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 #define BLOCK_GROUP_DATA EXTENT_WRITEBACK
30 #define BLOCK_GROUP_METADATA EXTENT_UPTODATE
31 #define BLOCK_GROUP_SYSTEM EXTENT_NEW
33 #define BLOCK_GROUP_DIRTY EXTENT_DIRTY
35 static int finish_current_insert(struct btrfs_trans_handle
*trans
, struct
36 btrfs_root
*extent_root
);
37 static int del_pending_extents(struct btrfs_trans_handle
*trans
, struct
38 btrfs_root
*extent_root
);
41 static int cache_block_group(struct btrfs_root
*root
,
42 struct btrfs_block_group_cache
*block_group
)
44 struct btrfs_path
*path
;
47 struct extent_buffer
*leaf
;
48 struct extent_io_tree
*free_space_cache
;
58 root
= root
->fs_info
->extent_root
;
59 free_space_cache
= &root
->fs_info
->free_space_cache
;
61 if (block_group
->cached
)
64 path
= btrfs_alloc_path();
69 first_free
= block_group
->key
.objectid
;
70 key
.objectid
= block_group
->key
.objectid
;
72 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
73 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
76 ret
= btrfs_previous_item(root
, path
, 0, BTRFS_EXTENT_ITEM_KEY
);
80 leaf
= path
->nodes
[0];
81 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
82 if (key
.objectid
+ key
.offset
> first_free
)
83 first_free
= key
.objectid
+ key
.offset
;
86 leaf
= path
->nodes
[0];
87 slot
= path
->slots
[0];
88 if (slot
>= btrfs_header_nritems(leaf
)) {
89 ret
= btrfs_next_leaf(root
, path
);
98 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
99 if (key
.objectid
< block_group
->key
.objectid
) {
102 if (key
.objectid
>= block_group
->key
.objectid
+
103 block_group
->key
.offset
) {
107 if (btrfs_key_type(&key
) == BTRFS_EXTENT_ITEM_KEY
) {
112 if (key
.objectid
> last
) {
113 hole_size
= key
.objectid
- last
;
114 set_extent_dirty(free_space_cache
, last
,
115 last
+ hole_size
- 1,
118 last
= key
.objectid
+ key
.offset
;
126 if (block_group
->key
.objectid
+
127 block_group
->key
.offset
> last
) {
128 hole_size
= block_group
->key
.objectid
+
129 block_group
->key
.offset
- last
;
130 set_extent_dirty(free_space_cache
, last
,
131 last
+ hole_size
- 1, GFP_NOFS
);
133 block_group
->cached
= 1;
135 btrfs_free_path(path
);
139 struct btrfs_block_group_cache
*btrfs_lookup_first_block_group(struct
143 struct extent_io_tree
*block_group_cache
;
144 struct btrfs_block_group_cache
*block_group
= NULL
;
150 bytenr
= max_t(u64
, bytenr
,
151 BTRFS_SUPER_INFO_OFFSET
+ BTRFS_SUPER_INFO_SIZE
);
152 block_group_cache
= &info
->block_group_cache
;
153 ret
= find_first_extent_bit(block_group_cache
,
154 bytenr
, &start
, &end
,
155 BLOCK_GROUP_DATA
| BLOCK_GROUP_METADATA
|
160 ret
= get_state_private(block_group_cache
, start
, &ptr
);
164 block_group
= (struct btrfs_block_group_cache
*)(unsigned long)ptr
;
168 struct btrfs_block_group_cache
*btrfs_lookup_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
;
194 if (block_group
->key
.objectid
<= bytenr
&& bytenr
<
195 block_group
->key
.objectid
+ block_group
->key
.offset
)
200 static int block_group_bits(struct btrfs_block_group_cache
*cache
, u64 bits
)
202 return (cache
->flags
& bits
) == bits
;
205 static int noinline
find_search_start(struct btrfs_root
*root
,
206 struct btrfs_block_group_cache
**cache_ret
,
207 u64
*start_ret
, u64 num
, int data
)
210 struct btrfs_block_group_cache
*cache
= *cache_ret
;
211 struct extent_io_tree
*free_space_cache
;
212 struct extent_state
*state
;
217 u64 search_start
= *start_ret
;
220 total_fs_bytes
= btrfs_super_total_bytes(&root
->fs_info
->super_copy
);
221 free_space_cache
= &root
->fs_info
->free_space_cache
;
227 ret
= cache_block_group(root
, cache
);
232 last
= max(search_start
, cache
->key
.objectid
);
233 if (!block_group_bits(cache
, data
) || cache
->ro
)
236 spin_lock_irq(&free_space_cache
->lock
);
237 state
= find_first_extent_bit_state(free_space_cache
, last
, EXTENT_DIRTY
);
242 spin_unlock_irq(&free_space_cache
->lock
);
246 start
= max(last
, state
->start
);
247 last
= state
->end
+ 1;
248 if (last
- start
< num
) {
250 state
= extent_state_next(state
);
251 } while(state
&& !(state
->state
& EXTENT_DIRTY
));
254 spin_unlock_irq(&free_space_cache
->lock
);
258 if (start
+ num
> cache
->key
.objectid
+ cache
->key
.offset
)
260 if (!block_group_bits(cache
, data
)) {
261 printk("block group bits don't match %Lu %d\n", cache
->flags
, data
);
267 cache
= btrfs_lookup_block_group(root
->fs_info
, search_start
);
269 printk("Unable to find block group for %Lu\n", search_start
);
275 last
= cache
->key
.objectid
+ cache
->key
.offset
;
277 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
278 if (!cache
|| cache
->key
.objectid
>= total_fs_bytes
) {
287 if (cache_miss
&& !cache
->cached
) {
288 cache_block_group(root
, cache
);
290 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
293 cache
= btrfs_find_block_group(root
, cache
, last
, data
, 0);
300 static u64
div_factor(u64 num
, int factor
)
309 static int block_group_state_bits(u64 flags
)
312 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
313 bits
|= BLOCK_GROUP_DATA
;
314 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
315 bits
|= BLOCK_GROUP_METADATA
;
316 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
317 bits
|= BLOCK_GROUP_SYSTEM
;
321 struct btrfs_block_group_cache
*btrfs_find_block_group(struct btrfs_root
*root
,
322 struct btrfs_block_group_cache
323 *hint
, u64 search_start
,
326 struct btrfs_block_group_cache
*cache
;
327 struct extent_io_tree
*block_group_cache
;
328 struct btrfs_block_group_cache
*found_group
= NULL
;
329 struct btrfs_fs_info
*info
= root
->fs_info
;
342 block_group_cache
= &info
->block_group_cache
;
344 if (data
& BTRFS_BLOCK_GROUP_METADATA
)
347 bit
= block_group_state_bits(data
);
350 struct btrfs_block_group_cache
*shint
;
351 shint
= btrfs_lookup_first_block_group(info
, search_start
);
352 if (shint
&& block_group_bits(shint
, data
) && !shint
->ro
) {
353 used
= btrfs_block_group_used(&shint
->item
);
354 if (used
+ shint
->pinned
<
355 div_factor(shint
->key
.offset
, factor
)) {
360 if (hint
&& !hint
->ro
&& block_group_bits(hint
, data
)) {
361 used
= btrfs_block_group_used(&hint
->item
);
362 if (used
+ hint
->pinned
<
363 div_factor(hint
->key
.offset
, factor
)) {
366 last
= hint
->key
.objectid
+ hint
->key
.offset
;
369 last
= max(hint
->key
.objectid
, search_start
);
375 ret
= find_first_extent_bit(block_group_cache
, last
,
380 ret
= get_state_private(block_group_cache
, start
, &ptr
);
386 cache
= (struct btrfs_block_group_cache
*)(unsigned long)ptr
;
387 last
= cache
->key
.objectid
+ cache
->key
.offset
;
388 used
= btrfs_block_group_used(&cache
->item
);
390 if (!cache
->ro
&& block_group_bits(cache
, data
)) {
391 free_check
= div_factor(cache
->key
.offset
, factor
);
392 if (used
+ cache
->pinned
< free_check
) {
404 if (!full_search
&& factor
< 10) {
414 static u64
hash_extent_ref(u64 root_objectid
, u64 ref_generation
,
415 u64 owner
, u64 owner_offset
)
417 u32 high_crc
= ~(u32
)0;
418 u32 low_crc
= ~(u32
)0;
420 lenum
= cpu_to_le64(root_objectid
);
421 high_crc
= btrfs_crc32c(high_crc
, &lenum
, sizeof(lenum
));
422 lenum
= cpu_to_le64(ref_generation
);
423 low_crc
= btrfs_crc32c(low_crc
, &lenum
, sizeof(lenum
));
424 if (owner
>= BTRFS_FIRST_FREE_OBJECTID
) {
425 lenum
= cpu_to_le64(owner
);
426 low_crc
= btrfs_crc32c(low_crc
, &lenum
, sizeof(lenum
));
427 lenum
= cpu_to_le64(owner_offset
);
428 low_crc
= btrfs_crc32c(low_crc
, &lenum
, sizeof(lenum
));
430 return ((u64
)high_crc
<< 32) | (u64
)low_crc
;
433 static int match_extent_ref(struct extent_buffer
*leaf
,
434 struct btrfs_extent_ref
*disk_ref
,
435 struct btrfs_extent_ref
*cpu_ref
)
440 if (cpu_ref
->objectid
)
441 len
= sizeof(*cpu_ref
);
443 len
= 2 * sizeof(u64
);
444 ret
= memcmp_extent_buffer(leaf
, cpu_ref
, (unsigned long)disk_ref
,
449 static int noinline
lookup_extent_backref(struct btrfs_trans_handle
*trans
,
450 struct btrfs_root
*root
,
451 struct btrfs_path
*path
, u64 bytenr
,
453 u64 ref_generation
, u64 owner
,
454 u64 owner_offset
, int del
)
457 struct btrfs_key key
;
458 struct btrfs_key found_key
;
459 struct btrfs_extent_ref ref
;
460 struct extent_buffer
*leaf
;
461 struct btrfs_extent_ref
*disk_ref
;
465 btrfs_set_stack_ref_root(&ref
, root_objectid
);
466 btrfs_set_stack_ref_generation(&ref
, ref_generation
);
467 btrfs_set_stack_ref_objectid(&ref
, owner
);
468 btrfs_set_stack_ref_offset(&ref
, owner_offset
);
470 hash
= hash_extent_ref(root_objectid
, ref_generation
, owner
,
473 key
.objectid
= bytenr
;
474 key
.type
= BTRFS_EXTENT_REF_KEY
;
477 ret
= btrfs_search_slot(trans
, root
, &key
, path
,
481 leaf
= path
->nodes
[0];
483 u32 nritems
= btrfs_header_nritems(leaf
);
484 if (path
->slots
[0] >= nritems
) {
485 ret2
= btrfs_next_leaf(root
, path
);
488 leaf
= path
->nodes
[0];
490 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
491 if (found_key
.objectid
!= bytenr
||
492 found_key
.type
!= BTRFS_EXTENT_REF_KEY
)
494 key
.offset
= found_key
.offset
;
496 btrfs_release_path(root
, path
);
500 disk_ref
= btrfs_item_ptr(path
->nodes
[0],
502 struct btrfs_extent_ref
);
503 if (match_extent_ref(path
->nodes
[0], disk_ref
, &ref
)) {
507 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
508 key
.offset
= found_key
.offset
+ 1;
509 btrfs_release_path(root
, path
);
516 * Back reference rules. Back refs have three main goals:
518 * 1) differentiate between all holders of references to an extent so that
519 * when a reference is dropped we can make sure it was a valid reference
520 * before freeing the extent.
