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.
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;
2774 /* We'll clean this up in btrfs_cleanup_transaction */
2778 if (root
== root
->fs_info
->extent_root
)
2779 root
= root
->fs_info
->tree_root
;
2781 delayed_refs
= &trans
->transaction
->delayed_refs
;
2783 count
= atomic_read(&delayed_refs
->num_entries
) * 2;
2788 #ifdef SCRAMBLE_DELAYED_REFS
2789 delayed_refs
->run_delayed_start
= find_middle(&delayed_refs
->root
);
2791 ret
= __btrfs_run_delayed_refs(trans
, root
, count
);
2793 btrfs_abort_transaction(trans
, root
, ret
);
2798 if (!list_empty(&trans
->new_bgs
))
2799 btrfs_create_pending_block_groups(trans
, root
);
2801 spin_lock(&delayed_refs
->lock
);
2802 node
= rb_first(&delayed_refs
->href_root
);
2804 spin_unlock(&delayed_refs
->lock
);
2807 count
= (unsigned long)-1;
2810 head
= rb_entry(node
, struct btrfs_delayed_ref_head
,
2812 if (btrfs_delayed_ref_is_head(&head
->node
)) {
2813 struct btrfs_delayed_ref_node
*ref
;
2816 atomic_inc(&ref
->refs
);
2818 spin_unlock(&delayed_refs
->lock
);
2820 * Mutex was contended, block until it's
2821 * released and try again
2823 mutex_lock(&head
->mutex
);
2824 mutex_unlock(&head
->mutex
);
2826 btrfs_put_delayed_ref(ref
);
2832 node
= rb_next(node
);
2834 spin_unlock(&delayed_refs
->lock
);
2839 ret
= btrfs_delayed_qgroup_accounting(trans
, root
->fs_info
);
2842 assert_qgroups_uptodate(trans
);
2846 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
2847 struct btrfs_root
*root
,
2848 u64 bytenr
, u64 num_bytes
, u64 flags
,
2849 int level
, int is_data
)
2851 struct btrfs_delayed_extent_op
*extent_op
;
2854 extent_op
= btrfs_alloc_delayed_extent_op();
2858 extent_op
->flags_to_set
= flags
;
2859 extent_op
->update_flags
= 1;
2860 extent_op
->update_key
= 0;
2861 extent_op
->is_data
= is_data
? 1 : 0;
2862 extent_op
->level
= level
;
2864 ret
= btrfs_add_delayed_extent_op(root
->fs_info
, trans
, bytenr
,
2865 num_bytes
, extent_op
);
2867 btrfs_free_delayed_extent_op(extent_op
);
2871 static noinline
int check_delayed_ref(struct btrfs_trans_handle
*trans
,
2872 struct btrfs_root
*root
,
2873 struct btrfs_path
*path
,
2874 u64 objectid
, u64 offset
, u64 bytenr
)
2876 struct btrfs_delayed_ref_head
*head
;
2877 struct btrfs_delayed_ref_node
*ref
;
2878 struct btrfs_delayed_data_ref
*data_ref
;
2879 struct btrfs_delayed_ref_root
*delayed_refs
;
2880 struct rb_node
*node
;
2883 delayed_refs
= &trans
->transaction
->delayed_refs
;
2884 spin_lock(&delayed_refs
->lock
);
2885 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
2887 spin_unlock(&delayed_refs
->lock
);
2891 if (!mutex_trylock(&head
->mutex
)) {
2892 atomic_inc(&head
->node
.refs
);
2893 spin_unlock(&delayed_refs
->lock
);
2895 btrfs_release_path(path
);
2898 * Mutex was contended, block until it's released and let
2901 mutex_lock(&head
->mutex
);
2902 mutex_unlock(&head
->mutex
);
2903 btrfs_put_delayed_ref(&head
->node
);
2906 spin_unlock(&delayed_refs
->lock
);
2908 spin_lock(&head
->lock
);
2909 node
= rb_first(&head
->ref_root
);
2911 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2912 node
= rb_next(node
);
2914 /* If it's a shared ref we know a cross reference exists */
2915 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
) {
2920 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
2923 * If our ref doesn't match the one we're currently looking at
2924 * then we have a cross reference.
2926 if (data_ref
->root
!= root
->root_key
.objectid
||
2927 data_ref
->objectid
!= objectid
||
2928 data_ref
->offset
!= offset
) {
2933 spin_unlock(&head
->lock
);
2934 mutex_unlock(&head
->mutex
);
2938 static noinline
int check_committed_ref(struct btrfs_trans_handle
*trans
,
2939 struct btrfs_root
*root
,
2940 struct btrfs_path
*path
,
2941 u64 objectid
, u64 offset
, u64 bytenr
)
2943 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2944 struct extent_buffer
*leaf
;
2945 struct btrfs_extent_data_ref
*ref
;
2946 struct btrfs_extent_inline_ref
*iref
;
2947 struct btrfs_extent_item
*ei
;
2948 struct btrfs_key key
;
2952 key
.objectid
= bytenr
;
2953 key
.offset
= (u64
)-1;
2954 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2956 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
2959 BUG_ON(ret
== 0); /* Corruption */
2962 if (path
->slots
[0] == 0)
2966 leaf
= path
->nodes
[0];
2967 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2969 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
2973 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2974 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2975 if (item_size
< sizeof(*ei
)) {
2976 WARN_ON(item_size
!= sizeof(struct btrfs_extent_item_v0
));
2980 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2982 if (item_size
!= sizeof(*ei
) +
2983 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
2986 if (btrfs_extent_generation(leaf
, ei
) <=
2987 btrfs_root_last_snapshot(&root
->root_item
))
2990 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
2991 if (btrfs_extent_inline_ref_type(leaf
, iref
) !=
2992 BTRFS_EXTENT_DATA_REF_KEY
)
2995 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
2996 if (btrfs_extent_refs(leaf
, ei
) !=
2997 btrfs_extent_data_ref_count(leaf
, ref
) ||
2998 btrfs_extent_data_ref_root(leaf
, ref
) !=
2999 root
->root_key
.objectid
||
3000 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
3001 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
3009 int btrfs_cross_ref_exist(struct btrfs_trans_handle
*trans
,
3010 struct btrfs_root
*root
,
3011 u64 objectid
, u64 offset
, u64 bytenr
)
3013 struct btrfs_path
*path
;
3017 path
= btrfs_alloc_path();
3022 ret
= check_committed_ref(trans
, root
, path
, objectid
,
3024 if (ret
&& ret
!= -ENOENT
)
3027 ret2
= check_delayed_ref(trans
, root
, path
, objectid
,
3029 } while (ret2
== -EAGAIN
);
3031 if (ret2
&& ret2
!= -ENOENT
) {
3036 if (ret
!= -ENOENT
|| ret2
!= -ENOENT
)
3039 btrfs_free_path(path
);
3040 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
3045 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
3046 struct btrfs_root
*root
,
3047 struct extent_buffer
*buf
,
3048 int full_backref
, int inc
)
3055 struct btrfs_key key
;
3056 struct btrfs_file_extent_item
*fi
;
3060 int (*process_func
)(struct btrfs_trans_handle
*, struct btrfs_root
*,
3061 u64
, u64
, u64
, u64
, u64
, u64
, int);
3064 if (btrfs_test_is_dummy_root(root
))
3067 ref_root
= btrfs_header_owner(buf
);
3068 nritems
= btrfs_header_nritems(buf
);
3069 level
= btrfs_header_level(buf
);
3071 if (!test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) && level
== 0)
3075 process_func
= btrfs_inc_extent_ref
;
3077 process_func
= btrfs_free_extent
;
3080 parent
= buf
->start
;
3084 for (i
= 0; i
< nritems
; i
++) {
3086 btrfs_item_key_to_cpu(buf
, &key
, i
);
3087 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
3089 fi
= btrfs_item_ptr(buf
, i
,
3090 struct btrfs_file_extent_item
);
3091 if (btrfs_file_extent_type(buf
, fi
) ==
3092 BTRFS_FILE_EXTENT_INLINE
)
3094 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
3098 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
3099 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
3100 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
3101 parent
, ref_root
, key
.objectid
,
3106 bytenr
= btrfs_node_blockptr(buf
, i
);
3107 num_bytes
= root
->nodesize
;
3108 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
3109 parent
, ref_root
, level
- 1, 0,
3120 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3121 struct extent_buffer
*buf
, int full_backref
)
3123 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1);
3126 int btrfs_dec_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3127 struct extent_buffer
*buf
, int full_backref
)
3129 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 0);
3132 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
3133 struct btrfs_root
*root
,
3134 struct btrfs_path
*path
,
3135 struct btrfs_block_group_cache
*cache
)
3138 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
3140 struct extent_buffer
*leaf
;
3142 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
3149 leaf
= path
->nodes
[0];
3150 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
3151 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
3152 btrfs_mark_buffer_dirty(leaf
);
3153 btrfs_release_path(path
);
3156 btrfs_abort_transaction(trans
, root
, ret
);
3161 static struct btrfs_block_group_cache
*
3162 next_block_group(struct btrfs_root
*root
,
3163 struct btrfs_block_group_cache
*cache
)
3165 struct rb_node
*node
;
3167 spin_lock(&root
->fs_info
->block_group_cache_lock
);
3169 /* If our block group was removed, we need a full search. */
3170 if (RB_EMPTY_NODE(&cache
->cache_node
)) {
3171 const u64 next_bytenr
= cache
->key
.objectid
+ cache
->key
.offset
;
3173 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
3174 btrfs_put_block_group(cache
);
3175 cache
= btrfs_lookup_first_block_group(root
->fs_info
,
3179 node
= rb_next(&cache
->cache_node
);
3180 btrfs_put_block_group(cache
);
3182 cache
= rb_entry(node
, struct btrfs_block_group_cache
,
3184 btrfs_get_block_group(cache
);
3187 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
3191 static int cache_save_setup(struct btrfs_block_group_cache
*block_group
,
3192 struct btrfs_trans_handle
*trans
,
3193 struct btrfs_path
*path
)
3195 struct btrfs_root
*root
= block_group
->fs_info
->tree_root
;
3196 struct inode
*inode
= NULL
;
3198 int dcs
= BTRFS_DC_ERROR
;
3204 * If this block group is smaller than 100 megs don't bother caching the
3207 if (block_group
->key
.offset
< (100 * 1024 * 1024)) {
3208 spin_lock(&block_group
->lock
);
3209 block_group
->disk_cache_state
= BTRFS_DC_WRITTEN
;
3210 spin_unlock(&block_group
->lock
);
3215 inode
= lookup_free_space_inode(root
, block_group
, path
);
3216 if (IS_ERR(inode
) && PTR_ERR(inode
) != -ENOENT
) {
3217 ret
= PTR_ERR(inode
);
3218 btrfs_release_path(path
);
3222 if (IS_ERR(inode
)) {
3226 if (block_group
->ro
)
3229 ret
= create_free_space_inode(root
, trans
, block_group
, path
);
3235 /* We've already setup this transaction, go ahead and exit */
3236 if (block_group
->cache_generation
== trans
->transid
&&
3237 i_size_read(inode
)) {
3238 dcs
= BTRFS_DC_SETUP
;
3243 * We want to set the generation to 0, that way if anything goes wrong
3244 * from here on out we know not to trust this cache when we load up next
3247 BTRFS_I(inode
)->generation
= 0;
3248 ret
= btrfs_update_inode(trans
, root
, inode
);
3251 if (i_size_read(inode
) > 0) {
3252 ret
= btrfs_check_trunc_cache_free_space(root
,
3253 &root
->fs_info
->global_block_rsv
);
3257 ret
= btrfs_truncate_free_space_cache(root
, trans
, inode
);
3262 spin_lock(&block_group
->lock
);
3263 if (block_group
->cached
!= BTRFS_CACHE_FINISHED
||
3264 !btrfs_test_opt(root
, SPACE_CACHE
) ||
3265 block_group
->delalloc_bytes
) {
3267 * don't bother trying to write stuff out _if_
3268 * a) we're not cached,
3269 * b) we're with nospace_cache mount option.
3271 dcs
= BTRFS_DC_WRITTEN
;
3272 spin_unlock(&block_group
->lock
);
3275 spin_unlock(&block_group
->lock
);
3278 * Try to preallocate enough space based on how big the block group is.
3279 * Keep in mind this has to include any pinned space which could end up
3280 * taking up quite a bit since it's not folded into the other space
3283 num_pages
= (int)div64_u64(block_group
->key
.offset
, 256 * 1024 * 1024);
3288 num_pages
*= PAGE_CACHE_SIZE
;
3290 ret
= btrfs_check_data_free_space(inode
, num_pages
);
3294 ret
= btrfs_prealloc_file_range_trans(inode
, trans
, 0, 0, num_pages
,
3295 num_pages
, num_pages
,
3298 dcs
= BTRFS_DC_SETUP
;
3299 btrfs_free_reserved_data_space(inode
, num_pages
);
3304 btrfs_release_path(path
);
3306 spin_lock(&block_group
->lock
);
3307 if (!ret
&& dcs
== BTRFS_DC_SETUP
)
3308 block_group
->cache_generation
= trans
->transid
;
3309 block_group
->disk_cache_state
= dcs
;
3310 spin_unlock(&block_group
->lock
);
3315 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
3316 struct btrfs_root
*root
)
3318 struct btrfs_block_group_cache
*cache
;
3319 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3321 struct btrfs_path
*path
;
3323 if (list_empty(&cur_trans
->dirty_bgs
))
3326 path
= btrfs_alloc_path();
3331 * We don't need the lock here since we are protected by the transaction
3332 * commit. We want to do the cache_save_setup first and then run the
3333 * delayed refs to make sure we have the best chance at doing this all
3336 while (!list_empty(&cur_trans
->dirty_bgs
)) {
3337 cache
= list_first_entry(&cur_trans
->dirty_bgs
,
3338 struct btrfs_block_group_cache
,
3340 list_del_init(&cache
->dirty_list
);
3341 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
)
3342 cache_save_setup(cache
, trans
, path
);
3344 ret
= btrfs_run_delayed_refs(trans
, root
,
3345 (unsigned long) -1);
3346 if (!ret
&& cache
->disk_cache_state
== BTRFS_DC_SETUP
)
3347 btrfs_write_out_cache(root
, trans
, cache
, path
);
3349 ret
= write_one_cache_group(trans
, root
, path
, cache
);
3350 btrfs_put_block_group(cache
);
3353 btrfs_free_path(path
);
3357 int btrfs_extent_readonly(struct btrfs_root
*root
, u64 bytenr
)
3359 struct btrfs_block_group_cache
*block_group
;
3362 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
3363 if (!block_group
|| block_group
->ro
)
3366 btrfs_put_block_group(block_group
);
3370 static const char *alloc_name(u64 flags
)
3373 case BTRFS_BLOCK_GROUP_METADATA
|BTRFS_BLOCK_GROUP_DATA
:
3375 case BTRFS_BLOCK_GROUP_METADATA
:
3377 case BTRFS_BLOCK_GROUP_DATA
:
3379 case BTRFS_BLOCK_GROUP_SYSTEM
:
3383 return "invalid-combination";
3387 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
3388 u64 total_bytes
, u64 bytes_used
,
3389 struct btrfs_space_info
**space_info
)
3391 struct btrfs_space_info
*found
;
3396 if (flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3397 BTRFS_BLOCK_GROUP_RAID10
))
3402 found
= __find_space_info(info
, flags
);
3404 spin_lock(&found
->lock
);
3405 found
->total_bytes
+= total_bytes
;
3406 found
->disk_total
+= total_bytes
* factor
;
3407 found
->bytes_used
+= bytes_used
;
3408 found
->disk_used
+= bytes_used
* factor
;
3410 spin_unlock(&found
->lock
);
3411 *space_info
= found
;
3414 found
= kzalloc(sizeof(*found
), GFP_NOFS
);
3418 ret
= percpu_counter_init(&found
->total_bytes_pinned
, 0, GFP_KERNEL
);
3424 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
3425 INIT_LIST_HEAD(&found
->block_groups
[i
]);
3426 init_rwsem(&found
->groups_sem
);
3427 spin_lock_init(&found
->lock
);
3428 found
->flags
= flags
& BTRFS_BLOCK_GROUP_TYPE_MASK
;
3429 found
->total_bytes
= total_bytes
;
3430 found
->disk_total
= total_bytes
* factor
;
3431 found
->bytes_used
= bytes_used
;
3432 found
->disk_used
= bytes_used
* factor
;
3433 found
->bytes_pinned
= 0;
3434 found
->bytes_reserved
= 0;
3435 found
->bytes_readonly
= 0;
3436 found
->bytes_may_use
= 0;
3438 found
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3439 found
->chunk_alloc
= 0;
3441 init_waitqueue_head(&found
->wait
);
3442 INIT_LIST_HEAD(&found
->ro_bgs
);
3444 ret
= kobject_init_and_add(&found
->kobj
, &space_info_ktype
,
3445 info
->space_info_kobj
, "%s",
3446 alloc_name(found
->flags
));
3452 *space_info
= found
;
3453 list_add_rcu(&found
->list
, &info
->space_info
);
3454 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3455 info
->data_sinfo
= found
;
3460 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
3462 u64 extra_flags
= chunk_to_extended(flags
) &
3463 BTRFS_EXTENDED_PROFILE_MASK
;
3465 write_seqlock(&fs_info
->profiles_lock
);
3466 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3467 fs_info
->avail_data_alloc_bits
|= extra_flags
;
3468 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3469 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
3470 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3471 fs_info
->avail_system_alloc_bits
|= extra_flags
;
3472 write_sequnlock(&fs_info
->profiles_lock
);
3476 * returns target flags in extended format or 0 if restripe for this
3477 * chunk_type is not in progress
3479 * should be called with either volume_mutex or balance_lock held
3481 static u64
get_restripe_target(struct btrfs_fs_info
*fs_info
, u64 flags
)
3483 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3489 if (flags
& BTRFS_BLOCK_GROUP_DATA
&&
3490 bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3491 target
= BTRFS_BLOCK_GROUP_DATA
| bctl
->data
.target
;
3492 } else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
&&
3493 bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3494 target
= BTRFS_BLOCK_GROUP_SYSTEM
| bctl
->sys
.target
;
3495 } else if (flags
& BTRFS_BLOCK_GROUP_METADATA
&&
3496 bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3497 target
= BTRFS_BLOCK_GROUP_METADATA
| bctl
->meta
.target
;
3504 * @flags: available profiles in extended format (see ctree.h)
3506 * Returns reduced profile in chunk format. If profile changing is in
3507 * progress (either running or paused) picks the target profile (if it's
3508 * already available), otherwise falls back to plain reducing.
3510 static u64
btrfs_reduce_alloc_profile(struct btrfs_root
*root
, u64 flags
)
3512 u64 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
3517 * see if restripe for this chunk_type is in progress, if so
3518 * try to reduce to the target profile
3520 spin_lock(&root
->fs_info
->balance_lock
);
3521 target
= get_restripe_target(root
->fs_info
, flags
);
3523 /* pick target profile only if it's already available */
3524 if ((flags
& target
) & BTRFS_EXTENDED_PROFILE_MASK
) {
3525 spin_unlock(&root
->fs_info
->balance_lock
);
3526 return extended_to_chunk(target
);
3529 spin_unlock(&root
->fs_info
->balance_lock
);
3531 /* First, mask out the RAID levels which aren't possible */
3532 if (num_devices
== 1)
3533 flags
&= ~(BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID0
|
3534 BTRFS_BLOCK_GROUP_RAID5
);
3535 if (num_devices
< 3)
3536 flags
&= ~BTRFS_BLOCK_GROUP_RAID6
;
3537 if (num_devices
< 4)
3538 flags
&= ~BTRFS_BLOCK_GROUP_RAID10
;
3540 tmp
= flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID0
|
3541 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID5
|
3542 BTRFS_BLOCK_GROUP_RAID6
| BTRFS_BLOCK_GROUP_RAID10
);
3545 if (tmp
& BTRFS_BLOCK_GROUP_RAID6
)
3546 tmp
= BTRFS_BLOCK_GROUP_RAID6
;
3547 else if (tmp
& BTRFS_BLOCK_GROUP_RAID5
)
3548 tmp
= BTRFS_BLOCK_GROUP_RAID5
;
3549 else if (tmp
& BTRFS_BLOCK_GROUP_RAID10
)
3550 tmp
= BTRFS_BLOCK_GROUP_RAID10
;
3551 else if (tmp
& BTRFS_BLOCK_GROUP_RAID1
)
3552 tmp
= BTRFS_BLOCK_GROUP_RAID1
;
3553 else if (tmp
& BTRFS_BLOCK_GROUP_RAID0
)
3554 tmp
= BTRFS_BLOCK_GROUP_RAID0
;
3556 return extended_to_chunk(flags
| tmp
);
3559 static u64
get_alloc_profile(struct btrfs_root
*root
, u64 orig_flags
)
3566 seq
= read_seqbegin(&root
->fs_info
->profiles_lock
);
3568 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3569 flags
|= root
->fs_info
->avail_data_alloc_bits
;
3570 else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3571 flags
|= root
->fs_info
->avail_system_alloc_bits
;
3572 else if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3573 flags
|= root
->fs_info
->avail_metadata_alloc_bits
;
3574 } while (read_seqretry(&root
->fs_info
->profiles_lock
, seq
));
3576 return btrfs_reduce_alloc_profile(root
, flags
);
3579 u64
btrfs_get_alloc_profile(struct btrfs_root
*root
, int data
)
3585 flags
= BTRFS_BLOCK_GROUP_DATA
;
3586 else if (root
== root
->fs_info
->chunk_root
)
3587 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
3589 flags
= BTRFS_BLOCK_GROUP_METADATA
;
3591 ret
= get_alloc_profile(root
, flags
);
3596 * This will check the space that the inode allocates from to make sure we have
3597 * enough space for bytes.
3599 int btrfs_check_data_free_space(struct inode
*inode
, u64 bytes
)
3601 struct btrfs_space_info
*data_sinfo
;
3602 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3603 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3605 int ret
= 0, committed
= 0, alloc_chunk
= 1;
3607 /* make sure bytes are sectorsize aligned */
3608 bytes
= ALIGN(bytes
, root
->sectorsize
);
3610 if (btrfs_is_free_space_inode(inode
)) {
3612 ASSERT(current
->journal_info
);
3615 data_sinfo
= fs_info
->data_sinfo
;
3620 /* make sure we have enough space to handle the data first */
3621 spin_lock(&data_sinfo
->lock
);
3622 used
= data_sinfo
->bytes_used
+ data_sinfo
->bytes_reserved
+
3623 data_sinfo
->bytes_pinned
+ data_sinfo
->bytes_readonly
+
3624 data_sinfo
->bytes_may_use
;
3626 if (used
+ bytes
> data_sinfo
->total_bytes
) {
3627 struct btrfs_trans_handle
*trans
;
3630 * if we don't have enough free bytes in this space then we need
3631 * to alloc a new chunk.
3633 if (!data_sinfo
->full
&& alloc_chunk
) {
3636 data_sinfo
->force_alloc
= CHUNK_ALLOC_FORCE
;
3637 spin_unlock(&data_sinfo
->lock
);
3639 alloc_target
= btrfs_get_alloc_profile(root
, 1);
3641 * It is ugly that we don't call nolock join
3642 * transaction for the free space inode case here.
3643 * But it is safe because we only do the data space
3644 * reservation for the free space cache in the
3645 * transaction context, the common join transaction
3646 * just increase the counter of the current transaction
3647 * handler, doesn't try to acquire the trans_lock of
3650 trans
= btrfs_join_transaction(root
);
3652 return PTR_ERR(trans
);
3654 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
3656 CHUNK_ALLOC_NO_FORCE
);
3657 btrfs_end_transaction(trans
, root
);
3666 data_sinfo
= fs_info
->data_sinfo
;
3672 * If we don't have enough pinned space to deal with this
3673 * allocation don't bother committing the transaction.
3675 if (percpu_counter_compare(&data_sinfo
->total_bytes_pinned
,
3678 spin_unlock(&data_sinfo
->lock
);
3680 /* commit the current transaction and try again */
3683 !atomic_read(&root
->fs_info
->open_ioctl_trans
)) {
3686 trans
= btrfs_join_transaction(root
);
3688 return PTR_ERR(trans
);
3689 ret
= btrfs_commit_transaction(trans
, root
);
3695 trace_btrfs_space_reservation(root
->fs_info
,
3696 "space_info:enospc",
3697 data_sinfo
->flags
, bytes
, 1);
3700 data_sinfo
->bytes_may_use
+= bytes
;
3701 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
3702 data_sinfo
->flags
, bytes
, 1);
3703 spin_unlock(&data_sinfo
->lock
);
3709 * Called if we need to clear a data reservation for this inode.
