2 * Copyright (C) 2011 Fujitsu. All rights reserved.
3 * Written by Miao Xie <miaox@cn.fujitsu.com>
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.
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.
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.
20 #include <linux/slab.h>
21 #include "delayed-inode.h"
23 #include "transaction.h"
26 #define BTRFS_DELAYED_WRITEBACK 512
27 #define BTRFS_DELAYED_BACKGROUND 128
28 #define BTRFS_DELAYED_BATCH 16
30 static struct kmem_cache *delayed_node_cache;
32 int __init btrfs_delayed_inode_init(void)
34 delayed_node_cache = kmem_cache_create("btrfs_delayed_node",
35 sizeof(struct btrfs_delayed_node),
37 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
39 if (!delayed_node_cache)
44 void btrfs_delayed_inode_exit(void)
46 kmem_cache_destroy(delayed_node_cache);
49 static inline void btrfs_init_delayed_node(
50 struct btrfs_delayed_node *delayed_node,
51 struct btrfs_root *root, u64 inode_id)
53 delayed_node->root = root;
54 delayed_node->inode_id = inode_id;
55 atomic_set(&delayed_node->refs, 0);
56 delayed_node->ins_root = RB_ROOT;
57 delayed_node->del_root = RB_ROOT;
58 mutex_init(&delayed_node->mutex);
59 INIT_LIST_HEAD(&delayed_node->n_list);
60 INIT_LIST_HEAD(&delayed_node->p_list);
63 static inline int btrfs_is_continuous_delayed_item(
64 struct btrfs_delayed_item *item1,
65 struct btrfs_delayed_item *item2)
67 if (item1->key.type == BTRFS_DIR_INDEX_KEY &&
68 item1->key.objectid == item2->key.objectid &&
69 item1->key.type == item2->key.type &&
70 item1->key.offset + 1 == item2->key.offset)
75 static inline struct btrfs_delayed_root *btrfs_get_delayed_root(
76 struct btrfs_root *root)
78 return root->fs_info->delayed_root;
81 static struct btrfs_delayed_node *btrfs_get_delayed_node(struct inode *inode)
83 struct btrfs_inode *btrfs_inode = BTRFS_I(inode);
84 struct btrfs_root *root = btrfs_inode->root;
85 u64 ino = btrfs_ino(inode);
86 struct btrfs_delayed_node *node;
88 node = ACCESS_ONCE(btrfs_inode->delayed_node);
90 atomic_inc(&node->refs);
94 spin_lock(&root->inode_lock);
95 node = radix_tree_lookup(&root->delayed_nodes_tree, ino);
97 if (btrfs_inode->delayed_node) {
98 atomic_inc(&node->refs); /* can be accessed */
99 BUG_ON(btrfs_inode->delayed_node != node);
100 spin_unlock(&root->inode_lock);
103 btrfs_inode->delayed_node = node;
104 /* can be accessed and cached in the inode */
105 atomic_add(2, &node->refs);
106 spin_unlock(&root->inode_lock);
109 spin_unlock(&root->inode_lock);
114 /* Will return either the node or PTR_ERR(-ENOMEM) */
115 static struct btrfs_delayed_node *btrfs_get_or_create_delayed_node(
118 struct btrfs_delayed_node *node;
119 struct btrfs_inode *btrfs_inode = BTRFS_I(inode);
120 struct btrfs_root *root = btrfs_inode->root;
121 u64 ino = btrfs_ino(inode);
125 node = btrfs_get_delayed_node(inode);
129 node = kmem_cache_zalloc(delayed_node_cache, GFP_NOFS);
131 return ERR_PTR(-ENOMEM);
132 btrfs_init_delayed_node(node, root, ino);
134 /* cached in the btrfs inode and can be accessed */
135 atomic_add(2, &node->refs);
137 ret = radix_tree_preload(GFP_NOFS);
139 kmem_cache_free(delayed_node_cache, node);
143 spin_lock(&root->inode_lock);
144 ret = radix_tree_insert(&root->delayed_nodes_tree, ino, node);
145 if (ret == -EEXIST) {
146 spin_unlock(&root->inode_lock);
147 kmem_cache_free(delayed_node_cache, node);
148 radix_tree_preload_end();
151 btrfs_inode->delayed_node = node;
152 spin_unlock(&root->inode_lock);
153 radix_tree_preload_end();
159 * Call it when holding delayed_node->mutex
161 * If mod = 1, add this node into the prepared list.
163 static void btrfs_queue_delayed_node(struct btrfs_delayed_root *root,
164 struct btrfs_delayed_node *node,
167 spin_lock(&root->lock);
168 if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
169 if (!list_empty(&node->p_list))
170 list_move_tail(&node->p_list, &root->prepare_list);
172 list_add_tail(&node->p_list, &root->prepare_list);
174 list_add_tail(&node->n_list, &root->node_list);
175 list_add_tail(&node->p_list, &root->prepare_list);
176 atomic_inc(&node->refs); /* inserted into list */
178 set_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
180 spin_unlock(&root->lock);
183 /* Call it when holding delayed_node->mutex */
184 static void btrfs_dequeue_delayed_node(struct btrfs_delayed_root *root,
185 struct btrfs_delayed_node *node)
187 spin_lock(&root->lock);
188 if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
190 atomic_dec(&node->refs); /* not in the list */
191 list_del_init(&node->n_list);
192 if (!list_empty(&node->p_list))
193 list_del_init(&node->p_list);
194 clear_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
196 spin_unlock(&root->lock);
199 static struct btrfs_delayed_node *btrfs_first_delayed_node(
200 struct btrfs_delayed_root *delayed_root)
203 struct btrfs_delayed_node *node = NULL;
205 spin_lock(&delayed_root->lock);
206 if (list_empty(&delayed_root->node_list))
209 p = delayed_root->node_list.next;
210 node = list_entry(p, struct btrfs_delayed_node, n_list);
211 atomic_inc(&node->refs);
213 spin_unlock(&delayed_root->lock);
218 static struct btrfs_delayed_node *btrfs_next_delayed_node(
219 struct btrfs_delayed_node *node)
221 struct btrfs_delayed_root *delayed_root;
223 struct btrfs_delayed_node *next = NULL;
225 delayed_root = node->root->fs_info->delayed_root;
226 spin_lock(&delayed_root->lock);
227 if (!test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
228 /* not in the list */
229 if (list_empty(&delayed_root->node_list))
231 p = delayed_root->node_list.next;
232 } else if (list_is_last(&node->n_list, &delayed_root->node_list))
235 p = node->n_list.next;
237 next = list_entry(p, struct btrfs_delayed_node, n_list);
238 atomic_inc(&next->refs);
240 spin_unlock(&delayed_root->lock);
