1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2011 Fujitsu. All rights reserved.
4 * Written by Miao Xie <miaox@cn.fujitsu.com>
7 #include <linux/slab.h>
8 #include <linux/iversion.h>
9 #include "delayed-inode.h"
11 #include "transaction.h"
15 #define BTRFS_DELAYED_WRITEBACK 512
16 #define BTRFS_DELAYED_BACKGROUND 128
17 #define BTRFS_DELAYED_BATCH 16
19 static struct kmem_cache *delayed_node_cache;
21 int __init btrfs_delayed_inode_init(void)
23 delayed_node_cache = kmem_cache_create("btrfs_delayed_node",
24 sizeof(struct btrfs_delayed_node),
28 if (!delayed_node_cache)
33 void __cold btrfs_delayed_inode_exit(void)
35 kmem_cache_destroy(delayed_node_cache);
38 static inline void btrfs_init_delayed_node(
39 struct btrfs_delayed_node *delayed_node,
40 struct btrfs_root *root, u64 inode_id)
42 delayed_node->root = root;
43 delayed_node->inode_id = inode_id;
44 refcount_set(&delayed_node->refs, 0);
45 delayed_node->ins_root = RB_ROOT;
46 delayed_node->del_root = RB_ROOT;
47 mutex_init(&delayed_node->mutex);
48 INIT_LIST_HEAD(&delayed_node->n_list);
49 INIT_LIST_HEAD(&delayed_node->p_list);
52 static inline int btrfs_is_continuous_delayed_item(
53 struct btrfs_delayed_item *item1,
54 struct btrfs_delayed_item *item2)
56 if (item1->key.type == BTRFS_DIR_INDEX_KEY &&
57 item1->key.objectid == item2->key.objectid &&
58 item1->key.type == item2->key.type &&
59 item1->key.offset + 1 == item2->key.offset)
64 static struct btrfs_delayed_node *btrfs_get_delayed_node(
65 struct btrfs_inode *btrfs_inode)
67 struct btrfs_root *root = btrfs_inode->root;
68 u64 ino = btrfs_ino(btrfs_inode);
69 struct btrfs_delayed_node *node;
71 node = READ_ONCE(btrfs_inode->delayed_node);
73 refcount_inc(&node->refs);
77 spin_lock(&root->inode_lock);
78 node = radix_tree_lookup(&root->delayed_nodes_tree, ino);
81 if (btrfs_inode->delayed_node) {
82 refcount_inc(&node->refs); /* can be accessed */
83 BUG_ON(btrfs_inode->delayed_node != node);
84 spin_unlock(&root->inode_lock);
89 * It's possible that we're racing into the middle of removing
90 * this node from the radix tree. In this case, the refcount
91 * was zero and it should never go back to one. Just return
92 * NULL like it was never in the radix at all; our release
93 * function is in the process of removing it.
95 * Some implementations of refcount_inc refuse to bump the
96 * refcount once it has hit zero. If we don't do this dance
97 * here, refcount_inc() may decide to just WARN_ONCE() instead
98 * of actually bumping the refcount.
100 * If this node is properly in the radix, we want to bump the
101 * refcount twice, once for the inode and once for this get
104 if (refcount_inc_not_zero(&node->refs)) {
105 refcount_inc(&node->refs);
106 btrfs_inode->delayed_node = node;
111 spin_unlock(&root->inode_lock);
114 spin_unlock(&root->inode_lock);
119 /* Will return either the node or PTR_ERR(-ENOMEM) */
120 static struct btrfs_delayed_node *btrfs_get_or_create_delayed_node(
121 struct btrfs_inode *btrfs_inode)
123 struct btrfs_delayed_node *node;
124 struct btrfs_root *root = btrfs_inode->root;
125 u64 ino = btrfs_ino(btrfs_inode);
129 node = btrfs_get_delayed_node(btrfs_inode);
133 node = kmem_cache_zalloc(delayed_node_cache, GFP_NOFS);
135 return ERR_PTR(-ENOMEM);
136 btrfs_init_delayed_node(node, root, ino);
138 /* cached in the btrfs inode and can be accessed */
139 refcount_set(&node->refs, 2);
141 ret = radix_tree_preload(GFP_NOFS);
143 kmem_cache_free(delayed_node_cache, node);
147 spin_lock(&root->inode_lock);
148 ret = radix_tree_insert(&root->delayed_nodes_tree, ino, node);
149 if (ret == -EEXIST) {
150 spin_unlock(&root->inode_lock);
151 kmem_cache_free(delayed_node_cache, node);
152 radix_tree_preload_end();
155 btrfs_inode->delayed_node = node;
156 spin_unlock(&root->inode_lock);
157 radix_tree_preload_end();
163 * Call it when holding delayed_node->mutex
165 * If mod = 1, add this node into the prepared list.
167 static void btrfs_queue_delayed_node(struct btrfs_delayed_root *root,
168 struct btrfs_delayed_node *node,
171 spin_lock(&root->lock);
172 if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
173 if (!list_empty(&node->p_list))
174 list_move_tail(&node->p_list, &root->prepare_list);
176 list_add_tail(&node->p_list, &root->prepare_list);
178 list_add_tail(&node->n_list, &root->node_list);
179 list_add_tail(&node->p_list, &root->prepare_list);
180 refcount_inc(&node->refs); /* inserted into list */
182 set_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
184 spin_unlock(&root->lock);
187 /* Call it when holding delayed_node->mutex */
188 static void btrfs_dequeue_delayed_node(struct btrfs_delayed_root *root,
189 struct btrfs_delayed_node *node)
191 spin_lock(&root->lock);
192 if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
194 refcount_dec(&node->refs); /* not in the list */
195 list_del_init(&node->n_list);
196 if (!list_empty(&node->p_list))
197 list_del_init(&node->p_list);
198 clear_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
200 spin_unlock(&root->lock);
203 static struct btrfs_delayed_node *btrfs_first_delayed_node(
204 struct btrfs_delayed_root *delayed_root)
207 struct btrfs_delayed_node *node = NULL;
209 spin_lock(&delayed_root->lock);
210 if (list_empty(&delayed_root->node_list))
213 p = delayed_root->node_list.next;
214 node = list_entry(p, struct btrfs_delayed_node, n_list);
215 refcount_inc(&node->refs);
217 spin_unlock(&delayed_root->lock);
222 static struct btrfs_delayed_node *btrfs_next_delayed_node(
223 struct btrfs_delayed_node *node)
225 struct btrfs_delayed_root *delayed_root;
227 struct btrfs_delayed_node *next = NULL;
229 delayed_root = node->root->fs_info->delayed_root;
230 spin_lock(&delayed_root->lock);
231 if (!test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
232 /* not in the list */
233 if (list_empty(&delayed_root->node_list))
235 p = delayed_root->node_list.next;
236 } else if (list_is_last(&node->n_list, &delayed_root->node_list))
239 p = node->n_list.next;
241 next = list_entry(p, struct btrfs_delayed_node, n_list);
242 refcount_inc(&next->refs);
244 spin_unlock(&delayed_root->lock);
249 static void __btrfs_release_delayed_node(
250 struct btrfs_delayed_node *delayed_node,
253 struct btrfs_delayed_root *delayed_root;
258 delayed_root = delayed_node->root->fs_info->delayed_root;
260 mutex_lock(&delayed_node->mutex);
261 if (delayed_node->count)
262 btrfs_queue_delayed_node(delayed_root, delayed_node, mod);
264 btrfs_dequeue_delayed_node(delayed_root, delayed_node);
265 mutex_unlock(&delayed_node->mutex);
267 if (refcount_dec_and_test(&delayed_node->refs)) {
268 struct btrfs_root *root = delayed_node->root;
270 spin_lock(&root->inode_lock);
272 * Once our refcount goes to zero, nobody is allowed to bump it
273 * back up. We can delete it now.
