Commit | Line | Data |
---|---|---|
16cdcec7 MX |
1 | /* |
2 | * Copyright (C) 2011 Fujitsu. All rights reserved. | |
3 | * Written by Miao Xie <miaox@cn.fujitsu.com> | |
4 | * | |
5 | * This program is free software; you can redistribute it and/or | |
6 | * modify it under the terms of the GNU General Public | |
7 | * License v2 as published by the Free Software Foundation. | |
8 | * | |
9 | * This program is distributed in the hope that it will be useful, | |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
12 | * General Public License for more details. | |
13 | * | |
14 | * You should have received a copy of the GNU General Public | |
15 | * License along with this program; if not, write to the | |
16 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
17 | * Boston, MA 021110-1307, USA. | |
18 | */ | |
19 | ||
20 | #include <linux/slab.h> | |
21 | #include "delayed-inode.h" | |
22 | #include "disk-io.h" | |
23 | #include "transaction.h" | |
3cae210f | 24 | #include "ctree.h" |
16cdcec7 | 25 | |
de3cb945 CM |
26 | #define BTRFS_DELAYED_WRITEBACK 512 |
27 | #define BTRFS_DELAYED_BACKGROUND 128 | |
28 | #define BTRFS_DELAYED_BATCH 16 | |
16cdcec7 MX |
29 | |
30 | static struct kmem_cache *delayed_node_cache; | |
31 | ||
32 | int __init btrfs_delayed_inode_init(void) | |
33 | { | |
837e1972 | 34 | delayed_node_cache = kmem_cache_create("btrfs_delayed_node", |
16cdcec7 MX |
35 | sizeof(struct btrfs_delayed_node), |
36 | 0, | |
37 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, | |
38 | NULL); | |
39 | if (!delayed_node_cache) | |
40 | return -ENOMEM; | |
41 | return 0; | |
42 | } | |
43 | ||
44 | void btrfs_delayed_inode_exit(void) | |
45 | { | |
46 | if (delayed_node_cache) | |
47 | kmem_cache_destroy(delayed_node_cache); | |
48 | } | |
49 | ||
50 | static inline void btrfs_init_delayed_node( | |
51 | struct btrfs_delayed_node *delayed_node, | |
52 | struct btrfs_root *root, u64 inode_id) | |
53 | { | |
54 | delayed_node->root = root; | |
55 | delayed_node->inode_id = inode_id; | |
56 | atomic_set(&delayed_node->refs, 0); | |
57 | delayed_node->count = 0; | |
7cf35d91 | 58 | delayed_node->flags = 0; |
16cdcec7 MX |
59 | delayed_node->ins_root = RB_ROOT; |
60 | delayed_node->del_root = RB_ROOT; | |
61 | mutex_init(&delayed_node->mutex); | |
62 | delayed_node->index_cnt = 0; | |
63 | INIT_LIST_HEAD(&delayed_node->n_list); | |
64 | INIT_LIST_HEAD(&delayed_node->p_list); | |
65 | delayed_node->bytes_reserved = 0; | |
293f7e07 | 66 | memset(&delayed_node->inode_item, 0, sizeof(delayed_node->inode_item)); |
16cdcec7 MX |
67 | } |
68 | ||
69 | static inline int btrfs_is_continuous_delayed_item( | |
70 | struct btrfs_delayed_item *item1, | |
71 | struct btrfs_delayed_item *item2) | |
72 | { | |
73 | if (item1->key.type == BTRFS_DIR_INDEX_KEY && | |
74 | item1->key.objectid == item2->key.objectid && | |
75 | item1->key.type == item2->key.type && | |
76 | item1->key.offset + 1 == item2->key.offset) | |
77 | return 1; | |
78 | return 0; | |
79 | } | |
80 | ||
81 | static inline struct btrfs_delayed_root *btrfs_get_delayed_root( | |
82 | struct btrfs_root *root) | |
83 | { | |
84 | return root->fs_info->delayed_root; | |
85 | } | |
86 | ||
2f7e33d4 | 87 | static struct btrfs_delayed_node *btrfs_get_delayed_node(struct inode *inode) |
16cdcec7 | 88 | { |
16cdcec7 MX |
89 | struct btrfs_inode *btrfs_inode = BTRFS_I(inode); |
90 | struct btrfs_root *root = btrfs_inode->root; | |
0d0ca30f | 91 | u64 ino = btrfs_ino(inode); |
2f7e33d4 | 92 | struct btrfs_delayed_node *node; |
16cdcec7 | 93 | |
16cdcec7 MX |
94 | node = ACCESS_ONCE(btrfs_inode->delayed_node); |
95 | if (node) { | |
2f7e33d4 | 96 | atomic_inc(&node->refs); |
16cdcec7 MX |
97 | return node; |
98 | } | |
99 | ||
100 | spin_lock(&root->inode_lock); | |
0d0ca30f | 101 | node = radix_tree_lookup(&root->delayed_nodes_tree, ino); |
16cdcec7 MX |
102 | if (node) { |
103 | if (btrfs_inode->delayed_node) { | |
2f7e33d4 MX |
104 | atomic_inc(&node->refs); /* can be accessed */ |
105 | BUG_ON(btrfs_inode->delayed_node != node); | |
16cdcec7 | 106 | spin_unlock(&root->inode_lock); |
2f7e33d4 | 107 | return node; |
16cdcec7 MX |
108 | } |
109 | btrfs_inode->delayed_node = node; | |
95e94d14 R |
110 | /* can be accessed and cached in the inode */ |
111 | atomic_add(2, &node->refs); | |
16cdcec7 MX |
112 | spin_unlock(&root->inode_lock); |
113 | return node; | |
114 | } | |
115 | spin_unlock(&root->inode_lock); | |
116 | ||
2f7e33d4 MX |
117 | return NULL; |
118 | } | |
119 | ||
79787eaa | 120 | /* Will return either the node or PTR_ERR(-ENOMEM) */ |
2f7e33d4 MX |
121 | static struct btrfs_delayed_node *btrfs_get_or_create_delayed_node( |
122 | struct inode *inode) | |
123 | { | |
124 | struct btrfs_delayed_node *node; | |
125 | struct btrfs_inode *btrfs_inode = BTRFS_I(inode); | |
126 | struct btrfs_root *root = btrfs_inode->root; | |
127 | u64 ino = btrfs_ino(inode); | |
128 | int ret; | |
129 | ||
130 | again: | |
131 | node = btrfs_get_delayed_node(inode); | |
132 | if (node) | |
133 | return node; | |
134 | ||
16cdcec7 MX |
135 | node = kmem_cache_alloc(delayed_node_cache, GFP_NOFS); |
136 | if (!node) | |
137 | return ERR_PTR(-ENOMEM); | |
0d0ca30f | 138 | btrfs_init_delayed_node(node, root, ino); |
16cdcec7 | 139 | |
95e94d14 R |
140 | /* cached in the btrfs inode and can be accessed */ |
141 | atomic_add(2, &node->refs); | |
16cdcec7 MX |
142 | |
143 | ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM); | |
144 | if (ret) { | |
145 | kmem_cache_free(delayed_node_cache, node); | |
146 | return ERR_PTR(ret); | |
147 | } | |
148 | ||
149 | spin_lock(&root->inode_lock); | |
0d0ca30f | 150 | ret = radix_tree_insert(&root->delayed_nodes_tree, ino, node); |
16cdcec7 MX |
151 | if (ret == -EEXIST) { |
152 | kmem_cache_free(delayed_node_cache, node); | |
153 | spin_unlock(&root->inode_lock); | |
154 | radix_tree_preload_end(); | |
155 | goto again; | |
156 | } | |
157 | btrfs_inode->delayed_node = node; | |
158 | spin_unlock(&root->inode_lock); | |
159 | radix_tree_preload_end(); | |
160 | ||
161 | return node; | |
162 | } | |
163 | ||
164 | /* | |
165 | * Call it when holding delayed_node->mutex | |
166 | * | |
167 | * If mod = 1, add this node into the prepared list. | |
168 | */ | |
169 | static void btrfs_queue_delayed_node(struct btrfs_delayed_root *root, | |
170 | struct btrfs_delayed_node *node, | |
171 | int mod) | |
172 | { | |
173 | spin_lock(&root->lock); | |
7cf35d91 | 174 | if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) { |
16cdcec7 MX |
175 | if (!list_empty(&node->p_list)) |
176 | list_move_tail(&node->p_list, &root->prepare_list); | |
177 | else if (mod) | |
178 | list_add_tail(&node->p_list, &root->prepare_list); | |
179 | } else { | |
180 | list_add_tail(&node->n_list, &root->node_list); | |
181 | list_add_tail(&node->p_list, &root->prepare_list); | |
182 | atomic_inc(&node->refs); /* inserted into list */ | |
183 | root->nodes++; | |
7cf35d91 | 184 | set_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags); |
16cdcec7 MX |
185 | } |
186 | spin_unlock(&root->lock); | |
187 | } | |
188 | ||
189 | /* Call it when holding delayed_node->mutex */ | |
190 | static void btrfs_dequeue_delayed_node(struct btrfs_delayed_root *root, | |
191 | struct btrfs_delayed_node *node) | |
192 | { | |
193 | spin_lock(&root->lock); | |
7cf35d91 | 194 | if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) { |
16cdcec7 MX |
195 | root->nodes--; |
196 | atomic_dec(&node->refs); /* not in the list */ | |
197 | list_del_init(&node->n_list); | |
198 | if (!list_empty(&node->p_list)) | |
199 | list_del_init(&node->p_list); | |
7cf35d91 | 200 | clear_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags); |
16cdcec7 MX |
201 | } |
202 | spin_unlock(&root->lock); | |
203 | } | |
204 | ||
48a3b636 | 205 | static struct btrfs_delayed_node *btrfs_first_delayed_node( |
16cdcec7 MX |
206 | struct btrfs_delayed_root *delayed_root) |
207 | { | |
208 | struct list_head *p; | |
209 | struct btrfs_delayed_node *node = NULL; | |
210 | ||
211 | spin_lock(&delayed_root->lock); | |
212 | if (list_empty(&delayed_root->node_list)) | |
213 | goto out; | |
214 | ||
215 | p = delayed_root->node_list.next; | |
216 | node = list_entry(p, struct btrfs_delayed_node, n_list); | |
217 | atomic_inc(&node->refs); | |
218 | out: | |
219 | spin_unlock(&delayed_root->lock); | |
220 | ||
221 | return node; | |
222 | } | |
223 | ||
48a3b636 | 224 | static struct btrfs_delayed_node *btrfs_next_delayed_node( |
16cdcec7 MX |
225 | struct btrfs_delayed_node *node) |
226 | { | |
227 | struct btrfs_delayed_root *delayed_root; | |
228 | struct list_head *p; | |
229 | struct btrfs_delayed_node *next = NULL; | |
230 | ||
231 | delayed_root = node->root->fs_info->delayed_root; | |
232 | spin_lock(&delayed_root->lock); | |
7cf35d91 MX |
233 | if (!test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) { |
234 | /* not in the list */ | |
16cdcec7 MX |
235 | if (list_empty(&delayed_root->node_list)) |
236 | goto out; | |
237 | p = delayed_root->node_list.next; | |
238 | } else if (list_is_last(&node->n_list, &delayed_root->node_list)) | |
239 | goto out; | |
240 | else | |
