?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
btrfs: Make btrfs_delayed_inode_reserve_metadata take btrfs_inode
[linux-2.6/btrfs-unstable.git] / fs / btrfs / delayed-inode.c
blobaa7032d1c5ae640af4bf9e96bf7ef8638e116929
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 struct btrfs_delayed_node *btrfs_get_delayed_node(struct btrfs_inode *btrfs_inode)
76 {
77 struct btrfs_root *root = btrfs_inode->root;
78 u64 ino = btrfs_ino(btrfs_inode);
79 struct btrfs_delayed_node *node;
80
81 node = READ_ONCE(btrfs_inode->delayed_node);
82 if (node) {
83 atomic_inc(&node->refs);
84 return node;
85 }
86
87 spin_lock(&root->inode_lock);
88 node = radix_tree_lookup(&root->delayed_nodes_tree, ino);
89 if (node) {
90 if (btrfs_inode->delayed_node) {
91 atomic_inc(&node->refs); /* can be accessed */
92 BUG_ON(btrfs_inode->delayed_node != node);
93 spin_unlock(&root->inode_lock);
94 return node;
95 }
96 btrfs_inode->delayed_node = node;
97 /* can be accessed and cached in the inode */
98 atomic_add(2, &node->refs);
99 spin_unlock(&root->inode_lock);
100 return node;
101 }
102 spin_unlock(&root->inode_lock);
103
104 return NULL;
105 }
106
107 /* Will return either the node or PTR_ERR(-ENOMEM) */
108 static struct btrfs_delayed_node *btrfs_get_or_create_delayed_node(
109 struct btrfs_inode *btrfs_inode)
110 {
111 struct btrfs_delayed_node *node;
112 struct btrfs_root *root = btrfs_inode->root;
113 u64 ino = btrfs_ino(btrfs_inode);
114 int ret;
115
116 again:
117 node = btrfs_get_delayed_node(btrfs_inode);
118 if (node)
119 return node;
120
121 node = kmem_cache_zalloc(delayed_node_cache, GFP_NOFS);
122 if (!node)
123 return ERR_PTR(-ENOMEM);
124 btrfs_init_delayed_node(node, root, ino);
125
126 /* cached in the btrfs inode and can be accessed */
127 atomic_add(2, &node->refs);
128
129 ret = radix_tree_preload(GFP_NOFS);
130 if (ret) {
131 kmem_cache_free(delayed_node_cache, node);
132 return ERR_PTR(ret);
133 }
134
135 spin_lock(&root->inode_lock);
136 ret = radix_tree_insert(&root->delayed_nodes_tree, ino, node);
137 if (ret == -EEXIST) {
138 spin_unlock(&root->inode_lock);
139 kmem_cache_free(delayed_node_cache, node);
140 radix_tree_preload_end();
141 goto again;
142 }
143 btrfs_inode->delayed_node = node;
144 spin_unlock(&root->inode_lock);
145 radix_tree_preload_end();
146
147 return node;
148 }
149
150 /*
151 * Call it when holding delayed_node->mutex
152 *
153 * If mod = 1, add this node into the prepared list.
154 */
155 static void btrfs_queue_delayed_node(struct btrfs_delayed_root *root,
156 struct btrfs_delayed_node *node,
157 int mod)
158 {
159 spin_lock(&root->lock);
160 if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
161 if (!list_empty(&node->p_list))
162 list_move_tail(&node->p_list, &root->prepare_list);
163 else if (mod)
164 list_add_tail(&node->p_list, &root->prepare_list);
165 } else {
166 list_add_tail(&node->n_list, &root->node_list);
167 list_add_tail(&node->p_list, &root->prepare_list);
168 atomic_inc(&node->refs); /* inserted into list */
169 root->nodes++;
170 set_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
171 }
172 spin_unlock(&root->lock);
173 }
174
175 /* Call it when holding delayed_node->mutex */
176 static void btrfs_dequeue_delayed_node(struct btrfs_delayed_root *root,
177 struct btrfs_delayed_node *node)
178 {
179 spin_lock(&root->lock);
180 if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
181 root->nodes--;
182 atomic_dec(&node->refs); /* not in the list */
183 list_del_init(&node->n_list);
184 if (!list_empty(&node->p_list))
185 list_del_init(&node->p_list);
186 clear_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
187 }
188 spin_unlock(&root->lock);
189 }
190
191 static struct btrfs_delayed_node *btrfs_first_delayed_node(
192 struct btrfs_delayed_root *delayed_root)
193 {
194 struct list_head *p;
195 struct btrfs_delayed_node *node = NULL;
196
197 spin_lock(&delayed_root->lock);
198 if (list_empty(&delayed_root->node_list))
199 goto out;
200
201 p = delayed_root->node_list.next;
202 node = list_entry(p, struct btrfs_delayed_node, n_list);
203 atomic_inc(&node->refs);
204 out:
205 spin_unlock(&delayed_root->lock);
206
207 return node;
208 }
209
210 static struct btrfs_delayed_node *btrfs_next_delayed_node(
211 struct btrfs_delayed_node *node)
212 {
213 struct btrfs_delayed_root *delayed_root;
214 struct list_head *p;
215 struct btrfs_delayed_node *next = NULL;
216
217 delayed_root = node->root->fs_info->delayed_root;
218 spin_lock(&delayed_root->lock);
219 if (!test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
220 /* not in the list */
221 if (list_empty(&delayed_root->node_list))
222 goto out;
223 p = delayed_root->node_list.next;
224 } else if (list_is_last(&node->n_list, &delayed_root->node_list))
225 goto out;
226 else
227 p = node->n_list.next;
228
229 next = list_entry(p, struct btrfs_delayed_node, n_list);
230 atomic_inc(&next->refs);
231 out:
232 spin_unlock(&delayed_root->lock);
233
234 return next;
235 }
236
237 static void __btrfs_release_delayed_node(
238 struct btrfs_delayed_node *delayed_node,
239 int mod)
240 {
241 struct btrfs_delayed_root *delayed_root;
242
243 if (!delayed_node)
244 return;
245
246 delayed_root = delayed_node->root->fs_info->delayed_root;
247
248 mutex_lock(&delayed_node->mutex);
249 if (delayed_node->count)
250 btrfs_queue_delayed_node(delayed_root, delayed_node, mod);
251 else
252 btrfs_dequeue_delayed_node(delayed_root, delayed_node);
253 mutex_unlock(&delayed_node->mutex);
254
255 if (atomic_dec_and_test(&delayed_node->refs)) {
256 bool free = false;
257 struct btrfs_root *root = delayed_node->root;
258 spin_lock(&root->inode_lock);
259 if (atomic_read(&delayed_node->refs) == 0) {
260 radix_tree_delete(&root->delayed_nodes_tree,
261 delayed_node->inode_id);
262 free = true;
263 }
264 spin_unlock(&root->inode_lock);
265 if (free)
266 kmem_cache_free(delayed_node_cache, delayed_node);
267 }
268 }
269
270 static inline void btrfs_release_delayed_node(struct btrfs_delayed_node *node)
271 {
272 __btrfs_release_delayed_node(node, 0);
273 }
274
275 static struct btrfs_delayed_node *btrfs_first_prepared_delayed_node(
276 struct btrfs_delayed_root *delayed_root)
277 {
278 struct list_head *p;
279 struct btrfs_delayed_node *node = NULL;
280
281 spin_lock(&delayed_root->lock);
282 if (list_empty(&delayed_root->prepare_list))
283 goto out;
284
285 p = delayed_root->prepare_list.next;
286 list_del_init(p);
287 node = list_entry(p, struct btrfs_delayed_node, p_list);
288 atomic_inc(&node->refs);
289 out:
290 spin_unlock(&delayed_root->lock);
291
292 return node;
293 }
294
295 static inline void btrfs_release_prepared_delayed_node(
296 struct btrfs_delayed_node *node)
297 {
298 __btrfs_release_delayed_node(node, 1);
299 }
300
301 static struct btrfs_delayed_item *btrfs_alloc_delayed_item(u32 data_len)
302 {
303 struct btrfs_delayed_item *item;
304 item = kmalloc(sizeof(*item) + data_len, GFP_NOFS);
305 if (item) {
306 item->data_len = data_len;
307 item->ins_or_del = 0;
308 item->bytes_reserved = 0;
309 item->delayed_node = NULL;
310 atomic_set(&item->refs, 1);
311 }
312 return item;
313 }
314
315 /*
316 * __btrfs_lookup_delayed_item - look up the delayed item by key
317 * @delayed_node: pointer to the delayed node
318 * @key: the key to look up
319 * @prev: used to store the prev item if the right item isn't found
320 * @next: used to store the next item if the right item isn't found
321 *
322 * Note: if we don't find the right item, we will return the prev item and
323 * the next item.
