Line data Source code
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 "ctree.h"
10 : #include "fs.h"
11 : #include "messages.h"
12 : #include "misc.h"
13 : #include "delayed-inode.h"
14 : #include "disk-io.h"
15 : #include "transaction.h"
16 : #include "qgroup.h"
17 : #include "locking.h"
18 : #include "inode-item.h"
19 : #include "space-info.h"
20 : #include "accessors.h"
21 : #include "file-item.h"
22 :
23 : #define BTRFS_DELAYED_WRITEBACK 512
24 : #define BTRFS_DELAYED_BACKGROUND 128
25 : #define BTRFS_DELAYED_BATCH 16
26 :
27 : static struct kmem_cache *delayed_node_cache;
28 :
29 11 : int __init btrfs_delayed_inode_init(void)
30 : {
31 11 : delayed_node_cache = kmem_cache_create("btrfs_delayed_node",
32 : sizeof(struct btrfs_delayed_node),
33 : 0,
34 : SLAB_MEM_SPREAD,
35 : NULL);
36 11 : if (!delayed_node_cache)
37 0 : return -ENOMEM;
38 : return 0;
39 : }
40 :
41 0 : void __cold btrfs_delayed_inode_exit(void)
42 : {
43 0 : kmem_cache_destroy(delayed_node_cache);
44 0 : }
45 :
46 1934969 : static inline void btrfs_init_delayed_node(
47 : struct btrfs_delayed_node *delayed_node,
48 : struct btrfs_root *root, u64 inode_id)
49 : {
50 1934969 : delayed_node->root = root;
51 1934969 : delayed_node->inode_id = inode_id;
52 1934969 : refcount_set(&delayed_node->refs, 0);
53 1934969 : delayed_node->ins_root = RB_ROOT_CACHED;
54 1934969 : delayed_node->del_root = RB_ROOT_CACHED;
55 1934969 : mutex_init(&delayed_node->mutex);
56 1934803 : INIT_LIST_HEAD(&delayed_node->n_list);
57 1934803 : INIT_LIST_HEAD(&delayed_node->p_list);
58 1934803 : }
59 :
60 49988820 : static struct btrfs_delayed_node *btrfs_get_delayed_node(
61 : struct btrfs_inode *btrfs_inode)
62 : {
63 49988820 : struct btrfs_root *root = btrfs_inode->root;
64 49988820 : u64 ino = btrfs_ino(btrfs_inode);
65 49988820 : struct btrfs_delayed_node *node;
66 :
67 49988820 : node = READ_ONCE(btrfs_inode->delayed_node);
68 49988820 : if (node) {
69 44397555 : refcount_inc(&node->refs);
70 44397555 : return node;
71 : }
72 :
73 5591265 : spin_lock(&root->inode_lock);
74 5592475 : node = radix_tree_lookup(&root->delayed_nodes_tree, ino);
75 :
76 5592459 : if (node) {
77 131 : if (btrfs_inode->delayed_node) {
78 2 : refcount_inc(&node->refs); /* can be accessed */
79 2 : BUG_ON(btrfs_inode->delayed_node != node);
80 2 : spin_unlock(&root->inode_lock);
81 2 : return node;
82 : }
83 :
84 : /*
85 : * It's possible that we're racing into the middle of removing
86 : * this node from the radix tree. In this case, the refcount
87 : * was zero and it should never go back to one. Just return
88 : * NULL like it was never in the radix at all; our release
89 : * function is in the process of removing it.
90 : *
91 : * Some implementations of refcount_inc refuse to bump the
92 : * refcount once it has hit zero. If we don't do this dance
93 : * here, refcount_inc() may decide to just WARN_ONCE() instead
94 : * of actually bumping the refcount.
95 : *
96 : * If this node is properly in the radix, we want to bump the
97 : * refcount twice, once for the inode and once for this get
98 : * operation.
99 : */
100 129 : if (refcount_inc_not_zero(&node->refs)) {
101 129 : refcount_inc(&node->refs);
102 129 : btrfs_inode->delayed_node = node;
103 : } else {
104 : node = NULL;
105 : }
106 :
107 129 : spin_unlock(&root->inode_lock);
108 129 : return node;
109 : }
110 5592328 : spin_unlock(&root->inode_lock);
111 :
112 5592328 : return NULL;
113 : }
114 :
115 : /* Will return either the node or PTR_ERR(-ENOMEM) */
116 43100464 : static struct btrfs_delayed_node *btrfs_get_or_create_delayed_node(
117 : struct btrfs_inode *btrfs_inode)
118 : {
119 43100464 : struct btrfs_delayed_node *node;
120 43100464 : struct btrfs_root *root = btrfs_inode->root;
121 43100464 : u64 ino = btrfs_ino(btrfs_inode);
122 43100476 : int ret;
123 :
124 43100476 : again:
125 43100476 : node = btrfs_get_delayed_node(btrfs_inode);
126 43101872 : if (node)
127 41166018 : return node;
128 :
129 1935854 : node = kmem_cache_zalloc(delayed_node_cache, GFP_NOFS);
130 1935065 : if (!node)
131 : return ERR_PTR(-ENOMEM);
132 1935065 : btrfs_init_delayed_node(node, root, ino);
133 :
134 : /* cached in the btrfs inode and can be accessed */
135 1935033 : refcount_set(&node->refs, 2);
136 :
137 1935033 : ret = radix_tree_preload(GFP_NOFS);
138 1933178 : if (ret) {
139 0 : kmem_cache_free(delayed_node_cache, node);
140 0 : return ERR_PTR(ret);
141 : }
142 :
143 1933178 : spin_lock(&root->inode_lock);
144 1935862 : ret = radix_tree_insert(&root->delayed_nodes_tree, ino, node);
145 1935862 : if (ret == -EEXIST) {
146 12 : spin_unlock(&root->inode_lock);
147 12 : kmem_cache_free(delayed_node_cache, node);
148 12 : radix_tree_preload_end();
149 12 : goto again;
150 : }
151 1935850 : btrfs_inode->delayed_node = node;
152 1935850 : spin_unlock(&root->inode_lock);
153 1935850 : radix_tree_preload_end();
154 :
155 1935850 : return node;
156 : }
157 :
158 : /*
159 : * Call it when holding delayed_node->mutex
160 : *
161 : * If mod = 1, add this node into the prepared list.
162 : */
163 43222035 : static void btrfs_queue_delayed_node(struct btrfs_delayed_root *root,
164 : struct btrfs_delayed_node *node,
165 : int mod)
166 : {
167 43222035 : spin_lock(&root->lock);
168 86450618 : if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
169 34647325 : if (!list_empty(&node->p_list))
170 34463359 : list_move_tail(&node->p_list, &root->prepare_list);
171 183966 : else if (mod)
172 169260 : list_add_tail(&node->p_list, &root->prepare_list);
173 : } else {
174 8577984 : list_add_tail(&node->n_list, &root->node_list);
175 8577984 : list_add_tail(&node->p_list, &root->prepare_list);
176 8577984 : refcount_inc(&node->refs); /* inserted into list */
177 8577984 : root->nodes++;
178 8577984 : set_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
179 : }
180 43225306 : spin_unlock(&root->lock);
181 43225175 : }
182 :
183 : /* Call it when holding delayed_node->mutex */
184 12088732 : static void btrfs_dequeue_delayed_node(struct btrfs_delayed_root *root,
185 : struct btrfs_delayed_node *node)
186 : {
187 12088732 : spin_lock(&root->lock);
188 24184634 : if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
189 8577984 : root->nodes--;
190 8577984 : refcount_dec(&node->refs); /* not in the list */
191 8577984 : list_del_init(&node->n_list);
192 8577984 : if (!list_empty(&node->p_list))
193 3824847 : list_del_init(&node->p_list);
194 8577984 : clear_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
195 : }
196 12092317 : spin_unlock(&root->lock);
197 12092207 : }
198 :
199 2699709 : static struct btrfs_delayed_node *btrfs_first_delayed_node(
200 : struct btrfs_delayed_root *delayed_root)
201 : {
202 2699709 : struct list_head *p;
203 2699709 : struct btrfs_delayed_node *node = NULL;
204 :
205 2699709 : spin_lock(&delayed_root->lock);
206 2699712 : if (list_empty(&delayed_root->node_list))
207 1652457 : goto out;
208 :
209 1047255 : p = delayed_root->node_list.next;
210 1047255 : node = list_entry(p, struct btrfs_delayed_node, n_list);
211 1047255 : refcount_inc(&node->refs);
212 2699712 : out:
213 2699712 : spin_unlock(&delayed_root->lock);
214 :
215 2699712 : return node;
216 : }
217 :
218 2591439 : static struct btrfs_delayed_node *btrfs_next_delayed_node(
219 : struct btrfs_delayed_node *node)
220 : {
221 2591439 : struct btrfs_delayed_root *delayed_root;
222 2591439 : struct list_head *p;
223 2591439 : struct btrfs_delayed_node *next = NULL;
224 :
225 2591439 : delayed_root = node->root->fs_info->delayed_root;
226 2591439 : spin_lock(&delayed_root->lock);
227 2591450 : if (!test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
228 : /* not in the list */
229 30091 : if (list_empty(&delayed_root->node_list))
230 705 : goto out;
231 29386 : p = delayed_root->node_list.next;
232 2561359 : } else if (list_is_last(&node->n_list, &delayed_root->node_list))
233 1028265 : goto out;
234 : else
235 : p = node->n_list.next;
236 :
237 1562480 : next = list_entry(p, struct btrfs_delayed_node, n_list);
238 1562480 : refcount_inc(&next->refs);
239 2591450 : out:
240 2591450 : spin_unlock(&delayed_root->lock);
241 :
242 2591438 : return next;
243 : }
244 :
245 55302863 : static void __btrfs_release_delayed_node(
246 : struct btrfs_delayed_node *delayed_node,
247 : int mod)
248 : {
249 55302863 : struct btrfs_delayed_root *delayed_root;
250 :
251 55302863 : if (!delayed_node)
252 : return;
253 :
254 55302863 : delayed_root = delayed_node->root->fs_info->delayed_root;
255 :
256 55302863 : mutex_lock(&delayed_node->mutex);
257 55313936 : if (delayed_node->count)
258 43223480 : btrfs_queue_delayed_node(delayed_root, delayed_node, mod);
259 : else
260 12090456 : btrfs_dequeue_delayed_node(delayed_root, delayed_node);
261 55317297 : mutex_unlock(&delayed_node->mutex);
262 :
263 55316876 : if (refcount_dec_and_test(&delayed_node->refs)) {
264 1935849 : struct btrfs_root *root = delayed_node->root;
265 :
266 1935849 : spin_lock(&root->inode_lock);
267 : /*
268 : * Once our refcount goes to zero, nobody is allowed to bump it
269 : * back up. We can delete it now.