241 | p = node->n_list.next; | |
242 | ||
243 | next = list_entry(p, struct btrfs_delayed_node, n_list); | |
244 | atomic_inc(&next->refs); | |
245 | out: | |
246 | spin_unlock(&delayed_root->lock); | |
247 | ||
248 | return next; | |
249 | } | |
250 | ||
251 | static void __btrfs_release_delayed_node( | |
252 | struct btrfs_delayed_node *delayed_node, | |
253 | int mod) | |
254 | { | |
255 | struct btrfs_delayed_root *delayed_root; | |
256 | ||
257 | if (!delayed_node) | |
258 | return; | |
259 | ||
260 | delayed_root = delayed_node->root->fs_info->delayed_root; | |
261 | ||
262 | mutex_lock(&delayed_node->mutex); | |
263 | if (delayed_node->count) | |
264 | btrfs_queue_delayed_node(delayed_root, delayed_node, mod); | |
265 | else | |
266 | btrfs_dequeue_delayed_node(delayed_root, delayed_node); | |
267 | mutex_unlock(&delayed_node->mutex); | |
268 | ||
269 | if (atomic_dec_and_test(&delayed_node->refs)) { | |
270 | struct btrfs_root *root = delayed_node->root; | |
271 | spin_lock(&root->inode_lock); | |
272 | if (atomic_read(&delayed_node->refs) == 0) { | |
273 | radix_tree_delete(&root->delayed_nodes_tree, | |
274 | delayed_node->inode_id); | |
275 | kmem_cache_free(delayed_node_cache, delayed_node); | |
276 | } | |
277 | spin_unlock(&root->inode_lock); | |
278 | } | |
279 | } | |
280 | ||
281 | static inline void btrfs_release_delayed_node(struct btrfs_delayed_node *node) | |
282 | { | |
283 | __btrfs_release_delayed_node(node, 0); | |
284 | } | |
285 | ||
48a3b636 | 286 | static struct btrfs_delayed_node *btrfs_first_prepared_delayed_node( |
16cdcec7 MX |
287 | struct btrfs_delayed_root *delayed_root) |
288 | { | |
289 | struct list_head *p; | |
290 | struct btrfs_delayed_node *node = NULL; | |
291 | ||
292 | spin_lock(&delayed_root->lock); | |
293 | if (list_empty(&delayed_root->prepare_list)) | |
294 | goto out; | |
295 | ||
296 | p = delayed_root->prepare_list.next; | |
297 | list_del_init(p); | |
298 | node = list_entry(p, struct btrfs_delayed_node, p_list); | |
299 | atomic_inc(&node->refs); | |
300 | out: | |
301 | spin_unlock(&delayed_root->lock); | |
302 | ||
303 | return node; | |
304 | } | |
305 | ||
306 | static inline void btrfs_release_prepared_delayed_node( | |
307 | struct btrfs_delayed_node *node) | |
308 | { | |
309 | __btrfs_release_delayed_node(node, 1); | |
310 | } | |
311 | ||
48a3b636 | 312 | static struct btrfs_delayed_item *btrfs_alloc_delayed_item(u32 data_len) |
16cdcec7 MX |
313 | { |
314 | struct btrfs_delayed_item *item; | |
315 | item = kmalloc(sizeof(*item) + data_len, GFP_NOFS); | |
316 | if (item) { | |
317 | item->data_len = data_len; | |
318 | item->ins_or_del = 0; | |
319 | item->bytes_reserved = 0; | |
16cdcec7 MX |
320 | item->delayed_node = NULL; |
321 | atomic_set(&item->refs, 1); | |
322 | } | |
323 | return item; | |
324 | } | |
325 | ||
326 | /* | |
327 | * __btrfs_lookup_delayed_item - look up the delayed item by key | |
328 | * @delayed_node: pointer to the delayed node | |
329 | * @key: the key to look up | |
330 | * @prev: used to store the prev item if the right item isn't found | |
331 | * @next: used to store the next item if the right item isn't found | |
332 | * | |
333 | * Note: if we don't find the right item, we will return the prev item and | |
334 | * the next item. | |
335 | */ | |
336 | static struct btrfs_delayed_item *__btrfs_lookup_delayed_item( | |
337 | struct rb_root *root, | |
338 | struct btrfs_key *key, | |
339 | struct btrfs_delayed_item **prev, | |
340 | struct btrfs_delayed_item **next) | |
341 | { | |
342 | struct rb_node *node, *prev_node = NULL; | |
343 | struct btrfs_delayed_item *delayed_item = NULL; | |
344 | int ret = 0; | |
345 | ||
346 | node = root->rb_node; | |
347 | ||
348 | while (node) { | |
349 | delayed_item = rb_entry(node, struct btrfs_delayed_item, | |
350 | rb_node); | |
351 | prev_node = node; | |
352 | ret = btrfs_comp_cpu_keys(&delayed_item->key, key); | |
353 | if (ret < 0) | |
354 | node = node->rb_right; | |
355 | else if (ret > 0) | |
356 | node = node->rb_left; | |
357 | else | |
358 | return delayed_item; | |
359 | } | |
360 | ||
361 | if (prev) { | |
362 | if (!prev_node) | |
363 | *prev = NULL; | |
364 | else if (ret < 0) | |
365 | *prev = delayed_item; | |
366 | else if ((node = rb_prev(prev_node)) != NULL) { | |
367 | *prev = rb_entry(node, struct btrfs_delayed_item, | |
368 | rb_node); | |
369 | } else | |
370 | *prev = NULL; | |
371 | } | |
372 | ||
373 | if (next) { | |
374 | if (!prev_node) | |
375 | *next = NULL; | |
376 | else if (ret > 0) | |
377 | *next = delayed_item; | |
378 | else if ((node = rb_next(prev_node)) != NULL) { | |
379 | *next = rb_entry(node, struct btrfs_delayed_item, | |
380 | rb_node); | |
381 | } else | |
382 | *next = NULL; | |
383 | } | |
384 | return NULL; | |
385 | } | |
386 | ||
48a3b636 | 387 | static struct btrfs_delayed_item *__btrfs_lookup_delayed_insertion_item( |
16cdcec7 MX |
388 | struct btrfs_delayed_node *delayed_node, |
389 | struct btrfs_key *key) | |
390 | { | |
391 | struct btrfs_delayed_item *item; | |
392 | ||
393 | item = __btrfs_lookup_delayed_item(&delayed_node->ins_root, key, | |
394 | NULL, NULL); | |
395 | return item; | |
396 | } | |
397 | ||
16cdcec7 MX |
398 | static int __btrfs_add_delayed_item(struct btrfs_delayed_node *delayed_node, |
399 | struct btrfs_delayed_item *ins, | |
400 | int action) | |
401 | { | |
402 | struct rb_node **p, *node; | |
403 | struct rb_node *parent_node = NULL; | |
404 | struct rb_root *root; | |
405 | struct btrfs_delayed_item *item; | |
406 | int cmp; | |
407 | ||
408 | if (action == BTRFS_DELAYED_INSERTION_ITEM) | |
409 | root = &delayed_node->ins_root; | |
410 | else if (action == BTRFS_DELAYED_DELETION_ITEM) | |
411 | root = &delayed_node->del_root; | |
412 | else | |
413 | BUG(); | |
414 | p = &root->rb_node; | |
415 | node = &ins->rb_node; | |
416 | ||
417 | while (*p) { | |
418 | parent_node = *p; | |
419 | item = rb_entry(parent_node, struct btrfs_delayed_item, | |
420 | rb_node); | |
421 | ||
422 | cmp = btrfs_comp_cpu_keys(&item->key, &ins->key); | |
423 | if (cmp < 0) | |
424 | p = &(*p)->rb_right; | |
425 | else if (cmp > 0) | |
426 | p = &(*p)->rb_left; | |
427 | else | |
428 | return -EEXIST; | |
429 | } | |
430 | ||
431 | rb_link_node(node, parent_node, p); | |
432 | rb_insert_color(node, root); | |
433 | ins->delayed_node = delayed_node; | |
434 | ins->ins_or_del = action; | |
435 | ||
436 | if (ins->key.type == BTRFS_DIR_INDEX_KEY && | |
437 | action == BTRFS_DELAYED_INSERTION_ITEM && | |
438 | ins->key.offset >= delayed_node->index_cnt) | |
439 | delayed_node->index_cnt = ins->key.offset + 1; | |
440 | ||
441 | delayed_node->count++; | |
442 | atomic_inc(&delayed_node->root->fs_info->delayed_root->items); | |
443 | return 0; | |
444 | } | |
445 | ||
446 | static int __btrfs_add_delayed_insertion_item(struct btrfs_delayed_node *node, | |
447 | struct btrfs_delayed_item *item) | |
448 | { | |
449 | return __btrfs_add_delayed_item(node, item, | |
450 | BTRFS_DELAYED_INSERTION_ITEM); | |
451 | } | |
452 | ||
453 | static int __btrfs_add_delayed_deletion_item(struct btrfs_delayed_node *node, | |
454 | struct btrfs_delayed_item *item) | |
455 | { | |
456 | return __btrfs_add_delayed_item(node, item, | |
457 | BTRFS_DELAYED_DELETION_ITEM); | |
458 | } | |
459 | ||
de3cb945 CM |
460 | static void finish_one_item(struct btrfs_delayed_root *delayed_root) |
461 | { | |
462 | int seq = atomic_inc_return(&delayed_root->items_seq); | |
463 | if ((atomic_dec_return(&delayed_root->items) < | |
464 | BTRFS_DELAYED_BACKGROUND || seq % BTRFS_DELAYED_BATCH == 0) && | |
465 | waitqueue_active(&delayed_root->wait)) | |
466 | wake_up(&delayed_root->wait); | |
467 | } | |
468 | ||
16cdcec7 MX |
469 | static void __btrfs_remove_delayed_item(struct btrfs_delayed_item *delayed_item) |
470 | { | |
471 | struct rb_root *root; | |
472 | struct btrfs_delayed_root *delayed_root; | |
473 | ||
474 | delayed_root = delayed_item->delayed_node->root->fs_info->delayed_root; | |
475 | ||
476 | BUG_ON(!delayed_root); | |
477 | BUG_ON(delayed_item->ins_or_del != BTRFS_DELAYED_DELETION_ITEM && | |
478 | delayed_item->ins_or_del != BTRFS_DELAYED_INSERTION_ITEM); | |
479 | ||
480 | if (delayed_item->ins_or_del == BTRFS_DELAYED_INSERTION_ITEM) | |
481 | root = &delayed_item->delayed_node->ins_root; | |
482 | else | |
483 | root = &delayed_item->delayed_node->del_root; | |
484 | ||
485 | rb_erase(&delayed_item->rb_node, root); | |
486 | delayed_item->delayed_node->count--; | |
de3cb945 CM |
487 | |
488 | finish_one_item(delayed_root); | |
16cdcec7 MX |
489 | } |
490 | ||
491 | static void btrfs_release_delayed_item(struct btrfs_delayed_item *item) | |
492 | { | |
493 | if (item) { | |
494 | __btrfs_remove_delayed_item(item); | |
495 | if (atomic_dec_and_test(&item->refs)) | |
496 | kfree(item); | |
497 | } | |
498 | } | |
499 | ||
48a3b636 | 500 | static struct btrfs_delayed_item *__btrfs_first_delayed_insertion_item( |
16cdcec7 MX |
501 | struct btrfs_delayed_node *delayed_node) |
502 | { | |
503 | struct rb_node *p; | |
504 | struct btrfs_delayed_item *item = NULL; | |
505 | ||
506 | p = rb_first(&delayed_node->ins_root); | |
