2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
28 #include "print-tree.h"
29 #include "transaction.h"
32 #include "free-space-cache.h"
34 static int update_reserved_extents(struct btrfs_root
*root
,
35 u64 bytenr
, u64 num
, int reserve
);
36 static int update_block_group(struct btrfs_trans_handle
*trans
,
37 struct btrfs_root
*root
,
38 u64 bytenr
, u64 num_bytes
, int alloc
,
40 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
41 struct btrfs_root
*root
,
42 u64 bytenr
, u64 num_bytes
, u64 parent
,
43 u64 root_objectid
, u64 owner_objectid
,
44 u64 owner_offset
, int refs_to_drop
,
45 struct btrfs_delayed_extent_op
*extra_op
);
46 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
47 struct extent_buffer
*leaf
,
48 struct btrfs_extent_item
*ei
);
49 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
50 struct btrfs_root
*root
,
51 u64 parent
, u64 root_objectid
,
52 u64 flags
, u64 owner
, u64 offset
,
53 struct btrfs_key
*ins
, int ref_mod
);
54 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
55 struct btrfs_root
*root
,
56 u64 parent
, u64 root_objectid
,
57 u64 flags
, struct btrfs_disk_key
*key
,
58 int level
, struct btrfs_key
*ins
);
60 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
61 struct btrfs_root
*extent_root
, u64 alloc_bytes
,
62 u64 flags
, int force
);
64 static int block_group_bits(struct btrfs_block_group_cache
*cache
, u64 bits
)
66 return (cache
->flags
& bits
) == bits
;
70 * this adds the block group to the fs_info rb tree for the block group
73 static int btrfs_add_block_group_cache(struct btrfs_fs_info
*info
,
74 struct btrfs_block_group_cache
*block_group
)
77 struct rb_node
*parent
= NULL
;
78 struct btrfs_block_group_cache
*cache
;
80 spin_lock(&info
->block_group_cache_lock
);
81 p
= &info
->block_group_cache_tree
.rb_node
;
85 cache
= rb_entry(parent
, struct btrfs_block_group_cache
,
87 if (block_group
->key
.objectid
< cache
->key
.objectid
) {
89 } else if (block_group
->key
.objectid
> cache
->key
.objectid
) {
92 spin_unlock(&info
->block_group_cache_lock
);
97 rb_link_node(&block_group
->cache_node
, parent
, p
);
98 rb_insert_color(&block_group
->cache_node
,
99 &info
->block_group_cache_tree
);
100 spin_unlock(&info
->block_group_cache_lock
);
106 * This will return the block group at or after bytenr if contains is 0, else
107 * it will return the block group that contains the bytenr
109 static struct btrfs_block_group_cache
*
110 block_group_cache_tree_search(struct btrfs_fs_info
*info
, u64 bytenr
,
113 struct btrfs_block_group_cache
*cache
, *ret
= NULL
;
117 spin_lock(&info
->block_group_cache_lock
);
118 n
= info
->block_group_cache_tree
.rb_node
;
121 cache
= rb_entry(n
, struct btrfs_block_group_cache
,
123 end
= cache
->key
.objectid
+ cache
->key
.offset
- 1;
124 start
= cache
->key
.objectid
;
126 if (bytenr
< start
) {
127 if (!contains
&& (!ret
|| start
< ret
->key
.objectid
))
130 } else if (bytenr
> start
) {
131 if (contains
&& bytenr
<= end
) {
142 atomic_inc(&ret
->count
);
143 spin_unlock(&info
->block_group_cache_lock
);
149 * this is only called by cache_block_group, since we could have freed extents
150 * we need to check the pinned_extents for any extents that can't be used yet
151 * since their free space will be released as soon as the transaction commits.
153 static int add_new_free_space(struct btrfs_block_group_cache
*block_group
,
154 struct btrfs_fs_info
*info
, u64 start
, u64 end
)
156 u64 extent_start
, extent_end
, size
;
159 while (start
< end
) {
160 ret
= find_first_extent_bit(&info
->pinned_extents
, start
,
161 &extent_start
, &extent_end
,
166 if (extent_start
== start
) {
167 start
= extent_end
+ 1;
168 } else if (extent_start
> start
&& extent_start
< end
) {
169 size
= extent_start
- start
;
170 ret
= btrfs_add_free_space(block_group
, start
,
173 start
= extent_end
+ 1;
181 ret
= btrfs_add_free_space(block_group
, start
, size
);
188 static int remove_sb_from_cache(struct btrfs_root
*root
,
189 struct btrfs_block_group_cache
*cache
)
196 for (i
= 0; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
197 bytenr
= btrfs_sb_offset(i
);
198 ret
= btrfs_rmap_block(&root
->fs_info
->mapping_tree
,
199 cache
->key
.objectid
, bytenr
, 0,
200 &logical
, &nr
, &stripe_len
);
203 btrfs_remove_free_space(cache
, logical
[nr
],
211 static int cache_block_group(struct btrfs_root
*root
,
212 struct btrfs_block_group_cache
*block_group
)
214 struct btrfs_path
*path
;
216 struct btrfs_key key
;
217 struct extent_buffer
*leaf
;
224 root
= root
->fs_info
->extent_root
;
226 if (block_group
->cached
)
229 path
= btrfs_alloc_path();
235 * we get into deadlocks with paths held by callers of this function.
236 * since the alloc_mutex is protecting things right now, just
237 * skip the locking here
239 path
->skip_locking
= 1;
240 last
= max_t(u64
, block_group
->key
.objectid
, BTRFS_SUPER_INFO_OFFSET
);
243 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
244 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
249 leaf
= path
->nodes
[0];
250 slot
= path
->slots
[0];
251 if (slot
>= btrfs_header_nritems(leaf
)) {
252 ret
= btrfs_next_leaf(root
, path
);
260 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
261 if (key
.objectid
< block_group
->key
.objectid
)
264 if (key
.objectid
>= block_group
->key
.objectid
+
265 block_group
->key
.offset
)
268 if (btrfs_key_type(&key
) == BTRFS_EXTENT_ITEM_KEY
) {
269 add_new_free_space(block_group
, root
->fs_info
, last
,
272 last
= key
.objectid
+ key
.offset
;
278 add_new_free_space(block_group
, root
->fs_info
, last
,
279 block_group
->key
.objectid
+
280 block_group
->key
.offset
);
282 block_group
->cached
= 1;
283 remove_sb_from_cache(root
, block_group
);
286 btrfs_free_path(path
);
291 * return the block group that starts at or after bytenr
293 static struct btrfs_block_group_cache
*
294 btrfs_lookup_first_block_group(struct btrfs_fs_info
*info
, u64 bytenr
)
296 struct btrfs_block_group_cache
*cache
;
298 cache
= block_group_cache_tree_search(info
, bytenr
, 0);
304 * return the block group that contains the given bytenr
306 struct btrfs_block_group_cache
*btrfs_lookup_block_group(
307 struct btrfs_fs_info
*info
,
310 struct btrfs_block_group_cache
*cache
;
312 cache
= block_group_cache_tree_search(info
, bytenr
, 1);
317 void btrfs_put_block_group(struct btrfs_block_group_cache
*cache
)
319 if (atomic_dec_and_test(&cache
->count
))
323 static struct btrfs_space_info
*__find_space_info(struct btrfs_fs_info
*info
,
326 struct list_head
*head
= &info
->space_info
;
327 struct btrfs_space_info
*found
;
330 list_for_each_entry_rcu(found
, head
, list
) {
331 if (found
->flags
== flags
) {
341 * after adding space to the filesystem, we need to clear the full flags
342 * on all the space infos.
344 void btrfs_clear_space_info_full(struct btrfs_fs_info
*info
)
346 struct list_head
*head
= &info
->space_info
;
347 struct btrfs_space_info
*found
;
350 list_for_each_entry_rcu(found
, head
, list
)
355 static u64
div_factor(u64 num
, int factor
)
364 u64
btrfs_find_block_group(struct btrfs_root
*root
,
365 u64 search_start
, u64 search_hint
, int owner
)
367 struct btrfs_block_group_cache
*cache
;
369 u64 last
= max(search_hint
, search_start
);
376 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
380 spin_lock(&cache
->lock
);
381 last
= cache
->key
.objectid
+ cache
->key
.offset
;
382 used
= btrfs_block_group_used(&cache
->item
);
384 if ((full_search
|| !cache
->ro
) &&
385 block_group_bits(cache
, BTRFS_BLOCK_GROUP_METADATA
)) {
386 if (used
+ cache
->pinned
+ cache
->reserved
<
387 div_factor(cache
->key
.offset
, factor
)) {
388 group_start
= cache
->key
.objectid
;
389 spin_unlock(&cache
->lock
);
390 btrfs_put_block_group(cache
);
394 spin_unlock(&cache
->lock
);
395 btrfs_put_block_group(cache
);
403 if (!full_search
&& factor
< 10) {
413 /* simple helper to search for an existing extent at a given offset */
414 int btrfs_lookup_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
417 struct btrfs_key key
;
418 struct btrfs_path
*path
;
420 path
= btrfs_alloc_path();
422 key
.objectid
= start
;
424 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
425 ret
= btrfs_search_slot(NULL
, root
->fs_info
->extent_root
, &key
, path
,
427 btrfs_free_path(path
);
432 * Back reference rules. Back refs have three main goals:
434 * 1) differentiate between all holders of references to an extent so that
435 * when a reference is dropped we can make sure it was a valid reference
436 * before freeing the extent.
438 * 2) Provide enough information to quickly find the holders of an extent
439 * if we notice a given block is corrupted or bad.
441 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
442 * maintenance. This is actually the same as #2, but with a slightly
443 * different use case.
445 * There are two kinds of back refs. The implicit back refs is optimized
446 * for pointers in non-shared tree blocks. For a given pointer in a block,
447 * back refs of this kind provide information about the block's owner tree
448 * and the pointer's key. These information allow us to find the block by
449 * b-tree searching. The full back refs is for pointers in tree blocks not
450 * referenced by their owner trees. The location of tree block is recorded
451 * in the back refs. Actually the full back refs is generic, and can be
452 * used in all cases the implicit back refs is used. The major shortcoming
453 * of the full back refs is its overhead. Every time a tree block gets
454 * COWed, we have to update back refs entry for all pointers in it.
456 * For a newly allocated tree block, we use implicit back refs for
457 * pointers in it. This means most tree related operations only involve
458 * implicit back refs. For a tree block created in old transaction, the
459 * only way to drop a reference to it is COW it. So we can detect the
460 * event that tree block loses its owner tree's reference and do the
461 * back refs conversion.
463 * When a tree block is COW'd through a tree, there are four cases:
465 * The reference count of the block is one and the tree is the block's
466 * owner tree. Nothing to do in this case.
468 * The reference count of the block is one and the tree is not the
469 * block's owner tree. In this case, full back refs is used for pointers
470 * in the block. Remove these full back refs, add implicit back refs for
471 * every pointers in the new block.
473 * The reference count of the block is greater than one and the tree is
474 * the block's owner tree. In this case, implicit back refs is used for
475 * pointers in the block. Add full back refs for every pointers in the
476 * block, increase lower level extents' reference counts. The original
477 * implicit back refs are entailed to the new block.
479 * The reference count of the block is greater than one and the tree is
480 * not the block's owner tree. Add implicit back refs for every pointer in
481 * the new block, increase lower level extents' reference count.
483 * Back Reference Key composing:
485 * The key objectid corresponds to the first byte in the extent,
486 * The key type is used to differentiate between types of back refs.
487 * There are different meanings of the key offset for different types
490 * File extents can be referenced by:
492 * - multiple snapshots, subvolumes, or different generations in one subvol
493 * - different files inside a single subvolume
494 * - different offsets inside a file (bookend extents in file.c)
496 * The extent ref structure for the implicit back refs has fields for:
498 * - Objectid of the subvolume root
499 * - objectid of the file holding the reference
500 * - original offset in the file
501 * - how many bookend extents
503 * The key offset for the implicit back refs is hash of the first
506 * The extent ref structure for the full back refs has field for:
508 * - number of pointers in the tree leaf
510 * The key offset for the implicit back refs is the first byte of
513 * When a file extent is allocated, The implicit back refs is used.
514 * the fields are filled in:
516 * (root_key.objectid, inode objectid, offset in file, 1)
518 * When a file extent is removed file truncation, we find the
519 * corresponding implicit back refs and check the following fields:
521 * (btrfs_header_owner(leaf), inode objectid, offset in file)
523 * Btree extents can be referenced by:
525 * - Different subvolumes
527 * Both the implicit back refs and the full back refs for tree blocks
528 * only consist of key. The key offset for the implicit back refs is
529 * objectid of block's owner tree. The key offset for the full back refs
530 * is the first byte of parent block.
532 * When implicit back refs is used, information about the lowest key and
533 * level of the tree block are required. These information are stored in
534 * tree block info structure.
537 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
538 static int convert_extent_item_v0(struct btrfs_trans_handle
*trans
,
539 struct btrfs_root
*root
,
540 struct btrfs_path
*path
,
541 u64 owner
, u32 extra_size
)
543 struct btrfs_extent_item
*item
;
544 struct btrfs_extent_item_v0
*ei0
;
545 struct btrfs_extent_ref_v0
*ref0
;
546 struct btrfs_tree_block_info
*bi
;
547 struct extent_buffer
*leaf
;
548 struct btrfs_key key
;
549 struct btrfs_key found_key
;
550 u32 new_size
= sizeof(*item
);
554 leaf
= path
->nodes
[0];
555 BUG_ON(btrfs_item_size_nr(leaf
, path
->slots
[0]) != sizeof(*ei0
));
557 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
558 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
559 struct btrfs_extent_item_v0
);
560 refs
= btrfs_extent_refs_v0(leaf
, ei0
);
562 if (owner
== (u64
)-1) {
564 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
565 ret
= btrfs_next_leaf(root
, path
);
569 leaf
= path
->nodes
[0];
571 btrfs_item_key_to_cpu(leaf
, &found_key
,
573 BUG_ON(key
.objectid
!= found_key
.objectid
);
574 if (found_key
.type
!= BTRFS_EXTENT_REF_V0_KEY
) {
578 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
579 struct btrfs_extent_ref_v0
);
580 owner
= btrfs_ref_objectid_v0(leaf
, ref0
);
584 btrfs_release_path(root
, path
);
586 if (owner
< BTRFS_FIRST_FREE_OBJECTID
)
587 new_size
+= sizeof(*bi
);
589 new_size
-= sizeof(*ei0
);
590 ret
= btrfs_search_slot(trans
, root
, &key
, path
,
591 new_size
+ extra_size
, 1);
596 ret
= btrfs_extend_item(trans
, root
, path
, new_size
);
599 leaf
= path
->nodes
[0];
600 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
601 btrfs_set_extent_refs(leaf
, item
, refs
);
602 /* FIXME: get real generation */
603 btrfs_set_extent_generation(leaf
, item
, 0);
604 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
605 btrfs_set_extent_flags(leaf
, item
,
606 BTRFS_EXTENT_FLAG_TREE_BLOCK
|
607 BTRFS_BLOCK_FLAG_FULL_BACKREF
);
608 bi
= (struct btrfs_tree_block_info
*)(item
+ 1);
609 /* FIXME: get first key of the block */
610 memset_extent_buffer(leaf
, 0, (unsigned long)bi
, sizeof(*bi
));
611 btrfs_set_tree_block_level(leaf
, bi
, (int)owner
);
613 btrfs_set_extent_flags(leaf
, item
, BTRFS_EXTENT_FLAG_DATA
);
615 btrfs_mark_buffer_dirty(leaf
);
620 static u64
hash_extent_data_ref(u64 root_objectid
, u64 owner
, u64 offset
)
622 u32 high_crc
= ~(u32
)0;
623 u32 low_crc
= ~(u32
)0;
626 lenum
= cpu_to_le64(root_objectid
);
627 high_crc
= crc32c(high_crc
, &lenum
, sizeof(lenum
));
628 lenum
= cpu_to_le64(owner
);
629 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
630 lenum
= cpu_to_le64(offset
);
631 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
633 return ((u64
)high_crc
<< 31) ^ (u64
)low_crc
;
636 static u64
hash_extent_data_ref_item(struct extent_buffer
*leaf
,
637 struct btrfs_extent_data_ref
*ref
)
639 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf
, ref
),
640 btrfs_extent_data_ref_objectid(leaf
, ref
),
641 btrfs_extent_data_ref_offset(leaf
, ref
));
644 static int match_extent_data_ref(struct extent_buffer
*leaf
,
645 struct btrfs_extent_data_ref
*ref
,
646 u64 root_objectid
, u64 owner
, u64 offset
)
648 if (btrfs_extent_data_ref_root(leaf
, ref
) != root_objectid
||
649 btrfs_extent_data_ref_objectid(leaf
, ref
) != owner
||
650 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
655 static noinline
int lookup_extent_data_ref(struct btrfs_trans_handle
*trans
,
656 struct btrfs_root
*root
,
657 struct btrfs_path
*path
,
658 u64 bytenr
, u64 parent
,
660 u64 owner
, u64 offset
)
662 struct btrfs_key key
;
663 struct btrfs_extent_data_ref
*ref
;
664 struct extent_buffer
*leaf
;
670 key
.objectid
= bytenr
;
672 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
675 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
676 key
.offset
= hash_extent_data_ref(root_objectid
,
681 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
690 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
691 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
692 btrfs_release_path(root
, path
);
693 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
704 leaf
= path
->nodes
[0];
705 nritems
= btrfs_header_nritems(leaf
);
707 if (path
->slots
[0] >= nritems
) {
708 ret
= btrfs_next_leaf(root
, path
);
714 leaf
= path
->nodes
[0];
715 nritems
= btrfs_header_nritems(leaf
);
719 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
720 if (key
.objectid
!= bytenr
||
721 key
.type
!= BTRFS_EXTENT_DATA_REF_KEY
)
724 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
725 struct btrfs_extent_data_ref
);
727 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
730 btrfs_release_path(root
, path
);
742 static noinline
int insert_extent_data_ref(struct btrfs_trans_handle
*trans
,
743 struct btrfs_root
*root
,
744 struct btrfs_path
*path
,
745 u64 bytenr
, u64 parent
,
746 u64 root_objectid
, u64 owner
,
747 u64 offset
, int refs_to_add
)
749 struct btrfs_key key
;
750 struct extent_buffer
*leaf
;
755 key
.objectid
= bytenr
;
757 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
759 size
= sizeof(struct btrfs_shared_data_ref
);
761 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
762 key
.offset
= hash_extent_data_ref(root_objectid
,
764 size
= sizeof(struct btrfs_extent_data_ref
);
767 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, size
);
768 if (ret
&& ret
!= -EEXIST
)
771 leaf
= path
->nodes
[0];
773 struct btrfs_shared_data_ref
*ref
;
774 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
775 struct btrfs_shared_data_ref
);
777 btrfs_set_shared_data_ref_count(leaf
, ref
, refs_to_add
);
779 num_refs
= btrfs_shared_data_ref_count(leaf
, ref
);
780 num_refs
+= refs_to_add
;
781 btrfs_set_shared_data_ref_count(leaf
, ref
, num_refs
);
784 struct btrfs_extent_data_ref
*ref
;
785 while (ret
== -EEXIST
) {
786 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
787 struct btrfs_extent_data_ref
);
788 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
791 btrfs_release_path(root
, path
);
793 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
795 if (ret
&& ret
!= -EEXIST
)
798 leaf
= path
->nodes
[0];
800 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
801 struct btrfs_extent_data_ref
);
803 btrfs_set_extent_data_ref_root(leaf
, ref
,
805 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
806 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
807 btrfs_set_extent_data_ref_count(leaf
, ref
, refs_to_add
);
809 num_refs
= btrfs_extent_data_ref_count(leaf
, ref
);
810 num_refs
+= refs_to_add
;
811 btrfs_set_extent_data_ref_count(leaf
, ref
, num_refs
);
814 btrfs_mark_buffer_dirty(leaf
);
817 btrfs_release_path(root
, path
);
821 static noinline
int remove_extent_data_ref(struct btrfs_trans_handle
*trans
,
822 struct btrfs_root
*root
,
823 struct btrfs_path
*path
,
826 struct btrfs_key key
;
827 struct btrfs_extent_data_ref
*ref1
= NULL
;
828 struct btrfs_shared_data_ref
*ref2
= NULL
;
829 struct extent_buffer
*leaf
;
833 leaf
= path
->nodes
[0];
834 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
836 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
837 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
838 struct btrfs_extent_data_ref
);
839 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
840 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
841 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
842 struct btrfs_shared_data_ref
);
843 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
844 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
845 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
846 struct btrfs_extent_ref_v0
*ref0
;
847 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
848 struct btrfs_extent_ref_v0
);
849 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
855 BUG_ON(num_refs
< refs_to_drop
);
856 num_refs
-= refs_to_drop
;
859 ret
= btrfs_del_item(trans
, root
, path
);
861 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
)
862 btrfs_set_extent_data_ref_count(leaf
, ref1
, num_refs
);
863 else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
)
864 btrfs_set_shared_data_ref_count(leaf
, ref2
, num_refs
);
865 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
867 struct btrfs_extent_ref_v0
*ref0
;
868 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
869 struct btrfs_extent_ref_v0
);
870 btrfs_set_ref_count_v0(leaf
, ref0
, num_refs
);
873 btrfs_mark_buffer_dirty(leaf
);
878 static noinline u32
extent_data_ref_count(struct btrfs_root
*root
,
879 struct btrfs_path
*path
,
880 struct btrfs_extent_inline_ref
*iref
)
882 struct btrfs_key key
;
883 struct extent_buffer
*leaf
;
884 struct btrfs_extent_data_ref
*ref1
;
885 struct btrfs_shared_data_ref
*ref2
;
888 leaf
= path
->nodes
[0];
889 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
891 if (btrfs_extent_inline_ref_type(leaf
, iref
) ==
892 BTRFS_EXTENT_DATA_REF_KEY
) {
893 ref1
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
894 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
896 ref2
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
897 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
899 } else if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
900 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
901 struct btrfs_extent_data_ref
);
902 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
903 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
904 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
905 struct btrfs_shared_data_ref
);
906 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
907 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
908 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
909 struct btrfs_extent_ref_v0
*ref0
;
910 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
911 struct btrfs_extent_ref_v0
);
912 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
920 static noinline
int lookup_tree_block_ref(struct btrfs_trans_handle
*trans
,
921 struct btrfs_root
*root
,
922 struct btrfs_path
*path
,
923 u64 bytenr
, u64 parent
,
926 struct btrfs_key key
;
929 key
.objectid
= bytenr
;
931 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
934 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
935 key
.offset
= root_objectid
;
938 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
941 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
942 if (ret
== -ENOENT
&& parent
) {
943 btrfs_release_path(root
, path
);
944 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
945 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
953 static noinline
int insert_tree_block_ref(struct btrfs_trans_handle
*trans
,
954 struct btrfs_root
*root
,
955 struct btrfs_path
*path
,
956 u64 bytenr
, u64 parent
,
959 struct btrfs_key key
;
962 key
.objectid
= bytenr
;
964 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
967 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
968 key
.offset
= root_objectid
;
971 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
972 btrfs_release_path(root
, path
);
976 static inline int extent_ref_type(u64 parent
, u64 owner
)
979 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
981 type
= BTRFS_SHARED_BLOCK_REF_KEY
;
983 type
= BTRFS_TREE_BLOCK_REF_KEY
;
986 type
= BTRFS_SHARED_DATA_REF_KEY
;
988 type
= BTRFS_EXTENT_DATA_REF_KEY
;
993 static int find_next_key(struct btrfs_path
*path
, int level
,
994 struct btrfs_key
*key
)
997 for (; level
< BTRFS_MAX_LEVEL
; level
++) {
998 if (!path
->nodes
[level
])
1000 if (path
->slots
[level
] + 1 >=
1001 btrfs_header_nritems(path
->nodes
[level
]))
1004 btrfs_item_key_to_cpu(path
->nodes
[level
], key
,
1005 path
->slots
[level
] + 1);
1007 btrfs_node_key_to_cpu(path
->nodes
[level
], key
,
1008 path
->slots
[level
] + 1);
1015 * look for inline back ref. if back ref is found, *ref_ret is set
1016 * to the address of inline back ref, and 0 is returned.
1018 * if back ref isn't found, *ref_ret is set to the address where it
1019 * should be inserted, and -ENOENT is returned.
1021 * if insert is true and there are too many inline back refs, the path
1022 * points to the extent item, and -EAGAIN is returned.
1024 * NOTE: inline back refs are ordered in the same way that back ref
1025 * items in the tree are ordered.
