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
);
3212 inode
= lookup_free_space_inode(root
, block_group
, path
);
3213 if (IS_ERR(inode
) && PTR_ERR(inode
) != -ENOENT
) {
3214 ret
= PTR_ERR(inode
);
3215 btrfs_release_path(path
);
3219 if (IS_ERR(inode
)) {
3223 if (block_group
->ro
)
3226 ret
= create_free_space_inode(root
, trans
, block_group
, path
);
3232 /* We've already setup this transaction, go ahead and exit */
3233 if (block_group
->cache_generation
== trans
->transid
&&
3234 i_size_read(inode
)) {
3235 dcs
= BTRFS_DC_SETUP
;
3240 * We want to set the generation to 0, that way if anything goes wrong
3241 * from here on out we know not to trust this cache when we load up next
3244 BTRFS_I(inode
)->generation
= 0;
3245 ret
= btrfs_update_inode(trans
, root
, inode
);
3248 if (i_size_read(inode
) > 0) {
3249 ret
= btrfs_check_trunc_cache_free_space(root
,
3250 &root
->fs_info
->global_block_rsv
);
3254 ret
= btrfs_truncate_free_space_cache(root
, trans
, inode
);
3259 spin_lock(&block_group
->lock
);
3260 if (block_group
->cached
!= BTRFS_CACHE_FINISHED
||
3261 !btrfs_test_opt(root
, SPACE_CACHE
) ||
3262 block_group
->delalloc_bytes
) {
3264 * don't bother trying to write stuff out _if_
3265 * a) we're not cached,
3266 * b) we're with nospace_cache mount option.
3268 dcs
= BTRFS_DC_WRITTEN
;
3269 spin_unlock(&block_group
->lock
);
3272 spin_unlock(&block_group
->lock
);
3275 * Try to preallocate enough space based on how big the block group is.
3276 * Keep in mind this has to include any pinned space which could end up
3277 * taking up quite a bit since it's not folded into the other space
3280 num_pages
= (int)div64_u64(block_group
->key
.offset
, 256 * 1024 * 1024);
3285 num_pages
*= PAGE_CACHE_SIZE
;
3287 ret
= btrfs_check_data_free_space(inode
, num_pages
);
3291 ret
= btrfs_prealloc_file_range_trans(inode
, trans
, 0, 0, num_pages
,
3292 num_pages
, num_pages
,
3295 dcs
= BTRFS_DC_SETUP
;
3296 btrfs_free_reserved_data_space(inode
, num_pages
);
3301 btrfs_release_path(path
);
3303 spin_lock(&block_group
->lock
);
3304 if (!ret
&& dcs
== BTRFS_DC_SETUP
)
3305 block_group
->cache_generation
= trans
->transid
;
3306 block_group
->disk_cache_state
= dcs
;
3307 spin_unlock(&block_group
->lock
);
3312 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
3313 struct btrfs_root
*root
)
3315 struct btrfs_block_group_cache
*cache
;
3316 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3318 struct btrfs_path
*path
;
3320 if (list_empty(&cur_trans
->dirty_bgs
))
3323 path
= btrfs_alloc_path();
3328 * We don't need the lock here since we are protected by the transaction
3329 * commit. We want to do the cache_save_setup first and then run the
3330 * delayed refs to make sure we have the best chance at doing this all
3333 while (!list_empty(&cur_trans
->dirty_bgs
)) {
3334 cache
= list_first_entry(&cur_trans
->dirty_bgs
,
3335 struct btrfs_block_group_cache
,
3337 list_del_init(&cache
->dirty_list
);
3338 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
)
3339 cache_save_setup(cache
, trans
, path
);
3341 ret
= btrfs_run_delayed_refs(trans
, root
,
3342 (unsigned long) -1);
3343 if (!ret
&& cache
->disk_cache_state
== BTRFS_DC_SETUP
)
3344 btrfs_write_out_cache(root
, trans
, cache
, path
);
3346 ret
= write_one_cache_group(trans
, root
, path
, cache
);
3347 btrfs_put_block_group(cache
);
3350 btrfs_free_path(path
);
3354 int btrfs_extent_readonly(struct btrfs_root
*root
, u64 bytenr
)
3356 struct btrfs_block_group_cache
*block_group
;
3359 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
3360 if (!block_group
|| block_group
->ro
)
3363 btrfs_put_block_group(block_group
);
3367 static const char *alloc_name(u64 flags
)
3370 case BTRFS_BLOCK_GROUP_METADATA
|BTRFS_BLOCK_GROUP_DATA
:
3372 case BTRFS_BLOCK_GROUP_METADATA
:
3374 case BTRFS_BLOCK_GROUP_DATA
:
3376 case BTRFS_BLOCK_GROUP_SYSTEM
:
3380 return "invalid-combination";
3384 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
3385 u64 total_bytes
, u64 bytes_used
,
3386 struct btrfs_space_info
**space_info
)
3388 struct btrfs_space_info
*found
;
3393 if (flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3394 BTRFS_BLOCK_GROUP_RAID10
))
3399 found
= __find_space_info(info
, flags
);
3401 spin_lock(&found
->lock
);
3402 found
->total_bytes
+= total_bytes
;
3403 found
->disk_total
+= total_bytes
* factor
;
3404 found
->bytes_used
+= bytes_used
;
3405 found
->disk_used
+= bytes_used
* factor
;
3407 spin_unlock(&found
->lock
);
3408 *space_info
= found
;
3411 found
= kzalloc(sizeof(*found
), GFP_NOFS
);
3415 ret
= percpu_counter_init(&found
->total_bytes_pinned
, 0, GFP_KERNEL
);
3421 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
3422 INIT_LIST_HEAD(&found
->block_groups
[i
]);
3423 init_rwsem(&found
->groups_sem
);
3424 spin_lock_init(&found
->lock
);
3425 found
->flags
= flags
& BTRFS_BLOCK_GROUP_TYPE_MASK
;
3426 found
->total_bytes
= total_bytes
;
3427 found
->disk_total
= total_bytes
* factor
;
3428 found
->bytes_used
= bytes_used
;
3429 found
->disk_used
= bytes_used
* factor
;
3430 found
->bytes_pinned
= 0;
3431 found
->bytes_reserved
= 0;
3432 found
->bytes_readonly
= 0;
3433 found
->bytes_may_use
= 0;
3435 found
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3436 found
->chunk_alloc
= 0;
3438 init_waitqueue_head(&found
->wait
);
3439 INIT_LIST_HEAD(&found
->ro_bgs
);
3441 ret
= kobject_init_and_add(&found
->kobj
, &space_info_ktype
,
3442 info
->space_info_kobj
, "%s",
3443 alloc_name(found
->flags
));
3449 *space_info
= found
;
3450 list_add_rcu(&found
->list
, &info
->space_info
);
3451 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3452 info
->data_sinfo
= found
;
3457 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
3459 u64 extra_flags
= chunk_to_extended(flags
) &
3460 BTRFS_EXTENDED_PROFILE_MASK
;
3462 write_seqlock(&fs_info
->profiles_lock
);
3463 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3464 fs_info
->avail_data_alloc_bits
|= extra_flags
;
3465 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3466 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
3467 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3468 fs_info
->avail_system_alloc_bits
|= extra_flags
;
3469 write_sequnlock(&fs_info
->profiles_lock
);
3473 * returns target flags in extended format or 0 if restripe for this
3474 * chunk_type is not in progress
3476 * should be called with either volume_mutex or balance_lock held
3478 static u64
get_restripe_target(struct btrfs_fs_info
*fs_info
, u64 flags
)
3480 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3486 if (flags
& BTRFS_BLOCK_GROUP_DATA
&&
3487 bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3488 target
= BTRFS_BLOCK_GROUP_DATA
| bctl
->data
.target
;
3489 } else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
&&
3490 bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3491 target
= BTRFS_BLOCK_GROUP_SYSTEM
| bctl
->sys
.target
;
3492 } else if (flags
& BTRFS_BLOCK_GROUP_METADATA
&&
3493 bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3494 target
= BTRFS_BLOCK_GROUP_METADATA
| bctl
->meta
.target
;
3501 * @flags: available profiles in extended format (see ctree.h)
3503 * Returns reduced profile in chunk format. If profile changing is in
3504 * progress (either running or paused) picks the target profile (if it's
3505 * already available), otherwise falls back to plain reducing.
3507 static u64
btrfs_reduce_alloc_profile(struct btrfs_root
*root
, u64 flags
)
3509 u64 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
3514 * see if restripe for this chunk_type is in progress, if so
3515 * try to reduce to the target profile
3517 spin_lock(&root
->fs_info
->balance_lock
);
3518 target
= get_restripe_target(root
->fs_info
, flags
);
3520 /* pick target profile only if it's already available */
3521 if ((flags
& target
) & BTRFS_EXTENDED_PROFILE_MASK
) {
3522 spin_unlock(&root
->fs_info
->balance_lock
);
3523 return extended_to_chunk(target
);
3526 spin_unlock(&root
->fs_info
->balance_lock
);
3528 /* First, mask out the RAID levels which aren't possible */
3529 if (num_devices
== 1)
3530 flags
&= ~(BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID0
|
3531 BTRFS_BLOCK_GROUP_RAID5
);
3532 if (num_devices
< 3)
3533 flags
&= ~BTRFS_BLOCK_GROUP_RAID6
;
3534 if (num_devices
< 4)
3535 flags
&= ~BTRFS_BLOCK_GROUP_RAID10
;
3537 tmp
= flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID0
|
3538 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID5
|
3539 BTRFS_BLOCK_GROUP_RAID6
| BTRFS_BLOCK_GROUP_RAID10
);
3542 if (tmp
& BTRFS_BLOCK_GROUP_RAID6
)
3543 tmp
= BTRFS_BLOCK_GROUP_RAID6
;
3544 else if (tmp
& BTRFS_BLOCK_GROUP_RAID5
)
3545 tmp
= BTRFS_BLOCK_GROUP_RAID5
;
3546 else if (tmp
& BTRFS_BLOCK_GROUP_RAID10
)
3547 tmp
= BTRFS_BLOCK_GROUP_RAID10
;
3548 else if (tmp
& BTRFS_BLOCK_GROUP_RAID1
)
3549 tmp
= BTRFS_BLOCK_GROUP_RAID1
;
3550 else if (tmp
& BTRFS_BLOCK_GROUP_RAID0
)
3551 tmp
= BTRFS_BLOCK_GROUP_RAID0
;
3553 return extended_to_chunk(flags
| tmp
);
3556 static u64
get_alloc_profile(struct btrfs_root
*root
, u64 orig_flags
)
3563 seq
= read_seqbegin(&root
->fs_info
->profiles_lock
);
3565 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3566 flags
|= root
->fs_info
->avail_data_alloc_bits
;
3567 else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3568 flags
|= root
->fs_info
->avail_system_alloc_bits
;
3569 else if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3570 flags
|= root
->fs_info
->avail_metadata_alloc_bits
;
3571 } while (read_seqretry(&root
->fs_info
->profiles_lock
, seq
));
3573 return btrfs_reduce_alloc_profile(root
, flags
);
3576 u64
btrfs_get_alloc_profile(struct btrfs_root
*root
, int data
)
3582 flags
= BTRFS_BLOCK_GROUP_DATA
;
3583 else if (root
== root
->fs_info
->chunk_root
)
3584 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
3586 flags
= BTRFS_BLOCK_GROUP_METADATA
;
3588 ret
= get_alloc_profile(root
, flags
);
3593 * This will check the space that the inode allocates from to make sure we have
3594 * enough space for bytes.
3596 int btrfs_check_data_free_space(struct inode
*inode
, u64 bytes
)
3598 struct btrfs_space_info
*data_sinfo
;
3599 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3600 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3602 int ret
= 0, committed
= 0, alloc_chunk
= 1;
3604 /* make sure bytes are sectorsize aligned */
3605 bytes
= ALIGN(bytes
, root
->sectorsize
);
3607 if (btrfs_is_free_space_inode(inode
)) {
3609 ASSERT(current
->journal_info
);
3612 data_sinfo
= fs_info
->data_sinfo
;
3617 /* make sure we have enough space to handle the data first */
3618 spin_lock(&data_sinfo
->lock
);
3619 used
= data_sinfo
->bytes_used
+ data_sinfo
->bytes_reserved
+
3620 data_sinfo
->bytes_pinned
+ data_sinfo
->bytes_readonly
+
3621 data_sinfo
->bytes_may_use
;
3623 if (used
+ bytes
> data_sinfo
->total_bytes
) {
3624 struct btrfs_trans_handle
*trans
;
3627 * if we don't have enough free bytes in this space then we need
3628 * to alloc a new chunk.
3630 if (!data_sinfo
->full
&& alloc_chunk
) {
3633 data_sinfo
->force_alloc
= CHUNK_ALLOC_FORCE
;
3634 spin_unlock(&data_sinfo
->lock
);
3636 alloc_target
= btrfs_get_alloc_profile(root
, 1);
3638 * It is ugly that we don't call nolock join
3639 * transaction for the free space inode case here.
3640 * But it is safe because we only do the data space
3641 * reservation for the free space cache in the
3642 * transaction context, the common join transaction
3643 * just increase the counter of the current transaction
3644 * handler, doesn't try to acquire the trans_lock of
3647 trans
= btrfs_join_transaction(root
);
3649 return PTR_ERR(trans
);
3651 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
3653 CHUNK_ALLOC_NO_FORCE
);
3654 btrfs_end_transaction(trans
, root
);
3663 data_sinfo
= fs_info
->data_sinfo
;
3669 * If we don't have enough pinned space to deal with this
3670 * allocation don't bother committing the transaction.
3672 if (percpu_counter_compare(&data_sinfo
->total_bytes_pinned
,
3675 spin_unlock(&data_sinfo
->lock
);
3677 /* commit the current transaction and try again */
3680 !atomic_read(&root
->fs_info
->open_ioctl_trans
)) {
3683 trans
= btrfs_join_transaction(root
);
3685 return PTR_ERR(trans
);
3686 ret
= btrfs_commit_transaction(trans
, root
);
3692 trace_btrfs_space_reservation(root
->fs_info
,
3693 "space_info:enospc",
3694 data_sinfo
->flags
, bytes
, 1);
3697 data_sinfo
->bytes_may_use
+= bytes
;
3698 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
3699 data_sinfo
->flags
, bytes
, 1);
3700 spin_unlock(&data_sinfo
->lock
);
3706 * Called if we need to clear a data reservation for this inode.
3708 void btrfs_free_reserved_data_space(struct inode
*inode
, u64 bytes
)
3710 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3711 struct btrfs_space_info
*data_sinfo
;
3713 /* make sure bytes are sectorsize aligned */
3714 bytes
= ALIGN(bytes
, root
->sectorsize
);
3716 data_sinfo
= root
->fs_info
->data_sinfo
;
3717 spin_lock(&data_sinfo
->lock
);
3718 WARN_ON(data_sinfo
->bytes_may_use
< bytes
);
3719 data_sinfo
->bytes_may_use
-= bytes
;
3720 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
3721 data_sinfo
->flags
, bytes
, 0);
3722 spin_unlock(&data_sinfo
->lock
);
3725 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
3727 struct list_head
*head
= &info
->space_info
;
3728 struct btrfs_space_info
*found
;
3731 list_for_each_entry_rcu(found
, head
, list
) {
3732 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
3733 found
->force_alloc
= CHUNK_ALLOC_FORCE
;
3738 static inline u64
calc_global_rsv_need_space(struct btrfs_block_rsv
*global
)
3740 return (global
->size
<< 1);
3743 static int should_alloc_chunk(struct btrfs_root
*root
,
3744 struct btrfs_space_info
*sinfo
, int force
)
3746 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
3747 u64 num_bytes
= sinfo
->total_bytes
- sinfo
->bytes_readonly
;
3748 u64 num_allocated
= sinfo
->bytes_used
+ sinfo
->bytes_reserved
;
3751 if (force
== CHUNK_ALLOC_FORCE
)
3755 * We need to take into account the global rsv because for all intents
3756 * and purposes it's used space. Don't worry about locking the
3757 * global_rsv, it doesn't change except when the transaction commits.
3759 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
3760 num_allocated
+= calc_global_rsv_need_space(global_rsv
);
3763 * in limited mode, we want to have some free space up to
3764 * about 1% of the FS size.
3766 if (force
== CHUNK_ALLOC_LIMITED
) {
3767 thresh
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
3768 thresh
= max_t(u64
, 64 * 1024 * 1024,
3769 div_factor_fine(thresh
, 1));
3771 if (num_bytes
- num_allocated
< thresh
)
3775 if (num_allocated
+ 2 * 1024 * 1024 < div_factor(num_bytes
, 8))
3780 static u64
get_system_chunk_thresh(struct btrfs_root
*root
, u64 type
)
3784 if (type
& (BTRFS_BLOCK_GROUP_RAID10
|
3785 BTRFS_BLOCK_GROUP_RAID0
|
3786 BTRFS_BLOCK_GROUP_RAID5
|
3787 BTRFS_BLOCK_GROUP_RAID6
))
3788 num_dev
= root
->fs_info
->fs_devices
->rw_devices
;
3789 else if (type
& BTRFS_BLOCK_GROUP_RAID1
)
3792 num_dev
= 1; /* DUP or single */
3794 /* metadata for updaing devices and chunk tree */
3795 return btrfs_calc_trans_metadata_size(root
, num_dev
+ 1);
3798 static void check_system_chunk(struct btrfs_trans_handle
*trans
,
3799 struct btrfs_root
*root
, u64 type
)
3801 struct btrfs_space_info
*info
;
3805 info
= __find_space_info(root
->fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
3806 spin_lock(&info
->lock
);
3807 left
= info
->total_bytes
- info
->bytes_used
- info
->bytes_pinned
-
3808 info
->bytes_reserved
- info
->bytes_readonly
;
3809 spin_unlock(&info
->lock
);
3811 thresh
= get_system_chunk_thresh(root
, type
);
3812 if (left
< thresh
&& btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
3813 btrfs_info(root
->fs_info
, "left=%llu, need=%llu, flags=%llu",
3814 left
, thresh
, type
);
3815 dump_space_info(info
, 0, 0);
3818 if (left
< thresh
) {
3821 flags
= btrfs_get_alloc_profile(root
->fs_info
->chunk_root
, 0);
3822 btrfs_alloc_chunk(trans
, root
, flags
);
3826 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
3827 struct btrfs_root
*extent_root
, u64 flags
, int force
)
3829 struct btrfs_space_info
*space_info
;
3830 struct btrfs_fs_info
*fs_info
= extent_root
->fs_info
;
3831 int wait_for_alloc
= 0;
3834 /* Don't re-enter if we're already allocating a chunk */
3835 if (trans
->allocating_chunk
)
3838 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
3840 ret
= update_space_info(extent_root
->fs_info
, flags
,
3842 BUG_ON(ret
); /* -ENOMEM */
3844 BUG_ON(!space_info
); /* Logic error */
3847 spin_lock(&space_info
->lock
);
3848 if (force
< space_info
->force_alloc
)
3849 force
= space_info
->force_alloc
;
3850 if (space_info
->full
) {
3851 if (should_alloc_chunk(extent_root
, space_info
, force
))
3855 spin_unlock(&space_info
->lock
);
3859 if (!should_alloc_chunk(extent_root
, space_info
, force
)) {
3860 spin_unlock(&space_info
->lock
);
3862 } else if (space_info
->chunk_alloc
) {
3865 space_info
->chunk_alloc
= 1;
3868 spin_unlock(&space_info
->lock
);
3870 mutex_lock(&fs_info
->chunk_mutex
);
3873 * The chunk_mutex is held throughout the entirety of a chunk
3874 * allocation, so once we've acquired the chunk_mutex we know that the
3875 * other guy is done and we need to recheck and see if we should
3878 if (wait_for_alloc
) {
3879 mutex_unlock(&fs_info
->chunk_mutex
);
3884 trans
->allocating_chunk
= true;
3887 * If we have mixed data/metadata chunks we want to make sure we keep
3888 * allocating mixed chunks instead of individual chunks.
3890 if (btrfs_mixed_space_info(space_info
))
3891 flags
|= (BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
);
3894 * if we're doing a data chunk, go ahead and make sure that
3895 * we keep a reasonable number of metadata chunks allocated in the
3898 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
3899 fs_info
->data_chunk_allocations
++;
3900 if (!(fs_info
->data_chunk_allocations
%
3901 fs_info
->metadata_ratio
))
3902 force_metadata_allocation(fs_info
);
3906 * Check if we have enough space in SYSTEM chunk because we may need
3907 * to update devices.
3909 check_system_chunk(trans
, extent_root
, flags
);
3911 ret
= btrfs_alloc_chunk(trans
, extent_root
, flags
);
3912 trans
->allocating_chunk
= false;
3914 spin_lock(&space_info
->lock
);
3915 if (ret
< 0 && ret
!= -ENOSPC
)
3918 space_info
->full
= 1;
3922 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3924 space_info
->chunk_alloc
= 0;
3925 spin_unlock(&space_info
->lock
);
3926 mutex_unlock(&fs_info
->chunk_mutex
);
3930 static int can_overcommit(struct btrfs_root
*root
,
3931 struct btrfs_space_info
*space_info
, u64 bytes
,
3932 enum btrfs_reserve_flush_enum flush
)
3934 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
3935 u64 profile
= btrfs_get_alloc_profile(root
, 0);
3940 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
3941 space_info
->bytes_pinned
+ space_info
->bytes_readonly
;
3944 * We only want to allow over committing if we have lots of actual space
3945 * free, but if we don't have enough space to handle the global reserve
3946 * space then we could end up having a real enospc problem when trying
3947 * to allocate a chunk or some other such important allocation.
3949 spin_lock(&global_rsv
->lock
);
3950 space_size
= calc_global_rsv_need_space(global_rsv
);
3951 spin_unlock(&global_rsv
->lock
);
3952 if (used
+ space_size
>= space_info
->total_bytes
)
3955 used
+= space_info
->bytes_may_use
;
3957 spin_lock(&root
->fs_info
->free_chunk_lock
);
3958 avail
= root
->fs_info
->free_chunk_space
;
3959 spin_unlock(&root
->fs_info
->free_chunk_lock
);
3962 * If we have dup, raid1 or raid10 then only half of the free
3963 * space is actually useable. For raid56, the space info used
3964 * doesn't include the parity drive, so we don't have to
3967 if (profile
& (BTRFS_BLOCK_GROUP_DUP
|
3968 BTRFS_BLOCK_GROUP_RAID1
|
3969 BTRFS_BLOCK_GROUP_RAID10
))
3973 * If we aren't flushing all things, let us overcommit up to
3974 * 1/2th of the space. If we can flush, don't let us overcommit
3975 * too much, let it overcommit up to 1/8 of the space.
