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.
19 #include <linux/sched.h>
20 #include <linux/crc32c.h>
21 #include <linux/pagemap.h>
25 #include "print-tree.h"
26 #include "transaction.h"
28 #define BLOCK_GROUP_DATA EXTENT_WRITEBACK
29 #define BLOCK_GROUP_METADATA EXTENT_UPTODATE
30 #define BLOCK_GROUP_DIRTY EXTENT_DIRTY
32 static int finish_current_insert(struct btrfs_trans_handle
*trans
, struct
33 btrfs_root
*extent_root
);
34 static int del_pending_extents(struct btrfs_trans_handle
*trans
, struct
35 btrfs_root
*extent_root
);
36 static int find_previous_extent(struct btrfs_root
*root
,
37 struct btrfs_path
*path
)
39 struct btrfs_key found_key
;
40 struct extent_buffer
*leaf
;
44 if (path
->slots
[0] == 0) {
45 ret
= btrfs_prev_leaf(root
, path
);
51 leaf
= path
->nodes
[0];
52 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
53 if (found_key
.type
== BTRFS_EXTENT_ITEM_KEY
)
59 static int cache_block_group(struct btrfs_root
*root
,
60 struct btrfs_block_group_cache
*block_group
)
62 struct btrfs_path
*path
;
65 struct extent_buffer
*leaf
;
66 struct extent_io_tree
*free_space_cache
;
76 root
= root
->fs_info
->extent_root
;
77 free_space_cache
= &root
->fs_info
->free_space_cache
;
79 if (block_group
->cached
)
82 path
= btrfs_alloc_path();
87 first_free
= block_group
->key
.objectid
;
88 key
.objectid
= block_group
->key
.objectid
;
90 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
91 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
94 ret
= find_previous_extent(root
, path
);
98 leaf
= path
->nodes
[0];
99 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
100 if (key
.objectid
+ key
.offset
> first_free
)
101 first_free
= key
.objectid
+ key
.offset
;
104 leaf
= path
->nodes
[0];
105 slot
= path
->slots
[0];
106 if (slot
>= btrfs_header_nritems(leaf
)) {
107 ret
= btrfs_next_leaf(root
, path
);
116 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
117 if (key
.objectid
< block_group
->key
.objectid
) {
120 if (key
.objectid
>= block_group
->key
.objectid
+
121 block_group
->key
.offset
) {
125 if (btrfs_key_type(&key
) == BTRFS_EXTENT_ITEM_KEY
) {
130 if (key
.objectid
> last
) {
131 hole_size
= key
.objectid
- last
;
132 set_extent_dirty(free_space_cache
, last
,
133 last
+ hole_size
- 1,
136 last
= key
.objectid
+ key
.offset
;
144 if (block_group
->key
.objectid
+
145 block_group
->key
.offset
> last
) {
146 hole_size
= block_group
->key
.objectid
+
147 block_group
->key
.offset
- last
;
148 set_extent_dirty(free_space_cache
, last
,
149 last
+ hole_size
- 1, GFP_NOFS
);
151 block_group
->cached
= 1;
153 btrfs_free_path(path
);
157 struct btrfs_block_group_cache
*btrfs_lookup_block_group(struct
161 struct extent_io_tree
*block_group_cache
;
162 struct btrfs_block_group_cache
*block_group
= NULL
;
168 block_group_cache
= &info
->block_group_cache
;
169 ret
= find_first_extent_bit(block_group_cache
,
170 bytenr
, &start
, &end
,
171 BLOCK_GROUP_DATA
| BLOCK_GROUP_METADATA
);
175 ret
= get_state_private(block_group_cache
, start
, &ptr
);
179 block_group
= (struct btrfs_block_group_cache
*)(unsigned long)ptr
;
180 if (block_group
->key
.objectid
<= bytenr
&& bytenr
<
181 block_group
->key
.objectid
+ block_group
->key
.offset
)
185 static u64 noinline
find_search_start(struct btrfs_root
*root
,
186 struct btrfs_block_group_cache
**cache_ret
,
187 u64 search_start
, int num
, int data
)
190 struct btrfs_block_group_cache
*cache
= *cache_ret
;
201 total_fs_bytes
= btrfs_super_total_bytes(&root
->fs_info
->super_copy
);
203 ret
= cache_block_group(root
, cache
);
207 last
= max(search_start
, cache
->key
.objectid
);
210 ret
= find_first_extent_bit(&root
->fs_info
->free_space_cache
,
211 last
, &start
, &end
, EXTENT_DIRTY
);
218 start
= max(last
, start
);
220 if (last
- start
< num
) {
221 if (last
== cache
->key
.objectid
+ cache
->key
.offset
)
225 if (data
!= BTRFS_BLOCK_GROUP_MIXED
&&
226 start
+ num
> cache
->key
.objectid
+ cache
->key
.offset
)
228 if (start
+ num
> total_fs_bytes
)
233 cache
= btrfs_lookup_block_group(root
->fs_info
, search_start
);
235 printk("Unable to find block group for %Lu\n",
243 last
= cache
->key
.objectid
+ cache
->key
.offset
;
245 cache
= btrfs_lookup_block_group(root
->fs_info
, last
);
246 if (!cache
|| cache
->key
.objectid
>= total_fs_bytes
) {
251 data
= BTRFS_BLOCK_GROUP_MIXED
;
256 if (cache_miss
&& !cache
->cached
) {
257 cache_block_group(root
, cache
);
259 cache
= btrfs_lookup_block_group(root
->fs_info
, last
);
261 cache
= btrfs_find_block_group(root
, cache
, last
, data
, 0);
269 static u64
div_factor(u64 num
, int factor
)
278 struct btrfs_block_group_cache
*btrfs_find_block_group(struct btrfs_root
*root
,
279 struct btrfs_block_group_cache
280 *hint
, u64 search_start
,
283 struct btrfs_block_group_cache
*cache
;
284 struct extent_io_tree
*block_group_cache
;
285 struct btrfs_block_group_cache
*found_group
= NULL
;
286 struct btrfs_fs_info
*info
= root
->fs_info
;
301 block_group_cache
= &info
->block_group_cache
;
302 total_fs_bytes
= btrfs_super_total_bytes(&root
->fs_info
->super_copy
);
307 if (data
== BTRFS_BLOCK_GROUP_MIXED
) {
308 bit
= BLOCK_GROUP_DATA
| BLOCK_GROUP_METADATA
;
311 bit
= BLOCK_GROUP_DATA
;
313 bit
= BLOCK_GROUP_METADATA
;
315 if (search_start
&& search_start
< total_fs_bytes
) {
316 struct btrfs_block_group_cache
*shint
;
317 shint
= btrfs_lookup_block_group(info
, search_start
);
318 if (shint
&& (shint
->data
== data
||
319 shint
->data
== BTRFS_BLOCK_GROUP_MIXED
)) {
320 used
= btrfs_block_group_used(&shint
->item
);
321 if (used
+ shint
->pinned
<
322 div_factor(shint
->key
.offset
, factor
)) {
327 if (hint
&& hint
->key
.objectid
< total_fs_bytes
&&
328 (hint
->data
== data
|| hint
->data
== BTRFS_BLOCK_GROUP_MIXED
)) {
329 used
= btrfs_block_group_used(&hint
->item
);
330 if (used
+ hint
->pinned
<
331 div_factor(hint
->key
.offset
, factor
)) {
334 last
= hint
->key
.objectid
+ hint
->key
.offset
;
338 hint_last
= max(hint
->key
.objectid
, search_start
);
340 hint_last
= search_start
;
342 if (hint_last
>= total_fs_bytes
)
343 hint_last
= search_start
;
348 ret
= find_first_extent_bit(block_group_cache
, last
,
353 ret
= get_state_private(block_group_cache
, start
, &ptr
);
357 cache
= (struct btrfs_block_group_cache
*)(unsigned long)ptr
;
358 last
= cache
->key
.objectid
+ cache
->key
.offset
;
359 used
= btrfs_block_group_used(&cache
->item
);
361 if (cache
->key
.objectid
> total_fs_bytes
)
365 free_check
= cache
->key
.offset
;
367 free_check
= div_factor(cache
->key
.offset
, factor
);
368 if (used
+ cache
->pinned
< free_check
) {
381 bit
= BLOCK_GROUP_DATA
| BLOCK_GROUP_METADATA
;
389 static u64
hash_extent_ref(u64 root_objectid
, u64 ref_generation
,
390 u64 owner
, u64 owner_offset
)
392 u32 high_crc
= ~(u32
)0;
393 u32 low_crc
= ~(u32
)0;
396 lenum
= cpu_to_le64(root_objectid
);
397 high_crc
= crc32c(high_crc
, &lenum
, sizeof(lenum
));
398 lenum
= cpu_to_le64(ref_generation
);
399 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
402 lenum
= cpu_to_le64(owner
);
403 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
404 lenum
= cpu_to_le64(owner_offset
);
405 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
407 return ((u64
)high_crc
<< 32) | (u64
)low_crc
;
410 static int match_extent_ref(struct extent_buffer
*leaf
,
411 struct btrfs_extent_ref
*disk_ref
,
412 struct btrfs_extent_ref
*cpu_ref
)
417 if (cpu_ref
->objectid
)
418 len
= sizeof(*cpu_ref
);
420 len
= 2 * sizeof(u64
);
421 ret
= memcmp_extent_buffer(leaf
, cpu_ref
, (unsigned long)disk_ref
,
426 static int noinline
lookup_extent_backref(struct btrfs_trans_handle
*trans
,
427 struct btrfs_root
*root
,
428 struct btrfs_path
*path
, u64 bytenr
,
430 u64 ref_generation
, u64 owner
,
431 u64 owner_offset
, int del
)
434 struct btrfs_key key
;
435 struct btrfs_key found_key
;
436 struct btrfs_extent_ref ref
;
437 struct extent_buffer
*leaf
;
438 struct btrfs_extent_ref
*disk_ref
;
442 btrfs_set_stack_ref_root(&ref
, root_objectid
);
443 btrfs_set_stack_ref_generation(&ref
, ref_generation
);
444 btrfs_set_stack_ref_objectid(&ref
, owner
);
445 btrfs_set_stack_ref_offset(&ref
, owner_offset
);
447 hash
= hash_extent_ref(root_objectid
, ref_generation
, owner
,
450 key
.objectid
= bytenr
;
451 key
.type
= BTRFS_EXTENT_REF_KEY
;
454 ret
= btrfs_search_slot(trans
, root
, &key
, path
,
458 leaf
= path
->nodes
[0];
460 u32 nritems
= btrfs_header_nritems(leaf
);
461 if (path
->slots
[0] >= nritems
) {
462 ret2
= btrfs_next_leaf(root
, path
);
465 leaf
= path
->nodes
[0];
467 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
468 if (found_key
.objectid
!= bytenr
||
469 found_key
.type
!= BTRFS_EXTENT_REF_KEY
)
471 key
.offset
= found_key
.offset
;
473 btrfs_release_path(root
, path
);
477 disk_ref
= btrfs_item_ptr(path
->nodes
[0],
479 struct btrfs_extent_ref
);
480 if (match_extent_ref(path
->nodes
[0], disk_ref
, &ref
)) {
484 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
485 key
.offset
= found_key
.offset
+ 1;
486 btrfs_release_path(root
, path
);
493 * Back reference rules. Back refs have three main goals:
495 * 1) differentiate between all holders of references to an extent so that
496 * when a reference is dropped we can make sure it was a valid reference
497 * before freeing the extent.
