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>
32 #include "print-tree.h"
33 #include "transaction.h"
37 #include "free-space-cache.h"
40 #undef SCRAMBLE_DELAYED_REFS
43 * control flags for do_chunk_alloc's force field
44 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
45 * if we really need one.
47 * CHUNK_ALLOC_LIMITED means to only try and allocate one
48 * if we have very few chunks already allocated. This is
49 * used as part of the clustering code to help make sure
50 * we have a good pool of storage to cluster in, without
51 * filling the FS with empty chunks
53 * CHUNK_ALLOC_FORCE means it must try to allocate one
57 CHUNK_ALLOC_NO_FORCE
= 0,
58 CHUNK_ALLOC_LIMITED
= 1,
59 CHUNK_ALLOC_FORCE
= 2,
63 * Control how reservations are dealt with.
65 * RESERVE_FREE - freeing a reservation.
66 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
68 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
69 * bytes_may_use as the ENOSPC accounting is done elsewhere
74 RESERVE_ALLOC_NO_ACCOUNT
= 2,
77 static int update_block_group(struct btrfs_root
*root
,
78 u64 bytenr
, u64 num_bytes
, int alloc
);
79 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
80 struct btrfs_root
*root
,
81 u64 bytenr
, u64 num_bytes
, u64 parent
,
82 u64 root_objectid
, u64 owner_objectid
,
83 u64 owner_offset
, int refs_to_drop
,
84 struct btrfs_delayed_extent_op
*extra_op
);
85 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
86 struct extent_buffer
*leaf
,
87 struct btrfs_extent_item
*ei
);
88 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
89 struct btrfs_root
*root
,
90 u64 parent
, u64 root_objectid
,
91 u64 flags
, u64 owner
, u64 offset
,
92 struct btrfs_key
*ins
, int ref_mod
);
93 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
94 struct btrfs_root
*root
,
95 u64 parent
, u64 root_objectid
,
96 u64 flags
, struct btrfs_disk_key
*key
,
97 int level
, struct btrfs_key
*ins
);
98 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
99 struct btrfs_root
*extent_root
, u64 flags
,
101 static int find_next_key(struct btrfs_path
*path
, int level
,
102 struct btrfs_key
*key
);
103 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
104 int dump_block_groups
);
105 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
106 u64 num_bytes
, int reserve
);
107 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
109 int btrfs_pin_extent(struct btrfs_root
*root
,
110 u64 bytenr
, u64 num_bytes
, int reserved
);
113 block_group_cache_done(struct btrfs_block_group_cache
*cache
)
116 return cache
->cached
== BTRFS_CACHE_FINISHED
||
117 cache
->cached
== BTRFS_CACHE_ERROR
;
120 static int block_group_bits(struct btrfs_block_group_cache
*cache
, u64 bits
)
122 return (cache
->flags
& bits
) == bits
;
125 static void btrfs_get_block_group(struct btrfs_block_group_cache
*cache
)
127 atomic_inc(&cache
->count
);
130 void btrfs_put_block_group(struct btrfs_block_group_cache
*cache
)
132 if (atomic_dec_and_test(&cache
->count
)) {
133 WARN_ON(cache
->pinned
> 0);
134 WARN_ON(cache
->reserved
> 0);
135 kfree(cache
->free_space_ctl
);
141 * this adds the block group to the fs_info rb tree for the block group
144 static int btrfs_add_block_group_cache(struct btrfs_fs_info
*info
,
145 struct btrfs_block_group_cache
*block_group
)
148 struct rb_node
*parent
= NULL
;
149 struct btrfs_block_group_cache
*cache
;
151 spin_lock(&info
->block_group_cache_lock
);
152 p
= &info
->block_group_cache_tree
.rb_node
;
156 cache
= rb_entry(parent
, struct btrfs_block_group_cache
,
158 if (block_group
->key
.objectid
< cache
->key
.objectid
) {
160 } else if (block_group
->key
.objectid
> cache
->key
.objectid
) {
163 spin_unlock(&info
->block_group_cache_lock
);
168 rb_link_node(&block_group
->cache_node
, parent
, p
);
169 rb_insert_color(&block_group
->cache_node
,
170 &info
->block_group_cache_tree
);
172 if (info
->first_logical_byte
> block_group
->key
.objectid
)
173 info
->first_logical_byte
= block_group
->key
.objectid
;
175 spin_unlock(&info
->block_group_cache_lock
);
181 * This will return the block group at or after bytenr if contains is 0, else
182 * it will return the block group that contains the bytenr
184 static struct btrfs_block_group_cache
*
185 block_group_cache_tree_search(struct btrfs_fs_info
*info
, u64 bytenr
,
188 struct btrfs_block_group_cache
*cache
, *ret
= NULL
;
192 spin_lock(&info
->block_group_cache_lock
);
193 n
= info
->block_group_cache_tree
.rb_node
;
196 cache
= rb_entry(n
, struct btrfs_block_group_cache
,
198 end
= cache
->key
.objectid
+ cache
->key
.offset
- 1;
199 start
= cache
->key
.objectid
;
201 if (bytenr
< start
) {
202 if (!contains
&& (!ret
|| start
< ret
->key
.objectid
))
205 } else if (bytenr
> start
) {
206 if (contains
&& bytenr
<= end
) {
217 btrfs_get_block_group(ret
);
218 if (bytenr
== 0 && info
->first_logical_byte
> ret
->key
.objectid
)
219 info
->first_logical_byte
= ret
->key
.objectid
;
221 spin_unlock(&info
->block_group_cache_lock
);
226 static int add_excluded_extent(struct btrfs_root
*root
,
227 u64 start
, u64 num_bytes
)
229 u64 end
= start
+ num_bytes
- 1;
230 set_extent_bits(&root
->fs_info
->freed_extents
[0],
231 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
232 set_extent_bits(&root
->fs_info
->freed_extents
[1],
233 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
237 static void free_excluded_extents(struct btrfs_root
*root
,
238 struct btrfs_block_group_cache
*cache
)
242 start
= cache
->key
.objectid
;
243 end
= start
+ cache
->key
.offset
- 1;
245 clear_extent_bits(&root
->fs_info
->freed_extents
[0],
246 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
247 clear_extent_bits(&root
->fs_info
->freed_extents
[1],
248 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
251 static int exclude_super_stripes(struct btrfs_root
*root
,
252 struct btrfs_block_group_cache
*cache
)
259 if (cache
->key
.objectid
< BTRFS_SUPER_INFO_OFFSET
) {
260 stripe_len
= BTRFS_SUPER_INFO_OFFSET
- cache
->key
.objectid
;
261 cache
->bytes_super
+= stripe_len
;
262 ret
= add_excluded_extent(root
, cache
->key
.objectid
,
268 for (i
= 0; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
269 bytenr
= btrfs_sb_offset(i
);
270 ret
= btrfs_rmap_block(&root
->fs_info
->mapping_tree
,
271 cache
->key
.objectid
, bytenr
,
272 0, &logical
, &nr
, &stripe_len
);
279 if (logical
[nr
] > cache
->key
.objectid
+
283 if (logical
[nr
] + stripe_len
<= cache
->key
.objectid
)
287 if (start
< cache
->key
.objectid
) {
288 start
= cache
->key
.objectid
;
289 len
= (logical
[nr
] + stripe_len
) - start
;
291 len
= min_t(u64
, stripe_len
,
292 cache
->key
.objectid
+
293 cache
->key
.offset
- start
);
296 cache
->bytes_super
+= len
;
297 ret
= add_excluded_extent(root
, start
, len
);
309 static struct btrfs_caching_control
*
310 get_caching_control(struct btrfs_block_group_cache
*cache
)
312 struct btrfs_caching_control
*ctl
;
314 spin_lock(&cache
->lock
);
315 if (cache
->cached
!= BTRFS_CACHE_STARTED
) {
316 spin_unlock(&cache
->lock
);
320 /* We're loading it the fast way, so we don't have a caching_ctl. */
321 if (!cache
->caching_ctl
) {
322 spin_unlock(&cache
->lock
);
326 ctl
= cache
->caching_ctl
;
327 atomic_inc(&ctl
->count
);
328 spin_unlock(&cache
->lock
);
332 static void put_caching_control(struct btrfs_caching_control
*ctl
)
334 if (atomic_dec_and_test(&ctl
->count
))
339 * this is only called by cache_block_group, since we could have freed extents
340 * we need to check the pinned_extents for any extents that can't be used yet
341 * since their free space will be released as soon as the transaction commits.
343 static u64
add_new_free_space(struct btrfs_block_group_cache
*block_group
,
344 struct btrfs_fs_info
*info
, u64 start
, u64 end
)
346 u64 extent_start
, extent_end
, size
, total_added
= 0;
349 while (start
< end
) {
350 ret
= find_first_extent_bit(info
->pinned_extents
, start
,
351 &extent_start
, &extent_end
,
352 EXTENT_DIRTY
| EXTENT_UPTODATE
,
357 if (extent_start
<= start
) {
358 start
= extent_end
+ 1;
359 } else if (extent_start
> start
&& extent_start
< end
) {
360 size
= extent_start
- start
;
362 ret
= btrfs_add_free_space(block_group
, start
,
364 BUG_ON(ret
); /* -ENOMEM or logic error */
365 start
= extent_end
+ 1;
374 ret
= btrfs_add_free_space(block_group
, start
, size
);
375 BUG_ON(ret
); /* -ENOMEM or logic error */
381 static noinline
void caching_thread(struct btrfs_work
*work
)
383 struct btrfs_block_group_cache
*block_group
;
384 struct btrfs_fs_info
*fs_info
;
385 struct btrfs_caching_control
*caching_ctl
;
386 struct btrfs_root
*extent_root
;
387 struct btrfs_path
*path
;
388 struct extent_buffer
*leaf
;
389 struct btrfs_key key
;
395 caching_ctl
= container_of(work
, struct btrfs_caching_control
, work
);
396 block_group
= caching_ctl
->block_group
;
397 fs_info
= block_group
->fs_info
;
398 extent_root
= fs_info
->extent_root
;
400 path
= btrfs_alloc_path();
404 last
= max_t(u64
, block_group
->key
.objectid
, BTRFS_SUPER_INFO_OFFSET
);
407 * We don't want to deadlock with somebody trying to allocate a new
408 * extent for the extent root while also trying to search the extent
409 * root to add free space. So we skip locking and search the commit
410 * root, since its read-only
412 path
->skip_locking
= 1;
413 path
->search_commit_root
= 1;
418 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
420 mutex_lock(&caching_ctl
->mutex
);
421 /* need to make sure the commit_root doesn't disappear */
422 down_read(&fs_info
->extent_commit_sem
);
425 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
429 leaf
= path
->nodes
[0];
430 nritems
= btrfs_header_nritems(leaf
);
433 if (btrfs_fs_closing(fs_info
) > 1) {
438 if (path
->slots
[0] < nritems
) {
439 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
441 ret
= find_next_key(path
, 0, &key
);
445 if (need_resched()) {
446 caching_ctl
->progress
= last
;
447 btrfs_release_path(path
);
448 up_read(&fs_info
->extent_commit_sem
);
449 mutex_unlock(&caching_ctl
->mutex
);
454 ret
= btrfs_next_leaf(extent_root
, path
);
459 leaf
= path
->nodes
[0];
460 nritems
= btrfs_header_nritems(leaf
);
464 if (key
.objectid
< last
) {
467 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
469 caching_ctl
->progress
= last
;
470 btrfs_release_path(path
);
474 if (key
.objectid
< block_group
->key
.objectid
) {
479 if (key
.objectid
>= block_group
->key
.objectid
+
480 block_group
->key
.offset
)
483 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
||
484 key
.type
== BTRFS_METADATA_ITEM_KEY
) {
485 total_found
+= add_new_free_space(block_group
,
488 if (key
.type
== BTRFS_METADATA_ITEM_KEY
)
489 last
= key
.objectid
+
490 fs_info
->tree_root
->leafsize
;
492 last
= key
.objectid
+ key
.offset
;
494 if (total_found
> (1024 * 1024 * 2)) {
496 wake_up(&caching_ctl
->wait
);
503 total_found
+= add_new_free_space(block_group
, fs_info
, last
,
504 block_group
->key
.objectid
+
505 block_group
->key
.offset
);
506 caching_ctl
->progress
= (u64
)-1;
508 spin_lock(&block_group
->lock
);
509 block_group
->caching_ctl
= NULL
;
510 block_group
->cached
= BTRFS_CACHE_FINISHED
;
511 spin_unlock(&block_group
->lock
);
514 btrfs_free_path(path
);
515 up_read(&fs_info
->extent_commit_sem
);
517 free_excluded_extents(extent_root
, block_group
);
519 mutex_unlock(&caching_ctl
->mutex
);
522 spin_lock(&block_group
->lock
);
523 block_group
->caching_ctl
= NULL
;
524 block_group
->cached
= BTRFS_CACHE_ERROR
;
525 spin_unlock(&block_group
->lock
);
527 wake_up(&caching_ctl
->wait
);
529 put_caching_control(caching_ctl
);
530 btrfs_put_block_group(block_group
);
533 static int cache_block_group(struct btrfs_block_group_cache
*cache
,
537 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
538 struct btrfs_caching_control
*caching_ctl
;
541 caching_ctl
= kzalloc(sizeof(*caching_ctl
), GFP_NOFS
);
545 INIT_LIST_HEAD(&caching_ctl
->list
);
546 mutex_init(&caching_ctl
->mutex
);
547 init_waitqueue_head(&caching_ctl
->wait
);
548 caching_ctl
->block_group
= cache
;
549 caching_ctl
->progress
= cache
->key
.objectid
;
550 atomic_set(&caching_ctl
->count
, 1);
551 caching_ctl
->work
.func
= caching_thread
;
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 ret
= load_free_space_cache(fs_info
, cache
);
594 spin_lock(&cache
->lock
);
596 cache
->caching_ctl
= NULL
;
597 cache
->cached
= BTRFS_CACHE_FINISHED
;
598 cache
->last_byte_to_unpin
= (u64
)-1;
600 if (load_cache_only
) {
601 cache
->caching_ctl
= NULL
;
602 cache
->cached
= BTRFS_CACHE_NO
;
604 cache
->cached
= BTRFS_CACHE_STARTED
;
607 spin_unlock(&cache
->lock
);
608 wake_up(&caching_ctl
->wait
);
610 put_caching_control(caching_ctl
);
611 free_excluded_extents(fs_info
->extent_root
, cache
);
616 * We are not going to do the fast caching, set cached to the
617 * appropriate value and wakeup any waiters.
619 spin_lock(&cache
->lock
);
620 if (load_cache_only
) {
621 cache
->caching_ctl
= NULL
;
622 cache
->cached
= BTRFS_CACHE_NO
;
624 cache
->cached
= BTRFS_CACHE_STARTED
;
626 spin_unlock(&cache
->lock
);
627 wake_up(&caching_ctl
->wait
);
630 if (load_cache_only
) {
631 put_caching_control(caching_ctl
);
635 down_write(&fs_info
->extent_commit_sem
);
636 atomic_inc(&caching_ctl
->count
);
637 list_add_tail(&caching_ctl
->list
, &fs_info
->caching_block_groups
);
638 up_write(&fs_info
->extent_commit_sem
);
640 btrfs_get_block_group(cache
);
642 btrfs_queue_worker(&fs_info
->caching_workers
, &caching_ctl
->work
);
648 * return the block group that starts at or after bytenr
650 static struct btrfs_block_group_cache
*
651 btrfs_lookup_first_block_group(struct btrfs_fs_info
*info
, u64 bytenr
)
653 struct btrfs_block_group_cache
*cache
;
655 cache
= block_group_cache_tree_search(info
, bytenr
, 0);
661 * return the block group that contains the given bytenr
663 struct btrfs_block_group_cache
*btrfs_lookup_block_group(
664 struct btrfs_fs_info
*info
,
667 struct btrfs_block_group_cache
*cache
;
669 cache
= block_group_cache_tree_search(info
, bytenr
, 1);
674 static struct btrfs_space_info
*__find_space_info(struct btrfs_fs_info
*info
,
677 struct list_head
*head
= &info
->space_info
;
678 struct btrfs_space_info
*found
;
680 flags
&= BTRFS_BLOCK_GROUP_TYPE_MASK
;
683 list_for_each_entry_rcu(found
, head
, list
) {
684 if (found
->flags
& flags
) {
694 * after adding space to the filesystem, we need to clear the full flags
695 * on all the space infos.
697 void btrfs_clear_space_info_full(struct btrfs_fs_info
*info
)
699 struct list_head
*head
= &info
->space_info
;
700 struct btrfs_space_info
*found
;
703 list_for_each_entry_rcu(found
, head
, list
)
708 /* simple helper to search for an existing extent at a given offset */
709 int btrfs_lookup_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
712 struct btrfs_key key
;
713 struct btrfs_path
*path
;
715 path
= btrfs_alloc_path();
719 key
.objectid
= start
;
721 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
722 ret
= btrfs_search_slot(NULL
, root
->fs_info
->extent_root
, &key
, path
,
725 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
726 if (key
.objectid
== start
&&
727 key
.type
== BTRFS_METADATA_ITEM_KEY
)
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
->leafsize
;
767 path
= btrfs_alloc_path();
772 key
.objectid
= bytenr
;
773 key
.type
= BTRFS_METADATA_ITEM_KEY
;
776 key
.objectid
= bytenr
;
777 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
782 path
->skip_locking
= 1;
783 path
->search_commit_root
= 1;
786 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
,
791 if (ret
> 0 && metadata
&& key
.type
== BTRFS_METADATA_ITEM_KEY
) {
793 if (path
->slots
[0]) {
795 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
797 if (key
.objectid
== bytenr
&&
798 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
799 key
.offset
== root
->leafsize
)
803 key
.objectid
= bytenr
;
804 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
805 key
.offset
= root
->leafsize
;
806 btrfs_release_path(path
);
812 leaf
= path
->nodes
[0];
813 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
814 if (item_size
>= sizeof(*ei
)) {
815 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
816 struct btrfs_extent_item
);
817 num_refs
= btrfs_extent_refs(leaf
, ei
);
818 extent_flags
= btrfs_extent_flags(leaf
, ei
);
820 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
821 struct btrfs_extent_item_v0
*ei0
;
822 BUG_ON(item_size
!= sizeof(*ei0
));
823 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
824 struct btrfs_extent_item_v0
);
825 num_refs
= btrfs_extent_refs_v0(leaf
, ei0
);
826 /* FIXME: this isn't correct for data */
827 extent_flags
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
832 BUG_ON(num_refs
== 0);
842 delayed_refs
= &trans
->transaction
->delayed_refs
;
843 spin_lock(&delayed_refs
->lock
);
844 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
846 if (!mutex_trylock(&head
->mutex
)) {
847 atomic_inc(&head
->node
.refs
);
848 spin_unlock(&delayed_refs
->lock
);
850 btrfs_release_path(path
);
853 * Mutex was contended, block until it's released and try
856 mutex_lock(&head
->mutex
);
857 mutex_unlock(&head
->mutex
);
858 btrfs_put_delayed_ref(&head
->node
);
861 if (head
->extent_op
&& head
->extent_op
->update_flags
)
862 extent_flags
|= head
->extent_op
->flags_to_set
;
864 BUG_ON(num_refs
== 0);
866 num_refs
+= head
->node
.ref_mod
;
867 mutex_unlock(&head
->mutex
);
869 spin_unlock(&delayed_refs
->lock
);
871 WARN_ON(num_refs
== 0);
875 *flags
= extent_flags
;
877 btrfs_free_path(path
);
882 * Back reference rules. Back refs have three main goals:
884 * 1) differentiate between all holders of references to an extent so that
885 * when a reference is dropped we can make sure it was a valid reference
886 * before freeing the extent.
888 * 2) Provide enough information to quickly find the holders of an extent
889 * if we notice a given block is corrupted or bad.
891 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
892 * maintenance. This is actually the same as #2, but with a slightly
893 * different use case.
895 * There are two kinds of back refs. The implicit back refs is optimized
896 * for pointers in non-shared tree blocks. For a given pointer in a block,
897 * back refs of this kind provide information about the block's owner tree
898 * and the pointer's key. These information allow us to find the block by
899 * b-tree searching. The full back refs is for pointers in tree blocks not
900 * referenced by their owner trees. The location of tree block is recorded
901 * in the back refs. Actually the full back refs is generic, and can be
902 * used in all cases the implicit back refs is used. The major shortcoming
903 * of the full back refs is its overhead. Every time a tree block gets
904 * COWed, we have to update back refs entry for all pointers in it.
906 * For a newly allocated tree block, we use implicit back refs for
907 * pointers in it. This means most tree related operations only involve
908 * implicit back refs. For a tree block created in old transaction, the
909 * only way to drop a reference to it is COW it. So we can detect the
910 * event that tree block loses its owner tree's reference and do the
911 * back refs conversion.
913 * When a tree block is COW'd through a tree, there are four cases:
915 * The reference count of the block is one and the tree is the block's
916 * owner tree. Nothing to do in this case.
918 * The reference count of the block is one and the tree is not the
919 * block's owner tree. In this case, full back refs is used for pointers
920 * in the block. Remove these full back refs, add implicit back refs for
921 * every pointers in the new block.
923 * The reference count of the block is greater than one and the tree is
924 * the block's owner tree. In this case, implicit back refs is used for
925 * pointers in the block. Add full back refs for every pointers in the
926 * block, increase lower level extents' reference counts. The original
927 * implicit back refs are entailed to the new block.
929 * The reference count of the block is greater than one and the tree is
930 * not the block's owner tree. Add implicit back refs for every pointer in
931 * the new block, increase lower level extents' reference count.
933 * Back Reference Key composing:
935 * The key objectid corresponds to the first byte in the extent,
936 * The key type is used to differentiate between types of back refs.
937 * There are different meanings of the key offset for different types
940 * File extents can be referenced by:
942 * - multiple snapshots, subvolumes, or different generations in one subvol
943 * - different files inside a single subvolume
944 * - different offsets inside a file (bookend extents in file.c)
946 * The extent ref structure for the implicit back refs has fields for:
948 * - Objectid of the subvolume root
949 * - objectid of the file holding the reference
950 * - original offset in the file
951 * - how many bookend extents
953 * The key offset for the implicit back refs is hash of the first
956 * The extent ref structure for the full back refs has field for:
958 * - number of pointers in the tree leaf
960 * The key offset for the implicit back refs is the first byte of
963 * When a file extent is allocated, The implicit back refs is used.
964 * the fields are filled in:
966 * (root_key.objectid, inode objectid, offset in file, 1)
968 * When a file extent is removed file truncation, we find the
969 * corresponding implicit back refs and check the following fields:
971 * (btrfs_header_owner(leaf), inode objectid, offset in file)
973 * Btree extents can be referenced by:
975 * - Different subvolumes
977 * Both the implicit back refs and the full back refs for tree blocks
978 * only consist of key. The key offset for the implicit back refs is
979 * objectid of block's owner tree. The key offset for the full back refs
980 * is the first byte of parent block.
982 * When implicit back refs is used, information about the lowest key and
983 * level of the tree block are required. These information are stored in
984 * tree block info structure.
987 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
988 static int convert_extent_item_v0(struct btrfs_trans_handle
*trans
,
989 struct btrfs_root
*root
,
990 struct btrfs_path
*path
,
991 u64 owner
, u32 extra_size
)
993 struct btrfs_extent_item
*item
;
994 struct btrfs_extent_item_v0
*ei0
;
995 struct btrfs_extent_ref_v0
*ref0
;
996 struct btrfs_tree_block_info
*bi
;
997 struct extent_buffer
*leaf
;
998 struct btrfs_key key
;
999 struct btrfs_key found_key
;
1000 u32 new_size
= sizeof(*item
);
1004 leaf
= path
->nodes
[0];
1005 BUG_ON(btrfs_item_size_nr(leaf
, path
->slots
[0]) != sizeof(*ei0
));
1007 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1008 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1009 struct btrfs_extent_item_v0
);
1010 refs
= btrfs_extent_refs_v0(leaf
, ei0
);
1012 if (owner
== (u64
)-1) {
1014 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1015 ret
= btrfs_next_leaf(root
, path
);
1018 BUG_ON(ret
> 0); /* Corruption */
1019 leaf
= path
->nodes
[0];
1021 btrfs_item_key_to_cpu(leaf
, &found_key
,
1023 BUG_ON(key
.objectid
!= found_key
.objectid
);
1024 if (found_key
.type
!= BTRFS_EXTENT_REF_V0_KEY
) {
1028 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1029 struct btrfs_extent_ref_v0
);
1030 owner
= btrfs_ref_objectid_v0(leaf
, ref0
);
1034 btrfs_release_path(path
);
1036 if (owner
< BTRFS_FIRST_FREE_OBJECTID
)
1037 new_size
+= sizeof(*bi
);
1039 new_size
-= sizeof(*ei0
);
1040 ret
= btrfs_search_slot(trans
, root
, &key
, path
,
1041 new_size
+ extra_size
, 1);
1044 BUG_ON(ret
); /* Corruption */
1046 btrfs_extend_item(root
, path
, new_size
);
1048 leaf
= path
->nodes
[0];
1049 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1050 btrfs_set_extent_refs(leaf
, item
, refs
);
1051 /* FIXME: get real generation */
1052 btrfs_set_extent_generation(leaf
, item
, 0);
1053 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1054 btrfs_set_extent_flags(leaf
, item
,
1055 BTRFS_EXTENT_FLAG_TREE_BLOCK
|
1056 BTRFS_BLOCK_FLAG_FULL_BACKREF
);
1057 bi
= (struct btrfs_tree_block_info
*)(item
+ 1);
1058 /* FIXME: get first key of the block */
1059 memset_extent_buffer(leaf
, 0, (unsigned long)bi
, sizeof(*bi
));
1060 btrfs_set_tree_block_level(leaf
, bi
, (int)owner
);
1062 btrfs_set_extent_flags(leaf
, item
, BTRFS_EXTENT_FLAG_DATA
);
1064 btrfs_mark_buffer_dirty(leaf
);
1069 static u64
hash_extent_data_ref(u64 root_objectid
, u64 owner
, u64 offset
)
1071 u32 high_crc
= ~(u32
)0;
1072 u32 low_crc
= ~(u32
)0;
1075 lenum
= cpu_to_le64(root_objectid
);
1076 high_crc
= crc32c(high_crc
, &lenum
, sizeof(lenum
));
1077 lenum
= cpu_to_le64(owner
);
1078 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
1079 lenum
= cpu_to_le64(offset
);
1080 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
1082 return ((u64
)high_crc
<< 31) ^ (u64
)low_crc
;
1085 static u64
hash_extent_data_ref_item(struct extent_buffer
*leaf
,
1086 struct btrfs_extent_data_ref
*ref
)
1088 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf
, ref
),
1089 btrfs_extent_data_ref_objectid(leaf
, ref
),
1090 btrfs_extent_data_ref_offset(leaf
, ref
));
1093 static int match_extent_data_ref(struct extent_buffer
*leaf
,
1094 struct btrfs_extent_data_ref
*ref
,
1095 u64 root_objectid
, u64 owner
, u64 offset
)
1097 if (btrfs_extent_data_ref_root(leaf
, ref
) != root_objectid
||
1098 btrfs_extent_data_ref_objectid(leaf
, ref
) != owner
||
1099 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
1104 static noinline
int lookup_extent_data_ref(struct btrfs_trans_handle
*trans
,
1105 struct btrfs_root
*root
,
1106 struct btrfs_path
*path
,
1107 u64 bytenr
, u64 parent
,
1109 u64 owner
, u64 offset
)
1111 struct btrfs_key key
;
1112 struct btrfs_extent_data_ref
*ref
;
1113 struct extent_buffer
*leaf
;
1119 key
.objectid
= bytenr
;
1121 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1122 key
.offset
= parent
;
1124 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1125 key
.offset
= hash_extent_data_ref(root_objectid
,
1130 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1139 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1140 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1141 btrfs_release_path(path
);
1142 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1153 leaf
= path
->nodes
[0];
1154 nritems
= btrfs_header_nritems(leaf
);
1156 if (path
->slots
[0] >= nritems
) {
1157 ret
= btrfs_next_leaf(root
, path
);
1163 leaf
= path
->nodes
[0];
1164 nritems
= btrfs_header_nritems(leaf
);
1168 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1169 if (key
.objectid
!= bytenr
||
1170 key
.type
!= BTRFS_EXTENT_DATA_REF_KEY
)
1173 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1174 struct btrfs_extent_data_ref
);
1176 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1179 btrfs_release_path(path
);
1191 static noinline
int insert_extent_data_ref(struct btrfs_trans_handle
*trans
,
1192 struct btrfs_root
*root
,
1193 struct btrfs_path
*path
,
1194 u64 bytenr
, u64 parent
,
1195 u64 root_objectid
, u64 owner
,
1196 u64 offset
, int refs_to_add
)
1198 struct btrfs_key key
;
1199 struct extent_buffer
*leaf
;
1204 key
.objectid
= bytenr
;
1206 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1207 key
.offset
= parent
;
1208 size
= sizeof(struct btrfs_shared_data_ref
);
1210 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1211 key
.offset
= hash_extent_data_ref(root_objectid
,
1213 size
= sizeof(struct btrfs_extent_data_ref
);
1216 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, size
);
1217 if (ret
&& ret
!= -EEXIST
)
1220 leaf
= path
->nodes
[0];
1222 struct btrfs_shared_data_ref
*ref
;
1223 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1224 struct btrfs_shared_data_ref
);
1226 btrfs_set_shared_data_ref_count(leaf
, ref
, refs_to_add
);
1228 num_refs
= btrfs_shared_data_ref_count(leaf
, ref
);
1229 num_refs
+= refs_to_add
;
1230 btrfs_set_shared_data_ref_count(leaf
, ref
, num_refs
);
1233 struct btrfs_extent_data_ref
*ref
;
1234 while (ret
== -EEXIST
) {
1235 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1236 struct btrfs_extent_data_ref
);
1237 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1240 btrfs_release_path(path
);
1242 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1244 if (ret
&& ret
!= -EEXIST
)
1247 leaf
= path
->nodes
[0];
1249 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1250 struct btrfs_extent_data_ref
);
1252 btrfs_set_extent_data_ref_root(leaf
, ref
,
1254 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
1255 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
1256 btrfs_set_extent_data_ref_count(leaf
, ref
, refs_to_add
);
1258 num_refs
= btrfs_extent_data_ref_count(leaf
, ref
);
1259 num_refs
+= refs_to_add
;
1260 btrfs_set_extent_data_ref_count(leaf
, ref
, num_refs
);
1263 btrfs_mark_buffer_dirty(leaf
);
1266 btrfs_release_path(path
);
1270 static noinline
int remove_extent_data_ref(struct btrfs_trans_handle
*trans
,
1271 struct btrfs_root
*root
,
1272 struct btrfs_path
*path
,
1275 struct btrfs_key key
;
1276 struct btrfs_extent_data_ref
*ref1
= NULL
;
1277 struct btrfs_shared_data_ref
*ref2
= NULL
;
1278 struct extent_buffer
*leaf
;
1282 leaf
= path
->nodes
[0];
1283 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1285 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1286 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1287 struct btrfs_extent_data_ref
);
1288 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1289 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1290 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1291 struct btrfs_shared_data_ref
);
1292 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1293 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1294 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1295 struct btrfs_extent_ref_v0
*ref0
;
1296 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1297 struct btrfs_extent_ref_v0
);
1298 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1304 BUG_ON(num_refs
< refs_to_drop
);
1305 num_refs
-= refs_to_drop
;
1307 if (num_refs
== 0) {
1308 ret
= btrfs_del_item(trans
, root
, path
);
1310 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
)
1311 btrfs_set_extent_data_ref_count(leaf
, ref1
, num_refs
);
1312 else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
)
1313 btrfs_set_shared_data_ref_count(leaf
, ref2
, num_refs
);
1314 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1316 struct btrfs_extent_ref_v0
*ref0
;
1317 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1318 struct btrfs_extent_ref_v0
);
1319 btrfs_set_ref_count_v0(leaf
, ref0
, num_refs
);
1322 btrfs_mark_buffer_dirty(leaf
);
1327 static noinline u32
extent_data_ref_count(struct btrfs_root
*root
,
1328 struct btrfs_path
*path
,
1329 struct btrfs_extent_inline_ref
*iref
)
1331 struct btrfs_key key
;
1332 struct extent_buffer
*leaf
;
1333 struct btrfs_extent_data_ref
*ref1
;
1334 struct btrfs_shared_data_ref
*ref2
;
1337 leaf
= path
->nodes
[0];
1338 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1340 if (btrfs_extent_inline_ref_type(leaf
, iref
) ==
1341 BTRFS_EXTENT_DATA_REF_KEY
) {
1342 ref1
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1343 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1345 ref2
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1346 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1348 } else if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1349 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1350 struct btrfs_extent_data_ref
);
1351 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1352 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1353 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1354 struct btrfs_shared_data_ref
);
1355 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1356 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1357 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1358 struct btrfs_extent_ref_v0
*ref0
;
1359 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1360 struct btrfs_extent_ref_v0
);
1361 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1369 static noinline
int lookup_tree_block_ref(struct btrfs_trans_handle
*trans
,
1370 struct btrfs_root
*root
,
1371 struct btrfs_path
*path
,
1372 u64 bytenr
, u64 parent
,
1375 struct btrfs_key key
;
1378 key
.objectid
= bytenr
;
1380 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1381 key
.offset
= parent
;
1383 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1384 key
.offset
= root_objectid
;
1387 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1390 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1391 if (ret
== -ENOENT
&& parent
) {
1392 btrfs_release_path(path
);
1393 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1394 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1402 static noinline
int insert_tree_block_ref(struct btrfs_trans_handle
*trans
,
1403 struct btrfs_root
*root
,
1404 struct btrfs_path
*path
,
1405 u64 bytenr
, u64 parent
,
1408 struct btrfs_key key
;
1411 key
.objectid
= bytenr
;
1413 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1414 key
.offset
= parent
;
1416 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1417 key
.offset
= root_objectid
;
1420 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
1421 btrfs_release_path(path
);
1425 static inline int extent_ref_type(u64 parent
, u64 owner
)
1428 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1430 type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1432 type
= BTRFS_TREE_BLOCK_REF_KEY
;
1435 type
= BTRFS_SHARED_DATA_REF_KEY
;
1437 type
= BTRFS_EXTENT_DATA_REF_KEY
;
1442 static int find_next_key(struct btrfs_path
*path
, int level
,
1443 struct btrfs_key
*key
)
1446 for (; level
< BTRFS_MAX_LEVEL
; level
++) {
1447 if (!path
->nodes
[level
])
1449 if (path
->slots
[level
] + 1 >=
1450 btrfs_header_nritems(path
->nodes
[level
]))
1453 btrfs_item_key_to_cpu(path
->nodes
[level
], key
,
1454 path
->slots
[level
] + 1);
1456 btrfs_node_key_to_cpu(path
->nodes
[level
], key
,
1457 path
->slots
[level
] + 1);
1464 * look for inline back ref. if back ref is found, *ref_ret is set
1465 * to the address of inline back ref, and 0 is returned.
