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_block_group(struct btrfs_trans_handle
*trans
,
36 struct btrfs_root
*root
,
37 u64 bytenr
, u64 num_bytes
, int alloc
,
39 static int update_reserved_extents(struct btrfs_block_group_cache
*cache
,
40 u64 num_bytes
, int reserve
);
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
);
60 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
61 struct btrfs_root
*extent_root
, u64 alloc_bytes
,
62 u64 flags
, int force
);
63 static int pin_down_bytes(struct btrfs_trans_handle
*trans
,
64 struct btrfs_root
*root
,
65 struct btrfs_path
*path
,
66 u64 bytenr
, u64 num_bytes
,
67 int is_data
, int reserved
,
68 struct extent_buffer
**must_clean
);
69 static int find_next_key(struct btrfs_path
*path
, int level
,
70 struct btrfs_key
*key
);
71 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
72 int dump_block_groups
);
75 block_group_cache_done(struct btrfs_block_group_cache
*cache
)
78 return cache
->cached
== BTRFS_CACHE_FINISHED
;
81 static int block_group_bits(struct btrfs_block_group_cache
*cache
, u64 bits
)
83 return (cache
->flags
& bits
) == bits
;
87 * this adds the block group to the fs_info rb tree for the block group
90 static int btrfs_add_block_group_cache(struct btrfs_fs_info
*info
,
91 struct btrfs_block_group_cache
*block_group
)
94 struct rb_node
*parent
= NULL
;
95 struct btrfs_block_group_cache
*cache
;
97 spin_lock(&info
->block_group_cache_lock
);
98 p
= &info
->block_group_cache_tree
.rb_node
;
102 cache
= rb_entry(parent
, struct btrfs_block_group_cache
,
104 if (block_group
->key
.objectid
< cache
->key
.objectid
) {
106 } else if (block_group
->key
.objectid
> cache
->key
.objectid
) {
109 spin_unlock(&info
->block_group_cache_lock
);
114 rb_link_node(&block_group
->cache_node
, parent
, p
);
115 rb_insert_color(&block_group
->cache_node
,
116 &info
->block_group_cache_tree
);
117 spin_unlock(&info
->block_group_cache_lock
);
123 * This will return the block group at or after bytenr if contains is 0, else
124 * it will return the block group that contains the bytenr
126 static struct btrfs_block_group_cache
*
127 block_group_cache_tree_search(struct btrfs_fs_info
*info
, u64 bytenr
,
130 struct btrfs_block_group_cache
*cache
, *ret
= NULL
;
134 spin_lock(&info
->block_group_cache_lock
);
135 n
= info
->block_group_cache_tree
.rb_node
;
138 cache
= rb_entry(n
, struct btrfs_block_group_cache
,
140 end
= cache
->key
.objectid
+ cache
->key
.offset
- 1;
141 start
= cache
->key
.objectid
;
143 if (bytenr
< start
) {
144 if (!contains
&& (!ret
|| start
< ret
->key
.objectid
))
147 } else if (bytenr
> start
) {
148 if (contains
&& bytenr
<= end
) {
159 atomic_inc(&ret
->count
);
160 spin_unlock(&info
->block_group_cache_lock
);
165 static int add_excluded_extent(struct btrfs_root
*root
,
166 u64 start
, u64 num_bytes
)
168 u64 end
= start
+ num_bytes
- 1;
169 set_extent_bits(&root
->fs_info
->freed_extents
[0],
170 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
171 set_extent_bits(&root
->fs_info
->freed_extents
[1],
172 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
176 static void free_excluded_extents(struct btrfs_root
*root
,
177 struct btrfs_block_group_cache
*cache
)
181 start
= cache
->key
.objectid
;
182 end
= start
+ cache
->key
.offset
- 1;
184 clear_extent_bits(&root
->fs_info
->freed_extents
[0],
185 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
186 clear_extent_bits(&root
->fs_info
->freed_extents
[1],
187 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
190 static int exclude_super_stripes(struct btrfs_root
*root
,
191 struct btrfs_block_group_cache
*cache
)
198 for (i
= 0; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
199 bytenr
= btrfs_sb_offset(i
);
200 ret
= btrfs_rmap_block(&root
->fs_info
->mapping_tree
,
201 cache
->key
.objectid
, bytenr
,
202 0, &logical
, &nr
, &stripe_len
);
206 cache
->bytes_super
+= stripe_len
;
207 ret
= add_excluded_extent(root
, logical
[nr
],
217 static struct btrfs_caching_control
*
218 get_caching_control(struct btrfs_block_group_cache
*cache
)
220 struct btrfs_caching_control
*ctl
;
222 spin_lock(&cache
->lock
);
223 if (cache
->cached
!= BTRFS_CACHE_STARTED
) {
224 spin_unlock(&cache
->lock
);
228 ctl
= cache
->caching_ctl
;
229 atomic_inc(&ctl
->count
);
230 spin_unlock(&cache
->lock
);
234 static void put_caching_control(struct btrfs_caching_control
*ctl
)
236 if (atomic_dec_and_test(&ctl
->count
))
241 * this is only called by cache_block_group, since we could have freed extents
242 * we need to check the pinned_extents for any extents that can't be used yet
243 * since their free space will be released as soon as the transaction commits.
245 static u64
add_new_free_space(struct btrfs_block_group_cache
*block_group
,
246 struct btrfs_fs_info
*info
, u64 start
, u64 end
)
248 u64 extent_start
, extent_end
, size
, total_added
= 0;
251 while (start
< end
) {
252 ret
= find_first_extent_bit(info
->pinned_extents
, start
,
253 &extent_start
, &extent_end
,
254 EXTENT_DIRTY
| EXTENT_UPTODATE
);
258 if (extent_start
== start
) {
259 start
= extent_end
+ 1;
260 } else if (extent_start
> start
&& extent_start
< end
) {
261 size
= extent_start
- start
;
263 ret
= btrfs_add_free_space(block_group
, start
,
266 start
= extent_end
+ 1;
275 ret
= btrfs_add_free_space(block_group
, start
, size
);
282 static int caching_kthread(void *data
)
284 struct btrfs_block_group_cache
*block_group
= data
;
285 struct btrfs_fs_info
*fs_info
= block_group
->fs_info
;
286 struct btrfs_caching_control
*caching_ctl
= block_group
->caching_ctl
;
287 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
288 struct btrfs_path
*path
;
289 struct extent_buffer
*leaf
;
290 struct btrfs_key key
;
296 path
= btrfs_alloc_path();
300 exclude_super_stripes(extent_root
, block_group
);
301 spin_lock(&block_group
->space_info
->lock
);
302 block_group
->space_info
->bytes_super
+= block_group
->bytes_super
;
303 spin_unlock(&block_group
->space_info
->lock
);
305 last
= max_t(u64
, block_group
->key
.objectid
, BTRFS_SUPER_INFO_OFFSET
);
308 * We don't want to deadlock with somebody trying to allocate a new
309 * extent for the extent root while also trying to search the extent
310 * root to add free space. So we skip locking and search the commit
311 * root, since its read-only
313 path
->skip_locking
= 1;
314 path
->search_commit_root
= 1;
319 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
321 mutex_lock(&caching_ctl
->mutex
);
322 /* need to make sure the commit_root doesn't disappear */
323 down_read(&fs_info
->extent_commit_sem
);
325 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
329 leaf
= path
->nodes
[0];
330 nritems
= btrfs_header_nritems(leaf
);
334 if (fs_info
->closing
> 1) {
339 if (path
->slots
[0] < nritems
) {
340 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
342 ret
= find_next_key(path
, 0, &key
);
346 caching_ctl
->progress
= last
;
347 btrfs_release_path(extent_root
, path
);
348 up_read(&fs_info
->extent_commit_sem
);
349 mutex_unlock(&caching_ctl
->mutex
);
350 if (btrfs_transaction_in_commit(fs_info
))
357 if (key
.objectid
< block_group
->key
.objectid
) {
362 if (key
.objectid
>= block_group
->key
.objectid
+
363 block_group
->key
.offset
)
366 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
) {
367 total_found
+= add_new_free_space(block_group
,
370 last
= key
.objectid
+ key
.offset
;
372 if (total_found
> (1024 * 1024 * 2)) {
374 wake_up(&caching_ctl
->wait
);
381 total_found
+= add_new_free_space(block_group
, fs_info
, last
,
382 block_group
->key
.objectid
+
383 block_group
->key
.offset
);
384 caching_ctl
->progress
= (u64
)-1;
386 spin_lock(&block_group
->lock
);
387 block_group
->caching_ctl
= NULL
;
388 block_group
->cached
= BTRFS_CACHE_FINISHED
;
389 spin_unlock(&block_group
->lock
);
392 btrfs_free_path(path
);
393 up_read(&fs_info
->extent_commit_sem
);
395 free_excluded_extents(extent_root
, block_group
);
397 mutex_unlock(&caching_ctl
->mutex
);
398 wake_up(&caching_ctl
->wait
);
400 put_caching_control(caching_ctl
);
401 atomic_dec(&block_group
->space_info
->caching_threads
);
405 static int cache_block_group(struct btrfs_block_group_cache
*cache
)
407 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
408 struct btrfs_caching_control
*caching_ctl
;
409 struct task_struct
*tsk
;
413 if (cache
->cached
!= BTRFS_CACHE_NO
)
416 caching_ctl
= kzalloc(sizeof(*caching_ctl
), GFP_KERNEL
);
417 BUG_ON(!caching_ctl
);
419 INIT_LIST_HEAD(&caching_ctl
->list
);
420 mutex_init(&caching_ctl
->mutex
);
421 init_waitqueue_head(&caching_ctl
->wait
);
422 caching_ctl
->block_group
= cache
;
423 caching_ctl
->progress
= cache
->key
.objectid
;
424 /* one for caching kthread, one for caching block group list */
425 atomic_set(&caching_ctl
->count
, 2);
427 spin_lock(&cache
->lock
);
428 if (cache
->cached
!= BTRFS_CACHE_NO
) {
429 spin_unlock(&cache
->lock
);
433 cache
->caching_ctl
= caching_ctl
;
434 cache
->cached
= BTRFS_CACHE_STARTED
;
435 spin_unlock(&cache
->lock
);
437 down_write(&fs_info
->extent_commit_sem
);
438 list_add_tail(&caching_ctl
->list
, &fs_info
->caching_block_groups
);
439 up_write(&fs_info
->extent_commit_sem
);
441 atomic_inc(&cache
->space_info
->caching_threads
);
443 tsk
= kthread_run(caching_kthread
, cache
, "btrfs-cache-%llu\n",
444 cache
->key
.objectid
);
447 printk(KERN_ERR
"error running thread %d\n", ret
);
455 * return the block group that starts at or after bytenr
457 static struct btrfs_block_group_cache
*
458 btrfs_lookup_first_block_group(struct btrfs_fs_info
*info
, u64 bytenr
)
460 struct btrfs_block_group_cache
*cache
;
462 cache
= block_group_cache_tree_search(info
, bytenr
, 0);
468 * return the block group that contains the given bytenr
470 struct btrfs_block_group_cache
*btrfs_lookup_block_group(
471 struct btrfs_fs_info
*info
,
474 struct btrfs_block_group_cache
*cache
;
476 cache
= block_group_cache_tree_search(info
, bytenr
, 1);
481 void btrfs_put_block_group(struct btrfs_block_group_cache
*cache
)
483 if (atomic_dec_and_test(&cache
->count
))
487 static struct btrfs_space_info
*__find_space_info(struct btrfs_fs_info
*info
,
490 struct list_head
*head
= &info
->space_info
;
491 struct btrfs_space_info
*found
;
494 list_for_each_entry_rcu(found
, head
, list
) {
495 if (found
->flags
== flags
) {
505 * after adding space to the filesystem, we need to clear the full flags
506 * on all the space infos.
508 void btrfs_clear_space_info_full(struct btrfs_fs_info
*info
)
510 struct list_head
*head
= &info
->space_info
;
511 struct btrfs_space_info
*found
;
514 list_for_each_entry_rcu(found
, head
, list
)
519 static u64
div_factor(u64 num
, int factor
)
528 u64
btrfs_find_block_group(struct btrfs_root
*root
,
529 u64 search_start
, u64 search_hint
, int owner
)
531 struct btrfs_block_group_cache
*cache
;
533 u64 last
= max(search_hint
, search_start
);
540 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
544 spin_lock(&cache
->lock
);
545 last
= cache
->key
.objectid
+ cache
->key
.offset
;
546 used
= btrfs_block_group_used(&cache
->item
);
548 if ((full_search
|| !cache
->ro
) &&
549 block_group_bits(cache
, BTRFS_BLOCK_GROUP_METADATA
)) {
550 if (used
+ cache
->pinned
+ cache
->reserved
<
551 div_factor(cache
->key
.offset
, factor
)) {
552 group_start
= cache
->key
.objectid
;
553 spin_unlock(&cache
->lock
);
554 btrfs_put_block_group(cache
);
558 spin_unlock(&cache
->lock
);
559 btrfs_put_block_group(cache
);
567 if (!full_search
&& factor
< 10) {
577 /* simple helper to search for an existing extent at a given offset */
578 int btrfs_lookup_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
581 struct btrfs_key key
;
582 struct btrfs_path
*path
;
584 path
= btrfs_alloc_path();
586 key
.objectid
= start
;
588 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
589 ret
= btrfs_search_slot(NULL
, root
->fs_info
->extent_root
, &key
, path
,
591 btrfs_free_path(path
);
596 * Back reference rules. Back refs have three main goals:
598 * 1) differentiate between all holders of references to an extent so that
599 * when a reference is dropped we can make sure it was a valid reference
600 * before freeing the extent.
602 * 2) Provide enough information to quickly find the holders of an extent
603 * if we notice a given block is corrupted or bad.
605 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
606 * maintenance. This is actually the same as #2, but with a slightly
607 * different use case.
609 * There are two kinds of back refs. The implicit back refs is optimized
610 * for pointers in non-shared tree blocks. For a given pointer in a block,
611 * back refs of this kind provide information about the block's owner tree
612 * and the pointer's key. These information allow us to find the block by
613 * b-tree searching. The full back refs is for pointers in tree blocks not
614 * referenced by their owner trees. The location of tree block is recorded
615 * in the back refs. Actually the full back refs is generic, and can be
616 * used in all cases the implicit back refs is used. The major shortcoming
617 * of the full back refs is its overhead. Every time a tree block gets
618 * COWed, we have to update back refs entry for all pointers in it.
620 * For a newly allocated tree block, we use implicit back refs for
621 * pointers in it. This means most tree related operations only involve
622 * implicit back refs. For a tree block created in old transaction, the
623 * only way to drop a reference to it is COW it. So we can detect the
624 * event that tree block loses its owner tree's reference and do the
625 * back refs conversion.
627 * When a tree block is COW'd through a tree, there are four cases:
629 * The reference count of the block is one and the tree is the block's
630 * owner tree. Nothing to do in this case.
632 * The reference count of the block is one and the tree is not the
633 * block's owner tree. In this case, full back refs is used for pointers
634 * in the block. Remove these full back refs, add implicit back refs for
635 * every pointers in the new block.
637 * The reference count of the block is greater than one and the tree is
638 * the block's owner tree. In this case, implicit back refs is used for
639 * pointers in the block. Add full back refs for every pointers in the
640 * block, increase lower level extents' reference counts. The original
641 * implicit back refs are entailed to the new block.
643 * The reference count of the block is greater than one and the tree is
644 * not the block's owner tree. Add implicit back refs for every pointer in
645 * the new block, increase lower level extents' reference count.
647 * Back Reference Key composing:
649 * The key objectid corresponds to the first byte in the extent,
650 * The key type is used to differentiate between types of back refs.
651 * There are different meanings of the key offset for different types
654 * File extents can be referenced by:
656 * - multiple snapshots, subvolumes, or different generations in one subvol
657 * - different files inside a single subvolume
658 * - different offsets inside a file (bookend extents in file.c)
660 * The extent ref structure for the implicit back refs has fields for:
662 * - Objectid of the subvolume root
663 * - objectid of the file holding the reference
664 * - original offset in the file
665 * - how many bookend extents
667 * The key offset for the implicit back refs is hash of the first
670 * The extent ref structure for the full back refs has field for:
672 * - number of pointers in the tree leaf
674 * The key offset for the implicit back refs is the first byte of
677 * When a file extent is allocated, The implicit back refs is used.
678 * the fields are filled in:
680 * (root_key.objectid, inode objectid, offset in file, 1)
682 * When a file extent is removed file truncation, we find the
683 * corresponding implicit back refs and check the following fields:
685 * (btrfs_header_owner(leaf), inode objectid, offset in file)
687 * Btree extents can be referenced by:
689 * - Different subvolumes
691 * Both the implicit back refs and the full back refs for tree blocks
692 * only consist of key. The key offset for the implicit back refs is
693 * objectid of block's owner tree. The key offset for the full back refs
694 * is the first byte of parent block.
696 * When implicit back refs is used, information about the lowest key and
697 * level of the tree block are required. These information are stored in
698 * tree block info structure.
701 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
702 static int convert_extent_item_v0(struct btrfs_trans_handle
*trans
,
703 struct btrfs_root
*root
,
704 struct btrfs_path
*path
,
705 u64 owner
, u32 extra_size
)
707 struct btrfs_extent_item
*item
;
708 struct btrfs_extent_item_v0
*ei0
;
709 struct btrfs_extent_ref_v0
*ref0
;
710 struct btrfs_tree_block_info
*bi
;
711 struct extent_buffer
*leaf
;
712 struct btrfs_key key
;
713 struct btrfs_key found_key
;
714 u32 new_size
= sizeof(*item
);
718 leaf
= path
->nodes
[0];
719 BUG_ON(btrfs_item_size_nr(leaf
, path
->slots
[0]) != sizeof(*ei0
));
721 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
722 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
723 struct btrfs_extent_item_v0
);
724 refs
= btrfs_extent_refs_v0(leaf
, ei0
);
726 if (owner
== (u64
)-1) {
728 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
729 ret
= btrfs_next_leaf(root
, path
);
733 leaf
= path
->nodes
[0];
735 btrfs_item_key_to_cpu(leaf
, &found_key
,
737 BUG_ON(key
.objectid
!= found_key
.objectid
);
738 if (found_key
.type
!= BTRFS_EXTENT_REF_V0_KEY
) {
742 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
743 struct btrfs_extent_ref_v0
);
744 owner
= btrfs_ref_objectid_v0(leaf
, ref0
);
748 btrfs_release_path(root
, path
);
750 if (owner
< BTRFS_FIRST_FREE_OBJECTID
)
751 new_size
+= sizeof(*bi
);
753 new_size
-= sizeof(*ei0
);
754 ret
= btrfs_search_slot(trans
, root
, &key
, path
,
755 new_size
+ extra_size
, 1);
760 ret
= btrfs_extend_item(trans
, root
, path
, new_size
);
763 leaf
= path
->nodes
[0];
764 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
765 btrfs_set_extent_refs(leaf
, item
, refs
);
766 /* FIXME: get real generation */
767 btrfs_set_extent_generation(leaf
, item
, 0);
768 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
769 btrfs_set_extent_flags(leaf
, item
,
770 BTRFS_EXTENT_FLAG_TREE_BLOCK
|
771 BTRFS_BLOCK_FLAG_FULL_BACKREF
);
772 bi
= (struct btrfs_tree_block_info
*)(item
+ 1);
773 /* FIXME: get first key of the block */
774 memset_extent_buffer(leaf
, 0, (unsigned long)bi
, sizeof(*bi
));
775 btrfs_set_tree_block_level(leaf
, bi
, (int)owner
);
777 btrfs_set_extent_flags(leaf
, item
, BTRFS_EXTENT_FLAG_DATA
);
779 btrfs_mark_buffer_dirty(leaf
);
784 static u64
hash_extent_data_ref(u64 root_objectid
, u64 owner
, u64 offset
)
786 u32 high_crc
= ~(u32
)0;
787 u32 low_crc
= ~(u32
)0;
790 lenum
= cpu_to_le64(root_objectid
);
791 high_crc
= crc32c(high_crc
, &lenum
, sizeof(lenum
));
792 lenum
= cpu_to_le64(owner
);
793 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
794 lenum
= cpu_to_le64(offset
);
795 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
797 return ((u64
)high_crc
<< 31) ^ (u64
)low_crc
;
800 static u64
hash_extent_data_ref_item(struct extent_buffer
*leaf
,
801 struct btrfs_extent_data_ref
*ref
)
803 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf
, ref
),
804 btrfs_extent_data_ref_objectid(leaf
, ref
),
805 btrfs_extent_data_ref_offset(leaf
, ref
));
808 static int match_extent_data_ref(struct extent_buffer
*leaf
,
809 struct btrfs_extent_data_ref
*ref
,
810 u64 root_objectid
, u64 owner
, u64 offset
)
812 if (btrfs_extent_data_ref_root(leaf
, ref
) != root_objectid
||
813 btrfs_extent_data_ref_objectid(leaf
, ref
) != owner
||
814 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
819 static noinline
int lookup_extent_data_ref(struct btrfs_trans_handle
*trans
,
820 struct btrfs_root
*root
,
821 struct btrfs_path
*path
,
822 u64 bytenr
, u64 parent
,
824 u64 owner
, u64 offset
)
826 struct btrfs_key key
;
827 struct btrfs_extent_data_ref
*ref
;
828 struct extent_buffer
*leaf
;
834 key
.objectid
= bytenr
;
836 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
839 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
840 key
.offset
= hash_extent_data_ref(root_objectid
,
845 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
854 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
855 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
856 btrfs_release_path(root
, path
);
857 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
868 leaf
= path
->nodes
[0];
869 nritems
= btrfs_header_nritems(leaf
);
871 if (path
->slots
[0] >= nritems
) {
872 ret
= btrfs_next_leaf(root
, path
);
878 leaf
= path
->nodes
[0];
879 nritems
= btrfs_header_nritems(leaf
);
883 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
884 if (key
.objectid
!= bytenr
||
885 key
.type
!= BTRFS_EXTENT_DATA_REF_KEY
)
888 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
889 struct btrfs_extent_data_ref
);
891 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
894 btrfs_release_path(root
, path
);
906 static noinline
int insert_extent_data_ref(struct btrfs_trans_handle
*trans
,
907 struct btrfs_root
*root
,
908 struct btrfs_path
*path
,
909 u64 bytenr
, u64 parent
,
910 u64 root_objectid
, u64 owner
,
911 u64 offset
, int refs_to_add
)
913 struct btrfs_key key
;
914 struct extent_buffer
*leaf
;
919 key
.objectid
= bytenr
;
921 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
923 size
= sizeof(struct btrfs_shared_data_ref
);
925 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
926 key
.offset
= hash_extent_data_ref(root_objectid
,
928 size
= sizeof(struct btrfs_extent_data_ref
);
931 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, size
);
932 if (ret
&& ret
!= -EEXIST
)
935 leaf
= path
->nodes
[0];
937 struct btrfs_shared_data_ref
*ref
;
938 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
939 struct btrfs_shared_data_ref
);
941 btrfs_set_shared_data_ref_count(leaf
, ref
, refs_to_add
);
943 num_refs
= btrfs_shared_data_ref_count(leaf
, ref
);
944 num_refs
+= refs_to_add
;
945 btrfs_set_shared_data_ref_count(leaf
, ref
, num_refs
);
948 struct btrfs_extent_data_ref
*ref
;
949 while (ret
== -EEXIST
) {
950 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
951 struct btrfs_extent_data_ref
);
952 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
955 btrfs_release_path(root
, path
);
957 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
959 if (ret
&& ret
!= -EEXIST
)
962 leaf
= path
->nodes
[0];
964 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
965 struct btrfs_extent_data_ref
);
967 btrfs_set_extent_data_ref_root(leaf
, ref
,
969 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
970 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
971 btrfs_set_extent_data_ref_count(leaf
, ref
, refs_to_add
);
973 num_refs
= btrfs_extent_data_ref_count(leaf
, ref
);
974 num_refs
+= refs_to_add
;
975 btrfs_set_extent_data_ref_count(leaf
, ref
, num_refs
);
978 btrfs_mark_buffer_dirty(leaf
);
981 btrfs_release_path(root
, path
);
985 static noinline
int remove_extent_data_ref(struct btrfs_trans_handle
*trans
,
986 struct btrfs_root
*root
,
987 struct btrfs_path
*path
,
990 struct btrfs_key key
;
991 struct btrfs_extent_data_ref
*ref1
= NULL
;
992 struct btrfs_shared_data_ref
*ref2
= NULL
;
993 struct extent_buffer
*leaf
;
997 leaf
= path
->nodes
[0];
998 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1000 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1001 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1002 struct btrfs_extent_data_ref
);
1003 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1004 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1005 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1006 struct btrfs_shared_data_ref
);
1007 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1008 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1009 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1010 struct btrfs_extent_ref_v0
*ref0
;
1011 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1012 struct btrfs_extent_ref_v0
);
1013 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1019 BUG_ON(num_refs
< refs_to_drop
);
1020 num_refs
-= refs_to_drop
;
1022 if (num_refs
== 0) {
1023 ret
= btrfs_del_item(trans
, root
, path
);
1025 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
)
1026 btrfs_set_extent_data_ref_count(leaf
, ref1
, num_refs
);
1027 else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
)
1028 btrfs_set_shared_data_ref_count(leaf
, ref2
, num_refs
);
1029 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1031 struct btrfs_extent_ref_v0
*ref0
;
1032 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1033 struct btrfs_extent_ref_v0
);
1034 btrfs_set_ref_count_v0(leaf
, ref0
, num_refs
);
1037 btrfs_mark_buffer_dirty(leaf
);
1042 static noinline u32
extent_data_ref_count(struct btrfs_root
*root
,
1043 struct btrfs_path
*path
,
1044 struct btrfs_extent_inline_ref
*iref
)
1046 struct btrfs_key key
;
1047 struct extent_buffer
*leaf
;
1048 struct btrfs_extent_data_ref
*ref1
;
1049 struct btrfs_shared_data_ref
*ref2
;
1052 leaf
= path
->nodes
[0];
1053 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1055 if (btrfs_extent_inline_ref_type(leaf
, iref
) ==
1056 BTRFS_EXTENT_DATA_REF_KEY
) {
1057 ref1
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1058 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1060 ref2
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1061 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1063 } else if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1064 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1065 struct btrfs_extent_data_ref
);
1066 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1067 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1068 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1069 struct btrfs_shared_data_ref
);
1070 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1071 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1072 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1073 struct btrfs_extent_ref_v0
*ref0
;
1074 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1075 struct btrfs_extent_ref_v0
);
1076 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1084 static noinline
int lookup_tree_block_ref(struct btrfs_trans_handle
*trans
,
1085 struct btrfs_root
*root
,
1086 struct btrfs_path
*path
,
1087 u64 bytenr
, u64 parent
,
1090 struct btrfs_key key
;
1093 key
.objectid
= bytenr
;
1095 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1096 key
.offset
= parent
;
1098 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1099 key
.offset
= root_objectid
;
1102 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1105 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1106 if (ret
== -ENOENT
&& parent
) {
1107 btrfs_release_path(root
, path
);
1108 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1109 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1117 static noinline
int insert_tree_block_ref(struct btrfs_trans_handle
*trans
,
1118 struct btrfs_root
*root
,
1119 struct btrfs_path
*path
,
1120 u64 bytenr
, u64 parent
,
1123 struct btrfs_key key
;
1126 key
.objectid
= bytenr
;
1128 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1129 key
.offset
= parent
;
1131 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1132 key
.offset
= root_objectid
;
1135 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
1136 btrfs_release_path(root
, path
);
1140 static inline int extent_ref_type(u64 parent
, u64 owner
)
1143 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1145 type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1147 type
= BTRFS_TREE_BLOCK_REF_KEY
;
1150 type
= BTRFS_SHARED_DATA_REF_KEY
;
1152 type
= BTRFS_EXTENT_DATA_REF_KEY
;
1157 static int find_next_key(struct btrfs_path
*path
, int level
,
1158 struct btrfs_key
*key
)
1161 for (; level
< BTRFS_MAX_LEVEL
; level
++) {
1162 if (!path
->nodes
[level
])
1164 if (path
->slots
[level
] + 1 >=
1165 btrfs_header_nritems(path
->nodes
[level
]))
1168 btrfs_item_key_to_cpu(path
->nodes
[level
], key
,
1169 path
->slots
[level
] + 1);
1171 btrfs_node_key_to_cpu(path
->nodes
[level
], key
,
1172 path
->slots
[level
] + 1);
1179 * look for inline back ref. if back ref is found, *ref_ret is set
1180 * to the address of inline back ref, and 0 is returned.
