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"
35 #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_trans_handle
*trans
,
78 struct btrfs_root
*root
, u64 bytenr
,
79 u64 num_bytes
, int alloc
);
80 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
81 struct btrfs_root
*root
,
82 u64 bytenr
, u64 num_bytes
, u64 parent
,
83 u64 root_objectid
, u64 owner_objectid
,
84 u64 owner_offset
, int refs_to_drop
,
85 struct btrfs_delayed_extent_op
*extra_op
,
87 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
88 struct extent_buffer
*leaf
,
89 struct btrfs_extent_item
*ei
);
90 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
91 struct btrfs_root
*root
,
92 u64 parent
, u64 root_objectid
,
93 u64 flags
, u64 owner
, u64 offset
,
94 struct btrfs_key
*ins
, int ref_mod
);
95 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
96 struct btrfs_root
*root
,
97 u64 parent
, u64 root_objectid
,
98 u64 flags
, struct btrfs_disk_key
*key
,
99 int level
, struct btrfs_key
*ins
,
101 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
102 struct btrfs_root
*extent_root
, u64 flags
,
104 static int find_next_key(struct btrfs_path
*path
, int level
,
105 struct btrfs_key
*key
);
106 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
107 int dump_block_groups
);
108 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
109 u64 num_bytes
, int reserve
,
111 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
113 int btrfs_pin_extent(struct btrfs_root
*root
,
114 u64 bytenr
, u64 num_bytes
, int reserved
);
117 block_group_cache_done(struct btrfs_block_group_cache
*cache
)
120 return cache
->cached
== BTRFS_CACHE_FINISHED
||
121 cache
->cached
== BTRFS_CACHE_ERROR
;
124 static int block_group_bits(struct btrfs_block_group_cache
*cache
, u64 bits
)
126 return (cache
->flags
& bits
) == bits
;
129 static void btrfs_get_block_group(struct btrfs_block_group_cache
*cache
)
131 atomic_inc(&cache
->count
);
134 void btrfs_put_block_group(struct btrfs_block_group_cache
*cache
)
136 if (atomic_dec_and_test(&cache
->count
)) {
137 WARN_ON(cache
->pinned
> 0);
138 WARN_ON(cache
->reserved
> 0);
139 kfree(cache
->free_space_ctl
);
145 * this adds the block group to the fs_info rb tree for the block group
148 static int btrfs_add_block_group_cache(struct btrfs_fs_info
*info
,
149 struct btrfs_block_group_cache
*block_group
)
152 struct rb_node
*parent
= NULL
;
153 struct btrfs_block_group_cache
*cache
;
155 spin_lock(&info
->block_group_cache_lock
);
156 p
= &info
->block_group_cache_tree
.rb_node
;
160 cache
= rb_entry(parent
, struct btrfs_block_group_cache
,
162 if (block_group
->key
.objectid
< cache
->key
.objectid
) {
164 } else if (block_group
->key
.objectid
> cache
->key
.objectid
) {
167 spin_unlock(&info
->block_group_cache_lock
);
172 rb_link_node(&block_group
->cache_node
, parent
, p
);
173 rb_insert_color(&block_group
->cache_node
,
174 &info
->block_group_cache_tree
);
176 if (info
->first_logical_byte
> block_group
->key
.objectid
)
177 info
->first_logical_byte
= block_group
->key
.objectid
;
179 spin_unlock(&info
->block_group_cache_lock
);
185 * This will return the block group at or after bytenr if contains is 0, else
186 * it will return the block group that contains the bytenr
188 static struct btrfs_block_group_cache
*
189 block_group_cache_tree_search(struct btrfs_fs_info
*info
, u64 bytenr
,
192 struct btrfs_block_group_cache
*cache
, *ret
= NULL
;
196 spin_lock(&info
->block_group_cache_lock
);
197 n
= info
->block_group_cache_tree
.rb_node
;
200 cache
= rb_entry(n
, struct btrfs_block_group_cache
,
202 end
= cache
->key
.objectid
+ cache
->key
.offset
- 1;
203 start
= cache
->key
.objectid
;
205 if (bytenr
< start
) {
206 if (!contains
&& (!ret
|| start
< ret
->key
.objectid
))
209 } else if (bytenr
> start
) {
210 if (contains
&& bytenr
<= end
) {
221 btrfs_get_block_group(ret
);
222 if (bytenr
== 0 && info
->first_logical_byte
> ret
->key
.objectid
)
223 info
->first_logical_byte
= ret
->key
.objectid
;
225 spin_unlock(&info
->block_group_cache_lock
);
230 static int add_excluded_extent(struct btrfs_root
*root
,
231 u64 start
, u64 num_bytes
)
233 u64 end
= start
+ num_bytes
- 1;
234 set_extent_bits(&root
->fs_info
->freed_extents
[0],
235 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
236 set_extent_bits(&root
->fs_info
->freed_extents
[1],
237 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
241 static void free_excluded_extents(struct btrfs_root
*root
,
242 struct btrfs_block_group_cache
*cache
)
246 start
= cache
->key
.objectid
;
247 end
= start
+ cache
->key
.offset
- 1;
249 clear_extent_bits(&root
->fs_info
->freed_extents
[0],
250 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
251 clear_extent_bits(&root
->fs_info
->freed_extents
[1],
252 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
255 static int exclude_super_stripes(struct btrfs_root
*root
,
256 struct btrfs_block_group_cache
*cache
)
263 if (cache
->key
.objectid
< BTRFS_SUPER_INFO_OFFSET
) {
264 stripe_len
= BTRFS_SUPER_INFO_OFFSET
- cache
->key
.objectid
;
265 cache
->bytes_super
+= stripe_len
;
266 ret
= add_excluded_extent(root
, cache
->key
.objectid
,
272 for (i
= 0; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
273 bytenr
= btrfs_sb_offset(i
);
274 ret
= btrfs_rmap_block(&root
->fs_info
->mapping_tree
,
275 cache
->key
.objectid
, bytenr
,
276 0, &logical
, &nr
, &stripe_len
);
283 if (logical
[nr
] > cache
->key
.objectid
+
287 if (logical
[nr
] + stripe_len
<= cache
->key
.objectid
)
291 if (start
< cache
->key
.objectid
) {
292 start
= cache
->key
.objectid
;
293 len
= (logical
[nr
] + stripe_len
) - start
;
295 len
= min_t(u64
, stripe_len
,
296 cache
->key
.objectid
+
297 cache
->key
.offset
- start
);
300 cache
->bytes_super
+= len
;
301 ret
= add_excluded_extent(root
, start
, len
);
313 static struct btrfs_caching_control
*
314 get_caching_control(struct btrfs_block_group_cache
*cache
)
316 struct btrfs_caching_control
*ctl
;
318 spin_lock(&cache
->lock
);
319 if (!cache
->caching_ctl
) {
320 spin_unlock(&cache
->lock
);
324 ctl
= cache
->caching_ctl
;
325 atomic_inc(&ctl
->count
);
326 spin_unlock(&cache
->lock
);
330 static void put_caching_control(struct btrfs_caching_control
*ctl
)
332 if (atomic_dec_and_test(&ctl
->count
))
337 * this is only called by cache_block_group, since we could have freed extents
338 * we need to check the pinned_extents for any extents that can't be used yet
339 * since their free space will be released as soon as the transaction commits.
341 static u64
add_new_free_space(struct btrfs_block_group_cache
*block_group
,
342 struct btrfs_fs_info
*info
, u64 start
, u64 end
)
344 u64 extent_start
, extent_end
, size
, total_added
= 0;
347 while (start
< end
) {
348 ret
= find_first_extent_bit(info
->pinned_extents
, start
,
349 &extent_start
, &extent_end
,
350 EXTENT_DIRTY
| EXTENT_UPTODATE
,
355 if (extent_start
<= start
) {
356 start
= extent_end
+ 1;
357 } else if (extent_start
> start
&& extent_start
< end
) {
358 size
= extent_start
- start
;
360 ret
= btrfs_add_free_space(block_group
, start
,
362 BUG_ON(ret
); /* -ENOMEM or logic error */
363 start
= extent_end
+ 1;
372 ret
= btrfs_add_free_space(block_group
, start
, size
);
373 BUG_ON(ret
); /* -ENOMEM or logic error */
379 static noinline
void caching_thread(struct btrfs_work
*work
)
381 struct btrfs_block_group_cache
*block_group
;
382 struct btrfs_fs_info
*fs_info
;
383 struct btrfs_caching_control
*caching_ctl
;
384 struct btrfs_root
*extent_root
;
385 struct btrfs_path
*path
;
386 struct extent_buffer
*leaf
;
387 struct btrfs_key key
;
393 caching_ctl
= container_of(work
, struct btrfs_caching_control
, work
);
394 block_group
= caching_ctl
->block_group
;
395 fs_info
= block_group
->fs_info
;
396 extent_root
= fs_info
->extent_root
;
398 path
= btrfs_alloc_path();
402 last
= max_t(u64
, block_group
->key
.objectid
, BTRFS_SUPER_INFO_OFFSET
);
405 * We don't want to deadlock with somebody trying to allocate a new
406 * extent for the extent root while also trying to search the extent
407 * root to add free space. So we skip locking and search the commit
408 * root, since its read-only
410 path
->skip_locking
= 1;
411 path
->search_commit_root
= 1;
416 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
418 mutex_lock(&caching_ctl
->mutex
);
419 /* need to make sure the commit_root doesn't disappear */
420 down_read(&fs_info
->commit_root_sem
);
423 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
427 leaf
= path
->nodes
[0];
428 nritems
= btrfs_header_nritems(leaf
);
431 if (btrfs_fs_closing(fs_info
) > 1) {
436 if (path
->slots
[0] < nritems
) {
437 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
439 ret
= find_next_key(path
, 0, &key
);
443 if (need_resched() ||
444 rwsem_is_contended(&fs_info
->commit_root_sem
)) {
445 caching_ctl
->progress
= last
;
446 btrfs_release_path(path
);
447 up_read(&fs_info
->commit_root_sem
);
448 mutex_unlock(&caching_ctl
->mutex
);
453 ret
= btrfs_next_leaf(extent_root
, path
);
458 leaf
= path
->nodes
[0];
459 nritems
= btrfs_header_nritems(leaf
);
463 if (key
.objectid
< last
) {
466 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
468 caching_ctl
->progress
= last
;
469 btrfs_release_path(path
);
473 if (key
.objectid
< block_group
->key
.objectid
) {
478 if (key
.objectid
>= block_group
->key
.objectid
+
479 block_group
->key
.offset
)
482 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
||
483 key
.type
== BTRFS_METADATA_ITEM_KEY
) {
484 total_found
+= add_new_free_space(block_group
,
487 if (key
.type
== BTRFS_METADATA_ITEM_KEY
)
488 last
= key
.objectid
+
489 fs_info
->tree_root
->nodesize
;
491 last
= key
.objectid
+ key
.offset
;
493 if (total_found
> (1024 * 1024 * 2)) {
495 wake_up(&caching_ctl
->wait
);
502 total_found
+= add_new_free_space(block_group
, fs_info
, last
,
503 block_group
->key
.objectid
+
504 block_group
->key
.offset
);
505 caching_ctl
->progress
= (u64
)-1;
507 spin_lock(&block_group
->lock
);
508 block_group
->caching_ctl
= NULL
;
509 block_group
->cached
= BTRFS_CACHE_FINISHED
;
510 spin_unlock(&block_group
->lock
);
513 btrfs_free_path(path
);
514 up_read(&fs_info
->commit_root_sem
);
516 free_excluded_extents(extent_root
, block_group
);
518 mutex_unlock(&caching_ctl
->mutex
);
521 spin_lock(&block_group
->lock
);
522 block_group
->caching_ctl
= NULL
;
523 block_group
->cached
= BTRFS_CACHE_ERROR
;
524 spin_unlock(&block_group
->lock
);
526 wake_up(&caching_ctl
->wait
);
528 put_caching_control(caching_ctl
);
529 btrfs_put_block_group(block_group
);
532 static int cache_block_group(struct btrfs_block_group_cache
*cache
,
536 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
537 struct btrfs_caching_control
*caching_ctl
;
540 caching_ctl
= kzalloc(sizeof(*caching_ctl
), GFP_NOFS
);
544 INIT_LIST_HEAD(&caching_ctl
->list
);
545 mutex_init(&caching_ctl
->mutex
);
546 init_waitqueue_head(&caching_ctl
->wait
);
547 caching_ctl
->block_group
= cache
;
548 caching_ctl
->progress
= cache
->key
.objectid
;
549 atomic_set(&caching_ctl
->count
, 1);
550 btrfs_init_work(&caching_ctl
->work
, btrfs_cache_helper
,
551 caching_thread
, NULL
, NULL
);
553 spin_lock(&cache
->lock
);
555 * This should be a rare occasion, but this could happen I think in the
556 * case where one thread starts to load the space cache info, and then
557 * some other thread starts a transaction commit which tries to do an
558 * allocation while the other thread is still loading the space cache
559 * info. The previous loop should have kept us from choosing this block
560 * group, but if we've moved to the state where we will wait on caching
561 * block groups we need to first check if we're doing a fast load here,
562 * so we can wait for it to finish, otherwise we could end up allocating
563 * from a block group who's cache gets evicted for one reason or
566 while (cache
->cached
== BTRFS_CACHE_FAST
) {
567 struct btrfs_caching_control
*ctl
;
569 ctl
= cache
->caching_ctl
;
570 atomic_inc(&ctl
->count
);
571 prepare_to_wait(&ctl
->wait
, &wait
, TASK_UNINTERRUPTIBLE
);
572 spin_unlock(&cache
->lock
);
576 finish_wait(&ctl
->wait
, &wait
);
577 put_caching_control(ctl
);
578 spin_lock(&cache
->lock
);
581 if (cache
->cached
!= BTRFS_CACHE_NO
) {
582 spin_unlock(&cache
->lock
);
586 WARN_ON(cache
->caching_ctl
);
587 cache
->caching_ctl
= caching_ctl
;
588 cache
->cached
= BTRFS_CACHE_FAST
;
589 spin_unlock(&cache
->lock
);
591 if (fs_info
->mount_opt
& BTRFS_MOUNT_SPACE_CACHE
) {
592 mutex_lock(&caching_ctl
->mutex
);
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;
600 caching_ctl
->progress
= (u64
)-1;
602 if (load_cache_only
) {
603 cache
->caching_ctl
= NULL
;
604 cache
->cached
= BTRFS_CACHE_NO
;
606 cache
->cached
= BTRFS_CACHE_STARTED
;
607 cache
->has_caching_ctl
= 1;
610 spin_unlock(&cache
->lock
);
611 mutex_unlock(&caching_ctl
->mutex
);
613 wake_up(&caching_ctl
->wait
);
615 put_caching_control(caching_ctl
);
616 free_excluded_extents(fs_info
->extent_root
, cache
);
621 * We are not going to do the fast caching, set cached to the
622 * appropriate value and wakeup any waiters.
624 spin_lock(&cache
->lock
);
625 if (load_cache_only
) {
626 cache
->caching_ctl
= NULL
;
627 cache
->cached
= BTRFS_CACHE_NO
;
629 cache
->cached
= BTRFS_CACHE_STARTED
;
630 cache
->has_caching_ctl
= 1;
632 spin_unlock(&cache
->lock
);
633 wake_up(&caching_ctl
->wait
);
636 if (load_cache_only
) {
637 put_caching_control(caching_ctl
);
641 down_write(&fs_info
->commit_root_sem
);
642 atomic_inc(&caching_ctl
->count
);
643 list_add_tail(&caching_ctl
->list
, &fs_info
->caching_block_groups
);
644 up_write(&fs_info
->commit_root_sem
);
646 btrfs_get_block_group(cache
);
648 btrfs_queue_work(fs_info
->caching_workers
, &caching_ctl
->work
);
654 * return the block group that starts at or after bytenr
656 static struct btrfs_block_group_cache
*
657 btrfs_lookup_first_block_group(struct btrfs_fs_info
*info
, u64 bytenr
)
659 struct btrfs_block_group_cache
*cache
;
661 cache
= block_group_cache_tree_search(info
, bytenr
, 0);
667 * return the block group that contains the given bytenr
669 struct btrfs_block_group_cache
*btrfs_lookup_block_group(
670 struct btrfs_fs_info
*info
,
673 struct btrfs_block_group_cache
*cache
;
675 cache
= block_group_cache_tree_search(info
, bytenr
, 1);
680 static struct btrfs_space_info
*__find_space_info(struct btrfs_fs_info
*info
,
683 struct list_head
*head
= &info
->space_info
;
684 struct btrfs_space_info
*found
;
686 flags
&= BTRFS_BLOCK_GROUP_TYPE_MASK
;
689 list_for_each_entry_rcu(found
, head
, list
) {
690 if (found
->flags
& flags
) {
700 * after adding space to the filesystem, we need to clear the full flags
701 * on all the space infos.
703 void btrfs_clear_space_info_full(struct btrfs_fs_info
*info
)
705 struct list_head
*head
= &info
->space_info
;
706 struct btrfs_space_info
*found
;
709 list_for_each_entry_rcu(found
, head
, list
)
714 /* simple helper to search for an existing data extent at a given offset */
715 int btrfs_lookup_data_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
718 struct btrfs_key key
;
719 struct btrfs_path
*path
;
721 path
= btrfs_alloc_path();
725 key
.objectid
= start
;
727 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
728 ret
= btrfs_search_slot(NULL
, root
->fs_info
->extent_root
, &key
, path
,
730 btrfs_free_path(path
);
735 * helper function to lookup reference count and flags of a tree block.
737 * the head node for delayed ref is used to store the sum of all the
738 * reference count modifications queued up in the rbtree. the head
739 * node may also store the extent flags to set. This way you can check
740 * to see what the reference count and extent flags would be if all of
741 * the delayed refs are not processed.
743 int btrfs_lookup_extent_info(struct btrfs_trans_handle
*trans
,
744 struct btrfs_root
*root
, u64 bytenr
,
745 u64 offset
, int metadata
, u64
*refs
, u64
*flags
)
747 struct btrfs_delayed_ref_head
*head
;
748 struct btrfs_delayed_ref_root
*delayed_refs
;
749 struct btrfs_path
*path
;
750 struct btrfs_extent_item
*ei
;
751 struct extent_buffer
*leaf
;
752 struct btrfs_key key
;
759 * If we don't have skinny metadata, don't bother doing anything
762 if (metadata
&& !btrfs_fs_incompat(root
->fs_info
, SKINNY_METADATA
)) {
763 offset
= root
->nodesize
;
767 path
= btrfs_alloc_path();
772 path
->skip_locking
= 1;
773 path
->search_commit_root
= 1;
777 key
.objectid
= bytenr
;
780 key
.type
= BTRFS_METADATA_ITEM_KEY
;
782 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
784 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
,
789 if (ret
> 0 && metadata
&& key
.type
== BTRFS_METADATA_ITEM_KEY
) {
790 if (path
->slots
[0]) {
792 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
794 if (key
.objectid
== bytenr
&&
795 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
796 key
.offset
== root
->nodesize
)
802 leaf
= path
->nodes
[0];
803 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
804 if (item_size
>= sizeof(*ei
)) {
805 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
806 struct btrfs_extent_item
);
807 num_refs
= btrfs_extent_refs(leaf
, ei
);
808 extent_flags
= btrfs_extent_flags(leaf
, ei
);
810 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
811 struct btrfs_extent_item_v0
*ei0
;
812 BUG_ON(item_size
!= sizeof(*ei0
));
813 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
814 struct btrfs_extent_item_v0
);
815 num_refs
= btrfs_extent_refs_v0(leaf
, ei0
);
816 /* FIXME: this isn't correct for data */
817 extent_flags
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
822 BUG_ON(num_refs
== 0);
832 delayed_refs
= &trans
->transaction
->delayed_refs
;
833 spin_lock(&delayed_refs
->lock
);
834 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
836 if (!mutex_trylock(&head
->mutex
)) {
837 atomic_inc(&head
->node
.refs
);
838 spin_unlock(&delayed_refs
->lock
);
840 btrfs_release_path(path
);
843 * Mutex was contended, block until it's released and try
846 mutex_lock(&head
->mutex
);
847 mutex_unlock(&head
->mutex
);
848 btrfs_put_delayed_ref(&head
->node
);
851 spin_lock(&head
->lock
);
852 if (head
->extent_op
&& head
->extent_op
->update_flags
)
853 extent_flags
|= head
->extent_op
->flags_to_set
;
855 BUG_ON(num_refs
== 0);
857 num_refs
+= head
->node
.ref_mod
;
858 spin_unlock(&head
->lock
);
859 mutex_unlock(&head
->mutex
);
861 spin_unlock(&delayed_refs
->lock
);
863 WARN_ON(num_refs
== 0);
867 *flags
= extent_flags
;
869 btrfs_free_path(path
);
874 * Back reference rules. Back refs have three main goals:
876 * 1) differentiate between all holders of references to an extent so that
877 * when a reference is dropped we can make sure it was a valid reference
878 * before freeing the extent.
880 * 2) Provide enough information to quickly find the holders of an extent
881 * if we notice a given block is corrupted or bad.
883 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
884 * maintenance. This is actually the same as #2, but with a slightly
885 * different use case.
887 * There are two kinds of back refs. The implicit back refs is optimized
888 * for pointers in non-shared tree blocks. For a given pointer in a block,
889 * back refs of this kind provide information about the block's owner tree
890 * and the pointer's key. These information allow us to find the block by
891 * b-tree searching. The full back refs is for pointers in tree blocks not
892 * referenced by their owner trees. The location of tree block is recorded
893 * in the back refs. Actually the full back refs is generic, and can be
894 * used in all cases the implicit back refs is used. The major shortcoming
895 * of the full back refs is its overhead. Every time a tree block gets
896 * COWed, we have to update back refs entry for all pointers in it.
898 * For a newly allocated tree block, we use implicit back refs for
899 * pointers in it. This means most tree related operations only involve
900 * implicit back refs. For a tree block created in old transaction, the
901 * only way to drop a reference to it is COW it. So we can detect the
902 * event that tree block loses its owner tree's reference and do the
903 * back refs conversion.
905 * When a tree block is COW'd through a tree, there are four cases:
907 * The reference count of the block is one and the tree is the block's
908 * owner tree. Nothing to do in this case.
910 * The reference count of the block is one and the tree is not the
911 * block's owner tree. In this case, full back refs is used for pointers
912 * in the block. Remove these full back refs, add implicit back refs for
913 * every pointers in the new block.
915 * The reference count of the block is greater than one and the tree is
916 * the block's owner tree. In this case, implicit back refs is used for
917 * pointers in the block. Add full back refs for every pointers in the
918 * block, increase lower level extents' reference counts. The original
919 * implicit back refs are entailed to the new block.
921 * The reference count of the block is greater than one and the tree is
922 * not the block's owner tree. Add implicit back refs for every pointer in
923 * the new block, increase lower level extents' reference count.
925 * Back Reference Key composing:
927 * The key objectid corresponds to the first byte in the extent,
928 * The key type is used to differentiate between types of back refs.
929 * There are different meanings of the key offset for different types
932 * File extents can be referenced by:
934 * - multiple snapshots, subvolumes, or different generations in one subvol
935 * - different files inside a single subvolume
936 * - different offsets inside a file (bookend extents in file.c)
938 * The extent ref structure for the implicit back refs has fields for:
940 * - Objectid of the subvolume root
941 * - objectid of the file holding the reference
942 * - original offset in the file
943 * - how many bookend extents
945 * The key offset for the implicit back refs is hash of the first
948 * The extent ref structure for the full back refs has field for:
950 * - number of pointers in the tree leaf
952 * The key offset for the implicit back refs is the first byte of
955 * When a file extent is allocated, The implicit back refs is used.
956 * the fields are filled in:
958 * (root_key.objectid, inode objectid, offset in file, 1)
960 * When a file extent is removed file truncation, we find the
961 * corresponding implicit back refs and check the following fields:
963 * (btrfs_header_owner(leaf), inode objectid, offset in file)
965 * Btree extents can be referenced by:
967 * - Different subvolumes
969 * Both the implicit back refs and the full back refs for tree blocks
970 * only consist of key. The key offset for the implicit back refs is
971 * objectid of block's owner tree. The key offset for the full back refs
972 * is the first byte of parent block.
974 * When implicit back refs is used, information about the lowest key and
975 * level of the tree block are required. These information are stored in
976 * tree block info structure.
979 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
980 static int convert_extent_item_v0(struct btrfs_trans_handle
*trans
,
981 struct btrfs_root
*root
,
982 struct btrfs_path
*path
,
983 u64 owner
, u32 extra_size
)
985 struct btrfs_extent_item
*item
;
986 struct btrfs_extent_item_v0
*ei0
;
987 struct btrfs_extent_ref_v0
*ref0
;
988 struct btrfs_tree_block_info
*bi
;
989 struct extent_buffer
*leaf
;
990 struct btrfs_key key
;
991 struct btrfs_key found_key
;
992 u32 new_size
= sizeof(*item
);
996 leaf
= path
->nodes
[0];
997 BUG_ON(btrfs_item_size_nr(leaf
, path
->slots
[0]) != sizeof(*ei0
));
999 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1000 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1001 struct btrfs_extent_item_v0
);
1002 refs
= btrfs_extent_refs_v0(leaf
, ei0
);
1004 if (owner
== (u64
)-1) {
1006 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1007 ret
= btrfs_next_leaf(root
, path
);
1010 BUG_ON(ret
> 0); /* Corruption */
1011 leaf
= path
->nodes
[0];
1013 btrfs_item_key_to_cpu(leaf
, &found_key
,
1015 BUG_ON(key
.objectid
!= found_key
.objectid
);
1016 if (found_key
.type
!= BTRFS_EXTENT_REF_V0_KEY
) {
1020 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1021 struct btrfs_extent_ref_v0
);
1022 owner
= btrfs_ref_objectid_v0(leaf
, ref0
);
1026 btrfs_release_path(path
);
1028 if (owner
< BTRFS_FIRST_FREE_OBJECTID
)
1029 new_size
+= sizeof(*bi
);
1031 new_size
-= sizeof(*ei0
);
1032 ret
= btrfs_search_slot(trans
, root
, &key
, path
,
1033 new_size
+ extra_size
, 1);
1036 BUG_ON(ret
); /* Corruption */
1038 btrfs_extend_item(root
, path
, new_size
);
1040 leaf
= path
->nodes
[0];
1041 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1042 btrfs_set_extent_refs(leaf
, item
, refs
);
1043 /* FIXME: get real generation */
1044 btrfs_set_extent_generation(leaf
, item
, 0);
1045 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1046 btrfs_set_extent_flags(leaf
, item
,
1047 BTRFS_EXTENT_FLAG_TREE_BLOCK
|
1048 BTRFS_BLOCK_FLAG_FULL_BACKREF
);
1049 bi
= (struct btrfs_tree_block_info
*)(item
+ 1);
1050 /* FIXME: get first key of the block */
1051 memset_extent_buffer(leaf
, 0, (unsigned long)bi
, sizeof(*bi
));
1052 btrfs_set_tree_block_level(leaf
, bi
, (int)owner
);
1054 btrfs_set_extent_flags(leaf
, item
, BTRFS_EXTENT_FLAG_DATA
);
1056 btrfs_mark_buffer_dirty(leaf
);
1061 static u64
hash_extent_data_ref(u64 root_objectid
, u64 owner
, u64 offset
)
1063 u32 high_crc
= ~(u32
)0;
1064 u32 low_crc
= ~(u32
)0;
1067 lenum
= cpu_to_le64(root_objectid
);
1068 high_crc
= btrfs_crc32c(high_crc
, &lenum
, sizeof(lenum
));
1069 lenum
= cpu_to_le64(owner
);
1070 low_crc
= btrfs_crc32c(low_crc
, &lenum
, sizeof(lenum
));
1071 lenum
= cpu_to_le64(offset
);
1072 low_crc
= btrfs_crc32c(low_crc
, &lenum
, sizeof(lenum
));
1074 return ((u64
)high_crc
<< 31) ^ (u64
)low_crc
;
1077 static u64
hash_extent_data_ref_item(struct extent_buffer
*leaf
,
1078 struct btrfs_extent_data_ref
*ref
)
1080 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf
, ref
),
1081 btrfs_extent_data_ref_objectid(leaf
, ref
),
1082 btrfs_extent_data_ref_offset(leaf
, ref
));
1085 static int match_extent_data_ref(struct extent_buffer
*leaf
,
1086 struct btrfs_extent_data_ref
*ref
,
1087 u64 root_objectid
, u64 owner
, u64 offset
)
1089 if (btrfs_extent_data_ref_root(leaf
, ref
) != root_objectid
||
1090 btrfs_extent_data_ref_objectid(leaf
, ref
) != owner
||
1091 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
1096 static noinline
int lookup_extent_data_ref(struct btrfs_trans_handle
*trans
,
1097 struct btrfs_root
*root
,
1098 struct btrfs_path
*path
,
1099 u64 bytenr
, u64 parent
,
1101 u64 owner
, u64 offset
)
1103 struct btrfs_key key
;
1104 struct btrfs_extent_data_ref
*ref
;
1105 struct extent_buffer
*leaf
;
1111 key
.objectid
= bytenr
;
1113 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1114 key
.offset
= parent
;
1116 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1117 key
.offset
= hash_extent_data_ref(root_objectid
,
1122 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1131 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1132 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1133 btrfs_release_path(path
);
1134 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1145 leaf
= path
->nodes
[0];
1146 nritems
= btrfs_header_nritems(leaf
);
1148 if (path
->slots
[0] >= nritems
) {
1149 ret
= btrfs_next_leaf(root
, path
);
1155 leaf
= path
->nodes
[0];
1156 nritems
= btrfs_header_nritems(leaf
);
1160 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1161 if (key
.objectid
!= bytenr
||
1162 key
.type
!= BTRFS_EXTENT_DATA_REF_KEY
)
1165 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1166 struct btrfs_extent_data_ref
);
1168 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1171 btrfs_release_path(path
);
1183 static noinline
int insert_extent_data_ref(struct btrfs_trans_handle
*trans
,
1184 struct btrfs_root
*root
,
1185 struct btrfs_path
*path
,
1186 u64 bytenr
, u64 parent
,
1187 u64 root_objectid
, u64 owner
,
1188 u64 offset
, int refs_to_add
)
1190 struct btrfs_key key
;
1191 struct extent_buffer
*leaf
;
1196 key
.objectid
= bytenr
;
1198 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1199 key
.offset
= parent
;
1200 size
= sizeof(struct btrfs_shared_data_ref
);
1202 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1203 key
.offset
= hash_extent_data_ref(root_objectid
,
1205 size
= sizeof(struct btrfs_extent_data_ref
);
1208 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, size
);
1209 if (ret
&& ret
!= -EEXIST
)
1212 leaf
= path
->nodes
[0];
1214 struct btrfs_shared_data_ref
*ref
;
1215 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1216 struct btrfs_shared_data_ref
);
1218 btrfs_set_shared_data_ref_count(leaf
, ref
, refs_to_add
);
1220 num_refs
= btrfs_shared_data_ref_count(leaf
, ref
);
1221 num_refs
+= refs_to_add
;
1222 btrfs_set_shared_data_ref_count(leaf
, ref
, num_refs
);
1225 struct btrfs_extent_data_ref
*ref
;
1226 while (ret
== -EEXIST
) {
1227 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1228 struct btrfs_extent_data_ref
);
1229 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1232 btrfs_release_path(path
);
1234 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1236 if (ret
&& ret
!= -EEXIST
)
1239 leaf
= path
->nodes
[0];
1241 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1242 struct btrfs_extent_data_ref
);
1244 btrfs_set_extent_data_ref_root(leaf
, ref
,
1246 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
1247 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
1248 btrfs_set_extent_data_ref_count(leaf
, ref
, refs_to_add
);
1250 num_refs
= btrfs_extent_data_ref_count(leaf
, ref
);
1251 num_refs
+= refs_to_add
;
1252 btrfs_set_extent_data_ref_count(leaf
, ref
, num_refs
);
1255 btrfs_mark_buffer_dirty(leaf
);
1258 btrfs_release_path(path
);
1262 static noinline
int remove_extent_data_ref(struct btrfs_trans_handle
*trans
,
1263 struct btrfs_root
*root
,
1264 struct btrfs_path
*path
,
1265 int refs_to_drop
, int *last_ref
)
1267 struct btrfs_key key
;
1268 struct btrfs_extent_data_ref
*ref1
= NULL
;
1269 struct btrfs_shared_data_ref
*ref2
= NULL
;
1270 struct extent_buffer
*leaf
;
1274 leaf
= path
->nodes
[0];
1275 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1277 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1278 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1279 struct btrfs_extent_data_ref
);
1280 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1281 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1282 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1283 struct btrfs_shared_data_ref
);
1284 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1285 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1286 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1287 struct btrfs_extent_ref_v0
*ref0
;
1288 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1289 struct btrfs_extent_ref_v0
);
1290 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1296 BUG_ON(num_refs
< refs_to_drop
);
1297 num_refs
-= refs_to_drop
;
1299 if (num_refs
== 0) {
1300 ret
= btrfs_del_item(trans
, root
, path
);
1303 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
)
1304 btrfs_set_extent_data_ref_count(leaf
, ref1
, num_refs
);
1305 else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
)
1306 btrfs_set_shared_data_ref_count(leaf
, ref2
, num_refs
);
1307 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1309 struct btrfs_extent_ref_v0
*ref0
;
1310 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1311 struct btrfs_extent_ref_v0
);
1312 btrfs_set_ref_count_v0(leaf
, ref0
, num_refs
);
1315 btrfs_mark_buffer_dirty(leaf
);
1320 static noinline u32
extent_data_ref_count(struct btrfs_root
*root
,
1321 struct btrfs_path
*path
,
1322 struct btrfs_extent_inline_ref
*iref
)
1324 struct btrfs_key key
;
1325 struct extent_buffer
*leaf
;
1326 struct btrfs_extent_data_ref
*ref1
;
1327 struct btrfs_shared_data_ref
*ref2
;
1330 leaf
= path
->nodes
[0];
1331 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1333 if (btrfs_extent_inline_ref_type(leaf
, iref
) ==
1334 BTRFS_EXTENT_DATA_REF_KEY
) {
1335 ref1
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1336 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1338 ref2
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1339 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1341 } else if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1342 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1343 struct btrfs_extent_data_ref
);
1344 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1345 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1346 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1347 struct btrfs_shared_data_ref
);
1348 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1349 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1350 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1351 struct btrfs_extent_ref_v0
*ref0
;
1352 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1353 struct btrfs_extent_ref_v0
);
1354 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1362 static noinline
int lookup_tree_block_ref(struct btrfs_trans_handle
*trans
,
1363 struct btrfs_root
*root
,
1364 struct btrfs_path
*path
,
1365 u64 bytenr
, u64 parent
,
1368 struct btrfs_key key
;
1371 key
.objectid
= bytenr
;
1373 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1374 key
.offset
= parent
;
1376 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1377 key
.offset
= root_objectid
;
1380 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1383 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1384 if (ret
== -ENOENT
&& parent
) {
1385 btrfs_release_path(path
);
1386 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1387 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1395 static noinline
int insert_tree_block_ref(struct btrfs_trans_handle
*trans
,
1396 struct btrfs_root
*root
,
1397 struct btrfs_path
*path
,
1398 u64 bytenr
, u64 parent
,
1401 struct btrfs_key key
;
1404 key
.objectid
= bytenr
;
1406 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1407 key
.offset
= parent
;
1409 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1410 key
.offset
= root_objectid
;
1413 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
1414 btrfs_release_path(path
);
1418 static inline int extent_ref_type(u64 parent
, u64 owner
)
1421 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1423 type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1425 type
= BTRFS_TREE_BLOCK_REF_KEY
;
1428 type
= BTRFS_SHARED_DATA_REF_KEY
;
1430 type
= BTRFS_EXTENT_DATA_REF_KEY
;
1435 static int find_next_key(struct btrfs_path
*path
, int level
,
1436 struct btrfs_key
*key
)
1439 for (; level
< BTRFS_MAX_LEVEL
; level
++) {
1440 if (!path
->nodes
[level
])
1442 if (path
->slots
[level
] + 1 >=
1443 btrfs_header_nritems(path
->nodes
[level
]))
1446 btrfs_item_key_to_cpu(path
->nodes
[level
], key
,
1447 path
->slots
[level
] + 1);
1449 btrfs_node_key_to_cpu(path
->nodes
[level
], key
,
1450 path
->slots
[level
] + 1);
1457 * look for inline back ref. if back ref is found, *ref_ret is set
1458 * to the address of inline back ref, and 0 is returned.
