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 <linux/sched/mm.h>
11 #include "delayed-inode.h"
13 #include "transaction.h"
18 #define BTRFS_DELAYED_WRITEBACK 512
19 #define BTRFS_DELAYED_BACKGROUND 128
20 #define BTRFS_DELAYED_BATCH 16
22 static struct kmem_cache *delayed_node_cache;
24 int __init btrfs_delayed_inode_init(void)
26 delayed_node_cache = kmem_cache_create("btrfs_delayed_node",
27 sizeof(struct btrfs_delayed_node),
31 if (!delayed_node_cache)
36 void __cold btrfs_delayed_inode_exit(void)
38 kmem_cache_destroy(delayed_node_cache);
41 static inline void btrfs_init_delayed_node(
42 struct btrfs_delayed_node *delayed_node,
43 struct btrfs_root *root, u64 inode_id)
45 delayed_node->root = root;
46 delayed_node->inode_id = inode_id;
47 refcount_set(&delayed_node->refs, 0);
48 delayed_node->ins_root = RB_ROOT_CACHED;
49 delayed_node->del_root = RB_ROOT_CACHED;
50 mutex_init(&delayed_node->mutex);
51 INIT_LIST_HEAD(&delayed_node->n_list);
52 INIT_LIST_HEAD(&delayed_node->p_list);
55 static inline int btrfs_is_continuous_delayed_item(
56 struct btrfs_delayed_item *item1,
57 struct btrfs_delayed_item *item2)
59 if (item1->key.type == BTRFS_DIR_INDEX_KEY &&
60 item1->key.objectid == item2->key.objectid &&
61 item1->key.type == item2->key.type &&
62 item1->key.offset + 1 == item2->key.offset)
67 static struct btrfs_delayed_node *btrfs_get_delayed_node(
68 struct btrfs_inode *btrfs_inode)
70 struct btrfs_root *root = btrfs_inode->root;
71 u64 ino = btrfs_ino(btrfs_inode);
72 struct btrfs_delayed_node *node;
74 node = READ_ONCE(btrfs_inode->delayed_node);
76 refcount_inc(&node->refs);
80 spin_lock(&root->inode_lock);
81 node = radix_tree_lookup(&root->delayed_nodes_tree, ino);
84 if (btrfs_inode->delayed_node) {
85 refcount_inc(&node->refs); /* can be accessed */
86 BUG_ON(btrfs_inode->delayed_node != node);
87 spin_unlock(&root->inode_lock);
92 * It's possible that we're racing into the middle of removing
93 * this node from the radix tree. In this case, the refcount
94 * was zero and it should never go back to one. Just return
95 * NULL like it was never in the radix at all; our release
96 * function is in the process of removing it.
98 * Some implementations of refcount_inc refuse to bump the
99 * refcount once it has hit zero. If we don't do this dance
100 * here, refcount_inc() may decide to just WARN_ONCE() instead
101 * of actually bumping the refcount.
103 * If this node is properly in the radix, we want to bump the
104 * refcount twice, once for the inode and once for this get
107 if (refcount_inc_not_zero(&node->refs)) {
108 refcount_inc(&node->refs);
109 btrfs_inode->delayed_node = node;
114 spin_unlock(&root->inode_lock);
117 spin_unlock(&root->inode_lock);
122 /* Will return either the node or PTR_ERR(-ENOMEM) */
123 static struct btrfs_delayed_node *btrfs_get_or_create_delayed_node(
124 struct btrfs_inode *btrfs_inode)
126 struct btrfs_delayed_node *node;
127 struct btrfs_root *root = btrfs_inode->root;
128 u64 ino = btrfs_ino(btrfs_inode);
132 node = btrfs_get_delayed_node(btrfs_inode);
136 node = kmem_cache_zalloc(delayed_node_cache, GFP_NOFS);
138 return ERR_PTR(-ENOMEM);
139 btrfs_init_delayed_node(node, root, ino);
141 /* cached in the btrfs inode and can be accessed */
142 refcount_set(&node->refs, 2);
144 ret = radix_tree_preload(GFP_NOFS);
146 kmem_cache_free(delayed_node_cache, node);
150 spin_lock(&root->inode_lock);
151 ret = radix_tree_insert(&root->delayed_nodes_tree, ino, node);
152 if (ret == -EEXIST) {
153 spin_unlock(&root->inode_lock);
154 kmem_cache_free(delayed_node_cache, node);
155 radix_tree_preload_end();
158 btrfs_inode->delayed_node = node;
159 spin_unlock(&root->inode_lock);
160 radix_tree_preload_end();
166 * Call it when holding delayed_node->mutex
168 * If mod = 1, add this node into the prepared list.
170 static void btrfs_queue_delayed_node(struct btrfs_delayed_root *root,
171 struct btrfs_delayed_node *node,
174 spin_lock(&root->lock);
175 if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
176 if (!list_empty(&node->p_list))
177 list_move_tail(&node->p_list, &root->prepare_list);
179 list_add_tail(&node->p_list, &root->prepare_list);
181 list_add_tail(&node->n_list, &root->node_list);
182 list_add_tail(&node->p_list, &root->prepare_list);
183 refcount_inc(&node->refs); /* inserted into list */
185 set_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
187 spin_unlock(&root->lock);
190 /* Call it when holding delayed_node->mutex */
191 static void btrfs_dequeue_delayed_node(struct btrfs_delayed_root *root,
192 struct btrfs_delayed_node *node)
194 spin_lock(&root->lock);
195 if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
197 refcount_dec(&node->refs); /* not in the list */
198 list_del_init(&node->n_list);
199 if (!list_empty(&node->p_list))
200 list_del_init(&node->p_list);
201 clear_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
203 spin_unlock(&root->lock);
206 static struct btrfs_delayed_node *btrfs_first_delayed_node(
207 struct btrfs_delayed_root *delayed_root)
210 struct btrfs_delayed_node *node = NULL;
212 spin_lock(&delayed_root->lock);
213 if (list_empty(&delayed_root->node_list))
216 p = delayed_root->node_list.next;
217 node = list_entry(p, struct btrfs_delayed_node, n_list);
218 refcount_inc(&node->refs);
220 spin_unlock(&delayed_root->lock);
225 static struct btrfs_delayed_node *btrfs_next_delayed_node(
226 struct btrfs_delayed_node *node)
228 struct btrfs_delayed_root *delayed_root;
230 struct btrfs_delayed_node *next = NULL;
232 delayed_root = node->root->fs_info->delayed_root;
233 spin_lock(&delayed_root->lock);
234 if (!test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
235 /* not in the list */
236 if (list_empty(&delayed_root->node_list))
238 p = delayed_root->node_list.next;
239 } else if (list_is_last(&node->n_list, &delayed_root->node_list))
242 p = node->n_list.next;
244 next = list_entry(p, struct btrfs_delayed_node, n_list);
245 refcount_inc(&next->refs);
247 spin_unlock(&delayed_root->lock);
252 static void __btrfs_release_delayed_node(
253 struct btrfs_delayed_node *delayed_node,
256 struct btrfs_delayed_root *delayed_root;
261 delayed_root = delayed_node->root->fs_info->delayed_root;
263 mutex_lock(&delayed_node->mutex);
264 if (delayed_node->count)
265 btrfs_queue_delayed_node(delayed_root, delayed_node, mod);
267 btrfs_dequeue_delayed_node(delayed_root, delayed_node);
268 mutex_unlock(&delayed_node->mutex);
270 if (refcount_dec_and_test(&delayed_node->refs)) {
271 struct btrfs_root *root = delayed_node->root;
273 spin_lock(&root->inode_lock);
275 * Once our refcount goes to zero, nobody is allowed to bump it
276 * back up. We can delete it now.