1182 * if back ref isn't found, *ref_ret is set to the address where it
1183 * should be inserted, and -ENOENT is returned.
1185 * if insert is true and there are too many inline back refs, the path
1186 * points to the extent item, and -EAGAIN is returned.
1188 * NOTE: inline back refs are ordered in the same way that back ref
1189 * items in the tree are ordered.
1191 static noinline_for_stack
1192 int lookup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1193 struct btrfs_root
*root
,
1194 struct btrfs_path
*path
,
1195 struct btrfs_extent_inline_ref
**ref_ret
,
1196 u64 bytenr
, u64 num_bytes
,
1197 u64 parent
, u64 root_objectid
,
1198 u64 owner
, u64 offset
, int insert
)
1200 struct btrfs_key key
;
1201 struct extent_buffer
*leaf
;
1202 struct btrfs_extent_item
*ei
;
1203 struct btrfs_extent_inline_ref
*iref
;
1214 key
.objectid
= bytenr
;
1215 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1216 key
.offset
= num_bytes
;
1218 want
= extent_ref_type(parent
, owner
);
1220 extra_size
= btrfs_extent_inline_ref_size(want
);
1221 path
->keep_locks
= 1;
1224 ret
= btrfs_search_slot(trans
, root
, &key
, path
, extra_size
, 1);
1231 leaf
= path
->nodes
[0];
1232 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1233 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1234 if (item_size
< sizeof(*ei
)) {
1239 ret
= convert_extent_item_v0(trans
, root
, path
, owner
,
1245 leaf
= path
->nodes
[0];
1246 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1249 BUG_ON(item_size
< sizeof(*ei
));
1251 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1252 flags
= btrfs_extent_flags(leaf
, ei
);
1254 ptr
= (unsigned long)(ei
+ 1);
1255 end
= (unsigned long)ei
+ item_size
;
1257 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
) {
1258 ptr
+= sizeof(struct btrfs_tree_block_info
);
1261 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_DATA
));
1270 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1271 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1275 ptr
+= btrfs_extent_inline_ref_size(type
);
1279 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1280 struct btrfs_extent_data_ref
*dref
;
1281 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1282 if (match_extent_data_ref(leaf
, dref
, root_objectid
,
1287 if (hash_extent_data_ref_item(leaf
, dref
) <
1288 hash_extent_data_ref(root_objectid
, owner
, offset
))
1292 ref_offset
= btrfs_extent_inline_ref_offset(leaf
, iref
);
1294 if (parent
== ref_offset
) {
1298 if (ref_offset
< parent
)
1301 if (root_objectid
== ref_offset
) {
1305 if (ref_offset
< root_objectid
)
1309 ptr
+= btrfs_extent_inline_ref_size(type
);
1311 if (err
== -ENOENT
&& insert
) {
1312 if (item_size
+ extra_size
>=
1313 BTRFS_MAX_EXTENT_ITEM_SIZE(root
)) {
1318 * To add new inline back ref, we have to make sure
1319 * there is no corresponding back ref item.
1320 * For simplicity, we just do not add new inline back
1321 * ref if there is any kind of item for this block
1323 if (find_next_key(path
, 0, &key
) == 0 &&
1324 key
.objectid
== bytenr
&&
1325 key
.type
< BTRFS_BLOCK_GROUP_ITEM_KEY
) {
1330 *ref_ret
= (struct btrfs_extent_inline_ref
*)ptr
;
1333 path
->keep_locks
= 0;
1334 btrfs_unlock_up_safe(path
, 1);
1340 * helper to add new inline back ref
1342 static noinline_for_stack
1343 int setup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1344 struct btrfs_root
*root
,
1345 struct btrfs_path
*path
,
1346 struct btrfs_extent_inline_ref
*iref
,
1347 u64 parent
, u64 root_objectid
,
1348 u64 owner
, u64 offset
, int refs_to_add
,
1349 struct btrfs_delayed_extent_op
*extent_op
)
1351 struct extent_buffer
*leaf
;
1352 struct btrfs_extent_item
*ei
;
1355 unsigned long item_offset
;
1361 leaf
= path
->nodes
[0];
1362 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1363 item_offset
= (unsigned long)iref
- (unsigned long)ei
;
1365 type
= extent_ref_type(parent
, owner
);
1366 size
= btrfs_extent_inline_ref_size(type
);
1368 ret
= btrfs_extend_item(trans
, root
, path
, size
);
1371 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1372 refs
= btrfs_extent_refs(leaf
, ei
);
1373 refs
+= refs_to_add
;
1374 btrfs_set_extent_refs(leaf
, ei
, refs
);
1376 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1378 ptr
= (unsigned long)ei
+ item_offset
;
1379 end
= (unsigned long)ei
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
1380 if (ptr
< end
- size
)
1381 memmove_extent_buffer(leaf
, ptr
+ size
, ptr
,
1384 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1385 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
1386 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1387 struct btrfs_extent_data_ref
*dref
;
1388 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1389 btrfs_set_extent_data_ref_root(leaf
, dref
, root_objectid
);
1390 btrfs_set_extent_data_ref_objectid(leaf
, dref
, owner
);
1391 btrfs_set_extent_data_ref_offset(leaf
, dref
, offset
);
1392 btrfs_set_extent_data_ref_count(leaf
, dref
, refs_to_add
);
1393 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1394 struct btrfs_shared_data_ref
*sref
;
1395 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1396 btrfs_set_shared_data_ref_count(leaf
, sref
, refs_to_add
);
1397 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1398 } else if (type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
1399 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1401 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
1403 btrfs_mark_buffer_dirty(leaf
);
1407 static int lookup_extent_backref(struct btrfs_trans_handle
*trans
,
1408 struct btrfs_root
*root
,
1409 struct btrfs_path
*path
,
1410 struct btrfs_extent_inline_ref
**ref_ret
,
1411 u64 bytenr
, u64 num_bytes
, u64 parent
,
1412 u64 root_objectid
, u64 owner
, u64 offset
)
1416 ret
= lookup_inline_extent_backref(trans
, root
, path
, ref_ret
,
1417 bytenr
, num_bytes
, parent
,
1418 root_objectid
, owner
, offset
, 0);
1422 btrfs_release_path(root
, path
);
1425 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1426 ret
= lookup_tree_block_ref(trans
, root
, path
, bytenr
, parent
,
1429 ret
= lookup_extent_data_ref(trans
, root
, path
, bytenr
, parent
,
1430 root_objectid
, owner
, offset
);
1436 * helper to update/remove inline back ref
1438 static noinline_for_stack
1439 int update_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1440 struct btrfs_root
*root
,
1441 struct btrfs_path
*path
,
1442 struct btrfs_extent_inline_ref
*iref
,
1444 struct btrfs_delayed_extent_op
*extent_op
)
1446 struct extent_buffer
*leaf
;
1447 struct btrfs_extent_item
*ei
;
1448 struct btrfs_extent_data_ref
*dref
= NULL
;
1449 struct btrfs_shared_data_ref
*sref
= NULL
;
1458 leaf
= path
->nodes
[0];
1459 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1460 refs
= btrfs_extent_refs(leaf
, ei
);
1461 WARN_ON(refs_to_mod
< 0 && refs
+ refs_to_mod
<= 0);
1462 refs
+= refs_to_mod
;
1463 btrfs_set_extent_refs(leaf
, ei
, refs
);
1465 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1467 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1469 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1470 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1471 refs
= btrfs_extent_data_ref_count(leaf
, dref
);
1472 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1473 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1474 refs
= btrfs_shared_data_ref_count(leaf
, sref
);
1477 BUG_ON(refs_to_mod
!= -1);
1480 BUG_ON(refs_to_mod
< 0 && refs
< -refs_to_mod
);
1481 refs
+= refs_to_mod
;
1484 if (type
== BTRFS_EXTENT_DATA_REF_KEY
)
1485 btrfs_set_extent_data_ref_count(leaf
, dref
, refs
);
1487 btrfs_set_shared_data_ref_count(leaf
, sref
, refs
);
1489 size
= btrfs_extent_inline_ref_size(type
);
1490 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1491 ptr
= (unsigned long)iref
;
1492 end
= (unsigned long)ei
+ item_size
;
1493 if (ptr
+ size
< end
)
1494 memmove_extent_buffer(leaf
, ptr
, ptr
+ size
,
1497 ret
= btrfs_truncate_item(trans
, root
, path
, item_size
, 1);
1500 btrfs_mark_buffer_dirty(leaf
);
1504 static noinline_for_stack
1505 int insert_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1506 struct btrfs_root
*root
,
1507 struct btrfs_path
*path
,
1508 u64 bytenr
, u64 num_bytes
, u64 parent
,
1509 u64 root_objectid
, u64 owner
,
1510 u64 offset
, int refs_to_add
,
1511 struct btrfs_delayed_extent_op
*extent_op
)
1513 struct btrfs_extent_inline_ref
*iref
;
1516 ret
= lookup_inline_extent_backref(trans
, root
, path
, &iref
,
1517 bytenr
, num_bytes
, parent
,
1518 root_objectid
, owner
, offset
, 1);
1520 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
);
1521 ret
= update_inline_extent_backref(trans
, root
, path
, iref
,
1522 refs_to_add
, extent_op
);
1523 } else if (ret
== -ENOENT
) {
1524 ret
= setup_inline_extent_backref(trans
, root
, path
, iref
,
1525 parent
, root_objectid
,
1526 owner
, offset
, refs_to_add
,
1532 static int insert_extent_backref(struct btrfs_trans_handle
*trans
,
1533 struct btrfs_root
*root
,
1534 struct btrfs_path
*path
,
1535 u64 bytenr
, u64 parent
, u64 root_objectid
,
1536 u64 owner
, u64 offset
, int refs_to_add
)
1539 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1540 BUG_ON(refs_to_add
!= 1);
1541 ret
= insert_tree_block_ref(trans
, root
, path
, bytenr
,
1542 parent
, root_objectid
);
1544 ret
= insert_extent_data_ref(trans
, root
, path
, bytenr
,
1545 parent
, root_objectid
,
1546 owner
, offset
, refs_to_add
);
1551 static int remove_extent_backref(struct btrfs_trans_handle
*trans
,
1552 struct btrfs_root
*root
,
1553 struct btrfs_path
*path
,
1554 struct btrfs_extent_inline_ref
*iref
,
1555 int refs_to_drop
, int is_data
)
1559 BUG_ON(!is_data
&& refs_to_drop
!= 1);
1561 ret
= update_inline_extent_backref(trans
, root
, path
, iref
,
1562 -refs_to_drop
, NULL
);
1563 } else if (is_data
) {
1564 ret
= remove_extent_data_ref(trans
, root
, path
, refs_to_drop
);
1566 ret
= btrfs_del_item(trans
, root
, path
);
1571 #ifdef BIO_RW_DISCARD
1572 static void btrfs_issue_discard(struct block_device
*bdev
,
1575 blkdev_issue_discard(bdev
, start
>> 9, len
>> 9, GFP_KERNEL
);
1579 static int btrfs_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
1582 #ifdef BIO_RW_DISCARD
1584 u64 map_length
= num_bytes
;
1585 struct btrfs_multi_bio
*multi
= NULL
;
1587 /* Tell the block device(s) that the sectors can be discarded */
1588 ret
= btrfs_map_block(&root
->fs_info
->mapping_tree
, READ
,
1589 bytenr
, &map_length
, &multi
, 0);
1591 struct btrfs_bio_stripe
*stripe
= multi
->stripes
;
1594 if (map_length
> num_bytes
)
1595 map_length
= num_bytes
;
1597 for (i
= 0; i
< multi
->num_stripes
; i
++, stripe
++) {
1598 btrfs_issue_discard(stripe
->dev
->bdev
,
1611 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1612 struct btrfs_root
*root
,
1613 u64 bytenr
, u64 num_bytes
, u64 parent
,
1614 u64 root_objectid
, u64 owner
, u64 offset
)
1617 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
&&
1618 root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
1620 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1621 ret
= btrfs_add_delayed_tree_ref(trans
, bytenr
, num_bytes
,
1622 parent
, root_objectid
, (int)owner
,
1623 BTRFS_ADD_DELAYED_REF
, NULL
);
1625 ret
= btrfs_add_delayed_data_ref(trans
, bytenr
, num_bytes
,
1626 parent
, root_objectid
, owner
, offset
,
1627 BTRFS_ADD_DELAYED_REF
, NULL
);
1632 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1633 struct btrfs_root
*root
,
1634 u64 bytenr
, u64 num_bytes
,
1635 u64 parent
, u64 root_objectid
,
1636 u64 owner
, u64 offset
, int refs_to_add
,
1637 struct btrfs_delayed_extent_op
*extent_op
)
1639 struct btrfs_path
*path
;
1640 struct extent_buffer
*leaf
;
1641 struct btrfs_extent_item
*item
;
1646 path
= btrfs_alloc_path();
1651 path
->leave_spinning
= 1;
1652 /* this will setup the path even if it fails to insert the back ref */
1653 ret
= insert_inline_extent_backref(trans
, root
->fs_info
->extent_root
,
1654 path
, bytenr
, num_bytes
, parent
,
1655 root_objectid
, owner
, offset
,
1656 refs_to_add
, extent_op
);
1660 if (ret
!= -EAGAIN
) {
1665 leaf
= path
->nodes
[0];
1666 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1667 refs
= btrfs_extent_refs(leaf
, item
);
1668 btrfs_set_extent_refs(leaf
, item
, refs
+ refs_to_add
);
1670 __run_delayed_extent_op(extent_op
, leaf
, item
);
1672 btrfs_mark_buffer_dirty(leaf
);
1673 btrfs_release_path(root
->fs_info
->extent_root
, path
);
1676 path
->leave_spinning
= 1;
1678 /* now insert the actual backref */
1679 ret
= insert_extent_backref(trans
, root
->fs_info
->extent_root
,
1680 path
, bytenr
, parent
, root_objectid
,
1681 owner
, offset
, refs_to_add
);
1684 btrfs_free_path(path
);
1688 static int run_delayed_data_ref(struct btrfs_trans_handle
*trans
,
1689 struct btrfs_root
*root
,
1690 struct btrfs_delayed_ref_node
*node
,
1691 struct btrfs_delayed_extent_op
*extent_op
,
1692 int insert_reserved
)
1695 struct btrfs_delayed_data_ref
*ref
;
1696 struct btrfs_key ins
;
1701 ins
.objectid
= node
->bytenr
;
1702 ins
.offset
= node
->num_bytes
;
1703 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1705 ref
= btrfs_delayed_node_to_data_ref(node
);
1706 if (node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
1707 parent
= ref
->parent
;
1709 ref_root
= ref
->root
;
1711 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
1713 BUG_ON(extent_op
->update_key
);
1714 flags
|= extent_op
->flags_to_set
;
1716 ret
= alloc_reserved_file_extent(trans
, root
,
1717 parent
, ref_root
, flags
,
1718 ref
->objectid
, ref
->offset
,
1719 &ins
, node
->ref_mod
);
1720 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
1721 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
1722 node
->num_bytes
, parent
,
1723 ref_root
, ref
->objectid
,
1724 ref
->offset
, node
->ref_mod
,
1726 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
1727 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
1728 node
->num_bytes
, parent
,
1729 ref_root
, ref
->objectid
,
1730 ref
->offset
, node
->ref_mod
,
1738 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
1739 struct extent_buffer
*leaf
,
1740 struct btrfs_extent_item
*ei
)
1742 u64 flags
= btrfs_extent_flags(leaf
, ei
);
1743 if (extent_op
->update_flags
) {
1744 flags
|= extent_op
->flags_to_set
;
1745 btrfs_set_extent_flags(leaf
, ei
, flags
);
1748 if (extent_op
->update_key
) {
1749 struct btrfs_tree_block_info
*bi
;
1750 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
));
1751 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
1752 btrfs_set_tree_block_key(leaf
, bi
, &extent_op
->key
);
1756 static int run_delayed_extent_op(struct btrfs_trans_handle
*trans
,
1757 struct btrfs_root
*root
,
1758 struct btrfs_delayed_ref_node
*node
,
1759 struct btrfs_delayed_extent_op
*extent_op
)
1761 struct btrfs_key key
;
1762 struct btrfs_path
*path
;
1763 struct btrfs_extent_item
*ei
;
1764 struct extent_buffer
*leaf
;
1769 path
= btrfs_alloc_path();
1773 key
.objectid
= node
->bytenr
;
1774 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1775 key
.offset
= node
->num_bytes
;
1778 path
->leave_spinning
= 1;
1779 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
,
1790 leaf
= path
->nodes
[0];
1791 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1792 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1793 if (item_size
< sizeof(*ei
)) {
1794 ret
= convert_extent_item_v0(trans
, root
->fs_info
->extent_root
,
1800 leaf
= path
->nodes
[0];
1801 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1804 BUG_ON(item_size
< sizeof(*ei
));
1805 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1806 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1808 btrfs_mark_buffer_dirty(leaf
);
1810 btrfs_free_path(path
);
1814 static int run_delayed_tree_ref(struct btrfs_trans_handle
*trans
,
1815 struct btrfs_root
*root
,
1816 struct btrfs_delayed_ref_node
*node
,
1817 struct btrfs_delayed_extent_op
*extent_op
,
1818 int insert_reserved
)
1821 struct btrfs_delayed_tree_ref
*ref
;
1822 struct btrfs_key ins
;
1826 ins
.objectid
= node
->bytenr
;
1827 ins
.offset
= node
->num_bytes
;
1828 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1830 ref
= btrfs_delayed_node_to_tree_ref(node
);
1831 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
1832 parent
= ref
->parent
;
1834 ref_root
= ref
->root
;
1836 BUG_ON(node
->ref_mod
!= 1);
1837 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
1838 BUG_ON(!extent_op
|| !extent_op
->update_flags
||
1839 !extent_op
->update_key
);
1840 ret
= alloc_reserved_tree_block(trans
, root
,
1842 extent_op
->flags_to_set
,
1845 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
1846 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
1847 node
->num_bytes
, parent
, ref_root
,
1848 ref
->level
, 0, 1, extent_op
);
1849 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
1850 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
1851 node
->num_bytes
, parent
, ref_root
,
1852 ref
->level
, 0, 1, extent_op
);
1860 /* helper function to actually process a single delayed ref entry */
1861 static int run_one_delayed_ref(struct btrfs_trans_handle
*trans
,
1862 struct btrfs_root
*root
,
1863 struct btrfs_delayed_ref_node
*node
,
1864 struct btrfs_delayed_extent_op
*extent_op
,
1865 int insert_reserved
)
1868 if (btrfs_delayed_ref_is_head(node
)) {
1869 struct btrfs_delayed_ref_head
*head
;
1871 * we've hit the end of the chain and we were supposed
1872 * to insert this extent into the tree. But, it got
1873 * deleted before we ever needed to insert it, so all
1874 * we have to do is clean up the accounting
1877 head
= btrfs_delayed_node_to_head(node
);
1878 if (insert_reserved
) {
1880 struct extent_buffer
*must_clean
= NULL
;
1882 ret
= pin_down_bytes(trans
, root
, NULL
,
1883 node
->bytenr
, node
->num_bytes
,
1884 head
->is_data
, 1, &must_clean
);
1889 clean_tree_block(NULL
, root
, must_clean
);
1890 btrfs_tree_unlock(must_clean
);
1891 free_extent_buffer(must_clean
);
1893 if (head
->is_data
) {
1894 ret
= btrfs_del_csums(trans
, root
,
1900 ret
= btrfs_free_reserved_extent(root
,
1906 mutex_unlock(&head
->mutex
);
1910 if (node
->type
== BTRFS_TREE_BLOCK_REF_KEY
||
1911 node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
1912 ret
= run_delayed_tree_ref(trans
, root
, node
, extent_op
,
1914 else if (node
->type
== BTRFS_EXTENT_DATA_REF_KEY
||
1915 node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
1916 ret
= run_delayed_data_ref(trans
, root
, node
, extent_op
,
1923 static noinline
struct btrfs_delayed_ref_node
*
1924 select_delayed_ref(struct btrfs_delayed_ref_head
*head
)
1926 struct rb_node
*node
;
1927 struct btrfs_delayed_ref_node
*ref
;
1928 int action
= BTRFS_ADD_DELAYED_REF
;
1931 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
1932 * this prevents ref count from going down to zero when
1933 * there still are pending delayed ref.
1935 node
= rb_prev(&head
->node
.rb_node
);
1939 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
1941 if (ref
->bytenr
!= head
->node
.bytenr
)
1943 if (ref
->action
== action
)
1945 node
= rb_prev(node
);
1947 if (action
== BTRFS_ADD_DELAYED_REF
) {
1948 action
= BTRFS_DROP_DELAYED_REF
;
1954 static noinline
int run_clustered_refs(struct btrfs_trans_handle
*trans
,
1955 struct btrfs_root
*root
,
1956 struct list_head
*cluster
)
1958 struct btrfs_delayed_ref_root
*delayed_refs
;
1959 struct btrfs_delayed_ref_node
*ref
;
1960 struct btrfs_delayed_ref_head
*locked_ref
= NULL
;
1961 struct btrfs_delayed_extent_op
*extent_op
;
1964 int must_insert_reserved
= 0;
1966 delayed_refs
= &trans
->transaction
->delayed_refs
;
1969 /* pick a new head ref from the cluster list */
1970 if (list_empty(cluster
))
1973 locked_ref
= list_entry(cluster
->next
,
1974 struct btrfs_delayed_ref_head
, cluster
);
1976 /* grab the lock that says we are going to process
1977 * all the refs for this head */
1978 ret
= btrfs_delayed_ref_lock(trans
, locked_ref
);
1981 * we may have dropped the spin lock to get the head
1982 * mutex lock, and that might have given someone else
1983 * time to free the head. If that's true, it has been
1984 * removed from our list and we can move on.
1986 if (ret
== -EAGAIN
) {
1994 * record the must insert reserved flag before we
1995 * drop the spin lock.
1997 must_insert_reserved
= locked_ref
->must_insert_reserved
;
1998 locked_ref
->must_insert_reserved
= 0;
2000 extent_op
= locked_ref
->extent_op
;
2001 locked_ref
->extent_op
= NULL
;
2004 * locked_ref is the head node, so we have to go one
2005 * node back for any delayed ref updates
2007 ref
= select_delayed_ref(locked_ref
);
2009 /* All delayed refs have been processed, Go ahead
2010 * and send the head node to run_one_delayed_ref,
2011 * so that any accounting fixes can happen
2013 ref
= &locked_ref
->node
;
2015 if (extent_op
&& must_insert_reserved
) {
2021 spin_unlock(&delayed_refs
->lock
);
2023 ret
= run_delayed_extent_op(trans
, root
,
2029 spin_lock(&delayed_refs
->lock
);
2033 list_del_init(&locked_ref
->cluster
);
2038 rb_erase(&ref
->rb_node
, &delayed_refs
->root
);
2039 delayed_refs
->num_entries
--;
2041 spin_unlock(&delayed_refs
->lock
);
2043 ret
= run_one_delayed_ref(trans
, root
, ref
, extent_op
,
2044 must_insert_reserved
);
2047 btrfs_put_delayed_ref(ref
);
2052 spin_lock(&delayed_refs
->lock
);
2058 * this starts processing the delayed reference count updates and
2059 * extent insertions we have queued up so far. count can be
2060 * 0, which means to process everything in the tree at the start
2061 * of the run (but not newly added entries), or it can be some target
2062 * number you'd like to process.
