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 "misc.h"
10 #include "delayed-inode.h"
11 #include "disk-io.h"
12 #include "transaction.h"
13 #include "ctree.h"
14 #include "qgroup.h"
15 #include "locking.h"
16 #include "inode-item.h"
17
18 #define BTRFS_DELAYED_WRITEBACK 512
19 #define BTRFS_DELAYED_BACKGROUND 128
20 #define BTRFS_DELAYED_BATCH 16
21
22 static struct kmem_cache *delayed_node_cache;
23
btrfs_delayed_inode_init(void)24 int __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
btrfs_delayed_inode_exit(void)36 void __cold btrfs_delayed_inode_exit(void)
37 {
38 kmem_cache_destroy(delayed_node_cache);
39 }
40
btrfs_init_delayed_node(struct btrfs_delayed_node * delayed_node,struct btrfs_root * root,u64 inode_id)41 static 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
btrfs_get_delayed_node(struct btrfs_inode * btrfs_inode)55 static struct btrfs_delayed_node *btrfs_get_delayed_node(
56 struct btrfs_inode *btrfs_inode)
57 {
58 struct btrfs_root *root = btrfs_inode->root;
59 u64 ino = btrfs_ino(btrfs_inode);
60 struct btrfs_delayed_node *node;
61
62 node = READ_ONCE(btrfs_inode->delayed_node);
63 if (node) {
64 refcount_inc(&node->refs);
65 return node;
66 }
67
68 spin_lock(&root->inode_lock);
69 node = radix_tree_lookup(&root->delayed_nodes_tree, ino);
70
71 if (node) {
72 if (btrfs_inode->delayed_node) {
73 refcount_inc(&node->refs); /* can be accessed */
74 BUG_ON(btrfs_inode->delayed_node != node);
75 spin_unlock(&root->inode_lock);
76 return node;
77 }
78
79 /*
80 * It's possible that we're racing into the middle of removing
81 * this node from the radix tree. In this case, the refcount
82 * was zero and it should never go back to one. Just return
83 * NULL like it was never in the radix at all; our release
84 * function is in the process of removing it.
85 *
86 * Some implementations of refcount_inc refuse to bump the
87 * refcount once it has hit zero. If we don't do this dance
88 * here, refcount_inc() may decide to just WARN_ONCE() instead
89 * of actually bumping the refcount.
90 *
91 * If this node is properly in the radix, we want to bump the
92 * refcount twice, once for the inode and once for this get
93 * operation.
94 */
95 if (refcount_inc_not_zero(&node->refs)) {
96 refcount_inc(&node->refs);
97 btrfs_inode->delayed_node = node;
98 } else {
99 node = NULL;
100 }
101
102 spin_unlock(&root->inode_lock);
103 return node;
104 }
105 spin_unlock(&root->inode_lock);
106
107 return NULL;
108 }
109
110 /* Will return either the node or PTR_ERR(-ENOMEM) */
btrfs_get_or_create_delayed_node(struct btrfs_inode * btrfs_inode)111 static struct btrfs_delayed_node *btrfs_get_or_create_delayed_node(
112 struct btrfs_inode *btrfs_inode)
113 {
114 struct btrfs_delayed_node *node;
115 struct btrfs_root *root = btrfs_inode->root;
116 u64 ino = btrfs_ino(btrfs_inode);
117 int ret;
118
119 again:
120 node = btrfs_get_delayed_node(btrfs_inode);
121 if (node)
122 return node;
123
124 node = kmem_cache_zalloc(delayed_node_cache, GFP_NOFS);
125 if (!node)
126 return ERR_PTR(-ENOMEM);
127 btrfs_init_delayed_node(node, root, ino);
128
129 /* cached in the btrfs inode and can be accessed */
130 refcount_set(&node->refs, 2);
131
132 ret = radix_tree_preload(GFP_NOFS);
133 if (ret) {
134 kmem_cache_free(delayed_node_cache, node);
135 return ERR_PTR(ret);
136 }
137
138 spin_lock(&root->inode_lock);
139 ret = radix_tree_insert(&root->delayed_nodes_tree, ino, node);
140 if (ret == -EEXIST) {
141 spin_unlock(&root->inode_lock);
142 kmem_cache_free(delayed_node_cache, node);
143 radix_tree_preload_end();
144 goto again;
145 }
146 btrfs_inode->delayed_node = node;
147 spin_unlock(&root->inode_lock);
148 radix_tree_preload_end();
149
150 return node;
151 }
152
153 /*
154 * Call it when holding delayed_node->mutex
155 *
156 * If mod = 1, add this node into the prepared list.
157 */
btrfs_queue_delayed_node(struct btrfs_delayed_root * root,struct btrfs_delayed_node * node,int mod)158 static void btrfs_queue_delayed_node(struct btrfs_delayed_root *root,
159 struct btrfs_delayed_node *node,
160 int mod)
161 {
162 spin_lock(&root->lock);
163 if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
164 if (!list_empty(&node->p_list))
165 list_move_tail(&node->p_list, &root->prepare_list);
166 else if (mod)
167 list_add_tail(&node->p_list, &root->prepare_list);
168 } else {
169 list_add_tail(&node->n_list, &root->node_list);
170 list_add_tail(&node->p_list, &root->prepare_list);
171 refcount_inc(&node->refs); /* inserted into list */
172 root->nodes++;
173 set_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
174 }
175 spin_unlock(&root->lock);
176 }
177
178 /* Call it when holding delayed_node->mutex */
btrfs_dequeue_delayed_node(struct btrfs_delayed_root * root,struct btrfs_delayed_node * node)179 static void btrfs_dequeue_delayed_node(struct btrfs_delayed_root *root,
180 struct btrfs_delayed_node *node)
181 {
182 spin_lock(&root->lock);
183 if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
184 root->nodes--;
185 refcount_dec(&node->refs); /* not in the list */
186 list_del_init(&node->n_list);
187 if (!list_empty(&node->p_list))
188 list_del_init(&node->p_list);
189 clear_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
190 }
191 spin_unlock(&root->lock);
192 }
193
btrfs_first_delayed_node(struct btrfs_delayed_root * delayed_root)194 static struct btrfs_delayed_node *btrfs_first_delayed_node(
195 struct btrfs_delayed_root *delayed_root)
196 {
197 struct list_head *p;
198 struct btrfs_delayed_node *node = NULL;
199
200 spin_lock(&delayed_root->lock);
201 if (list_empty(&delayed_root->node_list))
202 goto out;
203
204 p = delayed_root->node_list.next;
205 node = list_entry(p, struct btrfs_delayed_node, n_list);
206 refcount_inc(&node->refs);
207 out:
208 spin_unlock(&delayed_root->lock);
209
210 return node;
211 }
212
btrfs_next_delayed_node(struct btrfs_delayed_node * node)213 static struct btrfs_delayed_node *btrfs_next_delayed_node(
214 struct btrfs_delayed_node *node)
215 {
216 struct btrfs_delayed_root *delayed_root;
217 struct list_head *p;
218 struct btrfs_delayed_node *next = NULL;
219
220 delayed_root = node->root->fs_info->delayed_root;
221 spin_lock(&delayed_root->lock);
222 if (!test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
223 /* not in the list */
224 if (list_empty(&delayed_root->node_list))
225 goto out;
226 p = delayed_root->node_list.next;
227 } else if (list_is_last(&node->n_list, &delayed_root->node_list))
228 goto out;
229 else
230 p = node->n_list.next;
231
232 next = list_entry(p, struct btrfs_delayed_node, n_list);
233 refcount_inc(&next->refs);
234 out:
235 spin_unlock(&delayed_root->lock);
236
237 return next;
238 }
239
__btrfs_release_delayed_node(struct btrfs_delayed_node * delayed_node,int mod)240 static void __btrfs_release_delayed_node(
241 struct btrfs_delayed_node *delayed_node,
242 int mod)
243 {
244 struct btrfs_delayed_root *delayed_root;
245
246 if (!delayed_node)
247 return;
248
249 delayed_root = delayed_node->root->fs_info->delayed_root;
250
251 mutex_lock(&delayed_node->mutex);
252 if (delayed_node->count)
253 btrfs_queue_delayed_node(delayed_root, delayed_node, mod);
254 else
255 btrfs_dequeue_delayed_node(delayed_root, delayed_node);
256 mutex_unlock(&delayed_node->mutex);
257
258 if (refcount_dec_and_test(&delayed_node->refs)) {
259 struct btrfs_root *root = delayed_node->root;
260
261 spin_lock(&root->inode_lock);
262 /*
263 * Once our refcount goes to zero, nobody is allowed to bump it
264 * back up. We can delete it now.
265 */
266 ASSERT(refcount_read(&delayed_node->refs) == 0);
267 radix_tree_delete(&root->delayed_nodes_tree,
268 delayed_node->inode_id);
269 spin_unlock(&root->inode_lock);
270 kmem_cache_free(delayed_node_cache, delayed_node);
271 }
272 }
273
btrfs_release_delayed_node(struct btrfs_delayed_node * node)274 static inline void btrfs_release_delayed_node(struct btrfs_delayed_node *node)
275 {
276 __btrfs_release_delayed_node(node, 0);
277 }
278
btrfs_first_prepared_delayed_node(struct btrfs_delayed_root * delayed_root)279 static struct btrfs_delayed_node *btrfs_first_prepared_delayed_node(
280 struct btrfs_delayed_root *delayed_root)
281 {
282 struct list_head *p;
283 struct btrfs_delayed_node *node = NULL;
284
285 spin_lock(&delayed_root->lock);
286 if (list_empty(&delayed_root->prepare_list))
287 goto out;
288
289 p = delayed_root->prepare_list.next;
290 list_del_init(p);
291 node = list_entry(p, struct btrfs_delayed_node, p_list);
292 refcount_inc(&node->refs);
293 out:
294 spin_unlock(&delayed_root->lock);
295
296 return node;
297 }
298
btrfs_release_prepared_delayed_node(struct btrfs_delayed_node * node)299 static inline void btrfs_release_prepared_delayed_node(
300 struct btrfs_delayed_node *node)
301 {
302 __btrfs_release_delayed_node(node, 1);
303 }
304
btrfs_alloc_delayed_item(u16 data_len,struct btrfs_delayed_node * node,enum btrfs_delayed_item_type type)305 static struct btrfs_delayed_item *btrfs_alloc_delayed_item(u16 data_len,
306 struct btrfs_delayed_node *node,
307 enum btrfs_delayed_item_type type)
308 {
309 struct btrfs_delayed_item *item;
310
311 item = kmalloc(sizeof(*item) + data_len, GFP_NOFS);
312 if (item) {
313 item->data_len = data_len;
314 item->type = type;
315 item->bytes_reserved = 0;
316 item->delayed_node = node;
317 RB_CLEAR_NODE(&item->rb_node);
318 INIT_LIST_HEAD(&item->log_list);
319 item->logged = false;
320 refcount_set(&item->refs, 1);
321 }
322 return item;
323 }
324
325 /*
326 * __btrfs_lookup_delayed_item - look up the delayed item by key
327 * @delayed_node: pointer to the delayed node
328 * @index: the dir index value to lookup (offset of a dir index key)
329 *
330 * Note: if we don't find the right item, we will return the prev item and
331 * the next item.