3711 void btrfs_free_reserved_data_space(struct inode
*inode
, u64 bytes
)
3713 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3714 struct btrfs_space_info
*data_sinfo
;
3716 /* make sure bytes are sectorsize aligned */
3717 bytes
= ALIGN(bytes
, root
->sectorsize
);
3719 data_sinfo
= root
->fs_info
->data_sinfo
;
3720 spin_lock(&data_sinfo
->lock
);
3721 WARN_ON(data_sinfo
->bytes_may_use
< bytes
);
3722 data_sinfo
->bytes_may_use
-= bytes
;
3723 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
3724 data_sinfo
->flags
, bytes
, 0);
3725 spin_unlock(&data_sinfo
->lock
);
3728 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
3730 struct list_head
*head
= &info
->space_info
;
3731 struct btrfs_space_info
*found
;
3734 list_for_each_entry_rcu(found
, head
, list
) {
3735 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
3736 found
->force_alloc
= CHUNK_ALLOC_FORCE
;
3741 static inline u64
calc_global_rsv_need_space(struct btrfs_block_rsv
*global
)
3743 return (global
->size
<< 1);
3746 static int should_alloc_chunk(struct btrfs_root
*root
,
3747 struct btrfs_space_info
*sinfo
, int force
)
3749 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
3750 u64 num_bytes
= sinfo
->total_bytes
- sinfo
->bytes_readonly
;
3751 u64 num_allocated
= sinfo
->bytes_used
+ sinfo
->bytes_reserved
;
3754 if (force
== CHUNK_ALLOC_FORCE
)
3758 * We need to take into account the global rsv because for all intents
3759 * and purposes it's used space. Don't worry about locking the
3760 * global_rsv, it doesn't change except when the transaction commits.
3762 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
3763 num_allocated
+= calc_global_rsv_need_space(global_rsv
);
3766 * in limited mode, we want to have some free space up to
3767 * about 1% of the FS size.
3769 if (force
== CHUNK_ALLOC_LIMITED
) {
3770 thresh
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
3771 thresh
= max_t(u64
, 64 * 1024 * 1024,
3772 div_factor_fine(thresh
, 1));
3774 if (num_bytes
- num_allocated
< thresh
)
3778 if (num_allocated
+ 2 * 1024 * 1024 < div_factor(num_bytes
, 8))
3783 static u64
get_system_chunk_thresh(struct btrfs_root
*root
, u64 type
)
3787 if (type
& (BTRFS_BLOCK_GROUP_RAID10
|
3788 BTRFS_BLOCK_GROUP_RAID0
|
3789 BTRFS_BLOCK_GROUP_RAID5
|
3790 BTRFS_BLOCK_GROUP_RAID6
))
3791 num_dev
= root
->fs_info
->fs_devices
->rw_devices
;
3792 else if (type
& BTRFS_BLOCK_GROUP_RAID1
)
3795 num_dev
= 1; /* DUP or single */
3797 /* metadata for updaing devices and chunk tree */
3798 return btrfs_calc_trans_metadata_size(root
, num_dev
+ 1);
3801 static void check_system_chunk(struct btrfs_trans_handle
*trans
,
3802 struct btrfs_root
*root
, u64 type
)
3804 struct btrfs_space_info
*info
;
3808 info
= __find_space_info(root
->fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
3809 spin_lock(&info
->lock
);
3810 left
= info
->total_bytes
- info
->bytes_used
- info
->bytes_pinned
-
3811 info
->bytes_reserved
- info
->bytes_readonly
;
3812 spin_unlock(&info
->lock
);
3814 thresh
= get_system_chunk_thresh(root
, type
);
3815 if (left
< thresh
&& btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
3816 btrfs_info(root
->fs_info
, "left=%llu, need=%llu, flags=%llu",
3817 left
, thresh
, type
);
3818 dump_space_info(info
, 0, 0);
3821 if (left
< thresh
) {
3824 flags
= btrfs_get_alloc_profile(root
->fs_info
->chunk_root
, 0);
3825 btrfs_alloc_chunk(trans
, root
, flags
);
3829 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
3830 struct btrfs_root
*extent_root
, u64 flags
, int force
)
3832 struct btrfs_space_info
*space_info
;
3833 struct btrfs_fs_info
*fs_info
= extent_root
->fs_info
;
3834 int wait_for_alloc
= 0;
3837 /* Don't re-enter if we're already allocating a chunk */
3838 if (trans
->allocating_chunk
)
3841 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
3843 ret
= update_space_info(extent_root
->fs_info
, flags
,
3845 BUG_ON(ret
); /* -ENOMEM */
3847 BUG_ON(!space_info
); /* Logic error */
3850 spin_lock(&space_info
->lock
);
3851 if (force
< space_info
->force_alloc
)
3852 force
= space_info
->force_alloc
;
3853 if (space_info
->full
) {
3854 if (should_alloc_chunk(extent_root
, space_info
, force
))
3858 spin_unlock(&space_info
->lock
);
3862 if (!should_alloc_chunk(extent_root
, space_info
, force
)) {
3863 spin_unlock(&space_info
->lock
);
3865 } else if (space_info
->chunk_alloc
) {
3868 space_info
->chunk_alloc
= 1;
3871 spin_unlock(&space_info
->lock
);
3873 mutex_lock(&fs_info
->chunk_mutex
);
3876 * The chunk_mutex is held throughout the entirety of a chunk
3877 * allocation, so once we've acquired the chunk_mutex we know that the
3878 * other guy is done and we need to recheck and see if we should
3881 if (wait_for_alloc
) {
3882 mutex_unlock(&fs_info
->chunk_mutex
);
3887 trans
->allocating_chunk
= true;
3890 * If we have mixed data/metadata chunks we want to make sure we keep
3891 * allocating mixed chunks instead of individual chunks.
3893 if (btrfs_mixed_space_info(space_info
))
3894 flags
|= (BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
);
3897 * if we're doing a data chunk, go ahead and make sure that
3898 * we keep a reasonable number of metadata chunks allocated in the
3901 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
3902 fs_info
->data_chunk_allocations
++;
3903 if (!(fs_info
->data_chunk_allocations
%
3904 fs_info
->metadata_ratio
))
3905 force_metadata_allocation(fs_info
);
3909 * Check if we have enough space in SYSTEM chunk because we may need
3910 * to update devices.
3912 check_system_chunk(trans
, extent_root
, flags
);
3914 ret
= btrfs_alloc_chunk(trans
, extent_root
, flags
);
3915 trans
->allocating_chunk
= false;
3917 spin_lock(&space_info
->lock
);
3918 if (ret
< 0 && ret
!= -ENOSPC
)
3921 space_info
->full
= 1;
3925 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3927 space_info
->chunk_alloc
= 0;
3928 spin_unlock(&space_info
->lock
);
3929 mutex_unlock(&fs_info
->chunk_mutex
);
3933 static int can_overcommit(struct btrfs_root
*root
,
3934 struct btrfs_space_info
*space_info
, u64 bytes
,
3935 enum btrfs_reserve_flush_enum flush
)
3937 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
3938 u64 profile
= btrfs_get_alloc_profile(root
, 0);
3943 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
3944 space_info
->bytes_pinned
+ space_info
->bytes_readonly
;
3947 * We only want to allow over committing if we have lots of actual space
3948 * free, but if we don't have enough space to handle the global reserve
3949 * space then we could end up having a real enospc problem when trying
3950 * to allocate a chunk or some other such important allocation.
3952 spin_lock(&global_rsv
->lock
);
3953 space_size
= calc_global_rsv_need_space(global_rsv
);
3954 spin_unlock(&global_rsv
->lock
);
3955 if (used
+ space_size
>= space_info
->total_bytes
)
3958 used
+= space_info
->bytes_may_use
;
3960 spin_lock(&root
->fs_info
->free_chunk_lock
);
3961 avail
= root
->fs_info
->free_chunk_space
;
3962 spin_unlock(&root
->fs_info
->free_chunk_lock
);
3965 * If we have dup, raid1 or raid10 then only half of the free
3966 * space is actually useable. For raid56, the space info used
3967 * doesn't include the parity drive, so we don't have to
3970 if (profile
& (BTRFS_BLOCK_GROUP_DUP
|
3971 BTRFS_BLOCK_GROUP_RAID1
|
3972 BTRFS_BLOCK_GROUP_RAID10
))
3976 * If we aren't flushing all things, let us overcommit up to
3977 * 1/2th of the space. If we can flush, don't let us overcommit
3978 * too much, let it overcommit up to 1/8 of the space.
3980 if (flush
== BTRFS_RESERVE_FLUSH_ALL
)
3985 if (used
+ bytes
< space_info
->total_bytes
+ avail
)
3990 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root
*root
,
3991 unsigned long nr_pages
, int nr_items
)
3993 struct super_block
*sb
= root
->fs_info
->sb
;
3995 if (down_read_trylock(&sb
->s_umount
)) {
3996 writeback_inodes_sb_nr(sb
, nr_pages
, WB_REASON_FS_FREE_SPACE
);
3997 up_read(&sb
->s_umount
);
4000 * We needn't worry the filesystem going from r/w to r/o though
4001 * we don't acquire ->s_umount mutex, because the filesystem
4002 * should guarantee the delalloc inodes list be empty after
4003 * the filesystem is readonly(all dirty pages are written to
4006 btrfs_start_delalloc_roots(root
->fs_info
, 0, nr_items
);
4007 if (!current
->journal_info
)
4008 btrfs_wait_ordered_roots(root
->fs_info
, nr_items
);
4012 static inline int calc_reclaim_items_nr(struct btrfs_root
*root
, u64 to_reclaim
)
4017 bytes
= btrfs_calc_trans_metadata_size(root
, 1);
4018 nr
= (int)div64_u64(to_reclaim
, bytes
);
4024 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4027 * shrink metadata reservation for delalloc
4029 static void shrink_delalloc(struct btrfs_root
*root
, u64 to_reclaim
, u64 orig
,
4032 struct btrfs_block_rsv
*block_rsv
;
4033 struct btrfs_space_info
*space_info
;
4034 struct btrfs_trans_handle
*trans
;
4038 unsigned long nr_pages
;
4041 enum btrfs_reserve_flush_enum flush
;
4043 /* Calc the number of the pages we need flush for space reservation */
4044 items
= calc_reclaim_items_nr(root
, to_reclaim
);
4045 to_reclaim
= items
* EXTENT_SIZE_PER_ITEM
;
4047 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4048 block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
4049 space_info
= block_rsv
->space_info
;
4051 delalloc_bytes
= percpu_counter_sum_positive(
4052 &root
->fs_info
->delalloc_bytes
);
4053 if (delalloc_bytes
== 0) {
4057 btrfs_wait_ordered_roots(root
->fs_info
, items
);
4062 while (delalloc_bytes
&& loops
< 3) {
4063 max_reclaim
= min(delalloc_bytes
, to_reclaim
);
4064 nr_pages
= max_reclaim
>> PAGE_CACHE_SHIFT
;
4065 btrfs_writeback_inodes_sb_nr(root
, nr_pages
, items
);
4067 * We need to wait for the async pages to actually start before
4070 max_reclaim
= atomic_read(&root
->fs_info
->async_delalloc_pages
);
4074 if (max_reclaim
<= nr_pages
)
4077 max_reclaim
-= nr_pages
;
4079 wait_event(root
->fs_info
->async_submit_wait
,
4080 atomic_read(&root
->fs_info
->async_delalloc_pages
) <=
4084 flush
= BTRFS_RESERVE_FLUSH_ALL
;
4086 flush
= BTRFS_RESERVE_NO_FLUSH
;
4087 spin_lock(&space_info
->lock
);
4088 if (can_overcommit(root
, space_info
, orig
, flush
)) {
4089 spin_unlock(&space_info
->lock
);
4092 spin_unlock(&space_info
->lock
);
4095 if (wait_ordered
&& !trans
) {
4096 btrfs_wait_ordered_roots(root
->fs_info
, items
);
4098 time_left
= schedule_timeout_killable(1);
4102 delalloc_bytes
= percpu_counter_sum_positive(
4103 &root
->fs_info
->delalloc_bytes
);
4108 * maybe_commit_transaction - possibly commit the transaction if its ok to
4109 * @root - the root we're allocating for
4110 * @bytes - the number of bytes we want to reserve
4111 * @force - force the commit
4113 * This will check to make sure that committing the transaction will actually
4114 * get us somewhere and then commit the transaction if it does. Otherwise it
4115 * will return -ENOSPC.
4117 static int may_commit_transaction(struct btrfs_root
*root
,
4118 struct btrfs_space_info
*space_info
,
4119 u64 bytes
, int force
)
4121 struct btrfs_block_rsv
*delayed_rsv
= &root
->fs_info
->delayed_block_rsv
;
4122 struct btrfs_trans_handle
*trans
;
4124 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4131 /* See if there is enough pinned space to make this reservation */
4132 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4137 * See if there is some space in the delayed insertion reservation for
4140 if (space_info
!= delayed_rsv
->space_info
)
4143 spin_lock(&delayed_rsv
->lock
);
4144 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4145 bytes
- delayed_rsv
->size
) >= 0) {
4146 spin_unlock(&delayed_rsv
->lock
);
4149 spin_unlock(&delayed_rsv
->lock
);
4152 trans
= btrfs_join_transaction(root
);
4156 return btrfs_commit_transaction(trans
, root
);
4160 FLUSH_DELAYED_ITEMS_NR
= 1,
4161 FLUSH_DELAYED_ITEMS
= 2,
4163 FLUSH_DELALLOC_WAIT
= 4,
4168 static int flush_space(struct btrfs_root
*root
,
4169 struct btrfs_space_info
*space_info
, u64 num_bytes
,
4170 u64 orig_bytes
, int state
)
4172 struct btrfs_trans_handle
*trans
;
4177 case FLUSH_DELAYED_ITEMS_NR
:
4178 case FLUSH_DELAYED_ITEMS
:
4179 if (state
== FLUSH_DELAYED_ITEMS_NR
)
4180 nr
= calc_reclaim_items_nr(root
, num_bytes
) * 2;
4184 trans
= btrfs_join_transaction(root
);
4185 if (IS_ERR(trans
)) {
4186 ret
= PTR_ERR(trans
);
4189 ret
= btrfs_run_delayed_items_nr(trans
, root
, nr
);
4190 btrfs_end_transaction(trans
, root
);
4192 case FLUSH_DELALLOC
:
4193 case FLUSH_DELALLOC_WAIT
:
4194 shrink_delalloc(root
, num_bytes
* 2, orig_bytes
,
4195 state
== FLUSH_DELALLOC_WAIT
);
4198 trans
= btrfs_join_transaction(root
);
4199 if (IS_ERR(trans
)) {
4200 ret
= PTR_ERR(trans
);
4203 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
4204 btrfs_get_alloc_profile(root
, 0),
4205 CHUNK_ALLOC_NO_FORCE
);
4206 btrfs_end_transaction(trans
, root
);
4211 ret
= may_commit_transaction(root
, space_info
, orig_bytes
, 0);
4222 btrfs_calc_reclaim_metadata_size(struct btrfs_root
*root
,
4223 struct btrfs_space_info
*space_info
)
4229 to_reclaim
= min_t(u64
, num_online_cpus() * 1024 * 1024,
4231 spin_lock(&space_info
->lock
);
4232 if (can_overcommit(root
, space_info
, to_reclaim
,
4233 BTRFS_RESERVE_FLUSH_ALL
)) {
4238 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4239 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4240 space_info
->bytes_may_use
;
4241 if (can_overcommit(root
, space_info
, 1024 * 1024,
4242 BTRFS_RESERVE_FLUSH_ALL
))
4243 expected
= div_factor_fine(space_info
->total_bytes
, 95);
4245 expected
= div_factor_fine(space_info
->total_bytes
, 90);
4247 if (used
> expected
)
4248 to_reclaim
= used
- expected
;
4251 to_reclaim
= min(to_reclaim
, space_info
->bytes_may_use
+
4252 space_info
->bytes_reserved
);
4254 spin_unlock(&space_info
->lock
);
4259 static inline int need_do_async_reclaim(struct btrfs_space_info
*space_info
,
4260 struct btrfs_fs_info
*fs_info
, u64 used
)
4262 return (used
>= div_factor_fine(space_info
->total_bytes
, 98) &&
4263 !btrfs_fs_closing(fs_info
) &&
4264 !test_bit(BTRFS_FS_STATE_REMOUNTING
, &fs_info
->fs_state
));
4267 static int btrfs_need_do_async_reclaim(struct btrfs_space_info
*space_info
,
4268 struct btrfs_fs_info
*fs_info
,
4273 spin_lock(&space_info
->lock
);
4275 * We run out of space and have not got any free space via flush_space,
4276 * so don't bother doing async reclaim.
4278 if (flush_state
> COMMIT_TRANS
&& space_info
->full
) {
4279 spin_unlock(&space_info
->lock
);
4283 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4284 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4285 space_info
->bytes_may_use
;
4286 if (need_do_async_reclaim(space_info
, fs_info
, used
)) {
4287 spin_unlock(&space_info
->lock
);
4290 spin_unlock(&space_info
->lock
);
4295 static void btrfs_async_reclaim_metadata_space(struct work_struct
*work
)
4297 struct btrfs_fs_info
*fs_info
;
4298 struct btrfs_space_info
*space_info
;
4302 fs_info
= container_of(work
, struct btrfs_fs_info
, async_reclaim_work
);
4303 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4305 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
->fs_root
,
4310 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
4312 flush_space(fs_info
->fs_root
, space_info
, to_reclaim
,
4313 to_reclaim
, flush_state
);
4315 if (!btrfs_need_do_async_reclaim(space_info
, fs_info
,
4318 } while (flush_state
<= COMMIT_TRANS
);
4320 if (btrfs_need_do_async_reclaim(space_info
, fs_info
, flush_state
))
4321 queue_work(system_unbound_wq
, work
);
4324 void btrfs_init_async_reclaim_work(struct work_struct
*work
)
4326 INIT_WORK(work
, btrfs_async_reclaim_metadata_space
);
4330 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4331 * @root - the root we're allocating for
4332 * @block_rsv - the block_rsv we're allocating for
4333 * @orig_bytes - the number of bytes we want
4334 * @flush - whether or not we can flush to make our reservation
4336 * This will reserve orgi_bytes number of bytes from the space info associated
4337 * with the block_rsv. If there is not enough space it will make an attempt to
4338 * flush out space to make room. It will do this by flushing delalloc if
4339 * possible or committing the transaction. If flush is 0 then no attempts to
4340 * regain reservations will be made and this will fail if there is not enough
4343 static int reserve_metadata_bytes(struct btrfs_root
*root
,
4344 struct btrfs_block_rsv
*block_rsv
,
4346 enum btrfs_reserve_flush_enum flush
)
4348 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
4350 u64 num_bytes
= orig_bytes
;
4351 int flush_state
= FLUSH_DELAYED_ITEMS_NR
;
4353 bool flushing
= false;
4357 spin_lock(&space_info
->lock
);
4359 * We only want to wait if somebody other than us is flushing and we
4360 * are actually allowed to flush all things.
4362 while (flush
== BTRFS_RESERVE_FLUSH_ALL
&& !flushing
&&
4363 space_info
->flush
) {
4364 spin_unlock(&space_info
->lock
);
4366 * If we have a trans handle we can't wait because the flusher
4367 * may have to commit the transaction, which would mean we would
4368 * deadlock since we are waiting for the flusher to finish, but
4369 * hold the current transaction open.
4371 if (current
->journal_info
)
4373 ret
= wait_event_killable(space_info
->wait
, !space_info
->flush
);
4374 /* Must have been killed, return */
4378 spin_lock(&space_info
->lock
);
4382 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4383 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4384 space_info
->bytes_may_use
;
4387 * The idea here is that we've not already over-reserved the block group
4388 * then we can go ahead and save our reservation first and then start
4389 * flushing if we need to. Otherwise if we've already overcommitted
4390 * lets start flushing stuff first and then come back and try to make
4393 if (used
<= space_info
->total_bytes
) {
4394 if (used
+ orig_bytes
<= space_info
->total_bytes
) {
4395 space_info
->bytes_may_use
+= orig_bytes
;
4396 trace_btrfs_space_reservation(root
->fs_info
,
4397 "space_info", space_info
->flags
, orig_bytes
, 1);
4401 * Ok set num_bytes to orig_bytes since we aren't
4402 * overocmmitted, this way we only try and reclaim what
4405 num_bytes
= orig_bytes
;
4409 * Ok we're over committed, set num_bytes to the overcommitted
4410 * amount plus the amount of bytes that we need for this
4413 num_bytes
= used
- space_info
->total_bytes
+
4417 if (ret
&& can_overcommit(root
, space_info
, orig_bytes
, flush
)) {
4418 space_info
->bytes_may_use
+= orig_bytes
;
4419 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
4420 space_info
->flags
, orig_bytes
,
4426 * Couldn't make our reservation, save our place so while we're trying
4427 * to reclaim space we can actually use it instead of somebody else
4428 * stealing it from us.
4430 * We make the other tasks wait for the flush only when we can flush
4433 if (ret
&& flush
!= BTRFS_RESERVE_NO_FLUSH
) {
4435 space_info
->flush
= 1;
4436 } else if (!ret
&& space_info
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
4439 * We will do the space reservation dance during log replay,
4440 * which means we won't have fs_info->fs_root set, so don't do
4441 * the async reclaim as we will panic.
4443 if (!root
->fs_info
->log_root_recovering
&&
4444 need_do_async_reclaim(space_info
, root
->fs_info
, used
) &&
4445 !work_busy(&root
->fs_info
->async_reclaim_work
))
4446 queue_work(system_unbound_wq
,
4447 &root
->fs_info
->async_reclaim_work
);
4449 spin_unlock(&space_info
->lock
);
4451 if (!ret
|| flush
== BTRFS_RESERVE_NO_FLUSH
)
4454 ret
= flush_space(root
, space_info
, num_bytes
, orig_bytes
,
4459 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4460 * would happen. So skip delalloc flush.
