Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 2011 Fujitsu. All rights reserved.
4 * Written by Miao Xie <miaox@cn.fujitsu.com>
5 */
6
7#include <linux/slab.h>
8#include <linux/iversion.h>
9#include <linux/sched/mm.h>
10#include "misc.h"
11#include "delayed-inode.h"
12#include "disk-io.h"
13#include "transaction.h"
14#include "ctree.h"
15#include "qgroup.h"
16#include "locking.h"
17
18#define BTRFS_DELAYED_WRITEBACK 512
19#define BTRFS_DELAYED_BACKGROUND 128
20#define BTRFS_DELAYED_BATCH 16
21
22static struct kmem_cache *delayed_node_cache;
23
24int __init btrfs_delayed_inode_init(void)
25{
26 delayed_node_cache = kmem_cache_create("btrfs_delayed_node",
27 sizeof(struct btrfs_delayed_node),
28 0,
29 SLAB_MEM_SPREAD,
30 NULL);
31 if (!delayed_node_cache)
32 return -ENOMEM;
33 return 0;
34}
35
36void __cold btrfs_delayed_inode_exit(void)
37{
38 kmem_cache_destroy(delayed_node_cache);
39}
40
41static inline void btrfs_init_delayed_node(
42 struct btrfs_delayed_node *delayed_node,
43 struct btrfs_root *root, u64 inode_id)
44{
45 delayed_node->root = root;
46 delayed_node->inode_id = inode_id;
47 refcount_set(&delayed_node->refs, 0);
48 delayed_node->ins_root = RB_ROOT_CACHED;
49 delayed_node->del_root = RB_ROOT_CACHED;
50 mutex_init(&delayed_node->mutex);
51 INIT_LIST_HEAD(&delayed_node->n_list);
52 INIT_LIST_HEAD(&delayed_node->p_list);
53}
54
55static inline int btrfs_is_continuous_delayed_item(
56 struct btrfs_delayed_item *item1,
57 struct btrfs_delayed_item *item2)
58{
59 if (item1->key.type == BTRFS_DIR_INDEX_KEY &&
60 item1->key.objectid == item2->key.objectid &&
61 item1->key.type == item2->key.type &&
62 item1->key.offset + 1 == item2->key.offset)
63 return 1;
64 return 0;
65}
66
67static struct btrfs_delayed_node *btrfs_get_delayed_node(
68 struct btrfs_inode *btrfs_inode)
69{
70 struct btrfs_root *root = btrfs_inode->root;
71 u64 ino = btrfs_ino(btrfs_inode);
72 struct btrfs_delayed_node *node;
73
74 node = READ_ONCE(btrfs_inode->delayed_node);
75 if (node) {
76 refcount_inc(&node->refs);
77 return node;
78 }
79
80 spin_lock(&root->inode_lock);
81 node = radix_tree_lookup(&root->delayed_nodes_tree, ino);
82
83 if (node) {
84 if (btrfs_inode->delayed_node) {
85 refcount_inc(&node->refs); /* can be accessed */
86 BUG_ON(btrfs_inode->delayed_node != node);
87 spin_unlock(&root->inode_lock);
88 return node;
89 }
90
91 /*
92 * It's possible that we're racing into the middle of removing
93 * this node from the radix tree. In this case, the refcount
94 * was zero and it should never go back to one. Just return
95 * NULL like it was never in the radix at all; our release
96 * function is in the process of removing it.
97 *
98 * Some implementations of refcount_inc refuse to bump the
99 * refcount once it has hit zero. If we don't do this dance
100 * here, refcount_inc() may decide to just WARN_ONCE() instead
101 * of actually bumping the refcount.
102 *
103 * If this node is properly in the radix, we want to bump the
104 * refcount twice, once for the inode and once for this get
105 * operation.
106 */
107 if (refcount_inc_not_zero(&node->refs)) {
108 refcount_inc(&node->refs);
109 btrfs_inode->delayed_node = node;
110 } else {
111 node = NULL;
112 }
113
114 spin_unlock(&root->inode_lock);
115 return node;
116 }
117 spin_unlock(&root->inode_lock);
118
119 return NULL;
120}
121
122/* Will return either the node or PTR_ERR(-ENOMEM) */
123static struct btrfs_delayed_node *btrfs_get_or_create_delayed_node(
124 struct btrfs_inode *btrfs_inode)
125{
126 struct btrfs_delayed_node *node;
127 struct btrfs_root *root = btrfs_inode->root;
128 u64 ino = btrfs_ino(btrfs_inode);
129 int ret;
130
131again:
132 node = btrfs_get_delayed_node(btrfs_inode);
133 if (node)
134 return node;
135
136 node = kmem_cache_zalloc(delayed_node_cache, GFP_NOFS);
137 if (!node)
138 return ERR_PTR(-ENOMEM);
139 btrfs_init_delayed_node(node, root, ino);
140
141 /* cached in the btrfs inode and can be accessed */
142 refcount_set(&node->refs, 2);
143
144 ret = radix_tree_preload(GFP_NOFS);
145 if (ret) {
146 kmem_cache_free(delayed_node_cache, node);
147 return ERR_PTR(ret);
148 }
149
150 spin_lock(&root->inode_lock);
151 ret = radix_tree_insert(&root->delayed_nodes_tree, ino, node);
152 if (ret == -EEXIST) {
153 spin_unlock(&root->inode_lock);
154 kmem_cache_free(delayed_node_cache, node);
155 radix_tree_preload_end();
156 goto again;
157 }
158 btrfs_inode->delayed_node = node;
159 spin_unlock(&root->inode_lock);
160 radix_tree_preload_end();
161
162 return node;
163}
164
165/*
166 * Call it when holding delayed_node->mutex
167 *
168 * If mod = 1, add this node into the prepared list.
169 */
170static void btrfs_queue_delayed_node(struct btrfs_delayed_root *root,
171 struct btrfs_delayed_node *node,
172 int mod)
173{
174 spin_lock(&root->lock);
175 if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
176 if (!list_empty(&node->p_list))
177 list_move_tail(&node->p_list, &root->prepare_list);
178 else if (mod)
179 list_add_tail(&node->p_list, &root->prepare_list);
180 } else {
181 list_add_tail(&node->n_list, &root->node_list);
182 list_add_tail(&node->p_list, &root->prepare_list);
183 refcount_inc(&node->refs); /* inserted into list */
184 root->nodes++;
185 set_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
186 }
187 spin_unlock(&root->lock);
188}
189
190/* Call it when holding delayed_node->mutex */
191static void btrfs_dequeue_delayed_node(struct btrfs_delayed_root *root,
192 struct btrfs_delayed_node *node)
193{
194 spin_lock(&root->lock);
195 if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
196 root->nodes--;
197 refcount_dec(&node->refs); /* not in the list */
198 list_del_init(&node->n_list);
199 if (!list_empty(&node->p_list))
200 list_del_init(&node->p_list);
201 clear_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
202 }
203 spin_unlock(&root->lock);
204}
205
206static struct btrfs_delayed_node *btrfs_first_delayed_node(
207 struct btrfs_delayed_root *delayed_root)
208{
209 struct list_head *p;
210 struct btrfs_delayed_node *node = NULL;
211
212 spin_lock(&delayed_root->lock);
213 if (list_empty(&delayed_root->node_list))
214 goto out;
215
216 p = delayed_root->node_list.next;
217 node = list_entry(p, struct btrfs_delayed_node, n_list);
218 refcount_inc(&node->refs);
219out:
220 spin_unlock(&delayed_root->lock);
221
222 return node;
223}
224
225static struct btrfs_delayed_node *btrfs_next_delayed_node(
226 struct btrfs_delayed_node *node)
227{
228 struct btrfs_delayed_root *delayed_root;
229 struct list_head *p;
230 struct btrfs_delayed_node *next = NULL;
231
232 delayed_root = node->root->fs_info->delayed_root;
233 spin_lock(&delayed_root->lock);
234 if (!test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
235 /* not in the list */
236 if (list_empty(&delayed_root->node_list))
237 goto out;
238 p = delayed_root->node_list.next;
239 } else if (list_is_last(&node->n_list, &delayed_root->node_list))
240 goto out;
241 else
242 p = node->n_list.next;
243
244 next = list_entry(p, struct btrfs_delayed_node, n_list);
245 refcount_inc(&next->refs);
246out:
247 spin_unlock(&delayed_root->lock);
248
249 return next;
250}
251
252static void __btrfs_release_delayed_node(
253 struct btrfs_delayed_node *delayed_node,
254 int mod)
255{
256 struct btrfs_delayed_root *delayed_root;
257
258 if (!delayed_node)
259 return;
260
261 delayed_root = delayed_node->root->fs_info->delayed_root;
262
263 mutex_lock(&delayed_node->mutex);
264 if (delayed_node->count)
265 btrfs_queue_delayed_node(delayed_root, delayed_node, mod);
266 else
267 btrfs_dequeue_delayed_node(delayed_root, delayed_node);
268 mutex_unlock(&delayed_node->mutex);
269
270 if (refcount_dec_and_test(&delayed_node->refs)) {
271 struct btrfs_root *root = delayed_node->root;
272
273 spin_lock(&root->inode_lock);
274 /*
275 * Once our refcount goes to zero, nobody is allowed to bump it
276 * back up. We can delete it now.
277 */
278 ASSERT(refcount_read(&delayed_node->refs) == 0);
279 radix_tree_delete(&root->delayed_nodes_tree,
280 delayed_node->inode_id);
281 spin_unlock(&root->inode_lock);
282 kmem_cache_free(delayed_node_cache, delayed_node);
283 }
284}
285
286static inline void btrfs_release_delayed_node(struct btrfs_delayed_node *node)
287{
288 __btrfs_release_delayed_node(node, 0);
289}
290
291static struct btrfs_delayed_node *btrfs_first_prepared_delayed_node(
292 struct btrfs_delayed_root *delayed_root)
293{
294 struct list_head *p;
295 struct btrfs_delayed_node *node = NULL;
296
297 spin_lock(&delayed_root->lock);
298 if (list_empty(&delayed_root->prepare_list))
299 goto out;
300
301 p = delayed_root->prepare_list.next;
302 list_del_init(p);
303 node = list_entry(p, struct btrfs_delayed_node, p_list);
304 refcount_inc(&node->refs);
305out:
306 spin_unlock(&delayed_root->lock);
307
308 return node;
309}
310
311static inline void btrfs_release_prepared_delayed_node(
312 struct btrfs_delayed_node *node)
313{
314 __btrfs_release_delayed_node(node, 1);
315}
316
317static struct btrfs_delayed_item *btrfs_alloc_delayed_item(u32 data_len)
318{
319 struct btrfs_delayed_item *item;
320 item = kmalloc(sizeof(*item) + data_len, GFP_NOFS);
321 if (item) {
322 item->data_len = data_len;
323 item->ins_or_del = 0;
324 item->bytes_reserved = 0;
325 item->delayed_node = NULL;
326 refcount_set(&item->refs, 1);
327 }
328 return item;
329}
330
331/*
332 * __btrfs_lookup_delayed_item - look up the delayed item by key
333 * @delayed_node: pointer to the delayed node
334 * @key: the key to look up
335 * @prev: used to store the prev item if the right item isn't found
336 * @next: used to store the next item if the right item isn't found
337 *
338 * Note: if we don't find the right item, we will return the prev item and
339 * the next item.
340 */
341static struct btrfs_delayed_item *__btrfs_lookup_delayed_item(
342 struct rb_root *root,
343 struct btrfs_key *key,
344 struct btrfs_delayed_item **prev,
345 struct btrfs_delayed_item **next)
346{
347 struct rb_node *node, *prev_node = NULL;
348 struct btrfs_delayed_item *delayed_item = NULL;
349 int ret = 0;
350
351 node = root->rb_node;
352
353 while (node) {
354 delayed_item = rb_entry(node, struct btrfs_delayed_item,
355 rb_node);
356 prev_node = node;
357 ret = btrfs_comp_cpu_keys(&delayed_item->key, key);
358 if (ret < 0)
359 node = node->rb_right;
360 else if (ret > 0)
361 node = node->rb_left;
362 else
363 return delayed_item;
364 }
365
366 if (prev) {
367 if (!prev_node)
368 *prev = NULL;
369 else if (ret < 0)
370 *prev = delayed_item;
371 else if ((node = rb_prev(prev_node)) != NULL) {
372 *prev = rb_entry(node, struct btrfs_delayed_item,
373 rb_node);
374 } else
375 *prev = NULL;
376 }
377
378 if (next) {
379 if (!prev_node)
380 *next = NULL;
381 else if (ret > 0)
382 *next = delayed_item;
383 else if ((node = rb_next(prev_node)) != NULL) {
384 *next = rb_entry(node, struct btrfs_delayed_item,
385 rb_node);
386 } else
387 *next = NULL;
388 }
389 return NULL;
390}
391
392static struct btrfs_delayed_item *__btrfs_lookup_delayed_insertion_item(
393 struct btrfs_delayed_node *delayed_node,
394 struct btrfs_key *key)
395{
396 return __btrfs_lookup_delayed_item(&delayed_node->ins_root.rb_root, key,
397 NULL, NULL);
398}
399
400static int __btrfs_add_delayed_item(struct btrfs_delayed_node *delayed_node,
401 struct btrfs_delayed_item *ins,
402 int action)
403{
404 struct rb_node **p, *node;
405 struct rb_node *parent_node = NULL;
406 struct rb_root_cached *root;
407 struct btrfs_delayed_item *item;
408 int cmp;
409 bool leftmost = true;
410
411 if (action == BTRFS_DELAYED_INSERTION_ITEM)
412 root = &delayed_node->ins_root;
413 else if (action == BTRFS_DELAYED_DELETION_ITEM)
414 root = &delayed_node->del_root;
415 else
416 BUG();
417 p = &root->rb_root.rb_node;
418 node = &ins->rb_node;
419
420 while (*p) {
421 parent_node = *p;
422 item = rb_entry(parent_node, struct btrfs_delayed_item,
423 rb_node);
424
425 cmp = btrfs_comp_cpu_keys(&item->key, &ins->key);
426 if (cmp < 0) {
427 p = &(*p)->rb_right;
428 leftmost = false;
429 } else if (cmp > 0) {
430 p = &(*p)->rb_left;
431 } else {
432 return -EEXIST;
433 }
434 }
435
436 rb_link_node(node, parent_node, p);
437 rb_insert_color_cached(node, root, leftmost);
438 ins->delayed_node = delayed_node;
439 ins->ins_or_del = action;
440
441 if (ins->key.type == BTRFS_DIR_INDEX_KEY &&
442 action == BTRFS_DELAYED_INSERTION_ITEM &&
443 ins->key.offset >= delayed_node->index_cnt)
444 delayed_node->index_cnt = ins->key.offset + 1;
445
446 delayed_node->count++;
447 atomic_inc(&delayed_node->root->fs_info->delayed_root->items);
448 return 0;
449}
450
451static int __btrfs_add_delayed_insertion_item(struct btrfs_delayed_node *node,
452 struct btrfs_delayed_item *item)
453{
454 return __btrfs_add_delayed_item(node, item,
455 BTRFS_DELAYED_INSERTION_ITEM);
456}
457
458static int __btrfs_add_delayed_deletion_item(struct btrfs_delayed_node *node,
459 struct btrfs_delayed_item *item)
460{
461 return __btrfs_add_delayed_item(node, item,
462 BTRFS_DELAYED_DELETION_ITEM);
463}
464
465static void finish_one_item(struct btrfs_delayed_root *delayed_root)
466{
467 int seq = atomic_inc_return(&delayed_root->items_seq);
468
469 /* atomic_dec_return implies a barrier */
470 if ((atomic_dec_return(&delayed_root->items) <
471 BTRFS_DELAYED_BACKGROUND || seq % BTRFS_DELAYED_BATCH == 0))
472 cond_wake_up_nomb(&delayed_root->wait);
473}
474
475static void __btrfs_remove_delayed_item(struct btrfs_delayed_item *delayed_item)
476{
477 struct rb_root_cached *root;
478 struct btrfs_delayed_root *delayed_root;
479
480 /* Not associated with any delayed_node */
481 if (!delayed_item->delayed_node)
482 return;
483 delayed_root = delayed_item->delayed_node->root->fs_info->delayed_root;
484
485 BUG_ON(!delayed_root);
486 BUG_ON(delayed_item->ins_or_del != BTRFS_DELAYED_DELETION_ITEM &&
487 delayed_item->ins_or_del != BTRFS_DELAYED_INSERTION_ITEM);
488
489 if (delayed_item->ins_or_del == BTRFS_DELAYED_INSERTION_ITEM)
490 root = &delayed_item->delayed_node->ins_root;
491 else
492 root = &delayed_item->delayed_node->del_root;
493
494 rb_erase_cached(&delayed_item->rb_node, root);
495 delayed_item->delayed_node->count--;
496
497 finish_one_item(delayed_root);
498}
499
500static void btrfs_release_delayed_item(struct btrfs_delayed_item *item)
501{
502 if (item) {
503 __btrfs_remove_delayed_item(item);
504 if (refcount_dec_and_test(&item->refs))
505 kfree(item);
506 }
507}
508
509static struct btrfs_delayed_item *__btrfs_first_delayed_insertion_item(
510 struct btrfs_delayed_node *delayed_node)
511{
512 struct rb_node *p;
513 struct btrfs_delayed_item *item = NULL;
514
515 p = rb_first_cached(&delayed_node->ins_root);
516 if (p)
517 item = rb_entry(p, struct btrfs_delayed_item, rb_node);
518
519 return item;
520}
521
522static struct btrfs_delayed_item *__btrfs_first_delayed_deletion_item(
523 struct btrfs_delayed_node *delayed_node)
524{
525 struct rb_node *p;
526 struct btrfs_delayed_item *item = NULL;
527
528 p = rb_first_cached(&delayed_node->del_root);
529 if (p)
530 item = rb_entry(p, struct btrfs_delayed_item, rb_node);
531
532 return item;
533}
534
535static struct btrfs_delayed_item *__btrfs_next_delayed_item(
536 struct btrfs_delayed_item *item)
537{
538 struct rb_node *p;
539 struct btrfs_delayed_item *next = NULL;
540
541 p = rb_next(&item->rb_node);
542 if (p)
543 next = rb_entry(p, struct btrfs_delayed_item, rb_node);
544
545 return next;
546}
547
548static int btrfs_delayed_item_reserve_metadata(struct btrfs_trans_handle *trans,
549 struct btrfs_root *root,
550 struct btrfs_delayed_item *item)
551{
552 struct btrfs_block_rsv *src_rsv;
553 struct btrfs_block_rsv *dst_rsv;
554 struct btrfs_fs_info *fs_info = root->fs_info;
555 u64 num_bytes;
556 int ret;
557
558 if (!trans->bytes_reserved)
559 return 0;
560
561 src_rsv = trans->block_rsv;
562 dst_rsv = &fs_info->delayed_block_rsv;
563
564 num_bytes = btrfs_calc_insert_metadata_size(fs_info, 1);
565
566 /*
567 * Here we migrate space rsv from transaction rsv, since have already
568 * reserved space when starting a transaction. So no need to reserve
569 * qgroup space here.