270 : */
271 1935850 : ASSERT(refcount_read(&delayed_node->refs) == 0);
272 1935850 : radix_tree_delete(&root->delayed_nodes_tree,
273 1935850 : delayed_node->inode_id);
274 1935850 : spin_unlock(&root->inode_lock);
275 1935850 : kmem_cache_free(delayed_node_cache, delayed_node);
276 : }
277 : }
278 :
279 : static inline void btrfs_release_delayed_node(struct btrfs_delayed_node *node)
280 : {
281 50393325 : __btrfs_release_delayed_node(node, 0);
282 5917453 : }
283 :
284 4961496 : static struct btrfs_delayed_node *btrfs_first_prepared_delayed_node(
285 : struct btrfs_delayed_root *delayed_root)
286 : {
287 4961496 : struct list_head *p;
288 4961496 : struct btrfs_delayed_node *node = NULL;
289 :
290 4961496 : spin_lock(&delayed_root->lock);
291 4962240 : if (list_empty(&delayed_root->prepare_list))
292 39843 : goto out;
293 :
294 4922397 : p = delayed_root->prepare_list.next;
295 4922397 : list_del_init(p);
296 4922397 : node = list_entry(p, struct btrfs_delayed_node, p_list);
297 4922397 : refcount_inc(&node->refs);
298 4962240 : out:
299 4962240 : spin_unlock(&delayed_root->lock);
300 :
301 4962218 : return node;
302 : }
303 :
304 : static inline void btrfs_release_prepared_delayed_node(
305 : struct btrfs_delayed_node *node)
306 : {
307 4919866 : __btrfs_release_delayed_node(node, 1);
308 : }
309 :
310 4759577 : static struct btrfs_delayed_item *btrfs_alloc_delayed_item(u16 data_len,
311 : struct btrfs_delayed_node *node,
312 : enum btrfs_delayed_item_type type)
313 : {
314 4759577 : struct btrfs_delayed_item *item;
315 :
316 4759577 : item = kmalloc(sizeof(*item) + data_len, GFP_NOFS);
317 4759464 : if (item) {
318 4759464 : item->data_len = data_len;
319 4759464 : item->type = type;
320 4759464 : item->bytes_reserved = 0;
321 4759464 : item->delayed_node = node;
322 4759464 : RB_CLEAR_NODE(&item->rb_node);
323 4759464 : INIT_LIST_HEAD(&item->log_list);
324 4759464 : item->logged = false;
325 4759464 : refcount_set(&item->refs, 1);
326 : }
327 4759464 : return item;
328 : }
329 :
330 : /*
331 : * __btrfs_lookup_delayed_item - look up the delayed item by key
332 : * @delayed_node: pointer to the delayed node
333 : * @index: the dir index value to lookup (offset of a dir index key)
334 : *
335 : * Note: if we don't find the right item, we will return the prev item and
336 : * the next item.
337 : */
338 : static struct btrfs_delayed_item *__btrfs_lookup_delayed_item(
339 : struct rb_root *root,
340 : u64 index)
341 : {
342 1885582 : struct rb_node *node = root->rb_node;
343 1885582 : struct btrfs_delayed_item *delayed_item = NULL;
344 :
345 2233246 : while (node) {
346 851714 : delayed_item = rb_entry(node, struct btrfs_delayed_item,
347 : rb_node);
348 851714 : if (delayed_item->index < index)
349 180768 : node = node->rb_right;
350 670946 : else if (delayed_item->index > index)
351 166896 : node = node->rb_left;
352 : else
353 : return delayed_item;
354 : }
355 :
356 : return NULL;
357 : }
358 :
359 4759736 : static int __btrfs_add_delayed_item(struct btrfs_delayed_node *delayed_node,
360 : struct btrfs_delayed_item *ins)
361 : {
362 4759736 : struct rb_node **p, *node;
363 4759736 : struct rb_node *parent_node = NULL;
364 4759736 : struct rb_root_cached *root;
365 4759736 : struct btrfs_delayed_item *item;
366 4759736 : bool leftmost = true;
367 :
368 4759736 : if (ins->type == BTRFS_DELAYED_INSERTION_ITEM)
369 3378212 : root = &delayed_node->ins_root;
370 : else
371 1381524 : root = &delayed_node->del_root;
372 :
373 4759736 : p = &root->rb_root.rb_node;
374 4759736 : node = &ins->rb_node;
375 :
376 28638339 : while (*p) {
377 23878603 : parent_node = *p;
378 23878603 : item = rb_entry(parent_node, struct btrfs_delayed_item,
379 : rb_node);
380 :
381 23878603 : if (item->index < ins->index) {
382 23834147 : p = &(*p)->rb_right;
383 23834147 : leftmost = false;
384 44456 : } else if (item->index > ins->index) {
385 44456 : p = &(*p)->rb_left;
386 : } else {
387 : return -EEXIST;
388 : }
389 : }
390 :
391 4759736 : rb_link_node(node, parent_node, p);
392 4759736 : rb_insert_color_cached(node, root, leftmost);
393 :
394 4759505 : if (ins->type == BTRFS_DELAYED_INSERTION_ITEM &&
395 3378083 : ins->index >= delayed_node->index_cnt)
396 3378053 : delayed_node->index_cnt = ins->index + 1;
397 :
398 4759505 : delayed_node->count++;
399 4759505 : atomic_inc(&delayed_node->root->fs_info->delayed_root->items);
400 4759505 : return 0;
401 : }
402 :
403 14686233 : static void finish_one_item(struct btrfs_delayed_root *delayed_root)
404 : {
405 14686233 : int seq = atomic_inc_return(&delayed_root->items_seq);
406 :
407 : /* atomic_dec_return implies a barrier */
408 14685644 : if ((atomic_dec_return(&delayed_root->items) <
409 3072260 : BTRFS_DELAYED_BACKGROUND || seq % BTRFS_DELAYED_BATCH == 0))
410 11805917 : cond_wake_up_nomb(&delayed_root->wait);
411 14686075 : }
412 :
413 4759446 : static void __btrfs_remove_delayed_item(struct btrfs_delayed_item *delayed_item)
414 : {
415 4759446 : struct rb_root_cached *root;
416 4759446 : struct btrfs_delayed_root *delayed_root;
417 :
418 : /* Not inserted, ignore it. */
419 4759446 : if (RB_EMPTY_NODE(&delayed_item->rb_node))
420 : return;
421 :
422 4759446 : delayed_root = delayed_item->delayed_node->root->fs_info->delayed_root;
423 :
424 4759446 : BUG_ON(!delayed_root);
425 :
426 4759446 : if (delayed_item->type == BTRFS_DELAYED_INSERTION_ITEM)
427 3377967 : root = &delayed_item->delayed_node->ins_root;
428 : else
429 1381479 : root = &delayed_item->delayed_node->del_root;
430 :
431 4759446 : rb_erase_cached(&delayed_item->rb_node, root);
432 4759019 : RB_CLEAR_NODE(&delayed_item->rb_node);
433 4759019 : delayed_item->delayed_node->count--;
434 :
435 4759019 : finish_one_item(delayed_root);
436 : }
437 :
438 4759316 : static void btrfs_release_delayed_item(struct btrfs_delayed_item *item)
439 : {
440 4759316 : if (item) {
441 4759316 : __btrfs_remove_delayed_item(item);
442 4759480 : if (refcount_dec_and_test(&item->refs))
443 4759363 : kfree(item);
444 : }
445 4759613 : }
446 :
447 : static struct btrfs_delayed_item *__btrfs_first_delayed_insertion_item(
448 : struct btrfs_delayed_node *delayed_node)
449 : {
450 10026802 : struct rb_node *p;
451 10026802 : struct btrfs_delayed_item *item = NULL;
452 :
453 10026802 : p = rb_first_cached(&delayed_node->ins_root);
454 10026802 : if (p)
455 198715 : item = rb_entry(p, struct btrfs_delayed_item, rb_node);
456 :
457 8166539 : return item;
458 : }
459 :
460 : static struct btrfs_delayed_item *__btrfs_first_delayed_deletion_item(
461 : struct btrfs_delayed_node *delayed_node)
462 : {
463 9731659 : struct rb_node *p;
464 9731659 : struct btrfs_delayed_item *item = NULL;
465 :
466 9731659 : p = rb_first_cached(&delayed_node->del_root);
467 9731659 : if (p)
468 31398 : item = rb_entry(p, struct btrfs_delayed_item, rb_node);
469 :
470 7871403 : return item;
471 : }
472 :
473 : static struct btrfs_delayed_item *__btrfs_next_delayed_item(
474 : struct btrfs_delayed_item *item)
475 : {
476 21717320 : struct rb_node *p;
477 21717320 : struct btrfs_delayed_item *next = NULL;
478 :
479 21717320 : p = rb_next(&item->rb_node);
480 21717747 : if (p)
481 137236 : next = rb_entry(p, struct btrfs_delayed_item, rb_node);
482 :
483 3274737 : return next;
484 : }
485 :
486 2369059 : static int btrfs_delayed_item_reserve_metadata(struct btrfs_trans_handle *trans,
487 : struct btrfs_delayed_item *item)
488 : {
489 2369059 : struct btrfs_block_rsv *src_rsv;
490 2369059 : struct btrfs_block_rsv *dst_rsv;
491 2369059 : struct btrfs_fs_info *fs_info = trans->fs_info;
492 2369059 : u64 num_bytes;
493 2369059 : int ret;
494 :
495 2369059 : if (!trans->bytes_reserved)
496 : return 0;
497 :
498 2368902 : src_rsv = trans->block_rsv;
499 2368902 : dst_rsv = &fs_info->delayed_block_rsv;
500 :
501 2368902 : num_bytes = btrfs_calc_insert_metadata_size(fs_info, 1);
502 :
503 : /*
504 : * Here we migrate space rsv from transaction rsv, since have already
505 : * reserved space when starting a transaction. So no need to reserve
506 : * qgroup space here.
507 : */
508 2368902 : ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, true);
509 2368982 : if (!ret) {
510 2368984 : trace_btrfs_space_reservation(fs_info, "delayed_item",
511 2368984 : item->delayed_node->inode_id,
512 : num_bytes, 1);
513 : /*
514 : * For insertions we track reserved metadata space by accounting
515 : * for the number of leaves that will be used, based on the delayed
516 : * node's index_items_size field.
517 : */
518 2368968 : if (item->type == BTRFS_DELAYED_DELETION_ITEM)
519 1381503 : item->bytes_reserved = num_bytes;
520 : }
521 :
522 : return ret;
523 : }
524 :
525 156282 : static void btrfs_delayed_item_release_metadata(struct btrfs_root *root,
526 : struct btrfs_delayed_item *item)
527 : {
528 156282 : struct btrfs_block_rsv *rsv;
529 156282 : struct btrfs_fs_info *fs_info = root->fs_info;
530 :
531 156282 : if (!item->bytes_reserved)
532 : return;
533 :
534 156282 : rsv = &fs_info->delayed_block_rsv;
535 : /*
536 : * Check btrfs_delayed_item_reserve_metadata() to see why we don't need
537 : * to release/reserve qgroup space.
538 : */
539 156282 : trace_btrfs_space_reservation(fs_info, "delayed_item",
540 156282 : item->delayed_node->inode_id,
541 : item->bytes_reserved, 0);
542 156282 : btrfs_block_rsv_release(fs_info, rsv, item->bytes_reserved, NULL);
543 : }
544 :
545 987608 : static void btrfs_delayed_item_release_leaves(struct btrfs_delayed_node *node,
546 : unsigned int num_leaves)
547 : {
548 987608 : struct btrfs_fs_info *fs_info = node->root->fs_info;
549 987608 : const u64 bytes = btrfs_calc_insert_metadata_size(fs_info, num_leaves);
550 :
551 : /* There are no space reservations during log replay, bail out. */
552 1975216 : if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
553 : return;
554 :
555 987478 : trace_btrfs_space_reservation(fs_info, "delayed_item", node->inode_id,
556 : bytes, 0);
557 987477 : btrfs_block_rsv_release(fs_info, &fs_info->delayed_block_rsv, bytes, NULL);
558 : }
559 :
560 8678653 : static int btrfs_delayed_inode_reserve_metadata(
561 : struct btrfs_trans_handle *trans,
562 : struct btrfs_root *root,
563 : struct btrfs_delayed_node *node)
564 : {
565 8678653 : struct btrfs_fs_info *fs_info = root->fs_info;
566 8678653 : struct btrfs_block_rsv *src_rsv;
567 8678653 : struct btrfs_block_rsv *dst_rsv;
568 8678653 : u64 num_bytes;
569 8678653 : int ret;
570 :
571 8678653 : src_rsv = trans->block_rsv;
572 8678653 : dst_rsv = &fs_info->delayed_block_rsv;
573 :
574 8678653 : num_bytes = btrfs_calc_metadata_size(fs_info, 1);
575 :
576 : /*
577 : * btrfs_dirty_inode will update the inode under btrfs_join_transaction
578 : * which doesn't reserve space for speed. This is a problem since we
579 : * still need to reserve space for this update, so try to reserve the
580 : * space.