278 ASSERT(refcount_read(&delayed_node->refs) == 0);
279 radix_tree_delete(&root->delayed_nodes_tree,
280 delayed_node->inode_id);
281 spin_unlock(&root->inode_lock);
282 kmem_cache_free(delayed_node_cache, delayed_node);
286 static inline void btrfs_release_delayed_node(struct btrfs_delayed_node *node)
288 __btrfs_release_delayed_node(node, 0);
291 static struct btrfs_delayed_node *btrfs_first_prepared_delayed_node(
292 struct btrfs_delayed_root *delayed_root)
295 struct btrfs_delayed_node *node = NULL;
297 spin_lock(&delayed_root->lock);
298 if (list_empty(&delayed_root->prepare_list))
301 p = delayed_root->prepare_list.next;
303 node = list_entry(p, struct btrfs_delayed_node, p_list);
304 refcount_inc(&node->refs);
306 spin_unlock(&delayed_root->lock);
311 static inline void btrfs_release_prepared_delayed_node(
312 struct btrfs_delayed_node *node)
314 __btrfs_release_delayed_node(node, 1);
317 static struct btrfs_delayed_item *btrfs_alloc_delayed_item(u32 data_len)
319 struct btrfs_delayed_item *item;
320 item = kmalloc(sizeof(*item) + data_len, GFP_NOFS);
322 item->data_len = data_len;
323 item->ins_or_del = 0;
324 item->bytes_reserved = 0;
325 item->delayed_node = NULL;
326 refcount_set(&item->refs, 1);
332 * __btrfs_lookup_delayed_item - look up the delayed item by key
333 * @delayed_node: pointer to the delayed node
334 * @key: the key to look up
335 * @prev: used to store the prev item if the right item isn't found
336 * @next: used to store the next item if the right item isn't found
338 * Note: if we don't find the right item, we will return the prev item and
341 static struct btrfs_delayed_item *__btrfs_lookup_delayed_item(
342 struct rb_root *root,
343 struct btrfs_key *key,
344 struct btrfs_delayed_item **prev,
345 struct btrfs_delayed_item **next)
347 struct rb_node *node, *prev_node = NULL;
348 struct btrfs_delayed_item *delayed_item = NULL;
351 node = root->rb_node;
354 delayed_item = rb_entry(node, struct btrfs_delayed_item,
357 ret = btrfs_comp_cpu_keys(&delayed_item->key, key);
359 node = node->rb_right;
361 node = node->rb_left;
370 *prev = delayed_item;
371 else if ((node = rb_prev(prev_node)) != NULL) {
372 *prev = rb_entry(node, struct btrfs_delayed_item,
382 *next = delayed_item;
383 else if ((node = rb_next(prev_node)) != NULL) {
384 *next = rb_entry(node, struct btrfs_delayed_item,
392 static struct btrfs_delayed_item *__btrfs_lookup_delayed_insertion_item(
393 struct btrfs_delayed_node *delayed_node,
394 struct btrfs_key *key)
396 return __btrfs_lookup_delayed_item(&delayed_node->ins_root.rb_root, key,
400 static int __btrfs_add_delayed_item(struct btrfs_delayed_node *delayed_node,
401 struct btrfs_delayed_item *ins,
404 struct rb_node **p, *node;
405 struct rb_node *parent_node = NULL;
406 struct rb_root_cached *root;
407 struct btrfs_delayed_item *item;
409 bool leftmost = true;
411 if (action == BTRFS_DELAYED_INSERTION_ITEM)
412 root = &delayed_node->ins_root;
413 else if (action == BTRFS_DELAYED_DELETION_ITEM)
414 root = &delayed_node->del_root;
417 p = &root->rb_root.rb_node;
418 node = &ins->rb_node;
422 item = rb_entry(parent_node, struct btrfs_delayed_item,
425 cmp = btrfs_comp_cpu_keys(&item->key, &ins->key);
429 } else if (cmp > 0) {
436 rb_link_node(node, parent_node, p);
437 rb_insert_color_cached(node, root, leftmost);
438 ins->delayed_node = delayed_node;
439 ins->ins_or_del = action;
441 if (ins->key.type == BTRFS_DIR_INDEX_KEY &&
442 action == BTRFS_DELAYED_INSERTION_ITEM &&
443 ins->key.offset >= delayed_node->index_cnt)
444 delayed_node->index_cnt = ins->key.offset + 1;
446 delayed_node->count++;
447 atomic_inc(&delayed_node->root->fs_info->delayed_root->items);
451 static int __btrfs_add_delayed_insertion_item(struct btrfs_delayed_node *node,
452 struct btrfs_delayed_item *item)
454 return __btrfs_add_delayed_item(node, item,
455 BTRFS_DELAYED_INSERTION_ITEM);
458 static int __btrfs_add_delayed_deletion_item(struct btrfs_delayed_node *node,
459 struct btrfs_delayed_item *item)
461 return __btrfs_add_delayed_item(node, item,
462 BTRFS_DELAYED_DELETION_ITEM);
465 static void finish_one_item(struct btrfs_delayed_root *delayed_root)
467 int seq = atomic_inc_return(&delayed_root->items_seq);
469 /* atomic_dec_return implies a barrier */
470 if ((atomic_dec_return(&delayed_root->items) <
471 BTRFS_DELAYED_BACKGROUND || seq % BTRFS_DELAYED_BATCH == 0))
472 cond_wake_up_nomb(&delayed_root->wait);
475 static void __btrfs_remove_delayed_item(struct btrfs_delayed_item *delayed_item)
477 struct rb_root_cached *root;
478 struct btrfs_delayed_root *delayed_root;
480 /* Not associated with any delayed_node */
481 if (!delayed_item->delayed_node)
483 delayed_root = delayed_item->delayed_node->root->fs_info->delayed_root;
485 BUG_ON(!delayed_root);