332 */
__btrfs_lookup_delayed_item(struct rb_root * root,u64 index)333 static struct btrfs_delayed_item *__btrfs_lookup_delayed_item(
334 struct rb_root *root,
335 u64 index)
336 {
337 struct rb_node *node = root->rb_node;
338 struct btrfs_delayed_item *delayed_item = NULL;
339
340 while (node) {
341 delayed_item = rb_entry(node, struct btrfs_delayed_item,
342 rb_node);
343 if (delayed_item->index < index)
344 node = node->rb_right;
345 else if (delayed_item->index > index)
346 node = node->rb_left;
347 else
348 return delayed_item;
349 }
350
351 return NULL;
352 }
353
__btrfs_add_delayed_item(struct btrfs_delayed_node * delayed_node,struct btrfs_delayed_item * ins)354 static int __btrfs_add_delayed_item(struct btrfs_delayed_node *delayed_node,
355 struct btrfs_delayed_item *ins)
356 {
357 struct rb_node **p, *node;
358 struct rb_node *parent_node = NULL;
359 struct rb_root_cached *root;
360 struct btrfs_delayed_item *item;
361 bool leftmost = true;
362
363 if (ins->type == BTRFS_DELAYED_INSERTION_ITEM)
364 root = &delayed_node->ins_root;
365 else
366 root = &delayed_node->del_root;
367
368 p = &root->rb_root.rb_node;
369 node = &ins->rb_node;
370
371 while (*p) {
372 parent_node = *p;
373 item = rb_entry(parent_node, struct btrfs_delayed_item,
374 rb_node);
375
376 if (item->index < ins->index) {
377 p = &(*p)->rb_right;
378 leftmost = false;
379 } else if (item->index > ins->index) {
380 p = &(*p)->rb_left;
381 } else {
382 return -EEXIST;
383 }
384 }
385
386 rb_link_node(node, parent_node, p);
387 rb_insert_color_cached(node, root, leftmost);
388
389 if (ins->type == BTRFS_DELAYED_INSERTION_ITEM &&
390 ins->index >= delayed_node->index_cnt)
391 delayed_node->index_cnt = ins->index + 1;
392
393 delayed_node->count++;
394 atomic_inc(&delayed_node->root->fs_info->delayed_root->items);
395 return 0;
396 }
397
finish_one_item(struct btrfs_delayed_root * delayed_root)398 static void finish_one_item(struct btrfs_delayed_root *delayed_root)
399 {
400 int seq = atomic_inc_return(&delayed_root->items_seq);
401
402 /* atomic_dec_return implies a barrier */
403 if ((atomic_dec_return(&delayed_root->items) <
404 BTRFS_DELAYED_BACKGROUND || seq % BTRFS_DELAYED_BATCH == 0))
405 cond_wake_up_nomb(&delayed_root->wait);
406 }
407
__btrfs_remove_delayed_item(struct btrfs_delayed_item * delayed_item)408 static void __btrfs_remove_delayed_item(struct btrfs_delayed_item *delayed_item)
409 {
410 struct rb_root_cached *root;
411 struct btrfs_delayed_root *delayed_root;
412
413 /* Not inserted, ignore it. */
414 if (RB_EMPTY_NODE(&delayed_item->rb_node))
415 return;
416
417 delayed_root = delayed_item->delayed_node->root->fs_info->delayed_root;
418
419 BUG_ON(!delayed_root);
420
421 if (delayed_item->type == BTRFS_DELAYED_INSERTION_ITEM)
422 root = &delayed_item->delayed_node->ins_root;
423 else
424 root = &delayed_item->delayed_node->del_root;
425
426 rb_erase_cached(&delayed_item->rb_node, root);
427 RB_CLEAR_NODE(&delayed_item->rb_node);
428 delayed_item->delayed_node->count--;
429
430 finish_one_item(delayed_root);
431 }
432
btrfs_release_delayed_item(struct btrfs_delayed_item * item)433 static void btrfs_release_delayed_item(struct btrfs_delayed_item *item)
434 {
435 if (item) {
436 __btrfs_remove_delayed_item(item);
437 if (refcount_dec_and_test(&item->refs))
438 kfree(item);
439 }
440 }
441
__btrfs_first_delayed_insertion_item(struct btrfs_delayed_node * delayed_node)442 static struct btrfs_delayed_item *__btrfs_first_delayed_insertion_item(
443 struct btrfs_delayed_node *delayed_node)
444 {
445 struct rb_node *p;
446 struct btrfs_delayed_item *item = NULL;
447
448 p = rb_first_cached(&delayed_node->ins_root);
449 if (p)
450 item = rb_entry(p, struct btrfs_delayed_item, rb_node);
451
452 return item;
453 }
454
__btrfs_first_delayed_deletion_item(struct btrfs_delayed_node * delayed_node)455 static struct btrfs_delayed_item *__btrfs_first_delayed_deletion_item(
456 struct btrfs_delayed_node *delayed_node)
457 {
458 struct rb_node *p;
459 struct btrfs_delayed_item *item = NULL;
460
461 p = rb_first_cached(&delayed_node->del_root);
462 if (p)
463 item = rb_entry(p, struct btrfs_delayed_item, rb_node);
464
465 return item;
466 }
467
__btrfs_next_delayed_item(struct btrfs_delayed_item * item)468 static struct btrfs_delayed_item *__btrfs_next_delayed_item(
469 struct btrfs_delayed_item *item)
470 {
471 struct rb_node *p;
472 struct btrfs_delayed_item *next = NULL;
473
474 p = rb_next(&item->rb_node);
475 if (p)
476 next = rb_entry(p, struct btrfs_delayed_item, rb_node);
477
478 return next;
479 }
480
btrfs_delayed_item_reserve_metadata(struct btrfs_trans_handle * trans,struct btrfs_delayed_item * item)481 static int btrfs_delayed_item_reserve_metadata(struct btrfs_trans_handle *trans,
482 struct btrfs_delayed_item *item)
483 {
484 struct btrfs_block_rsv *src_rsv;
485 struct btrfs_block_rsv *dst_rsv;
486 struct btrfs_fs_info *fs_info = trans->fs_info;
487 u64 num_bytes;
488 int ret;
489
490 if (!trans->bytes_reserved)
491 return 0;
492
493 src_rsv = trans->block_rsv;
494 dst_rsv = &fs_info->delayed_block_rsv;
495
496 num_bytes = btrfs_calc_insert_metadata_size(fs_info, 1);
497
498 /*
499 * Here we migrate space rsv from transaction rsv, since have already
500 * reserved space when starting a transaction. So no need to reserve
501 * qgroup space here.
502 */
503 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, true);
504 if (!ret) {
505 trace_btrfs_space_reservation(fs_info, "delayed_item",
506 item->delayed_node->inode_id,
507 num_bytes, 1);
508 /*
509 * For insertions we track reserved metadata space by accounting
510 * for the number of leaves that will be used, based on the delayed
511 * node's index_items_size field.
512 */
513 if (item->type == BTRFS_DELAYED_DELETION_ITEM)
514 item->bytes_reserved = num_bytes;
515 }
516
517 return ret;
518 }
519
btrfs_delayed_item_release_metadata(struct btrfs_root * root,struct btrfs_delayed_item * item)520 static void btrfs_delayed_item_release_metadata(struct btrfs_root *root,
521 struct btrfs_delayed_item *item)
522 {
523 struct btrfs_block_rsv *rsv;
524 struct btrfs_fs_info *fs_info = root->fs_info;
525
526 if (!item->bytes_reserved)
527 return;
528
529 rsv = &fs_info->delayed_block_rsv;
530 /*
531 * Check btrfs_delayed_item_reserve_metadata() to see why we don't need
532 * to release/reserve qgroup space.
533 */
534 trace_btrfs_space_reservation(fs_info, "delayed_item",
535 item->delayed_node->inode_id,
536 item->bytes_reserved, 0);
537 btrfs_block_rsv_release(fs_info, rsv, item->bytes_reserved, NULL);
538 }
539
btrfs_delayed_item_release_leaves(struct btrfs_delayed_node * node,unsigned int num_leaves)540 static void btrfs_delayed_item_release_leaves(struct btrfs_delayed_node *node,
541 unsigned int num_leaves)
542 {
543 struct btrfs_fs_info *fs_info = node->root->fs_info;
544 const u64 bytes = btrfs_calc_insert_metadata_size(fs_info, num_leaves);
545
546 /* There are no space reservations during log replay, bail out. */
547 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
548 return;
549
550 trace_btrfs_space_reservation(fs_info, "delayed_item", node->inode_id,
551 bytes, 0);
552 btrfs_block_rsv_release(fs_info, &fs_info->delayed_block_rsv, bytes, NULL);
553 }
554
btrfs_delayed_inode_reserve_metadata(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_delayed_node * node)555 static int btrfs_delayed_inode_reserve_metadata(
556 struct btrfs_trans_handle *trans,
557 struct btrfs_root *root,
558 struct btrfs_delayed_node *node)
559 {
560 struct btrfs_fs_info *fs_info = root->fs_info;
561 struct btrfs_block_rsv *src_rsv;
562 struct btrfs_block_rsv *dst_rsv;
563 u64 num_bytes;
564 int ret;
565
566 src_rsv = trans->block_rsv;
567 dst_rsv = &fs_info->delayed_block_rsv;
568
569 num_bytes = btrfs_calc_metadata_size(fs_info, 1);
570
571 /*
572 * btrfs_dirty_inode will update the inode under btrfs_join_transaction
573 * which doesn't reserve space for speed. This is a problem since we
574 * still need to reserve space for this update, so try to reserve the
575 * space.