1460 * if back ref isn't found, *ref_ret is set to the address where it
1461 * should be inserted, and -ENOENT is returned.
1463 * if insert is true and there are too many inline back refs, the path
1464 * points to the extent item, and -EAGAIN is returned.
1466 * NOTE: inline back refs are ordered in the same way that back ref
1467 * items in the tree are ordered.
1469 static noinline_for_stack
1470 int lookup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1471 struct btrfs_root
*root
,
1472 struct btrfs_path
*path
,
1473 struct btrfs_extent_inline_ref
**ref_ret
,
1474 u64 bytenr
, u64 num_bytes
,
1475 u64 parent
, u64 root_objectid
,
1476 u64 owner
, u64 offset
, int insert
)
1478 struct btrfs_key key
;
1479 struct extent_buffer
*leaf
;
1480 struct btrfs_extent_item
*ei
;
1481 struct btrfs_extent_inline_ref
*iref
;
1491 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
1494 key
.objectid
= bytenr
;
1495 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1496 key
.offset
= num_bytes
;
1498 want
= extent_ref_type(parent
, owner
);
1500 extra_size
= btrfs_extent_inline_ref_size(want
);
1501 path
->keep_locks
= 1;
1506 * Owner is our parent level, so we can just add one to get the level
1507 * for the block we are interested in.
1509 if (skinny_metadata
&& owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1510 key
.type
= BTRFS_METADATA_ITEM_KEY
;
1515 ret
= btrfs_search_slot(trans
, root
, &key
, path
, extra_size
, 1);
1522 * We may be a newly converted file system which still has the old fat
1523 * extent entries for metadata, so try and see if we have one of those.
1525 if (ret
> 0 && skinny_metadata
) {
1526 skinny_metadata
= false;
1527 if (path
->slots
[0]) {
1529 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
1531 if (key
.objectid
== bytenr
&&
1532 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
1533 key
.offset
== num_bytes
)
1537 key
.objectid
= bytenr
;
1538 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1539 key
.offset
= num_bytes
;
1540 btrfs_release_path(path
);
1545 if (ret
&& !insert
) {
1548 } else if (WARN_ON(ret
)) {
1553 leaf
= path
->nodes
[0];
1554 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1555 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1556 if (item_size
< sizeof(*ei
)) {
1561 ret
= convert_extent_item_v0(trans
, root
, path
, owner
,
1567 leaf
= path
->nodes
[0];
1568 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1571 BUG_ON(item_size
< sizeof(*ei
));
1573 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1574 flags
= btrfs_extent_flags(leaf
, ei
);
1576 ptr
= (unsigned long)(ei
+ 1);
1577 end
= (unsigned long)ei
+ item_size
;
1579 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
&& !skinny_metadata
) {
1580 ptr
+= sizeof(struct btrfs_tree_block_info
);
1590 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1591 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1595 ptr
+= btrfs_extent_inline_ref_size(type
);
1599 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1600 struct btrfs_extent_data_ref
*dref
;
1601 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1602 if (match_extent_data_ref(leaf
, dref
, root_objectid
,
1607 if (hash_extent_data_ref_item(leaf
, dref
) <
1608 hash_extent_data_ref(root_objectid
, owner
, offset
))
1612 ref_offset
= btrfs_extent_inline_ref_offset(leaf
, iref
);
1614 if (parent
== ref_offset
) {
1618 if (ref_offset
< parent
)
1621 if (root_objectid
== ref_offset
) {
1625 if (ref_offset
< root_objectid
)
1629 ptr
+= btrfs_extent_inline_ref_size(type
);
1631 if (err
== -ENOENT
&& insert
) {
1632 if (item_size
+ extra_size
>=
1633 BTRFS_MAX_EXTENT_ITEM_SIZE(root
)) {
1638 * To add new inline back ref, we have to make sure
1639 * there is no corresponding back ref item.
1640 * For simplicity, we just do not add new inline back
1641 * ref if there is any kind of item for this block
1643 if (find_next_key(path
, 0, &key
) == 0 &&
1644 key
.objectid
== bytenr
&&
1645 key
.type
< BTRFS_BLOCK_GROUP_ITEM_KEY
) {
1650 *ref_ret
= (struct btrfs_extent_inline_ref
*)ptr
;
1653 path
->keep_locks
= 0;
1654 btrfs_unlock_up_safe(path
, 1);
1660 * helper to add new inline back ref
1662 static noinline_for_stack
1663 void setup_inline_extent_backref(struct btrfs_root
*root
,
1664 struct btrfs_path
*path
,
1665 struct btrfs_extent_inline_ref
*iref
,
1666 u64 parent
, u64 root_objectid
,
1667 u64 owner
, u64 offset
, int refs_to_add
,
1668 struct btrfs_delayed_extent_op
*extent_op
)
1670 struct extent_buffer
*leaf
;
1671 struct btrfs_extent_item
*ei
;
1674 unsigned long item_offset
;
1679 leaf
= path
->nodes
[0];
1680 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1681 item_offset
= (unsigned long)iref
- (unsigned long)ei
;
1683 type
= extent_ref_type(parent
, owner
);
1684 size
= btrfs_extent_inline_ref_size(type
);
1686 btrfs_extend_item(root
, path
, size
);
1688 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1689 refs
= btrfs_extent_refs(leaf
, ei
);
1690 refs
+= refs_to_add
;
1691 btrfs_set_extent_refs(leaf
, ei
, refs
);
1693 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1695 ptr
= (unsigned long)ei
+ item_offset
;
1696 end
= (unsigned long)ei
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
1697 if (ptr
< end
- size
)
1698 memmove_extent_buffer(leaf
, ptr
+ size
, ptr
,
1701 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1702 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
1703 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1704 struct btrfs_extent_data_ref
*dref
;
1705 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1706 btrfs_set_extent_data_ref_root(leaf
, dref
, root_objectid
);
1707 btrfs_set_extent_data_ref_objectid(leaf
, dref
, owner
);
1708 btrfs_set_extent_data_ref_offset(leaf
, dref
, offset
);
1709 btrfs_set_extent_data_ref_count(leaf
, dref
, refs_to_add
);
1710 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1711 struct btrfs_shared_data_ref
*sref
;
1712 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1713 btrfs_set_shared_data_ref_count(leaf
, sref
, refs_to_add
);
1714 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1715 } else if (type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
1716 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1718 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
1720 btrfs_mark_buffer_dirty(leaf
);
1723 static int lookup_extent_backref(struct btrfs_trans_handle
*trans
,
1724 struct btrfs_root
*root
,
1725 struct btrfs_path
*path
,
1726 struct btrfs_extent_inline_ref
**ref_ret
,
1727 u64 bytenr
, u64 num_bytes
, u64 parent
,
1728 u64 root_objectid
, u64 owner
, u64 offset
)
1732 ret
= lookup_inline_extent_backref(trans
, root
, path
, ref_ret
,
1733 bytenr
, num_bytes
, parent
,
1734 root_objectid
, owner
, offset
, 0);
1738 btrfs_release_path(path
);
1741 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1742 ret
= lookup_tree_block_ref(trans
, root
, path
, bytenr
, parent
,
1745 ret
= lookup_extent_data_ref(trans
, root
, path
, bytenr
, parent
,
1746 root_objectid
, owner
, offset
);
1752 * helper to update/remove inline back ref
1754 static noinline_for_stack
1755 void update_inline_extent_backref(struct btrfs_root
*root
,
1756 struct btrfs_path
*path
,
1757 struct btrfs_extent_inline_ref
*iref
,
1759 struct btrfs_delayed_extent_op
*extent_op
,
1762 struct extent_buffer
*leaf
;
1763 struct btrfs_extent_item
*ei
;
1764 struct btrfs_extent_data_ref
*dref
= NULL
;
1765 struct btrfs_shared_data_ref
*sref
= NULL
;
1773 leaf
= path
->nodes
[0];
1774 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1775 refs
= btrfs_extent_refs(leaf
, ei
);
1776 WARN_ON(refs_to_mod
< 0 && refs
+ refs_to_mod
<= 0);
1777 refs
+= refs_to_mod
;
1778 btrfs_set_extent_refs(leaf
, ei
, refs
);
1780 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1782 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1784 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1785 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1786 refs
= btrfs_extent_data_ref_count(leaf
, dref
);
1787 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1788 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1789 refs
= btrfs_shared_data_ref_count(leaf
, sref
);
1792 BUG_ON(refs_to_mod
!= -1);
1795 BUG_ON(refs_to_mod
< 0 && refs
< -refs_to_mod
);
1796 refs
+= refs_to_mod
;
1799 if (type
== BTRFS_EXTENT_DATA_REF_KEY
)
1800 btrfs_set_extent_data_ref_count(leaf
, dref
, refs
);
1802 btrfs_set_shared_data_ref_count(leaf
, sref
, refs
);
1805 size
= btrfs_extent_inline_ref_size(type
);
1806 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1807 ptr
= (unsigned long)iref
;
1808 end
= (unsigned long)ei
+ item_size
;
1809 if (ptr
+ size
< end
)
1810 memmove_extent_buffer(leaf
, ptr
, ptr
+ size
,
1813 btrfs_truncate_item(root
, path
, item_size
, 1);
1815 btrfs_mark_buffer_dirty(leaf
);
1818 static noinline_for_stack
1819 int insert_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1820 struct btrfs_root
*root
,
1821 struct btrfs_path
*path
,
1822 u64 bytenr
, u64 num_bytes
, u64 parent
,
1823 u64 root_objectid
, u64 owner
,
1824 u64 offset
, int refs_to_add
,
1825 struct btrfs_delayed_extent_op
*extent_op
)
1827 struct btrfs_extent_inline_ref
*iref
;
1830 ret
= lookup_inline_extent_backref(trans
, root
, path
, &iref
,
1831 bytenr
, num_bytes
, parent
,
1832 root_objectid
, owner
, offset
, 1);
1834 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
);
1835 update_inline_extent_backref(root
, path
, iref
,
1836 refs_to_add
, extent_op
, NULL
);
1837 } else if (ret
== -ENOENT
) {
1838 setup_inline_extent_backref(root
, path
, iref
, parent
,
1839 root_objectid
, owner
, offset
,
1840 refs_to_add
, extent_op
);
1846 static int insert_extent_backref(struct btrfs_trans_handle
*trans
,
1847 struct btrfs_root
*root
,
1848 struct btrfs_path
*path
,
1849 u64 bytenr
, u64 parent
, u64 root_objectid
,
1850 u64 owner
, u64 offset
, int refs_to_add
)
1853 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1854 BUG_ON(refs_to_add
!= 1);
1855 ret
= insert_tree_block_ref(trans
, root
, path
, bytenr
,
1856 parent
, root_objectid
);
1858 ret
= insert_extent_data_ref(trans
, root
, path
, bytenr
,
1859 parent
, root_objectid
,
1860 owner
, offset
, refs_to_add
);
1865 static int remove_extent_backref(struct btrfs_trans_handle
*trans
,
1866 struct btrfs_root
*root
,
1867 struct btrfs_path
*path
,
1868 struct btrfs_extent_inline_ref
*iref
,
1869 int refs_to_drop
, int is_data
, int *last_ref
)
1873 BUG_ON(!is_data
&& refs_to_drop
!= 1);
1875 update_inline_extent_backref(root
, path
, iref
,
1876 -refs_to_drop
, NULL
, last_ref
);
1877 } else if (is_data
) {
1878 ret
= remove_extent_data_ref(trans
, root
, path
, refs_to_drop
,
1882 ret
= btrfs_del_item(trans
, root
, path
);
1887 static int btrfs_issue_discard(struct block_device
*bdev
,
1890 return blkdev_issue_discard(bdev
, start
>> 9, len
>> 9, GFP_NOFS
, 0);
1893 int btrfs_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
1894 u64 num_bytes
, u64
*actual_bytes
)
1897 u64 discarded_bytes
= 0;
1898 struct btrfs_bio
*bbio
= NULL
;
1901 /* Tell the block device(s) that the sectors can be discarded */
1902 ret
= btrfs_map_block(root
->fs_info
, REQ_DISCARD
,
1903 bytenr
, &num_bytes
, &bbio
, 0);
1904 /* Error condition is -ENOMEM */
1906 struct btrfs_bio_stripe
*stripe
= bbio
->stripes
;
1910 for (i
= 0; i
< bbio
->num_stripes
; i
++, stripe
++) {
1911 if (!stripe
->dev
->can_discard
)
1914 ret
= btrfs_issue_discard(stripe
->dev
->bdev
,
1918 discarded_bytes
+= stripe
->length
;
1919 else if (ret
!= -EOPNOTSUPP
)
1920 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1923 * Just in case we get back EOPNOTSUPP for some reason,
1924 * just ignore the return value so we don't screw up
1925 * people calling discard_extent.
1929 btrfs_put_bbio(bbio
);
1933 *actual_bytes
= discarded_bytes
;
1936 if (ret
== -EOPNOTSUPP
)
1941 /* Can return -ENOMEM */
1942 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1943 struct btrfs_root
*root
,
1944 u64 bytenr
, u64 num_bytes
, u64 parent
,
1945 u64 root_objectid
, u64 owner
, u64 offset
,
1949 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1951 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
&&
1952 root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
1954 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1955 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
1957 parent
, root_objectid
, (int)owner
,
1958 BTRFS_ADD_DELAYED_REF
, NULL
, no_quota
);
1960 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
1962 parent
, root_objectid
, owner
, offset
,
1963 BTRFS_ADD_DELAYED_REF
, NULL
, no_quota
);
1968 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1969 struct btrfs_root
*root
,
1970 u64 bytenr
, u64 num_bytes
,
1971 u64 parent
, u64 root_objectid
,
1972 u64 owner
, u64 offset
, int refs_to_add
,
1974 struct btrfs_delayed_extent_op
*extent_op
)
1976 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1977 struct btrfs_path
*path
;
1978 struct extent_buffer
*leaf
;
1979 struct btrfs_extent_item
*item
;
1980 struct btrfs_key key
;
1983 enum btrfs_qgroup_operation_type type
= BTRFS_QGROUP_OPER_ADD_EXCL
;
1985 path
= btrfs_alloc_path();
1989 if (!is_fstree(root_objectid
) || !root
->fs_info
->quota_enabled
)
1993 path
->leave_spinning
= 1;
1994 /* this will setup the path even if it fails to insert the back ref */
1995 ret
= insert_inline_extent_backref(trans
, fs_info
->extent_root
, path
,
1996 bytenr
, num_bytes
, parent
,
1997 root_objectid
, owner
, offset
,
1998 refs_to_add
, extent_op
);
1999 if ((ret
< 0 && ret
!= -EAGAIN
) || (!ret
&& no_quota
))
2002 * Ok we were able to insert an inline extent and it appears to be a new
2003 * reference, deal with the qgroup accounting.
2005 if (!ret
&& !no_quota
) {
2006 ASSERT(root
->fs_info
->quota_enabled
);
2007 leaf
= path
->nodes
[0];
2008 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2009 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2010 struct btrfs_extent_item
);
2011 if (btrfs_extent_refs(leaf
, item
) > (u64
)refs_to_add
)
2012 type
= BTRFS_QGROUP_OPER_ADD_SHARED
;
2013 btrfs_release_path(path
);
2015 ret
= btrfs_qgroup_record_ref(trans
, fs_info
, root_objectid
,
2016 bytenr
, num_bytes
, type
, 0);
2021 * Ok we had -EAGAIN which means we didn't have space to insert and
2022 * inline extent ref, so just update the reference count and add a
2025 leaf
= path
->nodes
[0];
2026 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2027 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2028 refs
= btrfs_extent_refs(leaf
, item
);
2030 type
= BTRFS_QGROUP_OPER_ADD_SHARED
;
2031 btrfs_set_extent_refs(leaf
, item
, refs
+ refs_to_add
);
2033 __run_delayed_extent_op(extent_op
, leaf
, item
);
2035 btrfs_mark_buffer_dirty(leaf
);
2036 btrfs_release_path(path
);
2039 ret
= btrfs_qgroup_record_ref(trans
, fs_info
, root_objectid
,
2040 bytenr
, num_bytes
, type
, 0);
2046 path
->leave_spinning
= 1;
2047 /* now insert the actual backref */
2048 ret
= insert_extent_backref(trans
, root
->fs_info
->extent_root
,
2049 path
, bytenr
, parent
, root_objectid
,
2050 owner
, offset
, refs_to_add
);
2052 btrfs_abort_transaction(trans
, root
, ret
);
2054 btrfs_free_path(path
);
2058 static int run_delayed_data_ref(struct btrfs_trans_handle
*trans
,
2059 struct btrfs_root
*root
,
2060 struct btrfs_delayed_ref_node
*node
,
2061 struct btrfs_delayed_extent_op
*extent_op
,
2062 int insert_reserved
)
2065 struct btrfs_delayed_data_ref
*ref
;
2066 struct btrfs_key ins
;
2071 ins
.objectid
= node
->bytenr
;
2072 ins
.offset
= node
->num_bytes
;
2073 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2075 ref
= btrfs_delayed_node_to_data_ref(node
);
2076 trace_run_delayed_data_ref(node
, ref
, node
->action
);
2078 if (node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2079 parent
= ref
->parent
;
2080 ref_root
= ref
->root
;
2082 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2084 flags
|= extent_op
->flags_to_set
;
2085 ret
= alloc_reserved_file_extent(trans
, root
,
2086 parent
, ref_root
, flags
,
2087 ref
->objectid
, ref
->offset
,
2088 &ins
, node
->ref_mod
);
2089 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2090 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
2091 node
->num_bytes
, parent
,
2092 ref_root
, ref
->objectid
,
2093 ref
->offset
, node
->ref_mod
,
2094 node
->no_quota
, extent_op
);
2095 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2096 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
2097 node
->num_bytes
, parent
,
2098 ref_root
, ref
->objectid
,
2099 ref
->offset
, node
->ref_mod
,
2100 extent_op
, node
->no_quota
);
2107 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
2108 struct extent_buffer
*leaf
,
2109 struct btrfs_extent_item
*ei
)
2111 u64 flags
= btrfs_extent_flags(leaf
, ei
);
2112 if (extent_op
->update_flags
) {
2113 flags
|= extent_op
->flags_to_set
;
2114 btrfs_set_extent_flags(leaf
, ei
, flags
);
2117 if (extent_op
->update_key
) {
2118 struct btrfs_tree_block_info
*bi
;
2119 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
));
2120 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
2121 btrfs_set_tree_block_key(leaf
, bi
, &extent_op
->key
);
2125 static int run_delayed_extent_op(struct btrfs_trans_handle
*trans
,
2126 struct btrfs_root
*root
,
2127 struct btrfs_delayed_ref_node
*node
,
2128 struct btrfs_delayed_extent_op
*extent_op
)
2130 struct btrfs_key key
;
2131 struct btrfs_path
*path
;
2132 struct btrfs_extent_item
*ei
;
2133 struct extent_buffer
*leaf
;
2137 int metadata
= !extent_op
->is_data
;
2142 if (metadata
&& !btrfs_fs_incompat(root
->fs_info
, SKINNY_METADATA
))
2145 path
= btrfs_alloc_path();
2149 key
.objectid
= node
->bytenr
;
2152 key
.type
= BTRFS_METADATA_ITEM_KEY
;
2153 key
.offset
= extent_op
->level
;
2155 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2156 key
.offset
= node
->num_bytes
;
2161 path
->leave_spinning
= 1;
2162 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
,
2170 if (path
->slots
[0] > 0) {
2172 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
2174 if (key
.objectid
== node
->bytenr
&&
2175 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
2176 key
.offset
== node
->num_bytes
)
2180 btrfs_release_path(path
);
2183 key
.objectid
= node
->bytenr
;
2184 key
.offset
= node
->num_bytes
;
2185 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2194 leaf
= path
->nodes
[0];
2195 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2196 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2197 if (item_size
< sizeof(*ei
)) {
2198 ret
= convert_extent_item_v0(trans
, root
->fs_info
->extent_root
,
2204 leaf
= path
->nodes
[0];
2205 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2208 BUG_ON(item_size
< sizeof(*ei
));
2209 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2210 __run_delayed_extent_op(extent_op
, leaf
, ei
);
2212 btrfs_mark_buffer_dirty(leaf
);
2214 btrfs_free_path(path
);
2218 static int run_delayed_tree_ref(struct btrfs_trans_handle
*trans
,
2219 struct btrfs_root
*root
,
2220 struct btrfs_delayed_ref_node
*node
,
2221 struct btrfs_delayed_extent_op
*extent_op
,
2222 int insert_reserved
)
2225 struct btrfs_delayed_tree_ref
*ref
;
2226 struct btrfs_key ins
;
2229 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
2232 ref
= btrfs_delayed_node_to_tree_ref(node
);
2233 trace_run_delayed_tree_ref(node
, ref
, node
->action
);
2235 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2236 parent
= ref
->parent
;
2237 ref_root
= ref
->root
;
2239 ins
.objectid
= node
->bytenr
;
2240 if (skinny_metadata
) {
2241 ins
.offset
= ref
->level
;
2242 ins
.type
= BTRFS_METADATA_ITEM_KEY
;
2244 ins
.offset
= node
->num_bytes
;
2245 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2248 BUG_ON(node
->ref_mod
!= 1);
2249 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2250 BUG_ON(!extent_op
|| !extent_op
->update_flags
);
2251 ret
= alloc_reserved_tree_block(trans
, root
,
2253 extent_op
->flags_to_set
,
2257 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2258 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
2259 node
->num_bytes
, parent
, ref_root
,
2260 ref
->level
, 0, 1, node
->no_quota
,
2262 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2263 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
2264 node
->num_bytes
, parent
, ref_root
,
2265 ref
->level
, 0, 1, extent_op
,
2273 /* helper function to actually process a single delayed ref entry */
2274 static int run_one_delayed_ref(struct btrfs_trans_handle
*trans
,
2275 struct btrfs_root
*root
,
2276 struct btrfs_delayed_ref_node
*node
,
2277 struct btrfs_delayed_extent_op
*extent_op
,
2278 int insert_reserved
)
2282 if (trans
->aborted
) {
2283 if (insert_reserved
)
2284 btrfs_pin_extent(root
, node
->bytenr
,
2285 node
->num_bytes
, 1);
2289 if (btrfs_delayed_ref_is_head(node
)) {
2290 struct btrfs_delayed_ref_head
*head
;
2292 * we've hit the end of the chain and we were supposed
2293 * to insert this extent into the tree. But, it got
2294 * deleted before we ever needed to insert it, so all
2295 * we have to do is clean up the accounting
2298 head
= btrfs_delayed_node_to_head(node
);
2299 trace_run_delayed_ref_head(node
, head
, node
->action
);
2301 if (insert_reserved
) {
2302 btrfs_pin_extent(root
, node
->bytenr
,
2303 node
->num_bytes
, 1);
2304 if (head
->is_data
) {
2305 ret
= btrfs_del_csums(trans
, root
,
2313 if (node
->type
== BTRFS_TREE_BLOCK_REF_KEY
||
2314 node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2315 ret
= run_delayed_tree_ref(trans
, root
, node
, extent_op
,
2317 else if (node
->type
== BTRFS_EXTENT_DATA_REF_KEY
||
2318 node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2319 ret
= run_delayed_data_ref(trans
, root
, node
, extent_op
,
2326 static noinline
struct btrfs_delayed_ref_node
*
2327 select_delayed_ref(struct btrfs_delayed_ref_head
*head
)
2329 struct rb_node
*node
;
2330 struct btrfs_delayed_ref_node
*ref
, *last
= NULL
;;
2333 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2334 * this prevents ref count from going down to zero when
2335 * there still are pending delayed ref.
2337 node
= rb_first(&head
->ref_root
);
2339 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
2341 if (ref
->action
== BTRFS_ADD_DELAYED_REF
)
2343 else if (last
== NULL
)
2345 node
= rb_next(node
);
2351 * Returns 0 on success or if called with an already aborted transaction.
2352 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2354 static noinline
int __btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2355 struct btrfs_root
*root
,
2358 struct btrfs_delayed_ref_root
*delayed_refs
;
2359 struct btrfs_delayed_ref_node
*ref
;
2360 struct btrfs_delayed_ref_head
*locked_ref
= NULL
;
2361 struct btrfs_delayed_extent_op
*extent_op
;
2362 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2363 ktime_t start
= ktime_get();
2365 unsigned long count
= 0;
2366 unsigned long actual_count
= 0;
2367 int must_insert_reserved
= 0;
2369 delayed_refs
= &trans
->transaction
->delayed_refs
;
2375 spin_lock(&delayed_refs
->lock
);
2376 locked_ref
= btrfs_select_ref_head(trans
);
2378 spin_unlock(&delayed_refs
->lock
);
2382 /* grab the lock that says we are going to process
2383 * all the refs for this head */
2384 ret
= btrfs_delayed_ref_lock(trans
, locked_ref
);
2385 spin_unlock(&delayed_refs
->lock
);
2387 * we may have dropped the spin lock to get the head
2388 * mutex lock, and that might have given someone else
2389 * time to free the head. If that's true, it has been
2390 * removed from our list and we can move on.
2392 if (ret
== -EAGAIN
) {
2400 * We need to try and merge add/drops of the same ref since we
2401 * can run into issues with relocate dropping the implicit ref
2402 * and then it being added back again before the drop can
2403 * finish. If we merged anything we need to re-loop so we can
2406 spin_lock(&locked_ref
->lock
);
2407 btrfs_merge_delayed_refs(trans
, fs_info
, delayed_refs
,
2411 * locked_ref is the head node, so we have to go one
2412 * node back for any delayed ref updates
2414 ref
= select_delayed_ref(locked_ref
);
2416 if (ref
&& ref
->seq
&&
2417 btrfs_check_delayed_seq(fs_info
, delayed_refs
, ref
->seq
)) {
2418 spin_unlock(&locked_ref
->lock
);
2419 btrfs_delayed_ref_unlock(locked_ref
);
2420 spin_lock(&delayed_refs
->lock
);
2421 locked_ref
->processing
= 0;
2422 delayed_refs
->num_heads_ready
++;
2423 spin_unlock(&delayed_refs
->lock
);
2431 * record the must insert reserved flag before we
2432 * drop the spin lock.
2434 must_insert_reserved
= locked_ref
->must_insert_reserved
;
2435 locked_ref
->must_insert_reserved
= 0;
2437 extent_op
= locked_ref
->extent_op
;
2438 locked_ref
->extent_op
= NULL
;
2443 /* All delayed refs have been processed, Go ahead
2444 * and send the head node to run_one_delayed_ref,
2445 * so that any accounting fixes can happen
2447 ref
= &locked_ref
->node
;
2449 if (extent_op
&& must_insert_reserved
) {
2450 btrfs_free_delayed_extent_op(extent_op
);
2455 spin_unlock(&locked_ref
->lock
);
2456 ret
= run_delayed_extent_op(trans
, root
,
2458 btrfs_free_delayed_extent_op(extent_op
);
2462 * Need to reset must_insert_reserved if
2463 * there was an error so the abort stuff
2464 * can cleanup the reserved space
2467 if (must_insert_reserved
)
2468 locked_ref
->must_insert_reserved
= 1;
2469 locked_ref
->processing
= 0;
2470 btrfs_debug(fs_info
, "run_delayed_extent_op returned %d", ret
);
2471 btrfs_delayed_ref_unlock(locked_ref
);
2478 * Need to drop our head ref lock and re-aqcuire the
2479 * delayed ref lock and then re-check to make sure
2482 spin_unlock(&locked_ref
->lock
);
2483 spin_lock(&delayed_refs
->lock
);
2484 spin_lock(&locked_ref
->lock
);
2485 if (rb_first(&locked_ref
->ref_root
) ||
2486 locked_ref
->extent_op
) {
2487 spin_unlock(&locked_ref
->lock
);
2488 spin_unlock(&delayed_refs
->lock
);
2492 delayed_refs
->num_heads
--;
2493 rb_erase(&locked_ref
->href_node
,
2494 &delayed_refs
->href_root
);
2495 spin_unlock(&delayed_refs
->lock
);
2499 rb_erase(&ref
->rb_node
, &locked_ref
->ref_root
);
2501 atomic_dec(&delayed_refs
->num_entries
);
2503 if (!btrfs_delayed_ref_is_head(ref
)) {
2505 * when we play the delayed ref, also correct the
2508 switch (ref
->action
) {
2509 case BTRFS_ADD_DELAYED_REF
:
2510 case BTRFS_ADD_DELAYED_EXTENT
:
2511 locked_ref
->node
.ref_mod
-= ref
->ref_mod
;
2513 case BTRFS_DROP_DELAYED_REF
:
2514 locked_ref
->node
.ref_mod
+= ref
->ref_mod
;
2520 spin_unlock(&locked_ref
->lock
);
2522 ret
= run_one_delayed_ref(trans
, root
, ref
, extent_op
,
2523 must_insert_reserved
);
2525 btrfs_free_delayed_extent_op(extent_op
);
2527 locked_ref
->processing
= 0;
2528 btrfs_delayed_ref_unlock(locked_ref
);
2529 btrfs_put_delayed_ref(ref
);
2530 btrfs_debug(fs_info
, "run_one_delayed_ref returned %d", ret
);
2535 * If this node is a head, that means all the refs in this head
2536 * have been dealt with, and we will pick the next head to deal
2537 * with, so we must unlock the head and drop it from the cluster
2538 * list before we release it.
2540 if (btrfs_delayed_ref_is_head(ref
)) {
2541 btrfs_delayed_ref_unlock(locked_ref
);
2544 btrfs_put_delayed_ref(ref
);
2550 * We don't want to include ref heads since we can have empty ref heads
2551 * and those will drastically skew our runtime down since we just do
2552 * accounting, no actual extent tree updates.
2554 if (actual_count
> 0) {
2555 u64 runtime
= ktime_to_ns(ktime_sub(ktime_get(), start
));
2559 * We weigh the current average higher than our current runtime
2560 * to avoid large swings in the average.
