projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Btrfs: deal with errors in write_dev_supers
[deliverable/linux.git] / fs / btrfs / delayed-inode.c
1 /*
2 * Copyright (C) 2011 Fujitsu. All rights reserved.
3 * Written by Miao Xie <miaox@cn.fujitsu.com>
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public
7 * License v2 as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12 * General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public
15 * License along with this program; if not, write to the
16 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
17 * Boston, MA 021110-1307, USA.
18 */
19
20 #include <linux/slab.h>
21 #include "delayed-inode.h"
22 #include "disk-io.h"
23 #include "transaction.h"
24
25 #define BTRFS_DELAYED_WRITEBACK 512
26 #define BTRFS_DELAYED_BACKGROUND 128
27 #define BTRFS_DELAYED_BATCH 16
28
29 static struct kmem_cache *delayed_node_cache;
30
31 int __init btrfs_delayed_inode_init(void)
32 {
33 delayed_node_cache = kmem_cache_create("btrfs_delayed_node",
34 sizeof(struct btrfs_delayed_node),
35 0,
36 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
37 NULL);
38 if (!delayed_node_cache)
39 return -ENOMEM;
40 return 0;
41 }
42
43 void btrfs_delayed_inode_exit(void)
44 {
45 if (delayed_node_cache)
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->count = 0;
57 delayed_node->in_list = 0;
58 delayed_node->inode_dirty = 0;
59 delayed_node->ins_root = RB_ROOT;
60 delayed_node->del_root = RB_ROOT;
61 mutex_init(&delayed_node->mutex);
62 delayed_node->index_cnt = 0;
63 INIT_LIST_HEAD(&delayed_node->n_list);
64 INIT_LIST_HEAD(&delayed_node->p_list);
65 delayed_node->bytes_reserved = 0;
66 memset(&delayed_node->inode_item, 0, sizeof(delayed_node->inode_item));
67 }
68
69 static inline int btrfs_is_continuous_delayed_item(
70 struct btrfs_delayed_item *item1,
71 struct btrfs_delayed_item *item2)
72 {
73 if (item1->key.type == BTRFS_DIR_INDEX_KEY &&
74 item1->key.objectid == item2->key.objectid &&
75 item1->key.type == item2->key.type &&
76 item1->key.offset + 1 == item2->key.offset)
77 return 1;
78 return 0;
79 }
80
81 static inline struct btrfs_delayed_root *btrfs_get_delayed_root(
82 struct btrfs_root *root)
83 {
84 return root->fs_info->delayed_root;
85 }
86
87 static struct btrfs_delayed_node *btrfs_get_delayed_node(struct inode *inode)
88 {
89 struct btrfs_inode *btrfs_inode = BTRFS_I(inode);
90 struct btrfs_root *root = btrfs_inode->root;
91 u64 ino = btrfs_ino(inode);
92 struct btrfs_delayed_node *node;
93
94 node = ACCESS_ONCE(btrfs_inode->delayed_node);
95 if (node) {
96 atomic_inc(&node->refs);
97 return node;
98 }
99
100 spin_lock(&root->inode_lock);
101 node = radix_tree_lookup(&root->delayed_nodes_tree, ino);
102 if (node) {
103 if (btrfs_inode->delayed_node) {
104 atomic_inc(&node->refs); /* can be accessed */
105 BUG_ON(btrfs_inode->delayed_node != node);
106 spin_unlock(&root->inode_lock);
107 return node;
108 }
109 btrfs_inode->delayed_node = node;
110 atomic_inc(&node->refs); /* can be accessed */
111 atomic_inc(&node->refs); /* cached in the inode */
112 spin_unlock(&root->inode_lock);
113 return node;
114 }
115 spin_unlock(&root->inode_lock);
116
117 return NULL;
118 }
119
120 /* Will return either the node or PTR_ERR(-ENOMEM) */
121 static struct btrfs_delayed_node *btrfs_get_or_create_delayed_node(
122 struct inode *inode)
123 {
124 struct btrfs_delayed_node *node;
125 struct btrfs_inode *btrfs_inode = BTRFS_I(inode);
126 struct btrfs_root *root = btrfs_inode->root;
127 u64 ino = btrfs_ino(inode);
128 int ret;
129
130 again:
131 node = btrfs_get_delayed_node(inode);
132 if (node)
133 return node;
134
135 node = kmem_cache_alloc(delayed_node_cache, GFP_NOFS);
136 if (!node)
137 return ERR_PTR(-ENOMEM);
138 btrfs_init_delayed_node(node, root, ino);
139
140 atomic_inc(&node->refs); /* cached in the btrfs inode */
141 atomic_inc(&node->refs); /* can be accessed */
142
143 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
144 if (ret) {
145 kmem_cache_free(delayed_node_cache, node);
146 return ERR_PTR(ret);
147 }
148
149 spin_lock(&root->inode_lock);
150 ret = radix_tree_insert(&root->delayed_nodes_tree, ino, node);
151 if (ret == -EEXIST) {
152 kmem_cache_free(delayed_node_cache, node);
153 spin_unlock(&root->inode_lock);
154 radix_tree_preload_end();
155 goto again;
156 }
157 btrfs_inode->delayed_node = node;
158 spin_unlock(&root->inode_lock);
159 radix_tree_preload_end();
160
161 return node;
162 }
163
164 /*
165 * Call it when holding delayed_node->mutex
166 *
167 * If mod = 1, add this node into the prepared list.
168 */
169 static void btrfs_queue_delayed_node(struct btrfs_delayed_root *root,
170 struct btrfs_delayed_node *node,
171 int mod)
172 {
173 spin_lock(&root->lock);
174 if (node->in_list) {
175 if (!list_empty(&node->p_list))
176 list_move_tail(&node->p_list, &root->prepare_list);
177 else if (mod)
178 list_add_tail(&node->p_list, &root->prepare_list);
179 } else {
180 list_add_tail(&node->n_list, &root->node_list);
181 list_add_tail(&node->p_list, &root->prepare_list);
182 atomic_inc(&node->refs); /* inserted into list */
183 root->nodes++;
184 node->in_list = 1;
185 }
186 spin_unlock(&root->lock);
187 }
188
189 /* Call it when holding delayed_node->mutex */
190 static void btrfs_dequeue_delayed_node(struct btrfs_delayed_root *root,
191 struct btrfs_delayed_node *node)
192 {
193 spin_lock(&root->lock);
194 if (node->in_list) {
195 root->nodes--;
196 atomic_dec(&node->refs); /* not in the list */
197 list_del_init(&node->n_list);
198 if (!list_empty(&node->p_list))
199 list_del_init(&node->p_list);
200 node->in_list = 0;
201 }
202 spin_unlock(&root->lock);
203 }
204
205 struct btrfs_delayed_node *btrfs_first_delayed_node(
206 struct btrfs_delayed_root *delayed_root)
207 {
208 struct list_head *p;
209 struct btrfs_delayed_node *node = NULL;
210
211 spin_lock(&delayed_root->lock);
212 if (list_empty(&delayed_root->node_list))
213 goto out;
214
215 p = delayed_root->node_list.next;
216 node = list_entry(p, struct btrfs_delayed_node, n_list);
217 atomic_inc(&node->refs);
218 out:
219 spin_unlock(&delayed_root->lock);
220
221 return node;
222 }
223
224 struct btrfs_delayed_node *btrfs_next_delayed_node(
225 struct btrfs_delayed_node *node)
226 {
227 struct btrfs_delayed_root *delayed_root;
228 struct list_head *p;
229 struct btrfs_delayed_node *next = NULL;
230
231 delayed_root = node->root->fs_info->delayed_root;
232 spin_lock(&delayed_root->lock);
233 if (!node->in_list) { /* not in the list */
234 if (list_empty(&delayed_root->node_list))
235 goto out;
236 p = delayed_root->node_list.next;
237 } else if (list_is_last(&node->n_list, &delayed_root->node_list))
238 goto out;
239 else
240 p = node->n_list.next;
241
242 next = list_entry(p, struct btrfs_delayed_node, n_list);
243 atomic_inc(&next->refs);
244 out:
245 spin_unlock(&delayed_root->lock);
246
247 return next;
248 }
249
250 static void __btrfs_release_delayed_node(
251 struct btrfs_delayed_node *delayed_node,
252 int mod)
253 {
254 struct btrfs_delayed_root *delayed_root;
255
256 if (!delayed_node)
257 return;
258
259 delayed_root = delayed_node->root->fs_info->delayed_root;
260
261 mutex_lock(&delayed_node->mutex);
262 if (delayed_node->count)
263 btrfs_queue_delayed_node(delayed_root, delayed_node, mod);
264 else
265 btrfs_dequeue_delayed_node(delayed_root, delayed_node);
266 mutex_unlock(&delayed_node->mutex);
267
268 if (atomic_dec_and_test(&delayed_node->refs)) {
269 struct btrfs_root *root = delayed_node->root;
270 spin_lock(&root->inode_lock);
271 if (atomic_read(&delayed_node->refs) == 0) {
272 radix_tree_delete(&root->delayed_nodes_tree,
273 delayed_node->inode_id);
274 kmem_cache_free(delayed_node_cache, delayed_node);
275 }
276 spin_unlock(&root->inode_lock);
277 }
278 }
279
280 static inline void btrfs_release_delayed_node(struct btrfs_delayed_node *node)
281 {
282 __btrfs_release_delayed_node(node, 0);
283 }
284
285 struct btrfs_delayed_node *btrfs_first_prepared_delayed_node(
286 struct btrfs_delayed_root *delayed_root)
287 {
288 struct list_head *p;
289 struct btrfs_delayed_node *node = NULL;
290
291 spin_lock(&delayed_root->lock);
292 if (list_empty(&delayed_root->prepare_list))
293 goto out;
294
295 p = delayed_root->prepare_list.next;
296 list_del_init(p);
297 node = list_entry(p, struct btrfs_delayed_node, p_list);
298 atomic_inc(&node->refs);
299 out:
300 spin_unlock(&delayed_root->lock);
301
302 return node;
303 }
304
305 static inline void btrfs_release_prepared_delayed_node(
306 struct btrfs_delayed_node *node)
307 {
308 __btrfs_release_delayed_node(node, 1);
309 }
310
311 struct btrfs_delayed_item *btrfs_alloc_delayed_item(u32 data_len)
312 {
313 struct btrfs_delayed_item *item;
314 item = kmalloc(sizeof(*item) + data_len, GFP_NOFS);
315 if (item) {
316 item->data_len = data_len;
317 item->ins_or_del = 0;
318 item->bytes_reserved = 0;
319 item->delayed_node = NULL;
320 atomic_set(&item->refs, 1);
321 }
322 return item;
323 }
324
325 /*
326 * __btrfs_lookup_delayed_item - look up the delayed item by key
327 * @delayed_node: pointer to the delayed node
328 * @key: the key to look up
329 * @prev: used to store the prev item if the right item isn't found
330 * @next: used to store the next item if the right item isn't found
331 *
332 * Note: if we don't find the right item, we will return the prev item and
333 * the next item.
