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_inode *inode,
606 struct btrfs_delayed_node *node)
608 struct btrfs_fs_info *fs_info = root->fs_info;
609 struct btrfs_block_rsv *src_rsv;
610 struct btrfs_block_rsv *dst_rsv;
614 src_rsv = trans->block_rsv;
615 dst_rsv = &fs_info->delayed_block_rsv;
617 num_bytes = btrfs_calc_metadata_size(fs_info, 1);
620 * btrfs_dirty_inode will update the inode under btrfs_join_transaction
621 * which doesn't reserve space for speed. This is a problem since we
622 * still need to reserve space for this update, so try to reserve the
625 * Now if src_rsv == delalloc_block_rsv we'll let it just steal since
626 * we always reserve enough to update the inode item.
628 if (!src_rsv || (!trans->bytes_reserved &&
629 src_rsv->type != BTRFS_BLOCK_RSV_DELALLOC)) {
630 ret = btrfs_qgroup_reserve_meta(root, num_bytes,
631 BTRFS_QGROUP_RSV_META_PREALLOC, true);
634 ret = btrfs_block_rsv_add(root, dst_rsv, num_bytes,
635 BTRFS_RESERVE_NO_FLUSH);
637 * Since we're under a transaction reserve_metadata_bytes could
638 * try to commit the transaction which will make it return
639 * EAGAIN to make us stop the transaction we have, so return
640 * ENOSPC instead so that btrfs_dirty_inode knows what to do.
642 if (ret == -EAGAIN) {
644 btrfs_qgroup_free_meta_prealloc(root, num_bytes);
647 node->bytes_reserved = num_bytes;
648 trace_btrfs_space_reservation(fs_info,
653 btrfs_qgroup_free_meta_prealloc(root, num_bytes);
658 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, true);
660 trace_btrfs_space_reservation(fs_info, "delayed_inode",
661 btrfs_ino(inode), num_bytes, 1);
662 node->bytes_reserved = num_bytes;
668 static void btrfs_delayed_inode_release_metadata(struct btrfs_fs_info *fs_info,
669 struct btrfs_delayed_node *node,
672 struct btrfs_block_rsv *rsv;
674 if (!node->bytes_reserved)
677 rsv = &fs_info->delayed_block_rsv;
678 trace_btrfs_space_reservation(fs_info, "delayed_inode",
679 node->inode_id, node->bytes_reserved, 0);
680 btrfs_block_rsv_release(fs_info, rsv, node->bytes_reserved, NULL);
682 btrfs_qgroup_free_meta_prealloc(node->root,
683 node->bytes_reserved);
685 btrfs_qgroup_convert_reserved_meta(node->root,
686 node->bytes_reserved);
687 node->bytes_reserved = 0;
691 * This helper will insert some continuous items into the same leaf according
692 * to the free space of the leaf.
694 static int btrfs_batch_insert_items(struct btrfs_root *root,
695 struct btrfs_path *path,
696 struct btrfs_delayed_item *item)
698 struct btrfs_delayed_item *curr, *next;
700 int total_data_size = 0, total_size = 0;
701 struct extent_buffer *leaf;
703 struct btrfs_key *keys;
705 struct list_head head;
711 BUG_ON(!path->nodes[0]);
713 leaf = path->nodes[0];
714 free_space = btrfs_leaf_free_space(leaf);
715 INIT_LIST_HEAD(&head);
721 * count the number of the continuous items that we can insert in batch
723 while (total_size + next->data_len + sizeof(struct btrfs_item) <=
725 total_data_size += next->data_len;
726 total_size += next->data_len + sizeof(struct btrfs_item);
727 list_add_tail(&next->tree_list, &head);
731 next = __btrfs_next_delayed_item(curr);
735 if (!btrfs_is_continuous_delayed_item(curr, next))
745 * we need allocate some memory space, but it might cause the task
746 * to sleep, so we set all locked nodes in the path to blocking locks
749 btrfs_set_path_blocking(path);
751 keys = kmalloc_array(nitems, sizeof(struct btrfs_key), GFP_NOFS);
757 data_size = kmalloc_array(nitems, sizeof(u32), GFP_NOFS);
763 /* get keys of all the delayed items */
765 list_for_each_entry(next, &head, tree_list) {
767 data_size[i] = next->data_len;
771 /* insert the keys of the items */
772 setup_items_for_insert(root, path, keys, data_size, nitems);
774 /* insert the dir index items */
775 slot = path->slots[0];
776 list_for_each_entry_safe(curr, next, &head, tree_list) {
777 data_ptr = btrfs_item_ptr(leaf, slot, char);
778 write_extent_buffer(leaf, &curr->data,
779 (unsigned long)data_ptr,
783 btrfs_delayed_item_release_metadata(root, curr);
785 list_del(&curr->tree_list);
786 btrfs_release_delayed_item(curr);
797 * This helper can just do simple insertion that needn't extend item for new
798 * data, such as directory name index insertion, inode insertion.