507 | if (p) | |
508 | item = rb_entry(p, struct btrfs_delayed_item, rb_node); | |
509 | ||
510 | return item; | |
511 | } | |
512 | ||
48a3b636 | 513 | static struct btrfs_delayed_item *__btrfs_first_delayed_deletion_item( |
16cdcec7 MX |
514 | struct btrfs_delayed_node *delayed_node) |
515 | { | |
516 | struct rb_node *p; | |
517 | struct btrfs_delayed_item *item = NULL; | |
518 | ||
519 | p = rb_first(&delayed_node->del_root); | |
520 | if (p) | |
521 | item = rb_entry(p, struct btrfs_delayed_item, rb_node); | |
522 | ||
523 | return item; | |
524 | } | |
525 | ||
48a3b636 | 526 | static struct btrfs_delayed_item *__btrfs_next_delayed_item( |
16cdcec7 MX |
527 | struct btrfs_delayed_item *item) |
528 | { | |
529 | struct rb_node *p; | |
530 | struct btrfs_delayed_item *next = NULL; | |
531 | ||
532 | p = rb_next(&item->rb_node); | |
533 | if (p) | |
534 | next = rb_entry(p, struct btrfs_delayed_item, rb_node); | |
535 | ||
536 | return next; | |
537 | } | |
538 | ||
16cdcec7 MX |
539 | static int btrfs_delayed_item_reserve_metadata(struct btrfs_trans_handle *trans, |
540 | struct btrfs_root *root, | |
541 | struct btrfs_delayed_item *item) | |
542 | { | |
543 | struct btrfs_block_rsv *src_rsv; | |
544 | struct btrfs_block_rsv *dst_rsv; | |
545 | u64 num_bytes; | |
546 | int ret; | |
547 | ||
548 | if (!trans->bytes_reserved) | |
549 | return 0; | |
550 | ||
551 | src_rsv = trans->block_rsv; | |
6d668dda | 552 | dst_rsv = &root->fs_info->delayed_block_rsv; |
16cdcec7 MX |
553 | |
554 | num_bytes = btrfs_calc_trans_metadata_size(root, 1); | |
555 | ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes); | |
8c2a3ca2 JB |
556 | if (!ret) { |
557 | trace_btrfs_space_reservation(root->fs_info, "delayed_item", | |
558 | item->key.objectid, | |
559 | num_bytes, 1); | |
16cdcec7 | 560 | item->bytes_reserved = num_bytes; |
8c2a3ca2 | 561 | } |
16cdcec7 MX |
562 | |
563 | return ret; | |
564 | } | |
565 | ||
566 | static void btrfs_delayed_item_release_metadata(struct btrfs_root *root, | |
567 | struct btrfs_delayed_item *item) | |
568 | { | |
19fd2949 MX |
569 | struct btrfs_block_rsv *rsv; |
570 | ||
16cdcec7 MX |
571 | if (!item->bytes_reserved) |
572 | return; | |
573 | ||
6d668dda | 574 | rsv = &root->fs_info->delayed_block_rsv; |
8c2a3ca2 JB |
575 | trace_btrfs_space_reservation(root->fs_info, "delayed_item", |
576 | item->key.objectid, item->bytes_reserved, | |
577 | 0); | |
19fd2949 | 578 | btrfs_block_rsv_release(root, rsv, |
16cdcec7 MX |
579 | item->bytes_reserved); |
580 | } | |
581 | ||
582 | static int btrfs_delayed_inode_reserve_metadata( | |
583 | struct btrfs_trans_handle *trans, | |
584 | struct btrfs_root *root, | |
7fd2ae21 | 585 | struct inode *inode, |
16cdcec7 MX |
586 | struct btrfs_delayed_node *node) |
587 | { | |
588 | struct btrfs_block_rsv *src_rsv; | |
589 | struct btrfs_block_rsv *dst_rsv; | |
590 | u64 num_bytes; | |
591 | int ret; | |
8c2a3ca2 | 592 | bool release = false; |
16cdcec7 | 593 | |
16cdcec7 | 594 | src_rsv = trans->block_rsv; |
6d668dda | 595 | dst_rsv = &root->fs_info->delayed_block_rsv; |
16cdcec7 MX |
596 | |
597 | num_bytes = btrfs_calc_trans_metadata_size(root, 1); | |
c06a0e12 JB |
598 | |
599 | /* | |
600 | * btrfs_dirty_inode will update the inode under btrfs_join_transaction | |
601 | * which doesn't reserve space for speed. This is a problem since we | |
602 | * still need to reserve space for this update, so try to reserve the | |
603 | * space. | |
604 | * | |
605 | * Now if src_rsv == delalloc_block_rsv we'll let it just steal since | |
606 | * we're accounted for. | |
607 | */ | |
e755d9ab | 608 | if (!src_rsv || (!trans->bytes_reserved && |
66d8f3dd | 609 | src_rsv->type != BTRFS_BLOCK_RSV_DELALLOC)) { |
08e007d2 MX |
610 | ret = btrfs_block_rsv_add(root, dst_rsv, num_bytes, |
611 | BTRFS_RESERVE_NO_FLUSH); | |
c06a0e12 JB |
612 | /* |
613 | * Since we're under a transaction reserve_metadata_bytes could | |
614 | * try to commit the transaction which will make it return | |
615 | * EAGAIN to make us stop the transaction we have, so return | |
616 | * ENOSPC instead so that btrfs_dirty_inode knows what to do. | |
617 | */ | |
618 | if (ret == -EAGAIN) | |
619 | ret = -ENOSPC; | |
8c2a3ca2 | 620 | if (!ret) { |
c06a0e12 | 621 | node->bytes_reserved = num_bytes; |
8c2a3ca2 JB |
622 | trace_btrfs_space_reservation(root->fs_info, |
623 | "delayed_inode", | |
624 | btrfs_ino(inode), | |
625 | num_bytes, 1); | |
626 | } | |
c06a0e12 | 627 | return ret; |
66d8f3dd | 628 | } else if (src_rsv->type == BTRFS_BLOCK_RSV_DELALLOC) { |
7fd2ae21 | 629 | spin_lock(&BTRFS_I(inode)->lock); |
72ac3c0d JB |
630 | if (test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
631 | &BTRFS_I(inode)->runtime_flags)) { | |
7fd2ae21 JB |
632 | spin_unlock(&BTRFS_I(inode)->lock); |
633 | release = true; | |
634 | goto migrate; | |
635 | } | |
636 | spin_unlock(&BTRFS_I(inode)->lock); | |
637 | ||
638 | /* Ok we didn't have space pre-reserved. This shouldn't happen | |
639 | * too often but it can happen if we do delalloc to an existing | |
640 | * inode which gets dirtied because of the time update, and then | |
641 | * isn't touched again until after the transaction commits and | |
642 | * then we try to write out the data. First try to be nice and | |
643 | * reserve something strictly for us. If not be a pain and try | |
644 | * to steal from the delalloc block rsv. | |
645 | */ | |
08e007d2 MX |
646 | ret = btrfs_block_rsv_add(root, dst_rsv, num_bytes, |
647 | BTRFS_RESERVE_NO_FLUSH); | |
7fd2ae21 JB |
648 | if (!ret) |
649 | goto out; | |
650 | ||
651 | ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes); | |
fae7f21c | 652 | if (!WARN_ON(ret)) |
7fd2ae21 JB |
653 | goto out; |
654 | ||
655 | /* | |
656 | * Ok this is a problem, let's just steal from the global rsv | |
657 | * since this really shouldn't happen that often. | |
658 | */ | |
7fd2ae21 JB |
659 | ret = btrfs_block_rsv_migrate(&root->fs_info->global_block_rsv, |
660 | dst_rsv, num_bytes); | |
661 | goto out; | |
c06a0e12 JB |
662 | } |
663 | ||
7fd2ae21 | 664 | migrate: |
16cdcec7 | 665 | ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes); |
7fd2ae21 JB |
666 | |
667 | out: | |
668 | /* | |
669 | * Migrate only takes a reservation, it doesn't touch the size of the | |
670 | * block_rsv. This is to simplify people who don't normally have things | |
671 | * migrated from their block rsv. If they go to release their | |
672 | * reservation, that will decrease the size as well, so if migrate | |
673 | * reduced size we'd end up with a negative size. But for the | |
674 | * delalloc_meta_reserved stuff we will only know to drop 1 reservation, | |
675 | * but we could in fact do this reserve/migrate dance several times | |
676 | * between the time we did the original reservation and we'd clean it | |
677 | * up. So to take care of this, release the space for the meta | |
678 | * reservation here. I think it may be time for a documentation page on | |
679 | * how block rsvs. work. | |
680 | */ | |
8c2a3ca2 JB |
681 | if (!ret) { |
682 | trace_btrfs_space_reservation(root->fs_info, "delayed_inode", | |
683 | btrfs_ino(inode), num_bytes, 1); | |
16cdcec7 | 684 | node->bytes_reserved = num_bytes; |
8c2a3ca2 | 685 | } |
16cdcec7 | 686 | |
8c2a3ca2 JB |
687 | if (release) { |
688 | trace_btrfs_space_reservation(root->fs_info, "delalloc", | |
689 | btrfs_ino(inode), num_bytes, 0); | |
7fd2ae21 | 690 | btrfs_block_rsv_release(root, src_rsv, num_bytes); |
8c2a3ca2 | 691 | } |
16cdcec7 MX |
692 | |
693 | return ret; | |
694 | } | |
695 | ||
696 | static void btrfs_delayed_inode_release_metadata(struct btrfs_root *root, | |
697 | struct btrfs_delayed_node *node) | |
698 | { | |
699 | struct btrfs_block_rsv *rsv; | |
700 | ||
701 | if (!node->bytes_reserved) | |
702 | return; | |
703 | ||
6d668dda | 704 | rsv = &root->fs_info->delayed_block_rsv; |
8c2a3ca2 JB |
705 | trace_btrfs_space_reservation(root->fs_info, "delayed_inode", |
706 | node->inode_id, node->bytes_reserved, 0); | |
16cdcec7 MX |
707 | btrfs_block_rsv_release(root, rsv, |
708 | node->bytes_reserved); | |
709 | node->bytes_reserved = 0; | |
710 | } | |
711 | ||
712 | /* | |
713 | * This helper will insert some continuous items into the same leaf according | |
714 | * to the free space of the leaf. | |
715 | */ | |
afe5fea7 TI |
716 | static int btrfs_batch_insert_items(struct btrfs_root *root, |
717 | struct btrfs_path *path, | |
718 | struct btrfs_delayed_item *item) | |
16cdcec7 MX |
719 | { |
720 | struct btrfs_delayed_item *curr, *next; | |
721 | int free_space; | |
722 | int total_data_size = 0, total_size = 0; | |
723 | struct extent_buffer *leaf; | |
724 | char *data_ptr; | |
725 | struct btrfs_key *keys; | |
726 | u32 *data_size; | |
727 | struct list_head head; | |
728 | int slot; | |
729 | int nitems; | |
730 | int i; | |
731 | int ret = 0; | |
732 | ||
733 | BUG_ON(!path->nodes[0]); | |
734 | ||
735 | leaf = path->nodes[0]; | |
736 | free_space = btrfs_leaf_free_space(root, leaf); | |
737 | INIT_LIST_HEAD(&head); | |
738 | ||
739 | next = item; | |
17aca1c9 | 740 | nitems = 0; |
16cdcec7 MX |
741 | |
742 | /* | |
