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>
24 #include "print-tree.h"
25 #include "transaction.h"
28 #define BLOCK_GROUP_DATA EXTENT_WRITEBACK
29 #define BLOCK_GROUP_METADATA EXTENT_UPTODATE
30 #define BLOCK_GROUP_SYSTEM EXTENT_NEW
32 #define BLOCK_GROUP_DIRTY EXTENT_DIRTY
34 static int finish_current_insert(struct btrfs_trans_handle
*trans
, struct
35 btrfs_root
*extent_root
);
36 static int del_pending_extents(struct btrfs_trans_handle
*trans
, struct
37 btrfs_root
*extent_root
);
40 static int cache_block_group(struct btrfs_root
*root
,
41 struct btrfs_block_group_cache
*block_group
)
43 struct btrfs_path
*path
;
46 struct extent_buffer
*leaf
;
47 struct extent_io_tree
*free_space_cache
;
57 root
= root
->fs_info
->extent_root
;
58 free_space_cache
= &root
->fs_info
->free_space_cache
;
60 if (block_group
->cached
)
63 path
= btrfs_alloc_path();
68 first_free
= block_group
->key
.objectid
;
69 key
.objectid
= block_group
->key
.objectid
;
71 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
72 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
75 ret
= btrfs_previous_item(root
, path
, 0, BTRFS_EXTENT_ITEM_KEY
);
79 leaf
= path
->nodes
[0];
80 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
81 if (key
.objectid
+ key
.offset
> first_free
)
82 first_free
= key
.objectid
+ key
.offset
;
85 leaf
= path
->nodes
[0];
86 slot
= path
->slots
[0];
87 if (slot
>= btrfs_header_nritems(leaf
)) {
88 ret
= btrfs_next_leaf(root
, path
);
97 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
98 if (key
.objectid
< block_group
->key
.objectid
) {
101 if (key
.objectid
>= block_group
->key
.objectid
+
102 block_group
->key
.offset
) {
106 if (btrfs_key_type(&key
) == BTRFS_EXTENT_ITEM_KEY
) {
111 if (key
.objectid
> last
) {
112 hole_size
= key
.objectid
- last
;
113 set_extent_dirty(free_space_cache
, last
,
114 last
+ hole_size
- 1,
117 last
= key
.objectid
+ key
.offset
;
125 if (block_group
->key
.objectid
+
126 block_group
->key
.offset
> last
) {
127 hole_size
= block_group
->key
.objectid
+
128 block_group
->key
.offset
- last
;
129 set_extent_dirty(free_space_cache
, last
,
130 last
+ hole_size
- 1, GFP_NOFS
);
132 block_group
->cached
= 1;
134 btrfs_free_path(path
);
138 struct btrfs_block_group_cache
*btrfs_lookup_block_group(struct
142 struct extent_io_tree
*block_group_cache
;
143 struct btrfs_block_group_cache
*block_group
= NULL
;
149 block_group_cache
= &info
->block_group_cache
;
150 ret
= find_first_extent_bit(block_group_cache
,
151 bytenr
, &start
, &end
,
152 BLOCK_GROUP_DATA
| BLOCK_GROUP_METADATA
|
157 ret
= get_state_private(block_group_cache
, start
, &ptr
);
161 block_group
= (struct btrfs_block_group_cache
*)(unsigned long)ptr
;
162 if (block_group
->key
.objectid
<= bytenr
&& bytenr
<
163 block_group
->key
.objectid
+ block_group
->key
.offset
)
168 static int block_group_bits(struct btrfs_block_group_cache
*cache
, u64 bits
)
170 return (cache
->flags
& bits
) == bits
;
173 static int noinline
find_search_start(struct btrfs_root
*root
,
174 struct btrfs_block_group_cache
**cache_ret
,
175 u64
*start_ret
, int num
, int data
)
178 struct btrfs_block_group_cache
*cache
= *cache_ret
;
179 struct extent_io_tree
*free_space_cache
;
180 struct extent_state
*state
;
185 u64 search_start
= *start_ret
;
191 total_fs_bytes
= btrfs_super_total_bytes(&root
->fs_info
->super_copy
);
192 free_space_cache
= &root
->fs_info
->free_space_cache
;
195 ret
= cache_block_group(root
, cache
);
199 last
= max(search_start
, cache
->key
.objectid
);
200 if (!block_group_bits(cache
, data
) || cache
->ro
) {
204 spin_lock_irq(&free_space_cache
->lock
);
205 state
= find_first_extent_bit_state(free_space_cache
, last
, EXTENT_DIRTY
);
210 spin_unlock_irq(&free_space_cache
->lock
);
214 start
= max(last
, state
->start
);
215 last
= state
->end
+ 1;
216 if (last
- start
< num
) {
217 if (last
== cache
->key
.objectid
+ cache
->key
.offset
)
220 state
= extent_state_next(state
);
221 } while(state
&& !(state
->state
& EXTENT_DIRTY
));
224 spin_unlock_irq(&free_space_cache
->lock
);
227 if (start
+ num
> cache
->key
.objectid
+ cache
->key
.offset
)
229 if (start
+ num
> total_fs_bytes
)
231 if (!block_group_bits(cache
, data
)) {
232 printk("block group bits don't match %Lu %d\n", cache
->flags
, data
);
238 cache
= btrfs_lookup_block_group(root
->fs_info
, search_start
);
240 printk("Unable to find block group for %Lu\n", search_start
);
246 last
= cache
->key
.objectid
+ cache
->key
.offset
;
248 cache
= btrfs_lookup_block_group(root
->fs_info
, last
);
249 if (!cache
|| cache
->key
.objectid
>= total_fs_bytes
) {
258 if (cache_miss
&& !cache
->cached
) {
259 cache_block_group(root
, cache
);
261 cache
= btrfs_lookup_block_group(root
->fs_info
, last
);
263 cache
= btrfs_find_block_group(root
, cache
, last
, data
, 0);
271 static u64
div_factor(u64 num
, int factor
)
280 static int block_group_state_bits(u64 flags
)
283 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
284 bits
|= BLOCK_GROUP_DATA
;
285 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
286 bits
|= BLOCK_GROUP_METADATA
;
287 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
288 bits
|= BLOCK_GROUP_SYSTEM
;
292 struct btrfs_block_group_cache
*btrfs_find_block_group(struct btrfs_root
*root
,
293 struct btrfs_block_group_cache
294 *hint
, u64 search_start
,
297 struct btrfs_block_group_cache
*cache
;
298 struct extent_io_tree
*block_group_cache
;
299 struct btrfs_block_group_cache
*found_group
= NULL
;
300 struct btrfs_fs_info
*info
= root
->fs_info
;
314 block_group_cache
= &info
->block_group_cache
;
315 total_fs_bytes
= btrfs_super_total_bytes(&root
->fs_info
->super_copy
);
320 bit
= block_group_state_bits(data
);
322 if (search_start
&& search_start
< total_fs_bytes
) {
323 struct btrfs_block_group_cache
*shint
;
324 shint
= btrfs_lookup_block_group(info
, search_start
);
325 if (shint
&& block_group_bits(shint
, data
) && !shint
->ro
) {
326 used
= btrfs_block_group_used(&shint
->item
);
327 if (used
+ shint
->pinned
<
328 div_factor(shint
->key
.offset
, factor
)) {
333 if (hint
&& !hint
->ro
&& block_group_bits(hint
, data
) &&
334 hint
->key
.objectid
< total_fs_bytes
) {
335 used
= btrfs_block_group_used(&hint
->item
);
336 if (used
+ hint
->pinned
<
337 div_factor(hint
->key
.offset
, factor
)) {
340 last
= hint
->key
.objectid
+ hint
->key
.offset
;
344 hint_last
= max(hint
->key
.objectid
, search_start
);
346 hint_last
= search_start
;
348 if (hint_last
>= total_fs_bytes
)
349 hint_last
= search_start
;
354 ret
= find_first_extent_bit(block_group_cache
, last
,
359 ret
= get_state_private(block_group_cache
, start
, &ptr
);
363 cache
= (struct btrfs_block_group_cache
*)(unsigned long)ptr
;
364 last
= cache
->key
.objectid
+ cache
->key
.offset
;
365 used
= btrfs_block_group_used(&cache
->item
);
367 if (cache
->key
.objectid
> total_fs_bytes
)
370 if (!cache
->ro
&& block_group_bits(cache
, data
)) {
372 free_check
= cache
->key
.offset
;
374 free_check
= div_factor(cache
->key
.offset
,
377 if (used
+ cache
->pinned
< free_check
) {
393 static u64
hash_extent_ref(u64 root_objectid
, u64 ref_generation
,
394 u64 owner
, u64 owner_offset
)
396 u32 high_crc
= ~(u32
)0;
397 u32 low_crc
= ~(u32
)0;
399 lenum
= cpu_to_le64(root_objectid
);
400 high_crc
= btrfs_crc32c(high_crc
, &lenum
, sizeof(lenum
));
401 lenum
= cpu_to_le64(ref_generation
);
402 low_crc
= btrfs_crc32c(low_crc
, &lenum
, sizeof(lenum
));
403 if (owner
>= BTRFS_FIRST_FREE_OBJECTID
) {
404 lenum
= cpu_to_le64(owner
);
405 low_crc
= btrfs_crc32c(low_crc
, &lenum
, sizeof(lenum
));
406 lenum
= cpu_to_le64(owner_offset
);
407 low_crc
= btrfs_crc32c(low_crc
, &lenum
, sizeof(lenum
));
409 return ((u64
)high_crc
<< 32) | (u64
)low_crc
;
412 static int match_extent_ref(struct extent_buffer
*leaf
,
413 struct btrfs_extent_ref
*disk_ref
,
414 struct btrfs_extent_ref
*cpu_ref
)
419 if (cpu_ref
->objectid
)
420 len
= sizeof(*cpu_ref
);
422 len
= 2 * sizeof(u64
);
423 ret
= memcmp_extent_buffer(leaf
, cpu_ref
, (unsigned long)disk_ref
,
428 static int noinline
lookup_extent_backref(struct btrfs_trans_handle
*trans
,
429 struct btrfs_root
*root
,
430 struct btrfs_path
*path
, u64 bytenr
,
432 u64 ref_generation
, u64 owner
,
433 u64 owner_offset
, int del
)
436 struct btrfs_key key
;
437 struct btrfs_key found_key
;
438 struct btrfs_extent_ref ref
;
439 struct extent_buffer
*leaf
;
440 struct btrfs_extent_ref
*disk_ref
;
444 btrfs_set_stack_ref_root(&ref
, root_objectid
);
445 btrfs_set_stack_ref_generation(&ref
, ref_generation
);
446 btrfs_set_stack_ref_objectid(&ref
, owner
);
447 btrfs_set_stack_ref_offset(&ref
, owner_offset
);
449 hash
= hash_extent_ref(root_objectid
, ref_generation
, owner
,
452 key
.objectid
= bytenr
;
453 key
.type
= BTRFS_EXTENT_REF_KEY
;
456 ret
= btrfs_search_slot(trans
, root
, &key
, path
,
460 leaf
= path
->nodes
[0];
462 u32 nritems
= btrfs_header_nritems(leaf
);
463 if (path
->slots
[0] >= nritems
) {
464 ret2
= btrfs_next_leaf(root
, path
);
467 leaf
= path
->nodes
[0];
469 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
470 if (found_key
.objectid
!= bytenr
||
471 found_key
.type
!= BTRFS_EXTENT_REF_KEY
)
473 key
.offset
= found_key
.offset
;
475 btrfs_release_path(root
, path
);
479 disk_ref
= btrfs_item_ptr(path
->nodes
[0],
481 struct btrfs_extent_ref
);
482 if (match_extent_ref(path
->nodes
[0], disk_ref
, &ref
)) {
486 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
487 key
.offset
= found_key
.offset
+ 1;
488 btrfs_release_path(root
, path
);
495 * Back reference rules. Back refs have three main goals:
497 * 1) differentiate between all holders of references to an extent so that
498 * when a reference is dropped we can make sure it was a valid reference
499 * before freeing the extent.