570 */
571 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, true);
572 if (!ret) {
573 trace_btrfs_space_reservation(fs_info, "delayed_item",
574 item->key.objectid,
575 num_bytes, 1);
576 item->bytes_reserved = num_bytes;
577 }
578
579 return ret;
580}
581
582static void btrfs_delayed_item_release_metadata(struct btrfs_root *root,
583 struct btrfs_delayed_item *item)
584{
585 struct btrfs_block_rsv *rsv;
586 struct btrfs_fs_info *fs_info = root->fs_info;
587
588 if (!item->bytes_reserved)
589 return;
590
591 rsv = &fs_info->delayed_block_rsv;
592 /*
593 * Check btrfs_delayed_item_reserve_metadata() to see why we don't need
594 * to release/reserve qgroup space.
595 */
596 trace_btrfs_space_reservation(fs_info, "delayed_item",
597 item->key.objectid, item->bytes_reserved,
598 0);
599 btrfs_block_rsv_release(fs_info, rsv, item->bytes_reserved, NULL);
600}
601
602static int btrfs_delayed_inode_reserve_metadata(
603 struct btrfs_trans_handle *trans,
604 struct btrfs_root *root,
605 struct btrfs_delayed_node *node)
606{
607 struct btrfs_fs_info *fs_info = root->fs_info;
608 struct btrfs_block_rsv *src_rsv;
609 struct btrfs_block_rsv *dst_rsv;
610 u64 num_bytes;
611 int ret;
612
613 src_rsv = trans->block_rsv;
614 dst_rsv = &fs_info->delayed_block_rsv;
615
616 num_bytes = btrfs_calc_metadata_size(fs_info, 1);
617
618 /*
619 * btrfs_dirty_inode will update the inode under btrfs_join_transaction
620 * which doesn't reserve space for speed. This is a problem since we
621 * still need to reserve space for this update, so try to reserve the
622 * space.
623 *
624 * Now if src_rsv == delalloc_block_rsv we'll let it just steal since
625 * we always reserve enough to update the inode item.
626 */
627 if (!src_rsv || (!trans->bytes_reserved &&
628 src_rsv->type != BTRFS_BLOCK_RSV_DELALLOC)) {
629 ret = btrfs_qgroup_reserve_meta(root, num_bytes,
630 BTRFS_QGROUP_RSV_META_PREALLOC, true);
631 if (ret < 0)
632 return ret;
633 ret = btrfs_block_rsv_add(root, dst_rsv, num_bytes,
634 BTRFS_RESERVE_NO_FLUSH);
635 /* NO_FLUSH could only fail with -ENOSPC */
636 ASSERT(ret == 0 || ret == -ENOSPC);
637 if (ret)
638 btrfs_qgroup_free_meta_prealloc(root, num_bytes);
639 } else {
640 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, true);
641 }
642
643 if (!ret) {
644 trace_btrfs_space_reservation(fs_info, "delayed_inode",
645 node->inode_id, num_bytes, 1);
646 node->bytes_reserved = num_bytes;
647 }
648
649 return ret;
650}
651
652static void btrfs_delayed_inode_release_metadata(struct btrfs_fs_info *fs_info,
653 struct btrfs_delayed_node *node,
654 bool qgroup_free)
655{
656 struct btrfs_block_rsv *rsv;
657
658 if (!node->bytes_reserved)
659 return;
660
661 rsv = &fs_info->delayed_block_rsv;
662 trace_btrfs_space_reservation(fs_info, "delayed_inode",
663 node->inode_id, node->bytes_reserved, 0);
664 btrfs_block_rsv_release(fs_info, rsv, node->bytes_reserved, NULL);
665 if (qgroup_free)
666 btrfs_qgroup_free_meta_prealloc(node->root,
667 node->bytes_reserved);
668 else
669 btrfs_qgroup_convert_reserved_meta(node->root,
670 node->bytes_reserved);
671 node->bytes_reserved = 0;
672}
673
674/*
675 * This helper will insert some continuous items into the same leaf according
676 * to the free space of the leaf.
677 */
678static int btrfs_batch_insert_items(struct btrfs_root *root,
679 struct btrfs_path *path,
680 struct btrfs_delayed_item *item)
681{
682 struct btrfs_delayed_item *curr, *next;
683 int free_space;
684 int total_size = 0;
685 struct extent_buffer *leaf;
686 char *data_ptr;
687 struct btrfs_key *keys;
688 u32 *data_size;
689 struct list_head head;
690 int slot;
691 int nitems;
692 int i;
693 int ret = 0;
694
695 BUG_ON(!path->nodes[0]);
696
697 leaf = path->nodes[0];
698 free_space = btrfs_leaf_free_space(leaf);
699 INIT_LIST_HEAD(&head);
700
701 next = item;
702 nitems = 0;
703
704 /*
705 * count the number of the continuous items that we can insert in batch
706 */
707 while (total_size + next->data_len + sizeof(struct btrfs_item) <=
708 free_space) {
709 total_size += next->data_len + sizeof(struct btrfs_item);
710 list_add_tail(&next->tree_list, &head);
711 nitems++;
712
713 curr = next;
714 next = __btrfs_next_delayed_item(curr);
715 if (!next)
716 break;
717
718 if (!btrfs_is_continuous_delayed_item(curr, next))
719 break;
720 }
721
722 if (!nitems) {
723 ret = 0;
724 goto out;
725 }
726
727 keys = kmalloc_array(nitems, sizeof(struct btrfs_key), GFP_NOFS);
728 if (!keys) {
729 ret = -ENOMEM;
730 goto out;
731 }
732
733 data_size = kmalloc_array(nitems, sizeof(u32), GFP_NOFS);
734 if (!data_size) {
735 ret = -ENOMEM;
736 goto error;
737 }
738
739 /* get keys of all the delayed items */
740 i = 0;
741 list_for_each_entry(next, &head, tree_list) {
742 keys[i] = next->key;
743 data_size[i] = next->data_len;
744 i++;
745 }
746
747 /* insert the keys of the items */
748 setup_items_for_insert(root, path, keys, data_size, nitems);
749
750 /* insert the dir index items */
751 slot = path->slots[0];
752 list_for_each_entry_safe(curr, next, &head, tree_list) {
753 data_ptr = btrfs_item_ptr(leaf, slot, char);
754 write_extent_buffer(leaf, &curr->data,
755 (unsigned long)data_ptr,
756 curr->data_len);
757 slot++;
758
759 btrfs_delayed_item_release_metadata(root, curr);
760
761 list_del(&curr->tree_list);
762 btrfs_release_delayed_item(curr);
763 }
764
765error:
766 kfree(data_size);
767 kfree(keys);
768out:
769 return ret;
770}
771
772/*
773 * This helper can just do simple insertion that needn't extend item for new
774 * data, such as directory name index insertion, inode insertion.
775 */
776static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans,
777 struct btrfs_root *root,
778 struct btrfs_path *path,
779 struct btrfs_delayed_item *delayed_item)
780{
781 struct extent_buffer *leaf;
782 unsigned int nofs_flag;
783 char *ptr;
784 int ret;
785
786 nofs_flag = memalloc_nofs_save();
787 ret = btrfs_insert_empty_item(trans, root, path, &delayed_item->key,
788 delayed_item->data_len);
789 memalloc_nofs_restore(nofs_flag);
790 if (ret < 0 && ret != -EEXIST)
791 return ret;
792
793 leaf = path->nodes[0];
794
795 ptr = btrfs_item_ptr(leaf, path->slots[0], char);
796
797 write_extent_buffer(leaf, delayed_item->data, (unsigned long)ptr,
798 delayed_item->data_len);
799 btrfs_mark_buffer_dirty(leaf);
800
801 btrfs_delayed_item_release_metadata(root, delayed_item);
802 return 0;
803}
804
805/*
806 * we insert an item first, then if there are some continuous items, we try
807 * to insert those items into the same leaf.
808 */
809static int btrfs_insert_delayed_items(struct btrfs_trans_handle *trans,
810 struct btrfs_path *path,
811 struct btrfs_root *root,
812 struct btrfs_delayed_node *node)
813{
814 struct btrfs_delayed_item *curr, *prev;
815 int ret = 0;
816
817do_again:
818 mutex_lock(&node->mutex);
819 curr = __btrfs_first_delayed_insertion_item(node);
820 if (!curr)
821 goto insert_end;
822
823 ret = btrfs_insert_delayed_item(trans, root, path, curr);
824 if (ret < 0) {
825 btrfs_release_path(path);
826 goto insert_end;
827 }
828
829 prev = curr;
830 curr = __btrfs_next_delayed_item(prev);
831 if (curr && btrfs_is_continuous_delayed_item(prev, curr)) {
832 /* insert the continuous items into the same leaf */
833 path->slots[0]++;
834 btrfs_batch_insert_items(root, path, curr);
835 }
836 btrfs_release_delayed_item(prev);
837 btrfs_mark_buffer_dirty(path->nodes[0]);
838
839 btrfs_release_path(path);
840 mutex_unlock(&node->mutex);
841 goto do_again;
842
843insert_end:
844 mutex_unlock(&node->mutex);
845 return ret;
846}
847
848static int btrfs_batch_delete_items(struct btrfs_trans_handle *trans,
849 struct btrfs_root *root,
850 struct btrfs_path *path,
851 struct btrfs_delayed_item *item)
852{
853 struct btrfs_delayed_item *curr, *next;
854 struct extent_buffer *leaf;
855 struct btrfs_key key;
856 struct list_head head;
857 int nitems, i, last_item;
858 int ret = 0;
859
860 BUG_ON(!path->nodes[0]);
861
862 leaf = path->nodes[0];
863
864 i = path->slots[0];
865 last_item = btrfs_header_nritems(leaf) - 1;
866 if (i > last_item)
867 return -ENOENT; /* FIXME: Is errno suitable? */
868
869 next = item;
870 INIT_LIST_HEAD(&head);
871 btrfs_item_key_to_cpu(leaf, &key, i);
872 nitems = 0;
873 /*
874 * count the number of the dir index items that we can delete in batch
875 */
876 while (btrfs_comp_cpu_keys(&next->key, &key) == 0) {
877 list_add_tail(&next->tree_list, &head);
878 nitems++;
879
880 curr = next;
881 next = __btrfs_next_delayed_item(curr);
882 if (!next)
883 break;
884
885 if (!btrfs_is_continuous_delayed_item(curr, next))
886 break;
887
888 i++;
889 if (i > last_item)
890 break;
891 btrfs_item_key_to_cpu(leaf, &key, i);
892 }
893
894 if (!nitems)
895 return 0;
896
897 ret = btrfs_del_items(trans, root, path, path->slots[0], nitems);
898 if (ret)
899 goto out;
900
901 list_for_each_entry_safe(curr, next, &head, tree_list) {
902 btrfs_delayed_item_release_metadata(root, curr);
903 list_del(&curr->tree_list);
904 btrfs_release_delayed_item(curr);
905 }
906
907out:
908 return ret;
909}
910
911static int btrfs_delete_delayed_items(struct btrfs_trans_handle *trans,
912 struct btrfs_path *path,
913 struct btrfs_root *root,
914 struct btrfs_delayed_node *node)
915{
916 struct btrfs_delayed_item *curr, *prev;
917 unsigned int nofs_flag;
918 int ret = 0;
919
920do_again:
921 mutex_lock(&node->mutex);
922 curr = __btrfs_first_delayed_deletion_item(node);
923 if (!curr)
924 goto delete_fail;
925
926 nofs_flag = memalloc_nofs_save();
927 ret = btrfs_search_slot(trans, root, &curr->key, path, -1, 1);
928 memalloc_nofs_restore(nofs_flag);
929 if (ret < 0)
930 goto delete_fail;
931 else if (ret > 0) {
932 /*
933 * can't find the item which the node points to, so this node
934 * is invalid, just drop it.
935 */
936 prev = curr;
937 curr = __btrfs_next_delayed_item(prev);
938 btrfs_release_delayed_item(prev);
939 ret = 0;
940 btrfs_release_path(path);
941 if (curr) {
942 mutex_unlock(&node->mutex);
943 goto do_again;
944 } else
945 goto delete_fail;
946 }
947
948 btrfs_batch_delete_items(trans, root, path, curr);
949 btrfs_release_path(path);
950 mutex_unlock(&node->mutex);
951 goto do_again;
952
953delete_fail:
954 btrfs_release_path(path);
955 mutex_unlock(&node->mutex);
956 return ret;
957}
958
959static void btrfs_release_delayed_inode(struct btrfs_delayed_node *delayed_node)
960{
961 struct btrfs_delayed_root *delayed_root;
962
963 if (delayed_node &&
964 test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
965 BUG_ON(!delayed_node->root);
966 clear_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
967 delayed_node->count--;
968
969 delayed_root = delayed_node->root->fs_info->delayed_root;
970 finish_one_item(delayed_root);
971 }
972}
973
974static void btrfs_release_delayed_iref(struct btrfs_delayed_node *delayed_node)
975{
976
977 if (test_and_clear_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags)) {
978 struct btrfs_delayed_root *delayed_root;
979
980 ASSERT(delayed_node->root);
981 delayed_node->count--;
982
983 delayed_root = delayed_node->root->fs_info->delayed_root;
984 finish_one_item(delayed_root);
985 }
986}
987
988static int __btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
989 struct btrfs_root *root,
990 struct btrfs_path *path,
991 struct btrfs_delayed_node *node)
992{
993 struct btrfs_fs_info *fs_info = root->fs_info;
994 struct btrfs_key key;
995 struct btrfs_inode_item *inode_item;
996 struct extent_buffer *leaf;
997 unsigned int nofs_flag;
998 int mod;
999 int ret;
1000
1001 key.objectid = node->inode_id;
1002 key.type = BTRFS_INODE_ITEM_KEY;
1003 key.offset = 0;
1004
1005 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1006 mod = -1;
1007 else
1008 mod = 1;
1009
1010 nofs_flag = memalloc_nofs_save();
1011 ret = btrfs_lookup_inode(trans, root, path, &key, mod);
1012 memalloc_nofs_restore(nofs_flag);
1013 if (ret > 0)
1014 ret = -ENOENT;
1015 if (ret < 0)
1016 goto out;
1017
1018 leaf = path->nodes[0];
1019 inode_item = btrfs_item_ptr(leaf, path->slots[0],
1020 struct btrfs_inode_item);
1021 write_extent_buffer(leaf, &node->inode_item, (unsigned long)inode_item,
1022 sizeof(struct btrfs_inode_item));
1023 btrfs_mark_buffer_dirty(leaf);
1024
1025 if (!test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1026 goto out;
1027
1028 path->slots[0]++;
1029 if (path->slots[0] >= btrfs_header_nritems(leaf))
1030 goto search;
1031again:
1032 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1033 if (key.objectid != node->inode_id)
1034 goto out;
1035
1036 if (key.type != BTRFS_INODE_REF_KEY &&
1037 key.type != BTRFS_INODE_EXTREF_KEY)
1038 goto out;
1039
1040 /*
1041 * Delayed iref deletion is for the inode who has only one link,
1042 * so there is only one iref. The case that several irefs are
1043 * in the same item doesn't exist.