499 * 2) Provide enough information to quickly find the holders of an extent
500 * if we notice a given block is corrupted or bad.
502 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
503 * maintenance. This is actually the same as #2, but with a slightly
504 * different use case.
506 * File extents can be referenced by:
508 * - multiple snapshots, subvolumes, or different generations in one subvol
509 * - different files inside a single subvolume (in theory, not implemented yet)
510 * - different offsets inside a file (bookend extents in file.c)
512 * The extent ref structure has fields for:
514 * - Objectid of the subvolume root
515 * - Generation number of the tree holding the reference
516 * - objectid of the file holding the reference
517 * - offset in the file corresponding to the key holding the reference
519 * When a file extent is allocated the fields are filled in:
520 * (root_key.objectid, trans->transid, inode objectid, offset in file)
522 * When a leaf is cow'd new references are added for every file extent found
523 * in the leaf. It looks the same as the create case, but trans->transid
524 * will be different when the block is cow'd.
526 * (root_key.objectid, trans->transid, inode objectid, offset in file)
528 * When a file extent is removed either during snapshot deletion or file
529 * truncation, the corresponding back reference is found
532 * (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
533 * inode objectid, offset in file)
535 * Btree extents can be referenced by:
537 * - Different subvolumes
538 * - Different generations of the same subvolume
540 * Storing sufficient information for a full reverse mapping of a btree
541 * block would require storing the lowest key of the block in the backref,
542 * and it would require updating that lowest key either before write out or
543 * every time it changed. Instead, the objectid of the lowest key is stored
544 * along with the level of the tree block. This provides a hint
545 * about where in the btree the block can be found. Searches through the
546 * btree only need to look for a pointer to that block, so they stop one
547 * level higher than the level recorded in the backref.
549 * Some btrees do not do reference counting on their extents. These
550 * include the extent tree and the tree of tree roots. Backrefs for these
551 * trees always have a generation of zero.
553 * When a tree block is created, back references are inserted:
555 * (root->root_key.objectid, trans->transid or zero, level, lowest_key_objectid)
557 * When a tree block is cow'd in a reference counted root,
558 * new back references are added for all the blocks it points to.
559 * These are of the form (trans->transid will have increased since creation):
561 * (root->root_key.objectid, trans->transid, level, lowest_key_objectid)
563 * Because the lowest_key_objectid and the level are just hints
564 * they are not used when backrefs are deleted. When a backref is deleted:
566 * if backref was for a tree root:
567 * root_objectid = root->root_key.objectid
569 * root_objectid = btrfs_header_owner(parent)
571 * (root_objectid, btrfs_header_generation(parent) or zero, 0, 0)
573 * Back Reference Key hashing:
575 * Back references have four fields, each 64 bits long. Unfortunately,
576 * This is hashed into a single 64 bit number and placed into the key offset.
577 * The key objectid corresponds to the first byte in the extent, and the
578 * key type is set to BTRFS_EXTENT_REF_KEY
580 int btrfs_insert_extent_backref(struct btrfs_trans_handle
*trans
,
581 struct btrfs_root
*root
,
582 struct btrfs_path
*path
, u64 bytenr
,
583 u64 root_objectid
, u64 ref_generation
,
584 u64 owner
, u64 owner_offset
)
587 struct btrfs_key key
;
588 struct btrfs_extent_ref ref
;
589 struct btrfs_extent_ref
*disk_ref
;
592 btrfs_set_stack_ref_root(&ref
, root_objectid
);
593 btrfs_set_stack_ref_generation(&ref
, ref_generation
);
594 btrfs_set_stack_ref_objectid(&ref
, owner
);
595 btrfs_set_stack_ref_offset(&ref
, owner_offset
);
597 hash
= hash_extent_ref(root_objectid
, ref_generation
, owner
,
600 key
.objectid
= bytenr
;
601 key
.type
= BTRFS_EXTENT_REF_KEY
;
603 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, sizeof(ref
));
604 while (ret
== -EEXIST
) {
605 disk_ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
606 struct btrfs_extent_ref
);
607 if (match_extent_ref(path
->nodes
[0], disk_ref
, &ref
))
610 btrfs_release_path(root
, path
);
611 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
616 disk_ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
617 struct btrfs_extent_ref
);
618 write_extent_buffer(path
->nodes
[0], &ref
, (unsigned long)disk_ref
,
620 btrfs_mark_buffer_dirty(path
->nodes
[0]);
622 btrfs_release_path(root
, path
);
626 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
627 struct btrfs_root
*root
,
628 u64 bytenr
, u64 num_bytes
,
629 u64 root_objectid
, u64 ref_generation
,
630 u64 owner
, u64 owner_offset
)
632 struct btrfs_path
*path
;
634 struct btrfs_key key
;
635 struct extent_buffer
*l
;
636 struct btrfs_extent_item
*item
;
639 WARN_ON(num_bytes
< root
->sectorsize
);
640 path
= btrfs_alloc_path();
645 key
.objectid
= bytenr
;
646 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
647 key
.offset
= num_bytes
;
648 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
, path
,
657 item
= btrfs_item_ptr(l
, path
->slots
[0], struct btrfs_extent_item
);
658 refs
= btrfs_extent_refs(l
, item
);
659 btrfs_set_extent_refs(l
, item
, refs
+ 1);
660 btrfs_mark_buffer_dirty(path
->nodes
[0]);
662 btrfs_release_path(root
->fs_info
->extent_root
, path
);
665 ret
= btrfs_insert_extent_backref(trans
, root
->fs_info
->extent_root
,
666 path
, bytenr
, root_objectid
,
667 ref_generation
, owner
, owner_offset
);
669 finish_current_insert(trans
, root
->fs_info
->extent_root
);
670 del_pending_extents(trans
, root
->fs_info
->extent_root
);
672 btrfs_free_path(path
);
676 int btrfs_extent_post_op(struct btrfs_trans_handle
*trans
,
677 struct btrfs_root
*root
)
679 finish_current_insert(trans
, root
->fs_info
->extent_root
);
680 del_pending_extents(trans
, root
->fs_info
->extent_root
);
684 static int lookup_extent_ref(struct btrfs_trans_handle
*trans
,
685 struct btrfs_root
*root
, u64 bytenr
,
686 u64 num_bytes
, u32
*refs
)
688 struct btrfs_path
*path
;
690 struct btrfs_key key
;
691 struct extent_buffer
*l
;
692 struct btrfs_extent_item
*item
;
694 WARN_ON(num_bytes
< root
->sectorsize
);
695 path
= btrfs_alloc_path();
697 key
.objectid
= bytenr
;
698 key
.offset
= num_bytes
;
699 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
700 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
, path
,
705 btrfs_print_leaf(root
, path
->nodes
[0]);
706 printk("failed to find block number %Lu\n", bytenr
);
710 item
= btrfs_item_ptr(l
, path
->slots
[0], struct btrfs_extent_item
);
711 *refs
= btrfs_extent_refs(l
, item
);
713 btrfs_free_path(path
);
717 u32
btrfs_count_snapshots_in_path(struct btrfs_root
*root
,
718 struct btrfs_path
*count_path
,
721 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
722 struct btrfs_path
