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>
25 #include <linux/slab.h>
26 #include <linux/ratelimit.h>
27 #include <linux/percpu_counter.h>
31 #include "print-tree.h"
32 #include "transaction.h"
36 #include "free-space-cache.h"
40 #undef SCRAMBLE_DELAYED_REFS
43 * control flags for do_chunk_alloc's force field
44 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
45 * if we really need one.
47 * CHUNK_ALLOC_LIMITED means to only try and allocate one
48 * if we have very few chunks already allocated. This is
49 * used as part of the clustering code to help make sure
50 * we have a good pool of storage to cluster in, without
51 * filling the FS with empty chunks
53 * CHUNK_ALLOC_FORCE means it must try to allocate one
57 CHUNK_ALLOC_NO_FORCE
= 0,
58 CHUNK_ALLOC_LIMITED
= 1,
59 CHUNK_ALLOC_FORCE
= 2,
63 * Control how reservations are dealt with.
65 * RESERVE_FREE - freeing a reservation.
66 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
68 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
69 * bytes_may_use as the ENOSPC accounting is done elsewhere
74 RESERVE_ALLOC_NO_ACCOUNT
= 2,
77 static int update_block_group(struct btrfs_root
*root
,
78 u64 bytenr
, u64 num_bytes
, int alloc
);
79 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
80 struct btrfs_root
*root
,
81 u64 bytenr
, u64 num_bytes
, u64 parent
,
82 u64 root_objectid
, u64 owner_objectid
,
83 u64 owner_offset
, int refs_to_drop
,
84 struct btrfs_delayed_extent_op
*extra_op
);
85 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
86 struct extent_buffer
*leaf
,
87 struct btrfs_extent_item
*ei
);
88 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
89 struct btrfs_root
*root
,
90 u64 parent
, u64 root_objectid
,
91 u64 flags
, u64 owner
, u64 offset
,
92 struct btrfs_key
*ins
, int ref_mod
);
93 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
94 struct btrfs_root
*root
,
95 u64 parent
, u64 root_objectid
,
96 u64 flags
, struct btrfs_disk_key
*key
,
97 int level
, struct btrfs_key
*ins
);
98 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
99 struct btrfs_root
*extent_root
, u64 flags
,
101 static int find_next_key(struct btrfs_path
*path
, int level
,
102 struct btrfs_key
*key
);
103 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
104 int dump_block_groups
);
105 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
106 u64 num_bytes
, int reserve
);
107 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
109 int btrfs_pin_extent(struct btrfs_root
*root
,
110 u64 bytenr
, u64 num_bytes
, int reserved
);
113 block_group_cache_done(struct btrfs_block_group_cache
*cache
)
116 return cache
->cached
== BTRFS_CACHE_FINISHED
||
117 cache
->cached
== BTRFS_CACHE_ERROR
;
120 static int block_group_bits(struct btrfs_block_group_cache
*cache
, u64 bits
)
122 return (cache
->flags
& bits
) == bits
;
125 static void btrfs_get_block_group(struct btrfs_block_group_cache
*cache
)
127 atomic_inc(&cache
->count
);
130 void btrfs_put_block_group(struct btrfs_block_group_cache
*cache
)
132 if (atomic_dec_and_test(&cache
->count
)) {
133 WARN_ON(cache
->pinned
> 0);
134 WARN_ON(cache
->reserved
> 0);
135 kfree(cache
->free_space_ctl
);
141 * this adds the block group to the fs_info rb tree for the block group
144 static int btrfs_add_block_group_cache(struct btrfs_fs_info
*info
,
145 struct btrfs_block_group_cache
*block_group
)
148 struct rb_node
*parent
= NULL
;
149 struct btrfs_block_group_cache
*cache
;
151 spin_lock(&info
->block_group_cache_lock
);
152 p
= &info
->block_group_cache_tree
.rb_node
;
156 cache
= rb_entry(parent
, struct btrfs_block_group_cache
,
158 if (block_group
->key
.objectid
< cache
->key
.objectid
) {
160 } else if (block_group
->key
.objectid
> cache
->key
.objectid
) {
163 spin_unlock(&info
->block_group_cache_lock
);
168 rb_link_node(&block_group
->cache_node
, parent
, p
);
169 rb_insert_color(&block_group
->cache_node
,
170 &info
->block_group_cache_tree
);
172 if (info
->first_logical_byte
> block_group
->key
.objectid
)
173 info
->first_logical_byte
= block_group
->key
.objectid
;
175 spin_unlock(&info
->block_group_cache_lock
);
181 * This will return the block group at or after bytenr if contains is 0, else
182 * it will return the block group that contains the bytenr
184 static struct btrfs_block_group_cache
*
185 block_group_cache_tree_search(struct btrfs_fs_info
*info
, u64 bytenr
,
188 struct btrfs_block_group_cache
*cache
, *ret
= NULL
;
192 spin_lock(&info
->block_group_cache_lock
);
193 n
= info
->block_group_cache_tree
.rb_node
;
196 cache
= rb_entry(n
, struct btrfs_block_group_cache
,
198 end
= cache
->key
.objectid
+ cache
->key
.offset
- 1;
199 start
= cache
->key
.objectid
;
201 if (bytenr
< start
) {
202 if (!contains
&& (!ret
|| start
< ret
->key
.objectid
))
205 } else if (bytenr
> start
) {
206 if (contains
&& bytenr
<= end
) {
217 btrfs_get_block_group(ret
);
218 if (bytenr
== 0 && info
->first_logical_byte
> ret
->key
.objectid
)
219 info
->first_logical_byte
= ret
->key
.objectid
;
221 spin_unlock(&info
->block_group_cache_lock
);
226 static int add_excluded_extent(struct btrfs_root
*root
,
227 u64 start
, u64 num_bytes
)
229 u64 end
= start
+ num_bytes
- 1;
230 set_extent_bits(&root
->fs_info
->freed_extents
[0],
231 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
232 set_extent_bits(&root
->fs_info
->freed_extents
[1],
233 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
237 static void free_excluded_extents(struct btrfs_root
*root
,
238 struct btrfs_block_group_cache
*cache
)
242 start
= cache
->key
.objectid
;
243 end
= start
+ cache
->key
.offset
- 1;
245 clear_extent_bits(&root
->fs_info
->freed_extents
[0],
246 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
247 clear_extent_bits(&root
->fs_info
->freed_extents
[1],
248 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
251 static int exclude_super_stripes(struct btrfs_root
*root
,
252 struct btrfs_block_group_cache
*cache
)
259 if (cache
->key
.objectid
< BTRFS_SUPER_INFO_OFFSET
) {
260 stripe_len
= BTRFS_SUPER_INFO_OFFSET
- cache
->key
.objectid
;
261 cache
->bytes_super
+= stripe_len
;
262 ret
= add_excluded_extent(root
, cache
->key
.objectid
,
268 for (i
= 0; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
269 bytenr
= btrfs_sb_offset(i
);
270 ret
= btrfs_rmap_block(&root
->fs_info
->mapping_tree
,
271 cache
->key
.objectid
, bytenr
,
272 0, &logical
, &nr
, &stripe_len
);
279 if (logical
[nr
] > cache
->key
.objectid
+
283 if (logical
[nr
] + stripe_len
<= cache
->key
.objectid
)
287 if (start
< cache
->key
.objectid
) {
288 start
= cache
->key
.objectid
;
289 len
= (logical
[nr
] + stripe_len
) - start
;
291 len
= min_t(u64
, stripe_len
,
292 cache
->key
.objectid
+
293 cache
->key
.offset
- start
);
296 cache
->bytes_super
+= len
;
297 ret
= add_excluded_extent(root
, start
, len
);
309 static struct btrfs_caching_control
*
310 get_caching_control(struct btrfs_block_group_cache
*cache
)
312 struct btrfs_caching_control
*ctl
;
314 spin_lock(&cache
->lock
);
315 if (cache
->cached
!= BTRFS_CACHE_STARTED
) {
316 spin_unlock(&cache
->lock
);
320 /* We're loading it the fast way, so we don't have a caching_ctl. */
321 if (!cache
->caching_ctl
) {
322 spin_unlock(&cache
->lock
);
326 ctl
= cache
->caching_ctl
;
327 atomic_inc(&ctl
->count
);
328 spin_unlock(&cache
->lock
);
332 static void put_caching_control(struct btrfs_caching_control
*ctl
)
334 if (atomic_dec_and_test(&ctl
->count
))
339 * this is only called by cache_block_group, since we could have freed extents
340 * we need to check the pinned_extents for any extents that can't be used yet
341 * since their free space will be released as soon as the transaction commits.
343 static u64
add_new_free_space(struct btrfs_block_group_cache
*block_group
,
344 struct btrfs_fs_info
*info
, u64 start
, u64 end
)
346 u64 extent_start
, extent_end
, size
, total_added
= 0;
349 while (start
< end
) {
350 ret
= find_first_extent_bit(info
->pinned_extents
, start
,
351 &extent_start
, &extent_end
,
352 EXTENT_DIRTY
| EXTENT_UPTODATE
,
357 if (extent_start
<= start
) {
358 start
= extent_end
+ 1;
359 } else if (extent_start
> start
&& extent_start
< end
) {
360 size
= extent_start
- start
;
362 ret
= btrfs_add_free_space(block_group
, start
,
364 BUG_ON(ret
); /* -ENOMEM or logic error */
365 start
= extent_end
+ 1;
374 ret
= btrfs_add_free_space(block_group
, start
, size
);
375 BUG_ON(ret
); /* -ENOMEM or logic error */
381 static noinline
void caching_thread(struct btrfs_work
*work
)
383 struct btrfs_block_group_cache
*block_group
;
384 struct btrfs_fs_info
*fs_info
;
385 struct btrfs_caching_control
*caching_ctl
;
386 struct btrfs_root
*extent_root
;
387 struct btrfs_path
*path
;
388 struct extent_buffer
*leaf
;
389 struct btrfs_key key
;
395 caching_ctl
= container_of(work
, struct btrfs_caching_control
, work
);
396 block_group
= caching_ctl
->block_group
;
397 fs_info
= block_group
->fs_info
;
398 extent_root
= fs_info
->extent_root
;
400 path
= btrfs_alloc_path();
404 last
= max_t(u64
, block_group
->key
.objectid
, BTRFS_SUPER_INFO_OFFSET
);
407 * We don't want to deadlock with somebody trying to allocate a new
408 * extent for the extent root while also trying to search the extent
409 * root to add free space. So we skip locking and search the commit
410 * root, since its read-only
412 path
->skip_locking
= 1;
413 path
->search_commit_root
= 1;
418 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
420 mutex_lock(&caching_ctl
->mutex
);
421 /* need to make sure the commit_root doesn't disappear */
422 down_read(&fs_info
->commit_root_sem
);
425 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
429 leaf
= path
->nodes
[0];
430 nritems
= btrfs_header_nritems(leaf
);
433 if (btrfs_fs_closing(fs_info
) > 1) {
438 if (path
->slots
[0] < nritems
) {
439 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
441 ret
= find_next_key(path
, 0, &key
);
445 if (need_resched() ||
446 rwsem_is_contended(&fs_info
->commit_root_sem
)) {
447 caching_ctl
->progress
= last
;
448 btrfs_release_path(path
);
449 up_read(&fs_info
->commit_root_sem
);
450 mutex_unlock(&caching_ctl
->mutex
);
455 ret
= btrfs_next_leaf(extent_root
, path
);
460 leaf
= path
->nodes
[0];
461 nritems
= btrfs_header_nritems(leaf
);
465 if (key
.objectid
< last
) {
468 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
470 caching_ctl
->progress
= last
;
471 btrfs_release_path(path
);
475 if (key
.objectid
< block_group
->key
.objectid
) {
480 if (key
.objectid
>= block_group
->key
.objectid
+
481 block_group
->key
.offset
)
484 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
||
485 key
.type
== BTRFS_METADATA_ITEM_KEY
) {
486 total_found
+= add_new_free_space(block_group
,
489 if (key
.type
== BTRFS_METADATA_ITEM_KEY
)
490 last
= key
.objectid
+
491 fs_info
->tree_root
->leafsize
;
493 last
= key
.objectid
+ key
.offset
;
495 if (total_found
> (1024 * 1024 * 2)) {
497 wake_up(&caching_ctl
->wait
);
504 total_found
+= add_new_free_space(block_group
, fs_info
, last
,
505 block_group
->key
.objectid
+
506 block_group
->key
.offset
);
507 caching_ctl
->progress
= (u64
)-1;
509 spin_lock(&block_group
->lock
);
510 block_group
->caching_ctl
= NULL
;
511 block_group
->cached
= BTRFS_CACHE_FINISHED
;
512 spin_unlock(&block_group
->lock
);
515 btrfs_free_path(path
);
516 up_read(&fs_info
->commit_root_sem
);
518 free_excluded_extents(extent_root
, block_group
);
520 mutex_unlock(&caching_ctl
->mutex
);
523 spin_lock(&block_group
->lock
);
524 block_group
->caching_ctl
= NULL
;
525 block_group
->cached
= BTRFS_CACHE_ERROR
;
526 spin_unlock(&block_group
->lock
);
528 wake_up(&caching_ctl
->wait
);
530 put_caching_control(caching_ctl
);
531 btrfs_put_block_group(block_group
);
534 static int cache_block_group(struct btrfs_block_group_cache
*cache
,
538 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
539 struct btrfs_caching_control
*caching_ctl
;
542 caching_ctl
= kzalloc(sizeof(*caching_ctl
), GFP_NOFS
);
546 INIT_LIST_HEAD(&caching_ctl
->list
);
547 mutex_init(&caching_ctl
->mutex
);
548 init_waitqueue_head(&caching_ctl
->wait
);
549 caching_ctl
->block_group
= cache
;
550 caching_ctl
->progress
= cache
->key
.objectid
;
551 atomic_set(&caching_ctl
->count
, 1);
552 btrfs_init_work(&caching_ctl
->work
, caching_thread
, NULL
, NULL
);
554 spin_lock(&cache
->lock
);
556 * This should be a rare occasion, but this could happen I think in the
557 * case where one thread starts to load the space cache info, and then
558 * some other thread starts a transaction commit which tries to do an
559 * allocation while the other thread is still loading the space cache
560 * info. The previous loop should have kept us from choosing this block
561 * group, but if we've moved to the state where we will wait on caching
562 * block groups we need to first check if we're doing a fast load here,
563 * so we can wait for it to finish, otherwise we could end up allocating
564 * from a block group who's cache gets evicted for one reason or
567 while (cache
->cached
== BTRFS_CACHE_FAST
) {
568 struct btrfs_caching_control
*ctl
;
570 ctl
= cache
->caching_ctl
;
571 atomic_inc(&ctl
->count
);
572 prepare_to_wait(&ctl
->wait
, &wait
, TASK_UNINTERRUPTIBLE
);
573 spin_unlock(&cache
->lock
);
577 finish_wait(&ctl
->wait
, &wait
);
578 put_caching_control(ctl
);
579 spin_lock(&cache
->lock
);
582 if (cache
->cached
!= BTRFS_CACHE_NO
) {
583 spin_unlock(&cache
->lock
);
587 WARN_ON(cache
->caching_ctl
);
588 cache
->caching_ctl
= caching_ctl
;
589 cache
->cached
= BTRFS_CACHE_FAST
;
590 spin_unlock(&cache
->lock
);
592 if (fs_info
->mount_opt
& BTRFS_MOUNT_SPACE_CACHE
) {
593 ret
= load_free_space_cache(fs_info
, cache
);
595 spin_lock(&cache
->lock
);
597 cache
->caching_ctl
= NULL
;
598 cache
->cached
= BTRFS_CACHE_FINISHED
;
599 cache
->last_byte_to_unpin
= (u64
)-1;
601 if (load_cache_only
) {
602 cache
->caching_ctl
= NULL
;
603 cache
->cached
= BTRFS_CACHE_NO
;
605 cache
->cached
= BTRFS_CACHE_STARTED
;
608 spin_unlock(&cache
->lock
);
609 wake_up(&caching_ctl
->wait
);
611 put_caching_control(caching_ctl
);
612 free_excluded_extents(fs_info
->extent_root
, cache
);
617 * We are not going to do the fast caching, set cached to the
618 * appropriate value and wakeup any waiters.
620 spin_lock(&cache
->lock
);
621 if (load_cache_only
) {
622 cache
->caching_ctl
= NULL
;
623 cache
->cached
= BTRFS_CACHE_NO
;
625 cache
->cached
= BTRFS_CACHE_STARTED
;
627 spin_unlock(&cache
->lock
);
628 wake_up(&caching_ctl
->wait
);
631 if (load_cache_only
) {
632 put_caching_control(caching_ctl
);
636 down_write(&fs_info
->commit_root_sem
);
637 atomic_inc(&caching_ctl
->count
);
638 list_add_tail(&caching_ctl
->list
, &fs_info
->caching_block_groups
);
639 up_write(&fs_info
->commit_root_sem
);
641 btrfs_get_block_group(cache
);
643 btrfs_queue_work(fs_info
->caching_workers
, &caching_ctl
->work
);
649 * return the block group that starts at or after bytenr
651 static struct btrfs_block_group_cache
*
652 btrfs_lookup_first_block_group(struct btrfs_fs_info
*info
, u64 bytenr
)
654 struct btrfs_block_group_cache
*cache
;
656 cache
= block_group_cache_tree_search(info
, bytenr
, 0);
662 * return the block group that contains the given bytenr
664 struct btrfs_block_group_cache
*btrfs_lookup_block_group(
665 struct btrfs_fs_info
*info
,
668 struct btrfs_block_group_cache
*cache
;
670 cache
= block_group_cache_tree_search(info
, bytenr
, 1);
675 static struct btrfs_space_info
*__find_space_info(struct btrfs_fs_info
*info
,
678 struct list_head
*head
= &info
->space_info
;
679 struct btrfs_space_info
*found
;
681 flags
&= BTRFS_BLOCK_GROUP_TYPE_MASK
;
684 list_for_each_entry_rcu(found
, head
, list
) {
685 if (found
->flags
& flags
) {
695 * after adding space to the filesystem, we need to clear the full flags
696 * on all the space infos.
698 void btrfs_clear_space_info_full(struct btrfs_fs_info
*info
)
700 struct list_head
*head
= &info
->space_info
;
701 struct btrfs_space_info
*found
;
704 list_for_each_entry_rcu(found
, head
, list
)
709 /* simple helper to search for an existing extent at a given offset */
710 int btrfs_lookup_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
713 struct btrfs_key key
;
714 struct btrfs_path
*path
;
716 path
= btrfs_alloc_path();
720 key
.objectid
= start
;
722 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
723 ret
= btrfs_search_slot(NULL
, root
->fs_info
->extent_root
, &key
, path
,
726 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
727 if (key
.objectid
== start
&&
728 key
.type
== BTRFS_METADATA_ITEM_KEY
)
731 btrfs_free_path(path
);
736 * helper function to lookup reference count and flags of a tree block.
738 * the head node for delayed ref is used to store the sum of all the
739 * reference count modifications queued up in the rbtree. the head
740 * node may also store the extent flags to set. This way you can check
741 * to see what the reference count and extent flags would be if all of
742 * the delayed refs are not processed.
744 int btrfs_lookup_extent_info(struct btrfs_trans_handle
*trans
,
745 struct btrfs_root
*root
, u64 bytenr
,
746 u64 offset
, int metadata
, u64
*refs
, u64
*flags
)
748 struct btrfs_delayed_ref_head
*head
;
749 struct btrfs_delayed_ref_root
*delayed_refs
;
750 struct btrfs_path
*path
;
751 struct btrfs_extent_item
*ei
;
752 struct extent_buffer
*leaf
;
753 struct btrfs_key key
;
760 * If we don't have skinny metadata, don't bother doing anything
763 if (metadata
&& !btrfs_fs_incompat(root
->fs_info
, SKINNY_METADATA
)) {
764 offset
= root
->leafsize
;
768 path
= btrfs_alloc_path();
773 path
->skip_locking
= 1;
774 path
->search_commit_root
= 1;
778 key
.objectid
= bytenr
;
781 key
.type
= BTRFS_METADATA_ITEM_KEY
;
783 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
786 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
,
791 if (ret
> 0 && metadata
&& key
.type
== BTRFS_METADATA_ITEM_KEY
) {
792 if (path
->slots
[0]) {
794 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
796 if (key
.objectid
== bytenr
&&
797 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
798 key
.offset
== root
->leafsize
)
802 key
.objectid
= bytenr
;
803 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
804 key
.offset
= root
->leafsize
;
805 btrfs_release_path(path
);
811 leaf
= path
->nodes
[0];
812 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
813 if (item_size
>= sizeof(*ei
)) {
814 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
815 struct btrfs_extent_item
);
816 num_refs
= btrfs_extent_refs(leaf
, ei
);
817 extent_flags
= btrfs_extent_flags(leaf
, ei
);
819 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
820 struct btrfs_extent_item_v0
*ei0
;
821 BUG_ON(item_size
!= sizeof(*ei0
));
822 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
823 struct btrfs_extent_item_v0
);
824 num_refs
= btrfs_extent_refs_v0(leaf
, ei0
);
825 /* FIXME: this isn't correct for data */
826 extent_flags
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
831 BUG_ON(num_refs
== 0);
841 delayed_refs
= &trans
->transaction
->delayed_refs
;
842 spin_lock(&delayed_refs
->lock
);
843 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
845 if (!mutex_trylock(&head
->mutex
)) {
846 atomic_inc(&head
->node
.refs
);
847 spin_unlock(&delayed_refs
->lock
);
849 btrfs_release_path(path
);
852 * Mutex was contended, block until it's released and try
855 mutex_lock(&head
->mutex
);
856 mutex_unlock(&head
->mutex
);
857 btrfs_put_delayed_ref(&head
->node
);
860 spin_lock(&head
->lock
);
861 if (head
->extent_op
&& head
->extent_op
->update_flags
)
862 extent_flags
|= head
->extent_op
->flags_to_set
;
864 BUG_ON(num_refs
== 0);
866 num_refs
+= head
->node
.ref_mod
;
867 spin_unlock(&head
->lock
);
868 mutex_unlock(&head
->mutex
);
870 spin_unlock(&delayed_refs
->lock
);
872 WARN_ON(num_refs
== 0);
876 *flags
= extent_flags
;
878 btrfs_free_path(path
);
883 * Back reference rules. Back refs have three main goals:
885 * 1) differentiate between all holders of references to an extent so that
886 * when a reference is dropped we can make sure it was a valid reference
887 * before freeing the extent.
889 * 2) Provide enough information to quickly find the holders of an extent
890 * if we notice a given block is corrupted or bad.
892 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
893 * maintenance. This is actually the same as #2, but with a slightly
894 * different use case.
896 * There are two kinds of back refs. The implicit back refs is optimized
897 * for pointers in non-shared tree blocks. For a given pointer in a block,
898 * back refs of this kind provide information about the block's owner tree
899 * and the pointer's key. These information allow us to find the block by
900 * b-tree searching. The full back refs is for pointers in tree blocks not
901 * referenced by their owner trees. The location of tree block is recorded
902 * in the back refs. Actually the full back refs is generic, and can be
903 * used in all cases the implicit back refs is used. The major shortcoming
904 * of the full back refs is its overhead. Every time a tree block gets
905 * COWed, we have to update back refs entry for all pointers in it.
907 * For a newly allocated tree block, we use implicit back refs for
908 * pointers in it. This means most tree related operations only involve
909 * implicit back refs. For a tree block created in old transaction, the
910 * only way to drop a reference to it is COW it. So we can detect the
911 * event that tree block loses its owner tree's reference and do the
912 * back refs conversion.
914 * When a tree block is COW'd through a tree, there are four cases:
916 * The reference count of the block is one and the tree is the block's
917 * owner tree. Nothing to do in this case.
919 * The reference count of the block is one and the tree is not the
920 * block's owner tree. In this case, full back refs is used for pointers
921 * in the block. Remove these full back refs, add implicit back refs for
922 * every pointers in the new block.
924 * The reference count of the block is greater than one and the tree is
925 * the block's owner tree. In this case, implicit back refs is used for
926 * pointers in the block. Add full back refs for every pointers in the
927 * block, increase lower level extents' reference counts. The original
928 * implicit back refs are entailed to the new block.
930 * The reference count of the block is greater than one and the tree is
931 * not the block's owner tree. Add implicit back refs for every pointer in
932 * the new block, increase lower level extents' reference count.
934 * Back Reference Key composing:
936 * The key objectid corresponds to the first byte in the extent,
937 * The key type is used to differentiate between types of back refs.
938 * There are different meanings of the key offset for different types
941 * File extents can be referenced by:
943 * - multiple snapshots, subvolumes, or different generations in one subvol
944 * - different files inside a single subvolume
945 * - different offsets inside a file (bookend extents in file.c)
947 * The extent ref structure for the implicit back refs has fields for:
949 * - Objectid of the subvolume root
950 * - objectid of the file holding the reference
951 * - original offset in the file
952 * - how many bookend extents
954 * The key offset for the implicit back refs is hash of the first
957 * The extent ref structure for the full back refs has field for:
959 * - number of pointers in the tree leaf
961 * The key offset for the implicit back refs is the first byte of
964 * When a file extent is allocated, The implicit back refs is used.
965 * the fields are filled in:
967 * (root_key.objectid, inode objectid, offset in file, 1)
969 * When a file extent is removed file truncation, we find the
970 * corresponding implicit back refs and check the following fields:
972 * (btrfs_header_owner(leaf), inode objectid, offset in file)
974 * Btree extents can be referenced by:
976 * - Different subvolumes
978 * Both the implicit back refs and the full back refs for tree blocks
979 * only consist of key. The key offset for the implicit back refs is
980 * objectid of block's owner tree. The key offset for the full back refs
981 * is the first byte of parent block.
983 * When implicit back refs is used, information about the lowest key and
984 * level of the tree block are required. These information are stored in
985 * tree block info structure.