3977 if (flush
== BTRFS_RESERVE_FLUSH_ALL
)
3982 if (used
+ bytes
< space_info
->total_bytes
+ avail
)
3987 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root
*root
,
3988 unsigned long nr_pages
, int nr_items
)
3990 struct super_block
*sb
= root
->fs_info
->sb
;
3992 if (down_read_trylock(&sb
->s_umount
)) {
3993 writeback_inodes_sb_nr(sb
, nr_pages
, WB_REASON_FS_FREE_SPACE
);
3994 up_read(&sb
->s_umount
);
3997 * We needn't worry the filesystem going from r/w to r/o though
3998 * we don't acquire ->s_umount mutex, because the filesystem
3999 * should guarantee the delalloc inodes list be empty after
4000 * the filesystem is readonly(all dirty pages are written to
4003 btrfs_start_delalloc_roots(root
->fs_info
, 0, nr_items
);
4004 if (!current
->journal_info
)
4005 btrfs_wait_ordered_roots(root
->fs_info
, nr_items
);
4009 static inline int calc_reclaim_items_nr(struct btrfs_root
*root
, u64 to_reclaim
)
4014 bytes
= btrfs_calc_trans_metadata_size(root
, 1);
4015 nr
= (int)div64_u64(to_reclaim
, bytes
);
4021 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4024 * shrink metadata reservation for delalloc
4026 static void shrink_delalloc(struct btrfs_root
*root
, u64 to_reclaim
, u64 orig
,
4029 struct btrfs_block_rsv
*block_rsv
;
4030 struct btrfs_space_info
*space_info
;
4031 struct btrfs_trans_handle
*trans
;
4035 unsigned long nr_pages
;
4038 enum btrfs_reserve_flush_enum flush
;
4040 /* Calc the number of the pages we need flush for space reservation */
4041 items
= calc_reclaim_items_nr(root
, to_reclaim
);
4042 to_reclaim
= items
* EXTENT_SIZE_PER_ITEM
;
4044 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4045 block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
4046 space_info
= block_rsv
->space_info
;
4048 delalloc_bytes
= percpu_counter_sum_positive(
4049 &root
->fs_info
->delalloc_bytes
);
4050 if (delalloc_bytes
== 0) {
4054 btrfs_wait_ordered_roots(root
->fs_info
, items
);
4059 while (delalloc_bytes
&& loops
< 3) {
4060 max_reclaim
= min(delalloc_bytes
, to_reclaim
);
4061 nr_pages
= max_reclaim
>> PAGE_CACHE_SHIFT
;
4062 btrfs_writeback_inodes_sb_nr(root
, nr_pages
, items
);
4064 * We need to wait for the async pages to actually start before
4067 max_reclaim
= atomic_read(&root
->fs_info
->async_delalloc_pages
);
4071 if (max_reclaim
<= nr_pages
)
4074 max_reclaim
-= nr_pages
;
4076 wait_event(root
->fs_info
->async_submit_wait
,
4077 atomic_read(&root
->fs_info
->async_delalloc_pages
) <=
4081 flush
= BTRFS_RESERVE_FLUSH_ALL
;
4083 flush
= BTRFS_RESERVE_NO_FLUSH
;
4084 spin_lock(&space_info
->lock
);
4085 if (can_overcommit(root
, space_info
, orig
, flush
)) {
4086 spin_unlock(&space_info
->lock
);
4089 spin_unlock(&space_info
->lock
);
4092 if (wait_ordered
&& !trans
) {
4093 btrfs_wait_ordered_roots(root
->fs_info
, items
);
4095 time_left
= schedule_timeout_killable(1);
4099 delalloc_bytes
= percpu_counter_sum_positive(
4100 &root
->fs_info
->delalloc_bytes
);
4105 * maybe_commit_transaction - possibly commit the transaction if its ok to
4106 * @root - the root we're allocating for
4107 * @bytes - the number of bytes we want to reserve
4108 * @force - force the commit
4110 * This will check to make sure that committing the transaction will actually
4111 * get us somewhere and then commit the transaction if it does. Otherwise it
4112 * will return -ENOSPC.
4114 static int may_commit_transaction(struct btrfs_root
*root
,
4115 struct btrfs_space_info
*space_info
,
4116 u64 bytes
, int force
)
4118 struct btrfs_block_rsv
*delayed_rsv
= &root
->fs_info
->delayed_block_rsv
;
4119 struct btrfs_trans_handle
*trans
;
4121 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4128 /* See if there is enough pinned space to make this reservation */
4129 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4134 * See if there is some space in the delayed insertion reservation for
4137 if (space_info
!= delayed_rsv
->space_info
)
4140 spin_lock(&delayed_rsv
->lock
);
4141 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4142 bytes
- delayed_rsv
->size
) >= 0) {
4143 spin_unlock(&delayed_rsv
->lock
);
4146 spin_unlock(&delayed_rsv
->lock
);
4149 trans
= btrfs_join_transaction(root
);
4153 return btrfs_commit_transaction(trans
, root
);
4157 FLUSH_DELAYED_ITEMS_NR
= 1,
4158 FLUSH_DELAYED_ITEMS
= 2,
4160 FLUSH_DELALLOC_WAIT
= 4,
4165 static int flush_space(struct btrfs_root
*root
,
4166 struct btrfs_space_info
*space_info
, u64 num_bytes
,
4167 u64 orig_bytes
, int state
)
4169 struct btrfs_trans_handle
*trans
;
4174 case FLUSH_DELAYED_ITEMS_NR
:
4175 case FLUSH_DELAYED_ITEMS
:
4176 if (state
== FLUSH_DELAYED_ITEMS_NR
)
4177 nr
= calc_reclaim_items_nr(root
, num_bytes
) * 2;
4181 trans
= btrfs_join_transaction(root
);
4182 if (IS_ERR(trans
)) {
4183 ret
= PTR_ERR(trans
);
4186 ret
= btrfs_run_delayed_items_nr(trans
, root
, nr
);
4187 btrfs_end_transaction(trans
, root
);
4189 case FLUSH_DELALLOC
:
4190 case FLUSH_DELALLOC_WAIT
:
4191 shrink_delalloc(root
, num_bytes
* 2, orig_bytes
,
4192 state
== FLUSH_DELALLOC_WAIT
);
4195 trans
= btrfs_join_transaction(root
);
4196 if (IS_ERR(trans
)) {
4197 ret
= PTR_ERR(trans
);
4200 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
4201 btrfs_get_alloc_profile(root
, 0),
4202 CHUNK_ALLOC_NO_FORCE
);
4203 btrfs_end_transaction(trans
, root
);
4208 ret
= may_commit_transaction(root
, space_info
, orig_bytes
, 0);
4219 btrfs_calc_reclaim_metadata_size(struct btrfs_root
*root
,
4220 struct btrfs_space_info
*space_info
)
4226 to_reclaim
= min_t(u64
, num_online_cpus() * 1024 * 1024,
4228 spin_lock(&space_info
->lock
);
4229 if (can_overcommit(root
, space_info
, to_reclaim
,
4230 BTRFS_RESERVE_FLUSH_ALL
)) {
4235 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4236 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4237 space_info
->bytes_may_use
;
4238 if (can_overcommit(root
, space_info
, 1024 * 1024,
4239 BTRFS_RESERVE_FLUSH_ALL
))
4240 expected
= div_factor_fine(space_info
->total_bytes
, 95);
4242 expected
= div_factor_fine(space_info
->total_bytes
, 90);
4244 if (used
> expected
)
4245 to_reclaim
= used
- expected
;
4248 to_reclaim
= min(to_reclaim
, space_info
->bytes_may_use
+
4249 space_info
->bytes_reserved
);
4251 spin_unlock(&space_info
->lock
);
4256 static inline int need_do_async_reclaim(struct btrfs_space_info
*space_info
,
4257 struct btrfs_fs_info
*fs_info
, u64 used
)
4259 return (used
>= div_factor_fine(space_info
->total_bytes
, 98) &&
4260 !btrfs_fs_closing(fs_info
) &&
4261 !test_bit(BTRFS_FS_STATE_REMOUNTING
, &fs_info
->fs_state
));
4264 static int btrfs_need_do_async_reclaim(struct btrfs_space_info
*space_info
,
4265 struct btrfs_fs_info
*fs_info
,
4270 spin_lock(&space_info
->lock
);
4272 * We run out of space and have not got any free space via flush_space,
4273 * so don't bother doing async reclaim.
4275 if (flush_state
> COMMIT_TRANS
&& space_info
->full
) {
4276 spin_unlock(&space_info
->lock
);
4280 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4281 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4282 space_info
->bytes_may_use
;
4283 if (need_do_async_reclaim(space_info
, fs_info
, used
)) {
4284 spin_unlock(&space_info
->lock
);
4287 spin_unlock(&space_info
->lock
);
4292 static void btrfs_async_reclaim_metadata_space(struct work_struct
*work
)
4294 struct btrfs_fs_info
*fs_info
;
4295 struct btrfs_space_info
*space_info
;
4299 fs_info
= container_of(work
, struct btrfs_fs_info
, async_reclaim_work
);
4300 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4302 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
->fs_root
,
4307 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
4309 flush_space(fs_info
->fs_root
, space_info
, to_reclaim
,
4310 to_reclaim
, flush_state
);
4312 if (!btrfs_need_do_async_reclaim(space_info
, fs_info
,
4315 } while (flush_state
<= COMMIT_TRANS
);
4317 if (btrfs_need_do_async_reclaim(space_info
, fs_info
, flush_state
))
4318 queue_work(system_unbound_wq
, work
);
4321 void btrfs_init_async_reclaim_work(struct work_struct
*work
)
4323 INIT_WORK(work
, btrfs_async_reclaim_metadata_space
);
4327 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4328 * @root - the root we're allocating for
4329 * @block_rsv - the block_rsv we're allocating for
4330 * @orig_bytes - the number of bytes we want
4331 * @flush - whether or not we can flush to make our reservation
4333 * This will reserve orgi_bytes number of bytes from the space info associated
4334 * with the block_rsv. If there is not enough space it will make an attempt to
4335 * flush out space to make room. It will do this by flushing delalloc if
4336 * possible or committing the transaction. If flush is 0 then no attempts to
4337 * regain reservations will be made and this will fail if there is not enough
4340 static int reserve_metadata_bytes(struct btrfs_root
*root
,
4341 struct btrfs_block_rsv
*block_rsv
,
4343 enum btrfs_reserve_flush_enum flush
)
4345 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
4347 u64 num_bytes
= orig_bytes
;
4348 int flush_state
= FLUSH_DELAYED_ITEMS_NR
;
4350 bool flushing
= false;
4354 spin_lock(&space_info
->lock
);
4356 * We only want to wait if somebody other than us is flushing and we
4357 * are actually allowed to flush all things.
4359 while (flush
== BTRFS_RESERVE_FLUSH_ALL
&& !flushing
&&
4360 space_info
->flush
) {
4361 spin_unlock(&space_info
->lock
);
4363 * If we have a trans handle we can't wait because the flusher
4364 * may have to commit the transaction, which would mean we would
4365 * deadlock since we are waiting for the flusher to finish, but
4366 * hold the current transaction open.
4368 if (current
->journal_info
)
4370 ret
= wait_event_killable(space_info
->wait
, !space_info
->flush
);
4371 /* Must have been killed, return */
4375 spin_lock(&space_info
->lock
);
4379 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4380 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4381 space_info
->bytes_may_use
;
4384 * The idea here is that we've not already over-reserved the block group
4385 * then we can go ahead and save our reservation first and then start
4386 * flushing if we need to. Otherwise if we've already overcommitted
4387 * lets start flushing stuff first and then come back and try to make
4390 if (used
<= space_info
->total_bytes
) {
4391 if (used
+ orig_bytes
<= space_info
->total_bytes
) {
4392 space_info
->bytes_may_use
+= orig_bytes
;
4393 trace_btrfs_space_reservation(root
->fs_info
,
4394 "space_info", space_info
->flags
, orig_bytes
, 1);
4398 * Ok set num_bytes to orig_bytes since we aren't
4399 * overocmmitted, this way we only try and reclaim what
4402 num_bytes
= orig_bytes
;
4406 * Ok we're over committed, set num_bytes to the overcommitted
4407 * amount plus the amount of bytes that we need for this
4410 num_bytes
= used
- space_info
->total_bytes
+
4414 if (ret
&& can_overcommit(root
, space_info
, orig_bytes
, flush
)) {
4415 space_info
->bytes_may_use
+= orig_bytes
;
4416 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
4417 space_info
->flags
, orig_bytes
,
4423 * Couldn't make our reservation, save our place so while we're trying
4424 * to reclaim space we can actually use it instead of somebody else
4425 * stealing it from us.
4427 * We make the other tasks wait for the flush only when we can flush
4430 if (ret
&& flush
!= BTRFS_RESERVE_NO_FLUSH
) {
4432 space_info
->flush
= 1;
4433 } else if (!ret
&& space_info
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
4436 * We will do the space reservation dance during log replay,
4437 * which means we won't have fs_info->fs_root set, so don't do
4438 * the async reclaim as we will panic.
4440 if (!root
->fs_info
->log_root_recovering
&&
4441 need_do_async_reclaim(space_info
, root
->fs_info
, used
) &&
4442 !work_busy(&root
->fs_info
->async_reclaim_work
))
4443 queue_work(system_unbound_wq
,
4444 &root
->fs_info
->async_reclaim_work
);
4446 spin_unlock(&space_info
->lock
);
4448 if (!ret
|| flush
== BTRFS_RESERVE_NO_FLUSH
)
4451 ret
= flush_space(root
, space_info
, num_bytes
, orig_bytes
,
4456 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4457 * would happen. So skip delalloc flush.
4459 if (flush
== BTRFS_RESERVE_FLUSH_LIMIT
&&
4460 (flush_state
== FLUSH_DELALLOC
||
4461 flush_state
== FLUSH_DELALLOC_WAIT
))
4462 flush_state
= ALLOC_CHUNK
;
4466 else if (flush
== BTRFS_RESERVE_FLUSH_LIMIT
&&
4467 flush_state
< COMMIT_TRANS
)
4469 else if (flush
== BTRFS_RESERVE_FLUSH_ALL
&&
4470 flush_state
<= COMMIT_TRANS
)
4474 if (ret
== -ENOSPC
&&
4475 unlikely(root
->orphan_cleanup_state
== ORPHAN_CLEANUP_STARTED
)) {
4476 struct btrfs_block_rsv
*global_rsv
=
4477 &root
->fs_info
->global_block_rsv
;
4479 if (block_rsv
!= global_rsv
&&
4480 !block_rsv_use_bytes(global_rsv
, orig_bytes
))
4484 trace_btrfs_space_reservation(root
->fs_info
,
4485 "space_info:enospc",
4486 space_info
->flags
, orig_bytes
, 1);
4488 spin_lock(&space_info
->lock
);
4489 space_info
->flush
= 0;
4490 wake_up_all(&space_info
->wait
);
4491 spin_unlock(&space_info
->lock
);
4496 static struct btrfs_block_rsv
*get_block_rsv(
4497 const struct btrfs_trans_handle
*trans
,
4498 const struct btrfs_root
*root
)
4500 struct btrfs_block_rsv
*block_rsv
= NULL
;
4502 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
))
4503 block_rsv
= trans
->block_rsv
;
4505 if (root
== root
->fs_info
->csum_root
&& trans
->adding_csums
)
4506 block_rsv
= trans
->block_rsv
;
4508 if (root
== root
->fs_info
->uuid_root
)
4509 block_rsv
= trans
->block_rsv
;
4512 block_rsv
= root
->block_rsv
;
4515 block_rsv
= &root
->fs_info
->empty_block_rsv
;
4520 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
4524 spin_lock(&block_rsv
->lock
);
4525 if (block_rsv
->reserved
>= num_bytes
) {
4526 block_rsv
->reserved
-= num_bytes
;
4527 if (block_rsv
->reserved
< block_rsv
->size
)
4528 block_rsv
->full
= 0;
4531 spin_unlock(&block_rsv
->lock
);
4535 static void block_rsv_add_bytes(struct btrfs_block_rsv
*block_rsv
,
4536 u64 num_bytes
, int update_size
)
4538 spin_lock(&block_rsv
->lock
);
4539 block_rsv
->reserved
+= num_bytes
;
4541 block_rsv
->size
+= num_bytes
;
4542 else if (block_rsv
->reserved
>= block_rsv
->size
)
4543 block_rsv
->full
= 1;
4544 spin_unlock(&block_rsv
->lock
);
4547 int btrfs_cond_migrate_bytes(struct btrfs_fs_info
*fs_info
,
4548 struct btrfs_block_rsv
*dest
, u64 num_bytes
,
4551 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
4554 if (global_rsv
->space_info
!= dest
->space_info
)
4557 spin_lock(&global_rsv
->lock
);
4558 min_bytes
= div_factor(global_rsv
->size
, min_factor
);
4559 if (global_rsv
->reserved
< min_bytes
+ num_bytes
) {
4560 spin_unlock(&global_rsv
->lock
);
4563 global_rsv
->reserved
-= num_bytes
;
4564 if (global_rsv
->reserved
< global_rsv
->size
)
4565 global_rsv
->full
= 0;
4566 spin_unlock(&global_rsv
->lock
);
4568 block_rsv_add_bytes(dest
, num_bytes
, 1);
4572 static void block_rsv_release_bytes(struct btrfs_fs_info
*fs_info
,
4573 struct btrfs_block_rsv
*block_rsv
,
4574 struct btrfs_block_rsv
*dest
, u64 num_bytes
)
4576 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
4578 spin_lock(&block_rsv
->lock
);
4579 if (num_bytes
== (u64
)-1)
4580 num_bytes
= block_rsv
->size
;
4581 block_rsv
->size
-= num_bytes
;
4582 if (block_rsv
->reserved
>= block_rsv
->size
) {
4583 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
4584 block_rsv
->reserved
= block_rsv
->size
;
4585 block_rsv
->full
= 1;
4589 spin_unlock(&block_rsv
->lock
);
4591 if (num_bytes
> 0) {
4593 spin_lock(&dest
->lock
);
4597 bytes_to_add
= dest
->size
- dest
->reserved
;
4598 bytes_to_add
= min(num_bytes
, bytes_to_add
);
4599 dest
->reserved
+= bytes_to_add
;
4600 if (dest
->reserved
>= dest
->size
)
4602 num_bytes
-= bytes_to_add
;
4604 spin_unlock(&dest
->lock
);
4607 spin_lock(&space_info
->lock
);
4608 space_info
->bytes_may_use
-= num_bytes
;
4609 trace_btrfs_space_reservation(fs_info
, "space_info",
4610 space_info
->flags
, num_bytes
, 0);
4611 spin_unlock(&space_info
->lock
);
4616 static int block_rsv_migrate_bytes(struct btrfs_block_rsv
*src
,
4617 struct btrfs_block_rsv
*dst
, u64 num_bytes
)
4621 ret
= block_rsv_use_bytes(src
, num_bytes
);
4625 block_rsv_add_bytes(dst
, num_bytes
, 1);
4629 void btrfs_init_block_rsv(struct btrfs_block_rsv
*rsv
, unsigned short type
)
4631 memset(rsv
, 0, sizeof(*rsv
));
4632 spin_lock_init(&rsv
->lock
);
4636 struct btrfs_block_rsv
*btrfs_alloc_block_rsv(struct btrfs_root
*root
,
4637 unsigned short type
)
4639 struct btrfs_block_rsv
*block_rsv
;
4640 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4642 block_rsv
= kmalloc(sizeof(*block_rsv
), GFP_NOFS
);
4646 btrfs_init_block_rsv(block_rsv
, type
);
4647 block_rsv
->space_info
= __find_space_info(fs_info
,
4648 BTRFS_BLOCK_GROUP_METADATA
);
4652 void btrfs_free_block_rsv(struct btrfs_root
*root
,
4653 struct btrfs_block_rsv
*rsv
)
4657 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
4661 int btrfs_block_rsv_add(struct btrfs_root
*root
,
4662 struct btrfs_block_rsv
*block_rsv
, u64 num_bytes
,
4663 enum btrfs_reserve_flush_enum flush
)
4670 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
4672 block_rsv_add_bytes(block_rsv
, num_bytes
, 1);
4679 int btrfs_block_rsv_check(struct btrfs_root
*root
,
4680 struct btrfs_block_rsv
*block_rsv
, int min_factor
)
4688 spin_lock(&block_rsv
->lock
);
4689 num_bytes
= div_factor(block_rsv
->size
, min_factor
);
4690 if (block_rsv
->reserved
>= num_bytes
)
4692 spin_unlock(&block_rsv
->lock
);
4697 int btrfs_block_rsv_refill(struct btrfs_root
*root
,
4698 struct btrfs_block_rsv
*block_rsv
, u64 min_reserved
,
4699 enum btrfs_reserve_flush_enum flush
)
4707 spin_lock(&block_rsv
->lock
);
4708 num_bytes
= min_reserved
;
4709 if (block_rsv
->reserved
>= num_bytes
)
4712 num_bytes
-= block_rsv
->reserved
;
4713 spin_unlock(&block_rsv
->lock
);
4718 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
4720 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
4727 int btrfs_block_rsv_migrate(struct btrfs_block_rsv
*src_rsv
,
4728 struct btrfs_block_rsv
*dst_rsv
,
4731 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
4734 void btrfs_block_rsv_release(struct btrfs_root
*root
,
4735 struct btrfs_block_rsv
*block_rsv
,
4738 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
4739 if (global_rsv
== block_rsv
||
4740 block_rsv
->space_info
!= global_rsv
->space_info
)
4742 block_rsv_release_bytes(root
->fs_info
, block_rsv
, global_rsv
,
4747 * helper to calculate size of global block reservation.