581 : *
582 : * Now if src_rsv == delalloc_block_rsv we'll let it just steal since
583 : * we always reserve enough to update the inode item.
584 : */
585 8678653 : if (!src_rsv || (!trans->bytes_reserved &&
586 1792248 : src_rsv->type != BTRFS_BLOCK_RSV_DELALLOC)) {
587 2291299 : ret = btrfs_qgroup_reserve_meta(root, num_bytes,
588 : BTRFS_QGROUP_RSV_META_PREALLOC, true);
589 2290303 : if (ret < 0)
590 : return ret;
591 2287283 : ret = btrfs_block_rsv_add(fs_info, dst_rsv, num_bytes,
592 : BTRFS_RESERVE_NO_FLUSH);
593 : /* NO_FLUSH could only fail with -ENOSPC */
594 2289073 : ASSERT(ret == 0 || ret == -ENOSPC);
595 2289073 : if (ret)
596 86113 : btrfs_qgroup_free_meta_prealloc(root, num_bytes);
597 : } else {
598 6387354 : ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, true);
599 : }
600 :
601 8676655 : if (!ret) {
602 8590160 : trace_btrfs_space_reservation(fs_info, "delayed_inode",
603 : node->inode_id, num_bytes, 1);
604 8589919 : node->bytes_reserved = num_bytes;
605 : }
606 :
607 : return ret;
608 : }
609 :
610 8589875 : static void btrfs_delayed_inode_release_metadata(struct btrfs_fs_info *fs_info,
611 : struct btrfs_delayed_node *node,
612 : bool qgroup_free)
613 : {
614 8589875 : struct btrfs_block_rsv *rsv;
615 :
616 8589875 : if (!node->bytes_reserved)
617 : return;
618 :
619 8589859 : rsv = &fs_info->delayed_block_rsv;
620 8589859 : trace_btrfs_space_reservation(fs_info, "delayed_inode",
621 : node->inode_id, node->bytes_reserved, 0);
622 8588111 : btrfs_block_rsv_release(fs_info, rsv, node->bytes_reserved, NULL);
623 8590712 : if (qgroup_free)
624 6 : btrfs_qgroup_free_meta_prealloc(node->root,
625 6 : node->bytes_reserved);
626 : else
627 8590706 : btrfs_qgroup_convert_reserved_meta(node->root,
628 8590706 : node->bytes_reserved);
629 8590446 : node->bytes_reserved = 0;
630 : }
631 :
632 : /*
633 : * Insert a single delayed item or a batch of delayed items, as many as possible
634 : * that fit in a leaf. The delayed items (dir index keys) are sorted by their key
635 : * in the rbtree, and if there's a gap between two consecutive dir index items,
636 : * then it means at some point we had delayed dir indexes to add but they got
637 : * removed (by btrfs_delete_delayed_dir_index()) before we attempted to flush them
638 : * into the subvolume tree. Dir index keys also have their offsets coming from a
639 : * monotonically increasing counter, so we can't get new keys with an offset that
640 : * fits within a gap between delayed dir index items.
641 : */
642 653326 : static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans,
643 : struct btrfs_root *root,
644 : struct btrfs_path *path,
645 : struct btrfs_delayed_item *first_item)
646 : {
647 653326 : struct btrfs_fs_info *fs_info = root->fs_info;
648 653326 : struct btrfs_delayed_node *node = first_item->delayed_node;
649 653326 : LIST_HEAD(item_list);
650 653326 : struct btrfs_delayed_item *curr;
651 653326 : struct btrfs_delayed_item *next;
652 653326 : const int max_size = BTRFS_LEAF_DATA_SIZE(fs_info);
653 653326 : struct btrfs_item_batch batch;
654 653326 : struct btrfs_key first_key;
655 653326 : const u32 first_data_size = first_item->data_len;
656 653326 : int total_size;
657 653326 : char *ins_data = NULL;
658 653326 : int ret;
659 653326 : bool continuous_keys_only = false;
660 :
661 653326 : lockdep_assert_held(&node->mutex);
662 :
663 : /*
664 : * During normal operation the delayed index offset is continuously
665 : * increasing, so we can batch insert all items as there will not be any
666 : * overlapping keys in the tree.
667 : *
668 : * The exception to this is log replay, where we may have interleaved
669 : * offsets in the tree, so our batch needs to be continuous keys only in
670 : * order to ensure we do not end up with out of order items in our leaf.
671 : */
672 1306652 : if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
673 138 : continuous_keys_only = true;
674 :
675 : /*
676 : * For delayed items to insert, we track reserved metadata bytes based
677 : * on the number of leaves that we will use.
678 : * See btrfs_insert_delayed_dir_index() and
679 : * btrfs_delayed_item_reserve_metadata()).
680 : */
681 653326 : ASSERT(first_item->bytes_reserved == 0);
682 :
683 653326 : list_add_tail(&first_item->tree_list, &item_list);
684 653290 : batch.total_data_size = first_data_size;
685 653290 : batch.nr = 1;
686 653290 : total_size = first_data_size + sizeof(struct btrfs_item);
687 653290 : curr = first_item;
688 :
689 5091720 : while (true) {
690 2872505 : int next_size;
691 :
692 2872505 : next = __btrfs_next_delayed_item(curr);
693 2219904 : if (!next)
694 : break;
695 :
696 : /*
697 : * We cannot allow gaps in the key space if we're doing log
698 : * replay.
699 : */
700 2219904 : if (continuous_keys_only && (next->index != curr->index + 1))
701 : break;
702 :
703 2219897 : ASSERT(next->bytes_reserved == 0);
704 :
705 2219897 : next_size = next->data_len + sizeof(struct btrfs_item);
706 2219897 : if (total_size + next_size > max_size)
707 : break;
708 :
709 2219611 : list_add_tail(&next->tree_list, &item_list);
710 2219215 : batch.nr++;
711 2219215 : total_size += next_size;
712 2219215 : batch.total_data_size += next->data_len;
713 2219215 : curr = next;
714 : }
715 :
716 653313 : if (batch.nr == 1) {
717 552594 : first_key.objectid = node->inode_id;
718 552594 : first_key.type = BTRFS_DIR_INDEX_KEY;
719 552594 : first_key.offset = first_item->index;
720 552594 : batch.keys = &first_key;
721 552594 : batch.data_sizes = &first_data_size;
722 : } else {
723 100719 : struct btrfs_key *ins_keys;
724 100719 : u32 *ins_sizes;
725 100719 : int i = 0;
726 :
727 100719 : ins_data = kmalloc(batch.nr * sizeof(u32) +
728 : batch.nr * sizeof(struct btrfs_key), GFP_NOFS);
729 100716 : if (!ins_data) {
730 0 : ret = -ENOMEM;
731 0 : goto out;
732 : }
733 100716 : ins_sizes = (u32 *)ins_data;
734 100716 : ins_keys = (struct btrfs_key *)(ins_data + batch.nr * sizeof(u32));
735 100716 : batch.keys = ins_keys;
736 100716 : batch.data_sizes = ins_sizes;
737 2421035 : list_for_each_entry(curr, &item_list, tree_list) {
738 2320319 : ins_keys[i].objectid = node->inode_id;
739 2320319 : ins_keys[i].type = BTRFS_DIR_INDEX_KEY;
740 2320319 : ins_keys[i].offset = curr->index;
741 2320319 : ins_sizes[i] = curr->data_len;
742 2320319 : i++;
743 : }
744 : }
745 :
746 653310 : ret = btrfs_insert_empty_items(trans, root, path, &batch);
747 653321 : if (ret)
748 5 : goto out;
749 :
750 3527197 : list_for_each_entry(curr, &item_list, tree_list) {
751 2873874 : char *data_ptr;
752 :
753 2873874 : data_ptr = btrfs_item_ptr(path->nodes[0], path->slots[0], char);
754 2873875 : write_extent_buffer(path->nodes[0], &curr->data,
755 2873875 : (unsigned long)data_ptr, curr->data_len);
756 2873881 : path->slots[0]++;
757 : }
758 :
759 : /*
760 : * Now release our path before releasing the delayed items and their
761 : * metadata reservations, so that we don't block other tasks for more
762 : * time than needed.
763 : */
764 653323 : btrfs_release_path(path);
765 :
766 653326 : ASSERT(node->index_item_leaves > 0);
767 :
768 : /*
769 : * For normal operations we will batch an entire leaf's worth of delayed
770 : * items, so if there are more items to process we can decrement
771 : * index_item_leaves by 1 as we inserted 1 leaf's worth of items.
772 : *
773 : * However for log replay we may not have inserted an entire leaf's
774 : * worth of items, we may have not had continuous items, so decrementing
775 : * here would mess up the index_item_leaves accounting. For this case
776 : * only clean up the accounting when there are no items left.
777 : */
778 653326 : if (next && !continuous_keys_only) {
779 : /*
780 : * We inserted one batch of items into a leaf a there are more
781 : * items to flush in a future batch, now release one unit of
782 : * metadata space from the delayed block reserve, corresponding
783 : * the leaf we just flushed to.
784 : */
785 286 : btrfs_delayed_item_release_leaves(node, 1);
786 286 : node->index_item_leaves--;
787 653040 : } else if (!next) {
788 : /*
789 : * There are no more items to insert. We can have a number of
790 : * reserved leaves > 1 here - this happens when many dir index
791 : * items are added and then removed before they are flushed (file
792 : * names with a very short life, never span a transaction). So
793 : * release all remaining leaves.
794 : */
795 653035 : btrfs_delayed_item_release_leaves(node, node->index_item_leaves);
796 653046 : node->index_item_leaves = 0;
797 : }
798 :
799 3527341 : list_for_each_entry_safe(curr, next, &item_list, tree_list) {
800 2874025 : list_del(&curr->tree_list);
801 2873928 : btrfs_release_delayed_item(curr);
802 : }
803 653316 : out:
804 653321 : kfree(ins_data);
805 653314 : return ret;
806 : }
807 :
808 7512589 : static int btrfs_insert_delayed_items(struct btrfs_trans_handle *trans,
809 : struct btrfs_path *path,
810 : struct btrfs_root *root,
811 : struct btrfs_delayed_node *node)
812 : {
813 7512589 : int ret = 0;
814 :
815 8165775 : while (ret == 0) {
816 8165770 : struct btrfs_delayed_item *curr;
817 :
818 8165770 : mutex_lock(&node->mutex);
819 8166539 : curr = __btrfs_first_delayed_insertion_item(node);
820 8166539 : if (!curr) {
821 7513202 : mutex_unlock(&node->mutex);
822 7513202 : break;
823 : }
824 653337 : ret = btrfs_insert_delayed_item(trans, root, path, curr);
825 653303 : mutex_unlock(&node->mutex);
826 : }
827 :
828 7512195 : return ret;
829 : }
830 :
831 360309 : static int btrfs_batch_delete_items(struct btrfs_trans_handle *trans,
832 : struct btrfs_root *root,
833 : struct btrfs_path *path,
834 : struct btrfs_delayed_item *item)
835 : {
836 360309 : const u64 ino = item->delayed_node->inode_id;
837 360309 : struct btrfs_fs_info *fs_info = root->fs_info;
838 360309 : struct btrfs_delayed_item *curr, *next;
839 360309 : struct extent_buffer *leaf = path->nodes[0];
840 360309 : LIST_HEAD(batch_list);
841 360309 : int nitems, slot, last_slot;
842 360309 : int ret;
843 360309 : u64 total_reserved_size = item->bytes_reserved;
844 :
845 360309 : ASSERT(leaf != NULL);
846 :
847 360309 : slot = path->slots[0];
848 360309 : last_slot = btrfs_header_nritems(leaf) - 1;
849 : /*
850 : * Our caller always gives us a path pointing to an existing item, so
851 : * this can not happen.