*path
;
725 u64 root_objectid
= root
->root_key
.objectid
;
730 struct btrfs_key key
;
731 struct btrfs_key found_key
;
732 struct extent_buffer
*l
;
733 struct btrfs_extent_item
*item
;
734 struct btrfs_extent_ref
*ref_item
;
737 path
= btrfs_alloc_path();
740 bytenr
= first_extent
;
742 bytenr
= count_path
->nodes
[level
]->start
;
745 key
.objectid
= bytenr
;
748 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
749 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
755 btrfs_item_key_to_cpu(l
, &found_key
, path
->slots
[0]);
757 if (found_key
.objectid
!= bytenr
||
758 found_key
.type
!= BTRFS_EXTENT_ITEM_KEY
) {
762 item
= btrfs_item_ptr(l
, path
->slots
[0], struct btrfs_extent_item
);
765 nritems
= btrfs_header_nritems(l
);
766 if (path
->slots
[0] >= nritems
) {
767 ret
= btrfs_next_leaf(extent_root
, path
);
772 btrfs_item_key_to_cpu(l
, &found_key
, path
->slots
[0]);
773 if (found_key
.objectid
!= bytenr
)
776 if (found_key
.type
!= BTRFS_EXTENT_REF_KEY
) {
782 ref_item
= btrfs_item_ptr(l
, path
->slots
[0],
783 struct btrfs_extent_ref
);
784 found_objectid
= btrfs_ref_root(l
, ref_item
);
786 if (found_objectid
!= root_objectid
) {
793 if (cur_count
== 0) {
797 if (level
>= 0 && root
->node
== count_path
->nodes
[level
])
800 btrfs_release_path(root
, path
);
804 btrfs_free_path(path
);
807 int btrfs_inc_root_ref(struct btrfs_trans_handle
*trans
,
808 struct btrfs_root
*root
, u64 owner_objectid
)
814 struct btrfs_disk_key disk_key
;
816 level
= btrfs_header_level(root
->node
);
817 generation
= trans
->transid
;
818 nritems
= btrfs_header_nritems(root
->node
);
821 btrfs_item_key(root
->node
, &disk_key
, 0);
823 btrfs_node_key(root
->node
, &disk_key
, 0);
824 key_objectid
= btrfs_disk_key_objectid(&disk_key
);
828 return btrfs_inc_extent_ref(trans
, root
, root
->node
->start
,
829 root
->node
->len
, owner_objectid
,
830 generation
, level
, key_objectid
);
833 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
834 struct extent_buffer
*buf
)
838 struct btrfs_key key
;
839 struct btrfs_file_extent_item
*fi
;
848 level
= btrfs_header_level(buf
);
849 nritems
= btrfs_header_nritems(buf
);
850 for (i
= 0; i
< nritems
; i
++) {
853 btrfs_item_key_to_cpu(buf
, &key
, i
);
854 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
856 fi
= btrfs_item_ptr(buf
, i
,
857 struct btrfs_file_extent_item
);
858 if (btrfs_file_extent_type(buf
, fi
) ==
859 BTRFS_FILE_EXTENT_INLINE
)
861 disk_bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
862 if (disk_bytenr
== 0)
864 ret
= btrfs_inc_extent_ref(trans
, root
, disk_bytenr
,
865 btrfs_file_extent_disk_num_bytes(buf
, fi
),
866 root
->root_key
.objectid
, trans
->transid
,
867 key
.objectid
, key
.offset
);
873 bytenr
= btrfs_node_blockptr(buf
, i
);
874 btrfs_node_key_to_cpu(buf
, &key
, i
);
875 ret
= btrfs_inc_extent_ref(trans
, root
, bytenr
,
876 btrfs_level_size(root
, level
- 1),
877 root
->root_key
.objectid
,
879 level
- 1, key
.objectid
);
890 for (i
=0; i
< faili
; i
++) {
893 btrfs_item_key_to_cpu(buf
, &key
, i
);
894 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
896 fi
= btrfs_item_ptr(buf
, i
,
897 struct btrfs_file_extent_item
);
898 if (btrfs_file_extent_type(buf
, fi
) ==
899 BTRFS_FILE_EXTENT_INLINE
)
901 disk_bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
902 if (disk_bytenr
== 0)
904 err
= btrfs_free_extent(trans
, root
, disk_bytenr
,
905 btrfs_file_extent_disk_num_bytes(buf
,
909 bytenr
= btrfs_node_blockptr(buf
, i
);
910 err
= btrfs_free_extent(trans
, root
, bytenr
,
911 btrfs_level_size(root
, level
- 1), 0);
919 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
920 struct btrfs_root
*root
,
921 struct btrfs_path
*path
,
922 struct btrfs_block_group_cache
*cache
)
926 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
928 struct extent_buffer
*leaf
;
930 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
935 leaf
= path
->nodes
[0];
936 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
937 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
938 btrfs_mark_buffer_dirty(leaf
);
939 btrfs_release_path(extent_root
, path
);
941 finish_current_insert(trans
, extent_root
);
942 pending_ret
= del_pending_extents(trans
, extent_root
);
951 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
952 struct btrfs_root
*root
)
954 struct extent_io_tree
*block_group_cache
;
955 struct btrfs_block_group_cache
*cache
;
959 struct btrfs_path
*path
;
965 block_group_cache
= &root
->fs_info
->block_group_cache
;
966 path
= btrfs_alloc_path();
971 ret
= find_first_extent_bit(block_group_cache
, last
,
972 &start
, &end
, BLOCK_GROUP_DIRTY
);
977 ret
= get_state_private(block_group_cache
, start
, &ptr
);
981 cache
= (struct btrfs_block_group_cache
*)(unsigned long)ptr
;
982 err
= write_one_cache_group(trans
, root
,
985 * if we fail to write the cache group, we want
986 * to keep it marked dirty in hopes that a later
993 clear_extent_bits(block_group_cache
, start
, end
,
994 BLOCK_GROUP_DIRTY
, GFP_NOFS
);
996 btrfs_free_path(path
);
1000 static int update_block_group(struct btrfs_trans_handle
*trans
,
1001 struct btrfs_root
*root
,
1002 u64 bytenr
, u64 num_bytes
, int alloc
,
1003 int mark_free
, int data
)
1005 struct btrfs_block_group_cache
*cache
;
1006 struct btrfs_fs_info
*info
= root
->fs_info
;
1007 u64 total
= num_bytes
;
1014 cache
= btrfs_lookup_block_group(info
, bytenr
);
1018 byte_in_group
= bytenr
- cache
->key
.objectid
;
1019 WARN_ON(byte_in_group
> cache
->key
.offset
);
1020 start
= cache
->key
.objectid
;
1021 end
= start
+ cache
->key
.offset
- 1;
1022 set_extent_bits(&info
->block_group_cache
, start
, end
,
1023 BLOCK_GROUP_DIRTY
, GFP_NOFS
);
1025 old_val
= btrfs_block_group_used(&cache
->item
);
1026 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
1028 if (cache
->data
!= data
&&
1029 old_val
< (cache
->key
.offset
>> 1)) {
1034 bit_to_clear
= BLOCK_GROUP_METADATA
;
1035 bit_to_set
= BLOCK_GROUP_DATA
;
1036 cache
->item
.flags
&=
1037 ~BTRFS_BLOCK_GROUP_MIXED
;
1038 cache
->item
.flags
|=
1039 BTRFS_BLOCK_GROUP_DATA
;
1041 bit_to_clear
= BLOCK_GROUP_DATA
;
1042 bit_to_set
= BLOCK_GROUP_METADATA
;
1043 cache
->item
.flags
&=
1044 ~BTRFS_BLOCK_GROUP_MIXED
;
1045 cache
->item
.flags
&=
1046 ~BTRFS_BLOCK_GROUP_DATA
;
1048 clear_extent_bits(&info
->block_group_cache
,
1049 start
, end
, bit_to_clear
,
1051 set_extent_bits(&info
->block_group_cache
,
1052 start
, end
, bit_to_set
,
1054 } else if (cache
->data
!= data
&&
1055 cache
->data
!= BTRFS_BLOCK_GROUP_MIXED
) {
1056 cache
->data
= BTRFS_BLOCK_GROUP_MIXED
;
1057 set_extent_bits(&info
->block_group_cache
,
1060 BLOCK_GROUP_METADATA
,
1063 old_val
+= num_bytes
;
1065 old_val
-= num_bytes
;
1067 set_extent_dirty(&info
->free_space_cache
,
1068 bytenr
, bytenr
+ num_bytes
- 1,
1072 btrfs_set_block_group_used(&cache
->item
, old_val
);
1074 bytenr
+= num_bytes
;
1078 static int update_pinned_extents(struct btrfs_root
*root
,
1079 u64 bytenr
, u64 num
, int pin
)
1082 struct btrfs_block_group_cache
*cache
;
1083 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1086 set_extent_dirty(&fs_info