1044 */
1045 btrfs_del_item(trans, root, path);
1046out:
1047 btrfs_release_delayed_iref(node);
1048 btrfs_release_path(path);
1049err_out:
1050 btrfs_delayed_inode_release_metadata(fs_info, node, (ret < 0));
1051 btrfs_release_delayed_inode(node);
1052
1053 /*
1054 * If we fail to update the delayed inode we need to abort the
1055 * transaction, because we could leave the inode with the improper
1056 * counts behind.
1057 */
1058 if (ret && ret != -ENOENT)
1059 btrfs_abort_transaction(trans, ret);
1060
1061 return ret;
1062
1063search:
1064 btrfs_release_path(path);
1065
1066 key.type = BTRFS_INODE_EXTREF_KEY;
1067 key.offset = -1;
1068
1069 nofs_flag = memalloc_nofs_save();
1070 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1071 memalloc_nofs_restore(nofs_flag);
1072 if (ret < 0)
1073 goto err_out;
1074 ASSERT(ret);
1075
1076 ret = 0;
1077 leaf = path->nodes[0];
1078 path->slots[0]--;
1079 goto again;
1080}
1081
1082static inline int btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1083 struct btrfs_root *root,
1084 struct btrfs_path *path,
1085 struct btrfs_delayed_node *node)
1086{
1087 int ret;
1088
1089 mutex_lock(&node->mutex);
1090 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &node->flags)) {
1091 mutex_unlock(&node->mutex);
1092 return 0;
1093 }
1094
1095 ret = __btrfs_update_delayed_inode(trans, root, path, node);
1096 mutex_unlock(&node->mutex);
1097 return ret;
1098}
1099
1100static inline int
1101__btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1102 struct btrfs_path *path,
1103 struct btrfs_delayed_node *node)
1104{
1105 int ret;
1106
1107 ret = btrfs_insert_delayed_items(trans, path, node->root, node);
1108 if (ret)
1109 return ret;
1110
1111 ret = btrfs_delete_delayed_items(trans, path, node->root, node);
1112 if (ret)
1113 return ret;
1114
1115 ret = btrfs_update_delayed_inode(trans, node->root, path, node);
1116 return ret;
1117}
1118
1119/*
1120 * Called when committing the transaction.
1121 * Returns 0 on success.
1122 * Returns < 0 on error and returns with an aborted transaction with any
1123 * outstanding delayed items cleaned up.
1124 */
1125static int __btrfs_run_delayed_items(struct btrfs_trans_handle *trans, int nr)
1126{
1127 struct btrfs_fs_info *fs_info = trans->fs_info;
1128 struct btrfs_delayed_root *delayed_root;
1129 struct btrfs_delayed_node *curr_node, *prev_node;
1130 struct btrfs_path *path;
1131 struct btrfs_block_rsv *block_rsv;
1132 int ret = 0;
1133 bool count = (nr > 0);
1134
1135 if (TRANS_ABORTED(trans))
1136 return -EIO;
1137
1138 path = btrfs_alloc_path();
1139 if (!path)
1140 return -ENOMEM;
1141
1142 block_rsv = trans->block_rsv;
1143 trans->block_rsv = &fs_info->delayed_block_rsv;
1144
1145 delayed_root = fs_info->delayed_root;
1146
1147 curr_node = btrfs_first_delayed_node(delayed_root);
1148 while (curr_node && (!count || nr--)) {
1149 ret = __btrfs_commit_inode_delayed_items(trans, path,
1150 curr_node);
1151 if (ret) {
1152 btrfs_release_delayed_node(curr_node);
1153 curr_node = NULL;
1154 btrfs_abort_transaction(trans, ret);
1155 break;
1156 }
1157
1158 prev_node = curr_node;
1159 curr_node = btrfs_next_delayed_node(curr_node);
1160 btrfs_release_delayed_node(prev_node);
1161 }
1162
1163 if (curr_node)
1164 btrfs_release_delayed_node(curr_node);
1165 btrfs_free_path(path);
1166 trans->block_rsv = block_rsv;
1167
1168 return ret;
1169}
1170
1171int btrfs_run_delayed_items(struct btrfs_trans_handle *trans)
1172{
1173 return __btrfs_run_delayed_items(trans, -1);
1174}
1175
1176int btrfs_run_delayed_items_nr(struct btrfs_trans_handle *trans, int nr)
1177{
1178 return __btrfs_run_delayed_items(trans, nr);
1179}
1180
1181int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1182 struct btrfs_inode *inode)
1183{
1184 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1185 struct btrfs_path *path;
1186 struct btrfs_block_rsv *block_rsv;
1187 int ret;
1188
1189 if (!delayed_node)
1190 return 0;
1191
1192 mutex_lock(&delayed_node->mutex);
1193 if (!delayed_node->count) {
1194 mutex_unlock(&delayed_node->mutex);
1195 btrfs_release_delayed_node(delayed_node);
1196 return 0;
1197 }
1198 mutex_unlock(&delayed_node->mutex);
1199
1200 path = btrfs_alloc_path();
1201 if (!path) {
1202 btrfs_release_delayed_node(delayed_node);
1203 return -ENOMEM;
1204 }
1205
1206 block_rsv = trans->block_rsv;
1207 trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv;
1208
1209 ret = __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1210
1211 btrfs_release_delayed_node(delayed_node);
1212 btrfs_free_path(path);
1213 trans->block_rsv = block_rsv;
1214
1215 return ret;
1216}
1217
1218int btrfs_commit_inode_delayed_inode(struct btrfs_inode *inode)
1219{
1220 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1221 struct btrfs_trans_handle *trans;
1222 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1223 struct btrfs_path *path;
1224 struct btrfs_block_rsv *block_rsv;
1225 int ret;
1226
1227 if (!delayed_node)
1228 return 0;
1229
1230 mutex_lock(&delayed_node->mutex);
1231 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1232 mutex_unlock(&delayed_node->mutex);
1233 btrfs_release_delayed_node(delayed_node);
1234 return 0;
1235 }
1236 mutex_unlock(&delayed_node->mutex);
1237
1238 trans = btrfs_join_transaction(delayed_node->root);
1239 if (IS_ERR(trans)) {
1240 ret = PTR_ERR(trans);
1241 goto out;
1242 }
1243
1244 path = btrfs_alloc_path();
1245 if (!path) {
1246 ret = -ENOMEM;
1247 goto trans_out;
1248 }
1249
1250 block_rsv = trans->block_rsv;
1251 trans->block_rsv = &fs_info->delayed_block_rsv;
1252
1253 mutex_lock(&delayed_node->mutex);
1254 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags))
1255 ret = __btrfs_update_delayed_inode(trans, delayed_node->root,
1256 path, delayed_node);
1257 else
1258 ret = 0;
1259 mutex_unlock(&delayed_node->mutex);
1260
1261 btrfs_free_path(path);
1262 trans->block_rsv = block_rsv;
1263trans_out:
1264 btrfs_end_transaction(trans);
1265 btrfs_btree_balance_dirty(fs_info);
1266out:
1267 btrfs_release_delayed_node(delayed_node);
1268
1269 return ret;
1270}
1271
1272void btrfs_remove_delayed_node(struct btrfs_inode *inode)
1273{
1274 struct btrfs_delayed_node *delayed_node;
1275
1276 delayed_node = READ_ONCE(inode->delayed_node);
1277 if (!delayed_node)
1278 return;
1279
1280 inode->delayed_node = NULL;
1281 btrfs_release_delayed_node(delayed_node);
1282}
1283
1284struct btrfs_async_delayed_work {
1285 struct btrfs_delayed_root *delayed_root;
1286 int nr;
1287 struct btrfs_work work;
1288};
1289
1290static void btrfs_async_run_delayed_root(struct btrfs_work *work)
1291{
1292 struct btrfs_async_delayed_work *async_work;
1293 struct btrfs_delayed_root *delayed_root;
1294 struct btrfs_trans_handle *trans;
1295 struct btrfs_path *path;
1296 struct btrfs_delayed_node *delayed_node = NULL;
1297 struct btrfs_root *root;
1298 struct btrfs_block_rsv *block_rsv;
1299 int total_done = 0;
1300
1301 async_work = container_of(work, struct btrfs_async_delayed_work, work);
1302 delayed_root = async_work->delayed_root;
1303
1304 path = btrfs_alloc_path();
1305 if (!path)
1306 goto out;
1307
1308 do {
1309 if (atomic_read(&delayed_root->items) <
1310 BTRFS_DELAYED_BACKGROUND / 2)
1311 break;
1312
1313 delayed_node = btrfs_first_prepared_delayed_node(delayed_root);
1314 if (!delayed_node)
1315 break;
1316
1317 root = delayed_node->root;
1318
1319 trans = btrfs_join_transaction(root);
1320 if (IS_ERR(trans)) {
1321 btrfs_release_path(path);
1322 btrfs_release_prepared_delayed_node(delayed_node);
1323 total_done++;
1324 continue;
1325 }
1326
1327 block_rsv = trans->block_rsv;
1328 trans->block_rsv = &root->fs_info->delayed_block_rsv;
1329
1330 __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1331
1332 trans->block_rsv = block_rsv;
1333 btrfs_end_transaction(trans);
1334 btrfs_btree_balance_dirty_nodelay(root->fs_info);
1335
1336 btrfs_release_path(path);
1337 btrfs_release_prepared_delayed_node(delayed_node);
1338 total_done++;
1339
1340 } while ((async_work->nr == 0 && total_done < BTRFS_DELAYED_WRITEBACK)
1341 || total_done < async_work->nr);
1342
1343 btrfs_free_path(path);
1344out:
1345 wake_up(&delayed_root->wait);
1346 kfree(async_work);
1347}
1348
1349
1350static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root,
1351 struct btrfs_fs_info *fs_info, int nr)
1352{
1353 struct btrfs_async_delayed_work *async_work;
1354
1355 async_work = kmalloc(sizeof(*async_work), GFP_NOFS);
1356 if (!async_work)
1357 return -ENOMEM;
1358
1359 async_work->delayed_root = delayed_root;
1360 btrfs_init_work(&async_work->work, btrfs_async_run_delayed_root, NULL,
1361 NULL);
1362 async_work->nr = nr;
1363
1364 btrfs_queue_work(fs_info->delayed_workers, &async_work->work);
1365 return 0;
1366}
1367
1368void btrfs_assert_delayed_root_empty(struct btrfs_fs_info *fs_info)
1369{
1370 WARN_ON(btrfs_first_delayed_node(fs_info->delayed_root));
1371}
1372
1373static int could_end_wait(struct btrfs_delayed_root *delayed_root, int seq)
1374{
1375 int val = atomic_read(&delayed_root->items_seq);
1376
1377 if (val < seq || val >= seq + BTRFS_DELAYED_BATCH)
1378 return 1;
1379
1380 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1381 return 1;
1382
1383 return 0;
1384}
1385
1386void btrfs_balance_delayed_items(struct btrfs_fs_info *fs_info)
1387{
1388 struct btrfs_delayed_root *delayed_root = fs_info->delayed_root;
1389
1390 if ((atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND) ||
1391 btrfs_workqueue_normal_congested(fs_info->delayed_workers))
1392 return;
1393
1394 if (atomic_read(&delayed_root->items) >= BTRFS_DELAYED_WRITEBACK) {
1395 int seq;
1396 int ret;
1397
1398 seq = atomic_read(&delayed_root->items_seq);
1399
1400 ret = btrfs_wq_run_delayed_node(delayed_root, fs_info, 0);
1401 if (ret)
1402 return;
1403
1404 wait_event_interruptible(delayed_root->wait,
1405 could_end_wait(delayed_root, seq));
1406 return;
1407 }
1408
1409 btrfs_wq_run_delayed_node(delayed_root, fs_info, BTRFS_DELAYED_BATCH);
1410}
1411
1412/* Will return 0 or -ENOMEM */
1413int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans,
1414 const char *name, int name_len,
1415 struct btrfs_inode *dir,
1416 struct btrfs_disk_key *disk_key, u8 type,
1417 u64 index)
1418{
1419 struct btrfs_delayed_node *delayed_node;
1420 struct btrfs_delayed_item *delayed_item;
1421 struct btrfs_dir_item *dir_item;
1422 int ret;
1423
1424 delayed_node = btrfs_get_or_create_delayed_node(dir);
1425 if (IS_ERR(delayed_node))
1426 return PTR_ERR(delayed_node);
1427
1428 delayed_item = btrfs_alloc_delayed_item(sizeof(*dir_item) + name_len);
1429 if (!delayed_item) {
1430 ret = -ENOMEM;
1431 goto release_node;
1432 }
1433
1434 delayed_item->key.objectid = btrfs_ino(dir);
1435 delayed_item->key.type = BTRFS_DIR_INDEX_KEY;
1436 delayed_item->key.offset = index;
1437
1438 dir_item = (struct btrfs_dir_item *)delayed_item->data;
1439 dir_item->location = *disk_key;
1440 btrfs_set_stack_dir_transid(dir_item, trans->transid);
1441 btrfs_set_stack_dir_data_len(dir_item, 0);
1442 btrfs_set_stack_dir_name_len(dir_item, name_len);
1443 btrfs_set_stack_dir_type(dir_item, type);
1444 memcpy((char *)(dir_item + 1), name, name_len);
1445
1446 ret = btrfs_delayed_item_reserve_metadata(trans, dir->root, delayed_item);
1447 /*
1448 * we have reserved enough space when we start a new transaction,
1449 * so reserving metadata failure is impossible
1450 */
1451 BUG_ON(ret);
1452
1453 mutex_lock(&delayed_node->mutex);
1454 ret = __btrfs_add_delayed_insertion_item(delayed_node, delayed_item);
1455 if (unlikely(ret)) {
1456 btrfs_err(trans->fs_info,
1457 "err add delayed dir index item(name: %.*s) into the insertion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
1458 name_len, name, delayed_node->root->root_key.objectid,
1459 delayed_node->inode_id, ret);
1460 BUG();
1461 }
1462 mutex_unlock(&delayed_node->mutex);
1463
1464release_node:
1465 btrfs_release_delayed_node(delayed_node);
1466 return ret;
1467}
1468
1469static int btrfs_delete_delayed_insertion_item(struct btrfs_fs_info *fs_info,
1470 struct btrfs_delayed_node *node,
1471 struct btrfs_key *key)
1472{
1473 struct btrfs_delayed_item *item;
1474
1475 mutex_lock(&node->mutex);
1476 item = __btrfs_lookup_delayed_insertion_item(node, key);
1477 if (!item) {
1478 mutex_unlock(&node->mutex);
1479 return 1;
1480 }
1481
1482 btrfs_delayed_item_release_metadata(node->root, item);
1483 btrfs_release_delayed_item(item);
1484 mutex_unlock(&node->mutex);
1485 return 0;
1486}
1487
1488int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans,
1489 struct btrfs_inode *dir, u64 index)
1490{
1491 struct btrfs_delayed_node *node;
1492 struct btrfs_delayed_item *item;
1493 struct btrfs_key item_key;
1494 int ret;
1495
1496 node = btrfs_get_or_create_delayed_node(dir);
1497 if (IS_ERR(node))
1498 return PTR_ERR(node);
1499
1500 item_key.objectid = btrfs_ino(dir);
1501 item_key.type = BTRFS_DIR_INDEX_KEY;
1502 item_key.offset = index;
1503
1504 ret = btrfs_delete_delayed_insertion_item(trans->fs_info, node,
1505 &item_key);
1506 if (!ret)
1507 goto end;
1508
1509 item = btrfs_alloc_delayed_item(0);
1510 if (!item) {
1511 ret = -ENOMEM;
1512 goto end;
1513 }
1514
1515 item->key = item_key;
1516
1517 ret = btrfs_delayed_item_reserve_metadata(trans, dir->root, item);
1518 /*
1519 * we have reserved enough space when we start a new transaction,
1520 * so reserving metadata failure is impossible.
