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>
25 #include "print-tree.h"
26 #include "transaction.h"
29 #include "ref-cache.h"
31 #define BLOCK_GROUP_DATA EXTENT_WRITEBACK
32 #define BLOCK_GROUP_METADATA EXTENT_UPTODATE
33 #define BLOCK_GROUP_SYSTEM EXTENT_NEW
35 #define BLOCK_GROUP_DIRTY EXTENT_DIRTY
37 static int finish_current_insert(struct btrfs_trans_handle
*trans
, struct
38 btrfs_root
*extent_root
);
39 static int del_pending_extents(struct btrfs_trans_handle
*trans
, struct
40 btrfs_root
*extent_root
);
41 static struct btrfs_block_group_cache
*
42 __btrfs_find_block_group(struct btrfs_root
*root
,
43 struct btrfs_block_group_cache
*hint
,
44 u64 search_start
, int data
, int owner
);
46 void maybe_lock_mutex(struct btrfs_root
*root
)
48 if (root
!= root
->fs_info
->extent_root
&&
49 root
!= root
->fs_info
->chunk_root
&&
50 root
!= root
->fs_info
->dev_root
) {
51 mutex_lock(&root
->fs_info
->alloc_mutex
);
55 void maybe_unlock_mutex(struct btrfs_root
*root
)
57 if (root
!= root
->fs_info
->extent_root
&&
58 root
!= root
->fs_info
->chunk_root
&&
59 root
!= root
->fs_info
->dev_root
) {
60 mutex_unlock(&root
->fs_info
->alloc_mutex
);
64 static int cache_block_group(struct btrfs_root
*root
,
65 struct btrfs_block_group_cache
*block_group
)
67 struct btrfs_path
*path
;
70 struct extent_buffer
*leaf
;
71 struct extent_io_tree
*free_space_cache
;
81 root
= root
->fs_info
->extent_root
;
82 free_space_cache
= &root
->fs_info
->free_space_cache
;
84 if (block_group
->cached
)
87 path
= btrfs_alloc_path();
93 * we get into deadlocks with paths held by callers of this function.
94 * since the alloc_mutex is protecting things right now, just
95 * skip the locking here
97 path
->skip_locking
= 1;
98 first_free
= block_group
->key
.objectid
;
99 key
.objectid
= block_group
->key
.objectid
;
101 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
102 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
105 ret
= btrfs_previous_item(root
, path
, 0, BTRFS_EXTENT_ITEM_KEY
);
109 leaf
= path
->nodes
[0];
110 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
111 if (key
.objectid
+ key
.offset
> first_free
)
112 first_free
= key
.objectid
+ key
.offset
;
115 leaf
= path
->nodes
[0];
116 slot
= path
->slots
[0];
117 if (slot
>= btrfs_header_nritems(leaf
)) {
118 ret
= btrfs_next_leaf(root
, path
);
127 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
128 if (key
.objectid
< block_group
->key
.objectid
) {
131 if (key
.objectid
>= block_group
->key
.objectid
+
132 block_group
->key
.offset
) {
136 if (btrfs_key_type(&key
) == BTRFS_EXTENT_ITEM_KEY
) {
141 if (key
.objectid
> last
) {
142 hole_size
= key
.objectid
- last
;
143 set_extent_dirty(free_space_cache
, last
,
144 last
+ hole_size
- 1,
147 last
= key
.objectid
+ key
.offset
;
155 if (block_group
->key
.objectid
+
156 block_group
->key
.offset
> last
) {
157 hole_size
= block_group
->key
.objectid
+
158 block_group
->key
.offset
- last
;
159 set_extent_dirty(free_space_cache
, last
,
160 last
+ hole_size
- 1, GFP_NOFS
);
162 block_group
->cached
= 1;
164 btrfs_free_path(path
);
168 struct btrfs_block_group_cache
*btrfs_lookup_first_block_group(struct
172 struct extent_io_tree
*block_group_cache
;
173 struct btrfs_block_group_cache
*block_group
= NULL
;
179 bytenr
= max_t(u64
, bytenr
,
180 BTRFS_SUPER_INFO_OFFSET
+ BTRFS_SUPER_INFO_SIZE
);
181 block_group_cache
= &info
->block_group_cache
;
182 ret
= find_first_extent_bit(block_group_cache
,
183 bytenr
, &start
, &end
,
184 BLOCK_GROUP_DATA
| BLOCK_GROUP_METADATA
|
189 ret
= get_state_private(block_group_cache
, start
, &ptr
);
193 block_group
= (struct btrfs_block_group_cache
*)(unsigned long)ptr
;
197 struct btrfs_block_group_cache
*btrfs_lookup_block_group(struct
201 struct extent_io_tree
*block_group_cache
;
202 struct btrfs_block_group_cache
*block_group
= NULL
;
208 bytenr
= max_t(u64
, bytenr
,
209 BTRFS_SUPER_INFO_OFFSET
+ BTRFS_SUPER_INFO_SIZE
);
210 block_group_cache
= &info
->block_group_cache
;
211 ret
= find_first_extent_bit(block_group_cache
,
212 bytenr
, &start
, &end
,
213 BLOCK_GROUP_DATA
| BLOCK_GROUP_METADATA
|
218 ret
= get_state_private(block_group_cache
, start
, &ptr
);
222 block_group
= (struct btrfs_block_group_cache
*)(unsigned long)ptr
;
223 if (block_group
->key
.objectid
<= bytenr
&& bytenr
<
224 block_group
->key
.objectid
+ block_group
->key
.offset
)
229 static int block_group_bits(struct btrfs_block_group_cache
*cache
, u64 bits
)
231 return (cache
->flags
& bits
) == bits
;
234 static int noinline
find_search_start(struct btrfs_root
*root
,
235 struct btrfs_block_group_cache
**cache_ret
,
236 u64
*start_ret
, u64 num
, int data
)
239 struct btrfs_block_group_cache
*cache
= *cache_ret
;
240 struct extent_io_tree
*free_space_cache
;
241 struct extent_state
*state
;
246 u64 search_start
= *start_ret
;
249 WARN_ON(!mutex_is_locked(&root
->fs_info
->alloc_mutex
));
250 total_fs_bytes
= btrfs_super_total_bytes(&root
->fs_info
->super_copy
);
251 free_space_cache
= &root
->fs_info
->free_space_cache
;
257 ret
= cache_block_group(root
, cache
);
262 last
= max(search_start
, cache
->key
.objectid
);
263 if (!block_group_bits(cache
, data
) || cache
->ro
)
266 spin_lock_irq(&free_space_cache
->lock
);
267 state
= find_first_extent_bit_state(free_space_cache
, last
, EXTENT_DIRTY
);
272 spin_unlock_irq(&free_space_cache
->lock
);
276 start
= max(last
, state
->start
);
277 last
= state
->end
+ 1;
278 if (last
- start
< num
) {
280 state
= extent_state_next(state
);
281 } while(state
&& !(state
->state
& EXTENT_DIRTY
));
284 spin_unlock_irq(&free_space_cache
->lock
);
288 if (start
+ num
> cache
->key
.objectid
+ cache
->key
.offset
)
290 if (!block_group_bits(cache
, data
)) {
291 printk("block group bits don't match %Lu %d\n", cache
->flags
, data
);
297 cache
= btrfs_lookup_block_group(root
->fs_info
, search_start
);
299 printk("Unable to find block group for %Lu\n", search_start
);
305 last
= cache
->key
.objectid
+ cache
->key
.offset
;
307 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
308 if (!cache
|| cache
->key
.objectid
>= total_fs_bytes
) {
317 if (cache_miss
&& !cache
->cached
) {
318 cache_block_group(root
, cache
);
320 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
323 cache
= btrfs_find_block_group(root
, cache
, last
, data
, 0);
330 static u64
div_factor(u64 num
, int factor
)
339 static int block_group_state_bits(u64 flags
)
342 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
343 bits
|= BLOCK_GROUP_DATA
;
344 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
345 bits
|= BLOCK_GROUP_METADATA
;
346 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
347 bits
|= BLOCK_GROUP_SYSTEM
;
351 static struct btrfs_block_group_cache
*
352 __btrfs_find_block_group(struct btrfs_root
*root
,
353 struct btrfs_block_group_cache
*hint
,
354 u64 search_start
, int data
, int owner
)
356 struct btrfs_block_group_cache
*cache
;
357 struct extent_io_tree
*block_group_cache
;
358 struct btrfs_block_group_cache
*found_group
= NULL
;
359 struct btrfs_fs_info
*info
= root
->fs_info
;
372 block_group_cache
= &info
->block_group_cache
;
374 if (data
& BTRFS_BLOCK_GROUP_METADATA
)
377 bit
= block_group_state_bits(data
);
380 struct btrfs_block_group_cache
*shint
;
381 shint
= btrfs_lookup_first_block_group(info
, search_start
);
382 if (shint
&& block_group_bits(shint
, data
) && !shint
->ro
) {
383 spin_lock(&shint
->lock
);
384 used
= btrfs_block_group_used(&shint
->item
);
385 if (used
+ shint
->pinned
<
386 div_factor(shint
->key
.offset
, factor
)) {
387 spin_unlock(&shint
->lock
);
390 spin_unlock(&shint
->lock
);
393 if (hint
&& !hint
->ro
&& block_group_bits(hint
, data
)) {
394 spin_lock(&hint
->lock
);
395 used
= btrfs_block_group_used(&hint
->item
);
396 if (used
+ hint
->pinned
<
397 div_factor(hint
->key
.offset
, factor
)) {
398 spin_unlock(&hint
->lock
);
401 spin_unlock(&hint
->lock
);
402 last
= hint
->key
.objectid
+ hint
->key
.offset
;
405 last
= max(hint
->key
.objectid
, search_start
);
411 ret
= find_first_extent_bit(block_group_cache
, last
,
416 ret
= get_state_private(block_group_cache
, start
, &ptr
);
422 cache
= (struct btrfs_block_group_cache
*)(unsigned long)ptr
;
423 spin_lock(&cache
->lock
);
424 last
= cache
->key
.objectid
+ cache
->key
.offset
;
425 used
= btrfs_block_group_used(&cache
->item
);
427 if (!cache
->ro
&& block_group_bits(cache
, data
)) {
428 free_check
= div_factor(cache
->key
.offset
, factor
);
429 if (used
+ cache
->pinned
< free_check
) {
431 spin_unlock(&cache
->lock
);
435 spin_unlock(&cache
->lock
);
443 if (!full_search
&& factor
< 10) {
453 struct btrfs_block_group_cache
*btrfs_find_block_group(struct btrfs_root
*root
,
454 struct btrfs_block_group_cache
455 *hint
, u64 search_start
,
459 struct btrfs_block_group_cache
*ret
;
460 ret
= __btrfs_find_block_group(root
, hint
, search_start
, data
, owner
);
463 static u64
hash_extent_ref(u64 root_objectid
, u64 ref_generation
,
464 u64 owner
, u64 owner_offset
)
466 u32 high_crc
= ~(u32
)0;
467 u32 low_crc
= ~(u32
)0;
469 lenum
= cpu_to_le64(root_objectid
);
470 high_crc
= btrfs_crc32c(high_crc
, &lenum
, sizeof(lenum
));
471 lenum
= cpu_to_le64(ref_generation
);
472 low_crc
= btrfs_crc32c(low_crc
, &lenum
, sizeof(lenum
));
473 if (owner
>= BTRFS_FIRST_FREE_OBJECTID
) {
474 lenum
= cpu_to_le64(owner
);
475 low_crc
= btrfs_crc32c(low_crc
, &lenum
, sizeof(lenum
));
476 lenum
= cpu_to_le64(owner_offset
);
477 low_crc
= btrfs_crc32c(low_crc
, &lenum
, sizeof(lenum
));
479 return ((u64
)high_crc
<< 32) | (u64
)low_crc
;
482 static int match_extent_ref(struct extent_buffer
*leaf
,
483 struct btrfs_extent_ref
*disk_ref
,
484 struct btrfs_extent_ref
*cpu_ref
)
489 if (cpu_ref
->objectid
)
490 len
= sizeof(*cpu_ref
);
492 len
= 2 * sizeof(u64
);
493 ret
= memcmp_extent_buffer(leaf
, cpu_ref
, (unsigned long)disk_ref
,
498 static int noinline
lookup_extent_backref(struct btrfs_trans_handle
*trans
,
499 struct btrfs_root
*root
,
500 struct btrfs_path
*path
, u64 bytenr
,
502 u64 ref_generation
, u64 owner
,
503 u64 owner_offset
, int del
)
506 struct btrfs_key key
;
507 struct btrfs_key found_key
;
508 struct btrfs_extent_ref ref
;
509 struct extent_buffer
*leaf
;
510 struct btrfs_extent_ref
*disk_ref
;
514 btrfs_set_stack_ref_root(&ref
, root_objectid
);
515 btrfs_set_stack_ref_generation(&ref
, ref_generation
);
516 btrfs_set_stack_ref_objectid(&ref
, owner
);
517 btrfs_set_stack_ref_offset(&ref
, owner_offset
);
519 hash
= hash_extent_ref(root_objectid
, ref_generation
, owner
,
522 key
.objectid
= bytenr
;
523 key
.type
= BTRFS_EXTENT_REF_KEY
;
526 ret
= btrfs_search_slot(trans
, root
, &key
, path
,
530 leaf
= path
->nodes
[0];
532 u32 nritems
= btrfs_header_nritems(leaf
);
533 if (path
->slots
[0] >= nritems
) {
534 ret2
= btrfs_next_leaf(root
, path
);
537 leaf
= path
->nodes
[0];
539 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
540 if (found_key
.objectid
!= bytenr
||
541 found_key
.type
!= BTRFS_EXTENT_REF_KEY
)
543 key
.offset
= found_key
.offset
;
545 btrfs_release_path(root
, path
);
549 disk_ref
= btrfs_item_ptr(path
->nodes
[0],
551 struct btrfs_extent_ref
);
552 if (match_extent_ref(path
->nodes
[0], disk_ref
, &ref
)) {
556 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
557 key
.offset
= found_key
.offset
+ 1;
558 btrfs_release_path(root
, path
);
565 * Back reference rules. Back refs have three main goals:
567 * 1) differentiate between all holders of references to an extent so that
568 * when a reference is dropped we can make sure it was a valid reference
569 * before freeing the extent.