245 static void __btrfs_release_delayed_node(
246 struct btrfs_delayed_node *delayed_node,
249 struct btrfs_delayed_root *delayed_root;
254 delayed_root = delayed_node->root->fs_info->delayed_root;
256 mutex_lock(&delayed_node->mutex);
257 if (delayed_node->count)
258 btrfs_queue_delayed_node(delayed_root, delayed_node, mod);
260 btrfs_dequeue_delayed_node(delayed_root, delayed_node);
261 mutex_unlock(&delayed_node->mutex);
263 if (atomic_dec_and_test(&delayed_node->refs)) {
265 struct btrfs_root *root = delayed_node->root;
266 spin_lock(&root->inode_lock);
267 if (atomic_read(&delayed_node->refs) == 0) {
268 radix_tree_delete(&root->delayed_nodes_tree,
269 delayed_node->inode_id);
272 spin_unlock(&root->inode_lock);
274 kmem_cache_free(delayed_node_cache, delayed_node);
278 static inline void btrfs_release_delayed_node(struct btrfs_delayed_node *node)
280 __btrfs_release_delayed_node(node, 0);
283 static struct btrfs_delayed_node *btrfs_first_prepared_delayed_node(
284 struct btrfs_delayed_root *delayed_root)
287 struct btrfs_delayed_node *node = NULL;
289 spin_lock(&delayed_root->lock);
290 if (list_empty(&delayed_root->prepare_list))
293 p = delayed_root->prepare_list.next;
295 node = list_entry(p, struct btrfs_delayed_node, p_list);
296 atomic_inc(&node->refs);
298 spin_unlock(&delayed_root->lock);
303 static inline void btrfs_release_prepared_delayed_node(
304 struct btrfs_delayed_node *node)
306 __btrfs_release_delayed_node(node, 1);
309 static struct btrfs_delayed_item *btrfs_alloc_delayed_item(u32 data_len)
311 struct btrfs_delayed_item *item;
312 item = kmalloc(sizeof(*item) + data_len, GFP_NOFS);
314 item->data_len = data_len;
315 item->ins_or_del = 0;
316 item->bytes_reserved = 0;
317 item->delayed_node = NULL;
318 atomic_set(&item->refs, 1);
324 * __btrfs_lookup_delayed_item - look up the delayed item by key
325 * @delayed_node: pointer to the delayed node
326 * @key: the key to look up
327 * @prev: used to store the prev item if the right item isn't found
328 * @next: used to store the next item if the right item isn't found
330 * Note: if we don't find the right item, we will return the prev item and
333 static struct btrfs_delayed_item *__btrfs_lookup_delayed_item(
334 struct rb_root *root,
335 struct btrfs_key *key,
336 struct btrfs_delayed_item **prev,
337 struct btrfs_delayed_item **next)
339 struct rb_node *node, *prev_node = NULL;
340 struct btrfs_delayed_item *delayed_item = NULL;
343 node = root->rb_node;
346 delayed_item = rb_entry(node, struct btrfs_delayed_item,
349 ret = btrfs_comp_cpu_keys(&delayed_item->key, key);
351 node = node->rb_right;
353 node = node->rb_left;
362 *prev = delayed_item;
363 else if ((node = rb_prev(prev_node)) != NULL) {
364 *prev = rb_entry(node, struct btrfs_delayed_item,
374 *next = delayed_item;
375 else if ((node = rb_next(prev_node)) != NULL) {
376 *next = rb_entry(node, struct btrfs_delayed_item,
384 static struct btrfs_delayed_item *__btrfs_lookup_delayed_insertion_item(
385 struct btrfs_delayed_node *delayed_node,
386 struct btrfs_key *key)
388 struct btrfs_delayed_item *item;
390 item = __btrfs_lookup_delayed_item(&delayed_node->ins_root, key,
395 static int __btrfs_add_delayed_item(struct btrfs_delayed_node *delayed_node,
396 struct btrfs_delayed_item *ins,
399 struct rb_node **p, *node;
400 struct rb_node *parent_node = NULL;
401 struct rb_root *root;
402 struct btrfs_delayed_item *item;
405 if (action == BTRFS_DELAYED_INSERTION_ITEM)
406 root = &delayed_node->ins_root;
407 else if (action == BTRFS_DELAYED_DELETION_ITEM)
408 root = &delayed_node->del_root;
412 node = &ins->rb_node;
416 item = rb_entry(parent_node, struct btrfs_delayed_item,
419 cmp = btrfs_comp_cpu_keys(&item->key, &ins->key);
428 rb_link_node(node, parent_node, p);
429 rb_insert_color(node, root);
430 ins->delayed_node = delayed_node;
431 ins->ins_or_del = action;
433 if (ins->key.type == BTRFS_DIR_INDEX_KEY &&
434 action == BTRFS_DELAYED_INSERTION_ITEM &&
435 ins->key.offset >= delayed_node->index_cnt)
436 delayed_node->index_cnt = ins->key.offset + 1;
438 delayed_node->count++;
439 atomic_inc(&delayed_node->root->fs_info->delayed_root->items);
443 static int __btrfs_add_delayed_insertion_item(struct btrfs_delayed_node *node,
444 struct btrfs_delayed_item *item)
446 return __btrfs_add_delayed_item(node, item,
447 BTRFS_DELAYED_INSERTION_ITEM);
450 static int __btrfs_add_delayed_deletion_item(struct btrfs_delayed_node *node,
451 struct btrfs_delayed_item *item)
453 return __btrfs_add_delayed_item(node, item,
454 BTRFS_DELAYED_DELETION_ITEM);
457 static void finish_one_item(struct btrfs_delayed_root *delayed_root)
459 int seq = atomic_inc_return(&delayed_root->items_seq);
462 * atomic_dec_return implies a barrier for waitqueue_active
464 if ((atomic_dec_return(&delayed_root->items) <
465 BTRFS_DELAYED_BACKGROUND || seq % BTRFS_DELAYED_BATCH == 0) &&
466 waitqueue_active(&delayed_root->wait))
467 wake_up(&delayed_root->wait);
470 static void __btrfs_remove_delayed_item(struct btrfs_delayed_item *delayed_item)
472 struct rb_root *root;
473 struct btrfs_delayed_root *delayed_root;
475 delayed_root = delayed_item->delayed_node->root->fs_info->delayed_root;
477 BUG_ON(!delayed_root);
478 BUG_ON(delayed_item->ins_or_del != BTRFS_DELAYED_DELETION_ITEM &&
479 delayed_item->ins_or_del != BTRFS_DELAYED_INSERTION_ITEM);
481 if (delayed_item->ins_or_del == BTRFS_DELAYED_INSERTION_ITEM)
482 root = &delayed_item->delayed_node->ins_root;
484 root = &delayed_item->delayed_node->del_root;
486 rb_erase(&delayed_item->rb_node, root);
487 delayed_item->delayed_node->count--;
489 finish_one_item(delayed_root);
492 static void btrfs_release_delayed_item(struct btrfs_delayed_item *item)
495 __btrfs_remove_delayed_item(item);
496 if (atomic_dec_and_test(&item->refs))
