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 void maybe_lock_mutex(struct btrfs_root
*root
)
58 if (root
!= root
->fs_info
->extent_root
&&
59 root
!= root
->fs_info
->chunk_root
&&
60 root
!= root
->fs_info
->dev_root
) {
61 mutex_lock(&root
->fs_info
->alloc_mutex
);
65 void maybe_unlock_mutex(struct btrfs_root
*root
)
67 if (root
!= root
->fs_info
->extent_root
&&
68 root
!= root
->fs_info
->chunk_root
&&
69 root
!= root
->fs_info
->dev_root
) {
70 mutex_unlock(&root
->fs_info
->alloc_mutex
);
74 static int block_group_bits(struct btrfs_block_group_cache
*cache
, u64 bits
)
76 return (cache
->flags
& bits
) == bits
;
80 * this adds the block group to the fs_info rb tree for the block group
83 int btrfs_add_block_group_cache(struct btrfs_fs_info
*info
,
84 struct btrfs_block_group_cache
*block_group
)
87 struct rb_node
*parent
= NULL
;
88 struct btrfs_block_group_cache
*cache
;
90 spin_lock(&info
->block_group_cache_lock
);
91 p
= &info
->block_group_cache_tree
.rb_node
;
95 cache
= rb_entry(parent
, struct btrfs_block_group_cache
,
97 if (block_group
->key
.objectid
< cache
->key
.objectid
) {
99 } else if (block_group
->key
.objectid
> cache
->key
.objectid
) {
102 spin_unlock(&info
->block_group_cache_lock
);
107 rb_link_node(&block_group
->cache_node
, parent
, p
);
108 rb_insert_color(&block_group
->cache_node
,
109 &info
->block_group_cache_tree
);
110 spin_unlock(&info
->block_group_cache_lock
);
116 * This will return the block group at or after bytenr if contains is 0, else
117 * it will return the block group that contains the bytenr
119 static struct btrfs_block_group_cache
*
120 block_group_cache_tree_search(struct btrfs_fs_info
*info
, u64 bytenr
,
123 struct btrfs_block_group_cache
*cache
, *ret
= NULL
;
127 spin_lock(&info
->block_group_cache_lock
);
128 n
= info
->block_group_cache_tree
.rb_node
;
131 cache
= rb_entry(n
, struct btrfs_block_group_cache
,
133 end
= cache
->key
.objectid
+ cache
->key
.offset
- 1;
134 start
= cache
->key
.objectid
;
136 if (bytenr
< start
) {
137 if (!contains
&& (!ret
|| start
< ret
->key
.objectid
))
140 } else if (bytenr
> start
) {
141 if (contains
&& bytenr
<= end
) {
151 spin_unlock(&info
->block_group_cache_lock
);
157 * this is only called by cache_block_group, since we could have freed extents
158 * we need to check the pinned_extents for any extents that can't be used yet
159 * since their free space will be released as soon as the transaction commits.
161 static int add_new_free_space(struct btrfs_block_group_cache
*block_group
,
162 struct btrfs_fs_info
*info
, u64 start
, u64 end
)
164 u64 extent_start
, extent_end
, size
;
167 while (start
< end
) {
168 ret
= find_first_extent_bit(&info
->pinned_extents
, start
,
169 &extent_start
, &extent_end
,
174 if (extent_start
== start
) {
175 start
= extent_end
+ 1;
176 } else if (extent_start
> start
&& extent_start
< end
) {
177 size
= extent_start
- start
;
178 ret
= btrfs_add_free_space(block_group
, start
, size
);
180 start
= extent_end
+ 1;
188 ret
= btrfs_add_free_space(block_group
, start
, size
);
195 static int cache_block_group(struct btrfs_root
*root
,
196 struct btrfs_block_group_cache
*block_group
)
198 struct btrfs_path
*path
;
200 struct btrfs_key key
;
201 struct extent_buffer
*leaf
;
210 root
= root
->fs_info
->extent_root
;
212 if (block_group
->cached
)
215 path
= btrfs_alloc_path();
221 * we get into deadlocks with paths held by callers of this function.
222 * since the alloc_mutex is protecting things right now, just
223 * skip the locking here
225 path
->skip_locking
= 1;
226 first_free
= max_t(u64
, block_group
->key
.objectid
,
227 BTRFS_SUPER_INFO_OFFSET
+ BTRFS_SUPER_INFO_SIZE
);
228 key
.objectid
= block_group
->key
.objectid
;
230 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
231 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
234 ret
= btrfs_previous_item(root
, path
, 0, BTRFS_EXTENT_ITEM_KEY
);
238 leaf
= path
->nodes
[0];
239 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
240 if (key
.objectid
+ key
.offset
> first_free
)
241 first_free
= key
.objectid
+ key
.offset
;
244 leaf
= path
->nodes
[0];
245 slot
= path
->slots
[0];
246 if (slot
>= btrfs_header_nritems(leaf
)) {
247 ret
= btrfs_next_leaf(root
, path
);
255 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
256 if (key
.objectid
< block_group
->key
.objectid
)
259 if (key
.objectid
>= block_group
->key
.objectid
+
260 block_group
->key
.offset
)
263 if (btrfs_key_type(&key
) == BTRFS_EXTENT_ITEM_KEY
) {
269 add_new_free_space(block_group
, root
->fs_info
, last
,
272 last
= key
.objectid
+ key
.offset
;
281 add_new_free_space(block_group
, root
->fs_info
, last
,
282 block_group
->key
.objectid
+
283 block_group
->key
.offset
);
285 block_group
->cached
= 1;
288 btrfs_free_path(path
);
293 * return the block group that starts at or after bytenr
295 struct btrfs_block_group_cache
*btrfs_lookup_first_block_group(struct
299 struct btrfs_block_group_cache
*cache
;
301 cache
= block_group_cache_tree_search(info
, bytenr
, 0);
307 * return the block group that contains teh given bytenr
309 struct btrfs_block_group_cache
*btrfs_lookup_block_group(struct
313 struct btrfs_block_group_cache
*cache
;
315 cache
= block_group_cache_tree_search(info
, bytenr
, 1);
320 static int noinline
find_free_space(struct btrfs_root
*root
,
321 struct btrfs_block_group_cache
**cache_ret
,
322 u64
*start_ret
, u64 num
, int data
)
325 struct btrfs_block_group_cache
*cache
= *cache_ret
;
326 struct btrfs_free_space
*info
= NULL
;
328 u64 search_start
= *start_ret
;
330 WARN_ON(!mutex_is_locked(&root
->fs_info
->alloc_mutex
));
334 last
= max(search_start
, cache
->key
.objectid
);
337 ret
= cache_block_group(root
, cache
);
341 if (cache
->ro
|| !block_group_bits(cache
, data
))
344 info
= btrfs_find_free_space(cache
, last
, num
);
346 *start_ret
= info
->offset
;
351 last
= cache
->key
.objectid
+ cache
->key
.offset
;
353 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
364 static u64
div_factor(u64 num
, int factor
)
373 static struct btrfs_space_info
*__find_space_info(struct btrfs_fs_info
*info
,
376 struct list_head
*head
= &info
->space_info
;
377 struct list_head
*cur
;
378 struct btrfs_space_info
*found
;
379 list_for_each(cur
, head
) {
380 found
= list_entry(cur
, struct btrfs_space_info
, list
);
381 if (found
->flags
== flags
)
387 static struct btrfs_block_group_cache
*
388 __btrfs_find_block_group(struct btrfs_root
*root
,
389 struct btrfs_block_group_cache
*hint
,
390 u64 search_start
, int data
, int owner
)
392 struct btrfs_block_group_cache
*cache
;
393 struct btrfs_block_group_cache
*found_group
= NULL
;
394 struct btrfs_fs_info
*info
= root
->fs_info
;
402 if (data
& BTRFS_BLOCK_GROUP_METADATA
)
406 struct btrfs_block_group_cache
*shint
;
407 shint
= btrfs_lookup_first_block_group(info
, search_start
);
408 if (shint
&& block_group_bits(shint
, data
) && !shint
->ro
) {
409 spin_lock(&shint
->lock
);
410 used
= btrfs_block_group_used(&shint
->item
);
411 if (used
+ shint
->pinned
+ shint
->reserved
<
412 div_factor(shint
->key
.offset
, factor
)) {
413 spin_unlock(&shint
->lock
);
416 spin_unlock(&shint
->lock
);
419 if (hint
&& !hint
->ro
&& block_group_bits(hint
, data
)) {
420 spin_lock(&hint
->lock
);
421 used
= btrfs_block_group_used(&hint
->item
);
422 if (used
+ hint
->pinned
+ hint
->reserved
<
423 div_factor(hint
->key
.offset
, factor
)) {
424 spin_unlock(&hint
->lock
);
427 spin_unlock(&hint
->lock
);
428 last
= hint
->key
.objectid
+ hint
->key
.offset
;
431 last
= max(hint
->key
.objectid
, search_start
);
437 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
441 spin_lock(&cache
->lock
);
442 last
= cache
->key
.objectid
+ cache
->key
.offset
;
443 used
= btrfs_block_group_used(&cache
->item
);
445 if (!cache
->ro
&& block_group_bits(cache
, data
)) {
446 free_check
= div_factor(cache
->key
.offset
, factor
);
447 if (used
+ cache
->pinned
+ cache
->reserved
<
450 spin_unlock(&cache
->lock
);
454 spin_unlock(&cache
->lock
);
462 if (!full_search
&& factor
< 10) {
472 struct btrfs_block_group_cache
*btrfs_find_block_group(struct btrfs_root
*root
,
473 struct btrfs_block_group_cache
474 *hint
, u64 search_start
,
478 struct btrfs_block_group_cache
*ret
;
479 ret
= __btrfs_find_block_group(root
, hint
, search_start
, data
, owner
);
483 /* simple helper to search for an existing extent at a given offset */
484 int btrfs_lookup_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
487 struct btrfs_key key
;
488 struct btrfs_path
*path
;
490 path
= btrfs_alloc_path();
492 maybe_lock_mutex(root
);
493 key
.objectid
= start
;
495 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
496 ret
= btrfs_search_slot(NULL
, root
->fs_info
->extent_root
, &key
, path
,
498 maybe_unlock_mutex(root
);
499 btrfs_free_path(path
);
504 * Back reference rules. Back refs have three main goals:
506 * 1) differentiate between all holders of references to an extent so that
507 * when a reference is dropped we can make sure it was a valid reference
508 * before freeing the extent.
510 * 2) Provide enough information to quickly find the holders of an extent
511 * if we notice a given block is corrupted or bad.
513 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
514 * maintenance. This is actually the same as #2, but with a slightly
515 * different use case.
517 * File extents can be referenced by:
519 * - multiple snapshots, subvolumes, or different generations in one subvol
520 * - different files inside a single subvolume
521 * - different offsets inside a file (bookend extents in file.c)
523 * The extent ref structure has fields for:
525 * - Objectid of the subvolume root
526 * - Generation number of the tree holding the reference
527 * - objectid of the file holding the reference
528 * - offset in the file corresponding to the key holding the reference
529 * - number of references holding by parent node (alway 1 for tree blocks)
531 * Btree leaf may hold multiple references to a file extent. In most cases,
532 * these references are from same file and the corresponding offsets inside
533 * the file are close together. So inode objectid and offset in file are
534 * just hints, they provide hints about where in the btree the references
535 * can be found and when we can stop searching.
537 * When a file extent is allocated the fields are filled in:
538 * (root_key.objectid, trans->transid, inode objectid, offset in file, 1)
540 * When a leaf is cow'd new references are added for every file extent found
541 * in the leaf. It looks similar to the create case, but trans->transid will
542 * be different when the block is cow'd.
544 * (root_key.objectid, trans->transid, inode objectid, offset in file,
545 * number of references in the leaf)
547 * Because inode objectid and offset in file are just hints, they are not
548 * used when backrefs are deleted. When a file extent is removed either
549 * during snapshot deletion or file truncation, we find the corresponding
550 * back back reference and check the following fields.
552 * (btrfs_header_owner(leaf), btrfs_header_generation(leaf))
554 * Btree extents can be referenced by:
556 * - Different subvolumes
557 * - Different generations of the same subvolume
559 * When a tree block is created, back references are inserted:
561 * (root->root_key.objectid, trans->transid, level, 0, 1)
563 * When a tree block is cow'd, new back references are added for all the
564 * blocks it points to. If the tree block isn't in reference counted root,
565 * the old back references are removed. These new back references are of
566 * the form (trans->transid will have increased since creation):
568 * (root->root_key.objectid, trans->transid, level, 0, 1)
570 * When a backref is in deleting, the following fields are checked:
572 * if backref was for a tree root:
573 * (btrfs_header_owner(itself), btrfs_header_generation(itself))
575 * (btrfs_header_owner(parent), btrfs_header_generation(parent))
577 * Back Reference Key composing:
579 * The key objectid corresponds to the first byte in the extent, the key
580 * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first
581 * byte of parent extent. If a extent is tree root, the key offset is set
582 * to the key objectid.
585 static int noinline
lookup_extent_backref(struct btrfs_trans_handle
*trans
,
586 struct btrfs_root
*root
,
587 struct btrfs_path
*path
, u64 bytenr
,
588 u64 parent
, u64 ref_root
,
589 u64 ref_generation
, int del
)
591 struct btrfs_key key
;
592 struct btrfs_extent_ref
*ref
;
593 struct extent_buffer
*leaf
;
596 key
.objectid
= bytenr
;
597 key
.type
= BTRFS_EXTENT_REF_KEY
;
600 ret
= btrfs_search_slot(trans
, root
, &key
, path
, del
? -1 : 0, 1);
608 leaf
= path
->nodes
[0];
609 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_ref
);
610 if (btrfs_ref_root(leaf
, ref
) != ref_root
||
611 btrfs_ref_generation(leaf
, ref
) != ref_generation
) {
621 static int noinline
insert_extent_backref(struct btrfs_trans_handle
*trans
,
622 struct btrfs_root
*root
,
623 struct btrfs_path
*path
,
624 u64 bytenr
, u64 parent
,
625 u64 ref_root
, u64 ref_generation
,
626 u64 owner_objectid
, u64 owner_offset
)
628 struct btrfs_key key
;
629 struct extent_buffer
*leaf
;
630 struct btrfs_extent_ref
*ref
;
634 key
.objectid
= bytenr
;
635 key
.type
= BTRFS_EXTENT_REF_KEY
;
638 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, sizeof(*ref
));
640 leaf
= path
->nodes
[0];
641 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
642 struct btrfs_extent_ref
);
643 btrfs_set_ref_root(leaf
, ref
, ref_root
);
644 btrfs_set_ref_generation(leaf
, ref
, ref_generation
);
645 btrfs_set_ref_objectid(leaf
, ref
, owner_objectid
);
646 btrfs_set_ref_offset(leaf
, ref
, owner_offset
);
647 btrfs_set_ref_num_refs(leaf
, ref
, 1);
648 } else if (ret
== -EEXIST
) {
650 BUG_ON(owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
);
651 leaf
= path
->nodes
[0];
652 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
653 struct btrfs_extent_ref
);
654 if (btrfs_ref_root(leaf
, ref
) != ref_root
||
655 btrfs_ref_generation(leaf
, ref
) != ref_generation
) {
661 num_refs
= btrfs_ref_num_refs(leaf
, ref
);
662 BUG_ON(num_refs
== 0);
663 btrfs_set_ref_num_refs(leaf
, ref
, num_refs
+ 1);
665 existing_owner
= btrfs_ref_objectid(leaf
, ref
);
666 if (existing_owner
== owner_objectid
&&
667 btrfs_ref_offset(leaf
, ref
) > owner_offset
) {
668 btrfs_set_ref_offset(leaf
, ref
, owner_offset
);
669 } else if (existing_owner
!= owner_objectid
&&
670 existing_owner
!= BTRFS_MULTIPLE_OBJECTIDS
) {
671 btrfs_set_ref_objectid(leaf
, ref
,
672 BTRFS_MULTIPLE_OBJECTIDS
);
673 btrfs_set_ref_offset(leaf
, ref
, 0);
679 btrfs_mark_buffer_dirty(path
->nodes
[0]);
681 btrfs_release_path(root
, path
);
685 static int noinline
remove_extent_backref(struct btrfs_trans_handle
*trans
,
686 struct btrfs_root
*root
,
687 struct btrfs_path
*path
)
689 struct extent_buffer
*leaf
;
690 struct btrfs_extent_ref
*ref
;
694 leaf
= path
->nodes
[0];
695 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_ref
);
696 num_refs
= btrfs_ref_num_refs(leaf
, ref
);
697 BUG_ON(num_refs
== 0);
700 ret
= btrfs_del_item(trans
, root
, path
);
702 btrfs_set_ref_num_refs(leaf
, ref
, num_refs
);
703 btrfs_mark_buffer_dirty(leaf
);
705 btrfs_release_path(root
, path
);
709 static int __btrfs_update_extent_ref(struct btrfs_trans_handle
*trans
,
710 struct btrfs_root
*root
, u64 bytenr
,
711 u64 orig_parent
, u64 parent
,
712 u64 orig_root
, u64 ref_root
,
713 u64 orig_generation
, u64 ref_generation
,
714 u64 owner_objectid
, u64 owner_offset
)
717 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
718 struct btrfs_path
*path
;
720 if (root
== root
->fs_info
->extent_root
) {
721 struct pending_extent_op
*extent_op
;
724 BUG_ON(owner_objectid
>= BTRFS_MAX_LEVEL
);
725 num_bytes
= btrfs_level_size(root
, (int)owner_objectid
);
726 if (test_range_bit(&root
->fs_info
->extent_ins
, bytenr
,
727 bytenr
+ num_bytes
- 1, EXTENT_LOCKED
, 0)) {
729 ret
= get_state_private(&root
->fs_info
->extent_ins
,
732 extent_op
= (struct pending_extent_op
*)
734 BUG_ON(extent_op
->parent
!= orig_parent
);
735 BUG_ON(extent_op
->generation
!= orig_generation
);
736 extent_op
->parent
= parent
;
737 extent_op
->generation
= ref_generation
;
739 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
742 extent_op
->type
= PENDING_BACKREF_UPDATE
;
743 extent_op
->bytenr
= bytenr
;
744 extent_op
->num_bytes
= num_bytes
;
745 extent_op
->parent
= parent
;
746 extent_op
->orig_parent
= orig_parent
;
747 extent_op
