2 * Copyright (C) 2011 Fujitsu. All rights reserved.
3 * Written by Miao Xie <miaox@cn.fujitsu.com>
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public
7 * License v2 as published by the Free Software Foundation.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12 * General Public License for more details.
14 * You should have received a copy of the GNU General Public
15 * License along with this program; if not, write to the
16 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
17 * Boston, MA 021110-1307, USA.
20 #include <linux/slab.h>
21 #include "delayed-inode.h"
23 #include "transaction.h"
25 #define BTRFS_DELAYED_WRITEBACK 512
26 #define BTRFS_DELAYED_BACKGROUND 128
27 #define BTRFS_DELAYED_BATCH 16
29 static struct kmem_cache
*delayed_node_cache
;
31 int __init
btrfs_delayed_inode_init(void)
33 delayed_node_cache
= kmem_cache_create("btrfs_delayed_node",
34 sizeof(struct btrfs_delayed_node
),
36 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
,
38 if (!delayed_node_cache
)
43 void btrfs_delayed_inode_exit(void)
45 if (delayed_node_cache
)
46 kmem_cache_destroy(delayed_node_cache
);
49 static inline void btrfs_init_delayed_node(
50 struct btrfs_delayed_node
*delayed_node
,
51 struct btrfs_root
*root
, u64 inode_id
)
53 delayed_node
->root
= root
;
54 delayed_node
->inode_id
= inode_id
;
55 atomic_set(&delayed_node
->refs
, 0);
56 delayed_node
->count
= 0;
57 delayed_node
->in_list
= 0;
58 delayed_node
->inode_dirty
= 0;
59 delayed_node
->ins_root
= RB_ROOT
;
60 delayed_node
->del_root
= RB_ROOT
;
61 mutex_init(&delayed_node
->mutex
);
62 delayed_node
->index_cnt
= 0;
63 INIT_LIST_HEAD(&delayed_node
->n_list
);
64 INIT_LIST_HEAD(&delayed_node
->p_list
);
65 delayed_node
->bytes_reserved
= 0;
66 memset(&delayed_node
->inode_item
, 0, sizeof(delayed_node
->inode_item
));
69 static inline int btrfs_is_continuous_delayed_item(
70 struct btrfs_delayed_item
*item1
,
71 struct btrfs_delayed_item
*item2
)
73 if (item1
->key
.type
== BTRFS_DIR_INDEX_KEY
&&
74 item1
->key
.objectid
== item2
->key
.objectid
&&
75 item1
->key
.type
== item2
->key
.type
&&
76 item1
->key
.offset
+ 1 == item2
->key
.offset
)
81 static inline struct btrfs_delayed_root
*btrfs_get_delayed_root(
82 struct btrfs_root
*root
)
84 return root
->fs_info
->delayed_root
;
87 static struct btrfs_delayed_node
*btrfs_get_delayed_node(struct inode
*inode
)
89 struct btrfs_inode
*btrfs_inode
= BTRFS_I(inode
);
90 struct btrfs_root
*root
= btrfs_inode
->root
;
91 u64 ino
= btrfs_ino(inode
);
92 struct btrfs_delayed_node
*node
;
94 node
= ACCESS_ONCE(btrfs_inode
->delayed_node
);
96 atomic_inc(&node
->refs
);
100 spin_lock(&root
->inode_lock
);
101 node
= radix_tree_lookup(&root
->delayed_nodes_tree
, ino
);
103 if (btrfs_inode
->delayed_node
) {
104 atomic_inc(&node
->refs
); /* can be accessed */
105 BUG_ON(btrfs_inode
->delayed_node
!= node
);
106 spin_unlock(&root
->inode_lock
);
109 btrfs_inode
->delayed_node
= node
;
110 atomic_inc(&node
->refs
); /* can be accessed */
111 atomic_inc(&node
->refs
); /* cached in the inode */
112 spin_unlock(&root
->inode_lock
);
115 spin_unlock(&root
->inode_lock
);
120 /* Will return either the node or PTR_ERR(-ENOMEM) */
121 static struct btrfs_delayed_node
*btrfs_get_or_create_delayed_node(
124 struct btrfs_delayed_node
*node
;
125 struct btrfs_inode
*btrfs_inode
= BTRFS_I(inode
);
126 struct btrfs_root
*root
= btrfs_inode
->root
;
127 u64 ino
= btrfs_ino(inode
);
131 node
= btrfs_get_delayed_node(inode
);
135 node
= kmem_cache_alloc(delayed_node_cache
, GFP_NOFS
);
137 return ERR_PTR(-ENOMEM
);
138 btrfs_init_delayed_node(node
, root
, ino
);
140 atomic_inc(&node
->refs
); /* cached in the btrfs inode */
141 atomic_inc(&node
->refs
); /* can be accessed */
143 ret
= radix_tree_preload(GFP_NOFS
& ~__GFP_HIGHMEM
);
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 kmem_cache_free(delayed_node_cache
, node
);
153 spin_unlock(&root
->inode_lock
);
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
);
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 atomic_inc(&node
->refs
); /* inserted into list */
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
);
196 atomic_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
);
202 spin_unlock(&root
->lock
);
205 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 atomic_inc(&node
->refs
);
219 spin_unlock(&delayed_root
->lock
);
224 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 (!node
->in_list
) { /* not in the list */
234 if (list_empty(&delayed_root
->node_list
))
236 p
= delayed_root
->node_list
.next
;
237 } else if (list_is_last(&node
->n_list
, &delayed_root
->node_list
))
240 p
= node
->n_list
.next
;
242 next
= list_entry(p
, struct btrfs_delayed_node
, n_list
);
243 atomic_inc(&next
->refs
);
245 spin_unlock(&delayed_root
->lock
);
250 static void __btrfs_release_delayed_node(
251 struct btrfs_delayed_node
*delayed_node
,
254 struct btrfs_delayed_root
*delayed_root
;
259 delayed_root
= delayed_node
->root
->fs_info
->delayed_root
;
261 mutex_lock(&delayed_node
->mutex
);
262 if (delayed_node
->count
)
263 btrfs_queue_delayed_node(delayed_root
, delayed_node
, mod
);
265 btrfs_dequeue_delayed_node(delayed_root
, delayed_node
);
266 mutex_unlock(&delayed_node
->mutex
);
268 if (atomic_dec_and_test(&delayed_node
->refs
)) {
269 struct btrfs_root
*root
= delayed_node
->root
;
270 spin_lock(&root
->inode_lock
);
271 if (atomic_read(&delayed_node
->refs
) == 0) {
272 radix_tree_delete(&root
->delayed_nodes_tree
,
273 delayed_node
->inode_id
);
274 kmem_cache_free(delayed_node_cache
, delayed_node
);
276 spin_unlock(&root
->inode_lock
);
280 static inline void btrfs_release_delayed_node(struct btrfs_delayed_node
*node
)
282 __btrfs_release_delayed_node(node
, 0);
285 struct btrfs_delayed_node
*btrfs_first_prepared_delayed_node(
286 struct btrfs_delayed_root
*delayed_root
)
