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>
10 #include "delayed-inode.h"
12 #include "transaction.h"
17 #define BTRFS_DELAYED_WRITEBACK 512
18 #define BTRFS_DELAYED_BACKGROUND 128
19 #define BTRFS_DELAYED_BATCH 16
21 static struct kmem_cache
*delayed_node_cache
;
23 int __init
btrfs_delayed_inode_init(void)
25 delayed_node_cache
= kmem_cache_create("btrfs_delayed_node",
26 sizeof(struct btrfs_delayed_node
),
30 if (!delayed_node_cache
)
35 void __cold
btrfs_delayed_inode_exit(void)
37 kmem_cache_destroy(delayed_node_cache
);
40 static inline void btrfs_init_delayed_node(
41 struct btrfs_delayed_node
*delayed_node
,
42 struct btrfs_root
*root
, u64 inode_id
)
44 delayed_node
->root
= root
;
45 delayed_node
->inode_id
= inode_id
;
46 refcount_set(&delayed_node
->refs
, 0);
47 delayed_node
->ins_root
= RB_ROOT_CACHED
;
48 delayed_node
->del_root
= RB_ROOT_CACHED
;
49 mutex_init(&delayed_node
->mutex
);
50 INIT_LIST_HEAD(&delayed_node
->n_list
);
51 INIT_LIST_HEAD(&delayed_node
->p_list
);
54 static inline int btrfs_is_continuous_delayed_item(
55 struct btrfs_delayed_item
*item1
,
56 struct btrfs_delayed_item
*item2
)
58 if (item1
->key
.type
== BTRFS_DIR_INDEX_KEY
&&
59 item1
->key
.objectid
== item2
->key
.objectid
&&
60 item1
->key
.type
== item2
->key
.type
&&
61 item1
->key
.offset
+ 1 == item2
->key
.offset
)
66 static struct btrfs_delayed_node
*btrfs_get_delayed_node(
67 struct btrfs_inode
*btrfs_inode
)
69 struct btrfs_root
*root
= btrfs_inode
->root
;
70 u64 ino
= btrfs_ino(btrfs_inode
);
71 struct btrfs_delayed_node
*node
;
73 node
= READ_ONCE(btrfs_inode
->delayed_node
);
75 refcount_inc(&node
->refs
);
79 spin_lock(&root
->inode_lock
);
80 node
= radix_tree_lookup(&root
->delayed_nodes_tree
, ino
);
83 if (btrfs_inode
->delayed_node
) {
84 refcount_inc(&node
->refs
); /* can be accessed */
85 BUG_ON(btrfs_inode
->delayed_node
!= node
);
86 spin_unlock(&root
->inode_lock
);
91 * It's possible that we're racing into the middle of removing
92 * this node from the radix tree. In this case, the refcount
93 * was zero and it should never go back to one. Just return
94 * NULL like it was never in the radix at all; our release
95 * function is in the process of removing it.
97 * Some implementations of refcount_inc refuse to bump the
98 * refcount once it has hit zero. If we don't do this dance
99 * here, refcount_inc() may decide to just WARN_ONCE() instead
100 * of actually bumping the refcount.
102 * If this node is properly in the radix, we want to bump the
103 * refcount twice, once for the inode and once for this get
106 if (refcount_inc_not_zero(&node
->refs
)) {
107 refcount_inc(&node
->refs
);
108 btrfs_inode
->delayed_node
= node
;
113 spin_unlock(&root
->inode_lock
);
116 spin_unlock(&root
->inode_lock
);
121 /* Will return either the node or PTR_ERR(-ENOMEM) */
122 static struct btrfs_delayed_node
*btrfs_get_or_create_delayed_node(
123 struct btrfs_inode
*btrfs_inode
)
125 struct btrfs_delayed_node
*node
;
126 struct btrfs_root
*root
= btrfs_inode
->root
;
127 u64 ino
= btrfs_ino(btrfs_inode
);
131 node
= btrfs_get_delayed_node(btrfs_inode
);
135 node
= kmem_cache_zalloc(delayed_node_cache
, GFP_NOFS
);
137 return ERR_PTR(-ENOMEM
);
138 btrfs_init_delayed_node(node
, root
, ino
);
140 /* cached in the btrfs inode and can be accessed */
141 refcount_set(&node
->refs
, 2);
143 ret
= radix_tree_preload(GFP_NOFS
);
145 kmem_cache_free(delayed_node_cache
, node
);
149 spin_lock(&root
->inode_lock
);
150 ret
= radix_tree_insert(&root
->delayed_nodes_tree
, ino
, node
);
151 if (ret
== -EEXIST
) {
152 spin_unlock(&root
->inode_lock
);
153 kmem_cache_free(delayed_node_cache
, node
);
154 radix_tree_preload_end();
157 btrfs_inode
->delayed_node
= node
;
158 spin_unlock(&root
->inode_lock
);
159 radix_tree_preload_end();
165 * Call it when holding delayed_node->mutex
167 * If mod = 1, add this node into the prepared list.
169 static void btrfs_queue_delayed_node(struct btrfs_delayed_root
*root
,
170 struct btrfs_delayed_node
*node
,
173 spin_lock(&root
->lock
);
174 if (test_bit(BTRFS_DELAYED_NODE_IN_LIST
, &node
->flags
)) {
175 if (!list_empty(&node
->p_list
))
176 list_move_tail(&node
->p_list
, &root
->prepare_list
);
178 list_add_tail(&node
->p_list
, &root
->prepare_list
);
180 list_add_tail(&node
->n_list
, &root
->node_list
);
181 list_add_tail(&node
->p_list
, &root
->prepare_list
);
182 refcount_inc(&node
->refs
); /* inserted into list */
184 set_bit(BTRFS_DELAYED_NODE_IN_LIST
, &node
->flags
);
186 spin_unlock(&root
->lock
);
189 /* Call it when holding delayed_node->mutex */
190 static void btrfs_dequeue_delayed_node(struct btrfs_delayed_root
*root
,
191 struct btrfs_delayed_node
*node
)
193 spin_lock(&root
->lock
);
194 if (test_bit(BTRFS_DELAYED_NODE_IN_LIST
, &node
->flags
)) {
196 refcount_dec(&node
->refs
); /* not in the list */
197 list_del_init(&node
->n_list
);
198 if (!list_empty(&node
->p_list
))
199 list_del_init(&node
->p_list
);
200 clear_bit(BTRFS_DELAYED_NODE_IN_LIST
, &node
->flags
);
202 spin_unlock(&root
->lock
);
205 static struct btrfs_delayed_node
*btrfs_first_delayed_node(
206 struct btrfs_delayed_root
*delayed_root
)
209 struct btrfs_delayed_node
*node
= NULL
;
211 spin_lock(&delayed_root
->lock
);
212 if (list_empty(&delayed_root
->node_list
))
215 p
= delayed_root
->node_list
.next
;
216 node
= list_entry(p
, struct btrfs_delayed_node
, n_list
);
217 refcount_inc(&node
->refs
);
219 spin_unlock(&delayed_root
->lock
);
224 static struct btrfs_delayed_node
*btrfs_next_delayed_node(
225 struct btrfs_delayed_node
*node
)
227 struct btrfs_delayed_root
*delayed_root
;
229 struct btrfs_delayed_node
*next
= NULL
;
231 delayed_root
= node
->root
->fs_info
->delayed_root
;
232 spin_lock(&delayed_root
->lock
);
233 if (!test_bit(BTRFS_DELAYED_NODE_IN_LIST
, &node
->flags
)) {
234 /* not in the list */
235 if (list_empty(&delayed_root
->node_list
))
237 p
= delayed_root
->node_list
.next
;
238 } else if (list_is_last(&node
->n_list
, &delayed_root
->node_list
))
241 p
= node
->n_list
.next
;
243 next
= list_entry(p
, struct btrfs_delayed_node
, n_list
);
244 refcount_inc(&next
->refs
);
246 spin_unlock(&delayed_root
->lock
);
251 static void __btrfs_release_delayed_node(
252 struct btrfs_delayed_node
*delayed_node
,
255 struct btrfs_delayed_root
*delayed_root
;
260 delayed_root
= delayed_node
->root
->fs_info
->delayed_root
;
262 mutex_lock(&delayed_node
->mutex
);
263 if (delayed_node
->count
)
264 btrfs_queue_delayed_node(delayed_root
, delayed_node
, mod
);
266 btrfs_dequeue_delayed_node(delayed_root
, delayed_node
);
267 mutex_unlock(&delayed_node
->mutex
);
269 if (refcount_dec_and_test(&delayed_node
->refs
)) {
270 struct btrfs_root
*root
= delayed_node
->root
;
272 spin_lock(&root
->inode_lock
);
274 * Once our refcount goes to zero, nobody is allowed to bump it
275 * back up. We can delete it now.