988 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
989 static int convert_extent_item_v0(struct btrfs_trans_handle
*trans
,
990 struct btrfs_root
*root
,
991 struct btrfs_path
*path
,
992 u64 owner
, u32 extra_size
)
994 struct btrfs_extent_item
*item
;
995 struct btrfs_extent_item_v0
*ei0
;
996 struct btrfs_extent_ref_v0
*ref0
;
997 struct btrfs_tree_block_info
*bi
;
998 struct extent_buffer
*leaf
;
999 struct btrfs_key key
;
1000 struct btrfs_key found_key
;
1001 u32 new_size
= sizeof(*item
);
1005 leaf
= path
->nodes
[0];
1006 BUG_ON(btrfs_item_size_nr(leaf
, path
->slots
[0]) != sizeof(*ei0
));
1008 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1009 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1010 struct btrfs_extent_item_v0
);
1011 refs
= btrfs_extent_refs_v0(leaf
, ei0
);
1013 if (owner
== (u64
)-1) {
1015 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1016 ret
= btrfs_next_leaf(root
, path
);
1019 BUG_ON(ret
> 0); /* Corruption */
1020 leaf
= path
->nodes
[0];
1022 btrfs_item_key_to_cpu(leaf
, &found_key
,
1024 BUG_ON(key
.objectid
!= found_key
.objectid
);
1025 if (found_key
.type
!= BTRFS_EXTENT_REF_V0_KEY
) {
1029 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1030 struct btrfs_extent_ref_v0
);
1031 owner
= btrfs_ref_objectid_v0(leaf
, ref0
);
1035 btrfs_release_path(path
);
1037 if (owner
< BTRFS_FIRST_FREE_OBJECTID
)
1038 new_size
+= sizeof(*bi
);
1040 new_size
-= sizeof(*ei0
);
1041 ret
= btrfs_search_slot(trans
, root
, &key
, path
,
1042 new_size
+ extra_size
, 1);
1045 BUG_ON(ret
); /* Corruption */
1047 btrfs_extend_item(root
, path
, new_size
);
1049 leaf
= path
->nodes
[0];
1050 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1051 btrfs_set_extent_refs(leaf
, item
, refs
);
1052 /* FIXME: get real generation */
1053 btrfs_set_extent_generation(leaf
, item
, 0);
1054 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1055 btrfs_set_extent_flags(leaf
, item
,
1056 BTRFS_EXTENT_FLAG_TREE_BLOCK
|
1057 BTRFS_BLOCK_FLAG_FULL_BACKREF
);
1058 bi
= (struct btrfs_tree_block_info
*)(item
+ 1);
1059 /* FIXME: get first key of the block */
1060 memset_extent_buffer(leaf
, 0, (unsigned long)bi
, sizeof(*bi
));
1061 btrfs_set_tree_block_level(leaf
, bi
, (int)owner
);
1063 btrfs_set_extent_flags(leaf
, item
, BTRFS_EXTENT_FLAG_DATA
);
1065 btrfs_mark_buffer_dirty(leaf
);
1070 static u64
hash_extent_data_ref(u64 root_objectid
, u64 owner
, u64 offset
)
1072 u32 high_crc
= ~(u32
)0;
1073 u32 low_crc
= ~(u32
)0;
1076 lenum
= cpu_to_le64(root_objectid
);
1077 high_crc
= btrfs_crc32c(high_crc
, &lenum
, sizeof(lenum
));
1078 lenum
= cpu_to_le64(owner
);
1079 low_crc
= btrfs_crc32c(low_crc
, &lenum
, sizeof(lenum
));
1080 lenum
= cpu_to_le64(offset
);
1081 low_crc
= btrfs_crc32c(low_crc
, &lenum
, sizeof(lenum
));
1083 return ((u64
)high_crc
<< 31) ^ (u64
)low_crc
;
1086 static u64
hash_extent_data_ref_item(struct extent_buffer
*leaf
,
1087 struct btrfs_extent_data_ref
*ref
)
1089 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf
, ref
),
1090 btrfs_extent_data_ref_objectid(leaf
, ref
),
1091 btrfs_extent_data_ref_offset(leaf
, ref
));
1094 static int match_extent_data_ref(struct extent_buffer
*leaf
,
1095 struct btrfs_extent_data_ref
*ref
,
1096 u64 root_objectid
, u64 owner
, u64 offset
)
1098 if (btrfs_extent_data_ref_root(leaf
, ref
) != root_objectid
||
1099 btrfs_extent_data_ref_objectid(leaf
, ref
) != owner
||
1100 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
1105 static noinline
int lookup_extent_data_ref(struct btrfs_trans_handle
*trans
,
1106 struct btrfs_root
*root
,
1107 struct btrfs_path
*path
,
1108 u64 bytenr
, u64 parent
,
1110 u64 owner
, u64 offset
)
1112 struct btrfs_key key
;
1113 struct btrfs_extent_data_ref
*ref
;
1114 struct extent_buffer
*leaf
;
1120 key
.objectid
= bytenr
;
1122 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1123 key
.offset
= parent
;
1125 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1126 key
.offset
= hash_extent_data_ref(root_objectid
,
1131 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1140 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1141 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1142 btrfs_release_path(path
);
1143 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1154 leaf
= path
->nodes
[0];
1155 nritems
= btrfs_header_nritems(leaf
);
1157 if (path
->slots
[0] >= nritems
) {
1158 ret
= btrfs_next_leaf(root
, path
);
1164 leaf
= path
->nodes
[0];
1165 nritems
= btrfs_header_nritems(leaf
);
1169 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1170 if (key
.objectid
!= bytenr
||
1171 key
.type
!= BTRFS_EXTENT_DATA_REF_KEY
)
1174 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1175 struct btrfs_extent_data_ref
);
1177 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1180 btrfs_release_path(path
);
1192 static noinline
int insert_extent_data_ref(struct btrfs_trans_handle
*trans
,
1193 struct btrfs_root
*root
,
1194 struct btrfs_path
*path
,
1195 u64 bytenr
, u64 parent
,
1196 u64 root_objectid
, u64 owner
,
1197 u64 offset
, int refs_to_add
)
1199 struct btrfs_key key
;
1200 struct extent_buffer
*leaf
;
1205 key
.objectid
= bytenr
;
1207 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1208 key
.offset
= parent
;
1209 size
= sizeof(struct btrfs_shared_data_ref
);
1211 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1212 key
.offset
= hash_extent_data_ref(root_objectid
,
1214 size
= sizeof(struct btrfs_extent_data_ref
);
1217 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, size
);
1218 if (ret
&& ret
!= -EEXIST
)
1221 leaf
= path
->nodes
[0];
1223 struct btrfs_shared_data_ref
*ref
;
1224 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1225 struct btrfs_shared_data_ref
);
1227 btrfs_set_shared_data_ref_count(leaf
, ref
, refs_to_add
);
1229 num_refs
= btrfs_shared_data_ref_count(leaf
, ref
);
1230 num_refs
+= refs_to_add
;
1231 btrfs_set_shared_data_ref_count(leaf
, ref
, num_refs
);
1234 struct btrfs_extent_data_ref
*ref
;
1235 while (ret
== -EEXIST
) {
1236 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1237 struct btrfs_extent_data_ref
);
1238 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1241 btrfs_release_path(path
);
1243 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1245 if (ret
&& ret
!= -EEXIST
)
1248 leaf
= path
->nodes
[0];
1250 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1251 struct btrfs_extent_data_ref
);
1253 btrfs_set_extent_data_ref_root(leaf
, ref
,
1255 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
1256 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
1257 btrfs_set_extent_data_ref_count(leaf
, ref
, refs_to_add
);
1259 num_refs
= btrfs_extent_data_ref_count(leaf
, ref
);
1260 num_refs
+= refs_to_add
;
1261 btrfs_set_extent_data_ref_count(leaf
, ref
, num_refs
);
1264 btrfs_mark_buffer_dirty(leaf
);
1267 btrfs_release_path(path
);
1271 static noinline
int remove_extent_data_ref(struct btrfs_trans_handle
*trans
,
1272 struct btrfs_root
*root
,
1273 struct btrfs_path
*path
,
1276 struct btrfs_key key
;
1277 struct btrfs_extent_data_ref
*ref1
= NULL
;
1278 struct btrfs_shared_data_ref
*ref2
= NULL
;
1279 struct extent_buffer
*leaf
;
1283 leaf
= path
->nodes
[0];
1284 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1286 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1287 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1288 struct btrfs_extent_data_ref
);
1289 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1290 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1291 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1292 struct btrfs_shared_data_ref
);
1293 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1294 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1295 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1296 struct btrfs_extent_ref_v0
*ref0
;
1297 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1298 struct btrfs_extent_ref_v0
);
1299 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1305 BUG_ON(num_refs
< refs_to_drop
);
1306 num_refs
-= refs_to_drop
;
1308 if (num_refs
== 0) {
1309 ret
= btrfs_del_item(trans
, root
, path
);
1311 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
)
1312 btrfs_set_extent_data_ref_count(leaf
, ref1
, num_refs
);
1313 else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
)
1314 btrfs_set_shared_data_ref_count(leaf
, ref2
, num_refs
);
1315 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1317 struct btrfs_extent_ref_v0
*ref0
;
1318 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1319 struct btrfs_extent_ref_v0
);
1320 btrfs_set_ref_count_v0(leaf
, ref0
, num_refs
);
1323 btrfs_mark_buffer_dirty(leaf
);
1328 static noinline u32
extent_data_ref_count(struct btrfs_root
*root
,
1329 struct btrfs_path
*path
,
1330 struct btrfs_extent_inline_ref
*iref
)
1332 struct btrfs_key key
;
1333 struct extent_buffer
*leaf
;
1334 struct btrfs_extent_data_ref
*ref1
;
1335 struct btrfs_shared_data_ref
*ref2
;
1338 leaf
= path
->nodes
[0];
1339 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1341 if (btrfs_extent_inline_ref_type(leaf
, iref
) ==
1342 BTRFS_EXTENT_DATA_REF_KEY
) {
1343 ref1
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1344 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1346 ref2
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1347 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1349 } else if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1350 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1351 struct btrfs_extent_data_ref
);
1352 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1353 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1354 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1355 struct btrfs_shared_data_ref
);
1356 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1357 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1358 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1359 struct btrfs_extent_ref_v0
*ref0
;
1360 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1361 struct btrfs_extent_ref_v0
);
1362 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1370 static noinline
int lookup_tree_block_ref(struct btrfs_trans_handle
*trans
,
1371 struct btrfs_root
*root
,
1372 struct btrfs_path
*path
,
1373 u64 bytenr
, u64 parent
,
1376 struct btrfs_key key
;
1379 key
.objectid
= bytenr
;
1381 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1382 key
.offset
= parent
;
1384 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1385 key
.offset
= root_objectid
;
1388 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1391 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1392 if (ret
== -ENOENT
&& parent
) {
1393 btrfs_release_path(path
);
1394 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1395 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1403 static noinline
int insert_tree_block_ref(struct btrfs_trans_handle
*trans
,
1404 struct btrfs_root
*root
,
1405 struct btrfs_path
*path
,
1406 u64 bytenr
, u64 parent
,
1409 struct btrfs_key key
;
1412 key
.objectid
= bytenr
;
1414 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1415 key
.offset
= parent
;
1417 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1418 key
.offset
= root_objectid
;
1421 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
1422 btrfs_release_path(path
);
1426 static inline int extent_ref_type(u64 parent
, u64 owner
)
1429 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1431 type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1433 type
= BTRFS_TREE_BLOCK_REF_KEY
;
1436 type
= BTRFS_SHARED_DATA_REF_KEY
;
1438 type
= BTRFS_EXTENT_DATA_REF_KEY
;
1443 static int find_next_key(struct btrfs_path
*path
, int level
,
1444 struct btrfs_key
*key
)
1447 for (; level
< BTRFS_MAX_LEVEL
; level
++) {
1448 if (!path
->nodes
[level
])
1450 if (path
->slots
[level
] + 1 >=
1451 btrfs_header_nritems(path
->nodes
[level
]))
1454 btrfs_item_key_to_cpu(path
->nodes
[level
], key
,
1455 path
->slots
[level
] + 1);
1457 btrfs_node_key_to_cpu(path
->nodes
[level
], key
,
1458 path
->slots
[level
] + 1);
1465 * look for inline back ref. if back ref is found, *ref_ret is set
1466 * to the address of inline back ref, and 0 is returned.
1468 * if back ref isn't found, *ref_ret is set to the address where it
1469 * should be inserted, and -ENOENT is returned.
1471 * if insert is true and there are too many inline back refs, the path
1472 * points to the extent item, and -EAGAIN is returned.
1474 * NOTE: inline back refs are ordered in the same way that back ref
1475 * items in the tree are ordered.
1477 static noinline_for_stack
1478 int lookup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1479 struct btrfs_root
*root
,
1480 struct btrfs_path
*path
,
1481 struct btrfs_extent_inline_ref
**ref_ret
,
1482 u64 bytenr
, u64 num_bytes
,
1483 u64 parent
, u64 root_objectid
,
1484 u64 owner
, u64 offset
, int insert
)
1486 struct btrfs_key key
;
1487 struct extent_buffer
*leaf
;
1488 struct btrfs_extent_item
*ei
;
1489 struct btrfs_extent_inline_ref
*iref
;
1499 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
1502 key
.objectid
= bytenr
;
1503 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1504 key
.offset
= num_bytes
;
1506 want
= extent_ref_type(parent
, owner
);
1508 extra_size
= btrfs_extent_inline_ref_size(want
);
1509 path
->keep_locks
= 1;
1514 * Owner is our parent level, so we can just add one to get the level
1515 * for the block we are interested in.
1517 if (skinny_metadata
&& owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1518 key
.type
= BTRFS_METADATA_ITEM_KEY
;
1523 ret
= btrfs_search_slot(trans
, root
, &key
, path
, extra_size
, 1);
1530 * We may be a newly converted file system which still has the old fat
1531 * extent entries for metadata, so try and see if we have one of those.
1533 if (ret
> 0 && skinny_metadata
) {
1534 skinny_metadata
= false;
1535 if (path
->slots
[0]) {
1537 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
1539 if (key
.objectid
== bytenr
&&
1540 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
1541 key
.offset
== num_bytes
)
1545 key
.objectid
= bytenr
;
1546 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1547 key
.offset
= num_bytes
;
1548 btrfs_release_path(path
);
1553 if (ret
&& !insert
) {
1556 } else if (WARN_ON(ret
)) {
1561 leaf
= path
->nodes
[0];
1562 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1563 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1564 if (item_size
< sizeof(*ei
)) {
1569 ret
= convert_extent_item_v0(trans
, root
, path
, owner
,
1575 leaf
= path
->nodes
[0];
1576 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1579 BUG_ON(item_size
< sizeof(*ei
));
1581 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1582 flags
= btrfs_extent_flags(leaf
, ei
);
1584 ptr
= (unsigned long)(ei
+ 1);
1585 end
= (unsigned long)ei
+ item_size
;
1587 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
&& !skinny_metadata
) {
1588 ptr
+= sizeof(struct btrfs_tree_block_info
);
1598 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1599 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1603 ptr
+= btrfs_extent_inline_ref_size(type
);
1607 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1608 struct btrfs_extent_data_ref
*dref
;
1609 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1610 if (match_extent_data_ref(leaf
, dref
, root_objectid
,
1615 if (hash_extent_data_ref_item(leaf
, dref
) <
1616 hash_extent_data_ref(root_objectid
, owner
, offset
))
1620 ref_offset
= btrfs_extent_inline_ref_offset(leaf
, iref
);
1622 if (parent
== ref_offset
) {
1626 if (ref_offset
< parent
)
1629 if (root_objectid
== ref_offset
) {
1633 if (ref_offset
< root_objectid
)
1637 ptr
+= btrfs_extent_inline_ref_size(type
);
1639 if (err
== -ENOENT
&& insert
) {
1640 if (item_size
+ extra_size
>=
1641 BTRFS_MAX_EXTENT_ITEM_SIZE(root
)) {
1646 * To add new inline back ref, we have to make sure
1647 * there is no corresponding back ref item.
1648 * For simplicity, we just do not add new inline back
1649 * ref if there is any kind of item for this block
1651 if (find_next_key(path
, 0, &key
) == 0 &&
1652 key
.objectid
== bytenr
&&
1653 key
.type
< BTRFS_BLOCK_GROUP_ITEM_KEY
) {
1658 *ref_ret
= (struct btrfs_extent_inline_ref
*)ptr
;
1661 path
->keep_locks
= 0;
1662 btrfs_unlock_up_safe(path
, 1);
1668 * helper to add new inline back ref
1670 static noinline_for_stack
1671 void setup_inline_extent_backref(struct btrfs_root
*root
,
1672 struct btrfs_path
*path
,
1673 struct btrfs_extent_inline_ref
*iref
,
1674 u64 parent
, u64 root_objectid
,
1675 u64 owner
, u64 offset
, int refs_to_add
,
1676 struct btrfs_delayed_extent_op
*extent_op
)
1678 struct extent_buffer
*leaf
;
1679 struct btrfs_extent_item
*ei
;
1682 unsigned long item_offset
;
1687 leaf
= path
->nodes
[0];
1688 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1689 item_offset
= (unsigned long)iref
- (unsigned long)ei
;
1691 type
= extent_ref_type(parent
, owner
);
1692 size
= btrfs_extent_inline_ref_size(type
);
1694 btrfs_extend_item(root
, path
, size
);
1696 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1697 refs
= btrfs_extent_refs(leaf
, ei
);
1698 refs
+= refs_to_add
;
1699 btrfs_set_extent_refs(leaf
, ei
, refs
);
1701 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1703 ptr
= (unsigned long)ei
+ item_offset
;
1704 end
= (unsigned long)ei
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
1705 if (ptr
< end
- size
)
1706 memmove_extent_buffer(leaf
, ptr
+ size
, ptr
,
1709 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1710 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
1711 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1712 struct btrfs_extent_data_ref
*dref
;
1713 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1714 btrfs_set_extent_data_ref_root(leaf
, dref
, root_objectid
);
1715 btrfs_set_extent_data_ref_objectid(leaf
, dref
, owner
);
1716 btrfs_set_extent_data_ref_offset(leaf
, dref
, offset
);
1717 btrfs_set_extent_data_ref_count(leaf
, dref
, refs_to_add
);
1718 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1719 struct btrfs_shared_data_ref
*sref
;
1720 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1721 btrfs_set_shared_data_ref_count(leaf
, sref
, refs_to_add
);
1722 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1723 } else if (type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
1724 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1726 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
1728 btrfs_mark_buffer_dirty(leaf
);
1731 static int lookup_extent_backref(struct btrfs_trans_handle
*trans
,
1732 struct btrfs_root
*root
,
1733 struct btrfs_path
*path
,
1734 struct btrfs_extent_inline_ref
**ref_ret
,
1735 u64 bytenr
, u64 num_bytes
, u64 parent
,
1736 u64 root_objectid
, u64 owner
, u64 offset
)
1740 ret
= lookup_inline_extent_backref(trans
, root
, path
, ref_ret
,
1741 bytenr
, num_bytes
, parent
,
1742 root_objectid
, owner
, offset
, 0);
1746 btrfs_release_path(path
);
1749 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1750 ret
= lookup_tree_block_ref(trans
, root
, path
, bytenr
, parent
,
1753 ret
= lookup_extent_data_ref(trans
, root
, path
, bytenr
, parent
,
1754 root_objectid
, owner
, offset
);
1760 * helper to update/remove inline back ref
1762 static noinline_for_stack
1763 void update_inline_extent_backref(struct btrfs_root
*root
,
1764 struct btrfs_path
*path
,
1765 struct btrfs_extent_inline_ref
*iref
,
1767 struct btrfs_delayed_extent_op
*extent_op
)
1769 struct extent_buffer
*leaf
;
1770 struct btrfs_extent_item
*ei
;
1771 struct btrfs_extent_data_ref
*dref
= NULL
;
1772 struct btrfs_shared_data_ref
*sref
= NULL
;
1780 leaf
= path
->nodes
[0];
1781 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1782 refs
= btrfs_extent_refs(leaf
, ei
);
1783 WARN_ON(refs_to_mod
< 0 && refs
+ refs_to_mod
<= 0);
1784 refs
+= refs_to_mod
;
1785 btrfs_set_extent_refs(leaf
, ei
, refs
);
1787 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1789 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1791 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1792 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1793 refs
= btrfs_extent_data_ref_count(leaf
, dref
);
1794 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1795 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1796 refs
= btrfs_shared_data_ref_count(leaf
, sref
);
1799 BUG_ON(refs_to_mod
!= -1);
1802 BUG_ON(refs_to_mod
< 0 && refs
< -refs_to_mod
);
1803 refs
+= refs_to_mod
;
1806 if (type
== BTRFS_EXTENT_DATA_REF_KEY
)
1807 btrfs_set_extent_data_ref_count(leaf
, dref
, refs
);
1809 btrfs_set_shared_data_ref_count(leaf
, sref
, refs
);
1811 size
= btrfs_extent_inline_ref_size(type
);
1812 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1813 ptr
= (unsigned long)iref
;
1814 end
= (unsigned long)ei
+ item_size
;
1815 if (ptr
+ size
< end
)
1816 memmove_extent_buffer(leaf
, ptr
, ptr
+ size
,
1819 btrfs_truncate_item(root
, path
, item_size
, 1);
1821 btrfs_mark_buffer_dirty(leaf
);
1824 static noinline_for_stack
1825 int insert_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1826 struct btrfs_root
*root
,
1827 struct btrfs_path
*path
,
1828 u64 bytenr
, u64 num_bytes
, u64 parent
,
1829 u64 root_objectid
, u64 owner
,
1830 u64 offset
, int refs_to_add
,
1831 struct btrfs_delayed_extent_op
*extent_op
)
1833 struct btrfs_extent_inline_ref
*iref
;
1836 ret
= lookup_inline_extent_backref(trans
, root
, path
, &iref
,
1837 bytenr
, num_bytes
, parent
,
1838 root_objectid
, owner
, offset
, 1);
1840 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
);
1841 update_inline_extent_backref(root
, path
, iref
,
1842 refs_to_add
, extent_op
);
1843 } else if (ret
== -ENOENT
) {
1844 setup_inline_extent_backref(root
, path
, iref
, parent
,
1845 root_objectid
, owner
, offset
,
1846 refs_to_add
, extent_op
);
1852 static int insert_extent_backref(struct btrfs_trans_handle
*trans
,
1853 struct btrfs_root
*root
,
1854 struct btrfs_path
*path
,
1855 u64 bytenr
, u64 parent
, u64 root_objectid
,
1856 u64 owner
, u64 offset
, int refs_to_add
)
1859 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1860 BUG_ON(refs_to_add
!= 1);
1861 ret
= insert_tree_block_ref(trans
, root
, path
, bytenr
,
1862 parent
, root_objectid
);
1864 ret
= insert_extent_data_ref(trans
, root
, path
, bytenr
,
1865 parent
, root_objectid
,
1866 owner
, offset
, refs_to_add
);
1871 static int remove_extent_backref(struct btrfs_trans_handle
*trans
,
1872 struct btrfs_root
*root
,
1873 struct btrfs_path
*path
,
1874 struct btrfs_extent_inline_ref
*iref
,
1875 int refs_to_drop
, int is_data
)
1879 BUG_ON(!is_data
&& refs_to_drop
!= 1);
1881 update_inline_extent_backref(root
, path
, iref
,
1882 -refs_to_drop
, NULL
);
1883 } else if (is_data
) {
1884 ret
= remove_extent_data_ref(trans
, root
, path
, refs_to_drop
);
1886 ret
= btrfs_del_item(trans
, root
, path
);
1891 static int btrfs_issue_discard(struct block_device
*bdev
,
1894 return blkdev_issue_discard(bdev
, start
>> 9, len
>> 9, GFP_NOFS
, 0);
1897 static int btrfs_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
1898 u64 num_bytes
, u64
*actual_bytes
)
1901 u64 discarded_bytes
= 0;
1902 struct btrfs_bio
*bbio
= NULL
;
1905 /* Tell the block device(s) that the sectors can be discarded */
1906 ret
= btrfs_map_block(root
->fs_info
, REQ_DISCARD
,
1907 bytenr
, &num_bytes
, &bbio
, 0);
1908 /* Error condition is -ENOMEM */
1910 struct btrfs_bio_stripe
*stripe
= bbio
->stripes
;
1914 for (i
= 0; i
< bbio
->num_stripes
; i
++, stripe
++) {
1915 if (!stripe
->dev
->can_discard
)
1918 ret
= btrfs_issue_discard(stripe
->dev
->bdev
,
1922 discarded_bytes
+= stripe
->length
;
1923 else if (ret
!= -EOPNOTSUPP
)
1924 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1927 * Just in case we get back EOPNOTSUPP for some reason,
1928 * just ignore the return value so we don't screw up
1929 * people calling discard_extent.
1937 *actual_bytes
= discarded_bytes
;
1940 if (ret
== -EOPNOTSUPP
)
1945 /* Can return -ENOMEM */
1946 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1947 struct btrfs_root
*root
,
1948 u64 bytenr
, u64 num_bytes
, u64 parent
,
1949 u64 root_objectid
, u64 owner
, u64 offset
, int for_cow
)
1952 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1954 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
&&
1955 root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
1957 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1958 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
1960 parent
, root_objectid
, (int)owner
,
1961 BTRFS_ADD_DELAYED_REF
, NULL
, for_cow
);
1963 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
1965 parent
, root_objectid
, owner
, offset
,
1966 BTRFS_ADD_DELAYED_REF
, NULL
, for_cow
);
1971 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1972 struct btrfs_root
*root
,
1973 u64 bytenr
, u64 num_bytes
,
1974 u64 parent
, u64 root_objectid
,
1975 u64 owner
, u64 offset
, int refs_to_add
,
1976 struct btrfs_delayed_extent_op
*extent_op
)
1978 struct btrfs_path
*path
;
1979 struct extent_buffer
*leaf
;
1980 struct btrfs_extent_item
*item
;
1984 path
= btrfs_alloc_path();
1989 path
->leave_spinning
= 1;
1990 /* this will setup the path even if it fails to insert the back ref */
1991 ret
= insert_inline_extent_backref(trans
, root
->fs_info
->extent_root
,
1992 path
, bytenr
, num_bytes
, parent
,
1993 root_objectid
, owner
, offset
,
1994 refs_to_add
, extent_op
);
1998 leaf
= path
->nodes
[0];
1999 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2000 refs
= btrfs_extent_refs(leaf
, item
);
2001 btrfs_set_extent_refs(leaf
, item
, refs
+ refs_to_add
);
2003 __run_delayed_extent_op(extent_op
, leaf
, item
);
2005 btrfs_mark_buffer_dirty(leaf
);
2006 btrfs_release_path(path
);
2009 path
->leave_spinning
= 1;
2011 /* now insert the actual backref */
2012 ret
= insert_extent_backref(trans
, root
->fs_info
->extent_root
,
2013 path
, bytenr
, parent
, root_objectid
,
2014 owner
, offset
, refs_to_add
);
2016 btrfs_abort_transaction(trans
, root
, ret
);
2018 btrfs_free_path(path
);
2022 static int run_delayed_data_ref(struct btrfs_trans_handle
*trans
,
2023 struct btrfs_root
*root
,
2024 struct btrfs_delayed_ref_node
*node
,
2025 struct btrfs_delayed_extent_op
*extent_op
,
2026 int insert_reserved
)
2029 struct btrfs_delayed_data_ref
*ref
;
2030 struct btrfs_key ins
;
2035 ins
.objectid
= node
->bytenr
;
2036 ins
.offset
= node
->num_bytes
;
2037 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2039 ref
= btrfs_delayed_node_to_data_ref(node
);
2040 trace_run_delayed_data_ref(node
, ref
, node
->action
);
2042 if (node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2043 parent
= ref
->parent
;
2045 ref_root
= ref
->root
;
2047 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2049 flags
|= extent_op
->flags_to_set
;
2050 ret
= alloc_reserved_file_extent(trans
, root
,
2051 parent
, ref_root
, flags
,
2052 ref
->objectid
, ref
->offset
,
2053 &ins
, node
->ref_mod
);
2054 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2055 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
2056 node
->num_bytes
, parent
,
2057 ref_root
, ref
->objectid
,
2058 ref
->offset
, node
->ref_mod
,
2060 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2061 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
2062 node
->num_bytes
, parent
,
2063 ref_root
, ref
->objectid
,
2064 ref
->offset
, node
->ref_mod
,
2072 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
2073 struct extent_buffer
*leaf
,
2074 struct btrfs_extent_item
*ei
)
2076 u64 flags
= btrfs_extent_flags(leaf
, ei
);
2077 if (extent_op
->update_flags
) {
2078 flags
|= extent_op
->flags_to_set
;
2079 btrfs_set_extent_flags(leaf
, ei
, flags
);
2082 if (extent_op
->update_key
) {
2083 struct btrfs_tree_block_info
*bi
;
2084 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
));
2085 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
2086 btrfs_set_tree_block_key(leaf
, bi
, &extent_op
->key
);
2090 static int run_delayed_extent_op(struct btrfs_trans_handle
*trans
,
2091 struct btrfs_root
*root
,
2092 struct btrfs_delayed_ref_node
*node
,
2093 struct btrfs_delayed_extent_op
*extent_op
)
2095 struct btrfs_key key
;
2096 struct btrfs_path
*path
;
2097 struct btrfs_extent_item
*ei
;
2098 struct extent_buffer
*leaf
;
2102 int metadata
= !extent_op
->is_data
;
2107 if (metadata
&& !btrfs_fs_incompat(root
->fs_info
, SKINNY_METADATA
))
2110 path
= btrfs_alloc_path();
2114 key
.objectid
= node
->bytenr
;
2117 key
.type
= BTRFS_METADATA_ITEM_KEY
;
2118 key
.offset
= extent_op
->level
;
2120 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2121 key
.offset
= node
->num_bytes
;
2126 path
->leave_spinning
= 1;
2127 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
,
2135 if (path
->slots
[0] > 0) {
2137 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
2139 if (key
.objectid
== node
->bytenr
&&
2140 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
2141 key
.offset
== node
->num_bytes
)
2145 btrfs_release_path(path
);
2148 key
.objectid
= node
->bytenr
;
2149 key
.offset
= node
->num_bytes
;
2150 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2159 leaf
= path
->nodes
[0];
2160 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2161 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2162 if (item_size
< sizeof(*ei
)) {
2163 ret
= convert_extent_item_v0(trans
, root
->fs_info
->extent_root
,
2169 leaf
= path
->nodes
[0];
2170 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2173 BUG_ON(item_size
< sizeof(*ei
));
2174 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2175 __run_delayed_extent_op(extent_op
, leaf
, ei
);
2177 btrfs_mark_buffer_dirty(leaf
);
2179 btrfs_free_path(path
);
2183 static int run_delayed_tree_ref(struct btrfs_trans_handle
*trans
,
2184 struct btrfs_root
*root
,
2185 struct btrfs_delayed_ref_node
*node
,
2186 struct btrfs_delayed_extent_op
*extent_op
,
2187 int insert_reserved
)
2190 struct btrfs_delayed_tree_ref
*ref
;
2191 struct btrfs_key ins
;
2194 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
2197 ref
= btrfs_delayed_node_to_tree_ref(node
);
2198 trace_run_delayed_tree_ref(node
, ref
, node
->action
);
2200 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2201 parent
= ref
->parent
;
2203 ref_root
= ref
->root
;
2205 ins
.objectid
= node
->bytenr
;
2206 if (skinny_metadata
) {
2207 ins
.offset
= ref
->level
;
2208 ins
.type
= BTRFS_METADATA_ITEM_KEY
;
2210 ins
.offset
= node
->num_bytes
;
2211 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2214 BUG_ON(node
->ref_mod
!= 1);
2215 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2216 BUG_ON(!extent_op
|| !extent_op
->update_flags
);
2217 ret
= alloc_reserved_tree_block(trans
, root
,
2219 extent_op
->flags_to_set
,
2222 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2223 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
2224 node
->num_bytes
, parent
, ref_root
,
2225 ref
->level
, 0, 1, extent_op
);
2226 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2227 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
2228 node
->num_bytes
, parent
, ref_root
,
2229 ref
->level
, 0, 1, extent_op
);
2236 /* helper function to actually process a single delayed ref entry */
2237 static int run_one_delayed_ref(struct btrfs_trans_handle
*trans
,
2238 struct btrfs_root
*root
,
2239 struct btrfs_delayed_ref_node
*node
,
2240 struct btrfs_delayed_extent_op
*extent_op
,
2241 int insert_reserved
)
2245 if (trans
->aborted
) {
2246 if (insert_reserved
)
2247 btrfs_pin_extent(root
, node
->bytenr
,
2248 node
->num_bytes
, 1);
2252 if (btrfs_delayed_ref_is_head(node
)) {
2253 struct btrfs_delayed_ref_head
*head
;
2255 * we've hit the end of the chain and we were supposed
2256 * to insert this extent into the tree. But, it got
2257 * deleted before we ever needed to insert it, so all
2258 * we have to do is clean up the accounting
2261 head
= btrfs_delayed_node_to_head(node
);
2262 trace_run_delayed_ref_head(node
, head
, node
->action
);
2264 if (insert_reserved
) {
2265 btrfs_pin_extent(root
, node
->bytenr
,
2266 node
->num_bytes
, 1);
2267 if (head
->is_data
) {
2268 ret
= btrfs_del_csums(trans
, root
,
2276 if (node
->type
== BTRFS_TREE_BLOCK_REF_KEY
||
2277 node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2278 ret
= run_delayed_tree_ref(trans
, root
, node
, extent_op
,
2280 else if (node
->type
== BTRFS_EXTENT_DATA_REF_KEY
||
2281 node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2282 ret
= run_delayed_data_ref(trans
, root
, node
, extent_op
,
2289 static noinline
struct btrfs_delayed_ref_node
*
2290 select_delayed_ref(struct btrfs_delayed_ref_head
*head
)
2292 struct rb_node
*node
;
2293 struct btrfs_delayed_ref_node
*ref
, *last
= NULL
;;
2296 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2297 * this prevents ref count from going down to zero when
2298 * there still are pending delayed ref.
2300 node
= rb_first(&head
->ref_root
);
2302 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
2304 if (ref
->action
== BTRFS_ADD_DELAYED_REF
)
2306 else if (last
== NULL
)
2308 node
= rb_next(node
);
2314 * Returns 0 on success or if called with an already aborted transaction.
2315 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2317 static noinline
int __btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2318 struct btrfs_root
*root
,
2321 struct btrfs_delayed_ref_root
*delayed_refs
;
2322 struct btrfs_delayed_ref_node
*ref
;
2323 struct btrfs_delayed_ref_head
*locked_ref
= NULL
;
2324 struct btrfs_delayed_extent_op
*extent_op
;
2325 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2326 ktime_t start
= ktime_get();
2328 unsigned long count
= 0;
2329 unsigned long actual_count
= 0;
2330 int must_insert_reserved
= 0;
2332 delayed_refs
= &trans
->transaction
->delayed_refs
;
2338 spin_lock(&delayed_refs
->lock
);
2339 locked_ref
= btrfs_select_ref_head(trans
);
2341 spin_unlock(&delayed_refs
->lock
);
2345 /* grab the lock that says we are going to process
2346 * all the refs for this head */
2347 ret
= btrfs_delayed_ref_lock(trans
, locked_ref
);
2348 spin_unlock(&delayed_refs
->lock
);
2350 * we may have dropped the spin lock to get the head
2351 * mutex lock, and that might have given someone else
2352 * time to free the head. If that's true, it has been
2353 * removed from our list and we can move on.
