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>
24 #include <linux/kthread.h>
29 #include "print-tree.h"
30 #include "transaction.h"
33 #include "free-space-cache.h"
35 static int update_reserved_extents(struct btrfs_root
*root
,
36 u64 bytenr
, u64 num
, int reserve
);
37 static int update_block_group(struct btrfs_trans_handle
*trans
,
38 struct btrfs_root
*root
,
39 u64 bytenr
, u64 num_bytes
, int alloc
,
41 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
42 struct btrfs_root
*root
,
43 u64 bytenr
, u64 num_bytes
, u64 parent
,
44 u64 root_objectid
, u64 owner_objectid
,
45 u64 owner_offset
, int refs_to_drop
,
46 struct btrfs_delayed_extent_op
*extra_op
);
47 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
48 struct extent_buffer
*leaf
,
49 struct btrfs_extent_item
*ei
);
50 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
51 struct btrfs_root
*root
,
52 u64 parent
, u64 root_objectid
,
53 u64 flags
, u64 owner
, u64 offset
,
54 struct btrfs_key
*ins
, int ref_mod
);
55 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
56 struct btrfs_root
*root
,
57 u64 parent
, u64 root_objectid
,
58 u64 flags
, struct btrfs_disk_key
*key
,
59 int level
, struct btrfs_key
*ins
);
61 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
62 struct btrfs_root
*extent_root
, u64 alloc_bytes
,
63 u64 flags
, int force
);
66 block_group_cache_done(struct btrfs_block_group_cache
*cache
)
69 return cache
->cached
== BTRFS_CACHE_FINISHED
;
72 static int block_group_bits(struct btrfs_block_group_cache
*cache
, u64 bits
)
74 return (cache
->flags
& bits
) == bits
;
78 * this adds the block group to the fs_info rb tree for the block group
81 static int btrfs_add_block_group_cache(struct btrfs_fs_info
*info
,
82 struct btrfs_block_group_cache
*block_group
)
85 struct rb_node
*parent
= NULL
;
86 struct btrfs_block_group_cache
*cache
;
88 spin_lock(&info
->block_group_cache_lock
);
89 p
= &info
->block_group_cache_tree
.rb_node
;
93 cache
= rb_entry(parent
, struct btrfs_block_group_cache
,
95 if (block_group
->key
.objectid
< cache
->key
.objectid
) {
97 } else if (block_group
->key
.objectid
> cache
->key
.objectid
) {
100 spin_unlock(&info
->block_group_cache_lock
);
105 rb_link_node(&block_group
->cache_node
, parent
, p
);
106 rb_insert_color(&block_group
->cache_node
,
107 &info
->block_group_cache_tree
);
108 spin_unlock(&info
->block_group_cache_lock
);
114 * This will return the block group at or after bytenr if contains is 0, else
115 * it will return the block group that contains the bytenr
117 static struct btrfs_block_group_cache
*
118 block_group_cache_tree_search(struct btrfs_fs_info
*info
, u64 bytenr
,
121 struct btrfs_block_group_cache
*cache
, *ret
= NULL
;
125 spin_lock(&info
->block_group_cache_lock
);
126 n
= info
->block_group_cache_tree
.rb_node
;
129 cache
= rb_entry(n
, struct btrfs_block_group_cache
,
131 end
= cache
->key
.objectid
+ cache
->key
.offset
- 1;
132 start
= cache
->key
.objectid
;
134 if (bytenr
< start
) {
135 if (!contains
&& (!ret
|| start
< ret
->key
.objectid
))
138 } else if (bytenr
> start
) {
139 if (contains
&& bytenr
<= end
) {
150 atomic_inc(&ret
->count
);
151 spin_unlock(&info
->block_group_cache_lock
);
157 * We always set EXTENT_LOCKED for the super mirror extents so we don't
158 * overwrite them, so those bits need to be unset. Also, if we are unmounting
159 * with pinned extents still sitting there because we had a block group caching,
160 * we need to clear those now, since we are done.
162 void btrfs_free_pinned_extents(struct btrfs_fs_info
*info
)
164 u64 start
, end
, last
= 0;
168 ret
= find_first_extent_bit(&info
->pinned_extents
, last
,
170 EXTENT_LOCKED
|EXTENT_DIRTY
);
174 clear_extent_bits(&info
->pinned_extents
, start
, end
,
175 EXTENT_LOCKED
|EXTENT_DIRTY
, GFP_NOFS
);
180 static int remove_sb_from_cache(struct btrfs_root
*root
,
181 struct btrfs_block_group_cache
*cache
)
183 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
189 for (i
= 0; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
190 bytenr
= btrfs_sb_offset(i
);
191 ret
= btrfs_rmap_block(&root
->fs_info
->mapping_tree
,
192 cache
->key
.objectid
, bytenr
,
193 0, &logical
, &nr
, &stripe_len
);
196 try_lock_extent(&fs_info
->pinned_extents
,
198 logical
[nr
] + stripe_len
- 1, GFP_NOFS
);
207 * this is only called by cache_block_group, since we could have freed extents
208 * we need to check the pinned_extents for any extents that can't be used yet
209 * since their free space will be released as soon as the transaction commits.
211 static u64
add_new_free_space(struct btrfs_block_group_cache
*block_group
,
212 struct btrfs_fs_info
*info
, u64 start
, u64 end
)
214 u64 extent_start
, extent_end
, size
, total_added
= 0;
217 while (start
< end
) {
218 ret
= find_first_extent_bit(&info
->pinned_extents
, start
,
219 &extent_start
, &extent_end
,
220 EXTENT_DIRTY
|EXTENT_LOCKED
);
224 if (extent_start
== start
) {
225 start
= extent_end
+ 1;
226 } else if (extent_start
> start
&& extent_start
< end
) {
227 size
= extent_start
- start
;
229 ret
= btrfs_add_free_space(block_group
, start
,
232 start
= extent_end
+ 1;
241 ret
= btrfs_add_free_space(block_group
, start
, size
);
248 static int caching_kthread(void *data
)
250 struct btrfs_block_group_cache
*block_group
= data
;
251 struct btrfs_fs_info
*fs_info
= block_group
->fs_info
;
253 struct btrfs_path
*path
;
255 struct btrfs_key key
;
256 struct extent_buffer
*leaf
;
262 path
= btrfs_alloc_path();
266 atomic_inc(&block_group
->space_info
->caching_threads
);
267 last
= max_t(u64
, block_group
->key
.objectid
, BTRFS_SUPER_INFO_OFFSET
);
269 /* need to make sure the commit_root doesn't disappear */
270 down_read(&fs_info
->extent_root
->commit_root_sem
);
273 * We don't want to deadlock with somebody trying to allocate a new
274 * extent for the extent root while also trying to search the extent
275 * root to add free space. So we skip locking and search the commit
276 * root, since its read-only
278 path
->skip_locking
= 1;
279 path
->search_commit_root
= 1;
284 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
285 ret
= btrfs_search_slot(NULL
, fs_info
->extent_root
, &key
, path
, 0, 0);
291 if (block_group
->fs_info
->closing
> 1) {
296 leaf
= path
->nodes
[0];
297 slot
= path
->slots
[0];
298 if (slot
>= btrfs_header_nritems(leaf
)) {
299 ret
= btrfs_next_leaf(fs_info
->extent_root
, path
);
305 if (need_resched()) {
306 btrfs_release_path(fs_info
->extent_root
, path
);
307 up_read(&fs_info
->extent_root
->commit_root_sem
);
314 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
315 if (key
.objectid
< block_group
->key
.objectid
)
318 if (key
.objectid
>= block_group
->key
.objectid
+
319 block_group
->key
.offset
)
322 if (btrfs_key_type(&key
) == BTRFS_EXTENT_ITEM_KEY
) {
323 total_found
+= add_new_free_space(block_group
,
326 last
= key
.objectid
+ key
.offset
;
329 if (total_found
> (1024 * 1024 * 2)) {
331 wake_up(&block_group
->caching_q
);
338 total_found
+= add_new_free_space(block_group
, fs_info
, last
,
339 block_group
->key
.objectid
+
340 block_group
->key
.offset
);
342 spin_lock(&block_group
->lock
);
343 block_group
->cached
= BTRFS_CACHE_FINISHED
;
344 spin_unlock(&block_group
->lock
);
347 btrfs_free_path(path
);
348 up_read(&fs_info
->extent_root
->commit_root_sem
);
349 atomic_dec(&block_group
->space_info
->caching_threads
);
350 wake_up(&block_group
->caching_q
);
355 static int cache_block_group(struct btrfs_block_group_cache
*cache
)
357 struct task_struct
*tsk
;
360 spin_lock(&cache
->lock
);
361 if (cache
->cached
!= BTRFS_CACHE_NO
) {
362 spin_unlock(&cache
->lock
);
365 cache
->cached
= BTRFS_CACHE_STARTED
;
366 spin_unlock(&cache
->lock
);
368 tsk
= kthread_run(caching_kthread
, cache
, "btrfs-cache-%llu\n",
369 cache
->key
.objectid
);
372 printk(KERN_ERR
"error running thread %d\n", ret
);
380 * return the block group that starts at or after bytenr
382 static struct btrfs_block_group_cache
*
383 btrfs_lookup_first_block_group(struct btrfs_fs_info
*info
, u64 bytenr
)
385 struct btrfs_block_group_cache
*cache
;
387 cache
= block_group_cache_tree_search(info
, bytenr
, 0);
393 * return the block group that contains the given bytenr
395 struct btrfs_block_group_cache
*btrfs_lookup_block_group(
396 struct btrfs_fs_info
*info
,
399 struct btrfs_block_group_cache
*cache
;
401 cache
= block_group_cache_tree_search(info
, bytenr
, 1);
406 void btrfs_put_block_group(struct btrfs_block_group_cache
*cache
)
408 if (atomic_dec_and_test(&cache
->count
))
412 static struct btrfs_space_info
*__find_space_info(struct btrfs_fs_info
*info
,
415 struct list_head
*head
= &info
->space_info
;
416 struct btrfs_space_info
*found
;
419 list_for_each_entry_rcu(found
, head
, list
) {
420 if (found
->flags
== flags
) {
430 * after adding space to the filesystem, we need to clear the full flags
431 * on all the space infos.
433 void btrfs_clear_space_info_full(struct btrfs_fs_info
*info
)
435 struct list_head
*head
= &info
->space_info
;
436 struct btrfs_space_info
*found
;
439 list_for_each_entry_rcu(found
, head
, list
)
444 static u64
div_factor(u64 num
, int factor
)
453 u64
btrfs_find_block_group(struct btrfs_root
*root
,
454 u64 search_start
, u64 search_hint
, int owner
)
456 struct btrfs_block_group_cache
*cache
;
458 u64 last
= max(search_hint
, search_start
);
465 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
469 spin_lock(&cache
->lock
);
470 last
= cache
->key
.objectid
+ cache
->key
.offset
;
471 used
= btrfs_block_group_used(&cache
->item
);
473 if ((full_search
|| !cache
->ro
) &&
474 block_group_bits(cache
, BTRFS_BLOCK_GROUP_METADATA
)) {
475 if (used
+ cache
->pinned
+ cache
->reserved
<
476 div_factor(cache
->key
.offset
, factor
)) {
477 group_start
= cache
->key
.objectid
;
478 spin_unlock(&cache
->lock
);
479 btrfs_put_block_group(cache
);
483 spin_unlock(&cache
->lock
);
484 btrfs_put_block_group(cache
);
492 if (!full_search
&& factor
< 10) {
502 /* simple helper to search for an existing extent at a given offset */
503 int btrfs_lookup_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
506 struct btrfs_key key
;
507 struct btrfs_path
*path
;
509 path
= btrfs_alloc_path();
511 key
.objectid
= start
;
513 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
514 ret
= btrfs_search_slot(NULL
, root
->fs_info
->extent_root
, &key
, path
,
516 btrfs_free_path(path
);
521 * Back reference rules. Back refs have three main goals:
523 * 1) differentiate between all holders of references to an extent so that
524 * when a reference is dropped we can make sure it was a valid reference
525 * before freeing the extent.
527 * 2) Provide enough information to quickly find the holders of an extent
528 * if we notice a given block is corrupted or bad.
530 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
531 * maintenance. This is actually the same as #2, but with a slightly
532 * different use case.
534 * There are two kinds of back refs. The implicit back refs is optimized
535 * for pointers in non-shared tree blocks. For a given pointer in a block,
536 * back refs of this kind provide information about the block's owner tree
537 * and the pointer's key. These information allow us to find the block by
538 * b-tree searching. The full back refs is for pointers in tree blocks not
539 * referenced by their owner trees. The location of tree block is recorded
540 * in the back refs. Actually the full back refs is generic, and can be
541 * used in all cases the implicit back refs is used. The major shortcoming
542 * of the full back refs is its overhead. Every time a tree block gets
543 * COWed, we have to update back refs entry for all pointers in it.
545 * For a newly allocated tree block, we use implicit back refs for
546 * pointers in it. This means most tree related operations only involve
547 * implicit back refs. For a tree block created in old transaction, the
548 * only way to drop a reference to it is COW it. So we can detect the
549 * event that tree block loses its owner tree's reference and do the
550 * back refs conversion.
552 * When a tree block is COW'd through a tree, there are four cases:
554 * The reference count of the block is one and the tree is the block's
555 * owner tree. Nothing to do in this case.
557 * The reference count of the block is one and the tree is not the
558 * block's owner tree. In this case, full back refs is used for pointers
559 * in the block. Remove these full back refs, add implicit back refs for
560 * every pointers in the new block.
562 * The reference count of the block is greater than one and the tree is
563 * the block's owner tree. In this case, implicit back refs is used for
564 * pointers in the block. Add full back refs for every pointers in the
565 * block, increase lower level extents' reference counts. The original
566 * implicit back refs are entailed to the new block.
568 * The reference count of the block is greater than one and the tree is
569 * not the block's owner tree. Add implicit back refs for every pointer in
570 * the new block, increase lower level extents' reference count.
572 * Back Reference Key composing:
574 * The key objectid corresponds to the first byte in the extent,
575 * The key type is used to differentiate between types of back refs.
576 * There are different meanings of the key offset for different types
579 * File extents can be referenced by:
581 * - multiple snapshots, subvolumes, or different generations in one subvol
582 * - different files inside a single subvolume
583 * - different offsets inside a file (bookend extents in file.c)
585 * The extent ref structure for the implicit back refs has fields for:
587 * - Objectid of the subvolume root
588 * - objectid of the file holding the reference
589 * - original offset in the file
590 * - how many bookend extents
592 * The key offset for the implicit back refs is hash of the first
595 * The extent ref structure for the full back refs has field for:
597 * - number of pointers in the tree leaf
599 * The key offset for the implicit back refs is the first byte of
602 * When a file extent is allocated, The implicit back refs is used.
603 * the fields are filled in:
605 * (root_key.objectid, inode objectid, offset in file, 1)
607 * When a file extent is removed file truncation, we find the
608 * corresponding implicit back refs and check the following fields:
610 * (btrfs_header_owner(leaf), inode objectid, offset in file)
612 * Btree extents can be referenced by:
614 * - Different subvolumes
616 * Both the implicit back refs and the full back refs for tree blocks
617 * only consist of key. The key offset for the implicit back refs is
618 * objectid of block's owner tree. The key offset for the full back refs
619 * is the first byte of parent block.
621 * When implicit back refs is used, information about the lowest key and
622 * level of the tree block are required. These information are stored in
623 * tree block info structure.
626 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
627 static int convert_extent_item_v0(struct btrfs_trans_handle
*trans
,
628 struct btrfs_root
*root
,
629 struct btrfs_path
*path
,
630 u64 owner
, u32 extra_size
)
632 struct btrfs_extent_item
*item
;
633 struct btrfs_extent_item_v0
*ei0
;
634 struct btrfs_extent_ref_v0
*ref0
;
635 struct btrfs_tree_block_info
*bi
;
636 struct extent_buffer
*leaf
;
637 struct btrfs_key key
;
638 struct btrfs_key found_key
;
639 u32 new_size
= sizeof(*item
);
643 leaf
= path
->nodes
[0];
644 BUG_ON(btrfs_item_size_nr(leaf
, path
->slots
[0]) != sizeof(*ei0
));
646 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
647 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
648 struct btrfs_extent_item_v0
);
649 refs
= btrfs_extent_refs_v0(leaf
, ei0
);
651 if (owner
== (u64
)-1) {
653 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
654 ret
= btrfs_next_leaf(root
, path
);
658 leaf
= path
->nodes
[0];
660 btrfs_item_key_to_cpu(leaf
, &found_key
,
662 BUG_ON(key
.objectid
!= found_key
.objectid
);
663 if (found_key
.type
!= BTRFS_EXTENT_REF_V0_KEY
) {
667 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
668 struct btrfs_extent_ref_v0
);
669 owner
= btrfs_ref_objectid_v0(leaf
, ref0
);
673 btrfs_release_path(root
, path
);
675 if (owner
< BTRFS_FIRST_FREE_OBJECTID
)
676 new_size
+= sizeof(*bi
);
678 new_size
-= sizeof(*ei0
);
679 ret
= btrfs_search_slot(trans
, root
, &key
, path
,
680 new_size
+ extra_size
, 1);
685 ret
= btrfs_extend_item(trans
, root
, path
, new_size
);
688 leaf
= path
->nodes
[0];
689 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
690 btrfs_set_extent_refs(leaf
, item
, refs
);
691 /* FIXME: get real generation */
692 btrfs_set_extent_generation(leaf
, item
, 0);
693 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
694 btrfs_set_extent_flags(leaf
, item
,
695 BTRFS_EXTENT_FLAG_TREE_BLOCK
|
696 BTRFS_BLOCK_FLAG_FULL_BACKREF
);
697 bi
= (struct btrfs_tree_block_info
*)(item
+ 1);
698 /* FIXME: get first key of the block */
699 memset_extent_buffer(leaf
, 0, (unsigned long)bi
, sizeof(*bi
));
700 btrfs_set_tree_block_level(leaf
, bi
, (int)owner
);
702 btrfs_set_extent_flags(leaf
, item
, BTRFS_EXTENT_FLAG_DATA
);
704 btrfs_mark_buffer_dirty(leaf
);
709 static u64
hash_extent_data_ref(u64 root_objectid
, u64 owner
, u64 offset
)
711 u32 high_crc
= ~(u32
)0;
712 u32 low_crc
= ~(u32
)0;
715 lenum
= cpu_to_le64(root_objectid
);
716 high_crc
= crc32c(high_crc
, &lenum
, sizeof(lenum
));
717 lenum
= cpu_to_le64(owner
);
718 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
719 lenum
= cpu_to_le64(offset
);
720 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
722 return ((u64
)high_crc
<< 31) ^ (u64
)low_crc
;
725 static u64
hash_extent_data_ref_item(struct extent_buffer
*leaf
,
726 struct btrfs_extent_data_ref
*ref
)
728 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf
, ref
),
729 btrfs_extent_data_ref_objectid(leaf
, ref
),
730 btrfs_extent_data_ref_offset(leaf
, ref
));
733 static int match_extent_data_ref(struct extent_buffer
*leaf
,
734 struct btrfs_extent_data_ref
*ref
,
735 u64 root_objectid
, u64 owner
, u64 offset
)
737 if (btrfs_extent_data_ref_root(leaf
, ref
) != root_objectid
||
738 btrfs_extent_data_ref_objectid(leaf
, ref
) != owner
||
739 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
744 static noinline
int lookup_extent_data_ref(struct btrfs_trans_handle
*trans
,
745 struct btrfs_root
*root
,
746 struct btrfs_path
*path
,
747 u64 bytenr
, u64 parent
,
749 u64 owner
, u64 offset
)
751 struct btrfs_key key
;
752 struct btrfs_extent_data_ref
*ref
;
753 struct extent_buffer
*leaf
;
759 key
.objectid
= bytenr
;
761 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
764 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
765 key
.offset
= hash_extent_data_ref(root_objectid
,
770 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
779 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
780 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
781 btrfs_release_path(root
, path
);
782 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
793 leaf
= path
->nodes
[0];
794 nritems
= btrfs_header_nritems(leaf
);
796 if (path
->slots
[0] >= nritems
) {
797 ret
= btrfs_next_leaf(root
, path
);
803 leaf
= path
->nodes
[0];
804 nritems
= btrfs_header_nritems(leaf
);
808 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
809 if (key
.objectid
!= bytenr
||
810 key
.type
!= BTRFS_EXTENT_DATA_REF_KEY
)
813 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
814 struct btrfs_extent_data_ref
);
816 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
819 btrfs_release_path(root
, path
);
831 static noinline
int insert_extent_data_ref(struct btrfs_trans_handle
*trans
,
832 struct btrfs_root
*root
,
833 struct btrfs_path
*path
,
834 u64 bytenr
, u64 parent
,
835 u64 root_objectid
, u64 owner
,
836 u64 offset
, int refs_to_add
)
838 struct btrfs_key key
;
839 struct extent_buffer
*leaf
;
844 key
.objectid
= bytenr
;
846 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
848 size
= sizeof(struct btrfs_shared_data_ref
);
850 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
851 key
.offset
= hash_extent_data_ref(root_objectid
,
853 size
= sizeof(struct btrfs_extent_data_ref
);
856 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, size
);
857 if (ret
&& ret
!= -EEXIST
)
860 leaf
= path
->nodes
[0];
862 struct btrfs_shared_data_ref
*ref
;
863 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
864 struct btrfs_shared_data_ref
);
866 btrfs_set_shared_data_ref_count(leaf
, ref
, refs_to_add
);
868 num_refs
= btrfs_shared_data_ref_count(leaf
, ref
);
869 num_refs
+= refs_to_add
;
870 btrfs_set_shared_data_ref_count(leaf
, ref
, num_refs
);
873 struct btrfs_extent_data_ref
*ref
;
874 while (ret
== -EEXIST
) {
875 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
876 struct btrfs_extent_data_ref
);
877 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
880 btrfs_release_path(root
, path
);
882 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
884 if (ret
&& ret
!= -EEXIST
)
887 leaf
= path
->nodes
[0];
889 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
890 struct btrfs_extent_data_ref
);
892 btrfs_set_extent_data_ref_root(leaf
, ref
,
894 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
895 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
896 btrfs_set_extent_data_ref_count(leaf
, ref
, refs_to_add
);
898 num_refs
= btrfs_extent_data_ref_count(leaf
, ref
);
899 num_refs
+= refs_to_add
;
900 btrfs_set_extent_data_ref_count(leaf
, ref
, num_refs
);
903 btrfs_mark_buffer_dirty(leaf
);
906 btrfs_release_path(root
, path
);
910 static noinline
int remove_extent_data_ref(struct btrfs_trans_handle
*trans
,
911 struct btrfs_root
*root
,
912 struct btrfs_path
*path
,
915 struct btrfs_key key
;
916 struct btrfs_extent_data_ref
*ref1
= NULL
;
917 struct btrfs_shared_data_ref
*ref2
= NULL
;
918 struct extent_buffer
*leaf
;
922 leaf
= path
->nodes
[0];
923 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
925 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
926 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
927 struct btrfs_extent_data_ref
);
928 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
929 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
930 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
931 struct btrfs_shared_data_ref
);
932 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
933 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
934 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
935 struct btrfs_extent_ref_v0
*ref0
;
936 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
937 struct btrfs_extent_ref_v0
);
938 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
944 BUG_ON(num_refs
< refs_to_drop
);
945 num_refs
-= refs_to_drop
;
948 ret
= btrfs_del_item(trans
, root
, path
);
950 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
)
951 btrfs_set_extent_data_ref_count(leaf
, ref1
, num_refs
);
952 else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
)
953 btrfs_set_shared_data_ref_count(leaf
, ref2
, num_refs
);
954 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
956 struct btrfs_extent_ref_v0
*ref0
;
957 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
958 struct btrfs_extent_ref_v0
);
959 btrfs_set_ref_count_v0(leaf
, ref0
, num_refs
);
962 btrfs_mark_buffer_dirty(leaf
);
967 static noinline u32
extent_data_ref_count(struct btrfs_root
*root
,
968 struct btrfs_path
*path
,
969 struct btrfs_extent_inline_ref
*iref
)
971 struct btrfs_key key
;
972 struct extent_buffer
*leaf
;
973 struct btrfs_extent_data_ref
*ref1
;
974 struct btrfs_shared_data_ref
*ref2
;
977 leaf
= path
->nodes
[0];
978 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
980 if (btrfs_extent_inline_ref_type(leaf
, iref
) ==
981 BTRFS_EXTENT_DATA_REF_KEY
) {
982 ref1
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
983 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
985 ref2
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
986 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
988 } else if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
989 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
990 struct btrfs_extent_data_ref
);
991 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
992 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
993 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
994 struct btrfs_shared_data_ref
);
995 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
996 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
997 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
998 struct btrfs_extent_ref_v0
*ref0
;
999 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1000 struct btrfs_extent_ref_v0
);
1001 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1009 static noinline
int lookup_tree_block_ref(struct btrfs_trans_handle
*trans
,
1010 struct btrfs_root
*root
,
1011 struct btrfs_path
*path
,
1012 u64 bytenr
, u64 parent
,
1015 struct btrfs_key key
;
1018 key
.objectid
= bytenr
;
1020 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1021 key
.offset
= parent
;
1023 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1024 key
.offset
= root_objectid
;
1027 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1030 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1031 if (ret
== -ENOENT
&& parent
) {
1032 btrfs_release_path(root
, path
);
1033 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1034 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1042 static noinline
int insert_tree_block_ref(struct btrfs_trans_handle
*trans
,
1043 struct btrfs_root
*root
,
1044 struct btrfs_path
*path
,
1045 u64 bytenr
, u64 parent
,
1048 struct btrfs_key key
;
1051 key
.objectid
= bytenr
;
1053 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1054 key
.offset
= parent
;
1056 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1057 key
.offset
= root_objectid
;
1060 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
1061 btrfs_release_path(root
, path
);
1065 static inline int extent_ref_type(u64 parent
, u64 owner
)
1068 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1070 type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1072 type
= BTRFS_TREE_BLOCK_REF_KEY
;
1075 type
= BTRFS_SHARED_DATA_REF_KEY
;
1077 type
= BTRFS_EXTENT_DATA_REF_KEY
;
1082 static int find_next_key(struct btrfs_path
*path
, int level
,
1083 struct btrfs_key
*key
)
1086 for (; level
< BTRFS_MAX_LEVEL
; level
++) {
1087 if (!path
->nodes
[level
])
1089 if (path
->slots
[level
] + 1 >=
1090 btrfs_header_nritems(path
->nodes
[level
]))
1093 btrfs_item_key_to_cpu(path
->nodes
[level
], key
,
1094 path
->slots
[level
] + 1);
1096 btrfs_node_key_to_cpu(path
->nodes
[level
], key
,
1097 path
->slots
[level
] + 1);
1104 * look for inline back ref. if back ref is found, *ref_ret is set
1105 * to the address of inline back ref, and 0 is returned.