275 ASSERT(refcount_read(&delayed_node->refs) == 0);
276 radix_tree_delete(&root->delayed_nodes_tree,
277 delayed_node->inode_id);
278 spin_unlock(&root->inode_lock);
279 kmem_cache_free(delayed_node_cache, delayed_node);
283 static inline void btrfs_release_delayed_node(struct btrfs_delayed_node *node)
285 __btrfs_release_delayed_node(node, 0);
288 static struct btrfs_delayed_node *btrfs_first_prepared_delayed_node(
289 struct btrfs_delayed_root *delayed_root)
292 struct btrfs_delayed_node *node = NULL;
294 spin_lock(&delayed_root->lock);
295 if (list_empty(&delayed_root->prepare_list))
298 p = delayed_root->prepare_list.next;
300 node = list_entry(p, struct btrfs_delayed_node, p_list);
301 refcount_inc(&node->refs);
303 spin_unlock(&delayed_root->lock);
308 static inline void btrfs_release_prepared_delayed_node(
309 struct btrfs_delayed_node *node)
311 __btrfs_release_delayed_node(node, 1);
314 static struct btrfs_delayed_item *btrfs_alloc_delayed_item(u32 data_len)
316 struct btrfs_delayed_item *item;
317 item = kmalloc(sizeof(*item) + data_len, GFP_NOFS);
319 item->data_len = data_len;
320 item->ins_or_del = 0;
321 item->bytes_reserved = 0;
322 item->delayed_node = NULL;
323 refcount_set(&item->refs, 1);
329 * __btrfs_lookup_delayed_item - look up the delayed item by key
330 * @delayed_node: pointer to the delayed node
331 * @key: the key to look up
332 * @prev: used to store the prev item if the right item isn't found
333 * @next: used to store the next item if the right item isn't found
335 * Note: if we don't find the right item, we will return the prev item and
338 static struct btrfs_delayed_item *__btrfs_lookup_delayed_item(
339 struct rb_root *root,
340 struct btrfs_key *key,
341 struct btrfs_delayed_item **prev,
342 struct btrfs_delayed_item **next)
344 struct rb_node *node, *prev_node = NULL;
345 struct btrfs_delayed_item *delayed_item = NULL;
348 node = root->rb_node;
351 delayed_item = rb_entry(node, struct btrfs_delayed_item,
354 ret = btrfs_comp_cpu_keys(&delayed_item->key, key);
356 node = node->rb_right;
358 node = node->rb_left;
367 *prev = delayed_item;
368 else if ((node = rb_prev(prev_node)) != NULL) {
369 *prev = rb_entry(node, struct btrfs_delayed_item,
379 *next = delayed_item;
380 else if ((node = rb_next(prev_node)) != NULL) {
381 *next = rb_entry(node, struct btrfs_delayed_item,
389 static struct btrfs_delayed_item *__btrfs_lookup_delayed_insertion_item(
390 struct btrfs_delayed_node *delayed_node,
391 struct btrfs_key *key)
393 return __btrfs_lookup_delayed_item(&delayed_node->ins_root, key,
397 static int __btrfs_add_delayed_item(struct btrfs_delayed_node *delayed_node,
398 struct btrfs_delayed_item *ins,
401 struct rb_node **p, *node;
402 struct rb_node *parent_node = NULL;
403 struct rb_root *root;
404 struct btrfs_delayed_item *item;
407 if (action == BTRFS_DELAYED_INSERTION_ITEM)
408 root = &delayed_node->ins_root;
409 else if (action == BTRFS_DELAYED_DELETION_ITEM)
410 root = &delayed_node->del_root;
414 node = &ins->rb_node;
418 item = rb_entry(parent_node, struct btrfs_delayed_item,
421 cmp = btrfs_comp_cpu_keys(&item->key, &ins->key);
430 rb_link_node(node, parent_node, p);
431 rb_insert_color(node, root);
432 ins->delayed_node = delayed_node;
433 ins->ins_or_del = action;
435 if (ins->key.type == BTRFS_DIR_INDEX_KEY &&
436 action == BTRFS_DELAYED_INSERTION_ITEM &&
437 ins->key.offset >= delayed_node->index_cnt)
438 delayed_node->index_cnt = ins->key.offset + 1;
440 delayed_node->count++;
441 atomic_inc(&delayed_node->root->fs_info->delayed_root->items);
445 static int __btrfs_add_delayed_insertion_item(struct btrfs_delayed_node *node,
446 struct btrfs_delayed_item *item)
448 return __btrfs_add_delayed_item(node, item,
449 BTRFS_DELAYED_INSERTION_ITEM);
452 static int __btrfs_add_delayed_deletion_item(struct btrfs_delayed_node *node,
453 struct btrfs_delayed_item *item)
455 return __btrfs_add_delayed_item(node, item,
456 BTRFS_DELAYED_DELETION_ITEM);
459 static void finish_one_item(struct btrfs_delayed_root *delayed_root)
461 int seq = atomic_inc_return(&delayed_root->items_seq);
464 * atomic_dec_return implies a barrier for waitqueue_active
466 if ((atomic_dec_return(&delayed_root->items) <
467 BTRFS_DELAYED_BACKGROUND || seq % BTRFS_DELAYED_BATCH == 0) &&
468 waitqueue_active(&delayed_root->wait))
469 wake_up(&delayed_root->wait);
472 static void __btrfs_remove_delayed_item(struct btrfs_delayed_item *delayed_item)
474 struct rb_root *root;
475 struct btrfs_delayed_root *delayed_root;
477 delayed_root = delayed_item->delayed_node->root->fs_info->delayed_root;
479 BUG_ON(!delayed_root);
480 BUG_ON(delayed_item->ins_or_del != BTRFS_DELAYED_DELETION_ITEM &&
481 delayed_item->ins_or_del != BTRFS_DELAYED_INSERTION_ITEM);
483 if (delayed_item->ins_or_del == BTRFS_DELAYED_INSERTION_ITEM)
484 root = &delayed_item->delayed_node->ins_root;