4748 * the desired value is sum of space used by extent tree,
4749 * checksum tree and root tree
4751 static u64
calc_global_metadata_size(struct btrfs_fs_info
*fs_info
)
4753 struct btrfs_space_info
*sinfo
;
4757 int csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
4759 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_DATA
);
4760 spin_lock(&sinfo
->lock
);
4761 data_used
= sinfo
->bytes_used
;
4762 spin_unlock(&sinfo
->lock
);
4764 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4765 spin_lock(&sinfo
->lock
);
4766 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_DATA
)
4768 meta_used
= sinfo
->bytes_used
;
4769 spin_unlock(&sinfo
->lock
);
4771 num_bytes
= (data_used
>> fs_info
->sb
->s_blocksize_bits
) *
4773 num_bytes
+= div64_u64(data_used
+ meta_used
, 50);
4775 if (num_bytes
* 3 > meta_used
)
4776 num_bytes
= div64_u64(meta_used
, 3);
4778 return ALIGN(num_bytes
, fs_info
->extent_root
->nodesize
<< 10);
4781 static void update_global_block_rsv(struct btrfs_fs_info
*fs_info
)
4783 struct btrfs_block_rsv
*block_rsv
= &fs_info
->global_block_rsv
;
4784 struct btrfs_space_info
*sinfo
= block_rsv
->space_info
;
4787 num_bytes
= calc_global_metadata_size(fs_info
);
4789 spin_lock(&sinfo
->lock
);
4790 spin_lock(&block_rsv
->lock
);
4792 block_rsv
->size
= min_t(u64
, num_bytes
, 512 * 1024 * 1024);
4794 num_bytes
= sinfo
->bytes_used
+ sinfo
->bytes_pinned
+
4795 sinfo
->bytes_reserved
+ sinfo
->bytes_readonly
+
4796 sinfo
->bytes_may_use
;
4798 if (sinfo
->total_bytes
> num_bytes
) {
4799 num_bytes
= sinfo
->total_bytes
- num_bytes
;
4800 block_rsv
->reserved
+= num_bytes
;
4801 sinfo
->bytes_may_use
+= num_bytes
;
4802 trace_btrfs_space_reservation(fs_info
, "space_info",
4803 sinfo
->flags
, num_bytes
, 1);
4806 if (block_rsv
->reserved
>= block_rsv
->size
) {
4807 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
4808 sinfo
->bytes_may_use
-= num_bytes
;
4809 trace_btrfs_space_reservation(fs_info
, "space_info",
4810 sinfo
->flags
, num_bytes
, 0);
4811 block_rsv
->reserved
= block_rsv
->size
;
4812 block_rsv
->full
= 1;
4815 spin_unlock(&block_rsv
->lock
);
4816 spin_unlock(&sinfo
->lock
);
4819 static void init_global_block_rsv(struct btrfs_fs_info
*fs_info
)
4821 struct btrfs_space_info
*space_info
;
4823 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
4824 fs_info
->chunk_block_rsv
.space_info
= space_info
;
4826 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4827 fs_info
->global_block_rsv
.space_info
= space_info
;
4828 fs_info
->delalloc_block_rsv
.space_info
= space_info
;
4829 fs_info
->trans_block_rsv
.space_info
= space_info
;
4830 fs_info
->empty_block_rsv
.space_info
= space_info
;
4831 fs_info
->delayed_block_rsv
.space_info
= space_info
;
4833 fs_info
->extent_root
->block_rsv
= &fs_info
->global_block_rsv
;
4834 fs_info
->csum_root
->block_rsv
= &fs_info
->global_block_rsv
;
4835 fs_info
->dev_root
->block_rsv
= &fs_info
->global_block_rsv
;
4836 fs_info
->tree_root
->block_rsv
= &fs_info
->global_block_rsv
;
4837 if (fs_info
->quota_root
)
4838 fs_info
->quota_root
->block_rsv
= &fs_info
->global_block_rsv
;
4839 fs_info
->chunk_root
->block_rsv
= &fs_info
->chunk_block_rsv
;
4841 update_global_block_rsv(fs_info
);
4844 static void release_global_block_rsv(struct btrfs_fs_info
*fs_info
)
4846 block_rsv_release_bytes(fs_info
, &fs_info
->global_block_rsv
, NULL
,
4848 WARN_ON(fs_info
->delalloc_block_rsv
.size
> 0);
4849 WARN_ON(fs_info
->delalloc_block_rsv
.reserved
> 0);
4850 WARN_ON(fs_info
->trans_block_rsv
.size
> 0);
4851 WARN_ON(fs_info
->trans_block_rsv
.reserved
> 0);
4852 WARN_ON(fs_info
->chunk_block_rsv
.size
> 0);
4853 WARN_ON(fs_info
->chunk_block_rsv
.reserved
> 0);
4854 WARN_ON(fs_info
->delayed_block_rsv
.size
> 0);
4855 WARN_ON(fs_info
->delayed_block_rsv
.reserved
> 0);
4858 void btrfs_trans_release_metadata(struct btrfs_trans_handle
*trans
,
4859 struct btrfs_root
*root
)
4861 if (!trans
->block_rsv
)
4864 if (!trans
->bytes_reserved
)
4867 trace_btrfs_space_reservation(root
->fs_info
, "transaction",
4868 trans
->transid
, trans
->bytes_reserved
, 0);
4869 btrfs_block_rsv_release(root
, trans
->block_rsv
, trans
->bytes_reserved
);
4870 trans
->bytes_reserved
= 0;
4873 /* Can only return 0 or -ENOSPC */
4874 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle
*trans
,
4875 struct inode
*inode
)
4877 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4878 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
4879 struct btrfs_block_rsv
*dst_rsv
= root
->orphan_block_rsv
;
4882 * We need to hold space in order to delete our orphan item once we've
4883 * added it, so this takes the reservation so we can release it later
4884 * when we are truly done with the orphan item.
4886 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
4887 trace_btrfs_space_reservation(root
->fs_info
, "orphan",
4888 btrfs_ino(inode
), num_bytes
, 1);
4889 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
4892 void btrfs_orphan_release_metadata(struct inode
*inode
)
4894 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4895 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
4896 trace_btrfs_space_reservation(root
->fs_info
, "orphan",
4897 btrfs_ino(inode
), num_bytes
, 0);
4898 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
, num_bytes
);
4902 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4903 * root: the root of the parent directory
4904 * rsv: block reservation
4905 * items: the number of items that we need do reservation
4906 * qgroup_reserved: used to return the reserved size in qgroup
4908 * This function is used to reserve the space for snapshot/subvolume
4909 * creation and deletion. Those operations are different with the
4910 * common file/directory operations, they change two fs/file trees
4911 * and root tree, the number of items that the qgroup reserves is
4912 * different with the free space reservation. So we can not use
4913 * the space reseravtion mechanism in start_transaction().
4915 int btrfs_subvolume_reserve_metadata(struct btrfs_root
*root
,
4916 struct btrfs_block_rsv
*rsv
,
4918 u64
*qgroup_reserved
,
4919 bool use_global_rsv
)
4923 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
4925 if (root
->fs_info
->quota_enabled
) {
4926 /* One for parent inode, two for dir entries */
4927 num_bytes
= 3 * root
->nodesize
;
4928 ret
= btrfs_qgroup_reserve(root
, num_bytes
);
4935 *qgroup_reserved
= num_bytes
;
4937 num_bytes
= btrfs_calc_trans_metadata_size(root
, items
);
4938 rsv
->space_info
= __find_space_info(root
->fs_info
,
4939 BTRFS_BLOCK_GROUP_METADATA
);
4940 ret
= btrfs_block_rsv_add(root
, rsv
, num_bytes
,
4941 BTRFS_RESERVE_FLUSH_ALL
);
4943 if (ret
== -ENOSPC
&& use_global_rsv
)
4944 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
, num_bytes
);
4947 if (*qgroup_reserved
)
4948 btrfs_qgroup_free(root
, *qgroup_reserved
);
4954 void btrfs_subvolume_release_metadata(struct btrfs_root
*root
,
4955 struct btrfs_block_rsv
*rsv
,
4956 u64 qgroup_reserved
)
4958 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
4959 if (qgroup_reserved
)
4960 btrfs_qgroup_free(root
, qgroup_reserved
);
4964 * drop_outstanding_extent - drop an outstanding extent
4965 * @inode: the inode we're dropping the extent for
4967 * This is called when we are freeing up an outstanding extent, either called
4968 * after an error or after an extent is written. This will return the number of
4969 * reserved extents that need to be freed. This must be called with
4970 * BTRFS_I(inode)->lock held.
4972 static unsigned drop_outstanding_extent(struct inode
*inode
)
4974 unsigned drop_inode_space
= 0;
4975 unsigned dropped_extents
= 0;
4977 BUG_ON(!BTRFS_I(inode
)->outstanding_extents
);
4978 BTRFS_I(inode
)->outstanding_extents
--;
4980 if (BTRFS_I(inode
)->outstanding_extents
== 0 &&
4981 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
4982 &BTRFS_I(inode
)->runtime_flags
))
4983 drop_inode_space
= 1;
4986 * If we have more or the same amount of outsanding extents than we have
4987 * reserved then we need to leave the reserved extents count alone.
4989 if (BTRFS_I(inode
)->outstanding_extents
>=
4990 BTRFS_I(inode
)->reserved_extents
)
4991 return drop_inode_space
;
4993 dropped_extents
= BTRFS_I(inode
)->reserved_extents
-
4994 BTRFS_I(inode
)->outstanding_extents
;
4995 BTRFS_I(inode
)->reserved_extents
-= dropped_extents
;
4996 return dropped_extents
+ drop_inode_space
;
5000 * calc_csum_metadata_size - return the amount of metada space that must be
5001 * reserved/free'd for the given bytes.
5002 * @inode: the inode we're manipulating
5003 * @num_bytes: the number of bytes in question
5004 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5006 * This adjusts the number of csum_bytes in the inode and then returns the
5007 * correct amount of metadata that must either be reserved or freed. We
5008 * calculate how many checksums we can fit into one leaf and then divide the
5009 * number of bytes that will need to be checksumed by this value to figure out
5010 * how many checksums will be required. If we are adding bytes then the number
5011 * may go up and we will return the number of additional bytes that must be
5012 * reserved. If it is going down we will return the number of bytes that must
5015 * This must be called with BTRFS_I(inode)->lock held.
5017 static u64
calc_csum_metadata_size(struct inode
*inode
, u64 num_bytes
,
5020 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5022 int num_csums_per_leaf
;
5026 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
&&
5027 BTRFS_I(inode
)->csum_bytes
== 0)
5030 old_csums
= (int)div64_u64(BTRFS_I(inode
)->csum_bytes
, root
->sectorsize
);
5032 BTRFS_I(inode
)->csum_bytes
+= num_bytes
;
5034 BTRFS_I(inode
)->csum_bytes
-= num_bytes
;
5035 csum_size
= BTRFS_LEAF_DATA_SIZE(root
) - sizeof(struct btrfs_item
);
5036 num_csums_per_leaf
= (int)div64_u64(csum_size
,
5037 sizeof(struct btrfs_csum_item
) +
5038 sizeof(struct btrfs_disk_key
));
5039 num_csums
= (int)div64_u64(BTRFS_I(inode
)->csum_bytes
, root
->sectorsize
);
5040 num_csums
= num_csums
+ num_csums_per_leaf
- 1;
5041 num_csums
= num_csums
/ num_csums_per_leaf
;
5043 old_csums
= old_csums
+ num_csums_per_leaf
- 1;
5044 old_csums
= old_csums
/ num_csums_per_leaf
;
5046 /* No change, no need to reserve more */
5047 if (old_csums
== num_csums
)
5051 return btrfs_calc_trans_metadata_size(root
,
5052 num_csums
- old_csums
);
5054 return btrfs_calc_trans_metadata_size(root
, old_csums
- num_csums
);
5057 int btrfs_delalloc_reserve_metadata(struct inode
*inode
, u64 num_bytes
)
5059 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5060 struct btrfs_block_rsv
*block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
5063 unsigned nr_extents
= 0;
5064 int extra_reserve
= 0;
5065 enum btrfs_reserve_flush_enum flush
= BTRFS_RESERVE_FLUSH_ALL
;
5067 bool delalloc_lock
= true;
5071 /* If we are a free space inode we need to not flush since we will be in
5072 * the middle of a transaction commit. We also don't need the delalloc
5073 * mutex since we won't race with anybody. We need this mostly to make
5074 * lockdep shut its filthy mouth.
5076 if (btrfs_is_free_space_inode(inode
)) {
5077 flush
= BTRFS_RESERVE_NO_FLUSH
;
5078 delalloc_lock
= false;
5081 if (flush
!= BTRFS_RESERVE_NO_FLUSH
&&
5082 btrfs_transaction_in_commit(root
->fs_info
))
5083 schedule_timeout(1);
5086 mutex_lock(&BTRFS_I(inode
)->delalloc_mutex
);
5088 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
5090 spin_lock(&BTRFS_I(inode
)->lock
);
5091 BTRFS_I(inode
)->outstanding_extents
++;
5093 if (BTRFS_I(inode
)->outstanding_extents
>
5094 BTRFS_I(inode
)->reserved_extents
)
5095 nr_extents
= BTRFS_I(inode
)->outstanding_extents
-
5096 BTRFS_I(inode
)->reserved_extents
;
5099 * Add an item to reserve for updating the inode when we complete the
5102 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5103 &BTRFS_I(inode
)->runtime_flags
)) {
5108 to_reserve
= btrfs_calc_trans_metadata_size(root
, nr_extents
);
5109 to_reserve
+= calc_csum_metadata_size(inode
, num_bytes
, 1);
5110 csum_bytes
= BTRFS_I(inode
)->csum_bytes
;
5111 spin_unlock(&BTRFS_I(inode
)->lock
);
5113 if (root
->fs_info
->quota_enabled
) {
5114 ret
= btrfs_qgroup_reserve(root
, num_bytes
+
5115 nr_extents
* root
->nodesize
);
5120 ret
= reserve_metadata_bytes(root
, block_rsv
, to_reserve
, flush
);
5121 if (unlikely(ret
)) {
5122 if (root
->fs_info
->quota_enabled
)
5123 btrfs_qgroup_free(root
, num_bytes
+
5124 nr_extents
* root
->nodesize
);
5128 spin_lock(&BTRFS_I(inode
)->lock
);
5129 if (extra_reserve
) {
5130 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5131 &BTRFS_I(inode
)->runtime_flags
);
5134 BTRFS_I(inode
)->reserved_extents
+= nr_extents
;
5135 spin_unlock(&BTRFS_I(inode
)->lock
);
5138 mutex_unlock(&BTRFS_I(inode
)->delalloc_mutex
);
5141 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5142 btrfs_ino(inode
), to_reserve
, 1);
5143 block_rsv_add_bytes(block_rsv
, to_reserve
, 1);
5148 spin_lock(&BTRFS_I(inode
)->lock
);
5149 dropped
= drop_outstanding_extent(inode
);
5151 * If the inodes csum_bytes is the same as the original
5152 * csum_bytes then we know we haven't raced with any free()ers
5153 * so we can just reduce our inodes csum bytes and carry on.
5155 if (BTRFS_I(inode
)->csum_bytes
== csum_bytes
) {
5156 calc_csum_metadata_size(inode
, num_bytes
, 0);
5158 u64 orig_csum_bytes
= BTRFS_I(inode
)->csum_bytes
;
5162 * This is tricky, but first we need to figure out how much we
5163 * free'd from any free-ers that occured during this
5164 * reservation, so we reset ->csum_bytes to the csum_bytes
5165 * before we dropped our lock, and then call the free for the
5166 * number of bytes that were freed while we were trying our
5169 bytes
= csum_bytes
- BTRFS_I(inode
)->csum_bytes
;
5170 BTRFS_I(inode
)->csum_bytes
= csum_bytes
;
5171 to_free
= calc_csum_metadata_size(inode
, bytes
, 0);
5175 * Now we need to see how much we would have freed had we not
5176 * been making this reservation and our ->csum_bytes were not
5177 * artificially inflated.
5179 BTRFS_I(inode
)->csum_bytes
= csum_bytes
- num_bytes
;
5180 bytes
= csum_bytes
- orig_csum_bytes
;
5181 bytes
= calc_csum_metadata_size(inode
, bytes
, 0);
5184 * Now reset ->csum_bytes to what it should be. If bytes is
5185 * more than to_free then we would have free'd more space had we
5186 * not had an artificially high ->csum_bytes, so we need to free
5187 * the remainder. If bytes is the same or less then we don't
5188 * need to do anything, the other free-ers did the correct
5191 BTRFS_I(inode
)->csum_bytes
= orig_csum_bytes
- num_bytes
;
5192 if (bytes
> to_free
)
5193 to_free
= bytes
- to_free
;
5197 spin_unlock(&BTRFS_I(inode
)->lock
);
5199 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
5202 btrfs_block_rsv_release(root
, block_rsv
, to_free
);
5203 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5204 btrfs_ino(inode
), to_free
, 0);
5207 mutex_unlock(&BTRFS_I(inode
)->delalloc_mutex
);
5212 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5213 * @inode: the inode to release the reservation for
5214 * @num_bytes: the number of bytes we're releasing
5216 * This will release the metadata reservation for an inode. This can be called
5217 * once we complete IO for a given set of bytes to release their metadata
5220 void btrfs_delalloc_release_metadata(struct inode
*inode
, u64 num_bytes
)
5222 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5226 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
5227 spin_lock(&BTRFS_I(inode
)->lock
);
5228 dropped
= drop_outstanding_extent(inode
);
5231 to_free
= calc_csum_metadata_size(inode
, num_bytes
, 0);
5232 spin_unlock(&BTRFS_I(inode
)->lock
);
5234 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
5236 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5237 btrfs_ino(inode
), to_free
, 0);
5238 if (root
->fs_info
->quota_enabled
) {
5239 btrfs_qgroup_free(root
, num_bytes
+
5240 dropped
* root
->nodesize
);
5243 btrfs_block_rsv_release(root
, &root
->fs_info
->delalloc_block_rsv
,
5248 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5249 * @inode: inode we're writing to
5250 * @num_bytes: the number of bytes we want to allocate
5252 * This will do the following things
5254 * o reserve space in the data space info for num_bytes
5255 * o reserve space in the metadata space info based on number of outstanding
5256 * extents and how much csums will be needed
5257 * o add to the inodes ->delalloc_bytes
5258 * o add it to the fs_info's delalloc inodes list.
5260 * This will return 0 for success and -ENOSPC if there is no space left.
5262 int btrfs_delalloc_reserve_space(struct inode
*inode
, u64 num_bytes
)
5266 ret
= btrfs_check_data_free_space(inode
, num_bytes
);
5270 ret
= btrfs_delalloc_reserve_metadata(inode
, num_bytes
);
5272 btrfs_free_reserved_data_space(inode
, num_bytes
);
5280 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5281 * @inode: inode we're releasing space for
5282 * @num_bytes: the number of bytes we want to free up
5284 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5285 * called in the case that we don't need the metadata AND data reservations
5286 * anymore. So if there is an error or we insert an inline extent.
5288 * This function will release the metadata space that was not used and will
5289 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5290 * list if there are no delalloc bytes left.
5292 void btrfs_delalloc_release_space(struct inode
*inode
, u64 num_bytes
)
5294 btrfs_delalloc_release_metadata(inode
, num_bytes
);
5295 btrfs_free_reserved_data_space(inode
, num_bytes
);
5298 static int update_block_group(struct btrfs_trans_handle
*trans
,
5299 struct btrfs_root
*root
, u64 bytenr
,
5300 u64 num_bytes
, int alloc
)
5302 struct btrfs_block_group_cache
*cache
= NULL
;
5303 struct btrfs_fs_info
*info
= root
->fs_info
;
5304 u64 total
= num_bytes
;
5309 /* block accounting for super block */
5310 spin_lock(&info
->delalloc_root_lock
);
5311 old_val
= btrfs_super_bytes_used(info
->super_copy
);
5313 old_val
+= num_bytes
;
5315 old_val
-= num_bytes
;
5316 btrfs_set_super_bytes_used(info
->super_copy
, old_val
);
5317 spin_unlock(&info
->delalloc_root_lock
);
5320 cache
= btrfs_lookup_block_group(info
, bytenr
);
5323 if (cache
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
5324 BTRFS_BLOCK_GROUP_RAID1
|
5325 BTRFS_BLOCK_GROUP_RAID10
))
5330 * If this block group has free space cache written out, we
5331 * need to make sure to load it if we are removing space. This
5332 * is because we need the unpinning stage to actually add the
5333 * space back to the block group, otherwise we will leak space.
5335 if (!alloc
&& cache
->cached
== BTRFS_CACHE_NO
)
5336 cache_block_group(cache
, 1);
5338 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
5339 if (list_empty(&cache
->dirty_list
)) {
5340 list_add_tail(&cache
->dirty_list
,
5341 &trans
->transaction
->dirty_bgs
);
5342 btrfs_get_block_group(cache
);
5344 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
5346 byte_in_group
= bytenr
- cache
->key
.objectid
;
5347 WARN_ON(byte_in_group
> cache
->key
.offset
);
5349 spin_lock(&cache
->space_info
->lock
);
5350 spin_lock(&cache
->lock
);
5352 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
5353 cache
->disk_cache_state
< BTRFS_DC_CLEAR
)
5354 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
5356 old_val
= btrfs_block_group_used(&cache
->item
);
5357 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
5359 old_val
+= num_bytes
;
5360 btrfs_set_block_group_used(&cache
->item
, old_val
);
5361 cache
->reserved
-= num_bytes
;
5362 cache
->space_info
->bytes_reserved
-= num_bytes
;
5363 cache
->space_info
->bytes_used
+= num_bytes
;
5364 cache
->space_info
->disk_used
+= num_bytes
* factor
;
5365 spin_unlock(&cache
->lock
);
5366 spin_unlock(&cache
->space_info
->lock
);
5368 old_val
-= num_bytes
;
5369 btrfs_set_block_group_used(&cache
->item
, old_val
);
5370 cache
->pinned
+= num_bytes
;
5371 cache
->space_info
->bytes_pinned
+= num_bytes
;
5372 cache
->space_info
->bytes_used
-= num_bytes
;
5373 cache
->space_info
->disk_used
-= num_bytes
* factor
;
5374 spin_unlock(&cache
->lock
);
5375 spin_unlock(&cache
->space_info
->lock
);
5377 set_extent_dirty(info
->pinned_extents
,
5378 bytenr
, bytenr
+ num_bytes
- 1,
5379 GFP_NOFS
| __GFP_NOFAIL
);
5381 * No longer have used bytes in this block group, queue
5385 spin_lock(&info
->unused_bgs_lock
);
5386 if (list_empty(&cache
->bg_list
)) {
5387 btrfs_get_block_group(cache
);
5388 list_add_tail(&cache
->bg_list
,
5391 spin_unlock(&info
->unused_bgs_lock
);
5394 btrfs_put_block_group(cache
);
5396 bytenr
+= num_bytes
;
5401 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
5403 struct btrfs_block_group_cache
*cache
;
5406 spin_lock(&root
->fs_info
->block_group_cache_lock
);
5407 bytenr
= root
->fs_info
->first_logical_byte
;
5408 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
5410 if (bytenr
< (u64
)-1)
5413 cache
= btrfs_lookup_first_block_group(root
->fs_info
, search_start
);
5417 bytenr
= cache
->key
.objectid
;
5418 btrfs_put_block_group(cache
);
5423 static int pin_down_extent(struct btrfs_root
*root
,
5424 struct btrfs_block_group_cache
*cache
,
5425 u64 bytenr
, u64 num_bytes
, int reserved
)
5427 spin_lock(&cache
->space_info
->lock
);
5428 spin_lock(&cache
->lock
);
5429 cache
->pinned
+= num_bytes
;
5430 cache
->space_info
->bytes_pinned
+= num_bytes
;
5432 cache
->reserved
-= num_bytes
;
5433 cache
->space_info
->bytes_reserved
-= num_bytes
;
5435 spin_unlock(&cache
->lock
);
5436 spin_unlock(&cache
->space_info
->lock
);
5438 set_extent_dirty(root
->fs_info
->pinned_extents
, bytenr
,
5439 bytenr
+ num_bytes
- 1, GFP_NOFS
| __GFP_NOFAIL
);
5441 trace_btrfs_reserved_extent_free(root
, bytenr
, num_bytes
);
5446 * this function must be called within transaction
5448 int btrfs_pin_extent(struct btrfs_root
*root
,
5449 u64 bytenr
, u64 num_bytes
, int reserved
)
5451 struct btrfs_block_group_cache
*cache
;
5453 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
5454 BUG_ON(!cache
); /* Logic error */
5456 pin_down_extent(root
, cache
, bytenr
, num_bytes
, reserved
);
5458 btrfs_put_block_group(cache
);
5463 * this function must be called within transaction
5465 int btrfs_pin_extent_for_log_replay(struct btrfs_root
*root
,
5466 u64 bytenr
, u64 num_bytes
)
5468 struct btrfs_block_group_cache
*cache
;
5471 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
5476 * pull in the free space cache (if any) so that our pin
5477 * removes the free space from the cache. We have load_only set
5478 * to one because the slow code to read in the free extents does check
5479 * the pinned extents.