522 * 2) Provide enough information to quickly find the holders of an extent
523 * if we notice a given block is corrupted or bad.
525 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
526 * maintenance. This is actually the same as #2, but with a slightly
527 * different use case.
529 * File extents can be referenced by:
531 * - multiple snapshots, subvolumes, or different generations in one subvol
532 * - different files inside a single subvolume (in theory, not implemented yet)
533 * - different offsets inside a file (bookend extents in file.c)
535 * The extent ref structure has fields for:
537 * - Objectid of the subvolume root
538 * - Generation number of the tree holding the reference
539 * - objectid of the file holding the reference
540 * - offset in the file corresponding to the key holding the reference
542 * When a file extent is allocated the fields are filled in:
543 * (root_key.objectid, trans->transid, inode objectid, offset in file)
545 * When a leaf is cow'd new references are added for every file extent found
546 * in the leaf. It looks the same as the create case, but trans->transid
547 * will be different when the block is cow'd.
549 * (root_key.objectid, trans->transid, inode objectid, offset in file)
551 * When a file extent is removed either during snapshot deletion or file
552 * truncation, the corresponding back reference is found
555 * (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
556 * inode objectid, offset in file)
558 * Btree extents can be referenced by:
560 * - Different subvolumes
561 * - Different generations of the same subvolume
563 * Storing sufficient information for a full reverse mapping of a btree
564 * block would require storing the lowest key of the block in the backref,
565 * and it would require updating that lowest key either before write out or
566 * every time it changed. Instead, the objectid of the lowest key is stored
567 * along with the level of the tree block. This provides a hint
568 * about where in the btree the block can be found. Searches through the
569 * btree only need to look for a pointer to that block, so they stop one
570 * level higher than the level recorded in the backref.
572 * Some btrees do not do reference counting on their extents. These
573 * include the extent tree and the tree of tree roots. Backrefs for these
574 * trees always have a generation of zero.
576 * When a tree block is created, back references are inserted:
578 * (root->root_key.objectid, trans->transid or zero, level, lowest_key_objectid)
580 * When a tree block is cow'd in a reference counted root,
581 * new back references are added for all the blocks it points to.
582 * These are of the form (trans->transid will have increased since creation):
584 * (root->root_key.objectid, trans->transid, level, lowest_key_objectid)
586 * Because the lowest_key_objectid and the level are just hints
587 * they are not used when backrefs are deleted. When a backref is deleted:
589 * if backref was for a tree root:
590 * root_objectid = root->root_key.objectid
592 * root_objectid = btrfs_header_owner(parent)
594 * (root_objectid, btrfs_header_generation(parent) or zero, 0, 0)
596 * Back Reference Key hashing:
598 * Back references have four fields, each 64 bits long. Unfortunately,
599 * This is hashed into a single 64 bit number and placed into the key offset.
600 * The key objectid corresponds to the first byte in the extent, and the
601 * key type is set to BTRFS_EXTENT_REF_KEY
603 int btrfs_insert_extent_backref(struct btrfs_trans_handle
*trans
,
604 struct btrfs_root
*root
,
605 struct btrfs_path
*path
, u64 bytenr
,
606 u64 root_objectid
, u64 ref_generation
,
607 u64 owner
, u64 owner_offset
)
610 struct btrfs_key key
;
611 struct btrfs_extent_ref ref
;
612 struct btrfs_extent_ref
*disk_ref
;
615 btrfs_set_stack_ref_root(&ref
, root_objectid
);
616 btrfs_set_stack_ref_generation(&ref
, ref_generation
);
617 btrfs_set_stack_ref_objectid(&ref
, owner
);
618 btrfs_set_stack_ref_offset(&ref
, owner_offset
);
620 hash
= hash_extent_ref(root_objectid
, ref_generation
, owner
,
623 key
.objectid
= bytenr
;
624 key
.type
= BTRFS_EXTENT_REF_KEY
;
626 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, sizeof(ref
));
627 while (ret
== -EEXIST
) {
628 disk_ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
629 struct btrfs_extent_ref
);
630 if (match_extent_ref(path
->nodes
[0], disk_ref
, &ref
))
633 btrfs_release_path(root
, path
);
634 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
639 disk_ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
640 struct btrfs_extent_ref
);
641 write_extent_buffer(path
->nodes
[0], &ref
, (unsigned long)disk_ref
,
643 btrfs_mark_buffer_dirty(path
->nodes
[0]);
645 btrfs_release_path(root
, path
);
649 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
650 struct btrfs_root
*root
,
651 u64 bytenr
, u64 num_bytes
,
652 u64 root_objectid
, u64 ref_generation
,
653 u64 owner
, u64 owner_offset
)
655 struct btrfs_path
*path
;
657 struct btrfs_key key
;
658 struct extent_buffer
*l
;
659 struct btrfs_extent_item
*item
;
662 WARN_ON(num_bytes
< root
->sectorsize
);
663 path
= btrfs_alloc_path();
668 key
.objectid
= bytenr
;
669 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
670 key
.offset
= num_bytes
;
671 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
, path
,
680 item
= btrfs_item_ptr(l
, path
->slots
[0], struct btrfs_extent_item
);
681 refs
= btrfs_extent_refs(l
, item
);
682 btrfs_set_extent_refs(l
, item
, refs
+ 1);
683 btrfs_mark_buffer_dirty(path
->nodes
[0]);
685 btrfs_release_path(root
->fs_info
->extent_root
, path
);
688 ret
= btrfs_insert_extent_backref(trans
, root
->fs_info
->extent_root
,
689 path
, bytenr
, root_objectid
,
690 ref_generation
, owner
, owner_offset
);
692 finish_current_insert(trans
, root
->fs_info
->extent_root
);
693 del_pending_extents(trans
, root
->fs_info
->extent_root
);
695 btrfs_free_path(path
);
699 int btrfs_extent_post_op(struct btrfs_trans_handle
*trans
,
700 struct btrfs_root
*root
)
702 finish_current_insert(trans
, root
->fs_info
->extent_root
);
703 del_pending_extents(trans
, root
->fs_info
->extent_root
);
707 static int lookup_extent_ref(struct btrfs_trans_handle
*trans
,
708 struct btrfs_root
*root
, u64 bytenr
,
709 u64 num_bytes
, u32
*refs
)
711 struct btrfs_path
*path
;
713 struct btrfs_key key
;
714 struct extent_buffer
*l
;
715 struct btrfs_extent_item
*item
;
717 WARN_ON(num_bytes
< root
->sectorsize
);
718 path
= btrfs_alloc_path();
720 key
.objectid
= bytenr
;
721 key
.offset
= num_bytes
;
722 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
723 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
, path
,
728 btrfs_print_leaf(root
, path
->nodes
[0]);
729 printk("failed to find block number %Lu\n", bytenr
);
733 item
= btrfs_item_ptr(l
, path
->slots
[0], struct btrfs_extent_item
);
734 *refs
= btrfs_extent_refs(l
, item
);
736 btrfs_free_path(path
);
740 u32
btrfs_count_snapshots_in_path(struct btrfs_root
*root
,
741 struct btrfs_path
*count_path
,
745 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
746 struct btrfs_path
*path
;
750 u64 root_objectid
= root
->root_key
.objectid
;
756 struct btrfs_key key
;
757 struct btrfs_key found_key
;
758 struct extent_buffer
*l
;
759 struct btrfs_extent_item
*item
;
760 struct btrfs_extent_ref
*ref_item
;
763 path
= btrfs_alloc_path();
766 bytenr
= first_extent
;
768 bytenr
= count_path
->nodes
[level
]->start
;
771 key
.objectid
= bytenr
;
774 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
775 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
781 btrfs_item_key_to_cpu(l
, &found_key
, path
->slots
[0]);
783 if (found_key
.objectid
!= bytenr
||
784 found_key
.type
!= BTRFS_EXTENT_ITEM_KEY
) {
788 item
= btrfs_item_ptr(l
, path
->slots
[0], struct btrfs_extent_item
);
789 extent_refs
= btrfs_extent_refs(l
, item
);
792 nritems
= btrfs_header_nritems(l
);
793 if (path
->slots
[0] >= nritems
) {
794 ret
= btrfs_next_leaf(extent_root
, path
);
799 btrfs_item_key_to_cpu(l
, &found_key
, path
->slots
[0]);
800 if (found_key
.objectid
!= bytenr
)
803 if (found_key
.type
!= BTRFS_EXTENT_REF_KEY
) {
809 ref_item
= btrfs_item_ptr(l
, path
->slots
[0],
810 struct btrfs_extent_ref
);
811 found_objectid
= btrfs_ref_root(l
, ref_item
);
813 if (found_objectid
!= root_objectid
) {
818 found_owner
= btrfs_ref_objectid(l
, ref_item
);
819 if (found_owner
!= expected_owner
) {
824 * nasty. we don't count a reference held by
825 * the running transaction. This allows nodatacow
826 * to avoid cow most of the time
828 if (found_owner
>= BTRFS_FIRST_FREE_OBJECTID
&&
829 btrfs_ref_generation(l
, ref_item
) ==
830 root