1467 * if back ref isn't found, *ref_ret is set to the address where it
1468 * should be inserted, and -ENOENT is returned.
1470 * if insert is true and there are too many inline back refs, the path
1471 * points to the extent item, and -EAGAIN is returned.
1473 * NOTE: inline back refs are ordered in the same way that back ref
1474 * items in the tree are ordered.
1476 static noinline_for_stack
1477 int lookup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1478 struct btrfs_root
*root
,
1479 struct btrfs_path
*path
,
1480 struct btrfs_extent_inline_ref
**ref_ret
,
1481 u64 bytenr
, u64 num_bytes
,
1482 u64 parent
, u64 root_objectid
,
1483 u64 owner
, u64 offset
, int insert
)
1485 struct btrfs_key key
;
1486 struct extent_buffer
*leaf
;
1487 struct btrfs_extent_item
*ei
;
1488 struct btrfs_extent_inline_ref
*iref
;
1498 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
1501 key
.objectid
= bytenr
;
1502 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1503 key
.offset
= num_bytes
;
1505 want
= extent_ref_type(parent
, owner
);
1507 extra_size
= btrfs_extent_inline_ref_size(want
);
1508 path
->keep_locks
= 1;
1513 * Owner is our parent level, so we can just add one to get the level
1514 * for the block we are interested in.
1516 if (skinny_metadata
&& owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1517 key
.type
= BTRFS_METADATA_ITEM_KEY
;
1522 ret
= btrfs_search_slot(trans
, root
, &key
, path
, extra_size
, 1);
1529 * We may be a newly converted file system which still has the old fat
1530 * extent entries for metadata, so try and see if we have one of those.
1532 if (ret
> 0 && skinny_metadata
) {
1533 skinny_metadata
= false;
1534 if (path
->slots
[0]) {
1536 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
1538 if (key
.objectid
== bytenr
&&
1539 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
1540 key
.offset
== num_bytes
)
1544 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1545 key
.offset
= num_bytes
;
1546 btrfs_release_path(path
);
1551 if (ret
&& !insert
) {
1560 leaf
= path
->nodes
[0];
1561 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1562 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1563 if (item_size
< sizeof(*ei
)) {
1568 ret
= convert_extent_item_v0(trans
, root
, path
, owner
,
1574 leaf
= path
->nodes
[0];
1575 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1578 BUG_ON(item_size
< sizeof(*ei
));
1580 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1581 flags
= btrfs_extent_flags(leaf
, ei
);
1583 ptr
= (unsigned long)(ei
+ 1);
1584 end
= (unsigned long)ei
+ item_size
;
1586 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
&& !skinny_metadata
) {
1587 ptr
+= sizeof(struct btrfs_tree_block_info
);
1597 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1598 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1602 ptr
+= btrfs_extent_inline_ref_size(type
);
1606 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1607 struct btrfs_extent_data_ref
*dref
;
1608 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1609 if (match_extent_data_ref(leaf
, dref
, root_objectid
,
1614 if (hash_extent_data_ref_item(leaf
, dref
) <
1615 hash_extent_data_ref(root_objectid
, owner
, offset
))
1619 ref_offset
= btrfs_extent_inline_ref_offset(leaf
, iref
);
1621 if (parent
== ref_offset
) {
1625 if (ref_offset
< parent
)
1628 if (root_objectid
== ref_offset
) {
1632 if (ref_offset
< root_objectid
)
1636 ptr
+= btrfs_extent_inline_ref_size(type
);
1638 if (err
== -ENOENT
&& insert
) {
1639 if (item_size
+ extra_size
>=
1640 BTRFS_MAX_EXTENT_ITEM_SIZE(root
)) {
1645 * To add new inline back ref, we have to make sure
1646 * there is no corresponding back ref item.
1647 * For simplicity, we just do not add new inline back
1648 * ref if there is any kind of item for this block
1650 if (find_next_key(path
, 0, &key
) == 0 &&
1651 key
.objectid
== bytenr
&&
1652 key
.type
< BTRFS_BLOCK_GROUP_ITEM_KEY
) {
1657 *ref_ret
= (struct btrfs_extent_inline_ref
*)ptr
;
1660 path
->keep_locks
= 0;
1661 btrfs_unlock_up_safe(path
, 1);
1667 * helper to add new inline back ref
1669 static noinline_for_stack
1670 void setup_inline_extent_backref(struct btrfs_root
*root
,
1671 struct btrfs_path
*path
,
1672 struct btrfs_extent_inline_ref
*iref
,
1673 u64 parent
, u64 root_objectid
,
1674 u64 owner
, u64 offset
, int refs_to_add
,
1675 struct btrfs_delayed_extent_op
*extent_op
)
1677 struct extent_buffer
*leaf
;
1678 struct btrfs_extent_item
*ei
;
1681 unsigned long item_offset
;
1686 leaf
= path
->nodes
[0];
1687 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1688 item_offset
= (unsigned long)iref
- (unsigned long)ei
;
1690 type
= extent_ref_type(parent
, owner
);
1691 size
= btrfs_extent_inline_ref_size(type
);
1693 btrfs_extend_item(root
, path
, size
);
1695 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1696 refs
= btrfs_extent_refs(leaf
, ei
);
1697 refs
+= refs_to_add
;
1698 btrfs_set_extent_refs(leaf
, ei
, refs
);
1700 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1702 ptr
= (unsigned long)ei
+ item_offset
;
1703 end
= (unsigned long)ei
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
1704 if (ptr
< end
- size
)
1705 memmove_extent_buffer(leaf
, ptr
+ size
, ptr
,
1708 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1709 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
1710 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1711 struct btrfs_extent_data_ref
*dref
;
1712 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1713 btrfs_set_extent_data_ref_root(leaf
, dref
, root_objectid
);
1714 btrfs_set_extent_data_ref_objectid(leaf
, dref
, owner
);
1715 btrfs_set_extent_data_ref_offset(leaf
, dref
, offset
);
1716 btrfs_set_extent_data_ref_count(leaf
, dref
, refs_to_add
);
1717 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1718 struct btrfs_shared_data_ref
*sref
;
1719 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1720 btrfs_set_shared_data_ref_count(leaf
, sref
, refs_to_add
);
1721 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1722 } else if (type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
1723 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1725 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
1727 btrfs_mark_buffer_dirty(leaf
);
1730 static int lookup_extent_backref(struct btrfs_trans_handle
*trans
,
1731 struct btrfs_root
*root
,
1732 struct btrfs_path
*path
,
1733 struct btrfs_extent_inline_ref
**ref_ret
,
1734 u64 bytenr
, u64 num_bytes
, u64 parent
,
1735 u64 root_objectid
, u64 owner
, u64 offset
)
1739 ret
= lookup_inline_extent_backref(trans
, root
, path
, ref_ret
,
1740 bytenr
, num_bytes
, parent
,
1741 root_objectid
, owner
, offset
, 0);
1745 btrfs_release_path(path
);
1748 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1749 ret
= lookup_tree_block_ref(trans
, root
, path
, bytenr
, parent
,
1752 ret
= lookup_extent_data_ref(trans
, root
, path
, bytenr
, parent
,
1753 root_objectid
, owner
, offset
);
1759 * helper to update/remove inline back ref
1761 static noinline_for_stack
1762 void update_inline_extent_backref(struct btrfs_root
*root
,
1763 struct btrfs_path
*path
,
1764 struct btrfs_extent_inline_ref
*iref
,
1766 struct btrfs_delayed_extent_op
*extent_op
)
1768 struct extent_buffer
*leaf
;
1769 struct btrfs_extent_item
*ei
;
1770 struct btrfs_extent_data_ref
*dref
= NULL
;
1771 struct btrfs_shared_data_ref
*sref
= NULL
;
1779 leaf
= path
->nodes
[0];
1780 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1781 refs
= btrfs_extent_refs(leaf
, ei
);
1782 WARN_ON(refs_to_mod
< 0 && refs
+ refs_to_mod
<= 0);
1783 refs
+= refs_to_mod
;
1784 btrfs_set_extent_refs(leaf
, ei
, refs
);
1786 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1788 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1790 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1791 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1792 refs
= btrfs_extent_data_ref_count(leaf
, dref
);
1793 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1794 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1795 refs
= btrfs_shared_data_ref_count(leaf
, sref
);
1798 BUG_ON(refs_to_mod
!= -1);
1801 BUG_ON(refs_to_mod
< 0 && refs
< -refs_to_mod
);
1802 refs
+= refs_to_mod
;
1805 if (type
== BTRFS_EXTENT_DATA_REF_KEY
)
1806 btrfs_set_extent_data_ref_count(leaf
, dref
, refs
);
1808 btrfs_set_shared_data_ref_count(leaf
, sref
, refs
);
1810 size
= btrfs_extent_inline_ref_size(type
);
1811 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1812 ptr
= (unsigned long)iref
;
1813 end
= (unsigned long)ei
+ item_size
;
1814 if (ptr
+ size
< end
)
1815 memmove_extent_buffer(leaf
, ptr
, ptr
+ size
,
1818 btrfs_truncate_item(root
, path
, item_size
, 1);
1820 btrfs_mark_buffer_dirty(leaf
);
1823 static noinline_for_stack
1824 int insert_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1825 struct btrfs_root
*root
,
1826 struct btrfs_path
*path
,
1827 u64 bytenr
, u64 num_bytes
, u64 parent
,
1828 u64 root_objectid
, u64 owner
,
1829 u64 offset
, int refs_to_add
,
1830 struct btrfs_delayed_extent_op
*extent_op
)
1832 struct btrfs_extent_inline_ref
*iref
;
1835 ret
= lookup_inline_extent_backref(trans
, root
, path
, &iref
,
1836 bytenr
, num_bytes
, parent
,
1837 root_objectid
, owner
, offset
, 1);
1839 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
);
1840 update_inline_extent_backref(root
, path
, iref
,
1841 refs_to_add
, extent_op
);
1842 } else if (ret
== -ENOENT
) {
1843 setup_inline_extent_backref(root
, path
, iref
, parent
,
1844 root_objectid
, owner
, offset
,
1845 refs_to_add
, extent_op
);
1851 static int insert_extent_backref(struct btrfs_trans_handle
*trans
,
1852 struct btrfs_root
*root
,
1853 struct btrfs_path
*path
,
1854 u64 bytenr
, u64 parent
, u64 root_objectid
,
1855 u64 owner
, u64 offset
, int refs_to_add
)
1858 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1859 BUG_ON(refs_to_add
!= 1);
1860 ret
= insert_tree_block_ref(trans
, root
, path
, bytenr
,
1861 parent
, root_objectid
);
1863 ret
= insert_extent_data_ref(trans
, root
, path
, bytenr
,
1864 parent
, root_objectid
,
1865 owner
, offset
, refs_to_add
);
1870 static int remove_extent_backref(struct btrfs_trans_handle
*trans
,
1871 struct btrfs_root
*root
,
1872 struct btrfs_path
*path
,
1873 struct btrfs_extent_inline_ref
*iref
,
1874 int refs_to_drop
, int is_data
)
1878 BUG_ON(!is_data
&& refs_to_drop
!= 1);
1880 update_inline_extent_backref(root
, path
, iref
,
1881 -refs_to_drop
, NULL
);
1882 } else if (is_data
) {
1883 ret
= remove_extent_data_ref(trans
, root
, path
, refs_to_drop
);
1885 ret
= btrfs_del_item(trans
, root
, path
);
1890 static int btrfs_issue_discard(struct block_device
*bdev
,
1893 return blkdev_issue_discard(bdev
, start
>> 9, len
>> 9, GFP_NOFS
, 0);
1896 static int btrfs_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
1897 u64 num_bytes
, u64
*actual_bytes
)
1900 u64 discarded_bytes
= 0;
1901 struct btrfs_bio
*bbio
= NULL
;
1904 /* Tell the block device(s) that the sectors can be discarded */
1905 ret
= btrfs_map_block(root
->fs_info
, REQ_DISCARD
,
1906 bytenr
, &num_bytes
, &bbio
, 0);
1907 /* Error condition is -ENOMEM */
1909 struct btrfs_bio_stripe
*stripe
= bbio
->stripes
;
1913 for (i
= 0; i
< bbio
->num_stripes
; i
++, stripe
++) {
1914 if (!stripe
->dev
->can_discard
)
1917 ret
= btrfs_issue_discard(stripe
->dev
->bdev
,
1921 discarded_bytes
+= stripe
->length
;
1922 else if (ret
!= -EOPNOTSUPP
)
1923 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1926 * Just in case we get back EOPNOTSUPP for some reason,
1927 * just ignore the return value so we don't screw up
1928 * people calling discard_extent.
1936 *actual_bytes
= discarded_bytes
;
1939 if (ret
== -EOPNOTSUPP
)
1944 /* Can return -ENOMEM */
1945 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1946 struct btrfs_root
*root
,
1947 u64 bytenr
, u64 num_bytes
, u64 parent
,
1948 u64 root_objectid
, u64 owner
, u64 offset
, int for_cow
)
1951 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1953 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
&&
1954 root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
1956 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1957 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
1959 parent
, root_objectid
, (int)owner
,
1960 BTRFS_ADD_DELAYED_REF
, NULL
, for_cow
);
1962 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
1964 parent
, root_objectid
, owner
, offset
,
1965 BTRFS_ADD_DELAYED_REF
, NULL
, for_cow
);
1970 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1971 struct btrfs_root
*root
,
1972 u64 bytenr
, u64 num_bytes
,
1973 u64 parent
, u64 root_objectid
,
1974 u64 owner
, u64 offset
, int refs_to_add
,
1975 struct btrfs_delayed_extent_op
*extent_op
)
1977 struct btrfs_path
*path
;
1978 struct extent_buffer
*leaf
;
1979 struct btrfs_extent_item
*item
;
1984 path
= btrfs_alloc_path();
1989 path
->leave_spinning
= 1;
1990 /* this will setup the path even if it fails to insert the back ref */
1991 ret
= insert_inline_extent_backref(trans
, root
->fs_info
->extent_root
,
1992 path
, bytenr
, num_bytes
, parent
,
1993 root_objectid
, owner
, offset
,
1994 refs_to_add
, extent_op
);
1998 if (ret
!= -EAGAIN
) {
2003 leaf
= path
->nodes
[0];
2004 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2005 refs
= btrfs_extent_refs(leaf
, item
);
2006 btrfs_set_extent_refs(leaf
, item
, refs
+ refs_to_add
);
2008 __run_delayed_extent_op(extent_op
, leaf
, item
);
2010 btrfs_mark_buffer_dirty(leaf
);
2011 btrfs_release_path(path
);
2014 path
->leave_spinning
= 1;
2016 /* now insert the actual backref */
2017 ret
= insert_extent_backref(trans
, root
->fs_info
->extent_root
,
2018 path
, bytenr
, parent
, root_objectid
,
2019 owner
, offset
, refs_to_add
);
2021 btrfs_abort_transaction(trans
, root
, ret
);
2023 btrfs_free_path(path
);
2027 static int run_delayed_data_ref(struct btrfs_trans_handle
*trans
,
2028 struct btrfs_root
*root
,
2029 struct btrfs_delayed_ref_node
*node
,
2030 struct btrfs_delayed_extent_op
*extent_op
,
2031 int insert_reserved
)
2034 struct btrfs_delayed_data_ref
*ref
;
2035 struct btrfs_key ins
;
2040 ins
.objectid
= node
->bytenr
;
2041 ins
.offset
= node
->num_bytes
;
2042 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2044 ref
= btrfs_delayed_node_to_data_ref(node
);
2045 trace_run_delayed_data_ref(node
, ref
, node
->action
);
2047 if (node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2048 parent
= ref
->parent
;
2050 ref_root
= ref
->root
;
2052 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2054 flags
|= extent_op
->flags_to_set
;
2055 ret
= alloc_reserved_file_extent(trans
, root
,
2056 parent
, ref_root
, flags
,
2057 ref
->objectid
, ref
->offset
,
2058 &ins
, node
->ref_mod
);
2059 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2060 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
2061 node
->num_bytes
, parent
,
2062 ref_root
, ref
->objectid
,
2063 ref
->offset
, node
->ref_mod
,
2065 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2066 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
2067 node
->num_bytes
, parent
,
2068 ref_root
, ref
->objectid
,
2069 ref
->offset
, node
->ref_mod
,
2077 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
2078 struct extent_buffer
*leaf
,
2079 struct btrfs_extent_item
*ei
)
2081 u64 flags
= btrfs_extent_flags(leaf
, ei
);
2082 if (extent_op
->update_flags
) {
2083 flags
|= extent_op
->flags_to_set
;
2084 btrfs_set_extent_flags(leaf
, ei
, flags
);
2087 if (extent_op
->update_key
) {
2088 struct btrfs_tree_block_info
*bi
;
2089 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
));
2090 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
2091 btrfs_set_tree_block_key(leaf
, bi
, &extent_op
->key
);
2095 static int run_delayed_extent_op(struct btrfs_trans_handle
*trans
,
2096 struct btrfs_root
*root
,
2097 struct btrfs_delayed_ref_node
*node
,
2098 struct btrfs_delayed_extent_op
*extent_op
)
2100 struct btrfs_key key
;
2101 struct btrfs_path
*path
;
2102 struct btrfs_extent_item
*ei
;
2103 struct extent_buffer
*leaf
;
2107 int metadata
= !extent_op
->is_data
;
2112 if (metadata
&& !btrfs_fs_incompat(root
->fs_info
, SKINNY_METADATA
))
2115 path
= btrfs_alloc_path();
2119 key
.objectid
= node
->bytenr
;
2122 key
.type
= BTRFS_METADATA_ITEM_KEY
;
2123 key
.offset
= extent_op
->level
;
2125 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2126 key
.offset
= node
->num_bytes
;
2131 path
->leave_spinning
= 1;
2132 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
,
2140 btrfs_release_path(path
);
2143 key
.offset
= node
->num_bytes
;
2144 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2151 leaf
= path
->nodes
[0];
2152 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2153 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2154 if (item_size
< sizeof(*ei
)) {
2155 ret
= convert_extent_item_v0(trans
, root
->fs_info
->extent_root
,
2161 leaf
= path
->nodes
[0];
2162 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2165 BUG_ON(item_size
< sizeof(*ei
));
2166 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2167 __run_delayed_extent_op(extent_op
, leaf
, ei
);
2169 btrfs_mark_buffer_dirty(leaf
);
2171 btrfs_free_path(path
);
2175 static int run_delayed_tree_ref(struct btrfs_trans_handle
*trans
,
2176 struct btrfs_root
*root
,
2177 struct btrfs_delayed_ref_node
*node
,
2178 struct btrfs_delayed_extent_op
*extent_op
,
2179 int insert_reserved
)
2182 struct btrfs_delayed_tree_ref
*ref
;
2183 struct btrfs_key ins
;
2186 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
2189 ref
= btrfs_delayed_node_to_tree_ref(node
);
2190 trace_run_delayed_tree_ref(node
, ref
, node
->action
);
2192 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2193 parent
= ref
->parent
;
2195 ref_root
= ref
->root
;
2197 ins
.objectid
= node
->bytenr
;
2198 if (skinny_metadata
) {
2199 ins
.offset
= ref
->level
;
2200 ins
.type
= BTRFS_METADATA_ITEM_KEY
;
2202 ins
.offset
= node
->num_bytes
;
2203 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2206 BUG_ON(node
->ref_mod
!= 1);
2207 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2208 BUG_ON(!extent_op
|| !extent_op
->update_flags
);
2209 ret
= alloc_reserved_tree_block(trans
, root
,
2211 extent_op
->flags_to_set
,
2214 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2215 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
2216 node
->num_bytes
, parent
, ref_root
,
2217 ref
->level
, 0, 1, extent_op
);
2218 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2219 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
2220 node
->num_bytes
, parent
, ref_root
,
2221 ref
->level
, 0, 1, extent_op
);
2228 /* helper function to actually process a single delayed ref entry */
2229 static int run_one_delayed_ref(struct btrfs_trans_handle
*trans
,
2230 struct btrfs_root
*root
,
2231 struct btrfs_delayed_ref_node
*node
,
2232 struct btrfs_delayed_extent_op
*extent_op
,
2233 int insert_reserved
)
2240 if (btrfs_delayed_ref_is_head(node
)) {
2241 struct btrfs_delayed_ref_head
*head
;
2243 * we've hit the end of the chain and we were supposed
2244 * to insert this extent into the tree. But, it got
2245 * deleted before we ever needed to insert it, so all
2246 * we have to do is clean up the accounting
2249 head
= btrfs_delayed_node_to_head(node
);
2250 trace_run_delayed_ref_head(node
, head
, node
->action
);
2252 if (insert_reserved
) {
2253 btrfs_pin_extent(root
, node
->bytenr
,
2254 node
->num_bytes
, 1);
2255 if (head
->is_data
) {
2256 ret
= btrfs_del_csums(trans
, root
,
2264 if (node
->type
== BTRFS_TREE_BLOCK_REF_KEY
||
2265 node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2266 ret
= run_delayed_tree_ref(trans
, root
, node
, extent_op
,
2268 else if (node
->type
== BTRFS_EXTENT_DATA_REF_KEY
||
2269 node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2270 ret
= run_delayed_data_ref(trans
, root
, node
, extent_op
,
2277 static noinline
struct btrfs_delayed_ref_node
*
2278 select_delayed_ref(struct btrfs_delayed_ref_head
*head
)
2280 struct rb_node
*node
;
2281 struct btrfs_delayed_ref_node
*ref
;
2282 int action
= BTRFS_ADD_DELAYED_REF
;
2285 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2286 * this prevents ref count from going down to zero when
2287 * there still are pending delayed ref.
2289 node
= rb_prev(&head
->node
.rb_node
);
2293 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
2295 if (ref
->bytenr
!= head
->node
.bytenr
)
2297 if (ref
->action
== action
)
2299 node
= rb_prev(node
);
2301 if (action
== BTRFS_ADD_DELAYED_REF
) {
2302 action
= BTRFS_DROP_DELAYED_REF
;
2309 * Returns 0 on success or if called with an already aborted transaction.
2310 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2312 static noinline
int run_clustered_refs(struct btrfs_trans_handle
*trans
,
2313 struct btrfs_root
*root
,
2314 struct list_head
*cluster
)
2316 struct btrfs_delayed_ref_root
*delayed_refs
;
2317 struct btrfs_delayed_ref_node
*ref
;
2318 struct btrfs_delayed_ref_head
*locked_ref
= NULL
;
2319 struct btrfs_delayed_extent_op
*extent_op
;
2320 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2323 int must_insert_reserved
= 0;
2325 delayed_refs
= &trans
->transaction
->delayed_refs
;
2328 /* pick a new head ref from the cluster list */
2329 if (list_empty(cluster
))
2332 locked_ref
= list_entry(cluster
->next
,
2333 struct btrfs_delayed_ref_head
, cluster
);
2335 /* grab the lock that says we are going to process
2336 * all the refs for this head */
2337 ret
= btrfs_delayed_ref_lock(trans
, locked_ref
);
2340 * we may have dropped the spin lock to get the head
2341 * mutex lock, and that might have given someone else
2342 * time to free the head. If that's true, it has been
2343 * removed from our list and we can move on.
2345 if (ret
== -EAGAIN
) {
2353 * We need to try and merge add/drops of the same ref since we
2354 * can run into issues with relocate dropping the implicit ref
2355 * and then it being added back again before the drop can
2356 * finish. If we merged anything we need to re-loop so we can
2359 btrfs_merge_delayed_refs(trans
, fs_info
, delayed_refs
,
2363 * locked_ref is the head node, so we have to go one
2364 * node back for any delayed ref updates
2366 ref
= select_delayed_ref(locked_ref
);
2368 if (ref
&& ref
->seq
&&
2369 btrfs_check_delayed_seq(fs_info
, delayed_refs
, ref
->seq
)) {
2371 * there are still refs with lower seq numbers in the
2372 * process of being added. Don't run this ref yet.
2374 list_del_init(&locked_ref
->cluster
);
2375 btrfs_delayed_ref_unlock(locked_ref
);
2377 delayed_refs
->num_heads_ready
++;
2378 spin_unlock(&delayed_refs
->lock
);
2380 spin_lock(&delayed_refs
->lock
);
2385 * record the must insert reserved flag before we
2386 * drop the spin lock.