800 static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans,
801 struct btrfs_root *root,
802 struct btrfs_path *path,
803 struct btrfs_delayed_item *delayed_item)
805 struct extent_buffer *leaf;
806 unsigned int nofs_flag;
810 nofs_flag = memalloc_nofs_save();
811 ret = btrfs_insert_empty_item(trans, root, path, &delayed_item->key,
812 delayed_item->data_len);
813 memalloc_nofs_restore(nofs_flag);
814 if (ret < 0 && ret != -EEXIST)
817 leaf = path->nodes[0];
819 ptr = btrfs_item_ptr(leaf, path->slots[0], char);
821 write_extent_buffer(leaf, delayed_item->data, (unsigned long)ptr,
822 delayed_item->data_len);
823 btrfs_mark_buffer_dirty(leaf);
825 btrfs_delayed_item_release_metadata(root, delayed_item);
830 * we insert an item first, then if there are some continuous items, we try
831 * to insert those items into the same leaf.
833 static int btrfs_insert_delayed_items(struct btrfs_trans_handle *trans,
834 struct btrfs_path *path,
835 struct btrfs_root *root,
836 struct btrfs_delayed_node *node)
838 struct btrfs_delayed_item *curr, *prev;
842 mutex_lock(&node->mutex);
843 curr = __btrfs_first_delayed_insertion_item(node);
847 ret = btrfs_insert_delayed_item(trans, root, path, curr);
849 btrfs_release_path(path);
854 curr = __btrfs_next_delayed_item(prev);
855 if (curr && btrfs_is_continuous_delayed_item(prev, curr)) {
856 /* insert the continuous items into the same leaf */
858 btrfs_batch_insert_items(root, path, curr);
860 btrfs_release_delayed_item(prev);
861 btrfs_mark_buffer_dirty(path->nodes[0]);
863 btrfs_release_path(path);
864 mutex_unlock(&node->mutex);
868 mutex_unlock(&node->mutex);
872 static int btrfs_batch_delete_items(struct btrfs_trans_handle *trans,
873 struct btrfs_root *root,
874 struct btrfs_path *path,
875 struct btrfs_delayed_item *item)
877 struct btrfs_delayed_item *curr, *next;
878 struct extent_buffer *leaf;
879 struct btrfs_key key;
880 struct list_head head;
881 int nitems, i, last_item;
884 BUG_ON(!path->nodes[0]);
886 leaf = path->nodes[0];
889 last_item = btrfs_header_nritems(leaf) - 1;
891 return -ENOENT; /* FIXME: Is errno suitable? */
894 INIT_LIST_HEAD(&head);
895 btrfs_item_key_to_cpu(leaf, &key, i);
898 * count the number of the dir index items that we can delete in batch
900 while (btrfs_comp_cpu_keys(&next->key, &key) == 0) {
901 list_add_tail(&next->tree_list, &head);
905 next = __btrfs_next_delayed_item(curr);
909 if (!btrfs_is_continuous_delayed_item(curr, next))
915 btrfs_item_key_to_cpu(leaf, &key, i);
921 ret = btrfs_del_items(trans, root, path, path->slots[0], nitems);
925 list_for_each_entry_safe(curr, next, &head, tree_list) {
926 btrfs_delayed_item_release_metadata(root, curr);
927 list_del(&curr->tree_list);
928 btrfs_release_delayed_item(curr);
935 static int btrfs_delete_delayed_items(struct btrfs_trans_handle *trans,
936 struct btrfs_path *path,
937 struct btrfs_root *root,
938 struct btrfs_delayed_node *node)
940 struct btrfs_delayed_item *curr, *prev;
941 unsigned int nofs_flag;
945 mutex_lock(&node->mutex);
946 curr = __btrfs_first_delayed_deletion_item(node);
950 nofs_flag = memalloc_nofs_save();
951 ret = btrfs_search_slot(trans, root, &curr->key, path, -1, 1);
952 memalloc_nofs_restore(nofs_flag);
957 * can't find the item which the node points to, so this node
958 * is invalid, just drop it.