->generation
= ref_generation
;
748 extent_op
->orig_generation
= orig_generation
;
749 extent_op
->level
= (int)owner_objectid
;
751 set_extent_bits(&root
->fs_info
->extent_ins
,
752 bytenr
, bytenr
+ num_bytes
- 1,
753 EXTENT_LOCKED
, GFP_NOFS
);
754 set_state_private(&root
->fs_info
->extent_ins
,
755 bytenr
, (unsigned long)extent_op
);
760 path
= btrfs_alloc_path();
763 ret
= lookup_extent_backref(trans
, extent_root
, path
,
764 bytenr
, orig_parent
, orig_root
,
768 ret
= remove_extent_backref(trans
, extent_root
, path
);
771 ret
= insert_extent_backref(trans
, extent_root
, path
, bytenr
,
772 parent
, ref_root
, ref_generation
,
773 owner_objectid
, owner_offset
);
775 finish_current_insert(trans
, extent_root
);
776 del_pending_extents(trans
, extent_root
);
778 btrfs_free_path(path
);
782 int btrfs_update_extent_ref(struct btrfs_trans_handle
*trans
,
783 struct btrfs_root
*root
, u64 bytenr
,
784 u64 orig_parent
, u64 parent
,
785 u64 ref_root
, u64 ref_generation
,
786 u64 owner_objectid
, u64 owner_offset
)
789 if (ref_root
== BTRFS_TREE_LOG_OBJECTID
&&
790 owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
)
792 maybe_lock_mutex(root
);
793 ret
= __btrfs_update_extent_ref(trans
, root
, bytenr
, orig_parent
,
794 parent
, ref_root
, ref_root
,
795 ref_generation
, ref_generation
,
796 owner_objectid
, owner_offset
);
797 maybe_unlock_mutex(root
);
801 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
802 struct btrfs_root
*root
, u64 bytenr
,
803 u64 orig_parent
, u64 parent
,
804 u64 orig_root
, u64 ref_root
,
805 u64 orig_generation
, u64 ref_generation
,
806 u64 owner_objectid
, u64 owner_offset
)
808 struct btrfs_path
*path
;
810 struct btrfs_key key
;
811 struct extent_buffer
*l
;
812 struct btrfs_extent_item
*item
;
815 path
= btrfs_alloc_path();
820 key
.objectid
= bytenr
;
821 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
822 key
.offset
= (u64
)-1;
824 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
, path
,
828 BUG_ON(ret
== 0 || path
->slots
[0] == 0);
833 btrfs_item_key_to_cpu(l
, &key
, path
->slots
[0]);
834 BUG_ON(key
.objectid
!= bytenr
);
835 BUG_ON(key
.type
!= BTRFS_EXTENT_ITEM_KEY
);
837 item
= btrfs_item_ptr(l
, path
->slots
[0], struct btrfs_extent_item
);
838 refs
= btrfs_extent_refs(l
, item
);
839 btrfs_set_extent_refs(l
, item
, refs
+ 1);
840 btrfs_mark_buffer_dirty(path
->nodes
[0]);
842 btrfs_release_path(root
->fs_info
->extent_root
, path
);
845 ret
= insert_extent_backref(trans
, root
->fs_info
->extent_root
,
846 path
, bytenr
, parent
,
847 ref_root
, ref_generation
,
848 owner_objectid
, owner_offset
);
850 finish_current_insert(trans
, root
->fs_info
->extent_root
);
851 del_pending_extents(trans
, root
->fs_info
->extent_root
);
853 btrfs_free_path(path
);
857 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
858 struct btrfs_root
*root
,
859 u64 bytenr
, u64 num_bytes
, u64 parent
,
860 u64 ref_root
, u64 ref_generation
,
861 u64 owner_objectid
, u64 owner_offset
)
864 if (ref_root
== BTRFS_TREE_LOG_OBJECTID
&&
865 owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
)
867 maybe_lock_mutex(root
);
868 ret
= __btrfs_inc_extent_ref(trans
, root
, bytenr
, 0, parent
,
869 0, ref_root
, 0, ref_generation
,
870 owner_objectid
, owner_offset
);
871 maybe_unlock_mutex(root
);
875 int btrfs_extent_post_op(struct btrfs_trans_handle
*trans
,
876 struct btrfs_root
*root
)
878 finish_current_insert(trans
, root
->fs_info
->extent_root
);
879 del_pending_extents(trans
, root
->fs_info
->extent_root
);
883 int btrfs_lookup_extent_ref(struct btrfs_trans_handle
*trans
,
884 struct btrfs_root
*root
, u64 bytenr
,
885 u64 num_bytes
, u32
*refs
)
887 struct btrfs_path
*path
;
889 struct btrfs_key key
;
890 struct extent_buffer
*l
;
891 struct btrfs_extent_item
*item
;
893 WARN_ON(num_bytes
< root
->sectorsize
);
894 path
= btrfs_alloc_path();
896 key
.objectid
= bytenr
;
897 key
.offset
= num_bytes
;
898 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
899 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
, path
,
904 btrfs_print_leaf(root
, path
->nodes
[0]);
905 printk("failed to find block number %Lu\n", bytenr
);
909 item
= btrfs_item_ptr(l
, path
->slots
[0], struct btrfs_extent_item
);
910 *refs
= btrfs_extent_refs(l
, item
);
912 btrfs_free_path(path
);
916 static int get_reference_status(struct btrfs_root
*root
, u64 bytenr
,
917 u64 parent_gen
, u64 ref_objectid
,
918 u64
*min_generation
, u32
*ref_count
)
920 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
921 struct btrfs_path
*path
;
922 struct extent_buffer
*leaf
;
923 struct btrfs_extent_ref
*ref_item
;
924 struct btrfs_key key
;
925 struct btrfs_key found_key
;
926 u64 root_objectid
= root
->root_key
.objectid
;
931 key
.objectid
= bytenr
;
932 key
.offset
= (u64
)-1;
933 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
935 path
= btrfs_alloc_path();
936 mutex_lock(&root
->fs_info
->alloc_mutex
);
937 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
941 if (ret
< 0 || path
->slots
[0] == 0)
945 leaf
= path
->nodes
[0];
946 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
948 if (found_key
.objectid
!= bytenr
||
949 found_key
.type
!= BTRFS_EXTENT_ITEM_KEY
) {
955 *min_generation
= (u64
)-1;
958 leaf
= path
->nodes
[0];
959 nritems
= btrfs_header_nritems(leaf
);
960 if (path
->slots
[0] >= nritems
) {
961 ret
= btrfs_next_leaf(extent_root
, path
);
968 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
969 if (found_key
.objectid
!= bytenr
)
972 if (found_key
.type
!= BTRFS_EXTENT_REF_KEY
) {
977 ref_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
978 struct btrfs_extent_ref
);
979 ref_generation
= btrfs_ref_generation(leaf
, ref_item
);
981 * For (parent_gen > 0 && parent_gen > ref_generation):
983 * we reach here through the oldest root, therefore
984 * all other reference from same snapshot should have
985 * a larger generation.
987 if ((root_objectid
!= btrfs_ref_root(leaf
, ref_item
)) ||
988 (parent_gen
> 0 && parent_gen
> ref_generation
) ||
989 (ref_objectid
>= BTRFS_FIRST_FREE_OBJECTID
&&
990 ref_objectid
!= btrfs_ref_objectid(leaf
, ref_item
))) {
996 if (*min_generation
> ref_generation
)
997 *min_generation
= ref_generation
;
1003 mutex_unlock(&root
->fs_info
->alloc_mutex
);
1004 btrfs_free_path(path
);
1008 int btrfs_cross_ref_exists(struct btrfs_trans_handle
*trans
,
1009 struct btrfs_root
*root
,
1010 struct btrfs_key
*key
, u64 bytenr
)
1012 struct btrfs_root
*old_root
;
1013 struct btrfs_path
*path
= NULL
;
1014 struct extent_buffer
*eb
;
1015 struct btrfs_file_extent_item
*item
;
1023 BUG_ON(trans
== NULL
);
1024 BUG_ON(key
->type
!= BTRFS_EXTENT_DATA_KEY
);
1025 ret
= get_reference_status(root
, bytenr
, 0, key
->objectid
,
1026 &min_generation
, &ref_count
);
1033 old_root
= root
->dirty_root
->root
;
1034 ref_generation
= old_root
->root_key
.offset
;
1036 /* all references are created in running transaction */
1037 if (min_generation
> ref_generation
) {
1042 path
= btrfs_alloc_path();
1048 path
->skip_locking
= 1;
1049 /* if no item found, the extent is referenced by other snapshot */
1050 ret
= btrfs_search_slot(NULL
, old_root
, key
, path
, 0, 0);
1054 eb
= path
->nodes
[0];
1055 item
= btrfs_item_ptr(eb
, path
->slots
[0],
1056 struct btrfs_file_extent_item
);
1057 if (btrfs_file_extent_type(eb
, item
) != BTRFS_FILE_EXTENT_REG
||
1058 btrfs_file_extent_disk_bytenr(eb
, item
) != bytenr
) {
1063 for (level
= BTRFS_MAX_LEVEL
- 1; level
>= -1; level
--) {
1065 eb
= path
->nodes
[level
];
1068 extent_start
= eb
->start
;
1070 extent_start
= bytenr
;
1072 ret
= get_reference_status(root
, extent_start
, ref_generation
,
1073 0, &min_generation
, &ref_count
);
1077 if (ref_count
!= 1) {
1082 ref_generation
= btrfs_header_generation(eb
);
1087 btrfs_free_path(path
);
1091 int btrfs_cache_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
1092 struct extent_buffer
*buf
, u32 nr_extents
)
1094 struct btrfs_key key
;
1095 struct btrfs_file_extent_item
*fi
;
1103 if (!root
->ref_cows
)
1106 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
1108 root_gen
= root
->root_key
.offset
;
1111 root_gen
= trans
->transid
- 1;
1114 level
= btrfs_header_level(buf
);
1115 nritems
= btrfs_header_nritems(buf
);
1118 struct btrfs_leaf_ref
*ref
;
1119 struct btrfs_extent_info
*info
;
1121 ref
= btrfs_alloc_leaf_ref(root
, nr_extents
);
1127 ref
->root_gen
= root_gen
;
1128 ref
->bytenr
= buf
->start
;
1129 ref
->owner
= btrfs_header_owner(buf
);
1130 ref
->generation
= btrfs_header_generation(buf
);
1131 ref
->nritems
= nr_extents
;
1132 info
= ref
->extents
;
1134 for (i
= 0; nr_extents
> 0 && i
< nritems
; i
++) {
1136 btrfs_item_key_to_cpu(buf
, &key
, i
);
1137 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
1139 fi
= btrfs_item_ptr(buf
, i
,
1140 struct btrfs_file_extent_item
);
1141 if (btrfs_file_extent_type(buf
, fi
) ==
1142 BTRFS_FILE_EXTENT_INLINE
)
1144 disk_bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
1145 if (disk_bytenr
== 0)
1148 info
->bytenr
= disk_bytenr
;
1150 btrfs_file_extent_disk_num_bytes(buf
, fi
);
1151 info
->objectid
= key
.objectid
;
1152 info
->offset
= key
.offset
;
1156 ret
= btrfs_add_leaf_ref(root
, ref
, shared
);
1158 btrfs_free_leaf_ref(root
, ref
);
1164 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
1165 struct extent_buffer
*orig_buf
, struct extent_buffer
*buf
,
1172 u64 orig_generation
;
1174 u32 nr_file_extents
= 0;
1175 struct btrfs_key key
;
1176 struct btrfs_file_extent_item
*fi
;
1181 int (*process_func
)(struct btrfs_trans_handle
*, struct btrfs_root
*,
1182 u64
, u64
, u64
, u64
, u64
, u64
, u64
, u64
, u64
);
1184 ref_root
= btrfs_header_owner(buf
);
1185 ref_generation
= btrfs_header_generation(buf
);
1186 orig_root
= btrfs_header_owner(orig_buf
);
1187 orig_generation
= btrfs_header_generation(orig_buf
);
1189 nritems
= btrfs_header_nritems(buf
);
1190 level
= btrfs_header_level(buf
);
1192 if (root
->ref_cows
) {
1193 process_func
= __btrfs_inc_extent_ref
;
1196 root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
)
1199 root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
)
1201 process_func
= __btrfs_update_extent_ref
;
1204 for (i
= 0; i
< nritems
; i
++) {
1207 btrfs_item_key_to_cpu(buf
, &key
, i
);
1208 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
1210 fi
= btrfs_item_ptr(buf
, i
,
1211 struct btrfs_file_extent_item
);
1212 if (btrfs_file_extent_type(buf
, fi
) ==
1213 BTRFS_FILE_EXTENT_INLINE
)
1215 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
1221 maybe_lock_mutex(root
);
1222 ret
= process_func(trans
, root
, bytenr
,
1223 orig_buf
->start
, buf
->start
,
1224 orig_root
, ref_root
,
1225 orig_generation
, ref_generation
,
1226 key
.objectid
, key
.offset
);
1227 maybe_unlock_mutex(root
);
1235 bytenr
= btrfs_node_blockptr(buf
, i
);
1236 maybe_lock_mutex(root
);
1237 ret
= process_func(trans
, root
, bytenr
,
1238 orig_buf
->start
, buf
->start
,
1239 orig_root
, ref_root
,
1240 orig_generation
, ref_generation
,
1242 maybe_unlock_mutex(root
);
1253 *nr_extents
= nr_file_extents
;
1255 *nr_extents
= nritems
;
1263 int btrfs_update_ref(struct btrfs_trans_handle
*trans
,
1264 struct btrfs_root
*root
, struct extent_buffer
*orig_buf
,
1265 struct extent_buffer
*buf
, int start_slot
, int nr
)
1272 u64 orig_generation
;
1273 struct btrfs_key key
;
1274 struct btrfs_file_extent_item
*fi
;
1280 BUG_ON(start_slot
< 0);
1281 BUG_ON(start_slot
+ nr
> btrfs_header_nritems(buf
));
1283 ref_root
= btrfs_header_owner(buf
);
1284 ref_generation
= btrfs_header_generation(buf
);
1285 orig_root
= btrfs_header_owner(orig_buf
);
1286 orig_generation
= btrfs_header_generation(orig_buf
);
1287 level
= btrfs_header_level(buf
);
1289 if (!root
->ref_cows
) {
1291 root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
)
1294 root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
)
1298 for (i
= 0, slot
= start_slot
; i
< nr
; i
++, slot
++) {
1301 btrfs_item_key_to_cpu(buf
, &key
, slot
);
1302 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
1304 fi
= btrfs_item_ptr(buf
, slot
,
1305 struct btrfs_file_extent_item
);
1306 if (btrfs_file_extent_type(buf
, fi
) ==
1307 BTRFS_FILE_EXTENT_INLINE
)
1309 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
1312 maybe_lock_mutex(root
);
1313 ret
= __btrfs_update_extent_ref(trans
, root
, bytenr
,
1314 orig_buf
->start
, buf
->start
,
1315 orig_root
, ref_root
,
1316 orig_generation
, ref_generation
,
1317 key
.objectid
, key
.offset
);
1318 maybe_unlock_mutex(root
);
1322 bytenr
= btrfs_node_blockptr(buf
, slot
);
1323 maybe_lock_mutex(root
);
1324 ret
= __btrfs_update_extent_ref(trans
, root
, bytenr
,
1325 orig_buf
->start
, buf
->start
,
1326 orig_root
, ref_root
,
1327 orig_generation
, ref_generation
,
1329 maybe_unlock_mutex(root
);
1340 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
1341 struct btrfs_root
*root
,
1342 struct btrfs_path
*path
,
1343 struct btrfs_block_group_cache
*cache
)
1347 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
1349 struct extent_buffer
*leaf
;
1351 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
1356 leaf
= path
->nodes
[0];
1357 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
1358 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
1359 btrfs_mark_buffer_dirty(leaf
);
1360 btrfs_release_path(extent_root
, path
);
1362 finish_current_insert(trans
, extent_root
);
1363 pending_ret
= del_pending_extents(trans
, extent_root
);
1372 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
1373 struct btrfs_root
*root
)
1375 struct btrfs_block_group_cache
*cache
, *entry
;
1379 struct btrfs_path
*path
;
1382 path
= btrfs_alloc_path();
1386 mutex_lock(&root
->fs_info
->alloc_mutex
);
1389 spin_lock(&root
->fs_info
->block_group_cache_lock
);
1390 for (n
= rb_first(&root
->fs_info
->block_group_cache_tree
);
1391 n
; n
= rb_next(n
)) {
1392 entry
= rb_entry(n
, struct btrfs_block_group_cache
,
1399 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
1405 last
+= cache
->key
.offset
;
1407 err
= write_one_cache_group(trans
, root
,
1410 * if we fail to write the cache group, we want
1411 * to keep it marked dirty in hopes that a later
1419 btrfs_free_path(path
);
1420 mutex_unlock(&root
->fs_info
->alloc_mutex
);
1424 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
1425 u64 total_bytes
, u64 bytes_used
,
1426 struct btrfs_space_info
**space_info
)
1428 struct btrfs_space_info
*found
;
1430 found
= __find_space_info(info
, flags
);
1432 found
->total_bytes
+= total_bytes
;
1433 found
->bytes_used
+= bytes_used
;
1435 *space_info
= found
;
1438 found
= kmalloc(sizeof(*found
), GFP_NOFS
);
1442 list_add(&found
->list
, &info
->space_info
);
1443 INIT_LIST_HEAD(&found
->block_groups
);
1444 spin_lock_init(&found
->lock
);
1445 found
->flags
= flags
;
1446 found
->total_bytes
= total_bytes
;
1447 found
->bytes_used
= bytes_used
;
1448 found
->bytes_pinned
= 0;
1449 found
->bytes_reserved
= 0;
1451 found
->force_alloc
= 0;
1452 *space_info
= found
;
1456 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
1458 u64 extra_flags
= flags
& (BTRFS_BLOCK_GROUP_RAID0
|
1459 BTRFS_BLOCK_GROUP_RAID1
|
1460 BTRFS_BLOCK_GROUP_RAID10
|
1461 BTRFS_BLOCK_GROUP_DUP
);
1463 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
1464 fs_info
->avail_data_alloc_bits
|= extra_flags
;
1465 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
1466 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
1467 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
1468 fs_info
->avail_system_alloc_bits
|= extra_flags
;
1472 static u64
reduce_alloc_profile(struct btrfs_root
*root
, u64 flags
)
1474 u64 num_devices
= root
->fs_info
->fs_devices
->num_devices
;
1476 if (num_devices
== 1)
1477 flags
&= ~(BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID0
);
1478 if (num_devices
< 4)
1479 flags
&= ~BTRFS_BLOCK_GROUP_RAID10
;
1481 if ((flags
& BTRFS_BLOCK_GROUP_DUP
) &&
1482 (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
1483 BTRFS_BLOCK_GROUP_RAID10
))) {
1484 flags
&= ~BTRFS_BLOCK_GROUP_DUP
;
1487 if ((flags
& BTRFS_BLOCK_GROUP_RAID1
) &&
1488 (flags
& BTRFS_BLOCK_GROUP_RAID10
)) {
1489 flags
&= ~BTRFS_BLOCK_GROUP_RAID1
;
1492 if ((flags
& BTRFS_BLOCK_GROUP_RAID0
) &&
1493 ((flags
& BTRFS_BLOCK_GROUP_RAID1
) |
1494 (flags
& BTRFS_BLOCK_GROUP_RAID10
) |
1495 (flags
& BTRFS_BLOCK_GROUP_DUP
)))
1496 flags
&= ~BTRFS_BLOCK_GROUP_RAID0
;
1500 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
1501 struct btrfs_root
*extent_root
, u64 alloc_bytes
,
1502 u64 flags
, int force
)
1504 struct btrfs_space_info
