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),
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 refcount_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 struct btrfs_delayed_node *btrfs_get_delayed_node(
76 struct btrfs_inode *btrfs_inode)
78 struct btrfs_root *root = btrfs_inode->root;
79 u64 ino = btrfs_ino(btrfs_inode);
80 struct btrfs_delayed_node *node;
82 node = READ_ONCE(btrfs_inode->delayed_node);
84 refcount_inc(&node->refs);
88 spin_lock(&root->inode_lock);
89 node = radix_tree_lookup(&root->delayed_nodes_tree, ino);
92 if (btrfs_inode->delayed_node) {
93 refcount_inc(&node->refs); /* can be accessed */
94 BUG_ON(btrfs_inode->delayed_node != node);
95 spin_unlock(&root->inode_lock);
100 * It's possible that we're racing into the middle of removing
101 * this node from the radix tree. In this case, the refcount
102 * was zero and it should never go back to one. Just return
103 * NULL like it was never in the radix at all; our release
104 * function is in the process of removing it.
106 * Some implementations of refcount_inc refuse to bump the
107 * refcount once it has hit zero. If we don't do this dance
108 * here, refcount_inc() may decide to just WARN_ONCE() instead
109 * of actually bumping the refcount.
111 * If this node is properly in the radix, we want to bump the
112 * refcount twice, once for the inode and once for this get
115 if (refcount_inc_not_zero(&node->refs)) {
116 refcount_inc(&node->refs);
117 btrfs_inode->delayed_node = node;
122 spin_unlock(&root->inode_lock);
125 spin_unlock(&root->inode_lock);
130 /* Will return either the node or PTR_ERR(-ENOMEM) */
131 static struct btrfs_delayed_node *btrfs_get_or_create_delayed_node(
132 struct btrfs_inode *btrfs_inode)
134 struct btrfs_delayed_node *node;
135 struct btrfs_root *root = btrfs_inode->root;
136 u64 ino = btrfs_ino(btrfs_inode);
140 node = btrfs_get_delayed_node(btrfs_inode);
144 node = kmem_cache_zalloc(delayed_node_cache, GFP_NOFS);
146 return ERR_PTR(-ENOMEM);
147 btrfs_init_delayed_node(node, root, ino);
149 /* cached in the btrfs inode and can be accessed */
150 refcount_set(&node->refs, 2);
152 ret = radix_tree_preload(GFP_NOFS);
154 kmem_cache_free(delayed_node_cache, node);
158 spin_lock(&root->inode_lock);
159 ret = radix_tree_insert(&root->delayed_nodes_tree, ino, node);
160 if (ret == -EEXIST) {
161 spin_unlock(&root->inode_lock);
162 kmem_cache_free(delayed_node_cache, node);
163 radix_tree_preload_end();
166 btrfs_inode->delayed_node = node;
167 spin_unlock(&root->inode_lock);
168 radix_tree_preload_end();
174 * Call it when holding delayed_node->mutex
176 * If mod = 1, add this node into the prepared list.
178 static void btrfs_queue_delayed_node(struct btrfs_delayed_root *root,
179 struct btrfs_delayed_node *node,
182 spin_lock(&root->lock);
183 if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
184 if (!list_empty(&node->p_list))
185 list_move_tail(&node->p_list, &root->prepare_list);
187 list_add_tail(&node->p_list, &root->prepare_list);
189 list_add_tail(&node->n_list, &root->node_list);
190 list_add_tail(&node->p_list, &root->prepare_list);
191 refcount_inc(&node->refs); /* inserted into list */
193 set_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
195 spin_unlock(&root->lock);
198 /* Call it when holding delayed_node->mutex */
199 static void btrfs_dequeue_delayed_node(struct btrfs_delayed_root *root,
200 struct btrfs_delayed_node *node)
202 spin_lock(&root->lock);
203 if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
205 refcount_dec(&node->refs); /* not in the list */
206 list_del_init(&node->n_list);
207 if (!list_empty(&node->p_list))
208 list_del_init(&node->p_list);
209 clear_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
211 spin_unlock(&root->lock);
214 static struct btrfs_delayed_node *btrfs_first_delayed_node(
215 struct btrfs_delayed_root *delayed_root)
218 struct btrfs_delayed_node *node = NULL;
220 spin_lock(&delayed_root->lock);
221 if (list_empty(&delayed_root->node_list))
224 p = delayed_root->node_list.next;
225 node = list_entry(p, struct btrfs_delayed_node, n_list);
226 refcount_inc(&node->refs);
228 spin_unlock(&delayed_root->lock);
233 static struct btrfs_delayed_node *btrfs_next_delayed_node(
234 struct btrfs_delayed_node *node)
236 struct btrfs_delayed_root *delayed_root;
238 struct btrfs_delayed_node *next = NULL;
240 delayed_root = node->root->fs_info->delayed_root;
241 spin_lock(&delayed_root->lock);
242 if (!test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
243 /* not in the list */
244 if (list_empty(&delayed_root->node_list))
246 p = delayed_root->node_list.next;
247 } else if (list_is_last(&node->n_list, &delayed_root->node_list))
250 p = node->n_list.next;
252 next = list_entry(p, struct btrfs_delayed_node, n_list);
253 refcount_inc(&next->refs);
255 spin_unlock(&delayed_root->lock);
260 static void __btrfs_release_delayed_node(
261 struct btrfs_delayed_node *delayed_node,
264 struct btrfs_delayed_root *delayed_root;
269 delayed_root = delayed_node->root->fs_info->delayed_root;
271 mutex_lock(&delayed_node->mutex);
272 if (delayed_node->count)
273 btrfs_queue_delayed_node(delayed_root, delayed_node, mod);
275 btrfs_dequeue_delayed_node(delayed_root, delayed_node);
276 mutex_unlock(&delayed_node->mutex);
278 if (refcount_dec_and_test(&delayed_node->refs)) {
279 struct btrfs_root *root = delayed_node->root;
281 spin_lock(&root->inode_lock);
283 * Once our refcount goes to zero, nobody is allowed to bump it
284 * back up. We can delete it now.