2388 must_insert_reserved
= locked_ref
->must_insert_reserved
;
2389 locked_ref
->must_insert_reserved
= 0;
2391 extent_op
= locked_ref
->extent_op
;
2392 locked_ref
->extent_op
= NULL
;
2395 /* All delayed refs have been processed, Go ahead
2396 * and send the head node to run_one_delayed_ref,
2397 * so that any accounting fixes can happen
2399 ref
= &locked_ref
->node
;
2401 if (extent_op
&& must_insert_reserved
) {
2402 btrfs_free_delayed_extent_op(extent_op
);
2407 spin_unlock(&delayed_refs
->lock
);
2409 ret
= run_delayed_extent_op(trans
, root
,
2411 btrfs_free_delayed_extent_op(extent_op
);
2414 btrfs_debug(fs_info
, "run_delayed_extent_op returned %d", ret
);
2415 spin_lock(&delayed_refs
->lock
);
2416 btrfs_delayed_ref_unlock(locked_ref
);
2425 rb_erase(&ref
->rb_node
, &delayed_refs
->root
);
2426 delayed_refs
->num_entries
--;
2427 if (!btrfs_delayed_ref_is_head(ref
)) {
2429 * when we play the delayed ref, also correct the
2432 switch (ref
->action
) {
2433 case BTRFS_ADD_DELAYED_REF
:
2434 case BTRFS_ADD_DELAYED_EXTENT
:
2435 locked_ref
->node
.ref_mod
-= ref
->ref_mod
;
2437 case BTRFS_DROP_DELAYED_REF
:
2438 locked_ref
->node
.ref_mod
+= ref
->ref_mod
;
2444 list_del_init(&locked_ref
->cluster
);
2446 spin_unlock(&delayed_refs
->lock
);
2448 ret
= run_one_delayed_ref(trans
, root
, ref
, extent_op
,
2449 must_insert_reserved
);
2451 btrfs_free_delayed_extent_op(extent_op
);
2453 btrfs_delayed_ref_unlock(locked_ref
);
2454 btrfs_put_delayed_ref(ref
);
2455 btrfs_debug(fs_info
, "run_one_delayed_ref returned %d", ret
);
2456 spin_lock(&delayed_refs
->lock
);
2461 * If this node is a head, that means all the refs in this head
2462 * have been dealt with, and we will pick the next head to deal
2463 * with, so we must unlock the head and drop it from the cluster
2464 * list before we release it.
2466 if (btrfs_delayed_ref_is_head(ref
)) {
2467 btrfs_delayed_ref_unlock(locked_ref
);
2470 btrfs_put_delayed_ref(ref
);
2474 spin_lock(&delayed_refs
->lock
);
2479 #ifdef SCRAMBLE_DELAYED_REFS
2481 * Normally delayed refs get processed in ascending bytenr order. This
2482 * correlates in most cases to the order added. To expose dependencies on this
2483 * order, we start to process the tree in the middle instead of the beginning
2485 static u64
find_middle(struct rb_root
*root
)
2487 struct rb_node
*n
= root
->rb_node
;
2488 struct btrfs_delayed_ref_node
*entry
;
2491 u64 first
= 0, last
= 0;
2495 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2496 first
= entry
->bytenr
;
2500 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2501 last
= entry
->bytenr
;
2506 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2507 WARN_ON(!entry
->in_tree
);
2509 middle
= entry
->bytenr
;
2522 int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle
*trans
,
2523 struct btrfs_fs_info
*fs_info
)
2525 struct qgroup_update
*qgroup_update
;
2528 if (list_empty(&trans
->qgroup_ref_list
) !=
2529 !trans
->delayed_ref_elem
.seq
) {
2530 /* list without seq or seq without list */
2532 "qgroup accounting update error, list is%s empty, seq is %#x.%x",
2533 list_empty(&trans
->qgroup_ref_list
) ? "" : " not",
2534 (u32
)(trans
->delayed_ref_elem
.seq
>> 32),
2535 (u32
)trans
->delayed_ref_elem
.seq
);
2539 if (!trans
->delayed_ref_elem
.seq
)
2542 while (!list_empty(&trans
->qgroup_ref_list
)) {
2543 qgroup_update
= list_first_entry(&trans
->qgroup_ref_list
,
2544 struct qgroup_update
, list
);
2545 list_del(&qgroup_update
->list
);
2547 ret
= btrfs_qgroup_account_ref(
2548 trans
, fs_info
, qgroup_update
->node
,
2549 qgroup_update
->extent_op
);
2550 kfree(qgroup_update
);
2553 btrfs_put_tree_mod_seq(fs_info
, &trans
->delayed_ref_elem
);
2558 static int refs_newer(struct btrfs_delayed_ref_root
*delayed_refs
, int seq
,
2561 int val
= atomic_read(&delayed_refs
->ref_seq
);
2563 if (val
< seq
|| val
>= seq
+ count
)
2568 static inline u64
heads_to_leaves(struct btrfs_root
*root
, u64 heads
)
2572 num_bytes
= heads
* (sizeof(struct btrfs_extent_item
) +
2573 sizeof(struct btrfs_extent_inline_ref
));
2574 if (!btrfs_fs_incompat(root
->fs_info
, SKINNY_METADATA
))
2575 num_bytes
+= heads
* sizeof(struct btrfs_tree_block_info
);
2578 * We don't ever fill up leaves all the way so multiply by 2 just to be
2579 * closer to what we're really going to want to ouse.
2581 return div64_u64(num_bytes
, BTRFS_LEAF_DATA_SIZE(root
));
2584 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle
*trans
,
2585 struct btrfs_root
*root
)
2587 struct btrfs_block_rsv
*global_rsv
;
2588 u64 num_heads
= trans
->transaction
->delayed_refs
.num_heads_ready
;
2592 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
2593 num_heads
= heads_to_leaves(root
, num_heads
);
2595 num_bytes
+= (num_heads
- 1) * root
->leafsize
;
2597 global_rsv
= &root
->fs_info
->global_block_rsv
;
2600 * If we can't allocate any more chunks lets make sure we have _lots_ of
2601 * wiggle room since running delayed refs can create more delayed refs.
2603 if (global_rsv
->space_info
->full
)
2606 spin_lock(&global_rsv
->lock
);
2607 if (global_rsv
->reserved
<= num_bytes
)
2609 spin_unlock(&global_rsv
->lock
);
2614 * this starts processing the delayed reference count updates and
2615 * extent insertions we have queued up so far. count can be
2616 * 0, which means to process everything in the tree at the start
2617 * of the run (but not newly added entries), or it can be some target
2618 * number you'd like to process.
2620 * Returns 0 on success or if called with an aborted transaction
2621 * Returns <0 on error and aborts the transaction
2623 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2624 struct btrfs_root
*root
, unsigned long count
)
2626 struct rb_node
*node
;
2627 struct btrfs_delayed_ref_root
*delayed_refs
;
2628 struct btrfs_delayed_ref_node
*ref
;
2629 struct list_head cluster
;
2632 int run_all
= count
== (unsigned long)-1;
2636 /* We'll clean this up in btrfs_cleanup_transaction */
2640 if (root
== root
->fs_info
->extent_root
)
2641 root
= root
->fs_info
->tree_root
;
2643 btrfs_delayed_refs_qgroup_accounting(trans
, root
->fs_info
);
2645 delayed_refs
= &trans
->transaction
->delayed_refs
;
2646 INIT_LIST_HEAD(&cluster
);
2648 count
= delayed_refs
->num_entries
* 2;
2652 if (!run_all
&& !run_most
) {
2654 int seq
= atomic_read(&delayed_refs
->ref_seq
);
2657 old
= atomic_cmpxchg(&delayed_refs
->procs_running_refs
, 0, 1);
2659 DEFINE_WAIT(__wait
);
2660 if (delayed_refs
->flushing
||
2661 !btrfs_should_throttle_delayed_refs(trans
, root
))
2664 prepare_to_wait(&delayed_refs
->wait
, &__wait
,
2665 TASK_UNINTERRUPTIBLE
);
2667 old
= atomic_cmpxchg(&delayed_refs
->procs_running_refs
, 0, 1);
2670 finish_wait(&delayed_refs
->wait
, &__wait
);
2672 if (!refs_newer(delayed_refs
, seq
, 256))
2677 finish_wait(&delayed_refs
->wait
, &__wait
);
2683 atomic_inc(&delayed_refs
->procs_running_refs
);
2688 spin_lock(&delayed_refs
->lock
);
2690 #ifdef SCRAMBLE_DELAYED_REFS
2691 delayed_refs
->run_delayed_start
= find_middle(&delayed_refs
->root
);
2695 if (!(run_all
|| run_most
) &&
2696 !btrfs_should_throttle_delayed_refs(trans
, root
))
2700 * go find something we can process in the rbtree. We start at
2701 * the beginning of the tree, and then build a cluster
2702 * of refs to process starting at the first one we are able to
2705 delayed_start
= delayed_refs
->run_delayed_start
;
2706 ret
= btrfs_find_ref_cluster(trans
, &cluster
,
2707 delayed_refs
->run_delayed_start
);
2711 ret
= run_clustered_refs(trans
, root
, &cluster
);
2713 btrfs_release_ref_cluster(&cluster
);
2714 spin_unlock(&delayed_refs
->lock
);
2715 btrfs_abort_transaction(trans
, root
, ret
);
2716 atomic_dec(&delayed_refs
->procs_running_refs
);
2717 wake_up(&delayed_refs
->wait
);
2721 atomic_add(ret
, &delayed_refs
->ref_seq
);
2723 count
-= min_t(unsigned long, ret
, count
);
2728 if (delayed_start
>= delayed_refs
->run_delayed_start
) {
2731 * btrfs_find_ref_cluster looped. let's do one
2732 * more cycle. if we don't run any delayed ref
2733 * during that cycle (because we can't because
2734 * all of them are blocked), bail out.
2739 * no runnable refs left, stop trying
2746 /* refs were run, let's reset staleness detection */
2752 if (!list_empty(&trans
->new_bgs
)) {
2753 spin_unlock(&delayed_refs
->lock
);
2754 btrfs_create_pending_block_groups(trans
, root
);
2755 spin_lock(&delayed_refs
->lock
);
2758 node
= rb_first(&delayed_refs
->root
);
2761 count
= (unsigned long)-1;
2764 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
2766 if (btrfs_delayed_ref_is_head(ref
)) {
2767 struct btrfs_delayed_ref_head
*head
;
2769 head
= btrfs_delayed_node_to_head(ref
);
2770 atomic_inc(&ref
->refs
);
2772 spin_unlock(&delayed_refs
->lock
);
2774 * Mutex was contended, block until it's
2775 * released and try again
2777 mutex_lock(&head
->mutex
);
2778 mutex_unlock(&head
->mutex
);
2780 btrfs_put_delayed_ref(ref
);
2784 node
= rb_next(node
);
2786 spin_unlock(&delayed_refs
->lock
);
2787 schedule_timeout(1);
2791 atomic_dec(&delayed_refs
->procs_running_refs
);
2793 if (waitqueue_active(&delayed_refs
->wait
))
2794 wake_up(&delayed_refs
->wait
);
2796 spin_unlock(&delayed_refs
->lock
);
2797 assert_qgroups_uptodate(trans
);
2801 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
2802 struct btrfs_root
*root
,
2803 u64 bytenr
, u64 num_bytes
, u64 flags
,
2804 int level
, int is_data
)
2806 struct btrfs_delayed_extent_op
*extent_op
;
2809 extent_op
= btrfs_alloc_delayed_extent_op();
2813 extent_op
->flags_to_set
= flags
;
2814 extent_op
->update_flags
= 1;
2815 extent_op
->update_key
= 0;
2816 extent_op
->is_data
= is_data
? 1 : 0;
2817 extent_op
->level
= level
;
2819 ret
= btrfs_add_delayed_extent_op(root
->fs_info
, trans
, bytenr
,
2820 num_bytes
, extent_op
);
2822 btrfs_free_delayed_extent_op(extent_op
);
2826 static noinline
int check_delayed_ref(struct btrfs_trans_handle
*trans
,
2827 struct btrfs_root
*root
,
2828 struct btrfs_path
*path
,
2829 u64 objectid
, u64 offset
, u64 bytenr
)
2831 struct btrfs_delayed_ref_head
*head
;
2832 struct btrfs_delayed_ref_node
*ref
;
2833 struct btrfs_delayed_data_ref
*data_ref
;
2834 struct btrfs_delayed_ref_root
*delayed_refs
;
2835 struct rb_node
*node
;
2839 delayed_refs
= &trans
->transaction
->delayed_refs
;
2840 spin_lock(&delayed_refs
->lock
);
2841 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
2845 if (!mutex_trylock(&head
->mutex
)) {
2846 atomic_inc(&head
->node
.refs
);
2847 spin_unlock(&delayed_refs
->lock
);
2849 btrfs_release_path(path
);
2852 * Mutex was contended, block until it's released and let
2855 mutex_lock(&head
->mutex
);
2856 mutex_unlock(&head
->mutex
);
2857 btrfs_put_delayed_ref(&head
->node
);
2861 node
= rb_prev(&head
->node
.rb_node
);
2865 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2867 if (ref
->bytenr
!= bytenr
)
2871 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
)
2874 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
2876 node
= rb_prev(node
);
2880 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2881 if (ref
->bytenr
== bytenr
&& ref
->seq
== seq
)
2885 if (data_ref
->root
!= root
->root_key
.objectid
||
2886 data_ref
->objectid
!= objectid
|| data_ref
->offset
!= offset
)
2891 mutex_unlock(&head
->mutex
);
2893 spin_unlock(&delayed_refs
->lock
);
2897 static noinline
int check_committed_ref(struct btrfs_trans_handle
*trans
,
2898 struct btrfs_root
*root
,
2899 struct btrfs_path
*path
,
2900 u64 objectid
, u64 offset
, u64 bytenr
)
2902 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2903 struct extent_buffer
*leaf
;
2904 struct btrfs_extent_data_ref
*ref
;
2905 struct btrfs_extent_inline_ref
*iref
;
2906 struct btrfs_extent_item
*ei
;
2907 struct btrfs_key key
;
2911 key
.objectid
= bytenr
;
2912 key
.offset
= (u64
)-1;
2913 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2915 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
2918 BUG_ON(ret
== 0); /* Corruption */
2921 if (path
->slots
[0] == 0)
2925 leaf
= path
->nodes
[0];
2926 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2928 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
2932 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2933 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2934 if (item_size
< sizeof(*ei
)) {
2935 WARN_ON(item_size
!= sizeof(struct btrfs_extent_item_v0
));
2939 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2941 if (item_size
!= sizeof(*ei
) +
2942 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
2945 if (btrfs_extent_generation(leaf
, ei
) <=
2946 btrfs_root_last_snapshot(&root
->root_item
))
2949 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
2950 if (btrfs_extent_inline_ref_type(leaf
, iref
) !=
2951 BTRFS_EXTENT_DATA_REF_KEY
)
2954 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
2955 if (btrfs_extent_refs(leaf
, ei
) !=
2956 btrfs_extent_data_ref_count(leaf
, ref
) ||
2957 btrfs_extent_data_ref_root(leaf
, ref
) !=
2958 root
->root_key
.objectid
||
2959 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
2960 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
2968 int btrfs_cross_ref_exist(struct btrfs_trans_handle
*trans
,
2969 struct btrfs_root
*root
,
2970 u64 objectid
, u64 offset
, u64 bytenr
)
2972 struct btrfs_path
*path
;
2976 path
= btrfs_alloc_path();
2981 ret
= check_committed_ref(trans
, root
, path
, objectid
,
2983 if (ret
&& ret
!= -ENOENT
)
2986 ret2
= check_delayed_ref(trans
, root
, path
, objectid
,
2988 } while (ret2
== -EAGAIN
);
2990 if (ret2
&& ret2
!= -ENOENT
) {
2995 if (ret
!= -ENOENT
|| ret2
!= -ENOENT
)
2998 btrfs_free_path(path
);
2999 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
3004 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
3005 struct btrfs_root
*root
,
3006 struct extent_buffer
*buf
,
3007 int full_backref
, int inc
, int for_cow
)
3014 struct btrfs_key key
;
3015 struct btrfs_file_extent_item
*fi
;
3019 int (*process_func
)(struct btrfs_trans_handle
*, struct btrfs_root
*,
3020 u64
, u64
, u64
, u64
, u64
, u64
, int);
3022 ref_root
= btrfs_header_owner(buf
);
3023 nritems
= btrfs_header_nritems(buf
);
3024 level
= btrfs_header_level(buf
);
3026 if (!root
->ref_cows
&& level
== 0)
3030 process_func
= btrfs_inc_extent_ref
;
3032 process_func
= btrfs_free_extent
;
3035 parent
= buf
->start
;
3039 for (i
= 0; i
< nritems
; i
++) {
3041 btrfs_item_key_to_cpu(buf
, &key
, i
);
3042 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
3044 fi
= btrfs_item_ptr(buf
, i
,
3045 struct btrfs_file_extent_item
);
3046 if (btrfs_file_extent_type(buf
, fi
) ==
3047 BTRFS_FILE_EXTENT_INLINE
)
3049 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
3053 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
3054 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
3055 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
3056 parent
, ref_root
, key
.objectid
,
3057 key
.offset
, for_cow
);
3061 bytenr
= btrfs_node_blockptr(buf
, i
);
3062 num_bytes
= btrfs_level_size(root
, level
- 1);
3063 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
3064 parent
, ref_root
, level
- 1, 0,
3075 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3076 struct extent_buffer
*buf
, int full_backref
, int for_cow
)
3078 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1, for_cow
);
3081 int btrfs_dec_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3082 struct extent_buffer
*buf
, int full_backref
, int for_cow
)
3084 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 0, for_cow
);
3087 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
3088 struct btrfs_root
*root
,
3089 struct btrfs_path
*path
,
3090 struct btrfs_block_group_cache
*cache
)
3093 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
3095 struct extent_buffer
*leaf
;
3097 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
3100 BUG_ON(ret
); /* Corruption */
3102 leaf
= path
->nodes
[0];
3103 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
3104 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
3105 btrfs_mark_buffer_dirty(leaf
);
3106 btrfs_release_path(path
);
3109 btrfs_abort_transaction(trans
, root
, ret
);
3116 static struct btrfs_block_group_cache
*
3117 next_block_group(struct btrfs_root
*root
,
3118 struct btrfs_block_group_cache
*cache
)
3120 struct rb_node
*node
;
3121 spin_lock(&root
->fs_info
->block_group_cache_lock
);
3122 node
= rb_next(&cache
->cache_node
);
3123 btrfs_put_block_group(cache
);
3125 cache
= rb_entry(node
, struct btrfs_block_group_cache
,
3127 btrfs_get_block_group(cache
);
3130 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
3134 static int cache_save_setup(struct btrfs_block_group_cache
*block_group
,
3135 struct btrfs_trans_handle
*trans
,
3136 struct btrfs_path
*path
)
3138 struct btrfs_root
*root
= block_group
->fs_info
->tree_root
;
3139 struct inode
*inode
= NULL
;
3141 int dcs
= BTRFS_DC_ERROR
;
3147 * If this block group is smaller than 100 megs don't bother caching the
3150 if (block_group
->key
.offset
< (100 * 1024 * 1024)) {
3151 spin_lock(&block_group
->lock
);
3152 block_group
->disk_cache_state
= BTRFS_DC_WRITTEN
;
3153 spin_unlock(&block_group
->lock
);
3158 inode
= lookup_free_space_inode(root
, block_group
, path
);
3159 if (IS_ERR(inode
) && PTR_ERR(inode
) != -ENOENT
) {
3160 ret
= PTR_ERR(inode
);
3161 btrfs_release_path(path
);
3165 if (IS_ERR(inode
)) {
3169 if (block_group
->ro
)
3172 ret
= create_free_space_inode(root
, trans
, block_group
, path
);
3178 /* We've already setup this transaction, go ahead and exit */
3179 if (block_group
->cache_generation
== trans
->transid
&&
3180 i_size_read(inode
)) {
3181 dcs
= BTRFS_DC_SETUP
;
3186 * We want to set the generation to 0, that way if anything goes wrong
3187 * from here on out we know not to trust this cache when we load up next
3190 BTRFS_I(inode
)->generation
= 0;
3191 ret
= btrfs_update_inode(trans
, root
, inode
);
3194 if (i_size_read(inode
) > 0) {
3195 ret
= btrfs_check_trunc_cache_free_space(root
,
3196 &root
->fs_info
->global_block_rsv
);
3200 ret
= btrfs_truncate_free_space_cache(root
, trans
, inode
);
3205 spin_lock(&block_group
->lock
);
3206 if (block_group
->cached
!= BTRFS_CACHE_FINISHED
||
3207 !btrfs_test_opt(root
, SPACE_CACHE
)) {
3209 * don't bother trying to write stuff out _if_
3210 * a) we're not cached,
3211 * b) we're with nospace_cache mount option.
3213 dcs
= BTRFS_DC_WRITTEN
;
3214 spin_unlock(&block_group
->lock
);
3217 spin_unlock(&block_group
->lock
);
3220 * Try to preallocate enough space based on how big the block group is.
3221 * Keep in mind this has to include any pinned space which could end up
3222 * taking up quite a bit since it's not folded into the other space
3225 num_pages
= (int)div64_u64(block_group
->key
.offset
, 256 * 1024 * 1024);
3230 num_pages
*= PAGE_CACHE_SIZE
;
3232 ret
= btrfs_check_data_free_space(inode
, num_pages
);
3236 ret
= btrfs_prealloc_file_range_trans(inode
, trans
, 0, 0, num_pages
,
3237 num_pages
, num_pages
,
3240 dcs
= BTRFS_DC_SETUP
;
3241 btrfs_free_reserved_data_space(inode
, num_pages
);
3246 btrfs_release_path(path
);
3248 spin_lock(&block_group
->lock
);
3249 if (!ret
&& dcs
== BTRFS_DC_SETUP
)
3250 block_group
->cache_generation
= trans
->transid
;
3251 block_group
->disk_cache_state
= dcs
;
3252 spin_unlock(&block_group
->lock
);
3257 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
3258 struct btrfs_root
*root
)
3260 struct btrfs_block_group_cache
*cache
;
3262 struct btrfs_path
*path
;
3265 path
= btrfs_alloc_path();
3271 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
3273 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
)
3275 cache
= next_block_group(root
, cache
);
3283 err
= cache_save_setup(cache
, trans
, path
);
3284 last
= cache
->key
.objectid
+ cache
->key
.offset
;
3285 btrfs_put_block_group(cache
);
3290 err
= btrfs_run_delayed_refs(trans
, root
,
3292 if (err
) /* File system offline */
3296 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
3298 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
) {
3299 btrfs_put_block_group(cache
);
3305 cache
= next_block_group(root
, cache
);
3314 if (cache
->disk_cache_state
== BTRFS_DC_SETUP
)
3315 cache
->disk_cache_state
= BTRFS_DC_NEED_WRITE
;
3317 last
= cache
->key
.objectid
+ cache
->key
.offset
;
3319 err
= write_one_cache_group(trans
, root
, path
, cache
);
3320 btrfs_put_block_group(cache
);
3321 if (err
) /* File system offline */
3327 * I don't think this is needed since we're just marking our
3328 * preallocated extent as written, but just in case it can't
3332 err
= btrfs_run_delayed_refs(trans
, root
,
3334 if (err
) /* File system offline */
3338 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
3341 * Really this shouldn't happen, but it could if we
3342 * couldn't write the entire preallocated extent and
3343 * splitting the extent resulted in a new block.
3346 btrfs_put_block_group(cache
);
3349 if (cache
->disk_cache_state
== BTRFS_DC_NEED_WRITE
)
3351 cache
= next_block_group(root
, cache
);
3360 err
= btrfs_write_out_cache(root
, trans
, cache
, path
);
3363 * If we didn't have an error then the cache state is still
3364 * NEED_WRITE, so we can set it to WRITTEN.
3366 if (!err
&& cache
->disk_cache_state
== BTRFS_DC_NEED_WRITE
)
3367 cache
->disk_cache_state
= BTRFS_DC_WRITTEN
;
3368 last
= cache
->key
.objectid
+ cache
->key
.offset
;
3369 btrfs_put_block_group(cache
);
3373 btrfs_free_path(path
);
3377 int btrfs_extent_readonly(struct btrfs_root
*root
, u64 bytenr
)
3379 struct btrfs_block_group_cache
*block_group
;
3382 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
3383 if (!block_group
|| block_group
->ro
)
3386 btrfs_put_block_group(block_group
);
3390 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
3391 u64 total_bytes
, u64 bytes_used
,
3392 struct btrfs_space_info
**space_info
)
3394 struct btrfs_space_info
*found
;
3399 if (flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3400 BTRFS_BLOCK_GROUP_RAID10
))
3405 found
= __find_space_info(info
, flags
);
3407 spin_lock(&found
->lock
);
3408 found
->total_bytes
+= total_bytes
;
3409 found
->disk_total
+= total_bytes
* factor
;
3410 found
->bytes_used
+= bytes_used
;
3411 found
->disk_used
+= bytes_used
* factor
;
3413 spin_unlock(&found
->lock
);
3414 *space_info
= found
;
3417 found
= kzalloc(sizeof(*found
), GFP_NOFS
);
3421 ret
= percpu_counter_init(&found
->total_bytes_pinned
, 0);
3427 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
3428 INIT_LIST_HEAD(&found
->block_groups
[i
]);
3429 init_rwsem(&found
->groups_sem
);
3430 spin_lock_init(&found
->lock
);
3431 found
->flags
= flags
& BTRFS_BLOCK_GROUP_TYPE_MASK
;
3432 found
->total_bytes
= total_bytes
;
3433 found
->disk_total
= total_bytes
* factor
;
3434 found
->bytes_used
= bytes_used
;
3435 found
->disk_used
= bytes_used
* factor
;
3436 found
->bytes_pinned
= 0;
3437 found
->bytes_reserved
= 0;
3438 found
->bytes_readonly
= 0;
3439 found
->bytes_may_use
= 0;
3441 found
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3442 found
->chunk_alloc
= 0;
3444 init_waitqueue_head(&found
->wait
);
3445 *space_info
= found
;
3446 list_add_rcu(&found
->list
, &info
->space_info
);
3447 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3448 info
->data_sinfo
= found
;
3452 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
3454 u64 extra_flags
= chunk_to_extended(flags
) &
3455 BTRFS_EXTENDED_PROFILE_MASK
;
3457 write_seqlock(&fs_info
->profiles_lock
);
3458 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3459 fs_info
->avail_data_alloc_bits
|= extra_flags
;
3460 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3461 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
3462 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3463 fs_info
->avail_system_alloc_bits
|= extra_flags
;
3464 write_sequnlock(&fs_info
->profiles_lock
);
3468 * returns target flags in extended format or 0 if restripe for this
3469 * chunk_type is not in progress
3471 * should be called with either volume_mutex or balance_lock held
3473 static u64
get_restripe_target(struct btrfs_fs_info
*fs_info
, u64 flags
)
3475 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3481 if (flags
& BTRFS_BLOCK_GROUP_DATA
&&
3482 bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3483 target
= BTRFS_BLOCK_GROUP_DATA
| bctl
->data
.target
;
3484 } else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
&&
3485 bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3486 target
= BTRFS_BLOCK_GROUP_SYSTEM
| bctl
->sys
.target
;
3487 } else if (flags
& BTRFS_BLOCK_GROUP_METADATA
&&
3488 bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3489 target
= BTRFS_BLOCK_GROUP_METADATA
| bctl
->meta
.target
;
3496 * @flags: available profiles in extended format (see ctree.h)
3498 * Returns reduced profile in chunk format. If profile changing is in
3499 * progress (either running or paused) picks the target profile (if it's
3500 * already available), otherwise falls back to plain reducing.
3502 static u64
btrfs_reduce_alloc_profile(struct btrfs_root
*root
, u64 flags
)
3505 * we add in the count of missing devices because we want
3506 * to make sure that any RAID levels on a degraded FS
3507 * continue to be honored.
3509 u64 num_devices
= root
->fs_info
->fs_devices
->rw_devices
+
3510 root
->fs_info
->fs_devices
->missing_devices
;
3515 * see if restripe for this chunk_type is in progress, if so
3516 * try to reduce to the target profile
3518 spin_lock(&root
->fs_info
->balance_lock
);
3519 target
= get_restripe_target(root
->fs_info
, flags
);
3521 /* pick target profile only if it's already available */
3522 if ((flags
& target
) & BTRFS_EXTENDED_PROFILE_MASK
) {
3523 spin_unlock(&root
->fs_info
->balance_lock
);
3524 return extended_to_chunk(target
);
3527 spin_unlock(&root
->fs_info
->balance_lock
);
3529 /* First, mask out the RAID levels which aren't possible */
3530 if (num_devices
== 1)
3531 flags
&= ~(BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID0
|
3532 BTRFS_BLOCK_GROUP_RAID5
);
3533 if (num_devices
< 3)
3534 flags
&= ~BTRFS_BLOCK_GROUP_RAID6
;
3535 if (num_devices
< 4)
3536 flags
&= ~BTRFS_BLOCK_GROUP_RAID10
;
3538 tmp
= flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID0
|
3539 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID5
|
3540 BTRFS_BLOCK_GROUP_RAID6
| BTRFS_BLOCK_GROUP_RAID10
);
3543 if (tmp
& BTRFS_BLOCK_GROUP_RAID6
)
3544 tmp
= BTRFS_BLOCK_GROUP_RAID6
;
3545 else if (tmp
& BTRFS_BLOCK_GROUP_RAID5
)
3546 tmp
= BTRFS_BLOCK_GROUP_RAID5
;
3547 else if (tmp
& BTRFS_BLOCK_GROUP_RAID10
)
3548 tmp
= BTRFS_BLOCK_GROUP_RAID10
;
3549 else if (tmp
& BTRFS_BLOCK_GROUP_RAID1
)
3550 tmp
= BTRFS_BLOCK_GROUP_RAID1
;
3551 else if (tmp
& BTRFS_BLOCK_GROUP_RAID0
)
3552 tmp
= BTRFS_BLOCK_GROUP_RAID0
;
3554 return extended_to_chunk(flags
| tmp
);
3557 static u64
get_alloc_profile(struct btrfs_root
*root
, u64 flags
)
3562 seq
= read_seqbegin(&root
->fs_info
->profiles_lock
);
3564 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3565 flags
|= root
->fs_info
->avail_data_alloc_bits
;
3566 else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3567 flags
|= root
->fs_info
->avail_system_alloc_bits
;
3568 else if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3569 flags
|= root
->fs_info
->avail_metadata_alloc_bits
;
3570 } while (read_seqretry(&root
->fs_info
->profiles_lock
, seq
));
3572 return btrfs_reduce_alloc_profile(root
, flags
);
3575 u64
btrfs_get_alloc_profile(struct btrfs_root
*root
, int data
)
3581 flags
= BTRFS_BLOCK_GROUP_DATA
;
3582 else if (root
== root
->fs_info
->chunk_root
)
3583 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
3585 flags
= BTRFS_BLOCK_GROUP_METADATA
;
3587 ret
= get_alloc_profile(root
, flags
);
3592 * This will check the space that the inode allocates from to make sure we have
3593 * enough space for bytes.