*space_info
;
1510 flags
= reduce_alloc_profile(extent_root
, flags
);
1512 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
1514 ret
= update_space_info(extent_root
->fs_info
, flags
,
1518 BUG_ON(!space_info
);
1520 if (space_info
->force_alloc
) {
1522 space_info
->force_alloc
= 0;
1524 if (space_info
->full
)
1527 thresh
= div_factor(space_info
->total_bytes
, 6);
1529 (space_info
->bytes_used
+ space_info
->bytes_pinned
+
1530 space_info
->bytes_reserved
+ alloc_bytes
) < thresh
)
1533 mutex_lock(&extent_root
->fs_info
->chunk_mutex
);
1534 ret
= btrfs_alloc_chunk(trans
, extent_root
, &start
, &num_bytes
, flags
);
1535 if (ret
== -ENOSPC
) {
1536 printk("space info full %Lu\n", flags
);
1537 space_info
->full
= 1;
1542 ret
= btrfs_make_block_group(trans
, extent_root
, 0, flags
,
1543 BTRFS_FIRST_CHUNK_TREE_OBJECTID
, start
, num_bytes
);
1547 mutex_unlock(&extent_root
->fs_info
->chunk_mutex
);
1552 static int update_block_group(struct btrfs_trans_handle
*trans
,
1553 struct btrfs_root
*root
,
1554 u64 bytenr
, u64 num_bytes
, int alloc
,
1557 struct btrfs_block_group_cache
*cache
;
1558 struct btrfs_fs_info
*info
= root
->fs_info
;
1559 u64 total
= num_bytes
;
1563 WARN_ON(!mutex_is_locked(&root
->fs_info
->alloc_mutex
));
1565 cache
= btrfs_lookup_block_group(info
, bytenr
);
1569 byte_in_group
= bytenr
- cache
->key
.objectid
;
1570 WARN_ON(byte_in_group
> cache
->key
.offset
);
1572 spin_lock(&cache
->lock
);
1574 old_val
= btrfs_block_group_used(&cache
->item
);
1575 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
1577 old_val
+= num_bytes
;
1578 cache
->space_info
->bytes_used
+= num_bytes
;
1579 btrfs_set_block_group_used(&cache
->item
, old_val
);
1580 spin_unlock(&cache
->lock
);
1582 old_val
-= num_bytes
;
1583 cache
->space_info
->bytes_used
-= num_bytes
;
1584 btrfs_set_block_group_used(&cache
->item
, old_val
);
1585 spin_unlock(&cache
->lock
);
1588 ret
= btrfs_add_free_space(cache
, bytenr
,
1595 bytenr
+= num_bytes
;
1600 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
1602 struct btrfs_block_group_cache
*cache
;
1604 cache
= btrfs_lookup_first_block_group(root
->fs_info
, search_start
);
1608 return cache
->key
.objectid
;
1611 int btrfs_update_pinned_extents(struct btrfs_root
*root
,
1612 u64 bytenr
, u64 num
, int pin
)
1615 struct btrfs_block_group_cache
*cache
;
1616 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1618 WARN_ON(!mutex_is_locked(&root
->fs_info
->alloc_mutex
));
1620 set_extent_dirty(&fs_info
->pinned_extents
,
1621 bytenr
, bytenr
+ num
- 1, GFP_NOFS
);
1623 clear_extent_dirty(&fs_info
->pinned_extents
,
1624 bytenr
, bytenr
+ num
- 1, GFP_NOFS
);
1627 cache
= btrfs_lookup_block_group(fs_info
, bytenr
);
1629 len
= min(num
, cache
->key
.offset
-
1630 (bytenr
- cache
->key
.objectid
));
1632 spin_lock(&cache
->lock
);
1633 cache
->pinned
+= len
;
1634 cache
->space_info
->bytes_pinned
+= len
;
1635 spin_unlock(&cache
->lock
);
1636 fs_info
->total_pinned
+= len
;
1638 spin_lock(&cache
->lock
);
1639 cache
->pinned
-= len
;
1640 cache
->space_info
->bytes_pinned
-= len
;
1641 spin_unlock(&cache
->lock
);
1642 fs_info
->total_pinned
-= len
;
1650 static int update_reserved_extents(struct btrfs_root
*root
,
1651 u64 bytenr
, u64 num
, int reserve
)
1654 struct btrfs_block_group_cache
*cache
;
1655 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1657 WARN_ON(!mutex_is_locked(&root
->fs_info
->alloc_mutex
));
1659 cache
= btrfs_lookup_block_group(fs_info
, bytenr
);
1661 len
= min(num
, cache
->key
.offset
-
1662 (bytenr
- cache
->key
.objectid
));
1664 spin_lock(&cache
->lock
);
1665 cache
->reserved
+= len
;
1666 cache
->space_info
->bytes_reserved
+= len
;
1667 spin_unlock(&cache
->lock
);
1669 spin_lock(&cache
->lock
);
1670 cache
->reserved
-= len
;
1671 cache
->space_info
->bytes_reserved
-= len
;
1672 spin_unlock(&cache
->lock
);
1680 int btrfs_copy_pinned(struct btrfs_root
*root
, struct extent_io_tree
*copy
)
1685 struct extent_io_tree
*pinned_extents
= &root
->fs_info
->pinned_extents
;
1689 ret
= find_first_extent_bit(pinned_extents
, last
,
1690 &start
, &end
, EXTENT_DIRTY
);
1693 set_extent_dirty(copy
, start
, end
, GFP_NOFS
);
1699 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
1700 struct btrfs_root
*root
,
1701 struct extent_io_tree
*unpin
)
1706 struct btrfs_block_group_cache
*cache
;
1708 mutex_lock(&root
->fs_info
->alloc_mutex
);
1710 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
1714 btrfs_update_pinned_extents(root
, start
, end
+ 1 - start
, 0);
1715 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
1716 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
1718 btrfs_add_free_space(cache
, start
, end
- start
+ 1);
1719 if (need_resched()) {
1720 mutex_unlock(&root
->fs_info
->alloc_mutex
);
1722 mutex_lock(&root
->fs_info
->alloc_mutex
);
1725 mutex_unlock(&root
->fs_info
->alloc_mutex
);
1729 static int finish_current_insert(struct btrfs_trans_handle
*trans
,
1730 struct btrfs_root
*extent_root
)
1735 struct btrfs_fs_info
*info
= extent_root
->fs_info
;
1736 struct btrfs_path
*path
;
1737 struct btrfs_extent_ref
*ref
;
1738 struct pending_extent_op
*extent_op
;
1739 struct btrfs_key key
;
1740 struct btrfs_extent_item extent_item
;
1744 WARN_ON(!mutex_is_locked(&extent_root
->fs_info
->alloc_mutex
));
1745 btrfs_set_stack_extent_refs(&extent_item
, 1);
1746 path
= btrfs_alloc_path();
1749 ret
= find_first_extent_bit(&info
->extent_ins
, 0, &start
,
1750 &end
, EXTENT_LOCKED
);
1754 ret
= get_state_private(&info
->extent_ins
, start
, &priv
);
1756 extent_op
= (struct pending_extent_op
*)(unsigned long)priv
;
1758 if (extent_op
->type
== PENDING_EXTENT_INSERT
) {
1759 key
.objectid
= start
;
1760 key
.offset
= end
+ 1 - start
;
1761 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1762 err
= btrfs_insert_item(trans
, extent_root
, &key
,
1763 &extent_item
, sizeof(extent_item
));
1766 clear_extent_bits(&info
->extent_ins
, start
, end
,
1767 EXTENT_LOCKED
, GFP_NOFS
);
1769 err
= insert_extent_backref(trans
, extent_root
, path
,
1770 start
, extent_op
->parent
,
1771 extent_root
->root_key
.objectid
,
1772 extent_op
->generation
,
1773 extent_op
->level
, 0);
1775 } else if (extent_op
->type
== PENDING_BACKREF_UPDATE
) {
1776 err
= lookup_extent_backref(trans
, extent_root
, path
,
1777 start
, extent_op
->orig_parent
,
1778 extent_root
->root_key
.objectid
,
1779 extent_op
->orig_generation
, 0);
1782 clear_extent_bits(&info
->extent_ins
, start
, end
,
1783 EXTENT_LOCKED
, GFP_NOFS
);
1785 key
.objectid
= start
;
1786 key
.offset
= extent_op
->parent
;
1787 key
.type
= BTRFS_EXTENT_REF_KEY
;
1788 err
= btrfs_set_item_key_safe(trans
, extent_root
, path
,
1791 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1792 struct btrfs_extent_ref
);
1793 btrfs_set_ref_generation(path
->nodes
[0], ref
,
1794 extent_op
->generation
);
1795 btrfs_mark_buffer_dirty(path
->nodes
[0]);
1796 btrfs_release_path(extent_root
, path
);
1802 if (need_resched()) {
1803 mutex_unlock(&extent_root
->fs_info
->alloc_mutex
);
1805 mutex_lock(&extent_root
->fs_info
->alloc_mutex
);
1808 btrfs_free_path(path
);
1812 static int pin_down_bytes(struct btrfs_trans_handle
*trans
,
1813 struct btrfs_root
*root
,
1814 u64 bytenr
, u64 num_bytes
, int is_data
)
1817 struct extent_buffer
*buf
;
1819 WARN_ON(!mutex_is_locked(&root
->fs_info
->alloc_mutex
));
1823 buf
= btrfs_find_tree_block(root
, bytenr
, num_bytes
);
1827 /* we can reuse a block if it hasn't been written
1828 * and it is from this transaction. We can't
1829 * reuse anything from the tree log root because
1830 * it has tiny sub-transactions.
1832 if (btrfs_buffer_uptodate(buf
, 0) &&
1833 btrfs_try_tree_lock(buf
)) {
1834 u64 header_owner
= btrfs_header_owner(buf
);
1835 u64 header_transid
= btrfs_header_generation(buf
);
1836 if (header_owner
!= BTRFS_TREE_LOG_OBJECTID
&&
1837 header_owner
!= BTRFS_TREE_RELOC_OBJECTID
&&
1838 header_transid
== trans
->transid
&&
1839 !btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
1840 clean_tree_block(NULL
, root
, buf
);
1841 btrfs_tree_unlock(buf
);
1842 free_extent_buffer(buf
);
1845 btrfs_tree_unlock(buf
);
1847 free_extent_buffer(buf
);
1849 btrfs_update_pinned_extents(root
, bytenr
, num_bytes
, 1);
1856 * remove an extent from the root, returns 0 on success
1858 static int __free_extent(struct btrfs_trans_handle
*trans
,
1859 struct btrfs_root
*root
,
1860 u64 bytenr
, u64 num_bytes
, u64 parent
,
1861 u64 root_objectid
, u64 ref_generation
,
1862 u64 owner_objectid
, u64 owner_offset
,
1863 int pin
, int mark_free
)
1865 struct btrfs_path
*path
;
1866 struct btrfs_key key
;
1867 struct btrfs_fs_info
*info
= root
->fs_info
;
1868 struct btrfs_root
*extent_root
= info
->extent_root
;
1869 struct extent_buffer
*leaf
;
1871 int extent_slot
= 0;
1872 int found_extent
= 0;
1874 struct btrfs_extent_item
*ei
;
1877 WARN_ON(!mutex_is_locked(&root
->fs_info
->alloc_mutex
));
1878 key
.objectid
= bytenr
;
1879 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
1880 key
.offset
= num_bytes
;
1881 path
= btrfs_alloc_path();
1886 ret
= lookup_extent_backref(trans
, extent_root
, path
, bytenr
, parent
,
1887 root_objectid
, ref_generation
, 1);
1889 struct btrfs_key found_key
;
1890 extent_slot
= path
->slots
[0];
1891 while(extent_slot
> 0) {
1893 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1895 if (found_key
.objectid
!= bytenr
)
1897 if (found_key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
1898 found_key
.offset
== num_bytes
) {
1902 if (path
->slots
[0] - extent_slot
> 5)
1905 if (!found_extent
) {
1906 ret
= remove_extent_backref(trans
, extent_root
, path
);
1908 btrfs_release_path(extent_root
, path
);
1909 ret
= btrfs_search_slot(trans
, extent_root
,
1912 extent_slot
= path
->slots
[0];
1915 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
1917 printk("Unable to find ref byte nr %Lu root %Lu "
1918 " gen %Lu owner %Lu offset %Lu\n", bytenr
,
1919 root_objectid
, ref_generation
, owner_objectid
,
1923 leaf
= path
->nodes
[0];
1924 ei
= btrfs_item_ptr(leaf
, extent_slot
,
1925 struct btrfs_extent_item
);
1926 refs
= btrfs_extent_refs(leaf
, ei
);
1929 btrfs_set_extent_refs(leaf
, ei
, refs
);
1931 btrfs_mark_buffer_dirty(leaf
);
1933 if (refs
== 0 && found_extent
&& path
->slots
[0] == extent_slot
+ 1) {
1934 struct btrfs_extent_ref
*ref
;
1935 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1936 struct btrfs_extent_ref
);
1937 BUG_ON(btrfs_ref_num_refs(leaf
, ref
) != 1);
1938 /* if the back ref and the extent are next to each other
1939 * they get deleted below in one shot
1941 path
->slots
[0] = extent_slot
;
1943 } else if (found_extent
) {
1944 /* otherwise delete the extent back ref */
1945 ret
= remove_extent_backref(trans
, extent_root
, path
);
1947 /* if refs are 0, we need to setup the path for deletion */
1949 btrfs_release_path(extent_root
, path
);
1950 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
1959 #ifdef BIO_RW_DISCARD
1960 u64 map_length
= num_bytes
;
1961 struct btrfs_multi_bio
*multi
= NULL
;
1965 ret
= pin_down_bytes(trans
, root
, bytenr
, num_bytes
,
1966 owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
);
1972 /* block accounting for super block */
1973 spin_lock_irq(&info
->delalloc_lock
);
1974 super_used
= btrfs_super_bytes_used(&info
->super_copy
);
1975 btrfs_set_super_bytes_used(&info
->super_copy
,
1976 super_used
- num_bytes
);
1977 spin_unlock_irq(&info
->delalloc_lock
);
1979 /* block accounting for root item */
1980 root_used
= btrfs_root_used(&root
->root_item
);
1981 btrfs_set_root_used(&root
->root_item
,
1982 root_used
- num_bytes
);
1983 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
1986 ret
= update_block_group(trans
, root
, bytenr
, num_bytes
, 0,
1990 #ifdef BIO_RW_DISCARD
1991 /* Tell the block device(s) that the sectors can be discarded */
1992 ret
= btrfs_map_block(&root
->fs_info
->mapping_tree
, READ
,
1993 bytenr
, &map_length
, &multi
, 0);
1995 struct btrfs_bio_stripe
*stripe
= multi
->stripes
;
1998 if (map_length
> num_bytes
)
1999 map_length
= num_bytes
;
2001 for (i
= 0; i
< multi
->num_stripes
; i
++, stripe
++) {
2002 blkdev_issue_discard(stripe
->dev
->bdev
,
2003 stripe
->physical
>> 9,
2010 btrfs_free_path(path
);
2011 finish_current_insert(trans
, extent_root
);
2016 * find all the blocks marked as pending in the radix tree and remove
2017 * them from the extent map
2019 static int del_pending_extents(struct btrfs_trans_handle
*trans
, struct
2020 btrfs_root
*extent_root
)
2028 struct extent_io_tree
*pending_del
;
2029 struct extent_io_tree
*extent_ins
;
2030 struct pending_extent_op
*extent_op
;
2032 WARN_ON(!mutex_is_locked(&extent_root
->fs_info
->alloc_mutex
));
2033 extent_ins
= &extent_root
->fs_info
->extent_ins
;
2034 pending_del
= &extent_root
->fs_info
->pending_del
;
2037 ret
= find_first_extent_bit(pending_del
, 0, &start
, &end
,
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_LOCKED
,
2049 ret
= pin_down_bytes(trans
, extent_root
, start
,
2050 end
+ 1 - start
, 0);
2051 mark_free
= ret
> 0;
2052 if (!test_range_bit(extent_ins
, start
, end
,
2053 EXTENT_LOCKED
, 0)) {
2055 ret
= __free_extent(trans
, extent_root
,
2056 start
, end
+ 1 - start
,
2057 extent_op
->orig_parent
,
2058 extent_root
->root_key
.objectid
,
2059 extent_op
->orig_generation
,
2060 extent_op
->level
, 0, 0, mark_free
);
2064 ret
= get_state_private(extent_ins
, start
, &priv
);
2066 extent_op
= (struct pending_extent_op
*)
2067 (unsigned long)priv
;
2069 clear_extent_bits(extent_ins
, start
, end
,
2070 EXTENT_LOCKED
, GFP_NOFS
);
2072 if (extent_op
->type
== PENDING_BACKREF_UPDATE
)
2075 ret
= update_block_group(trans
, extent_root
, start
,
2076 end
+ 1 - start
, 0, mark_free
);
2083 if (need_resched()) {
2084 mutex_unlock(&extent_root
->fs_info
->alloc_mutex
);
2086 mutex_lock(&extent_root
->fs_info
->alloc_mutex
);
2093 * remove an extent from the root, returns 0 on success
2095 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
2096 struct btrfs_root
*root
,
2097 u64 bytenr
, u64 num_bytes
, u64 parent
,
2098 u64 root_objectid
, u64 ref_generation
,
2099 u64 owner_objectid
, u64 owner_offset
, int pin
)
2101 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2105 WARN_ON(num_bytes
< root
->sectorsize
);
2106 if (root
== extent_root
) {
2107 struct pending_extent_op
*extent_op
;
2109 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
2112 extent_op
->type
= PENDING_EXTENT_DELETE
;
2113 extent_op
->bytenr
= bytenr
;
2114 extent_op
->num_bytes
= num_bytes
;
2115 extent_op
->parent
= parent
;
2116 extent_op
->orig_parent
= parent
;
2117 extent_op
->generation
= ref_generation
;
2118 extent_op
->orig_generation
= ref_generation
;
2119 extent_op
->level
= (int)owner_objectid
;
2121 set_extent_bits(&root
->fs_info
->pending_del
,
2122 bytenr
, bytenr
+ num_bytes
- 1,
2123 EXTENT_LOCKED
, GFP_NOFS
);
2124 set_state_private(&root
->fs_info
->pending_del
,
2125 bytenr
, (unsigned long)extent_op
);
2128 /* if metadata always pin */
2129 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
) {
2130 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
2131 struct btrfs_block_group_cache
*cache
;
2133 /* btrfs_free_reserved_extent */
2134 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
2136 btrfs_add_free_space(cache
, bytenr
, num_bytes
);
2137 update_reserved_extents(root
, bytenr
, num_bytes
, 0);
2143 /* if data pin when any transaction has committed this */
2144 if (ref_generation
!= trans
->transid
)
2147 ret
= __free_extent(trans
, root
, bytenr
, num_bytes
, parent
,
2148 root_objectid
, ref_generation
, owner_objectid
,
2149 owner_offset
, pin
, pin
== 0);
2151 finish_current_insert(trans
, root
->fs_info
->extent_root
);
2152 pending_ret
= del_pending_extents(trans
, root
->fs_info
->extent_root
);
2153 return ret
? ret
: pending_ret
;
2156 int btrfs_free_extent(struct btrfs_trans_handle
*trans
,
2157 struct btrfs_root
*root
,
2158 u64 bytenr
, u64 num_bytes
, u64 parent
,
2159 u64 root_objectid
, u64 ref_generation
,
2160 u64 owner_objectid
, u64 owner_offset
, int pin
)
2164 maybe_lock_mutex(root
);
2165 ret
= __btrfs_free_extent(trans
, root
, bytenr
, num_bytes
, parent
,
2166 root_objectid
, ref_generation
,
2167 owner_objectid
, owner_offset
, pin
);
2168 maybe_unlock_mutex(root
);
2172 static u64
stripe_align(struct btrfs_root
*root
, u64 val
)
2174 u64 mask
= ((u64
)root
->stripesize
- 1);
2175 u64 ret
= (val
+ mask
) & ~mask
;
2180 * walks the btree of allocated extents and find a hole of a given size.