286 ASSERT(refcount_read(&delayed_node->refs) == 0);
287 radix_tree_delete(&root->delayed_nodes_tree,
288 delayed_node->inode_id);
289 spin_unlock(&root->inode_lock);
290 kmem_cache_free(delayed_node_cache, delayed_node);
294 static inline void btrfs_release_delayed_node(struct btrfs_delayed_node *node)
296 __btrfs_release_delayed_node(node, 0);
299 static struct btrfs_delayed_node *btrfs_first_prepared_delayed_node(
300 struct btrfs_delayed_root *delayed_root)
303 struct btrfs_delayed_node *node = NULL;
305 spin_lock(&delayed_root->lock);
306 if (list_empty(&delayed_root->prepare_list))
309 p = delayed_root->prepare_list.next;
311 node = list_entry(p, struct btrfs_delayed_node, p_list);
312 refcount_inc(&node->refs);
314 spin_unlock(&delayed_root->lock);
319 static inline void btrfs_release_prepared_delayed_node(
320 struct btrfs_delayed_node *node)
322 __btrfs_release_delayed_node(node, 1);
325 static struct btrfs_delayed_item *btrfs_alloc_delayed_item(u32 data_len)
327 struct btrfs_delayed_item *item;
328 item = kmalloc(sizeof(*item) + data_len, GFP_NOFS);
330 item->data_len = data_len;
331 item->ins_or_del = 0;
332 item->bytes_reserved = 0;
333 item->delayed_node = NULL;
334 refcount_set(&item->refs, 1);
340 * __btrfs_lookup_delayed_item - look up the delayed item by key
341 * @delayed_node: pointer to the delayed node
342 * @key: the key to look up
343 * @prev: used to store the prev item if the right item isn't found
344 * @next: used to store the next item if the right item isn't found
346 * Note: if we don't find the right item, we will return the prev item and
349 static struct btrfs_delayed_item *__btrfs_lookup_delayed_item(
350 struct rb_root *root,
351 struct btrfs_key *key,
352 struct btrfs_delayed_item **prev,
353 struct btrfs_delayed_item **next)
355 struct rb_node *node, *prev_node = NULL;
356 struct btrfs_delayed_item *delayed_item = NULL;
359 node = root->rb_node;
362 delayed_item = rb_entry(node, struct btrfs_delayed_item,
365 ret = btrfs_comp_cpu_keys(&delayed_item->key, key);
367 node = node->rb_right;
369 node = node->rb_left;
378 *prev = delayed_item;
379 else if ((node = rb_prev(prev_node)) != NULL) {
380 *prev = rb_entry(node, struct btrfs_delayed_item,
390 *next = delayed_item;
391 else if ((node = rb_next(prev_node)) != NULL) {
392 *next = rb_entry(node, struct btrfs_delayed_item,
400 static struct btrfs_delayed_item *__btrfs_lookup_delayed_insertion_item(
401 struct btrfs_delayed_node *delayed_node,
402 struct btrfs_key *key)
404 return __btrfs_lookup_delayed_item(&delayed_node->ins_root, key,
408 static int __btrfs_add_delayed_item(struct btrfs_delayed_node *delayed_node,
409 struct btrfs_delayed_item *ins,
412 struct rb_node **p, *node;
413 struct rb_node *parent_node = NULL;
414 struct rb_root *root;
415 struct btrfs_delayed_item *item;
418 if (action == BTRFS_DELAYED_INSERTION_ITEM)
419 root = &delayed_node->ins_root;
420 else if (action == BTRFS_DELAYED_DELETION_ITEM)
421 root = &delayed_node->del_root;
425 node = &ins->rb_node;
429 item = rb_entry(parent_node, struct btrfs_delayed_item,
432 cmp = btrfs_comp_cpu_keys(&item->key, &ins->key);
441 rb_link_node(node, parent_node, p);
442 rb_insert_color(node, root);
443 ins->delayed_node = delayed_node;
444 ins->ins_or_del = action;
446 if (ins->key.type == BTRFS_DIR_INDEX_KEY &&
447 action == BTRFS_DELAYED_INSERTION_ITEM &&
448 ins->key.offset >= delayed_node->index_cnt)
449 delayed_node->index_cnt = ins->key.offset + 1;
451 delayed_node->count++;
452 atomic_inc(&delayed_node->root->fs_info->delayed_root->items);
456 static int __btrfs_add_delayed_insertion_item(struct btrfs_delayed_node *node,
457 struct btrfs_delayed_item *item)
459 return __btrfs_add_delayed_item(node, item,
460 BTRFS_DELAYED_INSERTION_ITEM);
463 static int __btrfs_add_delayed_deletion_item(struct btrfs_delayed_node *node,
464 struct btrfs_delayed_item *item)
466 return __btrfs_add_delayed_item(node, item,
467 BTRFS_DELAYED_DELETION_ITEM);
470 static void finish_one_item(struct btrfs_delayed_root *delayed_root)
472 int seq = atomic_inc_return(&delayed_root->items_seq);
475 * atomic_dec_return implies a barrier for waitqueue_active
477 if ((atomic_dec_return(&delayed_root->items) <
478 BTRFS_DELAYED_BACKGROUND || seq % BTRFS_DELAYED_BATCH == 0) &&
479 waitqueue_active(&delayed_root->wait))
480 wake_up(&delayed_root->wait);
483 static void __btrfs_remove_delayed_item(struct btrfs_delayed_item *delayed_item)
485 struct rb_root *root;
486 struct btrfs_delayed_root *delayed_root;
488 delayed_root = delayed_item->delayed_node->root->fs_info->delayed_root;
490 BUG_ON(!delayed_root);
491 BUG_ON(delayed_item->ins_or_del != BTRFS_DELAYED_DELETION_ITEM &&
492 delayed_item->ins_or_del != BTRFS_DELAYED_INSERTION_ITEM);
494 if (delayed_item->ins_or_del == BTRFS_DELAYED_INSERTION_ITEM)
495 root = &delayed_item->delayed_node->ins_root;
497 root = &delayed_item->delayed_node->del_root;
499 rb_erase(&delayed_item->rb_node, root);
500 delayed_item->delayed_node->count--;
502 finish_one_item(delayed_root);
505 static void btrfs_release_delayed_item(struct btrfs_delayed_item *item)
508 __btrfs_remove_delayed_item(item);
509 if (refcount_dec_and_test(&item->refs))
514 static struct btrfs_delayed_item *__btrfs_first_delayed_insertion_item(
515 struct btrfs_delayed_node *delayed_node)
518 struct btrfs_delayed_item *item = NULL;
520 p = rb_first(&delayed_node->ins_root);
522 item = rb_entry(p, struct btrfs_delayed_item, rb_node);
527 static struct btrfs_delayed_item *__btrfs_first_delayed_deletion_item(
