]>
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" | |
24 | ||
25 | #define BTRFS_DELAYED_WRITEBACK 400 | |
26 | #define BTRFS_DELAYED_BACKGROUND 100 | |
27 | ||
28 | static struct kmem_cache *delayed_node_cache; | |
29 | ||
30 | int __init btrfs_delayed_inode_init(void) | |
31 | { | |
837e1972 | 32 | delayed_node_cache = kmem_cache_create("btrfs_delayed_node", |
16cdcec7 MX |
33 | sizeof(struct btrfs_delayed_node), |
34 | 0, | |
35 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, | |
36 | NULL); | |
37 | if (!delayed_node_cache) | |
38 | return -ENOMEM; | |
39 | return 0; | |
40 | } | |
41 | ||
42 | void btrfs_delayed_inode_exit(void) | |
43 | { | |
44 | if (delayed_node_cache) | |
45 | kmem_cache_destroy(delayed_node_cache); | |
46 | } | |
47 | ||
48 | static inline void btrfs_init_delayed_node( | |
49 | struct btrfs_delayed_node *delayed_node, | |
50 | struct btrfs_root *root, u64 inode_id) | |
51 | { | |
52 | delayed_node->root = root; | |
53 | delayed_node->inode_id = inode_id; | |
54 | atomic_set(&delayed_node->refs, 0); | |
55 | delayed_node->count = 0; | |
56 | delayed_node->in_list = 0; | |
57 | delayed_node->inode_dirty = 0; | |
58 | delayed_node->ins_root = RB_ROOT; | |
59 | delayed_node->del_root = RB_ROOT; | |
60 | mutex_init(&delayed_node->mutex); | |
61 | delayed_node->index_cnt = 0; | |
62 | INIT_LIST_HEAD(&delayed_node->n_list); | |
63 | INIT_LIST_HEAD(&delayed_node->p_list); | |
64 | delayed_node->bytes_reserved = 0; | |
293f7e07 | 65 | memset(&delayed_node->inode_item, 0, sizeof(delayed_node->inode_item)); |
16cdcec7 MX |
66 | } |
67 | ||
68 | static inline int btrfs_is_continuous_delayed_item( | |
69 | struct btrfs_delayed_item *item1, | |
70 | struct btrfs_delayed_item *item2) | |
71 | { | |
72 | if (item1->key.type == BTRFS_DIR_INDEX_KEY && | |
73 | item1->key.objectid == item2->key.objectid && | |
74 | item1->key.type == item2->key.type && | |
75 | item1->key.offset + 1 == item2->key.offset) | |
76 | return 1; | |
77 | return 0; | |
78 | } | |
79 | ||
80 | static inline struct btrfs_delayed_root *btrfs_get_delayed_root( | |
81 | struct btrfs_root *root) | |
82 | { | |
83 | return root->fs_info->delayed_root; | |
84 | } | |
85 | ||
2f7e33d4 | 86 | static struct btrfs_delayed_node *btrfs_get_delayed_node(struct inode *inode) |
16cdcec7 | 87 | { |
16cdcec7 MX |
88 | struct btrfs_inode *btrfs_inode = BTRFS_I(inode); |
89 | struct btrfs_root *root = btrfs_inode->root; | |
0d0ca30f | 90 | u64 ino = btrfs_ino(inode); |
2f7e33d4 | 91 | struct btrfs_delayed_node *node; |
16cdcec7 | 92 | |
16cdcec7 MX |
93 | node = ACCESS_ONCE(btrfs_inode->delayed_node); |
94 | if (node) { | |
2f7e33d4 | 95 | atomic_inc(&node->refs); |
16cdcec7 MX |
96 | return node; |
97 | } | |
98 | ||
99 | spin_lock(&root->inode_lock); | |
0d0ca30f | 100 | node = radix_tree_lookup(&root->delayed_nodes_tree, ino); |
16cdcec7 MX |
101 | if (node) { |
102 | if (btrfs_inode->delayed_node) { | |
2f7e33d4 MX |
103 | atomic_inc(&node->refs); /* can be accessed */ |
104 | BUG_ON(btrfs_inode->delayed_node != node); | |
16cdcec7 | 105 | spin_unlock(&root->inode_lock); |
2f7e33d4 | 106 | return node; |
16cdcec7 MX |
107 | } |
108 | btrfs_inode->delayed_node = node; | |
109 | atomic_inc(&node->refs); /* can be accessed */ | |
110 | atomic_inc(&node->refs); /* cached in the inode */ | |
111 | spin_unlock(&root->inode_lock); | |
112 | return node; | |
113 | } | |
114 | spin_unlock(&root->inode_lock); | |
115 | ||
2f7e33d4 MX |
116 | return NULL; |
117 | } | |
118 | ||
79787eaa | 119 | /* Will return either the node or PTR_ERR(-ENOMEM) */ |
2f7e33d4 MX |
120 | static struct btrfs_delayed_node *btrfs_get_or_create_delayed_node( |
121 | struct inode *inode) | |
122 | { | |
123 | struct btrfs_delayed_node *node; | |
124 | struct btrfs_inode *btrfs_inode = BTRFS_I(inode); | |
125 | struct btrfs_root *root = btrfs_inode->root; | |
126 | u64 ino = btrfs_ino(inode); | |
127 | int ret; | |
128 | ||
129 | again: | |
130 | node = btrfs_get_delayed_node(inode); | |
131 | if (node) | |
132 | return node; | |
133 | ||
16cdcec7 MX |
134 | node = kmem_cache_alloc(delayed_node_cache, GFP_NOFS); |
135 | if (!node) | |
136 | return ERR_PTR(-ENOMEM); | |
0d0ca30f | 137 | btrfs_init_delayed_node(node, root, ino); |
16cdcec7 MX |
138 | |
139 | atomic_inc(&node->refs); /* cached in the btrfs inode */ | |
140 | atomic_inc(&node->refs); /* can be accessed */ | |
141 | ||
142 | ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM); | |
143 | if (ret) { | |
144 | kmem_cache_free(delayed_node_cache, node); | |
145 | return ERR_PTR(ret); | |
146 | } | |
147 | ||
148 | spin_lock(&root->inode_lock); | |
0d0ca30f | 149 | ret = radix_tree_insert(&root->delayed_nodes_tree, ino, node); |
16cdcec7 MX |
150 | if (ret == -EEXIST) { |
151 | kmem_cache_free(delayed_node_cache, node); | |
152 | spin_unlock(&root->inode_lock); | |
153 | radix_tree_preload_end(); | |
154 | goto again; | |
155 | } | |
156 | btrfs_inode->delayed_node = node; | |
157 | spin_unlock(&root->inode_lock); | |
158 | radix_tree_preload_end(); | |
159 | ||
160 | return node; | |
161 | } | |
162 | ||
163 | /* | |
164 | * Call it when holding delayed_node->mutex | |
165 | * | |
166 | * If mod = 1, add this node into the prepared list. | |
167 | */ | |
168 | static void btrfs_queue_delayed_node(struct btrfs_delayed_root *root, | |
169 | struct btrfs_delayed_node *node, | |
170 | int mod) | |
171 | { | |
172 | spin_lock(&root->lock); | |
173 | if (node->in_list) { | |
174 | if (!list_empty(&node->p_list)) | |
175 | list_move_tail(&node->p_list, &root->prepare_list); | |
176 | else if (mod) | |
177 | list_add_tail(&node->p_list, &root->prepare_list); | |
178 | } else { | |
179 | list_add_tail(&node->n_list, &root->node_list); | |
180 | list_add_tail(&node->p_list, &root->prepare_list); | |
181 | atomic_inc(&node->refs); /* inserted into list */ | |
182 | root->nodes++; | |
183 | node->in_list = 1; | |
184 | } | |
185 | spin_unlock(&root->lock); | |
186 | } | |
187 | ||
188 | /* Call it when holding delayed_node->mutex */ | |
189 | static void btrfs_dequeue_delayed_node(struct btrfs_delayed_root *root, | |
190 | struct btrfs_delayed_node *node) | |
191 | { | |
192 | spin_lock(&root->lock); | |
193 | if (node->in_list) { | |
194 | root->nodes--; | |
195 | atomic_dec(&node->refs); /* not in the list */ | |
196 | list_del_init(&node->n_list); | |
197 | if (!list_empty(&node->p_list)) | |
198 | list_del_init(&node->p_list); | |
199 | node->in_list = 0; | |
200 | } | |
201 | spin_unlock(&root->lock); | |
202 | } | |
203 | ||
204 | struct btrfs_delayed_node *btrfs_first_delayed_node( | |
205 | struct btrfs_delayed_root *delayed_root) | |
206 | { | |
207 | struct list_head *p; | |
208 | struct btrfs_delayed_node *node = NULL; | |
209 | ||
210 | spin_lock(&delayed_root->lock); | |
211 | if (list_empty(&delayed_root->node_list)) | |
212 | goto out; | |
213 | ||
214 | p = delayed_root->node_list.next; | |
215 | node = list_entry(p, struct btrfs_delayed_node, n_list); | |
216 | atomic_inc(&node->refs); | |
217 | out: | |
218 | spin_unlock(&delayed_root->lock); | |
219 | ||
220 | return node; | |
221 | } | |
222 | ||
223 | struct btrfs_delayed_node *btrfs_next_delayed_node( | |
224 | struct btrfs_delayed_node *node) | |
225 | { | |
226 | struct btrfs_delayed_root *delayed_root; | |
227 | struct list_head *p; | |
228 | struct btrfs_delayed_node *next = NULL; | |
229 | ||
230 | delayed_root = node->root->fs_info->delayed_root; | |
231 | spin_lock(&delayed_root->lock); | |
232 | if (!node->in_list) { /* not in the list */ | |
233 | if (list_empty(&delayed_root->node_list)) | |
234 | goto out; | |
235 | p = delayed_root->node_list.next; | |
236 | } else if (list_is_last(&node->n_list, &delayed_root->node_list)) | |
237 | goto out; | |
238 | else | |
239 | p = node->n_list.next; | |
240 | ||
241 | next = list_entry(p, struct btrfs_delayed_node, n_list); | |
242 | atomic_inc(&next->refs); | |
243 | out: | |
244 | spin_unlock(&delayed_root->lock); | |
245 | ||
246 | return next; | |
247 | } | |
248 | ||
249 | static void __btrfs_release_delayed_node( | |
250 | struct btrfs_delayed_node *delayed_node, | |
251 | int mod) | |
252 | { | |
253 | struct btrfs_delayed_root *delayed_root; | |
254 | ||
255 | if (!delayed_node) | |
256 | return; | |
257 | ||
258 | delayed_root = delayed_node->root->fs_info->delayed_root; | |
259 | ||
260 | mutex_lock(&delayed_node->mutex); | |
261 | if (delayed_node->count) | |
262 | btrfs_queue_delayed_node(delayed_root, delayed_node, mod); | |
263 | else | |
264 | btrfs_dequeue_delayed_node(delayed_root, delayed_node); | |
265 | mutex_unlock(&delayed_node->mutex); | |
266 | ||
267 | if (atomic_dec_and_test(&delayed_node->refs)) { | |
268 | struct btrfs_root *root = delayed_node->root; | |