289 struct btrfs_delayed_node
*node
= NULL
;
291 spin_lock(&delayed_root
->lock
);
292 if (list_empty(&delayed_root
->prepare_list
))
295 p
= delayed_root
->prepare_list
.next
;
297 node
= list_entry(p
, struct btrfs_delayed_node
, p_list
);
298 atomic_inc(&node
->refs
);
300 spin_unlock(&delayed_root
->lock
);
305 static inline void btrfs_release_prepared_delayed_node(
306 struct btrfs_delayed_node
*node
)
308 __btrfs_release_delayed_node(node
, 1);
311 struct btrfs_delayed_item
*btrfs_alloc_delayed_item(u32 data_len
)
313 struct btrfs_delayed_item
*item
;
314 item
= kmalloc(sizeof(*item
) + data_len
, GFP_NOFS
);
316 item
->data_len
= data_len
;
317 item
->ins_or_del
= 0;
318 item
->bytes_reserved
= 0;
319 item
->delayed_node
= NULL
;
320 atomic_set(&item
->refs
, 1);
326 * __btrfs_lookup_delayed_item - look up the delayed item by key
327 * @delayed_node: pointer to the delayed node
328 * @key: the key to look up
329 * @prev: used to store the prev item if the right item isn't found
330 * @next: used to store the next item if the right item isn't found
332 * Note: if we don't find the right item, we will return the prev item and
335 static struct btrfs_delayed_item
*__btrfs_lookup_delayed_item(
336 struct rb_root
*root
,
337 struct btrfs_key
*key
,
338 struct btrfs_delayed_item
**prev
,
339 struct btrfs_delayed_item
**next
)
341 struct rb_node
*node
, *prev_node
= NULL
;
342 struct btrfs_delayed_item
*delayed_item
= NULL
;
345 node
= root
->rb_node
;
348 delayed_item
= rb_entry(node
, struct btrfs_delayed_item
,
351 ret
= btrfs_comp_cpu_keys(&delayed_item
->key
, key
);
353 node
= node
->rb_right
;
355 node
= node
->rb_left
;
364 *prev
= delayed_item
;
365 else if ((node
= rb_prev(prev_node
)) != NULL
) {
366 *prev
= rb_entry(node
, struct btrfs_delayed_item
,
376 *next
= delayed_item
;
377 else if ((node
= rb_next(prev_node
)) != NULL
) {
378 *next
= rb_entry(node
, struct btrfs_delayed_item
,
386 struct btrfs_delayed_item
*__btrfs_lookup_delayed_insertion_item(
387 struct btrfs_delayed_node
*delayed_node
,
388 struct btrfs_key
*key
)
390 struct btrfs_delayed_item
*item
;
392 item
= __btrfs_lookup_delayed_item(&delayed_node
->ins_root
, key
,
397 struct btrfs_delayed_item
*__btrfs_lookup_delayed_deletion_item(
398 struct btrfs_delayed_node
*delayed_node
,
399 struct btrfs_key
*key
)
401 struct btrfs_delayed_item
*item
;
403 item
= __btrfs_lookup_delayed_item(&delayed_node
->del_root
, key
,
408 struct btrfs_delayed_item
*__btrfs_search_delayed_insertion_item(
409 struct btrfs_delayed_node
*delayed_node
,
410 struct btrfs_key
*key
)
412 struct btrfs_delayed_item
*item
, *next
;
414 item
= __btrfs_lookup_delayed_item(&delayed_node
->ins_root
, key
,
422 struct btrfs_delayed_item
*__btrfs_search_delayed_deletion_item(
423 struct btrfs_delayed_node
*delayed_node
,
424 struct btrfs_key
*key
)
426 struct btrfs_delayed_item
*item
, *next
;
428 item
= __btrfs_lookup_delayed_item(&delayed_node
->del_root
, key
,
436 static int __btrfs_add_delayed_item(struct btrfs_delayed_node
*delayed_node
,
437 struct btrfs_delayed_item
*ins
,
440 struct rb_node
**p
, *node
;
441 struct rb_node
*parent_node
= NULL
;
442 struct rb_root
*root
;
443 struct btrfs_delayed_item
*item
;
446 if (action
== BTRFS_DELAYED_INSERTION_ITEM
)
447 root
= &delayed_node
->ins_root
;
448 else if (action
== BTRFS_DELAYED_DELETION_ITEM
)
449 root
= &delayed_node
->del_root
;
453 node
= &ins
->rb_node
;
457 item
= rb_entry(parent_node
, struct btrfs_delayed_item
,
460 cmp
= btrfs_comp_cpu_keys(&item
->key
, &ins
->key
);
469 rb_link_node(node
, parent_node
, p
);
470 rb_insert_color(node
, root
);
471 ins
->delayed_node
= delayed_node
;
472 ins
->ins_or_del
= action
;
474 if (ins
->key
.type
== BTRFS_DIR_INDEX_KEY
&&
475 action
== BTRFS_DELAYED_INSERTION_ITEM
&&
476 ins
->key
.offset
>= delayed_node
->index_cnt
)
477 delayed_node
->index_cnt
= ins
->key
.offset
+ 1;
479 delayed_node
->count
++;
480 atomic_inc(&delayed_node
->root
->fs_info
->delayed_root
->items
);
484 static int __btrfs_add_delayed_insertion_item(struct btrfs_delayed_node
*node
,
485 struct btrfs_delayed_item
*item
)
487 return __btrfs_add_delayed_item(node
, item
,
488 BTRFS_DELAYED_INSERTION_ITEM
);
491 static int __btrfs_add_delayed_deletion_item(struct btrfs_delayed_node
*node
,
492 struct btrfs_delayed_item
*item
)
494 return __btrfs_add_delayed_item(node
, item
,
495 BTRFS_DELAYED_DELETION_ITEM
);
498 static void finish_one_item(struct btrfs_delayed_root
*delayed_root
)
500 int seq
= atomic_inc_return(&delayed_root
->items_seq
);
501 if ((atomic_dec_return(&delayed_root
->items
) <
502 BTRFS_DELAYED_BACKGROUND
|| seq
% BTRFS_DELAYED_BATCH
== 0) &&
503 waitqueue_active(&delayed_root
->wait
))
504 wake_up(&delayed_root
->wait
);
507 static void __btrfs_remove_delayed_item(struct btrfs_delayed_item
*delayed_item
)
509 struct rb_root
*root
;
510 struct btrfs_delayed_root
*delayed_root
;
512 delayed_root
= delayed_item
->delayed_node
->root
->fs_info
->delayed_root
;
514 BUG_ON(!delayed_root
);
515 BUG_ON(delayed_item
->ins_or_del
!= BTRFS_DELAYED_DELETION_ITEM
&&
516 delayed_item
->ins_or_del
!= BTRFS_DELAYED_INSERTION_ITEM
);
518 if (delayed_item
->ins_or_del
== BTRFS_DELAYED_INSERTION_ITEM
)
519 root
= &delayed_item
->delayed_node
->ins_root
;
521 root
= &delayed_item
->delayed_node
->del_root
;
523 rb_erase(&delayed_item
->rb_node
, root
);
524 delayed_item
->delayed_node
->count
--;
526 finish_one_item(delayed_root
);
529 static void btrfs_release_delayed_item(struct btrfs_delayed_item
*item
)
532 __btrfs_remove_delayed_item(item
);
533 if (atomic_dec_and_test(&item
->refs
))
538 struct btrfs_delayed_item
*__btrfs_first_delayed_insertion_item(
539 struct btrfs_delayed_node
*delayed_node
)
542 struct btrfs_delayed_item
*item
= NULL
;
544 p
= rb_first(&delayed_node
->ins_root
);
546 item
= rb_entry(p
, struct btrfs_delayed_item
, rb_node
);