1027 static noinline_for_stack
1028 int lookup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1029 struct btrfs_root
*root
,
1030 struct btrfs_path
*path
,
1031 struct btrfs_extent_inline_ref
**ref_ret
,
1032 u64 bytenr
, u64 num_bytes
,
1033 u64 parent
, u64 root_objectid
,
1034 u64 owner
, u64 offset
, int insert
)
1036 struct btrfs_key key
;
1037 struct extent_buffer
*leaf
;
1038 struct btrfs_extent_item
*ei
;
1039 struct btrfs_extent_inline_ref
*iref
;
1050 key
.objectid
= bytenr
;
1051 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1052 key
.offset
= num_bytes
;
1054 want
= extent_ref_type(parent
, owner
);
1056 extra_size
= btrfs_extent_inline_ref_size(want
);
1057 path
->keep_locks
= 1;
1060 ret
= btrfs_search_slot(trans
, root
, &key
, path
, extra_size
, 1);
1067 leaf
= path
->nodes
[0];
1068 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1069 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1070 if (item_size
< sizeof(*ei
)) {
1075 ret
= convert_extent_item_v0(trans
, root
, path
, owner
,
1081 leaf
= path
->nodes
[0];
1082 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1085 BUG_ON(item_size
< sizeof(*ei
));
1087 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1088 flags
= btrfs_extent_flags(leaf
, ei
);
1090 ptr
= (unsigned long)(ei
+ 1);
1091 end
= (unsigned long)ei
+ item_size
;
1093 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
) {
1094 ptr
+= sizeof(struct btrfs_tree_block_info
);
1097 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_DATA
));
1106 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1107 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1111 ptr
+= btrfs_extent_inline_ref_size(type
);
1115 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1116 struct btrfs_extent_data_ref
*dref
;
1117 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1118 if (match_extent_data_ref(leaf
, dref
, root_objectid
,
1123 if (hash_extent_data_ref_item(leaf
, dref
) <
1124 hash_extent_data_ref(root_objectid
, owner
, offset
))
1128 ref_offset
= btrfs_extent_inline_ref_offset(leaf
, iref
);
1130 if (parent
== ref_offset
) {
1134 if (ref_offset
< parent
)
1137 if (root_objectid
== ref_offset
) {
1141 if (ref_offset
< root_objectid
)
1145 ptr
+= btrfs_extent_inline_ref_size(type
);
1147 if (err
== -ENOENT
&& insert
) {
1148 if (item_size
+ extra_size
>=
1149 BTRFS_MAX_EXTENT_ITEM_SIZE(root
)) {
1154 * To add new inline back ref, we have to make sure
1155 * there is no corresponding back ref item.
1156 * For simplicity, we just do not add new inline back
1157 * ref if there is any kind of item for this block
1159 if (find_next_key(path
, 0, &key
) == 0 &&
1160 key
.objectid
== bytenr
&&
1161 key
.type
< BTRFS_BLOCK_GROUP_ITEM_KEY
) {
1166 *ref_ret
= (struct btrfs_extent_inline_ref
*)ptr
;
1169 path
->keep_locks
= 0;
1170 btrfs_unlock_up_safe(path
, 1);
1176 * helper to add new inline back ref
1178 static noinline_for_stack
1179 int setup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1180 struct btrfs_root
*root
,
1181 struct btrfs_path
*path
,
1182 struct btrfs_extent_inline_ref
*iref
,
1183 u64 parent
, u64 root_objectid
,
1184 u64 owner
, u64 offset
, int refs_to_add
,
1185 struct btrfs_delayed_extent_op
*extent_op
)
1187 struct extent_buffer
*leaf
;
1188 struct btrfs_extent_item
*ei
;
1191 unsigned long item_offset
;
1197 leaf
= path
->nodes
[0];
1198 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1199 item_offset
= (unsigned long)iref
- (unsigned long)ei
;
1201 type
= extent_ref_type(parent
, owner
);
1202 size
= btrfs_extent_inline_ref_size(type
);
1204 ret
= btrfs_extend_item(trans
, root
, path
, size
);
1207 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1208 refs
= btrfs_extent_refs(leaf
, ei
);
1209 refs
+= refs_to_add
;
1210 btrfs_set_extent_refs(leaf
, ei
, refs
);
1212 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1214 ptr
= (unsigned long)ei
+ item_offset
;
1215 end
= (unsigned long)ei
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
1216 if (ptr
< end
- size
)
1217 memmove_extent_buffer(leaf
, ptr
+ size
, ptr
,
1220 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1221 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
1222 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1223 struct btrfs_extent_data_ref
*dref
;
1224 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1225 btrfs_set_extent_data_ref_root(leaf
, dref
, root_objectid
);
1226 btrfs_set_extent_data_ref_objectid(leaf
, dref
, owner
);
1227 btrfs_set_extent_data_ref_offset(leaf
, dref
, offset
);
1228 btrfs_set_extent_data_ref_count(leaf
, dref
, refs_to_add
);
1229 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1230 struct btrfs_shared_data_ref
*sref
;
1231 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1232 btrfs_set_shared_data_ref_count(leaf
, sref
, refs_to_add
);
1233 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1234 } else if (type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
1235 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1237 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
1239 btrfs_mark_buffer_dirty(leaf
);
1243 static int lookup_extent_backref(struct btrfs_trans_handle
*trans
,
1244 struct btrfs_root
*root
,
1245 struct btrfs_path
*path
,
1246 struct btrfs_extent_inline_ref
**ref_ret
,
1247 u64 bytenr
, u64 num_bytes
, u64 parent
,
1248 u64 root_objectid
, u64 owner
, u64 offset
)
1252 ret
= lookup_inline_extent_backref(trans
, root
, path
, ref_ret
,
1253 bytenr
, num_bytes
, parent
,
1254 root_objectid
, owner
, offset
, 0);
1258 btrfs_release_path(root
, path
);
1261 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1262 ret
= lookup_tree_block_ref(trans
, root
, path
, bytenr
, parent
,
1265 ret
= lookup_extent_data_ref(trans
, root
, path
, bytenr
, parent
,
1266 root_objectid
, owner
, offset
);
1272 * helper to update/remove inline back ref
1274 static noinline_for_stack
1275 int update_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1276 struct btrfs_root
*root
,
1277 struct btrfs_path
*path
,
1278 struct btrfs_extent_inline_ref
*iref
,
1280 struct btrfs_delayed_extent_op
*extent_op
)
1282 struct extent_buffer
*leaf
;
1283 struct btrfs_extent_item
*ei
;
1284 struct btrfs_extent_data_ref
*dref
= NULL
;
1285 struct btrfs_shared_data_ref
*sref
= NULL
;
1294 leaf
= path
->nodes
[0];
1295 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1296 refs
= btrfs_extent_refs(leaf
, ei
);
1297 WARN_ON(refs_to_mod
< 0 && refs
+ refs_to_mod
<= 0);
1298 refs
+= refs_to_mod
;
1299 btrfs_set_extent_refs(leaf
, ei
, refs
);
1301 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1303 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1305 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1306 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1307 refs
= btrfs_extent_data_ref_count(leaf
, dref
);
1308 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1309 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1310 refs
= btrfs_shared_data_ref_count(leaf
, sref
);
1313 BUG_ON(refs_to_mod
!= -1);
1316 BUG_ON(refs_to_mod
< 0 && refs
< -refs_to_mod
);
1317 refs
+= refs_to_mod
;
1320 if (type
== BTRFS_EXTENT_DATA_REF_KEY
)
1321 btrfs_set_extent_data_ref_count(leaf
, dref
, refs
);
1323 btrfs_set_shared_data_ref_count(leaf
, sref
, refs
);
1325 size
= btrfs_extent_inline_ref_size(type
);
1326 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1327 ptr
= (unsigned long)iref
;
1328 end
= (unsigned long)ei
+ item_size
;
1329 if (ptr
+ size
< end
)
1330 memmove_extent_buffer(leaf
, ptr
, ptr
+ size
,
1333 ret
= btrfs_truncate_item(trans
, root
, path
, item_size
, 1);
1336 btrfs_mark_buffer_dirty(leaf
);
1340 static noinline_for_stack
1341 int insert_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1342 struct btrfs_root
*root
,
1343 struct btrfs_path
*path
,
1344 u64 bytenr
, u64 num_bytes
, u64 parent
,
1345 u64 root_objectid
, u64 owner
,
1346 u64 offset
, int refs_to_add
,
1347 struct btrfs_delayed_extent_op
*extent_op
)
1349 struct btrfs_extent_inline_ref
*iref
;
1352 ret
= lookup_inline_extent_backref(trans
, root
, path
, &iref
,
1353 bytenr
, num_bytes
, parent
,
1354 root_objectid
, owner
, offset
, 1);
1356 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
);
1357 ret
= update_inline_extent_backref(trans
, root
, path
, iref
,
1358 refs_to_add
, extent_op
);
1359 } else if (ret
== -ENOENT
) {
1360 ret
= setup_inline_extent_backref(trans
, root
, path
, iref
,
1361 parent
, root_objectid
,
1362 owner
, offset
, refs_to_add
,
1368 static int insert_extent_backref(struct btrfs_trans_handle
*trans
,
1369 struct btrfs_root
*root
,
1370 struct btrfs_path
*path
,
1371 u64 bytenr
, u64 parent
, u64 root_objectid
,
1372 u64 owner
, u64 offset
, int refs_to_add
)
1375 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1376 BUG_ON(refs_to_add
!= 1);
1377 ret
= insert_tree_block_ref(trans
, root
, path
, bytenr
,
1378 parent
, root_objectid
);
1380 ret
= insert_extent_data_ref(trans
, root
, path
, bytenr
,
1381 parent
, root_objectid
,
1382 owner
, offset
, refs_to_add
);
1387 static int remove_extent_backref(struct btrfs_trans_handle
*trans
,
1388 struct btrfs_root
*root
,
1389 struct btrfs_path
*path
,
1390 struct btrfs_extent_inline_ref
*iref
,
1391 int refs_to_drop
, int is_data
)
1395 BUG_ON(!is_data
&& refs_to_drop
!= 1);
1397 ret
= update_inline_extent_backref(trans
, root
, path
, iref
,
1398 -refs_to_drop
, NULL
);
1399 } else if (is_data
) {
1400 ret
= remove_extent_data_ref(trans
, root
, path
, refs_to_drop
);
1402 ret
= btrfs_del_item(trans
, root
, path
);
1407 #ifdef BIO_RW_DISCARD
1408 static void btrfs_issue_discard(struct block_device
*bdev
,
1411 blkdev_issue_discard(bdev
, start
>> 9, len
>> 9, GFP_KERNEL
);
1415 static int btrfs_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
1418 #ifdef BIO_RW_DISCARD
1420 u64 map_length
= num_bytes
;
1421 struct btrfs_multi_bio
*multi
= NULL
;
1423 /* Tell the block device(s) that the sectors can be discarded */
1424 ret
= btrfs_map_block(&root
->fs_info
->mapping_tree
, READ
,
1425 bytenr
, &map_length
, &multi
, 0);
1427 struct btrfs_bio_stripe
*stripe
= multi
->stripes
;
1430 if (map_length
> num_bytes
)
1431 map_length
= num_bytes
;
1433 for (i
= 0; i
< multi
->num_stripes
; i
++, stripe
++) {
1434 btrfs_issue_discard(stripe
->dev
->bdev
,
1447 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1448 struct btrfs_root
*root
,
1449 u64 bytenr
, u64 num_bytes
, u64 parent
,
1450 u64 root_objectid
, u64 owner
, u64 offset
)
1453 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
&&
1454 root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
1456 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1457 ret
= btrfs_add_delayed_tree_ref(trans
, bytenr
, num_bytes
,
1458 parent
, root_objectid
, (int)owner
,
1459 BTRFS_ADD_DELAYED_REF
, NULL
);
1461 ret
= btrfs_add_delayed_data_ref(trans
, bytenr
, num_bytes
,
1462 parent
, root_objectid
, owner
, offset
,
1463 BTRFS_ADD_DELAYED_REF
, NULL
);
1468 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1469 struct btrfs_root
*root
,
1470 u64 bytenr
, u64 num_bytes
,
1471 u64 parent
, u64 root_objectid
,
1472 u64 owner
, u64 offset
, int refs_to_add
,
1473 struct btrfs_delayed_extent_op
*extent_op
)
1475 struct btrfs_path
*path
;
1476 struct extent_buffer
*leaf
;
1477 struct btrfs_extent_item
*item
;
1482 path
= btrfs_alloc_path();
1487 path
->leave_spinning
= 1;
1488 /* this will setup the path even if it fails to insert the back ref */
1489 ret
= insert_inline_extent_backref(trans
, root
->fs_info
->extent_root
,
1490 path
, bytenr
, num_bytes
, parent
,
1491 root_objectid
, owner
, offset
,
1492 refs_to_add
, extent_op
);
1496 if (ret
!= -EAGAIN
) {
1501 leaf
= path
->nodes
[0];
1502 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1503 refs
= btrfs_extent_refs(leaf
, item
);
1504 btrfs_set_extent_refs(leaf
, item
, refs
+ refs_to_add
);
1506 __run_delayed_extent_op(extent_op
, leaf
, item
);
1508 btrfs_mark_buffer_dirty(leaf
);
1509 btrfs_release_path(root
->fs_info
->extent_root
, path
);
1512 path
->leave_spinning
= 1;
1514 /* now insert the actual backref */
1515 ret
= insert_extent_backref(trans
, root
->fs_info
->extent_root
,
1516 path
, bytenr
, parent
, root_objectid
,
1517 owner
, offset
, refs_to_add
);
1520 btrfs_free_path(path
);
1524 static int run_delayed_data_ref(struct btrfs_trans_handle
*trans
,
1525 struct btrfs_root
*root
,
1526 struct btrfs_delayed_ref_node
*node
,
1527 struct btrfs_delayed_extent_op
*extent_op
,
1528 int insert_reserved
)
1531 struct btrfs_delayed_data_ref
*ref
;
1532 struct btrfs_key ins
;
1537 ins
.objectid
= node
->bytenr
;
1538 ins
.offset
= node
->num_bytes
;
1539 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1541 ref
= btrfs_delayed_node_to_data_ref(node
);
1542 if (node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
1543 parent
= ref
->parent
;
1545 ref_root
= ref
->root
;
1547 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
1549 BUG_ON(extent_op
->update_key
);
1550 flags
|= extent_op
->flags_to_set
;
1552 ret
= alloc_reserved_file_extent(trans
, root
,
1553 parent
, ref_root
, flags
,
1554 ref
->objectid
, ref
->offset
,
1555 &ins
, node
->ref_mod
);
1556 update_reserved_extents(root
, ins
.objectid
, ins
.offset
, 0);
1557 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
1558 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
1559 node
->num_bytes
, parent
,
1560 ref_root
, ref
->objectid
,
1561 ref
->offset
, node
->ref_mod
,
1563 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
1564 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
1565 node
->num_bytes
, parent
,
1566 ref_root
, ref
->objectid
,
1567 ref
->offset
, node
->ref_mod
,
1575 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
1576 struct extent_buffer
*leaf
,
1577 struct btrfs_extent_item
*ei
)
1579 u64 flags
= btrfs_extent_flags(leaf
, ei
);
1580 if (extent_op
->update_flags
) {
1581 flags
|= extent_op
->flags_to_set
;
1582 btrfs_set_extent_flags(leaf
, ei
, flags
);
1585 if (extent_op
->update_key
) {
1586 struct btrfs_tree_block_info
*bi
;
1587 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
));
1588 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
1589 btrfs_set_tree_block_key(leaf
, bi
, &extent_op
->key
);
1593 static int run_delayed_extent_op(struct btrfs_trans_handle
*trans
,
1594 struct btrfs_root
*root
,
1595 struct btrfs_delayed_ref_node
*node
,
1596 struct btrfs_delayed_extent_op
*extent_op
)
1598 struct btrfs_key key
;
1599 struct btrfs_path
*path
;
1600 struct btrfs_extent_item
*ei
;
1601 struct extent_buffer
*leaf
;
1606 path
= btrfs_alloc_path();
1610 key
.objectid
= node
->bytenr
;
1611 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1612 key
.offset
= node
->num_bytes
;
1615 path
->leave_spinning
= 1;
1616 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
,
1627 leaf
= path
->nodes
[0];
1628 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1629 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1630 if (item_size
< sizeof(*ei
)) {
1631 ret
= convert_extent_item_v0(trans
, root
->fs_info
->extent_root
,
1637 leaf
= path
->nodes
[0];
1638 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1641 BUG_ON(item_size
< sizeof(*ei
));
1642 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1643 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1645 btrfs_mark_buffer_dirty(leaf
);
1647 btrfs_free_path(path
);
1651 static int run_delayed_tree_ref(struct btrfs_trans_handle
*trans
,
1652 struct btrfs_root
*root
,
1653 struct btrfs_delayed_ref_node
*node
,
1654 struct btrfs_delayed_extent_op
*extent_op
,
1655 int insert_reserved
)
1658 struct btrfs_delayed_tree_ref
*ref
;
1659 struct btrfs_key ins
;
1663 ins
.objectid
= node
->bytenr
;
1664 ins
.offset
= node
->num_bytes
;
1665 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1667 ref
= btrfs_delayed_node_to_tree_ref(node
);
1668 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
1669 parent
= ref
->parent
;
1671 ref_root
= ref
->root
;
1673 BUG_ON(node
->ref_mod
!= 1);
1674 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
1675 BUG_ON(!extent_op
|| !extent_op
->update_flags
||
1676 !extent_op
->update_key
);
1677 ret
= alloc_reserved_tree_block(trans
, root
,
1679 extent_op
->flags_to_set
,
1682 update_reserved_extents(root
, ins
.objectid
, ins
.offset
, 0);
1683 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
1684 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
1685 node
->num_bytes
, parent
, ref_root
,
1686 ref
->level
, 0, 1, extent_op
);
1687 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
1688 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
1689 node
->num_bytes
, parent
, ref_root
,
1690 ref
->level
, 0, 1, extent_op
);
1698 /* helper function to actually process a single delayed ref entry */
1699 static int run_one_delayed_ref(struct btrfs_trans_handle
*trans
,
1700 struct btrfs_root
*root
,
1701 struct btrfs_delayed_ref_node
*node
,
1702 struct btrfs_delayed_extent_op
*extent_op
,
1703 int insert_reserved
)
1706 if (btrfs_delayed_ref_is_head(node
)) {
1707 struct btrfs_delayed_ref_head
*head
;
1709 * we've hit the end of the chain and we were supposed
1710 * to insert this extent into the tree. But, it got
1711 * deleted before we ever needed to insert it, so all
1712 * we have to do is clean up the accounting
1715 head
= btrfs_delayed_node_to_head(node
);
1716 if (insert_reserved
) {
1717 if (head
->is_data
) {
1718 ret
= btrfs_del_csums(trans
, root
,
1723 btrfs_update_pinned_extents(root
, node
->bytenr
,
1724 node
->num_bytes
, 1);
1725 update_reserved_extents(root
, node
->bytenr
,
1726 node
->num_bytes
, 0);
1728 mutex_unlock(&head
->mutex
);
1732 if (node
->type
== BTRFS_TREE_BLOCK_REF_KEY
||
1733 node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
1734 ret
= run_delayed_tree_ref(trans
, root
, node
, extent_op
,
1736 else if (node
->type
== BTRFS_EXTENT_DATA_REF_KEY
||
1737 node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
1738 ret
= run_delayed_data_ref(trans
, root
, node
, extent_op
,
1745 static noinline
struct btrfs_delayed_ref_node
*
1746 select_delayed_ref(struct btrfs_delayed_ref_head
*head
)
1748 struct rb_node
*node
;
1749 struct btrfs_delayed_ref_node
*ref
;
1750 int action
= BTRFS_ADD_DELAYED_REF
;
1753 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
1754 * this prevents ref count from going down to zero when
1755 * there still are pending delayed ref.
1757 node
= rb_prev(&head
->node
.rb_node
);
1761 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
1763 if (ref
->bytenr
!= head
->node
.bytenr
)
1765 if (ref
->action
== action
)
1767 node
= rb_prev(node
);
1769 if (action
== BTRFS_ADD_DELAYED_REF
) {
1770 action
= BTRFS_DROP_DELAYED_REF
;
1776 static noinline
int run_clustered_refs(struct btrfs_trans_handle
*trans
,
1777 struct btrfs_root
*root
,
1778 struct list_head
*cluster
)
1780 struct btrfs_delayed_ref_root
*delayed_refs
;
1781 struct btrfs_delayed_ref_node
*ref
;
1782 struct btrfs_delayed_ref_head
*locked_ref
= NULL
;
1783 struct btrfs_delayed_extent_op
*extent_op
;
1786 int must_insert_reserved
= 0;
1788 delayed_refs
= &trans
->transaction
->delayed_refs
;
1791 /* pick a new head ref from the cluster list */
1792 if (list_empty(cluster
))
1795 locked_ref
= list_entry(cluster
->next
,
1796 struct btrfs_delayed_ref_head
, cluster
);
1798 /* grab the lock that says we are going to process
1799 * all the refs for this head */
1800 ret
= btrfs_delayed_ref_lock(trans
, locked_ref
);
1803 * we may have dropped the spin lock to get the head
1804 * mutex lock, and that might have given someone else
1805 * time to free the head. If that's true, it has been
1806 * removed from our list and we can move on.
1808 if (ret
== -EAGAIN
) {
1816 * record the must insert reserved flag before we
1817 * drop the spin lock.
1819 must_insert_reserved
= locked_ref
->must_insert_reserved
;
1820 locked_ref
->must_insert_reserved
= 0;
1822 extent_op
= locked_ref
->extent_op
;
1823 locked_ref
->extent_op
= NULL
;
1826 * locked_ref is the head node, so we have to go one
1827 * node back for any delayed ref updates
1829 ref
= select_delayed_ref(locked_ref
);
1831 /* All delayed refs have been processed, Go ahead
1832 * and send the head node to run_one_delayed_ref,
1833 * so that any accounting fixes can happen
1835 ref
= &locked_ref
->node
;
1837 if (extent_op
&& must_insert_reserved
) {
1843 spin_unlock(&delayed_refs
->lock
);
1845 ret
= run_delayed_extent_op(trans
, root
,
1851 spin_lock(&delayed_refs
->lock
);
1855 list_del_init(&locked_ref
->cluster
);
1860 rb_erase(&ref
->rb_node
, &delayed_refs
->root
);
1861 delayed_refs
->num_entries
--;
1863 spin_unlock(&delayed_refs
->lock
);
1865 ret
= run_one_delayed_ref(trans
, root
, ref
, extent_op
,
1866 must_insert_reserved
);
1869 btrfs_put_delayed_ref(ref
);
1874 spin_lock(&delayed_refs
->lock
);
1880 * this starts processing the delayed reference count updates and
1881 * extent insertions we have queued up so far. count can be
1882 * 0, which means to process everything in the tree at the start
1883 * of the run (but not newly added entries), or it can be some target
1884 * number you'd like to process.