961 curr = __btrfs_next_delayed_item(prev);
962 btrfs_release_delayed_item(prev);
964 btrfs_release_path(path);
966 mutex_unlock(&node->mutex);
972 btrfs_batch_delete_items(trans, root, path, curr);
973 btrfs_release_path(path);
974 mutex_unlock(&node->mutex);
978 btrfs_release_path(path);
979 mutex_unlock(&node->mutex);
983 static void btrfs_release_delayed_inode(struct btrfs_delayed_node *delayed_node)
985 struct btrfs_delayed_root *delayed_root;
988 test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
989 BUG_ON(!delayed_node->root);
990 clear_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
991 delayed_node->count--;
993 delayed_root = delayed_node->root->fs_info->delayed_root;
994 finish_one_item(delayed_root);
998 static void btrfs_release_delayed_iref(struct btrfs_delayed_node *delayed_node)
1000 struct btrfs_delayed_root *delayed_root;
1002 ASSERT(delayed_node->root);
1003 clear_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
1004 delayed_node->count--;
1006 delayed_root = delayed_node->root->fs_info->delayed_root;
1007 finish_one_item(delayed_root);
1010 static int __btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1011 struct btrfs_root *root,
1012 struct btrfs_path *path,
1013 struct btrfs_delayed_node *node)
1015 struct btrfs_fs_info *fs_info = root->fs_info;
1016 struct btrfs_key key;
1017 struct btrfs_inode_item *inode_item;
1018 struct extent_buffer *leaf;
1019 unsigned int nofs_flag;
1023 key.objectid = node->inode_id;
1024 key.type = BTRFS_INODE_ITEM_KEY;
1027 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1032 nofs_flag = memalloc_nofs_save();
1033 ret = btrfs_lookup_inode(trans, root, path, &key, mod);
1034 memalloc_nofs_restore(nofs_flag);
1040 leaf = path->nodes[0];
1041 inode_item = btrfs_item_ptr(leaf, path->slots[0],
1042 struct btrfs_inode_item);
1043 write_extent_buffer(leaf, &node->inode_item, (unsigned long)inode_item,
1044 sizeof(struct btrfs_inode_item));
1045 btrfs_mark_buffer_dirty(leaf);
1047 if (!test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1051 if (path->slots[0] >= btrfs_header_nritems(leaf))
1054 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1055 if (key.objectid != node->inode_id)
1058 if (key.type != BTRFS_INODE_REF_KEY &&
1059 key.type != BTRFS_INODE_EXTREF_KEY)
1063 * Delayed iref deletion is for the inode who has only one link,
1064 * so there is only one iref. The case that several irefs are
1065 * in the same item doesn't exist.
1067 btrfs_del_item(trans, root, path);
1069 btrfs_release_delayed_iref(node);
1071 btrfs_release_path(path);
1073 btrfs_delayed_inode_release_metadata(fs_info, node, (ret < 0));
1074 btrfs_release_delayed_inode(node);
1077 * If we fail to update the delayed inode we need to abort the
1078 * transaction, because we could leave the inode with the improper
1081 if (ret && ret != -ENOENT)
1082 btrfs_abort_transaction(trans, ret);
1087 btrfs_release_path(path);
1089 key.type = BTRFS_INODE_EXTREF_KEY;
1092 nofs_flag = memalloc_nofs_save();
1093 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1094 memalloc_nofs_restore(nofs_flag);
1100 leaf = path->nodes[0];
1105 static inline int btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1106 struct btrfs_root *root,
1107 struct btrfs_path *path,
1108 struct btrfs_delayed_node *node)
1112 mutex_lock(&node->mutex);
1113 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &node->flags)) {
1114 mutex_unlock(&node->mutex);
1118 ret = __btrfs_update_delayed_inode(trans, root, path, node);
1119 mutex_unlock(&node->mutex);
1124 __btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1125 struct btrfs_path *path,
1126 struct btrfs_delayed_node *node)
1130 ret = btrfs_insert_delayed_items(trans, path, node->root, node);
1134 ret = btrfs_delete_delayed_items(trans, path, node->root, node);
1138 ret = btrfs_record_root_in_trans(trans, node->root);
1141 ret = btrfs_update_delayed_inode(trans, node->root, path, node);
1146 * Called when committing the transaction.
1147 * Returns 0 on success.
1148 * Returns < 0 on error and returns with an aborted transaction with any
1149 * outstanding delayed items cleaned up.
1151 static int __btrfs_run_delayed_items(struct btrfs_trans_handle *trans, int nr)
1153 struct btrfs_fs_info *fs_info = trans->fs_info;
1154 struct btrfs_delayed_root *delayed_root;
1155 struct btrfs_delayed_node *curr_node, *prev_node;
1156 struct btrfs_path *path;
1157 struct btrfs_block_rsv *block_rsv;
1159 bool count = (nr > 0);
1161 if (TRANS_ABORTED(trans))
1164 path = btrfs_alloc_path();
1167 path->leave_spinning = 1;
1169 block_rsv = trans->block_rsv;
1170 trans->block_rsv = &fs_info->delayed_block_rsv;
1172 delayed_root = fs_info->delayed_root;
1174 curr_node = btrfs_first_delayed_node(delayed_root);
1175 while (curr_node && (!count || (count && nr--))) {
1176 ret = __btrfs_commit_inode_delayed_items(trans, path,
1179 btrfs_abort_transaction(trans, ret);
1183 prev_node = curr_node;
1184 curr_node = btrfs_next_delayed_node(curr_node);
1186 * See the comment below about releasing path before releasing
1187 * node. If the commit of delayed items was successful the path
1188 * should always be released, but in case of an error, it may
1189 * point to locked extent buffers (a leaf at the very least).
1191 ASSERT(path->nodes[0] == NULL);
1192 btrfs_release_delayed_node(prev_node);
1196 * Release the path to avoid a potential deadlock and lockdep splat when
1197 * releasing the delayed node, as that requires taking the delayed node's
1198 * mutex. If another task starts running delayed items before we take
1199 * the mutex, it will first lock the mutex and then it may try to lock
1200 * the same btree path (leaf).