501 * 2) Provide enough information to quickly find the holders of an extent
502 * if we notice a given block is corrupted or bad.
504 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
505 * maintenance. This is actually the same as #2, but with a slightly
506 * different use case.
508 * File extents can be referenced by:
510 * - multiple snapshots, subvolumes, or different generations in one subvol
511 * - different files inside a single subvolume (in theory, not implemented yet)
512 * - different offsets inside a file (bookend extents in file.c)
514 * The extent ref structure has fields for:
516 * - Objectid of the subvolume root
517 * - Generation number of the tree holding the reference
518 * - objectid of the file holding the reference
519 * - offset in the file corresponding to the key holding the reference
521 * When a file extent is allocated the fields are filled in:
522 * (root_key.objectid, trans->transid, inode objectid, offset in file)
524 * When a leaf is cow'd new references are added for every file extent found
525 * in the leaf. It looks the same as the create case, but trans->transid
526 * will be different when the block is cow'd.
528 * (root_key.objectid, trans->transid, inode objectid, offset in file)
530 * When a file extent is removed either during snapshot deletion or file
531 * truncation, the corresponding back reference is found
534 * (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
535 * inode objectid, offset in file)
537 * Btree extents can be referenced by:
539 * - Different subvolumes
540 * - Different generations of the same subvolume
542 * Storing sufficient information for a full reverse mapping of a btree
543 * block would require storing the lowest key of the block in the backref,
544 * and it would require updating that lowest key either before write out or
545 * every time it changed. Instead, the objectid of the lowest key is stored
546 * along with the level of the tree block. This provides a hint
547 * about where in the btree the block can be found. Searches through the
548 * btree only need to look for a pointer to that block, so they stop one
549 * level higher than the level recorded in the backref.
551 * Some btrees do not do reference counting on their extents. These
552 * include the extent tree and the tree of tree roots. Backrefs for these
553 * trees always have a generation of zero.
555 * When a tree block is created, back references are inserted:
557 * (root->root_key.objectid, trans->transid or zero, level, lowest_key_objectid)
559 * When a tree block is cow'd in a reference counted root,
560 * new back references are added for all the blocks it points to.
561 * These are of the form (trans->transid will have increased since creation):
563 * (root->root_key.objectid, trans->transid, level, lowest_key_objectid)
565 * Because the lowest_key_objectid and the level are just hints
566 * they are not used when backrefs are deleted. When a backref is deleted:
568 * if backref was for a tree root:
569 * root_objectid = root->root_key.objectid
571 * root_objectid = btrfs_header_owner(parent)
573 * (root_objectid, btrfs_header_generation(parent) or zero, 0, 0)
575 * Back Reference Key hashing:
577 * Back references have four fields, each 64 bits long. Unfortunately,
578 * This is hashed into a single 64 bit number and placed into the key offset.
579 * The key objectid corresponds to the first byte in the extent, and the
580 * key type is set to BTRFS_EXTENT_REF_KEY
582 int btrfs_insert_extent_backref(struct btrfs_trans_handle
*trans
,
583 struct btrfs_root
*root
,
584 struct btrfs_path
*path
, u64 bytenr
,
585 u64 root_objectid
, u64 ref_generation
,
586 u64 owner
, u64 owner_offset
)
589 struct btrfs_key key
;
590 struct btrfs_extent_ref ref
;
591 struct btrfs_extent_ref
*disk_ref
;
594 btrfs_set_stack_ref_root(&ref
, root_objectid
);
595 btrfs_set_stack_ref_generation(&ref
, ref_generation
);
596 btrfs_set_stack_ref_objectid(&ref
, owner
);
597 btrfs_set_stack_ref_offset(&ref
, owner_offset
);
599 hash
= hash_extent_ref(root_objectid
, ref_generation
, owner
,
602 key
.objectid
= bytenr
;
603 key
.type
= BTRFS_EXTENT_REF_KEY
;
605 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, sizeof(ref
));
606 while (ret
== -EEXIST
) {
607 disk_ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
608 struct btrfs_extent_ref
);
609 if (match_extent_ref(path
->nodes
[0], disk_ref
, &ref
))
612 btrfs_release_path(root
, path
);
613 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
618 disk_ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
619 struct btrfs_extent_ref
);
620 write_extent_buffer(path
->nodes
[0], &ref
, (unsigned long)disk_ref
,
622 btrfs_mark_buffer_dirty(path
->nodes
[0]);
624 btrfs_release_path(root
, path
);
628 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
629 struct btrfs_root
*root
,
630 u64 bytenr
, u64 num_bytes
,
631 u64 root_objectid
, u64 ref_generation
,
632 u64 owner
, u64 owner_offset
)
634 struct btrfs_path
*path
;
636 struct btrfs_key key
;
637 struct extent_buffer
*l
;
638 struct btrfs_extent_item
*item
;
641 WARN_ON(num_bytes
< root
->sectorsize
);
642 path
= btrfs_alloc_path();
647 key
.objectid
= bytenr
;
648 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
649 key
.offset
= num_bytes
;
650 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
, path
,
659 item
= btrfs_item_ptr(l
, path
->slots
[0], struct btrfs_extent_item
);
660 refs
= btrfs_extent_refs(l
, item
);
661 btrfs_set_extent_refs(l
, item
, refs
+ 1);
662 btrfs_mark_buffer_dirty(path
->nodes
[0]);
664 btrfs_release_path(root
->fs_info
->extent_root
, path
);
667 ret
= btrfs_insert_extent_backref(trans
, root
->fs_info
->extent_root
,
668 path
, bytenr
, root_objectid
,
669 ref_generation
, owner
, owner_offset
);
671 finish_current_insert(trans
, root
->fs_info
->extent_root
);
672 del_pending_extents(trans
, root
->fs_info
->extent_root
);
674 btrfs_free_path(path
);
678 int btrfs_extent_post_op(struct btrfs_trans_handle
*trans
,
679 struct btrfs_root
*root
)
681 finish_current_insert(trans
, root
->fs_info
->extent_root
);
682 del_pending_extents(trans
, root
->fs_info
->extent_root
);
686 static int lookup_extent_ref(struct btrfs_trans_handle
*trans
,
687 struct btrfs_root
*root
, u64 bytenr
,
688 u64 num_bytes
, u32
*refs
)
690 struct btrfs_path
*path
;
692 struct btrfs_key key
;
693 struct extent_buffer
*l
;
694 struct btrfs_extent_item
*item
;
696 WARN_ON(num_bytes
< root
->sectorsize
);
697 path
= btrfs_alloc_path();
699 key
.objectid
= bytenr
;
700 key
.offset
= num_bytes
;
701 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
702 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
, path
,
707 btrfs_print_leaf(root
, path
->nodes
[0]);
708 printk("failed to find block number %Lu\n", bytenr
);
712 item
= btrfs_item_ptr(l
, path
->slots
[0], struct btrfs_extent_item
);
713 *refs
= btrfs_extent_refs(l
, item
);
715 btrfs_free_path(path
);
719 u32
btrfs_count_snapshots_in_path(struct btrfs_root
*root
,
720 struct btrfs_path
*count_path
,
723 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
724 struct btrfs_path
*path
;
727 u64 root_objectid
= root
->root_key
.objectid
;
732 struct btrfs_key key
;
733 struct btrfs_key found_key
;
734 struct extent_buffer
*l
;
735 struct btrfs_extent_item
*item
;
736 struct btrfs_extent_ref
*ref_item
;
739 path
= btrfs_alloc_path();
742 bytenr
= first_extent
;
744 bytenr
= count_path
->nodes
[level
]->start
;
747 key
.objectid
= bytenr
;
750 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
751 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
757 btrfs_item_key_to_cpu(l
, &found_key
, path
->slots
[0]);
759 if (found_key
.objectid
!= bytenr
||
760 found_key
.type
!= BTRFS_EXTENT_ITEM_KEY
) {
764 item
= btrfs_item_ptr(l
, path
->slots
[0], struct btrfs_extent_item
);
767 nritems
= btrfs_header_nritems(l
);
768 if (path
->slots
[0] >= nritems
) {
769 ret
= btrfs_next_leaf(extent_root
