2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
26 #include "print-tree.h"
27 #include "transaction.h"
30 #include "ref-cache.h"
32 #define PENDING_EXTENT_INSERT 0
33 #define PENDING_EXTENT_DELETE 1
34 #define PENDING_BACKREF_UPDATE 2
36 struct pending_extent_op
{
47 static int finish_current_insert(struct btrfs_trans_handle
*trans
, struct
48 btrfs_root
*extent_root
);
49 static int del_pending_extents(struct btrfs_trans_handle
*trans
, struct
50 btrfs_root
*extent_root
);
51 static struct btrfs_block_group_cache
*
52 __btrfs_find_block_group(struct btrfs_root
*root
,
53 struct btrfs_block_group_cache
*hint
,
54 u64 search_start
, int data
, int owner
);
56 static int block_group_bits(struct btrfs_block_group_cache
*cache
, u64 bits
)
58 return (cache
->flags
& bits
) == bits
;
62 * this adds the block group to the fs_info rb tree for the block group
65 int btrfs_add_block_group_cache(struct btrfs_fs_info
*info
,
66 struct btrfs_block_group_cache
*block_group
)
69 struct rb_node
*parent
= NULL
;
70 struct btrfs_block_group_cache
*cache
;
72 spin_lock(&info
->block_group_cache_lock
);
73 p
= &info
->block_group_cache_tree
.rb_node
;
77 cache
= rb_entry(parent
, struct btrfs_block_group_cache
,
79 if (block_group
->key
.objectid
< cache
->key
.objectid
) {
81 } else if (block_group
->key
.objectid
> cache
->key
.objectid
) {
84 spin_unlock(&info
->block_group_cache_lock
);
89 rb_link_node(&block_group
->cache_node
, parent
, p
);
90 rb_insert_color(&block_group
->cache_node
,
91 &info
->block_group_cache_tree
);
92 spin_unlock(&info
->block_group_cache_lock
);
98 * This will return the block group at or after bytenr if contains is 0, else
99 * it will return the block group that contains the bytenr
101 static struct btrfs_block_group_cache
*
102 block_group_cache_tree_search(struct btrfs_fs_info
*info
, u64 bytenr
,
105 struct btrfs_block_group_cache
*cache
, *ret
= NULL
;
109 spin_lock(&info
->block_group_cache_lock
);
110 n
= info
->block_group_cache_tree
.rb_node
;
113 cache
= rb_entry(n
, struct btrfs_block_group_cache
,
115 end
= cache
->key
.objectid
+ cache
->key
.offset
- 1;
116 start
= cache
->key
.objectid
;
118 if (bytenr
< start
) {
119 if (!contains
&& (!ret
|| start
< ret
->key
.objectid
))
122 } else if (bytenr
> start
) {
123 if (contains
&& bytenr
<= end
) {
133 spin_unlock(&info
->block_group_cache_lock
);
139 * this is only called by cache_block_group, since we could have freed extents
140 * we need to check the pinned_extents for any extents that can't be used yet
141 * since their free space will be released as soon as the transaction commits.
143 static int add_new_free_space(struct btrfs_block_group_cache
*block_group
,
144 struct btrfs_fs_info
*info
, u64 start
, u64 end
)
146 u64 extent_start
, extent_end
, size
;
149 mutex_lock(&info
->pinned_mutex
);
150 while (start
< end
) {
151 ret
= find_first_extent_bit(&info
->pinned_extents
, start
,
152 &extent_start
, &extent_end
,
157 if (extent_start
== start
) {
158 start
= extent_end
+ 1;
159 } else if (extent_start
> start
&& extent_start
< end
) {
160 size
= extent_start
- start
;
161 ret
= btrfs_add_free_space_lock(block_group
, start
,
164 start
= extent_end
+ 1;
172 ret
= btrfs_add_free_space_lock(block_group
, start
, size
);
175 mutex_unlock(&info
->pinned_mutex
);
180 static int cache_block_group(struct btrfs_root
*root
,
181 struct btrfs_block_group_cache
*block_group
)
183 struct btrfs_path
*path
;
185 struct btrfs_key key
;
186 struct extent_buffer
*leaf
;
195 root
= root
->fs_info
->extent_root
;
197 if (block_group
->cached
)
200 path
= btrfs_alloc_path();
206 * we get into deadlocks with paths held by callers of this function.
207 * since the alloc_mutex is protecting things right now, just
208 * skip the locking here
210 path
->skip_locking
= 1;
211 first_free
= max_t(u64
, block_group
->key
.objectid
,
212 BTRFS_SUPER_INFO_OFFSET
+ BTRFS_SUPER_INFO_SIZE
);
213 key
.objectid
= block_group
->key
.objectid
;
215 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
216 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
219 ret
= btrfs_previous_item(root
, path
, 0, BTRFS_EXTENT_ITEM_KEY
);
223 leaf
= path
->nodes
[0];
224 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
225 if (key
.objectid
+ key
.offset
> first_free
)
226 first_free
= key
.objectid
+ key
.offset
;
229 leaf
= path
->nodes
[0];
230 slot
= path
->slots
[0];
231 if (slot
>= btrfs_header_nritems(leaf
)) {
232 ret
= btrfs_next_leaf(root
, path
);
240 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
241 if (key
.objectid
< block_group
->key
.objectid
)
244 if (key
.objectid
>= block_group
->key
.objectid
+
245 block_group
->key
.offset
)
248 if (btrfs_key_type(&key
) == BTRFS_EXTENT_ITEM_KEY
) {
254 add_new_free_space(block_group
, root
->fs_info
, last
,
257 last
= key
.objectid
+ key
.offset
;
266 add_new_free_space(block_group
, root
->fs_info
, last
,
267 block_group
->key
.objectid
+
268 block_group
->key
.offset
);
270 block_group
->cached
= 1;
273 btrfs_free_path(path
);
278 * return the block group that starts at or after bytenr
280 struct btrfs_block_group_cache
*btrfs_lookup_first_block_group(struct
284 struct btrfs_block_group_cache
*cache
;
286 cache
= block_group_cache_tree_search(info
, bytenr
, 0);
292 * return the block group that contains teh given bytenr
294 struct btrfs_block_group_cache
*btrfs_lookup_block_group(struct
298 struct btrfs_block_group_cache
*cache
;
300 cache
= block_group_cache_tree_search(info
, bytenr
, 1);
305 static struct btrfs_space_info
*__find_space_info(struct btrfs_fs_info
*info
,
308 struct list_head
*head
= &info
->space_info
;
309 struct list_head
*cur
;
310 struct btrfs_space_info
*found
;
311 list_for_each(cur
, head
) {
312 found
= list_entry(cur
, struct btrfs_space_info
, list
);
313 if (found
->flags
== flags
)
319 static u64
div_factor(u64 num
, int factor
)
328 static struct btrfs_block_group_cache
*
329 __btrfs_find_block_group(struct btrfs_root
*root
,
330 struct btrfs_block_group_cache
*hint
,
331 u64 search_start
, int data
, int owner
)
333 struct btrfs_block_group_cache
*cache
;
334 struct btrfs_block_group_cache
*found_group
= NULL
;
335 struct btrfs_fs_info
*info
= root
->fs_info
;
343 if (data
& BTRFS_BLOCK_GROUP_METADATA
)
347 struct btrfs_block_group_cache
*shint
;
348 shint
= btrfs_lookup_first_block_group(info
, search_start
);
349 if (shint
&& block_group_bits(shint
, data
) && !shint
->ro
) {
350 spin_lock(&shint
->lock
);
351 used
= btrfs_block_group_used(&shint
->item
);
352 if (used
+ shint
->pinned
+ shint
->reserved
<
353 div_factor(shint
->key
.offset
, factor
)) {
354 spin_unlock(&shint
->lock
);
357 spin_unlock(&shint
->lock
);
360 if (hint
&& !hint
->ro
&& block_group_bits(hint
, data
)) {
361 spin_lock(&hint
->lock
);
362 used
= btrfs_block_group_used(&hint
->item
);
363 if (used
+ hint
->pinned
+ hint
->reserved
<
364 div_factor(hint
->key
.offset
, factor
)) {
365 spin_unlock(&hint
->lock
);
368 spin_unlock(&hint
->lock
);
369 last
= hint
->key
.objectid
+ hint
->key
.offset
;
372 last
= max(hint
->key
.objectid
, search_start
);
378 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
382 spin_lock(&cache
->lock
);
383 last
= cache
->key
.objectid
+ cache
->key
.offset
;
384 used
= btrfs_block_group_used(&cache
->item
);
386 if (!cache
->ro
&& block_group_bits(cache
, data
)) {
387 free_check
= div_factor(cache
->key
.offset
, factor
);
388 if (used
+ cache
->pinned
+ cache
->reserved
<
391 spin_unlock(&cache
->lock
);
395 spin_unlock(&cache
->lock
);
403 if (!full_search
&& factor
< 10) {
413 struct btrfs_block_group_cache
*btrfs_find_block_group(struct btrfs_root
*root
,
414 struct btrfs_block_group_cache
415 *hint
, u64 search_start
,
419 struct btrfs_block_group_cache
*ret
;
420 ret
= __btrfs_find_block_group(root
, hint
, search_start
, data
, owner
);
424 /* simple helper to search for an existing extent at a given offset */
425 int btrfs_lookup_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
428 struct btrfs_key key
;
429 struct btrfs_path
*path
;
431 path
= btrfs_alloc_path();
433 key
.objectid
= start
;
435 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
436 ret
= btrfs_search_slot(NULL
, root
->fs_info
->extent_root
, &key
, path
,
438 btrfs_free_path(path
);
443 * Back reference rules. Back refs have three main goals:
445 * 1) differentiate between all holders of references to an extent so that
446 * when a reference is dropped we can make sure it was a valid reference
447 * before freeing the extent.
449 * 2) Provide enough information to quickly find the holders of an extent
450 * if we notice a given block is corrupted or bad.
452 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
453 * maintenance. This is actually the same as #2, but with a slightly
454 * different use case.
456 * File extents can be referenced by:
458 * - multiple snapshots, subvolumes, or different generations in one subvol
459 * - different files inside a single subvolume
460 * - different offsets inside a file (bookend extents in file.c)
462 * The extent ref structure has fields for:
464 * - Objectid of the subvolume root
465 * - Generation number of the tree holding the reference
466 * - objectid of the file holding the reference
467 * - number of references holding by parent node (alway 1 for tree blocks)
469 * Btree leaf may hold multiple references to a file extent. In most cases,
470 * these references are from same file and the corresponding offsets inside
471 * the file are close together.
473 * When a file extent is allocated the fields are filled in:
474 * (root_key.objectid, trans->transid, inode objectid, 1)
476 * When a leaf is cow'd new references are added for every file extent found
477 * in the leaf. It looks similar to the create case, but trans->transid will
478 * be different when the block is cow'd.
480 * (root_key.objectid, trans->transid, inode objectid,
481 * number of references in the leaf)
483 * When a file extent is removed either during snapshot deletion or
484 * file truncation, we find the corresponding back reference and check
485 * the following fields:
487 * (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
490 * Btree extents can be referenced by:
492 * - Different subvolumes
493 * - Different generations of the same subvolume
495 * When a tree block is created, back references are inserted:
497 * (root->root_key.objectid, trans->transid, level, 1)
499 * When a tree block is cow'd, new back references are added for all the
500 * blocks it points to. If the tree block isn't in reference counted root,
501 * the old back references are removed. These new back references are of
502 * the form (trans->transid will have increased since creation):
504 * (root->root_key.objectid, trans->transid, level, 1)
506 * When a backref is in deleting, the following fields are checked:
508 * if backref was for a tree root:
509 * (btrfs_header_owner(itself), btrfs_header_generation(itself), level)
511 * (btrfs_header_owner(parent), btrfs_header_generation(parent), level)
513 * Back Reference Key composing:
515 * The key objectid corresponds to the first byte in the extent, the key
516 * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first
517 * byte of parent extent. If a extent is tree root, the key offset is set
518 * to the key objectid.
521 static int noinline
lookup_extent_backref(struct btrfs_trans_handle
*trans
,
522 struct btrfs_root
*root
,
523 struct btrfs_path
*path
,
524 u64 bytenr
, u64 parent
,
525 u64 ref_root
, u64 ref_generation
,
526 u64 owner_objectid
, int del
)
528 struct btrfs_key key
;
529 struct btrfs_extent_ref
*ref
;
530 struct extent_buffer
*leaf
;
534 key
.objectid
= bytenr
;
535 key
.type
= BTRFS_EXTENT_REF_KEY
;
538 ret
= btrfs_search_slot(trans
, root
, &key
, path
, del
? -1 : 0, 1);
546 leaf
= path
->nodes
[0];
547 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_ref
);
548 ref_objectid
= btrfs_ref_objectid(leaf
, ref
);
549 if (btrfs_ref_root(leaf
, ref
) != ref_root
||
550 btrfs_ref_generation(leaf
, ref
) != ref_generation
||
551 (ref_objectid
!= owner_objectid
&&
552 ref_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
)) {
562 static int noinline
insert_extent_backref(struct btrfs_trans_handle
*trans
,
563 struct btrfs_root
*root
,
564 struct btrfs_path
*path
,
565 u64 bytenr
, u64 parent
,
566 u64 ref_root
, u64 ref_generation
,
569 struct btrfs_key key
;
570 struct extent_buffer
*leaf
;
571 struct btrfs_extent_ref
*ref
;
575 key
.objectid
= bytenr
;
576 key
.type
= BTRFS_EXTENT_REF_KEY
;
579 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, sizeof(*ref
));
581 leaf
= path
->nodes
[0];
582 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
583 struct btrfs_extent_ref
);
584 btrfs_set_ref_root(leaf
, ref
, ref_root
);
585 btrfs_set_ref_generation(leaf
, ref
, ref_generation
);
586 btrfs_set_ref_objectid(leaf
, ref
, owner_objectid
);
587 btrfs_set_ref_num_refs(leaf
, ref
, 1);
588 } else if (ret
== -EEXIST
) {
590 BUG_ON(owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
);
591 leaf
= path
->nodes
[0];
592 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
593 struct btrfs_extent_ref
);
594 if (btrfs_ref_root(leaf
, ref
) != ref_root
||
595 btrfs_ref_generation(leaf
, ref
) != ref_generation
) {
601 num_refs
= btrfs_ref_num_refs(leaf
, ref
);
602 BUG_ON(num_refs
== 0);
603 btrfs_set_ref_num_refs(leaf
, ref
, num_refs
+ 1);
605 existing_owner
= btrfs_ref_objectid(leaf
, ref
);
606 if (existing_owner
!= owner_objectid
&&
607 existing_owner
!= BTRFS_MULTIPLE_OBJECTIDS
) {
608 btrfs_set_ref_objectid(leaf
, ref
,
609 BTRFS_MULTIPLE_OBJECTIDS
);
615 btrfs_mark_buffer_dirty(path
->nodes
[0]);
617 btrfs_release_path(root
, path
);
621 static int noinline
remove_extent_backref(struct btrfs_trans_handle
*trans
,
622 struct btrfs_root
*root
,
623 struct btrfs_path
*path
)
625 struct extent_buffer
*leaf
;
626 struct btrfs_extent_ref
*ref
;
630 leaf
= path
->nodes
[0];
631 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_ref
);
632 num_refs
= btrfs_ref_num_refs(leaf
, ref
);
633 BUG_ON(num_refs
== 0);
636 ret
= btrfs_del_item(trans
, root
, path
);
638 btrfs_set_ref_num_refs(leaf
, ref
, num_refs
);
639 btrfs_mark_buffer_dirty(leaf
);
641 btrfs_release_path(root
, path
);
645 static int __btrfs_update_extent_ref(struct btrfs_trans_handle
*trans
,
646 struct btrfs_root
*root
, u64 bytenr
,
647 u64 orig_parent
, u64 parent
,
648 u64 orig_root
, u64 ref_root
,
649 u64 orig_generation
, u64 ref_generation
,
653 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
654 struct btrfs_path
*path
;
656 if (root
== root
->fs_info
->extent_root
) {
657 struct pending_extent_op
*extent_op
;
660 BUG_ON(owner_objectid
>= BTRFS_MAX_LEVEL
);
661 num_bytes
= btrfs_level_size(root
, (int)owner_objectid
);
662 mutex_lock(&root
->fs_info
->extent_ins_mutex
);
663 if (test_range_bit(&root
->fs_info
->extent_ins
, bytenr
,
664 bytenr
+ num_bytes
- 1, EXTENT_WRITEBACK
, 0)) {
666 ret
= get_state_private(&root
->fs_info
->extent_ins
,
669 extent_op
= (struct pending_extent_op
*)
671 BUG_ON(extent_op
->parent
!= orig_parent
);
672 BUG_ON(extent_op
->generation
!= orig_generation
);
674 extent_op
->parent
= parent
;
675 extent_op
->generation
= ref_generation
;
677 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
680 extent_op
->type
= PENDING_BACKREF_UPDATE
;
681 extent_op
->bytenr
= bytenr
;
682 extent_op
->num_bytes
= num_bytes
;
683 extent_op
->parent
= parent
;
684 extent_op
->orig_parent
= orig_parent
;
685 extent_op
->generation
= ref_generation
;
686 extent_op
->orig_generation
= orig_generation
;
687 extent_op
->level
= (int)owner_objectid
;
689 set_extent_bits(&root
->fs_info
->extent_ins
,
690 bytenr
, bytenr
+ num_bytes
- 1,
691 EXTENT_WRITEBACK
, GFP_NOFS
);
692 set_state_private(&root
->fs_info
->extent_ins
,
693 bytenr
, (unsigned long)extent_op
);
695 mutex_unlock(&root
->fs_info
->extent_ins_mutex
);
699 path
= btrfs_alloc_path();
702 ret
= lookup_extent_backref(trans
, extent_root
, path
,
703 bytenr
, orig_parent
, orig_root
,
704 orig_generation
, owner_objectid
, 1);
707 ret
= remove_extent_backref(trans
, extent_root
, path
);
710 ret
= insert_extent_backref(trans
, extent_root
, path
, bytenr
,
711 parent
, ref_root
, ref_generation
,
714 finish_current_insert(trans
, extent_root
);
715 del_pending_extents(trans
, extent_root
);
717 btrfs_free_path(path
);
721 int btrfs_update_extent_ref(struct btrfs_trans_handle
*trans
,
722 struct btrfs_root
*root
, u64 bytenr
,
723 u64 orig_parent
, u64 parent
,
724 u64 ref_root
, u64 ref_generation
,
728 if (ref_root
== BTRFS_TREE_LOG_OBJECTID
&&
729 owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
)
731 ret
= __btrfs_update_extent_ref(trans
, root
, bytenr
, orig_parent
,
732 parent
, ref_root
, ref_root
,
733 ref_generation
, ref_generation
,
738 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
739 struct btrfs_root
*root
, u64 bytenr
,
740 u64 orig_parent
, u64 parent
,
741 u64 orig_root
, u64 ref_root
,
742 u64 orig_generation
, u64 ref_generation
,
745 struct btrfs_path
*path
;
747 struct btrfs_key key
;
748 struct extent_buffer
*l
;
749 struct btrfs_extent_item
*item
;
752 path
= btrfs_alloc_path();
757 key
.objectid
= bytenr
;
758 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
759 key
.offset
= (u64
)-1;
761 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
, path
,
765 BUG_ON(ret
== 0 || path
->slots
[0] == 0);
770 btrfs_item_key_to_cpu(l
, &key
, path
->slots
[0]);
771 BUG_ON(key
.objectid
!= bytenr
);
772 BUG_ON(key
.type
!= BTRFS_EXTENT_ITEM_KEY
);
774 item
= btrfs_item_ptr(l
, path
->slots
[0], struct btrfs_extent_item
);
775 refs
= btrfs_extent_refs(l
, item
);
776 btrfs_set_extent_refs(l
, item
, refs
+ 1);
777 btrfs_mark_buffer_dirty(path
->nodes
[0]);
779 btrfs_release_path(root
->fs_info
->extent_root
, path
);
782 ret
= insert_extent_backref(trans
, root
->fs_info
->extent_root
,
783 path
, bytenr
, parent
,
784 ref_root
, ref_generation
,
787 finish_current_insert(trans
, root
->fs_info
->extent_root
);
788 del_pending_extents(trans
, root
->fs_info
->extent_root
);
790 btrfs_free_path(path
);
794 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
795 struct btrfs_root
*root
,
796 u64 bytenr
, u64 num_bytes
, u64 parent
,
797 u64 ref_root
, u64 ref_generation
,
801 if (ref_root
== BTRFS_TREE_LOG_OBJECTID
&&
802 owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
)
804 ret
= __btrfs_inc_extent_ref(trans
, root
, bytenr
, 0, parent
,
805 0, ref_root
, 0, ref_generation
,
810 int btrfs_extent_post_op(struct btrfs_trans_handle
*trans
,
811 struct btrfs_root
*root
)
813 finish_current_insert(trans
, root
->fs_info
->extent_root
);
814 del_pending_extents(trans
, root
->fs_info
->extent_root
);
818 int btrfs_lookup_extent_ref(struct btrfs_trans_handle
*trans
,
819 struct btrfs_root
*root
, u64 bytenr
,
820 u64 num_bytes
, u32
*refs
)
822 struct btrfs_path
*path
;
824 struct btrfs_key key
;
825 struct extent_buffer
*l
;
826 struct btrfs_extent_item
*item
;
828 WARN_ON(num_bytes
< root
->sectorsize
);
829 path
= btrfs_alloc_path();
831 key
.objectid
= bytenr
;
832 key
.offset
= num_bytes
;
833 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
834 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
, path
,
839 btrfs_print_leaf(root
, path
->nodes
[0]);
840 printk("failed to find block number %Lu\n", bytenr
);
844 item
= btrfs_item_ptr(l
, path
->slots
[0], struct btrfs_extent_item
);
845 *refs
= btrfs_extent_refs(l
, item
);
847 btrfs_free_path(path
);
851 static int get_reference_status(struct btrfs_root
*root
, u64 bytenr
,
852 u64 parent_gen
, u64 ref_objectid
,
853 u64
*min_generation
, u32
*ref_count
)
855 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
856 struct btrfs_path
*path
;
857 struct extent_buffer
*leaf
;
858 struct btrfs_extent_ref
*ref_item
;
859 struct btrfs_key key
;
860 struct btrfs_key found_key
;
861 u64 root_objectid
= root
->root_key
.objectid
;
866 key
.objectid
= bytenr
;
867 key
.offset
= (u64
)-1;
868 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
870 path
= btrfs_alloc_path();
871 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
875 if (ret
< 0 || path
->slots
[0] == 0)
879 leaf
= path
->nodes
[0];
880 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
882 if (found_key
.objectid
!= bytenr
||
883 found_key
.type
!= BTRFS_EXTENT_ITEM_KEY
) {
889 *min_generation
= (u64
)-1;
892 leaf
= path
->nodes
[0];
893 nritems
= btrfs_header_nritems(leaf
);
894 if (path
->slots
[0] >= nritems
) {
895 ret
= btrfs_next_leaf(extent_root
, path
);
902 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
903 if (found_key
.objectid
!= bytenr
)
906 if (found_key
.type
!= BTRFS_EXTENT_REF_KEY
) {
911 ref_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
912 struct btrfs_extent_ref
);
913 ref_generation
= btrfs_ref_generation(leaf
, ref_item
);
915 * For (parent_gen > 0 && parent_gen > ref_generation):
917 * we reach here through the oldest root, therefore
918 * all other reference from same snapshot should have
919 * a larger generation.