1202 btrfs_free_path(path);
1205 btrfs_release_delayed_node(curr_node);
1206 trans->block_rsv = block_rsv;
1211 int btrfs_run_delayed_items(struct btrfs_trans_handle *trans)
1213 return __btrfs_run_delayed_items(trans, -1);
1216 int btrfs_run_delayed_items_nr(struct btrfs_trans_handle *trans, int nr)
1218 return __btrfs_run_delayed_items(trans, nr);
1221 int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1222 struct btrfs_inode *inode)
1224 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1225 struct btrfs_path *path;
1226 struct btrfs_block_rsv *block_rsv;
1232 mutex_lock(&delayed_node->mutex);
1233 if (!delayed_node->count) {
1234 mutex_unlock(&delayed_node->mutex);
1235 btrfs_release_delayed_node(delayed_node);
1238 mutex_unlock(&delayed_node->mutex);
1240 path = btrfs_alloc_path();
1242 btrfs_release_delayed_node(delayed_node);
1245 path->leave_spinning = 1;
1247 block_rsv = trans->block_rsv;
1248 trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv;
1250 ret = __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1252 btrfs_release_delayed_node(delayed_node);
1253 btrfs_free_path(path);
1254 trans->block_rsv = block_rsv;
1259 int btrfs_commit_inode_delayed_inode(struct btrfs_inode *inode)
1261 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1262 struct btrfs_trans_handle *trans;
1263 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1264 struct btrfs_path *path;
1265 struct btrfs_block_rsv *block_rsv;
1271 mutex_lock(&delayed_node->mutex);
1272 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1273 mutex_unlock(&delayed_node->mutex);
1274 btrfs_release_delayed_node(delayed_node);
1277 mutex_unlock(&delayed_node->mutex);
1279 trans = btrfs_join_transaction(delayed_node->root);
1280 if (IS_ERR(trans)) {
1281 ret = PTR_ERR(trans);
1285 path = btrfs_alloc_path();
1290 path->leave_spinning = 1;
1292 block_rsv = trans->block_rsv;
1293 trans->block_rsv = &fs_info->delayed_block_rsv;
1295 mutex_lock(&delayed_node->mutex);
1296 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags))
1297 ret = __btrfs_update_delayed_inode(trans, delayed_node->root,
1298 path, delayed_node);
1301 mutex_unlock(&delayed_node->mutex);
1303 btrfs_free_path(path);
1304 trans->block_rsv = block_rsv;
1306 btrfs_end_transaction(trans);
1307 btrfs_btree_balance_dirty(fs_info);
1309 btrfs_release_delayed_node(delayed_node);
1314 void btrfs_remove_delayed_node(struct btrfs_inode *inode)
1316 struct btrfs_delayed_node *delayed_node;
1318 delayed_node = READ_ONCE(inode->delayed_node);
1322 inode->delayed_node = NULL;
1323 btrfs_release_delayed_node(delayed_node);
1326 struct btrfs_async_delayed_work {
1327 struct btrfs_delayed_root *delayed_root;
1329 struct btrfs_work work;
1332 static void btrfs_async_run_delayed_root(struct btrfs_work *work)
1334 struct btrfs_async_delayed_work *async_work;
1335 struct btrfs_delayed_root *delayed_root;
1336 struct btrfs_trans_handle *trans;
1337 struct btrfs_path *path;
1338 struct btrfs_delayed_node *delayed_node = NULL;
1339 struct btrfs_root *root;
1340 struct btrfs_block_rsv *block_rsv;
1343 async_work = container_of(work, struct btrfs_async_delayed_work, work);
1344 delayed_root = async_work->delayed_root;
1346 path = btrfs_alloc_path();
1351 if (atomic_read(&delayed_root->items) <
1352 BTRFS_DELAYED_BACKGROUND / 2)
1355 delayed_node = btrfs_first_prepared_delayed_node(delayed_root);
1359 path->leave_spinning = 1;
1360 root = delayed_node->root;
1362 trans = btrfs_join_transaction(root);
1363 if (IS_ERR(trans)) {
1364 btrfs_release_path(path);
1365 btrfs_release_prepared_delayed_node(delayed_node);
1370 block_rsv = trans->block_rsv;
1371 trans->block_rsv = &root->fs_info->delayed_block_rsv;
1373 __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1375 trans->block_rsv = block_rsv;
1376 btrfs_end_transaction(trans);
1377 btrfs_btree_balance_dirty_nodelay(root->fs_info);
1379 btrfs_release_path(path);
1380 btrfs_release_prepared_delayed_node(delayed_node);
1383 } while ((async_work->nr == 0 && total_done < BTRFS_DELAYED_WRITEBACK)
1384 || total_done < async_work->nr);
1386 btrfs_free_path(path);
1388 wake_up(&delayed_root->wait);
1393 static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root,
1394 struct btrfs_fs_info *fs_info, int nr)