2562 spin_lock(&delayed_refs
->lock
);
2563 avg
= fs_info
->avg_delayed_ref_runtime
* 3 + runtime
;
2564 avg
= div64_u64(avg
, 4);
2565 fs_info
->avg_delayed_ref_runtime
= avg
;
2566 spin_unlock(&delayed_refs
->lock
);
2571 #ifdef SCRAMBLE_DELAYED_REFS
2573 * Normally delayed refs get processed in ascending bytenr order. This
2574 * correlates in most cases to the order added. To expose dependencies on this
2575 * order, we start to process the tree in the middle instead of the beginning
2577 static u64
find_middle(struct rb_root
*root
)
2579 struct rb_node
*n
= root
->rb_node
;
2580 struct btrfs_delayed_ref_node
*entry
;
2583 u64 first
= 0, last
= 0;
2587 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2588 first
= entry
->bytenr
;
2592 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2593 last
= entry
->bytenr
;
2598 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2599 WARN_ON(!entry
->in_tree
);
2601 middle
= entry
->bytenr
;
2614 static inline u64
heads_to_leaves(struct btrfs_root
*root
, u64 heads
)
2618 num_bytes
= heads
* (sizeof(struct btrfs_extent_item
) +
2619 sizeof(struct btrfs_extent_inline_ref
));
2620 if (!btrfs_fs_incompat(root
->fs_info
, SKINNY_METADATA
))
2621 num_bytes
+= heads
* sizeof(struct btrfs_tree_block_info
);
2624 * We don't ever fill up leaves all the way so multiply by 2 just to be
2625 * closer to what we're really going to want to ouse.
2627 return div64_u64(num_bytes
, BTRFS_LEAF_DATA_SIZE(root
));
2630 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle
*trans
,
2631 struct btrfs_root
*root
)
2633 struct btrfs_block_rsv
*global_rsv
;
2634 u64 num_heads
= trans
->transaction
->delayed_refs
.num_heads_ready
;
2638 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
2639 num_heads
= heads_to_leaves(root
, num_heads
);
2641 num_bytes
+= (num_heads
- 1) * root
->nodesize
;
2643 global_rsv
= &root
->fs_info
->global_block_rsv
;
2646 * If we can't allocate any more chunks lets make sure we have _lots_ of
2647 * wiggle room since running delayed refs can create more delayed refs.
2649 if (global_rsv
->space_info
->full
)
2652 spin_lock(&global_rsv
->lock
);
2653 if (global_rsv
->reserved
<= num_bytes
)
2655 spin_unlock(&global_rsv
->lock
);
2659 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle
*trans
,
2660 struct btrfs_root
*root
)
2662 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2664 atomic_read(&trans
->transaction
->delayed_refs
.num_entries
);
2669 avg_runtime
= fs_info
->avg_delayed_ref_runtime
;
2670 val
= num_entries
* avg_runtime
;
2671 if (num_entries
* avg_runtime
>= NSEC_PER_SEC
)
2673 if (val
>= NSEC_PER_SEC
/ 2)
2676 return btrfs_check_space_for_delayed_refs(trans
, root
);
2679 struct async_delayed_refs
{
2680 struct btrfs_root
*root
;
2684 struct completion wait
;
2685 struct btrfs_work work
;
2688 static void delayed_ref_async_start(struct btrfs_work
*work
)
2690 struct async_delayed_refs
*async
;
2691 struct btrfs_trans_handle
*trans
;
2694 async
= container_of(work
, struct async_delayed_refs
, work
);
2696 trans
= btrfs_join_transaction(async
->root
);
2697 if (IS_ERR(trans
)) {
2698 async
->error
= PTR_ERR(trans
);
2703 * trans->sync means that when we call end_transaciton, we won't
2704 * wait on delayed refs
2707 ret
= btrfs_run_delayed_refs(trans
, async
->root
, async
->count
);
2711 ret
= btrfs_end_transaction(trans
, async
->root
);
2712 if (ret
&& !async
->error
)
2716 complete(&async
->wait
);
2721 int btrfs_async_run_delayed_refs(struct btrfs_root
*root
,
2722 unsigned long count
, int wait
)
2724 struct async_delayed_refs
*async
;
2727 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
2731 async
->root
= root
->fs_info
->tree_root
;
2732 async
->count
= count
;
2738 init_completion(&async
->wait
);
2740 btrfs_init_work(&async
->work
, btrfs_extent_refs_helper
,
2741 delayed_ref_async_start
, NULL
, NULL
);
2743 btrfs_queue_work(root
->fs_info
->extent_workers
, &async
->work
);
2746 wait_for_completion(&async
->wait
);
2755 * this starts processing the delayed reference count updates and
2756 * extent insertions we have queued up so far. count can be
2757 * 0, which means to process everything in the tree at the start
2758 * of the run (but not newly added entries), or it can be some target
2759 * number you'd like to process.
2761 * Returns 0 on success or if called with an aborted transaction
2762 * Returns <0 on error and aborts the transaction
2764 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2765 struct btrfs_root
*root
, unsigned long count
)
2767 struct rb_node
*node
;
2768 struct btrfs_delayed_ref_root
*delayed_refs
;
2769 struct btrfs_delayed_ref_head
*head
;
2771 int run_all
= count
== (unsigned long)-1;
2773 /* We'll clean this up in btrfs_cleanup_transaction */
2777 if (root
== root
->fs_info
->extent_root
)
2778 root
= root
->fs_info
->tree_root
;
2780 delayed_refs
= &trans
->transaction
->delayed_refs
;
2782 count
= atomic_read(&delayed_refs
->num_entries
) * 2;
2785 #ifdef SCRAMBLE_DELAYED_REFS
2786 delayed_refs
->run_delayed_start
= find_middle(&delayed_refs
->root
);
2788 ret
= __btrfs_run_delayed_refs(trans
, root
, count
);
2790 btrfs_abort_transaction(trans
, root
, ret
);
2795 if (!list_empty(&trans
->new_bgs
))
2796 btrfs_create_pending_block_groups(trans
, root
);
2798 spin_lock(&delayed_refs
->lock
);
2799 node
= rb_first(&delayed_refs
->href_root
);
2801 spin_unlock(&delayed_refs
->lock
);
2804 count
= (unsigned long)-1;
2807 head
= rb_entry(node
, struct btrfs_delayed_ref_head
,
2809 if (btrfs_delayed_ref_is_head(&head
->node
)) {
2810 struct btrfs_delayed_ref_node
*ref
;
2813 atomic_inc(&ref
->refs
);
2815 spin_unlock(&delayed_refs
->lock
);
2817 * Mutex was contended, block until it's
2818 * released and try again
2820 mutex_lock(&head
->mutex
);
2821 mutex_unlock(&head
->mutex
);
2823 btrfs_put_delayed_ref(ref
);
2829 node
= rb_next(node
);
2831 spin_unlock(&delayed_refs
->lock
);
2836 ret
= btrfs_delayed_qgroup_accounting(trans
, root
->fs_info
);
2839 assert_qgroups_uptodate(trans
);
2843 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
2844 struct btrfs_root
*root
,
2845 u64 bytenr
, u64 num_bytes
, u64 flags
,
2846 int level
, int is_data
)
2848 struct btrfs_delayed_extent_op
*extent_op
;
2851 extent_op
= btrfs_alloc_delayed_extent_op();
2855 extent_op
->flags_to_set
= flags
;
2856 extent_op
->update_flags
= 1;
2857 extent_op
->update_key
= 0;
2858 extent_op
->is_data
= is_data
? 1 : 0;
2859 extent_op
->level
= level
;
2861 ret
= btrfs_add_delayed_extent_op(root
->fs_info
, trans
, bytenr
,
2862 num_bytes
, extent_op
);
2864 btrfs_free_delayed_extent_op(extent_op
);
2868 static noinline
int check_delayed_ref(struct btrfs_trans_handle
*trans
,
2869 struct btrfs_root
*root
,
2870 struct btrfs_path
*path
,
2871 u64 objectid
, u64 offset
, u64 bytenr
)
2873 struct btrfs_delayed_ref_head
*head
;
2874 struct btrfs_delayed_ref_node
*ref
;
2875 struct btrfs_delayed_data_ref
*data_ref
;
2876 struct btrfs_delayed_ref_root
*delayed_refs
;
2877 struct rb_node
*node
;
2880 delayed_refs
= &trans
->transaction
->delayed_refs
;
2881 spin_lock(&delayed_refs
->lock
);
2882 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
2884 spin_unlock(&delayed_refs
->lock
);
2888 if (!mutex_trylock(&head
->mutex
)) {
2889 atomic_inc(&head
->node
.refs
);
2890 spin_unlock(&delayed_refs
->lock
);
2892 btrfs_release_path(path
);
2895 * Mutex was contended, block until it's released and let
2898 mutex_lock(&head
->mutex
);
2899 mutex_unlock(&head
->mutex
);
2900 btrfs_put_delayed_ref(&head
->node
);
2903 spin_unlock(&delayed_refs
->lock
);
2905 spin_lock(&head
->lock
);
2906 node
= rb_first(&head
->ref_root
);
2908 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2909 node
= rb_next(node
);
2911 /* If it's a shared ref we know a cross reference exists */
2912 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
) {
2917 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
2920 * If our ref doesn't match the one we're currently looking at
2921 * then we have a cross reference.
2923 if (data_ref
->root
!= root
->root_key
.objectid
||
2924 data_ref
->objectid
!= objectid
||
2925 data_ref
->offset
!= offset
) {
2930 spin_unlock(&head
->lock
);
2931 mutex_unlock(&head
->mutex
);
2935 static noinline
int check_committed_ref(struct btrfs_trans_handle
*trans
,
2936 struct btrfs_root
*root
,
2937 struct btrfs_path
*path
,
2938 u64 objectid
, u64 offset
, u64 bytenr
)
2940 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2941 struct extent_buffer
*leaf
;
2942 struct btrfs_extent_data_ref
*ref
;
2943 struct btrfs_extent_inline_ref
*iref
;
2944 struct btrfs_extent_item
*ei
;
2945 struct btrfs_key key
;
2949 key
.objectid
= bytenr
;
2950 key
.offset
= (u64
)-1;
2951 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2953 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
2956 BUG_ON(ret
== 0); /* Corruption */
2959 if (path
->slots
[0] == 0)
2963 leaf
= path
->nodes
[0];
2964 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2966 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
2970 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2971 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2972 if (item_size
< sizeof(*ei
)) {
2973 WARN_ON(item_size
!= sizeof(struct btrfs_extent_item_v0
));
2977 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2979 if (item_size
!= sizeof(*ei
) +
2980 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
2983 if (btrfs_extent_generation(leaf
, ei
) <=
2984 btrfs_root_last_snapshot(&root
->root_item
))
2987 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
2988 if (btrfs_extent_inline_ref_type(leaf
, iref
) !=
2989 BTRFS_EXTENT_DATA_REF_KEY
)
2992 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
2993 if (btrfs_extent_refs(leaf
, ei
) !=
2994 btrfs_extent_data_ref_count(leaf
, ref
) ||
2995 btrfs_extent_data_ref_root(leaf
, ref
) !=
2996 root
->root_key
.objectid
||
2997 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
2998 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
3006 int btrfs_cross_ref_exist(struct btrfs_trans_handle
*trans
,
3007 struct btrfs_root
*root
,
3008 u64 objectid
, u64 offset
, u64 bytenr
)
3010 struct btrfs_path
*path
;
3014 path
= btrfs_alloc_path();
3019 ret
= check_committed_ref(trans
, root
, path
, objectid
,
3021 if (ret
&& ret
!= -ENOENT
)
3024 ret2
= check_delayed_ref(trans
, root
, path
, objectid
,
3026 } while (ret2
== -EAGAIN
);
3028 if (ret2
&& ret2
!= -ENOENT
) {
3033 if (ret
!= -ENOENT
|| ret2
!= -ENOENT
)
3036 btrfs_free_path(path
);
3037 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
3042 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
3043 struct btrfs_root
*root
,
3044 struct extent_buffer
*buf
,
3045 int full_backref
, int inc
)
3052 struct btrfs_key key
;
3053 struct btrfs_file_extent_item
*fi
;
3057 int (*process_func
)(struct btrfs_trans_handle
*, struct btrfs_root
*,
3058 u64
, u64
, u64
, u64
, u64
, u64
, int);
3061 if (btrfs_test_is_dummy_root(root
))
3064 ref_root
= btrfs_header_owner(buf
);
3065 nritems
= btrfs_header_nritems(buf
);
3066 level
= btrfs_header_level(buf
);
3068 if (!test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) && level
== 0)
3072 process_func
= btrfs_inc_extent_ref
;
3074 process_func
= btrfs_free_extent
;
3077 parent
= buf
->start
;
3081 for (i
= 0; i
< nritems
; i
++) {
3083 btrfs_item_key_to_cpu(buf
, &key
, i
);
3084 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
3086 fi
= btrfs_item_ptr(buf
, i
,
3087 struct btrfs_file_extent_item
);
3088 if (btrfs_file_extent_type(buf
, fi
) ==
3089 BTRFS_FILE_EXTENT_INLINE
)
3091 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
3095 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
3096 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
3097 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
3098 parent
, ref_root
, key
.objectid
,
3103 bytenr
= btrfs_node_blockptr(buf
, i
);
3104 num_bytes
= root
->nodesize
;
3105 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
3106 parent
, ref_root
, level
- 1, 0,
3117 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3118 struct extent_buffer
*buf
, int full_backref
)
3120 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1);
3123 int btrfs_dec_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3124 struct extent_buffer
*buf
, int full_backref
)
3126 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 0);
3129 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
3130 struct btrfs_root
*root
,
3131 struct btrfs_path
*path
,
3132 struct btrfs_block_group_cache
*cache
)
3135 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
3137 struct extent_buffer
*leaf
;
3139 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
3146 leaf
= path
->nodes
[0];
3147 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
3148 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
3149 btrfs_mark_buffer_dirty(leaf
);
3150 btrfs_release_path(path
);
3153 btrfs_abort_transaction(trans
, root
, ret
);
3158 static struct btrfs_block_group_cache
*
3159 next_block_group(struct btrfs_root
*root
,
3160 struct btrfs_block_group_cache
*cache
)
3162 struct rb_node
*node
;
3164 spin_lock(&root
->fs_info
->block_group_cache_lock
);
3166 /* If our block group was removed, we need a full search. */
3167 if (RB_EMPTY_NODE(&cache
->cache_node
)) {
3168 const u64 next_bytenr
= cache
->key
.objectid
+ cache
->key
.offset
;
3170 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
3171 btrfs_put_block_group(cache
);
3172 cache
= btrfs_lookup_first_block_group(root
->fs_info
,
3176 node
= rb_next(&cache
->cache_node
);
3177 btrfs_put_block_group(cache
);
3179 cache
= rb_entry(node
, struct btrfs_block_group_cache
,
3181 btrfs_get_block_group(cache
);
3184 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
3188 static int cache_save_setup(struct btrfs_block_group_cache
*block_group
,
3189 struct btrfs_trans_handle
*trans
,
3190 struct btrfs_path
*path
)
3192 struct btrfs_root
*root
= block_group
->fs_info
->tree_root
;
3193 struct inode
*inode
= NULL
;
3195 int dcs
= BTRFS_DC_ERROR
;
3201 * If this block group is smaller than 100 megs don't bother caching the
3204 if (block_group
->key
.offset
< (100 * 1024 * 1024)) {
3205 spin_lock(&block_group
->lock
);
3206 block_group
->disk_cache_state
= BTRFS_DC_WRITTEN
;
3207 spin_unlock(&block_group
->lock
);
3214 inode
= lookup_free_space_inode(root
, block_group
, path
);
3215 if (IS_ERR(inode
) && PTR_ERR(inode
) != -ENOENT
) {
3216 ret
= PTR_ERR(inode
);
3217 btrfs_release_path(path
);
3221 if (IS_ERR(inode
)) {
3225 if (block_group
->ro
)
3228 ret
= create_free_space_inode(root
, trans
, block_group
, path
);
3234 /* We've already setup this transaction, go ahead and exit */
3235 if (block_group
->cache_generation
== trans
->transid
&&
3236 i_size_read(inode
)) {
3237 dcs
= BTRFS_DC_SETUP
;
3242 * We want to set the generation to 0, that way if anything goes wrong
3243 * from here on out we know not to trust this cache when we load up next
3246 BTRFS_I(inode
)->generation
= 0;
3247 ret
= btrfs_update_inode(trans
, root
, inode
);
3250 * So theoretically we could recover from this, simply set the
3251 * super cache generation to 0 so we know to invalidate the
3252 * cache, but then we'd have to keep track of the block groups
3253 * that fail this way so we know we _have_ to reset this cache
3254 * before the next commit or risk reading stale cache. So to
3255 * limit our exposure to horrible edge cases lets just abort the
3256 * transaction, this only happens in really bad situations
3259 btrfs_abort_transaction(trans
, root
, ret
);
3264 if (i_size_read(inode
) > 0) {
3265 ret
= btrfs_check_trunc_cache_free_space(root
,
3266 &root
->fs_info
->global_block_rsv
);
3270 ret
= btrfs_truncate_free_space_cache(root
, trans
, inode
);
3275 spin_lock(&block_group
->lock
);
3276 if (block_group
->cached
!= BTRFS_CACHE_FINISHED
||
3277 !btrfs_test_opt(root
, SPACE_CACHE
) ||
3278 block_group
->delalloc_bytes
) {
3280 * don't bother trying to write stuff out _if_
3281 * a) we're not cached,
3282 * b) we're with nospace_cache mount option.
3284 dcs
= BTRFS_DC_WRITTEN
;
3285 spin_unlock(&block_group
->lock
);
3288 spin_unlock(&block_group
->lock
);
3291 * Try to preallocate enough space based on how big the block group is.
3292 * Keep in mind this has to include any pinned space which could end up
3293 * taking up quite a bit since it's not folded into the other space
3296 num_pages
= (int)div64_u64(block_group
->key
.offset
, 256 * 1024 * 1024);
3301 num_pages
*= PAGE_CACHE_SIZE
;
3303 ret
= btrfs_check_data_free_space(inode
, num_pages
);
3307 ret
= btrfs_prealloc_file_range_trans(inode
, trans
, 0, 0, num_pages
,
3308 num_pages
, num_pages
,
3311 dcs
= BTRFS_DC_SETUP
;
3312 btrfs_free_reserved_data_space(inode
, num_pages
);
3317 btrfs_release_path(path
);
3319 spin_lock(&block_group
->lock
);
3320 if (!ret
&& dcs
== BTRFS_DC_SETUP
)
3321 block_group
->cache_generation
= trans
->transid
;
3322 block_group
->disk_cache_state
= dcs
;
3323 spin_unlock(&block_group
->lock
);
3328 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
3329 struct btrfs_root
*root
)
3331 struct btrfs_block_group_cache
*cache
;
3332 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3334 struct btrfs_path
*path
;
3336 if (list_empty(&cur_trans
->dirty_bgs
))
3339 path
= btrfs_alloc_path();
3344 * We don't need the lock here since we are protected by the transaction
3345 * commit. We want to do the cache_save_setup first and then run the
3346 * delayed refs to make sure we have the best chance at doing this all
3349 while (!list_empty(&cur_trans
->dirty_bgs
)) {
3350 cache
= list_first_entry(&cur_trans
->dirty_bgs
,
3351 struct btrfs_block_group_cache
,
3353 list_del_init(&cache
->dirty_list
);
3354 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
)
3355 cache_save_setup(cache
, trans
, path
);
3357 ret
= btrfs_run_delayed_refs(trans
, root
,
3358 (unsigned long) -1);
3359 if (!ret
&& cache
->disk_cache_state
== BTRFS_DC_SETUP
)
3360 btrfs_write_out_cache(root
, trans
, cache
, path
);
3362 ret
= write_one_cache_group(trans
, root
, path
, cache
);
3363 btrfs_put_block_group(cache
);
3366 btrfs_free_path(path
);
3370 int btrfs_extent_readonly(struct btrfs_root
*root
, u64 bytenr
)
3372 struct btrfs_block_group_cache
*block_group
;
3375 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
3376 if (!block_group
|| block_group
->ro
)
3379 btrfs_put_block_group(block_group
);
3383 static const char *alloc_name(u64 flags
)
3386 case BTRFS_BLOCK_GROUP_METADATA
|BTRFS_BLOCK_GROUP_DATA
:
3388 case BTRFS_BLOCK_GROUP_METADATA
:
3390 case BTRFS_BLOCK_GROUP_DATA
:
3392 case BTRFS_BLOCK_GROUP_SYSTEM
:
3396 return "invalid-combination";
3400 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
3401 u64 total_bytes
, u64 bytes_used
,
3402 struct btrfs_space_info
**space_info
)
3404 struct btrfs_space_info
*found
;
3409 if (flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3410 BTRFS_BLOCK_GROUP_RAID10
))
3415 found
= __find_space_info(info
, flags
);
3417 spin_lock(&found
->lock
);
3418 found
->total_bytes
+= total_bytes
;
3419 found
->disk_total
+= total_bytes
* factor
;
3420 found
->bytes_used
+= bytes_used
;
3421 found
->disk_used
+= bytes_used
* factor
;
3423 spin_unlock(&found
->lock
);
3424 *space_info
= found
;
3427 found
= kzalloc(sizeof(*found
), GFP_NOFS
);
3431 ret
= percpu_counter_init(&found
->total_bytes_pinned
, 0, GFP_KERNEL
);
3437 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
3438 INIT_LIST_HEAD(&found
->block_groups
[i
]);
3439 init_rwsem(&found
->groups_sem
);
3440 spin_lock_init(&found
->lock
);
3441 found
->flags
= flags
& BTRFS_BLOCK_GROUP_TYPE_MASK
;
3442 found
->total_bytes
= total_bytes
;
3443 found
->disk_total
= total_bytes
* factor
;
3444 found
->bytes_used
= bytes_used
;
3445 found
->disk_used
= bytes_used
* factor
;
3446 found
->bytes_pinned
= 0;
3447 found
->bytes_reserved
= 0;
3448 found
->bytes_readonly
= 0;
3449 found
->bytes_may_use
= 0;
3451 found
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3452 found
->chunk_alloc
= 0;
3454 init_waitqueue_head(&found
->wait
);
3455 INIT_LIST_HEAD(&found
->ro_bgs
);
3457 ret
= kobject_init_and_add(&found
->kobj
, &space_info_ktype
,
3458 info
->space_info_kobj
, "%s",
3459 alloc_name(found
->flags
));
3465 *space_info
= found
;
3466 list_add_rcu(&found
->list
, &info
->space_info
);
3467 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3468 info
->data_sinfo
= found
;
3473 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
3475 u64 extra_flags
= chunk_to_extended(flags
) &
3476 BTRFS_EXTENDED_PROFILE_MASK
;
3478 write_seqlock(&fs_info
->profiles_lock
);
3479 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3480 fs_info
->avail_data_alloc_bits
|= extra_flags
;
3481 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3482 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
3483 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3484 fs_info
->avail_system_alloc_bits
|= extra_flags
;
3485 write_sequnlock(&fs_info
->profiles_lock
);
3489 * returns target flags in extended format or 0 if restripe for this
3490 * chunk_type is not in progress
3492 * should be called with either volume_mutex or balance_lock held
3494 static u64
get_restripe_target(struct btrfs_fs_info
*fs_info
, u64 flags
)
3496 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3502 if (flags
& BTRFS_BLOCK_GROUP_DATA
&&
3503 bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3504 target
= BTRFS_BLOCK_GROUP_DATA
| bctl
->data
.target
;
3505 } else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
&&
3506 bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3507 target
= BTRFS_BLOCK_GROUP_SYSTEM
| bctl
->sys
.target
;
3508 } else if (flags
& BTRFS_BLOCK_GROUP_METADATA
&&
3509 bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3510 target
= BTRFS_BLOCK_GROUP_METADATA
| bctl
->meta
.target
;
3517 * @flags: available profiles in extended format (see ctree.h)
3519 * Returns reduced profile in chunk format. If profile changing is in
3520 * progress (either running or paused) picks the target profile (if it's
3521 * already available), otherwise falls back to plain reducing.
3523 static u64
btrfs_reduce_alloc_profile(struct btrfs_root
*root
, u64 flags
)
3525 u64 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
3530 * see if restripe for this chunk_type is in progress, if so
3531 * try to reduce to the target profile
3533 spin_lock(&root
->fs_info
->balance_lock
);
3534 target
= get_restripe_target(root
->fs_info
, flags
);
3536 /* pick target profile only if it's already available */
3537 if ((flags
& target
) & BTRFS_EXTENDED_PROFILE_MASK
) {
3538 spin_unlock(&root
->fs_info
->balance_lock
);
3539 return extended_to_chunk(target
);
3542 spin_unlock(&root
->fs_info
->balance_lock
);
3544 /* First, mask out the RAID levels which aren't possible */
3545 if (num_devices
== 1)
3546 flags
&= ~(BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID0
|
3547 BTRFS_BLOCK_GROUP_RAID5
);
3548 if (num_devices
< 3)
3549 flags
&= ~BTRFS_BLOCK_GROUP_RAID6
;
3550 if (num_devices
< 4)
3551 flags
&= ~BTRFS_BLOCK_GROUP_RAID10
;
3553 tmp
= flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID0
|
3554 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID5
|
3555 BTRFS_BLOCK_GROUP_RAID6
| BTRFS_BLOCK_GROUP_RAID10
);
3558 if (tmp
& BTRFS_BLOCK_GROUP_RAID6
)
3559 tmp
= BTRFS_BLOCK_GROUP_RAID6
;
3560 else if (tmp
& BTRFS_BLOCK_GROUP_RAID5
)
3561 tmp
= BTRFS_BLOCK_GROUP_RAID5
;
3562 else if (tmp
& BTRFS_BLOCK_GROUP_RAID10
)
3563 tmp
= BTRFS_BLOCK_GROUP_RAID10
;
3564 else if (tmp
& BTRFS_BLOCK_GROUP_RAID1
)
3565 tmp
= BTRFS_BLOCK_GROUP_RAID1
;
3566 else if (tmp
& BTRFS_BLOCK_GROUP_RAID0
)
3567 tmp
= BTRFS_BLOCK_GROUP_RAID0
;
3569 return extended_to_chunk(flags
| tmp
);
3572 static u64
get_alloc_profile(struct btrfs_root
*root
, u64 orig_flags
)
3579 seq
= read_seqbegin(&root
->fs_info
->profiles_lock
);
3581 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3582 flags
|= root
->fs_info
->avail_data_alloc_bits
;
3583 else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3584 flags
|= root
->fs_info
->avail_system_alloc_bits
;
3585 else if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3586 flags
|= root
->fs_info
->avail_metadata_alloc_bits
;
3587 } while (read_seqretry(&root
->fs_info
->profiles_lock
, seq
));
3589 return btrfs_reduce_alloc_profile(root
, flags
);
3592 u64
btrfs_get_alloc_profile(struct btrfs_root
*root
, int data
)
3598 flags
= BTRFS_BLOCK_GROUP_DATA
;
3599 else if (root
== root
->fs_info
->chunk_root
)
3600 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
3602 flags
= BTRFS_BLOCK_GROUP_METADATA
;
3604 ret
= get_alloc_profile(root
, flags
);
3609 * This will check the space that the inode allocates from to make sure we have
3610 * enough space for bytes.
3612 int btrfs_check_data_free_space(struct inode
*inode
, u64 bytes
)
3614 struct btrfs_space_info
*data_sinfo
;
3615 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3616 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3618 int ret
= 0, committed
= 0, alloc_chunk
= 1;
3620 /* make sure bytes are sectorsize aligned */
3621 bytes
= ALIGN(bytes
, root
->sectorsize
);
3623 if (btrfs_is_free_space_inode(inode
)) {
3625 ASSERT(current
->journal_info
);
3628 data_sinfo
= fs_info
->data_sinfo
;
3633 /* make sure we have enough space to handle the data first */
3634 spin_lock(&data_sinfo
->lock
);
3635 used
= data_sinfo
->bytes_used
+ data_sinfo
->bytes_reserved
+
3636 data_sinfo
->bytes_pinned
+ data_sinfo
->bytes_readonly
+
3637 data_sinfo
->bytes_may_use
;
3639 if (used
+ bytes
> data_sinfo
->total_bytes
) {
3640 struct btrfs_trans_handle
*trans
;
3643 * if we don't have enough free bytes in this space then we need
3644 * to alloc a new chunk.
3646 if (!data_sinfo
->full
&& alloc_chunk
) {
3649 data_sinfo
->force_alloc
= CHUNK_ALLOC_FORCE
;
3650 spin_unlock(&data_sinfo
->lock
);
3652 alloc_target
= btrfs_get_alloc_profile(root
, 1);
3654 * It is ugly that we don't call nolock join
3655 * transaction for the free space inode case here.
3656 * But it is safe because we only do the data space
3657 * reservation for the free space cache in the
3658 * transaction context, the common join transaction
3659 * just increase the counter of the current transaction
3660 * handler, doesn't try to acquire the trans_lock of
3663 trans
= btrfs_join_transaction(root
);
3665 return PTR_ERR(trans
);
3667 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
3669 CHUNK_ALLOC_NO_FORCE
);
3670 btrfs_end_transaction(trans
, root
);
3679 data_sinfo
= fs_info
->data_sinfo
;
3685 * If we don't have enough pinned space to deal with this
3686 * allocation don't bother committing the transaction.
3688 if (percpu_counter_compare(&data_sinfo
->total_bytes_pinned
,
3691 spin_unlock(&data_sinfo
->lock
);
3693 /* commit the current transaction and try again */
3696 !atomic_read(&root
->fs_info
->open_ioctl_trans
)) {
3699 trans
= btrfs_join_transaction(root
);
3701 return PTR_ERR(trans
);
3702 ret
= btrfs_commit_transaction(trans
, root
);
3708 trace_btrfs_space_reservation(root
->fs_info
,
3709 "space_info:enospc",
3710 data_sinfo
->flags
, bytes
, 1);
3713 data_sinfo
->bytes_may_use
+= bytes
;
3714 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
3715 data_sinfo
->flags
, bytes
, 1);
3716 spin_unlock(&data_sinfo
->lock
);
3722 * Called if we need to clear a data reservation for this inode.
3724 void btrfs_free_reserved_data_space(struct inode
*inode
, u64 bytes
)
3726 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3727 struct btrfs_space_info
*data_sinfo
;
3729 /* make sure bytes are sectorsize aligned */
3730 bytes
= ALIGN(bytes
, root
->sectorsize
);
3732 data_sinfo
= root
->fs_info
->data_sinfo
;
3733 spin_lock(&data_sinfo
->lock
);
3734 WARN_ON(data_sinfo
->bytes_may_use
< bytes
);
3735 data_sinfo
->bytes_may_use
-= bytes
;
3736 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
3737 data_sinfo
->flags
, bytes
, 0);
3738 spin_unlock(&data_sinfo
->lock
);
3741 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
3743 struct list_head
*head
= &info
->space_info
;
3744 struct btrfs_space_info
*found
;
3747 list_for_each_entry_rcu(found
, head
, list
) {
3748 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
3749 found
->force_alloc
= CHUNK_ALLOC_FORCE
;
3754 static inline u64
calc_global_rsv_need_space(struct btrfs_block_rsv
*global
)
3756 return (global
->size
<< 1);
3759 static int should_alloc_chunk(struct btrfs_root
*root
,
3760 struct btrfs_space_info
*sinfo
, int force
)
3762 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
3763 u64 num_bytes
= sinfo
->total_bytes
- sinfo
->bytes_readonly
;
3764 u64 num_allocated
= sinfo
->bytes_used
+ sinfo
->bytes_reserved
;
3767 if (force
== CHUNK_ALLOC_FORCE
)
3771 * We need to take into account the global rsv because for all intents
3772 * and purposes it's used space. Don't worry about locking the
3773 * global_rsv, it doesn't change except when the transaction commits.
3775 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
3776 num_allocated
+= calc_global_rsv_need_space(global_rsv
);
3779 * in limited mode, we want to have some free space up to
3780 * about 1% of the FS size.
3782 if (force
== CHUNK_ALLOC_LIMITED
) {
3783 thresh
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
3784 thresh
= max_t(u64
, 64 * 1024 * 1024,
3785 div_factor_fine(thresh
, 1));
3787 if (num_bytes
- num_allocated
< thresh
)
3791 if (num_allocated
+ 2 * 1024 * 1024 < div_factor(num_bytes
, 8))
3796 static u64
get_system_chunk_thresh(struct btrfs_root
*root
, u64 type
)
3800 if (type
& (BTRFS_BLOCK_GROUP_RAID10
|
3801 BTRFS_BLOCK_GROUP_RAID0
|
3802 BTRFS_BLOCK_GROUP_RAID5
|
3803 BTRFS_BLOCK_GROUP_RAID6
))
3804 num_dev
= root
->fs_info
->fs_devices
->rw_devices
;
3805 else if (type
& BTRFS_BLOCK_GROUP_RAID1
)
3808 num_dev
= 1; /* DUP or single */
3810 /* metadata for updaing devices and chunk tree */
3811 return btrfs_calc_trans_metadata_size(root
, num_dev
+ 1);
3814 static void check_system_chunk(struct btrfs_trans_handle
*trans
,
3815 struct btrfs_root
*root
, u64 type
)
3817 struct btrfs_space_info
*info
;
3821 info
= __find_space_info(root
->fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
3822 spin_lock(&info
->lock
);
3823 left
= info
->total_bytes
- info
->bytes_used
- info
->bytes_pinned
-
3824 info
->bytes_reserved
- info
->bytes_readonly
;
3825 spin_unlock(&info
->lock
);
3827 thresh
= get_system_chunk_thresh(root
, type
);
3828 if (left
< thresh
&& btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
3829 btrfs_info(root
->fs_info
, "left=%llu, need=%llu, flags=%llu",
3830 left
, thresh
, type
);
3831 dump_space_info(info
, 0, 0);
3834 if (left
< thresh
) {
3837 flags
= btrfs_get_alloc_profile(root
->fs_info
->chunk_root
, 0);
3838 btrfs_alloc_chunk(trans
, root
, flags
);
3842 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
3843 struct btrfs_root
*extent_root
, u64 flags
, int force
)
3845 struct btrfs_space_info
*space_info
;
3846 struct btrfs_fs_info
*fs_info
= extent_root
->fs_info
;
3847 int wait_for_alloc
= 0;
3850 /* Don't re-enter if we're already allocating a chunk */
3851 if (trans
->allocating_chunk
)
3854 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
3856 ret
= update_space_info(extent_root
->fs_info
, flags
,
3858 BUG_ON(ret
); /* -ENOMEM */
3860 BUG_ON(!space_info
); /* Logic error */
3863 spin_lock(&space_info
->lock
);
3864 if (force
< space_info
->force_alloc
)
3865 force
= space_info
->force_alloc
;
3866 if (space_info
->full
) {
3867 if (should_alloc_chunk(extent_root
, space_info
, force
))
3871 spin_unlock(&space_info
->lock
);
3875 if (!should_alloc_chunk(extent_root
, space_info
, force
)) {
3876 spin_unlock(&space_info
->lock
);
3878 } else if (space_info
->chunk_alloc
) {
3881 space_info
->chunk_alloc
= 1;
3884 spin_unlock(&space_info
->lock
);
3886 mutex_lock(&fs_info
->chunk_mutex
);
3889 * The chunk_mutex is held throughout the entirety of a chunk
3890 * allocation, so once we've acquired the chunk_mutex we know that the
3891 * other guy is done and we need to recheck and see if we should
3894 if (wait_for_alloc
) {
3895 mutex_unlock(&fs_info
->chunk_mutex
);
3900 trans
->allocating_chunk
= true;
3903 * If we have mixed data/metadata chunks we want to make sure we keep
3904 * allocating mixed chunks instead of individual chunks.