486 BUG_ON(delayed_item->ins_or_del != BTRFS_DELAYED_DELETION_ITEM &&
487 delayed_item->ins_or_del != BTRFS_DELAYED_INSERTION_ITEM);
489 if (delayed_item->ins_or_del == BTRFS_DELAYED_INSERTION_ITEM)
490 root = &delayed_item->delayed_node->ins_root;
492 root = &delayed_item->delayed_node->del_root;
494 rb_erase_cached(&delayed_item->rb_node, root);
495 delayed_item->delayed_node->count--;
497 finish_one_item(delayed_root);
500 static void btrfs_release_delayed_item(struct btrfs_delayed_item *item)
503 __btrfs_remove_delayed_item(item);
504 if (refcount_dec_and_test(&item->refs))
509 static struct btrfs_delayed_item *__btrfs_first_delayed_insertion_item(
510 struct btrfs_delayed_node *delayed_node)
513 struct btrfs_delayed_item *item = NULL;
515 p = rb_first_cached(&delayed_node->ins_root);
517 item = rb_entry(p, struct btrfs_delayed_item, rb_node);
522 static struct btrfs_delayed_item *__btrfs_first_delayed_deletion_item(
523 struct btrfs_delayed_node *delayed_node)
526 struct btrfs_delayed_item *item = NULL;
528 p = rb_first_cached(&delayed_node->del_root);
530 item = rb_entry(p, struct btrfs_delayed_item, rb_node);
535 static struct btrfs_delayed_item *__btrfs_next_delayed_item(
536 struct btrfs_delayed_item *item)
539 struct btrfs_delayed_item *next = NULL;
541 p = rb_next(&item->rb_node);
543 next = rb_entry(p, struct btrfs_delayed_item, rb_node);
548 static int btrfs_delayed_item_reserve_metadata(struct btrfs_trans_handle *trans,
549 struct btrfs_root *root,
550 struct btrfs_delayed_item *item)
552 struct btrfs_block_rsv *src_rsv;
553 struct btrfs_block_rsv *dst_rsv;
554 struct btrfs_fs_info *fs_info = root->fs_info;
558 if (!trans->bytes_reserved)
561 src_rsv = trans->block_rsv;
562 dst_rsv = &fs_info->delayed_block_rsv;
564 num_bytes = btrfs_calc_insert_metadata_size(fs_info, 1);
567 * Here we migrate space rsv from transaction rsv, since have already
568 * reserved space when starting a transaction. So no need to reserve
571 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, true);
573 trace_btrfs_space_reservation(fs_info, "delayed_item",
576 item->bytes_reserved = num_bytes;
582 static void btrfs_delayed_item_release_metadata(struct btrfs_root *root,
583 struct btrfs_delayed_item *item)
585 struct btrfs_block_rsv *rsv;
586 struct btrfs_fs_info *fs_info = root->fs_info;
588 if (!item->bytes_reserved)
591 rsv = &fs_info->delayed_block_rsv;
593 * Check btrfs_delayed_item_reserve_metadata() to see why we don't need
594 * to release/reserve qgroup space.
596 trace_btrfs_space_reservation(fs_info, "delayed_item",
597 item->key.objectid, item->bytes_reserved,
599 btrfs_block_rsv_release(fs_info, rsv, item->bytes_reserved, NULL);
602 static int btrfs_delayed_inode_reserve_metadata(
603 struct btrfs_trans_handle *trans,
604 struct btrfs_root *root,
605 struct btrfs_delayed_node *node)
607 struct btrfs_fs_info *fs_info = root->fs_info;
608 struct btrfs_block_rsv *src_rsv;
609 struct btrfs_block_rsv *dst_rsv;
613 src_rsv = trans->block_rsv;
614 dst_rsv = &fs_info->delayed_block_rsv;
616 num_bytes = btrfs_calc_metadata_size(fs_info, 1);
619 * btrfs_dirty_inode will update the inode under btrfs_join_transaction
620 * which doesn't reserve space for speed. This is a problem since we
621 * still need to reserve space for this update, so try to reserve the
624 * Now if src_rsv == delalloc_block_rsv we'll let it just steal since
625 * we always reserve enough to update the inode item.
627 if (!src_rsv || (!trans->bytes_reserved &&
628 src_rsv->type != BTRFS_BLOCK_RSV_DELALLOC)) {
629 ret = btrfs_qgroup_reserve_meta(root, num_bytes,
630 BTRFS_QGROUP_RSV_META_PREALLOC, true);
633 ret = btrfs_block_rsv_add(root, dst_rsv, num_bytes,
634 BTRFS_RESERVE_NO_FLUSH);
635 /* NO_FLUSH could only fail with -ENOSPC */
636 ASSERT(ret == 0 || ret == -ENOSPC);
638 btrfs_qgroup_free_meta_prealloc(root, num_bytes);
640 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, true);
644 trace_btrfs_space_reservation(fs_info, "delayed_inode",
645 node->inode_id, num_bytes, 1);
646 node->bytes_reserved = num_bytes;
652 static void btrfs_delayed_inode_release_metadata(struct btrfs_fs_info *fs_info,
653 struct btrfs_delayed_node *node,
656 struct btrfs_block_rsv *rsv;
658 if (!node->bytes_reserved)
661 rsv = &fs_info->delayed_block_rsv;
662 trace_btrfs_space_reservation(fs_info, "delayed_inode",
663 node->inode_id, node->bytes_reserved, 0);
664 btrfs_block_rsv_release(fs_info, rsv, node->bytes_reserved, NULL);
666 btrfs_qgroup_free_meta_prealloc(node->root,
667 node->bytes_reserved);
669 btrfs_qgroup_convert_reserved_meta(node->root,
670 node->bytes_reserved);
671 node->bytes_reserved = 0;
675 * This helper will insert some continuous items into the same leaf according
676 * to the free space of the leaf.