269 | spin_lock(&root->inode_lock); | |
270 | if (atomic_read(&delayed_node->refs) == 0) { | |
271 | radix_tree_delete(&root->delayed_nodes_tree, | |
272 | delayed_node->inode_id); | |
273 | kmem_cache_free(delayed_node_cache, delayed_node); | |
274 | } | |
275 | spin_unlock(&root->inode_lock); | |
276 | } | |
277 | } | |
278 | ||
279 | static inline void btrfs_release_delayed_node(struct btrfs_delayed_node *node) | |
280 | { | |
281 | __btrfs_release_delayed_node(node, 0); | |
282 | } | |
283 | ||
284 | struct btrfs_delayed_node *btrfs_first_prepared_delayed_node( | |
285 | struct btrfs_delayed_root *delayed_root) | |
286 | { | |
287 | struct list_head *p; | |
288 | struct btrfs_delayed_node *node = NULL; | |
289 | ||
290 | spin_lock(&delayed_root->lock); | |
291 | if (list_empty(&delayed_root->prepare_list)) | |
292 | goto out; | |
293 | ||
294 | p = delayed_root->prepare_list.next; | |
295 | list_del_init(p); | |
296 | node = list_entry(p, struct btrfs_delayed_node, p_list); | |
297 | atomic_inc(&node->refs); | |
298 | out: | |
299 | spin_unlock(&delayed_root->lock); | |
300 | ||
301 | return node; | |
302 | } | |
303 | ||
304 | static inline void btrfs_release_prepared_delayed_node( | |
305 | struct btrfs_delayed_node *node) | |
306 | { | |
307 | __btrfs_release_delayed_node(node, 1); | |
308 | } | |
309 | ||
310 | struct btrfs_delayed_item *btrfs_alloc_delayed_item(u32 data_len) | |
311 | { | |
312 | struct btrfs_delayed_item *item; | |
313 | item = kmalloc(sizeof(*item) + data_len, GFP_NOFS); | |
314 | if (item) { | |
315 | item->data_len = data_len; | |
316 | item->ins_or_del = 0; | |
317 | item->bytes_reserved = 0; | |
16cdcec7 MX |
318 | item->delayed_node = NULL; |
319 | atomic_set(&item->refs, 1); | |
320 | } | |
321 | return item; | |
322 | } | |
323 | ||
324 | /* | |
325 | * __btrfs_lookup_delayed_item - look up the delayed item by key | |
326 | * @delayed_node: pointer to the delayed node | |
327 | * @key: the key to look up | |
328 | * @prev: used to store the prev item if the right item isn't found | |
329 | * @next: used to store the next item if the right item isn't found | |
330 | * | |
331 | * Note: if we don't find the right item, we will return the prev item and | |
332 | * the next item. | |
333 | */ | |
334 | static struct btrfs_delayed_item *__btrfs_lookup_delayed_item( | |
335 | struct rb_root *root, | |
336 | struct btrfs_key *key, | |
337 | struct btrfs_delayed_item **prev, | |
338 | struct btrfs_delayed_item **next) | |
339 | { | |
340 | struct rb_node *node, *prev_node = NULL; | |
341 | struct btrfs_delayed_item *delayed_item = NULL; | |
342 | int ret = 0; | |
343 | ||
344 | node = root->rb_node; | |
345 | ||
346 | while (node) { | |
347 | delayed_item = rb_entry(node, struct btrfs_delayed_item, | |
348 | rb_node); | |
349 | prev_node = node; | |
350 | ret = btrfs_comp_cpu_keys(&delayed_item->key, key); | |
351 | if (ret < 0) | |
352 | node = node->rb_right; | |
353 | else if (ret > 0) | |
354 | node = node->rb_left; | |
355 | else | |
356 | return delayed_item; | |
357 | } | |
358 | ||
359 | if (prev) { | |
360 | if (!prev_node) | |
361 | *prev = NULL; | |
362 | else if (ret < 0) | |
363 | *prev = delayed_item; | |
364 | else if ((node = rb_prev(prev_node)) != NULL) { | |
365 | *prev = rb_entry(node, struct btrfs_delayed_item, | |
366 | rb_node); | |
367 | } else | |
368 | *prev = NULL; | |
369 | } | |
370 | ||
371 | if (next) { | |
372 | if (!prev_node) | |
373 | *next = NULL; | |
374 | else if (ret > 0) | |
375 | *next = delayed_item; | |
376 | else if ((node = rb_next(prev_node)) != NULL) { | |
377 | *next = rb_entry(node, struct btrfs_delayed_item, | |
378 | rb_node); | |
379 | } else | |
380 | *next = NULL; | |
381 | } | |
382 | return NULL; | |
383 | } | |
384 | ||
385 | struct btrfs_delayed_item *__btrfs_lookup_delayed_insertion_item( | |
386 | struct btrfs_delayed_node *delayed_node, | |
387 | struct btrfs_key *key) | |
388 | { | |
389 | struct btrfs_delayed_item *item; | |
390 | ||
391 | item = __btrfs_lookup_delayed_item(&delayed_node->ins_root, key, | |
392 | NULL, NULL); | |
393 | return item; | |
394 | } | |
395 | ||
396 | struct btrfs_delayed_item *__btrfs_lookup_delayed_deletion_item( | |
397 | struct btrfs_delayed_node *delayed_node, | |
398 | struct btrfs_key *key) | |
399 | { | |
400 | struct btrfs_delayed_item *item; | |
401 | ||
402 | item = __btrfs_lookup_delayed_item(&delayed_node->del_root, key, | |
403 | NULL, NULL); | |
404 | return item; | |
405 | } | |
406 | ||
407 | struct btrfs_delayed_item *__btrfs_search_delayed_insertion_item( | |
408 | struct btrfs_delayed_node *delayed_node, | |
409 | struct btrfs_key *key) | |
410 | { | |
411 | struct btrfs_delayed_item *item, *next; | |
412 | ||
413 | item = __btrfs_lookup_delayed_item(&delayed_node->ins_root, key, | |
414 | NULL, &next); | |
415 | if (!item) | |
416 | item = next; | |
417 | ||
418 | return item; | |
419 | } | |
420 | ||
421 | struct btrfs_delayed_item *__btrfs_search_delayed_deletion_item( | |
422 | struct btrfs_delayed_node *delayed_node, | |
423 | struct btrfs_key *key) | |
424 | { | |
425 | struct btrfs_delayed_item *item, *next; | |
426 | ||
427 | item = __btrfs_lookup_delayed_item(&delayed_node->del_root, key, | |
428 | NULL, &next); | |
429 | if (!item) | |
430 | item = next; | |
431 | ||
432 | return item; | |
433 | } | |
434 | ||
435 | static int __btrfs_add_delayed_item(struct btrfs_delayed_node *delayed_node, | |
436 | struct btrfs_delayed_item *ins, | |
437 | int action) | |
438 | { | |
439 | struct rb_node **p, *node; | |
440 | struct rb_node *parent_node = NULL; | |
441 | struct rb_root *root; | |
442 | struct btrfs_delayed_item *item; | |
443 | int cmp; | |
444 | ||
445 | if (action == BTRFS_DELAYED_INSERTION_ITEM) | |
446 | root = &delayed_node->ins_root; | |
447 | else if (action == BTRFS_DELAYED_DELETION_ITEM) | |
448 | root = &delayed_node->del_root; | |
449 | else | |
450 | BUG(); | |
451 | p = &root->rb_node; | |
452 | node = &ins->rb_node; | |
453 | ||
454 | while (*p) { | |
455 | parent_node = *p; | |
456 | item = rb_entry(parent_node, struct btrfs_delayed_item, | |
457 | rb_node); | |
458 | ||
459 | cmp = btrfs_comp_cpu_keys(&item->key, &ins->key); | |
460 | if (cmp < 0) | |
461 | p = &(*p)->rb_right; | |
462 | else if (cmp > 0) | |
463 | p = &(*p)->rb_left; | |
464 | else | |
465 | return -EEXIST; | |
466 | } | |
467 | ||
468 | rb_link_node(node, parent_node, p); | |
469 | rb_insert_color(node, root); | |
470 | ins->delayed_node = delayed_node; | |
471 | ins->ins_or_del = action; | |
472 | ||
473 | if (ins->key.type == BTRFS_DIR_INDEX_KEY && | |
474 | action == BTRFS_DELAYED_INSERTION_ITEM && | |
475 | ins->key.offset >= delayed_node->index_cnt) | |
476 | delayed_node->index_cnt = ins->key.offset + 1; | |
477 | ||
478 | delayed_node->count++; | |
479 | atomic_inc(&delayed_node->root->fs_info->delayed_root->items); | |
480 | return 0; | |
481 | } | |
482 | ||
483 | static int __btrfs_add_delayed_insertion_item(struct btrfs_delayed_node *node, | |
484 | struct btrfs_delayed_item *item) | |
485 | { | |
486 | return __btrfs_add_delayed_item(node, item, | |
487 | BTRFS_DELAYED_INSERTION_ITEM); | |
488 | } | |
489 | ||
490 | static int __btrfs_add_delayed_deletion_item(struct btrfs_delayed_node *node, | |
491 | struct btrfs_delayed_item *item) | |
492 | { | |
493 | return __btrfs_add_delayed_item(node, item, | |
494 | BTRFS_DELAYED_DELETION_ITEM); | |
495 | } | |
496 | ||
497 | static void __btrfs_remove_delayed_item(struct btrfs_delayed_item *delayed_item) | |
498 | { | |
499 | struct rb_root *root; | |
500 | struct btrfs_delayed_root *delayed_root; | |
501 | ||
502 | delayed_root = delayed_item->delayed_node->root->fs_info->delayed_root; | |
503 | ||
504 | BUG_ON(!delayed_root); | |
505 | BUG_ON(delayed_item->ins_or_del != BTRFS_DELAYED_DELETION_ITEM && | |
506 | delayed_item->ins_or_del != BTRFS_DELAYED_INSERTION_ITEM); | |
507 | ||
508 | if (delayed_item->ins_or_del == BTRFS_DELAYED_INSERTION_ITEM) | |
509 | root = &delayed_item->delayed_node->ins_root; | |
510 | else | |
511 | root = &delayed_item->delayed_node->del_root; | |
512 | ||
513 | rb_erase(&delayed_item->rb_node, root); | |
514 | delayed_item->delayed_node->count--; | |
66657b31 JB |
515 | if (atomic_dec_return(&delayed_root->items) < |
516 | BTRFS_DELAYED_BACKGROUND && | |
16cdcec7 MX |
517 | waitqueue_active(&delayed_root->wait)) |
518 | wake_up(&delayed_root->wait); | |
519 | } | |
520 | ||
521 | static void btrfs_release_delayed_item(struct btrfs_delayed_item *item) | |
522 | { | |
523 | if (item) { | |
524 | __btrfs_remove_delayed_item(item); | |
525 | if (atomic_dec_and_test(&item->refs)) | |
526 | kfree(item); | |
527 | } | |
528 | } | |
529 | ||
530 | struct btrfs_delayed_item *__btrfs_first_delayed_insertion_item( | |