334 */
335 static struct btrfs_delayed_item *__btrfs_lookup_delayed_item(
336 struct rb_root *root,
337 struct btrfs_key *key,
338 struct btrfs_delayed_item **prev,
339 struct btrfs_delayed_item **next)
340 {
341 struct rb_node *node, *prev_node = NULL;
342 struct btrfs_delayed_item *delayed_item = NULL;
343 int ret = 0;
344
345 node = root->rb_node;
346
347 while (node) {
348 delayed_item = rb_entry(node, struct btrfs_delayed_item,
349 rb_node);
350 prev_node = node;
351 ret = btrfs_comp_cpu_keys(&delayed_item->key, key);
352 if (ret < 0)
353 node = node->rb_right;
354 else if (ret > 0)
355 node = node->rb_left;
356 else
357 return delayed_item;
358 }
359
360 if (prev) {
361 if (!prev_node)
362 *prev = NULL;
363 else if (ret < 0)
364 *prev = delayed_item;
365 else if ((node = rb_prev(prev_node)) != NULL) {
366 *prev = rb_entry(node, struct btrfs_delayed_item,
367 rb_node);
368 } else
369 *prev = NULL;
370 }
371
372 if (next) {
373 if (!prev_node)
374 *next = NULL;
375 else if (ret > 0)
376 *next = delayed_item;
377 else if ((node = rb_next(prev_node)) != NULL) {
378 *next = rb_entry(node, struct btrfs_delayed_item,
379 rb_node);
380 } else
381 *next = NULL;
382 }
383 return NULL;
384 }
385
386 struct btrfs_delayed_item *__btrfs_lookup_delayed_insertion_item(
387 struct btrfs_delayed_node *delayed_node,
388 struct btrfs_key *key)
389 {
390 struct btrfs_delayed_item *item;
391
392 item = __btrfs_lookup_delayed_item(&delayed_node->ins_root, key,
393 NULL, NULL);
394 return item;
395 }
396
397 struct btrfs_delayed_item *__btrfs_lookup_delayed_deletion_item(
398 struct btrfs_delayed_node *delayed_node,
399 struct btrfs_key *key)
400 {
401 struct btrfs_delayed_item *item;
402
403 item = __btrfs_lookup_delayed_item(&delayed_node->del_root, key,
404 NULL, NULL);
405 return item;
406 }
407
408 struct btrfs_delayed_item *__btrfs_search_delayed_insertion_item(
409 struct btrfs_delayed_node *delayed_node,
410 struct btrfs_key *key)
411 {
412 struct btrfs_delayed_item *item, *next;
413
414 item = __btrfs_lookup_delayed_item(&delayed_node->ins_root, key,
415 NULL, &next);
416 if (!item)
417 item = next;
418
419 return item;
420 }
421
422 struct btrfs_delayed_item *__btrfs_search_delayed_deletion_item(
423 struct btrfs_delayed_node *delayed_node,
424 struct btrfs_key *key)
425 {
426 struct btrfs_delayed_item *item, *next;
427
428 item = __btrfs_lookup_delayed_item(&delayed_node->del_root, key,
429 NULL, &next);
430 if (!item)
431 item = next;
432
433 return item;
434 }
435
436 static int __btrfs_add_delayed_item(struct btrfs_delayed_node *delayed_node,
437 struct btrfs_delayed_item *ins,
438 int action)
439 {
440 struct rb_node **p, *node;
441 struct rb_node *parent_node = NULL;
442 struct rb_root *root;
443 struct btrfs_delayed_item *item;
444 int cmp;
445
446 if (action == BTRFS_DELAYED_INSERTION_ITEM)
447 root = &delayed_node->ins_root;
448 else if (action == BTRFS_DELAYED_DELETION_ITEM)
449 root = &delayed_node->del_root;
450 else
451 BUG();
452 p = &root->rb_node;
453 node = &ins->rb_node;
454
455 while (*p) {
456 parent_node = *p;
457 item = rb_entry(parent_node, struct btrfs_delayed_item,
458 rb_node);
459
460 cmp = btrfs_comp_cpu_keys(&item->key, &ins->key);
461 if (cmp < 0)
462 p = &(*p)->rb_right;
463 else if (cmp > 0)
464 p = &(*p)->rb_left;
465 else
466 return -EEXIST;
467 }
468
469 rb_link_node(node, parent_node, p);
470 rb_insert_color(node, root);
471 ins->delayed_node = delayed_node;
472 ins->ins_or_del = action;
473
474 if (ins->key.type == BTRFS_DIR_INDEX_KEY &&
475 action == BTRFS_DELAYED_INSERTION_ITEM &&
476 ins->key.offset >= delayed_node->index_cnt)
477 delayed_node->index_cnt = ins->key.offset + 1;
478
479 delayed_node->count++;
480 atomic_inc(&delayed_node->root->fs_info->delayed_root->items);
481 return 0;
482 }
483
484 static int __btrfs_add_delayed_insertion_item(struct btrfs_delayed_node *node,
485 struct btrfs_delayed_item *item)
486 {
487 return __btrfs_add_delayed_item(node, item,
488 BTRFS_DELAYED_INSERTION_ITEM);
489 }
490
491 static int __btrfs_add_delayed_deletion_item(struct btrfs_delayed_node *node,
492 struct btrfs_delayed_item *item)
493 {
494 return __btrfs_add_delayed_item(node, item,
495 BTRFS_DELAYED_DELETION_ITEM);
496 }
497
498 static void finish_one_item(struct btrfs_delayed_root *delayed_root)
499 {
500 int seq = atomic_inc_return(&delayed_root->items_seq);
501 if ((atomic_dec_return(&delayed_root->items) <
502 BTRFS_DELAYED_BACKGROUND || seq % BTRFS_DELAYED_BATCH == 0) &&
503 waitqueue_active(&delayed_root->wait))
504 wake_up(&delayed_root->wait);
505 }
506
507 static void __btrfs_remove_delayed_item(struct btrfs_delayed_item *delayed_item)
508 {
509 struct rb_root *root;
510 struct btrfs_delayed_root *delayed_root;
511
512 delayed_root = delayed_item->delayed_node->root->fs_info->delayed_root;
513
514 BUG_ON(!delayed_root);
515 BUG_ON(delayed_item->ins_or_del != BTRFS_DELAYED_DELETION_ITEM &&
516 delayed_item->ins_or_del != BTRFS_DELAYED_INSERTION_ITEM);
517
518 if (delayed_item->ins_or_del == BTRFS_DELAYED_INSERTION_ITEM)
519 root = &delayed_item->delayed_node->ins_root;
520 else
521 root = &delayed_item->delayed_node->del_root;
522
523 rb_erase(&delayed_item->rb_node, root);
524 delayed_item->delayed_node->count--;
525
526 finish_one_item(delayed_root);
527 }
528
529 static void btrfs_release_delayed_item(struct btrfs_delayed_item *item)
530 {
531 if (item) {
532 __btrfs_remove_delayed_item(item);
533 if (atomic_dec_and_test(&item->refs))
534 kfree(item);
535 }
536 }
537
538 struct btrfs_delayed_item *__btrfs_first_delayed_insertion_item(
539 struct btrfs_delayed_node *delayed_node)
540 {
541 struct rb_node *p;
542 struct btrfs_delayed_item *item = NULL;
543
544 p = rb_first(&delayed_node->ins_root);
545 if (p)
546 item = rb_entry(p, struct btrfs_delayed_item, rb_node);
547
548 return item;
549 }
550
551 struct btrfs_delayed_item *__btrfs_first_delayed_deletion_item(
552 struct btrfs_delayed_node *delayed_node)
553 {
554 struct rb_node *p;
555 struct btrfs_delayed_item *item = NULL;
556
557 p = rb_first(&delayed_node->del_root);
558 if (p)
559 item = rb_entry(p, struct btrfs_delayed_item, rb_node);
560
561 return item;
562 }
563
564 struct btrfs_delayed_item *__btrfs_next_delayed_item(
565 struct btrfs_delayed_item *item)
566 {
567 struct rb_node *p;