->fs_info
->generation
) {
838 * if there is more than one reference against a data extent,
839 * we have to assume the other ref is another snapshot
841 if (level
== -1 && extent_refs
> 1) {
845 if (cur_count
== 0) {
849 if (level
>= 0 && root
->node
== count_path
->nodes
[level
])
852 btrfs_release_path(root
, path
);
856 btrfs_free_path(path
);
859 int btrfs_inc_root_ref(struct btrfs_trans_handle
*trans
,
860 struct btrfs_root
*root
, u64 owner_objectid
)
866 struct btrfs_disk_key disk_key
;
868 level
= btrfs_header_level(root
->node
);
869 generation
= trans
->transid
;
870 nritems
= btrfs_header_nritems(root
->node
);
873 btrfs_item_key(root
->node
, &disk_key
, 0);
875 btrfs_node_key(root
->node
, &disk_key
, 0);
876 key_objectid
= btrfs_disk_key_objectid(&disk_key
);
880 return btrfs_inc_extent_ref(trans
, root
, root
->node
->start
,
881 root
->node
->len
, owner_objectid
,
882 generation
, level
, key_objectid
);
885 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
886 struct extent_buffer
*buf
)
890 struct btrfs_key key
;
891 struct btrfs_file_extent_item
*fi
;
900 level
= btrfs_header_level(buf
);
901 nritems
= btrfs_header_nritems(buf
);
902 for (i
= 0; i
< nritems
; i
++) {
905 btrfs_item_key_to_cpu(buf
, &key
, i
);
906 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
908 fi
= btrfs_item_ptr(buf
, i
,
909 struct btrfs_file_extent_item
);
910 if (btrfs_file_extent_type(buf
, fi
) ==
911 BTRFS_FILE_EXTENT_INLINE
)
913 disk_bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
914 if (disk_bytenr
== 0)
916 ret
= btrfs_inc_extent_ref(trans
, root
, disk_bytenr
,
917 btrfs_file_extent_disk_num_bytes(buf
, fi
),
918 root
->root_key
.objectid
, trans
->transid
,
919 key
.objectid
, key
.offset
);
925 bytenr
= btrfs_node_blockptr(buf
, i
);
926 btrfs_node_key_to_cpu(buf
, &key
, i
);
927 ret
= btrfs_inc_extent_ref(trans
, root
, bytenr
,
928 btrfs_level_size(root
, level
- 1),
929 root
->root_key
.objectid
,
931 level
- 1, key
.objectid
);
942 for (i
=0; i
< faili
; i
++) {
945 btrfs_item_key_to_cpu(buf
, &key
, i
);
946 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
948 fi
= btrfs_item_ptr(buf
, i
,
949 struct btrfs_file_extent_item
);
950 if (btrfs_file_extent_type(buf
, fi
) ==
951 BTRFS_FILE_EXTENT_INLINE
)
953 disk_bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
954 if (disk_bytenr
== 0)
956 err
= btrfs_free_extent(trans
, root
, disk_bytenr
,
957 btrfs_file_extent_disk_num_bytes(buf
,
961 bytenr
= btrfs_node_blockptr(buf
, i
);
962 err
= btrfs_free_extent(trans
, root
, bytenr
,
963 btrfs_level_size(root
, level
- 1), 0);
971 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
972 struct btrfs_root
*root
,
973 struct btrfs_path
*path
,
974 struct btrfs_block_group_cache
*cache
)
978 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
980 struct extent_buffer
*leaf
;
982 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
987 leaf
= path
->nodes
[0];
988 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
989 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
990 btrfs_mark_buffer_dirty(leaf
);
991 btrfs_release_path(extent_root
, path
);
993 finish_current_insert(trans
, extent_root
);
994 pending_ret
= del_pending_extents(trans
, extent_root
);
1003 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
1004 struct btrfs_root
*root
)
1006 struct extent_io_tree
*block_group_cache
;
1007 struct btrfs_block_group_cache
*cache
;
1011 struct btrfs_path
*path
;
1017 block_group_cache
= &root
->fs_info
->block_group_cache
;
1018 path
= btrfs_alloc_path();
1023 ret
= find_first_extent_bit(block_group_cache
, last
,
1024 &start
, &end
, BLOCK_GROUP_DIRTY
);
1029 ret
= get_state_private(block_group_cache
, start
, &ptr
);
1032 cache
= (struct btrfs_block_group_cache
*)(unsigned long)ptr
;
1033 err
= write_one_cache_group(trans
, root
,
1036 * if we fail to write the cache group, we want
1037 * to keep it marked dirty in hopes that a later
1044 clear_extent_bits(block_group_cache
, start
, end
,
1045 BLOCK_GROUP_DIRTY
, GFP_NOFS
);
1047 btrfs_free_path(path
);
1051 static struct btrfs_space_info
*__find_space_info(struct btrfs_fs_info
*info
,
1054 struct list_head
*head
= &info
->space_info
;
1055 struct list_head
*cur
;
1056 struct btrfs_space_info
*found
;
1057 list_for_each(cur
, head
) {
1058 found
= list_entry(cur
, struct btrfs_space_info
, list
);
1059 if (found
->flags
== flags
)
1066 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
1067 u64 total_bytes
, u64 bytes_used
,
1068 struct btrfs_space_info
**space_info
)
1070 struct btrfs_space_info
*found
;
1072 found
= __find_space_info(info
, flags
);
1074 found
->total_bytes
+= total_bytes
;
1075 found
->bytes_used
+= bytes_used
;
1077 WARN_ON(found
->total_bytes
< found
->bytes_used
);
1078 *space_info
= found
;
1081 found
= kmalloc(sizeof(*found
), GFP_NOFS
);
1085 list_add(&found
->list
, &info
->space_info
);
1086 found
->flags
= flags
;
1087 found
->total_bytes
= total_bytes
;
1088 found
->bytes_used
= bytes_used
;
1089 found
->bytes_pinned
= 0;
1091 found
->force_alloc
= 0;
1092 *space_info
= found
;
1096 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
1098 u64 extra_flags
= flags
& (BTRFS_BLOCK_GROUP_RAID0
|
1099 BTRFS_BLOCK_GROUP_RAID1
|
1100 BTRFS_BLOCK_GROUP_RAID10
|
1101 BTRFS_BLOCK_GROUP_DUP
);
1103 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
1104 fs_info
->avail_data_alloc_bits
|= extra_flags
;
1105 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
1106 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
1107 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
1108 fs_info
->avail_system_alloc_bits
|= extra_flags
;
1112 static u64
reduce_alloc_profile(struct btrfs_root
*root
, u64 flags
)
1114 u64 num_devices
= root
->fs_info
->fs_devices
->num_devices
;
1116 if (num_devices
== 1)
1117 flags
&= ~(BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID0
);
1118 if (num_devices
< 4)
1119 flags
&= ~BTRFS_BLOCK_GROUP_RAID10
;
1121 if ((flags
& BTRFS_BLOCK_GROUP_DUP
) &&
1122 (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
1123 BTRFS_BLOCK_GROUP_RAID10
))) {
1124 flags
&= ~BTRFS_BLOCK_GROUP_DUP
;
1127 if ((flags
& BTRFS_BLOCK_GROUP_RAID1
) &&
1128 (flags
& BTRFS_BLOCK_GROUP_RAID10
)) {
1129 flags
&= ~BTRFS_BLOCK_GROUP_RAID1
;
1132 if ((flags
& BTRFS_BLOCK_GROUP_RAID0
) &&
1133 ((flags
& BTRFS_BLOCK_GROUP_RAID1
) |
1134 (flags
& BTRFS_BLOCK_GROUP_RAID10
) |
1135 (flags
& BTRFS_BLOCK_GROUP_DUP
)))
1136 flags
&= ~BTRFS_BLOCK_GROUP_RAID0
;
1140 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
1141 struct btrfs_root
*extent_root
, u64 alloc_bytes
,
1142 u64 flags
, int force
)
1144 struct btrfs_space_info
*space_info
;
1150 flags
= reduce_alloc_profile(extent_root
, flags
);
1152 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
1154 ret
= update_space_info(extent_root
->fs_info
, flags
,
1158 BUG_ON(!space_info
);
1160 if (space_info
->force_alloc
) {
1162 space_info
->force_alloc
= 0;
1164 if (space_info
->full
)
1167 thresh
= div_factor(space_info
->total_bytes
, 6);
1169 (space_info
->bytes_used
+ space_info
->bytes_pinned
+ alloc_bytes
) <
1173 ret
= btrfs_alloc_chunk(trans
, extent_root
, &start
, &num_bytes
, flags
);
1174 if (ret
== -ENOSPC
) {
1175 printk("space info full %Lu\n", flags
);
1176 space_info
->full
= 1;
1181 ret
= btrfs_make_block_group(trans
, extent_root
, 0, flags
,
1182 BTRFS_FIRST_CHUNK_TREE_OBJECTID
, start
, num_bytes
);
1188 static int update_block_group(struct btrfs_trans_handle
*trans
,
1189 struct btrfs_root
*root
,
1190 u64 bytenr
, u64 num_bytes
, int alloc
,
1193 struct btrfs_block_group_cache
*cache
;
1194 struct btrfs_fs_info
*info
= root
->fs_info
;
1195 u64 total
= num_bytes
;
1202 cache
= btrfs_lookup_block_group(info
, bytenr
);
1206 byte_in_group
= bytenr
- cache
->key
.objectid
;
1207 WARN_ON(byte_in_group
> cache
->key
.offset
);
1208 start
= cache
->key
.objectid
;
1209 end
= start
+ cache
->key
.offset
- 1;
1210 set_extent_bits(&info
->block_group_cache
, start
, end
,
1211 BLOCK_GROUP_DIRTY
, GFP_NOFS
);
1213 old_val
= btrfs_block_group_used(&cache
->item
);
1214 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
1216 old_val
+= num_bytes
;
1217 cache
->space_info
->bytes_used
+= num_bytes
;
1219 old_val
-= num_bytes
;
1220 cache
->space_info
->bytes_used
-= num_bytes
;
1222 set_extent_dirty(&info
->free_space_cache
,
1223 bytenr
, bytenr
+ num_bytes
- 1,
1227 btrfs_set_block_group_used(&cache
->item
, old_val
);
1229 bytenr
+= num_bytes
;
1234 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
1239 ret
= find_first_extent_bit(&root
->fs_info
->block_group_cache
,
1240 search_start
, &start
, &end
,
1241 BLOCK_GROUP_DATA