551 struct btrfs_delayed_item
*__btrfs_first_delayed_deletion_item(
552 struct btrfs_delayed_node
*delayed_node
)
555 struct btrfs_delayed_item
*item
= NULL
;
557 p
= rb_first(&delayed_node
->del_root
);
559 item
= rb_entry(p
, struct btrfs_delayed_item
, rb_node
);
564 struct btrfs_delayed_item
*__btrfs_next_delayed_item(
565 struct btrfs_delayed_item
*item
)
568 struct btrfs_delayed_item
*next
= NULL
;
570 p
= rb_next(&item
->rb_node
);
572 next
= rb_entry(p
, struct btrfs_delayed_item
, rb_node
);
577 static inline struct btrfs_root
*btrfs_get_fs_root(struct btrfs_root
*root
,
580 struct btrfs_key root_key
;
582 if (root
->objectid
== root_id
)
585 root_key
.objectid
= root_id
;
586 root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
587 root_key
.offset
= (u64
)-1;
588 return btrfs_read_fs_root_no_name(root
->fs_info
, &root_key
);
591 static int btrfs_delayed_item_reserve_metadata(struct btrfs_trans_handle
*trans
,
592 struct btrfs_root
*root
,
593 struct btrfs_delayed_item
*item
)
595 struct btrfs_block_rsv
*src_rsv
;
596 struct btrfs_block_rsv
*dst_rsv
;
600 if (!trans
->bytes_reserved
)
603 src_rsv
= trans
->block_rsv
;
604 dst_rsv
= &root
->fs_info
->delayed_block_rsv
;
606 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
607 ret
= btrfs_block_rsv_migrate(src_rsv
, dst_rsv
, num_bytes
);
609 trace_btrfs_space_reservation(root
->fs_info
, "delayed_item",
612 item
->bytes_reserved
= num_bytes
;
618 static void btrfs_delayed_item_release_metadata(struct btrfs_root
*root
,
619 struct btrfs_delayed_item
*item
)
621 struct btrfs_block_rsv
*rsv
;
623 if (!item
->bytes_reserved
)
626 rsv
= &root
->fs_info
->delayed_block_rsv
;
627 trace_btrfs_space_reservation(root
->fs_info
, "delayed_item",
628 item
->key
.objectid
, item
->bytes_reserved
,
630 btrfs_block_rsv_release(root
, rsv
,
631 item
->bytes_reserved
);
634 static int btrfs_delayed_inode_reserve_metadata(
635 struct btrfs_trans_handle
*trans
,
636 struct btrfs_root
*root
,
638 struct btrfs_delayed_node
*node
)
640 struct btrfs_block_rsv
*src_rsv
;
641 struct btrfs_block_rsv
*dst_rsv
;
644 bool release
= false;
646 src_rsv
= trans
->block_rsv
;
647 dst_rsv
= &root
->fs_info
->delayed_block_rsv
;
649 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
652 * btrfs_dirty_inode will update the inode under btrfs_join_transaction
653 * which doesn't reserve space for speed. This is a problem since we
654 * still need to reserve space for this update, so try to reserve the
657 * Now if src_rsv == delalloc_block_rsv we'll let it just steal since
658 * we're accounted for.
660 if (!src_rsv
|| (!trans
->bytes_reserved
&&
661 src_rsv
->type
!= BTRFS_BLOCK_RSV_DELALLOC
)) {
662 ret
= btrfs_block_rsv_add(root
, dst_rsv
, num_bytes
,
663 BTRFS_RESERVE_NO_FLUSH
);
665 * Since we're under a transaction reserve_metadata_bytes could
666 * try to commit the transaction which will make it return
667 * EAGAIN to make us stop the transaction we have, so return
668 * ENOSPC instead so that btrfs_dirty_inode knows what to do.
673 node
->bytes_reserved
= num_bytes
;
674 trace_btrfs_space_reservation(root
->fs_info
,
680 } else if (src_rsv
->type
== BTRFS_BLOCK_RSV_DELALLOC
) {
681 spin_lock(&BTRFS_I(inode
)->lock
);
682 if (test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
683 &BTRFS_I(inode
)->runtime_flags
)) {
684 spin_unlock(&BTRFS_I(inode
)->lock
);
688 spin_unlock(&BTRFS_I(inode
)->lock
);
690 /* Ok we didn't have space pre-reserved. This shouldn't happen
691 * too often but it can happen if we do delalloc to an existing
692 * inode which gets dirtied because of the time update, and then
693 * isn't touched again until after the transaction commits and
694 * then we try to write out the data. First try to be nice and
695 * reserve something strictly for us. If not be a pain and try
696 * to steal from the delalloc block rsv.
698 ret
= btrfs_block_rsv_add(root
, dst_rsv
, num_bytes
,
699 BTRFS_RESERVE_NO_FLUSH
);
703 ret
= btrfs_block_rsv_migrate(src_rsv
, dst_rsv
, num_bytes
);
708 * Ok this is a problem, let's just steal from the global rsv
709 * since this really shouldn't happen that often.
712 ret
= btrfs_block_rsv_migrate(&root
->fs_info
->global_block_rsv
,
718 ret
= btrfs_block_rsv_migrate(src_rsv
, dst_rsv
, num_bytes
);
722 * Migrate only takes a reservation, it doesn't touch the size of the
723 * block_rsv. This is to simplify people who don't normally have things
724 * migrated from their block rsv. If they go to release their
725 * reservation, that will decrease the size as well, so if migrate
726 * reduced size we'd end up with a negative size. But for the
727 * delalloc_meta_reserved stuff we will only know to drop 1 reservation,
728 * but we could in fact do this reserve/migrate dance several times
729 * between the time we did the original reservation and we'd clean it
730 * up. So to take care of this, release the space for the meta
731 * reservation here. I think it may be time for a documentation page on
732 * how block rsvs. work.
735 trace_btrfs_space_reservation(root
->fs_info
, "delayed_inode",
736 btrfs_ino(inode
), num_bytes
, 1);
737 node
->bytes_reserved
= num_bytes
;
741 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
742 btrfs_ino(inode
), num_bytes
, 0);
743 btrfs_block_rsv_release(root
, src_rsv
, num_bytes
);
749 static void btrfs_delayed_inode_release_metadata(struct btrfs_root
*root
,
750 struct btrfs_delayed_node
*node
)
752 struct btrfs_block_rsv
*rsv
;
754 if (!node
->bytes_reserved
)
757 rsv
= &root
->fs_info
->delayed_block_rsv
;
758 trace_btrfs_space_reservation(root
->fs_info
, "delayed_inode",
759 node
->inode_id
, node
->bytes_reserved
, 0);
760 btrfs_block_rsv_release(root
, rsv
,
761 node
->bytes_reserved
);
762 node
->bytes_reserved
= 0;
766 * This helper will insert some continuous items into the same leaf according
767 * to the free space of the leaf.