921 if ((root_objectid
!= btrfs_ref_root(leaf
, ref_item
)) ||
922 (parent_gen
> 0 && parent_gen
> ref_generation
) ||
923 (ref_objectid
>= BTRFS_FIRST_FREE_OBJECTID
&&
924 ref_objectid
!= btrfs_ref_objectid(leaf
, ref_item
))) {
930 if (*min_generation
> ref_generation
)
931 *min_generation
= ref_generation
;
937 btrfs_free_path(path
);
941 int btrfs_cross_ref_exists(struct btrfs_trans_handle
*trans
,
942 struct btrfs_root
*root
,
943 struct btrfs_key
*key
, u64 bytenr
)
945 struct btrfs_root
*old_root
;
946 struct btrfs_path
*path
= NULL
;
947 struct extent_buffer
*eb
;
948 struct btrfs_file_extent_item
*item
;
956 BUG_ON(trans
== NULL
);
957 BUG_ON(key
->type
!= BTRFS_EXTENT_DATA_KEY
);
958 ret
= get_reference_status(root
, bytenr
, 0, key
->objectid
,
959 &min_generation
, &ref_count
);
966 old_root
= root
->dirty_root
->root
;
967 ref_generation
= old_root
->root_key
.offset
;
969 /* all references are created in running transaction */
970 if (min_generation
> ref_generation
) {
975 path
= btrfs_alloc_path();
981 path
->skip_locking
= 1;
982 /* if no item found, the extent is referenced by other snapshot */
983 ret
= btrfs_search_slot(NULL
, old_root
, key
, path
, 0, 0);
988 item
= btrfs_item_ptr(eb
, path
->slots
[0],
989 struct btrfs_file_extent_item
);
990 if (btrfs_file_extent_type(eb
, item
) != BTRFS_FILE_EXTENT_REG
||
991 btrfs_file_extent_disk_bytenr(eb
, item
) != bytenr
) {
996 for (level
= BTRFS_MAX_LEVEL
- 1; level
>= -1; level
--) {
998 eb
= path
->nodes
[level
];
1001 extent_start
= eb
->start
;
1003 extent_start
= bytenr
;
1005 ret
= get_reference_status(root
, extent_start
, ref_generation
,
1006 0, &min_generation
, &ref_count
);
1010 if (ref_count
!= 1) {
1015 ref_generation
= btrfs_header_generation(eb
);
1020 btrfs_free_path(path
);
1024 int btrfs_cache_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
1025 struct extent_buffer
*buf
, u32 nr_extents
)
1027 struct btrfs_key key
;
1028 struct btrfs_file_extent_item
*fi
;
1036 if (!root
->ref_cows
)
1039 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
1041 root_gen
= root
->root_key
.offset
;
1044 root_gen
= trans
->transid
- 1;
1047 level
= btrfs_header_level(buf
);
1048 nritems
= btrfs_header_nritems(buf
);
1051 struct btrfs_leaf_ref
*ref
;
1052 struct btrfs_extent_info
*info
;
1054 ref
= btrfs_alloc_leaf_ref(root
, nr_extents
);
1060 ref
->root_gen
= root_gen
;
1061 ref
->bytenr
= buf
->start
;
1062 ref
->owner
= btrfs_header_owner(buf
);
1063 ref
->generation
= btrfs_header_generation(buf
);
1064 ref
->nritems
= nr_extents
;
1065 info
= ref
->extents
;
1067 for (i
= 0; nr_extents
> 0 && i
< nritems
; i
++) {
1069 btrfs_item_key_to_cpu(buf
, &key
, i
);
1070 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
1072 fi
= btrfs_item_ptr(buf
, i
,
1073 struct btrfs_file_extent_item
);
1074 if (btrfs_file_extent_type(buf
, fi
) ==
1075 BTRFS_FILE_EXTENT_INLINE
)
1077 disk_bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
1078 if (disk_bytenr
== 0)
1081 info
->bytenr
= disk_bytenr
;
1083 btrfs_file_extent_disk_num_bytes(buf
, fi
);
1084 info
->objectid
= key
.objectid
;
1085 info
->offset
= key
.offset
;
1089 ret
= btrfs_add_leaf_ref(root
, ref
, shared
);
1090 if (ret
== -EEXIST
&& shared
) {
1091 struct btrfs_leaf_ref
*old
;
1092 old
= btrfs_lookup_leaf_ref(root
, ref
->bytenr
);
1094 btrfs_remove_leaf_ref(root
, old
);
1095 btrfs_free_leaf_ref(root
, old
);
1096 ret
= btrfs_add_leaf_ref(root
, ref
, shared
);
1099 btrfs_free_leaf_ref(root
, ref
);
1105 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
1106 struct extent_buffer
*orig_buf
, struct extent_buffer
*buf
,
1113 u64 orig_generation
;
1115 u32 nr_file_extents
= 0;
1116 struct btrfs_key key
;
1117 struct btrfs_file_extent_item
*fi
;
1122 int (*process_func
)(struct btrfs_trans_handle
*, struct btrfs_root
*,
1123 u64
, u64
, u64
, u64
, u64
, u64
, u64
, u64
);
1125 ref_root
= btrfs_header_owner(buf
);
1126 ref_generation
= btrfs_header_generation(buf
);
1127 orig_root
= btrfs_header_owner(orig_buf
);
1128 orig_generation
= btrfs_header_generation(orig_buf
);
1130 nritems
= btrfs_header_nritems(buf
);
1131 level
= btrfs_header_level(buf
);
1133 if (root
->ref_cows
) {
1134 process_func
= __btrfs_inc_extent_ref
;
1137 root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
)
1140 root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
)
1142 process_func
= __btrfs_update_extent_ref
;
1145 for (i
= 0; i
< nritems
; i
++) {
1148 btrfs_item_key_to_cpu(buf
, &key
, i
);
1149 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
1151 fi
= btrfs_item_ptr(buf
, i
,
1152 struct btrfs_file_extent_item
);
1153 if (btrfs_file_extent_type(buf
, fi
) ==
1154 BTRFS_FILE_EXTENT_INLINE
)
1156 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
1162 ret
= process_func(trans
, root
, bytenr
,
1163 orig_buf
->start
, buf
->start
,
1164 orig_root
, ref_root
,
1165 orig_generation
, ref_generation
,
1174 bytenr
= btrfs_node_blockptr(buf
, i
);
1175 ret
= process_func(trans
, root
, bytenr
,
1176 orig_buf
->start
, buf
->start
,
1177 orig_root
, ref_root
,
1178 orig_generation
, ref_generation
,
1190 *nr_extents
= nr_file_extents
;
1192 *nr_extents
= nritems
;
1200 int btrfs_update_ref(struct btrfs_trans_handle
*trans
,
1201 struct btrfs_root
*root
, struct extent_buffer
*orig_buf
,
1202 struct extent_buffer
*buf
, int start_slot
, int nr
)
1209 u64 orig_generation
;
1210 struct btrfs_key key
;
1211 struct btrfs_file_extent_item
*fi
;
1217 BUG_ON(start_slot
< 0);
1218 BUG_ON(start_slot
+ nr
> btrfs_header_nritems(buf
));
1220 ref_root
= btrfs_header_owner(buf
);
1221 ref_generation
= btrfs_header_generation(buf
);
1222 orig_root
= btrfs_header_owner(orig_buf
);
1223 orig_generation
= btrfs_header_generation(orig_buf
);
1224 level
= btrfs_header_level(buf
);
1226 if (!root
->ref_cows
) {
1228 root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
)
1231 root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
)
1235 for (i
= 0, slot
= start_slot
; i
< nr
; i
++, slot
++) {
1238 btrfs_item_key_to_cpu(buf
, &key
, slot
);
1239 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
1241 fi
= btrfs_item_ptr(buf
, slot
,
1242 struct btrfs_file_extent_item
);
1243 if (btrfs_file_extent_type(buf
, fi
) ==
1244 BTRFS_FILE_EXTENT_INLINE
)
1246 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
1249 ret
= __btrfs_update_extent_ref(trans
, root
, bytenr
,
1250 orig_buf
->start
, buf
->start
,
1251 orig_root
, ref_root
,
1252 orig_generation
, ref_generation
,
1257 bytenr
= btrfs_node_blockptr(buf
, slot
);
1258 ret
= __btrfs_update_extent_ref(trans
, root
, bytenr
,
1259 orig_buf
->start
, buf
->start
,
1260 orig_root
, ref_root
,
1261 orig_generation
, ref_generation
,
1273 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
1274 struct btrfs_root
*root
,
1275 struct btrfs_path
*path
,
1276 struct btrfs_block_group_cache
*cache
)
1280 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
1282 struct extent_buffer
*leaf
;
1284 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
1289 leaf
= path
->nodes
[0];
1290 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
1291 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
1292 btrfs_mark_buffer_dirty(leaf
);
1293 btrfs_release_path(extent_root
, path
);
1295 finish_current_insert(trans
, extent_root
);
1296 pending_ret
= del_pending_extents(trans
, extent_root
);
1305 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
1306 struct btrfs_root
*root
)
1308 struct btrfs_block_group_cache
*cache
, *entry
;
1312 struct btrfs_path
*path
;
1315 path
= btrfs_alloc_path();
1321 spin_lock(&root
->fs_info
->block_group_cache_lock
);
1322 for (n
= rb_first(&root
->fs_info
->block_group_cache_tree
);
1323 n
; n
= rb_next(n
)) {
1324 entry
= rb_entry(n
, struct btrfs_block_group_cache
,
1331 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
1337 last
+= cache
->key
.offset
;
1339 err
= write_one_cache_group(trans
, root
,
1342 * if we fail to write the cache group, we want
1343 * to keep it marked dirty in hopes that a later
1351 btrfs_free_path(path
);
1355 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
1356 u64 total_bytes
, u64 bytes_used
,
1357 struct btrfs_space_info
**space_info
)
1359 struct btrfs_space_info
*found
;
1361 found
= __find_space_info(info
, flags
);
1363 spin_lock(&found
->lock
);
1364 found
->total_bytes
+= total_bytes
;
1365 found
->bytes_used
+= bytes_used
;
1367 spin_unlock(&found
->lock
);
1368 *space_info
= found
;
1371 found
= kmalloc(sizeof(*found
), GFP_NOFS
);
1375 list_add(&found
->list
, &info
->space_info
);
1376 INIT_LIST_HEAD(&found
->block_groups
);
1377 init_rwsem(&found
->groups_sem
);
1378 spin_lock_init(&found
->lock
);
1379 found
->flags
= flags
;
1380 found
->total_bytes
= total_bytes
;
1381 found
->bytes_used
= bytes_used
;
1382 found
->bytes_pinned
= 0;
1383 found
->bytes_reserved
= 0;
1385 found
->force_alloc
= 0;
1386 *space_info
= found
;
1390 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
1392 u64 extra_flags
= flags
& (BTRFS_BLOCK_GROUP_RAID0
|
1393 BTRFS_BLOCK_GROUP_RAID1
|
1394 BTRFS_BLOCK_GROUP_RAID10
|
1395 BTRFS_BLOCK_GROUP_DUP
);
1397 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
1398 fs_info
->avail_data_alloc_bits
|= extra_flags
;
1399 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
1400 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
1401 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
1402 fs_info
->avail_system_alloc_bits
|= extra_flags
;
1406 static u64
reduce_alloc_profile(struct btrfs_root
*root
, u64 flags
)
1408 u64 num_devices
= root
->fs_info
->fs_devices
->num_devices
;
1410 if (num_devices
== 1)
1411 flags
&= ~(BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID0
);
1412 if (num_devices
< 4)
1413 flags
&= ~BTRFS_BLOCK_GROUP_RAID10
;
1415 if ((flags
& BTRFS_BLOCK_GROUP_DUP
) &&
1416 (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
1417 BTRFS_BLOCK_GROUP_RAID10
))) {
1418 flags
&= ~BTRFS_BLOCK_GROUP_DUP
;
1421 if ((flags
& BTRFS_BLOCK_GROUP_RAID1
) &&
1422 (flags
& BTRFS_BLOCK_GROUP_RAID10
)) {
1423 flags
&= ~BTRFS_BLOCK_GROUP_RAID1
;
1426 if ((flags
& BTRFS_BLOCK_GROUP_RAID0
) &&
1427 ((flags
& BTRFS_BLOCK_GROUP_RAID1
) |
1428 (flags
& BTRFS_BLOCK_GROUP_RAID10
) |
1429 (flags
& BTRFS_BLOCK_GROUP_DUP
)))
1430 flags
&= ~BTRFS_BLOCK_GROUP_RAID0
;
1434 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
1435 struct btrfs_root
*extent_root
, u64 alloc_bytes
,
1436 u64 flags
, int force
)
1438 struct btrfs_space_info
*space_info
;
1442 int ret
= 0, waited
= 0;
1444 flags
= reduce_alloc_profile(extent_root
, flags
);
1446 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
1448 ret
= update_space_info(extent_root
->fs_info
, flags
,
1452 BUG_ON(!space_info
);
1454 spin_lock(&space_info
->lock
);
1455 if (space_info
->force_alloc
) {
1457 space_info
->force_alloc
= 0;
1459 if (space_info
->full
) {
1460 spin_unlock(&space_info
->lock
);
1464 thresh
= div_factor(space_info
->total_bytes
, 6);
1466 (space_info
->bytes_used
+ space_info
->bytes_pinned
+
1467 space_info
->bytes_reserved
+ alloc_bytes
) < thresh
) {
1468 spin_unlock(&space_info
->lock
);
1472 spin_unlock(&space_info
->lock
);
1474 ret
= mutex_trylock(&extent_root
->fs_info
->chunk_mutex
);
1475 if (!ret
&& !force
) {
1478 mutex_lock(&extent_root
->fs_info
->chunk_mutex
);
1483 spin_lock(&space_info
->lock
);
1484 if (space_info
->full
) {
1485 spin_unlock(&space_info
->lock
);
1488 spin_unlock(&space_info
->lock
);
1491 ret
= btrfs_alloc_chunk(trans
, extent_root
, &start
, &num_bytes
, flags
);
1493 printk("space info full %Lu\n", flags
);
1494 space_info
->full
= 1;
1498 ret
= btrfs_make_block_group(trans
, extent_root
, 0, flags
,
1499 BTRFS_FIRST_CHUNK_TREE_OBJECTID
, start
, num_bytes
);
1502 mutex_unlock(&extent_root
->fs_info
->chunk_mutex
);
1507 static int update_block_group(struct btrfs_trans_handle
*trans
,
1508 struct btrfs_root
*root
,
1509 u64 bytenr
, u64 num_bytes
, int alloc
,
1512 struct btrfs_block_group_cache
*cache
;
1513 struct btrfs_fs_info
*info
= root
->fs_info
;
1514 u64 total
= num_bytes
;
1519 cache
= btrfs_lookup_block_group(info
, bytenr
);
1523 byte_in_group
= bytenr
- cache
->key
.objectid
;
1524 WARN_ON(byte_in_group
> cache
->key
.offset
);
1526 spin_lock(&cache
->space_info
->lock
);
1527 spin_lock(&cache
->lock
);
1529 old_val
= btrfs_block_group_used(&cache
->item
);
1530 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
1532 old_val
+= num_bytes
;
1533 cache
->space_info
->bytes_used
+= num_bytes
;
1534 btrfs_set_block_group_used(&cache
->item
, old_val
);
1535 spin_unlock(&cache
->lock
);
1536 spin_unlock(&cache
->space_info
->lock
);
1538 old_val
-= num_bytes
;
1539 cache
->space_info
->bytes_used
-= num_bytes
;
1540 btrfs_set_block_group_used(&cache
->item
, old_val
);
1541 spin_unlock(&cache
->lock
);
1542 spin_unlock(&cache
->space_info
->lock
);
1545 ret
= btrfs_add_free_space(cache
, bytenr
,
1552 bytenr
+= num_bytes
;
1557 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
1559 struct btrfs_block_group_cache
*cache
;
1561 cache
= btrfs_lookup_first_block_group(root
->fs_info
, search_start
);
1565 return cache
->key
.objectid
;
1568 int btrfs_update_pinned_extents(struct btrfs_root
*root
,
1569 u64 bytenr
, u64 num
, int pin
)
1572 struct btrfs_block_group_cache
*cache
;
1573 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1575 WARN_ON(!mutex_is_locked(&root
->fs_info
->pinned_mutex
));
1577 set_extent_dirty(&fs_info
->pinned_extents
,
1578 bytenr
, bytenr
+ num
- 1, GFP_NOFS
);
1580 clear_extent_dirty(&fs_info
->pinned_extents
,
1581 bytenr
, bytenr
+ num
- 1, GFP_NOFS
);
1584 cache
= btrfs_lookup_block_group(fs_info
, bytenr
);
1586 len
= min(num
, cache
->key
.offset
-
1587 (bytenr
- cache
->key
.objectid
));
1589 spin_lock(&cache
->space_info
->lock
);
1590 spin_lock(&cache
->lock
);
1591 cache
->pinned
+= len
;
1592 cache
->space_info
->bytes_pinned
+= len
;
1593 spin_unlock(&cache
->lock
);
1594 spin_unlock(&cache
->space_info
->lock
);
1595 fs_info
->total_pinned
+= len
;
1597 spin_lock(&cache
->space_info
->lock
);
1598 spin_lock(&cache
->lock
);
1599 cache
->pinned
-= len
;
1600 cache
->space_info
->bytes_pinned
-= len
;
1601 spin_unlock(&cache
->lock
);