743 | * count the number of the continuous items that we can insert in batch | |
744 | */ | |
745 | while (total_size + next->data_len + sizeof(struct btrfs_item) <= | |
746 | free_space) { | |
747 | total_data_size += next->data_len; | |
748 | total_size += next->data_len + sizeof(struct btrfs_item); | |
749 | list_add_tail(&next->tree_list, &head); | |
750 | nitems++; | |
751 | ||
752 | curr = next; | |
753 | next = __btrfs_next_delayed_item(curr); | |
754 | if (!next) | |
755 | break; | |
756 | ||
757 | if (!btrfs_is_continuous_delayed_item(curr, next)) | |
758 | break; | |
759 | } | |
760 | ||
761 | if (!nitems) { | |
762 | ret = 0; | |
763 | goto out; | |
764 | } | |
765 | ||
766 | /* | |
767 | * we need allocate some memory space, but it might cause the task | |
768 | * to sleep, so we set all locked nodes in the path to blocking locks | |
769 | * first. | |
770 | */ | |
771 | btrfs_set_path_blocking(path); | |
772 | ||
d9b0d9ba | 773 | keys = kmalloc_array(nitems, sizeof(struct btrfs_key), GFP_NOFS); |
16cdcec7 MX |
774 | if (!keys) { |
775 | ret = -ENOMEM; | |
776 | goto out; | |
777 | } | |
778 | ||
d9b0d9ba | 779 | data_size = kmalloc_array(nitems, sizeof(u32), GFP_NOFS); |
16cdcec7 MX |
780 | if (!data_size) { |
781 | ret = -ENOMEM; | |
782 | goto error; | |
783 | } | |
784 | ||
785 | /* get keys of all the delayed items */ | |
786 | i = 0; | |
787 | list_for_each_entry(next, &head, tree_list) { | |
788 | keys[i] = next->key; | |
789 | data_size[i] = next->data_len; | |
790 | i++; | |
791 | } | |
792 | ||
793 | /* reset all the locked nodes in the patch to spinning locks. */ | |
bd681513 | 794 | btrfs_clear_path_blocking(path, NULL, 0); |
16cdcec7 MX |
795 | |
796 | /* insert the keys of the items */ | |
afe5fea7 | 797 | setup_items_for_insert(root, path, keys, data_size, |
143bede5 | 798 | total_data_size, total_size, nitems); |
16cdcec7 MX |
799 | |
800 | /* insert the dir index items */ | |
801 | slot = path->slots[0]; | |
802 | list_for_each_entry_safe(curr, next, &head, tree_list) { | |
803 | data_ptr = btrfs_item_ptr(leaf, slot, char); | |
804 | write_extent_buffer(leaf, &curr->data, | |
805 | (unsigned long)data_ptr, | |
806 | curr->data_len); | |
807 | slot++; | |
808 | ||
809 | btrfs_delayed_item_release_metadata(root, curr); | |
810 | ||
811 | list_del(&curr->tree_list); | |
812 | btrfs_release_delayed_item(curr); | |
813 | } | |
814 | ||
815 | error: | |
816 | kfree(data_size); | |
817 | kfree(keys); | |
818 | out: | |
819 | return ret; | |
820 | } | |
821 | ||
822 | /* | |
823 | * This helper can just do simple insertion that needn't extend item for new | |
824 | * data, such as directory name index insertion, inode insertion. | |
825 | */ | |
826 | static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans, | |
827 | struct btrfs_root *root, | |
828 | struct btrfs_path *path, | |
829 | struct btrfs_delayed_item *delayed_item) | |
830 | { | |
831 | struct extent_buffer *leaf; | |
16cdcec7 MX |
832 | char *ptr; |
833 | int ret; | |
834 | ||
835 | ret = btrfs_insert_empty_item(trans, root, path, &delayed_item->key, | |
836 | delayed_item->data_len); | |
837 | if (ret < 0 && ret != -EEXIST) | |
838 | return ret; | |
839 | ||
840 | leaf = path->nodes[0]; | |
841 | ||
16cdcec7 MX |
842 | ptr = btrfs_item_ptr(leaf, path->slots[0], char); |
843 | ||
844 | write_extent_buffer(leaf, delayed_item->data, (unsigned long)ptr, | |
845 | delayed_item->data_len); | |
846 | btrfs_mark_buffer_dirty(leaf); | |
847 | ||
848 | btrfs_delayed_item_release_metadata(root, delayed_item); | |
849 | return 0; | |
850 | } | |
851 | ||
852 | /* | |
853 | * we insert an item first, then if there are some continuous items, we try | |
854 | * to insert those items into the same leaf. | |
855 | */ | |
856 | static int btrfs_insert_delayed_items(struct btrfs_trans_handle *trans, | |
857 | struct btrfs_path *path, | |
858 | struct btrfs_root *root, | |
859 | struct btrfs_delayed_node *node) | |
860 | { | |
861 | struct btrfs_delayed_item *curr, *prev; | |
862 | int ret = 0; | |
863 | ||
864 | do_again: | |
865 | mutex_lock(&node->mutex); | |
866 | curr = __btrfs_first_delayed_insertion_item(node); | |
867 | if (!curr) | |
868 | goto insert_end; | |
869 | ||
870 | ret = btrfs_insert_delayed_item(trans, root, path, curr); | |
871 | if (ret < 0) { | |
945d8962 | 872 | btrfs_release_path(path); |
16cdcec7 MX |
873 | goto insert_end; |
874 | } | |
875 | ||
876 | prev = curr; | |
877 | curr = __btrfs_next_delayed_item(prev); | |
878 | if (curr && btrfs_is_continuous_delayed_item(prev, curr)) { | |
879 | /* insert the continuous items into the same leaf */ | |
880 | path->slots[0]++; | |
afe5fea7 | 881 | btrfs_batch_insert_items(root, path, curr); |
16cdcec7 MX |
882 | } |
883 | btrfs_release_delayed_item(prev); | |
884 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
885 | ||
945d8962 | 886 | btrfs_release_path(path); |
16cdcec7 MX |
887 | mutex_unlock(&node->mutex); |
888 | goto do_again; | |
889 | ||
890 | insert_end: | |
891 | mutex_unlock(&node->mutex); | |
892 | return ret; | |
893 | } | |
894 | ||
895 | static int btrfs_batch_delete_items(struct btrfs_trans_handle *trans, | |
896 | struct btrfs_root *root, | |
897 | struct btrfs_path *path, | |
898 | struct btrfs_delayed_item *item) | |
899 | { | |
900 | struct btrfs_delayed_item *curr, *next; | |
901 | struct extent_buffer *leaf; | |
902 | struct btrfs_key key; | |
903 | struct list_head head; | |
904 | int nitems, i, last_item; | |
905 | int ret = 0; | |
906 | ||
907 | BUG_ON(!path->nodes[0]); | |
908 | ||
909 | leaf = path->nodes[0]; | |
910 | ||
911 | i = path->slots[0]; | |
912 | last_item = btrfs_header_nritems(leaf) - 1; | |
913 | if (i > last_item) | |
914 | return -ENOENT; /* FIXME: Is errno suitable? */ | |
915 | ||
916 | next = item; | |
917 | INIT_LIST_HEAD(&head); | |
918 | btrfs_item_key_to_cpu(leaf, &key, i); | |
919 | nitems = 0; | |
920 | /* | |
921 | * count the number of the dir index items that we can delete in batch | |
922 | */ | |
923 | while (btrfs_comp_cpu_keys(&next->key, &key) == 0) { | |
924 | list_add_tail(&next->tree_list, &head); | |
925 | nitems++; | |
926 | ||
927 | curr = next; | |
928 | next = __btrfs_next_delayed_item(curr); | |
929 | if (!next) | |
930 | break; | |
931 | ||
932 | if (!btrfs_is_continuous_delayed_item(curr, next)) | |
933 | break; | |
934 | ||
935 | i++; | |
936 | if (i > last_item) | |
937 | break; | |
938 | btrfs_item_key_to_cpu(leaf, &key, i); | |
939 | } | |
940 | ||
941 | if (!nitems) | |
942 | return 0; | |
943 | ||
944 | ret = btrfs_del_items(trans, root, path, path->slots[0], nitems); | |
945 | if (ret) | |
946 | goto out; | |
947 | ||
948 | list_for_each_entry_safe(curr, next, &head, tree_list) { | |
949 | btrfs_delayed_item_release_metadata(root, curr); | |
950 | list_del(&curr->tree_list); | |
951 | btrfs_release_delayed_item(curr); | |
952 | } | |
953 | ||
954 | out: | |
955 | return ret; | |
956 | } | |
957 | ||
958 | static int btrfs_delete_delayed_items(struct btrfs_trans_handle *trans, | |
959 | struct btrfs_path *path, | |
960 | struct btrfs_root *root, | |
961 | struct btrfs_delayed_node *node) | |
962 | { | |
963 | struct btrfs_delayed_item *curr, *prev; | |
964 | int ret = 0; | |
965 | ||
966 | do_again: | |
967 | mutex_lock(&node->mutex); | |
968 | curr = __btrfs_first_delayed_deletion_item(node); | |
969 | if (!curr) | |
970 | goto delete_fail; | |
971 | ||
972 | ret = btrfs_search_slot(trans, root, &curr->key, path, -1, 1); | |
973 | if (ret < 0) | |
974 | goto delete_fail; | |
975 | else if (ret > 0) { | |
976 | /* | |
977 | * can't find the item which the node points to, so this node | |
978 | * is invalid, just drop it. | |
979 | */ | |
980 | prev = curr; | |
981 | curr = __btrfs_next_delayed_item(prev); | |
982 | btrfs_release_delayed_item(prev); | |
983 | ret = 0; | |
945d8962 | 984 | btrfs_release_path(path); |
62095265 FW |
985 | if (curr) { |
986 | mutex_unlock(&node->mutex); | |
16cdcec7 | 987 | goto do_again; |
62095265 | 988 | } else |
16cdcec7 MX |
989 | goto delete_fail; |
990 | } | |
991 | ||
992 | btrfs_batch_delete_items(trans, root, path, curr); | |
945d8962 | 993 | btrfs_release_path(path); |
16cdcec7 MX |
994 | mutex_unlock(&node->mutex); |
995 | goto do_again; | |
996 | ||
997 | delete_fail: | |
945d8962 | 998 | btrfs_release_path(path); |
16cdcec7 MX |
999 | mutex_unlock(&node->mutex); |
1000 | return ret; | |
1001 | } | |
1002 | ||
1003 | static void btrfs_release_delayed_inode(struct btrfs_delayed_node *delayed_node) | |
1004 | { | |
1005 | struct btrfs_delayed_root *delayed_root; | |
1006 | ||
7cf35d91 MX |
1007 | if (delayed_node && |
1008 | test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) { | |
16cdcec7 | 1009 | BUG_ON(!delayed_node->root); |
7cf35d91 | 1010 | clear_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags); |
16cdcec7 MX |
1011 | delayed_node->count--; |
1012 | ||
1013 | delayed_root = delayed_node->root->fs_info->delayed_root; | |
de3cb945 | 1014 | finish_one_item(delayed_root); |
16cdcec7 MX |
1015 | } |
1016 | } | |
1017 | ||
67de1176 MX |
1018 | static void btrfs_release_delayed_iref(struct btrfs_delayed_node *delayed_node) |
1019 | { | |