324 */
325 static struct btrfs_delayed_item *__btrfs_lookup_delayed_item(
326 struct rb_root *root,
327 struct btrfs_key *key,
328 struct btrfs_delayed_item **prev,
329 struct btrfs_delayed_item **next)
330 {
331 struct rb_node *node, *prev_node = NULL;
332 struct btrfs_delayed_item *delayed_item = NULL;
333 int ret = 0;
334
335 node = root->rb_node;
336
337 while (node) {
338 delayed_item = rb_entry(node, struct btrfs_delayed_item,
339 rb_node);
340 prev_node = node;
341 ret = btrfs_comp_cpu_keys(&delayed_item->key, key);
342 if (ret < 0)
343 node = node->rb_right;
344 else if (ret > 0)
345 node = node->rb_left;
346 else
347 return delayed_item;
348 }
349
350 if (prev) {
351 if (!prev_node)
352 *prev = NULL;
353 else if (ret < 0)
354 *prev = delayed_item;
355 else if ((node = rb_prev(prev_node)) != NULL) {
356 *prev = rb_entry(node, struct btrfs_delayed_item,
357 rb_node);
358 } else
359 *prev = NULL;
360 }
361
362 if (next) {
363 if (!prev_node)
364 *next = NULL;
365 else if (ret > 0)
366 *next = delayed_item;
367 else if ((node = rb_next(prev_node)) != NULL) {
368 *next = rb_entry(node, struct btrfs_delayed_item,
369 rb_node);
370 } else
371 *next = NULL;
372 }
373 return NULL;
374 }
375
376 static struct btrfs_delayed_item *__btrfs_lookup_delayed_insertion_item(
377 struct btrfs_delayed_node *delayed_node,
378 struct btrfs_key *key)
379 {
380 return __btrfs_lookup_delayed_item(&delayed_node->ins_root, key,
381 NULL, NULL);
382 }
383
384 static int __btrfs_add_delayed_item(struct btrfs_delayed_node *delayed_node,
385 struct btrfs_delayed_item *ins,
386 int action)
387 {
388 struct rb_node **p, *node;
389 struct rb_node *parent_node = NULL;
390 struct rb_root *root;
391 struct btrfs_delayed_item *item;
392 int cmp;
393
394 if (action == BTRFS_DELAYED_INSERTION_ITEM)
395 root = &delayed_node->ins_root;
396 else if (action == BTRFS_DELAYED_DELETION_ITEM)
397 root = &delayed_node->del_root;
398 else
399 BUG();
400 p = &root->rb_node;
401 node = &ins->rb_node;
402
403 while (*p) {
404 parent_node = *p;
405 item = rb_entry(parent_node, struct btrfs_delayed_item,
406 rb_node);
407
408 cmp = btrfs_comp_cpu_keys(&item->key, &ins->key);
409 if (cmp < 0)
410 p = &(*p)->rb_right;
411 else if (cmp > 0)
412 p = &(*p)->rb_left;
413 else
414 return -EEXIST;
415 }
416
417 rb_link_node(node, parent_node, p);
418 rb_insert_color(node, root);
419 ins->delayed_node = delayed_node;
420 ins->ins_or_del = action;
421
422 if (ins->key.type == BTRFS_DIR_INDEX_KEY &&
423 action == BTRFS_DELAYED_INSERTION_ITEM &&
424 ins->key.offset >= delayed_node->index_cnt)
425 delayed_node->index_cnt = ins->key.offset + 1;
426
427 delayed_node->count++;
428 atomic_inc(&delayed_node->root->fs_info->delayed_root->items);
429 return 0;
430 }
431
432 static int __btrfs_add_delayed_insertion_item(struct btrfs_delayed_node *node,
433 struct btrfs_delayed_item *item)
434 {
435 return __btrfs_add_delayed_item(node, item,
436 BTRFS_DELAYED_INSERTION_ITEM);
437 }
438
439 static int __btrfs_add_delayed_deletion_item(struct btrfs_delayed_node *node,
440 struct btrfs_delayed_item *item)
441 {
442 return __btrfs_add_delayed_item(node, item,
443 BTRFS_DELAYED_DELETION_ITEM);
444 }
445
446 static void finish_one_item(struct btrfs_delayed_root *delayed_root)
447 {
448 int seq = atomic_inc_return(&delayed_root->items_seq);
449
450 /*
451 * atomic_dec_return implies a barrier for waitqueue_active
452 */
453 if ((atomic_dec_return(&delayed_root->items) <
454 BTRFS_DELAYED_BACKGROUND || seq % BTRFS_DELAYED_BATCH == 0) &&
455 waitqueue_active(&delayed_root->wait))
456 wake_up(&delayed_root->wait);
457 }
458
459 static void __btrfs_remove_delayed_item(struct btrfs_delayed_item *delayed_item)
460 {
461 struct rb_root *root;
462 struct btrfs_delayed_root *delayed_root;
463
464 delayed_root = delayed_item->delayed_node->root->fs_info->delayed_root;
465
466 BUG_ON(!delayed_root);
467 BUG_ON(delayed_item->ins_or_del != BTRFS_DELAYED_DELETION_ITEM &&
468 delayed_item->ins_or_del != BTRFS_DELAYED_INSERTION_ITEM);
469
470 if (delayed_item->ins_or_del == BTRFS_DELAYED_INSERTION_ITEM)
471 root = &delayed_item->delayed_node->ins_root;
472 else
473 root = &delayed_item->delayed_node->del_root;
474
475 rb_erase(&delayed_item->rb_node, root);
476 delayed_item->delayed_node->count--;
477
478 finish_one_item(delayed_root);
479 }
480
481 static void btrfs_release_delayed_item(struct btrfs_delayed_item *item)
482 {
483 if (item) {
484 __btrfs_remove_delayed_item(item);
485 if (atomic_dec_and_test(&item->refs))
486 kfree(item);
487 }
488 }
489
490 static struct btrfs_delayed_item *__btrfs_first_delayed_insertion_item(
491 struct btrfs_delayed_node *delayed_node)
492 {
493 struct rb_node *p;
494 struct btrfs_delayed_item *item = NULL;
495
496 p = rb_first(&delayed_node->ins_root);
497 if (p)
498 item = rb_entry(p, struct btrfs_delayed_item, rb_node);
499
500 return item;
501 }
502
503 static struct btrfs_delayed_item *__btrfs_first_delayed_deletion_item(
504 struct btrfs_delayed_node *delayed_node)
505 {
506 struct rb_node *p;
507 struct btrfs_delayed_item *item = NULL;
508
509 p = rb_first(&delayed_node->del_root);
510 if (p)
511 item = rb_entry(p, struct btrfs_delayed_item, rb_node);
512
513 return item;
514 }
515
516 static struct btrfs_delayed_item *__btrfs_next_delayed_item(
517 struct btrfs_delayed_item *item)
518 {
519 struct rb_node *p;
520 struct btrfs_delayed_item *next = NULL;
521
522 p = rb_next(&item->rb_node);
523 if (p)
524 next = rb_entry(p, struct btrfs_delayed_item, rb_node);
525
526 return next;
527 }
528
529 static int btrfs_delayed_item_reserve_metadata(struct btrfs_trans_handle *trans,
530 struct btrfs_fs_info *fs_info,
531 struct btrfs_delayed_item *item)
532 {
533 struct btrfs_block_rsv *src_rsv;
534 struct btrfs_block_rsv *dst_rsv;
535 u64 num_bytes;
536 int ret;
537
538 if (!trans->bytes_reserved)
539 return 0;
540
541 src_rsv = trans->block_rsv;
542 dst_rsv = &fs_info->delayed_block_rsv;
543
544 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
545 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1);
546 if (!ret) {
547 trace_btrfs_space_reservation(fs_info, "delayed_item",
548 item->key.objectid,
549 num_bytes, 1);
550 item->bytes_reserved = num_bytes;
551 }
552
553 return ret;
554 }
555
556 static void btrfs_delayed_item_release_metadata(struct btrfs_fs_info *fs_info,
557 struct btrfs_delayed_item *item)
558 {
559 struct btrfs_block_rsv *rsv;
560
561 if (!item->bytes_reserved)
562 return;
563
564 rsv = &fs_info->delayed_block_rsv;
565 trace_btrfs_space_reservation(fs_info, "delayed_item",
566 item->key.objectid, item->bytes_reserved,
567 0);
568 btrfs_block_rsv_release(fs_info, rsv,
569 item->bytes_reserved);
570 }
571
572 static int btrfs_delayed_inode_reserve_metadata(
573 struct btrfs_trans_handle *trans,
574 struct btrfs_root *root,
575 struct btrfs_inode *inode,
576 struct btrfs_delayed_node *node)
577 {
578 struct btrfs_fs_info *fs_info = root->fs_info;
579 struct btrfs_block_rsv *src_rsv;
580 struct btrfs_block_rsv *dst_rsv;
581 u64 num_bytes;
582 int ret;
583 bool release = false;
584
585 src_rsv = trans->block_rsv;
586 dst_rsv = &fs_info->delayed_block_rsv;
587
588 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
589
590 /*
591 * If our block_rsv is the delalloc block reserve then check and see if
592 * we have our extra reservation for updating the inode. If not fall
593 * through and try to reserve space quickly.