| BLOCK_GROUP_METADATA
|
1242 BLOCK_GROUP_SYSTEM
);
1249 static int update_pinned_extents(struct btrfs_root
*root
,
1250 u64 bytenr
, u64 num
, int pin
)
1253 struct btrfs_block_group_cache
*cache
;
1254 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1257 set_extent_dirty(&fs_info
->pinned_extents
,
1258 bytenr
, bytenr
+ num
- 1, GFP_NOFS
);
1260 clear_extent_dirty(&fs_info
->pinned_extents
,
1261 bytenr
, bytenr
+ num
- 1, GFP_NOFS
);
1264 cache
= btrfs_lookup_block_group(fs_info
, bytenr
);
1266 u64 first
= first_logical_byte(root
, bytenr
);
1267 WARN_ON(first
< bytenr
);
1268 len
= min(first
- bytenr
, num
);
1270 len
= min(num
, cache
->key
.offset
-
1271 (bytenr
- cache
->key
.objectid
));
1275 cache
->pinned
+= len
;
1276 cache
->space_info
->bytes_pinned
+= len
;
1278 fs_info
->total_pinned
+= len
;
1281 cache
->pinned
-= len
;
1282 cache
->space_info
->bytes_pinned
-= len
;
1284 fs_info
->total_pinned
-= len
;
1292 int btrfs_copy_pinned(struct btrfs_root
*root
, struct extent_io_tree
*copy
)
1297 struct extent_io_tree
*pinned_extents
= &root
->fs_info
->pinned_extents
;
1301 ret
= find_first_extent_bit(pinned_extents
, last
,
1302 &start
, &end
, EXTENT_DIRTY
);
1305 set_extent_dirty(copy
, start
, end
, GFP_NOFS
);
1311 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
1312 struct btrfs_root
*root
,
1313 struct extent_io_tree
*unpin
)
1318 struct extent_io_tree
*free_space_cache
;
1319 free_space_cache
= &root
->fs_info
->free_space_cache
;
1322 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
1326 update_pinned_extents(root
, start
, end
+ 1 - start
, 0);
1327 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
1328 set_extent_dirty(free_space_cache
, start
, end
, GFP_NOFS
);
1333 static int finish_current_insert(struct btrfs_trans_handle
*trans
,
1334 struct btrfs_root
*extent_root
)
1338 struct btrfs_fs_info
*info
= extent_root
->fs_info
;
1339 struct extent_buffer
*eb
;
1340 struct btrfs_path
*path
;
1341 struct btrfs_key ins
;
1342 struct btrfs_disk_key first
;
1343 struct btrfs_extent_item extent_item
;
1348 btrfs_set_stack_extent_refs(&extent_item
, 1);
1349 btrfs_set_key_type(&ins
, BTRFS_EXTENT_ITEM_KEY
);
1350 path
= btrfs_alloc_path();
1353 ret
= find_first_extent_bit(&info
->extent_ins
, 0, &start
,
1354 &end
, EXTENT_LOCKED
);
1358 ins
.objectid
= start
;
1359 ins
.offset
= end
+ 1 - start
;
1360 err
= btrfs_insert_item(trans
, extent_root
, &ins
,
1361 &extent_item
, sizeof(extent_item
));
1362 clear_extent_bits(&info
->extent_ins
, start
, end
, EXTENT_LOCKED
,
1364 eb
= read_tree_block(extent_root
, ins
.objectid
, ins
.offset
,
1366 level
= btrfs_header_level(eb
);
1368 btrfs_item_key(eb
, &first
, 0);
1370 btrfs_node_key(eb
, &first
, 0);
1372 err
= btrfs_insert_extent_backref(trans
, extent_root
, path
,
1373 start
, extent_root
->root_key
.objectid
,
1375 btrfs_disk_key_objectid(&first
));
1377 free_extent_buffer(eb
);
1379 btrfs_free_path(path
);
1383 static int pin_down_bytes(struct btrfs_root
*root
, u64 bytenr
, u32 num_bytes
,
1387 struct extent_buffer
*buf
;
1390 buf
= btrfs_find_tree_block(root
, bytenr
, num_bytes
);
1392 if (btrfs_buffer_uptodate(buf
, 0)) {
1394 root
->fs_info
->running_transaction
->transid
;
1395 u64 header_transid
=
1396 btrfs_header_generation(buf
);
1397 if (header_transid
== transid
&&
1398 !btrfs_header_flag(buf
,
1399 BTRFS_HEADER_FLAG_WRITTEN
)) {
1400 clean_tree_block(NULL
, root
, buf
);
1401 free_extent_buffer(buf
);
1405 free_extent_buffer(buf
);
1407 update_pinned_extents(root
, bytenr
, num_bytes
, 1);
1409 set_extent_bits(&root
->fs_info
->pending_del
,
1410 bytenr
, bytenr
+ num_bytes
- 1,
1411 EXTENT_LOCKED
, GFP_NOFS
);
1418 * remove an extent from the root, returns 0 on success
1420 static int __free_extent(struct btrfs_trans_handle
*trans
, struct btrfs_root
1421 *root
, u64 bytenr
, u64 num_bytes
,
1422 u64 root_objectid
, u64 ref_generation
,
1423 u64 owner_objectid
, u64 owner_offset
, int pin
,
1426 struct btrfs_path
*path
;
1427 struct btrfs_key key
;
1428 struct btrfs_fs_info
*info
= root
->fs_info
;
1429 struct btrfs_root
*extent_root
= info
->extent_root
;
1430 struct extent_buffer
*leaf
;
1432 int extent_slot
= 0;
1433 int found_extent
= 0;
1435 struct btrfs_extent_item
*ei
;
1438 key
.objectid
= bytenr
;
1439 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
1440 key
.offset
= num_bytes
;
1441 path
= btrfs_alloc_path();
1446 ret
= lookup_extent_backref(trans
, extent_root
, path
,
1447 bytenr
, root_objectid
,
1449 owner_objectid
, owner_offset
, 1);
1451 struct btrfs_key found_key
;
1452 extent_slot
= path
->slots
[0];
1453 while(extent_slot
> 0) {
1455 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1457 if (found_key
.objectid
!= bytenr
)
1459 if (found_key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
1460 found_key
.offset
== num_bytes
) {
1464 if (path
->slots
[0] - extent_slot
> 5)
1468 ret
= btrfs_del_item(trans
, extent_root
, path
);
1470 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
1472 printk("Unable to find ref byte nr %Lu root %Lu "
1473 " gen %Lu owner %Lu offset %Lu\n", bytenr
,
1474 root_objectid
, ref_generation
, owner_objectid
,
1477 if (!found_extent
) {
1478 btrfs_release_path(extent_root
, path
);
1479 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
, -1, 1);
1483 extent_slot
= path
->slots
[0];
1486 leaf
= path
->nodes
[0];
1487 ei
= btrfs_item_ptr(leaf
, extent_slot
,
1488 struct btrfs_extent_item
);
1489 refs
= btrfs_extent_refs(leaf
, ei
);
1492 btrfs_set_extent_refs(leaf
, ei
, refs
);
1494 btrfs_mark_buffer_dirty(leaf
);
1496 if (refs
== 0 && found_extent
&& path
->slots
[0] == extent_slot
+ 1) {
1497 /* if the back ref and the extent are next to each other
1498 * they get deleted below in one shot
1500 path
->slots
[0] = extent_slot
;
1502 } else if (found_extent
) {
1503 /* otherwise delete the extent back ref */
1504 ret
= btrfs_del_item(trans
, extent_root
, path
);
1506 /* if refs are 0, we need to setup the path for deletion */
1508 btrfs_release_path(extent_root
, path
);
1509 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
1522 ret
= pin_down_bytes(root
, bytenr
, num_bytes
, 0);
1528 /* block accounting for super block */
1529 super_used
= btrfs_super_bytes_used(&info
->super_copy
);
1530 btrfs_set_super_bytes_used(&info
->super_copy
,
1531 super_used
- num_bytes
);
1533 /* block accounting for root item */
1534 root_used
= btrfs_root_used(&root
->root_item
);
1535 btrfs_set_root_used(&root
->root_item
,
1536 root_used
- num_bytes
);
1537 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
1542 ret
= update_block_group(trans
, root
, bytenr
, num_bytes
, 0,
1546 btrfs_free_path(path
);
1547 finish_current_insert(trans
, extent_root
);
1552 * find all the blocks marked as pending in the radix tree and remove
1553 * them from the extent map
1555 static int del_pending_extents(struct btrfs_trans_handle
*trans
, struct
1556 btrfs_root
*extent_root
)
1562 struct extent_io_tree
*pending_del
;
1563 struct extent_io_tree
*pinned_extents
;
1565 pending_del
= &extent_root
->fs_info
->pending_del
;
1566 pinned_extents
= &extent_root
->fs_info
->pinned_extents
;
1569 ret
= find_first_extent_bit(pending_del
, 0, &start
, &end
,
1573 update_pinned_extents(extent_root
, start
, end
+ 1 - start
, 1);
1574 clear_extent_bits(pending_del
, start
, end
, EXTENT_LOCKED
,
1576 ret
= __free_extent(trans
, extent_root
,
1577 start
, end
+ 1 - start
,
1578 extent_root
->root_key
.objectid
,
1587 * remove an extent from the root, returns 0 on success
1589 int btrfs_free_extent(struct btrfs_trans_handle
*trans
, struct btrfs_root
1590 *root
, u64 bytenr
, u64 num_bytes
,
1591 u64 root_objectid
, u64 ref_generation
,
1592 u64 owner_objectid
, u64 owner_offset
, int pin
)
1594 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
1598 WARN_ON(num_bytes
< root
->sectorsize
);
1599 if (!root
->ref_cows
)
1602 if (root
== extent_root
) {
1603 pin_down_bytes(root
, bytenr
, num_bytes
, 1);
1606 ret
= __free_extent(trans
, root
, bytenr
, num_bytes
, root_objectid
,
1607 ref_generation
, owner_objectid
, owner_offset
,
1609 pending_ret
= del_pending_extents(trans
, root
->fs_info
->extent_root
);
1610 return ret
? ret
: pending_ret
;
1613 static u64
stripe_align(struct btrfs_root
*root
, u64 val
)
1615 u64 mask
= ((u64
)root
->stripesize
- 1);
1616 u64 ret
= (val
+ mask
) & ~mask
;
1621 * walks the btree of allocated extents and find a hole of a given size.
1622 * The key ins is changed to record the hole:
1623 * ins->objectid == block start
1624 * ins->flags = BTRFS_EXTENT_ITEM_KEY
1625 * ins->offset == number of blocks
1626 * Any available blocks before search_start are skipped.
