blob: a2ae42720a6afe92701de402b837d56dd66d03b4 [file] [log] [blame]
Miao Xie16cdcec2011-04-22 18:12:22 +08001/*
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"
Qu Wenruo3cae2102013-07-16 11:19:18 +080024#include "ctree.h"
Miao Xie16cdcec2011-04-22 18:12:22 +080025
Chris Masonde3cb942013-03-04 17:13:31 -050026#define BTRFS_DELAYED_WRITEBACK 512
27#define BTRFS_DELAYED_BACKGROUND 128
28#define BTRFS_DELAYED_BATCH 16
Miao Xie16cdcec2011-04-22 18:12:22 +080029
30static struct kmem_cache *delayed_node_cache;
31
32int __init btrfs_delayed_inode_init(void)
33{
David Sterba837e1972012-09-07 03:00:48 -060034 delayed_node_cache = kmem_cache_create("btrfs_delayed_node",
Miao Xie16cdcec2011-04-22 18:12:22 +080035 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
44void btrfs_delayed_inode_exit(void)
45{
46 if (delayed_node_cache)
47 kmem_cache_destroy(delayed_node_cache);
48}
49
50static 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;
Miao Xie7cf35d92013-12-26 13:07:05 +080058 delayed_node->flags = 0;
Miao Xie16cdcec2011-04-22 18:12:22 +080059 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;
Li Zefan293f7e02012-07-10 00:58:58 -060066 memset(&delayed_node->inode_item, 0, sizeof(delayed_node->inode_item));
Miao Xie16cdcec2011-04-22 18:12:22 +080067}
68
69static 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
81static inline struct btrfs_delayed_root *btrfs_get_delayed_root(
82 struct btrfs_root *root)
83{
84 return root->fs_info->delayed_root;
85}
86
Miao Xie2f7e33d2011-06-23 07:27:13 +000087static struct btrfs_delayed_node *btrfs_get_delayed_node(struct inode *inode)
88{
89 struct btrfs_inode *btrfs_inode = BTRFS_I(inode);
90 struct btrfs_root *root = btrfs_inode->root;
91 u64 ino = btrfs_ino(inode);
92 struct btrfs_delayed_node *node;
93
94 node = ACCESS_ONCE(btrfs_inode->delayed_node);
95 if (node) {
96 atomic_inc(&node->refs);
97 return node;
98 }
99
100 spin_lock(&root->inode_lock);
101 node = radix_tree_lookup(&root->delayed_nodes_tree, ino);
102 if (node) {
103 if (btrfs_inode->delayed_node) {
104 atomic_inc(&node->refs); /* can be accessed */
105 BUG_ON(btrfs_inode->delayed_node != node);
106 spin_unlock(&root->inode_lock);
107 return node;
108 }
109 btrfs_inode->delayed_node = node;
Rashika95e94d12013-10-31 03:12:42 +0530110 /* can be accessed and cached in the inode */
111 atomic_add(2, &node->refs);
Miao Xie2f7e33d2011-06-23 07:27:13 +0000112 spin_unlock(&root->inode_lock);
113 return node;
114 }
115 spin_unlock(&root->inode_lock);
116
117 return NULL;
118}
119
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100120/* Will return either the node or PTR_ERR(-ENOMEM) */
Miao Xie16cdcec2011-04-22 18:12:22 +0800121static 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;
Chris Mason0d0ca302011-05-22 07:11:22 -0400127 u64 ino = btrfs_ino(inode);
Miao Xie16cdcec2011-04-22 18:12:22 +0800128 int ret;
129
130again:
Miao Xie2f7e33d2011-06-23 07:27:13 +0000131 node = btrfs_get_delayed_node(inode);
132 if (node)
Miao Xie16cdcec2011-04-22 18:12:22 +0800133 return node;
Miao Xie16cdcec2011-04-22 18:12:22 +0800134
135 node = kmem_cache_alloc(delayed_node_cache, GFP_NOFS);
136 if (!node)
137 return ERR_PTR(-ENOMEM);
Chris Mason0d0ca302011-05-22 07:11:22 -0400138 btrfs_init_delayed_node(node, root, ino);
Miao Xie16cdcec2011-04-22 18:12:22 +0800139
Rashika95e94d12013-10-31 03:12:42 +0530140 /* cached in the btrfs inode and can be accessed */
141 atomic_add(2, &node->refs);
Miao Xie16cdcec2011-04-22 18:12:22 +0800142
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);
Chris Mason0d0ca302011-05-22 07:11:22 -0400150 ret = radix_tree_insert(&root->delayed_nodes_tree, ino, node);
Miao Xie16cdcec2011-04-22 18:12:22 +0800151 if (ret == -EEXIST) {
Miao Xie16cdcec2011-04-22 18:12:22 +0800152 spin_unlock(&root->inode_lock);
Jeff Mahoney96493032014-05-27 13:53:20 -0400153 kmem_cache_free(delayed_node_cache, node);
Miao Xie16cdcec2011-04-22 18:12:22 +0800154 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 */
169static 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);
Miao Xie7cf35d92013-12-26 13:07:05 +0800174 if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
Miao Xie16cdcec2011-04-22 18:12:22 +0800175 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++;
Miao Xie7cf35d92013-12-26 13:07:05 +0800184 set_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
Miao Xie16cdcec2011-04-22 18:12:22 +0800185 }
186 spin_unlock(&root->lock);
187}
188
189/* Call it when holding delayed_node->mutex */
190static void btrfs_dequeue_delayed_node(struct btrfs_delayed_root *root,
191 struct btrfs_delayed_node *node)
192{
193 spin_lock(&root->lock);
Miao Xie7cf35d92013-12-26 13:07:05 +0800194 if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
Miao Xie16cdcec2011-04-22 18:12:22 +0800195 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);
Miao Xie7cf35d92013-12-26 13:07:05 +0800200 clear_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
Miao Xie16cdcec2011-04-22 18:12:22 +0800201 }
202 spin_unlock(&root->lock);
203}
204
Eric Sandeen48a3b632013-04-25 20:41:01 +0000205static struct btrfs_delayed_node *btrfs_first_delayed_node(
Miao Xie16cdcec2011-04-22 18:12:22 +0800206 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);
218out:
219 spin_unlock(&delayed_root->lock);
220
221 return node;
222}
223
Eric Sandeen48a3b632013-04-25 20:41:01 +0000224static struct btrfs_delayed_node *btrfs_next_delayed_node(
Miao Xie16cdcec2011-04-22 18:12:22 +0800225 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);
Miao Xie7cf35d92013-12-26 13:07:05 +0800233 if (!test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
234 /* not in the list */
Miao Xie16cdcec2011-04-22 18:12:22 +0800235 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);
245out:
246 spin_unlock(&delayed_root->lock);
247
248 return next;
249}
250
251static 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)) {
Jeff Mahoney96493032014-05-27 13:53:20 -0400270 bool free = false;
Miao Xie16cdcec2011-04-22 18:12:22 +0800271 struct btrfs_root *root = delayed_node->root;
272 spin_lock(&root->inode_lock);
273 if (atomic_read(&delayed_node->refs) == 0) {
274 radix_tree_delete(&root->delayed_nodes_tree,
275 delayed_node->inode_id);
Jeff Mahoney96493032014-05-27 13:53:20 -0400276 free = true;
Miao Xie16cdcec2011-04-22 18:12:22 +0800277 }
278 spin_unlock(&root->inode_lock);
Jeff Mahoney96493032014-05-27 13:53:20 -0400279 if (free)
280 kmem_cache_free(delayed_node_cache, delayed_node);
Miao Xie16cdcec2011-04-22 18:12:22 +0800281 }
282}
283
284static inline void btrfs_release_delayed_node(struct btrfs_delayed_node *node)
285{
286 __btrfs_release_delayed_node(node, 0);
287}
288
Eric Sandeen48a3b632013-04-25 20:41:01 +0000289static struct btrfs_delayed_node *btrfs_first_prepared_delayed_node(
Miao Xie16cdcec2011-04-22 18:12:22 +0800290 struct btrfs_delayed_root *delayed_root)
291{
292 struct list_head *p;
293 struct btrfs_delayed_node *node = NULL;
294
295 spin_lock(&delayed_root->lock);
296 if (list_empty(&delayed_root->prepare_list))
297 goto out;
298
299 p = delayed_root->prepare_list.next;
300 list_del_init(p);
301 node = list_entry(p, struct btrfs_delayed_node, p_list);
302 atomic_inc(&node->refs);
303out:
304 spin_unlock(&delayed_root->lock);
305
306 return node;
307}
308
309static inline void btrfs_release_prepared_delayed_node(
310 struct btrfs_delayed_node *node)
311{
312 __btrfs_release_delayed_node(node, 1);
313}
314
Eric Sandeen48a3b632013-04-25 20:41:01 +0000315static struct btrfs_delayed_item *btrfs_alloc_delayed_item(u32 data_len)
Miao Xie16cdcec2011-04-22 18:12:22 +0800316{
317 struct btrfs_delayed_item *item;
318 item = kmalloc(sizeof(*item) + data_len, GFP_NOFS);
319 if (item) {
320 item->data_len = data_len;
321 item->ins_or_del = 0;
322 item->bytes_reserved = 0;
Miao Xie16cdcec2011-04-22 18:12:22 +0800323 item->delayed_node = NULL;
324 atomic_set(&item->refs, 1);
325 }
326 return item;
327}
328
329/*
330 * __btrfs_lookup_delayed_item - look up the delayed item by key
331 * @delayed_node: pointer to the delayed node
332 * @key: the key to look up
333 * @prev: used to store the prev item if the right item isn't found
334 * @next: used to store the next item if the right item isn't found
335 *
336 * Note: if we don't find the right item, we will return the prev item and
337 * the next item.