1521 */
1522 if (ret < 0) {
1523 btrfs_err(trans->fs_info,
1524"metadata reservation failed for delayed dir item deltiona, should have been reserved");
1525 btrfs_release_delayed_item(item);
1526 goto end;
1527 }
1528
1529 mutex_lock(&node->mutex);
1530 ret = __btrfs_add_delayed_deletion_item(node, item);
1531 if (unlikely(ret)) {
1532 btrfs_err(trans->fs_info,
1533 "err add delayed dir index item(index: %llu) into the deletion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
1534 index, node->root->root_key.objectid,
1535 node->inode_id, ret);
1536 btrfs_delayed_item_release_metadata(dir->root, item);
1537 btrfs_release_delayed_item(item);
1538 }
1539 mutex_unlock(&node->mutex);
1540end:
1541 btrfs_release_delayed_node(node);
1542 return ret;
1543}
1544
1545int btrfs_inode_delayed_dir_index_count(struct btrfs_inode *inode)
1546{
1547 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1548
1549 if (!delayed_node)
1550 return -ENOENT;
1551
1552 /*
1553 * Since we have held i_mutex of this directory, it is impossible that
1554 * a new directory index is added into the delayed node and index_cnt
1555 * is updated now. So we needn't lock the delayed node.
1556 */
1557 if (!delayed_node->index_cnt) {
1558 btrfs_release_delayed_node(delayed_node);
1559 return -EINVAL;
1560 }
1561
1562 inode->index_cnt = delayed_node->index_cnt;
1563 btrfs_release_delayed_node(delayed_node);
1564 return 0;
1565}
1566
1567bool btrfs_readdir_get_delayed_items(struct inode *inode,
1568 struct list_head *ins_list,
1569 struct list_head *del_list)
1570{
1571 struct btrfs_delayed_node *delayed_node;
1572 struct btrfs_delayed_item *item;
1573
1574 delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
1575 if (!delayed_node)
1576 return false;
1577
1578 /*
1579 * We can only do one readdir with delayed items at a time because of
1580 * item->readdir_list.
1581 */
1582 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
1583 btrfs_inode_lock(inode, 0);
1584
1585 mutex_lock(&delayed_node->mutex);
1586 item = __btrfs_first_delayed_insertion_item(delayed_node);
1587 while (item) {
1588 refcount_inc(&item->refs);
1589 list_add_tail(&item->readdir_list, ins_list);
1590 item = __btrfs_next_delayed_item(item);
1591 }
1592
1593 item = __btrfs_first_delayed_deletion_item(delayed_node);
1594 while (item) {
1595 refcount_inc(&item->refs);
1596 list_add_tail(&item->readdir_list, del_list);
1597 item = __btrfs_next_delayed_item(item);
1598 }
1599 mutex_unlock(&delayed_node->mutex);
1600 /*
1601 * This delayed node is still cached in the btrfs inode, so refs
1602 * must be > 1 now, and we needn't check it is going to be freed
1603 * or not.
1604 *
1605 * Besides that, this function is used to read dir, we do not
1606 * insert/delete delayed items in this period. So we also needn't
1607 * requeue or dequeue this delayed node.
1608 */
1609 refcount_dec(&delayed_node->refs);
1610
1611 return true;
1612}
1613
1614void btrfs_readdir_put_delayed_items(struct inode *inode,
1615 struct list_head *ins_list,
1616 struct list_head *del_list)
1617{
1618 struct btrfs_delayed_item *curr, *next;
1619
1620 list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1621 list_del(&curr->readdir_list);
1622 if (refcount_dec_and_test(&curr->refs))
1623 kfree(curr);
1624 }
1625
1626 list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1627 list_del(&curr->readdir_list);
1628 if (refcount_dec_and_test(&curr->refs))
1629 kfree(curr);
1630 }
1631
1632 /*
1633 * The VFS is going to do up_read(), so we need to downgrade back to a
1634 * read lock.
1635 */
1636 downgrade_write(&inode->i_rwsem);
1637}
1638
1639int btrfs_should_delete_dir_index(struct list_head *del_list,
1640 u64 index)
1641{
1642 struct btrfs_delayed_item *curr;
1643 int ret = 0;
1644
1645 list_for_each_entry(curr, del_list, readdir_list) {
1646 if (curr->key.offset > index)
1647 break;
1648 if (curr->key.offset == index) {
1649 ret = 1;
1650 break;
1651 }
1652 }
1653 return ret;
1654}
1655
1656/*
1657 * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree
1658 *
1659 */
1660int btrfs_readdir_delayed_dir_index(struct dir_context *ctx,
1661 struct list_head *ins_list)
1662{
1663 struct btrfs_dir_item *di;
1664 struct btrfs_delayed_item *curr, *next;
1665 struct btrfs_key location;
1666 char *name;
1667 int name_len;
1668 int over = 0;
1669 unsigned char d_type;
1670
1671 if (list_empty(ins_list))
1672 return 0;
1673
1674 /*
1675 * Changing the data of the delayed item is impossible. So
1676 * we needn't lock them. And we have held i_mutex of the
1677 * directory, nobody can delete any directory indexes now.
1678 */
1679 list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1680 list_del(&curr->readdir_list);
1681
1682 if (curr->key.offset < ctx->pos) {
1683 if (refcount_dec_and_test(&curr->refs))
1684 kfree(curr);
1685 continue;
1686 }
1687
1688 ctx->pos = curr->key.offset;
1689
1690 di = (struct btrfs_dir_item *)curr->data;
1691 name = (char *)(di + 1);
1692 name_len = btrfs_stack_dir_name_len(di);
1693
1694 d_type = fs_ftype_to_dtype(di->type);
1695 btrfs_disk_key_to_cpu(&location, &di->location);
1696
1697 over = !dir_emit(ctx, name, name_len,
1698 location.objectid, d_type);
1699
1700 if (refcount_dec_and_test(&curr->refs))
1701 kfree(curr);
1702
1703 if (over)
1704 return 1;
1705 ctx->pos++;
1706 }
1707 return 0;
1708}
1709
1710static void fill_stack_inode_item(struct btrfs_trans_handle *trans,
1711 struct btrfs_inode_item *inode_item,
1712 struct inode *inode)
1713{
1714 btrfs_set_stack_inode_uid(inode_item, i_uid_read(inode));
1715 btrfs_set_stack_inode_gid(inode_item, i_gid_read(inode));
1716 btrfs_set_stack_inode_size(inode_item, BTRFS_I(inode)->disk_i_size);
1717 btrfs_set_stack_inode_mode(inode_item, inode->i_mode);
1718 btrfs_set_stack_inode_nlink(inode_item, inode->i_nlink);
1719 btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(inode));
1720 btrfs_set_stack_inode_generation(inode_item,
1721 BTRFS_I(inode)->generation);
1722 btrfs_set_stack_inode_sequence(inode_item,
1723 inode_peek_iversion(inode));
1724 btrfs_set_stack_inode_transid(inode_item, trans->transid);
1725 btrfs_set_stack_inode_rdev(inode_item, inode->i_rdev);
1726 btrfs_set_stack_inode_flags(inode_item, BTRFS_I(inode)->flags);
1727 btrfs_set_stack_inode_block_group(inode_item, 0);
1728
1729 btrfs_set_stack_timespec_sec(&inode_item->atime,
1730 inode->i_atime.tv_sec);
1731 btrfs_set_stack_timespec_nsec(&inode_item->atime,
1732 inode->i_atime.tv_nsec);
1733
1734 btrfs_set_stack_timespec_sec(&inode_item->mtime,
1735 inode->i_mtime.tv_sec);
1736 btrfs_set_stack_timespec_nsec(&inode_item->mtime,
1737 inode->i_mtime.tv_nsec);
1738
1739 btrfs_set_stack_timespec_sec(&inode_item->ctime,
1740 inode->i_ctime.tv_sec);
1741 btrfs_set_stack_timespec_nsec(&inode_item->ctime,
1742 inode->i_ctime.tv_nsec);
1743
1744 btrfs_set_stack_timespec_sec(&inode_item->otime,
1745 BTRFS_I(inode)->i_otime.tv_sec);
1746 btrfs_set_stack_timespec_nsec(&inode_item->otime,
1747 BTRFS_I(inode)->i_otime.tv_nsec);
1748}
1749
1750int btrfs_fill_inode(struct inode *inode, u32 *rdev)
1751{
1752 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1753 struct btrfs_delayed_node *delayed_node;
1754 struct btrfs_inode_item *inode_item;
1755
1756 delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
1757 if (!delayed_node)
1758 return -ENOENT;
1759
1760 mutex_lock(&delayed_node->mutex);
1761 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1762 mutex_unlock(&delayed_node->mutex);
1763 btrfs_release_delayed_node(delayed_node);
1764 return -ENOENT;
1765 }
1766
1767 inode_item = &delayed_node->inode_item;
1768
1769 i_uid_write(inode, btrfs_stack_inode_uid(inode_item));
1770 i_gid_write(inode, btrfs_stack_inode_gid(inode_item));
1771 btrfs_i_size_write(BTRFS_I(inode), btrfs_stack_inode_size(inode_item));
1772 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
1773 round_up(i_size_read(inode), fs_info->sectorsize));
1774 inode->i_mode = btrfs_stack_inode_mode(inode_item);
1775 set_nlink(inode, btrfs_stack_inode_nlink(inode_item));
1776 inode_set_bytes(inode, btrfs_stack_inode_nbytes(inode_item));
1777 BTRFS_I(inode)->generation = btrfs_stack_inode_generation(inode_item);
1778 BTRFS_I(inode)->last_trans = btrfs_stack_inode_transid(inode_item);
1779
1780 inode_set_iversion_queried(inode,
1781 btrfs_stack_inode_sequence(inode_item));
1782 inode->i_rdev = 0;
1783 *rdev = btrfs_stack_inode_rdev(inode_item);
1784 BTRFS_I(inode)->flags = btrfs_stack_inode_flags(inode_item);
1785
1786 inode->i_atime.tv_sec = btrfs_stack_timespec_sec(&inode_item->atime);
1787 inode->i_atime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->atime);
1788
1789 inode->i_mtime.tv_sec = btrfs_stack_timespec_sec(&inode_item->mtime);
1790 inode->i_mtime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->mtime);
1791
1792 inode->i_ctime.tv_sec = btrfs_stack_timespec_sec(&inode_item->ctime);
1793 inode->i_ctime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->ctime);
1794
1795 BTRFS_I(inode)->i_otime.tv_sec =
1796 btrfs_stack_timespec_sec(&inode_item->otime);
1797 BTRFS_I(inode)->i_otime.tv_nsec =
1798 btrfs_stack_timespec_nsec(&inode_item->otime);
1799
1800 inode->i_generation = BTRFS_I(inode)->generation;
1801 BTRFS_I(inode)->index_cnt = (u64)-1;
1802
1803 mutex_unlock(&delayed_node->mutex);
1804 btrfs_release_delayed_node(delayed_node);
1805 return 0;
1806}
1807
1808int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans,
1809 struct btrfs_root *root,
1810 struct btrfs_inode *inode)
1811{
1812 struct btrfs_delayed_node *delayed_node;
1813 int ret = 0;
1814
1815 delayed_node = btrfs_get_or_create_delayed_node(inode);
1816 if (IS_ERR(delayed_node))
1817 return PTR_ERR(delayed_node);
1818
1819 mutex_lock(&delayed_node->mutex);
1820 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1821 fill_stack_inode_item(trans, &delayed_node->inode_item,
1822 &inode->vfs_inode);
1823 goto release_node;
1824 }
1825
1826 ret = btrfs_delayed_inode_reserve_metadata(trans, root, delayed_node);
1827 if (ret)
1828 goto release_node;
1829
1830 fill_stack_inode_item(trans, &delayed_node->inode_item, &inode->vfs_inode);
1831 set_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
1832 delayed_node->count++;
1833 atomic_inc(&root->fs_info->delayed_root->items);
1834release_node:
1835 mutex_unlock(&delayed_node->mutex);
1836 btrfs_release_delayed_node(delayed_node);
1837 return ret;
1838}
1839
1840int btrfs_delayed_delete_inode_ref(struct btrfs_inode *inode)
1841{
1842 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1843 struct btrfs_delayed_node *delayed_node;
1844
1845 /*
1846 * we don't do delayed inode updates during log recovery because it
1847 * leads to enospc problems. This means we also can't do
1848 * delayed inode refs
1849 */
1850 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
1851 return -EAGAIN;
1852
1853 delayed_node = btrfs_get_or_create_delayed_node(inode);
1854 if (IS_ERR(delayed_node))
1855 return PTR_ERR(delayed_node);
1856
1857 /*
1858 * We don't reserve space for inode ref deletion is because:
1859 * - We ONLY do async inode ref deletion for the inode who has only
1860 * one link(i_nlink == 1), it means there is only one inode ref.
1861 * And in most case, the inode ref and the inode item are in the
1862 * same leaf, and we will deal with them at the same time.
1863 * Since we are sure we will reserve the space for the inode item,
1864 * it is unnecessary to reserve space for inode ref deletion.
1865 * - If the inode ref and the inode item are not in the same leaf,
1866 * We also needn't worry about enospc problem, because we reserve
1867 * much more space for the inode update than it needs.
1868 * - At the worst, we can steal some space from the global reservation.
1869 * It is very rare.
1870 */
1871 mutex_lock(&delayed_node->mutex);
1872 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1873 goto release_node;
1874
1875 set_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
1876 delayed_node->count++;
1877 atomic_inc(&fs_info->delayed_root->items);
1878release_node:
1879 mutex_unlock(&delayed_node->mutex);
1880 btrfs_release_delayed_node(delayed_node);
1881 return 0;
1882}
1883
1884static void __btrfs_kill_delayed_node(struct btrfs_delayed_node *delayed_node)
1885{
1886 struct btrfs_root *root = delayed_node->root;
1887 struct btrfs_fs_info *fs_info = root->fs_info;
1888 struct btrfs_delayed_item *curr_item, *prev_item;
1889
1890 mutex_lock(&delayed_node->mutex);
1891 curr_item = __btrfs_first_delayed_insertion_item(delayed_node);
1892 while (curr_item) {
1893 btrfs_delayed_item_release_metadata(root, curr_item);
1894 prev_item = curr_item;
1895 curr_item = __btrfs_next_delayed_item(prev_item);
1896 btrfs_release_delayed_item(prev_item);
1897 }
1898
1899 curr_item = __btrfs_first_delayed_deletion_item(delayed_node);
1900 while (curr_item) {
1901 btrfs_delayed_item_release_metadata(root, curr_item);
1902 prev_item = curr_item;
1903 curr_item = __btrfs_next_delayed_item(prev_item);
1904 btrfs_release_delayed_item(prev_item);
1905 }
1906
1907 btrfs_release_delayed_iref(delayed_node);
1908
1909 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1910 btrfs_delayed_inode_release_metadata(fs_info, delayed_node, false);
1911 btrfs_release_delayed_inode(delayed_node);
1912 }
1913 mutex_unlock(&delayed_node->mutex);
1914}
1915
1916void btrfs_kill_delayed_inode_items(struct btrfs_inode *inode)
1917{
1918 struct btrfs_delayed_node *delayed_node;
1919
1920 delayed_node = btrfs_get_delayed_node(inode);
1921 if (!delayed_node)
1922 return;
1923
1924 __btrfs_kill_delayed_node(delayed_node);
1925 btrfs_release_delayed_node(delayed_node);
1926}
1927
1928void btrfs_kill_all_delayed_nodes(struct btrfs_root *root)
1929{
1930 u64 inode_id = 0;
1931 struct btrfs_delayed_node *delayed_nodes[8];
1932 int i, n;
1933
1934 while (1) {
1935 spin_lock(&root->inode_lock);
1936 n = radix_tree_gang_lookup(&root->delayed_nodes_tree,
1937 (void **)delayed_nodes, inode_id,
1938 ARRAY_SIZE(delayed_nodes));
1939 if (!n) {
1940 spin_unlock(&root->inode_lock);
1941 break;
1942 }
1943
1944 inode_id = delayed_nodes[n - 1]->inode_id + 1;
1945 for (i = 0; i < n; i++) {
1946 /*
1947 * Don't increase refs in case the node is dead and
1948 * about to be removed from the tree in the loop below
1949 */
1950 if (!refcount_inc_not_zero(&delayed_nodes[i]->refs))
1951 delayed_nodes[i] = NULL;
1952 }
1953 spin_unlock(&root->inode_lock);
1954
1955 for (i = 0; i < n; i++) {
1956 if (!delayed_nodes[i])
1957 continue;
1958 __btrfs_kill_delayed_node(delayed_nodes[i]);
1959 btrfs_release_delayed_node(delayed_nodes[i]);
1960 }
1961 }
1962}
1963
1964void btrfs_destroy_delayed_inodes(struct btrfs_fs_info *fs_info)
1965{
1966 struct btrfs_delayed_node *curr_node, *prev_node;
1967
1968 curr_node = btrfs_first_delayed_node(fs_info->delayed_root);
1969 while (curr_node) {
1970 __btrfs_kill_delayed_node(curr_node);
1971
1972 prev_node = curr_node;
1973 curr_node = btrfs_next_delayed_node(curr_node);
1974 btrfs_release_delayed_node(prev_node);
1975 }
1976}
1977
1/*
2 * Copyright (C) 2011 Fujitsu. All rights reserved.