1602 spin_unlock(&cache
->space_info
->lock
);
1603 fs_info
->total_pinned
-= len
;
1611 static int update_reserved_extents(struct btrfs_root
*root
,
1612 u64 bytenr
, u64 num
, int reserve
)
1615 struct btrfs_block_group_cache
*cache
;
1616 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1619 cache
= btrfs_lookup_block_group(fs_info
, bytenr
);
1621 len
= min(num
, cache
->key
.offset
-
1622 (bytenr
- cache
->key
.objectid
));
1624 spin_lock(&cache
->space_info
->lock
);
1625 spin_lock(&cache
->lock
);
1627 cache
->reserved
+= len
;
1628 cache
->space_info
->bytes_reserved
+= len
;
1630 cache
->reserved
-= len
;
1631 cache
->space_info
->bytes_reserved
-= len
;
1633 spin_unlock(&cache
->lock
);
1634 spin_unlock(&cache
->space_info
->lock
);
1641 int btrfs_copy_pinned(struct btrfs_root
*root
, struct extent_io_tree
*copy
)
1646 struct extent_io_tree
*pinned_extents
= &root
->fs_info
->pinned_extents
;
1649 mutex_lock(&root
->fs_info
->pinned_mutex
);
1651 ret
= find_first_extent_bit(pinned_extents
, last
,
1652 &start
, &end
, EXTENT_DIRTY
);
1655 set_extent_dirty(copy
, start
, end
, GFP_NOFS
);
1658 mutex_unlock(&root
->fs_info
->pinned_mutex
);
1662 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
1663 struct btrfs_root
*root
,
1664 struct extent_io_tree
*unpin
)
1669 struct btrfs_block_group_cache
*cache
;
1671 mutex_lock(&root
->fs_info
->pinned_mutex
);
1673 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
1677 btrfs_update_pinned_extents(root
, start
, end
+ 1 - start
, 0);
1678 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
1679 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
1681 btrfs_add_free_space(cache
, start
, end
- start
+ 1);
1682 if (need_resched()) {
1683 mutex_unlock(&root
->fs_info
->pinned_mutex
);
1685 mutex_lock(&root
->fs_info
->pinned_mutex
);
1688 mutex_unlock(&root
->fs_info
->pinned_mutex
);
1692 static int finish_current_insert(struct btrfs_trans_handle
*trans
,
1693 struct btrfs_root
*extent_root
)
1699 struct btrfs_fs_info
*info
= extent_root
->fs_info
;
1700 struct btrfs_path
*path
;
1701 struct btrfs_extent_ref
*ref
;
1702 struct pending_extent_op
*extent_op
;
1703 struct btrfs_key key
;
1704 struct btrfs_extent_item extent_item
;
1708 btrfs_set_stack_extent_refs(&extent_item
, 1);
1709 path
= btrfs_alloc_path();
1712 mutex_lock(&info
->extent_ins_mutex
);
1713 ret
= find_first_extent_bit(&info
->extent_ins
, search
, &start
,
1714 &end
, EXTENT_WRITEBACK
);
1716 mutex_unlock(&info
->extent_ins_mutex
);
1724 ret
= try_lock_extent(&info
->extent_ins
, start
, end
, GFP_NOFS
);
1727 mutex_unlock(&info
->extent_ins_mutex
);
1733 ret
= get_state_private(&info
->extent_ins
, start
, &priv
);
1735 extent_op
= (struct pending_extent_op
*)(unsigned long)priv
;
1737 mutex_unlock(&info
->extent_ins_mutex
);
1739 if (extent_op
->type
== PENDING_EXTENT_INSERT
) {
1740 key
.objectid
= start
;
1741 key
.offset
= end
+ 1 - start
;
1742 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1743 err
= btrfs_insert_item(trans
, extent_root
, &key
,
1744 &extent_item
, sizeof(extent_item
));
1747 mutex_lock(&info
->extent_ins_mutex
);
1748 clear_extent_bits(&info
->extent_ins
, start
, end
,
1749 EXTENT_WRITEBACK
, GFP_NOFS
);
1750 mutex_unlock(&info
->extent_ins_mutex
);
1752 err
= insert_extent_backref(trans
, extent_root
, path
,
1753 start
, extent_op
->parent
,
1754 extent_root
->root_key
.objectid
,
1755 extent_op
->generation
,
1758 } else if (extent_op
->type
== PENDING_BACKREF_UPDATE
) {
1759 err
= lookup_extent_backref(trans
, extent_root
, path
,
1760 start
, extent_op
->orig_parent
,
1761 extent_root
->root_key
.objectid
,
1762 extent_op
->orig_generation
,
1763 extent_op
->level
, 0);
1766 mutex_lock(&info
->extent_ins_mutex
);
1767 clear_extent_bits(&info
->extent_ins
, start
, end
,
1768 EXTENT_WRITEBACK
, GFP_NOFS
);
1769 mutex_unlock(&info
->extent_ins_mutex
);
1771 key
.objectid
= start
;
1772 key
.offset
= extent_op
->parent
;
1773 key
.type
= BTRFS_EXTENT_REF_KEY
;
1774 err
= btrfs_set_item_key_safe(trans
, extent_root
, path
,
1777 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1778 struct btrfs_extent_ref
);
1779 btrfs_set_ref_generation(path
->nodes
[0], ref
,
1780 extent_op
->generation
);
1781 btrfs_mark_buffer_dirty(path
->nodes
[0]);
1782 btrfs_release_path(extent_root
, path
);
1787 unlock_extent(&info
->extent_ins
, start
, end
, GFP_NOFS
);
1792 btrfs_free_path(path
);
1796 static int pin_down_bytes(struct btrfs_trans_handle
*trans
,
1797 struct btrfs_root
*root
,
1798 u64 bytenr
, u64 num_bytes
, int is_data
)
1801 struct extent_buffer
*buf
;
1806 buf
= btrfs_find_tree_block(root
, bytenr
, num_bytes
);
1810 /* we can reuse a block if it hasn't been written
1811 * and it is from this transaction. We can't
1812 * reuse anything from the tree log root because
1813 * it has tiny sub-transactions.
1815 if (btrfs_buffer_uptodate(buf
, 0) &&
1816 btrfs_try_tree_lock(buf
)) {
1817 u64 header_owner
= btrfs_header_owner(buf
);
1818 u64 header_transid
= btrfs_header_generation(buf
);
1819 if (header_owner
!= BTRFS_TREE_LOG_OBJECTID
&&
1820 header_owner
!= BTRFS_TREE_RELOC_OBJECTID
&&
1821 header_transid
== trans
->transid
&&
1822 !btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
1823 clean_tree_block(NULL
, root
, buf
);
1824 btrfs_tree_unlock(buf
);
1825 free_extent_buffer(buf
);
1828 btrfs_tree_unlock(buf
);
1830 free_extent_buffer(buf
);
1832 btrfs_update_pinned_extents(root
, bytenr
, num_bytes
, 1);
1839 * remove an extent from the root, returns 0 on success
1841 static int __free_extent(struct btrfs_trans_handle
*trans
,
1842 struct btrfs_root
*root
,
1843 u64 bytenr
, u64 num_bytes
, u64 parent
,
1844 u64 root_objectid
, u64 ref_generation
,
1845 u64 owner_objectid
, int pin
, int mark_free
)
1847 struct btrfs_path
*path
;
1848 struct btrfs_key key
;
1849 struct btrfs_fs_info
*info
= root
->fs_info
;
1850 struct btrfs_root
*extent_root
= info
->extent_root
;
1851 struct extent_buffer
*leaf
;
1853 int extent_slot
= 0;
1854 int found_extent
= 0;
1856 struct btrfs_extent_item
*ei
;
1859 key
.objectid
= bytenr
;
1860 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
1861 key
.offset
= num_bytes
;
1862 path
= btrfs_alloc_path();
1867 ret
= lookup_extent_backref(trans
, extent_root
, path
,
1868 bytenr
, parent
, root_objectid
,
1869 ref_generation
, owner_objectid
, 1);
1871 struct btrfs_key found_key
;
1872 extent_slot
= path
->slots
[0];
1873 while(extent_slot
> 0) {
1875 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1877 if (found_key
.objectid
!= bytenr
)
1879 if (found_key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
1880 found_key
.offset
== num_bytes
) {
1884 if (path
->slots
[0] - extent_slot
> 5)
1887 if (!found_extent
) {
1888 ret
= remove_extent_backref(trans
, extent_root
, path
);
1890 btrfs_release_path(extent_root
, path
);
1891 ret
= btrfs_search_slot(trans
, extent_root
,
1894 extent_slot
= path
->slots
[0];
1897 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
1899 printk("Unable to find ref byte nr %Lu root %Lu "
1900 "gen %Lu owner %Lu\n", bytenr
,
1901 root_objectid
, ref_generation
, owner_objectid
);
1904 leaf
= path
->nodes
[0];
1905 ei
= btrfs_item_ptr(leaf
, extent_slot
,
1906 struct btrfs_extent_item
);
1907 refs
= btrfs_extent_refs(leaf
, ei
);
1910 btrfs_set_extent_refs(leaf
, ei
, refs
);
1912 btrfs_mark_buffer_dirty(leaf
);
1914 if (refs
== 0 && found_extent
&& path
->slots
[0] == extent_slot
+ 1) {
1915 struct btrfs_extent_ref
*ref
;
1916 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1917 struct btrfs_extent_ref
);
1918 BUG_ON(btrfs_ref_num_refs(leaf
, ref
) != 1);
1919 /* if the back ref and the extent are next to each other
1920 * they get deleted below in one shot
1922 path
->slots
[0] = extent_slot
;
1924 } else if (found_extent
) {
1925 /* otherwise delete the extent back ref */
1926 ret
= remove_extent_backref(trans
, extent_root
, path
);
1928 /* if refs are 0, we need to setup the path for deletion */
1930 btrfs_release_path(extent_root
, path
);
1931 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
1940 #ifdef BIO_RW_DISCARD
1941 u64 map_length
= num_bytes
;
1942 struct btrfs_multi_bio
*multi
= NULL
;
1946 mutex_lock(&root
->fs_info
->pinned_mutex
);
1947 ret
= pin_down_bytes(trans
, root
, bytenr
, num_bytes
,
1948 owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
);
1949 mutex_unlock(&root
->fs_info
->pinned_mutex
);
1955 /* block accounting for super block */
1956 spin_lock_irq(&info
->delalloc_lock
);
1957 super_used
= btrfs_super_bytes_used(&info
->super_copy
);
1958 btrfs_set_super_bytes_used(&info
->super_copy
,
1959 super_used
- num_bytes
);
1960 spin_unlock_irq(&info
->delalloc_lock
);
1962 /* block accounting for root item */
1963 root_used
= btrfs_root_used(&root
->root_item
);
1964 btrfs_set_root_used(&root
->root_item
,
1965 root_used
- num_bytes
);
1966 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
1969 btrfs_release_path(extent_root
, path
);
1970 ret
= update_block_group(trans
, root
, bytenr
, num_bytes
, 0,
1974 #ifdef BIO_RW_DISCARD
1975 /* Tell the block device(s) that the sectors can be discarded */
1976 ret
= btrfs_map_block(&root
->fs_info
->mapping_tree
, READ
,
1977 bytenr
, &map_length
, &multi
, 0);
1979 struct btrfs_bio_stripe
*stripe
= multi
->stripes
;
1982 if (map_length
> num_bytes
)
1983 map_length
= num_bytes
;
1985 for (i
= 0; i
< multi
->num_stripes
; i
++, stripe
++) {
1986 blkdev_issue_discard(stripe
->dev
->bdev
,
1987 stripe
->physical
>> 9,
1994 btrfs_free_path(path
);
1995 finish_current_insert(trans
, extent_root
);
2000 * find all the blocks marked as pending in the radix tree and remove
2001 * them from the extent map
2003 static int del_pending_extents(struct btrfs_trans_handle
*trans
, struct
2004 btrfs_root
*extent_root
)
2012 struct extent_io_tree
*pending_del
;
2013 struct extent_io_tree
*extent_ins
;
2014 struct pending_extent_op
*extent_op
;
2015 struct btrfs_fs_info
*info
= extent_root
->fs_info
;
2017 extent_ins
= &extent_root
->fs_info
->extent_ins
;
2018 pending_del
= &extent_root
->fs_info
->pending_del
;
2021 mutex_lock(&info
->extent_ins_mutex
);
2022 ret
= find_first_extent_bit(pending_del
, search
, &start
, &end
,
2025 mutex_unlock(&info
->extent_ins_mutex
);
2033 ret
= try_lock_extent(extent_ins
, start
, end
, GFP_NOFS
);
2036 mutex_unlock(&info
->extent_ins_mutex
);
2042 ret
= get_state_private(pending_del
, start
, &priv
);
2044 extent_op
= (struct pending_extent_op
*)(unsigned long)priv
;
2046 clear_extent_bits(pending_del
, start
, end
, EXTENT_WRITEBACK
,
2048 if (!test_range_bit(extent_ins
, start
, end
,
2049 EXTENT_WRITEBACK
, 0)) {
2050 mutex_unlock(&info
->extent_ins_mutex
);
2052 ret
= __free_extent(trans
, extent_root
,
2053 start
, end
+ 1 - start
,
2054 extent_op
->orig_parent
,
2055 extent_root
->root_key
.objectid
,
2056 extent_op
->orig_generation
,
2057 extent_op
->level
, 1, 0);
2062 ret
= get_state_private(&info
->extent_ins
, start
,
2065 extent_op
= (struct pending_extent_op
*)
2066 (unsigned long)priv
;
2068 clear_extent_bits(&info
->extent_ins
, start
, end
,
2069 EXTENT_WRITEBACK
, GFP_NOFS
);
2071 mutex_unlock(&info
->extent_ins_mutex
);
2073 if (extent_op
->type
== PENDING_BACKREF_UPDATE
)
2076 mutex_lock(&extent_root
->fs_info
->pinned_mutex
);
2077 ret
= pin_down_bytes(trans
, extent_root
, start
,
2078 end
+ 1 - start
, 0);
2079 mutex_unlock(&extent_root
->fs_info
->pinned_mutex
);
2081 ret
= update_block_group(trans
, extent_root
, start
,
2082 end
+ 1 - start
, 0, ret
> 0);
2089 unlock_extent(extent_ins
, start
, end
, GFP_NOFS
);
2098 * remove an extent from the root, returns 0 on success
2100 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
2101 struct btrfs_root
*root
,
2102 u64 bytenr
, u64 num_bytes
, u64 parent
,
2103 u64 root_objectid
, u64 ref_generation
,
2104 u64 owner_objectid
, int pin
)
2106 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2110 WARN_ON(num_bytes
< root
->sectorsize
);
2111 if (root
== extent_root
) {
2112 struct pending_extent_op
*extent_op
;
2114 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
2117 extent_op
->type
= PENDING_EXTENT_DELETE
;
2118 extent_op
->bytenr
= bytenr
;
2119 extent_op
->num_bytes
= num_bytes
;
2120 extent_op
->parent
= parent
;
2121 extent_op
->orig_parent
= parent
;
2122 extent_op
->generation
= ref_generation
;
2123 extent_op
->orig_generation
= ref_generation
;
2124 extent_op
->level
= (int)owner_objectid
;
2126 mutex_lock(&root
->fs_info
->extent_ins_mutex
);
2127 set_extent_bits(&root
->fs_info
->pending_del
,
2128 bytenr
, bytenr
+ num_bytes
- 1,
2129 EXTENT_WRITEBACK
, GFP_NOFS
);
2130 set_state_private(&root
->fs_info
->pending_del
,
2131 bytenr
, (unsigned long)extent_op
);
2132 mutex_unlock(&root
->fs_info
->extent_ins_mutex
);
2135 /* if metadata always pin */
2136 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
) {
2137 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
2138 struct btrfs_block_group_cache
*cache
;
2140 /* btrfs_free_reserved_extent */
2141 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
2143 btrfs_add_free_space(cache
, bytenr
, num_bytes
);
2144 update_reserved_extents(root
, bytenr
, num_bytes
, 0);
2150 /* if data pin when any transaction has committed this */
2151 if (ref_generation
!= trans
->transid
)
2154 ret
= __free_extent(trans
, root
, bytenr
, num_bytes
, parent
,
2155 root_objectid
, ref_generation
,
2156 owner_objectid
, pin
, pin
== 0);
2158 finish_current_insert(trans
, root
->fs_info
->extent_root
);
2159 pending_ret
= del_pending_extents(trans
, root
->fs_info
->extent_root
);
2160 return ret
? ret
: pending_ret
;
2163 int btrfs_free_extent(struct btrfs_trans_handle
*trans
,
2164 struct btrfs_root
*root
,
2165 u64 bytenr
, u64 num_bytes
, u64 parent
,
2166 u64 root_objectid
, u64 ref_generation
,
2167 u64 owner_objectid
, int pin
)
2171 ret
= __btrfs_free_extent(trans
, root
, bytenr
, num_bytes
, parent
,
2172 root_objectid
, ref_generation
,
2173 owner_objectid
, pin
);
2177 static u64
stripe_align(struct btrfs_root
*root
, u64 val
)
2179 u64 mask
= ((u64
)root
->stripesize
- 1);
2180 u64 ret
= (val
+ mask
) & ~mask
;
2185 * walks the btree of allocated extents and find a hole of a given size.