1396 struct btrfs_async_delayed_work *async_work;
1398 async_work = kmalloc(sizeof(*async_work), GFP_NOFS);
1402 async_work->delayed_root = delayed_root;
1403 btrfs_init_work(&async_work->work, btrfs_async_run_delayed_root, NULL,
1405 async_work->nr = nr;
1407 btrfs_queue_work(fs_info->delayed_workers, &async_work->work);
1411 void btrfs_assert_delayed_root_empty(struct btrfs_fs_info *fs_info)
1413 WARN_ON(btrfs_first_delayed_node(fs_info->delayed_root));
1416 static int could_end_wait(struct btrfs_delayed_root *delayed_root, int seq)
1418 int val = atomic_read(&delayed_root->items_seq);
1420 if (val < seq || val >= seq + BTRFS_DELAYED_BATCH)
1423 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1429 void btrfs_balance_delayed_items(struct btrfs_fs_info *fs_info)
1431 struct btrfs_delayed_root *delayed_root = fs_info->delayed_root;
1433 if ((atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND) ||
1434 btrfs_workqueue_normal_congested(fs_info->delayed_workers))
1437 if (atomic_read(&delayed_root->items) >= BTRFS_DELAYED_WRITEBACK) {
1441 seq = atomic_read(&delayed_root->items_seq);
1443 ret = btrfs_wq_run_delayed_node(delayed_root, fs_info, 0);
1447 wait_event_interruptible(delayed_root->wait,
1448 could_end_wait(delayed_root, seq));
1452 btrfs_wq_run_delayed_node(delayed_root, fs_info, BTRFS_DELAYED_BATCH);
1455 /* Will return 0 or -ENOMEM */
1456 int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans,
1457 const char *name, int name_len,
1458 struct btrfs_inode *dir,
1459 struct btrfs_disk_key *disk_key, u8 type,
1462 struct btrfs_delayed_node *delayed_node;
1463 struct btrfs_delayed_item *delayed_item;
1464 struct btrfs_dir_item *dir_item;
1467 delayed_node = btrfs_get_or_create_delayed_node(dir);
1468 if (IS_ERR(delayed_node))
1469 return PTR_ERR(delayed_node);
1471 delayed_item = btrfs_alloc_delayed_item(sizeof(*dir_item) + name_len);
1472 if (!delayed_item) {
1477 delayed_item->key.objectid = btrfs_ino(dir);
1478 delayed_item->key.type = BTRFS_DIR_INDEX_KEY;
1479 delayed_item->key.offset = index;
1481 dir_item = (struct btrfs_dir_item *)delayed_item->data;
1482 dir_item->location = *disk_key;
1483 btrfs_set_stack_dir_transid(dir_item, trans->transid);
1484 btrfs_set_stack_dir_data_len(dir_item, 0);
1485 btrfs_set_stack_dir_name_len(dir_item, name_len);
1486 btrfs_set_stack_dir_type(dir_item, type);
1487 memcpy((char *)(dir_item + 1), name, name_len);
1489 ret = btrfs_delayed_item_reserve_metadata(trans, dir->root, delayed_item);
1491 * we have reserved enough space when we start a new transaction,
1492 * so reserving metadata failure is impossible
1496 mutex_lock(&delayed_node->mutex);
1497 ret = __btrfs_add_delayed_insertion_item(delayed_node, delayed_item);
1498 if (unlikely(ret)) {
1499 btrfs_err(trans->fs_info,
1500 "err add delayed dir index item(name: %.*s) into the insertion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
1501 name_len, name, delayed_node->root->root_key.objectid,
1502 delayed_node->inode_id, ret);
1505 mutex_unlock(&delayed_node->mutex);
1508 btrfs_release_delayed_node(delayed_node);
1512 static int btrfs_delete_delayed_insertion_item(struct btrfs_fs_info *fs_info,
1513 struct btrfs_delayed_node *node,
1514 struct btrfs_key *key)
1516 struct btrfs_delayed_item *item;
1518 mutex_lock(&node->mutex);
1519 item = __btrfs_lookup_delayed_insertion_item(node, key);
1521 mutex_unlock(&node->mutex);
1525 btrfs_delayed_item_release_metadata(node->root, item);
1526 btrfs_release_delayed_item(item);
1527 mutex_unlock(&node->mutex);
1531 int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans,
1532 struct btrfs_inode *dir, u64 index)
1534 struct btrfs_delayed_node *node;
1535 struct btrfs_delayed_item *item;
1536 struct btrfs_key item_key;
1539 node = btrfs_get_or_create_delayed_node(dir);
1541 return PTR_ERR(node);
1543 item_key.objectid = btrfs_ino(dir);
1544 item_key.type = BTRFS_DIR_INDEX_KEY;
1545 item_key.offset = index;
1547 ret = btrfs_delete_delayed_insertion_item(trans->fs_info, node,
1552 item = btrfs_alloc_delayed_item(0);
1558 item->key = item_key;
1560 ret = btrfs_delayed_item_reserve_metadata(trans, dir->root, item);
1562 * we have reserved enough space when we start a new transaction,
1563 * so reserving metadata failure is impossible.