576 *
577 * Now if src_rsv == delalloc_block_rsv we'll let it just steal since
578 * we always reserve enough to update the inode item.
579 */
580 if (!src_rsv || (!trans->bytes_reserved &&
581 src_rsv->type != BTRFS_BLOCK_RSV_DELALLOC)) {
582 ret = btrfs_qgroup_reserve_meta(root, num_bytes,
583 BTRFS_QGROUP_RSV_META_PREALLOC, true);
584 if (ret < 0)
585 return ret;
586 ret = btrfs_block_rsv_add(fs_info, dst_rsv, num_bytes,
587 BTRFS_RESERVE_NO_FLUSH);
588 /* NO_FLUSH could only fail with -ENOSPC */
589 ASSERT(ret == 0 || ret == -ENOSPC);
590 if (ret)
591 btrfs_qgroup_free_meta_prealloc(root, num_bytes);
592 } else {
593 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, true);
594 }
595
596 if (!ret) {
597 trace_btrfs_space_reservation(fs_info, "delayed_inode",
598 node->inode_id, num_bytes, 1);
599 node->bytes_reserved = num_bytes;
600 }
601
602 return ret;
603 }
604
btrfs_delayed_inode_release_metadata(struct btrfs_fs_info * fs_info,struct btrfs_delayed_node * node,bool qgroup_free)605 static void btrfs_delayed_inode_release_metadata(struct btrfs_fs_info *fs_info,
606 struct btrfs_delayed_node *node,
607 bool qgroup_free)
608 {
609 struct btrfs_block_rsv *rsv;
610
611 if (!node->bytes_reserved)
612 return;
613
614 rsv = &fs_info->delayed_block_rsv;
615 trace_btrfs_space_reservation(fs_info, "delayed_inode",
616 node->inode_id, node->bytes_reserved, 0);
617 btrfs_block_rsv_release(fs_info, rsv, node->bytes_reserved, NULL);
618 if (qgroup_free)
619 btrfs_qgroup_free_meta_prealloc(node->root,
620 node->bytes_reserved);
621 else
622 btrfs_qgroup_convert_reserved_meta(node->root,
623 node->bytes_reserved);
624 node->bytes_reserved = 0;
625 }
626
627 /*
628 * Insert a single delayed item or a batch of delayed items, as many as possible
629 * that fit in a leaf. The delayed items (dir index keys) are sorted by their key
630 * in the rbtree, and if there's a gap between two consecutive dir index items,
631 * then it means at some point we had delayed dir indexes to add but they got
632 * removed (by btrfs_delete_delayed_dir_index()) before we attempted to flush them
633 * into the subvolume tree. Dir index keys also have their offsets coming from a
634 * monotonically increasing counter, so we can't get new keys with an offset that
635 * fits within a gap between delayed dir index items.
636 */
btrfs_insert_delayed_item(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,struct btrfs_delayed_item * first_item)637 static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans,
638 struct btrfs_root *root,
639 struct btrfs_path *path,
640 struct btrfs_delayed_item *first_item)
641 {
642 struct btrfs_fs_info *fs_info = root->fs_info;
643 struct btrfs_delayed_node *node = first_item->delayed_node;
644 LIST_HEAD(item_list);
645 struct btrfs_delayed_item *curr;
646 struct btrfs_delayed_item *next;
647 const int max_size = BTRFS_LEAF_DATA_SIZE(fs_info);
648 struct btrfs_item_batch batch;
649 struct btrfs_key first_key;
650 const u32 first_data_size = first_item->data_len;
651 int total_size;
652 char *ins_data = NULL;
653 int ret;
654 bool continuous_keys_only = false;
655
656 lockdep_assert_held(&node->mutex);
657
658 /*
659 * During normal operation the delayed index offset is continuously
660 * increasing, so we can batch insert all items as there will not be any
661 * overlapping keys in the tree.
662 *
663 * The exception to this is log replay, where we may have interleaved
664 * offsets in the tree, so our batch needs to be continuous keys only in
665 * order to ensure we do not end up with out of order items in our leaf.
666 */
667 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
668 continuous_keys_only = true;
669
670 /*
671 * For delayed items to insert, we track reserved metadata bytes based
672 * on the number of leaves that we will use.
673 * See btrfs_insert_delayed_dir_index() and
674 * btrfs_delayed_item_reserve_metadata()).
675 */
676 ASSERT(first_item->bytes_reserved == 0);
677
678 list_add_tail(&first_item->tree_list, &item_list);
679 batch.total_data_size = first_data_size;
680 batch.nr = 1;
681 total_size = first_data_size + sizeof(struct btrfs_item);
682 curr = first_item;
683
684 while (true) {
685 int next_size;
686
687 next = __btrfs_next_delayed_item(curr);
688 if (!next)
689 break;
690
691 /*
692 * We cannot allow gaps in the key space if we're doing log
693 * replay.
694 */
695 if (continuous_keys_only && (next->index != curr->index + 1))
696 break;
697
698 ASSERT(next->bytes_reserved == 0);
699
700 next_size = next->data_len + sizeof(struct btrfs_item);
701 if (total_size + next_size > max_size)
702 break;
703
704 list_add_tail(&next->tree_list, &item_list);
705 batch.nr++;
706 total_size += next_size;
707 batch.total_data_size += next->data_len;
708 curr = next;
709 }
710
711 if (batch.nr == 1) {
712 first_key.objectid = node->inode_id;
713 first_key.type = BTRFS_DIR_INDEX_KEY;
714 first_key.offset = first_item->index;
715 batch.keys = &first_key;
716 batch.data_sizes = &first_data_size;
717 } else {
718 struct btrfs_key *ins_keys;
719 u32 *ins_sizes;
720 int i = 0;
721
722 ins_data = kmalloc(batch.nr * sizeof(u32) +
723 batch.nr * sizeof(struct btrfs_key), GFP_NOFS);
724 if (!ins_data) {
725 ret = -ENOMEM;
726 goto out;
727 }
728 ins_sizes = (u32 *)ins_data;
729 ins_keys = (struct btrfs_key *)(ins_data + batch.nr * sizeof(u32));
730 batch.keys = ins_keys;
731 batch.data_sizes = ins_sizes;
732 list_for_each_entry(curr, &item_list, tree_list) {
733 ins_keys[i].objectid = node->inode_id;
734 ins_keys[i].type = BTRFS_DIR_INDEX_KEY;
735 ins_keys[i].offset = curr->index;
736 ins_sizes[i] = curr->data_len;
737 i++;
738 }
739 }
740
741 ret = btrfs_insert_empty_items(trans, root, path, &batch);
742 if (ret)
743 goto out;
744
745 list_for_each_entry(curr, &item_list, tree_list) {
746 char *data_ptr;
747
748 data_ptr = btrfs_item_ptr(path->nodes[0], path->slots[0], char);
749 write_extent_buffer(path->nodes[0], &curr->data,
750 (unsigned long)data_ptr, curr->data_len);
751 path->slots[0]++;
752 }
753
754 /*
755 * Now release our path before releasing the delayed items and their
756 * metadata reservations, so that we don't block other tasks for more
757 * time than needed.
758 */
759 btrfs_release_path(path);
760
761 ASSERT(node->index_item_leaves > 0);
762
763 /*
764 * For normal operations we will batch an entire leaf's worth of delayed
765 * items, so if there are more items to process we can decrement
766 * index_item_leaves by 1 as we inserted 1 leaf's worth of items.
767 *
768 * However for log replay we may not have inserted an entire leaf's
769 * worth of items, we may have not had continuous items, so decrementing
770 * here would mess up the index_item_leaves accounting. For this case
771 * only clean up the accounting when there are no items left.
772 */
773 if (next && !continuous_keys_only) {
774 /*
775 * We inserted one batch of items into a leaf a there are more
776 * items to flush in a future batch, now release one unit of
777 * metadata space from the delayed block reserve, corresponding
778 * the leaf we just flushed to.
779 */
780 btrfs_delayed_item_release_leaves(node, 1);
781 node->index_item_leaves--;
782 } else if (!next) {
783 /*
784 * There are no more items to insert. We can have a number of
785 * reserved leaves > 1 here - this happens when many dir index
786 * items are added and then removed before they are flushed (file
787 * names with a very short life, never span a transaction). So
788 * release all remaining leaves.
789 */
790 btrfs_delayed_item_release_leaves(node, node->index_item_leaves);
791 node->index_item_leaves = 0;
792 }
793
794 list_for_each_entry_safe(curr, next, &item_list, tree_list) {
795 list_del(&curr->tree_list);
796 btrfs_release_delayed_item(curr);
797 }
798 out:
799 kfree(ins_data);
800 return ret;
801 }
802
btrfs_insert_delayed_items(struct btrfs_trans_handle * trans,struct btrfs_path * path,struct btrfs_root * root,struct btrfs_delayed_node * node)803 static int btrfs_insert_delayed_items(struct btrfs_trans_handle *trans,
804 struct btrfs_path *path,
805 struct btrfs_root *root,
806 struct btrfs_delayed_node *node)
807 {
808 int ret = 0;
809
810 while (ret == 0) {
811 struct btrfs_delayed_item *curr;
812
813 mutex_lock(&node->mutex);
814 curr = __btrfs_first_delayed_insertion_item(node);
815 if (!curr) {
816 mutex_unlock(&node->mutex);
817 break;
818 }
819 ret = btrfs_insert_delayed_item(trans, root, path, curr);
820 mutex_unlock(&node->mutex);
821 }
822
823 return ret;
824 }
825
btrfs_batch_delete_items(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,struct btrfs_delayed_item * item)826 static int btrfs_batch_delete_items(struct btrfs_trans_handle *trans,
827 struct btrfs_root *root,
828 struct btrfs_path *path,
829 struct btrfs_delayed_item *item)
830 {
831 const u64 ino = item->delayed_node->inode_id;
832 struct btrfs_fs_info *fs_info = root->fs_info;
833 struct btrfs_delayed_item *curr, *next;
834 struct extent_buffer *leaf = path->nodes[0];
835 LIST_HEAD(batch_list);
836 int nitems, slot, last_slot;
837 int ret;
838 u64 total_reserved_size = item->bytes_reserved;
839
840 ASSERT(leaf != NULL);
841
842 slot = path->slots[0];
843 last_slot = btrfs_header_nritems(leaf) - 1;
844 /*
845 * Our caller always gives us a path pointing to an existing item, so
846 * this can not happen.