3906 if (btrfs_mixed_space_info(space_info
))
3907 flags
|= (BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
);
3910 * if we're doing a data chunk, go ahead and make sure that
3911 * we keep a reasonable number of metadata chunks allocated in the
3914 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
3915 fs_info
->data_chunk_allocations
++;
3916 if (!(fs_info
->data_chunk_allocations
%
3917 fs_info
->metadata_ratio
))
3918 force_metadata_allocation(fs_info
);
3922 * Check if we have enough space in SYSTEM chunk because we may need
3923 * to update devices.
3925 check_system_chunk(trans
, extent_root
, flags
);
3927 ret
= btrfs_alloc_chunk(trans
, extent_root
, flags
);
3928 trans
->allocating_chunk
= false;
3930 spin_lock(&space_info
->lock
);
3931 if (ret
< 0 && ret
!= -ENOSPC
)
3934 space_info
->full
= 1;
3938 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3940 space_info
->chunk_alloc
= 0;
3941 spin_unlock(&space_info
->lock
);
3942 mutex_unlock(&fs_info
->chunk_mutex
);
3946 static int can_overcommit(struct btrfs_root
*root
,
3947 struct btrfs_space_info
*space_info
, u64 bytes
,
3948 enum btrfs_reserve_flush_enum flush
)
3950 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
3951 u64 profile
= btrfs_get_alloc_profile(root
, 0);
3956 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
3957 space_info
->bytes_pinned
+ space_info
->bytes_readonly
;
3960 * We only want to allow over committing if we have lots of actual space
3961 * free, but if we don't have enough space to handle the global reserve
3962 * space then we could end up having a real enospc problem when trying
3963 * to allocate a chunk or some other such important allocation.
3965 spin_lock(&global_rsv
->lock
);
3966 space_size
= calc_global_rsv_need_space(global_rsv
);
3967 spin_unlock(&global_rsv
->lock
);
3968 if (used
+ space_size
>= space_info
->total_bytes
)
3971 used
+= space_info
->bytes_may_use
;
3973 spin_lock(&root
->fs_info
->free_chunk_lock
);
3974 avail
= root
->fs_info
->free_chunk_space
;
3975 spin_unlock(&root
->fs_info
->free_chunk_lock
);
3978 * If we have dup, raid1 or raid10 then only half of the free
3979 * space is actually useable. For raid56, the space info used
3980 * doesn't include the parity drive, so we don't have to
3983 if (profile
& (BTRFS_BLOCK_GROUP_DUP
|
3984 BTRFS_BLOCK_GROUP_RAID1
|
3985 BTRFS_BLOCK_GROUP_RAID10
))
3989 * If we aren't flushing all things, let us overcommit up to
3990 * 1/2th of the space. If we can flush, don't let us overcommit
3991 * too much, let it overcommit up to 1/8 of the space.
3993 if (flush
== BTRFS_RESERVE_FLUSH_ALL
)
3998 if (used
+ bytes
< space_info
->total_bytes
+ avail
)
4003 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root
*root
,
4004 unsigned long nr_pages
, int nr_items
)
4006 struct super_block
*sb
= root
->fs_info
->sb
;
4008 if (down_read_trylock(&sb
->s_umount
)) {
4009 writeback_inodes_sb_nr(sb
, nr_pages
, WB_REASON_FS_FREE_SPACE
);
4010 up_read(&sb
->s_umount
);
4013 * We needn't worry the filesystem going from r/w to r/o though
4014 * we don't acquire ->s_umount mutex, because the filesystem
4015 * should guarantee the delalloc inodes list be empty after
4016 * the filesystem is readonly(all dirty pages are written to
4019 btrfs_start_delalloc_roots(root
->fs_info
, 0, nr_items
);
4020 if (!current
->journal_info
)
4021 btrfs_wait_ordered_roots(root
->fs_info
, nr_items
);
4025 static inline int calc_reclaim_items_nr(struct btrfs_root
*root
, u64 to_reclaim
)
4030 bytes
= btrfs_calc_trans_metadata_size(root
, 1);
4031 nr
= (int)div64_u64(to_reclaim
, bytes
);
4037 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4040 * shrink metadata reservation for delalloc
4042 static void shrink_delalloc(struct btrfs_root
*root
, u64 to_reclaim
, u64 orig
,
4045 struct btrfs_block_rsv
*block_rsv
;
4046 struct btrfs_space_info
*space_info
;
4047 struct btrfs_trans_handle
*trans
;
4051 unsigned long nr_pages
;
4054 enum btrfs_reserve_flush_enum flush
;
4056 /* Calc the number of the pages we need flush for space reservation */
4057 items
= calc_reclaim_items_nr(root
, to_reclaim
);
4058 to_reclaim
= items
* EXTENT_SIZE_PER_ITEM
;
4060 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4061 block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
4062 space_info
= block_rsv
->space_info
;
4064 delalloc_bytes
= percpu_counter_sum_positive(
4065 &root
->fs_info
->delalloc_bytes
);
4066 if (delalloc_bytes
== 0) {
4070 btrfs_wait_ordered_roots(root
->fs_info
, items
);
4075 while (delalloc_bytes
&& loops
< 3) {
4076 max_reclaim
= min(delalloc_bytes
, to_reclaim
);
4077 nr_pages
= max_reclaim
>> PAGE_CACHE_SHIFT
;
4078 btrfs_writeback_inodes_sb_nr(root
, nr_pages
, items
);
4080 * We need to wait for the async pages to actually start before
4083 max_reclaim
= atomic_read(&root
->fs_info
->async_delalloc_pages
);
4087 if (max_reclaim
<= nr_pages
)
4090 max_reclaim
-= nr_pages
;
4092 wait_event(root
->fs_info
->async_submit_wait
,
4093 atomic_read(&root
->fs_info
->async_delalloc_pages
) <=
4097 flush
= BTRFS_RESERVE_FLUSH_ALL
;
4099 flush
= BTRFS_RESERVE_NO_FLUSH
;
4100 spin_lock(&space_info
->lock
);
4101 if (can_overcommit(root
, space_info
, orig
, flush
)) {
4102 spin_unlock(&space_info
->lock
);
4105 spin_unlock(&space_info
->lock
);
4108 if (wait_ordered
&& !trans
) {
4109 btrfs_wait_ordered_roots(root
->fs_info
, items
);
4111 time_left
= schedule_timeout_killable(1);
4115 delalloc_bytes
= percpu_counter_sum_positive(
4116 &root
->fs_info
->delalloc_bytes
);
4121 * maybe_commit_transaction - possibly commit the transaction if its ok to
4122 * @root - the root we're allocating for
4123 * @bytes - the number of bytes we want to reserve
4124 * @force - force the commit
4126 * This will check to make sure that committing the transaction will actually
4127 * get us somewhere and then commit the transaction if it does. Otherwise it
4128 * will return -ENOSPC.
4130 static int may_commit_transaction(struct btrfs_root
*root
,
4131 struct btrfs_space_info
*space_info
,
4132 u64 bytes
, int force
)
4134 struct btrfs_block_rsv
*delayed_rsv
= &root
->fs_info
->delayed_block_rsv
;
4135 struct btrfs_trans_handle
*trans
;
4137 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4144 /* See if there is enough pinned space to make this reservation */
4145 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4150 * See if there is some space in the delayed insertion reservation for
4153 if (space_info
!= delayed_rsv
->space_info
)
4156 spin_lock(&delayed_rsv
->lock
);
4157 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4158 bytes
- delayed_rsv
->size
) >= 0) {
4159 spin_unlock(&delayed_rsv
->lock
);
4162 spin_unlock(&delayed_rsv
->lock
);
4165 trans
= btrfs_join_transaction(root
);
4169 return btrfs_commit_transaction(trans
, root
);
4173 FLUSH_DELAYED_ITEMS_NR
= 1,
4174 FLUSH_DELAYED_ITEMS
= 2,
4176 FLUSH_DELALLOC_WAIT
= 4,
4181 static int flush_space(struct btrfs_root
*root
,
4182 struct btrfs_space_info
*space_info
, u64 num_bytes
,
4183 u64 orig_bytes
, int state
)
4185 struct btrfs_trans_handle
*trans
;
4190 case FLUSH_DELAYED_ITEMS_NR
:
4191 case FLUSH_DELAYED_ITEMS
:
4192 if (state
== FLUSH_DELAYED_ITEMS_NR
)
4193 nr
= calc_reclaim_items_nr(root
, num_bytes
) * 2;
4197 trans
= btrfs_join_transaction(root
);
4198 if (IS_ERR(trans
)) {
4199 ret
= PTR_ERR(trans
);
4202 ret
= btrfs_run_delayed_items_nr(trans
, root
, nr
);
4203 btrfs_end_transaction(trans
, root
);
4205 case FLUSH_DELALLOC
:
4206 case FLUSH_DELALLOC_WAIT
:
4207 shrink_delalloc(root
, num_bytes
* 2, orig_bytes
,
4208 state
== FLUSH_DELALLOC_WAIT
);
4211 trans
= btrfs_join_transaction(root
);
4212 if (IS_ERR(trans
)) {
4213 ret
= PTR_ERR(trans
);
4216 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
4217 btrfs_get_alloc_profile(root
, 0),
4218 CHUNK_ALLOC_NO_FORCE
);
4219 btrfs_end_transaction(trans
, root
);
4224 ret
= may_commit_transaction(root
, space_info
, orig_bytes
, 0);
4235 btrfs_calc_reclaim_metadata_size(struct btrfs_root
*root
,
4236 struct btrfs_space_info
*space_info
)
4242 to_reclaim
= min_t(u64
, num_online_cpus() * 1024 * 1024,
4244 spin_lock(&space_info
->lock
);
4245 if (can_overcommit(root
, space_info
, to_reclaim
,
4246 BTRFS_RESERVE_FLUSH_ALL
)) {
4251 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4252 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4253 space_info
->bytes_may_use
;
4254 if (can_overcommit(root
, space_info
, 1024 * 1024,
4255 BTRFS_RESERVE_FLUSH_ALL
))
4256 expected
= div_factor_fine(space_info
->total_bytes
, 95);
4258 expected
= div_factor_fine(space_info
->total_bytes
, 90);
4260 if (used
> expected
)
4261 to_reclaim
= used
- expected
;
4264 to_reclaim
= min(to_reclaim
, space_info
->bytes_may_use
+
4265 space_info
->bytes_reserved
);
4267 spin_unlock(&space_info
->lock
);
4272 static inline int need_do_async_reclaim(struct btrfs_space_info
*space_info
,
4273 struct btrfs_fs_info
*fs_info
, u64 used
)
4275 return (used
>= div_factor_fine(space_info
->total_bytes
, 98) &&
4276 !btrfs_fs_closing(fs_info
) &&
4277 !test_bit(BTRFS_FS_STATE_REMOUNTING
, &fs_info
->fs_state
));
4280 static int btrfs_need_do_async_reclaim(struct btrfs_space_info
*space_info
,
4281 struct btrfs_fs_info
*fs_info
,
4286 spin_lock(&space_info
->lock
);
4288 * We run out of space and have not got any free space via flush_space,
4289 * so don't bother doing async reclaim.
4291 if (flush_state
> COMMIT_TRANS
&& space_info
->full
) {
4292 spin_unlock(&space_info
->lock
);
4296 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4297 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4298 space_info
->bytes_may_use
;
4299 if (need_do_async_reclaim(space_info
, fs_info
, used
)) {
4300 spin_unlock(&space_info
->lock
);
4303 spin_unlock(&space_info
->lock
);
4308 static void btrfs_async_reclaim_metadata_space(struct work_struct
*work
)
4310 struct btrfs_fs_info
*fs_info
;
4311 struct btrfs_space_info
*space_info
;
4315 fs_info
= container_of(work
, struct btrfs_fs_info
, async_reclaim_work
);
4316 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4318 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
->fs_root
,
4323 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
4325 flush_space(fs_info
->fs_root
, space_info
, to_reclaim
,
4326 to_reclaim
, flush_state
);
4328 if (!btrfs_need_do_async_reclaim(space_info
, fs_info
,
4331 } while (flush_state
<= COMMIT_TRANS
);
4333 if (btrfs_need_do_async_reclaim(space_info
, fs_info
, flush_state
))
4334 queue_work(system_unbound_wq
, work
);
4337 void btrfs_init_async_reclaim_work(struct work_struct
*work
)
4339 INIT_WORK(work
, btrfs_async_reclaim_metadata_space
);
4343 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4344 * @root - the root we're allocating for
4345 * @block_rsv - the block_rsv we're allocating for
4346 * @orig_bytes - the number of bytes we want
4347 * @flush - whether or not we can flush to make our reservation
4349 * This will reserve orgi_bytes number of bytes from the space info associated
4350 * with the block_rsv. If there is not enough space it will make an attempt to
4351 * flush out space to make room. It will do this by flushing delalloc if
4352 * possible or committing the transaction. If flush is 0 then no attempts to
4353 * regain reservations will be made and this will fail if there is not enough
4356 static int reserve_metadata_bytes(struct btrfs_root
*root
,
4357 struct btrfs_block_rsv
*block_rsv
,
4359 enum btrfs_reserve_flush_enum flush
)
4361 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
4363 u64 num_bytes
= orig_bytes
;
4364 int flush_state
= FLUSH_DELAYED_ITEMS_NR
;
4366 bool flushing
= false;
4370 spin_lock(&space_info
->lock
);
4372 * We only want to wait if somebody other than us is flushing and we
4373 * are actually allowed to flush all things.
4375 while (flush
== BTRFS_RESERVE_FLUSH_ALL
&& !flushing
&&
4376 space_info
->flush
) {
4377 spin_unlock(&space_info
->lock
);
4379 * If we have a trans handle we can't wait because the flusher
4380 * may have to commit the transaction, which would mean we would
4381 * deadlock since we are waiting for the flusher to finish, but
4382 * hold the current transaction open.
4384 if (current
->journal_info
)
4386 ret
= wait_event_killable(space_info
->wait
, !space_info
->flush
);
4387 /* Must have been killed, return */
4391 spin_lock(&space_info
->lock
);
4395 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4396 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4397 space_info
->bytes_may_use
;
4400 * The idea here is that we've not already over-reserved the block group
4401 * then we can go ahead and save our reservation first and then start
4402 * flushing if we need to. Otherwise if we've already overcommitted
4403 * lets start flushing stuff first and then come back and try to make
4406 if (used
<= space_info
->total_bytes
) {
4407 if (used
+ orig_bytes
<= space_info
->total_bytes
) {
4408 space_info
->bytes_may_use
+= orig_bytes
;
4409 trace_btrfs_space_reservation(root
->fs_info
,
4410 "space_info", space_info
->flags
, orig_bytes
, 1);
4414 * Ok set num_bytes to orig_bytes since we aren't
4415 * overocmmitted, this way we only try and reclaim what
4418 num_bytes
= orig_bytes
;
4422 * Ok we're over committed, set num_bytes to the overcommitted
4423 * amount plus the amount of bytes that we need for this
4426 num_bytes
= used
- space_info
->total_bytes
+
4430 if (ret
&& can_overcommit(root
, space_info
, orig_bytes
, flush
)) {
4431 space_info
->bytes_may_use
+= orig_bytes
;
4432 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
4433 space_info
->flags
, orig_bytes
,
4439 * Couldn't make our reservation, save our place so while we're trying
4440 * to reclaim space we can actually use it instead of somebody else
4441 * stealing it from us.
4443 * We make the other tasks wait for the flush only when we can flush
4446 if (ret
&& flush
!= BTRFS_RESERVE_NO_FLUSH
) {
4448 space_info
->flush
= 1;
4449 } else if (!ret
&& space_info
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
4452 * We will do the space reservation dance during log replay,
4453 * which means we won't have fs_info->fs_root set, so don't do
4454 * the async reclaim as we will panic.
4456 if (!root
->fs_info
->log_root_recovering
&&
4457 need_do_async_reclaim(space_info
, root
->fs_info
, used
) &&
4458 !work_busy(&root
->fs_info
->async_reclaim_work
))
4459 queue_work(system_unbound_wq
,
4460 &root
->fs_info
->async_reclaim_work
);
4462 spin_unlock(&space_info
->lock
);
4464 if (!ret
|| flush
== BTRFS_RESERVE_NO_FLUSH
)
4467 ret
= flush_space(root
, space_info
, num_bytes
, orig_bytes
,
4472 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4473 * would happen. So skip delalloc flush.
4475 if (flush
== BTRFS_RESERVE_FLUSH_LIMIT
&&
4476 (flush_state
== FLUSH_DELALLOC
||
4477 flush_state
== FLUSH_DELALLOC_WAIT
))
4478 flush_state
= ALLOC_CHUNK
;
4482 else if (flush
== BTRFS_RESERVE_FLUSH_LIMIT
&&
4483 flush_state
< COMMIT_TRANS
)
4485 else if (flush
== BTRFS_RESERVE_FLUSH_ALL
&&
4486 flush_state
<= COMMIT_TRANS
)
4490 if (ret
== -ENOSPC
&&
4491 unlikely(root
->orphan_cleanup_state
== ORPHAN_CLEANUP_STARTED
)) {
4492 struct btrfs_block_rsv
*global_rsv
=
4493 &root
->fs_info
->global_block_rsv
;
4495 if (block_rsv
!= global_rsv
&&
4496 !block_rsv_use_bytes(global_rsv
, orig_bytes
))
4500 trace_btrfs_space_reservation(root
->fs_info
,
4501 "space_info:enospc",
4502 space_info
->flags
, orig_bytes
, 1);
4504 spin_lock(&space_info
->lock
);
4505 space_info
->flush
= 0;
4506 wake_up_all(&space_info
->wait
);
4507 spin_unlock(&space_info
->lock
);
4512 static struct btrfs_block_rsv
*get_block_rsv(
4513 const struct btrfs_trans_handle
*trans
,
4514 const struct btrfs_root
*root
)
4516 struct btrfs_block_rsv
*block_rsv
= NULL
;
4518 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
))
4519 block_rsv
= trans
->block_rsv
;
4521 if (root
== root
->fs_info
->csum_root
&& trans
->adding_csums
)
4522 block_rsv
= trans
->block_rsv
;
4524 if (root
== root
->fs_info
->uuid_root
)
4525 block_rsv
= trans
->block_rsv
;
4528 block_rsv
= root
->block_rsv
;
4531 block_rsv
= &root
->fs_info
->empty_block_rsv
;
4536 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
4540 spin_lock(&block_rsv
->lock
);
4541 if (block_rsv
->reserved
>= num_bytes
) {
4542 block_rsv
->reserved
-= num_bytes
;
4543 if (block_rsv
->reserved
< block_rsv
->size
)
4544 block_rsv
->full
= 0;
4547 spin_unlock(&block_rsv
->lock
);
4551 static void block_rsv_add_bytes(struct btrfs_block_rsv
*block_rsv
,
4552 u64 num_bytes
, int update_size
)
4554 spin_lock(&block_rsv
->lock
);
4555 block_rsv
->reserved
+= num_bytes
;
4557 block_rsv
->size
+= num_bytes
;
4558 else if (block_rsv
->reserved
>= block_rsv
->size
)
4559 block_rsv
->full
= 1;
4560 spin_unlock(&block_rsv
->lock
);
4563 int btrfs_cond_migrate_bytes(struct btrfs_fs_info
*fs_info
,
4564 struct btrfs_block_rsv
*dest
, u64 num_bytes
,
4567 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
4570 if (global_rsv
->space_info
!= dest
->space_info
)
4573 spin_lock(&global_rsv
->lock
);
4574 min_bytes
= div_factor(global_rsv
->size
, min_factor
);
4575 if (global_rsv
->reserved
< min_bytes
+ num_bytes
) {
4576 spin_unlock(&global_rsv
->lock
);
4579 global_rsv
->reserved
-= num_bytes
;
4580 if (global_rsv
->reserved
< global_rsv
->size
)
4581 global_rsv
->full
= 0;
4582 spin_unlock(&global_rsv
->lock
);
4584 block_rsv_add_bytes(dest
, num_bytes
, 1);
4588 static void block_rsv_release_bytes(struct btrfs_fs_info
*fs_info
,
4589 struct btrfs_block_rsv
*block_rsv
,
4590 struct btrfs_block_rsv
*dest
, u64 num_bytes
)
4592 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
4594 spin_lock(&block_rsv
->lock
);
4595 if (num_bytes
== (u64
)-1)
4596 num_bytes
= block_rsv
->size
;
4597 block_rsv
->size
-= num_bytes
;
4598 if (block_rsv
->reserved
>= block_rsv
->size
) {
4599 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
4600 block_rsv
->reserved
= block_rsv
->size
;
4601 block_rsv
->full
= 1;
4605 spin_unlock(&block_rsv
->lock
);
4607 if (num_bytes
> 0) {
4609 spin_lock(&dest
->lock
);
4613 bytes_to_add
= dest
->size
- dest
->reserved
;
4614 bytes_to_add
= min(num_bytes
, bytes_to_add
);
4615 dest
->reserved
+= bytes_to_add
;
4616 if (dest
->reserved
>= dest
->size
)
4618 num_bytes
-= bytes_to_add
;
4620 spin_unlock(&dest
->lock
);
4623 spin_lock(&space_info
->lock
);
4624 space_info
->bytes_may_use
-= num_bytes
;
4625 trace_btrfs_space_reservation(fs_info
, "space_info",
4626 space_info
->flags
, num_bytes
, 0);
4627 spin_unlock(&space_info
->lock
);
4632 static int block_rsv_migrate_bytes(struct btrfs_block_rsv
*src
,
4633 struct btrfs_block_rsv
*dst
, u64 num_bytes
)
4637 ret
= block_rsv_use_bytes(src
, num_bytes
);
4641 block_rsv_add_bytes(dst
, num_bytes
, 1);
4645 void btrfs_init_block_rsv(struct btrfs_block_rsv
*rsv
, unsigned short type
)
4647 memset(rsv
, 0, sizeof(*rsv
));
4648 spin_lock_init(&rsv
->lock
);
4652 struct btrfs_block_rsv
*btrfs_alloc_block_rsv(struct btrfs_root
*root
,
4653 unsigned short type
)
4655 struct btrfs_block_rsv
*block_rsv
;
4656 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4658 block_rsv
= kmalloc(sizeof(*block_rsv
), GFP_NOFS
);
4662 btrfs_init_block_rsv(block_rsv
, type
);
4663 block_rsv
->space_info
= __find_space_info(fs_info
,
4664 BTRFS_BLOCK_GROUP_METADATA
);
4668 void btrfs_free_block_rsv(struct btrfs_root
*root
,
4669 struct btrfs_block_rsv
*rsv
)
4673 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
4677 int btrfs_block_rsv_add(struct btrfs_root
*root
,
4678 struct btrfs_block_rsv
*block_rsv
, u64 num_bytes
,
4679 enum btrfs_reserve_flush_enum flush
)
4686 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
4688 block_rsv_add_bytes(block_rsv
, num_bytes
, 1);
4695 int btrfs_block_rsv_check(struct btrfs_root
*root
,
4696 struct btrfs_block_rsv
*block_rsv
, int min_factor
)
4704 spin_lock(&block_rsv
->lock
);
4705 num_bytes
= div_factor(block_rsv
->size
, min_factor
);
4706 if (block_rsv
->reserved
>= num_bytes
)
4708 spin_unlock(&block_rsv
->lock
);
4713 int btrfs_block_rsv_refill(struct btrfs_root
*root
,
4714 struct btrfs_block_rsv
*block_rsv
, u64 min_reserved
,
4715 enum btrfs_reserve_flush_enum flush
)
4723 spin_lock(&block_rsv
->lock
);
4724 num_bytes
= min_reserved
;
4725 if (block_rsv
->reserved
>= num_bytes
)
4728 num_bytes
-= block_rsv
->reserved
;
4729 spin_unlock(&block_rsv
->lock
);
4734 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
4736 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
4743 int btrfs_block_rsv_migrate(struct btrfs_block_rsv
*src_rsv
,
4744 struct btrfs_block_rsv
*dst_rsv
,
4747 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
4750 void btrfs_block_rsv_release(struct btrfs_root
*root
,
4751 struct btrfs_block_rsv
*block_rsv
,
4754 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
4755 if (global_rsv
== block_rsv
||
4756 block_rsv
->space_info
!= global_rsv
->space_info
)
4758 block_rsv_release_bytes(root
->fs_info
, block_rsv
, global_rsv
,
4763 * helper to calculate size of global block reservation.
4764 * the desired value is sum of space used by extent tree,
4765 * checksum tree and root tree
4767 static u64
calc_global_metadata_size(struct btrfs_fs_info
*fs_info
)
4769 struct btrfs_space_info
*sinfo
;
4773 int csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
4775 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_DATA
);
4776 spin_lock(&sinfo
->lock
);
4777 data_used
= sinfo
->bytes_used
;
4778 spin_unlock(&sinfo
->lock
);
4780 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4781 spin_lock(&sinfo
->lock
);
4782 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_DATA
)
4784 meta_used
= sinfo
->bytes_used
;
4785 spin_unlock(&sinfo
->lock
);
4787 num_bytes
= (data_used
>> fs_info
->sb
->s_blocksize_bits
) *
4789 num_bytes
+= div64_u64(data_used
+ meta_used
, 50);
4791 if (num_bytes
* 3 > meta_used
)
4792 num_bytes
= div64_u64(meta_used
, 3);
4794 return ALIGN(num_bytes
, fs_info
->extent_root
->nodesize
<< 10);
4797 static void update_global_block_rsv(struct btrfs_fs_info
*fs_info
)
4799 struct btrfs_block_rsv
*block_rsv
= &fs_info
->global_block_rsv
;
4800 struct btrfs_space_info
*sinfo
= block_rsv
->space_info
;
4803 num_bytes
= calc_global_metadata_size(fs_info
);
4805 spin_lock(&sinfo
->lock
);
4806 spin_lock(&block_rsv
->lock
);
4808 block_rsv
->size
= min_t(u64
, num_bytes
, 512 * 1024 * 1024);
4810 num_bytes
= sinfo
->bytes_used
+ sinfo
->bytes_pinned
+
4811 sinfo
->bytes_reserved
+ sinfo
->bytes_readonly
+
4812 sinfo
->bytes_may_use
;
4814 if (sinfo
->total_bytes
> num_bytes
) {
4815 num_bytes
= sinfo
->total_bytes
- num_bytes
;
4816 block_rsv
->reserved
+= num_bytes
;
4817 sinfo
->bytes_may_use
+= num_bytes
;
4818 trace_btrfs_space_reservation(fs_info
, "space_info",
4819 sinfo
->flags
, num_bytes
, 1);
4822 if (block_rsv
->reserved
>= block_rsv
->size
) {
4823 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
4824 sinfo
->bytes_may_use
-= num_bytes
;
4825 trace_btrfs_space_reservation(fs_info
, "space_info",
4826 sinfo
->flags
, num_bytes
, 0);
4827 block_rsv
->reserved
= block_rsv
->size
;
4828 block_rsv
->full
= 1;
4831 spin_unlock(&block_rsv
->lock
);
4832 spin_unlock(&sinfo
->lock
);
4835 static void init_global_block_rsv(struct btrfs_fs_info
*fs_info
)
4837 struct btrfs_space_info
*space_info
;
4839 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
4840 fs_info
->chunk_block_rsv
.space_info
= space_info
;
4842 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4843 fs_info
->global_block_rsv
.space_info
= space_info
;
4844 fs_info
->delalloc_block_rsv
.space_info
= space_info
;
4845 fs_info
->trans_block_rsv
.space_info
= space_info
;
4846 fs_info
->empty_block_rsv
.space_info
= space_info
;
4847 fs_info
->delayed_block_rsv
.space_info
= space_info
;
4849 fs_info
->extent_root
->block_rsv
= &fs_info
->global_block_rsv
;
4850 fs_info
->csum_root
->block_rsv
= &fs_info
->global_block_rsv
;
4851 fs_info
->dev_root
->block_rsv
= &fs_info
->global_block_rsv
;
4852 fs_info
->tree_root
->block_rsv
= &fs_info
->global_block_rsv
;
4853 if (fs_info
->quota_root
)
4854 fs_info
->quota_root
->block_rsv
= &fs_info
->global_block_rsv
;
4855 fs_info
->chunk_root
->block_rsv
= &fs_info
->chunk_block_rsv
;
4857 update_global_block_rsv(fs_info
);
4860 static void release_global_block_rsv(struct btrfs_fs_info
*fs_info
)
4862 block_rsv_release_bytes(fs_info
, &fs_info
->global_block_rsv
, NULL
,
4864 WARN_ON(fs_info
->delalloc_block_rsv
.size
> 0);
4865 WARN_ON(fs_info
->delalloc_block_rsv
.reserved
> 0);
4866 WARN_ON(fs_info
->trans_block_rsv
.size
> 0);
4867 WARN_ON(fs_info
->trans_block_rsv
.reserved
> 0);
4868 WARN_ON(fs_info
->chunk_block_rsv
.size
> 0);
4869 WARN_ON(fs_info
->chunk_block_rsv
.reserved
> 0);
4870 WARN_ON(fs_info
->delayed_block_rsv
.size
> 0);
4871 WARN_ON(fs_info
->delayed_block_rsv
.reserved
> 0);
4874 void btrfs_trans_release_metadata(struct btrfs_trans_handle
*trans
,
4875 struct btrfs_root
*root
)
4877 if (!trans
->block_rsv
)
4880 if (!trans
->bytes_reserved
)
4883 trace_btrfs_space_reservation(root
->fs_info
, "transaction",
4884 trans
->transid
, trans
->bytes_reserved
, 0);
4885 btrfs_block_rsv_release(root
, trans
->block_rsv
, trans
->bytes_reserved
);
4886 trans
->bytes_reserved
= 0;
4889 /* Can only return 0 or -ENOSPC */
4890 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle
*trans
,
4891 struct inode
*inode
)
4893 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4894 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
4895 struct btrfs_block_rsv
*dst_rsv
= root
->orphan_block_rsv
;
4898 * We need to hold space in order to delete our orphan item once we've
4899 * added it, so this takes the reservation so we can release it later
4900 * when we are truly done with the orphan item.
4902 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
4903 trace_btrfs_space_reservation(root
->fs_info
, "orphan",
4904 btrfs_ino(inode
), num_bytes
, 1);
4905 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
4908 void btrfs_orphan_release_metadata(struct inode
*inode
)
4910 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4911 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
4912 trace_btrfs_space_reservation(root
->fs_info
, "orphan",
4913 btrfs_ino(inode
), num_bytes
, 0);
4914 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
, num_bytes
);
4918 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4919 * root: the root of the parent directory
4920 * rsv: block reservation
4921 * items: the number of items that we need do reservation
4922 * qgroup_reserved: used to return the reserved size in qgroup
4924 * This function is used to reserve the space for snapshot/subvolume
4925 * creation and deletion. Those operations are different with the
4926 * common file/directory operations, they change two fs/file trees
4927 * and root tree, the number of items that the qgroup reserves is
4928 * different with the free space reservation. So we can not use
4929 * the space reseravtion mechanism in start_transaction().
4931 int btrfs_subvolume_reserve_metadata(struct btrfs_root
*root
,
4932 struct btrfs_block_rsv
*rsv
,
4934 u64
*qgroup_reserved
,
4935 bool use_global_rsv
)
4939 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
4941 if (root
->fs_info
->quota_enabled
) {
4942 /* One for parent inode, two for dir entries */
4943 num_bytes
= 3 * root
->nodesize
;
4944 ret
= btrfs_qgroup_reserve(root
, num_bytes
);
4951 *qgroup_reserved
= num_bytes
;
4953 num_bytes
= btrfs_calc_trans_metadata_size(root
, items
);
4954 rsv
->space_info
= __find_space_info(root
->fs_info
,
4955 BTRFS_BLOCK_GROUP_METADATA
);
4956 ret
= btrfs_block_rsv_add(root
, rsv
, num_bytes
,
4957 BTRFS_RESERVE_FLUSH_ALL
);
4959 if (ret
== -ENOSPC
&& use_global_rsv
)
4960 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
, num_bytes
);
4963 if (*qgroup_reserved
)
4964 btrfs_qgroup_free(root
, *qgroup_reserved
);
4970 void btrfs_subvolume_release_metadata(struct btrfs_root
*root
,
4971 struct btrfs_block_rsv
*rsv
,
4972 u64 qgroup_reserved
)
4974 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
4975 if (qgroup_reserved
)
4976 btrfs_qgroup_free(root
, qgroup_reserved
);
4980 * drop_outstanding_extent - drop an outstanding extent
4981 * @inode: the inode we're dropping the extent for
4982 * @num_bytes: the number of bytes we're relaseing.
