static void btrfs_async_run_delayed_root(struct btrfs_work *work)
1313 {
1314 struct btrfs_async_delayed_work *async_work;
1315 struct btrfs_delayed_root *delayed_root;
1316 struct btrfs_trans_handle *trans;
1317 struct btrfs_path *path;
1318 struct btrfs_delayed_node *delayed_node = NULL;
1319 struct btrfs_root *root;
1320 struct btrfs_block_rsv *block_rsv;
1321 int total_done = 0;
1322
1323 async_work = container_of(work, struct btrfs_async_delayed_work, work);
1324 delayed_root = async_work->delayed_root;
1325
1326 path = btrfs_alloc_path();
1327 if (!path)
1328 goto out;
1329
1330 again:
1331 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND / 2)
1332 goto free_path;
1333
1334 delayed_node = btrfs_first_prepared_delayed_node(delayed_root);
1335 if (!delayed_node)
1336 goto free_path;
1337
1338 path->leave_spinning = 1;
1339 root = delayed_node->root;
1340
1341 trans = btrfs_join_transaction(root);
1342 if (IS_ERR(trans))
1343 goto release_path;
1344
1345 block_rsv = trans->block_rsv;
1346 trans->block_rsv = &root->fs_info->delayed_block_rsv;
1347
1348 __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1349
1350 trans->block_rsv = block_rsv;
1351 btrfs_end_transaction(trans, root);
1352 btrfs_btree_balance_dirty_nodelay(root);
1353
1354 release_path:
1355 btrfs_release_path(path);
1356 total_done++;
1357
1358 btrfs_release_prepared_delayed_node(delayed_node);
1359 if (async_work->nr == 0 || total_done < async_work->nr)
1360 goto again;
1361
1362 free_path:
1363 btrfs_free_path(path);
1364 out:
1365 wake_up(&delayed_root->wait);
1366 kfree(async_work);
1367 }
1368
1369
1370 static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root,
1371 struct btrfs_fs_info *fs_info, int nr)
1372 {
1373 struct btrfs_async_delayed_work *async_work;
1374
1375 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1376 return 0;
1377
1378 async_work = kmalloc(sizeof(*async_work), GFP_NOFS);
1379 if (!async_work)
1380 return -ENOMEM;
1381
1382 async_work->delayed_root = delayed_root;
1383 btrfs_init_work(&async_work->work, btrfs_delayed_meta_helper,
1384 btrfs_async_run_delayed_root, NULL, NULL);
1385 async_work->nr = nr;
1386
1387 btrfs_queue_work(fs_info->delayed_workers, &async_work->work);
1388 return 0;
1389 }
1390
1391 void btrfs_assert_delayed_root_empty(struct btrfs_root *root)
1392 {
1393 struct btrfs_delayed_root *delayed_root;
1394 delayed_root = btrfs_get_delayed_root(root);
1395 WARN_ON(btrfs_first_delayed_node(delayed_root));
1396 }
1397
1398 static int could_end_wait(struct btrfs_delayed_root *delayed_root, int seq)
1399 {
1400 int val = atomic_read(&delayed_root->items_seq);
1401
1402 if (val < seq || val >= seq + BTRFS_DELAYED_BATCH)
1403 return 1;
1404
1405 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1406 return 1;
1407
1408 return 0;
1409 }
1410
1411 void btrfs_balance_delayed_items(struct btrfs_root *root)
1412 {
1413 struct btrfs_delayed_root *delayed_root;
1414 struct btrfs_fs_info *fs_info = root->fs_info;
1415
1416 delayed_root = btrfs_get_delayed_root(root);
1417
1418 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1419 return;
1420
1421 if (atomic_read(&delayed_root->items) >= BTRFS_DELAYED_WRITEBACK) {
1422 int seq;
1423 int ret;
1424
1425 seq = atomic_read(&delayed_root->items_seq);
1426
1427 ret = btrfs_wq_run_delayed_node(delayed_root, fs_info, 0);
1428 if (ret)
1429 return;
1430
1431 wait_event_interruptible(delayed_root->wait,
1432 could_end_wait(delayed_root, seq));
1433 return;
1434 }
1435