568 struct btrfs_delayed_item *next = NULL;
569
570 p = rb_next(&item->rb_node);
571 if (p)
572 next = rb_entry(p, struct btrfs_delayed_item, rb_node);
573
574 return next;
575 }
576
577 static inline struct btrfs_root *btrfs_get_fs_root(struct btrfs_root *root,
578 u64 root_id)
579 {
580 struct btrfs_key root_key;
581
582 if (root->objectid == root_id)
583 return root;
584
585 root_key.objectid = root_id;
586 root_key.type = BTRFS_ROOT_ITEM_KEY;
587 root_key.offset = (u64)-1;
588 return btrfs_read_fs_root_no_name(root->fs_info, &root_key);
589 }
590
591 static int btrfs_delayed_item_reserve_metadata(struct btrfs_trans_handle *trans,
592 struct btrfs_root *root,
593 struct btrfs_delayed_item *item)
594 {
595 struct btrfs_block_rsv *src_rsv;
596 struct btrfs_block_rsv *dst_rsv;
597 u64 num_bytes;
598 int ret;
599
600 if (!trans->bytes_reserved)
601 return 0;
602
603 src_rsv = trans->block_rsv;
604 dst_rsv = &root->fs_info->delayed_block_rsv;
605
606 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
607 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes);
608 if (!ret) {
609 trace_btrfs_space_reservation(root->fs_info, "delayed_item",
610 item->key.objectid,
611 num_bytes, 1);
612 item->bytes_reserved = num_bytes;
613 }
614
615 return ret;
616 }
617
618 static void btrfs_delayed_item_release_metadata(struct btrfs_root *root,
619 struct btrfs_delayed_item *item)
620 {
621 struct btrfs_block_rsv *rsv;
622
623 if (!item->bytes_reserved)
624 return;
625
626 rsv = &root->fs_info->delayed_block_rsv;
627 trace_btrfs_space_reservation(root->fs_info, "delayed_item",
628 item->key.objectid, item->bytes_reserved,
629 0);
630 btrfs_block_rsv_release(root, rsv,
631 item->bytes_reserved);
632 }
633
634 static int btrfs_delayed_inode_reserve_metadata(
635 struct btrfs_trans_handle *trans,
636 struct btrfs_root *root,
637 struct inode *inode,
638 struct btrfs_delayed_node *node)
639 {
640 struct btrfs_block_rsv *src_rsv;
641 struct btrfs_block_rsv *dst_rsv;
642 u64 num_bytes;
643 int ret;
644 bool release = false;
645
646 src_rsv = trans->block_rsv;
647 dst_rsv = &root->fs_info->delayed_block_rsv;
648
649 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
650
651 /*
652 * btrfs_dirty_inode will update the inode under btrfs_join_transaction
653 * which doesn't reserve space for speed. This is a problem since we
654 * still need to reserve space for this update, so try to reserve the
655 * space.
656 *
657 * Now if src_rsv == delalloc_block_rsv we'll let it just steal since
658 * we're accounted for.
659 */
660 if (!src_rsv || (!trans->bytes_reserved &&
661 src_rsv->type != BTRFS_BLOCK_RSV_DELALLOC)) {
662 ret = btrfs_block_rsv_add(root, dst_rsv, num_bytes,
663 BTRFS_RESERVE_NO_FLUSH);
664 /*
665 * Since we're under a transaction reserve_metadata_bytes could
666 * try to commit the transaction which will make it return
667 * EAGAIN to make us stop the transaction we have, so return
668 * ENOSPC instead so that btrfs_dirty_inode knows what to do.
669 */
670 if (ret == -EAGAIN)
671 ret = -ENOSPC;
672 if (!ret) {
673 node->bytes_reserved = num_bytes;
674 trace_btrfs_space_reservation(root->fs_info,
675 "delayed_inode",
676 btrfs_ino(inode),
677 num_bytes, 1);
678 }
679 return ret;
680 } else if (src_rsv->type == BTRFS_BLOCK_RSV_DELALLOC) {
681 spin_lock(&BTRFS_I(inode)->lock);
682 if (test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
683 &BTRFS_I(inode)->runtime_flags)) {
684 spin_unlock(&BTRFS_I(inode)->lock);
685 release = true;
686 goto migrate;
687 }
688 spin_unlock(&BTRFS_I(inode)->lock);
689
690 /* Ok we didn't have space pre-reserved. This shouldn't happen
691 * too often but it can happen if we do delalloc to an existing
692 * inode which gets dirtied because of the time update, and then
693 * isn't touched again until after the transaction commits and
694 * then we try to write out the data. First try to be nice and
695 * reserve something strictly for us. If not be a pain and try
696 * to steal from the delalloc block rsv.
697 */
698 ret = btrfs_block_rsv_add(root, dst_rsv, num_bytes,
699 BTRFS_RESERVE_NO_FLUSH);
700 if (!ret)
701 goto out;
702
703 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes);
704 if (!ret)
705 goto out;
706
707 /*
708 * Ok this is a problem, let's just steal from the global rsv
709 * since this really shouldn't happen that often.
710 */
711 WARN_ON(1);
712 ret = btrfs_block_rsv_migrate(&root->fs_info->global_block_rsv,
713 dst_rsv, num_bytes);
714 goto out;
715 }
716
717 migrate:
718 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes);
719
720 out:
721 /*
722 * Migrate only takes a reservation, it doesn't touch the size of the
723 * block_rsv. This is to simplify people who don't normally have things
724 * migrated from their block rsv. If they go to release their
725 * reservation, that will decrease the size as well, so if migrate
726 * reduced size we'd end up with a negative size. But for the
727 * delalloc_meta_reserved stuff we will only know to drop 1 reservation,
728 * but we could in fact do this reserve/migrate dance several times
729 * between the time we did the original reservation and we'd clean it
730 * up. So to take care of this, release the space for the meta
731 * reservation here. I think it may be time for a documentation page on
732 * how block rsvs. work.
733 */
734 if (!ret) {
735 trace_btrfs_space_reservation(root->fs_info, "delayed_inode",
736 btrfs_ino(inode), num_bytes, 1);
737 node->bytes_reserved = num_bytes;
738 }
739
740 if (release) {
741 trace_btrfs_space_reservation(root->fs_info, "delalloc",
742 btrfs_ino(inode), num_bytes, 0);
743 btrfs_block_rsv_release(root, src_rsv, num_bytes);
744 }
745
746 return ret;
747 }
748
749 static void btrfs_delayed_inode_release_metadata(struct btrfs_root *root,
750 struct btrfs_delayed_node *node)
751 {
752 struct btrfs_block_rsv *rsv;
753
754 if (!node->bytes_reserved)
755 return;
756
757 rsv = &root->fs_info->delayed_block_rsv;
758 trace_btrfs_space_reservation(root->fs_info, "delayed_inode",
759 node->inode_id, node->bytes_reserved, 0);
760 btrfs_block_rsv_release(root, rsv,
761 node->bytes_reserved);
762 node->bytes_reserved = 0;
763 }
764
765 /*
766 * This helper will insert some continuous items into the same leaf according
767 * to the free space of the leaf.