531 | struct btrfs_delayed_node *delayed_node) | |
532 | { | |
533 | struct rb_node *p; | |
534 | struct btrfs_delayed_item *item = NULL; | |
535 | ||
536 | p = rb_first(&delayed_node->ins_root); | |
537 | if (p) | |
538 | item = rb_entry(p, struct btrfs_delayed_item, rb_node); | |
539 | ||
540 | return item; | |
541 | } | |
542 | ||
543 | struct btrfs_delayed_item *__btrfs_first_delayed_deletion_item( | |
544 | struct btrfs_delayed_node *delayed_node) | |
545 | { | |
546 | struct rb_node *p; | |
547 | struct btrfs_delayed_item *item = NULL; | |
548 | ||
549 | p = rb_first(&delayed_node->del_root); | |
550 | if (p) | |
551 | item = rb_entry(p, struct btrfs_delayed_item, rb_node); | |
552 | ||
553 | return item; | |
554 | } | |
555 | ||
556 | struct btrfs_delayed_item *__btrfs_next_delayed_item( | |
557 | struct btrfs_delayed_item *item) | |
558 | { | |
559 | struct rb_node *p; | |
560 | struct btrfs_delayed_item *next = NULL; | |
561 | ||
562 | p = rb_next(&item->rb_node); | |
563 | if (p) | |
564 | next = rb_entry(p, struct btrfs_delayed_item, rb_node); | |
565 | ||
566 | return next; | |
567 | } | |
568 | ||
16cdcec7 MX |
569 | static inline struct btrfs_root *btrfs_get_fs_root(struct btrfs_root *root, |
570 | u64 root_id) | |
571 | { | |
572 | struct btrfs_key root_key; | |
573 | ||
574 | if (root->objectid == root_id) | |
575 | return root; | |
576 | ||
577 | root_key.objectid = root_id; | |
578 | root_key.type = BTRFS_ROOT_ITEM_KEY; | |
579 | root_key.offset = (u64)-1; | |
580 | return btrfs_read_fs_root_no_name(root->fs_info, &root_key); | |
581 | } | |
582 | ||
583 | static int btrfs_delayed_item_reserve_metadata(struct btrfs_trans_handle *trans, | |
584 | struct btrfs_root *root, | |
585 | struct btrfs_delayed_item *item) | |
586 | { | |
587 | struct btrfs_block_rsv *src_rsv; | |
588 | struct btrfs_block_rsv *dst_rsv; | |
589 | u64 num_bytes; | |
590 | int ret; | |
591 | ||
592 | if (!trans->bytes_reserved) | |
593 | return 0; | |
594 | ||
595 | src_rsv = trans->block_rsv; | |
6d668dda | 596 | dst_rsv = &root->fs_info->delayed_block_rsv; |
16cdcec7 MX |
597 | |
598 | num_bytes = btrfs_calc_trans_metadata_size(root, 1); | |
599 | ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes); | |
8c2a3ca2 JB |
600 | if (!ret) { |
601 | trace_btrfs_space_reservation(root->fs_info, "delayed_item", | |
602 | item->key.objectid, | |
603 | num_bytes, 1); | |
16cdcec7 | 604 | item->bytes_reserved = num_bytes; |
8c2a3ca2 | 605 | } |
16cdcec7 MX |
606 | |
607 | return ret; | |
608 | } | |
609 | ||
610 | static void btrfs_delayed_item_release_metadata(struct btrfs_root *root, | |
611 | struct btrfs_delayed_item *item) | |
612 | { | |
19fd2949 MX |
613 | struct btrfs_block_rsv *rsv; |
614 | ||
16cdcec7 MX |
615 | if (!item->bytes_reserved) |
616 | return; | |
617 | ||
6d668dda | 618 | rsv = &root->fs_info->delayed_block_rsv; |
8c2a3ca2 JB |
619 | trace_btrfs_space_reservation(root->fs_info, "delayed_item", |
620 | item->key.objectid, item->bytes_reserved, | |
621 | 0); | |
19fd2949 | 622 | btrfs_block_rsv_release(root, rsv, |
16cdcec7 MX |
623 | item->bytes_reserved); |
624 | } | |
625 | ||
626 | static int btrfs_delayed_inode_reserve_metadata( | |
627 | struct btrfs_trans_handle *trans, | |
628 | struct btrfs_root *root, | |
7fd2ae21 | 629 | struct inode *inode, |
16cdcec7 MX |
630 | struct btrfs_delayed_node *node) |
631 | { | |
632 | struct btrfs_block_rsv *src_rsv; | |
633 | struct btrfs_block_rsv *dst_rsv; | |
634 | u64 num_bytes; | |
635 | int ret; | |
8c2a3ca2 | 636 | bool release = false; |
16cdcec7 | 637 | |
16cdcec7 | 638 | src_rsv = trans->block_rsv; |
6d668dda | 639 | dst_rsv = &root->fs_info->delayed_block_rsv; |
16cdcec7 MX |
640 | |
641 | num_bytes = btrfs_calc_trans_metadata_size(root, 1); | |
c06a0e12 JB |
642 | |
643 | /* | |
644 | * btrfs_dirty_inode will update the inode under btrfs_join_transaction | |
645 | * which doesn't reserve space for speed. This is a problem since we | |
646 | * still need to reserve space for this update, so try to reserve the | |
647 | * space. | |
648 | * | |
649 | * Now if src_rsv == delalloc_block_rsv we'll let it just steal since | |
650 | * we're accounted for. | |
651 | */ | |
e755d9ab | 652 | if (!src_rsv || (!trans->bytes_reserved && |
66d8f3dd | 653 | src_rsv->type != BTRFS_BLOCK_RSV_DELALLOC)) { |
08e007d2 MX |
654 | ret = btrfs_block_rsv_add(root, dst_rsv, num_bytes, |
655 | BTRFS_RESERVE_NO_FLUSH); | |
c06a0e12 JB |
656 | /* |
657 | * Since we're under a transaction reserve_metadata_bytes could | |
658 | * try to commit the transaction which will make it return | |
659 | * EAGAIN to make us stop the transaction we have, so return | |
660 | * ENOSPC instead so that btrfs_dirty_inode knows what to do. | |
661 | */ | |
662 | if (ret == -EAGAIN) | |
663 | ret = -ENOSPC; | |
8c2a3ca2 | 664 | if (!ret) { |
c06a0e12 | 665 | node->bytes_reserved = num_bytes; |
8c2a3ca2 JB |
666 | trace_btrfs_space_reservation(root->fs_info, |
667 | "delayed_inode", | |
668 | btrfs_ino(inode), | |
669 | num_bytes, 1); | |
670 | } | |
c06a0e12 | 671 | return ret; |
66d8f3dd | 672 | } else if (src_rsv->type == BTRFS_BLOCK_RSV_DELALLOC) { |
7fd2ae21 | 673 | spin_lock(&BTRFS_I(inode)->lock); |
72ac3c0d JB |
674 | if (test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
675 | &BTRFS_I(inode)->runtime_flags)) { | |
7fd2ae21 JB |
676 | spin_unlock(&BTRFS_I(inode)->lock); |
677 | release = true; | |
678 | goto migrate; | |
679 | } | |
680 | spin_unlock(&BTRFS_I(inode)->lock); | |
681 | ||
682 | /* Ok we didn't have space pre-reserved. This shouldn't happen | |
683 | * too often but it can happen if we do delalloc to an existing | |
684 | * inode which gets dirtied because of the time update, and then | |
685 | * isn't touched again until after the transaction commits and | |
686 | * then we try to write out the data. First try to be nice and | |
687 | * reserve something strictly for us. If not be a pain and try | |
688 | * to steal from the delalloc block rsv. | |
689 | */ | |
08e007d2 MX |
690 | ret = btrfs_block_rsv_add(root, dst_rsv, num_bytes, |
691 | BTRFS_RESERVE_NO_FLUSH); | |
7fd2ae21 JB |
692 | if (!ret) |
693 | goto out; | |
694 | ||
695 | ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes); | |
696 | if (!ret) | |
697 | goto out; | |
698 | ||
699 | /* | |
700 | * Ok this is a problem, let's just steal from the global rsv | |
701 | * since this really shouldn't happen that often. | |
702 | */ | |
703 | WARN_ON(1); | |
704 | ret = btrfs_block_rsv_migrate(&root->fs_info->global_block_rsv, | |
705 | dst_rsv, num_bytes); | |
706 | goto out; | |
c06a0e12 JB |
707 | } |
708 | ||
7fd2ae21 | 709 | migrate: |
16cdcec7 | 710 | ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes); |
7fd2ae21 JB |
711 | |
712 | out: | |
713 | /* | |
714 | * Migrate only takes a reservation, it doesn't touch the size of the | |
715 | * block_rsv. This is to simplify people who don't normally have things | |
716 | * migrated from their block rsv. If they go to release their | |
717 | * reservation, that will decrease the size as well, so if migrate | |
718 | * reduced size we'd end up with a negative size. But for the | |
719 | * delalloc_meta_reserved stuff we will only know to drop 1 reservation, | |
720 | * but we could in fact do this reserve/migrate dance several times | |
721 | * between the time we did the original reservation and we'd clean it | |
722 | * up. So to take care of this, release the space for the meta | |
723 | * reservation here. I think it may be time for a documentation page on | |
724 | * how block rsvs. work. | |
725 | */ | |
8c2a3ca2 JB |
726 | if (!ret) { |
727 | trace_btrfs_space_reservation(root->fs_info, "delayed_inode", | |
728 | btrfs_ino(inode), num_bytes, 1); | |
16cdcec7 | 729 | node->bytes_reserved = num_bytes; |
8c2a3ca2 | 730 | } |
16cdcec7 | 731 | |
8c2a3ca2 JB |
732 | if (release) { |
733 | trace_btrfs_space_reservation(root->fs_info, "delalloc", | |
734 | btrfs_ino(inode), num_bytes, 0); | |
7fd2ae21 | 735 | btrfs_block_rsv_release(root, src_rsv, num_bytes); |
8c2a3ca2 | 736 | } |
16cdcec7 MX |
737 | |
738 | return ret; | |
739 | } | |
740 | ||
741 | static void btrfs_delayed_inode_release_metadata(struct btrfs_root *root, | |
742 | struct btrfs_delayed_node *node) | |
743 | { | |
744 | struct btrfs_block_rsv *rsv; | |
745 | ||
746 | if (!node->bytes_reserved) | |
747 | return; | |
748 | ||
6d668dda | 749 | rsv = &root->fs_info->delayed_block_rsv; |
8c2a3ca2 JB |
750 | trace_btrfs_space_reservation(root->fs_info, "delayed_inode", |
751 | node->inode_id, node->bytes_reserved, 0); | |
16cdcec7 MX |
752 | btrfs_block_rsv_release(root, rsv, |
753 | node->bytes_reserved); | |
754 | node->bytes_reserved = 0; | |
755 | } | |
756 | ||