3 * Written by Miao Xie <miaox@cn.fujitsu.com>
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public
7 * License v2 as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12 * General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public
15 * License along with this program; if not, write to the
16 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
17 * Boston, MA 021110-1307, USA.
18 */
19
20#include <linux/slab.h>
21#include "delayed-inode.h"
22#include "disk-io.h"
23#include "transaction.h"
24#include "ctree.h"
25
26#define BTRFS_DELAYED_WRITEBACK 512
27#define BTRFS_DELAYED_BACKGROUND 128
28#define BTRFS_DELAYED_BATCH 16
29
30static struct kmem_cache *delayed_node_cache;
31
32int __init btrfs_delayed_inode_init(void)
33{
34 delayed_node_cache = kmem_cache_create("btrfs_delayed_node",
35 sizeof(struct btrfs_delayed_node),
36 0,
37 SLAB_MEM_SPREAD,
38 NULL);
39 if (!delayed_node_cache)
40 return -ENOMEM;
41 return 0;
42}
43
44void btrfs_delayed_inode_exit(void)
45{
46 kmem_cache_destroy(delayed_node_cache);
47}
48
49static 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
63static 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
75static struct btrfs_delayed_node *btrfs_get_delayed_node(struct inode *inode)
76{
77 struct btrfs_inode *btrfs_inode = BTRFS_I(inode);
78 struct btrfs_root *root = btrfs_inode->root;
79 u64 ino = btrfs_ino(inode);
80 struct btrfs_delayed_node *node;
81
82 node = ACCESS_ONCE(btrfs_inode->delayed_node);
83 if (node) {
84 atomic_inc(&node->refs);
85 return node;
86 }
87
88 spin_lock(&root->inode_lock);
89 node = radix_tree_lookup(&root->delayed_nodes_tree, ino);
90 if (node) {
91 if (btrfs_inode->delayed_node) {
92 atomic_inc(&node->refs); /* can be accessed */
93 BUG_ON(btrfs_inode->delayed_node != node);
94 spin_unlock(&root->inode_lock);
95 return node;
96 }
97 btrfs_inode->delayed_node = node;
98 /* can be accessed and cached in the inode */
99 atomic_add(2, &node->refs);
100 spin_unlock(&root->inode_lock);
101 return node;
102 }
103 spin_unlock(&root->inode_lock);
104
105 return NULL;
106}
107
108/* Will return either the node or PTR_ERR(-ENOMEM) */
109static struct btrfs_delayed_node *btrfs_get_or_create_delayed_node(
110 struct inode *inode)
111{
112 struct btrfs_delayed_node *node;
113 struct btrfs_inode *btrfs_inode = BTRFS_I(inode);
114 struct btrfs_root *root = btrfs_inode->root;
115 u64 ino = btrfs_ino(inode);
116 int ret;
117
118again:
119 node = btrfs_get_delayed_node(inode);
120 if (node)
121 return node;
122
123 node = kmem_cache_zalloc(delayed_node_cache, GFP_NOFS);
124 if (!node)
125 return ERR_PTR(-ENOMEM);
126 btrfs_init_delayed_node(node, root, ino);
127
128 /* cached in the btrfs inode and can be accessed */
129 atomic_add(2, &node->refs);
130
131 ret = radix_tree_preload(GFP_NOFS);
132 if (ret) {
133 kmem_cache_free(delayed_node_cache, node);
134 return ERR_PTR(ret);
135 }
136
137 spin_lock(&root->inode_lock);
138 ret = radix_tree_insert(&root->delayed_nodes_tree, ino, node);
139 if (ret == -EEXIST) {
140 spin_unlock(&root->inode_lock);
141 kmem_cache_free(delayed_node_cache, node);
142 radix_tree_preload_end();
143 goto again;
144 }
145 btrfs_inode->delayed_node = node;
146 spin_unlock(&root->inode_lock);
147 radix_tree_preload_end();
148
149 return node;
150}
151
152/*
153 * Call it when holding delayed_node->mutex
154 *
155 * If mod = 1, add this node into the prepared list.
156 */
157static void btrfs_queue_delayed_node(struct btrfs_delayed_root *root,
158 struct btrfs_delayed_node *node,
159 int mod)
160{
161 spin_lock(&root->lock);
162 if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
163 if (!list_empty(&node->p_list))
164 list_move_tail(&node->p_list, &root->prepare_list);
165 else if (mod)
166 list_add_tail(&node->p_list, &root->prepare_list);
167 } else {
168 list_add_tail(&node->n_list, &root->node_list);
169 list_add_tail(&node->p_list, &root->prepare_list);
170 atomic_inc(&node->refs); /* inserted into list */
171 root->nodes++;
172 set_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
173 }
174 spin_unlock(&root->lock);
175}
176
177/* Call it when holding delayed_node->mutex */
178static void btrfs_dequeue_delayed_node(struct btrfs_delayed_root *root,
179 struct btrfs_delayed_node *node)
180{
181 spin_lock(&root->lock);
182 if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
183 root->nodes--;
184 atomic_dec(&node->refs); /* not in the list */
185 list_del_init(&node->n_list);
186 if (!list_empty(&node->p_list))
187 list_del_init(&node->p_list);
188 clear_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
189 }
190 spin_unlock(&root->lock);
191}
192
193static struct btrfs_delayed_node *btrfs_first_delayed_node(
194 struct btrfs_delayed_root *delayed_root)
195{
196 struct list_head *p;
197 struct btrfs_delayed_node *node = NULL;
198
199 spin_lock(&delayed_root->lock);
200 if (list_empty(&delayed_root->node_list))
201 goto out;
202
203 p = delayed_root->node_list.next;
204 node = list_entry(p, struct btrfs_delayed_node, n_list);
205 atomic_inc(&node->refs);
206out:
207 spin_unlock(&delayed_root->lock);
208
209 return node;
210}
211
212static struct btrfs_delayed_node *btrfs_next_delayed_node(
213 struct btrfs_delayed_node *node)
214{
215 struct btrfs_delayed_root *delayed_root;
216 struct list_head *p;
217 struct btrfs_delayed_node *next = NULL;
218
219 delayed_root = node->root->fs_info->delayed_root;
220 spin_lock(&delayed_root->lock);
221 if (!test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
222 /* not in the list */
223 if (list_empty(&delayed_root->node_list))
224 goto out;
225 p = delayed_root->node_list.next;
226 } else if (list_is_last(&node->n_list, &delayed_root->node_list))
227 goto out;
228 else
229 p = node->n_list.next;
230
231 next = list_entry(p, struct btrfs_delayed_node, n_list);
232 atomic_inc(&next->refs);
233out:
234 spin_unlock(&delayed_root->lock);
235
236 return next;
237}
238
239static void __btrfs_release_delayed_node(
240 struct btrfs_delayed_node *delayed_node,
241 int mod)
242{
243 struct btrfs_delayed_root *delayed_root;
244
245 if (!delayed_node)
246 return;
247
248 delayed_root = delayed_node->root->fs_info->delayed_root;
249
250 mutex_lock(&delayed_node->mutex);
251 if (delayed_node->count)
252 btrfs_queue_delayed_node(delayed_root, delayed_node, mod);
253 else
254 btrfs_dequeue_delayed_node(delayed_root, delayed_node);
255 mutex_unlock(&delayed_node->mutex);
256
257 if (atomic_dec_and_test(&delayed_node->refs)) {
258 bool free = false;
259 struct btrfs_root *root = delayed_node->root;
260 spin_lock(&root->inode_lock);
261 if (atomic_read(&delayed_node->refs) == 0) {
262 radix_tree_delete(&root->delayed_nodes_tree,
263 delayed_node->inode_id);
264 free = true;
265 }
266 spin_unlock(&root->inode_lock);
267 if (free)
268 kmem_cache_free(delayed_node_cache, delayed_node);
269 }
270}
271
272static inline void btrfs_release_delayed_node(struct btrfs_delayed_node *node)
273{
274 __btrfs_release_delayed_node(node, 0);
275}
276
277static struct btrfs_delayed_node *btrfs_first_prepared_delayed_node(
278 struct btrfs_delayed_root *delayed_root)
279{
280 struct list_head *p;
281 struct btrfs_delayed_node *node = NULL;
282
283 spin_lock(&delayed_root->lock);
284 if (list_empty(&delayed_root->prepare_list))
285 goto out;
286
287 p = delayed_root->prepare_list.next;
288 list_del_init(p);
289 node = list_entry(p, struct btrfs_delayed_node, p_list);
290 atomic_inc(&node->refs);
291out:
292 spin_unlock(&delayed_root->lock);
293
294 return node;
295}
296
297static inline void btrfs_release_prepared_delayed_node(
298 struct btrfs_delayed_node *node)
299{
300 __btrfs_release_delayed_node(node, 1);
301}
302
303static struct btrfs_delayed_item *btrfs_alloc_delayed_item(u32 data_len)
304{
305 struct btrfs_delayed_item *item;
306 item = kmalloc(sizeof(*item) + data_len, GFP_NOFS);
307 if (item) {
308 item->data_len = data_len;
309 item->ins_or_del = 0;
310 item->bytes_reserved = 0;
311 item->delayed_node = NULL;
312 atomic_set(&item->refs, 1);
313 }
314 return item;
315}
316
317/*
318 * __btrfs_lookup_delayed_item - look up the delayed item by key
319 * @delayed_node: pointer to the delayed node
320 * @key: the key to look up
321 * @prev: used to store the prev item if the right item isn't found
322 * @next: used to store the next item if the right item isn't found
323 *
324 * Note: if we don't find the right item, we will return the prev item and
325 * the next item.
326 */
327static struct btrfs_delayed_item *__btrfs_lookup_delayed_item(
328 struct rb_root *root,
329 struct btrfs_key *key,
330 struct btrfs_delayed_item **prev,
331 struct btrfs_delayed_item **next)
332{
333 struct rb_node *node, *prev_node = NULL;
334 struct btrfs_delayed_item *delayed_item = NULL;
335 int ret = 0;
336
337 node = root->rb_node;
338
339 while (node) {
340 delayed_item = rb_entry(node, struct btrfs_delayed_item,
341 rb_node);
342 prev_node = node;
343 ret = btrfs_comp_cpu_keys(&delayed_item->key, key);
344 if (ret < 0)
345 node = node->rb_right;
346 else if (ret > 0)
347 node = node->rb_left;
348 else
349 return delayed_item;
350 }
351
352 if (prev) {
353 if (!prev_node)
354 *prev = NULL;
355 else if (ret < 0)
356 *prev = delayed_item;
357 else if ((node = rb_prev(prev_node)) != NULL) {
358 *prev = rb_entry(node, struct btrfs_delayed_item,
359 rb_node);
360 } else
361 *prev = NULL;
362 }
363
364 if (next) {
365 if (!prev_node)
366 *next = NULL;
367 else if (ret > 0)
368 *next = delayed_item;
369 else if ((node = rb_next(prev_node)) != NULL) {
370 *next = rb_entry(node, struct btrfs_delayed_item,
371 rb_node);
372 } else
373 *next = NULL;
374 }
375 return NULL;
376}
377
378static struct btrfs_delayed_item *__btrfs_lookup_delayed_insertion_item(
379 struct btrfs_delayed_node *delayed_node,
380 struct btrfs_key *key)
381{
382 return __btrfs_lookup_delayed_item(&delayed_node->ins_root, key,
383 NULL, NULL);
384}
385
386static int __btrfs_add_delayed_item(struct btrfs_delayed_node *delayed_node,
387 struct btrfs_delayed_item *ins,
388 int action)
389{
390 struct rb_node **p, *node;
391 struct rb_node *parent_node = NULL;
392 struct rb_root *root;
393 struct btrfs_delayed_item *item;
394 int cmp;
395
396 if (action == BTRFS_DELAYED_INSERTION_ITEM)
397 root = &delayed_node->ins_root;
398 else if (action == BTRFS_DELAYED_DELETION_ITEM)
399 root = &delayed_node->del_root;
400 else
401 BUG();
402 p = &root->rb_node;
403 node = &ins->rb_node;
404
405 while (*p) {
406 parent_node = *p;
407 item = rb_entry(parent_node, struct btrfs_delayed_item,
408 rb_node);
409
410 cmp = btrfs_comp_cpu_keys(&item->key, &ins->key);
411 if (cmp < 0)
412 p = &(*p)->rb_right;
413 else if (cmp > 0)
414 p = &(*p)->rb_left;
415 else
416 return -EEXIST;
417 }
418
419 rb_link_node(node, parent_node, p);
420 rb_insert_color(node, root);
421 ins->delayed_node = delayed_node;
422 ins->ins_or_del = action;
423
424 if (ins->key.type == BTRFS_DIR_INDEX_KEY &&
425 action == BTRFS_DELAYED_INSERTION_ITEM &&
426 ins->key.offset >= delayed_node->index_cnt)
427 delayed_node->index_cnt = ins->key.offset + 1;
428
429 delayed_node->count++;
430 atomic_inc(&delayed_node->root->fs_info->delayed_root->items);
431 return 0;
432}
433
434static int __btrfs_add_delayed_insertion_item(struct btrfs_delayed_node *node,
435 struct btrfs_delayed_item *item)
436{
437 return __btrfs_add_delayed_item(node, item,
438 BTRFS_DELAYED_INSERTION_ITEM);
439}
440
441static int __btrfs_add_delayed_deletion_item(struct btrfs_delayed_node *node,
442 struct btrfs_delayed_item *item)
443{
444 return __btrfs_add_delayed_item(node, item,
445 BTRFS_DELAYED_DELETION_ITEM);
446}
447
448static void finish_one_item(struct btrfs_delayed_root *delayed_root)
449{
450 int seq = atomic_inc_return(&delayed_root->items_seq);
451
452 /*
453 * atomic_dec_return implies a barrier for waitqueue_active
454 */
455 if ((atomic_dec_return(&delayed_root->items) <
456 BTRFS_DELAYED_BACKGROUND || seq % BTRFS_DELAYED_BATCH == 0) &&
457 waitqueue_active(&delayed_root->wait))
458 wake_up(&delayed_root->wait);
459}
460
461static void __btrfs_remove_delayed_item(struct btrfs_delayed_item *delayed_item)
462{
463 struct rb_root *root;
464 struct btrfs_delayed_root *delayed_root;
465
466 delayed_root = delayed_item->delayed_node->root->fs_info->delayed_root;
467
468 BUG_ON(!delayed_root);
469 BUG_ON(delayed_item->ins_or_del != BTRFS_DELAYED_DELETION_ITEM &&
470 delayed_item->ins_or_del != BTRFS_DELAYED_INSERTION_ITEM);
471
472 if (delayed_item->ins_or_del == BTRFS_DELAYED_INSERTION_ITEM)
473 root = &delayed_item->delayed_node->ins_root;
474 else
475 root = &delayed_item->delayed_node->del_root;
476
477 rb_erase(&delayed_item->rb_node, root);
478 delayed_item->delayed_node->count--;
479
480 finish_one_item(delayed_root);
481}
482
483static void btrfs_release_delayed_item(struct btrfs_delayed_item *item)
484{
485 if (item) {
486 __btrfs_remove_delayed_item(item);
487 if (atomic_dec_and_test(&item->refs))
488 kfree(item);
489 }
490}
491
492static struct btrfs_delayed_item *__btrfs_first_delayed_insertion_item(
493 struct btrfs_delayed_node *delayed_node)
494{
495 struct rb_node *p;
496 struct btrfs_delayed_item *item = NULL;
497
498 p = rb_first(&delayed_node->ins_root);
499 if (p)
500 item = rb_entry(p, struct btrfs_delayed_item, rb_node);
501
502 return item;
503}
504
505static struct btrfs_delayed_item *__btrfs_first_delayed_deletion_item(
506 struct btrfs_delayed_node *delayed_node)
507{
508 struct rb_node *p;
509 struct btrfs_delayed_item *item = NULL;
510
511 p = rb_first(&delayed_node->del_root);
512 if (p)
513 item = rb_entry(p, struct btrfs_delayed_item, rb_node);
514
515 return item;
516}
517
518static struct btrfs_delayed_item *__btrfs_next_delayed_item(
519 struct btrfs_delayed_item *item)
520{
521 struct rb_node *p;
522 struct btrfs_delayed_item *next = NULL;
523
524 p = rb_next(&item->rb_node);
525 if (p)
526 next = rb_entry(p, struct btrfs_delayed_item, rb_node);
527
528 return next;
529}
530
531static int btrfs_delayed_item_reserve_metadata(struct btrfs_trans_handle *trans,
532 struct btrfs_fs_info *fs_info,
533 struct btrfs_delayed_item *item)
534{
535 struct btrfs_block_rsv *src_rsv;
536 struct btrfs_block_rsv *dst_rsv;
537 u64 num_bytes;
538 int ret;
539
540 if (!trans->bytes_reserved)
541 return 0;
542
543 src_rsv = trans->block_rsv;
544 dst_rsv = &fs_info->delayed_block_rsv;
545
546 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
547 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1);
548 if (!ret) {
549 trace_btrfs_space_reservation(fs_info, "delayed_item",
550 item->key.objectid,
551 num_bytes, 1);
552 item->bytes_reserved = num_bytes;
553 }
554
555 return ret;
556}
557
558static void btrfs_delayed_item_release_metadata(struct btrfs_fs_info *fs_info,
559 struct btrfs_delayed_item *item)
560{
561 struct btrfs_block_rsv *rsv;
562
563 if (!item->bytes_reserved)
564 return;
565
566 rsv = &fs_info->delayed_block_rsv;
567 trace_btrfs_space_reservation(fs_info, "delayed_item",
568 item->key.objectid, item->bytes_reserved,
569 0);
570 btrfs_block_rsv_release(fs_info, rsv,
571 item->bytes_reserved);
572}
573
574static int btrfs_delayed_inode_reserve_metadata(
575 struct btrfs_trans_handle *trans,
576 struct btrfs_root *root,
577 struct inode *inode,
578 struct btrfs_delayed_node *node)
579{
580 struct btrfs_fs_info *fs_info = root->fs_info;
581 struct btrfs_block_rsv *src_rsv;
582 struct btrfs_block_rsv *dst_rsv;
583 u64 num_bytes;
584 int ret;
585 bool release = false;
586
587 src_rsv = trans->block_rsv;
588 dst_rsv = &fs_info->delayed_block_rsv;
589
590 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
591
592 /*
593 * If our block_rsv is the delalloc block reserve then check and see if
594 * we have our extra reservation for updating the inode. If not fall
595 * through and try to reserve space quickly.