594 *
595 * We used to try and steal from the delalloc block rsv or the global
596 * reserve, but we'd steal a full reservation, which isn't kind. We are
597 * here through delalloc which means we've likely just cowed down close
598 * to the leaf that contains the inode, so we would steal less just
599 * doing the fallback inode update, so if we do end up having to steal
600 * from the global block rsv we hopefully only steal one or two blocks
601 * worth which is less likely to hurt us.
602 */
603 if (src_rsv && src_rsv->type == BTRFS_BLOCK_RSV_DELALLOC) {
604 spin_lock(&inode->lock);
605 if (test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
606 &inode->runtime_flags))
607 release = true;
608 else
609 src_rsv = NULL;
610 spin_unlock(&inode->lock);
611 }
612
613 /*
614 * btrfs_dirty_inode will update the inode under btrfs_join_transaction
615 * which doesn't reserve space for speed. This is a problem since we
616 * still need to reserve space for this update, so try to reserve the
617 * space.
618 *
619 * Now if src_rsv == delalloc_block_rsv we'll let it just steal since
620 * we're accounted for.
621 */
622 if (!src_rsv || (!trans->bytes_reserved &&
623 src_rsv->type != BTRFS_BLOCK_RSV_DELALLOC)) {
624 ret = btrfs_block_rsv_add(root, dst_rsv, num_bytes,
625 BTRFS_RESERVE_NO_FLUSH);
626 /*
627 * Since we're under a transaction reserve_metadata_bytes could
628 * try to commit the transaction which will make it return
629 * EAGAIN to make us stop the transaction we have, so return
630 * ENOSPC instead so that btrfs_dirty_inode knows what to do.
631 */
632 if (ret == -EAGAIN)
633 ret = -ENOSPC;
634 if (!ret) {
635 node->bytes_reserved = num_bytes;
636 trace_btrfs_space_reservation(fs_info,
637 "delayed_inode",
638 btrfs_ino(inode),
639 num_bytes, 1);
640 }
641 return ret;
642 }
643
644 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1);
645
646 /*
647 * Migrate only takes a reservation, it doesn't touch the size of the
648 * block_rsv. This is to simplify people who don't normally have things
649 * migrated from their block rsv. If they go to release their
650 * reservation, that will decrease the size as well, so if migrate
651 * reduced size we'd end up with a negative size. But for the
652 * delalloc_meta_reserved stuff we will only know to drop 1 reservation,
653 * but we could in fact do this reserve/migrate dance several times
654 * between the time we did the original reservation and we'd clean it
655 * up. So to take care of this, release the space for the meta
656 * reservation here. I think it may be time for a documentation page on
657 * how block rsvs. work.
658 */
659 if (!ret) {
660 trace_btrfs_space_reservation(fs_info, "delayed_inode",
661 btrfs_ino(inode), num_bytes, 1);
662 node->bytes_reserved = num_bytes;
663 }
664
665 if (release) {
666 trace_btrfs_space_reservation(fs_info, "delalloc",
667 btrfs_ino(inode), num_bytes, 0);
668 btrfs_block_rsv_release(fs_info, src_rsv, num_bytes);
669 }
670
671 return ret;
672 }
673
674 static void btrfs_delayed_inode_release_metadata(struct btrfs_fs_info *fs_info,
675 struct btrfs_delayed_node *node)
676 {
677 struct btrfs_block_rsv *rsv;
678
679 if (!node->bytes_reserved)
680 return;
681
682 rsv = &fs_info->delayed_block_rsv;
683 trace_btrfs_space_reservation(fs_info, "delayed_inode",
684 node->inode_id, node->bytes_reserved, 0);
685 btrfs_block_rsv_release(fs_info, rsv,
686 node->bytes_reserved);
687 node->bytes_reserved = 0;
688 }
689
690 /*
691 * This helper will insert some continuous items into the same leaf according
692 * to the free space of the leaf.
693 */
694 static int btrfs_batch_insert_items(struct btrfs_root *root,
695 struct btrfs_path *path,
696 struct btrfs_delayed_item *item)
697 {
698 struct btrfs_fs_info *fs_info = root->fs_info;
699 struct btrfs_delayed_item *curr, *next;
700 int free_space;
701 int total_data_size = 0, total_size = 0;
702 struct extent_buffer *leaf;
703 char *data_ptr;
704 struct btrfs_key *keys;
705 u32 *data_size;
706 struct list_head head;
707 int slot;
708 int nitems;
709 int i;
710 int ret = 0;
711
712 BUG_ON(!path->nodes[0]);
713
714 leaf = path->nodes[0];
715 free_space = btrfs_leaf_free_space(fs_info, leaf);
716 INIT_LIST_HEAD(&head);
717
718 next = item;
719 nitems = 0;
720
721 /*
722 * count the number of the continuous items that we can insert in batch
723 */
724 while (total_size + next->data_len + sizeof(struct btrfs_item) <=
725 free_space) {
726 total_data_size += next->data_len;
727 total_size += next->data_len + sizeof(struct btrfs_item);
728 list_add_tail(&next->tree_list, &head);
729 nitems++;
730
731 curr = next;
732 next = __btrfs_next_delayed_item(curr);
733 if (!next)
734 break;
735
736 if (!btrfs_is_continuous_delayed_item(curr, next))
737 break;
738 }
739
740 if (!nitems) {
741 ret = 0;
742 goto out;
743 }
744
745 /*
746 * we need allocate some memory space, but it might cause the task
747 * to sleep, so we set all locked nodes in the path to blocking locks
748 * first.
749 */
750 btrfs_set_path_blocking(path);
751
752 keys = kmalloc_array(nitems, sizeof(struct btrfs_key), GFP_NOFS);
753 if (!keys) {
754 ret = -ENOMEM;
755 goto out;
756 }
757
758 data_size = kmalloc_array(nitems, sizeof(u32), GFP_NOFS);
759 if (!data_size) {
760 ret = -ENOMEM;
761 goto error;
762 }
763
764 /* get keys of all the delayed items */
765 i = 0;
766 list_for_each_entry(next, &head, tree_list) {
767 keys[i] = next->key;
768 data_size[i] = next->data_len;
769 i++;
770 }
771
772 /* reset all the locked nodes in the patch to spinning locks. */
773 btrfs_clear_path_blocking(path, NULL, 0);
774
775 /* insert the keys of the items */
776 setup_items_for_insert(root, path, keys, data_size,
777 total_data_size, total_size, nitems);
778
779 /* insert the dir index items */
780 slot = path->slots[0];
781 list_for_each_entry_safe(curr, next, &head, tree_list) {
782 data_ptr = btrfs_item_ptr(leaf, slot, char);
783 write_extent_buffer(leaf, &curr->data,
784 (unsigned long)data_ptr,
785 curr->data_len);
786 slot++;
787
788 btrfs_delayed_item_release_metadata(fs_info, curr);
789
790 list_del(&curr->tree_list);
791 btrfs_release_delayed_item(curr);
792 }
793
794 error:
795 kfree(data_size);
796 kfree(keys);
797 out:
798 return ret;
799 }
800
801 /*
802 * This helper can just do simple insertion that needn't extend item for new
803 * data, such as directory name index insertion, inode insertion.