2064 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2065 struct btrfs_root
*root
, unsigned long count
)
2067 struct rb_node
*node
;
2068 struct btrfs_delayed_ref_root
*delayed_refs
;
2069 struct btrfs_delayed_ref_node
*ref
;
2070 struct list_head cluster
;
2072 int run_all
= count
== (unsigned long)-1;
2075 if (root
== root
->fs_info
->extent_root
)
2076 root
= root
->fs_info
->tree_root
;
2078 delayed_refs
= &trans
->transaction
->delayed_refs
;
2079 INIT_LIST_HEAD(&cluster
);
2081 spin_lock(&delayed_refs
->lock
);
2083 count
= delayed_refs
->num_entries
* 2;
2087 if (!(run_all
|| run_most
) &&
2088 delayed_refs
->num_heads_ready
< 64)
2092 * go find something we can process in the rbtree. We start at
2093 * the beginning of the tree, and then build a cluster
2094 * of refs to process starting at the first one we are able to
2097 ret
= btrfs_find_ref_cluster(trans
, &cluster
,
2098 delayed_refs
->run_delayed_start
);
2102 ret
= run_clustered_refs(trans
, root
, &cluster
);
2105 count
-= min_t(unsigned long, ret
, count
);
2112 node
= rb_first(&delayed_refs
->root
);
2115 count
= (unsigned long)-1;
2118 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
2120 if (btrfs_delayed_ref_is_head(ref
)) {
2121 struct btrfs_delayed_ref_head
*head
;
2123 head
= btrfs_delayed_node_to_head(ref
);
2124 atomic_inc(&ref
->refs
);
2126 spin_unlock(&delayed_refs
->lock
);
2127 mutex_lock(&head
->mutex
);
2128 mutex_unlock(&head
->mutex
);
2130 btrfs_put_delayed_ref(ref
);
2134 node
= rb_next(node
);
2136 spin_unlock(&delayed_refs
->lock
);
2137 schedule_timeout(1);
2141 spin_unlock(&delayed_refs
->lock
);
2145 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
2146 struct btrfs_root
*root
,
2147 u64 bytenr
, u64 num_bytes
, u64 flags
,
2150 struct btrfs_delayed_extent_op
*extent_op
;
2153 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
2157 extent_op
->flags_to_set
= flags
;
2158 extent_op
->update_flags
= 1;
2159 extent_op
->update_key
= 0;
2160 extent_op
->is_data
= is_data
? 1 : 0;
2162 ret
= btrfs_add_delayed_extent_op(trans
, bytenr
, num_bytes
, extent_op
);
2168 static noinline
int check_delayed_ref(struct btrfs_trans_handle
*trans
,
2169 struct btrfs_root
*root
,
2170 struct btrfs_path
*path
,
2171 u64 objectid
, u64 offset
, u64 bytenr
)
2173 struct btrfs_delayed_ref_head
*head
;
2174 struct btrfs_delayed_ref_node
*ref
;
2175 struct btrfs_delayed_data_ref
*data_ref
;
2176 struct btrfs_delayed_ref_root
*delayed_refs
;
2177 struct rb_node
*node
;
2181 delayed_refs
= &trans
->transaction
->delayed_refs
;
2182 spin_lock(&delayed_refs
->lock
);
2183 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
2187 if (!mutex_trylock(&head
->mutex
)) {
2188 atomic_inc(&head
->node
.refs
);
2189 spin_unlock(&delayed_refs
->lock
);
2191 btrfs_release_path(root
->fs_info
->extent_root
, path
);
2193 mutex_lock(&head
->mutex
);
2194 mutex_unlock(&head
->mutex
);
2195 btrfs_put_delayed_ref(&head
->node
);
2199 node
= rb_prev(&head
->node
.rb_node
);
2203 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2205 if (ref
->bytenr
!= bytenr
)
2209 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
)
2212 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
2214 node
= rb_prev(node
);
2216 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2217 if (ref
->bytenr
== bytenr
)
2221 if (data_ref
->root
!= root
->root_key
.objectid
||
2222 data_ref
->objectid
!= objectid
|| data_ref
->offset
!= offset
)
2227 mutex_unlock(&head
->mutex
);
2229 spin_unlock(&delayed_refs
->lock
);
2233 static noinline
int check_committed_ref(struct btrfs_trans_handle
*trans
,
2234 struct btrfs_root
*root
,
2235 struct btrfs_path
*path
,
2236 u64 objectid
, u64 offset
, u64 bytenr
)
2238 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2239 struct extent_buffer
*leaf
;
2240 struct btrfs_extent_data_ref
*ref
;
2241 struct btrfs_extent_inline_ref
*iref
;
2242 struct btrfs_extent_item
*ei
;
2243 struct btrfs_key key
;
2247 key
.objectid
= bytenr
;
2248 key
.offset
= (u64
)-1;
2249 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2251 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
2257 if (path
->slots
[0] == 0)
2261 leaf
= path
->nodes
[0];
2262 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2264 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
2268 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2269 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2270 if (item_size
< sizeof(*ei
)) {
2271 WARN_ON(item_size
!= sizeof(struct btrfs_extent_item_v0
));
2275 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2277 if (item_size
!= sizeof(*ei
) +
2278 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
2281 if (btrfs_extent_generation(leaf
, ei
) <=
2282 btrfs_root_last_snapshot(&root
->root_item
))
2285 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
2286 if (btrfs_extent_inline_ref_type(leaf
, iref
) !=
2287 BTRFS_EXTENT_DATA_REF_KEY
)
2290 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
2291 if (btrfs_extent_refs(leaf
, ei
) !=
2292 btrfs_extent_data_ref_count(leaf
, ref
) ||
2293 btrfs_extent_data_ref_root(leaf
, ref
) !=
2294 root
->root_key
.objectid
||
2295 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
2296 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
2304 int btrfs_cross_ref_exist(struct btrfs_trans_handle
*trans
,
2305 struct btrfs_root
*root
,
2306 u64 objectid
, u64 offset
, u64 bytenr
)
2308 struct btrfs_path
*path
;
2312 path
= btrfs_alloc_path();
2317 ret
= check_committed_ref(trans
, root
, path
, objectid
,
2319 if (ret
&& ret
!= -ENOENT
)
2322 ret2
= check_delayed_ref(trans
, root
, path
, objectid
,
2324 } while (ret2
== -EAGAIN
);
2326 if (ret2
&& ret2
!= -ENOENT
) {
2331 if (ret
!= -ENOENT
|| ret2
!= -ENOENT
)
2334 btrfs_free_path(path
);
2339 int btrfs_cache_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2340 struct extent_buffer
*buf
, u32 nr_extents
)
2342 struct btrfs_key key
;
2343 struct btrfs_file_extent_item
*fi
;
2351 if (!root
->ref_cows
)
2354 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
2356 root_gen
= root
->root_key
.offset
;
2359 root_gen
= trans
->transid
- 1;
2362 level
= btrfs_header_level(buf
);
2363 nritems
= btrfs_header_nritems(buf
);
2366 struct btrfs_leaf_ref
*ref
;
2367 struct btrfs_extent_info
*info
;
2369 ref
= btrfs_alloc_leaf_ref(root
, nr_extents
);
2375 ref
->root_gen
= root_gen
;
2376 ref
->bytenr
= buf
->start
;
2377 ref
->owner
= btrfs_header_owner(buf
);
2378 ref
->generation
= btrfs_header_generation(buf
);
2379 ref
->nritems
= nr_extents
;
2380 info
= ref
->extents
;
2382 for (i
= 0; nr_extents
> 0 && i
< nritems
; i
++) {
2384 btrfs_item_key_to_cpu(buf
, &key
, i
);
2385 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
2387 fi
= btrfs_item_ptr(buf
, i
,
2388 struct btrfs_file_extent_item
);
2389 if (btrfs_file_extent_type(buf
, fi
) ==
2390 BTRFS_FILE_EXTENT_INLINE
)
2392 disk_bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
2393 if (disk_bytenr
== 0)
2396 info
->bytenr
= disk_bytenr
;
2398 btrfs_file_extent_disk_num_bytes(buf
, fi
);
2399 info
->objectid
= key
.objectid
;
2400 info
->offset
= key
.offset
;
2404 ret
= btrfs_add_leaf_ref(root
, ref
, shared
);
2405 if (ret
== -EEXIST
&& shared
) {
2406 struct btrfs_leaf_ref
*old
;
2407 old
= btrfs_lookup_leaf_ref(root
, ref
->bytenr
);
2409 btrfs_remove_leaf_ref(root
, old
);
2410 btrfs_free_leaf_ref(root
, old
);
2411 ret
= btrfs_add_leaf_ref(root
, ref
, shared
);
2414 btrfs_free_leaf_ref(root
, ref
);
2420 /* when a block goes through cow, we update the reference counts of
2421 * everything that block points to. The internal pointers of the block
2422 * can be in just about any order, and it is likely to have clusters of
2423 * things that are close together and clusters of things that are not.
2425 * To help reduce the seeks that come with updating all of these reference
2426 * counts, sort them by byte number before actual updates are done.
2428 * struct refsort is used to match byte number to slot in the btree block.
2429 * we sort based on the byte number and then use the slot to actually
2432 * struct refsort is smaller than strcut btrfs_item and smaller than
2433 * struct btrfs_key_ptr. Since we're currently limited to the page size
2434 * for a btree block, there's no way for a kmalloc of refsorts for a
2435 * single node to be bigger than a page.
2443 * for passing into sort()
2445 static int refsort_cmp(const void *a_void
, const void *b_void
)
2447 const struct refsort
*a
= a_void
;
2448 const struct refsort
*b
= b_void
;
2450 if (a
->bytenr
< b
->bytenr
)
2452 if (a
->bytenr
> b
->bytenr
)
2458 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
2459 struct btrfs_root
*root
,
2460 struct extent_buffer
*buf
,
2461 int full_backref
, int inc
)
2468 struct btrfs_key key
;
2469 struct btrfs_file_extent_item
*fi
;
2473 int (*process_func
)(struct btrfs_trans_handle
*, struct btrfs_root
*,
2474 u64
, u64
, u64
, u64
, u64
, u64
);
2476 ref_root
= btrfs_header_owner(buf
);
2477 nritems
= btrfs_header_nritems(buf
);
2478 level
= btrfs_header_level(buf
);
2480 if (!root
->ref_cows
&& level
== 0)
2484 process_func
= btrfs_inc_extent_ref
;
2486 process_func
= btrfs_free_extent
;
2489 parent
= buf
->start
;
2493 for (i
= 0; i
< nritems
; i
++) {
2495 btrfs_item_key_to_cpu(buf
, &key
, i
);
2496 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
2498 fi
= btrfs_item_ptr(buf
, i
,
2499 struct btrfs_file_extent_item
);
2500 if (btrfs_file_extent_type(buf
, fi
) ==
2501 BTRFS_FILE_EXTENT_INLINE
)
2503 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
2507 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
2508 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
2509 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
2510 parent
, ref_root
, key
.objectid
,
2515 bytenr
= btrfs_node_blockptr(buf
, i
);
2516 num_bytes
= btrfs_level_size(root
, level
- 1);
2517 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
2518 parent
, ref_root
, level
- 1, 0);
2529 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2530 struct extent_buffer
*buf
, int full_backref
)
2532 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1);
2535 int btrfs_dec_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2536 struct extent_buffer
*buf
, int full_backref
)
2538 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 0);
2541 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
2542 struct btrfs_root
*root
,
2543 struct btrfs_path
*path
,
2544 struct btrfs_block_group_cache
*cache
)
2547 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2549 struct extent_buffer
*leaf
;
2551 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
2556 leaf
= path
->nodes
[0];
2557 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
2558 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
2559 btrfs_mark_buffer_dirty(leaf
);
2560 btrfs_release_path(extent_root
, path
);
2568 static struct btrfs_block_group_cache
*
2569 next_block_group(struct btrfs_root
*root
,
2570 struct btrfs_block_group_cache
*cache
)
2572 struct rb_node
*node
;
2573 spin_lock(&root
->fs_info
->block_group_cache_lock
);
2574 node
= rb_next(&cache
->cache_node
);
2575 btrfs_put_block_group(cache
);
2577 cache
= rb_entry(node
, struct btrfs_block_group_cache
,
2579 atomic_inc(&cache
->count
);
2582 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
2586 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
2587 struct btrfs_root
*root
)
2589 struct btrfs_block_group_cache
*cache
;
2591 struct btrfs_path
*path
;
2594 path
= btrfs_alloc_path();
2600 err
= btrfs_run_delayed_refs(trans
, root
,
2605 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
2609 cache
= next_block_group(root
, cache
);
2619 last
= cache
->key
.objectid
+ cache
->key
.offset
;
2621 err
= write_one_cache_group(trans
, root
, path
, cache
);
2623 btrfs_put_block_group(cache
);
2626 btrfs_free_path(path
);
2630 int btrfs_extent_readonly(struct btrfs_root
*root
, u64 bytenr
)
2632 struct btrfs_block_group_cache
*block_group
;
2635 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
2636 if (!block_group
|| block_group
->ro
)
2639 btrfs_put_block_group(block_group
);
2643 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
2644 u64 total_bytes
, u64 bytes_used
,
2645 struct btrfs_space_info
**space_info
)
2647 struct btrfs_space_info
*found
;
2649 found
= __find_space_info(info
, flags
);
2651 spin_lock(&found
->lock
);
2652 found
->total_bytes
+= total_bytes
;
2653 found
->bytes_used
+= bytes_used
;
2655 spin_unlock(&found
->lock
);
2656 *space_info
= found
;
2659 found
= kzalloc(sizeof(*found
), GFP_NOFS
);
2663 INIT_LIST_HEAD(&found
->block_groups
);
2664 init_rwsem(&found
->groups_sem
);
2665 spin_lock_init(&found
->lock
);
2666 found
->flags
= flags
;
2667 found
->total_bytes
= total_bytes
;
2668 found
->bytes_used
= bytes_used
;
2669 found
->bytes_pinned
= 0;
2670 found
->bytes_reserved
= 0;
2671 found
->bytes_readonly
= 0;
2672 found
->bytes_delalloc
= 0;
2674 found
->force_alloc
= 0;
2675 *space_info
= found
;
2676 list_add_rcu(&found
->list
, &info
->space_info
);
2677 atomic_set(&found
->caching_threads
, 0);
2681 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
2683 u64 extra_flags
= flags
& (BTRFS_BLOCK_GROUP_RAID0
|
2684 BTRFS_BLOCK_GROUP_RAID1
|
2685 BTRFS_BLOCK_GROUP_RAID10
|
2686 BTRFS_BLOCK_GROUP_DUP
);
2688 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
2689 fs_info
->avail_data_alloc_bits
|= extra_flags
;
2690 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
2691 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
2692 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
2693 fs_info
->avail_system_alloc_bits
|= extra_flags
;
2697 static void set_block_group_readonly(struct btrfs_block_group_cache
*cache
)
2699 spin_lock(&cache
->space_info
->lock
);
2700 spin_lock(&cache
->lock
);
2702 cache
->space_info
->bytes_readonly
+= cache
->key
.offset
-
2703 btrfs_block_group_used(&cache
->item
);
2706 spin_unlock(&cache
->lock
);
2707 spin_unlock(&cache
->space_info
->lock
);
2710 u64
btrfs_reduce_alloc_profile(struct btrfs_root
*root
, u64 flags
)
2712 u64 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
2714 if (num_devices
== 1)
2715 flags
&= ~(BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID0
);
2716 if (num_devices
< 4)
2717 flags
&= ~BTRFS_BLOCK_GROUP_RAID10
;
2719 if ((flags
& BTRFS_BLOCK_GROUP_DUP
) &&
2720 (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
2721 BTRFS_BLOCK_GROUP_RAID10
))) {
2722 flags
&= ~BTRFS_BLOCK_GROUP_DUP
;
2725 if ((flags
& BTRFS_BLOCK_GROUP_RAID1
) &&
2726 (flags
& BTRFS_BLOCK_GROUP_RAID10
)) {
2727 flags
&= ~BTRFS_BLOCK_GROUP_RAID1
;
2730 if ((flags
& BTRFS_BLOCK_GROUP_RAID0
) &&
2731 ((flags
& BTRFS_BLOCK_GROUP_RAID1
) |
2732 (flags
& BTRFS_BLOCK_GROUP_RAID10
) |
2733 (flags
& BTRFS_BLOCK_GROUP_DUP
)))
2734 flags
&= ~BTRFS_BLOCK_GROUP_RAID0
;
2738 static u64
btrfs_get_alloc_profile(struct btrfs_root
*root
, u64 data
)
2740 struct btrfs_fs_info
*info
= root
->fs_info
;
2744 alloc_profile
= info
->avail_data_alloc_bits
&
2745 info
->data_alloc_profile
;
2746 data
= BTRFS_BLOCK_GROUP_DATA
| alloc_profile
;
2747 } else if (root
== root
->fs_info
->chunk_root
) {
2748 alloc_profile
= info
->avail_system_alloc_bits
&
2749 info
->system_alloc_profile
;
2750 data
= BTRFS_BLOCK_GROUP_SYSTEM
| alloc_profile
;
2752 alloc_profile
= info
->avail_metadata_alloc_bits
&
2753 info
->metadata_alloc_profile
;
2754 data
= BTRFS_BLOCK_GROUP_METADATA
| alloc_profile
;
2757 return btrfs_reduce_alloc_profile(root
, data
);
2760 void btrfs_set_inode_space_info(struct btrfs_root
*root
, struct inode
*inode
)
2764 alloc_target
= btrfs_get_alloc_profile(root
, 1);
2765 BTRFS_I(inode
)->space_info
= __find_space_info(root
->fs_info
,
2769 static u64
calculate_bytes_needed(struct btrfs_root
*root
, int num_items
)
2774 level
= BTRFS_MAX_LEVEL
- 2;
2776 * NOTE: these calculations are absolutely the worst possible case.
2777 * This assumes that _every_ item we insert will require a new leaf, and
2778 * that the tree has grown to its maximum level size.
2782 * for every item we insert we could insert both an extent item and a
2783 * extent ref item. Then for ever item we insert, we will need to cow
2784 * both the original leaf, plus the leaf to the left and right of it.
2786 * Unless we are talking about the extent root, then we just want the
2787 * number of items * 2, since we just need the extent item plus its ref.
2789 if (root
== root
->fs_info
->extent_root
)
2790 num_bytes
= num_items
* 2;
2792 num_bytes
= (num_items
+ (2 * num_items
)) * 3;
2795 * num_bytes is total number of leaves we could need times the leaf
2796 * size, and then for every leaf we could end up cow'ing 2 nodes per
2797 * level, down to the leaf level.
2799 num_bytes
= (num_bytes
* root
->leafsize
) +
2800 (num_bytes
* (level
* 2)) * root
->nodesize
;
2806 * Unreserve metadata space for delalloc. If we have less reserved credits than
2807 * we have extents, this function does nothing.
2809 int btrfs_unreserve_metadata_for_delalloc(struct btrfs_root
*root
,
2810 struct inode
*inode
, int num_items
)
2812 struct btrfs_fs_info
*info
= root
->fs_info
;
2813 struct btrfs_space_info
*meta_sinfo
;
2818 /* get the space info for where the metadata will live */
2819 alloc_target
= btrfs_get_alloc_profile(root
, 0);
2820 meta_sinfo
= __find_space_info(info
, alloc_target
);
2822 num_bytes
= calculate_bytes_needed(root
->fs_info
->extent_root
,
2825 spin_lock(&meta_sinfo
->lock
);
2826 spin_lock(&BTRFS_I(inode
)->accounting_lock
);
2827 if (BTRFS_I(inode
)->reserved_extents
<=
2828 BTRFS_I(inode
)->outstanding_extents
) {
2829 spin_unlock(&BTRFS_I(inode
)->accounting_lock
);
2830 spin_unlock(&meta_sinfo
->lock
);
2833 spin_unlock(&BTRFS_I(inode
)->accounting_lock
);
2835 BTRFS_I(inode
)->reserved_extents
--;
2836 BUG_ON(BTRFS_I(inode
)->reserved_extents
< 0);
2838 if (meta_sinfo
->bytes_delalloc
< num_bytes
) {
2840 meta_sinfo
->bytes_delalloc
= 0;
2842 meta_sinfo
->bytes_delalloc
-= num_bytes
;
2844 spin_unlock(&meta_sinfo
->lock
);
2851 static void check_force_delalloc(struct btrfs_space_info
*meta_sinfo
)
2855 thresh
= meta_sinfo
->bytes_used
+ meta_sinfo
->bytes_reserved
+
2856 meta_sinfo
->bytes_pinned
+ meta_sinfo
->bytes_readonly
+
2857 meta_sinfo
->bytes_super
+ meta_sinfo
->bytes_root
+
2858 meta_sinfo
->bytes_may_use
;
2860 thresh
= meta_sinfo
->total_bytes
- thresh
;
2862 do_div(thresh
, 100);
2863 if (thresh
<= meta_sinfo
->bytes_delalloc
)
2864 meta_sinfo
->force_delalloc
= 1;
2866 meta_sinfo
->force_delalloc
= 0;
2869 struct async_flush
{
2870 struct btrfs_root
*root
;
2871 struct btrfs_space_info
*info
;
2872 struct btrfs_work work
;
2875 static noinline
void flush_delalloc_async(struct btrfs_work
*work
)
2877 struct async_flush
*async
;
2878 struct btrfs_root
*root
;
2879 struct btrfs_space_info
*info
;
2881 async
= container_of(work
, struct async_flush
, work
);
2885 btrfs_start_delalloc_inodes(root
);
2886 wake_up(&info
->flush_wait
);
2887 btrfs_wait_ordered_extents(root
, 0);
2889 spin_lock(&info
->lock
);
2891 spin_unlock(&info
->lock
);
2892 wake_up(&info
->flush_wait
);
2897 static void wait_on_flush(struct btrfs_space_info
*info
)
2903 prepare_to_wait(&info
->flush_wait
, &wait
,
2904 TASK_UNINTERRUPTIBLE
);
2905 spin_lock(&info
->lock
);
2906 if (!info
->flushing
) {
2907 spin_unlock(&info
->lock
);
2911 used
= info
->bytes_used
+ info
->bytes_reserved
+
2912 info
->bytes_pinned
+ info
->bytes_readonly
+
2913 info
->bytes_super
+ info
->bytes_root
+
2914 info
->bytes_may_use
+ info
->bytes_delalloc
;
2915 if (used
< info
->total_bytes
) {
2916 spin_unlock(&info
->lock
);
2919 spin_unlock(&info
->lock
);
2922 finish_wait(&info
->flush_wait
, &wait
);
2925 static void flush_delalloc(struct btrfs_root
*root
,
2926 struct btrfs_space_info
*info
)
2928 struct async_flush
*async
;
2931 spin_lock(&info
->lock
);
2933 if (!info
->flushing
) {
2935 init_waitqueue_head(&info
->flush_wait
);
2940 spin_unlock(&info
->lock
);
2943 wait_on_flush(info
);
2947 async
= kzalloc(sizeof(*async
), GFP_NOFS
);
2953 async
->work
.func
= flush_delalloc_async
;
2955 btrfs_queue_worker(&root
->fs_info
->enospc_workers
,
2957 wait_on_flush(info
);
2961 btrfs_start_delalloc_inodes(root
);
2962 btrfs_wait_ordered_extents(root
, 0);
2964 spin_lock(&info
->lock
);
2966 spin_unlock(&info
->lock
);
2967 wake_up(&info
->flush_wait
);
2970 static int maybe_allocate_chunk(struct btrfs_root
*root
,
2971 struct btrfs_space_info
*info
)
2973 struct btrfs_super_block
*disk_super
= &root
->fs_info
->super_copy
;
2974 struct btrfs_trans_handle
*trans
;
2980 free_space
= btrfs_super_total_bytes(disk_super
);
2982 * we allow the metadata to grow to a max of either 5gb or 5% of the
2983 * space in the volume.
2985 min_metadata
= min((u64
)5 * 1024 * 1024 * 1024,
2986 div64_u64(free_space
* 5, 100));
2987 if (info
->total_bytes
>= min_metadata
) {
2988 spin_unlock(&info
->lock
);
2993 spin_unlock(&info
->lock
);
2997 if (!info
->allocating_chunk
) {
2998 info
->force_alloc
= 1;
2999 info
->allocating_chunk
= 1;
3000 init_waitqueue_head(&info
->allocate_wait
);
3005 spin_unlock(&info
->lock
);
3008 wait_event(info
->allocate_wait
,
3009 !info
->allocating_chunk
);
3013 trans
= btrfs_start_transaction(root
, 1);
3019 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
3020 4096 + 2 * 1024 * 1024,
3022 btrfs_end_transaction(trans
, root
);
3026 spin_lock(&info
->lock
);
3027 info
->allocating_chunk
= 0;
3028 spin_unlock(&info
->lock
);
3029 wake_up(&info
->allocate_wait
);
3037 * Reserve metadata space for delalloc.