1566 btrfs_err(trans->fs_info,
1567 "metadata reservation failed for delayed dir item deltiona, should have been reserved");
1568 btrfs_release_delayed_item(item);
1572 mutex_lock(&node->mutex);
1573 ret = __btrfs_add_delayed_deletion_item(node, item);
1574 if (unlikely(ret)) {
1575 btrfs_err(trans->fs_info,
1576 "err add delayed dir index item(index: %llu) into the deletion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
1577 index, node->root->root_key.objectid,
1578 node->inode_id, ret);
1579 btrfs_delayed_item_release_metadata(dir->root, item);
1580 btrfs_release_delayed_item(item);
1582 mutex_unlock(&node->mutex);
1584 btrfs_release_delayed_node(node);
1588 int btrfs_inode_delayed_dir_index_count(struct btrfs_inode *inode)
1590 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1596 * Since we have held i_mutex of this directory, it is impossible that
1597 * a new directory index is added into the delayed node and index_cnt
1598 * is updated now. So we needn't lock the delayed node.
1600 if (!delayed_node->index_cnt) {
1601 btrfs_release_delayed_node(delayed_node);
1605 inode->index_cnt = delayed_node->index_cnt;
1606 btrfs_release_delayed_node(delayed_node);
1610 bool btrfs_readdir_get_delayed_items(struct inode *inode,
1611 struct list_head *ins_list,
1612 struct list_head *del_list)
1614 struct btrfs_delayed_node *delayed_node;
1615 struct btrfs_delayed_item *item;
1617 delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
1622 * We can only do one readdir with delayed items at a time because of
1623 * item->readdir_list.
1625 inode_unlock_shared(inode);
1628 mutex_lock(&delayed_node->mutex);
1629 item = __btrfs_first_delayed_insertion_item(delayed_node);
1631 refcount_inc(&item->refs);
1632 list_add_tail(&item->readdir_list, ins_list);
1633 item = __btrfs_next_delayed_item(item);
1636 item = __btrfs_first_delayed_deletion_item(delayed_node);
1638 refcount_inc(&item->refs);
1639 list_add_tail(&item->readdir_list, del_list);
1640 item = __btrfs_next_delayed_item(item);
1642 mutex_unlock(&delayed_node->mutex);
1644 * This delayed node is still cached in the btrfs inode, so refs
1645 * must be > 1 now, and we needn't check it is going to be freed
1648 * Besides that, this function is used to read dir, we do not
1649 * insert/delete delayed items in this period. So we also needn't
1650 * requeue or dequeue this delayed node.
1652 refcount_dec(&delayed_node->refs);
1657 void btrfs_readdir_put_delayed_items(struct inode *inode,
1658 struct list_head *ins_list,
1659 struct list_head *del_list)
1661 struct btrfs_delayed_item *curr, *next;
1663 list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1664 list_del(&curr->readdir_list);
1665 if (refcount_dec_and_test(&curr->refs))
1669 list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1670 list_del(&curr->readdir_list);
1671 if (refcount_dec_and_test(&curr->refs))
1676 * The VFS is going to do up_read(), so we need to downgrade back to a
1679 downgrade_write(&inode->i_rwsem);
1682 int btrfs_should_delete_dir_index(struct list_head *del_list,
1685 struct btrfs_delayed_item *curr;
1688 list_for_each_entry(curr, del_list, readdir_list) {
1689 if (curr->key.offset > index)
1691 if (curr->key.offset == index) {
1700 * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree
1703 int btrfs_readdir_delayed_dir_index(struct dir_context *ctx,
1704 struct list_head *ins_list)
1706 struct btrfs_dir_item *di;
1707 struct btrfs_delayed_item *curr, *next;
1708 struct btrfs_key location;
1712 unsigned char d_type;
1714 if (list_empty(ins_list))
1718 * Changing the data of the delayed item is impossible. So
1719 * we needn't lock them. And we have held i_mutex of the
1720 * directory, nobody can delete any directory indexes now.