769 static int btrfs_batch_insert_items(struct btrfs_trans_handle
*trans
,
770 struct btrfs_root
*root
,
771 struct btrfs_path
*path
,
772 struct btrfs_delayed_item
*item
)
774 struct btrfs_delayed_item
*curr
, *next
;
776 int total_data_size
= 0, total_size
= 0;
777 struct extent_buffer
*leaf
;
779 struct btrfs_key
*keys
;
781 struct list_head head
;
787 BUG_ON(!path
->nodes
[0]);
789 leaf
= path
->nodes
[0];
790 free_space
= btrfs_leaf_free_space(root
, leaf
);
791 INIT_LIST_HEAD(&head
);
797 * count the number of the continuous items that we can insert in batch
799 while (total_size
+ next
->data_len
+ sizeof(struct btrfs_item
) <=
801 total_data_size
+= next
->data_len
;
802 total_size
+= next
->data_len
+ sizeof(struct btrfs_item
);
803 list_add_tail(&next
->tree_list
, &head
);
807 next
= __btrfs_next_delayed_item(curr
);
811 if (!btrfs_is_continuous_delayed_item(curr
, next
))
821 * we need allocate some memory space, but it might cause the task
822 * to sleep, so we set all locked nodes in the path to blocking locks
825 btrfs_set_path_blocking(path
);
827 keys
= kmalloc(sizeof(struct btrfs_key
) * nitems
, GFP_NOFS
);
833 data_size
= kmalloc(sizeof(u32
) * nitems
, GFP_NOFS
);
839 /* get keys of all the delayed items */
841 list_for_each_entry(next
, &head
, tree_list
) {
843 data_size
[i
] = next
->data_len
;
847 /* reset all the locked nodes in the patch to spinning locks. */
848 btrfs_clear_path_blocking(path
, NULL
, 0);
850 /* insert the keys of the items */
851 setup_items_for_insert(trans
, root
, path
, keys
, data_size
,
852 total_data_size
, total_size
, nitems
);
854 /* insert the dir index items */
855 slot
= path
->slots
[0];
856 list_for_each_entry_safe(curr
, next
, &head
, tree_list
) {
857 data_ptr
= btrfs_item_ptr(leaf
, slot
, char);
858 write_extent_buffer(leaf
, &curr
->data
,
859 (unsigned long)data_ptr
,
863 btrfs_delayed_item_release_metadata(root
, curr
);
865 list_del(&curr
->tree_list
);
866 btrfs_release_delayed_item(curr
);
877 * This helper can just do simple insertion that needn't extend item for new
878 * data, such as directory name index insertion, inode insertion.
880 static int btrfs_insert_delayed_item(struct btrfs_trans_handle
*trans
,
881 struct btrfs_root
*root
,
882 struct btrfs_path
*path
,
883 struct btrfs_delayed_item
*delayed_item
)
885 struct extent_buffer
*leaf
;
889 ret
= btrfs_insert_empty_item(trans
, root
, path
, &delayed_item
->key
,
890 delayed_item
->data_len
);
891 if (ret
< 0 && ret
!= -EEXIST
)
894 leaf
= path
->nodes
[0];
896 ptr
= btrfs_item_ptr(leaf
, path
->slots
[0], char);
898 write_extent_buffer(leaf
, delayed_item
->data
, (unsigned long)ptr
,
899 delayed_item
->data_len
);
900 btrfs_mark_buffer_dirty(leaf
);
902 btrfs_delayed_item_release_metadata(root
, delayed_item
);
907 * we insert an item first, then if there are some continuous items, we try
908 * to insert those items into the same leaf.
910 static int btrfs_insert_delayed_items(struct btrfs_trans_handle
*trans
,
911 struct btrfs_path
*path
,
912 struct btrfs_root
*root
,
913 struct btrfs_delayed_node
*node
)
915 struct btrfs_delayed_item
*curr
, *prev
;
919 mutex_lock(&node
->mutex
);
920 curr
= __btrfs_first_delayed_insertion_item(node
);
924 ret
= btrfs_insert_delayed_item(trans
, root
, path
, curr
);
926 btrfs_release_path(path
);
931 curr
= __btrfs_next_delayed_item(prev
);
932 if (curr
&& btrfs_is_continuous_delayed_item(prev
, curr
)) {
933 /* insert the continuous items into the same leaf */
935 btrfs_batch_insert_items(trans
, root
, path
, curr
);
937 btrfs_release_delayed_item(prev
);
938 btrfs_mark_buffer_dirty(path
->nodes
[0]);
940 btrfs_release_path(path
);
941 mutex_unlock(&node
->mutex
);
945 mutex_unlock(&node
->mutex
);
949 static int btrfs_batch_delete_items(struct btrfs_trans_handle
*trans
,
950 struct btrfs_root
*root
,
951 struct btrfs_path
*path
,
952 struct btrfs_delayed_item
*item
)
954 struct btrfs_delayed_item
*curr
, *next
;
955 struct extent_buffer
*leaf
;
956 struct btrfs_key key
;
957 struct list_head head
;
958 int nitems
, i
, last_item
;
961 BUG_ON(!path
->nodes
[0]);
963 leaf
= path
->nodes
[0];
966 last_item
= btrfs_header_nritems(leaf
) - 1;
968 return -ENOENT
; /* FIXME: Is errno suitable? */
971 INIT_LIST_HEAD(&head
);
972 btrfs_item_key_to_cpu(leaf
, &key
, i
);
975 * count the number of the dir index items that we can delete in batch
977 while (btrfs_comp_cpu_keys(&next
->key
, &key
) == 0) {
978 list_add_tail(&next
->tree_list
, &head
);
982 next
= __btrfs_next_delayed_item(curr
);
986 if (!btrfs_is_continuous_delayed_item(curr
, next
))
992 btrfs_item_key_to_cpu(leaf
, &key
, i
);
998 ret
= btrfs_del_items(trans
, root
, path
, path
->slots
[0], nitems
);
1002 list_for_each_entry_safe(curr
, next
, &head
, tree_list
) {
1003 btrfs_delayed_item_release_metadata(root
, curr
);
1004 list_del(&curr
->tree_list
);
1005 btrfs_release_delayed_item(curr
);
1012 static int btrfs_delete_delayed_items(struct btrfs_trans_handle
*trans
,
1013 struct btrfs_path
*path
,
1014 struct btrfs_root
*root
,
1015 struct btrfs_delayed_node
*node
)
1017 struct btrfs_delayed_item
*curr
, *prev
;
1021 mutex_lock(&node
->mutex
);
1022 curr
= __btrfs_first_delayed_deletion_item(node
);
1026 ret
= btrfs_search_slot(trans
, root
, &curr
->key
, path
, -1, 1);
1031 * can't find the item which the node points to, so this node
1032 * is invalid, just drop it.