596 *
597 * We used to try and steal from the delalloc block rsv or the global
598 * reserve, but we'd steal a full reservation, which isn't kind. We are
599 * here through delalloc which means we've likely just cowed down close
600 * to the leaf that contains the inode, so we would steal less just
601 * doing the fallback inode update, so if we do end up having to steal
602 * from the global block rsv we hopefully only steal one or two blocks
603 * worth which is less likely to hurt us.
604 */
605 if (src_rsv && src_rsv->type == BTRFS_BLOCK_RSV_DELALLOC) {
606 spin_lock(&BTRFS_I(inode)->lock);
607 if (test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
608 &BTRFS_I(inode)->runtime_flags))
609 release = true;
610 else
611 src_rsv = NULL;
612 spin_unlock(&BTRFS_I(inode)->lock);
613 }
614
615 /*
616 * btrfs_dirty_inode will update the inode under btrfs_join_transaction
617 * which doesn't reserve space for speed. This is a problem since we
618 * still need to reserve space for this update, so try to reserve the
619 * space.
620 *
621 * Now if src_rsv == delalloc_block_rsv we'll let it just steal since
622 * we're accounted for.
623 */
624 if (!src_rsv || (!trans->bytes_reserved &&
625 src_rsv->type != BTRFS_BLOCK_RSV_DELALLOC)) {
626 ret = btrfs_block_rsv_add(root, dst_rsv, num_bytes,
627 BTRFS_RESERVE_NO_FLUSH);
628 /*
629 * Since we're under a transaction reserve_metadata_bytes could
630 * try to commit the transaction which will make it return
631 * EAGAIN to make us stop the transaction we have, so return
632 * ENOSPC instead so that btrfs_dirty_inode knows what to do.
633 */
634 if (ret == -EAGAIN)
635 ret = -ENOSPC;
636 if (!ret) {
637 node->bytes_reserved = num_bytes;
638 trace_btrfs_space_reservation(fs_info,
639 "delayed_inode",
640 btrfs_ino(inode),
641 num_bytes, 1);
642 }
643 return ret;
644 }
645
646 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1);
647
648 /*
649 * Migrate only takes a reservation, it doesn't touch the size of the
650 * block_rsv. This is to simplify people who don't normally have things
651 * migrated from their block rsv. If they go to release their
652 * reservation, that will decrease the size as well, so if migrate
653 * reduced size we'd end up with a negative size. But for the
654 * delalloc_meta_reserved stuff we will only know to drop 1 reservation,
655 * but we could in fact do this reserve/migrate dance several times
656 * between the time we did the original reservation and we'd clean it
657 * up. So to take care of this, release the space for the meta
658 * reservation here. I think it may be time for a documentation page on
659 * how block rsvs. work.
660 */
661 if (!ret) {
662 trace_btrfs_space_reservation(fs_info, "delayed_inode",
663 btrfs_ino(inode), num_bytes, 1);
664 node->bytes_reserved = num_bytes;
665 }
666
667 if (release) {
668 trace_btrfs_space_reservation(fs_info, "delalloc",
669 btrfs_ino(inode), num_bytes, 0);
670 btrfs_block_rsv_release(fs_info, src_rsv, num_bytes);
671 }
672
673 return ret;
674}
675
676static void btrfs_delayed_inode_release_metadata(struct btrfs_fs_info *fs_info,
677 struct btrfs_delayed_node *node)
678{
679 struct btrfs_block_rsv *rsv;
680
681 if (!node->bytes_reserved)
682 return;
683
684 rsv = &fs_info->delayed_block_rsv;
685 trace_btrfs_space_reservation(fs_info, "delayed_inode",
686 node->inode_id, node->bytes_reserved, 0);
687 btrfs_block_rsv_release(fs_info, rsv,
688 node->bytes_reserved);
689 node->bytes_reserved = 0;
690}
691
692/*
693 * This helper will insert some continuous items into the same leaf according
694 * to the free space of the leaf.
695 */
696static int btrfs_batch_insert_items(struct btrfs_root *root,
697 struct btrfs_path *path,
698 struct btrfs_delayed_item *item)
699{
700 struct btrfs_fs_info *fs_info = root->fs_info;
701 struct btrfs_delayed_item *curr, *next;
702 int free_space;
703 int total_data_size = 0, total_size = 0;
704 struct extent_buffer *leaf;
705 char *data_ptr;
706 struct btrfs_key *keys;
707 u32 *data_size;
708 struct list_head head;
709 int slot;
710 int nitems;
711 int i;
712 int ret = 0;
713
714 BUG_ON(!path->nodes[0]);
715
716 leaf = path->nodes[0];
717 free_space = btrfs_leaf_free_space(fs_info, leaf);
718 INIT_LIST_HEAD(&head);
719
720 next = item;
721 nitems = 0;
722
723 /*
724 * count the number of the continuous items that we can insert in batch
725 */
726 while (total_size + next->data_len + sizeof(struct btrfs_item) <=
727 free_space) {
728 total_data_size += next->data_len;
729 total_size += next->data_len + sizeof(struct btrfs_item);
730 list_add_tail(&next->tree_list, &head);
731 nitems++;
732
733 curr = next;
734 next = __btrfs_next_delayed_item(curr);
735 if (!next)
736 break;
737
738 if (!btrfs_is_continuous_delayed_item(curr, next))
739 break;
740 }
741
742 if (!nitems) {
743 ret = 0;
744 goto out;
745 }
746
747 /*
748 * we need allocate some memory space, but it might cause the task
749 * to sleep, so we set all locked nodes in the path to blocking locks
750 * first.
751 */
752 btrfs_set_path_blocking(path);
753
754 keys = kmalloc_array(nitems, sizeof(struct btrfs_key), GFP_NOFS);
755 if (!keys) {
756 ret = -ENOMEM;
757 goto out;
758 }
759
760 data_size = kmalloc_array(nitems, sizeof(u32), GFP_NOFS);
761 if (!data_size) {
762 ret = -ENOMEM;
763 goto error;
764 }
765
766 /* get keys of all the delayed items */
767 i = 0;
768 list_for_each_entry(next, &head, tree_list) {
769 keys[i] = next->key;
770 data_size[i] = next->data_len;
771 i++;
772 }
773
774 /* reset all the locked nodes in the patch to spinning locks. */
775 btrfs_clear_path_blocking(path, NULL, 0);
776
777 /* insert the keys of the items */
778 setup_items_for_insert(root, path, keys, data_size,
779 total_data_size, total_size, nitems);
780
781 /* insert the dir index items */
782 slot = path->slots[0];
783 list_for_each_entry_safe(curr, next, &head, tree_list) {
784 data_ptr = btrfs_item_ptr(leaf, slot, char);
785 write_extent_buffer(leaf, &curr->data,
786 (unsigned long)data_ptr,
787 curr->data_len);
788 slot++;
789
790 btrfs_delayed_item_release_metadata(fs_info, curr);
791
792 list_del(&curr->tree_list);
793 btrfs_release_delayed_item(curr);
794 }
795
796error:
797 kfree(data_size);
798 kfree(keys);
799out:
800 return ret;
801}
802
803/*
804 * This helper can just do simple insertion that needn't extend item for new
805 * data, such as directory name index insertion, inode insertion.
806 */
807static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans,
808 struct btrfs_root *root,
809 struct btrfs_path *path,
810 struct btrfs_delayed_item *delayed_item)
811{
812 struct btrfs_fs_info *fs_info = root->fs_info;
813 struct extent_buffer *leaf;
814 char *ptr;
815 int ret;
816
817 ret = btrfs_insert_empty_item(trans, root, path, &delayed_item->key,
818 delayed_item->data_len);
819 if (ret < 0 && ret != -EEXIST)
820 return ret;
821
822 leaf = path->nodes[0];
823
824 ptr = btrfs_item_ptr(leaf, path->slots[0], char);
825
826 write_extent_buffer(leaf, delayed_item->data, (unsigned long)ptr,
827 delayed_item->data_len);
828 btrfs_mark_buffer_dirty(leaf);
829
830 btrfs_delayed_item_release_metadata(fs_info, delayed_item);
831 return 0;
832}
833
834/*
835 * we insert an item first, then if there are some continuous items, we try
836 * to insert those items into the same leaf.
837 */
838static int btrfs_insert_delayed_items(struct btrfs_trans_handle *trans,
839 struct btrfs_path *path,
840 struct btrfs_root *root,
841 struct btrfs_delayed_node *node)
842{
843 struct btrfs_delayed_item *curr, *prev;
844 int ret = 0;
845
846do_again:
847 mutex_lock(&node->mutex);
848 curr = __btrfs_first_delayed_insertion_item(node);
849 if (!curr)
850 goto insert_end;
851
852 ret = btrfs_insert_delayed_item(trans, root, path, curr);
853 if (ret < 0) {
854 btrfs_release_path(path);
855 goto insert_end;
856 }
857
858 prev = curr;
859 curr = __btrfs_next_delayed_item(prev);
860 if (curr && btrfs_is_continuous_delayed_item(prev, curr)) {
861 /* insert the continuous items into the same leaf */
862 path->slots[0]++;
863 btrfs_batch_insert_items(root, path, curr);
864 }
865 btrfs_release_delayed_item(prev);
866 btrfs_mark_buffer_dirty(path->nodes[0]);
867
868 btrfs_release_path(path);
869 mutex_unlock(&node->mutex);
870 goto do_again;
871
872insert_end:
873 mutex_unlock(&node->mutex);
874 return ret;
875}
876
877static int btrfs_batch_delete_items(struct btrfs_trans_handle *trans,
878 struct btrfs_root *root,
879 struct btrfs_path *path,
880 struct btrfs_delayed_item *item)
881{
882 struct btrfs_fs_info *fs_info = root->fs_info;
883 struct btrfs_delayed_item *curr, *next;
884 struct extent_buffer *leaf;
885 struct btrfs_key key;
886 struct list_head head;
887 int nitems, i, last_item;
888 int ret = 0;
889
890 BUG_ON(!path->nodes[0]);
891
892 leaf = path->nodes[0];
893
894 i = path->slots[0];
895 last_item = btrfs_header_nritems(leaf) - 1;
896 if (i > last_item)
897 return -ENOENT; /* FIXME: Is errno suitable? */
898
899 next = item;
900 INIT_LIST_HEAD(&head);
901 btrfs_item_key_to_cpu(leaf, &key, i);
902 nitems = 0;
903 /*
904 * count the number of the dir index items that we can delete in batch
905 */
906 while (btrfs_comp_cpu_keys(&next->key, &key) == 0) {
907 list_add_tail(&next->tree_list, &head);
908 nitems++;
909
910 curr = next;
911 next = __btrfs_next_delayed_item(curr);
912 if (!next)
913 break;
914
915 if (!btrfs_is_continuous_delayed_item(curr, next))
916 break;
917
918 i++;
919 if (i > last_item)
920 break;
921 btrfs_item_key_to_cpu(leaf, &key, i);
922 }
923
924 if (!nitems)
925 return 0;
926
927 ret = btrfs_del_items(trans, root, path, path->slots[0], nitems);
928 if (ret)
929 goto out;
930
931 list_for_each_entry_safe(curr, next, &head, tree_list) {
932 btrfs_delayed_item_release_metadata(fs_info, curr);
933 list_del(&curr->tree_list);
934 btrfs_release_delayed_item(curr);
935 }
936
937out:
938 return ret;
939}
940
941static int btrfs_delete_delayed_items(struct btrfs_trans_handle *trans,
942 struct btrfs_path *path,
943 struct btrfs_root *root,
944 struct btrfs_delayed_node *node)
945{
946 struct btrfs_delayed_item *curr, *prev;
947 int ret = 0;
948
949do_again:
950 mutex_lock(&node->mutex);
951 curr = __btrfs_first_delayed_deletion_item(node);
952 if (!curr)
953 goto delete_fail;
954
955 ret = btrfs_search_slot(trans, root, &curr->key, path, -1, 1);
956 if (ret < 0)
957 goto delete_fail;
958 else if (ret > 0) {
959 /*
960 * can't find the item which the node points to, so this node
961 * is invalid, just drop it.
962 */
963 prev = curr;
964 curr = __btrfs_next_delayed_item(prev);
965 btrfs_release_delayed_item(prev);
966 ret = 0;
967 btrfs_release_path(path);
968 if (curr) {
969 mutex_unlock(&node->mutex);
970 goto do_again;
971 } else
972 goto delete_fail;
973 }
974
975 btrfs_batch_delete_items(trans, root, path, curr);
976 btrfs_release_path(path);
977 mutex_unlock(&node->mutex);
978 goto do_again;
979
980delete_fail:
981 btrfs_release_path(path);
982 mutex_unlock(&node->mutex);
983 return ret;
984}
985
986static void btrfs_release_delayed_inode(struct btrfs_delayed_node *delayed_node)
987{
988 struct btrfs_delayed_root *delayed_root;
989
990 if (delayed_node &&
991 test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
992 BUG_ON(!delayed_node->root);
993 clear_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
994 delayed_node->count--;
995
996 delayed_root = delayed_node->root->fs_info->delayed_root;
997 finish_one_item(delayed_root);
998 }
999}
1000
1001static void btrfs_release_delayed_iref(struct btrfs_delayed_node *delayed_node)
1002{
1003 struct btrfs_delayed_root *delayed_root;
1004
1005 ASSERT(delayed_node->root);
1006 clear_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
1007 delayed_node->count--;
1008
1009 delayed_root = delayed_node->root->fs_info->delayed_root;
1010 finish_one_item(delayed_root);
1011}
1012
1013static int __btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1014 struct btrfs_root *root,
1015 struct btrfs_path *path,
1016 struct btrfs_delayed_node *node)
1017{
1018 struct btrfs_fs_info *fs_info = root->fs_info;
1019 struct btrfs_key key;
1020 struct btrfs_inode_item *inode_item;
1021 struct extent_buffer *leaf;
1022 int mod;
1023 int ret;
1024
1025 key.objectid = node->inode_id;
1026 key.type = BTRFS_INODE_ITEM_KEY;
1027 key.offset = 0;
1028
1029 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1030 mod = -1;
1031 else
1032 mod = 1;
1033
1034 ret = btrfs_lookup_inode(trans, root, path, &key, mod);
1035 if (ret > 0) {
1036 btrfs_release_path(path);
1037 return -ENOENT;
1038 } else if (ret < 0) {
1039 return ret;
1040 }
1041
1042 leaf = path->nodes[0];
1043 inode_item = btrfs_item_ptr(leaf, path->slots[0],
1044 struct btrfs_inode_item);
1045 write_extent_buffer(leaf, &node->inode_item, (unsigned long)inode_item,
1046 sizeof(struct btrfs_inode_item));
1047 btrfs_mark_buffer_dirty(leaf);
1048
1049 if (!test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1050 goto no_iref;
1051
1052 path->slots[0]++;
1053 if (path->slots[0] >= btrfs_header_nritems(leaf))
1054 goto search;
1055again:
1056 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1057 if (key.objectid != node->inode_id)
1058 goto out;
1059
1060 if (key.type != BTRFS_INODE_REF_KEY &&
1061 key.type != BTRFS_INODE_EXTREF_KEY)
1062 goto out;
1063
1064 /*
1065 * Delayed iref deletion is for the inode who has only one link,
1066 * so there is only one iref. The case that several irefs are
1067 * in the same item doesn't exist.