1886 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
1887 struct btrfs_root
*root
, unsigned long count
)
1889 struct rb_node
*node
;
1890 struct btrfs_delayed_ref_root
*delayed_refs
;
1891 struct btrfs_delayed_ref_node
*ref
;
1892 struct list_head cluster
;
1894 int run_all
= count
== (unsigned long)-1;
1897 if (root
== root
->fs_info
->extent_root
)
1898 root
= root
->fs_info
->tree_root
;
1900 delayed_refs
= &trans
->transaction
->delayed_refs
;
1901 INIT_LIST_HEAD(&cluster
);
1903 spin_lock(&delayed_refs
->lock
);
1905 count
= delayed_refs
->num_entries
* 2;
1909 if (!(run_all
|| run_most
) &&
1910 delayed_refs
->num_heads_ready
< 64)
1914 * go find something we can process in the rbtree. We start at
1915 * the beginning of the tree, and then build a cluster
1916 * of refs to process starting at the first one we are able to
1919 ret
= btrfs_find_ref_cluster(trans
, &cluster
,
1920 delayed_refs
->run_delayed_start
);
1924 ret
= run_clustered_refs(trans
, root
, &cluster
);
1927 count
-= min_t(unsigned long, ret
, count
);
1934 node
= rb_first(&delayed_refs
->root
);
1937 count
= (unsigned long)-1;
1940 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
1942 if (btrfs_delayed_ref_is_head(ref
)) {
1943 struct btrfs_delayed_ref_head
*head
;
1945 head
= btrfs_delayed_node_to_head(ref
);
1946 atomic_inc(&ref
->refs
);
1948 spin_unlock(&delayed_refs
->lock
);
1949 mutex_lock(&head
->mutex
);
1950 mutex_unlock(&head
->mutex
);
1952 btrfs_put_delayed_ref(ref
);
1956 node
= rb_next(node
);
1958 spin_unlock(&delayed_refs
->lock
);
1959 schedule_timeout(1);
1963 spin_unlock(&delayed_refs
->lock
);
1967 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
1968 struct btrfs_root
*root
,
1969 u64 bytenr
, u64 num_bytes
, u64 flags
,
1972 struct btrfs_delayed_extent_op
*extent_op
;
1975 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
1979 extent_op
->flags_to_set
= flags
;
1980 extent_op
->update_flags
= 1;
1981 extent_op
->update_key
= 0;
1982 extent_op
->is_data
= is_data
? 1 : 0;
1984 ret
= btrfs_add_delayed_extent_op(trans
, bytenr
, num_bytes
, extent_op
);
1990 static noinline
int check_delayed_ref(struct btrfs_trans_handle
*trans
,
1991 struct btrfs_root
*root
,
1992 struct btrfs_path
*path
,
1993 u64 objectid
, u64 offset
, u64 bytenr
)
1995 struct btrfs_delayed_ref_head
*head
;
1996 struct btrfs_delayed_ref_node
*ref
;
1997 struct btrfs_delayed_data_ref
*data_ref
;
1998 struct btrfs_delayed_ref_root
*delayed_refs
;
1999 struct rb_node
*node
;
2003 delayed_refs
= &trans
->transaction
->delayed_refs
;
2004 spin_lock(&delayed_refs
->lock
);
2005 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
2009 if (!mutex_trylock(&head
->mutex
)) {
2010 atomic_inc(&head
->node
.refs
);
2011 spin_unlock(&delayed_refs
->lock
);
2013 btrfs_release_path(root
->fs_info
->extent_root
, path
);
2015 mutex_lock(&head
->mutex
);
2016 mutex_unlock(&head
->mutex
);
2017 btrfs_put_delayed_ref(&head
->node
);
2021 node
= rb_prev(&head
->node
.rb_node
);
2025 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2027 if (ref
->bytenr
!= bytenr
)
2031 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
)
2034 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
2036 node
= rb_prev(node
);
2038 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2039 if (ref
->bytenr
== bytenr
)
2043 if (data_ref
->root
!= root
->root_key
.objectid
||
2044 data_ref
->objectid
!= objectid
|| data_ref
->offset
!= offset
)
2049 mutex_unlock(&head
->mutex
);
2051 spin_unlock(&delayed_refs
->lock
);
2055 static noinline
int check_committed_ref(struct btrfs_trans_handle
*trans
,
2056 struct btrfs_root
*root
,
2057 struct btrfs_path
*path
,
2058 u64 objectid
, u64 offset
, u64 bytenr
)
2060 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2061 struct extent_buffer
*leaf
;
2062 struct btrfs_extent_data_ref
*ref
;
2063 struct btrfs_extent_inline_ref
*iref
;
2064 struct btrfs_extent_item
*ei
;
2065 struct btrfs_key key
;
2069 key
.objectid
= bytenr
;
2070 key
.offset
= (u64
)-1;
2071 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2073 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
2079 if (path
->slots
[0] == 0)
2083 leaf
= path
->nodes
[0];
2084 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2086 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
2090 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2091 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2092 if (item_size
< sizeof(*ei
)) {
2093 WARN_ON(item_size
!= sizeof(struct btrfs_extent_item_v0
));
2097 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2099 if (item_size
!= sizeof(*ei
) +
2100 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
2103 if (btrfs_extent_generation(leaf
, ei
) <=
2104 btrfs_root_last_snapshot(&root
->root_item
))
2107 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
2108 if (btrfs_extent_inline_ref_type(leaf
, iref
) !=
2109 BTRFS_EXTENT_DATA_REF_KEY
)
2112 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
2113 if (btrfs_extent_refs(leaf
, ei
) !=
2114 btrfs_extent_data_ref_count(leaf
, ref
) ||
2115 btrfs_extent_data_ref_root(leaf
, ref
) !=
2116 root
->root_key
.objectid
||
2117 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
2118 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
2126 int btrfs_cross_ref_exist(struct btrfs_trans_handle
*trans
,
2127 struct btrfs_root
*root
,
2128 u64 objectid
, u64 offset
, u64 bytenr
)
2130 struct btrfs_path
*path
;
2134 path
= btrfs_alloc_path();
2139 ret
= check_committed_ref(trans
, root
, path
, objectid
,
2141 if (ret
&& ret
!= -ENOENT
)
2144 ret2
= check_delayed_ref(trans
, root
, path
, objectid
,
2146 } while (ret2
== -EAGAIN
);
2148 if (ret2
&& ret2
!= -ENOENT
) {
2153 if (ret
!= -ENOENT
|| ret2
!= -ENOENT
)
2156 btrfs_free_path(path
);
2161 int btrfs_cache_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2162 struct extent_buffer
*buf
, u32 nr_extents
)
2164 struct btrfs_key key
;
2165 struct btrfs_file_extent_item
*fi
;
2173 if (!root
->ref_cows
)
2176 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
2178 root_gen
= root
->root_key
.offset
;
2181 root_gen
= trans
->transid
- 1;
2184 level
= btrfs_header_level(buf
);
2185 nritems
= btrfs_header_nritems(buf
);
2188 struct btrfs_leaf_ref
*ref
;
2189 struct btrfs_extent_info
*info
;
2191 ref
= btrfs_alloc_leaf_ref(root
, nr_extents
);
2197 ref
->root_gen
= root_gen
;
2198 ref
->bytenr
= buf
->start
;
2199 ref
->owner
= btrfs_header_owner(buf
);
2200 ref
->generation
= btrfs_header_generation(buf
);
2201 ref
->nritems
= nr_extents
;
2202 info
= ref
->extents
;
2204 for (i
= 0; nr_extents
> 0 && i
< nritems
; i
++) {
2206 btrfs_item_key_to_cpu(buf
, &key
, i
);
2207 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
2209 fi
= btrfs_item_ptr(buf
, i
,
2210 struct btrfs_file_extent_item
);
2211 if (btrfs_file_extent_type(buf
, fi
) ==
2212 BTRFS_FILE_EXTENT_INLINE
)
2214 disk_bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
2215 if (disk_bytenr
== 0)
2218 info
->bytenr
= disk_bytenr
;
2220 btrfs_file_extent_disk_num_bytes(buf
, fi
);
2221 info
->objectid
= key
.objectid
;
2222 info
->offset
= key
.offset
;
2226 ret
= btrfs_add_leaf_ref(root
, ref
, shared
);
2227 if (ret
== -EEXIST
&& shared
) {
2228 struct btrfs_leaf_ref
*old
;
2229 old
= btrfs_lookup_leaf_ref(root
, ref
->bytenr
);
2231 btrfs_remove_leaf_ref(root
, old
);
2232 btrfs_free_leaf_ref(root
, old
);
2233 ret
= btrfs_add_leaf_ref(root
, ref
, shared
);
2236 btrfs_free_leaf_ref(root
, ref
);
2242 /* when a block goes through cow, we update the reference counts of
2243 * everything that block points to. The internal pointers of the block
2244 * can be in just about any order, and it is likely to have clusters of
2245 * things that are close together and clusters of things that are not.
2247 * To help reduce the seeks that come with updating all of these reference
2248 * counts, sort them by byte number before actual updates are done.
2250 * struct refsort is used to match byte number to slot in the btree block.
2251 * we sort based on the byte number and then use the slot to actually
2254 * struct refsort is smaller than strcut btrfs_item and smaller than
2255 * struct btrfs_key_ptr. Since we're currently limited to the page size
2256 * for a btree block, there's no way for a kmalloc of refsorts for a
2257 * single node to be bigger than a page.
2265 * for passing into sort()
2267 static int refsort_cmp(const void *a_void
, const void *b_void
)
2269 const struct refsort
*a
= a_void
;
2270 const struct refsort
*b
= b_void
;
2272 if (a
->bytenr
< b
->bytenr
)
2274 if (a
->bytenr
> b
->bytenr
)
2280 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
2281 struct btrfs_root
*root
,
2282 struct extent_buffer
*buf
,
2283 int full_backref
, int inc
)
2290 struct btrfs_key key
;
2291 struct btrfs_file_extent_item
*fi
;
2295 int (*process_func
)(struct btrfs_trans_handle
*, struct btrfs_root
*,
2296 u64
, u64
, u64
, u64
, u64
, u64
);
2298 ref_root
= btrfs_header_owner(buf
);
2299 nritems
= btrfs_header_nritems(buf
);
2300 level
= btrfs_header_level(buf
);
2302 if (!root
->ref_cows
&& level
== 0)
2306 process_func
= btrfs_inc_extent_ref
;
2308 process_func
= btrfs_free_extent
;
2311 parent
= buf
->start
;
2315 for (i
= 0; i
< nritems
; i
++) {
2317 btrfs_item_key_to_cpu(buf
, &key
, i
);
2318 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
2320 fi
= btrfs_item_ptr(buf
, i
,
2321 struct btrfs_file_extent_item
);
2322 if (btrfs_file_extent_type(buf
, fi
) ==
2323 BTRFS_FILE_EXTENT_INLINE
)
2325 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
2329 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
2330 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
2331 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
2332 parent
, ref_root
, key
.objectid
,
2337 bytenr
= btrfs_node_blockptr(buf
, i
);
2338 num_bytes
= btrfs_level_size(root
, level
- 1);
2339 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
2340 parent
, ref_root
, level
- 1, 0);
2351 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2352 struct extent_buffer
*buf
, int full_backref
)
2354 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1);
2357 int btrfs_dec_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2358 struct extent_buffer
*buf
, int full_backref
)
2360 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 0);
2363 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
2364 struct btrfs_root
*root
,
2365 struct btrfs_path
*path
,
2366 struct btrfs_block_group_cache
*cache
)
2369 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2371 struct extent_buffer
*leaf
;
2373 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
2378 leaf
= path
->nodes
[0];
2379 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
2380 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
2381 btrfs_mark_buffer_dirty(leaf
);
2382 btrfs_release_path(extent_root
, path
);
2390 static struct btrfs_block_group_cache
*
2391 next_block_group(struct btrfs_root
*root
,
2392 struct btrfs_block_group_cache
*cache
)
2394 struct rb_node
*node
;
2395 spin_lock(&root
->fs_info
->block_group_cache_lock
);
2396 node
= rb_next(&cache
->cache_node
);
2397 btrfs_put_block_group(cache
);
2399 cache
= rb_entry(node
, struct btrfs_block_group_cache
,
2401 atomic_inc(&cache
->count
);
2404 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
2408 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
2409 struct btrfs_root
*root
)
2411 struct btrfs_block_group_cache
*cache
;
2413 struct btrfs_path
*path
;
2416 path
= btrfs_alloc_path();
2422 err
= btrfs_run_delayed_refs(trans
, root
,
2427 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
2431 cache
= next_block_group(root
, cache
);
2441 last
= cache
->key
.objectid
+ cache
->key
.offset
;
2443 err
= write_one_cache_group(trans
, root
, path
, cache
);
2445 btrfs_put_block_group(cache
);
2448 btrfs_free_path(path
);
2452 int btrfs_extent_readonly(struct btrfs_root
*root
, u64 bytenr
)
2454 struct btrfs_block_group_cache
*block_group
;
2457 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
2458 if (!block_group
|| block_group
->ro
)
2461 btrfs_put_block_group(block_group
);
2465 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
2466 u64 total_bytes
, u64 bytes_used
,
2467 struct btrfs_space_info
**space_info
)
2469 struct btrfs_space_info
*found
;
2471 found
= __find_space_info(info
, flags
);
2473 spin_lock(&found
->lock
);
2474 found
->total_bytes
+= total_bytes
;
2475 found
->bytes_used
+= bytes_used
;
2477 spin_unlock(&found
->lock
);
2478 *space_info
= found
;
2481 found
= kzalloc(sizeof(*found
), GFP_NOFS
);
2485 INIT_LIST_HEAD(&found
->block_groups
);
2486 init_rwsem(&found
->groups_sem
);
2487 spin_lock_init(&found
->lock
);
2488 found
->flags
= flags
;
2489 found
->total_bytes
= total_bytes
;
2490 found
->bytes_used
= bytes_used
;
2491 found
->bytes_pinned
= 0;
2492 found
->bytes_reserved
= 0;
2493 found
->bytes_readonly
= 0;
2494 found
->bytes_delalloc
= 0;
2496 found
->force_alloc
= 0;
2497 *space_info
= found
;
2498 list_add_rcu(&found
->list
, &info
->space_info
);
2502 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
2504 u64 extra_flags
= flags
& (BTRFS_BLOCK_GROUP_RAID0
|
2505 BTRFS_BLOCK_GROUP_RAID1
|
2506 BTRFS_BLOCK_GROUP_RAID10
|
2507 BTRFS_BLOCK_GROUP_DUP
);
2509 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
2510 fs_info
->avail_data_alloc_bits
|= extra_flags
;
2511 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
2512 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
2513 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
2514 fs_info
->avail_system_alloc_bits
|= extra_flags
;
2518 static void set_block_group_readonly(struct btrfs_block_group_cache
*cache
)
2520 spin_lock(&cache
->space_info
->lock
);
2521 spin_lock(&cache
->lock
);
2523 cache
->space_info
->bytes_readonly
+= cache
->key
.offset
-
2524 btrfs_block_group_used(&cache
->item
);
2527 spin_unlock(&cache
->lock
);
2528 spin_unlock(&cache
->space_info
->lock
);
2531 u64
btrfs_reduce_alloc_profile(struct btrfs_root
*root
, u64 flags
)
2533 u64 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
2535 if (num_devices
== 1)
2536 flags
&= ~(BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID0
);
2537 if (num_devices
< 4)
2538 flags
&= ~BTRFS_BLOCK_GROUP_RAID10
;
2540 if ((flags
& BTRFS_BLOCK_GROUP_DUP
) &&
2541 (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
2542 BTRFS_BLOCK_GROUP_RAID10
))) {
2543 flags
&= ~BTRFS_BLOCK_GROUP_DUP
;
2546 if ((flags
& BTRFS_BLOCK_GROUP_RAID1
) &&
2547 (flags
& BTRFS_BLOCK_GROUP_RAID10
)) {
2548 flags
&= ~BTRFS_BLOCK_GROUP_RAID1
;
2551 if ((flags
& BTRFS_BLOCK_GROUP_RAID0
) &&
2552 ((flags
& BTRFS_BLOCK_GROUP_RAID1
) |
2553 (flags
& BTRFS_BLOCK_GROUP_RAID10
) |
2554 (flags
& BTRFS_BLOCK_GROUP_DUP
)))
2555 flags
&= ~BTRFS_BLOCK_GROUP_RAID0
;
2559 static u64
btrfs_get_alloc_profile(struct btrfs_root
*root
, u64 data
)
2561 struct btrfs_fs_info
*info
= root
->fs_info
;
2565 alloc_profile
= info
->avail_data_alloc_bits
&
2566 info
->data_alloc_profile
;
2567 data
= BTRFS_BLOCK_GROUP_DATA
| alloc_profile
;
2568 } else if (root
== root
->fs_info
->chunk_root
) {
2569 alloc_profile
= info
->avail_system_alloc_bits
&
2570 info
->system_alloc_profile
;
2571 data
= BTRFS_BLOCK_GROUP_SYSTEM
| alloc_profile
;
2573 alloc_profile
= info
->avail_metadata_alloc_bits
&
2574 info
->metadata_alloc_profile
;
2575 data
= BTRFS_BLOCK_GROUP_METADATA
| alloc_profile
;
2578 return btrfs_reduce_alloc_profile(root
, data
);
2581 void btrfs_set_inode_space_info(struct btrfs_root
*root
, struct inode
*inode
)
2585 alloc_target
= btrfs_get_alloc_profile(root
, 1);
2586 BTRFS_I(inode
)->space_info
= __find_space_info(root
->fs_info
,
2591 * for now this just makes sure we have at least 5% of our metadata space free
2594 int btrfs_check_metadata_free_space(struct btrfs_root
*root
)
2596 struct btrfs_fs_info
*info
= root
->fs_info
;
2597 struct btrfs_space_info
*meta_sinfo
;
2598 u64 alloc_target
, thresh
;
2599 int committed
= 0, ret
;
2601 /* get the space info for where the metadata will live */
2602 alloc_target
= btrfs_get_alloc_profile(root
, 0);
2603 meta_sinfo
= __find_space_info(info
, alloc_target
);
2606 spin_lock(&meta_sinfo
->lock
);
2607 if (!meta_sinfo
->full
)
2608 thresh
= meta_sinfo
->total_bytes
* 80;
2610 thresh
= meta_sinfo
->total_bytes
* 95;
2612 do_div(thresh
, 100);
2614 if (meta_sinfo
->bytes_used
+ meta_sinfo
->bytes_reserved
+
2615 meta_sinfo
->bytes_pinned
+ meta_sinfo
->bytes_readonly
> thresh
) {
2616 struct btrfs_trans_handle
*trans
;
2617 if (!meta_sinfo
->full
) {
2618 meta_sinfo
->force_alloc
= 1;
2619 spin_unlock(&meta_sinfo
->lock
);
2621 trans
= btrfs_start_transaction(root
, 1);
2625 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
2626 2 * 1024 * 1024, alloc_target
, 0);
2627 btrfs_end_transaction(trans
, root
);
2630 spin_unlock(&meta_sinfo
->lock
);
2634 trans
= btrfs_join_transaction(root
, 1);
2637 ret
= btrfs_commit_transaction(trans
, root
);
2644 spin_unlock(&meta_sinfo
->lock
);
2650 * This will check the space that the inode allocates from to make sure we have
2651 * enough space for bytes.
2653 int btrfs_check_data_free_space(struct btrfs_root
*root
, struct inode
*inode
,
2656 struct btrfs_space_info
*data_sinfo
;
2657 int ret
= 0, committed
= 0;
2659 /* make sure bytes are sectorsize aligned */
2660 bytes
= (bytes
+ root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
2662 data_sinfo
= BTRFS_I(inode
)->space_info
;
2664 /* make sure we have enough space to handle the data first */
2665 spin_lock(&data_sinfo
->lock
);
2666 if (data_sinfo
->total_bytes
- data_sinfo
->bytes_used
-
2667 data_sinfo
->bytes_delalloc
- data_sinfo
->bytes_reserved
-
2668 data_sinfo
->bytes_pinned
- data_sinfo
->bytes_readonly
-
2669 data_sinfo
->bytes_may_use
< bytes
) {
2670 struct btrfs_trans_handle
*trans
;
2673 * if we don't have enough free bytes in this space then we need
2674 * to alloc a new chunk.
2676 if (!data_sinfo
->full
) {
2679 data_sinfo
->force_alloc
= 1;
2680 spin_unlock(&data_sinfo
->lock
);
2682 alloc_target
= btrfs_get_alloc_profile(root
, 1);
2683 trans
= btrfs_start_transaction(root
, 1);
2687 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
2688 bytes
+ 2 * 1024 * 1024,
2690 btrfs_end_transaction(trans
, root
);
2695 spin_unlock(&data_sinfo
->lock
);
2697 /* commit the current transaction and try again */
2700 trans
= btrfs_join_transaction(root
, 1);
2703 ret
= btrfs_commit_transaction(trans
, root
);
2709 printk(KERN_ERR
"no space left, need %llu, %llu delalloc bytes"
2710 ", %llu bytes_used, %llu bytes_reserved, "
2711 "%llu bytes_pinned, %llu bytes_readonly, %llu may use "
2712 "%llu total\n", (unsigned long long)bytes
,
2713 (unsigned long long)data_sinfo
->bytes_delalloc
,
2714 (unsigned long long)data_sinfo
->bytes_used
,
2715 (unsigned long long)data_sinfo
->bytes_reserved
,
2716 (unsigned long long)data_sinfo
->bytes_pinned
,
2717 (unsigned long long)data_sinfo
->bytes_readonly
,
2718 (unsigned long long)data_sinfo
->bytes_may_use
,
2719 (unsigned long long)data_sinfo
->total_bytes
);
2722 data_sinfo
->bytes_may_use
+= bytes
;
2723 BTRFS_I(inode
)->reserved_bytes
+= bytes
;
2724 spin_unlock(&data_sinfo
->lock
);
2726 return btrfs_check_metadata_free_space(root
);
2730 * if there was an error for whatever reason after calling
2731 * btrfs_check_data_free_space, call this so we can cleanup the counters.