804 */
805 static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans,
806 struct btrfs_root *root,
807 struct btrfs_path *path,
808 struct btrfs_delayed_item *delayed_item)
809 {
810 struct btrfs_fs_info *fs_info = root->fs_info;
811 struct extent_buffer *leaf;
812 char *ptr;
813 int ret;
814
815 ret = btrfs_insert_empty_item(trans, root, path, &delayed_item->key,
816 delayed_item->data_len);
817 if (ret < 0 && ret != -EEXIST)
818 return ret;
819
820 leaf = path->nodes[0];
821
822 ptr = btrfs_item_ptr(leaf, path->slots[0], char);
823
824 write_extent_buffer(leaf, delayed_item->data, (unsigned long)ptr,
825 delayed_item->data_len);
826 btrfs_mark_buffer_dirty(leaf);
827
828 btrfs_delayed_item_release_metadata(fs_info, delayed_item);
829 return 0;
830 }
831
832 /*
833 * we insert an item first, then if there are some continuous items, we try
834 * to insert those items into the same leaf.
835 */
836 static int btrfs_insert_delayed_items(struct btrfs_trans_handle *trans,
837 struct btrfs_path *path,
838 struct btrfs_root *root,
839 struct btrfs_delayed_node *node)
840 {
841 struct btrfs_delayed_item *curr, *prev;
842 int ret = 0;
843
844 do_again:
845 mutex_lock(&node->mutex);
846 curr = __btrfs_first_delayed_insertion_item(node);
847 if (!curr)
848 goto insert_end;
849
850 ret = btrfs_insert_delayed_item(trans, root, path, curr);
851 if (ret < 0) {
852 btrfs_release_path(path);
853 goto insert_end;
854 }
855
856 prev = curr;
857 curr = __btrfs_next_delayed_item(prev);
858 if (curr && btrfs_is_continuous_delayed_item(prev, curr)) {
859 /* insert the continuous items into the same leaf */
860 path->slots[0]++;
861 btrfs_batch_insert_items(root, path, curr);
862 }
863 btrfs_release_delayed_item(prev);
864 btrfs_mark_buffer_dirty(path->nodes[0]);
865
866 btrfs_release_path(path);
867 mutex_unlock(&node->mutex);
868 goto do_again;
869
870 insert_end:
871 mutex_unlock(&node->mutex);
872 return ret;
873 }
874
875 static int btrfs_batch_delete_items(struct btrfs_trans_handle *trans,
876 struct btrfs_root *root,
877 struct btrfs_path *path,
878 struct btrfs_delayed_item *item)
879 {
880 struct btrfs_fs_info *fs_info = root->fs_info;
881 struct btrfs_delayed_item *curr, *next;
882 struct extent_buffer *leaf;
883 struct btrfs_key key;
884 struct list_head head;
885 int nitems, i, last_item;
886 int ret = 0;
887
888 BUG_ON(!path->nodes[0]);
889
890 leaf = path->nodes[0];
891
892 i = path->slots[0];
893 last_item = btrfs_header_nritems(leaf) - 1;
894 if (i > last_item)
895 return -ENOENT; /* FIXME: Is errno suitable? */
896
897 next = item;
898 INIT_LIST_HEAD(&head);
899 btrfs_item_key_to_cpu(leaf, &key, i);
900 nitems = 0;
901 /*
902 * count the number of the dir index items that we can delete in batch
903 */
904 while (btrfs_comp_cpu_keys(&next->key, &key) == 0) {
905 list_add_tail(&next->tree_list, &head);
906 nitems++;
907
908 curr = next;
909 next = __btrfs_next_delayed_item(curr);
910 if (!next)
911 break;
912
913 if (!btrfs_is_continuous_delayed_item(curr, next))
914 break;
915
916 i++;
917 if (i > last_item)
918 break;
919 btrfs_item_key_to_cpu(leaf, &key, i);
920 }
921
922 if (!nitems)
923 return 0;
924
925 ret = btrfs_del_items(trans, root, path, path->slots[0], nitems);
926 if (ret)
927 goto out;
928
929 list_for_each_entry_safe(curr, next, &head, tree_list) {
930 btrfs_delayed_item_release_metadata(fs_info, curr);
931 list_del(&curr->tree_list);
932 btrfs_release_delayed_item(curr);
933 }
934
935 out:
936 return ret;
937 }
938
939 static int btrfs_delete_delayed_items(struct btrfs_trans_handle *trans,
940 struct btrfs_path *path,
941 struct btrfs_root *root,
942 struct btrfs_delayed_node *node)
943 {
944 struct btrfs_delayed_item *curr, *prev;
945 int ret = 0;
946
947 do_again:
948 mutex_lock(&node->mutex);
949 curr = __btrfs_first_delayed_deletion_item(node);
950 if (!curr)
951 goto delete_fail;
952
953 ret = btrfs_search_slot(trans, root, &curr->key, path, -1, 1);
954 if (ret < 0)
955 goto delete_fail;
956 else if (ret > 0) {
957 /*
958 * can't find the item which the node points to, so this node
959 * is invalid, just drop it.
960 */
961 prev = curr;
962 curr = __btrfs_next_delayed_item(prev);
963 btrfs_release_delayed_item(prev);
964 ret = 0;
965 btrfs_release_path(path);
966 if (curr) {
967 mutex_unlock(&node->mutex);
968 goto do_again;
969 } else
970 goto delete_fail;
971 }
972
973 btrfs_batch_delete_items(trans, root, path, curr);
974 btrfs_release_path(path);
975 mutex_unlock(&node->mutex);
976 goto do_again;
977
978 delete_fail:
979 btrfs_release_path(path);
980 mutex_unlock(&node->mutex);
981 return ret;
982 }
983
984 static void btrfs_release_delayed_inode(struct btrfs_delayed_node *delayed_node)
985 {
986 struct btrfs_delayed_root *delayed_root;
987
988 if (delayed_node &&
989 test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
990 BUG_ON(!delayed_node->root);
991 clear_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
992 delayed_node->count--;
993
994 delayed_root = delayed_node->root->fs_info->delayed_root;
995 finish_one_item(delayed_root);
996 }
997 }
998
999 static void btrfs_release_delayed_iref(struct btrfs_delayed_node *delayed_node)
1000 {
1001 struct btrfs_delayed_root *delayed_root;
1002
1003 ASSERT(delayed_node->root);
1004 clear_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
1005 delayed_node->count--;
1006
1007 delayed_root = delayed_node->root->fs_info->delayed_root;
1008 finish_one_item(delayed_root);
1009 }
1010
1011 static int __btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1012 struct btrfs_root *root,
1013 struct btrfs_path *path,
1014 struct btrfs_delayed_node *node)
1015 {
1016 struct btrfs_fs_info *fs_info = root->fs_info;
1017 struct btrfs_key key;
1018 struct btrfs_inode_item *inode_item;
1019 struct extent_buffer *leaf;
1020 int mod;
1021 int ret;
1022
1023 key.objectid = node->inode_id;
1024 key.type = BTRFS_INODE_ITEM_KEY;
1025 key.offset = 0;
1026
1027 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1028 mod = -1;
1029 else
1030 mod = 1;
1031
1032 ret = btrfs_lookup_inode(trans, root, path, &key, mod);
1033 if (ret > 0) {
1034 btrfs_release_path(path);
1035 return -ENOENT;
1036 } else if (ret < 0) {
1037 return ret;
1038 }
1039
1040 leaf = path->nodes[0];
1041 inode_item = btrfs_item_ptr(leaf, path->slots[0],
1042 struct btrfs_inode_item);
1043 write_extent_buffer(leaf, &node->inode_item, (unsigned long)inode_item,
1044 sizeof(struct btrfs_inode_item));
1045 btrfs_mark_buffer_dirty(leaf);
1046
1047 if (!test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1048 goto no_iref;
1049
1050 path->slots[0]++;
1051 if (path->slots[0] >= btrfs_header_nritems(leaf))
1052 goto search;
1053 again:
1054 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1055 if (key.objectid != node->inode_id)
1056 goto out;
1057
1058 if (key.type != BTRFS_INODE_REF_KEY &&
1059 key.type != BTRFS_INODE_EXTREF_KEY)
1060 goto out;
1061
1062 /*
1063 * Delayed iref deletion is for the inode who has only one link,
1064 * so there is only one iref. The case that several irefs are
1065 * in the same item doesn't exist.