1436 btrfs_wq_run_delayed_node(delayed_root, fs_info, BTRFS_DELAYED_BATCH);
1437 }
1438
1439 /* Will return 0 or -ENOMEM */
1440 int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans,
1441 struct btrfs_root *root, const char *name,
1442 int name_len, struct inode *dir,
1443 struct btrfs_disk_key *disk_key, u8 type,
1444 u64 index)
1445 {
1446 struct btrfs_delayed_node *delayed_node;
1447 struct btrfs_delayed_item *delayed_item;
1448 struct btrfs_dir_item *dir_item;
1449 int ret;
1450
1451 delayed_node = btrfs_get_or_create_delayed_node(dir);
1452 if (IS_ERR(delayed_node))
1453 return PTR_ERR(delayed_node);
1454
1455 delayed_item = btrfs_alloc_delayed_item(sizeof(*dir_item) + name_len);
1456 if (!delayed_item) {
1457 ret = -ENOMEM;
1458 goto release_node;
1459 }
1460
1461 delayed_item->key.objectid = btrfs_ino(dir);
1462 delayed_item->key.type = BTRFS_DIR_INDEX_KEY;
1463 delayed_item->key.offset = index;
1464
1465 dir_item = (struct btrfs_dir_item *)delayed_item->data;
1466 dir_item->location = *disk_key;
1467 btrfs_set_stack_dir_transid(dir_item, trans->transid);
1468 btrfs_set_stack_dir_data_len(dir_item, 0);
1469 btrfs_set_stack_dir_name_len(dir_item, name_len);
1470 btrfs_set_stack_dir_type(dir_item, type);
1471 memcpy((char *)(dir_item + 1), name, name_len);
1472
1473 ret = btrfs_delayed_item_reserve_metadata(trans, root, delayed_item);
1474 /*
1475 * we have reserved enough space when we start a new transaction,
1476 * so reserving metadata failure is impossible
1477 */
1478 BUG_ON(ret);
1479
1480
1481 mutex_lock(&delayed_node->mutex);
1482 ret = __btrfs_add_delayed_insertion_item(delayed_node, delayed_item);
1483 if (unlikely(ret)) {
1484 btrfs_err(root->fs_info, "err add delayed dir index item(name: %.*s) "
1485 "into the insertion tree of the delayed node"
1486 "(root id: %llu, inode id: %llu, errno: %d)",
1487 name_len, name, delayed_node->root->objectid,
1488 delayed_node->inode_id, ret);
1489 BUG();
1490 }
1491 mutex_unlock(&delayed_node->mutex);
1492
1493 release_node:
1494 btrfs_release_delayed_node(delayed_node);
1495 return ret;
1496 }
1497
1498 static int btrfs_delete_delayed_insertion_item(struct btrfs_root *root,
1499 struct btrfs_delayed_node *node,
1500 struct btrfs_key *key)
1501 {
1502 struct btrfs_delayed_item *item;
1503
1504 mutex_lock(&node->mutex);
1505 item = __btrfs_lookup_delayed_insertion_item(node, key);
1506 if (!item) {
1507 mutex_unlock(&node->mutex);
1508 return 1;
1509 }
1510
1511 btrfs_delayed_item_release_metadata(root, item);
1512 btrfs_release_delayed_item(item);
1513 mutex_unlock(&node->mutex);
1514 return 0;
1515 }
1516
1517 int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans,
1518 struct btrfs_root *root, struct inode *dir,
1519 u64 index)
1520 {
1521 struct btrfs_delayed_node *node;
1522 struct btrfs_delayed_item *item;
1523 struct btrfs_key item_key;
1524 int ret;
1525
1526 node = btrfs_get_or_create_delayed_node(dir);
1527 if (IS_ERR(node))
1528 return PTR_ERR(node);
1529
1530 item_key.objectid = btrfs_ino(dir);
1531 item_key.type = BTRFS_DIR_INDEX_KEY;
1532 item_key.offset = index;
1533
1534 ret = btrfs_delete_delayed_insertion_item(root, node, &item_key);
1535 if (!ret)
1536 goto end;
1537
1538 item = btrfs_alloc_delayed_item(0);
1539 if (!item) {
1540 ret = -ENOMEM;
1541 goto end;
1542 }
1543
1544 item->key = item_key;
1545
1546 ret = btrfs_delayed_item_reserve_metadata(trans, root, item);
1547 /*
1548 * we have reserved enough space when we start a new transaction,
1549 * so reserving metadata failure is impossible.