2733 void btrfs_free_reserved_data_space(struct btrfs_root
*root
,
2734 struct inode
*inode
, u64 bytes
)
2736 struct btrfs_space_info
*data_sinfo
;
2738 /* make sure bytes are sectorsize aligned */
2739 bytes
= (bytes
+ root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
2741 data_sinfo
= BTRFS_I(inode
)->space_info
;
2742 spin_lock(&data_sinfo
->lock
);
2743 data_sinfo
->bytes_may_use
-= bytes
;
2744 BTRFS_I(inode
)->reserved_bytes
-= bytes
;
2745 spin_unlock(&data_sinfo
->lock
);
2748 /* called when we are adding a delalloc extent to the inode's io_tree */
2749 void btrfs_delalloc_reserve_space(struct btrfs_root
*root
, struct inode
*inode
,
2752 struct btrfs_space_info
*data_sinfo
;
2754 /* get the space info for where this inode will be storing its data */
2755 data_sinfo
= BTRFS_I(inode
)->space_info
;
2757 /* make sure we have enough space to handle the data first */
2758 spin_lock(&data_sinfo
->lock
);
2759 data_sinfo
->bytes_delalloc
+= bytes
;
2762 * we are adding a delalloc extent without calling
2763 * btrfs_check_data_free_space first. This happens on a weird
2764 * writepage condition, but shouldn't hurt our accounting
2766 if (unlikely(bytes
> BTRFS_I(inode
)->reserved_bytes
)) {
2767 data_sinfo
->bytes_may_use
-= BTRFS_I(inode
)->reserved_bytes
;
2768 BTRFS_I(inode
)->reserved_bytes
= 0;
2770 data_sinfo
->bytes_may_use
-= bytes
;
2771 BTRFS_I(inode
)->reserved_bytes
-= bytes
;
2774 spin_unlock(&data_sinfo
->lock
);
2777 /* called when we are clearing an delalloc extent from the inode's io_tree */
2778 void btrfs_delalloc_free_space(struct btrfs_root
*root
, struct inode
*inode
,
2781 struct btrfs_space_info
*info
;
2783 info
= BTRFS_I(inode
)->space_info
;
2785 spin_lock(&info
->lock
);
2786 info
->bytes_delalloc
-= bytes
;
2787 spin_unlock(&info
->lock
);
2790 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
2792 struct list_head
*head
= &info
->space_info
;
2793 struct btrfs_space_info
*found
;
2796 list_for_each_entry_rcu(found
, head
, list
) {
2797 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
2798 found
->force_alloc
= 1;
2803 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
2804 struct btrfs_root
*extent_root
, u64 alloc_bytes
,
2805 u64 flags
, int force
)
2807 struct btrfs_space_info
*space_info
;
2808 struct btrfs_fs_info
*fs_info
= extent_root
->fs_info
;
2812 mutex_lock(&fs_info
->chunk_mutex
);
2814 flags
= btrfs_reduce_alloc_profile(extent_root
, flags
);
2816 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
2818 ret
= update_space_info(extent_root
->fs_info
, flags
,
2822 BUG_ON(!space_info
);
2824 spin_lock(&space_info
->lock
);
2825 if (space_info
->force_alloc
) {
2827 space_info
->force_alloc
= 0;
2829 if (space_info
->full
) {
2830 spin_unlock(&space_info
->lock
);
2834 thresh
= space_info
->total_bytes
- space_info
->bytes_readonly
;
2835 thresh
= div_factor(thresh
, 6);
2837 (space_info
->bytes_used
+ space_info
->bytes_pinned
+
2838 space_info
->bytes_reserved
+ alloc_bytes
) < thresh
) {
2839 spin_unlock(&space_info
->lock
);
2842 spin_unlock(&space_info
->lock
);
2845 * if we're doing a data chunk, go ahead and make sure that
2846 * we keep a reasonable number of metadata chunks allocated in the
2849 if (flags
& BTRFS_BLOCK_GROUP_DATA
) {
2850 fs_info
->data_chunk_allocations
++;
2851 if (!(fs_info
->data_chunk_allocations
%
2852 fs_info
->metadata_ratio
))
2853 force_metadata_allocation(fs_info
);
2856 ret
= btrfs_alloc_chunk(trans
, extent_root
, flags
);
2858 space_info
->full
= 1;
2860 mutex_unlock(&extent_root
->fs_info
->chunk_mutex
);
2864 static int update_block_group(struct btrfs_trans_handle
*trans
,
2865 struct btrfs_root
*root
,
2866 u64 bytenr
, u64 num_bytes
, int alloc
,
2869 struct btrfs_block_group_cache
*cache
;
2870 struct btrfs_fs_info
*info
= root
->fs_info
;
2871 u64 total
= num_bytes
;
2875 /* block accounting for super block */
2876 spin_lock(&info
->delalloc_lock
);
2877 old_val
= btrfs_super_bytes_used(&info
->super_copy
);
2879 old_val
+= num_bytes
;
2881 old_val
-= num_bytes
;
2882 btrfs_set_super_bytes_used(&info
->super_copy
, old_val
);
2884 /* block accounting for root item */
2885 old_val
= btrfs_root_used(&root
->root_item
);
2887 old_val
+= num_bytes
;
2889 old_val
-= num_bytes
;
2890 btrfs_set_root_used(&root
->root_item
, old_val
);
2891 spin_unlock(&info
->delalloc_lock
);
2894 cache
= btrfs_lookup_block_group(info
, bytenr
);
2897 byte_in_group
= bytenr
- cache
->key
.objectid
;
2898 WARN_ON(byte_in_group
> cache
->key
.offset
);
2900 spin_lock(&cache
->space_info
->lock
);
2901 spin_lock(&cache
->lock
);
2903 old_val
= btrfs_block_group_used(&cache
->item
);
2904 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
2906 old_val
+= num_bytes
;
2907 cache
->space_info
->bytes_used
+= num_bytes
;
2909 cache
->space_info
->bytes_readonly
-= num_bytes
;
2910 btrfs_set_block_group_used(&cache
->item
, old_val
);
2911 spin_unlock(&cache
->lock
);
2912 spin_unlock(&cache
->space_info
->lock
);
2914 old_val
-= num_bytes
;
2915 cache
->space_info
->bytes_used
-= num_bytes
;
2917 cache
->space_info
->bytes_readonly
+= num_bytes
;
2918 btrfs_set_block_group_used(&cache
->item
, old_val
);
2919 spin_unlock(&cache
->lock
);
2920 spin_unlock(&cache
->space_info
->lock
);
2924 ret
= btrfs_discard_extent(root
, bytenr
,
2928 ret
= btrfs_add_free_space(cache
, bytenr
,
2933 btrfs_put_block_group(cache
);
2935 bytenr
+= num_bytes
;
2940 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
2942 struct btrfs_block_group_cache
*cache
;
2945 cache
= btrfs_lookup_first_block_group(root
->fs_info
, search_start
);
2949 bytenr
= cache
->key
.objectid
;
2950 btrfs_put_block_group(cache
);
2955 int btrfs_update_pinned_extents(struct btrfs_root
*root
,
2956 u64 bytenr
, u64 num
, int pin
)
2959 struct btrfs_block_group_cache
*cache
;
2960 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2963 set_extent_dirty(&fs_info
->pinned_extents
,
2964 bytenr
, bytenr
+ num
- 1, GFP_NOFS
);
2966 clear_extent_dirty(&fs_info
->pinned_extents
,
2967 bytenr
, bytenr
+ num
- 1, GFP_NOFS
);
2971 cache
= btrfs_lookup_block_group(fs_info
, bytenr
);
2973 len
= min(num
, cache
->key
.offset
-
2974 (bytenr
- cache
->key
.objectid
));
2976 spin_lock(&cache
->space_info
->lock
);
2977 spin_lock(&cache
->lock
);
2978 cache
->pinned
+= len
;
2979 cache
->space_info
->bytes_pinned
+= len
;
2980 spin_unlock(&cache
->lock
);
2981 spin_unlock(&cache
->space_info
->lock
);
2982 fs_info
->total_pinned
+= len
;
2984 spin_lock(&cache
->space_info
->lock
);
2985 spin_lock(&cache
->lock
);
2986 cache
->pinned
-= len
;
2987 cache
->space_info
->bytes_pinned
-= len
;
2988 spin_unlock(&cache
->lock
);
2989 spin_unlock(&cache
->space_info
->lock
);
2990 fs_info
->total_pinned
-= len
;
2992 btrfs_add_free_space(cache
, bytenr
, len
);
2994 btrfs_put_block_group(cache
);
3001 static int update_reserved_extents(struct btrfs_root
*root
,
3002 u64 bytenr
, u64 num
, int reserve
)
3005 struct btrfs_block_group_cache
*cache
;
3006 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3009 cache
= btrfs_lookup_block_group(fs_info
, bytenr
);
3011 len
= min(num
, cache
->key
.offset
-
3012 (bytenr
- cache
->key
.objectid
));
3014 spin_lock(&cache
->space_info
->lock
);
3015 spin_lock(&cache
->lock
);
3017 cache
->reserved
+= len
;
3018 cache
->space_info
->bytes_reserved
+= len
;
3020 cache
->reserved
-= len
;
3021 cache
->space_info
->bytes_reserved
-= len
;
3023 spin_unlock(&cache
->lock
);
3024 spin_unlock(&cache
->space_info
->lock
);
3025 btrfs_put_block_group(cache
);
3032 int btrfs_copy_pinned(struct btrfs_root
*root
, struct extent_io_tree
*copy
)
3037 struct extent_io_tree
*pinned_extents
= &root
->fs_info
->pinned_extents
;
3041 ret
= find_first_extent_bit(pinned_extents
, last
,
3042 &start
, &end
, EXTENT_DIRTY
);
3045 set_extent_dirty(copy
, start
, end
, GFP_NOFS
);
3051 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
3052 struct btrfs_root
*root
,
3053 struct extent_io_tree
*unpin
)
3060 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
3065 ret
= btrfs_discard_extent(root
, start
, end
+ 1 - start
);
3067 /* unlocks the pinned mutex */
3068 btrfs_update_pinned_extents(root
, start
, end
+ 1 - start
, 0);
3069 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
3076 static int pin_down_bytes(struct btrfs_trans_handle
*trans
,
3077 struct btrfs_root
*root
,
3078 struct btrfs_path
*path
,
3079 u64 bytenr
, u64 num_bytes
, int is_data
,
3080 struct extent_buffer
**must_clean
)
3083 struct extent_buffer
*buf
;
3088 buf
= btrfs_find_tree_block(root
, bytenr
, num_bytes
);
3092 /* we can reuse a block if it hasn't been written
3093 * and it is from this transaction. We can't
3094 * reuse anything from the tree log root because
3095 * it has tiny sub-transactions.
3097 if (btrfs_buffer_uptodate(buf
, 0) &&
3098 btrfs_try_tree_lock(buf
)) {
3099 u64 header_owner
= btrfs_header_owner(buf
);
3100 u64 header_transid
= btrfs_header_generation(buf
);
3101 if (header_owner
!= BTRFS_TREE_LOG_OBJECTID
&&
3102 header_transid
== trans
->transid
&&
3103 !btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
3107 btrfs_tree_unlock(buf
);
3109 free_extent_buffer(buf
);
3111 btrfs_set_path_blocking(path
);
3112 /* unlocks the pinned mutex */
3113 btrfs_update_pinned_extents(root
, bytenr
, num_bytes
, 1);
3120 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
3121 struct btrfs_root
*root
,
3122 u64 bytenr
, u64 num_bytes
, u64 parent
,
3123 u64 root_objectid
, u64 owner_objectid
,
3124 u64 owner_offset
, int refs_to_drop
,
3125 struct btrfs_delayed_extent_op
*extent_op
)
3127 struct btrfs_key key
;
3128 struct btrfs_path
*path
;
3129 struct btrfs_fs_info
*info
= root
->fs_info
;
3130 struct btrfs_root
*extent_root
= info
->extent_root
;
3131 struct extent_buffer
*leaf
;
3132 struct btrfs_extent_item
*ei
;
3133 struct btrfs_extent_inline_ref
*iref
;
3136 int extent_slot
= 0;
3137 int found_extent
= 0;
3142 path
= btrfs_alloc_path();
3147 path
->leave_spinning
= 1;
3149 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
3150 BUG_ON(!is_data
&& refs_to_drop
!= 1);
3152 ret
= lookup_extent_backref(trans
, extent_root
, path
, &iref
,
3153 bytenr
, num_bytes
, parent
,
3154 root_objectid
, owner_objectid
,
3157 extent_slot
= path
->slots
[0];
3158 while (extent_slot
>= 0) {
3159 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
3161 if (key
.objectid
!= bytenr
)
3163 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
3164 key
.offset
== num_bytes
) {
3168 if (path
->slots
[0] - extent_slot
> 5)
3172 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3173 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
3174 if (found_extent
&& item_size
< sizeof(*ei
))
3177 if (!found_extent
) {
3179 ret
= remove_extent_backref(trans
, extent_root
, path
,
3183 btrfs_release_path(extent_root
, path
);
3184 path
->leave_spinning
= 1;
3186 key
.objectid
= bytenr
;
3187 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
3188 key
.offset
= num_bytes
;
3190 ret
= btrfs_search_slot(trans
, extent_root
,
3193 printk(KERN_ERR
"umm, got %d back from search"
3194 ", was looking for %llu\n", ret
,
3195 (unsigned long long)bytenr
);
3196 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
3199 extent_slot
= path
->slots
[0];
3202 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
3204 printk(KERN_ERR
"btrfs unable to find ref byte nr %llu "
3205 "parent %llu root %llu owner %llu offset %llu\n",
3206 (unsigned long long)bytenr
,
3207 (unsigned long long)parent
,
3208 (unsigned long long)root_objectid
,
3209 (unsigned long long)owner_objectid
,
3210 (unsigned long long)owner_offset
);
3213 leaf
= path
->nodes
[0];
3214 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
3215 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3216 if (item_size
< sizeof(*ei
)) {
3217 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
3218 ret
= convert_extent_item_v0(trans
, extent_root
, path
,
3222 btrfs_release_path(extent_root
, path
);
3223 path
->leave_spinning
= 1;
3225 key
.objectid
= bytenr
;
3226 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
3227 key
.offset
= num_bytes
;
3229 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
3232 printk(KERN_ERR
"umm, got %d back from search"
3233 ", was looking for %llu\n", ret
,
3234 (unsigned long long)bytenr
);
3235 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
3238 extent_slot
= path
->slots
[0];
3239 leaf
= path
->nodes
[0];
3240 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
3243 BUG_ON(item_size
< sizeof(*ei
));
3244 ei
= btrfs_item_ptr(leaf
, extent_slot
,
3245 struct btrfs_extent_item
);
3246 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
) {
3247 struct btrfs_tree_block_info
*bi
;
3248 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
3249 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
3250 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
3253 refs
= btrfs_extent_refs(leaf
, ei
);
3254 BUG_ON(refs
< refs_to_drop
);
3255 refs
-= refs_to_drop
;
3259 __run_delayed_extent_op(extent_op
, leaf
, ei
);
3261 * In the case of inline back ref, reference count will
3262 * be updated by remove_extent_backref
3265 BUG_ON(!found_extent
);
3267 btrfs_set_extent_refs(leaf
, ei
, refs
);
3268 btrfs_mark_buffer_dirty(leaf
);
3271 ret
= remove_extent_backref(trans
, extent_root
, path
,
3278 struct extent_buffer
*must_clean
= NULL
;
3281 BUG_ON(is_data
&& refs_to_drop
!=
3282 extent_data_ref_count(root
, path
, iref
));
3284 BUG_ON(path
->slots
[0] != extent_slot
);
3286 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
3287 path
->slots
[0] = extent_slot
;
3292 ret
= pin_down_bytes(trans
, root
, path
, bytenr
,
3293 num_bytes
, is_data
, &must_clean
);
3298 * it is going to be very rare for someone to be waiting
3299 * on the block we're freeing. del_items might need to
3300 * schedule, so rather than get fancy, just force it
3304 btrfs_set_lock_blocking(must_clean
);
3306 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
3309 btrfs_release_path(extent_root
, path
);
3312 clean_tree_block(NULL
, root
, must_clean
);
3313 btrfs_tree_unlock(must_clean
);
3314 free_extent_buffer(must_clean
);
3318 ret
= btrfs_del_csums(trans
, root
, bytenr
, num_bytes
);
3321 invalidate_mapping_pages(info
->btree_inode
->i_mapping
,
3322 bytenr
>> PAGE_CACHE_SHIFT
,
3323 (bytenr
+ num_bytes
- 1) >> PAGE_CACHE_SHIFT
);
3326 ret
= update_block_group(trans
, root
, bytenr
, num_bytes
, 0,
3330 btrfs_free_path(path
);
3335 * when we free an extent, it is possible (and likely) that we free the last
3336 * delayed ref for that extent as well. This searches the delayed ref tree for
3337 * a given extent, and if there are no other delayed refs to be processed, it
3338 * removes it from the tree.
3340 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
3341 struct btrfs_root
*root
, u64 bytenr
)
3343 struct btrfs_delayed_ref_head
*head
;
3344 struct btrfs_delayed_ref_root
*delayed_refs
;
3345 struct btrfs_delayed_ref_node
*ref
;
3346 struct rb_node
*node
;
3349 delayed_refs
= &trans
->transaction
->delayed_refs
;
3350 spin_lock(&delayed_refs
->lock
);
3351 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
3355 node
= rb_prev(&head
->node
.rb_node
);
3359 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
3361 /* there are still entries for this ref, we can't drop it */
3362 if (ref
->bytenr
== bytenr
)
3365 if (head
->extent_op
) {
3366 if (!head
->must_insert_reserved
)
3368 kfree(head
->extent_op
);
3369 head
->extent_op
= NULL
;
3373 * waiting for the lock here would deadlock. If someone else has it
3374 * locked they are already in the process of dropping it anyway
3376 if (!mutex_trylock(&head
->mutex
))
3380 * at this point we have a head with no other entries. Go
3381 * ahead and process it.
3383 head
->node
.in_tree
= 0;
3384 rb_erase(&head
->node
.rb_node
, &delayed_refs
->root
);
3386 delayed_refs
->num_entries
--;
3389 * we don't take a ref on the node because we're removing it from the
3390 * tree, so we just steal the ref the tree was holding.
3392 delayed_refs
->num_heads
--;
3393 if (list_empty(&head
->cluster
))
3394 delayed_refs
->num_heads_ready
--;
3396 list_del_init(&head
->cluster
);
3397 spin_unlock(&delayed_refs
->lock
);
3399 ret
= run_one_delayed_ref(trans
, root
->fs_info
->tree_root
,
3400 &head
->node
, head
->extent_op
,
3401 head
->must_insert_reserved
);
3403 btrfs_put_delayed_ref(&head
->node
);
3406 spin_unlock(&delayed_refs
->lock
);
3410 int btrfs_free_extent(struct btrfs_trans_handle
*trans
,
3411 struct btrfs_root
*root
,
3412 u64 bytenr
, u64 num_bytes
, u64 parent
,
3413 u64 root_objectid
, u64 owner
, u64 offset
)
3418 * tree log blocks never actually go into the extent allocation
3419 * tree, just update pinning info and exit early.
3421 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
3422 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
3423 /* unlocks the pinned mutex */
3424 btrfs_update_pinned_extents(root
, bytenr
, num_bytes
, 1);
3425 update_reserved_extents(root
, bytenr
, num_bytes
, 0);
3427 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
3428 ret
= btrfs_add_delayed_tree_ref(trans
, bytenr
, num_bytes
,
3429 parent
, root_objectid
, (int)owner
,
3430 BTRFS_DROP_DELAYED_REF
, NULL
);
3432 ret
= check_ref_cleanup(trans
, root
, bytenr
);
3435 ret
= btrfs_add_delayed_data_ref(trans
, bytenr
, num_bytes
,
3436 parent
, root_objectid
, owner
,
3437 offset
, BTRFS_DROP_DELAYED_REF
, NULL
);
3443 static u64
stripe_align(struct btrfs_root
*root
, u64 val
)
3445 u64 mask
= ((u64
)root
->stripesize
- 1);
3446 u64 ret
= (val
+ mask
) & ~mask
;
3451 * walks the btree of allocated extents and find a hole of a given size.
3452 * The key ins is changed to record the hole:
3453 * ins->objectid == block start
3454 * ins->flags = BTRFS_EXTENT_ITEM_KEY
3455 * ins->offset == number of blocks
3456 * Any available blocks before search_start are skipped.
3458 static noinline
int find_free_extent(struct btrfs_trans_handle
*trans
,
3459 struct btrfs_root
*orig_root
,
3460 u64 num_bytes
, u64 empty_size
,
3461 u64 search_start
, u64 search_end
,
3462 u64 hint_byte
, struct btrfs_key
*ins
,
3463 u64 exclude_start
, u64 exclude_nr
,
3467 struct btrfs_root
*root
= orig_root
->fs_info
->extent_root
;
3468 struct btrfs_free_cluster
*last_ptr
= NULL
;
3469 struct btrfs_block_group_cache
*block_group
= NULL
;
3470 int empty_cluster
= 2 * 1024 * 1024;
3471 int allowed_chunk_alloc
= 0;
3472 struct btrfs_space_info
*space_info
;
3473 int last_ptr_loop
= 0;
3476 WARN_ON(num_bytes
< root
->sectorsize
);
3477 btrfs_set_key_type(ins
, BTRFS_EXTENT_ITEM_KEY
);
3481 space_info
= __find_space_info(root
->fs_info
, data
);
3483 if (orig_root
->ref_cows
|| empty_size
)
3484 allowed_chunk_alloc
= 1;
3486 if (data
& BTRFS_BLOCK_GROUP_METADATA
) {
3487 last_ptr
= &root
->fs_info
->meta_alloc_cluster
;
3488 if (!btrfs_test_opt(root
, SSD
))
3489 empty_cluster
= 64 * 1024;
3492 if ((data
& BTRFS_BLOCK_GROUP_DATA
) && btrfs_test_opt(root
, SSD
)) {
3493 last_ptr
= &root
->fs_info
->data_alloc_cluster
;
3497 spin_lock(&last_ptr
->lock
);
3498 if (last_ptr
->block_group
)
3499 hint_byte
= last_ptr
->window_start
;
3500 spin_unlock(&last_ptr
->lock
);
3503 search_start
= max(search_start
, first_logical_byte(root
, 0));
3504 search_start
= max(search_start
, hint_byte
);
3511 if (search_start
== hint_byte
) {
3512 block_group
= btrfs_lookup_block_group(root
->fs_info
,
3514 if (block_group
&& block_group_bits(block_group
, data
)) {
3515 down_read(&space_info
->groups_sem
);
3516 if (list_empty(&block_group
->list
) ||
3519 * someone is removing this block group,
3520 * we can't jump into the have_block_group
3521 * target because our list pointers are not
3524 btrfs_put_block_group(block_group
);
3525 up_read(&space_info
->groups_sem
);
3527 goto have_block_group
;
3528 } else if (block_group
) {
3529 btrfs_put_block_group(block_group
);
3534 down_read(&space_info
->groups_sem
);
3535 list_for_each_entry(block_group
, &space_info
->block_groups
, list
) {
3538 atomic_inc(&block_group
->count
);
3539 search_start
= block_group
->key
.objectid
;
3542 if (unlikely(!block_group
->cached
)) {
3543 mutex_lock(&block_group
->cache_mutex
);
3544 ret
= cache_block_group(root
, block_group
);
3545 mutex_unlock(&block_group
->cache_mutex
);
3547 btrfs_put_block_group(block_group
);
3552 if (unlikely(block_group
->ro
))
3557 * the refill lock keeps out other
3558 * people trying to start a new cluster
3560 spin_lock(&last_ptr
->refill_lock
);
3561 if (last_ptr
->block_group
&&
3562 (last_ptr
->block_group
->ro
||
3563 !block_group_bits(last_ptr
->block_group
, data
))) {
3565 goto refill_cluster
;
3568 offset
= btrfs_alloc_from_cluster(block_group
, last_ptr
,
3569 num_bytes
, search_start
);
3571 /* we have a block, we're done */
3572 spin_unlock(&last_ptr
->refill_lock
);
3576 spin_lock(&last_ptr
->lock
);
3578 * whoops, this cluster doesn't actually point to
3579 * this block group. Get a ref on the block
3580 * group is does point to and try again
3582 if (!last_ptr_loop
&& last_ptr
->block_group
&&
3583 last_ptr
->block_group
!= block_group
) {
3585 btrfs_put_block_group(block_group
);
3586 block_group
= last_ptr
->block_group
;
3587 atomic_inc(&block_group
->count
);
3588 spin_unlock(&last_ptr
->lock
);
3589 spin_unlock(&last_ptr
->refill_lock
);
3592 search_start
= block_group
->key
.objectid
;
3594 * we know this block group is properly
3595 * in the list because
3596 * btrfs_remove_block_group, drops the
3597 * cluster before it removes the block
3598 * group from the list
3600 goto have_block_group
;
3602 spin_unlock(&last_ptr
->lock
);
3605 * this cluster didn't work out, free it and
3608 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
3612 /* allocate a cluster in this block group */
3613 ret
= btrfs_find_space_cluster(trans
, root
,
3614 block_group
, last_ptr
,
3616 empty_cluster
+ empty_size
);
3619 * now pull our allocation out of this
3622 offset
= btrfs_alloc_from_cluster(block_group
,
3623 last_ptr
, num_bytes
,
3626 /* we found one, proceed */
3627 spin_unlock(&last_ptr
->refill_lock
);
3632 * at this point we either didn't find a cluster
3633 * or we weren't able to allocate a block from our
3634 * cluster. Free the cluster we've been trying
3635 * to use, and go to the next block group
3638 btrfs_return_cluster_to_free_space(NULL
,
3640 spin_unlock(&last_ptr
->refill_lock
);
3643 spin_unlock(&last_ptr
->refill_lock
);
3646 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
3647 num_bytes
, empty_size
);
3651 search_start
= stripe_align(root
, offset
);
3652 /* move on to the next group */
3653 if (search_start
+ num_bytes
>= search_end
) {
3654 btrfs_add_free_space(block_group
, offset
, num_bytes
);
3658 /* move on to the next group */
3659 if (search_start
+ num_bytes
>
3660 block_group
->key
.objectid
+ block_group
->key
.offset
) {
3661 btrfs_add_free_space(block_group
, offset
, num_bytes
);
3665 if (exclude_nr
> 0 &&
3666 (search_start
+ num_bytes
> exclude_start
&&
3667 search_start
< exclude_start
+ exclude_nr
)) {
3668 search_start
= exclude_start
+ exclude_nr
;
3670 btrfs_add_free_space(block_group
, offset
, num_bytes
);
3672 * if search_start is still in this block group
3673 * then we just re-search this block group
3675 if (search_start
>= block_group
->key
.objectid
&&
3676 search_start
< (block_group
->key
.objectid
+
3677 block_group
->key
.offset
))
3678 goto have_block_group
;
3682 ins
->objectid
= search_start
;
3683 ins
->offset
= num_bytes
;
3685 if (offset
< search_start
)
3686 btrfs_add_free_space(block_group
, offset
,
3687 search_start
- offset
);
3688 BUG_ON(offset
> search_start
);
3690 /* we are all good, lets return */
3693 btrfs_put_block_group(block_group
);
3695 up_read(&space_info
->groups_sem
);
3697 /* loop == 0, try to find a clustered alloc in every block group