1066 */
1067 btrfs_del_item(trans, root, path);
1068 out:
1069 btrfs_release_delayed_iref(node);
1070 no_iref:
1071 btrfs_release_path(path);
1072 err_out:
1073 btrfs_delayed_inode_release_metadata(fs_info, node);
1074 btrfs_release_delayed_inode(node);
1075
1076 return ret;
1077
1078 search:
1079 btrfs_release_path(path);
1080
1081 key.type = BTRFS_INODE_EXTREF_KEY;
1082 key.offset = -1;
1083 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1084 if (ret < 0)
1085 goto err_out;
1086 ASSERT(ret);
1087
1088 ret = 0;
1089 leaf = path->nodes[0];
1090 path->slots[0]--;
1091 goto again;
1092 }
1093
1094 static inline int btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1095 struct btrfs_root *root,
1096 struct btrfs_path *path,
1097 struct btrfs_delayed_node *node)
1098 {
1099 int ret;
1100
1101 mutex_lock(&node->mutex);
1102 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &node->flags)) {
1103 mutex_unlock(&node->mutex);
1104 return 0;
1105 }
1106
1107 ret = __btrfs_update_delayed_inode(trans, root, path, node);
1108 mutex_unlock(&node->mutex);
1109 return ret;
1110 }
1111
1112 static inline int
1113 __btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1114 struct btrfs_path *path,
1115 struct btrfs_delayed_node *node)
1116 {
1117 int ret;
1118
1119 ret = btrfs_insert_delayed_items(trans, path, node->root, node);
1120 if (ret)
1121 return ret;
1122
1123 ret = btrfs_delete_delayed_items(trans, path, node->root, node);
1124 if (ret)
1125 return ret;
1126
1127 ret = btrfs_update_delayed_inode(trans, node->root, path, node);
1128 return ret;
1129 }
1130
1131 /*
1132 * Called when committing the transaction.
1133 * Returns 0 on success.
1134 * Returns < 0 on error and returns with an aborted transaction with any
1135 * outstanding delayed items cleaned up.
1136 */
1137 static int __btrfs_run_delayed_items(struct btrfs_trans_handle *trans,
1138 struct btrfs_fs_info *fs_info, int nr)
1139 {
1140 struct btrfs_delayed_root *delayed_root;
1141 struct btrfs_delayed_node *curr_node, *prev_node;
1142 struct btrfs_path *path;
1143 struct btrfs_block_rsv *block_rsv;
1144 int ret = 0;
1145 bool count = (nr > 0);
1146
1147 if (trans->aborted)
1148 return -EIO;
1149
1150 path = btrfs_alloc_path();
1151 if (!path)
1152 return -ENOMEM;
1153 path->leave_spinning = 1;
1154
1155 block_rsv = trans->block_rsv;
1156 trans->block_rsv = &fs_info->delayed_block_rsv;
1157
1158 delayed_root = fs_info->delayed_root;
1159
1160 curr_node = btrfs_first_delayed_node(delayed_root);
1161 while (curr_node && (!count || (count && nr--))) {
1162 ret = __btrfs_commit_inode_delayed_items(trans, path,
1163 curr_node);
1164 if (ret) {
1165 btrfs_release_delayed_node(curr_node);
1166 curr_node = NULL;
1167 btrfs_abort_transaction(trans, ret);
1168 break;
1169 }
1170
1171 prev_node = curr_node;
1172 curr_node = btrfs_next_delayed_node(curr_node);
1173 btrfs_release_delayed_node(prev_node);
1174 }
1175
1176 if (curr_node)
1177 btrfs_release_delayed_node(curr_node);
1178 btrfs_free_path(path);
1179 trans->block_rsv = block_rsv;
1180
1181 return ret;
1182 }
1183
1184 int btrfs_run_delayed_items(struct btrfs_trans_handle *trans,
1185 struct btrfs_fs_info *fs_info)
1186 {
1187 return __btrfs_run_delayed_items(trans, fs_info, -1);
1188 }
1189
1190 int btrfs_run_delayed_items_nr(struct btrfs_trans_handle *trans,
1191 struct btrfs_fs_info *fs_info, int nr)
1192 {
1193 return __btrfs_run_delayed_items(trans, fs_info, nr);
1194 }
1195
1196 int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1197 struct inode *inode)
1198 {
1199 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
1200 struct btrfs_path *path;
1201 struct btrfs_block_rsv *block_rsv;
1202 int ret;
1203
1204 if (!delayed_node)
1205 return 0;
1206
1207 mutex_lock(&delayed_node->mutex);
1208 if (!delayed_node->count) {
1209 mutex_unlock(&delayed_node->mutex);
1210 btrfs_release_delayed_node(delayed_node);
1211 return 0;
1212 }
1213 mutex_unlock(&delayed_node->mutex);
1214
1215 path = btrfs_alloc_path();
1216 if (!path) {
1217 btrfs_release_delayed_node(delayed_node);
1218 return -ENOMEM;
1219 }
1220 path->leave_spinning = 1;
1221
1222 block_rsv = trans->block_rsv;
1223 trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv;
1224
1225 ret = __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1226
1227 btrfs_release_delayed_node(delayed_node);
1228 btrfs_free_path(path);
1229 trans->block_rsv = block_rsv;
1230
1231 return ret;
1232 }
1233
1234 int btrfs_commit_inode_delayed_inode(struct inode *inode)
1235 {
1236 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1237 struct btrfs_trans_handle *trans;
1238 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
1239 struct btrfs_path *path;
1240 struct btrfs_block_rsv *block_rsv;
1241 int ret;
1242
1243 if (!delayed_node)
1244 return 0;
1245
1246 mutex_lock(&delayed_node->mutex);
1247 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1248 mutex_unlock(&delayed_node->mutex);
1249 btrfs_release_delayed_node(delayed_node);
1250 return 0;
1251 }
1252 mutex_unlock(&delayed_node->mutex);
1253
1254 trans = btrfs_join_transaction(delayed_node->root);
1255 if (IS_ERR(trans)) {
1256 ret = PTR_ERR(trans);
1257 goto out;
1258 }
1259
1260 path = btrfs_alloc_path();
1261 if (!path) {
1262 ret = -ENOMEM;
1263 goto trans_out;
1264 }
1265 path->leave_spinning = 1;
1266
1267 block_rsv = trans->block_rsv;
1268 trans->block_rsv = &fs_info->delayed_block_rsv;
1269
1270 mutex_lock(&delayed_node->mutex);
1271 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags))
1272 ret = __btrfs_update_delayed_inode(trans, delayed_node->root,
1273 path, delayed_node);
1274 else
1275 ret = 0;
1276 mutex_unlock(&delayed_node->mutex);
1277
1278 btrfs_free_path(path);
1279 trans->block_rsv = block_rsv;
1280 trans_out:
1281 btrfs_end_transaction(trans);
1282 btrfs_btree_balance_dirty(fs_info);
1283 out:
1284 btrfs_release_delayed_node(delayed_node);
1285
1286 return ret;
1287 }
1288
1289 void btrfs_remove_delayed_node(struct inode *inode)
1290 {
1291 struct btrfs_delayed_node *delayed_node;
1292
1293 delayed_node = READ_ONCE(BTRFS_I(inode)->delayed_node);
1294 if (!delayed_node)
1295 return;
1296
1297 BTRFS_I(inode)->delayed_node = NULL;
1298 btrfs_release_delayed_node(delayed_node);
1299 }
1300
1301 struct btrfs_async_delayed_work {
1302 struct btrfs_delayed_root *delayed_root;
1303 int nr;
1304 struct btrfs_work work;
1305 };
1306
1307 static void btrfs_async_run_delayed_root(struct btrfs_work *work)
1308 {
1309 struct btrfs_async_delayed_work *async_work;
1310 struct btrfs_delayed_root *delayed_root;
1311 struct btrfs_trans_handle *trans;
1312 struct btrfs_path *path;
1313 struct btrfs_delayed_node *delayed_node = NULL;
1314 struct btrfs_root *root;