486 root = &delayed_item->delayed_node->del_root;
488 rb_erase(&delayed_item->rb_node, root);
489 delayed_item->delayed_node->count--;
491 finish_one_item(delayed_root);
494 static void btrfs_release_delayed_item(struct btrfs_delayed_item *item)
497 __btrfs_remove_delayed_item(item);
498 if (refcount_dec_and_test(&item->refs))
503 static struct btrfs_delayed_item *__btrfs_first_delayed_insertion_item(
504 struct btrfs_delayed_node *delayed_node)
507 struct btrfs_delayed_item *item = NULL;
509 p = rb_first(&delayed_node->ins_root);
511 item = rb_entry(p, struct btrfs_delayed_item, rb_node);
516 static struct btrfs_delayed_item *__btrfs_first_delayed_deletion_item(
517 struct btrfs_delayed_node *delayed_node)
520 struct btrfs_delayed_item *item = NULL;
522 p = rb_first(&delayed_node->del_root);
524 item = rb_entry(p, struct btrfs_delayed_item, rb_node);
529 static struct btrfs_delayed_item *__btrfs_next_delayed_item(
530 struct btrfs_delayed_item *item)
533 struct btrfs_delayed_item *next = NULL;
535 p = rb_next(&item->rb_node);
537 next = rb_entry(p, struct btrfs_delayed_item, rb_node);
542 static int btrfs_delayed_item_reserve_metadata(struct btrfs_trans_handle *trans,
543 struct btrfs_root *root,
544 struct btrfs_delayed_item *item)
546 struct btrfs_block_rsv *src_rsv;
547 struct btrfs_block_rsv *dst_rsv;
548 struct btrfs_fs_info *fs_info = root->fs_info;
552 if (!trans->bytes_reserved)
555 src_rsv = trans->block_rsv;
556 dst_rsv = &fs_info->delayed_block_rsv;
558 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
561 * Here we migrate space rsv from transaction rsv, since have already
562 * reserved space when starting a transaction. So no need to reserve
565 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1);
567 trace_btrfs_space_reservation(fs_info, "delayed_item",
570 item->bytes_reserved = num_bytes;
576 static void btrfs_delayed_item_release_metadata(struct btrfs_root *root,
577 struct btrfs_delayed_item *item)
579 struct btrfs_block_rsv *rsv;
580 struct btrfs_fs_info *fs_info = root->fs_info;
582 if (!item->bytes_reserved)
585 rsv = &fs_info->delayed_block_rsv;
587 * Check btrfs_delayed_item_reserve_metadata() to see why we don't need
588 * to release/reserve qgroup space.
590 trace_btrfs_space_reservation(fs_info, "delayed_item",
591 item->key.objectid, item->bytes_reserved,
593 btrfs_block_rsv_release(fs_info, rsv,
594 item->bytes_reserved);
597 static int btrfs_delayed_inode_reserve_metadata(
598 struct btrfs_trans_handle *trans,
599 struct btrfs_root *root,
600 struct btrfs_inode *inode,
601 struct btrfs_delayed_node *node)
603 struct btrfs_fs_info *fs_info = root->fs_info;
604 struct btrfs_block_rsv *src_rsv;
605 struct btrfs_block_rsv *dst_rsv;
609 src_rsv = trans->block_rsv;
610 dst_rsv = &fs_info->delayed_block_rsv;
612 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
615 * btrfs_dirty_inode will update the inode under btrfs_join_transaction
616 * which doesn't reserve space for speed. This is a problem since we
617 * still need to reserve space for this update, so try to reserve the
620 * Now if src_rsv == delalloc_block_rsv we'll let it just steal since
621 * we always reserve enough to update the inode item.
623 if (!src_rsv || (!trans->bytes_reserved &&
624 src_rsv->type != BTRFS_BLOCK_RSV_DELALLOC)) {
625 ret = btrfs_qgroup_reserve_meta_prealloc(root,
626 fs_info->nodesize, true);
629 ret = btrfs_block_rsv_add(root, dst_rsv, num_bytes,
630 BTRFS_RESERVE_NO_FLUSH);
632 * Since we're under a transaction reserve_metadata_bytes could
633 * try to commit the transaction which will make it return
634 * EAGAIN to make us stop the transaction we have, so return
635 * ENOSPC instead so that btrfs_dirty_inode knows what to do.
637 if (ret == -EAGAIN) {
639 btrfs_qgroup_free_meta_prealloc(root, num_bytes);
642 node->bytes_reserved = num_bytes;
643 trace_btrfs_space_reservation(fs_info,
648 btrfs_qgroup_free_meta_prealloc(root, fs_info->nodesize);
653 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1);
655 trace_btrfs_space_reservation(fs_info, "delayed_inode",
656 btrfs_ino(inode), num_bytes, 1);
657 node->bytes_reserved = num_bytes;
663 static void btrfs_delayed_inode_release_metadata(struct btrfs_fs_info *fs_info,
664 struct btrfs_delayed_node *node,
667 struct btrfs_block_rsv *rsv;
669 if (!node->bytes_reserved)
672 rsv = &fs_info->delayed_block_rsv;
673 trace_btrfs_space_reservation(fs_info, "delayed_inode",
674 node->inode_id, node->bytes_reserved, 0);
675 btrfs_block_rsv_release(fs_info, rsv,
676 node->bytes_reserved);
678 btrfs_qgroup_free_meta_prealloc(node->root,
679 node->bytes_reserved);
681 btrfs_qgroup_convert_reserved_meta(node->root,
682 node->bytes_reserved);
683 node->bytes_reserved = 0;
687 * This helper will insert some continuous items into the same leaf according
688 * to the free space of the leaf.