757 | /* | |
758 | * This helper will insert some continuous items into the same leaf according | |
759 | * to the free space of the leaf. | |
760 | */ | |
761 | static int btrfs_batch_insert_items(struct btrfs_trans_handle *trans, | |
762 | struct btrfs_root *root, | |
763 | struct btrfs_path *path, | |
764 | struct btrfs_delayed_item *item) | |
765 | { | |
766 | struct btrfs_delayed_item *curr, *next; | |
767 | int free_space; | |
768 | int total_data_size = 0, total_size = 0; | |
769 | struct extent_buffer *leaf; | |
770 | char *data_ptr; | |
771 | struct btrfs_key *keys; | |
772 | u32 *data_size; | |
773 | struct list_head head; | |
774 | int slot; | |
775 | int nitems; | |
776 | int i; | |
777 | int ret = 0; | |
778 | ||
779 | BUG_ON(!path->nodes[0]); | |
780 | ||
781 | leaf = path->nodes[0]; | |
782 | free_space = btrfs_leaf_free_space(root, leaf); | |
783 | INIT_LIST_HEAD(&head); | |
784 | ||
785 | next = item; | |
17aca1c9 | 786 | nitems = 0; |
16cdcec7 MX |
787 | |
788 | /* | |
789 | * count the number of the continuous items that we can insert in batch | |
790 | */ | |
791 | while (total_size + next->data_len + sizeof(struct btrfs_item) <= | |
792 | free_space) { | |
793 | total_data_size += next->data_len; | |
794 | total_size += next->data_len + sizeof(struct btrfs_item); | |
795 | list_add_tail(&next->tree_list, &head); | |
796 | nitems++; | |
797 | ||
798 | curr = next; | |
799 | next = __btrfs_next_delayed_item(curr); | |
800 | if (!next) | |
801 | break; | |
802 | ||
803 | if (!btrfs_is_continuous_delayed_item(curr, next)) | |
804 | break; | |
805 | } | |
806 | ||
807 | if (!nitems) { | |
808 | ret = 0; | |
809 | goto out; | |
810 | } | |
811 | ||
812 | /* | |
813 | * we need allocate some memory space, but it might cause the task | |
814 | * to sleep, so we set all locked nodes in the path to blocking locks | |
815 | * first. | |
816 | */ | |
817 | btrfs_set_path_blocking(path); | |
818 | ||
819 | keys = kmalloc(sizeof(struct btrfs_key) * nitems, GFP_NOFS); | |
820 | if (!keys) { | |
821 | ret = -ENOMEM; | |
822 | goto out; | |
823 | } | |
824 | ||
825 | data_size = kmalloc(sizeof(u32) * nitems, GFP_NOFS); | |
826 | if (!data_size) { | |
827 | ret = -ENOMEM; | |
828 | goto error; | |
829 | } | |
830 | ||
831 | /* get keys of all the delayed items */ | |
832 | i = 0; | |
833 | list_for_each_entry(next, &head, tree_list) { | |
834 | keys[i] = next->key; | |
835 | data_size[i] = next->data_len; | |
836 | i++; | |
837 | } | |
838 | ||
839 | /* reset all the locked nodes in the patch to spinning locks. */ | |
bd681513 | 840 | btrfs_clear_path_blocking(path, NULL, 0); |
16cdcec7 MX |
841 | |
842 | /* insert the keys of the items */ | |
143bede5 JM |
843 | setup_items_for_insert(trans, root, path, keys, data_size, |
844 | total_data_size, total_size, nitems); | |
16cdcec7 MX |
845 | |
846 | /* insert the dir index items */ | |
847 | slot = path->slots[0]; | |
848 | list_for_each_entry_safe(curr, next, &head, tree_list) { | |
849 | data_ptr = btrfs_item_ptr(leaf, slot, char); | |
850 | write_extent_buffer(leaf, &curr->data, | |
851 | (unsigned long)data_ptr, | |
852 | curr->data_len); | |
853 | slot++; | |
854 | ||
855 | btrfs_delayed_item_release_metadata(root, curr); | |
856 | ||
857 | list_del(&curr->tree_list); | |
858 | btrfs_release_delayed_item(curr); | |
859 | } | |
860 | ||
861 | error: | |
862 | kfree(data_size); | |
863 | kfree(keys); | |
864 | out: | |
865 | return ret; | |
866 | } | |
867 | ||
868 | /* | |
869 | * This helper can just do simple insertion that needn't extend item for new | |
870 | * data, such as directory name index insertion, inode insertion. | |
871 | */ | |
872 | static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans, | |
873 | struct btrfs_root *root, | |
874 | struct btrfs_path *path, | |
875 | struct btrfs_delayed_item *delayed_item) | |
876 | { | |
877 | struct extent_buffer *leaf; | |
16cdcec7 MX |
878 | char *ptr; |
879 | int ret; | |
880 | ||
881 | ret = btrfs_insert_empty_item(trans, root, path, &delayed_item->key, | |
882 | delayed_item->data_len); | |
883 | if (ret < 0 && ret != -EEXIST) | |
884 | return ret; | |
885 | ||
886 | leaf = path->nodes[0]; | |
887 | ||
16cdcec7 MX |
888 | ptr = btrfs_item_ptr(leaf, path->slots[0], char); |
889 | ||
890 | write_extent_buffer(leaf, delayed_item->data, (unsigned long)ptr, | |
891 | delayed_item->data_len); | |
892 | btrfs_mark_buffer_dirty(leaf); | |
893 | ||
894 | btrfs_delayed_item_release_metadata(root, delayed_item); | |
895 | return 0; | |
896 | } | |
897 | ||
898 | /* | |
899 | * we insert an item first, then if there are some continuous items, we try | |
900 | * to insert those items into the same leaf. | |
901 | */ | |
902 | static int btrfs_insert_delayed_items(struct btrfs_trans_handle *trans, | |
903 | struct btrfs_path *path, | |
904 | struct btrfs_root *root, | |
905 | struct btrfs_delayed_node *node) | |
906 | { | |
907 | struct btrfs_delayed_item *curr, *prev; | |
908 | int ret = 0; | |
909 | ||
910 | do_again: | |
911 | mutex_lock(&node->mutex); | |
912 | curr = __btrfs_first_delayed_insertion_item(node); | |
913 | if (!curr) | |
914 | goto insert_end; | |
915 | ||
916 | ret = btrfs_insert_delayed_item(trans, root, path, curr); | |
917 | if (ret < 0) { | |
945d8962 | 918 | btrfs_release_path(path); |
16cdcec7 MX |
919 | goto insert_end; |
920 | } | |
921 | ||
922 | prev = curr; | |
923 | curr = __btrfs_next_delayed_item(prev); | |
924 | if (curr && btrfs_is_continuous_delayed_item(prev, curr)) { | |
925 | /* insert the continuous items into the same leaf */ | |
926 | path->slots[0]++; | |
927 | btrfs_batch_insert_items(trans, root, path, curr); | |
928 | } | |
929 | btrfs_release_delayed_item(prev); | |
930 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
931 | ||
945d8962 | 932 | btrfs_release_path(path); |
16cdcec7 MX |
933 | mutex_unlock(&node->mutex); |
934 | goto do_again; | |
935 | ||
936 | insert_end: | |
937 | mutex_unlock(&node->mutex); | |
938 | return ret; | |
939 | } | |
940 | ||
941 | static int btrfs_batch_delete_items(struct btrfs_trans_handle *trans, | |
942 | struct btrfs_root *root, | |
943 | struct btrfs_path *path, | |
944 | struct btrfs_delayed_item *item) | |
945 | { | |
946 | struct btrfs_delayed_item *curr, *next; | |
947 | struct extent_buffer *leaf; | |
948 | struct btrfs_key key; | |
949 | struct list_head head; | |
950 | int nitems, i, last_item; | |
951 | int ret = 0; | |
952 | ||
953 | BUG_ON(!path->nodes[0]); | |
954 | ||
955 | leaf = path->nodes[0]; | |
956 | ||
957 | i = path->slots[0]; | |
958 | last_item = btrfs_header_nritems(leaf) - 1; | |
959 | if (i > last_item) | |
960 | return -ENOENT; /* FIXME: Is errno suitable? */ | |
961 | ||
962 | next = item; | |
963 | INIT_LIST_HEAD(&head); | |
964 | btrfs_item_key_to_cpu(leaf, &key, i); | |
965 | nitems = 0; | |
966 | /* | |
967 | * count the number of the dir index items that we can delete in batch | |
968 | */ | |
969 | while (btrfs_comp_cpu_keys(&next->key, &key) == 0) { | |
970 | list_add_tail(&next->tree_list, &head); | |
971 | nitems++; | |
972 | ||
973 | curr = next; | |
974 | next = __btrfs_next_delayed_item(curr); | |
975 | if (!next) | |
976 | break; | |
977 | ||
978 | if (!btrfs_is_continuous_delayed_item(curr, next)) | |
979 | break; | |
980 | ||
981 | i++; | |
982 | if (i > last_item) | |
983 | break; | |
984 | btrfs_item_key_to_cpu(leaf, &key, i); | |
985 | } | |
986 | ||
987 | if (!nitems) | |
988 | return 0; | |
989 | ||
990 | ret = btrfs_del_items(trans, root, path, path->slots[0], nitems); | |
991 | if (ret) | |
992 | goto out; | |
993 | ||
994 | list_for_each_entry_safe(curr, next, &head, tree_list) { | |
995 | btrfs_delayed_item_release_metadata(root, curr); | |
996 | list_del(&curr->tree_list); | |
997 | btrfs_release_delayed_item(curr); | |
998 | } | |
999 | ||
1000 | out: | |
1001 | return ret; | |
1002 | } | |
1003 | ||
1004 | static int btrfs_delete_delayed_items(struct btrfs_trans_handle *trans, | |
1005 | struct btrfs_path *path, | |
1006 | struct btrfs_root *root, | |
1007 | struct btrfs_delayed_node *node) | |
1008 | { | |
1009 | struct btrfs_delayed_item *curr, *prev; | |
1010 | int ret = 0; | |
1011 | ||
1012 | do_again: | |
1013 | mutex_lock(&node->mutex); | |
1014 | curr = __btrfs_first_delayed_deletion_item(node); | |
1015 | if (!curr) | |
1016 | goto delete_fail; | |
1017 | ||
1018 | ret = btrfs_search_slot(trans, root, &curr->key, path, -1, 1); | |
1019 | if (ret < 0) | |
1020 | goto delete_fail; | |
1021 | else if (ret > 0) { | |
1022 | /* | |
1023 | * can't find the item which the node points to, so this node | |
1024 | * is invalid, just drop it. | |
1025 | */ | |
1026 | prev = curr; | |
1027 | curr = __btrfs_next_delayed_item(prev); | |
1028 | btrfs_release_delayed_item(prev); | |