, path
);
774 btrfs_item_key_to_cpu(l
, &found_key
, path
->slots
[0]);
775 if (found_key
.objectid
!= bytenr
)
778 if (found_key
.type
!= BTRFS_EXTENT_REF_KEY
) {
784 ref_item
= btrfs_item_ptr(l
, path
->slots
[0],
785 struct btrfs_extent_ref
);
786 found_objectid
= btrfs_ref_root(l
, ref_item
);
788 if (found_objectid
!= root_objectid
) {
795 if (cur_count
== 0) {
799 if (level
>= 0 && root
->node
== count_path
->nodes
[level
])
802 btrfs_release_path(root
, path
);
806 btrfs_free_path(path
);
809 int btrfs_inc_root_ref(struct btrfs_trans_handle
*trans
,
810 struct btrfs_root
*root
, u64 owner_objectid
)
816 struct btrfs_disk_key disk_key
;
818 level
= btrfs_header_level(root
->node
);
819 generation
= trans
->transid
;
820 nritems
= btrfs_header_nritems(root
->node
);
823 btrfs_item_key(root
->node
, &disk_key
, 0);
825 btrfs_node_key(root
->node
, &disk_key
, 0);
826 key_objectid
= btrfs_disk_key_objectid(&disk_key
);
830 return btrfs_inc_extent_ref(trans
, root
, root
->node
->start
,
831 root
->node
->len
, owner_objectid
,
832 generation
, level
, key_objectid
);
835 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
836 struct extent_buffer
*buf
)
840 struct btrfs_key key
;
841 struct btrfs_file_extent_item
*fi
;
850 level
= btrfs_header_level(buf
);
851 nritems
= btrfs_header_nritems(buf
);
852 for (i
= 0; i
< nritems
; i
++) {
855 btrfs_item_key_to_cpu(buf
, &key
, i
);
856 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
858 fi
= btrfs_item_ptr(buf
, i
,
859 struct btrfs_file_extent_item
);
860 if (btrfs_file_extent_type(buf
, fi
) ==
861 BTRFS_FILE_EXTENT_INLINE
)
863 disk_bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
864 if (disk_bytenr
== 0)
866 ret
= btrfs_inc_extent_ref(trans
, root
, disk_bytenr
,
867 btrfs_file_extent_disk_num_bytes(buf
, fi
),
868 root
->root_key
.objectid
, trans
->transid
,
869 key
.objectid
, key
.offset
);
875 bytenr
= btrfs_node_blockptr(buf
, i
);
876 btrfs_node_key_to_cpu(buf
, &key
, i
);
877 ret
= btrfs_inc_extent_ref(trans
, root
, bytenr
,
878 btrfs_level_size(root
, level
- 1),
879 root
->root_key
.objectid
,
881 level
- 1, key
.objectid
);
892 for (i
=0; i
< faili
; i
++) {
895 btrfs_item_key_to_cpu(buf
, &key
, i
);
896 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
898 fi
= btrfs_item_ptr(buf
, i
,
899 struct btrfs_file_extent_item
);
900 if (btrfs_file_extent_type(buf
, fi
) ==
901 BTRFS_FILE_EXTENT_INLINE
)
903 disk_bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
904 if (disk_bytenr
== 0)
906 err
= btrfs_free_extent(trans
, root
, disk_bytenr
,
907 btrfs_file_extent_disk_num_bytes(buf
,
911 bytenr
= btrfs_node_blockptr(buf
, i
);
912 err
= btrfs_free_extent(trans
, root
, bytenr
,
913 btrfs_level_size(root
, level
- 1), 0);
921 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
922 struct btrfs_root
*root
,
923 struct btrfs_path
*path
,
924 struct btrfs_block_group_cache
*cache
)
928 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
930 struct extent_buffer
*leaf
;
932 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
937 leaf
= path
->nodes
[0];
938 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
939 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
940 btrfs_mark_buffer_dirty(leaf
);
941 btrfs_release_path(extent_root
, path
);
943 finish_current_insert(trans
, extent_root
);
944 pending_ret
= del_pending_extents(trans
, extent_root
);
953 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
954 struct btrfs_root
*root
)
956 struct extent_io_tree
*block_group_cache
;
957 struct btrfs_block_group_cache
*cache
;
961 struct btrfs_path
*path
;
967 block_group_cache
= &root
->fs_info
->block_group_cache
;
968 path
= btrfs_alloc_path();
973 ret
= find_first_extent_bit(block_group_cache
, last
,
974 &start
, &end
, BLOCK_GROUP_DIRTY
);
979 ret
= get_state_private(block_group_cache
, start
, &ptr
);
982 cache
= (struct btrfs_block_group_cache
*)(unsigned long)ptr
;
983 err
= write_one_cache_group(trans
, root
,
986 * if we fail to write the cache group, we want
987 * to keep it marked dirty in hopes that a later
994 clear_extent_bits(block_group_cache
, start
, end
,
995 BLOCK_GROUP_DIRTY
, GFP_NOFS
);
997 btrfs_free_path(path
);
1001 static struct btrfs_space_info
*__find_space_info(struct btrfs_fs_info
*info
,
1004 struct list_head
*head
= &info
->space_info
;
1005 struct list_head
*cur
;
1006 struct btrfs_space_info
*found
;
1007 list_for_each(cur
, head
) {
1008 found
= list_entry(cur
, struct btrfs_space_info
, list
);
1009 if (found
->flags
== flags
)
1016 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
1017 u64 total_bytes
, u64 bytes_used
,
1018 struct btrfs_space_info
**space_info
)
1020 struct btrfs_space_info
*found
;
1022 found
= __find_space_info(info
, flags
);
1024 found
->total_bytes
+= total_bytes
;
1025 found
->bytes_used
+= bytes_used
;
1027 WARN_ON(found
->total_bytes
< found
->bytes_used
);
1028 *space_info
= found
;
1031 found
= kmalloc(sizeof(*found
), GFP_NOFS
);
1035 list_add(&found
->list
, &info
->space_info
);
1036 found
->flags
= flags
;
1037 found
->total_bytes
= total_bytes
;
1038 found
->bytes_used
= bytes_used
;
1039 found
->bytes_pinned
= 0;
1041 *space_info
= found
;
1045 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
1047 u64 extra_flags
= flags
& (BTRFS_BLOCK_GROUP_RAID0
|
1048 BTRFS_BLOCK_GROUP_RAID1
|
1049 BTRFS_BLOCK_GROUP_RAID10
|
1050 BTRFS_BLOCK_GROUP_DUP
);
1052 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
1053 fs_info
->avail_data_alloc_bits
|= extra_flags
;
1054 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
1055 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
1056 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
1057 fs_info
->avail_system_alloc_bits
|= extra_flags
;
1061 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
1062 struct btrfs_root
*extent_root
, u64 alloc_bytes
,
1065 struct btrfs_space_info
*space_info
;
1071 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
1073 ret
= update_space_info(extent_root
->fs_info
, flags
,
1077 BUG_ON(!space_info
);
1079 if (space_info
->full
)
1082 thresh
= div_factor(space_info
->total_bytes
, 6);
1083 if ((space_info
->bytes_used
+ space_info
->bytes_pinned
+ alloc_bytes
) <
1087 ret
= btrfs_alloc_chunk(trans
, extent_root
, &start
, &num_bytes
, flags
);
1088 if (ret
== -ENOSPC
) {
1089 printk("space info full %Lu\n", flags
);
1090 space_info
->full
= 1;
1096 ret
= btrfs_make_block_group(trans
, extent_root
, 0, flags
,
1097 BTRFS_FIRST_CHUNK_TREE_OBJECTID
, start
, num_bytes
);
1103 static int update_block_group(struct btrfs_trans_handle
*trans
,
1104 struct btrfs_root
*root
,
1105 u64 bytenr
, u64 num_bytes
, int alloc
,
1108 struct btrfs_block_group_cache
*cache
;
1109 struct btrfs_fs_info
*info
= root
->fs_info
;
1110 u64 total
= num_bytes
;
1117 cache
= btrfs_lookup_block_group(info
, bytenr
);
1121 byte_in_group
= bytenr
- cache
->key
.objectid
;
1122 WARN_ON(byte_in_group
> cache
->key
.offset
);
1123 start
= cache
->key
.objectid
;
1124 end
= start
+ cache
->key
.offset
- 1;
1125 set_extent_bits(&info
->block_group_cache
, start
, end
,
1126 BLOCK_GROUP_DIRTY
, GFP_NOFS
);
1128 old_val
= btrfs_block_group_used(&cache
->item
);
1129 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
1131 old_val
+= num_bytes
;
1132 cache
->space_info
->bytes_used
+= num_bytes
;
1134 old_val
-= num_bytes
;
1135 cache
->space_info
->bytes_used
-= num_bytes
;
1137 set_extent_dirty(&info
->free_space_cache
,
1138 bytenr
, bytenr
+ num_bytes
- 1,
1142 btrfs_set_block_group_used(&cache
->item
, old_val
);
1144 bytenr
+= num_bytes
;
1149 static int update_pinned_extents(struct btrfs_root
*root
,
1150 u64 bytenr
, u64 num
, int pin
)
1153 struct btrfs_block_group_cache
*cache
;
1154 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1157 set_extent_dirty(&fs_info
->pinned_extents
,
1158 bytenr
, bytenr
+ num
- 1, GFP_NOFS
);
1160 clear_extent_dirty(&fs_info
->pinned_extents