2355 if (ret
== -EAGAIN
) {
2363 * We need to try and merge add/drops of the same ref since we
2364 * can run into issues with relocate dropping the implicit ref
2365 * and then it being added back again before the drop can
2366 * finish. If we merged anything we need to re-loop so we can
2369 spin_lock(&locked_ref
->lock
);
2370 btrfs_merge_delayed_refs(trans
, fs_info
, delayed_refs
,
2374 * locked_ref is the head node, so we have to go one
2375 * node back for any delayed ref updates
2377 ref
= select_delayed_ref(locked_ref
);
2379 if (ref
&& ref
->seq
&&
2380 btrfs_check_delayed_seq(fs_info
, delayed_refs
, ref
->seq
)) {
2381 spin_unlock(&locked_ref
->lock
);
2382 btrfs_delayed_ref_unlock(locked_ref
);
2383 spin_lock(&delayed_refs
->lock
);
2384 locked_ref
->processing
= 0;
2385 delayed_refs
->num_heads_ready
++;
2386 spin_unlock(&delayed_refs
->lock
);
2394 * record the must insert reserved flag before we
2395 * drop the spin lock.
2397 must_insert_reserved
= locked_ref
->must_insert_reserved
;
2398 locked_ref
->must_insert_reserved
= 0;
2400 extent_op
= locked_ref
->extent_op
;
2401 locked_ref
->extent_op
= NULL
;
2406 /* All delayed refs have been processed, Go ahead
2407 * and send the head node to run_one_delayed_ref,
2408 * so that any accounting fixes can happen
2410 ref
= &locked_ref
->node
;
2412 if (extent_op
&& must_insert_reserved
) {
2413 btrfs_free_delayed_extent_op(extent_op
);
2418 spin_unlock(&locked_ref
->lock
);
2419 ret
= run_delayed_extent_op(trans
, root
,
2421 btrfs_free_delayed_extent_op(extent_op
);
2425 * Need to reset must_insert_reserved if
2426 * there was an error so the abort stuff
2427 * can cleanup the reserved space
2430 if (must_insert_reserved
)
2431 locked_ref
->must_insert_reserved
= 1;
2432 locked_ref
->processing
= 0;
2433 btrfs_debug(fs_info
, "run_delayed_extent_op returned %d", ret
);
2434 btrfs_delayed_ref_unlock(locked_ref
);
2441 * Need to drop our head ref lock and re-aqcuire the
2442 * delayed ref lock and then re-check to make sure
2445 spin_unlock(&locked_ref
->lock
);
2446 spin_lock(&delayed_refs
->lock
);
2447 spin_lock(&locked_ref
->lock
);
2448 if (rb_first(&locked_ref
->ref_root
) ||
2449 locked_ref
->extent_op
) {
2450 spin_unlock(&locked_ref
->lock
);
2451 spin_unlock(&delayed_refs
->lock
);
2455 delayed_refs
->num_heads
--;
2456 rb_erase(&locked_ref
->href_node
,
2457 &delayed_refs
->href_root
);
2458 spin_unlock(&delayed_refs
->lock
);
2462 rb_erase(&ref
->rb_node
, &locked_ref
->ref_root
);
2464 atomic_dec(&delayed_refs
->num_entries
);
2466 if (!btrfs_delayed_ref_is_head(ref
)) {
2468 * when we play the delayed ref, also correct the
2471 switch (ref
->action
) {
2472 case BTRFS_ADD_DELAYED_REF
:
2473 case BTRFS_ADD_DELAYED_EXTENT
:
2474 locked_ref
->node
.ref_mod
-= ref
->ref_mod
;
2476 case BTRFS_DROP_DELAYED_REF
:
2477 locked_ref
->node
.ref_mod
+= ref
->ref_mod
;
2483 spin_unlock(&locked_ref
->lock
);
2485 ret
= run_one_delayed_ref(trans
, root
, ref
, extent_op
,
2486 must_insert_reserved
);
2488 btrfs_free_delayed_extent_op(extent_op
);
2490 locked_ref
->processing
= 0;
2491 btrfs_delayed_ref_unlock(locked_ref
);
2492 btrfs_put_delayed_ref(ref
);
2493 btrfs_debug(fs_info
, "run_one_delayed_ref returned %d", ret
);
2498 * If this node is a head, that means all the refs in this head
2499 * have been dealt with, and we will pick the next head to deal
2500 * with, so we must unlock the head and drop it from the cluster
2501 * list before we release it.
2503 if (btrfs_delayed_ref_is_head(ref
)) {
2504 btrfs_delayed_ref_unlock(locked_ref
);
2507 btrfs_put_delayed_ref(ref
);
2513 * We don't want to include ref heads since we can have empty ref heads
2514 * and those will drastically skew our runtime down since we just do
2515 * accounting, no actual extent tree updates.
2517 if (actual_count
> 0) {
2518 u64 runtime
= ktime_to_ns(ktime_sub(ktime_get(), start
));
2522 * We weigh the current average higher than our current runtime
2523 * to avoid large swings in the average.
2525 spin_lock(&delayed_refs
->lock
);
2526 avg
= fs_info
->avg_delayed_ref_runtime
* 3 + runtime
;
2527 avg
= div64_u64(avg
, 4);
2528 fs_info
->avg_delayed_ref_runtime
= avg
;
2529 spin_unlock(&delayed_refs
->lock
);
2534 #ifdef SCRAMBLE_DELAYED_REFS
2536 * Normally delayed refs get processed in ascending bytenr order. This
2537 * correlates in most cases to the order added. To expose dependencies on this
2538 * order, we start to process the tree in the middle instead of the beginning
2540 static u64
find_middle(struct rb_root
*root
)
2542 struct rb_node
*n
= root
->rb_node
;
2543 struct btrfs_delayed_ref_node
*entry
;
2546 u64 first
= 0, last
= 0;
2550 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2551 first
= entry
->bytenr
;
2555 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2556 last
= entry
->bytenr
;
2561 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2562 WARN_ON(!entry
->in_tree
);
2564 middle
= entry
->bytenr
;
2577 int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle
*trans
,
2578 struct btrfs_fs_info
*fs_info
)
2580 struct qgroup_update
*qgroup_update
;
2583 if (list_empty(&trans
->qgroup_ref_list
) !=
2584 !trans
->delayed_ref_elem
.seq
) {
2585 /* list without seq or seq without list */
2587 "qgroup accounting update error, list is%s empty, seq is %#x.%x",
2588 list_empty(&trans
->qgroup_ref_list
) ? "" : " not",
2589 (u32
)(trans
->delayed_ref_elem
.seq
>> 32),
2590 (u32
)trans
->delayed_ref_elem
.seq
);
2594 if (!trans
->delayed_ref_elem
.seq
)
2597 while (!list_empty(&trans
->qgroup_ref_list
)) {
2598 qgroup_update
= list_first_entry(&trans
->qgroup_ref_list
,
2599 struct qgroup_update
, list
);
2600 list_del(&qgroup_update
->list
);
2602 ret
= btrfs_qgroup_account_ref(
2603 trans
, fs_info
, qgroup_update
->node
,
2604 qgroup_update
->extent_op
);
2605 kfree(qgroup_update
);
2608 btrfs_put_tree_mod_seq(fs_info
, &trans
->delayed_ref_elem
);
2613 static inline u64
heads_to_leaves(struct btrfs_root
*root
, u64 heads
)
2617 num_bytes
= heads
* (sizeof(struct btrfs_extent_item
) +
2618 sizeof(struct btrfs_extent_inline_ref
));
2619 if (!btrfs_fs_incompat(root
->fs_info
, SKINNY_METADATA
))
2620 num_bytes
+= heads
* sizeof(struct btrfs_tree_block_info
);
2623 * We don't ever fill up leaves all the way so multiply by 2 just to be
2624 * closer to what we're really going to want to ouse.
2626 return div64_u64(num_bytes
, BTRFS_LEAF_DATA_SIZE(root
));
2629 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle
*trans
,
2630 struct btrfs_root
*root
)
2632 struct btrfs_block_rsv
*global_rsv
;
2633 u64 num_heads
= trans
->transaction
->delayed_refs
.num_heads_ready
;
2637 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
2638 num_heads
= heads_to_leaves(root
, num_heads
);
2640 num_bytes
+= (num_heads
- 1) * root
->leafsize
;
2642 global_rsv
= &root
->fs_info
->global_block_rsv
;
2645 * If we can't allocate any more chunks lets make sure we have _lots_ of
2646 * wiggle room since running delayed refs can create more delayed refs.
2648 if (global_rsv
->space_info
->full
)
2651 spin_lock(&global_rsv
->lock
);
2652 if (global_rsv
->reserved
<= num_bytes
)
2654 spin_unlock(&global_rsv
->lock
);
2658 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle
*trans
,
2659 struct btrfs_root
*root
)
2661 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2663 atomic_read(&trans
->transaction
->delayed_refs
.num_entries
);
2667 avg_runtime
= fs_info
->avg_delayed_ref_runtime
;
2668 if (num_entries
* avg_runtime
>= NSEC_PER_SEC
)
2671 return btrfs_check_space_for_delayed_refs(trans
, root
);
2675 * this starts processing the delayed reference count updates and
2676 * extent insertions we have queued up so far. count can be
2677 * 0, which means to process everything in the tree at the start
2678 * of the run (but not newly added entries), or it can be some target
2679 * number you'd like to process.
2681 * Returns 0 on success or if called with an aborted transaction
2682 * Returns <0 on error and aborts the transaction
2684 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2685 struct btrfs_root
*root
, unsigned long count
)
2687 struct rb_node
*node
;
2688 struct btrfs_delayed_ref_root
*delayed_refs
;
2689 struct btrfs_delayed_ref_head
*head
;
2691 int run_all
= count
== (unsigned long)-1;
2694 /* We'll clean this up in btrfs_cleanup_transaction */
2698 if (root
== root
->fs_info
->extent_root
)
2699 root
= root
->fs_info
->tree_root
;
2701 btrfs_delayed_refs_qgroup_accounting(trans
, root
->fs_info
);
2703 delayed_refs
= &trans
->transaction
->delayed_refs
;
2705 count
= atomic_read(&delayed_refs
->num_entries
) * 2;
2710 #ifdef SCRAMBLE_DELAYED_REFS
2711 delayed_refs
->run_delayed_start
= find_middle(&delayed_refs
->root
);
2713 ret
= __btrfs_run_delayed_refs(trans
, root
, count
);
2715 btrfs_abort_transaction(trans
, root
, ret
);
2720 if (!list_empty(&trans
->new_bgs
))
2721 btrfs_create_pending_block_groups(trans
, root
);
2723 spin_lock(&delayed_refs
->lock
);
2724 node
= rb_first(&delayed_refs
->href_root
);
2726 spin_unlock(&delayed_refs
->lock
);
2729 count
= (unsigned long)-1;
2732 head
= rb_entry(node
, struct btrfs_delayed_ref_head
,
2734 if (btrfs_delayed_ref_is_head(&head
->node
)) {
2735 struct btrfs_delayed_ref_node
*ref
;
2738 atomic_inc(&ref
->refs
);
2740 spin_unlock(&delayed_refs
->lock
);
2742 * Mutex was contended, block until it's
2743 * released and try again
2745 mutex_lock(&head
->mutex
);
2746 mutex_unlock(&head
->mutex
);
2748 btrfs_put_delayed_ref(ref
);
2754 node
= rb_next(node
);
2756 spin_unlock(&delayed_refs
->lock
);
2761 assert_qgroups_uptodate(trans
);
2765 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
2766 struct btrfs_root
*root
,
2767 u64 bytenr
, u64 num_bytes
, u64 flags
,
2768 int level
, int is_data
)
2770 struct btrfs_delayed_extent_op
*extent_op
;
2773 extent_op
= btrfs_alloc_delayed_extent_op();
2777 extent_op
->flags_to_set
= flags
;
2778 extent_op
->update_flags
= 1;
2779 extent_op
->update_key
= 0;
2780 extent_op
->is_data
= is_data
? 1 : 0;
2781 extent_op
->level
= level
;
2783 ret
= btrfs_add_delayed_extent_op(root
->fs_info
, trans
, bytenr
,
2784 num_bytes
, extent_op
);
2786 btrfs_free_delayed_extent_op(extent_op
);
2790 static noinline
int check_delayed_ref(struct btrfs_trans_handle
*trans
,
2791 struct btrfs_root
*root
,
2792 struct btrfs_path
*path
,
2793 u64 objectid
, u64 offset
, u64 bytenr
)
2795 struct btrfs_delayed_ref_head
*head
;
2796 struct btrfs_delayed_ref_node
*ref
;
2797 struct btrfs_delayed_data_ref
*data_ref
;
2798 struct btrfs_delayed_ref_root
*delayed_refs
;
2799 struct rb_node
*node
;
2802 delayed_refs
= &trans
->transaction
->delayed_refs
;
2803 spin_lock(&delayed_refs
->lock
);
2804 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
2806 spin_unlock(&delayed_refs
->lock
);
2810 if (!mutex_trylock(&head
->mutex
)) {
2811 atomic_inc(&head
->node
.refs
);
2812 spin_unlock(&delayed_refs
->lock
);
2814 btrfs_release_path(path
);
2817 * Mutex was contended, block until it's released and let
2820 mutex_lock(&head
->mutex
);
2821 mutex_unlock(&head
->mutex
);
2822 btrfs_put_delayed_ref(&head
->node
);
2825 spin_unlock(&delayed_refs
->lock
);
2827 spin_lock(&head
->lock
);
2828 node
= rb_first(&head
->ref_root
);
2830 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2831 node
= rb_next(node
);
2833 /* If it's a shared ref we know a cross reference exists */
2834 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
) {
2839 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
2842 * If our ref doesn't match the one we're currently looking at
2843 * then we have a cross reference.
2845 if (data_ref
->root
!= root
->root_key
.objectid
||
2846 data_ref
->objectid
!= objectid
||
2847 data_ref
->offset
!= offset
) {
2852 spin_unlock(&head
->lock
);
2853 mutex_unlock(&head
->mutex
);
2857 static noinline
int check_committed_ref(struct btrfs_trans_handle
*trans
,
2858 struct btrfs_root
*root
,
2859 struct btrfs_path
*path
,
2860 u64 objectid
, u64 offset
, u64 bytenr
)
2862 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2863 struct extent_buffer
*leaf
;
2864 struct btrfs_extent_data_ref
*ref
;
2865 struct btrfs_extent_inline_ref
*iref
;
2866 struct btrfs_extent_item
*ei
;
2867 struct btrfs_key key
;
2871 key
.objectid
= bytenr
;
2872 key
.offset
= (u64
)-1;
2873 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2875 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
2878 BUG_ON(ret
== 0); /* Corruption */
2881 if (path
->slots
[0] == 0)
2885 leaf
= path
->nodes
[0];
2886 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2888 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
2892 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2893 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2894 if (item_size
< sizeof(*ei
)) {
2895 WARN_ON(item_size
!= sizeof(struct btrfs_extent_item_v0
));
2899 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2901 if (item_size
!= sizeof(*ei
) +
2902 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
2905 if (btrfs_extent_generation(leaf
, ei
) <=
2906 btrfs_root_last_snapshot(&root
->root_item
))
2909 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
2910 if (btrfs_extent_inline_ref_type(leaf
, iref
) !=
2911 BTRFS_EXTENT_DATA_REF_KEY
)
2914 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
2915 if (btrfs_extent_refs(leaf
, ei
) !=
2916 btrfs_extent_data_ref_count(leaf
, ref
) ||
2917 btrfs_extent_data_ref_root(leaf
, ref
) !=
2918 root
->root_key
.objectid
||
2919 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
2920 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
2928 int btrfs_cross_ref_exist(struct btrfs_trans_handle
*trans
,
2929 struct btrfs_root
*root
,
2930 u64 objectid
, u64 offset
, u64 bytenr
)
2932 struct btrfs_path
*path
;
2936 path
= btrfs_alloc_path();
2941 ret
= check_committed_ref(trans
, root
, path
, objectid
,
2943 if (ret
&& ret
!= -ENOENT
)
2946 ret2
= check_delayed_ref(trans
, root
, path
, objectid
,
2948 } while (ret2
== -EAGAIN
);
2950 if (ret2
&& ret2
!= -ENOENT
) {
2955 if (ret
!= -ENOENT
|| ret2
!= -ENOENT
)
2958 btrfs_free_path(path
);
2959 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
2964 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
2965 struct btrfs_root
*root
,
2966 struct extent_buffer
*buf
,
2967 int full_backref
, int inc
, int for_cow
)
2974 struct btrfs_key key
;
2975 struct btrfs_file_extent_item
*fi
;
2979 int (*process_func
)(struct btrfs_trans_handle
*, struct btrfs_root
*,
2980 u64
, u64
, u64
, u64
, u64
, u64
, int);
2982 ref_root
= btrfs_header_owner(buf
);
2983 nritems
= btrfs_header_nritems(buf
);
2984 level
= btrfs_header_level(buf
);
2986 if (!root
->ref_cows
&& level
== 0)
2990 process_func
= btrfs_inc_extent_ref
;
2992 process_func
= btrfs_free_extent
;
2995 parent
= buf
->start
;
2999 for (i
= 0; i
< nritems
; i
++) {
3001 btrfs_item_key_to_cpu(buf
, &key
, i
);
3002 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
3004 fi
= btrfs_item_ptr(buf
, i
,
3005 struct btrfs_file_extent_item
);
3006 if (btrfs_file_extent_type(buf
, fi
) ==
3007 BTRFS_FILE_EXTENT_INLINE
)
3009 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
3013 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
3014 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
3015 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
3016 parent
, ref_root
, key
.objectid
,
3017 key
.offset
, for_cow
);
3021 bytenr
= btrfs_node_blockptr(buf
, i
);
3022 num_bytes
= btrfs_level_size(root
, level
- 1);
3023 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
3024 parent
, ref_root
, level
- 1, 0,
3035 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3036 struct extent_buffer
*buf
, int full_backref
, int for_cow
)
3038 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1, for_cow
);
3041 int btrfs_dec_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3042 struct extent_buffer
*buf
, int full_backref
, int for_cow
)
3044 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 0, for_cow
);
3047 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
3048 struct btrfs_root
*root
,
3049 struct btrfs_path
*path
,
3050 struct btrfs_block_group_cache
*cache
)
3053 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
3055 struct extent_buffer
*leaf
;
3057 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
3060 BUG_ON(ret
); /* Corruption */
3062 leaf
= path
->nodes
[0];
3063 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
3064 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
3065 btrfs_mark_buffer_dirty(leaf
);
3066 btrfs_release_path(path
);
3069 btrfs_abort_transaction(trans
, root
, ret
);
3076 static struct btrfs_block_group_cache
*
3077 next_block_group(struct btrfs_root
*root
,
3078 struct btrfs_block_group_cache
*cache
)
3080 struct rb_node
*node
;
3081 spin_lock(&root
->fs_info
->block_group_cache_lock
);
3082 node
= rb_next(&cache
->cache_node
);
3083 btrfs_put_block_group(cache
);
3085 cache
= rb_entry(node
, struct btrfs_block_group_cache
,
3087 btrfs_get_block_group(cache
);
3090 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
3094 static int cache_save_setup(struct btrfs_block_group_cache
*block_group
,
3095 struct btrfs_trans_handle
*trans
,
3096 struct btrfs_path
*path
)
3098 struct btrfs_root
*root
= block_group
->fs_info
->tree_root
;
3099 struct inode
*inode
= NULL
;
3101 int dcs
= BTRFS_DC_ERROR
;
3107 * If this block group is smaller than 100 megs don't bother caching the
3110 if (block_group
->key
.offset
< (100 * 1024 * 1024)) {
3111 spin_lock(&block_group
->lock
);
3112 block_group
->disk_cache_state
= BTRFS_DC_WRITTEN
;
3113 spin_unlock(&block_group
->lock
);
3118 inode
= lookup_free_space_inode(root
, block_group
, path
);
3119 if (IS_ERR(inode
) && PTR_ERR(inode
) != -ENOENT
) {
3120 ret
= PTR_ERR(inode
);
3121 btrfs_release_path(path
);
3125 if (IS_ERR(inode
)) {
3129 if (block_group
->ro
)
3132 ret
= create_free_space_inode(root
, trans
, block_group
, path
);
3138 /* We've already setup this transaction, go ahead and exit */
3139 if (block_group
->cache_generation
== trans
->transid
&&
3140 i_size_read(inode
)) {
3141 dcs
= BTRFS_DC_SETUP
;
3146 * We want to set the generation to 0, that way if anything goes wrong
3147 * from here on out we know not to trust this cache when we load up next
3150 BTRFS_I(inode
)->generation
= 0;
3151 ret
= btrfs_update_inode(trans
, root
, inode
);
3154 if (i_size_read(inode
) > 0) {
3155 ret
= btrfs_check_trunc_cache_free_space(root
,
3156 &root
->fs_info
->global_block_rsv
);
3160 ret
= btrfs_truncate_free_space_cache(root
, trans
, inode
);
3165 spin_lock(&block_group
->lock
);
3166 if (block_group
->cached
!= BTRFS_CACHE_FINISHED
||
3167 !btrfs_test_opt(root
, SPACE_CACHE
)) {
3169 * don't bother trying to write stuff out _if_
3170 * a) we're not cached,
3171 * b) we're with nospace_cache mount option.
3173 dcs
= BTRFS_DC_WRITTEN
;
3174 spin_unlock(&block_group
->lock
);
3177 spin_unlock(&block_group
->lock
);
3180 * Try to preallocate enough space based on how big the block group is.
3181 * Keep in mind this has to include any pinned space which could end up
3182 * taking up quite a bit since it's not folded into the other space
3185 num_pages
= (int)div64_u64(block_group
->key
.offset
, 256 * 1024 * 1024);
3190 num_pages
*= PAGE_CACHE_SIZE
;
3192 ret
= btrfs_check_data_free_space(inode
, num_pages
);
3196 ret
= btrfs_prealloc_file_range_trans(inode
, trans
, 0, 0, num_pages
,
3197 num_pages
, num_pages
,
3200 dcs
= BTRFS_DC_SETUP
;
3201 btrfs_free_reserved_data_space(inode
, num_pages
);
3206 btrfs_release_path(path
);
3208 spin_lock(&block_group
->lock
);
3209 if (!ret
&& dcs
== BTRFS_DC_SETUP
)
3210 block_group
->cache_generation
= trans
->transid
;
3211 block_group
->disk_cache_state
= dcs
;
3212 spin_unlock(&block_group
->lock
);
3217 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
3218 struct btrfs_root
*root
)
3220 struct btrfs_block_group_cache
*cache
;
3222 struct btrfs_path
*path
;
3225 path
= btrfs_alloc_path();
3231 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
3233 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
)
3235 cache
= next_block_group(root
, cache
);
3243 err
= cache_save_setup(cache
, trans
, path
);
3244 last
= cache
->key
.objectid
+ cache
->key
.offset
;
3245 btrfs_put_block_group(cache
);
3250 err
= btrfs_run_delayed_refs(trans
, root
,
3252 if (err
) /* File system offline */
3256 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
3258 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
) {
3259 btrfs_put_block_group(cache
);
3265 cache
= next_block_group(root
, cache
);
3274 if (cache
->disk_cache_state
== BTRFS_DC_SETUP
)
3275 cache
->disk_cache_state
= BTRFS_DC_NEED_WRITE
;
3277 last
= cache
->key
.objectid
+ cache
->key
.offset
;
3279 err
= write_one_cache_group(trans
, root
, path
, cache
);
3280 btrfs_put_block_group(cache
);
3281 if (err
) /* File system offline */
3287 * I don't think this is needed since we're just marking our
3288 * preallocated extent as written, but just in case it can't
3292 err
= btrfs_run_delayed_refs(trans
, root
,
3294 if (err
) /* File system offline */
3298 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
3301 * Really this shouldn't happen, but it could if we
3302 * couldn't write the entire preallocated extent and
3303 * splitting the extent resulted in a new block.
3306 btrfs_put_block_group(cache
);
3309 if (cache
->disk_cache_state
== BTRFS_DC_NEED_WRITE
)
3311 cache
= next_block_group(root
, cache
);
3320 err
= btrfs_write_out_cache(root
, trans
, cache
, path
);
3323 * If we didn't have an error then the cache state is still
3324 * NEED_WRITE, so we can set it to WRITTEN.
3326 if (!err
&& cache
->disk_cache_state
== BTRFS_DC_NEED_WRITE
)
3327 cache
->disk_cache_state
= BTRFS_DC_WRITTEN
;
3328 last
= cache
->key
.objectid
+ cache
->key
.offset
;
3329 btrfs_put_block_group(cache
);
3333 btrfs_free_path(path
);
3337 int btrfs_extent_readonly(struct btrfs_root
*root
, u64 bytenr
)
3339 struct btrfs_block_group_cache
*block_group
;
3342 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
3343 if (!block_group
|| block_group
->ro
)
3346 btrfs_put_block_group(block_group
);
3350 static const char *alloc_name(u64 flags
)
3353 case BTRFS_BLOCK_GROUP_METADATA
|BTRFS_BLOCK_GROUP_DATA
:
3355 case BTRFS_BLOCK_GROUP_METADATA
:
3357 case BTRFS_BLOCK_GROUP_DATA
:
3359 case BTRFS_BLOCK_GROUP_SYSTEM
:
3363 return "invalid-combination";
3367 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
3368 u64 total_bytes
, u64 bytes_used
,
3369 struct btrfs_space_info
**space_info
)
3371 struct btrfs_space_info
*found
;
3376 if (flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3377 BTRFS_BLOCK_GROUP_RAID10
))
3382 found
= __find_space_info(info
, flags
);
3384 spin_lock(&found
->lock
);
3385 found
->total_bytes
+= total_bytes
;
3386 found
->disk_total
+= total_bytes
* factor
;
3387 found
->bytes_used
+= bytes_used
;
3388 found
->disk_used
+= bytes_used
* factor
;
3390 spin_unlock(&found
->lock
);
3391 *space_info
= found
;
3394 found
= kzalloc(sizeof(*found
), GFP_NOFS
);
3398 ret
= percpu_counter_init(&found
->total_bytes_pinned
, 0);
3404 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++) {
3405 INIT_LIST_HEAD(&found
->block_groups
[i
]);
3406 kobject_init(&found
->block_group_kobjs
[i
], &btrfs_raid_ktype
);
3408 init_rwsem(&found
->groups_sem
);
3409 spin_lock_init(&found
->lock
);
3410 found
->flags
= flags
& BTRFS_BLOCK_GROUP_TYPE_MASK
;
3411 found
->total_bytes
= total_bytes
;
3412 found
->disk_total
= total_bytes
* factor
;
3413 found
->bytes_used
= bytes_used
;
3414 found
->disk_used
= bytes_used
* factor
;
3415 found
->bytes_pinned
= 0;
3416 found
->bytes_reserved
= 0;
3417 found
->bytes_readonly
= 0;
3418 found
->bytes_may_use
= 0;
3420 found
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3421 found
->chunk_alloc
= 0;
3423 init_waitqueue_head(&found
->wait
);
3425 ret
= kobject_init_and_add(&found
->kobj
, &space_info_ktype
,
3426 info
->space_info_kobj
, "%s",
3427 alloc_name(found
->flags
));
3433 *space_info
= found
;
3434 list_add_rcu(&found
->list
, &info
->space_info
);
3435 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3436 info
->data_sinfo
= found
;
3441 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
3443 u64 extra_flags
= chunk_to_extended(flags
) &
3444 BTRFS_EXTENDED_PROFILE_MASK
;
3446 write_seqlock(&fs_info
->profiles_lock
);
3447 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3448 fs_info
->avail_data_alloc_bits
|= extra_flags
;
3449 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3450 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
3451 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3452 fs_info
->avail_system_alloc_bits
|= extra_flags
;
3453 write_sequnlock(&fs_info
->profiles_lock
);
3457 * returns target flags in extended format or 0 if restripe for this
3458 * chunk_type is not in progress
3460 * should be called with either volume_mutex or balance_lock held
3462 static u64
get_restripe_target(struct btrfs_fs_info
*fs_info
, u64 flags
)
3464 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3470 if (flags
& BTRFS_BLOCK_GROUP_DATA
&&
3471 bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3472 target
= BTRFS_BLOCK_GROUP_DATA
| bctl
->data
.target
;
3473 } else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
&&
3474 bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3475 target
= BTRFS_BLOCK_GROUP_SYSTEM
| bctl
->sys
.target
;
3476 } else if (flags
& BTRFS_BLOCK_GROUP_METADATA
&&
3477 bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3478 target
= BTRFS_BLOCK_GROUP_METADATA
| bctl
->meta
.target
;
3485 * @flags: available profiles in extended format (see ctree.h)
3487 * Returns reduced profile in chunk format. If profile changing is in
3488 * progress (either running or paused) picks the target profile (if it's
3489 * already available), otherwise falls back to plain reducing.
3491 static u64
btrfs_reduce_alloc_profile(struct btrfs_root
*root
, u64 flags
)
3494 * we add in the count of missing devices because we want
3495 * to make sure that any RAID levels on a degraded FS
3496 * continue to be honored.