1107 * if back ref isn't found, *ref_ret is set to the address where it
1108 * should be inserted, and -ENOENT is returned.
1110 * if insert is true and there are too many inline back refs, the path
1111 * points to the extent item, and -EAGAIN is returned.
1113 * NOTE: inline back refs are ordered in the same way that back ref
1114 * items in the tree are ordered.
1116 static noinline_for_stack
1117 int lookup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1118 struct btrfs_root
*root
,
1119 struct btrfs_path
*path
,
1120 struct btrfs_extent_inline_ref
**ref_ret
,
1121 u64 bytenr
, u64 num_bytes
,
1122 u64 parent
, u64 root_objectid
,
1123 u64 owner
, u64 offset
, int insert
)
1125 struct btrfs_key key
;
1126 struct extent_buffer
*leaf
;
1127 struct btrfs_extent_item
*ei
;
1128 struct btrfs_extent_inline_ref
*iref
;
1139 key
.objectid
= bytenr
;
1140 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1141 key
.offset
= num_bytes
;
1143 want
= extent_ref_type(parent
, owner
);
1145 extra_size
= btrfs_extent_inline_ref_size(want
);
1146 path
->keep_locks
= 1;
1149 ret
= btrfs_search_slot(trans
, root
, &key
, path
, extra_size
, 1);
1156 leaf
= path
->nodes
[0];
1157 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1158 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1159 if (item_size
< sizeof(*ei
)) {
1164 ret
= convert_extent_item_v0(trans
, root
, path
, owner
,
1170 leaf
= path
->nodes
[0];
1171 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1174 BUG_ON(item_size
< sizeof(*ei
));
1176 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1177 flags
= btrfs_extent_flags(leaf
, ei
);
1179 ptr
= (unsigned long)(ei
+ 1);
1180 end
= (unsigned long)ei
+ item_size
;
1182 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
) {
1183 ptr
+= sizeof(struct btrfs_tree_block_info
);
1186 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_DATA
));
1195 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1196 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1200 ptr
+= btrfs_extent_inline_ref_size(type
);
1204 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1205 struct btrfs_extent_data_ref
*dref
;
1206 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1207 if (match_extent_data_ref(leaf
, dref
, root_objectid
,
1212 if (hash_extent_data_ref_item(leaf
, dref
) <
1213 hash_extent_data_ref(root_objectid
, owner
, offset
))
1217 ref_offset
= btrfs_extent_inline_ref_offset(leaf
, iref
);
1219 if (parent
== ref_offset
) {
1223 if (ref_offset
< parent
)
1226 if (root_objectid
== ref_offset
) {
1230 if (ref_offset
< root_objectid
)
1234 ptr
+= btrfs_extent_inline_ref_size(type
);
1236 if (err
== -ENOENT
&& insert
) {
1237 if (item_size
+ extra_size
>=
1238 BTRFS_MAX_EXTENT_ITEM_SIZE(root
)) {
1243 * To add new inline back ref, we have to make sure
1244 * there is no corresponding back ref item.
1245 * For simplicity, we just do not add new inline back
1246 * ref if there is any kind of item for this block
1248 if (find_next_key(path
, 0, &key
) == 0 &&
1249 key
.objectid
== bytenr
&&
1250 key
.type
< BTRFS_BLOCK_GROUP_ITEM_KEY
) {
1255 *ref_ret
= (struct btrfs_extent_inline_ref
*)ptr
;
1258 path
->keep_locks
= 0;
1259 btrfs_unlock_up_safe(path
, 1);
1265 * helper to add new inline back ref
1267 static noinline_for_stack
1268 int setup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1269 struct btrfs_root
*root
,
1270 struct btrfs_path
*path
,
1271 struct btrfs_extent_inline_ref
*iref
,
1272 u64 parent
, u64 root_objectid
,
1273 u64 owner
, u64 offset
, int refs_to_add
,
1274 struct btrfs_delayed_extent_op
*extent_op
)
1276 struct extent_buffer
*leaf
;
1277 struct btrfs_extent_item
*ei
;
1280 unsigned long item_offset
;
1286 leaf
= path
->nodes
[0];
1287 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1288 item_offset
= (unsigned long)iref
- (unsigned long)ei
;
1290 type
= extent_ref_type(parent
, owner
);
1291 size
= btrfs_extent_inline_ref_size(type
);
1293 ret
= btrfs_extend_item(trans
, root
, path
, size
);
1296 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1297 refs
= btrfs_extent_refs(leaf
, ei
);
1298 refs
+= refs_to_add
;
1299 btrfs_set_extent_refs(leaf
, ei
, refs
);
1301 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1303 ptr
= (unsigned long)ei
+ item_offset
;
1304 end
= (unsigned long)ei
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
1305 if (ptr
< end
- size
)
1306 memmove_extent_buffer(leaf
, ptr
+ size
, ptr
,
1309 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1310 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
1311 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1312 struct btrfs_extent_data_ref
*dref
;
1313 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1314 btrfs_set_extent_data_ref_root(leaf
, dref
, root_objectid
);
1315 btrfs_set_extent_data_ref_objectid(leaf
, dref
, owner
);
1316 btrfs_set_extent_data_ref_offset(leaf
, dref
, offset
);
1317 btrfs_set_extent_data_ref_count(leaf
, dref
, refs_to_add
);
1318 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1319 struct btrfs_shared_data_ref
*sref
;
1320 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1321 btrfs_set_shared_data_ref_count(leaf
, sref
, refs_to_add
);
1322 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1323 } else if (type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
1324 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1326 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
1328 btrfs_mark_buffer_dirty(leaf
);
1332 static int lookup_extent_backref(struct btrfs_trans_handle
*trans
,
1333 struct btrfs_root
*root
,
1334 struct btrfs_path
*path
,
1335 struct btrfs_extent_inline_ref
**ref_ret
,
1336 u64 bytenr
, u64 num_bytes
, u64 parent
,
1337 u64 root_objectid
, u64 owner
, u64 offset
)
1341 ret
= lookup_inline_extent_backref(trans
, root
, path
, ref_ret
,
1342 bytenr
, num_bytes
, parent
,
1343 root_objectid
, owner
, offset
, 0);
1347 btrfs_release_path(root
, path
);
1350 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1351 ret
= lookup_tree_block_ref(trans
, root
, path
, bytenr
, parent
,
1354 ret
= lookup_extent_data_ref(trans
, root
, path
, bytenr
, parent
,
1355 root_objectid
, owner
, offset
);
1361 * helper to update/remove inline back ref
1363 static noinline_for_stack
1364 int update_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1365 struct btrfs_root
*root
,
1366 struct btrfs_path
*path
,
1367 struct btrfs_extent_inline_ref
*iref
,
1369 struct btrfs_delayed_extent_op
*extent_op
)
1371 struct extent_buffer
*leaf
;
1372 struct btrfs_extent_item
*ei
;
1373 struct btrfs_extent_data_ref
*dref
= NULL
;
1374 struct btrfs_shared_data_ref
*sref
= NULL
;
1383 leaf
= path
->nodes
[0];
1384 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1385 refs
= btrfs_extent_refs(leaf
, ei
);
1386 WARN_ON(refs_to_mod
< 0 && refs
+ refs_to_mod
<= 0);
1387 refs
+= refs_to_mod
;
1388 btrfs_set_extent_refs(leaf
, ei
, refs
);
1390 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1392 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1394 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1395 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1396 refs
= btrfs_extent_data_ref_count(leaf
, dref
);
1397 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1398 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1399 refs
= btrfs_shared_data_ref_count(leaf
, sref
);
1402 BUG_ON(refs_to_mod
!= -1);
1405 BUG_ON(refs_to_mod
< 0 && refs
< -refs_to_mod
);
1406 refs
+= refs_to_mod
;
1409 if (type
== BTRFS_EXTENT_DATA_REF_KEY
)
1410 btrfs_set_extent_data_ref_count(leaf
, dref
, refs
);
1412 btrfs_set_shared_data_ref_count(leaf
, sref
, refs
);
1414 size
= btrfs_extent_inline_ref_size(type
);
1415 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1416 ptr
= (unsigned long)iref
;
1417 end
= (unsigned long)ei
+ item_size
;
1418 if (ptr
+ size
< end
)
1419 memmove_extent_buffer(leaf
, ptr
, ptr
+ size
,
1422 ret
= btrfs_truncate_item(trans
, root
, path
, item_size
, 1);
1425 btrfs_mark_buffer_dirty(leaf
);
1429 static noinline_for_stack
1430 int insert_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1431 struct btrfs_root
*root
,
1432 struct btrfs_path
*path
,
1433 u64 bytenr
, u64 num_bytes
, u64 parent
,
1434 u64 root_objectid
, u64 owner
,
1435 u64 offset
, int refs_to_add
,
1436 struct btrfs_delayed_extent_op
*extent_op
)
1438 struct btrfs_extent_inline_ref
*iref
;
1441 ret
= lookup_inline_extent_backref(trans
, root
, path
, &iref
,
1442 bytenr
, num_bytes
, parent
,
1443 root_objectid
, owner
, offset
, 1);
1445 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
);
1446 ret
= update_inline_extent_backref(trans
, root
, path
, iref
,
1447 refs_to_add
, extent_op
);
1448 } else if (ret
== -ENOENT
) {
1449 ret
= setup_inline_extent_backref(trans
, root
, path
, iref
,
1450 parent
, root_objectid
,
1451 owner
, offset
, refs_to_add
,
1457 static int insert_extent_backref(struct btrfs_trans_handle
*trans
,
1458 struct btrfs_root
*root
,
1459 struct btrfs_path
*path
,
1460 u64 bytenr
, u64 parent
, u64 root_objectid
,
1461 u64 owner
, u64 offset
, int refs_to_add
)
1464 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1465 BUG_ON(refs_to_add
!= 1);
1466 ret
= insert_tree_block_ref(trans
, root
, path
, bytenr
,
1467 parent
, root_objectid
);
1469 ret
= insert_extent_data_ref(trans
, root
, path
, bytenr
,
1470 parent
, root_objectid
,
1471 owner
, offset
, refs_to_add
);
1476 static int remove_extent_backref(struct btrfs_trans_handle
*trans
,
1477 struct btrfs_root
*root
,
1478 struct btrfs_path
*path
,
1479 struct btrfs_extent_inline_ref
*iref
,
1480 int refs_to_drop
, int is_data
)
1484 BUG_ON(!is_data
&& refs_to_drop
!= 1);
1486 ret
= update_inline_extent_backref(trans
, root
, path
, iref
,
1487 -refs_to_drop
, NULL
);
1488 } else if (is_data
) {
1489 ret
= remove_extent_data_ref(trans
, root
, path
, refs_to_drop
);
1491 ret
= btrfs_del_item(trans
, root
, path
);
1496 #ifdef BIO_RW_DISCARD
1497 static void btrfs_issue_discard(struct block_device
*bdev
,
1500 blkdev_issue_discard(bdev
, start
>> 9, len
>> 9, GFP_KERNEL
);
1504 static int btrfs_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
1507 #ifdef BIO_RW_DISCARD
1509 u64 map_length
= num_bytes
;
1510 struct btrfs_multi_bio
*multi
= NULL
;
1512 /* Tell the block device(s) that the sectors can be discarded */
1513 ret
= btrfs_map_block(&root
->fs_info
->mapping_tree
, READ
,
1514 bytenr
, &map_length
, &multi
, 0);
1516 struct btrfs_bio_stripe
*stripe
= multi
->stripes
;
1519 if (map_length
> num_bytes
)
1520 map_length
= num_bytes
;
1522 for (i
= 0; i
< multi
->num_stripes
; i
++, stripe
++) {
1523 btrfs_issue_discard(stripe
->dev
->bdev
,
1536 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1537 struct btrfs_root
*root
,
1538 u64 bytenr
, u64 num_bytes
, u64 parent
,
1539 u64 root_objectid
, u64 owner
, u64 offset
)
1542 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
&&
1543 root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
1545 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1546 ret
= btrfs_add_delayed_tree_ref(trans
, bytenr
, num_bytes
,
1547 parent
, root_objectid
, (int)owner
,
1548 BTRFS_ADD_DELAYED_REF
, NULL
);
1550 ret
= btrfs_add_delayed_data_ref(trans
, bytenr
, num_bytes
,
1551 parent
, root_objectid
, owner
, offset
,
1552 BTRFS_ADD_DELAYED_REF
, NULL
);
1557 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1558 struct btrfs_root
*root
,
1559 u64 bytenr
, u64 num_bytes
,
1560 u64 parent
, u64 root_objectid
,
1561 u64 owner
, u64 offset
, int refs_to_add
,
1562 struct btrfs_delayed_extent_op
*extent_op
)
1564 struct btrfs_path
*path
;
1565 struct extent_buffer
*leaf
;
1566 struct btrfs_extent_item
*item
;
1571 path
= btrfs_alloc_path();
1576 path
->leave_spinning
= 1;
1577 /* this will setup the path even if it fails to insert the back ref */
1578 ret
= insert_inline_extent_backref(trans
, root
->fs_info
->extent_root
,
1579 path
, bytenr
, num_bytes
, parent
,
1580 root_objectid
, owner
, offset
,
1581 refs_to_add
, extent_op
);
1585 if (ret
!= -EAGAIN
) {
1590 leaf
= path
->nodes
[0];
1591 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1592 refs
= btrfs_extent_refs(leaf
, item
);
1593 btrfs_set_extent_refs(leaf
, item
, refs
+ refs_to_add
);
1595 __run_delayed_extent_op(extent_op
, leaf
, item
);
1597 btrfs_mark_buffer_dirty(leaf
);
1598 btrfs_release_path(root
->fs_info
->extent_root
, path
);
1601 path
->leave_spinning
= 1;
1603 /* now insert the actual backref */
1604 ret
= insert_extent_backref(trans
, root
->fs_info
->extent_root
,
1605 path
, bytenr
, parent
, root_objectid
,
1606 owner
, offset
, refs_to_add
);
1609 btrfs_free_path(path
);
1613 static int run_delayed_data_ref(struct btrfs_trans_handle
*trans
,
1614 struct btrfs_root
*root
,
1615 struct btrfs_delayed_ref_node
*node
,
1616 struct btrfs_delayed_extent_op
*extent_op
,
1617 int insert_reserved
)
1620 struct btrfs_delayed_data_ref
*ref
;
1621 struct btrfs_key ins
;
1626 ins
.objectid
= node
->bytenr
;
1627 ins
.offset
= node
->num_bytes
;
1628 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1630 ref
= btrfs_delayed_node_to_data_ref(node
);
1631 if (node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
1632 parent
= ref
->parent
;
1634 ref_root
= ref
->root
;
1636 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
1638 BUG_ON(extent_op
->update_key
);
1639 flags
|= extent_op
->flags_to_set
;
1641 ret
= alloc_reserved_file_extent(trans
, root
,
1642 parent
, ref_root
, flags
,
1643 ref
->objectid
, ref
->offset
,
1644 &ins
, node
->ref_mod
);
1645 update_reserved_extents(root
, ins
.objectid
, ins
.offset
, 0);
1646 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
1647 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
1648 node
->num_bytes
, parent
,
1649 ref_root
, ref
->objectid
,
1650 ref
->offset
, node
->ref_mod
,
1652 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
1653 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
1654 node
->num_bytes
, parent
,
1655 ref_root
, ref
->objectid
,
1656 ref
->offset
, node
->ref_mod
,
1664 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
1665 struct extent_buffer
*leaf
,
1666 struct btrfs_extent_item
*ei
)
1668 u64 flags
= btrfs_extent_flags(leaf
, ei
);
1669 if (extent_op
->update_flags
) {
1670 flags
|= extent_op
->flags_to_set
;
1671 btrfs_set_extent_flags(leaf
, ei
, flags
);
1674 if (extent_op
->update_key
) {
1675 struct btrfs_tree_block_info
*bi
;
1676 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
));
1677 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
1678 btrfs_set_tree_block_key(leaf
, bi
, &extent_op
->key
);
1682 static int run_delayed_extent_op(struct btrfs_trans_handle
*trans
,
1683 struct btrfs_root
*root
,
1684 struct btrfs_delayed_ref_node
*node
,
1685 struct btrfs_delayed_extent_op
*extent_op
)
1687 struct btrfs_key key
;
1688 struct btrfs_path
*path
;
1689 struct btrfs_extent_item
*ei
;
1690 struct extent_buffer
*leaf
;
1695 path
= btrfs_alloc_path();
1699 key
.objectid
= node
->bytenr
;
1700 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1701 key
.offset
= node
->num_bytes
;
1704 path
->leave_spinning
= 1;
1705 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
,
1716 leaf
= path
->nodes
[0];
1717 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1718 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1719 if (item_size
< sizeof(*ei
)) {
1720 ret
= convert_extent_item_v0(trans
, root
->fs_info
->extent_root
,
1726 leaf
= path
->nodes
[0];
1727 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1730 BUG_ON(item_size
< sizeof(*ei
));
1731 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1732 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1734 btrfs_mark_buffer_dirty(leaf
);
1736 btrfs_free_path(path
);
1740 static int run_delayed_tree_ref(struct btrfs_trans_handle
*trans
,
1741 struct btrfs_root
*root
,
1742 struct btrfs_delayed_ref_node
*node
,
1743 struct btrfs_delayed_extent_op
*extent_op
,
1744 int insert_reserved
)
1747 struct btrfs_delayed_tree_ref
*ref
;
1748 struct btrfs_key ins
;
1752 ins
.objectid
= node
->bytenr
;
1753 ins
.offset
= node
->num_bytes
;
1754 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1756 ref
= btrfs_delayed_node_to_tree_ref(node
);
1757 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
1758 parent
= ref
->parent
;
1760 ref_root
= ref
->root
;
1762 BUG_ON(node
->ref_mod
!= 1);
1763 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
1764 BUG_ON(!extent_op
|| !extent_op
->update_flags
||
1765 !extent_op
->update_key
);
1766 ret
= alloc_reserved_tree_block(trans
, root
,
1768 extent_op
->flags_to_set
,
1771 update_reserved_extents(root
, ins
.objectid
, ins
.offset
, 0);
1772 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
1773 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
1774 node
->num_bytes
, parent
, ref_root
,
1775 ref
->level
, 0, 1, extent_op
);
1776 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
1777 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
1778 node
->num_bytes
, parent
, ref_root
,
1779 ref
->level
, 0, 1, extent_op
);
1787 /* helper function to actually process a single delayed ref entry */
1788 static int run_one_delayed_ref(struct btrfs_trans_handle
*trans
,
1789 struct btrfs_root
*root
,
1790 struct btrfs_delayed_ref_node
*node
,
1791 struct btrfs_delayed_extent_op
*extent_op
,
1792 int insert_reserved
)
1795 if (btrfs_delayed_ref_is_head(node
)) {
1796 struct btrfs_delayed_ref_head
*head
;
1798 * we've hit the end of the chain and we were supposed
1799 * to insert this extent into the tree. But, it got
1800 * deleted before we ever needed to insert it, so all
1801 * we have to do is clean up the accounting
1804 head
= btrfs_delayed_node_to_head(node
);
1805 if (insert_reserved
) {
1806 if (head
->is_data
) {
1807 ret
= btrfs_del_csums(trans
, root
,
1812 btrfs_update_pinned_extents(root
, node
->bytenr
,
1813 node
->num_bytes
, 1);
1814 update_reserved_extents(root
, node
->bytenr
,
1815 node
->num_bytes
, 0);
1817 mutex_unlock(&head
->mutex
);
1821 if (node
->type
== BTRFS_TREE_BLOCK_REF_KEY
||
1822 node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
1823 ret
= run_delayed_tree_ref(trans
, root
, node
, extent_op
,
1825 else if (node
->type
== BTRFS_EXTENT_DATA_REF_KEY
||
1826 node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
1827 ret
= run_delayed_data_ref(trans
, root
, node
, extent_op
,
1834 static noinline
struct btrfs_delayed_ref_node
*
1835 select_delayed_ref(struct btrfs_delayed_ref_head
*head
)
1837 struct rb_node
*node
;
1838 struct btrfs_delayed_ref_node
*ref
;
1839 int action
= BTRFS_ADD_DELAYED_REF
;
1842 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
1843 * this prevents ref count from going down to zero when
1844 * there still are pending delayed ref.
1846 node
= rb_prev(&head
->node
.rb_node
);
1850 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
1852 if (ref
->bytenr
!= head
->node
.bytenr
)
1854 if (ref
->action
== action
)
1856 node
= rb_prev(node
);
1858 if (action
== BTRFS_ADD_DELAYED_REF
) {
1859 action
= BTRFS_DROP_DELAYED_REF
;
1865 static noinline
int run_clustered_refs(struct btrfs_trans_handle
*trans
,
1866 struct btrfs_root
*root
,
1867 struct list_head
*cluster
)
1869 struct btrfs_delayed_ref_root
*delayed_refs
;
1870 struct btrfs_delayed_ref_node
*ref
;
1871 struct btrfs_delayed_ref_head
*locked_ref
= NULL
;
1872 struct btrfs_delayed_extent_op
*extent_op
;
1875 int must_insert_reserved
= 0;
1877 delayed_refs
= &trans
->transaction
->delayed_refs
;
1880 /* pick a new head ref from the cluster list */
1881 if (list_empty(cluster
))
1884 locked_ref
= list_entry(cluster
->next
,
1885 struct btrfs_delayed_ref_head
, cluster
);
1887 /* grab the lock that says we are going to process
1888 * all the refs for this head */
1889 ret
= btrfs_delayed_ref_lock(trans
, locked_ref
);
1892 * we may have dropped the spin lock to get the head
1893 * mutex lock, and that might have given someone else
1894 * time to free the head. If that's true, it has been
1895 * removed from our list and we can move on.
1897 if (ret
== -EAGAIN
) {
1905 * record the must insert reserved flag before we
1906 * drop the spin lock.
1908 must_insert_reserved
= locked_ref
->must_insert_reserved
;
1909 locked_ref
->must_insert_reserved
= 0;
1911 extent_op
= locked_ref
->extent_op
;
1912 locked_ref
->extent_op
= NULL
;
1915 * locked_ref is the head node, so we have to go one
1916 * node back for any delayed ref updates
1918 ref
= select_delayed_ref(locked_ref
);
1920 /* All delayed refs have been processed, Go ahead
1921 * and send the head node to run_one_delayed_ref,
1922 * so that any accounting fixes can happen
1924 ref
= &locked_ref
->node
;
1926 if (extent_op
&& must_insert_reserved
) {
1932 spin_unlock(&delayed_refs
->lock
);
1934 ret
= run_delayed_extent_op(trans
, root
,
1940 spin_lock(&delayed_refs
->lock
);
1944 list_del_init(&locked_ref
->cluster
);
1949 rb_erase(&ref
->rb_node
, &delayed_refs
->root
);
1950 delayed_refs
->num_entries
--;
1952 spin_unlock(&delayed_refs
->lock
);
1954 ret
= run_one_delayed_ref(trans
, root
, ref
, extent_op
,
1955 must_insert_reserved
);
1958 btrfs_put_delayed_ref(ref
);
1963 spin_lock(&delayed_refs
->lock
);
1969 * this starts processing the delayed reference count updates and
1970 * extent insertions we have queued up so far. count can be
1971 * 0, which means to process everything in the tree at the start
1972 * of the run (but not newly added entries), or it can be some target
1973 * number you'd like to process.