3039 int btrfs_reserve_metadata_for_delalloc(struct btrfs_root
*root
,
3040 struct inode
*inode
, int num_items
)
3042 struct btrfs_fs_info
*info
= root
->fs_info
;
3043 struct btrfs_space_info
*meta_sinfo
;
3050 /* get the space info for where the metadata will live */
3051 alloc_target
= btrfs_get_alloc_profile(root
, 0);
3052 meta_sinfo
= __find_space_info(info
, alloc_target
);
3054 num_bytes
= calculate_bytes_needed(root
->fs_info
->extent_root
,
3057 spin_lock(&meta_sinfo
->lock
);
3059 force_delalloc
= meta_sinfo
->force_delalloc
;
3061 if (unlikely(!meta_sinfo
->bytes_root
))
3062 meta_sinfo
->bytes_root
= calculate_bytes_needed(root
, 6);
3065 meta_sinfo
->bytes_delalloc
+= num_bytes
;
3067 used
= meta_sinfo
->bytes_used
+ meta_sinfo
->bytes_reserved
+
3068 meta_sinfo
->bytes_pinned
+ meta_sinfo
->bytes_readonly
+
3069 meta_sinfo
->bytes_super
+ meta_sinfo
->bytes_root
+
3070 meta_sinfo
->bytes_may_use
+ meta_sinfo
->bytes_delalloc
;
3072 if (used
> meta_sinfo
->total_bytes
) {
3076 if (maybe_allocate_chunk(root
, meta_sinfo
))
3080 spin_unlock(&meta_sinfo
->lock
);
3084 filemap_flush(inode
->i_mapping
);
3086 } else if (flushed
== 3) {
3087 flush_delalloc(root
, meta_sinfo
);
3090 spin_lock(&meta_sinfo
->lock
);
3091 meta_sinfo
->bytes_delalloc
-= num_bytes
;
3092 spin_unlock(&meta_sinfo
->lock
);
3093 printk(KERN_ERR
"enospc, has %d, reserved %d\n",
3094 BTRFS_I(inode
)->outstanding_extents
,
3095 BTRFS_I(inode
)->reserved_extents
);
3096 dump_space_info(meta_sinfo
, 0, 0);
3100 BTRFS_I(inode
)->reserved_extents
++;
3101 check_force_delalloc(meta_sinfo
);
3102 spin_unlock(&meta_sinfo
->lock
);
3104 if (!flushed
&& force_delalloc
)
3105 filemap_flush(inode
->i_mapping
);
3111 * unreserve num_items number of items worth of metadata space. This needs to
3112 * be paired with btrfs_reserve_metadata_space.
3114 * NOTE: if you have the option, run this _AFTER_ you do a
3115 * btrfs_end_transaction, since btrfs_end_transaction will run delayed ref
3116 * oprations which will result in more used metadata, so we want to make sure we
3117 * can do that without issue.
3119 int btrfs_unreserve_metadata_space(struct btrfs_root
*root
, int num_items
)
3121 struct btrfs_fs_info
*info
= root
->fs_info
;
3122 struct btrfs_space_info
*meta_sinfo
;
3127 /* get the space info for where the metadata will live */
3128 alloc_target
= btrfs_get_alloc_profile(root
, 0);
3129 meta_sinfo
= __find_space_info(info
, alloc_target
);
3131 num_bytes
= calculate_bytes_needed(root
, num_items
);
3133 spin_lock(&meta_sinfo
->lock
);
3134 if (meta_sinfo
->bytes_may_use
< num_bytes
) {
3136 meta_sinfo
->bytes_may_use
= 0;
3138 meta_sinfo
->bytes_may_use
-= num_bytes
;
3140 spin_unlock(&meta_sinfo
->lock
);
3148 * Reserve some metadata space for use. We'll calculate the worste case number
3149 * of bytes that would be needed to modify num_items number of items. If we
3150 * have space, fantastic, if not, you get -ENOSPC. Please call
3151 * btrfs_unreserve_metadata_space when you are done for the _SAME_ number of
3152 * items you reserved, since whatever metadata you needed should have already
3155 * This will commit the transaction to make more space if we don't have enough
3156 * metadata space. THe only time we don't do this is if we're reserving space
3157 * inside of a transaction, then we will just return -ENOSPC and it is the
3158 * callers responsibility to handle it properly.
3160 int btrfs_reserve_metadata_space(struct btrfs_root
*root
, int num_items
)
3162 struct btrfs_fs_info
*info
= root
->fs_info
;
3163 struct btrfs_space_info
*meta_sinfo
;
3169 /* get the space info for where the metadata will live */
3170 alloc_target
= btrfs_get_alloc_profile(root
, 0);
3171 meta_sinfo
= __find_space_info(info
, alloc_target
);
3173 num_bytes
= calculate_bytes_needed(root
, num_items
);
3175 spin_lock(&meta_sinfo
->lock
);
3177 if (unlikely(!meta_sinfo
->bytes_root
))
3178 meta_sinfo
->bytes_root
= calculate_bytes_needed(root
, 6);
3181 meta_sinfo
->bytes_may_use
+= num_bytes
;
3183 used
= meta_sinfo
->bytes_used
+ meta_sinfo
->bytes_reserved
+
3184 meta_sinfo
->bytes_pinned
+ meta_sinfo
->bytes_readonly
+
3185 meta_sinfo
->bytes_super
+ meta_sinfo
->bytes_root
+
3186 meta_sinfo
->bytes_may_use
+ meta_sinfo
->bytes_delalloc
;
3188 if (used
> meta_sinfo
->total_bytes
) {
3191 if (maybe_allocate_chunk(root
, meta_sinfo
))
3195 spin_unlock(&meta_sinfo
->lock
);
3199 flush_delalloc(root
, meta_sinfo
);
3202 spin_lock(&meta_sinfo
->lock
);
3203 meta_sinfo
->bytes_may_use
-= num_bytes
;
3204 spin_unlock(&meta_sinfo
->lock
);
3206 dump_space_info(meta_sinfo
, 0, 0);
3210 check_force_delalloc(meta_sinfo
);
3211 spin_unlock(&meta_sinfo
->lock
);
3217 * This will check the space that the inode allocates from to make sure we have
3218 * enough space for bytes.
3220 int btrfs_check_data_free_space(struct btrfs_root
*root
, struct inode
*inode
,
3223 struct btrfs_space_info
*data_sinfo
;
3224 int ret
= 0, committed
= 0;
3226 /* make sure bytes are sectorsize aligned */
3227 bytes
= (bytes
+ root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
3229 data_sinfo
= BTRFS_I(inode
)->space_info
;
3234 /* make sure we have enough space to handle the data first */
3235 spin_lock(&data_sinfo
->lock
);
3236 if (data_sinfo
->total_bytes
- data_sinfo
->bytes_used
-
3237 data_sinfo
->bytes_delalloc
- data_sinfo
->bytes_reserved
-
3238 data_sinfo
->bytes_pinned
- data_sinfo
->bytes_readonly
-
3239 data_sinfo
->bytes_may_use
- data_sinfo
->bytes_super
< bytes
) {
3240 struct btrfs_trans_handle
*trans
;
3243 * if we don't have enough free bytes in this space then we need
3244 * to alloc a new chunk.
3246 if (!data_sinfo
->full
) {
3249 data_sinfo
->force_alloc
= 1;
3250 spin_unlock(&data_sinfo
->lock
);
3252 alloc_target
= btrfs_get_alloc_profile(root
, 1);
3253 trans
= btrfs_start_transaction(root
, 1);
3257 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
3258 bytes
+ 2 * 1024 * 1024,
3260 btrfs_end_transaction(trans
, root
);
3265 btrfs_set_inode_space_info(root
, inode
);
3266 data_sinfo
= BTRFS_I(inode
)->space_info
;
3270 spin_unlock(&data_sinfo
->lock
);
3272 /* commit the current transaction and try again */
3273 if (!committed
&& !root
->fs_info
->open_ioctl_trans
) {
3275 trans
= btrfs_join_transaction(root
, 1);
3278 ret
= btrfs_commit_transaction(trans
, root
);
3284 printk(KERN_ERR
"no space left, need %llu, %llu delalloc bytes"
3285 ", %llu bytes_used, %llu bytes_reserved, "
3286 "%llu bytes_pinned, %llu bytes_readonly, %llu may use "
3287 "%llu total\n", (unsigned long long)bytes
,
3288 (unsigned long long)data_sinfo
->bytes_delalloc
,
3289 (unsigned long long)data_sinfo
->bytes_used
,
3290 (unsigned long long)data_sinfo
->bytes_reserved
,
3291 (unsigned long long)data_sinfo
->bytes_pinned
,
3292 (unsigned long long)data_sinfo
->bytes_readonly
,
3293 (unsigned long long)data_sinfo
->bytes_may_use
,
3294 (unsigned long long)data_sinfo
->total_bytes
);
3297 data_sinfo
->bytes_may_use
+= bytes
;
3298 BTRFS_I(inode
)->reserved_bytes
+= bytes
;
3299 spin_unlock(&data_sinfo
->lock
);
3305 * if there was an error for whatever reason after calling
3306 * btrfs_check_data_free_space, call this so we can cleanup the counters.
3308 void btrfs_free_reserved_data_space(struct btrfs_root
*root
,
3309 struct inode
*inode
, u64 bytes
)
3311 struct btrfs_space_info
*data_sinfo
;
3313 /* make sure bytes are sectorsize aligned */
3314 bytes
= (bytes
+ root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
3316 data_sinfo
= BTRFS_I(inode
)->space_info
;
3317 spin_lock(&data_sinfo
->lock
);
3318 data_sinfo
->bytes_may_use
-= bytes
;
3319 BTRFS_I(inode
)->reserved_bytes
-= bytes
;
3320 spin_unlock(&data_sinfo
->lock
);
3323 /* called when we are adding a delalloc extent to the inode's io_tree */
3324 void btrfs_delalloc_reserve_space(struct btrfs_root
*root
, struct inode
*inode
,
3327 struct btrfs_space_info
*data_sinfo
;
3329 /* get the space info for where this inode will be storing its data */
3330 data_sinfo
= BTRFS_I(inode
)->space_info
;
3332 /* make sure we have enough space to handle the data first */
3333 spin_lock(&data_sinfo
->lock
);
3334 data_sinfo
->bytes_delalloc
+= bytes
;
3337 * we are adding a delalloc extent without calling
3338 * btrfs_check_data_free_space first. This happens on a weird
3339 * writepage condition, but shouldn't hurt our accounting
3341 if (unlikely(bytes
> BTRFS_I(inode
)->reserved_bytes
)) {
3342 data_sinfo
->bytes_may_use
-= BTRFS_I(inode
)->reserved_bytes
;
3343 BTRFS_I(inode
)->reserved_bytes
= 0;
3345 data_sinfo
->bytes_may_use
-= bytes
;
3346 BTRFS_I(inode
)->reserved_bytes
-= bytes
;
3349 spin_unlock(&data_sinfo
->lock
);
3352 /* called when we are clearing an delalloc extent from the inode's io_tree */
3353 void btrfs_delalloc_free_space(struct btrfs_root
*root
, struct inode
*inode
,
3356 struct btrfs_space_info
*info
;
3358 info
= BTRFS_I(inode
)->space_info
;
3360 spin_lock(&info
->lock
);
3361 info
->bytes_delalloc
-= bytes
;
3362 spin_unlock(&info
->lock
);
3365 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
3367 struct list_head
*head
= &info
->space_info
;
3368 struct btrfs_space_info
*found
;
3371 list_for_each_entry_rcu(found
, head
, list
) {
3372 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
3373 found
->force_alloc
= 1;
3378 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
3379 struct btrfs_root
*extent_root
, u64 alloc_bytes
,
3380 u64 flags
, int force
)
3382 struct btrfs_space_info
*space_info
;
3383 struct btrfs_fs_info
*fs_info
= extent_root
->fs_info
;
3387 mutex_lock(&fs_info
->chunk_mutex
);
3389 flags
= btrfs_reduce_alloc_profile(extent_root
, flags
);
3391 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
3393 ret
= update_space_info(extent_root
->fs_info
, flags
,
3397 BUG_ON(!space_info
);
3399 spin_lock(&space_info
->lock
);
3400 if (space_info
->force_alloc
)
3402 if (space_info
->full
) {
3403 spin_unlock(&space_info
->lock
);
3407 thresh
= space_info
->total_bytes
- space_info
->bytes_readonly
;
3408 thresh
= div_factor(thresh
, 8);
3410 (space_info
->bytes_used
+ space_info
->bytes_pinned
+
3411 space_info
->bytes_reserved
+ alloc_bytes
) < thresh
) {
3412 spin_unlock(&space_info
->lock
);
3415 spin_unlock(&space_info
->lock
);
3418 * if we're doing a data chunk, go ahead and make sure that
3419 * we keep a reasonable number of metadata chunks allocated in the
3422 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
3423 fs_info
->data_chunk_allocations
++;
3424 if (!(fs_info
->data_chunk_allocations
%
3425 fs_info
->metadata_ratio
))
3426 force_metadata_allocation(fs_info
);
3429 ret
= btrfs_alloc_chunk(trans
, extent_root
, flags
);
3430 spin_lock(&space_info
->lock
);
3432 space_info
->full
= 1;
3433 space_info
->force_alloc
= 0;
3434 spin_unlock(&space_info
->lock
);
3436 mutex_unlock(&extent_root
->fs_info
->chunk_mutex
);
3440 static int update_block_group(struct btrfs_trans_handle
*trans
,
3441 struct btrfs_root
*root
,
3442 u64 bytenr
, u64 num_bytes
, int alloc
,
3445 struct btrfs_block_group_cache
*cache
;
3446 struct btrfs_fs_info
*info
= root
->fs_info
;
3447 u64 total
= num_bytes
;
3451 /* block accounting for super block */
3452 spin_lock(&info
->delalloc_lock
);
3453 old_val
= btrfs_super_bytes_used(&info
->super_copy
);
3455 old_val
+= num_bytes
;
3457 old_val
-= num_bytes
;
3458 btrfs_set_super_bytes_used(&info
->super_copy
, old_val
);
3460 /* block accounting for root item */
3461 old_val
= btrfs_root_used(&root
->root_item
);
3463 old_val
+= num_bytes
;
3465 old_val
-= num_bytes
;
3466 btrfs_set_root_used(&root
->root_item
, old_val
);
3467 spin_unlock(&info
->delalloc_lock
);
3470 cache
= btrfs_lookup_block_group(info
, bytenr
);
3473 byte_in_group
= bytenr
- cache
->key
.objectid
;
3474 WARN_ON(byte_in_group
> cache
->key
.offset
);
3476 spin_lock(&cache
->space_info
->lock
);
3477 spin_lock(&cache
->lock
);
3479 old_val
= btrfs_block_group_used(&cache
->item
);
3480 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
3482 old_val
+= num_bytes
;
3483 btrfs_set_block_group_used(&cache
->item
, old_val
);
3484 cache
->reserved
-= num_bytes
;
3485 cache
->space_info
->bytes_used
+= num_bytes
;
3486 cache
->space_info
->bytes_reserved
-= num_bytes
;
3488 cache
->space_info
->bytes_readonly
-= num_bytes
;
3489 spin_unlock(&cache
->lock
);
3490 spin_unlock(&cache
->space_info
->lock
);
3492 old_val
-= num_bytes
;
3493 cache
->space_info
->bytes_used
-= num_bytes
;
3495 cache
->space_info
->bytes_readonly
+= num_bytes
;
3496 btrfs_set_block_group_used(&cache
->item
, old_val
);
3497 spin_unlock(&cache
->lock
);
3498 spin_unlock(&cache
->space_info
->lock
);
3502 ret
= btrfs_discard_extent(root
, bytenr
,
3506 ret
= btrfs_add_free_space(cache
, bytenr
,
3511 btrfs_put_block_group(cache
);
3513 bytenr
+= num_bytes
;
3518 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
3520 struct btrfs_block_group_cache
*cache
;
3523 cache
= btrfs_lookup_first_block_group(root
->fs_info
, search_start
);
3527 bytenr
= cache
->key
.objectid
;
3528 btrfs_put_block_group(cache
);
3534 * this function must be called within transaction
3536 int btrfs_pin_extent(struct btrfs_root
*root
,
3537 u64 bytenr
, u64 num_bytes
, int reserved
)
3539 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3540 struct btrfs_block_group_cache
*cache
;
3542 cache
= btrfs_lookup_block_group(fs_info
, bytenr
);
3545 spin_lock(&cache
->space_info
->lock
);
3546 spin_lock(&cache
->lock
);
3547 cache
->pinned
+= num_bytes
;
3548 cache
->space_info
->bytes_pinned
+= num_bytes
;
3550 cache
->reserved
-= num_bytes
;
3551 cache
->space_info
->bytes_reserved
-= num_bytes
;
3553 spin_unlock(&cache
->lock
);
3554 spin_unlock(&cache
->space_info
->lock
);
3556 btrfs_put_block_group(cache
);
3558 set_extent_dirty(fs_info
->pinned_extents
,
3559 bytenr
, bytenr
+ num_bytes
- 1, GFP_NOFS
);
3563 static int update_reserved_extents(struct btrfs_block_group_cache
*cache
,
3564 u64 num_bytes
, int reserve
)
3566 spin_lock(&cache
->space_info
->lock
);
3567 spin_lock(&cache
->lock
);
3569 cache
->reserved
+= num_bytes
;
3570 cache
->space_info
->bytes_reserved
+= num_bytes
;
3572 cache
->reserved
-= num_bytes
;
3573 cache
->space_info
->bytes_reserved
-= num_bytes
;
3575 spin_unlock(&cache
->lock
);
3576 spin_unlock(&cache
->space_info
->lock
);
3580 int btrfs_prepare_extent_commit(struct btrfs_trans_handle
*trans
,
3581 struct btrfs_root
*root
)
3583 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3584 struct btrfs_caching_control
*next
;
3585 struct btrfs_caching_control
*caching_ctl
;
3586 struct btrfs_block_group_cache
*cache
;
3588 down_write(&fs_info
->extent_commit_sem
);
3590 list_for_each_entry_safe(caching_ctl
, next
,
3591 &fs_info
->caching_block_groups
, list
) {
3592 cache
= caching_ctl
->block_group
;
3593 if (block_group_cache_done(cache
)) {
3594 cache
->last_byte_to_unpin
= (u64
)-1;
3595 list_del_init(&caching_ctl
->list
);
3596 put_caching_control(caching_ctl
);
3598 cache
->last_byte_to_unpin
= caching_ctl
->progress
;
3602 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
3603 fs_info
->pinned_extents
= &fs_info
->freed_extents
[1];
3605 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
3607 up_write(&fs_info
->extent_commit_sem
);
3611 static int unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
3613 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3614 struct btrfs_block_group_cache
*cache
= NULL
;
3617 while (start
<= end
) {
3619 start
>= cache
->key
.objectid
+ cache
->key
.offset
) {
3621 btrfs_put_block_group(cache
);
3622 cache
= btrfs_lookup_block_group(fs_info
, start
);
3626 len
= cache
->key
.objectid
+ cache
->key
.offset
- start
;
3627 len
= min(len
, end
+ 1 - start
);
3629 if (start
< cache
->last_byte_to_unpin
) {
3630 len
= min(len
, cache
->last_byte_to_unpin
- start
);
3631 btrfs_add_free_space(cache
, start
, len
);
3634 spin_lock(&cache
->space_info
->lock
);
3635 spin_lock(&cache
->lock
);
3636 cache
->pinned
-= len
;
3637 cache
->space_info
->bytes_pinned
-= len
;
3638 spin_unlock(&cache
->lock
);
3639 spin_unlock(&cache
->space_info
->lock
);
3645 btrfs_put_block_group(cache
);
3649 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
3650 struct btrfs_root
*root
)
3652 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3653 struct extent_io_tree
*unpin
;
3658 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
3659 unpin
= &fs_info
->freed_extents
[1];
3661 unpin
= &fs_info
->freed_extents
[0];
3664 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
3669 ret
= btrfs_discard_extent(root
, start
, end
+ 1 - start
);
3671 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
3672 unpin_extent_range(root
, start
, end
);
3679 static int pin_down_bytes(struct btrfs_trans_handle
*trans
,
3680 struct btrfs_root
*root
,
3681 struct btrfs_path
*path
,
3682 u64 bytenr
, u64 num_bytes
,
3683 int is_data
, int reserved
,
3684 struct extent_buffer
**must_clean
)
3687 struct extent_buffer
*buf
;
3692 buf
= btrfs_find_tree_block(root
, bytenr
, num_bytes
);
3696 /* we can reuse a block if it hasn't been written
3697 * and it is from this transaction. We can't
3698 * reuse anything from the tree log root because
3699 * it has tiny sub-transactions.
3701 if (btrfs_buffer_uptodate(buf
, 0) &&
3702 btrfs_try_tree_lock(buf
)) {
3703 u64 header_owner
= btrfs_header_owner(buf
);
3704 u64 header_transid
= btrfs_header_generation(buf
);
3705 if (header_owner
!= BTRFS_TREE_LOG_OBJECTID
&&
3706 header_transid
== trans
->transid
&&
3707 !btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
3711 btrfs_tree_unlock(buf
);
3713 free_extent_buffer(buf
);
3716 btrfs_set_path_blocking(path
);
3717 /* unlocks the pinned mutex */
3718 btrfs_pin_extent(root
, bytenr
, num_bytes
, reserved
);
3724 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
3725 struct btrfs_root
*root
,
3726 u64 bytenr
, u64 num_bytes
, u64 parent
,
3727 u64 root_objectid
, u64 owner_objectid
,
3728 u64 owner_offset
, int refs_to_drop
,
3729 struct btrfs_delayed_extent_op
*extent_op
)
3731 struct btrfs_key key
;
3732 struct btrfs_path
*path
;
3733 struct btrfs_fs_info
*info
= root
->fs_info
;
3734 struct btrfs_root
*extent_root
= info
->extent_root
;
3735 struct extent_buffer
*leaf
;
3736 struct btrfs_extent_item
*ei
;
3737 struct btrfs_extent_inline_ref
*iref
;
3740 int extent_slot
= 0;
3741 int found_extent
= 0;
3746 path
= btrfs_alloc_path();
3751 path
->leave_spinning
= 1;
3753 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
3754 BUG_ON(!is_data
&& refs_to_drop
!= 1);
3756 ret
= lookup_extent_backref(trans
, extent_root
, path
, &iref
,
3757 bytenr
, num_bytes
, parent
,
3758 root_objectid
, owner_objectid
,
3761 extent_slot
= path
->slots
[0];
3762 while (extent_slot
>= 0) {
3763 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
3765 if (key
.objectid
!= bytenr
)
3767 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
3768 key
.offset
== num_bytes
) {
3772 if (path
->slots
[0] - extent_slot
> 5)
3776 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3777 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
3778 if (found_extent
&& item_size
< sizeof(*ei
))
3781 if (!found_extent
) {
3783 ret
= remove_extent_backref(trans
, extent_root
, path
,
3787 btrfs_release_path(extent_root
, path
);
3788 path
->leave_spinning
= 1;
3790 key
.objectid
= bytenr
;
3791 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
3792 key
.offset
= num_bytes
;
3794 ret
= btrfs_search_slot(trans
, extent_root
,
3797 printk(KERN_ERR
"umm, got %d back from search"
3798 ", was looking for %llu\n", ret
,
3799 (unsigned long long)bytenr
);
3800 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
3803 extent_slot
= path
->slots
[0];
3806 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
3808 printk(KERN_ERR
"btrfs unable to find ref byte nr %llu "
3809 "parent %llu root %llu owner %llu offset %llu\n",
3810 (unsigned long long)bytenr
,
3811 (unsigned long long)parent
,
3812 (unsigned long long)root_objectid
,
3813 (unsigned long long)owner_objectid
,
3814 (unsigned long long)owner_offset
);
3817 leaf
= path
->nodes
[0];
3818 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
3819 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3820 if (item_size
< sizeof(*ei
)) {
3821 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
3822 ret
= convert_extent_item_v0(trans
, extent_root
, path
,
3826 btrfs_release_path(extent_root
, path
);
3827 path
->leave_spinning
= 1;
3829 key
.objectid
= bytenr
;
3830 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
3831 key
.offset
= num_bytes
;
3833 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
3836 printk(KERN_ERR
"umm, got %d back from search"
3837 ", was looking for %llu\n", ret
,
3838 (unsigned long long)bytenr
);
3839 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
3842 extent_slot
= path
->slots
[0];
3843 leaf
= path
->nodes
[0];
3844 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
3847 BUG_ON(item_size
< sizeof(*ei
));
3848 ei
= btrfs_item_ptr(leaf
, extent_slot
,
3849 struct btrfs_extent_item
);
3850 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
) {
3851 struct btrfs_tree_block_info
*bi
;
3852 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
3853 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
3854 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
3857 refs
= btrfs_extent_refs(leaf
, ei
);
3858 BUG_ON(refs
< refs_to_drop
);
3859 refs
-= refs_to_drop
;
3863 __run_delayed_extent_op(extent_op
, leaf
, ei
);
3865 * In the case of inline back ref, reference count will
3866 * be updated by remove_extent_backref
3869 BUG_ON(!found_extent
);
3871 btrfs_set_extent_refs(leaf
, ei
, refs
);
3872 btrfs_mark_buffer_dirty(leaf
);
3875 ret
= remove_extent_backref(trans
, extent_root
, path
,
3882 struct extent_buffer
*must_clean
= NULL
;
3885 BUG_ON(is_data
&& refs_to_drop
!=
3886 extent_data_ref_count(root
, path
, iref
));
3888 BUG_ON(path
->slots
[0] != extent_slot
);
3890 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
3891 path
->slots
[0] = extent_slot
;
3896 ret
= pin_down_bytes(trans
, root
, path
, bytenr
,
3897 num_bytes
, is_data
, 0, &must_clean
);
3902 * it is going to be very rare for someone to be waiting
3903 * on the block we're freeing. del_items might need to
3904 * schedule, so rather than get fancy, just force it
3908 btrfs_set_lock_blocking(must_clean
);
3910 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
3913 btrfs_release_path(extent_root
, path
);
3916 clean_tree_block(NULL
, root
, must_clean
);
3917 btrfs_tree_unlock(must_clean
);
3918 free_extent_buffer(must_clean
);
3922 ret
= btrfs_del_csums(trans
, root
, bytenr
, num_bytes
);
3925 invalidate_mapping_pages(info
->btree_inode
->i_mapping
,
3926 bytenr
>> PAGE_CACHE_SHIFT
,
3927 (bytenr
+ num_bytes
- 1) >> PAGE_CACHE_SHIFT
);
3930 ret
= update_block_group(trans
, root
, bytenr
, num_bytes
, 0,
3934 btrfs_free_path(path
);
3939 * when we free an extent, it is possible (and likely) that we free the last
3940 * delayed ref for that extent as well. This searches the delayed ref tree for
3941 * a given extent, and if there are no other delayed refs to be processed, it
3942 * removes it from the tree.