3498 u64 num_devices
= root
->fs_info
->fs_devices
->rw_devices
+
3499 root
->fs_info
->fs_devices
->missing_devices
;
3504 * see if restripe for this chunk_type is in progress, if so
3505 * try to reduce to the target profile
3507 spin_lock(&root
->fs_info
->balance_lock
);
3508 target
= get_restripe_target(root
->fs_info
, flags
);
3510 /* pick target profile only if it's already available */
3511 if ((flags
& target
) & BTRFS_EXTENDED_PROFILE_MASK
) {
3512 spin_unlock(&root
->fs_info
->balance_lock
);
3513 return extended_to_chunk(target
);
3516 spin_unlock(&root
->fs_info
->balance_lock
);
3518 /* First, mask out the RAID levels which aren't possible */
3519 if (num_devices
== 1)
3520 flags
&= ~(BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID0
|
3521 BTRFS_BLOCK_GROUP_RAID5
);
3522 if (num_devices
< 3)
3523 flags
&= ~BTRFS_BLOCK_GROUP_RAID6
;
3524 if (num_devices
< 4)
3525 flags
&= ~BTRFS_BLOCK_GROUP_RAID10
;
3527 tmp
= flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID0
|
3528 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID5
|
3529 BTRFS_BLOCK_GROUP_RAID6
| BTRFS_BLOCK_GROUP_RAID10
);
3532 if (tmp
& BTRFS_BLOCK_GROUP_RAID6
)
3533 tmp
= BTRFS_BLOCK_GROUP_RAID6
;
3534 else if (tmp
& BTRFS_BLOCK_GROUP_RAID5
)
3535 tmp
= BTRFS_BLOCK_GROUP_RAID5
;
3536 else if (tmp
& BTRFS_BLOCK_GROUP_RAID10
)
3537 tmp
= BTRFS_BLOCK_GROUP_RAID10
;
3538 else if (tmp
& BTRFS_BLOCK_GROUP_RAID1
)
3539 tmp
= BTRFS_BLOCK_GROUP_RAID1
;
3540 else if (tmp
& BTRFS_BLOCK_GROUP_RAID0
)
3541 tmp
= BTRFS_BLOCK_GROUP_RAID0
;
3543 return extended_to_chunk(flags
| tmp
);
3546 static u64
get_alloc_profile(struct btrfs_root
*root
, u64 flags
)
3551 seq
= read_seqbegin(&root
->fs_info
->profiles_lock
);
3553 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3554 flags
|= root
->fs_info
->avail_data_alloc_bits
;
3555 else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3556 flags
|= root
->fs_info
->avail_system_alloc_bits
;
3557 else if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3558 flags
|= root
->fs_info
->avail_metadata_alloc_bits
;
3559 } while (read_seqretry(&root
->fs_info
->profiles_lock
, seq
));
3561 return btrfs_reduce_alloc_profile(root
, flags
);
3564 u64
btrfs_get_alloc_profile(struct btrfs_root
*root
, int data
)
3570 flags
= BTRFS_BLOCK_GROUP_DATA
;
3571 else if (root
== root
->fs_info
->chunk_root
)
3572 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
3574 flags
= BTRFS_BLOCK_GROUP_METADATA
;
3576 ret
= get_alloc_profile(root
, flags
);
3581 * This will check the space that the inode allocates from to make sure we have
3582 * enough space for bytes.
3584 int btrfs_check_data_free_space(struct inode
*inode
, u64 bytes
)
3586 struct btrfs_space_info
*data_sinfo
;
3587 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3588 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3590 int ret
= 0, committed
= 0, alloc_chunk
= 1;
3592 /* make sure bytes are sectorsize aligned */
3593 bytes
= ALIGN(bytes
, root
->sectorsize
);
3595 if (btrfs_is_free_space_inode(inode
)) {
3597 ASSERT(current
->journal_info
);
3600 data_sinfo
= fs_info
->data_sinfo
;
3605 /* make sure we have enough space to handle the data first */
3606 spin_lock(&data_sinfo
->lock
);
3607 used
= data_sinfo
->bytes_used
+ data_sinfo
->bytes_reserved
+
3608 data_sinfo
->bytes_pinned
+ data_sinfo
->bytes_readonly
+
3609 data_sinfo
->bytes_may_use
;
3611 if (used
+ bytes
> data_sinfo
->total_bytes
) {
3612 struct btrfs_trans_handle
*trans
;
3615 * if we don't have enough free bytes in this space then we need
3616 * to alloc a new chunk.
3618 if (!data_sinfo
->full
&& alloc_chunk
) {
3621 data_sinfo
->force_alloc
= CHUNK_ALLOC_FORCE
;
3622 spin_unlock(&data_sinfo
->lock
);
3624 alloc_target
= btrfs_get_alloc_profile(root
, 1);
3626 * It is ugly that we don't call nolock join
3627 * transaction for the free space inode case here.
3628 * But it is safe because we only do the data space
3629 * reservation for the free space cache in the
3630 * transaction context, the common join transaction
3631 * just increase the counter of the current transaction
3632 * handler, doesn't try to acquire the trans_lock of
3635 trans
= btrfs_join_transaction(root
);
3637 return PTR_ERR(trans
);
3639 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
3641 CHUNK_ALLOC_NO_FORCE
);
3642 btrfs_end_transaction(trans
, root
);
3651 data_sinfo
= fs_info
->data_sinfo
;
3657 * If we don't have enough pinned space to deal with this
3658 * allocation don't bother committing the transaction.
3660 if (percpu_counter_compare(&data_sinfo
->total_bytes_pinned
,
3663 spin_unlock(&data_sinfo
->lock
);
3665 /* commit the current transaction and try again */
3668 !atomic_read(&root
->fs_info
->open_ioctl_trans
)) {
3671 trans
= btrfs_join_transaction(root
);
3673 return PTR_ERR(trans
);
3674 ret
= btrfs_commit_transaction(trans
, root
);
3680 trace_btrfs_space_reservation(root
->fs_info
,
3681 "space_info:enospc",
3682 data_sinfo
->flags
, bytes
, 1);
3685 data_sinfo
->bytes_may_use
+= bytes
;
3686 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
3687 data_sinfo
->flags
, bytes
, 1);
3688 spin_unlock(&data_sinfo
->lock
);
3694 * Called if we need to clear a data reservation for this inode.
3696 void btrfs_free_reserved_data_space(struct inode
*inode
, u64 bytes
)
3698 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3699 struct btrfs_space_info
*data_sinfo
;
3701 /* make sure bytes are sectorsize aligned */
3702 bytes
= ALIGN(bytes
, root
->sectorsize
);
3704 data_sinfo
= root
->fs_info
->data_sinfo
;
3705 spin_lock(&data_sinfo
->lock
);
3706 WARN_ON(data_sinfo
->bytes_may_use
< bytes
);
3707 data_sinfo
->bytes_may_use
-= bytes
;
3708 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
3709 data_sinfo
->flags
, bytes
, 0);
3710 spin_unlock(&data_sinfo
->lock
);
3713 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
3715 struct list_head
*head
= &info
->space_info
;
3716 struct btrfs_space_info
*found
;
3719 list_for_each_entry_rcu(found
, head
, list
) {
3720 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
3721 found
->force_alloc
= CHUNK_ALLOC_FORCE
;
3726 static inline u64
calc_global_rsv_need_space(struct btrfs_block_rsv
*global
)
3728 return (global
->size
<< 1);
3731 static int should_alloc_chunk(struct btrfs_root
*root
,
3732 struct btrfs_space_info
*sinfo
, int force
)
3734 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
3735 u64 num_bytes
= sinfo
->total_bytes
- sinfo
->bytes_readonly
;
3736 u64 num_allocated
= sinfo
->bytes_used
+ sinfo
->bytes_reserved
;
3739 if (force
== CHUNK_ALLOC_FORCE
)
3743 * We need to take into account the global rsv because for all intents
3744 * and purposes it's used space. Don't worry about locking the
3745 * global_rsv, it doesn't change except when the transaction commits.
3747 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
3748 num_allocated
+= calc_global_rsv_need_space(global_rsv
);
3751 * in limited mode, we want to have some free space up to
3752 * about 1% of the FS size.
3754 if (force
== CHUNK_ALLOC_LIMITED
) {
3755 thresh
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
3756 thresh
= max_t(u64
, 64 * 1024 * 1024,
3757 div_factor_fine(thresh
, 1));
3759 if (num_bytes
- num_allocated
< thresh
)
3763 if (num_allocated
+ 2 * 1024 * 1024 < div_factor(num_bytes
, 8))
3768 static u64
get_system_chunk_thresh(struct btrfs_root
*root
, u64 type
)
3772 if (type
& (BTRFS_BLOCK_GROUP_RAID10
|
3773 BTRFS_BLOCK_GROUP_RAID0
|
3774 BTRFS_BLOCK_GROUP_RAID5
|
3775 BTRFS_BLOCK_GROUP_RAID6
))
3776 num_dev
= root
->fs_info
->fs_devices
->rw_devices
;
3777 else if (type
& BTRFS_BLOCK_GROUP_RAID1
)
3780 num_dev
= 1; /* DUP or single */
3782 /* metadata for updaing devices and chunk tree */
3783 return btrfs_calc_trans_metadata_size(root
, num_dev
+ 1);
3786 static void check_system_chunk(struct btrfs_trans_handle
*trans
,
3787 struct btrfs_root
*root
, u64 type
)
3789 struct btrfs_space_info
*info
;
3793 info
= __find_space_info(root
->fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
3794 spin_lock(&info
->lock
);
3795 left
= info
->total_bytes
- info
->bytes_used
- info
->bytes_pinned
-
3796 info
->bytes_reserved
- info
->bytes_readonly
;
3797 spin_unlock(&info
->lock
);
3799 thresh
= get_system_chunk_thresh(root
, type
);
3800 if (left
< thresh
&& btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
3801 btrfs_info(root
->fs_info
, "left=%llu, need=%llu, flags=%llu",
3802 left
, thresh
, type
);
3803 dump_space_info(info
, 0, 0);
3806 if (left
< thresh
) {
3809 flags
= btrfs_get_alloc_profile(root
->fs_info
->chunk_root
, 0);
3810 btrfs_alloc_chunk(trans
, root
, flags
);
3814 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
3815 struct btrfs_root
*extent_root
, u64 flags
, int force
)
3817 struct btrfs_space_info
*space_info
;
3818 struct btrfs_fs_info
*fs_info
= extent_root
->fs_info
;
3819 int wait_for_alloc
= 0;
3822 /* Don't re-enter if we're already allocating a chunk */
3823 if (trans
->allocating_chunk
)
3826 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
3828 ret
= update_space_info(extent_root
->fs_info
, flags
,
3830 BUG_ON(ret
); /* -ENOMEM */
3832 BUG_ON(!space_info
); /* Logic error */
3835 spin_lock(&space_info
->lock
);
3836 if (force
< space_info
->force_alloc
)
3837 force
= space_info
->force_alloc
;
3838 if (space_info
->full
) {
3839 if (should_alloc_chunk(extent_root
, space_info
, force
))
3843 spin_unlock(&space_info
->lock
);
3847 if (!should_alloc_chunk(extent_root
, space_info
, force
)) {
3848 spin_unlock(&space_info
->lock
);
3850 } else if (space_info
->chunk_alloc
) {
3853 space_info
->chunk_alloc
= 1;
3856 spin_unlock(&space_info
->lock
);
3858 mutex_lock(&fs_info
->chunk_mutex
);
3861 * The chunk_mutex is held throughout the entirety of a chunk
3862 * allocation, so once we've acquired the chunk_mutex we know that the
3863 * other guy is done and we need to recheck and see if we should
3866 if (wait_for_alloc
) {
3867 mutex_unlock(&fs_info
->chunk_mutex
);
3872 trans
->allocating_chunk
= true;
3875 * If we have mixed data/metadata chunks we want to make sure we keep
3876 * allocating mixed chunks instead of individual chunks.
3878 if (btrfs_mixed_space_info(space_info
))
3879 flags
|= (BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
);
3882 * if we're doing a data chunk, go ahead and make sure that
3883 * we keep a reasonable number of metadata chunks allocated in the
3886 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
3887 fs_info
->data_chunk_allocations
++;
3888 if (!(fs_info
->data_chunk_allocations
%
3889 fs_info
->metadata_ratio
))
3890 force_metadata_allocation(fs_info
);
3894 * Check if we have enough space in SYSTEM chunk because we may need
3895 * to update devices.
3897 check_system_chunk(trans
, extent_root
, flags
);
3899 ret
= btrfs_alloc_chunk(trans
, extent_root
, flags
);
3900 trans
->allocating_chunk
= false;
3902 spin_lock(&space_info
->lock
);
3903 if (ret
< 0 && ret
!= -ENOSPC
)
3906 space_info
->full
= 1;
3910 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3912 space_info
->chunk_alloc
= 0;
3913 spin_unlock(&space_info
->lock
);
3914 mutex_unlock(&fs_info
->chunk_mutex
);
3918 static int can_overcommit(struct btrfs_root
*root
,
3919 struct btrfs_space_info
*space_info
, u64 bytes
,
3920 enum btrfs_reserve_flush_enum flush
)
3922 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
3923 u64 profile
= btrfs_get_alloc_profile(root
, 0);
3928 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
3929 space_info
->bytes_pinned
+ space_info
->bytes_readonly
;
3932 * We only want to allow over committing if we have lots of actual space
3933 * free, but if we don't have enough space to handle the global reserve
3934 * space then we could end up having a real enospc problem when trying
3935 * to allocate a chunk or some other such important allocation.
3937 spin_lock(&global_rsv
->lock
);
3938 space_size
= calc_global_rsv_need_space(global_rsv
);
3939 spin_unlock(&global_rsv
->lock
);
3940 if (used
+ space_size
>= space_info
->total_bytes
)
3943 used
+= space_info
->bytes_may_use
;
3945 spin_lock(&root
->fs_info
->free_chunk_lock
);
3946 avail
= root
->fs_info
->free_chunk_space
;
3947 spin_unlock(&root
->fs_info
->free_chunk_lock
);
3950 * If we have dup, raid1 or raid10 then only half of the free
3951 * space is actually useable. For raid56, the space info used
3952 * doesn't include the parity drive, so we don't have to
3955 if (profile
& (BTRFS_BLOCK_GROUP_DUP
|
3956 BTRFS_BLOCK_GROUP_RAID1
|
3957 BTRFS_BLOCK_GROUP_RAID10
))
3961 * If we aren't flushing all things, let us overcommit up to
3962 * 1/2th of the space. If we can flush, don't let us overcommit
3963 * too much, let it overcommit up to 1/8 of the space.
3965 if (flush
== BTRFS_RESERVE_FLUSH_ALL
)
3970 if (used
+ bytes
< space_info
->total_bytes
+ avail
)
3975 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root
*root
,
3976 unsigned long nr_pages
, int nr_items
)
3978 struct super_block
*sb
= root
->fs_info
->sb
;
3980 if (down_read_trylock(&sb
->s_umount
)) {
3981 writeback_inodes_sb_nr(sb
, nr_pages
, WB_REASON_FS_FREE_SPACE
);
3982 up_read(&sb
->s_umount
);
3985 * We needn't worry the filesystem going from r/w to r/o though
3986 * we don't acquire ->s_umount mutex, because the filesystem
3987 * should guarantee the delalloc inodes list be empty after
3988 * the filesystem is readonly(all dirty pages are written to
3991 btrfs_start_delalloc_roots(root
->fs_info
, 0, nr_items
);
3992 if (!current
->journal_info
)
3993 btrfs_wait_ordered_roots(root
->fs_info
, nr_items
);
3997 static inline int calc_reclaim_items_nr(struct btrfs_root
*root
, u64 to_reclaim
)
4002 bytes
= btrfs_calc_trans_metadata_size(root
, 1);
4003 nr
= (int)div64_u64(to_reclaim
, bytes
);
4009 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4012 * shrink metadata reservation for delalloc
4014 static void shrink_delalloc(struct btrfs_root
*root
, u64 to_reclaim
, u64 orig
,
4017 struct btrfs_block_rsv
*block_rsv
;
4018 struct btrfs_space_info
*space_info
;
4019 struct btrfs_trans_handle
*trans
;
4023 unsigned long nr_pages
;
4026 enum btrfs_reserve_flush_enum flush
;
4028 /* Calc the number of the pages we need flush for space reservation */
4029 items
= calc_reclaim_items_nr(root
, to_reclaim
);
4030 to_reclaim
= items
* EXTENT_SIZE_PER_ITEM
;
4032 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4033 block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
4034 space_info
= block_rsv
->space_info
;
4036 delalloc_bytes
= percpu_counter_sum_positive(
4037 &root
->fs_info
->delalloc_bytes
);
4038 if (delalloc_bytes
== 0) {
4042 btrfs_wait_ordered_roots(root
->fs_info
, items
);
4047 while (delalloc_bytes
&& loops
< 3) {
4048 max_reclaim
= min(delalloc_bytes
, to_reclaim
);
4049 nr_pages
= max_reclaim
>> PAGE_CACHE_SHIFT
;
4050 btrfs_writeback_inodes_sb_nr(root
, nr_pages
, items
);
4052 * We need to wait for the async pages to actually start before
4055 max_reclaim
= atomic_read(&root
->fs_info
->async_delalloc_pages
);
4059 if (max_reclaim
<= nr_pages
)
4062 max_reclaim
-= nr_pages
;
4064 wait_event(root
->fs_info
->async_submit_wait
,
4065 atomic_read(&root
->fs_info
->async_delalloc_pages
) <=
4069 flush
= BTRFS_RESERVE_FLUSH_ALL
;
4071 flush
= BTRFS_RESERVE_NO_FLUSH
;
4072 spin_lock(&space_info
->lock
);
4073 if (can_overcommit(root
, space_info
, orig
, flush
)) {
4074 spin_unlock(&space_info
->lock
);
4077 spin_unlock(&space_info
->lock
);
4080 if (wait_ordered
&& !trans
) {
4081 btrfs_wait_ordered_roots(root
->fs_info
, items
);
4083 time_left
= schedule_timeout_killable(1);
4087 delalloc_bytes
= percpu_counter_sum_positive(
4088 &root
->fs_info
->delalloc_bytes
);
4093 * maybe_commit_transaction - possibly commit the transaction if its ok to
4094 * @root - the root we're allocating for
4095 * @bytes - the number of bytes we want to reserve
4096 * @force - force the commit
4098 * This will check to make sure that committing the transaction will actually
4099 * get us somewhere and then commit the transaction if it does. Otherwise it
4100 * will return -ENOSPC.
4102 static int may_commit_transaction(struct btrfs_root
*root
,
4103 struct btrfs_space_info
*space_info
,
4104 u64 bytes
, int force
)
4106 struct btrfs_block_rsv
*delayed_rsv
= &root
->fs_info
->delayed_block_rsv
;
4107 struct btrfs_trans_handle
*trans
;
4109 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4116 /* See if there is enough pinned space to make this reservation */
4117 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4122 * See if there is some space in the delayed insertion reservation for
4125 if (space_info
!= delayed_rsv
->space_info
)
4128 spin_lock(&delayed_rsv
->lock
);
4129 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4130 bytes
- delayed_rsv
->size
) >= 0) {
4131 spin_unlock(&delayed_rsv
->lock
);
4134 spin_unlock(&delayed_rsv
->lock
);
4137 trans
= btrfs_join_transaction(root
);
4141 return btrfs_commit_transaction(trans
, root
);
4145 FLUSH_DELAYED_ITEMS_NR
= 1,
4146 FLUSH_DELAYED_ITEMS
= 2,
4148 FLUSH_DELALLOC_WAIT
= 4,
4153 static int flush_space(struct btrfs_root
*root
,
4154 struct btrfs_space_info
*space_info
, u64 num_bytes
,
4155 u64 orig_bytes
, int state
)
4157 struct btrfs_trans_handle
*trans
;
4162 case FLUSH_DELAYED_ITEMS_NR
:
4163 case FLUSH_DELAYED_ITEMS
:
4164 if (state
== FLUSH_DELAYED_ITEMS_NR
)
4165 nr
= calc_reclaim_items_nr(root
, num_bytes
) * 2;
4169 trans
= btrfs_join_transaction(root
);
4170 if (IS_ERR(trans
)) {
4171 ret
= PTR_ERR(trans
);
4174 ret
= btrfs_run_delayed_items_nr(trans
, root
, nr
);
4175 btrfs_end_transaction(trans
, root
);
4177 case FLUSH_DELALLOC
:
4178 case FLUSH_DELALLOC_WAIT
:
4179 shrink_delalloc(root
, num_bytes
* 2, orig_bytes
,
4180 state
== FLUSH_DELALLOC_WAIT
);
4183 trans
= btrfs_join_transaction(root
);
4184 if (IS_ERR(trans
)) {
4185 ret
= PTR_ERR(trans
);
4188 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
4189 btrfs_get_alloc_profile(root
, 0),
4190 CHUNK_ALLOC_NO_FORCE
);
4191 btrfs_end_transaction(trans
, root
);
4196 ret
= may_commit_transaction(root
, space_info
, orig_bytes
, 0);
4206 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4207 * @root - the root we're allocating for
4208 * @block_rsv - the block_rsv we're allocating for
4209 * @orig_bytes - the number of bytes we want
4210 * @flush - whether or not we can flush to make our reservation
4212 * This will reserve orgi_bytes number of bytes from the space info associated
4213 * with the block_rsv. If there is not enough space it will make an attempt to
4214 * flush out space to make room. It will do this by flushing delalloc if
4215 * possible or committing the transaction. If flush is 0 then no attempts to
4216 * regain reservations will be made and this will fail if there is not enough
4219 static int reserve_metadata_bytes(struct btrfs_root
*root
,
4220 struct btrfs_block_rsv
*block_rsv
,
4222 enum btrfs_reserve_flush_enum flush
)
4224 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
4226 u64 num_bytes
= orig_bytes
;
4227 int flush_state
= FLUSH_DELAYED_ITEMS_NR
;
4229 bool flushing
= false;
4233 spin_lock(&space_info
->lock
);
4235 * We only want to wait if somebody other than us is flushing and we
4236 * are actually allowed to flush all things.
4238 while (flush
== BTRFS_RESERVE_FLUSH_ALL
&& !flushing
&&
4239 space_info
->flush
) {
4240 spin_unlock(&space_info
->lock
);
4242 * If we have a trans handle we can't wait because the flusher
4243 * may have to commit the transaction, which would mean we would
4244 * deadlock since we are waiting for the flusher to finish, but
4245 * hold the current transaction open.
4247 if (current
->journal_info
)
4249 ret
= wait_event_killable(space_info
->wait
, !space_info
->flush
);
4250 /* Must have been killed, return */
4254 spin_lock(&space_info
->lock
);
4258 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4259 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4260 space_info
->bytes_may_use
;
4263 * The idea here is that we've not already over-reserved the block group
4264 * then we can go ahead and save our reservation first and then start
4265 * flushing if we need to. Otherwise if we've already overcommitted
4266 * lets start flushing stuff first and then come back and try to make
4269 if (used
<= space_info
->total_bytes
) {
4270 if (used
+ orig_bytes
<= space_info
->total_bytes
) {
4271 space_info
->bytes_may_use
+= orig_bytes
;
4272 trace_btrfs_space_reservation(root
->fs_info
,
4273 "space_info", space_info
->flags
, orig_bytes
, 1);
4277 * Ok set num_bytes to orig_bytes since we aren't
4278 * overocmmitted, this way we only try and reclaim what
4281 num_bytes
= orig_bytes
;
4285 * Ok we're over committed, set num_bytes to the overcommitted
4286 * amount plus the amount of bytes that we need for this
4289 num_bytes
= used
- space_info
->total_bytes
+
4293 if (ret
&& can_overcommit(root
, space_info
, orig_bytes
, flush
)) {
4294 space_info
->bytes_may_use
+= orig_bytes
;
4295 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
4296 space_info
->flags
, orig_bytes
,
4302 * Couldn't make our reservation, save our place so while we're trying
4303 * to reclaim space we can actually use it instead of somebody else
4304 * stealing it from us.
4306 * We make the other tasks wait for the flush only when we can flush
4309 if (ret
&& flush
!= BTRFS_RESERVE_NO_FLUSH
) {
4311 space_info
->flush
= 1;
4314 spin_unlock(&space_info
->lock
);
4316 if (!ret
|| flush
== BTRFS_RESERVE_NO_FLUSH
)
4319 ret
= flush_space(root
, space_info
, num_bytes
, orig_bytes
,
4324 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4325 * would happen. So skip delalloc flush.
4327 if (flush
== BTRFS_RESERVE_FLUSH_LIMIT
&&
4328 (flush_state
== FLUSH_DELALLOC
||
4329 flush_state
== FLUSH_DELALLOC_WAIT
))
4330 flush_state
= ALLOC_CHUNK
;
4334 else if (flush
== BTRFS_RESERVE_FLUSH_LIMIT
&&
4335 flush_state
< COMMIT_TRANS
)
4337 else if (flush
== BTRFS_RESERVE_FLUSH_ALL
&&
4338 flush_state
<= COMMIT_TRANS
)
4342 if (ret
== -ENOSPC
&&
4343 unlikely(root
->orphan_cleanup_state
== ORPHAN_CLEANUP_STARTED
)) {
4344 struct btrfs_block_rsv
*global_rsv
=
4345 &root
->fs_info
->global_block_rsv
;
4347 if (block_rsv
!= global_rsv
&&
4348 !block_rsv_use_bytes(global_rsv
, orig_bytes
))
4352 trace_btrfs_space_reservation(root
->fs_info
,
4353 "space_info:enospc",
4354 space_info
->flags
, orig_bytes
, 1);
4356 spin_lock(&space_info
->lock
);
4357 space_info
->flush
= 0;
4358 wake_up_all(&space_info
->wait
);
4359 spin_unlock(&space_info
->lock
);
4364 static struct btrfs_block_rsv
*get_block_rsv(
4365 const struct btrfs_trans_handle
*trans
,
4366 const struct btrfs_root
*root
)
4368 struct btrfs_block_rsv
*block_rsv
= NULL
;
4371 block_rsv
= trans
->block_rsv
;
4373 if (root
== root
->fs_info
->csum_root
&& trans
->adding_csums
)
4374 block_rsv
= trans
->block_rsv
;
4376 if (root
== root
->fs_info
->uuid_root
)
4377 block_rsv
= trans
->block_rsv
;
4380 block_rsv
= root
->block_rsv
;
4383 block_rsv
= &root
->fs_info
->empty_block_rsv
;
4388 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
4392 spin_lock(&block_rsv
->lock
);
4393 if (block_rsv
->reserved
>= num_bytes
) {
4394 block_rsv
->reserved
-= num_bytes
;
4395 if (block_rsv
->reserved
< block_rsv
->size
)
4396 block_rsv
->full
= 0;
4399 spin_unlock(&block_rsv
->lock
);
4403 static void block_rsv_add_bytes(struct btrfs_block_rsv
*block_rsv
,
4404 u64 num_bytes
, int update_size
)
4406 spin_lock(&block_rsv
->lock
);
4407 block_rsv
->reserved
+= num_bytes
;
4409 block_rsv
->size
+= num_bytes
;
4410 else if (block_rsv
->reserved
>= block_rsv
->size
)
4411 block_rsv
->full
= 1;
4412 spin_unlock(&block_rsv
->lock
);
4415 int btrfs_cond_migrate_bytes(struct btrfs_fs_info
*fs_info
,
4416 struct btrfs_block_rsv
*dest
, u64 num_bytes
,
4419 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
4422 if (global_rsv
->space_info
!= dest
->space_info
)
4425 spin_lock(&global_rsv
->lock
);
4426 min_bytes
= div_factor(global_rsv
->size
, min_factor
);
4427 if (global_rsv
->reserved
< min_bytes
+ num_bytes
) {
4428 spin_unlock(&global_rsv
->lock
);
4431 global_rsv
->reserved
-= num_bytes
;
4432 if (global_rsv
->reserved
< global_rsv
->size
)
4433 global_rsv
->full
= 0;
4434 spin_unlock(&global_rsv
->lock
);
4436 block_rsv_add_bytes(dest
, num_bytes
, 1);
4440 static void block_rsv_release_bytes(struct btrfs_fs_info
*fs_info
,
4441 struct btrfs_block_rsv
*block_rsv
,
4442 struct btrfs_block_rsv
*dest
, u64 num_bytes
)
4444 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
4446 spin_lock(&block_rsv
->lock
);
4447 if (num_bytes
== (u64
)-1)
4448 num_bytes
= block_rsv
->size
;
4449 block_rsv
->size
-= num_bytes
;
4450 if (block_rsv
->reserved
>= block_rsv
->size
) {
4451 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
4452 block_rsv
->reserved
= block_rsv
->size
;
4453 block_rsv
->full
= 1;
4457 spin_unlock(&block_rsv
->lock
);
4459 if (num_bytes
> 0) {
4461 spin_lock(&dest
->lock
);
4465 bytes_to_add
= dest
->size
- dest
->reserved
;
4466 bytes_to_add
= min(num_bytes
, bytes_to_add
);
4467 dest
->reserved
+= bytes_to_add
;
4468 if (dest
->reserved
>= dest
->size
)
4470 num_bytes
-= bytes_to_add
;
4472 spin_unlock(&dest
->lock
);
4475 spin_lock(&space_info
->lock
);
4476 space_info
->bytes_may_use
-= num_bytes
;
4477 trace_btrfs_space_reservation(fs_info
, "space_info",
4478 space_info
->flags
, num_bytes
, 0);
4479 spin_unlock(&space_info
->lock
);
4484 static int block_rsv_migrate_bytes(struct btrfs_block_rsv
*src
,
4485 struct btrfs_block_rsv
*dst
, u64 num_bytes
)
4489 ret
= block_rsv_use_bytes(src
, num_bytes
);
4493 block_rsv_add_bytes(dst
, num_bytes
, 1);
4497 void btrfs_init_block_rsv(struct btrfs_block_rsv
*rsv
, unsigned short type
)
4499 memset(rsv
, 0, sizeof(*rsv
));
4500 spin_lock_init(&rsv
->lock
);
4504 struct btrfs_block_rsv
*btrfs_alloc_block_rsv(struct btrfs_root
*root
,
4505 unsigned short type
)
4507 struct btrfs_block_rsv
*block_rsv
;
4508 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4510 block_rsv
= kmalloc(sizeof(*block_rsv
), GFP_NOFS
);
4514 btrfs_init_block_rsv(block_rsv
, type
);
4515 block_rsv
->space_info
= __find_space_info(fs_info
,
4516 BTRFS_BLOCK_GROUP_METADATA
);
4520 void btrfs_free_block_rsv(struct btrfs_root
*root
,
4521 struct btrfs_block_rsv
*rsv
)
4525 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
4529 int btrfs_block_rsv_add(struct btrfs_root
*root
,
4530 struct btrfs_block_rsv
*block_rsv
, u64 num_bytes
,
4531 enum btrfs_reserve_flush_enum flush
)
4538 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
4540 block_rsv_add_bytes(block_rsv
, num_bytes
, 1);
4547 int btrfs_block_rsv_check(struct btrfs_root
*root
,
4548 struct btrfs_block_rsv
*block_rsv
, int min_factor
)
4556 spin_lock(&block_rsv
->lock
);
4557 num_bytes
= div_factor(block_rsv
->size
, min_factor
);
4558 if (block_rsv
->reserved
>= num_bytes
)
4560 spin_unlock(&block_rsv
->lock
);
4565 int btrfs_block_rsv_refill(struct btrfs_root
*root
,
4566 struct btrfs_block_rsv
*block_rsv
, u64 min_reserved
,
4567 enum btrfs_reserve_flush_enum flush
)
4575 spin_lock(&block_rsv
->lock
);
4576 num_bytes
= min_reserved
;
4577 if (block_rsv
->reserved
>= num_bytes
)
4580 num_bytes
-= block_rsv
->reserved
;
4581 spin_unlock(&block_rsv
->lock
);
4586 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
4588 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
4595 int btrfs_block_rsv_migrate(struct btrfs_block_rsv
*src_rsv
,
4596 struct btrfs_block_rsv
*dst_rsv
,
4599 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
4602 void btrfs_block_rsv_release(struct btrfs_root
*root
,
4603 struct btrfs_block_rsv
*block_rsv
,
4606 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
4607 if (global_rsv
== block_rsv
||
4608 block_rsv
->space_info
!= global_rsv
->space_info
)
4610 block_rsv_release_bytes(root
->fs_info
, block_rsv
, global_rsv
,
4615 * helper to calculate size of global block reservation.