528 struct btrfs_delayed_node *delayed_node)
531 struct btrfs_delayed_item *item = NULL;
533 p = rb_first(&delayed_node->del_root);
535 item = rb_entry(p, struct btrfs_delayed_item, rb_node);
540 static struct btrfs_delayed_item *__btrfs_next_delayed_item(
541 struct btrfs_delayed_item *item)
544 struct btrfs_delayed_item *next = NULL;
546 p = rb_next(&item->rb_node);
548 next = rb_entry(p, struct btrfs_delayed_item, rb_node);
553 static int btrfs_delayed_item_reserve_metadata(struct btrfs_trans_handle *trans,
554 struct btrfs_fs_info *fs_info,
555 struct btrfs_delayed_item *item)
557 struct btrfs_block_rsv *src_rsv;
558 struct btrfs_block_rsv *dst_rsv;
562 if (!trans->bytes_reserved)
565 src_rsv = trans->block_rsv;
566 dst_rsv = &fs_info->delayed_block_rsv;
568 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
569 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1);
571 trace_btrfs_space_reservation(fs_info, "delayed_item",
574 item->bytes_reserved = num_bytes;
580 static void btrfs_delayed_item_release_metadata(struct btrfs_fs_info *fs_info,
581 struct btrfs_delayed_item *item)
583 struct btrfs_block_rsv *rsv;
585 if (!item->bytes_reserved)
588 rsv = &fs_info->delayed_block_rsv;
589 trace_btrfs_space_reservation(fs_info, "delayed_item",
590 item->key.objectid, item->bytes_reserved,
592 btrfs_block_rsv_release(fs_info, rsv,
593 item->bytes_reserved);
596 static int btrfs_delayed_inode_reserve_metadata(
597 struct btrfs_trans_handle *trans,
598 struct btrfs_root *root,
599 struct btrfs_inode *inode,
600 struct btrfs_delayed_node *node)
602 struct btrfs_fs_info *fs_info = root->fs_info;
603 struct btrfs_block_rsv *src_rsv;
604 struct btrfs_block_rsv *dst_rsv;
608 src_rsv = trans->block_rsv;
609 dst_rsv = &fs_info->delayed_block_rsv;
611 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
614 * btrfs_dirty_inode will update the inode under btrfs_join_transaction
615 * which doesn't reserve space for speed. This is a problem since we
616 * still need to reserve space for this update, so try to reserve the
619 * Now if src_rsv == delalloc_block_rsv we'll let it just steal since
620 * we always reserve enough to update the inode item.
622 if (!src_rsv || (!trans->bytes_reserved &&
623 src_rsv->type != BTRFS_BLOCK_RSV_DELALLOC)) {
624 ret = btrfs_block_rsv_add(root, dst_rsv, num_bytes,
625 BTRFS_RESERVE_NO_FLUSH);
627 * Since we're under a transaction reserve_metadata_bytes could
628 * try to commit the transaction which will make it return
629 * EAGAIN to make us stop the transaction we have, so return
630 * ENOSPC instead so that btrfs_dirty_inode knows what to do.
635 node->bytes_reserved = num_bytes;
636 trace_btrfs_space_reservation(fs_info,
644 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1);
646 trace_btrfs_space_reservation(fs_info, "delayed_inode",
647 btrfs_ino(inode), num_bytes, 1);
648 node->bytes_reserved = num_bytes;
654 static void btrfs_delayed_inode_release_metadata(struct btrfs_fs_info *fs_info,
655 struct btrfs_delayed_node *node)
657 struct btrfs_block_rsv *rsv;
659 if (!node->bytes_reserved)
662 rsv = &fs_info->delayed_block_rsv;
663 trace_btrfs_space_reservation(fs_info, "delayed_inode",
664 node->inode_id, node->bytes_reserved, 0);
665 btrfs_block_rsv_release(fs_info, rsv,
666 node->bytes_reserved);
667 node->bytes_reserved = 0;
671 * This helper will insert some continuous items into the same leaf according
672 * to the free space of the leaf.
674 static int btrfs_batch_insert_items(struct btrfs_root *root,
675 struct btrfs_path *path,
676 struct btrfs_delayed_item *item)
678 struct btrfs_fs_info *fs_info = root->fs_info;
679 struct btrfs_delayed_item *curr, *next;
681 int total_data_size = 0, total_size = 0;
682 struct extent_buffer *leaf;
684 struct btrfs_key *keys;
686 struct list_head head;
692 BUG_ON(!path->nodes[0]);
694 leaf = path->nodes[0];
695 free_space = btrfs_leaf_free_space(fs_info, leaf);
696 INIT_LIST_HEAD(&head);
702 * count the number of the continuous items that we can insert in batch
704 while (total_size + next->data_len + sizeof(struct btrfs_item) <=
706 total_data_size += next->data_len;
707 total_size += next->data_len + sizeof(struct btrfs_item);
708 list_add_tail(&next->tree_list, &head);
712 next = __btrfs_next_delayed_item(curr);
716 if (!btrfs_is_continuous_delayed_item(curr, next))
726 * we need allocate some memory space, but it might cause the task
727 * to sleep, so we set all locked nodes in the path to blocking locks
730 btrfs_set_path_blocking(path);
732 keys = kmalloc_array(nitems, sizeof(struct btrfs_key), GFP_NOFS);
738 data_size = kmalloc_array(nitems, sizeof(u32), GFP_NOFS);
744 /* get keys of all the delayed items */
746 list_for_each_entry(next, &head, tree_list) {
748 data_size[i] = next->data_len;
752 /* reset all the locked nodes in the patch to spinning locks. */
753 btrfs_clear_path_blocking(path, NULL, 0);
755 /* insert the keys of the items */
756 setup_items_for_insert(root, path, keys, data_size,
757 total_data_size, total_size, nitems);
759 /* insert the dir index items */
760 slot = path->slots[0];
761 list_for_each_entry_safe(curr, next, &head, tree_list) {
762 data_ptr = btrfs_item_ptr(leaf, slot, char);
763 write_extent_buffer(leaf, &curr->data,
764 (unsigned long)data_ptr,
768 btrfs_delayed_item_release_metadata(fs_info, curr);
770 list_del(&curr->tree_list);
771 btrfs_release_delayed_item(curr);
782 * This helper can just do simple insertion that needn't extend item for new
783 * data, such as directory name index insertion, inode insertion.