->pinned_extents
,
1087 bytenr
, bytenr
+ num
- 1, GFP_NOFS
);
1089 clear_extent_dirty(&fs_info
->pinned_extents
,
1090 bytenr
, bytenr
+ num
- 1, GFP_NOFS
);
1093 cache
= btrfs_lookup_block_group(fs_info
, bytenr
);
1095 len
= min(num
, cache
->key
.offset
-
1096 (bytenr
- cache
->key
.objectid
));
1098 cache
->pinned
+= len
;
1099 fs_info
->total_pinned
+= len
;
1101 cache
->pinned
-= len
;
1102 fs_info
->total_pinned
-= len
;
1110 int btrfs_copy_pinned(struct btrfs_root
*root
, struct extent_io_tree
*copy
)
1115 struct extent_io_tree
*pinned_extents
= &root
->fs_info
->pinned_extents
;
1119 ret
= find_first_extent_bit(pinned_extents
, last
,
1120 &start
, &end
, EXTENT_DIRTY
);
1123 set_extent_dirty(copy
, start
, end
, GFP_NOFS
);
1129 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
1130 struct btrfs_root
*root
,
1131 struct extent_io_tree
*unpin
)
1136 struct extent_io_tree
*free_space_cache
;
1137 free_space_cache
= &root
->fs_info
->free_space_cache
;
1140 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
1144 update_pinned_extents(root
, start
, end
+ 1 - start
, 0);
1145 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
1146 set_extent_dirty(free_space_cache
, start
, end
, GFP_NOFS
);
1151 static int finish_current_insert(struct btrfs_trans_handle
*trans
,
1152 struct btrfs_root
*extent_root
)
1156 struct btrfs_fs_info
*info
= extent_root
->fs_info
;
1157 struct extent_buffer
*eb
;
1158 struct btrfs_path
*path
;
1159 struct btrfs_key ins
;
1160 struct btrfs_disk_key first
;
1161 struct btrfs_extent_item extent_item
;
1166 btrfs_set_stack_extent_refs(&extent_item
, 1);
1167 btrfs_set_key_type(&ins
, BTRFS_EXTENT_ITEM_KEY
);
1168 path
= btrfs_alloc_path();
1171 ret
= find_first_extent_bit(&info
->extent_ins
, 0, &start
,
1172 &end
, EXTENT_LOCKED
);
1176 ins
.objectid
= start
;
1177 ins
.offset
= end
+ 1 - start
;
1178 err
= btrfs_insert_item(trans
, extent_root
, &ins
,
1179 &extent_item
, sizeof(extent_item
));
1180 clear_extent_bits(&info
->extent_ins
, start
, end
, EXTENT_LOCKED
,
1182 eb
= read_tree_block(extent_root
, ins
.objectid
, ins
.offset
);
1183 level
= btrfs_header_level(eb
);
1185 btrfs_item_key(eb
, &first
, 0);
1187 btrfs_node_key(eb
, &first
, 0);
1189 err
= btrfs_insert_extent_backref(trans
, extent_root
, path
,
1190 start
, extent_root
->root_key
.objectid
,
1192 btrfs_disk_key_objectid(&first
));
1194 free_extent_buffer(eb
);
1196 btrfs_free_path(path
);
1200 static int pin_down_bytes(struct btrfs_root
*root
, u64 bytenr
, u32 num_bytes
,
1204 struct extent_buffer
*buf
;
1207 buf
= btrfs_find_tree_block(root
, bytenr
, num_bytes
);
1209 if (btrfs_buffer_uptodate(buf
)) {
1211 root
->fs_info
->running_transaction
->transid
;
1212 u64 header_transid
=
1213 btrfs_header_generation(buf
);
1214 if (header_transid
== transid
) {
1215 clean_tree_block(NULL
, root
, buf
);
1216 free_extent_buffer(buf
);
1220 free_extent_buffer(buf
);
1222 update_pinned_extents(root
, bytenr
, num_bytes
, 1);
1224 set_extent_bits(&root
->fs_info
->pending_del
,
1225 bytenr
, bytenr
+ num_bytes
- 1,
1226 EXTENT_LOCKED
, GFP_NOFS
);
1233 * remove an extent from the root, returns 0 on success
1235 static int __free_extent(struct btrfs_trans_handle
*trans
, struct btrfs_root
1236 *root
, u64 bytenr
, u64 num_bytes
,
1237 u64 root_objectid
, u64 ref_generation
,
1238 u64 owner_objectid
, u64 owner_offset
, int pin
,
1241 struct btrfs_path
*path
;
1242 struct btrfs_key key
;
1243 struct btrfs_fs_info
*info
= root
->fs_info
;
1244 struct btrfs_root
*extent_root
= info
->extent_root
;
1245 struct extent_buffer
*leaf
;
1247 struct btrfs_extent_item
*ei
;
1250 key
.objectid
= bytenr
;
1251 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
1252 key
.offset
= num_bytes
;
1253 path
= btrfs_alloc_path();
1258 ret
= lookup_extent_backref(trans
, extent_root
, path
,
1259 bytenr
, root_objectid
,
1261 owner_objectid
, owner_offset
, 1);
1263 ret
= btrfs_del_item(trans
, extent_root
, path
);
1265 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
1267 printk("Unable to find ref byte nr %Lu root %Lu "
1268 " gen %Lu owner %Lu offset %Lu\n", bytenr
,
1269 root_objectid
, ref_generation
, owner_objectid
,
1272 btrfs_release_path(extent_root
, path
);
1273 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
, -1, 1);
1278 leaf
= path
->nodes
[0];
1279 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
1280 struct btrfs_extent_item
);
1281 refs
= btrfs_extent_refs(leaf
, ei
);
1284 btrfs_set_extent_refs(leaf
, ei
, refs
);
1285 btrfs_mark_buffer_dirty(leaf
);
1292 ret
= pin_down_bytes(root
, bytenr
, num_bytes
, 0);
1298 /* block accounting for super block */
1299 super_used
= btrfs_super_bytes_used(&info
->super_copy
);
1300 btrfs_set_super_bytes_used(&info
->super_copy
,
1301 super_used
- num_bytes
);
1303 /* block accounting for root item */
1304 root_used
= btrfs_root_used(&root
->root_item
);
1305 btrfs_set_root_used(&root
->root_item
,
1306 root_used
- num_bytes
);
1308 ret
= btrfs_del_item(trans
, extent_root
, path
);
1312 ret
= update_block_group(trans
, root
, bytenr
, num_bytes
, 0,
1316 btrfs_free_path(path
);
1317 finish_current_insert(trans
, extent_root
);
1322 * find all the blocks marked as pending in the radix tree and remove
1323 * them from the extent map
1325 static int del_pending_extents(struct btrfs_trans_handle
*trans
, struct
1326 btrfs_root
*extent_root
)
1332 struct extent_io_tree
*pending_del
;
1333 struct extent_io_tree
*pinned_extents
;
1335 pending_del
= &extent_root
->fs_info
->pending_del
;
1336 pinned_extents
= &extent_root
->fs_info
->pinned_extents
;
1339 ret
= find_first_extent_bit(pending_del
, 0, &start
, &end
,
1343 update_pinned_extents(extent_root
, start
, end
+ 1 - start
, 1);
1344 clear_extent_bits(pending_del
, start
, end
, EXTENT_LOCKED
,
1346 ret
= __free_extent(trans
, extent_root
,
1347 start
, end
+ 1 - start
,
1348 extent_root
->root_key
.objectid
,
1357 * remove an extent from the root, returns 0 on success
1359 int btrfs_free_extent(struct btrfs_trans_handle
*trans
, struct btrfs_root
1360 *root
, u64 bytenr
, u64 num_bytes
,
1361 u64 root_objectid
, u64 ref_generation
,
1362 u64 owner_objectid
, u64 owner_offset
, int pin
)
1364 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
1368 WARN_ON(num_bytes
< root
->sectorsize
);
1369 if (!root
->ref_cows
)
1372 if (root
== extent_root
) {
1373 pin_down_bytes(root
, bytenr
, num_bytes
, 1);
1376 ret
= __free_extent(trans
, root
, bytenr
, num_bytes
, root_objectid
,
1377 ref_generation
, owner_objectid
, owner_offset
,
1379 pending_ret
= del_pending_extents(trans
, root
->fs_info
->extent_root
);
1380 return ret
? ret
: pending_ret
;
1383 static u64
stripe_align(struct btrfs_root
*root
, u64 val
)
1385 u64 mask
= ((u64
)root
->stripesize
- 1);
1386 u64 ret
= (val
+ mask
) & ~mask
;
1391 * walks the btree of allocated extents and find a hole of a given size.
1392 * The key ins is changed to record the hole:
1393 * ins->objectid == block start
1394 * ins->flags = BTRFS_EXTENT_ITEM_KEY
1395 * ins->offset == number of blocks
1396 * Any available blocks before search_start are skipped.