4616 * the desired value is sum of space used by extent tree,
4617 * checksum tree and root tree
4619 static u64
calc_global_metadata_size(struct btrfs_fs_info
*fs_info
)
4621 struct btrfs_space_info
*sinfo
;
4625 int csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
4627 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_DATA
);
4628 spin_lock(&sinfo
->lock
);
4629 data_used
= sinfo
->bytes_used
;
4630 spin_unlock(&sinfo
->lock
);
4632 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4633 spin_lock(&sinfo
->lock
);
4634 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_DATA
)
4636 meta_used
= sinfo
->bytes_used
;
4637 spin_unlock(&sinfo
->lock
);
4639 num_bytes
= (data_used
>> fs_info
->sb
->s_blocksize_bits
) *
4641 num_bytes
+= div64_u64(data_used
+ meta_used
, 50);
4643 if (num_bytes
* 3 > meta_used
)
4644 num_bytes
= div64_u64(meta_used
, 3);
4646 return ALIGN(num_bytes
, fs_info
->extent_root
->leafsize
<< 10);
4649 static void update_global_block_rsv(struct btrfs_fs_info
*fs_info
)
4651 struct btrfs_block_rsv
*block_rsv
= &fs_info
->global_block_rsv
;
4652 struct btrfs_space_info
*sinfo
= block_rsv
->space_info
;
4655 num_bytes
= calc_global_metadata_size(fs_info
);
4657 spin_lock(&sinfo
->lock
);
4658 spin_lock(&block_rsv
->lock
);
4660 block_rsv
->size
= min_t(u64
, num_bytes
, 512 * 1024 * 1024);
4662 num_bytes
= sinfo
->bytes_used
+ sinfo
->bytes_pinned
+
4663 sinfo
->bytes_reserved
+ sinfo
->bytes_readonly
+
4664 sinfo
->bytes_may_use
;
4666 if (sinfo
->total_bytes
> num_bytes
) {
4667 num_bytes
= sinfo
->total_bytes
- num_bytes
;
4668 block_rsv
->reserved
+= num_bytes
;
4669 sinfo
->bytes_may_use
+= num_bytes
;
4670 trace_btrfs_space_reservation(fs_info
, "space_info",
4671 sinfo
->flags
, num_bytes
, 1);
4674 if (block_rsv
->reserved
>= block_rsv
->size
) {
4675 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
4676 sinfo
->bytes_may_use
-= num_bytes
;
4677 trace_btrfs_space_reservation(fs_info
, "space_info",
4678 sinfo
->flags
, num_bytes
, 0);
4679 block_rsv
->reserved
= block_rsv
->size
;
4680 block_rsv
->full
= 1;
4683 spin_unlock(&block_rsv
->lock
);
4684 spin_unlock(&sinfo
->lock
);
4687 static void init_global_block_rsv(struct btrfs_fs_info
*fs_info
)
4689 struct btrfs_space_info
*space_info
;
4691 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
4692 fs_info
->chunk_block_rsv
.space_info
= space_info
;
4694 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4695 fs_info
->global_block_rsv
.space_info
= space_info
;
4696 fs_info
->delalloc_block_rsv
.space_info
= space_info
;
4697 fs_info
->trans_block_rsv
.space_info
= space_info
;
4698 fs_info
->empty_block_rsv
.space_info
= space_info
;
4699 fs_info
->delayed_block_rsv
.space_info
= space_info
;
4701 fs_info
->extent_root
->block_rsv
= &fs_info
->global_block_rsv
;
4702 fs_info
->csum_root
->block_rsv
= &fs_info
->global_block_rsv
;
4703 fs_info
->dev_root
->block_rsv
= &fs_info
->global_block_rsv
;
4704 fs_info
->tree_root
->block_rsv
= &fs_info
->global_block_rsv
;
4705 if (fs_info
->quota_root
)
4706 fs_info
->quota_root
->block_rsv
= &fs_info
->global_block_rsv
;
4707 fs_info
->chunk_root
->block_rsv
= &fs_info
->chunk_block_rsv
;
4709 update_global_block_rsv(fs_info
);
4712 static void release_global_block_rsv(struct btrfs_fs_info
*fs_info
)
4714 block_rsv_release_bytes(fs_info
, &fs_info
->global_block_rsv
, NULL
,
4716 WARN_ON(fs_info
->delalloc_block_rsv
.size
> 0);
4717 WARN_ON(fs_info
->delalloc_block_rsv
.reserved
> 0);
4718 WARN_ON(fs_info
->trans_block_rsv
.size
> 0);
4719 WARN_ON(fs_info
->trans_block_rsv
.reserved
> 0);
4720 WARN_ON(fs_info
->chunk_block_rsv
.size
> 0);
4721 WARN_ON(fs_info
->chunk_block_rsv
.reserved
> 0);
4722 WARN_ON(fs_info
->delayed_block_rsv
.size
> 0);
4723 WARN_ON(fs_info
->delayed_block_rsv
.reserved
> 0);
4726 void btrfs_trans_release_metadata(struct btrfs_trans_handle
*trans
,
4727 struct btrfs_root
*root
)
4729 if (!trans
->block_rsv
)
4732 if (!trans
->bytes_reserved
)
4735 trace_btrfs_space_reservation(root
->fs_info
, "transaction",
4736 trans
->transid
, trans
->bytes_reserved
, 0);
4737 btrfs_block_rsv_release(root
, trans
->block_rsv
, trans
->bytes_reserved
);
4738 trans
->bytes_reserved
= 0;
4741 /* Can only return 0 or -ENOSPC */
4742 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle
*trans
,
4743 struct inode
*inode
)
4745 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4746 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
4747 struct btrfs_block_rsv
*dst_rsv
= root
->orphan_block_rsv
;
4750 * We need to hold space in order to delete our orphan item once we've
4751 * added it, so this takes the reservation so we can release it later
4752 * when we are truly done with the orphan item.
4754 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
4755 trace_btrfs_space_reservation(root
->fs_info
, "orphan",
4756 btrfs_ino(inode
), num_bytes
, 1);
4757 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
4760 void btrfs_orphan_release_metadata(struct inode
*inode
)
4762 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4763 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
4764 trace_btrfs_space_reservation(root
->fs_info
, "orphan",
4765 btrfs_ino(inode
), num_bytes
, 0);
4766 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
, num_bytes
);
4770 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4771 * root: the root of the parent directory
4772 * rsv: block reservation
4773 * items: the number of items that we need do reservation
4774 * qgroup_reserved: used to return the reserved size in qgroup
4776 * This function is used to reserve the space for snapshot/subvolume
4777 * creation and deletion. Those operations are different with the
4778 * common file/directory operations, they change two fs/file trees
4779 * and root tree, the number of items that the qgroup reserves is
4780 * different with the free space reservation. So we can not use
4781 * the space reseravtion mechanism in start_transaction().
4783 int btrfs_subvolume_reserve_metadata(struct btrfs_root
*root
,
4784 struct btrfs_block_rsv
*rsv
,
4786 u64
*qgroup_reserved
,
4787 bool use_global_rsv
)
4791 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
4793 if (root
->fs_info
->quota_enabled
) {
4794 /* One for parent inode, two for dir entries */
4795 num_bytes
= 3 * root
->leafsize
;
4796 ret
= btrfs_qgroup_reserve(root
, num_bytes
);
4803 *qgroup_reserved
= num_bytes
;
4805 num_bytes
= btrfs_calc_trans_metadata_size(root
, items
);
4806 rsv
->space_info
= __find_space_info(root
->fs_info
,
4807 BTRFS_BLOCK_GROUP_METADATA
);
4808 ret
= btrfs_block_rsv_add(root
, rsv
, num_bytes
,
4809 BTRFS_RESERVE_FLUSH_ALL
);
4811 if (ret
== -ENOSPC
&& use_global_rsv
)
4812 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
, num_bytes
);
4815 if (*qgroup_reserved
)
4816 btrfs_qgroup_free(root
, *qgroup_reserved
);
4822 void btrfs_subvolume_release_metadata(struct btrfs_root
*root
,
4823 struct btrfs_block_rsv
*rsv
,
4824 u64 qgroup_reserved
)
4826 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
4827 if (qgroup_reserved
)
4828 btrfs_qgroup_free(root
, qgroup_reserved
);
4832 * drop_outstanding_extent - drop an outstanding extent
4833 * @inode: the inode we're dropping the extent for
4835 * This is called when we are freeing up an outstanding extent, either called
4836 * after an error or after an extent is written. This will return the number of
4837 * reserved extents that need to be freed. This must be called with
4838 * BTRFS_I(inode)->lock held.
4840 static unsigned drop_outstanding_extent(struct inode
*inode
)
4842 unsigned drop_inode_space
= 0;
4843 unsigned dropped_extents
= 0;
4845 BUG_ON(!BTRFS_I(inode
)->outstanding_extents
);
4846 BTRFS_I(inode
)->outstanding_extents
--;
4848 if (BTRFS_I(inode
)->outstanding_extents
== 0 &&
4849 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
4850 &BTRFS_I(inode
)->runtime_flags
))
4851 drop_inode_space
= 1;
4854 * If we have more or the same amount of outsanding extents than we have
4855 * reserved then we need to leave the reserved extents count alone.
4857 if (BTRFS_I(inode
)->outstanding_extents
>=
4858 BTRFS_I(inode
)->reserved_extents
)
4859 return drop_inode_space
;
4861 dropped_extents
= BTRFS_I(inode
)->reserved_extents
-
4862 BTRFS_I(inode
)->outstanding_extents
;
4863 BTRFS_I(inode
)->reserved_extents
-= dropped_extents
;
4864 return dropped_extents
+ drop_inode_space
;
4868 * calc_csum_metadata_size - return the amount of metada space that must be
4869 * reserved/free'd for the given bytes.
4870 * @inode: the inode we're manipulating
4871 * @num_bytes: the number of bytes in question
4872 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4874 * This adjusts the number of csum_bytes in the inode and then returns the
4875 * correct amount of metadata that must either be reserved or freed. We
4876 * calculate how many checksums we can fit into one leaf and then divide the
4877 * number of bytes that will need to be checksumed by this value to figure out
4878 * how many checksums will be required. If we are adding bytes then the number
4879 * may go up and we will return the number of additional bytes that must be
4880 * reserved. If it is going down we will return the number of bytes that must
4883 * This must be called with BTRFS_I(inode)->lock held.
4885 static u64
calc_csum_metadata_size(struct inode
*inode
, u64 num_bytes
,
4888 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4890 int num_csums_per_leaf
;
4894 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
&&
4895 BTRFS_I(inode
)->csum_bytes
== 0)
4898 old_csums
= (int)div64_u64(BTRFS_I(inode
)->csum_bytes
, root
->sectorsize
);
4900 BTRFS_I(inode
)->csum_bytes
+= num_bytes
;
4902 BTRFS_I(inode
)->csum_bytes
-= num_bytes
;
4903 csum_size
= BTRFS_LEAF_DATA_SIZE(root
) - sizeof(struct btrfs_item
);
4904 num_csums_per_leaf
= (int)div64_u64(csum_size
,
4905 sizeof(struct btrfs_csum_item
) +
4906 sizeof(struct btrfs_disk_key
));
4907 num_csums
= (int)div64_u64(BTRFS_I(inode
)->csum_bytes
, root
->sectorsize
);
4908 num_csums
= num_csums
+ num_csums_per_leaf
- 1;
4909 num_csums
= num_csums
/ num_csums_per_leaf
;
4911 old_csums
= old_csums
+ num_csums_per_leaf
- 1;
4912 old_csums
= old_csums
/ num_csums_per_leaf
;
4914 /* No change, no need to reserve more */
4915 if (old_csums
== num_csums
)
4919 return btrfs_calc_trans_metadata_size(root
,
4920 num_csums
- old_csums
);
4922 return btrfs_calc_trans_metadata_size(root
, old_csums
- num_csums
);
4925 int btrfs_delalloc_reserve_metadata(struct inode
*inode
, u64 num_bytes
)
4927 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4928 struct btrfs_block_rsv
*block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
4931 unsigned nr_extents
= 0;
4932 int extra_reserve
= 0;
4933 enum btrfs_reserve_flush_enum flush
= BTRFS_RESERVE_FLUSH_ALL
;
4935 bool delalloc_lock
= true;
4939 /* If we are a free space inode we need to not flush since we will be in
4940 * the middle of a transaction commit. We also don't need the delalloc
4941 * mutex since we won't race with anybody. We need this mostly to make
4942 * lockdep shut its filthy mouth.
4944 if (btrfs_is_free_space_inode(inode
)) {
4945 flush
= BTRFS_RESERVE_NO_FLUSH
;
4946 delalloc_lock
= false;
4949 if (flush
!= BTRFS_RESERVE_NO_FLUSH
&&
4950 btrfs_transaction_in_commit(root
->fs_info
))
4951 schedule_timeout(1);
4954 mutex_lock(&BTRFS_I(inode
)->delalloc_mutex
);
4956 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
4958 spin_lock(&BTRFS_I(inode
)->lock
);
4959 BTRFS_I(inode
)->outstanding_extents
++;
4961 if (BTRFS_I(inode
)->outstanding_extents
>
4962 BTRFS_I(inode
)->reserved_extents
)
4963 nr_extents
= BTRFS_I(inode
)->outstanding_extents
-
4964 BTRFS_I(inode
)->reserved_extents
;
4967 * Add an item to reserve for updating the inode when we complete the
4970 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
4971 &BTRFS_I(inode
)->runtime_flags
)) {
4976 to_reserve
= btrfs_calc_trans_metadata_size(root
, nr_extents
);
4977 to_reserve
+= calc_csum_metadata_size(inode
, num_bytes
, 1);
4978 csum_bytes
= BTRFS_I(inode
)->csum_bytes
;
4979 spin_unlock(&BTRFS_I(inode
)->lock
);
4981 if (root
->fs_info
->quota_enabled
) {
4982 ret
= btrfs_qgroup_reserve(root
, num_bytes
+
4983 nr_extents
* root
->leafsize
);
4988 ret
= reserve_metadata_bytes(root
, block_rsv
, to_reserve
, flush
);
4989 if (unlikely(ret
)) {
4990 if (root
->fs_info
->quota_enabled
)
4991 btrfs_qgroup_free(root
, num_bytes
+
4992 nr_extents
* root
->leafsize
);
4996 spin_lock(&BTRFS_I(inode
)->lock
);
4997 if (extra_reserve
) {
4998 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
4999 &BTRFS_I(inode
)->runtime_flags
);
5002 BTRFS_I(inode
)->reserved_extents
+= nr_extents
;
5003 spin_unlock(&BTRFS_I(inode
)->lock
);
5006 mutex_unlock(&BTRFS_I(inode
)->delalloc_mutex
);
5009 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5010 btrfs_ino(inode
), to_reserve
, 1);
5011 block_rsv_add_bytes(block_rsv
, to_reserve
, 1);
5016 spin_lock(&BTRFS_I(inode
)->lock
);
5017 dropped
= drop_outstanding_extent(inode
);
5019 * If the inodes csum_bytes is the same as the original
5020 * csum_bytes then we know we haven't raced with any free()ers
5021 * so we can just reduce our inodes csum bytes and carry on.
5023 if (BTRFS_I(inode
)->csum_bytes
== csum_bytes
) {
5024 calc_csum_metadata_size(inode
, num_bytes
, 0);
5026 u64 orig_csum_bytes
= BTRFS_I(inode
)->csum_bytes
;
5030 * This is tricky, but first we need to figure out how much we
5031 * free'd from any free-ers that occured during this
5032 * reservation, so we reset ->csum_bytes to the csum_bytes
5033 * before we dropped our lock, and then call the free for the
5034 * number of bytes that were freed while we were trying our
5037 bytes
= csum_bytes
- BTRFS_I(inode
)->csum_bytes
;
5038 BTRFS_I(inode
)->csum_bytes
= csum_bytes
;
5039 to_free
= calc_csum_metadata_size(inode
, bytes
, 0);
5043 * Now we need to see how much we would have freed had we not
5044 * been making this reservation and our ->csum_bytes were not
5045 * artificially inflated.
5047 BTRFS_I(inode
)->csum_bytes
= csum_bytes
- num_bytes
;
5048 bytes
= csum_bytes
- orig_csum_bytes
;
5049 bytes
= calc_csum_metadata_size(inode
, bytes
, 0);
5052 * Now reset ->csum_bytes to what it should be. If bytes is
5053 * more than to_free then we would have free'd more space had we
5054 * not had an artificially high ->csum_bytes, so we need to free
5055 * the remainder. If bytes is the same or less then we don't
5056 * need to do anything, the other free-ers did the correct
5059 BTRFS_I(inode
)->csum_bytes
= orig_csum_bytes
- num_bytes
;
5060 if (bytes
> to_free
)
5061 to_free
= bytes
- to_free
;
5065 spin_unlock(&BTRFS_I(inode
)->lock
);
5067 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
5070 btrfs_block_rsv_release(root
, block_rsv
, to_free
);
5071 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5072 btrfs_ino(inode
), to_free
, 0);
5075 mutex_unlock(&BTRFS_I(inode
)->delalloc_mutex
);
5080 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5081 * @inode: the inode to release the reservation for
5082 * @num_bytes: the number of bytes we're releasing
5084 * This will release the metadata reservation for an inode. This can be called
5085 * once we complete IO for a given set of bytes to release their metadata
5088 void btrfs_delalloc_release_metadata(struct inode
*inode
, u64 num_bytes
)
5090 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5094 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
5095 spin_lock(&BTRFS_I(inode
)->lock
);
5096 dropped
= drop_outstanding_extent(inode
);
5099 to_free
= calc_csum_metadata_size(inode
, num_bytes
, 0);
5100 spin_unlock(&BTRFS_I(inode
)->lock
);
5102 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
5104 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5105 btrfs_ino(inode
), to_free
, 0);
5106 if (root
->fs_info
->quota_enabled
) {
5107 btrfs_qgroup_free(root
, num_bytes
+
5108 dropped
* root
->leafsize
);
5111 btrfs_block_rsv_release(root
, &root
->fs_info
->delalloc_block_rsv
,
5116 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5117 * @inode: inode we're writing to
5118 * @num_bytes: the number of bytes we want to allocate
5120 * This will do the following things
5122 * o reserve space in the data space info for num_bytes
5123 * o reserve space in the metadata space info based on number of outstanding
5124 * extents and how much csums will be needed
5125 * o add to the inodes ->delalloc_bytes
5126 * o add it to the fs_info's delalloc inodes list.
5128 * This will return 0 for success and -ENOSPC if there is no space left.
5130 int btrfs_delalloc_reserve_space(struct inode
*inode
, u64 num_bytes
)
5134 ret
= btrfs_check_data_free_space(inode
, num_bytes
);
5138 ret
= btrfs_delalloc_reserve_metadata(inode
, num_bytes
);
5140 btrfs_free_reserved_data_space(inode
, num_bytes
);
5148 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5149 * @inode: inode we're releasing space for
5150 * @num_bytes: the number of bytes we want to free up
5152 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5153 * called in the case that we don't need the metadata AND data reservations
5154 * anymore. So if there is an error or we insert an inline extent.
5156 * This function will release the metadata space that was not used and will
5157 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5158 * list if there are no delalloc bytes left.
5160 void btrfs_delalloc_release_space(struct inode
*inode
, u64 num_bytes
)
5162 btrfs_delalloc_release_metadata(inode
, num_bytes
);
5163 btrfs_free_reserved_data_space(inode
, num_bytes
);
5166 static int update_block_group(struct btrfs_root
*root
,
5167 u64 bytenr
, u64 num_bytes
, int alloc
)
5169 struct btrfs_block_group_cache
*cache
= NULL
;
5170 struct btrfs_fs_info
*info
= root
->fs_info
;
5171 u64 total
= num_bytes
;
5176 /* block accounting for super block */
5177 spin_lock(&info
->delalloc_root_lock
);
5178 old_val
= btrfs_super_bytes_used(info
->super_copy
);
5180 old_val
+= num_bytes
;
5182 old_val
-= num_bytes
;
5183 btrfs_set_super_bytes_used(info
->super_copy
, old_val
);
5184 spin_unlock(&info
->delalloc_root_lock
);
5187 cache
= btrfs_lookup_block_group(info
, bytenr
);
5190 if (cache
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
5191 BTRFS_BLOCK_GROUP_RAID1
|
5192 BTRFS_BLOCK_GROUP_RAID10
))
5197 * If this block group has free space cache written out, we
5198 * need to make sure to load it if we are removing space. This
5199 * is because we need the unpinning stage to actually add the
5200 * space back to the block group, otherwise we will leak space.
5202 if (!alloc
&& cache
->cached
== BTRFS_CACHE_NO
)
5203 cache_block_group(cache
, 1);
5205 byte_in_group
= bytenr
- cache
->key
.objectid
;
5206 WARN_ON(byte_in_group
> cache
->key
.offset
);
5208 spin_lock(&cache
->space_info
->lock
);
5209 spin_lock(&cache
->lock
);
5211 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
5212 cache
->disk_cache_state
< BTRFS_DC_CLEAR
)
5213 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
5216 old_val
= btrfs_block_group_used(&cache
->item
);
5217 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
5219 old_val
+= num_bytes
;
5220 btrfs_set_block_group_used(&cache
->item
, old_val
);
5221 cache
->reserved
-= num_bytes
;
5222 cache
->space_info
->bytes_reserved
-= num_bytes
;
5223 cache
->space_info
->bytes_used
+= num_bytes
;
5224 cache
->space_info
->disk_used
+= num_bytes
* factor
;
5225 spin_unlock(&cache
->lock
);
5226 spin_unlock(&cache
->space_info
->lock
);
5228 old_val
-= num_bytes
;
5229 btrfs_set_block_group_used(&cache
->item
, old_val
);
5230 cache
->pinned
+= num_bytes
;
5231 cache
->space_info
->bytes_pinned
+= num_bytes
;
5232 cache
->space_info
->bytes_used
-= num_bytes
;
5233 cache
->space_info
->disk_used
-= num_bytes
* factor
;
5234 spin_unlock(&cache
->lock
);
5235 spin_unlock(&cache
->space_info
->lock
);
5237 set_extent_dirty(info
->pinned_extents
,
5238 bytenr
, bytenr
+ num_bytes
- 1,
5239 GFP_NOFS
| __GFP_NOFAIL
);
5241 btrfs_put_block_group(cache
);
5243 bytenr
+= num_bytes
;
5248 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
5250 struct btrfs_block_group_cache
*cache
;
5253 spin_lock(&root
->fs_info
->block_group_cache_lock
);
5254 bytenr
= root
->fs_info
->first_logical_byte
;
5255 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
5257 if (bytenr
< (u64
)-1)
5260 cache
= btrfs_lookup_first_block_group(root
->fs_info
, search_start
);
5264 bytenr
= cache
->key
.objectid
;
5265 btrfs_put_block_group(cache
);
5270 static int pin_down_extent(struct btrfs_root
*root
,
5271 struct btrfs_block_group_cache
*cache
,
5272 u64 bytenr
, u64 num_bytes
, int reserved
)
5274 spin_lock(&cache
->space_info
->lock
);
5275 spin_lock(&cache
->lock
);
5276 cache
->pinned
+= num_bytes
;
5277 cache
->space_info
->bytes_pinned
+= num_bytes
;
5279 cache
->reserved
-= num_bytes
;
5280 cache
->space_info
->bytes_reserved
-= num_bytes
;
5282 spin_unlock(&cache
->lock
);
5283 spin_unlock(&cache
->space_info
->lock
);
5285 set_extent_dirty(root
->fs_info
->pinned_extents
, bytenr
,
5286 bytenr
+ num_bytes
- 1, GFP_NOFS
| __GFP_NOFAIL
);
5288 trace_btrfs_reserved_extent_free(root
, bytenr
, num_bytes
);
5293 * this function must be called within transaction
5295 int btrfs_pin_extent(struct btrfs_root
*root
,
5296 u64 bytenr
, u64 num_bytes
, int reserved
)
5298 struct btrfs_block_group_cache
*cache
;
5300 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
5301 BUG_ON(!cache
); /* Logic error */
5303 pin_down_extent(root
, cache
, bytenr
, num_bytes
, reserved
);
5305 btrfs_put_block_group(cache
);
5310 * this function must be called within transaction
5312 int btrfs_pin_extent_for_log_replay(struct btrfs_root
*root
,
5313 u64 bytenr
, u64 num_bytes
)
5315 struct btrfs_block_group_cache
*cache
;
5318 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
5323 * pull in the free space cache (if any) so that our pin
5324 * removes the free space from the cache. We have load_only set
5325 * to one because the slow code to read in the free extents does check
5326 * the pinned extents.
5328 cache_block_group(cache
, 1);
5330 pin_down_extent(root
, cache
, bytenr
, num_bytes
, 0);
5332 /* remove us from the free space cache (if we're there at all) */
5333 ret
= btrfs_remove_free_space(cache
, bytenr
, num_bytes
);
5334 btrfs_put_block_group(cache
);
5338 static int __exclude_logged_extent(struct btrfs_root
*root
, u64 start
, u64 num_bytes
)
5341 struct btrfs_block_group_cache
*block_group
;
5342 struct btrfs_caching_control
*caching_ctl
;
5344 block_group
= btrfs_lookup_block_group(root
->fs_info
, start
);
5348 cache_block_group(block_group
, 0);
5349 caching_ctl
= get_caching_control(block_group
);
5353 BUG_ON(!block_group_cache_done(block_group
));
5354 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
5356 mutex_lock(&caching_ctl
->mutex
);
5358 if (start
>= caching_ctl
->progress
) {
5359 ret
= add_excluded_extent(root
, start
, num_bytes
);
5360 } else if (start
+ num_bytes
<= caching_ctl
->progress
) {
5361 ret
= btrfs_remove_free_space(block_group
,
5364 num_bytes
= caching_ctl
->progress
- start
;
5365 ret
= btrfs_remove_free_space(block_group
,
5370 num_bytes
= (start
+ num_bytes
) -
5371 caching_ctl
->progress
;
5372 start
= caching_ctl
->progress
;
5373 ret
= add_excluded_extent(root
, start
, num_bytes
);
5376 mutex_unlock(&caching_ctl
->mutex
);
5377 put_caching_control(caching_ctl
);
5379 btrfs_put_block_group(block_group
);
5383 int btrfs_exclude_logged_extents(struct btrfs_root
*log
,
5384 struct extent_buffer
*eb
)
5386 struct btrfs_file_extent_item
*item
;
5387 struct btrfs_key key
;
5391 if (!btrfs_fs_incompat(log
->fs_info
, MIXED_GROUPS
))
5394 for (i
= 0; i
< btrfs_header_nritems(eb
); i
++) {
5395 btrfs_item_key_to_cpu(eb
, &key
, i
);
5396 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
5398 item
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
5399 found_type
= btrfs_file_extent_type(eb
, item
);
5400 if (found_type
== BTRFS_FILE_EXTENT_INLINE
)
5402 if (btrfs_file_extent_disk_bytenr(eb
, item
) == 0)
5404 key
.objectid
= btrfs_file_extent_disk_bytenr(eb
, item
);
5405 key
.offset
= btrfs_file_extent_disk_num_bytes(eb
, item
);
5406 __exclude_logged_extent(log
, key
.objectid
, key
.offset
);
5413 * btrfs_update_reserved_bytes - update the block_group and space info counters
5414 * @cache: The cache we are manipulating
5415 * @num_bytes: The number of bytes in question
5416 * @reserve: One of the reservation enums
5418 * This is called by the allocator when it reserves space, or by somebody who is
5419 * freeing space that was never actually used on disk. For example if you
5420 * reserve some space for a new leaf in transaction A and before transaction A
5421 * commits you free that leaf, you call this with reserve set to 0 in order to
5422 * clear the reservation.
5424 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5425 * ENOSPC accounting. For data we handle the reservation through clearing the
5426 * delalloc bits in the io_tree. We have to do this since we could end up
5427 * allocating less disk space for the amount of data we have reserved in the
5428 * case of compression.