4462 if (flush
== BTRFS_RESERVE_FLUSH_LIMIT
&&
4463 (flush_state
== FLUSH_DELALLOC
||
4464 flush_state
== FLUSH_DELALLOC_WAIT
))
4465 flush_state
= ALLOC_CHUNK
;
4469 else if (flush
== BTRFS_RESERVE_FLUSH_LIMIT
&&
4470 flush_state
< COMMIT_TRANS
)
4472 else if (flush
== BTRFS_RESERVE_FLUSH_ALL
&&
4473 flush_state
<= COMMIT_TRANS
)
4477 if (ret
== -ENOSPC
&&
4478 unlikely(root
->orphan_cleanup_state
== ORPHAN_CLEANUP_STARTED
)) {
4479 struct btrfs_block_rsv
*global_rsv
=
4480 &root
->fs_info
->global_block_rsv
;
4482 if (block_rsv
!= global_rsv
&&
4483 !block_rsv_use_bytes(global_rsv
, orig_bytes
))
4487 trace_btrfs_space_reservation(root
->fs_info
,
4488 "space_info:enospc",
4489 space_info
->flags
, orig_bytes
, 1);
4491 spin_lock(&space_info
->lock
);
4492 space_info
->flush
= 0;
4493 wake_up_all(&space_info
->wait
);
4494 spin_unlock(&space_info
->lock
);
4499 static struct btrfs_block_rsv
*get_block_rsv(
4500 const struct btrfs_trans_handle
*trans
,
4501 const struct btrfs_root
*root
)
4503 struct btrfs_block_rsv
*block_rsv
= NULL
;
4505 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
))
4506 block_rsv
= trans
->block_rsv
;
4508 if (root
== root
->fs_info
->csum_root
&& trans
->adding_csums
)
4509 block_rsv
= trans
->block_rsv
;
4511 if (root
== root
->fs_info
->uuid_root
)
4512 block_rsv
= trans
->block_rsv
;
4515 block_rsv
= root
->block_rsv
;
4518 block_rsv
= &root
->fs_info
->empty_block_rsv
;
4523 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
4527 spin_lock(&block_rsv
->lock
);
4528 if (block_rsv
->reserved
>= num_bytes
) {
4529 block_rsv
->reserved
-= num_bytes
;
4530 if (block_rsv
->reserved
< block_rsv
->size
)
4531 block_rsv
->full
= 0;
4534 spin_unlock(&block_rsv
->lock
);
4538 static void block_rsv_add_bytes(struct btrfs_block_rsv
*block_rsv
,
4539 u64 num_bytes
, int update_size
)
4541 spin_lock(&block_rsv
->lock
);
4542 block_rsv
->reserved
+= num_bytes
;
4544 block_rsv
->size
+= num_bytes
;
4545 else if (block_rsv
->reserved
>= block_rsv
->size
)
4546 block_rsv
->full
= 1;
4547 spin_unlock(&block_rsv
->lock
);
4550 int btrfs_cond_migrate_bytes(struct btrfs_fs_info
*fs_info
,
4551 struct btrfs_block_rsv
*dest
, u64 num_bytes
,
4554 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
4557 if (global_rsv
->space_info
!= dest
->space_info
)
4560 spin_lock(&global_rsv
->lock
);
4561 min_bytes
= div_factor(global_rsv
->size
, min_factor
);
4562 if (global_rsv
->reserved
< min_bytes
+ num_bytes
) {
4563 spin_unlock(&global_rsv
->lock
);
4566 global_rsv
->reserved
-= num_bytes
;
4567 if (global_rsv
->reserved
< global_rsv
->size
)
4568 global_rsv
->full
= 0;
4569 spin_unlock(&global_rsv
->lock
);
4571 block_rsv_add_bytes(dest
, num_bytes
, 1);
4575 static void block_rsv_release_bytes(struct btrfs_fs_info
*fs_info
,
4576 struct btrfs_block_rsv
*block_rsv
,
4577 struct btrfs_block_rsv
*dest
, u64 num_bytes
)
4579 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
4581 spin_lock(&block_rsv
->lock
);
4582 if (num_bytes
== (u64
)-1)
4583 num_bytes
= block_rsv
->size
;
4584 block_rsv
->size
-= num_bytes
;
4585 if (block_rsv
->reserved
>= block_rsv
->size
) {
4586 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
4587 block_rsv
->reserved
= block_rsv
->size
;
4588 block_rsv
->full
= 1;
4592 spin_unlock(&block_rsv
->lock
);
4594 if (num_bytes
> 0) {
4596 spin_lock(&dest
->lock
);
4600 bytes_to_add
= dest
->size
- dest
->reserved
;
4601 bytes_to_add
= min(num_bytes
, bytes_to_add
);
4602 dest
->reserved
+= bytes_to_add
;
4603 if (dest
->reserved
>= dest
->size
)
4605 num_bytes
-= bytes_to_add
;
4607 spin_unlock(&dest
->lock
);
4610 spin_lock(&space_info
->lock
);
4611 space_info
->bytes_may_use
-= num_bytes
;
4612 trace_btrfs_space_reservation(fs_info
, "space_info",
4613 space_info
->flags
, num_bytes
, 0);
4614 spin_unlock(&space_info
->lock
);
4619 static int block_rsv_migrate_bytes(struct btrfs_block_rsv
*src
,
4620 struct btrfs_block_rsv
*dst
, u64 num_bytes
)
4624 ret
= block_rsv_use_bytes(src
, num_bytes
);
4628 block_rsv_add_bytes(dst
, num_bytes
, 1);
4632 void btrfs_init_block_rsv(struct btrfs_block_rsv
*rsv
, unsigned short type
)
4634 memset(rsv
, 0, sizeof(*rsv
));
4635 spin_lock_init(&rsv
->lock
);
4639 struct btrfs_block_rsv
*btrfs_alloc_block_rsv(struct btrfs_root
*root
,
4640 unsigned short type
)
4642 struct btrfs_block_rsv
*block_rsv
;
4643 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4645 block_rsv
= kmalloc(sizeof(*block_rsv
), GFP_NOFS
);
4649 btrfs_init_block_rsv(block_rsv
, type
);
4650 block_rsv
->space_info
= __find_space_info(fs_info
,
4651 BTRFS_BLOCK_GROUP_METADATA
);
4655 void btrfs_free_block_rsv(struct btrfs_root
*root
,
4656 struct btrfs_block_rsv
*rsv
)
4660 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
4664 int btrfs_block_rsv_add(struct btrfs_root
*root
,
4665 struct btrfs_block_rsv
*block_rsv
, u64 num_bytes
,
4666 enum btrfs_reserve_flush_enum flush
)
4673 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
4675 block_rsv_add_bytes(block_rsv
, num_bytes
, 1);
4682 int btrfs_block_rsv_check(struct btrfs_root
*root
,
4683 struct btrfs_block_rsv
*block_rsv
, int min_factor
)
4691 spin_lock(&block_rsv
->lock
);
4692 num_bytes
= div_factor(block_rsv
->size
, min_factor
);
4693 if (block_rsv
->reserved
>= num_bytes
)
4695 spin_unlock(&block_rsv
->lock
);
4700 int btrfs_block_rsv_refill(struct btrfs_root
*root
,
4701 struct btrfs_block_rsv
*block_rsv
, u64 min_reserved
,
4702 enum btrfs_reserve_flush_enum flush
)
4710 spin_lock(&block_rsv
->lock
);
4711 num_bytes
= min_reserved
;
4712 if (block_rsv
->reserved
>= num_bytes
)
4715 num_bytes
-= block_rsv
->reserved
;
4716 spin_unlock(&block_rsv
->lock
);
4721 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
4723 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
4730 int btrfs_block_rsv_migrate(struct btrfs_block_rsv
*src_rsv
,
4731 struct btrfs_block_rsv
*dst_rsv
,
4734 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
4737 void btrfs_block_rsv_release(struct btrfs_root
*root
,
4738 struct btrfs_block_rsv
*block_rsv
,
4741 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
4742 if (global_rsv
== block_rsv
||
4743 block_rsv
->space_info
!= global_rsv
->space_info
)
4745 block_rsv_release_bytes(root
->fs_info
, block_rsv
, global_rsv
,
4750 * helper to calculate size of global block reservation.
4751 * the desired value is sum of space used by extent tree,
4752 * checksum tree and root tree
4754 static u64
calc_global_metadata_size(struct btrfs_fs_info
*fs_info
)
4756 struct btrfs_space_info
*sinfo
;
4760 int csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
4762 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_DATA
);
4763 spin_lock(&sinfo
->lock
);
4764 data_used
= sinfo
->bytes_used
;
4765 spin_unlock(&sinfo
->lock
);
4767 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4768 spin_lock(&sinfo
->lock
);
4769 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_DATA
)
4771 meta_used
= sinfo
->bytes_used
;
4772 spin_unlock(&sinfo
->lock
);
4774 num_bytes
= (data_used
>> fs_info
->sb
->s_blocksize_bits
) *
4776 num_bytes
+= div64_u64(data_used
+ meta_used
, 50);
4778 if (num_bytes
* 3 > meta_used
)
4779 num_bytes
= div64_u64(meta_used
, 3);
4781 return ALIGN(num_bytes
, fs_info
->extent_root
->nodesize
<< 10);
4784 static void update_global_block_rsv(struct btrfs_fs_info
*fs_info
)
4786 struct btrfs_block_rsv
*block_rsv
= &fs_info
->global_block_rsv
;
4787 struct btrfs_space_info
*sinfo
= block_rsv
->space_info
;
4790 num_bytes
= calc_global_metadata_size(fs_info
);
4792 spin_lock(&sinfo
->lock
);
4793 spin_lock(&block_rsv
->lock
);
4795 block_rsv
->size
= min_t(u64
, num_bytes
, 512 * 1024 * 1024);
4797 num_bytes
= sinfo
->bytes_used
+ sinfo
->bytes_pinned
+
4798 sinfo
->bytes_reserved
+ sinfo
->bytes_readonly
+
4799 sinfo
->bytes_may_use
;
4801 if (sinfo
->total_bytes
> num_bytes
) {
4802 num_bytes
= sinfo
->total_bytes
- num_bytes
;
4803 block_rsv
->reserved
+= num_bytes
;
4804 sinfo
->bytes_may_use
+= num_bytes
;
4805 trace_btrfs_space_reservation(fs_info
, "space_info",
4806 sinfo
->flags
, num_bytes
, 1);
4809 if (block_rsv
->reserved
>= block_rsv
->size
) {
4810 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
4811 sinfo
->bytes_may_use
-= num_bytes
;
4812 trace_btrfs_space_reservation(fs_info
, "space_info",
4813 sinfo
->flags
, num_bytes
, 0);
4814 block_rsv
->reserved
= block_rsv
->size
;
4815 block_rsv
->full
= 1;
4818 spin_unlock(&block_rsv
->lock
);
4819 spin_unlock(&sinfo
->lock
);
4822 static void init_global_block_rsv(struct btrfs_fs_info
*fs_info
)
4824 struct btrfs_space_info
*space_info
;
4826 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
4827 fs_info
->chunk_block_rsv
.space_info
= space_info
;
4829 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4830 fs_info
->global_block_rsv
.space_info
= space_info
;
4831 fs_info
->delalloc_block_rsv
.space_info
= space_info
;
4832 fs_info
->trans_block_rsv
.space_info
= space_info
;
4833 fs_info
->empty_block_rsv
.space_info
= space_info
;
4834 fs_info
->delayed_block_rsv
.space_info
= space_info
;
4836 fs_info
->extent_root
->block_rsv
= &fs_info
->global_block_rsv
;
4837 fs_info
->csum_root
->block_rsv
= &fs_info
->global_block_rsv
;
4838 fs_info
->dev_root
->block_rsv
= &fs_info
->global_block_rsv
;
4839 fs_info
->tree_root
->block_rsv
= &fs_info
->global_block_rsv
;
4840 if (fs_info
->quota_root
)
4841 fs_info
->quota_root
->block_rsv
= &fs_info
->global_block_rsv
;
4842 fs_info
->chunk_root
->block_rsv
= &fs_info
->chunk_block_rsv
;
4844 update_global_block_rsv(fs_info
);
4847 static void release_global_block_rsv(struct btrfs_fs_info
*fs_info
)
4849 block_rsv_release_bytes(fs_info
, &fs_info
->global_block_rsv
, NULL
,
4851 WARN_ON(fs_info
->delalloc_block_rsv
.size
> 0);
4852 WARN_ON(fs_info
->delalloc_block_rsv
.reserved
> 0);
4853 WARN_ON(fs_info
->trans_block_rsv
.size
> 0);
4854 WARN_ON(fs_info
->trans_block_rsv
.reserved
> 0);
4855 WARN_ON(fs_info
->chunk_block_rsv
.size
> 0);
4856 WARN_ON(fs_info
->chunk_block_rsv
.reserved
> 0);
4857 WARN_ON(fs_info
->delayed_block_rsv
.size
> 0);
4858 WARN_ON(fs_info
->delayed_block_rsv
.reserved
> 0);
4861 void btrfs_trans_release_metadata(struct btrfs_trans_handle
*trans
,
4862 struct btrfs_root
*root
)
4864 if (!trans
->block_rsv
)
4867 if (!trans
->bytes_reserved
)
4870 trace_btrfs_space_reservation(root
->fs_info
, "transaction",
4871 trans
->transid
, trans
->bytes_reserved
, 0);
4872 btrfs_block_rsv_release(root
, trans
->block_rsv
, trans
->bytes_reserved
);
4873 trans
->bytes_reserved
= 0;
4876 /* Can only return 0 or -ENOSPC */
4877 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle
*trans
,
4878 struct inode
*inode
)
4880 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4881 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
4882 struct btrfs_block_rsv
*dst_rsv
= root
->orphan_block_rsv
;
4885 * We need to hold space in order to delete our orphan item once we've
4886 * added it, so this takes the reservation so we can release it later
4887 * when we are truly done with the orphan item.
4889 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
4890 trace_btrfs_space_reservation(root
->fs_info
, "orphan",
4891 btrfs_ino(inode
), num_bytes
, 1);
4892 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
4895 void btrfs_orphan_release_metadata(struct inode
*inode
)
4897 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4898 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
4899 trace_btrfs_space_reservation(root
->fs_info
, "orphan",
4900 btrfs_ino(inode
), num_bytes
, 0);
4901 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
, num_bytes
);
4905 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4906 * root: the root of the parent directory
4907 * rsv: block reservation
4908 * items: the number of items that we need do reservation
4909 * qgroup_reserved: used to return the reserved size in qgroup
4911 * This function is used to reserve the space for snapshot/subvolume
4912 * creation and deletion. Those operations are different with the
4913 * common file/directory operations, they change two fs/file trees
4914 * and root tree, the number of items that the qgroup reserves is
4915 * different with the free space reservation. So we can not use
4916 * the space reseravtion mechanism in start_transaction().
4918 int btrfs_subvolume_reserve_metadata(struct btrfs_root
*root
,
4919 struct btrfs_block_rsv
*rsv
,
4921 u64
*qgroup_reserved
,
4922 bool use_global_rsv
)
4926 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
4928 if (root
->fs_info
->quota_enabled
) {
4929 /* One for parent inode, two for dir entries */
4930 num_bytes
= 3 * root
->nodesize
;
4931 ret
= btrfs_qgroup_reserve(root
, num_bytes
);
4938 *qgroup_reserved
= num_bytes
;
4940 num_bytes
= btrfs_calc_trans_metadata_size(root
, items
);
4941 rsv
->space_info
= __find_space_info(root
->fs_info
,
4942 BTRFS_BLOCK_GROUP_METADATA
);
4943 ret
= btrfs_block_rsv_add(root
, rsv
, num_bytes
,
4944 BTRFS_RESERVE_FLUSH_ALL
);
4946 if (ret
== -ENOSPC
&& use_global_rsv
)
4947 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
, num_bytes
);
4950 if (*qgroup_reserved
)
4951 btrfs_qgroup_free(root
, *qgroup_reserved
);
4957 void btrfs_subvolume_release_metadata(struct btrfs_root
*root
,
4958 struct btrfs_block_rsv
*rsv
,
4959 u64 qgroup_reserved
)
4961 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
4962 if (qgroup_reserved
)
4963 btrfs_qgroup_free(root
, qgroup_reserved
);
4967 * drop_outstanding_extent - drop an outstanding extent
4968 * @inode: the inode we're dropping the extent for
4970 * This is called when we are freeing up an outstanding extent, either called
4971 * after an error or after an extent is written. This will return the number of
4972 * reserved extents that need to be freed. This must be called with
4973 * BTRFS_I(inode)->lock held.
4975 static unsigned drop_outstanding_extent(struct inode
*inode
)
4977 unsigned drop_inode_space
= 0;
4978 unsigned dropped_extents
= 0;
4980 BUG_ON(!BTRFS_I(inode
)->outstanding_extents
);
4981 BTRFS_I(inode
)->outstanding_extents
--;
4983 if (BTRFS_I(inode
)->outstanding_extents
== 0 &&
4984 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
4985 &BTRFS_I(inode
)->runtime_flags
))
4986 drop_inode_space
= 1;
4989 * If we have more or the same amount of outsanding extents than we have
4990 * reserved then we need to leave the reserved extents count alone.
4992 if (BTRFS_I(inode
)->outstanding_extents
>=
4993 BTRFS_I(inode
)->reserved_extents
)
4994 return drop_inode_space
;
4996 dropped_extents
= BTRFS_I(inode
)->reserved_extents
-
4997 BTRFS_I(inode
)->outstanding_extents
;
4998 BTRFS_I(inode
)->reserved_extents
-= dropped_extents
;
4999 return dropped_extents
+ drop_inode_space
;
5003 * calc_csum_metadata_size - return the amount of metada space that must be
5004 * reserved/free'd for the given bytes.
5005 * @inode: the inode we're manipulating
5006 * @num_bytes: the number of bytes in question
5007 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5009 * This adjusts the number of csum_bytes in the inode and then returns the
5010 * correct amount of metadata that must either be reserved or freed. We
5011 * calculate how many checksums we can fit into one leaf and then divide the
5012 * number of bytes that will need to be checksumed by this value to figure out
5013 * how many checksums will be required. If we are adding bytes then the number
5014 * may go up and we will return the number of additional bytes that must be
5015 * reserved. If it is going down we will return the number of bytes that must
5018 * This must be called with BTRFS_I(inode)->lock held.
5020 static u64
calc_csum_metadata_size(struct inode
*inode
, u64 num_bytes
,
5023 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5025 int num_csums_per_leaf
;
5029 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
&&
5030 BTRFS_I(inode
)->csum_bytes
== 0)
5033 old_csums
= (int)div64_u64(BTRFS_I(inode
)->csum_bytes
, root
->sectorsize
);
5035 BTRFS_I(inode
)->csum_bytes
+= num_bytes
;
5037 BTRFS_I(inode
)->csum_bytes
-= num_bytes
;
5038 csum_size
= BTRFS_LEAF_DATA_SIZE(root
) - sizeof(struct btrfs_item
);
5039 num_csums_per_leaf
= (int)div64_u64(csum_size
,
5040 sizeof(struct btrfs_csum_item
) +
5041 sizeof(struct btrfs_disk_key
));
5042 num_csums
= (int)div64_u64(BTRFS_I(inode
)->csum_bytes
, root
->sectorsize
);
5043 num_csums
= num_csums
+ num_csums_per_leaf
- 1;
5044 num_csums
= num_csums
/ num_csums_per_leaf
;
5046 old_csums
= old_csums
+ num_csums_per_leaf
- 1;
5047 old_csums
= old_csums
/ num_csums_per_leaf
;
5049 /* No change, no need to reserve more */
5050 if (old_csums
== num_csums
)
5054 return btrfs_calc_trans_metadata_size(root
,
5055 num_csums
- old_csums
);
5057 return btrfs_calc_trans_metadata_size(root
, old_csums
- num_csums
);
5060 int btrfs_delalloc_reserve_metadata(struct inode
*inode
, u64 num_bytes
)
5062 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5063 struct btrfs_block_rsv
*block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
5066 unsigned nr_extents
= 0;
5067 int extra_reserve
= 0;
5068 enum btrfs_reserve_flush_enum flush
= BTRFS_RESERVE_FLUSH_ALL
;
5070 bool delalloc_lock
= true;
5074 /* If we are a free space inode we need to not flush since we will be in
5075 * the middle of a transaction commit. We also don't need the delalloc
5076 * mutex since we won't race with anybody. We need this mostly to make
5077 * lockdep shut its filthy mouth.
5079 if (btrfs_is_free_space_inode(inode
)) {
5080 flush
= BTRFS_RESERVE_NO_FLUSH
;
5081 delalloc_lock
= false;
5084 if (flush
!= BTRFS_RESERVE_NO_FLUSH
&&
5085 btrfs_transaction_in_commit(root
->fs_info
))
5086 schedule_timeout(1);
5089 mutex_lock(&BTRFS_I(inode
)->delalloc_mutex
);
5091 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
5093 spin_lock(&BTRFS_I(inode
)->lock
);
5094 BTRFS_I(inode
)->outstanding_extents
++;
5096 if (BTRFS_I(inode
)->outstanding_extents
>
5097 BTRFS_I(inode
)->reserved_extents
)
5098 nr_extents
= BTRFS_I(inode
)->outstanding_extents
-
5099 BTRFS_I(inode
)->reserved_extents
;
5102 * Add an item to reserve for updating the inode when we complete the
5105 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5106 &BTRFS_I(inode
)->runtime_flags
)) {
5111 to_reserve
= btrfs_calc_trans_metadata_size(root
, nr_extents
);
5112 to_reserve
+= calc_csum_metadata_size(inode
, num_bytes
, 1);
5113 csum_bytes
= BTRFS_I(inode
)->csum_bytes
;
5114 spin_unlock(&BTRFS_I(inode
)->lock
);
5116 if (root
->fs_info
->quota_enabled
) {
5117 ret
= btrfs_qgroup_reserve(root
, num_bytes
+
5118 nr_extents
* root
->nodesize
);
5123 ret
= reserve_metadata_bytes(root
, block_rsv
, to_reserve
, flush
);
5124 if (unlikely(ret
)) {
5125 if (root
->fs_info
->quota_enabled
)
5126 btrfs_qgroup_free(root
, num_bytes
+
5127 nr_extents
* root
->nodesize
);
5131 spin_lock(&BTRFS_I(inode
)->lock
);
5132 if (extra_reserve
) {
5133 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5134 &BTRFS_I(inode
)->runtime_flags
);
5137 BTRFS_I(inode
)->reserved_extents
+= nr_extents
;
5138 spin_unlock(&BTRFS_I(inode
)->lock
);
5141 mutex_unlock(&BTRFS_I(inode
)->delalloc_mutex
);
5144 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5145 btrfs_ino(inode
), to_reserve
, 1);
5146 block_rsv_add_bytes(block_rsv
, to_reserve
, 1);
5151 spin_lock(&BTRFS_I(inode
)->lock
);
5152 dropped
= drop_outstanding_extent(inode
);
5154 * If the inodes csum_bytes is the same as the original
5155 * csum_bytes then we know we haven't raced with any free()ers
5156 * so we can just reduce our inodes csum bytes and carry on.
5158 if (BTRFS_I(inode
)->csum_bytes
== csum_bytes
) {
5159 calc_csum_metadata_size(inode
, num_bytes
, 0);
5161 u64 orig_csum_bytes
= BTRFS_I(inode
)->csum_bytes
;
5165 * This is tricky, but first we need to figure out how much we
5166 * free'd from any free-ers that occured during this
5167 * reservation, so we reset ->csum_bytes to the csum_bytes
5168 * before we dropped our lock, and then call the free for the
5169 * number of bytes that were freed while we were trying our
5172 bytes
= csum_bytes
- BTRFS_I(inode
)->csum_bytes
;
5173 BTRFS_I(inode
)->csum_bytes
= csum_bytes
;
5174 to_free
= calc_csum_metadata_size(inode
, bytes
, 0);
5178 * Now we need to see how much we would have freed had we not
5179 * been making this reservation and our ->csum_bytes were not
5180 * artificially inflated.
5182 BTRFS_I(inode
)->csum_bytes
= csum_bytes
- num_bytes
;
5183 bytes
= csum_bytes
- orig_csum_bytes
;
5184 bytes
= calc_csum_metadata_size(inode
, bytes
, 0);
5187 * Now reset ->csum_bytes to what it should be. If bytes is
5188 * more than to_free then we would have free'd more space had we
5189 * not had an artificially high ->csum_bytes, so we need to free
5190 * the remainder. If bytes is the same or less then we don't
5191 * need to do anything, the other free-ers did the correct
5194 BTRFS_I(inode
)->csum_bytes
= orig_csum_bytes
- num_bytes
;
5195 if (bytes
> to_free
)
5196 to_free
= bytes
- to_free
;
5200 spin_unlock(&BTRFS_I(inode
)->lock
);
5202 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
5205 btrfs_block_rsv_release(root
, block_rsv
, to_free
);
5206 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5207 btrfs_ino(inode
), to_free
, 0);
5210 mutex_unlock(&BTRFS_I(inode
)->delalloc_mutex
);
5215 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5216 * @inode: the inode to release the reservation for
5217 * @num_bytes: the number of bytes we're releasing
5219 * This will release the metadata reservation for an inode. This can be called
5220 * once we complete IO for a given set of bytes to release their metadata
5223 void btrfs_delalloc_release_metadata(struct inode
*inode
, u64 num_bytes
)
5225 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5229 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
5230 spin_lock(&BTRFS_I(inode
)->lock
);
5231 dropped
= drop_outstanding_extent(inode
);
5234 to_free
= calc_csum_metadata_size(inode
, num_bytes
, 0);
5235 spin_unlock(&BTRFS_I(inode
)->lock
);
5237 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
5239 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5240 btrfs_ino(inode
), to_free
, 0);
5241 if (root
->fs_info
->quota_enabled
) {
5242 btrfs_qgroup_free(root
, num_bytes
+
5243 dropped
* root
->nodesize
);
5246 btrfs_block_rsv_release(root
, &root
->fs_info
->delalloc_block_rsv
,
5251 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5252 * @inode: inode we're writing to
5253 * @num_bytes: the number of bytes we want to allocate
5255 * This will do the following things
5257 * o reserve space in the data space info for num_bytes
5258 * o reserve space in the metadata space info based on number of outstanding
5259 * extents and how much csums will be needed
5260 * o add to the inodes ->delalloc_bytes
5261 * o add it to the fs_info's delalloc inodes list.