785 static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans,
786 struct btrfs_root *root,
787 struct btrfs_path *path,
788 struct btrfs_delayed_item *delayed_item)
790 struct btrfs_fs_info *fs_info = root->fs_info;
791 struct extent_buffer *leaf;
795 ret = btrfs_insert_empty_item(trans, root, path, &delayed_item->key,
796 delayed_item->data_len);
797 if (ret < 0 && ret != -EEXIST)
800 leaf = path->nodes[0];
802 ptr = btrfs_item_ptr(leaf, path->slots[0], char);
804 write_extent_buffer(leaf, delayed_item->data, (unsigned long)ptr,
805 delayed_item->data_len);
806 btrfs_mark_buffer_dirty(leaf);
808 btrfs_delayed_item_release_metadata(fs_info, delayed_item);
813 * we insert an item first, then if there are some continuous items, we try
814 * to insert those items into the same leaf.
816 static int btrfs_insert_delayed_items(struct btrfs_trans_handle *trans,
817 struct btrfs_path *path,
818 struct btrfs_root *root,
819 struct btrfs_delayed_node *node)
821 struct btrfs_delayed_item *curr, *prev;
825 mutex_lock(&node->mutex);
826 curr = __btrfs_first_delayed_insertion_item(node);
830 ret = btrfs_insert_delayed_item(trans, root, path, curr);
832 btrfs_release_path(path);
837 curr = __btrfs_next_delayed_item(prev);
838 if (curr && btrfs_is_continuous_delayed_item(prev, curr)) {
839 /* insert the continuous items into the same leaf */
841 btrfs_batch_insert_items(root, path, curr);
843 btrfs_release_delayed_item(prev);
844 btrfs_mark_buffer_dirty(path->nodes[0]);
846 btrfs_release_path(path);
847 mutex_unlock(&node->mutex);
851 mutex_unlock(&node->mutex);
855 static int btrfs_batch_delete_items(struct btrfs_trans_handle *trans,
856 struct btrfs_root *root,
857 struct btrfs_path *path,
858 struct btrfs_delayed_item *item)
860 struct btrfs_fs_info *fs_info = root->fs_info;
861 struct btrfs_delayed_item *curr, *next;
862 struct extent_buffer *leaf;
863 struct btrfs_key key;
864 struct list_head head;
865 int nitems, i, last_item;
868 BUG_ON(!path->nodes[0]);
870 leaf = path->nodes[0];
873 last_item = btrfs_header_nritems(leaf) - 1;
875 return -ENOENT; /* FIXME: Is errno suitable? */
878 INIT_LIST_HEAD(&head);
879 btrfs_item_key_to_cpu(leaf, &key, i);
882 * count the number of the dir index items that we can delete in batch
884 while (btrfs_comp_cpu_keys(&next->key, &key) == 0) {
885 list_add_tail(&next->tree_list, &head);
889 next = __btrfs_next_delayed_item(curr);
893 if (!btrfs_is_continuous_delayed_item(curr, next))
899 btrfs_item_key_to_cpu(leaf, &key, i);
905 ret = btrfs_del_items(trans, root, path, path->slots[0], nitems);
909 list_for_each_entry_safe(curr, next, &head, tree_list) {
910 btrfs_delayed_item_release_metadata(fs_info, curr);
911 list_del(&curr->tree_list);
912 btrfs_release_delayed_item(curr);
919 static int btrfs_delete_delayed_items(struct btrfs_trans_handle *trans,
920 struct btrfs_path *path,
921 struct btrfs_root *root,
922 struct btrfs_delayed_node *node)
924 struct btrfs_delayed_item *curr, *prev;
928 mutex_lock(&node->mutex);
929 curr = __btrfs_first_delayed_deletion_item(node);
933 ret = btrfs_search_slot(trans, root, &curr->key, path, -1, 1);
938 * can't find the item which the node points to, so this node
939 * is invalid, just drop it.
942 curr = __btrfs_next_delayed_item(prev);
943 btrfs_release_delayed_item(prev);
945 btrfs_release_path(path);
947 mutex_unlock(&node->mutex);
953 btrfs_batch_delete_items(trans, root, path, curr);
954 btrfs_release_path(path);
955 mutex_unlock(&node->mutex);
959 btrfs_release_path(path);
960 mutex_unlock(&node->mutex);
964 static void btrfs_release_delayed_inode(struct btrfs_delayed_node *delayed_node)
966 struct btrfs_delayed_root *delayed_root;
969 test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
970 BUG_ON(!delayed_node->root);
971 clear_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
972 delayed_node->count--;
974 delayed_root = delayed_node->root->fs_info->delayed_root;
975 finish_one_item(delayed_root);
979 static void btrfs_release_delayed_iref(struct btrfs_delayed_node *delayed_node)
981 struct btrfs_delayed_root *delayed_root;
983 ASSERT(delayed_node->root);
984 clear_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
985 delayed_node->count--;
987 delayed_root = delayed_node->root->fs_info->delayed_root;
988 finish_one_item(delayed_root);
991 static int __btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
992 struct btrfs_root *root,
993 struct btrfs_path *path,
994 struct btrfs_delayed_node *node)
996 struct btrfs_fs_info *fs_info = root->fs_info;
997 struct btrfs_key key;
998 struct btrfs_inode_item *inode_item;
999 struct extent_buffer *leaf;
1003 key.objectid = node->inode_id;
1004 key.type = BTRFS_INODE_ITEM_KEY;
1007 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1012 ret = btrfs_lookup_inode(trans, root, path, &key, mod);
1014 btrfs_release_path(path);
1016 } else if (ret < 0) {
1020 leaf = path->nodes[0];
1021 inode_item = btrfs_item_ptr(leaf, path->slots[0],
1022 struct btrfs_inode_item);
1023 write_extent_buffer(leaf, &node->inode_item, (unsigned long)inode_item,
1024 sizeof(struct btrfs_inode_item));
1025 btrfs_mark_buffer_dirty(leaf);
1027 if (!test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1031 if (path->slots[0] >= btrfs_header_nritems(leaf))
1034 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1035 if (key.objectid != node->inode_id)
1038 if (key.type != BTRFS_INODE_REF_KEY &&
1039 key.type != BTRFS_INODE_EXTREF_KEY)
1043 * Delayed iref deletion is for the inode who has only one link,
1044 * so there is only one iref. The case that several irefs are
1045 * in the same item doesn't exist.