5430 * If this is a reservation and the block group has become read only we cannot
5431 * make the reservation and return -EAGAIN, otherwise this function always
5434 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
5435 u64 num_bytes
, int reserve
)
5437 struct btrfs_space_info
*space_info
= cache
->space_info
;
5440 spin_lock(&space_info
->lock
);
5441 spin_lock(&cache
->lock
);
5442 if (reserve
!= RESERVE_FREE
) {
5446 cache
->reserved
+= num_bytes
;
5447 space_info
->bytes_reserved
+= num_bytes
;
5448 if (reserve
== RESERVE_ALLOC
) {
5449 trace_btrfs_space_reservation(cache
->fs_info
,
5450 "space_info", space_info
->flags
,
5452 space_info
->bytes_may_use
-= num_bytes
;
5457 space_info
->bytes_readonly
+= num_bytes
;
5458 cache
->reserved
-= num_bytes
;
5459 space_info
->bytes_reserved
-= num_bytes
;
5461 spin_unlock(&cache
->lock
);
5462 spin_unlock(&space_info
->lock
);
5466 void btrfs_prepare_extent_commit(struct btrfs_trans_handle
*trans
,
5467 struct btrfs_root
*root
)
5469 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5470 struct btrfs_caching_control
*next
;
5471 struct btrfs_caching_control
*caching_ctl
;
5472 struct btrfs_block_group_cache
*cache
;
5473 struct btrfs_space_info
*space_info
;
5475 down_write(&fs_info
->commit_root_sem
);
5477 list_for_each_entry_safe(caching_ctl
, next
,
5478 &fs_info
->caching_block_groups
, list
) {
5479 cache
= caching_ctl
->block_group
;
5480 if (block_group_cache_done(cache
)) {
5481 cache
->last_byte_to_unpin
= (u64
)-1;
5482 list_del_init(&caching_ctl
->list
);
5483 put_caching_control(caching_ctl
);
5485 cache
->last_byte_to_unpin
= caching_ctl
->progress
;
5489 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
5490 fs_info
->pinned_extents
= &fs_info
->freed_extents
[1];
5492 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
5494 up_write(&fs_info
->commit_root_sem
);
5496 list_for_each_entry_rcu(space_info
, &fs_info
->space_info
, list
)
5497 percpu_counter_set(&space_info
->total_bytes_pinned
, 0);
5499 update_global_block_rsv(fs_info
);
5502 static int unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
5504 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5505 struct btrfs_block_group_cache
*cache
= NULL
;
5506 struct btrfs_space_info
*space_info
;
5507 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5511 while (start
<= end
) {
5514 start
>= cache
->key
.objectid
+ cache
->key
.offset
) {
5516 btrfs_put_block_group(cache
);
5517 cache
= btrfs_lookup_block_group(fs_info
, start
);
5518 BUG_ON(!cache
); /* Logic error */
5521 len
= cache
->key
.objectid
+ cache
->key
.offset
- start
;
5522 len
= min(len
, end
+ 1 - start
);
5524 if (start
< cache
->last_byte_to_unpin
) {
5525 len
= min(len
, cache
->last_byte_to_unpin
- start
);
5526 btrfs_add_free_space(cache
, start
, len
);
5530 space_info
= cache
->space_info
;
5532 spin_lock(&space_info
->lock
);
5533 spin_lock(&cache
->lock
);
5534 cache
->pinned
-= len
;
5535 space_info
->bytes_pinned
-= len
;
5537 space_info
->bytes_readonly
+= len
;
5540 spin_unlock(&cache
->lock
);
5541 if (!readonly
&& global_rsv
->space_info
== space_info
) {
5542 spin_lock(&global_rsv
->lock
);
5543 if (!global_rsv
->full
) {
5544 len
= min(len
, global_rsv
->size
-
5545 global_rsv
->reserved
);
5546 global_rsv
->reserved
+= len
;
5547 space_info
->bytes_may_use
+= len
;
5548 if (global_rsv
->reserved
>= global_rsv
->size
)
5549 global_rsv
->full
= 1;
5551 spin_unlock(&global_rsv
->lock
);
5553 spin_unlock(&space_info
->lock
);
5557 btrfs_put_block_group(cache
);
5561 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
5562 struct btrfs_root
*root
)
5564 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5565 struct extent_io_tree
*unpin
;
5573 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
5574 unpin
= &fs_info
->freed_extents
[1];
5576 unpin
= &fs_info
->freed_extents
[0];
5579 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
5580 EXTENT_DIRTY
, NULL
);
5584 if (btrfs_test_opt(root
, DISCARD
))
5585 ret
= btrfs_discard_extent(root
, start
,
5586 end
+ 1 - start
, NULL
);
5588 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
5589 unpin_extent_range(root
, start
, end
);
5596 static void add_pinned_bytes(struct btrfs_fs_info
*fs_info
, u64 num_bytes
,
5597 u64 owner
, u64 root_objectid
)
5599 struct btrfs_space_info
*space_info
;
5602 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
5603 if (root_objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
5604 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
5606 flags
= BTRFS_BLOCK_GROUP_METADATA
;
5608 flags
= BTRFS_BLOCK_GROUP_DATA
;
5611 space_info
= __find_space_info(fs_info
, flags
);
5612 BUG_ON(!space_info
); /* Logic bug */
5613 percpu_counter_add(&space_info
->total_bytes_pinned
, num_bytes
);
5617 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
5618 struct btrfs_root
*root
,
5619 u64 bytenr
, u64 num_bytes
, u64 parent
,
5620 u64 root_objectid
, u64 owner_objectid
,
5621 u64 owner_offset
, int refs_to_drop
,
5622 struct btrfs_delayed_extent_op
*extent_op
)
5624 struct btrfs_key key
;
5625 struct btrfs_path
*path
;
5626 struct btrfs_fs_info
*info
= root
->fs_info
;
5627 struct btrfs_root
*extent_root
= info
->extent_root
;
5628 struct extent_buffer
*leaf
;
5629 struct btrfs_extent_item
*ei
;
5630 struct btrfs_extent_inline_ref
*iref
;
5633 int extent_slot
= 0;
5634 int found_extent
= 0;
5638 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
5641 path
= btrfs_alloc_path();
5646 path
->leave_spinning
= 1;
5648 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
5649 BUG_ON(!is_data
&& refs_to_drop
!= 1);
5652 skinny_metadata
= 0;
5654 ret
= lookup_extent_backref(trans
, extent_root
, path
, &iref
,
5655 bytenr
, num_bytes
, parent
,
5656 root_objectid
, owner_objectid
,
5659 extent_slot
= path
->slots
[0];
5660 while (extent_slot
>= 0) {
5661 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
5663 if (key
.objectid
!= bytenr
)
5665 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
5666 key
.offset
== num_bytes
) {
5670 if (key
.type
== BTRFS_METADATA_ITEM_KEY
&&
5671 key
.offset
== owner_objectid
) {
5675 if (path
->slots
[0] - extent_slot
> 5)
5679 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5680 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
5681 if (found_extent
&& item_size
< sizeof(*ei
))
5684 if (!found_extent
) {
5686 ret
= remove_extent_backref(trans
, extent_root
, path
,
5690 btrfs_abort_transaction(trans
, extent_root
, ret
);
5693 btrfs_release_path(path
);
5694 path
->leave_spinning
= 1;
5696 key
.objectid
= bytenr
;
5697 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
5698 key
.offset
= num_bytes
;
5700 if (!is_data
&& skinny_metadata
) {
5701 key
.type
= BTRFS_METADATA_ITEM_KEY
;
5702 key
.offset
= owner_objectid
;
5705 ret
= btrfs_search_slot(trans
, extent_root
,
5707 if (ret
> 0 && skinny_metadata
&& path
->slots
[0]) {
5709 * Couldn't find our skinny metadata item,
5710 * see if we have ye olde extent item.
5713 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
5715 if (key
.objectid
== bytenr
&&
5716 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
5717 key
.offset
== num_bytes
)
5721 if (ret
> 0 && skinny_metadata
) {
5722 skinny_metadata
= false;
5723 key
.objectid
= bytenr
;
5724 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
5725 key
.offset
= num_bytes
;
5726 btrfs_release_path(path
);
5727 ret
= btrfs_search_slot(trans
, extent_root
,
5732 btrfs_err(info
, "umm, got %d back from search, was looking for %llu",
5735 btrfs_print_leaf(extent_root
,
5739 btrfs_abort_transaction(trans
, extent_root
, ret
);
5742 extent_slot
= path
->slots
[0];
5744 } else if (WARN_ON(ret
== -ENOENT
)) {
5745 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
5747 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
5748 bytenr
, parent
, root_objectid
, owner_objectid
,
5750 btrfs_abort_transaction(trans
, extent_root
, ret
);
5753 btrfs_abort_transaction(trans
, extent_root
, ret
);
5757 leaf
= path
->nodes
[0];
5758 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
5759 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5760 if (item_size
< sizeof(*ei
)) {
5761 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
5762 ret
= convert_extent_item_v0(trans
, extent_root
, path
,
5765 btrfs_abort_transaction(trans
, extent_root
, ret
);
5769 btrfs_release_path(path
);
5770 path
->leave_spinning
= 1;
5772 key
.objectid
= bytenr
;
5773 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
5774 key
.offset
= num_bytes
;
5776 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
5779 btrfs_err(info
, "umm, got %d back from search, was looking for %llu",
5781 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
5784 btrfs_abort_transaction(trans
, extent_root
, ret
);
5788 extent_slot
= path
->slots
[0];
5789 leaf
= path
->nodes
[0];
5790 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
5793 BUG_ON(item_size
< sizeof(*ei
));
5794 ei
= btrfs_item_ptr(leaf
, extent_slot
,
5795 struct btrfs_extent_item
);
5796 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
5797 key
.type
== BTRFS_EXTENT_ITEM_KEY
) {
5798 struct btrfs_tree_block_info
*bi
;
5799 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
5800 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
5801 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
5804 refs
= btrfs_extent_refs(leaf
, ei
);
5805 if (refs
< refs_to_drop
) {
5806 btrfs_err(info
, "trying to drop %d refs but we only have %Lu "
5807 "for bytenr %Lu\n", refs_to_drop
, refs
, bytenr
);
5809 btrfs_abort_transaction(trans
, extent_root
, ret
);
5812 refs
-= refs_to_drop
;
5816 __run_delayed_extent_op(extent_op
, leaf
, ei
);
5818 * In the case of inline back ref, reference count will
5819 * be updated by remove_extent_backref
5822 BUG_ON(!found_extent
);
5824 btrfs_set_extent_refs(leaf
, ei
, refs
);
5825 btrfs_mark_buffer_dirty(leaf
);
5828 ret
= remove_extent_backref(trans
, extent_root
, path
,
5832 btrfs_abort_transaction(trans
, extent_root
, ret
);
5836 add_pinned_bytes(root
->fs_info
, -num_bytes
, owner_objectid
,
5840 BUG_ON(is_data
&& refs_to_drop
!=
5841 extent_data_ref_count(root
, path
, iref
));
5843 BUG_ON(path
->slots
[0] != extent_slot
);
5845 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
5846 path
->slots
[0] = extent_slot
;
5851 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
5854 btrfs_abort_transaction(trans
, extent_root
, ret
);
5857 btrfs_release_path(path
);
5860 ret
= btrfs_del_csums(trans
, root
, bytenr
, num_bytes
);
5862 btrfs_abort_transaction(trans
, extent_root
, ret
);
5867 ret
= update_block_group(root
, bytenr
, num_bytes
, 0);
5869 btrfs_abort_transaction(trans
, extent_root
, ret
);
5874 btrfs_free_path(path
);
5879 * when we free an block, it is possible (and likely) that we free the last
5880 * delayed ref for that extent as well. This searches the delayed ref tree for
5881 * a given extent, and if there are no other delayed refs to be processed, it
5882 * removes it from the tree.
5884 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
5885 struct btrfs_root
*root
, u64 bytenr
)
5887 struct btrfs_delayed_ref_head
*head
;
5888 struct btrfs_delayed_ref_root
*delayed_refs
;
5891 delayed_refs
= &trans
->transaction
->delayed_refs
;
5892 spin_lock(&delayed_refs
->lock
);
5893 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
5895 goto out_delayed_unlock
;
5897 spin_lock(&head
->lock
);
5898 if (rb_first(&head
->ref_root
))
5901 if (head
->extent_op
) {
5902 if (!head
->must_insert_reserved
)
5904 btrfs_free_delayed_extent_op(head
->extent_op
);
5905 head
->extent_op
= NULL
;
5909 * waiting for the lock here would deadlock. If someone else has it
5910 * locked they are already in the process of dropping it anyway
5912 if (!mutex_trylock(&head
->mutex
))
5916 * at this point we have a head with no other entries. Go
5917 * ahead and process it.
5919 head
->node
.in_tree
= 0;
5920 rb_erase(&head
->href_node
, &delayed_refs
->href_root
);
5922 atomic_dec(&delayed_refs
->num_entries
);
5925 * we don't take a ref on the node because we're removing it from the
5926 * tree, so we just steal the ref the tree was holding.
5928 delayed_refs
->num_heads
--;
5929 if (head
->processing
== 0)
5930 delayed_refs
->num_heads_ready
--;
5931 head
->processing
= 0;
5932 spin_unlock(&head
->lock
);
5933 spin_unlock(&delayed_refs
->lock
);
5935 BUG_ON(head
->extent_op
);
5936 if (head
->must_insert_reserved
)
5939 mutex_unlock(&head
->mutex
);
5940 btrfs_put_delayed_ref(&head
->node
);
5943 spin_unlock(&head
->lock
);
5946 spin_unlock(&delayed_refs
->lock
);
5950 void btrfs_free_tree_block(struct btrfs_trans_handle
*trans
,
5951 struct btrfs_root
*root
,
5952 struct extent_buffer
*buf
,
5953 u64 parent
, int last_ref
)
5955 struct btrfs_block_group_cache
*cache
= NULL
;
5959 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
5960 ret
= btrfs_add_delayed_tree_ref(root
->fs_info
, trans
,
5961 buf
->start
, buf
->len
,
5962 parent
, root
->root_key
.objectid
,
5963 btrfs_header_level(buf
),
5964 BTRFS_DROP_DELAYED_REF
, NULL
, 0);
5965 BUG_ON(ret
); /* -ENOMEM */
5971 cache
= btrfs_lookup_block_group(root
->fs_info
, buf
->start
);
5973 if (btrfs_header_generation(buf
) == trans
->transid
) {
5974 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
5975 ret
= check_ref_cleanup(trans
, root
, buf
->start
);
5980 if (btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
5981 pin_down_extent(root
, cache
, buf
->start
, buf
->len
, 1);
5985 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
));
5987 btrfs_add_free_space(cache
, buf
->start
, buf
->len
);
5988 btrfs_update_reserved_bytes(cache
, buf
->len
, RESERVE_FREE
);
5989 trace_btrfs_reserved_extent_free(root
, buf
->start
, buf
->len
);
5994 add_pinned_bytes(root
->fs_info
, buf
->len
,
5995 btrfs_header_level(buf
),
5996 root
->root_key
.objectid
);
5999 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6002 clear_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
);
6003 btrfs_put_block_group(cache
);
6006 /* Can return -ENOMEM */
6007 int btrfs_free_extent(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
6008 u64 bytenr
, u64 num_bytes
, u64 parent
, u64 root_objectid
,
6009 u64 owner
, u64 offset
, int for_cow
)
6012 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6014 add_pinned_bytes(root
->fs_info
, num_bytes
, owner
, root_objectid
);
6017 * tree log blocks never actually go into the extent allocation
6018 * tree, just update pinning info and exit early.
6020 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
6021 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
6022 /* unlocks the pinned mutex */
6023 btrfs_pin_extent(root
, bytenr
, num_bytes
, 1);
6025 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
6026 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
6028 parent
, root_objectid
, (int)owner
,
6029 BTRFS_DROP_DELAYED_REF
, NULL
, for_cow
);
6031 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
6033 parent
, root_objectid
, owner
,
6034 offset
, BTRFS_DROP_DELAYED_REF
,
6040 static u64
stripe_align(struct btrfs_root
*root
,
6041 struct btrfs_block_group_cache
*cache
,
6042 u64 val
, u64 num_bytes
)
6044 u64 ret
= ALIGN(val
, root
->stripesize
);
6049 * when we wait for progress in the block group caching, its because
6050 * our allocation attempt failed at least once. So, we must sleep
6051 * and let some progress happen before we try again.
6053 * This function will sleep at least once waiting for new free space to
6054 * show up, and then it will check the block group free space numbers
6055 * for our min num_bytes. Another option is to have it go ahead
6056 * and look in the rbtree for a free extent of a given size, but this
6059 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6060 * any of the information in this block group.
6062 static noinline
void
6063 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
6066 struct btrfs_caching_control
*caching_ctl
;
6068 caching_ctl
= get_caching_control(cache
);
6072 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
) ||
6073 (cache
->free_space_ctl
->free_space
>= num_bytes
));
6075 put_caching_control(caching_ctl
);
6079 wait_block_group_cache_done(struct btrfs_block_group_cache
*cache
)
6081 struct btrfs_caching_control
*caching_ctl
;
6084 caching_ctl
= get_caching_control(cache
);
6086 return (cache
->cached
== BTRFS_CACHE_ERROR
) ? -EIO
: 0;
6088 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
));
6089 if (cache
->cached
== BTRFS_CACHE_ERROR
)
6091 put_caching_control(caching_ctl
);
6095 int __get_raid_index(u64 flags
)
6097 if (flags
& BTRFS_BLOCK_GROUP_RAID10
)
6098 return BTRFS_RAID_RAID10
;
6099 else if (flags
& BTRFS_BLOCK_GROUP_RAID1
)
6100 return BTRFS_RAID_RAID1
;
6101 else if (flags
& BTRFS_BLOCK_GROUP_DUP
)
6102 return BTRFS_RAID_DUP
;
6103 else if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
6104 return BTRFS_RAID_RAID0
;
6105 else if (flags
& BTRFS_BLOCK_GROUP_RAID5
)
6106 return BTRFS_RAID_RAID5
;
6107 else if (flags
& BTRFS_BLOCK_GROUP_RAID6
)
6108 return BTRFS_RAID_RAID6
;
6110 return BTRFS_RAID_SINGLE
; /* BTRFS_BLOCK_GROUP_SINGLE */
6113 int get_block_group_index(struct btrfs_block_group_cache
*cache
)
6115 return __get_raid_index(cache
->flags
);
6118 static const char *btrfs_raid_type_names
[BTRFS_NR_RAID_TYPES
] = {
6119 [BTRFS_RAID_RAID10
] = "raid10",
6120 [BTRFS_RAID_RAID1
] = "raid1",
6121 [BTRFS_RAID_DUP
] = "dup",
6122 [BTRFS_RAID_RAID0
] = "raid0",
6123 [BTRFS_RAID_SINGLE
] = "single",
6124 [BTRFS_RAID_RAID5
] = "raid5",
6125 [BTRFS_RAID_RAID6
] = "raid6",
6128 static const char *get_raid_name(enum btrfs_raid_types type
)
6130 if (type
>= BTRFS_NR_RAID_TYPES
)
6133 return btrfs_raid_type_names
[type
];
6136 enum btrfs_loop_type
{
6137 LOOP_CACHING_NOWAIT
= 0,
6138 LOOP_CACHING_WAIT
= 1,
6139 LOOP_ALLOC_CHUNK
= 2,
6140 LOOP_NO_EMPTY_SIZE
= 3,
6144 * walks the btree of allocated extents and find a hole of a given size.
6145 * The key ins is changed to record the hole:
6146 * ins->objectid == start position
6147 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6148 * ins->offset == the size of the hole.
6149 * Any available blocks before search_start are skipped.
6151 * If there is no suitable free space, we will record the max size of
6152 * the free space extent currently.
6154 static noinline
int find_free_extent(struct btrfs_root
*orig_root
,
6155 u64 num_bytes
, u64 empty_size
,
6156 u64 hint_byte
, struct btrfs_key
*ins
,
6160 struct btrfs_root
*root
= orig_root
->fs_info
->extent_root
;
6161 struct btrfs_free_cluster
*last_ptr
= NULL
;
6162 struct btrfs_block_group_cache
*block_group
= NULL
;
6163 u64 search_start
= 0;
6164 u64 max_extent_size
= 0;
6165 int empty_cluster
= 2 * 1024 * 1024;
6166 struct btrfs_space_info
*space_info
;
6168 int index
= __get_raid_index(flags
);
6169 int alloc_type
= (flags
& BTRFS_BLOCK_GROUP_DATA
) ?
6170 RESERVE_ALLOC_NO_ACCOUNT
: RESERVE_ALLOC
;
6171 bool failed_cluster_refill
= false;
6172 bool failed_alloc
= false;
6173 bool use_cluster
= true;
6174 bool have_caching_bg
= false;
6176 WARN_ON(num_bytes
< root
->sectorsize
);
6177 btrfs_set_key_type(ins
, BTRFS_EXTENT_ITEM_KEY
);
6181 trace_find_free_extent(orig_root
, num_bytes
, empty_size
, flags
);
6183 space_info
= __find_space_info(root
->fs_info
, flags
);
6185 btrfs_err(root
->fs_info
, "No space info for %llu", flags
);
6190 * If the space info is for both data and metadata it means we have a
6191 * small filesystem and we can't use the clustering stuff.
6193 if (btrfs_mixed_space_info(space_info
))
6194 use_cluster
= false;
6196 if (flags
& BTRFS_BLOCK_GROUP_METADATA
&& use_cluster
) {
6197 last_ptr
= &root
->fs_info
->meta_alloc_cluster
;
6198 if (!btrfs_test_opt(root
, SSD
))
6199 empty_cluster
= 64 * 1024;
6202 if ((flags
& BTRFS_BLOCK_GROUP_DATA
) && use_cluster
&&
6203 btrfs_test_opt(root
, SSD
)) {
6204 last_ptr
= &root
->fs_info
->data_alloc_cluster
;
6208 spin_lock(&last_ptr
->lock
);
6209 if (last_ptr
->block_group
)
6210 hint_byte
= last_ptr
->window_start
;
6211 spin_unlock(&last_ptr
->lock
);
6214 search_start
= max(search_start
, first_logical_byte(root
, 0));
6215 search_start
= max(search_start
, hint_byte
);
6220 if (search_start
== hint_byte
) {
6221 block_group
= btrfs_lookup_block_group(root
->fs_info
,
6224 * we don't want to use the block group if it doesn't match our
6225 * allocation bits, or if its not cached.
6227 * However if we are re-searching with an ideal block group
6228 * picked out then we don't care that the block group is cached.
6230 if (block_group
&& block_group_bits(block_group
, flags
) &&
6231 block_group
->cached
!= BTRFS_CACHE_NO
) {
6232 down_read(&space_info
->groups_sem
);
6233 if (list_empty(&block_group
->list
) ||
6236 * someone is removing this block group,
6237 * we can't jump into the have_block_group
6238 * target because our list pointers are not
6241 btrfs_put_block_group(block_group
);
6242 up_read(&space_info
->groups_sem
);
6244 index
= get_block_group_index(block_group
);
6245 goto have_block_group
;
6247 } else if (block_group
) {
6248 btrfs_put_block_group(block_group
);
6252 have_caching_bg
= false;
6253 down_read(&space_info
->groups_sem
);
6254 list_for_each_entry(block_group
, &space_info
->block_groups
[index
],
6259 btrfs_get_block_group(block_group
);
6260 search_start
= block_group
->key
.objectid
;
6263 * this can happen if we end up cycling through all the
6264 * raid types, but we want to make sure we only allocate
6265 * for the proper type.
6267 if (!block_group_bits(block_group
, flags
)) {
6268 u64 extra
= BTRFS_BLOCK_GROUP_DUP
|
6269 BTRFS_BLOCK_GROUP_RAID1
|
6270 BTRFS_BLOCK_GROUP_RAID5
|
6271 BTRFS_BLOCK_GROUP_RAID6
|
6272 BTRFS_BLOCK_GROUP_RAID10
;
6275 * if they asked for extra copies and this block group
6276 * doesn't provide them, bail. This does allow us to
6277 * fill raid0 from raid1.
6279 if ((flags
& extra
) && !(block_group
->flags
& extra
))
6284 cached
= block_group_cache_done(block_group
);
6285 if (unlikely(!cached
)) {
6286 ret
= cache_block_group(block_group
, 0);
6291 if (unlikely(block_group
->cached
== BTRFS_CACHE_ERROR
))
6293 if (unlikely(block_group
->ro
))
6297 * Ok we want to try and use the cluster allocator, so
6301 struct btrfs_block_group_cache
*used_block_group
;
6302 unsigned long aligned_cluster
;
6304 * the refill lock keeps out other
6305 * people trying to start a new cluster
6307 spin_lock(&last_ptr
->refill_lock
);
6308 used_block_group
= last_ptr
->block_group
;
6309 if (used_block_group
!= block_group
&&
6310 (!used_block_group
||
6311 used_block_group
->ro
||
6312 !block_group_bits(used_block_group
, flags
)))
6313 goto refill_cluster
;
6315 if (used_block_group
!= block_group
)
6316 btrfs_get_block_group(used_block_group
);
6318 offset
= btrfs_alloc_from_cluster(used_block_group
,
6321 used_block_group
->key
.objectid
,
6324 /* we have a block, we're done */
6325 spin_unlock(&last_ptr
->refill_lock
);
6326 trace_btrfs_reserve_extent_cluster(root
,
6328 search_start
, num_bytes
);
6329 if (used_block_group
!= block_group
) {
6330 btrfs_put_block_group(block_group
);
6331 block_group
= used_block_group
;
6336 WARN_ON(last_ptr
->block_group
!= used_block_group
);
6337 if (used_block_group
!= block_group
)
6338 btrfs_put_block_group(used_block_group
);
6340 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6341 * set up a new clusters, so lets just skip it
6342 * and let the allocator find whatever block
6343 * it can find. If we reach this point, we
6344 * will have tried the cluster allocator
6345 * plenty of times and not have found
6346 * anything, so we are likely way too
6347 * fragmented for the clustering stuff to find
6350 * However, if the cluster is taken from the
6351 * current block group, release the cluster
6352 * first, so that we stand a better chance of
6353 * succeeding in the unclustered
6355 if (loop
>= LOOP_NO_EMPTY_SIZE
&&
6356 last_ptr
->block_group
!= block_group
) {
6357 spin_unlock(&last_ptr
->refill_lock
);
6358 goto unclustered_alloc
;
6362 * this cluster didn't work out, free it and
6365 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
6367 if (loop
>= LOOP_NO_EMPTY_SIZE
) {
6368 spin_unlock(&last_ptr
->refill_lock
);
6369 goto unclustered_alloc
;
6372 aligned_cluster
= max_t(unsigned long,
6373 empty_cluster
+ empty_size
,
6374 block_group
->full_stripe_len
);
6376 /* allocate a cluster in this block group */
6377 ret
= btrfs_find_space_cluster(root
, block_group
,
6378 last_ptr
, search_start
,
6383 * now pull our allocation out of this
6386 offset
= btrfs_alloc_from_cluster(block_group
,
6392 /* we found one, proceed */
6393 spin_unlock(&last_ptr
->refill_lock
);
6394 trace_btrfs_reserve_extent_cluster(root
,
6395 block_group
, search_start
,
6399 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
6400 && !failed_cluster_refill
) {
6401 spin_unlock(&last_ptr
->refill_lock
);
6403 failed_cluster_refill
= true;
6404 wait_block_group_cache_progress(block_group
,
6405 num_bytes
+ empty_cluster
+ empty_size
);
6406 goto have_block_group
;
6410 * at this point we either didn't find a cluster
6411 * or we weren't able to allocate a block from our
6412 * cluster. Free the cluster we've been trying
6413 * to use, and go to the next block group
6415 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
6416 spin_unlock(&last_ptr
->refill_lock
);
6421 spin_lock(&block_group
->free_space_ctl
->tree_lock
);
6423 block_group
->free_space_ctl
->free_space
<
6424 num_bytes
+ empty_cluster
+ empty_size
) {
6425 if (block_group
->free_space_ctl
->free_space
>
6428 block_group
->free_space_ctl
->free_space
;
6429 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
6432 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
6434 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
6435 num_bytes
, empty_size
,
6438 * If we didn't find a chunk, and we haven't failed on this
6439 * block group before, and this block group is in the middle of
6440 * caching and we are ok with waiting, then go ahead and wait
6441 * for progress to be made, and set failed_alloc to true.
6443 * If failed_alloc is true then we've already waited on this
6444 * block group once and should move on to the next block group.
6446 if (!offset
&& !failed_alloc
&& !cached
&&
6447 loop
> LOOP_CACHING_NOWAIT
) {
6448 wait_block_group_cache_progress(block_group
,
6449 num_bytes
+ empty_size
);
6450 failed_alloc
= true;
6451 goto have_block_group
;
6452 } else if (!offset
) {
6454 have_caching_bg
= true;
6458 search_start
= stripe_align(root
, block_group
,
6461 /* move on to the next group */
6462 if (search_start
+ num_bytes
>
6463 block_group
->key
.objectid
+ block_group
->key
.offset
) {
6464 btrfs_add_free_space(block_group
, offset
, num_bytes
);
6468 if (offset
< search_start
)
6469 btrfs_add_free_space(block_group
, offset
,
6470 search_start
- offset
);
6471 BUG_ON(offset
> search_start
);
6473 ret
= btrfs_update_reserved_bytes(block_group
, num_bytes
,
6475 if (ret
== -EAGAIN
) {
6476 btrfs_add_free_space(block_group
, offset
, num_bytes
);
6480 /* we are all good, lets return */
6481 ins
->objectid
= search_start
;
6482 ins
->offset
= num_bytes
;
6484 trace_btrfs_reserve_extent(orig_root
, block_group
,
6485 search_start
, num_bytes
);
6486 btrfs_put_block_group(block_group
);
6489 failed_cluster_refill
= false;
6490 failed_alloc
= false;
6491 BUG_ON(index
!= get_block_group_index(block_group
));
6492 btrfs_put_block_group(block_group
);
6494 up_read(&space_info
->groups_sem
);
6496 if (!ins
->objectid
&& loop
>= LOOP_CACHING_WAIT
&& have_caching_bg
)
6499 if (!ins
->objectid
&& ++index
< BTRFS_NR_RAID_TYPES
)
6503 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
6504 * caching kthreads as we move along
6505 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
6506 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
6507 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
6510 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
) {
6513 if (loop
== LOOP_ALLOC_CHUNK
) {
6514 struct btrfs_trans_handle
*trans
;
6516 trans
= btrfs_join_transaction(root
);
6517 if (IS_ERR(trans
)) {
6518 ret
= PTR_ERR(trans
);
6522 ret
= do_chunk_alloc(trans
, root
, flags
,
6525 * Do not bail out on ENOSPC since we
6526 * can do more things.