277 ASSERT(refcount_read(&delayed_node
->refs
) == 0);
278 radix_tree_delete(&root
->delayed_nodes_tree
,
279 delayed_node
->inode_id
);
280 spin_unlock(&root
->inode_lock
);
281 kmem_cache_free(delayed_node_cache
, delayed_node
);
285 static inline void btrfs_release_delayed_node(struct btrfs_delayed_node
*node
)
287 __btrfs_release_delayed_node(node
, 0);
290 static struct btrfs_delayed_node
*btrfs_first_prepared_delayed_node(
291 struct btrfs_delayed_root
*delayed_root
)
294 struct btrfs_delayed_node
*node
= NULL
;
296 spin_lock(&delayed_root
->lock
);
297 if (list_empty(&delayed_root
->prepare_list
))
300 p
= delayed_root
->prepare_list
.next
;
302 node
= list_entry(p
, struct btrfs_delayed_node
, p_list
);
303 refcount_inc(&node
->refs
);
305 spin_unlock(&delayed_root
->lock
);
310 static inline void btrfs_release_prepared_delayed_node(
311 struct btrfs_delayed_node
*node
)
313 __btrfs_release_delayed_node(node
, 1);
316 static struct btrfs_delayed_item
*btrfs_alloc_delayed_item(u32 data_len
)
318 struct btrfs_delayed_item
*item
;
319 item
= kmalloc(sizeof(*item
) + data_len
, GFP_NOFS
);
321 item
->data_len
= data_len
;
322 item
->ins_or_del
= 0;
323 item
->bytes_reserved
= 0;
324 item
->delayed_node
= NULL
;
325 refcount_set(&item
->refs
, 1);
331 * __btrfs_lookup_delayed_item - look up the delayed item by key
332 * @delayed_node: pointer to the delayed node
333 * @key: the key to look up
334 * @prev: used to store the prev item if the right item isn't found
335 * @next: used to store the next item if the right item isn't found
337 * Note: if we don't find the right item, we will return the prev item and
340 static struct btrfs_delayed_item
*__btrfs_lookup_delayed_item(
341 struct rb_root
*root
,
342 struct btrfs_key
*key
,
343 struct btrfs_delayed_item
**prev
,
344 struct btrfs_delayed_item
**next
)
346 struct rb_node
*node
, *prev_node
= NULL
;
347 struct btrfs_delayed_item
*delayed_item
= NULL
;
350 node
= root
->rb_node
;
353 delayed_item
= rb_entry(node
, struct btrfs_delayed_item
,
356 ret
= btrfs_comp_cpu_keys(&delayed_item
->key
, key
);
358 node
= node
->rb_right
;
360 node
= node
->rb_left
;
369 *prev
= delayed_item
;
370 else if ((node
= rb_prev(prev_node
)) != NULL
) {
371 *prev
= rb_entry(node
, struct btrfs_delayed_item
,
381 *next
= delayed_item
;
382 else if ((node
= rb_next(prev_node
)) != NULL
) {
383 *next
= rb_entry(node
, struct btrfs_delayed_item
,
391 static struct btrfs_delayed_item
*__btrfs_lookup_delayed_insertion_item(
392 struct btrfs_delayed_node
*delayed_node
,
393 struct btrfs_key
*key
)
395 return __btrfs_lookup_delayed_item(&delayed_node
->ins_root
.rb_root
, key
,
399 static int __btrfs_add_delayed_item(struct btrfs_delayed_node
*delayed_node
,
400 struct btrfs_delayed_item
*ins
,
403 struct rb_node
**p
, *node
;
404 struct rb_node
*parent_node
= NULL
;
405 struct rb_root_cached
*root
;
406 struct btrfs_delayed_item
*item
;
408 bool leftmost
= true;
410 if (action
== BTRFS_DELAYED_INSERTION_ITEM
)
411 root
= &delayed_node
->ins_root
;
412 else if (action
== BTRFS_DELAYED_DELETION_ITEM
)
413 root
= &delayed_node
->del_root
;
416 p
= &root
->rb_root
.rb_node
;
417 node
= &ins
->rb_node
;
421 item
= rb_entry(parent_node
, struct btrfs_delayed_item
,
424 cmp
= btrfs_comp_cpu_keys(&item
->key
, &ins
->key
);
428 } else if (cmp
> 0) {
435 rb_link_node(node
, parent_node
, p
);
436 rb_insert_color_cached(node
, root
, leftmost
);
437 ins
->delayed_node
= delayed_node
;
438 ins
->ins_or_del
= action
;
440 if (ins
->key
.type
== BTRFS_DIR_INDEX_KEY
&&
441 action
== BTRFS_DELAYED_INSERTION_ITEM
&&
442 ins
->key
.offset
>= delayed_node
->index_cnt
)
443 delayed_node
->index_cnt
= ins
->key
.offset
+ 1;
445 delayed_node
->count
++;
446 atomic_inc(&delayed_node
->root
->fs_info
->delayed_root
->items
);
450 static int __btrfs_add_delayed_insertion_item(struct btrfs_delayed_node
*node
,
451 struct btrfs_delayed_item
*item
)
453 return __btrfs_add_delayed_item(node
, item
,
454 BTRFS_DELAYED_INSERTION_ITEM
);
457 static int __btrfs_add_delayed_deletion_item(struct btrfs_delayed_node
*node
,
458 struct btrfs_delayed_item
*item
)
460 return __btrfs_add_delayed_item(node
, item
,
461 BTRFS_DELAYED_DELETION_ITEM
);
464 static void finish_one_item(struct btrfs_delayed_root
*delayed_root
)
466 int seq
= atomic_inc_return(&delayed_root
->items_seq
);
468 /* atomic_dec_return implies a barrier */
469 if ((atomic_dec_return(&delayed_root
->items
) <
470 BTRFS_DELAYED_BACKGROUND
|| seq
% BTRFS_DELAYED_BATCH
== 0))
471 cond_wake_up_nomb(&delayed_root
->wait
);
474 static void __btrfs_remove_delayed_item(struct btrfs_delayed_item
*delayed_item
)
476 struct rb_root_cached
*root
;
477 struct btrfs_delayed_root
*delayed_root
;
479 /* Not associated with any delayed_node */
480 if (!delayed_item
->delayed_node
)
482 delayed_root
= delayed_item
->delayed_node
->root
->fs_info
->delayed_root
;
484 BUG_ON(!delayed_root
);
485 BUG_ON(delayed_item
->ins_or_del
!= BTRFS_DELAYED_DELETION_ITEM
&&
486 delayed_item
->ins_or_del
!= BTRFS_DELAYED_INSERTION_ITEM
);
488 if (delayed_item
->ins_or_del
== BTRFS_DELAYED_INSERTION_ITEM
)
489 root
= &delayed_item
->delayed_node
->ins_root
;
491 root
= &delayed_item
->delayed_node
->del_root
;
493 rb_erase_cached(&delayed_item
->rb_node
, root
);
494 delayed_item
->delayed_node
->count
--;
496 finish_one_item(delayed_root
);
499 static void btrfs_release_delayed_item(struct btrfs_delayed_item
*item
)
502 __btrfs_remove_delayed_item(item
);
503 if (refcount_dec_and_test(&item