501 static struct btrfs_delayed_item *__btrfs_first_delayed_insertion_item(
502 struct btrfs_delayed_node *delayed_node)
505 struct btrfs_delayed_item *item = NULL;
507 p = rb_first(&delayed_node->ins_root);
509 item = rb_entry(p, struct btrfs_delayed_item, rb_node);
514 static struct btrfs_delayed_item *__btrfs_first_delayed_deletion_item(
515 struct btrfs_delayed_node *delayed_node)
518 struct btrfs_delayed_item *item = NULL;
520 p = rb_first(&delayed_node->del_root);
522 item = rb_entry(p, struct btrfs_delayed_item, rb_node);
527 static struct btrfs_delayed_item *__btrfs_next_delayed_item(
528 struct btrfs_delayed_item *item)
531 struct btrfs_delayed_item *next = NULL;
533 p = rb_next(&item->rb_node);
535 next = rb_entry(p, struct btrfs_delayed_item, rb_node);
540 static int btrfs_delayed_item_reserve_metadata(struct btrfs_trans_handle *trans,
541 struct btrfs_root *root,
542 struct btrfs_delayed_item *item)
544 struct btrfs_block_rsv *src_rsv;
545 struct btrfs_block_rsv *dst_rsv;
549 if (!trans->bytes_reserved)
552 src_rsv = trans->block_rsv;
553 dst_rsv = &root->fs_info->delayed_block_rsv;
555 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
556 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes);
558 trace_btrfs_space_reservation(root->fs_info, "delayed_item",
561 item->bytes_reserved = num_bytes;
567 static void btrfs_delayed_item_release_metadata(struct btrfs_root *root,
568 struct btrfs_delayed_item *item)
570 struct btrfs_block_rsv *rsv;
572 if (!item->bytes_reserved)
575 rsv = &root->fs_info->delayed_block_rsv;
576 trace_btrfs_space_reservation(root->fs_info, "delayed_item",
577 item->key.objectid, item->bytes_reserved,
579 btrfs_block_rsv_release(root, rsv,
580 item->bytes_reserved);
583 static int btrfs_delayed_inode_reserve_metadata(
584 struct btrfs_trans_handle *trans,
585 struct btrfs_root *root,
587 struct btrfs_delayed_node *node)
589 struct btrfs_block_rsv *src_rsv;
590 struct btrfs_block_rsv *dst_rsv;
593 bool release = false;
595 src_rsv = trans->block_rsv;
596 dst_rsv = &root->fs_info->delayed_block_rsv;
598 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
601 * btrfs_dirty_inode will update the inode under btrfs_join_transaction
602 * which doesn't reserve space for speed. This is a problem since we
603 * still need to reserve space for this update, so try to reserve the
606 * Now if src_rsv == delalloc_block_rsv we'll let it just steal since
607 * we're accounted for.
609 if (!src_rsv || (!trans->bytes_reserved &&
610 src_rsv->type != BTRFS_BLOCK_RSV_DELALLOC)) {
611 ret = btrfs_block_rsv_add(root, dst_rsv, num_bytes,
612 BTRFS_RESERVE_NO_FLUSH);
614 * Since we're under a transaction reserve_metadata_bytes could
615 * try to commit the transaction which will make it return
616 * EAGAIN to make us stop the transaction we have, so return
617 * ENOSPC instead so that btrfs_dirty_inode knows what to do.
622 node->bytes_reserved = num_bytes;
623 trace_btrfs_space_reservation(root->fs_info,
629 } else if (src_rsv->type == BTRFS_BLOCK_RSV_DELALLOC) {
630 spin_lock(&BTRFS_I(inode)->lock);
631 if (test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
632 &BTRFS_I(inode)->runtime_flags)) {
633 spin_unlock(&BTRFS_I(inode)->lock);
637 spin_unlock(&BTRFS_I(inode)->lock);
639 /* Ok we didn't have space pre-reserved. This shouldn't happen
640 * too often but it can happen if we do delalloc to an existing
641 * inode which gets dirtied because of the time update, and then
642 * isn't touched again until after the transaction commits and
643 * then we try to write out the data. First try to be nice and
644 * reserve something strictly for us. If not be a pain and try
645 * to steal from the delalloc block rsv.
647 ret = btrfs_block_rsv_add(root, dst_rsv, num_bytes,
648 BTRFS_RESERVE_NO_FLUSH);
652 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes);
656 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
657 btrfs_debug(root->fs_info,
658 "block rsv migrate returned %d", ret);
662 * Ok this is a problem, let's just steal from the global rsv
663 * since this really shouldn't happen that often.
665 ret = btrfs_block_rsv_migrate(&root->fs_info->global_block_rsv,
671 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes);
675 * Migrate only takes a reservation, it doesn't touch the size of the
676 * block_rsv. This is to simplify people who don't normally have things
677 * migrated from their block rsv. If they go to release their
678 * reservation, that will decrease the size as well, so if migrate
679 * reduced size we'd end up with a negative size. But for the
680 * delalloc_meta_reserved stuff we will only know to drop 1 reservation,
681 * but we could in fact do this reserve/migrate dance several times
682 * between the time we did the original reservation and we'd clean it
683 * up. So to take care of this, release the space for the meta
684 * reservation here. I think it may be time for a documentation page on
685 * how block rsvs. work.
688 trace_btrfs_space_reservation(root->fs_info, "delayed_inode",
689 btrfs_ino(inode), num_bytes, 1);
690 node->bytes_reserved = num_bytes;
694 trace_btrfs_space_reservation(root->fs_info, "delalloc",
695 btrfs_ino(inode), num_bytes, 0);
696 btrfs_block_rsv_release(root, src_rsv, num_bytes);
702 static void btrfs_delayed_inode_release_metadata(struct btrfs_root *root,
703 struct btrfs_delayed_node *node)
705 struct btrfs_block_rsv *rsv;
707 if (!node->bytes_reserved)
710 rsv = &root->fs_info->delayed_block_rsv;
711 trace_btrfs_space_reservation(root->fs_info, "delayed_inode",
712 node->inode_id, node->bytes_reserved, 0);
713 btrfs_block_rsv_release(root, rsv,
714 node->bytes_reserved);
715 node->bytes_reserved = 0;
719 * This helper will insert some continuous items into the same leaf according
720 * to the free space of the leaf.