3595 int btrfs_check_data_free_space(struct inode
*inode
, u64 bytes
)
3597 struct btrfs_space_info
*data_sinfo
;
3598 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3599 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3601 int ret
= 0, committed
= 0, alloc_chunk
= 1;
3603 /* make sure bytes are sectorsize aligned */
3604 bytes
= ALIGN(bytes
, root
->sectorsize
);
3606 if (root
== root
->fs_info
->tree_root
||
3607 BTRFS_I(inode
)->location
.objectid
== BTRFS_FREE_INO_OBJECTID
) {
3612 data_sinfo
= fs_info
->data_sinfo
;
3617 /* make sure we have enough space to handle the data first */
3618 spin_lock(&data_sinfo
->lock
);
3619 used
= data_sinfo
->bytes_used
+ data_sinfo
->bytes_reserved
+
3620 data_sinfo
->bytes_pinned
+ data_sinfo
->bytes_readonly
+
3621 data_sinfo
->bytes_may_use
;
3623 if (used
+ bytes
> data_sinfo
->total_bytes
) {
3624 struct btrfs_trans_handle
*trans
;
3627 * if we don't have enough free bytes in this space then we need
3628 * to alloc a new chunk.
3630 if (!data_sinfo
->full
&& alloc_chunk
) {
3633 data_sinfo
->force_alloc
= CHUNK_ALLOC_FORCE
;
3634 spin_unlock(&data_sinfo
->lock
);
3636 alloc_target
= btrfs_get_alloc_profile(root
, 1);
3637 trans
= btrfs_join_transaction(root
);
3639 return PTR_ERR(trans
);
3641 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
3643 CHUNK_ALLOC_NO_FORCE
);
3644 btrfs_end_transaction(trans
, root
);
3653 data_sinfo
= fs_info
->data_sinfo
;
3659 * If we don't have enough pinned space to deal with this
3660 * allocation don't bother committing the transaction.
3662 if (percpu_counter_compare(&data_sinfo
->total_bytes_pinned
,
3665 spin_unlock(&data_sinfo
->lock
);
3667 /* commit the current transaction and try again */
3670 !atomic_read(&root
->fs_info
->open_ioctl_trans
)) {
3673 trans
= btrfs_join_transaction(root
);
3675 return PTR_ERR(trans
);
3676 ret
= btrfs_commit_transaction(trans
, root
);
3684 data_sinfo
->bytes_may_use
+= bytes
;
3685 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
3686 data_sinfo
->flags
, bytes
, 1);
3687 spin_unlock(&data_sinfo
->lock
);
3693 * Called if we need to clear a data reservation for this inode.
3695 void btrfs_free_reserved_data_space(struct inode
*inode
, u64 bytes
)
3697 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3698 struct btrfs_space_info
*data_sinfo
;
3700 /* make sure bytes are sectorsize aligned */
3701 bytes
= ALIGN(bytes
, root
->sectorsize
);
3703 data_sinfo
= root
->fs_info
->data_sinfo
;
3704 spin_lock(&data_sinfo
->lock
);
3705 WARN_ON(data_sinfo
->bytes_may_use
< bytes
);
3706 data_sinfo
->bytes_may_use
-= bytes
;
3707 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
3708 data_sinfo
->flags
, bytes
, 0);
3709 spin_unlock(&data_sinfo
->lock
);
3712 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
3714 struct list_head
*head
= &info
->space_info
;
3715 struct btrfs_space_info
*found
;
3718 list_for_each_entry_rcu(found
, head
, list
) {
3719 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
3720 found
->force_alloc
= CHUNK_ALLOC_FORCE
;
3725 static inline u64
calc_global_rsv_need_space(struct btrfs_block_rsv
*global
)
3727 return (global
->size
<< 1);
3730 static int should_alloc_chunk(struct btrfs_root
*root
,
3731 struct btrfs_space_info
*sinfo
, int force
)
3733 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
3734 u64 num_bytes
= sinfo
->total_bytes
- sinfo
->bytes_readonly
;
3735 u64 num_allocated
= sinfo
->bytes_used
+ sinfo
->bytes_reserved
;
3738 if (force
== CHUNK_ALLOC_FORCE
)
3742 * We need to take into account the global rsv because for all intents
3743 * and purposes it's used space. Don't worry about locking the
3744 * global_rsv, it doesn't change except when the transaction commits.
3746 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
3747 num_allocated
+= calc_global_rsv_need_space(global_rsv
);
3750 * in limited mode, we want to have some free space up to
3751 * about 1% of the FS size.
3753 if (force
== CHUNK_ALLOC_LIMITED
) {
3754 thresh
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
3755 thresh
= max_t(u64
, 64 * 1024 * 1024,
3756 div_factor_fine(thresh
, 1));
3758 if (num_bytes
- num_allocated
< thresh
)
3762 if (num_allocated
+ 2 * 1024 * 1024 < div_factor(num_bytes
, 8))
3767 static u64
get_system_chunk_thresh(struct btrfs_root
*root
, u64 type
)
3771 if (type
& (BTRFS_BLOCK_GROUP_RAID10
|
3772 BTRFS_BLOCK_GROUP_RAID0
|
3773 BTRFS_BLOCK_GROUP_RAID5
|
3774 BTRFS_BLOCK_GROUP_RAID6
))
3775 num_dev
= root
->fs_info
->fs_devices
->rw_devices
;
3776 else if (type
& BTRFS_BLOCK_GROUP_RAID1
)
3779 num_dev
= 1; /* DUP or single */
3781 /* metadata for updaing devices and chunk tree */
3782 return btrfs_calc_trans_metadata_size(root
, num_dev
+ 1);
3785 static void check_system_chunk(struct btrfs_trans_handle
*trans
,
3786 struct btrfs_root
*root
, u64 type
)
3788 struct btrfs_space_info
*info
;
3792 info
= __find_space_info(root
->fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
3793 spin_lock(&info
->lock
);
3794 left
= info
->total_bytes
- info
->bytes_used
- info
->bytes_pinned
-
3795 info
->bytes_reserved
- info
->bytes_readonly
;
3796 spin_unlock(&info
->lock
);
3798 thresh
= get_system_chunk_thresh(root
, type
);
3799 if (left
< thresh
&& btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
3800 btrfs_info(root
->fs_info
, "left=%llu, need=%llu, flags=%llu",
3801 left
, thresh
, type
);
3802 dump_space_info(info
, 0, 0);
3805 if (left
< thresh
) {
3808 flags
= btrfs_get_alloc_profile(root
->fs_info
->chunk_root
, 0);
3809 btrfs_alloc_chunk(trans
, root
, flags
);
3813 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
3814 struct btrfs_root
*extent_root
, u64 flags
, int force
)
3816 struct btrfs_space_info
*space_info
;
3817 struct btrfs_fs_info
*fs_info
= extent_root
->fs_info
;
3818 int wait_for_alloc
= 0;
3821 /* Don't re-enter if we're already allocating a chunk */
3822 if (trans
->allocating_chunk
)
3825 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
3827 ret
= update_space_info(extent_root
->fs_info
, flags
,
3829 BUG_ON(ret
); /* -ENOMEM */
3831 BUG_ON(!space_info
); /* Logic error */
3834 spin_lock(&space_info
->lock
);
3835 if (force
< space_info
->force_alloc
)
3836 force
= space_info
->force_alloc
;
3837 if (space_info
->full
) {
3838 if (should_alloc_chunk(extent_root
, space_info
, force
))
3842 spin_unlock(&space_info
->lock
);
3846 if (!should_alloc_chunk(extent_root
, space_info
, force
)) {
3847 spin_unlock(&space_info
->lock
);
3849 } else if (space_info
->chunk_alloc
) {
3852 space_info
->chunk_alloc
= 1;
3855 spin_unlock(&space_info
->lock
);
3857 mutex_lock(&fs_info
->chunk_mutex
);
3860 * The chunk_mutex is held throughout the entirety of a chunk
3861 * allocation, so once we've acquired the chunk_mutex we know that the
3862 * other guy is done and we need to recheck and see if we should
3865 if (wait_for_alloc
) {
3866 mutex_unlock(&fs_info
->chunk_mutex
);
3871 trans
->allocating_chunk
= true;
3874 * If we have mixed data/metadata chunks we want to make sure we keep
3875 * allocating mixed chunks instead of individual chunks.
3877 if (btrfs_mixed_space_info(space_info
))
3878 flags
|= (BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
);
3881 * if we're doing a data chunk, go ahead and make sure that
3882 * we keep a reasonable number of metadata chunks allocated in the
3885 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
3886 fs_info
->data_chunk_allocations
++;
3887 if (!(fs_info
->data_chunk_allocations
%
3888 fs_info
->metadata_ratio
))
3889 force_metadata_allocation(fs_info
);
3893 * Check if we have enough space in SYSTEM chunk because we may need
3894 * to update devices.
3896 check_system_chunk(trans
, extent_root
, flags
);
3898 ret
= btrfs_alloc_chunk(trans
, extent_root
, flags
);
3899 trans
->allocating_chunk
= false;
3901 spin_lock(&space_info
->lock
);
3902 if (ret
< 0 && ret
!= -ENOSPC
)
3905 space_info
->full
= 1;
3909 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3911 space_info
->chunk_alloc
= 0;
3912 spin_unlock(&space_info
->lock
);
3913 mutex_unlock(&fs_info
->chunk_mutex
);
3917 static int can_overcommit(struct btrfs_root
*root
,
3918 struct btrfs_space_info
*space_info
, u64 bytes
,
3919 enum btrfs_reserve_flush_enum flush
)
3921 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
3922 u64 profile
= btrfs_get_alloc_profile(root
, 0);
3927 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
3928 space_info
->bytes_pinned
+ space_info
->bytes_readonly
;
3931 * We only want to allow over committing if we have lots of actual space
3932 * free, but if we don't have enough space to handle the global reserve
3933 * space then we could end up having a real enospc problem when trying
3934 * to allocate a chunk or some other such important allocation.
3936 spin_lock(&global_rsv
->lock
);
3937 space_size
= calc_global_rsv_need_space(global_rsv
);
3938 spin_unlock(&global_rsv
->lock
);
3939 if (used
+ space_size
>= space_info
->total_bytes
)
3942 used
+= space_info
->bytes_may_use
;
3944 spin_lock(&root
->fs_info
->free_chunk_lock
);
3945 avail
= root
->fs_info
->free_chunk_space
;
3946 spin_unlock(&root
->fs_info
->free_chunk_lock
);
3949 * If we have dup, raid1 or raid10 then only half of the free
3950 * space is actually useable. For raid56, the space info used
3951 * doesn't include the parity drive, so we don't have to
3954 if (profile
& (BTRFS_BLOCK_GROUP_DUP
|
3955 BTRFS_BLOCK_GROUP_RAID1
|
3956 BTRFS_BLOCK_GROUP_RAID10
))
3960 * If we aren't flushing all things, let us overcommit up to
3961 * 1/2th of the space. If we can flush, don't let us overcommit
3962 * too much, let it overcommit up to 1/8 of the space.
3964 if (flush
== BTRFS_RESERVE_FLUSH_ALL
)
3969 if (used
+ bytes
< space_info
->total_bytes
+ avail
)
3974 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root
*root
,
3975 unsigned long nr_pages
)
3977 struct super_block
*sb
= root
->fs_info
->sb
;
3979 if (down_read_trylock(&sb
->s_umount
)) {
3980 writeback_inodes_sb_nr(sb
, nr_pages
, WB_REASON_FS_FREE_SPACE
);
3981 up_read(&sb
->s_umount
);
3984 * We needn't worry the filesystem going from r/w to r/o though
3985 * we don't acquire ->s_umount mutex, because the filesystem
3986 * should guarantee the delalloc inodes list be empty after
3987 * the filesystem is readonly(all dirty pages are written to
3990 btrfs_start_all_delalloc_inodes(root
->fs_info
, 0);
3991 if (!current
->journal_info
)
3992 btrfs_wait_all_ordered_extents(root
->fs_info
);
3997 * shrink metadata reservation for delalloc
3999 static void shrink_delalloc(struct btrfs_root
*root
, u64 to_reclaim
, u64 orig
,
4002 struct btrfs_block_rsv
*block_rsv
;
4003 struct btrfs_space_info
*space_info
;
4004 struct btrfs_trans_handle
*trans
;
4008 unsigned long nr_pages
= (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT
;
4010 enum btrfs_reserve_flush_enum flush
;
4012 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4013 block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
4014 space_info
= block_rsv
->space_info
;
4017 delalloc_bytes
= percpu_counter_sum_positive(
4018 &root
->fs_info
->delalloc_bytes
);
4019 if (delalloc_bytes
== 0) {
4022 btrfs_wait_all_ordered_extents(root
->fs_info
);
4026 while (delalloc_bytes
&& loops
< 3) {
4027 max_reclaim
= min(delalloc_bytes
, to_reclaim
);
4028 nr_pages
= max_reclaim
>> PAGE_CACHE_SHIFT
;
4029 btrfs_writeback_inodes_sb_nr(root
, nr_pages
);
4031 * We need to wait for the async pages to actually start before
4034 wait_event(root
->fs_info
->async_submit_wait
,
4035 !atomic_read(&root
->fs_info
->async_delalloc_pages
));
4038 flush
= BTRFS_RESERVE_FLUSH_ALL
;
4040 flush
= BTRFS_RESERVE_NO_FLUSH
;
4041 spin_lock(&space_info
->lock
);
4042 if (can_overcommit(root
, space_info
, orig
, flush
)) {
4043 spin_unlock(&space_info
->lock
);
4046 spin_unlock(&space_info
->lock
);
4049 if (wait_ordered
&& !trans
) {
4050 btrfs_wait_all_ordered_extents(root
->fs_info
);
4052 time_left
= schedule_timeout_killable(1);
4057 delalloc_bytes
= percpu_counter_sum_positive(
4058 &root
->fs_info
->delalloc_bytes
);
4063 * maybe_commit_transaction - possibly commit the transaction if its ok to
4064 * @root - the root we're allocating for
4065 * @bytes - the number of bytes we want to reserve
4066 * @force - force the commit
4068 * This will check to make sure that committing the transaction will actually
4069 * get us somewhere and then commit the transaction if it does. Otherwise it
4070 * will return -ENOSPC.
4072 static int may_commit_transaction(struct btrfs_root
*root
,
4073 struct btrfs_space_info
*space_info
,
4074 u64 bytes
, int force
)
4076 struct btrfs_block_rsv
*delayed_rsv
= &root
->fs_info
->delayed_block_rsv
;
4077 struct btrfs_trans_handle
*trans
;
4079 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4086 /* See if there is enough pinned space to make this reservation */
4087 spin_lock(&space_info
->lock
);
4088 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4090 spin_unlock(&space_info
->lock
);
4093 spin_unlock(&space_info
->lock
);
4096 * See if there is some space in the delayed insertion reservation for
4099 if (space_info
!= delayed_rsv
->space_info
)
4102 spin_lock(&space_info
->lock
);
4103 spin_lock(&delayed_rsv
->lock
);
4104 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4105 bytes
- delayed_rsv
->size
) >= 0) {
4106 spin_unlock(&delayed_rsv
->lock
);
4107 spin_unlock(&space_info
->lock
);
4110 spin_unlock(&delayed_rsv
->lock
);
4111 spin_unlock(&space_info
->lock
);
4114 trans
= btrfs_join_transaction(root
);
4118 return btrfs_commit_transaction(trans
, root
);
4122 FLUSH_DELAYED_ITEMS_NR
= 1,
4123 FLUSH_DELAYED_ITEMS
= 2,
4125 FLUSH_DELALLOC_WAIT
= 4,
4130 static int flush_space(struct btrfs_root
*root
,
4131 struct btrfs_space_info
*space_info
, u64 num_bytes
,
4132 u64 orig_bytes
, int state
)
4134 struct btrfs_trans_handle
*trans
;
4139 case FLUSH_DELAYED_ITEMS_NR
:
4140 case FLUSH_DELAYED_ITEMS
:
4141 if (state
== FLUSH_DELAYED_ITEMS_NR
) {
4142 u64 bytes
= btrfs_calc_trans_metadata_size(root
, 1);
4144 nr
= (int)div64_u64(num_bytes
, bytes
);
4151 trans
= btrfs_join_transaction(root
);
4152 if (IS_ERR(trans
)) {
4153 ret
= PTR_ERR(trans
);
4156 ret
= btrfs_run_delayed_items_nr(trans
, root
, nr
);
4157 btrfs_end_transaction(trans
, root
);
4159 case FLUSH_DELALLOC
:
4160 case FLUSH_DELALLOC_WAIT
:
4161 shrink_delalloc(root
, num_bytes
, orig_bytes
,
4162 state
== FLUSH_DELALLOC_WAIT
);
4165 trans
= btrfs_join_transaction(root
);
4166 if (IS_ERR(trans
)) {
4167 ret
= PTR_ERR(trans
);
4170 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
4171 btrfs_get_alloc_profile(root
, 0),
4172 CHUNK_ALLOC_NO_FORCE
);
4173 btrfs_end_transaction(trans
, root
);
4178 ret
= may_commit_transaction(root
, space_info
, orig_bytes
, 0);
4188 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4189 * @root - the root we're allocating for
4190 * @block_rsv - the block_rsv we're allocating for
4191 * @orig_bytes - the number of bytes we want
4192 * @flush - whether or not we can flush to make our reservation
4194 * This will reserve orgi_bytes number of bytes from the space info associated
4195 * with the block_rsv. If there is not enough space it will make an attempt to
4196 * flush out space to make room. It will do this by flushing delalloc if
4197 * possible or committing the transaction. If flush is 0 then no attempts to
4198 * regain reservations will be made and this will fail if there is not enough
4201 static int reserve_metadata_bytes(struct btrfs_root
*root
,
4202 struct btrfs_block_rsv
*block_rsv
,
4204 enum btrfs_reserve_flush_enum flush
)
4206 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
4208 u64 num_bytes
= orig_bytes
;
4209 int flush_state
= FLUSH_DELAYED_ITEMS_NR
;
4211 bool flushing
= false;
4215 spin_lock(&space_info
->lock
);
4217 * We only want to wait if somebody other than us is flushing and we
4218 * are actually allowed to flush all things.
4220 while (flush
== BTRFS_RESERVE_FLUSH_ALL
&& !flushing
&&
4221 space_info
->flush
) {
4222 spin_unlock(&space_info
->lock
);
4224 * If we have a trans handle we can't wait because the flusher
4225 * may have to commit the transaction, which would mean we would
4226 * deadlock since we are waiting for the flusher to finish, but
4227 * hold the current transaction open.
4229 if (current
->journal_info
)
4231 ret
= wait_event_killable(space_info
->wait
, !space_info
->flush
);
4232 /* Must have been killed, return */
4236 spin_lock(&space_info
->lock
);
4240 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4241 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4242 space_info
->bytes_may_use
;
4245 * The idea here is that we've not already over-reserved the block group
4246 * then we can go ahead and save our reservation first and then start
4247 * flushing if we need to. Otherwise if we've already overcommitted
4248 * lets start flushing stuff first and then come back and try to make
4251 if (used
<= space_info
->total_bytes
) {
4252 if (used
+ orig_bytes
<= space_info
->total_bytes
) {
4253 space_info
->bytes_may_use
+= orig_bytes
;
4254 trace_btrfs_space_reservation(root
->fs_info
,
4255 "space_info", space_info
->flags
, orig_bytes
, 1);
4259 * Ok set num_bytes to orig_bytes since we aren't
4260 * overocmmitted, this way we only try and reclaim what
4263 num_bytes
= orig_bytes
;
4267 * Ok we're over committed, set num_bytes to the overcommitted
4268 * amount plus the amount of bytes that we need for this
4271 num_bytes
= used
- space_info
->total_bytes
+
4275 if (ret
&& can_overcommit(root
, space_info
, orig_bytes
, flush
)) {
4276 space_info
->bytes_may_use
+= orig_bytes
;
4277 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
4278 space_info
->flags
, orig_bytes
,
4284 * Couldn't make our reservation, save our place so while we're trying
4285 * to reclaim space we can actually use it instead of somebody else
4286 * stealing it from us.
4288 * We make the other tasks wait for the flush only when we can flush
4291 if (ret
&& flush
!= BTRFS_RESERVE_NO_FLUSH
) {
4293 space_info
->flush
= 1;
4296 spin_unlock(&space_info
->lock
);
4298 if (!ret
|| flush
== BTRFS_RESERVE_NO_FLUSH
)
4301 ret
= flush_space(root
, space_info
, num_bytes
, orig_bytes
,
4306 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4307 * would happen. So skip delalloc flush.
4309 if (flush
== BTRFS_RESERVE_FLUSH_LIMIT
&&
4310 (flush_state
== FLUSH_DELALLOC
||
4311 flush_state
== FLUSH_DELALLOC_WAIT
))
4312 flush_state
= ALLOC_CHUNK
;
4316 else if (flush
== BTRFS_RESERVE_FLUSH_LIMIT
&&
4317 flush_state
< COMMIT_TRANS
)
4319 else if (flush
== BTRFS_RESERVE_FLUSH_ALL
&&
4320 flush_state
<= COMMIT_TRANS
)
4324 if (ret
== -ENOSPC
&&
4325 unlikely(root
->orphan_cleanup_state
== ORPHAN_CLEANUP_STARTED
)) {
4326 struct btrfs_block_rsv
*global_rsv
=
4327 &root
->fs_info
->global_block_rsv
;
4329 if (block_rsv
!= global_rsv
&&
4330 !block_rsv_use_bytes(global_rsv
, orig_bytes
))
4334 spin_lock(&space_info
->lock
);
4335 space_info
->flush
= 0;
4336 wake_up_all(&space_info
->wait
);
4337 spin_unlock(&space_info
->lock
);
4342 static struct btrfs_block_rsv
*get_block_rsv(
4343 const struct btrfs_trans_handle
*trans
,
4344 const struct btrfs_root
*root
)
4346 struct btrfs_block_rsv
*block_rsv
= NULL
;
4349 block_rsv
= trans
->block_rsv
;
4351 if (root
== root
->fs_info
->csum_root
&& trans
->adding_csums
)
4352 block_rsv
= trans
->block_rsv
;
4354 if (root
== root
->fs_info
->uuid_root
)
4355 block_rsv
= trans
->block_rsv
;
4358 block_rsv
= root
->block_rsv
;
4361 block_rsv
= &root
->fs_info
->empty_block_rsv
;
4366 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
4370 spin_lock(&block_rsv
->lock
);
4371 if (block_rsv
->reserved
>= num_bytes
) {
4372 block_rsv
->reserved
-= num_bytes
;
4373 if (block_rsv
->reserved
< block_rsv
->size
)
4374 block_rsv
->full
= 0;
4377 spin_unlock(&block_rsv
->lock
);
4381 static void block_rsv_add_bytes(struct btrfs_block_rsv
*block_rsv
,
4382 u64 num_bytes
, int update_size
)
4384 spin_lock(&block_rsv
->lock
);
4385 block_rsv
->reserved
+= num_bytes
;
4387 block_rsv
->size
+= num_bytes
;
4388 else if (block_rsv
->reserved
>= block_rsv
->size
)
4389 block_rsv
->full
= 1;
4390 spin_unlock(&block_rsv
->lock
);
4393 int btrfs_cond_migrate_bytes(struct btrfs_fs_info
*fs_info
,
4394 struct btrfs_block_rsv
*dest
, u64 num_bytes
,
4397 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
4400 if (global_rsv
->space_info
!= dest
->space_info
)
4403 spin_lock(&global_rsv
->lock
);
4404 min_bytes
= div_factor(global_rsv
->size
, min_factor
);
4405 if (global_rsv
->reserved
< min_bytes
+ num_bytes
) {
4406 spin_unlock(&global_rsv
->lock
);
4409 global_rsv
->reserved
-= num_bytes
;
4410 if (global_rsv
->reserved
< global_rsv
->size
)
4411 global_rsv
->full
= 0;
4412 spin_unlock(&global_rsv
->lock
);
4414 block_rsv_add_bytes(dest
, num_bytes
, 1);
4418 static void block_rsv_release_bytes(struct btrfs_fs_info
*fs_info
,
4419 struct btrfs_block_rsv
*block_rsv
,
4420 struct btrfs_block_rsv
*dest
, u64 num_bytes
)
4422 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
4424 spin_lock(&block_rsv
->lock
);
4425 if (num_bytes
== (u64
)-1)
4426 num_bytes
= block_rsv
->size
;
4427 block_rsv
->size
-= num_bytes
;
4428 if (block_rsv
->reserved
>= block_rsv
->size
) {
4429 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
4430 block_rsv
->reserved
= block_rsv
->size
;
4431 block_rsv
->full
= 1;
4435 spin_unlock(&block_rsv
->lock
);
4437 if (num_bytes
> 0) {
4439 spin_lock(&dest
->lock
);
4443 bytes_to_add
= dest
->size
- dest
->reserved
;
4444 bytes_to_add
= min(num_bytes
, bytes_to_add
);
4445 dest
->reserved
+= bytes_to_add
;
4446 if (dest
->reserved
>= dest
->size
)
4448 num_bytes
-= bytes_to_add
;
4450 spin_unlock(&dest
->lock
);
4453 spin_lock(&space_info
->lock
);
4454 space_info
->bytes_may_use
-= num_bytes
;
4455 trace_btrfs_space_reservation(fs_info
, "space_info",
4456 space_info
->flags
, num_bytes
, 0);
4457 spin_unlock(&space_info
->lock
);
4462 static int block_rsv_migrate_bytes(struct btrfs_block_rsv
*src
,
4463 struct btrfs_block_rsv
*dst
, u64 num_bytes
)
4467 ret
= block_rsv_use_bytes(src
, num_bytes
);
4471 block_rsv_add_bytes(dst
, num_bytes
, 1);
4475 void btrfs_init_block_rsv(struct btrfs_block_rsv
*rsv
, unsigned short type
)
4477 memset(rsv
, 0, sizeof(*rsv
));
4478 spin_lock_init(&rsv
->lock
);
4482 struct btrfs_block_rsv
*btrfs_alloc_block_rsv(struct btrfs_root
*root
,
4483 unsigned short type
)
4485 struct btrfs_block_rsv
*block_rsv
;
4486 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4488 block_rsv
= kmalloc(sizeof(*block_rsv
), GFP_NOFS
);
4492 btrfs_init_block_rsv(block_rsv
, type
);
4493 block_rsv
->space_info
= __find_space_info(fs_info
,
4494 BTRFS_BLOCK_GROUP_METADATA
);
4498 void btrfs_free_block_rsv(struct btrfs_root
*root
,
4499 struct btrfs_block_rsv
*rsv
)
4503 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
4507 int btrfs_block_rsv_add(struct btrfs_root
*root
,
4508 struct btrfs_block_rsv
*block_rsv
, u64 num_bytes
,
4509 enum btrfs_reserve_flush_enum flush
)
4516 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
4518 block_rsv_add_bytes(block_rsv
, num_bytes
, 1);
4525 int btrfs_block_rsv_check(struct btrfs_root
*root
,
4526 struct btrfs_block_rsv
*block_rsv
, int min_factor
)
4534 spin_lock(&block_rsv
->lock
);
4535 num_bytes
= div_factor(block_rsv
->size
, min_factor
);
4536 if (block_rsv
->reserved
>= num_bytes
)
4538 spin_unlock(&block_rsv
->lock
);
4543 int btrfs_block_rsv_refill(struct btrfs_root
*root
,
4544 struct btrfs_block_rsv
*block_rsv
, u64 min_reserved
,
4545 enum btrfs_reserve_flush_enum flush
)
4553 spin_lock(&block_rsv
->lock
);
4554 num_bytes
= min_reserved
;
4555 if (block_rsv
->reserved
>= num_bytes
)
4558 num_bytes
-= block_rsv
->reserved
;
4559 spin_unlock(&block_rsv
->lock
);
4564 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
4566 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
4573 int btrfs_block_rsv_migrate(struct btrfs_block_rsv
*src_rsv
,
4574 struct btrfs_block_rsv
*dst_rsv
,
4577 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
4580 void btrfs_block_rsv_release(struct btrfs_root
*root
,
4581 struct btrfs_block_rsv
*block_rsv
,
4584 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
4585 if (global_rsv
->full
|| global_rsv
== block_rsv
||
4586 block_rsv
->space_info
!= global_rsv
->space_info
)
4588 block_rsv_release_bytes(root
->fs_info
, block_rsv
, global_rsv
,
4593 * helper to calculate size of global block reservation.