1035 curr
= __btrfs_next_delayed_item(prev
);
1036 btrfs_release_delayed_item(prev
);
1038 btrfs_release_path(path
);
1040 mutex_unlock(&node
->mutex
);
1046 btrfs_batch_delete_items(trans
, root
, path
, curr
);
1047 btrfs_release_path(path
);
1048 mutex_unlock(&node
->mutex
);
1052 btrfs_release_path(path
);
1053 mutex_unlock(&node
->mutex
);
1057 static void btrfs_release_delayed_inode(struct btrfs_delayed_node
*delayed_node
)
1059 struct btrfs_delayed_root
*delayed_root
;
1061 if (delayed_node
&& delayed_node
->inode_dirty
) {
1062 BUG_ON(!delayed_node
->root
);
1063 delayed_node
->inode_dirty
= 0;
1064 delayed_node
->count
--;
1066 delayed_root
= delayed_node
->root
->fs_info
->delayed_root
;
1067 finish_one_item(delayed_root
);
1071 static int __btrfs_update_delayed_inode(struct btrfs_trans_handle
*trans
,
1072 struct btrfs_root
*root
,
1073 struct btrfs_path
*path
,
1074 struct btrfs_delayed_node
*node
)
1076 struct btrfs_key key
;
1077 struct btrfs_inode_item
*inode_item
;
1078 struct extent_buffer
*leaf
;
1081 key
.objectid
= node
->inode_id
;
1082 btrfs_set_key_type(&key
, BTRFS_INODE_ITEM_KEY
);
1085 ret
= btrfs_lookup_inode(trans
, root
, path
, &key
, 1);
1087 btrfs_release_path(path
);
1089 } else if (ret
< 0) {
1093 btrfs_unlock_up_safe(path
, 1);
1094 leaf
= path
->nodes
[0];
1095 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
1096 struct btrfs_inode_item
);
1097 write_extent_buffer(leaf
, &node
->inode_item
, (unsigned long)inode_item
,
1098 sizeof(struct btrfs_inode_item
));
1099 btrfs_mark_buffer_dirty(leaf
);
1100 btrfs_release_path(path
);
1102 btrfs_delayed_inode_release_metadata(root
, node
);
1103 btrfs_release_delayed_inode(node
);
1108 static inline int btrfs_update_delayed_inode(struct btrfs_trans_handle
*trans
,
1109 struct btrfs_root
*root
,
1110 struct btrfs_path
*path
,
1111 struct btrfs_delayed_node
*node
)
1115 mutex_lock(&node
->mutex
);
1116 if (!node
->inode_dirty
) {
1117 mutex_unlock(&node
->mutex
);
1121 ret
= __btrfs_update_delayed_inode(trans
, root
, path
, node
);
1122 mutex_unlock(&node
->mutex
);
1127 __btrfs_commit_inode_delayed_items(struct btrfs_trans_handle
*trans
,
1128 struct btrfs_path
*path
,
1129 struct btrfs_delayed_node
*node
)
1133 ret
= btrfs_insert_delayed_items(trans
, path
, node
->root
, node
);
1137 ret
= btrfs_delete_delayed_items(trans
, path
, node
->root
, node
);
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
,
1152 struct btrfs_root
*root
, int nr
)
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);
1164 path
= btrfs_alloc_path();
1167 path
->leave_spinning
= 1;
1169 block_rsv
= trans
->block_rsv
;
1170 trans
->block_rsv
= &root
->fs_info
->delayed_block_rsv
;
1172 delayed_root
= btrfs_get_delayed_root(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_release_delayed_node(curr_node
);
1181 btrfs_abort_transaction(trans
, root
, ret
);
1185 prev_node
= curr_node
;
1186 curr_node
= btrfs_next_delayed_node(curr_node
);
1187 btrfs_release_delayed_node(prev_node
);
1191 btrfs_release_delayed_node(curr_node
);
1192 btrfs_free_path(path
);
1193 trans
->block_rsv
= block_rsv
;
1198 int btrfs_run_delayed_items(struct btrfs_trans_handle
*trans
,
1199 struct btrfs_root
*root
)
1201 return __btrfs_run_delayed_items(trans
, root
, -1);
1204 int btrfs_run_delayed_items_nr(struct btrfs_trans_handle
*trans
,
1205 struct btrfs_root
*root
, int nr
)
1207 return __btrfs_run_delayed_items(trans
, root
, nr
);
1210 int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle
*trans
,
1211 struct inode
*inode
)
1213 struct btrfs_delayed_node
*delayed_node
= btrfs_get_delayed_node(inode
);
1214 struct btrfs_path
*path
;
1215 struct btrfs_block_rsv
*block_rsv
;
1221 mutex_lock(&delayed_node
->mutex
);
1222 if (!delayed_node
->count
) {
1223 mutex_unlock(&delayed_node
->mutex
);
1224 btrfs_release_delayed_node(delayed_node
);
1227 mutex_unlock(&delayed_node
->mutex
);
1229 path
= btrfs_alloc_path();
1232 path
->leave_spinning
= 1;
1234 block_rsv
= trans
->block_rsv
;
1235 trans
->block_rsv
= &delayed_node
->root
->fs_info
->delayed_block_rsv
;
1237 ret
= __btrfs_commit_inode_delayed_items(trans
, path
, delayed_node
);
1239 btrfs_release_delayed_node(delayed_node
);
1240 btrfs_free_path(path
);
1241 trans
->block_rsv
= block_rsv
;
1246 int btrfs_commit_inode_delayed_inode(struct inode
*inode
)
1248 struct btrfs_trans_handle
*trans
;
1249 struct btrfs_delayed_node
*delayed_node
= btrfs_get_delayed_node(inode
);
1250 struct btrfs_path
*path
;
1251 struct btrfs_block_rsv
*block_rsv
;
1257 mutex_lock(&delayed_node
->mutex
);
1258 if (!delayed_node
->inode_dirty
) {
1259 mutex_unlock(&delayed_node
->mutex
);
1260 btrfs_release_delayed_node(delayed_node
);
1263 mutex_unlock(&delayed_node
->mutex
);
1265 trans
= btrfs_join_transaction(delayed_node
->root
);
1266 if (IS_ERR(trans
)) {
1267 ret
= PTR_ERR(trans
);
1271 path
= btrfs_alloc_path();
1276 path
->leave_spinning
= 1;
1278 block_rsv
= trans
->block_rsv
;
1279 trans
->block_rsv
= &delayed_node
->root
->fs_info
->delayed_block_rsv
;
1281 mutex_lock(&delayed_node
->mutex
);
1282 if (delayed_node
->inode_dirty
)
1283 ret
= __btrfs_update_delayed_inode(trans
, delayed_node
->root
,
1284 path
, delayed_node
);
1287 mutex_unlock(&delayed_node
->mutex
);
1289 btrfs_free_path(path
);
1290 trans
->block_rsv
= block_rsv
;
1292 btrfs_end_transaction(trans
, delayed_node
->root
);
1293 btrfs_btree_balance_dirty(delayed_node
->root
);
1295 btrfs_release_delayed_node(delayed_node
);
1300 void btrfs_remove_delayed_node(struct inode
*inode
)
1302 struct btrfs_delayed_node
*delayed_node
;
1304 delayed_node
= ACCESS_ONCE(BTRFS_I(inode
)->delayed_node
);
1308 BTRFS_I(inode
)->delayed_node
= NULL
;
1309 btrfs_release_delayed_node(delayed_node
);
1312 struct btrfs_async_delayed_work
{
1313 struct btrfs_delayed_root
*delayed_root
;
1315 struct btrfs_work work
;
1318 static void btrfs_async_run_delayed_root(struct btrfs_work
*work
)
1320 struct btrfs_async_delayed_work
*async_work
;
1321 struct btrfs_delayed_root
*delayed_root
;
1322 struct btrfs_trans_handle
*trans
;
1323 struct btrfs_path
*path
;
1324 struct btrfs_delayed_node
*delayed_node
= NULL
;
1325 struct btrfs_root
*root
;
1326 struct btrfs_block_rsv
*block_rsv
;
1329 async_work
= container_of(work
, struct btrfs_async_delayed_work
, work
);
1330 delayed_root
= async_work
->delayed_root
;
1332 path
= btrfs_alloc_path();
1337 if (atomic_read(&delayed_root
->items
) < BTRFS_DELAYED_BACKGROUND
/ 2)
1340 delayed_node
= btrfs_first_prepared_delayed_node(delayed_root
);
1344 path
->leave_spinning
= 1;
1345 root
= delayed_node
->root
;
1347 trans
= btrfs_join_transaction(root
);
1351 block_rsv
= trans
->block_rsv
;
1352 trans
->block_rsv
= &root
->fs_info
->delayed_block_rsv
;
1354 __btrfs_commit_inode_delayed_items(trans
, path
, delayed_node
);
1356 * Maybe new delayed items have been inserted, so we need requeue
1357 * the work. Besides that, we must dequeue the empty delayed nodes
1358 * to avoid the race between delayed items balance and the worker.
1359 * The race like this:
1360 * Task1 Worker thread
1361 * count == 0, needn't requeue
1362 * also needn't insert the
1363 * delayed node into prepare
1365 * add lots of delayed items
1366 * queue the delayed node
1367 * already in the list,
1368 * and not in the prepare
1369 * list, it means the delayed
1370 * node is being dealt with
1372 * do delayed items balance
1373 * the delayed node is being
1374 * dealt with by the worker
1376 * the worker goto idle.
1377 * Task1 will sleep until the transaction is commited.