2181 * The key ins is changed to record the hole:
2182 * ins->objectid == block start
2183 * ins->flags = BTRFS_EXTENT_ITEM_KEY
2184 * ins->offset == number of blocks
2185 * Any available blocks before search_start are skipped.
2187 static int noinline
find_free_extent(struct btrfs_trans_handle
*trans
,
2188 struct btrfs_root
*orig_root
,
2189 u64 num_bytes
, u64 empty_size
,
2190 u64 search_start
, u64 search_end
,
2191 u64 hint_byte
, struct btrfs_key
*ins
,
2192 u64 exclude_start
, u64 exclude_nr
,
2196 u64 orig_search_start
;
2197 struct btrfs_root
* root
= orig_root
->fs_info
->extent_root
;
2198 struct btrfs_fs_info
*info
= root
->fs_info
;
2199 u64 total_needed
= num_bytes
;
2200 u64
*last_ptr
= NULL
;
2201 struct btrfs_block_group_cache
*block_group
;
2202 int chunk_alloc_done
= 0;
2203 int empty_cluster
= 2 * 1024 * 1024;
2204 int allowed_chunk_alloc
= 0;
2206 WARN_ON(num_bytes
< root
->sectorsize
);
2207 btrfs_set_key_type(ins
, BTRFS_EXTENT_ITEM_KEY
);
2209 if (orig_root
->ref_cows
|| empty_size
)
2210 allowed_chunk_alloc
= 1;
2212 if (data
& BTRFS_BLOCK_GROUP_METADATA
) {
2213 last_ptr
= &root
->fs_info
->last_alloc
;
2214 empty_cluster
= 256 * 1024;
2217 if ((data
& BTRFS_BLOCK_GROUP_DATA
) && btrfs_test_opt(root
, SSD
))
2218 last_ptr
= &root
->fs_info
->last_data_alloc
;
2220 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
2221 last_ptr
= &root
->fs_info
->last_log_alloc
;
2222 if (!last_ptr
== 0 && root
->fs_info
->last_alloc
) {
2223 *last_ptr
= root
->fs_info
->last_alloc
+ empty_cluster
;
2229 hint_byte
= *last_ptr
;
2231 empty_size
+= empty_cluster
;
2234 search_start
= max(search_start
, first_logical_byte(root
, 0));
2235 orig_search_start
= search_start
;
2237 search_start
= max(search_start
, hint_byte
);
2238 total_needed
+= empty_size
;
2241 block_group
= btrfs_lookup_block_group(info
, search_start
);
2243 block_group
= btrfs_lookup_first_block_group(info
,
2247 * Ok this looks a little tricky, buts its really simple. First if we
2248 * didn't find a block group obviously we want to start over.
2249 * Secondly, if the block group we found does not match the type we
2250 * need, and we have a last_ptr and its not 0, chances are the last
2251 * allocation we made was at the end of the block group, so lets go
2252 * ahead and skip the looking through the rest of the block groups and
2253 * start at the beginning. This helps with metadata allocations,
2254 * since you are likely to have a bunch of data block groups to search
2255 * through first before you realize that you need to start over, so go
2256 * ahead and start over and save the time.
2258 if (!block_group
|| (!block_group_bits(block_group
, data
) &&
2259 last_ptr
&& *last_ptr
)) {
2260 if (search_start
!= orig_search_start
) {
2261 if (last_ptr
&& *last_ptr
) {
2262 total_needed
+= empty_cluster
;
2265 search_start
= orig_search_start
;
2267 } else if (!chunk_alloc_done
&& allowed_chunk_alloc
) {
2268 ret
= do_chunk_alloc(trans
, root
,
2269 num_bytes
+ 2 * 1024 * 1024,
2274 chunk_alloc_done
= 1;
2275 search_start
= orig_search_start
;
2284 * this is going to seach through all of the existing block groups it
2285 * can find, so if we don't find something we need to see if we can
2286 * allocate what we need.
2288 ret
= find_free_space(root
, &block_group
, &search_start
,
2289 total_needed
, data
);
2290 if (ret
== -ENOSPC
) {
2292 * instead of allocating, start at the original search start
2293 * and see if there is something to be found, if not then we
2296 if (search_start
!= orig_search_start
) {
2297 if (last_ptr
&& *last_ptr
) {
2299 total_needed
+= empty_cluster
;
2301 search_start
= orig_search_start
;
2306 * we've already allocated, we're pretty screwed
2308 if (chunk_alloc_done
) {
2310 } else if (!allowed_chunk_alloc
&& block_group
&&
2311 block_group_bits(block_group
, data
)) {
2312 block_group
->space_info
->force_alloc
= 1;
2314 } else if (!allowed_chunk_alloc
) {
2318 ret
= do_chunk_alloc(trans
, root
, num_bytes
+ 2 * 1024 * 1024,
2324 chunk_alloc_done
= 1;
2326 search_start
= block_group
->key
.objectid
+
2327 block_group
->key
.offset
;
2329 search_start
= orig_search_start
;
2336 search_start
= stripe_align(root
, search_start
);
2337 ins
->objectid
= search_start
;
2338 ins
->offset
= num_bytes
;
2340 if (ins
->objectid
+ num_bytes
>= search_end
) {
2341 search_start
= orig_search_start
;
2342 if (chunk_alloc_done
) {
2349 if (ins
->objectid
+ num_bytes
>
2350 block_group
->key
.objectid
+ block_group
->key
.offset
) {
2351 if (search_start
== orig_search_start
&& chunk_alloc_done
) {
2355 search_start
= block_group
->key
.objectid
+
2356 block_group
->key
.offset
;
2360 if (exclude_nr
> 0 && (ins
->objectid
+ num_bytes
> exclude_start
&&
2361 ins
->objectid
< exclude_start
+ exclude_nr
)) {
2362 search_start
= exclude_start
+ exclude_nr
;
2366 if (!(data
& BTRFS_BLOCK_GROUP_DATA
))
2367 trans
->block_group
= block_group
;
2369 ins
->offset
= num_bytes
;
2371 *last_ptr
= ins
->objectid
+ ins
->offset
;
2373 btrfs_super_total_bytes(&root
->fs_info
->super_copy
))
2382 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
)
2384 struct btrfs_block_group_cache
*cache
;
2385 struct list_head
*l
;
2387 printk(KERN_INFO
"space_info has %Lu free, is %sfull\n",
2388 info
->total_bytes
- info
->bytes_used
- info
->bytes_pinned
-
2389 info
->bytes_reserved
, (info
->full
) ? "" : "not ");
2391 spin_lock(&info
->lock
);
2392 list_for_each(l
, &info
->block_groups
) {
2393 cache
= list_entry(l
, struct btrfs_block_group_cache
, list
);
2394 spin_lock(&cache
->lock
);
2395 printk(KERN_INFO
"block group %Lu has %Lu bytes, %Lu used "
2396 "%Lu pinned %Lu reserved\n",
2397 cache
->key
.objectid
, cache
->key
.offset
,
2398 btrfs_block_group_used(&cache
->item
),
2399 cache
->pinned
, cache
->reserved
);
2400 btrfs_dump_free_space(cache
, bytes
);
2401 spin_unlock(&cache
->lock
);
2403 spin_unlock(&info
->lock
);
2406 static int __btrfs_reserve_extent(struct btrfs_trans_handle
*trans
,
2407 struct btrfs_root
*root
,
2408 u64 num_bytes
, u64 min_alloc_size
,
2409 u64 empty_size
, u64 hint_byte
,
2410 u64 search_end
, struct btrfs_key
*ins
,
2414 u64 search_start
= 0;
2416 struct btrfs_fs_info
*info
= root
->fs_info
;
2417 struct btrfs_block_group_cache
*cache
;
2420 alloc_profile
= info
->avail_data_alloc_bits
&
2421 info
->data_alloc_profile
;
2422 data
= BTRFS_BLOCK_GROUP_DATA
| alloc_profile
;
2423 } else if (root
== root
->fs_info
->chunk_root
) {
2424 alloc_profile
= info
->avail_system_alloc_bits
&
2425 info
->system_alloc_profile
;
2426 data
= BTRFS_BLOCK_GROUP_SYSTEM
| alloc_profile
;
2428 alloc_profile
= info
->avail_metadata_alloc_bits
&
2429 info
->metadata_alloc_profile
;
2430 data
= BTRFS_BLOCK_GROUP_METADATA
| alloc_profile
;
2433 data
= reduce_alloc_profile(root
, data
);
2435 * the only place that sets empty_size is btrfs_realloc_node, which
2436 * is not called recursively on allocations
2438 if (empty_size
|| root
->ref_cows
) {
2439 if (!(data
& BTRFS_BLOCK_GROUP_METADATA
)) {
2440 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
2442 BTRFS_BLOCK_GROUP_METADATA
|
2443 (info
->metadata_alloc_profile
&
2444 info
->avail_metadata_alloc_bits
), 0);
2446 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
2447 num_bytes
+ 2 * 1024 * 1024, data
, 0);
2450 WARN_ON(num_bytes
< root
->sectorsize
);
2451 ret
= find_free_extent(trans
, root
, num_bytes
, empty_size
,
2452 search_start
, search_end
, hint_byte
, ins
,
2453 trans
->alloc_exclude_start
,
2454 trans
->alloc_exclude_nr
, data
);
2456 if (ret
== -ENOSPC
&& num_bytes
> min_alloc_size
) {
2457 num_bytes
= num_bytes
>> 1;
2458 num_bytes
= num_bytes
& ~(root
->sectorsize
- 1);
2459 num_bytes
= max(num_bytes
, min_alloc_size
);
2460 do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
2461 num_bytes
, data
, 1);
2465 struct btrfs_space_info
*sinfo
;
2467 sinfo
= __find_space_info(root
->fs_info
, data
);
2468 printk("allocation failed flags %Lu, wanted %Lu\n",
2470 dump_space_info(sinfo
, num_bytes
);
2473 cache
= btrfs_lookup_block_group(root
->fs_info
, ins
->objectid
);
2475 printk(KERN_ERR
"Unable to find block group for %Lu\n", ins
->objectid
);
2479 ret
= btrfs_remove_free_space(cache
, ins
->objectid
, ins
->offset
);
2484 int btrfs_free_reserved_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
2486 struct btrfs_block_group_cache
*cache
;
2488 maybe_lock_mutex(root
);
2489 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
2491 printk(KERN_ERR
"Unable to find block group for %Lu\n", start
);
2492 maybe_unlock_mutex(root
);
2495 btrfs_add_free_space(cache
, start
, len
);
2496 update_reserved_extents(root
, start
, len
, 0);
2497 maybe_unlock_mutex(root
);
2501 int btrfs_reserve_extent(struct btrfs_trans_handle
*trans
,
2502 struct btrfs_root
*root
,
2503 u64 num_bytes
, u64 min_alloc_size
,
2504 u64 empty_size
, u64 hint_byte
,
2505 u64 search_end
, struct btrfs_key
*ins
,
2509 maybe_lock_mutex(root
);
2510 ret
= __btrfs_reserve_extent(trans
, root
, num_bytes
, min_alloc_size
,
2511 empty_size
, hint_byte
, search_end
, ins
,
2513 update_reserved_extents(root
, ins
->objectid
, ins
->offset
, 1);
2514 maybe_unlock_mutex(root
);
2518 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle
*trans
,
2519 struct btrfs_root
*root
, u64 parent
,
2520 u64 root_objectid
, u64 ref_generation
,
2521 u64 owner
, u64 owner_offset
,
2522 struct btrfs_key
*ins
)
2528 u64 num_bytes
= ins
->offset
;
2530 struct btrfs_fs_info
*info
= root
->fs_info
;
2531 struct btrfs_root
*extent_root
= info
->extent_root
;
2532 struct btrfs_extent_item
*extent_item
;
2533 struct btrfs_extent_ref
*ref
;
2534 struct btrfs_path
*path
;
2535 struct btrfs_key keys
[2];
2538 parent
= ins
->objectid
;
2540 /* block accounting for super block */
2541 spin_lock_irq(&info
->delalloc_lock
);
2542 super_used
= btrfs_super_bytes_used(&info
->super_copy
);
2543 btrfs_set_super_bytes_used(&info
->super_copy
, super_used
+ num_bytes
);
2544 spin_unlock_irq(&info
->delalloc_lock
);
2546 /* block accounting for root item */
2547 root_used
= btrfs_root_used(&root
->root_item
);
2548 btrfs_set_root_used(&root
->root_item
, root_used
+ num_bytes
);
2550 if (root
== extent_root
) {
2551 struct pending_extent_op
*extent_op
;
2553 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
2556 extent_op
->type
= PENDING_EXTENT_INSERT
;
2557 extent_op
->bytenr
= ins
->objectid
;
2558 extent_op
->num_bytes
= ins
->offset
;
2559 extent_op
->parent
= parent
;
2560 extent_op
->orig_parent
= 0;
2561 extent_op
->generation
= ref_generation
;
2562 extent_op
->orig_generation
= 0;
2563 extent_op
->level
= (int)owner
;
2565 set_extent_bits(&root
->fs_info
->extent_ins
, ins
->objectid
,
2566 ins
->objectid
+ ins
->offset
- 1,
2567 EXTENT_LOCKED
, GFP_NOFS
);
2568 set_state_private(&root
->fs_info
->extent_ins
,
2569 ins
->objectid
, (unsigned long)extent_op
);
2573 memcpy(&keys
[0], ins
, sizeof(*ins
));
2574 keys
[1].objectid
= ins
->objectid
;
2575 keys
[1].type
= BTRFS_EXTENT_REF_KEY
;
2576 keys
[1].offset
= parent
;
2577 sizes
[0] = sizeof(*extent_item
);
2578 sizes
[1] = sizeof(*ref
);
2580 path
= btrfs_alloc_path();
2583 ret
= btrfs_insert_empty_items(trans
, extent_root
, path
, keys
,
2587 extent_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2588 struct btrfs_extent_item
);
2589 btrfs_set_extent_refs(path
->nodes
[0], extent_item
, 1);
2590 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
2591 struct btrfs_extent_ref
);
2593 btrfs_set_ref_root(path
->nodes
[0], ref
, root_objectid
);
2594 btrfs_set_ref_generation(path
->nodes
[0], ref
, ref_generation
);
2595 btrfs_set_ref_objectid(path
->nodes
[0], ref
, owner
);
2596 btrfs_set_ref_offset(path
->nodes
[0], ref
, owner_offset
);
2597 btrfs_set_ref_num_refs(path
->nodes
[0], ref
, 1);
2599 btrfs_mark_buffer_dirty(path
->nodes
[0]);
2601 trans
->alloc_exclude_start
= 0;
2602 trans
->alloc_exclude_nr
= 0;
2603 btrfs_free_path(path
);
2604 finish_current_insert(trans
, extent_root
);
2605 pending_ret
= del_pending_extents(trans
, extent_root
);
2615 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1, 0);
2617 printk("update block group failed for %Lu %Lu\n",
2618 ins
->objectid
, ins
->offset
);
2625 int btrfs_alloc_reserved_extent(struct btrfs_trans_handle
*trans
,
2626 struct btrfs_root
*root
, u64 parent
,
2627 u64 root_objectid
, u64 ref_generation
,
2628 u64 owner
, u64 owner_offset
,
2629 struct btrfs_key
*ins
)
2633 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
)
2635 maybe_lock_mutex(root
);
2636 ret
= __btrfs_alloc_reserved_extent(trans
, root
, parent
,
2637 root_objectid
, ref_generation
,
2638 owner
, owner_offset
, ins
);
2639 update_reserved_extents(root
, ins
->objectid
, ins
->offset
, 0);
2640 maybe_unlock_mutex(root
);
2645 * this is used by the tree logging recovery code. It records that
2646 * an extent has been allocated and makes sure to clear the free
2647 * space cache bits as well
2649 int btrfs_alloc_logged_extent(struct btrfs_trans_handle
*trans
,
2650 struct btrfs_root
*root
, u64 parent
,
2651 u64 root_objectid
, u64 ref_generation
,
2652 u64 owner
, u64 owner_offset
,
2653 struct btrfs_key
*ins
)
2656 struct btrfs_block_group_cache
*block_group
;
2658 maybe_lock_mutex(root
);
2659 block_group
= btrfs_lookup_block_group(root
->fs_info
, ins
->objectid
);
2660 cache_block_group(root
, block_group
);
2662 ret
= btrfs_remove_free_space(block_group
, ins
->objectid
, ins
->offset
);
2664 ret
= __btrfs_alloc_reserved_extent(trans
, root
, parent
,
2665 root_objectid
, ref_generation
,
2666 owner
, owner_offset
, ins
);
2667 maybe_unlock_mutex(root
);
2672 * finds a free extent and does all the dirty work required for allocation
2673 * returns the key for the extent through ins, and a tree buffer for
2674 * the first block of the extent through buf.
2676 * returns 0 if everything worked, non-zero otherwise.