852 : */
853 360309 : ASSERT(slot <= last_slot);
854 360309 : if (WARN_ON(slot > last_slot))
855 : return -ENOENT;
856 :
857 360309 : nitems = 1;
858 360309 : curr = item;
859 360309 : list_add_tail(&curr->tree_list, &batch_list);
860 :
861 : /*
862 : * Keep checking if the next delayed item matches the next item in the
863 : * leaf - if so, we can add it to the batch of items to delete from the
864 : * leaf.
865 : */
866 1225233 : while (slot < last_slot) {
867 1218464 : struct btrfs_key key;
868 :
869 1218464 : next = __btrfs_next_delayed_item(curr);
870 898546 : if (!next)
871 : break;
872 :
873 898546 : slot++;
874 898546 : btrfs_item_key_to_cpu(leaf, &key, slot);
875 898555 : if (key.objectid != ino ||
876 898555 : key.type != BTRFS_DIR_INDEX_KEY ||
877 898555 : key.offset != next->index)
878 : break;
879 864939 : nitems++;
880 864939 : curr = next;
881 864939 : list_add_tail(&curr->tree_list, &batch_list);
882 864924 : total_reserved_size += curr->bytes_reserved;
883 : }
884 :
885 360309 : ret = btrfs_del_items(trans, root, path, path->slots[0], nitems);
886 360307 : if (ret)
887 : return ret;
888 :
889 : /* In case of BTRFS_FS_LOG_RECOVERING items won't have reserved space */
890 360307 : if (total_reserved_size > 0) {
891 : /*
892 : * Check btrfs_delayed_item_reserve_metadata() to see why we
893 : * don't need to release/reserve qgroup space.
894 : */
895 360287 : trace_btrfs_space_reservation(fs_info, "delayed_item", ino,
896 : total_reserved_size, 0);
897 360285 : btrfs_block_rsv_release(fs_info, &fs_info->delayed_block_rsv,
898 : total_reserved_size, NULL);
899 : }
900 :
901 1585541 : list_for_each_entry_safe(curr, next, &batch_list, tree_list) {
902 1225234 : list_del(&curr->tree_list);
903 1225207 : btrfs_release_delayed_item(curr);
904 : }
905 :
906 : return 0;
907 : }
908 :
909 7511708 : static int btrfs_delete_delayed_items(struct btrfs_trans_handle *trans,
910 : struct btrfs_path *path,
911 : struct btrfs_root *root,
912 : struct btrfs_delayed_node *node)
913 : {
914 7511708 : struct btrfs_key key;
915 7511708 : int ret = 0;
916 :
917 7511708 : key.objectid = node->inode_id;
918 7511708 : key.type = BTRFS_DIR_INDEX_KEY;
919 :
920 7871809 : while (ret == 0) {
921 7871809 : struct btrfs_delayed_item *item;
922 :
923 7871809 : mutex_lock(&node->mutex);
924 7871403 : item = __btrfs_first_delayed_deletion_item(node);
925 7871403 : if (!item) {
926 7511095 : mutex_unlock(&node->mutex);
927 7511095 : break;
928 : }
929 :
930 360308 : key.offset = item->index;
931 360308 : ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
932 360309 : if (ret > 0) {
933 : /*
934 : * There's no matching item in the leaf. This means we
935 : * have already deleted this item in a past run of the
936 : * delayed items. We ignore errors when running delayed
937 : * items from an async context, through a work queue job
938 : * running btrfs_async_run_delayed_root(), and don't
939 : * release delayed items that failed to complete. This
940 : * is because we will retry later, and at transaction
941 : * commit time we always run delayed items and will
942 : * then deal with errors if they fail to run again.
943 : *
944 : * So just release delayed items for which we can't find
945 : * an item in the tree, and move to the next item.
946 : */
947 0 : btrfs_release_path(path);
948 0 : btrfs_release_delayed_item(item);
949 0 : ret = 0;
950 360309 : } else if (ret == 0) {
951 360309 : ret = btrfs_batch_delete_items(trans, root, path, item);
952 360307 : btrfs_release_path(path);
953 : }
954 :
955 : /*
956 : * We unlock and relock on each iteration, this is to prevent
957 : * blocking other tasks for too long while we are being run from
958 : * the async context (work queue job). Those tasks are typically
959 : * running system calls like creat/mkdir/rename/unlink/etc which
960 : * need to add delayed items to this delayed node.
961 : */
962 360308 : mutex_unlock(&node->mutex);
963 : }
964 :
965 7510662 : return ret;
966 : }
967 :
968 8590176 : static void btrfs_release_delayed_inode(struct btrfs_delayed_node *delayed_node)
969 : {
970 8590176 : struct btrfs_delayed_root *delayed_root;
971 :
972 17180352 : if (delayed_node &&
973 8590176 : test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
974 8590140 : BUG_ON(!delayed_node->root);
975 8590140 : clear_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
976 8590604 : delayed_node->count--;
977 :
978 8590604 : delayed_root = delayed_node->root->fs_info->delayed_root;
979 8590604 : finish_one_item(delayed_root);
980 : }
981 8589664 : }
982 :
983 9960153 : static void btrfs_release_delayed_iref(struct btrfs_delayed_node *delayed_node)
984 : {
985 :
986 9960153 : if (test_and_clear_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags)) {
987 1337224 : struct btrfs_delayed_root *delayed_root;
988 :
989 1337224 : ASSERT(delayed_node->root);
990 1337224 : delayed_node->count--;
991 :
992 1337224 : delayed_root = delayed_node->root->fs_info->delayed_root;
993 1337224 : finish_one_item(delayed_root);
994 : }
995 9961350 : }
996 :
997 8588090 : static int __btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
998 : struct btrfs_root *root,
999 : struct btrfs_path *path,
1000 : struct btrfs_delayed_node *node)
1001 : {
1002 8588090 : struct btrfs_fs_info *fs_info = root->fs_info;
1003 8588090 : struct btrfs_key key;
1004 8588090 : struct btrfs_inode_item *inode_item;
1005 8588090 : struct extent_buffer *leaf;
1006 8588090 : int mod;
1007 8588090 : int ret;
1008 :
1009 8588090 : key.objectid = node->inode_id;
1010 8588090 : key.type = BTRFS_INODE_ITEM_KEY;
1011 8588090 : key.offset = 0;
1012 :
1013 17176180 : if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1014 : mod = -1;
1015 : else
1016 7249655 : mod = 1;
1017 :
1018 8588090 : ret = btrfs_lookup_inode(trans, root, path, &key, mod);
1019 8587415 : if (ret > 0)
1020 : ret = -ENOENT;
1021 8587530 : if (ret < 0)
1022 6 : goto out;
1023 :
1024 8587409 : leaf = path->nodes[0];
1025 8587409 : inode_item = btrfs_item_ptr(leaf, path->slots[0],
1026 : struct btrfs_inode_item);
1027 8586085 : write_extent_buffer(leaf, &node->inode_item, (unsigned long)inode_item,
1028 : sizeof(struct btrfs_inode_item));
1029 8585851 : btrfs_mark_buffer_dirty(leaf);
1030 :
1031 8588999 : if (!test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1032 7251773 : goto out;
1033 :
1034 1337226 : path->slots[0]++;
1035 1337226 : if (path->slots[0] >= btrfs_header_nritems(leaf))
1036 14197 : goto search;
1037 1323029 : again:
1038 1337226 : btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1039 1337227 : if (key.objectid != node->inode_id)
1040 0 : goto out;
1041 :
1042 1337227 : if (key.type != BTRFS_INODE_REF_KEY &&
1043 : key.type != BTRFS_INODE_EXTREF_KEY)
1044 0 : goto out;
1045 :
1046 : /*
1047 : * Delayed iref deletion is for the inode who has only one link,
1048 : * so there is only one iref. The case that several irefs are
1049 : * in the same item doesn't exist.
1050 : */
1051 1337227 : ret = btrfs_del_item(trans, root, path);
1052 8588997 : out:
1053 8588997 : btrfs_release_delayed_iref(node);
1054 8589447 : btrfs_release_path(path);
1055 8589972 : err_out:
1056 8589972 : btrfs_delayed_inode_release_metadata(fs_info, node, (ret < 0));
1057 8590301 : btrfs_release_delayed_inode(node);
1058 :
1059 : /*
1060 : * If we fail to update the delayed inode we need to abort the
1061 : * transaction, because we could leave the inode with the improper
1062 : * counts behind.
1063 : */
1064 8589634 : if (ret && ret != -ENOENT)
1065 6 : btrfs_abort_transaction(trans, ret);
1066 :
1067 8589634 : return ret;
1068 :
1069 : search:
1070 14197 : btrfs_release_path(path);
1071 :
1072 14197 : key.type = BTRFS_INODE_EXTREF_KEY;
1073 14197 : key.offset = -1;
1074 :
1075 14197 : ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1076 14197 : if (ret < 0)
1077 0 : goto err_out;
1078 14197 : ASSERT(ret);
1079 :
1080 14197 : ret = 0;
1081 14197 : leaf = path->nodes[0];
1082 14197 : path->slots[0]--;
1083 14197 : goto again;
1084 : }
1085 :
1086 7507225 : static inline int btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1087 : struct btrfs_root *root,
1088 : struct btrfs_path *path,
1089 : struct btrfs_delayed_node *node)
1090 : {
1091 7507225 : int ret;
1092 :
1093 7507225 : mutex_lock(&node->mutex);
1094 7512714 : if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &node->flags)) {
1095 283681 : mutex_unlock(&node->mutex);
1096 283681 : return 0;
1097 : }
1098 :
1099 7229033 : ret = __btrfs_update_delayed_inode(trans, root, path, node);
1100 7228949 : mutex_unlock(&node->mutex);
1101 7228949 : return ret;
1102 : }
1103 :
1104 : static inline int
1105 7512748 : __btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1106 : struct btrfs_path *path,
1107 : struct btrfs_delayed_node *node)
1108 : {
1109 7512748 : int ret;
1110 :
1111 7512748 : ret = btrfs_insert_delayed_items(trans, path, node->root, node);
1112 7512085 : if (ret)
1113 : return ret;
1114 :
1115 7512141 : ret = btrfs_delete_delayed_items(trans, path, node->root, node);
1116 7510359 : if (ret)
1117 : return ret;
1118 :
1119 7508663 : ret = btrfs_update_delayed_inode(trans, node->root, path, node);
1120 7508663 : return ret;
1121 : }
1122 :
1123 : /*
1124 : * Called when committing the transaction.
1125 : * Returns 0 on success.