3944 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
3945 struct btrfs_root
*root
, u64 bytenr
)
3947 struct btrfs_delayed_ref_head
*head
;
3948 struct btrfs_delayed_ref_root
*delayed_refs
;
3949 struct btrfs_delayed_ref_node
*ref
;
3950 struct rb_node
*node
;
3953 delayed_refs
= &trans
->transaction
->delayed_refs
;
3954 spin_lock(&delayed_refs
->lock
);
3955 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
3959 node
= rb_prev(&head
->node
.rb_node
);
3963 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
3965 /* there are still entries for this ref, we can't drop it */
3966 if (ref
->bytenr
== bytenr
)
3969 if (head
->extent_op
) {
3970 if (!head
->must_insert_reserved
)
3972 kfree(head
->extent_op
);
3973 head
->extent_op
= NULL
;
3977 * waiting for the lock here would deadlock. If someone else has it
3978 * locked they are already in the process of dropping it anyway
3980 if (!mutex_trylock(&head
->mutex
))
3984 * at this point we have a head with no other entries. Go
3985 * ahead and process it.
3987 head
->node
.in_tree
= 0;
3988 rb_erase(&head
->node
.rb_node
, &delayed_refs
->root
);
3990 delayed_refs
->num_entries
--;
3993 * we don't take a ref on the node because we're removing it from the
3994 * tree, so we just steal the ref the tree was holding.
3996 delayed_refs
->num_heads
--;
3997 if (list_empty(&head
->cluster
))
3998 delayed_refs
->num_heads_ready
--;
4000 list_del_init(&head
->cluster
);
4001 spin_unlock(&delayed_refs
->lock
);
4003 ret
= run_one_delayed_ref(trans
, root
->fs_info
->tree_root
,
4004 &head
->node
, head
->extent_op
,
4005 head
->must_insert_reserved
);
4007 btrfs_put_delayed_ref(&head
->node
);
4010 spin_unlock(&delayed_refs
->lock
);
4014 int btrfs_free_extent(struct btrfs_trans_handle
*trans
,
4015 struct btrfs_root
*root
,
4016 u64 bytenr
, u64 num_bytes
, u64 parent
,
4017 u64 root_objectid
, u64 owner
, u64 offset
)
4022 * tree log blocks never actually go into the extent allocation
4023 * tree, just update pinning info and exit early.
4025 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
4026 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
4027 /* unlocks the pinned mutex */
4028 btrfs_pin_extent(root
, bytenr
, num_bytes
, 1);
4030 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
4031 ret
= btrfs_add_delayed_tree_ref(trans
, bytenr
, num_bytes
,
4032 parent
, root_objectid
, (int)owner
,
4033 BTRFS_DROP_DELAYED_REF
, NULL
);
4035 ret
= check_ref_cleanup(trans
, root
, bytenr
);
4038 ret
= btrfs_add_delayed_data_ref(trans
, bytenr
, num_bytes
,
4039 parent
, root_objectid
, owner
,
4040 offset
, BTRFS_DROP_DELAYED_REF
, NULL
);
4046 static u64
stripe_align(struct btrfs_root
*root
, u64 val
)
4048 u64 mask
= ((u64
)root
->stripesize
- 1);
4049 u64 ret
= (val
+ mask
) & ~mask
;
4054 * when we wait for progress in the block group caching, its because
4055 * our allocation attempt failed at least once. So, we must sleep
4056 * and let some progress happen before we try again.
4058 * This function will sleep at least once waiting for new free space to
4059 * show up, and then it will check the block group free space numbers
4060 * for our min num_bytes. Another option is to have it go ahead
4061 * and look in the rbtree for a free extent of a given size, but this
4065 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
4068 struct btrfs_caching_control
*caching_ctl
;
4071 caching_ctl
= get_caching_control(cache
);
4075 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
) ||
4076 (cache
->free_space
>= num_bytes
));
4078 put_caching_control(caching_ctl
);
4083 wait_block_group_cache_done(struct btrfs_block_group_cache
*cache
)
4085 struct btrfs_caching_control
*caching_ctl
;
4088 caching_ctl
= get_caching_control(cache
);
4092 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
));
4094 put_caching_control(caching_ctl
);
4098 enum btrfs_loop_type
{
4099 LOOP_CACHED_ONLY
= 0,
4100 LOOP_CACHING_NOWAIT
= 1,
4101 LOOP_CACHING_WAIT
= 2,
4102 LOOP_ALLOC_CHUNK
= 3,
4103 LOOP_NO_EMPTY_SIZE
= 4,
4107 * walks the btree of allocated extents and find a hole of a given size.
4108 * The key ins is changed to record the hole:
4109 * ins->objectid == block start
4110 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4111 * ins->offset == number of blocks
4112 * Any available blocks before search_start are skipped.
4114 static noinline
int find_free_extent(struct btrfs_trans_handle
*trans
,
4115 struct btrfs_root
*orig_root
,
4116 u64 num_bytes
, u64 empty_size
,
4117 u64 search_start
, u64 search_end
,
4118 u64 hint_byte
, struct btrfs_key
*ins
,
4119 u64 exclude_start
, u64 exclude_nr
,
4123 struct btrfs_root
*root
= orig_root
->fs_info
->extent_root
;
4124 struct btrfs_free_cluster
*last_ptr
= NULL
;
4125 struct btrfs_block_group_cache
*block_group
= NULL
;
4126 int empty_cluster
= 2 * 1024 * 1024;
4127 int allowed_chunk_alloc
= 0;
4128 struct btrfs_space_info
*space_info
;
4129 int last_ptr_loop
= 0;
4131 bool found_uncached_bg
= false;
4132 bool failed_cluster_refill
= false;
4133 bool failed_alloc
= false;
4135 WARN_ON(num_bytes
< root
->sectorsize
);
4136 btrfs_set_key_type(ins
, BTRFS_EXTENT_ITEM_KEY
);
4140 space_info
= __find_space_info(root
->fs_info
, data
);
4142 if (orig_root
->ref_cows
|| empty_size
)
4143 allowed_chunk_alloc
= 1;
4145 if (data
& BTRFS_BLOCK_GROUP_METADATA
) {
4146 last_ptr
= &root
->fs_info
->meta_alloc_cluster
;
4147 if (!btrfs_test_opt(root
, SSD
))
4148 empty_cluster
= 64 * 1024;
4151 if ((data
& BTRFS_BLOCK_GROUP_DATA
) && btrfs_test_opt(root
, SSD
)) {
4152 last_ptr
= &root
->fs_info
->data_alloc_cluster
;
4156 spin_lock(&last_ptr
->lock
);
4157 if (last_ptr
->block_group
)
4158 hint_byte
= last_ptr
->window_start
;
4159 spin_unlock(&last_ptr
->lock
);
4162 search_start
= max(search_start
, first_logical_byte(root
, 0));
4163 search_start
= max(search_start
, hint_byte
);
4168 if (search_start
== hint_byte
) {
4169 block_group
= btrfs_lookup_block_group(root
->fs_info
,
4172 * we don't want to use the block group if it doesn't match our
4173 * allocation bits, or if its not cached.
4175 if (block_group
&& block_group_bits(block_group
, data
) &&
4176 block_group_cache_done(block_group
)) {
4177 down_read(&space_info
->groups_sem
);
4178 if (list_empty(&block_group
->list
) ||
4181 * someone is removing this block group,
4182 * we can't jump into the have_block_group
4183 * target because our list pointers are not
4186 btrfs_put_block_group(block_group
);
4187 up_read(&space_info
->groups_sem
);
4189 goto have_block_group
;
4190 } else if (block_group
) {
4191 btrfs_put_block_group(block_group
);
4196 down_read(&space_info
->groups_sem
);
4197 list_for_each_entry(block_group
, &space_info
->block_groups
, list
) {
4201 atomic_inc(&block_group
->count
);
4202 search_start
= block_group
->key
.objectid
;
4205 if (unlikely(block_group
->cached
== BTRFS_CACHE_NO
)) {
4207 * we want to start caching kthreads, but not too many
4208 * right off the bat so we don't overwhelm the system,
4209 * so only start them if there are less than 2 and we're
4210 * in the initial allocation phase.
4212 if (loop
> LOOP_CACHING_NOWAIT
||
4213 atomic_read(&space_info
->caching_threads
) < 2) {
4214 ret
= cache_block_group(block_group
);
4219 cached
= block_group_cache_done(block_group
);
4220 if (unlikely(!cached
)) {
4221 found_uncached_bg
= true;
4223 /* if we only want cached bgs, loop */
4224 if (loop
== LOOP_CACHED_ONLY
)
4228 if (unlikely(block_group
->ro
))
4232 * Ok we want to try and use the cluster allocator, so lets look
4233 * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
4234 * have tried the cluster allocator plenty of times at this
4235 * point and not have found anything, so we are likely way too
4236 * fragmented for the clustering stuff to find anything, so lets
4237 * just skip it and let the allocator find whatever block it can
4240 if (last_ptr
&& loop
< LOOP_NO_EMPTY_SIZE
) {
4242 * the refill lock keeps out other
4243 * people trying to start a new cluster
4245 spin_lock(&last_ptr
->refill_lock
);
4246 if (last_ptr
->block_group
&&
4247 (last_ptr
->block_group
->ro
||
4248 !block_group_bits(last_ptr
->block_group
, data
))) {
4250 goto refill_cluster
;
4253 offset
= btrfs_alloc_from_cluster(block_group
, last_ptr
,
4254 num_bytes
, search_start
);
4256 /* we have a block, we're done */
4257 spin_unlock(&last_ptr
->refill_lock
);
4261 spin_lock(&last_ptr
->lock
);
4263 * whoops, this cluster doesn't actually point to
4264 * this block group. Get a ref on the block
4265 * group is does point to and try again
4267 if (!last_ptr_loop
&& last_ptr
->block_group
&&
4268 last_ptr
->block_group
!= block_group
) {
4270 btrfs_put_block_group(block_group
);
4271 block_group
= last_ptr
->block_group
;
4272 atomic_inc(&block_group
->count
);
4273 spin_unlock(&last_ptr
->lock
);
4274 spin_unlock(&last_ptr
->refill_lock
);
4277 search_start
= block_group
->key
.objectid
;
4279 * we know this block group is properly
4280 * in the list because
4281 * btrfs_remove_block_group, drops the
4282 * cluster before it removes the block
4283 * group from the list
4285 goto have_block_group
;
4287 spin_unlock(&last_ptr
->lock
);
4290 * this cluster didn't work out, free it and
4293 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
4297 /* allocate a cluster in this block group */
4298 ret
= btrfs_find_space_cluster(trans
, root
,
4299 block_group
, last_ptr
,
4301 empty_cluster
+ empty_size
);
4304 * now pull our allocation out of this
4307 offset
= btrfs_alloc_from_cluster(block_group
,
4308 last_ptr
, num_bytes
,
4311 /* we found one, proceed */
4312 spin_unlock(&last_ptr
->refill_lock
);
4315 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
4316 && !failed_cluster_refill
) {
4317 spin_unlock(&last_ptr
->refill_lock
);
4319 failed_cluster_refill
= true;
4320 wait_block_group_cache_progress(block_group
,
4321 num_bytes
+ empty_cluster
+ empty_size
);
4322 goto have_block_group
;
4326 * at this point we either didn't find a cluster
4327 * or we weren't able to allocate a block from our
4328 * cluster. Free the cluster we've been trying
4329 * to use, and go to the next block group
4331 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
4332 spin_unlock(&last_ptr
->refill_lock
);
4336 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
4337 num_bytes
, empty_size
);
4339 * If we didn't find a chunk, and we haven't failed on this
4340 * block group before, and this block group is in the middle of
4341 * caching and we are ok with waiting, then go ahead and wait
4342 * for progress to be made, and set failed_alloc to true.
4344 * If failed_alloc is true then we've already waited on this
4345 * block group once and should move on to the next block group.
4347 if (!offset
&& !failed_alloc
&& !cached
&&
4348 loop
> LOOP_CACHING_NOWAIT
) {
4349 wait_block_group_cache_progress(block_group
,
4350 num_bytes
+ empty_size
);
4351 failed_alloc
= true;
4352 goto have_block_group
;
4353 } else if (!offset
) {
4357 search_start
= stripe_align(root
, offset
);
4358 /* move on to the next group */
4359 if (search_start
+ num_bytes
>= search_end
) {
4360 btrfs_add_free_space(block_group
, offset
, num_bytes
);
4364 /* move on to the next group */
4365 if (search_start
+ num_bytes
>
4366 block_group
->key
.objectid
+ block_group
->key
.offset
) {
4367 btrfs_add_free_space(block_group
, offset
, num_bytes
);
4371 if (exclude_nr
> 0 &&
4372 (search_start
+ num_bytes
> exclude_start
&&
4373 search_start
< exclude_start
+ exclude_nr
)) {
4374 search_start
= exclude_start
+ exclude_nr
;
4376 btrfs_add_free_space(block_group
, offset
, num_bytes
);
4378 * if search_start is still in this block group
4379 * then we just re-search this block group
4381 if (search_start
>= block_group
->key
.objectid
&&
4382 search_start
< (block_group
->key
.objectid
+
4383 block_group
->key
.offset
))
4384 goto have_block_group
;
4388 ins
->objectid
= search_start
;
4389 ins
->offset
= num_bytes
;
4391 if (offset
< search_start
)
4392 btrfs_add_free_space(block_group
, offset
,
4393 search_start
- offset
);
4394 BUG_ON(offset
> search_start
);
4396 update_reserved_extents(block_group
, num_bytes
, 1);
4398 /* we are all good, lets return */
4401 failed_cluster_refill
= false;
4402 failed_alloc
= false;
4403 btrfs_put_block_group(block_group
);
4405 up_read(&space_info
->groups_sem
);
4407 /* LOOP_CACHED_ONLY, only search fully cached block groups
4408 * LOOP_CACHING_NOWAIT, search partially cached block groups, but
4409 * dont wait foR them to finish caching
4410 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
4411 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
4412 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
4415 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
&&
4416 (found_uncached_bg
|| empty_size
|| empty_cluster
||
4417 allowed_chunk_alloc
)) {
4418 if (found_uncached_bg
) {
4419 found_uncached_bg
= false;
4420 if (loop
< LOOP_CACHING_WAIT
) {
4426 if (loop
== LOOP_ALLOC_CHUNK
) {
4431 if (allowed_chunk_alloc
) {
4432 ret
= do_chunk_alloc(trans
, root
, num_bytes
+
4433 2 * 1024 * 1024, data
, 1);
4434 allowed_chunk_alloc
= 0;
4436 space_info
->force_alloc
= 1;
4439 if (loop
< LOOP_NO_EMPTY_SIZE
) {
4444 } else if (!ins
->objectid
) {
4448 /* we found what we needed */
4449 if (ins
->objectid
) {
4450 if (!(data
& BTRFS_BLOCK_GROUP_DATA
))
4451 trans
->block_group
= block_group
->key
.objectid
;
4453 btrfs_put_block_group(block_group
);
4460 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
4461 int dump_block_groups
)
4463 struct btrfs_block_group_cache
*cache
;
4465 spin_lock(&info
->lock
);
4466 printk(KERN_INFO
"space_info has %llu free, is %sfull\n",
4467 (unsigned long long)(info
->total_bytes
- info
->bytes_used
-
4468 info
->bytes_pinned
- info
->bytes_reserved
-
4470 (info
->full
) ? "" : "not ");
4471 printk(KERN_INFO
"space_info total=%llu, pinned=%llu, delalloc=%llu,"
4472 " may_use=%llu, used=%llu, root=%llu, super=%llu, reserved=%llu"
4474 (unsigned long long)info
->total_bytes
,
4475 (unsigned long long)info
->bytes_pinned
,
4476 (unsigned long long)info
->bytes_delalloc
,
4477 (unsigned long long)info
->bytes_may_use
,
4478 (unsigned long long)info
->bytes_used
,
4479 (unsigned long long)info
->bytes_root
,
4480 (unsigned long long)info
->bytes_super
,
4481 (unsigned long long)info
->bytes_reserved
);
4482 spin_unlock(&info
->lock
);
4484 if (!dump_block_groups
)
4487 down_read(&info
->groups_sem
);
4488 list_for_each_entry(cache
, &info
->block_groups
, list
) {
4489 spin_lock(&cache
->lock
);
4490 printk(KERN_INFO
"block group %llu has %llu bytes, %llu used "
4491 "%llu pinned %llu reserved\n",
4492 (unsigned long long)cache
->key
.objectid
,
4493 (unsigned long long)cache
->key
.offset
,
4494 (unsigned long long)btrfs_block_group_used(&cache
->item
),
4495 (unsigned long long)cache
->pinned
,
4496 (unsigned long long)cache
->reserved
);
4497 btrfs_dump_free_space(cache
, bytes
);
4498 spin_unlock(&cache
->lock
);
4500 up_read(&info
->groups_sem
);
4503 int btrfs_reserve_extent(struct btrfs_trans_handle
*trans
,
4504 struct btrfs_root
*root
,
4505 u64 num_bytes
, u64 min_alloc_size
,
4506 u64 empty_size
, u64 hint_byte
,
4507 u64 search_end
, struct btrfs_key
*ins
,
4511 u64 search_start
= 0;
4512 struct btrfs_fs_info
*info
= root
->fs_info
;
4514 data
= btrfs_get_alloc_profile(root
, data
);
4517 * the only place that sets empty_size is btrfs_realloc_node, which
4518 * is not called recursively on allocations
4520 if (empty_size
|| root
->ref_cows
) {
4521 if (!(data
& BTRFS_BLOCK_GROUP_METADATA
)) {
4522 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
4524 BTRFS_BLOCK_GROUP_METADATA
|
4525 (info
->metadata_alloc_profile
&
4526 info
->avail_metadata_alloc_bits
), 0);
4528 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
4529 num_bytes
+ 2 * 1024 * 1024, data
, 0);
4532 WARN_ON(num_bytes
< root
->sectorsize
);
4533 ret
= find_free_extent(trans
, root
, num_bytes
, empty_size
,
4534 search_start
, search_end
, hint_byte
, ins
,
4535 trans
->alloc_exclude_start
,
4536 trans
->alloc_exclude_nr
, data
);
4538 if (ret
== -ENOSPC
&& num_bytes
> min_alloc_size
) {
4539 num_bytes
= num_bytes
>> 1;
4540 num_bytes
= num_bytes
& ~(root
->sectorsize
- 1);
4541 num_bytes
= max(num_bytes
, min_alloc_size
);
4542 do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
4543 num_bytes
, data
, 1);
4546 if (ret
== -ENOSPC
) {
4547 struct btrfs_space_info
*sinfo
;
4549 sinfo
= __find_space_info(root
->fs_info
, data
);
4550 printk(KERN_ERR
"btrfs allocation failed flags %llu, "
4551 "wanted %llu\n", (unsigned long long)data
,
4552 (unsigned long long)num_bytes
);
4553 dump_space_info(sinfo
, num_bytes
, 1);
4559 int btrfs_free_reserved_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
4561 struct btrfs_block_group_cache
*cache
;
4564 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
4566 printk(KERN_ERR
"Unable to find block group for %llu\n",
4567 (unsigned long long)start
);
4571 ret
= btrfs_discard_extent(root
, start
, len
);
4573 btrfs_add_free_space(cache
, start
, len
);
4574 update_reserved_extents(cache
, len
, 0);
4575 btrfs_put_block_group(cache
);
4580 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
4581 struct btrfs_root
*root
,
4582 u64 parent
, u64 root_objectid
,
4583 u64 flags
, u64 owner
, u64 offset
,
4584 struct btrfs_key
*ins
, int ref_mod
)
4587 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4588 struct btrfs_extent_item
*extent_item
;
4589 struct btrfs_extent_inline_ref
*iref
;
4590 struct btrfs_path
*path
;
4591 struct extent_buffer
*leaf
;
4596 type
= BTRFS_SHARED_DATA_REF_KEY
;
4598 type
= BTRFS_EXTENT_DATA_REF_KEY
;
4600 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
4602 path
= btrfs_alloc_path();
4605 path
->leave_spinning
= 1;
4606 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
4610 leaf
= path
->nodes
[0];
4611 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
4612 struct btrfs_extent_item
);
4613 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
4614 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
4615 btrfs_set_extent_flags(leaf
, extent_item
,
4616 flags
| BTRFS_EXTENT_FLAG_DATA
);
4618 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
4619 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
4621 struct btrfs_shared_data_ref
*ref
;
4622 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
4623 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
4624 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
4626 struct btrfs_extent_data_ref
*ref
;
4627 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
4628 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
4629 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
4630 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
4631 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
4634 btrfs_mark_buffer_dirty(path
->nodes
[0]);
4635 btrfs_free_path(path
);
4637 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
,
4640 printk(KERN_ERR
"btrfs update block group failed for %llu "
4641 "%llu\n", (unsigned long long)ins
->objectid
,
4642 (unsigned long long)ins
->offset
);
4648 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
4649 struct btrfs_root
*root
,
4650 u64 parent
, u64 root_objectid
,
4651 u64 flags
, struct btrfs_disk_key
*key
,
4652 int level
, struct btrfs_key
*ins
)
4655 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4656 struct btrfs_extent_item
*extent_item
;
4657 struct btrfs_tree_block_info
*block_info
;
4658 struct btrfs_extent_inline_ref
*iref
;
4659 struct btrfs_path
*path
;
4660 struct extent_buffer
*leaf
;
4661 u32 size
= sizeof(*extent_item
) + sizeof(*block_info
) + sizeof(*iref
);
4663 path
= btrfs_alloc_path();
4666 path
->leave_spinning
= 1;
4667 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
4671 leaf
= path
->nodes
[0];
4672 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
4673 struct btrfs_extent_item
);
4674 btrfs_set_extent_refs(leaf
, extent_item
, 1);
4675 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
4676 btrfs_set_extent_flags(leaf
, extent_item
,
4677 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
4678 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
4680 btrfs_set_tree_block_key(leaf
, block_info
, key
);
4681 btrfs_set_tree_block_level(leaf
, block_info
, level
);
4683 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
4685 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
4686 btrfs_set_extent_inline_ref_type(leaf
, iref
,
4687 BTRFS_SHARED_BLOCK_REF_KEY
);
4688 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
4690 btrfs_set_extent_inline_ref_type(leaf
, iref
,
4691 BTRFS_TREE_BLOCK_REF_KEY
);
4692 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
4695 btrfs_mark_buffer_dirty(leaf
);
4696 btrfs_free_path(path
);
4698 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
,
4701 printk(KERN_ERR
"btrfs update block group failed for %llu "
4702 "%llu\n", (unsigned long long)ins
->objectid
,
4703 (unsigned long long)ins
->offset
);
4709 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
4710 struct btrfs_root
*root
,
4711 u64 root_objectid
, u64 owner
,
4712 u64 offset
, struct btrfs_key
*ins
)
4716 BUG_ON(root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
4718 ret
= btrfs_add_delayed_data_ref(trans
, ins
->objectid
, ins
->offset
,
4719 0, root_objectid
, owner
, offset
,
4720 BTRFS_ADD_DELAYED_EXTENT
, NULL
);
4725 * this is used by the tree logging recovery code. It records that
4726 * an extent has been allocated and makes sure to clear the free
4727 * space cache bits as well
4729 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
4730 struct btrfs_root
*root
,
4731 u64 root_objectid
, u64 owner
, u64 offset
,
4732 struct btrfs_key
*ins
)
4735 struct btrfs_block_group_cache
*block_group
;
4736 struct btrfs_caching_control
*caching_ctl
;
4737 u64 start
= ins
->objectid
;
4738 u64 num_bytes
= ins
->offset
;
4740 block_group
= btrfs_lookup_block_group(root
->fs_info
, ins
->objectid
);
4741 cache_block_group(block_group
);
4742 caching_ctl
= get_caching_control(block_group
);
4745 BUG_ON(!block_group_cache_done(block_group
));
4746 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
4749 mutex_lock(&caching_ctl
->mutex
);
4751 if (start
>= caching_ctl
->progress
) {
4752 ret
= add_excluded_extent(root
, start
, num_bytes
);
4754 } else if (start
+ num_bytes
<= caching_ctl
->progress
) {
4755 ret
= btrfs_remove_free_space(block_group
,
4759 num_bytes
= caching_ctl
->progress
- start
;
4760 ret
= btrfs_remove_free_space(block_group
,
4764 start
= caching_ctl
->progress
;
4765 num_bytes
= ins
->objectid
+ ins
->offset
-
4766 caching_ctl
->progress
;
4767 ret
= add_excluded_extent(root
, start
, num_bytes
);
4771 mutex_unlock(&caching_ctl
->mutex
);
4772 put_caching_control(caching_ctl
);
4775 update_reserved_extents(block_group
, ins
->offset
, 1);
4776 btrfs_put_block_group(block_group
);
4777 ret
= alloc_reserved_file_extent(trans
, root
, 0, root_objectid
,
4778 0, owner
, offset
, ins
, 1);
4783 * finds a free extent and does all the dirty work required for allocation
4784 * returns the key for the extent through ins, and a tree buffer for
4785 * the first block of the extent through buf.