2678 int btrfs_alloc_extent(struct btrfs_trans_handle
*trans
,
2679 struct btrfs_root
*root
,
2680 u64 num_bytes
, u64 parent
, u64 min_alloc_size
,
2681 u64 root_objectid
, u64 ref_generation
,
2682 u64 owner_objectid
, u64 owner_offset
,
2683 u64 empty_size
, u64 hint_byte
,
2684 u64 search_end
, struct btrfs_key
*ins
, u64 data
)
2688 maybe_lock_mutex(root
);
2690 ret
= __btrfs_reserve_extent(trans
, root
, num_bytes
,
2691 min_alloc_size
, empty_size
, hint_byte
,
2692 search_end
, ins
, data
);
2694 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
2695 ret
= __btrfs_alloc_reserved_extent(trans
, root
, parent
,
2696 root_objectid
, ref_generation
,
2697 owner_objectid
, owner_offset
, ins
);
2701 update_reserved_extents(root
, ins
->objectid
, ins
->offset
, 1);
2703 maybe_unlock_mutex(root
);
2707 struct extent_buffer
*btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
,
2708 struct btrfs_root
*root
,
2709 u64 bytenr
, u32 blocksize
)
2711 struct extent_buffer
*buf
;
2713 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
2715 return ERR_PTR(-ENOMEM
);
2716 btrfs_set_header_generation(buf
, trans
->transid
);
2717 btrfs_tree_lock(buf
);
2718 clean_tree_block(trans
, root
, buf
);
2719 btrfs_set_buffer_uptodate(buf
);
2720 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
2721 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
2722 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
2724 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
2725 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
2727 trans
->blocks_used
++;
2732 * helper function to allocate a block for a given tree
2733 * returns the tree buffer or NULL.
2735 struct extent_buffer
*btrfs_alloc_free_block(struct btrfs_trans_handle
*trans
,
2736 struct btrfs_root
*root
,
2737 u32 blocksize
, u64 parent
,
2744 struct btrfs_key ins
;
2746 struct extent_buffer
*buf
;
2748 ret
= btrfs_alloc_extent(trans
, root
, blocksize
, parent
, blocksize
,
2749 root_objectid
, ref_generation
, level
, 0,
2750 empty_size
, hint
, (u64
)-1, &ins
, 0);
2753 return ERR_PTR(ret
);
2756 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
, blocksize
);
2760 int btrfs_drop_leaf_ref(struct btrfs_trans_handle
*trans
,
2761 struct btrfs_root
*root
, struct extent_buffer
*leaf
)
2764 u64 leaf_generation
;
2765 struct btrfs_key key
;
2766 struct btrfs_file_extent_item
*fi
;
2771 BUG_ON(!btrfs_is_leaf(leaf
));
2772 nritems
= btrfs_header_nritems(leaf
);
2773 leaf_owner
= btrfs_header_owner(leaf
);
2774 leaf_generation
= btrfs_header_generation(leaf
);
2776 for (i
= 0; i
< nritems
; i
++) {
2780 btrfs_item_key_to_cpu(leaf
, &key
, i
);
2781 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
2783 fi
= btrfs_item_ptr(leaf
, i
, struct btrfs_file_extent_item
);
2784 if (btrfs_file_extent_type(leaf
, fi
) ==
2785 BTRFS_FILE_EXTENT_INLINE
)
2788 * FIXME make sure to insert a trans record that
2789 * repeats the snapshot del on crash
2791 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
2792 if (disk_bytenr
== 0)
2795 mutex_lock(&root
->fs_info
->alloc_mutex
);
2796 ret
= __btrfs_free_extent(trans
, root
, disk_bytenr
,
2797 btrfs_file_extent_disk_num_bytes(leaf
, fi
),
2798 leaf
->start
, leaf_owner
, leaf_generation
,
2799 key
.objectid
, key
.offset
, 0);
2800 mutex_unlock(&root
->fs_info
->alloc_mutex
);
2803 atomic_inc(&root
->fs_info
->throttle_gen
);
2804 wake_up(&root
->fs_info
->transaction_throttle
);
2810 static int noinline
cache_drop_leaf_ref(struct btrfs_trans_handle
*trans
,
2811 struct btrfs_root
*root
,
2812 struct btrfs_leaf_ref
*ref
)
2816 struct btrfs_extent_info
*info
= ref
->extents
;
2818 for (i
= 0; i
< ref
->nritems
; i
++) {
2819 mutex_lock(&root
->fs_info
->alloc_mutex
);
2820 ret
= __btrfs_free_extent(trans
, root
, info
->bytenr
,
2821 info
->num_bytes
, ref
->bytenr
,
2822 ref
->owner
, ref
->generation
,
2823 info
->objectid
, info
->offset
, 0);
2824 mutex_unlock(&root
->fs_info
->alloc_mutex
);
2826 atomic_inc(&root
->fs_info
->throttle_gen
);
2827 wake_up(&root
->fs_info
->transaction_throttle
);
2837 int drop_snap_lookup_refcount(struct btrfs_root
*root
, u64 start
, u64 len
,
2842 ret
= btrfs_lookup_extent_ref(NULL
, root
, start
, len
, refs
);
2845 #if 0 // some debugging code in case we see problems here
2846 /* if the refs count is one, it won't get increased again. But
2847 * if the ref count is > 1, someone may be decreasing it at
2848 * the same time we are.
2851 struct extent_buffer
*eb
= NULL
;
2852 eb
= btrfs_find_create_tree_block(root
, start
, len
);
2854 btrfs_tree_lock(eb
);
2856 mutex_lock(&root
->fs_info
->alloc_mutex
);
2857 ret
= lookup_extent_ref(NULL
, root
, start
, len
, refs
);
2859 mutex_unlock(&root
->fs_info
->alloc_mutex
);
2862 btrfs_tree_unlock(eb
);
2863 free_extent_buffer(eb
);
2866 printk("block %llu went down to one during drop_snap\n",
2867 (unsigned long long)start
);
2878 * helper function for drop_snapshot, this walks down the tree dropping ref
2879 * counts as it goes.
2881 static int noinline
walk_down_tree(struct btrfs_trans_handle
*trans
,
2882 struct btrfs_root
*root
,
2883 struct btrfs_path
*path
, int *level
)
2889 struct extent_buffer
*next
;
2890 struct extent_buffer
*cur
;
2891 struct extent_buffer
*parent
;
2892 struct btrfs_leaf_ref
*ref
;
2897 WARN_ON(*level
< 0);
2898 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
2899 ret
= drop_snap_lookup_refcount(root
, path
->nodes
[*level
]->start
,
2900 path
->nodes
[*level
]->len
, &refs
);
2906 * walk down to the last node level and free all the leaves
2908 while(*level
>= 0) {
2909 WARN_ON(*level
< 0);
2910 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
2911 cur
= path
->nodes
[*level
];
2913 if (btrfs_header_level(cur
) != *level
)
2916 if (path
->slots
[*level
] >=
2917 btrfs_header_nritems(cur
))
2920 ret
= btrfs_drop_leaf_ref(trans
, root
, cur
);
2924 bytenr
= btrfs_node_blockptr(cur
, path
->slots
[*level
]);
2925 ptr_gen
= btrfs_node_ptr_generation(cur
, path
->slots
[*level
]);
2926 blocksize
= btrfs_level_size(root
, *level
- 1);
2928 ret
= drop_snap_lookup_refcount(root
, bytenr
, blocksize
, &refs
);
2931 parent
= path
->nodes
[*level
];
2932 root_owner
= btrfs_header_owner(parent
);
2933 root_gen
= btrfs_header_generation(parent
);
2934 path
->slots
[*level
]++;
2936 mutex_lock(&root
->fs_info
->alloc_mutex
);
2937 ret
= __btrfs_free_extent(trans
, root
, bytenr
,
2938 blocksize
, parent
->start
,
2939 root_owner
, root_gen
, 0, 0, 1);
2941 mutex_unlock(&root
->fs_info
->alloc_mutex
);
2943 atomic_inc(&root
->fs_info
->throttle_gen
);
2944 wake_up(&root
->fs_info
->transaction_throttle
);
2950 * at this point, we have a single ref, and since the
2951 * only place referencing this extent is a dead root
2952 * the reference count should never go higher.
2953 * So, we don't need to check it again
2956 ref
= btrfs_lookup_leaf_ref(root
, bytenr
);
2957 if (ref
&& ref
->generation
!= ptr_gen
) {
2958 btrfs_free_leaf_ref(root
, ref
);
2962 ret
= cache_drop_leaf_ref(trans
, root
, ref
);
2964 btrfs_remove_leaf_ref(root
, ref
);
2965 btrfs_free_leaf_ref(root
, ref
);
2969 if (printk_ratelimit())
2970 printk("leaf ref miss for bytenr %llu\n",
2971 (unsigned long long)bytenr
);
2973 next
= btrfs_find_tree_block(root
, bytenr
, blocksize
);
2974 if (!next
|| !btrfs_buffer_uptodate(next
, ptr_gen
)) {
2975 free_extent_buffer(next
);
2977 next
= read_tree_block(root
, bytenr
, blocksize
,
2982 * this is a debugging check and can go away
2983 * the ref should never go all the way down to 1
2986 ret
= lookup_extent_ref(NULL
, root
, bytenr
, blocksize
,
2992 WARN_ON(*level
<= 0);
2993 if (path
->nodes
[*level
-1])
2994 free_extent_buffer(path
->nodes
[*level
-1]);
2995 path
->nodes
[*level
-1] = next
;
2996 *level
= btrfs_header_level(next
);
2997 path
->slots
[*level
] = 0;
3001 WARN_ON(*level
< 0);
3002 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
3004 if (path
->nodes
[*level
] == root
->node
) {
3005 parent
= path
->nodes
[*level
];
3006 bytenr
= path
->nodes
[*level
]->start
;
3008 parent
= path
->nodes
[*level
+ 1];
3009 bytenr
= btrfs_node_blockptr(parent
, path
->slots
[*level
+ 1]);
3012 blocksize
= btrfs_level_size(root
, *level
);
3013 root_owner
= btrfs_header_owner(parent
);
3014 root_gen
= btrfs_header_generation(parent
);
3016 mutex_lock(&root
->fs_info
->alloc_mutex
);
3017 ret
= __btrfs_free_extent(trans
, root
, bytenr
, blocksize
,
3018 parent
->start
, root_owner
, root_gen
,
3020 mutex_unlock(&root
->fs_info
->alloc_mutex
);
3021 free_extent_buffer(path
->nodes
[*level
]);
3022 path
->nodes
[*level
] = NULL
;
3031 * helper for dropping snapshots. This walks back up the tree in the path
3032 * to find the first node higher up where we haven't yet gone through
3035 static int noinline
walk_up_tree(struct btrfs_trans_handle
*trans
,
3036 struct btrfs_root
*root
,
3037 struct btrfs_path
*path
, int *level
)
3041 struct btrfs_root_item
*root_item
= &root
->root_item
;
3046 for(i
= *level
; i
< BTRFS_MAX_LEVEL
- 1 && path
->nodes
[i
]; i
++) {
3047 slot
= path
->slots
[i
];
3048 if (slot
< btrfs_header_nritems(path
->nodes
[i
]) - 1) {
3049 struct extent_buffer
*node
;
3050 struct btrfs_disk_key disk_key
;
3051 node
= path
->nodes
[i
];
3054 WARN_ON(*level
== 0);
3055 btrfs_node_key(node
, &disk_key
, path
->slots
[i
]);
3056 memcpy(&root_item
->drop_progress
,
3057 &disk_key
, sizeof(disk_key
));
3058 root_item
->drop_level
= i
;
3061 struct extent_buffer
*parent
;
3062 if (path
->nodes
[*level
] == root
->node
)
3063 parent
= path
->nodes
[*level
];
3065 parent
= path
->nodes
[*level
+ 1];
3067 root_owner
= btrfs_header_owner(parent
);
3068 root_gen
= btrfs_header_generation(parent
);
3069 ret
= btrfs_free_extent(trans
, root
,
3070 path
->nodes
[*level
]->start
,
3071 path
->nodes
[*level
]->len
,
3073 root_owner
, root_gen
, 0, 0, 1);
3075 free_extent_buffer(path
->nodes
[*level
]);
3076 path
->nodes
[*level
] = NULL
;
3084 * drop the reference count on the tree rooted at 'snap'. This traverses
3085 * the tree freeing any blocks that have a ref count of zero after being
3088 int btrfs_drop_snapshot(struct btrfs_trans_handle
*trans
, struct btrfs_root
3094 struct btrfs_path
*path
;
3097 struct btrfs_root_item
*root_item
= &root
->root_item
;
3099 WARN_ON(!mutex_is_locked(&root
->fs_info
->drop_mutex
));
3100 path
= btrfs_alloc_path();
3103 level
= btrfs_header_level(root
->node
);
3105 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
3106 path
->nodes
[level
] = root
->node
;
3107 extent_buffer_get(root
->node
);
3108 path
->slots
[level
] = 0;
3110 struct btrfs_key key
;
3111 struct btrfs_disk_key found_key
;
3112 struct extent_buffer
*node
;
3114 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
3115 level
= root_item
->drop_level
;
3116 path
->lowest_level
= level
;
3117 wret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3122 node
= path
->nodes
[level
];
3123 btrfs_node_key(node
, &found_key
, path
->slots
[level
]);
3124 WARN_ON(memcmp(&found_key
, &root_item
->drop_progress
,
3125 sizeof(found_key
)));
3127 * unlock our path, this is safe because only this
3128 * function is allowed to delete this snapshot
3130 for (i
= 0; i
< BTRFS_MAX_LEVEL
; i
++) {
3131 if (path
->nodes
[i
] && path
->locks
[i
]) {
3133 btrfs_tree_unlock(path
->nodes
[i
]);
3138 wret
= walk_down_tree(trans
, root
, path
, &level
);
3144 wret
= walk_up_tree(trans
, root
, path
, &level
);
3149 if (trans
->transaction
->in_commit
) {
3153 atomic_inc(&root
->fs_info
->throttle_gen
);
3154 wake_up(&root
->fs_info
->transaction_throttle
);
3156 for (i
= 0; i
<= orig_level
; i
++) {
3157 if (path
->nodes
[i
]) {
3158 free_extent_buffer(path
->nodes
[i
]);
3159 path
->nodes
[i
] = NULL
;
3163 btrfs_free_path(path
);
3167 static unsigned long calc_ra(unsigned long start
, unsigned long last
,
3170 return min(last
, start
+ nr
- 1);
3173 static int noinline
relocate_inode_pages(struct inode
*inode
, u64 start
,
3178 unsigned long first_index
;
3179 unsigned long last_index
;
3182 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3183 struct file_ra_state
*ra
;
3184 struct btrfs_ordered_extent
*ordered
;
3185 unsigned int total_read
= 0;
3186 unsigned int total_dirty
= 0;
3189 ra
= kzalloc(sizeof(*ra
), GFP_NOFS
);
3191 mutex_lock(&inode
->i_mutex
);
3192 first_index
= start
>> PAGE_CACHE_SHIFT
;
3193 last_index
= (start
+ len
- 1) >> PAGE_CACHE_SHIFT
;
3195 /* make sure the dirty trick played by the caller work */
3196 ret
= invalidate_inode_pages2_range(inode
->i_mapping
,
3197 first_index
, last_index
);
3201 file_ra_state_init(ra
, inode
->i_mapping
);
3203 for (i
= first_index
; i
<= last_index
; i
++) {
3204 if (total_read
% ra
->ra_pages
== 0) {
3205 btrfs_force_ra(inode
->i_mapping
, ra
, NULL
, i
,
3206 calc_ra(i
, last_index
, ra
->ra_pages
));
3210 if (((u64
)i
<< PAGE_CACHE_SHIFT
) > i_size_read(inode
))
3212 page
= grab_cache_page(inode
->i_mapping
, i
);
3217 if (!PageUptodate(page
)) {
3218 btrfs_readpage(NULL
, page
);
3220 if (!PageUptodate(page
)) {
3222 page_cache_release(page
);
3227 wait_on_page_writeback(page
);
3229 page_start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
3230 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
3231 lock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
3233 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
3235 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
3237 page_cache_release(page
);
3238 btrfs_start_ordered_extent(inode
, ordered
, 1);
3239 btrfs_put_ordered_extent(ordered
);
3242 set_page_extent_mapped(page
);
3244 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
3245 if (i
== first_index
)
3246 set_extent_bits(io_tree
, page_start
, page_end
,
3247 EXTENT_BOUNDARY
, GFP_NOFS
);
3249 set_page_dirty(page
);
3252 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
3254 page_cache_release(page
);
3259 mutex_unlock(&inode
->i_mutex
);
3260 balance_dirty_pages_ratelimited_nr(inode
->i_mapping
, total_dirty
);
3264 static int noinline
relocate_data_extent(struct inode
*reloc_inode
,
3265 struct btrfs_key
*extent_key
,
3268 struct btrfs_root
*root
= BTRFS_I(reloc_inode
)->root
;
3269 struct extent_map_tree
*em_tree
= &BTRFS_I(reloc_inode
)->extent_tree
;
3270 struct extent_map
*em
;
3272 em
= alloc_extent_map(GFP_NOFS
);
3273 BUG_ON(!em
|| IS_ERR(em
));
3275 em
->start
= extent_key
->objectid
- offset
;
3276 em
->len
= extent_key
->offset
;
3277 em
->block_start
= extent_key
->objectid
;
3278 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
3279 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
3281 /* setup extent map to cheat btrfs_readpage */
3282 mutex_lock(&BTRFS_I(reloc_inode
)->extent_mutex
);
3285 spin_lock(&em_tree
->lock
);
3286 ret
= add_extent_mapping(em_tree
, em
);
3287 spin_unlock(&em_tree
->lock
);
3288 if (ret
!= -EEXIST
) {
3289 free_extent_map(em
);
3292 btrfs_drop_extent_cache(reloc_inode
, em
->start
,
3293 em
->start
+ em
->len
- 1, 0);
3295 mutex_unlock(&BTRFS_I(reloc_inode
)->extent_mutex
);
3297 return relocate_inode_pages(reloc_inode
, extent_key
->objectid
- offset
,
3298 extent_key
->offset
);
3301 struct btrfs_ref_path
{
3303 u64 nodes
[BTRFS_MAX_LEVEL
];
3305 u64 root_generation
;
3313 struct disk_extent
{
3320 static int is_cowonly_root(u64 root_objectid
)
3322 if (root_objectid
== BTRFS_ROOT_TREE_OBJECTID
||
3323 root_objectid
== BTRFS_EXTENT_TREE_OBJECTID
||
3324 root_objectid
== BTRFS_CHUNK_TREE_OBJECTID
||
3325 root_objectid
== BTRFS_DEV_TREE_OBJECTID
||
3326 root_objectid
== BTRFS_TREE_LOG_OBJECTID
)
3331 static int noinline
__next_ref_path(struct btrfs_trans_handle
*trans
,
3332 struct btrfs_root
*extent_root
,
3333 struct btrfs_ref_path
*ref_path
,
3336 struct extent_buffer
*leaf
;
3337 struct btrfs_path
*path
;
3338 struct btrfs_extent_ref
*ref
;
3339 struct btrfs_key key
;
3340 struct btrfs_key found_key
;
3346 path
= btrfs_alloc_path();
3350 mutex_lock(&extent_root
->fs_info
->alloc_mutex
);
3353 ref_path
->lowest_level
= -1;
3354 ref_path
->current_level
= -1;
3358 level
= ref_path
->current_level
- 1;
3359 while (level
>= -1) {
3361 if (level
< ref_path
->lowest_level
)
3365 bytenr
= ref_path
->nodes
[level
];
3367 bytenr
= ref_path
->extent_start
;
3369 BUG_ON(bytenr
== 0);
3371 parent
= ref_path
->nodes
[level
+ 1];
3372 ref_path
->nodes
[level
+ 1] = 0;
3373 ref_path
->current_level
= level
;
3374 BUG_ON(parent
== 0);
3376 key
.objectid
= bytenr
;
3377 key
.offset
= parent
+ 1;
3378 key
.type
= BTRFS_EXTENT_REF_KEY
;
3380 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
, 0, 0);
3385 leaf
= path
->nodes
[0];
3386 nritems
= btrfs_header_nritems(leaf
);
3387 if (path
->slots
[0] >= nritems
) {
3388 ret
= btrfs_next_leaf(extent_root
, path
);
3393 leaf
= path
->nodes
[0];
3396 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3397 if (found_key
.objectid
== bytenr
&&
3398 found_key
.type
== BTRFS_EXTENT_REF_KEY
)
3402 btrfs_release_path(extent_root
, path
);
3403 if (need_resched()) {
3404 mutex_unlock(&extent_root
->fs_info
->alloc_mutex
);
3406 mutex_lock(&extent_root
->fs_info
->alloc_mutex
);
3409 /* reached lowest level */
3413 level
= ref_path
->current_level
;
3414 while (level
< BTRFS_MAX_LEVEL
- 1) {
3417 bytenr
= ref_path
->nodes
[level
];
3419 bytenr
= ref_path
->extent_start
;
3421 BUG_ON(bytenr
== 0);
3423 key
.objectid
= bytenr
;
3425 key
.type
= BTRFS_EXTENT_REF_KEY
;
3427 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
, 0, 0);
3431 leaf
= path
->nodes
[0];
3432 nritems
= btrfs_header_nritems(leaf
);
3433 if (path
->slots
[0] >= nritems
) {
3434 ret
= btrfs_next_leaf(extent_root
, path
);
3438 /* the extent was freed by someone */
3439 if (ref_path
->lowest_level
== level
)
3441 btrfs_release_path(extent_root
, path
);
3444 leaf
= path
->nodes
[0];
3447 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3448 if (found_key
.objectid
!= bytenr
||
3449 found_key
.type
!= BTRFS_EXTENT_REF_KEY
) {
3450 /* the extent was freed by someone */
3451 if (ref_path
->lowest_level
== level
) {
3455 btrfs_release_path(extent_root
, path
);
3459 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
3460 struct btrfs_extent_ref
);
3461 ref_objectid
= btrfs_ref_objectid(leaf
, ref
);
3462 if (ref_objectid
< BTRFS_FIRST_FREE_OBJECTID
) {
3464 level
= (int)ref_objectid
;
3465 BUG_ON(level
>= BTRFS_MAX_LEVEL
);
3466 ref_path
->lowest_level
= level
;
3467 ref_path
->current_level
= level
;
3468 ref_path
->nodes
[level
] = bytenr
;
3470 WARN_ON(ref_objectid
!= level
);
3473 WARN_ON(level
!= -1);
3477 if (ref_path
->lowest_level
== level
) {
3478 ref_path
->owner_objectid
= ref_objectid
;
3479 ref_path
->owner_offset
= btrfs_ref_offset(leaf
, ref
);
3480 ref_path
->num_refs
= btrfs_ref_num_refs(leaf
, ref
);
3484 * the block is tree root or the block isn't in reference
3487 if (found_key
.objectid
== found_key
.offset
||
3488 is_cowonly_root(btrfs_ref_root(leaf
, ref
))) {
3489 ref_path
->root_objectid
= btrfs_ref_root(leaf
, ref
);
3490 ref_path
->root_generation
=
3491 btrfs_ref_generation(leaf
, ref
);
3493 /* special reference from the tree log */
3494 ref_path
->nodes
[0] = found_key
.offset
;
3495 ref_path
->current_level
= 0;
3502 BUG_ON(ref_path
->nodes
[level
] != 0);
3503 ref_path
->nodes
[level
] = found_key
.offset
;
3504 ref_path
->current_level
= level
;
3507 * the reference was created in the running transaction,
3508 * no need to continue walking up.