5263 * This will return 0 for success and -ENOSPC if there is no space left.
5265 int btrfs_delalloc_reserve_space(struct inode
*inode
, u64 num_bytes
)
5269 ret
= btrfs_check_data_free_space(inode
, num_bytes
);
5273 ret
= btrfs_delalloc_reserve_metadata(inode
, num_bytes
);
5275 btrfs_free_reserved_data_space(inode
, num_bytes
);
5283 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5284 * @inode: inode we're releasing space for
5285 * @num_bytes: the number of bytes we want to free up
5287 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5288 * called in the case that we don't need the metadata AND data reservations
5289 * anymore. So if there is an error or we insert an inline extent.
5291 * This function will release the metadata space that was not used and will
5292 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5293 * list if there are no delalloc bytes left.
5295 void btrfs_delalloc_release_space(struct inode
*inode
, u64 num_bytes
)
5297 btrfs_delalloc_release_metadata(inode
, num_bytes
);
5298 btrfs_free_reserved_data_space(inode
, num_bytes
);
5301 static int update_block_group(struct btrfs_trans_handle
*trans
,
5302 struct btrfs_root
*root
, u64 bytenr
,
5303 u64 num_bytes
, int alloc
)
5305 struct btrfs_block_group_cache
*cache
= NULL
;
5306 struct btrfs_fs_info
*info
= root
->fs_info
;
5307 u64 total
= num_bytes
;
5312 /* block accounting for super block */
5313 spin_lock(&info
->delalloc_root_lock
);
5314 old_val
= btrfs_super_bytes_used(info
->super_copy
);
5316 old_val
+= num_bytes
;
5318 old_val
-= num_bytes
;
5319 btrfs_set_super_bytes_used(info
->super_copy
, old_val
);
5320 spin_unlock(&info
->delalloc_root_lock
);
5323 cache
= btrfs_lookup_block_group(info
, bytenr
);
5326 if (cache
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
5327 BTRFS_BLOCK_GROUP_RAID1
|
5328 BTRFS_BLOCK_GROUP_RAID10
))
5333 * If this block group has free space cache written out, we
5334 * need to make sure to load it if we are removing space. This
5335 * is because we need the unpinning stage to actually add the
5336 * space back to the block group, otherwise we will leak space.
5338 if (!alloc
&& cache
->cached
== BTRFS_CACHE_NO
)
5339 cache_block_group(cache
, 1);
5341 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
5342 if (list_empty(&cache
->dirty_list
)) {
5343 list_add_tail(&cache
->dirty_list
,
5344 &trans
->transaction
->dirty_bgs
);
5345 btrfs_get_block_group(cache
);
5347 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
5349 byte_in_group
= bytenr
- cache
->key
.objectid
;
5350 WARN_ON(byte_in_group
> cache
->key
.offset
);
5352 spin_lock(&cache
->space_info
->lock
);
5353 spin_lock(&cache
->lock
);
5355 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
5356 cache
->disk_cache_state
< BTRFS_DC_CLEAR
)
5357 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
5359 old_val
= btrfs_block_group_used(&cache
->item
);
5360 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
5362 old_val
+= num_bytes
;
5363 btrfs_set_block_group_used(&cache
->item
, old_val
);
5364 cache
->reserved
-= num_bytes
;
5365 cache
->space_info
->bytes_reserved
-= num_bytes
;
5366 cache
->space_info
->bytes_used
+= num_bytes
;
5367 cache
->space_info
->disk_used
+= num_bytes
* factor
;
5368 spin_unlock(&cache
->lock
);
5369 spin_unlock(&cache
->space_info
->lock
);
5371 old_val
-= num_bytes
;
5372 btrfs_set_block_group_used(&cache
->item
, old_val
);
5373 cache
->pinned
+= num_bytes
;
5374 cache
->space_info
->bytes_pinned
+= num_bytes
;
5375 cache
->space_info
->bytes_used
-= num_bytes
;
5376 cache
->space_info
->disk_used
-= num_bytes
* factor
;
5377 spin_unlock(&cache
->lock
);
5378 spin_unlock(&cache
->space_info
->lock
);
5380 set_extent_dirty(info
->pinned_extents
,
5381 bytenr
, bytenr
+ num_bytes
- 1,
5382 GFP_NOFS
| __GFP_NOFAIL
);
5384 * No longer have used bytes in this block group, queue
5388 spin_lock(&info
->unused_bgs_lock
);
5389 if (list_empty(&cache
->bg_list
)) {
5390 btrfs_get_block_group(cache
);
5391 list_add_tail(&cache
->bg_list
,
5394 spin_unlock(&info
->unused_bgs_lock
);
5397 btrfs_put_block_group(cache
);
5399 bytenr
+= num_bytes
;
5404 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
5406 struct btrfs_block_group_cache
*cache
;
5409 spin_lock(&root
->fs_info
->block_group_cache_lock
);
5410 bytenr
= root
->fs_info
->first_logical_byte
;
5411 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
5413 if (bytenr
< (u64
)-1)
5416 cache
= btrfs_lookup_first_block_group(root
->fs_info
, search_start
);
5420 bytenr
= cache
->key
.objectid
;
5421 btrfs_put_block_group(cache
);
5426 static int pin_down_extent(struct btrfs_root
*root
,
5427 struct btrfs_block_group_cache
*cache
,
5428 u64 bytenr
, u64 num_bytes
, int reserved
)
5430 spin_lock(&cache
->space_info
->lock
);
5431 spin_lock(&cache
->lock
);
5432 cache
->pinned
+= num_bytes
;
5433 cache
->space_info
->bytes_pinned
+= num_bytes
;
5435 cache
->reserved
-= num_bytes
;
5436 cache
->space_info
->bytes_reserved
-= num_bytes
;
5438 spin_unlock(&cache
->lock
);
5439 spin_unlock(&cache
->space_info
->lock
);
5441 set_extent_dirty(root
->fs_info
->pinned_extents
, bytenr
,
5442 bytenr
+ num_bytes
- 1, GFP_NOFS
| __GFP_NOFAIL
);
5444 trace_btrfs_reserved_extent_free(root
, bytenr
, num_bytes
);
5449 * this function must be called within transaction
5451 int btrfs_pin_extent(struct btrfs_root
*root
,
5452 u64 bytenr
, u64 num_bytes
, int reserved
)
5454 struct btrfs_block_group_cache
*cache
;
5456 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
5457 BUG_ON(!cache
); /* Logic error */
5459 pin_down_extent(root
, cache
, bytenr
, num_bytes
, reserved
);
5461 btrfs_put_block_group(cache
);
5466 * this function must be called within transaction
5468 int btrfs_pin_extent_for_log_replay(struct btrfs_root
*root
,
5469 u64 bytenr
, u64 num_bytes
)
5471 struct btrfs_block_group_cache
*cache
;
5474 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
5479 * pull in the free space cache (if any) so that our pin
5480 * removes the free space from the cache. We have load_only set
5481 * to one because the slow code to read in the free extents does check
5482 * the pinned extents.
5484 cache_block_group(cache
, 1);
5486 pin_down_extent(root
, cache
, bytenr
, num_bytes
, 0);
5488 /* remove us from the free space cache (if we're there at all) */
5489 ret
= btrfs_remove_free_space(cache
, bytenr
, num_bytes
);
5490 btrfs_put_block_group(cache
);
5494 static int __exclude_logged_extent(struct btrfs_root
*root
, u64 start
, u64 num_bytes
)
5497 struct btrfs_block_group_cache
*block_group
;
5498 struct btrfs_caching_control
*caching_ctl
;
5500 block_group
= btrfs_lookup_block_group(root
->fs_info
, start
);
5504 cache_block_group(block_group
, 0);
5505 caching_ctl
= get_caching_control(block_group
);
5509 BUG_ON(!block_group_cache_done(block_group
));
5510 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
5512 mutex_lock(&caching_ctl
->mutex
);
5514 if (start
>= caching_ctl
->progress
) {
5515 ret
= add_excluded_extent(root
, start
, num_bytes
);
5516 } else if (start
+ num_bytes
<= caching_ctl
->progress
) {
5517 ret
= btrfs_remove_free_space(block_group
,
5520 num_bytes
= caching_ctl
->progress
- start
;
5521 ret
= btrfs_remove_free_space(block_group
,
5526 num_bytes
= (start
+ num_bytes
) -
5527 caching_ctl
->progress
;
5528 start
= caching_ctl
->progress
;
5529 ret
= add_excluded_extent(root
, start
, num_bytes
);
5532 mutex_unlock(&caching_ctl
->mutex
);
5533 put_caching_control(caching_ctl
);
5535 btrfs_put_block_group(block_group
);
5539 int btrfs_exclude_logged_extents(struct btrfs_root
*log
,
5540 struct extent_buffer
*eb
)
5542 struct btrfs_file_extent_item
*item
;
5543 struct btrfs_key key
;
5547 if (!btrfs_fs_incompat(log
->fs_info
, MIXED_GROUPS
))
5550 for (i
= 0; i
< btrfs_header_nritems(eb
); i
++) {
5551 btrfs_item_key_to_cpu(eb
, &key
, i
);
5552 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
5554 item
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
5555 found_type
= btrfs_file_extent_type(eb
, item
);
5556 if (found_type
== BTRFS_FILE_EXTENT_INLINE
)
5558 if (btrfs_file_extent_disk_bytenr(eb
, item
) == 0)
5560 key
.objectid
= btrfs_file_extent_disk_bytenr(eb
, item
);
5561 key
.offset
= btrfs_file_extent_disk_num_bytes(eb
, item
);
5562 __exclude_logged_extent(log
, key
.objectid
, key
.offset
);
5569 * btrfs_update_reserved_bytes - update the block_group and space info counters
5570 * @cache: The cache we are manipulating
5571 * @num_bytes: The number of bytes in question
5572 * @reserve: One of the reservation enums
5573 * @delalloc: The blocks are allocated for the delalloc write
5575 * This is called by the allocator when it reserves space, or by somebody who is
5576 * freeing space that was never actually used on disk. For example if you
5577 * reserve some space for a new leaf in transaction A and before transaction A
5578 * commits you free that leaf, you call this with reserve set to 0 in order to
5579 * clear the reservation.
5581 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5582 * ENOSPC accounting. For data we handle the reservation through clearing the
5583 * delalloc bits in the io_tree. We have to do this since we could end up
5584 * allocating less disk space for the amount of data we have reserved in the
5585 * case of compression.
5587 * If this is a reservation and the block group has become read only we cannot
5588 * make the reservation and return -EAGAIN, otherwise this function always
5591 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
5592 u64 num_bytes
, int reserve
, int delalloc
)
5594 struct btrfs_space_info
*space_info
= cache
->space_info
;
5597 spin_lock(&space_info
->lock
);
5598 spin_lock(&cache
->lock
);
5599 if (reserve
!= RESERVE_FREE
) {
5603 cache
->reserved
+= num_bytes
;
5604 space_info
->bytes_reserved
+= num_bytes
;
5605 if (reserve
== RESERVE_ALLOC
) {
5606 trace_btrfs_space_reservation(cache
->fs_info
,
5607 "space_info", space_info
->flags
,
5609 space_info
->bytes_may_use
-= num_bytes
;
5613 cache
->delalloc_bytes
+= num_bytes
;
5617 space_info
->bytes_readonly
+= num_bytes
;
5618 cache
->reserved
-= num_bytes
;
5619 space_info
->bytes_reserved
-= num_bytes
;
5622 cache
->delalloc_bytes
-= num_bytes
;
5624 spin_unlock(&cache
->lock
);
5625 spin_unlock(&space_info
->lock
);
5629 void btrfs_prepare_extent_commit(struct btrfs_trans_handle
*trans
,
5630 struct btrfs_root
*root
)
5632 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5633 struct btrfs_caching_control
*next
;
5634 struct btrfs_caching_control
*caching_ctl
;
5635 struct btrfs_block_group_cache
*cache
;
5637 down_write(&fs_info
->commit_root_sem
);
5639 list_for_each_entry_safe(caching_ctl
, next
,
5640 &fs_info
->caching_block_groups
, list
) {
5641 cache
= caching_ctl
->block_group
;
5642 if (block_group_cache_done(cache
)) {
5643 cache
->last_byte_to_unpin
= (u64
)-1;
5644 list_del_init(&caching_ctl
->list
);
5645 put_caching_control(caching_ctl
);
5647 cache
->last_byte_to_unpin
= caching_ctl
->progress
;
5651 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
5652 fs_info
->pinned_extents
= &fs_info
->freed_extents
[1];
5654 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
5656 up_write(&fs_info
->commit_root_sem
);
5658 update_global_block_rsv(fs_info
);
5661 static int unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
,
5662 const bool return_free_space
)
5664 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5665 struct btrfs_block_group_cache
*cache
= NULL
;
5666 struct btrfs_space_info
*space_info
;
5667 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5671 while (start
<= end
) {
5674 start
>= cache
->key
.objectid
+ cache
->key
.offset
) {
5676 btrfs_put_block_group(cache
);
5677 cache
= btrfs_lookup_block_group(fs_info
, start
);
5678 BUG_ON(!cache
); /* Logic error */
5681 len
= cache
->key
.objectid
+ cache
->key
.offset
- start
;
5682 len
= min(len
, end
+ 1 - start
);
5684 if (start
< cache
->last_byte_to_unpin
) {
5685 len
= min(len
, cache
->last_byte_to_unpin
- start
);
5686 if (return_free_space
)
5687 btrfs_add_free_space(cache
, start
, len
);
5691 space_info
= cache
->space_info
;
5693 spin_lock(&space_info
->lock
);
5694 spin_lock(&cache
->lock
);
5695 cache
->pinned
-= len
;
5696 space_info
->bytes_pinned
-= len
;
5697 percpu_counter_add(&space_info
->total_bytes_pinned
, -len
);
5699 space_info
->bytes_readonly
+= len
;
5702 spin_unlock(&cache
->lock
);
5703 if (!readonly
&& global_rsv
->space_info
== space_info
) {
5704 spin_lock(&global_rsv
->lock
);
5705 if (!global_rsv
->full
) {
5706 len
= min(len
, global_rsv
->size
-
5707 global_rsv
->reserved
);
5708 global_rsv
->reserved
+= len
;
5709 space_info
->bytes_may_use
+= len
;
5710 if (global_rsv
->reserved
>= global_rsv
->size
)
5711 global_rsv
->full
= 1;
5713 spin_unlock(&global_rsv
->lock
);
5715 spin_unlock(&space_info
->lock
);
5719 btrfs_put_block_group(cache
);
5723 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
5724 struct btrfs_root
*root
)
5726 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5727 struct extent_io_tree
*unpin
;
5735 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
5736 unpin
= &fs_info
->freed_extents
[1];
5738 unpin
= &fs_info
->freed_extents
[0];
5741 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
5742 EXTENT_DIRTY
, NULL
);
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);
5758 static void add_pinned_bytes(struct btrfs_fs_info
*fs_info
, u64 num_bytes
,
5759 u64 owner
, u64 root_objectid
)
5761 struct btrfs_space_info
*space_info
;
5764 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
5765 if (root_objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
5766 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
5768 flags
= BTRFS_BLOCK_GROUP_METADATA
;
5770 flags
= BTRFS_BLOCK_GROUP_DATA
;
5773 space_info
= __find_space_info(fs_info
, flags
);
5774 BUG_ON(!space_info
); /* Logic bug */
5775 percpu_counter_add(&space_info
->total_bytes_pinned
, num_bytes
);
5779 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
5780 struct btrfs_root
*root
,
5781 u64 bytenr
, u64 num_bytes
, u64 parent
,
5782 u64 root_objectid
, u64 owner_objectid
,
5783 u64 owner_offset
, int refs_to_drop
,
5784 struct btrfs_delayed_extent_op
*extent_op
,
5787 struct btrfs_key key
;
5788 struct btrfs_path
*path
;
5789 struct btrfs_fs_info
*info
= root
->fs_info
;
5790 struct btrfs_root
*extent_root
= info
->extent_root
;
5791 struct extent_buffer
*leaf
;
5792 struct btrfs_extent_item
*ei
;
5793 struct btrfs_extent_inline_ref
*iref
;
5796 int extent_slot
= 0;
5797 int found_extent
= 0;
5802 enum btrfs_qgroup_operation_type type
= BTRFS_QGROUP_OPER_SUB_EXCL
;
5803 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
5806 if (!info
->quota_enabled
|| !is_fstree(root_objectid
))
5809 path
= btrfs_alloc_path();
5814 path
->leave_spinning
= 1;
5816 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
5817 BUG_ON(!is_data
&& refs_to_drop
!= 1);
5820 skinny_metadata
= 0;
5822 ret
= lookup_extent_backref(trans
, extent_root
, path
, &iref
,
5823 bytenr
, num_bytes
, parent
,
5824 root_objectid
, owner_objectid
,
5827 extent_slot
= path
->slots
[0];
5828 while (extent_slot
>= 0) {
5829 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
5831 if (key
.objectid
!= bytenr
)
5833 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
5834 key
.offset
== num_bytes
) {
5838 if (key
.type
== BTRFS_METADATA_ITEM_KEY
&&
5839 key
.offset
== owner_objectid
) {
5843 if (path
->slots
[0] - extent_slot
> 5)
5847 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5848 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
5849 if (found_extent
&& item_size
< sizeof(*ei
))
5852 if (!found_extent
) {
5854 ret
= remove_extent_backref(trans
, extent_root
, path
,
5856 is_data
, &last_ref
);
5858 btrfs_abort_transaction(trans
, extent_root
, ret
);
5861 btrfs_release_path(path
);
5862 path
->leave_spinning
= 1;
5864 key
.objectid
= bytenr
;
5865 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
5866 key
.offset
= num_bytes
;
5868 if (!is_data
&& skinny_metadata
) {
5869 key
.type
= BTRFS_METADATA_ITEM_KEY
;
5870 key
.offset
= owner_objectid
;
5873 ret
= btrfs_search_slot(trans
, extent_root
,
5875 if (ret
> 0 && skinny_metadata
&& path
->slots
[0]) {
5877 * Couldn't find our skinny metadata item,
5878 * see if we have ye olde extent item.
5881 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
5883 if (key
.objectid
== bytenr
&&
5884 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
5885 key
.offset
== num_bytes
)
5889 if (ret
> 0 && skinny_metadata
) {
5890 skinny_metadata
= false;
5891 key
.objectid
= bytenr
;
5892 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
5893 key
.offset
= num_bytes
;
5894 btrfs_release_path(path
);
5895 ret
= btrfs_search_slot(trans
, extent_root
,
5900 btrfs_err(info
, "umm, got %d back from search, was looking for %llu",
5903 btrfs_print_leaf(extent_root
,
5907 btrfs_abort_transaction(trans
, extent_root
, ret
);
5910 extent_slot
= path
->slots
[0];
5912 } else if (WARN_ON(ret
== -ENOENT
)) {
5913 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
5915 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
5916 bytenr
, parent
, root_objectid
, owner_objectid
,
5918 btrfs_abort_transaction(trans
, extent_root
, ret
);
5921 btrfs_abort_transaction(trans
, extent_root
, ret
);
5925 leaf
= path
->nodes
[0];
5926 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
5927 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5928 if (item_size
< sizeof(*ei
)) {
5929 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
5930 ret
= convert_extent_item_v0(trans
, extent_root
, path
,
5933 btrfs_abort_transaction(trans
, extent_root
, ret
);
5937 btrfs_release_path(path
);
5938 path
->leave_spinning
= 1;
5940 key
.objectid
= bytenr
;
5941 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
5942 key
.offset
= num_bytes
;
5944 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
5947 btrfs_err(info
, "umm, got %d back from search, was looking for %llu",
5949 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
5952 btrfs_abort_transaction(trans
, extent_root
, ret
);
5956 extent_slot
= path
->slots
[0];
5957 leaf
= path
->nodes
[0];
5958 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
5961 BUG_ON(item_size
< sizeof(*ei
));
5962 ei
= btrfs_item_ptr(leaf
, extent_slot
,
5963 struct btrfs_extent_item
);
5964 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
5965 key
.type
== BTRFS_EXTENT_ITEM_KEY
) {
5966 struct btrfs_tree_block_info
*bi
;
5967 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
5968 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
5969 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
5972 refs
= btrfs_extent_refs(leaf
, ei
);
5973 if (refs
< refs_to_drop
) {
5974 btrfs_err(info
, "trying to drop %d refs but we only have %Lu "
5975 "for bytenr %Lu", refs_to_drop
, refs
, bytenr
);
5977 btrfs_abort_transaction(trans
, extent_root
, ret
);
5980 refs
-= refs_to_drop
;
5983 type
= BTRFS_QGROUP_OPER_SUB_SHARED
;
5985 __run_delayed_extent_op(extent_op
, leaf
, ei
);
5987 * In the case of inline back ref, reference count will
5988 * be updated by remove_extent_backref
5991 BUG_ON(!found_extent
);
5993 btrfs_set_extent_refs(leaf
, ei
, refs
);
5994 btrfs_mark_buffer_dirty(leaf
);
5997 ret
= remove_extent_backref(trans
, extent_root
, path
,
5999 is_data
, &last_ref
);
6001 btrfs_abort_transaction(trans
, extent_root
, ret
);
6005 add_pinned_bytes(root
->fs_info
, -num_bytes
, owner_objectid
,
6009 BUG_ON(is_data
&& refs_to_drop
!=
6010 extent_data_ref_count(root
, path
, iref
));
6012 BUG_ON(path
->slots
[0] != extent_slot
);
6014 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
6015 path
->slots
[0] = extent_slot
;
6021 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
6024 btrfs_abort_transaction(trans
, extent_root
, ret
);
6027 btrfs_release_path(path
);
6030 ret
= btrfs_del_csums(trans
, root
, bytenr
, num_bytes
);
6032 btrfs_abort_transaction(trans
, extent_root
, ret
);
6037 ret
= update_block_group(trans
, root
, bytenr
, num_bytes
, 0);
6039 btrfs_abort_transaction(trans
, extent_root
, ret
);
6043 btrfs_release_path(path
);
6045 /* Deal with the quota accounting */
6046 if (!ret
&& last_ref
&& !no_quota
) {
6049 if (owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
&&
6050 type
== BTRFS_QGROUP_OPER_SUB_SHARED
)
6053 ret
= btrfs_qgroup_record_ref(trans
, info
, root_objectid
,
6054 bytenr
, num_bytes
, type
,
6058 btrfs_free_path(path
);
6063 * when we free an block, it is possible (and likely) that we free the last
6064 * delayed ref for that extent as well. This searches the delayed ref tree for
6065 * a given extent, and if there are no other delayed refs to be processed, it
6066 * removes it from the tree.
6068 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
6069 struct btrfs_root
*root
, u64 bytenr
)
6071 struct btrfs_delayed_ref_head
*head
;
6072 struct btrfs_delayed_ref_root
*delayed_refs
;
6075 delayed_refs
= &trans
->transaction
->delayed_refs
;
6076 spin_lock(&delayed_refs
->lock
);
6077 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
6079 goto out_delayed_unlock
;
6081 spin_lock(&head
->lock
);
6082 if (rb_first(&head
->ref_root
))
6085 if (head
->extent_op
) {
6086 if (!head
->must_insert_reserved
)
6088 btrfs_free_delayed_extent_op(head
->extent_op
);
6089 head
->extent_op
= NULL
;
6093 * waiting for the lock here would deadlock. If someone else has it
6094 * locked they are already in the process of dropping it anyway
6096 if (!mutex_trylock(&head
->mutex
))
6100 * at this point we have a head with no other entries. Go
6101 * ahead and process it.
6103 head
->node
.in_tree
= 0;
6104 rb_erase(&head
->href_node
, &delayed_refs
->href_root
);
6106 atomic_dec(&delayed_refs
->num_entries
);
6109 * we don't take a ref on the node because we're removing it from the
6110 * tree, so we just steal the ref the tree was holding.