1975 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
1976 struct btrfs_root
*root
, unsigned long count
)
1978 struct rb_node
*node
;
1979 struct btrfs_delayed_ref_root
*delayed_refs
;
1980 struct btrfs_delayed_ref_node
*ref
;
1981 struct list_head cluster
;
1983 int run_all
= count
== (unsigned long)-1;
1986 if (root
== root
->fs_info
->extent_root
)
1987 root
= root
->fs_info
->tree_root
;
1989 delayed_refs
= &trans
->transaction
->delayed_refs
;
1990 INIT_LIST_HEAD(&cluster
);
1992 spin_lock(&delayed_refs
->lock
);
1994 count
= delayed_refs
->num_entries
* 2;
1998 if (!(run_all
|| run_most
) &&
1999 delayed_refs
->num_heads_ready
< 64)
2003 * go find something we can process in the rbtree. We start at
2004 * the beginning of the tree, and then build a cluster
2005 * of refs to process starting at the first one we are able to
2008 ret
= btrfs_find_ref_cluster(trans
, &cluster
,
2009 delayed_refs
->run_delayed_start
);
2013 ret
= run_clustered_refs(trans
, root
, &cluster
);
2016 count
-= min_t(unsigned long, ret
, count
);
2023 node
= rb_first(&delayed_refs
->root
);
2026 count
= (unsigned long)-1;
2029 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
2031 if (btrfs_delayed_ref_is_head(ref
)) {
2032 struct btrfs_delayed_ref_head
*head
;
2034 head
= btrfs_delayed_node_to_head(ref
);
2035 atomic_inc(&ref
->refs
);
2037 spin_unlock(&delayed_refs
->lock
);
2038 mutex_lock(&head
->mutex
);
2039 mutex_unlock(&head
->mutex
);
2041 btrfs_put_delayed_ref(ref
);
2045 node
= rb_next(node
);
2047 spin_unlock(&delayed_refs
->lock
);
2048 schedule_timeout(1);
2052 spin_unlock(&delayed_refs
->lock
);
2056 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
2057 struct btrfs_root
*root
,
2058 u64 bytenr
, u64 num_bytes
, u64 flags
,
2061 struct btrfs_delayed_extent_op
*extent_op
;
2064 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
2068 extent_op
->flags_to_set
= flags
;
2069 extent_op
->update_flags
= 1;
2070 extent_op
->update_key
= 0;
2071 extent_op
->is_data
= is_data
? 1 : 0;
2073 ret
= btrfs_add_delayed_extent_op(trans
, bytenr
, num_bytes
, extent_op
);
2079 static noinline
int check_delayed_ref(struct btrfs_trans_handle
*trans
,
2080 struct btrfs_root
*root
,
2081 struct btrfs_path
*path
,
2082 u64 objectid
, u64 offset
, u64 bytenr
)
2084 struct btrfs_delayed_ref_head
*head
;
2085 struct btrfs_delayed_ref_node
*ref
;
2086 struct btrfs_delayed_data_ref
*data_ref
;
2087 struct btrfs_delayed_ref_root
*delayed_refs
;
2088 struct rb_node
*node
;
2092 delayed_refs
= &trans
->transaction
->delayed_refs
;
2093 spin_lock(&delayed_refs
->lock
);
2094 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
2098 if (!mutex_trylock(&head
->mutex
)) {
2099 atomic_inc(&head
->node
.refs
);
2100 spin_unlock(&delayed_refs
->lock
);
2102 btrfs_release_path(root
->fs_info
->extent_root
, path
);
2104 mutex_lock(&head
->mutex
);
2105 mutex_unlock(&head
->mutex
);
2106 btrfs_put_delayed_ref(&head
->node
);
2110 node
= rb_prev(&head
->node
.rb_node
);
2114 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2116 if (ref
->bytenr
!= bytenr
)
2120 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
)
2123 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
2125 node
= rb_prev(node
);
2127 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2128 if (ref
->bytenr
== bytenr
)
2132 if (data_ref
->root
!= root
->root_key
.objectid
||
2133 data_ref
->objectid
!= objectid
|| data_ref
->offset
!= offset
)
2138 mutex_unlock(&head
->mutex
);
2140 spin_unlock(&delayed_refs
->lock
);
2144 static noinline
int check_committed_ref(struct btrfs_trans_handle
*trans
,
2145 struct btrfs_root
*root
,
2146 struct btrfs_path
*path
,
2147 u64 objectid
, u64 offset
, u64 bytenr
)
2149 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2150 struct extent_buffer
*leaf
;
2151 struct btrfs_extent_data_ref
*ref
;
2152 struct btrfs_extent_inline_ref
*iref
;
2153 struct btrfs_extent_item
*ei
;
2154 struct btrfs_key key
;
2158 key
.objectid
= bytenr
;
2159 key
.offset
= (u64
)-1;
2160 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2162 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
2168 if (path
->slots
[0] == 0)
2172 leaf
= path
->nodes
[0];
2173 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2175 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
2179 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2180 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2181 if (item_size
< sizeof(*ei
)) {
2182 WARN_ON(item_size
!= sizeof(struct btrfs_extent_item_v0
));
2186 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2188 if (item_size
!= sizeof(*ei
) +
2189 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
2192 if (btrfs_extent_generation(leaf
, ei
) <=
2193 btrfs_root_last_snapshot(&root
->root_item
))
2196 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
2197 if (btrfs_extent_inline_ref_type(leaf
, iref
) !=
2198 BTRFS_EXTENT_DATA_REF_KEY
)
2201 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
2202 if (btrfs_extent_refs(leaf
, ei
) !=
2203 btrfs_extent_data_ref_count(leaf
, ref
) ||
2204 btrfs_extent_data_ref_root(leaf
, ref
) !=
2205 root
->root_key
.objectid
||
2206 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
2207 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
2215 int btrfs_cross_ref_exist(struct btrfs_trans_handle
*trans
,
2216 struct btrfs_root
*root
,
2217 u64 objectid
, u64 offset
, u64 bytenr
)
2219 struct btrfs_path
*path
;
2223 path
= btrfs_alloc_path();
2228 ret
= check_committed_ref(trans
, root
, path
, objectid
,
2230 if (ret
&& ret
!= -ENOENT
)
2233 ret2
= check_delayed_ref(trans
, root
, path
, objectid
,
2235 } while (ret2
== -EAGAIN
);
2237 if (ret2
&& ret2
!= -ENOENT
) {
2242 if (ret
!= -ENOENT
|| ret2
!= -ENOENT
)
2245 btrfs_free_path(path
);
2250 int btrfs_cache_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2251 struct extent_buffer
*buf
, u32 nr_extents
)
2253 struct btrfs_key key
;
2254 struct btrfs_file_extent_item
*fi
;
2262 if (!root
->ref_cows
)
2265 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
2267 root_gen
= root
->root_key
.offset
;
2270 root_gen
= trans
->transid
- 1;
2273 level
= btrfs_header_level(buf
);
2274 nritems
= btrfs_header_nritems(buf
);
2277 struct btrfs_leaf_ref
*ref
;
2278 struct btrfs_extent_info
*info
;
2280 ref
= btrfs_alloc_leaf_ref(root
, nr_extents
);
2286 ref
->root_gen
= root_gen
;
2287 ref
->bytenr
= buf
->start
;
2288 ref
->owner
= btrfs_header_owner(buf
);
2289 ref
->generation
= btrfs_header_generation(buf
);
2290 ref
->nritems
= nr_extents
;
2291 info
= ref
->extents
;
2293 for (i
= 0; nr_extents
> 0 && i
< nritems
; i
++) {
2295 btrfs_item_key_to_cpu(buf
, &key
, i
);
2296 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
2298 fi
= btrfs_item_ptr(buf
, i
,
2299 struct btrfs_file_extent_item
);
2300 if (btrfs_file_extent_type(buf
, fi
) ==
2301 BTRFS_FILE_EXTENT_INLINE
)
2303 disk_bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
2304 if (disk_bytenr
== 0)
2307 info
->bytenr
= disk_bytenr
;
2309 btrfs_file_extent_disk_num_bytes(buf
, fi
);
2310 info
->objectid
= key
.objectid
;
2311 info
->offset
= key
.offset
;
2315 ret
= btrfs_add_leaf_ref(root
, ref
, shared
);
2316 if (ret
== -EEXIST
&& shared
) {
2317 struct btrfs_leaf_ref
*old
;
2318 old
= btrfs_lookup_leaf_ref(root
, ref
->bytenr
);
2320 btrfs_remove_leaf_ref(root
, old
);
2321 btrfs_free_leaf_ref(root
, old
);
2322 ret
= btrfs_add_leaf_ref(root
, ref
, shared
);
2325 btrfs_free_leaf_ref(root
, ref
);
2331 /* when a block goes through cow, we update the reference counts of
2332 * everything that block points to. The internal pointers of the block
2333 * can be in just about any order, and it is likely to have clusters of
2334 * things that are close together and clusters of things that are not.
2336 * To help reduce the seeks that come with updating all of these reference
2337 * counts, sort them by byte number before actual updates are done.
2339 * struct refsort is used to match byte number to slot in the btree block.
2340 * we sort based on the byte number and then use the slot to actually
2343 * struct refsort is smaller than strcut btrfs_item and smaller than
2344 * struct btrfs_key_ptr. Since we're currently limited to the page size
2345 * for a btree block, there's no way for a kmalloc of refsorts for a
2346 * single node to be bigger than a page.
2354 * for passing into sort()
2356 static int refsort_cmp(const void *a_void
, const void *b_void
)
2358 const struct refsort
*a
= a_void
;
2359 const struct refsort
*b
= b_void
;
2361 if (a
->bytenr
< b
->bytenr
)
2363 if (a
->bytenr
> b
->bytenr
)
2369 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
2370 struct btrfs_root
*root
,
2371 struct extent_buffer
*buf
,
2372 int full_backref
, int inc
)
2379 struct btrfs_key key
;
2380 struct btrfs_file_extent_item
*fi
;
2384 int (*process_func
)(struct btrfs_trans_handle
*, struct btrfs_root
*,
2385 u64
, u64
, u64
, u64
, u64
, u64
);
2387 ref_root
= btrfs_header_owner(buf
);
2388 nritems
= btrfs_header_nritems(buf
);
2389 level
= btrfs_header_level(buf
);
2391 if (!root
->ref_cows
&& level
== 0)
2395 process_func
= btrfs_inc_extent_ref
;
2397 process_func
= btrfs_free_extent
;
2400 parent
= buf
->start
;
2404 for (i
= 0; i
< nritems
; i
++) {
2406 btrfs_item_key_to_cpu(buf
, &key
, i
);
2407 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
2409 fi
= btrfs_item_ptr(buf
, i
,
2410 struct btrfs_file_extent_item
);
2411 if (btrfs_file_extent_type(buf
, fi
) ==
2412 BTRFS_FILE_EXTENT_INLINE
)
2414 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
2418 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
2419 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
2420 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
2421 parent
, ref_root
, key
.objectid
,
2426 bytenr
= btrfs_node_blockptr(buf
, i
);
2427 num_bytes
= btrfs_level_size(root
, level
- 1);
2428 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
2429 parent
, ref_root
, level
- 1, 0);
2440 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2441 struct extent_buffer
*buf
, int full_backref
)
2443 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1);
2446 int btrfs_dec_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2447 struct extent_buffer
*buf
, int full_backref
)
2449 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 0);
2452 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
2453 struct btrfs_root
*root
,
2454 struct btrfs_path
*path
,
2455 struct btrfs_block_group_cache
*cache
)
2458 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2460 struct extent_buffer
*leaf
;
2462 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
2467 leaf
= path
->nodes
[0];
2468 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
2469 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
2470 btrfs_mark_buffer_dirty(leaf
);
2471 btrfs_release_path(extent_root
, path
);
2479 static struct btrfs_block_group_cache
*
2480 next_block_group(struct btrfs_root
*root
,
2481 struct btrfs_block_group_cache
*cache
)
2483 struct rb_node
*node
;
2484 spin_lock(&root
->fs_info
->block_group_cache_lock
);
2485 node
= rb_next(&cache
->cache_node
);
2486 btrfs_put_block_group(cache
);
2488 cache
= rb_entry(node
, struct btrfs_block_group_cache
,
2490 atomic_inc(&cache
->count
);
2493 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
2497 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
2498 struct btrfs_root
*root
)
2500 struct btrfs_block_group_cache
*cache
;
2502 struct btrfs_path
*path
;
2505 path
= btrfs_alloc_path();
2511 err
= btrfs_run_delayed_refs(trans
, root
,
2516 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
2520 cache
= next_block_group(root
, cache
);
2530 last
= cache
->key
.objectid
+ cache
->key
.offset
;
2532 err
= write_one_cache_group(trans
, root
, path
, cache
);
2534 btrfs_put_block_group(cache
);
2537 btrfs_free_path(path
);
2541 int btrfs_extent_readonly(struct btrfs_root
*root
, u64 bytenr
)
2543 struct btrfs_block_group_cache
*block_group
;
2546 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
2547 if (!block_group
|| block_group
->ro
)
2550 btrfs_put_block_group(block_group
);
2554 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
2555 u64 total_bytes
, u64 bytes_used
,
2556 struct btrfs_space_info
**space_info
)
2558 struct btrfs_space_info
*found
;
2560 found
= __find_space_info(info
, flags
);
2562 spin_lock(&found
->lock
);
2563 found
->total_bytes
+= total_bytes
;
2564 found
->bytes_used
+= bytes_used
;
2566 spin_unlock(&found
->lock
);
2567 *space_info
= found
;
2570 found
= kzalloc(sizeof(*found
), GFP_NOFS
);
2574 INIT_LIST_HEAD(&found
->block_groups
);
2575 init_rwsem(&found
->groups_sem
);
2576 spin_lock_init(&found
->lock
);
2577 found
->flags
= flags
;
2578 found
->total_bytes
= total_bytes
;
2579 found
->bytes_used
= bytes_used
;
2580 found
->bytes_pinned
= 0;
2581 found
->bytes_reserved
= 0;
2582 found
->bytes_readonly
= 0;
2583 found
->bytes_delalloc
= 0;
2585 found
->force_alloc
= 0;
2586 *space_info
= found
;
2587 list_add_rcu(&found
->list
, &info
->space_info
);
2588 atomic_set(&found
->caching_threads
, 0);
2592 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
2594 u64 extra_flags
= flags
& (BTRFS_BLOCK_GROUP_RAID0
|
2595 BTRFS_BLOCK_GROUP_RAID1
|
2596 BTRFS_BLOCK_GROUP_RAID10
|
2597 BTRFS_BLOCK_GROUP_DUP
);
2599 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
2600 fs_info
->avail_data_alloc_bits
|= extra_flags
;
2601 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
2602 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
2603 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
2604 fs_info
->avail_system_alloc_bits
|= extra_flags
;
2608 static void set_block_group_readonly(struct btrfs_block_group_cache
*cache
)
2610 spin_lock(&cache
->space_info
->lock
);
2611 spin_lock(&cache
->lock
);
2613 cache
->space_info
->bytes_readonly
+= cache
->key
.offset
-
2614 btrfs_block_group_used(&cache
->item
);
2617 spin_unlock(&cache
->lock
);
2618 spin_unlock(&cache
->space_info
->lock
);
2621 u64
btrfs_reduce_alloc_profile(struct btrfs_root
*root
, u64 flags
)
2623 u64 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
2625 if (num_devices
== 1)
2626 flags
&= ~(BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID0
);
2627 if (num_devices
< 4)
2628 flags
&= ~BTRFS_BLOCK_GROUP_RAID10
;
2630 if ((flags
& BTRFS_BLOCK_GROUP_DUP
) &&
2631 (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
2632 BTRFS_BLOCK_GROUP_RAID10
))) {
2633 flags
&= ~BTRFS_BLOCK_GROUP_DUP
;
2636 if ((flags
& BTRFS_BLOCK_GROUP_RAID1
) &&
2637 (flags
& BTRFS_BLOCK_GROUP_RAID10
)) {
2638 flags
&= ~BTRFS_BLOCK_GROUP_RAID1
;
2641 if ((flags
& BTRFS_BLOCK_GROUP_RAID0
) &&
2642 ((flags
& BTRFS_BLOCK_GROUP_RAID1
) |
2643 (flags
& BTRFS_BLOCK_GROUP_RAID10
) |
2644 (flags
& BTRFS_BLOCK_GROUP_DUP
)))
2645 flags
&= ~BTRFS_BLOCK_GROUP_RAID0
;
2649 static u64
btrfs_get_alloc_profile(struct btrfs_root
*root
, u64 data
)
2651 struct btrfs_fs_info
*info
= root
->fs_info
;
2655 alloc_profile
= info
->avail_data_alloc_bits
&
2656 info
->data_alloc_profile
;
2657 data
= BTRFS_BLOCK_GROUP_DATA
| alloc_profile
;
2658 } else if (root
== root
->fs_info
->chunk_root
) {
2659 alloc_profile
= info
->avail_system_alloc_bits
&
2660 info
->system_alloc_profile
;
2661 data
= BTRFS_BLOCK_GROUP_SYSTEM
| alloc_profile
;
2663 alloc_profile
= info
->avail_metadata_alloc_bits
&
2664 info
->metadata_alloc_profile
;
2665 data
= BTRFS_BLOCK_GROUP_METADATA
| alloc_profile
;
2668 return btrfs_reduce_alloc_profile(root
, data
);
2671 void btrfs_set_inode_space_info(struct btrfs_root
*root
, struct inode
*inode
)
2675 alloc_target
= btrfs_get_alloc_profile(root
, 1);
2676 BTRFS_I(inode
)->space_info
= __find_space_info(root
->fs_info
,
2681 * for now this just makes sure we have at least 5% of our metadata space free
2684 int btrfs_check_metadata_free_space(struct btrfs_root
*root
)
2686 struct btrfs_fs_info
*info
= root
->fs_info
;
2687 struct btrfs_space_info
*meta_sinfo
;
2688 u64 alloc_target
, thresh
;
2689 int committed
= 0, ret
;
2691 /* get the space info for where the metadata will live */
2692 alloc_target
= btrfs_get_alloc_profile(root
, 0);
2693 meta_sinfo
= __find_space_info(info
, alloc_target
);
2696 spin_lock(&meta_sinfo
->lock
);
2697 if (!meta_sinfo
->full
)
2698 thresh
= meta_sinfo
->total_bytes
* 80;
2700 thresh
= meta_sinfo
->total_bytes
* 95;
2702 do_div(thresh
, 100);
2704 if (meta_sinfo
->bytes_used
+ meta_sinfo
->bytes_reserved
+
2705 meta_sinfo
->bytes_pinned
+ meta_sinfo
->bytes_readonly
> thresh
) {
2706 struct btrfs_trans_handle
*trans
;
2707 if (!meta_sinfo
->full
) {
2708 meta_sinfo
->force_alloc
= 1;
2709 spin_unlock(&meta_sinfo
->lock
);
2711 trans
= btrfs_start_transaction(root
, 1);
2715 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
2716 2 * 1024 * 1024, alloc_target
, 0);
2717 btrfs_end_transaction(trans
, root
);
2720 spin_unlock(&meta_sinfo
->lock
);
2724 trans
= btrfs_join_transaction(root
, 1);
2727 ret
= btrfs_commit_transaction(trans
, root
);
2734 spin_unlock(&meta_sinfo
->lock
);
2740 * This will check the space that the inode allocates from to make sure we have
2741 * enough space for bytes.
2743 int btrfs_check_data_free_space(struct btrfs_root
*root
, struct inode
*inode
,
2746 struct btrfs_space_info
*data_sinfo
;
2747 int ret
= 0, committed
= 0;
2749 /* make sure bytes are sectorsize aligned */
2750 bytes
= (bytes
+ root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
2752 data_sinfo
= BTRFS_I(inode
)->space_info
;
2754 /* make sure we have enough space to handle the data first */
2755 spin_lock(&data_sinfo
->lock
);
2756 if (data_sinfo
->total_bytes
- data_sinfo
->bytes_used
-
2757 data_sinfo
->bytes_delalloc
- data_sinfo
->bytes_reserved
-
2758 data_sinfo
->bytes_pinned
- data_sinfo
->bytes_readonly
-
2759 data_sinfo
->bytes_may_use
< bytes
) {
2760 struct btrfs_trans_handle
*trans
;
2763 * if we don't have enough free bytes in this space then we need
2764 * to alloc a new chunk.
2766 if (!data_sinfo
->full
) {
2769 data_sinfo
->force_alloc
= 1;
2770 spin_unlock(&data_sinfo
->lock
);
2772 alloc_target
= btrfs_get_alloc_profile(root
, 1);
2773 trans
= btrfs_start_transaction(root
, 1);
2777 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
2778 bytes
+ 2 * 1024 * 1024,
2780 btrfs_end_transaction(trans
, root
);
2785 spin_unlock(&data_sinfo
->lock
);
2787 /* commit the current transaction and try again */
2790 trans
= btrfs_join_transaction(root
, 1);
2793 ret
= btrfs_commit_transaction(trans
, root
);
2799 printk(KERN_ERR
"no space left, need %llu, %llu delalloc bytes"
2800 ", %llu bytes_used, %llu bytes_reserved, "
2801 "%llu bytes_pinned, %llu bytes_readonly, %llu may use "
2802 "%llu total\n", (unsigned long long)bytes
,
2803 (unsigned long long)data_sinfo
->bytes_delalloc
,
2804 (unsigned long long)data_sinfo
->bytes_used
,
2805 (unsigned long long)data_sinfo
->bytes_reserved
,
2806 (unsigned long long)data_sinfo
->bytes_pinned
,
2807 (unsigned long long)data_sinfo
->bytes_readonly
,
2808 (unsigned long long)data_sinfo
->bytes_may_use
,
2809 (unsigned long long)data_sinfo
->total_bytes
);
2812 data_sinfo
->bytes_may_use
+= bytes
;
2813 BTRFS_I(inode
)->reserved_bytes
+= bytes
;
2814 spin_unlock(&data_sinfo
->lock
);
2816 return btrfs_check_metadata_free_space(root
);
2820 * if there was an error for whatever reason after calling
2821 * btrfs_check_data_free_space, call this so we can cleanup the counters.