->refs
))
508 static struct btrfs_delayed_item
*__btrfs_first_delayed_insertion_item(
509 struct btrfs_delayed_node
*delayed_node
)
512 struct btrfs_delayed_item
*item
= NULL
;
514 p
= rb_first_cached(&delayed_node
->ins_root
);
516 item
= rb_entry(p
, struct btrfs_delayed_item
, rb_node
);
521 static struct btrfs_delayed_item
*__btrfs_first_delayed_deletion_item(
522 struct btrfs_delayed_node
*delayed_node
)
525 struct btrfs_delayed_item
*item
= NULL
;
527 p
= rb_first_cached(&delayed_node
->del_root
);
529 item
= rb_entry(p
, struct btrfs_delayed_item
, rb_node
);
534 static struct btrfs_delayed_item
*__btrfs_next_delayed_item(
535 struct btrfs_delayed_item
*item
)
538 struct btrfs_delayed_item
*next
= NULL
;
540 p
= rb_next(&item
->rb_node
);
542 next
= rb_entry(p
, struct btrfs_delayed_item
, rb_node
);
547 static int btrfs_delayed_item_reserve_metadata(struct btrfs_trans_handle
*trans
,
548 struct btrfs_root
*root
,
549 struct btrfs_delayed_item
*item
)
551 struct btrfs_block_rsv
*src_rsv
;
552 struct btrfs_block_rsv
*dst_rsv
;
553 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
557 if (!trans
->bytes_reserved
)
560 src_rsv
= trans
->block_rsv
;
561 dst_rsv
= &fs_info
->delayed_block_rsv
;
563 num_bytes
= btrfs_calc_insert_metadata_size(fs_info
, 1);
566 * Here we migrate space rsv from transaction rsv, since have already
567 * reserved space when starting a transaction. So no need to reserve
570 ret
= btrfs_block_rsv_migrate(src_rsv
, dst_rsv
, num_bytes
, true);
572 trace_btrfs_space_reservation(fs_info
, "delayed_item",
575 item
->bytes_reserved
= num_bytes
;
581 static void btrfs_delayed_item_release_metadata(struct btrfs_root
*root
,
582 struct btrfs_delayed_item
*item
)
584 struct btrfs_block_rsv
*rsv
;
585 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
587 if (!item
->bytes_reserved
)
590 rsv
= &fs_info
->delayed_block_rsv
;
592 * Check btrfs_delayed_item_reserve_metadata() to see why we don't need
593 * to release/reserve qgroup space.
595 trace_btrfs_space_reservation(fs_info
, "delayed_item",
596 item
->key
.objectid
, item
->bytes_reserved
,
598 btrfs_block_rsv_release(fs_info
, rsv
,
599 item
->bytes_reserved
);
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_prealloc(root
,
631 fs_info
->nodesize
, 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
, fs_info
->nodesize
);
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
,
681 node
->bytes_reserved
);
683 btrfs_qgroup_free_meta_prealloc(node
->root
,
684 node
->bytes_reserved
);
686 btrfs_qgroup_convert_reserved_meta(node
->root
,
687 node
->bytes_reserved
);
688 node
->bytes_reserved
= 0;
692 * This helper will insert some continuous items into the same leaf according
693 * to the free space of the leaf.
695 static int btrfs_batch_insert_items(struct btrfs_root
*root
,
696 struct btrfs_path
*path
,
697 struct btrfs_delayed_item
*item
)
699 struct btrfs_delayed_item
*curr
, *next
;
701 int total_data_size
= 0, total_size
= 0;
702 struct extent_buffer
*leaf
;
704 struct btrfs_key
*keys
;
706 struct list_head head
;
712 BUG_ON(!path
->nodes
[0]);
714 leaf
= path
->nodes
[0];
715 free_space
= btrfs_leaf_free_space(leaf
);
716 INIT_LIST_HEAD(&head
);
722 * count the number of the continuous items that we can insert in batch
724 while (total_size
+ next
->data_len
+ sizeof(struct btrfs_item
) <=
726 total_data_size
+= next
->data_len
;
727 total_size
+= next
->data_len
+ sizeof(struct btrfs_item
);
728 list_add_tail(&next
->tree_list
, &head
);
732 next
= __btrfs_next_delayed_item(curr
);
736 if (!btrfs_is_continuous_delayed_item(curr
, next
))
746 * we need allocate some memory space, but it might cause the task
747 * to sleep, so we set all locked nodes in the path to blocking locks
750 btrfs_set_path_blocking(path
);
752 keys
= kmalloc_array(nitems
, sizeof(struct btrfs_key
), GFP_NOFS
);
758 data_size
= kmalloc_array(nitems
, sizeof(u32
), GFP_NOFS
);
764 /* get keys of all the delayed items */
766 list_for_each_entry(next
, &head
, tree_list
) {
768 data_size
[i
] = next
->data_len
;
772 /* insert the keys of the items */
773 setup_items_for_insert(root
, path
, keys
, data_size
,
774 total_data_size
, total_size
, nitems
);
776 /* insert the dir index items */
777 slot
= path
->slots
[0];
778 list_for_each_entry_safe(curr
, next
, &head
, tree_list
) {
779 data_ptr
= btrfs_item_ptr(leaf
, slot
, char);
780 write_extent_buffer(leaf
, &curr
->data
,
781 (unsigned long)data_ptr
,
785 btrfs_delayed_item_release_metadata(root
, curr
);
787 list_del(&curr
->tree_list
);
788 btrfs_release_delayed_item(curr
);
799 * This helper can just do simple insertion that needn't extend item for new
800 * data, such as directory name index insertion, inode insertion.
802 static int btrfs_insert_delayed_item(struct btrfs_trans_handle
*trans
,
803 struct btrfs_root
*root
,
804 struct btrfs_path
*path
,
805 struct btrfs_delayed_item
*delayed_item
)
807 struct extent_buffer
*leaf
;
811 ret
= btrfs_insert_empty_item(trans
, root
, path
, &delayed_item
->key
,
812 delayed_item
->data_len
);
813 if (ret
< 0 && ret
!= -EEXIST
)
816 leaf
= path
->nodes
[0];
818 ptr
= btrfs_item_ptr(leaf
, path
->slots
[0], char);
820 write_extent_buffer(leaf
, delayed_item
->data
, (unsigned long)ptr
,
821 delayed_item
->data_len
);
822 btrfs_mark_buffer_dirty(leaf
);
824 btrfs_delayed_item_release_metadata(root
, delayed_item
);
829 * we insert an item first, then if there are some continuous items, we try
830 * to insert those items into the same leaf.