4594 * the desired value is sum of space used by extent tree,
4595 * checksum tree and root tree
4597 static u64
calc_global_metadata_size(struct btrfs_fs_info
*fs_info
)
4599 struct btrfs_space_info
*sinfo
;
4603 int csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
4605 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_DATA
);
4606 spin_lock(&sinfo
->lock
);
4607 data_used
= sinfo
->bytes_used
;
4608 spin_unlock(&sinfo
->lock
);
4610 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4611 spin_lock(&sinfo
->lock
);
4612 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_DATA
)
4614 meta_used
= sinfo
->bytes_used
;
4615 spin_unlock(&sinfo
->lock
);
4617 num_bytes
= (data_used
>> fs_info
->sb
->s_blocksize_bits
) *
4619 num_bytes
+= div64_u64(data_used
+ meta_used
, 50);
4621 if (num_bytes
* 3 > meta_used
)
4622 num_bytes
= div64_u64(meta_used
, 3);
4624 return ALIGN(num_bytes
, fs_info
->extent_root
->leafsize
<< 10);
4627 static void update_global_block_rsv(struct btrfs_fs_info
*fs_info
)
4629 struct btrfs_block_rsv
*block_rsv
= &fs_info
->global_block_rsv
;
4630 struct btrfs_space_info
*sinfo
= block_rsv
->space_info
;
4633 num_bytes
= calc_global_metadata_size(fs_info
);
4635 spin_lock(&sinfo
->lock
);
4636 spin_lock(&block_rsv
->lock
);
4638 block_rsv
->size
= min_t(u64
, num_bytes
, 512 * 1024 * 1024);
4640 num_bytes
= sinfo
->bytes_used
+ sinfo
->bytes_pinned
+
4641 sinfo
->bytes_reserved
+ sinfo
->bytes_readonly
+
4642 sinfo
->bytes_may_use
;
4644 if (sinfo
->total_bytes
> num_bytes
) {
4645 num_bytes
= sinfo
->total_bytes
- num_bytes
;
4646 block_rsv
->reserved
+= num_bytes
;
4647 sinfo
->bytes_may_use
+= num_bytes
;
4648 trace_btrfs_space_reservation(fs_info
, "space_info",
4649 sinfo
->flags
, num_bytes
, 1);
4652 if (block_rsv
->reserved
>= block_rsv
->size
) {
4653 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
4654 sinfo
->bytes_may_use
-= num_bytes
;
4655 trace_btrfs_space_reservation(fs_info
, "space_info",
4656 sinfo
->flags
, num_bytes
, 0);
4657 block_rsv
->reserved
= block_rsv
->size
;
4658 block_rsv
->full
= 1;
4661 spin_unlock(&block_rsv
->lock
);
4662 spin_unlock(&sinfo
->lock
);
4665 static void init_global_block_rsv(struct btrfs_fs_info
*fs_info
)
4667 struct btrfs_space_info
*space_info
;
4669 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
4670 fs_info
->chunk_block_rsv
.space_info
= space_info
;
4672 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4673 fs_info
->global_block_rsv
.space_info
= space_info
;
4674 fs_info
->delalloc_block_rsv
.space_info
= space_info
;
4675 fs_info
->trans_block_rsv
.space_info
= space_info
;
4676 fs_info
->empty_block_rsv
.space_info
= space_info
;
4677 fs_info
->delayed_block_rsv
.space_info
= space_info
;
4679 fs_info
->extent_root
->block_rsv
= &fs_info
->global_block_rsv
;
4680 fs_info
->csum_root
->block_rsv
= &fs_info
->global_block_rsv
;
4681 fs_info
->dev_root
->block_rsv
= &fs_info
->global_block_rsv
;
4682 fs_info
->tree_root
->block_rsv
= &fs_info
->global_block_rsv
;
4683 if (fs_info
->quota_root
)
4684 fs_info
->quota_root
->block_rsv
= &fs_info
->global_block_rsv
;
4685 fs_info
->chunk_root
->block_rsv
= &fs_info
->chunk_block_rsv
;
4687 update_global_block_rsv(fs_info
);
4690 static void release_global_block_rsv(struct btrfs_fs_info
*fs_info
)
4692 block_rsv_release_bytes(fs_info
, &fs_info
->global_block_rsv
, NULL
,
4694 WARN_ON(fs_info
->delalloc_block_rsv
.size
> 0);
4695 WARN_ON(fs_info
->delalloc_block_rsv
.reserved
> 0);
4696 WARN_ON(fs_info
->trans_block_rsv
.size
> 0);
4697 WARN_ON(fs_info
->trans_block_rsv
.reserved
> 0);
4698 WARN_ON(fs_info
->chunk_block_rsv
.size
> 0);
4699 WARN_ON(fs_info
->chunk_block_rsv
.reserved
> 0);
4700 WARN_ON(fs_info
->delayed_block_rsv
.size
> 0);
4701 WARN_ON(fs_info
->delayed_block_rsv
.reserved
> 0);
4704 void btrfs_trans_release_metadata(struct btrfs_trans_handle
*trans
,
4705 struct btrfs_root
*root
)
4707 if (!trans
->block_rsv
)
4710 if (!trans
->bytes_reserved
)
4713 trace_btrfs_space_reservation(root
->fs_info
, "transaction",
4714 trans
->transid
, trans
->bytes_reserved
, 0);
4715 btrfs_block_rsv_release(root
, trans
->block_rsv
, trans
->bytes_reserved
);
4716 trans
->bytes_reserved
= 0;
4719 /* Can only return 0 or -ENOSPC */
4720 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle
*trans
,
4721 struct inode
*inode
)
4723 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4724 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
4725 struct btrfs_block_rsv
*dst_rsv
= root
->orphan_block_rsv
;
4728 * We need to hold space in order to delete our orphan item once we've
4729 * added it, so this takes the reservation so we can release it later
4730 * when we are truly done with the orphan item.
4732 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
4733 trace_btrfs_space_reservation(root
->fs_info
, "orphan",
4734 btrfs_ino(inode
), num_bytes
, 1);
4735 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
4738 void btrfs_orphan_release_metadata(struct inode
*inode
)
4740 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4741 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
4742 trace_btrfs_space_reservation(root
->fs_info
, "orphan",
4743 btrfs_ino(inode
), num_bytes
, 0);
4744 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
, num_bytes
);
4748 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4749 * root: the root of the parent directory
4750 * rsv: block reservation
4751 * items: the number of items that we need do reservation
4752 * qgroup_reserved: used to return the reserved size in qgroup
4754 * This function is used to reserve the space for snapshot/subvolume
4755 * creation and deletion. Those operations are different with the
4756 * common file/directory operations, they change two fs/file trees
4757 * and root tree, the number of items that the qgroup reserves is
4758 * different with the free space reservation. So we can not use
4759 * the space reseravtion mechanism in start_transaction().
4761 int btrfs_subvolume_reserve_metadata(struct btrfs_root
*root
,
4762 struct btrfs_block_rsv
*rsv
,
4764 u64
*qgroup_reserved
,
4765 bool use_global_rsv
)
4769 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
4771 if (root
->fs_info
->quota_enabled
) {
4772 /* One for parent inode, two for dir entries */
4773 num_bytes
= 3 * root
->leafsize
;
4774 ret
= btrfs_qgroup_reserve(root
, num_bytes
);
4781 *qgroup_reserved
= num_bytes
;
4783 num_bytes
= btrfs_calc_trans_metadata_size(root
, items
);
4784 rsv
->space_info
= __find_space_info(root
->fs_info
,
4785 BTRFS_BLOCK_GROUP_METADATA
);
4786 ret
= btrfs_block_rsv_add(root
, rsv
, num_bytes
,
4787 BTRFS_RESERVE_FLUSH_ALL
);
4789 if (ret
== -ENOSPC
&& use_global_rsv
)
4790 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
, num_bytes
);
4793 if (*qgroup_reserved
)
4794 btrfs_qgroup_free(root
, *qgroup_reserved
);
4800 void btrfs_subvolume_release_metadata(struct btrfs_root
*root
,
4801 struct btrfs_block_rsv
*rsv
,
4802 u64 qgroup_reserved
)
4804 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
4805 if (qgroup_reserved
)
4806 btrfs_qgroup_free(root
, qgroup_reserved
);
4810 * drop_outstanding_extent - drop an outstanding extent
4811 * @inode: the inode we're dropping the extent for
4813 * This is called when we are freeing up an outstanding extent, either called
4814 * after an error or after an extent is written. This will return the number of
4815 * reserved extents that need to be freed. This must be called with
4816 * BTRFS_I(inode)->lock held.
4818 static unsigned drop_outstanding_extent(struct inode
*inode
)
4820 unsigned drop_inode_space
= 0;
4821 unsigned dropped_extents
= 0;
4823 BUG_ON(!BTRFS_I(inode
)->outstanding_extents
);
4824 BTRFS_I(inode
)->outstanding_extents
--;
4826 if (BTRFS_I(inode
)->outstanding_extents
== 0 &&
4827 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
4828 &BTRFS_I(inode
)->runtime_flags
))
4829 drop_inode_space
= 1;
4832 * If we have more or the same amount of outsanding extents than we have
4833 * reserved then we need to leave the reserved extents count alone.
4835 if (BTRFS_I(inode
)->outstanding_extents
>=
4836 BTRFS_I(inode
)->reserved_extents
)
4837 return drop_inode_space
;
4839 dropped_extents
= BTRFS_I(inode
)->reserved_extents
-
4840 BTRFS_I(inode
)->outstanding_extents
;
4841 BTRFS_I(inode
)->reserved_extents
-= dropped_extents
;
4842 return dropped_extents
+ drop_inode_space
;
4846 * calc_csum_metadata_size - return the amount of metada space that must be
4847 * reserved/free'd for the given bytes.
4848 * @inode: the inode we're manipulating
4849 * @num_bytes: the number of bytes in question
4850 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4852 * This adjusts the number of csum_bytes in the inode and then returns the
4853 * correct amount of metadata that must either be reserved or freed. We
4854 * calculate how many checksums we can fit into one leaf and then divide the
4855 * number of bytes that will need to be checksumed by this value to figure out
4856 * how many checksums will be required. If we are adding bytes then the number
4857 * may go up and we will return the number of additional bytes that must be
4858 * reserved. If it is going down we will return the number of bytes that must
4861 * This must be called with BTRFS_I(inode)->lock held.
4863 static u64
calc_csum_metadata_size(struct inode
*inode
, u64 num_bytes
,
4866 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4868 int num_csums_per_leaf
;
4872 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
&&
4873 BTRFS_I(inode
)->csum_bytes
== 0)
4876 old_csums
= (int)div64_u64(BTRFS_I(inode
)->csum_bytes
, root
->sectorsize
);
4878 BTRFS_I(inode
)->csum_bytes
+= num_bytes
;
4880 BTRFS_I(inode
)->csum_bytes
-= num_bytes
;
4881 csum_size
= BTRFS_LEAF_DATA_SIZE(root
) - sizeof(struct btrfs_item
);
4882 num_csums_per_leaf
= (int)div64_u64(csum_size
,
4883 sizeof(struct btrfs_csum_item
) +
4884 sizeof(struct btrfs_disk_key
));
4885 num_csums
= (int)div64_u64(BTRFS_I(inode
)->csum_bytes
, root
->sectorsize
);
4886 num_csums
= num_csums
+ num_csums_per_leaf
- 1;
4887 num_csums
= num_csums
/ num_csums_per_leaf
;
4889 old_csums
= old_csums
+ num_csums_per_leaf
- 1;
4890 old_csums
= old_csums
/ num_csums_per_leaf
;
4892 /* No change, no need to reserve more */
4893 if (old_csums
== num_csums
)
4897 return btrfs_calc_trans_metadata_size(root
,
4898 num_csums
- old_csums
);
4900 return btrfs_calc_trans_metadata_size(root
, old_csums
- num_csums
);
4903 int btrfs_delalloc_reserve_metadata(struct inode
*inode
, u64 num_bytes
)
4905 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4906 struct btrfs_block_rsv
*block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
4909 unsigned nr_extents
= 0;
4910 int extra_reserve
= 0;
4911 enum btrfs_reserve_flush_enum flush
= BTRFS_RESERVE_FLUSH_ALL
;
4913 bool delalloc_lock
= true;
4917 /* If we are a free space inode we need to not flush since we will be in
4918 * the middle of a transaction commit. We also don't need the delalloc
4919 * mutex since we won't race with anybody. We need this mostly to make
4920 * lockdep shut its filthy mouth.
4922 if (btrfs_is_free_space_inode(inode
)) {
4923 flush
= BTRFS_RESERVE_NO_FLUSH
;
4924 delalloc_lock
= false;
4927 if (flush
!= BTRFS_RESERVE_NO_FLUSH
&&
4928 btrfs_transaction_in_commit(root
->fs_info
))
4929 schedule_timeout(1);
4932 mutex_lock(&BTRFS_I(inode
)->delalloc_mutex
);
4934 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
4936 spin_lock(&BTRFS_I(inode
)->lock
);
4937 BTRFS_I(inode
)->outstanding_extents
++;
4939 if (BTRFS_I(inode
)->outstanding_extents
>
4940 BTRFS_I(inode
)->reserved_extents
)
4941 nr_extents
= BTRFS_I(inode
)->outstanding_extents
-
4942 BTRFS_I(inode
)->reserved_extents
;
4945 * Add an item to reserve for updating the inode when we complete the
4948 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
4949 &BTRFS_I(inode
)->runtime_flags
)) {
4954 to_reserve
= btrfs_calc_trans_metadata_size(root
, nr_extents
);
4955 to_reserve
+= calc_csum_metadata_size(inode
, num_bytes
, 1);
4956 csum_bytes
= BTRFS_I(inode
)->csum_bytes
;
4957 spin_unlock(&BTRFS_I(inode
)->lock
);
4959 if (root
->fs_info
->quota_enabled
) {
4960 ret
= btrfs_qgroup_reserve(root
, num_bytes
+
4961 nr_extents
* root
->leafsize
);
4966 ret
= reserve_metadata_bytes(root
, block_rsv
, to_reserve
, flush
);
4967 if (unlikely(ret
)) {
4968 if (root
->fs_info
->quota_enabled
)
4969 btrfs_qgroup_free(root
, num_bytes
+
4970 nr_extents
* root
->leafsize
);
4974 spin_lock(&BTRFS_I(inode
)->lock
);
4975 if (extra_reserve
) {
4976 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
4977 &BTRFS_I(inode
)->runtime_flags
);
4980 BTRFS_I(inode
)->reserved_extents
+= nr_extents
;
4981 spin_unlock(&BTRFS_I(inode
)->lock
);
4984 mutex_unlock(&BTRFS_I(inode
)->delalloc_mutex
);
4987 trace_btrfs_space_reservation(root
->fs_info
,"delalloc",
4988 btrfs_ino(inode
), to_reserve
, 1);
4989 block_rsv_add_bytes(block_rsv
, to_reserve
, 1);
4994 spin_lock(&BTRFS_I(inode
)->lock
);
4995 dropped
= drop_outstanding_extent(inode
);
4997 * If the inodes csum_bytes is the same as the original
4998 * csum_bytes then we know we haven't raced with any free()ers
4999 * so we can just reduce our inodes csum bytes and carry on.
5001 if (BTRFS_I(inode
)->csum_bytes
== csum_bytes
) {
5002 calc_csum_metadata_size(inode
, num_bytes
, 0);
5004 u64 orig_csum_bytes
= BTRFS_I(inode
)->csum_bytes
;
5008 * This is tricky, but first we need to figure out how much we
5009 * free'd from any free-ers that occured during this
5010 * reservation, so we reset ->csum_bytes to the csum_bytes
5011 * before we dropped our lock, and then call the free for the
5012 * number of bytes that were freed while we were trying our
5015 bytes
= csum_bytes
- BTRFS_I(inode
)->csum_bytes
;
5016 BTRFS_I(inode
)->csum_bytes
= csum_bytes
;
5017 to_free
= calc_csum_metadata_size(inode
, bytes
, 0);
5021 * Now we need to see how much we would have freed had we not
5022 * been making this reservation and our ->csum_bytes were not
5023 * artificially inflated.
5025 BTRFS_I(inode
)->csum_bytes
= csum_bytes
- num_bytes
;
5026 bytes
= csum_bytes
- orig_csum_bytes
;
5027 bytes
= calc_csum_metadata_size(inode
, bytes
, 0);
5030 * Now reset ->csum_bytes to what it should be. If bytes is
5031 * more than to_free then we would have free'd more space had we
5032 * not had an artificially high ->csum_bytes, so we need to free
5033 * the remainder. If bytes is the same or less then we don't
5034 * need to do anything, the other free-ers did the correct
5037 BTRFS_I(inode
)->csum_bytes
= orig_csum_bytes
- num_bytes
;
5038 if (bytes
> to_free
)
5039 to_free
= bytes
- to_free
;
5043 spin_unlock(&BTRFS_I(inode
)->lock
);
5045 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
5048 btrfs_block_rsv_release(root
, block_rsv
, to_free
);
5049 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5050 btrfs_ino(inode
), to_free
, 0);
5053 mutex_unlock(&BTRFS_I(inode
)->delalloc_mutex
);
5058 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5059 * @inode: the inode to release the reservation for
5060 * @num_bytes: the number of bytes we're releasing
5062 * This will release the metadata reservation for an inode. This can be called
5063 * once we complete IO for a given set of bytes to release their metadata
5066 void btrfs_delalloc_release_metadata(struct inode
*inode
, u64 num_bytes
)
5068 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5072 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
5073 spin_lock(&BTRFS_I(inode
)->lock
);
5074 dropped
= drop_outstanding_extent(inode
);
5077 to_free
= calc_csum_metadata_size(inode
, num_bytes
, 0);
5078 spin_unlock(&BTRFS_I(inode
)->lock
);
5080 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
5082 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5083 btrfs_ino(inode
), to_free
, 0);
5084 if (root
->fs_info
->quota_enabled
) {
5085 btrfs_qgroup_free(root
, num_bytes
+
5086 dropped
* root
->leafsize
);
5089 btrfs_block_rsv_release(root
, &root
->fs_info
->delalloc_block_rsv
,
5094 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5095 * @inode: inode we're writing to
5096 * @num_bytes: the number of bytes we want to allocate
5098 * This will do the following things
5100 * o reserve space in the data space info for num_bytes
5101 * o reserve space in the metadata space info based on number of outstanding
5102 * extents and how much csums will be needed
5103 * o add to the inodes ->delalloc_bytes
5104 * o add it to the fs_info's delalloc inodes list.
5106 * This will return 0 for success and -ENOSPC if there is no space left.
5108 int btrfs_delalloc_reserve_space(struct inode
*inode
, u64 num_bytes
)
5112 ret
= btrfs_check_data_free_space(inode
, num_bytes
);
5116 ret
= btrfs_delalloc_reserve_metadata(inode
, num_bytes
);
5118 btrfs_free_reserved_data_space(inode
, num_bytes
);
5126 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5127 * @inode: inode we're releasing space for
5128 * @num_bytes: the number of bytes we want to free up
5130 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5131 * called in the case that we don't need the metadata AND data reservations
5132 * anymore. So if there is an error or we insert an inline extent.
5134 * This function will release the metadata space that was not used and will
5135 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5136 * list if there are no delalloc bytes left.
5138 void btrfs_delalloc_release_space(struct inode
*inode
, u64 num_bytes
)
5140 btrfs_delalloc_release_metadata(inode
, num_bytes
);
5141 btrfs_free_reserved_data_space(inode
, num_bytes
);
5144 static int update_block_group(struct btrfs_root
*root
,
5145 u64 bytenr
, u64 num_bytes
, int alloc
)
5147 struct btrfs_block_group_cache
*cache
= NULL
;
5148 struct btrfs_fs_info
*info
= root
->fs_info
;
5149 u64 total
= num_bytes
;
5154 /* block accounting for super block */
5155 spin_lock(&info
->delalloc_root_lock
);
5156 old_val
= btrfs_super_bytes_used(info
->super_copy
);
5158 old_val
+= num_bytes
;
5160 old_val
-= num_bytes
;
5161 btrfs_set_super_bytes_used(info
->super_copy
, old_val
);
5162 spin_unlock(&info
->delalloc_root_lock
);
5165 cache
= btrfs_lookup_block_group(info
, bytenr
);
5168 if (cache
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
5169 BTRFS_BLOCK_GROUP_RAID1
|
5170 BTRFS_BLOCK_GROUP_RAID10
))
5175 * If this block group has free space cache written out, we
5176 * need to make sure to load it if we are removing space. This
5177 * is because we need the unpinning stage to actually add the
5178 * space back to the block group, otherwise we will leak space.
5180 if (!alloc
&& cache
->cached
== BTRFS_CACHE_NO
)
5181 cache_block_group(cache
, 1);
5183 byte_in_group
= bytenr
- cache
->key
.objectid
;
5184 WARN_ON(byte_in_group
> cache
->key
.offset
);
5186 spin_lock(&cache
->space_info
->lock
);
5187 spin_lock(&cache
->lock
);
5189 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
5190 cache
->disk_cache_state
< BTRFS_DC_CLEAR
)
5191 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
5194 old_val
= btrfs_block_group_used(&cache
->item
);
5195 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
5197 old_val
+= num_bytes
;
5198 btrfs_set_block_group_used(&cache
->item
, old_val
);
5199 cache
->reserved
-= num_bytes
;
5200 cache
->space_info
->bytes_reserved
-= num_bytes
;
5201 cache
->space_info
->bytes_used
+= num_bytes
;
5202 cache
->space_info
->disk_used
+= num_bytes
* factor
;
5203 spin_unlock(&cache
->lock
);
5204 spin_unlock(&cache
->space_info
->lock
);
5206 old_val
-= num_bytes
;
5207 btrfs_set_block_group_used(&cache
->item
, old_val
);
5208 cache
->pinned
+= num_bytes
;
5209 cache
->space_info
->bytes_pinned
+= num_bytes
;
5210 cache
->space_info
->bytes_used
-= num_bytes
;
5211 cache
->space_info
->disk_used
-= num_bytes
* factor
;
5212 spin_unlock(&cache
->lock
);
5213 spin_unlock(&cache
->space_info
->lock
);
5215 set_extent_dirty(info
->pinned_extents
,
5216 bytenr
, bytenr
+ num_bytes
- 1,
5217 GFP_NOFS
| __GFP_NOFAIL
);
5219 btrfs_put_block_group(cache
);
5221 bytenr
+= num_bytes
;
5226 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
5228 struct btrfs_block_group_cache
*cache
;
5231 spin_lock(&root
->fs_info
->block_group_cache_lock
);
5232 bytenr
= root
->fs_info
->first_logical_byte
;
5233 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
5235 if (bytenr
< (u64
)-1)
5238 cache
= btrfs_lookup_first_block_group(root
->fs_info
, search_start
);
5242 bytenr
= cache
->key
.objectid
;
5243 btrfs_put_block_group(cache
);
5248 static int pin_down_extent(struct btrfs_root
*root
,
5249 struct btrfs_block_group_cache
*cache
,
5250 u64 bytenr
, u64 num_bytes
, int reserved
)
5252 spin_lock(&cache
->space_info
->lock
);
5253 spin_lock(&cache
->lock
);
5254 cache
->pinned
+= num_bytes
;
5255 cache
->space_info
->bytes_pinned
+= num_bytes
;
5257 cache
->reserved
-= num_bytes
;
5258 cache
->space_info
->bytes_reserved
-= num_bytes
;
5260 spin_unlock(&cache
->lock
);
5261 spin_unlock(&cache
->space_info
->lock
);
5263 set_extent_dirty(root
->fs_info
->pinned_extents
, bytenr
,
5264 bytenr
+ num_bytes
- 1, GFP_NOFS
| __GFP_NOFAIL
);
5266 trace_btrfs_reserved_extent_free(root
, bytenr
, num_bytes
);
5271 * this function must be called within transaction
5273 int btrfs_pin_extent(struct btrfs_root
*root
,
5274 u64 bytenr
, u64 num_bytes
, int reserved
)
5276 struct btrfs_block_group_cache
*cache
;
5278 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
5279 BUG_ON(!cache
); /* Logic error */
5281 pin_down_extent(root
, cache
, bytenr
, num_bytes
, reserved
);
5283 btrfs_put_block_group(cache
);
5288 * this function must be called within transaction
5290 int btrfs_pin_extent_for_log_replay(struct btrfs_root
*root
,
5291 u64 bytenr
, u64 num_bytes
)
5293 struct btrfs_block_group_cache
*cache
;
5296 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
5301 * pull in the free space cache (if any) so that our pin
5302 * removes the free space from the cache. We have load_only set
5303 * to one because the slow code to read in the free extents does check
5304 * the pinned extents.
5306 cache_block_group(cache
, 1);
5308 pin_down_extent(root
, cache
, bytenr
, num_bytes
, 0);
5310 /* remove us from the free space cache (if we're there at all) */
5311 ret
= btrfs_remove_free_space(cache
, bytenr
, num_bytes
);
5312 btrfs_put_block_group(cache
);
5316 static int __exclude_logged_extent(struct btrfs_root
*root
, u64 start
, u64 num_bytes
)
5319 struct btrfs_block_group_cache
*block_group
;
5320 struct btrfs_caching_control
*caching_ctl
;
5322 block_group
= btrfs_lookup_block_group(root
->fs_info
, start
);
5326 cache_block_group(block_group
, 0);
5327 caching_ctl
= get_caching_control(block_group
);
5331 BUG_ON(!block_group_cache_done(block_group
));
5332 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
5334 mutex_lock(&caching_ctl
->mutex
);
5336 if (start
>= caching_ctl
->progress
) {
5337 ret
= add_excluded_extent(root
, start
, num_bytes
);
5338 } else if (start
+ num_bytes
<= caching_ctl
->progress
) {
5339 ret
= btrfs_remove_free_space(block_group
,
5342 num_bytes
= caching_ctl
->progress
- start
;
5343 ret
= btrfs_remove_free_space(block_group
,
5348 num_bytes
= (start
+ num_bytes
) -
5349 caching_ctl
->progress
;
5350 start
= caching_ctl
->progress
;
5351 ret
= add_excluded_extent(root
, start
, num_bytes
);
5354 mutex_unlock(&caching_ctl
->mutex
);
5355 put_caching_control(caching_ctl
);
5357 btrfs_put_block_group(block_group
);
5361 int btrfs_exclude_logged_extents(struct btrfs_root
*log
,
5362 struct extent_buffer
*eb
)
5364 struct btrfs_file_extent_item
*item
;
5365 struct btrfs_key key
;
5369 if (!btrfs_fs_incompat(log
->fs_info
, MIXED_GROUPS
))
5372 for (i
= 0; i
< btrfs_header_nritems(eb
); i
++) {
5373 btrfs_item_key_to_cpu(eb
, &key
, i
);
5374 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
5376 item
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
5377 found_type
= btrfs_file_extent_type(eb
, item
);
5378 if (found_type
== BTRFS_FILE_EXTENT_INLINE
)
5380 if (btrfs_file_extent_disk_bytenr(eb
, item
) == 0)
5382 key
.objectid
= btrfs_file_extent_disk_bytenr(eb
, item
);
5383 key
.offset
= btrfs_file_extent_disk_num_bytes(eb
, item
);
5384 __exclude_logged_extent(log
, key
.objectid
, key
.offset
);
5391 * btrfs_update_reserved_bytes - update the block_group and space info counters
5392 * @cache: The cache we are manipulating
5393 * @num_bytes: The number of bytes in question
5394 * @reserve: One of the reservation enums
5396 * This is called by the allocator when it reserves space, or by somebody who is
5397 * freeing space that was never actually used on disk. For example if you
5398 * reserve some space for a new leaf in transaction A and before transaction A
5399 * commits you free that leaf, you call this with reserve set to 0 in order to
5400 * clear the reservation.
5402 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5403 * ENOSPC accounting. For data we handle the reservation through clearing the
5404 * delalloc bits in the io_tree. We have to do this since we could end up
5405 * allocating less disk space for the amount of data we have reserved in the
5406 * case of compression.