1628 static int noinline
find_free_extent(struct btrfs_trans_handle
*trans
,
1629 struct btrfs_root
*orig_root
,
1630 u64 num_bytes
, u64 empty_size
,
1631 u64 search_start
, u64 search_end
,
1632 u64 hint_byte
, struct btrfs_key
*ins
,
1633 u64 exclude_start
, u64 exclude_nr
,
1637 u64 orig_search_start
;
1638 struct btrfs_root
* root
= orig_root
->fs_info
->extent_root
;
1639 struct btrfs_fs_info
*info
= root
->fs_info
;
1640 u64 total_needed
= num_bytes
;
1641 u64
*last_ptr
= NULL
;
1642 struct btrfs_block_group_cache
*block_group
;
1645 int chunk_alloc_done
= 0;
1646 int empty_cluster
= 2 * 1024 * 1024;
1647 int allowed_chunk_alloc
= 0;
1649 WARN_ON(num_bytes
< root
->sectorsize
);
1650 btrfs_set_key_type(ins
, BTRFS_EXTENT_ITEM_KEY
);
1652 if (orig_root
->ref_cows
|| empty_size
)
1653 allowed_chunk_alloc
= 1;
1655 if (data
& BTRFS_BLOCK_GROUP_METADATA
) {
1656 last_ptr
= &root
->fs_info
->last_alloc
;
1657 empty_cluster
= 256 * 1024;
1660 if ((data
& BTRFS_BLOCK_GROUP_DATA
) && btrfs_test_opt(root
, SSD
)) {
1661 last_ptr
= &root
->fs_info
->last_data_alloc
;
1666 hint_byte
= *last_ptr
;
1668 empty_size
+= empty_cluster
;
1672 search_start
= max(search_start
, first_logical_byte(root
, 0));
1673 orig_search_start
= search_start
;
1675 if (search_end
== (u64
)-1)
1676 search_end
= btrfs_super_total_bytes(&info
->super_copy
);
1679 block_group
= btrfs_lookup_first_block_group(info
, hint_byte
);
1681 hint_byte
= search_start
;
1682 block_group
= btrfs_find_block_group(root
, block_group
,
1683 hint_byte
, data
, 1);
1684 if (last_ptr
&& *last_ptr
== 0 && block_group
)
1685 hint_byte
= block_group
->key
.objectid
;
1687 block_group
= btrfs_find_block_group(root
,
1689 search_start
, data
, 1);
1691 search_start
= max(search_start
, hint_byte
);
1693 total_needed
+= empty_size
;
1697 block_group
= btrfs_lookup_first_block_group(info
,
1700 block_group
= btrfs_lookup_first_block_group(info
,
1703 if (full_scan
&& !chunk_alloc_done
) {
1704 if (allowed_chunk_alloc
) {
1705 do_chunk_alloc(trans
, root
,
1706 num_bytes
+ 2 * 1024 * 1024, data
, 1);
1707 allowed_chunk_alloc
= 0;
1708 } else if (block_group
&& block_group_bits(block_group
, data
)) {
1709 block_group
->space_info
->force_alloc
= 1;
1711 chunk_alloc_done
= 1;
1713 ret
= find_search_start(root
, &block_group
, &search_start
,
1714 total_needed
, data
);
1715 if (ret
== -ENOSPC
&& last_ptr
&& *last_ptr
) {
1717 block_group
= btrfs_lookup_first_block_group(info
,
1719 search_start
= orig_search_start
;
1720 ret
= find_search_start(root
, &block_group
, &search_start
,
1721 total_needed
, data
);
1728 if (last_ptr
&& *last_ptr
&& search_start
!= *last_ptr
) {
1731 empty_size
+= empty_cluster
;
1732 total_needed
+= empty_size
;
1734 block_group
= btrfs_lookup_first_block_group(info
,
1736 search_start
= orig_search_start
;
1737 ret
= find_search_start(root
, &block_group
,
1738 &search_start
, total_needed
, data
);
1745 search_start
= stripe_align(root
, search_start
);
1746 ins
->objectid
= search_start
;
1747 ins
->offset
= num_bytes
;
1749 if (ins
->objectid
+ num_bytes
>= search_end
)
1752 if (ins
->objectid
+ num_bytes
>
1753 block_group
->key
.objectid
+ block_group
->key
.offset
) {
1754 search_start
= block_group
->key
.objectid
+
1755 block_group
->key
.offset
;
1759 if (test_range_bit(&info
->extent_ins
, ins
->objectid
,
1760 ins
->objectid
+ num_bytes
-1, EXTENT_LOCKED
, 0)) {
1761 search_start
= ins
->objectid
+ num_bytes
;
1765 if (test_range_bit(&info
->pinned_extents
, ins
->objectid
,
1766 ins
->objectid
+ num_bytes
-1, EXTENT_DIRTY
, 0)) {
1767 search_start
= ins
->objectid
+ num_bytes
;
1771 if (exclude_nr
> 0 && (ins
->objectid
+ num_bytes
> exclude_start
&&
1772 ins
->objectid
< exclude_start
+ exclude_nr
)) {
1773 search_start
= exclude_start
+ exclude_nr
;
1777 if (!(data
& BTRFS_BLOCK_GROUP_DATA
)) {
1778 block_group
= btrfs_lookup_block_group(info
, ins
->objectid
);
1780 trans
->block_group
= block_group
;
1782 ins
->offset
= num_bytes
;
1784 *last_ptr
= ins
->objectid
+ ins
->offset
;
1786 btrfs_super_total_bytes(&root
->fs_info
->super_copy
)) {
1793 if (search_start
+ num_bytes
>= search_end
) {
1795 search_start
= orig_search_start
;
1802 total_needed
-= empty_size
;
1807 block_group
= btrfs_lookup_first_block_group(info
, search_start
);
1809 block_group
= btrfs_find_block_group(root
, block_group
,
1810 search_start
, data
, 0);
1818 * finds a free extent and does all the dirty work required for allocation
1819 * returns the key for the extent through ins, and a tree buffer for
1820 * the first block of the extent through buf.
1822 * returns 0 if everything worked, non-zero otherwise.
1824 int btrfs_alloc_extent(struct btrfs_trans_handle
*trans
,
1825 struct btrfs_root
*root
,
1826 u64 num_bytes
, u64 min_alloc_size
,
1827 u64 root_objectid
, u64 ref_generation
,
1828 u64 owner
, u64 owner_offset
,
1829 u64 empty_size
, u64 hint_byte
,
1830 u64 search_end
, struct btrfs_key
*ins
, u64 data
)
1836 u64 search_start
= 0;
1839 struct btrfs_fs_info
*info
= root
->fs_info
;
1840 struct btrfs_root
*extent_root
= info
->extent_root
;
1841 struct btrfs_extent_item
*extent_item
;
1842 struct btrfs_extent_ref
*ref
;
1843 struct btrfs_path
*path
;
1844 struct btrfs_key keys
[2];
1847 alloc_profile
= info
->avail_data_alloc_bits
&
1848 info
->data_alloc_profile
;
1849 data
= BTRFS_BLOCK_GROUP_DATA
| alloc_profile
;
1850 } else if (root
== root
->fs_info
->chunk_root
) {
1851 alloc_profile
= info
->avail_system_alloc_bits
&
1852 info
->system_alloc_profile
;
1853 data
= BTRFS_BLOCK_GROUP_SYSTEM
| alloc_profile
;
1855 alloc_profile
= info
->avail_metadata_alloc_bits
&
1856 info
->metadata_alloc_profile
;
1857 data
= BTRFS_BLOCK_GROUP_METADATA
| alloc_profile
;
1860 data
= reduce_alloc_profile(root
, data
);
1862 * the only place that sets empty_size is btrfs_realloc_node, which
1863 * is not called recursively on allocations
1865 if (empty_size
|| root
->ref_cows
) {
1866 if (!(data
& BTRFS_BLOCK_GROUP_METADATA
)) {
1867 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
1869 BTRFS_BLOCK_GROUP_METADATA
|
1870 (info
->metadata_alloc_profile
&
1871 info
->avail_metadata_alloc_bits
), 0);
1874 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
1875 num_bytes
+ 2 * 1024 * 1024, data
, 0);
1879 WARN_ON(num_bytes
< root
->sectorsize
);
1880 ret
= find_free_extent(trans
, root
, num_bytes
, empty_size
,
1881 search_start
, search_end
, hint_byte
, ins
,
1882 trans
->alloc_exclude_start
,
1883 trans
->alloc_exclude_nr
, data
);
1885 if (ret
== -ENOSPC
&& num_bytes
> min_alloc_size
) {
1886 num_bytes
= num_bytes
>> 1;
1887 num_bytes
= max(num_bytes
, min_alloc_size
);
1888 do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
1889 num_bytes
, data
, 1);
1893 printk("allocation failed flags %Lu\n", data
);
1899 /* block accounting for super block */
1900 super_used
= btrfs_super_bytes_used(&info
->super_copy
);
1901 btrfs_set_super_bytes_used(&info
->super_copy
, super_used
+ num_bytes
);
1903 /* block accounting for root item */
1904 root_used
= btrfs_root_used(&root
->root_item
);
1905 btrfs_set_root_used(&root
->root_item
, root_used
+ num_bytes
);
1907 clear_extent_dirty(&root
->fs_info
->free_space_cache
,
1908 ins
->objectid
, ins
->objectid
+ ins
->offset
- 1,
1911 if (root
== extent_root
) {
1912 set_extent_bits(&root
->fs_info
->extent_ins
, ins
->objectid
,
1913 ins
->objectid
+ ins
->offset
- 1,
1914 EXTENT_LOCKED
, GFP_NOFS
);
1918 WARN_ON(trans
->alloc_exclude_nr
);
1919 trans
->alloc_exclude_start
= ins
->objectid
;
1920 trans
->alloc_exclude_nr
= ins
->offset
;
1922 memcpy(&keys
[0], ins
, sizeof(*ins
));
1923 keys
[1].offset
= hash_extent_ref(root_objectid
, ref_generation
,
1924 owner
, owner_offset
);
1925 keys
[1].objectid
= ins
->objectid
;
1926 keys
[1].type
= BTRFS_EXTENT_REF_KEY
;
1927 sizes
[0] = sizeof(*extent_item
);
1928 sizes
[1] = sizeof(*ref
);
1930 path
= btrfs_alloc_path();
1933 ret
= btrfs_insert_empty_items(trans
, extent_root
, path
, keys
,
1937 extent_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1938 struct btrfs_extent_item
);
1939 btrfs_set_extent_refs(path
->nodes
[0], extent_item
, 1);
1940 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
1941 struct btrfs_extent_ref
);
1943 btrfs_set_ref_root(path
->nodes
[0], ref
, root_objectid
);
1944 btrfs_set_ref_generation(path
->nodes
[0], ref
, ref_generation
);
1945 btrfs_set_ref_objectid(path
->nodes
[0], ref
, owner
);
1946 btrfs_set_ref_offset(path
->nodes
[0], ref
, owner_offset
);
1948 btrfs_mark_buffer_dirty(path
->nodes
[0]);
1950 trans
->alloc_exclude_start
= 0;
1951 trans
->alloc_exclude_nr
= 0;
1952 btrfs_free_path(path
);
1953 finish_current_insert(trans
, extent_root
);
1954 pending_ret
= del_pending_extents(trans
, extent_root
);
1964 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1, 0);
1966 printk("update block group failed for %Lu %Lu\n",
1967 ins
->objectid
, ins
->offset
);
1974 * helper function to allocate a block for a given tree
1975 * returns the tree buffer or NULL.
1977 struct extent_buffer
*btrfs_alloc_free_block(struct btrfs_trans_handle
*trans
,
1978 struct btrfs_root
*root
,
1980 u64 root_objectid
, u64 hint
,
1986 ref_generation
= trans
->transid
;
1991 return __btrfs_alloc_free_block(trans
, root
, blocksize
, root_objectid
,
1992 ref_generation
, 0, 0, hint
, empty_size
);
1996 * helper function to allocate a block for a given tree
1997 * returns the tree buffer or NULL.