847 */
848 ASSERT(slot <= last_slot);
849 if (WARN_ON(slot > last_slot))
850 return -ENOENT;
851
852 nitems = 1;
853 curr = item;
854 list_add_tail(&curr->tree_list, &batch_list);
855
856 /*
857 * Keep checking if the next delayed item matches the next item in the
858 * leaf - if so, we can add it to the batch of items to delete from the
859 * leaf.
860 */
861 while (slot < last_slot) {
862 struct btrfs_key key;
863
864 next = __btrfs_next_delayed_item(curr);
865 if (!next)
866 break;
867
868 slot++;
869 btrfs_item_key_to_cpu(leaf, &key, slot);
870 if (key.objectid != ino ||
871 key.type != BTRFS_DIR_INDEX_KEY ||
872 key.offset != next->index)
873 break;
874 nitems++;
875 curr = next;
876 list_add_tail(&curr->tree_list, &batch_list);
877 total_reserved_size += curr->bytes_reserved;
878 }
879
880 ret = btrfs_del_items(trans, root, path, path->slots[0], nitems);
881 if (ret)
882 return ret;
883
884 /* In case of BTRFS_FS_LOG_RECOVERING items won't have reserved space */
885 if (total_reserved_size > 0) {
886 /*
887 * Check btrfs_delayed_item_reserve_metadata() to see why we
888 * don't need to release/reserve qgroup space.
889 */
890 trace_btrfs_space_reservation(fs_info, "delayed_item", ino,
891 total_reserved_size, 0);
892 btrfs_block_rsv_release(fs_info, &fs_info->delayed_block_rsv,
893 total_reserved_size, NULL);
894 }
895
896 list_for_each_entry_safe(curr, next, &batch_list, tree_list) {
897 list_del(&curr->tree_list);
898 btrfs_release_delayed_item(curr);
899 }
900
901 return 0;
902 }
903
btrfs_delete_delayed_items(struct btrfs_trans_handle * trans,struct btrfs_path * path,struct btrfs_root * root,struct btrfs_delayed_node * node)904 static int btrfs_delete_delayed_items(struct btrfs_trans_handle *trans,
905 struct btrfs_path *path,
906 struct btrfs_root *root,
907 struct btrfs_delayed_node *node)
908 {
909 struct btrfs_key key;
910 int ret = 0;
911
912 key.objectid = node->inode_id;
913 key.type = BTRFS_DIR_INDEX_KEY;
914
915 while (ret == 0) {
916 struct btrfs_delayed_item *item;
917
918 mutex_lock(&node->mutex);
919 item = __btrfs_first_delayed_deletion_item(node);
920 if (!item) {
921 mutex_unlock(&node->mutex);
922 break;
923 }
924
925 key.offset = item->index;
926 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
927 if (ret > 0) {
928 /*
929 * There's no matching item in the leaf. This means we
930 * have already deleted this item in a past run of the
931 * delayed items. We ignore errors when running delayed
932 * items from an async context, through a work queue job
933 * running btrfs_async_run_delayed_root(), and don't
934 * release delayed items that failed to complete. This
935 * is because we will retry later, and at transaction
936 * commit time we always run delayed items and will
937 * then deal with errors if they fail to run again.
938 *
939 * So just release delayed items for which we can't find
940 * an item in the tree, and move to the next item.
941 */
942 btrfs_release_path(path);
943 btrfs_release_delayed_item(item);
944 ret = 0;
945 } else if (ret == 0) {
946 ret = btrfs_batch_delete_items(trans, root, path, item);
947 btrfs_release_path(path);
948 }
949
950 /*
951 * We unlock and relock on each iteration, this is to prevent
952 * blocking other tasks for too long while we are being run from
953 * the async context (work queue job). Those tasks are typically
954 * running system calls like creat/mkdir/rename/unlink/etc which
955 * need to add delayed items to this delayed node.
956 */
957 mutex_unlock(&node->mutex);
958 }
959
960 return ret;
961 }
962
btrfs_release_delayed_inode(struct btrfs_delayed_node * delayed_node)963 static void btrfs_release_delayed_inode(struct btrfs_delayed_node *delayed_node)
964 {
965 struct btrfs_delayed_root *delayed_root;
966
967 if (delayed_node &&
968 test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
969 BUG_ON(!delayed_node->root);
970 clear_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
971 delayed_node->count--;
972
973 delayed_root = delayed_node->root->fs_info->delayed_root;
974 finish_one_item(delayed_root);
975 }
976 }
977
btrfs_release_delayed_iref(struct btrfs_delayed_node * delayed_node)978 static void btrfs_release_delayed_iref(struct btrfs_delayed_node *delayed_node)
979 {
980
981 if (test_and_clear_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags)) {
982 struct btrfs_delayed_root *delayed_root;
983
984 ASSERT(delayed_node->root);
985 delayed_node->count--;
986
987 delayed_root = delayed_node->root->fs_info->delayed_root;
988 finish_one_item(delayed_root);
989 }
990 }
991
__btrfs_update_delayed_inode(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,struct btrfs_delayed_node * node)992 static int __btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
993 struct btrfs_root *root,
994 struct btrfs_path *path,
995 struct btrfs_delayed_node *node)
996 {
997 struct btrfs_fs_info *fs_info = root->fs_info;
998 struct btrfs_key key;
999 struct btrfs_inode_item *inode_item;
1000 struct extent_buffer *leaf;
1001 int mod;
1002 int ret;
1003
1004 key.objectid = node->inode_id;
1005 key.type = BTRFS_INODE_ITEM_KEY;
1006 key.offset = 0;
1007
1008 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1009 mod = -1;
1010 else
1011 mod = 1;
1012
1013 ret = btrfs_lookup_inode(trans, root, path, &key, mod);
1014 if (ret > 0)
1015 ret = -ENOENT;
1016 if (ret < 0)
1017 goto out;
1018
1019 leaf = path->nodes[0];
1020 inode_item = btrfs_item_ptr(leaf, path->slots[0],
1021 struct btrfs_inode_item);
1022 write_extent_buffer(leaf, &node->inode_item, (unsigned long)inode_item,
1023 sizeof(struct btrfs_inode_item));
1024 btrfs_mark_buffer_dirty(leaf);
1025
1026 if (!test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1027 goto out;
1028
1029 path->slots[0]++;
1030 if (path->slots[0] >= btrfs_header_nritems(leaf))
1031 goto search;
1032 again:
1033 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1034 if (key.objectid != node->inode_id)
1035 goto out;
1036
1037 if (key.type != BTRFS_INODE_REF_KEY &&
1038 key.type != BTRFS_INODE_EXTREF_KEY)
1039 goto out;
1040
1041 /*
1042 * Delayed iref deletion is for the inode who has only one link,
1043 * so there is only one iref. The case that several irefs are
1044 * in the same item doesn't exist.
1045 */
1046 btrfs_del_item(trans, root, path);
1047 out:
1048 btrfs_release_delayed_iref(node);
1049 btrfs_release_path(path);
1050 err_out:
1051 btrfs_delayed_inode_release_metadata(fs_info, node, (ret < 0));
1052 btrfs_release_delayed_inode(node);
1053
1054 /*
1055 * If we fail to update the delayed inode we need to abort the
1056 * transaction, because we could leave the inode with the improper
1057 * counts behind.
1058 */
1059 if (ret && ret != -ENOENT)
1060 btrfs_abort_transaction(trans, ret);
1061
1062 return ret;
1063
1064 search:
1065 btrfs_release_path(path);
1066
1067 key.type = BTRFS_INODE_EXTREF_KEY;
1068 key.offset = -1;
1069
1070 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1071 if (ret < 0)
1072 goto err_out;
1073 ASSERT(ret);
1074
1075 ret = 0;
1076 leaf = path->nodes[0];
1077 path->slots[0]--;
1078 goto again;
1079 }
1080
btrfs_update_delayed_inode(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,struct btrfs_delayed_node * node)1081 static inline int btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1082 struct btrfs_root *root,
1083 struct btrfs_path *path,
1084 struct btrfs_delayed_node *node)
1085 {
1086 int ret;
1087
1088 mutex_lock(&node->mutex);
1089 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &node->flags)) {
1090 mutex_unlock(&node->mutex);
1091 return 0;
1092 }
1093
1094 ret = __btrfs_update_delayed_inode(trans, root, path, node);
1095 mutex_unlock(&node->mutex);
1096 return ret;
1097 }
1098
1099 static inline int
__btrfs_commit_inode_delayed_items(struct btrfs_trans_handle * trans,struct btrfs_path * path,struct btrfs_delayed_node * node)1100 __btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1101 struct btrfs_path *path,
1102 struct btrfs_delayed_node *node)
1103 {
1104 int ret;
1105
1106 ret = btrfs_insert_delayed_items(trans, path, node->root, node);
1107 if (ret)
1108 return ret;
1109
1110 ret = btrfs_delete_delayed_items(trans, path, node->root, node);
1111 if (ret)
1112 return ret;
1113
1114 ret = btrfs_update_delayed_inode(trans, node->root, path, node);
1115 return ret;
1116 }
1117
1118 /*
1119 * Called when committing the transaction.
1120 * Returns 0 on success.
1121 * Returns < 0 on error and returns with an aborted transaction with any
1122 * outstanding delayed items cleaned up.