571 * 2) Provide enough information to quickly find the holders of an extent
572 * if we notice a given block is corrupted or bad.
574 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
575 * maintenance. This is actually the same as #2, but with a slightly
576 * different use case.
578 * File extents can be referenced by:
580 * - multiple snapshots, subvolumes, or different generations in one subvol
581 * - different files inside a single subvolume (in theory, not implemented yet)
582 * - different offsets inside a file (bookend extents in file.c)
584 * The extent ref structure has fields for:
586 * - Objectid of the subvolume root
587 * - Generation number of the tree holding the reference
588 * - objectid of the file holding the reference
589 * - offset in the file corresponding to the key holding the reference
591 * When a file extent is allocated the fields are filled in:
592 * (root_key.objectid, trans->transid, inode objectid, offset in file)
594 * When a leaf is cow'd new references are added for every file extent found
595 * in the leaf. It looks the same as the create case, but trans->transid
596 * will be different when the block is cow'd.
598 * (root_key.objectid, trans->transid, inode objectid, offset in file)
600 * When a file extent is removed either during snapshot deletion or file
601 * truncation, the corresponding back reference is found
604 * (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
605 * inode objectid, offset in file)
607 * Btree extents can be referenced by:
609 * - Different subvolumes
610 * - Different generations of the same subvolume
612 * Storing sufficient information for a full reverse mapping of a btree
613 * block would require storing the lowest key of the block in the backref,
614 * and it would require updating that lowest key either before write out or
615 * every time it changed. Instead, the objectid of the lowest key is stored
616 * along with the level of the tree block. This provides a hint
617 * about where in the btree the block can be found. Searches through the
618 * btree only need to look for a pointer to that block, so they stop one
619 * level higher than the level recorded in the backref.
621 * Some btrees do not do reference counting on their extents. These
622 * include the extent tree and the tree of tree roots. Backrefs for these
623 * trees always have a generation of zero.
625 * When a tree block is created, back references are inserted:
627 * (root->root_key.objectid, trans->transid or zero, level, lowest_key_objectid)
629 * When a tree block is cow'd in a reference counted root,
630 * new back references are added for all the blocks it points to.
631 * These are of the form (trans->transid will have increased since creation):
633 * (root->root_key.objectid, trans->transid, level, lowest_key_objectid)
635 * Because the lowest_key_objectid and the level are just hints
636 * they are not used when backrefs are deleted. When a backref is deleted:
638 * if backref was for a tree root:
639 * root_objectid = root->root_key.objectid
641 * root_objectid = btrfs_header_owner(parent)
643 * (root_objectid, btrfs_header_generation(parent) or zero, 0, 0)
645 * Back Reference Key hashing:
647 * Back references have four fields, each 64 bits long. Unfortunately,
648 * This is hashed into a single 64 bit number and placed into the key offset.
649 * The key objectid corresponds to the first byte in the extent, and the
650 * key type is set to BTRFS_EXTENT_REF_KEY
652 int btrfs_insert_extent_backref(struct btrfs_trans_handle
*trans
,
653 struct btrfs_root
*root
,
654 struct btrfs_path
*path
, u64 bytenr
,
655 u64 root_objectid
, u64 ref_generation
,
656 u64 owner
, u64 owner_offset
)
659 struct btrfs_key key
;
660 struct btrfs_extent_ref ref
;
661 struct btrfs_extent_ref
*disk_ref
;
664 btrfs_set_stack_ref_root(&ref
, root_objectid
);
665 btrfs_set_stack_ref_generation(&ref
, ref_generation
);
666 btrfs_set_stack_ref_objectid(&ref
, owner
);
667 btrfs_set_stack_ref_offset(&ref
, owner_offset
);
669 hash
= hash_extent_ref(root_objectid
, ref_generation
, owner
,
672 key
.objectid
= bytenr
;
673 key
.type
= BTRFS_EXTENT_REF_KEY
;
675 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, sizeof(ref
));
676 while (ret
== -EEXIST
) {
677 disk_ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
678 struct btrfs_extent_ref
);
679 if (match_extent_ref(path
->nodes
[0], disk_ref
, &ref
))
682 btrfs_release_path(root
, path
);
683 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
688 disk_ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
689 struct btrfs_extent_ref
);
690 write_extent_buffer(path
->nodes
[0], &ref
, (unsigned long)disk_ref
,
692 btrfs_mark_buffer_dirty(path
->nodes
[0]);
694 btrfs_release_path(root
, path
);
698 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
699 struct btrfs_root
*root
,
700 u64 bytenr
, u64 num_bytes
,
701 u64 root_objectid
, u64 ref_generation
,
702 u64 owner
, u64 owner_offset
)
704 struct btrfs_path
*path
;
706 struct btrfs_key key
;
707 struct extent_buffer
*l
;
708 struct btrfs_extent_item
*item
;
711 WARN_ON(num_bytes
< root
->sectorsize
);
712 path
= btrfs_alloc_path();
717 key
.objectid
= bytenr
;
718 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
719 key
.offset
= num_bytes
;
720 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
, path
,
729 item
= btrfs_item_ptr(l
, path
->slots
[0], struct btrfs_extent_item
);
730 refs
= btrfs_extent_refs(l
, item
);
731 btrfs_set_extent_refs(l
, item
, refs
+ 1);
732 btrfs_mark_buffer_dirty(path
->nodes
[0]);
734 btrfs_release_path(root
->fs_info
->extent_root
, path
);
737 ret
= btrfs_insert_extent_backref(trans
, root
->fs_info
->extent_root
,
738 path
, bytenr
, root_objectid
,
739 ref_generation
, owner
, owner_offset
);
741 finish_current_insert(trans
, root
->fs_info
->extent_root
);
742 del_pending_extents(trans
, root
->fs_info
->extent_root
);
744 btrfs_free_path(path
);
748 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
749 struct btrfs_root
*root
,
750 u64 bytenr
, u64 num_bytes
,
751 u64 root_objectid
, u64 ref_generation
,
752 u64 owner
, u64 owner_offset
)
756 mutex_lock(&root
->fs_info
->alloc_mutex
);
757 ret
= __btrfs_inc_extent_ref(trans
, root
, bytenr
, num_bytes
,
758 root_objectid
, ref_generation
,
759 owner
, owner_offset
);
760 mutex_unlock(&root
->fs_info
->alloc_mutex
);
764 int btrfs_extent_post_op(struct btrfs_trans_handle
*trans
,
765 struct btrfs_root
*root
)
767 finish_current_insert(trans
, root
->fs_info
->extent_root
);
768 del_pending_extents(trans
, root
->fs_info
->extent_root
);
772 static int lookup_extent_ref(struct btrfs_trans_handle
*trans
,
773 struct btrfs_root
*root
, u64 bytenr
,
774 u64 num_bytes
, u32
*refs
)
776 struct btrfs_path
*path
;
778 struct btrfs_key key
;
779 struct extent_buffer
*l
;
780 struct btrfs_extent_item
*item
;
782 WARN_ON(num_bytes
< root
->sectorsize
);
783 path
= btrfs_alloc_path();
785 key
.objectid
= bytenr
;
786 key
.offset
= num_bytes
;
787 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
788 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
, path
,
793 btrfs_print_leaf(root
, path
->nodes
[0]);
794 printk("failed to find block number %Lu\n", bytenr
);
798 item
= btrfs_item_ptr(l
, path
->slots
[0], struct btrfs_extent_item
);
799 *refs
= btrfs_extent_refs(l
, item
);
801 btrfs_free_path(path
);
806 static int get_reference_status(struct btrfs_root
*root
, u64 bytenr
,
807 u64 parent_gen
, u64 ref_objectid
,
808 u64
*min_generation
, u32
*ref_count
)
810 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
811 struct btrfs_path
*path
;
812 struct extent_buffer
*leaf
;
813 struct btrfs_extent_ref
*ref_item
;
814 struct btrfs_key key
;
815 struct btrfs_key found_key
;
816 u64 root_objectid
= root
->root_key
.objectid
;
821 key
.objectid
= bytenr
;
823 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
825 path
= btrfs_alloc_path();
826 mutex_lock(&root
->fs_info
->alloc_mutex
);
827 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
832 leaf
= path
->nodes
[0];
833 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
835 if (found_key
.objectid
!= bytenr
||
836 found_key
.type
!= BTRFS_EXTENT_ITEM_KEY
) {
842 *min_generation
= (u64
)-1;
845 leaf
= path
->nodes
[0];
846 nritems
= btrfs_header_nritems(leaf
);
847 if (path
->slots
[0] >= nritems
) {
848 ret
= btrfs_next_leaf(extent_root
, path
);
855 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
856 if (found_key
.objectid
!= bytenr
)
859 if (found_key
.type
!= BTRFS_EXTENT_REF_KEY
) {
864 ref_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
865 struct btrfs_extent_ref
);
866 ref_generation
= btrfs_ref_generation(leaf
, ref_item
);
868 * For (parent_gen > 0 && parent_gen > ref_gen):
870 * we reach here through the oldest root, therefore
871 * all other reference from same snapshot should have
872 * a larger generation.