1315 struct btrfs_block_rsv *block_rsv;
1316 int total_done = 0;
1317
1318 async_work = container_of(work, struct btrfs_async_delayed_work, work);
1319 delayed_root = async_work->delayed_root;
1320
1321 path = btrfs_alloc_path();
1322 if (!path)
1323 goto out;
1324
1325 again:
1326 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND / 2)
1327 goto free_path;
1328
1329 delayed_node = btrfs_first_prepared_delayed_node(delayed_root);
1330 if (!delayed_node)
1331 goto free_path;
1332
1333 path->leave_spinning = 1;
1334 root = delayed_node->root;
1335
1336 trans = btrfs_join_transaction(root);
1337 if (IS_ERR(trans))
1338 goto release_path;
1339
1340 block_rsv = trans->block_rsv;
1341 trans->block_rsv = &root->fs_info->delayed_block_rsv;
1342
1343 __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1344
1345 trans->block_rsv = block_rsv;
1346 btrfs_end_transaction(trans);
1347 btrfs_btree_balance_dirty_nodelay(root->fs_info);
1348
1349 release_path:
1350 btrfs_release_path(path);
1351 total_done++;
1352
1353 btrfs_release_prepared_delayed_node(delayed_node);
1354 if ((async_work->nr == 0 && total_done < BTRFS_DELAYED_WRITEBACK) ||
1355 total_done < async_work->nr)
1356 goto again;
1357
1358 free_path:
1359 btrfs_free_path(path);
1360 out:
1361 wake_up(&delayed_root->wait);
1362 kfree(async_work);
1363 }
1364
1365
1366 static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root,
1367 struct btrfs_fs_info *fs_info, int nr)
1368 {
1369 struct btrfs_async_delayed_work *async_work;
1370
1371 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND ||
1372 btrfs_workqueue_normal_congested(fs_info->delayed_workers))
1373 return 0;
1374
1375 async_work = kmalloc(sizeof(*async_work), GFP_NOFS);
1376 if (!async_work)
1377 return -ENOMEM;
1378
1379 async_work->delayed_root = delayed_root;
1380 btrfs_init_work(&async_work->work, btrfs_delayed_meta_helper,
1381 btrfs_async_run_delayed_root, NULL, NULL);
1382 async_work->nr = nr;
1383
1384 btrfs_queue_work(fs_info->delayed_workers, &async_work->work);
1385 return 0;
1386 }
1387
1388 void btrfs_assert_delayed_root_empty(struct btrfs_fs_info *fs_info)
1389 {
1390 WARN_ON(btrfs_first_delayed_node(fs_info->delayed_root));
1391 }
1392
1393 static int could_end_wait(struct btrfs_delayed_root *delayed_root, int seq)
1394 {
1395 int val = atomic_read(&delayed_root->items_seq);
1396
1397 if (val < seq || val >= seq + BTRFS_DELAYED_BATCH)
1398 return 1;
1399
1400 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1401 return 1;
1402
1403 return 0;
1404 }
1405
1406 void btrfs_balance_delayed_items(struct btrfs_fs_info *fs_info)
1407 {
1408 struct btrfs_delayed_root *delayed_root = fs_info->delayed_root;
1409
1410 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1411 return;
1412
1413 if (atomic_read(&delayed_root->items) >= BTRFS_DELAYED_WRITEBACK) {
1414 int seq;
1415 int ret;
1416
1417 seq = atomic_read(&delayed_root->items_seq);
1418
1419 ret = btrfs_wq_run_delayed_node(delayed_root, fs_info, 0);
1420 if (ret)
1421 return;
1422
1423 wait_event_interruptible(delayed_root->wait,
1424 could_end_wait(delayed_root, seq));
1425 return;
1426 }
1427
1428 btrfs_wq_run_delayed_node(delayed_root, fs_info, BTRFS_DELAYED_BATCH);
1429 }
1430
1431 /* Will return 0 or -ENOMEM */
1432 int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans,
1433 struct btrfs_fs_info *fs_info,
1434 const char *name, int name_len,
1435 struct inode *dir,
1436 struct btrfs_disk_key *disk_key, u8 type,
1437 u64 index)
1438 {
1439 struct btrfs_delayed_node *delayed_node;
1440 struct btrfs_delayed_item *delayed_item;
1441 struct btrfs_dir_item *dir_item;
1442 int ret;
1443
1444 delayed_node = btrfs_get_or_create_delayed_node(BTRFS_I(dir));
1445 if (IS_ERR(delayed_node))
1446 return PTR_ERR(delayed_node);
1447
1448 delayed_item = btrfs_alloc_delayed_item(sizeof(*dir_item) + name_len);
1449 if (!delayed_item) {
1450 ret = -ENOMEM;
1451 goto release_node;
1452 }
1453
1454 delayed_item->key.objectid = btrfs_ino(BTRFS_I(dir));
1455 delayed_item->key.type = BTRFS_DIR_INDEX_KEY;
1456 delayed_item->key.offset = index;
1457
1458 dir_item = (struct btrfs_dir_item *)delayed_item->data;
1459 dir_item->location = *disk_key;
1460 btrfs_set_stack_dir_transid(dir_item, trans->transid);
1461 btrfs_set_stack_dir_data_len(dir_item, 0);
1462 btrfs_set_stack_dir_name_len(dir_item, name_len);
1463 btrfs_set_stack_dir_type(dir_item, type);
1464 memcpy((char *)(dir_item + 1), name, name_len);
1465
1466 ret = btrfs_delayed_item_reserve_metadata(trans, fs_info, delayed_item);
1467 /*
1468 * we have reserved enough space when we start a new transaction,
1469 * so reserving metadata failure is impossible
1470 */
1471 BUG_ON(ret);
1472
1473
1474 mutex_lock(&delayed_node->mutex);
1475 ret = __btrfs_add_delayed_insertion_item(delayed_node, delayed_item);
1476 if (unlikely(ret)) {
1477 btrfs_err(fs_info,
1478 "err add delayed dir index item(name: %.*s) into the insertion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
1479 name_len, name, delayed_node->root->objectid,
1480 delayed_node->inode_id, ret);
1481 BUG();
1482 }
1483 mutex_unlock(&delayed_node->mutex);
1484
1485 release_node:
1486 btrfs_release_delayed_node(delayed_node);
1487 return ret;
1488 }
1489
1490 static int btrfs_delete_delayed_insertion_item(struct btrfs_fs_info *fs_info,
1491 struct btrfs_delayed_node *node,
1492 struct btrfs_key *key)
1493 {
1494 struct btrfs_delayed_item *item;
1495
1496 mutex_lock(&node->mutex);
1497 item = __btrfs_lookup_delayed_insertion_item(node, key);
1498 if (!item) {
1499 mutex_unlock(&node->mutex);
1500 return 1;
1501 }
1502
1503 btrfs_delayed_item_release_metadata(fs_info, item);
1504 btrfs_release_delayed_item(item);
1505 mutex_unlock(&node->mutex);
1506 return 0;
1507 }
1508
1509 int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans,
1510 struct btrfs_fs_info *fs_info,
1511 struct inode *dir, u64 index)
1512 {
1513 struct btrfs_delayed_node *node;
1514 struct btrfs_delayed_item *item;
1515 struct btrfs_key item_key;
1516 int ret;
1517
1518 node = btrfs_get_or_create_delayed_node(BTRFS_I(dir));
1519 if (IS_ERR(node))
1520 return PTR_ERR(node);
1521
1522 item_key.objectid = btrfs_ino(BTRFS_I(dir));
1523 item_key.type = BTRFS_DIR_INDEX_KEY;
1524 item_key.offset = index;
1525
1526 ret = btrfs_delete_delayed_insertion_item(fs_info, node, &item_key);
1527 if (!ret)
1528 goto end;
1529
1530 item = btrfs_alloc_delayed_item(0);
1531 if (!item) {
1532 ret = -ENOMEM;
1533 goto end;
1534 }
1535
1536 item->key = item_key;
1537
1538 ret = btrfs_delayed_item_reserve_metadata(trans, fs_info, item);
1539 /*
1540 * we have reserved enough space when we start a new transaction,
1541 * so reserving metadata failure is impossible.