690 static int btrfs_batch_insert_items(struct btrfs_root *root,
691 struct btrfs_path *path,
692 struct btrfs_delayed_item *item)
694 struct btrfs_fs_info *fs_info = root->fs_info;
695 struct btrfs_delayed_item *curr, *next;
697 int total_data_size = 0, total_size = 0;
698 struct extent_buffer *leaf;
700 struct btrfs_key *keys;
702 struct list_head head;
708 BUG_ON(!path->nodes[0]);
710 leaf = path->nodes[0];
711 free_space = btrfs_leaf_free_space(fs_info, leaf);
712 INIT_LIST_HEAD(&head);
718 * count the number of the continuous items that we can insert in batch
720 while (total_size + next->data_len + sizeof(struct btrfs_item) <=
722 total_data_size += next->data_len;
723 total_size += next->data_len + sizeof(struct btrfs_item);
724 list_add_tail(&next->tree_list, &head);
728 next = __btrfs_next_delayed_item(curr);
732 if (!btrfs_is_continuous_delayed_item(curr, next))
742 * we need allocate some memory space, but it might cause the task
743 * to sleep, so we set all locked nodes in the path to blocking locks
746 btrfs_set_path_blocking(path);
748 keys = kmalloc_array(nitems, sizeof(struct btrfs_key), GFP_NOFS);
754 data_size = kmalloc_array(nitems, sizeof(u32), GFP_NOFS);
760 /* get keys of all the delayed items */
762 list_for_each_entry(next, &head, tree_list) {
764 data_size[i] = next->data_len;
768 /* reset all the locked nodes in the patch to spinning locks. */
769 btrfs_clear_path_blocking(path, NULL, 0);
771 /* insert the keys of the items */
772 setup_items_for_insert(root, path, keys, data_size,
773 total_data_size, total_size, nitems);
775 /* insert the dir index items */
776 slot = path->slots[0];
777 list_for_each_entry_safe(curr, next, &head, tree_list) {
778 data_ptr = btrfs_item_ptr(leaf, slot, char);
779 write_extent_buffer(leaf, &curr->data,
780 (unsigned long)data_ptr,
784 btrfs_delayed_item_release_metadata(root, curr);
786 list_del(&curr->tree_list);
787 btrfs_release_delayed_item(curr);
798 * This helper can just do simple insertion that needn't extend item for new
799 * data, such as directory name index insertion, inode insertion.
801 static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans,
802 struct btrfs_root *root,
803 struct btrfs_path *path,
804 struct btrfs_delayed_item *delayed_item)
806 struct extent_buffer *leaf;
810 ret = btrfs_insert_empty_item(trans, root, path, &delayed_item->key,
811 delayed_item->data_len);
812 if (ret < 0 && ret != -EEXIST)
815 leaf = path->nodes[0];
817 ptr = btrfs_item_ptr(leaf, path->slots[0], char);
819 write_extent_buffer(leaf, delayed_item->data, (unsigned long)ptr,
820 delayed_item->data_len);
821 btrfs_mark_buffer_dirty(leaf);
823 btrfs_delayed_item_release_metadata(root, delayed_item);
828 * we insert an item first, then if there are some continuous items, we try
829 * to insert those items into the same leaf.
831 static int btrfs_insert_delayed_items(struct btrfs_trans_handle *trans,
832 struct btrfs_path *path,
833 struct btrfs_root *root,
834 struct btrfs_delayed_node *node)
836 struct btrfs_delayed_item *curr, *prev;
840 mutex_lock(&node->mutex);
841 curr = __btrfs_first_delayed_insertion_item(node);
845 ret = btrfs_insert_delayed_item(trans, root, path, curr);
847 btrfs_release_path(path);
852 curr = __btrfs_next_delayed_item(prev);
853 if (curr && btrfs_is_continuous_delayed_item(prev, curr)) {
854 /* insert the continuous items into the same leaf */
856 btrfs_batch_insert_items(root, path, curr);
858 btrfs_release_delayed_item(prev);
859 btrfs_mark_buffer_dirty(path->nodes[0]);
861 btrfs_release_path(path);
862 mutex_unlock(&node->mutex);
866 mutex_unlock(&node->mutex);
870 static int btrfs_batch_delete_items(struct btrfs_trans_handle *trans,
871 struct btrfs_root *root,
872 struct btrfs_path *path,
873 struct btrfs_delayed_item *item)
875 struct btrfs_delayed_item *curr, *next;
876 struct extent_buffer *leaf;
877 struct btrfs_key key;
878 struct list_head head;
879 int nitems, i, last_item;
882 BUG_ON(!path->nodes[0]);
884 leaf = path->nodes[0];
887 last_item = btrfs_header_nritems(leaf) - 1;
889 return -ENOENT; /* FIXME: Is errno suitable? */
892 INIT_LIST_HEAD(&head);
893 btrfs_item_key_to_cpu(leaf, &key, i);
896 * count the number of the dir index items that we can delete in batch
898 while (btrfs_comp_cpu_keys(&next->key, &key) == 0) {
899 list_add_tail(&next->tree_list, &head);
903 next = __btrfs_next_delayed_item(curr);
907 if (!btrfs_is_continuous_delayed_item(curr, next))
913 btrfs_item_key_to_cpu(leaf, &key, i);
919 ret = btrfs_del_items(trans, root, path, path->slots[0], nitems);
923 list_for_each_entry_safe(curr, next, &head, tree_list) {
924 btrfs_delayed_item_release_metadata(root, curr);
925 list_del(&curr->tree_list);
926 btrfs_release_delayed_item(curr);
933 static int btrfs_delete_delayed_items(struct btrfs_trans_handle *trans,
934 struct btrfs_path *path,
935 struct btrfs_root *root,
936 struct btrfs_delayed_node *node)
938 struct btrfs_delayed_item *curr, *prev;
942 mutex_lock(&node->mutex);
943 curr = __btrfs_first_delayed_deletion_item(node);
947 ret = btrfs_search_slot(trans, root, &curr->key, path, -1, 1);
952 * can't find the item which the node points to, so this node
953 * is invalid, just drop it.
956 curr = __btrfs_next_delayed_item(prev);
957 btrfs_release_delayed_item(prev);
959 btrfs_release_path(path);
961 mutex_unlock(&node->mutex);
967 btrfs_batch_delete_items(trans, root, path, curr);
968 btrfs_release_path(path);
969 mutex_unlock(&node->mutex);
973 btrfs_release_path(path);
974 mutex_unlock(&node->mutex);
978 static void btrfs_release_delayed_inode(struct btrfs_delayed_node *delayed_node)
980 struct btrfs_delayed_root *delayed_root;
983 test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
984 BUG_ON(!delayed_node->root);
985 clear_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
986 delayed_node->count--;
988 delayed_root = delayed_node->root->fs_info->delayed_root;
989 finish_one_item(delayed_root);
993 static void btrfs_release_delayed_iref(struct btrfs_delayed_node *delayed_node)
995 struct btrfs_delayed_root *delayed_root;
997 ASSERT(delayed_node->root);
998 clear_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
999 delayed_node->count--;
1001 delayed_root = delayed_node->root->fs_info->delayed_root;
1002 finish_one_item(delayed_root);
1005 static int __btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1006 struct btrfs_root *root,
1007 struct btrfs_path *path,
1008 struct btrfs_delayed_node *node)
1010 struct btrfs_fs_info *fs_info = root->fs_info;
1011 struct btrfs_key key;
1012 struct btrfs_inode_item *inode_item;
1013 struct extent_buffer *leaf;
1017 key.objectid = node->inode_id;
1018 key.type = BTRFS_INODE_ITEM_KEY;
1021 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1026 ret = btrfs_lookup_inode(trans, root, path, &key, mod);
1028 btrfs_release_path(path);
1030 } else if (ret < 0) {
1034 leaf = path->nodes[0];
1035 inode_item = btrfs_item_ptr(leaf, path->slots[0],
1036 struct btrfs_inode_item);
1037 write_extent_buffer(leaf, &node->inode_item, (unsigned long)inode_item,
1038 sizeof(struct btrfs_inode_item));
1039 btrfs_mark_buffer_dirty(leaf);
1041 if (!test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1045 if (path->slots[0] >= btrfs_header_nritems(leaf))
1048 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1049 if (key.objectid != node->inode_id)
1052 if (key.type != BTRFS_INODE_REF_KEY &&
1053 key.type != BTRFS_INODE_EXTREF_KEY)
1057 * Delayed iref deletion is for the inode who has only one link,
1058 * so there is only one iref. The case that several irefs are
1059 * in the same item doesn't exist.