1126 : * Returns < 0 on error and returns with an aborted transaction with any
1127 : * outstanding delayed items cleaned up.
1128 : */
1129 2496640 : static int __btrfs_run_delayed_items(struct btrfs_trans_handle *trans, int nr)
1130 : {
1131 2496640 : struct btrfs_fs_info *fs_info = trans->fs_info;
1132 2496640 : struct btrfs_delayed_root *delayed_root;
1133 2496640 : struct btrfs_delayed_node *curr_node, *prev_node;
1134 2496640 : struct btrfs_path *path;
1135 2496640 : struct btrfs_block_rsv *block_rsv;
1136 2496640 : int ret = 0;
1137 2496640 : bool count = (nr > 0);
1138 :
1139 2496640 : if (TRANS_ABORTED(trans))
1140 : return -EIO;
1141 :
1142 2496640 : path = btrfs_alloc_path();
1143 2496637 : if (!path)
1144 : return -ENOMEM;
1145 :
1146 2496637 : block_rsv = trans->block_rsv;
1147 2496637 : trans->block_rsv = &fs_info->delayed_block_rsv;
1148 :
1149 2496637 : delayed_root = fs_info->delayed_root;
1150 :
1151 2496637 : curr_node = btrfs_first_delayed_node(delayed_root);
1152 5088071 : while (curr_node && (!count || nr--)) {
1153 2591446 : ret = __btrfs_commit_inode_delayed_items(trans, path,
1154 : curr_node);
1155 2591438 : if (ret) {
1156 11 : btrfs_release_delayed_node(curr_node);
1157 11 : curr_node = NULL;
1158 11 : btrfs_abort_transaction(trans, ret);
1159 11 : break;
1160 : }
1161 :
1162 2591427 : prev_node = curr_node;
1163 2591427 : curr_node = btrfs_next_delayed_node(curr_node);
1164 2591426 : btrfs_release_delayed_node(prev_node);
1165 : }
1166 :
1167 2496639 : if (curr_node)
1168 18274 : btrfs_release_delayed_node(curr_node);
1169 2496639 : btrfs_free_path(path);
1170 2496630 : trans->block_rsv = block_rsv;
1171 :
1172 2496630 : return ret;
1173 : }
1174 :
1175 611753 : int btrfs_run_delayed_items(struct btrfs_trans_handle *trans)
1176 : {
1177 611753 : return __btrfs_run_delayed_items(trans, -1);
1178 : }
1179 :
1180 1884887 : int btrfs_run_delayed_items_nr(struct btrfs_trans_handle *trans, int nr)
1181 : {
1182 1884887 : return __btrfs_run_delayed_items(trans, nr);
1183 : }
1184 :
1185 104 : int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1186 : struct btrfs_inode *inode)
1187 : {
1188 104 : struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1189 104 : struct btrfs_path *path;
1190 104 : struct btrfs_block_rsv *block_rsv;
1191 104 : int ret;
1192 :
1193 104 : if (!delayed_node)
1194 : return 0;
1195 :
1196 58 : mutex_lock(&delayed_node->mutex);
1197 58 : if (!delayed_node->count) {
1198 0 : mutex_unlock(&delayed_node->mutex);
1199 0 : btrfs_release_delayed_node(delayed_node);
1200 0 : return 0;
1201 : }
1202 58 : mutex_unlock(&delayed_node->mutex);
1203 :
1204 58 : path = btrfs_alloc_path();
1205 58 : if (!path) {
1206 0 : btrfs_release_delayed_node(delayed_node);
1207 0 : return -ENOMEM;
1208 : }
1209 :
1210 58 : block_rsv = trans->block_rsv;
1211 58 : trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv;
1212 :
1213 58 : ret = __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1214 :
1215 58 : btrfs_release_delayed_node(delayed_node);
1216 58 : btrfs_free_path(path);
1217 58 : trans->block_rsv = block_rsv;
1218 :
1219 58 : return ret;
1220 : }
1221 :
1222 1773671 : int btrfs_commit_inode_delayed_inode(struct btrfs_inode *inode)
1223 : {
1224 1773671 : struct btrfs_fs_info *fs_info = inode->root->fs_info;
1225 1773671 : struct btrfs_trans_handle *trans;
1226 1773671 : struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1227 1774168 : struct btrfs_path *path;
1228 1774168 : struct btrfs_block_rsv *block_rsv;
1229 1774168 : int ret;
1230 :
1231 1774168 : if (!delayed_node)
1232 : return 0;
1233 :
1234 1371435 : mutex_lock(&delayed_node->mutex);
1235 1371595 : if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1236 11163 : mutex_unlock(&delayed_node->mutex);
1237 11163 : btrfs_release_delayed_node(delayed_node);
1238 11163 : return 0;
1239 : }
1240 1360432 : mutex_unlock(&delayed_node->mutex);
1241 :
1242 1360369 : trans = btrfs_join_transaction(delayed_node->root);
1243 1360594 : if (IS_ERR(trans)) {
1244 0 : ret = PTR_ERR(trans);
1245 0 : goto out;
1246 : }
1247 :
1248 1360594 : path = btrfs_alloc_path();
1249 1360517 : if (!path) {
1250 0 : ret = -ENOMEM;
1251 0 : goto trans_out;
1252 : }
1253 :
1254 1360517 : block_rsv = trans->block_rsv;
1255 1360517 : trans->block_rsv = &fs_info->delayed_block_rsv;
1256 :
1257 1360517 : mutex_lock(&delayed_node->mutex);
1258 2720856 : if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags))
1259 1360362 : ret = __btrfs_update_delayed_inode(trans, delayed_node->root,
1260 : path, delayed_node);
1261 : else
1262 : ret = 0;
1263 1360646 : mutex_unlock(&delayed_node->mutex);
1264 :
1265 1360602 : btrfs_free_path(path);
1266 1360589 : trans->block_rsv = block_rsv;
1267 1360589 : trans_out:
1268 1360589 : btrfs_end_transaction(trans);
1269 1360593 : btrfs_btree_balance_dirty(fs_info);
1270 1360577 : out:
1271 1360577 : btrfs_release_delayed_node(delayed_node);
1272 :
1273 1360577 : return ret;
1274 : }
1275 :
1276 3865380 : void btrfs_remove_delayed_node(struct btrfs_inode *inode)
1277 : {
1278 3865380 : struct btrfs_delayed_node *delayed_node;
1279 :
1280 3865380 : delayed_node = READ_ONCE(inode->delayed_node);
1281 3865380 : if (!delayed_node)
1282 : return;
1283 :
1284 1935979 : inode->delayed_node = NULL;
1285 1935979 : btrfs_release_delayed_node(delayed_node);
1286 : }
1287 :
1288 : struct btrfs_async_delayed_work {
1289 : struct btrfs_delayed_root *delayed_root;
1290 : int nr;
1291 : struct btrfs_work work;
1292 : };
1293 :
1294 747538 : static void btrfs_async_run_delayed_root(struct btrfs_work *work)
1295 : {
1296 747538 : struct btrfs_async_delayed_work *async_work;
1297 747538 : struct btrfs_delayed_root *delayed_root;
1298 747538 : struct btrfs_trans_handle *trans;
1299 747538 : struct btrfs_path *path;
1300 747538 : struct btrfs_delayed_node *delayed_node = NULL;
1301 747538 : struct btrfs_root *root;
1302 747538 : struct btrfs_block_rsv *block_rsv;
1303 747538 : int total_done = 0;
1304 :
1305 747538 : async_work = container_of(work, struct btrfs_async_delayed_work, work);
1306 747538 : delayed_root = async_work->delayed_root;
1307 :
1308 747538 : path = btrfs_alloc_path();
1309 747530 : if (!path)
1310 0 : goto out;
1311 :
1312 5402012 : do {
1313 5402012 : if (atomic_read(&delayed_root->items) <
1314 : BTRFS_DELAYED_BACKGROUND / 2)
1315 : break;
1316 :
1317 4962197 : delayed_node = btrfs_first_prepared_delayed_node(delayed_root);
1318 4962193 : if (!delayed_node)
1319 : break;
1320 :
1321 4922356 : root = delayed_node->root;
1322 :
1323 4922356 : trans = btrfs_join_transaction(root);
1324 4922302 : if (IS_ERR(trans)) {
1325 0 : btrfs_release_path(path);
1326 0 : btrfs_release_prepared_delayed_node(delayed_node);
1327 0 : total_done++;
1328 0 : continue;
1329 : }
1330 :
1331 4922302 : block_rsv = trans->block_rsv;
1332 4922302 : trans->block_rsv = &root->fs_info->delayed_block_rsv;
1333 :
1334 4922302 : __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1335 :
1336 4921122 : trans->block_rsv = block_rsv;
1337 4921122 : btrfs_end_transaction(trans);
1338 4920251 : btrfs_btree_balance_dirty_nodelay(root->fs_info);
1339 :
1340 4916735 : btrfs_release_path(path);
1341 4919866 : btrfs_release_prepared_delayed_node(delayed_node);
1342 4922242 : total_done++;
1343 :
1344 285246 : } while ((async_work->nr == 0 && total_done < BTRFS_DELAYED_WRITEBACK)
1345 4922630 : || total_done < async_work->nr);
1346 :
1347 747412 : btrfs_free_path(path);
1348 747307 : out:
1349 747307 : wake_up(&delayed_root->wait);
1350 747609 : kfree(async_work);
1351 747376 : }
1352 :
1353 :
1354 746447 : static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root,
1355 : struct btrfs_fs_info *fs_info, int nr)
1356 : {
1357 746447 : struct btrfs_async_delayed_work *async_work;
1358 :
1359 746447 : async_work = kmalloc(sizeof(*async_work), GFP_NOFS);
1360 746197 : if (!async_work)
1361 : return -ENOMEM;
1362 :
1363 746197 : async_work->delayed_root = delayed_root;
1364 746197 : btrfs_init_work(&async_work->work, btrfs_async_run_delayed_root, NULL,
1365 : NULL);
1366 745959 : async_work->nr = nr;
1367 :
1368 745959 : btrfs_queue_work(fs_info->delayed_workers, &async_work->work);
1369 745959 : return 0;
1370 : }
1371 :
1372 203035 : void btrfs_assert_delayed_root_empty(struct btrfs_fs_info *fs_info)
1373 : {
1374 203035 : WARN_ON(btrfs_first_delayed_node(fs_info->delayed_root));
1375 203035 : }
1376 :
1377 : static int could_end_wait(struct btrfs_delayed_root *delayed_root, int seq)
1378 : {
1379 36415 : int val = atomic_read(&delayed_root->items_seq);
1380 :
1381 36415 : if (val < seq || val >= seq + BTRFS_DELAYED_BATCH)
1382 : return 1;
1383 :
1384 26841 : if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1385 : return 1;
1386 :
1387 : return 0;
1388 : }
1389 :
1390 16957381 : void btrfs_balance_delayed_items(struct btrfs_fs_info *fs_info)
1391 : {
1392 16957381 : struct btrfs_delayed_root *delayed_root = fs_info->delayed_root;
1393 :
1394 19221179 : if ((atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND) ||
1395 2264679 : btrfs_workqueue_normal_congested(fs_info->delayed_workers))
1396 16210073 : return;
1397 :
1398 746427 : if (atomic_read(&delayed_root->items) >= BTRFS_DELAYED_WRITEBACK) {
1399 9556 : int seq;
1400 9556 : int ret;
1401 :
1402 9556 : seq = atomic_read(&delayed_root->items_seq);
1403 :
1404 9556 : ret = btrfs_wq_run_delayed_node(delayed_root, fs_info, 0);
1405 9574 : if (ret)
1406 : return;
1407 :
1408 45867 : wait_event_interruptible(delayed_root->wait,
1409 : could_end_wait(delayed_root, seq));
1410 9574 : return;
1411 : }
1412 :
1413 736871 : btrfs_wq_run_delayed_node(delayed_root, fs_info, BTRFS_DELAYED_BATCH);
1414 : }
1415 :
1416 : /* Will return 0 or -ENOMEM */
1417 3378246 : int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans,
1418 : const char *name, int name_len,
1419 : struct btrfs_inode *dir,
1420 : struct btrfs_disk_key *disk_key, u8 flags,
1421 : u64 index)
1422 : {
1423 3378246 : struct btrfs_fs_info *fs_info = trans->fs_info;
1424 3378246 : const unsigned int leaf_data_size = BTRFS_LEAF_DATA_SIZE(fs_info);
1425 3378246 : struct btrfs_delayed_node *delayed_node;
1426 3378246 : struct btrfs_delayed_item *delayed_item;
1427 3378246 : struct btrfs_dir_item *dir_item;
1428 3378246 : bool reserve_leaf_space;
1429 3378246 : u32 data_len;
1430 3378246 : int ret;
1431 :
1432 3378246 : delayed_node = btrfs_get_or_create_delayed_node(dir);
1433 3378180 : if (IS_ERR(delayed_node))
1434 0 : return PTR_ERR(delayed_node);
1435 :
1436 3378180 : delayed_item = btrfs_alloc_delayed_item(sizeof(*dir_item) + name_len,
1437 : delayed_node,
1438 : BTRFS_DELAYED_INSERTION_ITEM);
1439 3377876 : if (!delayed_item) {
1440 0 : ret = -ENOMEM;
1441 0 : goto release_node;
1442 : }
1443 :
1444 3377876 : delayed_item->index = index;
1445 :
1446 3377876 : dir_item = (struct btrfs_dir_item *)delayed_item->data;
1447 3377876 : dir_item->location = *disk_key;
1448 3377876 : btrfs_set_stack_dir_transid(dir_item, trans->transid);
1449 3377876 : btrfs_set_stack_dir_data_len(dir_item, 0);
1450 3377876 : btrfs_set_stack_dir_name_len(dir_item, name_len);
1451 3377876 : btrfs_set_stack_dir_flags(dir_item, flags);
1452 6755752 : memcpy((char *)(dir_item + 1), name, name_len);
1453 :
1454 3377876 : data_len = delayed_item->data_len + sizeof(struct btrfs_item);
1455 :
1456 3377876 : mutex_lock(&delayed_node->mutex);
1457 :
1458 3378056 : if (delayed_node->index_item_leaves == 0 ||
1459 2390771 : delayed_node->curr_index_batch_size + data_len > leaf_data_size) {
1460 987616 : delayed_node->curr_index_batch_size = data_len;
1461 987616 : reserve_leaf_space = true;
1462 : } else {
1463 2390440 : delayed_node->curr_index_batch_size += data_len;
1464 2390440 : reserve_leaf_space = false;
1465 : }
1466 :
1467 3378056 : if (reserve_leaf_space) {
1468 987616 : ret = btrfs_delayed_item_reserve_metadata(trans, delayed_item);
1469 : /*
1470 : * Space was reserved for a dir index item insertion when we
1471 : * started the transaction, so getting a failure here should be
1472 : * impossible.