2186 * The key ins is changed to record the hole:
2187 * ins->objectid == block start
2188 * ins->flags = BTRFS_EXTENT_ITEM_KEY
2189 * ins->offset == number of blocks
2190 * Any available blocks before search_start are skipped.
2192 static int noinline
find_free_extent(struct btrfs_trans_handle
*trans
,
2193 struct btrfs_root
*orig_root
,
2194 u64 num_bytes
, u64 empty_size
,
2195 u64 search_start
, u64 search_end
,
2196 u64 hint_byte
, struct btrfs_key
*ins
,
2197 u64 exclude_start
, u64 exclude_nr
,
2201 struct btrfs_root
* root
= orig_root
->fs_info
->extent_root
;
2202 u64 total_needed
= num_bytes
;
2203 u64
*last_ptr
= NULL
;
2204 struct btrfs_block_group_cache
*block_group
= NULL
;
2205 int chunk_alloc_done
= 0;
2206 int empty_cluster
= 2 * 1024 * 1024;
2207 int allowed_chunk_alloc
= 0;
2208 struct list_head
*head
= NULL
, *cur
= NULL
;
2210 struct btrfs_space_info
*space_info
;
2212 WARN_ON(num_bytes
< root
->sectorsize
);
2213 btrfs_set_key_type(ins
, BTRFS_EXTENT_ITEM_KEY
);
2217 if (orig_root
->ref_cows
|| empty_size
)
2218 allowed_chunk_alloc
= 1;
2220 if (data
& BTRFS_BLOCK_GROUP_METADATA
) {
2221 last_ptr
= &root
->fs_info
->last_alloc
;
2222 empty_cluster
= 256 * 1024;
2225 if ((data
& BTRFS_BLOCK_GROUP_DATA
) && btrfs_test_opt(root
, SSD
))
2226 last_ptr
= &root
->fs_info
->last_data_alloc
;
2230 hint_byte
= *last_ptr
;
2232 empty_size
+= empty_cluster
;
2234 search_start
= max(search_start
, first_logical_byte(root
, 0));
2235 search_start
= max(search_start
, hint_byte
);
2236 total_needed
+= empty_size
;
2238 block_group
= btrfs_lookup_block_group(root
->fs_info
, search_start
);
2239 space_info
= __find_space_info(root
->fs_info
, data
);
2241 down_read(&space_info
->groups_sem
);
2243 struct btrfs_free_space
*free_space
;
2245 * the only way this happens if our hint points to a block
2246 * group thats not of the proper type, while looping this
2247 * should never happen
2249 WARN_ON(!block_group
);
2250 mutex_lock(&block_group
->alloc_mutex
);
2251 if (unlikely(!block_group_bits(block_group
, data
)))
2254 ret
= cache_block_group(root
, block_group
);
2256 mutex_unlock(&block_group
->alloc_mutex
);
2260 if (block_group
->ro
)
2263 free_space
= btrfs_find_free_space(block_group
, search_start
,
2266 u64 start
= block_group
->key
.objectid
;
2267 u64 end
= block_group
->key
.objectid
+
2268 block_group
->key
.offset
;
2270 search_start
= stripe_align(root
, free_space
->offset
);
2272 /* move on to the next group */
2273 if (search_start
+ num_bytes
>= search_end
)
2276 /* move on to the next group */
2277 if (search_start
+ num_bytes
> end
)
2280 if (exclude_nr
> 0 &&
2281 (search_start
+ num_bytes
> exclude_start
&&
2282 search_start
< exclude_start
+ exclude_nr
)) {
2283 search_start
= exclude_start
+ exclude_nr
;
2285 * if search_start is still in this block group
2286 * then we just re-search this block group
2288 if (search_start
>= start
&&
2289 search_start
< end
) {
2290 mutex_unlock(&block_group
->alloc_mutex
);
2294 /* else we go to the next block group */
2298 ins
->objectid
= search_start
;
2299 ins
->offset
= num_bytes
;
2301 btrfs_remove_free_space_lock(block_group
, search_start
,
2303 /* we are all good, lets return */
2304 mutex_unlock(&block_group
->alloc_mutex
);
2308 mutex_unlock(&block_group
->alloc_mutex
);
2310 * Here's how this works.
2311 * loop == 0: we were searching a block group via a hint
2312 * and didn't find anything, so we start at
2313 * the head of the block groups and keep searching
2314 * loop == 1: we're searching through all of the block groups
2315 * if we hit the head again we have searched
2316 * all of the block groups for this space and we
2317 * need to try and allocate, if we cant error out.
2318 * loop == 2: we allocated more space and are looping through
2319 * all of the block groups again.
2322 head
= &space_info
->block_groups
;
2325 if (last_ptr
&& *last_ptr
) {
2326 total_needed
+= empty_cluster
;
2330 } else if (loop
== 1 && cur
== head
) {
2331 if (allowed_chunk_alloc
&& !chunk_alloc_done
) {
2332 up_read(&space_info
->groups_sem
);
2333 ret
= do_chunk_alloc(trans
, root
, num_bytes
+
2334 2 * 1024 * 1024, data
, 1);
2337 down_read(&space_info
->groups_sem
);
2339 head
= &space_info
->block_groups
;
2341 chunk_alloc_done
= 1;
2342 } else if (!allowed_chunk_alloc
) {
2343 space_info
->force_alloc
= 1;
2348 } else if (cur
== head
) {
2352 block_group
= list_entry(cur
, struct btrfs_block_group_cache
,
2354 search_start
= block_group
->key
.objectid
;
2358 /* we found what we needed */
2359 if (ins
->objectid
) {
2360 if (!(data
& BTRFS_BLOCK_GROUP_DATA
))
2361 trans
->block_group
= block_group
;
2364 *last_ptr
= ins
->objectid
+ ins
->offset
;
2370 up_read(&space_info
->groups_sem
);
2374 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
)
2376 struct btrfs_block_group_cache
*cache
;
2377 struct list_head
*l
;
2379 printk(KERN_INFO
"space_info has %Lu free, is %sfull\n",
2380 info
->total_bytes
- info
->bytes_used
- info
->bytes_pinned
-
2381 info
->bytes_reserved
, (info
->full
) ? "" : "not ");
2383 down_read(&info
->groups_sem
);
2384 list_for_each(l
, &info
->block_groups
) {
2385 cache
= list_entry(l
, struct btrfs_block_group_cache
, list
);
2386 spin_lock(&cache
->lock
);
2387 printk(KERN_INFO
"block group %Lu has %Lu bytes, %Lu used "
2388 "%Lu pinned %Lu reserved\n",
2389 cache
->key
.objectid
, cache
->key
.offset
,
2390 btrfs_block_group_used(&cache
->item
),
2391 cache
->pinned
, cache
->reserved
);
2392 btrfs_dump_free_space(cache
, bytes
);
2393 spin_unlock(&cache
->lock
);
2395 up_read(&info
->groups_sem
);
2398 static int __btrfs_reserve_extent(struct btrfs_trans_handle
*trans
,
2399 struct btrfs_root
*root
,
2400 u64 num_bytes
, u64 min_alloc_size
,
2401 u64 empty_size
, u64 hint_byte
,
2402 u64 search_end
, struct btrfs_key
*ins
,
2406 u64 search_start
= 0;
2408 struct btrfs_fs_info
*info
= root
->fs_info
;
2411 alloc_profile
= info
->avail_data_alloc_bits
&
2412 info
->data_alloc_profile
;
2413 data
= BTRFS_BLOCK_GROUP_DATA
| alloc_profile
;
2414 } else if (root
== root
->fs_info
->chunk_root
) {
2415 alloc_profile
= info
->avail_system_alloc_bits
&
2416 info
->system_alloc_profile
;
2417 data
= BTRFS_BLOCK_GROUP_SYSTEM
| alloc_profile
;
2419 alloc_profile
= info
->avail_metadata_alloc_bits
&
2420 info
->metadata_alloc_profile
;
2421 data
= BTRFS_BLOCK_GROUP_METADATA
| alloc_profile
;
2424 data
= reduce_alloc_profile(root
, data
);
2426 * the only place that sets empty_size is btrfs_realloc_node, which
2427 * is not called recursively on allocations
2429 if (empty_size
|| root
->ref_cows
) {
2430 if (!(data
& BTRFS_BLOCK_GROUP_METADATA
)) {
2431 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
2433 BTRFS_BLOCK_GROUP_METADATA
|
2434 (info
->metadata_alloc_profile
&
2435 info
->avail_metadata_alloc_bits
), 0);
2437 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
2438 num_bytes
+ 2 * 1024 * 1024, data
, 0);
2441 WARN_ON(num_bytes
< root
->sectorsize
);
2442 ret
= find_free_extent(trans
, root
, num_bytes
, empty_size
,
2443 search_start
, search_end
, hint_byte
, ins
,
2444 trans
->alloc_exclude_start
,
2445 trans
->alloc_exclude_nr
, data
);
2447 if (ret
== -ENOSPC
&& num_bytes
> min_alloc_size
) {
2448 num_bytes
= num_bytes
>> 1;
2449 num_bytes
= num_bytes
& ~(root
->sectorsize
- 1);
2450 num_bytes
= max(num_bytes
, min_alloc_size
);
2451 do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
2452 num_bytes
, data
, 1);
2456 struct btrfs_space_info
*sinfo
;
2458 sinfo
= __find_space_info(root
->fs_info
, data
);
2459 printk("allocation failed flags %Lu, wanted %Lu\n",
2461 dump_space_info(sinfo
, num_bytes
);
2468 int btrfs_free_reserved_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
2470 struct btrfs_block_group_cache
*cache
;
2472 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
2474 printk(KERN_ERR
"Unable to find block group for %Lu\n", start
);
2477 btrfs_add_free_space(cache
, start
, len
);
2478 update_reserved_extents(root
, start
, len
, 0);
2482 int btrfs_reserve_extent(struct btrfs_trans_handle
*trans
,
2483 struct btrfs_root
*root
,
2484 u64 num_bytes
, u64 min_alloc_size
,
2485 u64 empty_size
, u64 hint_byte
,
2486 u64 search_end
, struct btrfs_key
*ins
,
2490 ret
= __btrfs_reserve_extent(trans
, root
, num_bytes
, min_alloc_size
,
2491 empty_size
, hint_byte
, search_end
, ins
,
2493 update_reserved_extents(root
, ins
->objectid
, ins
->offset
, 1);
2497 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle
*trans
,
2498 struct btrfs_root
*root
, u64 parent
,
2499 u64 root_objectid
, u64 ref_generation
,
2500 u64 owner
, struct btrfs_key
*ins
)
2506 u64 num_bytes
= ins
->offset
;
2508 struct btrfs_fs_info
*info
= root
->fs_info
;
2509 struct btrfs_root
*extent_root
= info
->extent_root
;
2510 struct btrfs_extent_item
*extent_item
;
2511 struct btrfs_extent_ref
*ref
;
2512 struct btrfs_path
*path
;
2513 struct btrfs_key keys
[2];
2516 parent
= ins
->objectid
;
2518 /* block accounting for super block */
2519 spin_lock_irq(&info
->delalloc_lock
);
2520 super_used
= btrfs_super_bytes_used(&info
->super_copy
);
2521 btrfs_set_super_bytes_used(&info
->super_copy
, super_used
+ num_bytes
);
2522 spin_unlock_irq(&info
->delalloc_lock
);
2524 /* block accounting for root item */
2525 root_used
= btrfs_root_used(&root
->root_item
);
2526 btrfs_set_root_used(&root
->root_item
, root_used
+ num_bytes
);
2528 if (root
== extent_root
) {
2529 struct pending_extent_op
*extent_op
;
2531 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
2534 extent_op
->type
= PENDING_EXTENT_INSERT
;
2535 extent_op
->bytenr
= ins
->objectid
;
2536 extent_op
->num_bytes
= ins
->offset
;
2537 extent_op
->parent
= parent
;
2538 extent_op
->orig_parent
= 0;
2539 extent_op
->generation
= ref_generation
;
2540 extent_op
->orig_generation
= 0;
2541 extent_op
->level
= (int)owner
;
2543 mutex_lock(&root
->fs_info
->extent_ins_mutex
);
2544 set_extent_bits(&root
->fs_info
->extent_ins
, ins
->objectid
,
2545 ins
->objectid
+ ins
->offset
- 1,
2546 EXTENT_WRITEBACK
, GFP_NOFS
);
2547 set_state_private(&root
->fs_info
->extent_ins
,
2548 ins
->objectid
, (unsigned long)extent_op
);
2549 mutex_unlock(&root
->fs_info
->extent_ins_mutex
);
2553 memcpy(&keys
[0], ins
, sizeof(*ins
));
2554 keys
[1].objectid
= ins
->objectid
;
2555 keys
[1].type
= BTRFS_EXTENT_REF_KEY
;
2556 keys
[1].offset
= parent
;
2557 sizes
[0] = sizeof(*extent_item
);
2558 sizes
[1] = sizeof(*ref
);
2560 path
= btrfs_alloc_path();
2563 ret
= btrfs_insert_empty_items(trans
, extent_root
, path
, keys
,
2567 extent_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2568 struct btrfs_extent_item
);
2569 btrfs_set_extent_refs(path
->nodes
[0], extent_item
, 1);
2570 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
2571 struct btrfs_extent_ref
);
2573 btrfs_set_ref_root(path
->nodes
[0], ref
, root_objectid
);
2574 btrfs_set_ref_generation(path
->nodes
[0], ref
, ref_generation
);
2575 btrfs_set_ref_objectid(path
->nodes
[0], ref
, owner
);
2576 btrfs_set_ref_num_refs(path
->nodes
[0], ref
, 1);
2578 btrfs_mark_buffer_dirty(path
->nodes
[0]);
2580 trans
->alloc_exclude_start
= 0;
2581 trans
->alloc_exclude_nr
= 0;
2582 btrfs_free_path(path
);
2583 finish_current_insert(trans
, extent_root
);
2584 pending_ret
= del_pending_extents(trans
, extent_root
);
2594 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1, 0);
2596 printk("update block group failed for %Lu %Lu\n",
2597 ins
->objectid
, ins
->offset
);
2604 int btrfs_alloc_reserved_extent(struct btrfs_trans_handle
*trans
,
2605 struct btrfs_root
*root
, u64 parent
,
2606 u64 root_objectid
, u64 ref_generation
,
2607 u64 owner
, struct btrfs_key
*ins
)
2611 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
)
2613 ret
= __btrfs_alloc_reserved_extent(trans
, root
, parent
, root_objectid
,
2614 ref_generation
, owner
, ins
);
2615 update_reserved_extents(root
, ins
->objectid
, ins
->offset
, 0);
2620 * this is used by the tree logging recovery code. It records that
2621 * an extent has been allocated and makes sure to clear the free
2622 * space cache bits as well
2624 int btrfs_alloc_logged_extent(struct btrfs_trans_handle
*trans
,
2625 struct btrfs_root
*root
, u64 parent
,
2626 u64 root_objectid
, u64 ref_generation
,
2627 u64 owner
, struct btrfs_key
*ins
)
2630 struct btrfs_block_group_cache
*block_group
;
2632 block_group
= btrfs_lookup_block_group(root
->fs_info
, ins
->objectid
);
2633 mutex_lock(&block_group
->alloc_mutex
);
2634 cache_block_group(root
, block_group
);
2636 ret
= btrfs_remove_free_space_lock(block_group
, ins
->objectid
,
2638 mutex_unlock(&block_group
->alloc_mutex
);
2640 ret
= __btrfs_alloc_reserved_extent(trans
, root
, parent
, root_objectid
,
2641 ref_generation
, owner
, ins
);
2646 * finds a free extent and does all the dirty work required for allocation
2647 * returns the key for the extent through ins, and a tree buffer for
2648 * the first block of the extent through buf.
2650 * returns 0 if everything worked, non-zero otherwise.
2652 int btrfs_alloc_extent(struct btrfs_trans_handle
*trans
,
2653 struct btrfs_root
*root
,
2654 u64 num_bytes
, u64 parent
, u64 min_alloc_size
,
2655 u64 root_objectid
, u64 ref_generation
,
2656 u64 owner_objectid
, u64 empty_size
, u64 hint_byte
,
2657 u64 search_end
, struct btrfs_key
*ins
, u64 data
)
2661 ret
= __btrfs_reserve_extent(trans
, root
, num_bytes
,
2662 min_alloc_size
, empty_size
, hint_byte
,
2663 search_end
, ins
, data
);
2665 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
2666 ret
= __btrfs_alloc_reserved_extent(trans
, root
, parent
,
2667 root_objectid
, ref_generation
,
2668 owner_objectid
, ins
);
2672 update_reserved_extents(root
, ins
->objectid
, ins
->offset
, 1);
2677 struct extent_buffer
*btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
,
2678 struct btrfs_root
*root
,
2679 u64 bytenr
, u32 blocksize
)
2681 struct extent_buffer
*buf
;
2683 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
2685 return ERR_PTR(-ENOMEM
);
2686 btrfs_set_header_generation(buf
, trans
->transid
);
2687 btrfs_tree_lock(buf
);
2688 clean_tree_block(trans
, root
, buf
);
2689 btrfs_set_buffer_uptodate(buf
);
2690 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
2691 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
2692 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
2694 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
2695 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
2697 trans
->blocks_used
++;
2702 * helper function to allocate a block for a given tree
2703 * returns the tree buffer or NULL.