874 if ((root_objectid
!= btrfs_ref_root(leaf
, ref_item
)) ||
875 (parent_gen
> 0 && parent_gen
> ref_generation
) ||
876 (ref_objectid
>= BTRFS_FIRST_FREE_OBJECTID
&&
877 ref_objectid
!= btrfs_ref_objectid(leaf
, ref_item
))) {
884 if (*min_generation
> ref_generation
)
885 *min_generation
= ref_generation
;
891 mutex_unlock(&root
->fs_info
->alloc_mutex
);
892 btrfs_free_path(path
);
896 int btrfs_cross_ref_exists(struct btrfs_trans_handle
*trans
,
897 struct btrfs_root
*root
,
898 struct btrfs_key
*key
, u64 bytenr
)
900 struct btrfs_root
*old_root
;
901 struct btrfs_path
*path
= NULL
;
902 struct extent_buffer
*eb
;
903 struct btrfs_file_extent_item
*item
;
911 BUG_ON(trans
== NULL
);
912 BUG_ON(key
->type
!= BTRFS_EXTENT_DATA_KEY
);
913 ret
= get_reference_status(root
, bytenr
, 0, key
->objectid
,
914 &min_generation
, &ref_count
);
921 old_root
= root
->dirty_root
->root
;
922 ref_generation
= old_root
->root_key
.offset
;
924 /* all references are created in running transaction */
925 if (min_generation
> ref_generation
) {
930 path
= btrfs_alloc_path();
936 path
->skip_locking
= 1;
937 /* if no item found, the extent is referenced by other snapshot */
938 ret
= btrfs_search_slot(NULL
, old_root
, key
, path
, 0, 0);
943 item
= btrfs_item_ptr(eb
, path
->slots
[0],
944 struct btrfs_file_extent_item
);
945 if (btrfs_file_extent_type(eb
, item
) != BTRFS_FILE_EXTENT_REG
||
946 btrfs_file_extent_disk_bytenr(eb
, item
) != bytenr
) {
951 for (level
= BTRFS_MAX_LEVEL
- 1; level
>= -1; level
--) {
953 eb
= path
->nodes
[level
];
956 extent_start
= eb
->start
;
958 extent_start
= bytenr
;
960 ret
= get_reference_status(root
, extent_start
, ref_generation
,
961 0, &min_generation
, &ref_count
);
965 if (ref_count
!= 1) {
970 ref_generation
= btrfs_header_generation(eb
);
975 btrfs_free_path(path
);
979 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
980 struct extent_buffer
*buf
, int cache_ref
)
984 struct btrfs_key key
;
985 struct btrfs_file_extent_item
*fi
;
990 int nr_file_extents
= 0;
995 level
= btrfs_header_level(buf
);
996 nritems
= btrfs_header_nritems(buf
);
997 for (i
= 0; i
< nritems
; i
++) {
1001 btrfs_item_key_to_cpu(buf
, &key
, i
);
1002 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
1004 fi
= btrfs_item_ptr(buf
, i
,
1005 struct btrfs_file_extent_item
);
1006 if (btrfs_file_extent_type(buf
, fi
) ==
1007 BTRFS_FILE_EXTENT_INLINE
)
1009 disk_bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
1010 if (disk_bytenr
== 0)
1013 if (buf
!= root
->commit_root
)
1016 mutex_lock(&root
->fs_info
->alloc_mutex
);
1017 ret
= __btrfs_inc_extent_ref(trans
, root
, disk_bytenr
,
1018 btrfs_file_extent_disk_num_bytes(buf
, fi
),
1019 root
->root_key
.objectid
, trans
->transid
,
1020 key
.objectid
, key
.offset
);
1021 mutex_unlock(&root
->fs_info
->alloc_mutex
);
1028 bytenr
= btrfs_node_blockptr(buf
, i
);
1029 btrfs_node_key_to_cpu(buf
, &key
, i
);
1031 mutex_lock(&root
->fs_info
->alloc_mutex
);
1032 ret
= __btrfs_inc_extent_ref(trans
, root
, bytenr
,
1033 btrfs_level_size(root
, level
- 1),
1034 root
->root_key
.objectid
,
1036 level
- 1, key
.objectid
);
1037 mutex_unlock(&root
->fs_info
->alloc_mutex
);
1045 /* cache orignal leaf block's references */
1046 if (level
== 0 && cache_ref
&& buf
!= root
->commit_root
) {
1047 struct btrfs_leaf_ref
*ref
;
1048 struct btrfs_extent_info
*info
;
1050 ref
= btrfs_alloc_leaf_ref(root
, nr_file_extents
);
1056 ref
->root_gen
= root
->root_key
.offset
;
1057 ref
->bytenr
= buf
->start
;
1058 ref
->owner
= btrfs_header_owner(buf
);
1059 ref
->generation
= btrfs_header_generation(buf
);
1060 ref
->nritems
= nr_file_extents
;
1061 info
= ref
->extents
;
1063 for (i
= 0; nr_file_extents
> 0 && i
< nritems
; i
++) {
1065 btrfs_item_key_to_cpu(buf
, &key
, i
);
1066 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
1068 fi
= btrfs_item_ptr(buf
, i
,
1069 struct btrfs_file_extent_item
);
1070 if (btrfs_file_extent_type(buf
, fi
) ==
1071 BTRFS_FILE_EXTENT_INLINE
)
1073 disk_bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
1074 if (disk_bytenr
== 0)
1077 info
->bytenr
= disk_bytenr
;
1079 btrfs_file_extent_disk_num_bytes(buf
, fi
);
1080 info
->objectid
= key
.objectid
;
1081 info
->offset
= key
.offset
;
1085 BUG_ON(!root
->ref_tree
);
1086 ret
= btrfs_add_leaf_ref(root
, ref
);
1088 btrfs_free_leaf_ref(root
, ref
);
1095 for (i
=0; i
< faili
; i
++) {
1098 btrfs_item_key_to_cpu(buf
, &key
, i
);
1099 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
1101 fi
= btrfs_item_ptr(buf
, i
,
1102 struct btrfs_file_extent_item
);
1103 if (btrfs_file_extent_type(buf
, fi
) ==
1104 BTRFS_FILE_EXTENT_INLINE
)
1106 disk_bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
1107 if (disk_bytenr
== 0)
1109 err
= btrfs_free_extent(trans
, root
, disk_bytenr
,
1110 btrfs_file_extent_disk_num_bytes(buf
,
1114 bytenr
= btrfs_node_blockptr(buf
, i
);
1115 err
= btrfs_free_extent(trans
, root
, bytenr
,
1116 btrfs_level_size(root
, level
- 1), 0);
1124 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
1125 struct btrfs_root
*root
,
1126 struct btrfs_path
*path
,
1127 struct btrfs_block_group_cache
*cache
)
1131 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
1133 struct extent_buffer
*leaf
;
1135 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
1140 leaf
= path
->nodes
[0];
1141 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
1142 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
1143 btrfs_mark_buffer_dirty(leaf
);
1144 btrfs_release_path(extent_root
, path
);
1146 finish_current_insert(trans
, extent_root
);
1147 pending_ret
= del_pending_extents(trans
, extent_root
);
1156 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
1157 struct btrfs_root
*root
)
1159 struct extent_io_tree
*block_group_cache
;
1160 struct btrfs_block_group_cache
*cache
;
1164 struct btrfs_path
*path
;
1170 block_group_cache
= &root
->fs_info
->block_group_cache
;
1171 path
= btrfs_alloc_path();
1175 mutex_lock(&root
->fs_info
->alloc_mutex
);
1177 ret
= find_first_extent_bit(block_group_cache
, last
,
1178 &start
, &end
, BLOCK_GROUP_DIRTY
);
1183 ret
= get_state_private(block_group_cache
, start
, &ptr
);
1186 cache
= (struct btrfs_block_group_cache
*)(unsigned long)ptr
;
1187 err
= write_one_cache_group(trans
, root
,
1190 * if we fail to write the cache group, we want
1191 * to keep it marked dirty in hopes that a later
1198 clear_extent_bits(block_group_cache
, start
, end
,
1199 BLOCK_GROUP_DIRTY
, GFP_NOFS
);
1201 btrfs_free_path(path
);
1202 mutex_unlock(&root
->fs_info
->alloc_mutex
);
1206 static struct btrfs_space_info
*__find_space_info(struct btrfs_fs_info
*info
,
1209 struct list_head
*head
= &info
->space_info
;
1210 struct list_head
*cur
;
1211 struct btrfs_space_info
*found
;
1212 list_for_each(cur
, head
) {
1213 found
= list_entry(cur
, struct btrfs_space_info
, list
);
1214 if (found
->flags
== flags
)
1221 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
1222 u64 total_bytes
, u64 bytes_used
,
1223 struct btrfs_space_info
**space_info
)
1225 struct btrfs_space_info
*found
;
1227 found
= __find_space_info(info
, flags
);
1229 found
->total_bytes
+= total_bytes
;
1230 found
->bytes_used
+= bytes_used
;
1232 *space_info
= found
;
1235 found
= kmalloc(sizeof(*found
), GFP_NOFS
);
1239 list_add(&found
->list
, &info
->space_info
);
1240 found
->flags
= flags
;
1241 found
->total_bytes
= total_bytes
;
1242 found
->bytes_used
= bytes_used
;
1243 found
->bytes_pinned
= 0;
1245 found
->force_alloc
= 0;
1246 *space_info
= found
;
1250 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
1252 u64 extra_flags
= flags
& (BTRFS_BLOCK_GROUP_RAID0
|
1253 BTRFS_BLOCK_GROUP_RAID1
|
1254 BTRFS_BLOCK_GROUP_RAID10
|
1255 BTRFS_BLOCK_GROUP_DUP
);
1257 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
1258 fs_info
->avail_data_alloc_bits
|= extra_flags
;
1259 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
1260 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
1261 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
1262 fs_info
->avail_system_alloc_bits
|= extra_flags
;
1266 static u64
reduce_alloc_profile(struct btrfs_root
*root
, u64 flags
)
1268 u64 num_devices
= root
->fs_info
->fs_devices
->num_devices
;
1270 if (num_devices
== 1)
1271 flags
&= ~(BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID0
);
1272 if (num_devices
< 4)
1273 flags
&= ~BTRFS_BLOCK_GROUP_RAID10
;
1275 if ((flags
& BTRFS_BLOCK_GROUP_DUP
) &&
1276 (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
1277 BTRFS_BLOCK_GROUP_RAID10
))) {
1278 flags
&= ~BTRFS_BLOCK_GROUP_DUP
;
1281 if ((flags
& BTRFS_BLOCK_GROUP_RAID1
) &&
1282 (flags
& BTRFS_BLOCK_GROUP_RAID10
)) {
1283 flags
&= ~BTRFS_BLOCK_GROUP_RAID1
;
1286 if ((flags
& BTRFS_BLOCK_GROUP_RAID0
) &&
1287 ((flags
& BTRFS_BLOCK_GROUP_RAID1
) |
1288 (flags
& BTRFS_BLOCK_GROUP_RAID10
) |
1289 (flags
& BTRFS_BLOCK_GROUP_DUP
)))
1290 flags
&= ~BTRFS_BLOCK_GROUP_RAID0
;
1294 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
1295 struct btrfs_root
*extent_root
, u64 alloc_bytes
,
1296 u64 flags
, int force
)
1298 struct btrfs_space_info
*space_info
;
1304 flags
= reduce_alloc_profile(extent_root
, flags
);
1306 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
1308 ret
= update_space_info(extent_root
->fs_info
, flags
,
1312 BUG_ON(!space_info
);
1314 if (space_info
->force_alloc
) {
1316 space_info
->force_alloc
= 0;
1318 if (space_info
->full
)
1321 thresh
= div_factor(space_info
->total_bytes
, 6);
1323 (space_info
->bytes_used
+ space_info
->bytes_pinned
+ alloc_bytes
) <
1327 mutex_lock(&extent_root
->fs_info
->chunk_mutex
);
1328 ret
= btrfs_alloc_chunk(trans
, extent_root
, &start
, &num_bytes
, flags
);
1329 if (ret
== -ENOSPC
) {
1330 printk("space info full %Lu\n", flags
);
1331 space_info
->full
= 1;
1336 ret
= btrfs_make_block_group(trans
, extent_root
, 0, flags
,
1337 BTRFS_FIRST_CHUNK_TREE_OBJECTID
, start
, num_bytes
);
1340 mutex_unlock(&extent_root
->fs_info
->chunk_mutex
);
1345 static int update_block_group(struct btrfs_trans_handle
*trans
,
1346 struct btrfs_root
*root
,
1347 u64 bytenr
, u64 num_bytes
, int alloc
,
1350 struct btrfs_block_group_cache
*cache
;
1351 struct btrfs_fs_info
*info
= root
->fs_info
;
1352 u64 total
= num_bytes
;
1358 WARN_ON(!mutex_is_locked(&root
->fs_info
->alloc_mutex
));
1360 cache
= btrfs_lookup_block_group(info
, bytenr
);
1364 byte_in_group
= bytenr
- cache
->key
.objectid
;
1365 WARN_ON(byte_in_group
> cache
->key
.offset
);
1366 start
= cache
->key
.objectid
;
1367 end
= start
+ cache
->key
.offset
- 1;
1368 set_extent_bits(&info
->block_group_cache
, start
, end
,
1369 BLOCK_GROUP_DIRTY
, GFP_NOFS
);
1371 spin_lock(&cache
->lock
);
1372 old_val
= btrfs_block_group_used(&cache
->item
);
1373 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
1375 old_val
+= num_bytes
;
1376 cache
->space_info
->bytes_used
+= num_bytes
;
1377 btrfs_set_block_group_used(&cache
->item
, old_val
);
1378 spin_unlock(&cache
->lock
);
1380 old_val
-= num_bytes
;
1381 cache
->space_info
->bytes_used
-= num_bytes
;
1382 btrfs_set_block_group_used(&cache
->item
, old_val
);
1383 spin_unlock(&cache
->lock
);
1385 set_extent_dirty(&info
->free_space_cache