1999 struct extent_buffer
*__btrfs_alloc_free_block(struct btrfs_trans_handle
*trans
,
2000 struct btrfs_root
*root
,
2009 struct btrfs_key ins
;
2011 struct extent_buffer
*buf
;
2013 ret
= btrfs_alloc_extent(trans
, root
, blocksize
, blocksize
,
2014 root_objectid
, ref_generation
,
2015 level
, first_objectid
, empty_size
, hint
,
2019 return ERR_PTR(ret
);
2021 buf
= btrfs_find_create_tree_block(root
, ins
.objectid
, blocksize
);
2023 btrfs_free_extent(trans
, root
, ins
.objectid
, blocksize
,
2024 root
->root_key
.objectid
, ref_generation
,
2026 return ERR_PTR(-ENOMEM
);
2028 btrfs_set_header_generation(buf
, trans
->transid
);
2029 clean_tree_block(trans
, root
, buf
);
2030 btrfs_set_buffer_uptodate(buf
);
2032 if (PageDirty(buf
->first_page
)) {
2033 printk("page %lu dirty\n", buf
->first_page
->index
);
2037 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
2038 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
2039 if (!btrfs_test_opt(root
, SSD
))
2040 btrfs_set_buffer_defrag(buf
);
2041 trans
->blocks_used
++;
2045 static int noinline
drop_leaf_ref(struct btrfs_trans_handle
*trans
,
2046 struct btrfs_root
*root
,
2047 struct extent_buffer
*leaf
)
2050 u64 leaf_generation
;
2051 struct btrfs_key key
;
2052 struct btrfs_file_extent_item
*fi
;
2057 BUG_ON(!btrfs_is_leaf(leaf
));
2058 nritems
= btrfs_header_nritems(leaf
);
2059 leaf_owner
= btrfs_header_owner(leaf
);
2060 leaf_generation
= btrfs_header_generation(leaf
);
2062 for (i
= 0; i
< nritems
; i
++) {
2065 btrfs_item_key_to_cpu(leaf
, &key
, i
);
2066 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
2068 fi
= btrfs_item_ptr(leaf
, i
, struct btrfs_file_extent_item
);
2069 if (btrfs_file_extent_type(leaf
, fi
) ==
2070 BTRFS_FILE_EXTENT_INLINE
)
2073 * FIXME make sure to insert a trans record that
2074 * repeats the snapshot del on crash
2076 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
2077 if (disk_bytenr
== 0)
2079 ret
= btrfs_free_extent(trans
, root
, disk_bytenr
,
2080 btrfs_file_extent_disk_num_bytes(leaf
, fi
),
2081 leaf_owner
, leaf_generation
,
2082 key
.objectid
, key
.offset
, 0);
2088 static void noinline
reada_walk_down(struct btrfs_root
*root
,
2089 struct extent_buffer
*node
,
2102 nritems
= btrfs_header_nritems(node
);
2103 level
= btrfs_header_level(node
);
2107 for (i
= slot
; i
< nritems
&& skipped
< 32; i
++) {
2108 bytenr
= btrfs_node_blockptr(node
, i
);
2109 if (last
&& ((bytenr
> last
&& bytenr
- last
> 32 * 1024) ||
2110 (last
> bytenr
&& last
- bytenr
> 32 * 1024))) {
2114 blocksize
= btrfs_level_size(root
, level
- 1);
2116 ret
= lookup_extent_ref(NULL
, root
, bytenr
,
2124 mutex_unlock(&root
->fs_info
->fs_mutex
);
2125 ret
= readahead_tree_block(root
, bytenr
, blocksize
,
2126 btrfs_node_ptr_generation(node
, i
));
2127 last
= bytenr
+ blocksize
;
2129 mutex_lock(&root
->fs_info
->fs_mutex
);
2136 * helper function for drop_snapshot, this walks down the tree dropping ref
2137 * counts as it goes.
2139 static int noinline
walk_down_tree(struct btrfs_trans_handle
*trans
,
2140 struct btrfs_root
*root
,
2141 struct btrfs_path
*path
, int *level
)
2147 struct extent_buffer
*next
;
2148 struct extent_buffer
*cur
;
2149 struct extent_buffer
*parent
;
2154 WARN_ON(*level
< 0);
2155 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
2156 ret
= lookup_extent_ref(trans
, root
,
2157 path
->nodes
[*level
]->start
,
2158 path
->nodes
[*level
]->len
, &refs
);
2164 * walk down to the last node level and free all the leaves
2166 while(*level
>= 0) {
2167 WARN_ON(*level
< 0);
2168 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
2169 cur
= path
->nodes
[*level
];
2171 if (btrfs_header_level(cur
) != *level
)
2174 if (path
->slots
[*level
] >=
2175 btrfs_header_nritems(cur
))
2178 ret
= drop_leaf_ref(trans
, root
, cur
);
2182 bytenr
= btrfs_node_blockptr(cur
, path
->slots
[*level
]);
2183 ptr_gen
= btrfs_node_ptr_generation(cur
, path
->slots
[*level
]);
2184 blocksize
= btrfs_level_size(root
, *level
- 1);
2185 ret
= lookup_extent_ref(trans
, root
, bytenr
, blocksize
, &refs
);
2188 parent
= path
->nodes
[*level
];
2189 root_owner
= btrfs_header_owner(parent
);
2190 root_gen
= btrfs_header_generation(parent
);
2191 path
->slots
[*level
]++;
2192 ret
= btrfs_free_extent(trans
, root
, bytenr
,
2193 blocksize
, root_owner
,
2198 next
= btrfs_find_tree_block(root
, bytenr
, blocksize
);
2199 if (!next
|| !btrfs_buffer_uptodate(next
, ptr_gen
)) {
2200 free_extent_buffer(next
);
2201 reada_walk_down(root
, cur
, path
->slots
[*level
]);
2203 mutex_unlock(&root
->fs_info
->fs_mutex
);
2204 next
= read_tree_block(root
, bytenr
, blocksize
,
2206 mutex_lock(&root
->fs_info
->fs_mutex
);
2208 /* we've dropped the lock, double check */
2209 ret
= lookup_extent_ref(trans
, root
, bytenr
,
2213 parent
= path
->nodes
[*level
];
2214 root_owner
= btrfs_header_owner(parent
);
2215 root_gen
= btrfs_header_generation(parent
);
2217 path
->slots
[*level
]++;
2218 free_extent_buffer(next
);
2219 ret
= btrfs_free_extent(trans
, root
, bytenr
,
2227 WARN_ON(*level
<= 0);
2228 if (path
->nodes
[*level
-1])
2229 free_extent_buffer(path
->nodes
[*level
-1]);
2230 path
->nodes
[*level
-1] = next
;
2231 *level
= btrfs_header_level(next
);
2232 path
->slots
[*level
] = 0;
2235 WARN_ON(*level
< 0);
2236 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
2238 if (path
->nodes
[*level
] == root
->node
) {
2239 root_owner
= root
->root_key
.objectid
;
2240 parent
= path
->nodes
[*level
];
2242 parent
= path
->nodes
[*level
+ 1];
2243 root_owner
= btrfs_header_owner(parent
);
2246 root_gen
= btrfs_header_generation(parent
);
2247 ret
= btrfs_free_extent(trans
, root
, path
->nodes
[*level
]->start
,
2248 path
->nodes
[*level
]->len
,
2249 root_owner
, root_gen
, 0, 0, 1);
2250 free_extent_buffer(path
->nodes
[*level
]);
2251 path
->nodes
[*level
] = NULL
;
2258 * helper for dropping snapshots. This walks back up the tree in the path
2259 * to find the first node higher up where we haven't yet gone through
2262 static int noinline
walk_up_tree(struct btrfs_trans_handle
*trans
,
2263 struct btrfs_root
*root
,
2264 struct btrfs_path
*path
, int *level
)
2268 struct btrfs_root_item
*root_item
= &root
->root_item
;
2273 for(i
= *level
; i
< BTRFS_MAX_LEVEL
- 1 && path
->nodes
[i
]; i
++) {
2274 slot
= path
->slots
[i
];
2275 if (slot
< btrfs_header_nritems(path
->nodes
[i
]) - 1) {
2276 struct extent_buffer
*node
;
2277 struct btrfs_disk_key disk_key
;
2278 node
= path
->nodes
[i
];
2281 WARN_ON(*level
== 0);
2282 btrfs_node_key(node
, &disk_key
, path
->slots
[i
]);
2283 memcpy(&root_item
->drop_progress
,
2284 &disk_key
, sizeof(disk_key
));
2285 root_item
->drop_level
= i
;
2288 if (path
->nodes
[*level
] == root
->node
) {
2289 root_owner
= root
->root_key
.objectid
;
2291 btrfs_header_generation(path
->nodes
[*level
]);
2293 struct extent_buffer
*node
;
2294 node
= path
->nodes
[*level
+ 1];
2295 root_owner
= btrfs_header_owner(node
);
2296 root_gen
= btrfs_header_generation(node
);
2298 ret
= btrfs_free_extent(trans
, root
,
2299 path
->nodes
[*level
]->start
,
2300 path
->nodes
[*level
]->len
,
2301 root_owner
, root_gen
, 0, 0, 1);
2303 free_extent_buffer(path
->nodes
[*level
]);
2304 path
->nodes
[*level
] = NULL
;
2312 * drop the reference count on the tree rooted at 'snap'. This traverses
2313 * the tree freeing any blocks that have a ref count of zero after being
2316 int btrfs_drop_snapshot(struct btrfs_trans_handle
*trans
, struct btrfs_root
2322 struct btrfs_path
*path
;
2325 struct btrfs_root_item
*root_item
= &root
->root_item
;
2327 path
= btrfs_alloc_path();
2330 level
= btrfs_header_level(root
->node
);
2332 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
2333 path
->nodes
[level
] = root
->node
;
2334 extent_buffer_get(root
->node
);
2335 path
->slots
[level
] = 0;
2337 struct btrfs_key key
;
2338 struct btrfs_disk_key found_key
;
2339 struct extent_buffer
*node
;
2341 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
2342 level
= root_item
->drop_level
;
2343 path
->lowest_level
= level
;
2344 wret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2349 node
= path
->nodes
[level
];
2350 btrfs_node_key(node
, &found_key
, path
->slots
[level
]);
2351 WARN_ON(memcmp(&found_key
, &root_item
->drop_progress
,
2352 sizeof(found_key
)));
2355 wret
= walk_down_tree(trans
, root
, path
, &level
);
2361 wret
= walk_up_tree(trans
, root
, path
, &level
);
2369 for (i
= 0; i
<= orig_level
; i
++) {
2370 if (path
->nodes
[i
]) {
2371 free_extent_buffer(path
->nodes
[i
]);
2372 path
->nodes
[i
] = NULL
;
2376 btrfs_free_path(path
);
2380 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
2387 ret
= find_first_extent_bit(&info
->block_group_cache
, 0,
2388 &start
, &end
, (unsigned int)-1);
2391 ret
= get_state_private(&info
->block_group_cache
, start
, &ptr
);
2393 kfree((void *)(unsigned long)ptr
);
2394 clear_extent_bits(&info
->block_group_cache
, start
,
2395 end
, (unsigned int)-1, GFP_NOFS
);
2398 ret
= find_first_extent_bit(&info
->free_space_cache
, 0,
2399 &start
, &end
, EXTENT_DIRTY
);
2402 clear_extent_dirty(&info
->free_space_cache
, start
,
2408 static unsigned long calc_ra(unsigned long start
, unsigned long last
,
2411 return min(last
, start
+ nr
- 1);
2414 static int noinline
relocate_inode_pages(struct inode
*inode
, u64 start
,
2419 unsigned long last_index
;
2422 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2423 struct file_ra_state
*ra
;
2424 unsigned long total_read
= 0;
2425 unsigned long ra_pages
;
2426 struct btrfs_trans_handle
*trans
;
2428 ra
= kzalloc(sizeof(*ra
), GFP_NOFS
);
2430 mutex_lock(&inode
->i_mutex
);
2431 i
= start
>> PAGE_CACHE_SHIFT
;
2432 last_index
= (start
+ len
- 1) >> PAGE_CACHE_SHIFT
;
2434 ra_pages
= BTRFS_I(inode
)->root
->fs_info
->bdi
.ra_pages
;
2436 file_ra_state_init(ra
, inode
->i_mapping
);
2438 for (; i
<= last_index
; i
++) {
2439 if (total_read
% ra_pages
== 0) {
2440 btrfs_force_ra(inode
->i_mapping
, ra
, NULL
, i
,
2441 calc_ra(i
, last_index
, ra_pages
));
2444 if (((u64
)i
<< PAGE_CACHE_SHIFT
) > inode
->i_size
)
2445 goto truncate_racing
;
2447 page
= grab_cache_page(inode
->i_mapping
, i
);
2451 if (!PageUptodate(page
)) {
2452 btrfs_readpage(NULL
, page
);
2454 if (!PageUptodate(page
)) {
2456 page_cache_release(page
);
2460 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
2461 ClearPageDirty(page
);
2463 cancel_dirty_page(page
, PAGE_CACHE_SIZE
);
2465 wait_on_page_writeback(page
);
2466 set_page_extent_mapped(page
);
2467 page_start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2468 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
2470 lock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
2472 set_extent_delalloc(io_tree
, page_start
,
2473 page_end
, GFP_NOFS
);
2474 set_page_dirty(page
);
2476 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
2478 page_cache_release(page
);
2480 balance_dirty_pages_ratelimited_nr(inode
->i_mapping
,
2485 trans
= btrfs_start_transaction(BTRFS_I(inode
)->root
, 1);
2487 btrfs_add_ordered_inode(inode
);
2488 btrfs_end_transaction(trans
, BTRFS_I(inode
)->root
);
2489 mark_inode_dirty(inode
);
2491 mutex_unlock(&inode
->i_mutex
);
2495 vmtruncate(inode
, inode
->i_size
);
2496 balance_dirty_pages_ratelimited_nr(inode
->i_mapping
,
2502 * The back references tell us which tree holds a ref on a block,
2503 * but it is possible for the tree root field in the reference to
2504 * reflect the original root before a snapshot was made. In this
2505 * case we should search through all the children of a given root
2506 * to find potential holders of references on a block.