1550 */
1551 BUG_ON(ret);
1552
1553 mutex_lock(&node->mutex);
1554 ret = __btrfs_add_delayed_deletion_item(node, item);
1555 if (unlikely(ret)) {
1556 btrfs_err(root->fs_info, "err add delayed dir index item(index: %llu) "
1557 "into the deletion tree of the delayed node"
1558 "(root id: %llu, inode id: %llu, errno: %d)",
1559 index, node->root->objectid, node->inode_id,
1560 ret);
1561 BUG();
1562 }
1563 mutex_unlock(&node->mutex);
1564 end:
1565 btrfs_release_delayed_node(node);
1566 return ret;
1567 }
1568
1569 int btrfs_inode_delayed_dir_index_count(struct inode *inode)
1570 {
1571 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1572
1573 if (!delayed_node)
1574 return -ENOENT;
1575
1576 /*
1577 * Since we have held i_mutex of this directory, it is impossible that
1578 * a new directory index is added into the delayed node and index_cnt
1579 * is updated now. So we needn't lock the delayed node.
1580 */
1581 if (!delayed_node->index_cnt) {
1582 btrfs_release_delayed_node(delayed_node);
1583 return -EINVAL;
1584 }
1585
1586 BTRFS_I(inode)->index_cnt = delayed_node->index_cnt;
1587 btrfs_release_delayed_node(delayed_node);
1588 return 0;
1589 }
1590
1591 bool btrfs_readdir_get_delayed_items(struct inode *inode,
1592 struct list_head *ins_list,
1593 struct list_head *del_list)
1594 {
1595 struct btrfs_delayed_node *delayed_node;
1596 struct btrfs_delayed_item *item;
1597
1598 delayed_node = btrfs_get_delayed_node(inode);
1599 if (!delayed_node)
1600 return false;
1601
1602 /*
1603 * We can only do one readdir with delayed items at a time because of
1604 * item->readdir_list.
1605 */
1606 inode_unlock_shared(inode);
1607 inode_lock(inode);
1608
1609 mutex_lock(&delayed_node->mutex);
1610 item = __btrfs_first_delayed_insertion_item(delayed_node);
1611 while (item) {
1612 atomic_inc(&item->refs);
1613 list_add_tail(&item->readdir_list, ins_list);
1614 item = __btrfs_next_delayed_item(item);
1615 }
1616
1617 item = __btrfs_first_delayed_deletion_item(delayed_node);
1618 while (item) {
1619 atomic_inc(&item->refs);
1620 list_add_tail(&item->readdir_list, del_list);
1621 item = __btrfs_next_delayed_item(item);
1622 }
1623 mutex_unlock(&delayed_node->mutex);
1624 /*
1625 * This delayed node is still cached in the btrfs inode, so refs
1626 * must be > 1 now, and we needn't check it is going to be freed
1627 * or not.
1628 *
1629 * Besides that, this function is used to read dir, we do not
1630 * insert/delete delayed items in this period. So we also needn't
1631 * requeue or dequeue this delayed node.
1632 */
1633 atomic_dec(&delayed_node->refs);
1634
1635 return true;
1636 }
1637
1638 void btrfs_readdir_put_delayed_items(struct inode *inode,
1639 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 /*
1657 * The VFS is going to do up_read(), so we need to downgrade back to a
1658 * read lock.
1659 */
1660 downgrade_write(&inode->i_rwsem);
1661 }
1662
1663 int btrfs_should_delete_dir_index(struct list_head *del_list,
1664 u64 index)
1665 {
1666 struct btrfs_delayed_item *curr, *next;
1667 int ret;
1668
1669 if (list_empty(del_list))
1670 return 0;
1671
1672 list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1673 if (curr->key.offset > index)
1674 break;
1675
1676 list_del(&curr->readdir_list);
1677 ret = (curr->key.offset == index);
1678
1679 if (atomic_dec_and_test(&curr->refs))
1680 kfree(curr);
1681
1682 if (ret)
1683 return 1;
1684 else
1685 continue;
1686 }
1687 return 0;
1688 }
1689
1690 /*
1691 * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree
1692 *
1693 */
1694 int btrfs_readdir_delayed_dir_index(struct dir_context *ctx,
1695 struct list_head *ins_list, bool *emitted)
1696 {
1697 struct btrfs_dir_item *di;
1698 struct btrfs_delayed_item *curr, *next;
1699 struct btrfs_key location;
1700 char *name;
1701 int name_len;
1702 int over = 0;
1703 unsigned char d_type;
1704
1705 if (list_empty(ins_list))
1706 return 0;
1707
1708 /*
1709 * Changing the data of the delayed item is impossible. So
1710 * we needn't lock them. And we have held i_mutex of the
1711 * directory, nobody can delete any directory indexes now.