1542 */
1543 BUG_ON(ret);
1544
1545 mutex_lock(&node->mutex);
1546 ret = __btrfs_add_delayed_deletion_item(node, item);
1547 if (unlikely(ret)) {
1548 btrfs_err(fs_info,
1549 "err add delayed dir index item(index: %llu) into the deletion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
1550 index, node->root->objectid, node->inode_id, ret);
1551 BUG();
1552 }
1553 mutex_unlock(&node->mutex);
1554 end:
1555 btrfs_release_delayed_node(node);
1556 return ret;
1557 }
1558
1559 int btrfs_inode_delayed_dir_index_count(struct inode *inode)
1560 {
1561 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
1562
1563 if (!delayed_node)
1564 return -ENOENT;
1565
1566 /*
1567 * Since we have held i_mutex of this directory, it is impossible that
1568 * a new directory index is added into the delayed node and index_cnt
1569 * is updated now. So we needn't lock the delayed node.
1570 */
1571 if (!delayed_node->index_cnt) {
1572 btrfs_release_delayed_node(delayed_node);
1573 return -EINVAL;
1574 }
1575
1576 BTRFS_I(inode)->index_cnt = delayed_node->index_cnt;
1577 btrfs_release_delayed_node(delayed_node);
1578 return 0;
1579 }
1580
1581 bool btrfs_readdir_get_delayed_items(struct inode *inode,
1582 struct list_head *ins_list,
1583 struct list_head *del_list)
1584 {
1585 struct btrfs_delayed_node *delayed_node;
1586 struct btrfs_delayed_item *item;
1587
1588 delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
1589 if (!delayed_node)
1590 return false;
1591
1592 /*
1593 * We can only do one readdir with delayed items at a time because of
1594 * item->readdir_list.
1595 */
1596 inode_unlock_shared(inode);
1597 inode_lock(inode);
1598
1599 mutex_lock(&delayed_node->mutex);
1600 item = __btrfs_first_delayed_insertion_item(delayed_node);
1601 while (item) {
1602 atomic_inc(&item->refs);
1603 list_add_tail(&item->readdir_list, ins_list);
1604 item = __btrfs_next_delayed_item(item);
1605 }
1606
1607 item = __btrfs_first_delayed_deletion_item(delayed_node);
1608 while (item) {
1609 atomic_inc(&item->refs);
1610 list_add_tail(&item->readdir_list, del_list);
1611 item = __btrfs_next_delayed_item(item);
1612 }
1613 mutex_unlock(&delayed_node->mutex);
1614 /*
1615 * This delayed node is still cached in the btrfs inode, so refs
1616 * must be > 1 now, and we needn't check it is going to be freed
1617 * or not.
1618 *
1619 * Besides that, this function is used to read dir, we do not
1620 * insert/delete delayed items in this period. So we also needn't
1621 * requeue or dequeue this delayed node.
1622 */
1623 atomic_dec(&delayed_node->refs);
1624
1625 return true;
1626 }
1627
1628 void btrfs_readdir_put_delayed_items(struct inode *inode,
1629 struct list_head *ins_list,
1630 struct list_head *del_list)
1631 {
1632 struct btrfs_delayed_item *curr, *next;
1633
1634 list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1635 list_del(&curr->readdir_list);
1636 if (atomic_dec_and_test(&curr->refs))
1637 kfree(curr);
1638 }
1639
1640 list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1641 list_del(&curr->readdir_list);
1642 if (atomic_dec_and_test(&curr->refs))
1643 kfree(curr);
1644 }
1645
1646 /*
1647 * The VFS is going to do up_read(), so we need to downgrade back to a
1648 * read lock.
1649 */
1650 downgrade_write(&inode->i_rwsem);
1651 }
1652
1653 int btrfs_should_delete_dir_index(struct list_head *del_list,
1654 u64 index)
1655 {
1656 struct btrfs_delayed_item *curr, *next;
1657 int ret;
1658
1659 if (list_empty(del_list))
1660 return 0;
1661
1662 list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1663 if (curr->key.offset > index)
1664 break;
1665
1666 list_del(&curr->readdir_list);
1667 ret = (curr->key.offset == index);
1668
1669 if (atomic_dec_and_test(&curr->refs))
1670 kfree(curr);
1671
1672 if (ret)
1673 return 1;
1674 else
1675 continue;
1676 }
1677 return 0;
1678 }
1679
1680 /*
1681 * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree
1682 *
1683 */
1684 int btrfs_readdir_delayed_dir_index(struct dir_context *ctx,
1685 struct list_head *ins_list)
1686 {
1687 struct btrfs_dir_item *di;
1688 struct btrfs_delayed_item *curr, *next;
1689 struct btrfs_key location;
1690 char *name;
1691 int name_len;
1692 int over = 0;
1693 unsigned char d_type;
1694
1695 if (list_empty(ins_list))
1696 return 0;
1697
1698 /*
1699 * Changing the data of the delayed item is impossible. So
1700 * we needn't lock them. And we have held i_mutex of the
1701 * directory, nobody can delete any directory indexes now.
1702 */
1703 list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1704 list_del(&curr->readdir_list);
1705
1706 if (curr->key.offset < ctx->pos) {
1707 if (atomic_dec_and_test(&curr->refs))
1708 kfree(curr);
1709 continue;
1710 }
1711
1712 ctx->pos = curr->key.offset;
1713
1714 di = (struct btrfs_dir_item *)curr->data;
1715 name = (char *)(di + 1);
1716 name_len = btrfs_stack_dir_name_len(di);
1717
1718 d_type = btrfs_filetype_table[di->type];
1719 btrfs_disk_key_to_cpu(&location, &di->location);
1720
1721 over = !dir_emit(ctx, name, name_len,
1722 location.objectid, d_type);
1723
1724 if (atomic_dec_and_test(&curr->refs))
1725 kfree(curr);
1726
1727 if (over)
1728 return 1;
1729 }
1730 return 0;
1731 }
1732
1733 static void fill_stack_inode_item(struct btrfs_trans_handle *trans,
1734 struct btrfs_inode_item *inode_item,
1735 struct inode *inode)
1736 {
1737 btrfs_set_stack_inode_uid(inode_item, i_uid_read(inode));
1738 btrfs_set_stack_inode_gid(inode_item, i_gid_read(inode));
1739 btrfs_set_stack_inode_size(inode_item, BTRFS_I(inode)->disk_i_size);
1740 btrfs_set_stack_inode_mode(inode_item, inode->i_mode);
1741 btrfs_set_stack_inode_nlink(inode_item, inode->i_nlink);
1742 btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(inode));
1743 btrfs_set_stack_inode_generation(inode_item,
1744 BTRFS_I(inode)->generation);
1745 btrfs_set_stack_inode_sequence(inode_item, inode->i_version);
1746 btrfs_set_stack_inode_transid(inode_item, trans->transid);
1747 btrfs_set_stack_inode_rdev(inode_item, inode->i_rdev);
1748 btrfs_set_stack_inode_flags(inode_item, BTRFS_I(inode)->flags);
1749 btrfs_set_stack_inode_block_group(inode_item, 0);
1750
1751 btrfs_set_stack_timespec_sec(&inode_item->atime,
1752 inode->i_atime.tv_sec);
1753 btrfs_set_stack_timespec_nsec(&inode_item->atime,
1754 inode->i_atime.tv_nsec);
1755
1756 btrfs_set_stack_timespec_sec(&inode_item->mtime,
1757 inode->i_mtime.tv_sec);
1758 btrfs_set_stack_timespec_nsec(&inode_item->mtime,
1759 inode->i_mtime.tv_nsec);
1760
1761 btrfs_set_stack_timespec_sec(&inode_item->ctime,
1762 inode->i_ctime.tv_sec);
1763 btrfs_set_stack_timespec_nsec(&inode_item->ctime,
1764 inode->i_ctime.tv_nsec);
1765
1766 btrfs_set_stack_timespec_sec(&inode_item->otime,
1767 BTRFS_I(inode)->i_otime.tv_sec);
1768 btrfs_set_stack_timespec_nsec(&inode_item->otime,
1769 BTRFS_I(inode)->i_otime.tv_nsec);
1770 }
1771
1772 int btrfs_fill_inode(struct inode *inode, u32 *rdev)
1773 {
1774 struct btrfs_delayed_node *delayed_node;
1775 struct btrfs_inode_item *inode_item;
1776
1777 delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
1778 if (!delayed_node)
1779 return -ENOENT;
1780
1781 mutex_lock(&delayed_node->mutex);
1782 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1783 mutex_unlock(&delayed_node->mutex);
