1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
6 #include <linux/sched.h>
7 #include <linux/sched/signal.h>
8 #include <linux/pagemap.h>
9 #include <linux/writeback.h>
10 #include <linux/blkdev.h>
11 #include <linux/sort.h>
12 #include <linux/rcupdate.h>
13 #include <linux/kthread.h>
14 #include <linux/slab.h>
15 #include <linux/ratelimit.h>
16 #include <linux/percpu_counter.h>
17 #include <linux/lockdep.h>
18 #include <linux/crc32c.h>
21 #include "print-tree.h"
25 #include "free-space-cache.h"
26 #include "free-space-tree.h"
30 #include "ref-verify.h"
32 #undef SCRAMBLE_DELAYED_REFS
35 * control flags for do_chunk_alloc's force field
36 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
37 * if we really need one.
39 * CHUNK_ALLOC_LIMITED means to only try and allocate one
40 * if we have very few chunks already allocated. This is
41 * used as part of the clustering code to help make sure
42 * we have a good pool of storage to cluster in, without
43 * filling the FS with empty chunks
45 * CHUNK_ALLOC_FORCE means it must try to allocate one
49 CHUNK_ALLOC_NO_FORCE
= 0,
50 CHUNK_ALLOC_LIMITED
= 1,
51 CHUNK_ALLOC_FORCE
= 2,
54 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
55 struct btrfs_delayed_ref_node
*node
, u64 parent
,
56 u64 root_objectid
, u64 owner_objectid
,
57 u64 owner_offset
, int refs_to_drop
,
58 struct btrfs_delayed_extent_op
*extra_op
);
59 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
60 struct extent_buffer
*leaf
,
61 struct btrfs_extent_item
*ei
);
62 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
63 u64 parent
, u64 root_objectid
,
64 u64 flags
, u64 owner
, u64 offset
,
65 struct btrfs_key
*ins
, int ref_mod
);
66 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
67 struct btrfs_delayed_ref_node
*node
,
68 struct btrfs_delayed_extent_op
*extent_op
);
69 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
, u64 flags
,
71 static int find_next_key(struct btrfs_path
*path
, int level
,
72 struct btrfs_key
*key
);
73 static void dump_space_info(struct btrfs_fs_info
*fs_info
,
74 struct btrfs_space_info
*info
, u64 bytes
,
75 int dump_block_groups
);
76 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
78 static void space_info_add_new_bytes(struct btrfs_fs_info
*fs_info
,
79 struct btrfs_space_info
*space_info
,
81 static void space_info_add_old_bytes(struct btrfs_fs_info
*fs_info
,
82 struct btrfs_space_info
*space_info
,
86 block_group_cache_done(struct btrfs_block_group_cache
*cache
)
89 return cache
->cached
== BTRFS_CACHE_FINISHED
||
90 cache
->cached
== BTRFS_CACHE_ERROR
;
93 static int block_group_bits(struct btrfs_block_group_cache
*cache
, u64 bits
)
95 return (cache
->flags
& bits
) == bits
;
98 void btrfs_get_block_group(struct btrfs_block_group_cache
*cache
)
100 atomic_inc(&cache
->count
);
103 void btrfs_put_block_group(struct btrfs_block_group_cache
*cache
)
105 if (atomic_dec_and_test(&cache
->count
)) {
106 WARN_ON(cache
->pinned
> 0);
107 WARN_ON(cache
->reserved
> 0);
110 * If not empty, someone is still holding mutex of
111 * full_stripe_lock, which can only be released by caller.
112 * And it will definitely cause use-after-free when caller
113 * tries to release full stripe lock.
115 * No better way to resolve, but only to warn.
117 WARN_ON(!RB_EMPTY_ROOT(&cache
->full_stripe_locks_root
.root
));
118 kfree(cache
->free_space_ctl
);
124 * this adds the block group to the fs_info rb tree for the block group
127 static int btrfs_add_block_group_cache(struct btrfs_fs_info
*info
,
128 struct btrfs_block_group_cache
*block_group
)
131 struct rb_node
*parent
= NULL
;
132 struct btrfs_block_group_cache
*cache
;
134 spin_lock(&info
->block_group_cache_lock
);
135 p
= &info
->block_group_cache_tree
.rb_node
;
139 cache
= rb_entry(parent
, struct btrfs_block_group_cache
,
141 if (block_group
->key
.objectid
< cache
->key
.objectid
) {
143 } else if (block_group
->key
.objectid
> cache
->key
.objectid
) {
146 spin_unlock(&info
->block_group_cache_lock
);
151 rb_link_node(&block_group
->cache_node
, parent
, p
);
152 rb_insert_color(&block_group
->cache_node
,
153 &info
->block_group_cache_tree
);
155 if (info
->first_logical_byte
> block_group
->key
.objectid
)
156 info
->first_logical_byte
= block_group
->key
.objectid
;
158 spin_unlock(&info
->block_group_cache_lock
);
164 * This will return the block group at or after bytenr if contains is 0, else
165 * it will return the block group that contains the bytenr
167 static struct btrfs_block_group_cache
*
168 block_group_cache_tree_search(struct btrfs_fs_info
*info
, u64 bytenr
,
171 struct btrfs_block_group_cache
*cache
, *ret
= NULL
;
175 spin_lock(&info
->block_group_cache_lock
);
176 n
= info
->block_group_cache_tree
.rb_node
;
179 cache
= rb_entry(n
, struct btrfs_block_group_cache
,
181 end
= cache
->key
.objectid
+ cache
->key
.offset
- 1;
182 start
= cache
->key
.objectid
;
184 if (bytenr
< start
) {
185 if (!contains
&& (!ret
|| start
< ret
->key
.objectid
))
188 } else if (bytenr
> start
) {
189 if (contains
&& bytenr
<= end
) {
200 btrfs_get_block_group(ret
);
201 if (bytenr
== 0 && info
->first_logical_byte
> ret
->key
.objectid
)
202 info
->first_logical_byte
= ret
->key
.objectid
;
204 spin_unlock(&info
->block_group_cache_lock
);
209 static int add_excluded_extent(struct btrfs_fs_info
*fs_info
,
210 u64 start
, u64 num_bytes
)
212 u64 end
= start
+ num_bytes
- 1;
213 set_extent_bits(&fs_info
->freed_extents
[0],
214 start
, end
, EXTENT_UPTODATE
);
215 set_extent_bits(&fs_info
->freed_extents
[1],
216 start
, end
, EXTENT_UPTODATE
);
220 static void free_excluded_extents(struct btrfs_block_group_cache
*cache
)
222 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
225 start
= cache
->key
.objectid
;
226 end
= start
+ cache
->key
.offset
- 1;
228 clear_extent_bits(&fs_info
->freed_extents
[0],
229 start
, end
, EXTENT_UPTODATE
);
230 clear_extent_bits(&fs_info
->freed_extents
[1],
231 start
, end
, EXTENT_UPTODATE
);
234 static int exclude_super_stripes(struct btrfs_block_group_cache
*cache
)
236 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
242 if (cache
->key
.objectid
< BTRFS_SUPER_INFO_OFFSET
) {
243 stripe_len
= BTRFS_SUPER_INFO_OFFSET
- cache
->key
.objectid
;
244 cache
->bytes_super
+= stripe_len
;
245 ret
= add_excluded_extent(fs_info
, cache
->key
.objectid
,
251 for (i
= 0; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
252 bytenr
= btrfs_sb_offset(i
);
253 ret
= btrfs_rmap_block(fs_info
, cache
->key
.objectid
,
254 bytenr
, &logical
, &nr
, &stripe_len
);
261 if (logical
[nr
] > cache
->key
.objectid
+
265 if (logical
[nr
] + stripe_len
<= cache
->key
.objectid
)
269 if (start
< cache
->key
.objectid
) {
270 start
= cache
->key
.objectid
;
271 len
= (logical
[nr
] + stripe_len
) - start
;
273 len
= min_t(u64
, stripe_len
,
274 cache
->key
.objectid
+
275 cache
->key
.offset
- start
);
278 cache
->bytes_super
+= len
;
279 ret
= add_excluded_extent(fs_info
, start
, len
);
291 static struct btrfs_caching_control
*
292 get_caching_control(struct btrfs_block_group_cache
*cache
)
294 struct btrfs_caching_control
*ctl
;
296 spin_lock(&cache
->lock
);
297 if (!cache
->caching_ctl
) {
298 spin_unlock(&cache
->lock
);
302 ctl
= cache
->caching_ctl
;
303 refcount_inc(&ctl
->count
);
304 spin_unlock(&cache
->lock
);
308 static void put_caching_control(struct btrfs_caching_control
*ctl
)
310 if (refcount_dec_and_test(&ctl
->count
))
314 #ifdef CONFIG_BTRFS_DEBUG
315 static void fragment_free_space(struct btrfs_block_group_cache
*block_group
)
317 struct btrfs_fs_info
*fs_info
= block_group
->fs_info
;
318 u64 start
= block_group
->key
.objectid
;
319 u64 len
= block_group
->key
.offset
;
320 u64 chunk
= block_group
->flags
& BTRFS_BLOCK_GROUP_METADATA
?
321 fs_info
->nodesize
: fs_info
->sectorsize
;
322 u64 step
= chunk
<< 1;
324 while (len
> chunk
) {
325 btrfs_remove_free_space(block_group
, start
, chunk
);
336 * this is only called by cache_block_group, since we could have freed extents
337 * we need to check the pinned_extents for any extents that can't be used yet
338 * since their free space will be released as soon as the transaction commits.
340 u64
add_new_free_space(struct btrfs_block_group_cache
*block_group
,
343 struct btrfs_fs_info
*info
= block_group
->fs_info
;
344 u64 extent_start
, extent_end
, size
, total_added
= 0;
347 while (start
< end
) {
348 ret
= find_first_extent_bit(info
->pinned_extents
, start
,
349 &extent_start
, &extent_end
,
350 EXTENT_DIRTY
| EXTENT_UPTODATE
,
355 if (extent_start
<= start
) {
356 start
= extent_end
+ 1;
357 } else if (extent_start
> start
&& extent_start
< end
) {
358 size
= extent_start
- start
;
360 ret
= btrfs_add_free_space(block_group
, start
,
362 BUG_ON(ret
); /* -ENOMEM or logic error */
363 start
= extent_end
+ 1;
372 ret
= btrfs_add_free_space(block_group
, start
, size
);
373 BUG_ON(ret
); /* -ENOMEM or logic error */
379 static int load_extent_tree_free(struct btrfs_caching_control
*caching_ctl
)
381 struct btrfs_block_group_cache
*block_group
= caching_ctl
->block_group
;
382 struct btrfs_fs_info
*fs_info
= block_group
->fs_info
;
383 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
384 struct btrfs_path
*path
;
385 struct extent_buffer
*leaf
;
386 struct btrfs_key key
;
393 path
= btrfs_alloc_path();
397 last
= max_t(u64
, block_group
->key
.objectid
, BTRFS_SUPER_INFO_OFFSET
);
399 #ifdef CONFIG_BTRFS_DEBUG
401 * If we're fragmenting we don't want to make anybody think we can
402 * allocate from this block group until we've had a chance to fragment
405 if (btrfs_should_fragment_free_space(block_group
))
409 * We don't want to deadlock with somebody trying to allocate a new
410 * extent for the extent root while also trying to search the extent
411 * root to add free space. So we skip locking and search the commit
412 * root, since its read-only
414 path
->skip_locking
= 1;
415 path
->search_commit_root
= 1;
416 path
->reada
= READA_FORWARD
;
420 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
423 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
427 leaf
= path
->nodes
[0];
428 nritems
= btrfs_header_nritems(leaf
);
431 if (btrfs_fs_closing(fs_info
) > 1) {
436 if (path
->slots
[0] < nritems
) {
437 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
439 ret
= find_next_key(path
, 0, &key
);
443 if (need_resched() ||
444 rwsem_is_contended(&fs_info
->commit_root_sem
)) {
446 caching_ctl
->progress
= last
;
447 btrfs_release_path(path
);
448 up_read(&fs_info
->commit_root_sem
);
449 mutex_unlock(&caching_ctl
->mutex
);
451 mutex_lock(&caching_ctl
->mutex
);
452 down_read(&fs_info
->commit_root_sem
);
456 ret
= btrfs_next_leaf(extent_root
, path
);
461 leaf
= path
->nodes
[0];
462 nritems
= btrfs_header_nritems(leaf
);
466 if (key
.objectid
< last
) {
469 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
472 caching_ctl
->progress
= last
;
473 btrfs_release_path(path
);
477 if (key
.objectid
< block_group
->key
.objectid
) {
482 if (key
.objectid
>= block_group
->key
.objectid
+
483 block_group
->key
.offset
)
486 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
||
487 key
.type
== BTRFS_METADATA_ITEM_KEY
) {
488 total_found
+= add_new_free_space(block_group
, last
,
490 if (key
.type
== BTRFS_METADATA_ITEM_KEY
)
491 last
= key
.objectid
+
494 last
= key
.objectid
+ key
.offset
;
496 if (total_found
> CACHING_CTL_WAKE_UP
) {
499 wake_up(&caching_ctl
->wait
);
506 total_found
+= add_new_free_space(block_group
, last
,
507 block_group
->key
.objectid
+
508 block_group
->key
.offset
);
509 caching_ctl
->progress
= (u64
)-1;
512 btrfs_free_path(path
);
516 static noinline
void caching_thread(struct btrfs_work
*work
)
518 struct btrfs_block_group_cache
*block_group
;
519 struct btrfs_fs_info
*fs_info
;
520 struct btrfs_caching_control
*caching_ctl
;
523 caching_ctl
= container_of(work
, struct btrfs_caching_control
, work
);
524 block_group
= caching_ctl
->block_group
;
525 fs_info
= block_group
->fs_info
;
527 mutex_lock(&caching_ctl
->mutex
);
528 down_read(&fs_info
->commit_root_sem
);
530 if (btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
))
531 ret
= load_free_space_tree(caching_ctl
);
533 ret
= load_extent_tree_free(caching_ctl
);
535 spin_lock(&block_group
->lock
);
536 block_group
->caching_ctl
= NULL
;
537 block_group
->cached
= ret
? BTRFS_CACHE_ERROR
: BTRFS_CACHE_FINISHED
;
538 spin_unlock(&block_group
->lock
);
540 #ifdef CONFIG_BTRFS_DEBUG
541 if (btrfs_should_fragment_free_space(block_group
)) {
544 spin_lock(&block_group
->space_info
->lock
);
545 spin_lock(&block_group
->lock
);
546 bytes_used
= block_group
->key
.offset
-
547 btrfs_block_group_used(&block_group
->item
);
548 block_group
->space_info
->bytes_used
+= bytes_used
>> 1;
549 spin_unlock(&block_group
->lock
);
550 spin_unlock(&block_group
->space_info
->lock
);
551 fragment_free_space(block_group
);
555 caching_ctl
->progress
= (u64
)-1;
557 up_read(&fs_info
->commit_root_sem
);
558 free_excluded_extents(block_group
);
559 mutex_unlock(&caching_ctl
->mutex
);
561 wake_up(&caching_ctl
->wait
);
563 put_caching_control(caching_ctl
);
564 btrfs_put_block_group(block_group
);
567 static int cache_block_group(struct btrfs_block_group_cache
*cache
,
571 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
572 struct btrfs_caching_control
*caching_ctl
;
575 caching_ctl
= kzalloc(sizeof(*caching_ctl
), GFP_NOFS
);
579 INIT_LIST_HEAD(&caching_ctl
->list
);
580 mutex_init(&caching_ctl
->mutex
);
581 init_waitqueue_head(&caching_ctl
->wait
);
582 caching_ctl
->block_group
= cache
;
583 caching_ctl
->progress
= cache
->key
.objectid
;
584 refcount_set(&caching_ctl
->count
, 1);
585 btrfs_init_work(&caching_ctl
->work
, btrfs_cache_helper
,
586 caching_thread
, NULL
, NULL
);
588 spin_lock(&cache
->lock
);
590 * This should be a rare occasion, but this could happen I think in the
591 * case where one thread starts to load the space cache info, and then
592 * some other thread starts a transaction commit which tries to do an
593 * allocation while the other thread is still loading the space cache
594 * info. The previous loop should have kept us from choosing this block
595 * group, but if we've moved to the state where we will wait on caching
596 * block groups we need to first check if we're doing a fast load here,
597 * so we can wait for it to finish, otherwise we could end up allocating
598 * from a block group who's cache gets evicted for one reason or
601 while (cache
->cached
== BTRFS_CACHE_FAST
) {
602 struct btrfs_caching_control
*ctl
;
604 ctl
= cache
->caching_ctl
;
605 refcount_inc(&ctl
->count
);
606 prepare_to_wait(&ctl
->wait
, &wait
, TASK_UNINTERRUPTIBLE
);
607 spin_unlock(&cache
->lock
);
611 finish_wait(&ctl
->wait
, &wait
);
612 put_caching_control(ctl
);
613 spin_lock(&cache
->lock
);
616 if (cache
->cached
!= BTRFS_CACHE_NO
) {
617 spin_unlock(&cache
->lock
);
621 WARN_ON(cache
->caching_ctl
);
622 cache
->caching_ctl
= caching_ctl
;
623 cache
->cached
= BTRFS_CACHE_FAST
;
624 spin_unlock(&cache
->lock
);
626 if (btrfs_test_opt(fs_info
, SPACE_CACHE
)) {
627 mutex_lock(&caching_ctl
->mutex
);
628 ret
= load_free_space_cache(fs_info
, cache
);
630 spin_lock(&cache
->lock
);
632 cache
->caching_ctl
= NULL
;
633 cache
->cached
= BTRFS_CACHE_FINISHED
;
634 cache
->last_byte_to_unpin
= (u64
)-1;
635 caching_ctl
->progress
= (u64
)-1;
637 if (load_cache_only
) {
638 cache
->caching_ctl
= NULL
;
639 cache
->cached
= BTRFS_CACHE_NO
;
641 cache
->cached
= BTRFS_CACHE_STARTED
;
642 cache
->has_caching_ctl
= 1;
645 spin_unlock(&cache
->lock
);
646 #ifdef CONFIG_BTRFS_DEBUG
648 btrfs_should_fragment_free_space(cache
)) {
651 spin_lock(&cache
->space_info
->lock
);
652 spin_lock(&cache
->lock
);
653 bytes_used
= cache
->key
.offset
-
654 btrfs_block_group_used(&cache
->item
);
655 cache
->space_info
->bytes_used
+= bytes_used
>> 1;
656 spin_unlock(&cache
->lock
);
657 spin_unlock(&cache
->space_info
->lock
);
658 fragment_free_space(cache
);
661 mutex_unlock(&caching_ctl
->mutex
);
663 wake_up(&caching_ctl
->wait
);
665 put_caching_control(caching_ctl
);
666 free_excluded_extents(cache
);
671 * We're either using the free space tree or no caching at all.
672 * Set cached to the appropriate value and wakeup any waiters.
674 spin_lock(&cache
->lock
);
675 if (load_cache_only
) {
676 cache
->caching_ctl
= NULL
;
677 cache
->cached
= BTRFS_CACHE_NO
;
679 cache
->cached
= BTRFS_CACHE_STARTED
;
680 cache
->has_caching_ctl
= 1;
682 spin_unlock(&cache
->lock
);
683 wake_up(&caching_ctl
->wait
);
686 if (load_cache_only
) {
687 put_caching_control(caching_ctl
);
691 down_write(&fs_info
->commit_root_sem
);
692 refcount_inc(&caching_ctl
->count
);
693 list_add_tail(&caching_ctl
->list
, &fs_info
->caching_block_groups
);
694 up_write(&fs_info
->commit_root_sem
);
696 btrfs_get_block_group(cache
);
698 btrfs_queue_work(fs_info
->caching_workers
, &caching_ctl
->work
);
704 * return the block group that starts at or after bytenr
706 static struct btrfs_block_group_cache
*
707 btrfs_lookup_first_block_group(struct btrfs_fs_info
*info
, u64 bytenr
)
709 return block_group_cache_tree_search(info
, bytenr
, 0);
713 * return the block group that contains the given bytenr
715 struct btrfs_block_group_cache
*btrfs_lookup_block_group(
716 struct btrfs_fs_info
*info
,
719 return block_group_cache_tree_search(info
, bytenr
, 1);
722 static struct btrfs_space_info
*__find_space_info(struct btrfs_fs_info
*info
,
725 struct list_head
*head
= &info
->space_info
;
726 struct btrfs_space_info
*found
;
728 flags
&= BTRFS_BLOCK_GROUP_TYPE_MASK
;
731 list_for_each_entry_rcu(found
, head
, list
) {
732 if (found
->flags
& flags
) {
741 static void add_pinned_bytes(struct btrfs_fs_info
*fs_info
, s64 num_bytes
,
742 bool metadata
, u64 root_objectid
)
744 struct btrfs_space_info
*space_info
;
748 if (root_objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
749 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
751 flags
= BTRFS_BLOCK_GROUP_METADATA
;
753 flags
= BTRFS_BLOCK_GROUP_DATA
;
756 space_info
= __find_space_info(fs_info
, flags
);
758 percpu_counter_add_batch(&space_info
->total_bytes_pinned
, num_bytes
,
759 BTRFS_TOTAL_BYTES_PINNED_BATCH
);
763 * after adding space to the filesystem, we need to clear the full flags
764 * on all the space infos.
766 void btrfs_clear_space_info_full(struct btrfs_fs_info
*info
)
768 struct list_head
*head
= &info
->space_info
;
769 struct btrfs_space_info
*found
;
772 list_for_each_entry_rcu(found
, head
, list
)
777 /* simple helper to search for an existing data extent at a given offset */
778 int btrfs_lookup_data_extent(struct btrfs_fs_info
*fs_info
, u64 start
, u64 len
)
781 struct btrfs_key key
;
782 struct btrfs_path
*path
;
784 path
= btrfs_alloc_path();
788 key
.objectid
= start
;
790 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
791 ret
= btrfs_search_slot(NULL
, fs_info
->extent_root
, &key
, path
, 0, 0);
792 btrfs_free_path(path
);
797 * helper function to lookup reference count and flags of a tree block.
799 * the head node for delayed ref is used to store the sum of all the
800 * reference count modifications queued up in the rbtree. the head
801 * node may also store the extent flags to set. This way you can check
802 * to see what the reference count and extent flags would be if all of
803 * the delayed refs are not processed.
805 int btrfs_lookup_extent_info(struct btrfs_trans_handle
*trans
,
806 struct btrfs_fs_info
*fs_info
, u64 bytenr
,
807 u64 offset
, int metadata
, u64
*refs
, u64
*flags
)
809 struct btrfs_delayed_ref_head
*head
;
810 struct btrfs_delayed_ref_root
*delayed_refs
;
811 struct btrfs_path
*path
;
812 struct btrfs_extent_item
*ei
;
813 struct extent_buffer
*leaf
;
814 struct btrfs_key key
;
821 * If we don't have skinny metadata, don't bother doing anything
824 if (metadata
&& !btrfs_fs_incompat(fs_info
, SKINNY_METADATA
)) {
825 offset
= fs_info
->nodesize
;
829 path
= btrfs_alloc_path();
834 path
->skip_locking
= 1;
835 path
->search_commit_root
= 1;
839 key
.objectid
= bytenr
;
842 key
.type
= BTRFS_METADATA_ITEM_KEY
;
844 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
846 ret
= btrfs_search_slot(trans
, fs_info
->extent_root
, &key
, path
, 0, 0);
850 if (ret
> 0 && metadata
&& key
.type
== BTRFS_METADATA_ITEM_KEY
) {
851 if (path
->slots
[0]) {
853 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
855 if (key
.objectid
== bytenr
&&
856 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
857 key
.offset
== fs_info
->nodesize
)
863 leaf
= path
->nodes
[0];
864 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
865 if (item_size
>= sizeof(*ei
)) {
866 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
867 struct btrfs_extent_item
);
868 num_refs
= btrfs_extent_refs(leaf
, ei
);
869 extent_flags
= btrfs_extent_flags(leaf
, ei
);
872 btrfs_print_v0_err(fs_info
);
874 btrfs_abort_transaction(trans
, ret
);
876 btrfs_handle_fs_error(fs_info
, ret
, NULL
);
881 BUG_ON(num_refs
== 0);
891 delayed_refs
= &trans
->transaction
->delayed_refs
;
892 spin_lock(&delayed_refs
->lock
);
893 head
= btrfs_find_delayed_ref_head(delayed_refs
, bytenr
);
895 if (!mutex_trylock(&head
->mutex
)) {
896 refcount_inc(&head
->refs
);
897 spin_unlock(&delayed_refs
->lock
);
899 btrfs_release_path(path
);
902 * Mutex was contended, block until it's released and try
905 mutex_lock(&head
->mutex
);
906 mutex_unlock(&head
->mutex
);
907 btrfs_put_delayed_ref_head(head
);
910 spin_lock(&head
->lock
);
911 if (head
->extent_op
&& head
->extent_op
->update_flags
)
912 extent_flags
|= head
->extent_op
->flags_to_set
;
914 BUG_ON(num_refs
== 0);
916 num_refs
+= head
->ref_mod
;
917 spin_unlock(&head
->lock
);
918 mutex_unlock(&head
->mutex
);
920 spin_unlock(&delayed_refs
->lock
);
922 WARN_ON(num_refs
== 0);
926 *flags
= extent_flags
;
928 btrfs_free_path(path
);
933 * Back reference rules. Back refs have three main goals:
935 * 1) differentiate between all holders of references to an extent so that
936 * when a reference is dropped we can make sure it was a valid reference
937 * before freeing the extent.
939 * 2) Provide enough information to quickly find the holders of an extent
940 * if we notice a given block is corrupted or bad.
942 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
943 * maintenance. This is actually the same as #2, but with a slightly
944 * different use case.
946 * There are two kinds of back refs. The implicit back refs is optimized
947 * for pointers in non-shared tree blocks. For a given pointer in a block,
948 * back refs of this kind provide information about the block's owner tree
949 * and the pointer's key. These information allow us to find the block by
950 * b-tree searching. The full back refs is for pointers in tree blocks not
951 * referenced by their owner trees. The location of tree block is recorded
952 * in the back refs. Actually the full back refs is generic, and can be
953 * used in all cases the implicit back refs is used. The major shortcoming
954 * of the full back refs is its overhead. Every time a tree block gets
955 * COWed, we have to update back refs entry for all pointers in it.
957 * For a newly allocated tree block, we use implicit back refs for
958 * pointers in it. This means most tree related operations only involve
959 * implicit back refs. For a tree block created in old transaction, the
960 * only way to drop a reference to it is COW it. So we can detect the
961 * event that tree block loses its owner tree's reference and do the
962 * back refs conversion.
964 * When a tree block is COWed through a tree, there are four cases:
966 * The reference count of the block is one and the tree is the block's
967 * owner tree. Nothing to do in this case.
969 * The reference count of the block is one and the tree is not the
970 * block's owner tree. In this case, full back refs is used for pointers
971 * in the block. Remove these full back refs, add implicit back refs for
972 * every pointers in the new block.