1712 */
1713 list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1714 list_del(&curr->readdir_list);
1715
1716 if (curr->key.offset < ctx->pos) {
1717 if (atomic_dec_and_test(&curr->refs))
1718 kfree(curr);
1719 continue;
1720 }
1721
1722 ctx->pos = curr->key.offset;
1723
1724 di = (struct btrfs_dir_item *)curr->data;
1725 name = (char *)(di + 1);
1726 name_len = btrfs_stack_dir_name_len(di);
1727
1728 d_type = btrfs_filetype_table[di->type];
1729 btrfs_disk_key_to_cpu(&location, &di->location);
1730
1731 over = !dir_emit(ctx, name, name_len,
1732 location.objectid, d_type);
1733
1734 if (atomic_dec_and_test(&curr->refs))
1735 kfree(curr);
1736
1737 if (over)
1738 return 1;
1739 *emitted = true;
1740 }
1741 return 0;
1742 }
1743
1744 static void fill_stack_inode_item(struct btrfs_trans_handle *trans,
1745 struct btrfs_inode_item *inode_item,
1746 struct inode *inode)
1747 {
1748 btrfs_set_stack_inode_uid(inode_item, i_uid_read(inode));
1749 btrfs_set_stack_inode_gid(inode_item, i_gid_read(inode));
1750 btrfs_set_stack_inode_size(inode_item, BTRFS_I(inode)->disk_i_size);
1751 btrfs_set_stack_inode_mode(inode_item, inode->i_mode);
1752 btrfs_set_stack_inode_nlink(inode_item, inode->i_nlink);
1753 btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(inode));
1754 btrfs_set_stack_inode_generation(inode_item,
1755 BTRFS_I(inode)->generation);
1756 btrfs_set_stack_inode_sequence(inode_item, inode->i_version);
1757 btrfs_set_stack_inode_transid(inode_item, trans->transid);
1758 btrfs_set_stack_inode_rdev(inode_item, inode->i_rdev);
1759 btrfs_set_stack_inode_flags(inode_item, BTRFS_I(inode)->flags);
1760 btrfs_set_stack_inode_block_group(inode_item, 0);
1761
1762 btrfs_set_stack_timespec_sec(&inode_item->atime,
1763 inode->i_atime.tv_sec);
1764 btrfs_set_stack_timespec_nsec(&inode_item->atime,
1765 inode->i_atime.tv_nsec);
1766
1767 btrfs_set_stack_timespec_sec(&inode_item->mtime,
1768 inode->i_mtime.tv_sec);
1769 btrfs_set_stack_timespec_nsec(&inode_item->mtime,
1770 inode->i_mtime.tv_nsec);
1771
1772 btrfs_set_stack_timespec_sec(&inode_item->ctime,
1773 inode->i_ctime.tv_sec);
1774 btrfs_set_stack_timespec_nsec(&inode_item->ctime,
1775 inode->i_ctime.tv_nsec);
1776
1777 btrfs_set_stack_timespec_sec(&inode_item->otime,
1778 BTRFS_I(inode)->i_otime.tv_sec);
1779 btrfs_set_stack_timespec_nsec(&inode_item->otime,
1780 BTRFS_I(inode)->i_otime.tv_nsec);
1781 }
1782
1783 int btrfs_fill_inode(struct inode *inode, u32 *rdev)
1784 {
1785 struct btrfs_delayed_node *delayed_node;
1786 struct btrfs_inode_item *inode_item;
1787
1788 delayed_node = btrfs_get_delayed_node(inode);
1789 if (!delayed_node)
1790 return -ENOENT;
1791
1792 mutex_lock(&delayed_node->mutex);
1793 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1794 mutex_unlock(&delayed_node->mutex);
1795 btrfs_release_delayed_node(delayed_node);
1796 return -ENOENT;
1797 }
1798
1799 inode_item = &delayed_node->inode_item;
1800
1801 i_uid_write(inode, btrfs_stack_inode_uid(inode_item));
1802 i_gid_write(inode, btrfs_stack_inode_gid(inode_item));
1803 btrfs_i_size_write(inode, btrfs_stack_inode_size(inode_item));
1804 inode->i_mode = btrfs_stack_inode_mode(inode_item);
1805 set_nlink(inode, btrfs_stack_inode_nlink(inode_item));