1020 | struct btrfs_delayed_root *delayed_root; | |
1021 | ||
1022 | ASSERT(delayed_node->root); | |
1023 | clear_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags); | |
1024 | delayed_node->count--; | |
1025 | ||
1026 | delayed_root = delayed_node->root->fs_info->delayed_root; | |
1027 | finish_one_item(delayed_root); | |
1028 | } | |
1029 | ||
0e8c36a9 MX |
1030 | static int __btrfs_update_delayed_inode(struct btrfs_trans_handle *trans, |
1031 | struct btrfs_root *root, | |
1032 | struct btrfs_path *path, | |
1033 | struct btrfs_delayed_node *node) | |
16cdcec7 MX |
1034 | { |
1035 | struct btrfs_key key; | |
1036 | struct btrfs_inode_item *inode_item; | |
1037 | struct extent_buffer *leaf; | |
67de1176 | 1038 | int mod; |
16cdcec7 MX |
1039 | int ret; |
1040 | ||
16cdcec7 MX |
1041 | key.objectid = node->inode_id; |
1042 | btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY); | |
1043 | key.offset = 0; | |
0e8c36a9 | 1044 | |
67de1176 MX |
1045 | if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags)) |
1046 | mod = -1; | |
1047 | else | |
1048 | mod = 1; | |
1049 | ||
1050 | ret = btrfs_lookup_inode(trans, root, path, &key, mod); | |
16cdcec7 | 1051 | if (ret > 0) { |
945d8962 | 1052 | btrfs_release_path(path); |
16cdcec7 MX |
1053 | return -ENOENT; |
1054 | } else if (ret < 0) { | |
16cdcec7 MX |
1055 | return ret; |
1056 | } | |
1057 | ||
16cdcec7 MX |
1058 | leaf = path->nodes[0]; |
1059 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
1060 | struct btrfs_inode_item); | |
1061 | write_extent_buffer(leaf, &node->inode_item, (unsigned long)inode_item, | |
1062 | sizeof(struct btrfs_inode_item)); | |
1063 | btrfs_mark_buffer_dirty(leaf); | |
16cdcec7 | 1064 | |
67de1176 MX |
1065 | if (!test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags)) |
1066 | goto no_iref; | |
1067 | ||
1068 | path->slots[0]++; | |
1069 | if (path->slots[0] >= btrfs_header_nritems(leaf)) | |
1070 | goto search; | |
1071 | again: | |
1072 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1073 | if (key.objectid != node->inode_id) | |
1074 | goto out; | |
1075 | ||
1076 | if (key.type != BTRFS_INODE_REF_KEY && | |
1077 | key.type != BTRFS_INODE_EXTREF_KEY) | |
1078 | goto out; | |
1079 | ||
1080 | /* | |
1081 | * Delayed iref deletion is for the inode who has only one link, | |
1082 | * so there is only one iref. The case that several irefs are | |
1083 | * in the same item doesn't exist. | |
1084 | */ | |
1085 | btrfs_del_item(trans, root, path); | |
1086 | out: | |
1087 | btrfs_release_delayed_iref(node); | |
1088 | no_iref: | |
1089 | btrfs_release_path(path); | |
1090 | err_out: | |
16cdcec7 MX |
1091 | btrfs_delayed_inode_release_metadata(root, node); |
1092 | btrfs_release_delayed_inode(node); | |
16cdcec7 | 1093 | |
67de1176 MX |
1094 | return ret; |
1095 | ||
1096 | search: | |
1097 | btrfs_release_path(path); | |
1098 | ||
1099 | btrfs_set_key_type(&key, BTRFS_INODE_EXTREF_KEY); | |
1100 | key.offset = -1; | |
1101 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1102 | if (ret < 0) | |
1103 | goto err_out; | |
1104 | ASSERT(ret); | |
1105 | ||
1106 | ret = 0; | |
1107 | leaf = path->nodes[0]; | |
1108 | path->slots[0]--; | |
1109 | goto again; | |
16cdcec7 MX |
1110 | } |
1111 | ||
0e8c36a9 MX |
1112 | static inline int btrfs_update_delayed_inode(struct btrfs_trans_handle *trans, |
1113 | struct btrfs_root *root, | |
1114 | struct btrfs_path *path, | |
1115 | struct btrfs_delayed_node *node) | |
1116 | { | |
1117 | int ret; | |
1118 | ||
1119 | mutex_lock(&node->mutex); | |
7cf35d91 | 1120 | if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &node->flags)) { |
0e8c36a9 MX |
1121 | mutex_unlock(&node->mutex); |
1122 | return 0; | |
1123 | } | |
1124 | ||
1125 | ret = __btrfs_update_delayed_inode(trans, root, path, node); | |
1126 | mutex_unlock(&node->mutex); | |
1127 | return ret; | |
1128 | } | |
1129 | ||
4ea41ce0 MX |
1130 | static inline int |
1131 | __btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans, | |
1132 | struct btrfs_path *path, | |
1133 | struct btrfs_delayed_node *node) | |
1134 | { | |
1135 | int ret; | |
1136 | ||
1137 | ret = btrfs_insert_delayed_items(trans, path, node->root, node); | |
1138 | if (ret) | |
1139 | return ret; | |
1140 | ||
1141 | ret = btrfs_delete_delayed_items(trans, path, node->root, node); | |
1142 | if (ret) | |
1143 | return ret; | |
1144 | ||
1145 | ret = btrfs_update_delayed_inode(trans, node->root, path, node); | |
1146 | return ret; | |
1147 | } | |
1148 | ||
79787eaa JM |
1149 | /* |
1150 | * Called when committing the transaction. | |
1151 | * Returns 0 on success. | |
1152 | * Returns < 0 on error and returns with an aborted transaction with any | |
1153 | * outstanding delayed items cleaned up. | |
1154 | */ | |
96c3f433 JB |
1155 | static int __btrfs_run_delayed_items(struct btrfs_trans_handle *trans, |
1156 | struct btrfs_root *root, int nr) | |
16cdcec7 MX |
1157 | { |
1158 | struct btrfs_delayed_root *delayed_root; | |
1159 | struct btrfs_delayed_node *curr_node, *prev_node; | |
1160 | struct btrfs_path *path; | |
19fd2949 | 1161 | struct btrfs_block_rsv *block_rsv; |
16cdcec7 | 1162 | int ret = 0; |
96c3f433 | 1163 | bool count = (nr > 0); |
16cdcec7 | 1164 | |
79787eaa JM |
1165 | if (trans->aborted) |
1166 | return -EIO; | |
1167 | ||
16cdcec7 MX |
1168 | path = btrfs_alloc_path(); |
1169 | if (!path) | |
1170 | return -ENOMEM; | |
1171 | path->leave_spinning = 1; | |
1172 | ||
19fd2949 | 1173 | block_rsv = trans->block_rsv; |
6d668dda | 1174 | trans->block_rsv = &root->fs_info->delayed_block_rsv; |
19fd2949 | 1175 | |
16cdcec7 MX |
1176 | delayed_root = btrfs_get_delayed_root(root); |
1177 | ||
1178 | curr_node = btrfs_first_delayed_node(delayed_root); | |
96c3f433 | 1179 | while (curr_node && (!count || (count && nr--))) { |
4ea41ce0 MX |
1180 | ret = __btrfs_commit_inode_delayed_items(trans, path, |
1181 | curr_node); | |
16cdcec7 MX |
1182 | if (ret) { |
1183 | btrfs_release_delayed_node(curr_node); | |
96c3f433 | 1184 | curr_node = NULL; |
79787eaa | 1185 | btrfs_abort_transaction(trans, root, ret); |
16cdcec7 MX |
1186 | break; |
1187 | } | |
1188 | ||
1189 | prev_node = curr_node; | |
1190 | curr_node = btrfs_next_delayed_node(curr_node); | |
1191 | btrfs_release_delayed_node(prev_node); | |
1192 | } | |
1193 | ||
96c3f433 JB |
1194 | if (curr_node) |
1195 | btrfs_release_delayed_node(curr_node); | |
16cdcec7 | 1196 | btrfs_free_path(path); |
19fd2949 | 1197 | trans->block_rsv = block_rsv; |
79787eaa | 1198 | |
16cdcec7 MX |
1199 | return ret; |
1200 | } | |
1201 | ||
96c3f433 JB |
1202 | int btrfs_run_delayed_items(struct btrfs_trans_handle *trans, |
1203 | struct btrfs_root *root) | |
1204 | { | |
1205 | return __btrfs_run_delayed_items(trans, root, -1); | |
1206 | } | |
1207 | ||
1208 | int btrfs_run_delayed_items_nr(struct btrfs_trans_handle *trans, | |
1209 | struct btrfs_root *root, int nr) | |
1210 | { | |
1211 | return __btrfs_run_delayed_items(trans, root, nr); | |
1212 | } | |
1213 | ||
16cdcec7 MX |
1214 | int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans, |
1215 | struct inode *inode) | |
1216 | { | |
1217 | struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode); | |
4ea41ce0 MX |
1218 | struct btrfs_path *path; |
1219 | struct btrfs_block_rsv *block_rsv; | |
16cdcec7 MX |
1220 | int ret; |
1221 | ||
1222 | if (!delayed_node) | |
1223 | return 0; | |
1224 | ||
1225 | mutex_lock(&delayed_node->mutex); | |
1226 | if (!delayed_node->count) { | |
1227 | mutex_unlock(&delayed_node->mutex); | |
1228 | btrfs_release_delayed_node(delayed_node); | |
1229 | return 0; | |
1230 | } | |
1231 | mutex_unlock(&delayed_node->mutex); | |
1232 | ||
4ea41ce0 | 1233 | path = btrfs_alloc_path(); |
3c77bd94 FDBM |
1234 | if (!path) { |
1235 | btrfs_release_delayed_node(delayed_node); | |
4ea41ce0 | 1236 | return -ENOMEM; |
3c77bd94 | 1237 | } |
4ea41ce0 MX |
1238 | path->leave_spinning = 1; |
1239 | ||
1240 | block_rsv = trans->block_rsv; | |
1241 | trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv; | |
1242 | ||
1243 | ret = __btrfs_commit_inode_delayed_items(trans, path, delayed_node); | |
1244 | ||
16cdcec7 | 1245 | btrfs_release_delayed_node(delayed_node); |
4ea41ce0 MX |
1246 | btrfs_free_path(path); |
1247 | trans->block_rsv = block_rsv; | |
1248 | ||
16cdcec7 MX |
1249 | return ret; |
1250 | } | |
1251 | ||
0e8c36a9 MX |
1252 | int btrfs_commit_inode_delayed_inode(struct inode *inode) |
1253 | { | |
1254 | struct btrfs_trans_handle *trans; | |
1255 | struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode); | |
1256 | struct btrfs_path *path; | |
1257 | struct btrfs_block_rsv *block_rsv; | |
1258 | int ret; | |
1259 | ||
1260 | if (!delayed_node) | |
1261 | return 0; | |
1262 | ||
1263 | mutex_lock(&delayed_node->mutex); | |
7cf35d91 | 1264 | if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) { |
0e8c36a9 MX |
1265 | mutex_unlock(&delayed_node->mutex); |
1266 | btrfs_release_delayed_node(delayed_node); | |
1267 | return 0; | |
1268 | } | |
1269 | mutex_unlock(&delayed_node->mutex); | |
1270 | ||
1271 | trans = btrfs_join_transaction(delayed_node->root); | |
1272 | if (IS_ERR(trans)) { | |