1473 : */
1474 987596 : if (WARN_ON(ret)) {
1475 0 : mutex_unlock(&delayed_node->mutex);
1476 0 : btrfs_release_delayed_item(delayed_item);
1477 0 : goto release_node;
1478 : }
1479 :
1480 987596 : delayed_node->index_item_leaves++;
1481 2390440 : } else if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
1482 2390352 : const u64 bytes = btrfs_calc_insert_metadata_size(fs_info, 1);
1483 :
1484 : /*
1485 : * Adding the new dir index item does not require touching another
1486 : * leaf, so we can release 1 unit of metadata that was previously
1487 : * reserved when starting the transaction. This applies only to
1488 : * the case where we had a transaction start and excludes the
1489 : * transaction join case (when replaying log trees).
1490 : */
1491 2390352 : trace_btrfs_space_reservation(fs_info, "transaction",
1492 : trans->transid, bytes, 0);
1493 2390319 : btrfs_block_rsv_release(fs_info, trans->block_rsv, bytes, NULL);
1494 2390591 : ASSERT(trans->bytes_reserved >= bytes);
1495 2390591 : trans->bytes_reserved -= bytes;
1496 : }
1497 :
1498 3378275 : ret = __btrfs_add_delayed_item(delayed_node, delayed_item);
1499 3378160 : if (unlikely(ret)) {
1500 0 : btrfs_err(trans->fs_info,
1501 : "err add delayed dir index item(name: %.*s) into the insertion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
1502 : name_len, name, delayed_node->root->root_key.objectid,
1503 : delayed_node->inode_id, ret);
1504 0 : BUG();
1505 : }
1506 3378160 : mutex_unlock(&delayed_node->mutex);
1507 :
1508 3378195 : release_node:
1509 3378195 : btrfs_release_delayed_node(delayed_node);
1510 3378195 : return ret;
1511 : }
1512 :
1513 1885586 : static int btrfs_delete_delayed_insertion_item(struct btrfs_fs_info *fs_info,
1514 : struct btrfs_delayed_node *node,
1515 : u64 index)
1516 : {
1517 1885586 : struct btrfs_delayed_item *item;
1518 :
1519 1885586 : mutex_lock(&node->mutex);
1520 1885582 : item = __btrfs_lookup_delayed_item(&node->ins_root.rb_root, index);
1521 1885582 : if (!item) {
1522 1381532 : mutex_unlock(&node->mutex);
1523 1381532 : return 1;
1524 : }
1525 :
1526 : /*
1527 : * For delayed items to insert, we track reserved metadata bytes based
1528 : * on the number of leaves that we will use.
1529 : * See btrfs_insert_delayed_dir_index() and
1530 : * btrfs_delayed_item_reserve_metadata()).
1531 : */
1532 504050 : ASSERT(item->bytes_reserved == 0);
1533 504050 : ASSERT(node->index_item_leaves > 0);
1534 :
1535 : /*
1536 : * If there's only one leaf reserved, we can decrement this item from the
1537 : * current batch, otherwise we can not because we don't know which leaf
1538 : * it belongs to. With the current limit on delayed items, we rarely
1539 : * accumulate enough dir index items to fill more than one leaf (even
1540 : * when using a leaf size of 4K).
1541 : */
1542 504050 : if (node->index_item_leaves == 1) {
1543 504050 : const u32 data_len = item->data_len + sizeof(struct btrfs_item);
1544 :
1545 504050 : ASSERT(node->curr_index_batch_size >= data_len);
1546 504050 : node->curr_index_batch_size -= data_len;
1547 : }
1548 :
1549 504050 : btrfs_release_delayed_item(item);
1550 :
1551 : /* If we now have no more dir index items, we can release all leaves. */
1552 504050 : if (RB_EMPTY_ROOT(&node->ins_root.rb_root)) {
1553 334283 : btrfs_delayed_item_release_leaves(node, node->index_item_leaves);
1554 334283 : node->index_item_leaves = 0;
1555 : }
1556 :
1557 504050 : mutex_unlock(&node->mutex);
1558 504050 : return 0;
1559 : }
1560 :
1561 1885587 : int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans,
1562 : struct btrfs_inode *dir, u64 index)
1563 : {
1564 1885587 : struct btrfs_delayed_node *node;
1565 1885587 : struct btrfs_delayed_item *item;
1566 1885587 : int ret;
1567 :
1568 1885587 : node = btrfs_get_or_create_delayed_node(dir);
1569 1885588 : if (IS_ERR(node))
1570 0 : return PTR_ERR(node);
1571 :
1572 1885588 : ret = btrfs_delete_delayed_insertion_item(trans->fs_info, node, index);
1573 1885584 : if (!ret)
1574 504050 : goto end;
1575 :
1576 1381534 : item = btrfs_alloc_delayed_item(0, node, BTRFS_DELAYED_DELETION_ITEM);
1577 1381504 : if (!item) {
1578 0 : ret = -ENOMEM;
1579 0 : goto end;
1580 : }
1581 :
1582 1381504 : item->index = index;
1583 :
1584 1381504 : ret = btrfs_delayed_item_reserve_metadata(trans, item);
1585 : /*
1586 : * we have reserved enough space when we start a new transaction,
1587 : * so reserving metadata failure is impossible.
1588 : */
1589 1381525 : if (ret < 0) {
1590 0 : btrfs_err(trans->fs_info,
1591 : "metadata reservation failed for delayed dir item deltiona, should have been reserved");
1592 0 : btrfs_release_delayed_item(item);
1593 0 : goto end;
1594 : }
1595 :
1596 1381525 : mutex_lock(&node->mutex);
1597 1381536 : ret = __btrfs_add_delayed_item(node, item);
1598 1381531 : if (unlikely(ret)) {
1599 0 : btrfs_err(trans->fs_info,
1600 : "err add delayed dir index item(index: %llu) into the deletion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
1601 : index, node->root->root_key.objectid,
1602 : node->inode_id, ret);
1603 0 : btrfs_delayed_item_release_metadata(dir->root, item);
1604 0 : btrfs_release_delayed_item(item);
1605 : }
1606 1381531 : mutex_unlock(&node->mutex);
1607 1885584 : end:
1608 1885584 : btrfs_release_delayed_node(node);
1609 1885584 : return ret;
1610 : }
1611 :
1612 6213 : int btrfs_inode_delayed_dir_index_count(struct btrfs_inode *inode)
1613 : {
1614 6213 : struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1615 :
1616 6213 : if (!delayed_node)
1617 : return -ENOENT;
1618 :
1619 : /*
1620 : * Since we have held i_mutex of this directory, it is impossible that
1621 : * a new directory index is added into the delayed node and index_cnt
1622 : * is updated now. So we needn't lock the delayed node.
1623 : */
1624 1204 : if (!delayed_node->index_cnt) {
1625 1204 : btrfs_release_delayed_node(delayed_node);
1626 1204 : return -EINVAL;
1627 : }
1628 :
1629 0 : inode->index_cnt = delayed_node->index_cnt;
1630 0 : btrfs_release_delayed_node(delayed_node);
1631 0 : return 0;
1632 : }
1633 :
1634 2773684 : bool btrfs_readdir_get_delayed_items(struct inode *inode,
1635 : struct list_head *ins_list,
1636 : struct list_head *del_list)
1637 : {
1638 2773684 : struct btrfs_delayed_node *delayed_node;
1639 2773684 : struct btrfs_delayed_item *item;
1640 :
1641 2773684 : delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
1642 2774015 : if (!delayed_node)
1643 : return false;
1644 :
1645 : /*
1646 : * We can only do one readdir with delayed items at a time because of
1647 : * item->readdir_list.
1648 : */
1649 487866 : btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_SHARED);
1650 487858 : btrfs_inode_lock(BTRFS_I(inode), 0);
1651 :
1652 487856 : mutex_lock(&delayed_node->mutex);
1653 487863 : item = __btrfs_first_delayed_insertion_item(delayed_node);
1654 17708760 : while (item) {
1655 17419135 : refcount_inc(&item->refs);
1656 17419135 : list_add_tail(&item->readdir_list, ins_list);
1657 17419133 : item = __btrfs_next_delayed_item(item);
1658 : }
1659 :
1660 487863 : item = __btrfs_first_delayed_deletion_item(delayed_node);
1661 525183 : while (item) {
1662 49469 : refcount_inc(&item->refs);
1663 49469 : list_add_tail(&item->readdir_list, del_list);
1664 49469 : item = __btrfs_next_delayed_item(item);
1665 : }
1666 487863 : mutex_unlock(&delayed_node->mutex);
1667 : /*
1668 : * This delayed node is still cached in the btrfs inode, so refs
1669 : * must be > 1 now, and we needn't check it is going to be freed
1670 : * or not.