1123 */
__btrfs_run_delayed_items(struct btrfs_trans_handle * trans,int nr)1124 static int __btrfs_run_delayed_items(struct btrfs_trans_handle *trans, int nr)
1125 {
1126 struct btrfs_fs_info *fs_info = trans->fs_info;
1127 struct btrfs_delayed_root *delayed_root;
1128 struct btrfs_delayed_node *curr_node, *prev_node;
1129 struct btrfs_path *path;
1130 struct btrfs_block_rsv *block_rsv;
1131 int ret = 0;
1132 bool count = (nr > 0);
1133
1134 if (TRANS_ABORTED(trans))
1135 return -EIO;
1136
1137 path = btrfs_alloc_path();
1138 if (!path)
1139 return -ENOMEM;
1140
1141 block_rsv = trans->block_rsv;
1142 trans->block_rsv = &fs_info->delayed_block_rsv;
1143
1144 delayed_root = fs_info->delayed_root;
1145
1146 curr_node = btrfs_first_delayed_node(delayed_root);
1147 while (curr_node && (!count || nr--)) {
1148 ret = __btrfs_commit_inode_delayed_items(trans, path,
1149 curr_node);
1150 if (ret) {
1151 btrfs_release_delayed_node(curr_node);
1152 curr_node = NULL;
1153 btrfs_abort_transaction(trans, ret);
1154 break;
1155 }
1156
1157 prev_node = curr_node;
1158 curr_node = btrfs_next_delayed_node(curr_node);
1159 btrfs_release_delayed_node(prev_node);
1160 }
1161
1162 if (curr_node)
1163 btrfs_release_delayed_node(curr_node);
1164 btrfs_free_path(path);
1165 trans->block_rsv = block_rsv;
1166
1167 return ret;
1168 }
1169
btrfs_run_delayed_items(struct btrfs_trans_handle * trans)1170 int btrfs_run_delayed_items(struct btrfs_trans_handle *trans)
1171 {
1172 return __btrfs_run_delayed_items(trans, -1);
1173 }
1174
btrfs_run_delayed_items_nr(struct btrfs_trans_handle * trans,int nr)1175 int btrfs_run_delayed_items_nr(struct btrfs_trans_handle *trans, int nr)
1176 {
1177 return __btrfs_run_delayed_items(trans, nr);
1178 }
1179
btrfs_commit_inode_delayed_items(struct btrfs_trans_handle * trans,struct btrfs_inode * inode)1180 int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1181 struct btrfs_inode *inode)
1182 {
1183 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1184 struct btrfs_path *path;
1185 struct btrfs_block_rsv *block_rsv;
1186 int ret;
1187
1188 if (!delayed_node)
1189 return 0;
1190
1191 mutex_lock(&delayed_node->mutex);
1192 if (!delayed_node->count) {
1193 mutex_unlock(&delayed_node->mutex);
1194 btrfs_release_delayed_node(delayed_node);
1195 return 0;
1196 }
1197 mutex_unlock(&delayed_node->mutex);
1198
1199 path = btrfs_alloc_path();
1200 if (!path) {
1201 btrfs_release_delayed_node(delayed_node);
1202 return -ENOMEM;
1203 }
1204
1205 block_rsv = trans->block_rsv;
1206 trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv;
1207
1208 ret = __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1209
1210 btrfs_release_delayed_node(delayed_node);
1211 btrfs_free_path(path);
1212 trans->block_rsv = block_rsv;
1213
1214 return ret;
1215 }
1216
btrfs_commit_inode_delayed_inode(struct btrfs_inode * inode)1217 int btrfs_commit_inode_delayed_inode(struct btrfs_inode *inode)
1218 {
1219 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1220 struct btrfs_trans_handle *trans;
1221 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1222 struct btrfs_path *path;
1223 struct btrfs_block_rsv *block_rsv;
1224 int ret;
1225
1226 if (!delayed_node)
1227 return 0;
1228
1229 mutex_lock(&delayed_node->mutex);
1230 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1231 mutex_unlock(&delayed_node->mutex);
1232 btrfs_release_delayed_node(delayed_node);
1233 return 0;
1234 }
1235 mutex_unlock(&delayed_node->mutex);
1236
1237 trans = btrfs_join_transaction(delayed_node->root);
1238 if (IS_ERR(trans)) {
1239 ret = PTR_ERR(trans);
1240 goto out;
1241 }
1242
1243 path = btrfs_alloc_path();
1244 if (!path) {
1245 ret = -ENOMEM;
1246 goto trans_out;
1247 }
1248
1249 block_rsv = trans->block_rsv;
1250 trans->block_rsv = &fs_info->delayed_block_rsv;
1251
1252 mutex_lock(&delayed_node->mutex);
1253 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags))
1254 ret = __btrfs_update_delayed_inode(trans, delayed_node->root,
1255 path, delayed_node);
1256 else
1257 ret = 0;
1258 mutex_unlock(&delayed_node->mutex);
1259
1260 btrfs_free_path(path);
1261 trans->block_rsv = block_rsv;
1262 trans_out:
1263 btrfs_end_transaction(trans);
1264 btrfs_btree_balance_dirty(fs_info);
1265 out:
1266 btrfs_release_delayed_node(delayed_node);
1267
1268 return ret;
1269 }
1270
btrfs_remove_delayed_node(struct btrfs_inode * inode)1271 void btrfs_remove_delayed_node(struct btrfs_inode *inode)
1272 {
1273 struct btrfs_delayed_node *delayed_node;
1274
1275 delayed_node = READ_ONCE(inode->delayed_node);
1276 if (!delayed_node)
1277 return;
1278
1279 inode->delayed_node = NULL;
1280 btrfs_release_delayed_node(delayed_node);
1281 }
1282
1283 struct btrfs_async_delayed_work {
1284 struct btrfs_delayed_root *delayed_root;
1285 int nr;
1286 struct btrfs_work work;
1287 };
1288
btrfs_async_run_delayed_root(struct btrfs_work * work)1289 static void btrfs_async_run_delayed_root(struct btrfs_work *work)
1290 {
1291 struct btrfs_async_delayed_work *async_work;
1292 struct btrfs_delayed_root *delayed_root;
1293 struct btrfs_trans_handle *trans;
1294 struct btrfs_path *path;
1295 struct btrfs_delayed_node *delayed_node = NULL;
1296 struct btrfs_root *root;
1297 struct btrfs_block_rsv *block_rsv;
1298 int total_done = 0;
1299
1300 async_work = container_of(work, struct btrfs_async_delayed_work, work);
1301 delayed_root = async_work->delayed_root;
1302
1303 path = btrfs_alloc_path();
1304 if (!path)
1305 goto out;
1306
1307 do {
1308 if (atomic_read(&delayed_root->items) <
1309 BTRFS_DELAYED_BACKGROUND / 2)
1310 break;
1311
1312 delayed_node = btrfs_first_prepared_delayed_node(delayed_root);
1313 if (!delayed_node)
1314 break;
1315
1316 root = delayed_node->root;
1317
1318 trans = btrfs_join_transaction(root);
1319 if (IS_ERR(trans)) {
1320 btrfs_release_path(path);
1321 btrfs_release_prepared_delayed_node(delayed_node);
1322 total_done++;
1323 continue;
1324 }
1325
1326 block_rsv = trans->block_rsv;
1327 trans->block_rsv = &root->fs_info->delayed_block_rsv;
1328
1329 __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1330
1331 trans->block_rsv = block_rsv;
1332 btrfs_end_transaction(trans);
1333 btrfs_btree_balance_dirty_nodelay(root->fs_info);
1334
1335 btrfs_release_path(path);
1336 btrfs_release_prepared_delayed_node(delayed_node);
1337 total_done++;
1338
1339 } while ((async_work->nr == 0 && total_done < BTRFS_DELAYED_WRITEBACK)
1340 || total_done < async_work->nr);
1341
1342 btrfs_free_path(path);
1343 out:
1344 wake_up(&delayed_root->wait);
1345 kfree(async_work);
1346 }
1347
1348
btrfs_wq_run_delayed_node(struct btrfs_delayed_root * delayed_root,struct btrfs_fs_info * fs_info,int nr)1349 static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root,
1350 struct btrfs_fs_info *fs_info, int nr)
1351 {
1352 struct btrfs_async_delayed_work *async_work;
1353
1354 async_work = kmalloc(sizeof(*async_work), GFP_NOFS);
1355 if (!async_work)
1356 return -ENOMEM;
1357
1358 async_work->delayed_root = delayed_root;
1359 btrfs_init_work(&async_work->work, btrfs_async_run_delayed_root, NULL,
1360 NULL);
1361 async_work->nr = nr;
1362
1363 btrfs_queue_work(fs_info->delayed_workers, &async_work->work);
1364 return 0;
1365 }
1366
btrfs_assert_delayed_root_empty(struct btrfs_fs_info * fs_info)1367 void btrfs_assert_delayed_root_empty(struct btrfs_fs_info *fs_info)
1368 {
1369 WARN_ON(btrfs_first_delayed_node(fs_info->delayed_root));
1370 }
1371
could_end_wait(struct btrfs_delayed_root * delayed_root,int seq)1372 static int could_end_wait(struct btrfs_delayed_root *delayed_root, int seq)
1373 {
1374 int val = atomic_read(&delayed_root->items_seq);
1375
1376 if (val < seq || val >= seq + BTRFS_DELAYED_BATCH)
1377 return 1;
1378
1379 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1380 return 1;
1381
1382 return 0;
1383 }
1384
btrfs_balance_delayed_items(struct btrfs_fs_info * fs_info)1385 void btrfs_balance_delayed_items(struct btrfs_fs_info *fs_info)
1386 {
1387 struct btrfs_delayed_root *delayed_root = fs_info->delayed_root;
1388
1389 if ((atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND) ||
1390 btrfs_workqueue_normal_congested(fs_info->delayed_workers))
1391 return;
1392
1393 if (atomic_read(&delayed_root->items) >= BTRFS_DELAYED_WRITEBACK) {
1394 int seq;
1395 int ret;
1396
1397 seq = atomic_read(&delayed_root->items_seq);
1398
1399 ret = btrfs_wq_run_delayed_node(delayed_root, fs_info, 0);
1400 if (ret)
1401 return;
1402
1403 wait_event_interruptible(delayed_root->wait,
1404 could_end_wait(delayed_root, seq));
1405 return;
1406 }
1407
1408 btrfs_wq_run_delayed_node(delayed_root, fs_info, BTRFS_DELAYED_BATCH);
1409 }
1410
1411 /* Will return 0 or -ENOMEM */
btrfs_insert_delayed_dir_index(struct btrfs_trans_handle * trans,const char * name,int name_len,struct btrfs_inode * dir,struct btrfs_disk_key * disk_key,u8 type,u64 index)1412 int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans,
1413 const char *name, int name_len,
1414 struct btrfs_inode *dir,
1415 struct btrfs_disk_key *disk_key, u8 type,
1416 u64 index)
1417 {
1418 struct btrfs_fs_info *fs_info = trans->fs_info;
1419 const unsigned int leaf_data_size = BTRFS_LEAF_DATA_SIZE(fs_info);
1420 struct btrfs_delayed_node *delayed_node;
1421 struct btrfs_delayed_item *delayed_item;
1422 struct btrfs_dir_item *dir_item;
1423 bool reserve_leaf_space;
1424 u32 data_len;
1425 int ret;
1426
1427 delayed_node = btrfs_get_or_create_delayed_node(dir);
1428 if (IS_ERR(delayed_node))
1429 return PTR_ERR(delayed_node);
1430
1431 delayed_item = btrfs_alloc_delayed_item(sizeof(*dir_item) + name_len,
1432 delayed_node,
1433 BTRFS_DELAYED_INSERTION_ITEM);
1434 if (!delayed_item) {
1435 ret = -ENOMEM;
1436 goto release_node;
1437 }
1438
1439 delayed_item->index = index;
1440
1441 dir_item = (struct btrfs_dir_item *)delayed_item->data;
1442 dir_item->location = *disk_key;
1443 btrfs_set_stack_dir_transid(dir_item, trans->transid);
1444 btrfs_set_stack_dir_data_len(dir_item, 0);
1445 btrfs_set_stack_dir_name_len(dir_item, name_len);
1446 btrfs_set_stack_dir_type(dir_item, type);
1447 memcpy((char *)(dir_item + 1), name, name_len);
1448
1449 data_len = delayed_item->data_len + sizeof(struct btrfs_item);
1450
1451 mutex_lock(&delayed_node->mutex);
1452
1453 if (delayed_node->index_item_leaves == 0 ||
1454 delayed_node->curr_index_batch_size + data_len > leaf_data_size) {
1455 delayed_node->curr_index_batch_size = data_len;
1456 reserve_leaf_space = true;
1457 } else {
1458 delayed_node->curr_index_batch_size += data_len;
1459 reserve_leaf_space = false;
1460 }
1461
1462 if (reserve_leaf_space) {
1463 ret = btrfs_delayed_item_reserve_metadata(trans, delayed_item);
1464 /*
1465 * Space was reserved for a dir index item insertion when we
1466 * started the transaction, so getting a failure here should be
1467 * impossible.