832 static int btrfs_insert_delayed_items(struct btrfs_trans_handle
*trans
,
833 struct btrfs_path
*path
,
834 struct btrfs_root
*root
,
835 struct btrfs_delayed_node
*node
)
837 struct btrfs_delayed_item
*curr
, *prev
;
841 mutex_lock(&node
->mutex
);
842 curr
= __btrfs_first_delayed_insertion_item(node
);
846 ret
= btrfs_insert_delayed_item(trans
, root
, path
, curr
);
848 btrfs_release_path(path
);
853 curr
= __btrfs_next_delayed_item(prev
);
854 if (curr
&& btrfs_is_continuous_delayed_item(prev
, curr
)) {
855 /* insert the continuous items into the same leaf */
857 btrfs_batch_insert_items(root
, path
, curr
);
859 btrfs_release_delayed_item(prev
);
860 btrfs_mark_buffer_dirty(path
->nodes
[0]);
862 btrfs_release_path(path
);
863 mutex_unlock(&node
->mutex
);
867 mutex_unlock(&node
->mutex
);
871 static int btrfs_batch_delete_items(struct btrfs_trans_handle
*trans
,
872 struct btrfs_root
*root
,
873 struct btrfs_path
*path
,
874 struct btrfs_delayed_item
*item
)
876 struct btrfs_delayed_item
*curr
, *next
;
877 struct extent_buffer
*leaf
;
878 struct btrfs_key key
;
879 struct list_head head
;
880 int nitems
, i
, last_item
;
883 BUG_ON(!path
->nodes
[0]);
885 leaf
= path
->nodes
[0];
888 last_item
= btrfs_header_nritems(leaf
) - 1;
890 return -ENOENT
; /* FIXME: Is errno suitable? */
893 INIT_LIST_HEAD(&head
);
894 btrfs_item_key_to_cpu(leaf
, &key
, i
);
897 * count the number of the dir index items that we can delete in batch
899 while (btrfs_comp_cpu_keys(&next
->key
, &key
) == 0) {
900 list_add_tail(&next
->tree_list
, &head
);
904 next
= __btrfs_next_delayed_item(curr
);
908 if (!btrfs_is_continuous_delayed_item(curr
, next
))
914 btrfs_item_key_to_cpu(leaf
, &key
, i
);
920 ret
= btrfs_del_items(trans
, root
, path
, path
->slots
[0], nitems
);
924 list_for_each_entry_safe(curr
, next
, &head
, tree_list
) {
925 btrfs_delayed_item_release_metadata(root
, curr
);
926 list_del(&curr
->tree_list
);
927 btrfs_release_delayed_item(curr
);
934 static int btrfs_delete_delayed_items(struct btrfs_trans_handle
*trans
,
935 struct btrfs_path
*path
,
936 struct btrfs_root
*root
,
937 struct btrfs_delayed_node
*node
)
939 struct btrfs_delayed_item
*curr
, *prev
;
943 mutex_lock(&node
->mutex
);
944 curr
= __btrfs_first_delayed_deletion_item(node
);
948 ret
= btrfs_search_slot(trans
, root
, &curr
->key
, path
, -1, 1);
953 * can't find the item which the node points to, so this node
954 * is invalid, just drop it.
957 curr
= __btrfs_next_delayed_item(prev
);
958 btrfs_release_delayed_item(prev
);
960 btrfs_release_path(path
);
962 mutex_unlock(&node
->mutex
);
968 btrfs_batch_delete_items(trans
, root
, path
, curr
);
969 btrfs_release_path(path
);
970 mutex_unlock(&node
->mutex
);
974 btrfs_release_path(path
);
975 mutex_unlock(&node
->mutex
);
979 static void btrfs_release_delayed_inode(struct btrfs_delayed_node
*delayed_node
)
981 struct btrfs_delayed_root
*delayed_root
;
984 test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY
, &delayed_node
->flags
)) {
985 BUG_ON(!delayed_node
->root
);
986 clear_bit(BTRFS_DELAYED_NODE_INODE_DIRTY
, &delayed_node
->flags
);
987 delayed_node
->count
--;
989 delayed_root
= delayed_node
->root
->fs_info
->delayed_root
;
990 finish_one_item(delayed_root
);
994 static void btrfs_release_delayed_iref(struct btrfs_delayed_node
*delayed_node
)
996 struct btrfs_delayed_root
*delayed_root
;
998 ASSERT(delayed_node
->root
);
999 clear_bit(BTRFS_DELAYED_NODE_DEL_IREF
, &delayed_node
->flags
);
1000 delayed_node
->count
--;
1002 delayed_root
= delayed_node
->root
->fs_info
->delayed_root
;
1003 finish_one_item(delayed_root
);
1006 static int __btrfs_update_delayed_inode(struct btrfs_trans_handle
*trans
,
1007 struct btrfs_root
*root
,
1008 struct btrfs_path
*path
,
1009 struct btrfs_delayed_node
*node
)
1011 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1012 struct btrfs_key key
;
1013 struct btrfs_inode_item
*inode_item
;
1014 struct extent_buffer
*leaf
;
1018 key
.objectid
= node
->inode_id
;
1019 key
.type
= BTRFS_INODE_ITEM_KEY
;
1022 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF
, &node
->flags
))
1027 ret
= btrfs_lookup_inode(trans
, root
, path
, &key
, mod
);
1029 btrfs_release_path(path
);
1031 } else if (ret
< 0) {
1035 leaf
= path
->nodes
[0];
1036 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
1037 struct btrfs_inode_item
);
1038 write_extent_buffer(leaf
, &node
->inode_item
, (unsigned long)inode_item
,
1039 sizeof(struct btrfs_inode_item
));
1040 btrfs_mark_buffer_dirty(leaf
);
1042 if (!test_bit(BTRFS_DELAYED_NODE_DEL_IREF
, &node
->flags
))
1046 if (path
->slots
[0] >= btrfs_header_nritems(leaf
))
1049 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1050 if (key
.objectid
!= node
->inode_id
)
1053 if (key
.type
!= BTRFS_INODE_REF_KEY
&&
1054 key
.type
!= BTRFS_INODE_EXTREF_KEY
)
1058 * Delayed iref deletion is for the inode who has only one link,
1059 * so there is only one iref. The case that several irefs are
1060 * in the same item doesn't exist.
1062 btrfs_del_item(trans
, root
, path
);
1064 btrfs_release_delayed_iref(node
);
1066 btrfs_release_path(path
);
1068 btrfs_delayed_inode_release_metadata(fs_info
, node
, (ret
< 0));
1069 btrfs_release_delayed_inode(node
);
1074 btrfs_release_path(path
);
1076 key
.type
= BTRFS_INODE_EXTREF_KEY
;
1078 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1084 leaf
= path
->nodes
[0];
1089 static inline int btrfs_update_delayed_inode(struct btrfs_trans_handle
*trans
,
1090 struct btrfs_root
*root
,
1091 struct btrfs_path
*path
,
1092 struct btrfs_delayed_node
*node
)
1096 mutex_lock(&node
->mutex
);
1097 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY
, &node
->flags
)) {
1098 mutex_unlock(&node
->mutex
);
1102 ret
= __btrfs_update_delayed_inode(trans
, root
, path
, node
);
1103 mutex_unlock(&node
->mutex
);
1108 __btrfs_commit_inode_delayed_items(struct btrfs_trans_handle
*trans
,
1109 struct btrfs_path
*path
,
1110 struct btrfs_delayed_node
*node
)
1114 ret
= btrfs_insert_delayed_items(trans
, path
, node
->root
, node
);
1118 ret
= btrfs_delete_delayed_items(trans
, path
, node
->root
, node
);
1122 ret
= btrfs_update_delayed_inode(trans
, node
->root
, path
, node
);
1127 * Called when committing the transaction.