5481 cache_block_group(cache
, 1);
5483 pin_down_extent(root
, cache
, bytenr
, num_bytes
, 0);
5485 /* remove us from the free space cache (if we're there at all) */
5486 ret
= btrfs_remove_free_space(cache
, bytenr
, num_bytes
);
5487 btrfs_put_block_group(cache
);
5491 static int __exclude_logged_extent(struct btrfs_root
*root
, u64 start
, u64 num_bytes
)
5494 struct btrfs_block_group_cache
*block_group
;
5495 struct btrfs_caching_control
*caching_ctl
;
5497 block_group
= btrfs_lookup_block_group(root
->fs_info
, start
);
5501 cache_block_group(block_group
, 0);
5502 caching_ctl
= get_caching_control(block_group
);
5506 BUG_ON(!block_group_cache_done(block_group
));
5507 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
5509 mutex_lock(&caching_ctl
->mutex
);
5511 if (start
>= caching_ctl
->progress
) {
5512 ret
= add_excluded_extent(root
, start
, num_bytes
);
5513 } else if (start
+ num_bytes
<= caching_ctl
->progress
) {
5514 ret
= btrfs_remove_free_space(block_group
,
5517 num_bytes
= caching_ctl
->progress
- start
;
5518 ret
= btrfs_remove_free_space(block_group
,
5523 num_bytes
= (start
+ num_bytes
) -
5524 caching_ctl
->progress
;
5525 start
= caching_ctl
->progress
;
5526 ret
= add_excluded_extent(root
, start
, num_bytes
);
5529 mutex_unlock(&caching_ctl
->mutex
);
5530 put_caching_control(caching_ctl
);
5532 btrfs_put_block_group(block_group
);
5536 int btrfs_exclude_logged_extents(struct btrfs_root
*log
,
5537 struct extent_buffer
*eb
)
5539 struct btrfs_file_extent_item
*item
;
5540 struct btrfs_key key
;
5544 if (!btrfs_fs_incompat(log
->fs_info
, MIXED_GROUPS
))
5547 for (i
= 0; i
< btrfs_header_nritems(eb
); i
++) {
5548 btrfs_item_key_to_cpu(eb
, &key
, i
);
5549 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
5551 item
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
5552 found_type
= btrfs_file_extent_type(eb
, item
);
5553 if (found_type
== BTRFS_FILE_EXTENT_INLINE
)
5555 if (btrfs_file_extent_disk_bytenr(eb
, item
) == 0)
5557 key
.objectid
= btrfs_file_extent_disk_bytenr(eb
, item
);
5558 key
.offset
= btrfs_file_extent_disk_num_bytes(eb
, item
);
5559 __exclude_logged_extent(log
, key
.objectid
, key
.offset
);
5566 * btrfs_update_reserved_bytes - update the block_group and space info counters
5567 * @cache: The cache we are manipulating
5568 * @num_bytes: The number of bytes in question
5569 * @reserve: One of the reservation enums
5570 * @delalloc: The blocks are allocated for the delalloc write
5572 * This is called by the allocator when it reserves space, or by somebody who is
5573 * freeing space that was never actually used on disk. For example if you
5574 * reserve some space for a new leaf in transaction A and before transaction A
5575 * commits you free that leaf, you call this with reserve set to 0 in order to
5576 * clear the reservation.
5578 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5579 * ENOSPC accounting. For data we handle the reservation through clearing the
5580 * delalloc bits in the io_tree. We have to do this since we could end up
5581 * allocating less disk space for the amount of data we have reserved in the
5582 * case of compression.
5584 * If this is a reservation and the block group has become read only we cannot
5585 * make the reservation and return -EAGAIN, otherwise this function always
5588 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
5589 u64 num_bytes
, int reserve
, int delalloc
)
5591 struct btrfs_space_info
*space_info
= cache
->space_info
;
5594 spin_lock(&space_info
->lock
);
5595 spin_lock(&cache
->lock
);
5596 if (reserve
!= RESERVE_FREE
) {
5600 cache
->reserved
+= num_bytes
;
5601 space_info
->bytes_reserved
+= num_bytes
;
5602 if (reserve
== RESERVE_ALLOC
) {
5603 trace_btrfs_space_reservation(cache
->fs_info
,
5604 "space_info", space_info
->flags
,
5606 space_info
->bytes_may_use
-= num_bytes
;
5610 cache
->delalloc_bytes
+= num_bytes
;
5614 space_info
->bytes_readonly
+= num_bytes
;
5615 cache
->reserved
-= num_bytes
;
5616 space_info
->bytes_reserved
-= num_bytes
;
5619 cache
->delalloc_bytes
-= num_bytes
;
5621 spin_unlock(&cache
->lock
);
5622 spin_unlock(&space_info
->lock
);
5626 void btrfs_prepare_extent_commit(struct btrfs_trans_handle
*trans
,
5627 struct btrfs_root
*root
)
5629 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5630 struct btrfs_caching_control
*next
;
5631 struct btrfs_caching_control
*caching_ctl
;
5632 struct btrfs_block_group_cache
*cache
;
5634 down_write(&fs_info
->commit_root_sem
);
5636 list_for_each_entry_safe(caching_ctl
, next
,
5637 &fs_info
->caching_block_groups
, list
) {
5638 cache
= caching_ctl
->block_group
;
5639 if (block_group_cache_done(cache
)) {
5640 cache
->last_byte_to_unpin
= (u64
)-1;
5641 list_del_init(&caching_ctl
->list
);
5642 put_caching_control(caching_ctl
);
5644 cache
->last_byte_to_unpin
= caching_ctl
->progress
;
5648 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
5649 fs_info
->pinned_extents
= &fs_info
->freed_extents
[1];
5651 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
5653 up_write(&fs_info
->commit_root_sem
);
5655 update_global_block_rsv(fs_info
);
5658 static int unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
,
5659 const bool return_free_space
)
5661 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5662 struct btrfs_block_group_cache
*cache
= NULL
;
5663 struct btrfs_space_info
*space_info
;
5664 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5668 while (start
<= end
) {
5671 start
>= cache
->key
.objectid
+ cache
->key
.offset
) {
5673 btrfs_put_block_group(cache
);
5674 cache
= btrfs_lookup_block_group(fs_info
, start
);
5675 BUG_ON(!cache
); /* Logic error */
5678 len
= cache
->key
.objectid
+ cache
->key
.offset
- start
;
5679 len
= min(len
, end
+ 1 - start
);
5681 if (start
< cache
->last_byte_to_unpin
) {
5682 len
= min(len
, cache
->last_byte_to_unpin
- start
);
5683 if (return_free_space
)
5684 btrfs_add_free_space(cache
, start
, len
);
5688 space_info
= cache
->space_info
;
5690 spin_lock(&space_info
->lock
);
5691 spin_lock(&cache
->lock
);
5692 cache
->pinned
-= len
;
5693 space_info
->bytes_pinned
-= len
;
5694 percpu_counter_add(&space_info
->total_bytes_pinned
, -len
);
5696 space_info
->bytes_readonly
+= len
;
5699 spin_unlock(&cache
->lock
);
5700 if (!readonly
&& global_rsv
->space_info
== space_info
) {
5701 spin_lock(&global_rsv
->lock
);
5702 if (!global_rsv
->full
) {
5703 len
= min(len
, global_rsv
->size
-
5704 global_rsv
->reserved
);
5705 global_rsv
->reserved
+= len
;
5706 space_info
->bytes_may_use
+= len
;
5707 if (global_rsv
->reserved
>= global_rsv
->size
)
5708 global_rsv
->full
= 1;
5710 spin_unlock(&global_rsv
->lock
);
5712 spin_unlock(&space_info
->lock
);
5716 btrfs_put_block_group(cache
);
5720 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
5721 struct btrfs_root
*root
)
5723 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5724 struct extent_io_tree
*unpin
;
5732 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
5733 unpin
= &fs_info
->freed_extents
[1];
5735 unpin
= &fs_info
->freed_extents
[0];
5738 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
5739 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
5740 EXTENT_DIRTY
, NULL
);
5742 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
5746 if (btrfs_test_opt(root
, DISCARD
))
5747 ret
= btrfs_discard_extent(root
, start
,
5748 end
+ 1 - start
, NULL
);
5750 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
5751 unpin_extent_range(root
, start
, end
, true);
5752 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
5759 static void add_pinned_bytes(struct btrfs_fs_info
*fs_info
, u64 num_bytes
,
5760 u64 owner
, u64 root_objectid
)
5762 struct btrfs_space_info
*space_info
;
5765 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
5766 if (root_objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
5767 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
5769 flags
= BTRFS_BLOCK_GROUP_METADATA
;
5771 flags
= BTRFS_BLOCK_GROUP_DATA
;
5774 space_info
= __find_space_info(fs_info
, flags
);
5775 BUG_ON(!space_info
); /* Logic bug */
5776 percpu_counter_add(&space_info
->total_bytes_pinned
, num_bytes
);
5780 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
5781 struct btrfs_root
*root
,
5782 u64 bytenr
, u64 num_bytes
, u64 parent
,
5783 u64 root_objectid
, u64 owner_objectid
,
5784 u64 owner_offset
, int refs_to_drop
,
5785 struct btrfs_delayed_extent_op
*extent_op
,
5788 struct btrfs_key key
;
5789 struct btrfs_path
*path
;
5790 struct btrfs_fs_info
*info
= root
->fs_info
;
5791 struct btrfs_root
*extent_root
= info
->extent_root
;
5792 struct extent_buffer
*leaf
;
5793 struct btrfs_extent_item
*ei
;
5794 struct btrfs_extent_inline_ref
*iref
;
5797 int extent_slot
= 0;
5798 int found_extent
= 0;
5803 enum btrfs_qgroup_operation_type type
= BTRFS_QGROUP_OPER_SUB_EXCL
;
5804 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
5807 if (!info
->quota_enabled
|| !is_fstree(root_objectid
))
5810 path
= btrfs_alloc_path();
5815 path
->leave_spinning
= 1;
5817 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
5818 BUG_ON(!is_data
&& refs_to_drop
!= 1);
5821 skinny_metadata
= 0;
5823 ret
= lookup_extent_backref(trans
, extent_root
, path
, &iref
,
5824 bytenr
, num_bytes
, parent
,
5825 root_objectid
, owner_objectid
,
5828 extent_slot
= path
->slots
[0];
5829 while (extent_slot
>= 0) {
5830 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
5832 if (key
.objectid
!= bytenr
)
5834 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
5835 key
.offset
== num_bytes
) {
5839 if (key
.type
== BTRFS_METADATA_ITEM_KEY
&&
5840 key
.offset
== owner_objectid
) {
5844 if (path
->slots
[0] - extent_slot
> 5)
5848 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5849 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
5850 if (found_extent
&& item_size
< sizeof(*ei
))
5853 if (!found_extent
) {
5855 ret
= remove_extent_backref(trans
, extent_root
, path
,
5857 is_data
, &last_ref
);
5859 btrfs_abort_transaction(trans
, extent_root
, ret
);
5862 btrfs_release_path(path
);
5863 path
->leave_spinning
= 1;
5865 key
.objectid
= bytenr
;
5866 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
5867 key
.offset
= num_bytes
;
5869 if (!is_data
&& skinny_metadata
) {
5870 key
.type
= BTRFS_METADATA_ITEM_KEY
;
5871 key
.offset
= owner_objectid
;
5874 ret
= btrfs_search_slot(trans
, extent_root
,
5876 if (ret
> 0 && skinny_metadata
&& path
->slots
[0]) {
5878 * Couldn't find our skinny metadata item,
5879 * see if we have ye olde extent item.
5882 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
5884 if (key
.objectid
== bytenr
&&
5885 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
5886 key
.offset
== num_bytes
)
5890 if (ret
> 0 && skinny_metadata
) {
5891 skinny_metadata
= false;
5892 key
.objectid
= bytenr
;
5893 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
5894 key
.offset
= num_bytes
;
5895 btrfs_release_path(path
);
5896 ret
= btrfs_search_slot(trans
, extent_root
,
5901 btrfs_err(info
, "umm, got %d back from search, was looking for %llu",
5904 btrfs_print_leaf(extent_root
,
5908 btrfs_abort_transaction(trans
, extent_root
, ret
);
5911 extent_slot
= path
->slots
[0];
5913 } else if (WARN_ON(ret
== -ENOENT
)) {
5914 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
5916 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
5917 bytenr
, parent
, root_objectid
, owner_objectid
,
5919 btrfs_abort_transaction(trans
, extent_root
, ret
);
5922 btrfs_abort_transaction(trans
, extent_root
, ret
);
5926 leaf
= path
->nodes
[0];
5927 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
5928 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5929 if (item_size
< sizeof(*ei
)) {
5930 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
5931 ret
= convert_extent_item_v0(trans
, extent_root
, path
,
5934 btrfs_abort_transaction(trans
, extent_root
, ret
);
5938 btrfs_release_path(path
);
5939 path
->leave_spinning
= 1;
5941 key
.objectid
= bytenr
;
5942 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
5943 key
.offset
= num_bytes
;
5945 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
5948 btrfs_err(info
, "umm, got %d back from search, was looking for %llu",
5950 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
5953 btrfs_abort_transaction(trans
, extent_root
, ret
);
5957 extent_slot
= path
->slots
[0];
5958 leaf
= path
->nodes
[0];
5959 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
5962 BUG_ON(item_size
< sizeof(*ei
));
5963 ei
= btrfs_item_ptr(leaf
, extent_slot
,
5964 struct btrfs_extent_item
);
5965 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
5966 key
.type
== BTRFS_EXTENT_ITEM_KEY
) {
5967 struct btrfs_tree_block_info
*bi
;
5968 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
5969 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
5970 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
5973 refs
= btrfs_extent_refs(leaf
, ei
);
5974 if (refs
< refs_to_drop
) {
5975 btrfs_err(info
, "trying to drop %d refs but we only have %Lu "
5976 "for bytenr %Lu", refs_to_drop
, refs
, bytenr
);
5978 btrfs_abort_transaction(trans
, extent_root
, ret
);
5981 refs
-= refs_to_drop
;
5984 type
= BTRFS_QGROUP_OPER_SUB_SHARED
;
5986 __run_delayed_extent_op(extent_op
, leaf
, ei
);
5988 * In the case of inline back ref, reference count will
5989 * be updated by remove_extent_backref
5992 BUG_ON(!found_extent
);
5994 btrfs_set_extent_refs(leaf
, ei
, refs
);
5995 btrfs_mark_buffer_dirty(leaf
);
5998 ret
= remove_extent_backref(trans
, extent_root
, path
,
6000 is_data
, &last_ref
);
6002 btrfs_abort_transaction(trans
, extent_root
, ret
);
6006 add_pinned_bytes(root
->fs_info
, -num_bytes
, owner_objectid
,
6010 BUG_ON(is_data
&& refs_to_drop
!=
6011 extent_data_ref_count(root
, path
, iref
));
6013 BUG_ON(path
->slots
[0] != extent_slot
);
6015 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
6016 path
->slots
[0] = extent_slot
;
6022 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
6025 btrfs_abort_transaction(trans
, extent_root
, ret
);
6028 btrfs_release_path(path
);
6031 ret
= btrfs_del_csums(trans
, root
, bytenr
, num_bytes
);
6033 btrfs_abort_transaction(trans
, extent_root
, ret
);
6038 ret
= update_block_group(trans
, root
, bytenr
, num_bytes
, 0);
6040 btrfs_abort_transaction(trans
, extent_root
, ret
);
6044 btrfs_release_path(path
);
6046 /* Deal with the quota accounting */
6047 if (!ret
&& last_ref
&& !no_quota
) {
6050 if (owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
&&
6051 type
== BTRFS_QGROUP_OPER_SUB_SHARED
)
6054 ret
= btrfs_qgroup_record_ref(trans
, info
, root_objectid
,
6055 bytenr
, num_bytes
, type
,
6059 btrfs_free_path(path
);
6064 * when we free an block, it is possible (and likely) that we free the last
6065 * delayed ref for that extent as well. This searches the delayed ref tree for
6066 * a given extent, and if there are no other delayed refs to be processed, it
6067 * removes it from the tree.
6069 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
6070 struct btrfs_root
*root
, u64 bytenr
)
6072 struct btrfs_delayed_ref_head
*head
;
6073 struct btrfs_delayed_ref_root
*delayed_refs
;
6076 delayed_refs
= &trans
->transaction
->delayed_refs
;
6077 spin_lock(&delayed_refs
->lock
);
6078 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
6080 goto out_delayed_unlock
;
6082 spin_lock(&head
->lock
);
6083 if (rb_first(&head
->ref_root
))
6086 if (head
->extent_op
) {
6087 if (!head
->must_insert_reserved
)
6089 btrfs_free_delayed_extent_op(head
->extent_op
);
6090 head
->extent_op
= NULL
;
6094 * waiting for the lock here would deadlock. If someone else has it
6095 * locked they are already in the process of dropping it anyway
6097 if (!mutex_trylock(&head
->mutex
))
6101 * at this point we have a head with no other entries. Go
6102 * ahead and process it.
6104 head
->node
.in_tree
= 0;
6105 rb_erase(&head
->href_node
, &delayed_refs
->href_root
);
6107 atomic_dec(&delayed_refs
->num_entries
);
6110 * we don't take a ref on the node because we're removing it from the
6111 * tree, so we just steal the ref the tree was holding.
6113 delayed_refs
->num_heads
--;
6114 if (head
->processing
== 0)
6115 delayed_refs
->num_heads_ready
--;
6116 head
->processing
= 0;
6117 spin_unlock(&head
->lock
);
6118 spin_unlock(&delayed_refs
->lock
);
6120 BUG_ON(head
->extent_op
);
6121 if (head
->must_insert_reserved
)
6124 mutex_unlock(&head
->mutex
);
6125 btrfs_put_delayed_ref(&head
->node
);
6128 spin_unlock(&head
->lock
);
6131 spin_unlock(&delayed_refs
->lock
);
6135 void btrfs_free_tree_block(struct btrfs_trans_handle
*trans
,
6136 struct btrfs_root
*root
,
6137 struct extent_buffer
*buf
,
6138 u64 parent
, int last_ref
)
6143 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
6144 ret
= btrfs_add_delayed_tree_ref(root
->fs_info
, trans
,
6145 buf
->start
, buf
->len
,
6146 parent
, root
->root_key
.objectid
,
6147 btrfs_header_level(buf
),
6148 BTRFS_DROP_DELAYED_REF
, NULL
, 0);
6149 BUG_ON(ret
); /* -ENOMEM */
6155 if (btrfs_header_generation(buf
) == trans
->transid
) {
6156 struct btrfs_block_group_cache
*cache
;
6158 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
6159 ret
= check_ref_cleanup(trans
, root
, buf
->start
);
6164 cache
= btrfs_lookup_block_group(root
->fs_info
, buf
->start
);
6166 if (btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
6167 pin_down_extent(root
, cache
, buf
->start
, buf
->len
, 1);
6168 btrfs_put_block_group(cache
);
6172 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
));
6174 btrfs_add_free_space(cache
, buf
->start
, buf
->len
);
6175 btrfs_update_reserved_bytes(cache
, buf
->len
, RESERVE_FREE
, 0);
6176 btrfs_put_block_group(cache
);
6177 trace_btrfs_reserved_extent_free(root
, buf
->start
, buf
->len
);
6182 add_pinned_bytes(root
->fs_info
, buf
->len
,
6183 btrfs_header_level(buf
),
6184 root
->root_key
.objectid
);
6187 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6190 clear_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
);
6193 /* Can return -ENOMEM */
6194 int btrfs_free_extent(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
6195 u64 bytenr
, u64 num_bytes
, u64 parent
, u64 root_objectid
,
6196 u64 owner
, u64 offset
, int no_quota
)
6199 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6201 if (btrfs_test_is_dummy_root(root
))
6204 add_pinned_bytes(root
->fs_info
, num_bytes
, owner
, root_objectid
);
6207 * tree log blocks never actually go into the extent allocation
6208 * tree, just update pinning info and exit early.
6210 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
6211 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
6212 /* unlocks the pinned mutex */
6213 btrfs_pin_extent(root
, bytenr
, num_bytes
, 1);
6215 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
6216 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
6218 parent
, root_objectid
, (int)owner
,
6219 BTRFS_DROP_DELAYED_REF
, NULL
, no_quota
);
6221 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
6223 parent
, root_objectid
, owner
,
6224 offset
, BTRFS_DROP_DELAYED_REF
,
6231 * when we wait for progress in the block group caching, its because
6232 * our allocation attempt failed at least once. So, we must sleep
6233 * and let some progress happen before we try again.
6235 * This function will sleep at least once waiting for new free space to
6236 * show up, and then it will check the block group free space numbers
6237 * for our min num_bytes. Another option is to have it go ahead
6238 * and look in the rbtree for a free extent of a given size, but this
6241 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6242 * any of the information in this block group.
6244 static noinline
void
6245 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
6248 struct btrfs_caching_control
*caching_ctl
;
6250 caching_ctl
= get_caching_control(cache
);
6254 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
) ||
6255 (cache
->free_space_ctl
->free_space
>= num_bytes
));
6257 put_caching_control(caching_ctl
);
6261 wait_block_group_cache_done(struct btrfs_block_group_cache
*cache
)
6263 struct btrfs_caching_control
*caching_ctl
;
6266 caching_ctl
= get_caching_control(cache
);
6268 return (cache
->cached
== BTRFS_CACHE_ERROR
) ? -EIO
: 0;
6270 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
));
6271 if (cache
->cached
== BTRFS_CACHE_ERROR
)
6273 put_caching_control(caching_ctl
);
6277 int __get_raid_index(u64 flags
)
6279 if (flags
& BTRFS_BLOCK_GROUP_RAID10
)
6280 return BTRFS_RAID_RAID10
;
6281 else if (flags
& BTRFS_BLOCK_GROUP_RAID1
)
6282 return BTRFS_RAID_RAID1
;
6283 else if (flags
& BTRFS_BLOCK_GROUP_DUP
)
6284 return BTRFS_RAID_DUP
;
6285 else if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
6286 return BTRFS_RAID_RAID0
;
6287 else if (flags
& BTRFS_BLOCK_GROUP_RAID5
)
6288 return BTRFS_RAID_RAID5
;
6289 else if (flags
& BTRFS_BLOCK_GROUP_RAID6
)
6290 return BTRFS_RAID_RAID6
;
6292 return BTRFS_RAID_SINGLE
; /* BTRFS_BLOCK_GROUP_SINGLE */
6295 int get_block_group_index(struct btrfs_block_group_cache
*cache
)
6297 return __get_raid_index(cache
->flags
);
6300 static const char *btrfs_raid_type_names
[BTRFS_NR_RAID_TYPES
] = {
6301 [BTRFS_RAID_RAID10
] = "raid10",
6302 [BTRFS_RAID_RAID1
] = "raid1",
6303 [BTRFS_RAID_DUP
] = "dup",
6304 [BTRFS_RAID_RAID0
] = "raid0",
6305 [BTRFS_RAID_SINGLE
] = "single",
6306 [BTRFS_RAID_RAID5
] = "raid5",
6307 [BTRFS_RAID_RAID6
] = "raid6",
6310 static const char *get_raid_name(enum btrfs_raid_types type
)
6312 if (type
>= BTRFS_NR_RAID_TYPES
)
6315 return btrfs_raid_type_names
[type
];
6318 enum btrfs_loop_type
{
6319 LOOP_CACHING_NOWAIT
= 0,
6320 LOOP_CACHING_WAIT
= 1,
6321 LOOP_ALLOC_CHUNK
= 2,
6322 LOOP_NO_EMPTY_SIZE
= 3,
6326 btrfs_lock_block_group(struct btrfs_block_group_cache
*cache
,
6330 down_read(&cache
->data_rwsem
);
6334 btrfs_grab_block_group(struct btrfs_block_group_cache
*cache
,
6337 btrfs_get_block_group(cache
);
6339 down_read(&cache
->data_rwsem
);
6342 static struct btrfs_block_group_cache
*
6343 btrfs_lock_cluster(struct btrfs_block_group_cache
*block_group
,
6344 struct btrfs_free_cluster
*cluster
,
6347 struct btrfs_block_group_cache
*used_bg
;
6348 bool locked
= false;
6350 spin_lock(&cluster
->refill_lock
);
6352 if (used_bg
== cluster
->block_group
)
6355 up_read(&used_bg
->data_rwsem
);
6356 btrfs_put_block_group(used_bg
);
6359 used_bg
= cluster
->block_group
;
6363 if (used_bg
== block_group
)
6366 btrfs_get_block_group(used_bg
);
6371 if (down_read_trylock(&used_bg
->data_rwsem
))
6374 spin_unlock(&cluster
->refill_lock
);
6375 down_read(&used_bg
->data_rwsem
);
6381 btrfs_release_block_group(struct btrfs_block_group_cache
*cache
,
6385 up_read(&cache
->data_rwsem
);
6386 btrfs_put_block_group(cache
);
6390 * walks the btree of allocated extents and find a hole of a given size.