1722 list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1723 list_del(&curr->readdir_list);
1725 if (curr->key.offset < ctx->pos) {
1726 if (refcount_dec_and_test(&curr->refs))
1731 ctx->pos = curr->key.offset;
1733 di = (struct btrfs_dir_item *)curr->data;
1734 name = (char *)(di + 1);
1735 name_len = btrfs_stack_dir_name_len(di);
1737 d_type = fs_ftype_to_dtype(di->type);
1738 btrfs_disk_key_to_cpu(&location, &di->location);
1740 over = !dir_emit(ctx, name, name_len,
1741 location.objectid, d_type);
1743 if (refcount_dec_and_test(&curr->refs))
1753 static void fill_stack_inode_item(struct btrfs_trans_handle *trans,
1754 struct btrfs_inode_item *inode_item,
1755 struct inode *inode)
1757 btrfs_set_stack_inode_uid(inode_item, i_uid_read(inode));
1758 btrfs_set_stack_inode_gid(inode_item, i_gid_read(inode));
1759 btrfs_set_stack_inode_size(inode_item, BTRFS_I(inode)->disk_i_size);
1760 btrfs_set_stack_inode_mode(inode_item, inode->i_mode);
1761 btrfs_set_stack_inode_nlink(inode_item, inode->i_nlink);
1762 btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(inode));
1763 btrfs_set_stack_inode_generation(inode_item,
1764 BTRFS_I(inode)->generation);
1765 btrfs_set_stack_inode_sequence(inode_item,
1766 inode_peek_iversion(inode));
1767 btrfs_set_stack_inode_transid(inode_item, trans->transid);
1768 btrfs_set_stack_inode_rdev(inode_item, inode->i_rdev);
1769 btrfs_set_stack_inode_flags(inode_item, BTRFS_I(inode)->flags);
1770 btrfs_set_stack_inode_block_group(inode_item, 0);
1772 btrfs_set_stack_timespec_sec(&inode_item->atime,
1773 inode->i_atime.tv_sec);
1774 btrfs_set_stack_timespec_nsec(&inode_item->atime,
1775 inode->i_atime.tv_nsec);
1777 btrfs_set_stack_timespec_sec(&inode_item->mtime,
1778 inode->i_mtime.tv_sec);
1779 btrfs_set_stack_timespec_nsec(&inode_item->mtime,
1780 inode->i_mtime.tv_nsec);
1782 btrfs_set_stack_timespec_sec(&inode_item->ctime,
1783 inode->i_ctime.tv_sec);
1784 btrfs_set_stack_timespec_nsec(&inode_item->ctime,
1785 inode->i_ctime.tv_nsec);
1787 btrfs_set_stack_timespec_sec(&inode_item->otime,
1788 BTRFS_I(inode)->i_otime.tv_sec);
1789 btrfs_set_stack_timespec_nsec(&inode_item->otime,
1790 BTRFS_I(inode)->i_otime.tv_nsec);
1793 int btrfs_fill_inode(struct inode *inode, u32 *rdev)
1795 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1796 struct btrfs_delayed_node *delayed_node;
1797 struct btrfs_inode_item *inode_item;
1799 delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
1803 mutex_lock(&delayed_node->mutex);
1804 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1805 mutex_unlock(&delayed_node->mutex);
1806 btrfs_release_delayed_node(delayed_node);
1810 inode_item = &delayed_node->inode_item;
1812 i_uid_write(inode, btrfs_stack_inode_uid(inode_item));
1813 i_gid_write(inode, btrfs_stack_inode_gid(inode_item));
1814 btrfs_i_size_write(BTRFS_I(inode), btrfs_stack_inode_size(inode_item));
1815 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
1816 round_up(i_size_read(inode), fs_info->sectorsize));
1817 inode->i_mode = btrfs_stack_inode_mode(inode_item);
1818 set_nlink(inode, btrfs_stack_inode_nlink(inode_item));
1819 inode_set_bytes(inode, btrfs_stack_inode_nbytes(inode_item));
1820 BTRFS_I(inode)->generation = btrfs_stack_inode_generation(inode_item);
1821 BTRFS_I(inode)->last_trans = btrfs_stack_inode_transid(inode_item);
1823 inode_set_iversion_queried(inode,
1824 btrfs_stack_inode_sequence(inode_item));
1826 *rdev = btrfs_stack_inode_rdev(inode_item);
1827 BTRFS_I(inode)->flags = btrfs_stack_inode_flags(inode_item);
1829 inode->i_atime.tv_sec = btrfs_stack_timespec_sec(&inode_item->atime);
1830 inode->i_atime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->atime);
1832 inode->i_mtime.tv_sec = btrfs_stack_timespec_sec(&inode_item->mtime);
1833 inode->i_mtime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->mtime);
1835 inode->i_ctime.tv_sec = btrfs_stack_timespec_sec(&inode_item->ctime);
1836 inode->i_ctime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->ctime);
1838 BTRFS_I(inode)->i_otime.tv_sec =
1839 btrfs_stack_timespec_sec(&inode_item->otime);
1840 BTRFS_I(inode)->i_otime.tv_nsec =
1841 btrfs_stack_timespec_nsec(&inode_item->otime);
1843 inode->i_generation = BTRFS_I(inode)->generation;
1844 BTRFS_I(inode)->index_cnt = (u64)-1;
1846 mutex_unlock(&delayed_node->mutex);
1847 btrfs_release_delayed_node(delayed_node);
1851 int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans,
1852 struct btrfs_root *root, struct inode *inode)
1854 struct btrfs_delayed_node *delayed_node;
1857 delayed_node = btrfs_get_or_create_delayed_node(BTRFS_I(inode));
1858 if (IS_ERR(delayed_node))
1859 return PTR_ERR(delayed_node);
1861 mutex_lock(&delayed_node->mutex);
1862 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1863 fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1867 ret = btrfs_delayed_inode_reserve_metadata(trans, root, BTRFS_I(inode),
1872 fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1873 set_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
1874 delayed_node->count++;
1875 atomic_inc(&root->fs_info->delayed_root->items);
1877 mutex_unlock(&delayed_node->mutex);
1878 btrfs_release_delayed_node(delayed_node);
1882 int btrfs_delayed_delete_inode_ref(struct btrfs_inode *inode)
1884 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1885 struct btrfs_delayed_node *delayed_node;
1888 * we don't do delayed inode updates during log recovery because it
1889 * leads to enospc problems. This means we also can't do
1890 * delayed inode refs
1892 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
1895 delayed_node = btrfs_get_or_create_delayed_node(inode);
1896 if (IS_ERR(delayed_node))
1897 return PTR_ERR(delayed_node);
1900 * We don't reserve space for inode ref deletion is because:
1901 * - We ONLY do async inode ref deletion for the inode who has only
1902 * one link(i_nlink == 1), it means there is only one inode ref.