1128 * Returns 0 on success.
1129 * Returns < 0 on error and returns with an aborted transaction with any
1130 * outstanding delayed items cleaned up.
1132 static int __btrfs_run_delayed_items(struct btrfs_trans_handle
*trans
, int nr
)
1134 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
1135 struct btrfs_delayed_root
*delayed_root
;
1136 struct btrfs_delayed_node
*curr_node
, *prev_node
;
1137 struct btrfs_path
*path
;
1138 struct btrfs_block_rsv
*block_rsv
;
1140 bool count
= (nr
> 0);
1145 path
= btrfs_alloc_path();
1148 path
->leave_spinning
= 1;
1150 block_rsv
= trans
->block_rsv
;
1151 trans
->block_rsv
= &fs_info
->delayed_block_rsv
;
1153 delayed_root
= fs_info
->delayed_root
;
1155 curr_node
= btrfs_first_delayed_node(delayed_root
);
1156 while (curr_node
&& (!count
|| (count
&& nr
--))) {
1157 ret
= __btrfs_commit_inode_delayed_items(trans
, path
,
1160 btrfs_release_delayed_node(curr_node
);
1162 btrfs_abort_transaction(trans
, ret
);
1166 prev_node
= curr_node
;
1167 curr_node
= btrfs_next_delayed_node(curr_node
);
1168 btrfs_release_delayed_node(prev_node
);
1172 btrfs_release_delayed_node(curr_node
);
1173 btrfs_free_path(path
);
1174 trans
->block_rsv
= block_rsv
;
1179 int btrfs_run_delayed_items(struct btrfs_trans_handle
*trans
)
1181 return __btrfs_run_delayed_items(trans
, -1);
1184 int btrfs_run_delayed_items_nr(struct btrfs_trans_handle
*trans
, int nr
)
1186 return __btrfs_run_delayed_items(trans
, nr
);
1189 int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle
*trans
,
1190 struct btrfs_inode
*inode
)
1192 struct btrfs_delayed_node
*delayed_node
= btrfs_get_delayed_node(inode
);
1193 struct btrfs_path
*path
;
1194 struct btrfs_block_rsv
*block_rsv
;
1200 mutex_lock(&delayed_node
->mutex
);
1201 if (!delayed_node
->count
) {
1202 mutex_unlock(&delayed_node
->mutex
);
1203 btrfs_release_delayed_node(delayed_node
);
1206 mutex_unlock(&delayed_node
->mutex
);
1208 path
= btrfs_alloc_path();
1210 btrfs_release_delayed_node(delayed_node
);
1213 path
->leave_spinning
= 1;
1215 block_rsv
= trans
->block_rsv
;
1216 trans
->block_rsv
= &delayed_node
->root
->fs_info
->delayed_block_rsv
;
1218 ret
= __btrfs_commit_inode_delayed_items(trans
, path
, delayed_node
);
1220 btrfs_release_delayed_node(delayed_node
);
1221 btrfs_free_path(path
);
1222 trans
->block_rsv
= block_rsv
;
1227 int btrfs_commit_inode_delayed_inode(struct btrfs_inode
*inode
)
1229 struct btrfs_fs_info
*fs_info
= inode
->root
->fs_info
;
1230 struct btrfs_trans_handle
*trans
;
1231 struct btrfs_delayed_node
*delayed_node
= btrfs_get_delayed_node(inode
);
1232 struct btrfs_path
*path
;
1233 struct btrfs_block_rsv
*block_rsv
;
1239 mutex_lock(&delayed_node
->mutex
);
1240 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY
, &delayed_node
->flags
)) {
1241 mutex_unlock(&delayed_node
->mutex
);
1242 btrfs_release_delayed_node(delayed_node
);
1245 mutex_unlock(&delayed_node
->mutex
);
1247 trans
= btrfs_join_transaction(delayed_node
->root
);
1248 if (IS_ERR(trans
)) {
1249 ret
= PTR_ERR(trans
);
1253 path
= btrfs_alloc_path();
1258 path
->leave_spinning
= 1;
1260 block_rsv
= trans
->block_rsv
;
1261 trans
->block_rsv
= &fs_info
->delayed_block_rsv
;
1263 mutex_lock(&delayed_node
->mutex
);
1264 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY
, &delayed_node
->flags
))
1265 ret
= __btrfs_update_delayed_inode(trans
, delayed_node
->root
,
1266 path
, delayed_node
);
1269 mutex_unlock(&delayed_node
->mutex
);
1271 btrfs_free_path(path
);
1272 trans
->block_rsv
= block_rsv
;
1274 btrfs_end_transaction(trans
);
1275 btrfs_btree_balance_dirty(fs_info
);
1277 btrfs_release_delayed_node(delayed_node
);
1282 void btrfs_remove_delayed_node(struct btrfs_inode
*inode
)
1284 struct btrfs_delayed_node
*delayed_node
;
1286 delayed_node
= READ_ONCE(inode
->delayed_node
);
1290 inode
->delayed_node
= NULL
;
1291 btrfs_release_delayed_node(delayed_node
);
1294 struct btrfs_async_delayed_work
{
1295 struct btrfs_delayed_root
*delayed_root
;
1297 struct btrfs_work work
;
1300 static void btrfs_async_run_delayed_root(struct btrfs_work
*work
)
1302 struct btrfs_async_delayed_work
*async_work
;
1303 struct btrfs_delayed_root
*delayed_root
;
1304 struct btrfs_trans_handle
*trans
;
1305 struct btrfs_path
*path
;
1306 struct btrfs_delayed_node
*delayed_node
= NULL
;
1307 struct btrfs_root
*root
;
1308 struct btrfs_block_rsv
*block_rsv
;
1311 async_work
= container_of(work
, struct btrfs_async_delayed_work
, work
);
1312 delayed_root
= async_work
->delayed_root
;
1314 path
= btrfs_alloc_path();
1319 if (atomic_read(&delayed_root
->items
) <
1320 BTRFS_DELAYED_BACKGROUND
/ 2)
1323 delayed_node
= btrfs_first_prepared_delayed_node(delayed_root
);
1327 path
->leave_spinning
= 1;
1328 root
= delayed_node
->root
;
1330 trans
= btrfs_join_transaction(root
);
1331 if (IS_ERR(trans
)) {
1332 btrfs_release_path(path
);
1333 btrfs_release_prepared_delayed_node(delayed_node
);
1338 block_rsv
= trans
->block_rsv
;
1339 trans
->block_rsv
= &root
->fs_info
->delayed_block_rsv
;
1341 __btrfs_commit_inode_delayed_items(trans
, path
, delayed_node
);
1343 trans
->block_rsv
= block_rsv
;
1344 btrfs_end_transaction(trans
);
1345 btrfs_btree_balance_dirty_nodelay(root
->fs_info
);
1347 btrfs_release_path(path
);
1348 btrfs_release_prepared_delayed_node(delayed_node
);
1351 } while ((async_work
->nr
== 0 && total_done
< BTRFS_DELAYED_WRITEBACK
)
1352 || total_done
< async_work
->nr
);
1354 btrfs_free_path(path
);
1356 wake_up(&delayed_root
->wait
);
1361 static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root
*delayed_root
,
1362 struct btrfs_fs_info
*fs_info
, int nr
)
1364 struct btrfs_async_delayed_work
*async_work
;
1366 async_work
= kmalloc(sizeof(*async_work
), GFP_NOFS
);
1370 async_work
->delayed_root
= delayed_root
;
1371 btrfs_init_work(&async_work
->work
, btrfs_async_run_delayed_root