2823 void btrfs_free_reserved_data_space(struct btrfs_root
*root
,
2824 struct inode
*inode
, u64 bytes
)
2826 struct btrfs_space_info
*data_sinfo
;
2828 /* make sure bytes are sectorsize aligned */
2829 bytes
= (bytes
+ root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
2831 data_sinfo
= BTRFS_I(inode
)->space_info
;
2832 spin_lock(&data_sinfo
->lock
);
2833 data_sinfo
->bytes_may_use
-= bytes
;
2834 BTRFS_I(inode
)->reserved_bytes
-= bytes
;
2835 spin_unlock(&data_sinfo
->lock
);
2838 /* called when we are adding a delalloc extent to the inode's io_tree */
2839 void btrfs_delalloc_reserve_space(struct btrfs_root
*root
, struct inode
*inode
,
2842 struct btrfs_space_info
*data_sinfo
;
2844 /* get the space info for where this inode will be storing its data */
2845 data_sinfo
= BTRFS_I(inode
)->space_info
;
2847 /* make sure we have enough space to handle the data first */
2848 spin_lock(&data_sinfo
->lock
);
2849 data_sinfo
->bytes_delalloc
+= bytes
;
2852 * we are adding a delalloc extent without calling
2853 * btrfs_check_data_free_space first. This happens on a weird
2854 * writepage condition, but shouldn't hurt our accounting
2856 if (unlikely(bytes
> BTRFS_I(inode
)->reserved_bytes
)) {
2857 data_sinfo
->bytes_may_use
-= BTRFS_I(inode
)->reserved_bytes
;
2858 BTRFS_I(inode
)->reserved_bytes
= 0;
2860 data_sinfo
->bytes_may_use
-= bytes
;
2861 BTRFS_I(inode
)->reserved_bytes
-= bytes
;
2864 spin_unlock(&data_sinfo
->lock
);
2867 /* called when we are clearing an delalloc extent from the inode's io_tree */
2868 void btrfs_delalloc_free_space(struct btrfs_root
*root
, struct inode
*inode
,
2871 struct btrfs_space_info
*info
;
2873 info
= BTRFS_I(inode
)->space_info
;
2875 spin_lock(&info
->lock
);
2876 info
->bytes_delalloc
-= bytes
;
2877 spin_unlock(&info
->lock
);
2880 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
2882 struct list_head
*head
= &info
->space_info
;
2883 struct btrfs_space_info
*found
;
2886 list_for_each_entry_rcu(found
, head
, list
) {
2887 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
2888 found
->force_alloc
= 1;
2893 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
2894 struct btrfs_root
*extent_root
, u64 alloc_bytes
,
2895 u64 flags
, int force
)
2897 struct btrfs_space_info
*space_info
;
2898 struct btrfs_fs_info
*fs_info
= extent_root
->fs_info
;
2902 mutex_lock(&fs_info
->chunk_mutex
);
2904 flags
= btrfs_reduce_alloc_profile(extent_root
, flags
);
2906 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
2908 ret
= update_space_info(extent_root
->fs_info
, flags
,
2912 BUG_ON(!space_info
);
2914 spin_lock(&space_info
->lock
);
2915 if (space_info
->force_alloc
) {
2917 space_info
->force_alloc
= 0;
2919 if (space_info
->full
) {
2920 spin_unlock(&space_info
->lock
);
2924 thresh
= space_info
->total_bytes
- space_info
->bytes_readonly
;
2925 thresh
= div_factor(thresh
, 6);
2927 (space_info
->bytes_used
+ space_info
->bytes_pinned
+
2928 space_info
->bytes_reserved
+ alloc_bytes
) < thresh
) {
2929 spin_unlock(&space_info
->lock
);
2932 spin_unlock(&space_info
->lock
);
2935 * if we're doing a data chunk, go ahead and make sure that
2936 * we keep a reasonable number of metadata chunks allocated in the
2939 if (flags
& BTRFS_BLOCK_GROUP_DATA
) {
2940 fs_info
->data_chunk_allocations
++;
2941 if (!(fs_info
->data_chunk_allocations
%
2942 fs_info
->metadata_ratio
))
2943 force_metadata_allocation(fs_info
);
2946 ret
= btrfs_alloc_chunk(trans
, extent_root
, flags
);
2948 space_info
->full
= 1;
2950 mutex_unlock(&extent_root
->fs_info
->chunk_mutex
);
2954 static int update_block_group(struct btrfs_trans_handle
*trans
,
2955 struct btrfs_root
*root
,
2956 u64 bytenr
, u64 num_bytes
, int alloc
,
2959 struct btrfs_block_group_cache
*cache
;
2960 struct btrfs_fs_info
*info
= root
->fs_info
;
2961 u64 total
= num_bytes
;
2965 /* block accounting for super block */
2966 spin_lock(&info
->delalloc_lock
);
2967 old_val
= btrfs_super_bytes_used(&info
->super_copy
);
2969 old_val
+= num_bytes
;
2971 old_val
-= num_bytes
;
2972 btrfs_set_super_bytes_used(&info
->super_copy
, old_val
);
2974 /* block accounting for root item */
2975 old_val
= btrfs_root_used(&root
->root_item
);
2977 old_val
+= num_bytes
;
2979 old_val
-= num_bytes
;
2980 btrfs_set_root_used(&root
->root_item
, old_val
);
2981 spin_unlock(&info
->delalloc_lock
);
2984 cache
= btrfs_lookup_block_group(info
, bytenr
);
2987 byte_in_group
= bytenr
- cache
->key
.objectid
;
2988 WARN_ON(byte_in_group
> cache
->key
.offset
);
2990 spin_lock(&cache
->space_info
->lock
);
2991 spin_lock(&cache
->lock
);
2993 old_val
= btrfs_block_group_used(&cache
->item
);
2994 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
2996 old_val
+= num_bytes
;
2997 cache
->space_info
->bytes_used
+= num_bytes
;
2999 cache
->space_info
->bytes_readonly
-= num_bytes
;
3000 btrfs_set_block_group_used(&cache
->item
, old_val
);
3001 spin_unlock(&cache
->lock
);
3002 spin_unlock(&cache
->space_info
->lock
);
3004 old_val
-= num_bytes
;
3005 cache
->space_info
->bytes_used
-= num_bytes
;
3007 cache
->space_info
->bytes_readonly
+= num_bytes
;
3008 btrfs_set_block_group_used(&cache
->item
, old_val
);
3009 spin_unlock(&cache
->lock
);
3010 spin_unlock(&cache
->space_info
->lock
);
3014 ret
= btrfs_discard_extent(root
, bytenr
,
3018 ret
= btrfs_add_free_space(cache
, bytenr
,
3023 btrfs_put_block_group(cache
);
3025 bytenr
+= num_bytes
;
3030 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
3032 struct btrfs_block_group_cache
*cache
;
3035 cache
= btrfs_lookup_first_block_group(root
->fs_info
, search_start
);
3039 bytenr
= cache
->key
.objectid
;
3040 btrfs_put_block_group(cache
);
3045 int btrfs_update_pinned_extents(struct btrfs_root
*root
,
3046 u64 bytenr
, u64 num
, int pin
)
3049 struct btrfs_block_group_cache
*cache
;
3050 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3053 set_extent_dirty(&fs_info
->pinned_extents
,
3054 bytenr
, bytenr
+ num
- 1, GFP_NOFS
);
3057 cache
= btrfs_lookup_block_group(fs_info
, bytenr
);
3059 len
= min(num
, cache
->key
.offset
-
3060 (bytenr
- cache
->key
.objectid
));
3062 spin_lock(&cache
->space_info
->lock
);
3063 spin_lock(&cache
->lock
);
3064 cache
->pinned
+= len
;
3065 cache
->space_info
->bytes_pinned
+= len
;
3066 spin_unlock(&cache
->lock
);
3067 spin_unlock(&cache
->space_info
->lock
);
3068 fs_info
->total_pinned
+= len
;
3073 * in order to not race with the block group caching, we
3074 * only want to unpin the extent if we are cached. If
3075 * we aren't cached, we want to start async caching this
3076 * block group so we can free the extent the next time
3079 spin_lock(&cache
->space_info
->lock
);
3080 spin_lock(&cache
->lock
);
3081 unpin
= (cache
->cached
== BTRFS_CACHE_FINISHED
);
3082 if (likely(unpin
)) {
3083 cache
->pinned
-= len
;
3084 cache
->space_info
->bytes_pinned
-= len
;
3085 fs_info
->total_pinned
-= len
;
3087 spin_unlock(&cache
->lock
);
3088 spin_unlock(&cache
->space_info
->lock
);
3091 clear_extent_dirty(&fs_info
->pinned_extents
,
3092 bytenr
, bytenr
+ len
-1,
3095 cache_block_group(cache
);
3098 btrfs_add_free_space(cache
, bytenr
, len
);
3100 btrfs_put_block_group(cache
);
3107 static int update_reserved_extents(struct btrfs_root
*root
,
3108 u64 bytenr
, u64 num
, int reserve
)
3111 struct btrfs_block_group_cache
*cache
;
3112 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3115 cache
= btrfs_lookup_block_group(fs_info
, bytenr
);
3117 len
= min(num
, cache
->key
.offset
-
3118 (bytenr
- cache
->key
.objectid
));
3120 spin_lock(&cache
->space_info
->lock
);
3121 spin_lock(&cache
->lock
);
3123 cache
->reserved
+= len
;
3124 cache
->space_info
->bytes_reserved
+= len
;
3126 cache
->reserved
-= len
;
3127 cache
->space_info
->bytes_reserved
-= len
;
3129 spin_unlock(&cache
->lock
);
3130 spin_unlock(&cache
->space_info
->lock
);
3131 btrfs_put_block_group(cache
);
3138 int btrfs_copy_pinned(struct btrfs_root
*root
, struct extent_io_tree
*copy
)
3143 struct extent_io_tree
*pinned_extents
= &root
->fs_info
->pinned_extents
;
3147 ret
= find_first_extent_bit(pinned_extents
, last
,
3148 &start
, &end
, EXTENT_DIRTY
);
3152 set_extent_dirty(copy
, start
, end
, GFP_NOFS
);
3158 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
3159 struct btrfs_root
*root
,
3160 struct extent_io_tree
*unpin
)
3167 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
3172 ret
= btrfs_discard_extent(root
, start
, end
+ 1 - start
);
3174 /* unlocks the pinned mutex */
3175 btrfs_update_pinned_extents(root
, start
, end
+ 1 - start
, 0);
3176 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
3184 static int pin_down_bytes(struct btrfs_trans_handle
*trans
,
3185 struct btrfs_root
*root
,
3186 struct btrfs_path
*path
,
3187 u64 bytenr
, u64 num_bytes
, int is_data
,
3188 struct extent_buffer
**must_clean
)
3191 struct extent_buffer
*buf
;
3196 buf
= btrfs_find_tree_block(root
, bytenr
, num_bytes
);
3200 /* we can reuse a block if it hasn't been written
3201 * and it is from this transaction. We can't
3202 * reuse anything from the tree log root because
3203 * it has tiny sub-transactions.
3205 if (btrfs_buffer_uptodate(buf
, 0) &&
3206 btrfs_try_tree_lock(buf
)) {
3207 u64 header_owner
= btrfs_header_owner(buf
);
3208 u64 header_transid
= btrfs_header_generation(buf
);
3209 if (header_owner
!= BTRFS_TREE_LOG_OBJECTID
&&
3210 header_transid
== trans
->transid
&&
3211 !btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
3215 btrfs_tree_unlock(buf
);
3217 free_extent_buffer(buf
);
3219 btrfs_set_path_blocking(path
);
3220 /* unlocks the pinned mutex */
3221 btrfs_update_pinned_extents(root
, bytenr
, num_bytes
, 1);
3228 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
3229 struct btrfs_root
*root
,
3230 u64 bytenr
, u64 num_bytes
, u64 parent
,
3231 u64 root_objectid
, u64 owner_objectid
,
3232 u64 owner_offset
, int refs_to_drop
,
3233 struct btrfs_delayed_extent_op
*extent_op
)
3235 struct btrfs_key key
;
3236 struct btrfs_path
*path
;
3237 struct btrfs_fs_info
*info
= root
->fs_info
;
3238 struct btrfs_root
*extent_root
= info
->extent_root
;
3239 struct extent_buffer
*leaf
;
3240 struct btrfs_extent_item
*ei
;
3241 struct btrfs_extent_inline_ref
*iref
;
3244 int extent_slot
= 0;
3245 int found_extent
= 0;
3250 path
= btrfs_alloc_path();
3255 path
->leave_spinning
= 1;
3257 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
3258 BUG_ON(!is_data
&& refs_to_drop
!= 1);
3260 ret
= lookup_extent_backref(trans
, extent_root
, path
, &iref
,
3261 bytenr
, num_bytes
, parent
,
3262 root_objectid
, owner_objectid
,
3265 extent_slot
= path
->slots
[0];
3266 while (extent_slot
>= 0) {
3267 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
3269 if (key
.objectid
!= bytenr
)
3271 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
3272 key
.offset
== num_bytes
) {
3276 if (path
->slots
[0] - extent_slot
> 5)
3280 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3281 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
3282 if (found_extent
&& item_size
< sizeof(*ei
))
3285 if (!found_extent
) {
3287 ret
= remove_extent_backref(trans
, extent_root
, path
,
3291 btrfs_release_path(extent_root
, path
);
3292 path
->leave_spinning
= 1;
3294 key
.objectid
= bytenr
;
3295 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
3296 key
.offset
= num_bytes
;
3298 ret
= btrfs_search_slot(trans
, extent_root
,
3301 printk(KERN_ERR
"umm, got %d back from search"
3302 ", was looking for %llu\n", ret
,
3303 (unsigned long long)bytenr
);
3304 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
3307 extent_slot
= path
->slots
[0];
3310 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
3312 printk(KERN_ERR
"btrfs unable to find ref byte nr %llu "
3313 "parent %llu root %llu owner %llu offset %llu\n",
3314 (unsigned long long)bytenr
,
3315 (unsigned long long)parent
,
3316 (unsigned long long)root_objectid
,
3317 (unsigned long long)owner_objectid
,
3318 (unsigned long long)owner_offset
);
3321 leaf
= path
->nodes
[0];
3322 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
3323 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3324 if (item_size
< sizeof(*ei
)) {
3325 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
3326 ret
= convert_extent_item_v0(trans
, extent_root
, path
,
3330 btrfs_release_path(extent_root
, path
);
3331 path
->leave_spinning
= 1;
3333 key
.objectid
= bytenr
;
3334 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
3335 key
.offset
= num_bytes
;
3337 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
3340 printk(KERN_ERR
"umm, got %d back from search"
3341 ", was looking for %llu\n", ret
,
3342 (unsigned long long)bytenr
);
3343 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
3346 extent_slot
= path
->slots
[0];
3347 leaf
= path
->nodes
[0];
3348 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
3351 BUG_ON(item_size
< sizeof(*ei
));
3352 ei
= btrfs_item_ptr(leaf
, extent_slot
,
3353 struct btrfs_extent_item
);
3354 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
) {
3355 struct btrfs_tree_block_info
*bi
;
3356 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
3357 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
3358 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
3361 refs
= btrfs_extent_refs(leaf
, ei
);
3362 BUG_ON(refs
< refs_to_drop
);
3363 refs
-= refs_to_drop
;
3367 __run_delayed_extent_op(extent_op
, leaf
, ei
);
3369 * In the case of inline back ref, reference count will
3370 * be updated by remove_extent_backref
3373 BUG_ON(!found_extent
);
3375 btrfs_set_extent_refs(leaf
, ei
, refs
);
3376 btrfs_mark_buffer_dirty(leaf
);
3379 ret
= remove_extent_backref(trans
, extent_root
, path
,
3386 struct extent_buffer
*must_clean
= NULL
;
3389 BUG_ON(is_data
&& refs_to_drop
!=
3390 extent_data_ref_count(root
, path
, iref
));
3392 BUG_ON(path
->slots
[0] != extent_slot
);
3394 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
3395 path
->slots
[0] = extent_slot
;
3400 ret
= pin_down_bytes(trans
, root
, path
, bytenr
,
3401 num_bytes
, is_data
, &must_clean
);
3406 * it is going to be very rare for someone to be waiting
3407 * on the block we're freeing. del_items might need to
3408 * schedule, so rather than get fancy, just force it
3412 btrfs_set_lock_blocking(must_clean
);
3414 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
3417 btrfs_release_path(extent_root
, path
);
3420 clean_tree_block(NULL
, root
, must_clean
);
3421 btrfs_tree_unlock(must_clean
);
3422 free_extent_buffer(must_clean
);
3426 ret
= btrfs_del_csums(trans
, root
, bytenr
, num_bytes
);
3429 invalidate_mapping_pages(info
->btree_inode
->i_mapping
,
3430 bytenr
>> PAGE_CACHE_SHIFT
,
3431 (bytenr
+ num_bytes
- 1) >> PAGE_CACHE_SHIFT
);
3434 ret
= update_block_group(trans
, root
, bytenr
, num_bytes
, 0,
3438 btrfs_free_path(path
);
3443 * when we free an extent, it is possible (and likely) that we free the last
3444 * delayed ref for that extent as well. This searches the delayed ref tree for
3445 * a given extent, and if there are no other delayed refs to be processed, it
3446 * removes it from the tree.
3448 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
3449 struct btrfs_root
*root
, u64 bytenr
)
3451 struct btrfs_delayed_ref_head
*head
;
3452 struct btrfs_delayed_ref_root
*delayed_refs
;
3453 struct btrfs_delayed_ref_node
*ref
;
3454 struct rb_node
*node
;
3457 delayed_refs
= &trans
->transaction
->delayed_refs
;
3458 spin_lock(&delayed_refs
->lock
);
3459 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
3463 node
= rb_prev(&head
->node
.rb_node
);
3467 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
3469 /* there are still entries for this ref, we can't drop it */
3470 if (ref
->bytenr
== bytenr
)
3473 if (head
->extent_op
) {
3474 if (!head
->must_insert_reserved
)
3476 kfree(head
->extent_op
);
3477 head
->extent_op
= NULL
;
3481 * waiting for the lock here would deadlock. If someone else has it
3482 * locked they are already in the process of dropping it anyway
3484 if (!mutex_trylock(&head
->mutex
))
3488 * at this point we have a head with no other entries. Go
3489 * ahead and process it.
3491 head
->node
.in_tree
= 0;
3492 rb_erase(&head
->node
.rb_node
, &delayed_refs
->root
);
3494 delayed_refs
->num_entries
--;
3497 * we don't take a ref on the node because we're removing it from the
3498 * tree, so we just steal the ref the tree was holding.
3500 delayed_refs
->num_heads
--;
3501 if (list_empty(&head
->cluster
))
3502 delayed_refs
->num_heads_ready
--;
3504 list_del_init(&head
->cluster
);
3505 spin_unlock(&delayed_refs
->lock
);
3507 ret
= run_one_delayed_ref(trans
, root
->fs_info
->tree_root
,
3508 &head
->node
, head
->extent_op
,
3509 head
->must_insert_reserved
);
3511 btrfs_put_delayed_ref(&head
->node
);
3514 spin_unlock(&delayed_refs
->lock
);
3518 int btrfs_free_extent(struct btrfs_trans_handle
*trans
,
3519 struct btrfs_root
*root
,
3520 u64 bytenr
, u64 num_bytes
, u64 parent
,
3521 u64 root_objectid
, u64 owner
, u64 offset
)
3526 * tree log blocks never actually go into the extent allocation
3527 * tree, just update pinning info and exit early.
3529 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
3530 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
3531 /* unlocks the pinned mutex */
3532 btrfs_update_pinned_extents(root
, bytenr
, num_bytes
, 1);
3533 update_reserved_extents(root
, bytenr
, num_bytes
, 0);
3535 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
3536 ret
= btrfs_add_delayed_tree_ref(trans
, bytenr
, num_bytes
,
3537 parent
, root_objectid
, (int)owner
,
3538 BTRFS_DROP_DELAYED_REF
, NULL
);
3540 ret
= check_ref_cleanup(trans
, root
, bytenr
);
3543 ret
= btrfs_add_delayed_data_ref(trans
, bytenr
, num_bytes
,
3544 parent
, root_objectid
, owner
,
3545 offset
, BTRFS_DROP_DELAYED_REF
, NULL
);
3551 static u64
stripe_align(struct btrfs_root
*root
, u64 val
)
3553 u64 mask
= ((u64
)root
->stripesize
- 1);
3554 u64 ret
= (val
+ mask
) & ~mask
;
3559 * when we wait for progress in the block group caching, its because
3560 * our allocation attempt failed at least once. So, we must sleep
3561 * and let some progress happen before we try again.
3563 * This function will sleep at least once waiting for new free space to
3564 * show up, and then it will check the block group free space numbers
3565 * for our min num_bytes. Another option is to have it go ahead
3566 * and look in the rbtree for a free extent of a given size, but this
3570 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
3575 prepare_to_wait(&cache
->caching_q
, &wait
, TASK_UNINTERRUPTIBLE
);
3577 if (block_group_cache_done(cache
)) {
3578 finish_wait(&cache
->caching_q
, &wait
);
3582 finish_wait(&cache
->caching_q
, &wait
);
3584 wait_event(cache
->caching_q
, block_group_cache_done(cache
) ||
3585 (cache
->free_space
>= num_bytes
));
3589 enum btrfs_loop_type
{
3590 LOOP_CACHED_ONLY
= 0,
3591 LOOP_CACHING_NOWAIT
= 1,
3592 LOOP_CACHING_WAIT
= 2,
3593 LOOP_ALLOC_CHUNK
= 3,
3594 LOOP_NO_EMPTY_SIZE
= 4,
3598 * walks the btree of allocated extents and find a hole of a given size.
3599 * The key ins is changed to record the hole:
3600 * ins->objectid == block start
3601 * ins->flags = BTRFS_EXTENT_ITEM_KEY
3602 * ins->offset == number of blocks
3603 * Any available blocks before search_start are skipped.
3605 static noinline
int find_free_extent(struct btrfs_trans_handle
*trans
,
3606 struct btrfs_root
*orig_root
,
3607 u64 num_bytes
, u64 empty_size
,
3608 u64 search_start
, u64 search_end
,
3609 u64 hint_byte
, struct btrfs_key
*ins
,
3610 u64 exclude_start
, u64 exclude_nr
,
3614 struct btrfs_root
*root
= orig_root
->fs_info
->extent_root
;
3615 struct btrfs_free_cluster
*last_ptr
= NULL
;
3616 struct btrfs_block_group_cache
*block_group
= NULL
;
3617 int empty_cluster
= 2 * 1024 * 1024;
3618 int allowed_chunk_alloc
= 0;
3619 struct btrfs_space_info
*space_info
;
3620 int last_ptr_loop
= 0;
3622 bool found_uncached_bg
= false;
3624 WARN_ON(num_bytes
< root
->sectorsize
);
3625 btrfs_set_key_type(ins
, BTRFS_EXTENT_ITEM_KEY
);
3629 space_info
= __find_space_info(root
->fs_info
, data
);
3631 if (orig_root
->ref_cows
|| empty_size
)
3632 allowed_chunk_alloc
= 1;
3634 if (data
& BTRFS_BLOCK_GROUP_METADATA
) {
3635 last_ptr
= &root
->fs_info
->meta_alloc_cluster
;
3636 if (!btrfs_test_opt(root
, SSD
))
3637 empty_cluster
= 64 * 1024;
3640 if ((data
& BTRFS_BLOCK_GROUP_DATA
) && btrfs_test_opt(root
, SSD
)) {
3641 last_ptr
= &root
->fs_info
->data_alloc_cluster
;
3645 spin_lock(&last_ptr
->lock
);
3646 if (last_ptr
->block_group
)
3647 hint_byte
= last_ptr
->window_start
;
3648 spin_unlock(&last_ptr
->lock
);
3651 search_start
= max(search_start
, first_logical_byte(root
, 0));
3652 search_start
= max(search_start
, hint_byte
);
3657 if (search_start
== hint_byte
) {
3658 block_group
= btrfs_lookup_block_group(root
->fs_info
,
3661 * we don't want to use the block group if it doesn't match our
3662 * allocation bits, or if its not cached.
3664 if (block_group
&& block_group_bits(block_group
, data
) &&
3665 block_group_cache_done(block_group
)) {
3666 down_read(&space_info
->groups_sem
);
3667 if (list_empty(&block_group
->list
) ||
3670 * someone is removing this block group,
3671 * we can't jump into the have_block_group
3672 * target because our list pointers are not
3675 btrfs_put_block_group(block_group
);
3676 up_read(&space_info
->groups_sem
);
3678 goto have_block_group
;
3679 } else if (block_group
) {
3680 btrfs_put_block_group(block_group
);
3685 down_read(&space_info
->groups_sem
);
3686 list_for_each_entry(block_group
, &space_info
->block_groups
, list
) {
3690 atomic_inc(&block_group
->count
);
3691 search_start
= block_group
->key
.objectid
;
3694 if (unlikely(block_group
->cached
== BTRFS_CACHE_NO
)) {
3696 * we want to start caching kthreads, but not too many
3697 * right off the bat so we don't overwhelm the system,
3698 * so only start them if there are less than 2 and we're
3699 * in the initial allocation phase.