1029 | ret = 0; | |
945d8962 | 1030 | btrfs_release_path(path); |
62095265 FW |
1031 | if (curr) { |
1032 | mutex_unlock(&node->mutex); | |
16cdcec7 | 1033 | goto do_again; |
62095265 | 1034 | } else |
16cdcec7 MX |
1035 | goto delete_fail; |
1036 | } | |
1037 | ||
1038 | btrfs_batch_delete_items(trans, root, path, curr); | |
945d8962 | 1039 | btrfs_release_path(path); |
16cdcec7 MX |
1040 | mutex_unlock(&node->mutex); |
1041 | goto do_again; | |
1042 | ||
1043 | delete_fail: | |
945d8962 | 1044 | btrfs_release_path(path); |
16cdcec7 MX |
1045 | mutex_unlock(&node->mutex); |
1046 | return ret; | |
1047 | } | |
1048 | ||
1049 | static void btrfs_release_delayed_inode(struct btrfs_delayed_node *delayed_node) | |
1050 | { | |
1051 | struct btrfs_delayed_root *delayed_root; | |
1052 | ||
1053 | if (delayed_node && delayed_node->inode_dirty) { | |
1054 | BUG_ON(!delayed_node->root); | |
1055 | delayed_node->inode_dirty = 0; | |
1056 | delayed_node->count--; | |
1057 | ||
1058 | delayed_root = delayed_node->root->fs_info->delayed_root; | |
66657b31 | 1059 | if (atomic_dec_return(&delayed_root->items) < |
16cdcec7 MX |
1060 | BTRFS_DELAYED_BACKGROUND && |
1061 | waitqueue_active(&delayed_root->wait)) | |
1062 | wake_up(&delayed_root->wait); | |
1063 | } | |
1064 | } | |
1065 | ||
0e8c36a9 MX |
1066 | static int __btrfs_update_delayed_inode(struct btrfs_trans_handle *trans, |
1067 | struct btrfs_root *root, | |
1068 | struct btrfs_path *path, | |
1069 | struct btrfs_delayed_node *node) | |
16cdcec7 MX |
1070 | { |
1071 | struct btrfs_key key; | |
1072 | struct btrfs_inode_item *inode_item; | |
1073 | struct extent_buffer *leaf; | |
1074 | int ret; | |
1075 | ||
16cdcec7 MX |
1076 | key.objectid = node->inode_id; |
1077 | btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY); | |
1078 | key.offset = 0; | |
0e8c36a9 | 1079 | |
16cdcec7 MX |
1080 | ret = btrfs_lookup_inode(trans, root, path, &key, 1); |
1081 | if (ret > 0) { | |
945d8962 | 1082 | btrfs_release_path(path); |
16cdcec7 MX |
1083 | return -ENOENT; |
1084 | } else if (ret < 0) { | |
16cdcec7 MX |
1085 | return ret; |
1086 | } | |
1087 | ||
1088 | btrfs_unlock_up_safe(path, 1); | |
1089 | leaf = path->nodes[0]; | |
1090 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
1091 | struct btrfs_inode_item); | |
1092 | write_extent_buffer(leaf, &node->inode_item, (unsigned long)inode_item, | |
1093 | sizeof(struct btrfs_inode_item)); | |
1094 | btrfs_mark_buffer_dirty(leaf); | |
945d8962 | 1095 | btrfs_release_path(path); |
16cdcec7 MX |
1096 | |
1097 | btrfs_delayed_inode_release_metadata(root, node); | |
1098 | btrfs_release_delayed_inode(node); | |
16cdcec7 MX |
1099 | |
1100 | return 0; | |
1101 | } | |
1102 | ||
0e8c36a9 MX |
1103 | static inline int btrfs_update_delayed_inode(struct btrfs_trans_handle *trans, |
1104 | struct btrfs_root *root, | |
1105 | struct btrfs_path *path, | |
1106 | struct btrfs_delayed_node *node) | |
1107 | { | |
1108 | int ret; | |
1109 | ||
1110 | mutex_lock(&node->mutex); | |
1111 | if (!node->inode_dirty) { | |
1112 | mutex_unlock(&node->mutex); | |
1113 | return 0; | |
1114 | } | |
1115 | ||
1116 | ret = __btrfs_update_delayed_inode(trans, root, path, node); | |
1117 | mutex_unlock(&node->mutex); | |
1118 | return ret; | |
1119 | } | |
1120 | ||
4ea41ce0 MX |
1121 | static inline int |
1122 | __btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans, | |
1123 | struct btrfs_path *path, | |
1124 | struct btrfs_delayed_node *node) | |
1125 | { | |
1126 | int ret; | |
1127 | ||
1128 | ret = btrfs_insert_delayed_items(trans, path, node->root, node); | |
1129 | if (ret) | |
1130 | return ret; | |
1131 | ||
1132 | ret = btrfs_delete_delayed_items(trans, path, node->root, node); | |
1133 | if (ret) | |
1134 | return ret; | |
1135 | ||
1136 | ret = btrfs_update_delayed_inode(trans, node->root, path, node); | |
1137 | return ret; | |
1138 | } | |
1139 | ||
79787eaa JM |
1140 | /* |
1141 | * Called when committing the transaction. | |
1142 | * Returns 0 on success. | |
1143 | * Returns < 0 on error and returns with an aborted transaction with any | |
1144 | * outstanding delayed items cleaned up. | |
1145 | */ | |
96c3f433 JB |
1146 | static int __btrfs_run_delayed_items(struct btrfs_trans_handle *trans, |
1147 | struct btrfs_root *root, int nr) | |
16cdcec7 MX |
1148 | { |
1149 | struct btrfs_delayed_root *delayed_root; | |
1150 | struct btrfs_delayed_node *curr_node, *prev_node; | |
1151 | struct btrfs_path *path; | |
19fd2949 | 1152 | struct btrfs_block_rsv *block_rsv; |
16cdcec7 | 1153 | int ret = 0; |
96c3f433 | 1154 | bool count = (nr > 0); |
16cdcec7 | 1155 | |
79787eaa JM |
1156 | if (trans->aborted) |
1157 | return -EIO; | |
1158 | ||
16cdcec7 MX |
1159 | path = btrfs_alloc_path(); |
1160 | if (!path) | |
1161 | return -ENOMEM; | |
1162 | path->leave_spinning = 1; | |
1163 | ||
19fd2949 | 1164 | block_rsv = trans->block_rsv; |
6d668dda | 1165 | trans->block_rsv = &root->fs_info->delayed_block_rsv; |
19fd2949 | 1166 | |
16cdcec7 MX |
1167 | delayed_root = btrfs_get_delayed_root(root); |
1168 | ||
1169 | curr_node = btrfs_first_delayed_node(delayed_root); | |
96c3f433 | 1170 | while (curr_node && (!count || (count && nr--))) { |
4ea41ce0 MX |
1171 | ret = __btrfs_commit_inode_delayed_items(trans, path, |
1172 | curr_node); | |
16cdcec7 MX |
1173 | if (ret) { |
1174 | btrfs_release_delayed_node(curr_node); | |
96c3f433 | 1175 | curr_node = NULL; |
79787eaa | 1176 | btrfs_abort_transaction(trans, root, ret); |
16cdcec7 MX |
1177 | break; |
1178 | } | |
1179 | ||
1180 | prev_node = curr_node; | |
1181 | curr_node = btrfs_next_delayed_node(curr_node); | |
1182 | btrfs_release_delayed_node(prev_node); | |
1183 | } | |
1184 | ||
96c3f433 JB |
1185 | if (curr_node) |
1186 | btrfs_release_delayed_node(curr_node); | |
16cdcec7 | 1187 | btrfs_free_path(path); |
19fd2949 | 1188 | trans->block_rsv = block_rsv; |
79787eaa | 1189 | |
16cdcec7 MX |
1190 | return ret; |
1191 | } | |
1192 | ||
96c3f433 JB |
1193 | int btrfs_run_delayed_items(struct btrfs_trans_handle *trans, |
1194 | struct btrfs_root *root) | |
1195 | { | |
1196 | return __btrfs_run_delayed_items(trans, root, -1); | |
1197 | } | |
1198 | ||
1199 | int btrfs_run_delayed_items_nr(struct btrfs_trans_handle *trans, | |
1200 | struct btrfs_root *root, int nr) | |
1201 | { | |
1202 | return __btrfs_run_delayed_items(trans, root, nr); | |
1203 | } | |
1204 | ||
16cdcec7 MX |
1205 | int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans, |
1206 | struct inode *inode) | |
1207 | { | |
1208 | struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode); | |
4ea41ce0 MX |
1209 | struct btrfs_path *path; |
1210 | struct btrfs_block_rsv *block_rsv; | |
16cdcec7 MX |
1211 | int ret; |
1212 | ||
1213 | if (!delayed_node) | |
1214 | return 0; | |
1215 | ||
1216 | mutex_lock(&delayed_node->mutex); | |
1217 | if (!delayed_node->count) { | |
1218 | mutex_unlock(&delayed_node->mutex); | |
1219 | btrfs_release_delayed_node(delayed_node); | |
1220 | return 0; | |
1221 | } | |
1222 | mutex_unlock(&delayed_node->mutex); | |
1223 | ||
4ea41ce0 MX |
1224 | path = btrfs_alloc_path(); |
1225 | if (!path) | |
1226 | return -ENOMEM; | |
1227 | path->leave_spinning = 1; | |
1228 | ||
1229 | block_rsv = trans->block_rsv; | |
1230 | trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv; | |
1231 | ||
1232 | ret = __btrfs_commit_inode_delayed_items(trans, path, delayed_node); | |
1233 | ||
16cdcec7 | 1234 | btrfs_release_delayed_node(delayed_node); |
4ea41ce0 MX |
1235 | btrfs_free_path(path); |
1236 | trans->block_rsv = block_rsv; | |
1237 | ||
16cdcec7 MX |
1238 | return ret; |
1239 | } | |
1240 | ||
0e8c36a9 MX |
1241 | int btrfs_commit_inode_delayed_inode(struct inode *inode) |
1242 | { | |
1243 | struct btrfs_trans_handle *trans; | |
1244 | struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode); | |
1245 | struct btrfs_path *path; | |
1246 | struct btrfs_block_rsv *block_rsv; | |
1247 | int ret; | |
1248 | ||
1249 | if (!delayed_node) | |
1250 | return 0; | |
1251 | ||
1252 | mutex_lock(&delayed_node->mutex); | |
1253 | if (!delayed_node->inode_dirty) { | |
1254 | mutex_unlock(&delayed_node->mutex); | |
1255 | btrfs_release_delayed_node(delayed_node); | |
1256 | return 0; | |
1257 | } | |
1258 | mutex_unlock(&delayed_node->mutex); | |
1259 | ||
1260 | trans = btrfs_join_transaction(delayed_node->root); | |
1261 | if (IS_ERR(trans)) { | |
1262 | ret = PTR_ERR(trans); | |
1263 | goto out; | |
1264 | } | |
1265 | ||
1266 | path = btrfs_alloc_path(); | |
1267 | if (!path) { | |
1268 | ret = -ENOMEM; | |
1269 | goto trans_out; | |
1270 | } | |
1271 | path->leave_spinning = 1; | |
1272 | ||
1273 | block_rsv = trans->block_rsv; | |
1274 | trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv; | |