, NULL
,
1373 async_work
->nr
= nr
;
1375 btrfs_queue_work(fs_info
->delayed_workers
, &async_work
->work
);
1379 void btrfs_assert_delayed_root_empty(struct btrfs_fs_info
*fs_info
)
1381 WARN_ON(btrfs_first_delayed_node(fs_info
->delayed_root
));
1384 static int could_end_wait(struct btrfs_delayed_root
*delayed_root
, int seq
)
1386 int val
= atomic_read(&delayed_root
->items_seq
);
1388 if (val
< seq
|| val
>= seq
+ BTRFS_DELAYED_BATCH
)
1391 if (atomic_read(&delayed_root
->items
) < BTRFS_DELAYED_BACKGROUND
)
1397 void btrfs_balance_delayed_items(struct btrfs_fs_info
*fs_info
)
1399 struct btrfs_delayed_root
*delayed_root
= fs_info
->delayed_root
;
1401 if ((atomic_read(&delayed_root
->items
) < BTRFS_DELAYED_BACKGROUND
) ||
1402 btrfs_workqueue_normal_congested(fs_info
->delayed_workers
))
1405 if (atomic_read(&delayed_root
->items
) >= BTRFS_DELAYED_WRITEBACK
) {
1409 seq
= atomic_read(&delayed_root
->items_seq
);
1411 ret
= btrfs_wq_run_delayed_node(delayed_root
, fs_info
, 0);
1415 wait_event_interruptible(delayed_root
->wait
,
1416 could_end_wait(delayed_root
, seq
));
1420 btrfs_wq_run_delayed_node(delayed_root
, fs_info
, BTRFS_DELAYED_BATCH
);
1423 /* Will return 0 or -ENOMEM */
1424 int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle
*trans
,
1425 const char *name
, int name_len
,
1426 struct btrfs_inode
*dir
,
1427 struct btrfs_disk_key
*disk_key
, u8 type
,
1430 struct btrfs_delayed_node
*delayed_node
;
1431 struct btrfs_delayed_item
*delayed_item
;
1432 struct btrfs_dir_item
*dir_item
;
1435 delayed_node
= btrfs_get_or_create_delayed_node(dir
);
1436 if (IS_ERR(delayed_node
))
1437 return PTR_ERR(delayed_node
);
1439 delayed_item
= btrfs_alloc_delayed_item(sizeof(*dir_item
) + name_len
);
1440 if (!delayed_item
) {
1445 delayed_item
->key
.objectid
= btrfs_ino(dir
);
1446 delayed_item
->key
.type
= BTRFS_DIR_INDEX_KEY
;
1447 delayed_item
->key
.offset
= index
;
1449 dir_item
= (struct btrfs_dir_item
*)delayed_item
->data
;
1450 dir_item
->location
= *disk_key
;
1451 btrfs_set_stack_dir_transid(dir_item
, trans
->transid
);
1452 btrfs_set_stack_dir_data_len(dir_item
, 0);
1453 btrfs_set_stack_dir_name_len(dir_item
, name_len
);
1454 btrfs_set_stack_dir_type(dir_item
, type
);
1455 memcpy((char *)(dir_item
+ 1), name
, name_len
);
1457 ret
= btrfs_delayed_item_reserve_metadata(trans
, dir
->root
, delayed_item
);
1459 * we have reserved enough space when we start a new transaction,
1460 * so reserving metadata failure is impossible
1464 mutex_lock(&delayed_node
->mutex
);
1465 ret
= __btrfs_add_delayed_insertion_item(delayed_node
, delayed_item
);
1466 if (unlikely(ret
)) {
1467 btrfs_err(trans
->fs_info
,
1468 "err add delayed dir index item(name: %.*s) into the insertion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
1469 name_len
, name
, delayed_node
->root
->root_key
.objectid
,
1470 delayed_node
->inode_id
, ret
);
1473 mutex_unlock(&delayed_node
->mutex
);
1476 btrfs_release_delayed_node(delayed_node
);
1480 static int btrfs_delete_delayed_insertion_item(struct btrfs_fs_info
*fs_info
,
1481 struct btrfs_delayed_node
*node
,
1482 struct btrfs_key
*key
)
1484 struct btrfs_delayed_item
*item
;
1486 mutex_lock(&node
->mutex
);
1487 item
= __btrfs_lookup_delayed_insertion_item(node
, key
);
1489 mutex_unlock(&node
->mutex
);
1493 btrfs_delayed_item_release_metadata(node
->root
, item
);
1494 btrfs_release_delayed_item(item
);
1495 mutex_unlock(&node
->mutex
);
1499 int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle
*trans
,
1500 struct btrfs_inode
*dir
, u64 index
)
1502 struct btrfs_delayed_node
*node
;
1503 struct btrfs_delayed_item
*item
;
1504 struct btrfs_key item_key
;
1507 node
= btrfs_get_or_create_delayed_node(dir
);
1509 return PTR_ERR(node
);
1511 item_key
.objectid
= btrfs_ino(dir
);
1512 item_key
.type
= BTRFS_DIR_INDEX_KEY
;
1513 item_key
.offset
= index
;
1515 ret
= btrfs_delete_delayed_insertion_item(trans
->fs_info
, node
,
1520 item
= btrfs_alloc_delayed_item(0);
1526 item
->key
= item_key
;
1528 ret
= btrfs_delayed_item_reserve_metadata(trans
, dir
->root
, item
);
1530 * we have reserved enough space when we start a new transaction,
1531 * so reserving metadata failure is impossible.
1534 btrfs_err(trans
->fs_info
,
1535 "metadata reservation failed for delayed dir item deltiona, should have been reserved");
1536 btrfs_release_delayed_item(item
);
1540 mutex_lock(&node
->mutex
);
1541 ret
= __btrfs_add_delayed_deletion_item(node
, item
);
1542 if (unlikely(ret
)) {
1543 btrfs_err(trans
->fs_info
,
1544 "err add delayed dir index item(index: %llu) into the deletion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
1545 index
, node
->root
->root_key
.objectid
,
1546 node
->inode_id
, ret
);
1547 btrfs_delayed_item_release_metadata(dir
->root
, item
);
1548 btrfs_release_delayed_item(item
);
1550 mutex_unlock(&node
->mutex
);
1552 btrfs_release_delayed_node(node
);
1556 int btrfs_inode_delayed_dir_index_count(struct btrfs_inode
*inode
)
1558 struct btrfs_delayed_node
*delayed_node
= btrfs_get_delayed_node(inode
);
1564 * Since we have held i_mutex of this directory, it is impossible that
1565 * a new directory index is added into the delayed node and index_cnt
1566 * is updated now. So we needn't lock the delayed node.
1568 if (!delayed_node
->index_cnt
) {
1569 btrfs_release_delayed_node(delayed_node
);
1573 inode
->index_cnt
= delayed_node
->index_cnt
;
1574 btrfs_release_delayed_node(delayed_node
);
1578 bool btrfs_readdir_get_delayed_items(struct inode
*inode
,
1579 struct list_head
*ins_list
,
1580 struct list_head
*del_list
)
1582 struct btrfs_delayed_node
*delayed_node
;
1583 struct btrfs_delayed_item
*item
;
1585 delayed_node
= btrfs_get_delayed_node(BTRFS_I(inode
));
1590 * We can only do one readdir with delayed items at a time because of
1591 * item->readdir_list.