6112 delayed_refs
->num_heads
--;
6113 if (head
->processing
== 0)
6114 delayed_refs
->num_heads_ready
--;
6115 head
->processing
= 0;
6116 spin_unlock(&head
->lock
);
6117 spin_unlock(&delayed_refs
->lock
);
6119 BUG_ON(head
->extent_op
);
6120 if (head
->must_insert_reserved
)
6123 mutex_unlock(&head
->mutex
);
6124 btrfs_put_delayed_ref(&head
->node
);
6127 spin_unlock(&head
->lock
);
6130 spin_unlock(&delayed_refs
->lock
);
6134 void btrfs_free_tree_block(struct btrfs_trans_handle
*trans
,
6135 struct btrfs_root
*root
,
6136 struct extent_buffer
*buf
,
6137 u64 parent
, int last_ref
)
6139 struct btrfs_block_group_cache
*cache
= NULL
;
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 cache
= btrfs_lookup_block_group(root
->fs_info
, buf
->start
);
6157 if (btrfs_header_generation(buf
) == trans
->transid
) {
6158 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
6159 ret
= check_ref_cleanup(trans
, root
, buf
->start
);
6164 if (btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
6165 pin_down_extent(root
, cache
, buf
->start
, buf
->len
, 1);
6169 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
));
6171 btrfs_add_free_space(cache
, buf
->start
, buf
->len
);
6172 btrfs_update_reserved_bytes(cache
, buf
->len
, RESERVE_FREE
, 0);
6173 trace_btrfs_reserved_extent_free(root
, buf
->start
, buf
->len
);
6178 add_pinned_bytes(root
->fs_info
, buf
->len
,
6179 btrfs_header_level(buf
),
6180 root
->root_key
.objectid
);
6183 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6186 clear_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
);
6187 btrfs_put_block_group(cache
);
6190 /* Can return -ENOMEM */
6191 int btrfs_free_extent(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
6192 u64 bytenr
, u64 num_bytes
, u64 parent
, u64 root_objectid
,
6193 u64 owner
, u64 offset
, int no_quota
)
6196 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6198 if (btrfs_test_is_dummy_root(root
))
6201 add_pinned_bytes(root
->fs_info
, num_bytes
, owner
, root_objectid
);
6204 * tree log blocks never actually go into the extent allocation
6205 * tree, just update pinning info and exit early.
6207 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
6208 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
6209 /* unlocks the pinned mutex */
6210 btrfs_pin_extent(root
, bytenr
, num_bytes
, 1);
6212 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
6213 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
6215 parent
, root_objectid
, (int)owner
,
6216 BTRFS_DROP_DELAYED_REF
, NULL
, no_quota
);
6218 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
6220 parent
, root_objectid
, owner
,
6221 offset
, BTRFS_DROP_DELAYED_REF
,
6228 * when we wait for progress in the block group caching, its because
6229 * our allocation attempt failed at least once. So, we must sleep
6230 * and let some progress happen before we try again.
6232 * This function will sleep at least once waiting for new free space to
6233 * show up, and then it will check the block group free space numbers
6234 * for our min num_bytes. Another option is to have it go ahead
6235 * and look in the rbtree for a free extent of a given size, but this
6238 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6239 * any of the information in this block group.
6241 static noinline
void
6242 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
6245 struct btrfs_caching_control
*caching_ctl
;
6247 caching_ctl
= get_caching_control(cache
);
6251 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
) ||
6252 (cache
->free_space_ctl
->free_space
>= num_bytes
));
6254 put_caching_control(caching_ctl
);
6258 wait_block_group_cache_done(struct btrfs_block_group_cache
*cache
)
6260 struct btrfs_caching_control
*caching_ctl
;
6263 caching_ctl
= get_caching_control(cache
);
6265 return (cache
->cached
== BTRFS_CACHE_ERROR
) ? -EIO
: 0;
6267 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
));
6268 if (cache
->cached
== BTRFS_CACHE_ERROR
)
6270 put_caching_control(caching_ctl
);
6274 int __get_raid_index(u64 flags
)
6276 if (flags
& BTRFS_BLOCK_GROUP_RAID10
)
6277 return BTRFS_RAID_RAID10
;
6278 else if (flags
& BTRFS_BLOCK_GROUP_RAID1
)
6279 return BTRFS_RAID_RAID1
;
6280 else if (flags
& BTRFS_BLOCK_GROUP_DUP
)
6281 return BTRFS_RAID_DUP
;
6282 else if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
6283 return BTRFS_RAID_RAID0
;
6284 else if (flags
& BTRFS_BLOCK_GROUP_RAID5
)
6285 return BTRFS_RAID_RAID5
;
6286 else if (flags
& BTRFS_BLOCK_GROUP_RAID6
)
6287 return BTRFS_RAID_RAID6
;
6289 return BTRFS_RAID_SINGLE
; /* BTRFS_BLOCK_GROUP_SINGLE */
6292 int get_block_group_index(struct btrfs_block_group_cache
*cache
)
6294 return __get_raid_index(cache
->flags
);
6297 static const char *btrfs_raid_type_names
[BTRFS_NR_RAID_TYPES
] = {
6298 [BTRFS_RAID_RAID10
] = "raid10",
6299 [BTRFS_RAID_RAID1
] = "raid1",
6300 [BTRFS_RAID_DUP
] = "dup",
6301 [BTRFS_RAID_RAID0
] = "raid0",
6302 [BTRFS_RAID_SINGLE
] = "single",
6303 [BTRFS_RAID_RAID5
] = "raid5",
6304 [BTRFS_RAID_RAID6
] = "raid6",
6307 static const char *get_raid_name(enum btrfs_raid_types type
)
6309 if (type
>= BTRFS_NR_RAID_TYPES
)
6312 return btrfs_raid_type_names
[type
];
6315 enum btrfs_loop_type
{
6316 LOOP_CACHING_NOWAIT
= 0,
6317 LOOP_CACHING_WAIT
= 1,
6318 LOOP_ALLOC_CHUNK
= 2,
6319 LOOP_NO_EMPTY_SIZE
= 3,
6323 btrfs_lock_block_group(struct btrfs_block_group_cache
*cache
,
6327 down_read(&cache
->data_rwsem
);
6331 btrfs_grab_block_group(struct btrfs_block_group_cache
*cache
,
6334 btrfs_get_block_group(cache
);
6336 down_read(&cache
->data_rwsem
);
6339 static struct btrfs_block_group_cache
*
6340 btrfs_lock_cluster(struct btrfs_block_group_cache
*block_group
,
6341 struct btrfs_free_cluster
*cluster
,
6344 struct btrfs_block_group_cache
*used_bg
;
6345 bool locked
= false;
6347 spin_lock(&cluster
->refill_lock
);
6349 if (used_bg
== cluster
->block_group
)
6352 up_read(&used_bg
->data_rwsem
);
6353 btrfs_put_block_group(used_bg
);
6356 used_bg
= cluster
->block_group
;
6360 if (used_bg
== block_group
)
6363 btrfs_get_block_group(used_bg
);
6368 if (down_read_trylock(&used_bg
->data_rwsem
))
6371 spin_unlock(&cluster
->refill_lock
);
6372 down_read(&used_bg
->data_rwsem
);
6378 btrfs_release_block_group(struct btrfs_block_group_cache
*cache
,
6382 up_read(&cache
->data_rwsem
);
6383 btrfs_put_block_group(cache
);
6387 * walks the btree of allocated extents and find a hole of a given size.
6388 * The key ins is changed to record the hole:
6389 * ins->objectid == start position
6390 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6391 * ins->offset == the size of the hole.
6392 * Any available blocks before search_start are skipped.
6394 * If there is no suitable free space, we will record the max size of
6395 * the free space extent currently.
6397 static noinline
int find_free_extent(struct btrfs_root
*orig_root
,
6398 u64 num_bytes
, u64 empty_size
,
6399 u64 hint_byte
, struct btrfs_key
*ins
,
6400 u64 flags
, int delalloc
)
6403 struct btrfs_root
*root
= orig_root
->fs_info
->extent_root
;
6404 struct btrfs_free_cluster
*last_ptr
= NULL
;
6405 struct btrfs_block_group_cache
*block_group
= NULL
;
6406 u64 search_start
= 0;
6407 u64 max_extent_size
= 0;
6408 int empty_cluster
= 2 * 1024 * 1024;
6409 struct btrfs_space_info
*space_info
;
6411 int index
= __get_raid_index(flags
);
6412 int alloc_type
= (flags
& BTRFS_BLOCK_GROUP_DATA
) ?
6413 RESERVE_ALLOC_NO_ACCOUNT
: RESERVE_ALLOC
;
6414 bool failed_cluster_refill
= false;
6415 bool failed_alloc
= false;
6416 bool use_cluster
= true;
6417 bool have_caching_bg
= false;
6419 WARN_ON(num_bytes
< root
->sectorsize
);
6420 ins
->type
= BTRFS_EXTENT_ITEM_KEY
;
6424 trace_find_free_extent(orig_root
, num_bytes
, empty_size
, flags
);
6426 space_info
= __find_space_info(root
->fs_info
, flags
);
6428 btrfs_err(root
->fs_info
, "No space info for %llu", flags
);
6433 * If the space info is for both data and metadata it means we have a
6434 * small filesystem and we can't use the clustering stuff.
6436 if (btrfs_mixed_space_info(space_info
))
6437 use_cluster
= false;
6439 if (flags
& BTRFS_BLOCK_GROUP_METADATA
&& use_cluster
) {
6440 last_ptr
= &root
->fs_info
->meta_alloc_cluster
;
6441 if (!btrfs_test_opt(root
, SSD
))
6442 empty_cluster
= 64 * 1024;
6445 if ((flags
& BTRFS_BLOCK_GROUP_DATA
) && use_cluster
&&
6446 btrfs_test_opt(root
, SSD
)) {
6447 last_ptr
= &root
->fs_info
->data_alloc_cluster
;
6451 spin_lock(&last_ptr
->lock
);
6452 if (last_ptr
->block_group
)
6453 hint_byte
= last_ptr
->window_start
;
6454 spin_unlock(&last_ptr
->lock
);
6457 search_start
= max(search_start
, first_logical_byte(root
, 0));
6458 search_start
= max(search_start
, hint_byte
);
6463 if (search_start
== hint_byte
) {
6464 block_group
= btrfs_lookup_block_group(root
->fs_info
,
6467 * we don't want to use the block group if it doesn't match our
6468 * allocation bits, or if its not cached.
6470 * However if we are re-searching with an ideal block group
6471 * picked out then we don't care that the block group is cached.
6473 if (block_group
&& block_group_bits(block_group
, flags
) &&
6474 block_group
->cached
!= BTRFS_CACHE_NO
) {
6475 down_read(&space_info
->groups_sem
);
6476 if (list_empty(&block_group
->list
) ||
6479 * someone is removing this block group,
6480 * we can't jump into the have_block_group
6481 * target because our list pointers are not
6484 btrfs_put_block_group(block_group
);
6485 up_read(&space_info
->groups_sem
);
6487 index
= get_block_group_index(block_group
);
6488 btrfs_lock_block_group(block_group
, delalloc
);
6489 goto have_block_group
;
6491 } else if (block_group
) {
6492 btrfs_put_block_group(block_group
);
6496 have_caching_bg
= false;
6497 down_read(&space_info
->groups_sem
);
6498 list_for_each_entry(block_group
, &space_info
->block_groups
[index
],
6503 btrfs_grab_block_group(block_group
, delalloc
);
6504 search_start
= block_group
->key
.objectid
;
6507 * this can happen if we end up cycling through all the
6508 * raid types, but we want to make sure we only allocate
6509 * for the proper type.
6511 if (!block_group_bits(block_group
, flags
)) {
6512 u64 extra
= BTRFS_BLOCK_GROUP_DUP
|
6513 BTRFS_BLOCK_GROUP_RAID1
|
6514 BTRFS_BLOCK_GROUP_RAID5
|
6515 BTRFS_BLOCK_GROUP_RAID6
|
6516 BTRFS_BLOCK_GROUP_RAID10
;
6519 * if they asked for extra copies and this block group
6520 * doesn't provide them, bail. This does allow us to
6521 * fill raid0 from raid1.
6523 if ((flags
& extra
) && !(block_group
->flags
& extra
))
6528 cached
= block_group_cache_done(block_group
);
6529 if (unlikely(!cached
)) {
6530 ret
= cache_block_group(block_group
, 0);
6535 if (unlikely(block_group
->cached
== BTRFS_CACHE_ERROR
))
6537 if (unlikely(block_group
->ro
))
6541 * Ok we want to try and use the cluster allocator, so
6545 struct btrfs_block_group_cache
*used_block_group
;
6546 unsigned long aligned_cluster
;
6548 * the refill lock keeps out other
6549 * people trying to start a new cluster
6551 used_block_group
= btrfs_lock_cluster(block_group
,
6554 if (!used_block_group
)
6555 goto refill_cluster
;
6557 if (used_block_group
!= block_group
&&
6558 (used_block_group
->ro
||
6559 !block_group_bits(used_block_group
, flags
)))
6560 goto release_cluster
;
6562 offset
= btrfs_alloc_from_cluster(used_block_group
,
6565 used_block_group
->key
.objectid
,
6568 /* we have a block, we're done */
6569 spin_unlock(&last_ptr
->refill_lock
);
6570 trace_btrfs_reserve_extent_cluster(root
,
6572 search_start
, num_bytes
);
6573 if (used_block_group
!= block_group
) {
6574 btrfs_release_block_group(block_group
,
6576 block_group
= used_block_group
;
6581 WARN_ON(last_ptr
->block_group
!= used_block_group
);
6583 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6584 * set up a new clusters, so lets just skip it
6585 * and let the allocator find whatever block
6586 * it can find. If we reach this point, we
6587 * will have tried the cluster allocator
6588 * plenty of times and not have found
6589 * anything, so we are likely way too
6590 * fragmented for the clustering stuff to find
6593 * However, if the cluster is taken from the
6594 * current block group, release the cluster
6595 * first, so that we stand a better chance of
6596 * succeeding in the unclustered
6598 if (loop
>= LOOP_NO_EMPTY_SIZE
&&
6599 used_block_group
!= block_group
) {
6600 spin_unlock(&last_ptr
->refill_lock
);
6601 btrfs_release_block_group(used_block_group
,
6603 goto unclustered_alloc
;
6607 * this cluster didn't work out, free it and
6610 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
6612 if (used_block_group
!= block_group
)
6613 btrfs_release_block_group(used_block_group
,
6616 if (loop
>= LOOP_NO_EMPTY_SIZE
) {
6617 spin_unlock(&last_ptr
->refill_lock
);
6618 goto unclustered_alloc
;
6621 aligned_cluster
= max_t(unsigned long,
6622 empty_cluster
+ empty_size
,
6623 block_group
->full_stripe_len
);
6625 /* allocate a cluster in this block group */
6626 ret
= btrfs_find_space_cluster(root
, block_group
,
6627 last_ptr
, search_start
,
6632 * now pull our allocation out of this
6635 offset
= btrfs_alloc_from_cluster(block_group
,
6641 /* we found one, proceed */
6642 spin_unlock(&last_ptr
->refill_lock
);
6643 trace_btrfs_reserve_extent_cluster(root
,
6644 block_group
, search_start
,
6648 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
6649 && !failed_cluster_refill
) {
6650 spin_unlock(&last_ptr
->refill_lock
);
6652 failed_cluster_refill
= true;
6653 wait_block_group_cache_progress(block_group
,
6654 num_bytes
+ empty_cluster
+ empty_size
);
6655 goto have_block_group
;
6659 * at this point we either didn't find a cluster
6660 * or we weren't able to allocate a block from our
6661 * cluster. Free the cluster we've been trying
6662 * to use, and go to the next block group
6664 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
6665 spin_unlock(&last_ptr
->refill_lock
);
6670 spin_lock(&block_group
->free_space_ctl
->tree_lock
);
6672 block_group
->free_space_ctl
->free_space
<
6673 num_bytes
+ empty_cluster
+ empty_size
) {
6674 if (block_group
->free_space_ctl
->free_space
>
6677 block_group
->free_space_ctl
->free_space
;
6678 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
6681 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
6683 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
6684 num_bytes
, empty_size
,
6687 * If we didn't find a chunk, and we haven't failed on this
6688 * block group before, and this block group is in the middle of
6689 * caching and we are ok with waiting, then go ahead and wait
6690 * for progress to be made, and set failed_alloc to true.
6692 * If failed_alloc is true then we've already waited on this
6693 * block group once and should move on to the next block group.
6695 if (!offset
&& !failed_alloc
&& !cached
&&
6696 loop
> LOOP_CACHING_NOWAIT
) {
6697 wait_block_group_cache_progress(block_group
,
6698 num_bytes
+ empty_size
);
6699 failed_alloc
= true;
6700 goto have_block_group
;
6701 } else if (!offset
) {
6703 have_caching_bg
= true;
6707 search_start
= ALIGN(offset
, root
->stripesize
);
6709 /* move on to the next group */
6710 if (search_start
+ num_bytes
>
6711 block_group
->key
.objectid
+ block_group
->key
.offset
) {
6712 btrfs_add_free_space(block_group
, offset
, num_bytes
);
6716 if (offset
< search_start
)
6717 btrfs_add_free_space(block_group
, offset
,
6718 search_start
- offset
);
6719 BUG_ON(offset
> search_start
);
6721 ret
= btrfs_update_reserved_bytes(block_group
, num_bytes
,
6722 alloc_type
, delalloc
);
6723 if (ret
== -EAGAIN
) {
6724 btrfs_add_free_space(block_group
, offset
, num_bytes
);
6728 /* we are all good, lets return */
6729 ins
->objectid
= search_start
;
6730 ins
->offset
= num_bytes
;
6732 trace_btrfs_reserve_extent(orig_root
, block_group
,
6733 search_start
, num_bytes
);
6734 btrfs_release_block_group(block_group
, delalloc
);
6737 failed_cluster_refill
= false;
6738 failed_alloc
= false;
6739 BUG_ON(index
!= get_block_group_index(block_group
));
6740 btrfs_release_block_group(block_group
, delalloc
);
6742 up_read(&space_info
->groups_sem
);
6744 if (!ins
->objectid
&& loop
>= LOOP_CACHING_WAIT
&& have_caching_bg
)
6747 if (!ins
->objectid
&& ++index
< BTRFS_NR_RAID_TYPES
)
6751 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
6752 * caching kthreads as we move along
6753 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
6754 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
6755 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
6758 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
) {
6761 if (loop
== LOOP_ALLOC_CHUNK
) {
6762 struct btrfs_trans_handle
*trans
;
6765 trans
= current
->journal_info
;
6769 trans
= btrfs_join_transaction(root
);
6771 if (IS_ERR(trans
)) {
6772 ret
= PTR_ERR(trans
);
6776 ret
= do_chunk_alloc(trans
, root
, flags
,
6779 * Do not bail out on ENOSPC since we
6780 * can do more things.
6782 if (ret
< 0 && ret
!= -ENOSPC
)
6783 btrfs_abort_transaction(trans
,
6788 btrfs_end_transaction(trans
, root
);
6793 if (loop
== LOOP_NO_EMPTY_SIZE
) {
6799 } else if (!ins
->objectid
) {
6801 } else if (ins
->objectid
) {
6806 ins
->offset
= max_extent_size
;
6810 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
6811 int dump_block_groups
)
6813 struct btrfs_block_group_cache
*cache
;
6816 spin_lock(&info
->lock
);
6817 printk(KERN_INFO
"BTRFS: space_info %llu has %llu free, is %sfull\n",
6819 info
->total_bytes
- info
->bytes_used
- info
->bytes_pinned
-
6820 info
->bytes_reserved
- info
->bytes_readonly
,
6821 (info
->full
) ? "" : "not ");
6822 printk(KERN_INFO
"BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
6823 "reserved=%llu, may_use=%llu, readonly=%llu\n",
6824 info
->total_bytes
, info
->bytes_used
, info
->bytes_pinned
,
6825 info
->bytes_reserved
, info
->bytes_may_use
,
6826 info
->bytes_readonly
);
6827 spin_unlock(&info
->lock
);
6829 if (!dump_block_groups
)
6832 down_read(&info
->groups_sem
);
6834 list_for_each_entry(cache
, &info
->block_groups
[index
], list
) {
6835 spin_lock(&cache
->lock
);
6836 printk(KERN_INFO
"BTRFS: "
6837 "block group %llu has %llu bytes, "
6838 "%llu used %llu pinned %llu reserved %s\n",
6839 cache
->key
.objectid
, cache
->key
.offset
,
6840 btrfs_block_group_used(&cache
->item
), cache
->pinned
,
6841 cache
->reserved
, cache
->ro
? "[readonly]" : "");
6842 btrfs_dump_free_space(cache
, bytes
);
6843 spin_unlock(&cache
->lock
);
6845 if (++index
< BTRFS_NR_RAID_TYPES
)
6847 up_read(&info
->groups_sem
);
6850 int btrfs_reserve_extent(struct btrfs_root
*root
,
6851 u64 num_bytes
, u64 min_alloc_size
,
6852 u64 empty_size
, u64 hint_byte
,
6853 struct btrfs_key
*ins
, int is_data
, int delalloc
)
6855 bool final_tried
= false;
6859 flags
= btrfs_get_alloc_profile(root
, is_data
);
6861 WARN_ON(num_bytes
< root
->sectorsize
);
6862 ret
= find_free_extent(root
, num_bytes
, empty_size
, hint_byte
, ins
,
6865 if (ret
== -ENOSPC
) {
6866 if (!final_tried
&& ins
->offset
) {
6867 num_bytes
= min(num_bytes
>> 1, ins
->offset
);
6868 num_bytes
= round_down(num_bytes
, root
->sectorsize
);
6869 num_bytes
= max(num_bytes
, min_alloc_size
);
6870 if (num_bytes
== min_alloc_size
)
6873 } else if (btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
6874 struct btrfs_space_info
*sinfo
;
6876 sinfo
= __find_space_info(root
->fs_info
, flags
);
6877 btrfs_err(root
->fs_info
, "allocation failed flags %llu, wanted %llu",
6880 dump_space_info(sinfo
, num_bytes
, 1);
6887 static int __btrfs_free_reserved_extent(struct btrfs_root
*root
,
6889 int pin
, int delalloc
)
6891 struct btrfs_block_group_cache
*cache
;
6894 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
6896 btrfs_err(root
->fs_info
, "Unable to find block group for %llu",
6901 if (btrfs_test_opt(root
, DISCARD
))
6902 ret
= btrfs_discard_extent(root
, start
, len
, NULL
);
6905 pin_down_extent(root
, cache
, start
, len
, 1);
6907 btrfs_add_free_space(cache
, start
, len
);
6908 btrfs_update_reserved_bytes(cache
, len
, RESERVE_FREE
, delalloc
);
6910 btrfs_put_block_group(cache
);
6912 trace_btrfs_reserved_extent_free(root
, start
, len
);
6917 int btrfs_free_reserved_extent(struct btrfs_root
*root
,
6918 u64 start
, u64 len
, int delalloc
)
6920 return __btrfs_free_reserved_extent(root
, start
, len
, 0, delalloc
);
6923 int btrfs_free_and_pin_reserved_extent(struct btrfs_root
*root
,
6926 return __btrfs_free_reserved_extent(root
, start
, len
, 1, 0);
6929 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
6930 struct btrfs_root
*root
,
6931 u64 parent
, u64 root_objectid
,
6932 u64 flags
, u64 owner
, u64 offset
,
6933 struct btrfs_key
*ins
, int ref_mod
)
6936 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6937 struct btrfs_extent_item
*extent_item
;
6938 struct btrfs_extent_inline_ref
*iref
;
6939 struct btrfs_path
*path
;
6940 struct extent_buffer
*leaf
;
6945 type
= BTRFS_SHARED_DATA_REF_KEY
;
6947 type
= BTRFS_EXTENT_DATA_REF_KEY
;
6949 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
6951 path
= btrfs_alloc_path();
6955 path
->leave_spinning
= 1;
6956 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
6959 btrfs_free_path(path
);
6963 leaf
= path
->nodes
[0];
6964 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
6965 struct btrfs_extent_item
);
6966 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
6967 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
6968 btrfs_set_extent_flags(leaf
, extent_item
,
6969 flags
| BTRFS_EXTENT_FLAG_DATA
);
6971 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
6972 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
6974 struct btrfs_shared_data_ref
*ref
;
6975 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
6976 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
6977 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
6979 struct btrfs_extent_data_ref
*ref
;
6980 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
6981 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
6982 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
6983 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
6984 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
6987 btrfs_mark_buffer_dirty(path
->nodes
[0]);
6988 btrfs_free_path(path
);
6990 /* Always set parent to 0 here since its exclusive anyway. */
6991 ret
= btrfs_qgroup_record_ref(trans
, fs_info
, root_objectid
,
6992 ins
->objectid
, ins
->offset
,
6993 BTRFS_QGROUP_OPER_ADD_EXCL
, 0);
6997 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1);
6998 if (ret
) { /* -ENOENT, logic error */
6999 btrfs_err(fs_info
, "update block group failed for %llu %llu",
7000 ins
->objectid
, ins
->offset
);
7003 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, ins
->offset
);
7007 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
7008 struct btrfs_root
*root
,
7009 u64 parent
, u64 root_objectid
,
7010 u64 flags
, struct btrfs_disk_key
*key
,
7011 int level
, struct btrfs_key
*ins
,
7015 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7016 struct btrfs_extent_item
*extent_item
;
7017 struct btrfs_tree_block_info
*block_info
;
7018 struct btrfs_extent_inline_ref
*iref
;
7019 struct btrfs_path
*path
;
7020 struct extent_buffer
*leaf
;
7021 u32 size
= sizeof(*extent_item
) + sizeof(*iref
);
7022 u64 num_bytes
= ins
->offset
;
7023 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
7026 if (!skinny_metadata
)
7027 size
+= sizeof(*block_info
);
7029 path
= btrfs_alloc_path();
7031 btrfs_free_and_pin_reserved_extent(root
, ins
->objectid
,
7036 path
->leave_spinning
= 1;
7037 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
7040 btrfs_free_and_pin_reserved_extent(root
, ins
->objectid
,
7042 btrfs_free_path(path
);
7046 leaf
= path
->nodes
[0];
7047 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
7048 struct btrfs_extent_item
);
7049 btrfs_set_extent_refs(leaf
, extent_item
, 1);
7050 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
7051 btrfs_set_extent_flags(leaf
, extent_item
,
7052 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
7054 if (skinny_metadata
) {
7055 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
7056 num_bytes
= root
->nodesize
;
7058 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
7059 btrfs_set_tree_block_key(leaf
, block_info
, key
);
7060 btrfs_set_tree_block_level(leaf
, block_info
, level
);
7061 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
7065 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
7066 btrfs_set_extent_inline_ref_type(leaf
, iref
,
7067 BTRFS_SHARED_BLOCK_REF_KEY
);
7068 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
7070 btrfs_set_extent_inline_ref_type(leaf
, iref
,
7071 BTRFS_TREE_BLOCK_REF_KEY
);
7072 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
7075 btrfs_mark_buffer_dirty(leaf
);
7076 btrfs_free_path(path
);
7079 ret
= btrfs_qgroup_record_ref(trans
, fs_info
, root_objectid
,
7080 ins
->objectid
, num_bytes
,
7081 BTRFS_QGROUP_OPER_ADD_EXCL
, 0);
7086 ret
= update_block_group(trans
, root
, ins
->objectid
, root
->nodesize
,
7088 if (ret
) { /* -ENOENT, logic error */
7089 btrfs_err(fs_info
, "update block group failed for %llu %llu",
7090 ins
->objectid
, ins
->offset
);
7094 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, root
->nodesize
);
7098 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
7099 struct btrfs_root
*root
,
7100 u64 root_objectid
, u64 owner
,
7101 u64 offset
, struct btrfs_key
*ins
)
7105 BUG_ON(root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
7107 ret
= btrfs_add_delayed_data_ref(root
->fs_info
, trans
, ins
->objectid
,
7109 root_objectid
, owner
, offset
,
7110 BTRFS_ADD_DELAYED_EXTENT
, NULL
, 0);
7115 * this is used by the tree logging recovery code. It records that
7116 * an extent has been allocated and makes sure to clear the free
7117 * space cache bits as well
7119 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
7120 struct btrfs_root
*root
,
7121 u64 root_objectid
, u64 owner
, u64 offset
,
7122 struct btrfs_key
*ins
)
7125 struct btrfs_block_group_cache
*block_group
;
7128 * Mixed block groups will exclude before processing the log so we only
7129 * need to do the exlude dance if this fs isn't mixed.