722 static int btrfs_batch_insert_items(struct btrfs_root *root,
723 struct btrfs_path *path,
724 struct btrfs_delayed_item *item)
726 struct btrfs_delayed_item *curr, *next;
728 int total_data_size = 0, total_size = 0;
729 struct extent_buffer *leaf;
731 struct btrfs_key *keys;
733 struct list_head head;
739 BUG_ON(!path->nodes[0]);
741 leaf = path->nodes[0];
742 free_space = btrfs_leaf_free_space(root, leaf);
743 INIT_LIST_HEAD(&head);
749 * count the number of the continuous items that we can insert in batch
751 while (total_size + next->data_len + sizeof(struct btrfs_item) <=
753 total_data_size += next->data_len;
754 total_size += next->data_len + sizeof(struct btrfs_item);
755 list_add_tail(&next->tree_list, &head);
759 next = __btrfs_next_delayed_item(curr);
763 if (!btrfs_is_continuous_delayed_item(curr, next))
773 * we need allocate some memory space, but it might cause the task
774 * to sleep, so we set all locked nodes in the path to blocking locks
777 btrfs_set_path_blocking(path);
779 keys = kmalloc_array(nitems, sizeof(struct btrfs_key), GFP_NOFS);
785 data_size = kmalloc_array(nitems, sizeof(u32), GFP_NOFS);
791 /* get keys of all the delayed items */
793 list_for_each_entry(next, &head, tree_list) {
795 data_size[i] = next->data_len;
799 /* reset all the locked nodes in the patch to spinning locks. */
800 btrfs_clear_path_blocking(path, NULL, 0);
802 /* insert the keys of the items */
803 setup_items_for_insert(root, path, keys, data_size,
804 total_data_size, total_size, nitems);
806 /* insert the dir index items */
807 slot = path->slots[0];
808 list_for_each_entry_safe(curr, next, &head, tree_list) {
809 data_ptr = btrfs_item_ptr(leaf, slot, char);
810 write_extent_buffer(leaf, &curr->data,
811 (unsigned long)data_ptr,
815 btrfs_delayed_item_release_metadata(root, curr);
817 list_del(&curr->tree_list);
818 btrfs_release_delayed_item(curr);
829 * This helper can just do simple insertion that needn't extend item for new
830 * data, such as directory name index insertion, inode insertion.
832 static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans,
833 struct btrfs_root *root,
834 struct btrfs_path *path,
835 struct btrfs_delayed_item *delayed_item)
837 struct extent_buffer *leaf;
841 ret = btrfs_insert_empty_item(trans, root, path, &delayed_item->key,
842 delayed_item->data_len);
843 if (ret < 0 && ret != -EEXIST)
846 leaf = path->nodes[0];
848 ptr = btrfs_item_ptr(leaf, path->slots[0], char);
850 write_extent_buffer(leaf, delayed_item->data, (unsigned long)ptr,
851 delayed_item->data_len);
852 btrfs_mark_buffer_dirty(leaf);
854 btrfs_delayed_item_release_metadata(root, delayed_item);
859 * we insert an item first, then if there are some continuous items, we try
860 * to insert those items into the same leaf.
862 static int btrfs_insert_delayed_items(struct btrfs_trans_handle *trans,
863 struct btrfs_path *path,
864 struct btrfs_root *root,
865 struct btrfs_delayed_node *node)
867 struct btrfs_delayed_item *curr, *prev;
871 mutex_lock(&node->mutex);
872 curr = __btrfs_first_delayed_insertion_item(node);
876 ret = btrfs_insert_delayed_item(trans, root, path, curr);
878 btrfs_release_path(path);
883 curr = __btrfs_next_delayed_item(prev);
884 if (curr && btrfs_is_continuous_delayed_item(prev, curr)) {
885 /* insert the continuous items into the same leaf */
887 btrfs_batch_insert_items(root, path, curr);
889 btrfs_release_delayed_item(prev);
890 btrfs_mark_buffer_dirty(path->nodes[0]);
892 btrfs_release_path(path);
893 mutex_unlock(&node->mutex);
897 mutex_unlock(&node->mutex);
901 static int btrfs_batch_delete_items(struct btrfs_trans_handle *trans,
902 struct btrfs_root *root,
903 struct btrfs_path *path,
904 struct btrfs_delayed_item *item)
906 struct btrfs_delayed_item *curr, *next;
907 struct extent_buffer *leaf;
908 struct btrfs_key key;
909 struct list_head head;
910 int nitems, i, last_item;
913 BUG_ON(!path->nodes[0]);
915 leaf = path->nodes[0];
918 last_item = btrfs_header_nritems(leaf) - 1;
920 return -ENOENT; /* FIXME: Is errno suitable? */
923 INIT_LIST_HEAD(&head);
924 btrfs_item_key_to_cpu(leaf, &key, i);
927 * count the number of the dir index items that we can delete in batch
929 while (btrfs_comp_cpu_keys(&next->key, &key) == 0) {
930 list_add_tail(&next->tree_list, &head);
934 next = __btrfs_next_delayed_item(curr);
938 if (!btrfs_is_continuous_delayed_item(curr, next))
944 btrfs_item_key_to_cpu(leaf, &key, i);
950 ret = btrfs_del_items(trans, root, path, path->slots[0], nitems);
954 list_for_each_entry_safe(curr, next, &head, tree_list) {
955 btrfs_delayed_item_release_metadata(root, curr);
956 list_del(&curr->tree_list);
957 btrfs_release_delayed_item(curr);
964 static int btrfs_delete_delayed_items(struct btrfs_trans_handle *trans,
965 struct btrfs_path *path,
966 struct btrfs_root *root,
967 struct btrfs_delayed_node *node)
969 struct btrfs_delayed_item *curr, *prev;
973 mutex_lock(&node->mutex);
974 curr = __btrfs_first_delayed_deletion_item(node);
978 ret = btrfs_search_slot(trans, root, &curr->key, path, -1, 1);
983 * can't find the item which the node points to, so this node
984 * is invalid, just drop it.
987 curr = __btrfs_next_delayed_item(prev);
988 btrfs_release_delayed_item(prev);
990 btrfs_release_path(path);
992 mutex_unlock(&node->mutex);
998 btrfs_batch_delete_items(trans, root, path, curr);
999 btrfs_release_path(path);
1000 mutex_unlock(&node->mutex);
1004 btrfs_release_path(path);
1005 mutex_unlock(&node->mutex);
1009 static void btrfs_release_delayed_inode(struct btrfs_delayed_node *delayed_node)
1011 struct btrfs_delayed_root *delayed_root;
1014 test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1015 BUG_ON(!delayed_node->root);
1016 clear_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
1017 delayed_node->count--;
1019 delayed_root = delayed_node->root->fs_info->delayed_root;
1020 finish_one_item(delayed_root);
1024 static void btrfs_release_delayed_iref(struct btrfs_delayed_node *delayed_node)
1026 struct btrfs_delayed_root *delayed_root;
1028 ASSERT(delayed_node->root);
1029 clear_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
1030 delayed_node->count--;
1032 delayed_root = delayed_node->root->fs_info->delayed_root;
1033 finish_one_item(delayed_root);
1036 static int __btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1037 struct btrfs_root *root,
1038 struct btrfs_path *path,
1039 struct btrfs_delayed_node *node)
1041 struct btrfs_key key;
1042 struct btrfs_inode_item *inode_item;
1043 struct extent_buffer *leaf;
1047 key.objectid = node->inode_id;
1048 key.type = BTRFS_INODE_ITEM_KEY;
1051 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1056 ret = btrfs_lookup_inode(trans, root, path, &key, mod);
1058 btrfs_release_path(path);
1060 } else if (ret < 0) {
1064 leaf = path->nodes[0];
1065 inode_item = btrfs_item_ptr(leaf, path->slots[0],
1066 struct btrfs_inode_item);
1067 write_extent_buffer(leaf, &node->inode_item, (unsigned long)inode_item,
1068 sizeof(struct btrfs_inode_item));
1069 btrfs_mark_buffer_dirty(leaf);
1071 if (!test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1075 if (path->slots[0] >= btrfs_header_nritems(leaf))
1078 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1079 if (key.objectid != node->inode_id)
1082 if (key.type != BTRFS_INODE_REF_KEY &&
1083 key.type != BTRFS_INODE_EXTREF_KEY)
1087 * Delayed iref deletion is for the inode who has only one link,
1088 * so there is only one iref. The case that several irefs are
1089 * in the same item doesn't exist.