678 static int btrfs_batch_insert_items(struct btrfs_root *root,
679 struct btrfs_path *path,
680 struct btrfs_delayed_item *item)
682 struct btrfs_delayed_item *curr, *next;
684 int total_data_size = 0, total_size = 0;
685 struct extent_buffer *leaf;
687 struct btrfs_key *keys;
689 struct list_head head;
695 BUG_ON(!path->nodes[0]);
697 leaf = path->nodes[0];
698 free_space = btrfs_leaf_free_space(leaf);
699 INIT_LIST_HEAD(&head);
705 * count the number of the continuous items that we can insert in batch
707 while (total_size + next->data_len + sizeof(struct btrfs_item) <=
709 total_data_size += next->data_len;
710 total_size += next->data_len + sizeof(struct btrfs_item);
711 list_add_tail(&next->tree_list, &head);
715 next = __btrfs_next_delayed_item(curr);
719 if (!btrfs_is_continuous_delayed_item(curr, next))
728 keys = kmalloc_array(nitems, sizeof(struct btrfs_key), GFP_NOFS);
734 data_size = kmalloc_array(nitems, sizeof(u32), GFP_NOFS);
740 /* get keys of all the delayed items */
742 list_for_each_entry(next, &head, tree_list) {
744 data_size[i] = next->data_len;
748 /* insert the keys of the items */
749 setup_items_for_insert(root, path, keys, data_size, nitems);
751 /* insert the dir index items */
752 slot = path->slots[0];
753 list_for_each_entry_safe(curr, next, &head, tree_list) {
754 data_ptr = btrfs_item_ptr(leaf, slot, char);
755 write_extent_buffer(leaf, &curr->data,
756 (unsigned long)data_ptr,
760 btrfs_delayed_item_release_metadata(root, curr);
762 list_del(&curr->tree_list);
763 btrfs_release_delayed_item(curr);
774 * This helper can just do simple insertion that needn't extend item for new
775 * data, such as directory name index insertion, inode insertion.
777 static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans,
778 struct btrfs_root *root,
779 struct btrfs_path *path,
780 struct btrfs_delayed_item *delayed_item)
782 struct extent_buffer *leaf;
783 unsigned int nofs_flag;
787 nofs_flag = memalloc_nofs_save();
788 ret = btrfs_insert_empty_item(trans, root, path, &delayed_item->key,
789 delayed_item->data_len);
790 memalloc_nofs_restore(nofs_flag);
791 if (ret < 0 && ret != -EEXIST)
794 leaf = path->nodes[0];
796 ptr = btrfs_item_ptr(leaf, path->slots[0], char);
798 write_extent_buffer(leaf, delayed_item->data, (unsigned long)ptr,
799 delayed_item->data_len);
800 btrfs_mark_buffer_dirty(leaf);
802 btrfs_delayed_item_release_metadata(root, delayed_item);
807 * we insert an item first, then if there are some continuous items, we try
808 * to insert those items into the same leaf.
810 static int btrfs_insert_delayed_items(struct btrfs_trans_handle *trans,
811 struct btrfs_path *path,
812 struct btrfs_root *root,
813 struct btrfs_delayed_node *node)
815 struct btrfs_delayed_item *curr, *prev;
819 mutex_lock(&node->mutex);
820 curr = __btrfs_first_delayed_insertion_item(node);
824 ret = btrfs_insert_delayed_item(trans, root, path, curr);
826 btrfs_release_path(path);
831 curr = __btrfs_next_delayed_item(prev);
832 if (curr && btrfs_is_continuous_delayed_item(prev, curr)) {
833 /* insert the continuous items into the same leaf */
835 btrfs_batch_insert_items(root, path, curr);
837 btrfs_release_delayed_item(prev);
838 btrfs_mark_buffer_dirty(path->nodes[0]);
840 btrfs_release_path(path);
841 mutex_unlock(&node->mutex);
845 mutex_unlock(&node->mutex);
849 static int btrfs_batch_delete_items(struct btrfs_trans_handle *trans,
850 struct btrfs_root *root,
851 struct btrfs_path *path,
852 struct btrfs_delayed_item *item)
854 struct btrfs_delayed_item *curr, *next;
855 struct extent_buffer *leaf;
856 struct btrfs_key key;
857 struct list_head head;
858 int nitems, i, last_item;
861 BUG_ON(!path->nodes[0]);
863 leaf = path->nodes[0];
866 last_item = btrfs_header_nritems(leaf) - 1;
868 return -ENOENT; /* FIXME: Is errno suitable? */
871 INIT_LIST_HEAD(&head);
872 btrfs_item_key_to_cpu(leaf, &key, i);
875 * count the number of the dir index items that we can delete in batch
877 while (btrfs_comp_cpu_keys(&next->key, &key) == 0) {
878 list_add_tail(&next->tree_list, &head);
882 next = __btrfs_next_delayed_item(curr);
886 if (!btrfs_is_continuous_delayed_item(curr, next))
892 btrfs_item_key_to_cpu(leaf, &key, i);
898 ret = btrfs_del_items(trans, root, path, path->slots[0], nitems);
902 list_for_each_entry_safe(curr, next, &head, tree_list) {
903 btrfs_delayed_item_release_metadata(root, curr);
904 list_del(&curr->tree_list);
905 btrfs_release_delayed_item(curr);
912 static int btrfs_delete_delayed_items(struct btrfs_trans_handle *trans,
913 struct btrfs_path *path,
914 struct btrfs_root *root,
915 struct btrfs_delayed_node *node)
917 struct btrfs_delayed_item *curr, *prev;
918 unsigned int nofs_flag;
922 mutex_lock(&node->mutex);
923 curr = __btrfs_first_delayed_deletion_item(node);
927 nofs_flag = memalloc_nofs_save();
928 ret = btrfs_search_slot(trans, root, &curr->key, path, -1, 1);
929 memalloc_nofs_restore(nofs_flag);
934 * can't find the item which the node points to, so this node
935 * is invalid, just drop it.