1593 inode_unlock_shared(inode
);
1596 mutex_lock(&delayed_node
->mutex
);
1597 item
= __btrfs_first_delayed_insertion_item(delayed_node
);
1599 refcount_inc(&item
->refs
);
1600 list_add_tail(&item
->readdir_list
, ins_list
);
1601 item
= __btrfs_next_delayed_item(item
);
1604 item
= __btrfs_first_delayed_deletion_item(delayed_node
);
1606 refcount_inc(&item
->refs
);
1607 list_add_tail(&item
->readdir_list
, del_list
);
1608 item
= __btrfs_next_delayed_item(item
);
1610 mutex_unlock(&delayed_node
->mutex
);
1612 * This delayed node is still cached in the btrfs inode, so refs
1613 * must be > 1 now, and we needn't check it is going to be freed
1616 * Besides that, this function is used to read dir, we do not
1617 * insert/delete delayed items in this period. So we also needn't
1618 * requeue or dequeue this delayed node.
1620 refcount_dec(&delayed_node
->refs
);
1625 void btrfs_readdir_put_delayed_items(struct inode
*inode
,
1626 struct list_head
*ins_list
,
1627 struct list_head
*del_list
)
1629 struct btrfs_delayed_item
*curr
, *next
;
1631 list_for_each_entry_safe(curr
, next
, ins_list
, readdir_list
) {
1632 list_del(&curr
->readdir_list
);
1633 if (refcount_dec_and_test(&curr
->refs
))
1637 list_for_each_entry_safe(curr
, next
, del_list
, readdir_list
) {
1638 list_del(&curr
->readdir_list
);
1639 if (refcount_dec_and_test(&curr
->refs
))
1644 * The VFS is going to do up_read(), so we need to downgrade back to a
1647 downgrade_write(&inode
->i_rwsem
);
1650 int btrfs_should_delete_dir_index(struct list_head
*del_list
,
1653 struct btrfs_delayed_item
*curr
;
1656 list_for_each_entry(curr
, del_list
, readdir_list
) {
1657 if (curr
->key
.offset
> index
)
1659 if (curr
->key
.offset
== index
) {
1668 * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree
1671 int btrfs_readdir_delayed_dir_index(struct dir_context
*ctx
,
1672 struct list_head
*ins_list
)
1674 struct btrfs_dir_item
*di
;
1675 struct btrfs_delayed_item
*curr
, *next
;
1676 struct btrfs_key location
;
1680 unsigned char d_type
;
1682 if (list_empty(ins_list
))
1686 * Changing the data of the delayed item is impossible. So
1687 * we needn't lock them. And we have held i_mutex of the
1688 * directory, nobody can delete any directory indexes now.
1690 list_for_each_entry_safe(curr
, next
, ins_list
, readdir_list
) {
1691 list_del(&curr
->readdir_list
);
1693 if (curr
->key
.offset
< ctx
->pos
) {
1694 if (refcount_dec_and_test(&curr
->refs
))
1699 ctx
->pos
= curr
->key
.offset
;
1701 di
= (struct btrfs_dir_item
*)curr
->data
;
1702 name
= (char *)(di
+ 1);
1703 name_len
= btrfs_stack_dir_name_len(di
);
1705 d_type
= fs_ftype_to_dtype(di
->type
);
1706 btrfs_disk_key_to_cpu(&location
, &di
->location
);
1708 over
= !dir_emit(ctx
, name
, name_len
,
1709 location
.objectid
, d_type
);
1711 if (refcount_dec_and_test(&curr
->refs
))
1721 static void fill_stack_inode_item(struct btrfs_trans_handle
*trans
,
1722 struct btrfs_inode_item
*inode_item
,
1723 struct inode
*inode
)
1725 btrfs_set_stack_inode_uid(inode_item
, i_uid_read(inode
));
1726 btrfs_set_stack_inode_gid(inode_item
, i_gid_read(inode
));
1727 btrfs_set_stack_inode_size(inode_item
, BTRFS_I(inode
)->disk_i_size
);
1728 btrfs_set_stack_inode_mode(inode_item
, inode
->i_mode
);
1729 btrfs_set_stack_inode_nlink(inode_item
, inode
->i_nlink
);
1730 btrfs_set_stack_inode_nbytes(inode_item
, inode_get_bytes(inode
));
1731 btrfs_set_stack_inode_generation(inode_item
,
1732 BTRFS_I(inode
)->generation
);
1733 btrfs_set_stack_inode_sequence(inode_item
,
1734 inode_peek_iversion(inode
));
1735 btrfs_set_stack_inode_transid(inode_item
, trans
->transid
);
1736 btrfs_set_stack_inode_rdev(inode_item
, inode
->i_rdev
);
1737 btrfs_set_stack_inode_flags(inode_item
, BTRFS_I(inode
)->flags
);
1738 btrfs_set_stack_inode_block_group(inode_item
, 0);
1740 btrfs_set_stack_timespec_sec(&inode_item
->atime
,
1741 inode
->i_atime
.tv_sec
);
1742 btrfs_set_stack_timespec_nsec(&inode_item
->atime
,
1743 inode
->i_atime
.tv_nsec
);
1745 btrfs_set_stack_timespec_sec(&inode_item
->mtime
,
1746 inode
->i_mtime
.tv_sec
);
1747 btrfs_set_stack_timespec_nsec(&inode_item
->mtime
,
1748 inode
->i_mtime
.tv_nsec
);
1750 btrfs_set_stack_timespec_sec(&inode_item
->ctime
,
1751 inode
->i_ctime
.tv_sec
);
1752 btrfs_set_stack_timespec_nsec(&inode_item
->ctime
,
1753 inode
->i_ctime
.tv_nsec
);
1755 btrfs_set_stack_timespec_sec(&inode_item
->otime
,
1756 BTRFS_I(inode
)->i_otime
.tv_sec
);
1757 btrfs_set_stack_timespec_nsec(&inode_item
->otime
,
1758 BTRFS_I(inode
)->i_otime
.tv_nsec
);
1761 int btrfs_fill_inode(struct inode
*inode
, u32
*rdev
)
1763 struct btrfs_delayed_node
*delayed_node
;
1764 struct btrfs_inode_item
*inode_item
;
1766 delayed_node
= btrfs_get_delayed_node(BTRFS_I(inode
));
1770 mutex_lock(&delayed_node
->mutex
);
1771 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY
, &delayed_node
->flags
)) {
1772 mutex_unlock(&delayed_node
->mutex
);
1773 btrfs_release_delayed_node(delayed_node
);
1777 inode_item
= &delayed_node
->inode_item
;
1779 i_uid_write(inode
, btrfs_stack_inode_uid(inode_item
));
1780 i_gid_write(inode
, btrfs_stack_inode_gid(inode_item
));
1781 btrfs_i_size_write(BTRFS_I(inode
), btrfs_stack_inode_size(inode_item
));
1782 inode
->i_mode
= btrfs_stack_inode_mode(inode_item
);
1783 set_nlink(inode
, btrfs_stack_inode_nlink(inode_item
));
1784 inode_set_bytes(inode
, btrfs_stack_inode_nbytes(inode_item
));
1785 BTRFS_I(inode
)->generation
= btrfs_stack_inode_generation(inode_item
);
1786 BTRFS_I(inode
)->last_trans
= btrfs_stack_inode_transid(inode_item
);
1788 inode_set_iversion_queried(inode
,
1789 btrfs_stack_inode_sequence(inode_item
));
1791 *rdev
= btrfs_stack_inode_rdev(inode_item
);
1792 BTRFS_I(inode
)->flags
= btrfs_stack_inode_flags(inode_item
);
1794 inode
->i_atime
.tv_sec
= btrfs_stack_timespec_sec(&inode_item
->atime
);
1795 inode
->i_atime
.tv_nsec
= btrfs_stack_timespec_nsec(&inode_item
->atime