,
1386 bytenr
, bytenr
+ num_bytes
- 1,
1391 bytenr
+= num_bytes
;
1396 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
1401 ret
= find_first_extent_bit(&root
->fs_info
->block_group_cache
,
1402 search_start
, &start
, &end
,
1403 BLOCK_GROUP_DATA
| BLOCK_GROUP_METADATA
|
1404 BLOCK_GROUP_SYSTEM
);
1411 static int update_pinned_extents(struct btrfs_root
*root
,
1412 u64 bytenr
, u64 num
, int pin
)
1415 struct btrfs_block_group_cache
*cache
;
1416 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1418 WARN_ON(!mutex_is_locked(&root
->fs_info
->alloc_mutex
));
1420 set_extent_dirty(&fs_info
->pinned_extents
,
1421 bytenr
, bytenr
+ num
- 1, GFP_NOFS
);
1423 clear_extent_dirty(&fs_info
->pinned_extents
,
1424 bytenr
, bytenr
+ num
- 1, GFP_NOFS
);
1427 cache
= btrfs_lookup_block_group(fs_info
, bytenr
);
1429 u64 first
= first_logical_byte(root
, bytenr
);
1430 WARN_ON(first
< bytenr
);
1431 len
= min(first
- bytenr
, num
);
1433 len
= min(num
, cache
->key
.offset
-
1434 (bytenr
- cache
->key
.objectid
));
1438 spin_lock(&cache
->lock
);
1439 cache
->pinned
+= len
;
1440 cache
->space_info
->bytes_pinned
+= len
;
1441 spin_unlock(&cache
->lock
);
1443 fs_info
->total_pinned
+= len
;
1446 spin_lock(&cache
->lock
);
1447 cache
->pinned
-= len
;
1448 cache
->space_info
->bytes_pinned
-= len
;
1449 spin_unlock(&cache
->lock
);
1451 fs_info
->total_pinned
-= len
;
1459 int btrfs_copy_pinned(struct btrfs_root
*root
, struct extent_io_tree
*copy
)
1464 struct extent_io_tree
*pinned_extents
= &root
->fs_info
->pinned_extents
;
1468 ret
= find_first_extent_bit(pinned_extents
, last
,
1469 &start
, &end
, EXTENT_DIRTY
);
1472 set_extent_dirty(copy
, start
, end
, GFP_NOFS
);
1478 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
1479 struct btrfs_root
*root
,
1480 struct extent_io_tree
*unpin
)
1485 struct extent_io_tree
*free_space_cache
;
1486 free_space_cache
= &root
->fs_info
->free_space_cache
;
1488 mutex_lock(&root
->fs_info
->alloc_mutex
);
1490 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
1494 update_pinned_extents(root
, start
, end
+ 1 - start
, 0);
1495 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
1496 set_extent_dirty(free_space_cache
, start
, end
, GFP_NOFS
);
1497 if (need_resched()) {
1498 mutex_unlock(&root
->fs_info
->alloc_mutex
);
1500 mutex_lock(&root
->fs_info
->alloc_mutex
);
1503 mutex_unlock(&root
->fs_info
->alloc_mutex
);
1507 static int finish_current_insert(struct btrfs_trans_handle
*trans
,
1508 struct btrfs_root
*extent_root
)
1512 struct btrfs_fs_info
*info
= extent_root
->fs_info
;
1513 struct extent_buffer
*eb
;
1514 struct btrfs_path
*path
;
1515 struct btrfs_key ins
;
1516 struct btrfs_disk_key first
;
1517 struct btrfs_extent_item extent_item
;
1522 WARN_ON(!mutex_is_locked(&extent_root
->fs_info
->alloc_mutex
));
1523 btrfs_set_stack_extent_refs(&extent_item
, 1);
1524 btrfs_set_key_type(&ins
, BTRFS_EXTENT_ITEM_KEY
);
1525 path
= btrfs_alloc_path();
1528 ret
= find_first_extent_bit(&info
->extent_ins
, 0, &start
,
1529 &end
, EXTENT_LOCKED
);
1533 ins
.objectid
= start
;
1534 ins
.offset
= end
+ 1 - start
;
1535 err
= btrfs_insert_item(trans
, extent_root
, &ins
,
1536 &extent_item
, sizeof(extent_item
));
1537 clear_extent_bits(&info
->extent_ins
, start
, end
, EXTENT_LOCKED
,
1540 eb
= btrfs_find_tree_block(extent_root
, ins
.objectid
,
1543 if (!btrfs_buffer_uptodate(eb
, trans
->transid
)) {
1544 mutex_unlock(&extent_root
->fs_info
->alloc_mutex
);
1545 btrfs_read_buffer(eb
, trans
->transid
);
1546 mutex_lock(&extent_root
->fs_info
->alloc_mutex
);
1549 btrfs_tree_lock(eb
);
1550 level
= btrfs_header_level(eb
);
1552 btrfs_item_key(eb
, &first
, 0);
1554 btrfs_node_key(eb
, &first
, 0);
1556 btrfs_tree_unlock(eb
);
1557 free_extent_buffer(eb
);
1559 * the first key is just a hint, so the race we've created
1560 * against reading it is fine
1562 err
= btrfs_insert_extent_backref(trans
, extent_root
, path
,
1563 start
, extent_root
->root_key
.objectid
,
1565 btrfs_disk_key_objectid(&first
));
1567 if (need_resched()) {
1568 mutex_unlock(&extent_root
->fs_info
->alloc_mutex
);
1570 mutex_lock(&extent_root
->fs_info
->alloc_mutex
);
1573 btrfs_free_path(path
);
1577 static int pin_down_bytes(struct btrfs_root
*root
, u64 bytenr
, u32 num_bytes
,
1582 WARN_ON(!mutex_is_locked(&root
->fs_info
->alloc_mutex
));
1584 struct extent_buffer
*buf
;
1585 buf
= btrfs_find_tree_block(root
, bytenr
, num_bytes
);
1587 if (btrfs_buffer_uptodate(buf
, 0) &&
1588 btrfs_try_tree_lock(buf
)) {
1590 root
->fs_info
->running_transaction
->transid
;
1591 u64 header_transid
=
1592 btrfs_header_generation(buf
);
1593 if (header_transid
== transid
&&
1594 !btrfs_header_flag(buf
,
1595 BTRFS_HEADER_FLAG_WRITTEN
)) {
1596 clean_tree_block(NULL
, root
, buf
);
1597 btrfs_tree_unlock(buf
);
1598 free_extent_buffer(buf
);
1601 btrfs_tree_unlock(buf
);
1603 free_extent_buffer(buf
);
1605 update_pinned_extents(root
, bytenr
, num_bytes
, 1);
1607 set_extent_bits(&root
->fs_info
->pending_del
,
1608 bytenr
, bytenr
+ num_bytes
- 1,
1609 EXTENT_LOCKED
, GFP_NOFS
);
1616 * remove an extent from the root, returns 0 on success
1618 static int __free_extent(struct btrfs_trans_handle
*trans
, struct btrfs_root
1619 *root
, u64 bytenr
, u64 num_bytes
,
1620 u64 root_objectid
, u64 ref_generation
,
1621 u64 owner_objectid
, u64 owner_offset
, int pin
,
1624 struct btrfs_path
*path
;
1625 struct btrfs_key key
;
1626 struct btrfs_fs_info
*info
= root
->fs_info
;
1627 struct btrfs_root
*extent_root
= info
->extent_root
;
1628 struct extent_buffer
*leaf
;
1630 int extent_slot
= 0;
1631 int found_extent
= 0;
1633 struct btrfs_extent_item
*ei
;
1636 WARN_ON(!mutex_is_locked(&root
->fs_info
->alloc_mutex
));
1637 key
.objectid
= bytenr
;
1638 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
1639 key
.offset
= num_bytes
;
1640 path
= btrfs_alloc_path();
1645 ret
= lookup_extent_backref(trans
, extent_root
, path
,
1646 bytenr
, root_objectid
,
1648 owner_objectid
, owner_offset
, 1);
1650 struct btrfs_key found_key
;
1651 extent_slot
= path
->slots
[0];
1652 while(extent_slot
> 0) {
1654 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1656 if (found_key
.objectid
!= bytenr
)
1658 if (found_key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
1659 found_key
.offset
== num_bytes
) {
1663 if (path
->slots
[0] - extent_slot
> 5)
1667 ret
= btrfs_del_item(trans
, extent_root
, path
);
1669 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
1671 printk("Unable to find ref byte nr %Lu root %Lu "
1672 " gen %Lu owner %Lu offset %Lu\n", bytenr
,
1673 root_objectid
, ref_generation
, owner_objectid
,
1676 if (!found_extent
) {
1677 btrfs_release_path(extent_root
, path
);
1678 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
, -1, 1);
1682 extent_slot
= path
->slots
[0];
1685 leaf
= path
->nodes
[0];
1686 ei
= btrfs_item_ptr(leaf
, extent_slot
,
1687 struct btrfs_extent_item
);
1688 refs
= btrfs_extent_refs(leaf
, ei
);
1691 btrfs_set_extent_refs(leaf
, ei
, refs
);
1693 btrfs_mark_buffer_dirty(leaf
);
1695 if (refs
== 0 && found_extent
&& path
->slots
[0] == extent_slot
+ 1) {
1696 /* if the back ref and the extent are next to each other
1697 * they get deleted below in one shot
1699 path
->slots
[0] = extent_slot
;
1701 } else if (found_extent
) {
1702 /* otherwise delete the extent back ref */
1703 ret
= btrfs_del_item(trans
, extent_root
, path
);
1705 /* if refs are 0, we need to setup the path for deletion */
1707 btrfs_release_path(extent_root
, path
);
1708 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
1721 ret
= pin_down_bytes(root
, bytenr
, num_bytes
, 0);
1727 /* block accounting for super block */
1728 spin_lock_irq(&info
->delalloc_lock
);
1729 super_used
= btrfs_super_bytes_used(&info
->super_copy
);
1730 btrfs_set_super_bytes_used(&info
->super_copy
,
1731 super_used
- num_bytes
);
1732 spin_unlock_irq(&info
->delalloc_lock
);
1734 /* block accounting for root item */
1735 root_used
= btrfs_root_used(&root
->root_item
);
1736 btrfs_set_root_used(&root
->root_item
,
1737 root_used
- num_bytes
);
1738 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
1743 ret
= update_block_group(trans
, root
, bytenr
, num_bytes
, 0,
1747 btrfs_free_path(path
);
1748 finish_current_insert(trans
, extent_root
);
1753 * find all the blocks marked as pending in the radix tree and remove
1754 * them from the extent map
1756 static int del_pending_extents(struct btrfs_trans_handle
*trans
, struct
1757 btrfs_root
*extent_root
)
1763 struct extent_io_tree
*pending_del
;
1764 struct extent_io_tree
*pinned_extents
;
1766 WARN_ON(!mutex_is_locked(&extent_root
->fs_info
->alloc_mutex
));
1767 pending_del
= &extent_root
->fs_info
->pending_del
;
1768 pinned_extents
= &extent_root
->fs_info
->pinned_extents
;
1771 ret
= find_first_extent_bit(pending_del
, 0, &start
, &end
,
1775 clear_extent_bits(pending_del
, start
, end
, EXTENT_LOCKED
,
1777 if (!test_range_bit(&extent_root
->fs_info
->extent_ins
,
1778 start
, end
, EXTENT_LOCKED
, 0)) {
1779 update_pinned_extents(extent_root
, start
,
1780 end
+ 1 - start
, 1);
1781 ret
= __free_extent(trans
, extent_root
,
1782 start
, end
+ 1 - start
,
1783 extent_root
->root_key
.objectid
,
1786 clear_extent_bits(&extent_root
->fs_info
->extent_ins
,
1787 start
, end
, EXTENT_LOCKED
, GFP_NOFS
);
1792 if (need_resched()) {
1793 mutex_unlock(&extent_root
->fs_info
->alloc_mutex
);
1795 mutex_lock(&extent_root
->fs_info
->alloc_mutex
);
1802 * remove an extent from the root, returns 0 on success
1804 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
1805 struct btrfs_root
*root
, u64 bytenr
,
1806 u64 num_bytes
, u64 root_objectid
,
1807 u64 ref_generation
, u64 owner_objectid
,
1808 u64 owner_offset
, int pin
)
1810 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
1814 WARN_ON(num_bytes
< root
->sectorsize
);
1815 if (!root
->ref_cows
)
1818 if (root
== extent_root
) {
1819 pin_down_bytes(root
, bytenr
, num_bytes
, 1);
1822 ret
= __free_extent(trans
, root
, bytenr
, num_bytes
, root_objectid
,
1823 ref_generation
, owner_objectid
, owner_offset
,
1826 finish_current_insert(trans
, root
->fs_info
->extent_root
);
1827 pending_ret
= del_pending_extents(trans
, root
->fs_info
->extent_root
);
1828 return ret
? ret
: pending_ret
;
1831 int btrfs_free_extent(struct btrfs_trans_handle
*trans
,
1832 struct btrfs_root
*root
, u64 bytenr
,
1833 u64 num_bytes
, u64 root_objectid
,
1834 u64 ref_generation
, u64 owner_objectid
,
1835 u64 owner_offset
, int pin
)
1839 maybe_lock_mutex(root
);
1840 ret
= __btrfs_free_extent(trans
, root
, bytenr
, num_bytes
,
1841 root_objectid
, ref_generation
,
1842 owner_objectid
, owner_offset
, pin
);
1843 maybe_unlock_mutex(root
);
1847 static u64
stripe_align(struct btrfs_root
*root
, u64 val
)
1849 u64 mask
= ((u64
)root
->stripesize
- 1);
1850 u64 ret
= (val
+ mask
) & ~mask
;
1855 * walks the btree of allocated extents and find a hole of a given size.
1856 * The key ins is changed to record the hole:
1857 * ins->objectid == block start
1858 * ins->flags = BTRFS_EXTENT_ITEM_KEY
1859 * ins->offset == number of blocks
1860 * Any available blocks before search_start are skipped.