1047 btrfs_del_item(trans, root, path);
1049 btrfs_release_delayed_iref(node);
1051 btrfs_release_path(path);
1053 btrfs_delayed_inode_release_metadata(fs_info, node);
1054 btrfs_release_delayed_inode(node);
1059 btrfs_release_path(path);
1061 key.type = BTRFS_INODE_EXTREF_KEY;
1063 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1069 leaf = path->nodes[0];
1074 static inline int btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1075 struct btrfs_root *root,
1076 struct btrfs_path *path,
1077 struct btrfs_delayed_node *node)
1081 mutex_lock(&node->mutex);
1082 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &node->flags)) {
1083 mutex_unlock(&node->mutex);
1087 ret = __btrfs_update_delayed_inode(trans, root, path, node);
1088 mutex_unlock(&node->mutex);
1093 __btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1094 struct btrfs_path *path,
1095 struct btrfs_delayed_node *node)
1099 ret = btrfs_insert_delayed_items(trans, path, node->root, node);
1103 ret = btrfs_delete_delayed_items(trans, path, node->root, node);
1107 ret = btrfs_update_delayed_inode(trans, node->root, path, node);
1112 * Called when committing the transaction.
1113 * Returns 0 on success.
1114 * Returns < 0 on error and returns with an aborted transaction with any
1115 * outstanding delayed items cleaned up.
1117 static int __btrfs_run_delayed_items(struct btrfs_trans_handle *trans,
1118 struct btrfs_fs_info *fs_info, int nr)
1120 struct btrfs_delayed_root *delayed_root;
1121 struct btrfs_delayed_node *curr_node, *prev_node;
1122 struct btrfs_path *path;
1123 struct btrfs_block_rsv *block_rsv;
1125 bool count = (nr > 0);
1130 path = btrfs_alloc_path();
1133 path->leave_spinning = 1;
1135 block_rsv = trans->block_rsv;
1136 trans->block_rsv = &fs_info->delayed_block_rsv;
1138 delayed_root = fs_info->delayed_root;
1140 curr_node = btrfs_first_delayed_node(delayed_root);
1141 while (curr_node && (!count || (count && nr--))) {
1142 ret = __btrfs_commit_inode_delayed_items(trans, path,
1145 btrfs_release_delayed_node(curr_node);
1147 btrfs_abort_transaction(trans, ret);
1151 prev_node = curr_node;
1152 curr_node = btrfs_next_delayed_node(curr_node);
1153 btrfs_release_delayed_node(prev_node);
1157 btrfs_release_delayed_node(curr_node);
1158 btrfs_free_path(path);
1159 trans->block_rsv = block_rsv;
1164 int btrfs_run_delayed_items(struct btrfs_trans_handle *trans,
1165 struct btrfs_fs_info *fs_info)
1167 return __btrfs_run_delayed_items(trans, fs_info, -1);
1170 int btrfs_run_delayed_items_nr(struct btrfs_trans_handle *trans,
1171 struct btrfs_fs_info *fs_info, int nr)
1173 return __btrfs_run_delayed_items(trans, fs_info, nr);
1176 int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1177 struct btrfs_inode *inode)
1179 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1180 struct btrfs_path *path;
1181 struct btrfs_block_rsv *block_rsv;
1187 mutex_lock(&delayed_node->mutex);
1188 if (!delayed_node->count) {
1189 mutex_unlock(&delayed_node->mutex);
1190 btrfs_release_delayed_node(delayed_node);
1193 mutex_unlock(&delayed_node->mutex);
1195 path = btrfs_alloc_path();
1197 btrfs_release_delayed_node(delayed_node);
1200 path->leave_spinning = 1;
1202 block_rsv = trans->block_rsv;
1203 trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv;
1205 ret = __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1207 btrfs_release_delayed_node(delayed_node);
1208 btrfs_free_path(path);
1209 trans->block_rsv = block_rsv;
1214 int btrfs_commit_inode_delayed_inode(struct btrfs_inode *inode)
1216 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1217 struct btrfs_trans_handle *trans;
1218 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1219 struct btrfs_path *path;
1220 struct btrfs_block_rsv *block_rsv;
1226 mutex_lock(&delayed_node->mutex);
1227 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1228 mutex_unlock(&delayed_node->mutex);
1229 btrfs_release_delayed_node(delayed_node);
1232 mutex_unlock(&delayed_node->mutex);
1234 trans = btrfs_join_transaction(delayed_node->root);
1235 if (IS_ERR(trans)) {
1236 ret = PTR_ERR(trans);
1240 path = btrfs_alloc_path();
1245 path->leave_spinning = 1;
1247 block_rsv = trans->block_rsv;
1248 trans->block_rsv = &fs_info->delayed_block_rsv;
1250 mutex_lock(&delayed_node->mutex);
1251 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags))
1252 ret = __btrfs_update_delayed_inode(trans, delayed_node->root,
1253 path, delayed_node);
1256 mutex_unlock(&delayed_node->mutex);
1258 btrfs_free_path(path);
1259 trans->block_rsv = block_rsv;
1261 btrfs_end_transaction(trans);
1262 btrfs_btree_balance_dirty(fs_info);
1264 btrfs_release_delayed_node(delayed_node);
1269 void btrfs_remove_delayed_node(struct btrfs_inode *inode)
1271 struct btrfs_delayed_node *delayed_node;
1273 delayed_node = READ_ONCE(inode->delayed_node);
1277 inode->delayed_node = NULL;
1278 btrfs_release_delayed_node(delayed_node);
1281 struct btrfs_async_delayed_work {
1282 struct btrfs_delayed_root *delayed_root;
1284 struct btrfs_work work;
1287 static void btrfs_async_run_delayed_root(struct btrfs_work *work)
1289 struct btrfs_async_delayed_work *async_work;
1290 struct btrfs_delayed_root *delayed_root;
1291 struct btrfs_trans_handle *trans;
1292 struct btrfs_path *path;
1293 struct btrfs_delayed_node *delayed_node = NULL;
1294 struct btrfs_root *root;
1295 struct btrfs_block_rsv *block_rsv;
1298 async_work = container_of(work, struct btrfs_async_delayed_work, work);
1299 delayed_root = async_work->delayed_root;
1301 path = btrfs_alloc_path();
1306 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND / 2)
1309 delayed_node = btrfs_first_prepared_delayed_node(delayed_root);
1313 path->leave_spinning = 1;
1314 root = delayed_node->root;
1316 trans = btrfs_join_transaction(root);
1320 block_rsv = trans->block_rsv;
1321 trans->block_rsv = &root->fs_info->delayed_block_rsv;
1323 __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1325 trans->block_rsv = block_rsv;
1326 btrfs_end_transaction(trans);
1327 btrfs_btree_balance_dirty_nodelay(root->fs_info);