3698 * loop == 1, try again after forcing a chunk allocation
3699 * loop == 2, set empty_size and empty_cluster to 0 and try again
3701 if (!ins
->objectid
&& loop
< 3 &&
3702 (empty_size
|| empty_cluster
|| allowed_chunk_alloc
)) {
3708 if (allowed_chunk_alloc
) {
3709 ret
= do_chunk_alloc(trans
, root
, num_bytes
+
3710 2 * 1024 * 1024, data
, 1);
3711 allowed_chunk_alloc
= 0;
3713 space_info
->force_alloc
= 1;
3721 } else if (!ins
->objectid
) {
3725 /* we found what we needed */
3726 if (ins
->objectid
) {
3727 if (!(data
& BTRFS_BLOCK_GROUP_DATA
))
3728 trans
->block_group
= block_group
->key
.objectid
;
3730 btrfs_put_block_group(block_group
);
3737 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
)
3739 struct btrfs_block_group_cache
*cache
;
3741 printk(KERN_INFO
"space_info has %llu free, is %sfull\n",
3742 (unsigned long long)(info
->total_bytes
- info
->bytes_used
-
3743 info
->bytes_pinned
- info
->bytes_reserved
),
3744 (info
->full
) ? "" : "not ");
3745 printk(KERN_INFO
"space_info total=%llu, pinned=%llu, delalloc=%llu,"
3746 " may_use=%llu, used=%llu\n",
3747 (unsigned long long)info
->total_bytes
,
3748 (unsigned long long)info
->bytes_pinned
,
3749 (unsigned long long)info
->bytes_delalloc
,
3750 (unsigned long long)info
->bytes_may_use
,
3751 (unsigned long long)info
->bytes_used
);
3753 down_read(&info
->groups_sem
);
3754 list_for_each_entry(cache
, &info
->block_groups
, list
) {
3755 spin_lock(&cache
->lock
);
3756 printk(KERN_INFO
"block group %llu has %llu bytes, %llu used "
3757 "%llu pinned %llu reserved\n",
3758 (unsigned long long)cache
->key
.objectid
,
3759 (unsigned long long)cache
->key
.offset
,
3760 (unsigned long long)btrfs_block_group_used(&cache
->item
),
3761 (unsigned long long)cache
->pinned
,
3762 (unsigned long long)cache
->reserved
);
3763 btrfs_dump_free_space(cache
, bytes
);
3764 spin_unlock(&cache
->lock
);
3766 up_read(&info
->groups_sem
);
3769 static int __btrfs_reserve_extent(struct btrfs_trans_handle
*trans
,
3770 struct btrfs_root
*root
,
3771 u64 num_bytes
, u64 min_alloc_size
,
3772 u64 empty_size
, u64 hint_byte
,
3773 u64 search_end
, struct btrfs_key
*ins
,
3777 u64 search_start
= 0;
3778 struct btrfs_fs_info
*info
= root
->fs_info
;
3780 data
= btrfs_get_alloc_profile(root
, data
);
3783 * the only place that sets empty_size is btrfs_realloc_node, which
3784 * is not called recursively on allocations
3786 if (empty_size
|| root
->ref_cows
) {
3787 if (!(data
& BTRFS_BLOCK_GROUP_METADATA
)) {
3788 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
3790 BTRFS_BLOCK_GROUP_METADATA
|
3791 (info
->metadata_alloc_profile
&
3792 info
->avail_metadata_alloc_bits
), 0);
3794 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
3795 num_bytes
+ 2 * 1024 * 1024, data
, 0);
3798 WARN_ON(num_bytes
< root
->sectorsize
);
3799 ret
= find_free_extent(trans
, root
, num_bytes
, empty_size
,
3800 search_start
, search_end
, hint_byte
, ins
,
3801 trans
->alloc_exclude_start
,
3802 trans
->alloc_exclude_nr
, data
);
3804 if (ret
== -ENOSPC
&& num_bytes
> min_alloc_size
) {
3805 num_bytes
= num_bytes
>> 1;
3806 num_bytes
= num_bytes
& ~(root
->sectorsize
- 1);
3807 num_bytes
= max(num_bytes
, min_alloc_size
);
3808 do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
3809 num_bytes
, data
, 1);
3813 struct btrfs_space_info
*sinfo
;
3815 sinfo
= __find_space_info(root
->fs_info
, data
);
3816 printk(KERN_ERR
"btrfs allocation failed flags %llu, "
3817 "wanted %llu\n", (unsigned long long)data
,
3818 (unsigned long long)num_bytes
);
3819 dump_space_info(sinfo
, num_bytes
);
3826 int btrfs_free_reserved_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
3828 struct btrfs_block_group_cache
*cache
;
3831 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
3833 printk(KERN_ERR
"Unable to find block group for %llu\n",
3834 (unsigned long long)start
);
3838 ret
= btrfs_discard_extent(root
, start
, len
);
3840 btrfs_add_free_space(cache
, start
, len
);
3841 btrfs_put_block_group(cache
);
3842 update_reserved_extents(root
, start
, len
, 0);
3847 int btrfs_reserve_extent(struct btrfs_trans_handle
*trans
,
3848 struct btrfs_root
*root
,
3849 u64 num_bytes
, u64 min_alloc_size
,
3850 u64 empty_size
, u64 hint_byte
,
3851 u64 search_end
, struct btrfs_key
*ins
,
3855 ret
= __btrfs_reserve_extent(trans
, root
, num_bytes
, min_alloc_size
,
3856 empty_size
, hint_byte
, search_end
, ins
,
3858 update_reserved_extents(root
, ins
->objectid
, ins
->offset
, 1);
3862 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
3863 struct btrfs_root
*root
,
3864 u64 parent
, u64 root_objectid
,
3865 u64 flags
, u64 owner
, u64 offset
,
3866 struct btrfs_key
*ins
, int ref_mod
)
3869 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3870 struct btrfs_extent_item
*extent_item
;
3871 struct btrfs_extent_inline_ref
*iref
;
3872 struct btrfs_path
*path
;
3873 struct extent_buffer
*leaf
;
3878 type
= BTRFS_SHARED_DATA_REF_KEY
;
3880 type
= BTRFS_EXTENT_DATA_REF_KEY
;
3882 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
3884 path
= btrfs_alloc_path();
3887 path
->leave_spinning
= 1;
3888 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
3892 leaf
= path
->nodes
[0];
3893 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
3894 struct btrfs_extent_item
);
3895 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
3896 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
3897 btrfs_set_extent_flags(leaf
, extent_item
,
3898 flags
| BTRFS_EXTENT_FLAG_DATA
);
3900 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
3901 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
3903 struct btrfs_shared_data_ref
*ref
;
3904 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
3905 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
3906 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
3908 struct btrfs_extent_data_ref
*ref
;
3909 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
3910 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
3911 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
3912 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
3913 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
3916 btrfs_mark_buffer_dirty(path
->nodes
[0]);
3917 btrfs_free_path(path
);
3919 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
,
3922 printk(KERN_ERR
"btrfs update block group failed for %llu "
3923 "%llu\n", (unsigned long long)ins
->objectid
,
3924 (unsigned long long)ins
->offset
);
3930 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
3931 struct btrfs_root
*root
,
3932 u64 parent
, u64 root_objectid
,
3933 u64 flags
, struct btrfs_disk_key
*key
,
3934 int level
, struct btrfs_key
*ins
)
3937 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3938 struct btrfs_extent_item
*extent_item
;
3939 struct btrfs_tree_block_info
*block_info
;
3940 struct btrfs_extent_inline_ref
*iref
;
3941 struct btrfs_path
*path
;
3942 struct extent_buffer
*leaf
;
3943 u32 size
= sizeof(*extent_item
) + sizeof(*block_info
) + sizeof(*iref
);
3945 path
= btrfs_alloc_path();
3948 path
->leave_spinning
= 1;
3949 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
3953 leaf
= path
->nodes
[0];
3954 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
3955 struct btrfs_extent_item
);
3956 btrfs_set_extent_refs(leaf
, extent_item
, 1);
3957 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
3958 btrfs_set_extent_flags(leaf
, extent_item
,
3959 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
3960 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
3962 btrfs_set_tree_block_key(leaf
, block_info
, key
);
3963 btrfs_set_tree_block_level(leaf
, block_info
, level
);
3965 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
3967 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
3968 btrfs_set_extent_inline_ref_type(leaf
, iref
,
3969 BTRFS_SHARED_BLOCK_REF_KEY
);
3970 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
3972 btrfs_set_extent_inline_ref_type(leaf
, iref
,
3973 BTRFS_TREE_BLOCK_REF_KEY
);
3974 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
3977 btrfs_mark_buffer_dirty(leaf
);
3978 btrfs_free_path(path
);
3980 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
,
3983 printk(KERN_ERR
"btrfs update block group failed for %llu "
3984 "%llu\n", (unsigned long long)ins
->objectid
,
3985 (unsigned long long)ins
->offset
);
3991 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
3992 struct btrfs_root
*root
,
3993 u64 root_objectid
, u64 owner
,
3994 u64 offset
, struct btrfs_key
*ins
)
3998 BUG_ON(root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
4000 ret
= btrfs_add_delayed_data_ref(trans
, ins
->objectid
, ins
->offset
,
4001 0, root_objectid
, owner
, offset
,
4002 BTRFS_ADD_DELAYED_EXTENT
, NULL
);
4007 * this is used by the tree logging recovery code. It records that
4008 * an extent has been allocated and makes sure to clear the free
4009 * space cache bits as well
4011 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
4012 struct btrfs_root
*root
,
4013 u64 root_objectid
, u64 owner
, u64 offset
,
4014 struct btrfs_key
*ins
)
4017 struct btrfs_block_group_cache
*block_group
;
4019 block_group
= btrfs_lookup_block_group(root
->fs_info
, ins
->objectid
);
4020 mutex_lock(&block_group
->cache_mutex
);
4021 cache_block_group(root
, block_group
);
4022 mutex_unlock(&block_group
->cache_mutex
);
4024 ret
= btrfs_remove_free_space(block_group
, ins
->objectid
,
4027 btrfs_put_block_group(block_group
);
4028 ret
= alloc_reserved_file_extent(trans
, root
, 0, root_objectid
,
4029 0, owner
, offset
, ins
, 1);
4034 * finds a free extent and does all the dirty work required for allocation
4035 * returns the key for the extent through ins, and a tree buffer for
4036 * the first block of the extent through buf.
4038 * returns 0 if everything worked, non-zero otherwise.
4040 static int alloc_tree_block(struct btrfs_trans_handle
*trans
,
4041 struct btrfs_root
*root
,
4042 u64 num_bytes
, u64 parent
, u64 root_objectid
,
4043 struct btrfs_disk_key
*key
, int level
,
4044 u64 empty_size
, u64 hint_byte
, u64 search_end
,
4045 struct btrfs_key
*ins
)
4050 ret
= __btrfs_reserve_extent(trans
, root
, num_bytes
, num_bytes
,
4051 empty_size
, hint_byte
, search_end
,
4055 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
4057 parent
= ins
->objectid
;
4058 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
4062 update_reserved_extents(root
, ins
->objectid
, ins
->offset
, 1);
4063 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
4064 struct btrfs_delayed_extent_op
*extent_op
;
4065 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
4068 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
4070 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
4071 extent_op
->flags_to_set
= flags
;
4072 extent_op
->update_key
= 1;
4073 extent_op
->update_flags
= 1;
4074 extent_op
->is_data
= 0;
4076 ret
= btrfs_add_delayed_tree_ref(trans
, ins
->objectid
,
4077 ins
->offset
, parent
, root_objectid
,
4078 level
, BTRFS_ADD_DELAYED_EXTENT
,
4085 struct extent_buffer
*btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
,
4086 struct btrfs_root
*root
,
4087 u64 bytenr
, u32 blocksize
,
4090 struct extent_buffer
*buf
;
4092 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
4094 return ERR_PTR(-ENOMEM
);
4095 btrfs_set_header_generation(buf
, trans
->transid
);
4096 btrfs_set_buffer_lockdep_class(buf
, level
);
4097 btrfs_tree_lock(buf
);
4098 clean_tree_block(trans
, root
, buf
);
4100 btrfs_set_lock_blocking(buf
);
4101 btrfs_set_buffer_uptodate(buf
);
4103 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
4104 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
4105 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
4107 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
4108 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
4110 trans
->blocks_used
++;
4111 /* this returns a buffer locked for blocking */
4116 * helper function to allocate a block for a given tree
4117 * returns the tree buffer or NULL.
4119 struct extent_buffer
*btrfs_alloc_free_block(struct btrfs_trans_handle
*trans
,
4120 struct btrfs_root
*root
, u32 blocksize
,
4121 u64 parent
, u64 root_objectid
,
4122 struct btrfs_disk_key
*key
, int level
,
4123 u64 hint
, u64 empty_size
)
4125 struct btrfs_key ins
;
4127 struct extent_buffer
*buf
;
4129 ret
= alloc_tree_block(trans
, root
, blocksize
, parent
, root_objectid
,
4130 key
, level
, empty_size
, hint
, (u64
)-1, &ins
);
4133 return ERR_PTR(ret
);
4136 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
,
4142 int btrfs_drop_leaf_ref(struct btrfs_trans_handle
*trans
,
4143 struct btrfs_root
*root
, struct extent_buffer
*leaf
)
4147 struct btrfs_key key
;
4148 struct btrfs_file_extent_item
*fi
;
4153 BUG_ON(!btrfs_is_leaf(leaf
));
4154 nritems
= btrfs_header_nritems(leaf
);
4156 for (i
= 0; i
< nritems
; i
++) {
4158 btrfs_item_key_to_cpu(leaf
, &key
, i
);
4160 /* only extents have references, skip everything else */
4161 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
4164 fi
= btrfs_item_ptr(leaf
, i
, struct btrfs_file_extent_item
);
4166 /* inline extents live in the btree, they don't have refs */
4167 if (btrfs_file_extent_type(leaf
, fi
) ==
4168 BTRFS_FILE_EXTENT_INLINE
)
4171 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
4173 /* holes don't have refs */
4174 if (disk_bytenr
== 0)
4177 num_bytes
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
4178 ret
= btrfs_free_extent(trans
, root
, disk_bytenr
, num_bytes
,
4179 leaf
->start
, 0, key
.objectid
, 0);
4185 static noinline
int cache_drop_leaf_ref(struct btrfs_trans_handle
*trans
,
4186 struct btrfs_root
*root
,
4187 struct btrfs_leaf_ref
*ref
)
4191 struct btrfs_extent_info
*info
;
4192 struct refsort
*sorted
;
4194 if (ref
->nritems
== 0)
4197 sorted
= kmalloc(sizeof(*sorted
) * ref
->nritems
, GFP_NOFS
);
4198 for (i
= 0; i
< ref
->nritems
; i
++) {
4199 sorted
[i
].bytenr
= ref
->extents
[i
].bytenr
;
4202 sort(sorted
, ref
->nritems
, sizeof(struct refsort
), refsort_cmp
, NULL
);
4205 * the items in the ref were sorted when the ref was inserted
4206 * into the ref cache, so this is already in order
4208 for (i
= 0; i
< ref
->nritems
; i
++) {
4209 info
= ref
->extents
+ sorted
[i
].slot
;
4210 ret
= btrfs_free_extent(trans
, root
, info
->bytenr
,
4211 info
->num_bytes
, ref
->bytenr
,
4212 ref
->owner
, ref
->generation
,
4215 atomic_inc(&root
->fs_info
->throttle_gen
);
4216 wake_up(&root
->fs_info
->transaction_throttle
);
4228 static int drop_snap_lookup_refcount(struct btrfs_trans_handle
*trans
,
4229 struct btrfs_root
*root
, u64 start
,
4234 ret
= btrfs_lookup_extent_refs(trans
, root
, start
, len
, refs
);
4237 #if 0 /* some debugging code in case we see problems here */
4238 /* if the refs count is one, it won't get increased again. But
4239 * if the ref count is > 1, someone may be decreasing it at
4240 * the same time we are.
4243 struct extent_buffer
*eb
= NULL
;
4244 eb
= btrfs_find_create_tree_block(root
, start
, len
);
4246 btrfs_tree_lock(eb
);
4248 mutex_lock(&root
->fs_info
->alloc_mutex
);
4249 ret
= lookup_extent_ref(NULL
, root
, start
, len
, refs
);
4251 mutex_unlock(&root
->fs_info
->alloc_mutex
);
4254 btrfs_tree_unlock(eb
);
4255 free_extent_buffer(eb
);
4258 printk(KERN_ERR
"btrfs block %llu went down to one "
4259 "during drop_snap\n", (unsigned long long)start
);
4271 * this is used while deleting old snapshots, and it drops the refs
4272 * on a whole subtree starting from a level 1 node.
4274 * The idea is to sort all the leaf pointers, and then drop the
4275 * ref on all the leaves in order. Most of the time the leaves
4276 * will have ref cache entries, so no leaf IOs will be required to
4277 * find the extents they have references on.
4279 * For each leaf, any references it has are also dropped in order
4281 * This ends up dropping the references in something close to optimal
4282 * order for reading and modifying the extent allocation tree.
4284 static noinline
int drop_level_one_refs(struct btrfs_trans_handle
*trans
,
4285 struct btrfs_root
*root
,
4286 struct btrfs_path
*path
)
4291 struct extent_buffer
*eb
= path
->nodes
[1];
4292 struct extent_buffer
*leaf
;
4293 struct btrfs_leaf_ref
*ref
;
4294 struct refsort
*sorted
= NULL
;
4295 int nritems
= btrfs_header_nritems(eb
);
4299 int slot
= path
->slots
[1];
4300 u32 blocksize
= btrfs_level_size(root
, 0);
4306 root_owner
= btrfs_header_owner(eb
);
4307 root_gen
= btrfs_header_generation(eb
);
4308 sorted
= kmalloc(sizeof(*sorted
) * nritems
, GFP_NOFS
);
4311 * step one, sort all the leaf pointers so we don't scribble
4312 * randomly into the extent allocation tree
4314 for (i
= slot
; i
< nritems
; i
++) {
4315 sorted
[refi
].bytenr
= btrfs_node_blockptr(eb
, i
);
4316 sorted
[refi
].slot
= i
;
4321 * nritems won't be zero, but if we're picking up drop_snapshot
4322 * after a crash, slot might be > 0, so double check things
4328 sort(sorted
, refi
, sizeof(struct refsort
), refsort_cmp
, NULL
);
4331 * the first loop frees everything the leaves point to
4333 for (i
= 0; i
< refi
; i
++) {
4336 bytenr
= sorted
[i
].bytenr
;
4339 * check the reference count on this leaf. If it is > 1
4340 * we just decrement it below and don't update any
4341 * of the refs the leaf points to.
4343 ret
= drop_snap_lookup_refcount(trans
, root
, bytenr
,
4349 ptr_gen
= btrfs_node_ptr_generation(eb
, sorted
[i
].slot
);
4352 * the leaf only had one reference, which means the
4353 * only thing pointing to this leaf is the snapshot
4354 * we're deleting. It isn't possible for the reference
4355 * count to increase again later
4357 * The reference cache is checked for the leaf,
4358 * and if found we'll be able to drop any refs held by
4359 * the leaf without needing to read it in.
4361 ref
= btrfs_lookup_leaf_ref(root
, bytenr
);
4362 if (ref
&& ref
->generation
!= ptr_gen
) {
4363 btrfs_free_leaf_ref(root
, ref
);
4367 ret
= cache_drop_leaf_ref(trans
, root
, ref
);
4369 btrfs_remove_leaf_ref(root
, ref
);
4370 btrfs_free_leaf_ref(root
, ref
);
4373 * the leaf wasn't in the reference cache, so
4374 * we have to read it.
4376 leaf
= read_tree_block(root
, bytenr
, blocksize
,
4378 ret
= btrfs_drop_leaf_ref(trans
, root
, leaf
);
4380 free_extent_buffer(leaf
);
4382 atomic_inc(&root
->fs_info
->throttle_gen
);
4383 wake_up(&root
->fs_info
->transaction_throttle
);
4388 * run through the loop again to free the refs on the leaves.
4389 * This is faster than doing it in the loop above because
4390 * the leaves are likely to be clustered together. We end up
4391 * working in nice chunks on the extent allocation tree.
4393 for (i
= 0; i
< refi
; i
++) {
4394 bytenr
= sorted
[i
].bytenr
;
4395 ret
= btrfs_free_extent(trans
, root
, bytenr
,
4396 blocksize
, eb
->start
,
4397 root_owner
, root_gen
, 0, 1);
4400 atomic_inc(&root
->fs_info
->throttle_gen
);
4401 wake_up(&root
->fs_info
->transaction_throttle
);
4408 * update the path to show we've processed the entire level 1
4409 * node. This will get saved into the root's drop_snapshot_progress
4410 * field so these drops are not repeated again if this transaction
4413 path
->slots
[1] = nritems
;
4418 * helper function for drop_snapshot, this walks down the tree dropping ref
4419 * counts as it goes.
4421 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
4422 struct btrfs_root
*root
,
4423 struct btrfs_path
*path
, int *level
)
4429 struct extent_buffer
*next
;
4430 struct extent_buffer
*cur
;
4431 struct extent_buffer
*parent
;
4436 WARN_ON(*level
< 0);
4437 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
4438 ret
= drop_snap_lookup_refcount(trans
, root
, path
->nodes
[*level
]->start
,
4439 path
->nodes
[*level
]->len
, &refs
);
4445 * walk down to the last node level and free all the leaves
4447 while (*level
>= 0) {
4448 WARN_ON(*level
< 0);
4449 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
4450 cur
= path
->nodes
[*level
];
4452 if (btrfs_header_level(cur
) != *level
)
4455 if (path
->slots
[*level
] >=
4456 btrfs_header_nritems(cur
))
4459 /* the new code goes down to level 1 and does all the
4460 * leaves pointed to that node in bulk. So, this check
4461 * for level 0 will always be false.
4463 * But, the disk format allows the drop_snapshot_progress
4464 * field in the root to leave things in a state where
4465 * a leaf will need cleaning up here. If someone crashes
4466 * with the old code and then boots with the new code,
4467 * we might find a leaf here.
4470 ret
= btrfs_drop_leaf_ref(trans
, root
, cur
);
4476 * once we get to level one, process the whole node
4477 * at once, including everything below it.
4480 ret
= drop_level_one_refs(trans
, root
, path
);
4485 bytenr
= btrfs_node_blockptr(cur
, path
->slots
[*level
]);
4486 ptr_gen
= btrfs_node_ptr_generation(cur
, path
->slots
[*level
]);
4487 blocksize
= btrfs_level_size(root
, *level
- 1);
4489 ret
= drop_snap_lookup_refcount(trans
, root
, bytenr
,
4494 * if there is more than one reference, we don't need
4495 * to read that node to drop any references it has. We
4496 * just drop the ref we hold on that node and move on to the
4497 * next slot in this level.
4500 parent
= path
->nodes
[*level
];
4501 root_owner
= btrfs_header_owner(parent
);
4502 root_gen
= btrfs_header_generation(parent
);
4503 path
->slots
[*level
]++;
4505 ret
= btrfs_free_extent(trans
, root
, bytenr
,
4506 blocksize
, parent
->start
,
4507 root_owner
, root_gen
,
4511 atomic_inc(&root
->fs_info
->throttle_gen
);
4512 wake_up(&root
->fs_info
->transaction_throttle
);
4519 * we need to keep freeing things in the next level down.
4520 * read the block and loop around to process it
4522 next
= read_tree_block(root
, bytenr
, blocksize
, ptr_gen
);
4523 WARN_ON(*level
<= 0);
4524 if (path
->nodes
[*level
-1])
4525 free_extent_buffer(path
->nodes
[*level
-1]);
4526 path
->nodes
[*level
-1] = next
;
4527 *level
= btrfs_header_level(next
);
4528 path
->slots
[*level
] = 0;
4532 WARN_ON(*level
< 0);
4533 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
4535 if (path
->nodes
[*level
] == root
->node
) {
4536 parent
= path
->nodes
[*level
];
4537 bytenr
= path
->nodes
[*level
]->start
;
4539 parent
= path
->nodes
[*level
+ 1];
4540 bytenr
= btrfs_node_blockptr(parent
, path
->slots
[*level
+ 1]);
4543 blocksize
= btrfs_level_size(root
, *level
);
4544 root_owner
= btrfs_header_owner(parent
);
4545 root_gen
= btrfs_header_generation(parent
);
4548 * cleanup and free the reference on the last node
4551 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
,
4552 parent
->start
, root_owner
, root_gen
,
4554 free_extent_buffer(path
->nodes
[*level
]);
4555 path
->nodes
[*level
] = NULL
;
4565 struct walk_control
{
4566 u64 refs
[BTRFS_MAX_LEVEL
];
4567 u64 flags
[BTRFS_MAX_LEVEL
];
4568 struct btrfs_key update_progress
;
4576 #define DROP_REFERENCE 1
4577 #define UPDATE_BACKREF 2
4580 * hepler to process tree block while walking down the tree.
4582 * when wc->stage == DROP_REFERENCE, this function checks
4583 * reference count of the block. if the block is shared and
4584 * we need update back refs for the subtree rooted at the
4585 * block, this function changes wc->stage to UPDATE_BACKREF
4587 * when wc->stage == UPDATE_BACKREF, this function updates
4588 * back refs for pointers in the block.
4590 * NOTE: return value 1 means we should stop walking down.
4592 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
4593 struct btrfs_root
*root
,
4594 struct btrfs_path
*path
,
4595 struct walk_control
*wc
)
4597 int level
= wc
->level
;
4598 struct extent_buffer
*eb
= path
->nodes
[level
];
4599 struct btrfs_key key
;
4600 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
4603 if (wc
->stage
== UPDATE_BACKREF
&&
4604 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
4608 * when reference count of tree block is 1, it won't increase
4609 * again. once full backref flag is set, we never clear it.