6391 * The key ins is changed to record the hole:
6392 * ins->objectid == start position
6393 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6394 * ins->offset == the size of the hole.
6395 * Any available blocks before search_start are skipped.
6397 * If there is no suitable free space, we will record the max size of
6398 * the free space extent currently.
6400 static noinline
int find_free_extent(struct btrfs_root
*orig_root
,
6401 u64 num_bytes
, u64 empty_size
,
6402 u64 hint_byte
, struct btrfs_key
*ins
,
6403 u64 flags
, int delalloc
)
6406 struct btrfs_root
*root
= orig_root
->fs_info
->extent_root
;
6407 struct btrfs_free_cluster
*last_ptr
= NULL
;
6408 struct btrfs_block_group_cache
*block_group
= NULL
;
6409 u64 search_start
= 0;
6410 u64 max_extent_size
= 0;
6411 int empty_cluster
= 2 * 1024 * 1024;
6412 struct btrfs_space_info
*space_info
;
6414 int index
= __get_raid_index(flags
);
6415 int alloc_type
= (flags
& BTRFS_BLOCK_GROUP_DATA
) ?
6416 RESERVE_ALLOC_NO_ACCOUNT
: RESERVE_ALLOC
;
6417 bool failed_cluster_refill
= false;
6418 bool failed_alloc
= false;
6419 bool use_cluster
= true;
6420 bool have_caching_bg
= false;
6422 WARN_ON(num_bytes
< root
->sectorsize
);
6423 ins
->type
= BTRFS_EXTENT_ITEM_KEY
;
6427 trace_find_free_extent(orig_root
, num_bytes
, empty_size
, flags
);
6429 space_info
= __find_space_info(root
->fs_info
, flags
);
6431 btrfs_err(root
->fs_info
, "No space info for %llu", flags
);
6436 * If the space info is for both data and metadata it means we have a
6437 * small filesystem and we can't use the clustering stuff.
6439 if (btrfs_mixed_space_info(space_info
))
6440 use_cluster
= false;
6442 if (flags
& BTRFS_BLOCK_GROUP_METADATA
&& use_cluster
) {
6443 last_ptr
= &root
->fs_info
->meta_alloc_cluster
;
6444 if (!btrfs_test_opt(root
, SSD
))
6445 empty_cluster
= 64 * 1024;
6448 if ((flags
& BTRFS_BLOCK_GROUP_DATA
) && use_cluster
&&
6449 btrfs_test_opt(root
, SSD
)) {
6450 last_ptr
= &root
->fs_info
->data_alloc_cluster
;
6454 spin_lock(&last_ptr
->lock
);
6455 if (last_ptr
->block_group
)
6456 hint_byte
= last_ptr
->window_start
;
6457 spin_unlock(&last_ptr
->lock
);
6460 search_start
= max(search_start
, first_logical_byte(root
, 0));
6461 search_start
= max(search_start
, hint_byte
);
6466 if (search_start
== hint_byte
) {
6467 block_group
= btrfs_lookup_block_group(root
->fs_info
,
6470 * we don't want to use the block group if it doesn't match our
6471 * allocation bits, or if its not cached.
6473 * However if we are re-searching with an ideal block group
6474 * picked out then we don't care that the block group is cached.
6476 if (block_group
&& block_group_bits(block_group
, flags
) &&
6477 block_group
->cached
!= BTRFS_CACHE_NO
) {
6478 down_read(&space_info
->groups_sem
);
6479 if (list_empty(&block_group
->list
) ||
6482 * someone is removing this block group,
6483 * we can't jump into the have_block_group
6484 * target because our list pointers are not
6487 btrfs_put_block_group(block_group
);
6488 up_read(&space_info
->groups_sem
);
6490 index
= get_block_group_index(block_group
);
6491 btrfs_lock_block_group(block_group
, delalloc
);
6492 goto have_block_group
;
6494 } else if (block_group
) {
6495 btrfs_put_block_group(block_group
);
6499 have_caching_bg
= false;
6500 down_read(&space_info
->groups_sem
);
6501 list_for_each_entry(block_group
, &space_info
->block_groups
[index
],
6506 btrfs_grab_block_group(block_group
, delalloc
);
6507 search_start
= block_group
->key
.objectid
;
6510 * this can happen if we end up cycling through all the
6511 * raid types, but we want to make sure we only allocate
6512 * for the proper type.
6514 if (!block_group_bits(block_group
, flags
)) {
6515 u64 extra
= BTRFS_BLOCK_GROUP_DUP
|
6516 BTRFS_BLOCK_GROUP_RAID1
|
6517 BTRFS_BLOCK_GROUP_RAID5
|
6518 BTRFS_BLOCK_GROUP_RAID6
|
6519 BTRFS_BLOCK_GROUP_RAID10
;
6522 * if they asked for extra copies and this block group
6523 * doesn't provide them, bail. This does allow us to
6524 * fill raid0 from raid1.
6526 if ((flags
& extra
) && !(block_group
->flags
& extra
))
6531 cached
= block_group_cache_done(block_group
);
6532 if (unlikely(!cached
)) {
6533 ret
= cache_block_group(block_group
, 0);
6538 if (unlikely(block_group
->cached
== BTRFS_CACHE_ERROR
))
6540 if (unlikely(block_group
->ro
))
6544 * Ok we want to try and use the cluster allocator, so
6548 struct btrfs_block_group_cache
*used_block_group
;
6549 unsigned long aligned_cluster
;
6551 * the refill lock keeps out other
6552 * people trying to start a new cluster
6554 used_block_group
= btrfs_lock_cluster(block_group
,
6557 if (!used_block_group
)
6558 goto refill_cluster
;
6560 if (used_block_group
!= block_group
&&
6561 (used_block_group
->ro
||
6562 !block_group_bits(used_block_group
, flags
)))
6563 goto release_cluster
;
6565 offset
= btrfs_alloc_from_cluster(used_block_group
,
6568 used_block_group
->key
.objectid
,
6571 /* we have a block, we're done */
6572 spin_unlock(&last_ptr
->refill_lock
);
6573 trace_btrfs_reserve_extent_cluster(root
,
6575 search_start
, num_bytes
);
6576 if (used_block_group
!= block_group
) {
6577 btrfs_release_block_group(block_group
,
6579 block_group
= used_block_group
;
6584 WARN_ON(last_ptr
->block_group
!= used_block_group
);
6586 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6587 * set up a new clusters, so lets just skip it
6588 * and let the allocator find whatever block
6589 * it can find. If we reach this point, we
6590 * will have tried the cluster allocator
6591 * plenty of times and not have found
6592 * anything, so we are likely way too
6593 * fragmented for the clustering stuff to find
6596 * However, if the cluster is taken from the
6597 * current block group, release the cluster
6598 * first, so that we stand a better chance of
6599 * succeeding in the unclustered
6601 if (loop
>= LOOP_NO_EMPTY_SIZE
&&
6602 used_block_group
!= block_group
) {
6603 spin_unlock(&last_ptr
->refill_lock
);
6604 btrfs_release_block_group(used_block_group
,
6606 goto unclustered_alloc
;
6610 * this cluster didn't work out, free it and
6613 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
6615 if (used_block_group
!= block_group
)
6616 btrfs_release_block_group(used_block_group
,
6619 if (loop
>= LOOP_NO_EMPTY_SIZE
) {
6620 spin_unlock(&last_ptr
->refill_lock
);
6621 goto unclustered_alloc
;
6624 aligned_cluster
= max_t(unsigned long,
6625 empty_cluster
+ empty_size
,
6626 block_group
->full_stripe_len
);
6628 /* allocate a cluster in this block group */
6629 ret
= btrfs_find_space_cluster(root
, block_group
,
6630 last_ptr
, search_start
,
6635 * now pull our allocation out of this
6638 offset
= btrfs_alloc_from_cluster(block_group
,
6644 /* we found one, proceed */
6645 spin_unlock(&last_ptr
->refill_lock
);
6646 trace_btrfs_reserve_extent_cluster(root
,
6647 block_group
, search_start
,
6651 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
6652 && !failed_cluster_refill
) {
6653 spin_unlock(&last_ptr
->refill_lock
);
6655 failed_cluster_refill
= true;
6656 wait_block_group_cache_progress(block_group
,
6657 num_bytes
+ empty_cluster
+ empty_size
);
6658 goto have_block_group
;
6662 * at this point we either didn't find a cluster
6663 * or we weren't able to allocate a block from our
6664 * cluster. Free the cluster we've been trying
6665 * to use, and go to the next block group
6667 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
6668 spin_unlock(&last_ptr
->refill_lock
);
6673 spin_lock(&block_group
->free_space_ctl
->tree_lock
);
6675 block_group
->free_space_ctl
->free_space
<
6676 num_bytes
+ empty_cluster
+ empty_size
) {
6677 if (block_group
->free_space_ctl
->free_space
>
6680 block_group
->free_space_ctl
->free_space
;
6681 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
6684 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
6686 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
6687 num_bytes
, empty_size
,
6690 * If we didn't find a chunk, and we haven't failed on this
6691 * block group before, and this block group is in the middle of
6692 * caching and we are ok with waiting, then go ahead and wait
6693 * for progress to be made, and set failed_alloc to true.
6695 * If failed_alloc is true then we've already waited on this
6696 * block group once and should move on to the next block group.
6698 if (!offset
&& !failed_alloc
&& !cached
&&
6699 loop
> LOOP_CACHING_NOWAIT
) {
6700 wait_block_group_cache_progress(block_group
,
6701 num_bytes
+ empty_size
);
6702 failed_alloc
= true;
6703 goto have_block_group
;
6704 } else if (!offset
) {
6706 have_caching_bg
= true;
6710 search_start
= ALIGN(offset
, root
->stripesize
);
6712 /* move on to the next group */
6713 if (search_start
+ num_bytes
>
6714 block_group
->key
.objectid
+ block_group
->key
.offset
) {
6715 btrfs_add_free_space(block_group
, offset
, num_bytes
);
6719 if (offset
< search_start
)
6720 btrfs_add_free_space(block_group
, offset
,
6721 search_start
- offset
);
6722 BUG_ON(offset
> search_start
);
6724 ret
= btrfs_update_reserved_bytes(block_group
, num_bytes
,
6725 alloc_type
, delalloc
);
6726 if (ret
== -EAGAIN
) {
6727 btrfs_add_free_space(block_group
, offset
, num_bytes
);
6731 /* we are all good, lets return */
6732 ins
->objectid
= search_start
;
6733 ins
->offset
= num_bytes
;
6735 trace_btrfs_reserve_extent(orig_root
, block_group
,
6736 search_start
, num_bytes
);
6737 btrfs_release_block_group(block_group
, delalloc
);
6740 failed_cluster_refill
= false;
6741 failed_alloc
= false;
6742 BUG_ON(index
!= get_block_group_index(block_group
));
6743 btrfs_release_block_group(block_group
, delalloc
);
6745 up_read(&space_info
->groups_sem
);
6747 if (!ins
->objectid
&& loop
>= LOOP_CACHING_WAIT
&& have_caching_bg
)
6750 if (!ins
->objectid
&& ++index
< BTRFS_NR_RAID_TYPES
)
6754 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
6755 * caching kthreads as we move along
6756 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
6757 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
6758 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
6761 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
) {
6764 if (loop
== LOOP_ALLOC_CHUNK
) {
6765 struct btrfs_trans_handle
*trans
;
6768 trans
= current
->journal_info
;
6772 trans
= btrfs_join_transaction(root
);
6774 if (IS_ERR(trans
)) {
6775 ret
= PTR_ERR(trans
);
6779 ret
= do_chunk_alloc(trans
, root
, flags
,
6782 * Do not bail out on ENOSPC since we
6783 * can do more things.
6785 if (ret
< 0 && ret
!= -ENOSPC
)
6786 btrfs_abort_transaction(trans
,
6791 btrfs_end_transaction(trans
, root
);
6796 if (loop
== LOOP_NO_EMPTY_SIZE
) {
6802 } else if (!ins
->objectid
) {
6804 } else if (ins
->objectid
) {
6809 ins
->offset
= max_extent_size
;
6813 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
6814 int dump_block_groups
)
6816 struct btrfs_block_group_cache
*cache
;
6819 spin_lock(&info
->lock
);
6820 printk(KERN_INFO
"BTRFS: space_info %llu has %llu free, is %sfull\n",
6822 info
->total_bytes
- info
->bytes_used
- info
->bytes_pinned
-
6823 info
->bytes_reserved
- info
->bytes_readonly
,
6824 (info
->full
) ? "" : "not ");
6825 printk(KERN_INFO
"BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
6826 "reserved=%llu, may_use=%llu, readonly=%llu\n",
6827 info
->total_bytes
, info
->bytes_used
, info
->bytes_pinned
,
6828 info
->bytes_reserved
, info
->bytes_may_use
,
6829 info
->bytes_readonly
);
6830 spin_unlock(&info
->lock
);
6832 if (!dump_block_groups
)
6835 down_read(&info
->groups_sem
);
6837 list_for_each_entry(cache
, &info
->block_groups
[index
], list
) {
6838 spin_lock(&cache
->lock
);
6839 printk(KERN_INFO
"BTRFS: "
6840 "block group %llu has %llu bytes, "
6841 "%llu used %llu pinned %llu reserved %s\n",
6842 cache
->key
.objectid
, cache
->key
.offset
,
6843 btrfs_block_group_used(&cache
->item
), cache
->pinned
,
6844 cache
->reserved
, cache
->ro
? "[readonly]" : "");
6845 btrfs_dump_free_space(cache
, bytes
);
6846 spin_unlock(&cache
->lock
);
6848 if (++index
< BTRFS_NR_RAID_TYPES
)
6850 up_read(&info
->groups_sem
);
6853 int btrfs_reserve_extent(struct btrfs_root
*root
,
6854 u64 num_bytes
, u64 min_alloc_size
,
6855 u64 empty_size
, u64 hint_byte
,
6856 struct btrfs_key
*ins
, int is_data
, int delalloc
)
6858 bool final_tried
= false;
6862 flags
= btrfs_get_alloc_profile(root
, is_data
);
6864 WARN_ON(num_bytes
< root
->sectorsize
);
6865 ret
= find_free_extent(root
, num_bytes
, empty_size
, hint_byte
, ins
,
6868 if (ret
== -ENOSPC
) {
6869 if (!final_tried
&& ins
->offset
) {
6870 num_bytes
= min(num_bytes
>> 1, ins
->offset
);
6871 num_bytes
= round_down(num_bytes
, root
->sectorsize
);
6872 num_bytes
= max(num_bytes
, min_alloc_size
);
6873 if (num_bytes
== min_alloc_size
)
6876 } else if (btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
6877 struct btrfs_space_info
*sinfo
;
6879 sinfo
= __find_space_info(root
->fs_info
, flags
);
6880 btrfs_err(root
->fs_info
, "allocation failed flags %llu, wanted %llu",
6883 dump_space_info(sinfo
, num_bytes
, 1);
6890 static int __btrfs_free_reserved_extent(struct btrfs_root
*root
,
6892 int pin
, int delalloc
)
6894 struct btrfs_block_group_cache
*cache
;
6897 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
6899 btrfs_err(root
->fs_info
, "Unable to find block group for %llu",
6904 if (btrfs_test_opt(root
, DISCARD
))
6905 ret
= btrfs_discard_extent(root
, start
, len
, NULL
);
6908 pin_down_extent(root
, cache
, start
, len
, 1);
6910 btrfs_add_free_space(cache
, start
, len
);
6911 btrfs_update_reserved_bytes(cache
, len
, RESERVE_FREE
, delalloc
);
6913 btrfs_put_block_group(cache
);
6915 trace_btrfs_reserved_extent_free(root
, start
, len
);
6920 int btrfs_free_reserved_extent(struct btrfs_root
*root
,
6921 u64 start
, u64 len
, int delalloc
)
6923 return __btrfs_free_reserved_extent(root
, start
, len
, 0, delalloc
);
6926 int btrfs_free_and_pin_reserved_extent(struct btrfs_root
*root
,
6929 return __btrfs_free_reserved_extent(root
, start
, len
, 1, 0);
6932 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
6933 struct btrfs_root
*root
,
6934 u64 parent
, u64 root_objectid
,
6935 u64 flags
, u64 owner
, u64 offset
,
6936 struct btrfs_key
*ins
, int ref_mod
)
6939 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6940 struct btrfs_extent_item
*extent_item
;
6941 struct btrfs_extent_inline_ref
*iref
;
6942 struct btrfs_path
*path
;
6943 struct extent_buffer
*leaf
;
6948 type
= BTRFS_SHARED_DATA_REF_KEY
;
6950 type
= BTRFS_EXTENT_DATA_REF_KEY
;
6952 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
6954 path
= btrfs_alloc_path();
6958 path
->leave_spinning
= 1;
6959 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
6962 btrfs_free_path(path
);
6966 leaf
= path
->nodes
[0];
6967 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
6968 struct btrfs_extent_item
);
6969 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
6970 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
6971 btrfs_set_extent_flags(leaf
, extent_item
,
6972 flags
| BTRFS_EXTENT_FLAG_DATA
);
6974 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
6975 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
6977 struct btrfs_shared_data_ref
*ref
;
6978 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
6979 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
6980 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
6982 struct btrfs_extent_data_ref
*ref
;
6983 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
6984 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
6985 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
6986 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
6987 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
6990 btrfs_mark_buffer_dirty(path
->nodes
[0]);
6991 btrfs_free_path(path
);
6993 /* Always set parent to 0 here since its exclusive anyway. */
6994 ret
= btrfs_qgroup_record_ref(trans
, fs_info
, root_objectid
,
6995 ins
->objectid
, ins
->offset
,
6996 BTRFS_QGROUP_OPER_ADD_EXCL
, 0);
7000 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1);
7001 if (ret
) { /* -ENOENT, logic error */
7002 btrfs_err(fs_info
, "update block group failed for %llu %llu",
7003 ins
->objectid
, ins
->offset
);
7006 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, ins
->offset
);
7010 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
7011 struct btrfs_root
*root
,
7012 u64 parent
, u64 root_objectid
,
7013 u64 flags
, struct btrfs_disk_key
*key
,
7014 int level
, struct btrfs_key
*ins
,
7018 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7019 struct btrfs_extent_item
*extent_item
;
7020 struct btrfs_tree_block_info
*block_info
;
7021 struct btrfs_extent_inline_ref
*iref
;
7022 struct btrfs_path
*path
;
7023 struct extent_buffer
*leaf
;
7024 u32 size
= sizeof(*extent_item
) + sizeof(*iref
);
7025 u64 num_bytes
= ins
->offset
;
7026 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
7029 if (!skinny_metadata
)
7030 size
+= sizeof(*block_info
);
7032 path
= btrfs_alloc_path();
7034 btrfs_free_and_pin_reserved_extent(root
, ins
->objectid
,
7039 path
->leave_spinning
= 1;
7040 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
7043 btrfs_free_and_pin_reserved_extent(root
, ins
->objectid
,
7045 btrfs_free_path(path
);
7049 leaf
= path
->nodes
[0];
7050 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
7051 struct btrfs_extent_item
);
7052 btrfs_set_extent_refs(leaf
, extent_item
, 1);
7053 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
7054 btrfs_set_extent_flags(leaf
, extent_item
,
7055 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
7057 if (skinny_metadata
) {
7058 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
7059 num_bytes
= root
->nodesize
;
7061 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
7062 btrfs_set_tree_block_key(leaf
, block_info
, key
);
7063 btrfs_set_tree_block_level(leaf
, block_info
, level
);
7064 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
7068 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
7069 btrfs_set_extent_inline_ref_type(leaf
, iref
,
7070 BTRFS_SHARED_BLOCK_REF_KEY
);
7071 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
7073 btrfs_set_extent_inline_ref_type(leaf
, iref
,
7074 BTRFS_TREE_BLOCK_REF_KEY
);
7075 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
7078 btrfs_mark_buffer_dirty(leaf
);
7079 btrfs_free_path(path
);
7082 ret
= btrfs_qgroup_record_ref(trans
, fs_info
, root_objectid
,
7083 ins
->objectid
, num_bytes
,
7084 BTRFS_QGROUP_OPER_ADD_EXCL
, 0);
7089 ret
= update_block_group(trans
, root
, ins
->objectid
, root
->nodesize
,
7091 if (ret
) { /* -ENOENT, logic error */
7092 btrfs_err(fs_info
, "update block group failed for %llu %llu",
7093 ins
->objectid
, ins
->offset
);
7097 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, root
->nodesize
);
7101 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
7102 struct btrfs_root
*root
,
7103 u64 root_objectid
, u64 owner
,
7104 u64 offset
, struct btrfs_key
*ins
)
7108 BUG_ON(root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
7110 ret
= btrfs_add_delayed_data_ref(root
->fs_info
, trans
, ins
->objectid
,
7112 root_objectid
, owner
, offset
,
7113 BTRFS_ADD_DELAYED_EXTENT
, NULL
, 0);
7118 * this is used by the tree logging recovery code. It records that
7119 * an extent has been allocated and makes sure to clear the free
7120 * space cache bits as well
7122 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
7123 struct btrfs_root
*root
,
7124 u64 root_objectid
, u64 owner
, u64 offset
,
7125 struct btrfs_key
*ins
)
7128 struct btrfs_block_group_cache
*block_group
;
7131 * Mixed block groups will exclude before processing the log so we only
7132 * need to do the exlude dance if this fs isn't mixed.