1398 static int noinline
find_free_extent(struct btrfs_trans_handle
*trans
,
1399 struct btrfs_root
*orig_root
,
1400 u64 num_bytes
, u64 empty_size
,
1401 u64 search_start
, u64 search_end
,
1402 u64 hint_byte
, struct btrfs_key
*ins
,
1403 u64 exclude_start
, u64 exclude_nr
,
1406 struct btrfs_path
*path
;
1407 struct btrfs_key key
;
1413 u64
*last_ptr
= NULL
;
1414 u64 orig_search_start
= search_start
;
1416 struct extent_buffer
*l
;
1417 struct btrfs_root
* root
= orig_root
->fs_info
->extent_root
;
1418 struct btrfs_fs_info
*info
= root
->fs_info
;
1419 u64 total_needed
= num_bytes
;
1421 struct btrfs_block_group_cache
*block_group
;
1426 WARN_ON(num_bytes
< root
->sectorsize
);
1427 btrfs_set_key_type(ins
, BTRFS_EXTENT_ITEM_KEY
);
1429 level
= btrfs_header_level(root
->node
);
1431 if (num_bytes
>= 32 * 1024 * 1024 && hint_byte
) {
1432 data
= BTRFS_BLOCK_GROUP_MIXED
;
1435 /* for SSD, cluster allocations together as much as possible */
1436 if (btrfs_test_opt(root
, SSD
)) {
1438 last_ptr
= &root
->fs_info
->last_data_alloc
;
1440 last_ptr
= &root
->fs_info
->last_alloc
;
1442 hint_byte
= *last_ptr
;
1445 hint_byte
= hint_byte
&
1446 ~((u64
)BTRFS_BLOCK_GROUP_SIZE
- 1);
1447 empty_size
+= 16 * 1024 * 1024;
1451 search_end
= min(search_end
,
1452 btrfs_super_total_bytes(&info
->super_copy
));
1454 block_group
= btrfs_lookup_block_group(info
, hint_byte
);
1456 hint_byte
= search_start
;
1457 block_group
= btrfs_find_block_group(root
, block_group
,
1458 hint_byte
, data
, 1);
1460 block_group
= btrfs_find_block_group(root
,
1462 search_start
, data
, 1);
1465 total_needed
+= empty_size
;
1466 path
= btrfs_alloc_path();
1469 block_group
= btrfs_lookup_block_group(info
, search_start
);
1471 block_group
= btrfs_lookup_block_group(info
,
1474 search_start
= find_search_start(root
, &block_group
, search_start
,
1475 total_needed
, data
);
1477 if (btrfs_test_opt(root
, SSD
) && *last_ptr
&&
1478 search_start
!= *last_ptr
) {
1479 info
->last_alloc
= 0;
1481 empty_size
+= 16 * 1024 * 1024;
1482 total_needed
+= empty_size
;
1484 search_start
= find_search_start(root
, &block_group
,
1485 search_start
, total_needed
,
1489 search_start
= stripe_align(root
, search_start
);
1490 cached_start
= search_start
;
1491 btrfs_init_path(path
);
1492 ins
->objectid
= search_start
;
1497 ret
= btrfs_search_slot(trans
, root
, ins
, path
, 0, 0);
1500 ret
= find_previous_extent(root
, path
);
1504 btrfs_item_key_to_cpu(l
, &key
, path
->slots
[0]);
1507 slot
= path
->slots
[0];
1508 if (slot
>= btrfs_header_nritems(l
)) {
1509 ret
= btrfs_next_leaf(root
, path
);
1515 search_start
= max(search_start
,
1516 block_group
->key
.objectid
);
1518 aligned
= stripe_align(root
, search_start
);
1519 ins
->objectid
= aligned
;
1520 if (aligned
>= search_end
) {
1524 ins
->offset
= search_end
- aligned
;
1528 ins
->objectid
= stripe_align(root
,
1529 last_byte
> search_start
?
1530 last_byte
: search_start
);
1531 if (search_end
<= ins
->objectid
) {
1535 ins
->offset
= search_end
- ins
->objectid
;
1536 BUG_ON(ins
->objectid
>= search_end
);
1539 btrfs_item_key_to_cpu(l
, &key
, slot
);
1541 if (key
.objectid
>= search_start
&& key
.objectid
> last_byte
&&
1543 if (last_byte
< search_start
)
1544 last_byte
= search_start
;
1545 aligned
= stripe_align(root
, last_byte
);
1546 hole_size
= key
.objectid
- aligned
;
1547 if (key
.objectid
> aligned
&& hole_size
>= num_bytes
) {
1548 ins
->objectid
= aligned
;
1549 ins
->offset
= hole_size
;
1553 if (btrfs_key_type(&key
) != BTRFS_EXTENT_ITEM_KEY
) {
1554 if (!start_found
&& btrfs_key_type(&key
) ==
1555 BTRFS_BLOCK_GROUP_ITEM_KEY
) {
1556 last_byte
= key
.objectid
;
1564 last_byte
= key
.objectid
+ key
.offset
;
1566 if (!full_scan
&& data
!= BTRFS_BLOCK_GROUP_MIXED
&&
1567 last_byte
>= block_group
->key
.objectid
+
1568 block_group
->key
.offset
) {
1569 btrfs_release_path(root
, path
);
1570 search_start
= block_group
->key
.objectid
+
1571 block_group
->key
.offset
;
1579 /* we have to make sure we didn't find an extent that has already
1580 * been allocated by the map tree or the original allocation
1582 btrfs_release_path(root
, path
);
1583 BUG_ON(ins
->objectid
< search_start
);
1585 if (ins
->objectid
+ num_bytes
>= search_end
)
1587 if (!full_scan
&& data
!= BTRFS_BLOCK_GROUP_MIXED
&&
1588 ins
->objectid
+ num_bytes
> block_group
->
1589 key
.objectid
+ block_group
->key
.offset
) {
1590 search_start
= block_group
->key
.objectid
+
1591 block_group
->key
.offset
;
1594 if (test_range_bit(&info
->extent_ins
, ins
->objectid
,
1595 ins
->objectid
+ num_bytes
-1, EXTENT_LOCKED
, 0)) {
1596 search_start
= ins
->objectid
+ num_bytes
;
1599 if (test_range_bit(&info
->pinned_extents
, ins
->objectid
,
1600 ins
->objectid
+ num_bytes
-1, EXTENT_DIRTY
, 0)) {
1601 search_start
= ins
->objectid
+ num_bytes
;
1604 if (exclude_nr
> 0 && (ins
->objectid
+ num_bytes
> exclude_start
&&
1605 ins
->objectid
< exclude_start
+ exclude_nr
)) {
1606 search_start
= exclude_start
+ exclude_nr
;
1610 block_group
= btrfs_lookup_block_group(info
, ins
->objectid
);
1612 trans
->block_group
= block_group
;
1614 ins
->offset
= num_bytes
;
1615 btrfs_free_path(path
);
1616 if (btrfs_test_opt(root
, SSD
))
1617 *last_ptr
= ins
->objectid
+ ins
->offset
;
1621 if (search_start
+ num_bytes
>= search_end
) {
1623 search_start
= orig_search_start
;
1630 total_needed
-= empty_size
;
1632 data
= BTRFS_BLOCK_GROUP_MIXED
;
1636 block_group
= btrfs_lookup_block_group(info
, search_start
);
1638 block_group
= btrfs_find_block_group(root
, block_group
,
1639 search_start
, data
, 0);
1643 btrfs_release_path(root
, path
);
1644 btrfs_free_path(path
);
1648 * finds a free extent and does all the dirty work required for allocation
1649 * returns the key for the extent through ins, and a tree buffer for
1650 * the first block of the extent through buf.
1652 * returns 0 if everything worked, non-zero otherwise.
1654 int btrfs_alloc_extent(struct btrfs_trans_handle
*trans
,
1655 struct btrfs_root
*root
,
1656 u64 num_bytes
, u64 root_objectid
, u64 ref_generation
,
1657 u64 owner
, u64 owner_offset
,
1658 u64 empty_size
, u64 hint_byte
,
1659 u64 search_end
, struct btrfs_key
*ins
, int data
)
1665 u64 search_start
= 0;
1668 struct btrfs_fs_info
*info
= root
->fs_info
;
1669 struct btrfs_root
*extent_root
= info
->extent_root
;
1670 struct btrfs_extent_item
*extent_item
;
1671 struct btrfs_extent_ref
*ref
;
1672 struct btrfs_path
*path
;
1673 struct btrfs_key keys
[2];
1675 new_hint
= max(hint_byte
, root
->fs_info
->alloc_start
);
1676 if (new_hint
< btrfs_super_total_bytes(&info
->super_copy
))
1677 hint_byte
= new_hint
;
1679 WARN_ON(num_bytes
< root
->sectorsize
);
1680 ret
= find_free_extent(trans
, root
, num_bytes
, empty_size
,
1681 search_start
, search_end
, hint_byte
, ins
,
1682 trans
->alloc_exclude_start
,
1683 trans
->alloc_exclude_nr
, data
);
1688 /* block accounting for super block */
1689 super_used
= btrfs_super_bytes_used(&info
->super_copy
);
1690 btrfs_set_super_bytes_used(&info
->super_copy
, super_used
+ num_bytes
);
1692 /* block accounting for root item */
1693 root_used
= btrfs_root_used(&root
->root_item
);
1694 btrfs_set_root_used(&root
->root_item
, root_used
+ num_bytes
);
1696 clear_extent_dirty(&root
->fs_info
->free_space_cache
,
1697 ins
->objectid
, ins
->objectid
+ ins
->offset
- 1,
1700 if (root
== extent_root
) {
1701 set_extent_bits(&root
->fs_info
->extent_ins
, ins
->objectid
,
1702 ins
->objectid
+ ins
->offset
- 1,
1703 EXTENT_LOCKED
, GFP_NOFS
);
1708 WARN_ON(trans
->alloc_exclude_nr
);
1709 trans
->alloc_exclude_start
= ins
->objectid
;
1710 trans
->alloc_exclude_nr
= ins
->offset
;
1712 memcpy(&keys
[0], ins
, sizeof(*ins
));
1713 keys
[1].offset
= hash_extent_ref(root_objectid
, ref_generation
,
1714 owner
, owner_offset
);
1715 keys
[1].objectid
= ins
->objectid
;
1716 keys
[1].type
= BTRFS_EXTENT_REF_KEY
;
1717 sizes
[0] = sizeof(*extent_item
);
1718 sizes
[1] = sizeof(*ref
);
1720 path
= btrfs_alloc_path();
1723 ret
= btrfs_insert_empty_items(trans
, extent_root
, path
, keys
,
1727 extent_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1728 struct btrfs_extent_item
);
1729 btrfs_set_extent_refs(path
->nodes
[0], extent_item
, 1);
1730 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
1731 struct btrfs_extent_ref
);
1733 btrfs_set_ref_root(path
->nodes
[0], ref
, root_objectid
);
1734 btrfs_set_ref_generation(path
->nodes
[0], ref
, ref_generation
);
1735 btrfs_set_ref_objectid(path
->nodes
[0], ref
, owner
);
1736 btrfs_set_ref_offset(path
->nodes
[0], ref
, owner_offset
);
1738 btrfs_mark_buffer_dirty(path
->nodes
[0]);
1740 trans
->alloc_exclude_start
= 0;
1741 trans
->alloc_exclude_nr
= 0;
1742 btrfs_free_path(path
);
1743 finish_current_insert(trans
, extent_root
);
1744 pending_ret
= del_pending_extents(trans
, extent_root
);
1754 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1, 0,
1761 * helper function to allocate a block for a given tree
1762 * returns the tree buffer or NULL.