4611 if ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
4612 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
))) {
4613 BUG_ON(!path
->locks
[level
]);
4614 ret
= btrfs_lookup_extent_info(trans
, root
,
4619 BUG_ON(wc
->refs
[level
] == 0);
4622 if (wc
->stage
== DROP_REFERENCE
&&
4623 wc
->update_ref
&& wc
->refs
[level
] > 1) {
4624 BUG_ON(eb
== root
->node
);
4625 BUG_ON(path
->slots
[level
] > 0);
4627 btrfs_item_key_to_cpu(eb
, &key
, path
->slots
[level
]);
4629 btrfs_node_key_to_cpu(eb
, &key
, path
->slots
[level
]);
4630 if (btrfs_header_owner(eb
) == root
->root_key
.objectid
&&
4631 btrfs_comp_cpu_keys(&key
, &wc
->update_progress
) >= 0) {
4632 wc
->stage
= UPDATE_BACKREF
;
4633 wc
->shared_level
= level
;
4637 if (wc
->stage
== DROP_REFERENCE
) {
4638 if (wc
->refs
[level
] > 1)
4641 if (path
->locks
[level
] && !wc
->keep_locks
) {
4642 btrfs_tree_unlock(eb
);
4643 path
->locks
[level
] = 0;
4648 /* wc->stage == UPDATE_BACKREF */
4649 if (!(wc
->flags
[level
] & flag
)) {
4650 BUG_ON(!path
->locks
[level
]);
4651 ret
= btrfs_inc_ref(trans
, root
, eb
, 1);
4653 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
4655 ret
= btrfs_set_disk_extent_flags(trans
, root
, eb
->start
,
4658 wc
->flags
[level
] |= flag
;
4662 * the block is shared by multiple trees, so it's not good to
4663 * keep the tree lock
4665 if (path
->locks
[level
] && level
> 0) {
4666 btrfs_tree_unlock(eb
);
4667 path
->locks
[level
] = 0;
4673 * hepler to process tree block while walking up the tree.
4675 * when wc->stage == DROP_REFERENCE, this function drops
4676 * reference count on the block.
4678 * when wc->stage == UPDATE_BACKREF, this function changes
4679 * wc->stage back to DROP_REFERENCE if we changed wc->stage
4680 * to UPDATE_BACKREF previously while processing the block.
4682 * NOTE: return value 1 means we should stop walking up.
4684 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
4685 struct btrfs_root
*root
,
4686 struct btrfs_path
*path
,
4687 struct walk_control
*wc
)
4690 int level
= wc
->level
;
4691 struct extent_buffer
*eb
= path
->nodes
[level
];
4694 if (wc
->stage
== UPDATE_BACKREF
) {
4695 BUG_ON(wc
->shared_level
< level
);
4696 if (level
< wc
->shared_level
)
4699 BUG_ON(wc
->refs
[level
] <= 1);
4700 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
4704 wc
->stage
= DROP_REFERENCE
;
4705 wc
->shared_level
= -1;
4706 path
->slots
[level
] = 0;
4709 * check reference count again if the block isn't locked.
4710 * we should start walking down the tree again if reference
4713 if (!path
->locks
[level
]) {
4715 btrfs_tree_lock(eb
);
4716 btrfs_set_lock_blocking(eb
);
4717 path
->locks
[level
] = 1;
4719 ret
= btrfs_lookup_extent_info(trans
, root
,
4724 BUG_ON(wc
->refs
[level
] == 0);
4725 if (wc
->refs
[level
] == 1) {
4726 btrfs_tree_unlock(eb
);
4727 path
->locks
[level
] = 0;
4735 /* wc->stage == DROP_REFERENCE */
4736 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
4738 if (wc
->refs
[level
] == 1) {
4740 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
4741 ret
= btrfs_dec_ref(trans
, root
, eb
, 1);
4743 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
4746 /* make block locked assertion in clean_tree_block happy */
4747 if (!path
->locks
[level
] &&
4748 btrfs_header_generation(eb
) == trans
->transid
) {
4749 btrfs_tree_lock(eb
);
4750 btrfs_set_lock_blocking(eb
);
4751 path
->locks
[level
] = 1;
4753 clean_tree_block(trans
, root
, eb
);
4756 if (eb
== root
->node
) {
4757 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
4760 BUG_ON(root
->root_key
.objectid
!=
4761 btrfs_header_owner(eb
));
4763 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
4764 parent
= path
->nodes
[level
+ 1]->start
;
4766 BUG_ON(root
->root_key
.objectid
!=
4767 btrfs_header_owner(path
->nodes
[level
+ 1]));
4770 ret
= btrfs_free_extent(trans
, root
, eb
->start
, eb
->len
, parent
,
4771 root
->root_key
.objectid
, level
, 0);
4774 wc
->refs
[level
] = 0;
4775 wc
->flags
[level
] = 0;
4779 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
4780 struct btrfs_root
*root
,
4781 struct btrfs_path
*path
,
4782 struct walk_control
*wc
)
4784 struct extent_buffer
*next
;
4785 struct extent_buffer
*cur
;
4789 int level
= wc
->level
;
4792 while (level
>= 0) {
4793 cur
= path
->nodes
[level
];
4794 BUG_ON(path
->slots
[level
] >= btrfs_header_nritems(cur
));
4796 ret
= walk_down_proc(trans
, root
, path
, wc
);
4803 bytenr
= btrfs_node_blockptr(cur
, path
->slots
[level
]);
4804 blocksize
= btrfs_level_size(root
, level
- 1);
4805 ptr_gen
= btrfs_node_ptr_generation(cur
, path
->slots
[level
]);
4807 next
= read_tree_block(root
, bytenr
, blocksize
, ptr_gen
);
4808 btrfs_tree_lock(next
);
4809 btrfs_set_lock_blocking(next
);
4812 BUG_ON(level
!= btrfs_header_level(next
));
4813 path
->nodes
[level
] = next
;
4814 path
->slots
[level
] = 0;
4815 path
->locks
[level
] = 1;
4821 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
4822 struct btrfs_root
*root
,
4823 struct btrfs_path
*path
,
4824 struct walk_control
*wc
, int max_level
)
4826 int level
= wc
->level
;
4829 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
4830 while (level
< max_level
&& path
->nodes
[level
]) {
4832 if (path
->slots
[level
] + 1 <
4833 btrfs_header_nritems(path
->nodes
[level
])) {
4834 path
->slots
[level
]++;
4837 ret
= walk_up_proc(trans
, root
, path
, wc
);
4841 if (path
->locks
[level
]) {
4842 btrfs_tree_unlock(path
->nodes
[level
]);
4843 path
->locks
[level
] = 0;
4845 free_extent_buffer(path
->nodes
[level
]);
4846 path
->nodes
[level
] = NULL
;
4854 * drop a subvolume tree.
4856 * this function traverses the tree freeing any blocks that only
4857 * referenced by the tree.
4859 * when a shared tree block is found. this function decreases its
4860 * reference count by one. if update_ref is true, this function
4861 * also make sure backrefs for the shared block and all lower level
4862 * blocks are properly updated.
4864 int btrfs_drop_snapshot(struct btrfs_root
*root
, int update_ref
)
4866 struct btrfs_path
*path
;
4867 struct btrfs_trans_handle
*trans
;
4868 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
4869 struct btrfs_root_item
*root_item
= &root
->root_item
;
4870 struct walk_control
*wc
;
4871 struct btrfs_key key
;
4876 path
= btrfs_alloc_path();
4879 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
4882 trans
= btrfs_start_transaction(tree_root
, 1);
4884 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
4885 level
= btrfs_header_level(root
->node
);
4886 path
->nodes
[level
] = btrfs_lock_root_node(root
);
4887 btrfs_set_lock_blocking(path
->nodes
[level
]);
4888 path
->slots
[level
] = 0;
4889 path
->locks
[level
] = 1;
4890 memset(&wc
->update_progress
, 0,
4891 sizeof(wc
->update_progress
));
4893 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
4894 memcpy(&wc
->update_progress
, &key
,
4895 sizeof(wc
->update_progress
));
4897 level
= root_item
->drop_level
;
4899 path
->lowest_level
= level
;
4900 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4901 path
->lowest_level
= 0;
4906 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
4907 path
->slots
[level
]);
4908 WARN_ON(memcmp(&key
, &wc
->update_progress
, sizeof(key
)));
4911 * unlock our path, this is safe because only this
4912 * function is allowed to delete this snapshot
4914 btrfs_unlock_up_safe(path
, 0);
4916 level
= btrfs_header_level(root
->node
);
4918 btrfs_tree_lock(path
->nodes
[level
]);
4919 btrfs_set_lock_blocking(path
->nodes
[level
]);
4921 ret
= btrfs_lookup_extent_info(trans
, root
,
4922 path
->nodes
[level
]->start
,
4923 path
->nodes
[level
]->len
,
4927 BUG_ON(wc
->refs
[level
] == 0);
4929 if (level
== root_item
->drop_level
)
4932 btrfs_tree_unlock(path
->nodes
[level
]);
4933 WARN_ON(wc
->refs
[level
] != 1);
4939 wc
->shared_level
= -1;
4940 wc
->stage
= DROP_REFERENCE
;
4941 wc
->update_ref
= update_ref
;
4945 ret
= walk_down_tree(trans
, root
, path
, wc
);
4951 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
4958 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
4962 if (wc
->stage
== DROP_REFERENCE
) {
4964 btrfs_node_key(path
->nodes
[level
],
4965 &root_item
->drop_progress
,
4966 path
->slots
[level
]);
4967 root_item
->drop_level
= level
;
4970 BUG_ON(wc
->level
== 0);
4971 if (trans
->transaction
->in_commit
||
4972 trans
->transaction
->delayed_refs
.flushing
) {
4973 ret
= btrfs_update_root(trans
, tree_root
,
4978 btrfs_end_transaction(trans
, tree_root
);
4979 trans
= btrfs_start_transaction(tree_root
, 1);
4981 unsigned long update
;
4982 update
= trans
->delayed_ref_updates
;
4983 trans
->delayed_ref_updates
= 0;
4985 btrfs_run_delayed_refs(trans
, tree_root
,
4989 btrfs_release_path(root
, path
);
4992 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
4995 free_extent_buffer(root
->node
);
4996 free_extent_buffer(root
->commit_root
);
4999 btrfs_end_transaction(trans
, tree_root
);
5001 btrfs_free_path(path
);
5006 * drop subtree rooted at tree block 'node'.
5008 * NOTE: this function will unlock and release tree block 'node'
5010 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
5011 struct btrfs_root
*root
,
5012 struct extent_buffer
*node
,
5013 struct extent_buffer
*parent
)
5015 struct btrfs_path
*path
;
5016 struct walk_control
*wc
;
5022 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
5024 path
= btrfs_alloc_path();
5027 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
5030 btrfs_assert_tree_locked(parent
);
5031 parent_level
= btrfs_header_level(parent
);
5032 extent_buffer_get(parent
);
5033 path
->nodes
[parent_level
] = parent
;
5034 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
5036 btrfs_assert_tree_locked(node
);
5037 level
= btrfs_header_level(node
);
5038 path
->nodes
[level
] = node
;
5039 path
->slots
[level
] = 0;
5040 path
->locks
[level
] = 1;
5042 wc
->refs
[parent_level
] = 1;
5043 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
5045 wc
->shared_level
= -1;
5046 wc
->stage
= DROP_REFERENCE
;
5051 wret
= walk_down_tree(trans
, root
, path
, wc
);
5057 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
5065 btrfs_free_path(path
);
5070 static unsigned long calc_ra(unsigned long start
, unsigned long last
,
5073 return min(last
, start
+ nr
- 1);
5076 static noinline
int relocate_inode_pages(struct inode
*inode
, u64 start
,
5081 unsigned long first_index
;
5082 unsigned long last_index
;
5085 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
5086 struct file_ra_state
*ra
;
5087 struct btrfs_ordered_extent
*ordered
;
5088 unsigned int total_read
= 0;
5089 unsigned int total_dirty
= 0;
5092 ra
= kzalloc(sizeof(*ra
), GFP_NOFS
);
5094 mutex_lock(&inode
->i_mutex
);
5095 first_index
= start
>> PAGE_CACHE_SHIFT
;
5096 last_index
= (start
+ len
- 1) >> PAGE_CACHE_SHIFT
;
5098 /* make sure the dirty trick played by the caller work */
5099 ret
= invalidate_inode_pages2_range(inode
->i_mapping
,
5100 first_index
, last_index
);
5104 file_ra_state_init(ra
, inode
->i_mapping
);
5106 for (i
= first_index
; i
<= last_index
; i
++) {
5107 if (total_read
% ra
->ra_pages
== 0) {
5108 btrfs_force_ra(inode
->i_mapping
, ra
, NULL
, i
,
5109 calc_ra(i
, last_index
, ra
->ra_pages
));
5113 if (((u64
)i
<< PAGE_CACHE_SHIFT
) > i_size_read(inode
))
5115 page
= grab_cache_page(inode
->i_mapping
, i
);
5120 if (!PageUptodate(page
)) {
5121 btrfs_readpage(NULL
, page
);
5123 if (!PageUptodate(page
)) {
5125 page_cache_release(page
);
5130 wait_on_page_writeback(page
);
5132 page_start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
5133 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
5134 lock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
5136 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
5138 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
5140 page_cache_release(page
);
5141 btrfs_start_ordered_extent(inode
, ordered
, 1);
5142 btrfs_put_ordered_extent(ordered
);
5145 set_page_extent_mapped(page
);
5147 if (i
== first_index
)
5148 set_extent_bits(io_tree
, page_start
, page_end
,
5149 EXTENT_BOUNDARY
, GFP_NOFS
);
5150 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
5152 set_page_dirty(page
);
5155 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
5157 page_cache_release(page
);
5162 mutex_unlock(&inode
->i_mutex
);
5163 balance_dirty_pages_ratelimited_nr(inode
->i_mapping
, total_dirty
);
5167 static noinline
int relocate_data_extent(struct inode
*reloc_inode
,
5168 struct btrfs_key
*extent_key
,
5171 struct btrfs_root
*root
= BTRFS_I(reloc_inode
)->root
;
5172 struct extent_map_tree
*em_tree
= &BTRFS_I(reloc_inode
)->extent_tree
;
5173 struct extent_map
*em
;
5174 u64 start
= extent_key
->objectid
- offset
;
5175 u64 end
= start
+ extent_key
->offset
- 1;
5177 em
= alloc_extent_map(GFP_NOFS
);
5178 BUG_ON(!em
|| IS_ERR(em
));
5181 em
->len
= extent_key
->offset
;
5182 em
->block_len
= extent_key
->offset
;
5183 em
->block_start
= extent_key
->objectid
;
5184 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
5185 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
5187 /* setup extent map to cheat btrfs_readpage */
5188 lock_extent(&BTRFS_I(reloc_inode
)->io_tree
, start
, end
, GFP_NOFS
);
5191 spin_lock(&em_tree
->lock
);
5192 ret
= add_extent_mapping(em_tree
, em
);
5193 spin_unlock(&em_tree
->lock
);
5194 if (ret
!= -EEXIST
) {
5195 free_extent_map(em
);
5198 btrfs_drop_extent_cache(reloc_inode
, start
, end
, 0);
5200 unlock_extent(&BTRFS_I(reloc_inode
)->io_tree
, start
, end
, GFP_NOFS
);
5202 return relocate_inode_pages(reloc_inode
, start
, extent_key
->offset
);
5205 struct btrfs_ref_path
{
5207 u64 nodes
[BTRFS_MAX_LEVEL
];
5209 u64 root_generation
;
5216 struct btrfs_key node_keys
[BTRFS_MAX_LEVEL
];
5217 u64 new_nodes
[BTRFS_MAX_LEVEL
];
5220 struct disk_extent
{
5231 static int is_cowonly_root(u64 root_objectid
)
5233 if (root_objectid
== BTRFS_ROOT_TREE_OBJECTID
||
5234 root_objectid
== BTRFS_EXTENT_TREE_OBJECTID
||
5235 root_objectid
== BTRFS_CHUNK_TREE_OBJECTID
||
5236 root_objectid
== BTRFS_DEV_TREE_OBJECTID
||
5237 root_objectid
== BTRFS_TREE_LOG_OBJECTID
||
5238 root_objectid
== BTRFS_CSUM_TREE_OBJECTID
)
5243 static noinline
int __next_ref_path(struct btrfs_trans_handle
*trans
,
5244 struct btrfs_root
*extent_root
,
5245 struct btrfs_ref_path
*ref_path
,
5248 struct extent_buffer
*leaf
;
5249 struct btrfs_path
*path
;
5250 struct btrfs_extent_ref
*ref
;
5251 struct btrfs_key key
;
5252 struct btrfs_key found_key
;
5258 path
= btrfs_alloc_path();
5263 ref_path
->lowest_level
= -1;
5264 ref_path
->current_level
= -1;
5265 ref_path
->shared_level
= -1;
5269 level
= ref_path
->current_level
- 1;
5270 while (level
>= -1) {
5272 if (level
< ref_path
->lowest_level
)
5276 bytenr
= ref_path
->nodes
[level
];
5278 bytenr
= ref_path
->extent_start
;
5279 BUG_ON(bytenr
== 0);
5281 parent
= ref_path
->nodes
[level
+ 1];
5282 ref_path
->nodes
[level
+ 1] = 0;
5283 ref_path
->current_level
= level
;
5284 BUG_ON(parent
== 0);
5286 key
.objectid
= bytenr
;
5287 key
.offset
= parent
+ 1;
5288 key
.type
= BTRFS_EXTENT_REF_KEY
;
5290 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
, 0, 0);
5295 leaf
= path
->nodes
[0];
5296 nritems
= btrfs_header_nritems(leaf
);
5297 if (path
->slots
[0] >= nritems
) {
5298 ret
= btrfs_next_leaf(extent_root
, path
);
5303 leaf
= path
->nodes
[0];
5306 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5307 if (found_key
.objectid
== bytenr
&&
5308 found_key
.type
== BTRFS_EXTENT_REF_KEY
) {
5309 if (level
< ref_path
->shared_level
)
5310 ref_path
->shared_level
= level
;
5315 btrfs_release_path(extent_root
, path
);
5318 /* reached lowest level */
5322 level
= ref_path
->current_level
;
5323 while (level
< BTRFS_MAX_LEVEL
- 1) {
5327 bytenr
= ref_path
->nodes
[level
];
5329 bytenr
= ref_path
->extent_start
;
5331 BUG_ON(bytenr
== 0);
5333 key
.objectid
= bytenr
;
5335 key
.type
= BTRFS_EXTENT_REF_KEY
;
5337 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
, 0, 0);
5341 leaf
= path
->nodes
[0];
5342 nritems
= btrfs_header_nritems(leaf
);
5343 if (path
->slots
[0] >= nritems
) {
5344 ret
= btrfs_next_leaf(extent_root
, path
);
5348 /* the extent was freed by someone */
5349 if (ref_path
->lowest_level
== level
)
5351 btrfs_release_path(extent_root
, path
);
5354 leaf
= path
->nodes
[0];
5357 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5358 if (found_key
.objectid
!= bytenr
||
5359 found_key
.type
!= BTRFS_EXTENT_REF_KEY
) {
5360 /* the extent was freed by someone */
5361 if (ref_path
->lowest_level
== level
) {
5365 btrfs_release_path(extent_root
, path
);
5369 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
5370 struct btrfs_extent_ref
);
5371 ref_objectid
= btrfs_ref_objectid(leaf
, ref
);
5372 if (ref_objectid
< BTRFS_FIRST_FREE_OBJECTID
) {
5374 level
= (int)ref_objectid
;
5375 BUG_ON(level
>= BTRFS_MAX_LEVEL
);
5376 ref_path
->lowest_level
= level
;
5377 ref_path
->current_level
= level
;
5378 ref_path
->nodes
[level
] = bytenr
;
5380 WARN_ON(ref_objectid
!= level
);
5383 WARN_ON(level
!= -1);
5387 if (ref_path
->lowest_level
== level
) {
5388 ref_path
->owner_objectid
= ref_objectid
;
5389 ref_path
->num_refs
= btrfs_ref_num_refs(leaf
, ref
);
5393 * the block is tree root or the block isn't in reference
5396 if (found_key
.objectid
== found_key
.offset
||
5397 is_cowonly_root(btrfs_ref_root(leaf
, ref
))) {
5398 ref_path
->root_objectid
= btrfs_ref_root(leaf
, ref
);
5399 ref_path
->root_generation
=
5400 btrfs_ref_generation(leaf
, ref
);
5402 /* special reference from the tree log */
5403 ref_path
->nodes
[0] = found_key
.offset
;
5404 ref_path
->current_level
= 0;
5411 BUG_ON(ref_path
->nodes
[level
] != 0);
5412 ref_path
->nodes
[level
] = found_key
.offset
;
5413 ref_path
->current_level
= level
;
5416 * the reference was created in the running transaction,
5417 * no need to continue walking up.
5419 if (btrfs_ref_generation(leaf
, ref
) == trans
->transid
) {
5420 ref_path
->root_objectid
= btrfs_ref_root(leaf
, ref
);
5421 ref_path
->root_generation
=
5422 btrfs_ref_generation(leaf
, ref
);
5427 btrfs_release_path(extent_root
, path
);
5430 /* reached max tree level, but no tree root found. */
5433 btrfs_free_path(path
);
5437 static int btrfs_first_ref_path(struct btrfs_trans_handle
*trans
,
5438 struct btrfs_root
*extent_root
,
5439 struct btrfs_ref_path
*ref_path
,
5442 memset(ref_path
, 0, sizeof(*ref_path
));
5443 ref_path
->extent_start
= extent_start
;
5445 return __next_ref_path(trans
, extent_root
, ref_path
, 1);
5448 static int btrfs_next_ref_path(struct btrfs_trans_handle
*trans
,
5449 struct btrfs_root
*extent_root
,
5450 struct btrfs_ref_path
*ref_path
)
5452 return __next_ref_path(trans
, extent_root
, ref_path
, 0);
5455 static noinline
int get_new_locations(struct inode
*reloc_inode
,
5456 struct btrfs_key
*extent_key
,
5457 u64 offset
, int no_fragment
,
5458 struct disk_extent
**extents
,
5461 struct btrfs_root
*root
= BTRFS_I(reloc_inode
)->root
;
5462 struct btrfs_path
*path
;
5463 struct btrfs_file_extent_item
*fi
;
5464 struct extent_buffer
*leaf
;
5465 struct disk_extent
*exts
= *extents
;
5466 struct btrfs_key found_key
;
5471 int max
= *nr_extents
;
5474 WARN_ON(!no_fragment
&& *extents
);
5477 exts
= kmalloc(sizeof(*exts
) * max
, GFP_NOFS
);
5482 path
= btrfs_alloc_path();
5485 cur_pos
= extent_key
->objectid
- offset
;
5486 last_byte
= extent_key
->objectid
+ extent_key
->offset
;
5487 ret
= btrfs_lookup_file_extent(NULL
, root
, path
, reloc_inode
->i_ino
,
5497 leaf
= path
->nodes
[0];
5498 nritems
= btrfs_header_nritems(leaf
);
5499 if (path
->slots
[0] >= nritems
) {
5500 ret
= btrfs_next_leaf(root
, path
);
5505 leaf
= path
->nodes
[0];
5508 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5509 if (found_key
.offset
!= cur_pos
||
5510 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
||
5511 found_key
.objectid
!= reloc_inode
->i_ino
)
5514 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
5515 struct btrfs_file_extent_item
);
5516 if (btrfs_file_extent_type(leaf
, fi
) !=
5517 BTRFS_FILE_EXTENT_REG
||
5518 btrfs_file_extent_disk_bytenr(leaf
, fi
) == 0)
5522 struct disk_extent
*old
= exts
;
5524 exts
= kzalloc(sizeof(*exts
) * max
, GFP_NOFS
);
5525 memcpy(exts
, old
, sizeof(*exts
) * nr
);
5526 if (old
!= *extents
)
5530 exts
[nr
].disk_bytenr
=
5531 btrfs_file_extent_disk_bytenr(leaf
, fi
);
5532 exts
[nr
].disk_num_bytes
=
5533 btrfs_file_extent_disk_num_bytes(leaf
, fi
);
5534 exts
[nr
].offset
= btrfs_file_extent_offset(leaf
, fi
);
5535 exts
[nr
].num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
5536 exts
[nr
].ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
5537 exts
[nr
].compression
= btrfs_file_extent_compression(leaf
, fi
);
5538 exts
[nr
].encryption
= btrfs_file_extent_encryption(leaf
, fi
);
5539 exts
[nr
].other_encoding
= btrfs_file_extent_other_encoding(leaf
,
5541 BUG_ON(exts
[nr
].offset
> 0);
5542 BUG_ON(exts
[nr
].compression
|| exts
[nr
].encryption
);
5543 BUG_ON(exts
[nr
].num_bytes
!= exts
[nr
].disk_num_bytes
);
5545 cur_pos
+= exts
[nr
].num_bytes
;
5548 if (cur_pos
+ offset
>= last_byte
)
5558 BUG_ON(cur_pos
+ offset
> last_byte
);
5559 if (cur_pos
+ offset
< last_byte
) {
5565 btrfs_free_path(path
);
5567 if (exts
!= *extents
)
5576 static noinline
int replace_one_extent(struct btrfs_trans_handle
*trans
,
5577 struct btrfs_root
*root
,
5578 struct btrfs_path
*path
,
5579 struct btrfs_key
*extent_key
,
5580 struct btrfs_key
*leaf_key
,
5581 struct btrfs_ref_path
*ref_path
,
5582 struct disk_extent
*new_extents
,
5585 struct extent_buffer
*leaf
;
5586 struct btrfs_file_extent_item
*fi
;
5587 struct inode
*inode
= NULL
;
5588 struct btrfs_key key
;
5593 u64 search_end
= (u64
)-1;
5596 int extent_locked
= 0;
5600 memcpy(&key
, leaf_key
, sizeof(key
));
5601 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
) {
5602 if (key
.objectid
< ref_path
->owner_objectid
||
5603 (key
.objectid
== ref_path
->owner_objectid
&&
5604 key
.type
< BTRFS_EXTENT_DATA_KEY
)) {
5605 key
.objectid
= ref_path
->owner_objectid
;
5606 key
.type
= BTRFS_EXTENT_DATA_KEY
;
5612 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
5616 leaf
= path
->nodes
[0];
5617 nritems
= btrfs_header_nritems(leaf
);
5619 if (extent_locked
&& ret
> 0) {
5621 * the file extent item was modified by someone
5622 * before the extent got locked.