7134 if (!btrfs_fs_incompat(root
->fs_info
, MIXED_GROUPS
)) {
7135 ret
= __exclude_logged_extent(root
, ins
->objectid
, ins
->offset
);
7140 block_group
= btrfs_lookup_block_group(root
->fs_info
, ins
->objectid
);
7144 ret
= btrfs_update_reserved_bytes(block_group
, ins
->offset
,
7145 RESERVE_ALLOC_NO_ACCOUNT
, 0);
7146 BUG_ON(ret
); /* logic error */
7147 ret
= alloc_reserved_file_extent(trans
, root
, 0, root_objectid
,
7148 0, owner
, offset
, ins
, 1);
7149 btrfs_put_block_group(block_group
);
7153 static struct extent_buffer
*
7154 btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
7155 u64 bytenr
, int level
)
7157 struct extent_buffer
*buf
;
7159 buf
= btrfs_find_create_tree_block(root
, bytenr
);
7161 return ERR_PTR(-ENOMEM
);
7162 btrfs_set_header_generation(buf
, trans
->transid
);
7163 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, buf
, level
);
7164 btrfs_tree_lock(buf
);
7165 clean_tree_block(trans
, root
, buf
);
7166 clear_bit(EXTENT_BUFFER_STALE
, &buf
->bflags
);
7168 btrfs_set_lock_blocking(buf
);
7169 btrfs_set_buffer_uptodate(buf
);
7171 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
7172 buf
->log_index
= root
->log_transid
% 2;
7174 * we allow two log transactions at a time, use different
7175 * EXENT bit to differentiate dirty pages.
7177 if (buf
->log_index
== 0)
7178 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
7179 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
7181 set_extent_new(&root
->dirty_log_pages
, buf
->start
,
7182 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
7184 buf
->log_index
= -1;
7185 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
7186 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
7188 trans
->blocks_used
++;
7189 /* this returns a buffer locked for blocking */
7193 static struct btrfs_block_rsv
*
7194 use_block_rsv(struct btrfs_trans_handle
*trans
,
7195 struct btrfs_root
*root
, u32 blocksize
)
7197 struct btrfs_block_rsv
*block_rsv
;
7198 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
7200 bool global_updated
= false;
7202 block_rsv
= get_block_rsv(trans
, root
);
7204 if (unlikely(block_rsv
->size
== 0))
7207 ret
= block_rsv_use_bytes(block_rsv
, blocksize
);
7211 if (block_rsv
->failfast
)
7212 return ERR_PTR(ret
);
7214 if (block_rsv
->type
== BTRFS_BLOCK_RSV_GLOBAL
&& !global_updated
) {
7215 global_updated
= true;
7216 update_global_block_rsv(root
->fs_info
);
7220 if (btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
7221 static DEFINE_RATELIMIT_STATE(_rs
,
7222 DEFAULT_RATELIMIT_INTERVAL
* 10,
7223 /*DEFAULT_RATELIMIT_BURST*/ 1);
7224 if (__ratelimit(&_rs
))
7226 "BTRFS: block rsv returned %d\n", ret
);
7229 ret
= reserve_metadata_bytes(root
, block_rsv
, blocksize
,
7230 BTRFS_RESERVE_NO_FLUSH
);
7234 * If we couldn't reserve metadata bytes try and use some from
7235 * the global reserve if its space type is the same as the global
7238 if (block_rsv
->type
!= BTRFS_BLOCK_RSV_GLOBAL
&&
7239 block_rsv
->space_info
== global_rsv
->space_info
) {
7240 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
7244 return ERR_PTR(ret
);
7247 static void unuse_block_rsv(struct btrfs_fs_info
*fs_info
,
7248 struct btrfs_block_rsv
*block_rsv
, u32 blocksize
)
7250 block_rsv_add_bytes(block_rsv
, blocksize
, 0);
7251 block_rsv_release_bytes(fs_info
, block_rsv
, NULL
, 0);
7255 * finds a free extent and does all the dirty work required for allocation
7256 * returns the key for the extent through ins, and a tree buffer for
7257 * the first block of the extent through buf.
7259 * returns the tree buffer or NULL.
7261 struct extent_buffer
*btrfs_alloc_tree_block(struct btrfs_trans_handle
*trans
,
7262 struct btrfs_root
*root
,
7263 u64 parent
, u64 root_objectid
,
7264 struct btrfs_disk_key
*key
, int level
,
7265 u64 hint
, u64 empty_size
)
7267 struct btrfs_key ins
;
7268 struct btrfs_block_rsv
*block_rsv
;
7269 struct extent_buffer
*buf
;
7272 u32 blocksize
= root
->nodesize
;
7273 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
7276 if (btrfs_test_is_dummy_root(root
)) {
7277 buf
= btrfs_init_new_buffer(trans
, root
, root
->alloc_bytenr
,
7280 root
->alloc_bytenr
+= blocksize
;
7284 block_rsv
= use_block_rsv(trans
, root
, blocksize
);
7285 if (IS_ERR(block_rsv
))
7286 return ERR_CAST(block_rsv
);
7288 ret
= btrfs_reserve_extent(root
, blocksize
, blocksize
,
7289 empty_size
, hint
, &ins
, 0, 0);
7291 unuse_block_rsv(root
->fs_info
, block_rsv
, blocksize
);
7292 return ERR_PTR(ret
);
7295 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
, level
);
7296 BUG_ON(IS_ERR(buf
)); /* -ENOMEM */
7298 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
7300 parent
= ins
.objectid
;
7301 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
7305 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
7306 struct btrfs_delayed_extent_op
*extent_op
;
7307 extent_op
= btrfs_alloc_delayed_extent_op();
7308 BUG_ON(!extent_op
); /* -ENOMEM */
7310 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
7312 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
7313 extent_op
->flags_to_set
= flags
;
7314 if (skinny_metadata
)
7315 extent_op
->update_key
= 0;
7317 extent_op
->update_key
= 1;
7318 extent_op
->update_flags
= 1;
7319 extent_op
->is_data
= 0;
7320 extent_op
->level
= level
;
7322 ret
= btrfs_add_delayed_tree_ref(root
->fs_info
, trans
,
7324 ins
.offset
, parent
, root_objectid
,
7325 level
, BTRFS_ADD_DELAYED_EXTENT
,
7327 BUG_ON(ret
); /* -ENOMEM */
7332 struct walk_control
{
7333 u64 refs
[BTRFS_MAX_LEVEL
];
7334 u64 flags
[BTRFS_MAX_LEVEL
];
7335 struct btrfs_key update_progress
;
7346 #define DROP_REFERENCE 1
7347 #define UPDATE_BACKREF 2
7349 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
7350 struct btrfs_root
*root
,
7351 struct walk_control
*wc
,
7352 struct btrfs_path
*path
)
7360 struct btrfs_key key
;
7361 struct extent_buffer
*eb
;
7366 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
7367 wc
->reada_count
= wc
->reada_count
* 2 / 3;
7368 wc
->reada_count
= max(wc
->reada_count
, 2);
7370 wc
->reada_count
= wc
->reada_count
* 3 / 2;
7371 wc
->reada_count
= min_t(int, wc
->reada_count
,
7372 BTRFS_NODEPTRS_PER_BLOCK(root
));
7375 eb
= path
->nodes
[wc
->level
];
7376 nritems
= btrfs_header_nritems(eb
);
7377 blocksize
= root
->nodesize
;
7379 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
7380 if (nread
>= wc
->reada_count
)
7384 bytenr
= btrfs_node_blockptr(eb
, slot
);
7385 generation
= btrfs_node_ptr_generation(eb
, slot
);
7387 if (slot
== path
->slots
[wc
->level
])
7390 if (wc
->stage
== UPDATE_BACKREF
&&
7391 generation
<= root
->root_key
.offset
)
7394 /* We don't lock the tree block, it's OK to be racy here */
7395 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
,
7396 wc
->level
- 1, 1, &refs
,
7398 /* We don't care about errors in readahead. */
7403 if (wc
->stage
== DROP_REFERENCE
) {
7407 if (wc
->level
== 1 &&
7408 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
7410 if (!wc
->update_ref
||
7411 generation
<= root
->root_key
.offset
)
7413 btrfs_node_key_to_cpu(eb
, &key
, slot
);
7414 ret
= btrfs_comp_cpu_keys(&key
,
7415 &wc
->update_progress
);
7419 if (wc
->level
== 1 &&
7420 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
7424 readahead_tree_block(root
, bytenr
);
7427 wc
->reada_slot
= slot
;
7430 static int account_leaf_items(struct btrfs_trans_handle
*trans
,
7431 struct btrfs_root
*root
,
7432 struct extent_buffer
*eb
)
7434 int nr
= btrfs_header_nritems(eb
);
7435 int i
, extent_type
, ret
;
7436 struct btrfs_key key
;
7437 struct btrfs_file_extent_item
*fi
;
7438 u64 bytenr
, num_bytes
;
7440 for (i
= 0; i
< nr
; i
++) {
7441 btrfs_item_key_to_cpu(eb
, &key
, i
);
7443 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
7446 fi
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
7447 /* filter out non qgroup-accountable extents */
7448 extent_type
= btrfs_file_extent_type(eb
, fi
);
7450 if (extent_type
== BTRFS_FILE_EXTENT_INLINE
)
7453 bytenr
= btrfs_file_extent_disk_bytenr(eb
, fi
);
7457 num_bytes
= btrfs_file_extent_disk_num_bytes(eb
, fi
);
7459 ret
= btrfs_qgroup_record_ref(trans
, root
->fs_info
,
7462 BTRFS_QGROUP_OPER_SUB_SUBTREE
, 0);
7470 * Walk up the tree from the bottom, freeing leaves and any interior
7471 * nodes which have had all slots visited. If a node (leaf or
7472 * interior) is freed, the node above it will have it's slot
7473 * incremented. The root node will never be freed.
7475 * At the end of this function, we should have a path which has all
7476 * slots incremented to the next position for a search. If we need to
7477 * read a new node it will be NULL and the node above it will have the
7478 * correct slot selected for a later read.
7480 * If we increment the root nodes slot counter past the number of
7481 * elements, 1 is returned to signal completion of the search.
7483 static int adjust_slots_upwards(struct btrfs_root
*root
,
7484 struct btrfs_path
*path
, int root_level
)
7488 struct extent_buffer
*eb
;
7490 if (root_level
== 0)
7493 while (level
<= root_level
) {
7494 eb
= path
->nodes
[level
];
7495 nr
= btrfs_header_nritems(eb
);
7496 path
->slots
[level
]++;
7497 slot
= path
->slots
[level
];
7498 if (slot
>= nr
|| level
== 0) {
7500 * Don't free the root - we will detect this
7501 * condition after our loop and return a
7502 * positive value for caller to stop walking the tree.
7504 if (level
!= root_level
) {
7505 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
7506 path
->locks
[level
] = 0;
7508 free_extent_buffer(eb
);
7509 path
->nodes
[level
] = NULL
;
7510 path
->slots
[level
] = 0;
7514 * We have a valid slot to walk back down
7515 * from. Stop here so caller can process these
7524 eb
= path
->nodes
[root_level
];
7525 if (path
->slots
[root_level
] >= btrfs_header_nritems(eb
))
7532 * root_eb is the subtree root and is locked before this function is called.
7534 static int account_shared_subtree(struct btrfs_trans_handle
*trans
,
7535 struct btrfs_root
*root
,
7536 struct extent_buffer
*root_eb
,
7542 struct extent_buffer
*eb
= root_eb
;
7543 struct btrfs_path
*path
= NULL
;
7545 BUG_ON(root_level
< 0 || root_level
> BTRFS_MAX_LEVEL
);
7546 BUG_ON(root_eb
== NULL
);
7548 if (!root
->fs_info
->quota_enabled
)
7551 if (!extent_buffer_uptodate(root_eb
)) {
7552 ret
= btrfs_read_buffer(root_eb
, root_gen
);
7557 if (root_level
== 0) {
7558 ret
= account_leaf_items(trans
, root
, root_eb
);
7562 path
= btrfs_alloc_path();
7567 * Walk down the tree. Missing extent blocks are filled in as
7568 * we go. Metadata is accounted every time we read a new
7571 * When we reach a leaf, we account for file extent items in it,
7572 * walk back up the tree (adjusting slot pointers as we go)
7573 * and restart the search process.
7575 extent_buffer_get(root_eb
); /* For path */
7576 path
->nodes
[root_level
] = root_eb
;
7577 path
->slots
[root_level
] = 0;
7578 path
->locks
[root_level
] = 0; /* so release_path doesn't try to unlock */
7581 while (level
>= 0) {
7582 if (path
->nodes
[level
] == NULL
) {
7587 /* We need to get child blockptr/gen from
7588 * parent before we can read it. */
7589 eb
= path
->nodes
[level
+ 1];
7590 parent_slot
= path
->slots
[level
+ 1];
7591 child_bytenr
= btrfs_node_blockptr(eb
, parent_slot
);
7592 child_gen
= btrfs_node_ptr_generation(eb
, parent_slot
);
7594 eb
= read_tree_block(root
, child_bytenr
, child_gen
);
7595 if (!eb
|| !extent_buffer_uptodate(eb
)) {
7600 path
->nodes
[level
] = eb
;
7601 path
->slots
[level
] = 0;
7603 btrfs_tree_read_lock(eb
);
7604 btrfs_set_lock_blocking_rw(eb
, BTRFS_READ_LOCK
);
7605 path
->locks
[level
] = BTRFS_READ_LOCK_BLOCKING
;
7607 ret
= btrfs_qgroup_record_ref(trans
, root
->fs_info
,
7611 BTRFS_QGROUP_OPER_SUB_SUBTREE
,
7619 ret
= account_leaf_items(trans
, root
, path
->nodes
[level
]);
7623 /* Nonzero return here means we completed our search */
7624 ret
= adjust_slots_upwards(root
, path
, root_level
);
7628 /* Restart search with new slots */
7637 btrfs_free_path(path
);
7643 * helper to process tree block while walking down the tree.
7645 * when wc->stage == UPDATE_BACKREF, this function updates
7646 * back refs for pointers in the block.
7648 * NOTE: return value 1 means we should stop walking down.
7650 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
7651 struct btrfs_root
*root
,
7652 struct btrfs_path
*path
,
7653 struct walk_control
*wc
, int lookup_info
)
7655 int level
= wc
->level
;
7656 struct extent_buffer
*eb
= path
->nodes
[level
];
7657 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
7660 if (wc
->stage
== UPDATE_BACKREF
&&
7661 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
7665 * when reference count of tree block is 1, it won't increase
7666 * again. once full backref flag is set, we never clear it.
7669 ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
7670 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
)))) {
7671 BUG_ON(!path
->locks
[level
]);
7672 ret
= btrfs_lookup_extent_info(trans
, root
,
7673 eb
->start
, level
, 1,
7676 BUG_ON(ret
== -ENOMEM
);
7679 BUG_ON(wc
->refs
[level
] == 0);
7682 if (wc
->stage
== DROP_REFERENCE
) {
7683 if (wc
->refs
[level
] > 1)
7686 if (path
->locks
[level
] && !wc
->keep_locks
) {
7687 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
7688 path
->locks
[level
] = 0;
7693 /* wc->stage == UPDATE_BACKREF */
7694 if (!(wc
->flags
[level
] & flag
)) {
7695 BUG_ON(!path
->locks
[level
]);
7696 ret
= btrfs_inc_ref(trans
, root
, eb
, 1);
7697 BUG_ON(ret
); /* -ENOMEM */
7698 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
7699 BUG_ON(ret
); /* -ENOMEM */
7700 ret
= btrfs_set_disk_extent_flags(trans
, root
, eb
->start
,
7702 btrfs_header_level(eb
), 0);
7703 BUG_ON(ret
); /* -ENOMEM */
7704 wc
->flags
[level
] |= flag
;
7708 * the block is shared by multiple trees, so it's not good to
7709 * keep the tree lock
7711 if (path
->locks
[level
] && level
> 0) {
7712 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
7713 path
->locks
[level
] = 0;
7719 * helper to process tree block pointer.
7721 * when wc->stage == DROP_REFERENCE, this function checks
7722 * reference count of the block pointed to. if the block
7723 * is shared and we need update back refs for the subtree
7724 * rooted at the block, this function changes wc->stage to
7725 * UPDATE_BACKREF. if the block is shared and there is no
7726 * need to update back, this function drops the reference
7729 * NOTE: return value 1 means we should stop walking down.
7731 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
7732 struct btrfs_root
*root
,
7733 struct btrfs_path
*path
,
7734 struct walk_control
*wc
, int *lookup_info
)
7740 struct btrfs_key key
;
7741 struct extent_buffer
*next
;
7742 int level
= wc
->level
;
7745 bool need_account
= false;
7747 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
7748 path
->slots
[level
]);
7750 * if the lower level block was created before the snapshot
7751 * was created, we know there is no need to update back refs
7754 if (wc
->stage
== UPDATE_BACKREF
&&
7755 generation
<= root
->root_key
.offset
) {
7760 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
7761 blocksize
= root
->nodesize
;
7763 next
= btrfs_find_tree_block(root
, bytenr
);
7765 next
= btrfs_find_create_tree_block(root
, bytenr
);
7768 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, next
,
7772 btrfs_tree_lock(next
);
7773 btrfs_set_lock_blocking(next
);
7775 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, level
- 1, 1,
7776 &wc
->refs
[level
- 1],
7777 &wc
->flags
[level
- 1]);
7779 btrfs_tree_unlock(next
);
7783 if (unlikely(wc
->refs
[level
- 1] == 0)) {
7784 btrfs_err(root
->fs_info
, "Missing references.");
7789 if (wc
->stage
== DROP_REFERENCE
) {
7790 if (wc
->refs
[level
- 1] > 1) {
7791 need_account
= true;
7793 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
7796 if (!wc
->update_ref
||
7797 generation
<= root
->root_key
.offset
)
7800 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
7801 path
->slots
[level
]);
7802 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
7806 wc
->stage
= UPDATE_BACKREF
;
7807 wc
->shared_level
= level
- 1;
7811 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
7815 if (!btrfs_buffer_uptodate(next
, generation
, 0)) {
7816 btrfs_tree_unlock(next
);
7817 free_extent_buffer(next
);
7823 if (reada
&& level
== 1)
7824 reada_walk_down(trans
, root
, wc
, path
);
7825 next
= read_tree_block(root
, bytenr
, generation
);
7826 if (!next
|| !extent_buffer_uptodate(next
)) {
7827 free_extent_buffer(next
);
7830 btrfs_tree_lock(next
);
7831 btrfs_set_lock_blocking(next
);
7835 BUG_ON(level
!= btrfs_header_level(next
));
7836 path
->nodes
[level
] = next
;
7837 path
->slots
[level
] = 0;
7838 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
7844 wc
->refs
[level
- 1] = 0;
7845 wc
->flags
[level
- 1] = 0;
7846 if (wc
->stage
== DROP_REFERENCE
) {
7847 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
7848 parent
= path
->nodes
[level
]->start
;
7850 BUG_ON(root
->root_key
.objectid
!=
7851 btrfs_header_owner(path
->nodes
[level
]));
7856 ret
= account_shared_subtree(trans
, root
, next
,
7857 generation
, level
- 1);
7859 printk_ratelimited(KERN_ERR
"BTRFS: %s Error "
7860 "%d accounting shared subtree. Quota "
7861 "is out of sync, rescan required.\n",
7862 root
->fs_info
->sb
->s_id
, ret
);
7865 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
, parent
,
7866 root
->root_key
.objectid
, level
- 1, 0, 0);
7867 BUG_ON(ret
); /* -ENOMEM */
7869 btrfs_tree_unlock(next
);
7870 free_extent_buffer(next
);
7876 * helper to process tree block while walking up the tree.
7878 * when wc->stage == DROP_REFERENCE, this function drops
7879 * reference count on the block.
7881 * when wc->stage == UPDATE_BACKREF, this function changes
7882 * wc->stage back to DROP_REFERENCE if we changed wc->stage
7883 * to UPDATE_BACKREF previously while processing the block.
7885 * NOTE: return value 1 means we should stop walking up.
7887 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
7888 struct btrfs_root
*root
,
7889 struct btrfs_path
*path
,
7890 struct walk_control
*wc
)
7893 int level
= wc
->level
;
7894 struct extent_buffer
*eb
= path
->nodes
[level
];
7897 if (wc
->stage
== UPDATE_BACKREF
) {
7898 BUG_ON(wc
->shared_level
< level
);
7899 if (level
< wc
->shared_level
)
7902 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
7906 wc
->stage
= DROP_REFERENCE
;
7907 wc
->shared_level
= -1;
7908 path
->slots
[level
] = 0;
7911 * check reference count again if the block isn't locked.
7912 * we should start walking down the tree again if reference
7915 if (!path
->locks
[level
]) {
7917 btrfs_tree_lock(eb
);
7918 btrfs_set_lock_blocking(eb
);
7919 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
7921 ret
= btrfs_lookup_extent_info(trans
, root
,
7922 eb
->start
, level
, 1,
7926 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
7927 path
->locks
[level
] = 0;
7930 BUG_ON(wc
->refs
[level
] == 0);
7931 if (wc
->refs
[level
] == 1) {
7932 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
7933 path
->locks
[level
] = 0;
7939 /* wc->stage == DROP_REFERENCE */
7940 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
7942 if (wc
->refs
[level
] == 1) {
7944 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
7945 ret
= btrfs_dec_ref(trans
, root
, eb
, 1);
7947 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
7948 BUG_ON(ret
); /* -ENOMEM */
7949 ret
= account_leaf_items(trans
, root
, eb
);
7951 printk_ratelimited(KERN_ERR
"BTRFS: %s Error "
7952 "%d accounting leaf items. Quota "
7953 "is out of sync, rescan required.\n",
7954 root
->fs_info
->sb
->s_id
, ret
);
7957 /* make block locked assertion in clean_tree_block happy */
7958 if (!path
->locks
[level
] &&
7959 btrfs_header_generation(eb
) == trans
->transid
) {
7960 btrfs_tree_lock(eb
);
7961 btrfs_set_lock_blocking(eb
);
7962 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
7964 clean_tree_block(trans
, root
, eb
);
7967 if (eb
== root
->node
) {
7968 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
7971 BUG_ON(root
->root_key
.objectid
!=
7972 btrfs_header_owner(eb
));
7974 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
7975 parent
= path
->nodes
[level
+ 1]->start
;
7977 BUG_ON(root
->root_key
.objectid
!=
7978 btrfs_header_owner(path
->nodes
[level
+ 1]));
7981 btrfs_free_tree_block(trans
, root
, eb
, parent
, wc
->refs
[level
] == 1);
7983 wc
->refs
[level
] = 0;
7984 wc
->flags
[level
] = 0;
7988 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
7989 struct btrfs_root
*root
,
7990 struct btrfs_path
*path
,
7991 struct walk_control
*wc
)
7993 int level
= wc
->level
;
7994 int lookup_info
= 1;
7997 while (level
>= 0) {
7998 ret
= walk_down_proc(trans
, root
, path
, wc
, lookup_info
);
8005 if (path
->slots
[level
] >=
8006 btrfs_header_nritems(path
->nodes
[level
]))
8009 ret
= do_walk_down(trans
, root
, path
, wc
, &lookup_info
);
8011 path
->slots
[level
]++;
8020 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
8021 struct btrfs_root
*root
,
8022 struct btrfs_path
*path
,
8023 struct walk_control
*wc
, int max_level
)
8025 int level
= wc
->level
;
8028 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
8029 while (level
< max_level
&& path
->nodes
[level
]) {
8031 if (path
->slots
[level
] + 1 <
8032 btrfs_header_nritems(path
->nodes
[level
])) {
8033 path
->slots
[level
]++;
8036 ret
= walk_up_proc(trans
, root
, path
, wc
);
8040 if (path
->locks
[level
]) {
8041 btrfs_tree_unlock_rw(path
->nodes
[level
],
8042 path
->locks
[level
]);
8043 path
->locks
[level
] = 0;
8045 free_extent_buffer(path
->nodes
[level
]);
8046 path
->nodes
[level
] = NULL
;
8054 * drop a subvolume tree.