338 */
339static struct btrfs_delayed_item *__btrfs_lookup_delayed_item(
340 struct rb_root *root,
341 struct btrfs_key *key,
342 struct btrfs_delayed_item **prev,
343 struct btrfs_delayed_item **next)
344{
345 struct rb_node *node, *prev_node = NULL;
346 struct btrfs_delayed_item *delayed_item = NULL;
347 int ret = 0;
348
349 node = root->rb_node;
350
351 while (node) {
352 delayed_item = rb_entry(node, struct btrfs_delayed_item,
353 rb_node);
354 prev_node = node;
355 ret = btrfs_comp_cpu_keys(&delayed_item->key, key);
356 if (ret < 0)
357 node = node->rb_right;
358 else if (ret > 0)
359 node = node->rb_left;
360 else
361 return delayed_item;
362 }
363
364 if (prev) {
365 if (!prev_node)
366 *prev = NULL;
367 else if (ret < 0)
368 *prev = delayed_item;
369 else if ((node = rb_prev(prev_node)) != NULL) {
370 *prev = rb_entry(node, struct btrfs_delayed_item,
371 rb_node);
372 } else
373 *prev = NULL;
374 }
375
376 if (next) {
377 if (!prev_node)
378 *next = NULL;
379 else if (ret > 0)
380 *next = delayed_item;
381 else if ((node = rb_next(prev_node)) != NULL) {
382 *next = rb_entry(node, struct btrfs_delayed_item,
383 rb_node);
384 } else
385 *next = NULL;
386 }
387 return NULL;
388}
389
Eric Sandeen48a3b632013-04-25 20:41:01 +0000390static struct btrfs_delayed_item *__btrfs_lookup_delayed_insertion_item(
Miao Xie16cdcec2011-04-22 18:12:22 +0800391 struct btrfs_delayed_node *delayed_node,
392 struct btrfs_key *key)
393{
394 struct btrfs_delayed_item *item;
395
396 item = __btrfs_lookup_delayed_item(&delayed_node->ins_root, key,
397 NULL, NULL);
398 return item;
399}
400
Miao Xie16cdcec2011-04-22 18:12:22 +0800401static int __btrfs_add_delayed_item(struct btrfs_delayed_node *delayed_node,
402 struct btrfs_delayed_item *ins,
403 int action)
404{
405 struct rb_node **p, *node;
406 struct rb_node *parent_node = NULL;
407 struct rb_root *root;
408 struct btrfs_delayed_item *item;
409 int cmp;
410
411 if (action == BTRFS_DELAYED_INSERTION_ITEM)
412 root = &delayed_node->ins_root;
413 else if (action == BTRFS_DELAYED_DELETION_ITEM)
414 root = &delayed_node->del_root;
415 else
416 BUG();
417 p = &root->rb_node;
418 node = &ins->rb_node;
419
420 while (*p) {
421 parent_node = *p;
422 item = rb_entry(parent_node, struct btrfs_delayed_item,
423 rb_node);
424
425 cmp = btrfs_comp_cpu_keys(&item->key, &ins->key);
426 if (cmp < 0)
427 p = &(*p)->rb_right;
428 else if (cmp > 0)
429 p = &(*p)->rb_left;
430 else
431 return -EEXIST;
432 }
433
434 rb_link_node(node, parent_node, p);
435 rb_insert_color(node, root);
436 ins->delayed_node = delayed_node;
437 ins->ins_or_del = action;
438
439 if (ins->key.type == BTRFS_DIR_INDEX_KEY &&
440 action == BTRFS_DELAYED_INSERTION_ITEM &&
441 ins->key.offset >= delayed_node->index_cnt)
442 delayed_node->index_cnt = ins->key.offset + 1;
443
444 delayed_node->count++;
445 atomic_inc(&delayed_node->root->fs_info->delayed_root->items);
446 return 0;
447}
448
449static int __btrfs_add_delayed_insertion_item(struct btrfs_delayed_node *node,
450 struct btrfs_delayed_item *item)
451{
452 return __btrfs_add_delayed_item(node, item,
453 BTRFS_DELAYED_INSERTION_ITEM);
454}
455
456static int __btrfs_add_delayed_deletion_item(struct btrfs_delayed_node *node,
457 struct btrfs_delayed_item *item)
458{
459 return __btrfs_add_delayed_item(node, item,
460 BTRFS_DELAYED_DELETION_ITEM);
461}
462
Chris Masonde3cb942013-03-04 17:13:31 -0500463static void finish_one_item(struct btrfs_delayed_root *delayed_root)
464{
465 int seq = atomic_inc_return(&delayed_root->items_seq);
466 if ((atomic_dec_return(&delayed_root->items) <
467 BTRFS_DELAYED_BACKGROUND || seq % BTRFS_DELAYED_BATCH == 0) &&
468 waitqueue_active(&delayed_root->wait))
469 wake_up(&delayed_root->wait);
470}
471
Miao Xie16cdcec2011-04-22 18:12:22 +0800472static void __btrfs_remove_delayed_item(struct btrfs_delayed_item *delayed_item)
473{
474 struct rb_root *root;
475 struct btrfs_delayed_root *delayed_root;
476
477 delayed_root = delayed_item->delayed_node->root->fs_info->delayed_root;
478
479 BUG_ON(!delayed_root);
480 BUG_ON(delayed_item->ins_or_del != BTRFS_DELAYED_DELETION_ITEM &&
481 delayed_item->ins_or_del != BTRFS_DELAYED_INSERTION_ITEM);
482
483 if (delayed_item->ins_or_del == BTRFS_DELAYED_INSERTION_ITEM)
484 root = &delayed_item->delayed_node->ins_root;
485 else
486 root = &delayed_item->delayed_node->del_root;
487
488 rb_erase(&delayed_item->rb_node, root);
489 delayed_item->delayed_node->count--;
Chris Masonde3cb942013-03-04 17:13:31 -0500490
491 finish_one_item(delayed_root);
Miao Xie16cdcec2011-04-22 18:12:22 +0800492}
493
494static void btrfs_release_delayed_item(struct btrfs_delayed_item *item)
495{
496 if (item) {
497 __btrfs_remove_delayed_item(item);
498 if (atomic_dec_and_test(&item->refs))
499 kfree(item);
500 }
501}
502
Eric Sandeen48a3b632013-04-25 20:41:01 +0000503static struct btrfs_delayed_item *__btrfs_first_delayed_insertion_item(
Miao Xie16cdcec2011-04-22 18:12:22 +0800504 struct btrfs_delayed_node *delayed_node)
505{
506 struct rb_node *p;
507 struct btrfs_delayed_item *item = NULL;
508
509 p = rb_first(&delayed_node->ins_root);
510 if (p)
511 item = rb_entry(p, struct btrfs_delayed_item, rb_node);
512
513 return item;
514}
515
Eric Sandeen48a3b632013-04-25 20:41:01 +0000516static struct btrfs_delayed_item *__btrfs_first_delayed_deletion_item(
Miao Xie16cdcec2011-04-22 18:12:22 +0800517 struct btrfs_delayed_node *delayed_node)
518{
519 struct rb_node *p;
520 struct btrfs_delayed_item *item = NULL;
521
522 p = rb_first(&delayed_node->del_root);
523 if (p)
524 item = rb_entry(p, struct btrfs_delayed_item, rb_node);
525
526 return item;
527}
528
Eric Sandeen48a3b632013-04-25 20:41:01 +0000529static struct btrfs_delayed_item *__btrfs_next_delayed_item(
Miao Xie16cdcec2011-04-22 18:12:22 +0800530 struct btrfs_delayed_item *item)
531{
532 struct rb_node *p;
533 struct btrfs_delayed_item *next = NULL;
534
535 p = rb_next(&item->rb_node);
536 if (p)
537 next = rb_entry(p, struct btrfs_delayed_item, rb_node);
538
539 return next;
540}
541
Miao Xie16cdcec2011-04-22 18:12:22 +0800542static int btrfs_delayed_item_reserve_metadata(struct btrfs_trans_handle *trans,
543 struct btrfs_root *root,
544 struct btrfs_delayed_item *item)
545{
546 struct btrfs_block_rsv *src_rsv;
547 struct btrfs_block_rsv *dst_rsv;
548 u64 num_bytes;
549 int ret;
550
551 if (!trans->bytes_reserved)
552 return 0;
553
554 src_rsv = trans->block_rsv;
Josef Bacik6d668dd2011-11-03 22:54:25 -0400555 dst_rsv = &root->fs_info->delayed_block_rsv;
Miao Xie16cdcec2011-04-22 18:12:22 +0800556
557 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
558 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes);
Josef Bacik8c2a3ca2012-01-10 10:31:31 -0500559 if (!ret) {
560 trace_btrfs_space_reservation(root->fs_info, "delayed_item",
561 item->key.objectid,
562 num_bytes, 1);
Miao Xie16cdcec2011-04-22 18:12:22 +0800563 item->bytes_reserved = num_bytes;
Josef Bacik8c2a3ca2012-01-10 10:31:31 -0500564 }
Miao Xie16cdcec2011-04-22 18:12:22 +0800565
566 return ret;
567}
568
569static void btrfs_delayed_item_release_metadata(struct btrfs_root *root,
570 struct btrfs_delayed_item *item)
571{
Miao Xie19fd2942011-06-15 10:47:30 +0000572 struct btrfs_block_rsv *rsv;
573
Miao Xie16cdcec2011-04-22 18:12:22 +0800574 if (!item->bytes_reserved)
575 return;
576
Josef Bacik6d668dd2011-11-03 22:54:25 -0400577 rsv = &root->fs_info->delayed_block_rsv;
Josef Bacik8c2a3ca2012-01-10 10:31:31 -0500578 trace_btrfs_space_reservation(root->fs_info, "delayed_item",
579 item->key.objectid, item->bytes_reserved,
580 0);
Miao Xie19fd2942011-06-15 10:47:30 +0000581 btrfs_block_rsv_release(root, rsv,
Miao Xie16cdcec2011-04-22 18:12:22 +0800582 item->bytes_reserved);
583}
584
585static int btrfs_delayed_inode_reserve_metadata(
586 struct btrfs_trans_handle *trans,
587 struct btrfs_root *root,
Josef Bacik7fd2ae22011-11-08 15:47:34 -0500588 struct inode *inode,
Miao Xie16cdcec2011-04-22 18:12:22 +0800589 struct btrfs_delayed_node *node)
590{
591 struct btrfs_block_rsv *src_rsv;
592 struct btrfs_block_rsv *dst_rsv;
593 u64 num_bytes;
594 int ret;
Josef Bacik8c2a3ca2012-01-10 10:31:31 -0500595 bool release = false;
Miao Xie16cdcec2011-04-22 18:12:22 +0800596
Miao Xie16cdcec2011-04-22 18:12:22 +0800597 src_rsv = trans->block_rsv;
Josef Bacik6d668dd2011-11-03 22:54:25 -0400598 dst_rsv = &root->fs_info->delayed_block_rsv;
Miao Xie16cdcec2011-04-22 18:12:22 +0800599
600 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
Josef Bacikc06a0e12011-11-04 19:56:02 -0400601
602 /*
603 * btrfs_dirty_inode will update the inode under btrfs_join_transaction
604 * which doesn't reserve space for speed. This is a problem since we
605 * still need to reserve space for this update, so try to reserve the
606 * space.
607 *
608 * Now if src_rsv == delalloc_block_rsv we'll let it just steal since
609 * we're accounted for.
610 */
Chris Masone755d9a2011-12-15 13:36:29 -0500611 if (!src_rsv || (!trans->bytes_reserved &&
Miao Xie66d8f3d2012-09-06 04:02:28 -0600612 src_rsv->type != BTRFS_BLOCK_RSV_DELALLOC)) {
Miao Xie08e007d2012-10-16 11:33:38 +0000613 ret = btrfs_block_rsv_add(root, dst_rsv, num_bytes,
614 BTRFS_RESERVE_NO_FLUSH);
Josef Bacikc06a0e12011-11-04 19:56:02 -0400615 /*
616 * Since we're under a transaction reserve_metadata_bytes could
617 * try to commit the transaction which will make it return
618 * EAGAIN to make us stop the transaction we have, so return
619 * ENOSPC instead so that btrfs_dirty_inode knows what to do.