2705 struct extent_buffer
*btrfs_alloc_free_block(struct btrfs_trans_handle
*trans
,
2706 struct btrfs_root
*root
,
2707 u32 blocksize
, u64 parent
,
2714 struct btrfs_key ins
;
2716 struct extent_buffer
*buf
;
2718 ret
= btrfs_alloc_extent(trans
, root
, blocksize
, parent
, blocksize
,
2719 root_objectid
, ref_generation
, level
,
2720 empty_size
, hint
, (u64
)-1, &ins
, 0);
2723 return ERR_PTR(ret
);
2726 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
, blocksize
);
2730 int btrfs_drop_leaf_ref(struct btrfs_trans_handle
*trans
,
2731 struct btrfs_root
*root
, struct extent_buffer
*leaf
)
2734 u64 leaf_generation
;
2735 struct btrfs_key key
;
2736 struct btrfs_file_extent_item
*fi
;
2741 BUG_ON(!btrfs_is_leaf(leaf
));
2742 nritems
= btrfs_header_nritems(leaf
);
2743 leaf_owner
= btrfs_header_owner(leaf
);
2744 leaf_generation
= btrfs_header_generation(leaf
);
2746 for (i
= 0; i
< nritems
; i
++) {
2750 btrfs_item_key_to_cpu(leaf
, &key
, i
);
2751 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
2753 fi
= btrfs_item_ptr(leaf
, i
, struct btrfs_file_extent_item
);
2754 if (btrfs_file_extent_type(leaf
, fi
) ==
2755 BTRFS_FILE_EXTENT_INLINE
)
2758 * FIXME make sure to insert a trans record that
2759 * repeats the snapshot del on crash
2761 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
2762 if (disk_bytenr
== 0)
2765 ret
= __btrfs_free_extent(trans
, root
, disk_bytenr
,
2766 btrfs_file_extent_disk_num_bytes(leaf
, fi
),
2767 leaf
->start
, leaf_owner
, leaf_generation
,
2771 atomic_inc(&root
->fs_info
->throttle_gen
);
2772 wake_up(&root
->fs_info
->transaction_throttle
);
2778 static int noinline
cache_drop_leaf_ref(struct btrfs_trans_handle
*trans
,
2779 struct btrfs_root
*root
,
2780 struct btrfs_leaf_ref
*ref
)
2784 struct btrfs_extent_info
*info
= ref
->extents
;
2786 for (i
= 0; i
< ref
->nritems
; i
++) {
2787 ret
= __btrfs_free_extent(trans
, root
, info
->bytenr
,
2788 info
->num_bytes
, ref
->bytenr
,
2789 ref
->owner
, ref
->generation
,
2792 atomic_inc(&root
->fs_info
->throttle_gen
);
2793 wake_up(&root
->fs_info
->transaction_throttle
);
2803 int drop_snap_lookup_refcount(struct btrfs_root
*root
, u64 start
, u64 len
,
2808 ret
= btrfs_lookup_extent_ref(NULL
, root
, start
, len
, refs
);
2811 #if 0 // some debugging code in case we see problems here
2812 /* if the refs count is one, it won't get increased again. But
2813 * if the ref count is > 1, someone may be decreasing it at
2814 * the same time we are.
2817 struct extent_buffer
*eb
= NULL
;
2818 eb
= btrfs_find_create_tree_block(root
, start
, len
);
2820 btrfs_tree_lock(eb
);
2822 mutex_lock(&root
->fs_info
->alloc_mutex
);
2823 ret
= lookup_extent_ref(NULL
, root
, start
, len
, refs
);
2825 mutex_unlock(&root
->fs_info
->alloc_mutex
);
2828 btrfs_tree_unlock(eb
);
2829 free_extent_buffer(eb
);
2832 printk("block %llu went down to one during drop_snap\n",
2833 (unsigned long long)start
);
2844 * helper function for drop_snapshot, this walks down the tree dropping ref
2845 * counts as it goes.
2847 static int noinline
walk_down_tree(struct btrfs_trans_handle
*trans
,
2848 struct btrfs_root
*root
,
2849 struct btrfs_path
*path
, int *level
)
2855 struct extent_buffer
*next
;
2856 struct extent_buffer
*cur
;
2857 struct extent_buffer
*parent
;
2858 struct btrfs_leaf_ref
*ref
;
2863 WARN_ON(*level
< 0);
2864 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
2865 ret
= drop_snap_lookup_refcount(root
, path
->nodes
[*level
]->start
,
2866 path
->nodes
[*level
]->len
, &refs
);
2872 * walk down to the last node level and free all the leaves
2874 while(*level
>= 0) {
2875 WARN_ON(*level
< 0);
2876 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
2877 cur
= path
->nodes
[*level
];
2879 if (btrfs_header_level(cur
) != *level
)
2882 if (path
->slots
[*level
] >=
2883 btrfs_header_nritems(cur
))
2886 ret
= btrfs_drop_leaf_ref(trans
, root
, cur
);
2890 bytenr
= btrfs_node_blockptr(cur
, path
->slots
[*level
]);
2891 ptr_gen
= btrfs_node_ptr_generation(cur
, path
->slots
[*level
]);
2892 blocksize
= btrfs_level_size(root
, *level
- 1);
2894 ret
= drop_snap_lookup_refcount(root
, bytenr
, blocksize
, &refs
);
2897 parent
= path
->nodes
[*level
];
2898 root_owner
= btrfs_header_owner(parent
);
2899 root_gen
= btrfs_header_generation(parent
);
2900 path
->slots
[*level
]++;
2902 ret
= __btrfs_free_extent(trans
, root
, bytenr
,
2903 blocksize
, parent
->start
,
2904 root_owner
, root_gen
,
2908 atomic_inc(&root
->fs_info
->throttle_gen
);
2909 wake_up(&root
->fs_info
->transaction_throttle
);
2915 * at this point, we have a single ref, and since the
2916 * only place referencing this extent is a dead root
2917 * the reference count should never go higher.
2918 * So, we don't need to check it again
2921 ref
= btrfs_lookup_leaf_ref(root
, bytenr
);
2922 if (ref
&& ref
->generation
!= ptr_gen
) {
2923 btrfs_free_leaf_ref(root
, ref
);
2927 ret
= cache_drop_leaf_ref(trans
, root
, ref
);
2929 btrfs_remove_leaf_ref(root
, ref
);
2930 btrfs_free_leaf_ref(root
, ref
);
2934 if (printk_ratelimit()) {
2935 printk("leaf ref miss for bytenr %llu\n",
2936 (unsigned long long)bytenr
);
2939 next
= btrfs_find_tree_block(root
, bytenr
, blocksize
);
2940 if (!next
|| !btrfs_buffer_uptodate(next
, ptr_gen
)) {
2941 free_extent_buffer(next
);
2943 next
= read_tree_block(root
, bytenr
, blocksize
,
2948 * this is a debugging check and can go away
2949 * the ref should never go all the way down to 1
2952 ret
= lookup_extent_ref(NULL
, root
, bytenr
, blocksize
,
2958 WARN_ON(*level
<= 0);
2959 if (path
->nodes
[*level
-1])
2960 free_extent_buffer(path
->nodes
[*level
-1]);
2961 path
->nodes
[*level
-1] = next
;
2962 *level
= btrfs_header_level(next
);
2963 path
->slots
[*level
] = 0;
2967 WARN_ON(*level
< 0);
2968 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
2970 if (path
->nodes
[*level
] == root
->node
) {
2971 parent
= path
->nodes
[*level
];
2972 bytenr
= path
->nodes
[*level
]->start
;
2974 parent
= path
->nodes
[*level
+ 1];
2975 bytenr
= btrfs_node_blockptr(parent
, path
->slots
[*level
+ 1]);
2978 blocksize
= btrfs_level_size(root
, *level
);
2979 root_owner
= btrfs_header_owner(parent
);
2980 root_gen
= btrfs_header_generation(parent
);
2982 ret
= __btrfs_free_extent(trans
, root
, bytenr
, blocksize
,
2983 parent
->start
, root_owner
, root_gen
,
2985 free_extent_buffer(path
->nodes
[*level
]);
2986 path
->nodes
[*level
] = NULL
;
2995 * helper function for drop_subtree, this function is similar to
2996 * walk_down_tree. The main difference is that it checks reference
2997 * counts while tree blocks are locked.
2999 static int noinline
walk_down_subtree(struct btrfs_trans_handle
*trans
,
3000 struct btrfs_root
*root
,
3001 struct btrfs_path
*path
, int *level
)
3003 struct extent_buffer
*next
;
3004 struct extent_buffer
*cur
;
3005 struct extent_buffer
*parent
;
3012 cur
= path
->nodes
[*level
];
3013 ret
= btrfs_lookup_extent_ref(trans
, root
, cur
->start
, cur
->len
,
3019 while (*level
>= 0) {
3020 cur
= path
->nodes
[*level
];
3022 ret
= btrfs_drop_leaf_ref(trans
, root
, cur
);
3024 clean_tree_block(trans
, root
, cur
);
3027 if (path
->slots
[*level
] >= btrfs_header_nritems(cur
)) {
3028 clean_tree_block(trans
, root
, cur
);
3032 bytenr
= btrfs_node_blockptr(cur
, path
->slots
[*level
]);
3033 blocksize
= btrfs_level_size(root
, *level
- 1);
3034 ptr_gen
= btrfs_node_ptr_generation(cur
, path
->slots
[*level
]);
3036 next
= read_tree_block(root
, bytenr
, blocksize
, ptr_gen
);
3037 btrfs_tree_lock(next
);
3039 ret
= btrfs_lookup_extent_ref(trans
, root
, bytenr
, blocksize
,
3043 parent
= path
->nodes
[*level
];
3044 ret
= btrfs_free_extent(trans
, root
, bytenr
,
3045 blocksize
, parent
->start
,
3046 btrfs_header_owner(parent
),
3047 btrfs_header_generation(parent
),
3050 path
->slots
[*level
]++;
3051 btrfs_tree_unlock(next
);
3052 free_extent_buffer(next
);
3056 *level
= btrfs_header_level(next
);
3057 path
->nodes
[*level
] = next
;
3058 path
->slots
[*level
] = 0;
3059 path
->locks
[*level
] = 1;
3063 parent
= path
->nodes
[*level
+ 1];
3064 bytenr
= path
->nodes
[*level
]->start
;
3065 blocksize
= path
->nodes
[*level
]->len
;
3067 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
,
3068 parent
->start
, btrfs_header_owner(parent
),
3069 btrfs_header_generation(parent
), *level
, 1);
3072 if (path
->locks
[*level
]) {
3073 btrfs_tree_unlock(path
->nodes
[*level
]);
3074 path
->locks
[*level
] = 0;
3076 free_extent_buffer(path
->nodes
[*level
]);
3077 path
->nodes
[*level
] = NULL
;
3084 * helper for dropping snapshots. This walks back up the tree in the path
3085 * to find the first node higher up where we haven't yet gone through
3088 static int noinline
walk_up_tree(struct btrfs_trans_handle
*trans
,
3089 struct btrfs_root
*root
,
3090 struct btrfs_path
*path
,
3091 int *level
, int max_level
)
3095 struct btrfs_root_item
*root_item
= &root
->root_item
;
3100 for (i
= *level
; i
< max_level
&& path
->nodes
[i
]; i
++) {
3101 slot
= path
->slots
[i
];
3102 if (slot
< btrfs_header_nritems(path
->nodes
[i
]) - 1) {
3103 struct extent_buffer
*node
;
3104 struct btrfs_disk_key disk_key
;
3105 node
= path
->nodes
[i
];
3108 WARN_ON(*level
== 0);
3109 btrfs_node_key(node
, &disk_key
, path
->slots
[i
]);
3110 memcpy(&root_item
->drop_progress
,
3111 &disk_key
, sizeof(disk_key
));
3112 root_item
->drop_level
= i
;
3115 struct extent_buffer
*parent
;
3116 if (path
->nodes
[*level
] == root
->node
)
3117 parent
= path
->nodes
[*level
];
3119 parent
= path
->nodes
[*level
+ 1];
3121 root_owner
= btrfs_header_owner(parent
);
3122 root_gen
= btrfs_header_generation(parent
);
3124 clean_tree_block(trans
, root
, path
->nodes
[*level
]);
3125 ret
= btrfs_free_extent(trans
, root
,
3126 path
->nodes
[*level
]->start
,
3127 path
->nodes
[*level
]->len
,
3128 parent
->start
, root_owner
,
3129 root_gen
, *level
, 1);
3131 if (path
->locks
[*level
]) {
3132 btrfs_tree_unlock(path
->nodes
[*level
]);
3133 path
->locks
[*level
] = 0;
3135 free_extent_buffer(path
->nodes
[*level
]);
3136 path
->nodes
[*level
] = NULL
;
3144 * drop the reference count on the tree rooted at 'snap'. This traverses
3145 * the tree freeing any blocks that have a ref count of zero after being
3148 int btrfs_drop_snapshot(struct btrfs_trans_handle
*trans
, struct btrfs_root
3154 struct btrfs_path
*path
;
3157 struct btrfs_root_item
*root_item
= &root
->root_item
;
3159 WARN_ON(!mutex_is_locked(&root
->fs_info
->drop_mutex
));
3160 path
= btrfs_alloc_path();
3163 level
= btrfs_header_level(root
->node
);
3165 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
3166 path
->nodes
[level
] = root
->node
;
3167 extent_buffer_get(root
->node
);
3168 path
->slots
[level
] = 0;
3170 struct btrfs_key key
;
3171 struct btrfs_disk_key found_key
;
3172 struct extent_buffer
*node
;
3174 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
3175 level
= root_item
->drop_level
;
3176 path
->lowest_level
= level
;
3177 wret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3182 node
= path
->nodes
[level
];
3183 btrfs_node_key(node
, &found_key
, path
->slots
[level
]);
3184 WARN_ON(memcmp(&found_key
, &root_item
->drop_progress
,
3185 sizeof(found_key
)));
3187 * unlock our path, this is safe because only this
3188 * function is allowed to delete this snapshot
3190 for (i
= 0; i
< BTRFS_MAX_LEVEL
; i
++) {
3191 if (path
->nodes
[i
] && path
->locks
[i
]) {
3193 btrfs_tree_unlock(path
->nodes
[i
]);
3198 wret
= walk_down_tree(trans
, root
, path
, &level
);
3204 wret
= walk_up_tree(trans
, root
, path
, &level
,
3210 if (trans
->transaction
->in_commit
) {
3214 atomic_inc(&root
->fs_info
->throttle_gen
);
3215 wake_up(&root
->fs_info
->transaction_throttle
);
3217 for (i
= 0; i
<= orig_level
; i
++) {
3218 if (path
->nodes
[i
]) {
3219 free_extent_buffer(path
->nodes
[i
]);
3220 path
->nodes
[i
] = NULL
;
3224 btrfs_free_path(path
);
3228 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
3229 struct btrfs_root
*root
,
3230 struct extent_buffer
*node
,
3231 struct extent_buffer
*parent
)
3233 struct btrfs_path
*path
;
3239 path
= btrfs_alloc_path();
3242 BUG_ON(!btrfs_tree_locked(parent
));
3243 parent_level
= btrfs_header_level(parent
);
3244 extent_buffer_get(parent
);
3245 path
->nodes
[parent_level
] = parent
;
3246 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
3248 BUG_ON(!btrfs_tree_locked(node
));
3249 level
= btrfs_header_level(node
);
3250 extent_buffer_get(node
);
3251 path
->nodes
[level
] = node
;
3252 path
->slots
[level
] = 0;
3255 wret
= walk_down_subtree(trans
, root
, path
, &level
);
3261 wret
= walk_up_tree(trans
, root
, path
, &level
, parent_level
);
3268 btrfs_free_path(path
);
3272 static unsigned long calc_ra(unsigned long start
, unsigned long last
,
3275 return min(last
, start
+ nr
- 1);
3278 static int noinline
relocate_inode_pages(struct inode
*inode
, u64 start
,
3283 unsigned long first_index
;
3284 unsigned long last_index
;
3287 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3288 struct file_ra_state
*ra
;
3289 struct btrfs_ordered_extent
*ordered
;
3290 unsigned int total_read
= 0;
3291 unsigned int total_dirty
= 0;
3294 ra
= kzalloc(sizeof(*ra
), GFP_NOFS
);
3296 mutex_lock(&inode
->i_mutex
);
3297 first_index
= start
>> PAGE_CACHE_SHIFT
;
3298 last_index
= (start
+ len
- 1) >> PAGE_CACHE_SHIFT
;
3300 /* make sure the dirty trick played by the caller work */
3301 ret
= invalidate_inode_pages2_range(inode
->i_mapping
,
3302 first_index
, last_index
);
3306 file_ra_state_init(ra
, inode
->i_mapping
);
3308 for (i
= first_index
; i
<= last_index
; i
++) {
3309 if (total_read
% ra
->ra_pages
== 0) {
3310 btrfs_force_ra(inode
->i_mapping
, ra
, NULL
, i
,
3311 calc_ra(i
, last_index
, ra
->ra_pages
));
3315 if (((u64
)i
<< PAGE_CACHE_SHIFT
) > i_size_read(inode
))
3317 page
= grab_cache_page(inode
->i_mapping
, i
);
3322 if (!PageUptodate(page
)) {
3323 btrfs_readpage(NULL
, page
);
3325 if (!PageUptodate(page
)) {
3327 page_cache_release(page
);
3332 wait_on_page_writeback(page
);
3334 page_start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
3335 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
3336 lock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
3338 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
3340 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
3342 page_cache_release(page
);
3343 btrfs_start_ordered_extent(inode
, ordered
, 1);
3344 btrfs_put_ordered_extent(ordered
);
3347 set_page_extent_mapped(page
);
3349 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
3350 if (i
== first_index
)
3351 set_extent_bits(io_tree
, page_start
, page_end
,
3352 EXTENT_BOUNDARY
, GFP_NOFS
);
3354 set_page_dirty(page
);
3357 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
3359 page_cache_release(page
);
3364 mutex_unlock(&inode
->i_mutex
);
3365 balance_dirty_pages_ratelimited_nr(inode
->i_mapping
, total_dirty
);
3369 static int noinline
relocate_data_extent(struct inode
*reloc_inode
,
3370 struct btrfs_key
*extent_key
,
3373 struct btrfs_root
*root
= BTRFS_I(reloc_inode
)->root
;
3374 struct extent_map_tree
*em_tree
= &BTRFS_I(reloc_inode
)->extent_tree
;
3375 struct extent_map
*em
;
3377 em
= alloc_extent_map(GFP_NOFS
);
3378 BUG_ON(!em
|| IS_ERR(em
));
3380 em
->start
= extent_key
->objectid
- offset
;
3381 em
->len
= extent_key
->offset
;
3382 em
->block_len
= extent_key
->offset
;
3383 em
->block_start
= extent_key
->objectid