938 curr = __btrfs_next_delayed_item(prev);
939 btrfs_release_delayed_item(prev);
941 btrfs_release_path(path);
943 mutex_unlock(&node->mutex);
949 btrfs_batch_delete_items(trans, root, path, curr);
950 btrfs_release_path(path);
951 mutex_unlock(&node->mutex);
955 btrfs_release_path(path);
956 mutex_unlock(&node->mutex);
960 static void btrfs_release_delayed_inode(struct btrfs_delayed_node *delayed_node)
962 struct btrfs_delayed_root *delayed_root;
965 test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
966 BUG_ON(!delayed_node->root);
967 clear_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
968 delayed_node->count--;
970 delayed_root = delayed_node->root->fs_info->delayed_root;
971 finish_one_item(delayed_root);
975 static void btrfs_release_delayed_iref(struct btrfs_delayed_node *delayed_node)
977 struct btrfs_delayed_root *delayed_root;
979 ASSERT(delayed_node->root);
980 clear_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
981 delayed_node->count--;
983 delayed_root = delayed_node->root->fs_info->delayed_root;
984 finish_one_item(delayed_root);
987 static int __btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
988 struct btrfs_root *root,
989 struct btrfs_path *path,
990 struct btrfs_delayed_node *node)
992 struct btrfs_fs_info *fs_info = root->fs_info;
993 struct btrfs_key key;
994 struct btrfs_inode_item *inode_item;
995 struct extent_buffer *leaf;
996 unsigned int nofs_flag;
1000 key.objectid = node->inode_id;
1001 key.type = BTRFS_INODE_ITEM_KEY;
1004 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1009 nofs_flag = memalloc_nofs_save();
1010 ret = btrfs_lookup_inode(trans, root, path, &key, mod);
1011 memalloc_nofs_restore(nofs_flag);
1013 btrfs_release_path(path);
1015 } else if (ret < 0) {
1019 leaf = path->nodes[0];
1020 inode_item = btrfs_item_ptr(leaf, path->slots[0],
1021 struct btrfs_inode_item);
1022 write_extent_buffer(leaf, &node->inode_item, (unsigned long)inode_item,
1023 sizeof(struct btrfs_inode_item));
1024 btrfs_mark_buffer_dirty(leaf);
1026 if (!test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1030 if (path->slots[0] >= btrfs_header_nritems(leaf))
1033 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1034 if (key.objectid != node->inode_id)
1037 if (key.type != BTRFS_INODE_REF_KEY &&
1038 key.type != BTRFS_INODE_EXTREF_KEY)
1042 * Delayed iref deletion is for the inode who has only one link,
1043 * so there is only one iref. The case that several irefs are
1044 * in the same item doesn't exist.
1046 btrfs_del_item(trans, root, path);
1048 btrfs_release_delayed_iref(node);
1050 btrfs_release_path(path);
1052 btrfs_delayed_inode_release_metadata(fs_info, node, (ret < 0));
1053 btrfs_release_delayed_inode(node);
1058 btrfs_release_path(path);
1060 key.type = BTRFS_INODE_EXTREF_KEY;
1063 nofs_flag = memalloc_nofs_save();
1064 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1065 memalloc_nofs_restore(nofs_flag);
1071 leaf = path->nodes[0];
1076 static inline int btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1077 struct btrfs_root *root,
1078 struct btrfs_path *path,
1079 struct btrfs_delayed_node *node)
1083 mutex_lock(&node->mutex);
1084 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &node->flags)) {
1085 mutex_unlock(&node->mutex);
1089 ret = __btrfs_update_delayed_inode(trans, root, path, node);
1090 mutex_unlock(&node->mutex);
1095 __btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1096 struct btrfs_path *path,
1097 struct btrfs_delayed_node *node)
1101 ret = btrfs_insert_delayed_items(trans, path, node->root, node);
1105 ret = btrfs_delete_delayed_items(trans, path, node->root, node);
1109 ret = btrfs_update_delayed_inode(trans, node->root, path, node);
1114 * Called when committing the transaction.
1115 * Returns 0 on success.
1116 * Returns < 0 on error and returns with an aborted transaction with any
1117 * outstanding delayed items cleaned up.
1119 static int __btrfs_run_delayed_items(struct btrfs_trans_handle *trans, int nr)
1121 struct btrfs_fs_info *fs_info = trans->fs_info;
1122 struct btrfs_delayed_root *delayed_root;
1123 struct btrfs_delayed_node *curr_node, *prev_node;
1124 struct btrfs_path *path;
1125 struct btrfs_block_rsv *block_rsv;
1127 bool count = (nr > 0);
1129 if (TRANS_ABORTED(trans))
1132 path = btrfs_alloc_path();
1136 block_rsv = trans->block_rsv;
1137 trans->block_rsv = &fs_info->delayed_block_rsv;
1139 delayed_root = fs_info->delayed_root;
1141 curr_node = btrfs_first_delayed_node(delayed_root);
1142 while (curr_node && (!count || nr--)) {
1143 ret = __btrfs_commit_inode_delayed_items(trans, path,
1146 btrfs_release_delayed_node(curr_node);
1148 btrfs_abort_transaction(trans, ret);
1152 prev_node = curr_node;
1153 curr_node = btrfs_next_delayed_node(curr_node);
1154 btrfs_release_delayed_node(prev_node);
1158 btrfs_release_delayed_node(curr_node);
1159 btrfs_free_path(path);
1160 trans->block_rsv = block_rsv;
1165 int btrfs_run_delayed_items(struct btrfs_trans_handle *trans)
1167 return __btrfs_run_delayed_items(trans, -1);
1170 int btrfs_run_delayed_items_nr(struct btrfs_trans_handle *trans, int nr)
1172 return __btrfs_run_delayed_items(trans, nr);
1175 int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1176 struct btrfs_inode *inode)
1178 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1179 struct btrfs_path *path;
1180 struct btrfs_block_rsv *block_rsv;
1186 mutex_lock(&delayed_node->mutex);
1187 if (!delayed_node->count) {
1188 mutex_unlock(&delayed_node->mutex);
1189 btrfs_release_delayed_node(delayed_node);
1192 mutex_unlock(&delayed_node->mutex);
1194 path = btrfs_alloc_path();
1196 btrfs_release_delayed_node(delayed_node);
1200 block_rsv = trans->block_rsv;
1201 trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv;
1203 ret = __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1205 btrfs_release_delayed_node(delayed_node);
1206 btrfs_free_path(path);
1207 trans->block_rsv = block_rsv;
1212 int btrfs_commit_inode_delayed_inode(struct btrfs_inode *inode)
1214 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1215 struct btrfs_trans_handle *trans;
1216 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1217 struct btrfs_path *path;
1218 struct btrfs_block_rsv *block_rsv;
1224 mutex_lock(&delayed_node->mutex);
1225 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1226 mutex_unlock(&delayed_node->mutex);
1227 btrfs_release_delayed_node(delayed_node);
1230 mutex_unlock(&delayed_node->mutex);
1232 trans = btrfs_join_transaction(delayed_node->root);
1233 if (IS_ERR(trans)) {
1234 ret = PTR_ERR(trans);
1238 path = btrfs_alloc_path();
1244 block_rsv = trans->block_rsv;