1862 static int noinline
find_free_extent(struct btrfs_trans_handle
*trans
,
1863 struct btrfs_root
*orig_root
,
1864 u64 num_bytes
, u64 empty_size
,
1865 u64 search_start
, u64 search_end
,
1866 u64 hint_byte
, struct btrfs_key
*ins
,
1867 u64 exclude_start
, u64 exclude_nr
,
1871 u64 orig_search_start
;
1872 struct btrfs_root
* root
= orig_root
->fs_info
->extent_root
;
1873 struct btrfs_fs_info
*info
= root
->fs_info
;
1874 u64 total_needed
= num_bytes
;
1875 u64
*last_ptr
= NULL
;
1876 struct btrfs_block_group_cache
*block_group
;
1879 int chunk_alloc_done
= 0;
1880 int empty_cluster
= 2 * 1024 * 1024;
1881 int allowed_chunk_alloc
= 0;
1883 WARN_ON(num_bytes
< root
->sectorsize
);
1884 btrfs_set_key_type(ins
, BTRFS_EXTENT_ITEM_KEY
);
1886 if (orig_root
->ref_cows
|| empty_size
)
1887 allowed_chunk_alloc
= 1;
1889 if (data
& BTRFS_BLOCK_GROUP_METADATA
) {
1890 last_ptr
= &root
->fs_info
->last_alloc
;
1891 empty_cluster
= 256 * 1024;
1894 if ((data
& BTRFS_BLOCK_GROUP_DATA
) && btrfs_test_opt(root
, SSD
)) {
1895 last_ptr
= &root
->fs_info
->last_data_alloc
;
1900 hint_byte
= *last_ptr
;
1902 empty_size
+= empty_cluster
;
1906 search_start
= max(search_start
, first_logical_byte(root
, 0));
1907 orig_search_start
= search_start
;
1909 if (search_end
== (u64
)-1)
1910 search_end
= btrfs_super_total_bytes(&info
->super_copy
);
1913 block_group
= btrfs_lookup_first_block_group(info
, hint_byte
);
1915 hint_byte
= search_start
;
1916 block_group
= btrfs_find_block_group(root
, block_group
,
1917 hint_byte
, data
, 1);
1918 if (last_ptr
&& *last_ptr
== 0 && block_group
)
1919 hint_byte
= block_group
->key
.objectid
;
1921 block_group
= btrfs_find_block_group(root
,
1923 search_start
, data
, 1);
1925 search_start
= max(search_start
, hint_byte
);
1927 total_needed
+= empty_size
;
1931 block_group
= btrfs_lookup_first_block_group(info
,
1934 block_group
= btrfs_lookup_first_block_group(info
,
1937 if (full_scan
&& !chunk_alloc_done
) {
1938 if (allowed_chunk_alloc
) {
1939 do_chunk_alloc(trans
, root
,
1940 num_bytes
+ 2 * 1024 * 1024, data
, 1);
1941 allowed_chunk_alloc
= 0;
1942 } else if (block_group
&& block_group_bits(block_group
, data
)) {
1943 block_group
->space_info
->force_alloc
= 1;
1945 chunk_alloc_done
= 1;
1947 ret
= find_search_start(root
, &block_group
, &search_start
,
1948 total_needed
, data
);
1949 if (ret
== -ENOSPC
&& last_ptr
&& *last_ptr
) {
1951 block_group
= btrfs_lookup_first_block_group(info
,
1953 search_start
= orig_search_start
;
1954 ret
= find_search_start(root
, &block_group
, &search_start
,
1955 total_needed
, data
);
1962 if (last_ptr
&& *last_ptr
&& search_start
!= *last_ptr
) {
1965 empty_size
+= empty_cluster
;
1966 total_needed
+= empty_size
;
1968 block_group
= btrfs_lookup_first_block_group(info
,
1970 search_start
= orig_search_start
;
1971 ret
= find_search_start(root
, &block_group
,
1972 &search_start
, total_needed
, data
);
1979 search_start
= stripe_align(root
, search_start
);
1980 ins
->objectid
= search_start
;
1981 ins
->offset
= num_bytes
;
1983 if (ins
->objectid
+ num_bytes
>= search_end
)
1986 if (ins
->objectid
+ num_bytes
>
1987 block_group
->key
.objectid
+ block_group
->key
.offset
) {
1988 search_start
= block_group
->key
.objectid
+
1989 block_group
->key
.offset
;
1993 if (test_range_bit(&info
->extent_ins
, ins
->objectid
,
1994 ins
->objectid
+ num_bytes
-1, EXTENT_LOCKED
, 0)) {
1995 search_start
= ins
->objectid
+ num_bytes
;
1999 if (test_range_bit(&info
->pinned_extents
, ins
->objectid
,
2000 ins
->objectid
+ num_bytes
-1, EXTENT_DIRTY
, 0)) {
2001 search_start
= ins
->objectid
+ num_bytes
;
2005 if (exclude_nr
> 0 && (ins
->objectid
+ num_bytes
> exclude_start
&&
2006 ins
->objectid
< exclude_start
+ exclude_nr
)) {
2007 search_start
= exclude_start
+ exclude_nr
;
2011 if (!(data
& BTRFS_BLOCK_GROUP_DATA
)) {
2012 block_group
= btrfs_lookup_block_group(info
, ins
->objectid
);
2014 trans
->block_group
= block_group
;
2016 ins
->offset
= num_bytes
;
2018 *last_ptr
= ins
->objectid
+ ins
->offset
;
2020 btrfs_super_total_bytes(&root
->fs_info
->super_copy
)) {
2027 if (search_start
+ num_bytes
>= search_end
) {
2029 search_start
= orig_search_start
;
2036 total_needed
-= empty_size
;
2041 block_group
= btrfs_lookup_first_block_group(info
, search_start
);
2043 block_group
= btrfs_find_block_group(root
, block_group
,
2044 search_start
, data
, 0);
2051 static int __btrfs_reserve_extent(struct btrfs_trans_handle
*trans
,
2052 struct btrfs_root
*root
,
2053 u64 num_bytes
, u64 min_alloc_size
,
2054 u64 empty_size
, u64 hint_byte
,
2055 u64 search_end
, struct btrfs_key
*ins
,
2059 u64 search_start
= 0;
2061 struct btrfs_fs_info
*info
= root
->fs_info
;
2064 alloc_profile
= info
->avail_data_alloc_bits
&
2065 info
->data_alloc_profile
;
2066 data
= BTRFS_BLOCK_GROUP_DATA
| alloc_profile
;
2067 } else if (root
== root
->fs_info
->chunk_root
) {
2068 alloc_profile
= info
->avail_system_alloc_bits
&
2069 info
->system_alloc_profile
;
2070 data
= BTRFS_BLOCK_GROUP_SYSTEM
| alloc_profile
;
2072 alloc_profile
= info
->avail_metadata_alloc_bits
&
2073 info
->metadata_alloc_profile
;
2074 data
= BTRFS_BLOCK_GROUP_METADATA
| alloc_profile
;
2077 data
= reduce_alloc_profile(root
, data
);
2079 * the only place that sets empty_size is btrfs_realloc_node, which
2080 * is not called recursively on allocations
2082 if (empty_size
|| root
->ref_cows
) {
2083 if (!(data
& BTRFS_BLOCK_GROUP_METADATA
)) {
2084 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
2086 BTRFS_BLOCK_GROUP_METADATA
|
2087 (info
->metadata_alloc_profile
&
2088 info
->avail_metadata_alloc_bits
), 0);
2091 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
2092 num_bytes
+ 2 * 1024 * 1024, data
, 0);
2096 WARN_ON(num_bytes
< root
->sectorsize
);
2097 ret
= find_free_extent(trans
, root
, num_bytes
, empty_size
,
2098 search_start
, search_end
, hint_byte
, ins
,
2099 trans
->alloc_exclude_start
,
2100 trans
->alloc_exclude_nr
, data
);
2102 if (ret
== -ENOSPC
&& num_bytes
> min_alloc_size
) {
2103 num_bytes
= num_bytes
>> 1;
2104 num_bytes
= max(num_bytes
, min_alloc_size
);
2105 do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
2106 num_bytes
, data
, 1);
2110 printk("allocation failed flags %Lu\n", data
);
2113 clear_extent_dirty(&root
->fs_info
->free_space_cache
,
2114 ins
->objectid
, ins
->objectid
+ ins
->offset
- 1,
2119 int btrfs_free_reserved_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
2121 maybe_lock_mutex(root
);
2122 set_extent_dirty(&root
->fs_info
->free_space_cache
,
2123 start
, start
+ len
- 1, GFP_NOFS
);
2124 maybe_unlock_mutex(root
);
2128 int btrfs_reserve_extent(struct btrfs_trans_handle
*trans
,
2129 struct btrfs_root
*root
,
2130 u64 num_bytes
, u64 min_alloc_size
,
2131 u64 empty_size
, u64 hint_byte
,
2132 u64 search_end
, struct btrfs_key
*ins
,
2136 maybe_lock_mutex(root
);
2137 ret
= __btrfs_reserve_extent(trans
, root
, num_bytes
, min_alloc_size
,
2138 empty_size
, hint_byte
, search_end
, ins
,
2140 maybe_unlock_mutex(root
);
2144 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle
*trans
,
2145 struct btrfs_root
*root
,
2146 u64 root_objectid
, u64 ref_generation
,
2147 u64 owner
, u64 owner_offset
,
2148 struct btrfs_key
*ins
)
2154 u64 num_bytes
= ins
->offset
;
2156 struct btrfs_fs_info
*info
= root
->fs_info
;
2157 struct btrfs_root
*extent_root
= info
->extent_root
;
2158 struct btrfs_extent_item
*extent_item
;
2159 struct btrfs_extent_ref
*ref
;
2160 struct btrfs_path
*path
;
2161 struct btrfs_key keys
[2];
2163 /* block accounting for super block */
2164 spin_lock_irq(&info
->delalloc_lock
);
2165 super_used
= btrfs_super_bytes_used(&info
->super_copy
);
2166 btrfs_set_super_bytes_used(&info
->super_copy
, super_used
+ num_bytes
);
2167 spin_unlock_irq(&info
->delalloc_lock
);
2169 /* block accounting for root item */
2170 root_used
= btrfs_root_used(&root
->root_item
);
2171 btrfs_set_root_used(&root
->root_item
, root_used
+ num_bytes
);
2173 if (root
== extent_root
) {
2174 set_extent_bits(&root
->fs_info
->extent_ins
, ins
->objectid
,
2175 ins
->objectid
+ ins
->offset
- 1,
2176 EXTENT_LOCKED
, GFP_NOFS
);
2180 memcpy(&keys
[0], ins
, sizeof(*ins
));
2181 keys
[1].offset
= hash_extent_ref(root_objectid
, ref_generation
,
2182 owner
, owner_offset
);
2183 keys
[1].objectid
= ins
->objectid
;
2184 keys
[1].type
= BTRFS_EXTENT_REF_KEY
;
2185 sizes
[0] = sizeof(*extent_item
);
2186 sizes
[1] = sizeof(*ref
);
2188 path
= btrfs_alloc_path();
2191 ret
= btrfs_insert_empty_items(trans
, extent_root
, path
, keys
,
2195 extent_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2196 struct btrfs_extent_item
);
2197 btrfs_set_extent_refs(path
->nodes
[0], extent_item
, 1);
2198 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
2199 struct btrfs_extent_ref
);
2201 btrfs_set_ref_root(path
->nodes
[0], ref
, root_objectid
);
2202 btrfs_set_ref_generation(path
->nodes
[0], ref
, ref_generation
);
2203 btrfs_set_ref_objectid(path
->nodes
[0], ref
, owner
);
2204 btrfs_set_ref_offset(path
->nodes
[0], ref
, owner_offset
);
2206 btrfs_mark_buffer_dirty(path
->nodes
[0]);
2208 trans
->alloc_exclude_start
= 0;
2209 trans
->alloc_exclude_nr
= 0;
2210 btrfs_free_path(path
);
2211 finish_current_insert(trans
, extent_root
);
2212 pending_ret
= del_pending_extents(trans
, extent_root
);
2222 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1, 0);
2224 printk("update block group failed for %Lu %Lu\n",
2225 ins
->objectid
, ins
->offset
);
2232 int btrfs_alloc_reserved_extent(struct btrfs_trans_handle
*trans
,
2233 struct btrfs_root
*root
,
2234 u64 root_objectid
, u64 ref_generation
,
2235 u64 owner
, u64 owner_offset
,
2236 struct btrfs_key
*ins
)
2239 maybe_lock_mutex(root
);
2240 ret
= __btrfs_alloc_reserved_extent(trans
, root
, root_objectid
,
2241 ref_generation
, owner
,
2243 maybe_unlock_mutex(root
);
2247 * finds a free extent and does all the dirty work required for allocation
2248 * returns the key for the extent through ins, and a tree buffer for
2249 * the first block of the extent through buf.
2251 * returns 0 if everything worked, non-zero otherwise.
2253 int btrfs_alloc_extent(struct btrfs_trans_handle
*trans
,
2254 struct btrfs_root
*root
,
2255 u64 num_bytes
, u64 min_alloc_size
,
2256 u64 root_objectid
, u64 ref_generation
,
2257 u64 owner
, u64 owner_offset
,
2258 u64 empty_size
, u64 hint_byte
,
2259 u64 search_end
, struct btrfs_key
*ins
, u64 data
)
2263 maybe_lock_mutex(root
);
2265 ret
= __btrfs_reserve_extent(trans
, root
, num_bytes
,
2266 min_alloc_size
, empty_size
, hint_byte
,
2267 search_end
, ins
, data
);
2269 ret
= __btrfs_alloc_reserved_extent(trans
, root
, root_objectid
,
2270 ref_generation
, owner
,
2274 maybe_unlock_mutex(root
);
2278 struct extent_buffer
*btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
,
2279 struct btrfs_root
*root
,
2280 u64 bytenr
, u32 blocksize
)
2282 struct extent_buffer
*buf
;
2284 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
2286 return ERR_PTR(-ENOMEM
);
2287 btrfs_set_header_generation(buf
, trans
->transid
);
2288 btrfs_tree_lock(buf
);
2289 clean_tree_block(trans
, root
, buf
);
2290 btrfs_set_buffer_uptodate(buf
);
2291 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
2292 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
2293 trans
->blocks_used
++;
2298 * helper function to allocate a block for a given tree
2299 * returns the tree buffer or NULL.