5408 * If this is a reservation and the block group has become read only we cannot
5409 * make the reservation and return -EAGAIN, otherwise this function always
5412 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
5413 u64 num_bytes
, int reserve
)
5415 struct btrfs_space_info
*space_info
= cache
->space_info
;
5418 spin_lock(&space_info
->lock
);
5419 spin_lock(&cache
->lock
);
5420 if (reserve
!= RESERVE_FREE
) {
5424 cache
->reserved
+= num_bytes
;
5425 space_info
->bytes_reserved
+= num_bytes
;
5426 if (reserve
== RESERVE_ALLOC
) {
5427 trace_btrfs_space_reservation(cache
->fs_info
,
5428 "space_info", space_info
->flags
,
5430 space_info
->bytes_may_use
-= num_bytes
;
5435 space_info
->bytes_readonly
+= num_bytes
;
5436 cache
->reserved
-= num_bytes
;
5437 space_info
->bytes_reserved
-= num_bytes
;
5439 spin_unlock(&cache
->lock
);
5440 spin_unlock(&space_info
->lock
);
5444 void btrfs_prepare_extent_commit(struct btrfs_trans_handle
*trans
,
5445 struct btrfs_root
*root
)
5447 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5448 struct btrfs_caching_control
*next
;
5449 struct btrfs_caching_control
*caching_ctl
;
5450 struct btrfs_block_group_cache
*cache
;
5451 struct btrfs_space_info
*space_info
;
5453 down_write(&fs_info
->extent_commit_sem
);
5455 list_for_each_entry_safe(caching_ctl
, next
,
5456 &fs_info
->caching_block_groups
, list
) {
5457 cache
= caching_ctl
->block_group
;
5458 if (block_group_cache_done(cache
)) {
5459 cache
->last_byte_to_unpin
= (u64
)-1;
5460 list_del_init(&caching_ctl
->list
);
5461 put_caching_control(caching_ctl
);
5463 cache
->last_byte_to_unpin
= caching_ctl
->progress
;
5467 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
5468 fs_info
->pinned_extents
= &fs_info
->freed_extents
[1];
5470 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
5472 up_write(&fs_info
->extent_commit_sem
);
5474 list_for_each_entry_rcu(space_info
, &fs_info
->space_info
, list
)
5475 percpu_counter_set(&space_info
->total_bytes_pinned
, 0);
5477 update_global_block_rsv(fs_info
);
5480 static int unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
5482 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5483 struct btrfs_block_group_cache
*cache
= NULL
;
5484 struct btrfs_space_info
*space_info
;
5485 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5489 while (start
<= end
) {
5492 start
>= cache
->key
.objectid
+ cache
->key
.offset
) {
5494 btrfs_put_block_group(cache
);
5495 cache
= btrfs_lookup_block_group(fs_info
, start
);
5496 BUG_ON(!cache
); /* Logic error */
5499 len
= cache
->key
.objectid
+ cache
->key
.offset
- start
;
5500 len
= min(len
, end
+ 1 - start
);
5502 if (start
< cache
->last_byte_to_unpin
) {
5503 len
= min(len
, cache
->last_byte_to_unpin
- start
);
5504 btrfs_add_free_space(cache
, start
, len
);
5508 space_info
= cache
->space_info
;
5510 spin_lock(&space_info
->lock
);
5511 spin_lock(&cache
->lock
);
5512 cache
->pinned
-= len
;
5513 space_info
->bytes_pinned
-= len
;
5515 space_info
->bytes_readonly
+= len
;
5518 spin_unlock(&cache
->lock
);
5519 if (!readonly
&& global_rsv
->space_info
== space_info
) {
5520 spin_lock(&global_rsv
->lock
);
5521 if (!global_rsv
->full
) {
5522 len
= min(len
, global_rsv
->size
-
5523 global_rsv
->reserved
);
5524 global_rsv
->reserved
+= len
;
5525 space_info
->bytes_may_use
+= len
;
5526 if (global_rsv
->reserved
>= global_rsv
->size
)
5527 global_rsv
->full
= 1;
5529 spin_unlock(&global_rsv
->lock
);
5531 spin_unlock(&space_info
->lock
);
5535 btrfs_put_block_group(cache
);
5539 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
5540 struct btrfs_root
*root
)
5542 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5543 struct extent_io_tree
*unpin
;
5551 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
5552 unpin
= &fs_info
->freed_extents
[1];
5554 unpin
= &fs_info
->freed_extents
[0];
5557 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
5558 EXTENT_DIRTY
, NULL
);
5562 if (btrfs_test_opt(root
, DISCARD
))
5563 ret
= btrfs_discard_extent(root
, start
,
5564 end
+ 1 - start
, NULL
);
5566 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
5567 unpin_extent_range(root
, start
, end
);
5574 static void add_pinned_bytes(struct btrfs_fs_info
*fs_info
, u64 num_bytes
,
5575 u64 owner
, u64 root_objectid
)
5577 struct btrfs_space_info
*space_info
;
5580 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
5581 if (root_objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
5582 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
5584 flags
= BTRFS_BLOCK_GROUP_METADATA
;
5586 flags
= BTRFS_BLOCK_GROUP_DATA
;
5589 space_info
= __find_space_info(fs_info
, flags
);
5590 BUG_ON(!space_info
); /* Logic bug */
5591 percpu_counter_add(&space_info
->total_bytes_pinned
, num_bytes
);
5595 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
5596 struct btrfs_root
*root
,
5597 u64 bytenr
, u64 num_bytes
, u64 parent
,
5598 u64 root_objectid
, u64 owner_objectid
,
5599 u64 owner_offset
, int refs_to_drop
,
5600 struct btrfs_delayed_extent_op
*extent_op
)
5602 struct btrfs_key key
;
5603 struct btrfs_path
*path
;
5604 struct btrfs_fs_info
*info
= root
->fs_info
;
5605 struct btrfs_root
*extent_root
= info
->extent_root
;
5606 struct extent_buffer
*leaf
;
5607 struct btrfs_extent_item
*ei
;
5608 struct btrfs_extent_inline_ref
*iref
;
5611 int extent_slot
= 0;
5612 int found_extent
= 0;
5616 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
5619 path
= btrfs_alloc_path();
5624 path
->leave_spinning
= 1;
5626 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
5627 BUG_ON(!is_data
&& refs_to_drop
!= 1);
5630 skinny_metadata
= 0;
5632 ret
= lookup_extent_backref(trans
, extent_root
, path
, &iref
,
5633 bytenr
, num_bytes
, parent
,
5634 root_objectid
, owner_objectid
,
5637 extent_slot
= path
->slots
[0];
5638 while (extent_slot
>= 0) {
5639 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
5641 if (key
.objectid
!= bytenr
)
5643 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
5644 key
.offset
== num_bytes
) {
5648 if (key
.type
== BTRFS_METADATA_ITEM_KEY
&&
5649 key
.offset
== owner_objectid
) {
5653 if (path
->slots
[0] - extent_slot
> 5)
5657 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5658 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
5659 if (found_extent
&& item_size
< sizeof(*ei
))
5662 if (!found_extent
) {
5664 ret
= remove_extent_backref(trans
, extent_root
, path
,
5668 btrfs_abort_transaction(trans
, extent_root
, ret
);
5671 btrfs_release_path(path
);
5672 path
->leave_spinning
= 1;
5674 key
.objectid
= bytenr
;
5675 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
5676 key
.offset
= num_bytes
;
5678 if (!is_data
&& skinny_metadata
) {
5679 key
.type
= BTRFS_METADATA_ITEM_KEY
;
5680 key
.offset
= owner_objectid
;
5683 ret
= btrfs_search_slot(trans
, extent_root
,
5685 if (ret
> 0 && skinny_metadata
&& path
->slots
[0]) {
5687 * Couldn't find our skinny metadata item,
5688 * see if we have ye olde extent item.
5691 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
5693 if (key
.objectid
== bytenr
&&
5694 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
5695 key
.offset
== num_bytes
)
5699 if (ret
> 0 && skinny_metadata
) {
5700 skinny_metadata
= false;
5701 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
5702 key
.offset
= num_bytes
;
5703 btrfs_release_path(path
);
5704 ret
= btrfs_search_slot(trans
, extent_root
,
5709 btrfs_err(info
, "umm, got %d back from search, was looking for %llu",
5712 btrfs_print_leaf(extent_root
,
5716 btrfs_abort_transaction(trans
, extent_root
, ret
);
5719 extent_slot
= path
->slots
[0];
5721 } else if (ret
== -ENOENT
) {
5722 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
5725 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
5726 bytenr
, parent
, root_objectid
, owner_objectid
,
5729 btrfs_abort_transaction(trans
, extent_root
, ret
);
5733 leaf
= path
->nodes
[0];
5734 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
5735 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5736 if (item_size
< sizeof(*ei
)) {
5737 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
5738 ret
= convert_extent_item_v0(trans
, extent_root
, path
,
5741 btrfs_abort_transaction(trans
, extent_root
, ret
);
5745 btrfs_release_path(path
);
5746 path
->leave_spinning
= 1;
5748 key
.objectid
= bytenr
;
5749 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
5750 key
.offset
= num_bytes
;
5752 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
5755 btrfs_err(info
, "umm, got %d back from search, was looking for %llu",
5757 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
5760 btrfs_abort_transaction(trans
, extent_root
, ret
);
5764 extent_slot
= path
->slots
[0];
5765 leaf
= path
->nodes
[0];
5766 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
5769 BUG_ON(item_size
< sizeof(*ei
));
5770 ei
= btrfs_item_ptr(leaf
, extent_slot
,
5771 struct btrfs_extent_item
);
5772 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
5773 key
.type
== BTRFS_EXTENT_ITEM_KEY
) {
5774 struct btrfs_tree_block_info
*bi
;
5775 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
5776 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
5777 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
5780 refs
= btrfs_extent_refs(leaf
, ei
);
5781 if (refs
< refs_to_drop
) {
5782 btrfs_err(info
, "trying to drop %d refs but we only have %Lu "
5783 "for bytenr %Lu\n", refs_to_drop
, refs
, bytenr
);
5785 btrfs_abort_transaction(trans
, extent_root
, ret
);
5788 refs
-= refs_to_drop
;
5792 __run_delayed_extent_op(extent_op
, leaf
, ei
);
5794 * In the case of inline back ref, reference count will
5795 * be updated by remove_extent_backref
5798 BUG_ON(!found_extent
);
5800 btrfs_set_extent_refs(leaf
, ei
, refs
);
5801 btrfs_mark_buffer_dirty(leaf
);
5804 ret
= remove_extent_backref(trans
, extent_root
, path
,
5808 btrfs_abort_transaction(trans
, extent_root
, ret
);
5812 add_pinned_bytes(root
->fs_info
, -num_bytes
, owner_objectid
,
5816 BUG_ON(is_data
&& refs_to_drop
!=
5817 extent_data_ref_count(root
, path
, iref
));
5819 BUG_ON(path
->slots
[0] != extent_slot
);
5821 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
5822 path
->slots
[0] = extent_slot
;
5827 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
5830 btrfs_abort_transaction(trans
, extent_root
, ret
);
5833 btrfs_release_path(path
);
5836 ret
= btrfs_del_csums(trans
, root
, bytenr
, num_bytes
);
5838 btrfs_abort_transaction(trans
, extent_root
, ret
);
5843 ret
= update_block_group(root
, bytenr
, num_bytes
, 0);
5845 btrfs_abort_transaction(trans
, extent_root
, ret
);
5850 btrfs_free_path(path
);
5855 * when we free an block, it is possible (and likely) that we free the last
5856 * delayed ref for that extent as well. This searches the delayed ref tree for
5857 * a given extent, and if there are no other delayed refs to be processed, it
5858 * removes it from the tree.
5860 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
5861 struct btrfs_root
*root
, u64 bytenr
)
5863 struct btrfs_delayed_ref_head
*head
;
5864 struct btrfs_delayed_ref_root
*delayed_refs
;
5865 struct btrfs_delayed_ref_node
*ref
;
5866 struct rb_node
*node
;
5869 delayed_refs
= &trans
->transaction
->delayed_refs
;
5870 spin_lock(&delayed_refs
->lock
);
5871 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
5875 node
= rb_prev(&head
->node
.rb_node
);
5879 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
5881 /* there are still entries for this ref, we can't drop it */
5882 if (ref
->bytenr
== bytenr
)
5885 if (head
->extent_op
) {
5886 if (!head
->must_insert_reserved
)
5888 btrfs_free_delayed_extent_op(head
->extent_op
);
5889 head
->extent_op
= NULL
;
5893 * waiting for the lock here would deadlock. If someone else has it
5894 * locked they are already in the process of dropping it anyway
5896 if (!mutex_trylock(&head
->mutex
))
5900 * at this point we have a head with no other entries. Go
5901 * ahead and process it.
5903 head
->node
.in_tree
= 0;
5904 rb_erase(&head
->node
.rb_node
, &delayed_refs
->root
);
5906 delayed_refs
->num_entries
--;
5909 * we don't take a ref on the node because we're removing it from the
5910 * tree, so we just steal the ref the tree was holding.
5912 delayed_refs
->num_heads
--;
5913 if (list_empty(&head
->cluster
))
5914 delayed_refs
->num_heads_ready
--;
5916 list_del_init(&head
->cluster
);
5917 spin_unlock(&delayed_refs
->lock
);
5919 BUG_ON(head
->extent_op
);
5920 if (head
->must_insert_reserved
)
5923 mutex_unlock(&head
->mutex
);
5924 btrfs_put_delayed_ref(&head
->node
);
5927 spin_unlock(&delayed_refs
->lock
);
5931 void btrfs_free_tree_block(struct btrfs_trans_handle
*trans
,
5932 struct btrfs_root
*root
,
5933 struct extent_buffer
*buf
,
5934 u64 parent
, int last_ref
)
5936 struct btrfs_block_group_cache
*cache
= NULL
;
5940 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
5941 ret
= btrfs_add_delayed_tree_ref(root
->fs_info
, trans
,
5942 buf
->start
, buf
->len
,
5943 parent
, root
->root_key
.objectid
,
5944 btrfs_header_level(buf
),
5945 BTRFS_DROP_DELAYED_REF
, NULL
, 0);
5946 BUG_ON(ret
); /* -ENOMEM */
5952 cache
= btrfs_lookup_block_group(root
->fs_info
, buf
->start
);
5954 if (btrfs_header_generation(buf
) == trans
->transid
) {
5955 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
5956 ret
= check_ref_cleanup(trans
, root
, buf
->start
);
5961 if (btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
5962 pin_down_extent(root
, cache
, buf
->start
, buf
->len
, 1);
5966 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
));
5968 btrfs_add_free_space(cache
, buf
->start
, buf
->len
);
5969 btrfs_update_reserved_bytes(cache
, buf
->len
, RESERVE_FREE
);
5970 trace_btrfs_reserved_extent_free(root
, buf
->start
, buf
->len
);
5975 add_pinned_bytes(root
->fs_info
, buf
->len
,
5976 btrfs_header_level(buf
),
5977 root
->root_key
.objectid
);
5980 * Deleting the buffer, clear the corrupt flag since it doesn't matter
5983 clear_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
);
5984 btrfs_put_block_group(cache
);
5987 /* Can return -ENOMEM */
5988 int btrfs_free_extent(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
5989 u64 bytenr
, u64 num_bytes
, u64 parent
, u64 root_objectid
,
5990 u64 owner
, u64 offset
, int for_cow
)
5993 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5995 add_pinned_bytes(root
->fs_info
, num_bytes
, owner
, root_objectid
);
5998 * tree log blocks never actually go into the extent allocation
5999 * tree, just update pinning info and exit early.
6001 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
6002 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
6003 /* unlocks the pinned mutex */
6004 btrfs_pin_extent(root
, bytenr
, num_bytes
, 1);
6006 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
6007 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
6009 parent
, root_objectid
, (int)owner
,
6010 BTRFS_DROP_DELAYED_REF
, NULL
, for_cow
);
6012 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
6014 parent
, root_objectid
, owner
,
6015 offset
, BTRFS_DROP_DELAYED_REF
,
6021 static u64
stripe_align(struct btrfs_root
*root
,
6022 struct btrfs_block_group_cache
*cache
,
6023 u64 val
, u64 num_bytes
)
6025 u64 ret
= ALIGN(val
, root
->stripesize
);
6030 * when we wait for progress in the block group caching, its because
6031 * our allocation attempt failed at least once. So, we must sleep
6032 * and let some progress happen before we try again.
6034 * This function will sleep at least once waiting for new free space to
6035 * show up, and then it will check the block group free space numbers
6036 * for our min num_bytes. Another option is to have it go ahead
6037 * and look in the rbtree for a free extent of a given size, but this
6040 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6041 * any of the information in this block group.
6043 static noinline
void
6044 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
6047 struct btrfs_caching_control
*caching_ctl
;
6049 caching_ctl
= get_caching_control(cache
);
6053 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
) ||
6054 (cache
->free_space_ctl
->free_space
>= num_bytes
));
6056 put_caching_control(caching_ctl
);
6060 wait_block_group_cache_done(struct btrfs_block_group_cache
*cache
)
6062 struct btrfs_caching_control
*caching_ctl
;
6065 caching_ctl
= get_caching_control(cache
);
6067 return (cache
->cached
== BTRFS_CACHE_ERROR
) ? -EIO
: 0;
6069 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
));
6070 if (cache
->cached
== BTRFS_CACHE_ERROR
)
6072 put_caching_control(caching_ctl
);
6076 int __get_raid_index(u64 flags
)
6078 if (flags
& BTRFS_BLOCK_GROUP_RAID10
)
6079 return BTRFS_RAID_RAID10
;
6080 else if (flags
& BTRFS_BLOCK_GROUP_RAID1
)
6081 return BTRFS_RAID_RAID1
;
6082 else if (flags
& BTRFS_BLOCK_GROUP_DUP
)
6083 return BTRFS_RAID_DUP
;
6084 else if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
6085 return BTRFS_RAID_RAID0
;
6086 else if (flags
& BTRFS_BLOCK_GROUP_RAID5
)
6087 return BTRFS_RAID_RAID5
;
6088 else if (flags
& BTRFS_BLOCK_GROUP_RAID6
)
6089 return BTRFS_RAID_RAID6
;
6091 return BTRFS_RAID_SINGLE
; /* BTRFS_BLOCK_GROUP_SINGLE */
6094 static int get_block_group_index(struct btrfs_block_group_cache
*cache
)
6096 return __get_raid_index(cache
->flags
);
6099 enum btrfs_loop_type
{
6100 LOOP_CACHING_NOWAIT
= 0,
6101 LOOP_CACHING_WAIT
= 1,
6102 LOOP_ALLOC_CHUNK
= 2,
6103 LOOP_NO_EMPTY_SIZE
= 3,
6107 * walks the btree of allocated extents and find a hole of a given size.
6108 * The key ins is changed to record the hole:
6109 * ins->objectid == start position
6110 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6111 * ins->offset == the size of the hole.
6112 * Any available blocks before search_start are skipped.
6114 * If there is no suitable free space, we will record the max size of
6115 * the free space extent currently.
6117 static noinline
int find_free_extent(struct btrfs_root
*orig_root
,
6118 u64 num_bytes
, u64 empty_size
,
6119 u64 hint_byte
, struct btrfs_key
*ins
,
6123 struct btrfs_root
*root
= orig_root
->fs_info
->extent_root
;
6124 struct btrfs_free_cluster
*last_ptr
= NULL
;
6125 struct btrfs_block_group_cache
*block_group
= NULL
;
6126 struct btrfs_block_group_cache
*used_block_group
;
6127 u64 search_start
= 0;
6128 u64 max_extent_size
= 0;
6129 int empty_cluster
= 2 * 1024 * 1024;
6130 struct btrfs_space_info
*space_info
;
6132 int index
= __get_raid_index(flags
);
6133 int alloc_type
= (flags
& BTRFS_BLOCK_GROUP_DATA
) ?
6134 RESERVE_ALLOC_NO_ACCOUNT
: RESERVE_ALLOC
;
6135 bool found_uncached_bg
= false;
6136 bool failed_cluster_refill
= false;
6137 bool failed_alloc
= false;
6138 bool use_cluster
= true;
6139 bool have_caching_bg
= false;
6141 WARN_ON(num_bytes
< root
->sectorsize
);
6142 btrfs_set_key_type(ins
, BTRFS_EXTENT_ITEM_KEY
);
6146 trace_find_free_extent(orig_root
, num_bytes
, empty_size
, flags
);
6148 space_info
= __find_space_info(root
->fs_info
, flags
);
6150 btrfs_err(root
->fs_info
, "No space info for %llu", flags
);
6155 * If the space info is for both data and metadata it means we have a
6156 * small filesystem and we can't use the clustering stuff.
6158 if (btrfs_mixed_space_info(space_info
))
6159 use_cluster
= false;
6161 if (flags
& BTRFS_BLOCK_GROUP_METADATA
&& use_cluster
) {
6162 last_ptr
= &root
->fs_info
->meta_alloc_cluster
;
6163 if (!btrfs_test_opt(root
, SSD
))
6164 empty_cluster
= 64 * 1024;
6167 if ((flags
& BTRFS_BLOCK_GROUP_DATA
) && use_cluster
&&
6168 btrfs_test_opt(root
, SSD
)) {
6169 last_ptr
= &root
->fs_info
->data_alloc_cluster
;
6173 spin_lock(&last_ptr
->lock
);
6174 if (last_ptr
->block_group
)
6175 hint_byte
= last_ptr
->window_start
;
6176 spin_unlock(&last_ptr
->lock
);
6179 search_start
= max(search_start
, first_logical_byte(root
, 0));
6180 search_start
= max(search_start
, hint_byte
);
6185 if (search_start
== hint_byte
) {
6186 block_group
= btrfs_lookup_block_group(root
->fs_info
,
6188 used_block_group
= block_group
;
6190 * we don't want to use the block group if it doesn't match our
6191 * allocation bits, or if its not cached.
6193 * However if we are re-searching with an ideal block group
6194 * picked out then we don't care that the block group is cached.
6196 if (block_group
&& block_group_bits(block_group
, flags
) &&
6197 block_group
->cached
!= BTRFS_CACHE_NO
) {
6198 down_read(&space_info
->groups_sem
);
6199 if (list_empty(&block_group
->list
) ||
6202 * someone is removing this block group,
6203 * we can't jump into the have_block_group
6204 * target because our list pointers are not
6207 btrfs_put_block_group(block_group
);
6208 up_read(&space_info
->groups_sem
);
6210 index
= get_block_group_index(block_group
);
6211 goto have_block_group
;
6213 } else if (block_group
) {
6214 btrfs_put_block_group(block_group
);
6218 have_caching_bg
= false;
6219 down_read(&space_info
->groups_sem
);
6220 list_for_each_entry(block_group
, &space_info
->block_groups
[index
],
6225 used_block_group
= block_group
;
6226 btrfs_get_block_group(block_group
);
6227 search_start
= block_group
->key
.objectid
;
6230 * this can happen if we end up cycling through all the
6231 * raid types, but we want to make sure we only allocate
6232 * for the proper type.
6234 if (!block_group_bits(block_group
, flags
)) {
6235 u64 extra
= BTRFS_BLOCK_GROUP_DUP
|
6236 BTRFS_BLOCK_GROUP_RAID1
|
6237 BTRFS_BLOCK_GROUP_RAID5
|
6238 BTRFS_BLOCK_GROUP_RAID6
|
6239 BTRFS_BLOCK_GROUP_RAID10
;
6242 * if they asked for extra copies and this block group
6243 * doesn't provide them, bail. This does allow us to
6244 * fill raid0 from raid1.
6246 if ((flags
& extra
) && !(block_group
->flags
& extra
))
6251 cached
= block_group_cache_done(block_group
);
6252 if (unlikely(!cached
)) {
6253 found_uncached_bg
= true;
6254 ret
= cache_block_group(block_group
, 0);
6259 if (unlikely(block_group
->cached
== BTRFS_CACHE_ERROR
))
6261 if (unlikely(block_group
->ro
))
6265 * Ok we want to try and use the cluster allocator, so
6269 unsigned long aligned_cluster
;
6271 * the refill lock keeps out other
6272 * people trying to start a new cluster
6274 spin_lock(&last_ptr
->refill_lock
);
6275 used_block_group
= last_ptr
->block_group
;
6276 if (used_block_group
!= block_group
&&
6277 (!used_block_group
||
6278 used_block_group
->ro
||
6279 !block_group_bits(used_block_group
, flags
))) {
6280 used_block_group
= block_group
;
6281 goto refill_cluster
;
6284 if (used_block_group
!= block_group
)
6285 btrfs_get_block_group(used_block_group
);
6287 offset
= btrfs_alloc_from_cluster(used_block_group
,
6290 used_block_group
->key
.objectid
,
6293 /* we have a block, we're done */
6294 spin_unlock(&last_ptr
->refill_lock
);
6295 trace_btrfs_reserve_extent_cluster(root
,
6296 block_group
, search_start
, num_bytes
);
6300 WARN_ON(last_ptr
->block_group
!= used_block_group
);
6301 if (used_block_group
!= block_group
) {
6302 btrfs_put_block_group(used_block_group
);
6303 used_block_group
= block_group
;
6306 BUG_ON(used_block_group
!= block_group
);
6307 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6308 * set up a new clusters, so lets just skip it
6309 * and let the allocator find whatever block
6310 * it can find. If we reach this point, we
6311 * will have tried the cluster allocator
6312 * plenty of times and not have found
6313 * anything, so we are likely way too
6314 * fragmented for the clustering stuff to find
6317 * However, if the cluster is taken from the
6318 * current block group, release the cluster
6319 * first, so that we stand a better chance of
6320 * succeeding in the unclustered
6322 if (loop
>= LOOP_NO_EMPTY_SIZE
&&
6323 last_ptr
->block_group
!= block_group
) {
6324 spin_unlock(&last_ptr
->refill_lock
);
6325 goto unclustered_alloc
;
6329 * this cluster didn't work out, free it and
6332 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
6334 if (loop
>= LOOP_NO_EMPTY_SIZE
) {
6335 spin_unlock(&last_ptr
->refill_lock
);
6336 goto unclustered_alloc
;
6339 aligned_cluster
= max_t(unsigned long,
6340 empty_cluster
+ empty_size
,
6341 block_group
->full_stripe_len
);
6343 /* allocate a cluster in this block group */
6344 ret
= btrfs_find_space_cluster(root
, block_group
,
6345 last_ptr
, search_start
,
6350 * now pull our allocation out of this
6353 offset
= btrfs_alloc_from_cluster(block_group
,
6359 /* we found one, proceed */
6360 spin_unlock(&last_ptr
->refill_lock
);
6361 trace_btrfs_reserve_extent_cluster(root
,
6362 block_group
, search_start
,
6366 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
6367 && !failed_cluster_refill
) {
6368 spin_unlock(&last_ptr
->refill_lock
);
6370 failed_cluster_refill
= true;
6371 wait_block_group_cache_progress(block_group
,
6372 num_bytes
+ empty_cluster
+ empty_size
);
6373 goto have_block_group
;
6377 * at this point we either didn't find a cluster
6378 * or we weren't able to allocate a block from our
6379 * cluster. Free the cluster we've been trying
6380 * to use, and go to the next block group
6382 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
6383 spin_unlock(&last_ptr
->refill_lock
);
6388 spin_lock(&block_group
->free_space_ctl
->tree_lock
);
6390 block_group
->free_space_ctl
->free_space
<
6391 num_bytes
+ empty_cluster
+ empty_size
) {
6392 if (block_group
->free_space_ctl
->free_space
>
6395 block_group
->free_space_ctl
->free_space
;
6396 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
6399 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
6401 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
6402 num_bytes
, empty_size
,
6405 * If we didn't find a chunk, and we haven't failed on this
6406 * block group before, and this block group is in the middle of
6407 * caching and we are ok with waiting, then go ahead and wait
6408 * for progress to be made, and set failed_alloc to true.
6410 * If failed_alloc is true then we've already waited on this
6411 * block group once and should move on to the next block group.
6413 if (!offset
&& !failed_alloc
&& !cached
&&
6414 loop
> LOOP_CACHING_NOWAIT
) {
6415 wait_block_group_cache_progress(block_group
,
6416 num_bytes
+ empty_size
);
6417 failed_alloc
= true;
6418 goto have_block_group
;
6419 } else if (!offset
) {
6421 have_caching_bg
= true;
6425 search_start
= stripe_align(root
, used_block_group
,
6428 /* move on to the next group */
6429 if (search_start
+ num_bytes
>
6430 used_block_group
->key
.objectid
+ used_block_group
->key
.offset
) {
6431 btrfs_add_free_space(used_block_group
, offset
, num_bytes
);
6435 if (offset
< search_start
)
6436 btrfs_add_free_space(used_block_group
, offset
,
6437 search_start
- offset
);
6438 BUG_ON(offset
> search_start
);
6440 ret
= btrfs_update_reserved_bytes(used_block_group
, num_bytes
,
6442 if (ret
== -EAGAIN
) {
6443 btrfs_add_free_space(used_block_group
, offset
, num_bytes
);
6447 /* we are all good, lets return */
6448 ins
->objectid
= search_start
;
6449 ins
->offset
= num_bytes
;
6451 trace_btrfs_reserve_extent(orig_root
, block_group
,
6452 search_start
, num_bytes
);
6453 if (used_block_group
!= block_group
)
6454 btrfs_put_block_group(used_block_group
);
6455 btrfs_put_block_group(block_group
);
6458 failed_cluster_refill
= false;
6459 failed_alloc
= false;
6460 BUG_ON(index
!= get_block_group_index(block_group
));
6461 if (used_block_group
!= block_group
)
6462 btrfs_put_block_group(used_block_group
);
6463 btrfs_put_block_group(block_group
);
6465 up_read(&space_info
->groups_sem
);
6467 if (!ins
->objectid
&& loop
>= LOOP_CACHING_WAIT
&& have_caching_bg
)
6470 if (!ins
->objectid
&& ++index
< BTRFS_NR_RAID_TYPES
)
6474 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
6475 * caching kthreads as we move along
6476 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
6477 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
6478 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
6481 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
) {
6484 if (loop
== LOOP_ALLOC_CHUNK
) {
6485 struct btrfs_trans_handle
*trans
;
6487 trans
= btrfs_join_transaction(root
);
6488 if (IS_ERR(trans
)) {
6489 ret
= PTR_ERR(trans
);
6493 ret
= do_chunk_alloc(trans
, root
, flags
,
6496 * Do not bail out on ENOSPC since we
6497 * can do more things.