;
3384 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
3385 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
3387 /* setup extent map to cheat btrfs_readpage */
3388 mutex_lock(&BTRFS_I(reloc_inode
)->extent_mutex
);
3391 spin_lock(&em_tree
->lock
);
3392 ret
= add_extent_mapping(em_tree
, em
);
3393 spin_unlock(&em_tree
->lock
);
3394 if (ret
!= -EEXIST
) {
3395 free_extent_map(em
);
3398 btrfs_drop_extent_cache(reloc_inode
, em
->start
,
3399 em
->start
+ em
->len
- 1, 0);
3401 mutex_unlock(&BTRFS_I(reloc_inode
)->extent_mutex
);
3403 return relocate_inode_pages(reloc_inode
, extent_key
->objectid
- offset
,
3404 extent_key
->offset
);
3407 struct btrfs_ref_path
{
3409 u64 nodes
[BTRFS_MAX_LEVEL
];
3411 u64 root_generation
;
3418 struct btrfs_key node_keys
[BTRFS_MAX_LEVEL
];
3419 u64 new_nodes
[BTRFS_MAX_LEVEL
];
3422 struct disk_extent
{
3433 static int is_cowonly_root(u64 root_objectid
)
3435 if (root_objectid
== BTRFS_ROOT_TREE_OBJECTID
||
3436 root_objectid
== BTRFS_EXTENT_TREE_OBJECTID
||
3437 root_objectid
== BTRFS_CHUNK_TREE_OBJECTID
||
3438 root_objectid
== BTRFS_DEV_TREE_OBJECTID
||
3439 root_objectid
== BTRFS_TREE_LOG_OBJECTID
)
3444 static int noinline
__next_ref_path(struct btrfs_trans_handle
*trans
,
3445 struct btrfs_root
*extent_root
,
3446 struct btrfs_ref_path
*ref_path
,
3449 struct extent_buffer
*leaf
;
3450 struct btrfs_path
*path
;
3451 struct btrfs_extent_ref
*ref
;
3452 struct btrfs_key key
;
3453 struct btrfs_key found_key
;
3459 path
= btrfs_alloc_path();
3464 ref_path
->lowest_level
= -1;
3465 ref_path
->current_level
= -1;
3466 ref_path
->shared_level
= -1;
3470 level
= ref_path
->current_level
- 1;
3471 while (level
>= -1) {
3473 if (level
< ref_path
->lowest_level
)
3477 bytenr
= ref_path
->nodes
[level
];
3479 bytenr
= ref_path
->extent_start
;
3481 BUG_ON(bytenr
== 0);
3483 parent
= ref_path
->nodes
[level
+ 1];
3484 ref_path
->nodes
[level
+ 1] = 0;
3485 ref_path
->current_level
= level
;
3486 BUG_ON(parent
== 0);
3488 key
.objectid
= bytenr
;
3489 key
.offset
= parent
+ 1;
3490 key
.type
= BTRFS_EXTENT_REF_KEY
;
3492 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
, 0, 0);
3497 leaf
= path
->nodes
[0];
3498 nritems
= btrfs_header_nritems(leaf
);
3499 if (path
->slots
[0] >= nritems
) {
3500 ret
= btrfs_next_leaf(extent_root
, path
);
3505 leaf
= path
->nodes
[0];
3508 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3509 if (found_key
.objectid
== bytenr
&&
3510 found_key
.type
== BTRFS_EXTENT_REF_KEY
) {
3511 if (level
< ref_path
->shared_level
)
3512 ref_path
->shared_level
= level
;
3517 btrfs_release_path(extent_root
, path
);
3518 if (need_resched()) {
3522 /* reached lowest level */
3526 level
= ref_path
->current_level
;
3527 while (level
< BTRFS_MAX_LEVEL
- 1) {
3530 bytenr
= ref_path
->nodes
[level
];
3532 bytenr
= ref_path
->extent_start
;
3534 BUG_ON(bytenr
== 0);
3536 key
.objectid
= bytenr
;
3538 key
.type
= BTRFS_EXTENT_REF_KEY
;
3540 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
, 0, 0);
3544 leaf
= path
->nodes
[0];
3545 nritems
= btrfs_header_nritems(leaf
);
3546 if (path
->slots
[0] >= nritems
) {
3547 ret
= btrfs_next_leaf(extent_root
, path
);
3551 /* the extent was freed by someone */
3552 if (ref_path
->lowest_level
== level
)
3554 btrfs_release_path(extent_root
, path
);
3557 leaf
= path
->nodes
[0];
3560 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3561 if (found_key
.objectid
!= bytenr
||
3562 found_key
.type
!= BTRFS_EXTENT_REF_KEY
) {
3563 /* the extent was freed by someone */
3564 if (ref_path
->lowest_level
== level
) {
3568 btrfs_release_path(extent_root
, path
);
3572 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
3573 struct btrfs_extent_ref
);
3574 ref_objectid
= btrfs_ref_objectid(leaf
, ref
);
3575 if (ref_objectid
< BTRFS_FIRST_FREE_OBJECTID
) {
3577 level
= (int)ref_objectid
;
3578 BUG_ON(level
>= BTRFS_MAX_LEVEL
);
3579 ref_path
->lowest_level
= level
;
3580 ref_path
->current_level
= level
;
3581 ref_path
->nodes
[level
] = bytenr
;
3583 WARN_ON(ref_objectid
!= level
);
3586 WARN_ON(level
!= -1);
3590 if (ref_path
->lowest_level
== level
) {
3591 ref_path
->owner_objectid
= ref_objectid
;
3592 ref_path
->num_refs
= btrfs_ref_num_refs(leaf
, ref
);
3596 * the block is tree root or the block isn't in reference
3599 if (found_key
.objectid
== found_key
.offset
||
3600 is_cowonly_root(btrfs_ref_root(leaf
, ref
))) {
3601 ref_path
->root_objectid
= btrfs_ref_root(leaf
, ref
);
3602 ref_path
->root_generation
=
3603 btrfs_ref_generation(leaf
, ref
);
3605 /* special reference from the tree log */
3606 ref_path
->nodes
[0] = found_key
.offset
;
3607 ref_path
->current_level
= 0;
3614 BUG_ON(ref_path
->nodes
[level
] != 0);
3615 ref_path
->nodes
[level
] = found_key
.offset
;
3616 ref_path
->current_level
= level
;
3619 * the reference was created in the running transaction,
3620 * no need to continue walking up.
3622 if (btrfs_ref_generation(leaf
, ref
) == trans
->transid
) {
3623 ref_path
->root_objectid
= btrfs_ref_root(leaf
, ref
);
3624 ref_path
->root_generation
=
3625 btrfs_ref_generation(leaf
, ref
);
3630 btrfs_release_path(extent_root
, path
);
3631 if (need_resched()) {
3635 /* reached max tree level, but no tree root found. */
3638 btrfs_free_path(path
);
3642 static int btrfs_first_ref_path(struct btrfs_trans_handle
*trans
,
3643 struct btrfs_root
*extent_root
,
3644 struct btrfs_ref_path
*ref_path
,
3647 memset(ref_path
, 0, sizeof(*ref_path
));
3648 ref_path
->extent_start
= extent_start
;
3650 return __next_ref_path(trans
, extent_root
, ref_path
, 1);
3653 static int btrfs_next_ref_path(struct btrfs_trans_handle
*trans
,
3654 struct btrfs_root
*extent_root
,
3655 struct btrfs_ref_path
*ref_path
)
3657 return __next_ref_path(trans
, extent_root
, ref_path
, 0);
3660 static int noinline
get_new_locations(struct inode
*reloc_inode
,
3661 struct btrfs_key
*extent_key
,
3662 u64 offset
, int no_fragment
,
3663 struct disk_extent
**extents
,
3666 struct btrfs_root
*root
= BTRFS_I(reloc_inode
)->root
;
3667 struct btrfs_path
*path
;
3668 struct btrfs_file_extent_item
*fi
;
3669 struct extent_buffer
*leaf
;
3670 struct disk_extent
*exts
= *extents
;
3671 struct btrfs_key found_key
;
3676 int max
= *nr_extents
;
3679 WARN_ON(!no_fragment
&& *extents
);
3682 exts
= kmalloc(sizeof(*exts
) * max
, GFP_NOFS
);
3687 path
= btrfs_alloc_path();
3690 cur_pos
= extent_key
->objectid
- offset
;
3691 last_byte
= extent_key
->objectid
+ extent_key
->offset
;
3692 ret
= btrfs_lookup_file_extent(NULL
, root
, path
, reloc_inode
->i_ino
,
3702 leaf
= path
->nodes
[0];
3703 nritems
= btrfs_header_nritems(leaf
);
3704 if (path
->slots
[0] >= nritems
) {
3705 ret
= btrfs_next_leaf(root
, path
);
3710 leaf
= path
->nodes
[0];
3713 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3714 if (found_key
.offset
!= cur_pos
||
3715 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
||
3716 found_key
.objectid
!= reloc_inode
->i_ino
)
3719 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
3720 struct btrfs_file_extent_item
);
3721 if (btrfs_file_extent_type(leaf
, fi
) !=
3722 BTRFS_FILE_EXTENT_REG
||
3723 btrfs_file_extent_disk_bytenr(leaf
, fi
) == 0)
3727 struct disk_extent
*old
= exts
;
3729 exts
= kzalloc(sizeof(*exts
) * max
, GFP_NOFS
);
3730 memcpy(exts
, old
, sizeof(*exts
) * nr
);
3731 if (old
!= *extents
)
3735 exts
[nr
].disk_bytenr
=
3736 btrfs_file_extent_disk_bytenr(leaf
, fi
);
3737 exts
[nr
].disk_num_bytes
=
3738 btrfs_file_extent_disk_num_bytes(leaf
, fi
);
3739 exts
[nr
].offset
= btrfs_file_extent_offset(leaf
, fi
);
3740 exts
[nr
].num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
3741 exts
[nr
].ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
3742 exts
[nr
].compression
= btrfs_file_extent_compression(leaf
, fi
);
3743 exts
[nr
].encryption
= btrfs_file_extent_encryption(leaf
, fi
);
3744 exts
[nr
].other_encoding
= btrfs_file_extent_other_encoding(leaf
,
3746 WARN_ON(exts
[nr
].offset
> 0);
3747 WARN_ON(exts
[nr
].num_bytes
!= exts
[nr
].disk_num_bytes
);
3749 cur_pos
+= exts
[nr
].num_bytes
;
3752 if (cur_pos
+ offset
>= last_byte
)
3762 WARN_ON(cur_pos
+ offset
> last_byte
);
3763 if (cur_pos
+ offset
< last_byte
) {
3769 btrfs_free_path(path
);
3771 if (exts
!= *extents
)
3780 static int noinline
replace_one_extent(struct btrfs_trans_handle
*trans
,
3781 struct btrfs_root
*root
,
3782 struct btrfs_path
*path
,
3783 struct btrfs_key
*extent_key
,
3784 struct btrfs_key
*leaf_key
,
3785 struct btrfs_ref_path
*ref_path
,
3786 struct disk_extent
*new_extents
,
3789 struct extent_buffer
*leaf
;
3790 struct btrfs_file_extent_item
*fi
;
3791 struct inode
*inode
= NULL
;
3792 struct btrfs_key key
;
3800 int extent_locked
= 0;
3803 memcpy(&key
, leaf_key
, sizeof(key
));
3804 first_pos
= INT_LIMIT(loff_t
) - extent_key
->offset
;
3805 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
) {
3806 if (key
.objectid
< ref_path
->owner_objectid
||
3807 (key
.objectid
== ref_path
->owner_objectid
&&
3808 key
.type
< BTRFS_EXTENT_DATA_KEY
)) {
3809 key
.objectid
= ref_path
->owner_objectid
;
3810 key
.type
= BTRFS_EXTENT_DATA_KEY
;
3816 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
3820 leaf
= path
->nodes
[0];
3821 nritems
= btrfs_header_nritems(leaf
);
3823 if (extent_locked
&& ret
> 0) {
3825 * the file extent item was modified by someone
3826 * before the extent got locked.
3828 mutex_unlock(&BTRFS_I(inode
)->extent_mutex
);
3829 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
3830 lock_end
, GFP_NOFS
);
3834 if (path
->slots
[0] >= nritems
) {
3835 if (++nr_scaned
> 2)
3838 BUG_ON(extent_locked
);
3839 ret
= btrfs_next_leaf(root
, path
);
3844 leaf
= path
->nodes
[0];
3845 nritems
= btrfs_header_nritems(leaf
);
3848 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
3850 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
) {
3851 if ((key
.objectid
> ref_path
->owner_objectid
) ||
3852 (key
.objectid
== ref_path
->owner_objectid
&&
3853 key
.type
> BTRFS_EXTENT_DATA_KEY
) ||
3854 (key
.offset
>= first_pos
+ extent_key
->offset
))
3858 if (inode
&& key
.objectid
!= inode
->i_ino
) {
3859 BUG_ON(extent_locked
);
3860 btrfs_release_path(root
, path
);
3861 mutex_unlock(&inode
->i_mutex
);
3867 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
3872 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
3873 struct btrfs_file_extent_item
);
3874 if ((btrfs_file_extent_type(leaf
, fi
) !=
3875 BTRFS_FILE_EXTENT_REG
) ||
3876 (btrfs_file_extent_disk_bytenr(leaf
, fi
) !=
3877 extent_key
->objectid
)) {
3883 num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
3884 ext_offset
= btrfs_file_extent_offset(leaf
, fi
);
3886 if (first_pos
> key
.offset
- ext_offset
)
3887 first_pos
= key
.offset
- ext_offset
;
3889 if (!extent_locked
) {
3890 lock_start
= key
.offset
;
3891 lock_end
= lock_start
+ num_bytes
- 1;
3893 BUG_ON(lock_start
!= key
.offset
);
3894 BUG_ON(lock_end
- lock_start
+ 1 < num_bytes
);
3898 btrfs_release_path(root
, path
);
3900 inode
= btrfs_iget_locked(root
->fs_info
->sb
,
3901 key
.objectid
, root
);
3902 if (inode
->i_state
& I_NEW
) {
3903 BTRFS_I(inode
)->root
= root
;
3904 BTRFS_I(inode
)->location
.objectid
=
3906 BTRFS_I(inode
)->location
.type
=
3907 BTRFS_INODE_ITEM_KEY
;
3908 BTRFS_I(inode
)->location
.offset
= 0;
3909 btrfs_read_locked_inode(inode
);
3910 unlock_new_inode(inode
);
3913 * some code call btrfs_commit_transaction while
3914 * holding the i_mutex, so we can't use mutex_lock
3917 if (is_bad_inode(inode
) ||
3918 !mutex_trylock(&inode
->i_mutex
)) {
3921 key
.offset
= (u64
)-1;
3926 if (!extent_locked
) {
3927 struct btrfs_ordered_extent
*ordered
;
3929 btrfs_release_path(root
, path
);
3931 lock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
3932 lock_end
, GFP_NOFS
);
3933 ordered
= btrfs_lookup_first_ordered_extent(inode
,
3936 ordered
->file_offset
<= lock_end
&&
3937 ordered
->file_offset
+ ordered
->len
> lock_start
) {
3938 unlock_extent(&BTRFS_I(inode
)->io_tree
,
3939 lock_start
, lock_end
, GFP_NOFS
);
3940 btrfs_start_ordered_extent(inode
, ordered
, 1);
3941 btrfs_put_ordered_extent(ordered
);
3942 key
.offset
+= num_bytes
;
3946 btrfs_put_ordered_extent(ordered
);
3948 mutex_lock(&BTRFS_I(inode
)->extent_mutex
);
3953 if (nr_extents
== 1) {
3954 /* update extent pointer in place */
3955 btrfs_set_file_extent_generation(leaf
, fi
,
3957 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
3958 new_extents
[0].disk_bytenr
);
3959 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
3960 new_extents
[0].disk_num_bytes
);
3961 btrfs_set_file_extent_ram_bytes(leaf
, fi
,
3962 new_extents
[0].ram_bytes
);
3963 ext_offset
+= new_extents
[0].offset
;
3964 btrfs_set_file_extent_offset(leaf
, fi
, ext_offset
);
3965 btrfs_mark_buffer_dirty(leaf
);
3967 btrfs_drop_extent_cache(inode
, key
.offset
,
3968 key
.offset
+ num_bytes
- 1, 0);
3970 ret
= btrfs_inc_extent_ref(trans
, root
,
3971 new_extents
[0].disk_bytenr
,
3972 new_extents
[0].disk_num_bytes
,
3974 root
->root_key
.objectid
,
3979 ret
= btrfs_free_extent(trans
, root
,
3980 extent_key
->objectid
,
3983 btrfs_header_owner(leaf
),
3984 btrfs_header_generation(leaf
),
3988 btrfs_release_path(root
, path
);
3989 key
.offset
+= num_bytes
;
3995 * drop old extent pointer at first, then insert the
3996 * new pointers one bye one
3998 btrfs_release_path(root
, path
);
3999 ret
= btrfs_drop_extents(trans
, root
, inode
, key
.offset
,
4000 key
.offset
+ num_bytes
,
4001 key
.offset
, &alloc_hint
);
4004 for (i
= 0; i
< nr_extents
; i
++) {
4005 if (ext_offset
>= new_extents
[i
].num_bytes
) {
4006 ext_offset
-= new_extents
[i
].num_bytes
;
4009 extent_len
= min(new_extents
[i
].num_bytes
-
4010 ext_offset
, num_bytes
);
4012 ret
= btrfs_insert_empty_item(trans
, root
,
4017 leaf
= path
->nodes
[0];
4018 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
4019 struct btrfs_file_extent_item
);
4020 btrfs_set_file_extent_generation(leaf
, fi
,
4022 btrfs_set_file_extent_type(leaf
, fi
,
4023 BTRFS_FILE_EXTENT_REG
);
4024 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
4025 new_extents
[i
].disk_bytenr
);
4026 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
4027 new_extents
[i
].disk_num_bytes
);
4028 btrfs_set_file_extent_ram_bytes(leaf
, fi
,
4029 new_extents
[i
].ram_bytes
);
4031 btrfs_set_file_extent_compression(leaf
, fi
,
4032 new_extents
[i
].compression
);
4033 btrfs_set_file_extent_encryption(leaf
, fi
,
4034 new_extents
[i
].encryption
);
4035 btrfs_set_file_extent_other_encoding(leaf
, fi
,
4036 new_extents
[i
].other_encoding
);
4038 btrfs_set_file_extent_num_bytes(leaf
, fi
,
4040 ext_offset
+= new_extents
[i
].offset
;
4041 btrfs_set_file_extent_offset(leaf
, fi
,
4043 btrfs_mark_buffer_dirty(leaf
);
4045 btrfs_drop_extent_cache(inode
, key
.offset
,
4046 key
.offset
+ extent_len
- 1, 0);
4048 ret
= btrfs_inc_extent_ref(trans
, root
,
4049 new_extents
[i
].disk_bytenr
,
4050 new_extents
[i
].disk_num_bytes
,
4052 root
->root_key
.objectid
,
4053 trans
->transid
, key
.objectid
);
4055 btrfs_release_path(root
, path
);
4057 inode_add_bytes(inode
, extent_len
);
4060 num_bytes
-= extent_len
;
4061 key
.offset
+= extent_len
;
4066 BUG_ON(i
>= nr_extents
);
4069 if (extent_locked
) {
4070 mutex_unlock(&BTRFS_I(inode
)->extent_mutex
);
4071 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
4072 lock_end
, GFP_NOFS
);
4076 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
&&
4077 key