2508 * Instead, we do something a little less fancy and just search
2509 * all the roots for a given key/block combination.
2511 static int find_root_for_ref(struct btrfs_root
*root
,
2512 struct btrfs_path
*path
,
2513 struct btrfs_key
*key0
,
2516 struct btrfs_root
**found_root
,
2519 struct btrfs_key root_location
;
2520 struct btrfs_root
*cur_root
= *found_root
;
2521 struct btrfs_file_extent_item
*file_extent
;
2522 u64 root_search_start
= BTRFS_FS_TREE_OBJECTID
;
2527 root_location
.offset
= (u64
)-1;
2528 root_location
.type
= BTRFS_ROOT_ITEM_KEY
;
2529 path
->lowest_level
= level
;
2532 ret
= btrfs_search_slot(NULL
, cur_root
, key0
, path
, 0, 0);
2534 if (ret
== 0 && file_key
) {
2535 struct extent_buffer
*leaf
= path
->nodes
[0];
2536 file_extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
2537 struct btrfs_file_extent_item
);
2538 if (btrfs_file_extent_type(leaf
, file_extent
) ==
2539 BTRFS_FILE_EXTENT_REG
) {
2541 btrfs_file_extent_disk_bytenr(leaf
,
2544 } else if (!file_key
) {
2545 if (path
->nodes
[level
])
2546 found_bytenr
= path
->nodes
[level
]->start
;
2549 for (i
= level
; i
< BTRFS_MAX_LEVEL
; i
++) {
2550 if (!path
->nodes
[i
])
2552 free_extent_buffer(path
->nodes
[i
]);
2553 path
->nodes
[i
] = NULL
;
2555 btrfs_release_path(cur_root
, path
);
2557 if (found_bytenr
== bytenr
) {
2558 *found_root
= cur_root
;
2562 ret
= btrfs_search_root(root
->fs_info
->tree_root
,
2563 root_search_start
, &root_search_start
);
2567 root_location
.objectid
= root_search_start
;
2568 cur_root
= btrfs_read_fs_root_no_name(root
->fs_info
,
2576 path
->lowest_level
= 0;
2581 * note, this releases the path
2583 static int noinline
relocate_one_reference(struct btrfs_root
*extent_root
,
2584 struct btrfs_path
*path
,
2585 struct btrfs_key
*extent_key
,
2586 u64
*last_file_objectid
,
2587 u64
*last_file_offset
,
2588 u64
*last_file_root
,
2591 struct inode
*inode
;
2592 struct btrfs_root
*found_root
;
2593 struct btrfs_key root_location
;
2594 struct btrfs_key found_key
;
2595 struct btrfs_extent_ref
*ref
;
2603 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2604 struct btrfs_extent_ref
);
2605 ref_root
= btrfs_ref_root(path
->nodes
[0], ref
);
2606 ref_gen
= btrfs_ref_generation(path
->nodes
[0], ref
);
2607 ref_objectid
= btrfs_ref_objectid(path
->nodes
[0], ref
);
2608 ref_offset
= btrfs_ref_offset(path
->nodes
[0], ref
);
2609 btrfs_release_path(extent_root
, path
);
2611 root_location
.objectid
= ref_root
;
2613 root_location
.offset
= 0;
2615 root_location
.offset
= (u64
)-1;
2616 root_location
.type
= BTRFS_ROOT_ITEM_KEY
;
2618 found_root
= btrfs_read_fs_root_no_name(extent_root
->fs_info
,
2620 BUG_ON(!found_root
);
2622 if (ref_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
2623 found_key
.objectid
= ref_objectid
;
2624 found_key
.type
= BTRFS_EXTENT_DATA_KEY
;
2625 found_key
.offset
= ref_offset
;
2628 if (last_extent
== extent_key
->objectid
&&
2629 *last_file_objectid
== ref_objectid
&&
2630 *last_file_offset
== ref_offset
&&
2631 *last_file_root
== ref_root
)
2634 ret
= find_root_for_ref(extent_root
, path
, &found_key
,
2635 level
, 1, &found_root
,
2636 extent_key
->objectid
);
2641 if (last_extent
== extent_key
->objectid
&&
2642 *last_file_objectid
== ref_objectid
&&
2643 *last_file_offset
== ref_offset
&&
2644 *last_file_root
== ref_root
)
2647 mutex_unlock(&extent_root
->fs_info
->fs_mutex
);
2648 inode
= btrfs_iget_locked(extent_root
->fs_info
->sb
,
2649 ref_objectid
, found_root
);
2650 if (inode
->i_state
& I_NEW
) {
2651 /* the inode and parent dir are two different roots */
2652 BTRFS_I(inode
)->root
= found_root
;
2653 BTRFS_I(inode
)->location
.objectid
= ref_objectid
;
2654 BTRFS_I(inode
)->location
.type
= BTRFS_INODE_ITEM_KEY
;
2655 BTRFS_I(inode
)->location
.offset
= 0;
2656 btrfs_read_locked_inode(inode
);
2657 unlock_new_inode(inode
);
2660 /* this can happen if the reference is not against
2661 * the latest version of the tree root
2663 if (is_bad_inode(inode
)) {
2664 mutex_lock(&extent_root
->fs_info
->fs_mutex
);
2667 *last_file_objectid
= inode
->i_ino
;
2668 *last_file_root
= found_root
->root_key
.objectid
;
2669 *last_file_offset
= ref_offset
;
2671 relocate_inode_pages(inode
, ref_offset
, extent_key
->offset
);
2673 mutex_lock(&extent_root
->fs_info
->fs_mutex
);
2675 struct btrfs_trans_handle
*trans
;
2676 struct extent_buffer
*eb
;
2679 eb
= read_tree_block(found_root
, extent_key
->objectid
,
2680 extent_key
->offset
, 0);
2681 level
= btrfs_header_level(eb
);
2684 btrfs_item_key_to_cpu(eb
, &found_key
, 0);
2686 btrfs_node_key_to_cpu(eb
, &found_key
, 0);
2688 free_extent_buffer(eb
);
2690 ret
= find_root_for_ref(extent_root
, path
, &found_key
,
2691 level
, 0, &found_root
,
2692 extent_key
->objectid
);
2697 trans
= btrfs_start_transaction(found_root
, 1);
2699 path
->lowest_level
= level
;
2701 ret
= btrfs_search_slot(trans
, found_root
, &found_key
, path
,
2703 path
->lowest_level
= 0;
2704 for (i
= level
; i
< BTRFS_MAX_LEVEL
; i
++) {
2705 if (!path
->nodes
[i
])
2707 free_extent_buffer(path
->nodes
[i
]);
2708 path
->nodes
[i
] = NULL
;
2710 btrfs_release_path(found_root
, path
);
2711 if (found_root
== found_root
->fs_info
->extent_root
)
2712 btrfs_extent_post_op(trans
, found_root
);
2713 btrfs_end_transaction(trans
, found_root
);
2720 static int noinline
del_extent_zero(struct btrfs_root
*extent_root
,
2721 struct btrfs_path
*path
,
2722 struct btrfs_key
*extent_key
)
2725 struct btrfs_trans_handle
*trans
;
2727 trans
= btrfs_start_transaction(extent_root
, 1);
2728 ret
= btrfs_search_slot(trans
, extent_root
, extent_key
, path
, -1, 1);
2735 ret
= btrfs_del_item(trans
, extent_root
, path
);
2737 btrfs_end_transaction(trans
, extent_root
);
2741 static int noinline
relocate_one_extent(struct btrfs_root
*extent_root
,
2742 struct btrfs_path
*path
,
2743 struct btrfs_key
*extent_key
)
2745 struct btrfs_key key
;
2746 struct btrfs_key found_key
;
2747 struct extent_buffer
*leaf
;
2748 u64 last_file_objectid
= 0;
2749 u64 last_file_root
= 0;
2750 u64 last_file_offset
= (u64
)-1;
2751 u64 last_extent
= 0;
2756 if (extent_key
->objectid
== 0) {
2757 ret
= del_extent_zero(extent_root
, path
, extent_key
);
2760 key
.objectid
= extent_key
->objectid
;
2761 key
.type
= BTRFS_EXTENT_REF_KEY
;
2765 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
2771 leaf
= path
->nodes
[0];
2772 nritems
= btrfs_header_nritems(leaf
);
2773 if (path
->slots
[0] == nritems
) {
2774 ret
= btrfs_next_leaf(extent_root
, path
);
2781 leaf
= path
->nodes
[0];
2784 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2785 if (found_key
.objectid
!= extent_key
->objectid
) {
2789 if (found_key
.type
!= BTRFS_EXTENT_REF_KEY
) {
2793 key
.offset
= found_key
.offset
+ 1;
2794 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2796 ret
= relocate_one_reference(extent_root
, path
, extent_key
,
2797 &last_file_objectid
,
2799 &last_file_root
, last_extent
);
2802 last_extent
= extent_key
->objectid
;
2806 btrfs_release_path(extent_root
, path
);
2810 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
2813 u64 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
2814 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
2816 num_devices
= root
->fs_info
->fs_devices
->num_devices
;
2817 if (num_devices
== 1) {
2818 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
2819 stripped
= flags
& ~stripped
;
2821 /* turn raid0 into single device chunks */
2822 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
2825 /* turn mirroring into duplication */
2826 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
2827 BTRFS_BLOCK_GROUP_RAID10
))
2828 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
2831 /* they already had raid on here, just return */
2832 if (flags
& stripped
)
2835 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
2836 stripped
= flags
& ~stripped
;
2838 /* switch duplicated blocks with raid1 */
2839 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
2840 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
2842 /* turn single device chunks into raid0 */
2843 return stripped
| BTRFS_BLOCK_GROUP_RAID0
;
2848 int __alloc_chunk_for_shrink(struct btrfs_root
*root
,
2849 struct btrfs_block_group_cache
*shrink_block_group
,
2852 struct btrfs_trans_handle
*trans
;
2853 u64 new_alloc_flags
;
2856 if (btrfs_block_group_used(&shrink_block_group
->item
) > 0) {
2858 trans
= btrfs_start_transaction(root
, 1);
2859 new_alloc_flags
= update_block_group_flags(root
,
2860 shrink_block_group
->flags