3510 if (btrfs_ref_generation(leaf
, ref
) == trans
->transid
) {
3511 ref_path
->root_objectid
= btrfs_ref_root(leaf
, ref
);
3512 ref_path
->root_generation
=
3513 btrfs_ref_generation(leaf
, ref
);
3518 btrfs_release_path(extent_root
, path
);
3519 if (need_resched()) {
3520 mutex_unlock(&extent_root
->fs_info
->alloc_mutex
);
3522 mutex_lock(&extent_root
->fs_info
->alloc_mutex
);
3525 /* reached max tree level, but no tree root found. */
3528 mutex_unlock(&extent_root
->fs_info
->alloc_mutex
);
3529 btrfs_free_path(path
);
3533 static int btrfs_first_ref_path(struct btrfs_trans_handle
*trans
,
3534 struct btrfs_root
*extent_root
,
3535 struct btrfs_ref_path
*ref_path
,
3538 memset(ref_path
, 0, sizeof(*ref_path
));
3539 ref_path
->extent_start
= extent_start
;
3541 return __next_ref_path(trans
, extent_root
, ref_path
, 1);
3544 static int btrfs_next_ref_path(struct btrfs_trans_handle
*trans
,
3545 struct btrfs_root
*extent_root
,
3546 struct btrfs_ref_path
*ref_path
)
3548 return __next_ref_path(trans
, extent_root
, ref_path
, 0);
3551 static int noinline
get_new_locations(struct inode
*reloc_inode
,
3552 struct btrfs_key
*extent_key
,
3553 u64 offset
, int no_fragment
,
3554 struct disk_extent
**extents
,
3557 struct btrfs_root
*root
= BTRFS_I(reloc_inode
)->root
;
3558 struct btrfs_path
*path
;
3559 struct btrfs_file_extent_item
*fi
;
3560 struct extent_buffer
*leaf
;
3561 struct disk_extent
*exts
= *extents
;
3562 struct btrfs_key found_key
;
3567 int max
= *nr_extents
;
3570 WARN_ON(!no_fragment
&& *extents
);
3573 exts
= kmalloc(sizeof(*exts
) * max
, GFP_NOFS
);
3578 path
= btrfs_alloc_path();
3581 cur_pos
= extent_key
->objectid
- offset
;
3582 last_byte
= extent_key
->objectid
+ extent_key
->offset
;
3583 ret
= btrfs_lookup_file_extent(NULL
, root
, path
, reloc_inode
->i_ino
,
3593 leaf
= path
->nodes
[0];
3594 nritems
= btrfs_header_nritems(leaf
);
3595 if (path
->slots
[0] >= nritems
) {
3596 ret
= btrfs_next_leaf(root
, path
);
3601 leaf
= path
->nodes
[0];
3604 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3605 if (found_key
.offset
!= cur_pos
||
3606 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
||
3607 found_key
.objectid
!= reloc_inode
->i_ino
)
3610 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
3611 struct btrfs_file_extent_item
);
3612 if (btrfs_file_extent_type(leaf
, fi
) !=
3613 BTRFS_FILE_EXTENT_REG
||
3614 btrfs_file_extent_disk_bytenr(leaf
, fi
) == 0)
3618 struct disk_extent
*old
= exts
;
3620 exts
= kzalloc(sizeof(*exts
) * max
, GFP_NOFS
);
3621 memcpy(exts
, old
, sizeof(*exts
) * nr
);
3622 if (old
!= *extents
)
3626 exts
[nr
].disk_bytenr
=
3627 btrfs_file_extent_disk_bytenr(leaf
, fi
);
3628 exts
[nr
].disk_num_bytes
=
3629 btrfs_file_extent_disk_num_bytes(leaf
, fi
);
3630 exts
[nr
].offset
= btrfs_file_extent_offset(leaf
, fi
);
3631 exts
[nr
].num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
3632 WARN_ON(exts
[nr
].offset
> 0);
3633 WARN_ON(exts
[nr
].num_bytes
!= exts
[nr
].disk_num_bytes
);
3635 cur_pos
+= exts
[nr
].num_bytes
;
3638 if (cur_pos
+ offset
>= last_byte
)
3648 WARN_ON(cur_pos
+ offset
> last_byte
);
3649 if (cur_pos
+ offset
< last_byte
) {
3655 btrfs_free_path(path
);
3657 if (exts
!= *extents
)
3666 static int noinline
replace_one_extent(struct btrfs_trans_handle
*trans
,
3667 struct btrfs_root
*root
,
3668 struct btrfs_path
*path
,
3669 struct btrfs_key
*extent_key
,
3670 struct btrfs_key
*leaf_key
,
3671 struct btrfs_ref_path
*ref_path
,
3672 struct disk_extent
*new_extents
,
3675 struct extent_buffer
*leaf
;
3676 struct btrfs_file_extent_item
*fi
;
3677 struct inode
*inode
= NULL
;
3678 struct btrfs_key key
;
3685 int extent_locked
= 0;
3688 first_pos
= ref_path
->owner_offset
;
3689 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
) {
3690 key
.objectid
= ref_path
->owner_objectid
;
3691 key
.offset
= ref_path
->owner_offset
;
3692 key
.type
= BTRFS_EXTENT_DATA_KEY
;
3694 memcpy(&key
, leaf_key
, sizeof(key
));
3698 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
3702 leaf
= path
->nodes
[0];
3703 nritems
= btrfs_header_nritems(leaf
);
3705 if (extent_locked
&& ret
> 0) {
3707 * the file extent item was modified by someone
3708 * before the extent got locked.
3710 mutex_unlock(&BTRFS_I(inode
)->extent_mutex
);
3711 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
3712 lock_end
, GFP_NOFS
);
3716 if (path
->slots
[0] >= nritems
) {
3717 if (ref_path
->owner_objectid
==
3718 BTRFS_MULTIPLE_OBJECTIDS
)
3721 BUG_ON(extent_locked
);
3722 ret
= btrfs_next_leaf(root
, path
);
3727 leaf
= path
->nodes
[0];
3728 nritems
= btrfs_header_nritems(leaf
);
3731 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
3733 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
) {
3734 if ((key
.objectid
> ref_path
->owner_objectid
) ||
3735 (key
.objectid
== ref_path
->owner_objectid
&&
3736 key
.type
> BTRFS_EXTENT_DATA_KEY
) ||
3737 (key
.offset
>= first_pos
+ extent_key
->offset
))
3741 if (inode
&& key
.objectid
!= inode
->i_ino
) {
3742 BUG_ON(extent_locked
);
3743 btrfs_release_path(root
, path
);
3744 mutex_unlock(&inode
->i_mutex
);
3750 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
3755 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
3756 struct btrfs_file_extent_item
);
3757 if ((btrfs_file_extent_type(leaf
, fi
) !=
3758 BTRFS_FILE_EXTENT_REG
) ||
3759 (btrfs_file_extent_disk_bytenr(leaf
, fi
) !=
3760 extent_key
->objectid
)) {
3766 num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
3767 ext_offset
= btrfs_file_extent_offset(leaf
, fi
);
3769 if (first_pos
> key
.offset
- ext_offset
)
3770 first_pos
= key
.offset
- ext_offset
;
3772 if (!extent_locked
) {
3773 lock_start
= key
.offset
;
3774 lock_end
= lock_start
+ num_bytes
- 1;
3776 BUG_ON(lock_start
!= key
.offset
);
3777 BUG_ON(lock_end
- lock_start
+ 1 < num_bytes
);
3781 btrfs_release_path(root
, path
);
3783 inode
= btrfs_iget_locked(root
->fs_info
->sb
,
3784 key
.objectid
, root
);
3785 if (inode
->i_state
& I_NEW
) {
3786 BTRFS_I(inode
)->root
= root
;
3787 BTRFS_I(inode
)->location
.objectid
=
3789 BTRFS_I(inode
)->location
.type
=
3790 BTRFS_INODE_ITEM_KEY
;
3791 BTRFS_I(inode
)->location
.offset
= 0;
3792 btrfs_read_locked_inode(inode
);
3793 unlock_new_inode(inode
);
3796 * some code call btrfs_commit_transaction while
3797 * holding the i_mutex, so we can't use mutex_lock
3800 if (is_bad_inode(inode
) ||
3801 !mutex_trylock(&inode
->i_mutex
)) {
3804 key
.offset
= (u64
)-1;
3809 if (!extent_locked
) {
3810 struct btrfs_ordered_extent
*ordered
;
3812 btrfs_release_path(root
, path
);
3814 lock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
3815 lock_end
, GFP_NOFS
);
3816 ordered
= btrfs_lookup_first_ordered_extent(inode
,
3819 ordered
->file_offset
<= lock_end
&&
3820 ordered
->file_offset
+ ordered
->len
> lock_start
) {
3821 unlock_extent(&BTRFS_I(inode
)->io_tree
,
3822 lock_start
, lock_end
, GFP_NOFS
);
3823 btrfs_start_ordered_extent(inode
, ordered
, 1);
3824 btrfs_put_ordered_extent(ordered
);
3825 key
.offset
+= num_bytes
;
3829 btrfs_put_ordered_extent(ordered
);
3831 mutex_lock(&BTRFS_I(inode
)->extent_mutex
);
3836 if (nr_extents
== 1) {
3837 /* update extent pointer in place */
3838 btrfs_set_file_extent_generation(leaf
, fi
,
3840 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
3841 new_extents
[0].disk_bytenr
);
3842 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
3843 new_extents
[0].disk_num_bytes
);
3844 ext_offset
+= new_extents
[0].offset
;
3845 btrfs_set_file_extent_offset(leaf
, fi
, ext_offset
);
3846 btrfs_mark_buffer_dirty(leaf
);
3848 btrfs_drop_extent_cache(inode
, key
.offset
,
3849 key
.offset
+ num_bytes
- 1, 0);
3851 ret
= btrfs_inc_extent_ref(trans
, root
,
3852 new_extents
[0].disk_bytenr
,
3853 new_extents
[0].disk_num_bytes
,
3855 root
->root_key
.objectid
,
3857 key
.objectid
, key
.offset
);
3860 ret
= btrfs_free_extent(trans
, root
,
3861 extent_key
->objectid
,
3864 btrfs_header_owner(leaf
),
3865 btrfs_header_generation(leaf
),
3866 key
.objectid
, key
.offset
, 0);
3869 btrfs_release_path(root
, path
);
3870 key
.offset
+= num_bytes
;
3876 * drop old extent pointer at first, then insert the
3877 * new pointers one bye one
3879 btrfs_release_path(root
, path
);
3880 ret
= btrfs_drop_extents(trans
, root
, inode
, key
.offset
,
3881 key
.offset
+ num_bytes
,
3882 key
.offset
, &alloc_hint
);
3885 for (i
= 0; i
< nr_extents
; i
++) {
3886 if (ext_offset
>= new_extents
[i
].num_bytes
) {
3887 ext_offset
-= new_extents
[i
].num_bytes
;
3890 extent_len
= min(new_extents
[i
].num_bytes
-
3891 ext_offset
, num_bytes
);
3893 ret
= btrfs_insert_empty_item(trans
, root
,
3898 leaf
= path
->nodes
[0];
3899 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
3900 struct btrfs_file_extent_item
);
3901 btrfs_set_file_extent_generation(leaf
, fi
,
3903 btrfs_set_file_extent_type(leaf
, fi
,
3904 BTRFS_FILE_EXTENT_REG
);
3905 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
3906 new_extents
[i
].disk_bytenr
);
3907 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
3908 new_extents
[i
].disk_num_bytes
);
3909 btrfs_set_file_extent_num_bytes(leaf
, fi
,
3911 ext_offset
+= new_extents
[i
].offset
;
3912 btrfs_set_file_extent_offset(leaf
, fi
,
3914 btrfs_mark_buffer_dirty(leaf
);
3916 btrfs_drop_extent_cache(inode
, key
.offset
,
3917 key
.offset
+ extent_len
- 1, 0);
3919 ret
= btrfs_inc_extent_ref(trans
, root
,
3920 new_extents
[i
].disk_bytenr
,
3921 new_extents
[i
].disk_num_bytes
,
3923 root
->root_key
.objectid
,
3925 key
.objectid
, key
.offset
);
3927 btrfs_release_path(root
, path
);
3929 inode
->i_blocks
+= extent_len
>> 9;
3932 num_bytes
-= extent_len
;
3933 key
.offset
+= extent_len
;
3938 BUG_ON(i
>= nr_extents
);
3941 if (extent_locked
) {
3942 mutex_unlock(&BTRFS_I(inode
)->extent_mutex
);
3943 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
3944 lock_end
, GFP_NOFS
);
3948 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
&&
3949 key
.offset
>= first_pos
+ extent_key
->offset
)
3956 btrfs_release_path(root
, path
);
3958 mutex_unlock(&inode
->i_mutex
);
3959 if (extent_locked
) {
3960 mutex_unlock(&BTRFS_I(inode
)->extent_mutex
);
3961 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
3962 lock_end
, GFP_NOFS
);
3969 int btrfs_add_reloc_mapping(struct btrfs_root
*root
, u64 orig_bytenr
,
3970 u64 num_bytes
, u64 new_bytenr
)
3972 set_extent_bits(&root
->fs_info
->reloc_mapping_tree
,
3973 orig_bytenr
, orig_bytenr
+ num_bytes
- 1,
3974 EXTENT_LOCKED
, GFP_NOFS
);
3975 set_state_private(&root
->fs_info
->reloc_mapping_tree
,
3976 orig_bytenr
, new_bytenr
);
3980 int btrfs_get_reloc_mapping(struct btrfs_root
*root
, u64 orig_bytenr
,
3981 u64 num_bytes
, u64
*new_bytenr
)
3984 u64 cur_bytenr
= orig_bytenr
;
3985 u64 prev_bytenr
= orig_bytenr
;
3989 ret
= get_state_private(&root
->fs_info
->reloc_mapping_tree
,
3990 cur_bytenr
, &bytenr
);
3993 prev_bytenr
= cur_bytenr
;
3994 cur_bytenr
= bytenr
;
3997 if (orig_bytenr
== cur_bytenr
)
4000 if (prev_bytenr
!= orig_bytenr
) {
4001 set_state_private(&root
->fs_info
->reloc_mapping_tree
,
4002 orig_bytenr
, cur_bytenr
);
4004 *new_bytenr
= cur_bytenr
;
4008 void btrfs_free_reloc_mappings(struct btrfs_root
*root
)
4010 clear_extent_bits(&root
->fs_info
->reloc_mapping_tree
,
4011 0, (u64
)-1, -1, GFP_NOFS
);
4014 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle
*trans
,
4015 struct btrfs_root
*root
,
4016 struct extent_buffer
*buf
, u64 orig_start
)
4021 BUG_ON(btrfs_header_generation(buf
) != trans
->transid
);
4022 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
4024 level
= btrfs_header_level(buf
);
4026 struct btrfs_leaf_ref
*ref
;
4027 struct btrfs_leaf_ref
*orig_ref
;
4029 orig_ref
= btrfs_lookup_leaf_ref(root
, orig_start
);
4033 ref
= btrfs_alloc_leaf_ref(root
, orig_ref
->nritems
);
4035 btrfs_free_leaf_ref(root
, orig_ref
);
4039 ref
->nritems
= orig_ref
->nritems
;
4040 memcpy(ref
->extents
, orig_ref
->extents
,
4041 sizeof(ref
->extents
[0]) * ref
->nritems
);
4043 btrfs_free_leaf_ref(root
, orig_ref
);
4045 ref
->root_gen
= trans
->transid
;
4046 ref
->bytenr
= buf
->start
;
4047 ref
->owner
= btrfs_header_owner(buf
);
4048 ref
->generation
= btrfs_header_generation(buf
);
4049 ret
= btrfs_add_leaf_ref(root
, ref
, 0);
4051 btrfs_free_leaf_ref(root
, ref
);
4056 static int noinline
invalidate_extent_cache(struct btrfs_root
*root
,
4057 struct extent_buffer
*leaf
,
4058 struct btrfs_block_group_cache
*group
,
4059 struct btrfs_root
*target_root
)
4061 struct btrfs_key key
;
4062 struct inode
*inode
= NULL
;
4063 struct btrfs_file_extent_item
*fi
;
4065 u64 skip_objectid
= 0;
4069 nritems
= btrfs_header_nritems(leaf
);
4070 for (i
= 0; i
< nritems
; i
++) {
4071 btrfs_item_key_to_cpu(leaf
, &key
, i
);
4072 if (key
.objectid
== skip_objectid
||
4073 key
.type
!= BTRFS_EXTENT_DATA_KEY
)
4075 fi
= btrfs_item_ptr(leaf
, i
, struct btrfs_file_extent_item
);
4076 if (btrfs_file_extent_type(leaf
, fi
) ==
4077 BTRFS_FILE_EXTENT_INLINE
)
4079 if (btrfs_file_extent_disk_bytenr(leaf
, fi
) == 0)
4081 if (!inode
|| inode
->i_ino
!= key
.objectid
) {
4083 inode
= btrfs_ilookup(target_root
->fs_info
->sb
,
4084 key
.objectid
, target_root
, 1);
4087 skip_objectid
= key
.objectid
;
4090 num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
4092 lock_extent(&BTRFS_I(inode
)->io_tree
, key
.offset
,
4093 key
.offset
+ num_bytes
- 1, GFP_NOFS
);
4094 mutex_lock(&BTRFS_I(inode
)->extent_mutex
);
4095 btrfs_drop_extent_cache(inode
, key
.offset
,
4096 key
.offset
+ num_bytes
- 1, 1);