.offset
>= first_pos
+ extent_key
->offset
)
4084 btrfs_release_path(root
, path
);
4086 mutex_unlock(&inode
->i_mutex
);
4087 if (extent_locked
) {
4088 mutex_unlock(&BTRFS_I(inode
)->extent_mutex
);
4089 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
4090 lock_end
, GFP_NOFS
);
4097 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle
*trans
,
4098 struct btrfs_root
*root
,
4099 struct extent_buffer
*buf
, u64 orig_start
)
4104 BUG_ON(btrfs_header_generation(buf
) != trans
->transid
);
4105 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
4107 level
= btrfs_header_level(buf
);
4109 struct btrfs_leaf_ref
*ref
;
4110 struct btrfs_leaf_ref
*orig_ref
;
4112 orig_ref
= btrfs_lookup_leaf_ref(root
, orig_start
);
4116 ref
= btrfs_alloc_leaf_ref(root
, orig_ref
->nritems
);
4118 btrfs_free_leaf_ref(root
, orig_ref
);
4122 ref
->nritems
= orig_ref
->nritems
;
4123 memcpy(ref
->extents
, orig_ref
->extents
,
4124 sizeof(ref
->extents
[0]) * ref
->nritems
);
4126 btrfs_free_leaf_ref(root
, orig_ref
);
4128 ref
->root_gen
= trans
->transid
;
4129 ref
->bytenr
= buf
->start
;
4130 ref
->owner
= btrfs_header_owner(buf
);
4131 ref
->generation
= btrfs_header_generation(buf
);
4132 ret
= btrfs_add_leaf_ref(root
, ref
, 0);
4134 btrfs_free_leaf_ref(root
, ref
);
4139 static int noinline
invalidate_extent_cache(struct btrfs_root
*root
,
4140 struct extent_buffer
*leaf
,
4141 struct btrfs_block_group_cache
*group
,
4142 struct btrfs_root
*target_root
)
4144 struct btrfs_key key
;
4145 struct inode
*inode
= NULL
;
4146 struct btrfs_file_extent_item
*fi
;
4148 u64 skip_objectid
= 0;
4152 nritems
= btrfs_header_nritems(leaf
);
4153 for (i
= 0; i
< nritems
; i
++) {
4154 btrfs_item_key_to_cpu(leaf
, &key
, i
);
4155 if (key
.objectid
== skip_objectid
||
4156 key
.type
!= BTRFS_EXTENT_DATA_KEY
)
4158 fi
= btrfs_item_ptr(leaf
, i
, struct btrfs_file_extent_item
);
4159 if (btrfs_file_extent_type(leaf
, fi
) ==
4160 BTRFS_FILE_EXTENT_INLINE
)
4162 if (btrfs_file_extent_disk_bytenr(leaf
, fi
) == 0)
4164 if (!inode
|| inode
->i_ino
!= key
.objectid
) {
4166 inode
= btrfs_ilookup(target_root
->fs_info
->sb
,
4167 key
.objectid
, target_root
, 1);
4170 skip_objectid
= key
.objectid
;
4173 num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
4175 lock_extent(&BTRFS_I(inode
)->io_tree
, key
.offset
,
4176 key
.offset
+ num_bytes
- 1, GFP_NOFS
);
4177 mutex_lock(&BTRFS_I(inode
)->extent_mutex
);
4178 btrfs_drop_extent_cache(inode
, key
.offset
,
4179 key
.offset
+ num_bytes
- 1, 1);
4180 mutex_unlock(&BTRFS_I(inode
)->extent_mutex
);
4181 unlock_extent(&BTRFS_I(inode
)->io_tree
, key
.offset
,
4182 key
.offset
+ num_bytes
- 1, GFP_NOFS
);
4189 static int noinline
replace_extents_in_leaf(struct btrfs_trans_handle
*trans
,
4190 struct btrfs_root
*root
,
4191 struct extent_buffer
*leaf
,
4192 struct btrfs_block_group_cache
*group
,
4193 struct inode
*reloc_inode
)
4195 struct btrfs_key key
;
4196 struct btrfs_key extent_key
;
4197 struct btrfs_file_extent_item
*fi
;
4198 struct btrfs_leaf_ref
*ref
;
4199 struct disk_extent
*new_extent
;
4208 new_extent
= kmalloc(sizeof(*new_extent
), GFP_NOFS
);
4209 BUG_ON(!new_extent
);
4211 ref
= btrfs_lookup_leaf_ref(root
, leaf
->start
);
4215 nritems
= btrfs_header_nritems(leaf
);
4216 for (i
= 0; i
< nritems
; i
++) {
4217 btrfs_item_key_to_cpu(leaf
, &key
, i
);
4218 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
4220 fi
= btrfs_item_ptr(leaf
, i
, struct btrfs_file_extent_item
);
4221 if (btrfs_file_extent_type(leaf
, fi
) ==
4222 BTRFS_FILE_EXTENT_INLINE
)
4224 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
4225 num_bytes
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
4230 if (bytenr
>= group
->key
.objectid
+ group
->key
.offset
||
4231 bytenr
+ num_bytes
<= group
->key
.objectid
)
4234 extent_key
.objectid
= bytenr
;
4235 extent_key
.offset
= num_bytes
;
4236 extent_key
.type
= BTRFS_EXTENT_ITEM_KEY
;
4238 ret
= get_new_locations(reloc_inode
, &extent_key
,
4239 group
->key
.objectid
, 1,
4240 &new_extent
, &nr_extent
);
4245 BUG_ON(ref
->extents
[ext_index
].bytenr
!= bytenr
);
4246 BUG_ON(ref
->extents
[ext_index
].num_bytes
!= num_bytes
);
4247 ref
->extents
[ext_index
].bytenr
= new_extent
->disk_bytenr
;
4248 ref
->extents
[ext_index
].num_bytes
= new_extent
->disk_num_bytes
;
4250 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
4251 btrfs_set_file_extent_ram_bytes(leaf
, fi
,
4252 new_extent
->ram_bytes
);
4253 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
4254 new_extent
->disk_bytenr
);
4255 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
4256 new_extent
->disk_num_bytes
);
4257 new_extent
->offset
+= btrfs_file_extent_offset(leaf
, fi
);
4258 btrfs_set_file_extent_offset(leaf
, fi
, new_extent
->offset
);
4259 btrfs_mark_buffer_dirty(leaf
);
4261 ret
= btrfs_inc_extent_ref(trans
, root
,
4262 new_extent
->disk_bytenr
,
4263 new_extent
->disk_num_bytes
,
4265 root
->root_key
.objectid
,
4266 trans
->transid
, key
.objectid
);
4268 ret
= btrfs_free_extent(trans
, root
,
4269 bytenr
, num_bytes
, leaf
->start
,
4270 btrfs_header_owner(leaf
),
4271 btrfs_header_generation(leaf
),
4277 BUG_ON(ext_index
+ 1 != ref
->nritems
);
4278 btrfs_free_leaf_ref(root
, ref
);
4282 int btrfs_free_reloc_root(struct btrfs_trans_handle
*trans
,
4283 struct btrfs_root
*root
)
4285 struct btrfs_root
*reloc_root
;
4288 if (root
->reloc_root
) {
4289 reloc_root
= root
->reloc_root
;
4290 root
->reloc_root
= NULL
;
4291 list_add(&reloc_root
->dead_list
,
4292 &root
->fs_info
->dead_reloc_roots
);
4294 btrfs_set_root_bytenr(&reloc_root
->root_item
,
4295 reloc_root
->node
->start
);
4296 btrfs_set_root_level(&root
->root_item
,
4297 btrfs_header_level(reloc_root
->node
));
4298 memset(&reloc_root
->root_item
.drop_progress
, 0,
4299 sizeof(struct btrfs_disk_key
));
4300 reloc_root
->root_item
.drop_level
= 0;
4302 ret
= btrfs_update_root(trans
, root
->fs_info
->tree_root
,
4303 &reloc_root
->root_key
,
4304 &reloc_root
->root_item
);
4310 int btrfs_drop_dead_reloc_roots(struct btrfs_root
*root
)
4312 struct btrfs_trans_handle
*trans
;
4313 struct btrfs_root
*reloc_root
;
4314 struct btrfs_root
*prev_root
= NULL
;
4315 struct list_head dead_roots
;
4319 INIT_LIST_HEAD(&dead_roots
);
4320 list_splice_init(&root
->fs_info
->dead_reloc_roots
, &dead_roots
);
4322 while (!list_empty(&dead_roots
)) {
4323 reloc_root
= list_entry(dead_roots
.prev
,
4324 struct btrfs_root
, dead_list
);
4325 list_del_init(&reloc_root
->dead_list
);
4327 BUG_ON(reloc_root
->commit_root
!= NULL
);
4329 trans
= btrfs_join_transaction(root
, 1);
4332 mutex_lock(&root
->fs_info
->drop_mutex
);
4333 ret
= btrfs_drop_snapshot(trans
, reloc_root
);
4336 mutex_unlock(&root
->fs_info
->drop_mutex
);
4338 nr
= trans
->blocks_used
;
4339 ret
= btrfs_end_transaction(trans
, root
);
4341 btrfs_btree_balance_dirty(root
, nr
);
4344 free_extent_buffer(reloc_root
->node
);
4346 ret
= btrfs_del_root(trans
, root
->fs_info
->tree_root
,
4347 &reloc_root
->root_key
);
4349 mutex_unlock(&root
->fs_info
->drop_mutex
);
4351 nr
= trans
->blocks_used
;
4352 ret
= btrfs_end_transaction(trans
, root
);
4354 btrfs_btree_balance_dirty(root
, nr
);
4357 prev_root
= reloc_root
;
4360 btrfs_remove_leaf_refs(prev_root
, (u64
)-1, 0);
4366 int btrfs_add_dead_reloc_root(struct btrfs_root
*root
)
4368 list_add(&root
->dead_list
, &root
->fs_info
->dead_reloc_roots
);
4372 int btrfs_cleanup_reloc_trees(struct btrfs_root
*root
)
4374 struct btrfs_root
*reloc_root
;
4375 struct btrfs_trans_handle
*trans
;
4376 struct btrfs_key location
;
4380 mutex_lock(&root
->fs_info
->tree_reloc_mutex
);
4381 ret
= btrfs_find_dead_roots(root
, BTRFS_TREE_RELOC_OBJECTID
, NULL
);
4383 found
= !list_empty(&root
->fs_info
->dead_reloc_roots
);
4384 mutex_unlock(&root
->fs_info
->tree_reloc_mutex
);
4387 trans
= btrfs_start_transaction(root
, 1);
4389 ret
= btrfs_commit_transaction(trans
, root
);
4393 location
.objectid
= BTRFS_DATA_RELOC_TREE_OBJECTID
;
4394 location
.offset
= (u64
)-1;
4395 location
.type
= BTRFS_ROOT_ITEM_KEY
;
4397 reloc_root
= btrfs_read_fs_root_no_name(root
->fs_info
, &location
);
4398 BUG_ON(!reloc_root
);
4399 btrfs_orphan_cleanup(reloc_root
);
4403 static int noinline
init_reloc_tree(struct btrfs_trans_handle
*trans
,
4404 struct btrfs_root
*root
)
4406 struct btrfs_root
*reloc_root
;
4407 struct extent_buffer
*eb
;
4408 struct btrfs_root_item
*root_item
;
4409 struct btrfs_key root_key
;
4412 BUG_ON(!root
->ref_cows
);
4413 if (root
->reloc_root
)
4416 root_item
= kmalloc(sizeof(*root_item
), GFP_NOFS
);
4419 ret
= btrfs_copy_root(trans
, root
, root
->commit_root
,
4420 &eb
, BTRFS_TREE_RELOC_OBJECTID
);
4423 root_key
.objectid
= BTRFS_TREE_RELOC_OBJECTID
;
4424 root_key
.offset
= root
->root_key
.objectid
;
4425 root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
4427 memcpy(root_item
, &root
->root_item
, sizeof(root_item
));
4428 btrfs_set_root_refs(root_item
, 0);
4429 btrfs_set_root_bytenr(root_item
, eb
->start
);
4430 btrfs_set_root_level(root_item
, btrfs_header_level(eb
));
4431 btrfs_set_root_generation(root_item
, trans
->transid
);
4433 btrfs_tree_unlock(eb
);
4434 free_extent_buffer(eb
);
4436 ret
= btrfs_insert_root(trans
, root
->fs_info
->tree_root
,
4437 &root_key
, root_item
);
4441 reloc_root
= btrfs_read_fs_root_no_radix(root
->fs_info
->tree_root
,
4443 BUG_ON(!reloc_root
);
4444 reloc_root
->last_trans
= trans
->transid
;
4445 reloc_root
->commit_root
= NULL
;
4446 reloc_root
->ref_tree
= &root
->fs_info
->reloc_ref_tree
;
4448 root
->reloc_root
= reloc_root
;
4453 * Core function of space balance.
4455 * The idea is using reloc trees to relocate tree blocks in reference
4456 * counted roots. There is one reloc tree for each subvol, and all
4457 * reloc trees share same root key objectid. Reloc trees are snapshots
4458 * of the latest committed roots of subvols (root->commit_root).
4460 * To relocate a tree block referenced by a subvol, there are two steps.
4461 * COW the block through subvol's reloc tree, then update block pointer
4462 * in the subvol to point to the new block. Since all reloc trees share
4463 * same root key objectid, doing special handing for tree blocks owned
4464 * by them is easy. Once a tree block has been COWed in one reloc tree,
4465 * we can use the resulting new block directly when the same block is
4466 * required to COW again through other reloc trees. By this way, relocated
4467 * tree blocks are shared between reloc trees, so they are also shared
4470 static int noinline
relocate_one_path(struct btrfs_trans_handle
*trans
,
4471 struct btrfs_root
*root
,
4472 struct btrfs_path
*path
,
4473 struct btrfs_key
*first_key
,
4474 struct btrfs_ref_path
*ref_path
,
4475 struct btrfs_block_group_cache
*group
,
4476 struct inode
*reloc_inode
)
4478 struct btrfs_root
*reloc_root
;
4479 struct extent_buffer
*eb
= NULL
;
4480 struct btrfs_key
*keys
;
4484 int lowest_level
= 0;
4487 if (ref_path
->owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
)
4488 lowest_level
= ref_path
->owner_objectid
;
4490 if (!root
->ref_cows
) {
4491 path
->lowest_level
= lowest_level
;
4492 ret
= btrfs_search_slot(trans
, root
, first_key
, path
, 0, 1);
4494 path
->lowest_level
= 0;
4495 btrfs_release_path(root
, path
);
4499 mutex_lock(&root
->fs_info
->tree_reloc_mutex
);
4500 ret
= init_reloc_tree(trans
, root
);
4502 reloc_root
= root
->reloc_root
;
4504 shared_level
= ref_path
->shared_level
;
4505 ref_path
->shared_level
= BTRFS_MAX_LEVEL
- 1;
4507 keys
= ref_path
->node_keys
;
4508 nodes
= ref_path
->new_nodes
;
4509 memset(&keys
[shared_level
+ 1], 0,
4510 sizeof(*keys
) * (BTRFS_MAX_LEVEL
- shared_level
- 1));
4511 memset(&nodes
[shared_level
+ 1], 0,
4512 sizeof(*nodes
) * (BTRFS_MAX_LEVEL
- shared_level
- 1));
4514 if (nodes
[lowest_level
] == 0) {
4515 path
->lowest_level
= lowest_level
;
4516 ret
= btrfs_search_slot(trans
, reloc_root
, first_key
, path
,
4519 for (level
= lowest_level
; level
< BTRFS_MAX_LEVEL
; level
++) {
4520 eb
= path
->nodes
[level
];
4521 if (!eb
|| eb
== reloc_root
->node
)
4523 nodes
[level
] = eb
->start
;
4525 btrfs_item_key_to_cpu(eb
, &keys
[level
], 0);
4527 btrfs_node_key_to_cpu(eb
, &keys
[level
], 0);
4529 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
4530 eb
= path
->nodes
[0];
4531 ret
= replace_extents_in_leaf(trans
, reloc_root
, eb
,
4532 group
, reloc_inode
);
4535 btrfs_release_path(reloc_root
, path
);
4537 ret
= btrfs_merge_path(trans
, reloc_root
, keys
, nodes
,
4543 * replace tree blocks in the fs tree with tree blocks in
4546 ret
= btrfs_merge_path(trans
, root
, keys
, nodes
, lowest_level
);
4549 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
4550 ret
= btrfs_search_slot(trans
, reloc_root
, first_key
, path
,
4553 extent_buffer_get(path
->nodes
[0]);
4554 eb
= path
->nodes
[0];
4555 btrfs_release_path(reloc_root
, path
);
4556 ret
= invalidate_extent_cache(reloc_root
, eb
, group
, root
);
4558 free_extent_buffer(eb
);
4561 mutex_unlock(&root
->fs_info
->tree_reloc_mutex
);
4562 path
->lowest_level
= 0;
4566 static int noinline
relocate_tree_block(struct btrfs_trans_handle
*trans
,
4567 struct btrfs_root
*root
,
4568 struct btrfs_path
*path
,
4569 struct btrfs_key
*first_key
,
4570 struct btrfs_ref_path
*ref_path
)
4574 ret
= relocate_one_path(trans
, root
, path
, first_key
,
4575 ref_path
, NULL
, NULL
);
4578 if (root
== root
->fs_info
->extent_root
)
4579 btrfs_extent_post_op(trans
, root
);
4584 static int noinline
del_extent_zero(struct btrfs_trans_handle
*trans
,
4585 struct btrfs_root
*extent_root
,
4586 struct btrfs_path
*path
,
4587 struct btrfs_key
*extent_key
)
4591 ret
= btrfs_search_slot(trans
, extent_root
, extent_key
, path
, -1, 1);
4594 ret
= btrfs_del_item(trans
, extent_root
, path
);
4596 btrfs_release_path(extent_root
, path
);
4600 static struct btrfs_root noinline
*read_ref_root(struct btrfs_fs_info
*fs_info
,
4601 struct btrfs_ref_path
*ref_path
)
4603 struct btrfs_key root_key
;
4605 root_key
.objectid
= ref_path
->root_objectid
;
4606 root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
4607 if (is_cowonly_root(ref_path
->root_objectid
))
4608 root_key
.offset
= 0;
4610 root_key
.offset
= (u64
)-1;
4612 return btrfs_read_fs_root_no_name(fs_info
, &root_key
);
4615 static int noinline
relocate_one_extent(struct btrfs_root
*extent_root
,
4616 struct btrfs_path
*path
,
4617 struct btrfs_key
*extent_key
,
4618 struct btrfs_block_group_cache
*group
,
4619 struct inode
*reloc_inode
, int pass
)
4621 struct btrfs_trans_handle
*trans
;
4622 struct btrfs_root
*found_root
;
4623 struct btrfs_ref_path
*ref_path
= NULL
;
4624 struct disk_extent
*new_extents
= NULL
;
4629 struct btrfs_key first_key
;
4633 trans
= btrfs_start_transaction(extent_root
, 1);
4636 if (extent_key
->objectid
== 0) {
4637 ret
= del_extent_zero(trans
, extent_root
, path
, extent_key
);
4641 ref_path
= kmalloc(sizeof(*ref_path
), GFP_NOFS
);
4647 for (loops
= 0; ; loops
++) {
4649 ret
= btrfs_first_ref_path(trans
, extent_root
, ref_path
,
4650 extent_key
->objectid
);
4652 ret
= btrfs_next_ref_path(trans
, extent_root
, ref_path
);
4659 if (ref_path
->root_objectid
== BTRFS_TREE_LOG_OBJECTID
||
4660 ref_path
->root_objectid
== BTRFS_TREE_RELOC_OBJECTID
)
4663 found_root
= read_ref_root(extent_root
->fs_info
, ref_path
);
4664 BUG_ON(!found_root
);
4666 * for reference counted tree, only process reference paths
4667 * rooted at the latest committed root.