3701 if (loop
> LOOP_CACHING_NOWAIT
||
3702 atomic_read(&space_info
->caching_threads
) < 2) {
3703 ret
= cache_block_group(block_group
);
3708 cached
= block_group_cache_done(block_group
);
3709 if (unlikely(!cached
)) {
3710 found_uncached_bg
= true;
3712 /* if we only want cached bgs, loop */
3713 if (loop
== LOOP_CACHED_ONLY
)
3717 if (unlikely(block_group
->ro
))
3722 * the refill lock keeps out other
3723 * people trying to start a new cluster
3725 spin_lock(&last_ptr
->refill_lock
);
3726 if (last_ptr
->block_group
&&
3727 (last_ptr
->block_group
->ro
||
3728 !block_group_bits(last_ptr
->block_group
, data
))) {
3730 goto refill_cluster
;
3733 offset
= btrfs_alloc_from_cluster(block_group
, last_ptr
,
3734 num_bytes
, search_start
);
3736 /* we have a block, we're done */
3737 spin_unlock(&last_ptr
->refill_lock
);
3741 spin_lock(&last_ptr
->lock
);
3743 * whoops, this cluster doesn't actually point to
3744 * this block group. Get a ref on the block
3745 * group is does point to and try again
3747 if (!last_ptr_loop
&& last_ptr
->block_group
&&
3748 last_ptr
->block_group
!= block_group
) {
3750 btrfs_put_block_group(block_group
);
3751 block_group
= last_ptr
->block_group
;
3752 atomic_inc(&block_group
->count
);
3753 spin_unlock(&last_ptr
->lock
);
3754 spin_unlock(&last_ptr
->refill_lock
);
3757 search_start
= block_group
->key
.objectid
;
3759 * we know this block group is properly
3760 * in the list because
3761 * btrfs_remove_block_group, drops the
3762 * cluster before it removes the block
3763 * group from the list
3765 goto have_block_group
;
3767 spin_unlock(&last_ptr
->lock
);
3770 * this cluster didn't work out, free it and
3773 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
3777 /* allocate a cluster in this block group */
3778 ret
= btrfs_find_space_cluster(trans
, root
,
3779 block_group
, last_ptr
,
3781 empty_cluster
+ empty_size
);
3784 * now pull our allocation out of this
3787 offset
= btrfs_alloc_from_cluster(block_group
,
3788 last_ptr
, num_bytes
,
3791 /* we found one, proceed */
3792 spin_unlock(&last_ptr
->refill_lock
);
3795 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
) {
3796 spin_unlock(&last_ptr
->refill_lock
);
3798 wait_block_group_cache_progress(block_group
,
3799 num_bytes
+ empty_cluster
+ empty_size
);
3800 goto have_block_group
;
3804 * at this point we either didn't find a cluster
3805 * or we weren't able to allocate a block from our
3806 * cluster. Free the cluster we've been trying
3807 * to use, and go to the next block group
3809 if (loop
< LOOP_NO_EMPTY_SIZE
) {
3810 btrfs_return_cluster_to_free_space(NULL
,
3812 spin_unlock(&last_ptr
->refill_lock
);
3815 spin_unlock(&last_ptr
->refill_lock
);
3818 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
3819 num_bytes
, empty_size
);
3820 if (!offset
&& (cached
|| (!cached
&&
3821 loop
== LOOP_CACHING_NOWAIT
))) {
3823 } else if (!offset
&& (!cached
&&
3824 loop
> LOOP_CACHING_NOWAIT
)) {
3825 wait_block_group_cache_progress(block_group
,
3826 num_bytes
+ empty_size
);
3827 goto have_block_group
;
3830 search_start
= stripe_align(root
, offset
);
3831 /* move on to the next group */
3832 if (search_start
+ num_bytes
>= search_end
) {
3833 btrfs_add_free_space(block_group
, offset
, num_bytes
);
3837 /* move on to the next group */
3838 if (search_start
+ num_bytes
>
3839 block_group
->key
.objectid
+ block_group
->key
.offset
) {
3840 btrfs_add_free_space(block_group
, offset
, num_bytes
);
3844 if (exclude_nr
> 0 &&
3845 (search_start
+ num_bytes
> exclude_start
&&
3846 search_start
< exclude_start
+ exclude_nr
)) {
3847 search_start
= exclude_start
+ exclude_nr
;
3849 btrfs_add_free_space(block_group
, offset
, num_bytes
);
3851 * if search_start is still in this block group
3852 * then we just re-search this block group
3854 if (search_start
>= block_group
->key
.objectid
&&
3855 search_start
< (block_group
->key
.objectid
+
3856 block_group
->key
.offset
))
3857 goto have_block_group
;
3861 ins
->objectid
= search_start
;
3862 ins
->offset
= num_bytes
;
3864 if (offset
< search_start
)
3865 btrfs_add_free_space(block_group
, offset
,
3866 search_start
- offset
);
3867 BUG_ON(offset
> search_start
);
3869 /* we are all good, lets return */
3872 btrfs_put_block_group(block_group
);
3874 up_read(&space_info
->groups_sem
);
3876 /* LOOP_CACHED_ONLY, only search fully cached block groups
3877 * LOOP_CACHING_NOWAIT, search partially cached block groups, but
3878 * dont wait foR them to finish caching
3879 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
3880 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
3881 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
3884 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
&&
3885 (found_uncached_bg
|| empty_size
|| empty_cluster
||
3886 allowed_chunk_alloc
)) {
3887 if (found_uncached_bg
) {
3888 found_uncached_bg
= false;
3889 if (loop
< LOOP_CACHING_WAIT
) {
3895 if (loop
== LOOP_ALLOC_CHUNK
) {
3900 if (allowed_chunk_alloc
) {
3901 ret
= do_chunk_alloc(trans
, root
, num_bytes
+
3902 2 * 1024 * 1024, data
, 1);
3903 allowed_chunk_alloc
= 0;
3905 space_info
->force_alloc
= 1;
3908 if (loop
< LOOP_NO_EMPTY_SIZE
) {
3913 } else if (!ins
->objectid
) {
3917 /* we found what we needed */
3918 if (ins
->objectid
) {
3919 if (!(data
& BTRFS_BLOCK_GROUP_DATA
))
3920 trans
->block_group
= block_group
->key
.objectid
;
3922 btrfs_put_block_group(block_group
);
3929 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
)
3931 struct btrfs_block_group_cache
*cache
;
3933 printk(KERN_INFO
"space_info has %llu free, is %sfull\n",
3934 (unsigned long long)(info
->total_bytes
- info
->bytes_used
-
3935 info
->bytes_pinned
- info
->bytes_reserved
),
3936 (info
->full
) ? "" : "not ");
3937 printk(KERN_INFO
"space_info total=%llu, pinned=%llu, delalloc=%llu,"
3938 " may_use=%llu, used=%llu\n",
3939 (unsigned long long)info
->total_bytes
,
3940 (unsigned long long)info
->bytes_pinned
,
3941 (unsigned long long)info
->bytes_delalloc
,
3942 (unsigned long long)info
->bytes_may_use
,
3943 (unsigned long long)info
->bytes_used
);
3945 down_read(&info
->groups_sem
);
3946 list_for_each_entry(cache
, &info
->block_groups
, list
) {
3947 spin_lock(&cache
->lock
);
3948 printk(KERN_INFO
"block group %llu has %llu bytes, %llu used "
3949 "%llu pinned %llu reserved\n",
3950 (unsigned long long)cache
->key
.objectid
,
3951 (unsigned long long)cache
->key
.offset
,
3952 (unsigned long long)btrfs_block_group_used(&cache
->item
),
3953 (unsigned long long)cache
->pinned
,
3954 (unsigned long long)cache
->reserved
);
3955 btrfs_dump_free_space(cache
, bytes
);
3956 spin_unlock(&cache
->lock
);
3958 up_read(&info
->groups_sem
);
3961 static int __btrfs_reserve_extent(struct btrfs_trans_handle
*trans
,
3962 struct btrfs_root
*root
,
3963 u64 num_bytes
, u64 min_alloc_size
,
3964 u64 empty_size
, u64 hint_byte
,
3965 u64 search_end
, struct btrfs_key
*ins
,
3969 u64 search_start
= 0;
3970 struct btrfs_fs_info
*info
= root
->fs_info
;
3972 data
= btrfs_get_alloc_profile(root
, data
);
3975 * the only place that sets empty_size is btrfs_realloc_node, which
3976 * is not called recursively on allocations
3978 if (empty_size
|| root
->ref_cows
) {
3979 if (!(data
& BTRFS_BLOCK_GROUP_METADATA
)) {
3980 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
3982 BTRFS_BLOCK_GROUP_METADATA
|
3983 (info
->metadata_alloc_profile
&
3984 info
->avail_metadata_alloc_bits
), 0);
3986 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
3987 num_bytes
+ 2 * 1024 * 1024, data
, 0);
3990 WARN_ON(num_bytes
< root
->sectorsize
);
3991 ret
= find_free_extent(trans
, root
, num_bytes
, empty_size
,
3992 search_start
, search_end
, hint_byte
, ins
,
3993 trans
->alloc_exclude_start
,
3994 trans
->alloc_exclude_nr
, data
);
3996 if (ret
== -ENOSPC
&& num_bytes
> min_alloc_size
) {
3997 num_bytes
= num_bytes
>> 1;
3998 num_bytes
= num_bytes
& ~(root
->sectorsize
- 1);
3999 num_bytes
= max(num_bytes
, min_alloc_size
);
4000 do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
4001 num_bytes
, data
, 1);
4004 if (ret
== -ENOSPC
) {
4005 struct btrfs_space_info
*sinfo
;
4007 sinfo
= __find_space_info(root
->fs_info
, data
);
4008 printk(KERN_ERR
"btrfs allocation failed flags %llu, "
4009 "wanted %llu\n", (unsigned long long)data
,
4010 (unsigned long long)num_bytes
);
4011 dump_space_info(sinfo
, num_bytes
);
4017 int btrfs_free_reserved_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
4019 struct btrfs_block_group_cache
*cache
;
4022 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
4024 printk(KERN_ERR
"Unable to find block group for %llu\n",
4025 (unsigned long long)start
);
4029 ret
= btrfs_discard_extent(root
, start
, len
);
4031 btrfs_add_free_space(cache
, start
, len
);
4032 btrfs_put_block_group(cache
);
4033 update_reserved_extents(root
, start
, len
, 0);
4038 int btrfs_reserve_extent(struct btrfs_trans_handle
*trans
,
4039 struct btrfs_root
*root
,
4040 u64 num_bytes
, u64 min_alloc_size
,
4041 u64 empty_size
, u64 hint_byte
,
4042 u64 search_end
, struct btrfs_key
*ins
,
4046 ret
= __btrfs_reserve_extent(trans
, root
, num_bytes
, min_alloc_size
,
4047 empty_size
, hint_byte
, search_end
, ins
,
4050 update_reserved_extents(root
, ins
->objectid
, ins
->offset
, 1);
4055 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
4056 struct btrfs_root
*root
,
4057 u64 parent
, u64 root_objectid
,
4058 u64 flags
, u64 owner
, u64 offset
,
4059 struct btrfs_key
*ins
, int ref_mod
)
4062 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4063 struct btrfs_extent_item
*extent_item
;
4064 struct btrfs_extent_inline_ref
*iref
;
4065 struct btrfs_path
*path
;
4066 struct extent_buffer
*leaf
;
4071 type
= BTRFS_SHARED_DATA_REF_KEY
;
4073 type
= BTRFS_EXTENT_DATA_REF_KEY
;
4075 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
4077 path
= btrfs_alloc_path();
4080 path
->leave_spinning
= 1;
4081 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
4085 leaf
= path
->nodes
[0];
4086 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
4087 struct btrfs_extent_item
);
4088 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
4089 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
4090 btrfs_set_extent_flags(leaf
, extent_item
,
4091 flags
| BTRFS_EXTENT_FLAG_DATA
);
4093 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
4094 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
4096 struct btrfs_shared_data_ref
*ref
;
4097 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
4098 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
4099 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
4101 struct btrfs_extent_data_ref
*ref
;
4102 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
4103 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
4104 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
4105 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
4106 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
4109 btrfs_mark_buffer_dirty(path
->nodes
[0]);
4110 btrfs_free_path(path
);
4112 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
,
4115 printk(KERN_ERR
"btrfs update block group failed for %llu "
4116 "%llu\n", (unsigned long long)ins
->objectid
,
4117 (unsigned long long)ins
->offset
);
4123 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
4124 struct btrfs_root
*root
,
4125 u64 parent
, u64 root_objectid
,
4126 u64 flags
, struct btrfs_disk_key
*key
,
4127 int level
, struct btrfs_key
*ins
)
4130 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4131 struct btrfs_extent_item
*extent_item
;
4132 struct btrfs_tree_block_info
*block_info
;
4133 struct btrfs_extent_inline_ref
*iref
;
4134 struct btrfs_path
*path
;
4135 struct extent_buffer
*leaf
;
4136 u32 size
= sizeof(*extent_item
) + sizeof(*block_info
) + sizeof(*iref
);
4138 path
= btrfs_alloc_path();
4141 path
->leave_spinning
= 1;
4142 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
4146 leaf
= path
->nodes
[0];
4147 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
4148 struct btrfs_extent_item
);
4149 btrfs_set_extent_refs(leaf
, extent_item
, 1);
4150 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
4151 btrfs_set_extent_flags(leaf
, extent_item
,
4152 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
4153 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
4155 btrfs_set_tree_block_key(leaf
, block_info
, key
);
4156 btrfs_set_tree_block_level(leaf
, block_info
, level
);
4158 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
4160 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
4161 btrfs_set_extent_inline_ref_type(leaf
, iref
,
4162 BTRFS_SHARED_BLOCK_REF_KEY
);
4163 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
4165 btrfs_set_extent_inline_ref_type(leaf
, iref
,
4166 BTRFS_TREE_BLOCK_REF_KEY
);
4167 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
4170 btrfs_mark_buffer_dirty(leaf
);
4171 btrfs_free_path(path
);
4173 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
,
4176 printk(KERN_ERR
"btrfs update block group failed for %llu "
4177 "%llu\n", (unsigned long long)ins
->objectid
,
4178 (unsigned long long)ins
->offset
);
4184 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
4185 struct btrfs_root
*root
,
4186 u64 root_objectid
, u64 owner
,
4187 u64 offset
, struct btrfs_key
*ins
)
4191 BUG_ON(root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
4193 ret
= btrfs_add_delayed_data_ref(trans
, ins
->objectid
, ins
->offset
,
4194 0, root_objectid
, owner
, offset
,
4195 BTRFS_ADD_DELAYED_EXTENT
, NULL
);
4200 * this is used by the tree logging recovery code. It records that
4201 * an extent has been allocated and makes sure to clear the free
4202 * space cache bits as well
4204 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
4205 struct btrfs_root
*root
,
4206 u64 root_objectid
, u64 owner
, u64 offset
,
4207 struct btrfs_key
*ins
)
4210 struct btrfs_block_group_cache
*block_group
;
4212 block_group
= btrfs_lookup_block_group(root
->fs_info
, ins
->objectid
);
4213 cache_block_group(block_group
);
4214 wait_event(block_group
->caching_q
,
4215 block_group_cache_done(block_group
));
4217 ret
= btrfs_remove_free_space(block_group
, ins
->objectid
,
4220 btrfs_put_block_group(block_group
);
4221 ret
= alloc_reserved_file_extent(trans
, root
, 0, root_objectid
,
4222 0, owner
, offset
, ins
, 1);
4227 * finds a free extent and does all the dirty work required for allocation
4228 * returns the key for the extent through ins, and a tree buffer for
4229 * the first block of the extent through buf.
4231 * returns 0 if everything worked, non-zero otherwise.
4233 static int alloc_tree_block(struct btrfs_trans_handle
*trans
,
4234 struct btrfs_root
*root
,
4235 u64 num_bytes
, u64 parent
, u64 root_objectid
,
4236 struct btrfs_disk_key
*key
, int level
,
4237 u64 empty_size
, u64 hint_byte
, u64 search_end
,
4238 struct btrfs_key
*ins
)
4243 ret
= __btrfs_reserve_extent(trans
, root
, num_bytes
, num_bytes
,
4244 empty_size
, hint_byte
, search_end
,
4249 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
4251 parent
= ins
->objectid
;
4252 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
4256 update_reserved_extents(root
, ins
->objectid
, ins
->offset
, 1);
4257 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
4258 struct btrfs_delayed_extent_op
*extent_op
;
4259 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
4262 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
4264 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
4265 extent_op
->flags_to_set
= flags
;
4266 extent_op
->update_key
= 1;
4267 extent_op
->update_flags
= 1;
4268 extent_op
->is_data
= 0;
4270 ret
= btrfs_add_delayed_tree_ref(trans
, ins
->objectid
,
4271 ins
->offset
, parent
, root_objectid
,
4272 level
, BTRFS_ADD_DELAYED_EXTENT
,
4279 struct extent_buffer
*btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
,
4280 struct btrfs_root
*root
,
4281 u64 bytenr
, u32 blocksize
,
4284 struct extent_buffer
*buf
;
4286 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
4288 return ERR_PTR(-ENOMEM
);
4289 btrfs_set_header_generation(buf
, trans
->transid
);
4290 btrfs_set_buffer_lockdep_class(buf
, level
);
4291 btrfs_tree_lock(buf
);
4292 clean_tree_block(trans
, root
, buf
);
4294 btrfs_set_lock_blocking(buf
);
4295 btrfs_set_buffer_uptodate(buf
);
4297 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
4298 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
4299 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
4301 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
4302 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
4304 trans
->blocks_used
++;
4305 /* this returns a buffer locked for blocking */
4310 * helper function to allocate a block for a given tree
4311 * returns the tree buffer or NULL.
4313 struct extent_buffer
*btrfs_alloc_free_block(struct btrfs_trans_handle
*trans
,
4314 struct btrfs_root
*root
, u32 blocksize
,
4315 u64 parent
, u64 root_objectid
,
4316 struct btrfs_disk_key
*key
, int level
,
4317 u64 hint
, u64 empty_size
)
4319 struct btrfs_key ins
;
4321 struct extent_buffer
*buf
;
4323 ret
= alloc_tree_block(trans
, root
, blocksize
, parent
, root_objectid
,
4324 key
, level
, empty_size
, hint
, (u64
)-1, &ins
);
4327 return ERR_PTR(ret
);
4330 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
,
4336 int btrfs_drop_leaf_ref(struct btrfs_trans_handle
*trans
,
4337 struct btrfs_root
*root
, struct extent_buffer
*leaf
)
4341 struct btrfs_key key
;
4342 struct btrfs_file_extent_item
*fi
;
4347 BUG_ON(!btrfs_is_leaf(leaf
));
4348 nritems
= btrfs_header_nritems(leaf
);
4350 for (i
= 0; i
< nritems
; i
++) {
4352 btrfs_item_key_to_cpu(leaf
, &key
, i
);
4354 /* only extents have references, skip everything else */
4355 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
4358 fi
= btrfs_item_ptr(leaf
, i
, struct btrfs_file_extent_item
);
4360 /* inline extents live in the btree, they don't have refs */
4361 if (btrfs_file_extent_type(leaf
, fi
) ==
4362 BTRFS_FILE_EXTENT_INLINE
)
4365 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
4367 /* holes don't have refs */
4368 if (disk_bytenr
== 0)
4371 num_bytes
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
4372 ret
= btrfs_free_extent(trans
, root
, disk_bytenr
, num_bytes
,
4373 leaf
->start
, 0, key
.objectid
, 0);
4379 static noinline
int cache_drop_leaf_ref(struct btrfs_trans_handle
*trans
,
4380 struct btrfs_root
*root
,
4381 struct btrfs_leaf_ref
*ref
)
4385 struct btrfs_extent_info
*info
;
4386 struct refsort
*sorted
;
4388 if (ref
->nritems
== 0)
4391 sorted
= kmalloc(sizeof(*sorted
) * ref
->nritems
, GFP_NOFS
);
4392 for (i
= 0; i
< ref
->nritems
; i
++) {
4393 sorted
[i
].bytenr
= ref
->extents
[i
].bytenr
;
4396 sort(sorted
, ref
->nritems
, sizeof(struct refsort
), refsort_cmp
, NULL
);
4399 * the items in the ref were sorted when the ref was inserted
4400 * into the ref cache, so this is already in order
4402 for (i
= 0; i
< ref
->nritems
; i
++) {
4403 info
= ref
->extents
+ sorted
[i
].slot
;
4404 ret
= btrfs_free_extent(trans
, root
, info
->bytenr
,
4405 info
->num_bytes
, ref
->bytenr
,
4406 ref
->owner
, ref
->generation
,
4409 atomic_inc(&root
->fs_info
->throttle_gen
);
4410 wake_up(&root
->fs_info
->transaction_throttle
);
4422 static int drop_snap_lookup_refcount(struct btrfs_trans_handle
*trans
,
4423 struct btrfs_root
*root
, u64 start
,
4428 ret
= btrfs_lookup_extent_refs(trans
, root
, start
, len
, refs
);
4431 #if 0 /* some debugging code in case we see problems here */
4432 /* if the refs count is one, it won't get increased again. But
4433 * if the ref count is > 1, someone may be decreasing it at
4434 * the same time we are.
4437 struct extent_buffer
*eb
= NULL
;
4438 eb
= btrfs_find_create_tree_block(root
, start
, len
);
4440 btrfs_tree_lock(eb
);
4442 mutex_lock(&root
->fs_info
->alloc_mutex
);
4443 ret
= lookup_extent_ref(NULL
, root
, start
, len
, refs
);
4445 mutex_unlock(&root
->fs_info
->alloc_mutex
);
4448 btrfs_tree_unlock(eb
);
4449 free_extent_buffer(eb
);
4452 printk(KERN_ERR
"btrfs block %llu went down to one "
4453 "during drop_snap\n", (unsigned long long)start
);
4465 * this is used while deleting old snapshots, and it drops the refs
4466 * on a whole subtree starting from a level 1 node.
4468 * The idea is to sort all the leaf pointers, and then drop the
4469 * ref on all the leaves in order. Most of the time the leaves
4470 * will have ref cache entries, so no leaf IOs will be required to
4471 * find the extents they have references on.
4473 * For each leaf, any references it has are also dropped in order
4475 * This ends up dropping the references in something close to optimal
4476 * order for reading and modifying the extent allocation tree.
4478 static noinline
int drop_level_one_refs(struct btrfs_trans_handle
*trans
,
4479 struct btrfs_root
*root
,
4480 struct btrfs_path
*path
)
4485 struct extent_buffer
*eb
= path
->nodes
[1];
4486 struct extent_buffer
*leaf
;
4487 struct btrfs_leaf_ref
*ref
;
4488 struct refsort
*sorted
= NULL
;
4489 int nritems
= btrfs_header_nritems(eb
);
4493 int slot
= path
->slots
[1];
4494 u32 blocksize
= btrfs_level_size(root
, 0);
4500 root_owner
= btrfs_header_owner(eb
);
4501 root_gen
= btrfs_header_generation(eb
);
4502 sorted
= kmalloc(sizeof(*sorted
) * nritems
, GFP_NOFS
);
4505 * step one, sort all the leaf pointers so we don't scribble
4506 * randomly into the extent allocation tree
4508 for (i
= slot
; i
< nritems
; i
++) {
4509 sorted
[refi
].bytenr
= btrfs_node_blockptr(eb
, i
);
4510 sorted
[refi
].slot
= i
;
4515 * nritems won't be zero, but if we're picking up drop_snapshot
4516 * after a crash, slot might be > 0, so double check things
4522 sort(sorted
, refi
, sizeof(struct refsort
), refsort_cmp
, NULL
);
4525 * the first loop frees everything the leaves point to
4527 for (i
= 0; i
< refi
; i
++) {
4530 bytenr
= sorted
[i
].bytenr
;
4533 * check the reference count on this leaf. If it is > 1
4534 * we just decrement it below and don't update any
4535 * of the refs the leaf points to.
4537 ret
= drop_snap_lookup_refcount(trans
, root
, bytenr
,
4543 ptr_gen
= btrfs_node_ptr_generation(eb
, sorted
[i
].slot
);
4546 * the leaf only had one reference, which means the
4547 * only thing pointing to this leaf is the snapshot
4548 * we're deleting. It isn't possible for the reference
4549 * count to increase again later
4551 * The reference cache is checked for the leaf,
4552 * and if found we'll be able to drop any refs held by
4553 * the leaf without needing to read it in.
4555 ref
= btrfs_lookup_leaf_ref(root
, bytenr
);
4556 if (ref
&& ref
->generation
!= ptr_gen
) {
4557 btrfs_free_leaf_ref(root
, ref
);
4561 ret
= cache_drop_leaf_ref(trans
, root
, ref
);
4563 btrfs_remove_leaf_ref(root
, ref
);
4564 btrfs_free_leaf_ref(root
, ref
);
4567 * the leaf wasn't in the reference cache, so
4568 * we have to read it.
4570 leaf
= read_tree_block(root
, bytenr
, blocksize
,
4572 ret
= btrfs_drop_leaf_ref(trans
, root
, leaf
);
4574 free_extent_buffer(leaf
);
4576 atomic_inc(&root
->fs_info
->throttle_gen
);
4577 wake_up(&root
->fs_info
->transaction_throttle
);
4582 * run through the loop again to free the refs on the leaves.
4583 * This is faster than doing it in the loop above because
4584 * the leaves are likely to be clustered together. We end up
4585 * working in nice chunks on the extent allocation tree.
4587 for (i
= 0; i
< refi
; i
++) {
4588 bytenr
= sorted
[i
].bytenr
;
4589 ret
= btrfs_free_extent(trans
, root
, bytenr
,
4590 blocksize
, eb
->start
,
4591 root_owner
, root_gen
, 0, 1);
4594 atomic_inc(&root
->fs_info
->throttle_gen
);
4595 wake_up(&root
->fs_info
->transaction_throttle
);
4602 * update the path to show we've processed the entire level 1
4603 * node. This will get saved into the root's drop_snapshot_progress
4604 * field so these drops are not repeated again if this transaction
4607 path
->slots
[1] = nritems
;
4612 * helper function for drop_snapshot, this walks down the tree dropping ref
4613 * counts as it goes.
4615 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
4616 struct btrfs_root
*root
,
4617 struct btrfs_path
*path
, int *level
)
4623 struct extent_buffer
*next
;
4624 struct extent_buffer
*cur
;
4625 struct extent_buffer
*parent
;
4630 WARN_ON(*level
< 0);
4631 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
4632 ret
= drop_snap_lookup_refcount(trans
, root
, path
->nodes
[*level
]->start
,
4633 path
->nodes
[*level
]->len
, &refs
);
4639 * walk down to the last node level and free all the leaves
4641 while (*level
>= 0) {
4642 WARN_ON(*level
< 0);
4643 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
4644 cur
= path
->nodes
[*level
];
4646 if (btrfs_header_level(cur
) != *level
)
4649 if (path
->slots
[*level
] >=
4650 btrfs_header_nritems(cur
))
4653 /* the new code goes down to level 1 and does all the
4654 * leaves pointed to that node in bulk. So, this check
4655 * for level 0 will always be false.
4657 * But, the disk format allows the drop_snapshot_progress
4658 * field in the root to leave things in a state where
4659 * a leaf will need cleaning up here. If someone crashes
4660 * with the old code and then boots with the new code,
4661 * we might find a leaf here.
4664 ret
= btrfs_drop_leaf_ref(trans
, root
, cur
);
4670 * once we get to level one, process the whole node
4671 * at once, including everything below it.
4674 ret
= drop_level_one_refs(trans
, root
, path
);
4679 bytenr
= btrfs_node_blockptr(cur
, path
->slots
[*level
]);
4680 ptr_gen
= btrfs_node_ptr_generation(cur
, path
->slots
[*level
]);
4681 blocksize
= btrfs_level_size(root
, *level
- 1);
4683 ret
= drop_snap_lookup_refcount(trans
, root
, bytenr
,
4688 * if there is more than one reference, we don't need
4689 * to read that node to drop any references it has. We
4690 * just drop the ref we hold on that node and move on to the
4691 * next slot in this level.
4694 parent
= path
->nodes
[*level
];
4695 root_owner
= btrfs_header_owner(parent
);
4696 root_gen
= btrfs_header_generation(parent
);
4697 path
->slots
[*level
]++;
4699 ret
= btrfs_free_extent(trans
, root
, bytenr
,
4700 blocksize
, parent
->start
,
4701 root_owner
, root_gen
,
4705 atomic_inc(&root
->fs_info
->throttle_gen
);
4706 wake_up(&root
->fs_info
->transaction_throttle
);
4713 * we need to keep freeing things in the next level down.