4097 mutex_unlock(&BTRFS_I(inode
)->extent_mutex
);
4098 unlock_extent(&BTRFS_I(inode
)->io_tree
, key
.offset
,
4099 key
.offset
+ num_bytes
- 1, GFP_NOFS
);
4106 static int noinline
replace_extents_in_leaf(struct btrfs_trans_handle
*trans
,
4107 struct btrfs_root
*root
,
4108 struct extent_buffer
*leaf
,
4109 struct btrfs_block_group_cache
*group
,
4110 struct inode
*reloc_inode
)
4112 struct btrfs_key key
;
4113 struct btrfs_key extent_key
;
4114 struct btrfs_file_extent_item
*fi
;
4115 struct btrfs_leaf_ref
*ref
;
4116 struct disk_extent
*new_extent
;
4125 new_extent
= kmalloc(sizeof(*new_extent
), GFP_NOFS
);
4126 BUG_ON(!new_extent
);
4128 ref
= btrfs_lookup_leaf_ref(root
, leaf
->start
);
4132 nritems
= btrfs_header_nritems(leaf
);
4133 for (i
= 0; i
< nritems
; i
++) {
4134 btrfs_item_key_to_cpu(leaf
, &key
, i
);
4135 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
4137 fi
= btrfs_item_ptr(leaf
, i
, struct btrfs_file_extent_item
);
4138 if (btrfs_file_extent_type(leaf
, fi
) ==
4139 BTRFS_FILE_EXTENT_INLINE
)
4141 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
4142 num_bytes
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
4147 if (bytenr
>= group
->key
.objectid
+ group
->key
.offset
||
4148 bytenr
+ num_bytes
<= group
->key
.objectid
)
4151 extent_key
.objectid
= bytenr
;
4152 extent_key
.offset
= num_bytes
;
4153 extent_key
.type
= BTRFS_EXTENT_ITEM_KEY
;
4155 ret
= get_new_locations(reloc_inode
, &extent_key
,
4156 group
->key
.objectid
, 1,
4157 &new_extent
, &nr_extent
);
4162 BUG_ON(ref
->extents
[ext_index
].bytenr
!= bytenr
);
4163 BUG_ON(ref
->extents
[ext_index
].num_bytes
!= num_bytes
);
4164 ref
->extents
[ext_index
].bytenr
= new_extent
->disk_bytenr
;
4165 ref
->extents
[ext_index
].num_bytes
= new_extent
->disk_num_bytes
;
4167 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
4168 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
4169 new_extent
->disk_bytenr
);
4170 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
4171 new_extent
->disk_num_bytes
);
4172 new_extent
->offset
+= btrfs_file_extent_offset(leaf
, fi
);
4173 btrfs_set_file_extent_offset(leaf
, fi
, new_extent
->offset
);
4174 btrfs_mark_buffer_dirty(leaf
);
4176 ret
= btrfs_inc_extent_ref(trans
, root
,
4177 new_extent
->disk_bytenr
,
4178 new_extent
->disk_num_bytes
,
4180 root
->root_key
.objectid
,
4182 key
.objectid
, key
.offset
);
4184 ret
= btrfs_free_extent(trans
, root
,
4185 bytenr
, num_bytes
, leaf
->start
,
4186 btrfs_header_owner(leaf
),
4187 btrfs_header_generation(leaf
),
4188 key
.objectid
, key
.offset
, 0);
4193 BUG_ON(ext_index
+ 1 != ref
->nritems
);
4194 btrfs_free_leaf_ref(root
, ref
);
4198 int btrfs_free_reloc_root(struct btrfs_root
*root
)
4200 struct btrfs_root
*reloc_root
;
4202 if (root
->reloc_root
) {
4203 reloc_root
= root
->reloc_root
;
4204 root
->reloc_root
= NULL
;
4205 list_add(&reloc_root
->dead_list
,
4206 &root
->fs_info
->dead_reloc_roots
);
4211 int btrfs_drop_dead_reloc_roots(struct btrfs_root
*root
)
4213 struct btrfs_trans_handle
*trans
;
4214 struct btrfs_root
*reloc_root
;
4215 struct btrfs_root
*prev_root
= NULL
;
4216 struct list_head dead_roots
;
4220 INIT_LIST_HEAD(&dead_roots
);
4221 list_splice_init(&root
->fs_info
->dead_reloc_roots
, &dead_roots
);
4223 while (!list_empty(&dead_roots
)) {
4224 reloc_root
= list_entry(dead_roots
.prev
,
4225 struct btrfs_root
, dead_list
);
4226 list_del_init(&reloc_root
->dead_list
);
4228 BUG_ON(reloc_root
->commit_root
!= NULL
);
4230 trans
= btrfs_join_transaction(root
, 1);
4233 mutex_lock(&root
->fs_info
->drop_mutex
);
4234 ret
= btrfs_drop_snapshot(trans
, reloc_root
);
4237 mutex_unlock(&root
->fs_info
->drop_mutex
);
4239 nr
= trans
->blocks_used
;
4240 ret
= btrfs_end_transaction(trans
, root
);
4242 btrfs_btree_balance_dirty(root
, nr
);
4245 free_extent_buffer(reloc_root
->node
);
4247 ret
= btrfs_del_root(trans
, root
->fs_info
->tree_root
,
4248 &reloc_root
->root_key
);
4250 mutex_unlock(&root
->fs_info
->drop_mutex
);
4252 nr
= trans
->blocks_used
;
4253 ret
= btrfs_end_transaction(trans
, root
);
4255 btrfs_btree_balance_dirty(root
, nr
);
4258 prev_root
= reloc_root
;
4261 btrfs_remove_leaf_refs(prev_root
, (u64
)-1, 0);
4267 int btrfs_add_dead_reloc_root(struct btrfs_root
*root
)
4269 list_add(&root
->dead_list
, &root
->fs_info
->dead_reloc_roots
);
4273 int btrfs_cleanup_reloc_trees(struct btrfs_root
*root
)
4275 struct btrfs_root
*reloc_root
;
4276 struct btrfs_trans_handle
*trans
;
4277 struct btrfs_key location
;
4281 mutex_lock(&root
->fs_info
->tree_reloc_mutex
);
4282 ret
= btrfs_find_dead_roots(root
, BTRFS_TREE_RELOC_OBJECTID
, NULL
);
4284 found
= !list_empty(&root
->fs_info
->dead_reloc_roots
);
4285 mutex_unlock(&root
->fs_info
->tree_reloc_mutex
);
4288 trans
= btrfs_start_transaction(root
, 1);
4290 ret
= btrfs_commit_transaction(trans
, root
);
4294 location
.objectid
= BTRFS_DATA_RELOC_TREE_OBJECTID
;
4295 location
.offset
= (u64
)-1;
4296 location
.type
= BTRFS_ROOT_ITEM_KEY
;
4298 reloc_root
= btrfs_read_fs_root_no_name(root
->fs_info
, &location
);
4299 BUG_ON(!reloc_root
);
4300 btrfs_orphan_cleanup(reloc_root
);
4304 static int noinline
init_reloc_tree(struct btrfs_trans_handle
*trans
,
4305 struct btrfs_root
*root
)
4307 struct btrfs_root
*reloc_root
;
4308 struct extent_buffer
*eb
;
4309 struct btrfs_root_item
*root_item
;
4310 struct btrfs_key root_key
;
4313 BUG_ON(!root
->ref_cows
);
4314 if (root
->reloc_root
)
4317 root_item
= kmalloc(sizeof(*root_item
), GFP_NOFS
);
4320 ret
= btrfs_copy_root(trans
, root
, root
->commit_root
,
4321 &eb
, BTRFS_TREE_RELOC_OBJECTID
);
4324 root_key
.objectid
= BTRFS_TREE_RELOC_OBJECTID
;
4325 root_key
.offset
= root
->root_key
.objectid
;
4326 root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
4328 memcpy(root_item
, &root
->root_item
, sizeof(root_item
));
4329 btrfs_set_root_refs(root_item
, 0);
4330 btrfs_set_root_bytenr(root_item
, eb
->start
);
4331 btrfs_set_root_level(root_item
, btrfs_header_level(eb
));
4332 memset(&root_item
->drop_progress
, 0, sizeof(root_item
->drop_progress
));
4333 root_item
->drop_level
= 0;
4335 btrfs_tree_unlock(eb
);
4336 free_extent_buffer(eb
);
4338 ret
= btrfs_insert_root(trans
, root
->fs_info
->tree_root
,
4339 &root_key
, root_item
);
4343 reloc_root
= btrfs_read_fs_root_no_radix(root
->fs_info
->tree_root
,
4345 BUG_ON(!reloc_root
);
4346 reloc_root
->last_trans
= trans
->transid
;
4347 reloc_root
->commit_root
= NULL
;
4348 reloc_root
->ref_tree
= &root
->fs_info
->reloc_ref_tree
;
4350 root
->reloc_root
= reloc_root
;
4355 * Core function of space balance.
4357 * The idea is using reloc trees to relocate tree blocks in reference
4358 * counted roots. There is one reloc tree for each subvol, all reloc
4359 * trees share same key objectid. Reloc trees are snapshots of the
4360 * latest committed roots (subvol root->commit_root). To relocate a tree
4361 * block referenced by a subvol, the code COW the block through the reloc
4362 * tree, then update pointer in the subvol to point to the new block.
4363 * Since all reloc trees share same key objectid, we can easily do special
4364 * handing to share tree blocks between reloc trees. Once a tree block has
4365 * been COWed in one reloc tree, we can use the result when the same block
4366 * is COWed again through other reloc trees.
4368 static int noinline
relocate_one_path(struct btrfs_trans_handle
*trans
,
4369 struct btrfs_root
*root
,
4370 struct btrfs_path
*path
,
4371 struct btrfs_key
*first_key
,
4372 struct btrfs_ref_path
*ref_path
,
4373 struct btrfs_block_group_cache
*group
,
4374 struct inode
*reloc_inode
)
4376 struct btrfs_root
*reloc_root
;
4377 struct extent_buffer
*eb
= NULL
;
4378 struct btrfs_key
*keys
;
4382 int lowest_level
= 0;
4386 if (ref_path
->owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
)
4387 lowest_level
= ref_path
->owner_objectid
;
4389 if (is_cowonly_root(ref_path
->root_objectid
)) {
4390 path
->lowest_level
= lowest_level
;
4391 ret
= btrfs_search_slot(trans
, root
, first_key
, path
, 0, 1);
4393 path
->lowest_level
= 0;
4394 btrfs_release_path(root
, path
);
4398 keys
= kzalloc(sizeof(*keys
) * BTRFS_MAX_LEVEL
, GFP_NOFS
);
4400 nodes
= kzalloc(sizeof(*nodes
) * BTRFS_MAX_LEVEL
, GFP_NOFS
);
4403 mutex_lock(&root
->fs_info
->tree_reloc_mutex
);
4404 ret
= init_reloc_tree(trans
, root
);
4406 reloc_root
= root
->reloc_root
;
4408 path
->lowest_level
= lowest_level
;
4409 ret
= btrfs_search_slot(trans
, reloc_root
, first_key
, path
, 0, 0);
4412 * get relocation mapping for tree blocks in the path
4414 lowest_merge
= BTRFS_MAX_LEVEL
;
4415 for (level
= BTRFS_MAX_LEVEL
- 1; level
>= lowest_level
; level
--) {
4417 eb
= path
->nodes
[level
];
4418 if (!eb
|| eb
== reloc_root
->node
)
4420 ret
= btrfs_get_reloc_mapping(reloc_root
, eb
->start
, eb
->len
,
4425 btrfs_item_key_to_cpu(eb
, &keys
[level
], 0);
4427 btrfs_node_key_to_cpu(eb
, &keys
[level
], 0);
4428 nodes
[level
] = new_bytenr
;
4429 lowest_merge
= level
;
4433 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
4434 eb
= path
->nodes
[0];
4435 if (btrfs_header_generation(eb
) < trans
->transid
)
4439 btrfs_release_path(reloc_root
, path
);
4441 * merge tree blocks that already relocated in other reloc trees
4443 if (lowest_merge
!= BTRFS_MAX_LEVEL
) {
4444 ret
= btrfs_merge_path(trans
, reloc_root
, keys
, nodes
,
4449 * cow any tree blocks that still haven't been relocated
4451 ret
= btrfs_search_slot(trans
, reloc_root
, first_key
, path
, 0, 1);
4454 * if we are relocating data block group, update extent pointers
4455 * in the newly created tree leaf.
4457 eb
= path
->nodes
[0];
4458 if (update_refs
&& nodes
[0] != eb
->start
) {
4459 ret
= replace_extents_in_leaf(trans
, reloc_root
, eb
, group
,
4464 memset(keys
, 0, sizeof(*keys
) * BTRFS_MAX_LEVEL
);
4465 memset(nodes
, 0, sizeof(*nodes
) * BTRFS_MAX_LEVEL
);
4466 for (level
= BTRFS_MAX_LEVEL
- 1; level
>= lowest_level
; level
--) {
4467 eb
= path
->nodes
[level
];
4468 if (!eb
|| eb
== reloc_root
->node
)
4470 BUG_ON(btrfs_header_owner(eb
) != BTRFS_TREE_RELOC_OBJECTID
);
4471 nodes
[level
] = eb
->start
;
4473 btrfs_item_key_to_cpu(eb
, &keys
[level
], 0);
4475 btrfs_node_key_to_cpu(eb
, &keys
[level
], 0);
4478 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
4479 eb
= path
->nodes
[0];
4480 extent_buffer_get(eb
);
4482 btrfs_release_path(reloc_root
, path
);
4484 * replace tree blocks in the fs tree with tree blocks in
4487 ret
= btrfs_merge_path(trans
, root
, keys
, nodes
, lowest_level
);
4490 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
4491 ret
= invalidate_extent_cache(reloc_root
, eb
, group
, root
);
4493 free_extent_buffer(eb
);
4495 mutex_unlock(&root
->fs_info
->tree_reloc_mutex
);
4497 path
->lowest_level
= 0;
4503 static int noinline
relocate_tree_block(struct btrfs_trans_handle
*trans
,
4504 struct btrfs_root
*root
,
4505 struct btrfs_path
*path
,
4506 struct btrfs_key
*first_key
,
4507 struct btrfs_ref_path
*ref_path
)
4512 if (root
== root
->fs_info
->extent_root
||
4513 root
== root
->fs_info
->chunk_root
||
4514 root
== root
->fs_info
->dev_root
) {
4516 mutex_lock(&root
->fs_info
->alloc_mutex
);
4519 ret
= relocate_one_path(trans
, root
, path
, first_key
,
4520 ref_path
, NULL
, NULL
);
4523 if (root
== root
->fs_info
->extent_root
)
4524 btrfs_extent_post_op(trans
, root
);
4526 mutex_unlock(&root
->fs_info
->alloc_mutex
);
4531 static int noinline
del_extent_zero(struct btrfs_trans_handle
*trans
,
4532 struct btrfs_root
*extent_root
,
4533 struct btrfs_path
*path
,
4534 struct btrfs_key
*extent_key
)
4538 mutex_lock(&extent_root
->fs_info
->alloc_mutex
);
4539 ret
= btrfs_search_slot(trans
, extent_root
, extent_key
, path
, -1, 1);
4542 ret
= btrfs_del_item(trans
, extent_root
, path
);
4544 btrfs_release_path(extent_root
, path
);
4545 mutex_unlock(&extent_root
->fs_info
->alloc_mutex
);
4549 static struct btrfs_root noinline
*read_ref_root(struct btrfs_fs_info
*fs_info
,
4550 struct btrfs_ref_path
*ref_path
)
4552 struct btrfs_key root_key
;
4554 root_key
.objectid
= ref_path
->root_objectid
;
4555 root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
4556 if (is_cowonly_root(ref_path
->root_objectid
))
4557 root_key
.offset
= 0;
4559 root_key
.offset
= (u64
)-1;
4561 return btrfs_read_fs_root_no_name(fs_info
, &root_key
);
4564 static int noinline
relocate_one_extent(struct btrfs_root
*extent_root
,
4565 struct btrfs_path
*path
,
4566 struct btrfs_key
*extent_key
,
4567 struct btrfs_block_group_cache
*group
,
4568 struct inode
*reloc_inode
, int pass
)
4570 struct btrfs_trans_handle
*trans
;
4571 struct btrfs_root
*found_root
;
4572 struct btrfs_ref_path
*ref_path
= NULL
;
4573 struct disk_extent
*new_extents
= NULL
;
4578 struct btrfs_key first_key
;
4581 mutex_unlock(&extent_root
->fs_info
->alloc_mutex
);
4583 trans
= btrfs_start_transaction(extent_root
, 1);
4586 if (extent_key
->objectid
== 0) {
4587 ret
= del_extent_zero(trans
, extent_root
, path
, extent_key
);
4591 ref_path
= kmalloc(sizeof(*ref_path
), GFP_NOFS
);
4597 for (loops
= 0; ; loops
++) {
4599 ret
= btrfs_first_ref_path(trans
, extent_root
, ref_path
,
4600 extent_key
->objectid
);
4602 ret
= btrfs_next_ref_path(trans
, extent_root
, ref_path
);
4609 if (ref_path
->root_objectid
== BTRFS_TREE_LOG_OBJECTID
||
4610 ref_path
->root_objectid
== BTRFS_TREE_RELOC_OBJECTID
)
4613 found_root
= read_ref_root(extent_root
->fs_info
, ref_path
);
4614 BUG_ON(!found_root
);
4616 * for reference counted tree, only process reference paths
4617 * rooted at the latest committed root.