1764 struct extent_buffer
*btrfs_alloc_free_block(struct btrfs_trans_handle
*trans
,
1765 struct btrfs_root
*root
,
1767 u64 root_objectid
, u64 hint
,
1773 ref_generation
= trans
->transid
;
1778 return __btrfs_alloc_free_block(trans
, root
, blocksize
, root_objectid
,
1779 ref_generation
, 0, 0, hint
, empty_size
);
1783 * helper function to allocate a block for a given tree
1784 * returns the tree buffer or NULL.
1786 struct extent_buffer
*__btrfs_alloc_free_block(struct btrfs_trans_handle
*trans
,
1787 struct btrfs_root
*root
,
1796 struct btrfs_key ins
;
1798 struct extent_buffer
*buf
;
1800 ret
= btrfs_alloc_extent(trans
, root
, blocksize
,
1801 root_objectid
, ref_generation
,
1802 level
, first_objectid
, empty_size
, hint
,
1806 return ERR_PTR(ret
);
1808 buf
= btrfs_find_create_tree_block(root
, ins
.objectid
, blocksize
);
1810 btrfs_free_extent(trans
, root
, ins
.objectid
, blocksize
,
1811 root
->root_key
.objectid
, ref_generation
,
1813 return ERR_PTR(-ENOMEM
);
1815 btrfs_set_header_generation(buf
, trans
->transid
);
1816 clean_tree_block(trans
, root
, buf
);
1817 wait_on_tree_block_writeback(root
, buf
);
1818 btrfs_set_buffer_uptodate(buf
);
1820 if (PageDirty(buf
->first_page
)) {
1821 printk("page %lu dirty\n", buf
->first_page
->index
);
1825 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
1826 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
1827 set_extent_bits(&BTRFS_I(root
->fs_info
->btree_inode
)->io_tree
,
1828 buf
->start
, buf
->start
+ buf
->len
- 1,
1829 EXTENT_CSUM
, GFP_NOFS
);
1830 buf
->flags
|= EXTENT_CSUM
;
1831 if (!btrfs_test_opt(root
, SSD
))
1832 btrfs_set_buffer_defrag(buf
);
1833 trans
->blocks_used
++;
1837 static int noinline
drop_leaf_ref(struct btrfs_trans_handle
*trans
,
1838 struct btrfs_root
*root
,
1839 struct extent_buffer
*leaf
)
1842 u64 leaf_generation
;
1843 struct btrfs_key key
;
1844 struct btrfs_file_extent_item
*fi
;
1849 BUG_ON(!btrfs_is_leaf(leaf
));
1850 nritems
= btrfs_header_nritems(leaf
);
1851 leaf_owner
= btrfs_header_owner(leaf
);
1852 leaf_generation
= btrfs_header_generation(leaf
);
1854 for (i
= 0; i
< nritems
; i
++) {
1857 btrfs_item_key_to_cpu(leaf
, &key
, i
);
1858 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
1860 fi
= btrfs_item_ptr(leaf
, i
, struct btrfs_file_extent_item
);
1861 if (btrfs_file_extent_type(leaf
, fi
) ==
1862 BTRFS_FILE_EXTENT_INLINE
)
1865 * FIXME make sure to insert a trans record that
1866 * repeats the snapshot del on crash
1868 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
1869 if (disk_bytenr
== 0)
1871 ret
= btrfs_free_extent(trans
, root
, disk_bytenr
,
1872 btrfs_file_extent_disk_num_bytes(leaf
, fi
),
1873 leaf_owner
, leaf_generation
,
1874 key
.objectid
, key
.offset
, 0);
1880 static void noinline
reada_walk_down(struct btrfs_root
*root
,
1881 struct extent_buffer
*node
,
1894 nritems
= btrfs_header_nritems(node
);
1895 level
= btrfs_header_level(node
);
1899 for (i
= slot
; i
< nritems
&& skipped
< 32; i
++) {
1900 bytenr
= btrfs_node_blockptr(node
, i
);
1901 if (last
&& ((bytenr
> last
&& bytenr
- last
> 32 * 1024) ||
1902 (last
> bytenr
&& last
- bytenr
> 32 * 1024))) {
1906 blocksize
= btrfs_level_size(root
, level
- 1);
1908 ret
= lookup_extent_ref(NULL
, root
, bytenr
,
1916 mutex_unlock(&root
->fs_info
->fs_mutex
);
1917 ret
= readahead_tree_block(root
, bytenr
, blocksize
);
1918 last
= bytenr
+ blocksize
;
1920 mutex_lock(&root
->fs_info
->fs_mutex
);
1927 * helper function for drop_snapshot, this walks down the tree dropping ref
1928 * counts as it goes.
1930 static int noinline
walk_down_tree(struct btrfs_trans_handle
*trans
,
1931 struct btrfs_root
*root
,
1932 struct btrfs_path
*path
, int *level
)
1937 struct extent_buffer
*next
;
1938 struct extent_buffer
*cur
;
1939 struct extent_buffer
*parent
;
1944 WARN_ON(*level
< 0);
1945 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
1946 ret
= lookup_extent_ref(trans
, root
,
1947 path
->nodes
[*level
]->start
,
1948 path
->nodes
[*level
]->len
, &refs
);
1954 * walk down to the last node level and free all the leaves
1956 while(*level
>= 0) {
1957 WARN_ON(*level
< 0);
1958 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
1959 cur
= path
->nodes
[*level
];
1961 if (btrfs_header_level(cur
) != *level
)
1964 if (path
->slots
[*level
] >=
1965 btrfs_header_nritems(cur
))
1968 ret
= drop_leaf_ref(trans
, root
, cur
);
1972 bytenr
= btrfs_node_blockptr(cur
, path
->slots
[*level
]);
1973 blocksize
= btrfs_level_size(root
, *level
- 1);
1974 ret
= lookup_extent_ref(trans
, root
, bytenr
, blocksize
, &refs
);
1977 parent
= path
->nodes
[*level
];
1978 root_owner
= btrfs_header_owner(parent
);
1979 root_gen
= btrfs_header_generation(parent
);
1980 path
->slots
[*level
]++;
1981 ret
= btrfs_free_extent(trans
, root
, bytenr
,
1982 blocksize
, root_owner
,
1987 next
= btrfs_find_tree_block(root
, bytenr
, blocksize
);
1988 if (!next
|| !btrfs_buffer_uptodate(next
)) {
1989 free_extent_buffer(next
);
1990 reada_walk_down(root
, cur
, path
->slots
[*level
]);
1991 mutex_unlock(&root
->fs_info
->fs_mutex
);
1992 next
= read_tree_block(root
, bytenr
, blocksize
);
1993 mutex_lock(&root
->fs_info
->fs_mutex
);
1995 /* we dropped the lock, check one more time */
1996 ret
= lookup_extent_ref(trans
, root
, bytenr
,
2000 parent
= path
->nodes
[*level
];
2001 root_owner
= btrfs_header_owner(parent
);
2002 root_gen
= btrfs_header_generation(parent
);
2004 path
->slots
[*level
]++;
2005 free_extent_buffer(next
);
2006 ret
= btrfs_free_extent(trans
, root
, bytenr
,
2014 WARN_ON(*level
<= 0);
2015 if (path
->nodes
[*level
-1])
2016 free_extent_buffer(path
->nodes
[*level
-1]);
2017 path
->nodes
[*level
-1] = next
;
2018 *level
= btrfs_header_level(next
);
2019 path
->slots
[*level
] = 0;
2022 WARN_ON(*level
< 0);
2023 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
2025 if (path
->nodes
[*level
] == root
->node
) {
2026 root_owner
= root
->root_key
.objectid
;
2027 parent
= path
->nodes
[*level
];
2029 parent
= path
->nodes
[*level
+ 1];
2030 root_owner
= btrfs_header_owner(parent
);
2033 root_gen
= btrfs_header_generation(parent
);
2034 ret
= btrfs_free_extent(trans
, root
, path
->nodes
[*level
]->start
,
2035 path
->nodes
[*level
]->len
,
2036 root_owner
, root_gen
, 0, 0, 1);
2037 free_extent_buffer(path
->nodes
[*level
]);
2038 path
->nodes
[*level
] = NULL
;
2045 * helper for dropping snapshots. This walks back up the tree in the path
2046 * to find the first node higher up where we haven't yet gone through
2049 static int noinline
walk_up_tree(struct btrfs_trans_handle
*trans
,
2050 struct btrfs_root
*root
,
2051 struct btrfs_path
*path
, int *level
)
2055 struct btrfs_root_item
*root_item
= &root
->root_item
;
2060 for(i
= *level
; i
< BTRFS_MAX_LEVEL
- 1 && path
->nodes
[i
]; i
++) {
2061 slot
= path
->slots
[i
];
2062 if (slot
< btrfs_header_nritems(path
->nodes
[i
]) - 1) {
2063 struct extent_buffer
*node
;
2064 struct btrfs_disk_key disk_key
;
2065 node
= path
->nodes
[i
];
2068 WARN_ON(*level
== 0);
2069 btrfs_node_key(node
, &disk_key
, path
->slots
[i
]);
2070 memcpy(&root_item
->drop_progress
,
2071 &disk_key
, sizeof(disk_key
));
2072 root_item
->drop_level
= i
;
2075 if (path
->nodes
[*level
] == root
->node
) {
2076 root_owner
= root
->root_key
.objectid
;
2078 btrfs_header_generation(path
->nodes
[*level
]);
2080 struct extent_buffer
*node
;
2081 node
= path
->nodes
[*level
+ 1];
2082 root_owner
= btrfs_header_owner(node
);
2083 root_gen
= btrfs_header_generation(node