);
1797 inode
->i_mtime
.tv_sec
= btrfs_stack_timespec_sec(&inode_item
->mtime
);
1798 inode
->i_mtime
.tv_nsec
= btrfs_stack_timespec_nsec(&inode_item
->mtime
);
1800 inode
->i_ctime
.tv_sec
= btrfs_stack_timespec_sec(&inode_item
->ctime
);
1801 inode
->i_ctime
.tv_nsec
= btrfs_stack_timespec_nsec(&inode_item
->ctime
);
1803 BTRFS_I(inode
)->i_otime
.tv_sec
=
1804 btrfs_stack_timespec_sec(&inode_item
->otime
);
1805 BTRFS_I(inode
)->i_otime
.tv_nsec
=
1806 btrfs_stack_timespec_nsec(&inode_item
->otime
);
1808 inode
->i_generation
= BTRFS_I(inode
)->generation
;
1809 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
1811 mutex_unlock(&delayed_node
->mutex
);
1812 btrfs_release_delayed_node(delayed_node
);
1816 int btrfs_delayed_update_inode(struct btrfs_trans_handle
*trans
,
1817 struct btrfs_root
*root
, struct inode
*inode
)
1819 struct btrfs_delayed_node
*delayed_node
;
1822 delayed_node
= btrfs_get_or_create_delayed_node(BTRFS_I(inode
));
1823 if (IS_ERR(delayed_node
))
1824 return PTR_ERR(delayed_node
);
1826 mutex_lock(&delayed_node
->mutex
);
1827 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY
, &delayed_node
->flags
)) {
1828 fill_stack_inode_item(trans
, &delayed_node
->inode_item
, inode
);
1832 ret
= btrfs_delayed_inode_reserve_metadata(trans
, root
, BTRFS_I(inode
),
1837 fill_stack_inode_item(trans
, &delayed_node
->inode_item
, inode
);
1838 set_bit(BTRFS_DELAYED_NODE_INODE_DIRTY
, &delayed_node
->flags
);
1839 delayed_node
->count
++;
1840 atomic_inc(&root
->fs_info
->delayed_root
->items
);
1842 mutex_unlock(&delayed_node
->mutex
);
1843 btrfs_release_delayed_node(delayed_node
);
1847 int btrfs_delayed_delete_inode_ref(struct btrfs_inode
*inode
)
1849 struct btrfs_fs_info
*fs_info
= inode
->root
->fs_info
;
1850 struct btrfs_delayed_node
*delayed_node
;
1853 * we don't do delayed inode updates during log recovery because it
1854 * leads to enospc problems. This means we also can't do
1855 * delayed inode refs
1857 if (test_bit(BTRFS_FS_LOG_RECOVERING
, &fs_info
->flags
))
1860 delayed_node
= btrfs_get_or_create_delayed_node(inode
);
1861 if (IS_ERR(delayed_node
))
1862 return PTR_ERR(delayed_node
);
1865 * We don't reserve space for inode ref deletion is because:
1866 * - We ONLY do async inode ref deletion for the inode who has only
1867 * one link(i_nlink == 1), it means there is only one inode ref.
1868 * And in most case, the inode ref and the inode item are in the
1869 * same leaf, and we will deal with them at the same time.
1870 * Since we are sure we will reserve the space for the inode item,
1871 * it is unnecessary to reserve space for inode ref deletion.
1872 * - If the inode ref and the inode item are not in the same leaf,
1873 * We also needn't worry about enospc problem, because we reserve
1874 * much more space for the inode update than it needs.
1875 * - At the worst, we can steal some space from the global reservation.
1878 mutex_lock(&delayed_node
->mutex
);
1879 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF
, &delayed_node
->flags
))
1882 set_bit(BTRFS_DELAYED_NODE_DEL_IREF
, &delayed_node
->flags
);
1883 delayed_node
->count
++;
1884 atomic_inc(&fs_info
->delayed_root
->items
);
1886 mutex_unlock(&delayed_node
->mutex
);
1887 btrfs_release_delayed_node(delayed_node
);
1891 static void __btrfs_kill_delayed_node(struct btrfs_delayed_node
*delayed_node
)
1893 struct btrfs_root
*root
= delayed_node
->root
;
1894 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1895 struct btrfs_delayed_item
*curr_item
, *prev_item
;
1897 mutex_lock(&delayed_node
->mutex
);
1898 curr_item
= __btrfs_first_delayed_insertion_item(delayed_node
);
1900 btrfs_delayed_item_release_metadata(root
, curr_item
);
1901 prev_item
= curr_item
;
1902 curr_item
= __btrfs_next_delayed_item(prev_item
);
1903 btrfs_release_delayed_item(prev_item
);
1906 curr_item
= __btrfs_first_delayed_deletion_item(delayed_node
);
1908 btrfs_delayed_item_release_metadata(root
, curr_item
);
1909 prev_item
= curr_item
;
1910 curr_item
= __btrfs_next_delayed_item(prev_item
);
1911 btrfs_release_delayed_item(prev_item
);
1914 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF
, &delayed_node
->flags
))
1915 btrfs_release_delayed_iref(delayed_node
);
1917 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY
, &delayed_node
->flags
)) {
1918 btrfs_delayed_inode_release_metadata(fs_info
, delayed_node
, false);
1919 btrfs_release_delayed_inode(delayed_node
);
1921 mutex_unlock(&delayed_node
->mutex
);
1924 void btrfs_kill_delayed_inode_items(struct btrfs_inode
*inode
)
1926 struct btrfs_delayed_node
*delayed_node
;
1928 delayed_node
= btrfs_get_delayed_node(inode
);
1932 __btrfs_kill_delayed_node(delayed_node
);
1933 btrfs_release_delayed_node(delayed_node
);
1936 void btrfs_kill_all_delayed_nodes(struct btrfs_root
*root
)
1939 struct btrfs_delayed_node
*delayed_nodes
[8];
1943 spin_lock(&root
->inode_lock
);
1944 n
= radix_tree_gang_lookup(&root
->delayed_nodes_tree
,
1945 (void **)delayed_nodes
, inode_id
,
1946 ARRAY_SIZE(delayed_nodes
));
1948 spin_unlock(&root
->inode_lock
);
1952 inode_id
= delayed_nodes
[n
- 1]->inode_id
+ 1;
1953 for (i
= 0; i
< n
; i
++) {
1955 * Don't increase refs in case the node is dead and
1956 * about to be removed from the tree in the loop below
1958 if (!refcount_inc_not_zero(&delayed_nodes
[i
]->refs
))
1959 delayed_nodes
[i
] = NULL
;
1961 spin_unlock(&root
->inode_lock
);
1963 for (i
= 0; i
< n
; i
++) {
1964 if (!delayed_nodes
[i
])
1966 __btrfs_kill_delayed_node(delayed_nodes
[i
]);
1967 btrfs_release_delayed_node(delayed_nodes
[i
]);
1972 void btrfs_destroy_delayed_inodes(struct btrfs_fs_info
*fs_info
)
1974 struct btrfs_delayed_node
*curr_node
, *prev_node
;
1976 curr_node
= btrfs_first_delayed_node(fs_info
->delayed_root
);
1978 __btrfs_kill_delayed_node(curr_node
);
1980 prev_node
= curr_node
;
1981 curr_node
= btrfs_next_delayed_node(curr_node
);
1982 btrfs_release_delayed_node(prev_node
);