1379 mutex_lock(&delayed_node
->mutex
);
1380 btrfs_dequeue_delayed_node(root
->fs_info
->delayed_root
, delayed_node
);
1381 mutex_unlock(&delayed_node
->mutex
);
1383 trans
->block_rsv
= block_rsv
;
1384 btrfs_end_transaction_dmeta(trans
, root
);
1385 btrfs_btree_balance_dirty_nodelay(root
);
1388 btrfs_release_path(path
);
1391 btrfs_release_prepared_delayed_node(delayed_node
);
1392 if (async_work
->nr
== 0 || total_done
< async_work
->nr
)
1396 btrfs_free_path(path
);
1398 wake_up(&delayed_root
->wait
);
1403 static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root
*delayed_root
,
1404 struct btrfs_root
*root
, int nr
)
1406 struct btrfs_async_delayed_work
*async_work
;
1408 if (atomic_read(&delayed_root
->items
) < BTRFS_DELAYED_BACKGROUND
)
1411 async_work
= kmalloc(sizeof(*async_work
), GFP_NOFS
);
1415 async_work
->delayed_root
= delayed_root
;
1416 async_work
->work
.func
= btrfs_async_run_delayed_root
;
1417 async_work
->work
.flags
= 0;
1418 async_work
->nr
= nr
;
1420 btrfs_queue_worker(&root
->fs_info
->delayed_workers
, &async_work
->work
);
1424 void btrfs_assert_delayed_root_empty(struct btrfs_root
*root
)
1426 struct btrfs_delayed_root
*delayed_root
;
1427 delayed_root
= btrfs_get_delayed_root(root
);
1428 WARN_ON(btrfs_first_delayed_node(delayed_root
));
1431 static int refs_newer(struct btrfs_delayed_root
*delayed_root
,
1434 int val
= atomic_read(&delayed_root
->items_seq
);
1436 if (val
< seq
|| val
>= seq
+ count
)
1441 void btrfs_balance_delayed_items(struct btrfs_root
*root
)
1443 struct btrfs_delayed_root
*delayed_root
;
1446 delayed_root
= btrfs_get_delayed_root(root
);
1448 if (atomic_read(&delayed_root
->items
) < BTRFS_DELAYED_BACKGROUND
)
1451 seq
= atomic_read(&delayed_root
->items_seq
);
1453 if (atomic_read(&delayed_root
->items
) >= BTRFS_DELAYED_WRITEBACK
) {
1455 DEFINE_WAIT(__wait
);
1457 ret
= btrfs_wq_run_delayed_node(delayed_root
, root
, 0);
1462 prepare_to_wait(&delayed_root
->wait
, &__wait
,
1463 TASK_INTERRUPTIBLE
);
1465 if (refs_newer(delayed_root
, seq
,
1466 BTRFS_DELAYED_BATCH
) ||
1467 atomic_read(&delayed_root
->items
) <
1468 BTRFS_DELAYED_BACKGROUND
) {
1471 if (!signal_pending(current
))
1476 finish_wait(&delayed_root
->wait
, &__wait
);
1479 btrfs_wq_run_delayed_node(delayed_root
, root
, BTRFS_DELAYED_BATCH
);
1482 /* Will return 0 or -ENOMEM */
1483 int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle
*trans
,
1484 struct btrfs_root
*root
, const char *name
,
1485 int name_len
, struct inode
*dir
,
1486 struct btrfs_disk_key
*disk_key
, u8 type
,
1489 struct btrfs_delayed_node
*delayed_node
;
1490 struct btrfs_delayed_item
*delayed_item
;
1491 struct btrfs_dir_item
*dir_item
;
1494 delayed_node
= btrfs_get_or_create_delayed_node(dir
);
1495 if (IS_ERR(delayed_node
))
1496 return PTR_ERR(delayed_node
);
1498 delayed_item
= btrfs_alloc_delayed_item(sizeof(*dir_item
) + name_len
);
1499 if (!delayed_item
) {
1504 delayed_item
->key
.objectid
= btrfs_ino(dir
);
1505 btrfs_set_key_type(&delayed_item
->key
, BTRFS_DIR_INDEX_KEY
);
1506 delayed_item
->key
.offset
= index
;
1508 dir_item
= (struct btrfs_dir_item
*)delayed_item
->data
;
1509 dir_item
->location
= *disk_key
;
1510 dir_item
->transid
= cpu_to_le64(trans
->transid
);
1511 dir_item
->data_len
= 0;
1512 dir_item
->name_len
= cpu_to_le16(name_len
);
1513 dir_item
->type
= type
;
1514 memcpy((char *)(dir_item
+ 1), name
, name_len
);
1516 ret
= btrfs_delayed_item_reserve_metadata(trans
, root
, delayed_item
);
1518 * we have reserved enough space when we start a new transaction,
1519 * so reserving metadata failure is impossible
1524 mutex_lock(&delayed_node
->mutex
);
1525 ret
= __btrfs_add_delayed_insertion_item(delayed_node
, delayed_item
);
1526 if (unlikely(ret
)) {
1527 printk(KERN_ERR
"err add delayed dir index item(name: %s) into "
1528 "the insertion tree of the delayed node"
1529 "(root id: %llu, inode id: %llu, errno: %d)\n",
1531 (unsigned long long)delayed_node
->root
->objectid
,
1532 (unsigned long long)delayed_node
->inode_id
,
1536 mutex_unlock(&delayed_node
->mutex
);
1539 btrfs_release_delayed_node(delayed_node
);
1543 static int btrfs_delete_delayed_insertion_item(struct btrfs_root
*root
,
1544 struct btrfs_delayed_node
*node
,
1545 struct btrfs_key
*key
)
1547 struct btrfs_delayed_item
*item
;
1549 mutex_lock(&node
->mutex
);
1550 item
= __btrfs_lookup_delayed_insertion_item(node
, key
);
1552 mutex_unlock(&node
->mutex
);
1556 btrfs_delayed_item_release_metadata(root
, item
);
1557 btrfs_release_delayed_item(item
);
1558 mutex_unlock(&node
->mutex
);
1562 int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle
*trans
,
1563 struct btrfs_root
*root
, struct inode
*dir
,
1566 struct btrfs_delayed_node
*node
;
1567 struct btrfs_delayed_item
*item
;
1568 struct btrfs_key item_key
;
1571 node
= btrfs_get_or_create_delayed_node(dir
);
1573 return PTR_ERR(node
);
1575 item_key
.objectid
= btrfs_ino(dir
);
1576 btrfs_set_key_type(&item_key
, BTRFS_DIR_INDEX_KEY
);
1577 item_key
.offset
= index
;
1579 ret
= btrfs_delete_delayed_insertion_item(root
, node
, &item_key
);
1583 item
= btrfs_alloc_delayed_item(0);
1589 item
->key
= item_key
;
1591 ret
= btrfs_delayed_item_reserve_metadata(trans
, root
, item
);
1593 * we have reserved enough space when we start a new transaction,
1594 * so reserving metadata failure is impossible.
1598 mutex_lock(&node
->mutex
);
1599 ret
= __btrfs_add_delayed_deletion_item(node
, item
);
1600 if (unlikely(ret
)) {
1601 printk(KERN_ERR
"err add delayed dir index item(index: %llu) "
1602 "into the deletion tree of the delayed node"
1603 "(root id: %llu, inode id: %llu, errno: %d)\n",
1604 (unsigned long long)index
,
1605 (unsigned long long)node
->root
->objectid
,
1606 (unsigned long long)node
->inode_id
,
1610 mutex_unlock(&node
->mutex
);
1612 btrfs_release_delayed_node(node
);
1616 int btrfs_inode_delayed_dir_index_count(struct inode
*inode
)
1618 struct btrfs_delayed_node
*delayed_node
= btrfs_get_delayed_node(inode
);
1624 * Since we have held i_mutex of this directory, it is impossible that
1625 * a new directory index is added into the delayed node and index_cnt
1626 * is updated now. So we needn't lock the delayed node.