2301 struct extent_buffer
*btrfs_alloc_free_block(struct btrfs_trans_handle
*trans
,
2302 struct btrfs_root
*root
,
2311 struct btrfs_key ins
;
2313 struct extent_buffer
*buf
;
2315 ret
= btrfs_alloc_extent(trans
, root
, blocksize
, blocksize
,
2316 root_objectid
, ref_generation
,
2317 level
, first_objectid
, empty_size
, hint
,
2321 return ERR_PTR(ret
);
2324 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
, blocksize
);
2328 static int noinline
drop_leaf_ref_no_cache(struct btrfs_trans_handle
*trans
,
2329 struct btrfs_root
*root
,
2330 struct extent_buffer
*leaf
)
2333 u64 leaf_generation
;
2334 struct btrfs_key key
;
2335 struct btrfs_file_extent_item
*fi
;
2340 BUG_ON(!btrfs_is_leaf(leaf
));
2341 nritems
= btrfs_header_nritems(leaf
);
2342 leaf_owner
= btrfs_header_owner(leaf
);
2343 leaf_generation
= btrfs_header_generation(leaf
);
2345 for (i
= 0; i
< nritems
; i
++) {
2349 btrfs_item_key_to_cpu(leaf
, &key
, i
);
2350 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
2352 fi
= btrfs_item_ptr(leaf
, i
, struct btrfs_file_extent_item
);
2353 if (btrfs_file_extent_type(leaf
, fi
) ==
2354 BTRFS_FILE_EXTENT_INLINE
)
2357 * FIXME make sure to insert a trans record that
2358 * repeats the snapshot del on crash
2360 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
2361 if (disk_bytenr
== 0)
2364 mutex_lock(&root
->fs_info
->alloc_mutex
);
2365 ret
= __btrfs_free_extent(trans
, root
, disk_bytenr
,
2366 btrfs_file_extent_disk_num_bytes(leaf
, fi
),
2367 leaf_owner
, leaf_generation
,
2368 key
.objectid
, key
.offset
, 0);
2369 mutex_unlock(&root
->fs_info
->alloc_mutex
);
2371 atomic_inc(&root
->fs_info
->throttle_gen
);
2372 wake_up(&root
->fs_info
->transaction_throttle
);
2380 static int noinline
drop_leaf_ref(struct btrfs_trans_handle
*trans
,
2381 struct btrfs_root
*root
,
2382 struct btrfs_leaf_ref
*ref
)
2386 struct btrfs_extent_info
*info
= ref
->extents
;
2388 for (i
= 0; i
< ref
->nritems
; i
++) {
2389 mutex_lock(&root
->fs_info
->alloc_mutex
);
2390 ret
= __btrfs_free_extent(trans
, root
,
2391 info
->bytenr
, info
->num_bytes
,
2392 ref
->owner
, ref
->generation
,
2393 info
->objectid
, info
->offset
, 0);
2394 mutex_unlock(&root
->fs_info
->alloc_mutex
);
2396 atomic_inc(&root
->fs_info
->throttle_gen
);
2397 wake_up(&root
->fs_info
->transaction_throttle
);
2407 int drop_snap_lookup_refcount(struct btrfs_root
*root
, u64 start
, u64 len
,
2412 ret
= lookup_extent_ref(NULL
, root
, start
, len
, refs
);
2415 #if 0 // some debugging code in case we see problems here
2416 /* if the refs count is one, it won't get increased again. But
2417 * if the ref count is > 1, someone may be decreasing it at
2418 * the same time we are.
2421 struct extent_buffer
*eb
= NULL
;
2422 eb
= btrfs_find_create_tree_block(root
, start
, len
);
2424 btrfs_tree_lock(eb
);
2426 mutex_lock(&root
->fs_info
->alloc_mutex
);
2427 ret
= lookup_extent_ref(NULL
, root
, start
, len
, refs
);
2429 mutex_unlock(&root
->fs_info
->alloc_mutex
);
2432 btrfs_tree_unlock(eb
);
2433 free_extent_buffer(eb
);
2436 printk("block %llu went down to one during drop_snap\n",
2437 (unsigned long long)start
);
2448 * helper function for drop_snapshot, this walks down the tree dropping ref
2449 * counts as it goes.
2451 static int noinline
walk_down_tree(struct btrfs_trans_handle
*trans
,
2452 struct btrfs_root
*root
,
2453 struct btrfs_path
*path
, int *level
)
2459 struct extent_buffer
*next
;
2460 struct extent_buffer
*cur
;
2461 struct extent_buffer
*parent
;
2462 struct btrfs_leaf_ref
*ref
;
2467 WARN_ON(*level
< 0);
2468 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
2469 ret
= drop_snap_lookup_refcount(root
, path
->nodes
[*level
]->start
,
2470 path
->nodes
[*level
]->len
, &refs
);
2476 * walk down to the last node level and free all the leaves
2478 while(*level
>= 0) {
2479 WARN_ON(*level
< 0);
2480 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
2481 cur
= path
->nodes
[*level
];
2483 if (btrfs_header_level(cur
) != *level
)
2486 if (path
->slots
[*level
] >=
2487 btrfs_header_nritems(cur
))
2490 ret
= drop_leaf_ref_no_cache(trans
, root
, cur
);
2494 bytenr
= btrfs_node_blockptr(cur
, path
->slots
[*level
]);
2495 ptr_gen
= btrfs_node_ptr_generation(cur
, path
->slots
[*level
]);
2496 blocksize
= btrfs_level_size(root
, *level
- 1);
2498 ret
= drop_snap_lookup_refcount(root
, bytenr
, blocksize
, &refs
);
2501 parent
= path
->nodes
[*level
];
2502 root_owner
= btrfs_header_owner(parent
);
2503 root_gen
= btrfs_header_generation(parent
);
2504 path
->slots
[*level
]++;
2506 mutex_lock(&root
->fs_info
->alloc_mutex
);
2507 ret
= __btrfs_free_extent(trans
, root
, bytenr
,
2508 blocksize
, root_owner
,
2511 mutex_unlock(&root
->fs_info
->alloc_mutex
);
2513 atomic_inc(&root
->fs_info
->throttle_gen
);
2514 wake_up(&root
->fs_info
->transaction_throttle
);
2520 * at this point, we have a single ref, and since the
2521 * only place referencing this extent is a dead root
2522 * the reference count should never go higher.
2523 * So, we don't need to check it again
2526 struct btrfs_key key
;
2527 btrfs_node_key_to_cpu(cur
, &key
, path
->slots
[*level
]);
2528 ref
= btrfs_lookup_leaf_ref(root
, bytenr
);
2530 ret
= drop_leaf_ref(trans
, root
, ref
);
2532 btrfs_remove_leaf_ref(root
, ref
);
2533 btrfs_free_leaf_ref(root
, ref
);
2537 if (printk_ratelimit())
2538 printk("leaf ref miss for bytenr %llu\n",
2539 (unsigned long long)bytenr
);
2541 next
= btrfs_find_tree_block(root
, bytenr
, blocksize
);
2542 if (!next
|| !btrfs_buffer_uptodate(next
, ptr_gen
)) {
2543 free_extent_buffer(next
);
2545 next
= read_tree_block(root
, bytenr
, blocksize
,
2550 * this is a debugging check and can go away
2551 * the ref should never go all the way down to 1
2554 ret
= lookup_extent_ref(NULL
, root
, bytenr
, blocksize
,
2560 WARN_ON(*level
<= 0);
2561 if (path
->nodes
[*level
-1])
2562 free_extent_buffer(path
->nodes
[*level
-1]);
2563 path
->nodes
[*level
-1] = next
;
2564 *level
= btrfs_header_level(next
);
2565 path
->slots
[*level
] = 0;
2569 WARN_ON(*level
< 0);
2570 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
2572 if (path
->nodes
[*level
] == root
->node
) {
2573 parent
= path
->nodes
[*level
];
2574 bytenr
= path
->nodes
[*level
]->start
;
2576 parent
= path
->nodes
[*level
+ 1];
2577 bytenr
= btrfs_node_blockptr(parent
, path
->slots
[*level
+ 1]);
2580 blocksize
= btrfs_level_size(root
, *level
);
2581 root_owner
= btrfs_header_owner(parent
);
2582 root_gen
= btrfs_header_generation(parent
);
2584 mutex_lock(&root
->fs_info
->alloc_mutex
);
2585 ret
= __btrfs_free_extent(trans
, root
, bytenr
, blocksize
,
2586 root_owner
, root_gen
, 0, 0, 1);
2587 free_extent_buffer(path
->nodes
[*level
]);
2588 path
->nodes
[*level
] = NULL
;
2591 mutex_unlock(&root
->fs_info
->alloc_mutex
);
2598 * helper for dropping snapshots. This walks back up the tree in the path
2599 * to find the first node higher up where we haven't yet gone through
2602 static int noinline
walk_up_tree(struct btrfs_trans_handle
*trans
,
2603 struct btrfs_root
*root
,
2604 struct btrfs_path
*path
, int *level
)
2608 struct btrfs_root_item
*root_item
= &root
->root_item
;
2613 for(i
= *level
; i
< BTRFS_MAX_LEVEL
- 1 && path
->nodes
[i
]; i
++) {
2614 slot
= path
->slots
[i
];
2615 if (slot
< btrfs_header_nritems(path
->nodes
[i
]) - 1) {
2616 struct extent_buffer
*node
;
2617 struct btrfs_disk_key disk_key
;
2618 node
= path
->nodes
[i
];
2621 WARN_ON(*level
== 0);
2622 btrfs_node_key(node
, &disk_key
, path
->slots
[i
]);
2623 memcpy(&root_item
->drop_progress
,
2624 &disk_key
, sizeof(disk_key
));
2625 root_item
->drop_level
= i
;
2628 if (path
->nodes
[*level
] == root
->node
) {
2629 root_owner
= root
->root_key
.objectid
;
2631 btrfs_header_generation(path
->nodes
[*level
]);
2633 struct extent_buffer
*node
;
2634 node
= path
->nodes
[*level
+ 1];
2635 root_owner
= btrfs_header_owner(node
);
2636 root_gen
= btrfs_header_generation(node
);
2638 ret
= btrfs_free_extent(trans
, root
,
2639 path
->nodes
[*level
]->start
,
2640 path
->nodes
[*level
]->len
,
2641 root_owner
, root_gen
, 0, 0, 1);
2643 free_extent_buffer(path
->nodes
[*level
]);
2644 path
->nodes
[*level
] = NULL
;
2652 * drop the reference count on the tree rooted at 'snap'. This traverses
2653 * the tree freeing any blocks that have a ref count of zero after being
2656 int btrfs_drop_snapshot(struct btrfs_trans_handle
*trans
, struct btrfs_root
2662 struct btrfs_path
*path
;
2665 struct btrfs_root_item
*root_item
= &root
->root_item
;
2667 WARN_ON(!mutex_is_locked(&root
->fs_info
->drop_mutex
));
2668 path
= btrfs_alloc_path();
2671 level
= btrfs_header_level(root
->node
);
2673 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
2674 path
->nodes
[level
] = root
->node
;
2675 extent_buffer_get(root
->node
);
2676 path
->slots
[level
] = 0;
2678 struct btrfs_key key
;
2679 struct btrfs_disk_key found_key
;
2680 struct extent_buffer
*node
;
2682 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
2683 level
= root_item
->drop_level
;
2684 path
->lowest_level
= level
;
2685 wret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2690 node
= path
->nodes
[level
];
2691 btrfs_node_key(node
, &found_key
, path
->slots
[level
]);
2692 WARN_ON(memcmp(&found_key
, &root_item
->drop_progress
,
2693 sizeof(found_key
)));
2695 * unlock our path, this is safe because only this
2696 * function is allowed to delete this snapshot
2698 for (i
= 0; i
< BTRFS_MAX_LEVEL
; i
++) {
2699 if (path
->nodes
[i
] && path
->locks
[i
]) {
2701 btrfs_tree_unlock(path
->nodes
[i
]);
2706 wret
= walk_down_tree(trans
, root
, path
, &level
);
2712 wret
= walk_up_tree(trans
, root
, path
, &level
);
2717 if (trans
->transaction
->in_commit
) {
2721 atomic_inc(&root
->fs_info
->throttle_gen
);
2722 wake_up(&root
->fs_info
->transaction_throttle
);
2724 for (i
= 0; i
<= orig_level
; i
++) {
2725 if (path
->nodes
[i
]) {
2726 free_extent_buffer(path
->nodes
[i
]);
2727 path
->nodes
[i
] = NULL
;
2731 btrfs_free_path(path
);
2735 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
2742 mutex_lock(&info
->alloc_mutex
);
2744 ret
= find_first_extent_bit(&info
->block_group_cache
, 0,
2745 &start
, &end
, (unsigned int)-1);
2748 ret
= get_state_private(&info
->block_group_cache
, start
, &ptr
);
2750 kfree((void *)(unsigned long)ptr
);
2751 clear_extent_bits(&info
->block_group_cache
, start
,
2752 end
, (unsigned int)-1, GFP_NOFS
);
2755 ret
= find_first_extent_bit(&info
->free_space_cache
, 0,
2756 &start
, &end
, EXTENT_DIRTY
);
2759 clear_extent_dirty(&info
->free_space_cache
, start
,
2762 mutex_unlock(&info
->alloc_mutex
);
2766 static unsigned long calc_ra(unsigned long start
, unsigned long last
,
2769 return min(last
, start
+ nr
- 1);
2772 static int noinline
relocate_inode_pages(struct inode
*inode
, u64 start
,
2777 unsigned long last_index
;
2780 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2781 struct file_ra_state
*ra
;
2782 unsigned long total_read
= 0;
2783 unsigned long ra_pages
;
2784 struct btrfs_ordered_extent
*ordered
;
2785 struct btrfs_trans_handle
*trans
;
2787 ra
= kzalloc(sizeof(*ra
), GFP_NOFS
);
2789 mutex_lock(&inode
->i_mutex
);
2790 i
= start
>> PAGE_CACHE_SHIFT
;
2791 last_index
= (start
+ len
- 1) >> PAGE_CACHE_SHIFT
;
2793 ra_pages
= BTRFS_I(inode
)->root
->fs_info
->bdi
.ra_pages
;
2795 file_ra_state_init(ra
, inode
->i_mapping
);
2797 for (; i
<= last_index
; i
++) {
2798 if (total_read
% ra_pages
== 0) {
2799 btrfs_force_ra(inode
->i_mapping
, ra
, NULL
, i
,
2800 calc_ra(i
, last_index
, ra_pages
));
2804 if (((u64
)i
<< PAGE_CACHE_SHIFT
) > i_size_read(inode
))
2805 goto truncate_racing
;
2806 page
= grab_cache_page(inode
->i_mapping
, i
);
2810 if (!PageUptodate(page
)) {
2811 btrfs_readpage(NULL
, page
);
2813 if (!PageUptodate(page
)) {
2815 page_cache_release(page
);
2819 wait_on_page_writeback(page
);
2821 page_start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2822 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
2823 lock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
2825 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
2827 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
2829 page_cache_release(page
);
2830 btrfs_start_ordered_extent(inode
, ordered
, 1);
2831 btrfs_put_ordered_extent(ordered
);
2834 set_page_extent_mapped(page
);
2837 * make sure page_mkwrite is called for this page if userland
2838 * wants to change it from mmap
2840 clear_page_dirty_for_io(page
);
2842 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
2843 set_page_dirty(page
);
2845 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
2847 page_cache_release(page
);
2851 /* we have to start the IO in order to get the ordered extents
2852 * instantiated. This allows the relocation to code to wait
2853 * for all the ordered extents to hit the disk.