974 * The reference count of the block is greater than one and the tree is
975 * the block's owner tree. In this case, implicit back refs is used for
976 * pointers in the block. Add full back refs for every pointers in the
977 * block, increase lower level extents' reference counts. The original
978 * implicit back refs are entailed to the new block.
980 * The reference count of the block is greater than one and the tree is
981 * not the block's owner tree. Add implicit back refs for every pointer in
982 * the new block, increase lower level extents' reference count.
984 * Back Reference Key composing:
986 * The key objectid corresponds to the first byte in the extent,
987 * The key type is used to differentiate between types of back refs.
988 * There are different meanings of the key offset for different types
991 * File extents can be referenced by:
993 * - multiple snapshots, subvolumes, or different generations in one subvol
994 * - different files inside a single subvolume
995 * - different offsets inside a file (bookend extents in file.c)
997 * The extent ref structure for the implicit back refs has fields for:
999 * - Objectid of the subvolume root
1000 * - objectid of the file holding the reference
1001 * - original offset in the file
1002 * - how many bookend extents
1004 * The key offset for the implicit back refs is hash of the first
1007 * The extent ref structure for the full back refs has field for:
1009 * - number of pointers in the tree leaf
1011 * The key offset for the implicit back refs is the first byte of
1014 * When a file extent is allocated, The implicit back refs is used.
1015 * the fields are filled in:
1017 * (root_key.objectid, inode objectid, offset in file, 1)
1019 * When a file extent is removed file truncation, we find the
1020 * corresponding implicit back refs and check the following fields:
1022 * (btrfs_header_owner(leaf), inode objectid, offset in file)
1024 * Btree extents can be referenced by:
1026 * - Different subvolumes
1028 * Both the implicit back refs and the full back refs for tree blocks
1029 * only consist of key. The key offset for the implicit back refs is
1030 * objectid of block's owner tree. The key offset for the full back refs
1031 * is the first byte of parent block.
1033 * When implicit back refs is used, information about the lowest key and
1034 * level of the tree block are required. These information are stored in
1035 * tree block info structure.
1039 * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required,
1040 * is_data == BTRFS_REF_TYPE_DATA, data type is requried,
1041 * is_data == BTRFS_REF_TYPE_ANY, either type is OK.
1043 int btrfs_get_extent_inline_ref_type(const struct extent_buffer
*eb
,
1044 struct btrfs_extent_inline_ref
*iref
,
1045 enum btrfs_inline_ref_type is_data
)
1047 int type
= btrfs_extent_inline_ref_type(eb
, iref
);
1048 u64 offset
= btrfs_extent_inline_ref_offset(eb
, iref
);
1050 if (type
== BTRFS_TREE_BLOCK_REF_KEY
||
1051 type
== BTRFS_SHARED_BLOCK_REF_KEY
||
1052 type
== BTRFS_SHARED_DATA_REF_KEY
||
1053 type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1054 if (is_data
== BTRFS_REF_TYPE_BLOCK
) {
1055 if (type
== BTRFS_TREE_BLOCK_REF_KEY
)
1057 if (type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
1058 ASSERT(eb
->fs_info
);
1060 * Every shared one has parent tree
1061 * block, which must be aligned to
1065 IS_ALIGNED(offset
, eb
->fs_info
->nodesize
))
1068 } else if (is_data
== BTRFS_REF_TYPE_DATA
) {
1069 if (type
== BTRFS_EXTENT_DATA_REF_KEY
)
1071 if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1072 ASSERT(eb
->fs_info
);
1074 * Every shared one has parent tree
1075 * block, which must be aligned to
1079 IS_ALIGNED(offset
, eb
->fs_info
->nodesize
))
1083 ASSERT(is_data
== BTRFS_REF_TYPE_ANY
);
1088 btrfs_print_leaf((struct extent_buffer
*)eb
);
1089 btrfs_err(eb
->fs_info
, "eb %llu invalid extent inline ref type %d",
1093 return BTRFS_REF_TYPE_INVALID
;
1096 static u64
hash_extent_data_ref(u64 root_objectid
, u64 owner
, u64 offset
)
1098 u32 high_crc
= ~(u32
)0;
1099 u32 low_crc
= ~(u32
)0;
1102 lenum
= cpu_to_le64(root_objectid
);
1103 high_crc
= crc32c(high_crc
, &lenum
, sizeof(lenum
));
1104 lenum
= cpu_to_le64(owner
);
1105 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
1106 lenum
= cpu_to_le64(offset
);
1107 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
1109 return ((u64
)high_crc
<< 31) ^ (u64
)low_crc
;
1112 static u64
hash_extent_data_ref_item(struct extent_buffer
*leaf
,
1113 struct btrfs_extent_data_ref
*ref
)
1115 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf
, ref
),
1116 btrfs_extent_data_ref_objectid(leaf
, ref
),
1117 btrfs_extent_data_ref_offset(leaf
, ref
));
1120 static int match_extent_data_ref(struct extent_buffer
*leaf
,
1121 struct btrfs_extent_data_ref
*ref
,
1122 u64 root_objectid
, u64 owner
, u64 offset
)
1124 if (btrfs_extent_data_ref_root(leaf
, ref
) != root_objectid
||
1125 btrfs_extent_data_ref_objectid(leaf
, ref
) != owner
||
1126 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
1131 static noinline
int lookup_extent_data_ref(struct btrfs_trans_handle
*trans
,
1132 struct btrfs_path
*path
,
1133 u64 bytenr
, u64 parent
,
1135 u64 owner
, u64 offset
)
1137 struct btrfs_root
*root
= trans
->fs_info
->extent_root
;
1138 struct btrfs_key key
;
1139 struct btrfs_extent_data_ref
*ref
;
1140 struct extent_buffer
*leaf
;
1146 key
.objectid
= bytenr
;
1148 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1149 key
.offset
= parent
;
1151 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1152 key
.offset
= hash_extent_data_ref(root_objectid
,
1157 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1169 leaf
= path
->nodes
[0];
1170 nritems
= btrfs_header_nritems(leaf
);
1172 if (path
->slots
[0] >= nritems
) {
1173 ret
= btrfs_next_leaf(root
, path
);
1179 leaf
= path
->nodes
[0];
1180 nritems
= btrfs_header_nritems(leaf
);
1184 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1185 if (key
.objectid
!= bytenr
||
1186 key
.type
!= BTRFS_EXTENT_DATA_REF_KEY
)
1189 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1190 struct btrfs_extent_data_ref
);
1192 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1195 btrfs_release_path(path
);
1207 static noinline
int insert_extent_data_ref(struct btrfs_trans_handle
*trans
,
1208 struct btrfs_path
*path
,
1209 u64 bytenr
, u64 parent
,
1210 u64 root_objectid
, u64 owner
,
1211 u64 offset
, int refs_to_add
)
1213 struct btrfs_root
*root
= trans
->fs_info
->extent_root
;
1214 struct btrfs_key key
;
1215 struct extent_buffer
*leaf
;
1220 key
.objectid
= bytenr
;
1222 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1223 key
.offset
= parent
;
1224 size
= sizeof(struct btrfs_shared_data_ref
);
1226 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1227 key
.offset
= hash_extent_data_ref(root_objectid
,
1229 size
= sizeof(struct btrfs_extent_data_ref
);
1232 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, size
);
1233 if (ret
&& ret
!= -EEXIST
)
1236 leaf
= path
->nodes
[0];
1238 struct btrfs_shared_data_ref
*ref
;
1239 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1240 struct btrfs_shared_data_ref
);
1242 btrfs_set_shared_data_ref_count(leaf
, ref
, refs_to_add
);
1244 num_refs
= btrfs_shared_data_ref_count(leaf
, ref
);
1245 num_refs
+= refs_to_add
;
1246 btrfs_set_shared_data_ref_count(leaf
, ref
, num_refs
);
1249 struct btrfs_extent_data_ref
*ref
;
1250 while (ret
== -EEXIST
) {
1251 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1252 struct btrfs_extent_data_ref
);
1253 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1256 btrfs_release_path(path
);
1258 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1260 if (ret
&& ret
!= -EEXIST
)
1263 leaf
= path
->nodes
[0];
1265 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1266 struct btrfs_extent_data_ref
);
1268 btrfs_set_extent_data_ref_root(leaf
, ref
,
1270 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
1271 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
1272 btrfs_set_extent_data_ref_count(leaf
, ref
, refs_to_add
);
1274 num_refs
= btrfs_extent_data_ref_count(leaf
, ref
);
1275 num_refs
+= refs_to_add
;
1276 btrfs_set_extent_data_ref_count(leaf
, ref
, num_refs
);
1279 btrfs_mark_buffer_dirty(leaf
);
1282 btrfs_release_path(path
);
1286 static noinline
int remove_extent_data_ref(struct btrfs_trans_handle
*trans
,
1287 struct btrfs_path
*path
,
1288 int refs_to_drop
, int *last_ref
)
1290 struct btrfs_key key
;
1291 struct btrfs_extent_data_ref
*ref1
= NULL
;
1292 struct btrfs_shared_data_ref
*ref2
= NULL
;
1293 struct extent_buffer
*leaf
;
1297 leaf
= path
->nodes
[0];
1298 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1300 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1301 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1302 struct btrfs_extent_data_ref
);
1303 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1304 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1305 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1306 struct btrfs_shared_data_ref
);
1307 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1308 } else if (unlikely(key
.type
== BTRFS_EXTENT_REF_V0_KEY
)) {
1309 btrfs_print_v0_err(trans
->fs_info
);
1310 btrfs_abort_transaction(trans
, -EINVAL
);
1316 BUG_ON(num_refs
< refs_to_drop
);
1317 num_refs
-= refs_to_drop
;
1319 if (num_refs
== 0) {
1320 ret
= btrfs_del_item(trans
, trans
->fs_info
->extent_root
, path
);
1323 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
)
1324 btrfs_set_extent_data_ref_count(leaf
, ref1
, num_refs
);
1325 else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
)
1326 btrfs_set_shared_data_ref_count(leaf
, ref2
, num_refs
);
1327 btrfs_mark_buffer_dirty(leaf
);
1332 static noinline u32
extent_data_ref_count(struct btrfs_path
*path
,
1333 struct btrfs_extent_inline_ref
*iref
)
1335 struct btrfs_key key
;
1336 struct extent_buffer
*leaf
;
1337 struct btrfs_extent_data_ref
*ref1
;
1338 struct btrfs_shared_data_ref
*ref2
;
1342 leaf
= path
->nodes
[0];
1343 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1345 BUG_ON(key
.type
== BTRFS_EXTENT_REF_V0_KEY
);
1348 * If type is invalid, we should have bailed out earlier than
1351 type
= btrfs_get_extent_inline_ref_type(leaf
, iref
, BTRFS_REF_TYPE_DATA
);
1352 ASSERT(type
!= BTRFS_REF_TYPE_INVALID
);
1353 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1354 ref1
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1355 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1357 ref2
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1358 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1360 } else if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1361 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1362 struct btrfs_extent_data_ref
);
1363 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1364 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1365 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1366 struct btrfs_shared_data_ref
);
1367 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1374 static noinline
int lookup_tree_block_ref(struct btrfs_trans_handle
*trans
,
1375 struct btrfs_path
*path
,
1376 u64 bytenr
, u64 parent
,
1379 struct btrfs_root
*root
= trans
->fs_info
->extent_root
;
1380 struct btrfs_key key
;
1383 key
.objectid
= bytenr
;
1385 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1386 key
.offset
= parent
;
1388 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1389 key
.offset
= root_objectid
;
1392 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1398 static noinline
int insert_tree_block_ref(struct btrfs_trans_handle
*trans
,
1399 struct btrfs_path
*path
,
1400 u64 bytenr
, u64 parent
,
1403 struct btrfs_key key
;
1406 key
.objectid
= bytenr
;
1408 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1409 key
.offset
= parent
;
1411 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1412 key
.offset
= root_objectid
;
1415 ret
= btrfs_insert_empty_item(trans
, trans
->fs_info
->extent_root
,
1417 btrfs_release_path(path
);
1421 static inline int extent_ref_type(u64 parent
, u64 owner
)
1424 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1426 type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1428 type
= BTRFS_TREE_BLOCK_REF_KEY
;
1431 type
= BTRFS_SHARED_DATA_REF_KEY
;
1433 type
= BTRFS_EXTENT_DATA_REF_KEY
;
1438 static int find_next_key(struct btrfs_path
*path
, int level
,
1439 struct btrfs_key
*key
)
1442 for (; level
< BTRFS_MAX_LEVEL
; level
++) {
1443 if (!path
->nodes
[level
])
1445 if (path
->slots
[level
] + 1 >=
1446 btrfs_header_nritems(path
->nodes
[level
]))
1449 btrfs_item_key_to_cpu(path
->nodes
[level
], key
,
1450 path
->slots
[level
] + 1);
1452 btrfs_node_key_to_cpu(path
->nodes
[level
], key
,
1453 path
->slots
[level
] + 1);
1460 * look for inline back ref. if back ref is found, *ref_ret is set
1461 * to the address of inline back ref, and 0 is returned.
1463 * if back ref isn't found, *ref_ret is set to the address where it
1464 * should be inserted, and -ENOENT is returned.
1466 * if insert is true and there are too many inline back refs, the path
1467 * points to the extent item, and -EAGAIN is returned.
1469 * NOTE: inline back refs are ordered in the same way that back ref
1470 * items in the tree are ordered.
1472 static noinline_for_stack
1473 int lookup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1474 struct btrfs_path
*path
,
1475 struct btrfs_extent_inline_ref
**ref_ret
,
1476 u64 bytenr
, u64 num_bytes
,
1477 u64 parent
, u64 root_objectid
,
1478 u64 owner
, u64 offset
, int insert
)
1480 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
1481 struct btrfs_root
*root
= fs_info
->extent_root
;
1482 struct btrfs_key key
;
1483 struct extent_buffer
*leaf
;
1484 struct btrfs_extent_item
*ei
;
1485 struct btrfs_extent_inline_ref
*iref
;
1495 bool skinny_metadata
= btrfs_fs_incompat(fs_info
, SKINNY_METADATA
);
1498 key
.objectid
= bytenr
;
1499 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1500 key
.offset
= num_bytes
;
1502 want
= extent_ref_type(parent
, owner
);
1504 extra_size
= btrfs_extent_inline_ref_size(want
);
1505 path
->keep_locks
= 1;
1510 * Owner is our level, so we can just add one to get the level for the
1511 * block we are interested in.
1513 if (skinny_metadata
&& owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1514 key
.type
= BTRFS_METADATA_ITEM_KEY
;
1519 ret
= btrfs_search_slot(trans
, root
, &key
, path
, extra_size
, 1);
1526 * We may be a newly converted file system which still has the old fat
1527 * extent entries for metadata, so try and see if we have one of those.
1529 if (ret
> 0 && skinny_metadata
) {
1530 skinny_metadata
= false;
1531 if (path
->slots
[0]) {
1533 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
1535 if (key
.objectid
== bytenr
&&
1536 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
1537 key
.offset
== num_bytes
)
1541 key
.objectid
= bytenr
;
1542 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1543 key
.offset
= num_bytes
;
1544 btrfs_release_path(path
);
1549 if (ret
&& !insert
) {
1552 } else if (WARN_ON(ret
)) {
1557 leaf
= path
->nodes
[0];
1558 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1559 if (unlikely(item_size
< sizeof(*ei
))) {
1561 btrfs_print_v0_err(fs_info
);
1562 btrfs_abort_transaction(trans
, err
);
1566 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1567 flags
= btrfs_extent_flags(leaf
, ei
);
1569 ptr
= (unsigned long)(ei
+ 1);
1570 end
= (unsigned long)ei
+ item_size
;
1572 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
&& !skinny_metadata
) {
1573 ptr
+= sizeof(struct btrfs_tree_block_info
);
1577 if (owner
>= BTRFS_FIRST_FREE_OBJECTID
)
1578 needed
= BTRFS_REF_TYPE_DATA
;
1580 needed
= BTRFS_REF_TYPE_BLOCK
;
1588 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1589 type
= btrfs_get_extent_inline_ref_type(leaf
, iref
, needed
);
1590 if (type
== BTRFS_REF_TYPE_INVALID
) {
1598 ptr
+= btrfs_extent_inline_ref_size(type
);
1602 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1603 struct btrfs_extent_data_ref
*dref
;
1604 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1605 if (match_extent_data_ref(leaf
, dref
, root_objectid
,
1610 if (hash_extent_data_ref_item(leaf
, dref
) <
1611 hash_extent_data_ref(root_objectid
, owner
, offset
))
1615 ref_offset
= btrfs_extent_inline_ref_offset(leaf
, iref
);
1617 if (parent
== ref_offset
) {
1621 if (ref_offset
< parent
)
1624 if (root_objectid
== ref_offset
) {
1628 if (ref_offset
< root_objectid
)
1632 ptr
+= btrfs_extent_inline_ref_size(type
);
1634 if (err
== -ENOENT
&& insert
) {
1635 if (item_size
+ extra_size
>=
1636 BTRFS_MAX_EXTENT_ITEM_SIZE(root
)) {
1641 * To add new inline back ref, we have to make sure
1642 * there is no corresponding back ref item.
1643 * For simplicity, we just do not add new inline back
1644 * ref if there is any kind of item for this block
1646 if (find_next_key(path
, 0, &key
) == 0 &&
1647 key
.objectid
== bytenr
&&
1648 key
.type
< BTRFS_BLOCK_GROUP_ITEM_KEY
) {
1653 *ref_ret
= (struct btrfs_extent_inline_ref
*)ptr
;
1656 path
->keep_locks
= 0;
1657 btrfs_unlock_up_safe(path
, 1);
1663 * helper to add new inline back ref
1665 static noinline_for_stack
1666 void setup_inline_extent_backref(struct btrfs_fs_info
*fs_info
,
1667 struct btrfs_path
*path
,
1668 struct btrfs_extent_inline_ref
*iref
,
1669 u64 parent
, u64 root_objectid
,
1670 u64 owner
, u64 offset
, int refs_to_add
,
1671 struct btrfs_delayed_extent_op
*extent_op
)
1673 struct extent_buffer
*leaf
;
1674 struct btrfs_extent_item
*ei
;
1677 unsigned long item_offset
;
1682 leaf
= path
->nodes
[0];
1683 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1684 item_offset
= (unsigned long)iref
- (unsigned long)ei
;
1686 type
= extent_ref_type(parent
, owner
);
1687 size
= btrfs_extent_inline_ref_size(type
);
1689 btrfs_extend_item(fs_info
, path
, size
);
1691 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1692 refs
= btrfs_extent_refs(leaf
, ei
);
1693 refs
+= refs_to_add
;
1694 btrfs_set_extent_refs(leaf
, ei
, refs
);
1696 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1698 ptr
= (unsigned long)ei
+ item_offset
;
1699 end
= (unsigned long)ei
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
1700 if (ptr
< end
- size
)
1701 memmove_extent_buffer(leaf
, ptr
+ size
, ptr
,
1704 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1705 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
1706 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1707 struct btrfs_extent_data_ref
*dref
;
1708 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1709 btrfs_set_extent_data_ref_root(leaf
, dref
, root_objectid
);
1710 btrfs_set_extent_data_ref_objectid(leaf
, dref
, owner
);
1711 btrfs_set_extent_data_ref_offset(leaf
, dref
, offset
);
1712 btrfs_set_extent_data_ref_count(leaf
, dref
, refs_to_add
);
1713 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1714 struct btrfs_shared_data_ref
*sref
;
1715 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1716 btrfs_set_shared_data_ref_count(leaf
, sref
, refs_to_add
);
1717 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1718 } else if (type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
1719 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1721 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
1723 btrfs_mark_buffer_dirty(leaf
);
1726 static int lookup_extent_backref(struct btrfs_trans_handle
*trans
,
1727 struct btrfs_path
*path
,
1728 struct btrfs_extent_inline_ref
**ref_ret
,
1729 u64 bytenr
, u64 num_bytes
, u64 parent
,
1730 u64 root_objectid
, u64 owner
, u64 offset
)
1734 ret
= lookup_inline_extent_backref(trans
, path
, ref_ret
, bytenr
,
1735 num_bytes
, parent
, root_objectid
,
1740 btrfs_release_path(path
);
1743 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1744 ret
= lookup_tree_block_ref(trans
, path
, bytenr
, parent
,
1747 ret
= lookup_extent_data_ref(trans
, path
, bytenr
, parent
,
1748 root_objectid
, owner
, offset
);
1754 * helper to update/remove inline back ref
1756 static noinline_for_stack
1757 void update_inline_extent_backref(struct btrfs_path
*path
,
1758 struct btrfs_extent_inline_ref
*iref
,
1760 struct btrfs_delayed_extent_op
*extent_op
,
1763 struct extent_buffer
*leaf
= path
->nodes
[0];
1764 struct btrfs_fs_info
*fs_info
= leaf
->fs_info
;
1765 struct btrfs_extent_item
*ei
;
1766 struct btrfs_extent_data_ref
*dref
= NULL
;
1767 struct btrfs_shared_data_ref
*sref
= NULL
;
1775 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1776 refs
= btrfs_extent_refs(leaf
, ei
);
1777 WARN_ON(refs_to_mod
< 0 && refs
+ refs_to_mod
<= 0);
1778 refs
+= refs_to_mod
;
1779 btrfs_set_extent_refs(leaf
, ei
, refs
);
1781 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1784 * If type is invalid, we should have bailed out after
1785 * lookup_inline_extent_backref().
1787 type
= btrfs_get_extent_inline_ref_type(leaf
, iref
, BTRFS_REF_TYPE_ANY
);
1788 ASSERT(type
!= BTRFS_REF_TYPE_INVALID
);
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
);
1811 size
= btrfs_extent_inline_ref_size(type
);
1812 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1813 ptr
= (unsigned long)iref
;
1814 end
= (unsigned long)ei
+ item_size
;
1815 if (ptr
+ size
< end
)
1816 memmove_extent_buffer(leaf
, ptr
, ptr
+ size
,
1819 btrfs_truncate_item(fs_info
, path
, item_size
, 1);
1821 btrfs_mark_buffer_dirty(leaf
);
1824 static noinline_for_stack
1825 int insert_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1826 struct btrfs_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
, path
, &iref
, bytenr
,
1836 num_bytes
, parent
, root_objectid
,
1839 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
);
1840 update_inline_extent_backref(path
, iref
, refs_to_add
,
1842 } else if (ret
== -ENOENT
) {
1843 setup_inline_extent_backref(trans
->fs_info
, 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_path
*path
,
1853 u64 bytenr
, u64 parent
, u64 root_objectid
,
1854 u64 owner
, u64 offset
, int refs_to_add
)
1857 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1858 BUG_ON(refs_to_add
!= 1);
1859 ret
= insert_tree_block_ref(trans
, path
, bytenr
, parent
,
1862 ret
= insert_extent_data_ref(trans
, path
, bytenr
, parent
,
1863 root_objectid
, owner
, offset
,
1869 static int remove_extent_backref(struct btrfs_trans_handle
*trans
,
1870 struct btrfs_path
*path
,
1871 struct btrfs_extent_inline_ref
*iref
,
1872 int refs_to_drop
, int is_data
, int *last_ref
)
1876 BUG_ON(!is_data
&& refs_to_drop
!= 1);
1878 update_inline_extent_backref(path
, iref
, -refs_to_drop
, NULL
,
1880 } else if (is_data
) {
1881 ret
= remove_extent_data_ref(trans
, path
, refs_to_drop
,
1885 ret
= btrfs_del_item(trans
, trans
->fs_info
->extent_root
, path
);
1890 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
1891 static int btrfs_issue_discard(struct block_device
*bdev
, u64 start
, u64 len
,
1892 u64
*discarded_bytes
)
1895 u64 bytes_left
, end
;
1896 u64 aligned_start
= ALIGN(start
, 1 << 9);
1898 if (WARN_ON(start
!= aligned_start
)) {
1899 len
-= aligned_start
- start
;
1900 len
= round_down(len
, 1 << 9);
1901 start
= aligned_start
;
1904 *discarded_bytes
= 0;
1912 /* Skip any superblocks on this device. */
1913 for (j
= 0; j
< BTRFS_SUPER_MIRROR_MAX
; j
++) {
1914 u64 sb_start
= btrfs_sb_offset(j
);
1915 u64 sb_end
= sb_start
+ BTRFS_SUPER_INFO_SIZE
;
1916 u64 size
= sb_start
- start
;
1918 if (!in_range(sb_start
, start
, bytes_left
) &&
1919 !in_range(sb_end
, start
, bytes_left
) &&
1920 !in_range(start
, sb_start
, BTRFS_SUPER_INFO_SIZE
))
1924 * Superblock spans beginning of range. Adjust start and
1927 if (sb_start
<= start
) {
1928 start
+= sb_end
- start
;
1933 bytes_left
= end
- start
;
1938 ret
= blkdev_issue_discard(bdev
, start
>> 9, size
>> 9,
1941 *discarded_bytes
+= size
;
1942 else if (ret
!= -EOPNOTSUPP
)
1951 bytes_left
= end
- start
;
1955 ret
= blkdev_issue_discard(bdev
, start
>> 9, bytes_left
>> 9,
1958 *discarded_bytes
+= bytes_left
;
1963 int btrfs_discard_extent(struct btrfs_fs_info
*fs_info
, u64 bytenr
,
1964 u64 num_bytes
, u64
*actual_bytes
)
1967 u64 discarded_bytes
= 0;
1968 struct btrfs_bio
*bbio
= NULL
;
1972 * Avoid races with device replace and make sure our bbio has devices
1973 * associated to its stripes that don't go away while we are discarding.
1975 btrfs_bio_counter_inc_blocked(fs_info
);
1976 /* Tell the block device(s) that the sectors can be discarded */
1977 ret
= btrfs_map_block(fs_info
, BTRFS_MAP_DISCARD
, bytenr
, &num_bytes
,
1979 /* Error condition is -ENOMEM */
1981 struct btrfs_bio_stripe
*stripe
= bbio
->stripes
;
1985 for (i
= 0; i
< bbio
->num_stripes
; i
++, stripe
++) {
1987 struct request_queue
*req_q
;
1989 if (!stripe
->dev
->bdev
) {
1990 ASSERT(btrfs_test_opt(fs_info
, DEGRADED
));
1993 req_q
= bdev_get_queue(stripe
->dev
->bdev
);
1994 if (!blk_queue_discard(req_q
))
1997 ret
= btrfs_issue_discard(stripe
->dev
->bdev
,
2002 discarded_bytes
+= bytes
;
2003 else if (ret
!= -EOPNOTSUPP
)
2004 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2007 * Just in case we get back EOPNOTSUPP for some reason,
2008 * just ignore the return value so we don't screw up
2009 * people calling discard_extent.