1068 */
1069 btrfs_del_item(trans, root, path);
1070out:
1071 btrfs_release_delayed_iref(node);
1072no_iref:
1073 btrfs_release_path(path);
1074err_out:
1075 btrfs_delayed_inode_release_metadata(fs_info, node);
1076 btrfs_release_delayed_inode(node);
1077
1078 return ret;
1079
1080search:
1081 btrfs_release_path(path);
1082
1083 key.type = BTRFS_INODE_EXTREF_KEY;
1084 key.offset = -1;
1085 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1086 if (ret < 0)
1087 goto err_out;
1088 ASSERT(ret);
1089
1090 ret = 0;
1091 leaf = path->nodes[0];
1092 path->slots[0]--;
1093 goto again;
1094}
1095
1096static inline int btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1097 struct btrfs_root *root,
1098 struct btrfs_path *path,
1099 struct btrfs_delayed_node *node)
1100{
1101 int ret;
1102
1103 mutex_lock(&node->mutex);
1104 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &node->flags)) {
1105 mutex_unlock(&node->mutex);
1106 return 0;
1107 }
1108
1109 ret = __btrfs_update_delayed_inode(trans, root, path, node);
1110 mutex_unlock(&node->mutex);
1111 return ret;
1112}
1113
1114static inline int
1115__btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1116 struct btrfs_path *path,
1117 struct btrfs_delayed_node *node)
1118{
1119 int ret;
1120
1121 ret = btrfs_insert_delayed_items(trans, path, node->root, node);
1122 if (ret)
1123 return ret;
1124
1125 ret = btrfs_delete_delayed_items(trans, path, node->root, node);
1126 if (ret)
1127 return ret;
1128
1129 ret = btrfs_update_delayed_inode(trans, node->root, path, node);
1130 return ret;
1131}
1132
1133/*
1134 * Called when committing the transaction.
1135 * Returns 0 on success.
1136 * Returns < 0 on error and returns with an aborted transaction with any
1137 * outstanding delayed items cleaned up.
1138 */
1139static int __btrfs_run_delayed_items(struct btrfs_trans_handle *trans,
1140 struct btrfs_fs_info *fs_info, int nr)
1141{
1142 struct btrfs_delayed_root *delayed_root;
1143 struct btrfs_delayed_node *curr_node, *prev_node;
1144 struct btrfs_path *path;
1145 struct btrfs_block_rsv *block_rsv;
1146 int ret = 0;
1147 bool count = (nr > 0);
1148
1149 if (trans->aborted)
1150 return -EIO;
1151
1152 path = btrfs_alloc_path();
1153 if (!path)
1154 return -ENOMEM;
1155 path->leave_spinning = 1;
1156
1157 block_rsv = trans->block_rsv;
1158 trans->block_rsv = &fs_info->delayed_block_rsv;
1159
1160 delayed_root = fs_info->delayed_root;
1161
1162 curr_node = btrfs_first_delayed_node(delayed_root);
1163 while (curr_node && (!count || (count && nr--))) {
1164 ret = __btrfs_commit_inode_delayed_items(trans, path,
1165 curr_node);
1166 if (ret) {
1167 btrfs_release_delayed_node(curr_node);
1168 curr_node = NULL;
1169 btrfs_abort_transaction(trans, ret);
1170 break;
1171 }
1172
1173 prev_node = curr_node;
1174 curr_node = btrfs_next_delayed_node(curr_node);
1175 btrfs_release_delayed_node(prev_node);
1176 }
1177
1178 if (curr_node)
1179 btrfs_release_delayed_node(curr_node);
1180 btrfs_free_path(path);
1181 trans->block_rsv = block_rsv;
1182
1183 return ret;
1184}
1185
1186int btrfs_run_delayed_items(struct btrfs_trans_handle *trans,
1187 struct btrfs_fs_info *fs_info)
1188{
1189 return __btrfs_run_delayed_items(trans, fs_info, -1);
1190}
1191
1192int btrfs_run_delayed_items_nr(struct btrfs_trans_handle *trans,
1193 struct btrfs_fs_info *fs_info, int nr)
1194{
1195 return __btrfs_run_delayed_items(trans, fs_info, nr);
1196}
1197
1198int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1199 struct inode *inode)
1200{
1201 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1202 struct btrfs_path *path;
1203 struct btrfs_block_rsv *block_rsv;
1204 int ret;
1205
1206 if (!delayed_node)
1207 return 0;
1208
1209 mutex_lock(&delayed_node->mutex);
1210 if (!delayed_node->count) {
1211 mutex_unlock(&delayed_node->mutex);
1212 btrfs_release_delayed_node(delayed_node);
1213 return 0;
1214 }
1215 mutex_unlock(&delayed_node->mutex);
1216
1217 path = btrfs_alloc_path();
1218 if (!path) {
1219 btrfs_release_delayed_node(delayed_node);
1220 return -ENOMEM;
1221 }
1222 path->leave_spinning = 1;
1223
1224 block_rsv = trans->block_rsv;
1225 trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv;
1226
1227 ret = __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1228
1229 btrfs_release_delayed_node(delayed_node);
1230 btrfs_free_path(path);
1231 trans->block_rsv = block_rsv;
1232
1233 return ret;
1234}
1235
1236int btrfs_commit_inode_delayed_inode(struct inode *inode)
1237{
1238 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1239 struct btrfs_trans_handle *trans;
1240 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1241 struct btrfs_path *path;
1242 struct btrfs_block_rsv *block_rsv;
1243 int ret;
1244
1245 if (!delayed_node)
1246 return 0;
1247
1248 mutex_lock(&delayed_node->mutex);
1249 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1250 mutex_unlock(&delayed_node->mutex);
1251 btrfs_release_delayed_node(delayed_node);
1252 return 0;
1253 }
1254 mutex_unlock(&delayed_node->mutex);
1255
1256 trans = btrfs_join_transaction(delayed_node->root);
1257 if (IS_ERR(trans)) {
1258 ret = PTR_ERR(trans);
1259 goto out;
1260 }
1261
1262 path = btrfs_alloc_path();
1263 if (!path) {
1264 ret = -ENOMEM;
1265 goto trans_out;
1266 }
1267 path->leave_spinning = 1;
1268
1269 block_rsv = trans->block_rsv;
1270 trans->block_rsv = &fs_info->delayed_block_rsv;
1271
1272 mutex_lock(&delayed_node->mutex);
1273 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags))
1274 ret = __btrfs_update_delayed_inode(trans, delayed_node->root,
1275 path, delayed_node);
1276 else
1277 ret = 0;
1278 mutex_unlock(&delayed_node->mutex);
1279
1280 btrfs_free_path(path);
1281 trans->block_rsv = block_rsv;
1282trans_out:
1283 btrfs_end_transaction(trans);
1284 btrfs_btree_balance_dirty(fs_info);
1285out:
1286 btrfs_release_delayed_node(delayed_node);
1287
1288 return ret;
1289}
1290
1291void btrfs_remove_delayed_node(struct inode *inode)
1292{
1293 struct btrfs_delayed_node *delayed_node;
1294
1295 delayed_node = ACCESS_ONCE(BTRFS_I(inode)->delayed_node);
1296 if (!delayed_node)
1297 return;
1298
1299 BTRFS_I(inode)->delayed_node = NULL;
1300 btrfs_release_delayed_node(delayed_node);
1301}
1302
1303struct btrfs_async_delayed_work {
1304 struct btrfs_delayed_root *delayed_root;
1305 int nr;
1306 struct btrfs_work work;
1307};
1308
1309static void btrfs_async_run_delayed_root(struct btrfs_work *work)
1310{
1311 struct btrfs_async_delayed_work *async_work;
1312 struct btrfs_delayed_root *delayed_root;
1313 struct btrfs_trans_handle *trans;
1314 struct btrfs_path *path;
1315 struct btrfs_delayed_node *delayed_node = NULL;
1316 struct btrfs_root *root;
1317 struct btrfs_block_rsv *block_rsv;
1318 int total_done = 0;
1319
1320 async_work = container_of(work, struct btrfs_async_delayed_work, work);
1321 delayed_root = async_work->delayed_root;
1322
1323 path = btrfs_alloc_path();
1324 if (!path)
1325 goto out;
1326
1327again:
1328 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND / 2)
1329 goto free_path;
1330
1331 delayed_node = btrfs_first_prepared_delayed_node(delayed_root);
1332 if (!delayed_node)
1333 goto free_path;
1334
1335 path->leave_spinning = 1;
1336 root = delayed_node->root;
1337
1338 trans = btrfs_join_transaction(root);
1339 if (IS_ERR(trans))
1340 goto release_path;
1341
1342 block_rsv = trans->block_rsv;
1343 trans->block_rsv = &root->fs_info->delayed_block_rsv;
1344
1345 __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1346
1347 trans->block_rsv = block_rsv;
1348 btrfs_end_transaction(trans);
1349 btrfs_btree_balance_dirty_nodelay(root->fs_info);
1350
1351release_path:
1352 btrfs_release_path(path);
1353 total_done++;
1354
1355 btrfs_release_prepared_delayed_node(delayed_node);
1356 if ((async_work->nr == 0 && total_done < BTRFS_DELAYED_WRITEBACK) ||
1357 total_done < async_work->nr)
1358 goto again;
1359
1360free_path:
1361 btrfs_free_path(path);
1362out:
1363 wake_up(&delayed_root->wait);
1364 kfree(async_work);
1365}
1366
1367
1368static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root,
1369 struct btrfs_fs_info *fs_info, int nr)
1370{
1371 struct btrfs_async_delayed_work *async_work;
1372
1373 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND ||
1374 btrfs_workqueue_normal_congested(fs_info->delayed_workers))
1375 return 0;
1376
1377 async_work = kmalloc(sizeof(*async_work), GFP_NOFS);
1378 if (!async_work)
1379 return -ENOMEM;
1380
1381 async_work->delayed_root = delayed_root;
1382 btrfs_init_work(&async_work->work, btrfs_delayed_meta_helper,
1383 btrfs_async_run_delayed_root, NULL, NULL);
1384 async_work->nr = nr;
1385
1386 btrfs_queue_work(fs_info->delayed_workers, &async_work->work);
1387 return 0;
1388}
1389
1390void btrfs_assert_delayed_root_empty(struct btrfs_fs_info *fs_info)
1391{
1392 WARN_ON(btrfs_first_delayed_node(fs_info->delayed_root));
1393}
1394
1395static int could_end_wait(struct btrfs_delayed_root *delayed_root, int seq)
1396{
1397 int val = atomic_read(&delayed_root->items_seq);
1398
1399 if (val < seq || val >= seq + BTRFS_DELAYED_BATCH)
1400 return 1;
1401
1402 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1403 return 1;
1404
1405 return 0;
1406}
1407
1408void btrfs_balance_delayed_items(struct btrfs_fs_info *fs_info)
1409{
1410 struct btrfs_delayed_root *delayed_root = fs_info->delayed_root;
1411
1412 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1413 return;
1414
1415 if (atomic_read(&delayed_root->items) >= BTRFS_DELAYED_WRITEBACK) {
1416 int seq;
1417 int ret;
1418
1419 seq = atomic_read(&delayed_root->items_seq);
1420
1421 ret = btrfs_wq_run_delayed_node(delayed_root, fs_info, 0);
1422 if (ret)
1423 return;
1424
1425 wait_event_interruptible(delayed_root->wait,
1426 could_end_wait(delayed_root, seq));
1427 return;
1428 }
1429
1430 btrfs_wq_run_delayed_node(delayed_root, fs_info, BTRFS_DELAYED_BATCH);
1431}
1432
1433/* Will return 0 or -ENOMEM */
1434int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans,
1435 struct btrfs_fs_info *fs_info,
1436 const char *name, int name_len,
1437 struct inode *dir,
1438 struct btrfs_disk_key *disk_key, u8 type,
1439 u64 index)
1440{
1441 struct btrfs_delayed_node *delayed_node;
1442 struct btrfs_delayed_item *delayed_item;
1443 struct btrfs_dir_item *dir_item;
1444 int ret;
1445
1446 delayed_node = btrfs_get_or_create_delayed_node(dir);
1447 if (IS_ERR(delayed_node))
1448 return PTR_ERR(delayed_node);
1449
1450 delayed_item = btrfs_alloc_delayed_item(sizeof(*dir_item) + name_len);
1451 if (!delayed_item) {
1452 ret = -ENOMEM;
1453 goto release_node;
1454 }
1455
1456 delayed_item->key.objectid = btrfs_ino(dir);
1457 delayed_item->key.type = BTRFS_DIR_INDEX_KEY;
1458 delayed_item->key.offset = index;
1459
1460 dir_item = (struct btrfs_dir_item *)delayed_item->data;
1461 dir_item->location = *disk_key;
1462 btrfs_set_stack_dir_transid(dir_item, trans->transid);
1463 btrfs_set_stack_dir_data_len(dir_item, 0);
1464 btrfs_set_stack_dir_name_len(dir_item, name_len);
1465 btrfs_set_stack_dir_type(dir_item, type);
1466 memcpy((char *)(dir_item + 1), name, name_len);
1467
1468 ret = btrfs_delayed_item_reserve_metadata(trans, fs_info, delayed_item);
1469 /*
1470 * we have reserved enough space when we start a new transaction,
1471 * so reserving metadata failure is impossible
1472 */
1473 BUG_ON(ret);
1474
1475
1476 mutex_lock(&delayed_node->mutex);
1477 ret = __btrfs_add_delayed_insertion_item(delayed_node, delayed_item);
1478 if (unlikely(ret)) {
1479 btrfs_err(fs_info,
1480 "err add delayed dir index item(name: %.*s) into the insertion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
1481 name_len, name, delayed_node->root->objectid,
1482 delayed_node->inode_id, ret);
1483 BUG();
1484 }
1485 mutex_unlock(&delayed_node->mutex);
1486
1487release_node:
1488 btrfs_release_delayed_node(delayed_node);
1489 return ret;
1490}
1491
1492static int btrfs_delete_delayed_insertion_item(struct btrfs_fs_info *fs_info,
1493 struct btrfs_delayed_node *node,
1494 struct btrfs_key *key)
1495{
1496 struct btrfs_delayed_item *item;
1497
1498 mutex_lock(&node->mutex);
1499 item = __btrfs_lookup_delayed_insertion_item(node, key);
1500 if (!item) {
1501 mutex_unlock(&node->mutex);
1502 return 1;
1503 }
1504
1505 btrfs_delayed_item_release_metadata(fs_info, item);
1506 btrfs_release_delayed_item(item);
1507 mutex_unlock(&node->mutex);
1508 return 0;
1509}
1510
1511int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans,
1512 struct btrfs_fs_info *fs_info,
1513 struct inode *dir, u64 index)
1514{
1515 struct btrfs_delayed_node *node;
1516 struct btrfs_delayed_item *item;
1517 struct btrfs_key item_key;
1518 int ret;
1519
1520 node = btrfs_get_or_create_delayed_node(dir);
1521 if (IS_ERR(node))
1522 return PTR_ERR(node);
1523
1524 item_key.objectid = btrfs_ino(dir);
1525 item_key.type = BTRFS_DIR_INDEX_KEY;
1526 item_key.offset = index;
1527
1528 ret = btrfs_delete_delayed_insertion_item(fs_info, node, &item_key);
1529 if (!ret)
1530 goto end;
1531
1532 item = btrfs_alloc_delayed_item(0);
1533 if (!item) {
1534 ret = -ENOMEM;
1535 goto end;
1536 }
1537
1538 item->key = item_key;
1539
1540 ret = btrfs_delayed_item_reserve_metadata(trans, fs_info, item);
1541 /*
1542 * we have reserved enough space when we start a new transaction,
1543 * so reserving metadata failure is impossible.