620 */
621 if (ret == -EAGAIN)
622 ret = -ENOSPC;
Josef Bacik8c2a3ca2012-01-10 10:31:31 -0500623 if (!ret) {
Josef Bacikc06a0e12011-11-04 19:56:02 -0400624 node->bytes_reserved = num_bytes;
Josef Bacik8c2a3ca2012-01-10 10:31:31 -0500625 trace_btrfs_space_reservation(root->fs_info,
626 "delayed_inode",
627 btrfs_ino(inode),
628 num_bytes, 1);
629 }
Josef Bacikc06a0e12011-11-04 19:56:02 -0400630 return ret;
Miao Xie66d8f3d2012-09-06 04:02:28 -0600631 } else if (src_rsv->type == BTRFS_BLOCK_RSV_DELALLOC) {
Josef Bacik7fd2ae22011-11-08 15:47:34 -0500632 spin_lock(&BTRFS_I(inode)->lock);
Josef Bacik72ac3c02012-05-23 14:13:11 -0400633 if (test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
634 &BTRFS_I(inode)->runtime_flags)) {
Josef Bacik7fd2ae22011-11-08 15:47:34 -0500635 spin_unlock(&BTRFS_I(inode)->lock);
636 release = true;
637 goto migrate;
638 }
639 spin_unlock(&BTRFS_I(inode)->lock);
640
641 /* Ok we didn't have space pre-reserved. This shouldn't happen
642 * too often but it can happen if we do delalloc to an existing
643 * inode which gets dirtied because of the time update, and then
644 * isn't touched again until after the transaction commits and
645 * then we try to write out the data. First try to be nice and
646 * reserve something strictly for us. If not be a pain and try
647 * to steal from the delalloc block rsv.
648 */
Miao Xie08e007d2012-10-16 11:33:38 +0000649 ret = btrfs_block_rsv_add(root, dst_rsv, num_bytes,
650 BTRFS_RESERVE_NO_FLUSH);
Josef Bacik7fd2ae22011-11-08 15:47:34 -0500651 if (!ret)
652 goto out;
653
654 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes);
Dulshani Gunawardhanafae7f212013-10-31 10:30:08 +0530655 if (!WARN_ON(ret))
Josef Bacik7fd2ae22011-11-08 15:47:34 -0500656 goto out;
657
658 /*
659 * Ok this is a problem, let's just steal from the global rsv
660 * since this really shouldn't happen that often.
661 */
Josef Bacik7fd2ae22011-11-08 15:47:34 -0500662 ret = btrfs_block_rsv_migrate(&root->fs_info->global_block_rsv,
663 dst_rsv, num_bytes);
664 goto out;
Josef Bacikc06a0e12011-11-04 19:56:02 -0400665 }
666
Josef Bacik7fd2ae22011-11-08 15:47:34 -0500667migrate:
Miao Xie16cdcec2011-04-22 18:12:22 +0800668 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes);
Josef Bacik7fd2ae22011-11-08 15:47:34 -0500669
670out:
671 /*
672 * Migrate only takes a reservation, it doesn't touch the size of the
673 * block_rsv. This is to simplify people who don't normally have things
674 * migrated from their block rsv. If they go to release their
675 * reservation, that will decrease the size as well, so if migrate
676 * reduced size we'd end up with a negative size. But for the
677 * delalloc_meta_reserved stuff we will only know to drop 1 reservation,
678 * but we could in fact do this reserve/migrate dance several times
679 * between the time we did the original reservation and we'd clean it
680 * up. So to take care of this, release the space for the meta
681 * reservation here. I think it may be time for a documentation page on
682 * how block rsvs. work.
683 */
Josef Bacik8c2a3ca2012-01-10 10:31:31 -0500684 if (!ret) {
685 trace_btrfs_space_reservation(root->fs_info, "delayed_inode",
686 btrfs_ino(inode), num_bytes, 1);
Miao Xie16cdcec2011-04-22 18:12:22 +0800687 node->bytes_reserved = num_bytes;
Josef Bacik8c2a3ca2012-01-10 10:31:31 -0500688 }
Miao Xie16cdcec2011-04-22 18:12:22 +0800689
Josef Bacik8c2a3ca2012-01-10 10:31:31 -0500690 if (release) {
691 trace_btrfs_space_reservation(root->fs_info, "delalloc",
692 btrfs_ino(inode), num_bytes, 0);
Josef Bacik7fd2ae22011-11-08 15:47:34 -0500693 btrfs_block_rsv_release(root, src_rsv, num_bytes);
Josef Bacik8c2a3ca2012-01-10 10:31:31 -0500694 }
Miao Xie16cdcec2011-04-22 18:12:22 +0800695
696 return ret;
697}
698
699static void btrfs_delayed_inode_release_metadata(struct btrfs_root *root,
700 struct btrfs_delayed_node *node)
701{
702 struct btrfs_block_rsv *rsv;
703
704 if (!node->bytes_reserved)
705 return;
706
Josef Bacik6d668dd2011-11-03 22:54:25 -0400707 rsv = &root->fs_info->delayed_block_rsv;
Josef Bacik8c2a3ca2012-01-10 10:31:31 -0500708 trace_btrfs_space_reservation(root->fs_info, "delayed_inode",
709 node->inode_id, node->bytes_reserved, 0);
Miao Xie16cdcec2011-04-22 18:12:22 +0800710 btrfs_block_rsv_release(root, rsv,
711 node->bytes_reserved);
712 node->bytes_reserved = 0;
713}
714
715/*
716 * This helper will insert some continuous items into the same leaf according
717 * to the free space of the leaf.
718 */
Tsutomu Itohafe5fea2013-04-16 05:18:22 +0000719static int btrfs_batch_insert_items(struct btrfs_root *root,
720 struct btrfs_path *path,
721 struct btrfs_delayed_item *item)
Miao Xie16cdcec2011-04-22 18:12:22 +0800722{
723 struct btrfs_delayed_item *curr, *next;
724 int free_space;
725 int total_data_size = 0, total_size = 0;
726 struct extent_buffer *leaf;
727 char *data_ptr;
728 struct btrfs_key *keys;
729 u32 *data_size;
730 struct list_head head;
731 int slot;
732 int nitems;
733 int i;
734 int ret = 0;
735
736 BUG_ON(!path->nodes[0]);
737
738 leaf = path->nodes[0];
739 free_space = btrfs_leaf_free_space(root, leaf);
740 INIT_LIST_HEAD(&head);
741
742 next = item;
Chris Mason17aca1c2011-06-03 01:13:45 -0400743 nitems = 0;
Miao Xie16cdcec2011-04-22 18:12:22 +0800744
745 /*
746 * count the number of the continuous items that we can insert in batch
747 */
748 while (total_size + next->data_len + sizeof(struct btrfs_item) <=
749 free_space) {
750 total_data_size += next->data_len;
751 total_size += next->data_len + sizeof(struct btrfs_item);
752 list_add_tail(&next->tree_list, &head);
753 nitems++;
754
755 curr = next;
756 next = __btrfs_next_delayed_item(curr);
757 if (!next)
758 break;
759
760 if (!btrfs_is_continuous_delayed_item(curr, next))
761 break;
762 }
763
764 if (!nitems) {
765 ret = 0;
766 goto out;
767 }
768
769 /*
770 * we need allocate some memory space, but it might cause the task
771 * to sleep, so we set all locked nodes in the path to blocking locks
772 * first.
773 */
774 btrfs_set_path_blocking(path);
775
Dulshani Gunawardhanad9b0d9b2013-10-31 10:32:18 +0530776 keys = kmalloc_array(nitems, sizeof(struct btrfs_key), GFP_NOFS);
Miao Xie16cdcec2011-04-22 18:12:22 +0800777 if (!keys) {
778 ret = -ENOMEM;
779 goto out;
780 }
781
Dulshani Gunawardhanad9b0d9b2013-10-31 10:32:18 +0530782 data_size = kmalloc_array(nitems, sizeof(u32), GFP_NOFS);
Miao Xie16cdcec2011-04-22 18:12:22 +0800783 if (!data_size) {
784 ret = -ENOMEM;
785 goto error;
786 }
787
788 /* get keys of all the delayed items */
789 i = 0;
790 list_for_each_entry(next, &head, tree_list) {
791 keys[i] = next->key;
792 data_size[i] = next->data_len;
793 i++;
794 }
795
796 /* reset all the locked nodes in the patch to spinning locks. */
Chris Masonbd681512011-07-16 15:23:14 -0400797 btrfs_clear_path_blocking(path, NULL, 0);
Miao Xie16cdcec2011-04-22 18:12:22 +0800798
799 /* insert the keys of the items */
Tsutomu Itohafe5fea2013-04-16 05:18:22 +0000800 setup_items_for_insert(root, path, keys, data_size,
Jeff Mahoney143bede2012-03-01 14:56:26 +0100801 total_data_size, total_size, nitems);
Miao Xie16cdcec2011-04-22 18:12:22 +0800802
803 /* insert the dir index items */
804 slot = path->slots[0];
805 list_for_each_entry_safe(curr, next, &head, tree_list) {
806 data_ptr = btrfs_item_ptr(leaf, slot, char);
807 write_extent_buffer(leaf, &curr->data,
808 (unsigned long)data_ptr,
809 curr->data_len);
810 slot++;
811
812 btrfs_delayed_item_release_metadata(root, curr);
813
814 list_del(&curr->tree_list);
815 btrfs_release_delayed_item(curr);
816 }
817
818error:
819 kfree(data_size);
820 kfree(keys);
821out:
822 return ret;
823}
824
825/*
826 * This helper can just do simple insertion that needn't extend item for new
827 * data, such as directory name index insertion, inode insertion.
828 */
829static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans,
830 struct btrfs_root *root,
831 struct btrfs_path *path,
832 struct btrfs_delayed_item *delayed_item)
833{
834 struct extent_buffer *leaf;
Miao Xie16cdcec2011-04-22 18:12:22 +0800835 char *ptr;
836 int ret;
837
838 ret = btrfs_insert_empty_item(trans, root, path, &delayed_item->key,
839 delayed_item->data_len);
840 if (ret < 0 && ret != -EEXIST)
841 return ret;
842
843 leaf = path->nodes[0];
844
Miao Xie16cdcec2011-04-22 18:12:22 +0800845 ptr = btrfs_item_ptr(leaf, path->slots[0], char);
846
847 write_extent_buffer(leaf, delayed_item->data, (unsigned long)ptr,
848 delayed_item->data_len);
849 btrfs_mark_buffer_dirty(leaf);
850
851 btrfs_delayed_item_release_metadata(root, delayed_item);
852 return 0;
853}
854
855/*
856 * we insert an item first, then if there are some continuous items, we try
857 * to insert those items into the same leaf.