2013 btrfs_put_bbio(bbio
);
2015 btrfs_bio_counter_dec(fs_info
);
2018 *actual_bytes
= discarded_bytes
;
2021 if (ret
== -EOPNOTSUPP
)
2026 /* Can return -ENOMEM */
2027 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
2028 struct btrfs_root
*root
,
2029 u64 bytenr
, u64 num_bytes
, u64 parent
,
2030 u64 root_objectid
, u64 owner
, u64 offset
)
2032 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2033 int old_ref_mod
, new_ref_mod
;
2036 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
&&
2037 root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
2039 btrfs_ref_tree_mod(root
, bytenr
, num_bytes
, parent
, root_objectid
,
2040 owner
, offset
, BTRFS_ADD_DELAYED_REF
);
2042 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
2043 ret
= btrfs_add_delayed_tree_ref(trans
, bytenr
,
2045 root_objectid
, (int)owner
,
2046 BTRFS_ADD_DELAYED_REF
, NULL
,
2047 &old_ref_mod
, &new_ref_mod
);
2049 ret
= btrfs_add_delayed_data_ref(trans
, bytenr
,
2051 root_objectid
, owner
, offset
,
2052 0, BTRFS_ADD_DELAYED_REF
,
2053 &old_ref_mod
, &new_ref_mod
);
2056 if (ret
== 0 && old_ref_mod
< 0 && new_ref_mod
>= 0) {
2057 bool metadata
= owner
< BTRFS_FIRST_FREE_OBJECTID
;
2059 add_pinned_bytes(fs_info
, -num_bytes
, metadata
, root_objectid
);
2066 * __btrfs_inc_extent_ref - insert backreference for a given extent
2068 * @trans: Handle of transaction
2070 * @node: The delayed ref node used to get the bytenr/length for
2071 * extent whose references are incremented.
2073 * @parent: If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/
2074 * BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical
2075 * bytenr of the parent block. Since new extents are always
2076 * created with indirect references, this will only be the case
2077 * when relocating a shared extent. In that case, root_objectid
2078 * will be BTRFS_TREE_RELOC_OBJECTID. Otheriwse, parent must
2081 * @root_objectid: The id of the root where this modification has originated,
2082 * this can be either one of the well-known metadata trees or
2083 * the subvolume id which references this extent.
2085 * @owner: For data extents it is the inode number of the owning file.
2086 * For metadata extents this parameter holds the level in the
2087 * tree of the extent.
2089 * @offset: For metadata extents the offset is ignored and is currently
2090 * always passed as 0. For data extents it is the fileoffset
2091 * this extent belongs to.
2093 * @refs_to_add Number of references to add
2095 * @extent_op Pointer to a structure, holding information necessary when
2096 * updating a tree block's flags
2099 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
2100 struct btrfs_delayed_ref_node
*node
,
2101 u64 parent
, u64 root_objectid
,
2102 u64 owner
, u64 offset
, int refs_to_add
,
2103 struct btrfs_delayed_extent_op
*extent_op
)
2105 struct btrfs_path
*path
;
2106 struct extent_buffer
*leaf
;
2107 struct btrfs_extent_item
*item
;
2108 struct btrfs_key key
;
2109 u64 bytenr
= node
->bytenr
;
2110 u64 num_bytes
= node
->num_bytes
;
2114 path
= btrfs_alloc_path();
2118 path
->reada
= READA_FORWARD
;
2119 path
->leave_spinning
= 1;
2120 /* this will setup the path even if it fails to insert the back ref */
2121 ret
= insert_inline_extent_backref(trans
, path
, bytenr
, num_bytes
,
2122 parent
, root_objectid
, owner
,
2123 offset
, refs_to_add
, extent_op
);
2124 if ((ret
< 0 && ret
!= -EAGAIN
) || !ret
)
2128 * Ok we had -EAGAIN which means we didn't have space to insert and
2129 * inline extent ref, so just update the reference count and add a
2132 leaf
= path
->nodes
[0];
2133 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2134 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2135 refs
= btrfs_extent_refs(leaf
, item
);
2136 btrfs_set_extent_refs(leaf
, item
, refs
+ refs_to_add
);
2138 __run_delayed_extent_op(extent_op
, leaf
, item
);
2140 btrfs_mark_buffer_dirty(leaf
);
2141 btrfs_release_path(path
);
2143 path
->reada
= READA_FORWARD
;
2144 path
->leave_spinning
= 1;
2145 /* now insert the actual backref */
2146 ret
= insert_extent_backref(trans
, path
, bytenr
, parent
, root_objectid
,
2147 owner
, offset
, refs_to_add
);
2149 btrfs_abort_transaction(trans
, ret
);
2151 btrfs_free_path(path
);
2155 static int run_delayed_data_ref(struct btrfs_trans_handle
*trans
,
2156 struct btrfs_delayed_ref_node
*node
,
2157 struct btrfs_delayed_extent_op
*extent_op
,
2158 int insert_reserved
)
2161 struct btrfs_delayed_data_ref
*ref
;
2162 struct btrfs_key ins
;
2167 ins
.objectid
= node
->bytenr
;
2168 ins
.offset
= node
->num_bytes
;
2169 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2171 ref
= btrfs_delayed_node_to_data_ref(node
);
2172 trace_run_delayed_data_ref(trans
->fs_info
, node
, ref
, node
->action
);
2174 if (node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2175 parent
= ref
->parent
;
2176 ref_root
= ref
->root
;
2178 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2180 flags
|= extent_op
->flags_to_set
;
2181 ret
= alloc_reserved_file_extent(trans
, parent
, ref_root
,
2182 flags
, ref
->objectid
,
2185 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2186 ret
= __btrfs_inc_extent_ref(trans
, node
, parent
, ref_root
,
2187 ref
->objectid
, ref
->offset
,
2188 node
->ref_mod
, extent_op
);
2189 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2190 ret
= __btrfs_free_extent(trans
, node
, parent
,
2191 ref_root
, ref
->objectid
,
2192 ref
->offset
, node
->ref_mod
,
2200 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
2201 struct extent_buffer
*leaf
,
2202 struct btrfs_extent_item
*ei
)
2204 u64 flags
= btrfs_extent_flags(leaf
, ei
);
2205 if (extent_op
->update_flags
) {
2206 flags
|= extent_op
->flags_to_set
;
2207 btrfs_set_extent_flags(leaf
, ei
, flags
);
2210 if (extent_op
->update_key
) {
2211 struct btrfs_tree_block_info
*bi
;
2212 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
));
2213 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
2214 btrfs_set_tree_block_key(leaf
, bi
, &extent_op
->key
);
2218 static int run_delayed_extent_op(struct btrfs_trans_handle
*trans
,
2219 struct btrfs_delayed_ref_head
*head
,
2220 struct btrfs_delayed_extent_op
*extent_op
)
2222 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2223 struct btrfs_key key
;
2224 struct btrfs_path
*path
;
2225 struct btrfs_extent_item
*ei
;
2226 struct extent_buffer
*leaf
;
2230 int metadata
= !extent_op
->is_data
;
2235 if (metadata
&& !btrfs_fs_incompat(fs_info
, SKINNY_METADATA
))
2238 path
= btrfs_alloc_path();
2242 key
.objectid
= head
->bytenr
;
2245 key
.type
= BTRFS_METADATA_ITEM_KEY
;
2246 key
.offset
= extent_op
->level
;
2248 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2249 key
.offset
= head
->num_bytes
;
2253 path
->reada
= READA_FORWARD
;
2254 path
->leave_spinning
= 1;
2255 ret
= btrfs_search_slot(trans
, fs_info
->extent_root
, &key
, path
, 0, 1);
2262 if (path
->slots
[0] > 0) {
2264 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
2266 if (key
.objectid
== head
->bytenr
&&
2267 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
2268 key
.offset
== head
->num_bytes
)
2272 btrfs_release_path(path
);
2275 key
.objectid
= head
->bytenr
;
2276 key
.offset
= head
->num_bytes
;
2277 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2286 leaf
= path
->nodes
[0];
2287 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2289 if (unlikely(item_size
< sizeof(*ei
))) {
2291 btrfs_print_v0_err(fs_info
);
2292 btrfs_abort_transaction(trans
, err
);
2296 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2297 __run_delayed_extent_op(extent_op
, leaf
, ei
);
2299 btrfs_mark_buffer_dirty(leaf
);
2301 btrfs_free_path(path
);
2305 static int run_delayed_tree_ref(struct btrfs_trans_handle
*trans
,
2306 struct btrfs_delayed_ref_node
*node
,
2307 struct btrfs_delayed_extent_op
*extent_op
,
2308 int insert_reserved
)
2311 struct btrfs_delayed_tree_ref
*ref
;
2315 ref
= btrfs_delayed_node_to_tree_ref(node
);
2316 trace_run_delayed_tree_ref(trans
->fs_info
, node
, ref
, node
->action
);
2318 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2319 parent
= ref
->parent
;
2320 ref_root
= ref
->root
;
2322 if (node
->ref_mod
!= 1) {
2323 btrfs_err(trans
->fs_info
,
2324 "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu",
2325 node
->bytenr
, node
->ref_mod
, node
->action
, ref_root
,
2329 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2330 BUG_ON(!extent_op
|| !extent_op
->update_flags
);
2331 ret
= alloc_reserved_tree_block(trans
, node
, extent_op
);
2332 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2333 ret
= __btrfs_inc_extent_ref(trans
, node
, parent
, ref_root
,
2334 ref
->level
, 0, 1, extent_op
);
2335 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2336 ret
= __btrfs_free_extent(trans
, node
, parent
, ref_root
,
2337 ref
->level
, 0, 1, extent_op
);
2344 /* helper function to actually process a single delayed ref entry */
2345 static int run_one_delayed_ref(struct btrfs_trans_handle
*trans
,
2346 struct btrfs_delayed_ref_node
*node
,
2347 struct btrfs_delayed_extent_op
*extent_op
,
2348 int insert_reserved
)
2352 if (trans
->aborted
) {
2353 if (insert_reserved
)
2354 btrfs_pin_extent(trans
->fs_info
, node
->bytenr
,
2355 node
->num_bytes
, 1);
2359 if (node
->type
== BTRFS_TREE_BLOCK_REF_KEY
||
2360 node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2361 ret
= run_delayed_tree_ref(trans
, node
, extent_op
,
2363 else if (node
->type
== BTRFS_EXTENT_DATA_REF_KEY
||
2364 node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2365 ret
= run_delayed_data_ref(trans
, node
, extent_op
,
2372 static inline struct btrfs_delayed_ref_node
*
2373 select_delayed_ref(struct btrfs_delayed_ref_head
*head
)
2375 struct btrfs_delayed_ref_node
*ref
;
2377 if (RB_EMPTY_ROOT(&head
->ref_tree
))
2381 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2382 * This is to prevent a ref count from going down to zero, which deletes
2383 * the extent item from the extent tree, when there still are references
2384 * to add, which would fail because they would not find the extent item.
2386 if (!list_empty(&head
->ref_add_list
))
2387 return list_first_entry(&head
->ref_add_list
,
2388 struct btrfs_delayed_ref_node
, add_list
);
2390 ref
= rb_entry(rb_first(&head
->ref_tree
),
2391 struct btrfs_delayed_ref_node
, ref_node
);
2392 ASSERT(list_empty(&ref
->add_list
));
2396 static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root
*delayed_refs
,
2397 struct btrfs_delayed_ref_head
*head
)
2399 spin_lock(&delayed_refs
->lock
);
2400 head
->processing
= 0;
2401 delayed_refs
->num_heads_ready
++;
2402 spin_unlock(&delayed_refs
->lock
);
2403 btrfs_delayed_ref_unlock(head
);
2406 static int cleanup_extent_op(struct btrfs_trans_handle
*trans
,
2407 struct btrfs_delayed_ref_head
*head
)
2409 struct btrfs_delayed_extent_op
*extent_op
= head
->extent_op
;
2414 head
->extent_op
= NULL
;
2415 if (head
->must_insert_reserved
) {
2416 btrfs_free_delayed_extent_op(extent_op
);
2419 spin_unlock(&head
->lock
);
2420 ret
= run_delayed_extent_op(trans
, head
, extent_op
);
2421 btrfs_free_delayed_extent_op(extent_op
);
2422 return ret
? ret
: 1;
2425 static int cleanup_ref_head(struct btrfs_trans_handle
*trans
,
2426 struct btrfs_delayed_ref_head
*head
)
2429 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2430 struct btrfs_delayed_ref_root
*delayed_refs
;
2433 delayed_refs
= &trans
->transaction
->delayed_refs
;
2435 ret
= cleanup_extent_op(trans
, head
);
2437 unselect_delayed_ref_head(delayed_refs
, head
);
2438 btrfs_debug(fs_info
, "run_delayed_extent_op returned %d", ret
);
2445 * Need to drop our head ref lock and re-acquire the delayed ref lock
2446 * and then re-check to make sure nobody got added.
2448 spin_unlock(&head
->lock
);
2449 spin_lock(&delayed_refs
->lock
);
2450 spin_lock(&head
->lock
);
2451 if (!RB_EMPTY_ROOT(&head
->ref_tree
) || head
->extent_op
) {
2452 spin_unlock(&head
->lock
);
2453 spin_unlock(&delayed_refs
->lock
);
2456 delayed_refs
->num_heads
--;
2457 rb_erase(&head
->href_node
, &delayed_refs
->href_root
);
2458 RB_CLEAR_NODE(&head
->href_node
);
2459 spin_unlock(&head
->lock
);
2460 spin_unlock(&delayed_refs
->lock
);
2461 atomic_dec(&delayed_refs
->num_entries
);
2463 trace_run_delayed_ref_head(fs_info
, head
, 0);
2465 if (head
->total_ref_mod
< 0) {
2466 struct btrfs_space_info
*space_info
;
2470 flags
= BTRFS_BLOCK_GROUP_DATA
;
2471 else if (head
->is_system
)
2472 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
2474 flags
= BTRFS_BLOCK_GROUP_METADATA
;
2475 space_info
= __find_space_info(fs_info
, flags
);
2477 percpu_counter_add_batch(&space_info
->total_bytes_pinned
,
2479 BTRFS_TOTAL_BYTES_PINNED_BATCH
);
2481 if (head
->is_data
) {
2482 spin_lock(&delayed_refs
->lock
);
2483 delayed_refs
->pending_csums
-= head
->num_bytes
;
2484 spin_unlock(&delayed_refs
->lock
);
2488 if (head
->must_insert_reserved
) {
2489 btrfs_pin_extent(fs_info
, head
->bytenr
,
2490 head
->num_bytes
, 1);
2491 if (head
->is_data
) {
2492 ret
= btrfs_del_csums(trans
, fs_info
, head
->bytenr
,
2497 /* Also free its reserved qgroup space */
2498 btrfs_qgroup_free_delayed_ref(fs_info
, head
->qgroup_ref_root
,
2499 head
->qgroup_reserved
);
2500 btrfs_delayed_ref_unlock(head
);
2501 btrfs_put_delayed_ref_head(head
);
2506 * Returns 0 on success or if called with an already aborted transaction.
2507 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2509 static noinline
int __btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2512 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2513 struct btrfs_delayed_ref_root
*delayed_refs
;
2514 struct btrfs_delayed_ref_node
*ref
;
2515 struct btrfs_delayed_ref_head
*locked_ref
= NULL
;
2516 struct btrfs_delayed_extent_op
*extent_op
;
2517 ktime_t start
= ktime_get();
2519 unsigned long count
= 0;
2520 unsigned long actual_count
= 0;
2521 int must_insert_reserved
= 0;
2523 delayed_refs
= &trans
->transaction
->delayed_refs
;
2529 spin_lock(&delayed_refs
->lock
);
2530 locked_ref
= btrfs_select_ref_head(trans
);
2532 spin_unlock(&delayed_refs
->lock
);
2536 /* grab the lock that says we are going to process
2537 * all the refs for this head */
2538 ret
= btrfs_delayed_ref_lock(trans
, locked_ref
);
2539 spin_unlock(&delayed_refs
->lock
);
2541 * we may have dropped the spin lock to get the head
2542 * mutex lock, and that might have given someone else
2543 * time to free the head. If that's true, it has been
2544 * removed from our list and we can move on.
2546 if (ret
== -EAGAIN
) {
2554 * We need to try and merge add/drops of the same ref since we
2555 * can run into issues with relocate dropping the implicit ref
2556 * and then it being added back again before the drop can
2557 * finish. If we merged anything we need to re-loop so we can
2559 * Or we can get node references of the same type that weren't
2560 * merged when created due to bumps in the tree mod seq, and
2561 * we need to merge them to prevent adding an inline extent
2562 * backref before dropping it (triggering a BUG_ON at
2563 * insert_inline_extent_backref()).
2565 spin_lock(&locked_ref
->lock
);
2566 btrfs_merge_delayed_refs(trans
, delayed_refs
, locked_ref
);
2568 ref
= select_delayed_ref(locked_ref
);
2570 if (ref
&& ref
->seq
&&
2571 btrfs_check_delayed_seq(fs_info
, ref
->seq
)) {
2572 spin_unlock(&locked_ref
->lock
);
2573 unselect_delayed_ref_head(delayed_refs
, locked_ref
);
2581 * We're done processing refs in this ref_head, clean everything
2582 * up and move on to the next ref_head.
2585 ret
= cleanup_ref_head(trans
, locked_ref
);
2587 /* We dropped our lock, we need to loop. */
2600 rb_erase(&ref
->ref_node
, &locked_ref
->ref_tree
);
2601 RB_CLEAR_NODE(&ref
->ref_node
);
2602 if (!list_empty(&ref
->add_list
))
2603 list_del(&ref
->add_list
);
2605 * When we play the delayed ref, also correct the ref_mod on
2608 switch (ref
->action
) {
2609 case BTRFS_ADD_DELAYED_REF
:
2610 case BTRFS_ADD_DELAYED_EXTENT
:
2611 locked_ref
->ref_mod
-= ref
->ref_mod
;
2613 case BTRFS_DROP_DELAYED_REF
:
2614 locked_ref
->ref_mod
+= ref
->ref_mod
;
2619 atomic_dec(&delayed_refs
->num_entries
);
2622 * Record the must-insert_reserved flag before we drop the spin
2625 must_insert_reserved
= locked_ref
->must_insert_reserved
;
2626 locked_ref
->must_insert_reserved
= 0;
2628 extent_op
= locked_ref
->extent_op
;
2629 locked_ref
->extent_op
= NULL
;
2630 spin_unlock(&locked_ref
->lock
);
2632 ret
= run_one_delayed_ref(trans
, ref
, extent_op
,
2633 must_insert_reserved
);
2635 btrfs_free_delayed_extent_op(extent_op
);
2637 unselect_delayed_ref_head(delayed_refs
, locked_ref
);
2638 btrfs_put_delayed_ref(ref
);
2639 btrfs_debug(fs_info
, "run_one_delayed_ref returned %d",
2644 btrfs_put_delayed_ref(ref
);
2650 * We don't want to include ref heads since we can have empty ref heads
2651 * and those will drastically skew our runtime down since we just do
2652 * accounting, no actual extent tree updates.
2654 if (actual_count
> 0) {
2655 u64 runtime
= ktime_to_ns(ktime_sub(ktime_get(), start
));
2659 * We weigh the current average higher than our current runtime
2660 * to avoid large swings in the average.
2662 spin_lock(&delayed_refs
->lock
);
2663 avg
= fs_info
->avg_delayed_ref_runtime
* 3 + runtime
;
2664 fs_info
->avg_delayed_ref_runtime
= avg
>> 2; /* div by 4 */
2665 spin_unlock(&delayed_refs
->lock
);
2670 #ifdef SCRAMBLE_DELAYED_REFS
2672 * Normally delayed refs get processed in ascending bytenr order. This
2673 * correlates in most cases to the order added. To expose dependencies on this
2674 * order, we start to process the tree in the middle instead of the beginning
2676 static u64
find_middle(struct rb_root
*root
)
2678 struct rb_node
*n
= root
->rb_node
;
2679 struct btrfs_delayed_ref_node
*entry
;
2682 u64 first
= 0, last
= 0;
2686 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2687 first
= entry
->bytenr
;
2691 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2692 last
= entry
->bytenr
;
2697 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2698 WARN_ON(!entry
->in_tree
);
2700 middle
= entry
->bytenr
;
2713 static inline u64
heads_to_leaves(struct btrfs_fs_info
*fs_info
, u64 heads
)
2717 num_bytes
= heads
* (sizeof(struct btrfs_extent_item
) +
2718 sizeof(struct btrfs_extent_inline_ref
));
2719 if (!btrfs_fs_incompat(fs_info
, SKINNY_METADATA
))
2720 num_bytes
+= heads
* sizeof(struct btrfs_tree_block_info
);
2723 * We don't ever fill up leaves all the way so multiply by 2 just to be
2724 * closer to what we're really going to want to use.
2726 return div_u64(num_bytes
, BTRFS_LEAF_DATA_SIZE(fs_info
));
2730 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2731 * would require to store the csums for that many bytes.
2733 u64
btrfs_csum_bytes_to_leaves(struct btrfs_fs_info
*fs_info
, u64 csum_bytes
)
2736 u64 num_csums_per_leaf
;
2739 csum_size
= BTRFS_MAX_ITEM_SIZE(fs_info
);
2740 num_csums_per_leaf
= div64_u64(csum_size
,
2741 (u64
)btrfs_super_csum_size(fs_info
->super_copy
));
2742 num_csums
= div64_u64(csum_bytes
, fs_info
->sectorsize
);
2743 num_csums
+= num_csums_per_leaf
- 1;
2744 num_csums
= div64_u64(num_csums
, num_csums_per_leaf
);
2748 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle
*trans
,
2749 struct btrfs_fs_info
*fs_info
)
2751 struct btrfs_block_rsv
*global_rsv
;
2752 u64 num_heads
= trans
->transaction
->delayed_refs
.num_heads_ready
;
2753 u64 csum_bytes
= trans
->transaction
->delayed_refs
.pending_csums
;
2754 unsigned int num_dirty_bgs
= trans
->transaction
->num_dirty_bgs
;
2755 u64 num_bytes
, num_dirty_bgs_bytes
;
2758 num_bytes
= btrfs_calc_trans_metadata_size(fs_info
, 1);
2759 num_heads
= heads_to_leaves(fs_info
, num_heads
);
2761 num_bytes
+= (num_heads
- 1) * fs_info
->nodesize
;
2763 num_bytes
+= btrfs_csum_bytes_to_leaves(fs_info
, csum_bytes
) *
2765 num_dirty_bgs_bytes
= btrfs_calc_trans_metadata_size(fs_info
,
2767 global_rsv
= &fs_info
->global_block_rsv
;
2770 * If we can't allocate any more chunks lets make sure we have _lots_ of
2771 * wiggle room since running delayed refs can create more delayed refs.
2773 if (global_rsv
->space_info
->full
) {
2774 num_dirty_bgs_bytes
<<= 1;
2778 spin_lock(&global_rsv
->lock
);
2779 if (global_rsv
->reserved
<= num_bytes
+ num_dirty_bgs_bytes
)
2781 spin_unlock(&global_rsv
->lock
);
2785 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle
*trans
,
2786 struct btrfs_fs_info
*fs_info
)
2789 atomic_read(&trans
->transaction
->delayed_refs
.num_entries
);
2794 avg_runtime
= fs_info
->avg_delayed_ref_runtime
;
2795 val
= num_entries
* avg_runtime
;
2796 if (val
>= NSEC_PER_SEC
)
2798 if (val
>= NSEC_PER_SEC
/ 2)
2801 return btrfs_check_space_for_delayed_refs(trans
, fs_info
);
2804 struct async_delayed_refs
{
2805 struct btrfs_root
*root
;
2810 struct completion wait
;
2811 struct btrfs_work work
;
2814 static inline struct async_delayed_refs
*
2815 to_async_delayed_refs(struct btrfs_work
*work
)
2817 return container_of(work
, struct async_delayed_refs
, work
);
2820 static void delayed_ref_async_start(struct btrfs_work
*work
)
2822 struct async_delayed_refs
*async
= to_async_delayed_refs(work
);
2823 struct btrfs_trans_handle
*trans
;
2824 struct btrfs_fs_info
*fs_info
= async
->root
->fs_info
;
2827 /* if the commit is already started, we don't need to wait here */
2828 if (btrfs_transaction_blocked(fs_info
))
2831 trans
= btrfs_join_transaction(async
->root
);
2832 if (IS_ERR(trans
)) {
2833 async
->error
= PTR_ERR(trans
);
2838 * trans->sync means that when we call end_transaction, we won't
2839 * wait on delayed refs
2843 /* Don't bother flushing if we got into a different transaction */
2844 if (trans
->transid
> async
->transid
)
2847 ret
= btrfs_run_delayed_refs(trans
, async
->count
);
2851 ret
= btrfs_end_transaction(trans
);
2852 if (ret
&& !async
->error
)
2856 complete(&async
->wait
);
2861 int btrfs_async_run_delayed_refs(struct btrfs_fs_info
*fs_info
,
2862 unsigned long count
, u64 transid
, int wait
)
2864 struct async_delayed_refs
*async
;
2867 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
2871 async
->root
= fs_info
->tree_root
;
2872 async
->count
= count
;
2874 async
->transid
= transid
;
2879 init_completion(&async
->wait
);
2881 btrfs_init_work(&async
->work
, btrfs_extent_refs_helper
,
2882 delayed_ref_async_start
, NULL
, NULL
);
2884 btrfs_queue_work(fs_info
->extent_workers
, &async
->work
);
2887 wait_for_completion(&async
->wait
);
2896 * this starts processing the delayed reference count updates and
2897 * extent insertions we have queued up so far. count can be
2898 * 0, which means to process everything in the tree at the start
2899 * of the run (but not newly added entries), or it can be some target
2900 * number you'd like to process.