1091 btrfs_del_item(trans, root, path);
1093 btrfs_release_delayed_iref(node);
1095 btrfs_release_path(path);
1097 btrfs_delayed_inode_release_metadata(root, node);
1098 btrfs_release_delayed_inode(node);
1103 btrfs_release_path(path);
1105 key.type = BTRFS_INODE_EXTREF_KEY;
1107 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1113 leaf = path->nodes[0];
1118 static inline int btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1119 struct btrfs_root *root,
1120 struct btrfs_path *path,
1121 struct btrfs_delayed_node *node)
1125 mutex_lock(&node->mutex);
1126 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &node->flags)) {
1127 mutex_unlock(&node->mutex);
1131 ret = __btrfs_update_delayed_inode(trans, root, path, node);
1132 mutex_unlock(&node->mutex);
1137 __btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1138 struct btrfs_path *path,
1139 struct btrfs_delayed_node *node)
1143 ret = btrfs_insert_delayed_items(trans, path, node->root, node);
1147 ret = btrfs_delete_delayed_items(trans, path, node->root, node);
1151 ret = btrfs_update_delayed_inode(trans, node->root, path, node);
1156 * Called when committing the transaction.
1157 * Returns 0 on success.
1158 * Returns < 0 on error and returns with an aborted transaction with any
1159 * outstanding delayed items cleaned up.
1161 static int __btrfs_run_delayed_items(struct btrfs_trans_handle *trans,
1162 struct btrfs_root *root, int nr)
1164 struct btrfs_delayed_root *delayed_root;
1165 struct btrfs_delayed_node *curr_node, *prev_node;
1166 struct btrfs_path *path;
1167 struct btrfs_block_rsv *block_rsv;
1169 bool count = (nr > 0);
1174 path = btrfs_alloc_path();
1177 path->leave_spinning = 1;
1179 block_rsv = trans->block_rsv;
1180 trans->block_rsv = &root->fs_info->delayed_block_rsv;
1182 delayed_root = btrfs_get_delayed_root(root);
1184 curr_node = btrfs_first_delayed_node(delayed_root);
1185 while (curr_node && (!count || (count && nr--))) {
1186 ret = __btrfs_commit_inode_delayed_items(trans, path,
1189 btrfs_release_delayed_node(curr_node);
1191 btrfs_abort_transaction(trans, root, ret);
1195 prev_node = curr_node;
1196 curr_node = btrfs_next_delayed_node(curr_node);
1197 btrfs_release_delayed_node(prev_node);
1201 btrfs_release_delayed_node(curr_node);
1202 btrfs_free_path(path);
1203 trans->block_rsv = block_rsv;
1208 int btrfs_run_delayed_items(struct btrfs_trans_handle *trans,
1209 struct btrfs_root *root)
1211 return __btrfs_run_delayed_items(trans, root, -1);
1214 int btrfs_run_delayed_items_nr(struct btrfs_trans_handle *trans,
1215 struct btrfs_root *root, int nr)
1217 return __btrfs_run_delayed_items(trans, root, nr);
1220 int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1221 struct inode *inode)
1223 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1224 struct btrfs_path *path;
1225 struct btrfs_block_rsv *block_rsv;
1231 mutex_lock(&delayed_node->mutex);
1232 if (!delayed_node->count) {
1233 mutex_unlock(&delayed_node->mutex);
1234 btrfs_release_delayed_node(delayed_node);
1237 mutex_unlock(&delayed_node->mutex);
1239 path = btrfs_alloc_path();
1241 btrfs_release_delayed_node(delayed_node);
1244 path->leave_spinning = 1;
1246 block_rsv = trans->block_rsv;
1247 trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv;
1249 ret = __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1251 btrfs_release_delayed_node(delayed_node);
1252 btrfs_free_path(path);
1253 trans->block_rsv = block_rsv;
1258 int btrfs_commit_inode_delayed_inode(struct inode *inode)
1260 struct btrfs_trans_handle *trans;
1261 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1262 struct btrfs_path *path;
1263 struct btrfs_block_rsv *block_rsv;
1269 mutex_lock(&delayed_node->mutex);
1270 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1271 mutex_unlock(&delayed_node->mutex);
1272 btrfs_release_delayed_node(delayed_node);
1275 mutex_unlock(&delayed_node->mutex);
1277 trans = btrfs_join_transaction(delayed_node->root);
1278 if (IS_ERR(trans)) {
1279 ret = PTR_ERR(trans);
1283 path = btrfs_alloc_path();
1288 path->leave_spinning = 1;
1290 block_rsv = trans->block_rsv;
1291 trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv;
1293 mutex_lock(&delayed_node->mutex);
1294 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags))
1295 ret = __btrfs_update_delayed_inode(trans, delayed_node->root,
1296 path, delayed_node);
1299 mutex_unlock(&delayed_node->mutex);
1301 btrfs_free_path(path);
1302 trans->block_rsv = block_rsv;
1304 btrfs_end_transaction(trans, delayed_node->root);
1305 btrfs_btree_balance_dirty(delayed_node->root);
1307 btrfs_release_delayed_node(delayed_node);
1312 void btrfs_remove_delayed_node(struct inode *inode)
1314 struct btrfs_delayed_node *delayed_node;
1316 delayed_node = ACCESS_ONCE(BTRFS_I(inode)->delayed_node);
1320 BTRFS_I(inode)->delayed_node = NULL;
1321 btrfs_release_delayed_node(delayed_node);
1324 struct btrfs_async_delayed_work {
1325 struct btrfs_delayed_root *delayed_root;
1327 struct btrfs_work work;
1330 static void btrfs_async_run_delayed_root(struct btrfs_work *work)
1332 struct btrfs_async_delayed_work *async_work;
1333 struct btrfs_delayed_root *delayed_root;
1334 struct btrfs_trans_handle *trans;
1335 struct btrfs_path *path;
1336 struct btrfs_delayed_node *delayed_node = NULL;
1337 struct btrfs_root *root;
1338 struct btrfs_block_rsv *block_rsv;
1341 async_work = container_of(work, struct btrfs_async_delayed_work, work);
1342 delayed_root = async_work->delayed_root;
1344 path = btrfs_alloc_path();
1349 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND / 2)
1352 delayed_node = btrfs_first_prepared_delayed_node(delayed_root);
1356 path->leave_spinning = 1;
1357 root = delayed_node->root;