1468 */
1469 if (WARN_ON(ret)) {
1470 mutex_unlock(&delayed_node->mutex);
1471 btrfs_release_delayed_item(delayed_item);
1472 goto release_node;
1473 }
1474
1475 delayed_node->index_item_leaves++;
1476 } else if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
1477 const u64 bytes = btrfs_calc_insert_metadata_size(fs_info, 1);
1478
1479 /*
1480 * Adding the new dir index item does not require touching another
1481 * leaf, so we can release 1 unit of metadata that was previously
1482 * reserved when starting the transaction. This applies only to
1483 * the case where we had a transaction start and excludes the
1484 * transaction join case (when replaying log trees).
1485 */
1486 trace_btrfs_space_reservation(fs_info, "transaction",
1487 trans->transid, bytes, 0);
1488 btrfs_block_rsv_release(fs_info, trans->block_rsv, bytes, NULL);
1489 ASSERT(trans->bytes_reserved >= bytes);
1490 trans->bytes_reserved -= bytes;
1491 }
1492
1493 ret = __btrfs_add_delayed_item(delayed_node, delayed_item);
1494 if (unlikely(ret)) {
1495 btrfs_err(trans->fs_info,
1496 "err add delayed dir index item(name: %.*s) into the insertion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
1497 name_len, name, delayed_node->root->root_key.objectid,
1498 delayed_node->inode_id, ret);
1499 BUG();
1500 }
1501 mutex_unlock(&delayed_node->mutex);
1502
1503 release_node:
1504 btrfs_release_delayed_node(delayed_node);
1505 return ret;
1506 }
1507
btrfs_delete_delayed_insertion_item(struct btrfs_fs_info * fs_info,struct btrfs_delayed_node * node,u64 index)1508 static int btrfs_delete_delayed_insertion_item(struct btrfs_fs_info *fs_info,
1509 struct btrfs_delayed_node *node,
1510 u64 index)
1511 {
1512 struct btrfs_delayed_item *item;
1513
1514 mutex_lock(&node->mutex);
1515 item = __btrfs_lookup_delayed_item(&node->ins_root.rb_root, index);
1516 if (!item) {
1517 mutex_unlock(&node->mutex);
1518 return 1;
1519 }
1520
1521 /*
1522 * For delayed items to insert, we track reserved metadata bytes based
1523 * on the number of leaves that we will use.
1524 * See btrfs_insert_delayed_dir_index() and
1525 * btrfs_delayed_item_reserve_metadata()).
1526 */
1527 ASSERT(item->bytes_reserved == 0);
1528 ASSERT(node->index_item_leaves > 0);
1529
1530 /*
1531 * If there's only one leaf reserved, we can decrement this item from the
1532 * current batch, otherwise we can not because we don't know which leaf
1533 * it belongs to. With the current limit on delayed items, we rarely
1534 * accumulate enough dir index items to fill more than one leaf (even
1535 * when using a leaf size of 4K).
1536 */
1537 if (node->index_item_leaves == 1) {
1538 const u32 data_len = item->data_len + sizeof(struct btrfs_item);
1539
1540 ASSERT(node->curr_index_batch_size >= data_len);
1541 node->curr_index_batch_size -= data_len;
1542 }
1543
1544 btrfs_release_delayed_item(item);
1545
1546 /* If we now have no more dir index items, we can release all leaves. */
1547 if (RB_EMPTY_ROOT(&node->ins_root.rb_root)) {
1548 btrfs_delayed_item_release_leaves(node, node->index_item_leaves);
1549 node->index_item_leaves = 0;
1550 }
1551
1552 mutex_unlock(&node->mutex);
1553 return 0;
1554 }
1555
btrfs_delete_delayed_dir_index(struct btrfs_trans_handle * trans,struct btrfs_inode * dir,u64 index)1556 int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans,
1557 struct btrfs_inode *dir, u64 index)
1558 {
1559 struct btrfs_delayed_node *node;
1560 struct btrfs_delayed_item *item;
1561 int ret;
1562
1563 node = btrfs_get_or_create_delayed_node(dir);
1564 if (IS_ERR(node))
1565 return PTR_ERR(node);
1566
1567 ret = btrfs_delete_delayed_insertion_item(trans->fs_info, node, index);
1568 if (!ret)
1569 goto end;
1570
1571 item = btrfs_alloc_delayed_item(0, node, BTRFS_DELAYED_DELETION_ITEM);
1572 if (!item) {
1573 ret = -ENOMEM;
1574 goto end;
1575 }
1576
1577 item->index = index;
1578
1579 ret = btrfs_delayed_item_reserve_metadata(trans, item);
1580 /*
1581 * we have reserved enough space when we start a new transaction,
1582 * so reserving metadata failure is impossible.
1583 */
1584 if (ret < 0) {
1585 btrfs_err(trans->fs_info,
1586 "metadata reservation failed for delayed dir item deltiona, should have been reserved");
1587 btrfs_release_delayed_item(item);
1588 goto end;
1589 }
1590
1591 mutex_lock(&node->mutex);
1592 ret = __btrfs_add_delayed_item(node, item);
1593 if (unlikely(ret)) {
1594 btrfs_err(trans->fs_info,
1595 "err add delayed dir index item(index: %llu) into the deletion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
1596 index, node->root->root_key.objectid,
1597 node->inode_id, ret);
1598 btrfs_delayed_item_release_metadata(dir->root, item);
1599 btrfs_release_delayed_item(item);
1600 }
1601 mutex_unlock(&node->mutex);
1602 end:
1603 btrfs_release_delayed_node(node);
1604 return ret;
1605 }
1606
btrfs_inode_delayed_dir_index_count(struct btrfs_inode * inode)1607 int btrfs_inode_delayed_dir_index_count(struct btrfs_inode *inode)
1608 {
1609 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1610
1611 if (!delayed_node)
1612 return -ENOENT;
1613
1614 /*
1615 * Since we have held i_mutex of this directory, it is impossible that
1616 * a new directory index is added into the delayed node and index_cnt
1617 * is updated now. So we needn't lock the delayed node.
1618 */
1619 if (!delayed_node->index_cnt) {
1620 btrfs_release_delayed_node(delayed_node);
1621 return -EINVAL;
1622 }
1623
1624 inode->index_cnt = delayed_node->index_cnt;
1625 btrfs_release_delayed_node(delayed_node);
1626 return 0;
1627 }
1628
btrfs_readdir_get_delayed_items(struct inode * inode,struct list_head * ins_list,struct list_head * del_list)1629 bool btrfs_readdir_get_delayed_items(struct inode *inode,
1630 struct list_head *ins_list,
1631 struct list_head *del_list)
1632 {
1633 struct btrfs_delayed_node *delayed_node;
1634 struct btrfs_delayed_item *item;
1635
1636 delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
1637 if (!delayed_node)
1638 return false;
1639
1640 /*
1641 * We can only do one readdir with delayed items at a time because of
1642 * item->readdir_list.
1643 */
1644 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
1645 btrfs_inode_lock(inode, 0);
1646
1647 mutex_lock(&delayed_node->mutex);
1648 item = __btrfs_first_delayed_insertion_item(delayed_node);
1649 while (item) {
1650 refcount_inc(&item->refs);
1651 list_add_tail(&item->readdir_list, ins_list);
1652 item = __btrfs_next_delayed_item(item);
1653 }
1654
1655 item = __btrfs_first_delayed_deletion_item(delayed_node);
1656 while (item) {
1657 refcount_inc(&item->refs);
1658 list_add_tail(&item->readdir_list, del_list);
1659 item = __btrfs_next_delayed_item(item);
1660 }
1661 mutex_unlock(&delayed_node->mutex);
1662 /*
1663 * This delayed node is still cached in the btrfs inode, so refs
1664 * must be > 1 now, and we needn't check it is going to be freed
1665 * or not.