,
1161 bytenr
, bytenr
+ num
- 1, GFP_NOFS
);
1164 cache
= btrfs_lookup_block_group(fs_info
, bytenr
);
1166 len
= min(num
, cache
->key
.offset
-
1167 (bytenr
- cache
->key
.objectid
));
1169 cache
->pinned
+= len
;
1170 cache
->space_info
->bytes_pinned
+= len
;
1171 fs_info
->total_pinned
+= len
;
1173 cache
->pinned
-= len
;
1174 cache
->space_info
->bytes_pinned
-= len
;
1175 fs_info
->total_pinned
-= len
;
1183 int btrfs_copy_pinned(struct btrfs_root
*root
, struct extent_io_tree
*copy
)
1188 struct extent_io_tree
*pinned_extents
= &root
->fs_info
->pinned_extents
;
1192 ret
= find_first_extent_bit(pinned_extents
, last
,
1193 &start
, &end
, EXTENT_DIRTY
);
1196 set_extent_dirty(copy
, start
, end
, GFP_NOFS
);
1202 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
1203 struct btrfs_root
*root
,
1204 struct extent_io_tree
*unpin
)
1209 struct extent_io_tree
*free_space_cache
;
1210 free_space_cache
= &root
->fs_info
->free_space_cache
;
1213 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
1217 update_pinned_extents(root
, start
, end
+ 1 - start
, 0);
1218 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
1219 set_extent_dirty(free_space_cache
, start
, end
, GFP_NOFS
);
1224 static int finish_current_insert(struct btrfs_trans_handle
*trans
,
1225 struct btrfs_root
*extent_root
)
1229 struct btrfs_fs_info
*info
= extent_root
->fs_info
;
1230 struct extent_buffer
*eb
;
1231 struct btrfs_path
*path
;
1232 struct btrfs_key ins
;
1233 struct btrfs_disk_key first
;
1234 struct btrfs_extent_item extent_item
;
1239 btrfs_set_stack_extent_refs(&extent_item
, 1);
1240 btrfs_set_key_type(&ins
, BTRFS_EXTENT_ITEM_KEY
);
1241 path
= btrfs_alloc_path();
1244 ret
= find_first_extent_bit(&info
->extent_ins
, 0, &start
,
1245 &end
, EXTENT_LOCKED
);
1249 ins
.objectid
= start
;
1250 ins
.offset
= end
+ 1 - start
;
1251 err
= btrfs_insert_item(trans
, extent_root
, &ins
,
1252 &extent_item
, sizeof(extent_item
));
1253 clear_extent_bits(&info
->extent_ins
, start
, end
, EXTENT_LOCKED
,
1255 eb
= read_tree_block(extent_root
, ins
.objectid
, ins
.offset
);
1256 level
= btrfs_header_level(eb
);
1258 btrfs_item_key(eb
, &first
, 0);
1260 btrfs_node_key(eb
, &first
, 0);
1262 err
= btrfs_insert_extent_backref(trans
, extent_root
, path
,
1263 start
, extent_root
->root_key
.objectid
,
1265 btrfs_disk_key_objectid(&first
));
1267 free_extent_buffer(eb
);
1269 btrfs_free_path(path
);
1273 static int pin_down_bytes(struct btrfs_root
*root
, u64 bytenr
, u32 num_bytes
,
1277 struct extent_buffer
*buf
;
1280 buf
= btrfs_find_tree_block(root
, bytenr
, num_bytes
);
1282 if (btrfs_buffer_uptodate(buf
)) {
1284 root
->fs_info
->running_transaction
->transid
;
1285 u64 header_transid
=
1286 btrfs_header_generation(buf
);
1287 if (header_transid
== transid
&&
1288 !btrfs_header_flag(buf
,
1289 BTRFS_HEADER_FLAG_WRITTEN
)) {
1290 clean_tree_block(NULL
, root
, buf
);
1291 free_extent_buffer(buf
);
1295 free_extent_buffer(buf
);
1297 update_pinned_extents(root
, bytenr
, num_bytes
, 1);
1299 set_extent_bits(&root
->fs_info
->pending_del
,
1300 bytenr
, bytenr
+ num_bytes
- 1,
1301 EXTENT_LOCKED
, GFP_NOFS
);
1308 * remove an extent from the root, returns 0 on success
1310 static int __free_extent(struct btrfs_trans_handle
*trans
, struct btrfs_root
1311 *root
, u64 bytenr
, u64 num_bytes
,
1312 u64 root_objectid
, u64 ref_generation
,
1313 u64 owner_objectid
, u64 owner_offset
, int pin
,
1316 struct btrfs_path
*path
;
1317 struct btrfs_key key
;
1318 struct btrfs_fs_info
*info
= root
->fs_info
;
1319 struct btrfs_root
*extent_root
= info
->extent_root
;
1320 struct extent_buffer
*leaf
;
1322 int extent_slot
= 0;
1323 int found_extent
= 0;
1325 struct btrfs_extent_item
*ei
;
1328 key
.objectid
= bytenr
;
1329 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
1330 key
.offset
= num_bytes
;
1331 path
= btrfs_alloc_path();
1336 ret
= lookup_extent_backref(trans
, extent_root
, path
,
1337 bytenr
, root_objectid
,
1339 owner_objectid
, owner_offset
, 1);
1341 struct btrfs_key found_key
;
1342 extent_slot
= path
->slots
[0];
1343 while(extent_slot
> 0) {
1345 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1347 if (found_key
.objectid
!= bytenr
)
1349 if (found_key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
1350 found_key
.offset
== num_bytes
) {
1354 if (path
->slots
[0] - extent_slot
> 5)
1358 ret
= btrfs_del_item(trans
, extent_root
, path
);
1360 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
1362 printk("Unable to find ref byte nr %Lu root %Lu "
1363 " gen %Lu owner %Lu offset %Lu\n", bytenr
,
1364 root_objectid
, ref_generation
, owner_objectid
,
1367 if (!found_extent
) {
1368 btrfs_release_path(extent_root
, path
);
1369 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
, -1, 1);
1373 extent_slot
= path
->slots
[0];
1376 leaf
= path
->nodes
[0];
1377 ei
= btrfs_item_ptr(leaf
, extent_slot
,
1378 struct btrfs_extent_item
);
1379 refs
= btrfs_extent_refs(leaf
, ei
);
1382 btrfs_set_extent_refs(leaf
, ei
, refs
);
1384 btrfs_mark_buffer_dirty(leaf
);
1386 if (refs
== 0 && found_extent
&& path
->slots
[0] == extent_slot
+ 1) {
1387 /* if the back ref and the extent are next to each other
1388 * they get deleted below in one shot
1390 path
->slots
[0] = extent_slot
;
1392 } else if (found_extent
) {
1393 /* otherwise delete the extent back ref */
1394 ret
= btrfs_del_item(trans
, extent_root
, path
);
1396 /* if refs are 0, we need to setup the path for deletion */
1398 btrfs_release_path(extent_root
, path
);
1399 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
1412 ret
= pin_down_bytes(root
, bytenr
, num_bytes
, 0);
1418 /* block accounting for super block */
1419 super_used
= btrfs_super_bytes_used(&info
->super_copy
);
1420 btrfs_set_super_bytes_used(&info
->super_copy
,
1421 super_used
- num_bytes
);
1423 /* block accounting for root item */
1424 root_used
= btrfs_root_used(&root
->root_item
);
1425 btrfs_set_root_used(&root
->root_item
,
1426 root_used
- num_bytes
);
1427 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
1432 ret
= update_block_group(trans
, root
, bytenr
, num_bytes
, 0,
1436 btrfs_free_path(path
);
1437 finish_current_insert(trans
, extent_root
);
1442 * find all the blocks marked as pending in the radix tree and remove
1443 * them from the extent map
1445 static int del_pending_extents(struct btrfs_trans_handle
*trans
, struct
1446 btrfs_root
*extent_root
)
1452 struct extent_io_tree
*pending_del
;
1453 struct extent_io_tree
*pinned_extents
;
1455 pending_del
= &extent_root
->fs_info
->pending_del
;
1456 pinned_extents
= &extent_root
->fs_info
->pinned_extents
;
1459 ret
= find_first_extent_bit(pending_del
, 0, &start
, &end
,
1463 update_pinned_extents(extent_root
, start
, end
+ 1 - start
, 1);
1464 clear_extent_bits(pending_del
, start
, end
, EXTENT_LOCKED
,
1466 ret
= __free_extent(trans
, extent_root
,
1467 start
, end
+ 1 - start
,
1468 extent_root
->root_key
.objectid
,
1477 * remove an extent from the root, returns 0 on success
1479 int btrfs_free_extent(struct btrfs_trans_handle
*trans
, struct btrfs_root
1480 *root
, u64 bytenr
, u64 num_bytes
,
1481 u64 root_objectid
, u64 ref_generation
,
1482 u64 owner_objectid
, u64 owner_offset
, int pin
)
1484 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
1488 WARN_ON(num_bytes
< root
->sectorsize
);
1489 if (!root
->ref_cows
)
1492 if (root
== extent_root
) {
1493 pin_down_bytes(root
, bytenr
, num_bytes
, 1);
1496 ret
= __free_extent(trans
, root
, bytenr
, num_bytes
, root_objectid
,
1497 ref_generation
, owner_objectid
, owner_offset
,
1499 pending_ret
= del_pending_extents(trans
, root
->fs_info
->extent_root
);
1500 return ret
? ret
: pending_ret
;
1503 static u64
stripe_align(struct btrfs_root
*root
, u64 val
)
1505 u64 mask
= ((u64
)root
->stripesize
- 1);
1506 u64 ret
= (val
+ mask
) & ~mask
;
1511 * walks the btree of allocated extents and find a hole of a given size.