1061 btrfs_del_item(trans, root, path);
1063 btrfs_release_delayed_iref(node);
1065 btrfs_release_path(path);
1067 btrfs_delayed_inode_release_metadata(fs_info, node, (ret < 0));
1068 btrfs_release_delayed_inode(node);
1073 btrfs_release_path(path);
1075 key.type = BTRFS_INODE_EXTREF_KEY;
1077 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1083 leaf = path->nodes[0];
1088 static inline int btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1089 struct btrfs_root *root,
1090 struct btrfs_path *path,
1091 struct btrfs_delayed_node *node)
1095 mutex_lock(&node->mutex);
1096 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &node->flags)) {
1097 mutex_unlock(&node->mutex);
1101 ret = __btrfs_update_delayed_inode(trans, root, path, node);
1102 mutex_unlock(&node->mutex);
1107 __btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1108 struct btrfs_path *path,
1109 struct btrfs_delayed_node *node)
1113 ret = btrfs_insert_delayed_items(trans, path, node->root, node);
1117 ret = btrfs_delete_delayed_items(trans, path, node->root, node);
1121 ret = btrfs_update_delayed_inode(trans, node->root, path, node);
1126 * Called when committing the transaction.
1127 * Returns 0 on success.
1128 * Returns < 0 on error and returns with an aborted transaction with any
1129 * outstanding delayed items cleaned up.
1131 static int __btrfs_run_delayed_items(struct btrfs_trans_handle *trans, int nr)
1133 struct btrfs_fs_info *fs_info = trans->fs_info;
1134 struct btrfs_delayed_root *delayed_root;
1135 struct btrfs_delayed_node *curr_node, *prev_node;
1136 struct btrfs_path *path;
1137 struct btrfs_block_rsv *block_rsv;
1139 bool count = (nr > 0);
1144 path = btrfs_alloc_path();
1147 path->leave_spinning = 1;
1149 block_rsv = trans->block_rsv;
1150 trans->block_rsv = &fs_info->delayed_block_rsv;
1152 delayed_root = fs_info->delayed_root;
1154 curr_node = btrfs_first_delayed_node(delayed_root);
1155 while (curr_node && (!count || (count && nr--))) {
1156 ret = __btrfs_commit_inode_delayed_items(trans, path,
1159 btrfs_release_delayed_node(curr_node);
1161 btrfs_abort_transaction(trans, ret);
1165 prev_node = curr_node;
1166 curr_node = btrfs_next_delayed_node(curr_node);
1167 btrfs_release_delayed_node(prev_node);
1171 btrfs_release_delayed_node(curr_node);
1172 btrfs_free_path(path);
1173 trans->block_rsv = block_rsv;
1178 int btrfs_run_delayed_items(struct btrfs_trans_handle *trans)
1180 return __btrfs_run_delayed_items(trans, -1);
1183 int btrfs_run_delayed_items_nr(struct btrfs_trans_handle *trans, int nr)
1185 return __btrfs_run_delayed_items(trans, nr);
1188 int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1189 struct btrfs_inode *inode)
1191 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1192 struct btrfs_path *path;
1193 struct btrfs_block_rsv *block_rsv;
1199 mutex_lock(&delayed_node->mutex);
1200 if (!delayed_node->count) {
1201 mutex_unlock(&delayed_node->mutex);
1202 btrfs_release_delayed_node(delayed_node);
1205 mutex_unlock(&delayed_node->mutex);
1207 path = btrfs_alloc_path();
1209 btrfs_release_delayed_node(delayed_node);
1212 path->leave_spinning = 1;
1214 block_rsv = trans->block_rsv;
1215 trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv;
1217 ret = __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1219 btrfs_release_delayed_node(delayed_node);
1220 btrfs_free_path(path);
1221 trans->block_rsv = block_rsv;
1226 int btrfs_commit_inode_delayed_inode(struct btrfs_inode *inode)
1228 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1229 struct btrfs_trans_handle *trans;
1230 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1231 struct btrfs_path *path;
1232 struct btrfs_block_rsv *block_rsv;
1238 mutex_lock(&delayed_node->mutex);
1239 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1240 mutex_unlock(&delayed_node->mutex);
1241 btrfs_release_delayed_node(delayed_node);
1244 mutex_unlock(&delayed_node->mutex);
1246 trans = btrfs_join_transaction(delayed_node->root);
1247 if (IS_ERR(trans)) {
1248 ret = PTR_ERR(trans);
1252 path = btrfs_alloc_path();
1257 path->leave_spinning = 1;
1259 block_rsv = trans->block_rsv;
1260 trans->block_rsv = &fs_info->delayed_block_rsv;
1262 mutex_lock(&delayed_node->mutex);
1263 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags))
1264 ret = __btrfs_update_delayed_inode(trans, delayed_node->root,
1265 path, delayed_node);
1268 mutex_unlock(&delayed_node->mutex);
1270 btrfs_free_path(path);
1271 trans->block_rsv = block_rsv;
1273 btrfs_end_transaction(trans);
1274 btrfs_btree_balance_dirty(fs_info);
1276 btrfs_release_delayed_node(delayed_node);
1281 void btrfs_remove_delayed_node(struct btrfs_inode *inode)
1283 struct btrfs_delayed_node *delayed_node;
1285 delayed_node = READ_ONCE(inode->delayed_node);
1289 inode->delayed_node = NULL;
1290 btrfs_release_delayed_node(delayed_node);
1293 struct btrfs_async_delayed_work {
1294 struct btrfs_delayed_root *delayed_root;
1296 struct btrfs_work work;
1299 static void btrfs_async_run_delayed_root(struct btrfs_work *work)
1301 struct btrfs_async_delayed_work *async_work;
1302 struct btrfs_delayed_root *delayed_root;
1303 struct btrfs_trans_handle *trans;
1304 struct btrfs_path *path;
1305 struct btrfs_delayed_node *delayed_node = NULL;
1306 struct btrfs_root *root;
1307 struct btrfs_block_rsv *block_rsv;
1310 async_work = container_of(work, struct btrfs_async_delayed_work, work);
1311 delayed_root = async_work->delayed_root;