5624 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
5625 lock_end
, GFP_NOFS
);
5629 if (path
->slots
[0] >= nritems
) {
5630 if (++nr_scaned
> 2)
5633 BUG_ON(extent_locked
);
5634 ret
= btrfs_next_leaf(root
, path
);
5639 leaf
= path
->nodes
[0];
5640 nritems
= btrfs_header_nritems(leaf
);
5643 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
5645 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
) {
5646 if ((key
.objectid
> ref_path
->owner_objectid
) ||
5647 (key
.objectid
== ref_path
->owner_objectid
&&
5648 key
.type
> BTRFS_EXTENT_DATA_KEY
) ||
5649 key
.offset
>= search_end
)
5653 if (inode
&& key
.objectid
!= inode
->i_ino
) {
5654 BUG_ON(extent_locked
);
5655 btrfs_release_path(root
, path
);
5656 mutex_unlock(&inode
->i_mutex
);
5662 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
5667 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
5668 struct btrfs_file_extent_item
);
5669 extent_type
= btrfs_file_extent_type(leaf
, fi
);
5670 if ((extent_type
!= BTRFS_FILE_EXTENT_REG
&&
5671 extent_type
!= BTRFS_FILE_EXTENT_PREALLOC
) ||
5672 (btrfs_file_extent_disk_bytenr(leaf
, fi
) !=
5673 extent_key
->objectid
)) {
5679 num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
5680 ext_offset
= btrfs_file_extent_offset(leaf
, fi
);
5682 if (search_end
== (u64
)-1) {
5683 search_end
= key
.offset
- ext_offset
+
5684 btrfs_file_extent_ram_bytes(leaf
, fi
);
5687 if (!extent_locked
) {
5688 lock_start
= key
.offset
;
5689 lock_end
= lock_start
+ num_bytes
- 1;
5691 if (lock_start
> key
.offset
||
5692 lock_end
+ 1 < key
.offset
+ num_bytes
) {
5693 unlock_extent(&BTRFS_I(inode
)->io_tree
,
5694 lock_start
, lock_end
, GFP_NOFS
);
5700 btrfs_release_path(root
, path
);
5702 inode
= btrfs_iget_locked(root
->fs_info
->sb
,
5703 key
.objectid
, root
);
5704 if (inode
->i_state
& I_NEW
) {
5705 BTRFS_I(inode
)->root
= root
;
5706 BTRFS_I(inode
)->location
.objectid
=
5708 BTRFS_I(inode
)->location
.type
=
5709 BTRFS_INODE_ITEM_KEY
;
5710 BTRFS_I(inode
)->location
.offset
= 0;
5711 btrfs_read_locked_inode(inode
);
5712 unlock_new_inode(inode
);
5715 * some code call btrfs_commit_transaction while
5716 * holding the i_mutex, so we can't use mutex_lock
5719 if (is_bad_inode(inode
) ||
5720 !mutex_trylock(&inode
->i_mutex
)) {
5723 key
.offset
= (u64
)-1;
5728 if (!extent_locked
) {
5729 struct btrfs_ordered_extent
*ordered
;
5731 btrfs_release_path(root
, path
);
5733 lock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
5734 lock_end
, GFP_NOFS
);
5735 ordered
= btrfs_lookup_first_ordered_extent(inode
,
5738 ordered
->file_offset
<= lock_end
&&
5739 ordered
->file_offset
+ ordered
->len
> lock_start
) {
5740 unlock_extent(&BTRFS_I(inode
)->io_tree
,
5741 lock_start
, lock_end
, GFP_NOFS
);
5742 btrfs_start_ordered_extent(inode
, ordered
, 1);
5743 btrfs_put_ordered_extent(ordered
);
5744 key
.offset
+= num_bytes
;
5748 btrfs_put_ordered_extent(ordered
);
5754 if (nr_extents
== 1) {
5755 /* update extent pointer in place */
5756 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
5757 new_extents
[0].disk_bytenr
);
5758 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
5759 new_extents
[0].disk_num_bytes
);
5760 btrfs_mark_buffer_dirty(leaf
);
5762 btrfs_drop_extent_cache(inode
, key
.offset
,
5763 key
.offset
+ num_bytes
- 1, 0);
5765 ret
= btrfs_inc_extent_ref(trans
, root
,
5766 new_extents
[0].disk_bytenr
,
5767 new_extents
[0].disk_num_bytes
,
5769 root
->root_key
.objectid
,
5774 ret
= btrfs_free_extent(trans
, root
,
5775 extent_key
->objectid
,
5778 btrfs_header_owner(leaf
),
5779 btrfs_header_generation(leaf
),
5783 btrfs_release_path(root
, path
);
5784 key
.offset
+= num_bytes
;
5792 * drop old extent pointer at first, then insert the
5793 * new pointers one bye one
5795 btrfs_release_path(root
, path
);
5796 ret
= btrfs_drop_extents(trans
, root
, inode
, key
.offset
,
5797 key
.offset
+ num_bytes
,
5798 key
.offset
, &alloc_hint
);
5801 for (i
= 0; i
< nr_extents
; i
++) {
5802 if (ext_offset
>= new_extents
[i
].num_bytes
) {
5803 ext_offset
-= new_extents
[i
].num_bytes
;
5806 extent_len
= min(new_extents
[i
].num_bytes
-
5807 ext_offset
, num_bytes
);
5809 ret
= btrfs_insert_empty_item(trans
, root
,
5814 leaf
= path
->nodes
[0];
5815 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
5816 struct btrfs_file_extent_item
);
5817 btrfs_set_file_extent_generation(leaf
, fi
,
5819 btrfs_set_file_extent_type(leaf
, fi
,
5820 BTRFS_FILE_EXTENT_REG
);
5821 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
5822 new_extents
[i
].disk_bytenr
);
5823 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
5824 new_extents
[i
].disk_num_bytes
);
5825 btrfs_set_file_extent_ram_bytes(leaf
, fi
,
5826 new_extents
[i
].ram_bytes
);
5828 btrfs_set_file_extent_compression(leaf
, fi
,
5829 new_extents
[i
].compression
);
5830 btrfs_set_file_extent_encryption(leaf
, fi
,
5831 new_extents
[i
].encryption
);
5832 btrfs_set_file_extent_other_encoding(leaf
, fi
,
5833 new_extents
[i
].other_encoding
);
5835 btrfs_set_file_extent_num_bytes(leaf
, fi
,
5837 ext_offset
+= new_extents
[i
].offset
;
5838 btrfs_set_file_extent_offset(leaf
, fi
,
5840 btrfs_mark_buffer_dirty(leaf
);
5842 btrfs_drop_extent_cache(inode
, key
.offset
,
5843 key
.offset
+ extent_len
- 1, 0);
5845 ret
= btrfs_inc_extent_ref(trans
, root
,
5846 new_extents
[i
].disk_bytenr
,
5847 new_extents
[i
].disk_num_bytes
,
5849 root
->root_key
.objectid
,
5850 trans
->transid
, key
.objectid
);
5852 btrfs_release_path(root
, path
);
5854 inode_add_bytes(inode
, extent_len
);
5857 num_bytes
-= extent_len
;
5858 key
.offset
+= extent_len
;
5863 BUG_ON(i
>= nr_extents
);
5867 if (extent_locked
) {
5868 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
5869 lock_end
, GFP_NOFS
);
5873 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
&&
5874 key
.offset
>= search_end
)
5881 btrfs_release_path(root
, path
);
5883 mutex_unlock(&inode
->i_mutex
);
5884 if (extent_locked
) {
5885 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
5886 lock_end
, GFP_NOFS
);
5893 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle
*trans
,
5894 struct btrfs_root
*root
,
5895 struct extent_buffer
*buf
, u64 orig_start
)
5900 BUG_ON(btrfs_header_generation(buf
) != trans
->transid
);
5901 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
5903 level
= btrfs_header_level(buf
);
5905 struct btrfs_leaf_ref
*ref
;
5906 struct btrfs_leaf_ref
*orig_ref
;
5908 orig_ref
= btrfs_lookup_leaf_ref(root
, orig_start
);
5912 ref
= btrfs_alloc_leaf_ref(root
, orig_ref
->nritems
);
5914 btrfs_free_leaf_ref(root
, orig_ref
);
5918 ref
->nritems
= orig_ref
->nritems
;
5919 memcpy(ref
->extents
, orig_ref
->extents
,
5920 sizeof(ref
->extents
[0]) * ref
->nritems
);
5922 btrfs_free_leaf_ref(root
, orig_ref
);
5924 ref
->root_gen
= trans
->transid
;
5925 ref
->bytenr
= buf
->start
;
5926 ref
->owner
= btrfs_header_owner(buf
);
5927 ref
->generation
= btrfs_header_generation(buf
);
5929 ret
= btrfs_add_leaf_ref(root
, ref
, 0);
5931 btrfs_free_leaf_ref(root
, ref
);
5936 static noinline
int invalidate_extent_cache(struct btrfs_root
*root
,
5937 struct extent_buffer
*leaf
,
5938 struct btrfs_block_group_cache
*group
,
5939 struct btrfs_root
*target_root
)
5941 struct btrfs_key key
;
5942 struct inode
*inode
= NULL
;
5943 struct btrfs_file_extent_item
*fi
;
5945 u64 skip_objectid
= 0;
5949 nritems
= btrfs_header_nritems(leaf
);
5950 for (i
= 0; i
< nritems
; i
++) {
5951 btrfs_item_key_to_cpu(leaf
, &key
, i
);
5952 if (key
.objectid
== skip_objectid
||
5953 key
.type
!= BTRFS_EXTENT_DATA_KEY
)
5955 fi
= btrfs_item_ptr(leaf
, i
, struct btrfs_file_extent_item
);
5956 if (btrfs_file_extent_type(leaf
, fi
) ==
5957 BTRFS_FILE_EXTENT_INLINE
)
5959 if (btrfs_file_extent_disk_bytenr(leaf
, fi
) == 0)
5961 if (!inode
|| inode
->i_ino
!= key
.objectid
) {
5963 inode
= btrfs_ilookup(target_root
->fs_info
->sb
,
5964 key
.objectid
, target_root
, 1);
5967 skip_objectid
= key
.objectid
;
5970 num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
5972 lock_extent(&BTRFS_I(inode
)->io_tree
, key
.offset
,
5973 key
.offset
+ num_bytes
- 1, GFP_NOFS
);
5974 btrfs_drop_extent_cache(inode
, key
.offset
,
5975 key
.offset
+ num_bytes
- 1, 1);
5976 unlock_extent(&BTRFS_I(inode
)->io_tree
, key
.offset
,
5977 key
.offset
+ num_bytes
- 1, GFP_NOFS
);
5984 static noinline
int replace_extents_in_leaf(struct btrfs_trans_handle
*trans
,
5985 struct btrfs_root
*root
,
5986 struct extent_buffer
*leaf
,
5987 struct btrfs_block_group_cache
*group
,
5988 struct inode
*reloc_inode
)
5990 struct btrfs_key key
;
5991 struct btrfs_key extent_key
;
5992 struct btrfs_file_extent_item
*fi
;
5993 struct btrfs_leaf_ref
*ref
;
5994 struct disk_extent
*new_extent
;
6003 new_extent
= kmalloc(sizeof(*new_extent
), GFP_NOFS
);
6004 BUG_ON(!new_extent
);
6006 ref
= btrfs_lookup_leaf_ref(root
, leaf
->start
);
6010 nritems
= btrfs_header_nritems(leaf
);
6011 for (i
= 0; i
< nritems
; i
++) {
6012 btrfs_item_key_to_cpu(leaf
, &key
, i
);
6013 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
6015 fi
= btrfs_item_ptr(leaf
, i
, struct btrfs_file_extent_item
);
6016 if (btrfs_file_extent_type(leaf
, fi
) ==
6017 BTRFS_FILE_EXTENT_INLINE
)
6019 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
6020 num_bytes
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
6025 if (bytenr
>= group
->key
.objectid
+ group
->key
.offset
||
6026 bytenr
+ num_bytes
<= group
->key
.objectid
)
6029 extent_key
.objectid
= bytenr
;
6030 extent_key
.offset
= num_bytes
;
6031 extent_key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6033 ret
= get_new_locations(reloc_inode
, &extent_key
,
6034 group
->key
.objectid
, 1,
6035 &new_extent
, &nr_extent
);
6040 BUG_ON(ref
->extents
[ext_index
].bytenr
!= bytenr
);
6041 BUG_ON(ref
->extents
[ext_index
].num_bytes
!= num_bytes
);
6042 ref
->extents
[ext_index
].bytenr
= new_extent
->disk_bytenr
;
6043 ref
->extents
[ext_index
].num_bytes
= new_extent
->disk_num_bytes
;
6045 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
6046 new_extent
->disk_bytenr
);
6047 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
6048 new_extent
->disk_num_bytes
);
6049 btrfs_mark_buffer_dirty(leaf
);
6051 ret
= btrfs_inc_extent_ref(trans
, root
,
6052 new_extent
->disk_bytenr
,
6053 new_extent
->disk_num_bytes
,
6055 root
->root_key
.objectid
,
6056 trans
->transid
, key
.objectid
);
6059 ret
= btrfs_free_extent(trans
, root
,
6060 bytenr
, num_bytes
, leaf
->start
,
6061 btrfs_header_owner(leaf
),
6062 btrfs_header_generation(leaf
),
6068 BUG_ON(ext_index
+ 1 != ref
->nritems
);
6069 btrfs_free_leaf_ref(root
, ref
);
6073 int btrfs_free_reloc_root(struct btrfs_trans_handle
*trans
,
6074 struct btrfs_root
*root
)
6076 struct btrfs_root
*reloc_root
;
6079 if (root
->reloc_root
) {
6080 reloc_root
= root
->reloc_root
;
6081 root
->reloc_root
= NULL
;
6082 list_add(&reloc_root
->dead_list
,
6083 &root
->fs_info
->dead_reloc_roots
);
6085 btrfs_set_root_bytenr(&reloc_root
->root_item
,
6086 reloc_root
->node
->start
);
6087 btrfs_set_root_level(&root
->root_item
,
6088 btrfs_header_level(reloc_root
->node
));
6089 memset(&reloc_root
->root_item
.drop_progress
, 0,
6090 sizeof(struct btrfs_disk_key
));
6091 reloc_root
->root_item
.drop_level
= 0;
6093 ret
= btrfs_update_root(trans
, root
->fs_info
->tree_root
,
6094 &reloc_root
->root_key
,
6095 &reloc_root
->root_item
);
6101 int btrfs_drop_dead_reloc_roots(struct btrfs_root
*root
)
6103 struct btrfs_trans_handle
*trans
;
6104 struct btrfs_root
*reloc_root
;
6105 struct btrfs_root
*prev_root
= NULL
;
6106 struct list_head dead_roots
;
6110 INIT_LIST_HEAD(&dead_roots
);
6111 list_splice_init(&root
->fs_info
->dead_reloc_roots
, &dead_roots
);
6113 while (!list_empty(&dead_roots
)) {
6114 reloc_root
= list_entry(dead_roots
.prev
,
6115 struct btrfs_root
, dead_list
);
6116 list_del_init(&reloc_root
->dead_list
);
6118 BUG_ON(reloc_root
->commit_root
!= NULL
);
6120 trans
= btrfs_join_transaction(root
, 1);
6123 mutex_lock(&root
->fs_info
->drop_mutex
);
6124 ret
= btrfs_drop_snapshot(trans
, reloc_root
);
6127 mutex_unlock(&root
->fs_info
->drop_mutex
);
6129 nr
= trans
->blocks_used
;
6130 ret
= btrfs_end_transaction(trans
, root
);
6132 btrfs_btree_balance_dirty(root
, nr
);
6135 free_extent_buffer(reloc_root
->node
);
6137 ret
= btrfs_del_root(trans
, root
->fs_info
->tree_root
,
6138 &reloc_root
->root_key
);
6140 mutex_unlock(&root
->fs_info
->drop_mutex
);
6142 nr
= trans
->blocks_used
;
6143 ret
= btrfs_end_transaction(trans
, root
);
6145 btrfs_btree_balance_dirty(root
, nr
);
6148 prev_root
= reloc_root
;
6151 btrfs_remove_leaf_refs(prev_root
, (u64
)-1, 0);
6157 int btrfs_add_dead_reloc_root(struct btrfs_root
*root
)
6159 list_add(&root
->dead_list
, &root
->fs_info
->dead_reloc_roots
);
6163 int btrfs_cleanup_reloc_trees(struct btrfs_root
*root
)
6165 struct btrfs_root
*reloc_root
;
6166 struct btrfs_trans_handle
*trans
;
6167 struct btrfs_key location
;
6171 mutex_lock(&root
->fs_info
->tree_reloc_mutex
);
6172 ret
= btrfs_find_dead_roots(root
, BTRFS_TREE_RELOC_OBJECTID
, NULL
);
6174 found
= !list_empty(&root
->fs_info
->dead_reloc_roots
);
6175 mutex_unlock(&root
->fs_info
->tree_reloc_mutex
);
6178 trans
= btrfs_start_transaction(root
, 1);
6180 ret
= btrfs_commit_transaction(trans
, root
);
6184 location
.objectid
= BTRFS_DATA_RELOC_TREE_OBJECTID
;
6185 location
.offset
= (u64
)-1;
6186 location
.type
= BTRFS_ROOT_ITEM_KEY
;
6188 reloc_root
= btrfs_read_fs_root_no_name(root
->fs_info
, &location
);
6189 BUG_ON(!reloc_root
);
6190 btrfs_orphan_cleanup(reloc_root
);
6194 static noinline
int init_reloc_tree(struct btrfs_trans_handle
*trans
,
6195 struct btrfs_root
*root
)
6197 struct btrfs_root
*reloc_root
;
6198 struct extent_buffer
*eb
;
6199 struct btrfs_root_item
*root_item
;
6200 struct btrfs_key root_key
;
6203 BUG_ON(!root
->ref_cows
);
6204 if (root
->reloc_root
)
6207 root_item
= kmalloc(sizeof(*root_item
), GFP_NOFS
);
6210 ret
= btrfs_copy_root(trans
, root
, root
->commit_root
,
6211 &eb
, BTRFS_TREE_RELOC_OBJECTID
);
6214 root_key
.objectid
= BTRFS_TREE_RELOC_OBJECTID
;
6215 root_key
.offset
= root
->root_key
.objectid
;
6216 root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
6218 memcpy(root_item
, &root
->root_item
, sizeof(root_item
));
6219 btrfs_set_root_refs(root_item
, 0);
6220 btrfs_set_root_bytenr(root_item
, eb
->start
);
6221 btrfs_set_root_level(root_item
, btrfs_header_level(eb
));
6222 btrfs_set_root_generation(root_item
, trans
->transid
);
6224 btrfs_tree_unlock(eb
);
6225 free_extent_buffer(eb
);
6227 ret
= btrfs_insert_root(trans
, root
->fs_info
->tree_root
,
6228 &root_key
, root_item
);
6232 reloc_root
= btrfs_read_fs_root_no_radix(root
->fs_info
->tree_root
,
6234 BUG_ON(!reloc_root
);
6235 reloc_root
->last_trans
= trans
->transid
;
6236 reloc_root
->commit_root
= NULL
;
6237 reloc_root
->ref_tree
= &root
->fs_info
->reloc_ref_tree
;
6239 root
->reloc_root
= reloc_root
;
6244 * Core function of space balance.
6246 * The idea is using reloc trees to relocate tree blocks in reference
6247 * counted roots. There is one reloc tree for each subvol, and all
6248 * reloc trees share same root key objectid. Reloc trees are snapshots
6249 * of the latest committed roots of subvols (root->commit_root).
6251 * To relocate a tree block referenced by a subvol, there are two steps.