858 */
859static int btrfs_insert_delayed_items(struct btrfs_trans_handle *trans,
860 struct btrfs_path *path,
861 struct btrfs_root *root,
862 struct btrfs_delayed_node *node)
863{
864 struct btrfs_delayed_item *curr, *prev;
865 int ret = 0;
866
867do_again:
868 mutex_lock(&node->mutex);
869 curr = __btrfs_first_delayed_insertion_item(node);
870 if (!curr)
871 goto insert_end;
872
873 ret = btrfs_insert_delayed_item(trans, root, path, curr);
874 if (ret < 0) {
Chris Mason945d8962011-05-22 12:33:42 -0400875 btrfs_release_path(path);
Miao Xie16cdcec2011-04-22 18:12:22 +0800876 goto insert_end;
877 }
878
879 prev = curr;
880 curr = __btrfs_next_delayed_item(prev);
881 if (curr && btrfs_is_continuous_delayed_item(prev, curr)) {
882 /* insert the continuous items into the same leaf */
883 path->slots[0]++;
Tsutomu Itohafe5fea2013-04-16 05:18:22 +0000884 btrfs_batch_insert_items(root, path, curr);
Miao Xie16cdcec2011-04-22 18:12:22 +0800885 }
886 btrfs_release_delayed_item(prev);
887 btrfs_mark_buffer_dirty(path->nodes[0]);
888
Chris Mason945d8962011-05-22 12:33:42 -0400889 btrfs_release_path(path);
Miao Xie16cdcec2011-04-22 18:12:22 +0800890 mutex_unlock(&node->mutex);
891 goto do_again;
892
893insert_end:
894 mutex_unlock(&node->mutex);
895 return ret;
896}
897
898static int btrfs_batch_delete_items(struct btrfs_trans_handle *trans,
899 struct btrfs_root *root,
900 struct btrfs_path *path,
901 struct btrfs_delayed_item *item)
902{
903 struct btrfs_delayed_item *curr, *next;
904 struct extent_buffer *leaf;
905 struct btrfs_key key;
906 struct list_head head;
907 int nitems, i, last_item;
908 int ret = 0;
909
910 BUG_ON(!path->nodes[0]);
911
912 leaf = path->nodes[0];
913
914 i = path->slots[0];
915 last_item = btrfs_header_nritems(leaf) - 1;
916 if (i > last_item)
917 return -ENOENT; /* FIXME: Is errno suitable? */
918
919 next = item;
920 INIT_LIST_HEAD(&head);
921 btrfs_item_key_to_cpu(leaf, &key, i);
922 nitems = 0;
923 /*
924 * count the number of the dir index items that we can delete in batch
925 */
926 while (btrfs_comp_cpu_keys(&next->key, &key) == 0) {
927 list_add_tail(&next->tree_list, &head);
928 nitems++;
929
930 curr = next;
931 next = __btrfs_next_delayed_item(curr);
932 if (!next)
933 break;
934
935 if (!btrfs_is_continuous_delayed_item(curr, next))
936 break;
937
938 i++;
939 if (i > last_item)
940 break;
941 btrfs_item_key_to_cpu(leaf, &key, i);
942 }
943
944 if (!nitems)
945 return 0;
946
947 ret = btrfs_del_items(trans, root, path, path->slots[0], nitems);
948 if (ret)
949 goto out;
950
951 list_for_each_entry_safe(curr, next, &head, tree_list) {
952 btrfs_delayed_item_release_metadata(root, curr);
953 list_del(&curr->tree_list);
954 btrfs_release_delayed_item(curr);
955 }
956
957out:
958 return ret;
959}
960
961static int btrfs_delete_delayed_items(struct btrfs_trans_handle *trans,
962 struct btrfs_path *path,
963 struct btrfs_root *root,
964 struct btrfs_delayed_node *node)
965{
966 struct btrfs_delayed_item *curr, *prev;
967 int ret = 0;
968
969do_again:
970 mutex_lock(&node->mutex);
971 curr = __btrfs_first_delayed_deletion_item(node);
972 if (!curr)
973 goto delete_fail;
974
975 ret = btrfs_search_slot(trans, root, &curr->key, path, -1, 1);
976 if (ret < 0)
977 goto delete_fail;
978 else if (ret > 0) {
979 /*
980 * can't find the item which the node points to, so this node
981 * is invalid, just drop it.
982 */
983 prev = curr;
984 curr = __btrfs_next_delayed_item(prev);
985 btrfs_release_delayed_item(prev);
986 ret = 0;
Chris Mason945d8962011-05-22 12:33:42 -0400987 btrfs_release_path(path);
Fengguang Wu62095262012-08-04 01:45:02 -0600988 if (curr) {
989 mutex_unlock(&node->mutex);
Miao Xie16cdcec2011-04-22 18:12:22 +0800990 goto do_again;
Fengguang Wu62095262012-08-04 01:45:02 -0600991 } else
Miao Xie16cdcec2011-04-22 18:12:22 +0800992 goto delete_fail;
993 }
994
995 btrfs_batch_delete_items(trans, root, path, curr);
Chris Mason945d8962011-05-22 12:33:42 -0400996 btrfs_release_path(path);
Miao Xie16cdcec2011-04-22 18:12:22 +0800997 mutex_unlock(&node->mutex);
998 goto do_again;
999
1000delete_fail:
Chris Mason945d8962011-05-22 12:33:42 -04001001 btrfs_release_path(path);
Miao Xie16cdcec2011-04-22 18:12:22 +08001002 mutex_unlock(&node->mutex);
1003 return ret;
1004}
1005
1006static void btrfs_release_delayed_inode(struct btrfs_delayed_node *delayed_node)
1007{
1008 struct btrfs_delayed_root *delayed_root;
1009
Miao Xie7cf35d92013-12-26 13:07:05 +08001010 if (delayed_node &&
1011 test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
Miao Xie16cdcec2011-04-22 18:12:22 +08001012 BUG_ON(!delayed_node->root);
Miao Xie7cf35d92013-12-26 13:07:05 +08001013 clear_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
Miao Xie16cdcec2011-04-22 18:12:22 +08001014 delayed_node->count--;
1015
1016 delayed_root = delayed_node->root->fs_info->delayed_root;
Chris Masonde3cb942013-03-04 17:13:31 -05001017 finish_one_item(delayed_root);
Miao Xie16cdcec2011-04-22 18:12:22 +08001018 }
1019}
1020
Miao Xie67de1172013-12-26 13:07:06 +08001021static void btrfs_release_delayed_iref(struct btrfs_delayed_node *delayed_node)
1022{
1023 struct btrfs_delayed_root *delayed_root;
1024
1025 ASSERT(delayed_node->root);
1026 clear_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
1027 delayed_node->count--;
1028
1029 delayed_root = delayed_node->root->fs_info->delayed_root;
1030 finish_one_item(delayed_root);
1031}
1032
Miao Xie0e8c36a2012-12-19 06:59:51 +00001033static int __btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1034 struct btrfs_root *root,
1035 struct btrfs_path *path,
1036 struct btrfs_delayed_node *node)
Miao Xie16cdcec2011-04-22 18:12:22 +08001037{
1038 struct btrfs_key key;
1039 struct btrfs_inode_item *inode_item;
1040 struct extent_buffer *leaf;
Miao Xie67de1172013-12-26 13:07:06 +08001041 int mod;
Miao Xie16cdcec2011-04-22 18:12:22 +08001042 int ret;
1043
Miao Xie16cdcec2011-04-22 18:12:22 +08001044 key.objectid = node->inode_id;
David Sterba962a2982014-06-04 18:41:45 +02001045 key.type = BTRFS_INODE_ITEM_KEY;
Miao Xie16cdcec2011-04-22 18:12:22 +08001046 key.offset = 0;
Miao Xie0e8c36a2012-12-19 06:59:51 +00001047
Miao Xie67de1172013-12-26 13:07:06 +08001048 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1049 mod = -1;
1050 else
1051 mod = 1;
1052
1053 ret = btrfs_lookup_inode(trans, root, path, &key, mod);
Miao Xie16cdcec2011-04-22 18:12:22 +08001054 if (ret > 0) {
Chris Mason945d8962011-05-22 12:33:42 -04001055 btrfs_release_path(path);
Miao Xie16cdcec2011-04-22 18:12:22 +08001056 return -ENOENT;
1057 } else if (ret < 0) {
Miao Xie16cdcec2011-04-22 18:12:22 +08001058 return ret;
1059 }
1060
Miao Xie16cdcec2011-04-22 18:12:22 +08001061 leaf = path->nodes[0];
1062 inode_item = btrfs_item_ptr(leaf, path->slots[0],
1063 struct btrfs_inode_item);
1064 write_extent_buffer(leaf, &node->inode_item, (unsigned long)inode_item,
1065 sizeof(struct btrfs_inode_item));
1066 btrfs_mark_buffer_dirty(leaf);
Miao Xie16cdcec2011-04-22 18:12:22 +08001067
Miao Xie67de1172013-12-26 13:07:06 +08001068 if (!test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1069 goto no_iref;
1070
1071 path->slots[0]++;
1072 if (path->slots[0] >= btrfs_header_nritems(leaf))
1073 goto search;
1074again:
1075 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1076 if (key.objectid != node->inode_id)
1077 goto out;
1078
1079 if (key.type != BTRFS_INODE_REF_KEY &&
1080 key.type != BTRFS_INODE_EXTREF_KEY)
1081 goto out;
1082
1083 /*
1084 * Delayed iref deletion is for the inode who has only one link,
1085 * so there is only one iref. The case that several irefs are
1086 * in the same item doesn't exist.
1087 */
1088 btrfs_del_item(trans, root, path);
1089out:
1090 btrfs_release_delayed_iref(node);
1091no_iref:
1092 btrfs_release_path(path);
1093err_out:
Miao Xie16cdcec2011-04-22 18:12:22 +08001094 btrfs_delayed_inode_release_metadata(root, node);
1095 btrfs_release_delayed_inode(node);
Miao Xie16cdcec2011-04-22 18:12:22 +08001096
Miao Xie67de1172013-12-26 13:07:06 +08001097 return ret;
1098
1099search:
1100 btrfs_release_path(path);
1101
David Sterba962a2982014-06-04 18:41:45 +02001102 key.type = BTRFS_INODE_EXTREF_KEY;
Miao Xie67de1172013-12-26 13:07:06 +08001103 key.offset = -1;
1104 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1105 if (ret < 0)
1106 goto err_out;
1107 ASSERT(ret);
1108
1109 ret = 0;
1110 leaf = path->nodes[0];
1111 path->slots[0]--;
1112 goto again;
Miao Xie16cdcec2011-04-22 18:12:22 +08001113}
1114
Miao Xie0e8c36a2012-12-19 06:59:51 +00001115static inline int btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1116 struct btrfs_root *root,
1117 struct btrfs_path *path,
1118 struct btrfs_delayed_node *node)
1119{
1120 int ret;
1121
1122 mutex_lock(&node->mutex);
Miao Xie7cf35d92013-12-26 13:07:05 +08001123 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &node->flags)) {
Miao Xie0e8c36a2012-12-19 06:59:51 +00001124 mutex_unlock(&node->mutex);
1125 return 0;
1126 }
1127
1128 ret = __btrfs_update_delayed_inode(trans, root, path, node);
1129 mutex_unlock(&node->mutex);
1130 return ret;
1131}
1132
Miao Xie4ea41ce2012-12-19 06:59:03 +00001133static inline int
1134__btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1135 struct btrfs_path *path,
1136 struct btrfs_delayed_node *node)
1137{
1138 int ret;
1139
1140 ret = btrfs_insert_delayed_items(trans, path, node->root, node);
1141 if (ret)
1142 return ret;
1143
1144 ret = btrfs_delete_delayed_items(trans, path, node->root, node);
1145 if (ret)
1146 return ret;
1147
1148 ret = btrfs_update_delayed_inode(trans, node->root, path, node);
1149 return ret;
1150}
1151
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001152/*
1153 * Called when committing the transaction.