1313 path = btrfs_alloc_path();
1318 if (atomic_read(&delayed_root->items) <
1319 BTRFS_DELAYED_BACKGROUND / 2)
1322 delayed_node = btrfs_first_prepared_delayed_node(delayed_root);
1326 path->leave_spinning = 1;
1327 root = delayed_node->root;
1329 trans = btrfs_join_transaction(root);
1330 if (IS_ERR(trans)) {
1331 btrfs_release_path(path);
1332 btrfs_release_prepared_delayed_node(delayed_node);
1337 block_rsv = trans->block_rsv;
1338 trans->block_rsv = &root->fs_info->delayed_block_rsv;
1340 __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1342 trans->block_rsv = block_rsv;
1343 btrfs_end_transaction(trans);
1344 btrfs_btree_balance_dirty_nodelay(root->fs_info);
1346 btrfs_release_path(path);
1347 btrfs_release_prepared_delayed_node(delayed_node);
1350 } while ((async_work->nr == 0 && total_done < BTRFS_DELAYED_WRITEBACK)
1351 || total_done < async_work->nr);
1353 btrfs_free_path(path);
1355 wake_up(&delayed_root->wait);
1360 static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root,
1361 struct btrfs_fs_info *fs_info, int nr)
1363 struct btrfs_async_delayed_work *async_work;
1365 async_work = kmalloc(sizeof(*async_work), GFP_NOFS);
1369 async_work->delayed_root = delayed_root;
1370 btrfs_init_work(&async_work->work, btrfs_delayed_meta_helper,
1371 btrfs_async_run_delayed_root, NULL, NULL);
1372 async_work->nr = nr;
1374 btrfs_queue_work(fs_info->delayed_workers, &async_work->work);
1378 void btrfs_assert_delayed_root_empty(struct btrfs_fs_info *fs_info)
1380 WARN_ON(btrfs_first_delayed_node(fs_info->delayed_root));
1383 static int could_end_wait(struct btrfs_delayed_root *delayed_root, int seq)
1385 int val = atomic_read(&delayed_root->items_seq);
1387 if (val < seq || val >= seq + BTRFS_DELAYED_BATCH)
1390 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1396 void btrfs_balance_delayed_items(struct btrfs_fs_info *fs_info)
1398 struct btrfs_delayed_root *delayed_root = fs_info->delayed_root;
1400 if ((atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND) ||
1401 btrfs_workqueue_normal_congested(fs_info->delayed_workers))
1404 if (atomic_read(&delayed_root->items) >= BTRFS_DELAYED_WRITEBACK) {
1408 seq = atomic_read(&delayed_root->items_seq);
1410 ret = btrfs_wq_run_delayed_node(delayed_root, fs_info, 0);
1414 wait_event_interruptible(delayed_root->wait,
1415 could_end_wait(delayed_root, seq));
1419 btrfs_wq_run_delayed_node(delayed_root, fs_info, BTRFS_DELAYED_BATCH);
1422 /* Will return 0 or -ENOMEM */
1423 int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans,
1424 struct btrfs_fs_info *fs_info,
1425 const char *name, int name_len,
1426 struct btrfs_inode *dir,
1427 struct btrfs_disk_key *disk_key, u8 type,
1430 struct btrfs_delayed_node *delayed_node;
1431 struct btrfs_delayed_item *delayed_item;
1432 struct btrfs_dir_item *dir_item;
1435 delayed_node = btrfs_get_or_create_delayed_node(dir);
1436 if (IS_ERR(delayed_node))
1437 return PTR_ERR(delayed_node);
1439 delayed_item = btrfs_alloc_delayed_item(sizeof(*dir_item) + name_len);
1440 if (!delayed_item) {
1445 delayed_item->key.objectid = btrfs_ino(dir);
1446 delayed_item->key.type = BTRFS_DIR_INDEX_KEY;
1447 delayed_item->key.offset = index;
1449 dir_item = (struct btrfs_dir_item *)delayed_item->data;
1450 dir_item->location = *disk_key;
1451 btrfs_set_stack_dir_transid(dir_item, trans->transid);
1452 btrfs_set_stack_dir_data_len(dir_item, 0);
1453 btrfs_set_stack_dir_name_len(dir_item, name_len);
1454 btrfs_set_stack_dir_type(dir_item, type);
1455 memcpy((char *)(dir_item + 1), name, name_len);
1457 ret = btrfs_delayed_item_reserve_metadata(trans, dir->root, delayed_item);
1459 * we have reserved enough space when we start a new transaction,
1460 * so reserving metadata failure is impossible
1465 mutex_lock(&delayed_node->mutex);
1466 ret = __btrfs_add_delayed_insertion_item(delayed_node, delayed_item);
1467 if (unlikely(ret)) {
1469 "err add delayed dir index item(name: %.*s) into the insertion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
1470 name_len, name, delayed_node->root->objectid,
1471 delayed_node->inode_id, ret);
1474 mutex_unlock(&delayed_node->mutex);
1477 btrfs_release_delayed_node(delayed_node);
1481 static int btrfs_delete_delayed_insertion_item(struct btrfs_fs_info *fs_info,
1482 struct btrfs_delayed_node *node,
1483 struct btrfs_key *key)
1485 struct btrfs_delayed_item *item;
1487 mutex_lock(&node->mutex);
1488 item = __btrfs_lookup_delayed_insertion_item(node, key);
1490 mutex_unlock(&node->mutex);
1494 btrfs_delayed_item_release_metadata(node->root, item);
1495 btrfs_release_delayed_item(item);
1496 mutex_unlock(&node->mutex);
1500 int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans,
1501 struct btrfs_fs_info *fs_info,
1502 struct btrfs_inode *dir, u64 index)
1504 struct btrfs_delayed_node *node;
1505 struct btrfs_delayed_item *item;
1506 struct btrfs_key item_key;
1509 node = btrfs_get_or_create_delayed_node(dir);
1511 return PTR_ERR(node);
1513 item_key.objectid = btrfs_ino(dir);
1514 item_key.type = BTRFS_DIR_INDEX_KEY;
1515 item_key.offset = index;
1517 ret = btrfs_delete_delayed_insertion_item(fs_info, node, &item_key);
1521 item = btrfs_alloc_delayed_item(0);
1527 item->key = item_key;
1529 ret = btrfs_delayed_item_reserve_metadata(trans, dir->root, item);
1531 * we have reserved enough space when we start a new transaction,
1532 * so reserving metadata failure is impossible.