4984 * This is called when we are freeing up an outstanding extent, either called
4985 * after an error or after an extent is written. This will return the number of
4986 * reserved extents that need to be freed. This must be called with
4987 * BTRFS_I(inode)->lock held.
4989 static unsigned drop_outstanding_extent(struct inode
*inode
, u64 num_bytes
)
4991 unsigned drop_inode_space
= 0;
4992 unsigned dropped_extents
= 0;
4993 unsigned num_extents
= 0;
4995 num_extents
= (unsigned)div64_u64(num_bytes
+
4996 BTRFS_MAX_EXTENT_SIZE
- 1,
4997 BTRFS_MAX_EXTENT_SIZE
);
4998 ASSERT(num_extents
);
4999 ASSERT(BTRFS_I(inode
)->outstanding_extents
>= num_extents
);
5000 BTRFS_I(inode
)->outstanding_extents
-= num_extents
;
5002 if (BTRFS_I(inode
)->outstanding_extents
== 0 &&
5003 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5004 &BTRFS_I(inode
)->runtime_flags
))
5005 drop_inode_space
= 1;
5008 * If we have more or the same amount of outsanding extents than we have
5009 * reserved then we need to leave the reserved extents count alone.
5011 if (BTRFS_I(inode
)->outstanding_extents
>=
5012 BTRFS_I(inode
)->reserved_extents
)
5013 return drop_inode_space
;
5015 dropped_extents
= BTRFS_I(inode
)->reserved_extents
-
5016 BTRFS_I(inode
)->outstanding_extents
;
5017 BTRFS_I(inode
)->reserved_extents
-= dropped_extents
;
5018 return dropped_extents
+ drop_inode_space
;
5022 * calc_csum_metadata_size - return the amount of metada space that must be
5023 * reserved/free'd for the given bytes.
5024 * @inode: the inode we're manipulating
5025 * @num_bytes: the number of bytes in question
5026 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5028 * This adjusts the number of csum_bytes in the inode and then returns the
5029 * correct amount of metadata that must either be reserved or freed. We
5030 * calculate how many checksums we can fit into one leaf and then divide the
5031 * number of bytes that will need to be checksumed by this value to figure out
5032 * how many checksums will be required. If we are adding bytes then the number
5033 * may go up and we will return the number of additional bytes that must be
5034 * reserved. If it is going down we will return the number of bytes that must
5037 * This must be called with BTRFS_I(inode)->lock held.
5039 static u64
calc_csum_metadata_size(struct inode
*inode
, u64 num_bytes
,
5042 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5044 int num_csums_per_leaf
;
5048 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
&&
5049 BTRFS_I(inode
)->csum_bytes
== 0)
5052 old_csums
= (int)div64_u64(BTRFS_I(inode
)->csum_bytes
, root
->sectorsize
);
5054 BTRFS_I(inode
)->csum_bytes
+= num_bytes
;
5056 BTRFS_I(inode
)->csum_bytes
-= num_bytes
;
5057 csum_size
= BTRFS_LEAF_DATA_SIZE(root
) - sizeof(struct btrfs_item
);
5058 num_csums_per_leaf
= (int)div64_u64(csum_size
,
5059 sizeof(struct btrfs_csum_item
) +
5060 sizeof(struct btrfs_disk_key
));
5061 num_csums
= (int)div64_u64(BTRFS_I(inode
)->csum_bytes
, root
->sectorsize
);
5062 num_csums
= num_csums
+ num_csums_per_leaf
- 1;
5063 num_csums
= num_csums
/ num_csums_per_leaf
;
5065 old_csums
= old_csums
+ num_csums_per_leaf
- 1;
5066 old_csums
= old_csums
/ num_csums_per_leaf
;
5068 /* No change, no need to reserve more */
5069 if (old_csums
== num_csums
)
5073 return btrfs_calc_trans_metadata_size(root
,
5074 num_csums
- old_csums
);
5076 return btrfs_calc_trans_metadata_size(root
, old_csums
- num_csums
);
5079 int btrfs_delalloc_reserve_metadata(struct inode
*inode
, u64 num_bytes
)
5081 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5082 struct btrfs_block_rsv
*block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
5085 unsigned nr_extents
= 0;
5086 int extra_reserve
= 0;
5087 enum btrfs_reserve_flush_enum flush
= BTRFS_RESERVE_FLUSH_ALL
;
5089 bool delalloc_lock
= true;
5093 /* If we are a free space inode we need to not flush since we will be in
5094 * the middle of a transaction commit. We also don't need the delalloc
5095 * mutex since we won't race with anybody. We need this mostly to make
5096 * lockdep shut its filthy mouth.
5098 if (btrfs_is_free_space_inode(inode
)) {
5099 flush
= BTRFS_RESERVE_NO_FLUSH
;
5100 delalloc_lock
= false;
5103 if (flush
!= BTRFS_RESERVE_NO_FLUSH
&&
5104 btrfs_transaction_in_commit(root
->fs_info
))
5105 schedule_timeout(1);
5108 mutex_lock(&BTRFS_I(inode
)->delalloc_mutex
);
5110 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
5112 spin_lock(&BTRFS_I(inode
)->lock
);
5113 nr_extents
= (unsigned)div64_u64(num_bytes
+
5114 BTRFS_MAX_EXTENT_SIZE
- 1,
5115 BTRFS_MAX_EXTENT_SIZE
);
5116 BTRFS_I(inode
)->outstanding_extents
+= nr_extents
;
5119 if (BTRFS_I(inode
)->outstanding_extents
>
5120 BTRFS_I(inode
)->reserved_extents
)
5121 nr_extents
= BTRFS_I(inode
)->outstanding_extents
-
5122 BTRFS_I(inode
)->reserved_extents
;
5125 * Add an item to reserve for updating the inode when we complete the
5128 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5129 &BTRFS_I(inode
)->runtime_flags
)) {
5134 to_reserve
= btrfs_calc_trans_metadata_size(root
, nr_extents
);
5135 to_reserve
+= calc_csum_metadata_size(inode
, num_bytes
, 1);
5136 csum_bytes
= BTRFS_I(inode
)->csum_bytes
;
5137 spin_unlock(&BTRFS_I(inode
)->lock
);
5139 if (root
->fs_info
->quota_enabled
) {
5140 ret
= btrfs_qgroup_reserve(root
, num_bytes
+
5141 nr_extents
* root
->nodesize
);
5146 ret
= reserve_metadata_bytes(root
, block_rsv
, to_reserve
, flush
);
5147 if (unlikely(ret
)) {
5148 if (root
->fs_info
->quota_enabled
)
5149 btrfs_qgroup_free(root
, num_bytes
+
5150 nr_extents
* root
->nodesize
);
5154 spin_lock(&BTRFS_I(inode
)->lock
);
5155 if (extra_reserve
) {
5156 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5157 &BTRFS_I(inode
)->runtime_flags
);
5160 BTRFS_I(inode
)->reserved_extents
+= nr_extents
;
5161 spin_unlock(&BTRFS_I(inode
)->lock
);
5164 mutex_unlock(&BTRFS_I(inode
)->delalloc_mutex
);
5167 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5168 btrfs_ino(inode
), to_reserve
, 1);
5169 block_rsv_add_bytes(block_rsv
, to_reserve
, 1);
5174 spin_lock(&BTRFS_I(inode
)->lock
);
5175 dropped
= drop_outstanding_extent(inode
, num_bytes
);
5177 * If the inodes csum_bytes is the same as the original
5178 * csum_bytes then we know we haven't raced with any free()ers
5179 * so we can just reduce our inodes csum bytes and carry on.
5181 if (BTRFS_I(inode
)->csum_bytes
== csum_bytes
) {
5182 calc_csum_metadata_size(inode
, num_bytes
, 0);
5184 u64 orig_csum_bytes
= BTRFS_I(inode
)->csum_bytes
;
5188 * This is tricky, but first we need to figure out how much we
5189 * free'd from any free-ers that occured during this
5190 * reservation, so we reset ->csum_bytes to the csum_bytes
5191 * before we dropped our lock, and then call the free for the
5192 * number of bytes that were freed while we were trying our
5195 bytes
= csum_bytes
- BTRFS_I(inode
)->csum_bytes
;
5196 BTRFS_I(inode
)->csum_bytes
= csum_bytes
;
5197 to_free
= calc_csum_metadata_size(inode
, bytes
, 0);
5201 * Now we need to see how much we would have freed had we not
5202 * been making this reservation and our ->csum_bytes were not
5203 * artificially inflated.
5205 BTRFS_I(inode
)->csum_bytes
= csum_bytes
- num_bytes
;
5206 bytes
= csum_bytes
- orig_csum_bytes
;
5207 bytes
= calc_csum_metadata_size(inode
, bytes
, 0);
5210 * Now reset ->csum_bytes to what it should be. If bytes is
5211 * more than to_free then we would have free'd more space had we
5212 * not had an artificially high ->csum_bytes, so we need to free
5213 * the remainder. If bytes is the same or less then we don't
5214 * need to do anything, the other free-ers did the correct
5217 BTRFS_I(inode
)->csum_bytes
= orig_csum_bytes
- num_bytes
;
5218 if (bytes
> to_free
)
5219 to_free
= bytes
- to_free
;
5223 spin_unlock(&BTRFS_I(inode
)->lock
);
5225 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
5228 btrfs_block_rsv_release(root
, block_rsv
, to_free
);
5229 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5230 btrfs_ino(inode
), to_free
, 0);
5233 mutex_unlock(&BTRFS_I(inode
)->delalloc_mutex
);
5238 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5239 * @inode: the inode to release the reservation for
5240 * @num_bytes: the number of bytes we're releasing
5242 * This will release the metadata reservation for an inode. This can be called
5243 * once we complete IO for a given set of bytes to release their metadata
5246 void btrfs_delalloc_release_metadata(struct inode
*inode
, u64 num_bytes
)
5248 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5252 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
5253 spin_lock(&BTRFS_I(inode
)->lock
);
5254 dropped
= drop_outstanding_extent(inode
, num_bytes
);
5257 to_free
= calc_csum_metadata_size(inode
, num_bytes
, 0);
5258 spin_unlock(&BTRFS_I(inode
)->lock
);
5260 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
5262 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5263 btrfs_ino(inode
), to_free
, 0);
5264 if (root
->fs_info
->quota_enabled
) {
5265 btrfs_qgroup_free(root
, num_bytes
+
5266 dropped
* root
->nodesize
);
5269 btrfs_block_rsv_release(root
, &root
->fs_info
->delalloc_block_rsv
,
5274 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5275 * @inode: inode we're writing to
5276 * @num_bytes: the number of bytes we want to allocate
5278 * This will do the following things
5280 * o reserve space in the data space info for num_bytes
5281 * o reserve space in the metadata space info based on number of outstanding
5282 * extents and how much csums will be needed
5283 * o add to the inodes ->delalloc_bytes
5284 * o add it to the fs_info's delalloc inodes list.
5286 * This will return 0 for success and -ENOSPC if there is no space left.
5288 int btrfs_delalloc_reserve_space(struct inode
*inode
, u64 num_bytes
)
5292 ret
= btrfs_check_data_free_space(inode
, num_bytes
);
5296 ret
= btrfs_delalloc_reserve_metadata(inode
, num_bytes
);
5298 btrfs_free_reserved_data_space(inode
, num_bytes
);
5306 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5307 * @inode: inode we're releasing space for
5308 * @num_bytes: the number of bytes we want to free up
5310 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5311 * called in the case that we don't need the metadata AND data reservations
5312 * anymore. So if there is an error or we insert an inline extent.
5314 * This function will release the metadata space that was not used and will
5315 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5316 * list if there are no delalloc bytes left.
5318 void btrfs_delalloc_release_space(struct inode
*inode
, u64 num_bytes
)
5320 btrfs_delalloc_release_metadata(inode
, num_bytes
);
5321 btrfs_free_reserved_data_space(inode
, num_bytes
);
5324 static int update_block_group(struct btrfs_trans_handle
*trans
,
5325 struct btrfs_root
*root
, u64 bytenr
,
5326 u64 num_bytes
, int alloc
)
5328 struct btrfs_block_group_cache
*cache
= NULL
;
5329 struct btrfs_fs_info
*info
= root
->fs_info
;
5330 u64 total
= num_bytes
;
5335 /* block accounting for super block */
5336 spin_lock(&info
->delalloc_root_lock
);
5337 old_val
= btrfs_super_bytes_used(info
->super_copy
);
5339 old_val
+= num_bytes
;
5341 old_val
-= num_bytes
;
5342 btrfs_set_super_bytes_used(info
->super_copy
, old_val
);
5343 spin_unlock(&info
->delalloc_root_lock
);
5346 cache
= btrfs_lookup_block_group(info
, bytenr
);
5349 if (cache
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
5350 BTRFS_BLOCK_GROUP_RAID1
|
5351 BTRFS_BLOCK_GROUP_RAID10
))
5356 * If this block group has free space cache written out, we
5357 * need to make sure to load it if we are removing space. This
5358 * is because we need the unpinning stage to actually add the
5359 * space back to the block group, otherwise we will leak space.
5361 if (!alloc
&& cache
->cached
== BTRFS_CACHE_NO
)
5362 cache_block_group(cache
, 1);
5364 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
5365 if (list_empty(&cache
->dirty_list
)) {
5366 list_add_tail(&cache
->dirty_list
,
5367 &trans
->transaction
->dirty_bgs
);
5368 btrfs_get_block_group(cache
);
5370 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
5372 byte_in_group
= bytenr
- cache
->key
.objectid
;
5373 WARN_ON(byte_in_group
> cache
->key
.offset
);
5375 spin_lock(&cache
->space_info
->lock
);
5376 spin_lock(&cache
->lock
);
5378 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
5379 cache
->disk_cache_state
< BTRFS_DC_CLEAR
)
5380 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
5382 old_val
= btrfs_block_group_used(&cache
->item
);
5383 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
5385 old_val
+= num_bytes
;
5386 btrfs_set_block_group_used(&cache
->item
, old_val
);
5387 cache
->reserved
-= num_bytes
;
5388 cache
->space_info
->bytes_reserved
-= num_bytes
;
5389 cache
->space_info
->bytes_used
+= num_bytes
;
5390 cache
->space_info
->disk_used
+= num_bytes
* factor
;
5391 spin_unlock(&cache
->lock
);
5392 spin_unlock(&cache
->space_info
->lock
);
5394 old_val
-= num_bytes
;
5395 btrfs_set_block_group_used(&cache
->item
, old_val
);
5396 cache
->pinned
+= num_bytes
;
5397 cache
->space_info
->bytes_pinned
+= num_bytes
;
5398 cache
->space_info
->bytes_used
-= num_bytes
;
5399 cache
->space_info
->disk_used
-= num_bytes
* factor
;
5400 spin_unlock(&cache
->lock
);
5401 spin_unlock(&cache
->space_info
->lock
);
5403 set_extent_dirty(info
->pinned_extents
,
5404 bytenr
, bytenr
+ num_bytes
- 1,
5405 GFP_NOFS
| __GFP_NOFAIL
);
5407 * No longer have used bytes in this block group, queue
5411 spin_lock(&info
->unused_bgs_lock
);
5412 if (list_empty(&cache
->bg_list
)) {
5413 btrfs_get_block_group(cache
);
5414 list_add_tail(&cache
->bg_list
,
5417 spin_unlock(&info
->unused_bgs_lock
);
5420 btrfs_put_block_group(cache
);
5422 bytenr
+= num_bytes
;
5427 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
5429 struct btrfs_block_group_cache
*cache
;
5432 spin_lock(&root
->fs_info
->block_group_cache_lock
);
5433 bytenr
= root
->fs_info
->first_logical_byte
;
5434 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
5436 if (bytenr
< (u64
)-1)
5439 cache
= btrfs_lookup_first_block_group(root
->fs_info
, search_start
);
5443 bytenr
= cache
->key
.objectid
;
5444 btrfs_put_block_group(cache
);
5449 static int pin_down_extent(struct btrfs_root
*root
,
5450 struct btrfs_block_group_cache
*cache
,
5451 u64 bytenr
, u64 num_bytes
, int reserved
)
5453 spin_lock(&cache
->space_info
->lock
);
5454 spin_lock(&cache
->lock
);
5455 cache
->pinned
+= num_bytes
;
5456 cache
->space_info
->bytes_pinned
+= num_bytes
;
5458 cache
->reserved
-= num_bytes
;
5459 cache
->space_info
->bytes_reserved
-= num_bytes
;
5461 spin_unlock(&cache
->lock
);
5462 spin_unlock(&cache
->space_info
->lock
);
5464 set_extent_dirty(root
->fs_info
->pinned_extents
, bytenr
,
5465 bytenr
+ num_bytes
- 1, GFP_NOFS
| __GFP_NOFAIL
);
5467 trace_btrfs_reserved_extent_free(root
, bytenr
, num_bytes
);
5472 * this function must be called within transaction
5474 int btrfs_pin_extent(struct btrfs_root
*root
,
5475 u64 bytenr
, u64 num_bytes
, int reserved
)
5477 struct btrfs_block_group_cache
*cache
;
5479 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
5480 BUG_ON(!cache
); /* Logic error */
5482 pin_down_extent(root
, cache
, bytenr
, num_bytes
, reserved
);
5484 btrfs_put_block_group(cache
);
5489 * this function must be called within transaction
5491 int btrfs_pin_extent_for_log_replay(struct btrfs_root
*root
,
5492 u64 bytenr
, u64 num_bytes
)
5494 struct btrfs_block_group_cache
*cache
;
5497 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
5502 * pull in the free space cache (if any) so that our pin
5503 * removes the free space from the cache. We have load_only set
5504 * to one because the slow code to read in the free extents does check
5505 * the pinned extents.
5507 cache_block_group(cache
, 1);
5509 pin_down_extent(root
, cache
, bytenr
, num_bytes
, 0);
5511 /* remove us from the free space cache (if we're there at all) */
5512 ret
= btrfs_remove_free_space(cache
, bytenr
, num_bytes
);
5513 btrfs_put_block_group(cache
);
5517 static int __exclude_logged_extent(struct btrfs_root
*root
, u64 start
, u64 num_bytes
)
5520 struct btrfs_block_group_cache
*block_group
;
5521 struct btrfs_caching_control
*caching_ctl
;
5523 block_group
= btrfs_lookup_block_group(root
->fs_info
, start
);
5527 cache_block_group(block_group
, 0);
5528 caching_ctl
= get_caching_control(block_group
);
5532 BUG_ON(!block_group_cache_done(block_group
));
5533 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
5535 mutex_lock(&caching_ctl
->mutex
);
5537 if (start
>= caching_ctl
->progress
) {
5538 ret
= add_excluded_extent(root
, start
, num_bytes
);
5539 } else if (start
+ num_bytes
<= caching_ctl
->progress
) {
5540 ret
= btrfs_remove_free_space(block_group
,
5543 num_bytes
= caching_ctl
->progress
- start
;
5544 ret
= btrfs_remove_free_space(block_group
,
5549 num_bytes
= (start
+ num_bytes
) -
5550 caching_ctl
->progress
;
5551 start
= caching_ctl
->progress
;
5552 ret
= add_excluded_extent(root
, start
, num_bytes
);
5555 mutex_unlock(&caching_ctl
->mutex
);
5556 put_caching_control(caching_ctl
);
5558 btrfs_put_block_group(block_group
);
5562 int btrfs_exclude_logged_extents(struct btrfs_root
*log
,
5563 struct extent_buffer
*eb
)
5565 struct btrfs_file_extent_item
*item
;
5566 struct btrfs_key key
;
5570 if (!btrfs_fs_incompat(log
->fs_info
, MIXED_GROUPS
))
5573 for (i
= 0; i
< btrfs_header_nritems(eb
); i
++) {
5574 btrfs_item_key_to_cpu(eb
, &key
, i
);
5575 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
5577 item
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
5578 found_type
= btrfs_file_extent_type(eb
, item
);
5579 if (found_type
== BTRFS_FILE_EXTENT_INLINE
)
5581 if (btrfs_file_extent_disk_bytenr(eb
, item
) == 0)
5583 key
.objectid
= btrfs_file_extent_disk_bytenr(eb
, item
);
5584 key
.offset
= btrfs_file_extent_disk_num_bytes(eb
, item
);
5585 __exclude_logged_extent(log
, key
.objectid
, key
.offset
);
5592 * btrfs_update_reserved_bytes - update the block_group and space info counters
5593 * @cache: The cache we are manipulating
5594 * @num_bytes: The number of bytes in question
5595 * @reserve: One of the reservation enums
5596 * @delalloc: The blocks are allocated for the delalloc write
5598 * This is called by the allocator when it reserves space, or by somebody who is
5599 * freeing space that was never actually used on disk. For example if you
5600 * reserve some space for a new leaf in transaction A and before transaction A
5601 * commits you free that leaf, you call this with reserve set to 0 in order to
5602 * clear the reservation.
5604 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5605 * ENOSPC accounting. For data we handle the reservation through clearing the
5606 * delalloc bits in the io_tree. We have to do this since we could end up
5607 * allocating less disk space for the amount of data we have reserved in the
5608 * case of compression.
5610 * If this is a reservation and the block group has become read only we cannot
5611 * make the reservation and return -EAGAIN, otherwise this function always
5614 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
5615 u64 num_bytes
, int reserve
, int delalloc
)
5617 struct btrfs_space_info
*space_info
= cache
->space_info
;
5620 spin_lock(&space_info
->lock
);
5621 spin_lock(&cache
->lock
);
5622 if (reserve
!= RESERVE_FREE
) {
5626 cache
->reserved
+= num_bytes
;
5627 space_info
->bytes_reserved
+= num_bytes
;
5628 if (reserve
== RESERVE_ALLOC
) {
5629 trace_btrfs_space_reservation(cache
->fs_info
,
5630 "space_info", space_info
->flags
,
5632 space_info
->bytes_may_use
-= num_bytes
;
5636 cache
->delalloc_bytes
+= num_bytes
;
5640 space_info
->bytes_readonly
+= num_bytes
;
5641 cache
->reserved
-= num_bytes
;
5642 space_info
->bytes_reserved
-= num_bytes
;
5645 cache
->delalloc_bytes
-= num_bytes
;
5647 spin_unlock(&cache
->lock
);
5648 spin_unlock(&space_info
->lock
);
5652 void btrfs_prepare_extent_commit(struct btrfs_trans_handle
*trans
,
5653 struct btrfs_root
*root
)
5655 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5656 struct btrfs_caching_control
*next
;
5657 struct btrfs_caching_control
*caching_ctl
;
5658 struct btrfs_block_group_cache
*cache
;
5660 down_write(&fs_info
->commit_root_sem
);
5662 list_for_each_entry_safe(caching_ctl
, next
,
5663 &fs_info
->caching_block_groups
, list
) {
5664 cache
= caching_ctl
->block_group
;
5665 if (block_group_cache_done(cache
)) {
5666 cache
->last_byte_to_unpin
= (u64
)-1;
5667 list_del_init(&caching_ctl
->list
);
5668 put_caching_control(caching_ctl
);
5670 cache
->last_byte_to_unpin
= caching_ctl
->progress
;
5674 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
5675 fs_info
->pinned_extents
= &fs_info
->freed_extents
[1];
5677 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
5679 up_write(&fs_info
->commit_root_sem
);
5681 update_global_block_rsv(fs_info
);
5684 static int unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
,
5685 const bool return_free_space
)
5687 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5688 struct btrfs_block_group_cache
*cache
= NULL
;
5689 struct btrfs_space_info
*space_info
;
5690 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5694 while (start
<= end
) {
5697 start
>= cache
->key
.objectid
+ cache
->key
.offset
) {
5699 btrfs_put_block_group(cache
);
5700 cache
= btrfs_lookup_block_group(fs_info
, start
);
5701 BUG_ON(!cache
); /* Logic error */
5704 len
= cache
->key
.objectid
+ cache
->key
.offset
- start
;
5705 len
= min(len
, end
+ 1 - start
);
5707 if (start
< cache
->last_byte_to_unpin
) {
5708 len
= min(len
, cache
->last_byte_to_unpin
- start
);
5709 if (return_free_space
)
5710 btrfs_add_free_space(cache
, start
, len
);
5714 space_info
= cache
->space_info
;
5716 spin_lock(&space_info
->lock
);
5717 spin_lock(&cache
->lock
);
5718 cache
->pinned
-= len
;
5719 space_info
->bytes_pinned
-= len
;
5720 percpu_counter_add(&space_info
->total_bytes_pinned
, -len
);
5722 space_info
->bytes_readonly
+= len
;
5725 spin_unlock(&cache
->lock
);
5726 if (!readonly
&& global_rsv
->space_info
== space_info
) {
5727 spin_lock(&global_rsv
->lock
);
5728 if (!global_rsv
->full
) {
5729 len
= min(len
, global_rsv
->size
-
5730 global_rsv
->reserved
);
5731 global_rsv
->reserved
+= len
;
5732 space_info
->bytes_may_use
+= len
;
5733 if (global_rsv
->reserved
>= global_rsv
->size
)
5734 global_rsv
->full
= 1;
5736 spin_unlock(&global_rsv
->lock
);
5738 spin_unlock(&space_info
->lock
);
5742 btrfs_put_block_group(cache
);
5746 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
5747 struct btrfs_root
*root
)
5749 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5750 struct extent_io_tree
*unpin
;
5758 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
5759 unpin
= &fs_info
->freed_extents
[1];
5761 unpin
= &fs_info
->freed_extents
[0];
5764 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
5765 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
5766 EXTENT_DIRTY
, NULL
);
5768 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
5772 if (btrfs_test_opt(root
, DISCARD
))
5773 ret
= btrfs_discard_extent(root
, start
,
5774 end
+ 1 - start
, NULL
);
5776 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
5777 unpin_extent_range(root
, start
, end
, true);
5778 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
5785 static void add_pinned_bytes(struct btrfs_fs_info
*fs_info
, u64 num_bytes
,
5786 u64 owner
, u64 root_objectid
)
5788 struct btrfs_space_info
*space_info
;
5791 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
5792 if (root_objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
5793 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
5795 flags
= BTRFS_BLOCK_GROUP_METADATA
;
5797 flags
= BTRFS_BLOCK_GROUP_DATA
;
5800 space_info
= __find_space_info(fs_info
, flags
);
5801 BUG_ON(!space_info
); /* Logic bug */
5802 percpu_counter_add(&space_info
->total_bytes_pinned
, num_bytes
);
5806 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
5807 struct btrfs_root
*root
,
5808 u64 bytenr
, u64 num_bytes
, u64 parent
,
5809 u64 root_objectid
, u64 owner_objectid
,
5810 u64 owner_offset
, int refs_to_drop
,
5811 struct btrfs_delayed_extent_op
*extent_op
,
5814 struct btrfs_key key
;
5815 struct btrfs_path
*path
;
5816 struct btrfs_fs_info
*info
= root
->fs_info
;
5817 struct btrfs_root
*extent_root
= info
->extent_root
;
5818 struct extent_buffer
*leaf
;
5819 struct btrfs_extent_item
*ei
;
5820 struct btrfs_extent_inline_ref
*iref
;
5823 int extent_slot
= 0;
5824 int found_extent
= 0;
5829 enum btrfs_qgroup_operation_type type
= BTRFS_QGROUP_OPER_SUB_EXCL
;
5830 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
5833 if (!info
->quota_enabled
|| !is_fstree(root_objectid
))
5836 path
= btrfs_alloc_path();
5841 path
->leave_spinning
= 1;
5843 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
5844 BUG_ON(!is_data
&& refs_to_drop
!= 1);
5847 skinny_metadata
= 0;
5849 ret
= lookup_extent_backref(trans
, extent_root
, path
, &iref
,
5850 bytenr
, num_bytes
, parent
,
5851 root_objectid
, owner_objectid
,
5854 extent_slot
= path
->slots
[0];
5855 while (extent_slot
>= 0) {
5856 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
5858 if (key
.objectid
!= bytenr
)
5860 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
5861 key
.offset
== num_bytes
) {
5865 if (key
.type
== BTRFS_METADATA_ITEM_KEY
&&
5866 key
.offset
== owner_objectid
) {
5870 if (path
->slots
[0] - extent_slot
> 5)
5874 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5875 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
5876 if (found_extent
&& item_size
< sizeof(*ei
))
5879 if (!found_extent
) {
5881 ret
= remove_extent_backref(trans
, extent_root
, path
,
5883 is_data
, &last_ref
);
5885 btrfs_abort_transaction(trans
, extent_root
, ret
);
5888 btrfs_release_path(path
);
5889 path
->leave_spinning
= 1;
5891 key
.objectid
= bytenr
;
5892 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
5893 key
.offset
= num_bytes
;
5895 if (!is_data
&& skinny_metadata
) {
5896 key
.type
= BTRFS_METADATA_ITEM_KEY
;
5897 key
.offset
= owner_objectid
;
5900 ret
= btrfs_search_slot(trans
, extent_root
,
5902 if (ret
> 0 && skinny_metadata
&& path
->slots
[0]) {
5904 * Couldn't find our skinny metadata item,
5905 * see if we have ye olde extent item.
5908 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
5910 if (key
.objectid
== bytenr
&&
5911 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
5912 key
.offset
== num_bytes
)
5916 if (ret
> 0 && skinny_metadata
) {
5917 skinny_metadata
= false;
5918 key
.objectid
= bytenr
;
5919 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
5920 key
.offset
= num_bytes
;
5921 btrfs_release_path(path
);
5922 ret
= btrfs_search_slot(trans
, extent_root
,
5927 btrfs_err(info
, "umm, got %d back from search, was looking for %llu",
5930 btrfs_print_leaf(extent_root
,
5934 btrfs_abort_transaction(trans
, extent_root
, ret
);
5937 extent_slot
= path
->slots
[0];
5939 } else if (WARN_ON(ret
== -ENOENT
)) {
5940 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
5942 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
5943 bytenr
, parent
, root_objectid
, owner_objectid
,
5945 btrfs_abort_transaction(trans
, extent_root
, ret
);
5948 btrfs_abort_transaction(trans
, extent_root
, ret
);
5952 leaf
= path
->nodes
[0];
5953 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
5954 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5955 if (item_size
< sizeof(*ei
)) {
5956 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
5957 ret
= convert_extent_item_v0(trans
, extent_root
, path
,
5960 btrfs_abort_transaction(trans
, extent_root
, ret
);
5964 btrfs_release_path(path
);
5965 path
->leave_spinning
= 1;
5967 key
.objectid
= bytenr
;
5968 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
5969 key
.offset
= num_bytes
;
5971 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
5974 btrfs_err(info
, "umm, got %d back from search, was looking for %llu",
5976 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
5979 btrfs_abort_transaction(trans
, extent_root
, ret
);
5983 extent_slot
= path
->slots
[0];
5984 leaf
= path
->nodes
[0];
5985 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
5988 BUG_ON(item_size
< sizeof(*ei
));
5989 ei
= btrfs_item_ptr(leaf
, extent_slot
,
5990 struct btrfs_extent_item
);
5991 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
5992 key
.type
== BTRFS_EXTENT_ITEM_KEY
) {
5993 struct btrfs_tree_block_info
*bi
;
5994 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
5995 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
5996 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
5999 refs
= btrfs_extent_refs(leaf
, ei
);
6000 if (refs
< refs_to_drop
) {
6001 btrfs_err(info
, "trying to drop %d refs but we only have %Lu "
6002 "for bytenr %Lu", refs_to_drop
, refs
, bytenr
);
6004 btrfs_abort_transaction(trans
, extent_root
, ret
);
6007 refs
-= refs_to_drop
;
6010 type
= BTRFS_QGROUP_OPER_SUB_SHARED
;
6012 __run_delayed_extent_op(extent_op
, leaf
, ei
);
6014 * In the case of inline back ref, reference count will
6015 * be updated by remove_extent_backref
6018 BUG_ON(!found_extent
);
6020 btrfs_set_extent_refs(leaf
, ei
, refs
);
6021 btrfs_mark_buffer_dirty(leaf
);
6024 ret
= remove_extent_backref(trans
, extent_root
, path
,
6026 is_data
, &last_ref
);
6028 btrfs_abort_transaction(trans
, extent_root
, ret
);
6032 add_pinned_bytes(root
->fs_info
, -num_bytes
, owner_objectid
,
6036 BUG_ON(is_data
&& refs_to_drop
!=
6037 extent_data_ref_count(root
, path
, iref
));
6039 BUG_ON(path
->slots
[0] != extent_slot
);
6041 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
6042 path
->slots
[0] = extent_slot
;
6048 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
6051 btrfs_abort_transaction(trans
, extent_root
, ret
);
6054 btrfs_release_path(path
);
6057 ret
= btrfs_del_csums(trans
, root
, bytenr
, num_bytes
);
6059 btrfs_abort_transaction(trans
, extent_root
, ret
);
6064 ret
= update_block_group(trans
, root
, bytenr
, num_bytes
, 0);
6066 btrfs_abort_transaction(trans
, extent_root
, ret
);
6070 btrfs_release_path(path
);
6072 /* Deal with the quota accounting */
6073 if (!ret
&& last_ref
&& !no_quota
) {
6076 if (owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
&&
6077 type
== BTRFS_QGROUP_OPER_SUB_SHARED
)
6080 ret
= btrfs_qgroup_record_ref(trans
, info
, root_objectid
,
6081 bytenr
, num_bytes
, type
,
6085 btrfs_free_path(path
);
6090 * when we free an block, it is possible (and likely) that we free the last
6091 * delayed ref for that extent as well. This searches the delayed ref tree for
6092 * a given extent, and if there are no other delayed refs to be processed, it
6093 * removes it from the tree.