2902 * Returns 0 on success or if called with an aborted transaction
2903 * Returns <0 on error and aborts the transaction
2905 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2906 unsigned long count
)
2908 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2909 struct rb_node
*node
;
2910 struct btrfs_delayed_ref_root
*delayed_refs
;
2911 struct btrfs_delayed_ref_head
*head
;
2913 int run_all
= count
== (unsigned long)-1;
2914 bool can_flush_pending_bgs
= trans
->can_flush_pending_bgs
;
2916 /* We'll clean this up in btrfs_cleanup_transaction */
2920 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE
, &fs_info
->flags
))
2923 delayed_refs
= &trans
->transaction
->delayed_refs
;
2925 count
= atomic_read(&delayed_refs
->num_entries
) * 2;
2928 #ifdef SCRAMBLE_DELAYED_REFS
2929 delayed_refs
->run_delayed_start
= find_middle(&delayed_refs
->root
);
2931 trans
->can_flush_pending_bgs
= false;
2932 ret
= __btrfs_run_delayed_refs(trans
, count
);
2934 btrfs_abort_transaction(trans
, ret
);
2939 if (!list_empty(&trans
->new_bgs
))
2940 btrfs_create_pending_block_groups(trans
);
2942 spin_lock(&delayed_refs
->lock
);
2943 node
= rb_first(&delayed_refs
->href_root
);
2945 spin_unlock(&delayed_refs
->lock
);
2948 head
= rb_entry(node
, struct btrfs_delayed_ref_head
,
2950 refcount_inc(&head
->refs
);
2951 spin_unlock(&delayed_refs
->lock
);
2953 /* Mutex was contended, block until it's released and retry. */
2954 mutex_lock(&head
->mutex
);
2955 mutex_unlock(&head
->mutex
);
2957 btrfs_put_delayed_ref_head(head
);
2962 trans
->can_flush_pending_bgs
= can_flush_pending_bgs
;
2966 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
2967 struct btrfs_fs_info
*fs_info
,
2968 u64 bytenr
, u64 num_bytes
, u64 flags
,
2969 int level
, int is_data
)
2971 struct btrfs_delayed_extent_op
*extent_op
;
2974 extent_op
= btrfs_alloc_delayed_extent_op();
2978 extent_op
->flags_to_set
= flags
;
2979 extent_op
->update_flags
= true;
2980 extent_op
->update_key
= false;
2981 extent_op
->is_data
= is_data
? true : false;
2982 extent_op
->level
= level
;
2984 ret
= btrfs_add_delayed_extent_op(fs_info
, trans
, bytenr
,
2985 num_bytes
, extent_op
);
2987 btrfs_free_delayed_extent_op(extent_op
);
2991 static noinline
int check_delayed_ref(struct btrfs_root
*root
,
2992 struct btrfs_path
*path
,
2993 u64 objectid
, u64 offset
, u64 bytenr
)
2995 struct btrfs_delayed_ref_head
*head
;
2996 struct btrfs_delayed_ref_node
*ref
;
2997 struct btrfs_delayed_data_ref
*data_ref
;
2998 struct btrfs_delayed_ref_root
*delayed_refs
;
2999 struct btrfs_transaction
*cur_trans
;
3000 struct rb_node
*node
;
3003 spin_lock(&root
->fs_info
->trans_lock
);
3004 cur_trans
= root
->fs_info
->running_transaction
;
3006 refcount_inc(&cur_trans
->use_count
);
3007 spin_unlock(&root
->fs_info
->trans_lock
);
3011 delayed_refs
= &cur_trans
->delayed_refs
;
3012 spin_lock(&delayed_refs
->lock
);
3013 head
= btrfs_find_delayed_ref_head(delayed_refs
, bytenr
);
3015 spin_unlock(&delayed_refs
->lock
);
3016 btrfs_put_transaction(cur_trans
);
3020 if (!mutex_trylock(&head
->mutex
)) {
3021 refcount_inc(&head
->refs
);
3022 spin_unlock(&delayed_refs
->lock
);
3024 btrfs_release_path(path
);
3027 * Mutex was contended, block until it's released and let
3030 mutex_lock(&head
->mutex
);
3031 mutex_unlock(&head
->mutex
);
3032 btrfs_put_delayed_ref_head(head
);
3033 btrfs_put_transaction(cur_trans
);
3036 spin_unlock(&delayed_refs
->lock
);
3038 spin_lock(&head
->lock
);
3040 * XXX: We should replace this with a proper search function in the
3043 for (node
= rb_first(&head
->ref_tree
); node
; node
= rb_next(node
)) {
3044 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, ref_node
);
3045 /* If it's a shared ref we know a cross reference exists */
3046 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
) {
3051 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
3054 * If our ref doesn't match the one we're currently looking at
3055 * then we have a cross reference.
3057 if (data_ref
->root
!= root
->root_key
.objectid
||
3058 data_ref
->objectid
!= objectid
||
3059 data_ref
->offset
!= offset
) {
3064 spin_unlock(&head
->lock
);
3065 mutex_unlock(&head
->mutex
);
3066 btrfs_put_transaction(cur_trans
);
3070 static noinline
int check_committed_ref(struct btrfs_root
*root
,
3071 struct btrfs_path
*path
,
3072 u64 objectid
, u64 offset
, u64 bytenr
)
3074 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3075 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
3076 struct extent_buffer
*leaf
;
3077 struct btrfs_extent_data_ref
*ref
;
3078 struct btrfs_extent_inline_ref
*iref
;
3079 struct btrfs_extent_item
*ei
;
3080 struct btrfs_key key
;
3085 key
.objectid
= bytenr
;
3086 key
.offset
= (u64
)-1;
3087 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
3089 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
3092 BUG_ON(ret
== 0); /* Corruption */
3095 if (path
->slots
[0] == 0)
3099 leaf
= path
->nodes
[0];
3100 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
3102 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
3106 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
3107 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
3109 if (item_size
!= sizeof(*ei
) +
3110 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
3113 if (btrfs_extent_generation(leaf
, ei
) <=
3114 btrfs_root_last_snapshot(&root
->root_item
))
3117 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
3119 type
= btrfs_get_extent_inline_ref_type(leaf
, iref
, BTRFS_REF_TYPE_DATA
);
3120 if (type
!= BTRFS_EXTENT_DATA_REF_KEY
)
3123 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
3124 if (btrfs_extent_refs(leaf
, ei
) !=
3125 btrfs_extent_data_ref_count(leaf
, ref
) ||
3126 btrfs_extent_data_ref_root(leaf
, ref
) !=
3127 root
->root_key
.objectid
||
3128 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
3129 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
3137 int btrfs_cross_ref_exist(struct btrfs_root
*root
, u64 objectid
, u64 offset
,
3140 struct btrfs_path
*path
;
3144 path
= btrfs_alloc_path();
3149 ret
= check_committed_ref(root
, path
, objectid
,
3151 if (ret
&& ret
!= -ENOENT
)
3154 ret2
= check_delayed_ref(root
, path
, objectid
,
3156 } while (ret2
== -EAGAIN
);
3158 if (ret2
&& ret2
!= -ENOENT
) {
3163 if (ret
!= -ENOENT
|| ret2
!= -ENOENT
)
3166 btrfs_free_path(path
);
3167 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
3172 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
3173 struct btrfs_root
*root
,
3174 struct extent_buffer
*buf
,
3175 int full_backref
, int inc
)
3177 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3183 struct btrfs_key key
;
3184 struct btrfs_file_extent_item
*fi
;
3188 int (*process_func
)(struct btrfs_trans_handle
*,
3189 struct btrfs_root
*,
3190 u64
, u64
, u64
, u64
, u64
, u64
);
3193 if (btrfs_is_testing(fs_info
))
3196 ref_root
= btrfs_header_owner(buf
);
3197 nritems
= btrfs_header_nritems(buf
);
3198 level
= btrfs_header_level(buf
);
3200 if (!test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) && level
== 0)
3204 process_func
= btrfs_inc_extent_ref
;
3206 process_func
= btrfs_free_extent
;
3209 parent
= buf
->start
;
3213 for (i
= 0; i
< nritems
; i
++) {
3215 btrfs_item_key_to_cpu(buf
, &key
, i
);
3216 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
3218 fi
= btrfs_item_ptr(buf
, i
,
3219 struct btrfs_file_extent_item
);
3220 if (btrfs_file_extent_type(buf
, fi
) ==
3221 BTRFS_FILE_EXTENT_INLINE
)
3223 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
3227 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
3228 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
3229 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
3230 parent
, ref_root
, key
.objectid
,
3235 bytenr
= btrfs_node_blockptr(buf
, i
);
3236 num_bytes
= fs_info
->nodesize
;
3237 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
3238 parent
, ref_root
, level
- 1, 0);
3248 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3249 struct extent_buffer
*buf
, int full_backref
)
3251 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1);
3254 int btrfs_dec_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3255 struct extent_buffer
*buf
, int full_backref
)
3257 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 0);
3260 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
3261 struct btrfs_fs_info
*fs_info
,
3262 struct btrfs_path
*path
,
3263 struct btrfs_block_group_cache
*cache
)
3266 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
3268 struct extent_buffer
*leaf
;
3270 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
3277 leaf
= path
->nodes
[0];
3278 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
3279 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
3280 btrfs_mark_buffer_dirty(leaf
);
3282 btrfs_release_path(path
);
3287 static struct btrfs_block_group_cache
*
3288 next_block_group(struct btrfs_fs_info
*fs_info
,
3289 struct btrfs_block_group_cache
*cache
)
3291 struct rb_node
*node
;
3293 spin_lock(&fs_info
->block_group_cache_lock
);
3295 /* If our block group was removed, we need a full search. */
3296 if (RB_EMPTY_NODE(&cache
->cache_node
)) {
3297 const u64 next_bytenr
= cache
->key
.objectid
+ cache
->key
.offset
;
3299 spin_unlock(&fs_info
->block_group_cache_lock
);
3300 btrfs_put_block_group(cache
);
3301 cache
= btrfs_lookup_first_block_group(fs_info
, next_bytenr
); return cache
;
3303 node
= rb_next(&cache
->cache_node
);
3304 btrfs_put_block_group(cache
);
3306 cache
= rb_entry(node
, struct btrfs_block_group_cache
,
3308 btrfs_get_block_group(cache
);
3311 spin_unlock(&fs_info
->block_group_cache_lock
);
3315 static int cache_save_setup(struct btrfs_block_group_cache
*block_group
,
3316 struct btrfs_trans_handle
*trans
,
3317 struct btrfs_path
*path
)
3319 struct btrfs_fs_info
*fs_info
= block_group
->fs_info
;
3320 struct btrfs_root
*root
= fs_info
->tree_root
;
3321 struct inode
*inode
= NULL
;
3322 struct extent_changeset
*data_reserved
= NULL
;
3324 int dcs
= BTRFS_DC_ERROR
;
3330 * If this block group is smaller than 100 megs don't bother caching the
3333 if (block_group
->key
.offset
< (100 * SZ_1M
)) {
3334 spin_lock(&block_group
->lock
);
3335 block_group
->disk_cache_state
= BTRFS_DC_WRITTEN
;
3336 spin_unlock(&block_group
->lock
);
3343 inode
= lookup_free_space_inode(fs_info
, block_group
, path
);
3344 if (IS_ERR(inode
) && PTR_ERR(inode
) != -ENOENT
) {
3345 ret
= PTR_ERR(inode
);
3346 btrfs_release_path(path
);
3350 if (IS_ERR(inode
)) {
3354 if (block_group
->ro
)
3357 ret
= create_free_space_inode(fs_info
, trans
, block_group
,
3365 * We want to set the generation to 0, that way if anything goes wrong
3366 * from here on out we know not to trust this cache when we load up next
3369 BTRFS_I(inode
)->generation
= 0;
3370 ret
= btrfs_update_inode(trans
, root
, inode
);
3373 * So theoretically we could recover from this, simply set the
3374 * super cache generation to 0 so we know to invalidate the
3375 * cache, but then we'd have to keep track of the block groups
3376 * that fail this way so we know we _have_ to reset this cache
3377 * before the next commit or risk reading stale cache. So to
3378 * limit our exposure to horrible edge cases lets just abort the
3379 * transaction, this only happens in really bad situations
3382 btrfs_abort_transaction(trans
, ret
);
3387 /* We've already setup this transaction, go ahead and exit */
3388 if (block_group
->cache_generation
== trans
->transid
&&
3389 i_size_read(inode
)) {
3390 dcs
= BTRFS_DC_SETUP
;
3394 if (i_size_read(inode
) > 0) {
3395 ret
= btrfs_check_trunc_cache_free_space(fs_info
,
3396 &fs_info
->global_block_rsv
);
3400 ret
= btrfs_truncate_free_space_cache(trans
, NULL
, inode
);
3405 spin_lock(&block_group
->lock
);
3406 if (block_group
->cached
!= BTRFS_CACHE_FINISHED
||
3407 !btrfs_test_opt(fs_info
, SPACE_CACHE
)) {
3409 * don't bother trying to write stuff out _if_
3410 * a) we're not cached,
3411 * b) we're with nospace_cache mount option,
3412 * c) we're with v2 space_cache (FREE_SPACE_TREE).
3414 dcs
= BTRFS_DC_WRITTEN
;
3415 spin_unlock(&block_group
->lock
);
3418 spin_unlock(&block_group
->lock
);
3421 * We hit an ENOSPC when setting up the cache in this transaction, just
3422 * skip doing the setup, we've already cleared the cache so we're safe.
3424 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC
, &trans
->transaction
->flags
)) {
3430 * Try to preallocate enough space based on how big the block group is.
3431 * Keep in mind this has to include any pinned space which could end up
3432 * taking up quite a bit since it's not folded into the other space
3435 num_pages
= div_u64(block_group
->key
.offset
, SZ_256M
);
3440 num_pages
*= PAGE_SIZE
;
3442 ret
= btrfs_check_data_free_space(inode
, &data_reserved
, 0, num_pages
);
3446 ret
= btrfs_prealloc_file_range_trans(inode
, trans
, 0, 0, num_pages
,
3447 num_pages
, num_pages
,
3450 * Our cache requires contiguous chunks so that we don't modify a bunch
3451 * of metadata or split extents when writing the cache out, which means
3452 * we can enospc if we are heavily fragmented in addition to just normal
3453 * out of space conditions. So if we hit this just skip setting up any
3454 * other block groups for this transaction, maybe we'll unpin enough
3455 * space the next time around.
3458 dcs
= BTRFS_DC_SETUP
;
3459 else if (ret
== -ENOSPC
)
3460 set_bit(BTRFS_TRANS_CACHE_ENOSPC
, &trans
->transaction
->flags
);
3465 btrfs_release_path(path
);
3467 spin_lock(&block_group
->lock
);
3468 if (!ret
&& dcs
== BTRFS_DC_SETUP
)
3469 block_group
->cache_generation
= trans
->transid
;
3470 block_group
->disk_cache_state
= dcs
;
3471 spin_unlock(&block_group
->lock
);
3473 extent_changeset_free(data_reserved
);
3477 int btrfs_setup_space_cache(struct btrfs_trans_handle
*trans
,
3478 struct btrfs_fs_info
*fs_info
)
3480 struct btrfs_block_group_cache
*cache
, *tmp
;
3481 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3482 struct btrfs_path
*path
;
3484 if (list_empty(&cur_trans
->dirty_bgs
) ||
3485 !btrfs_test_opt(fs_info
, SPACE_CACHE
))
3488 path
= btrfs_alloc_path();
3492 /* Could add new block groups, use _safe just in case */
3493 list_for_each_entry_safe(cache
, tmp
, &cur_trans
->dirty_bgs
,
3495 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
)
3496 cache_save_setup(cache
, trans
, path
);
3499 btrfs_free_path(path
);
3504 * transaction commit does final block group cache writeback during a
3505 * critical section where nothing is allowed to change the FS. This is
3506 * required in order for the cache to actually match the block group,
3507 * but can introduce a lot of latency into the commit.
3509 * So, btrfs_start_dirty_block_groups is here to kick off block group
3510 * cache IO. There's a chance we'll have to redo some of it if the
3511 * block group changes again during the commit, but it greatly reduces
3512 * the commit latency by getting rid of the easy block groups while
3513 * we're still allowing others to join the commit.
3515 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle
*trans
)
3517 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
3518 struct btrfs_block_group_cache
*cache
;
3519 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3522 struct btrfs_path
*path
= NULL
;
3524 struct list_head
*io
= &cur_trans
->io_bgs
;
3525 int num_started
= 0;
3528 spin_lock(&cur_trans
->dirty_bgs_lock
);
3529 if (list_empty(&cur_trans
->dirty_bgs
)) {
3530 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3533 list_splice_init(&cur_trans
->dirty_bgs
, &dirty
);
3534 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3538 * make sure all the block groups on our dirty list actually
3541 btrfs_create_pending_block_groups(trans
);
3544 path
= btrfs_alloc_path();
3550 * cache_write_mutex is here only to save us from balance or automatic
3551 * removal of empty block groups deleting this block group while we are
3552 * writing out the cache
3554 mutex_lock(&trans
->transaction
->cache_write_mutex
);
3555 while (!list_empty(&dirty
)) {
3556 cache
= list_first_entry(&dirty
,
3557 struct btrfs_block_group_cache
,
3560 * this can happen if something re-dirties a block
3561 * group that is already under IO. Just wait for it to
3562 * finish and then do it all again
3564 if (!list_empty(&cache
->io_list
)) {
3565 list_del_init(&cache
->io_list
);
3566 btrfs_wait_cache_io(trans
, cache
, path
);
3567 btrfs_put_block_group(cache
);
3572 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3573 * if it should update the cache_state. Don't delete
3574 * until after we wait.
3576 * Since we're not running in the commit critical section
3577 * we need the dirty_bgs_lock to protect from update_block_group
3579 spin_lock(&cur_trans
->dirty_bgs_lock
);
3580 list_del_init(&cache
->dirty_list
);
3581 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3585 cache_save_setup(cache
, trans
, path
);
3587 if (cache
->disk_cache_state
== BTRFS_DC_SETUP
) {
3588 cache
->io_ctl
.inode
= NULL
;
3589 ret
= btrfs_write_out_cache(fs_info
, trans
,
3591 if (ret
== 0 && cache
->io_ctl
.inode
) {
3596 * The cache_write_mutex is protecting the
3597 * io_list, also refer to the definition of
3598 * btrfs_transaction::io_bgs for more details
3600 list_add_tail(&cache
->io_list
, io
);
3603 * if we failed to write the cache, the
3604 * generation will be bad and life goes on
3610 ret
= write_one_cache_group(trans
, fs_info
,
3613 * Our block group might still be attached to the list
3614 * of new block groups in the transaction handle of some
3615 * other task (struct btrfs_trans_handle->new_bgs). This
3616 * means its block group item isn't yet in the extent
3617 * tree. If this happens ignore the error, as we will
3618 * try again later in the critical section of the
3619 * transaction commit.
3621 if (ret
== -ENOENT
) {
3623 spin_lock(&cur_trans
->dirty_bgs_lock
);
3624 if (list_empty(&cache
->dirty_list
)) {
3625 list_add_tail(&cache
->dirty_list
,
3626 &cur_trans
->dirty_bgs
);
3627 btrfs_get_block_group(cache
);
3629 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3631 btrfs_abort_transaction(trans
, ret
);
3635 /* if its not on the io list, we need to put the block group */
3637 btrfs_put_block_group(cache
);
3643 * Avoid blocking other tasks for too long. It might even save
3644 * us from writing caches for block groups that are going to be
3647 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
3648 mutex_lock(&trans
->transaction
->cache_write_mutex
);
3650 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
3653 * go through delayed refs for all the stuff we've just kicked off
3654 * and then loop back (just once)
3656 ret
= btrfs_run_delayed_refs(trans
, 0);
3657 if (!ret
&& loops
== 0) {
3659 spin_lock(&cur_trans
->dirty_bgs_lock
);
3660 list_splice_init(&cur_trans
->dirty_bgs
, &dirty
);
3662 * dirty_bgs_lock protects us from concurrent block group
3663 * deletes too (not just cache_write_mutex).
3665 if (!list_empty(&dirty
)) {
3666 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3669 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3670 } else if (ret
< 0) {
3671 btrfs_cleanup_dirty_bgs(cur_trans
, fs_info
);
3674 btrfs_free_path(path
);
3678 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
3679 struct btrfs_fs_info
*fs_info
)
3681 struct btrfs_block_group_cache
*cache
;
3682 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3685 struct btrfs_path
*path
;
3686 struct list_head
*io
= &cur_trans
->io_bgs
;
3687 int num_started
= 0;
3689 path
= btrfs_alloc_path();
3694 * Even though we are in the critical section of the transaction commit,
3695 * we can still have concurrent tasks adding elements to this
3696 * transaction's list of dirty block groups. These tasks correspond to
3697 * endio free space workers started when writeback finishes for a
3698 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3699 * allocate new block groups as a result of COWing nodes of the root
3700 * tree when updating the free space inode. The writeback for the space
3701 * caches is triggered by an earlier call to
3702 * btrfs_start_dirty_block_groups() and iterations of the following
3704 * Also we want to do the cache_save_setup first and then run the
3705 * delayed refs to make sure we have the best chance at doing this all
3708 spin_lock(&cur_trans
->dirty_bgs_lock
);
3709 while (!list_empty(&cur_trans
->dirty_bgs
)) {
3710 cache
= list_first_entry(&cur_trans
->dirty_bgs
,
3711 struct btrfs_block_group_cache
,
3715 * this can happen if cache_save_setup re-dirties a block
3716 * group that is already under IO. Just wait for it to
3717 * finish and then do it all again
3719 if (!list_empty(&cache
->io_list
)) {
3720 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3721 list_del_init(&cache
->io_list
);
3722 btrfs_wait_cache_io(trans
, cache
, path
);
3723 btrfs_put_block_group(cache
);
3724 spin_lock(&cur_trans
->dirty_bgs_lock
);
3728 * don't remove from the dirty list until after we've waited
3731 list_del_init(&cache
->dirty_list
);
3732 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3735 cache_save_setup(cache
, trans
, path
);
3738 ret
= btrfs_run_delayed_refs(trans
,
3739 (unsigned long) -1);
3741 if (!ret
&& cache
->disk_cache_state
== BTRFS_DC_SETUP
) {
3742 cache
->io_ctl
.inode
= NULL
;
3743 ret
= btrfs_write_out_cache(fs_info
, trans
,
3745 if (ret
== 0 && cache
->io_ctl
.inode
) {
3748 list_add_tail(&cache
->io_list
, io
);
3751 * if we failed to write the cache, the
3752 * generation will be bad and life goes on
3758 ret
= write_one_cache_group(trans
, fs_info
,
3761 * One of the free space endio workers might have
3762 * created a new block group while updating a free space
3763 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3764 * and hasn't released its transaction handle yet, in
3765 * which case the new block group is still attached to
3766 * its transaction handle and its creation has not
3767 * finished yet (no block group item in the extent tree
3768 * yet, etc). If this is the case, wait for all free
3769 * space endio workers to finish and retry. This is a
3770 * a very rare case so no need for a more efficient and
3773 if (ret
== -ENOENT
) {
3774 wait_event(cur_trans
->writer_wait
,
3775 atomic_read(&cur_trans
->num_writers
) == 1);
3776 ret
= write_one_cache_group(trans
, fs_info
,
3780 btrfs_abort_transaction(trans
, ret
);
3783 /* if its not on the io list, we need to put the block group */
3785 btrfs_put_block_group(cache
);
3786 spin_lock(&cur_trans
->dirty_bgs_lock
);
3788 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3791 * Refer to the definition of io_bgs member for details why it's safe
3792 * to use it without any locking
3794 while (!list_empty(io
)) {
3795 cache
= list_first_entry(io
, struct btrfs_block_group_cache
,
3797 list_del_init(&cache
->io_list
);
3798 btrfs_wait_cache_io(trans
, cache
, path
);
3799 btrfs_put_block_group(cache
);
3802 btrfs_free_path(path
);
3806 int btrfs_extent_readonly(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
3808 struct btrfs_block_group_cache
*block_group
;
3811 block_group
= btrfs_lookup_block_group(fs_info
, bytenr
);
3812 if (!block_group
|| block_group
->ro
)
3815 btrfs_put_block_group(block_group
);
3819 bool btrfs_inc_nocow_writers(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
3821 struct btrfs_block_group_cache
*bg
;
3824 bg
= btrfs_lookup_block_group(fs_info
, bytenr
);
3828 spin_lock(&bg
->lock
);
3832 atomic_inc(&bg
->nocow_writers
);
3833 spin_unlock(&bg
->lock
);
3835 /* no put on block group, done by btrfs_dec_nocow_writers */
3837 btrfs_put_block_group(bg
);
3843 void btrfs_dec_nocow_writers(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
3845 struct btrfs_block_group_cache
*bg
;
3847 bg
= btrfs_lookup_block_group(fs_info
, bytenr
);
3849 if (atomic_dec_and_test(&bg
->nocow_writers
))
3850 wake_up_var(&bg
->nocow_writers
);
3852 * Once for our lookup and once for the lookup done by a previous call
3853 * to btrfs_inc_nocow_writers()
3855 btrfs_put_block_group(bg
);
3856 btrfs_put_block_group(bg
);
3859 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache
*bg
)
3861 wait_var_event(&bg
->nocow_writers
, !atomic_read(&bg
->nocow_writers
));
3864 static const char *alloc_name(u64 flags
)
3867 case BTRFS_BLOCK_GROUP_METADATA
|BTRFS_BLOCK_GROUP_DATA
:
3869 case BTRFS_BLOCK_GROUP_METADATA
:
3871 case BTRFS_BLOCK_GROUP_DATA
:
3873 case BTRFS_BLOCK_GROUP_SYSTEM
:
3877 return "invalid-combination";
3881 static int create_space_info(struct btrfs_fs_info
*info
, u64 flags
)
3884 struct btrfs_space_info
*space_info
;
3888 space_info
= kzalloc(sizeof(*space_info
), GFP_NOFS
);
3892 ret
= percpu_counter_init(&space_info
->total_bytes_pinned
, 0,
3899 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
3900 INIT_LIST_HEAD(&space_info
->block_groups
[i
]);
3901 init_rwsem(&space_info
->groups_sem
);
3902 spin_lock_init(&space_info
->lock
);
3903 space_info
->flags
= flags
& BTRFS_BLOCK_GROUP_TYPE_MASK
;
3904 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3905 init_waitqueue_head(&space_info
->wait
);
3906 INIT_LIST_HEAD(&space_info
->ro_bgs
);
3907 INIT_LIST_HEAD(&space_info
->tickets
);
3908 INIT_LIST_HEAD(&space_info
->priority_tickets
);
3910 ret
= kobject_init_and_add(&space_info
->kobj
, &space_info_ktype
,
3911 info
->space_info_kobj
, "%s",
3912 alloc_name(space_info
->flags
));
3914 percpu_counter_destroy(&space_info
->total_bytes_pinned
);
3919 list_add_rcu(&space_info
->list
, &info
->space_info
);
3920 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3921 info
->data_sinfo
= space_info
;
3926 static void update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
3927 u64 total_bytes
, u64 bytes_used
,
3929 struct btrfs_space_info
**space_info
)
3931 struct btrfs_space_info
*found
;
3934 if (flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3935 BTRFS_BLOCK_GROUP_RAID10
))
3940 found
= __find_space_info(info
, flags
);
3942 spin_lock(&found
->lock
);
3943 found
->total_bytes
+= total_bytes
;
3944 found
->disk_total
+= total_bytes
* factor
;
3945 found
->bytes_used
+= bytes_used
;
3946 found
->disk_used
+= bytes_used
* factor
;
3947 found
->bytes_readonly
+= bytes_readonly
;
3948 if (total_bytes
> 0)
3950 space_info_add_new_bytes(info
, found
, total_bytes
-
3951 bytes_used
- bytes_readonly
);
3952 spin_unlock(&found
->lock
);
3953 *space_info
= found
;
3956 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
3958 u64 extra_flags
= chunk_to_extended(flags
) &
3959 BTRFS_EXTENDED_PROFILE_MASK
;
3961 write_seqlock(&fs_info
->profiles_lock
);
3962 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3963 fs_info
->avail_data_alloc_bits
|= extra_flags
;
3964 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3965 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
3966 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3967 fs_info
->avail_system_alloc_bits
|= extra_flags
;
3968 write_sequnlock(&fs_info
->profiles_lock
);
3972 * returns target flags in extended format or 0 if restripe for this
3973 * chunk_type is not in progress
3975 * should be called with balance_lock held
3977 static u64
get_restripe_target(struct btrfs_fs_info
*fs_info
, u64 flags
)
3979 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3985 if (flags
& BTRFS_BLOCK_GROUP_DATA
&&
3986 bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3987 target
= BTRFS_BLOCK_GROUP_DATA
| bctl
->data
.target
;
3988 } else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
&&
3989 bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3990 target
= BTRFS_BLOCK_GROUP_SYSTEM
| bctl
->sys
.target
;
3991 } else if (flags
& BTRFS_BLOCK_GROUP_METADATA
&&
3992 bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3993 target
= BTRFS_BLOCK_GROUP_METADATA
| bctl
->meta
.target
;
4000 * @flags: available profiles in extended format (see ctree.h)
4002 * Returns reduced profile in chunk format. If profile changing is in
4003 * progress (either running or paused) picks the target profile (if it's
4004 * already available), otherwise falls back to plain reducing.