4787 * returns 0 if everything worked, non-zero otherwise.
4789 static int alloc_tree_block(struct btrfs_trans_handle
*trans
,
4790 struct btrfs_root
*root
,
4791 u64 num_bytes
, u64 parent
, u64 root_objectid
,
4792 struct btrfs_disk_key
*key
, int level
,
4793 u64 empty_size
, u64 hint_byte
, u64 search_end
,
4794 struct btrfs_key
*ins
)
4799 ret
= btrfs_reserve_extent(trans
, root
, num_bytes
, num_bytes
,
4800 empty_size
, hint_byte
, search_end
,
4805 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
4807 parent
= ins
->objectid
;
4808 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
4812 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
4813 struct btrfs_delayed_extent_op
*extent_op
;
4814 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
4817 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
4819 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
4820 extent_op
->flags_to_set
= flags
;
4821 extent_op
->update_key
= 1;
4822 extent_op
->update_flags
= 1;
4823 extent_op
->is_data
= 0;
4825 ret
= btrfs_add_delayed_tree_ref(trans
, ins
->objectid
,
4826 ins
->offset
, parent
, root_objectid
,
4827 level
, BTRFS_ADD_DELAYED_EXTENT
,
4834 struct extent_buffer
*btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
,
4835 struct btrfs_root
*root
,
4836 u64 bytenr
, u32 blocksize
,
4839 struct extent_buffer
*buf
;
4841 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
4843 return ERR_PTR(-ENOMEM
);
4844 btrfs_set_header_generation(buf
, trans
->transid
);
4845 btrfs_set_buffer_lockdep_class(buf
, level
);
4846 btrfs_tree_lock(buf
);
4847 clean_tree_block(trans
, root
, buf
);
4849 btrfs_set_lock_blocking(buf
);
4850 btrfs_set_buffer_uptodate(buf
);
4852 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
4853 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
4854 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
4856 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
4857 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
4859 trans
->blocks_used
++;
4860 /* this returns a buffer locked for blocking */
4865 * helper function to allocate a block for a given tree
4866 * returns the tree buffer or NULL.
4868 struct extent_buffer
*btrfs_alloc_free_block(struct btrfs_trans_handle
*trans
,
4869 struct btrfs_root
*root
, u32 blocksize
,
4870 u64 parent
, u64 root_objectid
,
4871 struct btrfs_disk_key
*key
, int level
,
4872 u64 hint
, u64 empty_size
)
4874 struct btrfs_key ins
;
4876 struct extent_buffer
*buf
;
4878 ret
= alloc_tree_block(trans
, root
, blocksize
, parent
, root_objectid
,
4879 key
, level
, empty_size
, hint
, (u64
)-1, &ins
);
4882 return ERR_PTR(ret
);
4885 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
,
4890 struct walk_control
{
4891 u64 refs
[BTRFS_MAX_LEVEL
];
4892 u64 flags
[BTRFS_MAX_LEVEL
];
4893 struct btrfs_key update_progress
;
4903 #define DROP_REFERENCE 1
4904 #define UPDATE_BACKREF 2
4906 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
4907 struct btrfs_root
*root
,
4908 struct walk_control
*wc
,
4909 struct btrfs_path
*path
)
4917 struct btrfs_key key
;
4918 struct extent_buffer
*eb
;
4923 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
4924 wc
->reada_count
= wc
->reada_count
* 2 / 3;
4925 wc
->reada_count
= max(wc
->reada_count
, 2);
4927 wc
->reada_count
= wc
->reada_count
* 3 / 2;
4928 wc
->reada_count
= min_t(int, wc
->reada_count
,
4929 BTRFS_NODEPTRS_PER_BLOCK(root
));
4932 eb
= path
->nodes
[wc
->level
];
4933 nritems
= btrfs_header_nritems(eb
);
4934 blocksize
= btrfs_level_size(root
, wc
->level
- 1);
4936 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
4937 if (nread
>= wc
->reada_count
)
4941 bytenr
= btrfs_node_blockptr(eb
, slot
);
4942 generation
= btrfs_node_ptr_generation(eb
, slot
);
4944 if (slot
== path
->slots
[wc
->level
])
4947 if (wc
->stage
== UPDATE_BACKREF
&&
4948 generation
<= root
->root_key
.offset
)
4951 if (wc
->stage
== DROP_REFERENCE
) {
4952 ret
= btrfs_lookup_extent_info(trans
, root
,
4960 if (!wc
->update_ref
||
4961 generation
<= root
->root_key
.offset
)
4963 btrfs_node_key_to_cpu(eb
, &key
, slot
);
4964 ret
= btrfs_comp_cpu_keys(&key
,
4965 &wc
->update_progress
);
4970 ret
= readahead_tree_block(root
, bytenr
, blocksize
,
4974 last
= bytenr
+ blocksize
;
4977 wc
->reada_slot
= slot
;
4981 * hepler to process tree block while walking down the tree.
4983 * when wc->stage == UPDATE_BACKREF, this function updates
4984 * back refs for pointers in the block.
4986 * NOTE: return value 1 means we should stop walking down.
4988 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
4989 struct btrfs_root
*root
,
4990 struct btrfs_path
*path
,
4991 struct walk_control
*wc
)
4993 int level
= wc
->level
;
4994 struct extent_buffer
*eb
= path
->nodes
[level
];
4995 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
4998 if (wc
->stage
== UPDATE_BACKREF
&&
4999 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
5003 * when reference count of tree block is 1, it won't increase
5004 * again. once full backref flag is set, we never clear it.
5006 if ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
5007 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
))) {
5008 BUG_ON(!path
->locks
[level
]);
5009 ret
= btrfs_lookup_extent_info(trans
, root
,
5014 BUG_ON(wc
->refs
[level
] == 0);
5017 if (wc
->stage
== DROP_REFERENCE
) {
5018 if (wc
->refs
[level
] > 1)
5021 if (path
->locks
[level
] && !wc
->keep_locks
) {
5022 btrfs_tree_unlock(eb
);
5023 path
->locks
[level
] = 0;
5028 /* wc->stage == UPDATE_BACKREF */
5029 if (!(wc
->flags
[level
] & flag
)) {
5030 BUG_ON(!path
->locks
[level
]);
5031 ret
= btrfs_inc_ref(trans
, root
, eb
, 1);
5033 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
5035 ret
= btrfs_set_disk_extent_flags(trans
, root
, eb
->start
,
5038 wc
->flags
[level
] |= flag
;
5042 * the block is shared by multiple trees, so it's not good to
5043 * keep the tree lock
5045 if (path
->locks
[level
] && level
> 0) {
5046 btrfs_tree_unlock(eb
);
5047 path
->locks
[level
] = 0;
5053 * hepler to process tree block pointer.
5055 * when wc->stage == DROP_REFERENCE, this function checks
5056 * reference count of the block pointed to. if the block
5057 * is shared and we need update back refs for the subtree
5058 * rooted at the block, this function changes wc->stage to
5059 * UPDATE_BACKREF. if the block is shared and there is no
5060 * need to update back, this function drops the reference
5063 * NOTE: return value 1 means we should stop walking down.
5065 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
5066 struct btrfs_root
*root
,
5067 struct btrfs_path
*path
,
5068 struct walk_control
*wc
)
5074 struct btrfs_key key
;
5075 struct extent_buffer
*next
;
5076 int level
= wc
->level
;
5080 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
5081 path
->slots
[level
]);
5083 * if the lower level block was created before the snapshot
5084 * was created, we know there is no need to update back refs
5087 if (wc
->stage
== UPDATE_BACKREF
&&
5088 generation
<= root
->root_key
.offset
)
5091 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
5092 blocksize
= btrfs_level_size(root
, level
- 1);
5094 next
= btrfs_find_tree_block(root
, bytenr
, blocksize
);
5096 next
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
5099 btrfs_tree_lock(next
);
5100 btrfs_set_lock_blocking(next
);
5102 if (wc
->stage
== DROP_REFERENCE
) {
5103 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, blocksize
,
5104 &wc
->refs
[level
- 1],
5105 &wc
->flags
[level
- 1]);
5107 BUG_ON(wc
->refs
[level
- 1] == 0);
5109 if (wc
->refs
[level
- 1] > 1) {
5110 if (!wc
->update_ref
||
5111 generation
<= root
->root_key
.offset
)
5114 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
5115 path
->slots
[level
]);
5116 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
5120 wc
->stage
= UPDATE_BACKREF
;
5121 wc
->shared_level
= level
- 1;
5125 if (!btrfs_buffer_uptodate(next
, generation
)) {
5126 btrfs_tree_unlock(next
);
5127 free_extent_buffer(next
);
5132 if (reada
&& level
== 1)
5133 reada_walk_down(trans
, root
, wc
, path
);
5134 next
= read_tree_block(root
, bytenr
, blocksize
, generation
);
5135 btrfs_tree_lock(next
);
5136 btrfs_set_lock_blocking(next
);
5140 BUG_ON(level
!= btrfs_header_level(next
));
5141 path
->nodes
[level
] = next
;
5142 path
->slots
[level
] = 0;
5143 path
->locks
[level
] = 1;
5149 wc
->refs
[level
- 1] = 0;
5150 wc
->flags
[level
- 1] = 0;
5152 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
5153 parent
= path
->nodes
[level
]->start
;
5155 BUG_ON(root
->root_key
.objectid
!=
5156 btrfs_header_owner(path
->nodes
[level
]));
5160 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
, parent
,
5161 root
->root_key
.objectid
, level
- 1, 0);
5164 btrfs_tree_unlock(next
);
5165 free_extent_buffer(next
);
5170 * hepler to process tree block while walking up the tree.
5172 * when wc->stage == DROP_REFERENCE, this function drops
5173 * reference count on the block.
5175 * when wc->stage == UPDATE_BACKREF, this function changes
5176 * wc->stage back to DROP_REFERENCE if we changed wc->stage
5177 * to UPDATE_BACKREF previously while processing the block.
5179 * NOTE: return value 1 means we should stop walking up.
5181 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
5182 struct btrfs_root
*root
,
5183 struct btrfs_path
*path
,
5184 struct walk_control
*wc
)
5187 int level
= wc
->level
;
5188 struct extent_buffer
*eb
= path
->nodes
[level
];
5191 if (wc
->stage
== UPDATE_BACKREF
) {
5192 BUG_ON(wc
->shared_level
< level
);
5193 if (level
< wc
->shared_level
)
5196 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
5200 wc
->stage
= DROP_REFERENCE
;
5201 wc
->shared_level
= -1;
5202 path
->slots
[level
] = 0;
5205 * check reference count again if the block isn't locked.
5206 * we should start walking down the tree again if reference
5209 if (!path
->locks
[level
]) {
5211 btrfs_tree_lock(eb
);
5212 btrfs_set_lock_blocking(eb
);
5213 path
->locks
[level
] = 1;
5215 ret
= btrfs_lookup_extent_info(trans
, root
,
5220 BUG_ON(wc
->refs
[level
] == 0);
5221 if (wc
->refs
[level
] == 1) {
5222 btrfs_tree_unlock(eb
);
5223 path
->locks
[level
] = 0;
5229 /* wc->stage == DROP_REFERENCE */
5230 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
5232 if (wc
->refs
[level
] == 1) {
5234 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
5235 ret
= btrfs_dec_ref(trans
, root
, eb
, 1);
5237 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
5240 /* make block locked assertion in clean_tree_block happy */
5241 if (!path
->locks
[level
] &&
5242 btrfs_header_generation(eb
) == trans
->transid
) {
5243 btrfs_tree_lock(eb
);
5244 btrfs_set_lock_blocking(eb
);
5245 path
->locks
[level
] = 1;
5247 clean_tree_block(trans
, root
, eb
);
5250 if (eb
== root
->node
) {
5251 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
5254 BUG_ON(root
->root_key
.objectid
!=
5255 btrfs_header_owner(eb
));
5257 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
5258 parent
= path
->nodes
[level
+ 1]->start
;
5260 BUG_ON(root
->root_key
.objectid
!=
5261 btrfs_header_owner(path
->nodes
[level
+ 1]));
5264 ret
= btrfs_free_extent(trans
, root
, eb
->start
, eb
->len
, parent
,
5265 root
->root_key
.objectid
, level
, 0);
5268 wc
->refs
[level
] = 0;
5269 wc
->flags
[level
] = 0;
5273 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
5274 struct btrfs_root
*root
,
5275 struct btrfs_path
*path
,
5276 struct walk_control
*wc
)
5278 int level
= wc
->level
;
5281 while (level
>= 0) {
5282 if (path
->slots
[level
] >=
5283 btrfs_header_nritems(path
->nodes
[level
]))
5286 ret
= walk_down_proc(trans
, root
, path
, wc
);
5293 ret
= do_walk_down(trans
, root
, path
, wc
);
5295 path
->slots
[level
]++;
5303 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
5304 struct btrfs_root
*root
,
5305 struct btrfs_path
*path
,
5306 struct walk_control
*wc
, int max_level
)
5308 int level
= wc
->level
;
5311 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
5312 while (level
< max_level
&& path
->nodes
[level
]) {
5314 if (path
->slots
[level
] + 1 <
5315 btrfs_header_nritems(path
->nodes
[level
])) {
5316 path
->slots
[level
]++;
5319 ret
= walk_up_proc(trans
, root
, path
, wc
);
5323 if (path
->locks
[level
]) {
5324 btrfs_tree_unlock(path
->nodes
[level
]);
5325 path
->locks
[level
] = 0;
5327 free_extent_buffer(path
->nodes
[level
]);
5328 path
->nodes
[level
] = NULL
;
5336 * drop a subvolume tree.
5338 * this function traverses the tree freeing any blocks that only
5339 * referenced by the tree.
5341 * when a shared tree block is found. this function decreases its
5342 * reference count by one. if update_ref is true, this function
5343 * also make sure backrefs for the shared block and all lower level
5344 * blocks are properly updated.
5346 int btrfs_drop_snapshot(struct btrfs_root
*root
, int update_ref
)
5348 struct btrfs_path
*path
;
5349 struct btrfs_trans_handle
*trans
;
5350 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
5351 struct btrfs_root_item
*root_item
= &root
->root_item
;
5352 struct walk_control
*wc
;
5353 struct btrfs_key key
;
5358 path
= btrfs_alloc_path();
5361 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
5364 trans
= btrfs_start_transaction(tree_root
, 1);
5366 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
5367 level
= btrfs_header_level(root
->node
);
5368 path
->nodes
[level
] = btrfs_lock_root_node(root
);
5369 btrfs_set_lock_blocking(path
->nodes
[level
]);
5370 path
->slots
[level
] = 0;
5371 path
->locks
[level
] = 1;
5372 memset(&wc
->update_progress
, 0,
5373 sizeof(wc
->update_progress
));
5375 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
5376 memcpy(&wc
->update_progress
, &key
,
5377 sizeof(wc
->update_progress
));
5379 level
= root_item
->drop_level
;
5381 path
->lowest_level
= level
;
5382 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5383 path
->lowest_level
= 0;
5391 * unlock our path, this is safe because only this
5392 * function is allowed to delete this snapshot
5394 btrfs_unlock_up_safe(path
, 0);
5396 level
= btrfs_header_level(root
->node
);
5398 btrfs_tree_lock(path
->nodes
[level
]);
5399 btrfs_set_lock_blocking(path
->nodes
[level
]);
5401 ret
= btrfs_lookup_extent_info(trans
, root
,
5402 path
->nodes
[level
]->start
,
5403 path
->nodes
[level
]->len
,
5407 BUG_ON(wc
->refs
[level
] == 0);
5409 if (level
== root_item
->drop_level
)
5412 btrfs_tree_unlock(path
->nodes
[level
]);
5413 WARN_ON(wc
->refs
[level
] != 1);
5419 wc
->shared_level
= -1;
5420 wc
->stage
= DROP_REFERENCE
;
5421 wc
->update_ref
= update_ref
;
5423 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
5426 ret
= walk_down_tree(trans
, root
, path
, wc
);
5432 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
5439 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
5443 if (wc
->stage
== DROP_REFERENCE
) {
5445 btrfs_node_key(path
->nodes
[level
],
5446 &root_item
->drop_progress
,
5447 path
->slots
[level
]);
5448 root_item
->drop_level
= level
;
5451 BUG_ON(wc
->level
== 0);
5452 if (trans
->transaction
->in_commit
||
5453 trans
->transaction
->delayed_refs
.flushing
) {
5454 ret
= btrfs_update_root(trans
, tree_root
,
5459 btrfs_end_transaction(trans
, tree_root
);
5460 trans
= btrfs_start_transaction(tree_root
, 1);
5462 unsigned long update
;
5463 update
= trans
->delayed_ref_updates
;
5464 trans
->delayed_ref_updates
= 0;
5466 btrfs_run_delayed_refs(trans
, tree_root
,
5470 btrfs_release_path(root
, path
);
5473 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
5476 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
5477 ret
= btrfs_find_last_root(tree_root
, root
->root_key
.objectid
,
5481 ret
= btrfs_del_orphan_item(trans
, tree_root
,
5482 root
->root_key
.objectid
);
5487 if (root
->in_radix
) {
5488 btrfs_free_fs_root(tree_root
->fs_info
, root
);
5490 free_extent_buffer(root
->node
);
5491 free_extent_buffer(root
->commit_root
);
5495 btrfs_end_transaction(trans
, tree_root
);
5497 btrfs_free_path(path
);
5502 * drop subtree rooted at tree block 'node'.
5504 * NOTE: this function will unlock and release tree block 'node'
5506 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
5507 struct btrfs_root
*root
,
5508 struct extent_buffer
*node
,
5509 struct extent_buffer
*parent
)
5511 struct btrfs_path
*path
;
5512 struct walk_control
*wc
;
5518 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
5520 path
= btrfs_alloc_path();
5523 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
5526 btrfs_assert_tree_locked(parent
);
5527 parent_level
= btrfs_header_level(parent
);
5528 extent_buffer_get(parent
);
5529 path
->nodes
[parent_level
] = parent
;
5530 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
5532 btrfs_assert_tree_locked(node
);
5533 level
= btrfs_header_level(node
);
5534 path
->nodes
[level
] = node
;
5535 path
->slots
[level
] = 0;
5536 path
->locks
[level
] = 1;
5538 wc
->refs
[parent_level
] = 1;
5539 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
5541 wc
->shared_level
= -1;
5542 wc
->stage
= DROP_REFERENCE
;
5545 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
5548 wret
= walk_down_tree(trans
, root
, path
, wc
);
5554 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
5562 btrfs_free_path(path
);
5567 static unsigned long calc_ra(unsigned long start
, unsigned long last
,
5570 return min(last
, start
+ nr
- 1);
5573 static noinline
int relocate_inode_pages(struct inode
*inode
, u64 start
,
5578 unsigned long first_index
;
5579 unsigned long last_index
;
5582 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
5583 struct file_ra_state
*ra
;
5584 struct btrfs_ordered_extent
*ordered
;
5585 unsigned int total_read
= 0;
5586 unsigned int total_dirty
= 0;
5589 ra
= kzalloc(sizeof(*ra
), GFP_NOFS
);
5591 mutex_lock(&inode
->i_mutex
);
5592 first_index
= start
>> PAGE_CACHE_SHIFT
;
5593 last_index
= (start
+ len
- 1) >> PAGE_CACHE_SHIFT
;
5595 /* make sure the dirty trick played by the caller work */
5596 ret
= invalidate_inode_pages2_range(inode
->i_mapping
,
5597 first_index
, last_index
);
5601 file_ra_state_init(ra
, inode
->i_mapping
);
5603 for (i
= first_index
; i
<= last_index
; i
++) {
5604 if (total_read
% ra
->ra_pages
== 0) {
5605 btrfs_force_ra(inode
->i_mapping
, ra
, NULL
, i
,
5606 calc_ra(i
, last_index
, ra
->ra_pages
));
5610 if (((u64
)i
<< PAGE_CACHE_SHIFT
) > i_size_read(inode
))
5612 page
= grab_cache_page(inode
->i_mapping
, i
);
5617 if (!PageUptodate(page
)) {
5618 btrfs_readpage(NULL
, page
);
5620 if (!PageUptodate(page
)) {
5622 page_cache_release(page
);
5627 wait_on_page_writeback(page
);
5629 page_start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
5630 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
5631 lock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
5633 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
5635 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
5637 page_cache_release(page
);
5638 btrfs_start_ordered_extent(inode
, ordered
, 1);
5639 btrfs_put_ordered_extent(ordered
);
5642 set_page_extent_mapped(page
);
5644 if (i
== first_index
)
5645 set_extent_bits(io_tree
, page_start
, page_end
,
5646 EXTENT_BOUNDARY
, GFP_NOFS
);
5647 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
5649 set_page_dirty(page
);
5652 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
5654 page_cache_release(page
);
5659 mutex_unlock(&inode
->i_mutex
);
5660 balance_dirty_pages_ratelimited_nr(inode
->i_mapping
, total_dirty
);
5664 static noinline
int relocate_data_extent(struct inode
*reloc_inode
,
5665 struct btrfs_key
*extent_key
,
5668 struct btrfs_root
*root
= BTRFS_I(reloc_inode
)->root
;
5669 struct extent_map_tree
*em_tree
= &BTRFS_I(reloc_inode
)->extent_tree
;
5670 struct extent_map
*em
;
5671 u64 start
= extent_key
->objectid
- offset
;
5672 u64 end
= start
+ extent_key
->offset
- 1;
5674 em
= alloc_extent_map(GFP_NOFS
);
5675 BUG_ON(!em
|| IS_ERR(em
));
5678 em
->len
= extent_key
->offset
;
5679 em
->block_len
= extent_key
->offset
;
5680 em
->block_start
= extent_key
->objectid
;
5681 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
5682 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
5684 /* setup extent map to cheat btrfs_readpage */
5685 lock_extent(&BTRFS_I(reloc_inode
)->io_tree
, start
, end
, GFP_NOFS
);
5688 write_lock(&em_tree
->lock
);
5689 ret
= add_extent_mapping(em_tree
, em
);
5690 write_unlock(&em_tree
->lock
);
5691 if (ret
!= -EEXIST
) {
5692 free_extent_map(em
);
5695 btrfs_drop_extent_cache(reloc_inode
, start
, end
, 0);
5697 unlock_extent(&BTRFS_I(reloc_inode
)->io_tree
, start
, end
, GFP_NOFS
);
5699 return relocate_inode_pages(reloc_inode
, start
, extent_key
->offset
);
5702 struct btrfs_ref_path
{
5704 u64 nodes
[BTRFS_MAX_LEVEL
];
5706 u64 root_generation
;
5713 struct btrfs_key node_keys
[BTRFS_MAX_LEVEL
];
5714 u64 new_nodes
[BTRFS_MAX_LEVEL
];
5717 struct disk_extent
{
5728 static int is_cowonly_root(u64 root_objectid
)
5730 if (root_objectid
== BTRFS_ROOT_TREE_OBJECTID
||
5731 root_objectid
== BTRFS_EXTENT_TREE_OBJECTID
||
5732 root_objectid
== BTRFS_CHUNK_TREE_OBJECTID
||
5733 root_objectid
== BTRFS_DEV_TREE_OBJECTID
||
5734 root_objectid
== BTRFS_TREE_LOG_OBJECTID
||
5735 root_objectid
== BTRFS_CSUM_TREE_OBJECTID
)
5740 static noinline
int __next_ref_path(struct btrfs_trans_handle
*trans
,
5741 struct btrfs_root
*extent_root
,
5742 struct btrfs_ref_path
*ref_path
,
5745 struct extent_buffer
*leaf
;
5746 struct btrfs_path
*path
;
5747 struct btrfs_extent_ref
*ref
;
5748 struct btrfs_key key
;
5749 struct btrfs_key found_key
;
5755 path
= btrfs_alloc_path();
5760 ref_path
->lowest_level
= -1;
5761 ref_path
->current_level
= -1;
5762 ref_path
->shared_level
= -1;
5766 level
= ref_path
->current_level
- 1;
5767 while (level
>= -1) {
5769 if (level
< ref_path
->lowest_level
)
5773 bytenr
= ref_path
->nodes
[level
];
5775 bytenr
= ref_path
->extent_start
;
5776 BUG_ON(bytenr
== 0);
5778 parent
= ref_path
->nodes
[level
+ 1];
5779 ref_path
->nodes
[level
+ 1] = 0;
5780 ref_path
->current_level
= level
;
5781 BUG_ON(parent
== 0);
5783 key
.objectid
= bytenr
;
5784 key
.offset
= parent
+ 1;
5785 key
.type
= BTRFS_EXTENT_REF_KEY
;
5787 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
, 0, 0);
5792 leaf
= path
->nodes
[0];
5793 nritems
= btrfs_header_nritems(leaf
);
5794 if (path
->slots
[0] >= nritems
) {
5795 ret
= btrfs_next_leaf(extent_root
, path
);
5800 leaf
= path
->nodes
[0];
5803 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5804 if (found_key
.objectid
== bytenr
&&
5805 found_key
.type
== BTRFS_EXTENT_REF_KEY
) {
5806 if (level
< ref_path
->shared_level
)
5807 ref_path
->shared_level
= level
;
5812 btrfs_release_path(extent_root
, path
);
5815 /* reached lowest level */
5819 level
= ref_path
->current_level
;
5820 while (level
< BTRFS_MAX_LEVEL
- 1) {
5824 bytenr
= ref_path
->nodes
[level
];
5826 bytenr
= ref_path
->extent_start
;
5828 BUG_ON(bytenr
== 0);
5830 key
.objectid
= bytenr
;
5832 key
.type
= BTRFS_EXTENT_REF_KEY
;
5834 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
, 0, 0);
5838 leaf
= path
->nodes
[0];
5839 nritems
= btrfs_header_nritems(leaf
);
5840 if (path
->slots
[0] >= nritems
) {
5841 ret
= btrfs_next_leaf(extent_root
, path
);
5845 /* the extent was freed by someone */
5846 if (ref_path
->lowest_level
== level
)
5848 btrfs_release_path(extent_root
, path
);
5851 leaf
= path
->nodes
[0];
5854 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5855 if (found_key
.objectid
!= bytenr
||
5856 found_key
.type
!= BTRFS_EXTENT_REF_KEY
) {
5857 /* the extent was freed by someone */
5858 if (ref_path
->lowest_level
== level
) {
5862 btrfs_release_path(extent_root
, path
);
5866 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
5867 struct btrfs_extent_ref
);
5868 ref_objectid
= btrfs_ref_objectid(leaf
, ref
);
5869 if (ref_objectid
< BTRFS_FIRST_FREE_OBJECTID
) {
5871 level
= (int)ref_objectid
;
5872 BUG_ON(level
>= BTRFS_MAX_LEVEL
);
5873 ref_path
->lowest_level
= level
;
5874 ref_path
->current_level
= level
;
5875 ref_path
->nodes
[level
] = bytenr
;
5877 WARN_ON(ref_objectid
!= level
);
5880 WARN_ON(level
!= -1);
5884 if (ref_path
->lowest_level
== level
) {
5885 ref_path
->owner_objectid
= ref_objectid
;
5886 ref_path
->num_refs
= btrfs_ref_num_refs(leaf
, ref
);
5890 * the block is tree root or the block isn't in reference
5893 if (found_key
.objectid
== found_key
.offset
||
5894 is_cowonly_root(btrfs_ref_root(leaf
, ref
))) {
5895 ref_path
->root_objectid
= btrfs_ref_root(leaf
, ref
);
5896 ref_path
->root_generation
=
5897 btrfs_ref_generation(leaf
, ref
);
5899 /* special reference from the tree log */
5900 ref_path
->nodes
[0] = found_key
.offset
;
5901 ref_path
->current_level
= 0;
5908 BUG_ON(ref_path
->nodes
[level
] != 0);
5909 ref_path
->nodes
[level
] = found_key
.offset
;
5910 ref_path
->current_level
= level
;
5913 * the reference was created in the running transaction,
5914 * no need to continue walking up.