1806 inode_set_bytes(inode, btrfs_stack_inode_nbytes(inode_item));
1807 BTRFS_I(inode)->generation = btrfs_stack_inode_generation(inode_item);
1808 BTRFS_I(inode)->last_trans = btrfs_stack_inode_transid(inode_item);
1809
1810 inode->i_version = btrfs_stack_inode_sequence(inode_item);
1811 inode->i_rdev = 0;
1812 *rdev = btrfs_stack_inode_rdev(inode_item);
1813 BTRFS_I(inode)->flags = btrfs_stack_inode_flags(inode_item);
1814
1815 inode->i_atime.tv_sec = btrfs_stack_timespec_sec(&inode_item->atime);
1816 inode->i_atime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->atime);
1817
1818 inode->i_mtime.tv_sec = btrfs_stack_timespec_sec(&inode_item->mtime);
1819 inode->i_mtime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->mtime);
1820
1821 inode->i_ctime.tv_sec = btrfs_stack_timespec_sec(&inode_item->ctime);
1822 inode->i_ctime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->ctime);
1823
1824 BTRFS_I(inode)->i_otime.tv_sec =
1825 btrfs_stack_timespec_sec(&inode_item->otime);
1826 BTRFS_I(inode)->i_otime.tv_nsec =
1827 btrfs_stack_timespec_nsec(&inode_item->otime);
1828
1829 inode->i_generation = BTRFS_I(inode)->generation;
1830 BTRFS_I(inode)->index_cnt = (u64)-1;
1831
1832 mutex_unlock(&delayed_node->mutex);
1833 btrfs_release_delayed_node(delayed_node);
1834 return 0;
1835 }
1836
1837 int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans,
1838 struct btrfs_root *root, struct inode *inode)
1839 {
1840 struct btrfs_delayed_node *delayed_node;
1841 int ret = 0;
1842
1843 delayed_node = btrfs_get_or_create_delayed_node(inode);
1844 if (IS_ERR(delayed_node))
1845 return PTR_ERR(delayed_node);
1846
1847 mutex_lock(&delayed_node->mutex);
1848 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1849 fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1850 goto release_node;
1851 }
1852
1853 ret = btrfs_delayed_inode_reserve_metadata(trans, root, inode,
1854 delayed_node);
1855 if (ret)
1856 goto release_node;
1857
1858 fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1859 set_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
1860 delayed_node->count++;
1861 atomic_inc(&root->fs_info->delayed_root->items);
1862 release_node:
1863 mutex_unlock(&delayed_node->mutex);
1864 btrfs_release_delayed_node(delayed_node);
1865 return ret;
1866 }
1867
1868 int btrfs_delayed_delete_inode_ref(struct inode *inode)
1869 {
1870 struct btrfs_delayed_node *delayed_node;
1871
1872 /*
1873 * we don't do delayed inode updates during log recovery because it
1874 * leads to enospc problems. This means we also can't do
1875 * delayed inode refs
1876 */
1877 if (BTRFS_I(inode)->root->fs_info->log_root_recovering)
1878 return -EAGAIN;
1879
1880 delayed_node = btrfs_get_or_create_delayed_node(inode);
1881 if (IS_ERR(delayed_node))
1882 return PTR_ERR(delayed_node);
1883
1884 /*
1885 * We don't reserve space for inode ref deletion is because:
1886 * - We ONLY do async inode ref deletion for the inode who has only
1887 * one link(i_nlink == 1), it means there is only one inode ref.
1888 * And in most case, the inode ref and the inode item are in the
1889 * same leaf, and we will deal with them at the same time.
1890 * Since we are sure we will reserve the space for the inode item,
1891 * it is unnecessary to reserve space for inode ref deletion.