1512 * The key ins is changed to record the hole:
1513 * ins->objectid == block start
1514 * ins->flags = BTRFS_EXTENT_ITEM_KEY
1515 * ins->offset == number of blocks
1516 * Any available blocks before search_start are skipped.
1518 static int noinline
find_free_extent(struct btrfs_trans_handle
*trans
,
1519 struct btrfs_root
*orig_root
,
1520 u64 num_bytes
, u64 empty_size
,
1521 u64 search_start
, u64 search_end
,
1522 u64 hint_byte
, struct btrfs_key
*ins
,
1523 u64 exclude_start
, u64 exclude_nr
,
1527 u64 orig_search_start
= search_start
;
1528 struct btrfs_root
* root
= orig_root
->fs_info
->extent_root
;
1529 struct btrfs_fs_info
*info
= root
->fs_info
;
1530 u64 total_needed
= num_bytes
;
1531 u64
*last_ptr
= NULL
;
1532 struct btrfs_block_group_cache
*block_group
;
1535 int empty_cluster
= 2 * 1024 * 1024;
1537 WARN_ON(num_bytes
< root
->sectorsize
);
1538 btrfs_set_key_type(ins
, BTRFS_EXTENT_ITEM_KEY
);
1540 if (data
& BTRFS_BLOCK_GROUP_METADATA
) {
1541 last_ptr
= &root
->fs_info
->last_alloc
;
1542 empty_cluster
= 256 * 1024;
1545 if ((data
& BTRFS_BLOCK_GROUP_DATA
) && btrfs_test_opt(root
, SSD
)) {
1546 last_ptr
= &root
->fs_info
->last_data_alloc
;
1551 hint_byte
= *last_ptr
;
1553 empty_size
+= empty_cluster
;
1557 if (search_end
== (u64
)-1)
1558 search_end
= btrfs_super_total_bytes(&info
->super_copy
);
1561 block_group
= btrfs_lookup_block_group(info
, hint_byte
);
1563 hint_byte
= search_start
;
1564 block_group
= btrfs_find_block_group(root
, block_group
,
1565 hint_byte
, data
, 1);
1566 if (last_ptr
&& *last_ptr
== 0 && block_group
)
1567 hint_byte
= block_group
->key
.objectid
;
1569 block_group
= btrfs_find_block_group(root
,
1571 search_start
, data
, 1);
1573 search_start
= max(search_start
, hint_byte
);
1575 total_needed
+= empty_size
;
1579 block_group
= btrfs_lookup_block_group(info
, search_start
);
1581 block_group
= btrfs_lookup_block_group(info
,
1584 ret
= find_search_start(root
, &block_group
, &search_start
,
1585 total_needed
, data
);
1586 if (ret
== -ENOSPC
&& last_ptr
&& *last_ptr
) {
1588 block_group
= btrfs_lookup_block_group(info
,
1590 search_start
= orig_search_start
;
1591 ret
= find_search_start(root
, &block_group
, &search_start
,
1592 total_needed
, data
);
1599 if (last_ptr
&& *last_ptr
&& search_start
!= *last_ptr
) {
1602 empty_size
+= empty_cluster
;
1603 total_needed
+= empty_size
;
1605 block_group
= btrfs_lookup_block_group(info
,
1607 search_start
= orig_search_start
;
1608 ret
= find_search_start(root
, &block_group
,
1609 &search_start
, total_needed
, data
);
1616 search_start
= stripe_align(root
, search_start
);
1617 ins
->objectid
= search_start
;
1618 ins
->offset
= num_bytes
;
1620 if (ins
->objectid
+ num_bytes
>= search_end
)
1623 if (ins
->objectid
+ num_bytes
>
1624 block_group
->key
.objectid
+ block_group
->key
.offset
) {
1625 search_start
= block_group
->key
.objectid
+
1626 block_group
->key
.offset
;
1630 if (test_range_bit(&info
->extent_ins
, ins
->objectid
,
1631 ins
->objectid
+ num_bytes
-1, EXTENT_LOCKED
, 0)) {
1632 search_start
= ins
->objectid
+ num_bytes
;
1636 if (test_range_bit(&info
->pinned_extents
, ins
->objectid
,
1637 ins
->objectid
+ num_bytes
-1, EXTENT_DIRTY
, 0)) {
1638 search_start
= ins
->objectid
+ num_bytes
;
1642 if (exclude_nr
> 0 && (ins
->objectid
+ num_bytes
> exclude_start
&&
1643 ins
->objectid
< exclude_start
+ exclude_nr
)) {
1644 search_start
= exclude_start
+ exclude_nr
;
1648 if (!(data
& BTRFS_BLOCK_GROUP_DATA
)) {
1649 block_group
= btrfs_lookup_block_group(info
, ins
->objectid
);
1651 trans
->block_group
= block_group
;
1653 ins
->offset
= num_bytes
;
1655 *last_ptr
= ins
->objectid
+ ins
->offset
;
1657 btrfs_super_total_bytes(&root
->fs_info
->super_copy
)) {
1664 if (search_start
+ num_bytes
>= search_end
) {
1666 search_start
= orig_search_start
;
1673 total_needed
-= empty_size
;
1678 block_group
= btrfs_lookup_block_group(info
, search_start
);
1680 block_group
= btrfs_find_block_group(root
, block_group
,
1681 search_start
, data
, 0);
1688 * finds a free extent and does all the dirty work required for allocation
1689 * returns the key for the extent through ins, and a tree buffer for
1690 * the first block of the extent through buf.
1692 * returns 0 if everything worked, non-zero otherwise.
1694 int btrfs_alloc_extent(struct btrfs_trans_handle
*trans
,
1695 struct btrfs_root
*root
,
1696 u64 num_bytes
, u64 min_alloc_size
,
1697 u64 root_objectid
, u64 ref_generation
,
1698 u64 owner
, u64 owner_offset
,
1699 u64 empty_size
, u64 hint_byte
,
1700 u64 search_end
, struct btrfs_key
*ins
, int data
)
1706 u64 search_start
= 0;
1709 struct btrfs_fs_info
*info
= root
->fs_info
;
1710 struct btrfs_root
*extent_root
= info
->extent_root
;
1711 struct btrfs_extent_item
*extent_item
;
1712 struct btrfs_extent_ref
*ref
;
1713 struct btrfs_path
*path
;
1714 struct btrfs_key keys
[2];
1717 alloc_profile
= info
->avail_data_alloc_bits
&
1718 info
->data_alloc_profile
;
1719 data
= BTRFS_BLOCK_GROUP_DATA
| alloc_profile
;
1720 } else if (root
== root
->fs_info
->chunk_root
) {
1721 alloc_profile
= info
->avail_system_alloc_bits
&
1722 info
->system_alloc_profile
;
1723 data
= BTRFS_BLOCK_GROUP_SYSTEM
| alloc_profile
;
1725 alloc_profile
= info
->avail_metadata_alloc_bits
&
1726 info
->metadata_alloc_profile
;
1727 data
= BTRFS_BLOCK_GROUP_METADATA
| alloc_profile
;
1730 if (root
->ref_cows
) {
1731 if (!(data
& BTRFS_BLOCK_GROUP_METADATA
)) {
1732 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
1734 BTRFS_BLOCK_GROUP_METADATA
|
1735 (info
->metadata_alloc_profile
&
1736 info
->avail_metadata_alloc_bits
));
1739 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
1740 num_bytes
+ 2 * 1024 * 1024, data
);
1744 WARN_ON(num_bytes
< root
->sectorsize
);
1745 ret
= find_free_extent(trans
, root
, num_bytes
, empty_size
,
1746 search_start
, search_end
, hint_byte
, ins
,
1747 trans
->alloc_exclude_start
,
1748 trans
->alloc_exclude_nr
, data
);
1750 if (ret
== -ENOSPC
&& num_bytes
> min_alloc_size
) {
1751 num_bytes
= num_bytes
>> 1;
1752 num_bytes
= max(num_bytes
, min_alloc_size
);
1759 /* block accounting for super block */
1760 super_used
= btrfs_super_bytes_used(&info
->super_copy
);
1761 btrfs_set_super_bytes_used(&info
->super_copy
, super_used
+ num_bytes
);
1763 /* block accounting for root item */
1764 root_used
= btrfs_root_used(&root
->root_item
);
1765 btrfs_set_root_used(&root
->root_item
, root_used
+ num_bytes
);
1767 clear_extent_dirty(&root
->fs_info
->free_space_cache
,
1768 ins
->objectid
, ins
->objectid
+ ins
->offset
- 1,
1771 if (root
== extent_root
) {
1772 set_extent_bits(&root
->fs_info
->extent_ins
, ins
->objectid
,
1773 ins
->objectid
+ ins
->offset
- 1,
1774 EXTENT_LOCKED
, GFP_NOFS
);
1778 WARN_ON(trans
->alloc_exclude_nr
);
1779 trans
->alloc_exclude_start
= ins
->objectid
;
1780 trans
->alloc_exclude_nr
= ins
->offset
;
1782 memcpy(&keys
[0], ins
, sizeof(*ins
));
1783 keys
[1].offset
= hash_extent_ref(root_objectid
, ref_generation
,
1784 owner
, owner_offset
);
1785 keys
[1].objectid
= ins
->objectid
;
1786 keys
[1].type
= BTRFS_EXTENT_REF_KEY
;
1787 sizes
[0] = sizeof(*extent_item
);
1788 sizes
[1] = sizeof(*ref
);
1790 path
= btrfs_alloc_path();
1793 ret
= btrfs_insert_empty_items(trans
, extent_root
, path
, keys
,
1797 extent_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1798 struct btrfs_extent_item
);
1799 btrfs_set_extent_refs(path
->nodes
[0], extent_item
, 1);
1800 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
1801 struct btrfs_extent_ref
);
1803 btrfs_set_ref_root(path
->nodes
[0], ref
, root_objectid
);
1804 btrfs_set_ref_generation(path
->nodes
[0], ref
, ref_generation
);
1805 btrfs_set_ref_objectid(path
->nodes
[0], ref
, owner
);
1806 btrfs_set_ref_offset(path
->nodes
[0], ref
, owner_offset
);
1808 btrfs_mark_buffer_dirty(path
->nodes
[0]);
1810 trans
->alloc_exclude_start
= 0;
1811 trans
->alloc_exclude_nr
= 0;
1812 btrfs_free_path(path
);
1813 finish_current_insert(trans
, extent_root
);
1814 pending_ret
= del_pending_extents(trans
, extent_root
);
1824 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1, 0);
1826 printk("update block group failed for %Lu %Lu\n",
1827 ins
->objectid
, ins
->offset
);
1834 * helper function to allocate a block for a given tree
1835 * returns the tree buffer or NULL.