8056 * this function traverses the tree freeing any blocks that only
8057 * referenced by the tree.
8059 * when a shared tree block is found. this function decreases its
8060 * reference count by one. if update_ref is true, this function
8061 * also make sure backrefs for the shared block and all lower level
8062 * blocks are properly updated.
8064 * If called with for_reloc == 0, may exit early with -EAGAIN
8066 int btrfs_drop_snapshot(struct btrfs_root
*root
,
8067 struct btrfs_block_rsv
*block_rsv
, int update_ref
,
8070 struct btrfs_path
*path
;
8071 struct btrfs_trans_handle
*trans
;
8072 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
8073 struct btrfs_root_item
*root_item
= &root
->root_item
;
8074 struct walk_control
*wc
;
8075 struct btrfs_key key
;
8079 bool root_dropped
= false;
8081 btrfs_debug(root
->fs_info
, "Drop subvolume %llu", root
->objectid
);
8083 path
= btrfs_alloc_path();
8089 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
8091 btrfs_free_path(path
);
8096 trans
= btrfs_start_transaction(tree_root
, 0);
8097 if (IS_ERR(trans
)) {
8098 err
= PTR_ERR(trans
);
8103 trans
->block_rsv
= block_rsv
;
8105 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
8106 level
= btrfs_header_level(root
->node
);
8107 path
->nodes
[level
] = btrfs_lock_root_node(root
);
8108 btrfs_set_lock_blocking(path
->nodes
[level
]);
8109 path
->slots
[level
] = 0;
8110 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8111 memset(&wc
->update_progress
, 0,
8112 sizeof(wc
->update_progress
));
8114 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
8115 memcpy(&wc
->update_progress
, &key
,
8116 sizeof(wc
->update_progress
));
8118 level
= root_item
->drop_level
;
8120 path
->lowest_level
= level
;
8121 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
8122 path
->lowest_level
= 0;
8130 * unlock our path, this is safe because only this
8131 * function is allowed to delete this snapshot
8133 btrfs_unlock_up_safe(path
, 0);
8135 level
= btrfs_header_level(root
->node
);
8137 btrfs_tree_lock(path
->nodes
[level
]);
8138 btrfs_set_lock_blocking(path
->nodes
[level
]);
8139 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8141 ret
= btrfs_lookup_extent_info(trans
, root
,
8142 path
->nodes
[level
]->start
,
8143 level
, 1, &wc
->refs
[level
],
8149 BUG_ON(wc
->refs
[level
] == 0);
8151 if (level
== root_item
->drop_level
)
8154 btrfs_tree_unlock(path
->nodes
[level
]);
8155 path
->locks
[level
] = 0;
8156 WARN_ON(wc
->refs
[level
] != 1);
8162 wc
->shared_level
= -1;
8163 wc
->stage
= DROP_REFERENCE
;
8164 wc
->update_ref
= update_ref
;
8166 wc
->for_reloc
= for_reloc
;
8167 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
8171 ret
= walk_down_tree(trans
, root
, path
, wc
);
8177 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
8184 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
8188 if (wc
->stage
== DROP_REFERENCE
) {
8190 btrfs_node_key(path
->nodes
[level
],
8191 &root_item
->drop_progress
,
8192 path
->slots
[level
]);
8193 root_item
->drop_level
= level
;
8196 BUG_ON(wc
->level
== 0);
8197 if (btrfs_should_end_transaction(trans
, tree_root
) ||
8198 (!for_reloc
&& btrfs_need_cleaner_sleep(root
))) {
8199 ret
= btrfs_update_root(trans
, tree_root
,
8203 btrfs_abort_transaction(trans
, tree_root
, ret
);
8209 * Qgroup update accounting is run from
8210 * delayed ref handling. This usually works
8211 * out because delayed refs are normally the
8212 * only way qgroup updates are added. However,
8213 * we may have added updates during our tree
8214 * walk so run qgroups here to make sure we
8215 * don't lose any updates.
8217 ret
= btrfs_delayed_qgroup_accounting(trans
,
8220 printk_ratelimited(KERN_ERR
"BTRFS: Failure %d "
8221 "running qgroup updates "
8222 "during snapshot delete. "
8223 "Quota is out of sync, "
8224 "rescan required.\n", ret
);
8226 btrfs_end_transaction_throttle(trans
, tree_root
);
8227 if (!for_reloc
&& btrfs_need_cleaner_sleep(root
)) {
8228 pr_debug("BTRFS: drop snapshot early exit\n");
8233 trans
= btrfs_start_transaction(tree_root
, 0);
8234 if (IS_ERR(trans
)) {
8235 err
= PTR_ERR(trans
);
8239 trans
->block_rsv
= block_rsv
;
8242 btrfs_release_path(path
);
8246 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
8248 btrfs_abort_transaction(trans
, tree_root
, ret
);
8252 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
8253 ret
= btrfs_find_root(tree_root
, &root
->root_key
, path
,
8256 btrfs_abort_transaction(trans
, tree_root
, ret
);
8259 } else if (ret
> 0) {
8260 /* if we fail to delete the orphan item this time
8261 * around, it'll get picked up the next time.
8263 * The most common failure here is just -ENOENT.
8265 btrfs_del_orphan_item(trans
, tree_root
,
8266 root
->root_key
.objectid
);
8270 if (test_bit(BTRFS_ROOT_IN_RADIX
, &root
->state
)) {
8271 btrfs_drop_and_free_fs_root(tree_root
->fs_info
, root
);
8273 free_extent_buffer(root
->node
);
8274 free_extent_buffer(root
->commit_root
);
8275 btrfs_put_fs_root(root
);
8277 root_dropped
= true;
8279 ret
= btrfs_delayed_qgroup_accounting(trans
, tree_root
->fs_info
);
8281 printk_ratelimited(KERN_ERR
"BTRFS: Failure %d "
8282 "running qgroup updates "
8283 "during snapshot delete. "
8284 "Quota is out of sync, "
8285 "rescan required.\n", ret
);
8287 btrfs_end_transaction_throttle(trans
, tree_root
);
8290 btrfs_free_path(path
);
8293 * So if we need to stop dropping the snapshot for whatever reason we
8294 * need to make sure to add it back to the dead root list so that we
8295 * keep trying to do the work later. This also cleans up roots if we
8296 * don't have it in the radix (like when we recover after a power fail
8297 * or unmount) so we don't leak memory.
8299 if (!for_reloc
&& root_dropped
== false)
8300 btrfs_add_dead_root(root
);
8301 if (err
&& err
!= -EAGAIN
)
8302 btrfs_std_error(root
->fs_info
, err
);
8307 * drop subtree rooted at tree block 'node'.
8309 * NOTE: this function will unlock and release tree block 'node'
8310 * only used by relocation code
8312 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
8313 struct btrfs_root
*root
,
8314 struct extent_buffer
*node
,
8315 struct extent_buffer
*parent
)
8317 struct btrfs_path
*path
;
8318 struct walk_control
*wc
;
8324 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
8326 path
= btrfs_alloc_path();
8330 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
8332 btrfs_free_path(path
);
8336 btrfs_assert_tree_locked(parent
);
8337 parent_level
= btrfs_header_level(parent
);
8338 extent_buffer_get(parent
);
8339 path
->nodes
[parent_level
] = parent
;
8340 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
8342 btrfs_assert_tree_locked(node
);
8343 level
= btrfs_header_level(node
);
8344 path
->nodes
[level
] = node
;
8345 path
->slots
[level
] = 0;
8346 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8348 wc
->refs
[parent_level
] = 1;
8349 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
8351 wc
->shared_level
= -1;
8352 wc
->stage
= DROP_REFERENCE
;
8356 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
8359 wret
= walk_down_tree(trans
, root
, path
, wc
);
8365 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
8373 btrfs_free_path(path
);
8377 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
8383 * if restripe for this chunk_type is on pick target profile and
8384 * return, otherwise do the usual balance
8386 stripped
= get_restripe_target(root
->fs_info
, flags
);
8388 return extended_to_chunk(stripped
);
8390 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
8392 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
8393 BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
|
8394 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
8396 if (num_devices
== 1) {
8397 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
8398 stripped
= flags
& ~stripped
;
8400 /* turn raid0 into single device chunks */
8401 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
8404 /* turn mirroring into duplication */
8405 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
8406 BTRFS_BLOCK_GROUP_RAID10
))
8407 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
8409 /* they already had raid on here, just return */
8410 if (flags
& stripped
)
8413 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
8414 stripped
= flags
& ~stripped
;
8416 /* switch duplicated blocks with raid1 */
8417 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
8418 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
8420 /* this is drive concat, leave it alone */
8426 static int set_block_group_ro(struct btrfs_block_group_cache
*cache
, int force
)
8428 struct btrfs_space_info
*sinfo
= cache
->space_info
;
8430 u64 min_allocable_bytes
;
8435 * We need some metadata space and system metadata space for
8436 * allocating chunks in some corner cases until we force to set
8437 * it to be readonly.
8440 (BTRFS_BLOCK_GROUP_SYSTEM
| BTRFS_BLOCK_GROUP_METADATA
)) &&
8442 min_allocable_bytes
= 1 * 1024 * 1024;
8444 min_allocable_bytes
= 0;
8446 spin_lock(&sinfo
->lock
);
8447 spin_lock(&cache
->lock
);
8454 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
8455 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
8457 if (sinfo
->bytes_used
+ sinfo
->bytes_reserved
+ sinfo
->bytes_pinned
+
8458 sinfo
->bytes_may_use
+ sinfo
->bytes_readonly
+ num_bytes
+
8459 min_allocable_bytes
<= sinfo
->total_bytes
) {
8460 sinfo
->bytes_readonly
+= num_bytes
;
8462 list_add_tail(&cache
->ro_list
, &sinfo
->ro_bgs
);
8466 spin_unlock(&cache
->lock
);
8467 spin_unlock(&sinfo
->lock
);
8471 int btrfs_set_block_group_ro(struct btrfs_root
*root
,
8472 struct btrfs_block_group_cache
*cache
)
8475 struct btrfs_trans_handle
*trans
;
8481 trans
= btrfs_join_transaction(root
);
8483 return PTR_ERR(trans
);
8485 alloc_flags
= update_block_group_flags(root
, cache
->flags
);
8486 if (alloc_flags
!= cache
->flags
) {
8487 ret
= do_chunk_alloc(trans
, root
, alloc_flags
,
8493 ret
= set_block_group_ro(cache
, 0);
8496 alloc_flags
= get_alloc_profile(root
, cache
->space_info
->flags
);
8497 ret
= do_chunk_alloc(trans
, root
, alloc_flags
,
8501 ret
= set_block_group_ro(cache
, 0);
8503 btrfs_end_transaction(trans
, root
);
8507 int btrfs_force_chunk_alloc(struct btrfs_trans_handle
*trans
,
8508 struct btrfs_root
*root
, u64 type
)
8510 u64 alloc_flags
= get_alloc_profile(root
, type
);
8511 return do_chunk_alloc(trans
, root
, alloc_flags
,
8516 * helper to account the unused space of all the readonly block group in the
8517 * space_info. takes mirrors into account.
8519 u64
btrfs_account_ro_block_groups_free_space(struct btrfs_space_info
*sinfo
)
8521 struct btrfs_block_group_cache
*block_group
;
8525 /* It's df, we don't care if it's racey */
8526 if (list_empty(&sinfo
->ro_bgs
))
8529 spin_lock(&sinfo
->lock
);
8530 list_for_each_entry(block_group
, &sinfo
->ro_bgs
, ro_list
) {
8531 spin_lock(&block_group
->lock
);
8533 if (!block_group
->ro
) {
8534 spin_unlock(&block_group
->lock
);
8538 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_RAID1
|
8539 BTRFS_BLOCK_GROUP_RAID10
|
8540 BTRFS_BLOCK_GROUP_DUP
))
8545 free_bytes
+= (block_group
->key
.offset
-
8546 btrfs_block_group_used(&block_group
->item
)) *
8549 spin_unlock(&block_group
->lock
);
8551 spin_unlock(&sinfo
->lock
);
8556 void btrfs_set_block_group_rw(struct btrfs_root
*root
,
8557 struct btrfs_block_group_cache
*cache
)
8559 struct btrfs_space_info
*sinfo
= cache
->space_info
;
8564 spin_lock(&sinfo
->lock
);
8565 spin_lock(&cache
->lock
);
8566 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
8567 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
8568 sinfo
->bytes_readonly
-= num_bytes
;
8570 list_del_init(&cache
->ro_list
);
8571 spin_unlock(&cache
->lock
);
8572 spin_unlock(&sinfo
->lock
);
8576 * checks to see if its even possible to relocate this block group.
8578 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8579 * ok to go ahead and try.
8581 int btrfs_can_relocate(struct btrfs_root
*root
, u64 bytenr
)
8583 struct btrfs_block_group_cache
*block_group
;
8584 struct btrfs_space_info
*space_info
;
8585 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
8586 struct btrfs_device
*device
;
8587 struct btrfs_trans_handle
*trans
;
8596 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
8598 /* odd, couldn't find the block group, leave it alone */
8602 min_free
= btrfs_block_group_used(&block_group
->item
);
8604 /* no bytes used, we're good */
8608 space_info
= block_group
->space_info
;
8609 spin_lock(&space_info
->lock
);
8611 full
= space_info
->full
;
8614 * if this is the last block group we have in this space, we can't
8615 * relocate it unless we're able to allocate a new chunk below.
8617 * Otherwise, we need to make sure we have room in the space to handle
8618 * all of the extents from this block group. If we can, we're good
8620 if ((space_info
->total_bytes
!= block_group
->key
.offset
) &&
8621 (space_info
->bytes_used
+ space_info
->bytes_reserved
+
8622 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
8623 min_free
< space_info
->total_bytes
)) {
8624 spin_unlock(&space_info
->lock
);
8627 spin_unlock(&space_info
->lock
);
8630 * ok we don't have enough space, but maybe we have free space on our
8631 * devices to allocate new chunks for relocation, so loop through our
8632 * alloc devices and guess if we have enough space. if this block
8633 * group is going to be restriped, run checks against the target
8634 * profile instead of the current one.
8646 target
= get_restripe_target(root
->fs_info
, block_group
->flags
);
8648 index
= __get_raid_index(extended_to_chunk(target
));
8651 * this is just a balance, so if we were marked as full
8652 * we know there is no space for a new chunk
8657 index
= get_block_group_index(block_group
);
8660 if (index
== BTRFS_RAID_RAID10
) {
8664 } else if (index
== BTRFS_RAID_RAID1
) {
8666 } else if (index
== BTRFS_RAID_DUP
) {
8669 } else if (index
== BTRFS_RAID_RAID0
) {
8670 dev_min
= fs_devices
->rw_devices
;
8671 do_div(min_free
, dev_min
);
8674 /* We need to do this so that we can look at pending chunks */
8675 trans
= btrfs_join_transaction(root
);
8676 if (IS_ERR(trans
)) {
8677 ret
= PTR_ERR(trans
);
8681 mutex_lock(&root
->fs_info
->chunk_mutex
);
8682 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
8686 * check to make sure we can actually find a chunk with enough
8687 * space to fit our block group in.
8689 if (device
->total_bytes
> device
->bytes_used
+ min_free
&&
8690 !device
->is_tgtdev_for_dev_replace
) {
8691 ret
= find_free_dev_extent(trans
, device
, min_free
,
8696 if (dev_nr
>= dev_min
)
8702 mutex_unlock(&root
->fs_info
->chunk_mutex
);
8703 btrfs_end_transaction(trans
, root
);
8705 btrfs_put_block_group(block_group
);
8709 static int find_first_block_group(struct btrfs_root
*root
,
8710 struct btrfs_path
*path
, struct btrfs_key
*key
)
8713 struct btrfs_key found_key
;
8714 struct extent_buffer
*leaf
;
8717 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
8722 slot
= path
->slots
[0];
8723 leaf
= path
->nodes
[0];
8724 if (slot
>= btrfs_header_nritems(leaf
)) {
8725 ret
= btrfs_next_leaf(root
, path
);
8732 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
8734 if (found_key
.objectid
>= key
->objectid
&&
8735 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
8745 void btrfs_put_block_group_cache(struct btrfs_fs_info
*info
)
8747 struct btrfs_block_group_cache
*block_group
;
8751 struct inode
*inode
;
8753 block_group
= btrfs_lookup_first_block_group(info
, last
);
8754 while (block_group
) {
8755 spin_lock(&block_group
->lock
);
8756 if (block_group
->iref
)
8758 spin_unlock(&block_group
->lock
);
8759 block_group
= next_block_group(info
->tree_root
,
8769 inode
= block_group
->inode
;
8770 block_group
->iref
= 0;
8771 block_group
->inode
= NULL
;
8772 spin_unlock(&block_group
->lock
);
8774 last
= block_group
->key
.objectid
+ block_group
->key
.offset
;
8775 btrfs_put_block_group(block_group
);
8779 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
8781 struct btrfs_block_group_cache
*block_group
;
8782 struct btrfs_space_info
*space_info
;
8783 struct btrfs_caching_control
*caching_ctl
;
8786 down_write(&info
->commit_root_sem
);
8787 while (!list_empty(&info
->caching_block_groups
)) {
8788 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
8789 struct btrfs_caching_control
, list
);
8790 list_del(&caching_ctl
->list
);
8791 put_caching_control(caching_ctl
);
8793 up_write(&info
->commit_root_sem
);
8795 spin_lock(&info
->unused_bgs_lock
);
8796 while (!list_empty(&info
->unused_bgs
)) {
8797 block_group
= list_first_entry(&info
->unused_bgs
,
8798 struct btrfs_block_group_cache
,
8800 list_del_init(&block_group
->bg_list
);
8801 btrfs_put_block_group(block_group
);
8803 spin_unlock(&info
->unused_bgs_lock
);
8805 spin_lock(&info
->block_group_cache_lock
);
8806 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
8807 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
8809 rb_erase(&block_group
->cache_node
,
8810 &info
->block_group_cache_tree
);
8811 RB_CLEAR_NODE(&block_group
->cache_node
);
8812 spin_unlock(&info
->block_group_cache_lock
);
8814 down_write(&block_group
->space_info
->groups_sem
);
8815 list_del(&block_group
->list
);
8816 up_write(&block_group
->space_info
->groups_sem
);
8818 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
8819 wait_block_group_cache_done(block_group
);
8822 * We haven't cached this block group, which means we could
8823 * possibly have excluded extents on this block group.
8825 if (block_group
->cached
== BTRFS_CACHE_NO
||
8826 block_group
->cached
== BTRFS_CACHE_ERROR
)
8827 free_excluded_extents(info
->extent_root
, block_group
);
8829 btrfs_remove_free_space_cache(block_group
);
8830 btrfs_put_block_group(block_group
);
8832 spin_lock(&info
->block_group_cache_lock
);
8834 spin_unlock(&info
->block_group_cache_lock
);
8836 /* now that all the block groups are freed, go through and
8837 * free all the space_info structs. This is only called during
8838 * the final stages of unmount, and so we know nobody is
8839 * using them. We call synchronize_rcu() once before we start,
8840 * just to be on the safe side.
8844 release_global_block_rsv(info
);
8846 while (!list_empty(&info
->space_info
)) {
8849 space_info
= list_entry(info
->space_info
.next
,
8850 struct btrfs_space_info
,
8852 if (btrfs_test_opt(info
->tree_root
, ENOSPC_DEBUG
)) {
8853 if (WARN_ON(space_info
->bytes_pinned
> 0 ||
8854 space_info
->bytes_reserved
> 0 ||
8855 space_info
->bytes_may_use
> 0)) {
8856 dump_space_info(space_info
, 0, 0);
8859 list_del(&space_info
->list
);
8860 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++) {
8861 struct kobject
*kobj
;
8862 kobj
= space_info
->block_group_kobjs
[i
];
8863 space_info
->block_group_kobjs
[i
] = NULL
;
8869 kobject_del(&space_info
->kobj
);
8870 kobject_put(&space_info
->kobj
);
8875 static void __link_block_group(struct btrfs_space_info
*space_info
,
8876 struct btrfs_block_group_cache
*cache
)
8878 int index
= get_block_group_index(cache
);
8881 down_write(&space_info
->groups_sem
);
8882 if (list_empty(&space_info
->block_groups
[index
]))
8884 list_add_tail(&cache
->list
, &space_info
->block_groups
[index
]);
8885 up_write(&space_info
->groups_sem
);
8888 struct raid_kobject
*rkobj
;
8891 rkobj
= kzalloc(sizeof(*rkobj
), GFP_NOFS
);
8894 rkobj
->raid_type
= index
;
8895 kobject_init(&rkobj
->kobj
, &btrfs_raid_ktype
);
8896 ret
= kobject_add(&rkobj
->kobj
, &space_info
->kobj
,
8897 "%s", get_raid_name(index
));
8899 kobject_put(&rkobj
->kobj
);
8902 space_info
->block_group_kobjs
[index
] = &rkobj
->kobj
;
8907 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
8910 static struct btrfs_block_group_cache
*
8911 btrfs_create_block_group_cache(struct btrfs_root
*root
, u64 start
, u64 size
)
8913 struct btrfs_block_group_cache
*cache
;
8915 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
8919 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
8921 if (!cache
->free_space_ctl
) {
8926 cache
->key
.objectid
= start
;
8927 cache
->key
.offset
= size
;
8928 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
8930 cache
->sectorsize
= root
->sectorsize
;
8931 cache
->fs_info
= root
->fs_info
;
8932 cache
->full_stripe_len
= btrfs_full_stripe_len(root
,
8933 &root
->fs_info
->mapping_tree
,
8935 atomic_set(&cache
->count
, 1);
8936 spin_lock_init(&cache
->lock
);
8937 init_rwsem(&cache
->data_rwsem
);
8938 INIT_LIST_HEAD(&cache
->list
);
8939 INIT_LIST_HEAD(&cache
->cluster_list
);
8940 INIT_LIST_HEAD(&cache
->bg_list
);
8941 INIT_LIST_HEAD(&cache
->ro_list
);
8942 INIT_LIST_HEAD(&cache
->dirty_list
);
8943 btrfs_init_free_space_ctl(cache
);
8944 atomic_set(&cache
->trimming
, 0);
8949 int btrfs_read_block_groups(struct btrfs_root
*root
)
8951 struct btrfs_path
*path
;
8953 struct btrfs_block_group_cache
*cache
;
8954 struct btrfs_fs_info
*info
= root
->fs_info
;
8955 struct btrfs_space_info
*space_info
;
8956 struct btrfs_key key
;
8957 struct btrfs_key found_key
;
8958 struct extent_buffer
*leaf
;
8962 root
= info
->extent_root
;
8965 key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
8966 path
= btrfs_alloc_path();
8971 cache_gen
= btrfs_super_cache_generation(root
->fs_info
->super_copy
);
8972 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
8973 btrfs_super_generation(root
->fs_info
->super_copy
) != cache_gen
)
8975 if (btrfs_test_opt(root
, CLEAR_CACHE
))
8979 ret
= find_first_block_group(root
, path
, &key
);
8985 leaf
= path
->nodes
[0];
8986 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
8988 cache
= btrfs_create_block_group_cache(root
, found_key
.objectid
,
8997 * When we mount with old space cache, we need to
8998 * set BTRFS_DC_CLEAR and set dirty flag.