6095 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
6096 struct btrfs_root
*root
, u64 bytenr
)
6098 struct btrfs_delayed_ref_head
*head
;
6099 struct btrfs_delayed_ref_root
*delayed_refs
;
6102 delayed_refs
= &trans
->transaction
->delayed_refs
;
6103 spin_lock(&delayed_refs
->lock
);
6104 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
6106 goto out_delayed_unlock
;
6108 spin_lock(&head
->lock
);
6109 if (rb_first(&head
->ref_root
))
6112 if (head
->extent_op
) {
6113 if (!head
->must_insert_reserved
)
6115 btrfs_free_delayed_extent_op(head
->extent_op
);
6116 head
->extent_op
= NULL
;
6120 * waiting for the lock here would deadlock. If someone else has it
6121 * locked they are already in the process of dropping it anyway
6123 if (!mutex_trylock(&head
->mutex
))
6127 * at this point we have a head with no other entries. Go
6128 * ahead and process it.
6130 head
->node
.in_tree
= 0;
6131 rb_erase(&head
->href_node
, &delayed_refs
->href_root
);
6133 atomic_dec(&delayed_refs
->num_entries
);
6136 * we don't take a ref on the node because we're removing it from the
6137 * tree, so we just steal the ref the tree was holding.
6139 delayed_refs
->num_heads
--;
6140 if (head
->processing
== 0)
6141 delayed_refs
->num_heads_ready
--;
6142 head
->processing
= 0;
6143 spin_unlock(&head
->lock
);
6144 spin_unlock(&delayed_refs
->lock
);
6146 BUG_ON(head
->extent_op
);
6147 if (head
->must_insert_reserved
)
6150 mutex_unlock(&head
->mutex
);
6151 btrfs_put_delayed_ref(&head
->node
);
6154 spin_unlock(&head
->lock
);
6157 spin_unlock(&delayed_refs
->lock
);
6161 void btrfs_free_tree_block(struct btrfs_trans_handle
*trans
,
6162 struct btrfs_root
*root
,
6163 struct extent_buffer
*buf
,
6164 u64 parent
, int last_ref
)
6169 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
6170 ret
= btrfs_add_delayed_tree_ref(root
->fs_info
, trans
,
6171 buf
->start
, buf
->len
,
6172 parent
, root
->root_key
.objectid
,
6173 btrfs_header_level(buf
),
6174 BTRFS_DROP_DELAYED_REF
, NULL
, 0);
6175 BUG_ON(ret
); /* -ENOMEM */
6181 if (btrfs_header_generation(buf
) == trans
->transid
) {
6182 struct btrfs_block_group_cache
*cache
;
6184 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
6185 ret
= check_ref_cleanup(trans
, root
, buf
->start
);
6190 cache
= btrfs_lookup_block_group(root
->fs_info
, buf
->start
);
6192 if (btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
6193 pin_down_extent(root
, cache
, buf
->start
, buf
->len
, 1);
6194 btrfs_put_block_group(cache
);
6198 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
));
6200 btrfs_add_free_space(cache
, buf
->start
, buf
->len
);
6201 btrfs_update_reserved_bytes(cache
, buf
->len
, RESERVE_FREE
, 0);
6202 btrfs_put_block_group(cache
);
6203 trace_btrfs_reserved_extent_free(root
, buf
->start
, buf
->len
);
6208 add_pinned_bytes(root
->fs_info
, buf
->len
,
6209 btrfs_header_level(buf
),
6210 root
->root_key
.objectid
);
6213 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6216 clear_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
);
6219 /* Can return -ENOMEM */
6220 int btrfs_free_extent(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
6221 u64 bytenr
, u64 num_bytes
, u64 parent
, u64 root_objectid
,
6222 u64 owner
, u64 offset
, int no_quota
)
6225 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6227 if (btrfs_test_is_dummy_root(root
))
6230 add_pinned_bytes(root
->fs_info
, num_bytes
, owner
, root_objectid
);
6233 * tree log blocks never actually go into the extent allocation
6234 * tree, just update pinning info and exit early.
6236 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
6237 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
6238 /* unlocks the pinned mutex */
6239 btrfs_pin_extent(root
, bytenr
, num_bytes
, 1);
6241 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
6242 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
6244 parent
, root_objectid
, (int)owner
,
6245 BTRFS_DROP_DELAYED_REF
, NULL
, no_quota
);
6247 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
6249 parent
, root_objectid
, owner
,
6250 offset
, BTRFS_DROP_DELAYED_REF
,
6257 * when we wait for progress in the block group caching, its because
6258 * our allocation attempt failed at least once. So, we must sleep
6259 * and let some progress happen before we try again.
6261 * This function will sleep at least once waiting for new free space to
6262 * show up, and then it will check the block group free space numbers
6263 * for our min num_bytes. Another option is to have it go ahead
6264 * and look in the rbtree for a free extent of a given size, but this
6267 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6268 * any of the information in this block group.
6270 static noinline
void
6271 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
6274 struct btrfs_caching_control
*caching_ctl
;
6276 caching_ctl
= get_caching_control(cache
);
6280 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
) ||
6281 (cache
->free_space_ctl
->free_space
>= num_bytes
));
6283 put_caching_control(caching_ctl
);
6287 wait_block_group_cache_done(struct btrfs_block_group_cache
*cache
)
6289 struct btrfs_caching_control
*caching_ctl
;
6292 caching_ctl
= get_caching_control(cache
);
6294 return (cache
->cached
== BTRFS_CACHE_ERROR
) ? -EIO
: 0;
6296 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
));
6297 if (cache
->cached
== BTRFS_CACHE_ERROR
)
6299 put_caching_control(caching_ctl
);
6303 int __get_raid_index(u64 flags
)
6305 if (flags
& BTRFS_BLOCK_GROUP_RAID10
)
6306 return BTRFS_RAID_RAID10
;
6307 else if (flags
& BTRFS_BLOCK_GROUP_RAID1
)
6308 return BTRFS_RAID_RAID1
;
6309 else if (flags
& BTRFS_BLOCK_GROUP_DUP
)
6310 return BTRFS_RAID_DUP
;
6311 else if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
6312 return BTRFS_RAID_RAID0
;
6313 else if (flags
& BTRFS_BLOCK_GROUP_RAID5
)
6314 return BTRFS_RAID_RAID5
;
6315 else if (flags
& BTRFS_BLOCK_GROUP_RAID6
)
6316 return BTRFS_RAID_RAID6
;
6318 return BTRFS_RAID_SINGLE
; /* BTRFS_BLOCK_GROUP_SINGLE */
6321 int get_block_group_index(struct btrfs_block_group_cache
*cache
)
6323 return __get_raid_index(cache
->flags
);
6326 static const char *btrfs_raid_type_names
[BTRFS_NR_RAID_TYPES
] = {
6327 [BTRFS_RAID_RAID10
] = "raid10",
6328 [BTRFS_RAID_RAID1
] = "raid1",
6329 [BTRFS_RAID_DUP
] = "dup",
6330 [BTRFS_RAID_RAID0
] = "raid0",
6331 [BTRFS_RAID_SINGLE
] = "single",
6332 [BTRFS_RAID_RAID5
] = "raid5",
6333 [BTRFS_RAID_RAID6
] = "raid6",
6336 static const char *get_raid_name(enum btrfs_raid_types type
)
6338 if (type
>= BTRFS_NR_RAID_TYPES
)
6341 return btrfs_raid_type_names
[type
];
6344 enum btrfs_loop_type
{
6345 LOOP_CACHING_NOWAIT
= 0,
6346 LOOP_CACHING_WAIT
= 1,
6347 LOOP_ALLOC_CHUNK
= 2,
6348 LOOP_NO_EMPTY_SIZE
= 3,
6352 btrfs_lock_block_group(struct btrfs_block_group_cache
*cache
,
6356 down_read(&cache
->data_rwsem
);
6360 btrfs_grab_block_group(struct btrfs_block_group_cache
*cache
,
6363 btrfs_get_block_group(cache
);
6365 down_read(&cache
->data_rwsem
);
6368 static struct btrfs_block_group_cache
*
6369 btrfs_lock_cluster(struct btrfs_block_group_cache
*block_group
,
6370 struct btrfs_free_cluster
*cluster
,
6373 struct btrfs_block_group_cache
*used_bg
;
6374 bool locked
= false;
6376 spin_lock(&cluster
->refill_lock
);
6378 if (used_bg
== cluster
->block_group
)
6381 up_read(&used_bg
->data_rwsem
);
6382 btrfs_put_block_group(used_bg
);
6385 used_bg
= cluster
->block_group
;
6389 if (used_bg
== block_group
)
6392 btrfs_get_block_group(used_bg
);
6397 if (down_read_trylock(&used_bg
->data_rwsem
))
6400 spin_unlock(&cluster
->refill_lock
);
6401 down_read(&used_bg
->data_rwsem
);
6407 btrfs_release_block_group(struct btrfs_block_group_cache
*cache
,
6411 up_read(&cache
->data_rwsem
);
6412 btrfs_put_block_group(cache
);
6416 * walks the btree of allocated extents and find a hole of a given size.
6417 * The key ins is changed to record the hole:
6418 * ins->objectid == start position
6419 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6420 * ins->offset == the size of the hole.
6421 * Any available blocks before search_start are skipped.
6423 * If there is no suitable free space, we will record the max size of
6424 * the free space extent currently.
6426 static noinline
int find_free_extent(struct btrfs_root
*orig_root
,
6427 u64 num_bytes
, u64 empty_size
,
6428 u64 hint_byte
, struct btrfs_key
*ins
,
6429 u64 flags
, int delalloc
)
6432 struct btrfs_root
*root
= orig_root
->fs_info
->extent_root
;
6433 struct btrfs_free_cluster
*last_ptr
= NULL
;
6434 struct btrfs_block_group_cache
*block_group
= NULL
;
6435 u64 search_start
= 0;
6436 u64 max_extent_size
= 0;
6437 int empty_cluster
= 2 * 1024 * 1024;
6438 struct btrfs_space_info
*space_info
;
6440 int index
= __get_raid_index(flags
);
6441 int alloc_type
= (flags
& BTRFS_BLOCK_GROUP_DATA
) ?
6442 RESERVE_ALLOC_NO_ACCOUNT
: RESERVE_ALLOC
;
6443 bool failed_cluster_refill
= false;
6444 bool failed_alloc
= false;
6445 bool use_cluster
= true;
6446 bool have_caching_bg
= false;
6448 WARN_ON(num_bytes
< root
->sectorsize
);
6449 ins
->type
= BTRFS_EXTENT_ITEM_KEY
;
6453 trace_find_free_extent(orig_root
, num_bytes
, empty_size
, flags
);
6455 space_info
= __find_space_info(root
->fs_info
, flags
);
6457 btrfs_err(root
->fs_info
, "No space info for %llu", flags
);
6462 * If the space info is for both data and metadata it means we have a
6463 * small filesystem and we can't use the clustering stuff.
6465 if (btrfs_mixed_space_info(space_info
))
6466 use_cluster
= false;
6468 if (flags
& BTRFS_BLOCK_GROUP_METADATA
&& use_cluster
) {
6469 last_ptr
= &root
->fs_info
->meta_alloc_cluster
;
6470 if (!btrfs_test_opt(root
, SSD
))
6471 empty_cluster
= 64 * 1024;
6474 if ((flags
& BTRFS_BLOCK_GROUP_DATA
) && use_cluster
&&
6475 btrfs_test_opt(root
, SSD
)) {
6476 last_ptr
= &root
->fs_info
->data_alloc_cluster
;
6480 spin_lock(&last_ptr
->lock
);
6481 if (last_ptr
->block_group
)
6482 hint_byte
= last_ptr
->window_start
;
6483 spin_unlock(&last_ptr
->lock
);
6486 search_start
= max(search_start
, first_logical_byte(root
, 0));
6487 search_start
= max(search_start
, hint_byte
);
6492 if (search_start
== hint_byte
) {
6493 block_group
= btrfs_lookup_block_group(root
->fs_info
,
6496 * we don't want to use the block group if it doesn't match our
6497 * allocation bits, or if its not cached.
6499 * However if we are re-searching with an ideal block group
6500 * picked out then we don't care that the block group is cached.
6502 if (block_group
&& block_group_bits(block_group
, flags
) &&
6503 block_group
->cached
!= BTRFS_CACHE_NO
) {
6504 down_read(&space_info
->groups_sem
);
6505 if (list_empty(&block_group
->list
) ||
6508 * someone is removing this block group,
6509 * we can't jump into the have_block_group
6510 * target because our list pointers are not
6513 btrfs_put_block_group(block_group
);
6514 up_read(&space_info
->groups_sem
);
6516 index
= get_block_group_index(block_group
);
6517 btrfs_lock_block_group(block_group
, delalloc
);
6518 goto have_block_group
;
6520 } else if (block_group
) {
6521 btrfs_put_block_group(block_group
);
6525 have_caching_bg
= false;
6526 down_read(&space_info
->groups_sem
);
6527 list_for_each_entry(block_group
, &space_info
->block_groups
[index
],
6532 btrfs_grab_block_group(block_group
, delalloc
);
6533 search_start
= block_group
->key
.objectid
;
6536 * this can happen if we end up cycling through all the
6537 * raid types, but we want to make sure we only allocate
6538 * for the proper type.
6540 if (!block_group_bits(block_group
, flags
)) {
6541 u64 extra
= BTRFS_BLOCK_GROUP_DUP
|
6542 BTRFS_BLOCK_GROUP_RAID1
|
6543 BTRFS_BLOCK_GROUP_RAID5
|
6544 BTRFS_BLOCK_GROUP_RAID6
|
6545 BTRFS_BLOCK_GROUP_RAID10
;
6548 * if they asked for extra copies and this block group
6549 * doesn't provide them, bail. This does allow us to
6550 * fill raid0 from raid1.
6552 if ((flags
& extra
) && !(block_group
->flags
& extra
))
6557 cached
= block_group_cache_done(block_group
);
6558 if (unlikely(!cached
)) {
6559 ret
= cache_block_group(block_group
, 0);
6564 if (unlikely(block_group
->cached
== BTRFS_CACHE_ERROR
))
6566 if (unlikely(block_group
->ro
))
6570 * Ok we want to try and use the cluster allocator, so
6574 struct btrfs_block_group_cache
*used_block_group
;
6575 unsigned long aligned_cluster
;
6577 * the refill lock keeps out other
6578 * people trying to start a new cluster
6580 used_block_group
= btrfs_lock_cluster(block_group
,
6583 if (!used_block_group
)
6584 goto refill_cluster
;
6586 if (used_block_group
!= block_group
&&
6587 (used_block_group
->ro
||
6588 !block_group_bits(used_block_group
, flags
)))
6589 goto release_cluster
;
6591 offset
= btrfs_alloc_from_cluster(used_block_group
,
6594 used_block_group
->key
.objectid
,
6597 /* we have a block, we're done */
6598 spin_unlock(&last_ptr
->refill_lock
);
6599 trace_btrfs_reserve_extent_cluster(root
,
6601 search_start
, num_bytes
);
6602 if (used_block_group
!= block_group
) {
6603 btrfs_release_block_group(block_group
,
6605 block_group
= used_block_group
;
6610 WARN_ON(last_ptr
->block_group
!= used_block_group
);
6612 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6613 * set up a new clusters, so lets just skip it
6614 * and let the allocator find whatever block
6615 * it can find. If we reach this point, we
6616 * will have tried the cluster allocator
6617 * plenty of times and not have found
6618 * anything, so we are likely way too
6619 * fragmented for the clustering stuff to find
6622 * However, if the cluster is taken from the
6623 * current block group, release the cluster
6624 * first, so that we stand a better chance of
6625 * succeeding in the unclustered
6627 if (loop
>= LOOP_NO_EMPTY_SIZE
&&
6628 used_block_group
!= block_group
) {
6629 spin_unlock(&last_ptr
->refill_lock
);
6630 btrfs_release_block_group(used_block_group
,
6632 goto unclustered_alloc
;
6636 * this cluster didn't work out, free it and
6639 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
6641 if (used_block_group
!= block_group
)
6642 btrfs_release_block_group(used_block_group
,
6645 if (loop
>= LOOP_NO_EMPTY_SIZE
) {
6646 spin_unlock(&last_ptr
->refill_lock
);
6647 goto unclustered_alloc
;
6650 aligned_cluster
= max_t(unsigned long,
6651 empty_cluster
+ empty_size
,
6652 block_group
->full_stripe_len
);
6654 /* allocate a cluster in this block group */
6655 ret
= btrfs_find_space_cluster(root
, block_group
,
6656 last_ptr
, search_start
,
6661 * now pull our allocation out of this
6664 offset
= btrfs_alloc_from_cluster(block_group
,
6670 /* we found one, proceed */
6671 spin_unlock(&last_ptr
->refill_lock
);
6672 trace_btrfs_reserve_extent_cluster(root
,
6673 block_group
, search_start
,
6677 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
6678 && !failed_cluster_refill
) {
6679 spin_unlock(&last_ptr
->refill_lock
);
6681 failed_cluster_refill
= true;
6682 wait_block_group_cache_progress(block_group
,
6683 num_bytes
+ empty_cluster
+ empty_size
);
6684 goto have_block_group
;
6688 * at this point we either didn't find a cluster
6689 * or we weren't able to allocate a block from our
6690 * cluster. Free the cluster we've been trying
6691 * to use, and go to the next block group
6693 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
6694 spin_unlock(&last_ptr
->refill_lock
);
6699 spin_lock(&block_group
->free_space_ctl
->tree_lock
);
6701 block_group
->free_space_ctl
->free_space
<
6702 num_bytes
+ empty_cluster
+ empty_size
) {
6703 if (block_group
->free_space_ctl
->free_space
>
6706 block_group
->free_space_ctl
->free_space
;
6707 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
6710 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
6712 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
6713 num_bytes
, empty_size
,
6716 * If we didn't find a chunk, and we haven't failed on this
6717 * block group before, and this block group is in the middle of
6718 * caching and we are ok with waiting, then go ahead and wait
6719 * for progress to be made, and set failed_alloc to true.
6721 * If failed_alloc is true then we've already waited on this
6722 * block group once and should move on to the next block group.
6724 if (!offset
&& !failed_alloc
&& !cached
&&
6725 loop
> LOOP_CACHING_NOWAIT
) {
6726 wait_block_group_cache_progress(block_group
,
6727 num_bytes
+ empty_size
);
6728 failed_alloc
= true;
6729 goto have_block_group
;
6730 } else if (!offset
) {
6732 have_caching_bg
= true;
6736 search_start
= ALIGN(offset
, root
->stripesize
);
6738 /* move on to the next group */
6739 if (search_start
+ num_bytes
>
6740 block_group
->key
.objectid
+ block_group
->key
.offset
) {
6741 btrfs_add_free_space(block_group
, offset
, num_bytes
);
6745 if (offset
< search_start
)
6746 btrfs_add_free_space(block_group
, offset
,
6747 search_start
- offset
);
6748 BUG_ON(offset
> search_start
);
6750 ret
= btrfs_update_reserved_bytes(block_group
, num_bytes
,
6751 alloc_type
, delalloc
);
6752 if (ret
== -EAGAIN
) {
6753 btrfs_add_free_space(block_group
, offset
, num_bytes
);
6757 /* we are all good, lets return */
6758 ins
->objectid
= search_start
;
6759 ins
->offset
= num_bytes
;
6761 trace_btrfs_reserve_extent(orig_root
, block_group
,
6762 search_start
, num_bytes
);
6763 btrfs_release_block_group(block_group
, delalloc
);
6766 failed_cluster_refill
= false;
6767 failed_alloc
= false;
6768 BUG_ON(index
!= get_block_group_index(block_group
));
6769 btrfs_release_block_group(block_group
, delalloc
);
6771 up_read(&space_info
->groups_sem
);
6773 if (!ins
->objectid
&& loop
>= LOOP_CACHING_WAIT
&& have_caching_bg
)
6776 if (!ins
->objectid
&& ++index
< BTRFS_NR_RAID_TYPES
)
6780 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
6781 * caching kthreads as we move along
6782 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
6783 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
6784 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
6787 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
) {
6790 if (loop
== LOOP_ALLOC_CHUNK
) {
6791 struct btrfs_trans_handle
*trans
;
6794 trans
= current
->journal_info
;
6798 trans
= btrfs_join_transaction(root
);
6800 if (IS_ERR(trans
)) {
6801 ret
= PTR_ERR(trans
);
6805 ret
= do_chunk_alloc(trans
, root
, flags
,
6808 * Do not bail out on ENOSPC since we
6809 * can do more things.
6811 if (ret
< 0 && ret
!= -ENOSPC
)
6812 btrfs_abort_transaction(trans
,
6817 btrfs_end_transaction(trans
, root
);
6822 if (loop
== LOOP_NO_EMPTY_SIZE
) {
6828 } else if (!ins
->objectid
) {
6830 } else if (ins
->objectid
) {
6835 ins
->offset
= max_extent_size
;
6839 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
6840 int dump_block_groups
)
6842 struct btrfs_block_group_cache
*cache
;
6845 spin_lock(&info
->lock
);
6846 printk(KERN_INFO
"BTRFS: space_info %llu has %llu free, is %sfull\n",
6848 info
->total_bytes
- info
->bytes_used
- info
->bytes_pinned
-
6849 info
->bytes_reserved
- info
->bytes_readonly
,
6850 (info
->full
) ? "" : "not ");
6851 printk(KERN_INFO
"BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
6852 "reserved=%llu, may_use=%llu, readonly=%llu\n",
6853 info
->total_bytes
, info
->bytes_used
, info
->bytes_pinned
,
6854 info
->bytes_reserved
, info
->bytes_may_use
,
6855 info
->bytes_readonly
);
6856 spin_unlock(&info
->lock
);
6858 if (!dump_block_groups
)
6861 down_read(&info
->groups_sem
);
6863 list_for_each_entry(cache
, &info
->block_groups
[index
], list
) {
6864 spin_lock(&cache
->lock
);
6865 printk(KERN_INFO
"BTRFS: "
6866 "block group %llu has %llu bytes, "
6867 "%llu used %llu pinned %llu reserved %s\n",
6868 cache
->key
.objectid
, cache
->key
.offset
,
6869 btrfs_block_group_used(&cache
->item
), cache
->pinned
,
6870 cache
->reserved
, cache
->ro
? "[readonly]" : "");
6871 btrfs_dump_free_space(cache
, bytes
);
6872 spin_unlock(&cache
->lock
);
6874 if (++index
< BTRFS_NR_RAID_TYPES
)
6876 up_read(&info
->groups_sem
);
6879 int btrfs_reserve_extent(struct btrfs_root
*root
,
6880 u64 num_bytes
, u64 min_alloc_size
,
6881 u64 empty_size
, u64 hint_byte
,
6882 struct btrfs_key
*ins
, int is_data
, int delalloc
)
6884 bool final_tried
= false;
6888 flags
= btrfs_get_alloc_profile(root
, is_data
);
6890 WARN_ON(num_bytes
< root
->sectorsize
);
6891 ret
= find_free_extent(root
, num_bytes
, empty_size
, hint_byte
, ins
,
6894 if (ret
== -ENOSPC
) {
6895 if (!final_tried
&& ins
->offset
) {
6896 num_bytes
= min(num_bytes
>> 1, ins
->offset
);
6897 num_bytes
= round_down(num_bytes
, root
->sectorsize
);
6898 num_bytes
= max(num_bytes
, min_alloc_size
);
6899 if (num_bytes
== min_alloc_size
)
6902 } else if (btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
6903 struct btrfs_space_info
*sinfo
;
6905 sinfo
= __find_space_info(root
->fs_info
, flags
);
6906 btrfs_err(root
->fs_info
, "allocation failed flags %llu, wanted %llu",
6909 dump_space_info(sinfo
, num_bytes
, 1);
6916 static int __btrfs_free_reserved_extent(struct btrfs_root
*root
,
6918 int pin
, int delalloc
)
6920 struct btrfs_block_group_cache
*cache
;
6923 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
6925 btrfs_err(root
->fs_info
, "Unable to find block group for %llu",
6930 if (btrfs_test_opt(root
, DISCARD
))
6931 ret
= btrfs_discard_extent(root
, start
, len
, NULL
);
6934 pin_down_extent(root
, cache
, start
, len
, 1);
6936 btrfs_add_free_space(cache
, start
, len
);
6937 btrfs_update_reserved_bytes(cache
, len
, RESERVE_FREE
, delalloc
);
6939 btrfs_put_block_group(cache
);
6941 trace_btrfs_reserved_extent_free(root
, start
, len
);
6946 int btrfs_free_reserved_extent(struct btrfs_root
*root
,
6947 u64 start
, u64 len
, int delalloc
)
6949 return __btrfs_free_reserved_extent(root
, start
, len
, 0, delalloc
);
6952 int btrfs_free_and_pin_reserved_extent(struct btrfs_root
*root
,
6955 return __btrfs_free_reserved_extent(root
, start
, len
, 1, 0);
6958 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
6959 struct btrfs_root
*root
,
6960 u64 parent
, u64 root_objectid
,
6961 u64 flags
, u64 owner
, u64 offset
,
6962 struct btrfs_key
*ins
, int ref_mod
)
6965 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6966 struct btrfs_extent_item
*extent_item
;
6967 struct btrfs_extent_inline_ref
*iref
;
6968 struct btrfs_path
*path
;
6969 struct extent_buffer
*leaf
;
6974 type
= BTRFS_SHARED_DATA_REF_KEY
;
6976 type
= BTRFS_EXTENT_DATA_REF_KEY
;
6978 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
6980 path
= btrfs_alloc_path();
6984 path
->leave_spinning
= 1;
6985 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
6988 btrfs_free_path(path
);
6992 leaf
= path
->nodes
[0];
6993 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
6994 struct btrfs_extent_item
);
6995 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
6996 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
6997 btrfs_set_extent_flags(leaf
, extent_item
,
6998 flags
| BTRFS_EXTENT_FLAG_DATA
);
7000 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
7001 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
7003 struct btrfs_shared_data_ref
*ref
;
7004 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
7005 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
7006 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
7008 struct btrfs_extent_data_ref
*ref
;
7009 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
7010 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
7011 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
7012 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
7013 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
7016 btrfs_mark_buffer_dirty(path
->nodes
[0]);
7017 btrfs_free_path(path
);
7019 /* Always set parent to 0 here since its exclusive anyway. */
7020 ret
= btrfs_qgroup_record_ref(trans
, fs_info
, root_objectid
,
7021 ins
->objectid
, ins
->offset
,
7022 BTRFS_QGROUP_OPER_ADD_EXCL
, 0);
7026 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1);
7027 if (ret
) { /* -ENOENT, logic error */
7028 btrfs_err(fs_info
, "update block group failed for %llu %llu",
7029 ins
->objectid
, ins
->offset
);
7032 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, ins
->offset
);
7036 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
7037 struct btrfs_root
*root
,
7038 u64 parent
, u64 root_objectid
,
7039 u64 flags
, struct btrfs_disk_key
*key
,
7040 int level
, struct btrfs_key
*ins
,
7044 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7045 struct btrfs_extent_item
*extent_item
;
7046 struct btrfs_tree_block_info
*block_info
;
7047 struct btrfs_extent_inline_ref
*iref
;
7048 struct btrfs_path
*path
;
7049 struct extent_buffer
*leaf
;
7050 u32 size
= sizeof(*extent_item
) + sizeof(*iref
);
7051 u64 num_bytes
= ins
->offset
;
7052 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
7055 if (!skinny_metadata
)
7056 size
+= sizeof(*block_info
);
7058 path
= btrfs_alloc_path();
7060 btrfs_free_and_pin_reserved_extent(root
, ins
->objectid
,
7065 path
->leave_spinning
= 1;
7066 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
7069 btrfs_free_and_pin_reserved_extent(root
, ins
->objectid
,
7071 btrfs_free_path(path
);
7075 leaf
= path
->nodes
[0];
7076 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
7077 struct btrfs_extent_item
);
7078 btrfs_set_extent_refs(leaf
, extent_item
, 1);
7079 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
7080 btrfs_set_extent_flags(leaf
, extent_item
,
7081 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
7083 if (skinny_metadata
) {
7084 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
7085 num_bytes
= root
->nodesize
;
7087 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
7088 btrfs_set_tree_block_key(leaf
, block_info
, key
);
7089 btrfs_set_tree_block_level(leaf
, block_info
, level
);
7090 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
7094 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
7095 btrfs_set_extent_inline_ref_type(leaf
, iref
,
7096 BTRFS_SHARED_BLOCK_REF_KEY
);
7097 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
7099 btrfs_set_extent_inline_ref_type(leaf
, iref
,
7100 BTRFS_TREE_BLOCK_REF_KEY
);
7101 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
7104 btrfs_mark_buffer_dirty(leaf
);
7105 btrfs_free_path(path
);
7108 ret
= btrfs_qgroup_record_ref(trans
, fs_info
, root_objectid
,
7109 ins
->objectid
, num_bytes
,
7110 BTRFS_QGROUP_OPER_ADD_EXCL
, 0);
7115 ret
= update_block_group(trans
, root
, ins
->objectid
, root
->nodesize
,
7117 if (ret
) { /* -ENOENT, logic error */
7118 btrfs_err(fs_info
, "update block group failed for %llu %llu",
7119 ins
->objectid
, ins
->offset
);
7123 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, root
->nodesize
);
7127 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
7128 struct btrfs_root
*root
,
7129 u64 root_objectid
, u64 owner
,
7130 u64 offset
, struct btrfs_key
*ins
)
7134 BUG_ON(root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
7136 ret
= btrfs_add_delayed_data_ref(root
->fs_info
, trans
, ins
->objectid
,
7138 root_objectid
, owner
, offset
,
7139 BTRFS_ADD_DELAYED_EXTENT
, NULL
, 0);
7144 * this is used by the tree logging recovery code. It records that
7145 * an extent has been allocated and makes sure to clear the free
7146 * space cache bits as well
7148 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
7149 struct btrfs_root
*root
,
7150 u64 root_objectid
, u64 owner
, u64 offset
,
7151 struct btrfs_key
*ins
)
7154 struct btrfs_block_group_cache
*block_group
;
7157 * Mixed block groups will exclude before processing the log so we only
7158 * need to do the exlude dance if this fs isn't mixed.
7160 if (!btrfs_fs_incompat(root
->fs_info
, MIXED_GROUPS
)) {
7161 ret
= __exclude_logged_extent(root
, ins
->objectid
, ins
->offset
);
7166 block_group
= btrfs_lookup_block_group(root
->fs_info
, ins
->objectid
);
7170 ret
= btrfs_update_reserved_bytes(block_group
, ins
->offset
,
7171 RESERVE_ALLOC_NO_ACCOUNT
, 0);
7172 BUG_ON(ret
); /* logic error */
7173 ret
= alloc_reserved_file_extent(trans
, root
, 0, root_objectid
,
7174 0, owner
, offset
, ins
, 1);
7175 btrfs_put_block_group(block_group
);
7179 static struct extent_buffer
*
7180 btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
7181 u64 bytenr
, int level
)
7183 struct extent_buffer
*buf
;
7185 buf
= btrfs_find_create_tree_block(root
, bytenr
);
7187 return ERR_PTR(-ENOMEM
);
7188 btrfs_set_header_generation(buf
, trans
->transid
);
7189 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, buf
, level
);
7190 btrfs_tree_lock(buf
);
7191 clean_tree_block(trans
, root
, buf
);
7192 clear_bit(EXTENT_BUFFER_STALE
, &buf
->bflags
);
7194 btrfs_set_lock_blocking(buf
);
7195 btrfs_set_buffer_uptodate(buf
);
7197 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
7198 buf
->log_index
= root
->log_transid
% 2;
7200 * we allow two log transactions at a time, use different
7201 * EXENT bit to differentiate dirty pages.