1671 : *
1672 : * Besides that, this function is used to read dir, we do not
1673 : * insert/delete delayed items in this period. So we also needn't
1674 : * requeue or dequeue this delayed node.
1675 : */
1676 487865 : refcount_dec(&delayed_node->refs);
1677 :
1678 487865 : return true;
1679 : }
1680 :
1681 487869 : void btrfs_readdir_put_delayed_items(struct inode *inode,
1682 : struct list_head *ins_list,
1683 : struct list_head *del_list)
1684 : {
1685 487869 : struct btrfs_delayed_item *curr, *next;
1686 :
1687 8456543 : list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1688 7968677 : list_del(&curr->readdir_list);
1689 7968674 : if (refcount_dec_and_test(&curr->refs))
1690 100 : kfree(curr);
1691 : }
1692 :
1693 537335 : list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1694 49471 : list_del(&curr->readdir_list);
1695 49469 : if (refcount_dec_and_test(&curr->refs))
1696 12 : kfree(curr);
1697 : }
1698 :
1699 : /*
1700 : * The VFS is going to do up_read(), so we need to downgrade back to a
1701 : * read lock.
1702 : */
1703 487864 : downgrade_write(&inode->i_rwsem);
1704 487862 : }
1705 :
1706 2660161504 : int btrfs_should_delete_dir_index(struct list_head *del_list,
1707 : u64 index)
1708 : {
1709 2660161504 : struct btrfs_delayed_item *curr;
1710 2660161504 : int ret = 0;
1711 :
1712 2678715893 : list_for_each_entry(curr, del_list, readdir_list) {
1713 24279350 : if (curr->index > index)
1714 : break;
1715 18563708 : if (curr->index == index) {
1716 : ret = 1;
1717 : break;
1718 : }
1719 : }
1720 2660161504 : return ret;
1721 : }
1722 :
1723 : /*
1724 : * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree
1725 : *
1726 : */
1727 377972 : int btrfs_readdir_delayed_dir_index(struct dir_context *ctx,
1728 : struct list_head *ins_list)
1729 : {
1730 377972 : struct btrfs_dir_item *di;
1731 377972 : struct btrfs_delayed_item *curr, *next;
1732 377972 : struct btrfs_key location;
1733 377972 : char *name;
1734 377972 : int name_len;
1735 377972 : int over = 0;
1736 377972 : unsigned char d_type;
1737 :
1738 377972 : if (list_empty(ins_list))
1739 : return 0;
1740 :
1741 : /*
1742 : * Changing the data of the delayed item is impossible. So
1743 : * we needn't lock them. And we have held i_mutex of the
1744 : * directory, nobody can delete any directory indexes now.
1745 : */
1746 9526893 : list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1747 9450461 : list_del(&curr->readdir_list);
1748 :
1749 9450460 : if (curr->index < ctx->pos) {
1750 6158906 : if (refcount_dec_and_test(&curr->refs))
1751 3 : kfree(curr);
1752 6158906 : continue;
1753 : }
1754 :
1755 3291554 : ctx->pos = curr->index;
1756 :
1757 3291554 : di = (struct btrfs_dir_item *)curr->data;
1758 3291554 : name = (char *)(di + 1);
1759 3291554 : name_len = btrfs_stack_dir_name_len(di);
1760 :
1761 3291554 : d_type = fs_ftype_to_dtype(btrfs_dir_flags_to_ftype(di->type));
1762 3291554 : btrfs_disk_key_to_cpu(&location, &di->location);
1763 :
1764 3291554 : over = !dir_emit(ctx, name, name_len,
1765 : location.objectid, d_type);
1766 :
1767 3291554 : if (refcount_dec_and_test(&curr->refs))
1768 0 : kfree(curr);
1769 :
1770 3291554 : if (over)
1771 : return 1;
1772 3259491 : ctx->pos++;
1773 : }
1774 : return 0;
1775 : }
1776 :
1777 36410121 : static void fill_stack_inode_item(struct btrfs_trans_handle *trans,
1778 : struct btrfs_inode_item *inode_item,
1779 : struct inode *inode)
1780 : {
1781 36410121 : u64 flags;
1782 :
1783 36410121 : btrfs_set_stack_inode_uid(inode_item, i_uid_read(inode));
1784 36409651 : btrfs_set_stack_inode_gid(inode_item, i_gid_read(inode));
1785 36407159 : btrfs_set_stack_inode_size(inode_item, BTRFS_I(inode)->disk_i_size);
1786 36407159 : btrfs_set_stack_inode_mode(inode_item, inode->i_mode);
1787 36407159 : btrfs_set_stack_inode_nlink(inode_item, inode->i_nlink);
1788 36407159 : btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(inode));
1789 36410958 : btrfs_set_stack_inode_generation(inode_item,
1790 : BTRFS_I(inode)->generation);
1791 36410958 : btrfs_set_stack_inode_sequence(inode_item,
1792 : inode_peek_iversion(inode));
1793 36410958 : btrfs_set_stack_inode_transid(inode_item, trans->transid);
1794 36410958 : btrfs_set_stack_inode_rdev(inode_item, inode->i_rdev);
1795 36410958 : flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
1796 : BTRFS_I(inode)->ro_flags);
1797 36410958 : btrfs_set_stack_inode_flags(inode_item, flags);
1798 36410958 : btrfs_set_stack_inode_block_group(inode_item, 0);
1799 :
1800 36410958 : btrfs_set_stack_timespec_sec(&inode_item->atime,
1801 36410958 : inode->i_atime.tv_sec);
1802 36410958 : btrfs_set_stack_timespec_nsec(&inode_item->atime,
1803 36410958 : inode->i_atime.tv_nsec);
1804 :
1805 36410958 : btrfs_set_stack_timespec_sec(&inode_item->mtime,
1806 36410958 : inode->i_mtime.tv_sec);
1807 36410958 : btrfs_set_stack_timespec_nsec(&inode_item->mtime,
1808 36410958 : inode->i_mtime.tv_nsec);
1809 :
1810 36410958 : btrfs_set_stack_timespec_sec(&inode_item->ctime,
1811 36410958 : inode->i_ctime.tv_sec);
1812 36410958 : btrfs_set_stack_timespec_nsec(&inode_item->ctime,
1813 36410958 : inode->i_ctime.tv_nsec);
1814 :
1815 36410958 : btrfs_set_stack_timespec_sec(&inode_item->otime,
1816 36410958 : BTRFS_I(inode)->i_otime.tv_sec);
1817 36410958 : btrfs_set_stack_timespec_nsec(&inode_item->otime,
1818 36410958 : BTRFS_I(inode)->i_otime.tv_nsec);
1819 36410958 : }
1820 :
1821 559711 : int btrfs_fill_inode(struct inode *inode, u32 *rdev)
1822 : {
1823 559711 : struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1824 559711 : struct btrfs_delayed_node *delayed_node;
1825 559711 : struct btrfs_inode_item *inode_item;
1826 :
1827 559711 : delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
1828 559706 : if (!delayed_node)
1829 : return -ENOENT;
1830 :
1831 129 : mutex_lock(&delayed_node->mutex);
1832 129 : if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1833 115 : mutex_unlock(&delayed_node->mutex);
1834 115 : btrfs_release_delayed_node(delayed_node);
1835 115 : return -ENOENT;
1836 : }
1837 :
1838 14 : inode_item = &delayed_node->inode_item;
1839 :
1840 14 : i_uid_write(inode, btrfs_stack_inode_uid(inode_item));
1841 14 : i_gid_write(inode, btrfs_stack_inode_gid(inode_item));
1842 14 : btrfs_i_size_write(BTRFS_I(inode), btrfs_stack_inode_size(inode_item));
1843 14 : btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
1844 14 : round_up(i_size_read(inode), fs_info->sectorsize));
1845 14 : inode->i_mode = btrfs_stack_inode_mode(inode_item);
1846 14 : set_nlink(inode, btrfs_stack_inode_nlink(inode_item));
1847 14 : inode_set_bytes(inode, btrfs_stack_inode_nbytes(inode_item));
1848 14 : BTRFS_I(inode)->generation = btrfs_stack_inode_generation(inode_item);
1849 14 : BTRFS_I(inode)->last_trans = btrfs_stack_inode_transid(inode_item);
1850 :
1851 14 : inode_set_iversion_queried(inode,
1852 : btrfs_stack_inode_sequence(inode_item));
1853 14 : inode->i_rdev = 0;
1854 14 : *rdev = btrfs_stack_inode_rdev(inode_item);
1855 14 : btrfs_inode_split_flags(btrfs_stack_inode_flags(inode_item),
1856 : &BTRFS_I(inode)->flags, &BTRFS_I(inode)->ro_flags);
1857 :
1858 14 : inode->i_atime.tv_sec = btrfs_stack_timespec_sec(&inode_item->atime);
1859 14 : inode->i_atime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->atime);
1860 :
1861 14 : inode->i_mtime.tv_sec = btrfs_stack_timespec_sec(&inode_item->mtime);
1862 14 : inode->i_mtime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->mtime);
1863 :
1864 14 : inode->i_ctime.tv_sec = btrfs_stack_timespec_sec(&inode_item->ctime);
1865 14 : inode->i_ctime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->ctime);
1866 :
1867 14 : BTRFS_I(inode)->i_otime.tv_sec =
1868 14 : btrfs_stack_timespec_sec(&inode_item->otime);
1869 14 : BTRFS_I(inode)->i_otime.tv_nsec =
1870 14 : btrfs_stack_timespec_nsec(&inode_item->otime);
1871 :
1872 14 : inode->i_generation = BTRFS_I(inode)->generation;
1873 14 : BTRFS_I(inode)->index_cnt = (u64)-1;
1874 :
1875 14 : mutex_unlock(&delayed_node->mutex);
1876 14 : btrfs_release_delayed_node(delayed_node);
1877 14 : return 0;
1878 : }
1879 :
1880 36501522 : int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans,
1881 : struct btrfs_root *root,
1882 : struct btrfs_inode *inode)
1883 : {
1884 36501522 : struct btrfs_delayed_node *delayed_node;
1885 36501522 : int ret = 0;
1886 :
1887 36501522 : delayed_node = btrfs_get_or_create_delayed_node(inode);
1888 36501254 : if (IS_ERR(delayed_node))
1889 0 : return PTR_ERR(delayed_node);
1890 :
1891 36501254 : mutex_lock(&delayed_node->mutex);
1892 73003574 : if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1893 27822691 : fill_stack_inode_item(trans, &delayed_node->inode_item,
1894 : &inode->vfs_inode);
1895 27821469 : goto release_node;
1896 : }
1897 :
1898 8679096 : ret = btrfs_delayed_inode_reserve_metadata(trans, root, delayed_node);
1899 8679013 : if (ret)
1900 89082 : goto release_node;
1901 :
1902 8589931 : fill_stack_inode_item(trans, &delayed_node->inode_item, &inode->vfs_inode);
1903 8588345 : set_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
1904 8590331 : delayed_node->count++;
1905 8590331 : atomic_inc(&root->fs_info->delayed_root->items);
1906 36500022 : release_node:
1907 36500022 : mutex_unlock(&delayed_node->mutex);
1908 36501735 : btrfs_release_delayed_node(delayed_node);
1909 36501735 : return ret;
1910 : }
1911 :
1912 1337229 : int btrfs_delayed_delete_inode_ref(struct btrfs_inode *inode)
1913 : {
1914 1337229 : struct btrfs_fs_info *fs_info = inode->root->fs_info;
1915 1337229 : struct btrfs_delayed_node *delayed_node;
1916 :
1917 : /*
1918 : * we don't do delayed inode updates during log recovery because it
1919 : * leads to enospc problems. This means we also can't do
1920 : * delayed inode refs
1921 : */
1922 2674458 : if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
1923 : return -EAGAIN;
1924 :
1925 1337222 : delayed_node = btrfs_get_or_create_delayed_node(inode);
1926 1337225 : if (IS_ERR(delayed_node))
1927 0 : return PTR_ERR(delayed_node);
1928 :
1929 : /*
1930 : * We don't reserve space for inode ref deletion is because:
1931 : * - We ONLY do async inode ref deletion for the inode who has only
1932 : * one link(i_nlink == 1), it means there is only one inode ref.