1628 if (!delayed_node
->index_cnt
) {
1629 btrfs_release_delayed_node(delayed_node
);
1633 BTRFS_I(inode
)->index_cnt
= delayed_node
->index_cnt
;
1634 btrfs_release_delayed_node(delayed_node
);
1638 void btrfs_get_delayed_items(struct inode
*inode
, struct list_head
*ins_list
,
1639 struct list_head
*del_list
)
1641 struct btrfs_delayed_node
*delayed_node
;
1642 struct btrfs_delayed_item
*item
;
1644 delayed_node
= btrfs_get_delayed_node(inode
);
1648 mutex_lock(&delayed_node
->mutex
);
1649 item
= __btrfs_first_delayed_insertion_item(delayed_node
);
1651 atomic_inc(&item
->refs
);
1652 list_add_tail(&item
->readdir_list
, ins_list
);
1653 item
= __btrfs_next_delayed_item(item
);
1656 item
= __btrfs_first_delayed_deletion_item(delayed_node
);
1658 atomic_inc(&item
->refs
);
1659 list_add_tail(&item
->readdir_list
, del_list
);
1660 item
= __btrfs_next_delayed_item(item
);
1662 mutex_unlock(&delayed_node
->mutex
);
1664 * This delayed node is still cached in the btrfs inode, so refs
1665 * must be > 1 now, and we needn't check it is going to be freed
1668 * Besides that, this function is used to read dir, we do not
1669 * insert/delete delayed items in this period. So we also needn't
1670 * requeue or dequeue this delayed node.
1672 atomic_dec(&delayed_node
->refs
);
1675 void btrfs_put_delayed_items(struct list_head
*ins_list
,
1676 struct list_head
*del_list
)
1678 struct btrfs_delayed_item
*curr
, *next
;
1680 list_for_each_entry_safe(curr
, next
, ins_list
, readdir_list
) {
1681 list_del(&curr
->readdir_list
);
1682 if (atomic_dec_and_test(&curr
->refs
))
1686 list_for_each_entry_safe(curr
, next
, del_list
, readdir_list
) {
1687 list_del(&curr
->readdir_list
);
1688 if (atomic_dec_and_test(&curr
->refs
))
1693 int btrfs_should_delete_dir_index(struct list_head
*del_list
,
1696 struct btrfs_delayed_item
*curr
, *next
;
1699 if (list_empty(del_list
))
1702 list_for_each_entry_safe(curr
, next
, del_list
, readdir_list
) {
1703 if (curr
->key
.offset
> index
)
1706 list_del(&curr
->readdir_list
);
1707 ret
= (curr
->key
.offset
== index
);
1709 if (atomic_dec_and_test(&curr
->refs
))
1721 * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree
1724 int btrfs_readdir_delayed_dir_index(struct file
*filp
, void *dirent
,
1726 struct list_head
*ins_list
)
1728 struct btrfs_dir_item
*di
;
1729 struct btrfs_delayed_item
*curr
, *next
;
1730 struct btrfs_key location
;
1734 unsigned char d_type
;
1736 if (list_empty(ins_list
))
1740 * Changing the data of the delayed item is impossible. So
1741 * we needn't lock them. And we have held i_mutex of the
1742 * directory, nobody can delete any directory indexes now.
1744 list_for_each_entry_safe(curr
, next
, ins_list
, readdir_list
) {
1745 list_del(&curr
->readdir_list
);
1747 if (curr
->key
.offset
< filp
->f_pos
) {
1748 if (atomic_dec_and_test(&curr
->refs
))
1753 filp
->f_pos
= curr
->key
.offset
;
1755 di
= (struct btrfs_dir_item
*)curr
->data
;
1756 name
= (char *)(di
+ 1);
1757 name_len
= le16_to_cpu(di
->name_len
);
1759 d_type
= btrfs_filetype_table
[di
->type
];
1760 btrfs_disk_key_to_cpu(&location
, &di
->location
);
1762 over
= filldir(dirent
, name
, name_len
, curr
->key
.offset
,
1763 location
.objectid
, d_type
);
1765 if (atomic_dec_and_test(&curr
->refs
))
1774 BTRFS_SETGET_STACK_FUNCS(stack_inode_generation
, struct btrfs_inode_item
,
1776 BTRFS_SETGET_STACK_FUNCS(stack_inode_sequence
, struct btrfs_inode_item
,
1778 BTRFS_SETGET_STACK_FUNCS(stack_inode_transid
, struct btrfs_inode_item
,
1780 BTRFS_SETGET_STACK_FUNCS(stack_inode_size
, struct btrfs_inode_item
, size
, 64);
1781 BTRFS_SETGET_STACK_FUNCS(stack_inode_nbytes
, struct btrfs_inode_item
,
1783 BTRFS_SETGET_STACK_FUNCS(stack_inode_block_group
, struct btrfs_inode_item
,
1785 BTRFS_SETGET_STACK_FUNCS(stack_inode_nlink
, struct btrfs_inode_item
, nlink
, 32);
1786 BTRFS_SETGET_STACK_FUNCS(stack_inode_uid
, struct btrfs_inode_item
, uid
, 32);
1787 BTRFS_SETGET_STACK_FUNCS(stack_inode_gid
, struct btrfs_inode_item
, gid
, 32);
1788 BTRFS_SETGET_STACK_FUNCS(stack_inode_mode
, struct btrfs_inode_item
, mode
, 32);
1789 BTRFS_SETGET_STACK_FUNCS(stack_inode_rdev
, struct btrfs_inode_item
, rdev
, 64);
1790 BTRFS_SETGET_STACK_FUNCS(stack_inode_flags
, struct btrfs_inode_item
, flags
, 64);
1792 BTRFS_SETGET_STACK_FUNCS(stack_timespec_sec
, struct btrfs_timespec
, sec
, 64);
1793 BTRFS_SETGET_STACK_FUNCS(stack_timespec_nsec
, struct btrfs_timespec
, nsec
, 32);
1795 static void fill_stack_inode_item(struct btrfs_trans_handle
*trans
,
1796 struct btrfs_inode_item
*inode_item
,
1797 struct inode
*inode
)
1799 btrfs_set_stack_inode_uid(inode_item
, i_uid_read(inode
));
1800 btrfs_set_stack_inode_gid(inode_item
, i_gid_read(inode
));
1801 btrfs_set_stack_inode_size(inode_item
, BTRFS_I(inode
)->disk_i_size
);
1802 btrfs_set_stack_inode_mode(inode_item
, inode
->i_mode
);
1803 btrfs_set_stack_inode_nlink(inode_item
, inode
->i_nlink
);
1804 btrfs_set_stack_inode_nbytes(inode_item
, inode_get_bytes(inode
));
1805 btrfs_set_stack_inode_generation(inode_item
,
1806 BTRFS_I(inode
)->generation
);
1807 btrfs_set_stack_inode_sequence(inode_item
, inode
->i_version
);
1808 btrfs_set_stack_inode_transid(inode_item
, trans
->transid
);
1809 btrfs_set_stack_inode_rdev(inode_item
, inode
->i_rdev
);
1810 btrfs_set_stack_inode_flags(inode_item
, BTRFS_I(inode
)->flags
);
1811 btrfs_set_stack_inode_block_group(inode_item
, 0);
1813 btrfs_set_stack_timespec_sec(btrfs_inode_atime(inode_item