7203 if (buf
->log_index
== 0)
7204 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
7205 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
7207 set_extent_new(&root
->dirty_log_pages
, buf
->start
,
7208 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
7210 buf
->log_index
= -1;
7211 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
7212 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
7214 trans
->blocks_used
++;
7215 /* this returns a buffer locked for blocking */
7219 static struct btrfs_block_rsv
*
7220 use_block_rsv(struct btrfs_trans_handle
*trans
,
7221 struct btrfs_root
*root
, u32 blocksize
)
7223 struct btrfs_block_rsv
*block_rsv
;
7224 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
7226 bool global_updated
= false;
7228 block_rsv
= get_block_rsv(trans
, root
);
7230 if (unlikely(block_rsv
->size
== 0))
7233 ret
= block_rsv_use_bytes(block_rsv
, blocksize
);
7237 if (block_rsv
->failfast
)
7238 return ERR_PTR(ret
);
7240 if (block_rsv
->type
== BTRFS_BLOCK_RSV_GLOBAL
&& !global_updated
) {
7241 global_updated
= true;
7242 update_global_block_rsv(root
->fs_info
);
7246 if (btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
7247 static DEFINE_RATELIMIT_STATE(_rs
,
7248 DEFAULT_RATELIMIT_INTERVAL
* 10,
7249 /*DEFAULT_RATELIMIT_BURST*/ 1);
7250 if (__ratelimit(&_rs
))
7252 "BTRFS: block rsv returned %d\n", ret
);
7255 ret
= reserve_metadata_bytes(root
, block_rsv
, blocksize
,
7256 BTRFS_RESERVE_NO_FLUSH
);
7260 * If we couldn't reserve metadata bytes try and use some from
7261 * the global reserve if its space type is the same as the global
7264 if (block_rsv
->type
!= BTRFS_BLOCK_RSV_GLOBAL
&&
7265 block_rsv
->space_info
== global_rsv
->space_info
) {
7266 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
7270 return ERR_PTR(ret
);
7273 static void unuse_block_rsv(struct btrfs_fs_info
*fs_info
,
7274 struct btrfs_block_rsv
*block_rsv
, u32 blocksize
)
7276 block_rsv_add_bytes(block_rsv
, blocksize
, 0);
7277 block_rsv_release_bytes(fs_info
, block_rsv
, NULL
, 0);
7281 * finds a free extent and does all the dirty work required for allocation
7282 * returns the key for the extent through ins, and a tree buffer for
7283 * the first block of the extent through buf.
7285 * returns the tree buffer or NULL.
7287 struct extent_buffer
*btrfs_alloc_tree_block(struct btrfs_trans_handle
*trans
,
7288 struct btrfs_root
*root
,
7289 u64 parent
, u64 root_objectid
,
7290 struct btrfs_disk_key
*key
, int level
,
7291 u64 hint
, u64 empty_size
)
7293 struct btrfs_key ins
;
7294 struct btrfs_block_rsv
*block_rsv
;
7295 struct extent_buffer
*buf
;
7298 u32 blocksize
= root
->nodesize
;
7299 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
7302 if (btrfs_test_is_dummy_root(root
)) {
7303 buf
= btrfs_init_new_buffer(trans
, root
, root
->alloc_bytenr
,
7306 root
->alloc_bytenr
+= blocksize
;
7310 block_rsv
= use_block_rsv(trans
, root
, blocksize
);
7311 if (IS_ERR(block_rsv
))
7312 return ERR_CAST(block_rsv
);
7314 ret
= btrfs_reserve_extent(root
, blocksize
, blocksize
,
7315 empty_size
, hint
, &ins
, 0, 0);
7317 unuse_block_rsv(root
->fs_info
, block_rsv
, blocksize
);
7318 return ERR_PTR(ret
);
7321 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
, level
);
7322 BUG_ON(IS_ERR(buf
)); /* -ENOMEM */
7324 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
7326 parent
= ins
.objectid
;
7327 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
7331 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
7332 struct btrfs_delayed_extent_op
*extent_op
;
7333 extent_op
= btrfs_alloc_delayed_extent_op();
7334 BUG_ON(!extent_op
); /* -ENOMEM */
7336 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
7338 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
7339 extent_op
->flags_to_set
= flags
;
7340 if (skinny_metadata
)
7341 extent_op
->update_key
= 0;
7343 extent_op
->update_key
= 1;
7344 extent_op
->update_flags
= 1;
7345 extent_op
->is_data
= 0;
7346 extent_op
->level
= level
;
7348 ret
= btrfs_add_delayed_tree_ref(root
->fs_info
, trans
,
7350 ins
.offset
, parent
, root_objectid
,
7351 level
, BTRFS_ADD_DELAYED_EXTENT
,
7353 BUG_ON(ret
); /* -ENOMEM */
7358 struct walk_control
{
7359 u64 refs
[BTRFS_MAX_LEVEL
];
7360 u64 flags
[BTRFS_MAX_LEVEL
];
7361 struct btrfs_key update_progress
;
7372 #define DROP_REFERENCE 1
7373 #define UPDATE_BACKREF 2
7375 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
7376 struct btrfs_root
*root
,
7377 struct walk_control
*wc
,
7378 struct btrfs_path
*path
)
7386 struct btrfs_key key
;
7387 struct extent_buffer
*eb
;
7392 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
7393 wc
->reada_count
= wc
->reada_count
* 2 / 3;
7394 wc
->reada_count
= max(wc
->reada_count
, 2);
7396 wc
->reada_count
= wc
->reada_count
* 3 / 2;
7397 wc
->reada_count
= min_t(int, wc
->reada_count
,
7398 BTRFS_NODEPTRS_PER_BLOCK(root
));
7401 eb
= path
->nodes
[wc
->level
];
7402 nritems
= btrfs_header_nritems(eb
);
7403 blocksize
= root
->nodesize
;
7405 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
7406 if (nread
>= wc
->reada_count
)
7410 bytenr
= btrfs_node_blockptr(eb
, slot
);
7411 generation
= btrfs_node_ptr_generation(eb
, slot
);
7413 if (slot
== path
->slots
[wc
->level
])
7416 if (wc
->stage
== UPDATE_BACKREF
&&
7417 generation
<= root
->root_key
.offset
)
7420 /* We don't lock the tree block, it's OK to be racy here */
7421 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
,
7422 wc
->level
- 1, 1, &refs
,
7424 /* We don't care about errors in readahead. */
7429 if (wc
->stage
== DROP_REFERENCE
) {
7433 if (wc
->level
== 1 &&
7434 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
7436 if (!wc
->update_ref
||
7437 generation
<= root
->root_key
.offset
)
7439 btrfs_node_key_to_cpu(eb
, &key
, slot
);
7440 ret
= btrfs_comp_cpu_keys(&key
,
7441 &wc
->update_progress
);
7445 if (wc
->level
== 1 &&
7446 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
7450 readahead_tree_block(root
, bytenr
);
7453 wc
->reada_slot
= slot
;
7456 static int account_leaf_items(struct btrfs_trans_handle
*trans
,
7457 struct btrfs_root
*root
,
7458 struct extent_buffer
*eb
)
7460 int nr
= btrfs_header_nritems(eb
);
7461 int i
, extent_type
, ret
;
7462 struct btrfs_key key
;
7463 struct btrfs_file_extent_item
*fi
;
7464 u64 bytenr
, num_bytes
;
7466 for (i
= 0; i
< nr
; i
++) {
7467 btrfs_item_key_to_cpu(eb
, &key
, i
);
7469 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
7472 fi
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
7473 /* filter out non qgroup-accountable extents */
7474 extent_type
= btrfs_file_extent_type(eb
, fi
);
7476 if (extent_type
== BTRFS_FILE_EXTENT_INLINE
)
7479 bytenr
= btrfs_file_extent_disk_bytenr(eb
, fi
);
7483 num_bytes
= btrfs_file_extent_disk_num_bytes(eb
, fi
);
7485 ret
= btrfs_qgroup_record_ref(trans
, root
->fs_info
,
7488 BTRFS_QGROUP_OPER_SUB_SUBTREE
, 0);
7496 * Walk up the tree from the bottom, freeing leaves and any interior
7497 * nodes which have had all slots visited. If a node (leaf or
7498 * interior) is freed, the node above it will have it's slot
7499 * incremented. The root node will never be freed.
7501 * At the end of this function, we should have a path which has all
7502 * slots incremented to the next position for a search. If we need to
7503 * read a new node it will be NULL and the node above it will have the
7504 * correct slot selected for a later read.
7506 * If we increment the root nodes slot counter past the number of
7507 * elements, 1 is returned to signal completion of the search.
7509 static int adjust_slots_upwards(struct btrfs_root
*root
,
7510 struct btrfs_path
*path
, int root_level
)
7514 struct extent_buffer
*eb
;
7516 if (root_level
== 0)
7519 while (level
<= root_level
) {
7520 eb
= path
->nodes
[level
];
7521 nr
= btrfs_header_nritems(eb
);
7522 path
->slots
[level
]++;
7523 slot
= path
->slots
[level
];
7524 if (slot
>= nr
|| level
== 0) {
7526 * Don't free the root - we will detect this
7527 * condition after our loop and return a
7528 * positive value for caller to stop walking the tree.
7530 if (level
!= root_level
) {
7531 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
7532 path
->locks
[level
] = 0;
7534 free_extent_buffer(eb
);
7535 path
->nodes
[level
] = NULL
;
7536 path
->slots
[level
] = 0;
7540 * We have a valid slot to walk back down
7541 * from. Stop here so caller can process these
7550 eb
= path
->nodes
[root_level
];
7551 if (path
->slots
[root_level
] >= btrfs_header_nritems(eb
))
7558 * root_eb is the subtree root and is locked before this function is called.
7560 static int account_shared_subtree(struct btrfs_trans_handle
*trans
,
7561 struct btrfs_root
*root
,
7562 struct extent_buffer
*root_eb
,
7568 struct extent_buffer
*eb
= root_eb
;
7569 struct btrfs_path
*path
= NULL
;
7571 BUG_ON(root_level
< 0 || root_level
> BTRFS_MAX_LEVEL
);
7572 BUG_ON(root_eb
== NULL
);
7574 if (!root
->fs_info
->quota_enabled
)
7577 if (!extent_buffer_uptodate(root_eb
)) {
7578 ret
= btrfs_read_buffer(root_eb
, root_gen
);
7583 if (root_level
== 0) {
7584 ret
= account_leaf_items(trans
, root
, root_eb
);
7588 path
= btrfs_alloc_path();
7593 * Walk down the tree. Missing extent blocks are filled in as
7594 * we go. Metadata is accounted every time we read a new
7597 * When we reach a leaf, we account for file extent items in it,
7598 * walk back up the tree (adjusting slot pointers as we go)
7599 * and restart the search process.
7601 extent_buffer_get(root_eb
); /* For path */
7602 path
->nodes
[root_level
] = root_eb
;
7603 path
->slots
[root_level
] = 0;
7604 path
->locks
[root_level
] = 0; /* so release_path doesn't try to unlock */
7607 while (level
>= 0) {
7608 if (path
->nodes
[level
] == NULL
) {
7613 /* We need to get child blockptr/gen from
7614 * parent before we can read it. */
7615 eb
= path
->nodes
[level
+ 1];
7616 parent_slot
= path
->slots
[level
+ 1];
7617 child_bytenr
= btrfs_node_blockptr(eb
, parent_slot
);
7618 child_gen
= btrfs_node_ptr_generation(eb
, parent_slot
);
7620 eb
= read_tree_block(root
, child_bytenr
, child_gen
);
7621 if (!eb
|| !extent_buffer_uptodate(eb
)) {
7626 path
->nodes
[level
] = eb
;
7627 path
->slots
[level
] = 0;
7629 btrfs_tree_read_lock(eb
);
7630 btrfs_set_lock_blocking_rw(eb
, BTRFS_READ_LOCK
);
7631 path
->locks
[level
] = BTRFS_READ_LOCK_BLOCKING
;
7633 ret
= btrfs_qgroup_record_ref(trans
, root
->fs_info
,
7637 BTRFS_QGROUP_OPER_SUB_SUBTREE
,
7645 ret
= account_leaf_items(trans
, root
, path
->nodes
[level
]);
7649 /* Nonzero return here means we completed our search */
7650 ret
= adjust_slots_upwards(root
, path
, root_level
);
7654 /* Restart search with new slots */
7663 btrfs_free_path(path
);
7669 * helper to process tree block while walking down the tree.
7671 * when wc->stage == UPDATE_BACKREF, this function updates
7672 * back refs for pointers in the block.
7674 * NOTE: return value 1 means we should stop walking down.
7676 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
7677 struct btrfs_root
*root
,
7678 struct btrfs_path
*path
,
7679 struct walk_control
*wc
, int lookup_info
)
7681 int level
= wc
->level
;
7682 struct extent_buffer
*eb
= path
->nodes
[level
];
7683 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
7686 if (wc
->stage
== UPDATE_BACKREF
&&
7687 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
7691 * when reference count of tree block is 1, it won't increase
7692 * again. once full backref flag is set, we never clear it.
7695 ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
7696 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
)))) {
7697 BUG_ON(!path
->locks
[level
]);
7698 ret
= btrfs_lookup_extent_info(trans
, root
,
7699 eb
->start
, level
, 1,
7702 BUG_ON(ret
== -ENOMEM
);
7705 BUG_ON(wc
->refs
[level
] == 0);
7708 if (wc
->stage
== DROP_REFERENCE
) {
7709 if (wc
->refs
[level
] > 1)
7712 if (path
->locks
[level
] && !wc
->keep_locks
) {
7713 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
7714 path
->locks
[level
] = 0;
7719 /* wc->stage == UPDATE_BACKREF */
7720 if (!(wc
->flags
[level
] & flag
)) {
7721 BUG_ON(!path
->locks
[level
]);
7722 ret
= btrfs_inc_ref(trans
, root
, eb
, 1);
7723 BUG_ON(ret
); /* -ENOMEM */
7724 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
7725 BUG_ON(ret
); /* -ENOMEM */
7726 ret
= btrfs_set_disk_extent_flags(trans
, root
, eb
->start
,
7728 btrfs_header_level(eb
), 0);
7729 BUG_ON(ret
); /* -ENOMEM */
7730 wc
->flags
[level
] |= flag
;
7734 * the block is shared by multiple trees, so it's not good to
7735 * keep the tree lock
7737 if (path
->locks
[level
] && level
> 0) {
7738 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
7739 path
->locks
[level
] = 0;
7745 * helper to process tree block pointer.
7747 * when wc->stage == DROP_REFERENCE, this function checks
7748 * reference count of the block pointed to. if the block
7749 * is shared and we need update back refs for the subtree
7750 * rooted at the block, this function changes wc->stage to
7751 * UPDATE_BACKREF. if the block is shared and there is no
7752 * need to update back, this function drops the reference
7755 * NOTE: return value 1 means we should stop walking down.
7757 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
7758 struct btrfs_root
*root
,
7759 struct btrfs_path
*path
,
7760 struct walk_control
*wc
, int *lookup_info
)
7766 struct btrfs_key key
;
7767 struct extent_buffer
*next
;
7768 int level
= wc
->level
;
7771 bool need_account
= false;
7773 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
7774 path
->slots
[level
]);
7776 * if the lower level block was created before the snapshot
7777 * was created, we know there is no need to update back refs
7780 if (wc
->stage
== UPDATE_BACKREF
&&
7781 generation
<= root
->root_key
.offset
) {
7786 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
7787 blocksize
= root
->nodesize
;
7789 next
= btrfs_find_tree_block(root
, bytenr
);
7791 next
= btrfs_find_create_tree_block(root
, bytenr
);
7794 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, next
,
7798 btrfs_tree_lock(next
);
7799 btrfs_set_lock_blocking(next
);
7801 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, level
- 1, 1,
7802 &wc
->refs
[level
- 1],
7803 &wc
->flags
[level
- 1]);
7805 btrfs_tree_unlock(next
);
7809 if (unlikely(wc
->refs
[level
- 1] == 0)) {
7810 btrfs_err(root
->fs_info
, "Missing references.");
7815 if (wc
->stage
== DROP_REFERENCE
) {
7816 if (wc
->refs
[level
- 1] > 1) {
7817 need_account
= true;
7819 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
7822 if (!wc
->update_ref
||
7823 generation
<= root
->root_key
.offset
)
7826 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
7827 path
->slots
[level
]);
7828 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
7832 wc
->stage
= UPDATE_BACKREF
;
7833 wc
->shared_level
= level
- 1;
7837 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
7841 if (!btrfs_buffer_uptodate(next
, generation
, 0)) {
7842 btrfs_tree_unlock(next
);
7843 free_extent_buffer(next
);
7849 if (reada
&& level
== 1)
7850 reada_walk_down(trans
, root
, wc
, path
);
7851 next
= read_tree_block(root
, bytenr
, generation
);
7852 if (!next
|| !extent_buffer_uptodate(next
)) {
7853 free_extent_buffer(next
);
7856 btrfs_tree_lock(next
);
7857 btrfs_set_lock_blocking(next
);
7861 BUG_ON(level
!= btrfs_header_level(next
));
7862 path
->nodes
[level
] = next
;
7863 path
->slots
[level
] = 0;
7864 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
7870 wc
->refs
[level
- 1] = 0;
7871 wc
->flags
[level
- 1] = 0;
7872 if (wc
->stage
== DROP_REFERENCE
) {
7873 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
7874 parent
= path
->nodes
[level
]->start
;
7876 BUG_ON(root
->root_key
.objectid
!=
7877 btrfs_header_owner(path
->nodes
[level
]));
7882 ret
= account_shared_subtree(trans
, root
, next
,
7883 generation
, level
- 1);
7885 printk_ratelimited(KERN_ERR
"BTRFS: %s Error "
7886 "%d accounting shared subtree. Quota "
7887 "is out of sync, rescan required.\n",
7888 root
->fs_info
->sb
->s_id
, ret
);
7891 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
, parent
,
7892 root
->root_key
.objectid
, level
- 1, 0, 0);
7893 BUG_ON(ret
); /* -ENOMEM */
7895 btrfs_tree_unlock(next
);
7896 free_extent_buffer(next
);
7902 * helper to process tree block while walking up the tree.
7904 * when wc->stage == DROP_REFERENCE, this function drops
7905 * reference count on the block.
7907 * when wc->stage == UPDATE_BACKREF, this function changes
7908 * wc->stage back to DROP_REFERENCE if we changed wc->stage
7909 * to UPDATE_BACKREF previously while processing the block.
7911 * NOTE: return value 1 means we should stop walking up.
7913 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
7914 struct btrfs_root
*root
,
7915 struct btrfs_path
*path
,
7916 struct walk_control
*wc
)
7919 int level
= wc
->level
;
7920 struct extent_buffer
*eb
= path
->nodes
[level
];
7923 if (wc
->stage
== UPDATE_BACKREF
) {
7924 BUG_ON(wc
->shared_level
< level
);
7925 if (level
< wc
->shared_level
)
7928 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
7932 wc
->stage
= DROP_REFERENCE
;
7933 wc
->shared_level
= -1;
7934 path
->slots
[level
] = 0;
7937 * check reference count again if the block isn't locked.
7938 * we should start walking down the tree again if reference
7941 if (!path
->locks
[level
]) {
7943 btrfs_tree_lock(eb
);
7944 btrfs_set_lock_blocking(eb
);
7945 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
7947 ret
= btrfs_lookup_extent_info(trans
, root
,
7948 eb
->start
, level
, 1,
7952 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
7953 path
->locks
[level
] = 0;
7956 BUG_ON(wc
->refs
[level
] == 0);
7957 if (wc
->refs
[level
] == 1) {
7958 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
7959 path
->locks
[level
] = 0;
7965 /* wc->stage == DROP_REFERENCE */
7966 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
7968 if (wc
->refs
[level
] == 1) {
7970 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
7971 ret
= btrfs_dec_ref(trans
, root
, eb
, 1);
7973 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
7974 BUG_ON(ret
); /* -ENOMEM */
7975 ret
= account_leaf_items(trans
, root
, eb
);
7977 printk_ratelimited(KERN_ERR
"BTRFS: %s Error "
7978 "%d accounting leaf items. Quota "
7979 "is out of sync, rescan required.\n",
7980 root
->fs_info
->sb
->s_id
, ret
);
7983 /* make block locked assertion in clean_tree_block happy */
7984 if (!path
->locks
[level
] &&
7985 btrfs_header_generation(eb
) == trans
->transid
) {
7986 btrfs_tree_lock(eb
);
7987 btrfs_set_lock_blocking(eb
);
7988 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
7990 clean_tree_block(trans
, root
, eb
);
7993 if (eb
== root
->node
) {
7994 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
7997 BUG_ON(root
->root_key
.objectid
!=
7998 btrfs_header_owner(eb
));
8000 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8001 parent
= path
->nodes
[level
+ 1]->start
;
8003 BUG_ON(root
->root_key
.objectid
!=
8004 btrfs_header_owner(path
->nodes
[level
+ 1]));
8007 btrfs_free_tree_block(trans
, root
, eb
, parent
, wc
->refs
[level
] == 1);
8009 wc
->refs
[level
] = 0;
8010 wc
->flags
[level
] = 0;
8014 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
8015 struct btrfs_root
*root
,
8016 struct btrfs_path
*path
,
8017 struct walk_control
*wc
)
8019 int level
= wc
->level
;
8020 int lookup_info
= 1;
8023 while (level
>= 0) {
8024 ret
= walk_down_proc(trans
, root
, path
, wc
, lookup_info
);
8031 if (path
->slots
[level
] >=
8032 btrfs_header_nritems(path
->nodes
[level
]))
8035 ret
= do_walk_down(trans
, root
, path
, wc
, &lookup_info
);
8037 path
->slots
[level
]++;
8046 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
8047 struct btrfs_root
*root
,
8048 struct btrfs_path
*path
,
8049 struct walk_control
*wc
, int max_level
)
8051 int level
= wc
->level
;
8054 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
8055 while (level
< max_level
&& path
->nodes
[level
]) {
8057 if (path
->slots
[level
] + 1 <
8058 btrfs_header_nritems(path
->nodes
[level
])) {
8059 path
->slots
[level
]++;
8062 ret
= walk_up_proc(trans
, root
, path
, wc
);
8066 if (path
->locks
[level
]) {
8067 btrfs_tree_unlock_rw(path
->nodes
[level
],
8068 path
->locks
[level
]);
8069 path
->locks
[level
] = 0;
8071 free_extent_buffer(path
->nodes
[level
]);
8072 path
->nodes
[level
] = NULL
;
8080 * drop a subvolume tree.
8082 * this function traverses the tree freeing any blocks that only
8083 * referenced by the tree.
8085 * when a shared tree block is found. this function decreases its
8086 * reference count by one. if update_ref is true, this function
8087 * also make sure backrefs for the shared block and all lower level
8088 * blocks are properly updated.
8090 * If called with for_reloc == 0, may exit early with -EAGAIN
8092 int btrfs_drop_snapshot(struct btrfs_root
*root
,
8093 struct btrfs_block_rsv
*block_rsv
, int update_ref
,
8096 struct btrfs_path
*path
;
8097 struct btrfs_trans_handle
*trans
;
8098 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
8099 struct btrfs_root_item
*root_item
= &root
->root_item
;
8100 struct walk_control
*wc
;
8101 struct btrfs_key key
;
8105 bool root_dropped
= false;
8107 btrfs_debug(root
->fs_info
, "Drop subvolume %llu", root
->objectid
);
8109 path
= btrfs_alloc_path();
8115 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
8117 btrfs_free_path(path
);
8122 trans
= btrfs_start_transaction(tree_root
, 0);
8123 if (IS_ERR(trans
)) {
8124 err
= PTR_ERR(trans
);
8129 trans
->block_rsv
= block_rsv
;
8131 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
8132 level
= btrfs_header_level(root
->node
);
8133 path
->nodes
[level
] = btrfs_lock_root_node(root
);
8134 btrfs_set_lock_blocking(path
->nodes
[level
]);
8135 path
->slots
[level
] = 0;
8136 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8137 memset(&wc
->update_progress
, 0,
8138 sizeof(wc
->update_progress
));
8140 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
8141 memcpy(&wc
->update_progress
, &key
,
8142 sizeof(wc
->update_progress
));
8144 level
= root_item
->drop_level
;
8146 path
->lowest_level
= level
;
8147 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
8148 path
->lowest_level
= 0;
8156 * unlock our path, this is safe because only this
8157 * function is allowed to delete this snapshot
8159 btrfs_unlock_up_safe(path
, 0);
8161 level
= btrfs_header_level(root
->node
);
8163 btrfs_tree_lock(path
->nodes
[level
]);
8164 btrfs_set_lock_blocking(path
->nodes
[level
]);
8165 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8167 ret
= btrfs_lookup_extent_info(trans
, root
,
8168 path
->nodes
[level
]->start
,
8169 level
, 1, &wc
->refs
[level
],
8175 BUG_ON(wc
->refs
[level
] == 0);
8177 if (level
== root_item
->drop_level
)
8180 btrfs_tree_unlock(path
->nodes
[level
]);
8181 path
->locks
[level
] = 0;
8182 WARN_ON(wc
->refs
[level
] != 1);
8188 wc
->shared_level
= -1;
8189 wc
->stage
= DROP_REFERENCE
;
8190 wc
->update_ref
= update_ref
;
8192 wc
->for_reloc
= for_reloc
;
8193 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
8197 ret
= walk_down_tree(trans
, root
, path
, wc
);
8203 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
8210 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
8214 if (wc
->stage
== DROP_REFERENCE
) {
8216 btrfs_node_key(path
->nodes
[level
],
8217 &root_item
->drop_progress
,
8218 path
->slots
[level
]);
8219 root_item
->drop_level
= level
;
8222 BUG_ON(wc
->level
== 0);
8223 if (btrfs_should_end_transaction(trans
, tree_root
) ||
8224 (!for_reloc
&& btrfs_need_cleaner_sleep(root
))) {
8225 ret
= btrfs_update_root(trans
, tree_root
,
8229 btrfs_abort_transaction(trans
, tree_root
, ret
);
8235 * Qgroup update accounting is run from
8236 * delayed ref handling. This usually works
8237 * out because delayed refs are normally the
8238 * only way qgroup updates are added. However,
8239 * we may have added updates during our tree
8240 * walk so run qgroups here to make sure we
8241 * don't lose any updates.
8243 ret
= btrfs_delayed_qgroup_accounting(trans
,
8246 printk_ratelimited(KERN_ERR
"BTRFS: Failure %d "
8247 "running qgroup updates "
8248 "during snapshot delete. "
8249 "Quota is out of sync, "
8250 "rescan required.\n", ret
);
8252 btrfs_end_transaction_throttle(trans
, tree_root
);
8253 if (!for_reloc
&& btrfs_need_cleaner_sleep(root
)) {
8254 pr_debug("BTRFS: drop snapshot early exit\n");
8259 trans
= btrfs_start_transaction(tree_root
, 0);
8260 if (IS_ERR(trans
)) {
8261 err
= PTR_ERR(trans
);
8265 trans
->block_rsv
= block_rsv
;
8268 btrfs_release_path(path
);
8272 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
8274 btrfs_abort_transaction(trans
, tree_root
, ret
);
8278 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
8279 ret
= btrfs_find_root(tree_root
, &root
->root_key
, path
,
8282 btrfs_abort_transaction(trans
, tree_root
, ret
);
8285 } else if (ret
> 0) {
8286 /* if we fail to delete the orphan item this time
8287 * around, it'll get picked up the next time.
8289 * The most common failure here is just -ENOENT.
8291 btrfs_del_orphan_item(trans
, tree_root
,
8292 root
->root_key
.objectid
);
8296 if (test_bit(BTRFS_ROOT_IN_RADIX
, &root
->state
)) {
8297 btrfs_drop_and_free_fs_root(tree_root
->fs_info
, root
);
8299 free_extent_buffer(root
->node
);
8300 free_extent_buffer(root
->commit_root
);
8301 btrfs_put_fs_root(root
);
8303 root_dropped
= true;
8305 ret
= btrfs_delayed_qgroup_accounting(trans
, tree_root
->fs_info
);
8307 printk_ratelimited(KERN_ERR
"BTRFS: Failure %d "
8308 "running qgroup updates "
8309 "during snapshot delete. "
8310 "Quota is out of sync, "
8311 "rescan required.\n", ret
);
8313 btrfs_end_transaction_throttle(trans
, tree_root
);
8316 btrfs_free_path(path
);
8319 * So if we need to stop dropping the snapshot for whatever reason we
8320 * need to make sure to add it back to the dead root list so that we
8321 * keep trying to do the work later. This also cleans up roots if we
8322 * don't have it in the radix (like when we recover after a power fail
8323 * or unmount) so we don't leak memory.
8325 if (!for_reloc
&& root_dropped
== false)
8326 btrfs_add_dead_root(root
);
8327 if (err
&& err
!= -EAGAIN
)
8328 btrfs_std_error(root
->fs_info
, err
);
8333 * drop subtree rooted at tree block 'node'.
8335 * NOTE: this function will unlock and release tree block 'node'
8336 * only used by relocation code
8338 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
8339 struct btrfs_root
*root
,
8340 struct extent_buffer
*node
,
8341 struct extent_buffer
*parent
)
8343 struct btrfs_path
*path
;
8344 struct walk_control
*wc
;
8350 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
8352 path
= btrfs_alloc_path();
8356 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
8358 btrfs_free_path(path
);
8362 btrfs_assert_tree_locked(parent
);
8363 parent_level
= btrfs_header_level(parent
);
8364 extent_buffer_get(parent
);
8365 path
->nodes
[parent_level
] = parent
;
8366 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
8368 btrfs_assert_tree_locked(node
);
8369 level
= btrfs_header_level(node
);
8370 path
->nodes
[level
] = node
;
8371 path
->slots
[level
] = 0;
8372 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8374 wc
->refs
[parent_level
] = 1;
8375 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
8377 wc
->shared_level
= -1;
8378 wc
->stage
= DROP_REFERENCE
;
8382 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
8385 wret
= walk_down_tree(trans
, root
, path
, wc
);
8391 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
8399 btrfs_free_path(path
);
8403 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
8409 * if restripe for this chunk_type is on pick target profile and
8410 * return, otherwise do the usual balance
8412 stripped
= get_restripe_target(root
->fs_info
, flags
);
8414 return extended_to_chunk(stripped
);
8416 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
8418 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
8419 BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
|
8420 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
8422 if (num_devices
== 1) {
8423 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
8424 stripped
= flags
& ~stripped
;
8426 /* turn raid0 into single device chunks */
8427 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
8430 /* turn mirroring into duplication */
8431 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
8432 BTRFS_BLOCK_GROUP_RAID10
))
8433 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
8435 /* they already had raid on here, just return */
8436 if (flags
& stripped
)
8439 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
8440 stripped
= flags
& ~stripped
;
8442 /* switch duplicated blocks with raid1 */
8443 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
8444 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
8446 /* this is drive concat, leave it alone */
8452 static int set_block_group_ro(struct btrfs_block_group_cache
*cache
, int force
)
8454 struct btrfs_space_info
*sinfo
= cache
->space_info
;
8456 u64 min_allocable_bytes
;
8461 * We need some metadata space and system metadata space for
8462 * allocating chunks in some corner cases until we force to set
8463 * it to be readonly.
8466 (BTRFS_BLOCK_GROUP_SYSTEM
| BTRFS_BLOCK_GROUP_METADATA
)) &&
8468 min_allocable_bytes
= 1 * 1024 * 1024;
8470 min_allocable_bytes
= 0;
8472 spin_lock(&sinfo
->lock
);
8473 spin_lock(&cache
->lock
);
8480 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
8481 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
8483 if (sinfo
->bytes_used
+ sinfo
->bytes_reserved
+ sinfo
->bytes_pinned
+
8484 sinfo
->bytes_may_use
+ sinfo
->bytes_readonly
+ num_bytes
+
8485 min_allocable_bytes
<= sinfo
->total_bytes
) {
8486 sinfo
->bytes_readonly
+= num_bytes
;
8488 list_add_tail(&cache
->ro_list
, &sinfo
->ro_bgs
);
8492 spin_unlock(&cache
->lock
);
8493 spin_unlock(&sinfo
->lock
);
8497 int btrfs_set_block_group_ro(struct btrfs_root
*root
,
8498 struct btrfs_block_group_cache
*cache
)
8501 struct btrfs_trans_handle
*trans
;
8507 trans
= btrfs_join_transaction(root
);
8509 return PTR_ERR(trans
);
8511 ret
= set_block_group_ro(cache
, 0);
8514 alloc_flags
= get_alloc_profile(root
, cache
->space_info
->flags
);
8515 ret
= do_chunk_alloc(trans
, root
, alloc_flags
,
8519 ret
= set_block_group_ro(cache
, 0);
8521 if (cache
->flags
& BTRFS_BLOCK_GROUP_SYSTEM
) {
8522 alloc_flags
= update_block_group_flags(root
, cache
->flags
);
8523 check_system_chunk(trans
, root
, alloc_flags
);
8526 btrfs_end_transaction(trans
, root
);
8530 int btrfs_force_chunk_alloc(struct btrfs_trans_handle
*trans
,
8531 struct btrfs_root
*root
, u64 type
)
8533 u64 alloc_flags
= get_alloc_profile(root
, type
);
8534 return do_chunk_alloc(trans
, root
, alloc_flags
,
8539 * helper to account the unused space of all the readonly block group in the
8540 * space_info. takes mirrors into account.
8542 u64
btrfs_account_ro_block_groups_free_space(struct btrfs_space_info
*sinfo
)
8544 struct btrfs_block_group_cache
*block_group
;
8548 /* It's df, we don't care if it's racey */
8549 if (list_empty(&sinfo
->ro_bgs
))
8552 spin_lock(&sinfo
->lock
);
8553 list_for_each_entry(block_group
, &sinfo
->ro_bgs
, ro_list
) {
8554 spin_lock(&block_group
->lock
);
8556 if (!block_group
->ro
) {
8557 spin_unlock(&block_group
->lock
);
8561 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_RAID1
|
8562 BTRFS_BLOCK_GROUP_RAID10
|
8563 BTRFS_BLOCK_GROUP_DUP
))
8568 free_bytes
+= (block_group
->key
.offset
-
8569 btrfs_block_group_used(&block_group
->item
)) *
8572 spin_unlock(&block_group
->lock
);
8574 spin_unlock(&sinfo
->lock
);
8579 void btrfs_set_block_group_rw(struct btrfs_root
*root
,
8580 struct btrfs_block_group_cache
*cache
)
8582 struct btrfs_space_info
*sinfo
= cache
->space_info
;
8587 spin_lock(&sinfo
->lock
);
8588 spin_lock(&cache
->lock
);
8589 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
8590 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
8591 sinfo
->bytes_readonly
-= num_bytes
;
8593 list_del_init(&cache
->ro_list
);
8594 spin_unlock(&cache
->lock
);
8595 spin_unlock(&sinfo
->lock
);
8599 * checks to see if its even possible to relocate this block group.