6528 if (ret
< 0 && ret
!= -ENOSPC
)
6529 btrfs_abort_transaction(trans
,
6533 btrfs_end_transaction(trans
, root
);
6538 if (loop
== LOOP_NO_EMPTY_SIZE
) {
6544 } else if (!ins
->objectid
) {
6546 } else if (ins
->objectid
) {
6551 ins
->offset
= max_extent_size
;
6555 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
6556 int dump_block_groups
)
6558 struct btrfs_block_group_cache
*cache
;
6561 spin_lock(&info
->lock
);
6562 printk(KERN_INFO
"BTRFS: space_info %llu has %llu free, is %sfull\n",
6564 info
->total_bytes
- info
->bytes_used
- info
->bytes_pinned
-
6565 info
->bytes_reserved
- info
->bytes_readonly
,
6566 (info
->full
) ? "" : "not ");
6567 printk(KERN_INFO
"BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
6568 "reserved=%llu, may_use=%llu, readonly=%llu\n",
6569 info
->total_bytes
, info
->bytes_used
, info
->bytes_pinned
,
6570 info
->bytes_reserved
, info
->bytes_may_use
,
6571 info
->bytes_readonly
);
6572 spin_unlock(&info
->lock
);
6574 if (!dump_block_groups
)
6577 down_read(&info
->groups_sem
);
6579 list_for_each_entry(cache
, &info
->block_groups
[index
], list
) {
6580 spin_lock(&cache
->lock
);
6581 printk(KERN_INFO
"BTRFS: "
6582 "block group %llu has %llu bytes, "
6583 "%llu used %llu pinned %llu reserved %s\n",
6584 cache
->key
.objectid
, cache
->key
.offset
,
6585 btrfs_block_group_used(&cache
->item
), cache
->pinned
,
6586 cache
->reserved
, cache
->ro
? "[readonly]" : "");
6587 btrfs_dump_free_space(cache
, bytes
);
6588 spin_unlock(&cache
->lock
);
6590 if (++index
< BTRFS_NR_RAID_TYPES
)
6592 up_read(&info
->groups_sem
);
6595 int btrfs_reserve_extent(struct btrfs_root
*root
,
6596 u64 num_bytes
, u64 min_alloc_size
,
6597 u64 empty_size
, u64 hint_byte
,
6598 struct btrfs_key
*ins
, int is_data
)
6600 bool final_tried
= false;
6604 flags
= btrfs_get_alloc_profile(root
, is_data
);
6606 WARN_ON(num_bytes
< root
->sectorsize
);
6607 ret
= find_free_extent(root
, num_bytes
, empty_size
, hint_byte
, ins
,
6610 if (ret
== -ENOSPC
) {
6611 if (!final_tried
&& ins
->offset
) {
6612 num_bytes
= min(num_bytes
>> 1, ins
->offset
);
6613 num_bytes
= round_down(num_bytes
, root
->sectorsize
);
6614 num_bytes
= max(num_bytes
, min_alloc_size
);
6615 if (num_bytes
== min_alloc_size
)
6618 } else if (btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
6619 struct btrfs_space_info
*sinfo
;
6621 sinfo
= __find_space_info(root
->fs_info
, flags
);
6622 btrfs_err(root
->fs_info
, "allocation failed flags %llu, wanted %llu",
6625 dump_space_info(sinfo
, num_bytes
, 1);
6632 static int __btrfs_free_reserved_extent(struct btrfs_root
*root
,
6633 u64 start
, u64 len
, int pin
)
6635 struct btrfs_block_group_cache
*cache
;
6638 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
6640 btrfs_err(root
->fs_info
, "Unable to find block group for %llu",
6645 if (btrfs_test_opt(root
, DISCARD
))
6646 ret
= btrfs_discard_extent(root
, start
, len
, NULL
);
6649 pin_down_extent(root
, cache
, start
, len
, 1);
6651 btrfs_add_free_space(cache
, start
, len
);
6652 btrfs_update_reserved_bytes(cache
, len
, RESERVE_FREE
);
6654 btrfs_put_block_group(cache
);
6656 trace_btrfs_reserved_extent_free(root
, start
, len
);
6661 int btrfs_free_reserved_extent(struct btrfs_root
*root
,
6664 return __btrfs_free_reserved_extent(root
, start
, len
, 0);
6667 int btrfs_free_and_pin_reserved_extent(struct btrfs_root
*root
,
6670 return __btrfs_free_reserved_extent(root
, start
, len
, 1);
6673 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
6674 struct btrfs_root
*root
,
6675 u64 parent
, u64 root_objectid
,
6676 u64 flags
, u64 owner
, u64 offset
,
6677 struct btrfs_key
*ins
, int ref_mod
)
6680 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6681 struct btrfs_extent_item
*extent_item
;
6682 struct btrfs_extent_inline_ref
*iref
;
6683 struct btrfs_path
*path
;
6684 struct extent_buffer
*leaf
;
6689 type
= BTRFS_SHARED_DATA_REF_KEY
;
6691 type
= BTRFS_EXTENT_DATA_REF_KEY
;
6693 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
6695 path
= btrfs_alloc_path();
6699 path
->leave_spinning
= 1;
6700 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
6703 btrfs_free_path(path
);
6707 leaf
= path
->nodes
[0];
6708 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
6709 struct btrfs_extent_item
);
6710 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
6711 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
6712 btrfs_set_extent_flags(leaf
, extent_item
,
6713 flags
| BTRFS_EXTENT_FLAG_DATA
);
6715 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
6716 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
6718 struct btrfs_shared_data_ref
*ref
;
6719 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
6720 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
6721 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
6723 struct btrfs_extent_data_ref
*ref
;
6724 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
6725 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
6726 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
6727 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
6728 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
6731 btrfs_mark_buffer_dirty(path
->nodes
[0]);
6732 btrfs_free_path(path
);
6734 ret
= update_block_group(root
, ins
->objectid
, ins
->offset
, 1);
6735 if (ret
) { /* -ENOENT, logic error */
6736 btrfs_err(fs_info
, "update block group failed for %llu %llu",
6737 ins
->objectid
, ins
->offset
);
6740 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, ins
->offset
);
6744 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
6745 struct btrfs_root
*root
,
6746 u64 parent
, u64 root_objectid
,
6747 u64 flags
, struct btrfs_disk_key
*key
,
6748 int level
, struct btrfs_key
*ins
)
6751 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6752 struct btrfs_extent_item
*extent_item
;
6753 struct btrfs_tree_block_info
*block_info
;
6754 struct btrfs_extent_inline_ref
*iref
;
6755 struct btrfs_path
*path
;
6756 struct extent_buffer
*leaf
;
6757 u32 size
= sizeof(*extent_item
) + sizeof(*iref
);
6758 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
6761 if (!skinny_metadata
)
6762 size
+= sizeof(*block_info
);
6764 path
= btrfs_alloc_path();
6766 btrfs_free_and_pin_reserved_extent(root
, ins
->objectid
,
6771 path
->leave_spinning
= 1;
6772 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
6775 btrfs_free_and_pin_reserved_extent(root
, ins
->objectid
,
6777 btrfs_free_path(path
);
6781 leaf
= path
->nodes
[0];
6782 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
6783 struct btrfs_extent_item
);
6784 btrfs_set_extent_refs(leaf
, extent_item
, 1);
6785 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
6786 btrfs_set_extent_flags(leaf
, extent_item
,
6787 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
6789 if (skinny_metadata
) {
6790 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
6792 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
6793 btrfs_set_tree_block_key(leaf
, block_info
, key
);
6794 btrfs_set_tree_block_level(leaf
, block_info
, level
);
6795 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
6799 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
6800 btrfs_set_extent_inline_ref_type(leaf
, iref
,
6801 BTRFS_SHARED_BLOCK_REF_KEY
);
6802 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
6804 btrfs_set_extent_inline_ref_type(leaf
, iref
,
6805 BTRFS_TREE_BLOCK_REF_KEY
);
6806 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
6809 btrfs_mark_buffer_dirty(leaf
);
6810 btrfs_free_path(path
);
6812 ret
= update_block_group(root
, ins
->objectid
, root
->leafsize
, 1);
6813 if (ret
) { /* -ENOENT, logic error */
6814 btrfs_err(fs_info
, "update block group failed for %llu %llu",
6815 ins
->objectid
, ins
->offset
);
6819 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, root
->leafsize
);
6823 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
6824 struct btrfs_root
*root
,
6825 u64 root_objectid
, u64 owner
,
6826 u64 offset
, struct btrfs_key
*ins
)
6830 BUG_ON(root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
6832 ret
= btrfs_add_delayed_data_ref(root
->fs_info
, trans
, ins
->objectid
,
6834 root_objectid
, owner
, offset
,
6835 BTRFS_ADD_DELAYED_EXTENT
, NULL
, 0);
6840 * this is used by the tree logging recovery code. It records that
6841 * an extent has been allocated and makes sure to clear the free
6842 * space cache bits as well
6844 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
6845 struct btrfs_root
*root
,
6846 u64 root_objectid
, u64 owner
, u64 offset
,
6847 struct btrfs_key
*ins
)
6850 struct btrfs_block_group_cache
*block_group
;
6853 * Mixed block groups will exclude before processing the log so we only
6854 * need to do the exlude dance if this fs isn't mixed.
6856 if (!btrfs_fs_incompat(root
->fs_info
, MIXED_GROUPS
)) {
6857 ret
= __exclude_logged_extent(root
, ins
->objectid
, ins
->offset
);
6862 block_group
= btrfs_lookup_block_group(root
->fs_info
, ins
->objectid
);
6866 ret
= btrfs_update_reserved_bytes(block_group
, ins
->offset
,
6867 RESERVE_ALLOC_NO_ACCOUNT
);
6868 BUG_ON(ret
); /* logic error */
6869 ret
= alloc_reserved_file_extent(trans
, root
, 0, root_objectid
,
6870 0, owner
, offset
, ins
, 1);
6871 btrfs_put_block_group(block_group
);
6875 static struct extent_buffer
*
6876 btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
6877 u64 bytenr
, u32 blocksize
, int level
)
6879 struct extent_buffer
*buf
;
6881 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
6883 return ERR_PTR(-ENOMEM
);
6884 btrfs_set_header_generation(buf
, trans
->transid
);
6885 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, buf
, level
);
6886 btrfs_tree_lock(buf
);
6887 clean_tree_block(trans
, root
, buf
);
6888 clear_bit(EXTENT_BUFFER_STALE
, &buf
->bflags
);
6890 btrfs_set_lock_blocking(buf
);
6891 btrfs_set_buffer_uptodate(buf
);
6893 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
6895 * we allow two log transactions at a time, use different
6896 * EXENT bit to differentiate dirty pages.
6898 if (root
->log_transid
% 2 == 0)
6899 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
6900 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
6902 set_extent_new(&root
->dirty_log_pages
, buf
->start
,
6903 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
6905 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
6906 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
6908 trans
->blocks_used
++;
6909 /* this returns a buffer locked for blocking */
6913 static struct btrfs_block_rsv
*
6914 use_block_rsv(struct btrfs_trans_handle
*trans
,
6915 struct btrfs_root
*root
, u32 blocksize
)
6917 struct btrfs_block_rsv
*block_rsv
;
6918 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
6920 bool global_updated
= false;
6922 block_rsv
= get_block_rsv(trans
, root
);
6924 if (unlikely(block_rsv
->size
== 0))
6927 ret
= block_rsv_use_bytes(block_rsv
, blocksize
);
6931 if (block_rsv
->failfast
)
6932 return ERR_PTR(ret
);
6934 if (block_rsv
->type
== BTRFS_BLOCK_RSV_GLOBAL
&& !global_updated
) {
6935 global_updated
= true;
6936 update_global_block_rsv(root
->fs_info
);
6940 if (btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
6941 static DEFINE_RATELIMIT_STATE(_rs
,
6942 DEFAULT_RATELIMIT_INTERVAL
* 10,
6943 /*DEFAULT_RATELIMIT_BURST*/ 1);
6944 if (__ratelimit(&_rs
))
6946 "BTRFS: block rsv returned %d\n", ret
);
6949 ret
= reserve_metadata_bytes(root
, block_rsv
, blocksize
,
6950 BTRFS_RESERVE_NO_FLUSH
);
6954 * If we couldn't reserve metadata bytes try and use some from
6955 * the global reserve if its space type is the same as the global
6958 if (block_rsv
->type
!= BTRFS_BLOCK_RSV_GLOBAL
&&
6959 block_rsv
->space_info
== global_rsv
->space_info
) {
6960 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
6964 return ERR_PTR(ret
);
6967 static void unuse_block_rsv(struct btrfs_fs_info
*fs_info
,
6968 struct btrfs_block_rsv
*block_rsv
, u32 blocksize
)
6970 block_rsv_add_bytes(block_rsv
, blocksize
, 0);
6971 block_rsv_release_bytes(fs_info
, block_rsv
, NULL
, 0);
6975 * finds a free extent and does all the dirty work required for allocation
6976 * returns the key for the extent through ins, and a tree buffer for
6977 * the first block of the extent through buf.
6979 * returns the tree buffer or NULL.
6981 struct extent_buffer
*btrfs_alloc_free_block(struct btrfs_trans_handle
*trans
,
6982 struct btrfs_root
*root
, u32 blocksize
,
6983 u64 parent
, u64 root_objectid
,
6984 struct btrfs_disk_key
*key
, int level
,
6985 u64 hint
, u64 empty_size
)
6987 struct btrfs_key ins
;
6988 struct btrfs_block_rsv
*block_rsv
;
6989 struct extent_buffer
*buf
;
6992 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
6995 block_rsv
= use_block_rsv(trans
, root
, blocksize
);
6996 if (IS_ERR(block_rsv
))
6997 return ERR_CAST(block_rsv
);
6999 ret
= btrfs_reserve_extent(root
, blocksize
, blocksize
,
7000 empty_size
, hint
, &ins
, 0);
7002 unuse_block_rsv(root
->fs_info
, block_rsv
, blocksize
);
7003 return ERR_PTR(ret
);
7006 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
,
7008 BUG_ON(IS_ERR(buf
)); /* -ENOMEM */
7010 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
7012 parent
= ins
.objectid
;
7013 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
7017 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
7018 struct btrfs_delayed_extent_op
*extent_op
;
7019 extent_op
= btrfs_alloc_delayed_extent_op();
7020 BUG_ON(!extent_op
); /* -ENOMEM */
7022 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
7024 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
7025 extent_op
->flags_to_set
= flags
;
7026 if (skinny_metadata
)
7027 extent_op
->update_key
= 0;
7029 extent_op
->update_key
= 1;
7030 extent_op
->update_flags
= 1;
7031 extent_op
->is_data
= 0;
7032 extent_op
->level
= level
;
7034 ret
= btrfs_add_delayed_tree_ref(root
->fs_info
, trans
,
7036 ins
.offset
, parent
, root_objectid
,
7037 level
, BTRFS_ADD_DELAYED_EXTENT
,
7039 BUG_ON(ret
); /* -ENOMEM */
7044 struct walk_control
{
7045 u64 refs
[BTRFS_MAX_LEVEL
];
7046 u64 flags
[BTRFS_MAX_LEVEL
];
7047 struct btrfs_key update_progress
;
7058 #define DROP_REFERENCE 1
7059 #define UPDATE_BACKREF 2
7061 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
7062 struct btrfs_root
*root
,
7063 struct walk_control
*wc
,
7064 struct btrfs_path
*path
)
7072 struct btrfs_key key
;
7073 struct extent_buffer
*eb
;
7078 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
7079 wc
->reada_count
= wc
->reada_count
* 2 / 3;
7080 wc
->reada_count
= max(wc
->reada_count
, 2);
7082 wc
->reada_count
= wc
->reada_count
* 3 / 2;
7083 wc
->reada_count
= min_t(int, wc
->reada_count
,
7084 BTRFS_NODEPTRS_PER_BLOCK(root
));
7087 eb
= path
->nodes
[wc
->level
];
7088 nritems
= btrfs_header_nritems(eb
);
7089 blocksize
= btrfs_level_size(root
, wc
->level
- 1);
7091 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
7092 if (nread
>= wc
->reada_count
)
7096 bytenr
= btrfs_node_blockptr(eb
, slot
);
7097 generation
= btrfs_node_ptr_generation(eb
, slot
);
7099 if (slot
== path
->slots
[wc
->level
])
7102 if (wc
->stage
== UPDATE_BACKREF
&&
7103 generation
<= root
->root_key
.offset
)
7106 /* We don't lock the tree block, it's OK to be racy here */
7107 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
,
7108 wc
->level
- 1, 1, &refs
,
7110 /* We don't care about errors in readahead. */
7115 if (wc
->stage
== DROP_REFERENCE
) {
7119 if (wc
->level
== 1 &&
7120 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
7122 if (!wc
->update_ref
||
7123 generation
<= root
->root_key
.offset
)
7125 btrfs_node_key_to_cpu(eb
, &key
, slot
);
7126 ret
= btrfs_comp_cpu_keys(&key
,
7127 &wc
->update_progress
);
7131 if (wc
->level
== 1 &&
7132 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
7136 ret
= readahead_tree_block(root
, bytenr
, blocksize
,
7142 wc
->reada_slot
= slot
;
7146 * helper to process tree block while walking down the tree.
7148 * when wc->stage == UPDATE_BACKREF, this function updates
7149 * back refs for pointers in the block.
7151 * NOTE: return value 1 means we should stop walking down.
7153 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
7154 struct btrfs_root
*root
,
7155 struct btrfs_path
*path
,
7156 struct walk_control
*wc
, int lookup_info
)
7158 int level
= wc
->level
;
7159 struct extent_buffer
*eb
= path
->nodes
[level
];
7160 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
7163 if (wc
->stage
== UPDATE_BACKREF
&&
7164 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
7168 * when reference count of tree block is 1, it won't increase
7169 * again. once full backref flag is set, we never clear it.
7172 ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
7173 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
)))) {
7174 BUG_ON(!path
->locks
[level
]);
7175 ret
= btrfs_lookup_extent_info(trans
, root
,
7176 eb
->start
, level
, 1,
7179 BUG_ON(ret
== -ENOMEM
);
7182 BUG_ON(wc
->refs
[level
] == 0);
7185 if (wc
->stage
== DROP_REFERENCE
) {
7186 if (wc
->refs
[level
] > 1)
7189 if (path
->locks
[level
] && !wc
->keep_locks
) {
7190 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
7191 path
->locks
[level
] = 0;
7196 /* wc->stage == UPDATE_BACKREF */
7197 if (!(wc
->flags
[level
] & flag
)) {
7198 BUG_ON(!path
->locks
[level
]);
7199 ret
= btrfs_inc_ref(trans
, root
, eb
, 1, wc
->for_reloc
);
7200 BUG_ON(ret
); /* -ENOMEM */
7201 ret
= btrfs_dec_ref(trans
, root
, eb
, 0, wc
->for_reloc
);
7202 BUG_ON(ret
); /* -ENOMEM */
7203 ret
= btrfs_set_disk_extent_flags(trans
, root
, eb
->start
,
7205 btrfs_header_level(eb
), 0);
7206 BUG_ON(ret
); /* -ENOMEM */
7207 wc
->flags
[level
] |= flag
;
7211 * the block is shared by multiple trees, so it's not good to
7212 * keep the tree lock
7214 if (path
->locks
[level
] && level
> 0) {
7215 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
7216 path
->locks
[level
] = 0;
7222 * helper to process tree block pointer.
7224 * when wc->stage == DROP_REFERENCE, this function checks
7225 * reference count of the block pointed to. if the block
7226 * is shared and we need update back refs for the subtree
7227 * rooted at the block, this function changes wc->stage to
7228 * UPDATE_BACKREF. if the block is shared and there is no
7229 * need to update back, this function drops the reference
7232 * NOTE: return value 1 means we should stop walking down.
7234 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
7235 struct btrfs_root
*root
,
7236 struct btrfs_path
*path
,
7237 struct walk_control
*wc
, int *lookup_info
)
7243 struct btrfs_key key
;
7244 struct extent_buffer
*next
;
7245 int level
= wc
->level
;
7249 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
7250 path
->slots
[level
]);
7252 * if the lower level block was created before the snapshot
7253 * was created, we know there is no need to update back refs
7256 if (wc
->stage
== UPDATE_BACKREF
&&
7257 generation
<= root
->root_key
.offset
) {
7262 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
7263 blocksize
= btrfs_level_size(root
, level
- 1);
7265 next
= btrfs_find_tree_block(root
, bytenr
, blocksize
);
7267 next
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
7270 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, next
,
7274 btrfs_tree_lock(next
);
7275 btrfs_set_lock_blocking(next
);
7277 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, level
- 1, 1,
7278 &wc
->refs
[level
- 1],
7279 &wc
->flags
[level
- 1]);
7281 btrfs_tree_unlock(next
);
7285 if (unlikely(wc
->refs
[level
- 1] == 0)) {
7286 btrfs_err(root
->fs_info
, "Missing references.");
7291 if (wc
->stage
== DROP_REFERENCE
) {
7292 if (wc
->refs
[level
- 1] > 1) {
7294 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
7297 if (!wc
->update_ref
||
7298 generation
<= root
->root_key
.offset
)
7301 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
7302 path
->slots
[level
]);
7303 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
7307 wc
->stage
= UPDATE_BACKREF
;
7308 wc
->shared_level
= level
- 1;
7312 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
7316 if (!btrfs_buffer_uptodate(next
, generation
, 0)) {
7317 btrfs_tree_unlock(next
);
7318 free_extent_buffer(next
);
7324 if (reada
&& level
== 1)
7325 reada_walk_down(trans
, root
, wc
, path
);
7326 next
= read_tree_block(root
, bytenr
, blocksize
, generation
);
7327 if (!next
|| !extent_buffer_uptodate(next
)) {
7328 free_extent_buffer(next
);
7331 btrfs_tree_lock(next
);
7332 btrfs_set_lock_blocking(next
);
7336 BUG_ON(level
!= btrfs_header_level(next
));
7337 path
->nodes
[level
] = next
;
7338 path
->slots
[level
] = 0;
7339 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
7345 wc
->refs
[level
- 1] = 0;
7346 wc
->flags
[level
- 1] = 0;
7347 if (wc
->stage
== DROP_REFERENCE
) {
7348 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
7349 parent
= path
->nodes
[level
]->start
;
7351 BUG_ON(root
->root_key
.objectid
!=
7352 btrfs_header_owner(path
->nodes
[level
]));
7356 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
, parent
,
7357 root
->root_key
.objectid
, level
- 1, 0, 0);
7358 BUG_ON(ret
); /* -ENOMEM */
7360 btrfs_tree_unlock(next
);
7361 free_extent_buffer(next
);
7367 * helper to process tree block while walking up the tree.
7369 * when wc->stage == DROP_REFERENCE, this function drops
7370 * reference count on the block.
7372 * when wc->stage == UPDATE_BACKREF, this function changes
7373 * wc->stage back to DROP_REFERENCE if we changed wc->stage
7374 * to UPDATE_BACKREF previously while processing the block.
7376 * NOTE: return value 1 means we should stop walking up.
7378 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
7379 struct btrfs_root
*root
,
7380 struct btrfs_path
*path
,
7381 struct walk_control
*wc
)
7384 int level
= wc
->level
;
7385 struct extent_buffer
*eb
= path
->nodes
[level
];
7388 if (wc
->stage
== UPDATE_BACKREF
) {
7389 BUG_ON(wc
->shared_level
< level
);
7390 if (level
< wc
->shared_level
)
7393 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
7397 wc
->stage
= DROP_REFERENCE
;
7398 wc
->shared_level
= -1;
7399 path
->slots
[level
] = 0;
7402 * check reference count again if the block isn't locked.
7403 * we should start walking down the tree again if reference
7406 if (!path
->locks
[level
]) {
7408 btrfs_tree_lock(eb
);
7409 btrfs_set_lock_blocking(eb
);
7410 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
7412 ret
= btrfs_lookup_extent_info(trans
, root
,
7413 eb
->start
, level
, 1,
7417 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
7418 path
->locks
[level
] = 0;
7421 BUG_ON(wc
->refs
[level
] == 0);
7422 if (wc
->refs
[level
] == 1) {
7423 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
7424 path
->locks
[level
] = 0;
7430 /* wc->stage == DROP_REFERENCE */
7431 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
7433 if (wc
->refs
[level
] == 1) {
7435 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
7436 ret
= btrfs_dec_ref(trans
, root
, eb
, 1,
7439 ret
= btrfs_dec_ref(trans
, root
, eb
, 0,
7441 BUG_ON(ret
); /* -ENOMEM */
7443 /* make block locked assertion in clean_tree_block happy */
7444 if (!path
->locks
[level
] &&
7445 btrfs_header_generation(eb
) == trans
->transid
) {
7446 btrfs_tree_lock(eb
);
7447 btrfs_set_lock_blocking(eb
);
7448 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
7450 clean_tree_block(trans
, root
, eb
);
7453 if (eb
== root
->node
) {
7454 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
7457 BUG_ON(root
->root_key
.objectid
!=
7458 btrfs_header_owner(eb
));
7460 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
7461 parent
= path
->nodes
[level
+ 1]->start
;
7463 BUG_ON(root
->root_key
.objectid
!=
7464 btrfs_header_owner(path
->nodes
[level
+ 1]));
7467 btrfs_free_tree_block(trans
, root
, eb
, parent
, wc
->refs
[level
] == 1);
7469 wc
->refs
[level
] = 0;
7470 wc
->flags
[level
] = 0;
7474 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
7475 struct btrfs_root
*root
,
7476 struct btrfs_path
*path
,
7477 struct walk_control
*wc
)
7479 int level
= wc
->level
;
7480 int lookup_info
= 1;
7483 while (level
>= 0) {
7484 ret
= walk_down_proc(trans
, root
, path
, wc
, lookup_info
);
7491 if (path
->slots
[level
] >=
7492 btrfs_header_nritems(path
->nodes
[level
]))
7495 ret
= do_walk_down(trans
, root
, path
, wc
, &lookup_info
);
7497 path
->slots
[level
]++;
7506 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
7507 struct btrfs_root
*root
,
7508 struct btrfs_path
*path
,
7509 struct walk_control
*wc
, int max_level
)
7511 int level
= wc
->level
;
7514 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
7515 while (level
< max_level
&& path
->nodes
[level
]) {
7517 if (path
->slots
[level
] + 1 <
7518 btrfs_header_nritems(path
->nodes
[level
])) {
7519 path
->slots
[level
]++;
7522 ret
= walk_up_proc(trans
, root
, path
, wc
);
7526 if (path
->locks
[level
]) {
7527 btrfs_tree_unlock_rw(path
->nodes
[level
],
7528 path
->locks
[level
]);
7529 path
->locks
[level
] = 0;
7531 free_extent_buffer(path
->nodes
[level
]);
7532 path
->nodes
[level
] = NULL
;
7540 * drop a subvolume tree.
7542 * this function traverses the tree freeing any blocks that only
7543 * referenced by the tree.
7545 * when a shared tree block is found. this function decreases its
7546 * reference count by one. if update_ref is true, this function
7547 * also make sure backrefs for the shared block and all lower level
7548 * blocks are properly updated.
7550 * If called with for_reloc == 0, may exit early with -EAGAIN
7552 int btrfs_drop_snapshot(struct btrfs_root
*root
,
7553 struct btrfs_block_rsv
*block_rsv
, int update_ref
,
7556 struct btrfs_path
*path
;
7557 struct btrfs_trans_handle
*trans
;
7558 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
7559 struct btrfs_root_item
*root_item
= &root
->root_item
;
7560 struct walk_control
*wc
;
7561 struct btrfs_key key
;
7565 bool root_dropped
= false;
7567 path
= btrfs_alloc_path();
7573 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
7575 btrfs_free_path(path
);
7580 trans
= btrfs_start_transaction(tree_root
, 0);
7581 if (IS_ERR(trans
)) {
7582 err
= PTR_ERR(trans
);
7587 trans
->block_rsv
= block_rsv
;
7589 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
7590 level
= btrfs_header_level(root
->node
);
7591 path
->nodes
[level
] = btrfs_lock_root_node(root
);
7592 btrfs_set_lock_blocking(path
->nodes
[level
]);
7593 path
->slots
[level
] = 0;
7594 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
7595 memset(&wc
->update_progress
, 0,
7596 sizeof(wc
->update_progress
));
7598 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
7599 memcpy(&wc
->update_progress
, &key
,
7600 sizeof(wc
->update_progress
));
7602 level
= root_item
->drop_level
;
7604 path
->lowest_level
= level
;
7605 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
7606 path
->lowest_level
= 0;
7614 * unlock our path, this is safe because only this
7615 * function is allowed to delete this snapshot
7617 btrfs_unlock_up_safe(path
, 0);
7619 level
= btrfs_header_level(root
->node
);
7621 btrfs_tree_lock(path
->nodes
[level
]);
7622 btrfs_set_lock_blocking(path
->nodes
[level
]);
7623 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
7625 ret
= btrfs_lookup_extent_info(trans
, root
,
7626 path
->nodes
[level
]->start
,
7627 level
, 1, &wc
->refs
[level
],
7633 BUG_ON(wc
->refs
[level
] == 0);
7635 if (level
== root_item
->drop_level
)
7638 btrfs_tree_unlock(path
->nodes
[level
]);
7639 path
->locks
[level
] = 0;
7640 WARN_ON(wc
->refs
[level
] != 1);
7646 wc
->shared_level
= -1;
7647 wc
->stage
= DROP_REFERENCE
;
7648 wc
->update_ref
= update_ref
;
7650 wc
->for_reloc
= for_reloc
;
7651 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
7655 ret
= walk_down_tree(trans
, root
, path
, wc
);
7661 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
7668 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
7672 if (wc
->stage
== DROP_REFERENCE
) {
7674 btrfs_node_key(path
->nodes
[level
],
7675 &root_item
->drop_progress
,
7676 path
->slots
[level
]);
7677 root_item
->drop_level
= level
;
7680 BUG_ON(wc
->level
== 0);
7681 if (btrfs_should_end_transaction(trans
, tree_root
) ||
7682 (!for_reloc
&& btrfs_need_cleaner_sleep(root
))) {
7683 ret
= btrfs_update_root(trans
, tree_root
,
7687 btrfs_abort_transaction(trans
, tree_root
, ret
);
7692 btrfs_end_transaction_throttle(trans
, tree_root
);
7693 if (!for_reloc
&& btrfs_need_cleaner_sleep(root
)) {
7694 pr_debug("BTRFS: drop snapshot early exit\n");
7699 trans
= btrfs_start_transaction(tree_root
, 0);
7700 if (IS_ERR(trans
)) {
7701 err
= PTR_ERR(trans
);
7705 trans
->block_rsv
= block_rsv
;
7708 btrfs_release_path(path
);
7712 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
7714 btrfs_abort_transaction(trans
, tree_root
, ret
);
7718 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
7719 ret
= btrfs_find_root(tree_root
, &root
->root_key
, path
,
7722 btrfs_abort_transaction(trans
, tree_root
, ret
);
7725 } else if (ret
> 0) {
7726 /* if we fail to delete the orphan item this time
7727 * around, it'll get picked up the next time.
7729 * The most common failure here is just -ENOENT.
7731 btrfs_del_orphan_item(trans
, tree_root
,
7732 root
->root_key
.objectid
);
7736 if (root
->in_radix
) {
7737 btrfs_drop_and_free_fs_root(tree_root
->fs_info
, root
);
7739 free_extent_buffer(root
->node
);
7740 free_extent_buffer(root
->commit_root
);
7741 btrfs_put_fs_root(root
);
7743 root_dropped
= true;
7745 btrfs_end_transaction_throttle(trans
, tree_root
);
7748 btrfs_free_path(path
);
7751 * So if we need to stop dropping the snapshot for whatever reason we
7752 * need to make sure to add it back to the dead root list so that we
7753 * keep trying to do the work later. This also cleans up roots if we
7754 * don't have it in the radix (like when we recover after a power fail
7755 * or unmount) so we don't leak memory.
7757 if (!for_reloc
&& root_dropped
== false)
7758 btrfs_add_dead_root(root
);
7759 if (err
&& err
!= -EAGAIN
)
7760 btrfs_std_error(root
->fs_info
, err
);
7765 * drop subtree rooted at tree block 'node'.