1784 btrfs_release_delayed_node(delayed_node);
1785 return -ENOENT;
1786 }
1787
1788 inode_item = &delayed_node->inode_item;
1789
1790 i_uid_write(inode, btrfs_stack_inode_uid(inode_item));
1791 i_gid_write(inode, btrfs_stack_inode_gid(inode_item));
1792 btrfs_i_size_write(inode, btrfs_stack_inode_size(inode_item));
1793 inode->i_mode = btrfs_stack_inode_mode(inode_item);
1794 set_nlink(inode, btrfs_stack_inode_nlink(inode_item));
1795 inode_set_bytes(inode, btrfs_stack_inode_nbytes(inode_item));
1796 BTRFS_I(inode)->generation = btrfs_stack_inode_generation(inode_item);
1797 BTRFS_I(inode)->last_trans = btrfs_stack_inode_transid(inode_item);
1798
1799 inode->i_version = btrfs_stack_inode_sequence(inode_item);
1800 inode->i_rdev = 0;
1801 *rdev = btrfs_stack_inode_rdev(inode_item);
1802 BTRFS_I(inode)->flags = btrfs_stack_inode_flags(inode_item);
1803
1804 inode->i_atime.tv_sec = btrfs_stack_timespec_sec(&inode_item->atime);
1805 inode->i_atime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->atime);
1806
1807 inode->i_mtime.tv_sec = btrfs_stack_timespec_sec(&inode_item->mtime);
1808 inode->i_mtime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->mtime);
1809
1810 inode->i_ctime.tv_sec = btrfs_stack_timespec_sec(&inode_item->ctime);
1811 inode->i_ctime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->ctime);
1812
1813 BTRFS_I(inode)->i_otime.tv_sec =
1814 btrfs_stack_timespec_sec(&inode_item->otime);
1815 BTRFS_I(inode)->i_otime.tv_nsec =
1816 btrfs_stack_timespec_nsec(&inode_item->otime);
1817
1818 inode->i_generation = BTRFS_I(inode)->generation;
1819 BTRFS_I(inode)->index_cnt = (u64)-1;
1820
1821 mutex_unlock(&delayed_node->mutex);
1822 btrfs_release_delayed_node(delayed_node);
1823 return 0;
1824 }
1825
1826 int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans,
1827 struct btrfs_root *root, struct inode *inode)
1828 {
1829 struct btrfs_delayed_node *delayed_node;
1830 int ret = 0;
1831
1832 delayed_node = btrfs_get_or_create_delayed_node(BTRFS_I(inode));
1833 if (IS_ERR(delayed_node))
1834 return PTR_ERR(delayed_node);
1835
1836 mutex_lock(&delayed_node->mutex);
1837 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1838 fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1839 goto release_node;
1840 }
1841
1842 ret = btrfs_delayed_inode_reserve_metadata(trans, root, BTRFS_I(inode),
1843 delayed_node);
1844 if (ret)
1845 goto release_node;
1846
1847 fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1848 set_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
1849 delayed_node->count++;
1850 atomic_inc(&root->fs_info->delayed_root->items);
1851 release_node:
1852 mutex_unlock(&delayed_node->mutex);
1853 btrfs_release_delayed_node(delayed_node);
1854 return ret;
1855 }
1856
1857 int btrfs_delayed_delete_inode_ref(struct inode *inode)
1858 {
1859 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1860 struct btrfs_delayed_node *delayed_node;
1861
1862 /*
1863 * we don't do delayed inode updates during log recovery because it
1864 * leads to enospc problems. This means we also can't do
1865 * delayed inode refs
1866 */
1867 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
1868 return -EAGAIN;
1869
1870 delayed_node = btrfs_get_or_create_delayed_node(BTRFS_I(inode));
1871 if (IS_ERR(delayed_node))
1872 return PTR_ERR(delayed_node);
1873
1874 /*
1875 * We don't reserve space for inode ref deletion is because:
1876 * - We ONLY do async inode ref deletion for the inode who has only
1877 * one link(i_nlink == 1), it means there is only one inode ref.
1878 * And in most case, the inode ref and the inode item are in the
1879 * same leaf, and we will deal with them at the same time.
1880 * Since we are sure we will reserve the space for the inode item,
1881 * it is unnecessary to reserve space for inode ref deletion.
1882 * - If the inode ref and the inode item are not in the same leaf,
1883 * We also needn't worry about enospc problem, because we reserve
1884 * much more space for the inode update than it needs.
1885 * - At the worst, we can steal some space from the global reservation.
1886 * It is very rare.
1887 */
1888 mutex_lock(&delayed_node->mutex);
1889 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1890 goto release_node;
1891
1892 set_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
1893 delayed_node->count++;
1894 atomic_inc(&fs_info->delayed_root->items);
1895 release_node:
1896 mutex_unlock(&delayed_node->mutex);
1897 btrfs_release_delayed_node(delayed_node);
1898 return 0;
1899 }
1900
1901 static void __btrfs_kill_delayed_node(struct btrfs_delayed_node *delayed_node)
1902 {
1903 struct btrfs_root *root = delayed_node->root;
1904 struct btrfs_fs_info *fs_info = root->fs_info;
1905 struct btrfs_delayed_item *curr_item, *prev_item;
1906
1907 mutex_lock(&delayed_node->mutex);
1908 curr_item = __btrfs_first_delayed_insertion_item(delayed_node);
1909 while (curr_item) {
1910 btrfs_delayed_item_release_metadata(fs_info, curr_item);
1911 prev_item = curr_item;
1912 curr_item = __btrfs_next_delayed_item(prev_item);
1913 btrfs_release_delayed_item(prev_item);
1914 }
1915
1916 curr_item = __btrfs_first_delayed_deletion_item(delayed_node);
1917 while (curr_item) {
1918 btrfs_delayed_item_release_metadata(fs_info, curr_item);
1919 prev_item = curr_item;
1920 curr_item = __btrfs_next_delayed_item(prev_item);
1921 btrfs_release_delayed_item(prev_item);
1922 }
1923
1924 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1925 btrfs_release_delayed_iref(delayed_node);
1926
1927 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1928 btrfs_delayed_inode_release_metadata(fs_info, delayed_node);
1929 btrfs_release_delayed_inode(delayed_node);
1930 }
1931 mutex_unlock(&delayed_node->mutex);
1932 }
1933
1934 void btrfs_kill_delayed_inode_items(struct inode *inode)
1935 {
1936 struct btrfs_delayed_node *delayed_node;
1937
1938 delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
1939 if (!delayed_node)
1940 return;
1941
1942 __btrfs_kill_delayed_node(delayed_node);
1943 btrfs_release_delayed_node(delayed_node);
1944 }
1945
1946 void btrfs_kill_all_delayed_nodes(struct btrfs_root *root)
1947 {
1948 u64 inode_id = 0;
1949 struct btrfs_delayed_node *delayed_nodes[8];
1950 int i, n;
1951
1952 while (1) {
1953 spin_lock(&root->inode_lock);
1954 n = radix_tree_gang_lookup(&root->delayed_nodes_tree,
1955 (void **)delayed_nodes, inode_id,
1956 ARRAY_SIZE(delayed_nodes));
1957 if (!n) {
1958 spin_unlock(&root->inode_lock);
1959 break;
1960 }
1961
1962 inode_id = delayed_nodes[n - 1]->inode_id + 1;
1963
1964 for (i = 0; i < n; i++)
1965 atomic_inc(&delayed_nodes[i]->refs);
1966 spin_unlock(&root->inode_lock);
1967
1968 for (i = 0; i < n; i++) {
1969 __btrfs_kill_delayed_node(delayed_nodes[i]);
1970 btrfs_release_delayed_node(delayed_nodes[i]);
1971 }
1972 }
1973 }
1974
1975 void btrfs_destroy_delayed_inodes(struct btrfs_fs_info *fs_info)
1976 {
1977 struct btrfs_delayed_node *curr_node, *prev_node;
1978
1979 curr_node = btrfs_first_delayed_node(fs_info->delayed_root);
1980 while (curr_node) {
1981 __btrfs_kill_delayed_node(curr_node);
1982
1983 prev_node = curr_node;
1984 curr_node = btrfs_next_delayed_node(curr_node);
1985 btrfs_release_delayed_node(prev_node);
1986 }
1987 }
1988
(deprecated) Btrfs devel tree