6252 * COW the block through subvol's reloc tree, then update block pointer
6253 * in the subvol to point to the new block. Since all reloc trees share
6254 * same root key objectid, doing special handing for tree blocks owned
6255 * by them is easy. Once a tree block has been COWed in one reloc tree,
6256 * we can use the resulting new block directly when the same block is
6257 * required to COW again through other reloc trees. By this way, relocated
6258 * tree blocks are shared between reloc trees, so they are also shared
6261 static noinline
int relocate_one_path(struct btrfs_trans_handle
*trans
,
6262 struct btrfs_root
*root
,
6263 struct btrfs_path
*path
,
6264 struct btrfs_key
*first_key
,
6265 struct btrfs_ref_path
*ref_path
,
6266 struct btrfs_block_group_cache
*group
,
6267 struct inode
*reloc_inode
)
6269 struct btrfs_root
*reloc_root
;
6270 struct extent_buffer
*eb
= NULL
;
6271 struct btrfs_key
*keys
;
6275 int lowest_level
= 0;
6278 if (ref_path
->owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
)
6279 lowest_level
= ref_path
->owner_objectid
;
6281 if (!root
->ref_cows
) {
6282 path
->lowest_level
= lowest_level
;
6283 ret
= btrfs_search_slot(trans
, root
, first_key
, path
, 0, 1);
6285 path
->lowest_level
= 0;
6286 btrfs_release_path(root
, path
);
6290 mutex_lock(&root
->fs_info
->tree_reloc_mutex
);
6291 ret
= init_reloc_tree(trans
, root
);
6293 reloc_root
= root
->reloc_root
;
6295 shared_level
= ref_path
->shared_level
;
6296 ref_path
->shared_level
= BTRFS_MAX_LEVEL
- 1;
6298 keys
= ref_path
->node_keys
;
6299 nodes
= ref_path
->new_nodes
;
6300 memset(&keys
[shared_level
+ 1], 0,
6301 sizeof(*keys
) * (BTRFS_MAX_LEVEL
- shared_level
- 1));
6302 memset(&nodes
[shared_level
+ 1], 0,
6303 sizeof(*nodes
) * (BTRFS_MAX_LEVEL
- shared_level
- 1));
6305 if (nodes
[lowest_level
] == 0) {
6306 path
->lowest_level
= lowest_level
;
6307 ret
= btrfs_search_slot(trans
, reloc_root
, first_key
, path
,
6310 for (level
= lowest_level
; level
< BTRFS_MAX_LEVEL
; level
++) {
6311 eb
= path
->nodes
[level
];
6312 if (!eb
|| eb
== reloc_root
->node
)
6314 nodes
[level
] = eb
->start
;
6316 btrfs_item_key_to_cpu(eb
, &keys
[level
], 0);
6318 btrfs_node_key_to_cpu(eb
, &keys
[level
], 0);
6321 ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
6322 eb
= path
->nodes
[0];
6323 ret
= replace_extents_in_leaf(trans
, reloc_root
, eb
,
6324 group
, reloc_inode
);
6327 btrfs_release_path(reloc_root
, path
);
6329 ret
= btrfs_merge_path(trans
, reloc_root
, keys
, nodes
,
6335 * replace tree blocks in the fs tree with tree blocks in
6338 ret
= btrfs_merge_path(trans
, root
, keys
, nodes
, lowest_level
);
6341 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
6342 ret
= btrfs_search_slot(trans
, reloc_root
, first_key
, path
,
6345 extent_buffer_get(path
->nodes
[0]);
6346 eb
= path
->nodes
[0];
6347 btrfs_release_path(reloc_root
, path
);
6348 ret
= invalidate_extent_cache(reloc_root
, eb
, group
, root
);
6350 free_extent_buffer(eb
);
6353 mutex_unlock(&root
->fs_info
->tree_reloc_mutex
);
6354 path
->lowest_level
= 0;
6358 static noinline
int relocate_tree_block(struct btrfs_trans_handle
*trans
,
6359 struct btrfs_root
*root
,
6360 struct btrfs_path
*path
,
6361 struct btrfs_key
*first_key
,
6362 struct btrfs_ref_path
*ref_path
)
6366 ret
= relocate_one_path(trans
, root
, path
, first_key
,
6367 ref_path
, NULL
, NULL
);
6373 static noinline
int del_extent_zero(struct btrfs_trans_handle
*trans
,
6374 struct btrfs_root
*extent_root
,
6375 struct btrfs_path
*path
,
6376 struct btrfs_key
*extent_key
)
6380 ret
= btrfs_search_slot(trans
, extent_root
, extent_key
, path
, -1, 1);
6383 ret
= btrfs_del_item(trans
, extent_root
, path
);
6385 btrfs_release_path(extent_root
, path
);
6389 static noinline
struct btrfs_root
*read_ref_root(struct btrfs_fs_info
*fs_info
,
6390 struct btrfs_ref_path
*ref_path
)
6392 struct btrfs_key root_key
;
6394 root_key
.objectid
= ref_path
->root_objectid
;
6395 root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
6396 if (is_cowonly_root(ref_path
->root_objectid
))
6397 root_key
.offset
= 0;
6399 root_key
.offset
= (u64
)-1;
6401 return btrfs_read_fs_root_no_name(fs_info
, &root_key
);
6404 static noinline
int relocate_one_extent(struct btrfs_root
*extent_root
,
6405 struct btrfs_path
*path
,
6406 struct btrfs_key
*extent_key
,
6407 struct btrfs_block_group_cache
*group
,
6408 struct inode
*reloc_inode
, int pass
)
6410 struct btrfs_trans_handle
*trans
;
6411 struct btrfs_root
*found_root
;
6412 struct btrfs_ref_path
*ref_path
= NULL
;
6413 struct disk_extent
*new_extents
= NULL
;
6418 struct btrfs_key first_key
;
6422 trans
= btrfs_start_transaction(extent_root
, 1);
6425 if (extent_key
->objectid
== 0) {
6426 ret
= del_extent_zero(trans
, extent_root
, path
, extent_key
);
6430 ref_path
= kmalloc(sizeof(*ref_path
), GFP_NOFS
);
6436 for (loops
= 0; ; loops
++) {
6438 ret
= btrfs_first_ref_path(trans
, extent_root
, ref_path
,
6439 extent_key
->objectid
);
6441 ret
= btrfs_next_ref_path(trans
, extent_root
, ref_path
);
6448 if (ref_path
->root_objectid
== BTRFS_TREE_LOG_OBJECTID
||
6449 ref_path
->root_objectid
== BTRFS_TREE_RELOC_OBJECTID
)
6452 found_root
= read_ref_root(extent_root
->fs_info
, ref_path
);
6453 BUG_ON(!found_root
);
6455 * for reference counted tree, only process reference paths
6456 * rooted at the latest committed root.
6458 if (found_root
->ref_cows
&&
6459 ref_path
->root_generation
!= found_root
->root_key
.offset
)
6462 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
6465 * copy data extents to new locations
6467 u64 group_start
= group
->key
.objectid
;
6468 ret
= relocate_data_extent(reloc_inode
,
6477 level
= ref_path
->owner_objectid
;
6480 if (prev_block
!= ref_path
->nodes
[level
]) {
6481 struct extent_buffer
*eb
;
6482 u64 block_start
= ref_path
->nodes
[level
];
6483 u64 block_size
= btrfs_level_size(found_root
, level
);
6485 eb
= read_tree_block(found_root
, block_start
,
6487 btrfs_tree_lock(eb
);
6488 BUG_ON(level
!= btrfs_header_level(eb
));
6491 btrfs_item_key_to_cpu(eb
, &first_key
, 0);
6493 btrfs_node_key_to_cpu(eb
, &first_key
, 0);
6495 btrfs_tree_unlock(eb
);
6496 free_extent_buffer(eb
);
6497 prev_block
= block_start
;
6500 mutex_lock(&extent_root
->fs_info
->trans_mutex
);
6501 btrfs_record_root_in_trans(found_root
);
6502 mutex_unlock(&extent_root
->fs_info
->trans_mutex
);
6503 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
6505 * try to update data extent references while
6506 * keeping metadata shared between snapshots.
6509 ret
= relocate_one_path(trans
, found_root
,
6510 path
, &first_key
, ref_path
,
6511 group
, reloc_inode
);
6517 * use fallback method to process the remaining
6521 u64 group_start
= group
->key
.objectid
;
6522 new_extents
= kmalloc(sizeof(*new_extents
),
6525 ret
= get_new_locations(reloc_inode
,
6533 ret
= replace_one_extent(trans
, found_root
,
6535 &first_key
, ref_path
,
6536 new_extents
, nr_extents
);
6538 ret
= relocate_tree_block(trans
, found_root
, path
,
6539 &first_key
, ref_path
);
6546 btrfs_end_transaction(trans
, extent_root
);
6553 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
6556 u64 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
6557 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
6559 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
6560 if (num_devices
== 1) {
6561 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
6562 stripped
= flags
& ~stripped
;
6564 /* turn raid0 into single device chunks */
6565 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
6568 /* turn mirroring into duplication */
6569 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
6570 BTRFS_BLOCK_GROUP_RAID10
))
6571 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
6574 /* they already had raid on here, just return */
6575 if (flags
& stripped
)
6578 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
6579 stripped
= flags
& ~stripped
;
6581 /* switch duplicated blocks with raid1 */
6582 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
6583 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
6585 /* turn single device chunks into raid0 */
6586 return stripped
| BTRFS_BLOCK_GROUP_RAID0
;
6591 static int __alloc_chunk_for_shrink(struct btrfs_root
*root
,
6592 struct btrfs_block_group_cache
*shrink_block_group
,
6595 struct btrfs_trans_handle
*trans
;
6596 u64 new_alloc_flags
;
6599 spin_lock(&shrink_block_group
->lock
);
6600 if (btrfs_block_group_used(&shrink_block_group
->item
) +
6601 shrink_block_group
->reserved
> 0) {
6602 spin_unlock(&shrink_block_group
->lock
);
6604 trans
= btrfs_start_transaction(root
, 1);
6605 spin_lock(&shrink_block_group
->lock
);
6607 new_alloc_flags
= update_block_group_flags(root
,
6608 shrink_block_group
->flags
);
6609 if (new_alloc_flags
!= shrink_block_group
->flags
) {
6611 btrfs_block_group_used(&shrink_block_group
->item
);
6613 calc
= shrink_block_group
->key
.offset
;
6615 spin_unlock(&shrink_block_group
->lock
);
6617 do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
6618 calc
+ 2 * 1024 * 1024, new_alloc_flags
, force
);
6620 btrfs_end_transaction(trans
, root
);
6622 spin_unlock(&shrink_block_group
->lock
);
6627 int btrfs_prepare_block_group_relocation(struct btrfs_root
*root
,
6628 struct btrfs_block_group_cache
*group
)
6631 __alloc_chunk_for_shrink(root
, group
, 1);
6632 set_block_group_readonly(group
);
6637 static int __insert_orphan_inode(struct btrfs_trans_handle
*trans
,
6638 struct btrfs_root
*root
,
6639 u64 objectid
, u64 size
)
6641 struct btrfs_path
*path
;
6642 struct btrfs_inode_item
*item
;
6643 struct extent_buffer
*leaf
;
6646 path
= btrfs_alloc_path();
6650 path
->leave_spinning
= 1;
6651 ret
= btrfs_insert_empty_inode(trans
, root
, path
, objectid
);
6655 leaf
= path
->nodes
[0];
6656 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_inode_item
);
6657 memset_extent_buffer(leaf
, 0, (unsigned long)item
, sizeof(*item
));
6658 btrfs_set_inode_generation(leaf
, item
, 1);
6659 btrfs_set_inode_size(leaf
, item
, size
);
6660 btrfs_set_inode_mode(leaf
, item
, S_IFREG
| 0600);
6661 btrfs_set_inode_flags(leaf
, item
, BTRFS_INODE_NOCOMPRESS
);
6662 btrfs_mark_buffer_dirty(leaf
);
6663 btrfs_release_path(root
, path
);
6665 btrfs_free_path(path
);
6669 static noinline
struct inode
*create_reloc_inode(struct btrfs_fs_info
*fs_info
,
6670 struct btrfs_block_group_cache
*group
)
6672 struct inode
*inode
= NULL
;
6673 struct btrfs_trans_handle
*trans
;
6674 struct btrfs_root
*root
;
6675 struct btrfs_key root_key
;
6676 u64 objectid
= BTRFS_FIRST_FREE_OBJECTID
;
6679 root_key
.objectid
= BTRFS_DATA_RELOC_TREE_OBJECTID
;
6680 root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
6681 root_key
.offset
= (u64
)-1;
6682 root
= btrfs_read_fs_root_no_name(fs_info
, &root_key
);
6684 return ERR_CAST(root
);
6686 trans
= btrfs_start_transaction(root
, 1);
6689 err
= btrfs_find_free_objectid(trans
, root
, objectid
, &objectid
);
6693 err
= __insert_orphan_inode(trans
, root
, objectid
, group
->key
.offset
);
6696 err
= btrfs_insert_file_extent(trans
, root
, objectid
, 0, 0, 0,
6697 group
->key
.offset
, 0, group
->key
.offset
,
6701 inode
= btrfs_iget_locked(root
->fs_info
->sb
, objectid
, root
);
6702 if (inode
->i_state
& I_NEW
) {
6703 BTRFS_I(inode
)->root
= root
;
6704 BTRFS_I(inode
)->location
.objectid
= objectid
;
6705 BTRFS_I(inode
)->location
.type
= BTRFS_INODE_ITEM_KEY
;
6706 BTRFS_I(inode
)->location
.offset
= 0;
6707 btrfs_read_locked_inode(inode
);
6708 unlock_new_inode(inode
);
6709 BUG_ON(is_bad_inode(inode
));
6713 BTRFS_I(inode
)->index_cnt
= group
->key
.objectid
;
6715 err
= btrfs_orphan_add(trans
, inode
);
6717 btrfs_end_transaction(trans
, root
);
6721 inode
= ERR_PTR(err
);
6726 int btrfs_reloc_clone_csums(struct inode
*inode
, u64 file_pos
, u64 len
)
6729 struct btrfs_ordered_sum
*sums
;
6730 struct btrfs_sector_sum
*sector_sum
;
6731 struct btrfs_ordered_extent
*ordered
;
6732 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6733 struct list_head list
;
6738 INIT_LIST_HEAD(&list
);
6740 ordered
= btrfs_lookup_ordered_extent(inode
, file_pos
);
6741 BUG_ON(ordered
->file_offset
!= file_pos
|| ordered
->len
!= len
);
6743 disk_bytenr
= file_pos
+ BTRFS_I(inode
)->index_cnt
;
6744 ret
= btrfs_lookup_csums_range(root
->fs_info
->csum_root
, disk_bytenr
,
6745 disk_bytenr
+ len
- 1, &list
);
6747 while (!list_empty(&list
)) {
6748 sums
= list_entry(list
.next
, struct btrfs_ordered_sum
, list
);
6749 list_del_init(&sums
->list
);
6751 sector_sum
= sums
->sums
;
6752 sums
->bytenr
= ordered
->start
;
6755 while (offset
< sums
->len
) {
6756 sector_sum
->bytenr
+= ordered
->start
- disk_bytenr
;
6758 offset
+= root
->sectorsize
;
6761 btrfs_add_ordered_sum(inode
, ordered
, sums
);
6763 btrfs_put_ordered_extent(ordered
);
6767 int btrfs_relocate_block_group(struct btrfs_root
*root
, u64 group_start
)
6769 struct btrfs_trans_handle
*trans
;
6770 struct btrfs_path
*path
;
6771 struct btrfs_fs_info
*info
= root
->fs_info
;
6772 struct extent_buffer
*leaf
;
6773 struct inode
*reloc_inode
;
6774 struct btrfs_block_group_cache
*block_group
;
6775 struct btrfs_key key
;
6784 root
= root
->fs_info
->extent_root
;
6786 block_group
= btrfs_lookup_block_group(info
, group_start
);
6787 BUG_ON(!block_group
);
6789 printk(KERN_INFO
"btrfs relocating block group %llu flags %llu\n",
6790 (unsigned long long)block_group
->key
.objectid
,
6791 (unsigned long long)block_group
->flags
);
6793 path
= btrfs_alloc_path();
6796 reloc_inode
= create_reloc_inode(info
, block_group
);
6797 BUG_ON(IS_ERR(reloc_inode
));
6799 __alloc_chunk_for_shrink(root
, block_group
, 1);
6800 set_block_group_readonly(block_group
);
6802 btrfs_start_delalloc_inodes(info
->tree_root
);
6803 btrfs_wait_ordered_extents(info
->tree_root
, 0);
6808 key
.objectid
= block_group
->key
.objectid
;
6811 cur_byte
= key
.objectid
;
6813 trans
= btrfs_start_transaction(info
->tree_root
, 1);
6814 btrfs_commit_transaction(trans
, info
->tree_root
);
6816 mutex_lock(&root
->fs_info
->cleaner_mutex
);
6817 btrfs_clean_old_snapshots(info
->tree_root
);
6818 btrfs_remove_leaf_refs(info
->tree_root
, (u64
)-1, 1);
6819 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
6821 trans
= btrfs_start_transaction(info
->tree_root
, 1);
6822 btrfs_commit_transaction(trans
, info
->tree_root
);
6825 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6829 leaf
= path
->nodes
[0];
6830 nritems
= btrfs_header_nritems(leaf
);
6831 if (path
->slots
[0] >= nritems
) {
6832 ret
= btrfs_next_leaf(root
, path
);
6839 leaf
= path
->nodes
[0];
6840 nritems
= btrfs_header_nritems(leaf
);
6843 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
6845 if (key
.objectid
>= block_group
->key
.objectid
+
6846 block_group
->key
.offset
)
6849 if (progress
&& need_resched()) {
6850 btrfs_release_path(root
, path
);
6857 if (btrfs_key_type(&key
) != BTRFS_EXTENT_ITEM_KEY
||
6858 key
.objectid
+ key
.offset
<= cur_byte
) {
6864 cur_byte
= key
.objectid
+ key
.offset
;
6865 btrfs_release_path(root
, path
);
6867 __alloc_chunk_for_shrink(root
, block_group
, 0);
6868 ret
= relocate_one_extent(root
, path
, &key
, block_group
,
6874 key
.objectid
= cur_byte
;
6879 btrfs_release_path(root
, path
);
6882 btrfs_wait_ordered_range(reloc_inode
, 0, (u64
)-1);
6883 invalidate_mapping_pages(reloc_inode
->i_mapping
, 0, -1);
6886 if (total_found
> 0) {
6887 printk(KERN_INFO
"btrfs found %llu extents in pass %d\n",
6888 (unsigned long long)total_found
, pass
);
6890 if (total_found
== skipped
&& pass
> 2) {
6892 reloc_inode
= create_reloc_inode(info
, block_group
);
6898 /* delete reloc_inode */
6901 /* unpin extents in this range */
6902 trans
= btrfs_start_transaction(info
->tree_root
, 1);
6903 btrfs_commit_transaction(trans
, info
->tree_root
);
6905 spin_lock(&block_group
->lock
);
6906 WARN_ON(block_group
->pinned
> 0);
6907 WARN_ON(block_group
->reserved
> 0);
6908 WARN_ON(btrfs_block_group_used(&block_group
->item
) > 0);
6909 spin_unlock(&block_group
->lock
);
6910 btrfs_put_block_group(block_group
);
6913 btrfs_free_path(path
);
6918 static int find_first_block_group(struct btrfs_root
*root
,
6919 struct btrfs_path
*path
, struct btrfs_key
*key
)
6922 struct btrfs_key found_key
;
6923 struct extent_buffer
*leaf
;
6926 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
6931 slot
= path
->slots
[0];
6932 leaf
= path
->nodes
[0];
6933 if (slot
>= btrfs_header_nritems(leaf
)) {
6934 ret
= btrfs_next_leaf(root
, path
);
6941 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
6943 if (found_key
.objectid
>= key
->objectid
&&
6944 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
6955 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
6957 struct btrfs_block_group_cache
*block_group
;
6958 struct btrfs_space_info
*space_info
;
6961 spin_lock(&info
->block_group_cache_lock
);
6962 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
6963 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
6965 rb_erase(&block_group
->cache_node
,
6966 &info
->block_group_cache_tree
);
6967 spin_unlock(&info
->block_group_cache_lock
);
6969 btrfs_remove_free_space_cache(block_group
);
6970 down_write(&block_group
->space_info
->groups_sem
);
6971 list_del(&block_group
->list
);
6972 up_write(&block_group
->space_info
->groups_sem
);
6974 WARN_ON(atomic_read(&block_group
->count
) != 1);
6977 spin_lock(&info
->block_group_cache_lock
);
6979 spin_unlock(&info
->block_group_cache_lock
);
6981 /* now that all the block groups are freed, go through and
6982 * free all the space_info structs. This is only called during
6983 * the final stages of unmount, and so we know nobody is
6984 * using them. We call synchronize_rcu() once before we start,
6985 * just to be on the safe side.
6989 while(!list_empty(&info
->space_info
)) {
6990 space_info
= list_entry(info
->space_info
.next
,
6991 struct btrfs_space_info
,
6994 list_del(&space_info
->list
);
7000 int btrfs_read_block_groups(struct btrfs_root
*root
)
7002 struct btrfs_path
*path
;
7004 struct btrfs_block_group_cache
*cache
;
7005 struct btrfs_fs_info
*info
= root
->fs_info
;
7006 struct btrfs_space_info
*space_info
;
7007 struct btrfs_key key
;
7008 struct btrfs_key found_key
;
7009 struct extent_buffer
*leaf
;
7011 root
= info
->extent_root
;
7014 btrfs_set_key_type(&key
, BTRFS_BLOCK_GROUP_ITEM_KEY
);
7015 path
= btrfs_alloc_path();
7020 ret
= find_first_block_group(root
, path
, &key
);
7028 leaf
= path
->nodes
[0];
7029 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
7030 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
7036 atomic_set(&cache
->count
, 1);
7037 spin_lock_init(&cache
->lock
);
7038 spin_lock_init(&cache
->tree_lock
);
7039 mutex_init(&cache
->cache_mutex
);
7040 INIT_LIST_HEAD(&cache
->list
);
7041 INIT_LIST_HEAD(&cache
->cluster_list
);
7042 cache
->sectorsize
= root
->sectorsize
;
7045 * we only want to have 32k of ram per block group for keeping
7046 * track of free space, and if we pass 1/2 of that we want to
7047 * start converting things over to using bitmaps
7049 cache
->extents_thresh
= ((1024 * 32) / 2) /
7050 sizeof(struct btrfs_free_space
);
7052 read_extent_buffer(leaf
, &cache
->item
,
7053 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
7054 sizeof(cache
->item
));
7055 memcpy(&cache
->key
, &found_key
, sizeof(found_key
));
7057 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
7058 btrfs_release_path(root
, path
);
7059 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
7061 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
7062 btrfs_block_group_used(&cache
->item
),
7065 cache
->space_info
= space_info
;
7066 down_write(&space_info
->groups_sem
);
7067 list_add_tail(&cache
->list
, &space_info
->block_groups
);
7068 up_write(&space_info
->groups_sem
);
7070 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
7073 set_avail_alloc_bits(root
->fs_info
, cache
->flags
);
7074 if (btrfs_chunk_readonly(root
, cache
->key
.objectid
))
7075 set_block_group_readonly(cache
);
7079 btrfs_free_path(path
);
7083 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
7084 struct btrfs_root
*root
, u64 bytes_used
,
7085 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
7089 struct btrfs_root
*extent_root
;
7090 struct btrfs_block_group_cache
*cache
;
7092 extent_root
= root
->fs_info
->extent_root
;
7094 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
7096 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
7100 cache
->key
.objectid
= chunk_offset
;
7101 cache
->key
.offset
= size
;
7102 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
7103 cache
->sectorsize
= root
->sectorsize
;
7106 * we only want to have 32k of ram per block group for keeping track
7107 * of free space, and if we pass 1/2 of that we want to start
7108 * converting things over to using bitmaps
7110 cache
->extents_thresh
= ((1024 * 32) / 2) /
7111 sizeof(struct btrfs_free_space
);
7112 atomic_set(&cache
->count
, 1);
7113 spin_lock_init(&cache
->lock
);
7114 spin_lock_init(&cache
->tree_lock
);
7115 mutex_init(&cache
->cache_mutex
);
7116 INIT_LIST_HEAD(&cache
->list
);
7117 INIT_LIST_HEAD(&cache
->cluster_list
);
7119 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
7120 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
7121 cache
->flags
= type
;
7122 btrfs_set_block_group_flags(&cache
->item
, type
);
7125 ret
= btrfs_add_free_space(cache
, chunk_offset
, size
);
7127 remove_sb_from_cache(root
, cache
);
7129 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
7130 &cache
->space_info
);
7132 down_write(&cache
->space_info
->groups_sem
);
7133 list_add_tail(&cache
->list
, &cache
->space_info
->block_groups
);
7134 up_write(&cache
->space_info
->groups_sem
);
7136 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
7139 ret
= btrfs_insert_item(trans
, extent_root
, &cache
->key
, &cache
->item
,
7140 sizeof(cache
->item
));
7143 set_avail_alloc_bits(extent_root
->fs_info
, type
);
7148 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
7149 struct btrfs_root
*root
, u64 group_start
)
7151 struct btrfs_path
*path
;
7152 struct btrfs_block_group_cache
*block_group
;
7153 struct btrfs_free_cluster
*cluster
;
7154 struct btrfs_key key
;
7157 root
= root
->fs_info
->extent_root
;
7159 block_group
= btrfs_lookup_block_group(root
->fs_info
, group_start
);
7160 BUG_ON(!block_group
);
7161 BUG_ON(!block_group
->ro
);
7163 memcpy(&key
, &block_group
->key
, sizeof(key
));
7165 /* make sure this block group isn't part of an allocation cluster */
7166 cluster
= &root
->fs_info
->data_alloc_cluster
;
7167 spin_lock(&cluster
->refill_lock
);
7168 btrfs_return_cluster_to_free_space(block_group
, cluster
);
7169 spin_unlock(&cluster
->refill_lock
);
7172 * make sure this block group isn't part of a metadata
7173 * allocation cluster
7175 cluster
= &root
->fs_info
->meta_alloc_cluster
;
7176 spin_lock(&cluster
->refill_lock
);
7177 btrfs_return_cluster_to_free_space(block_group
, cluster
);
7178 spin_unlock(&cluster
->refill_lock
);
7180 path
= btrfs_alloc_path();
7183 spin_lock(&root
->fs_info
->block_group_cache_lock
);
7184 rb_erase(&block_group
->cache_node
,
7185 &root
->fs_info
->block_group_cache_tree
);
7186 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
7187 btrfs_remove_free_space_cache(block_group
);
7188 down_write(&block_group
->space_info
->groups_sem
);
7190 * we must use list_del_init so people can check to see if they
7191 * are still on the list after taking the semaphore
7193 list_del_init(&block_group
->list
);
7194 up_write(&block_group
->space_info
->groups_sem
);
7196 spin_lock(&block_group
->space_info
->lock
);
7197 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
7198 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
7199 spin_unlock(&block_group
->space_info
->lock
);
7200 block_group
->space_info
->full
= 0;
7202 btrfs_put_block_group(block_group
);
7203 btrfs_put_block_group(block_group
);
7205 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
7211 ret
= btrfs_del_item(trans
, root
, path
);
7213 btrfs_free_path(path
);