1154 * Returns 0 on success.
1155 * Returns < 0 on error and returns with an aborted transaction with any
1156 * outstanding delayed items cleaned up.
1157 */
Josef Bacik96c3f432012-06-21 14:05:49 -04001158static int __btrfs_run_delayed_items(struct btrfs_trans_handle *trans,
1159 struct btrfs_root *root, int nr)
Miao Xie16cdcec2011-04-22 18:12:22 +08001160{
1161 struct btrfs_delayed_root *delayed_root;
1162 struct btrfs_delayed_node *curr_node, *prev_node;
1163 struct btrfs_path *path;
Miao Xie19fd2942011-06-15 10:47:30 +00001164 struct btrfs_block_rsv *block_rsv;
Miao Xie16cdcec2011-04-22 18:12:22 +08001165 int ret = 0;
Josef Bacik96c3f432012-06-21 14:05:49 -04001166 bool count = (nr > 0);
Miao Xie16cdcec2011-04-22 18:12:22 +08001167
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001168 if (trans->aborted)
1169 return -EIO;
1170
Miao Xie16cdcec2011-04-22 18:12:22 +08001171 path = btrfs_alloc_path();
1172 if (!path)
1173 return -ENOMEM;
1174 path->leave_spinning = 1;
1175
Miao Xie19fd2942011-06-15 10:47:30 +00001176 block_rsv = trans->block_rsv;
Josef Bacik6d668dd2011-11-03 22:54:25 -04001177 trans->block_rsv = &root->fs_info->delayed_block_rsv;
Miao Xie19fd2942011-06-15 10:47:30 +00001178
Miao Xie16cdcec2011-04-22 18:12:22 +08001179 delayed_root = btrfs_get_delayed_root(root);
1180
1181 curr_node = btrfs_first_delayed_node(delayed_root);
Josef Bacik96c3f432012-06-21 14:05:49 -04001182 while (curr_node && (!count || (count && nr--))) {
Miao Xie4ea41ce2012-12-19 06:59:03 +00001183 ret = __btrfs_commit_inode_delayed_items(trans, path,
1184 curr_node);
Miao Xie16cdcec2011-04-22 18:12:22 +08001185 if (ret) {
1186 btrfs_release_delayed_node(curr_node);
Josef Bacik96c3f432012-06-21 14:05:49 -04001187 curr_node = NULL;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001188 btrfs_abort_transaction(trans, root, ret);
Miao Xie16cdcec2011-04-22 18:12:22 +08001189 break;
1190 }
1191
1192 prev_node = curr_node;
1193 curr_node = btrfs_next_delayed_node(curr_node);
1194 btrfs_release_delayed_node(prev_node);
1195 }
1196
Josef Bacik96c3f432012-06-21 14:05:49 -04001197 if (curr_node)
1198 btrfs_release_delayed_node(curr_node);
Miao Xie16cdcec2011-04-22 18:12:22 +08001199 btrfs_free_path(path);
Miao Xie19fd2942011-06-15 10:47:30 +00001200 trans->block_rsv = block_rsv;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001201
Miao Xie16cdcec2011-04-22 18:12:22 +08001202 return ret;
1203}
1204
Josef Bacik96c3f432012-06-21 14:05:49 -04001205int btrfs_run_delayed_items(struct btrfs_trans_handle *trans,
1206 struct btrfs_root *root)
1207{
1208 return __btrfs_run_delayed_items(trans, root, -1);
1209}
1210
1211int btrfs_run_delayed_items_nr(struct btrfs_trans_handle *trans,
1212 struct btrfs_root *root, int nr)
1213{
1214 return __btrfs_run_delayed_items(trans, root, nr);
1215}
1216
Miao Xie16cdcec2011-04-22 18:12:22 +08001217int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1218 struct inode *inode)
1219{
1220 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
Miao Xie4ea41ce2012-12-19 06:59:03 +00001221 struct btrfs_path *path;
1222 struct btrfs_block_rsv *block_rsv;
Miao Xie16cdcec2011-04-22 18:12:22 +08001223 int ret;
1224
1225 if (!delayed_node)
1226 return 0;
1227
1228 mutex_lock(&delayed_node->mutex);
1229 if (!delayed_node->count) {
1230 mutex_unlock(&delayed_node->mutex);
1231 btrfs_release_delayed_node(delayed_node);
1232 return 0;
1233 }
1234 mutex_unlock(&delayed_node->mutex);
1235
Miao Xie4ea41ce2012-12-19 06:59:03 +00001236 path = btrfs_alloc_path();
Filipe David Borba Manana3c77bd92013-10-12 20:32:59 +01001237 if (!path) {
1238 btrfs_release_delayed_node(delayed_node);
Miao Xie4ea41ce2012-12-19 06:59:03 +00001239 return -ENOMEM;
Filipe David Borba Manana3c77bd92013-10-12 20:32:59 +01001240 }
Miao Xie4ea41ce2012-12-19 06:59:03 +00001241 path->leave_spinning = 1;
1242
1243 block_rsv = trans->block_rsv;
1244 trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv;
1245
1246 ret = __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1247
Miao Xie16cdcec2011-04-22 18:12:22 +08001248 btrfs_release_delayed_node(delayed_node);
Miao Xie4ea41ce2012-12-19 06:59:03 +00001249 btrfs_free_path(path);
1250 trans->block_rsv = block_rsv;
1251
Miao Xie16cdcec2011-04-22 18:12:22 +08001252 return ret;
1253}
1254
Miao Xie0e8c36a2012-12-19 06:59:51 +00001255int btrfs_commit_inode_delayed_inode(struct inode *inode)
1256{
1257 struct btrfs_trans_handle *trans;
1258 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1259 struct btrfs_path *path;
1260 struct btrfs_block_rsv *block_rsv;
1261 int ret;
1262
1263 if (!delayed_node)
1264 return 0;
1265
1266 mutex_lock(&delayed_node->mutex);
Miao Xie7cf35d92013-12-26 13:07:05 +08001267 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
Miao Xie0e8c36a2012-12-19 06:59:51 +00001268 mutex_unlock(&delayed_node->mutex);
1269 btrfs_release_delayed_node(delayed_node);
1270 return 0;
1271 }
1272 mutex_unlock(&delayed_node->mutex);
1273
1274 trans = btrfs_join_transaction(delayed_node->root);
1275 if (IS_ERR(trans)) {
1276 ret = PTR_ERR(trans);
1277 goto out;
1278 }
1279
1280 path = btrfs_alloc_path();
1281 if (!path) {
1282 ret = -ENOMEM;
1283 goto trans_out;
1284 }
1285 path->leave_spinning = 1;
1286
1287 block_rsv = trans->block_rsv;
1288 trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv;
1289
1290 mutex_lock(&delayed_node->mutex);
Miao Xie7cf35d92013-12-26 13:07:05 +08001291 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags))
Miao Xie0e8c36a2012-12-19 06:59:51 +00001292 ret = __btrfs_update_delayed_inode(trans, delayed_node->root,
1293 path, delayed_node);
1294 else
1295 ret = 0;
1296 mutex_unlock(&delayed_node->mutex);
1297
1298 btrfs_free_path(path);
1299 trans->block_rsv = block_rsv;
1300trans_out:
1301 btrfs_end_transaction(trans, delayed_node->root);
1302 btrfs_btree_balance_dirty(delayed_node->root);
1303out:
1304 btrfs_release_delayed_node(delayed_node);
1305
1306 return ret;
1307}
1308
Miao Xie16cdcec2011-04-22 18:12:22 +08001309void btrfs_remove_delayed_node(struct inode *inode)
1310{
1311 struct btrfs_delayed_node *delayed_node;
1312
1313 delayed_node = ACCESS_ONCE(BTRFS_I(inode)->delayed_node);
1314 if (!delayed_node)
1315 return;
1316
1317 BTRFS_I(inode)->delayed_node = NULL;
1318 btrfs_release_delayed_node(delayed_node);
1319}
1320
Chris Masonde3cb942013-03-04 17:13:31 -05001321struct btrfs_async_delayed_work {
1322 struct btrfs_delayed_root *delayed_root;
1323 int nr;
Qu Wenruod458b052014-02-28 10:46:19 +08001324 struct btrfs_work work;
Miao Xie16cdcec2011-04-22 18:12:22 +08001325};
1326
Qu Wenruod458b052014-02-28 10:46:19 +08001327static void btrfs_async_run_delayed_root(struct btrfs_work *work)
Miao Xie16cdcec2011-04-22 18:12:22 +08001328{
Chris Masonde3cb942013-03-04 17:13:31 -05001329 struct btrfs_async_delayed_work *async_work;
1330 struct btrfs_delayed_root *delayed_root;
Miao Xie16cdcec2011-04-22 18:12:22 +08001331 struct btrfs_trans_handle *trans;
1332 struct btrfs_path *path;
1333 struct btrfs_delayed_node *delayed_node = NULL;
1334 struct btrfs_root *root;
Miao Xie19fd2942011-06-15 10:47:30 +00001335 struct btrfs_block_rsv *block_rsv;
Chris Masonde3cb942013-03-04 17:13:31 -05001336 int total_done = 0;
Miao Xie16cdcec2011-04-22 18:12:22 +08001337
Chris Masonde3cb942013-03-04 17:13:31 -05001338 async_work = container_of(work, struct btrfs_async_delayed_work, work);
1339 delayed_root = async_work->delayed_root;
Miao Xie16cdcec2011-04-22 18:12:22 +08001340
1341 path = btrfs_alloc_path();
1342 if (!path)
1343 goto out;
Miao Xie16cdcec2011-04-22 18:12:22 +08001344
Chris Masonde3cb942013-03-04 17:13:31 -05001345again:
1346 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND / 2)
1347 goto free_path;
1348
1349 delayed_node = btrfs_first_prepared_delayed_node(delayed_root);
1350 if (!delayed_node)
1351 goto free_path;
1352
1353 path->leave_spinning = 1;
Miao Xie16cdcec2011-04-22 18:12:22 +08001354 root = delayed_node->root;
1355
Chris Masonff5714c2011-05-28 07:00:39 -04001356 trans = btrfs_join_transaction(root);
Miao Xie16cdcec2011-04-22 18:12:22 +08001357 if (IS_ERR(trans))
Chris Masonde3cb942013-03-04 17:13:31 -05001358 goto release_path;
Miao Xie16cdcec2011-04-22 18:12:22 +08001359
Miao Xie19fd2942011-06-15 10:47:30 +00001360 block_rsv = trans->block_rsv;
Josef Bacik6d668dd2011-11-03 22:54:25 -04001361 trans->block_rsv = &root->fs_info->delayed_block_rsv;
Miao Xie19fd2942011-06-15 10:47:30 +00001362