1359 trans = btrfs_join_transaction(root);
1363 block_rsv = trans->block_rsv;
1364 trans->block_rsv = &root->fs_info->delayed_block_rsv;
1366 __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1368 trans->block_rsv = block_rsv;
1369 btrfs_end_transaction(trans, root);
1370 btrfs_btree_balance_dirty_nodelay(root);
1373 btrfs_release_path(path);
1376 btrfs_release_prepared_delayed_node(delayed_node);
1377 if (async_work->nr == 0 || total_done < async_work->nr)
1381 btrfs_free_path(path);
1383 wake_up(&delayed_root->wait);
1388 static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root,
1389 struct btrfs_fs_info *fs_info, int nr)
1391 struct btrfs_async_delayed_work *async_work;
1393 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1396 async_work = kmalloc(sizeof(*async_work), GFP_NOFS);
1400 async_work->delayed_root = delayed_root;
1401 btrfs_init_work(&async_work->work, btrfs_delayed_meta_helper,
1402 btrfs_async_run_delayed_root, NULL, NULL);
1403 async_work->nr = nr;
1405 btrfs_queue_work(fs_info->delayed_workers, &async_work->work);
1409 void btrfs_assert_delayed_root_empty(struct btrfs_root *root)
1411 struct btrfs_delayed_root *delayed_root;
1412 delayed_root = btrfs_get_delayed_root(root);
1413 WARN_ON(btrfs_first_delayed_node(delayed_root));
1416 static int could_end_wait(struct btrfs_delayed_root *delayed_root, int seq)
1418 int val = atomic_read(&delayed_root->items_seq);
1420 if (val < seq || val >= seq + BTRFS_DELAYED_BATCH)
1423 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1429 void btrfs_balance_delayed_items(struct btrfs_root *root)
1431 struct btrfs_delayed_root *delayed_root;
1432 struct btrfs_fs_info *fs_info = root->fs_info;
1434 delayed_root = btrfs_get_delayed_root(root);
1436 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1439 if (atomic_read(&delayed_root->items) >= BTRFS_DELAYED_WRITEBACK) {
1443 seq = atomic_read(&delayed_root->items_seq);
1445 ret = btrfs_wq_run_delayed_node(delayed_root, fs_info, 0);
1449 wait_event_interruptible(delayed_root->wait,
1450 could_end_wait(delayed_root, seq));
1454 btrfs_wq_run_delayed_node(delayed_root, fs_info, BTRFS_DELAYED_BATCH);
1457 /* Will return 0 or -ENOMEM */
1458 int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans,
1459 struct btrfs_root *root, const char *name,
1460 int name_len, struct inode *dir,
1461 struct btrfs_disk_key *disk_key, u8 type,
1464 struct btrfs_delayed_node *delayed_node;
1465 struct btrfs_delayed_item *delayed_item;
1466 struct btrfs_dir_item *dir_item;
1469 delayed_node = btrfs_get_or_create_delayed_node(dir);
1470 if (IS_ERR(delayed_node))
1471 return PTR_ERR(delayed_node);
1473 delayed_item = btrfs_alloc_delayed_item(sizeof(*dir_item) + name_len);
1474 if (!delayed_item) {
1479 delayed_item->key.objectid = btrfs_ino(dir);
1480 delayed_item->key.type = BTRFS_DIR_INDEX_KEY;
1481 delayed_item->key.offset = index;
1483 dir_item = (struct btrfs_dir_item *)delayed_item->data;
1484 dir_item->location = *disk_key;
1485 btrfs_set_stack_dir_transid(dir_item, trans->transid);
1486 btrfs_set_stack_dir_data_len(dir_item, 0);
1487 btrfs_set_stack_dir_name_len(dir_item, name_len);
1488 btrfs_set_stack_dir_type(dir_item, type);
1489 memcpy((char *)(dir_item + 1), name, name_len);
1491 ret = btrfs_delayed_item_reserve_metadata(trans, root, delayed_item);
1493 * we have reserved enough space when we start a new transaction,
1494 * so reserving metadata failure is impossible
1499 mutex_lock(&delayed_node->mutex);
1500 ret = __btrfs_add_delayed_insertion_item(delayed_node, delayed_item);
1501 if (unlikely(ret)) {
1502 btrfs_err(root->fs_info, "err add delayed dir index item(name: %.*s) "
1503 "into the insertion tree of the delayed node"
1504 "(root id: %llu, inode id: %llu, errno: %d)",
1505 name_len, name, delayed_node->root->objectid,
1506 delayed_node->inode_id, ret);
1509 mutex_unlock(&delayed_node->mutex);
1512 btrfs_release_delayed_node(delayed_node);
1516 static int btrfs_delete_delayed_insertion_item(struct btrfs_root *root,
1517 struct btrfs_delayed_node *node,
1518 struct btrfs_key *key)
1520 struct btrfs_delayed_item *item;
1522 mutex_lock(&node->mutex);
1523 item = __btrfs_lookup_delayed_insertion_item(node, key);
1525 mutex_unlock(&node->mutex);
1529 btrfs_delayed_item_release_metadata(root, item);
1530 btrfs_release_delayed_item(item);
1531 mutex_unlock(&node->mutex);
1535 int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans,
1536 struct btrfs_root *root, struct inode *dir,
1539 struct btrfs_delayed_node *node;
1540 struct btrfs_delayed_item *item;
1541 struct btrfs_key item_key;
1544 node = btrfs_get_or_create_delayed_node(dir);
1546 return PTR_ERR(node);
1548 item_key.objectid = btrfs_ino(dir);
1549 item_key.type = BTRFS_DIR_INDEX_KEY;
1550 item_key.offset = index;
1552 ret = btrfs_delete_delayed_insertion_item(root, node, &item_key);
1556 item = btrfs_alloc_delayed_item(0);
1562 item->key = item_key;
1564 ret = btrfs_delayed_item_reserve_metadata(trans, root, item);
1566 * we have reserved enough space when we start a new transaction,
1567 * so reserving metadata failure is impossible.
1571 mutex_lock(&node->mutex);
1572 ret = __btrfs_add_delayed_deletion_item(node, item);
1573 if (unlikely(ret)) {
1574 btrfs_err(root->fs_info, "err add delayed dir index item(index: %llu) "
1575 "into the deletion tree of the delayed node"
1576 "(root id: %llu, inode id: %llu, errno: %d)",
1577 index, node->root->objectid, node->inode_id,
1581 mutex_unlock(&node->mutex);
1583 btrfs_release_delayed_node(node);
1587 int btrfs_inode_delayed_dir_index_count(struct inode *inode)
1589 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
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.