768 */
769 static int btrfs_batch_insert_items(struct btrfs_root *root,
770 struct btrfs_path *path,
771 struct btrfs_delayed_item *item)
772 {
773 struct btrfs_delayed_item *curr, *next;
774 int free_space;
775 int total_data_size = 0, total_size = 0;
776 struct extent_buffer *leaf;
777 char *data_ptr;
778 struct btrfs_key *keys;
779 u32 *data_size;
780 struct list_head head;
781 int slot;
782 int nitems;
783 int i;
784 int ret = 0;
785
786 BUG_ON(!path->nodes[0]);
787
788 leaf = path->nodes[0];
789 free_space = btrfs_leaf_free_space(root, leaf);
790 INIT_LIST_HEAD(&head);
791
792 next = item;
793 nitems = 0;
794
795 /*
796 * count the number of the continuous items that we can insert in batch
797 */
798 while (total_size + next->data_len + sizeof(struct btrfs_item) <=
799 free_space) {
800 total_data_size += next->data_len;
801 total_size += next->data_len + sizeof(struct btrfs_item);
802 list_add_tail(&next->tree_list, &head);
803 nitems++;
804
805 curr = next;
806 next = __btrfs_next_delayed_item(curr);
807 if (!next)
808 break;
809
810 if (!btrfs_is_continuous_delayed_item(curr, next))
811 break;
812 }
813
814 if (!nitems) {
815 ret = 0;
816 goto out;
817 }
818
819 /*
820 * we need allocate some memory space, but it might cause the task
821 * to sleep, so we set all locked nodes in the path to blocking locks
822 * first.
823 */
824 btrfs_set_path_blocking(path);
825
826 keys = kmalloc(sizeof(struct btrfs_key) * nitems, GFP_NOFS);
827 if (!keys) {
828 ret = -ENOMEM;
829 goto out;
830 }
831
832 data_size = kmalloc(sizeof(u32) * nitems, GFP_NOFS);
833 if (!data_size) {
834 ret = -ENOMEM;
835 goto error;
836 }
837
838 /* get keys of all the delayed items */
839 i = 0;
840 list_for_each_entry(next, &head, tree_list) {
841 keys[i] = next->key;
842 data_size[i] = next->data_len;
843 i++;
844 }
845
846 /* reset all the locked nodes in the patch to spinning locks. */
847 btrfs_clear_path_blocking(path, NULL, 0);
848
849 /* insert the keys of the items */
850 setup_items_for_insert(root, path, keys, data_size,
851 total_data_size, total_size, nitems);
852
853 /* insert the dir index items */
854 slot = path->slots[0];
855 list_for_each_entry_safe(curr, next, &head, tree_list) {
856 data_ptr = btrfs_item_ptr(leaf, slot, char);
857 write_extent_buffer(leaf, &curr->data,
858 (unsigned long)data_ptr,
859 curr->data_len);
860 slot++;
861
862 btrfs_delayed_item_release_metadata(root, curr);
863
864 list_del(&curr->tree_list);
865 btrfs_release_delayed_item(curr);
866 }
867
868 error:
869 kfree(data_size);
870 kfree(keys);
871 out:
872 return ret;
873 }
874
875 /*
876 * This helper can just do simple insertion that needn't extend item for new
877 * data, such as directory name index insertion, inode insertion.
878 */
879 static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans,
880 struct btrfs_root *root,
881 struct btrfs_path *path,
882 struct btrfs_delayed_item *delayed_item)
883 {
884 struct extent_buffer *leaf;
885 char *ptr;
886 int ret;
887
888 ret = btrfs_insert_empty_item(trans, root, path, &delayed_item->key,
889 delayed_item->data_len);
890 if (ret < 0 && ret != -EEXIST)
891 return ret;
892
893 leaf = path->nodes[0];
894
895 ptr = btrfs_item_ptr(leaf, path->slots[0], char);
896
897 write_extent_buffer(leaf, delayed_item->data, (unsigned long)ptr,
898 delayed_item->data_len);
899 btrfs_mark_buffer_dirty(leaf);
900
901 btrfs_delayed_item_release_metadata(root, delayed_item);
902 return 0;
903 }
904
905 /*
906 * we insert an item first, then if there are some continuous items, we try
907 * to insert those items into the same leaf.
908 */
909 static int btrfs_insert_delayed_items(struct btrfs_trans_handle *trans,
910 struct btrfs_path *path,
911 struct btrfs_root *root,
912 struct btrfs_delayed_node *node)
913 {
914 struct btrfs_delayed_item *curr, *prev;
915 int ret = 0;
916
917 do_again:
918 mutex_lock(&node->mutex);
919 curr = __btrfs_first_delayed_insertion_item(node);
920 if (!curr)
921 goto insert_end;
922
923 ret = btrfs_insert_delayed_item(trans, root, path, curr);
924 if (ret < 0) {
925 btrfs_release_path(path);
926 goto insert_end;
927 }
928
929 prev = curr;
930 curr = __btrfs_next_delayed_item(prev);
931 if (curr && btrfs_is_continuous_delayed_item(prev, curr)) {
932 /* insert the continuous items into the same leaf */
933 path->slots[0]++;
934 btrfs_batch_insert_items(root, path, curr);
935 }
936 btrfs_release_delayed_item(prev);
937 btrfs_mark_buffer_dirty(path->nodes[0]);
938
939 btrfs_release_path(path);
940 mutex_unlock(&node->mutex);
941 goto do_again;
942
943 insert_end:
944 mutex_unlock(&node->mutex);
945 return ret;
946 }
947
948 static int btrfs_batch_delete_items(struct btrfs_trans_handle *trans,
949 struct btrfs_root *root,
950 struct btrfs_path *path,
951 struct btrfs_delayed_item *item)
952 {
953 struct btrfs_delayed_item *curr, *next;
954 struct extent_buffer *leaf;
955 struct btrfs_key key;
956 struct list_head head;
957 int nitems, i, last_item;
958 int ret = 0;
959
960 BUG_ON(!path->nodes[0]);
961
962 leaf = path->nodes[0];
963
964 i = path->slots[0];
965 last_item = btrfs_header_nritems(leaf) - 1;
966 if (i > last_item)
967 return -ENOENT; /* FIXME: Is errno suitable? */
968
969 next = item;
970 INIT_LIST_HEAD(&head);
971 btrfs_item_key_to_cpu(leaf, &key, i);
972 nitems = 0;
973 /*
974 * count the number of the dir index items that we can delete in batch
975 */
976 while (btrfs_comp_cpu_keys(&next->key, &key) == 0) {
977 list_add_tail(&next->tree_list, &head);
978 nitems++;
979
980 curr = next;
981 next = __btrfs_next_delayed_item(curr);
982 if (!next)
983 break;
984
985 if (!btrfs_is_continuous_delayed_item(curr, next))
986 break;
987
988 i++;
989 if (i > last_item)
990 break;
991 btrfs_item_key_to_cpu(leaf, &key, i);
992 }
993
994 if (!nitems)
995 return 0;
996
997 ret = btrfs_del_items(trans, root, path, path->slots[0], nitems);
998 if (ret)
999 goto out;
1000
1001 list_for_each_entry_safe(curr, next, &head, tree_list) {
1002 btrfs_delayed_item_release_metadata(root, curr);
1003 list_del(&curr->tree_list);
1004 btrfs_release_delayed_item(curr);
1005 }
1006
1007 out:
1008 return ret;
1009 }
1010
1011 static int btrfs_delete_delayed_items(struct btrfs_trans_handle *trans,
1012 struct btrfs_path *path,
1013 struct btrfs_root *root,
1014 struct btrfs_delayed_node *node)
1015 {
1016 struct btrfs_delayed_item *curr, *prev;
1017 int ret = 0;
1018
1019 do_again:
1020 mutex_lock(&node->mutex);
1021 curr = __btrfs_first_delayed_deletion_item(node);
1022 if (!curr)
1023 goto delete_fail;
1024
1025 ret = btrfs_search_slot(trans, root, &curr->key, path, -1, 1);
1026 if (ret < 0)
1027 goto delete_fail;
1028 else if (ret > 0) {
1029 /*
1030 * can't find the item which the node points to, so this node
1031 * is invalid, just drop it.