1330 btrfs_release_path(path);
1333 btrfs_release_prepared_delayed_node(delayed_node);
1334 if ((async_work->nr == 0 && total_done < BTRFS_DELAYED_WRITEBACK) ||
1335 total_done < async_work->nr)
1339 btrfs_free_path(path);
1341 wake_up(&delayed_root->wait);
1346 static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root,
1347 struct btrfs_fs_info *fs_info, int nr)
1349 struct btrfs_async_delayed_work *async_work;
1351 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND ||
1352 btrfs_workqueue_normal_congested(fs_info->delayed_workers))
1355 async_work = kmalloc(sizeof(*async_work), GFP_NOFS);
1359 async_work->delayed_root = delayed_root;
1360 btrfs_init_work(&async_work->work, btrfs_delayed_meta_helper,
1361 btrfs_async_run_delayed_root, NULL, NULL);
1362 async_work->nr = nr;
1364 btrfs_queue_work(fs_info->delayed_workers, &async_work->work);
1368 void btrfs_assert_delayed_root_empty(struct btrfs_fs_info *fs_info)
1370 WARN_ON(btrfs_first_delayed_node(fs_info->delayed_root));
1373 static int could_end_wait(struct btrfs_delayed_root *delayed_root, int seq)
1375 int val = atomic_read(&delayed_root->items_seq);
1377 if (val < seq || val >= seq + BTRFS_DELAYED_BATCH)
1380 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1386 void btrfs_balance_delayed_items(struct btrfs_fs_info *fs_info)
1388 struct btrfs_delayed_root *delayed_root = fs_info->delayed_root;
1390 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1393 if (atomic_read(&delayed_root->items) >= BTRFS_DELAYED_WRITEBACK) {
1397 seq = atomic_read(&delayed_root->items_seq);
1399 ret = btrfs_wq_run_delayed_node(delayed_root, fs_info, 0);
1403 wait_event_interruptible(delayed_root->wait,
1404 could_end_wait(delayed_root, seq));
1408 btrfs_wq_run_delayed_node(delayed_root, fs_info, BTRFS_DELAYED_BATCH);
1411 /* Will return 0 or -ENOMEM */
1412 int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans,
1413 struct btrfs_fs_info *fs_info,
1414 const char *name, int name_len,
1415 struct btrfs_inode *dir,
1416 struct btrfs_disk_key *disk_key, u8 type,
1419 struct btrfs_delayed_node *delayed_node;
1420 struct btrfs_delayed_item *delayed_item;
1421 struct btrfs_dir_item *dir_item;
1424 delayed_node = btrfs_get_or_create_delayed_node(dir);
1425 if (IS_ERR(delayed_node))
1426 return PTR_ERR(delayed_node);
1428 delayed_item = btrfs_alloc_delayed_item(sizeof(*dir_item) + name_len);
1429 if (!delayed_item) {
1434 delayed_item->key.objectid = btrfs_ino(dir);
1435 delayed_item->key.type = BTRFS_DIR_INDEX_KEY;
1436 delayed_item->key.offset = index;
1438 dir_item = (struct btrfs_dir_item *)delayed_item->data;
1439 dir_item->location = *disk_key;
1440 btrfs_set_stack_dir_transid(dir_item, trans->transid);
1441 btrfs_set_stack_dir_data_len(dir_item, 0);
1442 btrfs_set_stack_dir_name_len(dir_item, name_len);
1443 btrfs_set_stack_dir_type(dir_item, type);
1444 memcpy((char *)(dir_item + 1), name, name_len);
1446 ret = btrfs_delayed_item_reserve_metadata(trans, fs_info, delayed_item);
1448 * we have reserved enough space when we start a new transaction,
1449 * so reserving metadata failure is impossible
1454 mutex_lock(&delayed_node->mutex);
1455 ret = __btrfs_add_delayed_insertion_item(delayed_node, delayed_item);
1456 if (unlikely(ret)) {
1458 "err add delayed dir index item(name: %.*s) into the insertion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
1459 name_len, name, delayed_node->root->objectid,
1460 delayed_node->inode_id, ret);
1463 mutex_unlock(&delayed_node->mutex);
1466 btrfs_release_delayed_node(delayed_node);
1470 static int btrfs_delete_delayed_insertion_item(struct btrfs_fs_info *fs_info,
1471 struct btrfs_delayed_node *node,
1472 struct btrfs_key *key)
1474 struct btrfs_delayed_item *item;
1476 mutex_lock(&node->mutex);
1477 item = __btrfs_lookup_delayed_insertion_item(node, key);
1479 mutex_unlock(&node->mutex);
1483 btrfs_delayed_item_release_metadata(fs_info, item);
1484 btrfs_release_delayed_item(item);
1485 mutex_unlock(&node->mutex);
1489 int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans,
1490 struct btrfs_fs_info *fs_info,
1491 struct btrfs_inode *dir, u64 index)
1493 struct btrfs_delayed_node *node;
1494 struct btrfs_delayed_item *item;
1495 struct btrfs_key item_key;
1498 node = btrfs_get_or_create_delayed_node(dir);
1500 return PTR_ERR(node);
1502 item_key.objectid = btrfs_ino(dir);
1503 item_key.type = BTRFS_DIR_INDEX_KEY;
1504 item_key.offset = index;
1506 ret = btrfs_delete_delayed_insertion_item(fs_info, node, &item_key);
1510 item = btrfs_alloc_delayed_item(0);
1516 item->key = item_key;
1518 ret = btrfs_delayed_item_reserve_metadata(trans, fs_info, item);
1520 * we have reserved enough space when we start a new transaction,
1521 * so reserving metadata failure is impossible.
1525 mutex_lock(&node->mutex);
1526 ret = __btrfs_add_delayed_deletion_item(node, item);
1527 if (unlikely(ret)) {
1529 "err add delayed dir index item(index: %llu) into the deletion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
1530 index, node->root->objectid, node->inode_id, ret);
1533 mutex_unlock(&node->mutex);
1535 btrfs_release_delayed_node(node);
1539 int btrfs_inode_delayed_dir_index_count(struct btrfs_inode *inode)
1541 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1547 * Since we have held i_mutex of this directory, it is impossible that
1548 * a new directory index is added into the delayed node and index_cnt
1549 * is updated now. So we needn't lock the delayed node.
1551 if (!delayed_node->index_cnt) {
1552 btrfs_release_delayed_node(delayed_node);
1556 inode->index_cnt = delayed_node->index_cnt;
1557 btrfs_release_delayed_node(delayed_node);
1561 bool btrfs_readdir_get_delayed_items(struct inode *inode,
1562 struct list_head *ins_list,
1563 struct list_head *del_list)
1565 struct btrfs_delayed_node *delayed_node;
1566 struct btrfs_delayed_item *item;
1568 delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
1573 * We can only do one readdir with delayed items at a time because of
1574 * item->readdir_list.