1536 mutex_lock(&node->mutex);
1537 ret = __btrfs_add_delayed_deletion_item(node, item);
1538 if (unlikely(ret)) {
1540 "err add delayed dir index item(index: %llu) into the deletion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
1541 index, node->root->objectid, node->inode_id, ret);
1544 mutex_unlock(&node->mutex);
1546 btrfs_release_delayed_node(node);
1550 int btrfs_inode_delayed_dir_index_count(struct btrfs_inode *inode)
1552 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1558 * Since we have held i_mutex of this directory, it is impossible that
1559 * a new directory index is added into the delayed node and index_cnt
1560 * is updated now. So we needn't lock the delayed node.
1562 if (!delayed_node->index_cnt) {
1563 btrfs_release_delayed_node(delayed_node);
1567 inode->index_cnt = delayed_node->index_cnt;
1568 btrfs_release_delayed_node(delayed_node);
1572 bool btrfs_readdir_get_delayed_items(struct inode *inode,
1573 struct list_head *ins_list,
1574 struct list_head *del_list)
1576 struct btrfs_delayed_node *delayed_node;
1577 struct btrfs_delayed_item *item;
1579 delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
1584 * We can only do one readdir with delayed items at a time because of
1585 * item->readdir_list.
1587 inode_unlock_shared(inode);
1590 mutex_lock(&delayed_node->mutex);
1591 item = __btrfs_first_delayed_insertion_item(delayed_node);
1593 refcount_inc(&item->refs);
1594 list_add_tail(&item->readdir_list, ins_list);
1595 item = __btrfs_next_delayed_item(item);
1598 item = __btrfs_first_delayed_deletion_item(delayed_node);
1600 refcount_inc(&item->refs);
1601 list_add_tail(&item->readdir_list, del_list);
1602 item = __btrfs_next_delayed_item(item);
1604 mutex_unlock(&delayed_node->mutex);
1606 * This delayed node is still cached in the btrfs inode, so refs
1607 * must be > 1 now, and we needn't check it is going to be freed
1610 * Besides that, this function is used to read dir, we do not
1611 * insert/delete delayed items in this period. So we also needn't
1612 * requeue or dequeue this delayed node.
1614 refcount_dec(&delayed_node->refs);
1619 void btrfs_readdir_put_delayed_items(struct inode *inode,
1620 struct list_head *ins_list,
1621 struct list_head *del_list)
1623 struct btrfs_delayed_item *curr, *next;
1625 list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1626 list_del(&curr->readdir_list);
1627 if (refcount_dec_and_test(&curr->refs))
1631 list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1632 list_del(&curr->readdir_list);
1633 if (refcount_dec_and_test(&curr->refs))
1638 * The VFS is going to do up_read(), so we need to downgrade back to a
1641 downgrade_write(&inode->i_rwsem);
1644 int btrfs_should_delete_dir_index(struct list_head *del_list,
1647 struct btrfs_delayed_item *curr;
1650 list_for_each_entry(curr, del_list, readdir_list) {
1651 if (curr->key.offset > index)
1653 if (curr->key.offset == index) {
1662 * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree
1665 int btrfs_readdir_delayed_dir_index(struct dir_context *ctx,
1666 struct list_head *ins_list)
1668 struct btrfs_dir_item *di;
1669 struct btrfs_delayed_item *curr, *next;
1670 struct btrfs_key location;
1674 unsigned char d_type;
1676 if (list_empty(ins_list))
1680 * Changing the data of the delayed item is impossible. So
1681 * we needn't lock them. And we have held i_mutex of the
1682 * directory, nobody can delete any directory indexes now.
1684 list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1685 list_del(&curr->readdir_list);
1687 if (curr->key.offset < ctx->pos) {
1688 if (refcount_dec_and_test(&curr->refs))
1693 ctx->pos = curr->key.offset;
1695 di = (struct btrfs_dir_item *)curr->data;
1696 name = (char *)(di + 1);
1697 name_len = btrfs_stack_dir_name_len(di);
1699 d_type = btrfs_filetype_table[di->type];
1700 btrfs_disk_key_to_cpu(&location, &di->location);
1702 over = !dir_emit(ctx, name, name_len,
1703 location.objectid, d_type);
1705 if (refcount_dec_and_test(&curr->refs))
1715 static void fill_stack_inode_item(struct btrfs_trans_handle *trans,
1716 struct btrfs_inode_item *inode_item,
1717 struct inode *inode)
1719 btrfs_set_stack_inode_uid(inode_item, i_uid_read(inode));
1720 btrfs_set_stack_inode_gid(inode_item, i_gid_read(inode));
1721 btrfs_set_stack_inode_size(inode_item, BTRFS_I(inode)->disk_i_size);
1722 btrfs_set_stack_inode_mode(inode_item, inode->i_mode);
1723 btrfs_set_stack_inode_nlink(inode_item, inode->i_nlink);
1724 btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(inode));
1725 btrfs_set_stack_inode_generation(inode_item,
1726 BTRFS_I(inode)->generation);
1727 btrfs_set_stack_inode_sequence(inode_item,
1728 inode_peek_iversion(inode));
1729 btrfs_set_stack_inode_transid(inode_item, trans->transid);
1730 btrfs_set_stack_inode_rdev(inode_item, inode->i_rdev);
1731 btrfs_set_stack_inode_flags(inode_item, BTRFS_I(inode)->flags);
1732 btrfs_set_stack_inode_block_group(inode_item, 0);
1734 btrfs_set_stack_timespec_sec(&inode_item->atime,
1735 inode->i_atime.tv_sec);
1736 btrfs_set_stack_timespec_nsec(&inode_item->atime,
1737 inode->i_atime.tv_nsec);
1739 btrfs_set_stack_timespec_sec(&inode_item->mtime,
1740 inode->i_mtime.tv_sec);
1741 btrfs_set_stack_timespec_nsec(&inode_item->mtime,
1742 inode->i_mtime.tv_nsec);
1744 btrfs_set_stack_timespec_sec(&inode_item->ctime,
1745 inode->i_ctime.tv_sec);
1746 btrfs_set_stack_timespec_nsec(&inode_item->ctime,
1747 inode->i_ctime.tv_nsec);
1749 btrfs_set_stack_timespec_sec(&inode_item->otime,
1750 BTRFS_I(inode)->i_otime.tv_sec);
1751 btrfs_set_stack_timespec_nsec(&inode_item->otime,
1752 BTRFS_I(inode)->i_otime.tv_nsec);
1755 int btrfs_fill_inode(struct inode *inode, u32 *rdev)
1757 struct btrfs_delayed_node *delayed_node;
1758 struct btrfs_inode_item *inode_item;
1760 delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
1764 mutex_lock(&delayed_node->mutex);
1765 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1766 mutex_unlock(&delayed_node->mutex);
1767 btrfs_release_delayed_node(delayed_node);
1771 inode_item = &delayed_node->inode_item;
1773 i_uid_write(inode, btrfs_stack_inode_uid(inode_item));
1774 i_gid_write(inode, btrfs_stack_inode_gid(inode_item));