5916 if (btrfs_ref_generation(leaf
, ref
) == trans
->transid
) {
5917 ref_path
->root_objectid
= btrfs_ref_root(leaf
, ref
);
5918 ref_path
->root_generation
=
5919 btrfs_ref_generation(leaf
, ref
);
5924 btrfs_release_path(extent_root
, path
);
5927 /* reached max tree level, but no tree root found. */
5930 btrfs_free_path(path
);
5934 static int btrfs_first_ref_path(struct btrfs_trans_handle
*trans
,
5935 struct btrfs_root
*extent_root
,
5936 struct btrfs_ref_path
*ref_path
,
5939 memset(ref_path
, 0, sizeof(*ref_path
));
5940 ref_path
->extent_start
= extent_start
;
5942 return __next_ref_path(trans
, extent_root
, ref_path
, 1);
5945 static int btrfs_next_ref_path(struct btrfs_trans_handle
*trans
,
5946 struct btrfs_root
*extent_root
,
5947 struct btrfs_ref_path
*ref_path
)
5949 return __next_ref_path(trans
, extent_root
, ref_path
, 0);
5952 static noinline
int get_new_locations(struct inode
*reloc_inode
,
5953 struct btrfs_key
*extent_key
,
5954 u64 offset
, int no_fragment
,
5955 struct disk_extent
**extents
,
5958 struct btrfs_root
*root
= BTRFS_I(reloc_inode
)->root
;
5959 struct btrfs_path
*path
;
5960 struct btrfs_file_extent_item
*fi
;
5961 struct extent_buffer
*leaf
;
5962 struct disk_extent
*exts
= *extents
;
5963 struct btrfs_key found_key
;
5968 int max
= *nr_extents
;
5971 WARN_ON(!no_fragment
&& *extents
);
5974 exts
= kmalloc(sizeof(*exts
) * max
, GFP_NOFS
);
5979 path
= btrfs_alloc_path();
5982 cur_pos
= extent_key
->objectid
- offset
;
5983 last_byte
= extent_key
->objectid
+ extent_key
->offset
;
5984 ret
= btrfs_lookup_file_extent(NULL
, root
, path
, reloc_inode
->i_ino
,
5994 leaf
= path
->nodes
[0];
5995 nritems
= btrfs_header_nritems(leaf
);
5996 if (path
->slots
[0] >= nritems
) {
5997 ret
= btrfs_next_leaf(root
, path
);
6002 leaf
= path
->nodes
[0];
6005 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6006 if (found_key
.offset
!= cur_pos
||
6007 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
||
6008 found_key
.objectid
!= reloc_inode
->i_ino
)
6011 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
6012 struct btrfs_file_extent_item
);
6013 if (btrfs_file_extent_type(leaf
, fi
) !=
6014 BTRFS_FILE_EXTENT_REG
||
6015 btrfs_file_extent_disk_bytenr(leaf
, fi
) == 0)
6019 struct disk_extent
*old
= exts
;
6021 exts
= kzalloc(sizeof(*exts
) * max
, GFP_NOFS
);
6022 memcpy(exts
, old
, sizeof(*exts
) * nr
);
6023 if (old
!= *extents
)
6027 exts
[nr
].disk_bytenr
=
6028 btrfs_file_extent_disk_bytenr(leaf
, fi
);
6029 exts
[nr
].disk_num_bytes
=
6030 btrfs_file_extent_disk_num_bytes(leaf
, fi
);
6031 exts
[nr
].offset
= btrfs_file_extent_offset(leaf
, fi
);
6032 exts
[nr
].num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
6033 exts
[nr
].ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
6034 exts
[nr
].compression
= btrfs_file_extent_compression(leaf
, fi
);
6035 exts
[nr
].encryption
= btrfs_file_extent_encryption(leaf
, fi
);
6036 exts
[nr
].other_encoding
= btrfs_file_extent_other_encoding(leaf
,
6038 BUG_ON(exts
[nr
].offset
> 0);
6039 BUG_ON(exts
[nr
].compression
|| exts
[nr
].encryption
);
6040 BUG_ON(exts
[nr
].num_bytes
!= exts
[nr
].disk_num_bytes
);
6042 cur_pos
+= exts
[nr
].num_bytes
;
6045 if (cur_pos
+ offset
>= last_byte
)
6055 BUG_ON(cur_pos
+ offset
> last_byte
);
6056 if (cur_pos
+ offset
< last_byte
) {
6062 btrfs_free_path(path
);
6064 if (exts
!= *extents
)
6073 static noinline
int replace_one_extent(struct btrfs_trans_handle
*trans
,
6074 struct btrfs_root
*root
,
6075 struct btrfs_path
*path
,
6076 struct btrfs_key
*extent_key
,
6077 struct btrfs_key
*leaf_key
,
6078 struct btrfs_ref_path
*ref_path
,
6079 struct disk_extent
*new_extents
,
6082 struct extent_buffer
*leaf
;
6083 struct btrfs_file_extent_item
*fi
;
6084 struct inode
*inode
= NULL
;
6085 struct btrfs_key key
;
6090 u64 search_end
= (u64
)-1;
6093 int extent_locked
= 0;
6097 memcpy(&key
, leaf_key
, sizeof(key
));
6098 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
) {
6099 if (key
.objectid
< ref_path
->owner_objectid
||
6100 (key
.objectid
== ref_path
->owner_objectid
&&
6101 key
.type
< BTRFS_EXTENT_DATA_KEY
)) {
6102 key
.objectid
= ref_path
->owner_objectid
;
6103 key
.type
= BTRFS_EXTENT_DATA_KEY
;
6109 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
6113 leaf
= path
->nodes
[0];
6114 nritems
= btrfs_header_nritems(leaf
);
6116 if (extent_locked
&& ret
> 0) {
6118 * the file extent item was modified by someone
6119 * before the extent got locked.
6121 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
6122 lock_end
, GFP_NOFS
);
6126 if (path
->slots
[0] >= nritems
) {
6127 if (++nr_scaned
> 2)
6130 BUG_ON(extent_locked
);
6131 ret
= btrfs_next_leaf(root
, path
);
6136 leaf
= path
->nodes
[0];
6137 nritems
= btrfs_header_nritems(leaf
);
6140 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
6142 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
) {
6143 if ((key
.objectid
> ref_path
->owner_objectid
) ||
6144 (key
.objectid
== ref_path
->owner_objectid
&&
6145 key
.type
> BTRFS_EXTENT_DATA_KEY
) ||
6146 key
.offset
>= search_end
)
6150 if (inode
&& key
.objectid
!= inode
->i_ino
) {
6151 BUG_ON(extent_locked
);
6152 btrfs_release_path(root
, path
);
6153 mutex_unlock(&inode
->i_mutex
);
6159 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
6164 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
6165 struct btrfs_file_extent_item
);
6166 extent_type
= btrfs_file_extent_type(leaf
, fi
);
6167 if ((extent_type
!= BTRFS_FILE_EXTENT_REG
&&
6168 extent_type
!= BTRFS_FILE_EXTENT_PREALLOC
) ||
6169 (btrfs_file_extent_disk_bytenr(leaf
, fi
) !=
6170 extent_key
->objectid
)) {
6176 num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
6177 ext_offset
= btrfs_file_extent_offset(leaf
, fi
);
6179 if (search_end
== (u64
)-1) {
6180 search_end
= key
.offset
- ext_offset
+
6181 btrfs_file_extent_ram_bytes(leaf
, fi
);
6184 if (!extent_locked
) {
6185 lock_start
= key
.offset
;
6186 lock_end
= lock_start
+ num_bytes
- 1;
6188 if (lock_start
> key
.offset
||
6189 lock_end
+ 1 < key
.offset
+ num_bytes
) {
6190 unlock_extent(&BTRFS_I(inode
)->io_tree
,
6191 lock_start
, lock_end
, GFP_NOFS
);
6197 btrfs_release_path(root
, path
);
6199 inode
= btrfs_iget_locked(root
->fs_info
->sb
,
6200 key
.objectid
, root
);
6201 if (inode
->i_state
& I_NEW
) {
6202 BTRFS_I(inode
)->root
= root
;
6203 BTRFS_I(inode
)->location
.objectid
=
6205 BTRFS_I(inode
)->location
.type
=
6206 BTRFS_INODE_ITEM_KEY
;
6207 BTRFS_I(inode
)->location
.offset
= 0;
6208 btrfs_read_locked_inode(inode
);
6209 unlock_new_inode(inode
);
6212 * some code call btrfs_commit_transaction while
6213 * holding the i_mutex, so we can't use mutex_lock
6216 if (is_bad_inode(inode
) ||
6217 !mutex_trylock(&inode
->i_mutex
)) {
6220 key
.offset
= (u64
)-1;
6225 if (!extent_locked
) {
6226 struct btrfs_ordered_extent
*ordered
;
6228 btrfs_release_path(root
, path
);
6230 lock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
6231 lock_end
, GFP_NOFS
);
6232 ordered
= btrfs_lookup_first_ordered_extent(inode
,
6235 ordered
->file_offset
<= lock_end
&&
6236 ordered
->file_offset
+ ordered
->len
> lock_start
) {
6237 unlock_extent(&BTRFS_I(inode
)->io_tree
,
6238 lock_start
, lock_end
, GFP_NOFS
);
6239 btrfs_start_ordered_extent(inode
, ordered
, 1);
6240 btrfs_put_ordered_extent(ordered
);
6241 key
.offset
+= num_bytes
;
6245 btrfs_put_ordered_extent(ordered
);
6251 if (nr_extents
== 1) {
6252 /* update extent pointer in place */
6253 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
6254 new_extents
[0].disk_bytenr
);
6255 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
6256 new_extents
[0].disk_num_bytes
);
6257 btrfs_mark_buffer_dirty(leaf
);
6259 btrfs_drop_extent_cache(inode
, key
.offset
,
6260 key
.offset
+ num_bytes
- 1, 0);
6262 ret
= btrfs_inc_extent_ref(trans
, root
,
6263 new_extents
[0].disk_bytenr
,
6264 new_extents
[0].disk_num_bytes
,
6266 root
->root_key
.objectid
,
6271 ret
= btrfs_free_extent(trans
, root
,
6272 extent_key
->objectid
,
6275 btrfs_header_owner(leaf
),
6276 btrfs_header_generation(leaf
),
6280 btrfs_release_path(root
, path
);
6281 key
.offset
+= num_bytes
;
6289 * drop old extent pointer at first, then insert the
6290 * new pointers one bye one
6292 btrfs_release_path(root
, path
);
6293 ret
= btrfs_drop_extents(trans
, root
, inode
, key
.offset
,
6294 key
.offset
+ num_bytes
,
6295 key
.offset
, &alloc_hint
);
6298 for (i
= 0; i
< nr_extents
; i
++) {
6299 if (ext_offset
>= new_extents
[i
].num_bytes
) {
6300 ext_offset
-= new_extents
[i
].num_bytes
;
6303 extent_len
= min(new_extents
[i
].num_bytes
-
6304 ext_offset
, num_bytes
);
6306 ret
= btrfs_insert_empty_item(trans
, root
,
6311 leaf
= path
->nodes
[0];
6312 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
6313 struct btrfs_file_extent_item
);
6314 btrfs_set_file_extent_generation(leaf
, fi
,
6316 btrfs_set_file_extent_type(leaf
, fi
,
6317 BTRFS_FILE_EXTENT_REG
);
6318 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
6319 new_extents
[i
].disk_bytenr
);
6320 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
6321 new_extents
[i
].disk_num_bytes
);
6322 btrfs_set_file_extent_ram_bytes(leaf
, fi
,
6323 new_extents
[i
].ram_bytes
);
6325 btrfs_set_file_extent_compression(leaf
, fi
,
6326 new_extents
[i
].compression
);
6327 btrfs_set_file_extent_encryption(leaf
, fi
,
6328 new_extents
[i
].encryption
);
6329 btrfs_set_file_extent_other_encoding(leaf
, fi
,
6330 new_extents
[i
].other_encoding
);
6332 btrfs_set_file_extent_num_bytes(leaf
, fi
,
6334 ext_offset
+= new_extents
[i
].offset
;
6335 btrfs_set_file_extent_offset(leaf
, fi
,
6337 btrfs_mark_buffer_dirty(leaf
);
6339 btrfs_drop_extent_cache(inode
, key
.offset
,
6340 key
.offset
+ extent_len
- 1, 0);
6342 ret
= btrfs_inc_extent_ref(trans
, root
,
6343 new_extents
[i
].disk_bytenr
,
6344 new_extents
[i
].disk_num_bytes
,
6346 root
->root_key
.objectid
,
6347 trans
->transid
, key
.objectid
);
6349 btrfs_release_path(root
, path
);
6351 inode_add_bytes(inode
, extent_len
);
6354 num_bytes
-= extent_len
;
6355 key
.offset
+= extent_len
;
6360 BUG_ON(i
>= nr_extents
);
6364 if (extent_locked
) {
6365 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
6366 lock_end
, GFP_NOFS
);
6370 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
&&
6371 key
.offset
>= search_end
)
6378 btrfs_release_path(root
, path
);
6380 mutex_unlock(&inode
->i_mutex
);
6381 if (extent_locked
) {
6382 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
6383 lock_end
, GFP_NOFS
);
6390 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle
*trans
,
6391 struct btrfs_root
*root
,
6392 struct extent_buffer
*buf
, u64 orig_start
)
6397 BUG_ON(btrfs_header_generation(buf
) != trans
->transid
);
6398 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
6400 level
= btrfs_header_level(buf
);
6402 struct btrfs_leaf_ref
*ref
;
6403 struct btrfs_leaf_ref
*orig_ref
;
6405 orig_ref
= btrfs_lookup_leaf_ref(root
, orig_start
);
6409 ref
= btrfs_alloc_leaf_ref(root
, orig_ref
->nritems
);
6411 btrfs_free_leaf_ref(root
, orig_ref
);
6415 ref
->nritems
= orig_ref
->nritems
;
6416 memcpy(ref
->extents
, orig_ref
->extents
,
6417 sizeof(ref
->extents
[0]) * ref
->nritems
);
6419 btrfs_free_leaf_ref(root
, orig_ref
);
6421 ref
->root_gen
= trans
->transid
;
6422 ref
->bytenr
= buf
->start
;
6423 ref
->owner
= btrfs_header_owner(buf
);
6424 ref
->generation
= btrfs_header_generation(buf
);
6426 ret
= btrfs_add_leaf_ref(root
, ref
, 0);
6428 btrfs_free_leaf_ref(root
, ref
);
6433 static noinline
int invalidate_extent_cache(struct btrfs_root
*root
,
6434 struct extent_buffer
*leaf
,
6435 struct btrfs_block_group_cache
*group
,
6436 struct btrfs_root
*target_root
)
6438 struct btrfs_key key
;
6439 struct inode
*inode
= NULL
;
6440 struct btrfs_file_extent_item
*fi
;
6442 u64 skip_objectid
= 0;
6446 nritems
= btrfs_header_nritems(leaf
);
6447 for (i
= 0; i
< nritems
; i
++) {
6448 btrfs_item_key_to_cpu(leaf
, &key
, i
);
6449 if (key
.objectid
== skip_objectid
||
6450 key
.type
!= BTRFS_EXTENT_DATA_KEY
)
6452 fi
= btrfs_item_ptr(leaf
, i
, struct btrfs_file_extent_item
);
6453 if (btrfs_file_extent_type(leaf
, fi
) ==
6454 BTRFS_FILE_EXTENT_INLINE
)
6456 if (btrfs_file_extent_disk_bytenr(leaf
, fi
) == 0)
6458 if (!inode
|| inode
->i_ino
!= key
.objectid
) {
6460 inode
= btrfs_ilookup(target_root
->fs_info
->sb
,
6461 key
.objectid
, target_root
, 1);
6464 skip_objectid
= key
.objectid
;
6467 num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
6469 lock_extent(&BTRFS_I(inode
)->io_tree
, key
.offset
,
6470 key
.offset
+ num_bytes
- 1, GFP_NOFS
);
6471 btrfs_drop_extent_cache(inode
, key
.offset
,
6472 key
.offset
+ num_bytes
- 1, 1);
6473 unlock_extent(&BTRFS_I(inode
)->io_tree
, key
.offset
,
6474 key
.offset
+ num_bytes
- 1, GFP_NOFS
);
6481 static noinline
int replace_extents_in_leaf(struct btrfs_trans_handle
*trans
,
6482 struct btrfs_root
*root
,
6483 struct extent_buffer
*leaf
,
6484 struct btrfs_block_group_cache
*group
,
6485 struct inode
*reloc_inode
)
6487 struct btrfs_key key
;
6488 struct btrfs_key extent_key
;
6489 struct btrfs_file_extent_item
*fi
;
6490 struct btrfs_leaf_ref
*ref
;
6491 struct disk_extent
*new_extent
;
6500 new_extent
= kmalloc(sizeof(*new_extent
), GFP_NOFS
);
6501 BUG_ON(!new_extent
);
6503 ref
= btrfs_lookup_leaf_ref(root
, leaf
->start
);
6507 nritems
= btrfs_header_nritems(leaf
);
6508 for (i
= 0; i
< nritems
; i
++) {
6509 btrfs_item_key_to_cpu(leaf
, &key
, i
);
6510 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
6512 fi
= btrfs_item_ptr(leaf
, i
, struct btrfs_file_extent_item
);
6513 if (btrfs_file_extent_type(leaf
, fi
) ==
6514 BTRFS_FILE_EXTENT_INLINE
)
6516 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
6517 num_bytes
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
6522 if (bytenr
>= group
->key
.objectid
+ group
->key
.offset
||
6523 bytenr
+ num_bytes
<= group
->key
.objectid
)
6526 extent_key
.objectid
= bytenr
;
6527 extent_key
.offset
= num_bytes
;
6528 extent_key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6530 ret
= get_new_locations(reloc_inode
, &extent_key
,
6531 group
->key
.objectid
, 1,
6532 &new_extent
, &nr_extent
);
6537 BUG_ON(ref
->extents
[ext_index
].bytenr
!= bytenr
);
6538 BUG_ON(ref
->extents
[ext_index
].num_bytes
!= num_bytes
);
6539 ref
->extents
[ext_index
].bytenr
= new_extent
->disk_bytenr
;
6540 ref
->extents
[ext_index
].num_bytes
= new_extent
->disk_num_bytes
;
6542 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
6543 new_extent
->disk_bytenr
);
6544 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
6545 new_extent
->disk_num_bytes
);
6546 btrfs_mark_buffer_dirty(leaf
);
6548 ret
= btrfs_inc_extent_ref(trans
, root
,
6549 new_extent
->disk_bytenr
,
6550 new_extent
->disk_num_bytes
,
6552 root
->root_key
.objectid
,
6553 trans
->transid
, key
.objectid
);
6556 ret
= btrfs_free_extent(trans
, root
,
6557 bytenr
, num_bytes
, leaf
->start
,
6558 btrfs_header_owner(leaf
),
6559 btrfs_header_generation(leaf
),
6565 BUG_ON(ext_index
+ 1 != ref
->nritems
);
6566 btrfs_free_leaf_ref(root
, ref
);
6570 int btrfs_free_reloc_root(struct btrfs_trans_handle
*trans
,
6571 struct btrfs_root
*root
)
6573 struct btrfs_root
*reloc_root
;
6576 if (root
->reloc_root
) {
6577 reloc_root
= root
->reloc_root
;
6578 root
->reloc_root
= NULL
;
6579 list_add(&reloc_root
->dead_list
,
6580 &root
->fs_info
->dead_reloc_roots
);
6582 btrfs_set_root_bytenr(&reloc_root
->root_item
,
6583 reloc_root
->node
->start
);
6584 btrfs_set_root_level(&root
->root_item
,
6585 btrfs_header_level(reloc_root
->node
));
6586 memset(&reloc_root
->root_item
.drop_progress
, 0,
6587 sizeof(struct btrfs_disk_key
));
6588 reloc_root
->root_item
.drop_level
= 0;
6590 ret
= btrfs_update_root(trans
, root
->fs_info
->tree_root
,
6591 &reloc_root
->root_key
,
6592 &reloc_root
->root_item
);
6598 int btrfs_drop_dead_reloc_roots(struct btrfs_root
*root
)
6600 struct btrfs_trans_handle
*trans
;
6601 struct btrfs_root
*reloc_root
;
6602 struct btrfs_root
*prev_root
= NULL
;
6603 struct list_head dead_roots
;
6607 INIT_LIST_HEAD(&dead_roots
);
6608 list_splice_init(&root
->fs_info
->dead_reloc_roots
, &dead_roots
);
6610 while (!list_empty(&dead_roots
)) {
6611 reloc_root
= list_entry(dead_roots
.prev
,
6612 struct btrfs_root
, dead_list
);
6613 list_del_init(&reloc_root
->dead_list
);
6615 BUG_ON(reloc_root
->commit_root
!= NULL
);
6617 trans
= btrfs_join_transaction(root
, 1);
6620 mutex_lock(&root
->fs_info
->drop_mutex
);
6621 ret
= btrfs_drop_snapshot(trans
, reloc_root
);
6624 mutex_unlock(&root
->fs_info
->drop_mutex
);
6626 nr
= trans
->blocks_used
;
6627 ret
= btrfs_end_transaction(trans
, root
);
6629 btrfs_btree_balance_dirty(root
, nr
);
6632 free_extent_buffer(reloc_root
->node
);
6634 ret
= btrfs_del_root(trans
, root
->fs_info
->tree_root
,
6635 &reloc_root
->root_key
);
6637 mutex_unlock(&root
->fs_info
->drop_mutex
);
6639 nr
= trans
->blocks_used
;
6640 ret
= btrfs_end_transaction(trans
, root
);
6642 btrfs_btree_balance_dirty(root
, nr
);
6645 prev_root
= reloc_root
;
6648 btrfs_remove_leaf_refs(prev_root
, (u64
)-1, 0);
6654 int btrfs_add_dead_reloc_root(struct btrfs_root
*root
)
6656 list_add(&root
->dead_list
, &root
->fs_info
->dead_reloc_roots
);
6660 int btrfs_cleanup_reloc_trees(struct btrfs_root
*root
)
6662 struct btrfs_root
*reloc_root
;
6663 struct btrfs_trans_handle
*trans
;
6664 struct btrfs_key location
;
6668 mutex_lock(&root
->fs_info
->tree_reloc_mutex
);
6669 ret
= btrfs_find_dead_roots(root
, BTRFS_TREE_RELOC_OBJECTID
, NULL
);
6671 found
= !list_empty(&root
->fs_info
->dead_reloc_roots
);
6672 mutex_unlock(&root
->fs_info
->tree_reloc_mutex
);
6675 trans
= btrfs_start_transaction(root
, 1);
6677 ret
= btrfs_commit_transaction(trans
, root
);
6681 location
.objectid
= BTRFS_DATA_RELOC_TREE_OBJECTID
;
6682 location
.offset
= (u64
)-1;
6683 location
.type
= BTRFS_ROOT_ITEM_KEY
;
6685 reloc_root
= btrfs_read_fs_root_no_name(root
->fs_info
, &location
);
6686 BUG_ON(!reloc_root
);
6687 btrfs_orphan_cleanup(reloc_root
);
6691 static noinline
int init_reloc_tree(struct btrfs_trans_handle
*trans
,
6692 struct btrfs_root
*root
)
6694 struct btrfs_root
*reloc_root
;
6695 struct extent_buffer
*eb
;
6696 struct btrfs_root_item
*root_item
;
6697 struct btrfs_key root_key
;
6700 BUG_ON(!root
->ref_cows
);
6701 if (root
->reloc_root
)
6704 root_item
= kmalloc(sizeof(*root_item
), GFP_NOFS
);
6707 ret
= btrfs_copy_root(trans
, root
, root
->commit_root
,
6708 &eb
, BTRFS_TREE_RELOC_OBJECTID
);
6711 root_key
.objectid
= BTRFS_TREE_RELOC_OBJECTID
;
6712 root_key
.offset
= root
->root_key
.objectid
;
6713 root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
6715 memcpy(root_item
, &root
->root_item
, sizeof(root_item
));
6716 btrfs_set_root_refs(root_item
, 0);
6717 btrfs_set_root_bytenr(root_item
, eb
->start
);
6718 btrfs_set_root_level(root_item
, btrfs_header_level(eb
));
6719 btrfs_set_root_generation(root_item
, trans
->transid
);
6721 btrfs_tree_unlock(eb
);
6722 free_extent_buffer(eb
);
6724 ret
= btrfs_insert_root(trans
, root
->fs_info
->tree_root
,
6725 &root_key
, root_item
);
6729 reloc_root
= btrfs_read_fs_root_no_radix(root
->fs_info
->tree_root
,
6731 BUG_ON(!reloc_root
);
6732 reloc_root
->last_trans
= trans
->transid
;
6733 reloc_root
->commit_root
= NULL
;
6734 reloc_root
->ref_tree
= &root
->fs_info
->reloc_ref_tree
;
6736 root
->reloc_root
= reloc_root
;
6741 * Core function of space balance.