7131 if (!btrfs_fs_incompat(root
->fs_info
, MIXED_GROUPS
)) {
7132 ret
= __exclude_logged_extent(root
, ins
->objectid
, ins
->offset
);
7137 block_group
= btrfs_lookup_block_group(root
->fs_info
, ins
->objectid
);
7141 ret
= btrfs_update_reserved_bytes(block_group
, ins
->offset
,
7142 RESERVE_ALLOC_NO_ACCOUNT
, 0);
7143 BUG_ON(ret
); /* logic error */
7144 ret
= alloc_reserved_file_extent(trans
, root
, 0, root_objectid
,
7145 0, owner
, offset
, ins
, 1);
7146 btrfs_put_block_group(block_group
);
7150 static struct extent_buffer
*
7151 btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
7152 u64 bytenr
, int level
)
7154 struct extent_buffer
*buf
;
7156 buf
= btrfs_find_create_tree_block(root
, bytenr
);
7158 return ERR_PTR(-ENOMEM
);
7159 btrfs_set_header_generation(buf
, trans
->transid
);
7160 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, buf
, level
);
7161 btrfs_tree_lock(buf
);
7162 clean_tree_block(trans
, root
, buf
);
7163 clear_bit(EXTENT_BUFFER_STALE
, &buf
->bflags
);
7165 btrfs_set_lock_blocking(buf
);
7166 btrfs_set_buffer_uptodate(buf
);
7168 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
7169 buf
->log_index
= root
->log_transid
% 2;
7171 * we allow two log transactions at a time, use different
7172 * EXENT bit to differentiate dirty pages.
7174 if (buf
->log_index
== 0)
7175 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
7176 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
7178 set_extent_new(&root
->dirty_log_pages
, buf
->start
,
7179 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
7181 buf
->log_index
= -1;
7182 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
7183 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
7185 trans
->blocks_used
++;
7186 /* this returns a buffer locked for blocking */
7190 static struct btrfs_block_rsv
*
7191 use_block_rsv(struct btrfs_trans_handle
*trans
,
7192 struct btrfs_root
*root
, u32 blocksize
)
7194 struct btrfs_block_rsv
*block_rsv
;
7195 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
7197 bool global_updated
= false;
7199 block_rsv
= get_block_rsv(trans
, root
);
7201 if (unlikely(block_rsv
->size
== 0))
7204 ret
= block_rsv_use_bytes(block_rsv
, blocksize
);
7208 if (block_rsv
->failfast
)
7209 return ERR_PTR(ret
);
7211 if (block_rsv
->type
== BTRFS_BLOCK_RSV_GLOBAL
&& !global_updated
) {
7212 global_updated
= true;
7213 update_global_block_rsv(root
->fs_info
);
7217 if (btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
7218 static DEFINE_RATELIMIT_STATE(_rs
,
7219 DEFAULT_RATELIMIT_INTERVAL
* 10,
7220 /*DEFAULT_RATELIMIT_BURST*/ 1);
7221 if (__ratelimit(&_rs
))
7223 "BTRFS: block rsv returned %d\n", ret
);
7226 ret
= reserve_metadata_bytes(root
, block_rsv
, blocksize
,
7227 BTRFS_RESERVE_NO_FLUSH
);
7231 * If we couldn't reserve metadata bytes try and use some from
7232 * the global reserve if its space type is the same as the global
7235 if (block_rsv
->type
!= BTRFS_BLOCK_RSV_GLOBAL
&&
7236 block_rsv
->space_info
== global_rsv
->space_info
) {
7237 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
7241 return ERR_PTR(ret
);
7244 static void unuse_block_rsv(struct btrfs_fs_info
*fs_info
,
7245 struct btrfs_block_rsv
*block_rsv
, u32 blocksize
)
7247 block_rsv_add_bytes(block_rsv
, blocksize
, 0);
7248 block_rsv_release_bytes(fs_info
, block_rsv
, NULL
, 0);
7252 * finds a free extent and does all the dirty work required for allocation
7253 * returns the key for the extent through ins, and a tree buffer for
7254 * the first block of the extent through buf.
7256 * returns the tree buffer or NULL.
7258 struct extent_buffer
*btrfs_alloc_tree_block(struct btrfs_trans_handle
*trans
,
7259 struct btrfs_root
*root
,
7260 u64 parent
, u64 root_objectid
,
7261 struct btrfs_disk_key
*key
, int level
,
7262 u64 hint
, u64 empty_size
)
7264 struct btrfs_key ins
;
7265 struct btrfs_block_rsv
*block_rsv
;
7266 struct extent_buffer
*buf
;
7269 u32 blocksize
= root
->nodesize
;
7270 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
7273 if (btrfs_test_is_dummy_root(root
)) {
7274 buf
= btrfs_init_new_buffer(trans
, root
, root
->alloc_bytenr
,
7277 root
->alloc_bytenr
+= blocksize
;
7281 block_rsv
= use_block_rsv(trans
, root
, blocksize
);
7282 if (IS_ERR(block_rsv
))
7283 return ERR_CAST(block_rsv
);
7285 ret
= btrfs_reserve_extent(root
, blocksize
, blocksize
,
7286 empty_size
, hint
, &ins
, 0, 0);
7288 unuse_block_rsv(root
->fs_info
, block_rsv
, blocksize
);
7289 return ERR_PTR(ret
);
7292 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
, level
);
7293 BUG_ON(IS_ERR(buf
)); /* -ENOMEM */
7295 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
7297 parent
= ins
.objectid
;
7298 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
7302 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
7303 struct btrfs_delayed_extent_op
*extent_op
;
7304 extent_op
= btrfs_alloc_delayed_extent_op();
7305 BUG_ON(!extent_op
); /* -ENOMEM */
7307 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
7309 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
7310 extent_op
->flags_to_set
= flags
;
7311 if (skinny_metadata
)
7312 extent_op
->update_key
= 0;
7314 extent_op
->update_key
= 1;
7315 extent_op
->update_flags
= 1;
7316 extent_op
->is_data
= 0;
7317 extent_op
->level
= level
;
7319 ret
= btrfs_add_delayed_tree_ref(root
->fs_info
, trans
,
7321 ins
.offset
, parent
, root_objectid
,
7322 level
, BTRFS_ADD_DELAYED_EXTENT
,
7324 BUG_ON(ret
); /* -ENOMEM */
7329 struct walk_control
{
7330 u64 refs
[BTRFS_MAX_LEVEL
];
7331 u64 flags
[BTRFS_MAX_LEVEL
];
7332 struct btrfs_key update_progress
;
7343 #define DROP_REFERENCE 1
7344 #define UPDATE_BACKREF 2
7346 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
7347 struct btrfs_root
*root
,
7348 struct walk_control
*wc
,
7349 struct btrfs_path
*path
)
7357 struct btrfs_key key
;
7358 struct extent_buffer
*eb
;
7363 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
7364 wc
->reada_count
= wc
->reada_count
* 2 / 3;
7365 wc
->reada_count
= max(wc
->reada_count
, 2);
7367 wc
->reada_count
= wc
->reada_count
* 3 / 2;
7368 wc
->reada_count
= min_t(int, wc
->reada_count
,
7369 BTRFS_NODEPTRS_PER_BLOCK(root
));
7372 eb
= path
->nodes
[wc
->level
];
7373 nritems
= btrfs_header_nritems(eb
);
7374 blocksize
= root
->nodesize
;
7376 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
7377 if (nread
>= wc
->reada_count
)
7381 bytenr
= btrfs_node_blockptr(eb
, slot
);
7382 generation
= btrfs_node_ptr_generation(eb
, slot
);
7384 if (slot
== path
->slots
[wc
->level
])
7387 if (wc
->stage
== UPDATE_BACKREF
&&
7388 generation
<= root
->root_key
.offset
)
7391 /* We don't lock the tree block, it's OK to be racy here */
7392 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
,
7393 wc
->level
- 1, 1, &refs
,
7395 /* We don't care about errors in readahead. */
7400 if (wc
->stage
== DROP_REFERENCE
) {
7404 if (wc
->level
== 1 &&
7405 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
7407 if (!wc
->update_ref
||
7408 generation
<= root
->root_key
.offset
)
7410 btrfs_node_key_to_cpu(eb
, &key
, slot
);
7411 ret
= btrfs_comp_cpu_keys(&key
,
7412 &wc
->update_progress
);
7416 if (wc
->level
== 1 &&
7417 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
7421 readahead_tree_block(root
, bytenr
);
7424 wc
->reada_slot
= slot
;
7427 static int account_leaf_items(struct btrfs_trans_handle
*trans
,
7428 struct btrfs_root
*root
,
7429 struct extent_buffer
*eb
)
7431 int nr
= btrfs_header_nritems(eb
);
7432 int i
, extent_type
, ret
;
7433 struct btrfs_key key
;
7434 struct btrfs_file_extent_item
*fi
;
7435 u64 bytenr
, num_bytes
;
7437 for (i
= 0; i
< nr
; i
++) {
7438 btrfs_item_key_to_cpu(eb
, &key
, i
);
7440 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
7443 fi
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
7444 /* filter out non qgroup-accountable extents */
7445 extent_type
= btrfs_file_extent_type(eb
, fi
);
7447 if (extent_type
== BTRFS_FILE_EXTENT_INLINE
)
7450 bytenr
= btrfs_file_extent_disk_bytenr(eb
, fi
);
7454 num_bytes
= btrfs_file_extent_disk_num_bytes(eb
, fi
);
7456 ret
= btrfs_qgroup_record_ref(trans
, root
->fs_info
,
7459 BTRFS_QGROUP_OPER_SUB_SUBTREE
, 0);
7467 * Walk up the tree from the bottom, freeing leaves and any interior
7468 * nodes which have had all slots visited. If a node (leaf or
7469 * interior) is freed, the node above it will have it's slot
7470 * incremented. The root node will never be freed.
7472 * At the end of this function, we should have a path which has all
7473 * slots incremented to the next position for a search. If we need to
7474 * read a new node it will be NULL and the node above it will have the
7475 * correct slot selected for a later read.
7477 * If we increment the root nodes slot counter past the number of
7478 * elements, 1 is returned to signal completion of the search.
7480 static int adjust_slots_upwards(struct btrfs_root
*root
,
7481 struct btrfs_path
*path
, int root_level
)
7485 struct extent_buffer
*eb
;
7487 if (root_level
== 0)
7490 while (level
<= root_level
) {
7491 eb
= path
->nodes
[level
];
7492 nr
= btrfs_header_nritems(eb
);
7493 path
->slots
[level
]++;
7494 slot
= path
->slots
[level
];
7495 if (slot
>= nr
|| level
== 0) {
7497 * Don't free the root - we will detect this
7498 * condition after our loop and return a
7499 * positive value for caller to stop walking the tree.
7501 if (level
!= root_level
) {
7502 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
7503 path
->locks
[level
] = 0;
7505 free_extent_buffer(eb
);
7506 path
->nodes
[level
] = NULL
;
7507 path
->slots
[level
] = 0;
7511 * We have a valid slot to walk back down
7512 * from. Stop here so caller can process these
7521 eb
= path
->nodes
[root_level
];
7522 if (path
->slots
[root_level
] >= btrfs_header_nritems(eb
))
7529 * root_eb is the subtree root and is locked before this function is called.
7531 static int account_shared_subtree(struct btrfs_trans_handle
*trans
,
7532 struct btrfs_root
*root
,
7533 struct extent_buffer
*root_eb
,
7539 struct extent_buffer
*eb
= root_eb
;
7540 struct btrfs_path
*path
= NULL
;
7542 BUG_ON(root_level
< 0 || root_level
> BTRFS_MAX_LEVEL
);
7543 BUG_ON(root_eb
== NULL
);
7545 if (!root
->fs_info
->quota_enabled
)
7548 if (!extent_buffer_uptodate(root_eb
)) {
7549 ret
= btrfs_read_buffer(root_eb
, root_gen
);
7554 if (root_level
== 0) {
7555 ret
= account_leaf_items(trans
, root
, root_eb
);
7559 path
= btrfs_alloc_path();
7564 * Walk down the tree. Missing extent blocks are filled in as
7565 * we go. Metadata is accounted every time we read a new
7568 * When we reach a leaf, we account for file extent items in it,
7569 * walk back up the tree (adjusting slot pointers as we go)
7570 * and restart the search process.
7572 extent_buffer_get(root_eb
); /* For path */
7573 path
->nodes
[root_level
] = root_eb
;
7574 path
->slots
[root_level
] = 0;
7575 path
->locks
[root_level
] = 0; /* so release_path doesn't try to unlock */
7578 while (level
>= 0) {
7579 if (path
->nodes
[level
] == NULL
) {
7584 /* We need to get child blockptr/gen from
7585 * parent before we can read it. */
7586 eb
= path
->nodes
[level
+ 1];
7587 parent_slot
= path
->slots
[level
+ 1];
7588 child_bytenr
= btrfs_node_blockptr(eb
, parent_slot
);
7589 child_gen
= btrfs_node_ptr_generation(eb
, parent_slot
);
7591 eb
= read_tree_block(root
, child_bytenr
, child_gen
);
7592 if (!eb
|| !extent_buffer_uptodate(eb
)) {
7597 path
->nodes
[level
] = eb
;
7598 path
->slots
[level
] = 0;
7600 btrfs_tree_read_lock(eb
);
7601 btrfs_set_lock_blocking_rw(eb
, BTRFS_READ_LOCK
);
7602 path
->locks
[level
] = BTRFS_READ_LOCK_BLOCKING
;
7604 ret
= btrfs_qgroup_record_ref(trans
, root
->fs_info
,
7608 BTRFS_QGROUP_OPER_SUB_SUBTREE
,
7616 ret
= account_leaf_items(trans
, root
, path
->nodes
[level
]);
7620 /* Nonzero return here means we completed our search */
7621 ret
= adjust_slots_upwards(root
, path
, root_level
);
7625 /* Restart search with new slots */
7634 btrfs_free_path(path
);
7640 * helper to process tree block while walking down the tree.
7642 * when wc->stage == UPDATE_BACKREF, this function updates
7643 * back refs for pointers in the block.
7645 * NOTE: return value 1 means we should stop walking down.
7647 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
7648 struct btrfs_root
*root
,
7649 struct btrfs_path
*path
,
7650 struct walk_control
*wc
, int lookup_info
)
7652 int level
= wc
->level
;
7653 struct extent_buffer
*eb
= path
->nodes
[level
];
7654 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
7657 if (wc
->stage
== UPDATE_BACKREF
&&
7658 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
7662 * when reference count of tree block is 1, it won't increase
7663 * again. once full backref flag is set, we never clear it.
7666 ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
7667 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
)))) {
7668 BUG_ON(!path
->locks
[level
]);
7669 ret
= btrfs_lookup_extent_info(trans
, root
,
7670 eb
->start
, level
, 1,
7673 BUG_ON(ret
== -ENOMEM
);
7676 BUG_ON(wc
->refs
[level
] == 0);
7679 if (wc
->stage
== DROP_REFERENCE
) {
7680 if (wc
->refs
[level
] > 1)
7683 if (path
->locks
[level
] && !wc
->keep_locks
) {
7684 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
7685 path
->locks
[level
] = 0;
7690 /* wc->stage == UPDATE_BACKREF */
7691 if (!(wc
->flags
[level
] & flag
)) {
7692 BUG_ON(!path
->locks
[level
]);
7693 ret
= btrfs_inc_ref(trans
, root
, eb
, 1);
7694 BUG_ON(ret
); /* -ENOMEM */
7695 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
7696 BUG_ON(ret
); /* -ENOMEM */
7697 ret
= btrfs_set_disk_extent_flags(trans
, root
, eb
->start
,
7699 btrfs_header_level(eb
), 0);
7700 BUG_ON(ret
); /* -ENOMEM */
7701 wc
->flags
[level
] |= flag
;
7705 * the block is shared by multiple trees, so it's not good to
7706 * keep the tree lock
7708 if (path
->locks
[level
] && level
> 0) {
7709 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
7710 path
->locks
[level
] = 0;
7716 * helper to process tree block pointer.
7718 * when wc->stage == DROP_REFERENCE, this function checks
7719 * reference count of the block pointed to. if the block
7720 * is shared and we need update back refs for the subtree
7721 * rooted at the block, this function changes wc->stage to
7722 * UPDATE_BACKREF. if the block is shared and there is no
7723 * need to update back, this function drops the reference
7726 * NOTE: return value 1 means we should stop walking down.
7728 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
7729 struct btrfs_root
*root
,
7730 struct btrfs_path
*path
,
7731 struct walk_control
*wc
, int *lookup_info
)
7737 struct btrfs_key key
;
7738 struct extent_buffer
*next
;
7739 int level
= wc
->level
;
7742 bool need_account
= false;
7744 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
7745 path
->slots
[level
]);
7747 * if the lower level block was created before the snapshot
7748 * was created, we know there is no need to update back refs
7751 if (wc
->stage
== UPDATE_BACKREF
&&
7752 generation
<= root
->root_key
.offset
) {
7757 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
7758 blocksize
= root
->nodesize
;
7760 next
= btrfs_find_tree_block(root
, bytenr
);
7762 next
= btrfs_find_create_tree_block(root
, bytenr
);
7765 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, next
,
7769 btrfs_tree_lock(next
);
7770 btrfs_set_lock_blocking(next
);
7772 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, level
- 1, 1,
7773 &wc
->refs
[level
- 1],
7774 &wc
->flags
[level
- 1]);
7776 btrfs_tree_unlock(next
);
7780 if (unlikely(wc
->refs
[level
- 1] == 0)) {
7781 btrfs_err(root
->fs_info
, "Missing references.");
7786 if (wc
->stage
== DROP_REFERENCE
) {
7787 if (wc
->refs
[level
- 1] > 1) {
7788 need_account
= true;
7790 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
7793 if (!wc
->update_ref
||
7794 generation
<= root
->root_key
.offset
)
7797 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
7798 path
->slots
[level
]);
7799 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
7803 wc
->stage
= UPDATE_BACKREF
;
7804 wc
->shared_level
= level
- 1;
7808 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
7812 if (!btrfs_buffer_uptodate(next
, generation
, 0)) {
7813 btrfs_tree_unlock(next
);
7814 free_extent_buffer(next
);
7820 if (reada
&& level
== 1)
7821 reada_walk_down(trans
, root
, wc
, path
);
7822 next
= read_tree_block(root
, bytenr
, generation
);
7823 if (!next
|| !extent_buffer_uptodate(next
)) {
7824 free_extent_buffer(next
);
7827 btrfs_tree_lock(next
);
7828 btrfs_set_lock_blocking(next
);
7832 BUG_ON(level
!= btrfs_header_level(next
));
7833 path
->nodes
[level
] = next
;
7834 path
->slots
[level
] = 0;
7835 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
7841 wc
->refs
[level
- 1] = 0;
7842 wc
->flags
[level
- 1] = 0;
7843 if (wc
->stage
== DROP_REFERENCE
) {
7844 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
7845 parent
= path
->nodes
[level
]->start
;
7847 BUG_ON(root
->root_key
.objectid
!=
7848 btrfs_header_owner(path
->nodes
[level
]));
7853 ret
= account_shared_subtree(trans
, root
, next
,
7854 generation
, level
- 1);
7856 printk_ratelimited(KERN_ERR
"BTRFS: %s Error "
7857 "%d accounting shared subtree. Quota "
7858 "is out of sync, rescan required.\n",
7859 root
->fs_info
->sb
->s_id
, ret
);
7862 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
, parent
,
7863 root
->root_key
.objectid
, level
- 1, 0, 0);
7864 BUG_ON(ret
); /* -ENOMEM */
7866 btrfs_tree_unlock(next
);
7867 free_extent_buffer(next
);
7873 * helper to process tree block while walking up the tree.
7875 * when wc->stage == DROP_REFERENCE, this function drops
7876 * reference count on the block.
7878 * when wc->stage == UPDATE_BACKREF, this function changes
7879 * wc->stage back to DROP_REFERENCE if we changed wc->stage
7880 * to UPDATE_BACKREF previously while processing the block.
7882 * NOTE: return value 1 means we should stop walking up.
7884 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
7885 struct btrfs_root
*root
,
7886 struct btrfs_path
*path
,
7887 struct walk_control
*wc
)
7890 int level
= wc
->level
;
7891 struct extent_buffer
*eb
= path
->nodes
[level
];
7894 if (wc
->stage
== UPDATE_BACKREF
) {
7895 BUG_ON(wc
->shared_level
< level
);
7896 if (level
< wc
->shared_level
)
7899 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
7903 wc
->stage
= DROP_REFERENCE
;
7904 wc
->shared_level
= -1;
7905 path
->slots
[level
] = 0;
7908 * check reference count again if the block isn't locked.