1032 */
1033 prev = curr;
1034 curr = __btrfs_next_delayed_item(prev);
1035 btrfs_release_delayed_item(prev);
1036 ret = 0;
1037 btrfs_release_path(path);
1038 if (curr) {
1039 mutex_unlock(&node->mutex);
1040 goto do_again;
1041 } else
1042 goto delete_fail;
1043 }
1044
1045 btrfs_batch_delete_items(trans, root, path, curr);
1046 btrfs_release_path(path);
1047 mutex_unlock(&node->mutex);
1048 goto do_again;
1049
1050 delete_fail:
1051 btrfs_release_path(path);
1052 mutex_unlock(&node->mutex);
1053 return ret;
1054 }
1055
1056 static void btrfs_release_delayed_inode(struct btrfs_delayed_node *delayed_node)
1057 {
1058 struct btrfs_delayed_root *delayed_root;
1059
1060 if (delayed_node && delayed_node->inode_dirty) {
1061 BUG_ON(!delayed_node->root);
1062 delayed_node->inode_dirty = 0;
1063 delayed_node->count--;
1064
1065 delayed_root = delayed_node->root->fs_info->delayed_root;
1066 finish_one_item(delayed_root);
1067 }
1068 }
1069
1070 static int __btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1071 struct btrfs_root *root,
1072 struct btrfs_path *path,
1073 struct btrfs_delayed_node *node)
1074 {
1075 struct btrfs_key key;
1076 struct btrfs_inode_item *inode_item;
1077 struct extent_buffer *leaf;
1078 int ret;
1079
1080 key.objectid = node->inode_id;
1081 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
1082 key.offset = 0;
1083
1084 ret = btrfs_lookup_inode(trans, root, path, &key, 1);
1085 if (ret > 0) {
1086 btrfs_release_path(path);
1087 return -ENOENT;
1088 } else if (ret < 0) {
1089 return ret;
1090 }
1091
1092 btrfs_unlock_up_safe(path, 1);
1093 leaf = path->nodes[0];
1094 inode_item = btrfs_item_ptr(leaf, path->slots[0],
1095 struct btrfs_inode_item);
1096 write_extent_buffer(leaf, &node->inode_item, (unsigned long)inode_item,
1097 sizeof(struct btrfs_inode_item));
1098 btrfs_mark_buffer_dirty(leaf);
1099 btrfs_release_path(path);
1100
1101 btrfs_delayed_inode_release_metadata(root, node);
1102 btrfs_release_delayed_inode(node);
1103
1104 return 0;
1105 }
1106
1107 static inline int btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1108 struct btrfs_root *root,
1109 struct btrfs_path *path,
1110 struct btrfs_delayed_node *node)
1111 {
1112 int ret;
1113
1114 mutex_lock(&node->mutex);
1115 if (!node->inode_dirty) {
1116 mutex_unlock(&node->mutex);
1117 return 0;
1118 }
1119
1120 ret = __btrfs_update_delayed_inode(trans, root, path, node);
1121 mutex_unlock(&node->mutex);
1122 return ret;
1123 }
1124
1125 static inline int
1126 __btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1127 struct btrfs_path *path,
1128 struct btrfs_delayed_node *node)
1129 {
1130 int ret;
1131
1132 ret = btrfs_insert_delayed_items(trans, path, node->root, node);
1133 if (ret)
1134 return ret;
1135
1136 ret = btrfs_delete_delayed_items(trans, path, node->root, node);
1137 if (ret)
1138 return ret;
1139
1140 ret = btrfs_update_delayed_inode(trans, node->root, path, node);
1141 return ret;
1142 }
1143
1144 /*
1145 * Called when committing the transaction.
1146 * Returns 0 on success.
1147 * Returns < 0 on error and returns with an aborted transaction with any
1148 * outstanding delayed items cleaned up.
1149 */
1150 static int __btrfs_run_delayed_items(struct btrfs_trans_handle *trans,
1151 struct btrfs_root *root, int nr)
1152 {
1153 struct btrfs_delayed_root *delayed_root;
1154 struct btrfs_delayed_node *curr_node, *prev_node;
1155 struct btrfs_path *path;
1156 struct btrfs_block_rsv *block_rsv;
1157 int ret = 0;
1158 bool count = (nr > 0);
1159
1160 if (trans->aborted)
1161 return -EIO;
1162
1163 path = btrfs_alloc_path();
1164 if (!path)
1165 return -ENOMEM;
1166 path->leave_spinning = 1;
1167
1168 block_rsv = trans->block_rsv;
1169 trans->block_rsv = &root->fs_info->delayed_block_rsv;
1170
1171 delayed_root = btrfs_get_delayed_root(root);
1172
1173 curr_node = btrfs_first_delayed_node(delayed_root);
1174 while (curr_node && (!count || (count && nr--))) {
1175 ret = __btrfs_commit_inode_delayed_items(trans, path,
1176 curr_node);
1177 if (ret) {
1178 btrfs_release_delayed_node(curr_node);
1179 curr_node = NULL;
1180 btrfs_abort_transaction(trans, root, ret);
1181 break;
1182 }
1183
1184 prev_node = curr_node;
1185 curr_node = btrfs_next_delayed_node(curr_node);
1186 btrfs_release_delayed_node(prev_node);
1187 }
1188
1189 if (curr_node)
1190 btrfs_release_delayed_node(curr_node);
1191 btrfs_free_path(path);
1192 trans->block_rsv = block_rsv;
1193
1194 return ret;
1195 }
1196
1197 int btrfs_run_delayed_items(struct btrfs_trans_handle *trans,
1198 struct btrfs_root *root)
1199 {
1200 return __btrfs_run_delayed_items(trans, root, -1);
1201 }
1202
1203 int btrfs_run_delayed_items_nr(struct btrfs_trans_handle *trans,
1204 struct btrfs_root *root, int nr)
1205 {
1206 return __btrfs_run_delayed_items(trans, root, nr);
1207 }
1208
1209 int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1210 struct inode *inode)
1211 {
1212 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1213 struct btrfs_path *path;
1214 struct btrfs_block_rsv *block_rsv;
1215 int ret;
1216
1217 if (!delayed_node)
1218 return 0;
1219
1220 mutex_lock(&delayed_node->mutex);
1221 if (!delayed_node->count) {
1222 mutex_unlock(&delayed_node->mutex);
1223 btrfs_release_delayed_node(delayed_node);
1224 return 0;
1225 }
1226 mutex_unlock(&delayed_node->mutex);
1227
1228 path = btrfs_alloc_path();
1229 if (!path)
1230 return -ENOMEM;
1231 path->leave_spinning = 1;
1232
1233 block_rsv = trans->block_rsv;
1234 trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv;
1235
1236 ret = __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1237
1238 btrfs_release_delayed_node(delayed_node);
1239 btrfs_free_path(path);
1240 trans->block_rsv = block_rsv;
1241
1242 return ret;
1243 }
1244
1245 int btrfs_commit_inode_delayed_inode(struct inode *inode)
1246 {
1247 struct btrfs_trans_handle *trans;
1248 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1249 struct btrfs_path *path;
1250 struct btrfs_block_rsv *block_rsv;
1251 int ret;
1252
1253 if (!delayed_node)
1254 return 0;
1255
1256 mutex_lock(&delayed_node->mutex);
1257 if (!delayed_node->inode_dirty) {
1258 mutex_unlock(&delayed_node->mutex);
1259 btrfs_release_delayed_node(delayed_node);
1260 return 0;
1261 }
1262 mutex_unlock(&delayed_node->mutex);
1263
1264 trans = btrfs_join_transaction(delayed_node->root);
1265 if (IS_ERR(trans)) {
1266 ret = PTR_ERR(trans);
1267 goto out;
1268 }
1269
1270 path = btrfs_alloc_path();
1271 if (!path) {
1272 ret = -ENOMEM;
1273 goto trans_out;
1274 }
1275 path->leave_spinning = 1;
1276
1277 block_rsv = trans->block_rsv;
1278 trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv;
1279
1280 mutex_lock(&delayed_node->mutex);
1281 if (delayed_node->inode_dirty)
1282 ret = __btrfs_update_delayed_inode(trans, delayed_node->root,
1283 path, delayed_node);
1284 else
1285 ret = 0;
1286 mutex_unlock(&delayed_node->mutex);
1287
1288 btrfs_free_path(path);
1289 trans->block_rsv = block_rsv;
1290 trans_out:
1291 btrfs_end_transaction(trans, delayed_node->root);
1292 btrfs_btree_balance_dirty(delayed_node->root);
1293 out:
1294 btrfs_release_delayed_node(delayed_node);
1295
1296 return ret;
1297 }
1298
1299 void btrfs_remove_delayed_node(struct inode *inode)
1300 {
1301 struct btrfs_delayed_node *delayed_node;
1302
1303 delayed_node = ACCESS_ONCE(BTRFS_I(inode)->delayed_node);
1304 if (!delayed_node)
1305 return;
1306
1307 BTRFS_I(inode)->delayed_node = NULL;
1308 btrfs_release_delayed_node(delayed_node);
1309 }
1310
1311 struct btrfs_async_delayed_work {
1312 struct btrfs_delayed_root *delayed_root;
1313 int nr;
1314 struct btrfs_work work;
1315 };
1316
1317 static void btrfs_async_run_delayed_root(struct btrfs_work *work)
1318 {
1319 struct btrfs_async_delayed_work *async_work;
1320 struct btrfs_delayed_root *delayed_root;
1321 struct btrfs_trans_handle *trans;
1322 struct btrfs_path *path;
1323 struct btrfs_delayed_node *delayed_node = NULL;
1324 struct btrfs_root *root;
1325 struct btrfs_block_rsv *block_rsv;
1326 int total_done = 0;
1327
1328 async_work = container_of(work, struct btrfs_async_delayed_work, work);
1329 delayed_root = async_work->delayed_root;
1330
1331 path = btrfs_alloc_path();
1332 if (!path)
1333 goto out;
1334
1335 again:
1336 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND / 2)
1337 goto free_path;
1338
1339 delayed_node = btrfs_first_prepared_delayed_node(delayed_root);
1340 if (!delayed_node)
1341 goto free_path;
1342
1343 path->leave_spinning = 1;
1344 root = delayed_node->root;
1345
1346 trans = btrfs_join_transaction(root);
1347 if (IS_ERR(trans))
1348 goto release_path;
1349
1350 block_rsv = trans->block_rsv;
1351 trans->block_rsv = &root->fs_info->delayed_block_rsv;
1352
1353 __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1354 /*
1355 * Maybe new delayed items have been inserted, so we need requeue
1356 * the work. Besides that, we must dequeue the empty delayed nodes
1357 * to avoid the race between delayed items balance and the worker.