4714 * read the block and loop around to process it
4716 next
= read_tree_block(root
, bytenr
, blocksize
, ptr_gen
);
4717 WARN_ON(*level
<= 0);
4718 if (path
->nodes
[*level
-1])
4719 free_extent_buffer(path
->nodes
[*level
-1]);
4720 path
->nodes
[*level
-1] = next
;
4721 *level
= btrfs_header_level(next
);
4722 path
->slots
[*level
] = 0;
4726 WARN_ON(*level
< 0);
4727 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
4729 if (path
->nodes
[*level
] == root
->node
) {
4730 parent
= path
->nodes
[*level
];
4731 bytenr
= path
->nodes
[*level
]->start
;
4733 parent
= path
->nodes
[*level
+ 1];
4734 bytenr
= btrfs_node_blockptr(parent
, path
->slots
[*level
+ 1]);
4737 blocksize
= btrfs_level_size(root
, *level
);
4738 root_owner
= btrfs_header_owner(parent
);
4739 root_gen
= btrfs_header_generation(parent
);
4742 * cleanup and free the reference on the last node
4745 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
,
4746 parent
->start
, root_owner
, root_gen
,
4748 free_extent_buffer(path
->nodes
[*level
]);
4749 path
->nodes
[*level
] = NULL
;
4759 struct walk_control
{
4760 u64 refs
[BTRFS_MAX_LEVEL
];
4761 u64 flags
[BTRFS_MAX_LEVEL
];
4762 struct btrfs_key update_progress
;
4770 #define DROP_REFERENCE 1
4771 #define UPDATE_BACKREF 2
4774 * hepler to process tree block while walking down the tree.
4776 * when wc->stage == DROP_REFERENCE, this function checks
4777 * reference count of the block. if the block is shared and
4778 * we need update back refs for the subtree rooted at the
4779 * block, this function changes wc->stage to UPDATE_BACKREF
4781 * when wc->stage == UPDATE_BACKREF, this function updates
4782 * back refs for pointers in the block.
4784 * NOTE: return value 1 means we should stop walking down.
4786 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
4787 struct btrfs_root
*root
,
4788 struct btrfs_path
*path
,
4789 struct walk_control
*wc
)
4791 int level
= wc
->level
;
4792 struct extent_buffer
*eb
= path
->nodes
[level
];
4793 struct btrfs_key key
;
4794 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
4797 if (wc
->stage
== UPDATE_BACKREF
&&
4798 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
4802 * when reference count of tree block is 1, it won't increase
4803 * again. once full backref flag is set, we never clear it.
4805 if ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
4806 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
))) {
4807 BUG_ON(!path
->locks
[level
]);
4808 ret
= btrfs_lookup_extent_info(trans
, root
,
4813 BUG_ON(wc
->refs
[level
] == 0);
4816 if (wc
->stage
== DROP_REFERENCE
&&
4817 wc
->update_ref
&& wc
->refs
[level
] > 1) {
4818 BUG_ON(eb
== root
->node
);
4819 BUG_ON(path
->slots
[level
] > 0);
4821 btrfs_item_key_to_cpu(eb
, &key
, path
->slots
[level
]);
4823 btrfs_node_key_to_cpu(eb
, &key
, path
->slots
[level
]);
4824 if (btrfs_header_owner(eb
) == root
->root_key
.objectid
&&
4825 btrfs_comp_cpu_keys(&key
, &wc
->update_progress
) >= 0) {
4826 wc
->stage
= UPDATE_BACKREF
;
4827 wc
->shared_level
= level
;
4831 if (wc
->stage
== DROP_REFERENCE
) {
4832 if (wc
->refs
[level
] > 1)
4835 if (path
->locks
[level
] && !wc
->keep_locks
) {
4836 btrfs_tree_unlock(eb
);
4837 path
->locks
[level
] = 0;
4842 /* wc->stage == UPDATE_BACKREF */
4843 if (!(wc
->flags
[level
] & flag
)) {
4844 BUG_ON(!path
->locks
[level
]);
4845 ret
= btrfs_inc_ref(trans
, root
, eb
, 1);
4847 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
4849 ret
= btrfs_set_disk_extent_flags(trans
, root
, eb
->start
,
4852 wc
->flags
[level
] |= flag
;
4856 * the block is shared by multiple trees, so it's not good to
4857 * keep the tree lock
4859 if (path
->locks
[level
] && level
> 0) {
4860 btrfs_tree_unlock(eb
);
4861 path
->locks
[level
] = 0;
4867 * hepler to process tree block while walking up the tree.
4869 * when wc->stage == DROP_REFERENCE, this function drops
4870 * reference count on the block.
4872 * when wc->stage == UPDATE_BACKREF, this function changes
4873 * wc->stage back to DROP_REFERENCE if we changed wc->stage
4874 * to UPDATE_BACKREF previously while processing the block.
4876 * NOTE: return value 1 means we should stop walking up.
4878 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
4879 struct btrfs_root
*root
,
4880 struct btrfs_path
*path
,
4881 struct walk_control
*wc
)
4884 int level
= wc
->level
;
4885 struct extent_buffer
*eb
= path
->nodes
[level
];
4888 if (wc
->stage
== UPDATE_BACKREF
) {
4889 BUG_ON(wc
->shared_level
< level
);
4890 if (level
< wc
->shared_level
)
4893 BUG_ON(wc
->refs
[level
] <= 1);
4894 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
4898 wc
->stage
= DROP_REFERENCE
;
4899 wc
->shared_level
= -1;
4900 path
->slots
[level
] = 0;
4903 * check reference count again if the block isn't locked.
4904 * we should start walking down the tree again if reference
4907 if (!path
->locks
[level
]) {
4909 btrfs_tree_lock(eb
);
4910 btrfs_set_lock_blocking(eb
);
4911 path
->locks
[level
] = 1;
4913 ret
= btrfs_lookup_extent_info(trans
, root
,
4918 BUG_ON(wc
->refs
[level
] == 0);
4919 if (wc
->refs
[level
] == 1) {
4920 btrfs_tree_unlock(eb
);
4921 path
->locks
[level
] = 0;
4929 /* wc->stage == DROP_REFERENCE */
4930 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
4932 if (wc
->refs
[level
] == 1) {
4934 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
4935 ret
= btrfs_dec_ref(trans
, root
, eb
, 1);
4937 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
4940 /* make block locked assertion in clean_tree_block happy */
4941 if (!path
->locks
[level
] &&
4942 btrfs_header_generation(eb
) == trans
->transid
) {
4943 btrfs_tree_lock(eb
);
4944 btrfs_set_lock_blocking(eb
);
4945 path
->locks
[level
] = 1;
4947 clean_tree_block(trans
, root
, eb
);
4950 if (eb
== root
->node
) {
4951 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
4954 BUG_ON(root
->root_key
.objectid
!=
4955 btrfs_header_owner(eb
));
4957 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
4958 parent
= path
->nodes
[level
+ 1]->start
;
4960 BUG_ON(root
->root_key
.objectid
!=
4961 btrfs_header_owner(path
->nodes
[level
+ 1]));
4964 ret
= btrfs_free_extent(trans
, root
, eb
->start
, eb
->len
, parent
,
4965 root
->root_key
.objectid
, level
, 0);
4968 wc
->refs
[level
] = 0;
4969 wc
->flags
[level
] = 0;
4973 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
4974 struct btrfs_root
*root
,
4975 struct btrfs_path
*path
,
4976 struct walk_control
*wc
)
4978 struct extent_buffer
*next
;
4979 struct extent_buffer
*cur
;
4983 int level
= wc
->level
;
4986 while (level
>= 0) {
4987 cur
= path
->nodes
[level
];
4988 BUG_ON(path
->slots
[level
] >= btrfs_header_nritems(cur
));
4990 ret
= walk_down_proc(trans
, root
, path
, wc
);
4997 bytenr
= btrfs_node_blockptr(cur
, path
->slots
[level
]);
4998 blocksize
= btrfs_level_size(root
, level
- 1);
4999 ptr_gen
= btrfs_node_ptr_generation(cur
, path
->slots
[level
]);
5001 next
= read_tree_block(root
, bytenr
, blocksize
, ptr_gen
);
5002 btrfs_tree_lock(next
);
5003 btrfs_set_lock_blocking(next
);
5006 BUG_ON(level
!= btrfs_header_level(next
));
5007 path
->nodes
[level
] = next
;
5008 path
->slots
[level
] = 0;
5009 path
->locks
[level
] = 1;
5015 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
5016 struct btrfs_root
*root
,
5017 struct btrfs_path
*path
,
5018 struct walk_control
*wc
, int max_level
)
5020 int level
= wc
->level
;
5023 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
5024 while (level
< max_level
&& path
->nodes
[level
]) {
5026 if (path
->slots
[level
] + 1 <
5027 btrfs_header_nritems(path
->nodes
[level
])) {
5028 path
->slots
[level
]++;
5031 ret
= walk_up_proc(trans
, root
, path
, wc
);
5035 if (path
->locks
[level
]) {
5036 btrfs_tree_unlock(path
->nodes
[level
]);
5037 path
->locks
[level
] = 0;
5039 free_extent_buffer(path
->nodes
[level
]);
5040 path
->nodes
[level
] = NULL
;
5048 * drop a subvolume tree.
5050 * this function traverses the tree freeing any blocks that only
5051 * referenced by the tree.
5053 * when a shared tree block is found. this function decreases its
5054 * reference count by one. if update_ref is true, this function
5055 * also make sure backrefs for the shared block and all lower level
5056 * blocks are properly updated.
5058 int btrfs_drop_snapshot(struct btrfs_root
*root
, int update_ref
)
5060 struct btrfs_path
*path
;
5061 struct btrfs_trans_handle
*trans
;
5062 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
5063 struct btrfs_root_item
*root_item
= &root
->root_item
;
5064 struct walk_control
*wc
;
5065 struct btrfs_key key
;
5070 path
= btrfs_alloc_path();
5073 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
5076 trans
= btrfs_start_transaction(tree_root
, 1);
5078 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
5079 level
= btrfs_header_level(root
->node
);
5080 path
->nodes
[level
] = btrfs_lock_root_node(root
);
5081 btrfs_set_lock_blocking(path
->nodes
[level
]);
5082 path
->slots
[level
] = 0;
5083 path
->locks
[level
] = 1;
5084 memset(&wc
->update_progress
, 0,
5085 sizeof(wc
->update_progress
));
5087 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
5088 memcpy(&wc
->update_progress
, &key
,
5089 sizeof(wc
->update_progress
));
5091 level
= root_item
->drop_level
;
5093 path
->lowest_level
= level
;
5094 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5095 path
->lowest_level
= 0;
5100 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
5101 path
->slots
[level
]);
5102 WARN_ON(memcmp(&key
, &wc
->update_progress
, sizeof(key
)));
5105 * unlock our path, this is safe because only this
5106 * function is allowed to delete this snapshot
5108 btrfs_unlock_up_safe(path
, 0);
5110 level
= btrfs_header_level(root
->node
);
5112 btrfs_tree_lock(path
->nodes
[level
]);
5113 btrfs_set_lock_blocking(path
->nodes
[level
]);
5115 ret
= btrfs_lookup_extent_info(trans
, root
,
5116 path
->nodes
[level
]->start
,
5117 path
->nodes
[level
]->len
,
5121 BUG_ON(wc
->refs
[level
] == 0);
5123 if (level
== root_item
->drop_level
)
5126 btrfs_tree_unlock(path
->nodes
[level
]);
5127 WARN_ON(wc
->refs
[level
] != 1);
5133 wc
->shared_level
= -1;
5134 wc
->stage
= DROP_REFERENCE
;
5135 wc
->update_ref
= update_ref
;
5139 ret
= walk_down_tree(trans
, root
, path
, wc
);
5145 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
5152 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
5156 if (wc
->stage
== DROP_REFERENCE
) {
5158 btrfs_node_key(path
->nodes
[level
],
5159 &root_item
->drop_progress
,
5160 path
->slots
[level
]);
5161 root_item
->drop_level
= level
;
5164 BUG_ON(wc
->level
== 0);
5165 if (trans
->transaction
->in_commit
||
5166 trans
->transaction
->delayed_refs
.flushing
) {
5167 ret
= btrfs_update_root(trans
, tree_root
,
5172 btrfs_end_transaction(trans
, tree_root
);
5173 trans
= btrfs_start_transaction(tree_root
, 1);
5175 unsigned long update
;
5176 update
= trans
->delayed_ref_updates
;
5177 trans
->delayed_ref_updates
= 0;
5179 btrfs_run_delayed_refs(trans
, tree_root
,
5183 btrfs_release_path(root
, path
);
5186 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
5189 free_extent_buffer(root
->node
);
5190 free_extent_buffer(root
->commit_root
);
5193 btrfs_end_transaction(trans
, tree_root
);
5195 btrfs_free_path(path
);
5200 * drop subtree rooted at tree block 'node'.
5202 * NOTE: this function will unlock and release tree block 'node'
5204 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
5205 struct btrfs_root
*root
,
5206 struct extent_buffer
*node
,
5207 struct extent_buffer
*parent
)
5209 struct btrfs_path
*path
;
5210 struct walk_control
*wc
;
5216 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
5218 path
= btrfs_alloc_path();
5221 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
5224 btrfs_assert_tree_locked(parent
);
5225 parent_level
= btrfs_header_level(parent
);
5226 extent_buffer_get(parent
);
5227 path
->nodes
[parent_level
] = parent
;
5228 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
5230 btrfs_assert_tree_locked(node
);
5231 level
= btrfs_header_level(node
);
5232 path
->nodes
[level
] = node
;
5233 path
->slots
[level
] = 0;
5234 path
->locks
[level
] = 1;
5236 wc
->refs
[parent_level
] = 1;
5237 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
5239 wc
->shared_level
= -1;
5240 wc
->stage
= DROP_REFERENCE
;
5245 wret
= walk_down_tree(trans
, root
, path
, wc
);
5251 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
5259 btrfs_free_path(path
);
5264 static unsigned long calc_ra(unsigned long start
, unsigned long last
,
5267 return min(last
, start
+ nr
- 1);
5270 static noinline
int relocate_inode_pages(struct inode
*inode
, u64 start
,
5275 unsigned long first_index
;
5276 unsigned long last_index
;
5279 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
5280 struct file_ra_state
*ra
;
5281 struct btrfs_ordered_extent
*ordered
;
5282 unsigned int total_read
= 0;
5283 unsigned int total_dirty
= 0;
5286 ra
= kzalloc(sizeof(*ra
), GFP_NOFS
);
5288 mutex_lock(&inode
->i_mutex
);
5289 first_index
= start
>> PAGE_CACHE_SHIFT
;
5290 last_index
= (start
+ len
- 1) >> PAGE_CACHE_SHIFT
;
5292 /* make sure the dirty trick played by the caller work */
5293 ret
= invalidate_inode_pages2_range(inode
->i_mapping
,
5294 first_index
, last_index
);
5298 file_ra_state_init(ra
, inode
->i_mapping
);
5300 for (i
= first_index
; i
<= last_index
; i
++) {
5301 if (total_read
% ra
->ra_pages
== 0) {
5302 btrfs_force_ra(inode
->i_mapping
, ra
, NULL
, i
,
5303 calc_ra(i
, last_index
, ra
->ra_pages
));
5307 if (((u64
)i
<< PAGE_CACHE_SHIFT
) > i_size_read(inode
))
5309 page
= grab_cache_page(inode
->i_mapping
, i
);
5314 if (!PageUptodate(page
)) {
5315 btrfs_readpage(NULL
, page
);
5317 if (!PageUptodate(page
)) {
5319 page_cache_release(page
);
5324 wait_on_page_writeback(page
);
5326 page_start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
5327 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
5328 lock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
5330 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
5332 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
5334 page_cache_release(page
);
5335 btrfs_start_ordered_extent(inode
, ordered
, 1);
5336 btrfs_put_ordered_extent(ordered
);
5339 set_page_extent_mapped(page
);
5341 if (i
== first_index
)
5342 set_extent_bits(io_tree
, page_start
, page_end
,
5343 EXTENT_BOUNDARY
, GFP_NOFS
);
5344 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
5346 set_page_dirty(page
);
5349 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
5351 page_cache_release(page
);
5356 mutex_unlock(&inode
->i_mutex
);
5357 balance_dirty_pages_ratelimited_nr(inode
->i_mapping
, total_dirty
);
5361 static noinline
int relocate_data_extent(struct inode
*reloc_inode
,
5362 struct btrfs_key
*extent_key
,
5365 struct btrfs_root
*root
= BTRFS_I(reloc_inode
)->root
;
5366 struct extent_map_tree
*em_tree
= &BTRFS_I(reloc_inode
)->extent_tree
;
5367 struct extent_map
*em
;
5368 u64 start
= extent_key
->objectid
- offset
;
5369 u64 end
= start
+ extent_key
->offset
- 1;
5371 em
= alloc_extent_map(GFP_NOFS
);
5372 BUG_ON(!em
|| IS_ERR(em
));
5375 em
->len
= extent_key
->offset
;
5376 em
->block_len
= extent_key
->offset
;
5377 em
->block_start
= extent_key
->objectid
;
5378 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
5379 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
5381 /* setup extent map to cheat btrfs_readpage */
5382 lock_extent(&BTRFS_I(reloc_inode
)->io_tree
, start
, end
, GFP_NOFS
);
5385 spin_lock(&em_tree
->lock
);
5386 ret
= add_extent_mapping(em_tree
, em
);
5387 spin_unlock(&em_tree
->lock
);
5388 if (ret
!= -EEXIST
) {
5389 free_extent_map(em
);
5392 btrfs_drop_extent_cache(reloc_inode
, start
, end
, 0);
5394 unlock_extent(&BTRFS_I(reloc_inode
)->io_tree
, start
, end
, GFP_NOFS
);
5396 return relocate_inode_pages(reloc_inode
, start
, extent_key
->offset
);
5399 struct btrfs_ref_path
{
5401 u64 nodes
[BTRFS_MAX_LEVEL
];
5403 u64 root_generation
;
5410 struct btrfs_key node_keys
[BTRFS_MAX_LEVEL
];
5411 u64 new_nodes
[BTRFS_MAX_LEVEL
];
5414 struct disk_extent
{
5425 static int is_cowonly_root(u64 root_objectid
)
5427 if (root_objectid
== BTRFS_ROOT_TREE_OBJECTID
||
5428 root_objectid
== BTRFS_EXTENT_TREE_OBJECTID
||
5429 root_objectid
== BTRFS_CHUNK_TREE_OBJECTID
||
5430 root_objectid
== BTRFS_DEV_TREE_OBJECTID
||
5431 root_objectid
== BTRFS_TREE_LOG_OBJECTID
||
5432 root_objectid
== BTRFS_CSUM_TREE_OBJECTID
)
5437 static noinline
int __next_ref_path(struct btrfs_trans_handle
*trans
,
5438 struct btrfs_root
*extent_root
,
5439 struct btrfs_ref_path
*ref_path
,
5442 struct extent_buffer
*leaf
;
5443 struct btrfs_path
*path
;
5444 struct btrfs_extent_ref
*ref
;
5445 struct btrfs_key key
;
5446 struct btrfs_key found_key
;
5452 path
= btrfs_alloc_path();
5457 ref_path
->lowest_level
= -1;
5458 ref_path
->current_level
= -1;
5459 ref_path
->shared_level
= -1;
5463 level
= ref_path
->current_level
- 1;
5464 while (level
>= -1) {
5466 if (level
< ref_path
->lowest_level
)
5470 bytenr
= ref_path
->nodes
[level
];
5472 bytenr
= ref_path
->extent_start
;
5473 BUG_ON(bytenr
== 0);
5475 parent
= ref_path
->nodes
[level
+ 1];
5476 ref_path
->nodes
[level
+ 1] = 0;
5477 ref_path
->current_level
= level
;
5478 BUG_ON(parent
== 0);
5480 key
.objectid
= bytenr
;
5481 key
.offset
= parent
+ 1;
5482 key
.type
= BTRFS_EXTENT_REF_KEY
;
5484 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
, 0, 0);
5489 leaf
= path
->nodes
[0];
5490 nritems
= btrfs_header_nritems(leaf
);
5491 if (path
->slots
[0] >= nritems
) {
5492 ret
= btrfs_next_leaf(extent_root
, path
);
5497 leaf
= path
->nodes
[0];
5500 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5501 if (found_key
.objectid
== bytenr
&&
5502 found_key
.type
== BTRFS_EXTENT_REF_KEY
) {
5503 if (level
< ref_path
->shared_level
)
5504 ref_path
->shared_level
= level
;
5509 btrfs_release_path(extent_root
, path
);
5512 /* reached lowest level */
5516 level
= ref_path
->current_level
;
5517 while (level
< BTRFS_MAX_LEVEL
- 1) {
5521 bytenr
= ref_path
->nodes
[level
];
5523 bytenr
= ref_path
->extent_start
;
5525 BUG_ON(bytenr
== 0);
5527 key
.objectid
= bytenr
;
5529 key
.type
= BTRFS_EXTENT_REF_KEY
;
5531 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
, 0, 0);
5535 leaf
= path
->nodes
[0];
5536 nritems
= btrfs_header_nritems(leaf
);
5537 if (path
->slots
[0] >= nritems
) {
5538 ret
= btrfs_next_leaf(extent_root
, path
);
5542 /* the extent was freed by someone */
5543 if (ref_path
->lowest_level
== level
)
5545 btrfs_release_path(extent_root
, path
);
5548 leaf
= path
->nodes
[0];
5551 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5552 if (found_key
.objectid
!= bytenr
||
5553 found_key
.type
!= BTRFS_EXTENT_REF_KEY
) {
5554 /* the extent was freed by someone */
5555 if (ref_path
->lowest_level
== level
) {
5559 btrfs_release_path(extent_root
, path
);
5563 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
5564 struct btrfs_extent_ref
);
5565 ref_objectid
= btrfs_ref_objectid(leaf
, ref
);
5566 if (ref_objectid
< BTRFS_FIRST_FREE_OBJECTID
) {
5568 level
= (int)ref_objectid
;
5569 BUG_ON(level
>= BTRFS_MAX_LEVEL
);
5570 ref_path
->lowest_level
= level
;
5571 ref_path
->current_level
= level
;
5572 ref_path
->nodes
[level
] = bytenr
;
5574 WARN_ON(ref_objectid
!= level
);
5577 WARN_ON(level
!= -1);
5581 if (ref_path
->lowest_level
== level
) {
5582 ref_path
->owner_objectid
= ref_objectid
;
5583 ref_path
->num_refs
= btrfs_ref_num_refs(leaf
, ref
);
5587 * the block is tree root or the block isn't in reference
5590 if (found_key
.objectid
== found_key
.offset
||
5591 is_cowonly_root(btrfs_ref_root(leaf
, ref
))) {
5592 ref_path
->root_objectid
= btrfs_ref_root(leaf
, ref
);
5593 ref_path
->root_generation
=
5594 btrfs_ref_generation(leaf
, ref
);
5596 /* special reference from the tree log */
5597 ref_path
->nodes
[0] = found_key
.offset
;
5598 ref_path
->current_level
= 0;
5605 BUG_ON(ref_path
->nodes
[level
] != 0);
5606 ref_path
->nodes
[level
] = found_key
.offset
;
5607 ref_path
->current_level
= level
;
5610 * the reference was created in the running transaction,
5611 * no need to continue walking up.
5613 if (btrfs_ref_generation(leaf
, ref
) == trans
->transid
) {
5614 ref_path
->root_objectid
= btrfs_ref_root(leaf
, ref
);
5615 ref_path
->root_generation
=
5616 btrfs_ref_generation(leaf
, ref
);
5621 btrfs_release_path(extent_root
, path
);
5624 /* reached max tree level, but no tree root found. */
5627 btrfs_free_path(path
);
5631 static int btrfs_first_ref_path(struct btrfs_trans_handle
*trans
,
5632 struct btrfs_root
*extent_root
,
5633 struct btrfs_ref_path
*ref_path
,
5636 memset(ref_path
, 0, sizeof(*ref_path
));
5637 ref_path
->extent_start
= extent_start
;
5639 return __next_ref_path(trans
, extent_root
, ref_path
, 1);
5642 static int btrfs_next_ref_path(struct btrfs_trans_handle
*trans
,
5643 struct btrfs_root
*extent_root
,
5644 struct btrfs_ref_path
*ref_path
)
5646 return __next_ref_path(trans
, extent_root
, ref_path
, 0);
5649 static noinline
int get_new_locations(struct inode
*reloc_inode
,
5650 struct btrfs_key
*extent_key
,
5651 u64 offset
, int no_fragment
,
5652 struct disk_extent
**extents
,
5655 struct btrfs_root
*root
= BTRFS_I(reloc_inode
)->root
;
5656 struct btrfs_path
*path
;
5657 struct btrfs_file_extent_item
*fi
;
5658 struct extent_buffer
*leaf
;
5659 struct disk_extent
*exts
= *extents
;
5660 struct btrfs_key found_key
;
5665 int max
= *nr_extents
;
5668 WARN_ON(!no_fragment
&& *extents
);
5671 exts
= kmalloc(sizeof(*exts
) * max
, GFP_NOFS
);
5676 path
= btrfs_alloc_path();
5679 cur_pos
= extent_key
->objectid
- offset
;
5680 last_byte
= extent_key
->objectid
+ extent_key
->offset
;
5681 ret
= btrfs_lookup_file_extent(NULL
, root
, path
, reloc_inode
->i_ino
,
5691 leaf
= path
->nodes
[0];
5692 nritems
= btrfs_header_nritems(leaf
);
5693 if (path
->slots
[0] >= nritems
) {
5694 ret
= btrfs_next_leaf(root
, path
);
5699 leaf
= path
->nodes
[0];
5702 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5703 if (found_key
.offset
!= cur_pos
||
5704 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
||
5705 found_key
.objectid
!= reloc_inode
->i_ino
)
5708 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
5709 struct btrfs_file_extent_item
);
5710 if (btrfs_file_extent_type(leaf
, fi
) !=
5711 BTRFS_FILE_EXTENT_REG
||
5712 btrfs_file_extent_disk_bytenr(leaf
, fi
) == 0)
5716 struct disk_extent
*old
= exts
;
5718 exts
= kzalloc(sizeof(*exts
) * max
, GFP_NOFS
);
5719 memcpy(exts
, old
, sizeof(*exts
) * nr
);
5720 if (old
!= *extents
)
5724 exts
[nr
].disk_bytenr
=
5725 btrfs_file_extent_disk_bytenr(leaf
, fi
);
5726 exts
[nr
].disk_num_bytes
=
5727 btrfs_file_extent_disk_num_bytes(leaf
, fi
);
5728 exts
[nr
].offset
= btrfs_file_extent_offset(leaf
, fi
);
5729 exts
[nr
].num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
5730 exts
[nr
].ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
5731 exts
[nr
].compression
= btrfs_file_extent_compression(leaf
, fi
);
5732 exts
[nr
].encryption
= btrfs_file_extent_encryption(leaf
, fi
);
5733 exts
[nr
].other_encoding
= btrfs_file_extent_other_encoding(leaf
,
5735 BUG_ON(exts
[nr
].offset
> 0);
5736 BUG_ON(exts
[nr
].compression
|| exts
[nr
].encryption
);
5737 BUG_ON(exts
[nr
].num_bytes
!= exts
[nr
].disk_num_bytes
);
5739 cur_pos
+= exts
[nr
].num_bytes
;
5742 if (cur_pos
+ offset
>= last_byte
)
5752 BUG_ON(cur_pos
+ offset
> last_byte
);
5753 if (cur_pos
+ offset
< last_byte
) {
5759 btrfs_free_path(path
);
5761 if (exts
!= *extents
)
5770 static noinline
int replace_one_extent(struct btrfs_trans_handle
*trans
,
5771 struct btrfs_root
*root
,
5772 struct btrfs_path
*path
,
5773 struct btrfs_key
*extent_key
,
5774 struct btrfs_key
*leaf_key
,
5775 struct btrfs_ref_path
*ref_path
,
5776 struct disk_extent
*new_extents
,
5779 struct extent_buffer
*leaf
;
5780 struct btrfs_file_extent_item
*fi
;
5781 struct inode
*inode
= NULL
;
5782 struct btrfs_key key
;
5787 u64 search_end
= (u64
)-1;
5790 int extent_locked
= 0;
5794 memcpy(&key
, leaf_key
, sizeof(key
));
5795 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
) {
5796 if (key
.objectid
< ref_path
->owner_objectid
||
5797 (key
.objectid
== ref_path
->owner_objectid
&&
5798 key
.type
< BTRFS_EXTENT_DATA_KEY
)) {
5799 key
.objectid
= ref_path
->owner_objectid
;
5800 key
.type
= BTRFS_EXTENT_DATA_KEY
;
5806 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
5810 leaf
= path
->nodes
[0];
5811 nritems
= btrfs_header_nritems(leaf
);
5813 if (extent_locked
&& ret
> 0) {
5815 * the file extent item was modified by someone
5816 * before the extent got locked.