Miao Xie4ea41ce2012-12-19 06:59:03 +00001363 __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
Miao Xie16cdcec2011-04-22 18:12:22 +08001364
Miao Xie19fd2942011-06-15 10:47:30 +00001365 trans->block_rsv = block_rsv;
Miao Xiea56dbd82013-12-26 13:07:04 +08001366 btrfs_end_transaction(trans, root);
Liu Bob53d3f52012-11-14 14:34:34 +00001367 btrfs_btree_balance_dirty_nodelay(root);
Chris Masonde3cb942013-03-04 17:13:31 -05001368
1369release_path:
1370 btrfs_release_path(path);
1371 total_done++;
1372
1373 btrfs_release_prepared_delayed_node(delayed_node);
1374 if (async_work->nr == 0 || total_done < async_work->nr)
1375 goto again;
1376
Miao Xie16cdcec2011-04-22 18:12:22 +08001377free_path:
1378 btrfs_free_path(path);
1379out:
Chris Masonde3cb942013-03-04 17:13:31 -05001380 wake_up(&delayed_root->wait);
1381 kfree(async_work);
Miao Xie16cdcec2011-04-22 18:12:22 +08001382}
1383
Miao Xie16cdcec2011-04-22 18:12:22 +08001384
Chris Masonde3cb942013-03-04 17:13:31 -05001385static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root,
Daniel Dresslera585e942014-11-17 22:05:02 +09001386 struct btrfs_fs_info *fs_info, int nr)
Chris Masonde3cb942013-03-04 17:13:31 -05001387{
1388 struct btrfs_async_delayed_work *async_work;
1389
1390 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
Miao Xie16cdcec2011-04-22 18:12:22 +08001391 return 0;
1392
Chris Masonde3cb942013-03-04 17:13:31 -05001393 async_work = kmalloc(sizeof(*async_work), GFP_NOFS);
1394 if (!async_work)
Miao Xie16cdcec2011-04-22 18:12:22 +08001395 return -ENOMEM;
Miao Xie16cdcec2011-04-22 18:12:22 +08001396
Chris Masonde3cb942013-03-04 17:13:31 -05001397 async_work->delayed_root = delayed_root;
Liu Bo9e0af232014-08-15 23:36:53 +08001398 btrfs_init_work(&async_work->work, btrfs_delayed_meta_helper,
1399 btrfs_async_run_delayed_root, NULL, NULL);
Chris Masonde3cb942013-03-04 17:13:31 -05001400 async_work->nr = nr;
Miao Xie16cdcec2011-04-22 18:12:22 +08001401
Daniel Dresslera585e942014-11-17 22:05:02 +09001402 btrfs_queue_work(fs_info->delayed_workers, &async_work->work);
Miao Xie16cdcec2011-04-22 18:12:22 +08001403 return 0;
1404}
1405
Chris Masone9993762011-06-17 16:14:09 -04001406void btrfs_assert_delayed_root_empty(struct btrfs_root *root)
1407{
1408 struct btrfs_delayed_root *delayed_root;
1409 delayed_root = btrfs_get_delayed_root(root);
1410 WARN_ON(btrfs_first_delayed_node(delayed_root));
1411}
1412
Miao Xie03538082013-12-26 13:07:03 +08001413static int could_end_wait(struct btrfs_delayed_root *delayed_root, int seq)
Chris Masonde3cb942013-03-04 17:13:31 -05001414{
1415 int val = atomic_read(&delayed_root->items_seq);
1416
Miao Xie03538082013-12-26 13:07:03 +08001417 if (val < seq || val >= seq + BTRFS_DELAYED_BATCH)
Chris Masonde3cb942013-03-04 17:13:31 -05001418 return 1;
Miao Xie03538082013-12-26 13:07:03 +08001419
1420 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1421 return 1;
1422
Chris Masonde3cb942013-03-04 17:13:31 -05001423 return 0;
1424}
1425
Miao Xie16cdcec2011-04-22 18:12:22 +08001426void btrfs_balance_delayed_items(struct btrfs_root *root)
1427{
1428 struct btrfs_delayed_root *delayed_root;
Daniel Dresslera585e942014-11-17 22:05:02 +09001429 struct btrfs_fs_info *fs_info = root->fs_info;
Miao Xie16cdcec2011-04-22 18:12:22 +08001430
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) {
Miao Xie03538082013-12-26 13:07:03 +08001437 int seq;
Miao Xie16cdcec2011-04-22 18:12:22 +08001438 int ret;
Miao Xie03538082013-12-26 13:07:03 +08001439
1440 seq = atomic_read(&delayed_root->items_seq);
Chris Masonde3cb942013-03-04 17:13:31 -05001441
Daniel Dresslera585e942014-11-17 22:05:02 +09001442 ret = btrfs_wq_run_delayed_node(delayed_root, fs_info, 0);
Miao Xie16cdcec2011-04-22 18:12:22 +08001443 if (ret)
1444 return;
1445
Miao Xie03538082013-12-26 13:07:03 +08001446 wait_event_interruptible(delayed_root->wait,
1447 could_end_wait(delayed_root, seq));
Miao Xie4dd466d2013-12-26 13:07:02 +08001448 return;
Miao Xie16cdcec2011-04-22 18:12:22 +08001449 }
1450
Daniel Dresslera585e942014-11-17 22:05:02 +09001451 btrfs_wq_run_delayed_node(delayed_root, fs_info, BTRFS_DELAYED_BATCH);
Miao Xie16cdcec2011-04-22 18:12:22 +08001452}
1453
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001454/* Will return 0 or -ENOMEM */
Miao Xie16cdcec2011-04-22 18:12:22 +08001455int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans,
1456 struct btrfs_root *root, const char *name,
1457 int name_len, struct inode *dir,
1458 struct btrfs_disk_key *disk_key, u8 type,
1459 u64 index)
1460{
1461 struct btrfs_delayed_node *delayed_node;
1462 struct btrfs_delayed_item *delayed_item;
1463 struct btrfs_dir_item *dir_item;
1464 int ret;
1465
1466 delayed_node = btrfs_get_or_create_delayed_node(dir);
1467 if (IS_ERR(delayed_node))
1468 return PTR_ERR(delayed_node);
1469
1470 delayed_item = btrfs_alloc_delayed_item(sizeof(*dir_item) + name_len);
1471 if (!delayed_item) {
1472 ret = -ENOMEM;
1473 goto release_node;
1474 }
1475
Chris Mason0d0ca302011-05-22 07:11:22 -04001476 delayed_item->key.objectid = btrfs_ino(dir);
David Sterba962a2982014-06-04 18:41:45 +02001477 delayed_item->key.type = BTRFS_DIR_INDEX_KEY;
Miao Xie16cdcec2011-04-22 18:12:22 +08001478 delayed_item->key.offset = index;
1479
1480 dir_item = (struct btrfs_dir_item *)delayed_item->data;
1481 dir_item->location = *disk_key;
Qu Wenruo3cae2102013-07-16 11:19:18 +08001482 btrfs_set_stack_dir_transid(dir_item, trans->transid);
1483 btrfs_set_stack_dir_data_len(dir_item, 0);
1484 btrfs_set_stack_dir_name_len(dir_item, name_len);
1485 btrfs_set_stack_dir_type(dir_item, type);
Miao Xie16cdcec2011-04-22 18:12:22 +08001486 memcpy((char *)(dir_item + 1), name, name_len);
1487
Josef Bacik8c2a3ca2012-01-10 10:31:31 -05001488 ret = btrfs_delayed_item_reserve_metadata(trans, root, delayed_item);
1489 /*
1490 * we have reserved enough space when we start a new transaction,
1491 * so reserving metadata failure is impossible
1492 */
1493 BUG_ON(ret);
1494
1495
Miao Xie16cdcec2011-04-22 18:12:22 +08001496 mutex_lock(&delayed_node->mutex);
1497 ret = __btrfs_add_delayed_insertion_item(delayed_node, delayed_item);
1498 if (unlikely(ret)) {
Frank Holtonefe120a2013-12-20 11:37:06 -05001499 btrfs_err(root->fs_info, "err add delayed dir index item(name: %.*s) "
Filipe David Borba Mananabdab49d2013-08-20 17:51:48 +01001500 "into the insertion tree of the delayed node"
Frank Holtonefe120a2013-12-20 11:37:06 -05001501 "(root id: %llu, inode id: %llu, errno: %d)",
Filipe David Borba Mananabdab49d2013-08-20 17:51:48 +01001502 name_len, name, delayed_node->root->objectid,
Geert Uytterhoevenc1c9ff72013-08-20 13:20:07 +02001503 delayed_node->inode_id, ret);
Miao Xie16cdcec2011-04-22 18:12:22 +08001504 BUG();
1505 }
1506 mutex_unlock(&delayed_node->mutex);
1507
1508release_node:
1509 btrfs_release_delayed_node(delayed_node);
1510 return ret;
1511}
1512
1513static int btrfs_delete_delayed_insertion_item(struct btrfs_root *root,
1514 struct btrfs_delayed_node *node,
1515 struct btrfs_key *key)
1516{
1517 struct btrfs_delayed_item *item;
1518
1519 mutex_lock(&node->mutex);
1520 item = __btrfs_lookup_delayed_insertion_item(node, key);
1521 if (!item) {
1522 mutex_unlock(&node->mutex);
1523 return 1;
1524 }
1525
1526 btrfs_delayed_item_release_metadata(root, item);
1527 btrfs_release_delayed_item(item);
1528 mutex_unlock(&node->mutex);
1529 return 0;
1530}
1531
1532int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans,
1533 struct btrfs_root *root, struct inode *dir,
1534 u64 index)
1535{
1536 struct btrfs_delayed_node *node;
1537 struct btrfs_delayed_item *item;
1538 struct btrfs_key item_key;
1539 int ret;
1540
1541 node = btrfs_get_or_create_delayed_node(dir);
1542 if (IS_ERR(node))
1543 return PTR_ERR(node);
1544
Chris Mason0d0ca302011-05-22 07:11:22 -04001545 item_key.objectid = btrfs_ino(dir);
David Sterba962a2982014-06-04 18:41:45 +02001546 item_key.type = BTRFS_DIR_INDEX_KEY;
Miao Xie16cdcec2011-04-22 18:12:22 +08001547 item_key.offset = index;
1548
1549 ret = btrfs_delete_delayed_insertion_item(root, node, &item_key);
1550 if (!ret)
1551 goto end;
1552
1553 item = btrfs_alloc_delayed_item(0);
1554 if (!item) {
1555 ret = -ENOMEM;
1556 goto end;
1557 }
1558
1559 item->key = item_key;
1560
1561 ret = btrfs_delayed_item_reserve_metadata(trans, root, item);
1562 /*
1563 * we have reserved enough space when we start a new transaction,
1564 * so reserving metadata failure is impossible.