1837 struct extent_buffer
*btrfs_alloc_free_block(struct btrfs_trans_handle
*trans
,
1838 struct btrfs_root
*root
,
1840 u64 root_objectid
, u64 hint
,
1846 ref_generation
= trans
->transid
;
1851 return __btrfs_alloc_free_block(trans
, root
, blocksize
, root_objectid
,
1852 ref_generation
, 0, 0, hint
, empty_size
);
1856 * helper function to allocate a block for a given tree
1857 * returns the tree buffer or NULL.
1859 struct extent_buffer
*__btrfs_alloc_free_block(struct btrfs_trans_handle
*trans
,
1860 struct btrfs_root
*root
,
1869 struct btrfs_key ins
;
1871 struct extent_buffer
*buf
;
1873 ret
= btrfs_alloc_extent(trans
, root
, blocksize
, blocksize
,
1874 root_objectid
, ref_generation
,
1875 level
, first_objectid
, empty_size
, hint
,
1879 return ERR_PTR(ret
);
1881 buf
= btrfs_find_create_tree_block(root
, ins
.objectid
, blocksize
);
1883 btrfs_free_extent(trans
, root
, ins
.objectid
, blocksize
,
1884 root
->root_key
.objectid
, ref_generation
,
1886 return ERR_PTR(-ENOMEM
);
1888 btrfs_set_header_generation(buf
, trans
->transid
);
1889 clean_tree_block(trans
, root
, buf
);
1890 btrfs_set_buffer_uptodate(buf
);
1892 if (PageDirty(buf
->first_page
)) {
1893 printk("page %lu dirty\n", buf
->first_page
->index
);
1897 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
1898 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
1899 if (!btrfs_test_opt(root
, SSD
))
1900 btrfs_set_buffer_defrag(buf
);
1901 trans
->blocks_used
++;
1905 static int noinline
drop_leaf_ref(struct btrfs_trans_handle
*trans
,
1906 struct btrfs_root
*root
,
1907 struct extent_buffer
*leaf
)
1910 u64 leaf_generation
;
1911 struct btrfs_key key
;
1912 struct btrfs_file_extent_item
*fi
;
1917 BUG_ON(!btrfs_is_leaf(leaf
));
1918 nritems
= btrfs_header_nritems(leaf
);
1919 leaf_owner
= btrfs_header_owner(leaf
);
1920 leaf_generation
= btrfs_header_generation(leaf
);
1922 for (i
= 0; i
< nritems
; i
++) {
1925 btrfs_item_key_to_cpu(leaf
, &key
, i
);
1926 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
1928 fi
= btrfs_item_ptr(leaf
, i
, struct btrfs_file_extent_item
);
1929 if (btrfs_file_extent_type(leaf
, fi
) ==
1930 BTRFS_FILE_EXTENT_INLINE
)
1933 * FIXME make sure to insert a trans record that
1934 * repeats the snapshot del on crash
1936 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
1937 if (disk_bytenr
== 0)
1939 ret
= btrfs_free_extent(trans
, root
, disk_bytenr
,
1940 btrfs_file_extent_disk_num_bytes(leaf
, fi
),
1941 leaf_owner
, leaf_generation
,
1942 key
.objectid
, key
.offset
, 0);
1948 static void noinline
reada_walk_down(struct btrfs_root
*root
,
1949 struct extent_buffer
*node
,
1962 nritems
= btrfs_header_nritems(node
);
1963 level
= btrfs_header_level(node
);
1967 for (i
= slot
; i
< nritems
&& skipped
< 32; i
++) {
1968 bytenr
= btrfs_node_blockptr(node
, i
);
1969 if (last
&& ((bytenr
> last
&& bytenr
- last
> 32 * 1024) ||
1970 (last
> bytenr
&& last
- bytenr
> 32 * 1024))) {
1974 blocksize
= btrfs_level_size(root
, level
- 1);
1976 ret
= lookup_extent_ref(NULL
, root
, bytenr
,
1984 mutex_unlock(&root
->fs_info
->fs_mutex
);
1985 ret
= readahead_tree_block(root
, bytenr
, blocksize
);
1986 last
= bytenr
+ blocksize
;
1988 mutex_lock(&root
->fs_info
->fs_mutex
);
1995 * helper function for drop_snapshot, this walks down the tree dropping ref
1996 * counts as it goes.
1998 static int noinline
walk_down_tree(struct btrfs_trans_handle
*trans
,
1999 struct btrfs_root
*root
,
2000 struct btrfs_path
*path
, int *level
)
2005 struct extent_buffer
*next
;
2006 struct extent_buffer
*cur
;
2007 struct extent_buffer
*parent
;
2012 WARN_ON(*level
< 0);
2013 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
2014 ret
= lookup_extent_ref(trans
, root
,
2015 path
->nodes
[*level
]->start
,
2016 path
->nodes
[*level
]->len
, &refs
);
2022 * walk down to the last node level and free all the leaves
2024 while(*level
>= 0) {
2025 WARN_ON(*level
< 0);
2026 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
2027 cur
= path
->nodes
[*level
];
2029 if (btrfs_header_level(cur
) != *level
)
2032 if (path
->slots
[*level
] >=
2033 btrfs_header_nritems(cur
))
2036 ret
= drop_leaf_ref(trans
, root
, cur
);
2040 bytenr
= btrfs_node_blockptr(cur
, path
->slots
[*level
]);
2041 blocksize
= btrfs_level_size(root
, *level
- 1);
2042 ret
= lookup_extent_ref(trans
, root
, bytenr
, blocksize
, &refs
);
2045 parent
= path
->nodes
[*level
];
2046 root_owner
= btrfs_header_owner(parent
);
2047 root_gen
= btrfs_header_generation(parent
);
2048 path
->slots
[*level
]++;
2049 ret
= btrfs_free_extent(trans
, root
, bytenr
,
2050 blocksize
, root_owner
,
2055 next
= btrfs_find_tree_block(root
, bytenr
, blocksize
);
2056 if (!next
|| !btrfs_buffer_uptodate(next
)) {
2057 free_extent_buffer(next
);
2058 reada_walk_down(root
, cur
, path
->slots
[*level
]);
2060 mutex_unlock(&root
->fs_info
->fs_mutex
);
2061 next
= read_tree_block(root
, bytenr
, blocksize
);
2062 mutex_lock(&root
->fs_info
->fs_mutex
);
2064 /* we've dropped the lock, double check */
2065 ret
= lookup_extent_ref(trans
, root
, bytenr
,
2069 parent
= path
->nodes
[*level
];
2070 root_owner
= btrfs_header_owner(parent
);
2071 root_gen
= btrfs_header_generation(parent
);
2073 path
->slots
[*level
]++;
2074 free_extent_buffer(next
);
2075 ret
= btrfs_free_extent(trans
, root
, bytenr
,
2083 btrfs_verify_block_csum(root
, next
);
2085 WARN_ON(*level
<= 0);
2086 if (path
->nodes
[*level
-1])
2087 free_extent_buffer(path
->nodes
[*level
-1]);
2088 path
->nodes
[*level
-1] = next
;
2089 *level
= btrfs_header_level(next
);
2090 path
->slots
[*level
] = 0;
2093 WARN_ON(*level
< 0);
2094 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
2096 if (path
->nodes
[*level
] == root
->node
) {
2097 root_owner
= root
->root_key
.objectid
;
2098 parent
= path
->nodes
[*level
];
2100 parent
= path
->nodes
[*level
+ 1];
2101 root_owner
= btrfs_header_owner(parent
);
2104 root_gen
= btrfs_header_generation(parent
);
2105 ret
= btrfs_free_extent(trans
, root
, path
->nodes
[*level
]->start
,
2106 path
->nodes
[*level
]->len
,
2107 root_owner
, root_gen
, 0, 0, 1);
2108 free_extent_buffer(path
->nodes
[*level
]);
2109 path
->nodes
[*level
] = NULL
;
2116 * helper for dropping snapshots. This walks back up the tree in the path
2117 * to find the first node higher up where we haven't yet gone through
2120 static int noinline
walk_up_tree(struct btrfs_trans_handle
*trans
,
2121 struct btrfs_root
*root
,
2122 struct btrfs_path
*path
, int *level
)
2126 struct btrfs_root_item
*root_item
= &root
->root_item
;
2131 for(i
= *level
; i
< BTRFS_MAX_LEVEL
- 1 && path
->nodes
[i
]; i
++) {
2132 slot
= path
->slots
[i
];
2133 if (slot
< btrfs_header_nritems(path
->nodes
[i
]) - 1) {
2134 struct extent_buffer
*node
;
2135 struct btrfs_disk_key disk_key
;
2136 node
= path
->nodes
[i
];
2139 WARN_ON(*level
== 0);
2140 btrfs_node_key(node
, &disk_key
, path
->slots
[i
]);
2141 memcpy(&root_item
->drop_progress
,
2142 &disk_key
, sizeof(disk_key
));
2143 root_item
->drop_level
= i
;
2146 if (path
->nodes
[*level
] == root
->node
) {
2147 root_owner
= root
->root_key
.objectid
;
2149 btrfs_header_generation(path
->nodes
[*level
]);
2151 struct extent_buffer
*node
;
2152 node
= path
->nodes
[*level
+ 1];
2153 root_owner
= btrfs_header_owner(node
);
2154 root_gen
= btrfs_header_generation(node
);
2156 ret
= btrfs_free_extent(trans
, root
,
2157 path
->nodes
[*level
]->start
,
2158 path
->nodes
[*level
]->len
,
2159 root_owner
, root_gen
, 0, 0, 1);
2161 free_extent_buffer(path
->nodes
[*level
]);
2162 path
->nodes
[*level
] = NULL
;