);
2861 if (new_alloc_flags
!= shrink_block_group
->flags
) {
2863 btrfs_block_group_used(&shrink_block_group
->item
);
2865 calc
= shrink_block_group
->key
.offset
;
2867 do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
2868 calc
+ 2 * 1024 * 1024, new_alloc_flags
, force
);
2869 btrfs_end_transaction(trans
, root
);
2874 int btrfs_shrink_extent_tree(struct btrfs_root
*root
, u64 shrink_start
)
2876 struct btrfs_trans_handle
*trans
;
2877 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
2878 struct btrfs_path
*path
;
2881 u64 shrink_last_byte
;
2882 struct btrfs_block_group_cache
*shrink_block_group
;
2883 struct btrfs_fs_info
*info
= root
->fs_info
;
2884 struct btrfs_key key
;
2885 struct btrfs_key found_key
;
2886 struct extent_buffer
*leaf
;
2891 shrink_block_group
= btrfs_lookup_block_group(root
->fs_info
,
2893 BUG_ON(!shrink_block_group
);
2895 shrink_last_byte
= shrink_block_group
->key
.objectid
+
2896 shrink_block_group
->key
.offset
;
2898 shrink_block_group
->space_info
->total_bytes
-=
2899 shrink_block_group
->key
.offset
;
2900 path
= btrfs_alloc_path();
2901 root
= root
->fs_info
->extent_root
;
2904 printk("btrfs relocating block group %llu flags %llu\n",
2905 (unsigned long long)shrink_start
,
2906 (unsigned long long)shrink_block_group
->flags
);
2908 __alloc_chunk_for_shrink(root
, shrink_block_group
, 1);
2912 shrink_block_group
->ro
= 1;
2916 key
.objectid
= shrink_start
;
2919 cur_byte
= key
.objectid
;
2921 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2925 ret
= btrfs_previous_item(root
, path
, 0, BTRFS_EXTENT_ITEM_KEY
);
2930 leaf
= path
->nodes
[0];
2931 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2932 if (found_key
.objectid
+ found_key
.offset
> shrink_start
&&
2933 found_key
.objectid
< shrink_last_byte
) {
2934 cur_byte
= found_key
.objectid
;
2935 key
.objectid
= cur_byte
;
2938 btrfs_release_path(root
, path
);
2941 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2945 leaf
= path
->nodes
[0];
2946 nritems
= btrfs_header_nritems(leaf
);
2948 if (path
->slots
[0] >= nritems
) {
2949 ret
= btrfs_next_leaf(root
, path
);
2956 leaf
= path
->nodes
[0];
2957 nritems
= btrfs_header_nritems(leaf
);
2960 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2962 if (found_key
.objectid
>= shrink_last_byte
)
2965 if (progress
&& need_resched()) {
2966 memcpy(&key
, &found_key
, sizeof(key
));
2967 mutex_unlock(&root
->fs_info
->fs_mutex
);
2969 mutex_lock(&root
->fs_info
->fs_mutex
);
2970 btrfs_release_path(root
, path
);
2971 btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2977 if (btrfs_key_type(&found_key
) != BTRFS_EXTENT_ITEM_KEY
||
2978 found_key
.objectid
+ found_key
.offset
<= cur_byte
) {
2979 memcpy(&key
, &found_key
, sizeof(key
));
2986 cur_byte
= found_key
.objectid
+ found_key
.offset
;
2987 key
.objectid
= cur_byte
;
2988 btrfs_release_path(root
, path
);
2989 ret
= relocate_one_extent(root
, path
, &found_key
);
2990 __alloc_chunk_for_shrink(root
, shrink_block_group
, 0);
2993 btrfs_release_path(root
, path
);
2995 if (total_found
> 0) {
2996 printk("btrfs relocate found %llu last extent was %llu\n",
2997 (unsigned long long)total_found
,
2998 (unsigned long long)found_key
.objectid
);
2999 trans
= btrfs_start_transaction(tree_root
, 1);
3000 btrfs_commit_transaction(trans
, tree_root
);
3002 mutex_unlock(&root
->fs_info
->fs_mutex
);
3003 btrfs_clean_old_snapshots(tree_root
);
3004 mutex_lock(&root
->fs_info
->fs_mutex
);
3006 trans
= btrfs_start_transaction(tree_root
, 1);
3007 btrfs_commit_transaction(trans
, tree_root
);
3012 * we've freed all the extents, now remove the block
3013 * group item from the tree
3015 trans
= btrfs_start_transaction(root
, 1);
3016 memcpy(&key
, &shrink_block_group
->key
, sizeof(key
));
3018 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
3024 clear_extent_bits(&info
->block_group_cache
, key
.objectid
,
3025 key
.objectid
+ key
.offset
- 1,
3026 (unsigned int)-1, GFP_NOFS
);
3029 clear_extent_bits(&info
->free_space_cache
,
3030 key
.objectid
, key
.objectid
+ key
.offset
- 1,
3031 (unsigned int)-1, GFP_NOFS
);
3033 memset(shrink_block_group
, 0, sizeof(*shrink_block_group
));
3034 kfree(shrink_block_group
);
3036 btrfs_del_item(trans
, root
, path
);
3037 btrfs_commit_transaction(trans
, root
);
3039 /* the code to unpin extents might set a few bits in the free
3040 * space cache for this range again
3042 clear_extent_bits(&info
->free_space_cache
,
3043 key
.objectid
, key
.objectid
+ key
.offset
- 1,
3044 (unsigned int)-1, GFP_NOFS
);
3046 btrfs_free_path(path
);
3050 int find_first_block_group(struct btrfs_root
*root
, struct btrfs_path
*path
,
3051 struct btrfs_key
*key
)
3054 struct btrfs_key found_key
;
3055 struct extent_buffer
*leaf
;
3058 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
3062 slot
= path
->slots
[0];
3063 leaf
= path
->nodes
[0];
3064 if (slot
>= btrfs_header_nritems(leaf
)) {
3065 ret
= btrfs_next_leaf(root
, path
);
3072 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3074 if (found_key
.objectid
>= key
->objectid
&&
3075 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
)
3084 int btrfs_read_block_groups(struct btrfs_root
*root
)
3086 struct btrfs_path
*path
;
3089 struct btrfs_block_group_cache
*cache
;
3090 struct btrfs_fs_info
*info
= root
->fs_info
;
3091 struct btrfs_space_info
*space_info
;
3092 struct extent_io_tree
*block_group_cache
;
3093 struct btrfs_key key
;
3094 struct btrfs_key found_key
;
3095 struct extent_buffer
*leaf
;
3097 block_group_cache
= &info
->block_group_cache
;
3098 root
= info
->extent_root
;
3101 btrfs_set_key_type(&key
, BTRFS_BLOCK_GROUP_ITEM_KEY
);
3102 path
= btrfs_alloc_path();
3107 ret
= find_first_block_group(root
, path
, &key
);
3115 leaf
= path
->nodes
[0];
3116 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3117 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
3123 read_extent_buffer(leaf
, &cache
->item
,
3124 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
3125 sizeof(cache
->item
));
3126 memcpy(&cache
->key
, &found_key
, sizeof(found_key
));
3128 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
3129 btrfs_release_path(root
, path
);
3130 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
3132 if (cache
->flags
& BTRFS_BLOCK_GROUP_DATA
) {
3133 bit
= BLOCK_GROUP_DATA
;
3134 } else if (cache
->flags
& BTRFS_BLOCK_GROUP_SYSTEM
) {
3135 bit
= BLOCK_GROUP_SYSTEM
;
3136 } else if (cache
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
3137 bit
= BLOCK_GROUP_METADATA
;
3139 set_avail_alloc_bits(info
, cache
->flags
);
3141 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
3142 btrfs_block_group_used(&cache
->item
),
3145 cache
->space_info
= space_info
;
3147 /* use EXTENT_LOCKED to prevent merging */
3148 set_extent_bits(block_group_cache
, found_key
.objectid
,
3149 found_key
.objectid
+ found_key
.offset
- 1,
3150 bit
| EXTENT_LOCKED
, GFP_NOFS
);
3151 set_state_private(block_group_cache
, found_key
.objectid
,
3152 (unsigned long)cache
);
3155 btrfs_super_total_bytes(&info
->super_copy
))
3160 btrfs_free_path(path
);
3164 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
3165 struct btrfs_root
*root
, u64 bytes_used
,
3166 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
3171 struct btrfs_root
*extent_root
;
3172 struct btrfs_block_group_cache
*cache
;
3173 struct extent_io_tree
*block_group_cache
;
3175 extent_root
= root
->fs_info
->extent_root
;
3176 block_group_cache
= &root
->fs_info
->block_group_cache
;
3178 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
3180 cache
->key
.objectid
= chunk_offset
;
3181 cache
->key
.offset
= size
;
3182 btrfs_set_key_type(&cache
->key
, BTRFS_BLOCK_GROUP_ITEM_KEY
);
3184 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
3185 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
3186 cache
->flags
= type
;
3187 btrfs_set_block_group_flags(&cache
->item
, type
);
3189 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
3190 &cache
->space_info
);
3193 bit
= block_group_state_bits(type
);
3194 set_extent_bits(block_group_cache
, chunk_offset
,
3195 chunk_offset
+ size
- 1,
3196 bit
| EXTENT_LOCKED
, GFP_NOFS
);
3198 set_state_private(block_group_cache
, chunk_offset
,
3199 (unsigned long)cache
);
3200 ret
= btrfs_insert_item(trans
, extent_root
, &cache
->key
, &cache
->item
,
3201 sizeof(cache
->item
));
3204 finish_current_insert(trans
, extent_root
);
3205 ret
= del_pending_extents(trans
, extent_root
);
3207 set_avail_alloc_bits(extent_root
->fs_info
, type
);