1245 trans->block_rsv = &fs_info->delayed_block_rsv;
1247 mutex_lock(&delayed_node->mutex);
1248 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags))
1249 ret = __btrfs_update_delayed_inode(trans, delayed_node->root,
1250 path, delayed_node);
1253 mutex_unlock(&delayed_node->mutex);
1255 btrfs_free_path(path);
1256 trans->block_rsv = block_rsv;
1258 btrfs_end_transaction(trans);
1259 btrfs_btree_balance_dirty(fs_info);
1261 btrfs_release_delayed_node(delayed_node);
1266 void btrfs_remove_delayed_node(struct btrfs_inode *inode)
1268 struct btrfs_delayed_node *delayed_node;
1270 delayed_node = READ_ONCE(inode->delayed_node);
1274 inode->delayed_node = NULL;
1275 btrfs_release_delayed_node(delayed_node);
1278 struct btrfs_async_delayed_work {
1279 struct btrfs_delayed_root *delayed_root;
1281 struct btrfs_work work;
1284 static void btrfs_async_run_delayed_root(struct btrfs_work *work)
1286 struct btrfs_async_delayed_work *async_work;
1287 struct btrfs_delayed_root *delayed_root;
1288 struct btrfs_trans_handle *trans;
1289 struct btrfs_path *path;
1290 struct btrfs_delayed_node *delayed_node = NULL;
1291 struct btrfs_root *root;
1292 struct btrfs_block_rsv *block_rsv;
1295 async_work = container_of(work, struct btrfs_async_delayed_work, work);
1296 delayed_root = async_work->delayed_root;
1298 path = btrfs_alloc_path();
1303 if (atomic_read(&delayed_root->items) <
1304 BTRFS_DELAYED_BACKGROUND / 2)
1307 delayed_node = btrfs_first_prepared_delayed_node(delayed_root);
1311 root = delayed_node->root;
1313 trans = btrfs_join_transaction(root);
1314 if (IS_ERR(trans)) {
1315 btrfs_release_path(path);
1316 btrfs_release_prepared_delayed_node(delayed_node);
1321 block_rsv = trans->block_rsv;
1322 trans->block_rsv = &root->fs_info->delayed_block_rsv;
1324 __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1326 trans->block_rsv = block_rsv;
1327 btrfs_end_transaction(trans);
1328 btrfs_btree_balance_dirty_nodelay(root->fs_info);
1330 btrfs_release_path(path);
1331 btrfs_release_prepared_delayed_node(delayed_node);
1334 } while ((async_work->nr == 0 && total_done < BTRFS_DELAYED_WRITEBACK)
1335 || total_done < async_work->nr);
1337 btrfs_free_path(path);
1339 wake_up(&delayed_root->wait);
1344 static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root,
1345 struct btrfs_fs_info *fs_info, int nr)
1347 struct btrfs_async_delayed_work *async_work;
1349 async_work = kmalloc(sizeof(*async_work), GFP_NOFS);
1353 async_work->delayed_root = delayed_root;
1354 btrfs_init_work(&async_work->work, btrfs_async_run_delayed_root, NULL,
1356 async_work->nr = nr;
1358 btrfs_queue_work(fs_info->delayed_workers, &async_work->work);
1362 void btrfs_assert_delayed_root_empty(struct btrfs_fs_info *fs_info)
1364 WARN_ON(btrfs_first_delayed_node(fs_info->delayed_root));
1367 static int could_end_wait(struct btrfs_delayed_root *delayed_root, int seq)
1369 int val = atomic_read(&delayed_root->items_seq);
1371 if (val < seq || val >= seq + BTRFS_DELAYED_BATCH)
1374 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1380 void btrfs_balance_delayed_items(struct btrfs_fs_info *fs_info)
1382 struct btrfs_delayed_root *delayed_root = fs_info->delayed_root;
1384 if ((atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND) ||
1385 btrfs_workqueue_normal_congested(fs_info->delayed_workers))
1388 if (atomic_read(&delayed_root->items) >= BTRFS_DELAYED_WRITEBACK) {
1392 seq = atomic_read(&delayed_root->items_seq);
1394 ret = btrfs_wq_run_delayed_node(delayed_root, fs_info, 0);
1398 wait_event_interruptible(delayed_root->wait,
1399 could_end_wait(delayed_root, seq));
1403 btrfs_wq_run_delayed_node(delayed_root, fs_info, BTRFS_DELAYED_BATCH);
1406 /* Will return 0 or -ENOMEM */
1407 int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans,
1408 const char *name, int name_len,
1409 struct btrfs_inode *dir,
1410 struct btrfs_disk_key *disk_key, u8 type,
1413 struct btrfs_delayed_node *delayed_node;
1414 struct btrfs_delayed_item *delayed_item;
1415 struct btrfs_dir_item *dir_item;
1418 delayed_node = btrfs_get_or_create_delayed_node(dir);
1419 if (IS_ERR(delayed_node))
1420 return PTR_ERR(delayed_node);
1422 delayed_item = btrfs_alloc_delayed_item(sizeof(*dir_item) + name_len);
1423 if (!delayed_item) {
1428 delayed_item->key.objectid = btrfs_ino(dir);
1429 delayed_item->key.type = BTRFS_DIR_INDEX_KEY;
1430 delayed_item->key.offset = index;
1432 dir_item = (struct btrfs_dir_item *)delayed_item->data;
1433 dir_item->location = *disk_key;
1434 btrfs_set_stack_dir_transid(dir_item, trans->transid);
1435 btrfs_set_stack_dir_data_len(dir_item, 0);
1436 btrfs_set_stack_dir_name_len(dir_item, name_len);
1437 btrfs_set_stack_dir_type(dir_item, type);
1438 memcpy((char *)(dir_item + 1), name, name_len);
1440 ret = btrfs_delayed_item_reserve_metadata(trans, dir->root, delayed_item);
1442 * we have reserved enough space when we start a new transaction,
1443 * so reserving metadata failure is impossible
1447 mutex_lock(&delayed_node->mutex);
1448 ret = __btrfs_add_delayed_insertion_item(delayed_node, delayed_item);
1449 if (unlikely(ret)) {
1450 btrfs_err(trans->fs_info,
1451 "err add delayed dir index item(name: %.*s) into the insertion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
1452 name_len, name, delayed_node->root->root_key.objectid,
1453 delayed_node->inode_id, ret);
1456 mutex_unlock(&delayed_node->mutex);
1459 btrfs_release_delayed_node(delayed_node);
1463 static int btrfs_delete_delayed_insertion_item(struct btrfs_fs_info *fs_info,
1464 struct btrfs_delayed_node *node,
1465 struct btrfs_key *key)
1467 struct btrfs_delayed_item *item;
1469 mutex_lock(&node->mutex);
1470 item = __btrfs_lookup_delayed_insertion_item(node, key);
1472 mutex_unlock(&node->mutex);
1476 btrfs_delayed_item_release_metadata(node->root, item);
1477 btrfs_release_delayed_item(item);
1478 mutex_unlock(&node->mutex);
1482 int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans,
1483 struct btrfs_inode *dir, u64 index)
1485 struct btrfs_delayed_node *node;
1486 struct btrfs_delayed_item *item;
1487 struct btrfs_key item_key;
1490 node = btrfs_get_or_create_delayed_node(dir);
1492 return PTR_ERR(node);
1494 item_key.objectid = btrfs_ino(dir);
1495 item_key.type = BTRFS_DIR_INDEX_KEY;
1496 item_key.offset = index;
1498 ret = btrfs_delete_delayed_insertion_item(trans->fs_info, node,
1503 item = btrfs_alloc_delayed_item(0);
1509 item->key = item_key;
1511 ret = btrfs_delayed_item_reserve_metadata(trans, dir->root, item);
1513 * we have reserved enough space when we start a new transaction,
1514 * so reserving metadata failure is impossible.