7767 * NOTE: this function will unlock and release tree block 'node'
7768 * only used by relocation code
7770 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
7771 struct btrfs_root
*root
,
7772 struct extent_buffer
*node
,
7773 struct extent_buffer
*parent
)
7775 struct btrfs_path
*path
;
7776 struct walk_control
*wc
;
7782 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
7784 path
= btrfs_alloc_path();
7788 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
7790 btrfs_free_path(path
);
7794 btrfs_assert_tree_locked(parent
);
7795 parent_level
= btrfs_header_level(parent
);
7796 extent_buffer_get(parent
);
7797 path
->nodes
[parent_level
] = parent
;
7798 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
7800 btrfs_assert_tree_locked(node
);
7801 level
= btrfs_header_level(node
);
7802 path
->nodes
[level
] = node
;
7803 path
->slots
[level
] = 0;
7804 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
7806 wc
->refs
[parent_level
] = 1;
7807 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
7809 wc
->shared_level
= -1;
7810 wc
->stage
= DROP_REFERENCE
;
7814 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
7817 wret
= walk_down_tree(trans
, root
, path
, wc
);
7823 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
7831 btrfs_free_path(path
);
7835 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
7841 * if restripe for this chunk_type is on pick target profile and
7842 * return, otherwise do the usual balance
7844 stripped
= get_restripe_target(root
->fs_info
, flags
);
7846 return extended_to_chunk(stripped
);
7849 * we add in the count of missing devices because we want
7850 * to make sure that any RAID levels on a degraded FS
7851 * continue to be honored.
7853 num_devices
= root
->fs_info
->fs_devices
->rw_devices
+
7854 root
->fs_info
->fs_devices
->missing_devices
;
7856 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
7857 BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
|
7858 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
7860 if (num_devices
== 1) {
7861 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
7862 stripped
= flags
& ~stripped
;
7864 /* turn raid0 into single device chunks */
7865 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
7868 /* turn mirroring into duplication */
7869 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
7870 BTRFS_BLOCK_GROUP_RAID10
))
7871 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
7873 /* they already had raid on here, just return */
7874 if (flags
& stripped
)
7877 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
7878 stripped
= flags
& ~stripped
;
7880 /* switch duplicated blocks with raid1 */
7881 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
7882 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
7884 /* this is drive concat, leave it alone */
7890 static int set_block_group_ro(struct btrfs_block_group_cache
*cache
, int force
)
7892 struct btrfs_space_info
*sinfo
= cache
->space_info
;
7894 u64 min_allocable_bytes
;
7899 * We need some metadata space and system metadata space for
7900 * allocating chunks in some corner cases until we force to set
7901 * it to be readonly.
7904 (BTRFS_BLOCK_GROUP_SYSTEM
| BTRFS_BLOCK_GROUP_METADATA
)) &&
7906 min_allocable_bytes
= 1 * 1024 * 1024;
7908 min_allocable_bytes
= 0;
7910 spin_lock(&sinfo
->lock
);
7911 spin_lock(&cache
->lock
);
7918 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
7919 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
7921 if (sinfo
->bytes_used
+ sinfo
->bytes_reserved
+ sinfo
->bytes_pinned
+
7922 sinfo
->bytes_may_use
+ sinfo
->bytes_readonly
+ num_bytes
+
7923 min_allocable_bytes
<= sinfo
->total_bytes
) {
7924 sinfo
->bytes_readonly
+= num_bytes
;
7929 spin_unlock(&cache
->lock
);
7930 spin_unlock(&sinfo
->lock
);
7934 int btrfs_set_block_group_ro(struct btrfs_root
*root
,
7935 struct btrfs_block_group_cache
*cache
)
7938 struct btrfs_trans_handle
*trans
;
7944 trans
= btrfs_join_transaction(root
);
7946 return PTR_ERR(trans
);
7948 alloc_flags
= update_block_group_flags(root
, cache
->flags
);
7949 if (alloc_flags
!= cache
->flags
) {
7950 ret
= do_chunk_alloc(trans
, root
, alloc_flags
,
7956 ret
= set_block_group_ro(cache
, 0);
7959 alloc_flags
= get_alloc_profile(root
, cache
->space_info
->flags
);
7960 ret
= do_chunk_alloc(trans
, root
, alloc_flags
,
7964 ret
= set_block_group_ro(cache
, 0);
7966 btrfs_end_transaction(trans
, root
);
7970 int btrfs_force_chunk_alloc(struct btrfs_trans_handle
*trans
,
7971 struct btrfs_root
*root
, u64 type
)
7973 u64 alloc_flags
= get_alloc_profile(root
, type
);
7974 return do_chunk_alloc(trans
, root
, alloc_flags
,
7979 * helper to account the unused space of all the readonly block group in the
7980 * list. takes mirrors into account.
7982 static u64
__btrfs_get_ro_block_group_free_space(struct list_head
*groups_list
)
7984 struct btrfs_block_group_cache
*block_group
;
7988 list_for_each_entry(block_group
, groups_list
, list
) {
7989 spin_lock(&block_group
->lock
);
7991 if (!block_group
->ro
) {
7992 spin_unlock(&block_group
->lock
);
7996 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_RAID1
|
7997 BTRFS_BLOCK_GROUP_RAID10
|
7998 BTRFS_BLOCK_GROUP_DUP
))
8003 free_bytes
+= (block_group
->key
.offset
-
8004 btrfs_block_group_used(&block_group
->item
)) *
8007 spin_unlock(&block_group
->lock
);
8014 * helper to account the unused space of all the readonly block group in the
8015 * space_info. takes mirrors into account.
8017 u64
btrfs_account_ro_block_groups_free_space(struct btrfs_space_info
*sinfo
)
8022 spin_lock(&sinfo
->lock
);
8024 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
8025 if (!list_empty(&sinfo
->block_groups
[i
]))
8026 free_bytes
+= __btrfs_get_ro_block_group_free_space(
8027 &sinfo
->block_groups
[i
]);
8029 spin_unlock(&sinfo
->lock
);
8034 void btrfs_set_block_group_rw(struct btrfs_root
*root
,
8035 struct btrfs_block_group_cache
*cache
)
8037 struct btrfs_space_info
*sinfo
= cache
->space_info
;
8042 spin_lock(&sinfo
->lock
);
8043 spin_lock(&cache
->lock
);
8044 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
8045 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
8046 sinfo
->bytes_readonly
-= num_bytes
;
8048 spin_unlock(&cache
->lock
);
8049 spin_unlock(&sinfo
->lock
);
8053 * checks to see if its even possible to relocate this block group.
8055 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8056 * ok to go ahead and try.
8058 int btrfs_can_relocate(struct btrfs_root
*root
, u64 bytenr
)
8060 struct btrfs_block_group_cache
*block_group
;
8061 struct btrfs_space_info
*space_info
;
8062 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
8063 struct btrfs_device
*device
;
8064 struct btrfs_trans_handle
*trans
;
8073 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
8075 /* odd, couldn't find the block group, leave it alone */
8079 min_free
= btrfs_block_group_used(&block_group
->item
);
8081 /* no bytes used, we're good */
8085 space_info
= block_group
->space_info
;
8086 spin_lock(&space_info
->lock
);
8088 full
= space_info
->full
;
8091 * if this is the last block group we have in this space, we can't
8092 * relocate it unless we're able to allocate a new chunk below.
8094 * Otherwise, we need to make sure we have room in the space to handle
8095 * all of the extents from this block group. If we can, we're good
8097 if ((space_info
->total_bytes
!= block_group
->key
.offset
) &&
8098 (space_info
->bytes_used
+ space_info
->bytes_reserved
+
8099 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
8100 min_free
< space_info
->total_bytes
)) {
8101 spin_unlock(&space_info
->lock
);
8104 spin_unlock(&space_info
->lock
);
8107 * ok we don't have enough space, but maybe we have free space on our
8108 * devices to allocate new chunks for relocation, so loop through our
8109 * alloc devices and guess if we have enough space. if this block
8110 * group is going to be restriped, run checks against the target
8111 * profile instead of the current one.
8123 target
= get_restripe_target(root
->fs_info
, block_group
->flags
);
8125 index
= __get_raid_index(extended_to_chunk(target
));
8128 * this is just a balance, so if we were marked as full
8129 * we know there is no space for a new chunk
8134 index
= get_block_group_index(block_group
);
8137 if (index
== BTRFS_RAID_RAID10
) {
8141 } else if (index
== BTRFS_RAID_RAID1
) {
8143 } else if (index
== BTRFS_RAID_DUP
) {
8146 } else if (index
== BTRFS_RAID_RAID0
) {
8147 dev_min
= fs_devices
->rw_devices
;
8148 do_div(min_free
, dev_min
);
8151 /* We need to do this so that we can look at pending chunks */
8152 trans
= btrfs_join_transaction(root
);
8153 if (IS_ERR(trans
)) {
8154 ret
= PTR_ERR(trans
);
8158 mutex_lock(&root
->fs_info
->chunk_mutex
);
8159 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
8163 * check to make sure we can actually find a chunk with enough
8164 * space to fit our block group in.
8166 if (device
->total_bytes
> device
->bytes_used
+ min_free
&&
8167 !device
->is_tgtdev_for_dev_replace
) {
8168 ret
= find_free_dev_extent(trans
, device
, min_free
,
8173 if (dev_nr
>= dev_min
)
8179 mutex_unlock(&root
->fs_info
->chunk_mutex
);
8180 btrfs_end_transaction(trans
, root
);
8182 btrfs_put_block_group(block_group
);
8186 static int find_first_block_group(struct btrfs_root
*root
,
8187 struct btrfs_path
*path
, struct btrfs_key
*key
)
8190 struct btrfs_key found_key
;
8191 struct extent_buffer
*leaf
;
8194 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
8199 slot
= path
->slots
[0];
8200 leaf
= path
->nodes
[0];
8201 if (slot
>= btrfs_header_nritems(leaf
)) {
8202 ret
= btrfs_next_leaf(root
, path
);
8209 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
8211 if (found_key
.objectid
>= key
->objectid
&&
8212 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
8222 void btrfs_put_block_group_cache(struct btrfs_fs_info
*info
)
8224 struct btrfs_block_group_cache
*block_group
;
8228 struct inode
*inode
;
8230 block_group
= btrfs_lookup_first_block_group(info
, last
);
8231 while (block_group
) {
8232 spin_lock(&block_group
->lock
);
8233 if (block_group
->iref
)
8235 spin_unlock(&block_group
->lock
);
8236 block_group
= next_block_group(info
->tree_root
,
8246 inode
= block_group
->inode
;
8247 block_group
->iref
= 0;
8248 block_group
->inode
= NULL
;
8249 spin_unlock(&block_group
->lock
);
8251 last
= block_group
->key
.objectid
+ block_group
->key
.offset
;
8252 btrfs_put_block_group(block_group
);
8256 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
8258 struct btrfs_block_group_cache
*block_group
;
8259 struct btrfs_space_info
*space_info
;
8260 struct btrfs_caching_control
*caching_ctl
;
8263 down_write(&info
->commit_root_sem
);
8264 while (!list_empty(&info
->caching_block_groups
)) {
8265 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
8266 struct btrfs_caching_control
, list
);
8267 list_del(&caching_ctl
->list
);
8268 put_caching_control(caching_ctl
);
8270 up_write(&info
->commit_root_sem
);
8272 spin_lock(&info
->block_group_cache_lock
);
8273 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
8274 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
8276 rb_erase(&block_group
->cache_node
,
8277 &info
->block_group_cache_tree
);
8278 spin_unlock(&info
->block_group_cache_lock
);
8280 down_write(&block_group
->space_info
->groups_sem
);
8281 list_del(&block_group
->list
);
8282 up_write(&block_group
->space_info
->groups_sem
);
8284 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
8285 wait_block_group_cache_done(block_group
);
8288 * We haven't cached this block group, which means we could
8289 * possibly have excluded extents on this block group.
8291 if (block_group
->cached
== BTRFS_CACHE_NO
||
8292 block_group
->cached
== BTRFS_CACHE_ERROR
)
8293 free_excluded_extents(info
->extent_root
, block_group
);
8295 btrfs_remove_free_space_cache(block_group
);
8296 btrfs_put_block_group(block_group
);
8298 spin_lock(&info
->block_group_cache_lock
);
8300 spin_unlock(&info
->block_group_cache_lock
);
8302 /* now that all the block groups are freed, go through and
8303 * free all the space_info structs. This is only called during
8304 * the final stages of unmount, and so we know nobody is
8305 * using them. We call synchronize_rcu() once before we start,
8306 * just to be on the safe side.
8310 release_global_block_rsv(info
);
8312 while (!list_empty(&info
->space_info
)) {
8315 space_info
= list_entry(info
->space_info
.next
,
8316 struct btrfs_space_info
,
8318 if (btrfs_test_opt(info
->tree_root
, ENOSPC_DEBUG
)) {
8319 if (WARN_ON(space_info
->bytes_pinned
> 0 ||
8320 space_info
->bytes_reserved
> 0 ||
8321 space_info
->bytes_may_use
> 0)) {
8322 dump_space_info(space_info
, 0, 0);
8325 list_del(&space_info
->list
);
8326 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++) {
8327 struct kobject
*kobj
;
8328 kobj
= &space_info
->block_group_kobjs
[i
];
8334 kobject_del(&space_info
->kobj
);
8335 kobject_put(&space_info
->kobj
);
8340 static void __link_block_group(struct btrfs_space_info
*space_info
,
8341 struct btrfs_block_group_cache
*cache
)
8343 int index
= get_block_group_index(cache
);
8346 down_write(&space_info
->groups_sem
);
8347 if (list_empty(&space_info
->block_groups
[index
]))
8349 list_add_tail(&cache
->list
, &space_info
->block_groups
[index
]);
8350 up_write(&space_info
->groups_sem
);
8353 struct kobject
*kobj
= &space_info
->block_group_kobjs
[index
];
8356 kobject_get(&space_info
->kobj
); /* put in release */
8357 ret
= kobject_add(kobj
, &space_info
->kobj
, "%s",
8358 get_raid_name(index
));
8360 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
8361 kobject_put(&space_info
->kobj
);
8366 static struct btrfs_block_group_cache
*
8367 btrfs_create_block_group_cache(struct btrfs_root
*root
, u64 start
, u64 size
)
8369 struct btrfs_block_group_cache
*cache
;
8371 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
8375 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
8377 if (!cache
->free_space_ctl
) {
8382 cache
->key
.objectid
= start
;
8383 cache
->key
.offset
= size
;
8384 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
8386 cache
->sectorsize
= root
->sectorsize
;
8387 cache
->fs_info
= root
->fs_info
;
8388 cache
->full_stripe_len
= btrfs_full_stripe_len(root
,
8389 &root
->fs_info
->mapping_tree
,
8391 atomic_set(&cache
->count
, 1);
8392 spin_lock_init(&cache
->lock
);
8393 INIT_LIST_HEAD(&cache
->list
);
8394 INIT_LIST_HEAD(&cache
->cluster_list
);
8395 INIT_LIST_HEAD(&cache
->new_bg_list
);
8396 btrfs_init_free_space_ctl(cache
);
8401 int btrfs_read_block_groups(struct btrfs_root
*root
)
8403 struct btrfs_path
*path
;
8405 struct btrfs_block_group_cache
*cache
;
8406 struct btrfs_fs_info
*info
= root
->fs_info
;
8407 struct btrfs_space_info
*space_info
;
8408 struct btrfs_key key
;
8409 struct btrfs_key found_key
;
8410 struct extent_buffer
*leaf
;
8414 root
= info
->extent_root
;
8417 btrfs_set_key_type(&key
, BTRFS_BLOCK_GROUP_ITEM_KEY
);
8418 path
= btrfs_alloc_path();
8423 cache_gen
= btrfs_super_cache_generation(root
->fs_info
->super_copy
);
8424 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
8425 btrfs_super_generation(root
->fs_info
->super_copy
) != cache_gen
)
8427 if (btrfs_test_opt(root
, CLEAR_CACHE
))
8431 ret
= find_first_block_group(root
, path
, &key
);
8437 leaf
= path
->nodes
[0];
8438 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
8440 cache
= btrfs_create_block_group_cache(root
, found_key
.objectid
,
8449 * When we mount with old space cache, we need to
8450 * set BTRFS_DC_CLEAR and set dirty flag.
8452 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
8453 * truncate the old free space cache inode and
8455 * b) Setting 'dirty flag' makes sure that we flush
8456 * the new space cache info onto disk.
8458 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
8459 if (btrfs_test_opt(root
, SPACE_CACHE
))
8463 read_extent_buffer(leaf
, &cache
->item
,
8464 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
8465 sizeof(cache
->item
));
8466 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
8468 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
8469 btrfs_release_path(path
);
8472 * We need to exclude the super stripes now so that the space
8473 * info has super bytes accounted for, otherwise we'll think
8474 * we have more space than we actually do.
8476 ret
= exclude_super_stripes(root
, cache
);
8479 * We may have excluded something, so call this just in
8482 free_excluded_extents(root
, cache
);
8483 btrfs_put_block_group(cache
);
8488 * check for two cases, either we are full, and therefore
8489 * don't need to bother with the caching work since we won't
8490 * find any space, or we are empty, and we can just add all
8491 * the space in and be done with it. This saves us _alot_ of
8492 * time, particularly in the full case.
8494 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
8495 cache
->last_byte_to_unpin
= (u64
)-1;
8496 cache
->cached
= BTRFS_CACHE_FINISHED
;
8497 free_excluded_extents(root
, cache
);
8498 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
8499 cache
->last_byte_to_unpin
= (u64
)-1;
8500 cache
->cached
= BTRFS_CACHE_FINISHED
;
8501 add_new_free_space(cache
, root
->fs_info
,
8503 found_key
.objectid
+
8505 free_excluded_extents(root
, cache
);
8508 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
8510 btrfs_remove_free_space_cache(cache
);
8511 btrfs_put_block_group(cache
);
8515 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
8516 btrfs_block_group_used(&cache
->item
),
8519 btrfs_remove_free_space_cache(cache
);
8520 spin_lock(&info
->block_group_cache_lock
);
8521 rb_erase(&cache
->cache_node
,
8522 &info
->block_group_cache_tree
);
8523 spin_unlock(&info
->block_group_cache_lock
);
8524 btrfs_put_block_group(cache
);
8528 cache
->space_info
= space_info
;
8529 spin_lock(&cache
->space_info
->lock
);
8530 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
8531 spin_unlock(&cache
->space_info
->lock
);
8533 __link_block_group(space_info
, cache
);
8535 set_avail_alloc_bits(root
->fs_info
, cache
->flags
);
8536 if (btrfs_chunk_readonly(root
, cache
->key
.objectid
))
8537 set_block_group_ro(cache
, 1);
8540 list_for_each_entry_rcu(space_info
, &root
->fs_info
->space_info
, list
) {
8541 if (!(get_alloc_profile(root
, space_info
->flags
) &
8542 (BTRFS_BLOCK_GROUP_RAID10
|
8543 BTRFS_BLOCK_GROUP_RAID1
|
8544 BTRFS_BLOCK_GROUP_RAID5
|
8545 BTRFS_BLOCK_GROUP_RAID6
|
8546 BTRFS_BLOCK_GROUP_DUP
)))
8549 * avoid allocating from un-mirrored block group if there are
8550 * mirrored block groups.
8552 list_for_each_entry(cache
,
8553 &space_info
->block_groups
[BTRFS_RAID_RAID0
],
8555 set_block_group_ro(cache
, 1);
8556 list_for_each_entry(cache
,
8557 &space_info
->block_groups
[BTRFS_RAID_SINGLE
],
8559 set_block_group_ro(cache
, 1);
8562 init_global_block_rsv(info
);
8565 btrfs_free_path(path
);
8569 void btrfs_create_pending_block_groups(struct btrfs_trans_handle
*trans
,
8570 struct btrfs_root
*root
)
8572 struct btrfs_block_group_cache
*block_group
, *tmp
;
8573 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
8574 struct btrfs_block_group_item item
;
8575 struct btrfs_key key
;
8578 list_for_each_entry_safe(block_group
, tmp
, &trans
->new_bgs
,
8580 list_del_init(&block_group
->new_bg_list
);
8585 spin_lock(&block_group
->lock
);
8586 memcpy(&item
, &block_group
->item
, sizeof(item
));
8587 memcpy(&key
, &block_group
->key
, sizeof(key
));
8588 spin_unlock(&block_group
->lock
);
8590 ret
= btrfs_insert_item(trans
, extent_root
, &key
, &item
,
8593 btrfs_abort_transaction(trans
, extent_root
, ret
);
8594 ret
= btrfs_finish_chunk_alloc(trans
, extent_root
,
8595 key
.objectid
, key
.offset
);
8597 btrfs_abort_transaction(trans
, extent_root
, ret
);
8601 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
8602 struct btrfs_root
*root
, u64 bytes_used
,
8603 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
8607 struct btrfs_root
*extent_root
;
8608 struct btrfs_block_group_cache
*cache
;
8610 extent_root
= root
->fs_info
->extent_root
;
8612 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
8614 cache
= btrfs_create_block_group_cache(root
, chunk_offset
, size
);
8618 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
8619 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
8620 btrfs_set_block_group_flags(&cache
->item
, type
);
8622 cache
->flags
= type
;
8623 cache
->last_byte_to_unpin
= (u64
)-1;
8624 cache
->cached
= BTRFS_CACHE_FINISHED
;
8625 ret
= exclude_super_stripes(root
, cache
);
8628 * We may have excluded something, so call this just in
8631 free_excluded_extents(root
, cache
);
8632 btrfs_put_block_group(cache
);
8636 add_new_free_space(cache
, root
->fs_info
, chunk_offset
,
8637 chunk_offset
+ size
);
8639 free_excluded_extents(root
, cache
);
8641 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
8643 btrfs_remove_free_space_cache(cache
);
8644 btrfs_put_block_group(cache
);
8648 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
8649 &cache
->space_info
);
8651 btrfs_remove_free_space_cache(cache
);
8652 spin_lock(&root
->fs_info
->block_group_cache_lock
);
8653 rb_erase(&cache
->cache_node
,
8654 &root
->fs_info
->block_group_cache_tree
);
8655 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
8656 btrfs_put_block_group(cache
);
8659 update_global_block_rsv(root
->fs_info
);
8661 spin_lock(&cache
->space_info
->lock
);
8662 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
8663 spin_unlock(&cache
->space_info
->lock
);
8665 __link_block_group(cache
->space_info
, cache
);
8667 list_add_tail(&cache
->new_bg_list
, &trans
->new_bgs
);
8669 set_avail_alloc_bits(extent_root
->fs_info
, type
);
8674 static void clear_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
8676 u64 extra_flags
= chunk_to_extended(flags
) &
8677 BTRFS_EXTENDED_PROFILE_MASK
;
8679 write_seqlock(&fs_info
->profiles_lock
);
8680 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
8681 fs_info
->avail_data_alloc_bits
&= ~extra_flags
;
8682 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
8683 fs_info
->avail_metadata_alloc_bits
&= ~extra_flags
;
8684 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
8685 fs_info
->avail_system_alloc_bits
&= ~extra_flags
;
8686 write_sequnlock(&fs_info
->profiles_lock
);
8689 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
8690 struct btrfs_root
*root
, u64 group_start
)
8692 struct btrfs_path
*path
;
8693 struct btrfs_block_group_cache
*block_group
;
8694 struct btrfs_free_cluster
*cluster
;
8695 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
8696 struct btrfs_key key
;
8697 struct inode
*inode
;
8702 root
= root
->fs_info
->extent_root
;
8704 block_group
= btrfs_lookup_block_group(root
->fs_info
, group_start
);
8705 BUG_ON(!block_group
);
8706 BUG_ON(!block_group
->ro
);
8709 * Free the reserved super bytes from this block group before
8712 free_excluded_extents(root
, block_group
);
8714 memcpy(&key
, &block_group
->key
, sizeof(key
));
8715 index
= get_block_group_index(block_group
);
8716 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
8717 BTRFS_BLOCK_GROUP_RAID1
|
8718 BTRFS_BLOCK_GROUP_RAID10
))
8723 /* make sure this block group isn't part of an allocation cluster */
8724 cluster
= &root
->fs_info
->data_alloc_cluster
;
8725 spin_lock(&cluster
->refill_lock
);
8726 btrfs_return_cluster_to_free_space(block_group
, cluster
);
8727 spin_unlock(&cluster
->refill_lock
);
8730 * make sure this block group isn't part of a metadata
8731 * allocation cluster
8733 cluster
= &root
->fs_info
->meta_alloc_cluster
;
8734 spin_lock(&cluster
->refill_lock
);
8735 btrfs_return_cluster_to_free_space(block_group
, cluster
);
8736 spin_unlock(&cluster
->refill_lock
);
8738 path
= btrfs_alloc_path();
8744 inode
= lookup_free_space_inode(tree_root
, block_group
, path
);
8745 if (!IS_ERR(inode
)) {
8746 ret
= btrfs_orphan_add(trans
, inode
);
8748 btrfs_add_delayed_iput(inode
);
8752 /* One for the block groups ref */
8753 spin_lock(&block_group
->lock
);
8754 if (block_group
->iref
) {
8755 block_group
->iref
= 0;
8756 block_group
->inode
= NULL
;
8757 spin_unlock(&block_group
->lock
);
8760 spin_unlock(&block_group
->lock
);
8762 /* One for our lookup ref */
8763 btrfs_add_delayed_iput(inode
);
8766 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
8767 key
.offset
= block_group
->key
.objectid
;
8770 ret
= btrfs_search_slot(trans
, tree_root
, &key
, path
, -1, 1);
8774 btrfs_release_path(path
);
8776 ret
= btrfs_del_item(trans
, tree_root
, path
);
8779 btrfs_release_path(path
);
8782 spin_lock(&root
->fs_info
->block_group_cache_lock
);
8783 rb_erase(&block_group
->cache_node
,
8784 &root
->fs_info
->block_group_cache_tree
);
8786 if (root
->fs_info
->first_logical_byte
== block_group
->key
.objectid
)
8787 root
->fs_info
->first_logical_byte
= (u64
)-1;
8788 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
8790 down_write(&block_group
->space_info
->groups_sem
);
8792 * we must use list_del_init so people can check to see if they
8793 * are still on the list after taking the semaphore
8795 list_del_init(&block_group
->list
);
8796 if (list_empty(&block_group
->space_info
->block_groups
[index
])) {
8797 kobject_del(&block_group
->space_info
->block_group_kobjs
[index
]);
8798 kobject_put(&block_group
->space_info
->block_group_kobjs
[index
]);
8799 clear_avail_alloc_bits(root
->fs_info
, block_group
->flags
);
8801 up_write(&block_group
->space_info
->groups_sem
);
8803 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
8804 wait_block_group_cache_done(block_group
);
8806 btrfs_remove_free_space_cache(block_group
);
8808 spin_lock(&block_group
->space_info
->lock
);
8809 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
8810 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
8811 block_group
->space_info
->disk_total
-= block_group
->key
.offset
* factor
;
8812 spin_unlock(&block_group
->space_info
->lock
);
8814 memcpy(&key
, &block_group
->key
, sizeof(key
));
8816 btrfs_clear_space_info_full(root
->fs_info
);
8818 btrfs_put_block_group(block_group
);
8819 btrfs_put_block_group(block_group
);
8821 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
8827 ret
= btrfs_del_item(trans
, root
, path
);
8829 btrfs_free_path(path
);
8833 int btrfs_init_space_info(struct btrfs_fs_info
*fs_info
)
8835 struct btrfs_space_info
*space_info
;
8836 struct btrfs_super_block
*disk_super
;
8842 disk_super
= fs_info
->super_copy
;
8843 if (!btrfs_super_root(disk_super
))
8846 features
= btrfs_super_incompat_flags(disk_super
);
8847 if (features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
8850 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
8851 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
8856 flags
= BTRFS_BLOCK_GROUP_METADATA
| BTRFS_BLOCK_GROUP_DATA
;
8857 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
8859 flags
= BTRFS_BLOCK_GROUP_METADATA
;
8860 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
8864 flags
= BTRFS_BLOCK_GROUP_DATA
;
8865 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
8871 int btrfs_error_unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
8873 return unpin_extent_range(root
, start
, end
);
8876 int btrfs_error_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
8877 u64 num_bytes
, u64
*actual_bytes
)
8879 return btrfs_discard_extent(root
, bytenr
, num_bytes
, actual_bytes
);
8882 int btrfs_trim_fs(struct btrfs_root
*root
, struct fstrim_range
*range
)
8884 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8885 struct btrfs_block_group_cache
*cache
= NULL
;
8890 u64 total_bytes
= btrfs_super_total_bytes(fs_info
->super_copy
);
8894 * try to trim all FS space, our block group may start from non-zero.
8896 if (range
->len
== total_bytes
)
8897 cache
= btrfs_lookup_first_block_group(fs_info
, range
->start
);
8899 cache
= btrfs_lookup_block_group(fs_info
, range
->start
);
8902 if (cache
->key
.objectid
>= (range
->start
+ range
->len
)) {
8903 btrfs_put_block_group(cache
);
8907 start
= max(range
->start
, cache
->key
.objectid
);
8908 end
= min(range
->start
+ range
->len
,
8909 cache
->key
.objectid
+ cache
->key
.offset
);
8911 if (end
- start
>= range
->minlen
) {
8912 if (!block_group_cache_done(cache
)) {
8913 ret
= cache_block_group(cache
, 0);
8915 btrfs_put_block_group(cache
);
8918 ret
= wait_block_group_cache_done(cache
);
8920 btrfs_put_block_group(cache
);
8924 ret
= btrfs_trim_block_group(cache
,
8930 trimmed
+= group_trimmed
;
8932 btrfs_put_block_group(cache
);
8937 cache
= next_block_group(fs_info
->tree_root
, cache
);
8940 range
->len
= trimmed
;
8945 * btrfs_{start,end}_write() is similar to mnt_{want, drop}_write(),
8946 * they are used to prevent the some tasks writing data into the page cache
8947 * by nocow before the subvolume is snapshoted, but flush the data into
8948 * the disk after the snapshot creation.
8950 void btrfs_end_nocow_write(struct btrfs_root
*root
)
8952 percpu_counter_dec(&root
->subv_writers
->counter
);
8954 * Make sure counter is updated before we wake up
8958 if (waitqueue_active(&root
->subv_writers
->wait
))
8959 wake_up(&root
->subv_writers
->wait
);
8962 int btrfs_start_nocow_write(struct btrfs_root
*root
)
8964 if (unlikely(atomic_read(&root
->will_be_snapshoted
)))
8967 percpu_counter_inc(&root
->subv_writers
->counter
);
8969 * Make sure counter is updated before we check for snapshot creation.
8972 if (unlikely(atomic_read(&root
->will_be_snapshoted
))) {
8973 btrfs_end_nocow_write(root
);