1358 * The race like this:
1359 * Task1 Worker thread
1360 * count == 0, needn't requeue
1361 * also needn't insert the
1362 * delayed node into prepare
1363 * list again.
1364 * add lots of delayed items
1365 * queue the delayed node
1366 * already in the list,
1367 * and not in the prepare
1368 * list, it means the delayed
1369 * node is being dealt with
1370 * by the worker.
1371 * do delayed items balance
1372 * the delayed node is being
1373 * dealt with by the worker
1374 * now, just wait.
1375 * the worker goto idle.
1376 * Task1 will sleep until the transaction is commited.
1377 */
1378 mutex_lock(&delayed_node->mutex);
1379 btrfs_dequeue_delayed_node(root->fs_info->delayed_root, delayed_node);
1380 mutex_unlock(&delayed_node->mutex);
1381
1382 trans->block_rsv = block_rsv;
1383 btrfs_end_transaction_dmeta(trans, root);
1384 btrfs_btree_balance_dirty_nodelay(root);
1385
1386 release_path:
1387 btrfs_release_path(path);
1388 total_done++;
1389
1390 btrfs_release_prepared_delayed_node(delayed_node);
1391 if (async_work->nr == 0 || total_done < async_work->nr)
1392 goto again;
1393
1394 free_path:
1395 btrfs_free_path(path);
1396 out:
1397 wake_up(&delayed_root->wait);
1398 kfree(async_work);
1399 }
1400
1401
1402 static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root,
1403 struct btrfs_root *root, int nr)
1404 {
1405 struct btrfs_async_delayed_work *async_work;
1406
1407 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1408 return 0;
1409
1410 async_work = kmalloc(sizeof(*async_work), GFP_NOFS);
1411 if (!async_work)
1412 return -ENOMEM;
1413
1414 async_work->delayed_root = delayed_root;
1415 async_work->work.func = btrfs_async_run_delayed_root;
1416 async_work->work.flags = 0;
1417 async_work->nr = nr;
1418
1419 btrfs_queue_worker(&root->fs_info->delayed_workers, &async_work->work);
1420 return 0;
1421 }
1422
1423 void btrfs_assert_delayed_root_empty(struct btrfs_root *root)
1424 {
1425 struct btrfs_delayed_root *delayed_root;
1426 delayed_root = btrfs_get_delayed_root(root);
1427 WARN_ON(btrfs_first_delayed_node(delayed_root));
1428 }
1429
1430 static int refs_newer(struct btrfs_delayed_root *delayed_root,
1431 int seq, int count)
1432 {
1433 int val = atomic_read(&delayed_root->items_seq);
1434
1435 if (val < seq || val >= seq + count)
1436 return 1;
1437 return 0;
1438 }
1439
1440 void btrfs_balance_delayed_items(struct btrfs_root *root)
1441 {
1442 struct btrfs_delayed_root *delayed_root;
1443 int seq;
1444
1445 delayed_root = btrfs_get_delayed_root(root);
1446
1447 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1448 return;
1449
1450 seq = atomic_read(&delayed_root->items_seq);
1451
1452 if (atomic_read(&delayed_root->items) >= BTRFS_DELAYED_WRITEBACK) {
1453 int ret;
1454 DEFINE_WAIT(__wait);
1455
1456 ret = btrfs_wq_run_delayed_node(delayed_root, root, 0);
1457 if (ret)
1458 return;
1459
1460 while (1) {
1461 prepare_to_wait(&delayed_root->wait, &__wait,
1462 TASK_INTERRUPTIBLE);
1463
1464 if (refs_newer(delayed_root, seq,
1465 BTRFS_DELAYED_BATCH) ||
1466 atomic_read(&delayed_root->items) <
1467 BTRFS_DELAYED_BACKGROUND) {
1468 break;
1469 }
1470 if (!signal_pending(current))
1471 schedule();
1472 else
1473 break;
1474 }
1475 finish_wait(&delayed_root->wait, &__wait);
1476 }
1477
1478 btrfs_wq_run_delayed_node(delayed_root, root, BTRFS_DELAYED_BATCH);
1479 }
1480
1481 /* Will return 0 or -ENOMEM */
1482 int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans,
1483 struct btrfs_root *root, const char *name,
1484 int name_len, struct inode *dir,
1485 struct btrfs_disk_key *disk_key, u8 type,
1486 u64 index)
1487 {
1488 struct btrfs_delayed_node *delayed_node;
1489 struct btrfs_delayed_item *delayed_item;
1490 struct btrfs_dir_item *dir_item;
1491 int ret;
1492
1493 delayed_node = btrfs_get_or_create_delayed_node(dir);
1494 if (IS_ERR(delayed_node))
1495 return PTR_ERR(delayed_node);
1496
1497 delayed_item = btrfs_alloc_delayed_item(sizeof(*dir_item) + name_len);
1498 if (!delayed_item) {
1499 ret = -ENOMEM;
1500 goto release_node;
1501 }
1502
1503 delayed_item->key.objectid = btrfs_ino(dir);
1504 btrfs_set_key_type(&delayed_item->key, BTRFS_DIR_INDEX_KEY);
1505 delayed_item->key.offset = index;
1506
1507 dir_item = (struct btrfs_dir_item *)delayed_item->data;
1508 dir_item->location = *disk_key;
1509 dir_item->transid = cpu_to_le64(trans->transid);
1510 dir_item->data_len = 0;
1511 dir_item->name_len = cpu_to_le16(name_len);
1512 dir_item->type = type;
1513 memcpy((char *)(dir_item + 1), name, name_len);
1514
1515 ret = btrfs_delayed_item_reserve_metadata(trans, root, delayed_item);
1516 /*
1517 * we have reserved enough space when we start a new transaction,
1518 * so reserving metadata failure is impossible
1519 */
1520 BUG_ON(ret);
1521
1522
1523 mutex_lock(&delayed_node->mutex);
1524 ret = __btrfs_add_delayed_insertion_item(delayed_node, delayed_item);
1525 if (unlikely(ret)) {
1526 printk(KERN_ERR "err add delayed dir index item(name: %s) into "
1527 "the insertion tree of the delayed node"
1528 "(root id: %llu, inode id: %llu, errno: %d)\n",
1529 name,
1530 (unsigned long long)delayed_node->root->objectid,
1531 (unsigned long long)delayed_node->inode_id,
1532 ret);
1533 BUG();
1534 }
1535 mutex_unlock(&delayed_node->mutex);
1536
1537 release_node:
1538 btrfs_release_delayed_node(delayed_node);
1539 return ret;
1540 }
1541
1542 static int btrfs_delete_delayed_insertion_item(struct btrfs_root *root,
1543 struct btrfs_delayed_node *node,
1544 struct btrfs_key *key)
1545 {
1546 struct btrfs_delayed_item *item;
1547
1548 mutex_lock(&node->mutex);
1549 item = __btrfs_lookup_delayed_insertion_item(node, key);
1550 if (!item) {
1551 mutex_unlock(&node->mutex);
1552 return 1;
1553 }
1554
1555 btrfs_delayed_item_release_metadata(root, item);
1556 btrfs_release_delayed_item(item);
1557 mutex_unlock(&node->mutex);
1558 return 0;
1559 }
1560
1561 int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans,
1562 struct btrfs_root *root, struct inode *dir,
1563 u64 index)
1564 {
1565 struct btrfs_delayed_node *node;
1566 struct btrfs_delayed_item *item;
1567 struct btrfs_key item_key;
1568 int ret;
1569
1570 node = btrfs_get_or_create_delayed_node(dir);
1571 if (IS_ERR(node))
1572 return PTR_ERR(node);
1573
1574 item_key.objectid = btrfs_ino(dir);
1575 btrfs_set_key_type(&item_key, BTRFS_DIR_INDEX_KEY);
1576 item_key.offset = index;
1577
1578 ret = btrfs_delete_delayed_insertion_item(root, node, &item_key);
1579 if (!ret)
1580 goto end;
1581
1582 item = btrfs_alloc_delayed_item(0);
1583 if (!item) {
1584 ret = -ENOMEM;
1585 goto end;
1586 }
1587
1588 item->key = item_key;
1589
1590 ret = btrfs_delayed_item_reserve_metadata(trans, root, item);
1591 /*
1592 * we have reserved enough space when we start a new transaction,
1593 * so reserving metadata failure is impossible.