7909 * we should start walking down the tree again if reference
7912 if (!path
->locks
[level
]) {
7914 btrfs_tree_lock(eb
);
7915 btrfs_set_lock_blocking(eb
);
7916 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
7918 ret
= btrfs_lookup_extent_info(trans
, root
,
7919 eb
->start
, level
, 1,
7923 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
7924 path
->locks
[level
] = 0;
7927 BUG_ON(wc
->refs
[level
] == 0);
7928 if (wc
->refs
[level
] == 1) {
7929 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
7930 path
->locks
[level
] = 0;
7936 /* wc->stage == DROP_REFERENCE */
7937 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
7939 if (wc
->refs
[level
] == 1) {
7941 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
7942 ret
= btrfs_dec_ref(trans
, root
, eb
, 1);
7944 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
7945 BUG_ON(ret
); /* -ENOMEM */
7946 ret
= account_leaf_items(trans
, root
, eb
);
7948 printk_ratelimited(KERN_ERR
"BTRFS: %s Error "
7949 "%d accounting leaf items. Quota "
7950 "is out of sync, rescan required.\n",
7951 root
->fs_info
->sb
->s_id
, ret
);
7954 /* make block locked assertion in clean_tree_block happy */
7955 if (!path
->locks
[level
] &&
7956 btrfs_header_generation(eb
) == trans
->transid
) {
7957 btrfs_tree_lock(eb
);
7958 btrfs_set_lock_blocking(eb
);
7959 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
7961 clean_tree_block(trans
, root
, eb
);
7964 if (eb
== root
->node
) {
7965 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
7968 BUG_ON(root
->root_key
.objectid
!=
7969 btrfs_header_owner(eb
));
7971 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
7972 parent
= path
->nodes
[level
+ 1]->start
;
7974 BUG_ON(root
->root_key
.objectid
!=
7975 btrfs_header_owner(path
->nodes
[level
+ 1]));
7978 btrfs_free_tree_block(trans
, root
, eb
, parent
, wc
->refs
[level
] == 1);
7980 wc
->refs
[level
] = 0;
7981 wc
->flags
[level
] = 0;
7985 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
7986 struct btrfs_root
*root
,
7987 struct btrfs_path
*path
,
7988 struct walk_control
*wc
)
7990 int level
= wc
->level
;
7991 int lookup_info
= 1;
7994 while (level
>= 0) {
7995 ret
= walk_down_proc(trans
, root
, path
, wc
, lookup_info
);
8002 if (path
->slots
[level
] >=
8003 btrfs_header_nritems(path
->nodes
[level
]))
8006 ret
= do_walk_down(trans
, root
, path
, wc
, &lookup_info
);
8008 path
->slots
[level
]++;
8017 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
8018 struct btrfs_root
*root
,
8019 struct btrfs_path
*path
,
8020 struct walk_control
*wc
, int max_level
)
8022 int level
= wc
->level
;
8025 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
8026 while (level
< max_level
&& path
->nodes
[level
]) {
8028 if (path
->slots
[level
] + 1 <
8029 btrfs_header_nritems(path
->nodes
[level
])) {
8030 path
->slots
[level
]++;
8033 ret
= walk_up_proc(trans
, root
, path
, wc
);
8037 if (path
->locks
[level
]) {
8038 btrfs_tree_unlock_rw(path
->nodes
[level
],
8039 path
->locks
[level
]);
8040 path
->locks
[level
] = 0;
8042 free_extent_buffer(path
->nodes
[level
]);
8043 path
->nodes
[level
] = NULL
;
8051 * drop a subvolume tree.
8053 * this function traverses the tree freeing any blocks that only
8054 * referenced by the tree.
8056 * when a shared tree block is found. this function decreases its
8057 * reference count by one. if update_ref is true, this function
8058 * also make sure backrefs for the shared block and all lower level
8059 * blocks are properly updated.
8061 * If called with for_reloc == 0, may exit early with -EAGAIN
8063 int btrfs_drop_snapshot(struct btrfs_root
*root
,
8064 struct btrfs_block_rsv
*block_rsv
, int update_ref
,
8067 struct btrfs_path
*path
;
8068 struct btrfs_trans_handle
*trans
;
8069 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
8070 struct btrfs_root_item
*root_item
= &root
->root_item
;
8071 struct walk_control
*wc
;
8072 struct btrfs_key key
;
8076 bool root_dropped
= false;
8078 btrfs_debug(root
->fs_info
, "Drop subvolume %llu", root
->objectid
);
8080 path
= btrfs_alloc_path();
8086 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
8088 btrfs_free_path(path
);
8093 trans
= btrfs_start_transaction(tree_root
, 0);
8094 if (IS_ERR(trans
)) {
8095 err
= PTR_ERR(trans
);
8100 trans
->block_rsv
= block_rsv
;
8102 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
8103 level
= btrfs_header_level(root
->node
);
8104 path
->nodes
[level
] = btrfs_lock_root_node(root
);
8105 btrfs_set_lock_blocking(path
->nodes
[level
]);
8106 path
->slots
[level
] = 0;
8107 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8108 memset(&wc
->update_progress
, 0,
8109 sizeof(wc
->update_progress
));
8111 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
8112 memcpy(&wc
->update_progress
, &key
,
8113 sizeof(wc
->update_progress
));
8115 level
= root_item
->drop_level
;
8117 path
->lowest_level
= level
;
8118 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
8119 path
->lowest_level
= 0;
8127 * unlock our path, this is safe because only this
8128 * function is allowed to delete this snapshot
8130 btrfs_unlock_up_safe(path
, 0);
8132 level
= btrfs_header_level(root
->node
);
8134 btrfs_tree_lock(path
->nodes
[level
]);
8135 btrfs_set_lock_blocking(path
->nodes
[level
]);
8136 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8138 ret
= btrfs_lookup_extent_info(trans
, root
,
8139 path
->nodes
[level
]->start
,
8140 level
, 1, &wc
->refs
[level
],
8146 BUG_ON(wc
->refs
[level
] == 0);
8148 if (level
== root_item
->drop_level
)
8151 btrfs_tree_unlock(path
->nodes
[level
]);
8152 path
->locks
[level
] = 0;
8153 WARN_ON(wc
->refs
[level
] != 1);
8159 wc
->shared_level
= -1;
8160 wc
->stage
= DROP_REFERENCE
;
8161 wc
->update_ref
= update_ref
;
8163 wc
->for_reloc
= for_reloc
;
8164 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
8168 ret
= walk_down_tree(trans
, root
, path
, wc
);
8174 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
8181 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
8185 if (wc
->stage
== DROP_REFERENCE
) {
8187 btrfs_node_key(path
->nodes
[level
],
8188 &root_item
->drop_progress
,
8189 path
->slots
[level
]);
8190 root_item
->drop_level
= level
;
8193 BUG_ON(wc
->level
== 0);
8194 if (btrfs_should_end_transaction(trans
, tree_root
) ||
8195 (!for_reloc
&& btrfs_need_cleaner_sleep(root
))) {
8196 ret
= btrfs_update_root(trans
, tree_root
,
8200 btrfs_abort_transaction(trans
, tree_root
, ret
);
8206 * Qgroup update accounting is run from
8207 * delayed ref handling. This usually works
8208 * out because delayed refs are normally the
8209 * only way qgroup updates are added. However,
8210 * we may have added updates during our tree
8211 * walk so run qgroups here to make sure we
8212 * don't lose any updates.
8214 ret
= btrfs_delayed_qgroup_accounting(trans
,
8217 printk_ratelimited(KERN_ERR
"BTRFS: Failure %d "
8218 "running qgroup updates "
8219 "during snapshot delete. "
8220 "Quota is out of sync, "
8221 "rescan required.\n", ret
);
8223 btrfs_end_transaction_throttle(trans
, tree_root
);
8224 if (!for_reloc
&& btrfs_need_cleaner_sleep(root
)) {
8225 pr_debug("BTRFS: drop snapshot early exit\n");
8230 trans
= btrfs_start_transaction(tree_root
, 0);
8231 if (IS_ERR(trans
)) {
8232 err
= PTR_ERR(trans
);
8236 trans
->block_rsv
= block_rsv
;
8239 btrfs_release_path(path
);
8243 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
8245 btrfs_abort_transaction(trans
, tree_root
, ret
);
8249 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
8250 ret
= btrfs_find_root(tree_root
, &root
->root_key
, path
,
8253 btrfs_abort_transaction(trans
, tree_root
, ret
);
8256 } else if (ret
> 0) {
8257 /* if we fail to delete the orphan item this time
8258 * around, it'll get picked up the next time.
8260 * The most common failure here is just -ENOENT.
8262 btrfs_del_orphan_item(trans
, tree_root
,
8263 root
->root_key
.objectid
);
8267 if (test_bit(BTRFS_ROOT_IN_RADIX
, &root
->state
)) {
8268 btrfs_drop_and_free_fs_root(tree_root
->fs_info
, root
);
8270 free_extent_buffer(root
->node
);
8271 free_extent_buffer(root
->commit_root
);
8272 btrfs_put_fs_root(root
);
8274 root_dropped
= true;
8276 ret
= btrfs_delayed_qgroup_accounting(trans
, tree_root
->fs_info
);
8278 printk_ratelimited(KERN_ERR
"BTRFS: Failure %d "
8279 "running qgroup updates "
8280 "during snapshot delete. "
8281 "Quota is out of sync, "
8282 "rescan required.\n", ret
);
8284 btrfs_end_transaction_throttle(trans
, tree_root
);
8287 btrfs_free_path(path
);
8290 * So if we need to stop dropping the snapshot for whatever reason we
8291 * need to make sure to add it back to the dead root list so that we
8292 * keep trying to do the work later. This also cleans up roots if we
8293 * don't have it in the radix (like when we recover after a power fail
8294 * or unmount) so we don't leak memory.
8296 if (!for_reloc
&& root_dropped
== false)
8297 btrfs_add_dead_root(root
);
8298 if (err
&& err
!= -EAGAIN
)
8299 btrfs_std_error(root
->fs_info
, err
);
8304 * drop subtree rooted at tree block 'node'.
8306 * NOTE: this function will unlock and release tree block 'node'
8307 * only used by relocation code
8309 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
8310 struct btrfs_root
*root
,
8311 struct extent_buffer
*node
,
8312 struct extent_buffer
*parent
)
8314 struct btrfs_path
*path
;
8315 struct walk_control
*wc
;
8321 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
8323 path
= btrfs_alloc_path();
8327 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
8329 btrfs_free_path(path
);
8333 btrfs_assert_tree_locked(parent
);
8334 parent_level
= btrfs_header_level(parent
);
8335 extent_buffer_get(parent
);
8336 path
->nodes
[parent_level
] = parent
;
8337 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
8339 btrfs_assert_tree_locked(node
);
8340 level
= btrfs_header_level(node
);
8341 path
->nodes
[level
] = node
;
8342 path
->slots
[level
] = 0;
8343 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8345 wc
->refs
[parent_level
] = 1;
8346 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
8348 wc
->shared_level
= -1;
8349 wc
->stage
= DROP_REFERENCE
;
8353 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
8356 wret
= walk_down_tree(trans
, root
, path
, wc
);
8362 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
8370 btrfs_free_path(path
);
8374 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
8380 * if restripe for this chunk_type is on pick target profile and
8381 * return, otherwise do the usual balance
8383 stripped
= get_restripe_target(root
->fs_info
, flags
);
8385 return extended_to_chunk(stripped
);
8387 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
8389 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
8390 BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
|
8391 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
8393 if (num_devices
== 1) {
8394 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
8395 stripped
= flags
& ~stripped
;
8397 /* turn raid0 into single device chunks */
8398 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
8401 /* turn mirroring into duplication */
8402 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
8403 BTRFS_BLOCK_GROUP_RAID10
))
8404 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
8406 /* they already had raid on here, just return */
8407 if (flags
& stripped
)
8410 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
8411 stripped
= flags
& ~stripped
;
8413 /* switch duplicated blocks with raid1 */
8414 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
8415 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
8417 /* this is drive concat, leave it alone */
8423 static int set_block_group_ro(struct btrfs_block_group_cache
*cache
, int force
)
8425 struct btrfs_space_info
*sinfo
= cache
->space_info
;
8427 u64 min_allocable_bytes
;
8432 * We need some metadata space and system metadata space for
8433 * allocating chunks in some corner cases until we force to set
8434 * it to be readonly.
8437 (BTRFS_BLOCK_GROUP_SYSTEM
| BTRFS_BLOCK_GROUP_METADATA
)) &&
8439 min_allocable_bytes
= 1 * 1024 * 1024;
8441 min_allocable_bytes
= 0;
8443 spin_lock(&sinfo
->lock
);
8444 spin_lock(&cache
->lock
);
8451 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
8452 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
8454 if (sinfo
->bytes_used
+ sinfo
->bytes_reserved
+ sinfo
->bytes_pinned
+
8455 sinfo
->bytes_may_use
+ sinfo
->bytes_readonly
+ num_bytes
+
8456 min_allocable_bytes
<= sinfo
->total_bytes
) {
8457 sinfo
->bytes_readonly
+= num_bytes
;
8459 list_add_tail(&cache
->ro_list
, &sinfo
->ro_bgs
);
8463 spin_unlock(&cache
->lock
);
8464 spin_unlock(&sinfo
->lock
);
8468 int btrfs_set_block_group_ro(struct btrfs_root
*root
,
8469 struct btrfs_block_group_cache
*cache
)
8472 struct btrfs_trans_handle
*trans
;
8478 trans
= btrfs_join_transaction(root
);
8480 return PTR_ERR(trans
);
8482 alloc_flags
= update_block_group_flags(root
, cache
->flags
);
8483 if (alloc_flags
!= cache
->flags
) {
8484 ret
= do_chunk_alloc(trans
, root
, alloc_flags
,
8490 ret
= set_block_group_ro(cache
, 0);
8493 alloc_flags
= get_alloc_profile(root
, cache
->space_info
->flags
);
8494 ret
= do_chunk_alloc(trans
, root
, alloc_flags
,
8498 ret
= set_block_group_ro(cache
, 0);
8500 btrfs_end_transaction(trans
, root
);
8504 int btrfs_force_chunk_alloc(struct btrfs_trans_handle
*trans
,
8505 struct btrfs_root
*root
, u64 type
)
8507 u64 alloc_flags
= get_alloc_profile(root
, type
);
8508 return do_chunk_alloc(trans
, root
, alloc_flags
,
8513 * helper to account the unused space of all the readonly block group in the
8514 * space_info. takes mirrors into account.
8516 u64
btrfs_account_ro_block_groups_free_space(struct btrfs_space_info
*sinfo
)
8518 struct btrfs_block_group_cache
*block_group
;
8522 /* It's df, we don't care if it's racey */
8523 if (list_empty(&sinfo
->ro_bgs
))
8526 spin_lock(&sinfo
->lock
);
8527 list_for_each_entry(block_group
, &sinfo
->ro_bgs
, ro_list
) {
8528 spin_lock(&block_group
->lock
);
8530 if (!block_group
->ro
) {
8531 spin_unlock(&block_group
->lock
);
8535 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_RAID1
|
8536 BTRFS_BLOCK_GROUP_RAID10
|
8537 BTRFS_BLOCK_GROUP_DUP
))
8542 free_bytes
+= (block_group
->key
.offset
-
8543 btrfs_block_group_used(&block_group
->item
)) *
8546 spin_unlock(&block_group
->lock
);
8548 spin_unlock(&sinfo
->lock
);
8553 void btrfs_set_block_group_rw(struct btrfs_root
*root
,
8554 struct btrfs_block_group_cache
*cache
)
8556 struct btrfs_space_info
*sinfo
= cache
->space_info
;
8561 spin_lock(&sinfo
->lock
);
8562 spin_lock(&cache
->lock
);
8563 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
8564 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
8565 sinfo
->bytes_readonly
-= num_bytes
;
8567 list_del_init(&cache
->ro_list
);
8568 spin_unlock(&cache
->lock
);
8569 spin_unlock(&sinfo
->lock
);
8573 * checks to see if its even possible to relocate this block group.
8575 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8576 * ok to go ahead and try.
8578 int btrfs_can_relocate(struct btrfs_root
*root
, u64 bytenr
)
8580 struct btrfs_block_group_cache
*block_group
;
8581 struct btrfs_space_info
*space_info
;
8582 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
8583 struct btrfs_device
*device
;
8584 struct btrfs_trans_handle
*trans
;
8593 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
8595 /* odd, couldn't find the block group, leave it alone */
8599 min_free
= btrfs_block_group_used(&block_group
->item
);
8601 /* no bytes used, we're good */
8605 space_info
= block_group
->space_info
;
8606 spin_lock(&space_info
->lock
);
8608 full
= space_info
->full
;
8611 * if this is the last block group we have in this space, we can't
8612 * relocate it unless we're able to allocate a new chunk below.
8614 * Otherwise, we need to make sure we have room in the space to handle
8615 * all of the extents from this block group. If we can, we're good
8617 if ((space_info
->total_bytes
!= block_group
->key
.offset
) &&
8618 (space_info
->bytes_used
+ space_info
->bytes_reserved
+
8619 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
8620 min_free
< space_info
->total_bytes
)) {
8621 spin_unlock(&space_info
->lock
);
8624 spin_unlock(&space_info
->lock
);
8627 * ok we don't have enough space, but maybe we have free space on our
8628 * devices to allocate new chunks for relocation, so loop through our
8629 * alloc devices and guess if we have enough space. if this block
8630 * group is going to be restriped, run checks against the target
8631 * profile instead of the current one.
8643 target
= get_restripe_target(root
->fs_info
, block_group
->flags
);
8645 index
= __get_raid_index(extended_to_chunk(target
));
8648 * this is just a balance, so if we were marked as full
8649 * we know there is no space for a new chunk
8654 index
= get_block_group_index(block_group
);
8657 if (index
== BTRFS_RAID_RAID10
) {
8661 } else if (index
== BTRFS_RAID_RAID1
) {
8663 } else if (index
== BTRFS_RAID_DUP
) {
8666 } else if (index
== BTRFS_RAID_RAID0
) {
8667 dev_min
= fs_devices
->rw_devices
;
8668 do_div(min_free
, dev_min
);
8671 /* We need to do this so that we can look at pending chunks */
8672 trans
= btrfs_join_transaction(root
);
8673 if (IS_ERR(trans
)) {
8674 ret
= PTR_ERR(trans
);
8678 mutex_lock(&root
->fs_info
->chunk_mutex
);
8679 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
8683 * check to make sure we can actually find a chunk with enough
8684 * space to fit our block group in.
8686 if (device
->total_bytes
> device
->bytes_used
+ min_free
&&
8687 !device
->is_tgtdev_for_dev_replace
) {
8688 ret
= find_free_dev_extent(trans
, device
, min_free
,
8693 if (dev_nr
>= dev_min
)
8699 mutex_unlock(&root
->fs_info
->chunk_mutex
);
8700 btrfs_end_transaction(trans
, root
);
8702 btrfs_put_block_group(block_group
);
8706 static int find_first_block_group(struct btrfs_root
*root
,
8707 struct btrfs_path
*path
, struct btrfs_key
*key
)
8710 struct btrfs_key found_key
;
8711 struct extent_buffer
*leaf
;
8714 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
8719 slot
= path
->slots
[0];
8720 leaf
= path
->nodes
[0];
8721 if (slot
>= btrfs_header_nritems(leaf
)) {
8722 ret
= btrfs_next_leaf(root
, path
);
8729 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
8731 if (found_key
.objectid
>= key
->objectid
&&
8732 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
8742 void btrfs_put_block_group_cache(struct btrfs_fs_info
*info
)
8744 struct btrfs_block_group_cache
*block_group
;
8748 struct inode
*inode
;
8750 block_group
= btrfs_lookup_first_block_group(info
, last
);
8751 while (block_group
) {
8752 spin_lock(&block_group
->lock
);
8753 if (block_group
->iref
)
8755 spin_unlock(&block_group
->lock
);
8756 block_group
= next_block_group(info
->tree_root
,
8766 inode
= block_group
->inode
;
8767 block_group
->iref
= 0;
8768 block_group
->inode
= NULL
;
8769 spin_unlock(&block_group
->lock
);
8771 last
= block_group
->key
.objectid
+ block_group
->key
.offset
;
8772 btrfs_put_block_group(block_group
);
8776 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
8778 struct btrfs_block_group_cache
*block_group
;
8779 struct btrfs_space_info
*space_info
;
8780 struct btrfs_caching_control
*caching_ctl
;
8783 down_write(&info
->commit_root_sem
);
8784 while (!list_empty(&info
->caching_block_groups
)) {
8785 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
8786 struct btrfs_caching_control
, list
);
8787 list_del(&caching_ctl
->list
);
8788 put_caching_control(caching_ctl
);
8790 up_write(&info
->commit_root_sem
);
8792 spin_lock(&info
->unused_bgs_lock
);
8793 while (!list_empty(&info
->unused_bgs
)) {
8794 block_group
= list_first_entry(&info
->unused_bgs
,
8795 struct btrfs_block_group_cache
,
8797 list_del_init(&block_group
->bg_list
);
8798 btrfs_put_block_group(block_group
);
8800 spin_unlock(&info
->unused_bgs_lock
);
8802 spin_lock(&info
->block_group_cache_lock
);
8803 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
8804 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
8806 rb_erase(&block_group
->cache_node
,
8807 &info
->block_group_cache_tree
);
8808 RB_CLEAR_NODE(&block_group
->cache_node
);
8809 spin_unlock(&info
->block_group_cache_lock
);
8811 down_write(&block_group
->space_info
->groups_sem
);
8812 list_del(&block_group
->list
);
8813 up_write(&block_group
->space_info
->groups_sem
);
8815 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
8816 wait_block_group_cache_done(block_group
);
8819 * We haven't cached this block group, which means we could
8820 * possibly have excluded extents on this block group.
8822 if (block_group
->cached
== BTRFS_CACHE_NO
||
8823 block_group
->cached
== BTRFS_CACHE_ERROR
)
8824 free_excluded_extents(info
->extent_root
, block_group
);
8826 btrfs_remove_free_space_cache(block_group
);
8827 btrfs_put_block_group(block_group
);
8829 spin_lock(&info
->block_group_cache_lock
);
8831 spin_unlock(&info
->block_group_cache_lock
);
8833 /* now that all the block groups are freed, go through and
8834 * free all the space_info structs. This is only called during
8835 * the final stages of unmount, and so we know nobody is
8836 * using them. We call synchronize_rcu() once before we start,
8837 * just to be on the safe side.