);
2085 ret
= btrfs_free_extent(trans
, root
,
2086 path
->nodes
[*level
]->start
,
2087 path
->nodes
[*level
]->len
,
2088 root_owner
, root_gen
, 0, 0, 1);
2090 free_extent_buffer(path
->nodes
[*level
]);
2091 path
->nodes
[*level
] = NULL
;
2099 * drop the reference count on the tree rooted at 'snap'. This traverses
2100 * the tree freeing any blocks that have a ref count of zero after being
2103 int btrfs_drop_snapshot(struct btrfs_trans_handle
*trans
, struct btrfs_root
2109 struct btrfs_path
*path
;
2112 struct btrfs_root_item
*root_item
= &root
->root_item
;
2114 path
= btrfs_alloc_path();
2117 level
= btrfs_header_level(root
->node
);
2119 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
2120 path
->nodes
[level
] = root
->node
;
2121 extent_buffer_get(root
->node
);
2122 path
->slots
[level
] = 0;
2124 struct btrfs_key key
;
2125 struct btrfs_disk_key found_key
;
2126 struct extent_buffer
*node
;
2128 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
2129 level
= root_item
->drop_level
;
2130 path
->lowest_level
= level
;
2131 wret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2136 node
= path
->nodes
[level
];
2137 btrfs_node_key(node
, &found_key
, path
->slots
[level
]);
2138 WARN_ON(memcmp(&found_key
, &root_item
->drop_progress
,
2139 sizeof(found_key
)));
2142 wret
= walk_down_tree(trans
, root
, path
, &level
);
2148 wret
= walk_up_tree(trans
, root
, path
, &level
);
2156 for (i
= 0; i
<= orig_level
; i
++) {
2157 if (path
->nodes
[i
]) {
2158 free_extent_buffer(path
->nodes
[i
]);
2159 path
->nodes
[i
] = NULL
;
2163 btrfs_free_path(path
);
2167 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
2174 ret
= find_first_extent_bit(&info
->block_group_cache
, 0,
2175 &start
, &end
, (unsigned int)-1);
2178 ret
= get_state_private(&info
->block_group_cache
, start
, &ptr
);
2180 kfree((void *)(unsigned long)ptr
);
2181 clear_extent_bits(&info
->block_group_cache
, start
,
2182 end
, (unsigned int)-1, GFP_NOFS
);
2185 ret
= find_first_extent_bit(&info
->free_space_cache
, 0,
2186 &start
, &end
, EXTENT_DIRTY
);
2189 clear_extent_dirty(&info
->free_space_cache
, start
,
2195 static int noinline
relocate_inode_pages(struct inode
*inode
, u64 start
,
2201 u64 existing_delalloc
;
2202 unsigned long last_index
;
2205 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2206 struct file_ra_state
*ra
;
2208 ra
= kzalloc(sizeof(*ra
), GFP_NOFS
);
2210 mutex_lock(&inode
->i_mutex
);
2211 i
= start
>> PAGE_CACHE_SHIFT
;
2212 last_index
= (start
+ len
- 1) >> PAGE_CACHE_SHIFT
;
2214 file_ra_state_init(ra
, inode
->i_mapping
);
2215 btrfs_force_ra(inode
->i_mapping
, ra
, NULL
, i
, last_index
);
2218 for (; i
<= last_index
; i
++) {
2219 page
= grab_cache_page(inode
->i_mapping
, i
);
2222 if (!PageUptodate(page
)) {
2223 btrfs_readpage(NULL
, page
);
2225 if (!PageUptodate(page
)) {
2227 page_cache_release(page
);
2231 page_start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
2232 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
2234 lock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
2236 delalloc_start
= page_start
;
2237 existing_delalloc
= count_range_bits(io_tree
,
2238 &delalloc_start
, page_end
,
2239 PAGE_CACHE_SIZE
, EXTENT_DELALLOC
);
2241 set_extent_delalloc(io_tree
, page_start
,
2242 page_end
, GFP_NOFS
);
2244 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
2245 set_page_dirty(page
);
2247 page_cache_release(page
);
2251 mutex_unlock(&inode
->i_mutex
);
2256 * note, this releases the path
2258 static int noinline
relocate_one_reference(struct btrfs_root
*extent_root
,
2259 struct btrfs_path
*path
,
2260 struct btrfs_key
*extent_key
)
2262 struct inode
*inode
;
2263 struct btrfs_root
*found_root
;
2264 struct btrfs_key
*root_location
;
2265 struct btrfs_extent_ref
*ref
;
2272 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2273 struct btrfs_extent_ref
);
2274 ref_root
= btrfs_ref_root(path
->nodes
[0], ref
);
2275 ref_gen
= btrfs_ref_generation(path
->nodes
[0], ref
);
2276 ref_objectid
= btrfs_ref_objectid(path
->nodes
[0], ref
);
2277 ref_offset
= btrfs_ref_offset(path
->nodes
[0], ref
);
2278 btrfs_release_path(extent_root
, path
);
2280 root_location
= kmalloc(sizeof(*root_location
), GFP_NOFS
);
2281 root_location
->objectid
= ref_root
;
2283 root_location
->offset
= 0;
2285 root_location
->offset
= (u64
)-1;
2286 root_location
->type
= BTRFS_ROOT_ITEM_KEY
;
2288 found_root
= btrfs_read_fs_root_no_name(extent_root
->fs_info
,
2290 BUG_ON(!found_root
);
2291 kfree(root_location
);
2293 if (ref_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
2294 mutex_unlock(&extent_root
->fs_info
->fs_mutex
);
2295 inode
= btrfs_iget_locked(extent_root
->fs_info
->sb
,
2296 ref_objectid
, found_root
);
2297 if (inode
->i_state
& I_NEW
) {
2298 /* the inode and parent dir are two different roots */
2299 BTRFS_I(inode
)->root
= found_root
;
2300 BTRFS_I(inode
)->location
.objectid
= ref_objectid
;
2301 BTRFS_I(inode
)->location
.type
= BTRFS_INODE_ITEM_KEY
;
2302 BTRFS_I(inode
)->location
.offset
= 0;
2303 btrfs_read_locked_inode(inode
);
2304 unlock_new_inode(inode
);
2307 /* this can happen if the reference is not against
2308 * the latest version of the tree root
2310 if (is_bad_inode(inode
)) {
2311 mutex_lock(&extent_root
->fs_info
->fs_mutex
);
2314 relocate_inode_pages(inode
, ref_offset
, extent_key
->offset
);
2315 /* FIXME, data=ordered will help get rid of this */
2316 filemap_fdatawrite(inode
->i_mapping
);
2318 mutex_lock(&extent_root
->fs_info
->fs_mutex
);
2320 struct btrfs_trans_handle
*trans
;
2321 struct btrfs_key found_key
;
2322 struct extent_buffer
*eb
;
2326 trans
= btrfs_start_transaction(found_root
, 1);
2327 eb
= read_tree_block(found_root
, extent_key
->objectid
,
2328 extent_key
->offset
);
2329 level
= btrfs_header_level(eb
);
2332 btrfs_item_key_to_cpu(eb
, &found_key
, 0);
2334 btrfs_node_key_to_cpu(eb
, &found_key
, 0);
2336 free_extent_buffer(eb
);
2338 path
->lowest_level
= level
;
2340 ret
= btrfs_search_slot(trans
, found_root
, &found_key
, path
,
2342 path
->lowest_level
= 0;
2343 for (i
= level
; i
< BTRFS_MAX_LEVEL
; i
++) {
2344 if (!path
->nodes
[i
])
2346 free_extent_buffer(path
->nodes
[i
]);
2347 path
->nodes
[i
] = NULL
;
2349 btrfs_release_path(found_root
, path
);
2350 btrfs_end_transaction(trans
, found_root
);
2357 static int noinline
relocate_one_extent(struct btrfs_root
*extent_root
,
2358 struct btrfs_path
*path
,
2359 struct btrfs_key
*extent_key
)
2361 struct btrfs_key key
;
2362 struct btrfs_key found_key
;
2363 struct extent_buffer
*leaf
;
2368 key
.objectid
= extent_key
->objectid
;
2369 key
.type
= BTRFS_EXTENT_REF_KEY
;
2373 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
2379 leaf
= path
->nodes
[0];
2380 nritems
= btrfs_header_nritems(leaf
);
2381 if (path
->slots
[0] == nritems
)
2384 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2385 if (found_key
.objectid
!= extent_key
->objectid
)
2388 if (found_key
.type
!= BTRFS_EXTENT_REF_KEY
)
2391 key
.offset
= found_key
.offset
+ 1;
2392 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2394 ret
= relocate_one_reference(extent_root
, path
, extent_key
);
2400 btrfs_release_path(extent_root
, path
);
2404 int btrfs_shrink_extent_tree(struct btrfs_root
*root
, u64 new_size
)
2406 struct btrfs_trans_handle
*trans
;
2407 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
2408 struct btrfs_path
*path
;
2411 struct btrfs_fs_info
*info
= root
->fs_info
;
2412 struct extent_io_tree
*block_group_cache
;