1273 | ret = PTR_ERR(trans); | |
1274 | goto out; | |
1275 | } | |
1276 | ||
1277 | path = btrfs_alloc_path(); | |
1278 | if (!path) { | |
1279 | ret = -ENOMEM; | |
1280 | goto trans_out; | |
1281 | } | |
1282 | path->leave_spinning = 1; | |
1283 | ||
1284 | block_rsv = trans->block_rsv; | |
1285 | trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv; | |
1286 | ||
1287 | mutex_lock(&delayed_node->mutex); | |
7cf35d91 | 1288 | if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) |
0e8c36a9 MX |
1289 | ret = __btrfs_update_delayed_inode(trans, delayed_node->root, |
1290 | path, delayed_node); | |
1291 | else | |
1292 | ret = 0; | |
1293 | mutex_unlock(&delayed_node->mutex); | |
1294 | ||
1295 | btrfs_free_path(path); | |
1296 | trans->block_rsv = block_rsv; | |
1297 | trans_out: | |
1298 | btrfs_end_transaction(trans, delayed_node->root); | |
1299 | btrfs_btree_balance_dirty(delayed_node->root); | |
1300 | out: | |
1301 | btrfs_release_delayed_node(delayed_node); | |
1302 | ||
1303 | return ret; | |
1304 | } | |
1305 | ||
16cdcec7 MX |
1306 | void btrfs_remove_delayed_node(struct inode *inode) |
1307 | { | |
1308 | struct btrfs_delayed_node *delayed_node; | |
1309 | ||
1310 | delayed_node = ACCESS_ONCE(BTRFS_I(inode)->delayed_node); | |
1311 | if (!delayed_node) | |
1312 | return; | |
1313 | ||
1314 | BTRFS_I(inode)->delayed_node = NULL; | |
1315 | btrfs_release_delayed_node(delayed_node); | |
1316 | } | |
1317 | ||
de3cb945 CM |
1318 | struct btrfs_async_delayed_work { |
1319 | struct btrfs_delayed_root *delayed_root; | |
1320 | int nr; | |
d458b054 | 1321 | struct btrfs_work work; |
16cdcec7 MX |
1322 | }; |
1323 | ||
d458b054 | 1324 | static void btrfs_async_run_delayed_root(struct btrfs_work *work) |
16cdcec7 | 1325 | { |
de3cb945 CM |
1326 | struct btrfs_async_delayed_work *async_work; |
1327 | struct btrfs_delayed_root *delayed_root; | |
16cdcec7 MX |
1328 | struct btrfs_trans_handle *trans; |
1329 | struct btrfs_path *path; | |
1330 | struct btrfs_delayed_node *delayed_node = NULL; | |
1331 | struct btrfs_root *root; | |
19fd2949 | 1332 | struct btrfs_block_rsv *block_rsv; |
de3cb945 | 1333 | int total_done = 0; |
16cdcec7 | 1334 | |
de3cb945 CM |
1335 | async_work = container_of(work, struct btrfs_async_delayed_work, work); |
1336 | delayed_root = async_work->delayed_root; | |
16cdcec7 MX |
1337 | |
1338 | path = btrfs_alloc_path(); | |
1339 | if (!path) | |
1340 | goto out; | |
16cdcec7 | 1341 | |
de3cb945 CM |
1342 | again: |
1343 | if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND / 2) | |
1344 | goto free_path; | |
1345 | ||
1346 | delayed_node = btrfs_first_prepared_delayed_node(delayed_root); | |
1347 | if (!delayed_node) | |
1348 | goto free_path; | |
1349 | ||
1350 | path->leave_spinning = 1; | |
16cdcec7 MX |
1351 | root = delayed_node->root; |
1352 | ||
ff5714cc | 1353 | trans = btrfs_join_transaction(root); |
16cdcec7 | 1354 | if (IS_ERR(trans)) |
de3cb945 | 1355 | goto release_path; |
16cdcec7 | 1356 | |
19fd2949 | 1357 | block_rsv = trans->block_rsv; |
6d668dda | 1358 | trans->block_rsv = &root->fs_info->delayed_block_rsv; |
19fd2949 | 1359 | |
4ea41ce0 | 1360 | __btrfs_commit_inode_delayed_items(trans, path, delayed_node); |
16cdcec7 | 1361 | |
19fd2949 | 1362 | trans->block_rsv = block_rsv; |
a56dbd89 | 1363 | btrfs_end_transaction(trans, root); |
b53d3f5d | 1364 | btrfs_btree_balance_dirty_nodelay(root); |
de3cb945 CM |
1365 | |
1366 | release_path: | |
1367 | btrfs_release_path(path); | |
1368 | total_done++; | |
1369 | ||
1370 | btrfs_release_prepared_delayed_node(delayed_node); | |
1371 | if (async_work->nr == 0 || total_done < async_work->nr) | |
1372 | goto again; | |
1373 | ||
16cdcec7 MX |
1374 | free_path: |
1375 | btrfs_free_path(path); | |
1376 | out: | |
de3cb945 CM |
1377 | wake_up(&delayed_root->wait); |
1378 | kfree(async_work); | |
16cdcec7 MX |
1379 | } |
1380 | ||
de3cb945 | 1381 | |
16cdcec7 | 1382 | static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root, |
de3cb945 | 1383 | struct btrfs_root *root, int nr) |
16cdcec7 | 1384 | { |
de3cb945 | 1385 | struct btrfs_async_delayed_work *async_work; |
16cdcec7 | 1386 | |
de3cb945 | 1387 | if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND) |
16cdcec7 MX |
1388 | return 0; |
1389 | ||
de3cb945 CM |
1390 | async_work = kmalloc(sizeof(*async_work), GFP_NOFS); |
1391 | if (!async_work) | |
16cdcec7 | 1392 | return -ENOMEM; |
16cdcec7 | 1393 | |
de3cb945 | 1394 | async_work->delayed_root = delayed_root; |
5b3bc44e QW |
1395 | btrfs_init_work(&async_work->work, btrfs_async_run_delayed_root, |
1396 | NULL, NULL); | |
de3cb945 | 1397 | async_work->nr = nr; |
16cdcec7 | 1398 | |
5b3bc44e | 1399 | btrfs_queue_work(root->fs_info->delayed_workers, &async_work->work); |
16cdcec7 MX |
1400 | return 0; |
1401 | } | |
1402 | ||
e999376f CM |
1403 | void btrfs_assert_delayed_root_empty(struct btrfs_root *root) |
1404 | { | |
1405 | struct btrfs_delayed_root *delayed_root; | |
1406 | delayed_root = btrfs_get_delayed_root(root); | |
1407 | WARN_ON(btrfs_first_delayed_node(delayed_root)); | |
1408 | } | |
1409 | ||
0353808c | 1410 | static int could_end_wait(struct btrfs_delayed_root *delayed_root, int seq) |
de3cb945 CM |
1411 | { |
1412 | int val = atomic_read(&delayed_root->items_seq); | |
1413 | ||
0353808c | 1414 | if (val < seq || val >= seq + BTRFS_DELAYED_BATCH) |
de3cb945 | 1415 | return 1; |
0353808c MX |
1416 | |
1417 | if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND) | |
1418 | return 1; | |
1419 | ||
de3cb945 CM |
1420 | return 0; |
1421 | } | |
1422 | ||
16cdcec7 MX |
1423 | void btrfs_balance_delayed_items(struct btrfs_root *root) |
1424 | { | |
1425 | struct btrfs_delayed_root *delayed_root; | |
1426 | ||
1427 | delayed_root = btrfs_get_delayed_root(root); | |
1428 | ||
1429 | if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND) | |
1430 | return; | |
1431 | ||
1432 | if (atomic_read(&delayed_root->items) >= BTRFS_DELAYED_WRITEBACK) { | |
0353808c | 1433 | int seq; |
16cdcec7 | 1434 | int ret; |
0353808c MX |
1435 | |
1436 | seq = atomic_read(&delayed_root->items_seq); | |
de3cb945 CM |
1437 | |
1438 | ret = btrfs_wq_run_delayed_node(delayed_root, root, 0); | |
16cdcec7 MX |
1439 | if (ret) |
1440 | return; | |
1441 | ||
0353808c MX |
1442 | wait_event_interruptible(delayed_root->wait, |
1443 | could_end_wait(delayed_root, seq)); | |
4dd466d3 | 1444 | return; |
16cdcec7 MX |
1445 | } |
1446 | ||
de3cb945 | 1447 | btrfs_wq_run_delayed_node(delayed_root, root, BTRFS_DELAYED_BATCH); |
16cdcec7 MX |
1448 | } |
1449 | ||
79787eaa | 1450 | /* Will return 0 or -ENOMEM */ |
16cdcec7 MX |
1451 | int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans, |
1452 | struct btrfs_root *root, const char *name, | |
1453 | int name_len, struct inode *dir, | |
1454 | struct btrfs_disk_key *disk_key, u8 type, | |
1455 | u64 index) | |
1456 | { | |
1457 | struct btrfs_delayed_node *delayed_node; | |
1458 | struct btrfs_delayed_item *delayed_item; | |
1459 | struct btrfs_dir_item *dir_item; | |
1460 | int ret; | |
1461 | ||
1462 | delayed_node = btrfs_get_or_create_delayed_node(dir); | |
1463 | if (IS_ERR(delayed_node)) | |
1464 | return PTR_ERR(delayed_node); | |
1465 | ||
1466 | delayed_item = btrfs_alloc_delayed_item(sizeof(*dir_item) + name_len); | |
1467 | if (!delayed_item) { | |
1468 | ret = -ENOMEM; | |
1469 | goto release_node; | |
1470 | } | |
1471 | ||
0d0ca30f | 1472 | delayed_item->key.objectid = btrfs_ino(dir); |
16cdcec7 MX |
1473 | btrfs_set_key_type(&delayed_item->key, BTRFS_DIR_INDEX_KEY); |
1474 | delayed_item->key.offset = index; | |
1475 | ||
1476 | dir_item = (struct btrfs_dir_item *)delayed_item->data; | |
1477 | dir_item->location = *disk_key; | |
3cae210f QW |
1478 | btrfs_set_stack_dir_transid(dir_item, trans->transid); |
1479 | btrfs_set_stack_dir_data_len(dir_item, 0); | |
1480 | btrfs_set_stack_dir_name_len(dir_item, name_len); | |
1481 | btrfs_set_stack_dir_type(dir_item, type); | |
16cdcec7 MX |
1482 | memcpy((char *)(dir_item + 1), name, name_len); |
1483 | ||
8c2a3ca2 JB |
1484 | ret = btrfs_delayed_item_reserve_metadata(trans, root, delayed_item); |
1485 | /* | |
1486 | * we have reserved enough space when we start a new transaction, | |
1487 | * so reserving metadata failure is impossible | |
1488 | */ | |
1489 | BUG_ON(ret); | |
1490 | ||
1491 | ||
16cdcec7 MX |
1492 | mutex_lock(&delayed_node->mutex); |
1493 | ret = __btrfs_add_delayed_insertion_item(delayed_node, delayed_item); | |
1494 | if (unlikely(ret)) { | |
efe120a0 | 1495 | btrfs_err(root->fs_info, "err add delayed dir index item(name: %.*s) " |
bdab49d7 | 1496 | "into the insertion tree of the delayed node" |
efe120a0 | 1497 | "(root id: %llu, inode id: %llu, errno: %d)", |
bdab49d7 | 1498 | name_len, name, delayed_node->root->objectid, |
c1c9ff7c | 1499 | delayed_node->inode_id, ret); |
16cdcec7 MX |
1500 | BUG(); |
1501 | } | |
1502 | mutex_unlock(&delayed_node->mutex); | |
1503 | ||
1504 | release_node: | |
1505 | btrfs_release_delayed_node(delayed_node); | |
1506 | return ret; | |
1507 | } | |
1508 | ||
1509 | static int btrfs_delete_delayed_insertion_item(struct btrfs_root *root, | |