1275 | ||
1276 | mutex_lock(&delayed_node->mutex); | |
1277 | if (delayed_node->inode_dirty) | |
1278 | ret = __btrfs_update_delayed_inode(trans, delayed_node->root, | |
1279 | path, delayed_node); | |
1280 | else | |
1281 | ret = 0; | |
1282 | mutex_unlock(&delayed_node->mutex); | |
1283 | ||
1284 | btrfs_free_path(path); | |
1285 | trans->block_rsv = block_rsv; | |
1286 | trans_out: | |
1287 | btrfs_end_transaction(trans, delayed_node->root); | |
1288 | btrfs_btree_balance_dirty(delayed_node->root); | |
1289 | out: | |
1290 | btrfs_release_delayed_node(delayed_node); | |
1291 | ||
1292 | return ret; | |
1293 | } | |
1294 | ||
16cdcec7 MX |
1295 | void btrfs_remove_delayed_node(struct inode *inode) |
1296 | { | |
1297 | struct btrfs_delayed_node *delayed_node; | |
1298 | ||
1299 | delayed_node = ACCESS_ONCE(BTRFS_I(inode)->delayed_node); | |
1300 | if (!delayed_node) | |
1301 | return; | |
1302 | ||
1303 | BTRFS_I(inode)->delayed_node = NULL; | |
1304 | btrfs_release_delayed_node(delayed_node); | |
1305 | } | |
1306 | ||
1307 | struct btrfs_async_delayed_node { | |
1308 | struct btrfs_root *root; | |
1309 | struct btrfs_delayed_node *delayed_node; | |
1310 | struct btrfs_work work; | |
1311 | }; | |
1312 | ||
1313 | static void btrfs_async_run_delayed_node_done(struct btrfs_work *work) | |
1314 | { | |
1315 | struct btrfs_async_delayed_node *async_node; | |
1316 | struct btrfs_trans_handle *trans; | |
1317 | struct btrfs_path *path; | |
1318 | struct btrfs_delayed_node *delayed_node = NULL; | |
1319 | struct btrfs_root *root; | |
19fd2949 | 1320 | struct btrfs_block_rsv *block_rsv; |
16cdcec7 | 1321 | int need_requeue = 0; |
16cdcec7 MX |
1322 | |
1323 | async_node = container_of(work, struct btrfs_async_delayed_node, work); | |
1324 | ||
1325 | path = btrfs_alloc_path(); | |
1326 | if (!path) | |
1327 | goto out; | |
1328 | path->leave_spinning = 1; | |
1329 | ||
1330 | delayed_node = async_node->delayed_node; | |
1331 | root = delayed_node->root; | |
1332 | ||
ff5714cc | 1333 | trans = btrfs_join_transaction(root); |
16cdcec7 MX |
1334 | if (IS_ERR(trans)) |
1335 | goto free_path; | |
1336 | ||
19fd2949 | 1337 | block_rsv = trans->block_rsv; |
6d668dda | 1338 | trans->block_rsv = &root->fs_info->delayed_block_rsv; |
19fd2949 | 1339 | |
4ea41ce0 | 1340 | __btrfs_commit_inode_delayed_items(trans, path, delayed_node); |
16cdcec7 MX |
1341 | /* |
1342 | * Maybe new delayed items have been inserted, so we need requeue | |
1343 | * the work. Besides that, we must dequeue the empty delayed nodes | |
1344 | * to avoid the race between delayed items balance and the worker. | |
1345 | * The race like this: | |
1346 | * Task1 Worker thread | |
1347 | * count == 0, needn't requeue | |
1348 | * also needn't insert the | |
1349 | * delayed node into prepare | |
1350 | * list again. | |
1351 | * add lots of delayed items | |
1352 | * queue the delayed node | |
1353 | * already in the list, | |
1354 | * and not in the prepare | |
1355 | * list, it means the delayed | |
1356 | * node is being dealt with | |
1357 | * by the worker. | |
1358 | * do delayed items balance | |
1359 | * the delayed node is being | |
1360 | * dealt with by the worker | |
1361 | * now, just wait. | |
1362 | * the worker goto idle. | |
1363 | * Task1 will sleep until the transaction is commited. | |
1364 | */ | |
1365 | mutex_lock(&delayed_node->mutex); | |
1366 | if (delayed_node->count) | |
1367 | need_requeue = 1; | |
1368 | else | |
1369 | btrfs_dequeue_delayed_node(root->fs_info->delayed_root, | |
1370 | delayed_node); | |
1371 | mutex_unlock(&delayed_node->mutex); | |
1372 | ||
19fd2949 | 1373 | trans->block_rsv = block_rsv; |
16cdcec7 | 1374 | btrfs_end_transaction_dmeta(trans, root); |
b53d3f5d | 1375 | btrfs_btree_balance_dirty_nodelay(root); |
16cdcec7 MX |
1376 | free_path: |
1377 | btrfs_free_path(path); | |
1378 | out: | |
1379 | if (need_requeue) | |
1380 | btrfs_requeue_work(&async_node->work); | |
1381 | else { | |
1382 | btrfs_release_prepared_delayed_node(delayed_node); | |
1383 | kfree(async_node); | |
1384 | } | |
1385 | } | |
1386 | ||
1387 | static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root, | |
1388 | struct btrfs_root *root, int all) | |
1389 | { | |
1390 | struct btrfs_async_delayed_node *async_node; | |
1391 | struct btrfs_delayed_node *curr; | |
1392 | int count = 0; | |
1393 | ||
1394 | again: | |
1395 | curr = btrfs_first_prepared_delayed_node(delayed_root); | |
1396 | if (!curr) | |
1397 | return 0; | |
1398 | ||
1399 | async_node = kmalloc(sizeof(*async_node), GFP_NOFS); | |
1400 | if (!async_node) { | |
1401 | btrfs_release_prepared_delayed_node(curr); | |
1402 | return -ENOMEM; | |
1403 | } | |
1404 | ||
1405 | async_node->root = root; | |
1406 | async_node->delayed_node = curr; | |
1407 | ||
1408 | async_node->work.func = btrfs_async_run_delayed_node_done; | |
1409 | async_node->work.flags = 0; | |
1410 | ||
1411 | btrfs_queue_worker(&root->fs_info->delayed_workers, &async_node->work); | |
1412 | count++; | |
1413 | ||
1414 | if (all || count < 4) | |
1415 | goto again; | |
1416 | ||
1417 | return 0; | |
1418 | } | |
1419 | ||
e999376f CM |
1420 | void btrfs_assert_delayed_root_empty(struct btrfs_root *root) |
1421 | { | |
1422 | struct btrfs_delayed_root *delayed_root; | |
1423 | delayed_root = btrfs_get_delayed_root(root); | |
1424 | WARN_ON(btrfs_first_delayed_node(delayed_root)); | |
1425 | } | |
1426 | ||
16cdcec7 MX |
1427 | void btrfs_balance_delayed_items(struct btrfs_root *root) |
1428 | { | |
1429 | struct btrfs_delayed_root *delayed_root; | |
1430 | ||
1431 | delayed_root = btrfs_get_delayed_root(root); | |
1432 | ||
1433 | if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND) | |
1434 | return; | |
1435 | ||
1436 | if (atomic_read(&delayed_root->items) >= BTRFS_DELAYED_WRITEBACK) { | |
1437 | int ret; | |
1438 | ret = btrfs_wq_run_delayed_node(delayed_root, root, 1); | |
1439 | if (ret) | |
1440 | return; | |
1441 | ||
1442 | wait_event_interruptible_timeout( | |
1443 | delayed_root->wait, | |
1444 | (atomic_read(&delayed_root->items) < | |
1445 | BTRFS_DELAYED_BACKGROUND), | |
1446 | HZ); | |
1447 | return; | |
1448 | } | |
1449 | ||
1450 | btrfs_wq_run_delayed_node(delayed_root, root, 0); | |
1451 | } | |
1452 | ||
79787eaa | 1453 | /* Will return 0 or -ENOMEM */ |
16cdcec7 MX |
1454 | int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans, |
1455 | struct btrfs_root *root, const char *name, | |
1456 | int name_len, struct inode *dir, | |
1457 | struct btrfs_disk_key *disk_key, u8 type, | |
1458 | u64 index) | |
1459 | { | |
1460 | struct btrfs_delayed_node *delayed_node; | |
1461 | struct btrfs_delayed_item *delayed_item; | |
1462 | struct btrfs_dir_item *dir_item; | |
1463 | int ret; | |
1464 | ||
1465 | delayed_node = btrfs_get_or_create_delayed_node(dir); | |
1466 | if (IS_ERR(delayed_node)) | |
1467 | return PTR_ERR(delayed_node); | |
1468 | ||
1469 | delayed_item = btrfs_alloc_delayed_item(sizeof(*dir_item) + name_len); | |
1470 | if (!delayed_item) { | |
1471 | ret = -ENOMEM; | |
1472 | goto release_node; | |
1473 | } | |
1474 | ||
0d0ca30f | 1475 | delayed_item->key.objectid = btrfs_ino(dir); |
16cdcec7 MX |
1476 | btrfs_set_key_type(&delayed_item->key, BTRFS_DIR_INDEX_KEY); |
1477 | delayed_item->key.offset = index; | |
1478 | ||
1479 | dir_item = (struct btrfs_dir_item *)delayed_item->data; | |
1480 | dir_item->location = *disk_key; | |
1481 | dir_item->transid = cpu_to_le64(trans->transid); | |
1482 | dir_item->data_len = 0; | |
1483 | dir_item->name_len = cpu_to_le16(name_len); | |
1484 | dir_item->type = type; | |
1485 | memcpy((char *)(dir_item + 1), name, name_len); | |
1486 | ||
8c2a3ca2 JB |
1487 | ret = btrfs_delayed_item_reserve_metadata(trans, root, delayed_item); |
1488 | /* | |
1489 | * we have reserved enough space when we start a new transaction, | |
1490 | * so reserving metadata failure is impossible | |
1491 | */ | |
1492 | BUG_ON(ret); | |
1493 | ||
1494 | ||
16cdcec7 MX |
1495 | mutex_lock(&delayed_node->mutex); |
1496 | ret = __btrfs_add_delayed_insertion_item(delayed_node, delayed_item); | |
1497 | if (unlikely(ret)) { | |
1498 | printk(KERN_ERR "err add delayed dir index item(name: %s) into " | |
1499 | "the insertion tree of the delayed node" | |
1500 | "(root id: %llu, inode id: %llu, errno: %d)\n", | |
1501 | name, | |
1502 | (unsigned long long)delayed_node->root->objectid, | |
1503 | (unsigned long long)delayed_node->inode_id, | |
1504 | ret); | |
1505 | BUG(); | |
1506 | } | |
1507 | mutex_unlock(&delayed_node->mutex); | |
1508 | ||
1509 | release_node: | |
1510 | btrfs_release_delayed_node(delayed_node); | |
1511 | return ret; | |
1512 | } | |
1513 | ||
1514 | static int btrfs_delete_delayed_insertion_item(struct btrfs_root *root, | |