2170 * drop the reference count on the tree rooted at 'snap'. This traverses
2171 * the tree freeing any blocks that have a ref count of zero after being
2174 int btrfs_drop_snapshot(struct btrfs_trans_handle
*trans
, struct btrfs_root
2180 struct btrfs_path
*path
;
2183 struct btrfs_root_item
*root_item
= &root
->root_item
;
2185 path
= btrfs_alloc_path();
2188 level
= btrfs_header_level(root
->node
);
2190 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
2191 path
->nodes
[level
] = root
->node
;
2192 extent_buffer_get(root
->node
);
2193 path
->slots
[level
] = 0;
2195 struct btrfs_key key
;
2196 struct btrfs_disk_key found_key
;
2197 struct extent_buffer
*node
;
2199 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
2200 level
= root_item
->drop_level
;
2201 path
->lowest_level
= level
;
2202 wret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2207 node
= path
->nodes
[level
];
2208 btrfs_node_key(node
, &found_key
, path
->slots
[level
]);
2209 WARN_ON(memcmp(&found_key
, &root_item
->drop_progress
,
2210 sizeof(found_key
)));
2213 wret
= walk_down_tree(trans
, root
, path
, &level
);
2219 wret
= walk_up_tree(trans
, root
, path
, &level
);
2227 for (i
= 0; i
<= orig_level
; i
++) {
2228 if (path
->nodes
[i
]) {
2229 free_extent_buffer(path
->nodes
[i
]);
2230 path
->nodes
[i
] = NULL
;
2234 btrfs_free_path(path
);
2238 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
2245 ret
= find_first_extent_bit(&info
->block_group_cache
, 0,
2246 &start
, &end
, (unsigned int)-1);
2249 ret
= get_state_private(&info
->block_group_cache
, start
, &ptr
);
2251 kfree((void *)(unsigned long)ptr
);
2252 clear_extent_bits(&info
->block_group_cache
, start
,
2253 end
, (unsigned int)-1, GFP_NOFS
);
2256 ret
= find_first_extent_bit(&info
->free_space_cache
, 0,
2257 &start
, &end
, EXTENT_DIRTY
);
2260 clear_extent_dirty(&info
->free_space_cache
, start
,
2266 static unsigned long calc_ra(unsigned long start
, unsigned long last
,
2269 return min(last
, start
+ nr
- 1);
2272 static int noinline
relocate_inode_pages(struct inode
*inode
, u64 start
,
2278 u64 existing_delalloc
;
2279 unsigned long last_index
;
2282 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2283 struct file_ra_state
*ra
;
2284 unsigned long total_read
= 0;
2285 unsigned long ra_pages
;
2287 ra
= kzalloc(sizeof(*ra
), GFP_NOFS
);
2289 mutex_lock(&inode
->i_mutex
);
2290 i
= start
>> PAGE_CACHE_SHIFT
;
2291 last_index
= (start
+ len
- 1) >> PAGE_CACHE_SHIFT
;
2293 ra_pages
= BTRFS_I(inode
)->root
->fs_info
->bdi
.ra_pages
;
2295 file_ra_state_init(ra
, inode
->i_mapping
);
2298 for (; i
<= last_index
; i
++) {
2299 if (total_read
% ra_pages
== 0) {
2300 btrfs_force_ra(inode
->i_mapping
, ra
, NULL
, i
,
2301 calc_ra(i
, last_index
, ra_pages
));
2304 page
= grab_cache_page(inode
->i_mapping
, i
);
2307 if (!PageUptodate(page
)) {
2308 btrfs_readpage(NULL
, page
);
2310 if (!PageUptodate(page
)) {
2312 page_cache_release(page
);
2316 page_start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2317 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
2319 lock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
2321 delalloc_start
= page_start
;
2322 existing_delalloc
= count_range_bits(io_tree
,
2323 &delalloc_start
, page_end
,
2324 PAGE_CACHE_SIZE
, EXTENT_DELALLOC
);
2326 set_extent_delalloc(io_tree
, page_start
,
2327 page_end
, GFP_NOFS
);
2329 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
2330 set_page_dirty(page
);
2332 page_cache_release(page
);
2336 mutex_unlock(&inode
->i_mutex
);
2341 * note, this releases the path
2343 static int noinline
relocate_one_reference(struct btrfs_root
*extent_root
,
2344 struct btrfs_path
*path
,
2345 struct btrfs_key
*extent_key
)
2347 struct inode
*inode
;
2348 struct btrfs_root
*found_root
;
2349 struct btrfs_key
*root_location
;
2350 struct btrfs_extent_ref
*ref
;
2357 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2358 struct btrfs_extent_ref
);
2359 ref_root
= btrfs_ref_root(path
->nodes
[0], ref
);
2360 ref_gen
= btrfs_ref_generation(path
->nodes
[0], ref
);
2361 ref_objectid
= btrfs_ref_objectid(path
->nodes
[0], ref
);
2362 ref_offset
= btrfs_ref_offset(path
->nodes
[0], ref
);
2363 btrfs_release_path(extent_root
, path
);
2365 root_location
= kmalloc(sizeof(*root_location
), GFP_NOFS
);
2366 root_location
->objectid
= ref_root
;
2368 root_location
->offset
= 0;
2370 root_location
->offset
= (u64
)-1;
2371 root_location
->type
= BTRFS_ROOT_ITEM_KEY
;
2373 found_root
= btrfs_read_fs_root_no_name(extent_root
->fs_info
,
2375 BUG_ON(!found_root
);
2376 kfree(root_location
);
2378 if (ref_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
2379 mutex_unlock(&extent_root
->fs_info
->fs_mutex
);
2380 inode
= btrfs_iget_locked(extent_root
->fs_info
->sb
,
2381 ref_objectid
, found_root
);
2382 if (inode
->i_state
& I_NEW
) {
2383 /* the inode and parent dir are two different roots */
2384 BTRFS_I(inode
)->root
= found_root
;
2385 BTRFS_I(inode
)->location
.objectid
= ref_objectid
;
2386 BTRFS_I(inode
)->location
.type
= BTRFS_INODE_ITEM_KEY
;
2387 BTRFS_I(inode
)->location
.offset
= 0;
2388 btrfs_read_locked_inode(inode
);
2389 unlock_new_inode(inode
);
2392 /* this can happen if the reference is not against
2393 * the latest version of the tree root
2395 if (is_bad_inode(inode
)) {
2396 mutex_lock(&extent_root
->fs_info
->fs_mutex
);
2399 relocate_inode_pages(inode
, ref_offset
, extent_key
->offset
);
2400 /* FIXME, data=ordered will help get rid of this */
2401 filemap_fdatawrite(inode
->i_mapping
);
2403 mutex_lock(&extent_root
->fs_info
->fs_mutex
);
2405 struct btrfs_trans_handle
*trans
;
2406 struct btrfs_key found_key
;
2407 struct extent_buffer
*eb
;
2411 trans
= btrfs_start_transaction(found_root
, 1);
2412 eb
= read_tree_block(found_root
, extent_key
->objectid
,
2413 extent_key
->offset
);
2414 level
= btrfs_header_level(eb
);
2417 btrfs_item_key_to_cpu(eb
, &found_key
, 0);
2419 btrfs_node_key_to_cpu(eb
, &found_key
, 0);
2421 free_extent_buffer(eb
);
2423 path
->lowest_level
= level
;
2425 ret
= btrfs_search_slot(trans
, found_root
, &found_key
, path
,
2427 path
->lowest_level
= 0;
2428 for (i
= level
; i
< BTRFS_MAX_LEVEL
; i
++) {
2429 if (!path
->nodes
[i
])
2431 free_extent_buffer(path
->nodes
[i
]);
2432 path
->nodes
[i
] = NULL
;
2434 btrfs_release_path(found_root
, path
);
2435 btrfs_end_transaction(trans
, found_root
);
2442 static int noinline
relocate_one_extent(struct btrfs_root
*extent_root
,
2443 struct btrfs_path
*path
,
2444 struct btrfs_key
*extent_key
)
2446 struct btrfs_key key
;
2447 struct btrfs_key found_key
;
2448 struct extent_buffer
*leaf
;
2453 key
.objectid
= extent_key
->objectid
;
2454 key
.type
= BTRFS_EXTENT_REF_KEY
;
2458 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
2464 leaf
= path
->nodes
[0];
2465 nritems
= btrfs_header_nritems(leaf
);
2466 if (path
->slots
[0] == nritems
)
2469 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2470 if (found_key
.objectid
!= extent_key
->objectid
)
2473 if (found_key
.type
!= BTRFS_EXTENT_REF_KEY
)
2476 key
.offset
= found_key
.offset
+ 1;
2477 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2479 ret
= relocate_one_reference(extent_root
, path
, extent_key
);
2485 btrfs_release_path(extent_root
, path
);
2489 int btrfs_shrink_extent_tree(struct btrfs_root
*root
, u64 shrink_start
)
2491 struct btrfs_trans_handle
*trans
;
2492 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
2493 struct btrfs_path
*path
;
2496 u64 shrink_last_byte
;
2497 struct btrfs_block_group_cache
*shrink_block_group
;
2498 struct btrfs_fs_info
*info
= root