1510 | struct btrfs_delayed_node *node, | |
1511 | struct btrfs_key *key) | |
1512 | { | |
1513 | struct btrfs_delayed_item *item; | |
1514 | ||
1515 | mutex_lock(&node->mutex); | |
1516 | item = __btrfs_lookup_delayed_insertion_item(node, key); | |
1517 | if (!item) { | |
1518 | mutex_unlock(&node->mutex); | |
1519 | return 1; | |
1520 | } | |
1521 | ||
1522 | btrfs_delayed_item_release_metadata(root, item); | |
1523 | btrfs_release_delayed_item(item); | |
1524 | mutex_unlock(&node->mutex); | |
1525 | return 0; | |
1526 | } | |
1527 | ||
1528 | int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans, | |
1529 | struct btrfs_root *root, struct inode *dir, | |
1530 | u64 index) | |
1531 | { | |
1532 | struct btrfs_delayed_node *node; | |
1533 | struct btrfs_delayed_item *item; | |
1534 | struct btrfs_key item_key; | |
1535 | int ret; | |
1536 | ||
1537 | node = btrfs_get_or_create_delayed_node(dir); | |
1538 | if (IS_ERR(node)) | |
1539 | return PTR_ERR(node); | |
1540 | ||
0d0ca30f | 1541 | item_key.objectid = btrfs_ino(dir); |
16cdcec7 MX |
1542 | btrfs_set_key_type(&item_key, BTRFS_DIR_INDEX_KEY); |
1543 | item_key.offset = index; | |
1544 | ||
1545 | ret = btrfs_delete_delayed_insertion_item(root, node, &item_key); | |
1546 | if (!ret) | |
1547 | goto end; | |
1548 | ||
1549 | item = btrfs_alloc_delayed_item(0); | |
1550 | if (!item) { | |
1551 | ret = -ENOMEM; | |
1552 | goto end; | |
1553 | } | |
1554 | ||
1555 | item->key = item_key; | |
1556 | ||
1557 | ret = btrfs_delayed_item_reserve_metadata(trans, root, item); | |
1558 | /* | |
1559 | * we have reserved enough space when we start a new transaction, | |
1560 | * so reserving metadata failure is impossible. | |
1561 | */ | |
1562 | BUG_ON(ret); | |
1563 | ||
1564 | mutex_lock(&node->mutex); | |
1565 | ret = __btrfs_add_delayed_deletion_item(node, item); | |
1566 | if (unlikely(ret)) { | |
efe120a0 | 1567 | btrfs_err(root->fs_info, "err add delayed dir index item(index: %llu) " |
16cdcec7 | 1568 | "into the deletion tree of the delayed node" |
efe120a0 | 1569 | "(root id: %llu, inode id: %llu, errno: %d)", |
c1c9ff7c | 1570 | index, node->root->objectid, node->inode_id, |
16cdcec7 MX |
1571 | ret); |
1572 | BUG(); | |
1573 | } | |
1574 | mutex_unlock(&node->mutex); | |
1575 | end: | |
1576 | btrfs_release_delayed_node(node); | |
1577 | return ret; | |
1578 | } | |
1579 | ||
1580 | int btrfs_inode_delayed_dir_index_count(struct inode *inode) | |
1581 | { | |
2f7e33d4 | 1582 | struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode); |
16cdcec7 MX |
1583 | |
1584 | if (!delayed_node) | |
1585 | return -ENOENT; | |
1586 | ||
1587 | /* | |
1588 | * Since we have held i_mutex of this directory, it is impossible that | |
1589 | * a new directory index is added into the delayed node and index_cnt | |
1590 | * is updated now. So we needn't lock the delayed node. | |
1591 | */ | |
2f7e33d4 MX |
1592 | if (!delayed_node->index_cnt) { |
1593 | btrfs_release_delayed_node(delayed_node); | |
16cdcec7 | 1594 | return -EINVAL; |
2f7e33d4 | 1595 | } |
16cdcec7 MX |
1596 | |
1597 | BTRFS_I(inode)->index_cnt = delayed_node->index_cnt; | |
2f7e33d4 MX |
1598 | btrfs_release_delayed_node(delayed_node); |
1599 | return 0; | |
16cdcec7 MX |
1600 | } |
1601 | ||
1602 | void btrfs_get_delayed_items(struct inode *inode, struct list_head *ins_list, | |
1603 | struct list_head *del_list) | |
1604 | { | |
1605 | struct btrfs_delayed_node *delayed_node; | |
1606 | struct btrfs_delayed_item *item; | |
1607 | ||
1608 | delayed_node = btrfs_get_delayed_node(inode); | |
1609 | if (!delayed_node) | |
1610 | return; | |
1611 | ||
1612 | mutex_lock(&delayed_node->mutex); | |
1613 | item = __btrfs_first_delayed_insertion_item(delayed_node); | |
1614 | while (item) { | |
1615 | atomic_inc(&item->refs); | |
1616 | list_add_tail(&item->readdir_list, ins_list); | |
1617 | item = __btrfs_next_delayed_item(item); | |
1618 | } | |
1619 | ||
1620 | item = __btrfs_first_delayed_deletion_item(delayed_node); | |
1621 | while (item) { | |
1622 | atomic_inc(&item->refs); | |
1623 | list_add_tail(&item->readdir_list, del_list); | |
1624 | item = __btrfs_next_delayed_item(item); | |
1625 | } | |
1626 | mutex_unlock(&delayed_node->mutex); | |
1627 | /* | |
1628 | * This delayed node is still cached in the btrfs inode, so refs | |
1629 | * must be > 1 now, and we needn't check it is going to be freed | |
1630 | * or not. | |
1631 | * | |
1632 | * Besides that, this function is used to read dir, we do not | |
1633 | * insert/delete delayed items in this period. So we also needn't | |
1634 | * requeue or dequeue this delayed node. | |
1635 | */ | |
1636 | atomic_dec(&delayed_node->refs); | |
1637 | } | |
1638 | ||
1639 | void btrfs_put_delayed_items(struct list_head *ins_list, | |
1640 | struct list_head *del_list) | |
1641 | { | |
1642 | struct btrfs_delayed_item *curr, *next; | |
1643 | ||
1644 | list_for_each_entry_safe(curr, next, ins_list, readdir_list) { | |
1645 | list_del(&curr->readdir_list); | |
1646 | if (atomic_dec_and_test(&curr->refs)) | |
1647 | kfree(curr); | |
1648 | } | |
1649 | ||
1650 | list_for_each_entry_safe(curr, next, del_list, readdir_list) { | |
1651 | list_del(&curr->readdir_list); | |
1652 | if (atomic_dec_and_test(&curr->refs)) | |
1653 | kfree(curr); | |
1654 | } | |
1655 | } | |
1656 | ||
1657 | int btrfs_should_delete_dir_index(struct list_head *del_list, | |
1658 | u64 index) | |
1659 | { | |
1660 | struct btrfs_delayed_item *curr, *next; | |
1661 | int ret; | |
1662 | ||
1663 | if (list_empty(del_list)) | |
1664 | return 0; | |
1665 | ||
1666 | list_for_each_entry_safe(curr, next, del_list, readdir_list) { | |
1667 | if (curr->key.offset > index) | |
1668 | break; | |
1669 | ||
1670 | list_del(&curr->readdir_list); | |
1671 | ret = (curr->key.offset == index); | |
1672 | ||
1673 | if (atomic_dec_and_test(&curr->refs)) | |
1674 | kfree(curr); | |
1675 | ||
1676 | if (ret) | |
1677 | return 1; | |
1678 | else | |
1679 | continue; | |
1680 | } | |
1681 | return 0; | |
1682 | } | |
1683 | ||
1684 | /* | |
1685 | * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree | |
1686 | * | |
1687 | */ | |
9cdda8d3 | 1688 | int btrfs_readdir_delayed_dir_index(struct dir_context *ctx, |
16cdcec7 MX |
1689 | struct list_head *ins_list) |
1690 | { | |
1691 | struct btrfs_dir_item *di; | |
1692 | struct btrfs_delayed_item *curr, *next; | |
1693 | struct btrfs_key location; | |
1694 | char *name; | |
1695 | int name_len; | |
1696 | int over = 0; | |
1697 | unsigned char d_type; | |
1698 | ||
1699 | if (list_empty(ins_list)) | |
1700 | return 0; | |
1701 | ||
1702 | /* | |
1703 | * Changing the data of the delayed item is impossible. So | |
1704 | * we needn't lock them. And we have held i_mutex of the | |
1705 | * directory, nobody can delete any directory indexes now. | |
1706 | */ | |
1707 | list_for_each_entry_safe(curr, next, ins_list, readdir_list) { | |
1708 | list_del(&curr->readdir_list); | |
1709 | ||
9cdda8d3 | 1710 | if (curr->key.offset < ctx->pos) { |
16cdcec7 MX |
1711 | if (atomic_dec_and_test(&curr->refs)) |
1712 | kfree(curr); | |
1713 | continue; | |
1714 | } | |
1715 | ||
9cdda8d3 | 1716 | ctx->pos = curr->key.offset; |
16cdcec7 MX |
1717 | |
1718 | di = (struct btrfs_dir_item *)curr->data; | |
1719 | name = (char *)(di + 1); | |
3cae210f | 1720 | name_len = btrfs_stack_dir_name_len(di); |
16cdcec7 MX |
1721 | |
1722 | d_type = btrfs_filetype_table[di->type]; | |
1723 | btrfs_disk_key_to_cpu(&location, &di->location); | |
1724 | ||
9cdda8d3 | 1725 | over = !dir_emit(ctx, name, name_len, |
16cdcec7 MX |
1726 | location.objectid, d_type); |
1727 | ||
1728 | if (atomic_dec_and_test(&curr->refs)) | |
1729 | kfree(curr); | |
1730 | ||
1731 | if (over) | |
1732 | return 1; | |
1733 | } | |
1734 | return 0; | |
1735 | } | |
1736 | ||
16cdcec7 MX |
1737 | static void fill_stack_inode_item(struct btrfs_trans_handle *trans, |
1738 | struct btrfs_inode_item *inode_item, | |
1739 | struct inode *inode) | |
1740 | { | |
2f2f43d3 EB |
1741 | btrfs_set_stack_inode_uid(inode_item, i_uid_read(inode)); |
1742 | btrfs_set_stack_inode_gid(inode_item, i_gid_read(inode)); | |
16cdcec7 MX |
1743 | btrfs_set_stack_inode_size(inode_item, BTRFS_I(inode)->disk_i_size); |
1744 | btrfs_set_stack_inode_mode(inode_item, inode->i_mode); | |
1745 | btrfs_set_stack_inode_nlink(inode_item, inode->i_nlink); | |
1746 | btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(inode)); | |
1747 | btrfs_set_stack_inode_generation(inode_item, | |
1748 | BTRFS_I(inode)->generation); | |
0c4d2d95 | 1749 | btrfs_set_stack_inode_sequence(inode_item, inode->i_version); |
16cdcec7 MX |
1750 | btrfs_set_stack_inode_transid(inode_item, trans->transid); |
1751 | btrfs_set_stack_inode_rdev(inode_item, inode->i_rdev); | |
1752 | btrfs_set_stack_inode_flags(inode_item, BTRFS_I(inode)->flags); | |
ff5714cc | 1753 | btrfs_set_stack_inode_block_group(inode_item, 0); |
16cdcec7 MX |
1754 | |
1755 | btrfs_set_stack_timespec_sec(btrfs_inode_atime(inode_item), | |
1756 | inode->i_atime.tv_sec); | |
1757 | btrfs_set_stack_timespec_nsec(btrfs_inode_atime(inode_item), | |
1758 | inode->i_atime.tv_nsec); | |