4006 static u64
btrfs_reduce_alloc_profile(struct btrfs_fs_info
*fs_info
, u64 flags
)
4008 u64 num_devices
= fs_info
->fs_devices
->rw_devices
;
4014 * see if restripe for this chunk_type is in progress, if so
4015 * try to reduce to the target profile
4017 spin_lock(&fs_info
->balance_lock
);
4018 target
= get_restripe_target(fs_info
, flags
);
4020 /* pick target profile only if it's already available */
4021 if ((flags
& target
) & BTRFS_EXTENDED_PROFILE_MASK
) {
4022 spin_unlock(&fs_info
->balance_lock
);
4023 return extended_to_chunk(target
);
4026 spin_unlock(&fs_info
->balance_lock
);
4028 /* First, mask out the RAID levels which aren't possible */
4029 for (raid_type
= 0; raid_type
< BTRFS_NR_RAID_TYPES
; raid_type
++) {
4030 if (num_devices
>= btrfs_raid_array
[raid_type
].devs_min
)
4031 allowed
|= btrfs_raid_array
[raid_type
].bg_flag
;
4035 if (allowed
& BTRFS_BLOCK_GROUP_RAID6
)
4036 allowed
= BTRFS_BLOCK_GROUP_RAID6
;
4037 else if (allowed
& BTRFS_BLOCK_GROUP_RAID5
)
4038 allowed
= BTRFS_BLOCK_GROUP_RAID5
;
4039 else if (allowed
& BTRFS_BLOCK_GROUP_RAID10
)
4040 allowed
= BTRFS_BLOCK_GROUP_RAID10
;
4041 else if (allowed
& BTRFS_BLOCK_GROUP_RAID1
)
4042 allowed
= BTRFS_BLOCK_GROUP_RAID1
;
4043 else if (allowed
& BTRFS_BLOCK_GROUP_RAID0
)
4044 allowed
= BTRFS_BLOCK_GROUP_RAID0
;
4046 flags
&= ~BTRFS_BLOCK_GROUP_PROFILE_MASK
;
4048 return extended_to_chunk(flags
| allowed
);
4051 static u64
get_alloc_profile(struct btrfs_fs_info
*fs_info
, u64 orig_flags
)
4058 seq
= read_seqbegin(&fs_info
->profiles_lock
);
4060 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
4061 flags
|= fs_info
->avail_data_alloc_bits
;
4062 else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
4063 flags
|= fs_info
->avail_system_alloc_bits
;
4064 else if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
4065 flags
|= fs_info
->avail_metadata_alloc_bits
;
4066 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
4068 return btrfs_reduce_alloc_profile(fs_info
, flags
);
4071 static u64
get_alloc_profile_by_root(struct btrfs_root
*root
, int data
)
4073 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4078 flags
= BTRFS_BLOCK_GROUP_DATA
;
4079 else if (root
== fs_info
->chunk_root
)
4080 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
4082 flags
= BTRFS_BLOCK_GROUP_METADATA
;
4084 ret
= get_alloc_profile(fs_info
, flags
);
4088 u64
btrfs_data_alloc_profile(struct btrfs_fs_info
*fs_info
)
4090 return get_alloc_profile(fs_info
, BTRFS_BLOCK_GROUP_DATA
);
4093 u64
btrfs_metadata_alloc_profile(struct btrfs_fs_info
*fs_info
)
4095 return get_alloc_profile(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4098 u64
btrfs_system_alloc_profile(struct btrfs_fs_info
*fs_info
)
4100 return get_alloc_profile(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
4103 static u64
btrfs_space_info_used(struct btrfs_space_info
*s_info
,
4104 bool may_use_included
)
4107 return s_info
->bytes_used
+ s_info
->bytes_reserved
+
4108 s_info
->bytes_pinned
+ s_info
->bytes_readonly
+
4109 (may_use_included
? s_info
->bytes_may_use
: 0);
4112 int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode
*inode
, u64 bytes
)
4114 struct btrfs_root
*root
= inode
->root
;
4115 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4116 struct btrfs_space_info
*data_sinfo
= fs_info
->data_sinfo
;
4119 int need_commit
= 2;
4120 int have_pinned_space
;
4122 /* make sure bytes are sectorsize aligned */
4123 bytes
= ALIGN(bytes
, fs_info
->sectorsize
);
4125 if (btrfs_is_free_space_inode(inode
)) {
4127 ASSERT(current
->journal_info
);
4131 /* make sure we have enough space to handle the data first */
4132 spin_lock(&data_sinfo
->lock
);
4133 used
= btrfs_space_info_used(data_sinfo
, true);
4135 if (used
+ bytes
> data_sinfo
->total_bytes
) {
4136 struct btrfs_trans_handle
*trans
;
4139 * if we don't have enough free bytes in this space then we need
4140 * to alloc a new chunk.
4142 if (!data_sinfo
->full
) {
4145 data_sinfo
->force_alloc
= CHUNK_ALLOC_FORCE
;
4146 spin_unlock(&data_sinfo
->lock
);
4148 alloc_target
= btrfs_data_alloc_profile(fs_info
);
4150 * It is ugly that we don't call nolock join
4151 * transaction for the free space inode case here.
4152 * But it is safe because we only do the data space
4153 * reservation for the free space cache in the
4154 * transaction context, the common join transaction
4155 * just increase the counter of the current transaction
4156 * handler, doesn't try to acquire the trans_lock of
4159 trans
= btrfs_join_transaction(root
);
4161 return PTR_ERR(trans
);
4163 ret
= do_chunk_alloc(trans
, alloc_target
,
4164 CHUNK_ALLOC_NO_FORCE
);
4165 btrfs_end_transaction(trans
);
4170 have_pinned_space
= 1;
4179 * If we don't have enough pinned space to deal with this
4180 * allocation, and no removed chunk in current transaction,
4181 * don't bother committing the transaction.
4183 have_pinned_space
= __percpu_counter_compare(
4184 &data_sinfo
->total_bytes_pinned
,
4185 used
+ bytes
- data_sinfo
->total_bytes
,
4186 BTRFS_TOTAL_BYTES_PINNED_BATCH
);
4187 spin_unlock(&data_sinfo
->lock
);
4189 /* commit the current transaction and try again */
4194 if (need_commit
> 0) {
4195 btrfs_start_delalloc_roots(fs_info
, -1);
4196 btrfs_wait_ordered_roots(fs_info
, U64_MAX
, 0,
4200 trans
= btrfs_join_transaction(root
);
4202 return PTR_ERR(trans
);
4203 if (have_pinned_space
>= 0 ||
4204 test_bit(BTRFS_TRANS_HAVE_FREE_BGS
,
4205 &trans
->transaction
->flags
) ||
4207 ret
= btrfs_commit_transaction(trans
);
4211 * The cleaner kthread might still be doing iput
4212 * operations. Wait for it to finish so that
4213 * more space is released.
4215 mutex_lock(&fs_info
->cleaner_delayed_iput_mutex
);
4216 mutex_unlock(&fs_info
->cleaner_delayed_iput_mutex
);
4219 btrfs_end_transaction(trans
);
4223 trace_btrfs_space_reservation(fs_info
,
4224 "space_info:enospc",
4225 data_sinfo
->flags
, bytes
, 1);
4228 data_sinfo
->bytes_may_use
+= bytes
;
4229 trace_btrfs_space_reservation(fs_info
, "space_info",
4230 data_sinfo
->flags
, bytes
, 1);
4231 spin_unlock(&data_sinfo
->lock
);
4236 int btrfs_check_data_free_space(struct inode
*inode
,
4237 struct extent_changeset
**reserved
, u64 start
, u64 len
)
4239 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
4242 /* align the range */
4243 len
= round_up(start
+ len
, fs_info
->sectorsize
) -
4244 round_down(start
, fs_info
->sectorsize
);
4245 start
= round_down(start
, fs_info
->sectorsize
);
4247 ret
= btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode
), len
);
4251 /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */
4252 ret
= btrfs_qgroup_reserve_data(inode
, reserved
, start
, len
);
4254 btrfs_free_reserved_data_space_noquota(inode
, start
, len
);
4261 * Called if we need to clear a data reservation for this inode
4262 * Normally in a error case.
4264 * This one will *NOT* use accurate qgroup reserved space API, just for case
4265 * which we can't sleep and is sure it won't affect qgroup reserved space.
4266 * Like clear_bit_hook().
4268 void btrfs_free_reserved_data_space_noquota(struct inode
*inode
, u64 start
,
4271 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
4272 struct btrfs_space_info
*data_sinfo
;
4274 /* Make sure the range is aligned to sectorsize */
4275 len
= round_up(start
+ len
, fs_info
->sectorsize
) -
4276 round_down(start
, fs_info
->sectorsize
);
4277 start
= round_down(start
, fs_info
->sectorsize
);
4279 data_sinfo
= fs_info
->data_sinfo
;
4280 spin_lock(&data_sinfo
->lock
);
4281 if (WARN_ON(data_sinfo
->bytes_may_use
< len
))
4282 data_sinfo
->bytes_may_use
= 0;
4284 data_sinfo
->bytes_may_use
-= len
;
4285 trace_btrfs_space_reservation(fs_info
, "space_info",
4286 data_sinfo
->flags
, len
, 0);
4287 spin_unlock(&data_sinfo
->lock
);
4291 * Called if we need to clear a data reservation for this inode
4292 * Normally in a error case.
4294 * This one will handle the per-inode data rsv map for accurate reserved
4297 void btrfs_free_reserved_data_space(struct inode
*inode
,
4298 struct extent_changeset
*reserved
, u64 start
, u64 len
)
4300 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4302 /* Make sure the range is aligned to sectorsize */
4303 len
= round_up(start
+ len
, root
->fs_info
->sectorsize
) -
4304 round_down(start
, root
->fs_info
->sectorsize
);
4305 start
= round_down(start
, root
->fs_info
->sectorsize
);
4307 btrfs_free_reserved_data_space_noquota(inode
, start
, len
);
4308 btrfs_qgroup_free_data(inode
, reserved
, start
, len
);
4311 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
4313 struct list_head
*head
= &info
->space_info
;
4314 struct btrfs_space_info
*found
;
4317 list_for_each_entry_rcu(found
, head
, list
) {
4318 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
4319 found
->force_alloc
= CHUNK_ALLOC_FORCE
;
4324 static inline u64
calc_global_rsv_need_space(struct btrfs_block_rsv
*global
)
4326 return (global
->size
<< 1);
4329 static int should_alloc_chunk(struct btrfs_fs_info
*fs_info
,
4330 struct btrfs_space_info
*sinfo
, int force
)
4332 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
4333 u64 bytes_used
= btrfs_space_info_used(sinfo
, false);
4336 if (force
== CHUNK_ALLOC_FORCE
)
4340 * We need to take into account the global rsv because for all intents
4341 * and purposes it's used space. Don't worry about locking the
4342 * global_rsv, it doesn't change except when the transaction commits.
4344 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
4345 bytes_used
+= calc_global_rsv_need_space(global_rsv
);
4348 * in limited mode, we want to have some free space up to
4349 * about 1% of the FS size.
4351 if (force
== CHUNK_ALLOC_LIMITED
) {
4352 thresh
= btrfs_super_total_bytes(fs_info
->super_copy
);
4353 thresh
= max_t(u64
, SZ_64M
, div_factor_fine(thresh
, 1));
4355 if (sinfo
->total_bytes
- bytes_used
< thresh
)
4359 if (bytes_used
+ SZ_2M
< div_factor(sinfo
->total_bytes
, 8))
4364 static u64
get_profile_num_devs(struct btrfs_fs_info
*fs_info
, u64 type
)
4368 if (type
& (BTRFS_BLOCK_GROUP_RAID10
|
4369 BTRFS_BLOCK_GROUP_RAID0
|
4370 BTRFS_BLOCK_GROUP_RAID5
|
4371 BTRFS_BLOCK_GROUP_RAID6
))
4372 num_dev
= fs_info
->fs_devices
->rw_devices
;
4373 else if (type
& BTRFS_BLOCK_GROUP_RAID1
)
4376 num_dev
= 1; /* DUP or single */
4382 * If @is_allocation is true, reserve space in the system space info necessary
4383 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4386 void check_system_chunk(struct btrfs_trans_handle
*trans
, u64 type
)
4388 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
4389 struct btrfs_space_info
*info
;
4396 * Needed because we can end up allocating a system chunk and for an
4397 * atomic and race free space reservation in the chunk block reserve.
4399 lockdep_assert_held(&fs_info
->chunk_mutex
);
4401 info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
4402 spin_lock(&info
->lock
);
4403 left
= info
->total_bytes
- btrfs_space_info_used(info
, true);
4404 spin_unlock(&info
->lock
);
4406 num_devs
= get_profile_num_devs(fs_info
, type
);
4408 /* num_devs device items to update and 1 chunk item to add or remove */
4409 thresh
= btrfs_calc_trunc_metadata_size(fs_info
, num_devs
) +
4410 btrfs_calc_trans_metadata_size(fs_info
, 1);
4412 if (left
< thresh
&& btrfs_test_opt(fs_info
, ENOSPC_DEBUG
)) {
4413 btrfs_info(fs_info
, "left=%llu, need=%llu, flags=%llu",
4414 left
, thresh
, type
);
4415 dump_space_info(fs_info
, info
, 0, 0);
4418 if (left
< thresh
) {
4419 u64 flags
= btrfs_system_alloc_profile(fs_info
);
4422 * Ignore failure to create system chunk. We might end up not
4423 * needing it, as we might not need to COW all nodes/leafs from
4424 * the paths we visit in the chunk tree (they were already COWed
4425 * or created in the current transaction for example).
4427 ret
= btrfs_alloc_chunk(trans
, flags
);
4431 ret
= btrfs_block_rsv_add(fs_info
->chunk_root
,
4432 &fs_info
->chunk_block_rsv
,
4433 thresh
, BTRFS_RESERVE_NO_FLUSH
);
4435 trans
->chunk_bytes_reserved
+= thresh
;
4440 * If force is CHUNK_ALLOC_FORCE:
4441 * - return 1 if it successfully allocates a chunk,
4442 * - return errors including -ENOSPC otherwise.
4443 * If force is NOT CHUNK_ALLOC_FORCE:
4444 * - return 0 if it doesn't need to allocate a new chunk,
4445 * - return 1 if it successfully allocates a chunk,
4446 * - return errors including -ENOSPC otherwise.
4448 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
, u64 flags
,
4451 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
4452 struct btrfs_space_info
*space_info
;
4453 int wait_for_alloc
= 0;
4456 /* Don't re-enter if we're already allocating a chunk */
4457 if (trans
->allocating_chunk
)
4460 space_info
= __find_space_info(fs_info
, flags
);
4464 spin_lock(&space_info
->lock
);
4465 if (force
< space_info
->force_alloc
)
4466 force
= space_info
->force_alloc
;
4467 if (space_info
->full
) {
4468 if (should_alloc_chunk(fs_info
, space_info
, force
))
4472 spin_unlock(&space_info
->lock
);
4476 if (!should_alloc_chunk(fs_info
, space_info
, force
)) {
4477 spin_unlock(&space_info
->lock
);
4479 } else if (space_info
->chunk_alloc
) {
4482 space_info
->chunk_alloc
= 1;
4485 spin_unlock(&space_info
->lock
);
4487 mutex_lock(&fs_info
->chunk_mutex
);
4490 * The chunk_mutex is held throughout the entirety of a chunk
4491 * allocation, so once we've acquired the chunk_mutex we know that the
4492 * other guy is done and we need to recheck and see if we should
4495 if (wait_for_alloc
) {
4496 mutex_unlock(&fs_info
->chunk_mutex
);
4502 trans
->allocating_chunk
= true;
4505 * If we have mixed data/metadata chunks we want to make sure we keep
4506 * allocating mixed chunks instead of individual chunks.
4508 if (btrfs_mixed_space_info(space_info
))
4509 flags
|= (BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
);
4512 * if we're doing a data chunk, go ahead and make sure that
4513 * we keep a reasonable number of metadata chunks allocated in the
4516 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
4517 fs_info
->data_chunk_allocations
++;
4518 if (!(fs_info
->data_chunk_allocations
%
4519 fs_info
->metadata_ratio
))
4520 force_metadata_allocation(fs_info
);
4524 * Check if we have enough space in SYSTEM chunk because we may need
4525 * to update devices.
4527 check_system_chunk(trans
, flags
);
4529 ret
= btrfs_alloc_chunk(trans
, flags
);
4530 trans
->allocating_chunk
= false;
4532 spin_lock(&space_info
->lock
);
4535 space_info
->full
= 1;
4542 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
4544 space_info
->chunk_alloc
= 0;
4545 spin_unlock(&space_info
->lock
);
4546 mutex_unlock(&fs_info
->chunk_mutex
);
4548 * When we allocate a new chunk we reserve space in the chunk block
4549 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4550 * add new nodes/leafs to it if we end up needing to do it when
4551 * inserting the chunk item and updating device items as part of the
4552 * second phase of chunk allocation, performed by
4553 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4554 * large number of new block groups to create in our transaction
4555 * handle's new_bgs list to avoid exhausting the chunk block reserve
4556 * in extreme cases - like having a single transaction create many new
4557 * block groups when starting to write out the free space caches of all
4558 * the block groups that were made dirty during the lifetime of the
4561 if (trans
->can_flush_pending_bgs
&&
4562 trans
->chunk_bytes_reserved
>= (u64
)SZ_2M
) {
4563 btrfs_create_pending_block_groups(trans
);
4564 btrfs_trans_release_chunk_metadata(trans
);
4569 static int can_overcommit(struct btrfs_fs_info
*fs_info
,
4570 struct btrfs_space_info
*space_info
, u64 bytes
,
4571 enum btrfs_reserve_flush_enum flush
,
4574 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
4580 /* Don't overcommit when in mixed mode. */
4581 if (space_info
->flags
& BTRFS_BLOCK_GROUP_DATA
)
4585 profile
= btrfs_system_alloc_profile(fs_info
);
4587 profile
= btrfs_metadata_alloc_profile(fs_info
);
4589 used
= btrfs_space_info_used(space_info
, false);
4592 * We only want to allow over committing if we have lots of actual space
4593 * free, but if we don't have enough space to handle the global reserve
4594 * space then we could end up having a real enospc problem when trying
4595 * to allocate a chunk or some other such important allocation.
4597 spin_lock(&global_rsv
->lock
);
4598 space_size
= calc_global_rsv_need_space(global_rsv
);
4599 spin_unlock(&global_rsv
->lock
);
4600 if (used
+ space_size
>= space_info
->total_bytes
)
4603 used
+= space_info
->bytes_may_use
;
4605 avail
= atomic64_read(&fs_info
->free_chunk_space
);
4608 * If we have dup, raid1 or raid10 then only half of the free
4609 * space is actually useable. For raid56, the space info used
4610 * doesn't include the parity drive, so we don't have to
4613 if (profile
& (BTRFS_BLOCK_GROUP_DUP
|
4614 BTRFS_BLOCK_GROUP_RAID1
|
4615 BTRFS_BLOCK_GROUP_RAID10
))
4619 * If we aren't flushing all things, let us overcommit up to
4620 * 1/2th of the space. If we can flush, don't let us overcommit
4621 * too much, let it overcommit up to 1/8 of the space.
4623 if (flush
== BTRFS_RESERVE_FLUSH_ALL
)
4628 if (used
+ bytes
< space_info
->total_bytes
+ avail
)
4633 static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info
*fs_info
,
4634 unsigned long nr_pages
, int nr_items
)
4636 struct super_block
*sb
= fs_info
->sb
;
4638 if (down_read_trylock(&sb
->s_umount
)) {
4639 writeback_inodes_sb_nr(sb
, nr_pages
, WB_REASON_FS_FREE_SPACE
);
4640 up_read(&sb
->s_umount
);
4643 * We needn't worry the filesystem going from r/w to r/o though
4644 * we don't acquire ->s_umount mutex, because the filesystem
4645 * should guarantee the delalloc inodes list be empty after
4646 * the filesystem is readonly(all dirty pages are written to
4649 btrfs_start_delalloc_roots(fs_info
, nr_items
);
4650 if (!current
->journal_info
)
4651 btrfs_wait_ordered_roots(fs_info
, nr_items
, 0, (u64
)-1);
4655 static inline u64
calc_reclaim_items_nr(struct btrfs_fs_info
*fs_info
,
4661 bytes
= btrfs_calc_trans_metadata_size(fs_info
, 1);
4662 nr
= div64_u64(to_reclaim
, bytes
);
4668 #define EXTENT_SIZE_PER_ITEM SZ_256K
4671 * shrink metadata reservation for delalloc
4673 static void shrink_delalloc(struct btrfs_fs_info
*fs_info
, u64 to_reclaim
,
4674 u64 orig
, bool wait_ordered
)
4676 struct btrfs_space_info
*space_info
;
4677 struct btrfs_trans_handle
*trans
;
4682 unsigned long nr_pages
;
4685 /* Calc the number of the pages we need flush for space reservation */
4686 items
= calc_reclaim_items_nr(fs_info
, to_reclaim
);
4687 to_reclaim
= items
* EXTENT_SIZE_PER_ITEM
;
4689 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4690 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4692 delalloc_bytes
= percpu_counter_sum_positive(
4693 &fs_info
->delalloc_bytes
);
4694 if (delalloc_bytes
== 0) {
4698 btrfs_wait_ordered_roots(fs_info
, items
, 0, (u64
)-1);
4703 while (delalloc_bytes
&& loops
< 3) {
4704 max_reclaim
= min(delalloc_bytes
, to_reclaim
);
4705 nr_pages
= max_reclaim
>> PAGE_SHIFT
;
4706 btrfs_writeback_inodes_sb_nr(fs_info
, nr_pages
, items
);
4708 * We need to wait for the async pages to actually start before
4711 max_reclaim
= atomic_read(&fs_info
->async_delalloc_pages
);
4715 if (max_reclaim
<= nr_pages
)
4718 max_reclaim
-= nr_pages
;
4720 wait_event(fs_info
->async_submit_wait
,
4721 atomic_read(&fs_info
->async_delalloc_pages
) <=
4724 spin_lock(&space_info
->lock
);
4725 if (list_empty(&space_info
->tickets
) &&
4726 list_empty(&space_info
->priority_tickets
)) {
4727 spin_unlock(&space_info
->lock
);
4730 spin_unlock(&space_info
->lock
);
4733 if (wait_ordered
&& !trans
) {
4734 btrfs_wait_ordered_roots(fs_info
, items
, 0, (u64
)-1);
4736 time_left
= schedule_timeout_killable(1);
4740 delalloc_bytes
= percpu_counter_sum_positive(
4741 &fs_info
->delalloc_bytes
);
4745 struct reserve_ticket
{
4748 struct list_head list
;
4749 wait_queue_head_t wait
;
4753 * maybe_commit_transaction - possibly commit the transaction if its ok to
4754 * @root - the root we're allocating for
4755 * @bytes - the number of bytes we want to reserve
4756 * @force - force the commit
4758 * This will check to make sure that committing the transaction will actually
4759 * get us somewhere and then commit the transaction if it does. Otherwise it
4760 * will return -ENOSPC.
4762 static int may_commit_transaction(struct btrfs_fs_info
*fs_info
,
4763 struct btrfs_space_info
*space_info
)
4765 struct reserve_ticket
*ticket
= NULL
;
4766 struct btrfs_block_rsv
*delayed_rsv
= &fs_info
->delayed_block_rsv
;
4767 struct btrfs_trans_handle
*trans
;
4770 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4774 spin_lock(&space_info
->lock
);
4775 if (!list_empty(&space_info
->priority_tickets
))
4776 ticket
= list_first_entry(&space_info
->priority_tickets
,
4777 struct reserve_ticket
, list
);
4778 else if (!list_empty(&space_info
->tickets
))
4779 ticket
= list_first_entry(&space_info
->tickets
,
4780 struct reserve_ticket
, list
);
4781 bytes
= (ticket
) ? ticket
->bytes
: 0;
4782 spin_unlock(&space_info
->lock
);
4787 /* See if there is enough pinned space to make this reservation */
4788 if (__percpu_counter_compare(&space_info
->total_bytes_pinned
,
4790 BTRFS_TOTAL_BYTES_PINNED_BATCH
) >= 0)
4794 * See if there is some space in the delayed insertion reservation for
4797 if (space_info
!= delayed_rsv
->space_info
)
4800 spin_lock(&delayed_rsv
->lock
);
4801 if (delayed_rsv
->size
> bytes
)
4804 bytes
-= delayed_rsv
->size
;
4805 spin_unlock(&delayed_rsv
->lock
);
4807 if (__percpu_counter_compare(&space_info
->total_bytes_pinned
,
4809 BTRFS_TOTAL_BYTES_PINNED_BATCH
) < 0) {
4814 trans
= btrfs_join_transaction(fs_info
->extent_root
);
4818 return btrfs_commit_transaction(trans
);
4822 * Try to flush some data based on policy set by @state. This is only advisory
4823 * and may fail for various reasons. The caller is supposed to examine the
4824 * state of @space_info to detect the outcome.