1517 btrfs_err(trans->fs_info,
1518 "metadata reservation failed for delayed dir item deltiona, should have been reserved");
1519 btrfs_release_delayed_item(item);
1523 mutex_lock(&node->mutex);
1524 ret = __btrfs_add_delayed_deletion_item(node, item);
1525 if (unlikely(ret)) {
1526 btrfs_err(trans->fs_info,
1527 "err add delayed dir index item(index: %llu) into the deletion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
1528 index, node->root->root_key.objectid,
1529 node->inode_id, ret);
1530 btrfs_delayed_item_release_metadata(dir->root, item);
1531 btrfs_release_delayed_item(item);
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 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
1577 btrfs_inode_lock(inode, 0);
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;
1639 list_for_each_entry(curr, del_list, readdir_list) {
1640 if (curr->key.offset > index)
1642 if (curr->key.offset == index) {
1651 * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree
1654 int btrfs_readdir_delayed_dir_index(struct dir_context *ctx,
1655 struct list_head *ins_list)
1657 struct btrfs_dir_item *di;
1658 struct btrfs_delayed_item *curr, *next;
1659 struct btrfs_key location;
1663 unsigned char d_type;
1665 if (list_empty(ins_list))
1669 * Changing the data of the delayed item is impossible. So
1670 * we needn't lock them. And we have held i_mutex of the
1671 * directory, nobody can delete any directory indexes now.
1673 list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1674 list_del(&curr->readdir_list);
1676 if (curr->key.offset < ctx->pos) {
1677 if (refcount_dec_and_test(&curr->refs))
1682 ctx->pos = curr->key.offset;
1684 di = (struct btrfs_dir_item *)curr->data;
1685 name = (char *)(di + 1);
1686 name_len = btrfs_stack_dir_name_len(di);
1688 d_type = fs_ftype_to_dtype(di->type);
1689 btrfs_disk_key_to_cpu(&location, &di->location);
1691 over = !dir_emit(ctx, name, name_len,
1692 location.objectid, d_type);
1694 if (refcount_dec_and_test(&curr->refs))
1704 static void fill_stack_inode_item(struct btrfs_trans_handle *trans,
1705 struct btrfs_inode_item *inode_item,
1706 struct inode *inode)
1708 btrfs_set_stack_inode_uid(inode_item, i_uid_read(inode));
1709 btrfs_set_stack_inode_gid(inode_item, i_gid_read(inode));
1710 btrfs_set_stack_inode_size(inode_item, BTRFS_I(inode)->disk_i_size);
1711 btrfs_set_stack_inode_mode(inode_item, inode->i_mode);
1712 btrfs_set_stack_inode_nlink(inode_item, inode->i_nlink);
1713 btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(inode));
1714 btrfs_set_stack_inode_generation(inode_item,
1715 BTRFS_I(inode)->generation);
1716 btrfs_set_stack_inode_sequence(inode_item,
1717 inode_peek_iversion(inode));
1718 btrfs_set_stack_inode_transid(inode_item, trans->transid);
1719 btrfs_set_stack_inode_rdev(inode_item, inode->i_rdev);
1720 btrfs_set_stack_inode_flags(inode_item, BTRFS_I(inode)->flags);
1721 btrfs_set_stack_inode_block_group(inode_item, 0);
1723 btrfs_set_stack_timespec_sec(&inode_item->atime,
1724 inode->i_atime.tv_sec);
1725 btrfs_set_stack_timespec_nsec(&inode_item->atime,
1726 inode->i_atime.tv_nsec);
1728 btrfs_set_stack_timespec_sec(&inode_item->mtime,
1729 inode->i_mtime.tv_sec);
1730 btrfs_set_stack_timespec_nsec(&inode_item->mtime,
1731 inode->i_mtime.tv_nsec);
1733 btrfs_set_stack_timespec_sec(&inode_item->ctime,
1734 inode->i_ctime.tv_sec);
1735 btrfs_set_stack_timespec_nsec(&inode_item->ctime,
1736 inode->i_ctime.tv_nsec);
1738 btrfs_set_stack_timespec_sec(&inode_item->otime,
1739 BTRFS_I(inode)->i_otime.tv_sec);
1740 btrfs_set_stack_timespec_nsec(&inode_item->otime,
1741 BTRFS_I(inode)->i_otime.tv_nsec);
1744 int btrfs_fill_inode(struct inode *inode, u32 *rdev)
1746 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1747 struct btrfs_delayed_node *delayed_node;
1748 struct btrfs_inode_item *inode_item;
1750 delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
1754 mutex_lock(&delayed_node->mutex);
1755 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1756 mutex_unlock(&delayed_node->mutex);
1757 btrfs_release_delayed_node(delayed_node);
1761 inode_item = &delayed_node->inode_item;
1763 i_uid_write(inode, btrfs_stack_inode_uid(inode_item));
1764 i_gid_write(inode, btrfs_stack_inode_gid(inode_item));
1765 btrfs_i_size_write(BTRFS_I(inode), btrfs_stack_inode_size(inode_item));
1766 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
1767 round_up(i_size_read(inode), fs_info->sectorsize));
1768 inode->i_mode = btrfs_stack_inode_mode(inode_item);
1769 set_nlink(inode, btrfs_stack_inode_nlink(inode_item));
1770 inode_set_bytes(inode, btrfs_stack_inode_nbytes(inode_item));
1771 BTRFS_I(inode)->generation = btrfs_stack_inode_generation(inode_item);
1772 BTRFS_I(inode)->last_trans = btrfs_stack_inode_transid(inode_item);
1774 inode_set_iversion_queried(inode,
1775 btrfs_stack_inode_sequence(inode_item));
1777 *rdev = btrfs_stack_inode_rdev(inode_item);
1778 BTRFS_I(inode)->flags = btrfs_stack_inode_flags(inode_item);
1780 inode->i_atime.tv_sec = btrfs_stack_timespec_sec(&inode_item->atime);
1781 inode->i_atime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->atime);
1783 inode->i_mtime.tv_sec = btrfs_stack_timespec_sec(&inode_item->mtime);