2855 * Otherwise, it would constantly loop over the same extents
2856 * because the old ones don't get deleted until the IO is
2859 btrfs_fdatawrite_range(inode
->i_mapping
, start
, start
+ len
- 1,
2862 trans
= btrfs_start_transaction(BTRFS_I(inode
)->root
, 1);
2864 btrfs_end_transaction(trans
, BTRFS_I(inode
)->root
);
2865 mark_inode_dirty(inode
);
2867 mutex_unlock(&inode
->i_mutex
);
2871 vmtruncate(inode
, inode
->i_size
);
2872 balance_dirty_pages_ratelimited_nr(inode
->i_mapping
,
2878 * The back references tell us which tree holds a ref on a block,
2879 * but it is possible for the tree root field in the reference to
2880 * reflect the original root before a snapshot was made. In this
2881 * case we should search through all the children of a given root
2882 * to find potential holders of references on a block.
2884 * Instead, we do something a little less fancy and just search
2885 * all the roots for a given key/block combination.
2887 static int find_root_for_ref(struct btrfs_root
*root
,
2888 struct btrfs_path
*path
,
2889 struct btrfs_key
*key0
,
2892 struct btrfs_root
**found_root
,
2895 struct btrfs_key root_location
;
2896 struct btrfs_root
*cur_root
= *found_root
;
2897 struct btrfs_file_extent_item
*file_extent
;
2898 u64 root_search_start
= BTRFS_FS_TREE_OBJECTID
;
2902 root_location
.offset
= (u64
)-1;
2903 root_location
.type
= BTRFS_ROOT_ITEM_KEY
;
2904 path
->lowest_level
= level
;
2907 ret
= btrfs_search_slot(NULL
, cur_root
, key0
, path
, 0, 0);
2909 if (ret
== 0 && file_key
) {
2910 struct extent_buffer
*leaf
= path
->nodes
[0];
2911 file_extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
2912 struct btrfs_file_extent_item
);
2913 if (btrfs_file_extent_type(leaf
, file_extent
) ==
2914 BTRFS_FILE_EXTENT_REG
) {
2916 btrfs_file_extent_disk_bytenr(leaf
,
2919 } else if (!file_key
) {
2920 if (path
->nodes
[level
])
2921 found_bytenr
= path
->nodes
[level
]->start
;
2924 btrfs_release_path(cur_root
, path
);
2926 if (found_bytenr
== bytenr
) {
2927 *found_root
= cur_root
;
2931 ret
= btrfs_search_root(root
->fs_info
->tree_root
,
2932 root_search_start
, &root_search_start
);
2936 root_location
.objectid
= root_search_start
;
2937 cur_root
= btrfs_read_fs_root_no_name(root
->fs_info
,
2945 path
->lowest_level
= 0;
2950 * note, this releases the path
2952 static int noinline
relocate_one_reference(struct btrfs_root
*extent_root
,
2953 struct btrfs_path
*path
,
2954 struct btrfs_key
*extent_key
,
2955 u64
*last_file_objectid
,
2956 u64
*last_file_offset
,
2957 u64
*last_file_root
,
2960 struct inode
*inode
;
2961 struct btrfs_root
*found_root
;
2962 struct btrfs_key root_location
;
2963 struct btrfs_key found_key
;
2964 struct btrfs_extent_ref
*ref
;
2972 WARN_ON(!mutex_is_locked(&extent_root
->fs_info
->alloc_mutex
));
2974 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2975 struct btrfs_extent_ref
);
2976 ref_root
= btrfs_ref_root(path
->nodes
[0], ref
);
2977 ref_gen
= btrfs_ref_generation(path
->nodes
[0], ref
);
2978 ref_objectid
= btrfs_ref_objectid(path
->nodes
[0], ref
);
2979 ref_offset
= btrfs_ref_offset(path
->nodes
[0], ref
);
2980 btrfs_release_path(extent_root
, path
);
2982 root_location
.objectid
= ref_root
;
2984 root_location
.offset
= 0;
2986 root_location
.offset
= (u64
)-1;
2987 root_location
.type
= BTRFS_ROOT_ITEM_KEY
;
2989 found_root
= btrfs_read_fs_root_no_name(extent_root
->fs_info
,
2991 BUG_ON(!found_root
);
2992 mutex_unlock(&extent_root
->fs_info
->alloc_mutex
);
2994 if (ref_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
2995 found_key
.objectid
= ref_objectid
;
2996 found_key
.type
= BTRFS_EXTENT_DATA_KEY
;
2997 found_key
.offset
= ref_offset
;
3000 if (last_extent
== extent_key
->objectid
&&
3001 *last_file_objectid
== ref_objectid
&&
3002 *last_file_offset
== ref_offset
&&
3003 *last_file_root
== ref_root
)
3006 ret
= find_root_for_ref(extent_root
, path
, &found_key
,
3007 level
, 1, &found_root
,
3008 extent_key
->objectid
);
3013 if (last_extent
== extent_key
->objectid
&&
3014 *last_file_objectid
== ref_objectid
&&
3015 *last_file_offset
== ref_offset
&&
3016 *last_file_root
== ref_root
)
3019 inode
= btrfs_iget_locked(extent_root
->fs_info
->sb
,
3020 ref_objectid
, found_root
);
3021 if (inode
->i_state
& I_NEW
) {
3022 /* the inode and parent dir are two different roots */
3023 BTRFS_I(inode
)->root
= found_root
;
3024 BTRFS_I(inode
)->location
.objectid
= ref_objectid
;
3025 BTRFS_I(inode
)->location
.type
= BTRFS_INODE_ITEM_KEY
;
3026 BTRFS_I(inode
)->location
.offset
= 0;
3027 btrfs_read_locked_inode(inode
);
3028 unlock_new_inode(inode
);
3031 /* this can happen if the reference is not against
3032 * the latest version of the tree root
3034 if (is_bad_inode(inode
))
3037 *last_file_objectid
= inode
->i_ino
;
3038 *last_file_root
= found_root
->root_key
.objectid
;
3039 *last_file_offset
= ref_offset
;
3041 relocate_inode_pages(inode
, ref_offset
, extent_key
->offset
);
3044 struct btrfs_trans_handle
*trans
;
3045 struct extent_buffer
*eb
;
3048 eb
= read_tree_block(found_root
, extent_key
->objectid
,
3049 extent_key
->offset
, 0);
3050 btrfs_tree_lock(eb
);
3051 level
= btrfs_header_level(eb
);
3054 btrfs_item_key_to_cpu(eb
, &found_key
, 0);
3056 btrfs_node_key_to_cpu(eb
, &found_key
, 0);
3058 btrfs_tree_unlock(eb
);
3059 free_extent_buffer(eb
);
3061 ret
= find_root_for_ref(extent_root
, path
, &found_key
,
3062 level
, 0, &found_root
,
3063 extent_key
->objectid
);
3069 * right here almost anything could happen to our key,
3070 * but that's ok. The cow below will either relocate it
3071 * or someone else will have relocated it. Either way,
3072 * it is in a different spot than it was before and
3076 trans
= btrfs_start_transaction(found_root
, 1);
3078 if (found_root
== extent_root
->fs_info
->extent_root
||
3079 found_root
== extent_root
->fs_info
->chunk_root
||
3080 found_root
== extent_root
->fs_info
->dev_root
) {
3082 mutex_lock(&extent_root
->fs_info
->alloc_mutex
);
3085 path
->lowest_level
= level
;
3087 ret
= btrfs_search_slot(trans
, found_root
, &found_key
, path
,
3089 path
->lowest_level
= 0;
3090 btrfs_release_path(found_root
, path
);
3092 if (found_root
== found_root
->fs_info
->extent_root
)
3093 btrfs_extent_post_op(trans
, found_root
);
3095 mutex_unlock(&extent_root
->fs_info
->alloc_mutex
);
3097 btrfs_end_transaction(trans
, found_root
);
3101 mutex_lock(&extent_root
->fs_info
->alloc_mutex
);
3105 static int noinline
del_extent_zero(struct btrfs_root
*extent_root
,
3106 struct btrfs_path
*path
,
3107 struct btrfs_key
*extent_key
)
3110 struct btrfs_trans_handle
*trans
;
3112 trans
= btrfs_start_transaction(extent_root
, 1);
3113 ret
= btrfs_search_slot(trans
, extent_root
, extent_key
, path
, -1, 1);
3120 ret
= btrfs_del_item(trans
, extent_root
, path
);
3122 btrfs_end_transaction(trans
, extent_root
);
3126 static int noinline
relocate_one_extent(struct btrfs_root
*extent_root
,
3127 struct btrfs_path
*path
,
3128 struct btrfs_key
*extent_key
)
3130 struct btrfs_key key
;
3131 struct btrfs_key found_key
;
3132 struct extent_buffer
*leaf
;
3133 u64 last_file_objectid
= 0;
3134 u64 last_file_root
= 0;
3135 u64 last_file_offset
= (u64
)-1;
3136 u64 last_extent
= 0;
3141 if (extent_key
->objectid
== 0) {
3142 ret
= del_extent_zero(extent_root
, path
, extent_key
);
3145 key
.objectid
= extent_key
->objectid
;
3146 key
.type
= BTRFS_EXTENT_REF_KEY
;
3150 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
3156 leaf
= path
->nodes
[0];
3157 nritems
= btrfs_header_nritems(leaf
);
3158 if (path
->slots
[0] == nritems
) {
3159 ret
= btrfs_next_leaf(extent_root
, path
);
3166 leaf
= path
->nodes
[0];
3169 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3170 if (found_key
.objectid
!= extent_key
->objectid
) {
3174 if (found_key
.type
!= BTRFS_EXTENT_REF_KEY
) {
3178 key
.offset
= found_key
.offset
+ 1;
3179 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
3181 ret
= relocate_one_reference(extent_root
, path
, extent_key
,
3182 &last_file_objectid
,
3184 &last_file_root
, last_extent
);
3187 last_extent
= extent_key
->objectid
;
3191 btrfs_release_path(extent_root
, path
);
3195 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
3198 u64 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
3199 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
3201 num_devices
= root
->fs_info
->fs_devices
->num_devices
;
3202 if (num_devices
== 1) {
3203 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
3204 stripped
= flags
& ~stripped
;
3206 /* turn raid0 into single device chunks */
3207 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
3210 /* turn mirroring into duplication */
3211 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
3212 BTRFS_BLOCK_GROUP_RAID10
))
3213 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
3216 /* they already had raid on here, just return */
3217 if (flags
& stripped
)
3220 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
3221 stripped
= flags
& ~stripped
;
3223 /* switch duplicated blocks with raid1 */
3224 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
3225 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
3227 /* turn single device chunks into raid0 */
3228 return stripped
| BTRFS_BLOCK_GROUP_RAID0
;
3233 int __alloc_chunk_for_shrink(struct btrfs_root
*root
,
3234 struct btrfs_block_group_cache
*shrink_block_group
,
3237 struct btrfs_trans_handle
*trans
;
3238 u64 new_alloc_flags
;
3241 spin_lock(&shrink_block_group
->lock
);
3242 if (btrfs_block_group_used(&shrink_block_group
->item
) > 0) {
3243 spin_unlock(&shrink_block_group
->lock
);
3244 mutex_unlock(&root
->fs_info
->alloc_mutex
);
3246 trans
= btrfs_start_transaction(root
, 1);
3247 mutex_lock(&root
->fs_info
->alloc_mutex
);
3248 spin_lock(&shrink_block_group
->lock
);
3250 new_alloc_flags
= update_block_group_flags(root