9000 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9001 * truncate the old free space cache inode and
9003 * b) Setting 'dirty flag' makes sure that we flush
9004 * the new space cache info onto disk.
9006 if (btrfs_test_opt(root
, SPACE_CACHE
))
9007 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
9010 read_extent_buffer(leaf
, &cache
->item
,
9011 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
9012 sizeof(cache
->item
));
9013 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
9015 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
9016 btrfs_release_path(path
);
9019 * We need to exclude the super stripes now so that the space
9020 * info has super bytes accounted for, otherwise we'll think
9021 * we have more space than we actually do.
9023 ret
= exclude_super_stripes(root
, cache
);
9026 * We may have excluded something, so call this just in
9029 free_excluded_extents(root
, cache
);
9030 btrfs_put_block_group(cache
);
9035 * check for two cases, either we are full, and therefore
9036 * don't need to bother with the caching work since we won't
9037 * find any space, or we are empty, and we can just add all
9038 * the space in and be done with it. This saves us _alot_ of
9039 * time, particularly in the full case.
9041 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
9042 cache
->last_byte_to_unpin
= (u64
)-1;
9043 cache
->cached
= BTRFS_CACHE_FINISHED
;
9044 free_excluded_extents(root
, cache
);
9045 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
9046 cache
->last_byte_to_unpin
= (u64
)-1;
9047 cache
->cached
= BTRFS_CACHE_FINISHED
;
9048 add_new_free_space(cache
, root
->fs_info
,
9050 found_key
.objectid
+
9052 free_excluded_extents(root
, cache
);
9055 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
9057 btrfs_remove_free_space_cache(cache
);
9058 btrfs_put_block_group(cache
);
9062 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
9063 btrfs_block_group_used(&cache
->item
),
9066 btrfs_remove_free_space_cache(cache
);
9067 spin_lock(&info
->block_group_cache_lock
);
9068 rb_erase(&cache
->cache_node
,
9069 &info
->block_group_cache_tree
);
9070 RB_CLEAR_NODE(&cache
->cache_node
);
9071 spin_unlock(&info
->block_group_cache_lock
);
9072 btrfs_put_block_group(cache
);
9076 cache
->space_info
= space_info
;
9077 spin_lock(&cache
->space_info
->lock
);
9078 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
9079 spin_unlock(&cache
->space_info
->lock
);
9081 __link_block_group(space_info
, cache
);
9083 set_avail_alloc_bits(root
->fs_info
, cache
->flags
);
9084 if (btrfs_chunk_readonly(root
, cache
->key
.objectid
)) {
9085 set_block_group_ro(cache
, 1);
9086 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
9087 spin_lock(&info
->unused_bgs_lock
);
9088 /* Should always be true but just in case. */
9089 if (list_empty(&cache
->bg_list
)) {
9090 btrfs_get_block_group(cache
);
9091 list_add_tail(&cache
->bg_list
,
9094 spin_unlock(&info
->unused_bgs_lock
);
9098 list_for_each_entry_rcu(space_info
, &root
->fs_info
->space_info
, list
) {
9099 if (!(get_alloc_profile(root
, space_info
->flags
) &
9100 (BTRFS_BLOCK_GROUP_RAID10
|
9101 BTRFS_BLOCK_GROUP_RAID1
|
9102 BTRFS_BLOCK_GROUP_RAID5
|
9103 BTRFS_BLOCK_GROUP_RAID6
|
9104 BTRFS_BLOCK_GROUP_DUP
)))
9107 * avoid allocating from un-mirrored block group if there are
9108 * mirrored block groups.
9110 list_for_each_entry(cache
,
9111 &space_info
->block_groups
[BTRFS_RAID_RAID0
],
9113 set_block_group_ro(cache
, 1);
9114 list_for_each_entry(cache
,
9115 &space_info
->block_groups
[BTRFS_RAID_SINGLE
],
9117 set_block_group_ro(cache
, 1);
9120 init_global_block_rsv(info
);
9123 btrfs_free_path(path
);
9127 void btrfs_create_pending_block_groups(struct btrfs_trans_handle
*trans
,
9128 struct btrfs_root
*root
)
9130 struct btrfs_block_group_cache
*block_group
, *tmp
;
9131 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
9132 struct btrfs_block_group_item item
;
9133 struct btrfs_key key
;
9136 list_for_each_entry_safe(block_group
, tmp
, &trans
->new_bgs
, bg_list
) {
9140 spin_lock(&block_group
->lock
);
9141 memcpy(&item
, &block_group
->item
, sizeof(item
));
9142 memcpy(&key
, &block_group
->key
, sizeof(key
));
9143 spin_unlock(&block_group
->lock
);
9145 ret
= btrfs_insert_item(trans
, extent_root
, &key
, &item
,
9148 btrfs_abort_transaction(trans
, extent_root
, ret
);
9149 ret
= btrfs_finish_chunk_alloc(trans
, extent_root
,
9150 key
.objectid
, key
.offset
);
9152 btrfs_abort_transaction(trans
, extent_root
, ret
);
9154 list_del_init(&block_group
->bg_list
);
9158 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
9159 struct btrfs_root
*root
, u64 bytes_used
,
9160 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
9164 struct btrfs_root
*extent_root
;
9165 struct btrfs_block_group_cache
*cache
;
9167 extent_root
= root
->fs_info
->extent_root
;
9169 btrfs_set_log_full_commit(root
->fs_info
, trans
);
9171 cache
= btrfs_create_block_group_cache(root
, chunk_offset
, size
);
9175 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
9176 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
9177 btrfs_set_block_group_flags(&cache
->item
, type
);
9179 cache
->flags
= type
;
9180 cache
->last_byte_to_unpin
= (u64
)-1;
9181 cache
->cached
= BTRFS_CACHE_FINISHED
;
9182 ret
= exclude_super_stripes(root
, cache
);
9185 * We may have excluded something, so call this just in
9188 free_excluded_extents(root
, cache
);
9189 btrfs_put_block_group(cache
);
9193 add_new_free_space(cache
, root
->fs_info
, chunk_offset
,
9194 chunk_offset
+ size
);
9196 free_excluded_extents(root
, cache
);
9198 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
9200 btrfs_remove_free_space_cache(cache
);
9201 btrfs_put_block_group(cache
);
9205 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
9206 &cache
->space_info
);
9208 btrfs_remove_free_space_cache(cache
);
9209 spin_lock(&root
->fs_info
->block_group_cache_lock
);
9210 rb_erase(&cache
->cache_node
,
9211 &root
->fs_info
->block_group_cache_tree
);
9212 RB_CLEAR_NODE(&cache
->cache_node
);
9213 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
9214 btrfs_put_block_group(cache
);
9217 update_global_block_rsv(root
->fs_info
);
9219 spin_lock(&cache
->space_info
->lock
);
9220 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
9221 spin_unlock(&cache
->space_info
->lock
);
9223 __link_block_group(cache
->space_info
, cache
);
9225 list_add_tail(&cache
->bg_list
, &trans
->new_bgs
);
9227 set_avail_alloc_bits(extent_root
->fs_info
, type
);
9232 static void clear_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
9234 u64 extra_flags
= chunk_to_extended(flags
) &
9235 BTRFS_EXTENDED_PROFILE_MASK
;
9237 write_seqlock(&fs_info
->profiles_lock
);
9238 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
9239 fs_info
->avail_data_alloc_bits
&= ~extra_flags
;
9240 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
9241 fs_info
->avail_metadata_alloc_bits
&= ~extra_flags
;
9242 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
9243 fs_info
->avail_system_alloc_bits
&= ~extra_flags
;
9244 write_sequnlock(&fs_info
->profiles_lock
);
9247 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
9248 struct btrfs_root
*root
, u64 group_start
,
9249 struct extent_map
*em
)
9251 struct btrfs_path
*path
;
9252 struct btrfs_block_group_cache
*block_group
;
9253 struct btrfs_free_cluster
*cluster
;
9254 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
9255 struct btrfs_key key
;
9256 struct inode
*inode
;
9257 struct kobject
*kobj
= NULL
;
9261 struct btrfs_caching_control
*caching_ctl
= NULL
;
9264 root
= root
->fs_info
->extent_root
;
9266 block_group
= btrfs_lookup_block_group(root
->fs_info
, group_start
);
9267 BUG_ON(!block_group
);
9268 BUG_ON(!block_group
->ro
);
9271 * Free the reserved super bytes from this block group before
9274 free_excluded_extents(root
, block_group
);
9276 memcpy(&key
, &block_group
->key
, sizeof(key
));
9277 index
= get_block_group_index(block_group
);
9278 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
9279 BTRFS_BLOCK_GROUP_RAID1
|
9280 BTRFS_BLOCK_GROUP_RAID10
))
9285 /* make sure this block group isn't part of an allocation cluster */
9286 cluster
= &root
->fs_info
->data_alloc_cluster
;
9287 spin_lock(&cluster
->refill_lock
);
9288 btrfs_return_cluster_to_free_space(block_group
, cluster
);
9289 spin_unlock(&cluster
->refill_lock
);
9292 * make sure this block group isn't part of a metadata
9293 * allocation cluster
9295 cluster
= &root
->fs_info
->meta_alloc_cluster
;
9296 spin_lock(&cluster
->refill_lock
);
9297 btrfs_return_cluster_to_free_space(block_group
, cluster
);
9298 spin_unlock(&cluster
->refill_lock
);
9300 path
= btrfs_alloc_path();
9306 inode
= lookup_free_space_inode(tree_root
, block_group
, path
);
9307 if (!IS_ERR(inode
)) {
9308 ret
= btrfs_orphan_add(trans
, inode
);
9310 btrfs_add_delayed_iput(inode
);
9314 /* One for the block groups ref */
9315 spin_lock(&block_group
->lock
);
9316 if (block_group
->iref
) {
9317 block_group
->iref
= 0;
9318 block_group
->inode
= NULL
;
9319 spin_unlock(&block_group
->lock
);
9322 spin_unlock(&block_group
->lock
);
9324 /* One for our lookup ref */
9325 btrfs_add_delayed_iput(inode
);
9328 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
9329 key
.offset
= block_group
->key
.objectid
;
9332 ret
= btrfs_search_slot(trans
, tree_root
, &key
, path
, -1, 1);
9336 btrfs_release_path(path
);
9338 ret
= btrfs_del_item(trans
, tree_root
, path
);
9341 btrfs_release_path(path
);
9344 spin_lock(&root
->fs_info
->block_group_cache_lock
);
9345 rb_erase(&block_group
->cache_node
,
9346 &root
->fs_info
->block_group_cache_tree
);
9347 RB_CLEAR_NODE(&block_group
->cache_node
);
9349 if (root
->fs_info
->first_logical_byte
== block_group
->key
.objectid
)
9350 root
->fs_info
->first_logical_byte
= (u64
)-1;
9351 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
9353 down_write(&block_group
->space_info
->groups_sem
);
9355 * we must use list_del_init so people can check to see if they
9356 * are still on the list after taking the semaphore
9358 list_del_init(&block_group
->list
);
9359 list_del_init(&block_group
->ro_list
);
9360 if (list_empty(&block_group
->space_info
->block_groups
[index
])) {
9361 kobj
= block_group
->space_info
->block_group_kobjs
[index
];
9362 block_group
->space_info
->block_group_kobjs
[index
] = NULL
;
9363 clear_avail_alloc_bits(root
->fs_info
, block_group
->flags
);
9365 up_write(&block_group
->space_info
->groups_sem
);
9371 if (block_group
->has_caching_ctl
)
9372 caching_ctl
= get_caching_control(block_group
);
9373 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
9374 wait_block_group_cache_done(block_group
);
9375 if (block_group
->has_caching_ctl
) {
9376 down_write(&root
->fs_info
->commit_root_sem
);
9378 struct btrfs_caching_control
*ctl
;
9380 list_for_each_entry(ctl
,
9381 &root
->fs_info
->caching_block_groups
, list
)
9382 if (ctl
->block_group
== block_group
) {
9384 atomic_inc(&caching_ctl
->count
);
9389 list_del_init(&caching_ctl
->list
);
9390 up_write(&root
->fs_info
->commit_root_sem
);
9392 /* Once for the caching bgs list and once for us. */
9393 put_caching_control(caching_ctl
);
9394 put_caching_control(caching_ctl
);
9398 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
9399 if (!list_empty(&block_group
->dirty_list
)) {
9400 list_del_init(&block_group
->dirty_list
);
9401 btrfs_put_block_group(block_group
);
9403 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
9405 btrfs_remove_free_space_cache(block_group
);
9407 spin_lock(&block_group
->space_info
->lock
);
9408 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
9409 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
9410 block_group
->space_info
->disk_total
-= block_group
->key
.offset
* factor
;
9411 spin_unlock(&block_group
->space_info
->lock
);
9413 memcpy(&key
, &block_group
->key
, sizeof(key
));
9416 if (!list_empty(&em
->list
)) {
9417 /* We're in the transaction->pending_chunks list. */
9418 free_extent_map(em
);
9420 spin_lock(&block_group
->lock
);
9421 block_group
->removed
= 1;
9423 * At this point trimming can't start on this block group, because we
9424 * removed the block group from the tree fs_info->block_group_cache_tree
9425 * so no one can't find it anymore and even if someone already got this
9426 * block group before we removed it from the rbtree, they have already
9427 * incremented block_group->trimming - if they didn't, they won't find
9428 * any free space entries because we already removed them all when we
9429 * called btrfs_remove_free_space_cache().
9431 * And we must not remove the extent map from the fs_info->mapping_tree
9432 * to prevent the same logical address range and physical device space
9433 * ranges from being reused for a new block group. This is because our
9434 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
9435 * completely transactionless, so while it is trimming a range the
9436 * currently running transaction might finish and a new one start,
9437 * allowing for new block groups to be created that can reuse the same
9438 * physical device locations unless we take this special care.
9440 remove_em
= (atomic_read(&block_group
->trimming
) == 0);
9442 * Make sure a trimmer task always sees the em in the pinned_chunks list
9443 * if it sees block_group->removed == 1 (needs to lock block_group->lock
9444 * before checking block_group->removed).
9448 * Our em might be in trans->transaction->pending_chunks which
9449 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
9450 * and so is the fs_info->pinned_chunks list.
9452 * So at this point we must be holding the chunk_mutex to avoid
9453 * any races with chunk allocation (more specifically at
9454 * volumes.c:contains_pending_extent()), to ensure it always
9455 * sees the em, either in the pending_chunks list or in the
9456 * pinned_chunks list.
9458 list_move_tail(&em
->list
, &root
->fs_info
->pinned_chunks
);
9460 spin_unlock(&block_group
->lock
);
9463 struct extent_map_tree
*em_tree
;
9465 em_tree
= &root
->fs_info
->mapping_tree
.map_tree
;
9466 write_lock(&em_tree
->lock
);
9468 * The em might be in the pending_chunks list, so make sure the
9469 * chunk mutex is locked, since remove_extent_mapping() will
9470 * delete us from that list.
9472 remove_extent_mapping(em_tree
, em
);
9473 write_unlock(&em_tree
->lock
);
9474 /* once for the tree */
9475 free_extent_map(em
);
9478 unlock_chunks(root
);
9480 btrfs_put_block_group(block_group
);
9481 btrfs_put_block_group(block_group
);
9483 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
9489 ret
= btrfs_del_item(trans
, root
, path
);
9491 btrfs_free_path(path
);
9496 * Process the unused_bgs list and remove any that don't have any allocated
9497 * space inside of them.
9499 void btrfs_delete_unused_bgs(struct btrfs_fs_info
*fs_info
)
9501 struct btrfs_block_group_cache
*block_group
;
9502 struct btrfs_space_info
*space_info
;
9503 struct btrfs_root
*root
= fs_info
->extent_root
;
9504 struct btrfs_trans_handle
*trans
;
9510 spin_lock(&fs_info
->unused_bgs_lock
);
9511 while (!list_empty(&fs_info
->unused_bgs
)) {
9514 block_group
= list_first_entry(&fs_info
->unused_bgs
,
9515 struct btrfs_block_group_cache
,
9517 space_info
= block_group
->space_info
;
9518 list_del_init(&block_group
->bg_list
);
9519 if (ret
|| btrfs_mixed_space_info(space_info
)) {
9520 btrfs_put_block_group(block_group
);
9523 spin_unlock(&fs_info
->unused_bgs_lock
);
9525 /* Don't want to race with allocators so take the groups_sem */
9526 down_write(&space_info
->groups_sem
);
9527 spin_lock(&block_group
->lock
);
9528 if (block_group
->reserved
||
9529 btrfs_block_group_used(&block_group
->item
) ||
9532 * We want to bail if we made new allocations or have
9533 * outstanding allocations in this block group. We do
9534 * the ro check in case balance is currently acting on
9537 spin_unlock(&block_group
->lock
);
9538 up_write(&space_info
->groups_sem
);
9541 spin_unlock(&block_group
->lock
);
9543 /* We don't want to force the issue, only flip if it's ok. */
9544 ret
= set_block_group_ro(block_group
, 0);
9545 up_write(&space_info
->groups_sem
);
9552 * Want to do this before we do anything else so we can recover
9553 * properly if we fail to join the transaction.
9555 trans
= btrfs_join_transaction(root
);
9556 if (IS_ERR(trans
)) {
9557 btrfs_set_block_group_rw(root
, block_group
);
9558 ret
= PTR_ERR(trans
);
9563 * We could have pending pinned extents for this block group,
9564 * just delete them, we don't care about them anymore.
9566 start
= block_group
->key
.objectid
;
9567 end
= start
+ block_group
->key
.offset
- 1;
9569 * Hold the unused_bg_unpin_mutex lock to avoid racing with
9570 * btrfs_finish_extent_commit(). If we are at transaction N,
9571 * another task might be running finish_extent_commit() for the
9572 * previous transaction N - 1, and have seen a range belonging
9573 * to the block group in freed_extents[] before we were able to
9574 * clear the whole block group range from freed_extents[]. This
9575 * means that task can lookup for the block group after we
9576 * unpinned it from freed_extents[] and removed it, leading to
9577 * a BUG_ON() at btrfs_unpin_extent_range().
9579 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
9580 ret
= clear_extent_bits(&fs_info
->freed_extents
[0], start
, end
,
9581 EXTENT_DIRTY
, GFP_NOFS
);
9583 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
9584 btrfs_set_block_group_rw(root
, block_group
);
9587 ret
= clear_extent_bits(&fs_info
->freed_extents
[1], start
, end
,
9588 EXTENT_DIRTY
, GFP_NOFS
);
9590 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
9591 btrfs_set_block_group_rw(root
, block_group
);
9594 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
9596 /* Reset pinned so btrfs_put_block_group doesn't complain */
9597 block_group
->pinned
= 0;
9600 * Btrfs_remove_chunk will abort the transaction if things go
9603 ret
= btrfs_remove_chunk(trans
, root
,
9604 block_group
->key
.objectid
);
9606 btrfs_end_transaction(trans
, root
);
9608 btrfs_put_block_group(block_group
);
9609 spin_lock(&fs_info
->unused_bgs_lock
);
9611 spin_unlock(&fs_info
->unused_bgs_lock
);
9614 int btrfs_init_space_info(struct btrfs_fs_info
*fs_info
)
9616 struct btrfs_space_info
*space_info
;
9617 struct btrfs_super_block
*disk_super
;
9623 disk_super
= fs_info
->super_copy
;
9624 if (!btrfs_super_root(disk_super
))
9627 features
= btrfs_super_incompat_flags(disk_super
);
9628 if (features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
9631 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
9632 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
9637 flags
= BTRFS_BLOCK_GROUP_METADATA
| BTRFS_BLOCK_GROUP_DATA
;
9638 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
9640 flags
= BTRFS_BLOCK_GROUP_METADATA
;
9641 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
9645 flags
= BTRFS_BLOCK_GROUP_DATA
;
9646 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
9652 int btrfs_error_unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
9654 return unpin_extent_range(root
, start
, end
, false);
9657 int btrfs_trim_fs(struct btrfs_root
*root
, struct fstrim_range
*range
)
9659 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
9660 struct btrfs_block_group_cache
*cache
= NULL
;
9665 u64 total_bytes
= btrfs_super_total_bytes(fs_info
->super_copy
);
9669 * try to trim all FS space, our block group may start from non-zero.
9671 if (range
->len
== total_bytes
)
9672 cache
= btrfs_lookup_first_block_group(fs_info
, range
->start
);
9674 cache
= btrfs_lookup_block_group(fs_info
, range
->start
);
9677 if (cache
->key
.objectid
>= (range
->start
+ range
->len
)) {
9678 btrfs_put_block_group(cache
);
9682 start
= max(range
->start
, cache
->key
.objectid
);
9683 end
= min(range
->start
+ range
->len
,
9684 cache
->key
.objectid
+ cache
->key
.offset
);
9686 if (end
- start
>= range
->minlen
) {
9687 if (!block_group_cache_done(cache
)) {
9688 ret
= cache_block_group(cache
, 0);
9690 btrfs_put_block_group(cache
);
9693 ret
= wait_block_group_cache_done(cache
);
9695 btrfs_put_block_group(cache
);
9699 ret
= btrfs_trim_block_group(cache
,
9705 trimmed
+= group_trimmed
;
9707 btrfs_put_block_group(cache
);
9712 cache
= next_block_group(fs_info
->tree_root
, cache
);
9715 range
->len
= trimmed
;
9720 * btrfs_{start,end}_write_no_snapshoting() are similar to
9721 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
9722 * data into the page cache through nocow before the subvolume is snapshoted,
9723 * but flush the data into disk after the snapshot creation, or to prevent
9724 * operations while snapshoting is ongoing and that cause the snapshot to be
9725 * inconsistent (writes followed by expanding truncates for example).
9727 void btrfs_end_write_no_snapshoting(struct btrfs_root
*root
)
9729 percpu_counter_dec(&root
->subv_writers
->counter
);
9731 * Make sure counter is updated before we wake up
9735 if (waitqueue_active(&root
->subv_writers
->wait
))
9736 wake_up(&root
->subv_writers
->wait
);
9739 int btrfs_start_write_no_snapshoting(struct btrfs_root
*root
)
9741 if (atomic_read(&root
->will_be_snapshoted
))
9744 percpu_counter_inc(&root
->subv_writers
->counter
);
9746 * Make sure counter is updated before we check for snapshot creation.
9749 if (atomic_read(&root
->will_be_snapshoted
)) {
9750 btrfs_end_write_no_snapshoting(root
);