5818 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
5819 lock_end
, GFP_NOFS
);
5823 if (path
->slots
[0] >= nritems
) {
5824 if (++nr_scaned
> 2)
5827 BUG_ON(extent_locked
);
5828 ret
= btrfs_next_leaf(root
, path
);
5833 leaf
= path
->nodes
[0];
5834 nritems
= btrfs_header_nritems(leaf
);
5837 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
5839 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
) {
5840 if ((key
.objectid
> ref_path
->owner_objectid
) ||
5841 (key
.objectid
== ref_path
->owner_objectid
&&
5842 key
.type
> BTRFS_EXTENT_DATA_KEY
) ||
5843 key
.offset
>= search_end
)
5847 if (inode
&& key
.objectid
!= inode
->i_ino
) {
5848 BUG_ON(extent_locked
);
5849 btrfs_release_path(root
, path
);
5850 mutex_unlock(&inode
->i_mutex
);
5856 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
5861 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
5862 struct btrfs_file_extent_item
);
5863 extent_type
= btrfs_file_extent_type(leaf
, fi
);
5864 if ((extent_type
!= BTRFS_FILE_EXTENT_REG
&&
5865 extent_type
!= BTRFS_FILE_EXTENT_PREALLOC
) ||
5866 (btrfs_file_extent_disk_bytenr(leaf
, fi
) !=
5867 extent_key
->objectid
)) {
5873 num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
5874 ext_offset
= btrfs_file_extent_offset(leaf
, fi
);
5876 if (search_end
== (u64
)-1) {
5877 search_end
= key
.offset
- ext_offset
+
5878 btrfs_file_extent_ram_bytes(leaf
, fi
);
5881 if (!extent_locked
) {
5882 lock_start
= key
.offset
;
5883 lock_end
= lock_start
+ num_bytes
- 1;
5885 if (lock_start
> key
.offset
||
5886 lock_end
+ 1 < key
.offset
+ num_bytes
) {
5887 unlock_extent(&BTRFS_I(inode
)->io_tree
,
5888 lock_start
, lock_end
, GFP_NOFS
);
5894 btrfs_release_path(root
, path
);
5896 inode
= btrfs_iget_locked(root
->fs_info
->sb
,
5897 key
.objectid
, root
);
5898 if (inode
->i_state
& I_NEW
) {
5899 BTRFS_I(inode
)->root
= root
;
5900 BTRFS_I(inode
)->location
.objectid
=
5902 BTRFS_I(inode
)->location
.type
=
5903 BTRFS_INODE_ITEM_KEY
;
5904 BTRFS_I(inode
)->location
.offset
= 0;
5905 btrfs_read_locked_inode(inode
);
5906 unlock_new_inode(inode
);
5909 * some code call btrfs_commit_transaction while
5910 * holding the i_mutex, so we can't use mutex_lock
5913 if (is_bad_inode(inode
) ||
5914 !mutex_trylock(&inode
->i_mutex
)) {
5917 key
.offset
= (u64
)-1;
5922 if (!extent_locked
) {
5923 struct btrfs_ordered_extent
*ordered
;
5925 btrfs_release_path(root
, path
);
5927 lock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
5928 lock_end
, GFP_NOFS
);
5929 ordered
= btrfs_lookup_first_ordered_extent(inode
,
5932 ordered
->file_offset
<= lock_end
&&
5933 ordered
->file_offset
+ ordered
->len
> lock_start
) {
5934 unlock_extent(&BTRFS_I(inode
)->io_tree
,
5935 lock_start
, lock_end
, GFP_NOFS
);
5936 btrfs_start_ordered_extent(inode
, ordered
, 1);
5937 btrfs_put_ordered_extent(ordered
);
5938 key
.offset
+= num_bytes
;
5942 btrfs_put_ordered_extent(ordered
);
5948 if (nr_extents
== 1) {
5949 /* update extent pointer in place */
5950 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
5951 new_extents
[0].disk_bytenr
);
5952 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
5953 new_extents
[0].disk_num_bytes
);
5954 btrfs_mark_buffer_dirty(leaf
);
5956 btrfs_drop_extent_cache(inode
, key
.offset
,
5957 key
.offset
+ num_bytes
- 1, 0);
5959 ret
= btrfs_inc_extent_ref(trans
, root
,
5960 new_extents
[0].disk_bytenr
,
5961 new_extents
[0].disk_num_bytes
,
5963 root
->root_key
.objectid
,
5968 ret
= btrfs_free_extent(trans
, root
,
5969 extent_key
->objectid
,
5972 btrfs_header_owner(leaf
),
5973 btrfs_header_generation(leaf
),
5977 btrfs_release_path(root
, path
);
5978 key
.offset
+= num_bytes
;
5986 * drop old extent pointer at first, then insert the
5987 * new pointers one bye one
5989 btrfs_release_path(root
, path
);
5990 ret
= btrfs_drop_extents(trans
, root
, inode
, key
.offset
,
5991 key
.offset
+ num_bytes
,
5992 key
.offset
, &alloc_hint
);
5995 for (i
= 0; i
< nr_extents
; i
++) {
5996 if (ext_offset
>= new_extents
[i
].num_bytes
) {
5997 ext_offset
-= new_extents
[i
].num_bytes
;
6000 extent_len
= min(new_extents
[i
].num_bytes
-
6001 ext_offset
, num_bytes
);
6003 ret
= btrfs_insert_empty_item(trans
, root
,
6008 leaf
= path
->nodes
[0];
6009 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
6010 struct btrfs_file_extent_item
);
6011 btrfs_set_file_extent_generation(leaf
, fi
,
6013 btrfs_set_file_extent_type(leaf
, fi
,
6014 BTRFS_FILE_EXTENT_REG
);
6015 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
6016 new_extents
[i
].disk_bytenr
);
6017 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
6018 new_extents
[i
].disk_num_bytes
);
6019 btrfs_set_file_extent_ram_bytes(leaf
, fi
,
6020 new_extents
[i
].ram_bytes
);
6022 btrfs_set_file_extent_compression(leaf
, fi
,
6023 new_extents
[i
].compression
);
6024 btrfs_set_file_extent_encryption(leaf
, fi
,
6025 new_extents
[i
].encryption
);
6026 btrfs_set_file_extent_other_encoding(leaf
, fi
,
6027 new_extents
[i
].other_encoding
);
6029 btrfs_set_file_extent_num_bytes(leaf
, fi
,
6031 ext_offset
+= new_extents
[i
].offset
;
6032 btrfs_set_file_extent_offset(leaf
, fi
,
6034 btrfs_mark_buffer_dirty(leaf
);
6036 btrfs_drop_extent_cache(inode
, key
.offset
,
6037 key
.offset
+ extent_len
- 1, 0);
6039 ret
= btrfs_inc_extent_ref(trans
, root
,
6040 new_extents
[i
].disk_bytenr
,
6041 new_extents
[i
].disk_num_bytes
,
6043 root
->root_key
.objectid
,
6044 trans
->transid
, key
.objectid
);
6046 btrfs_release_path(root
, path
);
6048 inode_add_bytes(inode
, extent_len
);
6051 num_bytes
-= extent_len
;
6052 key
.offset
+= extent_len
;
6057 BUG_ON(i
>= nr_extents
);
6061 if (extent_locked
) {
6062 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
6063 lock_end
, GFP_NOFS
);
6067 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
&&
6068 key
.offset
>= search_end
)
6075 btrfs_release_path(root
, path
);
6077 mutex_unlock(&inode
->i_mutex
);
6078 if (extent_locked
) {
6079 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
6080 lock_end
, GFP_NOFS
);
6087 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle
*trans
,
6088 struct btrfs_root
*root
,
6089 struct extent_buffer
*buf
, u64 orig_start
)
6094 BUG_ON(btrfs_header_generation(buf
) != trans
->transid
);
6095 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
6097 level
= btrfs_header_level(buf
);
6099 struct btrfs_leaf_ref
*ref
;
6100 struct btrfs_leaf_ref
*orig_ref
;
6102 orig_ref
= btrfs_lookup_leaf_ref(root
, orig_start
);
6106 ref
= btrfs_alloc_leaf_ref(root
, orig_ref
->nritems
);
6108 btrfs_free_leaf_ref(root
, orig_ref
);
6112 ref
->nritems
= orig_ref
->nritems
;
6113 memcpy(ref
->extents
, orig_ref
->extents
,
6114 sizeof(ref
->extents
[0]) * ref
->nritems
);
6116 btrfs_free_leaf_ref(root
, orig_ref
);
6118 ref
->root_gen
= trans
->transid
;
6119 ref
->bytenr
= buf
->start
;
6120 ref
->owner
= btrfs_header_owner(buf
);
6121 ref
->generation
= btrfs_header_generation(buf
);
6123 ret
= btrfs_add_leaf_ref(root
, ref
, 0);
6125 btrfs_free_leaf_ref(root
, ref
);
6130 static noinline
int invalidate_extent_cache(struct btrfs_root
*root
,
6131 struct extent_buffer
*leaf
,
6132 struct btrfs_block_group_cache
*group
,
6133 struct btrfs_root
*target_root
)
6135 struct btrfs_key key
;
6136 struct inode
*inode
= NULL
;
6137 struct btrfs_file_extent_item
*fi
;
6139 u64 skip_objectid
= 0;
6143 nritems
= btrfs_header_nritems(leaf
);
6144 for (i
= 0; i
< nritems
; i
++) {
6145 btrfs_item_key_to_cpu(leaf
, &key
, i
);
6146 if (key
.objectid
== skip_objectid
||
6147 key
.type
!= BTRFS_EXTENT_DATA_KEY
)
6149 fi
= btrfs_item_ptr(leaf
, i
, struct btrfs_file_extent_item
);
6150 if (btrfs_file_extent_type(leaf
, fi
) ==
6151 BTRFS_FILE_EXTENT_INLINE
)
6153 if (btrfs_file_extent_disk_bytenr(leaf
, fi
) == 0)
6155 if (!inode
|| inode
->i_ino
!= key
.objectid
) {
6157 inode
= btrfs_ilookup(target_root
->fs_info
->sb
,
6158 key
.objectid
, target_root
, 1);
6161 skip_objectid
= key
.objectid
;
6164 num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
6166 lock_extent(&BTRFS_I(inode
)->io_tree
, key
.offset
,
6167 key
.offset
+ num_bytes
- 1, GFP_NOFS
);
6168 btrfs_drop_extent_cache(inode
, key
.offset
,
6169 key
.offset
+ num_bytes
- 1, 1);
6170 unlock_extent(&BTRFS_I(inode
)->io_tree
, key
.offset
,
6171 key
.offset
+ num_bytes
- 1, GFP_NOFS
);
6178 static noinline
int replace_extents_in_leaf(struct btrfs_trans_handle
*trans
,
6179 struct btrfs_root
*root
,
6180 struct extent_buffer
*leaf
,
6181 struct btrfs_block_group_cache
*group
,
6182 struct inode
*reloc_inode
)
6184 struct btrfs_key key
;
6185 struct btrfs_key extent_key
;
6186 struct btrfs_file_extent_item
*fi
;
6187 struct btrfs_leaf_ref
*ref
;
6188 struct disk_extent
*new_extent
;
6197 new_extent
= kmalloc(sizeof(*new_extent
), GFP_NOFS
);
6198 BUG_ON(!new_extent
);
6200 ref
= btrfs_lookup_leaf_ref(root
, leaf
->start
);
6204 nritems
= btrfs_header_nritems(leaf
);
6205 for (i
= 0; i
< nritems
; i
++) {
6206 btrfs_item_key_to_cpu(leaf
, &key
, i
);
6207 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
6209 fi
= btrfs_item_ptr(leaf
, i
, struct btrfs_file_extent_item
);
6210 if (btrfs_file_extent_type(leaf
, fi
) ==
6211 BTRFS_FILE_EXTENT_INLINE
)
6213 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
6214 num_bytes
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
6219 if (bytenr
>= group
->key
.objectid
+ group
->key
.offset
||
6220 bytenr
+ num_bytes
<= group
->key
.objectid
)
6223 extent_key
.objectid
= bytenr
;
6224 extent_key
.offset
= num_bytes
;
6225 extent_key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6227 ret
= get_new_locations(reloc_inode
, &extent_key
,
6228 group
->key
.objectid
, 1,
6229 &new_extent
, &nr_extent
);
6234 BUG_ON(ref
->extents
[ext_index
].bytenr
!= bytenr
);
6235 BUG_ON(ref
->extents
[ext_index
].num_bytes
!= num_bytes
);
6236 ref
->extents
[ext_index
].bytenr
= new_extent
->disk_bytenr
;
6237 ref
->extents
[ext_index
].num_bytes
= new_extent
->disk_num_bytes
;
6239 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
6240 new_extent
->disk_bytenr
);
6241 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
6242 new_extent
->disk_num_bytes
);
6243 btrfs_mark_buffer_dirty(leaf
);
6245 ret
= btrfs_inc_extent_ref(trans
, root
,
6246 new_extent
->disk_bytenr
,
6247 new_extent
->disk_num_bytes
,
6249 root
->root_key
.objectid
,
6250 trans
->transid
, key
.objectid
);
6253 ret
= btrfs_free_extent(trans
, root
,
6254 bytenr
, num_bytes
, leaf
->start
,
6255 btrfs_header_owner(leaf
),
6256 btrfs_header_generation(leaf
),
6262 BUG_ON(ext_index
+ 1 != ref
->nritems
);
6263 btrfs_free_leaf_ref(root
, ref
);
6267 int btrfs_free_reloc_root(struct btrfs_trans_handle
*trans
,
6268 struct btrfs_root
*root
)
6270 struct btrfs_root
*reloc_root
;
6273 if (root
->reloc_root
) {
6274 reloc_root
= root
->reloc_root
;
6275 root
->reloc_root
= NULL
;
6276 list_add(&reloc_root
->dead_list
,
6277 &root
->fs_info
->dead_reloc_roots
);
6279 btrfs_set_root_bytenr(&reloc_root
->root_item
,
6280 reloc_root
->node
->start
);
6281 btrfs_set_root_level(&root
->root_item
,
6282 btrfs_header_level(reloc_root
->node
));
6283 memset(&reloc_root
->root_item
.drop_progress
, 0,
6284 sizeof(struct btrfs_disk_key
));
6285 reloc_root
->root_item
.drop_level
= 0;
6287 ret
= btrfs_update_root(trans
, root
->fs_info
->tree_root
,
6288 &reloc_root
->root_key
,
6289 &reloc_root
->root_item
);
6295 int btrfs_drop_dead_reloc_roots(struct btrfs_root
*root
)
6297 struct btrfs_trans_handle
*trans
;
6298 struct btrfs_root
*reloc_root
;
6299 struct btrfs_root
*prev_root
= NULL
;
6300 struct list_head dead_roots
;
6304 INIT_LIST_HEAD(&dead_roots
);
6305 list_splice_init(&root
->fs_info
->dead_reloc_roots
, &dead_roots
);
6307 while (!list_empty(&dead_roots
)) {
6308 reloc_root
= list_entry(dead_roots
.prev
,
6309 struct btrfs_root
, dead_list
);
6310 list_del_init(&reloc_root
->dead_list
);
6312 BUG_ON(reloc_root
->commit_root
!= NULL
);
6314 trans
= btrfs_join_transaction(root
, 1);
6317 mutex_lock(&root
->fs_info
->drop_mutex
);
6318 ret
= btrfs_drop_snapshot(trans
, reloc_root
);
6321 mutex_unlock(&root
->fs_info
->drop_mutex
);
6323 nr
= trans
->blocks_used
;
6324 ret
= btrfs_end_transaction(trans
, root
);
6326 btrfs_btree_balance_dirty(root
, nr
);
6329 free_extent_buffer(reloc_root
->node
);
6331 ret
= btrfs_del_root(trans
, root
->fs_info
->tree_root
,
6332 &reloc_root
->root_key
);
6334 mutex_unlock(&root
->fs_info
->drop_mutex
);
6336 nr
= trans
->blocks_used
;
6337 ret
= btrfs_end_transaction(trans
, root
);
6339 btrfs_btree_balance_dirty(root
, nr
);
6342 prev_root
= reloc_root
;
6345 btrfs_remove_leaf_refs(prev_root
, (u64
)-1, 0);
6351 int btrfs_add_dead_reloc_root(struct btrfs_root
*root
)
6353 list_add(&root
->dead_list
, &root
->fs_info
->dead_reloc_roots
);
6357 int btrfs_cleanup_reloc_trees(struct btrfs_root
*root
)
6359 struct btrfs_root
*reloc_root
;
6360 struct btrfs_trans_handle
*trans
;
6361 struct btrfs_key location
;
6365 mutex_lock(&root
->fs_info
->tree_reloc_mutex
);
6366 ret
= btrfs_find_dead_roots(root
, BTRFS_TREE_RELOC_OBJECTID
, NULL
);
6368 found
= !list_empty(&root
->fs_info
->dead_reloc_roots
);
6369 mutex_unlock(&root
->fs_info
->tree_reloc_mutex
);
6372 trans
= btrfs_start_transaction(root
, 1);
6374 ret
= btrfs_commit_transaction(trans
, root
);
6378 location
.objectid
= BTRFS_DATA_RELOC_TREE_OBJECTID
;
6379 location
.offset
= (u64
)-1;
6380 location
.type
= BTRFS_ROOT_ITEM_KEY
;
6382 reloc_root
= btrfs_read_fs_root_no_name(root
->fs_info
, &location
);
6383 BUG_ON(!reloc_root
);
6384 btrfs_orphan_cleanup(reloc_root
);
6388 static noinline
int init_reloc_tree(struct btrfs_trans_handle
*trans
,
6389 struct btrfs_root
*root
)
6391 struct btrfs_root
*reloc_root
;
6392 struct extent_buffer
*eb
;
6393 struct btrfs_root_item
*root_item
;
6394 struct btrfs_key root_key
;
6397 BUG_ON(!root
->ref_cows
);
6398 if (root
->reloc_root
)
6401 root_item
= kmalloc(sizeof(*root_item
), GFP_NOFS
);
6404 ret
= btrfs_copy_root(trans
, root
, root
->commit_root
,
6405 &eb
, BTRFS_TREE_RELOC_OBJECTID
);
6408 root_key
.objectid
= BTRFS_TREE_RELOC_OBJECTID
;
6409 root_key
.offset
= root
->root_key
.objectid
;
6410 root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
6412 memcpy(root_item
, &root
->root_item
, sizeof(root_item
));
6413 btrfs_set_root_refs(root_item
, 0);
6414 btrfs_set_root_bytenr(root_item
, eb
->start
);
6415 btrfs_set_root_level(root_item
, btrfs_header_level(eb
));
6416 btrfs_set_root_generation(root_item
, trans
->transid
);
6418 btrfs_tree_unlock(eb
);
6419 free_extent_buffer(eb
);
6421 ret
= btrfs_insert_root(trans
, root
->fs_info
->tree_root
,
6422 &root_key
, root_item
);
6426 reloc_root
= btrfs_read_fs_root_no_radix(root
->fs_info
->tree_root
,
6428 BUG_ON(!reloc_root
);
6429 reloc_root
->last_trans
= trans
->transid
;
6430 reloc_root
->commit_root
= NULL
;
6431 reloc_root
->ref_tree
= &root
->fs_info
->reloc_ref_tree
;
6433 root
->reloc_root
= reloc_root
;
6438 * Core function of space balance.
6440 * The idea is using reloc trees to relocate tree blocks in reference
6441 * counted roots. There is one reloc tree for each subvol, and all
6442 * reloc trees share same root key objectid. Reloc trees are snapshots
6443 * of the latest committed roots of subvols (root->commit_root).
6445 * To relocate a tree block referenced by a subvol, there are two steps.