1576 inode_unlock_shared(inode);
1579 mutex_lock(&delayed_node->mutex);
1580 item = __btrfs_first_delayed_insertion_item(delayed_node);
1582 refcount_inc(&item->refs);
1583 list_add_tail(&item->readdir_list, ins_list);
1584 item = __btrfs_next_delayed_item(item);
1587 item = __btrfs_first_delayed_deletion_item(delayed_node);
1589 refcount_inc(&item->refs);
1590 list_add_tail(&item->readdir_list, del_list);
1591 item = __btrfs_next_delayed_item(item);
1593 mutex_unlock(&delayed_node->mutex);
1595 * This delayed node is still cached in the btrfs inode, so refs
1596 * must be > 1 now, and we needn't check it is going to be freed
1599 * Besides that, this function is used to read dir, we do not
1600 * insert/delete delayed items in this period. So we also needn't
1601 * requeue or dequeue this delayed node.
1603 refcount_dec(&delayed_node->refs);
1608 void btrfs_readdir_put_delayed_items(struct inode *inode,
1609 struct list_head *ins_list,
1610 struct list_head *del_list)
1612 struct btrfs_delayed_item *curr, *next;
1614 list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1615 list_del(&curr->readdir_list);
1616 if (refcount_dec_and_test(&curr->refs))
1620 list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1621 list_del(&curr->readdir_list);
1622 if (refcount_dec_and_test(&curr->refs))
1627 * The VFS is going to do up_read(), so we need to downgrade back to a
1630 downgrade_write(&inode->i_rwsem);
1633 int btrfs_should_delete_dir_index(struct list_head *del_list,
1636 struct btrfs_delayed_item *curr, *next;
1639 if (list_empty(del_list))
1642 list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1643 if (curr->key.offset > index)
1646 list_del(&curr->readdir_list);
1647 ret = (curr->key.offset == index);
1649 if (refcount_dec_and_test(&curr->refs))
1661 * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree
1664 int btrfs_readdir_delayed_dir_index(struct dir_context *ctx,
1665 struct list_head *ins_list)
1667 struct btrfs_dir_item *di;
1668 struct btrfs_delayed_item *curr, *next;
1669 struct btrfs_key location;
1673 unsigned char d_type;
1675 if (list_empty(ins_list))
1679 * Changing the data of the delayed item is impossible. So
1680 * we needn't lock them. And we have held i_mutex of the
1681 * directory, nobody can delete any directory indexes now.
1683 list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1684 list_del(&curr->readdir_list);
1686 if (curr->key.offset < ctx->pos) {
1687 if (refcount_dec_and_test(&curr->refs))
1692 ctx->pos = curr->key.offset;
1694 di = (struct btrfs_dir_item *)curr->data;
1695 name = (char *)(di + 1);
1696 name_len = btrfs_stack_dir_name_len(di);
1698 d_type = btrfs_filetype_table[di->type];
1699 btrfs_disk_key_to_cpu(&location, &di->location);
1701 over = !dir_emit(ctx, name, name_len,
1702 location.objectid, d_type);
1704 if (refcount_dec_and_test(&curr->refs))
1714 static void fill_stack_inode_item(struct btrfs_trans_handle *trans,
1715 struct btrfs_inode_item *inode_item,
1716 struct inode *inode)
1718 btrfs_set_stack_inode_uid(inode_item, i_uid_read(inode));
1719 btrfs_set_stack_inode_gid(inode_item, i_gid_read(inode));
1720 btrfs_set_stack_inode_size(inode_item, BTRFS_I(inode)->disk_i_size);
1721 btrfs_set_stack_inode_mode(inode_item, inode->i_mode);
1722 btrfs_set_stack_inode_nlink(inode_item, inode->i_nlink);
1723 btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(inode));
1724 btrfs_set_stack_inode_generation(inode_item,
1725 BTRFS_I(inode)->generation);
1726 btrfs_set_stack_inode_sequence(inode_item, inode->i_version);
1727 btrfs_set_stack_inode_transid(inode_item, trans->transid);
1728 btrfs_set_stack_inode_rdev(inode_item, inode->i_rdev);
1729 btrfs_set_stack_inode_flags(inode_item, BTRFS_I(inode)->flags);
1730 btrfs_set_stack_inode_block_group(inode_item, 0);
1732 btrfs_set_stack_timespec_sec(&inode_item->atime,
1733 inode->i_atime.tv_sec);
1734 btrfs_set_stack_timespec_nsec(&inode_item->atime,
1735 inode->i_atime.tv_nsec);
1737 btrfs_set_stack_timespec_sec(&inode_item->mtime,
1738 inode->i_mtime.tv_sec);
1739 btrfs_set_stack_timespec_nsec(&inode_item->mtime,
1740 inode->i_mtime.tv_nsec);
1742 btrfs_set_stack_timespec_sec(&inode_item->ctime,
1743 inode->i_ctime.tv_sec);
1744 btrfs_set_stack_timespec_nsec(&inode_item->ctime,
1745 inode->i_ctime.tv_nsec);
1747 btrfs_set_stack_timespec_sec(&inode_item->otime,
1748 BTRFS_I(inode)->i_otime.tv_sec);
1749 btrfs_set_stack_timespec_nsec(&inode_item->otime,
1750 BTRFS_I(inode)->i_otime.tv_nsec);
1753 int btrfs_fill_inode(struct inode *inode, u32 *rdev)
1755 struct btrfs_delayed_node *delayed_node;
1756 struct btrfs_inode_item *inode_item;
1758 delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
1762 mutex_lock(&delayed_node->mutex);
1763 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1764 mutex_unlock(&delayed_node->mutex);
1765 btrfs_release_delayed_node(delayed_node);
1769 inode_item = &delayed_node->inode_item;
1771 i_uid_write(inode, btrfs_stack_inode_uid(inode_item));
1772 i_gid_write(inode, btrfs_stack_inode_gid(inode_item));
1773 btrfs_i_size_write(BTRFS_I(inode), btrfs_stack_inode_size(inode_item));
1774 inode->i_mode = btrfs_stack_inode_mode(inode_item);
1775 set_nlink(inode, btrfs_stack_inode_nlink(inode_item));
1776 inode_set_bytes(inode, btrfs_stack_inode_nbytes(inode_item));
1777 BTRFS_I(inode)->generation = btrfs_stack_inode_generation(inode_item);
1778 BTRFS_I(inode)->last_trans = btrfs_stack_inode_transid(inode_item);
1780 inode->i_version = btrfs_stack_inode_sequence(inode_item);
1782 *rdev = btrfs_stack_inode_rdev(inode_item);
1783 BTRFS_I(inode)->flags = btrfs_stack_inode_flags(inode_item);