1599 if (!delayed_node->index_cnt) {
1600 btrfs_release_delayed_node(delayed_node);
1604 BTRFS_I(inode)->index_cnt = delayed_node->index_cnt;
1605 btrfs_release_delayed_node(delayed_node);
1609 bool btrfs_readdir_get_delayed_items(struct inode *inode,
1610 struct list_head *ins_list,
1611 struct list_head *del_list)
1613 struct btrfs_delayed_node *delayed_node;
1614 struct btrfs_delayed_item *item;
1616 delayed_node = btrfs_get_delayed_node(inode);
1621 * We can only do one readdir with delayed items at a time because of
1622 * item->readdir_list.
1624 inode_unlock_shared(inode);
1627 mutex_lock(&delayed_node->mutex);
1628 item = __btrfs_first_delayed_insertion_item(delayed_node);
1630 atomic_inc(&item->refs);
1631 list_add_tail(&item->readdir_list, ins_list);
1632 item = __btrfs_next_delayed_item(item);
1635 item = __btrfs_first_delayed_deletion_item(delayed_node);
1637 atomic_inc(&item->refs);
1638 list_add_tail(&item->readdir_list, del_list);
1639 item = __btrfs_next_delayed_item(item);
1641 mutex_unlock(&delayed_node->mutex);
1643 * This delayed node is still cached in the btrfs inode, so refs
1644 * must be > 1 now, and we needn't check it is going to be freed
1647 * Besides that, this function is used to read dir, we do not
1648 * insert/delete delayed items in this period. So we also needn't
1649 * requeue or dequeue this delayed node.
1651 atomic_dec(&delayed_node->refs);
1656 void btrfs_readdir_put_delayed_items(struct inode *inode,
1657 struct list_head *ins_list,
1658 struct list_head *del_list)
1660 struct btrfs_delayed_item *curr, *next;
1662 list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1663 list_del(&curr->readdir_list);
1664 if (atomic_dec_and_test(&curr->refs))
1668 list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1669 list_del(&curr->readdir_list);
1670 if (atomic_dec_and_test(&curr->refs))
1675 * The VFS is going to do up_read(), so we need to downgrade back to a
1678 downgrade_write(&inode->i_rwsem);
1681 int btrfs_should_delete_dir_index(struct list_head *del_list,
1684 struct btrfs_delayed_item *curr, *next;
1687 if (list_empty(del_list))
1690 list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1691 if (curr->key.offset > index)
1694 list_del(&curr->readdir_list);
1695 ret = (curr->key.offset == index);
1697 if (atomic_dec_and_test(&curr->refs))
1709 * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree
1712 int btrfs_readdir_delayed_dir_index(struct dir_context *ctx,
1713 struct list_head *ins_list, bool *emitted)
1715 struct btrfs_dir_item *di;
1716 struct btrfs_delayed_item *curr, *next;
1717 struct btrfs_key location;
1721 unsigned char d_type;
1723 if (list_empty(ins_list))
1727 * Changing the data of the delayed item is impossible. So
1728 * we needn't lock them. And we have held i_mutex of the
1729 * directory, nobody can delete any directory indexes now.
1731 list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1732 list_del(&curr->readdir_list);
1734 if (curr->key.offset < ctx->pos) {
1735 if (atomic_dec_and_test(&curr->refs))
1740 ctx->pos = curr->key.offset;
1742 di = (struct btrfs_dir_item *)curr->data;
1743 name = (char *)(di + 1);
1744 name_len = btrfs_stack_dir_name_len(di);
1746 d_type = btrfs_filetype_table[di->type];
1747 btrfs_disk_key_to_cpu(&location, &di->location);
1749 over = !dir_emit(ctx, name, name_len,
1750 location.objectid, d_type);
1752 if (atomic_dec_and_test(&curr->refs))
1762 static void fill_stack_inode_item(struct btrfs_trans_handle *trans,
1763 struct btrfs_inode_item *inode_item,
1764 struct inode *inode)
1766 btrfs_set_stack_inode_uid(inode_item, i_uid_read(inode));
1767 btrfs_set_stack_inode_gid(inode_item, i_gid_read(inode));
1768 btrfs_set_stack_inode_size(inode_item, BTRFS_I(inode)->disk_i_size);
1769 btrfs_set_stack_inode_mode(inode_item, inode->i_mode);
1770 btrfs_set_stack_inode_nlink(inode_item, inode->i_nlink);
1771 btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(inode));
1772 btrfs_set_stack_inode_generation(inode_item,
1773 BTRFS_I(inode)->generation);
1774 btrfs_set_stack_inode_sequence(inode_item, inode->i_version);
1775 btrfs_set_stack_inode_transid(inode_item, trans->transid);
1776 btrfs_set_stack_inode_rdev(inode_item, inode->i_rdev);
1777 btrfs_set_stack_inode_flags(inode_item, BTRFS_I(inode)->flags);
1778 btrfs_set_stack_inode_block_group(inode_item, 0);
1780 btrfs_set_stack_timespec_sec(&inode_item->atime,
1781 inode->i_atime.tv_sec);
1782 btrfs_set_stack_timespec_nsec(&inode_item->atime,
1783 inode->i_atime.tv_nsec);
1785 btrfs_set_stack_timespec_sec(&inode_item->mtime,
1786 inode->i_mtime.tv_sec);
1787 btrfs_set_stack_timespec_nsec(&inode_item->mtime,
1788 inode->i_mtime.tv_nsec);
1790 btrfs_set_stack_timespec_sec(&inode_item->ctime,
1791 inode->i_ctime.tv_sec);
1792 btrfs_set_stack_timespec_nsec(&inode_item->ctime,
1793 inode->i_ctime.tv_nsec);
1795 btrfs_set_stack_timespec_sec(&inode_item->otime,
1796 BTRFS_I(inode)->i_otime.tv_sec);
1797 btrfs_set_stack_timespec_nsec(&inode_item->otime,
1798 BTRFS_I(inode)->i_otime.tv_nsec);
1801 int btrfs_fill_inode(struct inode *inode, u32 *rdev)
1803 struct btrfs_delayed_node *delayed_node;
1804 struct btrfs_inode_item *inode_item;
1806 delayed_node = btrfs_get_delayed_node(inode);
1810 mutex_lock(&delayed_node->mutex);
1811 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1812 mutex_unlock(&delayed_node->mutex);
1813 btrfs_release_delayed_node(delayed_node);
1817 inode_item = &delayed_node->inode_item;
1819 i_uid_write(inode, btrfs_stack_inode_uid(inode_item));
1820 i_gid_write(inode, btrfs_stack_inode_gid(inode_item));
1821 btrfs_i_size_write(inode, btrfs_stack_inode_size(inode_item));
1822 inode->i_mode = btrfs_stack_inode_mode(inode_item);
1823 set_nlink(inode, btrfs_stack_inode_nlink(inode_item));
1824 inode_set_bytes(inode, btrfs_stack_inode_nbytes(inode_item));