1933 : * And in most case, the inode ref and the inode item are in the
1934 : * same leaf, and we will deal with them at the same time.
1935 : * Since we are sure we will reserve the space for the inode item,
1936 : * it is unnecessary to reserve space for inode ref deletion.
1937 : * - If the inode ref and the inode item are not in the same leaf,
1938 : * We also needn't worry about enospc problem, because we reserve
1939 : * much more space for the inode update than it needs.
1940 : * - At the worst, we can steal some space from the global reservation.
1941 : * It is very rare.
1942 : */
1943 1337225 : mutex_lock(&delayed_node->mutex);
1944 2674424 : if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1945 0 : goto release_node;
1946 :
1947 1337212 : set_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
1948 1337222 : delayed_node->count++;
1949 1337222 : atomic_inc(&fs_info->delayed_root->items);
1950 1337220 : release_node:
1951 1337220 : mutex_unlock(&delayed_node->mutex);
1952 1337225 : btrfs_release_delayed_node(delayed_node);
1953 1337225 : return 0;
1954 : }
1955 :
1956 1371767 : static void __btrfs_kill_delayed_node(struct btrfs_delayed_node *delayed_node)
1957 : {
1958 1371767 : struct btrfs_root *root = delayed_node->root;
1959 1371767 : struct btrfs_fs_info *fs_info = root->fs_info;
1960 1371767 : struct btrfs_delayed_item *curr_item, *prev_item;
1961 :
1962 1371767 : mutex_lock(&delayed_node->mutex);
1963 1371768 : curr_item = __btrfs_first_delayed_insertion_item(delayed_node);
1964 1371766 : while (curr_item) {
1965 5 : prev_item = curr_item;
1966 5 : curr_item = __btrfs_next_delayed_item(prev_item);
1967 5 : btrfs_release_delayed_item(prev_item);
1968 : }
1969 :
1970 1371761 : if (delayed_node->index_item_leaves > 0) {
1971 5 : btrfs_delayed_item_release_leaves(delayed_node,
1972 : delayed_node->index_item_leaves);
1973 5 : delayed_node->index_item_leaves = 0;
1974 : }
1975 :
1976 1371761 : curr_item = __btrfs_first_delayed_deletion_item(delayed_node);
1977 1528036 : while (curr_item) {
1978 156282 : btrfs_delayed_item_release_metadata(root, curr_item);
1979 156282 : prev_item = curr_item;
1980 156282 : curr_item = __btrfs_next_delayed_item(prev_item);
1981 156282 : btrfs_release_delayed_item(prev_item);
1982 : }
1983 :
1984 1371754 : btrfs_release_delayed_iref(delayed_node);
1985 :
1986 2743526 : if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1987 12 : btrfs_delayed_inode_release_metadata(fs_info, delayed_node, false);
1988 12 : btrfs_release_delayed_inode(delayed_node);
1989 : }
1990 1371763 : mutex_unlock(&delayed_node->mutex);
1991 1371767 : }
1992 :
1993 1774487 : void btrfs_kill_delayed_inode_items(struct btrfs_inode *inode)
1994 : {
1995 1774487 : struct btrfs_delayed_node *delayed_node;
1996 :
1997 1774487 : delayed_node = btrfs_get_delayed_node(inode);
1998 1774498 : if (!delayed_node)
1999 : return;
2000 :
2001 1371764 : __btrfs_kill_delayed_node(delayed_node);
2002 1371753 : btrfs_release_delayed_node(delayed_node);
2003 : }
2004 :
2005 118 : void btrfs_kill_all_delayed_nodes(struct btrfs_root *root)
2006 : {
2007 118 : u64 inode_id = 0;
2008 118 : struct btrfs_delayed_node *delayed_nodes[8];
2009 118 : int i, n;
2010 :
2011 118 : while (1) {
2012 118 : spin_lock(&root->inode_lock);
2013 118 : n = radix_tree_gang_lookup(&root->delayed_nodes_tree,
2014 : (void **)delayed_nodes, inode_id,
2015 : ARRAY_SIZE(delayed_nodes));
2016 118 : if (!n) {
2017 118 : spin_unlock(&root->inode_lock);
2018 118 : break;
2019 : }
2020 :
2021 0 : inode_id = delayed_nodes[n - 1]->inode_id + 1;
2022 0 : for (i = 0; i < n; i++) {
2023 : /*
2024 : * Don't increase refs in case the node is dead and
2025 : * about to be removed from the tree in the loop below
2026 : */
2027 0 : if (!refcount_inc_not_zero(&delayed_nodes[i]->refs))
2028 0 : delayed_nodes[i] = NULL;
2029 : }
2030 0 : spin_unlock(&root->inode_lock);
2031 :
2032 0 : for (i = 0; i < n; i++) {
2033 0 : if (!delayed_nodes[i])
2034 0 : continue;
2035 0 : __btrfs_kill_delayed_node(delayed_nodes[i]);
2036 0 : btrfs_release_delayed_node(delayed_nodes[i]);
2037 : }
2038 : }
2039 118 : }
2040 :
2041 37 : void btrfs_destroy_delayed_inodes(struct btrfs_fs_info *fs_info)
2042 : {
2043 37 : struct btrfs_delayed_node *curr_node, *prev_node;
2044 :
2045 37 : curr_node = btrfs_first_delayed_node(fs_info->delayed_root);
2046 49 : while (curr_node) {
2047 12 : __btrfs_kill_delayed_node(curr_node);
2048 :
2049 12 : prev_node = curr_node;
2050 12 : curr_node = btrfs_next_delayed_node(curr_node);
2051 12 : btrfs_release_delayed_node(prev_node);
2052 : }
2053 37 : }
2054 :
2055 705 : void btrfs_log_get_delayed_items(struct btrfs_inode *inode,
2056 : struct list_head *ins_list,
2057 : struct list_head *del_list)
2058 : {
2059 705 : struct btrfs_delayed_node *node;
2060 705 : struct btrfs_delayed_item *item;
2061 :
2062 705 : node = btrfs_get_delayed_node(inode);
2063 705 : if (!node)
2064 : return;
2065 :
2066 632 : mutex_lock(&node->mutex);
2067 632 : item = __btrfs_first_delayed_insertion_item(node);
2068 1310 : while (item) {
2069 : /*
2070 : * It's possible that the item is already in a log list. This
2071 : * can happen in case two tasks are trying to log the same
2072 : * directory. For example if we have tasks A and task B:
2073 : *
2074 : * Task A collected the delayed items into a log list while
2075 : * under the inode's log_mutex (at btrfs_log_inode()), but it
2076 : * only releases the items after logging the inodes they point
2077 : * to (if they are new inodes), which happens after unlocking
2078 : * the log mutex;
2079 : *
2080 : * Task B enters btrfs_log_inode() and acquires the log_mutex
2081 : * of the same directory inode, before task B releases the
2082 : * delayed items. This can happen for example when logging some
2083 : * inode we need to trigger logging of its parent directory, so
2084 : * logging two files that have the same parent directory can
2085 : * lead to this.
2086 : *
2087 : * If this happens, just ignore delayed items already in a log
2088 : * list. All the tasks logging the directory are under a log
2089 : * transaction and whichever finishes first can not sync the log
2090 : * before the other completes and leaves the log transaction.
2091 : */
2092 1150 : if (!item->logged && list_empty(&item->log_list)) {
2093 1130 : refcount_inc(&item->refs);
2094 1130 : list_add_tail(&item->log_list, ins_list);
2095 : }
2096 1150 : item = __btrfs_next_delayed_item(item);
2097 : }
2098 :
2099 632 : item = __btrfs_first_delayed_deletion_item(node);
2100 741 : while (item) {
2101 : /* It may be non-empty, for the same reason mentioned above. */
2102 312 : if (!item->logged && list_empty(&item->log_list)) {
2103 304 : refcount_inc(&item->refs);
2104 304 : list_add_tail(&item->log_list, del_list);
2105 : }
2106 312 : item = __btrfs_next_delayed_item(item);
2107 : }
2108 632 : mutex_unlock(&node->mutex);
2109 :
2110 : /*
2111 : * We are called during inode logging, which means the inode is in use
2112 : * and can not be evicted before we finish logging the inode. So we never
2113 : * have the last reference on the delayed inode.
2114 : * Also, we don't use btrfs_release_delayed_node() because that would
2115 : * requeue the delayed inode (change its order in the list of prepared
2116 : * nodes) and we don't want to do such change because we don't create or
2117 : * delete delayed items.
2118 : */
2119 632 : ASSERT(refcount_read(&node->refs) > 1);
2120 632 : refcount_dec(&node->refs);
2121 : }
2122 :
2123 1343 : void btrfs_log_put_delayed_items(struct btrfs_inode *inode,
2124 : struct list_head *ins_list,
2125 : struct list_head *del_list)
2126 : {
2127 1343 : struct btrfs_delayed_node *node;
2128 1343 : struct btrfs_delayed_item *item;
2129 1343 : struct btrfs_delayed_item *next;
2130 :
2131 1343 : node = btrfs_get_delayed_node(inode);
2132 1343 : if (!node)
2133 : return;
2134 :
2135 1270 : mutex_lock(&node->mutex);
2136 :
2137 2400 : list_for_each_entry_safe(item, next, ins_list, log_list) {
2138 1130 : item->logged = true;
2139 1130 : list_del_init(&item->log_list);
2140 1130 : if (refcount_dec_and_test(&item->refs))
2141 10 : kfree(item);
2142 : }
2143 :
2144 1574 : list_for_each_entry_safe(item, next, del_list, log_list) {
2145 304 : item->logged = true;
2146 304 : list_del_init(&item->log_list);
2147 304 : if (refcount_dec_and_test(&item->refs))
2148 8 : kfree(item);
2149 : }
2150 :
2151 1270 : mutex_unlock(&node->mutex);
2152 :
2153 : /*
2154 : * We are called during inode logging, which means the inode is in use
2155 : * and can not be evicted before we finish logging the inode. So we never
2156 : * have the last reference on the delayed inode.
2157 : * Also, we don't use btrfs_release_delayed_node() because that would
2158 : * requeue the delayed inode (change its order in the list of prepared
2159 : * nodes) and we don't want to do such change because we don't create or
2160 : * delete delayed items.
2161 : */
2162 1270 : ASSERT(refcount_read(&node->refs) > 1);
2163 1270 : refcount_dec(&node->refs);
2164 : }
|