1785 inode->i_atime.tv_sec = btrfs_stack_timespec_sec(&inode_item->atime);
1786 inode->i_atime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->atime);
1788 inode->i_mtime.tv_sec = btrfs_stack_timespec_sec(&inode_item->mtime);
1789 inode->i_mtime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->mtime);
1791 inode->i_ctime.tv_sec = btrfs_stack_timespec_sec(&inode_item->ctime);
1792 inode->i_ctime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->ctime);
1794 BTRFS_I(inode)->i_otime.tv_sec =
1795 btrfs_stack_timespec_sec(&inode_item->otime);
1796 BTRFS_I(inode)->i_otime.tv_nsec =
1797 btrfs_stack_timespec_nsec(&inode_item->otime);
1799 inode->i_generation = BTRFS_I(inode)->generation;
1800 BTRFS_I(inode)->index_cnt = (u64)-1;
1802 mutex_unlock(&delayed_node->mutex);
1803 btrfs_release_delayed_node(delayed_node);
1807 int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans,
1808 struct btrfs_root *root, struct inode *inode)
1810 struct btrfs_delayed_node *delayed_node;
1813 delayed_node = btrfs_get_or_create_delayed_node(BTRFS_I(inode));
1814 if (IS_ERR(delayed_node))
1815 return PTR_ERR(delayed_node);
1817 mutex_lock(&delayed_node->mutex);
1818 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1819 fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1823 ret = btrfs_delayed_inode_reserve_metadata(trans, root, BTRFS_I(inode),
1828 fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1829 set_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
1830 delayed_node->count++;
1831 atomic_inc(&root->fs_info->delayed_root->items);
1833 mutex_unlock(&delayed_node->mutex);
1834 btrfs_release_delayed_node(delayed_node);
1838 int btrfs_delayed_delete_inode_ref(struct btrfs_inode *inode)
1840 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1841 struct btrfs_delayed_node *delayed_node;
1844 * we don't do delayed inode updates during log recovery because it
1845 * leads to enospc problems. This means we also can't do
1846 * delayed inode refs
1848 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
1851 delayed_node = btrfs_get_or_create_delayed_node(inode);
1852 if (IS_ERR(delayed_node))
1853 return PTR_ERR(delayed_node);
1856 * We don't reserve space for inode ref deletion is because:
1857 * - We ONLY do async inode ref deletion for the inode who has only
1858 * one link(i_nlink == 1), it means there is only one inode ref.
1859 * And in most case, the inode ref and the inode item are in the
1860 * same leaf, and we will deal with them at the same time.
1861 * Since we are sure we will reserve the space for the inode item,
1862 * it is unnecessary to reserve space for inode ref deletion.
1863 * - If the inode ref and the inode item are not in the same leaf,
1864 * We also needn't worry about enospc problem, because we reserve
1865 * much more space for the inode update than it needs.
1866 * - At the worst, we can steal some space from the global reservation.
1869 mutex_lock(&delayed_node->mutex);
1870 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1873 set_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
1874 delayed_node->count++;
1875 atomic_inc(&fs_info->delayed_root->items);
1877 mutex_unlock(&delayed_node->mutex);
1878 btrfs_release_delayed_node(delayed_node);
1882 static void __btrfs_kill_delayed_node(struct btrfs_delayed_node *delayed_node)
1884 struct btrfs_root *root = delayed_node->root;
1885 struct btrfs_fs_info *fs_info = root->fs_info;
1886 struct btrfs_delayed_item *curr_item, *prev_item;
1888 mutex_lock(&delayed_node->mutex);
1889 curr_item = __btrfs_first_delayed_insertion_item(delayed_node);
1891 btrfs_delayed_item_release_metadata(fs_info, curr_item);
1892 prev_item = curr_item;
1893 curr_item = __btrfs_next_delayed_item(prev_item);
1894 btrfs_release_delayed_item(prev_item);
1897 curr_item = __btrfs_first_delayed_deletion_item(delayed_node);
1899 btrfs_delayed_item_release_metadata(fs_info, curr_item);
1900 prev_item = curr_item;
1901 curr_item = __btrfs_next_delayed_item(prev_item);
1902 btrfs_release_delayed_item(prev_item);
1905 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1906 btrfs_release_delayed_iref(delayed_node);
1908 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1909 btrfs_delayed_inode_release_metadata(fs_info, delayed_node);
1910 btrfs_release_delayed_inode(delayed_node);
1912 mutex_unlock(&delayed_node->mutex);
1915 void btrfs_kill_delayed_inode_items(struct btrfs_inode *inode)
1917 struct btrfs_delayed_node *delayed_node;
1919 delayed_node = btrfs_get_delayed_node(inode);
1923 __btrfs_kill_delayed_node(delayed_node);
1924 btrfs_release_delayed_node(delayed_node);
1927 void btrfs_kill_all_delayed_nodes(struct btrfs_root *root)
1930 struct btrfs_delayed_node *delayed_nodes[8];
1934 spin_lock(&root->inode_lock);
1935 n = radix_tree_gang_lookup(&root->delayed_nodes_tree,
1936 (void **)delayed_nodes, inode_id,
1937 ARRAY_SIZE(delayed_nodes));
1939 spin_unlock(&root->inode_lock);
1943 inode_id = delayed_nodes[n - 1]->inode_id + 1;
1945 for (i = 0; i < n; i++)
1946 refcount_inc(&delayed_nodes[i]->refs);
1947 spin_unlock(&root->inode_lock);
1949 for (i = 0; i < n; i++) {
1950 __btrfs_kill_delayed_node(delayed_nodes[i]);
1951 btrfs_release_delayed_node(delayed_nodes[i]);
1956 void btrfs_destroy_delayed_inodes(struct btrfs_fs_info *fs_info)
1958 struct btrfs_delayed_node *curr_node, *prev_node;
1960 curr_node = btrfs_first_delayed_node(fs_info->delayed_root);
1962 __btrfs_kill_delayed_node(curr_node);
1964 prev_node = curr_node;
1965 curr_node = btrfs_next_delayed_node(curr_node);
1966 btrfs_release_delayed_node(prev_node);