1892 * - If the inode ref and the inode item are not in the same leaf,
1893 * We also needn't worry about enospc problem, because we reserve
1894 * much more space for the inode update than it needs.
1895 * - At the worst, we can steal some space from the global reservation.
1896 * It is very rare.
1897 */
1898 mutex_lock(&delayed_node->mutex);
1899 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1900 goto release_node;
1901
1902 set_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
1903 delayed_node->count++;
1904 atomic_inc(&BTRFS_I(inode)->root->fs_info->delayed_root->items);
1905 release_node:
1906 mutex_unlock(&delayed_node->mutex);
1907 btrfs_release_delayed_node(delayed_node);
1908 return 0;
1909 }
1910
1911 static void __btrfs_kill_delayed_node(struct btrfs_delayed_node *delayed_node)
1912 {
1913 struct btrfs_root *root = delayed_node->root;
1914 struct btrfs_delayed_item *curr_item, *prev_item;
1915
1916 mutex_lock(&delayed_node->mutex);
1917 curr_item = __btrfs_first_delayed_insertion_item(delayed_node);
1918 while (curr_item) {
1919 btrfs_delayed_item_release_metadata(root, curr_item);
1920 prev_item = curr_item;
1921 curr_item = __btrfs_next_delayed_item(prev_item);
1922 btrfs_release_delayed_item(prev_item);
1923 }
1924
1925 curr_item = __btrfs_first_delayed_deletion_item(delayed_node);
1926 while (curr_item) {
1927 btrfs_delayed_item_release_metadata(root, curr_item);
1928 prev_item = curr_item;
1929 curr_item = __btrfs_next_delayed_item(prev_item);
1930 btrfs_release_delayed_item(prev_item);
1931 }
1932
1933 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1934 btrfs_release_delayed_iref(delayed_node);
1935
1936 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1937 btrfs_delayed_inode_release_metadata(root, delayed_node);
1938 btrfs_release_delayed_inode(delayed_node);
1939 }
1940 mutex_unlock(&delayed_node->mutex);
1941 }
1942
1943 void btrfs_kill_delayed_inode_items(struct inode *inode)
1944 {
1945 struct btrfs_delayed_node *delayed_node;
1946
1947 delayed_node = btrfs_get_delayed_node(inode);
1948 if (!delayed_node)
1949 return;
1950
1951 __btrfs_kill_delayed_node(delayed_node);
1952 btrfs_release_delayed_node(delayed_node);
1953 }
1954
1955 void btrfs_kill_all_delayed_nodes(struct btrfs_root *root)
1956 {
1957 u64 inode_id = 0;
1958 struct btrfs_delayed_node *delayed_nodes[8];
1959 int i, n;
1960
1961 while (1) {
1962 spin_lock(&root->inode_lock);
1963 n = radix_tree_gang_lookup(&root->delayed_nodes_tree,
1964 (void **)delayed_nodes, inode_id,
1965 ARRAY_SIZE(delayed_nodes));
1966 if (!n) {
1967 spin_unlock(&root->inode_lock);
1968 break;
1969 }
1970
1971 inode_id = delayed_nodes[n - 1]->inode_id + 1;
1972
1973 for (i = 0; i < n; i++)
1974 atomic_inc(&delayed_nodes[i]->refs);
1975 spin_unlock(&root->inode_lock);
1976
1977 for (i = 0; i < n; i++) {
1978 __btrfs_kill_delayed_node(delayed_nodes[i]);
1979 btrfs_release_delayed_node(delayed_nodes[i]);
1980 }
1981 }
1982 }
1983
1984 void btrfs_destroy_delayed_inodes(struct btrfs_root *root)
1985 {
1986 struct btrfs_delayed_root *delayed_root;
1987 struct btrfs_delayed_node *curr_node, *prev_node;
1988
1989 delayed_root = btrfs_get_delayed_root(root);
1990
1991 curr_node = btrfs_first_delayed_node(delayed_root);
1992 while (curr_node) {
1993 __btrfs_kill_delayed_node(curr_node);
1994
1995 prev_node = curr_node;
1996 curr_node = btrfs_next_delayed_node(curr_node);
1997 btrfs_release_delayed_node(prev_node);
1998 }
1999 }
2000
Linux with added FPC-III support