1775 btrfs_i_size_write(BTRFS_I(inode), btrfs_stack_inode_size(inode_item));
1776 inode->i_mode = btrfs_stack_inode_mode(inode_item);
1777 set_nlink(inode, btrfs_stack_inode_nlink(inode_item));
1778 inode_set_bytes(inode, btrfs_stack_inode_nbytes(inode_item));
1779 BTRFS_I(inode)->generation = btrfs_stack_inode_generation(inode_item);
1780 BTRFS_I(inode)->last_trans = btrfs_stack_inode_transid(inode_item);
1782 inode_set_iversion_queried(inode,
1783 btrfs_stack_inode_sequence(inode_item));
1785 *rdev = btrfs_stack_inode_rdev(inode_item);
1786 BTRFS_I(inode)->flags = btrfs_stack_inode_flags(inode_item);
1788 inode->i_atime.tv_sec = btrfs_stack_timespec_sec(&inode_item->atime);
1789 inode->i_atime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->atime);
1791 inode->i_mtime.tv_sec = btrfs_stack_timespec_sec(&inode_item->mtime);
1792 inode->i_mtime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->mtime);
1794 inode->i_ctime.tv_sec = btrfs_stack_timespec_sec(&inode_item->ctime);
1795 inode->i_ctime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->ctime);
1797 BTRFS_I(inode)->i_otime.tv_sec =
1798 btrfs_stack_timespec_sec(&inode_item->otime);
1799 BTRFS_I(inode)->i_otime.tv_nsec =
1800 btrfs_stack_timespec_nsec(&inode_item->otime);
1802 inode->i_generation = BTRFS_I(inode)->generation;
1803 BTRFS_I(inode)->index_cnt = (u64)-1;
1805 mutex_unlock(&delayed_node->mutex);
1806 btrfs_release_delayed_node(delayed_node);
1810 int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans,
1811 struct btrfs_root *root, struct inode *inode)
1813 struct btrfs_delayed_node *delayed_node;
1816 delayed_node = btrfs_get_or_create_delayed_node(BTRFS_I(inode));
1817 if (IS_ERR(delayed_node))
1818 return PTR_ERR(delayed_node);
1820 mutex_lock(&delayed_node->mutex);
1821 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1822 fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1826 ret = btrfs_delayed_inode_reserve_metadata(trans, root, BTRFS_I(inode),
1831 fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1832 set_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
1833 delayed_node->count++;
1834 atomic_inc(&root->fs_info->delayed_root->items);
1836 mutex_unlock(&delayed_node->mutex);
1837 btrfs_release_delayed_node(delayed_node);
1841 int btrfs_delayed_delete_inode_ref(struct btrfs_inode *inode)
1843 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1844 struct btrfs_delayed_node *delayed_node;
1847 * we don't do delayed inode updates during log recovery because it
1848 * leads to enospc problems. This means we also can't do
1849 * delayed inode refs
1851 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
1854 delayed_node = btrfs_get_or_create_delayed_node(inode);
1855 if (IS_ERR(delayed_node))
1856 return PTR_ERR(delayed_node);
1859 * We don't reserve space for inode ref deletion is because:
1860 * - We ONLY do async inode ref deletion for the inode who has only
1861 * one link(i_nlink == 1), it means there is only one inode ref.
1862 * And in most case, the inode ref and the inode item are in the
1863 * same leaf, and we will deal with them at the same time.
1864 * Since we are sure we will reserve the space for the inode item,
1865 * it is unnecessary to reserve space for inode ref deletion.
1866 * - If the inode ref and the inode item are not in the same leaf,
1867 * We also needn't worry about enospc problem, because we reserve
1868 * much more space for the inode update than it needs.
1869 * - At the worst, we can steal some space from the global reservation.
1872 mutex_lock(&delayed_node->mutex);
1873 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1876 set_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
1877 delayed_node->count++;
1878 atomic_inc(&fs_info->delayed_root->items);
1880 mutex_unlock(&delayed_node->mutex);
1881 btrfs_release_delayed_node(delayed_node);
1885 static void __btrfs_kill_delayed_node(struct btrfs_delayed_node *delayed_node)
1887 struct btrfs_root *root = delayed_node->root;
1888 struct btrfs_fs_info *fs_info = root->fs_info;
1889 struct btrfs_delayed_item *curr_item, *prev_item;
1891 mutex_lock(&delayed_node->mutex);
1892 curr_item = __btrfs_first_delayed_insertion_item(delayed_node);
1894 btrfs_delayed_item_release_metadata(root, curr_item);
1895 prev_item = curr_item;
1896 curr_item = __btrfs_next_delayed_item(prev_item);
1897 btrfs_release_delayed_item(prev_item);
1900 curr_item = __btrfs_first_delayed_deletion_item(delayed_node);
1902 btrfs_delayed_item_release_metadata(root, curr_item);
1903 prev_item = curr_item;
1904 curr_item = __btrfs_next_delayed_item(prev_item);
1905 btrfs_release_delayed_item(prev_item);
1908 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1909 btrfs_release_delayed_iref(delayed_node);
1911 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1912 btrfs_delayed_inode_release_metadata(fs_info, delayed_node, false);
1913 btrfs_release_delayed_inode(delayed_node);
1915 mutex_unlock(&delayed_node->mutex);
1918 void btrfs_kill_delayed_inode_items(struct btrfs_inode *inode)
1920 struct btrfs_delayed_node *delayed_node;
1922 delayed_node = btrfs_get_delayed_node(inode);
1926 __btrfs_kill_delayed_node(delayed_node);
1927 btrfs_release_delayed_node(delayed_node);
1930 void btrfs_kill_all_delayed_nodes(struct btrfs_root *root)
1933 struct btrfs_delayed_node *delayed_nodes[8];
1937 spin_lock(&root->inode_lock);
1938 n = radix_tree_gang_lookup(&root->delayed_nodes_tree,
1939 (void **)delayed_nodes, inode_id,
1940 ARRAY_SIZE(delayed_nodes));
1942 spin_unlock(&root->inode_lock);
1946 inode_id = delayed_nodes[n - 1]->inode_id + 1;
1948 for (i = 0; i < n; i++)
1949 refcount_inc(&delayed_nodes[i]->refs);
1950 spin_unlock(&root->inode_lock);
1952 for (i = 0; i < n; i++) {
1953 __btrfs_kill_delayed_node(delayed_nodes[i]);
1954 btrfs_release_delayed_node(delayed_nodes[i]);
1959 void btrfs_destroy_delayed_inodes(struct btrfs_fs_info *fs_info)
1961 struct btrfs_delayed_node *curr_node, *prev_node;
1963 curr_node = btrfs_first_delayed_node(fs_info->delayed_root);
1965 __btrfs_kill_delayed_node(curr_node);
1967 prev_node = curr_node;
1968 curr_node = btrfs_next_delayed_node(curr_node);
1969 btrfs_release_delayed_node(prev_node);