1515 | struct btrfs_delayed_node *node, | |
1516 | struct btrfs_key *key) | |
1517 | { | |
1518 | struct btrfs_delayed_item *item; | |
1519 | ||
1520 | mutex_lock(&node->mutex); | |
1521 | item = __btrfs_lookup_delayed_insertion_item(node, key); | |
1522 | if (!item) { | |
1523 | mutex_unlock(&node->mutex); | |
1524 | return 1; | |
1525 | } | |
1526 | ||
1527 | btrfs_delayed_item_release_metadata(root, item); | |
1528 | btrfs_release_delayed_item(item); | |
1529 | mutex_unlock(&node->mutex); | |
1530 | return 0; | |
1531 | } | |
1532 | ||
1533 | int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans, | |
1534 | struct btrfs_root *root, struct inode *dir, | |
1535 | u64 index) | |
1536 | { | |
1537 | struct btrfs_delayed_node *node; | |
1538 | struct btrfs_delayed_item *item; | |
1539 | struct btrfs_key item_key; | |
1540 | int ret; | |
1541 | ||
1542 | node = btrfs_get_or_create_delayed_node(dir); | |
1543 | if (IS_ERR(node)) | |
1544 | return PTR_ERR(node); | |
1545 | ||
0d0ca30f | 1546 | item_key.objectid = btrfs_ino(dir); |
16cdcec7 MX |
1547 | btrfs_set_key_type(&item_key, BTRFS_DIR_INDEX_KEY); |
1548 | item_key.offset = index; | |
1549 | ||
1550 | ret = btrfs_delete_delayed_insertion_item(root, node, &item_key); | |
1551 | if (!ret) | |
1552 | goto end; | |
1553 | ||
1554 | item = btrfs_alloc_delayed_item(0); | |
1555 | if (!item) { | |
1556 | ret = -ENOMEM; | |
1557 | goto end; | |
1558 | } | |
1559 | ||
1560 | item->key = item_key; | |
1561 | ||
1562 | ret = btrfs_delayed_item_reserve_metadata(trans, root, item); | |
1563 | /* | |
1564 | * we have reserved enough space when we start a new transaction, | |
1565 | * so reserving metadata failure is impossible. | |
1566 | */ | |
1567 | BUG_ON(ret); | |
1568 | ||
1569 | mutex_lock(&node->mutex); | |
1570 | ret = __btrfs_add_delayed_deletion_item(node, item); | |
1571 | if (unlikely(ret)) { | |
1572 | printk(KERN_ERR "err add delayed dir index item(index: %llu) " | |
1573 | "into the deletion tree of the delayed node" | |
1574 | "(root id: %llu, inode id: %llu, errno: %d)\n", | |
1575 | (unsigned long long)index, | |
1576 | (unsigned long long)node->root->objectid, | |
1577 | (unsigned long long)node->inode_id, | |
1578 | ret); | |
1579 | BUG(); | |
1580 | } | |
1581 | mutex_unlock(&node->mutex); | |
1582 | end: | |
1583 | btrfs_release_delayed_node(node); | |
1584 | return ret; | |
1585 | } | |
1586 | ||
1587 | int btrfs_inode_delayed_dir_index_count(struct inode *inode) | |
1588 | { | |
2f7e33d4 | 1589 | struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode); |
16cdcec7 MX |
1590 | |
1591 | if (!delayed_node) | |
1592 | return -ENOENT; | |
1593 | ||
1594 | /* | |
1595 | * Since we have held i_mutex of this directory, it is impossible that | |
1596 | * a new directory index is added into the delayed node and index_cnt | |
1597 | * is updated now. So we needn't lock the delayed node. | |
1598 | */ | |
2f7e33d4 MX |
1599 | if (!delayed_node->index_cnt) { |
1600 | btrfs_release_delayed_node(delayed_node); | |
16cdcec7 | 1601 | return -EINVAL; |
2f7e33d4 | 1602 | } |
16cdcec7 MX |
1603 | |
1604 | BTRFS_I(inode)->index_cnt = delayed_node->index_cnt; | |
2f7e33d4 MX |
1605 | btrfs_release_delayed_node(delayed_node); |
1606 | return 0; | |
16cdcec7 MX |
1607 | } |
1608 | ||
1609 | void btrfs_get_delayed_items(struct inode *inode, struct list_head *ins_list, | |
1610 | struct list_head *del_list) | |
1611 | { | |
1612 | struct btrfs_delayed_node *delayed_node; | |
1613 | struct btrfs_delayed_item *item; | |
1614 | ||
1615 | delayed_node = btrfs_get_delayed_node(inode); | |
1616 | if (!delayed_node) | |
1617 | return; | |
1618 | ||
1619 | mutex_lock(&delayed_node->mutex); | |
1620 | item = __btrfs_first_delayed_insertion_item(delayed_node); | |
1621 | while (item) { | |
1622 | atomic_inc(&item->refs); | |
1623 | list_add_tail(&item->readdir_list, ins_list); | |
1624 | item = __btrfs_next_delayed_item(item); | |
1625 | } | |
1626 | ||
1627 | item = __btrfs_first_delayed_deletion_item(delayed_node); | |
1628 | while (item) { | |
1629 | atomic_inc(&item->refs); | |
1630 | list_add_tail(&item->readdir_list, del_list); | |
1631 | item = __btrfs_next_delayed_item(item); | |
1632 | } | |
1633 | mutex_unlock(&delayed_node->mutex); | |
1634 | /* | |
1635 | * This delayed node is still cached in the btrfs inode, so refs | |
1636 | * must be > 1 now, and we needn't check it is going to be freed | |
1637 | * or not. | |
1638 | * | |
1639 | * Besides that, this function is used to read dir, we do not | |
1640 | * insert/delete delayed items in this period. So we also needn't | |
1641 | * requeue or dequeue this delayed node. | |
1642 | */ | |
1643 | atomic_dec(&delayed_node->refs); | |
1644 | } | |
1645 | ||
1646 | void btrfs_put_delayed_items(struct list_head *ins_list, | |
1647 | struct list_head *del_list) | |
1648 | { | |
1649 | struct btrfs_delayed_item *curr, *next; | |
1650 | ||
1651 | list_for_each_entry_safe(curr, next, ins_list, readdir_list) { | |
1652 | list_del(&curr->readdir_list); | |
1653 | if (atomic_dec_and_test(&curr->refs)) | |
1654 | kfree(curr); | |
1655 | } | |
1656 | ||
1657 | list_for_each_entry_safe(curr, next, del_list, readdir_list) { | |
1658 | list_del(&curr->readdir_list); | |
1659 | if (atomic_dec_and_test(&curr->refs)) | |
1660 | kfree(curr); | |
1661 | } | |
1662 | } | |
1663 | ||
1664 | int btrfs_should_delete_dir_index(struct list_head *del_list, | |
1665 | u64 index) | |
1666 | { | |
1667 | struct btrfs_delayed_item *curr, *next; | |
1668 | int ret; | |
1669 | ||
1670 | if (list_empty(del_list)) | |
1671 | return 0; | |
1672 | ||
1673 | list_for_each_entry_safe(curr, next, del_list, readdir_list) { | |
1674 | if (curr->key.offset > index) | |
1675 | break; | |
1676 | ||
1677 | list_del(&curr->readdir_list); | |
1678 | ret = (curr->key.offset == index); | |
1679 | ||
1680 | if (atomic_dec_and_test(&curr->refs)) | |
1681 | kfree(curr); | |
1682 | ||
1683 | if (ret) | |
1684 | return 1; | |
1685 | else | |
1686 | continue; | |
1687 | } | |
1688 | return 0; | |
1689 | } | |
1690 | ||
1691 | /* | |
1692 | * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree | |
1693 | * | |
1694 | */ | |
1695 | int btrfs_readdir_delayed_dir_index(struct file *filp, void *dirent, | |
1696 | filldir_t filldir, | |
1697 | struct list_head *ins_list) | |
1698 | { | |
1699 | struct btrfs_dir_item *di; | |
1700 | struct btrfs_delayed_item *curr, *next; | |
1701 | struct btrfs_key location; | |
1702 | char *name; | |
1703 | int name_len; | |
1704 | int over = 0; | |
1705 | unsigned char d_type; | |
1706 | ||
1707 | if (list_empty(ins_list)) | |
1708 | return 0; | |
1709 | ||
1710 | /* | |
1711 | * Changing the data of the delayed item is impossible. So | |
1712 | * we needn't lock them. And we have held i_mutex of the | |
1713 | * directory, nobody can delete any directory indexes now. | |
1714 | */ | |
1715 | list_for_each_entry_safe(curr, next, ins_list, readdir_list) { | |
1716 | list_del(&curr->readdir_list); | |
1717 | ||
1718 | if (curr->key.offset < filp->f_pos) { | |
1719 | if (atomic_dec_and_test(&curr->refs)) | |
1720 | kfree(curr); | |
1721 | continue; | |
1722 | } | |
1723 | ||
1724 | filp->f_pos = curr->key.offset; | |
1725 | ||
1726 | di = (struct btrfs_dir_item *)curr->data; | |
1727 | name = (char *)(di + 1); | |
1728 | name_len = le16_to_cpu(di->name_len); | |
1729 | ||
1730 | d_type = btrfs_filetype_table[di->type]; | |
1731 | btrfs_disk_key_to_cpu(&location, &di->location); | |
1732 | ||
1733 | over = filldir(dirent, name, name_len, curr->key.offset, | |
1734 | location.objectid, d_type); | |
1735 | ||
1736 | if (atomic_dec_and_test(&curr->refs)) | |
1737 | kfree(curr); | |
1738 | ||
1739 | if (over) | |
1740 | return 1; | |
1741 | } | |
1742 | return 0; | |
1743 | } | |
1744 | ||
1745 | BTRFS_SETGET_STACK_FUNCS(stack_inode_generation, struct btrfs_inode_item, | |
1746 | generation, 64); | |
1747 | BTRFS_SETGET_STACK_FUNCS(stack_inode_sequence, struct btrfs_inode_item, | |
1748 | sequence, 64); | |
1749 | BTRFS_SETGET_STACK_FUNCS(stack_inode_transid, struct btrfs_inode_item, | |
1750 | transid, 64); | |
1751 | BTRFS_SETGET_STACK_FUNCS(stack_inode_size, struct btrfs_inode_item, size, 64); | |
1752 | BTRFS_SETGET_STACK_FUNCS(stack_inode_nbytes, struct btrfs_inode_item, | |
1753 | nbytes, 64); | |
1754 | BTRFS_SETGET_STACK_FUNCS(stack_inode_block_group, struct btrfs_inode_item, | |
1755 | block_group, 64); | |
1756 | BTRFS_SETGET_STACK_FUNCS(stack_inode_nlink, struct btrfs_inode_item, nlink, 32); | |
1757 | BTRFS_SETGET_STACK_FUNCS(stack_inode_uid, struct btrfs_inode_item, uid, 32); | |
1758 | BTRFS_SETGET_STACK_FUNCS(stack_inode_gid, struct btrfs_inode_item, gid, 32); | |