6446 * COW the block through subvol's reloc tree, then update block pointer
6447 * in the subvol to point to the new block. Since all reloc trees share
6448 * same root key objectid, doing special handing for tree blocks owned
6449 * by them is easy. Once a tree block has been COWed in one reloc tree,
6450 * we can use the resulting new block directly when the same block is
6451 * required to COW again through other reloc trees. By this way, relocated
6452 * tree blocks are shared between reloc trees, so they are also shared
6455 static noinline
int relocate_one_path(struct btrfs_trans_handle
*trans
,
6456 struct btrfs_root
*root
,
6457 struct btrfs_path
*path
,
6458 struct btrfs_key
*first_key
,
6459 struct btrfs_ref_path
*ref_path
,
6460 struct btrfs_block_group_cache
*group
,
6461 struct inode
*reloc_inode
)
6463 struct btrfs_root
*reloc_root
;
6464 struct extent_buffer
*eb
= NULL
;
6465 struct btrfs_key
*keys
;
6469 int lowest_level
= 0;
6472 if (ref_path
->owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
)
6473 lowest_level
= ref_path
->owner_objectid
;
6475 if (!root
->ref_cows
) {
6476 path
->lowest_level
= lowest_level
;
6477 ret
= btrfs_search_slot(trans
, root
, first_key
, path
, 0, 1);
6479 path
->lowest_level
= 0;
6480 btrfs_release_path(root
, path
);
6484 mutex_lock(&root
->fs_info
->tree_reloc_mutex
);
6485 ret
= init_reloc_tree(trans
, root
);
6487 reloc_root
= root
->reloc_root
;
6489 shared_level
= ref_path
->shared_level
;
6490 ref_path
->shared_level
= BTRFS_MAX_LEVEL
- 1;
6492 keys
= ref_path
->node_keys
;
6493 nodes
= ref_path
->new_nodes
;
6494 memset(&keys
[shared_level
+ 1], 0,
6495 sizeof(*keys
) * (BTRFS_MAX_LEVEL
- shared_level
- 1));
6496 memset(&nodes
[shared_level
+ 1], 0,
6497 sizeof(*nodes
) * (BTRFS_MAX_LEVEL
- shared_level
- 1));
6499 if (nodes
[lowest_level
] == 0) {
6500 path
->lowest_level
= lowest_level
;
6501 ret
= btrfs_search_slot(trans
, reloc_root
, first_key
, path
,
6504 for (level
= lowest_level
; level
< BTRFS_MAX_LEVEL
; level
++) {
6505 eb
= path
->nodes
[level
];
6506 if (!eb
|| eb
== reloc_root
->node
)
6508 nodes
[level
] = eb
->start
;
6510 btrfs_item_key_to_cpu(eb
, &keys
[level
], 0);
6512 btrfs_node_key_to_cpu(eb
, &keys
[level
], 0);
6515 ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
6516 eb
= path
->nodes
[0];
6517 ret
= replace_extents_in_leaf(trans
, reloc_root
, eb
,
6518 group
, reloc_inode
);
6521 btrfs_release_path(reloc_root
, path
);
6523 ret
= btrfs_merge_path(trans
, reloc_root
, keys
, nodes
,
6529 * replace tree blocks in the fs tree with tree blocks in
6532 ret
= btrfs_merge_path(trans
, root
, keys
, nodes
, lowest_level
);
6535 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
6536 ret
= btrfs_search_slot(trans
, reloc_root
, first_key
, path
,
6539 extent_buffer_get(path
->nodes
[0]);
6540 eb
= path
->nodes
[0];
6541 btrfs_release_path(reloc_root
, path
);
6542 ret
= invalidate_extent_cache(reloc_root
, eb
, group
, root
);
6544 free_extent_buffer(eb
);
6547 mutex_unlock(&root
->fs_info
->tree_reloc_mutex
);
6548 path
->lowest_level
= 0;
6552 static noinline
int relocate_tree_block(struct btrfs_trans_handle
*trans
,
6553 struct btrfs_root
*root
,
6554 struct btrfs_path
*path
,
6555 struct btrfs_key
*first_key
,
6556 struct btrfs_ref_path
*ref_path
)
6560 ret
= relocate_one_path(trans
, root
, path
, first_key
,
6561 ref_path
, NULL
, NULL
);
6567 static noinline
int del_extent_zero(struct btrfs_trans_handle
*trans
,
6568 struct btrfs_root
*extent_root
,
6569 struct btrfs_path
*path
,
6570 struct btrfs_key
*extent_key
)
6574 ret
= btrfs_search_slot(trans
, extent_root
, extent_key
, path
, -1, 1);
6577 ret
= btrfs_del_item(trans
, extent_root
, path
);
6579 btrfs_release_path(extent_root
, path
);
6583 static noinline
struct btrfs_root
*read_ref_root(struct btrfs_fs_info
*fs_info
,
6584 struct btrfs_ref_path
*ref_path
)
6586 struct btrfs_key root_key
;
6588 root_key
.objectid
= ref_path
->root_objectid
;
6589 root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
6590 if (is_cowonly_root(ref_path
->root_objectid
))
6591 root_key
.offset
= 0;
6593 root_key
.offset
= (u64
)-1;
6595 return btrfs_read_fs_root_no_name(fs_info
, &root_key
);
6598 static noinline
int relocate_one_extent(struct btrfs_root
*extent_root
,
6599 struct btrfs_path
*path
,
6600 struct btrfs_key
*extent_key
,
6601 struct btrfs_block_group_cache
*group
,
6602 struct inode
*reloc_inode
, int pass
)
6604 struct btrfs_trans_handle
*trans
;
6605 struct btrfs_root
*found_root
;
6606 struct btrfs_ref_path
*ref_path
= NULL
;
6607 struct disk_extent
*new_extents
= NULL
;
6612 struct btrfs_key first_key
;
6616 trans
= btrfs_start_transaction(extent_root
, 1);
6619 if (extent_key
->objectid
== 0) {
6620 ret
= del_extent_zero(trans
, extent_root
, path
, extent_key
);
6624 ref_path
= kmalloc(sizeof(*ref_path
), GFP_NOFS
);
6630 for (loops
= 0; ; loops
++) {
6632 ret
= btrfs_first_ref_path(trans
, extent_root
, ref_path
,
6633 extent_key
->objectid
);
6635 ret
= btrfs_next_ref_path(trans
, extent_root
, ref_path
);
6642 if (ref_path
->root_objectid
== BTRFS_TREE_LOG_OBJECTID
||
6643 ref_path
->root_objectid
== BTRFS_TREE_RELOC_OBJECTID
)
6646 found_root
= read_ref_root(extent_root
->fs_info
, ref_path
);
6647 BUG_ON(!found_root
);
6649 * for reference counted tree, only process reference paths
6650 * rooted at the latest committed root.
6652 if (found_root
->ref_cows
&&
6653 ref_path
->root_generation
!= found_root
->root_key
.offset
)
6656 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
6659 * copy data extents to new locations
6661 u64 group_start
= group
->key
.objectid
;
6662 ret
= relocate_data_extent(reloc_inode
,
6671 level
= ref_path
->owner_objectid
;
6674 if (prev_block
!= ref_path
->nodes
[level
]) {
6675 struct extent_buffer
*eb
;
6676 u64 block_start
= ref_path
->nodes
[level
];
6677 u64 block_size
= btrfs_level_size(found_root
, level
);
6679 eb
= read_tree_block(found_root
, block_start
,
6681 btrfs_tree_lock(eb
);
6682 BUG_ON(level
!= btrfs_header_level(eb
));
6685 btrfs_item_key_to_cpu(eb
, &first_key
, 0);
6687 btrfs_node_key_to_cpu(eb
, &first_key
, 0);
6689 btrfs_tree_unlock(eb
);
6690 free_extent_buffer(eb
);
6691 prev_block
= block_start
;
6694 mutex_lock(&extent_root
->fs_info
->trans_mutex
);
6695 btrfs_record_root_in_trans(found_root
);
6696 mutex_unlock(&extent_root
->fs_info
->trans_mutex
);
6697 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
6699 * try to update data extent references while
6700 * keeping metadata shared between snapshots.
6703 ret
= relocate_one_path(trans
, found_root
,
6704 path
, &first_key
, ref_path
,
6705 group
, reloc_inode
);
6711 * use fallback method to process the remaining
6715 u64 group_start
= group
->key
.objectid
;
6716 new_extents
= kmalloc(sizeof(*new_extents
),
6719 ret
= get_new_locations(reloc_inode
,
6727 ret
= replace_one_extent(trans
, found_root
,
6729 &first_key
, ref_path
,
6730 new_extents
, nr_extents
);
6732 ret
= relocate_tree_block(trans
, found_root
, path
,
6733 &first_key
, ref_path
);
6740 btrfs_end_transaction(trans
, extent_root
);
6747 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
6750 u64 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
6751 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
6753 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
6754 if (num_devices
== 1) {
6755 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
6756 stripped
= flags
& ~stripped
;
6758 /* turn raid0 into single device chunks */
6759 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
6762 /* turn mirroring into duplication */
6763 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
6764 BTRFS_BLOCK_GROUP_RAID10
))
6765 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
6768 /* they already had raid on here, just return */
6769 if (flags
& stripped
)
6772 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
6773 stripped
= flags
& ~stripped
;
6775 /* switch duplicated blocks with raid1 */
6776 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
6777 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
6779 /* turn single device chunks into raid0 */
6780 return stripped
| BTRFS_BLOCK_GROUP_RAID0
;
6785 static int __alloc_chunk_for_shrink(struct btrfs_root
*root
,
6786 struct btrfs_block_group_cache
*shrink_block_group
,
6789 struct btrfs_trans_handle
*trans
;
6790 u64 new_alloc_flags
;
6793 spin_lock(&shrink_block_group
->lock
);
6794 if (btrfs_block_group_used(&shrink_block_group
->item
) +
6795 shrink_block_group
->reserved
> 0) {
6796 spin_unlock(&shrink_block_group
->lock
);
6798 trans
= btrfs_start_transaction(root
, 1);
6799 spin_lock(&shrink_block_group
->lock
);
6801 new_alloc_flags
= update_block_group_flags(root
,
6802 shrink_block_group
->flags
);
6803 if (new_alloc_flags
!= shrink_block_group
->flags
) {
6805 btrfs_block_group_used(&shrink_block_group
->item
);
6807 calc
= shrink_block_group
->key
.offset
;
6809 spin_unlock(&shrink_block_group
->lock
);
6811 do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
6812 calc
+ 2 * 1024 * 1024, new_alloc_flags
, force
);
6814 btrfs_end_transaction(trans
, root
);
6816 spin_unlock(&shrink_block_group
->lock
);
6821 int btrfs_prepare_block_group_relocation(struct btrfs_root
*root
,
6822 struct btrfs_block_group_cache
*group
)
6825 __alloc_chunk_for_shrink(root
, group
, 1);
6826 set_block_group_readonly(group
);
6831 static int __insert_orphan_inode(struct btrfs_trans_handle
*trans
,
6832 struct btrfs_root
*root
,
6833 u64 objectid
, u64 size
)
6835 struct btrfs_path
*path
;
6836 struct btrfs_inode_item
*item
;
6837 struct extent_buffer
*leaf
;
6840 path
= btrfs_alloc_path();
6844 path
->leave_spinning
= 1;
6845 ret
= btrfs_insert_empty_inode(trans
, root
, path
, objectid
);
6849 leaf
= path
->nodes
[0];
6850 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_inode_item
);
6851 memset_extent_buffer(leaf
, 0, (unsigned long)item
, sizeof(*item
));
6852 btrfs_set_inode_generation(leaf
, item
, 1);
6853 btrfs_set_inode_size(leaf
, item
, size
);
6854 btrfs_set_inode_mode(leaf
, item
, S_IFREG
| 0600);
6855 btrfs_set_inode_flags(leaf
, item
, BTRFS_INODE_NOCOMPRESS
);
6856 btrfs_mark_buffer_dirty(leaf
);
6857 btrfs_release_path(root
, path
);
6859 btrfs_free_path(path
);
6863 static noinline
struct inode
*create_reloc_inode(struct btrfs_fs_info
*fs_info
,
6864 struct btrfs_block_group_cache
*group
)
6866 struct inode
*inode
= NULL
;
6867 struct btrfs_trans_handle
*trans
;
6868 struct btrfs_root
*root
;
6869 struct btrfs_key root_key
;
6870 u64 objectid
= BTRFS_FIRST_FREE_OBJECTID
;
6873 root_key
.objectid
= BTRFS_DATA_RELOC_TREE_OBJECTID
;
6874 root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
6875 root_key
.offset
= (u64
)-1;
6876 root
= btrfs_read_fs_root_no_name(fs_info
, &root_key
);
6878 return ERR_CAST(root
);
6880 trans
= btrfs_start_transaction(root
, 1);
6883 err
= btrfs_find_free_objectid(trans
, root
, objectid
, &objectid
);
6887 err
= __insert_orphan_inode(trans
, root
, objectid
, group
->key
.offset
);
6890 err
= btrfs_insert_file_extent(trans
, root
, objectid
, 0, 0, 0,
6891 group
->key
.offset
, 0, group
->key
.offset
,
6895 inode
= btrfs_iget_locked(root
->fs_info
->sb
, objectid
, root
);
6896 if (inode
->i_state
& I_NEW
) {
6897 BTRFS_I(inode
)->root
= root
;
6898 BTRFS_I(inode
)->location
.objectid
= objectid
;
6899 BTRFS_I(inode
)->location
.type
= BTRFS_INODE_ITEM_KEY
;
6900 BTRFS_I(inode
)->location
.offset
= 0;
6901 btrfs_read_locked_inode(inode
);
6902 unlock_new_inode(inode
);
6903 BUG_ON(is_bad_inode(inode
));
6907 BTRFS_I(inode
)->index_cnt
= group
->key
.objectid
;
6909 err
= btrfs_orphan_add(trans
, inode
);
6911 btrfs_end_transaction(trans
, root
);
6915 inode
= ERR_PTR(err
);
6920 int btrfs_reloc_clone_csums(struct inode
*inode
, u64 file_pos
, u64 len
)
6923 struct btrfs_ordered_sum
*sums
;
6924 struct btrfs_sector_sum
*sector_sum
;
6925 struct btrfs_ordered_extent
*ordered
;
6926 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6927 struct list_head list
;
6932 INIT_LIST_HEAD(&list
);
6934 ordered
= btrfs_lookup_ordered_extent(inode
, file_pos
);
6935 BUG_ON(ordered
->file_offset
!= file_pos
|| ordered
->len
!= len
);
6937 disk_bytenr
= file_pos
+ BTRFS_I(inode
)->index_cnt
;
6938 ret
= btrfs_lookup_csums_range(root
->fs_info
->csum_root
, disk_bytenr
,
6939 disk_bytenr
+ len
- 1, &list
);
6941 while (!list_empty(&list
)) {
6942 sums
= list_entry(list
.next
, struct btrfs_ordered_sum
, list
);
6943 list_del_init(&sums
->list
);
6945 sector_sum
= sums
->sums
;
6946 sums
->bytenr
= ordered
->start
;
6949 while (offset
< sums
->len
) {
6950 sector_sum
->bytenr
+= ordered
->start
- disk_bytenr
;
6952 offset
+= root
->sectorsize
;
6955 btrfs_add_ordered_sum(inode
, ordered
, sums
);
6957 btrfs_put_ordered_extent(ordered
);
6961 int btrfs_relocate_block_group(struct btrfs_root
*root
, u64 group_start
)
6963 struct btrfs_trans_handle
*trans
;
6964 struct btrfs_path
*path
;
6965 struct btrfs_fs_info
*info
= root
->fs_info
;
6966 struct extent_buffer
*leaf
;
6967 struct inode
*reloc_inode
;
6968 struct btrfs_block_group_cache
*block_group
;
6969 struct btrfs_key key
;
6978 root
= root
->fs_info
->extent_root
;
6980 block_group
= btrfs_lookup_block_group(info
, group_start
);
6981 BUG_ON(!block_group
);
6983 printk(KERN_INFO
"btrfs relocating block group %llu flags %llu\n",
6984 (unsigned long long)block_group
->key
.objectid
,
6985 (unsigned long long)block_group
->flags
);
6987 path
= btrfs_alloc_path();
6990 reloc_inode
= create_reloc_inode(info
, block_group
);
6991 BUG_ON(IS_ERR(reloc_inode
));
6993 __alloc_chunk_for_shrink(root
, block_group
, 1);
6994 set_block_group_readonly(block_group
);
6996 btrfs_start_delalloc_inodes(info
->tree_root
);
6997 btrfs_wait_ordered_extents(info
->tree_root
, 0);
7002 key
.objectid
= block_group
->key
.objectid
;
7005 cur_byte
= key
.objectid
;
7007 trans
= btrfs_start_transaction(info
->tree_root
, 1);
7008 btrfs_commit_transaction(trans
, info
->tree_root
);
7010 mutex_lock(&root
->fs_info
->cleaner_mutex
);
7011 btrfs_clean_old_snapshots(info
->tree_root
);
7012 btrfs_remove_leaf_refs(info
->tree_root
, (u64
)-1, 1);
7013 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
7015 trans
= btrfs_start_transaction(info
->tree_root
, 1);
7016 btrfs_commit_transaction(trans
, info
->tree_root
);
7019 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
7023 leaf
= path
->nodes
[0];
7024 nritems
= btrfs_header_nritems(leaf
);
7025 if (path
->slots
[0] >= nritems
) {
7026 ret
= btrfs_next_leaf(root
, path
);
7033 leaf
= path
->nodes
[0];
7034 nritems
= btrfs_header_nritems(leaf
);
7037 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
7039 if (key
.objectid
>= block_group
->key
.objectid
+
7040 block_group
->key
.offset
)
7043 if (progress
&& need_resched()) {
7044 btrfs_release_path(root
, path
);
7051 if (btrfs_key_type(&key
) != BTRFS_EXTENT_ITEM_KEY
||
7052 key
.objectid
+ key
.offset
<= cur_byte
) {
7058 cur_byte
= key
.objectid
+ key
.offset
;
7059 btrfs_release_path(root
, path
);
7061 __alloc_chunk_for_shrink(root
, block_group
, 0);
7062 ret
= relocate_one_extent(root
, path
, &key
, block_group
,
7068 key
.objectid
= cur_byte
;
7073 btrfs_release_path(root
, path
);
7076 btrfs_wait_ordered_range(reloc_inode
, 0, (u64
)-1);
7077 invalidate_mapping_pages(reloc_inode
->i_mapping
, 0, -1);
7080 if (total_found
> 0) {
7081 printk(KERN_INFO
"btrfs found %llu extents in pass %d\n",
7082 (unsigned long long)total_found
, pass
);
7084 if (total_found
== skipped
&& pass
> 2) {
7086 reloc_inode
= create_reloc_inode(info
, block_group
);
7092 /* delete reloc_inode */
7095 /* unpin extents in this range */
7096 trans
= btrfs_start_transaction(info
->tree_root
, 1);
7097 btrfs_commit_transaction(trans
, info
->tree_root
);
7099 spin_lock(&block_group
->lock
);
7100 WARN_ON(block_group
->pinned
> 0);
7101 WARN_ON(block_group
->reserved
> 0);
7102 WARN_ON(btrfs_block_group_used(&block_group
->item
) > 0);
7103 spin_unlock(&block_group
->lock
);
7104 btrfs_put_block_group(block_group
);
7107 btrfs_free_path(path
);
7112 static int find_first_block_group(struct btrfs_root
*root
,
7113 struct btrfs_path
*path
, struct btrfs_key
*key
)
7116 struct btrfs_key found_key
;
7117 struct extent_buffer
*leaf
;
7120 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
7125 slot
= path
->slots
[0];
7126 leaf
= path
->nodes
[0];
7127 if (slot
>= btrfs_header_nritems(leaf
)) {
7128 ret
= btrfs_next_leaf(root
, path
);
7135 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
7137 if (found_key
.objectid
>= key
->objectid
&&
7138 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
7149 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
7151 struct btrfs_block_group_cache
*block_group
;
7152 struct btrfs_space_info
*space_info
;
7155 spin_lock(&info
->block_group_cache_lock
);
7156 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
7157 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
7159 rb_erase(&block_group
->cache_node
,
7160 &info
->block_group_cache_tree
);
7161 spin_unlock(&info
->block_group_cache_lock
);
7163 down_write(&block_group
->space_info
->groups_sem
);
7164 list_del(&block_group
->list
);
7165 up_write(&block_group
->space_info
->groups_sem
);
7167 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
7168 wait_event(block_group
->caching_q
,
7169 block_group_cache_done(block_group
));
7171 btrfs_remove_free_space_cache(block_group
);
7173 WARN_ON(atomic_read(&block_group
->count
) != 1);
7176 spin_lock(&info
->block_group_cache_lock
);
7178 spin_unlock(&info
->block_group_cache_lock
);
7180 /* now that all the block groups are freed, go through and
7181 * free all the space_info structs. This is only called during
7182 * the final stages of unmount, and so we know nobody is
7183 * using them. We call synchronize_rcu() once before we start,
7184 * just to be on the safe side.
7188 while(!list_empty(&info
->space_info
)) {
7189 space_info
= list_entry(info
->space_info
.next
,
7190 struct btrfs_space_info
,
7193 list_del(&space_info
->list
);
7199 int btrfs_read_block_groups(struct btrfs_root
*root
)
7201 struct btrfs_path
*path
;
7203 struct btrfs_block_group_cache
*cache
;
7204 struct btrfs_fs_info
*info
= root
->fs_info
;
7205 struct btrfs_space_info
*space_info
;
7206 struct btrfs_key key
;
7207 struct btrfs_key found_key
;
7208 struct extent_buffer
*leaf
;
7210 root
= info
->extent_root
;
7213 btrfs_set_key_type(&key
, BTRFS_BLOCK_GROUP_ITEM_KEY
);
7214 path
= btrfs_alloc_path();
7219 ret
= find_first_block_group(root
, path
, &key
);
7227 leaf
= path
->nodes
[0];
7228 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
7229 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
7235 atomic_set(&cache
->count
, 1);
7236 spin_lock_init(&cache
->lock
);
7237 spin_lock_init(&cache
->tree_lock
);
7238 cache
->fs_info
= info
;
7239 init_waitqueue_head(&cache
->caching_q
);
7240 INIT_LIST_HEAD(&cache
->list
);
7241 INIT_LIST_HEAD(&cache
->cluster_list
);
7244 * we only want to have 32k of ram per block group for keeping
7245 * track of free space, and if we pass 1/2 of that we want to
7246 * start converting things over to using bitmaps
7248 cache
->extents_thresh
= ((1024 * 32) / 2) /
7249 sizeof(struct btrfs_free_space
);
7251 read_extent_buffer(leaf
, &cache
->item
,
7252 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
7253 sizeof(cache
->item
));
7254 memcpy(&cache
->key
, &found_key
, sizeof(found_key
));
7256 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
7257 btrfs_release_path(root
, path
);
7258 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
7259 cache
->sectorsize
= root
->sectorsize
;
7261 remove_sb_from_cache(root
, cache
);
7264 * check for two cases, either we are full, and therefore
7265 * don't need to bother with the caching work since we won't
7266 * find any space, or we are empty, and we can just add all
7267 * the space in and be done with it. This saves us _alot_ of
7268 * time, particularly in the full case.
7270 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
7271 cache
->cached
= BTRFS_CACHE_FINISHED
;
7272 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
7273 cache
->cached
= BTRFS_CACHE_FINISHED
;
7274 add_new_free_space(cache
, root
->fs_info
,
7276 found_key
.objectid
+
7280 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
7281 btrfs_block_group_used(&cache
->item
),
7284 cache
->space_info
= space_info
;
7285 down_write(&space_info
->groups_sem
);
7286 list_add_tail(&cache
->list
, &space_info
->block_groups
);
7287 up_write(&space_info
->groups_sem
);
7289 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
7292 set_avail_alloc_bits(root
->fs_info
, cache
->flags
);
7293 if (btrfs_chunk_readonly(root
, cache
->key
.objectid
))
7294 set_block_group_readonly(cache
);
7298 btrfs_free_path(path
);
7302 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
7303 struct btrfs_root
*root
, u64 bytes_used
,
7304 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
7308 struct btrfs_root
*extent_root
;
7309 struct btrfs_block_group_cache
*cache
;
7311 extent_root
= root
->fs_info
->extent_root
;
7313 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
7315 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
7319 cache
->key
.objectid
= chunk_offset
;
7320 cache
->key
.offset
= size
;
7321 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
7322 cache
->sectorsize
= root
->sectorsize
;
7325 * we only want to have 32k of ram per block group for keeping track
7326 * of free space, and if we pass 1/2 of that we want to start
7327 * converting things over to using bitmaps
7329 cache
->extents_thresh
= ((1024 * 32) / 2) /
7330 sizeof(struct btrfs_free_space
);
7331 atomic_set(&cache
->count
, 1);
7332 spin_lock_init(&cache
->lock
);
7333 spin_lock_init(&cache
->tree_lock
);
7334 init_waitqueue_head(&cache
->caching_q
);
7335 INIT_LIST_HEAD(&cache
->list
);
7336 INIT_LIST_HEAD(&cache
->cluster_list
);
7338 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
7339 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
7340 cache
->flags
= type
;
7341 btrfs_set_block_group_flags(&cache
->item
, type
);
7343 cache
->cached
= BTRFS_CACHE_FINISHED
;
7344 remove_sb_from_cache(root
, cache
);
7346 add_new_free_space(cache
, root
->fs_info
, chunk_offset
,
7347 chunk_offset
+ size
);
7349 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
7350 &cache
->space_info
);
7352 down_write(&cache
->space_info
->groups_sem
);
7353 list_add_tail(&cache
->list
, &cache
->space_info
->block_groups
);
7354 up_write(&cache
->space_info
->groups_sem
);
7356 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
7359 ret
= btrfs_insert_item(trans
, extent_root
, &cache
->key
, &cache
->item
,
7360 sizeof(cache
->item
));
7363 set_avail_alloc_bits(extent_root
->fs_info
, type
);
7368 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
7369 struct btrfs_root
*root
, u64 group_start
)
7371 struct btrfs_path
*path
;
7372 struct btrfs_block_group_cache
*block_group
;
7373 struct btrfs_free_cluster
*cluster
;
7374 struct btrfs_key key
;
7377 root
= root
->fs_info
->extent_root
;
7379 block_group
= btrfs_lookup_block_group(root
->fs_info
, group_start
);
7380 BUG_ON(!block_group
);
7381 BUG_ON(!block_group
->ro
);
7383 memcpy(&key
, &block_group
->key
, sizeof(key
));
7385 /* make sure this block group isn't part of an allocation cluster */
7386 cluster
= &root
->fs_info
->data_alloc_cluster
;
7387 spin_lock(&cluster
->refill_lock
);
7388 btrfs_return_cluster_to_free_space(block_group
, cluster
);
7389 spin_unlock(&cluster
->refill_lock
);
7392 * make sure this block group isn't part of a metadata
7393 * allocation cluster
7395 cluster
= &root
->fs_info
->meta_alloc_cluster
;
7396 spin_lock(&cluster
->refill_lock
);
7397 btrfs_return_cluster_to_free_space(block_group
, cluster
);
7398 spin_unlock(&cluster
->refill_lock
);
7400 path
= btrfs_alloc_path();
7403 spin_lock(&root
->fs_info
->block_group_cache_lock
);
7404 rb_erase(&block_group
->cache_node
,
7405 &root
->fs_info
->block_group_cache_tree
);
7406 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
7408 down_write(&block_group
->space_info
->groups_sem
);
7410 * we must use list_del_init so people can check to see if they
7411 * are still on the list after taking the semaphore
7413 list_del_init(&block_group
->list
);
7414 up_write(&block_group
->space_info
->groups_sem
);
7416 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
7417 wait_event(block_group
->caching_q
,
7418 block_group_cache_done(block_group
));
7420 btrfs_remove_free_space_cache(block_group
);
7422 spin_lock(&block_group
->space_info
->lock
);
7423 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
7424 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
7425 spin_unlock(&block_group
->space_info
->lock
);
7427 btrfs_clear_space_info_full(root
->fs_info
);
7429 btrfs_put_block_group(block_group
);
7430 btrfs_put_block_group(block_group
);
7432 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
7438 ret
= btrfs_del_item(trans
, root
, path
);
7440 btrfs_free_path(path
);