,
3251 shrink_block_group
->flags
);
3252 if (new_alloc_flags
!= shrink_block_group
->flags
) {
3254 btrfs_block_group_used(&shrink_block_group
->item
);
3256 calc
= shrink_block_group
->key
.offset
;
3258 spin_unlock(&shrink_block_group
->lock
);
3260 do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
3261 calc
+ 2 * 1024 * 1024, new_alloc_flags
, force
);
3263 mutex_unlock(&root
->fs_info
->alloc_mutex
);
3264 btrfs_end_transaction(trans
, root
);
3265 mutex_lock(&root
->fs_info
->alloc_mutex
);
3267 spin_unlock(&shrink_block_group
->lock
);
3271 int btrfs_shrink_extent_tree(struct btrfs_root
*root
, u64 shrink_start
)
3273 struct btrfs_trans_handle
*trans
;
3274 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
3275 struct btrfs_path
*path
;
3278 u64 shrink_last_byte
;
3279 struct btrfs_block_group_cache
*shrink_block_group
;
3280 struct btrfs_fs_info
*info
= root
->fs_info
;
3281 struct btrfs_key key
;
3282 struct btrfs_key found_key
;
3283 struct extent_buffer
*leaf
;
3288 mutex_lock(&root
->fs_info
->alloc_mutex
);
3289 shrink_block_group
= btrfs_lookup_block_group(root
->fs_info
,
3291 BUG_ON(!shrink_block_group
);
3293 shrink_last_byte
= shrink_block_group
->key
.objectid
+
3294 shrink_block_group
->key
.offset
;
3296 shrink_block_group
->space_info
->total_bytes
-=
3297 shrink_block_group
->key
.offset
;
3298 path
= btrfs_alloc_path();
3299 root
= root
->fs_info
->extent_root
;
3302 printk("btrfs relocating block group %llu flags %llu\n",
3303 (unsigned long long)shrink_start
,
3304 (unsigned long long)shrink_block_group
->flags
);
3306 __alloc_chunk_for_shrink(root
, shrink_block_group
, 1);
3310 shrink_block_group
->ro
= 1;
3314 key
.objectid
= shrink_start
;
3317 cur_byte
= key
.objectid
;
3319 mutex_unlock(&root
->fs_info
->alloc_mutex
);
3321 btrfs_start_delalloc_inodes(root
);
3322 btrfs_wait_ordered_extents(tree_root
, 0);
3324 mutex_lock(&root
->fs_info
->alloc_mutex
);
3326 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3330 ret
= btrfs_previous_item(root
, path
, 0, BTRFS_EXTENT_ITEM_KEY
);
3335 leaf
= path
->nodes
[0];
3336 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3337 if (found_key
.objectid
+ found_key
.offset
> shrink_start
&&
3338 found_key
.objectid
< shrink_last_byte
) {
3339 cur_byte
= found_key
.objectid
;
3340 key
.objectid
= cur_byte
;
3343 btrfs_release_path(root
, path
);
3346 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3351 leaf
= path
->nodes
[0];
3352 nritems
= btrfs_header_nritems(leaf
);
3353 if (path
->slots
[0] >= nritems
) {
3354 ret
= btrfs_next_leaf(root
, path
);
3361 leaf
= path
->nodes
[0];
3362 nritems
= btrfs_header_nritems(leaf
);
3365 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3367 if (found_key
.objectid
>= shrink_last_byte
)
3370 if (progress
&& need_resched()) {
3371 memcpy(&key
, &found_key
, sizeof(key
));
3373 btrfs_release_path(root
, path
);
3374 btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3380 if (btrfs_key_type(&found_key
) != BTRFS_EXTENT_ITEM_KEY
||
3381 found_key
.objectid
+ found_key
.offset
<= cur_byte
) {
3382 memcpy(&key
, &found_key
, sizeof(key
));
3389 cur_byte
= found_key
.objectid
+ found_key
.offset
;
3390 key
.objectid
= cur_byte
;
3391 btrfs_release_path(root
, path
);
3392 ret
= relocate_one_extent(root
, path
, &found_key
);
3393 __alloc_chunk_for_shrink(root
, shrink_block_group
, 0);
3396 btrfs_release_path(root
, path
);
3398 if (total_found
> 0) {
3399 printk("btrfs relocate found %llu last extent was %llu\n",
3400 (unsigned long long)total_found
,
3401 (unsigned long long)found_key
.objectid
);
3402 mutex_unlock(&root
->fs_info
->alloc_mutex
);
3403 trans
= btrfs_start_transaction(tree_root
, 1);
3404 btrfs_commit_transaction(trans
, tree_root
);
3406 btrfs_clean_old_snapshots(tree_root
);
3408 btrfs_start_delalloc_inodes(root
);
3409 btrfs_wait_ordered_extents(tree_root
, 0);
3411 trans
= btrfs_start_transaction(tree_root
, 1);
3412 btrfs_commit_transaction(trans
, tree_root
);
3413 mutex_lock(&root
->fs_info
->alloc_mutex
);
3418 * we've freed all the extents, now remove the block
3419 * group item from the tree
3421 mutex_unlock(&root
->fs_info
->alloc_mutex
);
3423 trans
= btrfs_start_transaction(root
, 1);
3425 mutex_lock(&root
->fs_info
->alloc_mutex
);
3426 memcpy(&key
, &shrink_block_group
->key
, sizeof(key
));
3428 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
3432 btrfs_end_transaction(trans
, root
);
3436 clear_extent_bits(&info
->block_group_cache
, key
.objectid
,
3437 key
.objectid
+ key
.offset
- 1,
3438 (unsigned int)-1, GFP_NOFS
);
3441 clear_extent_bits(&info
->free_space_cache
,
3442 key
.objectid
, key
.objectid
+ key
.offset
- 1,
3443 (unsigned int)-1, GFP_NOFS
);
3446 memset(shrink_block_group, 0, sizeof(*shrink_block_group));
3447 kfree(shrink_block_group);
3450 btrfs_del_item(trans
, root
, path
);
3451 btrfs_release_path(root
, path
);
3452 mutex_unlock(&root
->fs_info
->alloc_mutex
);
3453 btrfs_commit_transaction(trans
, root
);
3455 mutex_lock(&root
->fs_info
->alloc_mutex
);
3457 /* the code to unpin extents might set a few bits in the free
3458 * space cache for this range again
3460 clear_extent_bits(&info
->free_space_cache
,
3461 key
.objectid
, key
.objectid
+ key
.offset
- 1,
3462 (unsigned int)-1, GFP_NOFS
);
3464 btrfs_free_path(path
);
3465 mutex_unlock(&root
->fs_info
->alloc_mutex
);
3469 int find_first_block_group(struct btrfs_root
*root
, struct btrfs_path
*path
,
3470 struct btrfs_key
*key
)
3473 struct btrfs_key found_key
;
3474 struct extent_buffer
*leaf
;
3477 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
3482 slot
= path
->slots
[0];
3483 leaf
= path
->nodes
[0];
3484 if (slot
>= btrfs_header_nritems(leaf
)) {
3485 ret
= btrfs_next_leaf(root
, path
);
3492 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3494 if (found_key
.objectid
>= key
->objectid
&&
3495 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
3506 int btrfs_read_block_groups(struct btrfs_root
*root
)
3508 struct btrfs_path
*path
;
3511 struct btrfs_block_group_cache
*cache
;
3512 struct btrfs_fs_info
*info
= root
->fs_info
;
3513 struct btrfs_space_info
*space_info
;
3514 struct extent_io_tree
*block_group_cache
;
3515 struct btrfs_key key
;
3516 struct btrfs_key found_key
;
3517 struct extent_buffer
*leaf
;
3519 block_group_cache
= &info
->block_group_cache
;
3520 root
= info
->extent_root
;
3523 btrfs_set_key_type(&key
, BTRFS_BLOCK_GROUP_ITEM_KEY
);
3524 path
= btrfs_alloc_path();
3528 mutex_lock(&root
->fs_info
->alloc_mutex
);
3530 ret
= find_first_block_group(root
, path
, &key
);
3538 leaf
= path
->nodes
[0];
3539 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3540 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
3546 spin_lock_init(&cache
->lock
);
3547 read_extent_buffer(leaf
, &cache
->item
,
3548 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
3549 sizeof(cache
->item
));
3550 memcpy(&cache
->key
, &found_key
, sizeof(found_key
));
3552 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
3553 btrfs_release_path(root
, path
);
3554 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
3556 if (cache
->flags
& BTRFS_BLOCK_GROUP_DATA
) {
3557 bit
= BLOCK_GROUP_DATA
;
3558 } else if (cache
->flags
& BTRFS_BLOCK_GROUP_SYSTEM
) {
3559 bit
= BLOCK_GROUP_SYSTEM
;
3560 } else if (cache
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
3561 bit
= BLOCK_GROUP_METADATA
;
3563 set_avail_alloc_bits(info
, cache
->flags
);
3565 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
3566 btrfs_block_group_used(&cache
->item
),
3569 cache
->space_info
= space_info
;
3571 /* use EXTENT_LOCKED to prevent merging */
3572 set_extent_bits(block_group_cache
, found_key
.objectid
,
3573 found_key
.objectid
+ found_key
.offset
- 1,
3574 EXTENT_LOCKED
, GFP_NOFS
);
3575 set_state_private(block_group_cache
, found_key
.objectid
,
3576 (unsigned long)cache
);
3577 set_extent_bits(block_group_cache
, found_key
.objectid
,
3578 found_key
.objectid
+ found_key
.offset
- 1,
3579 bit
| EXTENT_LOCKED
, GFP_NOFS
);
3581 btrfs_super_total_bytes(&info
->super_copy
))
3586 btrfs_free_path(path
);
3587 mutex_unlock(&root
->fs_info
->alloc_mutex
);
3591 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
3592 struct btrfs_root
*root
, u64 bytes_used
,
3593 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
3598 struct btrfs_root
*extent_root
;
3599 struct btrfs_block_group_cache
*cache
;
3600 struct extent_io_tree
*block_group_cache
;
3602 WARN_ON(!mutex_is_locked(&root
->fs_info
->alloc_mutex
));
3603 extent_root
= root
->fs_info
->extent_root
;
3604 block_group_cache
= &root
->fs_info
->block_group_cache
;
3606 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
3608 cache
->key
.objectid
= chunk_offset
;
3609 cache
->key
.offset
= size
;
3610 spin_lock_init(&cache
->lock
);
3611 btrfs_set_key_type(&cache
->key
, BTRFS_BLOCK_GROUP_ITEM_KEY
);
3613 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
3614 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
3615 cache
->flags
= type
;
3616 btrfs_set_block_group_flags(&cache
->item
, type
);
3618 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
3619 &cache
->space_info
);
3622 bit
= block_group_state_bits(type
);
3623 set_extent_bits(block_group_cache
, chunk_offset
,
3624 chunk_offset
+ size
- 1,
3625 EXTENT_LOCKED
, GFP_NOFS
);
3626 set_state_private(block_group_cache
, chunk_offset
,
3627 (unsigned long)cache
);
3628 set_extent_bits(block_group_cache
, chunk_offset
,
3629 chunk_offset
+ size
- 1,
3630 bit
| EXTENT_LOCKED
, GFP_NOFS
);
3632 ret
= btrfs_insert_item(trans
, extent_root
, &cache
->key
, &cache
->item
,
3633 sizeof(cache
->item
));
3636 finish_current_insert(trans
, extent_root
);
3637 ret
= del_pending_extents(trans
, extent_root
);
3639 set_avail_alloc_bits(extent_root
->fs_info
, type
);