1759 | ||
1760 | btrfs_set_stack_timespec_sec(btrfs_inode_mtime(inode_item), | |
1761 | inode->i_mtime.tv_sec); | |
1762 | btrfs_set_stack_timespec_nsec(btrfs_inode_mtime(inode_item), | |
1763 | inode->i_mtime.tv_nsec); | |
1764 | ||
1765 | btrfs_set_stack_timespec_sec(btrfs_inode_ctime(inode_item), | |
1766 | inode->i_ctime.tv_sec); | |
1767 | btrfs_set_stack_timespec_nsec(btrfs_inode_ctime(inode_item), | |
1768 | inode->i_ctime.tv_nsec); | |
1769 | } | |
1770 | ||
2f7e33d4 MX |
1771 | int btrfs_fill_inode(struct inode *inode, u32 *rdev) |
1772 | { | |
1773 | struct btrfs_delayed_node *delayed_node; | |
1774 | struct btrfs_inode_item *inode_item; | |
1775 | struct btrfs_timespec *tspec; | |
1776 | ||
1777 | delayed_node = btrfs_get_delayed_node(inode); | |
1778 | if (!delayed_node) | |
1779 | return -ENOENT; | |
1780 | ||
1781 | mutex_lock(&delayed_node->mutex); | |
7cf35d91 | 1782 | if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) { |
2f7e33d4 MX |
1783 | mutex_unlock(&delayed_node->mutex); |
1784 | btrfs_release_delayed_node(delayed_node); | |
1785 | return -ENOENT; | |
1786 | } | |
1787 | ||
1788 | inode_item = &delayed_node->inode_item; | |
1789 | ||
2f2f43d3 EB |
1790 | i_uid_write(inode, btrfs_stack_inode_uid(inode_item)); |
1791 | i_gid_write(inode, btrfs_stack_inode_gid(inode_item)); | |
2f7e33d4 MX |
1792 | btrfs_i_size_write(inode, btrfs_stack_inode_size(inode_item)); |
1793 | inode->i_mode = btrfs_stack_inode_mode(inode_item); | |
bfe86848 | 1794 | set_nlink(inode, btrfs_stack_inode_nlink(inode_item)); |
2f7e33d4 MX |
1795 | inode_set_bytes(inode, btrfs_stack_inode_nbytes(inode_item)); |
1796 | BTRFS_I(inode)->generation = btrfs_stack_inode_generation(inode_item); | |
0c4d2d95 | 1797 | inode->i_version = btrfs_stack_inode_sequence(inode_item); |
2f7e33d4 MX |
1798 | inode->i_rdev = 0; |
1799 | *rdev = btrfs_stack_inode_rdev(inode_item); | |
1800 | BTRFS_I(inode)->flags = btrfs_stack_inode_flags(inode_item); | |
1801 | ||
1802 | tspec = btrfs_inode_atime(inode_item); | |
1803 | inode->i_atime.tv_sec = btrfs_stack_timespec_sec(tspec); | |
1804 | inode->i_atime.tv_nsec = btrfs_stack_timespec_nsec(tspec); | |
1805 | ||
1806 | tspec = btrfs_inode_mtime(inode_item); | |
1807 | inode->i_mtime.tv_sec = btrfs_stack_timespec_sec(tspec); | |
1808 | inode->i_mtime.tv_nsec = btrfs_stack_timespec_nsec(tspec); | |
1809 | ||
1810 | tspec = btrfs_inode_ctime(inode_item); | |
1811 | inode->i_ctime.tv_sec = btrfs_stack_timespec_sec(tspec); | |
1812 | inode->i_ctime.tv_nsec = btrfs_stack_timespec_nsec(tspec); | |
1813 | ||
1814 | inode->i_generation = BTRFS_I(inode)->generation; | |
1815 | BTRFS_I(inode)->index_cnt = (u64)-1; | |
1816 | ||
1817 | mutex_unlock(&delayed_node->mutex); | |
1818 | btrfs_release_delayed_node(delayed_node); | |
1819 | return 0; | |
1820 | } | |
1821 | ||
16cdcec7 MX |
1822 | int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans, |
1823 | struct btrfs_root *root, struct inode *inode) | |
1824 | { | |
1825 | struct btrfs_delayed_node *delayed_node; | |
aa0467d8 | 1826 | int ret = 0; |
16cdcec7 MX |
1827 | |
1828 | delayed_node = btrfs_get_or_create_delayed_node(inode); | |
1829 | if (IS_ERR(delayed_node)) | |
1830 | return PTR_ERR(delayed_node); | |
1831 | ||
1832 | mutex_lock(&delayed_node->mutex); | |
7cf35d91 | 1833 | if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) { |
16cdcec7 MX |
1834 | fill_stack_inode_item(trans, &delayed_node->inode_item, inode); |
1835 | goto release_node; | |
1836 | } | |
1837 | ||
7fd2ae21 JB |
1838 | ret = btrfs_delayed_inode_reserve_metadata(trans, root, inode, |
1839 | delayed_node); | |
c06a0e12 JB |
1840 | if (ret) |
1841 | goto release_node; | |
16cdcec7 MX |
1842 | |
1843 | fill_stack_inode_item(trans, &delayed_node->inode_item, inode); | |
7cf35d91 | 1844 | set_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags); |
16cdcec7 MX |
1845 | delayed_node->count++; |
1846 | atomic_inc(&root->fs_info->delayed_root->items); | |
1847 | release_node: | |
1848 | mutex_unlock(&delayed_node->mutex); | |
1849 | btrfs_release_delayed_node(delayed_node); | |
1850 | return ret; | |
1851 | } | |
1852 | ||
67de1176 MX |
1853 | int btrfs_delayed_delete_inode_ref(struct inode *inode) |
1854 | { | |
1855 | struct btrfs_delayed_node *delayed_node; | |
1856 | ||
1857 | delayed_node = btrfs_get_or_create_delayed_node(inode); | |
1858 | if (IS_ERR(delayed_node)) | |
1859 | return PTR_ERR(delayed_node); | |
1860 | ||
1861 | /* | |
1862 | * We don't reserve space for inode ref deletion is because: | |
1863 | * - We ONLY do async inode ref deletion for the inode who has only | |
1864 | * one link(i_nlink == 1), it means there is only one inode ref. | |
1865 | * And in most case, the inode ref and the inode item are in the | |
1866 | * same leaf, and we will deal with them at the same time. | |
1867 | * Since we are sure we will reserve the space for the inode item, | |
1868 | * it is unnecessary to reserve space for inode ref deletion. | |
1869 | * - If the inode ref and the inode item are not in the same leaf, | |
1870 | * We also needn't worry about enospc problem, because we reserve | |
1871 | * much more space for the inode update than it needs. | |
1872 | * - At the worst, we can steal some space from the global reservation. | |
1873 | * It is very rare. | |
1874 | */ | |
1875 | mutex_lock(&delayed_node->mutex); | |
1876 | if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags)) | |
1877 | goto release_node; | |
1878 | ||
1879 | set_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags); | |
1880 | delayed_node->count++; | |
1881 | atomic_inc(&BTRFS_I(inode)->root->fs_info->delayed_root->items); | |
1882 | release_node: | |
1883 | mutex_unlock(&delayed_node->mutex); | |
1884 | btrfs_release_delayed_node(delayed_node); | |
1885 | return 0; | |
1886 | } | |
1887 | ||
16cdcec7 MX |
1888 | static void __btrfs_kill_delayed_node(struct btrfs_delayed_node *delayed_node) |
1889 | { | |
1890 | struct btrfs_root *root = delayed_node->root; | |
1891 | struct btrfs_delayed_item *curr_item, *prev_item; | |
1892 | ||
1893 | mutex_lock(&delayed_node->mutex); | |
1894 | curr_item = __btrfs_first_delayed_insertion_item(delayed_node); | |
1895 | while (curr_item) { | |
1896 | btrfs_delayed_item_release_metadata(root, curr_item); | |
1897 | prev_item = curr_item; | |
1898 | curr_item = __btrfs_next_delayed_item(prev_item); | |
1899 | btrfs_release_delayed_item(prev_item); | |
1900 | } | |
1901 | ||
1902 | curr_item = __btrfs_first_delayed_deletion_item(delayed_node); | |
1903 | while (curr_item) { | |
1904 | btrfs_delayed_item_release_metadata(root, curr_item); | |
1905 | prev_item = curr_item; | |
1906 | curr_item = __btrfs_next_delayed_item(prev_item); | |
1907 | btrfs_release_delayed_item(prev_item); | |
1908 | } | |
1909 | ||
67de1176 MX |
1910 | if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags)) |
1911 | btrfs_release_delayed_iref(delayed_node); | |
1912 | ||
7cf35d91 | 1913 | if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) { |
16cdcec7 MX |
1914 | btrfs_delayed_inode_release_metadata(root, delayed_node); |
1915 | btrfs_release_delayed_inode(delayed_node); | |
1916 | } | |
1917 | mutex_unlock(&delayed_node->mutex); | |
1918 | } | |
1919 | ||
1920 | void btrfs_kill_delayed_inode_items(struct inode *inode) | |
1921 | { | |
1922 | struct btrfs_delayed_node *delayed_node; | |
1923 | ||
1924 | delayed_node = btrfs_get_delayed_node(inode); | |
1925 | if (!delayed_node) | |
1926 | return; | |
1927 | ||
1928 | __btrfs_kill_delayed_node(delayed_node); | |
1929 | btrfs_release_delayed_node(delayed_node); | |
1930 | } | |
1931 | ||
1932 | void btrfs_kill_all_delayed_nodes(struct btrfs_root *root) | |
1933 | { | |
1934 | u64 inode_id = 0; | |
1935 | struct btrfs_delayed_node *delayed_nodes[8]; | |
1936 | int i, n; | |
1937 | ||
1938 | while (1) { | |
1939 | spin_lock(&root->inode_lock); | |
1940 | n = radix_tree_gang_lookup(&root->delayed_nodes_tree, | |
1941 | (void **)delayed_nodes, inode_id, | |
1942 | ARRAY_SIZE(delayed_nodes)); | |
1943 | if (!n) { | |
1944 | spin_unlock(&root->inode_lock); | |
1945 | break; | |
1946 | } | |
1947 | ||
1948 | inode_id = delayed_nodes[n - 1]->inode_id + 1; | |
1949 | ||
1950 | for (i = 0; i < n; i++) | |
1951 | atomic_inc(&delayed_nodes[i]->refs); | |
1952 | spin_unlock(&root->inode_lock); | |
1953 | ||
1954 | for (i = 0; i < n; i++) { | |
1955 | __btrfs_kill_delayed_node(delayed_nodes[i]); | |
1956 | btrfs_release_delayed_node(delayed_nodes[i]); | |
1957 | } | |
1958 | } | |
1959 | } | |
67cde344 MX |
1960 | |
1961 | void btrfs_destroy_delayed_inodes(struct btrfs_root *root) | |
1962 | { | |
1963 | struct btrfs_delayed_root *delayed_root; | |
1964 | struct btrfs_delayed_node *curr_node, *prev_node; | |
1965 | ||
1966 | delayed_root = btrfs_get_delayed_root(root); | |
1967 | ||
1968 | curr_node = btrfs_first_delayed_node(delayed_root); | |
1969 | while (curr_node) { | |
1970 | __btrfs_kill_delayed_node(curr_node); | |
1971 | ||
1972 | prev_node = curr_node; | |
1973 | curr_node = btrfs_next_delayed_node(curr_node); | |
1974 | btrfs_release_delayed_node(prev_node); | |
1975 | } | |
1976 | } | |
1977 |