2413 struct btrfs_key key
;
2414 struct btrfs_key found_key
;
2415 struct extent_buffer
*leaf
;
2420 btrfs_set_super_total_bytes(&info
->super_copy
, new_size
);
2421 clear_extent_dirty(&info
->free_space_cache
, new_size
, (u64
)-1,
2423 block_group_cache
= &info
->block_group_cache
;
2424 path
= btrfs_alloc_path();
2425 root
= root
->fs_info
->extent_root
;
2430 key
.objectid
= new_size
;
2433 cur_byte
= key
.objectid
;
2435 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2439 ret
= find_previous_extent(root
, path
);
2443 leaf
= path
->nodes
[0];
2444 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2445 if (found_key
.objectid
+ found_key
.offset
> new_size
) {
2446 cur_byte
= found_key
.objectid
;
2447 key
.objectid
= cur_byte
;
2450 btrfs_release_path(root
, path
);
2453 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2457 leaf
= path
->nodes
[0];
2458 nritems
= btrfs_header_nritems(leaf
);
2460 if (path
->slots
[0] >= nritems
) {
2461 ret
= btrfs_next_leaf(root
, path
);
2468 leaf
= path
->nodes
[0];
2469 nritems
= btrfs_header_nritems(leaf
);
2472 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2474 if (progress
&& need_resched()) {
2475 memcpy(&key
, &found_key
, sizeof(key
));
2476 mutex_unlock(&root
->fs_info
->fs_mutex
);
2478 mutex_lock(&root
->fs_info
->fs_mutex
);
2479 btrfs_release_path(root
, path
);
2480 btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2486 if (btrfs_key_type(&found_key
) != BTRFS_EXTENT_ITEM_KEY
||
2487 found_key
.objectid
+ found_key
.offset
<= cur_byte
) {
2493 cur_byte
= found_key
.objectid
+ found_key
.offset
;
2494 key
.objectid
= cur_byte
;
2495 btrfs_release_path(root
, path
);
2496 ret
= relocate_one_extent(root
, path
, &found_key
);
2499 btrfs_release_path(root
, path
);
2501 if (total_found
> 0) {
2502 trans
= btrfs_start_transaction(tree_root
, 1);
2503 btrfs_commit_transaction(trans
, tree_root
);
2505 mutex_unlock(&root
->fs_info
->fs_mutex
);
2506 btrfs_clean_old_snapshots(tree_root
);
2507 mutex_lock(&root
->fs_info
->fs_mutex
);
2509 trans
= btrfs_start_transaction(tree_root
, 1);
2510 btrfs_commit_transaction(trans
, tree_root
);
2514 trans
= btrfs_start_transaction(root
, 1);
2515 key
.objectid
= new_size
;
2521 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2525 leaf
= path
->nodes
[0];
2526 nritems
= btrfs_header_nritems(leaf
);
2528 if (path
->slots
[0] >= nritems
) {
2529 ret
= btrfs_next_leaf(root
, path
);
2536 leaf
= path
->nodes
[0];
2537 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2540 * btrfs_next_leaf doesn't cow buffers, we have to
2541 * do the search again
2543 memcpy(&key
, &found_key
, sizeof(key
));
2544 btrfs_release_path(root
, path
);
2548 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2549 if (btrfs_key_type(&found_key
) != BTRFS_BLOCK_GROUP_ITEM_KEY
) {
2550 printk("shrinker found key %Lu %u %Lu\n",
2551 found_key
.objectid
, found_key
.type
,
2556 ret
= get_state_private(&info
->block_group_cache
,
2557 found_key
.objectid
, &ptr
);
2559 kfree((void *)(unsigned long)ptr
);
2561 clear_extent_bits(&info
->block_group_cache
, found_key
.objectid
,
2562 found_key
.objectid
+ found_key
.offset
- 1,
2563 (unsigned int)-1, GFP_NOFS
);
2565 key
.objectid
= found_key
.objectid
+ 1;
2566 btrfs_del_item(trans
, root
, path
);
2567 btrfs_release_path(root
, path
);
2569 if (need_resched()) {
2570 mutex_unlock(&root
->fs_info
->fs_mutex
);
2572 mutex_lock(&root
->fs_info
->fs_mutex
);
2575 clear_extent_dirty(&info
->free_space_cache
, new_size
, (u64
)-1,
2577 btrfs_commit_transaction(trans
, root
);
2579 btrfs_free_path(path
);
2583 int btrfs_grow_extent_tree(struct btrfs_trans_handle
*trans
,
2584 struct btrfs_root
*root
, u64 new_size
)
2586 struct btrfs_path
*path
;
2591 struct btrfs_block_group_cache
*cache
;
2592 struct btrfs_block_group_item
*item
;
2593 struct btrfs_fs_info
*info
= root
->fs_info
;
2594 struct extent_io_tree
*block_group_cache
;
2595 struct btrfs_key key
;
2596 struct extent_buffer
*leaf
;
2600 old_size
= btrfs_super_total_bytes(&info
->super_copy
);
2601 block_group_cache
= &info
->block_group_cache
;
2603 root
= info
->extent_root
;
2605 cache
= btrfs_lookup_block_group(root
->fs_info
, old_size
- 1);
2607 cur_byte
= cache
->key
.objectid
+ cache
->key
.offset
;
2608 if (cur_byte
>= new_size
)
2611 key
.offset
= BTRFS_BLOCK_GROUP_SIZE
;
2612 btrfs_set_key_type(&key
, BTRFS_BLOCK_GROUP_ITEM_KEY
);
2614 path
= btrfs_alloc_path();
2618 while(cur_byte
< new_size
) {
2619 key
.objectid
= cur_byte
;
2620 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
2621 sizeof(struct btrfs_block_group_item
));
2623 leaf
= path
->nodes
[0];
2624 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2625 struct btrfs_block_group_item
);
2627 btrfs_set_disk_block_group_used(leaf
, item
, 0);
2628 div_long_long_rem(nr
, 3, &rem
);
2630 btrfs_set_disk_block_group_flags(leaf
, item
,
2631 BTRFS_BLOCK_GROUP_DATA
);
2633 btrfs_set_disk_block_group_flags(leaf
, item
, 0);
2637 cache
= kmalloc(sizeof(*cache
), GFP_NOFS
);
2640 read_extent_buffer(leaf
, &cache
->item
, (unsigned long)item
,
2641 sizeof(cache
->item
));
2643 memcpy(&cache
->key
, &key
, sizeof(key
));
2646 cur_byte
= key
.objectid
+ key
.offset
;
2647 btrfs_release_path(root
, path
);
2649 if (cache
->item
.flags
& BTRFS_BLOCK_GROUP_DATA
) {
2650 bit
= BLOCK_GROUP_DATA
;
2651 cache
->data
= BTRFS_BLOCK_GROUP_DATA
;
2653 bit
= BLOCK_GROUP_METADATA
;
2657 /* use EXTENT_LOCKED to prevent merging */
2658 set_extent_bits(block_group_cache
, key
.objectid
,
2659 key
.objectid
+ key
.offset
- 1,
2660 bit
| EXTENT_LOCKED
, GFP_NOFS
);
2661 set_state_private(block_group_cache
, key
.objectid
,
2662 (unsigned long)cache
);
2664 btrfs_free_path(path
);
2666 btrfs_set_super_total_bytes(&info
->super_copy
, new_size
);
2670 int btrfs_read_block_groups(struct btrfs_root
*root
)
2672 struct btrfs_path
*path
;
2676 struct btrfs_block_group_cache
*cache
;
2677 struct btrfs_fs_info
*info
= root
->fs_info
;
2678 struct extent_io_tree
*block_group_cache
;
2679 struct btrfs_key key
;
2680 struct btrfs_key found_key
;
2681 struct extent_buffer
*leaf
;
2683 block_group_cache
= &info
->block_group_cache
;
2685 root
= info
->extent_root
;
2687 key
.offset
= BTRFS_BLOCK_GROUP_SIZE
;
2688 btrfs_set_key_type(&key
, BTRFS_BLOCK_GROUP_ITEM_KEY
);
2690 path
= btrfs_alloc_path();
2695 ret
= btrfs_search_slot(NULL
, info
->extent_root
,
2701 leaf
= path
->nodes
[0];
2702 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2703 cache
= kmalloc(sizeof(*cache
), GFP_NOFS
);
2709 read_extent_buffer(leaf
, &cache
->item
,
2710 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
2711 sizeof(cache
->item
));
2712 memcpy(&cache
->key
, &found_key
, sizeof(found_key
));
2715 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
2716 btrfs_release_path(root
, path
);
2718 if (cache
->item
.flags
& BTRFS_BLOCK_GROUP_MIXED
) {
2719 bit
= BLOCK_GROUP_DATA
| BLOCK_GROUP_METADATA
;
2720 cache
->data
= BTRFS_BLOCK_GROUP_MIXED
;
2721 } else if (cache
->item
.flags
& BTRFS_BLOCK_GROUP_DATA
) {
2722 bit
= BLOCK_GROUP_DATA
;
2723 cache
->data
= BTRFS_BLOCK_GROUP_DATA
;
2725 bit
= BLOCK_GROUP_METADATA
;
2729 /* use EXTENT_LOCKED to prevent merging */
2730 set_extent_bits(block_group_cache
, found_key
.objectid
,
2731 found_key
.objectid
+ found_key
.offset
- 1,
2732 bit
| EXTENT_LOCKED
, GFP_NOFS
);
2733 set_state_private(block_group_cache
, found_key
.objectid
,
2734 (unsigned long)cache
);
2737 btrfs_super_total_bytes(&info
->super_copy
))
2741 btrfs_free_path(path
);