8841 release_global_block_rsv(info
);
8843 while (!list_empty(&info
->space_info
)) {
8846 space_info
= list_entry(info
->space_info
.next
,
8847 struct btrfs_space_info
,
8849 if (btrfs_test_opt(info
->tree_root
, ENOSPC_DEBUG
)) {
8850 if (WARN_ON(space_info
->bytes_pinned
> 0 ||
8851 space_info
->bytes_reserved
> 0 ||
8852 space_info
->bytes_may_use
> 0)) {
8853 dump_space_info(space_info
, 0, 0);
8856 list_del(&space_info
->list
);
8857 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++) {
8858 struct kobject
*kobj
;
8859 kobj
= space_info
->block_group_kobjs
[i
];
8860 space_info
->block_group_kobjs
[i
] = NULL
;
8866 kobject_del(&space_info
->kobj
);
8867 kobject_put(&space_info
->kobj
);
8872 static void __link_block_group(struct btrfs_space_info
*space_info
,
8873 struct btrfs_block_group_cache
*cache
)
8875 int index
= get_block_group_index(cache
);
8878 down_write(&space_info
->groups_sem
);
8879 if (list_empty(&space_info
->block_groups
[index
]))
8881 list_add_tail(&cache
->list
, &space_info
->block_groups
[index
]);
8882 up_write(&space_info
->groups_sem
);
8885 struct raid_kobject
*rkobj
;
8888 rkobj
= kzalloc(sizeof(*rkobj
), GFP_NOFS
);
8891 rkobj
->raid_type
= index
;
8892 kobject_init(&rkobj
->kobj
, &btrfs_raid_ktype
);
8893 ret
= kobject_add(&rkobj
->kobj
, &space_info
->kobj
,
8894 "%s", get_raid_name(index
));
8896 kobject_put(&rkobj
->kobj
);
8899 space_info
->block_group_kobjs
[index
] = &rkobj
->kobj
;
8904 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
8907 static struct btrfs_block_group_cache
*
8908 btrfs_create_block_group_cache(struct btrfs_root
*root
, u64 start
, u64 size
)
8910 struct btrfs_block_group_cache
*cache
;
8912 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
8916 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
8918 if (!cache
->free_space_ctl
) {
8923 cache
->key
.objectid
= start
;
8924 cache
->key
.offset
= size
;
8925 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
8927 cache
->sectorsize
= root
->sectorsize
;
8928 cache
->fs_info
= root
->fs_info
;
8929 cache
->full_stripe_len
= btrfs_full_stripe_len(root
,
8930 &root
->fs_info
->mapping_tree
,
8932 atomic_set(&cache
->count
, 1);
8933 spin_lock_init(&cache
->lock
);
8934 init_rwsem(&cache
->data_rwsem
);
8935 INIT_LIST_HEAD(&cache
->list
);
8936 INIT_LIST_HEAD(&cache
->cluster_list
);
8937 INIT_LIST_HEAD(&cache
->bg_list
);
8938 INIT_LIST_HEAD(&cache
->ro_list
);
8939 INIT_LIST_HEAD(&cache
->dirty_list
);
8940 btrfs_init_free_space_ctl(cache
);
8941 atomic_set(&cache
->trimming
, 0);
8946 int btrfs_read_block_groups(struct btrfs_root
*root
)
8948 struct btrfs_path
*path
;
8950 struct btrfs_block_group_cache
*cache
;
8951 struct btrfs_fs_info
*info
= root
->fs_info
;
8952 struct btrfs_space_info
*space_info
;
8953 struct btrfs_key key
;
8954 struct btrfs_key found_key
;
8955 struct extent_buffer
*leaf
;
8959 root
= info
->extent_root
;
8962 key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
8963 path
= btrfs_alloc_path();
8968 cache_gen
= btrfs_super_cache_generation(root
->fs_info
->super_copy
);
8969 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
8970 btrfs_super_generation(root
->fs_info
->super_copy
) != cache_gen
)
8972 if (btrfs_test_opt(root
, CLEAR_CACHE
))
8976 ret
= find_first_block_group(root
, path
, &key
);
8982 leaf
= path
->nodes
[0];
8983 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
8985 cache
= btrfs_create_block_group_cache(root
, found_key
.objectid
,
8994 * When we mount with old space cache, we need to
8995 * set BTRFS_DC_CLEAR and set dirty flag.
8997 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
8998 * truncate the old free space cache inode and
9000 * b) Setting 'dirty flag' makes sure that we flush
9001 * the new space cache info onto disk.
9003 if (btrfs_test_opt(root
, SPACE_CACHE
))
9004 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
9007 read_extent_buffer(leaf
, &cache
->item
,
9008 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
9009 sizeof(cache
->item
));
9010 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
9012 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
9013 btrfs_release_path(path
);
9016 * We need to exclude the super stripes now so that the space
9017 * info has super bytes accounted for, otherwise we'll think
9018 * we have more space than we actually do.
9020 ret
= exclude_super_stripes(root
, cache
);
9023 * We may have excluded something, so call this just in
9026 free_excluded_extents(root
, cache
);
9027 btrfs_put_block_group(cache
);
9032 * check for two cases, either we are full, and therefore
9033 * don't need to bother with the caching work since we won't
9034 * find any space, or we are empty, and we can just add all
9035 * the space in and be done with it. This saves us _alot_ of
9036 * time, particularly in the full case.
9038 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
9039 cache
->last_byte_to_unpin
= (u64
)-1;
9040 cache
->cached
= BTRFS_CACHE_FINISHED
;
9041 free_excluded_extents(root
, cache
);
9042 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
9043 cache
->last_byte_to_unpin
= (u64
)-1;
9044 cache
->cached
= BTRFS_CACHE_FINISHED
;
9045 add_new_free_space(cache
, root
->fs_info
,
9047 found_key
.objectid
+
9049 free_excluded_extents(root
, cache
);
9052 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
9054 btrfs_remove_free_space_cache(cache
);
9055 btrfs_put_block_group(cache
);
9059 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
9060 btrfs_block_group_used(&cache
->item
),
9063 btrfs_remove_free_space_cache(cache
);
9064 spin_lock(&info
->block_group_cache_lock
);
9065 rb_erase(&cache
->cache_node
,
9066 &info
->block_group_cache_tree
);
9067 RB_CLEAR_NODE(&cache
->cache_node
);
9068 spin_unlock(&info
->block_group_cache_lock
);
9069 btrfs_put_block_group(cache
);
9073 cache
->space_info
= space_info
;
9074 spin_lock(&cache
->space_info
->lock
);
9075 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
9076 spin_unlock(&cache
->space_info
->lock
);
9078 __link_block_group(space_info
, cache
);
9080 set_avail_alloc_bits(root
->fs_info
, cache
->flags
);
9081 if (btrfs_chunk_readonly(root
, cache
->key
.objectid
)) {
9082 set_block_group_ro(cache
, 1);
9083 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
9084 spin_lock(&info
->unused_bgs_lock
);
9085 /* Should always be true but just in case. */
9086 if (list_empty(&cache
->bg_list
)) {
9087 btrfs_get_block_group(cache
);
9088 list_add_tail(&cache
->bg_list
,
9091 spin_unlock(&info
->unused_bgs_lock
);
9095 list_for_each_entry_rcu(space_info
, &root
->fs_info
->space_info
, list
) {
9096 if (!(get_alloc_profile(root
, space_info
->flags
) &
9097 (BTRFS_BLOCK_GROUP_RAID10
|
9098 BTRFS_BLOCK_GROUP_RAID1
|
9099 BTRFS_BLOCK_GROUP_RAID5
|
9100 BTRFS_BLOCK_GROUP_RAID6
|
9101 BTRFS_BLOCK_GROUP_DUP
)))
9104 * avoid allocating from un-mirrored block group if there are
9105 * mirrored block groups.
9107 list_for_each_entry(cache
,
9108 &space_info
->block_groups
[BTRFS_RAID_RAID0
],
9110 set_block_group_ro(cache
, 1);
9111 list_for_each_entry(cache
,
9112 &space_info
->block_groups
[BTRFS_RAID_SINGLE
],
9114 set_block_group_ro(cache
, 1);
9117 init_global_block_rsv(info
);
9120 btrfs_free_path(path
);
9124 void btrfs_create_pending_block_groups(struct btrfs_trans_handle
*trans
,
9125 struct btrfs_root
*root
)
9127 struct btrfs_block_group_cache
*block_group
, *tmp
;
9128 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
9129 struct btrfs_block_group_item item
;
9130 struct btrfs_key key
;
9133 list_for_each_entry_safe(block_group
, tmp
, &trans
->new_bgs
, bg_list
) {
9137 spin_lock(&block_group
->lock
);
9138 memcpy(&item
, &block_group
->item
, sizeof(item
));
9139 memcpy(&key
, &block_group
->key
, sizeof(key
));
9140 spin_unlock(&block_group
->lock
);
9142 ret
= btrfs_insert_item(trans
, extent_root
, &key
, &item
,
9145 btrfs_abort_transaction(trans
, extent_root
, ret
);
9146 ret
= btrfs_finish_chunk_alloc(trans
, extent_root
,
9147 key
.objectid
, key
.offset
);
9149 btrfs_abort_transaction(trans
, extent_root
, ret
);
9151 list_del_init(&block_group
->bg_list
);
9155 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
9156 struct btrfs_root
*root
, u64 bytes_used
,
9157 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
9161 struct btrfs_root
*extent_root
;
9162 struct btrfs_block_group_cache
*cache
;
9164 extent_root
= root
->fs_info
->extent_root
;
9166 btrfs_set_log_full_commit(root
->fs_info
, trans
);
9168 cache
= btrfs_create_block_group_cache(root
, chunk_offset
, size
);
9172 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
9173 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
9174 btrfs_set_block_group_flags(&cache
->item
, type
);
9176 cache
->flags
= type
;
9177 cache
->last_byte_to_unpin
= (u64
)-1;
9178 cache
->cached
= BTRFS_CACHE_FINISHED
;
9179 ret
= exclude_super_stripes(root
, cache
);
9182 * We may have excluded something, so call this just in
9185 free_excluded_extents(root
, cache
);
9186 btrfs_put_block_group(cache
);
9190 add_new_free_space(cache
, root
->fs_info
, chunk_offset
,
9191 chunk_offset
+ size
);
9193 free_excluded_extents(root
, cache
);
9195 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
9197 btrfs_remove_free_space_cache(cache
);
9198 btrfs_put_block_group(cache
);
9202 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
9203 &cache
->space_info
);
9205 btrfs_remove_free_space_cache(cache
);
9206 spin_lock(&root
->fs_info
->block_group_cache_lock
);
9207 rb_erase(&cache
->cache_node
,
9208 &root
->fs_info
->block_group_cache_tree
);
9209 RB_CLEAR_NODE(&cache
->cache_node
);
9210 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
9211 btrfs_put_block_group(cache
);
9214 update_global_block_rsv(root
->fs_info
);
9216 spin_lock(&cache
->space_info
->lock
);
9217 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
9218 spin_unlock(&cache
->space_info
->lock
);
9220 __link_block_group(cache
->space_info
, cache
);
9222 list_add_tail(&cache
->bg_list
, &trans
->new_bgs
);
9224 set_avail_alloc_bits(extent_root
->fs_info
, type
);
9229 static void clear_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
9231 u64 extra_flags
= chunk_to_extended(flags
) &
9232 BTRFS_EXTENDED_PROFILE_MASK
;
9234 write_seqlock(&fs_info
->profiles_lock
);
9235 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
9236 fs_info
->avail_data_alloc_bits
&= ~extra_flags
;
9237 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
9238 fs_info
->avail_metadata_alloc_bits
&= ~extra_flags
;
9239 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
9240 fs_info
->avail_system_alloc_bits
&= ~extra_flags
;
9241 write_sequnlock(&fs_info
->profiles_lock
);
9244 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
9245 struct btrfs_root
*root
, u64 group_start
,
9246 struct extent_map
*em
)
9248 struct btrfs_path
*path
;
9249 struct btrfs_block_group_cache
*block_group
;
9250 struct btrfs_free_cluster
*cluster
;
9251 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
9252 struct btrfs_key key
;
9253 struct inode
*inode
;
9254 struct kobject
*kobj
= NULL
;
9258 struct btrfs_caching_control
*caching_ctl
= NULL
;
9261 root
= root
->fs_info
->extent_root
;
9263 block_group
= btrfs_lookup_block_group(root
->fs_info
, group_start
);
9264 BUG_ON(!block_group
);
9265 BUG_ON(!block_group
->ro
);
9268 * Free the reserved super bytes from this block group before
9271 free_excluded_extents(root
, block_group
);
9273 memcpy(&key
, &block_group
->key
, sizeof(key
));
9274 index
= get_block_group_index(block_group
);
9275 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
9276 BTRFS_BLOCK_GROUP_RAID1
|
9277 BTRFS_BLOCK_GROUP_RAID10
))
9282 /* make sure this block group isn't part of an allocation cluster */
9283 cluster
= &root
->fs_info
->data_alloc_cluster
;
9284 spin_lock(&cluster
->refill_lock
);
9285 btrfs_return_cluster_to_free_space(block_group
, cluster
);
9286 spin_unlock(&cluster
->refill_lock
);
9289 * make sure this block group isn't part of a metadata
9290 * allocation cluster
9292 cluster
= &root
->fs_info
->meta_alloc_cluster
;
9293 spin_lock(&cluster
->refill_lock
);
9294 btrfs_return_cluster_to_free_space(block_group
, cluster
);
9295 spin_unlock(&cluster
->refill_lock
);
9297 path
= btrfs_alloc_path();
9303 inode
= lookup_free_space_inode(tree_root
, block_group
, path
);
9304 if (!IS_ERR(inode
)) {
9305 ret
= btrfs_orphan_add(trans
, inode
);
9307 btrfs_add_delayed_iput(inode
);
9311 /* One for the block groups ref */
9312 spin_lock(&block_group
->lock
);
9313 if (block_group
->iref
) {
9314 block_group
->iref
= 0;
9315 block_group
->inode
= NULL
;
9316 spin_unlock(&block_group
->lock
);
9319 spin_unlock(&block_group
->lock
);
9321 /* One for our lookup ref */
9322 btrfs_add_delayed_iput(inode
);
9325 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
9326 key
.offset
= block_group
->key
.objectid
;
9329 ret
= btrfs_search_slot(trans
, tree_root
, &key
, path
, -1, 1);
9333 btrfs_release_path(path
);
9335 ret
= btrfs_del_item(trans
, tree_root
, path
);
9338 btrfs_release_path(path
);
9341 spin_lock(&root
->fs_info
->block_group_cache_lock
);
9342 rb_erase(&block_group
->cache_node
,
9343 &root
->fs_info
->block_group_cache_tree
);
9344 RB_CLEAR_NODE(&block_group
->cache_node
);
9346 if (root
->fs_info
->first_logical_byte
== block_group
->key
.objectid
)
9347 root
->fs_info
->first_logical_byte
= (u64
)-1;
9348 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
9350 down_write(&block_group
->space_info
->groups_sem
);
9352 * we must use list_del_init so people can check to see if they
9353 * are still on the list after taking the semaphore
9355 list_del_init(&block_group
->list
);
9356 list_del_init(&block_group
->ro_list
);
9357 if (list_empty(&block_group
->space_info
->block_groups
[index
])) {
9358 kobj
= block_group
->space_info
->block_group_kobjs
[index
];
9359 block_group
->space_info
->block_group_kobjs
[index
] = NULL
;
9360 clear_avail_alloc_bits(root
->fs_info
, block_group
->flags
);
9362 up_write(&block_group
->space_info
->groups_sem
);
9368 if (block_group
->has_caching_ctl
)
9369 caching_ctl
= get_caching_control(block_group
);
9370 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
9371 wait_block_group_cache_done(block_group
);
9372 if (block_group
->has_caching_ctl
) {
9373 down_write(&root
->fs_info
->commit_root_sem
);
9375 struct btrfs_caching_control
*ctl
;
9377 list_for_each_entry(ctl
,
9378 &root
->fs_info
->caching_block_groups
, list
)
9379 if (ctl
->block_group
== block_group
) {
9381 atomic_inc(&caching_ctl
->count
);
9386 list_del_init(&caching_ctl
->list
);
9387 up_write(&root
->fs_info
->commit_root_sem
);
9389 /* Once for the caching bgs list and once for us. */
9390 put_caching_control(caching_ctl
);
9391 put_caching_control(caching_ctl
);
9395 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
9396 if (!list_empty(&block_group
->dirty_list
)) {
9397 list_del_init(&block_group
->dirty_list
);
9398 btrfs_put_block_group(block_group
);
9400 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
9402 btrfs_remove_free_space_cache(block_group
);
9404 spin_lock(&block_group
->space_info
->lock
);
9405 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
9406 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
9407 block_group
->space_info
->disk_total
-= block_group
->key
.offset
* factor
;
9408 spin_unlock(&block_group
->space_info
->lock
);
9410 memcpy(&key
, &block_group
->key
, sizeof(key
));
9413 if (!list_empty(&em
->list
)) {
9414 /* We're in the transaction->pending_chunks list. */
9415 free_extent_map(em
);
9417 spin_lock(&block_group
->lock
);
9418 block_group
->removed
= 1;
9420 * At this point trimming can't start on this block group, because we
9421 * removed the block group from the tree fs_info->block_group_cache_tree
9422 * so no one can't find it anymore and even if someone already got this
9423 * block group before we removed it from the rbtree, they have already
9424 * incremented block_group->trimming - if they didn't, they won't find
9425 * any free space entries because we already removed them all when we
9426 * called btrfs_remove_free_space_cache().
9428 * And we must not remove the extent map from the fs_info->mapping_tree
9429 * to prevent the same logical address range and physical device space
9430 * ranges from being reused for a new block group. This is because our
9431 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
9432 * completely transactionless, so while it is trimming a range the
9433 * currently running transaction might finish and a new one start,
9434 * allowing for new block groups to be created that can reuse the same
9435 * physical device locations unless we take this special care.
9437 remove_em
= (atomic_read(&block_group
->trimming
) == 0);
9439 * Make sure a trimmer task always sees the em in the pinned_chunks list
9440 * if it sees block_group->removed == 1 (needs to lock block_group->lock
9441 * before checking block_group->removed).
9445 * Our em might be in trans->transaction->pending_chunks which
9446 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
9447 * and so is the fs_info->pinned_chunks list.
9449 * So at this point we must be holding the chunk_mutex to avoid
9450 * any races with chunk allocation (more specifically at
9451 * volumes.c:contains_pending_extent()), to ensure it always
9452 * sees the em, either in the pending_chunks list or in the
9453 * pinned_chunks list.
9455 list_move_tail(&em
->list
, &root
->fs_info
->pinned_chunks
);
9457 spin_unlock(&block_group
->lock
);
9460 struct extent_map_tree
*em_tree
;
9462 em_tree
= &root
->fs_info
->mapping_tree
.map_tree
;
9463 write_lock(&em_tree
->lock
);
9465 * The em might be in the pending_chunks list, so make sure the
9466 * chunk mutex is locked, since remove_extent_mapping() will
9467 * delete us from that list.
9469 remove_extent_mapping(em_tree
, em
);
9470 write_unlock(&em_tree
->lock
);
9471 /* once for the tree */
9472 free_extent_map(em
);
9475 unlock_chunks(root
);
9477 btrfs_put_block_group(block_group
);
9478 btrfs_put_block_group(block_group
);
9480 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
9486 ret
= btrfs_del_item(trans
, root
, path
);
9488 btrfs_free_path(path
);
9493 * Process the unused_bgs list and remove any that don't have any allocated
9494 * space inside of them.
9496 void btrfs_delete_unused_bgs(struct btrfs_fs_info
*fs_info
)
9498 struct btrfs_block_group_cache
*block_group
;
9499 struct btrfs_space_info
*space_info
;
9500 struct btrfs_root
*root
= fs_info
->extent_root
;
9501 struct btrfs_trans_handle
*trans
;
9507 spin_lock(&fs_info
->unused_bgs_lock
);
9508 while (!list_empty(&fs_info
->unused_bgs
)) {
9511 block_group
= list_first_entry(&fs_info
->unused_bgs
,
9512 struct btrfs_block_group_cache
,
9514 space_info
= block_group
->space_info
;
9515 list_del_init(&block_group
->bg_list
);
9516 if (ret
|| btrfs_mixed_space_info(space_info
)) {
9517 btrfs_put_block_group(block_group
);
9520 spin_unlock(&fs_info
->unused_bgs_lock
);
9522 /* Don't want to race with allocators so take the groups_sem */
9523 down_write(&space_info
->groups_sem
);
9524 spin_lock(&block_group
->lock
);
9525 if (block_group
->reserved
||
9526 btrfs_block_group_used(&block_group
->item
) ||
9529 * We want to bail if we made new allocations or have
9530 * outstanding allocations in this block group. We do
9531 * the ro check in case balance is currently acting on
9534 spin_unlock(&block_group
->lock
);
9535 up_write(&space_info
->groups_sem
);
9538 spin_unlock(&block_group
->lock
);
9540 /* We don't want to force the issue, only flip if it's ok. */
9541 ret
= set_block_group_ro(block_group
, 0);
9542 up_write(&space_info
->groups_sem
);
9549 * Want to do this before we do anything else so we can recover
9550 * properly if we fail to join the transaction.
9552 trans
= btrfs_join_transaction(root
);
9553 if (IS_ERR(trans
)) {
9554 btrfs_set_block_group_rw(root
, block_group
);
9555 ret
= PTR_ERR(trans
);
9560 * We could have pending pinned extents for this block group,
9561 * just delete them, we don't care about them anymore.
9563 start
= block_group
->key
.objectid
;
9564 end
= start
+ block_group
->key
.offset
- 1;
9565 ret
= clear_extent_bits(&fs_info
->freed_extents
[0], start
, end
,
9566 EXTENT_DIRTY
, GFP_NOFS
);
9568 btrfs_set_block_group_rw(root
, block_group
);
9571 ret
= clear_extent_bits(&fs_info
->freed_extents
[1], start
, end
,
9572 EXTENT_DIRTY
, GFP_NOFS
);
9574 btrfs_set_block_group_rw(root
, block_group
);
9578 /* Reset pinned so btrfs_put_block_group doesn't complain */
9579 block_group
->pinned
= 0;
9582 * Btrfs_remove_chunk will abort the transaction if things go
9585 ret
= btrfs_remove_chunk(trans
, root
,
9586 block_group
->key
.objectid
);
9588 btrfs_end_transaction(trans
, root
);
9590 btrfs_put_block_group(block_group
);
9591 spin_lock(&fs_info
->unused_bgs_lock
);
9593 spin_unlock(&fs_info
->unused_bgs_lock
);
9596 int btrfs_init_space_info(struct btrfs_fs_info
*fs_info
)
9598 struct btrfs_space_info
*space_info
;
9599 struct btrfs_super_block
*disk_super
;
9605 disk_super
= fs_info
->super_copy
;
9606 if (!btrfs_super_root(disk_super
))
9609 features
= btrfs_super_incompat_flags(disk_super
);
9610 if (features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
9613 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
9614 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
9619 flags
= BTRFS_BLOCK_GROUP_METADATA
| BTRFS_BLOCK_GROUP_DATA
;
9620 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
9622 flags
= BTRFS_BLOCK_GROUP_METADATA
;
9623 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
9627 flags
= BTRFS_BLOCK_GROUP_DATA
;
9628 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
9634 int btrfs_error_unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
9636 return unpin_extent_range(root
, start
, end
, false);
9639 int btrfs_trim_fs(struct btrfs_root
*root
, struct fstrim_range
*range
)
9641 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
9642 struct btrfs_block_group_cache
*cache
= NULL
;
9647 u64 total_bytes
= btrfs_super_total_bytes(fs_info
->super_copy
);
9651 * try to trim all FS space, our block group may start from non-zero.
9653 if (range
->len
== total_bytes
)
9654 cache
= btrfs_lookup_first_block_group(fs_info
, range
->start
);
9656 cache
= btrfs_lookup_block_group(fs_info
, range
->start
);
9659 if (cache
->key
.objectid
>= (range
->start
+ range
->len
)) {
9660 btrfs_put_block_group(cache
);
9664 start
= max(range
->start
, cache
->key
.objectid
);
9665 end
= min(range
->start
+ range
->len
,
9666 cache
->key
.objectid
+ cache
->key
.offset
);
9668 if (end
- start
>= range
->minlen
) {
9669 if (!block_group_cache_done(cache
)) {
9670 ret
= cache_block_group(cache
, 0);
9672 btrfs_put_block_group(cache
);
9675 ret
= wait_block_group_cache_done(cache
);
9677 btrfs_put_block_group(cache
);
9681 ret
= btrfs_trim_block_group(cache
,
9687 trimmed
+= group_trimmed
;
9689 btrfs_put_block_group(cache
);
9694 cache
= next_block_group(fs_info
->tree_root
, cache
);
9697 range
->len
= trimmed
;
9702 * btrfs_{start,end}_write_no_snapshoting() are similar to
9703 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
9704 * data into the page cache through nocow before the subvolume is snapshoted,
9705 * but flush the data into disk after the snapshot creation, or to prevent
9706 * operations while snapshoting is ongoing and that cause the snapshot to be
9707 * inconsistent (writes followed by expanding truncates for example).
9709 void btrfs_end_write_no_snapshoting(struct btrfs_root
*root
)
9711 percpu_counter_dec(&root
->subv_writers
->counter
);
9713 * Make sure counter is updated before we wake up
9717 if (waitqueue_active(&root
->subv_writers
->wait
))
9718 wake_up(&root
->subv_writers
->wait
);
9721 int btrfs_start_write_no_snapshoting(struct btrfs_root
*root
)
9723 if (atomic_read(&root
->will_be_snapshoted
))
9726 percpu_counter_inc(&root
->subv_writers
->counter
);
9728 * Make sure counter is updated before we check for snapshot creation.
9731 if (atomic_read(&root
->will_be_snapshoted
)) {
9732 btrfs_end_write_no_snapshoting(root
);