6499 if (ret
< 0 && ret
!= -ENOSPC
)
6500 btrfs_abort_transaction(trans
,
6504 btrfs_end_transaction(trans
, root
);
6509 if (loop
== LOOP_NO_EMPTY_SIZE
) {
6515 } else if (!ins
->objectid
) {
6517 } else if (ins
->objectid
) {
6522 ins
->offset
= max_extent_size
;
6526 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
6527 int dump_block_groups
)
6529 struct btrfs_block_group_cache
*cache
;
6532 spin_lock(&info
->lock
);
6533 printk(KERN_INFO
"space_info %llu has %llu free, is %sfull\n",
6535 info
->total_bytes
- info
->bytes_used
- info
->bytes_pinned
-
6536 info
->bytes_reserved
- info
->bytes_readonly
,
6537 (info
->full
) ? "" : "not ");
6538 printk(KERN_INFO
"space_info total=%llu, used=%llu, pinned=%llu, "
6539 "reserved=%llu, may_use=%llu, readonly=%llu\n",
6540 info
->total_bytes
, info
->bytes_used
, info
->bytes_pinned
,
6541 info
->bytes_reserved
, info
->bytes_may_use
,
6542 info
->bytes_readonly
);
6543 spin_unlock(&info
->lock
);
6545 if (!dump_block_groups
)
6548 down_read(&info
->groups_sem
);
6550 list_for_each_entry(cache
, &info
->block_groups
[index
], list
) {
6551 spin_lock(&cache
->lock
);
6552 printk(KERN_INFO
"block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n",
6553 cache
->key
.objectid
, cache
->key
.offset
,
6554 btrfs_block_group_used(&cache
->item
), cache
->pinned
,
6555 cache
->reserved
, cache
->ro
? "[readonly]" : "");
6556 btrfs_dump_free_space(cache
, bytes
);
6557 spin_unlock(&cache
->lock
);
6559 if (++index
< BTRFS_NR_RAID_TYPES
)
6561 up_read(&info
->groups_sem
);
6564 int btrfs_reserve_extent(struct btrfs_root
*root
,
6565 u64 num_bytes
, u64 min_alloc_size
,
6566 u64 empty_size
, u64 hint_byte
,
6567 struct btrfs_key
*ins
, int is_data
)
6569 bool final_tried
= false;
6573 flags
= btrfs_get_alloc_profile(root
, is_data
);
6575 WARN_ON(num_bytes
< root
->sectorsize
);
6576 ret
= find_free_extent(root
, num_bytes
, empty_size
, hint_byte
, ins
,
6579 if (ret
== -ENOSPC
) {
6580 if (!final_tried
&& ins
->offset
) {
6581 num_bytes
= min(num_bytes
>> 1, ins
->offset
);
6582 num_bytes
= round_down(num_bytes
, root
->sectorsize
);
6583 num_bytes
= max(num_bytes
, min_alloc_size
);
6584 if (num_bytes
== min_alloc_size
)
6587 } else if (btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
6588 struct btrfs_space_info
*sinfo
;
6590 sinfo
= __find_space_info(root
->fs_info
, flags
);
6591 btrfs_err(root
->fs_info
, "allocation failed flags %llu, wanted %llu",
6594 dump_space_info(sinfo
, num_bytes
, 1);
6601 static int __btrfs_free_reserved_extent(struct btrfs_root
*root
,
6602 u64 start
, u64 len
, int pin
)
6604 struct btrfs_block_group_cache
*cache
;
6607 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
6609 btrfs_err(root
->fs_info
, "Unable to find block group for %llu",
6614 if (btrfs_test_opt(root
, DISCARD
))
6615 ret
= btrfs_discard_extent(root
, start
, len
, NULL
);
6618 pin_down_extent(root
, cache
, start
, len
, 1);
6620 btrfs_add_free_space(cache
, start
, len
);
6621 btrfs_update_reserved_bytes(cache
, len
, RESERVE_FREE
);
6623 btrfs_put_block_group(cache
);
6625 trace_btrfs_reserved_extent_free(root
, start
, len
);
6630 int btrfs_free_reserved_extent(struct btrfs_root
*root
,
6633 return __btrfs_free_reserved_extent(root
, start
, len
, 0);
6636 int btrfs_free_and_pin_reserved_extent(struct btrfs_root
*root
,
6639 return __btrfs_free_reserved_extent(root
, start
, len
, 1);
6642 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
6643 struct btrfs_root
*root
,
6644 u64 parent
, u64 root_objectid
,
6645 u64 flags
, u64 owner
, u64 offset
,
6646 struct btrfs_key
*ins
, int ref_mod
)
6649 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6650 struct btrfs_extent_item
*extent_item
;
6651 struct btrfs_extent_inline_ref
*iref
;
6652 struct btrfs_path
*path
;
6653 struct extent_buffer
*leaf
;
6658 type
= BTRFS_SHARED_DATA_REF_KEY
;
6660 type
= BTRFS_EXTENT_DATA_REF_KEY
;
6662 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
6664 path
= btrfs_alloc_path();
6668 path
->leave_spinning
= 1;
6669 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
6672 btrfs_free_path(path
);
6676 leaf
= path
->nodes
[0];
6677 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
6678 struct btrfs_extent_item
);
6679 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
6680 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
6681 btrfs_set_extent_flags(leaf
, extent_item
,
6682 flags
| BTRFS_EXTENT_FLAG_DATA
);
6684 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
6685 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
6687 struct btrfs_shared_data_ref
*ref
;
6688 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
6689 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
6690 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
6692 struct btrfs_extent_data_ref
*ref
;
6693 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
6694 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
6695 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
6696 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
6697 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
6700 btrfs_mark_buffer_dirty(path
->nodes
[0]);
6701 btrfs_free_path(path
);
6703 ret
= update_block_group(root
, ins
->objectid
, ins
->offset
, 1);
6704 if (ret
) { /* -ENOENT, logic error */
6705 btrfs_err(fs_info
, "update block group failed for %llu %llu",
6706 ins
->objectid
, ins
->offset
);
6709 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, ins
->offset
);
6713 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
6714 struct btrfs_root
*root
,
6715 u64 parent
, u64 root_objectid
,
6716 u64 flags
, struct btrfs_disk_key
*key
,
6717 int level
, struct btrfs_key
*ins
)
6720 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6721 struct btrfs_extent_item
*extent_item
;
6722 struct btrfs_tree_block_info
*block_info
;
6723 struct btrfs_extent_inline_ref
*iref
;
6724 struct btrfs_path
*path
;
6725 struct extent_buffer
*leaf
;
6726 u32 size
= sizeof(*extent_item
) + sizeof(*iref
);
6727 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
6730 if (!skinny_metadata
)
6731 size
+= sizeof(*block_info
);
6733 path
= btrfs_alloc_path();
6737 path
->leave_spinning
= 1;
6738 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
6741 btrfs_free_path(path
);
6745 leaf
= path
->nodes
[0];
6746 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
6747 struct btrfs_extent_item
);
6748 btrfs_set_extent_refs(leaf
, extent_item
, 1);
6749 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
6750 btrfs_set_extent_flags(leaf
, extent_item
,
6751 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
6753 if (skinny_metadata
) {
6754 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
6756 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
6757 btrfs_set_tree_block_key(leaf
, block_info
, key
);
6758 btrfs_set_tree_block_level(leaf
, block_info
, level
);
6759 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
6763 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
6764 btrfs_set_extent_inline_ref_type(leaf
, iref
,
6765 BTRFS_SHARED_BLOCK_REF_KEY
);
6766 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
6768 btrfs_set_extent_inline_ref_type(leaf
, iref
,
6769 BTRFS_TREE_BLOCK_REF_KEY
);
6770 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
6773 btrfs_mark_buffer_dirty(leaf
);
6774 btrfs_free_path(path
);
6776 ret
= update_block_group(root
, ins
->objectid
, root
->leafsize
, 1);
6777 if (ret
) { /* -ENOENT, logic error */
6778 btrfs_err(fs_info
, "update block group failed for %llu %llu",
6779 ins
->objectid
, ins
->offset
);
6783 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, root
->leafsize
);
6787 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
6788 struct btrfs_root
*root
,
6789 u64 root_objectid
, u64 owner
,
6790 u64 offset
, struct btrfs_key
*ins
)
6794 BUG_ON(root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
6796 ret
= btrfs_add_delayed_data_ref(root
->fs_info
, trans
, ins
->objectid
,
6798 root_objectid
, owner
, offset
,
6799 BTRFS_ADD_DELAYED_EXTENT
, NULL
, 0);
6804 * this is used by the tree logging recovery code. It records that
6805 * an extent has been allocated and makes sure to clear the free
6806 * space cache bits as well
6808 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
6809 struct btrfs_root
*root
,
6810 u64 root_objectid
, u64 owner
, u64 offset
,
6811 struct btrfs_key
*ins
)
6814 struct btrfs_block_group_cache
*block_group
;
6817 * Mixed block groups will exclude before processing the log so we only
6818 * need to do the exlude dance if this fs isn't mixed.
6820 if (!btrfs_fs_incompat(root
->fs_info
, MIXED_GROUPS
)) {
6821 ret
= __exclude_logged_extent(root
, ins
->objectid
, ins
->offset
);
6826 block_group
= btrfs_lookup_block_group(root
->fs_info
, ins
->objectid
);
6830 ret
= btrfs_update_reserved_bytes(block_group
, ins
->offset
,
6831 RESERVE_ALLOC_NO_ACCOUNT
);
6832 BUG_ON(ret
); /* logic error */
6833 ret
= alloc_reserved_file_extent(trans
, root
, 0, root_objectid
,
6834 0, owner
, offset
, ins
, 1);
6835 btrfs_put_block_group(block_group
);
6839 static struct extent_buffer
*
6840 btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
6841 u64 bytenr
, u32 blocksize
, int level
)
6843 struct extent_buffer
*buf
;
6845 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
6847 return ERR_PTR(-ENOMEM
);
6848 btrfs_set_header_generation(buf
, trans
->transid
);
6849 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, buf
, level
);
6850 btrfs_tree_lock(buf
);
6851 clean_tree_block(trans
, root
, buf
);
6852 clear_bit(EXTENT_BUFFER_STALE
, &buf
->bflags
);
6854 btrfs_set_lock_blocking(buf
);
6855 btrfs_set_buffer_uptodate(buf
);
6857 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
6859 * we allow two log transactions at a time, use different
6860 * EXENT bit to differentiate dirty pages.
6862 if (root
->log_transid
% 2 == 0)
6863 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
6864 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
6866 set_extent_new(&root
->dirty_log_pages
, buf
->start
,
6867 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
6869 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
6870 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
6872 trans
->blocks_used
++;
6873 /* this returns a buffer locked for blocking */
6877 static struct btrfs_block_rsv
*
6878 use_block_rsv(struct btrfs_trans_handle
*trans
,
6879 struct btrfs_root
*root
, u32 blocksize
)
6881 struct btrfs_block_rsv
*block_rsv
;
6882 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
6884 bool global_updated
= false;
6886 block_rsv
= get_block_rsv(trans
, root
);
6888 if (unlikely(block_rsv
->size
== 0))
6891 ret
= block_rsv_use_bytes(block_rsv
, blocksize
);
6895 if (block_rsv
->failfast
)
6896 return ERR_PTR(ret
);
6898 if (block_rsv
->type
== BTRFS_BLOCK_RSV_GLOBAL
&& !global_updated
) {
6899 global_updated
= true;
6900 update_global_block_rsv(root
->fs_info
);
6904 if (btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
6905 static DEFINE_RATELIMIT_STATE(_rs
,
6906 DEFAULT_RATELIMIT_INTERVAL
* 10,
6907 /*DEFAULT_RATELIMIT_BURST*/ 1);
6908 if (__ratelimit(&_rs
))
6910 "btrfs: block rsv returned %d\n", ret
);
6913 ret
= reserve_metadata_bytes(root
, block_rsv
, blocksize
,
6914 BTRFS_RESERVE_NO_FLUSH
);
6918 * If we couldn't reserve metadata bytes try and use some from
6919 * the global reserve if its space type is the same as the global
6922 if (block_rsv
->type
!= BTRFS_BLOCK_RSV_GLOBAL
&&
6923 block_rsv
->space_info
== global_rsv
->space_info
) {
6924 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
6928 return ERR_PTR(ret
);
6931 static void unuse_block_rsv(struct btrfs_fs_info
*fs_info
,
6932 struct btrfs_block_rsv
*block_rsv
, u32 blocksize
)
6934 block_rsv_add_bytes(block_rsv
, blocksize
, 0);
6935 block_rsv_release_bytes(fs_info
, block_rsv
, NULL
, 0);
6939 * finds a free extent and does all the dirty work required for allocation
6940 * returns the key for the extent through ins, and a tree buffer for
6941 * the first block of the extent through buf.
6943 * returns the tree buffer or NULL.
6945 struct extent_buffer
*btrfs_alloc_free_block(struct btrfs_trans_handle
*trans
,
6946 struct btrfs_root
*root
, u32 blocksize
,
6947 u64 parent
, u64 root_objectid
,
6948 struct btrfs_disk_key
*key
, int level
,
6949 u64 hint
, u64 empty_size
)
6951 struct btrfs_key ins
;
6952 struct btrfs_block_rsv
*block_rsv
;
6953 struct extent_buffer
*buf
;
6956 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
6959 block_rsv
= use_block_rsv(trans
, root
, blocksize
);
6960 if (IS_ERR(block_rsv
))
6961 return ERR_CAST(block_rsv
);
6963 ret
= btrfs_reserve_extent(root
, blocksize
, blocksize
,
6964 empty_size
, hint
, &ins
, 0);
6966 unuse_block_rsv(root
->fs_info
, block_rsv
, blocksize
);
6967 return ERR_PTR(ret
);
6970 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
,
6972 BUG_ON(IS_ERR(buf
)); /* -ENOMEM */
6974 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
6976 parent
= ins
.objectid
;
6977 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
6981 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
6982 struct btrfs_delayed_extent_op
*extent_op
;
6983 extent_op
= btrfs_alloc_delayed_extent_op();
6984 BUG_ON(!extent_op
); /* -ENOMEM */
6986 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
6988 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
6989 extent_op
->flags_to_set
= flags
;
6990 if (skinny_metadata
)
6991 extent_op
->update_key
= 0;
6993 extent_op
->update_key
= 1;
6994 extent_op
->update_flags
= 1;
6995 extent_op
->is_data
= 0;
6996 extent_op
->level
= level
;
6998 ret
= btrfs_add_delayed_tree_ref(root
->fs_info
, trans
,
7000 ins
.offset
, parent
, root_objectid
,
7001 level
, BTRFS_ADD_DELAYED_EXTENT
,
7003 BUG_ON(ret
); /* -ENOMEM */
7008 struct walk_control
{
7009 u64 refs
[BTRFS_MAX_LEVEL
];
7010 u64 flags
[BTRFS_MAX_LEVEL
];
7011 struct btrfs_key update_progress
;
7022 #define DROP_REFERENCE 1
7023 #define UPDATE_BACKREF 2
7025 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
7026 struct btrfs_root
*root
,
7027 struct walk_control
*wc
,
7028 struct btrfs_path
*path
)
7036 struct btrfs_key key
;
7037 struct extent_buffer
*eb
;
7042 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
7043 wc
->reada_count
= wc
->reada_count
* 2 / 3;
7044 wc
->reada_count
= max(wc
->reada_count
, 2);
7046 wc
->reada_count
= wc
->reada_count
* 3 / 2;
7047 wc
->reada_count
= min_t(int, wc
->reada_count
,
7048 BTRFS_NODEPTRS_PER_BLOCK(root
));
7051 eb
= path
->nodes
[wc
->level
];
7052 nritems
= btrfs_header_nritems(eb
);
7053 blocksize
= btrfs_level_size(root
, wc
->level
- 1);
7055 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
7056 if (nread
>= wc
->reada_count
)
7060 bytenr
= btrfs_node_blockptr(eb
, slot
);
7061 generation
= btrfs_node_ptr_generation(eb
, slot
);
7063 if (slot
== path
->slots
[wc
->level
])
7066 if (wc
->stage
== UPDATE_BACKREF
&&
7067 generation
<= root
->root_key
.offset
)
7070 /* We don't lock the tree block, it's OK to be racy here */
7071 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
,
7072 wc
->level
- 1, 1, &refs
,
7074 /* We don't care about errors in readahead. */
7079 if (wc
->stage
== DROP_REFERENCE
) {
7083 if (wc
->level
== 1 &&
7084 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
7086 if (!wc
->update_ref
||
7087 generation
<= root
->root_key
.offset
)
7089 btrfs_node_key_to_cpu(eb
, &key
, slot
);
7090 ret
= btrfs_comp_cpu_keys(&key
,
7091 &wc
->update_progress
);
7095 if (wc
->level
== 1 &&
7096 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
7100 ret
= readahead_tree_block(root
, bytenr
, blocksize
,
7106 wc
->reada_slot
= slot
;
7110 * helper to process tree block while walking down the tree.
7112 * when wc->stage == UPDATE_BACKREF, this function updates
7113 * back refs for pointers in the block.
7115 * NOTE: return value 1 means we should stop walking down.
7117 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
7118 struct btrfs_root
*root
,
7119 struct btrfs_path
*path
,
7120 struct walk_control
*wc
, int lookup_info
)
7122 int level
= wc
->level
;
7123 struct extent_buffer
*eb
= path
->nodes
[level
];
7124 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
7127 if (wc
->stage
== UPDATE_BACKREF
&&
7128 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
7132 * when reference count of tree block is 1, it won't increase
7133 * again. once full backref flag is set, we never clear it.
7136 ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
7137 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
)))) {
7138 BUG_ON(!path
->locks
[level
]);
7139 ret
= btrfs_lookup_extent_info(trans
, root
,
7140 eb
->start
, level
, 1,
7143 BUG_ON(ret
== -ENOMEM
);
7146 BUG_ON(wc
->refs
[level
] == 0);
7149 if (wc
->stage
== DROP_REFERENCE
) {
7150 if (wc
->refs
[level
] > 1)
7153 if (path
->locks
[level
] && !wc
->keep_locks
) {
7154 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
7155 path
->locks
[level
] = 0;
7160 /* wc->stage == UPDATE_BACKREF */
7161 if (!(wc
->flags
[level
] & flag
)) {
7162 BUG_ON(!path
->locks
[level
]);
7163 ret
= btrfs_inc_ref(trans
, root
, eb
, 1, wc
->for_reloc
);
7164 BUG_ON(ret
); /* -ENOMEM */
7165 ret
= btrfs_dec_ref(trans
, root
, eb
, 0, wc
->for_reloc
);
7166 BUG_ON(ret
); /* -ENOMEM */
7167 ret
= btrfs_set_disk_extent_flags(trans
, root
, eb
->start
,
7169 btrfs_header_level(eb
), 0);
7170 BUG_ON(ret
); /* -ENOMEM */
7171 wc
->flags
[level
] |= flag
;
7175 * the block is shared by multiple trees, so it's not good to
7176 * keep the tree lock
7178 if (path
->locks
[level
] && level
> 0) {
7179 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
7180 path
->locks
[level
] = 0;
7186 * helper to process tree block pointer.
7188 * when wc->stage == DROP_REFERENCE, this function checks
7189 * reference count of the block pointed to. if the block
7190 * is shared and we need update back refs for the subtree
7191 * rooted at the block, this function changes wc->stage to
7192 * UPDATE_BACKREF. if the block is shared and there is no
7193 * need to update back, this function drops the reference
7196 * NOTE: return value 1 means we should stop walking down.
7198 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
7199 struct btrfs_root
*root
,
7200 struct btrfs_path
*path
,
7201 struct walk_control
*wc
, int *lookup_info
)
7207 struct btrfs_key key
;
7208 struct extent_buffer
*next
;
7209 int level
= wc
->level
;
7213 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
7214 path
->slots
[level
]);
7216 * if the lower level block was created before the snapshot
7217 * was created, we know there is no need to update back refs
7220 if (wc
->stage
== UPDATE_BACKREF
&&
7221 generation
<= root
->root_key
.offset
) {
7226 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
7227 blocksize
= btrfs_level_size(root
, level
- 1);
7229 next
= btrfs_find_tree_block(root
, bytenr
, blocksize
);
7231 next
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
7234 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, next
,
7238 btrfs_tree_lock(next
);
7239 btrfs_set_lock_blocking(next
);
7241 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, level
- 1, 1,
7242 &wc
->refs
[level
- 1],
7243 &wc
->flags
[level
- 1]);
7245 btrfs_tree_unlock(next
);
7249 if (unlikely(wc
->refs
[level
- 1] == 0)) {
7250 btrfs_err(root
->fs_info
, "Missing references.");
7255 if (wc
->stage
== DROP_REFERENCE
) {
7256 if (wc
->refs
[level
- 1] > 1) {
7258 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
7261 if (!wc
->update_ref
||
7262 generation
<= root
->root_key
.offset
)
7265 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
7266 path
->slots
[level
]);
7267 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
7271 wc
->stage
= UPDATE_BACKREF
;
7272 wc
->shared_level
= level
- 1;
7276 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
7280 if (!btrfs_buffer_uptodate(next
, generation
, 0)) {
7281 btrfs_tree_unlock(next
);
7282 free_extent_buffer(next
);
7288 if (reada
&& level
== 1)
7289 reada_walk_down(trans
, root
, wc
, path
);
7290 next
= read_tree_block(root
, bytenr
, blocksize
, generation
);
7291 if (!next
|| !extent_buffer_uptodate(next
)) {
7292 free_extent_buffer(next
);
7295 btrfs_tree_lock(next
);
7296 btrfs_set_lock_blocking(next
);
7300 BUG_ON(level
!= btrfs_header_level(next
));
7301 path
->nodes
[level
] = next
;
7302 path
->slots
[level
] = 0;
7303 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
7309 wc
->refs
[level
- 1] = 0;
7310 wc
->flags
[level
- 1] = 0;
7311 if (wc
->stage
== DROP_REFERENCE
) {
7312 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
7313 parent
= path
->nodes
[level
]->start
;
7315 BUG_ON(root
->root_key
.objectid
!=
7316 btrfs_header_owner(path
->nodes
[level
]));
7320 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
, parent
,
7321 root
->root_key
.objectid
, level
- 1, 0, 0);
7322 BUG_ON(ret
); /* -ENOMEM */
7324 btrfs_tree_unlock(next
);
7325 free_extent_buffer(next
);
7331 * helper to process tree block while walking up the tree.
7333 * when wc->stage == DROP_REFERENCE, this function drops
7334 * reference count on the block.
7336 * when wc->stage == UPDATE_BACKREF, this function changes
7337 * wc->stage back to DROP_REFERENCE if we changed wc->stage
7338 * to UPDATE_BACKREF previously while processing the block.
7340 * NOTE: return value 1 means we should stop walking up.
7342 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
7343 struct btrfs_root
*root
,
7344 struct btrfs_path
*path
,
7345 struct walk_control
*wc
)
7348 int level
= wc
->level
;
7349 struct extent_buffer
*eb
= path
->nodes
[level
];
7352 if (wc
->stage
== UPDATE_BACKREF
) {
7353 BUG_ON(wc
->shared_level
< level
);
7354 if (level
< wc
->shared_level
)
7357 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
7361 wc
->stage
= DROP_REFERENCE
;
7362 wc
->shared_level
= -1;
7363 path
->slots
[level
] = 0;
7366 * check reference count again if the block isn't locked.
7367 * we should start walking down the tree again if reference
7370 if (!path
->locks
[level
]) {
7372 btrfs_tree_lock(eb
);
7373 btrfs_set_lock_blocking(eb
);
7374 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
7376 ret
= btrfs_lookup_extent_info(trans
, root
,
7377 eb
->start
, level
, 1,
7381 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
7382 path
->locks
[level
] = 0;
7385 BUG_ON(wc
->refs
[level
] == 0);
7386 if (wc
->refs
[level
] == 1) {
7387 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
7388 path
->locks
[level
] = 0;
7394 /* wc->stage == DROP_REFERENCE */
7395 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
7397 if (wc
->refs
[level
] == 1) {
7399 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
7400 ret
= btrfs_dec_ref(trans
, root
, eb
, 1,
7403 ret
= btrfs_dec_ref(trans
, root
, eb
, 0,
7405 BUG_ON(ret
); /* -ENOMEM */
7407 /* make block locked assertion in clean_tree_block happy */
7408 if (!path
->locks
[level
] &&
7409 btrfs_header_generation(eb
) == trans
->transid
) {
7410 btrfs_tree_lock(eb
);
7411 btrfs_set_lock_blocking(eb
);
7412 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
7414 clean_tree_block(trans
, root
, eb
);
7417 if (eb
== root
->node
) {
7418 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
7421 BUG_ON(root
->root_key
.objectid
!=
7422 btrfs_header_owner(eb
));
7424 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
7425 parent
= path
->nodes
[level
+ 1]->start
;
7427 BUG_ON(root
->root_key
.objectid
!=
7428 btrfs_header_owner(path
->nodes
[level
+ 1]));
7431 btrfs_free_tree_block(trans
, root
, eb
, parent
, wc
->refs
[level
] == 1);
7433 wc
->refs
[level
] = 0;
7434 wc
->flags
[level
] = 0;
7438 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
7439 struct btrfs_root
*root
,
7440 struct btrfs_path
*path
,
7441 struct walk_control
*wc
)
7443 int level
= wc
->level
;
7444 int lookup_info
= 1;
7447 while (level
>= 0) {
7448 ret
= walk_down_proc(trans
, root
, path
, wc
, lookup_info
);
7455 if (path
->slots
[level
] >=
7456 btrfs_header_nritems(path
->nodes
[level
]))
7459 ret
= do_walk_down(trans
, root
, path
, wc
, &lookup_info
);
7461 path
->slots
[level
]++;
7470 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
7471 struct btrfs_root
*root
,
7472 struct btrfs_path
*path
,
7473 struct walk_control
*wc
, int max_level
)
7475 int level
= wc
->level
;
7478 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
7479 while (level
< max_level
&& path
->nodes
[level
]) {
7481 if (path
->slots
[level
] + 1 <
7482 btrfs_header_nritems(path
->nodes
[level
])) {
7483 path
->slots
[level
]++;
7486 ret
= walk_up_proc(trans
, root
, path
, wc
);
7490 if (path
->locks
[level
]) {
7491 btrfs_tree_unlock_rw(path
->nodes
[level
],
7492 path
->locks
[level
]);
7493 path
->locks
[level
] = 0;
7495 free_extent_buffer(path
->nodes
[level
]);
7496 path
->nodes
[level
] = NULL
;
7504 * drop a subvolume tree.
7506 * this function traverses the tree freeing any blocks that only
7507 * referenced by the tree.
7509 * when a shared tree block is found. this function decreases its
7510 * reference count by one. if update_ref is true, this function
7511 * also make sure backrefs for the shared block and all lower level
7512 * blocks are properly updated.
7514 * If called with for_reloc == 0, may exit early with -EAGAIN
7516 int btrfs_drop_snapshot(struct btrfs_root
*root
,
7517 struct btrfs_block_rsv
*block_rsv
, int update_ref
,
7520 struct btrfs_path
*path
;
7521 struct btrfs_trans_handle
*trans
;
7522 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
7523 struct btrfs_root_item
*root_item
= &root
->root_item
;
7524 struct walk_control
*wc
;
7525 struct btrfs_key key
;
7529 bool root_dropped
= false;
7531 path
= btrfs_alloc_path();
7537 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
7539 btrfs_free_path(path
);
7544 trans
= btrfs_start_transaction(tree_root
, 0);
7545 if (IS_ERR(trans
)) {
7546 err
= PTR_ERR(trans
);
7551 trans
->block_rsv
= block_rsv
;
7553 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
7554 level
= btrfs_header_level(root
->node
);
7555 path
->nodes
[level
] = btrfs_lock_root_node(root
);
7556 btrfs_set_lock_blocking(path
->nodes
[level
]);
7557 path
->slots
[level
] = 0;
7558 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
7559 memset(&wc
->update_progress
, 0,
7560 sizeof(wc
->update_progress
));
7562 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
7563 memcpy(&wc
->update_progress
, &key
,
7564 sizeof(wc
->update_progress
));
7566 level
= root_item
->drop_level
;
7568 path
->lowest_level
= level
;
7569 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
7570 path
->lowest_level
= 0;
7578 * unlock our path, this is safe because only this
7579 * function is allowed to delete this snapshot
7581 btrfs_unlock_up_safe(path
, 0);
7583 level
= btrfs_header_level(root
->node
);
7585 btrfs_tree_lock(path
->nodes
[level
]);
7586 btrfs_set_lock_blocking(path
->nodes
[level
]);
7587 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
7589 ret
= btrfs_lookup_extent_info(trans
, root
,
7590 path
->nodes
[level
]->start
,
7591 level
, 1, &wc
->refs
[level
],
7597 BUG_ON(wc
->refs
[level
] == 0);
7599 if (level
== root_item
->drop_level
)
7602 btrfs_tree_unlock(path
->nodes
[level
]);
7603 path
->locks
[level
] = 0;
7604 WARN_ON(wc
->refs
[level
] != 1);
7610 wc
->shared_level
= -1;
7611 wc
->stage
= DROP_REFERENCE
;
7612 wc
->update_ref
= update_ref
;
7614 wc
->for_reloc
= for_reloc
;
7615 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
7619 ret
= walk_down_tree(trans
, root
, path
, wc
);
7625 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
7632 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
7636 if (wc
->stage
== DROP_REFERENCE
) {
7638 btrfs_node_key(path
->nodes
[level
],
7639 &root_item
->drop_progress
,
7640 path
->slots
[level
]);
7641 root_item
->drop_level
= level
;
7644 BUG_ON(wc
->level
== 0);
7645 if (btrfs_should_end_transaction(trans
, tree_root
) ||
7646 (!for_reloc
&& btrfs_need_cleaner_sleep(root
))) {
7647 ret
= btrfs_update_root(trans
, tree_root
,
7651 btrfs_abort_transaction(trans
, tree_root
, ret
);
7656 btrfs_end_transaction_throttle(trans
, tree_root
);
7657 if (!for_reloc
&& btrfs_need_cleaner_sleep(root
)) {
7658 pr_debug("btrfs: drop snapshot early exit\n");
7663 trans
= btrfs_start_transaction(tree_root
, 0);
7664 if (IS_ERR(trans
)) {
7665 err
= PTR_ERR(trans
);
7669 trans
->block_rsv
= block_rsv
;
7672 btrfs_release_path(path
);
7676 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
7678 btrfs_abort_transaction(trans
, tree_root
, ret
);
7682 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
7683 ret
= btrfs_find_root(tree_root
, &root
->root_key
, path
,
7686 btrfs_abort_transaction(trans
, tree_root
, ret
);
7689 } else if (ret
> 0) {
7690 /* if we fail to delete the orphan item this time
7691 * around, it'll get picked up the next time.
7693 * The most common failure here is just -ENOENT.
7695 btrfs_del_orphan_item(trans
, tree_root
,
7696 root
->root_key
.objectid
);
7700 if (root
->in_radix
) {
7701 btrfs_drop_and_free_fs_root(tree_root
->fs_info
, root
);
7703 free_extent_buffer(root
->node
);
7704 free_extent_buffer(root
->commit_root
);
7705 btrfs_put_fs_root(root
);
7707 root_dropped
= true;
7709 btrfs_end_transaction_throttle(trans
, tree_root
);
7712 btrfs_free_path(path
);
7715 * So if we need to stop dropping the snapshot for whatever reason we
7716 * need to make sure to add it back to the dead root list so that we
7717 * keep trying to do the work later. This also cleans up roots if we
7718 * don't have it in the radix (like when we recover after a power fail
7719 * or unmount) so we don't leak memory.