1825 BTRFS_I(inode)->generation = btrfs_stack_inode_generation(inode_item);
1826 BTRFS_I(inode)->last_trans = btrfs_stack_inode_transid(inode_item);
1828 inode->i_version = btrfs_stack_inode_sequence(inode_item);
1830 *rdev = btrfs_stack_inode_rdev(inode_item);
1831 BTRFS_I(inode)->flags = btrfs_stack_inode_flags(inode_item);
1833 inode->i_atime.tv_sec = btrfs_stack_timespec_sec(&inode_item->atime);
1834 inode->i_atime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->atime);
1836 inode->i_mtime.tv_sec = btrfs_stack_timespec_sec(&inode_item->mtime);
1837 inode->i_mtime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->mtime);
1839 inode->i_ctime.tv_sec = btrfs_stack_timespec_sec(&inode_item->ctime);
1840 inode->i_ctime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->ctime);
1842 BTRFS_I(inode)->i_otime.tv_sec =
1843 btrfs_stack_timespec_sec(&inode_item->otime);
1844 BTRFS_I(inode)->i_otime.tv_nsec =
1845 btrfs_stack_timespec_nsec(&inode_item->otime);
1847 inode->i_generation = BTRFS_I(inode)->generation;
1848 BTRFS_I(inode)->index_cnt = (u64)-1;
1850 mutex_unlock(&delayed_node->mutex);
1851 btrfs_release_delayed_node(delayed_node);
1855 int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans,
1856 struct btrfs_root *root, struct inode *inode)
1858 struct btrfs_delayed_node *delayed_node;
1861 delayed_node = btrfs_get_or_create_delayed_node(inode);
1862 if (IS_ERR(delayed_node))
1863 return PTR_ERR(delayed_node);
1865 mutex_lock(&delayed_node->mutex);
1866 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1867 fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1871 ret = btrfs_delayed_inode_reserve_metadata(trans, root, inode,
1876 fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1877 set_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
1878 delayed_node->count++;
1879 atomic_inc(&root->fs_info->delayed_root->items);
1881 mutex_unlock(&delayed_node->mutex);
1882 btrfs_release_delayed_node(delayed_node);
1886 int btrfs_delayed_delete_inode_ref(struct inode *inode)
1888 struct btrfs_delayed_node *delayed_node;
1891 * we don't do delayed inode updates during log recovery because it
1892 * leads to enospc problems. This means we also can't do
1893 * delayed inode refs
1895 if (BTRFS_I(inode)->root->fs_info->log_root_recovering)
1898 delayed_node = btrfs_get_or_create_delayed_node(inode);
1899 if (IS_ERR(delayed_node))
1900 return PTR_ERR(delayed_node);
1903 * We don't reserve space for inode ref deletion is because:
1904 * - We ONLY do async inode ref deletion for the inode who has only
1905 * one link(i_nlink == 1), it means there is only one inode ref.
1906 * And in most case, the inode ref and the inode item are in the
1907 * same leaf, and we will deal with them at the same time.
1908 * Since we are sure we will reserve the space for the inode item,
1909 * it is unnecessary to reserve space for inode ref deletion.
1910 * - If the inode ref and the inode item are not in the same leaf,
1911 * We also needn't worry about enospc problem, because we reserve
1912 * much more space for the inode update than it needs.
1913 * - At the worst, we can steal some space from the global reservation.
1916 mutex_lock(&delayed_node->mutex);
1917 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1920 set_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
1921 delayed_node->count++;
1922 atomic_inc(&BTRFS_I(inode)->root->fs_info->delayed_root->items);
1924 mutex_unlock(&delayed_node->mutex);
1925 btrfs_release_delayed_node(delayed_node);
1929 static void __btrfs_kill_delayed_node(struct btrfs_delayed_node *delayed_node)
1931 struct btrfs_root *root = delayed_node->root;
1932 struct btrfs_delayed_item *curr_item, *prev_item;
1934 mutex_lock(&delayed_node->mutex);
1935 curr_item = __btrfs_first_delayed_insertion_item(delayed_node);
1937 btrfs_delayed_item_release_metadata(root, curr_item);
1938 prev_item = curr_item;
1939 curr_item = __btrfs_next_delayed_item(prev_item);
1940 btrfs_release_delayed_item(prev_item);
1943 curr_item = __btrfs_first_delayed_deletion_item(delayed_node);
1945 btrfs_delayed_item_release_metadata(root, curr_item);
1946 prev_item = curr_item;
1947 curr_item = __btrfs_next_delayed_item(prev_item);
1948 btrfs_release_delayed_item(prev_item);
1951 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1952 btrfs_release_delayed_iref(delayed_node);
1954 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1955 btrfs_delayed_inode_release_metadata(root, delayed_node);
1956 btrfs_release_delayed_inode(delayed_node);
1958 mutex_unlock(&delayed_node->mutex);
1961 void btrfs_kill_delayed_inode_items(struct inode *inode)
1963 struct btrfs_delayed_node *delayed_node;
1965 delayed_node = btrfs_get_delayed_node(inode);
1969 __btrfs_kill_delayed_node(delayed_node);
1970 btrfs_release_delayed_node(delayed_node);
1973 void btrfs_kill_all_delayed_nodes(struct btrfs_root *root)
1976 struct btrfs_delayed_node *delayed_nodes[8];
1980 spin_lock(&root->inode_lock);
1981 n = radix_tree_gang_lookup(&root->delayed_nodes_tree,
1982 (void **)delayed_nodes, inode_id,
1983 ARRAY_SIZE(delayed_nodes));
1985 spin_unlock(&root->inode_lock);
1989 inode_id = delayed_nodes[n - 1]->inode_id + 1;
1991 for (i = 0; i < n; i++)
1992 atomic_inc(&delayed_nodes[i]->refs);
1993 spin_unlock(&root->inode_lock);
1995 for (i = 0; i < n; i++) {
1996 __btrfs_kill_delayed_node(delayed_nodes[i]);
1997 btrfs_release_delayed_node(delayed_nodes[i]);
2002 void btrfs_destroy_delayed_inodes(struct btrfs_root *root)
2004 struct btrfs_delayed_root *delayed_root;
2005 struct btrfs_delayed_node *curr_node, *prev_node;
2007 delayed_root = btrfs_get_delayed_root(root);
2009 curr_node = btrfs_first_delayed_node(delayed_root);
2011 __btrfs_kill_delayed_node(curr_node);
2013 prev_node = curr_node;
2014 curr_node = btrfs_next_delayed_node(curr_node);
2015 btrfs_release_delayed_node(prev_node);