1759 | BTRFS_SETGET_STACK_FUNCS(stack_inode_mode, struct btrfs_inode_item, mode, 32); | |
1760 | BTRFS_SETGET_STACK_FUNCS(stack_inode_rdev, struct btrfs_inode_item, rdev, 64); | |
1761 | BTRFS_SETGET_STACK_FUNCS(stack_inode_flags, struct btrfs_inode_item, flags, 64); | |
1762 | ||
1763 | BTRFS_SETGET_STACK_FUNCS(stack_timespec_sec, struct btrfs_timespec, sec, 64); | |
1764 | BTRFS_SETGET_STACK_FUNCS(stack_timespec_nsec, struct btrfs_timespec, nsec, 32); | |
1765 | ||
1766 | static void fill_stack_inode_item(struct btrfs_trans_handle *trans, | |
1767 | struct btrfs_inode_item *inode_item, | |
1768 | struct inode *inode) | |
1769 | { | |
2f2f43d3 EB |
1770 | btrfs_set_stack_inode_uid(inode_item, i_uid_read(inode)); |
1771 | btrfs_set_stack_inode_gid(inode_item, i_gid_read(inode)); | |
16cdcec7 MX |
1772 | btrfs_set_stack_inode_size(inode_item, BTRFS_I(inode)->disk_i_size); |
1773 | btrfs_set_stack_inode_mode(inode_item, inode->i_mode); | |
1774 | btrfs_set_stack_inode_nlink(inode_item, inode->i_nlink); | |
1775 | btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(inode)); | |
1776 | btrfs_set_stack_inode_generation(inode_item, | |
1777 | BTRFS_I(inode)->generation); | |
0c4d2d95 | 1778 | btrfs_set_stack_inode_sequence(inode_item, inode->i_version); |
16cdcec7 MX |
1779 | btrfs_set_stack_inode_transid(inode_item, trans->transid); |
1780 | btrfs_set_stack_inode_rdev(inode_item, inode->i_rdev); | |
1781 | btrfs_set_stack_inode_flags(inode_item, BTRFS_I(inode)->flags); | |
ff5714cc | 1782 | btrfs_set_stack_inode_block_group(inode_item, 0); |
16cdcec7 MX |
1783 | |
1784 | btrfs_set_stack_timespec_sec(btrfs_inode_atime(inode_item), | |
1785 | inode->i_atime.tv_sec); | |
1786 | btrfs_set_stack_timespec_nsec(btrfs_inode_atime(inode_item), | |
1787 | inode->i_atime.tv_nsec); | |
1788 | ||
1789 | btrfs_set_stack_timespec_sec(btrfs_inode_mtime(inode_item), | |
1790 | inode->i_mtime.tv_sec); | |
1791 | btrfs_set_stack_timespec_nsec(btrfs_inode_mtime(inode_item), | |
1792 | inode->i_mtime.tv_nsec); | |
1793 | ||
1794 | btrfs_set_stack_timespec_sec(btrfs_inode_ctime(inode_item), | |
1795 | inode->i_ctime.tv_sec); | |
1796 | btrfs_set_stack_timespec_nsec(btrfs_inode_ctime(inode_item), | |
1797 | inode->i_ctime.tv_nsec); | |
1798 | } | |
1799 | ||
2f7e33d4 MX |
1800 | int btrfs_fill_inode(struct inode *inode, u32 *rdev) |
1801 | { | |
1802 | struct btrfs_delayed_node *delayed_node; | |
1803 | struct btrfs_inode_item *inode_item; | |
1804 | struct btrfs_timespec *tspec; | |
1805 | ||
1806 | delayed_node = btrfs_get_delayed_node(inode); | |
1807 | if (!delayed_node) | |
1808 | return -ENOENT; | |
1809 | ||
1810 | mutex_lock(&delayed_node->mutex); | |
1811 | if (!delayed_node->inode_dirty) { | |
1812 | mutex_unlock(&delayed_node->mutex); | |
1813 | btrfs_release_delayed_node(delayed_node); | |
1814 | return -ENOENT; | |
1815 | } | |
1816 | ||
1817 | inode_item = &delayed_node->inode_item; | |
1818 | ||
2f2f43d3 EB |
1819 | i_uid_write(inode, btrfs_stack_inode_uid(inode_item)); |
1820 | i_gid_write(inode, btrfs_stack_inode_gid(inode_item)); | |
2f7e33d4 MX |
1821 | btrfs_i_size_write(inode, btrfs_stack_inode_size(inode_item)); |
1822 | inode->i_mode = btrfs_stack_inode_mode(inode_item); | |
bfe86848 | 1823 | set_nlink(inode, btrfs_stack_inode_nlink(inode_item)); |
2f7e33d4 MX |
1824 | inode_set_bytes(inode, btrfs_stack_inode_nbytes(inode_item)); |
1825 | BTRFS_I(inode)->generation = btrfs_stack_inode_generation(inode_item); | |
0c4d2d95 | 1826 | inode->i_version = btrfs_stack_inode_sequence(inode_item); |
2f7e33d4 MX |
1827 | inode->i_rdev = 0; |
1828 | *rdev = btrfs_stack_inode_rdev(inode_item); | |
1829 | BTRFS_I(inode)->flags = btrfs_stack_inode_flags(inode_item); | |
1830 | ||
1831 | tspec = btrfs_inode_atime(inode_item); | |
1832 | inode->i_atime.tv_sec = btrfs_stack_timespec_sec(tspec); | |
1833 | inode->i_atime.tv_nsec = btrfs_stack_timespec_nsec(tspec); | |
1834 | ||
1835 | tspec = btrfs_inode_mtime(inode_item); | |
1836 | inode->i_mtime.tv_sec = btrfs_stack_timespec_sec(tspec); | |
1837 | inode->i_mtime.tv_nsec = btrfs_stack_timespec_nsec(tspec); | |
1838 | ||
1839 | tspec = btrfs_inode_ctime(inode_item); | |
1840 | inode->i_ctime.tv_sec = btrfs_stack_timespec_sec(tspec); | |
1841 | inode->i_ctime.tv_nsec = btrfs_stack_timespec_nsec(tspec); | |
1842 | ||
1843 | inode->i_generation = BTRFS_I(inode)->generation; | |
1844 | BTRFS_I(inode)->index_cnt = (u64)-1; | |
1845 | ||
1846 | mutex_unlock(&delayed_node->mutex); | |
1847 | btrfs_release_delayed_node(delayed_node); | |
1848 | return 0; | |
1849 | } | |
1850 | ||
16cdcec7 MX |
1851 | int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans, |
1852 | struct btrfs_root *root, struct inode *inode) | |
1853 | { | |
1854 | struct btrfs_delayed_node *delayed_node; | |
aa0467d8 | 1855 | int ret = 0; |
16cdcec7 MX |
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 | mutex_lock(&delayed_node->mutex); | |
1862 | if (delayed_node->inode_dirty) { | |
1863 | fill_stack_inode_item(trans, &delayed_node->inode_item, inode); | |
1864 | goto release_node; | |
1865 | } | |
1866 | ||
7fd2ae21 JB |
1867 | ret = btrfs_delayed_inode_reserve_metadata(trans, root, inode, |
1868 | delayed_node); | |
c06a0e12 JB |
1869 | if (ret) |
1870 | goto release_node; | |
16cdcec7 MX |
1871 | |
1872 | fill_stack_inode_item(trans, &delayed_node->inode_item, inode); | |
1873 | delayed_node->inode_dirty = 1; | |
1874 | delayed_node->count++; | |
1875 | atomic_inc(&root->fs_info->delayed_root->items); | |
1876 | release_node: | |
1877 | mutex_unlock(&delayed_node->mutex); | |
1878 | btrfs_release_delayed_node(delayed_node); | |
1879 | return ret; | |
1880 | } | |
1881 | ||
1882 | static void __btrfs_kill_delayed_node(struct btrfs_delayed_node *delayed_node) | |
1883 | { | |
1884 | struct btrfs_root *root = delayed_node->root; | |
1885 | struct btrfs_delayed_item *curr_item, *prev_item; | |
1886 | ||
1887 | mutex_lock(&delayed_node->mutex); | |
1888 | curr_item = __btrfs_first_delayed_insertion_item(delayed_node); | |
1889 | while (curr_item) { | |
1890 | btrfs_delayed_item_release_metadata(root, curr_item); | |
1891 | prev_item = curr_item; | |
1892 | curr_item = __btrfs_next_delayed_item(prev_item); | |
1893 | btrfs_release_delayed_item(prev_item); | |
1894 | } | |
1895 | ||
1896 | curr_item = __btrfs_first_delayed_deletion_item(delayed_node); | |
1897 | while (curr_item) { | |
1898 | btrfs_delayed_item_release_metadata(root, curr_item); | |
1899 | prev_item = curr_item; | |
1900 | curr_item = __btrfs_next_delayed_item(prev_item); | |
1901 | btrfs_release_delayed_item(prev_item); | |
1902 | } | |
1903 | ||
1904 | if (delayed_node->inode_dirty) { | |
1905 | btrfs_delayed_inode_release_metadata(root, delayed_node); | |
1906 | btrfs_release_delayed_inode(delayed_node); | |
1907 | } | |
1908 | mutex_unlock(&delayed_node->mutex); | |
1909 | } | |
1910 | ||
1911 | void btrfs_kill_delayed_inode_items(struct inode *inode) | |
1912 | { | |
1913 | struct btrfs_delayed_node *delayed_node; | |
1914 | ||
1915 | delayed_node = btrfs_get_delayed_node(inode); | |
1916 | if (!delayed_node) | |
1917 | return; | |
1918 | ||
1919 | __btrfs_kill_delayed_node(delayed_node); | |
1920 | btrfs_release_delayed_node(delayed_node); | |
1921 | } | |
1922 | ||
1923 | void btrfs_kill_all_delayed_nodes(struct btrfs_root *root) | |
1924 | { | |
1925 | u64 inode_id = 0; | |
1926 | struct btrfs_delayed_node *delayed_nodes[8]; | |
1927 | int i, n; | |
1928 | ||
1929 | while (1) { | |
1930 | spin_lock(&root->inode_lock); | |
1931 | n = radix_tree_gang_lookup(&root->delayed_nodes_tree, | |
1932 | (void **)delayed_nodes, inode_id, | |
1933 | ARRAY_SIZE(delayed_nodes)); | |
1934 | if (!n) { | |
1935 | spin_unlock(&root->inode_lock); | |
1936 | break; | |
1937 | } | |
1938 | ||
1939 | inode_id = delayed_nodes[n - 1]->inode_id + 1; | |
1940 | ||
1941 | for (i = 0; i < n; i++) | |
1942 | atomic_inc(&delayed_nodes[i]->refs); | |
1943 | spin_unlock(&root->inode_lock); | |
1944 | ||
1945 | for (i = 0; i < n; i++) { | |
1946 | __btrfs_kill_delayed_node(delayed_nodes[i]); | |
1947 | btrfs_release_delayed_node(delayed_nodes[i]); | |
1948 | } | |
1949 | } | |
1950 | } | |
67cde344 MX |
1951 | |
1952 | void btrfs_destroy_delayed_inodes(struct btrfs_root *root) | |
1953 | { | |
1954 | struct btrfs_delayed_root *delayed_root; | |
1955 | struct btrfs_delayed_node *curr_node, *prev_node; | |
1956 | ||
1957 | delayed_root = btrfs_get_delayed_root(root); | |
1958 | ||
1959 | curr_node = btrfs_first_delayed_node(delayed_root); | |
1960 | while (curr_node) { | |
1961 | __btrfs_kill_delayed_node(curr_node); | |
1962 | ||
1963 | prev_node = curr_node; | |
1964 | curr_node = btrfs_next_delayed_node(curr_node); | |
1965 | btrfs_release_delayed_node(prev_node); | |
1966 | } | |
1967 | } | |
1968 |