1565 */
1566 BUG_ON(ret);
1567
1568 mutex_lock(&node->mutex);
1569 ret = __btrfs_add_delayed_deletion_item(node, item);
1570 if (unlikely(ret)) {
Frank Holtonefe120a2013-12-20 11:37:06 -05001571 btrfs_err(root->fs_info, "err add delayed dir index item(index: %llu) "
Miao Xie16cdcec2011-04-22 18:12:22 +08001572 "into the deletion tree of the delayed node"
Frank Holtonefe120a2013-12-20 11:37:06 -05001573 "(root id: %llu, inode id: %llu, errno: %d)",
Geert Uytterhoevenc1c9ff72013-08-20 13:20:07 +02001574 index, node->root->objectid, node->inode_id,
Miao Xie16cdcec2011-04-22 18:12:22 +08001575 ret);
1576 BUG();
1577 }
1578 mutex_unlock(&node->mutex);
1579end:
1580 btrfs_release_delayed_node(node);
1581 return ret;
1582}
1583
1584int btrfs_inode_delayed_dir_index_count(struct inode *inode)
1585{
Miao Xie2f7e33d2011-06-23 07:27:13 +00001586 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
Miao Xie16cdcec2011-04-22 18:12:22 +08001587
1588 if (!delayed_node)
1589 return -ENOENT;
1590
1591 /*
1592 * Since we have held i_mutex of this directory, it is impossible that
1593 * a new directory index is added into the delayed node and index_cnt
1594 * is updated now. So we needn't lock the delayed node.
1595 */
Miao Xie2f7e33d2011-06-23 07:27:13 +00001596 if (!delayed_node->index_cnt) {
1597 btrfs_release_delayed_node(delayed_node);
Miao Xie16cdcec2011-04-22 18:12:22 +08001598 return -EINVAL;
Miao Xie2f7e33d2011-06-23 07:27:13 +00001599 }
Miao Xie16cdcec2011-04-22 18:12:22 +08001600
1601 BTRFS_I(inode)->index_cnt = delayed_node->index_cnt;
Miao Xie2f7e33d2011-06-23 07:27:13 +00001602 btrfs_release_delayed_node(delayed_node);
1603 return 0;
Miao Xie16cdcec2011-04-22 18:12:22 +08001604}
1605
1606void btrfs_get_delayed_items(struct inode *inode, struct list_head *ins_list,
1607 struct list_head *del_list)
1608{
1609 struct btrfs_delayed_node *delayed_node;
1610 struct btrfs_delayed_item *item;
1611
1612 delayed_node = btrfs_get_delayed_node(inode);
1613 if (!delayed_node)
1614 return;
1615
1616 mutex_lock(&delayed_node->mutex);
1617 item = __btrfs_first_delayed_insertion_item(delayed_node);
1618 while (item) {
1619 atomic_inc(&item->refs);
1620 list_add_tail(&item->readdir_list, ins_list);
1621 item = __btrfs_next_delayed_item(item);
1622 }
1623
1624 item = __btrfs_first_delayed_deletion_item(delayed_node);
1625 while (item) {
1626 atomic_inc(&item->refs);
1627 list_add_tail(&item->readdir_list, del_list);
1628 item = __btrfs_next_delayed_item(item);
1629 }
1630 mutex_unlock(&delayed_node->mutex);
1631 /*
1632 * This delayed node is still cached in the btrfs inode, so refs
1633 * must be > 1 now, and we needn't check it is going to be freed
1634 * or not.
1635 *
1636 * Besides that, this function is used to read dir, we do not
1637 * insert/delete delayed items in this period. So we also needn't
1638 * requeue or dequeue this delayed node.
1639 */
1640 atomic_dec(&delayed_node->refs);
1641}
1642
1643void btrfs_put_delayed_items(struct list_head *ins_list,
1644 struct list_head *del_list)
1645{
1646 struct btrfs_delayed_item *curr, *next;
1647
1648 list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1649 list_del(&curr->readdir_list);
1650 if (atomic_dec_and_test(&curr->refs))
1651 kfree(curr);
1652 }
1653
1654 list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1655 list_del(&curr->readdir_list);
1656 if (atomic_dec_and_test(&curr->refs))
1657 kfree(curr);
1658 }
1659}
1660
1661int btrfs_should_delete_dir_index(struct list_head *del_list,
1662 u64 index)
1663{
1664 struct btrfs_delayed_item *curr, *next;
1665 int ret;
1666
1667 if (list_empty(del_list))
1668 return 0;
1669
1670 list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1671 if (curr->key.offset > index)
1672 break;
1673
1674 list_del(&curr->readdir_list);
1675 ret = (curr->key.offset == index);
1676
1677 if (atomic_dec_and_test(&curr->refs))
1678 kfree(curr);
1679
1680 if (ret)
1681 return 1;
1682 else
1683 continue;
1684 }
1685 return 0;
1686}
1687
1688/*
1689 * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree
1690 *
1691 */
Al Viro9cdda8d2013-05-22 16:48:09 -04001692int btrfs_readdir_delayed_dir_index(struct dir_context *ctx,
Miao Xie16cdcec2011-04-22 18:12:22 +08001693 struct list_head *ins_list)
1694{
1695 struct btrfs_dir_item *di;
1696 struct btrfs_delayed_item *curr, *next;
1697 struct btrfs_key location;
1698 char *name;
1699 int name_len;
1700 int over = 0;
1701 unsigned char d_type;
1702
1703 if (list_empty(ins_list))
1704 return 0;
1705
1706 /*
1707 * Changing the data of the delayed item is impossible. So
1708 * we needn't lock them. And we have held i_mutex of the
1709 * directory, nobody can delete any directory indexes now.
1710 */
1711 list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1712 list_del(&curr->readdir_list);
1713
Al Viro9cdda8d2013-05-22 16:48:09 -04001714 if (curr->key.offset < ctx->pos) {
Miao Xie16cdcec2011-04-22 18:12:22 +08001715 if (atomic_dec_and_test(&curr->refs))
1716 kfree(curr);
1717 continue;
1718 }
1719
Al Viro9cdda8d2013-05-22 16:48:09 -04001720 ctx->pos = curr->key.offset;
Miao Xie16cdcec2011-04-22 18:12:22 +08001721
1722 di = (struct btrfs_dir_item *)curr->data;
1723 name = (char *)(di + 1);
Qu Wenruo3cae2102013-07-16 11:19:18 +08001724 name_len = btrfs_stack_dir_name_len(di);
Miao Xie16cdcec2011-04-22 18:12:22 +08001725
1726 d_type = btrfs_filetype_table[di->type];
1727 btrfs_disk_key_to_cpu(&location, &di->location);
1728
Al Viro9cdda8d2013-05-22 16:48:09 -04001729 over = !dir_emit(ctx, name, name_len,
Miao Xie16cdcec2011-04-22 18:12:22 +08001730 location.objectid, d_type);
1731
1732 if (atomic_dec_and_test(&curr->refs))
1733 kfree(curr);
1734
1735 if (over)
1736 return 1;
1737 }
1738 return 0;
1739}
1740
Miao Xie16cdcec2011-04-22 18:12:22 +08001741static void fill_stack_inode_item(struct btrfs_trans_handle *trans,
1742 struct btrfs_inode_item *inode_item,
1743 struct inode *inode)
1744{
Eric W. Biederman2f2f43d2012-02-10 11:05:07 -08001745 btrfs_set_stack_inode_uid(inode_item, i_uid_read(inode));
1746 btrfs_set_stack_inode_gid(inode_item, i_gid_read(inode));
Miao Xie16cdcec2011-04-22 18:12:22 +08001747 btrfs_set_stack_inode_size(inode_item, BTRFS_I(inode)->disk_i_size);
1748 btrfs_set_stack_inode_mode(inode_item, inode->i_mode);
1749 btrfs_set_stack_inode_nlink(inode_item, inode->i_nlink);
1750 btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(inode));
1751 btrfs_set_stack_inode_generation(inode_item,
1752 BTRFS_I(inode)->generation);
Josef Bacik0c4d2d92012-04-05 15:03:02 -04001753 btrfs_set_stack_inode_sequence(inode_item, inode->i_version);
Miao Xie16cdcec2011-04-22 18:12:22 +08001754 btrfs_set_stack_inode_transid(inode_item, trans->transid);
1755 btrfs_set_stack_inode_rdev(inode_item, inode->i_rdev);
1756 btrfs_set_stack_inode_flags(inode_item, BTRFS_I(inode)->flags);
Chris Masonff5714c2011-05-28 07:00:39 -04001757 btrfs_set_stack_inode_block_group(inode_item, 0);
Miao Xie16cdcec2011-04-22 18:12:22 +08001758
David Sterbaa937b972014-12-12 17:39:12 +01001759 btrfs_set_stack_timespec_sec(&inode_item->atime,
Miao Xie16cdcec2011-04-22 18:12:22 +08001760 inode->i_atime.tv_sec);
David Sterbaa937b972014-12-12 17:39:12 +01001761 btrfs_set_stack_timespec_nsec(&inode_item->atime,
Miao Xie16cdcec2011-04-22 18:12:22 +08001762 inode->i_atime.tv_nsec);
1763
David Sterbaa937b972014-12-12 17:39:12 +01001764 btrfs_set_stack_timespec_sec(&inode_item->mtime,
Miao Xie16cdcec2011-04-22 18:12:22 +08001765 inode->i_mtime.tv_sec);
David Sterbaa937b972014-12-12 17:39:12 +01001766 btrfs_set_stack_timespec_nsec(&inode_item->mtime,
Miao Xie16cdcec2011-04-22 18:12:22 +08001767 inode->i_mtime.tv_nsec);
1768
David Sterbaa937b972014-12-12 17:39:12 +01001769 btrfs_set_stack_timespec_sec(&inode_item->ctime,
Miao Xie16cdcec2011-04-22 18:12:22 +08001770 inode->i_ctime.tv_sec);
David Sterbaa937b972014-12-12 17:39:12 +01001771 btrfs_set_stack_timespec_nsec(&inode_item->ctime,
Miao Xie16cdcec2011-04-22 18:12:22 +08001772 inode->i_ctime.tv_nsec);
chandan r9cc97d62012-07-04 12:48:07 +05301773
1774 btrfs_set_stack_timespec_sec(&inode_item->otime,
1775 BTRFS_I(inode)->i_otime.tv_sec);
1776 btrfs_set_stack_timespec_nsec(&inode_item->otime,
1777 BTRFS_I(inode)->i_otime.tv_nsec);
Miao Xie16cdcec2011-04-22 18:12:22 +08001778}
1779
Miao Xie2f7e33d2011-06-23 07:27:13 +00001780int btrfs_fill_inode(struct inode *inode, u32 *rdev)
1781{
1782 struct btrfs_delayed_node *delayed_node;
1783 struct btrfs_inode_item *inode_item;
Miao Xie2f7e33d2011-06-23 07:27:13 +00001784
1785 delayed_node = btrfs_get_delayed_node(inode);
1786 if (!delayed_node)
1787 return -ENOENT;
1788
1789 mutex_lock(&delayed_node->mutex);