4826 static void flush_space(struct btrfs_fs_info
*fs_info
,
4827 struct btrfs_space_info
*space_info
, u64 num_bytes
,
4830 struct btrfs_root
*root
= fs_info
->extent_root
;
4831 struct btrfs_trans_handle
*trans
;
4836 case FLUSH_DELAYED_ITEMS_NR
:
4837 case FLUSH_DELAYED_ITEMS
:
4838 if (state
== FLUSH_DELAYED_ITEMS_NR
)
4839 nr
= calc_reclaim_items_nr(fs_info
, num_bytes
) * 2;
4843 trans
= btrfs_join_transaction(root
);
4844 if (IS_ERR(trans
)) {
4845 ret
= PTR_ERR(trans
);
4848 ret
= btrfs_run_delayed_items_nr(trans
, nr
);
4849 btrfs_end_transaction(trans
);
4851 case FLUSH_DELALLOC
:
4852 case FLUSH_DELALLOC_WAIT
:
4853 shrink_delalloc(fs_info
, num_bytes
* 2, num_bytes
,
4854 state
== FLUSH_DELALLOC_WAIT
);
4857 trans
= btrfs_join_transaction(root
);
4858 if (IS_ERR(trans
)) {
4859 ret
= PTR_ERR(trans
);
4862 ret
= do_chunk_alloc(trans
,
4863 btrfs_metadata_alloc_profile(fs_info
),
4864 CHUNK_ALLOC_NO_FORCE
);
4865 btrfs_end_transaction(trans
);
4866 if (ret
> 0 || ret
== -ENOSPC
)
4870 ret
= may_commit_transaction(fs_info
, space_info
);
4877 trace_btrfs_flush_space(fs_info
, space_info
->flags
, num_bytes
, state
,
4883 btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info
*fs_info
,
4884 struct btrfs_space_info
*space_info
,
4887 struct reserve_ticket
*ticket
;
4892 list_for_each_entry(ticket
, &space_info
->tickets
, list
)
4893 to_reclaim
+= ticket
->bytes
;
4894 list_for_each_entry(ticket
, &space_info
->priority_tickets
, list
)
4895 to_reclaim
+= ticket
->bytes
;
4899 to_reclaim
= min_t(u64
, num_online_cpus() * SZ_1M
, SZ_16M
);
4900 if (can_overcommit(fs_info
, space_info
, to_reclaim
,
4901 BTRFS_RESERVE_FLUSH_ALL
, system_chunk
))
4904 used
= btrfs_space_info_used(space_info
, true);
4906 if (can_overcommit(fs_info
, space_info
, SZ_1M
,
4907 BTRFS_RESERVE_FLUSH_ALL
, system_chunk
))
4908 expected
= div_factor_fine(space_info
->total_bytes
, 95);
4910 expected
= div_factor_fine(space_info
->total_bytes
, 90);
4912 if (used
> expected
)
4913 to_reclaim
= used
- expected
;
4916 to_reclaim
= min(to_reclaim
, space_info
->bytes_may_use
+
4917 space_info
->bytes_reserved
);
4921 static inline int need_do_async_reclaim(struct btrfs_fs_info
*fs_info
,
4922 struct btrfs_space_info
*space_info
,
4923 u64 used
, bool system_chunk
)
4925 u64 thresh
= div_factor_fine(space_info
->total_bytes
, 98);
4927 /* If we're just plain full then async reclaim just slows us down. */
4928 if ((space_info
->bytes_used
+ space_info
->bytes_reserved
) >= thresh
)
4931 if (!btrfs_calc_reclaim_metadata_size(fs_info
, space_info
,
4935 return (used
>= thresh
&& !btrfs_fs_closing(fs_info
) &&
4936 !test_bit(BTRFS_FS_STATE_REMOUNTING
, &fs_info
->fs_state
));
4939 static void wake_all_tickets(struct list_head
*head
)
4941 struct reserve_ticket
*ticket
;
4943 while (!list_empty(head
)) {
4944 ticket
= list_first_entry(head
, struct reserve_ticket
, list
);
4945 list_del_init(&ticket
->list
);
4946 ticket
->error
= -ENOSPC
;
4947 wake_up(&ticket
->wait
);
4952 * This is for normal flushers, we can wait all goddamned day if we want to. We
4953 * will loop and continuously try to flush as long as we are making progress.
4954 * We count progress as clearing off tickets each time we have to loop.
4956 static void btrfs_async_reclaim_metadata_space(struct work_struct
*work
)
4958 struct btrfs_fs_info
*fs_info
;
4959 struct btrfs_space_info
*space_info
;
4962 int commit_cycles
= 0;
4963 u64 last_tickets_id
;
4965 fs_info
= container_of(work
, struct btrfs_fs_info
, async_reclaim_work
);
4966 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4968 spin_lock(&space_info
->lock
);
4969 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
, space_info
,
4972 space_info
->flush
= 0;
4973 spin_unlock(&space_info
->lock
);
4976 last_tickets_id
= space_info
->tickets_id
;
4977 spin_unlock(&space_info
->lock
);
4979 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
4981 flush_space(fs_info
, space_info
, to_reclaim
, flush_state
);
4982 spin_lock(&space_info
->lock
);
4983 if (list_empty(&space_info
->tickets
)) {
4984 space_info
->flush
= 0;
4985 spin_unlock(&space_info
->lock
);
4988 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
,
4991 if (last_tickets_id
== space_info
->tickets_id
) {
4994 last_tickets_id
= space_info
->tickets_id
;
4995 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
5000 if (flush_state
> COMMIT_TRANS
) {
5002 if (commit_cycles
> 2) {
5003 wake_all_tickets(&space_info
->tickets
);
5004 space_info
->flush
= 0;
5006 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
5009 spin_unlock(&space_info
->lock
);
5010 } while (flush_state
<= COMMIT_TRANS
);
5013 void btrfs_init_async_reclaim_work(struct work_struct
*work
)
5015 INIT_WORK(work
, btrfs_async_reclaim_metadata_space
);
5018 static void priority_reclaim_metadata_space(struct btrfs_fs_info
*fs_info
,
5019 struct btrfs_space_info
*space_info
,
5020 struct reserve_ticket
*ticket
)
5023 int flush_state
= FLUSH_DELAYED_ITEMS_NR
;
5025 spin_lock(&space_info
->lock
);
5026 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
, space_info
,
5029 spin_unlock(&space_info
->lock
);
5032 spin_unlock(&space_info
->lock
);
5035 flush_space(fs_info
, space_info
, to_reclaim
, flush_state
);
5037 spin_lock(&space_info
->lock
);
5038 if (ticket
->bytes
== 0) {
5039 spin_unlock(&space_info
->lock
);
5042 spin_unlock(&space_info
->lock
);
5045 * Priority flushers can't wait on delalloc without
5048 if (flush_state
== FLUSH_DELALLOC
||
5049 flush_state
== FLUSH_DELALLOC_WAIT
)
5050 flush_state
= ALLOC_CHUNK
;
5051 } while (flush_state
< COMMIT_TRANS
);
5054 static int wait_reserve_ticket(struct btrfs_fs_info
*fs_info
,
5055 struct btrfs_space_info
*space_info
,
5056 struct reserve_ticket
*ticket
, u64 orig_bytes
)
5062 spin_lock(&space_info
->lock
);
5063 while (ticket
->bytes
> 0 && ticket
->error
== 0) {
5064 ret
= prepare_to_wait_event(&ticket
->wait
, &wait
, TASK_KILLABLE
);
5069 spin_unlock(&space_info
->lock
);
5073 finish_wait(&ticket
->wait
, &wait
);
5074 spin_lock(&space_info
->lock
);
5077 ret
= ticket
->error
;
5078 if (!list_empty(&ticket
->list
))
5079 list_del_init(&ticket
->list
);
5080 if (ticket
->bytes
&& ticket
->bytes
< orig_bytes
) {
5081 u64 num_bytes
= orig_bytes
- ticket
->bytes
;
5082 space_info
->bytes_may_use
-= num_bytes
;
5083 trace_btrfs_space_reservation(fs_info
, "space_info",
5084 space_info
->flags
, num_bytes
, 0);
5086 spin_unlock(&space_info
->lock
);
5092 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5093 * @root - the root we're allocating for
5094 * @space_info - the space info we want to allocate from
5095 * @orig_bytes - the number of bytes we want
5096 * @flush - whether or not we can flush to make our reservation
5098 * This will reserve orig_bytes number of bytes from the space info associated
5099 * with the block_rsv. If there is not enough space it will make an attempt to
5100 * flush out space to make room. It will do this by flushing delalloc if
5101 * possible or committing the transaction. If flush is 0 then no attempts to
5102 * regain reservations will be made and this will fail if there is not enough
5105 static int __reserve_metadata_bytes(struct btrfs_fs_info
*fs_info
,
5106 struct btrfs_space_info
*space_info
,
5108 enum btrfs_reserve_flush_enum flush
,
5111 struct reserve_ticket ticket
;
5116 ASSERT(!current
->journal_info
|| flush
!= BTRFS_RESERVE_FLUSH_ALL
);
5118 spin_lock(&space_info
->lock
);
5120 used
= btrfs_space_info_used(space_info
, true);
5123 * If we have enough space then hooray, make our reservation and carry
5124 * on. If not see if we can overcommit, and if we can, hooray carry on.
5125 * If not things get more complicated.
5127 if (used
+ orig_bytes
<= space_info
->total_bytes
) {
5128 space_info
->bytes_may_use
+= orig_bytes
;
5129 trace_btrfs_space_reservation(fs_info
, "space_info",
5130 space_info
->flags
, orig_bytes
, 1);
5132 } else if (can_overcommit(fs_info
, space_info
, orig_bytes
, flush
,
5134 space_info
->bytes_may_use
+= orig_bytes
;
5135 trace_btrfs_space_reservation(fs_info
, "space_info",
5136 space_info
->flags
, orig_bytes
, 1);
5141 * If we couldn't make a reservation then setup our reservation ticket
5142 * and kick the async worker if it's not already running.
5144 * If we are a priority flusher then we just need to add our ticket to
5145 * the list and we will do our own flushing further down.
5147 if (ret
&& flush
!= BTRFS_RESERVE_NO_FLUSH
) {
5148 ticket
.bytes
= orig_bytes
;
5150 init_waitqueue_head(&ticket
.wait
);
5151 if (flush
== BTRFS_RESERVE_FLUSH_ALL
) {
5152 list_add_tail(&ticket
.list
, &space_info
->tickets
);
5153 if (!space_info
->flush
) {
5154 space_info
->flush
= 1;
5155 trace_btrfs_trigger_flush(fs_info
,
5159 queue_work(system_unbound_wq
,
5160 &fs_info
->async_reclaim_work
);
5163 list_add_tail(&ticket
.list
,
5164 &space_info
->priority_tickets
);
5166 } else if (!ret
&& space_info
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
5169 * We will do the space reservation dance during log replay,
5170 * which means we won't have fs_info->fs_root set, so don't do
5171 * the async reclaim as we will panic.
5173 if (!test_bit(BTRFS_FS_LOG_RECOVERING
, &fs_info
->flags
) &&
5174 need_do_async_reclaim(fs_info
, space_info
,
5175 used
, system_chunk
) &&
5176 !work_busy(&fs_info
->async_reclaim_work
)) {
5177 trace_btrfs_trigger_flush(fs_info
, space_info
->flags
,
5178 orig_bytes
, flush
, "preempt");
5179 queue_work(system_unbound_wq
,
5180 &fs_info
->async_reclaim_work
);
5183 spin_unlock(&space_info
->lock
);
5184 if (!ret
|| flush
== BTRFS_RESERVE_NO_FLUSH
)
5187 if (flush
== BTRFS_RESERVE_FLUSH_ALL
)
5188 return wait_reserve_ticket(fs_info
, space_info
, &ticket
,
5192 priority_reclaim_metadata_space(fs_info
, space_info
, &ticket
);
5193 spin_lock(&space_info
->lock
);
5195 if (ticket
.bytes
< orig_bytes
) {
5196 u64 num_bytes
= orig_bytes
- ticket
.bytes
;
5197 space_info
->bytes_may_use
-= num_bytes
;
5198 trace_btrfs_space_reservation(fs_info
, "space_info",
5203 list_del_init(&ticket
.list
);
5206 spin_unlock(&space_info
->lock
);
5207 ASSERT(list_empty(&ticket
.list
));
5212 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5213 * @root - the root we're allocating for
5214 * @block_rsv - the block_rsv we're allocating for
5215 * @orig_bytes - the number of bytes we want
5216 * @flush - whether or not we can flush to make our reservation
5218 * This will reserve orgi_bytes number of bytes from the space info associated
5219 * with the block_rsv. If there is not enough space it will make an attempt to
5220 * flush out space to make room. It will do this by flushing delalloc if
5221 * possible or committing the transaction. If flush is 0 then no attempts to
5222 * regain reservations will be made and this will fail if there is not enough
5225 static int reserve_metadata_bytes(struct btrfs_root
*root
,
5226 struct btrfs_block_rsv
*block_rsv
,
5228 enum btrfs_reserve_flush_enum flush
)
5230 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5231 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5233 bool system_chunk
= (root
== fs_info
->chunk_root
);
5235 ret
= __reserve_metadata_bytes(fs_info
, block_rsv
->space_info
,
5236 orig_bytes
, flush
, system_chunk
);
5237 if (ret
== -ENOSPC
&&
5238 unlikely(root
->orphan_cleanup_state
== ORPHAN_CLEANUP_STARTED
)) {
5239 if (block_rsv
!= global_rsv
&&
5240 !block_rsv_use_bytes(global_rsv
, orig_bytes
))
5243 if (ret
== -ENOSPC
) {
5244 trace_btrfs_space_reservation(fs_info
, "space_info:enospc",
5245 block_rsv
->space_info
->flags
,
5248 if (btrfs_test_opt(fs_info
, ENOSPC_DEBUG
))
5249 dump_space_info(fs_info
, block_rsv
->space_info
,
5255 static struct btrfs_block_rsv
*get_block_rsv(
5256 const struct btrfs_trans_handle
*trans
,
5257 const struct btrfs_root
*root
)
5259 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5260 struct btrfs_block_rsv
*block_rsv
= NULL
;
5262 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
5263 (root
== fs_info
->csum_root
&& trans
->adding_csums
) ||
5264 (root
== fs_info
->uuid_root
))
5265 block_rsv
= trans
->block_rsv
;
5268 block_rsv
= root
->block_rsv
;
5271 block_rsv
= &fs_info
->empty_block_rsv
;
5276 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
5280 spin_lock(&block_rsv
->lock
);
5281 if (block_rsv
->reserved
>= num_bytes
) {
5282 block_rsv
->reserved
-= num_bytes
;
5283 if (block_rsv
->reserved
< block_rsv
->size
)
5284 block_rsv
->full
= 0;
5287 spin_unlock(&block_rsv
->lock
);
5291 static void block_rsv_add_bytes(struct btrfs_block_rsv
*block_rsv
,
5292 u64 num_bytes
, int update_size
)
5294 spin_lock(&block_rsv
->lock
);
5295 block_rsv
->reserved
+= num_bytes
;
5297 block_rsv
->size
+= num_bytes
;
5298 else if (block_rsv
->reserved
>= block_rsv
->size
)
5299 block_rsv
->full
= 1;
5300 spin_unlock(&block_rsv
->lock
);
5303 int btrfs_cond_migrate_bytes(struct btrfs_fs_info
*fs_info
,
5304 struct btrfs_block_rsv
*dest
, u64 num_bytes
,
5307 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5310 if (global_rsv
->space_info
!= dest
->space_info
)
5313 spin_lock(&global_rsv
->lock
);
5314 min_bytes
= div_factor(global_rsv
->size
, min_factor
);
5315 if (global_rsv
->reserved
< min_bytes
+ num_bytes
) {
5316 spin_unlock(&global_rsv
->lock
);
5319 global_rsv
->reserved
-= num_bytes
;
5320 if (global_rsv
->reserved
< global_rsv
->size
)
5321 global_rsv
->full
= 0;
5322 spin_unlock(&global_rsv
->lock
);
5324 block_rsv_add_bytes(dest
, num_bytes
, 1);
5329 * This is for space we already have accounted in space_info->bytes_may_use, so
5330 * basically when we're returning space from block_rsv's.
5332 static void space_info_add_old_bytes(struct btrfs_fs_info
*fs_info
,
5333 struct btrfs_space_info
*space_info
,
5336 struct reserve_ticket
*ticket
;
5337 struct list_head
*head
;
5339 enum btrfs_reserve_flush_enum flush
= BTRFS_RESERVE_NO_FLUSH
;
5340 bool check_overcommit
= false;
5342 spin_lock(&space_info
->lock
);
5343 head
= &space_info
->priority_tickets
;
5346 * If we are over our limit then we need to check and see if we can
5347 * overcommit, and if we can't then we just need to free up our space
5348 * and not satisfy any requests.
5350 used
= btrfs_space_info_used(space_info
, true);
5351 if (used
- num_bytes
>= space_info
->total_bytes
)
5352 check_overcommit
= true;
5354 while (!list_empty(head
) && num_bytes
) {
5355 ticket
= list_first_entry(head
, struct reserve_ticket
,
5358 * We use 0 bytes because this space is already reserved, so
5359 * adding the ticket space would be a double count.
5361 if (check_overcommit
&&
5362 !can_overcommit(fs_info
, space_info
, 0, flush
, false))
5364 if (num_bytes
>= ticket
->bytes
) {
5365 list_del_init(&ticket
->list
);
5366 num_bytes
-= ticket
->bytes
;
5368 space_info
->tickets_id
++;
5369 wake_up(&ticket
->wait
);
5371 ticket
->bytes
-= num_bytes
;
5376 if (num_bytes
&& head
== &space_info
->priority_tickets
) {
5377 head
= &space_info
->tickets
;
5378 flush
= BTRFS_RESERVE_FLUSH_ALL
;
5381 space_info
->bytes_may_use
-= num_bytes
;
5382 trace_btrfs_space_reservation(fs_info
, "space_info",
5383 space_info
->flags
, num_bytes
, 0);
5384 spin_unlock(&space_info
->lock
);
5388 * This is for newly allocated space that isn't accounted in
5389 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5390 * we use this helper.
5392 static void space_info_add_new_bytes(struct btrfs_fs_info
*fs_info
,
5393 struct btrfs_space_info
*space_info
,
5396 struct reserve_ticket
*ticket
;
5397 struct list_head
*head
= &space_info
->priority_tickets
;
5400 while (!list_empty(head
) && num_bytes
) {
5401 ticket
= list_first_entry(head
, struct reserve_ticket
,
5403 if (num_bytes
>= ticket
->bytes
) {
5404 trace_btrfs_space_reservation(fs_info
, "space_info",
5407 list_del_init(&ticket
->list
);
5408 num_bytes
-= ticket
->bytes
;
5409 space_info
->bytes_may_use
+= ticket
->bytes
;
5411 space_info
->tickets_id
++;
5412 wake_up(&ticket
->wait
);
5414 trace_btrfs_space_reservation(fs_info
, "space_info",
5417 space_info
->bytes_may_use
+= num_bytes
;
5418 ticket
->bytes
-= num_bytes
;
5423 if (num_bytes
&& head
== &space_info
->priority_tickets
) {
5424 head
= &space_info
->tickets
;
5429 static u64
block_rsv_release_bytes(struct btrfs_fs_info
*fs_info
,
5430 struct btrfs_block_rsv
*block_rsv
,
5431 struct btrfs_block_rsv
*dest
, u64 num_bytes
,
5432 u64
*qgroup_to_release_ret
)
5434 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
5435 u64 qgroup_to_release
= 0;
5438 spin_lock(&block_rsv
->lock
);
5439 if (num_bytes
== (u64
)-1) {
5440 num_bytes
= block_rsv
->size
;
5441 qgroup_to_release
= block_rsv
->qgroup_rsv_size
;
5443 block_rsv
->size
-= num_bytes
;
5444 if (block_rsv
->reserved
>= block_rsv
->size
) {
5445 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
5446 block_rsv
->reserved
= block_rsv
->size
;
5447 block_rsv
->full
= 1;
5451 if (block_rsv
->qgroup_rsv_reserved
>= block_rsv
->qgroup_rsv_size
) {
5452 qgroup_to_release
= block_rsv
->qgroup_rsv_reserved
-
5453 block_rsv
->qgroup_rsv_size
;
5454 block_rsv
->qgroup_rsv_reserved
= block_rsv
->qgroup_rsv_size
;
5456 qgroup_to_release
= 0;
5458 spin_unlock(&block_rsv
->lock
);
5461 if (num_bytes
> 0) {
5463 spin_lock(&dest
->lock
);
5467 bytes_to_add
= dest
->size
- dest
->reserved
;
5468 bytes_to_add
= min(num_bytes
, bytes_to_add
);
5469 dest
->reserved
+= bytes_to_add
;
5470 if (dest
->reserved
>= dest
->size
)
5472 num_bytes
-= bytes_to_add
;
5474 spin_unlock(&dest
->lock
);
5477 space_info_add_old_bytes(fs_info
, space_info
,
5480 if (qgroup_to_release_ret
)
5481 *qgroup_to_release_ret
= qgroup_to_release
;
5485 int btrfs_block_rsv_migrate(struct btrfs_block_rsv
*src
,
5486 struct btrfs_block_rsv
*dst
, u64 num_bytes
,
5491 ret
= block_rsv_use_bytes(src
, num_bytes
);
5495 block_rsv_add_bytes(dst
, num_bytes
, update_size
);
5499 void btrfs_init_block_rsv(struct btrfs_block_rsv
*rsv
, unsigned short type
)
5501 memset(rsv
, 0, sizeof(*rsv
));
5502 spin_lock_init(&rsv
->lock
);
5506 void btrfs_init_metadata_block_rsv(struct btrfs_fs_info
*fs_info
,
5507 struct btrfs_block_rsv
*rsv
,
5508 unsigned short type
)
5510 btrfs_init_block_rsv(rsv
, type
);
5511 rsv
->space_info
= __find_space_info(fs_info
,
5512 BTRFS_BLOCK_GROUP_METADATA
);
5515 struct btrfs_block_rsv
*btrfs_alloc_block_rsv(struct btrfs_fs_info
*fs_info
,
5516 unsigned short type
)
5518 struct btrfs_block_rsv
*block_rsv
;
5520 block_rsv
= kmalloc(sizeof(*block_rsv
), GFP_NOFS
);
5524 btrfs_init_metadata_block_rsv(fs_info
, block_rsv
, type
);
5528 void btrfs_free_block_rsv(struct btrfs_fs_info
*fs_info
,
5529 struct btrfs_block_rsv
*rsv
)
5533 btrfs_block_rsv_release(fs_info
, rsv
, (u64
)-1);
5537 void __btrfs_free_block_rsv(struct btrfs_block_rsv
*rsv
)
5542 int btrfs_block_rsv_add(struct btrfs_root
*root
,
5543 struct btrfs_block_rsv
*block_rsv
, u64 num_bytes
,
5544 enum btrfs_reserve_flush_enum flush
)
5551 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
5553 block_rsv_add_bytes(block_rsv
, num_bytes
, 1);
5560 int btrfs_block_rsv_check(struct btrfs_block_rsv
*block_rsv
, int min_factor
)
5568 spin_lock(&block_rsv
->lock
);
5569 num_bytes
= div_factor(block_rsv
->size
, min_factor
);
5570 if (block_rsv
->reserved
>= num_bytes
)
5572 spin_unlock(&block_rsv
->lock
);
5577 int btrfs_block_rsv_refill(struct btrfs_root
*root
,
5578 struct btrfs_block_rsv
*block_rsv
, u64 min_reserved
,
5579 enum btrfs_reserve_flush_enum flush
)
5587 spin_lock(&block_rsv
->lock
);
5588 num_bytes
= min_reserved
;
5589 if (block_rsv
->reserved
>= num_bytes
)
5592 num_bytes
-= block_rsv
->reserved
;
5593 spin_unlock(&block_rsv
->lock
);
5598 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
5600 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
5608 * btrfs_inode_rsv_refill - refill the inode block rsv.
5609 * @inode - the inode we are refilling.
5610 * @flush - the flusing restriction.
5612 * Essentially the same as btrfs_block_rsv_refill, except it uses the
5613 * block_rsv->size as the minimum size. We'll either refill the missing amount
5614 * or return if we already have enough space. This will also handle the resreve
5615 * tracepoint for the reserved amount.
5617 static int btrfs_inode_rsv_refill(struct btrfs_inode
*inode
,
5618 enum btrfs_reserve_flush_enum flush
)
5620 struct btrfs_root
*root
= inode
->root
;
5621 struct btrfs_block_rsv
*block_rsv
= &inode
->block_rsv
;
5623 u64 qgroup_num_bytes
= 0;
5626 spin_lock(&block_rsv
->lock
);
5627 if (block_rsv
->reserved
< block_rsv
->size
)
5628 num_bytes
= block_rsv
->size
- block_rsv
->reserved
;
5629 if (block_rsv
->qgroup_rsv_reserved
< block_rsv
->qgroup_rsv_size
)
5630 qgroup_num_bytes
= block_rsv
->qgroup_rsv_size
-
5631 block_rsv
->qgroup_rsv_reserved
;
5632 spin_unlock(&block_rsv
->lock
);
5637 ret
= btrfs_qgroup_reserve_meta_prealloc(root
, qgroup_num_bytes
, true);
5640 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
5642 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
5643 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5644 btrfs_ino(inode
), num_bytes
, 1);
5646 /* Don't forget to increase qgroup_rsv_reserved */
5647 spin_lock(&block_rsv
->lock
);
5648 block_rsv
->qgroup_rsv_reserved
+= qgroup_num_bytes
;
5649 spin_unlock(&block_rsv
->lock
);
5651 btrfs_qgroup_free_meta_prealloc(root
, qgroup_num_bytes
);
5656 * btrfs_inode_rsv_release - release any excessive reservation.
5657 * @inode - the inode we need to release from.
5658 * @qgroup_free - free or convert qgroup meta.
5659 * Unlike normal operation, qgroup meta reservation needs to know if we are
5660 * freeing qgroup reservation or just converting it into per-trans. Normally
5661 * @qgroup_free is true for error handling, and false for normal release.
5663 * This is the same as btrfs_block_rsv_release, except that it handles the
5664 * tracepoint for the reservation.
5666 static void btrfs_inode_rsv_release(struct btrfs_inode
*inode
, bool qgroup_free
)
5668 struct btrfs_fs_info
*fs_info
= inode
->root
->fs_info
;
5669 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5670 struct btrfs_block_rsv
*block_rsv
= &inode
->block_rsv
;
5672 u64 qgroup_to_release
= 0;
5675 * Since we statically set the block_rsv->size we just want to say we
5676 * are releasing 0 bytes, and then we'll just get the reservation over
5679 released
= block_rsv_release_bytes(fs_info
, block_rsv
, global_rsv
, 0,
5680 &qgroup_to_release
);
5682 trace_btrfs_space_reservation(fs_info
, "delalloc",
5683 btrfs_ino(inode
), released
, 0);
5685 btrfs_qgroup_free_meta_prealloc(inode
->root
, qgroup_to_release
);
5687 btrfs_qgroup_convert_reserved_meta(inode
->root
,
5691 void btrfs_block_rsv_release(struct btrfs_fs_info
*fs_info
,
5692 struct btrfs_block_rsv
*block_rsv
,
5695 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5697 if (global_rsv
== block_rsv
||
5698 block_rsv
->space_info
!= global_rsv
->space_info
)
5700 block_rsv_release_bytes(fs_info
, block_rsv
, global_rsv
, num_bytes
, NULL
);
5703 static void update_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5705 struct btrfs_block_rsv
*block_rsv
= &fs_info
->global_block_rsv
;
5706 struct btrfs_space_info
*sinfo
= block_rsv
->space_info
;
5710 * The global block rsv is based on the size of the extent tree, the
5711 * checksum tree and the root tree. If the fs is empty we want to set
5712 * it to a minimal amount for safety.