6743 * The idea is using reloc trees to relocate tree blocks in reference
6744 * counted roots. There is one reloc tree for each subvol, and all
6745 * reloc trees share same root key objectid. Reloc trees are snapshots
6746 * of the latest committed roots of subvols (root->commit_root).
6748 * To relocate a tree block referenced by a subvol, there are two steps.
6749 * COW the block through subvol's reloc tree, then update block pointer
6750 * in the subvol to point to the new block. Since all reloc trees share
6751 * same root key objectid, doing special handing for tree blocks owned
6752 * by them is easy. Once a tree block has been COWed in one reloc tree,
6753 * we can use the resulting new block directly when the same block is
6754 * required to COW again through other reloc trees. By this way, relocated
6755 * tree blocks are shared between reloc trees, so they are also shared
6758 static noinline
int relocate_one_path(struct btrfs_trans_handle
*trans
,
6759 struct btrfs_root
*root
,
6760 struct btrfs_path
*path
,
6761 struct btrfs_key
*first_key
,
6762 struct btrfs_ref_path
*ref_path
,
6763 struct btrfs_block_group_cache
*group
,
6764 struct inode
*reloc_inode
)
6766 struct btrfs_root
*reloc_root
;
6767 struct extent_buffer
*eb
= NULL
;
6768 struct btrfs_key
*keys
;
6772 int lowest_level
= 0;
6775 if (ref_path
->owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
)
6776 lowest_level
= ref_path
->owner_objectid
;
6778 if (!root
->ref_cows
) {
6779 path
->lowest_level
= lowest_level
;
6780 ret
= btrfs_search_slot(trans
, root
, first_key
, path
, 0, 1);
6782 path
->lowest_level
= 0;
6783 btrfs_release_path(root
, path
);
6787 mutex_lock(&root
->fs_info
->tree_reloc_mutex
);
6788 ret
= init_reloc_tree(trans
, root
);
6790 reloc_root
= root
->reloc_root
;
6792 shared_level
= ref_path
->shared_level
;
6793 ref_path
->shared_level
= BTRFS_MAX_LEVEL
- 1;
6795 keys
= ref_path
->node_keys
;
6796 nodes
= ref_path
->new_nodes
;
6797 memset(&keys
[shared_level
+ 1], 0,
6798 sizeof(*keys
) * (BTRFS_MAX_LEVEL
- shared_level
- 1));
6799 memset(&nodes
[shared_level
+ 1], 0,
6800 sizeof(*nodes
) * (BTRFS_MAX_LEVEL
- shared_level
- 1));
6802 if (nodes
[lowest_level
] == 0) {
6803 path
->lowest_level
= lowest_level
;
6804 ret
= btrfs_search_slot(trans
, reloc_root
, first_key
, path
,
6807 for (level
= lowest_level
; level
< BTRFS_MAX_LEVEL
; level
++) {
6808 eb
= path
->nodes
[level
];
6809 if (!eb
|| eb
== reloc_root
->node
)
6811 nodes
[level
] = eb
->start
;
6813 btrfs_item_key_to_cpu(eb
, &keys
[level
], 0);
6815 btrfs_node_key_to_cpu(eb
, &keys
[level
], 0);
6818 ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
6819 eb
= path
->nodes
[0];
6820 ret
= replace_extents_in_leaf(trans
, reloc_root
, eb
,
6821 group
, reloc_inode
);
6824 btrfs_release_path(reloc_root
, path
);
6826 ret
= btrfs_merge_path(trans
, reloc_root
, keys
, nodes
,
6832 * replace tree blocks in the fs tree with tree blocks in
6835 ret
= btrfs_merge_path(trans
, root
, keys
, nodes
, lowest_level
);
6838 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
6839 ret
= btrfs_search_slot(trans
, reloc_root
, first_key
, path
,
6842 extent_buffer_get(path
->nodes
[0]);
6843 eb
= path
->nodes
[0];
6844 btrfs_release_path(reloc_root
, path
);
6845 ret
= invalidate_extent_cache(reloc_root
, eb
, group
, root
);
6847 free_extent_buffer(eb
);
6850 mutex_unlock(&root
->fs_info
->tree_reloc_mutex
);
6851 path
->lowest_level
= 0;
6855 static noinline
int relocate_tree_block(struct btrfs_trans_handle
*trans
,
6856 struct btrfs_root
*root
,
6857 struct btrfs_path
*path
,
6858 struct btrfs_key
*first_key
,
6859 struct btrfs_ref_path
*ref_path
)
6863 ret
= relocate_one_path(trans
, root
, path
, first_key
,
6864 ref_path
, NULL
, NULL
);
6870 static noinline
int del_extent_zero(struct btrfs_trans_handle
*trans
,
6871 struct btrfs_root
*extent_root
,
6872 struct btrfs_path
*path
,
6873 struct btrfs_key
*extent_key
)
6877 ret
= btrfs_search_slot(trans
, extent_root
, extent_key
, path
, -1, 1);
6880 ret
= btrfs_del_item(trans
, extent_root
, path
);
6882 btrfs_release_path(extent_root
, path
);
6886 static noinline
struct btrfs_root
*read_ref_root(struct btrfs_fs_info
*fs_info
,
6887 struct btrfs_ref_path
*ref_path
)
6889 struct btrfs_key root_key
;
6891 root_key
.objectid
= ref_path
->root_objectid
;
6892 root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
6893 if (is_cowonly_root(ref_path
->root_objectid
))
6894 root_key
.offset
= 0;
6896 root_key
.offset
= (u64
)-1;
6898 return btrfs_read_fs_root_no_name(fs_info
, &root_key
);
6901 static noinline
int relocate_one_extent(struct btrfs_root
*extent_root
,
6902 struct btrfs_path
*path
,
6903 struct btrfs_key
*extent_key
,
6904 struct btrfs_block_group_cache
*group
,
6905 struct inode
*reloc_inode
, int pass
)
6907 struct btrfs_trans_handle
*trans
;
6908 struct btrfs_root
*found_root
;
6909 struct btrfs_ref_path
*ref_path
= NULL
;
6910 struct disk_extent
*new_extents
= NULL
;
6915 struct btrfs_key first_key
;
6919 trans
= btrfs_start_transaction(extent_root
, 1);
6922 if (extent_key
->objectid
== 0) {
6923 ret
= del_extent_zero(trans
, extent_root
, path
, extent_key
);
6927 ref_path
= kmalloc(sizeof(*ref_path
), GFP_NOFS
);
6933 for (loops
= 0; ; loops
++) {
6935 ret
= btrfs_first_ref_path(trans
, extent_root
, ref_path
,
6936 extent_key
->objectid
);
6938 ret
= btrfs_next_ref_path(trans
, extent_root
, ref_path
);
6945 if (ref_path
->root_objectid
== BTRFS_TREE_LOG_OBJECTID
||
6946 ref_path
->root_objectid
== BTRFS_TREE_RELOC_OBJECTID
)
6949 found_root
= read_ref_root(extent_root
->fs_info
, ref_path
);
6950 BUG_ON(!found_root
);
6952 * for reference counted tree, only process reference paths
6953 * rooted at the latest committed root.
6955 if (found_root
->ref_cows
&&
6956 ref_path
->root_generation
!= found_root
->root_key
.offset
)
6959 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
6962 * copy data extents to new locations
6964 u64 group_start
= group
->key
.objectid
;
6965 ret
= relocate_data_extent(reloc_inode
,
6974 level
= ref_path
->owner_objectid
;
6977 if (prev_block
!= ref_path
->nodes
[level
]) {
6978 struct extent_buffer
*eb
;
6979 u64 block_start
= ref_path
->nodes
[level
];
6980 u64 block_size
= btrfs_level_size(found_root
, level
);
6982 eb
= read_tree_block(found_root
, block_start
,
6984 btrfs_tree_lock(eb
);
6985 BUG_ON(level
!= btrfs_header_level(eb
));
6988 btrfs_item_key_to_cpu(eb
, &first_key
, 0);
6990 btrfs_node_key_to_cpu(eb
, &first_key
, 0);
6992 btrfs_tree_unlock(eb
);
6993 free_extent_buffer(eb
);
6994 prev_block
= block_start
;
6997 mutex_lock(&extent_root
->fs_info
->trans_mutex
);
6998 btrfs_record_root_in_trans(found_root
);
6999 mutex_unlock(&extent_root
->fs_info
->trans_mutex
);
7000 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
7002 * try to update data extent references while
7003 * keeping metadata shared between snapshots.
7006 ret
= relocate_one_path(trans
, found_root
,
7007 path
, &first_key
, ref_path
,
7008 group
, reloc_inode
);
7014 * use fallback method to process the remaining
7018 u64 group_start
= group
->key
.objectid
;
7019 new_extents
= kmalloc(sizeof(*new_extents
),
7022 ret
= get_new_locations(reloc_inode
,
7030 ret
= replace_one_extent(trans
, found_root
,
7032 &first_key
, ref_path
,
7033 new_extents
, nr_extents
);
7035 ret
= relocate_tree_block(trans
, found_root
, path
,
7036 &first_key
, ref_path
);
7043 btrfs_end_transaction(trans
, extent_root
);
7050 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
7053 u64 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
7054 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
7056 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
7057 if (num_devices
== 1) {
7058 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
7059 stripped
= flags
& ~stripped
;
7061 /* turn raid0 into single device chunks */
7062 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
7065 /* turn mirroring into duplication */
7066 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
7067 BTRFS_BLOCK_GROUP_RAID10
))
7068 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
7071 /* they already had raid on here, just return */
7072 if (flags
& stripped
)
7075 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
7076 stripped
= flags
& ~stripped
;
7078 /* switch duplicated blocks with raid1 */
7079 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
7080 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
7082 /* turn single device chunks into raid0 */
7083 return stripped
| BTRFS_BLOCK_GROUP_RAID0
;
7088 static int __alloc_chunk_for_shrink(struct btrfs_root
*root
,
7089 struct btrfs_block_group_cache
*shrink_block_group
,
7092 struct btrfs_trans_handle
*trans
;
7093 u64 new_alloc_flags
;
7096 spin_lock(&shrink_block_group
->lock
);
7097 if (btrfs_block_group_used(&shrink_block_group
->item
) +
7098 shrink_block_group
->reserved
> 0) {
7099 spin_unlock(&shrink_block_group
->lock
);
7101 trans
= btrfs_start_transaction(root
, 1);
7102 spin_lock(&shrink_block_group
->lock
);
7104 new_alloc_flags
= update_block_group_flags(root
,
7105 shrink_block_group
->flags
);
7106 if (new_alloc_flags
!= shrink_block_group
->flags
) {
7108 btrfs_block_group_used(&shrink_block_group
->item
);
7110 calc
= shrink_block_group
->key
.offset
;
7112 spin_unlock(&shrink_block_group
->lock
);
7114 do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
7115 calc
+ 2 * 1024 * 1024, new_alloc_flags
, force
);
7117 btrfs_end_transaction(trans
, root
);
7119 spin_unlock(&shrink_block_group
->lock
);
7124 int btrfs_prepare_block_group_relocation(struct btrfs_root
*root
,
7125 struct btrfs_block_group_cache
*group
)
7128 __alloc_chunk_for_shrink(root
, group
, 1);
7129 set_block_group_readonly(group
);
7134 * checks to see if its even possible to relocate this block group.
7136 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7137 * ok to go ahead and try.
7139 int btrfs_can_relocate(struct btrfs_root
*root
, u64 bytenr
)
7141 struct btrfs_block_group_cache
*block_group
;
7142 struct btrfs_space_info
*space_info
;
7143 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
7144 struct btrfs_device
*device
;
7148 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
7150 /* odd, couldn't find the block group, leave it alone */
7154 /* no bytes used, we're good */
7155 if (!btrfs_block_group_used(&block_group
->item
))
7158 space_info
= block_group
->space_info
;
7159 spin_lock(&space_info
->lock
);
7161 full
= space_info
->full
;
7164 * if this is the last block group we have in this space, we can't
7165 * relocate it unless we're able to allocate a new chunk below.
7167 * Otherwise, we need to make sure we have room in the space to handle
7168 * all of the extents from this block group. If we can, we're good
7170 if ((space_info
->total_bytes
!= block_group
->key
.offset
) &&
7171 (space_info
->bytes_used
+ space_info
->bytes_reserved
+
7172 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
7173 btrfs_block_group_used(&block_group
->item
) <
7174 space_info
->total_bytes
)) {
7175 spin_unlock(&space_info
->lock
);
7178 spin_unlock(&space_info
->lock
);
7181 * ok we don't have enough space, but maybe we have free space on our
7182 * devices to allocate new chunks for relocation, so loop through our
7183 * alloc devices and guess if we have enough space. However, if we
7184 * were marked as full, then we know there aren't enough chunks, and we
7191 mutex_lock(&root
->fs_info
->chunk_mutex
);
7192 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
7193 u64 min_free
= btrfs_block_group_used(&block_group
->item
);
7194 u64 dev_offset
, max_avail
;
7197 * check to make sure we can actually find a chunk with enough
7198 * space to fit our block group in.
7200 if (device
->total_bytes
> device
->bytes_used
+ min_free
) {
7201 ret
= find_free_dev_extent(NULL
, device
, min_free
,
7202 &dev_offset
, &max_avail
);
7208 mutex_unlock(&root
->fs_info
->chunk_mutex
);
7210 btrfs_put_block_group(block_group
);
7214 static int find_first_block_group(struct btrfs_root
*root
,
7215 struct btrfs_path
*path
, struct btrfs_key
*key
)
7218 struct btrfs_key found_key
;
7219 struct extent_buffer
*leaf
;
7222 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
7227 slot
= path
->slots
[0];
7228 leaf
= path
->nodes
[0];
7229 if (slot
>= btrfs_header_nritems(leaf
)) {
7230 ret
= btrfs_next_leaf(root
, path
);
7237 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
7239 if (found_key
.objectid
>= key
->objectid
&&
7240 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
7251 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
7253 struct btrfs_block_group_cache
*block_group
;
7254 struct btrfs_space_info
*space_info
;
7255 struct btrfs_caching_control
*caching_ctl
;
7258 down_write(&info
->extent_commit_sem
);
7259 while (!list_empty(&info
->caching_block_groups
)) {
7260 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
7261 struct btrfs_caching_control
, list
);
7262 list_del(&caching_ctl
->list
);
7263 put_caching_control(caching_ctl
);
7265 up_write(&info
->extent_commit_sem
);
7267 spin_lock(&info
->block_group_cache_lock
);
7268 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
7269 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
7271 rb_erase(&block_group
->cache_node
,
7272 &info
->block_group_cache_tree
);
7273 spin_unlock(&info
->block_group_cache_lock
);
7275 down_write(&block_group
->space_info
->groups_sem
);
7276 list_del(&block_group
->list
);
7277 up_write(&block_group
->space_info
->groups_sem
);
7279 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
7280 wait_block_group_cache_done(block_group
);
7282 btrfs_remove_free_space_cache(block_group
);
7284 WARN_ON(atomic_read(&block_group
->count
) != 1);
7287 spin_lock(&info
->block_group_cache_lock
);
7289 spin_unlock(&info
->block_group_cache_lock
);
7291 /* now that all the block groups are freed, go through and
7292 * free all the space_info structs. This is only called during
7293 * the final stages of unmount, and so we know nobody is
7294 * using them. We call synchronize_rcu() once before we start,
7295 * just to be on the safe side.
7299 while(!list_empty(&info
->space_info
)) {
7300 space_info
= list_entry(info
->space_info
.next
,
7301 struct btrfs_space_info
,
7304 list_del(&space_info
->list
);
7310 int btrfs_read_block_groups(struct btrfs_root
*root
)
7312 struct btrfs_path
*path
;
7314 struct btrfs_block_group_cache
*cache
;
7315 struct btrfs_fs_info
*info
= root
->fs_info
;
7316 struct btrfs_space_info
*space_info
;
7317 struct btrfs_key key
;
7318 struct btrfs_key found_key
;
7319 struct extent_buffer
*leaf
;
7321 root
= info
->extent_root
;
7324 btrfs_set_key_type(&key
, BTRFS_BLOCK_GROUP_ITEM_KEY
);
7325 path
= btrfs_alloc_path();
7330 ret
= find_first_block_group(root
, path
, &key
);
7338 leaf
= path
->nodes
[0];
7339 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
7340 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
7346 atomic_set(&cache
->count
, 1);
7347 spin_lock_init(&cache
->lock
);
7348 spin_lock_init(&cache
->tree_lock
);
7349 cache
->fs_info
= info
;
7350 INIT_LIST_HEAD(&cache
->list
);
7351 INIT_LIST_HEAD(&cache
->cluster_list
);
7354 * we only want to have 32k of ram per block group for keeping
7355 * track of free space, and if we pass 1/2 of that we want to
7356 * start converting things over to using bitmaps
7358 cache
->extents_thresh
= ((1024 * 32) / 2) /
7359 sizeof(struct btrfs_free_space
);
7361 read_extent_buffer(leaf
, &cache
->item
,
7362 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
7363 sizeof(cache
->item
));
7364 memcpy(&cache
->key
, &found_key
, sizeof(found_key
));
7366 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
7367 btrfs_release_path(root
, path
);
7368 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
7369 cache
->sectorsize
= root
->sectorsize
;
7372 * check for two cases, either we are full, and therefore
7373 * don't need to bother with the caching work since we won't
7374 * find any space, or we are empty, and we can just add all
7375 * the space in and be done with it. This saves us _alot_ of
7376 * time, particularly in the full case.
7378 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
7379 exclude_super_stripes(root
, cache
);
7380 cache
->last_byte_to_unpin
= (u64
)-1;
7381 cache
->cached
= BTRFS_CACHE_FINISHED
;
7382 free_excluded_extents(root
, cache
);
7383 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
7384 exclude_super_stripes(root
, cache
);
7385 cache
->last_byte_to_unpin
= (u64
)-1;
7386 cache
->cached
= BTRFS_CACHE_FINISHED
;
7387 add_new_free_space(cache
, root
->fs_info
,
7389 found_key
.objectid
+
7391 free_excluded_extents(root
, cache
);
7394 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
7395 btrfs_block_group_used(&cache
->item
),
7398 cache
->space_info
= space_info
;
7399 spin_lock(&cache
->space_info
->lock
);
7400 cache
->space_info
->bytes_super
+= cache
->bytes_super
;
7401 spin_unlock(&cache
->space_info
->lock
);
7403 down_write(&space_info
->groups_sem
);
7404 list_add_tail(&cache
->list
, &space_info
->block_groups
);
7405 up_write(&space_info
->groups_sem
);
7407 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
7410 set_avail_alloc_bits(root
->fs_info
, cache
->flags
);
7411 if (btrfs_chunk_readonly(root
, cache
->key
.objectid
))
7412 set_block_group_readonly(cache
);
7416 btrfs_free_path(path
);
7420 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
7421 struct btrfs_root
*root
, u64 bytes_used
,
7422 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
7426 struct btrfs_root
*extent_root
;
7427 struct btrfs_block_group_cache
*cache
;
7429 extent_root
= root
->fs_info
->extent_root
;
7431 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
7433 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
7437 cache
->key
.objectid
= chunk_offset
;
7438 cache
->key
.offset
= size
;
7439 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
7440 cache
->sectorsize
= root
->sectorsize
;
7443 * we only want to have 32k of ram per block group for keeping track
7444 * of free space, and if we pass 1/2 of that we want to start
7445 * converting things over to using bitmaps
7447 cache
->extents_thresh
= ((1024 * 32) / 2) /
7448 sizeof(struct btrfs_free_space
);
7449 atomic_set(&cache
->count
, 1);
7450 spin_lock_init(&cache
->lock
);
7451 spin_lock_init(&cache
->tree_lock
);
7452 INIT_LIST_HEAD(&cache
->list
);
7453 INIT_LIST_HEAD(&cache
->cluster_list
);
7455 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
7456 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
7457 cache
->flags
= type
;
7458 btrfs_set_block_group_flags(&cache
->item
, type
);
7460 cache
->last_byte_to_unpin
= (u64
)-1;
7461 cache
->cached
= BTRFS_CACHE_FINISHED
;
7462 exclude_super_stripes(root
, cache
);
7464 add_new_free_space(cache
, root
->fs_info
, chunk_offset
,
7465 chunk_offset
+ size
);
7467 free_excluded_extents(root
, cache
);
7469 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
7470 &cache
->space_info
);
7473 spin_lock(&cache
->space_info
->lock
);
7474 cache
->space_info
->bytes_super
+= cache
->bytes_super
;
7475 spin_unlock(&cache
->space_info
->lock
);
7477 down_write(&cache
->space_info
->groups_sem
);
7478 list_add_tail(&cache
->list
, &cache
->space_info
->block_groups
);
7479 up_write(&cache
->space_info
->groups_sem
);
7481 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
7484 ret
= btrfs_insert_item(trans
, extent_root
, &cache
->key
, &cache
->item
,
7485 sizeof(cache
->item
));
7488 set_avail_alloc_bits(extent_root
->fs_info
, type
);
7493 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
7494 struct btrfs_root
*root
, u64 group_start
)
7496 struct btrfs_path
*path
;
7497 struct btrfs_block_group_cache
*block_group
;
7498 struct btrfs_free_cluster
*cluster
;
7499 struct btrfs_key key
;
7502 root
= root
->fs_info
->extent_root
;
7504 block_group
= btrfs_lookup_block_group(root
->fs_info
, group_start
);
7505 BUG_ON(!block_group
);
7506 BUG_ON(!block_group
->ro
);
7508 memcpy(&key
, &block_group
->key
, sizeof(key
));
7510 /* make sure this block group isn't part of an allocation cluster */
7511 cluster
= &root
->fs_info
->data_alloc_cluster
;
7512 spin_lock(&cluster
->refill_lock
);
7513 btrfs_return_cluster_to_free_space(block_group
, cluster
);
7514 spin_unlock(&cluster
->refill_lock
);
7517 * make sure this block group isn't part of a metadata
7518 * allocation cluster
7520 cluster
= &root
->fs_info
->meta_alloc_cluster
;
7521 spin_lock(&cluster
->refill_lock
);
7522 btrfs_return_cluster_to_free_space(block_group
, cluster
);
7523 spin_unlock(&cluster
->refill_lock
);
7525 path
= btrfs_alloc_path();
7528 spin_lock(&root
->fs_info
->block_group_cache_lock
);
7529 rb_erase(&block_group
->cache_node
,
7530 &root
->fs_info
->block_group_cache_tree
);
7531 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
7533 down_write(&block_group
->space_info
->groups_sem
);
7535 * we must use list_del_init so people can check to see if they
7536 * are still on the list after taking the semaphore
7538 list_del_init(&block_group
->list
);
7539 up_write(&block_group
->space_info
->groups_sem
);
7541 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
7542 wait_block_group_cache_done(block_group
);
7544 btrfs_remove_free_space_cache(block_group
);
7546 spin_lock(&block_group
->space_info
->lock
);
7547 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
7548 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
7549 spin_unlock(&block_group
->space_info
->lock
);
7551 btrfs_clear_space_info_full(root
->fs_info
);
7553 btrfs_put_block_group(block_group
);
7554 btrfs_put_block_group(block_group
);
7556 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
7562 ret
= btrfs_del_item(trans
, root
, path
);
7564 btrfs_free_path(path
);