->fs_info
;
2499 struct btrfs_key key
;
2500 struct btrfs_key found_key
;
2501 struct extent_buffer
*leaf
;
2506 shrink_block_group
= btrfs_lookup_block_group(root
->fs_info
,
2508 BUG_ON(!shrink_block_group
);
2510 shrink_last_byte
= shrink_start
+ shrink_block_group
->key
.offset
;
2512 shrink_block_group
->space_info
->total_bytes
-=
2513 shrink_block_group
->key
.offset
;
2514 printk("shrink_extent_tree %Lu -> %Lu type %Lu\n", shrink_start
, shrink_last_byte
, shrink_block_group
->flags
);
2515 path
= btrfs_alloc_path();
2516 root
= root
->fs_info
->extent_root
;
2520 trans
= btrfs_start_transaction(root
, 1);
2521 do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
2522 btrfs_block_group_used(&shrink_block_group
->item
) +
2523 2 * 1024 * 1024, shrink_block_group
->flags
);
2524 btrfs_end_transaction(trans
, root
);
2525 shrink_block_group
->ro
= 1;
2528 key
.objectid
= shrink_start
;
2531 cur_byte
= key
.objectid
;
2533 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2537 ret
= btrfs_previous_item(root
, path
, 0, BTRFS_EXTENT_ITEM_KEY
);
2542 leaf
= path
->nodes
[0];
2543 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2544 if (found_key
.objectid
+ found_key
.offset
> shrink_start
&&
2545 found_key
.objectid
< shrink_last_byte
) {
2546 cur_byte
= found_key
.objectid
;
2547 key
.objectid
= cur_byte
;
2550 btrfs_release_path(root
, path
);
2553 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2557 leaf
= path
->nodes
[0];
2558 nritems
= btrfs_header_nritems(leaf
);
2560 if (path
->slots
[0] >= nritems
) {
2561 ret
= btrfs_next_leaf(root
, path
);
2568 leaf
= path
->nodes
[0];
2569 nritems
= btrfs_header_nritems(leaf
);
2572 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2574 if (found_key
.objectid
>= shrink_last_byte
)
2577 if (progress
&& need_resched()) {
2578 memcpy(&key
, &found_key
, sizeof(key
));
2579 mutex_unlock(&root
->fs_info
->fs_mutex
);
2581 mutex_lock(&root
->fs_info
->fs_mutex
);
2582 btrfs_release_path(root
, path
);
2583 btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2589 if (btrfs_key_type(&found_key
) != BTRFS_EXTENT_ITEM_KEY
||
2590 found_key
.objectid
+ found_key
.offset
<= cur_byte
) {
2596 cur_byte
= found_key
.objectid
+ found_key
.offset
;
2597 key
.objectid
= cur_byte
;
2598 btrfs_release_path(root
, path
);
2599 ret
= relocate_one_extent(root
, path
, &found_key
);
2602 btrfs_release_path(root
, path
);
2604 if (total_found
> 0) {
2605 trans
= btrfs_start_transaction(tree_root
, 1);
2606 btrfs_commit_transaction(trans
, tree_root
);
2608 mutex_unlock(&root
->fs_info
->fs_mutex
);
2609 btrfs_clean_old_snapshots(tree_root
);
2610 mutex_lock(&root
->fs_info
->fs_mutex
);
2612 trans
= btrfs_start_transaction(tree_root
, 1);
2613 btrfs_commit_transaction(trans
, tree_root
);
2618 * we've freed all the extents, now remove the block
2619 * group item from the tree
2621 trans
= btrfs_start_transaction(root
, 1);
2622 memcpy(&key
, &shrink_block_group
->key
, sizeof(key
));
2624 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2630 leaf
= path
->nodes
[0];
2631 nritems
= btrfs_header_nritems(leaf
);
2632 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2633 kfree(shrink_block_group
);
2635 clear_extent_bits(&info
->block_group_cache
, found_key
.objectid
,
2636 found_key
.objectid
+ found_key
.offset
- 1,
2637 (unsigned int)-1, GFP_NOFS
);
2639 btrfs_del_item(trans
, root
, path
);
2640 clear_extent_dirty(&info
->free_space_cache
,
2641 shrink_start
, shrink_last_byte
- 1,
2643 btrfs_commit_transaction(trans
, root
);
2645 btrfs_free_path(path
);
2649 int find_first_block_group(struct btrfs_root
*root
, struct btrfs_path
*path
,
2650 struct btrfs_key
*key
)
2653 struct btrfs_key found_key
;
2654 struct extent_buffer
*leaf
;
2657 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
2661 slot
= path
->slots
[0];
2662 leaf
= path
->nodes
[0];
2663 if (slot
>= btrfs_header_nritems(leaf
)) {
2664 ret
= btrfs_next_leaf(root
, path
);
2671 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
2673 if (found_key
.objectid
>= key
->objectid
&&
2674 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
)
2683 int btrfs_read_block_groups(struct btrfs_root
*root
)
2685 struct btrfs_path
*path
;
2688 struct btrfs_block_group_cache
*cache
;
2689 struct btrfs_fs_info
*info
= root
->fs_info
;
2690 struct btrfs_space_info
*space_info
;
2691 struct extent_io_tree
*block_group_cache
;
2692 struct btrfs_key key
;
2693 struct btrfs_key found_key
;
2694 struct extent_buffer
*leaf
;
2696 block_group_cache
= &info
->block_group_cache
;
2697 root
= info
->extent_root
;
2700 btrfs_set_key_type(&key
, BTRFS_BLOCK_GROUP_ITEM_KEY
);
2701 path
= btrfs_alloc_path();
2706 ret
= find_first_block_group(root
, path
, &key
);
2714 leaf
= path
->nodes
[0];
2715 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2716 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
2722 read_extent_buffer(leaf
, &cache
->item
,
2723 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
2724 sizeof(cache
->item
));
2725 memcpy(&cache
->key
, &found_key
, sizeof(found_key
));
2727 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
2728 btrfs_release_path(root
, path
);
2729 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
2731 if (cache
->flags
& BTRFS_BLOCK_GROUP_DATA
) {
2732 bit
= BLOCK_GROUP_DATA
;
2733 } else if (cache
->flags
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2734 bit
= BLOCK_GROUP_SYSTEM
;
2735 } else if (cache
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
2736 bit
= BLOCK_GROUP_METADATA
;
2738 set_avail_alloc_bits(info
, cache
->flags
);
2740 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
2741 btrfs_block_group_used(&cache
->item
),
2744 cache
->space_info
= space_info
;
2746 /* use EXTENT_LOCKED to prevent merging */
2747 set_extent_bits(block_group_cache
, found_key
.objectid
,
2748 found_key
.objectid
+ found_key
.offset
- 1,
2749 bit
| EXTENT_LOCKED
, GFP_NOFS
);
2750 set_state_private(block_group_cache
, found_key
.objectid
,
2751 (unsigned long)cache
);
2754 btrfs_super_total_bytes(&info
->super_copy
))
2759 btrfs_free_path(path
);
2763 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
2764 struct btrfs_root
*root
, u64 bytes_used
,
2765 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
2770 struct btrfs_root
*extent_root
;
2771 struct btrfs_block_group_cache
*cache
;
2772 struct extent_io_tree
*block_group_cache
;
2774 extent_root
= root
->fs_info
->extent_root
;
2775 block_group_cache
= &root
->fs_info
->block_group_cache
;
2777 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
2779 cache
->key
.objectid
= chunk_offset
;
2780 cache
->key
.offset
= size
;
2782 btrfs_set_key_type(&cache
->key
, BTRFS_BLOCK_GROUP_ITEM_KEY
);
2783 memset(&cache
->item
, 0, sizeof(cache
->item
));
2784 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
2785 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
2786 cache
->flags
= type
;
2787 btrfs_set_block_group_flags(&cache
->item
, type
);
2789 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
2790 &cache
->space_info
);
2793 bit
= block_group_state_bits(type
);
2794 set_extent_bits(block_group_cache
, chunk_offset
,
2795 chunk_offset
+ size
- 1,
2796 bit
| EXTENT_LOCKED
, GFP_NOFS
);
2798 set_state_private(block_group_cache
, chunk_offset
,
2799 (unsigned long)cache
);
2800 ret
= btrfs_insert_item(trans
, extent_root
, &cache
->key
, &cache
->item
,
2801 sizeof(cache
->item
));
2804 finish_current_insert(trans
, extent_root
);
2805 ret
= del_pending_extents(trans
, extent_root
);
2807 set_avail_alloc_bits(extent_root
->fs_info
, type
);