5714 num_bytes
= btrfs_root_used(&fs_info
->extent_root
->root_item
) +
5715 btrfs_root_used(&fs_info
->csum_root
->root_item
) +
5716 btrfs_root_used(&fs_info
->tree_root
->root_item
);
5717 num_bytes
= max_t(u64
, num_bytes
, SZ_16M
);
5719 spin_lock(&sinfo
->lock
);
5720 spin_lock(&block_rsv
->lock
);
5722 block_rsv
->size
= min_t(u64
, num_bytes
, SZ_512M
);
5724 if (block_rsv
->reserved
< block_rsv
->size
) {
5725 num_bytes
= btrfs_space_info_used(sinfo
, true);
5726 if (sinfo
->total_bytes
> num_bytes
) {
5727 num_bytes
= sinfo
->total_bytes
- num_bytes
;
5728 num_bytes
= min(num_bytes
,
5729 block_rsv
->size
- block_rsv
->reserved
);
5730 block_rsv
->reserved
+= num_bytes
;
5731 sinfo
->bytes_may_use
+= num_bytes
;
5732 trace_btrfs_space_reservation(fs_info
, "space_info",
5733 sinfo
->flags
, num_bytes
,
5736 } else if (block_rsv
->reserved
> block_rsv
->size
) {
5737 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
5738 sinfo
->bytes_may_use
-= num_bytes
;
5739 trace_btrfs_space_reservation(fs_info
, "space_info",
5740 sinfo
->flags
, num_bytes
, 0);
5741 block_rsv
->reserved
= block_rsv
->size
;
5744 if (block_rsv
->reserved
== block_rsv
->size
)
5745 block_rsv
->full
= 1;
5747 block_rsv
->full
= 0;
5749 spin_unlock(&block_rsv
->lock
);
5750 spin_unlock(&sinfo
->lock
);
5753 static void init_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5755 struct btrfs_space_info
*space_info
;
5757 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
5758 fs_info
->chunk_block_rsv
.space_info
= space_info
;
5760 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
5761 fs_info
->global_block_rsv
.space_info
= space_info
;
5762 fs_info
->trans_block_rsv
.space_info
= space_info
;
5763 fs_info
->empty_block_rsv
.space_info
= space_info
;
5764 fs_info
->delayed_block_rsv
.space_info
= space_info
;
5766 fs_info
->extent_root
->block_rsv
= &fs_info
->global_block_rsv
;
5767 fs_info
->csum_root
->block_rsv
= &fs_info
->global_block_rsv
;
5768 fs_info
->dev_root
->block_rsv
= &fs_info
->global_block_rsv
;
5769 fs_info
->tree_root
->block_rsv
= &fs_info
->global_block_rsv
;
5770 if (fs_info
->quota_root
)
5771 fs_info
->quota_root
->block_rsv
= &fs_info
->global_block_rsv
;
5772 fs_info
->chunk_root
->block_rsv
= &fs_info
->chunk_block_rsv
;
5774 update_global_block_rsv(fs_info
);
5777 static void release_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5779 block_rsv_release_bytes(fs_info
, &fs_info
->global_block_rsv
, NULL
,
5781 WARN_ON(fs_info
->trans_block_rsv
.size
> 0);
5782 WARN_ON(fs_info
->trans_block_rsv
.reserved
> 0);
5783 WARN_ON(fs_info
->chunk_block_rsv
.size
> 0);
5784 WARN_ON(fs_info
->chunk_block_rsv
.reserved
> 0);
5785 WARN_ON(fs_info
->delayed_block_rsv
.size
> 0);
5786 WARN_ON(fs_info
->delayed_block_rsv
.reserved
> 0);
5791 * To be called after all the new block groups attached to the transaction
5792 * handle have been created (btrfs_create_pending_block_groups()).
5794 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle
*trans
)
5796 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
5798 if (!trans
->chunk_bytes_reserved
)
5801 WARN_ON_ONCE(!list_empty(&trans
->new_bgs
));
5803 block_rsv_release_bytes(fs_info
, &fs_info
->chunk_block_rsv
, NULL
,
5804 trans
->chunk_bytes_reserved
, NULL
);
5805 trans
->chunk_bytes_reserved
= 0;
5809 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5810 * root: the root of the parent directory
5811 * rsv: block reservation
5812 * items: the number of items that we need do reservation
5813 * qgroup_reserved: used to return the reserved size in qgroup
5815 * This function is used to reserve the space for snapshot/subvolume
5816 * creation and deletion. Those operations are different with the
5817 * common file/directory operations, they change two fs/file trees
5818 * and root tree, the number of items that the qgroup reserves is
5819 * different with the free space reservation. So we can not use
5820 * the space reservation mechanism in start_transaction().
5822 int btrfs_subvolume_reserve_metadata(struct btrfs_root
*root
,
5823 struct btrfs_block_rsv
*rsv
,
5825 bool use_global_rsv
)
5829 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5830 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5832 if (test_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
)) {
5833 /* One for parent inode, two for dir entries */
5834 num_bytes
= 3 * fs_info
->nodesize
;
5835 ret
= btrfs_qgroup_reserve_meta_prealloc(root
, num_bytes
, true);
5842 num_bytes
= btrfs_calc_trans_metadata_size(fs_info
, items
);
5843 rsv
->space_info
= __find_space_info(fs_info
,
5844 BTRFS_BLOCK_GROUP_METADATA
);
5845 ret
= btrfs_block_rsv_add(root
, rsv
, num_bytes
,
5846 BTRFS_RESERVE_FLUSH_ALL
);
5848 if (ret
== -ENOSPC
&& use_global_rsv
)
5849 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
, num_bytes
, 1);
5851 if (ret
&& num_bytes
)
5852 btrfs_qgroup_free_meta_prealloc(root
, num_bytes
);
5857 void btrfs_subvolume_release_metadata(struct btrfs_fs_info
*fs_info
,
5858 struct btrfs_block_rsv
*rsv
)
5860 btrfs_block_rsv_release(fs_info
, rsv
, (u64
)-1);
5863 static void btrfs_calculate_inode_block_rsv_size(struct btrfs_fs_info
*fs_info
,
5864 struct btrfs_inode
*inode
)
5866 struct btrfs_block_rsv
*block_rsv
= &inode
->block_rsv
;
5867 u64 reserve_size
= 0;
5868 u64 qgroup_rsv_size
= 0;
5870 unsigned outstanding_extents
;
5872 lockdep_assert_held(&inode
->lock
);
5873 outstanding_extents
= inode
->outstanding_extents
;
5874 if (outstanding_extents
)
5875 reserve_size
= btrfs_calc_trans_metadata_size(fs_info
,
5876 outstanding_extents
+ 1);
5877 csum_leaves
= btrfs_csum_bytes_to_leaves(fs_info
,
5879 reserve_size
+= btrfs_calc_trans_metadata_size(fs_info
,
5882 * For qgroup rsv, the calculation is very simple:
5883 * account one nodesize for each outstanding extent
5885 * This is overestimating in most cases.
5887 qgroup_rsv_size
= outstanding_extents
* fs_info
->nodesize
;
5889 spin_lock(&block_rsv
->lock
);
5890 block_rsv
->size
= reserve_size
;
5891 block_rsv
->qgroup_rsv_size
= qgroup_rsv_size
;
5892 spin_unlock(&block_rsv
->lock
);
5895 int btrfs_delalloc_reserve_metadata(struct btrfs_inode
*inode
, u64 num_bytes
)
5897 struct btrfs_fs_info
*fs_info
= inode
->root
->fs_info
;
5898 unsigned nr_extents
;
5899 enum btrfs_reserve_flush_enum flush
= BTRFS_RESERVE_FLUSH_ALL
;
5901 bool delalloc_lock
= true;
5903 /* If we are a free space inode we need to not flush since we will be in
5904 * the middle of a transaction commit. We also don't need the delalloc
5905 * mutex since we won't race with anybody. We need this mostly to make
5906 * lockdep shut its filthy mouth.
5908 * If we have a transaction open (can happen if we call truncate_block
5909 * from truncate), then we need FLUSH_LIMIT so we don't deadlock.
5911 if (btrfs_is_free_space_inode(inode
)) {
5912 flush
= BTRFS_RESERVE_NO_FLUSH
;
5913 delalloc_lock
= false;
5915 if (current
->journal_info
)
5916 flush
= BTRFS_RESERVE_FLUSH_LIMIT
;
5918 if (btrfs_transaction_in_commit(fs_info
))
5919 schedule_timeout(1);
5923 mutex_lock(&inode
->delalloc_mutex
);
5925 num_bytes
= ALIGN(num_bytes
, fs_info
->sectorsize
);
5927 /* Add our new extents and calculate the new rsv size. */
5928 spin_lock(&inode
->lock
);
5929 nr_extents
= count_max_extents(num_bytes
);
5930 btrfs_mod_outstanding_extents(inode
, nr_extents
);
5931 inode
->csum_bytes
+= num_bytes
;
5932 btrfs_calculate_inode_block_rsv_size(fs_info
, inode
);
5933 spin_unlock(&inode
->lock
);
5935 ret
= btrfs_inode_rsv_refill(inode
, flush
);
5940 mutex_unlock(&inode
->delalloc_mutex
);
5944 spin_lock(&inode
->lock
);
5945 nr_extents
= count_max_extents(num_bytes
);
5946 btrfs_mod_outstanding_extents(inode
, -nr_extents
);
5947 inode
->csum_bytes
-= num_bytes
;
5948 btrfs_calculate_inode_block_rsv_size(fs_info
, inode
);
5949 spin_unlock(&inode
->lock
);
5951 btrfs_inode_rsv_release(inode
, true);
5953 mutex_unlock(&inode
->delalloc_mutex
);
5958 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5959 * @inode: the inode to release the reservation for.
5960 * @num_bytes: the number of bytes we are releasing.
5961 * @qgroup_free: free qgroup reservation or convert it to per-trans reservation
5963 * This will release the metadata reservation for an inode. This can be called
5964 * once we complete IO for a given set of bytes to release their metadata
5965 * reservations, or on error for the same reason.
5967 void btrfs_delalloc_release_metadata(struct btrfs_inode
*inode
, u64 num_bytes
,
5970 struct btrfs_fs_info
*fs_info
= inode
->root
->fs_info
;
5972 num_bytes
= ALIGN(num_bytes
, fs_info
->sectorsize
);
5973 spin_lock(&inode
->lock
);
5974 inode
->csum_bytes
-= num_bytes
;
5975 btrfs_calculate_inode_block_rsv_size(fs_info
, inode
);
5976 spin_unlock(&inode
->lock
);
5978 if (btrfs_is_testing(fs_info
))
5981 btrfs_inode_rsv_release(inode
, qgroup_free
);
5985 * btrfs_delalloc_release_extents - release our outstanding_extents
5986 * @inode: the inode to balance the reservation for.
5987 * @num_bytes: the number of bytes we originally reserved with
5988 * @qgroup_free: do we need to free qgroup meta reservation or convert them.
5990 * When we reserve space we increase outstanding_extents for the extents we may
5991 * add. Once we've set the range as delalloc or created our ordered extents we
5992 * have outstanding_extents to track the real usage, so we use this to free our
5993 * temporarily tracked outstanding_extents. This _must_ be used in conjunction
5994 * with btrfs_delalloc_reserve_metadata.
5996 void btrfs_delalloc_release_extents(struct btrfs_inode
*inode
, u64 num_bytes
,
5999 struct btrfs_fs_info
*fs_info
= inode
->root
->fs_info
;
6000 unsigned num_extents
;
6002 spin_lock(&inode
->lock
);
6003 num_extents
= count_max_extents(num_bytes
);
6004 btrfs_mod_outstanding_extents(inode
, -num_extents
);
6005 btrfs_calculate_inode_block_rsv_size(fs_info
, inode
);
6006 spin_unlock(&inode
->lock
);
6008 if (btrfs_is_testing(fs_info
))
6011 btrfs_inode_rsv_release(inode
, qgroup_free
);
6015 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6017 * @inode: inode we're writing to
6018 * @start: start range we are writing to
6019 * @len: how long the range we are writing to
6020 * @reserved: mandatory parameter, record actually reserved qgroup ranges of
6021 * current reservation.
6023 * This will do the following things
6025 * o reserve space in data space info for num bytes
6026 * and reserve precious corresponding qgroup space
6027 * (Done in check_data_free_space)
6029 * o reserve space for metadata space, based on the number of outstanding
6030 * extents and how much csums will be needed
6031 * also reserve metadata space in a per root over-reserve method.
6032 * o add to the inodes->delalloc_bytes
6033 * o add it to the fs_info's delalloc inodes list.
6034 * (Above 3 all done in delalloc_reserve_metadata)
6036 * Return 0 for success
6037 * Return <0 for error(-ENOSPC or -EQUOT)
6039 int btrfs_delalloc_reserve_space(struct inode
*inode
,
6040 struct extent_changeset
**reserved
, u64 start
, u64 len
)
6044 ret
= btrfs_check_data_free_space(inode
, reserved
, start
, len
);
6047 ret
= btrfs_delalloc_reserve_metadata(BTRFS_I(inode
), len
);
6049 btrfs_free_reserved_data_space(inode
, *reserved
, start
, len
);
6054 * btrfs_delalloc_release_space - release data and metadata space for delalloc
6055 * @inode: inode we're releasing space for
6056 * @start: start position of the space already reserved
6057 * @len: the len of the space already reserved
6058 * @release_bytes: the len of the space we consumed or didn't use
6060 * This function will release the metadata space that was not used and will
6061 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
6062 * list if there are no delalloc bytes left.
6063 * Also it will handle the qgroup reserved space.
6065 void btrfs_delalloc_release_space(struct inode
*inode
,
6066 struct extent_changeset
*reserved
,
6067 u64 start
, u64 len
, bool qgroup_free
)
6069 btrfs_delalloc_release_metadata(BTRFS_I(inode
), len
, qgroup_free
);
6070 btrfs_free_reserved_data_space(inode
, reserved
, start
, len
);
6073 static int update_block_group(struct btrfs_trans_handle
*trans
,
6074 struct btrfs_fs_info
*info
, u64 bytenr
,
6075 u64 num_bytes
, int alloc
)
6077 struct btrfs_block_group_cache
*cache
= NULL
;
6078 u64 total
= num_bytes
;
6083 /* block accounting for super block */
6084 spin_lock(&info
->delalloc_root_lock
);
6085 old_val
= btrfs_super_bytes_used(info
->super_copy
);
6087 old_val
+= num_bytes
;
6089 old_val
-= num_bytes
;
6090 btrfs_set_super_bytes_used(info
->super_copy
, old_val
);
6091 spin_unlock(&info
->delalloc_root_lock
);
6094 cache
= btrfs_lookup_block_group(info
, bytenr
);
6097 if (cache
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
6098 BTRFS_BLOCK_GROUP_RAID1
|
6099 BTRFS_BLOCK_GROUP_RAID10
))
6104 * If this block group has free space cache written out, we
6105 * need to make sure to load it if we are removing space. This
6106 * is because we need the unpinning stage to actually add the
6107 * space back to the block group, otherwise we will leak space.
6109 if (!alloc
&& cache
->cached
== BTRFS_CACHE_NO
)
6110 cache_block_group(cache
, 1);
6112 byte_in_group
= bytenr
- cache
->key
.objectid
;
6113 WARN_ON(byte_in_group
> cache
->key
.offset
);
6115 spin_lock(&cache
->space_info
->lock
);
6116 spin_lock(&cache
->lock
);
6118 if (btrfs_test_opt(info
, SPACE_CACHE
) &&
6119 cache
->disk_cache_state
< BTRFS_DC_CLEAR
)
6120 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
6122 old_val
= btrfs_block_group_used(&cache
->item
);
6123 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
6125 old_val
+= num_bytes
;
6126 btrfs_set_block_group_used(&cache
->item
, old_val
);
6127 cache
->reserved
-= num_bytes
;
6128 cache
->space_info
->bytes_reserved
-= num_bytes
;
6129 cache
->space_info
->bytes_used
+= num_bytes
;
6130 cache
->space_info
->disk_used
+= num_bytes
* factor
;
6131 spin_unlock(&cache
->lock
);
6132 spin_unlock(&cache
->space_info
->lock
);
6134 old_val
-= num_bytes
;
6135 btrfs_set_block_group_used(&cache
->item
, old_val
);
6136 cache
->pinned
+= num_bytes
;
6137 cache
->space_info
->bytes_pinned
+= num_bytes
;
6138 cache
->space_info
->bytes_used
-= num_bytes
;
6139 cache
->space_info
->disk_used
-= num_bytes
* factor
;
6140 spin_unlock(&cache
->lock
);
6141 spin_unlock(&cache
->space_info
->lock
);
6143 trace_btrfs_space_reservation(info
, "pinned",
6144 cache
->space_info
->flags
,
6146 percpu_counter_add_batch(&cache
->space_info
->total_bytes_pinned
,
6148 BTRFS_TOTAL_BYTES_PINNED_BATCH
);
6149 set_extent_dirty(info
->pinned_extents
,
6150 bytenr
, bytenr
+ num_bytes
- 1,
6151 GFP_NOFS
| __GFP_NOFAIL
);
6154 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
6155 if (list_empty(&cache
->dirty_list
)) {
6156 list_add_tail(&cache
->dirty_list
,
6157 &trans
->transaction
->dirty_bgs
);
6158 trans
->transaction
->num_dirty_bgs
++;
6159 btrfs_get_block_group(cache
);
6161 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
6164 * No longer have used bytes in this block group, queue it for
6165 * deletion. We do this after adding the block group to the
6166 * dirty list to avoid races between cleaner kthread and space
6169 if (!alloc
&& old_val
== 0) {
6170 spin_lock(&info
->unused_bgs_lock
);
6171 if (list_empty(&cache
->bg_list
)) {
6172 btrfs_get_block_group(cache
);
6173 trace_btrfs_add_unused_block_group(cache
);
6174 list_add_tail(&cache
->bg_list
,
6177 spin_unlock(&info
->unused_bgs_lock
);
6180 btrfs_put_block_group(cache
);
6182 bytenr
+= num_bytes
;
6187 static u64
first_logical_byte(struct btrfs_fs_info
*fs_info
, u64 search_start
)
6189 struct btrfs_block_group_cache
*cache
;
6192 spin_lock(&fs_info
->block_group_cache_lock
);
6193 bytenr
= fs_info
->first_logical_byte
;
6194 spin_unlock(&fs_info
->block_group_cache_lock
);
6196 if (bytenr
< (u64
)-1)
6199 cache
= btrfs_lookup_first_block_group(fs_info
, search_start
);
6203 bytenr
= cache
->key
.objectid
;
6204 btrfs_put_block_group(cache
);
6209 static int pin_down_extent(struct btrfs_fs_info
*fs_info
,
6210 struct btrfs_block_group_cache
*cache
,
6211 u64 bytenr
, u64 num_bytes
, int reserved
)
6213 spin_lock(&cache
->space_info
->lock
);
6214 spin_lock(&cache
->lock
);
6215 cache
->pinned
+= num_bytes
;
6216 cache
->space_info
->bytes_pinned
+= num_bytes
;
6218 cache
->reserved
-= num_bytes
;
6219 cache
->space_info
->bytes_reserved
-= num_bytes
;
6221 spin_unlock(&cache
->lock
);
6222 spin_unlock(&cache
->space_info
->lock
);
6224 trace_btrfs_space_reservation(fs_info
, "pinned",
6225 cache
->space_info
->flags
, num_bytes
, 1);
6226 percpu_counter_add_batch(&cache
->space_info
->total_bytes_pinned
,
6227 num_bytes
, BTRFS_TOTAL_BYTES_PINNED_BATCH
);
6228 set_extent_dirty(fs_info
->pinned_extents
, bytenr
,
6229 bytenr
+ num_bytes
- 1, GFP_NOFS
| __GFP_NOFAIL
);
6234 * this function must be called within transaction
6236 int btrfs_pin_extent(struct btrfs_fs_info
*fs_info
,
6237 u64 bytenr
, u64 num_bytes
, int reserved
)
6239 struct btrfs_block_group_cache
*cache
;
6241 cache
= btrfs_lookup_block_group(fs_info
, bytenr
);
6242 BUG_ON(!cache
); /* Logic error */
6244 pin_down_extent(fs_info
, cache
, bytenr
, num_bytes
, reserved
);
6246 btrfs_put_block_group(cache
);
6251 * this function must be called within transaction
6253 int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info
*fs_info
,
6254 u64 bytenr
, u64 num_bytes
)
6256 struct btrfs_block_group_cache
*cache
;
6259 cache
= btrfs_lookup_block_group(fs_info
, bytenr
);
6264 * pull in the free space cache (if any) so that our pin
6265 * removes the free space from the cache. We have load_only set
6266 * to one because the slow code to read in the free extents does check
6267 * the pinned extents.
6269 cache_block_group(cache
, 1);
6271 pin_down_extent(fs_info
, cache
, bytenr
, num_bytes
, 0);
6273 /* remove us from the free space cache (if we're there at all) */
6274 ret
= btrfs_remove_free_space(cache
, bytenr
, num_bytes
);
6275 btrfs_put_block_group(cache
);
6279 static int __exclude_logged_extent(struct btrfs_fs_info
*fs_info
,
6280 u64 start
, u64 num_bytes
)
6283 struct btrfs_block_group_cache
*block_group
;
6284 struct btrfs_caching_control
*caching_ctl
;
6286 block_group
= btrfs_lookup_block_group(fs_info
, start
);
6290 cache_block_group(block_group
, 0);
6291 caching_ctl
= get_caching_control(block_group
);
6295 BUG_ON(!block_group_cache_done(block_group
));
6296 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
6298 mutex_lock(&caching_ctl
->mutex
);
6300 if (start
>= caching_ctl
->progress
) {
6301 ret
= add_excluded_extent(fs_info
, start
, num_bytes
);
6302 } else if (start
+ num_bytes
<= caching_ctl
->progress
) {
6303 ret
= btrfs_remove_free_space(block_group
,
6306 num_bytes
= caching_ctl
->progress
- start
;
6307 ret
= btrfs_remove_free_space(block_group
,
6312 num_bytes
= (start
+ num_bytes
) -
6313 caching_ctl
->progress
;
6314 start
= caching_ctl
->progress
;
6315 ret
= add_excluded_extent(fs_info
, start
, num_bytes
);
6318 mutex_unlock(&caching_ctl
->mutex
);
6319 put_caching_control(caching_ctl
);
6321 btrfs_put_block_group(block_group
);
6325 int btrfs_exclude_logged_extents(struct btrfs_fs_info
*fs_info
,
6326 struct extent_buffer
*eb
)
6328 struct btrfs_file_extent_item
*item
;
6329 struct btrfs_key key
;
6334 if (!btrfs_fs_incompat(fs_info
, MIXED_GROUPS
))
6337 for (i
= 0; i
< btrfs_header_nritems(eb
); i
++) {
6338 btrfs_item_key_to_cpu(eb
, &key
, i
);
6339 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
6341 item
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
6342 found_type
= btrfs_file_extent_type(eb
, item
);
6343 if (found_type
== BTRFS_FILE_EXTENT_INLINE
)
6345 if (btrfs_file_extent_disk_bytenr(eb
, item
) == 0)
6347 key
.objectid
= btrfs_file_extent_disk_bytenr(eb
, item
);
6348 key
.offset
= btrfs_file_extent_disk_num_bytes(eb
, item
);
6349 ret
= __exclude_logged_extent(fs_info
, key
.objectid
, key
.offset
);
6358 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache
*bg
)
6360 atomic_inc(&bg
->reservations
);
6363 void btrfs_dec_block_group_reservations(struct btrfs_fs_info
*fs_info
,
6366 struct btrfs_block_group_cache
*bg
;
6368 bg
= btrfs_lookup_block_group(fs_info
, start
);
6370 if (atomic_dec_and_test(&bg
->reservations
))
6371 wake_up_var(&bg
->reservations
);
6372 btrfs_put_block_group(bg
);
6375 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache
*bg
)
6377 struct btrfs_space_info
*space_info
= bg
->space_info
;
6381 if (!(bg
->flags
& BTRFS_BLOCK_GROUP_DATA
))
6385 * Our block group is read only but before we set it to read only,
6386 * some task might have had allocated an extent from it already, but it
6387 * has not yet created a respective ordered extent (and added it to a
6388 * root's list of ordered extents).
6389 * Therefore wait for any task currently allocating extents, since the
6390 * block group's reservations counter is incremented while a read lock
6391 * on the groups' semaphore is held and decremented after releasing
6392 * the read access on that semaphore and creating the ordered extent.
6394 down_write(&space_info
->groups_sem
);
6395 up_write(&space_info
->groups_sem
);
6397 wait_var_event(&bg
->reservations
, !atomic_read(&bg
->reservations
));
6401 * btrfs_add_reserved_bytes - update the block_group and space info counters
6402 * @cache: The cache we are manipulating
6403 * @ram_bytes: The number of bytes of file content, and will be same to
6404 * @num_bytes except for the compress path.
6405 * @num_bytes: The number of bytes in question
6406 * @delalloc: The blocks are allocated for the delalloc write
6408 * This is called by the allocator when it reserves space. If this is a
6409 * reservation and the block group has become read only we cannot make the
6410 * reservation and return -EAGAIN, otherwise this function always succeeds.
6412 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache
*cache
,
6413 u64 ram_bytes
, u64 num_bytes
, int delalloc
)
6415 struct btrfs_space_info
*space_info
= cache
->space_info
;
6418 spin_lock(&space_info
->lock
);
6419 spin_lock(&cache
->lock
);
6423 cache
->reserved
+= num_bytes
;
6424 space_info
->bytes_reserved
+= num_bytes
;
6426 trace_btrfs_space_reservation(cache
->fs_info
,
6427 "space_info", space_info
->flags
,
6429 space_info
->bytes_may_use
-= ram_bytes
;
6431 cache
->delalloc_bytes
+= num_bytes
;
6433 spin_unlock(&cache
->lock
);
6434 spin_unlock(&space_info
->lock
);
6439 * btrfs_free_reserved_bytes - update the block_group and space info counters
6440 * @cache: The cache we are manipulating
6441 * @num_bytes: The number of bytes in question
6442 * @delalloc: The blocks are allocated for the delalloc write
6444 * This is called by somebody who is freeing space that was never actually used
6445 * on disk. For example if you reserve some space for a new leaf in transaction
6446 * A and before transaction A commits you free that leaf, you call this with
6447 * reserve set to 0 in order to clear the reservation.
6450 static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache
*cache
,
6451 u64 num_bytes
, int delalloc
)
6453 struct btrfs_space_info
*space_info
= cache
->space_info
;
6456 spin_lock(&space_info
->lock
);
6457 spin_lock(&cache
->lock
);
6459 space_info
->bytes_readonly
+= num_bytes
;
6460 cache
->reserved
-= num_bytes
;
6461 space_info
->bytes_reserved
-= num_bytes
;
6464 cache
->delalloc_bytes
-= num_bytes
;
6465 spin_unlock(&cache
->lock
);
6466 spin_unlock(&space_info
->lock
);
6469 void btrfs_prepare_extent_commit(struct btrfs_fs_info
*fs_info
)
6471 struct btrfs_caching_control
*next
;
6472 struct btrfs_caching_control
*caching_ctl
;
6473 struct btrfs_block_group_cache
*cache
;
6475 down_write(&fs_info
->commit_root_sem
);
6477 list_for_each_entry_safe(caching_ctl
, next
,
6478 &fs_info
->caching_block_groups
, list
) {
6479 cache
= caching_ctl
->block_group
;
6480 if (block_group_cache_done(cache
)) {
6481 cache
->last_byte_to_unpin
= (u64
)-1;
6482 list_del_init(&caching_ctl
->list
);
6483 put_caching_control(caching_ctl
);
6485 cache
->last_byte_to_unpin
= caching_ctl
->progress
;
6489 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
6490 fs_info
->pinned_extents
= &fs_info
->freed_extents
[1];
6492 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
6494 up_write(&fs_info
->commit_root_sem
);
6496 update_global_block_rsv(fs_info
);
6500 * Returns the free cluster for the given space info and sets empty_cluster to
6501 * what it should be based on the mount options.