1903 * And in most case, the inode ref and the inode item are in the
1904 * same leaf, and we will deal with them at the same time.
1905 * Since we are sure we will reserve the space for the inode item,
1906 * it is unnecessary to reserve space for inode ref deletion.
1907 * - If the inode ref and the inode item are not in the same leaf,
1908 * We also needn't worry about enospc problem, because we reserve
1909 * much more space for the inode update than it needs.
1910 * - At the worst, we can steal some space from the global reservation.
1913 mutex_lock(&delayed_node->mutex);
1914 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1917 set_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
1918 delayed_node->count++;
1919 atomic_inc(&fs_info->delayed_root->items);
1921 mutex_unlock(&delayed_node->mutex);
1922 btrfs_release_delayed_node(delayed_node);
1926 static void __btrfs_kill_delayed_node(struct btrfs_delayed_node *delayed_node)
1928 struct btrfs_root *root = delayed_node->root;
1929 struct btrfs_fs_info *fs_info = root->fs_info;
1930 struct btrfs_delayed_item *curr_item, *prev_item;
1932 mutex_lock(&delayed_node->mutex);
1933 curr_item = __btrfs_first_delayed_insertion_item(delayed_node);
1935 btrfs_delayed_item_release_metadata(root, curr_item);
1936 prev_item = curr_item;
1937 curr_item = __btrfs_next_delayed_item(prev_item);
1938 btrfs_release_delayed_item(prev_item);
1941 curr_item = __btrfs_first_delayed_deletion_item(delayed_node);
1943 btrfs_delayed_item_release_metadata(root, curr_item);
1944 prev_item = curr_item;
1945 curr_item = __btrfs_next_delayed_item(prev_item);
1946 btrfs_release_delayed_item(prev_item);
1949 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1950 btrfs_release_delayed_iref(delayed_node);
1952 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1953 btrfs_delayed_inode_release_metadata(fs_info, delayed_node, false);
1954 btrfs_release_delayed_inode(delayed_node);
1956 mutex_unlock(&delayed_node->mutex);
1959 void btrfs_kill_delayed_inode_items(struct btrfs_inode *inode)
1961 struct btrfs_delayed_node *delayed_node;
1963 delayed_node = btrfs_get_delayed_node(inode);
1967 __btrfs_kill_delayed_node(delayed_node);
1968 btrfs_release_delayed_node(delayed_node);
1971 void btrfs_kill_all_delayed_nodes(struct btrfs_root *root)
1974 struct btrfs_delayed_node *delayed_nodes[8];
1978 spin_lock(&root->inode_lock);
1979 n = radix_tree_gang_lookup(&root->delayed_nodes_tree,
1980 (void **)delayed_nodes, inode_id,
1981 ARRAY_SIZE(delayed_nodes));
1983 spin_unlock(&root->inode_lock);
1987 inode_id = delayed_nodes[n - 1]->inode_id + 1;
1988 for (i = 0; i < n; i++) {
1990 * Don't increase refs in case the node is dead and
1991 * about to be removed from the tree in the loop below
1993 if (!refcount_inc_not_zero(&delayed_nodes[i]->refs))
1994 delayed_nodes[i] = NULL;
1996 spin_unlock(&root->inode_lock);
1998 for (i = 0; i < n; i++) {
1999 if (!delayed_nodes[i])
2001 __btrfs_kill_delayed_node(delayed_nodes[i]);
2002 btrfs_release_delayed_node(delayed_nodes[i]);
2007 void btrfs_destroy_delayed_inodes(struct btrfs_fs_info *fs_info)
2009 struct btrfs_delayed_node *curr_node, *prev_node;
2011 curr_node = btrfs_first_delayed_node(fs_info->delayed_root);
2013 __btrfs_kill_delayed_node(curr_node);
2015 prev_node = curr_node;
2016 curr_node = btrfs_next_delayed_node(curr_node);
2017 btrfs_release_delayed_node(prev_node);