4669 if (found_root
->ref_cows
&&
4670 ref_path
->root_generation
!= found_root
->root_key
.offset
)
4673 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
4676 * copy data extents to new locations
4678 u64 group_start
= group
->key
.objectid
;
4679 ret
= relocate_data_extent(reloc_inode
,
4688 level
= ref_path
->owner_objectid
;
4691 if (prev_block
!= ref_path
->nodes
[level
]) {
4692 struct extent_buffer
*eb
;
4693 u64 block_start
= ref_path
->nodes
[level
];
4694 u64 block_size
= btrfs_level_size(found_root
, level
);
4696 eb
= read_tree_block(found_root
, block_start
,
4698 btrfs_tree_lock(eb
);
4699 BUG_ON(level
!= btrfs_header_level(eb
));
4702 btrfs_item_key_to_cpu(eb
, &first_key
, 0);
4704 btrfs_node_key_to_cpu(eb
, &first_key
, 0);
4706 btrfs_tree_unlock(eb
);
4707 free_extent_buffer(eb
);
4708 prev_block
= block_start
;
4711 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
&&
4714 * use fallback method to process the remaining
4718 u64 group_start
= group
->key
.objectid
;
4719 ret
= get_new_locations(reloc_inode
,
4727 btrfs_record_root_in_trans(found_root
);
4728 ret
= replace_one_extent(trans
, found_root
,
4730 &first_key
, ref_path
,
4731 new_extents
, nr_extents
);
4737 btrfs_record_root_in_trans(found_root
);
4738 if (ref_path
->owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
) {
4739 ret
= relocate_tree_block(trans
, found_root
, path
,
4740 &first_key
, ref_path
);
4743 * try to update data extent references while
4744 * keeping metadata shared between snapshots.
4746 ret
= relocate_one_path(trans
, found_root
, path
,
4747 &first_key
, ref_path
,
4748 group
, reloc_inode
);
4755 btrfs_end_transaction(trans
, extent_root
);
4761 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
4764 u64 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
4765 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
4767 num_devices
= root
->fs_info
->fs_devices
->num_devices
;
4768 if (num_devices
== 1) {
4769 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
4770 stripped
= flags
& ~stripped
;
4772 /* turn raid0 into single device chunks */
4773 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
4776 /* turn mirroring into duplication */
4777 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
4778 BTRFS_BLOCK_GROUP_RAID10
))
4779 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
4782 /* they already had raid on here, just return */
4783 if (flags
& stripped
)
4786 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
4787 stripped
= flags
& ~stripped
;
4789 /* switch duplicated blocks with raid1 */
4790 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
4791 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
4793 /* turn single device chunks into raid0 */
4794 return stripped
| BTRFS_BLOCK_GROUP_RAID0
;
4799 int __alloc_chunk_for_shrink(struct btrfs_root
*root
,
4800 struct btrfs_block_group_cache
*shrink_block_group
,
4803 struct btrfs_trans_handle
*trans
;
4804 u64 new_alloc_flags
;
4807 spin_lock(&shrink_block_group
->lock
);
4808 if (btrfs_block_group_used(&shrink_block_group
->item
) > 0) {
4809 spin_unlock(&shrink_block_group
->lock
);
4811 trans
= btrfs_start_transaction(root
, 1);
4812 spin_lock(&shrink_block_group
->lock
);
4814 new_alloc_flags
= update_block_group_flags(root
,
4815 shrink_block_group
->flags
);
4816 if (new_alloc_flags
!= shrink_block_group
->flags
) {
4818 btrfs_block_group_used(&shrink_block_group
->item
);
4820 calc
= shrink_block_group
->key
.offset
;
4822 spin_unlock(&shrink_block_group
->lock
);
4824 do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
4825 calc
+ 2 * 1024 * 1024, new_alloc_flags
, force
);
4827 btrfs_end_transaction(trans
, root
);
4829 spin_unlock(&shrink_block_group
->lock
);
4833 static int __insert_orphan_inode(struct btrfs_trans_handle
*trans
,
4834 struct btrfs_root
*root
,
4835 u64 objectid
, u64 size
)
4837 struct btrfs_path
*path
;
4838 struct btrfs_inode_item
*item
;
4839 struct extent_buffer
*leaf
;
4842 path
= btrfs_alloc_path();
4846 ret
= btrfs_insert_empty_inode(trans
, root
, path
, objectid
);
4850 leaf
= path
->nodes
[0];
4851 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_inode_item
);
4852 memset_extent_buffer(leaf
, 0, (unsigned long)item
, sizeof(*item
));
4853 btrfs_set_inode_generation(leaf
, item
, 1);
4854 btrfs_set_inode_size(leaf
, item
, size
);
4855 btrfs_set_inode_mode(leaf
, item
, S_IFREG
| 0600);
4856 btrfs_set_inode_flags(leaf
, item
, BTRFS_INODE_NODATASUM
);
4857 btrfs_mark_buffer_dirty(leaf
);
4858 btrfs_release_path(root
, path
);
4860 btrfs_free_path(path
);
4864 static struct inode noinline
*create_reloc_inode(struct btrfs_fs_info
*fs_info
,
4865 struct btrfs_block_group_cache
*group
)
4867 struct inode
*inode
= NULL
;
4868 struct btrfs_trans_handle
*trans
;
4869 struct btrfs_root
*root
;
4870 struct btrfs_key root_key
;
4871 u64 objectid
= BTRFS_FIRST_FREE_OBJECTID
;
4874 root_key
.objectid
= BTRFS_DATA_RELOC_TREE_OBJECTID
;
4875 root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
4876 root_key
.offset
= (u64
)-1;
4877 root
= btrfs_read_fs_root_no_name(fs_info
, &root_key
);
4879 return ERR_CAST(root
);
4881 trans
= btrfs_start_transaction(root
, 1);
4884 err
= btrfs_find_free_objectid(trans
, root
, objectid
, &objectid
);
4888 err
= __insert_orphan_inode(trans
, root
, objectid
, group
->key
.offset
);
4891 err
= btrfs_insert_file_extent(trans
, root
, objectid
, 0, 0, 0,
4892 group
->key
.offset
, 0, group
->key
.offset
,
4896 inode
= btrfs_iget_locked(root
->fs_info
->sb
, objectid
, root
);
4897 if (inode
->i_state
& I_NEW
) {
4898 BTRFS_I(inode
)->root
= root
;
4899 BTRFS_I(inode
)->location
.objectid
= objectid
;
4900 BTRFS_I(inode
)->location
.type
= BTRFS_INODE_ITEM_KEY
;
4901 BTRFS_I(inode
)->location
.offset
= 0;
4902 btrfs_read_locked_inode(inode
);
4903 unlock_new_inode(inode
);
4904 BUG_ON(is_bad_inode(inode
));
4909 err
= btrfs_orphan_add(trans
, inode
);
4911 btrfs_end_transaction(trans
, root
);
4915 inode
= ERR_PTR(err
);
4920 int btrfs_relocate_block_group(struct btrfs_root
*root
, u64 group_start
)
4922 struct btrfs_trans_handle
*trans
;
4923 struct btrfs_path
*path
;
4924 struct btrfs_fs_info
*info
= root
->fs_info
;
4925 struct extent_buffer
*leaf
;
4926 struct inode
*reloc_inode
;
4927 struct btrfs_block_group_cache
*block_group
;
4928 struct btrfs_key key
;
4936 root
= root
->fs_info
->extent_root
;
4938 block_group
= btrfs_lookup_block_group(info
, group_start
);
4939 BUG_ON(!block_group
);
4941 printk("btrfs relocating block group %llu flags %llu\n",
4942 (unsigned long long)block_group
->key
.objectid
,
4943 (unsigned long long)block_group
->flags
);
4945 path
= btrfs_alloc_path();
4948 reloc_inode
= create_reloc_inode(info
, block_group
);
4949 BUG_ON(IS_ERR(reloc_inode
));
4951 __alloc_chunk_for_shrink(root
, block_group
, 1);
4952 block_group
->ro
= 1;
4953 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
4955 btrfs_start_delalloc_inodes(info
->tree_root
);
4956 btrfs_wait_ordered_extents(info
->tree_root
, 0);
4960 key
.objectid
= block_group
->key
.objectid
;
4963 cur_byte
= key
.objectid
;
4965 trans
= btrfs_start_transaction(info
->tree_root
, 1);
4966 btrfs_commit_transaction(trans
, info
->tree_root
);
4968 mutex_lock(&root
->fs_info
->cleaner_mutex
);
4969 btrfs_clean_old_snapshots(info
->tree_root
);
4970 btrfs_remove_leaf_refs(info
->tree_root
, (u64
)-1, 1);
4971 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
4974 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4978 leaf
= path
->nodes
[0];
4979 nritems
= btrfs_header_nritems(leaf
);
4980 if (path
->slots
[0] >= nritems
) {
4981 ret
= btrfs_next_leaf(root
, path
);
4988 leaf
= path
->nodes
[0];
4989 nritems
= btrfs_header_nritems(leaf
);
4992 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
4994 if (key
.objectid
>= block_group
->key
.objectid
+
4995 block_group
->key
.offset
)
4998 if (progress
&& need_resched()) {
4999 btrfs_release_path(root
, path
);
5006 if (btrfs_key_type(&key
) != BTRFS_EXTENT_ITEM_KEY
||
5007 key
.objectid
+ key
.offset
<= cur_byte
) {
5013 cur_byte
= key
.objectid
+ key
.offset
;
5014 btrfs_release_path(root
, path
);
5016 __alloc_chunk_for_shrink(root
, block_group
, 0);
5017 ret
= relocate_one_extent(root
, path
, &key
, block_group
,
5021 key
.objectid
= cur_byte
;
5026 btrfs_release_path(root
, path
);
5029 btrfs_wait_ordered_range(reloc_inode
, 0, (u64
)-1);
5030 invalidate_mapping_pages(reloc_inode
->i_mapping
, 0, -1);
5031 WARN_ON(reloc_inode
->i_mapping
->nrpages
);
5034 if (total_found
> 0) {
5035 printk("btrfs found %llu extents in pass %d\n",
5036 (unsigned long long)total_found
, pass
);
5041 /* delete reloc_inode */
5044 /* unpin extents in this range */
5045 trans
= btrfs_start_transaction(info
->tree_root
, 1);
5046 btrfs_commit_transaction(trans
, info
->tree_root
);
5048 spin_lock(&block_group
->lock
);
5049 WARN_ON(block_group
->pinned
> 0);
5050 WARN_ON(block_group
->reserved
> 0);
5051 WARN_ON(btrfs_block_group_used(&block_group
->item
) > 0);
5052 spin_unlock(&block_group
->lock
);
5055 btrfs_free_path(path
);
5059 int find_first_block_group(struct btrfs_root
*root
, struct btrfs_path
*path
,
5060 struct btrfs_key
*key
)
5063 struct btrfs_key found_key
;
5064 struct extent_buffer
*leaf
;
5067 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
5072 slot
= path
->slots
[0];
5073 leaf
= path
->nodes
[0];
5074 if (slot
>= btrfs_header_nritems(leaf
)) {
5075 ret
= btrfs_next_leaf(root
, path
);
5082 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
5084 if (found_key
.objectid
>= key
->objectid
&&
5085 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
5096 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
5098 struct btrfs_block_group_cache
*block_group
;
5101 spin_lock(&info
->block_group_cache_lock
);
5102 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
5103 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
5106 spin_unlock(&info
->block_group_cache_lock
);
5107 btrfs_remove_free_space_cache(block_group
);
5108 spin_lock(&info
->block_group_cache_lock
);
5110 rb_erase(&block_group
->cache_node
,
5111 &info
->block_group_cache_tree
);
5112 down_write(&block_group
->space_info
->groups_sem
);
5113 list_del(&block_group
->list
);
5114 up_write(&block_group
->space_info
->groups_sem
);
5117 spin_unlock(&info
->block_group_cache_lock
);
5121 int btrfs_read_block_groups(struct btrfs_root
*root
)
5123 struct btrfs_path
*path
;
5125 struct btrfs_block_group_cache
*cache
;
5126 struct btrfs_fs_info
*info
= root
->fs_info
;
5127 struct btrfs_space_info
*space_info
;
5128 struct btrfs_key key
;
5129 struct btrfs_key found_key
;
5130 struct extent_buffer
*leaf
;
5132 root
= info
->extent_root
;
5135 btrfs_set_key_type(&key
, BTRFS_BLOCK_GROUP_ITEM_KEY
);
5136 path
= btrfs_alloc_path();
5141 ret
= find_first_block_group(root
, path
, &key
);
5149 leaf
= path
->nodes
[0];
5150 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5151 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
5157 spin_lock_init(&cache
->lock
);
5158 mutex_init(&cache
->alloc_mutex
);
5159 INIT_LIST_HEAD(&cache
->list
);
5160 read_extent_buffer(leaf
, &cache
->item
,
5161 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
5162 sizeof(cache
->item
));
5163 memcpy(&cache
->key
, &found_key
, sizeof(found_key
));
5165 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
5166 btrfs_release_path(root
, path
);
5167 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
5169 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
5170 btrfs_block_group_used(&cache
->item
),
5173 cache
->space_info
= space_info
;
5174 down_write(&space_info
->groups_sem
);
5175 list_add_tail(&cache
->list
, &space_info
->block_groups
);
5176 up_write(&space_info
->groups_sem
);
5178 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
5181 set_avail_alloc_bits(root
->fs_info
, cache
->flags
);
5185 btrfs_free_path(path
);
5189 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
5190 struct btrfs_root
*root
, u64 bytes_used
,
5191 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
5195 struct btrfs_root
*extent_root
;
5196 struct btrfs_block_group_cache
*cache
;
5198 extent_root
= root
->fs_info
->extent_root
;
5200 root
->fs_info
->last_trans_new_blockgroup
= trans
->transid
;
5202 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
5206 cache
->key
.objectid
= chunk_offset
;
5207 cache
->key
.offset
= size
;
5208 spin_lock_init(&cache
->lock
);
5209 mutex_init(&cache
->alloc_mutex
);
5210 INIT_LIST_HEAD(&cache
->list
);
5211 btrfs_set_key_type(&cache
->key
, BTRFS_BLOCK_GROUP_ITEM_KEY
);
5213 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
5214 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
5215 cache
->flags
= type
;
5216 btrfs_set_block_group_flags(&cache
->item
, type
);
5218 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
5219 &cache
->space_info
);
5221 down_write(&cache
->space_info
->groups_sem
);
5222 list_add_tail(&cache
->list
, &cache
->space_info
->block_groups
);
5223 up_write(&cache
->space_info
->groups_sem
);
5225 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
5228 ret
= btrfs_insert_item(trans
, extent_root
, &cache
->key
, &cache
->item
,
5229 sizeof(cache
->item
));
5232 finish_current_insert(trans
, extent_root
);
5233 ret
= del_pending_extents(trans
, extent_root
);
5235 set_avail_alloc_bits(extent_root
->fs_info
, type
);
5240 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
5241 struct btrfs_root
*root
, u64 group_start
)
5243 struct btrfs_path
*path
;
5244 struct btrfs_block_group_cache
*block_group
;
5245 struct btrfs_key key
;
5248 root
= root
->fs_info
->extent_root
;
5250 block_group
= btrfs_lookup_block_group(root
->fs_info
, group_start
);
5251 BUG_ON(!block_group
);
5253 memcpy(&key
, &block_group
->key
, sizeof(key
));
5255 path
= btrfs_alloc_path();
5258 btrfs_remove_free_space_cache(block_group
);
5259 rb_erase(&block_group
->cache_node
,
5260 &root
->fs_info
->block_group_cache_tree
);
5261 down_write(&block_group
->space_info
->groups_sem
);
5262 list_del(&block_group
->list
);
5263 up_write(&block_group
->space_info
->groups_sem
);
5266 memset(shrink_block_group, 0, sizeof(*shrink_block_group));
5267 kfree(shrink_block_group);
5270 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
5276 ret
= btrfs_del_item(trans
, root
, path
);
5278 btrfs_free_path(path
);