),
1814 inode
->i_atime
.tv_sec
);
1815 btrfs_set_stack_timespec_nsec(btrfs_inode_atime(inode_item
),
1816 inode
->i_atime
.tv_nsec
);
1818 btrfs_set_stack_timespec_sec(btrfs_inode_mtime(inode_item
),
1819 inode
->i_mtime
.tv_sec
);
1820 btrfs_set_stack_timespec_nsec(btrfs_inode_mtime(inode_item
),
1821 inode
->i_mtime
.tv_nsec
);
1823 btrfs_set_stack_timespec_sec(btrfs_inode_ctime(inode_item
),
1824 inode
->i_ctime
.tv_sec
);
1825 btrfs_set_stack_timespec_nsec(btrfs_inode_ctime(inode_item
),
1826 inode
->i_ctime
.tv_nsec
);
1829 int btrfs_fill_inode(struct inode
*inode
, u32
*rdev
)
1831 struct btrfs_delayed_node
*delayed_node
;
1832 struct btrfs_inode_item
*inode_item
;
1833 struct btrfs_timespec
*tspec
;
1835 delayed_node
= btrfs_get_delayed_node(inode
);
1839 mutex_lock(&delayed_node
->mutex
);
1840 if (!delayed_node
->inode_dirty
) {
1841 mutex_unlock(&delayed_node
->mutex
);
1842 btrfs_release_delayed_node(delayed_node
);
1846 inode_item
= &delayed_node
->inode_item
;
1848 i_uid_write(inode
, btrfs_stack_inode_uid(inode_item
));
1849 i_gid_write(inode
, btrfs_stack_inode_gid(inode_item
));
1850 btrfs_i_size_write(inode
, btrfs_stack_inode_size(inode_item
));
1851 inode
->i_mode
= btrfs_stack_inode_mode(inode_item
);
1852 set_nlink(inode
, btrfs_stack_inode_nlink(inode_item
));
1853 inode_set_bytes(inode
, btrfs_stack_inode_nbytes(inode_item
));
1854 BTRFS_I(inode
)->generation
= btrfs_stack_inode_generation(inode_item
);
1855 inode
->i_version
= btrfs_stack_inode_sequence(inode_item
);
1857 *rdev
= btrfs_stack_inode_rdev(inode_item
);
1858 BTRFS_I(inode
)->flags
= btrfs_stack_inode_flags(inode_item
);
1860 tspec
= btrfs_inode_atime(inode_item
);
1861 inode
->i_atime
.tv_sec
= btrfs_stack_timespec_sec(tspec
);
1862 inode
->i_atime
.tv_nsec
= btrfs_stack_timespec_nsec(tspec
);
1864 tspec
= btrfs_inode_mtime(inode_item
);
1865 inode
->i_mtime
.tv_sec
= btrfs_stack_timespec_sec(tspec
);
1866 inode
->i_mtime
.tv_nsec
= btrfs_stack_timespec_nsec(tspec
);
1868 tspec
= btrfs_inode_ctime(inode_item
);
1869 inode
->i_ctime
.tv_sec
= btrfs_stack_timespec_sec(tspec
);
1870 inode
->i_ctime
.tv_nsec
= btrfs_stack_timespec_nsec(tspec
);
1872 inode
->i_generation
= BTRFS_I(inode
)->generation
;
1873 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
1875 mutex_unlock(&delayed_node
->mutex
);
1876 btrfs_release_delayed_node(delayed_node
);
1880 int btrfs_delayed_update_inode(struct btrfs_trans_handle
*trans
,
1881 struct btrfs_root
*root
, struct inode
*inode
)
1883 struct btrfs_delayed_node
*delayed_node
;
1886 delayed_node
= btrfs_get_or_create_delayed_node(inode
);
1887 if (IS_ERR(delayed_node
))
1888 return PTR_ERR(delayed_node
);
1890 mutex_lock(&delayed_node
->mutex
);
1891 if (delayed_node
->inode_dirty
) {
1892 fill_stack_inode_item(trans
, &delayed_node
->inode_item
, inode
);
1896 ret
= btrfs_delayed_inode_reserve_metadata(trans
, root
, inode
,
1901 fill_stack_inode_item(trans
, &delayed_node
->inode_item
, inode
);
1902 delayed_node
->inode_dirty
= 1;
1903 delayed_node
->count
++;
1904 atomic_inc(&root
->fs_info
->delayed_root
->items
);
1906 mutex_unlock(&delayed_node
->mutex
);
1907 btrfs_release_delayed_node(delayed_node
);
1911 static void __btrfs_kill_delayed_node(struct btrfs_delayed_node
*delayed_node
)
1913 struct btrfs_root
*root
= delayed_node
->root
;
1914 struct btrfs_delayed_item
*curr_item
, *prev_item
;
1916 mutex_lock(&delayed_node
->mutex
);
1917 curr_item
= __btrfs_first_delayed_insertion_item(delayed_node
);
1919 btrfs_delayed_item_release_metadata(root
, curr_item
);
1920 prev_item
= curr_item
;
1921 curr_item
= __btrfs_next_delayed_item(prev_item
);
1922 btrfs_release_delayed_item(prev_item
);
1925 curr_item
= __btrfs_first_delayed_deletion_item(delayed_node
);
1927 btrfs_delayed_item_release_metadata(root
, curr_item
);
1928 prev_item
= curr_item
;
1929 curr_item
= __btrfs_next_delayed_item(prev_item
);
1930 btrfs_release_delayed_item(prev_item
);
1933 if (delayed_node
->inode_dirty
) {
1934 btrfs_delayed_inode_release_metadata(root
, delayed_node
);
1935 btrfs_release_delayed_inode(delayed_node
);
1937 mutex_unlock(&delayed_node
->mutex
);
1940 void btrfs_kill_delayed_inode_items(struct inode
*inode
)
1942 struct btrfs_delayed_node
*delayed_node
;
1944 delayed_node
= btrfs_get_delayed_node(inode
);
1948 __btrfs_kill_delayed_node(delayed_node
);
1949 btrfs_release_delayed_node(delayed_node
);
1952 void btrfs_kill_all_delayed_nodes(struct btrfs_root
*root
)
1955 struct btrfs_delayed_node
*delayed_nodes
[8];
1959 spin_lock(&root
->inode_lock
);
1960 n
= radix_tree_gang_lookup(&root
->delayed_nodes_tree
,
1961 (void **)delayed_nodes
, inode_id
,
1962 ARRAY_SIZE(delayed_nodes
));
1964 spin_unlock(&root
->inode_lock
);
1968 inode_id
= delayed_nodes
[n
- 1]->inode_id
+ 1;
1970 for (i
= 0; i
< n
; i
++)
1971 atomic_inc(&delayed_nodes
[i
]->refs
);
1972 spin_unlock(&root
->inode_lock
);
1974 for (i
= 0; i
< n
; i
++) {
1975 __btrfs_kill_delayed_node(delayed_nodes
[i
]);
1976 btrfs_release_delayed_node(delayed_nodes
[i
]);
1981 void btrfs_destroy_delayed_inodes(struct btrfs_root
*root
)
1983 struct btrfs_delayed_root
*delayed_root
;
1984 struct btrfs_delayed_node
*curr_node
, *prev_node
;
1986 delayed_root
= btrfs_get_delayed_root(root
);
1988 curr_node
= btrfs_first_delayed_node(delayed_root
);
1990 __btrfs_kill_delayed_node(curr_node
);
1992 prev_node
= curr_node
;
1993 curr_node
= btrfs_next_delayed_node(curr_node
);
1994 btrfs_release_delayed_node(prev_node
);