8601 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8602 * ok to go ahead and try.
8604 int btrfs_can_relocate(struct btrfs_root
*root
, u64 bytenr
)
8606 struct btrfs_block_group_cache
*block_group
;
8607 struct btrfs_space_info
*space_info
;
8608 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
8609 struct btrfs_device
*device
;
8610 struct btrfs_trans_handle
*trans
;
8619 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
8621 /* odd, couldn't find the block group, leave it alone */
8625 min_free
= btrfs_block_group_used(&block_group
->item
);
8627 /* no bytes used, we're good */
8631 space_info
= block_group
->space_info
;
8632 spin_lock(&space_info
->lock
);
8634 full
= space_info
->full
;
8637 * if this is the last block group we have in this space, we can't
8638 * relocate it unless we're able to allocate a new chunk below.
8640 * Otherwise, we need to make sure we have room in the space to handle
8641 * all of the extents from this block group. If we can, we're good
8643 if ((space_info
->total_bytes
!= block_group
->key
.offset
) &&
8644 (space_info
->bytes_used
+ space_info
->bytes_reserved
+
8645 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
8646 min_free
< space_info
->total_bytes
)) {
8647 spin_unlock(&space_info
->lock
);
8650 spin_unlock(&space_info
->lock
);
8653 * ok we don't have enough space, but maybe we have free space on our
8654 * devices to allocate new chunks for relocation, so loop through our
8655 * alloc devices and guess if we have enough space. if this block
8656 * group is going to be restriped, run checks against the target
8657 * profile instead of the current one.
8669 target
= get_restripe_target(root
->fs_info
, block_group
->flags
);
8671 index
= __get_raid_index(extended_to_chunk(target
));
8674 * this is just a balance, so if we were marked as full
8675 * we know there is no space for a new chunk
8680 index
= get_block_group_index(block_group
);
8683 if (index
== BTRFS_RAID_RAID10
) {
8687 } else if (index
== BTRFS_RAID_RAID1
) {
8689 } else if (index
== BTRFS_RAID_DUP
) {
8692 } else if (index
== BTRFS_RAID_RAID0
) {
8693 dev_min
= fs_devices
->rw_devices
;
8694 do_div(min_free
, dev_min
);
8697 /* We need to do this so that we can look at pending chunks */
8698 trans
= btrfs_join_transaction(root
);
8699 if (IS_ERR(trans
)) {
8700 ret
= PTR_ERR(trans
);
8704 mutex_lock(&root
->fs_info
->chunk_mutex
);
8705 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
8709 * check to make sure we can actually find a chunk with enough
8710 * space to fit our block group in.
8712 if (device
->total_bytes
> device
->bytes_used
+ min_free
&&
8713 !device
->is_tgtdev_for_dev_replace
) {
8714 ret
= find_free_dev_extent(trans
, device
, min_free
,
8719 if (dev_nr
>= dev_min
)
8725 mutex_unlock(&root
->fs_info
->chunk_mutex
);
8726 btrfs_end_transaction(trans
, root
);
8728 btrfs_put_block_group(block_group
);
8732 static int find_first_block_group(struct btrfs_root
*root
,
8733 struct btrfs_path
*path
, struct btrfs_key
*key
)
8736 struct btrfs_key found_key
;
8737 struct extent_buffer
*leaf
;
8740 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
8745 slot
= path
->slots
[0];
8746 leaf
= path
->nodes
[0];
8747 if (slot
>= btrfs_header_nritems(leaf
)) {
8748 ret
= btrfs_next_leaf(root
, path
);
8755 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
8757 if (found_key
.objectid
>= key
->objectid
&&
8758 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
8768 void btrfs_put_block_group_cache(struct btrfs_fs_info
*info
)
8770 struct btrfs_block_group_cache
*block_group
;
8774 struct inode
*inode
;
8776 block_group
= btrfs_lookup_first_block_group(info
, last
);
8777 while (block_group
) {
8778 spin_lock(&block_group
->lock
);
8779 if (block_group
->iref
)
8781 spin_unlock(&block_group
->lock
);
8782 block_group
= next_block_group(info
->tree_root
,
8792 inode
= block_group
->inode
;
8793 block_group
->iref
= 0;
8794 block_group
->inode
= NULL
;
8795 spin_unlock(&block_group
->lock
);
8797 last
= block_group
->key
.objectid
+ block_group
->key
.offset
;
8798 btrfs_put_block_group(block_group
);
8802 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
8804 struct btrfs_block_group_cache
*block_group
;
8805 struct btrfs_space_info
*space_info
;
8806 struct btrfs_caching_control
*caching_ctl
;
8809 down_write(&info
->commit_root_sem
);
8810 while (!list_empty(&info
->caching_block_groups
)) {
8811 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
8812 struct btrfs_caching_control
, list
);
8813 list_del(&caching_ctl
->list
);
8814 put_caching_control(caching_ctl
);
8816 up_write(&info
->commit_root_sem
);
8818 spin_lock(&info
->unused_bgs_lock
);
8819 while (!list_empty(&info
->unused_bgs
)) {
8820 block_group
= list_first_entry(&info
->unused_bgs
,
8821 struct btrfs_block_group_cache
,
8823 list_del_init(&block_group
->bg_list
);
8824 btrfs_put_block_group(block_group
);
8826 spin_unlock(&info
->unused_bgs_lock
);
8828 spin_lock(&info
->block_group_cache_lock
);
8829 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
8830 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
8832 rb_erase(&block_group
->cache_node
,
8833 &info
->block_group_cache_tree
);
8834 RB_CLEAR_NODE(&block_group
->cache_node
);
8835 spin_unlock(&info
->block_group_cache_lock
);
8837 down_write(&block_group
->space_info
->groups_sem
);
8838 list_del(&block_group
->list
);
8839 up_write(&block_group
->space_info
->groups_sem
);
8841 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
8842 wait_block_group_cache_done(block_group
);
8845 * We haven't cached this block group, which means we could
8846 * possibly have excluded extents on this block group.
8848 if (block_group
->cached
== BTRFS_CACHE_NO
||
8849 block_group
->cached
== BTRFS_CACHE_ERROR
)
8850 free_excluded_extents(info
->extent_root
, block_group
);
8852 btrfs_remove_free_space_cache(block_group
);
8853 btrfs_put_block_group(block_group
);
8855 spin_lock(&info
->block_group_cache_lock
);
8857 spin_unlock(&info
->block_group_cache_lock
);
8859 /* now that all the block groups are freed, go through and
8860 * free all the space_info structs. This is only called during
8861 * the final stages of unmount, and so we know nobody is
8862 * using them. We call synchronize_rcu() once before we start,
8863 * just to be on the safe side.
8867 release_global_block_rsv(info
);
8869 while (!list_empty(&info
->space_info
)) {
8872 space_info
= list_entry(info
->space_info
.next
,
8873 struct btrfs_space_info
,
8875 if (btrfs_test_opt(info
->tree_root
, ENOSPC_DEBUG
)) {
8876 if (WARN_ON(space_info
->bytes_pinned
> 0 ||
8877 space_info
->bytes_reserved
> 0 ||
8878 space_info
->bytes_may_use
> 0)) {
8879 dump_space_info(space_info
, 0, 0);
8882 list_del(&space_info
->list
);
8883 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++) {
8884 struct kobject
*kobj
;
8885 kobj
= space_info
->block_group_kobjs
[i
];
8886 space_info
->block_group_kobjs
[i
] = NULL
;
8892 kobject_del(&space_info
->kobj
);
8893 kobject_put(&space_info
->kobj
);
8898 static void __link_block_group(struct btrfs_space_info
*space_info
,
8899 struct btrfs_block_group_cache
*cache
)
8901 int index
= get_block_group_index(cache
);
8904 down_write(&space_info
->groups_sem
);
8905 if (list_empty(&space_info
->block_groups
[index
]))
8907 list_add_tail(&cache
->list
, &space_info
->block_groups
[index
]);
8908 up_write(&space_info
->groups_sem
);
8911 struct raid_kobject
*rkobj
;
8914 rkobj
= kzalloc(sizeof(*rkobj
), GFP_NOFS
);
8917 rkobj
->raid_type
= index
;
8918 kobject_init(&rkobj
->kobj
, &btrfs_raid_ktype
);
8919 ret
= kobject_add(&rkobj
->kobj
, &space_info
->kobj
,
8920 "%s", get_raid_name(index
));
8922 kobject_put(&rkobj
->kobj
);
8925 space_info
->block_group_kobjs
[index
] = &rkobj
->kobj
;
8930 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
8933 static struct btrfs_block_group_cache
*
8934 btrfs_create_block_group_cache(struct btrfs_root
*root
, u64 start
, u64 size
)
8936 struct btrfs_block_group_cache
*cache
;
8938 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
8942 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
8944 if (!cache
->free_space_ctl
) {
8949 cache
->key
.objectid
= start
;
8950 cache
->key
.offset
= size
;
8951 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
8953 cache
->sectorsize
= root
->sectorsize
;
8954 cache
->fs_info
= root
->fs_info
;
8955 cache
->full_stripe_len
= btrfs_full_stripe_len(root
,
8956 &root
->fs_info
->mapping_tree
,
8958 atomic_set(&cache
->count
, 1);
8959 spin_lock_init(&cache
->lock
);
8960 init_rwsem(&cache
->data_rwsem
);
8961 INIT_LIST_HEAD(&cache
->list
);
8962 INIT_LIST_HEAD(&cache
->cluster_list
);
8963 INIT_LIST_HEAD(&cache
->bg_list
);
8964 INIT_LIST_HEAD(&cache
->ro_list
);
8965 INIT_LIST_HEAD(&cache
->dirty_list
);
8966 btrfs_init_free_space_ctl(cache
);
8967 atomic_set(&cache
->trimming
, 0);
8972 int btrfs_read_block_groups(struct btrfs_root
*root
)
8974 struct btrfs_path
*path
;
8976 struct btrfs_block_group_cache
*cache
;
8977 struct btrfs_fs_info
*info
= root
->fs_info
;
8978 struct btrfs_space_info
*space_info
;
8979 struct btrfs_key key
;
8980 struct btrfs_key found_key
;
8981 struct extent_buffer
*leaf
;
8985 root
= info
->extent_root
;
8988 key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
8989 path
= btrfs_alloc_path();
8994 cache_gen
= btrfs_super_cache_generation(root
->fs_info
->super_copy
);
8995 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
8996 btrfs_super_generation(root
->fs_info
->super_copy
) != cache_gen
)
8998 if (btrfs_test_opt(root
, CLEAR_CACHE
))
9002 ret
= find_first_block_group(root
, path
, &key
);
9008 leaf
= path
->nodes
[0];
9009 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
9011 cache
= btrfs_create_block_group_cache(root
, found_key
.objectid
,
9020 * When we mount with old space cache, we need to
9021 * set BTRFS_DC_CLEAR and set dirty flag.
9023 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9024 * truncate the old free space cache inode and
9026 * b) Setting 'dirty flag' makes sure that we flush
9027 * the new space cache info onto disk.
9029 if (btrfs_test_opt(root
, SPACE_CACHE
))
9030 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
9033 read_extent_buffer(leaf
, &cache
->item
,
9034 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
9035 sizeof(cache
->item
));
9036 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
9038 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
9039 btrfs_release_path(path
);
9042 * We need to exclude the super stripes now so that the space
9043 * info has super bytes accounted for, otherwise we'll think
9044 * we have more space than we actually do.
9046 ret
= exclude_super_stripes(root
, cache
);
9049 * We may have excluded something, so call this just in
9052 free_excluded_extents(root
, cache
);
9053 btrfs_put_block_group(cache
);
9058 * check for two cases, either we are full, and therefore
9059 * don't need to bother with the caching work since we won't
9060 * find any space, or we are empty, and we can just add all
9061 * the space in and be done with it. This saves us _alot_ of
9062 * time, particularly in the full case.
9064 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
9065 cache
->last_byte_to_unpin
= (u64
)-1;
9066 cache
->cached
= BTRFS_CACHE_FINISHED
;
9067 free_excluded_extents(root
, cache
);
9068 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
9069 cache
->last_byte_to_unpin
= (u64
)-1;
9070 cache
->cached
= BTRFS_CACHE_FINISHED
;
9071 add_new_free_space(cache
, root
->fs_info
,
9073 found_key
.objectid
+
9075 free_excluded_extents(root
, cache
);
9078 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
9080 btrfs_remove_free_space_cache(cache
);
9081 btrfs_put_block_group(cache
);
9085 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
9086 btrfs_block_group_used(&cache
->item
),
9089 btrfs_remove_free_space_cache(cache
);
9090 spin_lock(&info
->block_group_cache_lock
);
9091 rb_erase(&cache
->cache_node
,
9092 &info
->block_group_cache_tree
);
9093 RB_CLEAR_NODE(&cache
->cache_node
);
9094 spin_unlock(&info
->block_group_cache_lock
);
9095 btrfs_put_block_group(cache
);
9099 cache
->space_info
= space_info
;
9100 spin_lock(&cache
->space_info
->lock
);
9101 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
9102 spin_unlock(&cache
->space_info
->lock
);
9104 __link_block_group(space_info
, cache
);
9106 set_avail_alloc_bits(root
->fs_info
, cache
->flags
);
9107 if (btrfs_chunk_readonly(root
, cache
->key
.objectid
)) {
9108 set_block_group_ro(cache
, 1);
9109 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
9110 spin_lock(&info
->unused_bgs_lock
);
9111 /* Should always be true but just in case. */
9112 if (list_empty(&cache
->bg_list
)) {
9113 btrfs_get_block_group(cache
);
9114 list_add_tail(&cache
->bg_list
,
9117 spin_unlock(&info
->unused_bgs_lock
);
9121 list_for_each_entry_rcu(space_info
, &root
->fs_info
->space_info
, list
) {
9122 if (!(get_alloc_profile(root
, space_info
->flags
) &
9123 (BTRFS_BLOCK_GROUP_RAID10
|
9124 BTRFS_BLOCK_GROUP_RAID1
|
9125 BTRFS_BLOCK_GROUP_RAID5
|
9126 BTRFS_BLOCK_GROUP_RAID6
|
9127 BTRFS_BLOCK_GROUP_DUP
)))
9130 * avoid allocating from un-mirrored block group if there are
9131 * mirrored block groups.
9133 list_for_each_entry(cache
,
9134 &space_info
->block_groups
[BTRFS_RAID_RAID0
],
9136 set_block_group_ro(cache
, 1);
9137 list_for_each_entry(cache
,
9138 &space_info
->block_groups
[BTRFS_RAID_SINGLE
],
9140 set_block_group_ro(cache
, 1);
9143 init_global_block_rsv(info
);
9146 btrfs_free_path(path
);
9150 void btrfs_create_pending_block_groups(struct btrfs_trans_handle
*trans
,
9151 struct btrfs_root
*root
)
9153 struct btrfs_block_group_cache
*block_group
, *tmp
;
9154 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
9155 struct btrfs_block_group_item item
;
9156 struct btrfs_key key
;
9159 list_for_each_entry_safe(block_group
, tmp
, &trans
->new_bgs
, bg_list
) {
9163 spin_lock(&block_group
->lock
);
9164 memcpy(&item
, &block_group
->item
, sizeof(item
));
9165 memcpy(&key
, &block_group
->key
, sizeof(key
));
9166 spin_unlock(&block_group
->lock
);
9168 ret
= btrfs_insert_item(trans
, extent_root
, &key
, &item
,
9171 btrfs_abort_transaction(trans
, extent_root
, ret
);
9172 ret
= btrfs_finish_chunk_alloc(trans
, extent_root
,
9173 key
.objectid
, key
.offset
);
9175 btrfs_abort_transaction(trans
, extent_root
, ret
);
9177 list_del_init(&block_group
->bg_list
);
9181 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
9182 struct btrfs_root
*root
, u64 bytes_used
,
9183 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
9187 struct btrfs_root
*extent_root
;
9188 struct btrfs_block_group_cache
*cache
;
9190 extent_root
= root
->fs_info
->extent_root
;
9192 btrfs_set_log_full_commit(root
->fs_info
, trans
);
9194 cache
= btrfs_create_block_group_cache(root
, chunk_offset
, size
);
9198 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
9199 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
9200 btrfs_set_block_group_flags(&cache
->item
, type
);
9202 cache
->flags
= type
;
9203 cache
->last_byte_to_unpin
= (u64
)-1;
9204 cache
->cached
= BTRFS_CACHE_FINISHED
;
9205 ret
= exclude_super_stripes(root
, cache
);
9208 * We may have excluded something, so call this just in
9211 free_excluded_extents(root
, cache
);
9212 btrfs_put_block_group(cache
);
9216 add_new_free_space(cache
, root
->fs_info
, chunk_offset
,
9217 chunk_offset
+ size
);
9219 free_excluded_extents(root
, cache
);
9221 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
9223 btrfs_remove_free_space_cache(cache
);
9224 btrfs_put_block_group(cache
);
9228 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
9229 &cache
->space_info
);
9231 btrfs_remove_free_space_cache(cache
);
9232 spin_lock(&root
->fs_info
->block_group_cache_lock
);
9233 rb_erase(&cache
->cache_node
,
9234 &root
->fs_info
->block_group_cache_tree
);
9235 RB_CLEAR_NODE(&cache
->cache_node
);
9236 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
9237 btrfs_put_block_group(cache
);
9240 update_global_block_rsv(root
->fs_info
);
9242 spin_lock(&cache
->space_info
->lock
);
9243 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
9244 spin_unlock(&cache
->space_info
->lock
);
9246 __link_block_group(cache
->space_info
, cache
);
9248 list_add_tail(&cache
->bg_list
, &trans
->new_bgs
);
9250 set_avail_alloc_bits(extent_root
->fs_info
, type
);
9255 static void clear_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
9257 u64 extra_flags
= chunk_to_extended(flags
) &
9258 BTRFS_EXTENDED_PROFILE_MASK
;
9260 write_seqlock(&fs_info
->profiles_lock
);
9261 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
9262 fs_info
->avail_data_alloc_bits
&= ~extra_flags
;
9263 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
9264 fs_info
->avail_metadata_alloc_bits
&= ~extra_flags
;
9265 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
9266 fs_info
->avail_system_alloc_bits
&= ~extra_flags
;
9267 write_sequnlock(&fs_info
->profiles_lock
);
9270 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
9271 struct btrfs_root
*root
, u64 group_start
,
9272 struct extent_map
*em
)
9274 struct btrfs_path
*path
;
9275 struct btrfs_block_group_cache
*block_group
;
9276 struct btrfs_free_cluster
*cluster
;
9277 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
9278 struct btrfs_key key
;
9279 struct inode
*inode
;
9280 struct kobject
*kobj
= NULL
;
9284 struct btrfs_caching_control
*caching_ctl
= NULL
;
9287 root
= root
->fs_info
->extent_root
;
9289 block_group
= btrfs_lookup_block_group(root
->fs_info
, group_start
);
9290 BUG_ON(!block_group
);
9291 BUG_ON(!block_group
->ro
);
9294 * Free the reserved super bytes from this block group before
9297 free_excluded_extents(root
, block_group
);
9299 memcpy(&key
, &block_group
->key
, sizeof(key
));
9300 index
= get_block_group_index(block_group
);
9301 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
9302 BTRFS_BLOCK_GROUP_RAID1
|
9303 BTRFS_BLOCK_GROUP_RAID10
))
9308 /* make sure this block group isn't part of an allocation cluster */
9309 cluster
= &root
->fs_info
->data_alloc_cluster
;
9310 spin_lock(&cluster
->refill_lock
);
9311 btrfs_return_cluster_to_free_space(block_group
, cluster
);
9312 spin_unlock(&cluster
->refill_lock
);
9315 * make sure this block group isn't part of a metadata
9316 * allocation cluster
9318 cluster
= &root
->fs_info
->meta_alloc_cluster
;
9319 spin_lock(&cluster
->refill_lock
);
9320 btrfs_return_cluster_to_free_space(block_group
, cluster
);
9321 spin_unlock(&cluster
->refill_lock
);
9323 path
= btrfs_alloc_path();
9329 inode
= lookup_free_space_inode(tree_root
, block_group
, path
);
9330 if (!IS_ERR(inode
)) {
9331 ret
= btrfs_orphan_add(trans
, inode
);
9333 btrfs_add_delayed_iput(inode
);
9337 /* One for the block groups ref */
9338 spin_lock(&block_group
->lock
);
9339 if (block_group
->iref
) {
9340 block_group
->iref
= 0;
9341 block_group
->inode
= NULL
;
9342 spin_unlock(&block_group
->lock
);
9345 spin_unlock(&block_group
->lock
);
9347 /* One for our lookup ref */
9348 btrfs_add_delayed_iput(inode
);
9351 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
9352 key
.offset
= block_group
->key
.objectid
;
9355 ret
= btrfs_search_slot(trans
, tree_root
, &key
, path
, -1, 1);
9359 btrfs_release_path(path
);
9361 ret
= btrfs_del_item(trans
, tree_root
, path
);
9364 btrfs_release_path(path
);
9367 spin_lock(&root
->fs_info
->block_group_cache_lock
);
9368 rb_erase(&block_group
->cache_node
,
9369 &root
->fs_info
->block_group_cache_tree
);
9370 RB_CLEAR_NODE(&block_group
->cache_node
);
9372 if (root
->fs_info
->first_logical_byte
== block_group
->key
.objectid
)
9373 root
->fs_info
->first_logical_byte
= (u64
)-1;
9374 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
9376 down_write(&block_group
->space_info
->groups_sem
);
9378 * we must use list_del_init so people can check to see if they
9379 * are still on the list after taking the semaphore
9381 list_del_init(&block_group
->list
);
9382 list_del_init(&block_group
->ro_list
);
9383 if (list_empty(&block_group
->space_info
->block_groups
[index
])) {
9384 kobj
= block_group
->space_info
->block_group_kobjs
[index
];
9385 block_group
->space_info
->block_group_kobjs
[index
] = NULL
;
9386 clear_avail_alloc_bits(root
->fs_info
, block_group
->flags
);
9388 up_write(&block_group
->space_info
->groups_sem
);
9394 if (block_group
->has_caching_ctl
)
9395 caching_ctl
= get_caching_control(block_group
);
9396 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
9397 wait_block_group_cache_done(block_group
);
9398 if (block_group
->has_caching_ctl
) {
9399 down_write(&root
->fs_info
->commit_root_sem
);
9401 struct btrfs_caching_control
*ctl
;
9403 list_for_each_entry(ctl
,
9404 &root
->fs_info
->caching_block_groups
, list
)
9405 if (ctl
->block_group
== block_group
) {
9407 atomic_inc(&caching_ctl
->count
);
9412 list_del_init(&caching_ctl
->list
);
9413 up_write(&root
->fs_info
->commit_root_sem
);
9415 /* Once for the caching bgs list and once for us. */
9416 put_caching_control(caching_ctl
);
9417 put_caching_control(caching_ctl
);
9421 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
9422 if (!list_empty(&block_group
->dirty_list
)) {
9423 list_del_init(&block_group
->dirty_list
);
9424 btrfs_put_block_group(block_group
);
9426 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
9428 btrfs_remove_free_space_cache(block_group
);
9430 spin_lock(&block_group
->space_info
->lock
);
9431 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
9432 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
9433 block_group
->space_info
->disk_total
-= block_group
->key
.offset
* factor
;
9434 spin_unlock(&block_group
->space_info
->lock
);
9436 memcpy(&key
, &block_group
->key
, sizeof(key
));
9439 if (!list_empty(&em
->list
)) {
9440 /* We're in the transaction->pending_chunks list. */
9441 free_extent_map(em
);
9443 spin_lock(&block_group
->lock
);
9444 block_group
->removed
= 1;
9446 * At this point trimming can't start on this block group, because we
9447 * removed the block group from the tree fs_info->block_group_cache_tree
9448 * so no one can't find it anymore and even if someone already got this
9449 * block group before we removed it from the rbtree, they have already
9450 * incremented block_group->trimming - if they didn't, they won't find
9451 * any free space entries because we already removed them all when we
9452 * called btrfs_remove_free_space_cache().
9454 * And we must not remove the extent map from the fs_info->mapping_tree
9455 * to prevent the same logical address range and physical device space
9456 * ranges from being reused for a new block group. This is because our
9457 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
9458 * completely transactionless, so while it is trimming a range the
9459 * currently running transaction might finish and a new one start,
9460 * allowing for new block groups to be created that can reuse the same
9461 * physical device locations unless we take this special care.
9463 remove_em
= (atomic_read(&block_group
->trimming
) == 0);
9465 * Make sure a trimmer task always sees the em in the pinned_chunks list
9466 * if it sees block_group->removed == 1 (needs to lock block_group->lock
9467 * before checking block_group->removed).
9471 * Our em might be in trans->transaction->pending_chunks which
9472 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
9473 * and so is the fs_info->pinned_chunks list.
9475 * So at this point we must be holding the chunk_mutex to avoid
9476 * any races with chunk allocation (more specifically at
9477 * volumes.c:contains_pending_extent()), to ensure it always
9478 * sees the em, either in the pending_chunks list or in the
9479 * pinned_chunks list.
9481 list_move_tail(&em
->list
, &root
->fs_info
->pinned_chunks
);
9483 spin_unlock(&block_group
->lock
);
9486 struct extent_map_tree
*em_tree
;
9488 em_tree
= &root
->fs_info
->mapping_tree
.map_tree
;
9489 write_lock(&em_tree
->lock
);
9491 * The em might be in the pending_chunks list, so make sure the
9492 * chunk mutex is locked, since remove_extent_mapping() will
9493 * delete us from that list.
9495 remove_extent_mapping(em_tree
, em
);
9496 write_unlock(&em_tree
->lock
);
9497 /* once for the tree */
9498 free_extent_map(em
);
9501 unlock_chunks(root
);
9503 btrfs_put_block_group(block_group
);
9504 btrfs_put_block_group(block_group
);
9506 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
9512 ret
= btrfs_del_item(trans
, root
, path
);
9514 btrfs_free_path(path
);
9519 * Process the unused_bgs list and remove any that don't have any allocated
9520 * space inside of them.
9522 void btrfs_delete_unused_bgs(struct btrfs_fs_info
*fs_info
)
9524 struct btrfs_block_group_cache
*block_group
;
9525 struct btrfs_space_info
*space_info
;
9526 struct btrfs_root
*root
= fs_info
->extent_root
;
9527 struct btrfs_trans_handle
*trans
;
9533 spin_lock(&fs_info
->unused_bgs_lock
);
9534 while (!list_empty(&fs_info
->unused_bgs
)) {
9537 block_group
= list_first_entry(&fs_info
->unused_bgs
,
9538 struct btrfs_block_group_cache
,
9540 space_info
= block_group
->space_info
;
9541 list_del_init(&block_group
->bg_list
);
9542 if (ret
|| btrfs_mixed_space_info(space_info
)) {
9543 btrfs_put_block_group(block_group
);
9546 spin_unlock(&fs_info
->unused_bgs_lock
);
9548 /* Don't want to race with allocators so take the groups_sem */
9549 down_write(&space_info
->groups_sem
);
9550 spin_lock(&block_group
->lock
);
9551 if (block_group
->reserved
||
9552 btrfs_block_group_used(&block_group
->item
) ||
9555 * We want to bail if we made new allocations or have
9556 * outstanding allocations in this block group. We do
9557 * the ro check in case balance is currently acting on
9560 spin_unlock(&block_group
->lock
);
9561 up_write(&space_info
->groups_sem
);
9564 spin_unlock(&block_group
->lock
);
9566 /* We don't want to force the issue, only flip if it's ok. */
9567 ret
= set_block_group_ro(block_group
, 0);
9568 up_write(&space_info
->groups_sem
);
9575 * Want to do this before we do anything else so we can recover
9576 * properly if we fail to join the transaction.
9578 /* 1 for btrfs_orphan_reserve_metadata() */
9579 trans
= btrfs_start_transaction(root
, 1);
9580 if (IS_ERR(trans
)) {
9581 btrfs_set_block_group_rw(root
, block_group
);
9582 ret
= PTR_ERR(trans
);
9587 * We could have pending pinned extents for this block group,
9588 * just delete them, we don't care about them anymore.
9590 start
= block_group
->key
.objectid
;
9591 end
= start
+ block_group
->key
.offset
- 1;
9593 * Hold the unused_bg_unpin_mutex lock to avoid racing with
9594 * btrfs_finish_extent_commit(). If we are at transaction N,
9595 * another task might be running finish_extent_commit() for the
9596 * previous transaction N - 1, and have seen a range belonging
9597 * to the block group in freed_extents[] before we were able to
9598 * clear the whole block group range from freed_extents[]. This
9599 * means that task can lookup for the block group after we
9600 * unpinned it from freed_extents[] and removed it, leading to
9601 * a BUG_ON() at btrfs_unpin_extent_range().
9603 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
9604 ret
= clear_extent_bits(&fs_info
->freed_extents
[0], start
, end
,
9605 EXTENT_DIRTY
, GFP_NOFS
);
9607 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
9608 btrfs_set_block_group_rw(root
, block_group
);
9611 ret
= clear_extent_bits(&fs_info
->freed_extents
[1], start
, end
,
9612 EXTENT_DIRTY
, GFP_NOFS
);
9614 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
9615 btrfs_set_block_group_rw(root
, block_group
);
9618 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
9620 /* Reset pinned so btrfs_put_block_group doesn't complain */
9621 block_group
->pinned
= 0;
9624 * Btrfs_remove_chunk will abort the transaction if things go
9627 ret
= btrfs_remove_chunk(trans
, root
,
9628 block_group
->key
.objectid
);
9630 btrfs_end_transaction(trans
, root
);
9632 btrfs_put_block_group(block_group
);
9633 spin_lock(&fs_info
->unused_bgs_lock
);
9635 spin_unlock(&fs_info
->unused_bgs_lock
);
9638 int btrfs_init_space_info(struct btrfs_fs_info
*fs_info
)
9640 struct btrfs_space_info
*space_info
;
9641 struct btrfs_super_block
*disk_super
;
9647 disk_super
= fs_info
->super_copy
;
9648 if (!btrfs_super_root(disk_super
))
9651 features
= btrfs_super_incompat_flags(disk_super
);
9652 if (features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
9655 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
9656 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
9661 flags
= BTRFS_BLOCK_GROUP_METADATA
| BTRFS_BLOCK_GROUP_DATA
;
9662 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
9664 flags
= BTRFS_BLOCK_GROUP_METADATA
;
9665 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
9669 flags
= BTRFS_BLOCK_GROUP_DATA
;
9670 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
9676 int btrfs_error_unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
9678 return unpin_extent_range(root
, start
, end
, false);
9681 int btrfs_trim_fs(struct btrfs_root
*root
, struct fstrim_range
*range
)
9683 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
9684 struct btrfs_block_group_cache
*cache
= NULL
;
9689 u64 total_bytes
= btrfs_super_total_bytes(fs_info
->super_copy
);
9693 * try to trim all FS space, our block group may start from non-zero.
9695 if (range
->len
== total_bytes
)
9696 cache
= btrfs_lookup_first_block_group(fs_info
, range
->start
);
9698 cache
= btrfs_lookup_block_group(fs_info
, range
->start
);
9701 if (cache
->key
.objectid
>= (range
->start
+ range
->len
)) {
9702 btrfs_put_block_group(cache
);
9706 start
= max(range
->start
, cache
->key
.objectid
);
9707 end
= min(range
->start
+ range
->len
,
9708 cache
->key
.objectid
+ cache
->key
.offset
);
9710 if (end
- start
>= range
->minlen
) {
9711 if (!block_group_cache_done(cache
)) {
9712 ret
= cache_block_group(cache
, 0);
9714 btrfs_put_block_group(cache
);
9717 ret
= wait_block_group_cache_done(cache
);
9719 btrfs_put_block_group(cache
);
9723 ret
= btrfs_trim_block_group(cache
,
9729 trimmed
+= group_trimmed
;
9731 btrfs_put_block_group(cache
);
9736 cache
= next_block_group(fs_info
->tree_root
, cache
);
9739 range
->len
= trimmed
;
9744 * btrfs_{start,end}_write_no_snapshoting() are similar to
9745 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
9746 * data into the page cache through nocow before the subvolume is snapshoted,
9747 * but flush the data into disk after the snapshot creation, or to prevent
9748 * operations while snapshoting is ongoing and that cause the snapshot to be
9749 * inconsistent (writes followed by expanding truncates for example).
9751 void btrfs_end_write_no_snapshoting(struct btrfs_root
*root
)
9753 percpu_counter_dec(&root
->subv_writers
->counter
);
9755 * Make sure counter is updated before we wake up
9759 if (waitqueue_active(&root
->subv_writers
->wait
))
9760 wake_up(&root
->subv_writers
->wait
);
9763 int btrfs_start_write_no_snapshoting(struct btrfs_root
*root
)
9765 if (atomic_read(&root
->will_be_snapshoted
))
9768 percpu_counter_inc(&root
->subv_writers
->counter
);
9770 * Make sure counter is updated before we check for snapshot creation.
9773 if (atomic_read(&root
->will_be_snapshoted
)) {
9774 btrfs_end_write_no_snapshoting(root
);