1594 */
1595 BUG_ON(ret);
1596
1597 mutex_lock(&node->mutex);
1598 ret = __btrfs_add_delayed_deletion_item(node, item);
1599 if (unlikely(ret)) {
1600 printk(KERN_ERR "err add delayed dir index item(index: %llu) "
1601 "into the deletion tree of the delayed node"
1602 "(root id: %llu, inode id: %llu, errno: %d)\n",
1603 (unsigned long long)index,
1604 (unsigned long long)node->root->objectid,
1605 (unsigned long long)node->inode_id,
1606 ret);
1607 BUG();
1608 }
1609 mutex_unlock(&node->mutex);
1610 end:
1611 btrfs_release_delayed_node(node);
1612 return ret;
1613 }
1614
1615 int btrfs_inode_delayed_dir_index_count(struct inode *inode)
1616 {
1617 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1618
1619 if (!delayed_node)
1620 return -ENOENT;
1621
1622 /*
1623 * Since we have held i_mutex of this directory, it is impossible that
1624 * a new directory index is added into the delayed node and index_cnt
1625 * is updated now. So we needn't lock the delayed node.
1626 */
1627 if (!delayed_node->index_cnt) {
1628 btrfs_release_delayed_node(delayed_node);
1629 return -EINVAL;
1630 }
1631
1632 BTRFS_I(inode)->index_cnt = delayed_node->index_cnt;
1633 btrfs_release_delayed_node(delayed_node);
1634 return 0;
1635 }
1636
1637 void btrfs_get_delayed_items(struct inode *inode, struct list_head *ins_list,
1638 struct list_head *del_list)
1639 {
1640 struct btrfs_delayed_node *delayed_node;
1641 struct btrfs_delayed_item *item;
1642
1643 delayed_node = btrfs_get_delayed_node(inode);
1644 if (!delayed_node)
1645 return;
1646
1647 mutex_lock(&delayed_node->mutex);
1648 item = __btrfs_first_delayed_insertion_item(delayed_node);
1649 while (item) {
1650 atomic_inc(&item->refs);
1651 list_add_tail(&item->readdir_list, ins_list);
1652 item = __btrfs_next_delayed_item(item);
1653 }
1654
1655 item = __btrfs_first_delayed_deletion_item(delayed_node);
1656 while (item) {
1657 atomic_inc(&item->refs);
1658 list_add_tail(&item->readdir_list, del_list);
1659 item = __btrfs_next_delayed_item(item);
1660 }
1661 mutex_unlock(&delayed_node->mutex);
1662 /*
1663 * This delayed node is still cached in the btrfs inode, so refs
1664 * must be > 1 now, and we needn't check it is going to be freed
1665 * or not.
1666 *
1667 * Besides that, this function is used to read dir, we do not
1668 * insert/delete delayed items in this period. So we also needn't
1669 * requeue or dequeue this delayed node.
1670 */
1671 atomic_dec(&delayed_node->refs);
1672 }
1673
1674 void btrfs_put_delayed_items(struct list_head *ins_list,
1675 struct list_head *del_list)
1676 {
1677 struct btrfs_delayed_item *curr, *next;
1678
1679 list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1680 list_del(&curr->readdir_list);
1681 if (atomic_dec_and_test(&curr->refs))
1682 kfree(curr);
1683 }
1684
1685 list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1686 list_del(&curr->readdir_list);
1687 if (atomic_dec_and_test(&curr->refs))
1688 kfree(curr);
1689 }
1690 }
1691
1692 int btrfs_should_delete_dir_index(struct list_head *del_list,
1693 u64 index)
1694 {
1695 struct btrfs_delayed_item *curr, *next;
1696 int ret;
1697
1698 if (list_empty(del_list))
1699 return 0;
1700
1701 list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1702 if (curr->key.offset > index)
1703 break;
1704
1705 list_del(&curr->readdir_list);
1706 ret = (curr->key.offset == index);
1707
1708 if (atomic_dec_and_test(&curr->refs))
1709 kfree(curr);
1710
1711 if (ret)
1712 return 1;
1713 else
1714 continue;
1715 }
1716 return 0;
1717 }
1718
1719 /*
1720 * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree
1721 *
1722 */
1723 int btrfs_readdir_delayed_dir_index(struct file *filp, void *dirent,
1724 filldir_t filldir,
1725 struct list_head *ins_list)
1726 {
1727 struct btrfs_dir_item *di;
1728 struct btrfs_delayed_item *curr, *next;
1729 struct btrfs_key location;
1730 char *name;
1731 int name_len;
1732 int over = 0;
1733 unsigned char d_type;
1734
1735 if (list_empty(ins_list))
1736 return 0;
1737
1738 /*
1739 * Changing the data of the delayed item is impossible. So
1740 * we needn't lock them. And we have held i_mutex of the
1741 * directory, nobody can delete any directory indexes now.
1742 */
1743 list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1744 list_del(&curr->readdir_list);
1745
1746 if (curr->key.offset < filp->f_pos) {
1747 if (atomic_dec_and_test(&curr->refs))
1748 kfree(curr);
1749 continue;
1750 }
1751
1752 filp->f_pos = curr->key.offset;
1753
1754 di = (struct btrfs_dir_item *)curr->data;
1755 name = (char *)(di + 1);
1756 name_len = le16_to_cpu(di->name_len);
1757
1758 d_type = btrfs_filetype_table[di->type];
1759 btrfs_disk_key_to_cpu(&location, &di->location);
1760
1761 over = filldir(dirent, name, name_len, curr->key.offset,
1762 location.objectid, d_type);
1763
1764 if (atomic_dec_and_test(&curr->refs))
1765 kfree(curr);
1766
1767 if (over)
1768 return 1;
1769 }
1770 return 0;
1771 }
1772
1773 BTRFS_SETGET_STACK_FUNCS(stack_inode_generation, struct btrfs_inode_item,
1774 generation, 64);
1775 BTRFS_SETGET_STACK_FUNCS(stack_inode_sequence, struct btrfs_inode_item,
1776 sequence, 64);
1777 BTRFS_SETGET_STACK_FUNCS(stack_inode_transid, struct btrfs_inode_item,
1778 transid, 64);
1779 BTRFS_SETGET_STACK_FUNCS(stack_inode_size, struct btrfs_inode_item, size, 64);
1780 BTRFS_SETGET_STACK_FUNCS(stack_inode_nbytes, struct btrfs_inode_item,
1781 nbytes, 64);
1782 BTRFS_SETGET_STACK_FUNCS(stack_inode_block_group, struct btrfs_inode_item,
1783 block_group, 64);
1784 BTRFS_SETGET_STACK_FUNCS(stack_inode_nlink, struct btrfs_inode_item, nlink, 32);
1785 BTRFS_SETGET_STACK_FUNCS(stack_inode_uid, struct btrfs_inode_item, uid, 32);
1786 BTRFS_SETGET_STACK_FUNCS(stack_inode_gid, struct btrfs_inode_item, gid, 32);
1787 BTRFS_SETGET_STACK_FUNCS(stack_inode_mode, struct btrfs_inode_item, mode, 32);
1788 BTRFS_SETGET_STACK_FUNCS(stack_inode_rdev, struct btrfs_inode_item, rdev, 64);
1789 BTRFS_SETGET_STACK_FUNCS(stack_inode_flags, struct btrfs_inode_item, flags, 64);
1790
1791 BTRFS_SETGET_STACK_FUNCS(stack_timespec_sec, struct btrfs_timespec, sec, 64);
1792 BTRFS_SETGET_STACK_FUNCS(stack_timespec_nsec, struct btrfs_timespec, nsec, 32);
1793
1794 static void fill_stack_inode_item(struct btrfs_trans_handle *trans,
1795 struct btrfs_inode_item *inode_item,
1796 struct inode *inode)
1797 {
1798 btrfs_set_stack_inode_uid(inode_item, i_uid_read(inode));
1799 btrfs_set_stack_inode_gid(inode_item, i_gid_read(inode));
1800 btrfs_set_stack_inode_size(inode_item, BTRFS_I(inode)->disk_i_size);
1801 btrfs_set_stack_inode_mode(inode_item, inode->i_mode);