1784 inode->i_mtime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->mtime);
1786 inode->i_ctime.tv_sec = btrfs_stack_timespec_sec(&inode_item->ctime);
1787 inode->i_ctime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->ctime);
1789 BTRFS_I(inode)->i_otime.tv_sec =
1790 btrfs_stack_timespec_sec(&inode_item->otime);
1791 BTRFS_I(inode)->i_otime.tv_nsec =
1792 btrfs_stack_timespec_nsec(&inode_item->otime);
1794 inode->i_generation = BTRFS_I(inode)->generation;
1795 BTRFS_I(inode)->index_cnt = (u64)-1;
1797 mutex_unlock(&delayed_node->mutex);
1798 btrfs_release_delayed_node(delayed_node);
1802 int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans,
1803 struct btrfs_root *root,
1804 struct btrfs_inode *inode)
1806 struct btrfs_delayed_node *delayed_node;
1809 delayed_node = btrfs_get_or_create_delayed_node(inode);
1810 if (IS_ERR(delayed_node))
1811 return PTR_ERR(delayed_node);
1813 mutex_lock(&delayed_node->mutex);
1814 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1815 fill_stack_inode_item(trans, &delayed_node->inode_item,
1820 ret = btrfs_delayed_inode_reserve_metadata(trans, root, delayed_node);
1824 fill_stack_inode_item(trans, &delayed_node->inode_item, &inode->vfs_inode);
1825 set_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
1826 delayed_node->count++;
1827 atomic_inc(&root->fs_info->delayed_root->items);
1829 mutex_unlock(&delayed_node->mutex);
1830 btrfs_release_delayed_node(delayed_node);
1834 int btrfs_delayed_delete_inode_ref(struct btrfs_inode *inode)
1836 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1837 struct btrfs_delayed_node *delayed_node;
1840 * we don't do delayed inode updates during log recovery because it
1841 * leads to enospc problems. This means we also can't do
1842 * delayed inode refs
1844 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
1847 delayed_node = btrfs_get_or_create_delayed_node(inode);
1848 if (IS_ERR(delayed_node))
1849 return PTR_ERR(delayed_node);
1852 * We don't reserve space for inode ref deletion is because:
1853 * - We ONLY do async inode ref deletion for the inode who has only
1854 * one link(i_nlink == 1), it means there is only one inode ref.
1855 * And in most case, the inode ref and the inode item are in the
1856 * same leaf, and we will deal with them at the same time.
1857 * Since we are sure we will reserve the space for the inode item,
1858 * it is unnecessary to reserve space for inode ref deletion.
1859 * - If the inode ref and the inode item are not in the same leaf,
1860 * We also needn't worry about enospc problem, because we reserve
1861 * much more space for the inode update than it needs.
1862 * - At the worst, we can steal some space from the global reservation.
1865 mutex_lock(&delayed_node->mutex);
1866 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1869 set_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
1870 delayed_node->count++;
1871 atomic_inc(&fs_info->delayed_root->items);
1873 mutex_unlock(&delayed_node->mutex);
1874 btrfs_release_delayed_node(delayed_node);
1878 static void __btrfs_kill_delayed_node(struct btrfs_delayed_node *delayed_node)
1880 struct btrfs_root *root = delayed_node->root;
1881 struct btrfs_fs_info *fs_info = root->fs_info;
1882 struct btrfs_delayed_item *curr_item, *prev_item;
1884 mutex_lock(&delayed_node->mutex);
1885 curr_item = __btrfs_first_delayed_insertion_item(delayed_node);
1887 btrfs_delayed_item_release_metadata(root, curr_item);
1888 prev_item = curr_item;
1889 curr_item = __btrfs_next_delayed_item(prev_item);
1890 btrfs_release_delayed_item(prev_item);
1893 curr_item = __btrfs_first_delayed_deletion_item(delayed_node);
1895 btrfs_delayed_item_release_metadata(root, curr_item);
1896 prev_item = curr_item;
1897 curr_item = __btrfs_next_delayed_item(prev_item);
1898 btrfs_release_delayed_item(prev_item);
1901 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1902 btrfs_release_delayed_iref(delayed_node);
1904 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1905 btrfs_delayed_inode_release_metadata(fs_info, delayed_node, false);
1906 btrfs_release_delayed_inode(delayed_node);
1908 mutex_unlock(&delayed_node->mutex);
1911 void btrfs_kill_delayed_inode_items(struct btrfs_inode *inode)
1913 struct btrfs_delayed_node *delayed_node;
1915 delayed_node = btrfs_get_delayed_node(inode);
1919 __btrfs_kill_delayed_node(delayed_node);
1920 btrfs_release_delayed_node(delayed_node);
1923 void btrfs_kill_all_delayed_nodes(struct btrfs_root *root)
1926 struct btrfs_delayed_node *delayed_nodes[8];
1930 spin_lock(&root->inode_lock);
1931 n = radix_tree_gang_lookup(&root->delayed_nodes_tree,
1932 (void **)delayed_nodes, inode_id,
1933 ARRAY_SIZE(delayed_nodes));
1935 spin_unlock(&root->inode_lock);
1939 inode_id = delayed_nodes[n - 1]->inode_id + 1;
1940 for (i = 0; i < n; i++) {
1942 * Don't increase refs in case the node is dead and
1943 * about to be removed from the tree in the loop below
1945 if (!refcount_inc_not_zero(&delayed_nodes[i]->refs))
1946 delayed_nodes[i] = NULL;
1948 spin_unlock(&root->inode_lock);
1950 for (i = 0; i < n; i++) {
1951 if (!delayed_nodes[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);