1666 *
1667 * Besides that, this function is used to read dir, we do not
1668 * insert/delete delayed items in this period. So we also needn't
1669 * requeue or dequeue this delayed node.
1670 */
1671 refcount_dec(&delayed_node->refs);
1672
1673 return true;
1674 }
1675
btrfs_readdir_put_delayed_items(struct inode * inode,struct list_head * ins_list,struct list_head * del_list)1676 void btrfs_readdir_put_delayed_items(struct inode *inode,
1677 struct list_head *ins_list,
1678 struct list_head *del_list)
1679 {
1680 struct btrfs_delayed_item *curr, *next;
1681
1682 list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1683 list_del(&curr->readdir_list);
1684 if (refcount_dec_and_test(&curr->refs))
1685 kfree(curr);
1686 }
1687
1688 list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1689 list_del(&curr->readdir_list);
1690 if (refcount_dec_and_test(&curr->refs))
1691 kfree(curr);
1692 }
1693
1694 /*
1695 * The VFS is going to do up_read(), so we need to downgrade back to a
1696 * read lock.
1697 */
1698 downgrade_write(&inode->i_rwsem);
1699 }
1700
btrfs_should_delete_dir_index(struct list_head * del_list,u64 index)1701 int btrfs_should_delete_dir_index(struct list_head *del_list,
1702 u64 index)
1703 {
1704 struct btrfs_delayed_item *curr;
1705 int ret = 0;
1706
1707 list_for_each_entry(curr, del_list, readdir_list) {
1708 if (curr->index > index)
1709 break;
1710 if (curr->index == index) {
1711 ret = 1;
1712 break;
1713 }
1714 }
1715 return ret;
1716 }
1717
1718 /*
1719 * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree
1720 *
1721 */
btrfs_readdir_delayed_dir_index(struct dir_context * ctx,struct list_head * ins_list)1722 int btrfs_readdir_delayed_dir_index(struct dir_context *ctx,
1723 struct list_head *ins_list)
1724 {
1725 struct btrfs_dir_item *di;
1726 struct btrfs_delayed_item *curr, *next;
1727 struct btrfs_key location;
1728 char *name;
1729 int name_len;
1730 int over = 0;
1731 unsigned char d_type;
1732
1733 if (list_empty(ins_list))
1734 return 0;
1735
1736 /*
1737 * Changing the data of the delayed item is impossible. So
1738 * we needn't lock them. And we have held i_mutex of the
1739 * directory, nobody can delete any directory indexes now.
1740 */
1741 list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1742 list_del(&curr->readdir_list);
1743
1744 if (curr->index < ctx->pos) {
1745 if (refcount_dec_and_test(&curr->refs))
1746 kfree(curr);
1747 continue;
1748 }
1749
1750 ctx->pos = curr->index;
1751
1752 di = (struct btrfs_dir_item *)curr->data;
1753 name = (char *)(di + 1);
1754 name_len = btrfs_stack_dir_name_len(di);
1755
1756 d_type = fs_ftype_to_dtype(di->type);
1757 btrfs_disk_key_to_cpu(&location, &di->location);
1758
1759 over = !dir_emit(ctx, name, name_len,
1760 location.objectid, d_type);
1761
1762 if (refcount_dec_and_test(&curr->refs))
1763 kfree(curr);
1764
1765 if (over)
1766 return 1;
1767 ctx->pos++;
1768 }
1769 return 0;
1770 }
1771
fill_stack_inode_item(struct btrfs_trans_handle * trans,struct btrfs_inode_item * inode_item,struct inode * inode)1772 static void fill_stack_inode_item(struct btrfs_trans_handle *trans,
1773 struct btrfs_inode_item *inode_item,
1774 struct inode *inode)
1775 {
1776 u64 flags;
1777
1778 btrfs_set_stack_inode_uid(inode_item, i_uid_read(inode));
1779 btrfs_set_stack_inode_gid(inode_item, i_gid_read(inode));
1780 btrfs_set_stack_inode_size(inode_item, BTRFS_I(inode)->disk_i_size);
1781 btrfs_set_stack_inode_mode(inode_item, inode->i_mode);
1782 btrfs_set_stack_inode_nlink(inode_item, inode->i_nlink);
1783 btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(inode));
1784 btrfs_set_stack_inode_generation(inode_item,
1785 BTRFS_I(inode)->generation);
1786 btrfs_set_stack_inode_sequence(inode_item,
1787 inode_peek_iversion(inode));
1788 btrfs_set_stack_inode_transid(inode_item, trans->transid);
1789 btrfs_set_stack_inode_rdev(inode_item, inode->i_rdev);
1790 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
1791 BTRFS_I(inode)->ro_flags);
1792 btrfs_set_stack_inode_flags(inode_item, flags);
1793 btrfs_set_stack_inode_block_group(inode_item, 0);
1794
1795 btrfs_set_stack_timespec_sec(&inode_item->atime,
1796 inode->i_atime.tv_sec);
1797 btrfs_set_stack_timespec_nsec(&inode_item->atime,
1798 inode->i_atime.tv_nsec);
1799
1800 btrfs_set_stack_timespec_sec(&inode_item->mtime,
1801 inode->i_mtime.tv_sec);
1802 btrfs_set_stack_timespec_nsec(&inode_item->mtime,
1803 inode->i_mtime.tv_nsec);
1804
1805 btrfs_set_stack_timespec_sec(&inode_item->ctime,
1806 inode->i_ctime.tv_sec);
1807 btrfs_set_stack_timespec_nsec(&inode_item->ctime,
1808 inode->i_ctime.tv_nsec);
1809
1810 btrfs_set_stack_timespec_sec(&inode_item->otime,
1811 BTRFS_I(inode)->i_otime.tv_sec);
1812 btrfs_set_stack_timespec_nsec(&inode_item->otime,
1813 BTRFS_I(inode)->i_otime.tv_nsec);
1814 }
1815
btrfs_fill_inode(struct inode * inode,u32 * rdev)1816 int btrfs_fill_inode(struct inode *inode, u32 *rdev)
1817 {
1818 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1819 struct btrfs_delayed_node *delayed_node;
1820 struct btrfs_inode_item *inode_item;
1821
1822 delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
1823 if (!delayed_node)
1824 return -ENOENT;
1825
1826 mutex_lock(&delayed_node->mutex);
1827 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1828 mutex_unlock(&delayed_node->mutex);
1829 btrfs_release_delayed_node(delayed_node);
1830 return -ENOENT;
1831 }
1832
1833 inode_item = &delayed_node->inode_item;
1834
1835 i_uid_write(inode, btrfs_stack_inode_uid(inode_item));
1836 i_gid_write(inode, btrfs_stack_inode_gid(inode_item));
1837 btrfs_i_size_write(BTRFS_I(inode), btrfs_stack_inode_size(inode_item));
1838 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
1839 round_up(i_size_read(inode), fs_info->sectorsize));
1840 inode->i_mode = btrfs_stack_inode_mode(inode_item);
1841 set_nlink(inode, btrfs_stack_inode_nlink(inode_item));
1842 inode_set_bytes(inode, btrfs_stack_inode_nbytes(inode_item));
1843 BTRFS_I(inode)->generation = btrfs_stack_inode_generation(inode_item);
1844 BTRFS_I(inode)->last_trans = btrfs_stack_inode_transid(inode_item);
1845
1846 inode_set_iversion_queried(inode,
1847 btrfs_stack_inode_sequence(inode_item));
1848 inode->i_rdev = 0;
1849 *rdev = btrfs_stack_inode_rdev(inode_item);
1850 btrfs_inode_split_flags(btrfs_stack_inode_flags(inode_item),
1851 &BTRFS_I(inode)->flags, &BTRFS_I(inode)->ro_flags);
1852
1853 inode->i_atime.tv_sec = btrfs_stack_timespec_sec(&inode_item->atime);
1854 inode->i_atime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->atime);
1855
1856 inode->i_mtime.tv_sec = btrfs_stack_timespec_sec(&inode_item->mtime);
1857 inode->i_mtime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->mtime);
1858
1859 inode->i_ctime.tv_sec = btrfs_stack_timespec_sec(&inode_item->ctime);
1860 inode->i_ctime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->ctime);
1861
1862 BTRFS_I(inode)->i_otime.tv_sec =
1863 btrfs_stack_timespec_sec(&inode_item->otime);
1864 BTRFS_I(inode)->i_otime.tv_nsec =
1865 btrfs_stack_timespec_nsec(&inode_item->otime);
1866
1867 inode->i_generation = BTRFS_I(inode)->generation;
1868 BTRFS_I(inode)->index_cnt = (u64)-1;
1869
1870 mutex_unlock(&delayed_node->mutex);
1871 btrfs_release_delayed_node(delayed_node);
1872 return 0;
1873 }
1874
btrfs_delayed_update_inode(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_inode * inode)1875 int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans,
1876 struct btrfs_root *root,
1877 struct btrfs_inode *inode)
1878 {
1879 struct btrfs_delayed_node *delayed_node;
1880 int ret = 0;
1881
1882 delayed_node = btrfs_get_or_create_delayed_node(inode);
1883 if (IS_ERR(delayed_node))
1884 return PTR_ERR(delayed_node);
1885
1886 mutex_lock(&delayed_node->mutex);
1887 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1888 fill_stack_inode_item(trans, &delayed_node->inode_item,
1889 &inode->vfs_inode);
1890 goto release_node;
1891 }
1892
1893 ret = btrfs_delayed_inode_reserve_metadata(trans, root, delayed_node);
1894 if (ret)
1895 goto release_node;
1896
1897 fill_stack_inode_item(trans, &delayed_node->inode_item, &inode->vfs_inode);
1898 set_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
1899 delayed_node->count++;
1900 atomic_inc(&root->fs_info->delayed_root->items);
1901 release_node:
1902 mutex_unlock(&delayed_node->mutex);
1903 btrfs_release_delayed_node(delayed_node);
1904 return ret;
1905 }
1906
btrfs_delayed_delete_inode_ref(struct btrfs_inode * inode)1907 int btrfs_delayed_delete_inode_ref(struct btrfs_inode *inode)
1908 {
1909 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1910 struct btrfs_delayed_node *delayed_node;
1911
1912 /*
1913 * we don't do delayed inode updates during log recovery because it
1914 * leads to enospc problems. This means we also can't do
1915 * delayed inode refs
1916 */
1917 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
1918 return -EAGAIN;
1919
1920 delayed_node = btrfs_get_or_create_delayed_node(inode);
1921 if (IS_ERR(delayed_node))
1922 return PTR_ERR(delayed_node);
1923
1924 /*
1925 * We don't reserve space for inode ref deletion is because:
1926 * - We ONLY do async inode ref deletion for the inode who has only
1927 * one link(i_nlink == 1), it means there is only one inode ref.