1802 btrfs_set_stack_inode_nlink(inode_item, inode->i_nlink);
1803 btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(inode));
1804 btrfs_set_stack_inode_generation(inode_item,
1805 BTRFS_I(inode)->generation);
1806 btrfs_set_stack_inode_sequence(inode_item, inode->i_version);
1807 btrfs_set_stack_inode_transid(inode_item, trans->transid);
1808 btrfs_set_stack_inode_rdev(inode_item, inode->i_rdev);
1809 btrfs_set_stack_inode_flags(inode_item, BTRFS_I(inode)->flags);
1810 btrfs_set_stack_inode_block_group(inode_item, 0);
1811
1812 btrfs_set_stack_timespec_sec(btrfs_inode_atime(inode_item),
1813 inode->i_atime.tv_sec);
1814 btrfs_set_stack_timespec_nsec(btrfs_inode_atime(inode_item),
1815 inode->i_atime.tv_nsec);
1816
1817 btrfs_set_stack_timespec_sec(btrfs_inode_mtime(inode_item),
1818 inode->i_mtime.tv_sec);
1819 btrfs_set_stack_timespec_nsec(btrfs_inode_mtime(inode_item),
1820 inode->i_mtime.tv_nsec);
1821
1822 btrfs_set_stack_timespec_sec(btrfs_inode_ctime(inode_item),
1823 inode->i_ctime.tv_sec);
1824 btrfs_set_stack_timespec_nsec(btrfs_inode_ctime(inode_item),
1825 inode->i_ctime.tv_nsec);
1826 }
1827
1828 int btrfs_fill_inode(struct inode *inode, u32 *rdev)
1829 {
1830 struct btrfs_delayed_node *delayed_node;
1831 struct btrfs_inode_item *inode_item;
1832 struct btrfs_timespec *tspec;
1833
1834 delayed_node = btrfs_get_delayed_node(inode);
1835 if (!delayed_node)
1836 return -ENOENT;
1837
1838 mutex_lock(&delayed_node->mutex);
1839 if (!delayed_node->inode_dirty) {
1840 mutex_unlock(&delayed_node->mutex);
1841 btrfs_release_delayed_node(delayed_node);
1842 return -ENOENT;
1843 }
1844
1845 inode_item = &delayed_node->inode_item;
1846
1847 i_uid_write(inode, btrfs_stack_inode_uid(inode_item));
1848 i_gid_write(inode, btrfs_stack_inode_gid(inode_item));
1849 btrfs_i_size_write(inode, btrfs_stack_inode_size(inode_item));
1850 inode->i_mode = btrfs_stack_inode_mode(inode_item);
1851 set_nlink(inode, btrfs_stack_inode_nlink(inode_item));
1852 inode_set_bytes(inode, btrfs_stack_inode_nbytes(inode_item));
1853 BTRFS_I(inode)->generation = btrfs_stack_inode_generation(inode_item);
1854 inode->i_version = btrfs_stack_inode_sequence(inode_item);
1855 inode->i_rdev = 0;
1856 *rdev = btrfs_stack_inode_rdev(inode_item);
1857 BTRFS_I(inode)->flags = btrfs_stack_inode_flags(inode_item);
1858
1859 tspec = btrfs_inode_atime(inode_item);
1860 inode->i_atime.tv_sec = btrfs_stack_timespec_sec(tspec);
1861 inode->i_atime.tv_nsec = btrfs_stack_timespec_nsec(tspec);
1862
1863 tspec = btrfs_inode_mtime(inode_item);
1864 inode->i_mtime.tv_sec = btrfs_stack_timespec_sec(tspec);
1865 inode->i_mtime.tv_nsec = btrfs_stack_timespec_nsec(tspec);
1866
1867 tspec = btrfs_inode_ctime(inode_item);
1868 inode->i_ctime.tv_sec = btrfs_stack_timespec_sec(tspec);
1869 inode->i_ctime.tv_nsec = btrfs_stack_timespec_nsec(tspec);
1870
1871 inode->i_generation = BTRFS_I(inode)->generation;
1872 BTRFS_I(inode)->index_cnt = (u64)-1;
1873
1874 mutex_unlock(&delayed_node->mutex);
1875 btrfs_release_delayed_node(delayed_node);
1876 return 0;
1877 }
1878
1879 int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans,
1880 struct btrfs_root *root, struct inode *inode)
1881 {
1882 struct btrfs_delayed_node *delayed_node;
1883 int ret = 0;
1884
1885 delayed_node = btrfs_get_or_create_delayed_node(inode);
1886 if (IS_ERR(delayed_node))
1887 return PTR_ERR(delayed_node);
1888
1889 mutex_lock(&delayed_node->mutex);
1890 if (delayed_node->inode_dirty) {
1891 fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1892 goto release_node;
1893 }
1894
1895 ret = btrfs_delayed_inode_reserve_metadata(trans, root, inode,
1896 delayed_node);
1897 if (ret)
1898 goto release_node;
1899
1900 fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1901 delayed_node->inode_dirty = 1;
1902 delayed_node->count++;
1903 atomic_inc(&root->fs_info->delayed_root->items);
1904 release_node:
1905 mutex_unlock(&delayed_node->mutex);
1906 btrfs_release_delayed_node(delayed_node);
1907 return ret;
1908 }
1909
1910 static void __btrfs_kill_delayed_node(struct btrfs_delayed_node *delayed_node)
1911 {
1912 struct btrfs_root *root = delayed_node->root;
1913 struct btrfs_delayed_item *curr_item, *prev_item;
1914
1915 mutex_lock(&delayed_node->mutex);
1916 curr_item = __btrfs_first_delayed_insertion_item(delayed_node);
1917 while (curr_item) {
1918 btrfs_delayed_item_release_metadata(root, 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 curr_item = __btrfs_first_delayed_deletion_item(delayed_node);
1925 while (curr_item) {
1926 btrfs_delayed_item_release_metadata(root, curr_item);
1927 prev_item = curr_item;
1928 curr_item = __btrfs_next_delayed_item(prev_item);
1929 btrfs_release_delayed_item(prev_item);
1930 }
1931
1932 if (delayed_node->inode_dirty) {
1933 btrfs_delayed_inode_release_metadata(root, delayed_node);
1934 btrfs_release_delayed_inode(delayed_node);
1935 }
1936 mutex_unlock(&delayed_node->mutex);
1937 }
1938
1939 void btrfs_kill_delayed_inode_items(struct inode *inode)
1940 {
1941 struct btrfs_delayed_node *delayed_node;
1942
1943 delayed_node = btrfs_get_delayed_node(inode);
1944 if (!delayed_node)
1945 return;
1946
1947 __btrfs_kill_delayed_node(delayed_node);
1948 btrfs_release_delayed_node(delayed_node);
1949 }
1950
1951 void btrfs_kill_all_delayed_nodes(struct btrfs_root *root)
1952 {
1953 u64 inode_id = 0;
1954 struct btrfs_delayed_node *delayed_nodes[8];
1955 int i, n;
1956
1957 while (1) {
1958 spin_lock(&root->inode_lock);
1959 n = radix_tree_gang_lookup(&root->delayed_nodes_tree,
1960 (void **)delayed_nodes, inode_id,
1961 ARRAY_SIZE(delayed_nodes));
1962 if (!n) {
1963 spin_unlock(&root->inode_lock);
1964 break;
1965 }
1966
1967 inode_id = delayed_nodes[n - 1]->inode_id + 1;
1968
1969 for (i = 0; i < n; i++)
1970 atomic_inc(&delayed_nodes[i]->refs);
1971 spin_unlock(&root->inode_lock);
1972
1973 for (i = 0; i < n; i++) {
1974 __btrfs_kill_delayed_node(delayed_nodes[i]);
1975 btrfs_release_delayed_node(delayed_nodes[i]);
1976 }
1977 }
1978 }
1979
1980 void btrfs_destroy_delayed_inodes(struct btrfs_root *root)
1981 {
1982 struct btrfs_delayed_root *delayed_root;
1983 struct btrfs_delayed_node *curr_node, *prev_node;
1984
1985 delayed_root = btrfs_get_delayed_root(root);
1986
1987 curr_node = btrfs_first_delayed_node(delayed_root);
1988 while (curr_node) {
1989 __btrfs_kill_delayed_node(curr_node);
1990
1991 prev_node = curr_node;
1992 curr_node = btrfs_next_delayed_node(curr_node);
1993 btrfs_release_delayed_node(prev_node);
1994 }
1995 }
1996
Linux kernel - Deliverables
This page took 0.088794 seconds and 5 git commands to generate.