Miao Xie7cf35d92013-12-26 13:07:05 +08001790 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
Miao Xie2f7e33d2011-06-23 07:27:13 +00001791 mutex_unlock(&delayed_node->mutex);
1792 btrfs_release_delayed_node(delayed_node);
1793 return -ENOENT;
1794 }
1795
1796 inode_item = &delayed_node->inode_item;
1797
Eric W. Biederman2f2f43d2012-02-10 11:05:07 -08001798 i_uid_write(inode, btrfs_stack_inode_uid(inode_item));
1799 i_gid_write(inode, btrfs_stack_inode_gid(inode_item));
Miao Xie2f7e33d2011-06-23 07:27:13 +00001800 btrfs_i_size_write(inode, btrfs_stack_inode_size(inode_item));
1801 inode->i_mode = btrfs_stack_inode_mode(inode_item);
Miklos Szeredibfe86842011-10-28 14:13:29 +02001802 set_nlink(inode, btrfs_stack_inode_nlink(inode_item));
Miao Xie2f7e33d2011-06-23 07:27:13 +00001803 inode_set_bytes(inode, btrfs_stack_inode_nbytes(inode_item));
1804 BTRFS_I(inode)->generation = btrfs_stack_inode_generation(inode_item);
Yang Dongsheng6e17d302015-04-09 12:08:43 +08001805 BTRFS_I(inode)->last_trans = btrfs_stack_inode_transid(inode_item);
1806
Josef Bacik0c4d2d92012-04-05 15:03:02 -04001807 inode->i_version = btrfs_stack_inode_sequence(inode_item);
Miao Xie2f7e33d2011-06-23 07:27:13 +00001808 inode->i_rdev = 0;
1809 *rdev = btrfs_stack_inode_rdev(inode_item);
1810 BTRFS_I(inode)->flags = btrfs_stack_inode_flags(inode_item);
1811
David Sterbaa937b972014-12-12 17:39:12 +01001812 inode->i_atime.tv_sec = btrfs_stack_timespec_sec(&inode_item->atime);
1813 inode->i_atime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->atime);
Miao Xie2f7e33d2011-06-23 07:27:13 +00001814
David Sterbaa937b972014-12-12 17:39:12 +01001815 inode->i_mtime.tv_sec = btrfs_stack_timespec_sec(&inode_item->mtime);
1816 inode->i_mtime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->mtime);
Miao Xie2f7e33d2011-06-23 07:27:13 +00001817
David Sterbaa937b972014-12-12 17:39:12 +01001818 inode->i_ctime.tv_sec = btrfs_stack_timespec_sec(&inode_item->ctime);
1819 inode->i_ctime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->ctime);
Miao Xie2f7e33d2011-06-23 07:27:13 +00001820
chandan r9cc97d62012-07-04 12:48:07 +05301821 BTRFS_I(inode)->i_otime.tv_sec =
1822 btrfs_stack_timespec_sec(&inode_item->otime);
1823 BTRFS_I(inode)->i_otime.tv_nsec =
1824 btrfs_stack_timespec_nsec(&inode_item->otime);
1825
Miao Xie2f7e33d2011-06-23 07:27:13 +00001826 inode->i_generation = BTRFS_I(inode)->generation;
1827 BTRFS_I(inode)->index_cnt = (u64)-1;
1828
1829 mutex_unlock(&delayed_node->mutex);
1830 btrfs_release_delayed_node(delayed_node);
1831 return 0;
1832}
1833
Miao Xie16cdcec2011-04-22 18:12:22 +08001834int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans,
1835 struct btrfs_root *root, struct inode *inode)
1836{
1837 struct btrfs_delayed_node *delayed_node;
David Sterbaaa0467d2011-06-03 16:29:08 +02001838 int ret = 0;
Miao Xie16cdcec2011-04-22 18:12:22 +08001839
1840 delayed_node = btrfs_get_or_create_delayed_node(inode);
1841 if (IS_ERR(delayed_node))
1842 return PTR_ERR(delayed_node);
1843
1844 mutex_lock(&delayed_node->mutex);
Miao Xie7cf35d92013-12-26 13:07:05 +08001845 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
Miao Xie16cdcec2011-04-22 18:12:22 +08001846 fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1847 goto release_node;
1848 }
1849
Josef Bacik7fd2ae22011-11-08 15:47:34 -05001850 ret = btrfs_delayed_inode_reserve_metadata(trans, root, inode,
1851 delayed_node);
Josef Bacikc06a0e12011-11-04 19:56:02 -04001852 if (ret)
1853 goto release_node;
Miao Xie16cdcec2011-04-22 18:12:22 +08001854
1855 fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
Miao Xie7cf35d92013-12-26 13:07:05 +08001856 set_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
Miao Xie16cdcec2011-04-22 18:12:22 +08001857 delayed_node->count++;
1858 atomic_inc(&root->fs_info->delayed_root->items);
1859release_node:
1860 mutex_unlock(&delayed_node->mutex);
1861 btrfs_release_delayed_node(delayed_node);
1862 return ret;
1863}
1864
Miao Xie67de1172013-12-26 13:07:06 +08001865int btrfs_delayed_delete_inode_ref(struct inode *inode)
1866{
1867 struct btrfs_delayed_node *delayed_node;
1868
Chris Mason6f896052014-12-31 12:18:29 -05001869 /*
1870 * we don't do delayed inode updates during log recovery because it
1871 * leads to enospc problems. This means we also can't do
1872 * delayed inode refs
1873 */
1874 if (BTRFS_I(inode)->root->fs_info->log_root_recovering)
1875 return -EAGAIN;
1876
Miao Xie67de1172013-12-26 13:07:06 +08001877 delayed_node = btrfs_get_or_create_delayed_node(inode);
1878 if (IS_ERR(delayed_node))
1879 return PTR_ERR(delayed_node);
1880
1881 /*
1882 * We don't reserve space for inode ref deletion is because:
1883 * - We ONLY do async inode ref deletion for the inode who has only
1884 * one link(i_nlink == 1), it means there is only one inode ref.
1885 * And in most case, the inode ref and the inode item are in the
1886 * same leaf, and we will deal with them at the same time.
1887 * Since we are sure we will reserve the space for the inode item,
1888 * it is unnecessary to reserve space for inode ref deletion.
1889 * - If the inode ref and the inode item are not in the same leaf,
1890 * We also needn't worry about enospc problem, because we reserve
1891 * much more space for the inode update than it needs.
1892 * - At the worst, we can steal some space from the global reservation.
1893 * It is very rare.
1894 */
1895 mutex_lock(&delayed_node->mutex);
1896 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1897 goto release_node;
1898
1899 set_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
1900 delayed_node->count++;
1901 atomic_inc(&BTRFS_I(inode)->root->fs_info->delayed_root->items);
1902release_node:
1903 mutex_unlock(&delayed_node->mutex);
1904 btrfs_release_delayed_node(delayed_node);
1905 return 0;
1906}
1907
Miao Xie16cdcec2011-04-22 18:12:22 +08001908static void __btrfs_kill_delayed_node(struct btrfs_delayed_node *delayed_node)
1909{
1910 struct btrfs_root *root = delayed_node->root;
1911 struct btrfs_delayed_item *curr_item, *prev_item;
1912
1913 mutex_lock(&delayed_node->mutex);
1914 curr_item = __btrfs_first_delayed_insertion_item(delayed_node);
1915 while (curr_item) {
1916 btrfs_delayed_item_release_metadata(root, curr_item);
1917 prev_item = curr_item;
1918 curr_item = __btrfs_next_delayed_item(prev_item);
1919 btrfs_release_delayed_item(prev_item);
1920 }
1921
1922 curr_item = __btrfs_first_delayed_deletion_item(delayed_node);
1923 while (curr_item) {
1924 btrfs_delayed_item_release_metadata(root, curr_item);
1925 prev_item = curr_item;
1926 curr_item = __btrfs_next_delayed_item(prev_item);
1927 btrfs_release_delayed_item(prev_item);
1928 }
1929
Miao Xie67de1172013-12-26 13:07:06 +08001930 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1931 btrfs_release_delayed_iref(delayed_node);
1932
Miao Xie7cf35d92013-12-26 13:07:05 +08001933 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
Miao Xie16cdcec2011-04-22 18:12:22 +08001934 btrfs_delayed_inode_release_metadata(root, delayed_node);
1935 btrfs_release_delayed_inode(delayed_node);
1936 }
1937 mutex_unlock(&delayed_node->mutex);
1938}
1939
1940void btrfs_kill_delayed_inode_items(struct inode *inode)
1941{
1942 struct btrfs_delayed_node *delayed_node;
1943
1944 delayed_node = btrfs_get_delayed_node(inode);
1945 if (!delayed_node)
1946 return;
1947
1948 __btrfs_kill_delayed_node(delayed_node);
1949 btrfs_release_delayed_node(delayed_node);
1950}
1951
1952void btrfs_kill_all_delayed_nodes(struct btrfs_root *root)
1953{
1954 u64 inode_id = 0;
1955 struct btrfs_delayed_node *delayed_nodes[8];
1956 int i, n;
1957
1958 while (1) {
1959 spin_lock(&root->inode_lock);
1960 n = radix_tree_gang_lookup(&root->delayed_nodes_tree,
1961 (void **)delayed_nodes, inode_id,
1962 ARRAY_SIZE(delayed_nodes));
1963 if (!n) {
1964 spin_unlock(&root->inode_lock);
1965 break;
1966 }
1967
1968 inode_id = delayed_nodes[n - 1]->inode_id + 1;
1969
1970 for (i = 0; i < n; i++)
1971 atomic_inc(&delayed_nodes[i]->refs);
1972 spin_unlock(&root->inode_lock);
1973
1974 for (i = 0; i < n; i++) {
1975 __btrfs_kill_delayed_node(delayed_nodes[i]);
1976 btrfs_release_delayed_node(delayed_nodes[i]);
1977 }
1978 }
1979}
Miao Xie67cde342012-06-14 02:23:22 -06001980
1981void btrfs_destroy_delayed_inodes(struct btrfs_root *root)
1982{
1983 struct btrfs_delayed_root *delayed_root;
1984 struct btrfs_delayed_node *curr_node, *prev_node;
1985
1986 delayed_root = btrfs_get_delayed_root(root);
1987
1988 curr_node = btrfs_first_delayed_node(delayed_root);
1989 while (curr_node) {
1990 __btrfs_kill_delayed_node(curr_node);
1991
1992 prev_node = curr_node;
1993 curr_node = btrfs_next_delayed_node(curr_node);
1994 btrfs_release_delayed_node(prev_node);
1995 }
1996}
1997