1544 */
1545 BUG_ON(ret);
1546
1547 mutex_lock(&node->mutex);
1548 ret = __btrfs_add_delayed_deletion_item(node, item);
1549 if (unlikely(ret)) {
1550 btrfs_err(fs_info,
1551 "err add delayed dir index item(index: %llu) into the deletion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
1552 index, node->root->objectid, node->inode_id, ret);
1553 BUG();
1554 }
1555 mutex_unlock(&node->mutex);
1556end:
1557 btrfs_release_delayed_node(node);
1558 return ret;
1559}
1560
1561int btrfs_inode_delayed_dir_index_count(struct inode *inode)
1562{
1563 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1564
1565 if (!delayed_node)
1566 return -ENOENT;
1567
1568 /*
1569 * Since we have held i_mutex of this directory, it is impossible that
1570 * a new directory index is added into the delayed node and index_cnt
1571 * is updated now. So we needn't lock the delayed node.
1572 */
1573 if (!delayed_node->index_cnt) {
1574 btrfs_release_delayed_node(delayed_node);
1575 return -EINVAL;
1576 }
1577
1578 BTRFS_I(inode)->index_cnt = delayed_node->index_cnt;
1579 btrfs_release_delayed_node(delayed_node);
1580 return 0;
1581}
1582
1583bool btrfs_readdir_get_delayed_items(struct inode *inode,
1584 struct list_head *ins_list,
1585 struct list_head *del_list)
1586{
1587 struct btrfs_delayed_node *delayed_node;
1588 struct btrfs_delayed_item *item;
1589
1590 delayed_node = btrfs_get_delayed_node(inode);
1591 if (!delayed_node)
1592 return false;
1593
1594 /*
1595 * We can only do one readdir with delayed items at a time because of
1596 * item->readdir_list.
1597 */
1598 inode_unlock_shared(inode);
1599 inode_lock(inode);
1600
1601 mutex_lock(&delayed_node->mutex);
1602 item = __btrfs_first_delayed_insertion_item(delayed_node);
1603 while (item) {
1604 atomic_inc(&item->refs);
1605 list_add_tail(&item->readdir_list, ins_list);
1606 item = __btrfs_next_delayed_item(item);
1607 }
1608
1609 item = __btrfs_first_delayed_deletion_item(delayed_node);
1610 while (item) {
1611 atomic_inc(&item->refs);
1612 list_add_tail(&item->readdir_list, del_list);
1613 item = __btrfs_next_delayed_item(item);
1614 }
1615 mutex_unlock(&delayed_node->mutex);
1616 /*
1617 * This delayed node is still cached in the btrfs inode, so refs
1618 * must be > 1 now, and we needn't check it is going to be freed
1619 * or not.
1620 *
1621 * Besides that, this function is used to read dir, we do not
1622 * insert/delete delayed items in this period. So we also needn't
1623 * requeue or dequeue this delayed node.
1624 */
1625 atomic_dec(&delayed_node->refs);
1626
1627 return true;
1628}
1629
1630void btrfs_readdir_put_delayed_items(struct inode *inode,
1631 struct list_head *ins_list,
1632 struct list_head *del_list)
1633{
1634 struct btrfs_delayed_item *curr, *next;
1635
1636 list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1637 list_del(&curr->readdir_list);
1638 if (atomic_dec_and_test(&curr->refs))
1639 kfree(curr);
1640 }
1641
1642 list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1643 list_del(&curr->readdir_list);
1644 if (atomic_dec_and_test(&curr->refs))
1645 kfree(curr);
1646 }
1647
1648 /*
1649 * The VFS is going to do up_read(), so we need to downgrade back to a
1650 * read lock.
1651 */
1652 downgrade_write(&inode->i_rwsem);
1653}
1654
1655int btrfs_should_delete_dir_index(struct list_head *del_list,
1656 u64 index)
1657{
1658 struct btrfs_delayed_item *curr, *next;
1659 int ret;
1660
1661 if (list_empty(del_list))
1662 return 0;
1663
1664 list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1665 if (curr->key.offset > index)
1666 break;
1667
1668 list_del(&curr->readdir_list);
1669 ret = (curr->key.offset == index);
1670
1671 if (atomic_dec_and_test(&curr->refs))
1672 kfree(curr);
1673
1674 if (ret)
1675 return 1;
1676 else
1677 continue;
1678 }
1679 return 0;
1680}
1681
1682/*
1683 * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree
1684 *
1685 */
1686int btrfs_readdir_delayed_dir_index(struct dir_context *ctx,
1687 struct list_head *ins_list)
1688{
1689 struct btrfs_dir_item *di;
1690 struct btrfs_delayed_item *curr, *next;
1691 struct btrfs_key location;
1692 char *name;
1693 int name_len;
1694 int over = 0;
1695 unsigned char d_type;
1696
1697 if (list_empty(ins_list))
1698 return 0;
1699
1700 /*
1701 * Changing the data of the delayed item is impossible. So
1702 * we needn't lock them. And we have held i_mutex of the
1703 * directory, nobody can delete any directory indexes now.
1704 */
1705 list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1706 list_del(&curr->readdir_list);
1707
1708 if (curr->key.offset < ctx->pos) {
1709 if (atomic_dec_and_test(&curr->refs))
1710 kfree(curr);
1711 continue;
1712 }
1713
1714 ctx->pos = curr->key.offset;
1715
1716 di = (struct btrfs_dir_item *)curr->data;
1717 name = (char *)(di + 1);
1718 name_len = btrfs_stack_dir_name_len(di);
1719
1720 d_type = btrfs_filetype_table[di->type];
1721 btrfs_disk_key_to_cpu(&location, &di->location);
1722
1723 over = !dir_emit(ctx, name, name_len,
1724 location.objectid, d_type);
1725
1726 if (atomic_dec_and_test(&curr->refs))
1727 kfree(curr);
1728
1729 if (over)
1730 return 1;
1731 }
1732 return 0;
1733}
1734
1735static void fill_stack_inode_item(struct btrfs_trans_handle *trans,
1736 struct btrfs_inode_item *inode_item,
1737 struct inode *inode)
1738{
1739 btrfs_set_stack_inode_uid(inode_item, i_uid_read(inode));
1740 btrfs_set_stack_inode_gid(inode_item, i_gid_read(inode));
1741 btrfs_set_stack_inode_size(inode_item, BTRFS_I(inode)->disk_i_size);
1742 btrfs_set_stack_inode_mode(inode_item, inode->i_mode);
1743 btrfs_set_stack_inode_nlink(inode_item, inode->i_nlink);
1744 btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(inode));
1745 btrfs_set_stack_inode_generation(inode_item,
1746 BTRFS_I(inode)->generation);
1747 btrfs_set_stack_inode_sequence(inode_item, inode->i_version);
1748 btrfs_set_stack_inode_transid(inode_item, trans->transid);
1749 btrfs_set_stack_inode_rdev(inode_item, inode->i_rdev);
1750 btrfs_set_stack_inode_flags(inode_item, BTRFS_I(inode)->flags);
1751 btrfs_set_stack_inode_block_group(inode_item, 0);
1752
1753 btrfs_set_stack_timespec_sec(&inode_item->atime,
1754 inode->i_atime.tv_sec);
1755 btrfs_set_stack_timespec_nsec(&inode_item->atime,
1756 inode->i_atime.tv_nsec);
1757
1758 btrfs_set_stack_timespec_sec(&inode_item->mtime,
1759 inode->i_mtime.tv_sec);
1760 btrfs_set_stack_timespec_nsec(&inode_item->mtime,
1761 inode->i_mtime.tv_nsec);
1762
1763 btrfs_set_stack_timespec_sec(&inode_item->ctime,
1764 inode->i_ctime.tv_sec);
1765 btrfs_set_stack_timespec_nsec(&inode_item->ctime,
1766 inode->i_ctime.tv_nsec);
1767
1768 btrfs_set_stack_timespec_sec(&inode_item->otime,
1769 BTRFS_I(inode)->i_otime.tv_sec);
1770 btrfs_set_stack_timespec_nsec(&inode_item->otime,
1771 BTRFS_I(inode)->i_otime.tv_nsec);
1772}
1773
1774int btrfs_fill_inode(struct inode *inode, u32 *rdev)
1775{
1776 struct btrfs_delayed_node *delayed_node;
1777 struct btrfs_inode_item *inode_item;
1778
1779 delayed_node = btrfs_get_delayed_node(inode);
1780 if (!delayed_node)
1781 return -ENOENT;
1782
1783 mutex_lock(&delayed_node->mutex);
1784 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1785 mutex_unlock(&delayed_node->mutex);
1786 btrfs_release_delayed_node(delayed_node);
1787 return -ENOENT;
1788 }
1789
1790 inode_item = &delayed_node->inode_item;
1791
1792 i_uid_write(inode, btrfs_stack_inode_uid(inode_item));
1793 i_gid_write(inode, btrfs_stack_inode_gid(inode_item));
1794 btrfs_i_size_write(inode, btrfs_stack_inode_size(inode_item));
1795 inode->i_mode = btrfs_stack_inode_mode(inode_item);
1796 set_nlink(inode, btrfs_stack_inode_nlink(inode_item));
1797 inode_set_bytes(inode, btrfs_stack_inode_nbytes(inode_item));
1798 BTRFS_I(inode)->generation = btrfs_stack_inode_generation(inode_item);
1799 BTRFS_I(inode)->last_trans = btrfs_stack_inode_transid(inode_item);
1800
1801 inode->i_version = btrfs_stack_inode_sequence(inode_item);
1802 inode->i_rdev = 0;
1803 *rdev = btrfs_stack_inode_rdev(inode_item);
1804 BTRFS_I(inode)->flags = btrfs_stack_inode_flags(inode_item);
1805
1806 inode->i_atime.tv_sec = btrfs_stack_timespec_sec(&inode_item->atime);
1807 inode->i_atime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->atime);
1808
1809 inode->i_mtime.tv_sec = btrfs_stack_timespec_sec(&inode_item->mtime);
1810 inode->i_mtime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->mtime);
1811
1812 inode->i_ctime.tv_sec = btrfs_stack_timespec_sec(&inode_item->ctime);
1813 inode->i_ctime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->ctime);
1814
1815 BTRFS_I(inode)->i_otime.tv_sec =
1816 btrfs_stack_timespec_sec(&inode_item->otime);
1817 BTRFS_I(inode)->i_otime.tv_nsec =
1818 btrfs_stack_timespec_nsec(&inode_item->otime);
1819
1820 inode->i_generation = BTRFS_I(inode)->generation;
1821 BTRFS_I(inode)->index_cnt = (u64)-1;
1822
1823 mutex_unlock(&delayed_node->mutex);
1824 btrfs_release_delayed_node(delayed_node);
1825 return 0;
1826}
1827
1828int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans,
1829 struct btrfs_root *root, struct inode *inode)
1830{
1831 struct btrfs_delayed_node *delayed_node;
1832 int ret = 0;
1833
1834 delayed_node = btrfs_get_or_create_delayed_node(inode);
1835 if (IS_ERR(delayed_node))
1836 return PTR_ERR(delayed_node);
1837
1838 mutex_lock(&delayed_node->mutex);
1839 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1840 fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1841 goto release_node;
1842 }
1843
1844 ret = btrfs_delayed_inode_reserve_metadata(trans, root, inode,
1845 delayed_node);
1846 if (ret)
1847 goto release_node;
1848
1849 fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1850 set_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
1851 delayed_node->count++;
1852 atomic_inc(&root->fs_info->delayed_root->items);
1853release_node:
1854 mutex_unlock(&delayed_node->mutex);
1855 btrfs_release_delayed_node(delayed_node);
1856 return ret;
1857}
1858
1859int btrfs_delayed_delete_inode_ref(struct inode *inode)
1860{
1861 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1862 struct btrfs_delayed_node *delayed_node;
1863
1864 /*
1865 * we don't do delayed inode updates during log recovery because it
1866 * leads to enospc problems. This means we also can't do
1867 * delayed inode refs
1868 */
1869 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
1870 return -EAGAIN;
1871
1872 delayed_node = btrfs_get_or_create_delayed_node(inode);
1873 if (IS_ERR(delayed_node))
1874 return PTR_ERR(delayed_node);
1875
1876 /*
1877 * We don't reserve space for inode ref deletion is because:
1878 * - We ONLY do async inode ref deletion for the inode who has only
1879 * one link(i_nlink == 1), it means there is only one inode ref.
1880 * And in most case, the inode ref and the inode item are in the
1881 * same leaf, and we will deal with them at the same time.
1882 * Since we are sure we will reserve the space for the inode item,
1883 * it is unnecessary to reserve space for inode ref deletion.
1884 * - If the inode ref and the inode item are not in the same leaf,
1885 * We also needn't worry about enospc problem, because we reserve
1886 * much more space for the inode update than it needs.
1887 * - At the worst, we can steal some space from the global reservation.
1888 * It is very rare.
1889 */
1890 mutex_lock(&delayed_node->mutex);
1891 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1892 goto release_node;
1893
1894 set_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
1895 delayed_node->count++;
1896 atomic_inc(&fs_info->delayed_root->items);
1897release_node:
1898 mutex_unlock(&delayed_node->mutex);
1899 btrfs_release_delayed_node(delayed_node);
1900 return 0;
1901}
1902
1903static void __btrfs_kill_delayed_node(struct btrfs_delayed_node *delayed_node)
1904{
1905 struct btrfs_root *root = delayed_node->root;
1906 struct btrfs_fs_info *fs_info = root->fs_info;
1907 struct btrfs_delayed_item *curr_item, *prev_item;
1908
1909 mutex_lock(&delayed_node->mutex);
1910 curr_item = __btrfs_first_delayed_insertion_item(delayed_node);
1911 while (curr_item) {
1912 btrfs_delayed_item_release_metadata(fs_info, curr_item);
1913 prev_item = curr_item;
1914 curr_item = __btrfs_next_delayed_item(prev_item);
1915 btrfs_release_delayed_item(prev_item);
1916 }
1917
1918 curr_item = __btrfs_first_delayed_deletion_item(delayed_node);
1919 while (curr_item) {
1920 btrfs_delayed_item_release_metadata(fs_info, curr_item);
1921 prev_item = curr_item;
1922 curr_item = __btrfs_next_delayed_item(prev_item);
1923 btrfs_release_delayed_item(prev_item);
1924 }
1925
1926 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1927 btrfs_release_delayed_iref(delayed_node);
1928
1929 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1930 btrfs_delayed_inode_release_metadata(fs_info, delayed_node);
1931 btrfs_release_delayed_inode(delayed_node);
1932 }
1933 mutex_unlock(&delayed_node->mutex);
1934}
1935
1936void btrfs_kill_delayed_inode_items(struct inode *inode)
1937{
1938 struct btrfs_delayed_node *delayed_node;
1939
1940 delayed_node = btrfs_get_delayed_node(inode);
1941 if (!delayed_node)
1942 return;
1943
1944 __btrfs_kill_delayed_node(delayed_node);
1945 btrfs_release_delayed_node(delayed_node);
1946}
1947
1948void btrfs_kill_all_delayed_nodes(struct btrfs_root *root)
1949{
1950 u64 inode_id = 0;
1951 struct btrfs_delayed_node *delayed_nodes[8];
1952 int i, n;
1953
1954 while (1) {
1955 spin_lock(&root->inode_lock);
1956 n = radix_tree_gang_lookup(&root->delayed_nodes_tree,
1957 (void **)delayed_nodes, inode_id,
1958 ARRAY_SIZE(delayed_nodes));
1959 if (!n) {
1960 spin_unlock(&root->inode_lock);
1961 break;
1962 }
1963
1964 inode_id = delayed_nodes[n - 1]->inode_id + 1;
1965
1966 for (i = 0; i < n; i++)
1967 atomic_inc(&delayed_nodes[i]->refs);
1968 spin_unlock(&root->inode_lock);
1969
1970 for (i = 0; i < n; i++) {
1971 __btrfs_kill_delayed_node(delayed_nodes[i]);
1972 btrfs_release_delayed_node(delayed_nodes[i]);
1973 }
1974 }
1975}
1976
1977void btrfs_destroy_delayed_inodes(struct btrfs_fs_info *fs_info)
1978{
1979 struct btrfs_delayed_node *curr_node, *prev_node;
1980
1981 curr_node = btrfs_first_delayed_node(fs_info->delayed_root);
1982 while (curr_node) {
1983 __btrfs_kill_delayed_node(curr_node);
1984
1985 prev_node = curr_node;
1986 curr_node = btrfs_next_delayed_node(curr_node);
1987 btrfs_release_delayed_node(prev_node);
1988 }
1989}
1990