1928 * And in most case, the inode ref and the inode item are in the
1929 * same leaf, and we will deal with them at the same time.
1930 * Since we are sure we will reserve the space for the inode item,
1931 * it is unnecessary to reserve space for inode ref deletion.
1932 * - If the inode ref and the inode item are not in the same leaf,
1933 * We also needn't worry about enospc problem, because we reserve
1934 * much more space for the inode update than it needs.
1935 * - At the worst, we can steal some space from the global reservation.
1936 * It is very rare.
1937 */
1938 mutex_lock(&delayed_node->mutex);
1939 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1940 goto release_node;
1941
1942 set_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
1943 delayed_node->count++;
1944 atomic_inc(&fs_info->delayed_root->items);
1945 release_node:
1946 mutex_unlock(&delayed_node->mutex);
1947 btrfs_release_delayed_node(delayed_node);
1948 return 0;
1949 }
1950
__btrfs_kill_delayed_node(struct btrfs_delayed_node * delayed_node)1951 static void __btrfs_kill_delayed_node(struct btrfs_delayed_node *delayed_node)
1952 {
1953 struct btrfs_root *root = delayed_node->root;
1954 struct btrfs_fs_info *fs_info = root->fs_info;
1955 struct btrfs_delayed_item *curr_item, *prev_item;
1956
1957 mutex_lock(&delayed_node->mutex);
1958 curr_item = __btrfs_first_delayed_insertion_item(delayed_node);
1959 while (curr_item) {
1960 prev_item = curr_item;
1961 curr_item = __btrfs_next_delayed_item(prev_item);
1962 btrfs_release_delayed_item(prev_item);
1963 }
1964
1965 if (delayed_node->index_item_leaves > 0) {
1966 btrfs_delayed_item_release_leaves(delayed_node,
1967 delayed_node->index_item_leaves);
1968 delayed_node->index_item_leaves = 0;
1969 }
1970
1971 curr_item = __btrfs_first_delayed_deletion_item(delayed_node);
1972 while (curr_item) {
1973 btrfs_delayed_item_release_metadata(root, curr_item);
1974 prev_item = curr_item;
1975 curr_item = __btrfs_next_delayed_item(prev_item);
1976 btrfs_release_delayed_item(prev_item);
1977 }
1978
1979 btrfs_release_delayed_iref(delayed_node);
1980
1981 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1982 btrfs_delayed_inode_release_metadata(fs_info, delayed_node, false);
1983 btrfs_release_delayed_inode(delayed_node);
1984 }
1985 mutex_unlock(&delayed_node->mutex);
1986 }
1987
btrfs_kill_delayed_inode_items(struct btrfs_inode * inode)1988 void btrfs_kill_delayed_inode_items(struct btrfs_inode *inode)
1989 {
1990 struct btrfs_delayed_node *delayed_node;
1991
1992 delayed_node = btrfs_get_delayed_node(inode);
1993 if (!delayed_node)
1994 return;
1995
1996 __btrfs_kill_delayed_node(delayed_node);
1997 btrfs_release_delayed_node(delayed_node);
1998 }
1999
btrfs_kill_all_delayed_nodes(struct btrfs_root * root)2000 void btrfs_kill_all_delayed_nodes(struct btrfs_root *root)
2001 {
2002 u64 inode_id = 0;
2003 struct btrfs_delayed_node *delayed_nodes[8];
2004 int i, n;
2005
2006 while (1) {
2007 spin_lock(&root->inode_lock);
2008 n = radix_tree_gang_lookup(&root->delayed_nodes_tree,
2009 (void **)delayed_nodes, inode_id,
2010 ARRAY_SIZE(delayed_nodes));
2011 if (!n) {
2012 spin_unlock(&root->inode_lock);
2013 break;
2014 }
2015
2016 inode_id = delayed_nodes[n - 1]->inode_id + 1;
2017 for (i = 0; i < n; i++) {
2018 /*
2019 * Don't increase refs in case the node is dead and
2020 * about to be removed from the tree in the loop below
2021 */
2022 if (!refcount_inc_not_zero(&delayed_nodes[i]->refs))
2023 delayed_nodes[i] = NULL;
2024 }
2025 spin_unlock(&root->inode_lock);
2026
2027 for (i = 0; i < n; i++) {
2028 if (!delayed_nodes[i])
2029 continue;
2030 __btrfs_kill_delayed_node(delayed_nodes[i]);
2031 btrfs_release_delayed_node(delayed_nodes[i]);
2032 }
2033 }
2034 }
2035
btrfs_destroy_delayed_inodes(struct btrfs_fs_info * fs_info)2036 void btrfs_destroy_delayed_inodes(struct btrfs_fs_info *fs_info)
2037 {
2038 struct btrfs_delayed_node *curr_node, *prev_node;
2039
2040 curr_node = btrfs_first_delayed_node(fs_info->delayed_root);
2041 while (curr_node) {
2042 __btrfs_kill_delayed_node(curr_node);
2043
2044 prev_node = curr_node;
2045 curr_node = btrfs_next_delayed_node(curr_node);
2046 btrfs_release_delayed_node(prev_node);
2047 }
2048 }
2049
btrfs_log_get_delayed_items(struct btrfs_inode * inode,struct list_head * ins_list,struct list_head * del_list)2050 void btrfs_log_get_delayed_items(struct btrfs_inode *inode,
2051 struct list_head *ins_list,
2052 struct list_head *del_list)
2053 {
2054 struct btrfs_delayed_node *node;
2055 struct btrfs_delayed_item *item;
2056
2057 node = btrfs_get_delayed_node(inode);
2058 if (!node)
2059 return;
2060
2061 mutex_lock(&node->mutex);
2062 item = __btrfs_first_delayed_insertion_item(node);
2063 while (item) {
2064 /*
2065 * It's possible that the item is already in a log list. This
2066 * can happen in case two tasks are trying to log the same
2067 * directory. For example if we have tasks A and task B:
2068 *
2069 * Task A collected the delayed items into a log list while
2070 * under the inode's log_mutex (at btrfs_log_inode()), but it
2071 * only releases the items after logging the inodes they point
2072 * to (if they are new inodes), which happens after unlocking
2073 * the log mutex;
2074 *
2075 * Task B enters btrfs_log_inode() and acquires the log_mutex
2076 * of the same directory inode, before task B releases the
2077 * delayed items. This can happen for example when logging some
2078 * inode we need to trigger logging of its parent directory, so
2079 * logging two files that have the same parent directory can
2080 * lead to this.
2081 *
2082 * If this happens, just ignore delayed items already in a log
2083 * list. All the tasks logging the directory are under a log
2084 * transaction and whichever finishes first can not sync the log
2085 * before the other completes and leaves the log transaction.
2086 */
2087 if (!item->logged && list_empty(&item->log_list)) {
2088 refcount_inc(&item->refs);
2089 list_add_tail(&item->log_list, ins_list);
2090 }
2091 item = __btrfs_next_delayed_item(item);
2092 }
2093
2094 item = __btrfs_first_delayed_deletion_item(node);
2095 while (item) {
2096 /* It may be non-empty, for the same reason mentioned above. */
2097 if (!item->logged && list_empty(&item->log_list)) {
2098 refcount_inc(&item->refs);
2099 list_add_tail(&item->log_list, del_list);
2100 }
2101 item = __btrfs_next_delayed_item(item);
2102 }
2103 mutex_unlock(&node->mutex);
2104
2105 /*
2106 * We are called during inode logging, which means the inode is in use
2107 * and can not be evicted before we finish logging the inode. So we never
2108 * have the last reference on the delayed inode.
2109 * Also, we don't use btrfs_release_delayed_node() because that would
2110 * requeue the delayed inode (change its order in the list of prepared
2111 * nodes) and we don't want to do such change because we don't create or
2112 * delete delayed items.
2113 */
2114 ASSERT(refcount_read(&node->refs) > 1);
2115 refcount_dec(&node->refs);
2116 }
2117
btrfs_log_put_delayed_items(struct btrfs_inode * inode,struct list_head * ins_list,struct list_head * del_list)2118 void btrfs_log_put_delayed_items(struct btrfs_inode *inode,
2119 struct list_head *ins_list,
2120 struct list_head *del_list)
2121 {
2122 struct btrfs_delayed_node *node;
2123 struct btrfs_delayed_item *item;
2124 struct btrfs_delayed_item *next;
2125
2126 node = btrfs_get_delayed_node(inode);
2127 if (!node)
2128 return;
2129
2130 mutex_lock(&node->mutex);
2131
2132 list_for_each_entry_safe(item, next, ins_list, log_list) {
2133 item->logged = true;
2134 list_del_init(&item->log_list);
2135 if (refcount_dec_and_test(&item->refs))
2136 kfree(item);
2137 }
2138
2139 list_for_each_entry_safe(item, next, del_list, log_list) {
2140 item->logged = true;
2141 list_del_init(&item->log_list);
2142 if (refcount_dec_and_test(&item->refs))
2143 kfree(item);
2144 }
2145
2146 mutex_unlock(&node->mutex);
2147
2148 /*
2149 * We are called during inode logging, which means the inode is in use
2150 * and can not be evicted before we finish logging the inode. So we never
2151 * have the last reference on the delayed inode.
2152 * Also, we don't use btrfs_release_delayed_node() because that would
2153 * requeue the delayed inode (change its order in the list of prepared
2154 * nodes) and we don't want to do such change because we don't create or
2155 * delete delayed items.
2156 */
2157 ASSERT(refcount_read(&node->refs) > 1);
2158 refcount_dec(&node->refs);
2159 }
2160