4619 if (found_root
->ref_cows
&&
4620 ref_path
->root_generation
!= found_root
->root_key
.offset
)
4623 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
4626 * copy data extents to new locations
4628 u64 group_start
= group
->key
.objectid
;
4629 ret
= relocate_data_extent(reloc_inode
,
4638 level
= ref_path
->owner_objectid
;
4641 if (prev_block
!= ref_path
->nodes
[level
]) {
4642 struct extent_buffer
*eb
;
4643 u64 block_start
= ref_path
->nodes
[level
];
4644 u64 block_size
= btrfs_level_size(found_root
, level
);
4646 eb
= read_tree_block(found_root
, block_start
,
4648 btrfs_tree_lock(eb
);
4649 BUG_ON(level
!= btrfs_header_level(eb
));
4652 btrfs_item_key_to_cpu(eb
, &first_key
, 0);
4654 btrfs_node_key_to_cpu(eb
, &first_key
, 0);
4656 btrfs_tree_unlock(eb
);
4657 free_extent_buffer(eb
);
4658 prev_block
= block_start
;
4661 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
&&
4664 * use fallback method to process the remaining
4668 u64 group_start
= group
->key
.objectid
;
4669 ret
= get_new_locations(reloc_inode
,
4677 btrfs_record_root_in_trans(found_root
);
4678 ret
= replace_one_extent(trans
, found_root
,
4680 &first_key
, ref_path
,
4681 new_extents
, nr_extents
);
4687 btrfs_record_root_in_trans(found_root
);
4688 if (ref_path
->owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
) {
4689 ret
= relocate_tree_block(trans
, found_root
, path
,
4690 &first_key
, ref_path
);
4693 * try to update data extent references while
4694 * keeping metadata shared between snapshots.
4696 ret
= relocate_one_path(trans
, found_root
, path
,
4697 &first_key
, ref_path
,
4698 group
, reloc_inode
);
4705 btrfs_end_transaction(trans
, extent_root
);
4708 mutex_lock(&extent_root
->fs_info
->alloc_mutex
);
4712 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
4715 u64 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
4716 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
4718 num_devices
= root
->fs_info
->fs_devices
->num_devices
;
4719 if (num_devices
== 1) {
4720 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
4721 stripped
= flags
& ~stripped
;
4723 /* turn raid0 into single device chunks */
4724 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
4727 /* turn mirroring into duplication */
4728 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
4729 BTRFS_BLOCK_GROUP_RAID10
))
4730 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
4733 /* they already had raid on here, just return */
4734 if (flags
& stripped
)
4737 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
4738 stripped
= flags
& ~stripped
;
4740 /* switch duplicated blocks with raid1 */
4741 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
4742 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
4744 /* turn single device chunks into raid0 */
4745 return stripped
| BTRFS_BLOCK_GROUP_RAID0
;
4750 int __alloc_chunk_for_shrink(struct btrfs_root
*root
,
4751 struct btrfs_block_group_cache
*shrink_block_group
,
4754 struct btrfs_trans_handle
*trans
;
4755 u64 new_alloc_flags
;
4758 spin_lock(&shrink_block_group
->lock
);
4759 if (btrfs_block_group_used(&shrink_block_group
->item
) > 0) {
4760 spin_unlock(&shrink_block_group
->lock
);
4761 mutex_unlock(&root
->fs_info
->alloc_mutex
);
4763 trans
= btrfs_start_transaction(root
, 1);
4764 mutex_lock(&root
->fs_info
->alloc_mutex
);
4765 spin_lock(&shrink_block_group
->lock
);
4767 new_alloc_flags
= update_block_group_flags(root
,
4768 shrink_block_group
->flags
);
4769 if (new_alloc_flags
!= shrink_block_group
->flags
) {
4771 btrfs_block_group_used(&shrink_block_group
->item
);
4773 calc
= shrink_block_group
->key
.offset
;
4775 spin_unlock(&shrink_block_group
->lock
);
4777 do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
4778 calc
+ 2 * 1024 * 1024, new_alloc_flags
, force
);
4780 mutex_unlock(&root
->fs_info
->alloc_mutex
);
4781 btrfs_end_transaction(trans
, root
);
4782 mutex_lock(&root
->fs_info
->alloc_mutex
);
4784 spin_unlock(&shrink_block_group
->lock
);
4788 static int __insert_orphan_inode(struct btrfs_trans_handle
*trans
,
4789 struct btrfs_root
*root
,
4790 u64 objectid
, u64 size
)
4792 struct btrfs_path
*path
;
4793 struct btrfs_inode_item
*item
;
4794 struct extent_buffer
*leaf
;
4797 path
= btrfs_alloc_path();
4801 ret
= btrfs_insert_empty_inode(trans
, root
, path
, objectid
);
4805 leaf
= path
->nodes
[0];
4806 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_inode_item
);
4807 memset_extent_buffer(leaf
, 0, (unsigned long)item
, sizeof(*item
));
4808 btrfs_set_inode_generation(leaf
, item
, 1);
4809 btrfs_set_inode_size(leaf
, item
, size
);
4810 btrfs_set_inode_mode(leaf
, item
, S_IFREG
| 0600);
4811 btrfs_set_inode_flags(leaf
, item
, BTRFS_INODE_NODATASUM
);
4812 btrfs_mark_buffer_dirty(leaf
);
4813 btrfs_release_path(root
, path
);
4815 btrfs_free_path(path
);
4819 static struct inode noinline
*create_reloc_inode(struct btrfs_fs_info
*fs_info
,
4820 struct btrfs_block_group_cache
*group
)
4822 struct inode
*inode
= NULL
;
4823 struct btrfs_trans_handle
*trans
;
4824 struct btrfs_root
*root
;
4825 struct btrfs_key root_key
;
4826 u64 objectid
= BTRFS_FIRST_FREE_OBJECTID
;
4829 root_key
.objectid
= BTRFS_DATA_RELOC_TREE_OBJECTID
;
4830 root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
4831 root_key
.offset
= (u64
)-1;
4832 root
= btrfs_read_fs_root_no_name(fs_info
, &root_key
);
4834 return ERR_CAST(root
);
4836 trans
= btrfs_start_transaction(root
, 1);
4839 err
= btrfs_find_free_objectid(trans
, root
, objectid
, &objectid
);
4843 err
= __insert_orphan_inode(trans
, root
, objectid
, group
->key
.offset
);
4846 err
= btrfs_insert_file_extent(trans
, root
, objectid
, 0, 0, 0,
4847 group
->key
.offset
, 0);
4850 inode
= btrfs_iget_locked(root
->fs_info
->sb
, objectid
, root
);
4851 if (inode
->i_state
& I_NEW
) {
4852 BTRFS_I(inode
)->root
= root
;
4853 BTRFS_I(inode
)->location
.objectid
= objectid
;
4854 BTRFS_I(inode
)->location
.type
= BTRFS_INODE_ITEM_KEY
;
4855 BTRFS_I(inode
)->location
.offset
= 0;
4856 btrfs_read_locked_inode(inode
);
4857 unlock_new_inode(inode
);
4858 BUG_ON(is_bad_inode(inode
));
4863 err
= btrfs_orphan_add(trans
, inode
);
4865 btrfs_end_transaction(trans
, root
);
4869 inode
= ERR_PTR(err
);
4874 int btrfs_relocate_block_group(struct btrfs_root
*root
, u64 group_start
)
4876 struct btrfs_trans_handle
*trans
;
4877 struct btrfs_path
*path
;
4878 struct btrfs_fs_info
*info
= root
->fs_info
;
4879 struct extent_buffer
*leaf
;
4880 struct inode
*reloc_inode
;
4881 struct btrfs_block_group_cache
*block_group
;
4882 struct btrfs_key key
;
4890 root
= root
->fs_info
->extent_root
;
4892 block_group
= btrfs_lookup_block_group(info
, group_start
);
4893 BUG_ON(!block_group
);
4895 printk("btrfs relocating block group %llu flags %llu\n",
4896 (unsigned long long)block_group
->key
.objectid
,
4897 (unsigned long long)block_group
->flags
);
4899 path
= btrfs_alloc_path();
4902 reloc_inode
= create_reloc_inode(info
, block_group
);
4903 BUG_ON(IS_ERR(reloc_inode
));
4905 mutex_lock(&root
->fs_info
->alloc_mutex
);
4907 __alloc_chunk_for_shrink(root
, block_group
, 1);
4908 block_group
->ro
= 1;
4909 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
4911 mutex_unlock(&root
->fs_info
->alloc_mutex
);
4913 btrfs_start_delalloc_inodes(info
->tree_root
);
4914 btrfs_wait_ordered_extents(info
->tree_root
, 0);
4918 key
.objectid
= block_group
->key
.objectid
;
4921 cur_byte
= key
.objectid
;
4923 trans
= btrfs_start_transaction(info
->tree_root
, 1);
4924 btrfs_commit_transaction(trans
, info
->tree_root
);
4926 mutex_lock(&root
->fs_info
->cleaner_mutex
);
4927 btrfs_clean_old_snapshots(info
->tree_root
);
4928 btrfs_remove_leaf_refs(info
->tree_root
, (u64
)-1, 1);
4929 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
4931 mutex_lock(&root
->fs_info
->alloc_mutex
);
4934 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4938 leaf
= path
->nodes
[0];
4939 nritems
= btrfs_header_nritems(leaf
);
4940 if (path
->slots
[0] >= nritems
) {
4941 ret
= btrfs_next_leaf(root
, path
);
4948 leaf
= path
->nodes
[0];
4949 nritems
= btrfs_header_nritems(leaf
);
4952 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
4954 if (key
.objectid
>= block_group
->key
.objectid
+
4955 block_group
->key
.offset
)
4958 if (progress
&& need_resched()) {
4959 btrfs_release_path(root
, path
);
4960 mutex_unlock(&root
->fs_info
->alloc_mutex
);
4962 mutex_lock(&root
->fs_info
->alloc_mutex
);
4968 if (btrfs_key_type(&key
) != BTRFS_EXTENT_ITEM_KEY
||
4969 key
.objectid
+ key
.offset
<= cur_byte
) {
4975 cur_byte
= key
.objectid
+ key
.offset
;
4976 btrfs_release_path(root
, path
);
4978 __alloc_chunk_for_shrink(root
, block_group
, 0);
4979 ret
= relocate_one_extent(root
, path
, &key
, block_group
,
4983 key
.objectid
= cur_byte
;
4988 btrfs_release_path(root
, path
);
4989 mutex_unlock(&root
->fs_info
->alloc_mutex
);
4992 btrfs_wait_ordered_range(reloc_inode
, 0, (u64
)-1);
4993 invalidate_mapping_pages(reloc_inode
->i_mapping
, 0, -1);
4994 WARN_ON(reloc_inode
->i_mapping
->nrpages
);
4997 if (total_found
> 0) {
4998 printk("btrfs found %llu extents in pass %d\n",
4999 (unsigned long long)total_found
, pass
);
5004 /* delete reloc_inode */
5007 /* unpin extents in this range */
5008 trans
= btrfs_start_transaction(info
->tree_root
, 1);
5009 btrfs_commit_transaction(trans
, info
->tree_root
);
5011 mutex_lock(&root
->fs_info
->alloc_mutex
);
5013 spin_lock(&block_group
->lock
);
5014 WARN_ON(block_group
->pinned
> 0);
5015 WARN_ON(block_group
->reserved
> 0);
5016 WARN_ON(btrfs_block_group_used(&block_group
->item
) > 0);
5017 spin_unlock(&block_group
->lock
);
5020 mutex_unlock(&root
->fs_info
->alloc_mutex
);
5021 btrfs_free_path(path
);
5025 int find_first_block_group(struct btrfs_root
*root
, struct btrfs_path
*path
,
5026 struct btrfs_key
*key
)
5029 struct btrfs_key found_key
;
5030 struct extent_buffer
*leaf
;
5033 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
5038 slot
= path
->slots
[0];
5039 leaf
= path
->nodes
[0];
5040 if (slot
>= btrfs_header_nritems(leaf
)) {
5041 ret
= btrfs_next_leaf(root
, path
);
5048 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
5050 if (found_key
.objectid
>= key
->objectid
&&
5051 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
5062 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
5064 struct btrfs_block_group_cache
*block_group
;
5067 mutex_lock(&info
->alloc_mutex
);
5068 spin_lock(&info
->block_group_cache_lock
);
5069 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
5070 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
5073 spin_unlock(&info
->block_group_cache_lock
);
5074 btrfs_remove_free_space_cache(block_group
);
5075 spin_lock(&info
->block_group_cache_lock
);
5077 rb_erase(&block_group
->cache_node
,
5078 &info
->block_group_cache_tree
);
5079 spin_lock(&block_group
->space_info
->lock
);
5080 list_del(&block_group
->list
);
5081 spin_unlock(&block_group
->space_info
->lock
);
5084 spin_unlock(&info
->block_group_cache_lock
);
5085 mutex_unlock(&info
->alloc_mutex
);
5089 int btrfs_read_block_groups(struct btrfs_root
*root
)
5091 struct btrfs_path
*path
;
5093 struct btrfs_block_group_cache
*cache
;
5094 struct btrfs_fs_info
*info
= root
->fs_info
;
5095 struct btrfs_space_info
*space_info
;
5096 struct btrfs_key key
;
5097 struct btrfs_key found_key
;
5098 struct extent_buffer
*leaf
;
5100 root
= info
->extent_root
;
5103 btrfs_set_key_type(&key
, BTRFS_BLOCK_GROUP_ITEM_KEY
);
5104 path
= btrfs_alloc_path();
5108 mutex_lock(&root
->fs_info
->alloc_mutex
);
5110 ret
= find_first_block_group(root
, path
, &key
);
5118 leaf
= path
->nodes
[0];
5119 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5120 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
5126 spin_lock_init(&cache
->lock
);
5127 INIT_LIST_HEAD(&cache
->list
);
5128 read_extent_buffer(leaf
, &cache
->item
,
5129 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
5130 sizeof(cache
->item
));
5131 memcpy(&cache
->key
, &found_key
, sizeof(found_key
));
5133 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
5134 btrfs_release_path(root
, path
);
5135 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
5137 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
5138 btrfs_block_group_used(&cache
->item
),
5141 cache
->space_info
= space_info
;
5142 spin_lock(&space_info
->lock
);
5143 list_add(&cache
->list
, &space_info
->block_groups
);
5144 spin_unlock(&space_info
->lock
);
5146 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
5149 set_avail_alloc_bits(root
->fs_info
, cache
->flags
);
5153 btrfs_free_path(path
);
5154 mutex_unlock(&root
->fs_info
->alloc_mutex
);
5158 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
5159 struct btrfs_root
*root
, u64 bytes_used
,
5160 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
5164 struct btrfs_root
*extent_root
;
5165 struct btrfs_block_group_cache
*cache
;
5167 WARN_ON(!mutex_is_locked(&root
->fs_info
->alloc_mutex
));
5168 extent_root
= root
->fs_info
->extent_root
;
5170 root
->fs_info
->last_trans_new_blockgroup
= trans
->transid
;
5172 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
5176 cache
->key
.objectid
= chunk_offset
;
5177 cache
->key
.offset
= size
;
5178 spin_lock_init(&cache
->lock
);
5179 INIT_LIST_HEAD(&cache
->list
);
5180 btrfs_set_key_type(&cache
->key
, BTRFS_BLOCK_GROUP_ITEM_KEY
);
5182 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
5183 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
5184 cache
->flags
= type
;
5185 btrfs_set_block_group_flags(&cache
->item
, type
);
5187 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
5188 &cache
->space_info
);
5190 spin_lock(&cache
->space_info
->lock
);
5191 list_add(&cache
->list
, &cache
->space_info
->block_groups
);
5192 spin_unlock(&cache
->space_info
->lock
);
5194 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
5197 ret
= btrfs_insert_item(trans
, extent_root
, &cache
->key
, &cache
->item
,
5198 sizeof(cache
->item
));
5201 finish_current_insert(trans
, extent_root
);
5202 ret
= del_pending_extents(trans
, extent_root
);
5204 set_avail_alloc_bits(extent_root
->fs_info
, type
);
5209 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
5210 struct btrfs_root
*root
, u64 group_start
)
5212 struct btrfs_path
*path
;
5213 struct btrfs_block_group_cache
*block_group
;
5214 struct btrfs_key key
;
5217 BUG_ON(!mutex_is_locked(&root
->fs_info
->alloc_mutex
));
5218 root
= root
->fs_info
->extent_root
;
5220 block_group
= btrfs_lookup_block_group(root
->fs_info
, group_start
);
5221 BUG_ON(!block_group
);
5223 memcpy(&key
, &block_group
->key
, sizeof(key
));
5225 path
= btrfs_alloc_path();
5228 btrfs_remove_free_space_cache(block_group
);
5229 rb_erase(&block_group
->cache_node
,
5230 &root
->fs_info
->block_group_cache_tree
);
5231 spin_lock(&block_group
->space_info
->lock
);
5232 list_del(&block_group
->list
);
5233 spin_unlock(&block_group
->space_info
->lock
);
5236 memset(shrink_block_group, 0, sizeof(*shrink_block_group));
5237 kfree(shrink_block_group);
5240 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
5246 ret
= btrfs_del_item(trans
, root
, path
);
5248 btrfs_free_path(path
);