7721 if (!for_reloc
&& root_dropped
== false)
7722 btrfs_add_dead_root(root
);
7724 btrfs_std_error(root
->fs_info
, err
);
7729 * drop subtree rooted at tree block 'node'.
7731 * NOTE: this function will unlock and release tree block 'node'
7732 * only used by relocation code
7734 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
7735 struct btrfs_root
*root
,
7736 struct extent_buffer
*node
,
7737 struct extent_buffer
*parent
)
7739 struct btrfs_path
*path
;
7740 struct walk_control
*wc
;
7746 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
7748 path
= btrfs_alloc_path();
7752 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
7754 btrfs_free_path(path
);
7758 btrfs_assert_tree_locked(parent
);
7759 parent_level
= btrfs_header_level(parent
);
7760 extent_buffer_get(parent
);
7761 path
->nodes
[parent_level
] = parent
;
7762 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
7764 btrfs_assert_tree_locked(node
);
7765 level
= btrfs_header_level(node
);
7766 path
->nodes
[level
] = node
;
7767 path
->slots
[level
] = 0;
7768 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
7770 wc
->refs
[parent_level
] = 1;
7771 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
7773 wc
->shared_level
= -1;
7774 wc
->stage
= DROP_REFERENCE
;
7778 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
7781 wret
= walk_down_tree(trans
, root
, path
, wc
);
7787 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
7795 btrfs_free_path(path
);
7799 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
7805 * if restripe for this chunk_type is on pick target profile and
7806 * return, otherwise do the usual balance
7808 stripped
= get_restripe_target(root
->fs_info
, flags
);
7810 return extended_to_chunk(stripped
);
7813 * we add in the count of missing devices because we want
7814 * to make sure that any RAID levels on a degraded FS
7815 * continue to be honored.
7817 num_devices
= root
->fs_info
->fs_devices
->rw_devices
+
7818 root
->fs_info
->fs_devices
->missing_devices
;
7820 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
7821 BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
|
7822 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
7824 if (num_devices
== 1) {
7825 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
7826 stripped
= flags
& ~stripped
;
7828 /* turn raid0 into single device chunks */
7829 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
7832 /* turn mirroring into duplication */
7833 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
7834 BTRFS_BLOCK_GROUP_RAID10
))
7835 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
7837 /* they already had raid on here, just return */
7838 if (flags
& stripped
)
7841 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
7842 stripped
= flags
& ~stripped
;
7844 /* switch duplicated blocks with raid1 */
7845 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
7846 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
7848 /* this is drive concat, leave it alone */
7854 static int set_block_group_ro(struct btrfs_block_group_cache
*cache
, int force
)
7856 struct btrfs_space_info
*sinfo
= cache
->space_info
;
7858 u64 min_allocable_bytes
;
7863 * We need some metadata space and system metadata space for
7864 * allocating chunks in some corner cases until we force to set
7865 * it to be readonly.
7868 (BTRFS_BLOCK_GROUP_SYSTEM
| BTRFS_BLOCK_GROUP_METADATA
)) &&
7870 min_allocable_bytes
= 1 * 1024 * 1024;
7872 min_allocable_bytes
= 0;
7874 spin_lock(&sinfo
->lock
);
7875 spin_lock(&cache
->lock
);
7882 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
7883 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
7885 if (sinfo
->bytes_used
+ sinfo
->bytes_reserved
+ sinfo
->bytes_pinned
+
7886 sinfo
->bytes_may_use
+ sinfo
->bytes_readonly
+ num_bytes
+
7887 min_allocable_bytes
<= sinfo
->total_bytes
) {
7888 sinfo
->bytes_readonly
+= num_bytes
;
7893 spin_unlock(&cache
->lock
);
7894 spin_unlock(&sinfo
->lock
);
7898 int btrfs_set_block_group_ro(struct btrfs_root
*root
,
7899 struct btrfs_block_group_cache
*cache
)
7902 struct btrfs_trans_handle
*trans
;
7908 trans
= btrfs_join_transaction(root
);
7910 return PTR_ERR(trans
);
7912 alloc_flags
= update_block_group_flags(root
, cache
->flags
);
7913 if (alloc_flags
!= cache
->flags
) {
7914 ret
= do_chunk_alloc(trans
, root
, alloc_flags
,
7920 ret
= set_block_group_ro(cache
, 0);
7923 alloc_flags
= get_alloc_profile(root
, cache
->space_info
->flags
);
7924 ret
= do_chunk_alloc(trans
, root
, alloc_flags
,
7928 ret
= set_block_group_ro(cache
, 0);
7930 btrfs_end_transaction(trans
, root
);
7934 int btrfs_force_chunk_alloc(struct btrfs_trans_handle
*trans
,
7935 struct btrfs_root
*root
, u64 type
)
7937 u64 alloc_flags
= get_alloc_profile(root
, type
);
7938 return do_chunk_alloc(trans
, root
, alloc_flags
,
7943 * helper to account the unused space of all the readonly block group in the
7944 * list. takes mirrors into account.
7946 static u64
__btrfs_get_ro_block_group_free_space(struct list_head
*groups_list
)
7948 struct btrfs_block_group_cache
*block_group
;
7952 list_for_each_entry(block_group
, groups_list
, list
) {
7953 spin_lock(&block_group
->lock
);
7955 if (!block_group
->ro
) {
7956 spin_unlock(&block_group
->lock
);
7960 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_RAID1
|
7961 BTRFS_BLOCK_GROUP_RAID10
|
7962 BTRFS_BLOCK_GROUP_DUP
))
7967 free_bytes
+= (block_group
->key
.offset
-
7968 btrfs_block_group_used(&block_group
->item
)) *
7971 spin_unlock(&block_group
->lock
);
7978 * helper to account the unused space of all the readonly block group in the
7979 * space_info. takes mirrors into account.
7981 u64
btrfs_account_ro_block_groups_free_space(struct btrfs_space_info
*sinfo
)
7986 spin_lock(&sinfo
->lock
);
7988 for(i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
7989 if (!list_empty(&sinfo
->block_groups
[i
]))
7990 free_bytes
+= __btrfs_get_ro_block_group_free_space(
7991 &sinfo
->block_groups
[i
]);
7993 spin_unlock(&sinfo
->lock
);
7998 void btrfs_set_block_group_rw(struct btrfs_root
*root
,
7999 struct btrfs_block_group_cache
*cache
)
8001 struct btrfs_space_info
*sinfo
= cache
->space_info
;
8006 spin_lock(&sinfo
->lock
);
8007 spin_lock(&cache
->lock
);
8008 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
8009 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
8010 sinfo
->bytes_readonly
-= num_bytes
;
8012 spin_unlock(&cache
->lock
);
8013 spin_unlock(&sinfo
->lock
);
8017 * checks to see if its even possible to relocate this block group.
8019 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8020 * ok to go ahead and try.
8022 int btrfs_can_relocate(struct btrfs_root
*root
, u64 bytenr
)
8024 struct btrfs_block_group_cache
*block_group
;
8025 struct btrfs_space_info
*space_info
;
8026 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
8027 struct btrfs_device
*device
;
8028 struct btrfs_trans_handle
*trans
;
8037 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
8039 /* odd, couldn't find the block group, leave it alone */
8043 min_free
= btrfs_block_group_used(&block_group
->item
);
8045 /* no bytes used, we're good */
8049 space_info
= block_group
->space_info
;
8050 spin_lock(&space_info
->lock
);
8052 full
= space_info
->full
;
8055 * if this is the last block group we have in this space, we can't
8056 * relocate it unless we're able to allocate a new chunk below.
8058 * Otherwise, we need to make sure we have room in the space to handle
8059 * all of the extents from this block group. If we can, we're good
8061 if ((space_info
->total_bytes
!= block_group
->key
.offset
) &&
8062 (space_info
->bytes_used
+ space_info
->bytes_reserved
+
8063 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
8064 min_free
< space_info
->total_bytes
)) {
8065 spin_unlock(&space_info
->lock
);
8068 spin_unlock(&space_info
->lock
);
8071 * ok we don't have enough space, but maybe we have free space on our
8072 * devices to allocate new chunks for relocation, so loop through our
8073 * alloc devices and guess if we have enough space. if this block
8074 * group is going to be restriped, run checks against the target
8075 * profile instead of the current one.
8087 target
= get_restripe_target(root
->fs_info
, block_group
->flags
);
8089 index
= __get_raid_index(extended_to_chunk(target
));
8092 * this is just a balance, so if we were marked as full
8093 * we know there is no space for a new chunk
8098 index
= get_block_group_index(block_group
);
8101 if (index
== BTRFS_RAID_RAID10
) {
8105 } else if (index
== BTRFS_RAID_RAID1
) {
8107 } else if (index
== BTRFS_RAID_DUP
) {
8110 } else if (index
== BTRFS_RAID_RAID0
) {
8111 dev_min
= fs_devices
->rw_devices
;
8112 do_div(min_free
, dev_min
);
8115 /* We need to do this so that we can look at pending chunks */
8116 trans
= btrfs_join_transaction(root
);
8117 if (IS_ERR(trans
)) {
8118 ret
= PTR_ERR(trans
);
8122 mutex_lock(&root
->fs_info
->chunk_mutex
);
8123 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
8127 * check to make sure we can actually find a chunk with enough
8128 * space to fit our block group in.
8130 if (device
->total_bytes
> device
->bytes_used
+ min_free
&&
8131 !device
->is_tgtdev_for_dev_replace
) {
8132 ret
= find_free_dev_extent(trans
, device
, min_free
,
8137 if (dev_nr
>= dev_min
)
8143 mutex_unlock(&root
->fs_info
->chunk_mutex
);
8144 btrfs_end_transaction(trans
, root
);
8146 btrfs_put_block_group(block_group
);
8150 static int find_first_block_group(struct btrfs_root
*root
,
8151 struct btrfs_path
*path
, struct btrfs_key
*key
)
8154 struct btrfs_key found_key
;
8155 struct extent_buffer
*leaf
;
8158 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
8163 slot
= path
->slots
[0];
8164 leaf
= path
->nodes
[0];
8165 if (slot
>= btrfs_header_nritems(leaf
)) {
8166 ret
= btrfs_next_leaf(root
, path
);
8173 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
8175 if (found_key
.objectid
>= key
->objectid
&&
8176 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
8186 void btrfs_put_block_group_cache(struct btrfs_fs_info
*info
)
8188 struct btrfs_block_group_cache
*block_group
;
8192 struct inode
*inode
;
8194 block_group
= btrfs_lookup_first_block_group(info
, last
);
8195 while (block_group
) {
8196 spin_lock(&block_group
->lock
);
8197 if (block_group
->iref
)
8199 spin_unlock(&block_group
->lock
);
8200 block_group
= next_block_group(info
->tree_root
,
8210 inode
= block_group
->inode
;
8211 block_group
->iref
= 0;
8212 block_group
->inode
= NULL
;
8213 spin_unlock(&block_group
->lock
);
8215 last
= block_group
->key
.objectid
+ block_group
->key
.offset
;
8216 btrfs_put_block_group(block_group
);
8220 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
8222 struct btrfs_block_group_cache
*block_group
;
8223 struct btrfs_space_info
*space_info
;
8224 struct btrfs_caching_control
*caching_ctl
;
8227 down_write(&info
->extent_commit_sem
);
8228 while (!list_empty(&info
->caching_block_groups
)) {
8229 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
8230 struct btrfs_caching_control
, list
);
8231 list_del(&caching_ctl
->list
);
8232 put_caching_control(caching_ctl
);
8234 up_write(&info
->extent_commit_sem
);
8236 spin_lock(&info
->block_group_cache_lock
);
8237 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
8238 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
8240 rb_erase(&block_group
->cache_node
,
8241 &info
->block_group_cache_tree
);
8242 spin_unlock(&info
->block_group_cache_lock
);
8244 down_write(&block_group
->space_info
->groups_sem
);
8245 list_del(&block_group
->list
);
8246 up_write(&block_group
->space_info
->groups_sem
);
8248 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
8249 wait_block_group_cache_done(block_group
);
8252 * We haven't cached this block group, which means we could
8253 * possibly have excluded extents on this block group.
8255 if (block_group
->cached
== BTRFS_CACHE_NO
||
8256 block_group
->cached
== BTRFS_CACHE_ERROR
)
8257 free_excluded_extents(info
->extent_root
, block_group
);
8259 btrfs_remove_free_space_cache(block_group
);
8260 btrfs_put_block_group(block_group
);
8262 spin_lock(&info
->block_group_cache_lock
);
8264 spin_unlock(&info
->block_group_cache_lock
);
8266 /* now that all the block groups are freed, go through and
8267 * free all the space_info structs. This is only called during
8268 * the final stages of unmount, and so we know nobody is
8269 * using them. We call synchronize_rcu() once before we start,
8270 * just to be on the safe side.
8274 release_global_block_rsv(info
);
8276 while(!list_empty(&info
->space_info
)) {
8277 space_info
= list_entry(info
->space_info
.next
,
8278 struct btrfs_space_info
,
8280 if (btrfs_test_opt(info
->tree_root
, ENOSPC_DEBUG
)) {
8281 if (space_info
->bytes_pinned
> 0 ||
8282 space_info
->bytes_reserved
> 0 ||
8283 space_info
->bytes_may_use
> 0) {
8285 dump_space_info(space_info
, 0, 0);
8288 percpu_counter_destroy(&space_info
->total_bytes_pinned
);
8289 list_del(&space_info
->list
);
8295 static void __link_block_group(struct btrfs_space_info
*space_info
,
8296 struct btrfs_block_group_cache
*cache
)
8298 int index
= get_block_group_index(cache
);
8300 down_write(&space_info
->groups_sem
);
8301 list_add_tail(&cache
->list
, &space_info
->block_groups
[index
]);
8302 up_write(&space_info
->groups_sem
);
8305 int btrfs_read_block_groups(struct btrfs_root
*root
)
8307 struct btrfs_path
*path
;
8309 struct btrfs_block_group_cache
*cache
;
8310 struct btrfs_fs_info
*info
= root
->fs_info
;
8311 struct btrfs_space_info
*space_info
;
8312 struct btrfs_key key
;
8313 struct btrfs_key found_key
;
8314 struct extent_buffer
*leaf
;
8318 root
= info
->extent_root
;
8321 btrfs_set_key_type(&key
, BTRFS_BLOCK_GROUP_ITEM_KEY
);
8322 path
= btrfs_alloc_path();
8327 cache_gen
= btrfs_super_cache_generation(root
->fs_info
->super_copy
);
8328 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
8329 btrfs_super_generation(root
->fs_info
->super_copy
) != cache_gen
)
8331 if (btrfs_test_opt(root
, CLEAR_CACHE
))
8335 ret
= find_first_block_group(root
, path
, &key
);
8340 leaf
= path
->nodes
[0];
8341 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
8342 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
8347 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
8349 if (!cache
->free_space_ctl
) {
8355 atomic_set(&cache
->count
, 1);
8356 spin_lock_init(&cache
->lock
);
8357 cache
->fs_info
= info
;
8358 INIT_LIST_HEAD(&cache
->list
);
8359 INIT_LIST_HEAD(&cache
->cluster_list
);
8363 * When we mount with old space cache, we need to
8364 * set BTRFS_DC_CLEAR and set dirty flag.
8366 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
8367 * truncate the old free space cache inode and
8369 * b) Setting 'dirty flag' makes sure that we flush
8370 * the new space cache info onto disk.
8372 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
8373 if (btrfs_test_opt(root
, SPACE_CACHE
))
8377 read_extent_buffer(leaf
, &cache
->item
,
8378 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
8379 sizeof(cache
->item
));
8380 memcpy(&cache
->key
, &found_key
, sizeof(found_key
));
8382 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
8383 btrfs_release_path(path
);
8384 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
8385 cache
->sectorsize
= root
->sectorsize
;
8386 cache
->full_stripe_len
= btrfs_full_stripe_len(root
,
8387 &root
->fs_info
->mapping_tree
,
8388 found_key
.objectid
);
8389 btrfs_init_free_space_ctl(cache
);
8392 * We need to exclude the super stripes now so that the space
8393 * info has super bytes accounted for, otherwise we'll think
8394 * we have more space than we actually do.
8396 ret
= exclude_super_stripes(root
, cache
);
8399 * We may have excluded something, so call this just in
8402 free_excluded_extents(root
, cache
);
8403 kfree(cache
->free_space_ctl
);
8409 * check for two cases, either we are full, and therefore
8410 * don't need to bother with the caching work since we won't
8411 * find any space, or we are empty, and we can just add all
8412 * the space in and be done with it. This saves us _alot_ of
8413 * time, particularly in the full case.
8415 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
8416 cache
->last_byte_to_unpin
= (u64
)-1;
8417 cache
->cached
= BTRFS_CACHE_FINISHED
;
8418 free_excluded_extents(root
, cache
);
8419 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
8420 cache
->last_byte_to_unpin
= (u64
)-1;
8421 cache
->cached
= BTRFS_CACHE_FINISHED
;
8422 add_new_free_space(cache
, root
->fs_info
,
8424 found_key
.objectid
+
8426 free_excluded_extents(root
, cache
);
8429 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
8431 btrfs_remove_free_space_cache(cache
);
8432 btrfs_put_block_group(cache
);
8436 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
8437 btrfs_block_group_used(&cache
->item
),
8440 btrfs_remove_free_space_cache(cache
);
8441 spin_lock(&info
->block_group_cache_lock
);
8442 rb_erase(&cache
->cache_node
,
8443 &info
->block_group_cache_tree
);
8444 spin_unlock(&info
->block_group_cache_lock
);
8445 btrfs_put_block_group(cache
);
8449 cache
->space_info
= space_info
;
8450 spin_lock(&cache
->space_info
->lock
);
8451 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
8452 spin_unlock(&cache
->space_info
->lock
);
8454 __link_block_group(space_info
, cache
);
8456 set_avail_alloc_bits(root
->fs_info
, cache
->flags
);
8457 if (btrfs_chunk_readonly(root
, cache
->key
.objectid
))
8458 set_block_group_ro(cache
, 1);
8461 list_for_each_entry_rcu(space_info
, &root
->fs_info
->space_info
, list
) {
8462 if (!(get_alloc_profile(root
, space_info
->flags
) &
8463 (BTRFS_BLOCK_GROUP_RAID10
|
8464 BTRFS_BLOCK_GROUP_RAID1
|
8465 BTRFS_BLOCK_GROUP_RAID5
|
8466 BTRFS_BLOCK_GROUP_RAID6
|
8467 BTRFS_BLOCK_GROUP_DUP
)))
8470 * avoid allocating from un-mirrored block group if there are
8471 * mirrored block groups.
8473 list_for_each_entry(cache
,
8474 &space_info
->block_groups
[BTRFS_RAID_RAID0
],
8476 set_block_group_ro(cache
, 1);
8477 list_for_each_entry(cache
,
8478 &space_info
->block_groups
[BTRFS_RAID_SINGLE
],
8480 set_block_group_ro(cache
, 1);
8483 init_global_block_rsv(info
);
8486 btrfs_free_path(path
);
8490 void btrfs_create_pending_block_groups(struct btrfs_trans_handle
*trans
,
8491 struct btrfs_root
*root
)
8493 struct btrfs_block_group_cache
*block_group
, *tmp
;
8494 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
8495 struct btrfs_block_group_item item
;
8496 struct btrfs_key key
;
8499 list_for_each_entry_safe(block_group
, tmp
, &trans
->new_bgs
,
8501 list_del_init(&block_group
->new_bg_list
);
8506 spin_lock(&block_group
->lock
);
8507 memcpy(&item
, &block_group
->item
, sizeof(item
));
8508 memcpy(&key
, &block_group
->key
, sizeof(key
));
8509 spin_unlock(&block_group
->lock
);
8511 ret
= btrfs_insert_item(trans
, extent_root
, &key
, &item
,
8514 btrfs_abort_transaction(trans
, extent_root
, ret
);
8515 ret
= btrfs_finish_chunk_alloc(trans
, extent_root
,
8516 key
.objectid
, key
.offset
);
8518 btrfs_abort_transaction(trans
, extent_root
, ret
);
8522 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
8523 struct btrfs_root
*root
, u64 bytes_used
,
8524 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
8528 struct btrfs_root
*extent_root
;
8529 struct btrfs_block_group_cache
*cache
;
8531 extent_root
= root
->fs_info
->extent_root
;
8533 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
8535 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
8538 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
8540 if (!cache
->free_space_ctl
) {
8545 cache
->key
.objectid
= chunk_offset
;
8546 cache
->key
.offset
= size
;
8547 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
8548 cache
->sectorsize
= root
->sectorsize
;
8549 cache
->fs_info
= root
->fs_info
;
8550 cache
->full_stripe_len
= btrfs_full_stripe_len(root
,
8551 &root
->fs_info
->mapping_tree
,
8554 atomic_set(&cache
->count
, 1);
8555 spin_lock_init(&cache
->lock
);
8556 INIT_LIST_HEAD(&cache
->list
);
8557 INIT_LIST_HEAD(&cache
->cluster_list
);
8558 INIT_LIST_HEAD(&cache
->new_bg_list
);
8560 btrfs_init_free_space_ctl(cache
);
8562 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
8563 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
8564 cache
->flags
= type
;
8565 btrfs_set_block_group_flags(&cache
->item
, type
);
8567 cache
->last_byte_to_unpin
= (u64
)-1;
8568 cache
->cached
= BTRFS_CACHE_FINISHED
;
8569 ret
= exclude_super_stripes(root
, cache
);
8572 * We may have excluded something, so call this just in
8575 free_excluded_extents(root
, cache
);
8576 kfree(cache
->free_space_ctl
);
8581 add_new_free_space(cache
, root
->fs_info
, chunk_offset
,
8582 chunk_offset
+ size
);
8584 free_excluded_extents(root
, cache
);
8586 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
8588 btrfs_remove_free_space_cache(cache
);
8589 btrfs_put_block_group(cache
);
8593 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
8594 &cache
->space_info
);
8596 btrfs_remove_free_space_cache(cache
);
8597 spin_lock(&root
->fs_info
->block_group_cache_lock
);
8598 rb_erase(&cache
->cache_node
,
8599 &root
->fs_info
->block_group_cache_tree
);
8600 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
8601 btrfs_put_block_group(cache
);
8604 update_global_block_rsv(root
->fs_info
);
8606 spin_lock(&cache
->space_info
->lock
);
8607 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
8608 spin_unlock(&cache
->space_info
->lock
);
8610 __link_block_group(cache
->space_info
, cache
);
8612 list_add_tail(&cache
->new_bg_list
, &trans
->new_bgs
);
8614 set_avail_alloc_bits(extent_root
->fs_info
, type
);
8619 static void clear_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
8621 u64 extra_flags
= chunk_to_extended(flags
) &
8622 BTRFS_EXTENDED_PROFILE_MASK
;
8624 write_seqlock(&fs_info
->profiles_lock
);
8625 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
8626 fs_info
->avail_data_alloc_bits
&= ~extra_flags
;
8627 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
8628 fs_info
->avail_metadata_alloc_bits
&= ~extra_flags
;
8629 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
8630 fs_info
->avail_system_alloc_bits
&= ~extra_flags
;
8631 write_sequnlock(&fs_info
->profiles_lock
);
8634 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
8635 struct btrfs_root
*root
, u64 group_start
)
8637 struct btrfs_path
*path
;
8638 struct btrfs_block_group_cache
*block_group
;
8639 struct btrfs_free_cluster
*cluster
;
8640 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
8641 struct btrfs_key key
;
8642 struct inode
*inode
;
8647 root
= root
->fs_info
->extent_root
;
8649 block_group
= btrfs_lookup_block_group(root
->fs_info
, group_start
);
8650 BUG_ON(!block_group
);
8651 BUG_ON(!block_group
->ro
);
8654 * Free the reserved super bytes from this block group before
8657 free_excluded_extents(root
, block_group
);
8659 memcpy(&key
, &block_group
->key
, sizeof(key
));
8660 index
= get_block_group_index(block_group
);
8661 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
8662 BTRFS_BLOCK_GROUP_RAID1
|
8663 BTRFS_BLOCK_GROUP_RAID10
))
8668 /* make sure this block group isn't part of an allocation cluster */
8669 cluster
= &root
->fs_info
->data_alloc_cluster
;
8670 spin_lock(&cluster
->refill_lock
);
8671 btrfs_return_cluster_to_free_space(block_group
, cluster
);
8672 spin_unlock(&cluster
->refill_lock
);
8675 * make sure this block group isn't part of a metadata
8676 * allocation cluster
8678 cluster
= &root
->fs_info
->meta_alloc_cluster
;
8679 spin_lock(&cluster
->refill_lock
);
8680 btrfs_return_cluster_to_free_space(block_group
, cluster
);
8681 spin_unlock(&cluster
->refill_lock
);
8683 path
= btrfs_alloc_path();
8689 inode
= lookup_free_space_inode(tree_root
, block_group
, path
);
8690 if (!IS_ERR(inode
)) {
8691 ret
= btrfs_orphan_add(trans
, inode
);
8693 btrfs_add_delayed_iput(inode
);
8697 /* One for the block groups ref */
8698 spin_lock(&block_group
->lock
);
8699 if (block_group
->iref
) {
8700 block_group
->iref
= 0;
8701 block_group
->inode
= NULL
;
8702 spin_unlock(&block_group
->lock
);
8705 spin_unlock(&block_group
->lock
);
8707 /* One for our lookup ref */
8708 btrfs_add_delayed_iput(inode
);
8711 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
8712 key
.offset
= block_group
->key
.objectid
;
8715 ret
= btrfs_search_slot(trans
, tree_root
, &key
, path
, -1, 1);
8719 btrfs_release_path(path
);
8721 ret
= btrfs_del_item(trans
, tree_root
, path
);
8724 btrfs_release_path(path
);
8727 spin_lock(&root
->fs_info
->block_group_cache_lock
);
8728 rb_erase(&block_group
->cache_node
,
8729 &root
->fs_info
->block_group_cache_tree
);
8731 if (root
->fs_info
->first_logical_byte
== block_group
->key
.objectid
)
8732 root
->fs_info
->first_logical_byte
= (u64
)-1;
8733 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
8735 down_write(&block_group
->space_info
->groups_sem
);
8737 * we must use list_del_init so people can check to see if they
8738 * are still on the list after taking the semaphore
8740 list_del_init(&block_group
->list
);
8741 if (list_empty(&block_group
->space_info
->block_groups
[index
]))
8742 clear_avail_alloc_bits(root
->fs_info
, block_group
->flags
);
8743 up_write(&block_group
->space_info
->groups_sem
);
8745 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
8746 wait_block_group_cache_done(block_group
);
8748 btrfs_remove_free_space_cache(block_group
);
8750 spin_lock(&block_group
->space_info
->lock
);
8751 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
8752 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
8753 block_group
->space_info
->disk_total
-= block_group
->key
.offset
* factor
;
8754 spin_unlock(&block_group
->space_info
->lock
);
8756 memcpy(&key
, &block_group
->key
, sizeof(key
));
8758 btrfs_clear_space_info_full(root
->fs_info
);
8760 btrfs_put_block_group(block_group
);
8761 btrfs_put_block_group(block_group
);
8763 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
8769 ret
= btrfs_del_item(trans
, root
, path
);
8771 btrfs_free_path(path
);
8775 int btrfs_init_space_info(struct btrfs_fs_info
*fs_info
)
8777 struct btrfs_space_info
*space_info
;
8778 struct btrfs_super_block
*disk_super
;
8784 disk_super
= fs_info
->super_copy
;
8785 if (!btrfs_super_root(disk_super
))
8788 features
= btrfs_super_incompat_flags(disk_super
);
8789 if (features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
8792 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
8793 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
8798 flags
= BTRFS_BLOCK_GROUP_METADATA
| BTRFS_BLOCK_GROUP_DATA
;
8799 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
8801 flags
= BTRFS_BLOCK_GROUP_METADATA
;
8802 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
8806 flags
= BTRFS_BLOCK_GROUP_DATA
;
8807 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
8813 int btrfs_error_unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
8815 return unpin_extent_range(root
, start
, end
);
8818 int btrfs_error_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
8819 u64 num_bytes
, u64
*actual_bytes
)
8821 return btrfs_discard_extent(root
, bytenr
, num_bytes
, actual_bytes
);
8824 int btrfs_trim_fs(struct btrfs_root
*root
, struct fstrim_range
*range
)
8826 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8827 struct btrfs_block_group_cache
*cache
= NULL
;
8832 u64 total_bytes
= btrfs_super_total_bytes(fs_info
->super_copy
);
8836 * try to trim all FS space, our block group may start from non-zero.
8838 if (range
->len
== total_bytes
)
8839 cache
= btrfs_lookup_first_block_group(fs_info
, range
->start
);
8841 cache
= btrfs_lookup_block_group(fs_info
, range
->start
);
8844 if (cache
->key
.objectid
>= (range
->start
+ range
->len
)) {
8845 btrfs_put_block_group(cache
);
8849 start
= max(range
->start
, cache
->key
.objectid
);
8850 end
= min(range
->start
+ range
->len
,
8851 cache
->key
.objectid
+ cache
->key
.offset
);
8853 if (end
- start
>= range
->minlen
) {
8854 if (!block_group_cache_done(cache
)) {
8855 ret
= cache_block_group(cache
, 0);
8857 btrfs_put_block_group(cache
);
8860 ret
= wait_block_group_cache_done(cache
);
8862 btrfs_put_block_group(cache
);
8866 ret
= btrfs_trim_block_group(cache
,
8872 trimmed
+= group_trimmed
;
8874 btrfs_put_block_group(cache
);
8879 cache
= next_block_group(fs_info
->tree_root
, cache
);
8882 range
->len
= trimmed
;