6503 static struct btrfs_free_cluster
*
6504 fetch_cluster_info(struct btrfs_fs_info
*fs_info
,
6505 struct btrfs_space_info
*space_info
, u64
*empty_cluster
)
6507 struct btrfs_free_cluster
*ret
= NULL
;
6510 if (btrfs_mixed_space_info(space_info
))
6513 if (space_info
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
6514 ret
= &fs_info
->meta_alloc_cluster
;
6515 if (btrfs_test_opt(fs_info
, SSD
))
6516 *empty_cluster
= SZ_2M
;
6518 *empty_cluster
= SZ_64K
;
6519 } else if ((space_info
->flags
& BTRFS_BLOCK_GROUP_DATA
) &&
6520 btrfs_test_opt(fs_info
, SSD_SPREAD
)) {
6521 *empty_cluster
= SZ_2M
;
6522 ret
= &fs_info
->data_alloc_cluster
;
6528 static int unpin_extent_range(struct btrfs_fs_info
*fs_info
,
6530 const bool return_free_space
)
6532 struct btrfs_block_group_cache
*cache
= NULL
;
6533 struct btrfs_space_info
*space_info
;
6534 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
6535 struct btrfs_free_cluster
*cluster
= NULL
;
6537 u64 total_unpinned
= 0;
6538 u64 empty_cluster
= 0;
6541 while (start
<= end
) {
6544 start
>= cache
->key
.objectid
+ cache
->key
.offset
) {
6546 btrfs_put_block_group(cache
);
6548 cache
= btrfs_lookup_block_group(fs_info
, start
);
6549 BUG_ON(!cache
); /* Logic error */
6551 cluster
= fetch_cluster_info(fs_info
,
6554 empty_cluster
<<= 1;
6557 len
= cache
->key
.objectid
+ cache
->key
.offset
- start
;
6558 len
= min(len
, end
+ 1 - start
);
6560 if (start
< cache
->last_byte_to_unpin
) {
6561 len
= min(len
, cache
->last_byte_to_unpin
- start
);
6562 if (return_free_space
)
6563 btrfs_add_free_space(cache
, start
, len
);
6567 total_unpinned
+= len
;
6568 space_info
= cache
->space_info
;
6571 * If this space cluster has been marked as fragmented and we've
6572 * unpinned enough in this block group to potentially allow a
6573 * cluster to be created inside of it go ahead and clear the
6576 if (cluster
&& cluster
->fragmented
&&
6577 total_unpinned
> empty_cluster
) {
6578 spin_lock(&cluster
->lock
);
6579 cluster
->fragmented
= 0;
6580 spin_unlock(&cluster
->lock
);
6583 spin_lock(&space_info
->lock
);
6584 spin_lock(&cache
->lock
);
6585 cache
->pinned
-= len
;
6586 space_info
->bytes_pinned
-= len
;
6588 trace_btrfs_space_reservation(fs_info
, "pinned",
6589 space_info
->flags
, len
, 0);
6590 space_info
->max_extent_size
= 0;
6591 percpu_counter_add_batch(&space_info
->total_bytes_pinned
,
6592 -len
, BTRFS_TOTAL_BYTES_PINNED_BATCH
);
6594 space_info
->bytes_readonly
+= len
;
6597 spin_unlock(&cache
->lock
);
6598 if (!readonly
&& return_free_space
&&
6599 global_rsv
->space_info
== space_info
) {
6602 spin_lock(&global_rsv
->lock
);
6603 if (!global_rsv
->full
) {
6604 to_add
= min(len
, global_rsv
->size
-
6605 global_rsv
->reserved
);
6606 global_rsv
->reserved
+= to_add
;
6607 space_info
->bytes_may_use
+= to_add
;
6608 if (global_rsv
->reserved
>= global_rsv
->size
)
6609 global_rsv
->full
= 1;
6610 trace_btrfs_space_reservation(fs_info
,
6616 spin_unlock(&global_rsv
->lock
);
6617 /* Add to any tickets we may have */
6619 space_info_add_new_bytes(fs_info
, space_info
,
6622 spin_unlock(&space_info
->lock
);
6626 btrfs_put_block_group(cache
);
6630 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
)
6632 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
6633 struct btrfs_block_group_cache
*block_group
, *tmp
;
6634 struct list_head
*deleted_bgs
;
6635 struct extent_io_tree
*unpin
;
6640 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
6641 unpin
= &fs_info
->freed_extents
[1];
6643 unpin
= &fs_info
->freed_extents
[0];
6645 while (!trans
->aborted
) {
6646 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
6647 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
6648 EXTENT_DIRTY
, NULL
);
6650 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
6654 if (btrfs_test_opt(fs_info
, DISCARD
))
6655 ret
= btrfs_discard_extent(fs_info
, start
,
6656 end
+ 1 - start
, NULL
);
6658 clear_extent_dirty(unpin
, start
, end
);
6659 unpin_extent_range(fs_info
, start
, end
, true);
6660 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
6665 * Transaction is finished. We don't need the lock anymore. We
6666 * do need to clean up the block groups in case of a transaction
6669 deleted_bgs
= &trans
->transaction
->deleted_bgs
;
6670 list_for_each_entry_safe(block_group
, tmp
, deleted_bgs
, bg_list
) {
6674 if (!trans
->aborted
)
6675 ret
= btrfs_discard_extent(fs_info
,
6676 block_group
->key
.objectid
,
6677 block_group
->key
.offset
,
6680 list_del_init(&block_group
->bg_list
);
6681 btrfs_put_block_group_trimming(block_group
);
6682 btrfs_put_block_group(block_group
);
6685 const char *errstr
= btrfs_decode_error(ret
);
6687 "discard failed while removing blockgroup: errno=%d %s",
6695 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
6696 struct btrfs_delayed_ref_node
*node
, u64 parent
,
6697 u64 root_objectid
, u64 owner_objectid
,
6698 u64 owner_offset
, int refs_to_drop
,
6699 struct btrfs_delayed_extent_op
*extent_op
)
6701 struct btrfs_fs_info
*info
= trans
->fs_info
;
6702 struct btrfs_key key
;
6703 struct btrfs_path
*path
;
6704 struct btrfs_root
*extent_root
= info
->extent_root
;
6705 struct extent_buffer
*leaf
;
6706 struct btrfs_extent_item
*ei
;
6707 struct btrfs_extent_inline_ref
*iref
;
6710 int extent_slot
= 0;
6711 int found_extent
= 0;
6715 u64 bytenr
= node
->bytenr
;
6716 u64 num_bytes
= node
->num_bytes
;
6718 bool skinny_metadata
= btrfs_fs_incompat(info
, SKINNY_METADATA
);
6720 path
= btrfs_alloc_path();
6724 path
->reada
= READA_FORWARD
;
6725 path
->leave_spinning
= 1;
6727 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
6728 BUG_ON(!is_data
&& refs_to_drop
!= 1);
6731 skinny_metadata
= false;
6733 ret
= lookup_extent_backref(trans
, path
, &iref
, bytenr
, num_bytes
,
6734 parent
, root_objectid
, owner_objectid
,
6737 extent_slot
= path
->slots
[0];
6738 while (extent_slot
>= 0) {
6739 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
6741 if (key
.objectid
!= bytenr
)
6743 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
6744 key
.offset
== num_bytes
) {
6748 if (key
.type
== BTRFS_METADATA_ITEM_KEY
&&
6749 key
.offset
== owner_objectid
) {
6753 if (path
->slots
[0] - extent_slot
> 5)
6758 if (!found_extent
) {
6760 ret
= remove_extent_backref(trans
, path
, NULL
,
6762 is_data
, &last_ref
);
6764 btrfs_abort_transaction(trans
, ret
);
6767 btrfs_release_path(path
);
6768 path
->leave_spinning
= 1;
6770 key
.objectid
= bytenr
;
6771 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6772 key
.offset
= num_bytes
;
6774 if (!is_data
&& skinny_metadata
) {
6775 key
.type
= BTRFS_METADATA_ITEM_KEY
;
6776 key
.offset
= owner_objectid
;
6779 ret
= btrfs_search_slot(trans
, extent_root
,
6781 if (ret
> 0 && skinny_metadata
&& path
->slots
[0]) {
6783 * Couldn't find our skinny metadata item,
6784 * see if we have ye olde extent item.
6787 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
6789 if (key
.objectid
== bytenr
&&
6790 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
6791 key
.offset
== num_bytes
)
6795 if (ret
> 0 && skinny_metadata
) {
6796 skinny_metadata
= false;
6797 key
.objectid
= bytenr
;
6798 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6799 key
.offset
= num_bytes
;
6800 btrfs_release_path(path
);
6801 ret
= btrfs_search_slot(trans
, extent_root
,
6807 "umm, got %d back from search, was looking for %llu",
6810 btrfs_print_leaf(path
->nodes
[0]);
6813 btrfs_abort_transaction(trans
, ret
);
6816 extent_slot
= path
->slots
[0];
6818 } else if (WARN_ON(ret
== -ENOENT
)) {
6819 btrfs_print_leaf(path
->nodes
[0]);
6821 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6822 bytenr
, parent
, root_objectid
, owner_objectid
,
6824 btrfs_abort_transaction(trans
, ret
);
6827 btrfs_abort_transaction(trans
, ret
);
6831 leaf
= path
->nodes
[0];
6832 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
6833 if (unlikely(item_size
< sizeof(*ei
))) {
6835 btrfs_print_v0_err(info
);
6836 btrfs_abort_transaction(trans
, ret
);
6839 ei
= btrfs_item_ptr(leaf
, extent_slot
,
6840 struct btrfs_extent_item
);
6841 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
6842 key
.type
== BTRFS_EXTENT_ITEM_KEY
) {
6843 struct btrfs_tree_block_info
*bi
;
6844 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
6845 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
6846 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
6849 refs
= btrfs_extent_refs(leaf
, ei
);
6850 if (refs
< refs_to_drop
) {
6852 "trying to drop %d refs but we only have %Lu for bytenr %Lu",
6853 refs_to_drop
, refs
, bytenr
);
6855 btrfs_abort_transaction(trans
, ret
);
6858 refs
-= refs_to_drop
;
6862 __run_delayed_extent_op(extent_op
, leaf
, ei
);
6864 * In the case of inline back ref, reference count will
6865 * be updated by remove_extent_backref
6868 BUG_ON(!found_extent
);
6870 btrfs_set_extent_refs(leaf
, ei
, refs
);
6871 btrfs_mark_buffer_dirty(leaf
);
6874 ret
= remove_extent_backref(trans
, path
, iref
,
6875 refs_to_drop
, is_data
,
6878 btrfs_abort_transaction(trans
, ret
);
6884 BUG_ON(is_data
&& refs_to_drop
!=
6885 extent_data_ref_count(path
, iref
));
6887 BUG_ON(path
->slots
[0] != extent_slot
);
6889 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
6890 path
->slots
[0] = extent_slot
;
6896 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
6899 btrfs_abort_transaction(trans
, ret
);
6902 btrfs_release_path(path
);
6905 ret
= btrfs_del_csums(trans
, info
, bytenr
, num_bytes
);
6907 btrfs_abort_transaction(trans
, ret
);
6912 ret
= add_to_free_space_tree(trans
, bytenr
, num_bytes
);
6914 btrfs_abort_transaction(trans
, ret
);
6918 ret
= update_block_group(trans
, info
, bytenr
, num_bytes
, 0);
6920 btrfs_abort_transaction(trans
, ret
);
6924 btrfs_release_path(path
);
6927 btrfs_free_path(path
);
6932 * when we free an block, it is possible (and likely) that we free the last
6933 * delayed ref for that extent as well. This searches the delayed ref tree for
6934 * a given extent, and if there are no other delayed refs to be processed, it
6935 * removes it from the tree.
6937 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
6940 struct btrfs_delayed_ref_head
*head
;
6941 struct btrfs_delayed_ref_root
*delayed_refs
;
6944 delayed_refs
= &trans
->transaction
->delayed_refs
;
6945 spin_lock(&delayed_refs
->lock
);
6946 head
= btrfs_find_delayed_ref_head(delayed_refs
, bytenr
);
6948 goto out_delayed_unlock
;
6950 spin_lock(&head
->lock
);
6951 if (!RB_EMPTY_ROOT(&head
->ref_tree
))
6954 if (head
->extent_op
) {
6955 if (!head
->must_insert_reserved
)
6957 btrfs_free_delayed_extent_op(head
->extent_op
);
6958 head
->extent_op
= NULL
;
6962 * waiting for the lock here would deadlock. If someone else has it
6963 * locked they are already in the process of dropping it anyway
6965 if (!mutex_trylock(&head
->mutex
))
6969 * at this point we have a head with no other entries. Go
6970 * ahead and process it.
6972 rb_erase(&head
->href_node
, &delayed_refs
->href_root
);
6973 RB_CLEAR_NODE(&head
->href_node
);
6974 atomic_dec(&delayed_refs
->num_entries
);
6977 * we don't take a ref on the node because we're removing it from the
6978 * tree, so we just steal the ref the tree was holding.
6980 delayed_refs
->num_heads
--;
6981 if (head
->processing
== 0)
6982 delayed_refs
->num_heads_ready
--;
6983 head
->processing
= 0;
6984 spin_unlock(&head
->lock
);
6985 spin_unlock(&delayed_refs
->lock
);
6987 BUG_ON(head
->extent_op
);
6988 if (head
->must_insert_reserved
)
6991 mutex_unlock(&head
->mutex
);
6992 btrfs_put_delayed_ref_head(head
);
6995 spin_unlock(&head
->lock
);
6998 spin_unlock(&delayed_refs
->lock
);
7002 void btrfs_free_tree_block(struct btrfs_trans_handle
*trans
,
7003 struct btrfs_root
*root
,
7004 struct extent_buffer
*buf
,
7005 u64 parent
, int last_ref
)
7007 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7011 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
7012 int old_ref_mod
, new_ref_mod
;
7014 btrfs_ref_tree_mod(root
, buf
->start
, buf
->len
, parent
,
7015 root
->root_key
.objectid
,
7016 btrfs_header_level(buf
), 0,
7017 BTRFS_DROP_DELAYED_REF
);
7018 ret
= btrfs_add_delayed_tree_ref(trans
, buf
->start
,
7020 root
->root_key
.objectid
,
7021 btrfs_header_level(buf
),
7022 BTRFS_DROP_DELAYED_REF
, NULL
,
7023 &old_ref_mod
, &new_ref_mod
);
7024 BUG_ON(ret
); /* -ENOMEM */
7025 pin
= old_ref_mod
>= 0 && new_ref_mod
< 0;
7028 if (last_ref
&& btrfs_header_generation(buf
) == trans
->transid
) {
7029 struct btrfs_block_group_cache
*cache
;
7031 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
7032 ret
= check_ref_cleanup(trans
, buf
->start
);
7038 cache
= btrfs_lookup_block_group(fs_info
, buf
->start
);
7040 if (btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
7041 pin_down_extent(fs_info
, cache
, buf
->start
,
7043 btrfs_put_block_group(cache
);
7047 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
));
7049 btrfs_add_free_space(cache
, buf
->start
, buf
->len
);
7050 btrfs_free_reserved_bytes(cache
, buf
->len
, 0);
7051 btrfs_put_block_group(cache
);
7052 trace_btrfs_reserved_extent_free(fs_info
, buf
->start
, buf
->len
);
7056 add_pinned_bytes(fs_info
, buf
->len
, true,
7057 root
->root_key
.objectid
);
7061 * Deleting the buffer, clear the corrupt flag since it doesn't
7064 clear_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
);
7068 /* Can return -ENOMEM */
7069 int btrfs_free_extent(struct btrfs_trans_handle
*trans
,
7070 struct btrfs_root
*root
,
7071 u64 bytenr
, u64 num_bytes
, u64 parent
, u64 root_objectid
,
7072 u64 owner
, u64 offset
)
7074 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7075 int old_ref_mod
, new_ref_mod
;
7078 if (btrfs_is_testing(fs_info
))
7081 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
)
7082 btrfs_ref_tree_mod(root
, bytenr
, num_bytes
, parent
,
7083 root_objectid
, owner
, offset
,
7084 BTRFS_DROP_DELAYED_REF
);
7087 * tree log blocks never actually go into the extent allocation
7088 * tree, just update pinning info and exit early.
7090 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
7091 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
7092 /* unlocks the pinned mutex */
7093 btrfs_pin_extent(fs_info
, bytenr
, num_bytes
, 1);
7094 old_ref_mod
= new_ref_mod
= 0;
7096 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
7097 ret
= btrfs_add_delayed_tree_ref(trans
, bytenr
,
7099 root_objectid
, (int)owner
,
7100 BTRFS_DROP_DELAYED_REF
, NULL
,
7101 &old_ref_mod
, &new_ref_mod
);
7103 ret
= btrfs_add_delayed_data_ref(trans
, bytenr
,
7105 root_objectid
, owner
, offset
,
7106 0, BTRFS_DROP_DELAYED_REF
,
7107 &old_ref_mod
, &new_ref_mod
);
7110 if (ret
== 0 && old_ref_mod
>= 0 && new_ref_mod
< 0) {
7111 bool metadata
= owner
< BTRFS_FIRST_FREE_OBJECTID
;
7113 add_pinned_bytes(fs_info
, num_bytes
, metadata
, root_objectid
);
7120 * when we wait for progress in the block group caching, its because
7121 * our allocation attempt failed at least once. So, we must sleep
7122 * and let some progress happen before we try again.
7124 * This function will sleep at least once waiting for new free space to
7125 * show up, and then it will check the block group free space numbers
7126 * for our min num_bytes. Another option is to have it go ahead
7127 * and look in the rbtree for a free extent of a given size, but this
7130 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7131 * any of the information in this block group.
7133 static noinline
void
7134 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
7137 struct btrfs_caching_control
*caching_ctl
;
7139 caching_ctl
= get_caching_control(cache
);
7143 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
) ||
7144 (cache
->free_space_ctl
->free_space
>= num_bytes
));
7146 put_caching_control(caching_ctl
);
7150 wait_block_group_cache_done(struct btrfs_block_group_cache
*cache
)
7152 struct btrfs_caching_control
*caching_ctl
;
7155 caching_ctl
= get_caching_control(cache
);
7157 return (cache
->cached
== BTRFS_CACHE_ERROR
) ? -EIO
: 0;
7159 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
));
7160 if (cache
->cached
== BTRFS_CACHE_ERROR
)
7162 put_caching_control(caching_ctl
);
7166 enum btrfs_loop_type
{
7167 LOOP_CACHING_NOWAIT
= 0,
7168 LOOP_CACHING_WAIT
= 1,
7169 LOOP_ALLOC_CHUNK
= 2,
7170 LOOP_NO_EMPTY_SIZE
= 3,
7174 btrfs_lock_block_group(struct btrfs_block_group_cache
*cache
,
7178 down_read(&cache
->data_rwsem
);
7182 btrfs_grab_block_group(struct btrfs_block_group_cache
*cache
,
7185 btrfs_get_block_group(cache
);
7187 down_read(&cache
->data_rwsem
);
7190 static struct btrfs_block_group_cache
*
7191 btrfs_lock_cluster(struct btrfs_block_group_cache
*block_group
,
7192 struct btrfs_free_cluster
*cluster
,
7195 struct btrfs_block_group_cache
*used_bg
= NULL
;
7197 spin_lock(&cluster
->refill_lock
);
7199 used_bg
= cluster
->block_group
;
7203 if (used_bg
== block_group
)
7206 btrfs_get_block_group(used_bg
);
7211 if (down_read_trylock(&used_bg
->data_rwsem
))
7214 spin_unlock(&cluster
->refill_lock
);
7216 /* We should only have one-level nested. */
7217 down_read_nested(&used_bg
->data_rwsem
, SINGLE_DEPTH_NESTING
);
7219 spin_lock(&cluster
->refill_lock
);
7220 if (used_bg
== cluster
->block_group
)
7223 up_read(&used_bg
->data_rwsem
);
7224 btrfs_put_block_group(used_bg
);
7229 btrfs_release_block_group(struct btrfs_block_group_cache
*cache
,
7233 up_read(&cache
->data_rwsem
);
7234 btrfs_put_block_group(cache
);
7238 * walks the btree of allocated extents and find a hole of a given size.
7239 * The key ins is changed to record the hole:
7240 * ins->objectid == start position
7241 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7242 * ins->offset == the size of the hole.
7243 * Any available blocks before search_start are skipped.
7245 * If there is no suitable free space, we will record the max size of
7246 * the free space extent currently.
7248 static noinline
int find_free_extent(struct btrfs_fs_info
*fs_info
,
7249 u64 ram_bytes
, u64 num_bytes
, u64 empty_size
,
7250 u64 hint_byte
, struct btrfs_key
*ins
,
7251 u64 flags
, int delalloc
)
7254 struct btrfs_root
*root
= fs_info
->extent_root
;
7255 struct btrfs_free_cluster
*last_ptr
= NULL
;
7256 struct btrfs_block_group_cache
*block_group
= NULL
;
7257 u64 search_start
= 0;
7258 u64 max_extent_size
= 0;
7259 u64 empty_cluster
= 0;
7260 struct btrfs_space_info
*space_info
;
7262 int index
= btrfs_bg_flags_to_raid_index(flags
);
7263 bool failed_cluster_refill
= false;
7264 bool failed_alloc
= false;
7265 bool use_cluster
= true;
7266 bool have_caching_bg
= false;
7267 bool orig_have_caching_bg
= false;
7268 bool full_search
= false;
7270 WARN_ON(num_bytes
< fs_info
->sectorsize
);
7271 ins
->type
= BTRFS_EXTENT_ITEM_KEY
;
7275 trace_find_free_extent(fs_info
, num_bytes
, empty_size
, flags
);
7277 space_info
= __find_space_info(fs_info
, flags
);
7279 btrfs_err(fs_info
, "No space info for %llu", flags
);
7284 * If our free space is heavily fragmented we may not be able to make
7285 * big contiguous allocations, so instead of doing the expensive search
7286 * for free space, simply return ENOSPC with our max_extent_size so we
7287 * can go ahead and search for a more manageable chunk.
7289 * If our max_extent_size is large enough for our allocation simply
7290 * disable clustering since we will likely not be able to find enough
7291 * space to create a cluster and induce latency trying.
7293 if (unlikely(space_info
->max_extent_size
)) {
7294 spin_lock(&space_info
->lock
);
7295 if (space_info
->max_extent_size
&&
7296 num_bytes
> space_info
->max_extent_size
) {
7297 ins
->offset
= space_info
->max_extent_size
;
7298 spin_unlock(&space_info
->lock
);
7300 } else if (space_info
->max_extent_size
) {
7301 use_cluster
= false;
7303 spin_unlock(&space_info
->lock
);
7306 last_ptr
= fetch_cluster_info(fs_info
, space_info
, &empty_cluster
);
7308 spin_lock(&last_ptr
->lock
);
7309 if (last_ptr
->block_group
)
7310 hint_byte
= last_ptr
->window_start
;
7311 if (last_ptr
->fragmented
) {
7313 * We still set window_start so we can keep track of the
7314 * last place we found an allocation to try and save
7317 hint_byte
= last_ptr
->window_start
;
7318 use_cluster
= false;
7320 spin_unlock(&last_ptr
->lock
);
7323 search_start
= max(search_start
, first_logical_byte(fs_info
, 0));
7324 search_start
= max(search_start
, hint_byte
);
7325 if (search_start
== hint_byte
) {
7326 block_group
= btrfs_lookup_block_group(fs_info
, search_start
);
7328 * we don't want to use the block group if it doesn't match our
7329 * allocation bits, or if its not cached.
7331 * However if we are re-searching with an ideal block group
7332 * picked out then we don't care that the block group is cached.
7334 if (block_group
&& block_group_bits(block_group
, flags
) &&
7335 block_group
->cached
!= BTRFS_CACHE_NO
) {
7336 down_read(&space_info
->groups_sem
);
7337 if (list_empty(&block_group
->list
) ||
7340 * someone is removing this block group,
7341 * we can't jump into the have_block_group
7342 * target because our list pointers are not
7345 btrfs_put_block_group(block_group
);
7346 up_read(&space_info
->groups_sem
);
7348 index
= btrfs_bg_flags_to_raid_index(
7349 block_group
->flags
);
7350 btrfs_lock_block_group(block_group
, delalloc
);
7351 goto have_block_group
;
7353 } else if (block_group
) {
7354 btrfs_put_block_group(block_group
);
7358 have_caching_bg
= false;
7359 if (index
== 0 || index
== btrfs_bg_flags_to_raid_index(flags
))
7361 down_read(&space_info
->groups_sem
);
7362 list_for_each_entry(block_group
, &space_info
->block_groups
[index
],
7367 /* If the block group is read-only, we can skip it entirely. */
7368 if (unlikely(block_group
->ro
))
7371 btrfs_grab_block_group(block_group
, delalloc
);
7372 search_start
= block_group
->key
.objectid
;
7375 * this can happen if we end up cycling through all the
7376 * raid types, but we want to make sure we only allocate
7377 * for the proper type.
7379 if (!block_group_bits(block_group
, flags
)) {
7380 u64 extra
= BTRFS_BLOCK_GROUP_DUP
|
7381 BTRFS_BLOCK_GROUP_RAID1
|
7382 BTRFS_BLOCK_GROUP_RAID5
|
7383 BTRFS_BLOCK_GROUP_RAID6
|
7384 BTRFS_BLOCK_GROUP_RAID10
;
7387 * if they asked for extra copies and this block group
7388 * doesn't provide them, bail. This does allow us to
7389 * fill raid0 from raid1.
7391 if ((flags
& extra
) && !(block_group
->flags
& extra
))
7396 cached
= block_group_cache_done(block_group
);
7397 if (unlikely(!cached
)) {
7398 have_caching_bg
= true;
7399 ret
= cache_block_group(block_group
, 0);
7404 if (unlikely(block_group
->cached
== BTRFS_CACHE_ERROR
))
7408 * Ok we want to try and use the cluster allocator, so
7411 if (last_ptr
&& use_cluster
) {
7412 struct btrfs_block_group_cache
*used_block_group
;
7413 unsigned long aligned_cluster
;
7415 * the refill lock keeps out other
7416 * people trying to start a new cluster
7418 used_block_group
= btrfs_lock_cluster(block_group
,
7421 if (!used_block_group
)
7422 goto refill_cluster
;
7424 if (used_block_group
!= block_group
&&
7425 (used_block_group
->ro
||
7426 !block_group_bits(used_block_group
, flags
)))
7427 goto release_cluster
;
7429 offset
= btrfs_alloc_from_cluster(used_block_group
,
7432 used_block_group
->key
.objectid
,
7435 /* we have a block, we're done */
7436 spin_unlock(&last_ptr
->refill_lock
);
7437 trace_btrfs_reserve_extent_cluster(
7439 search_start
, num_bytes
);
7440 if (used_block_group
!= block_group
) {
7441 btrfs_release_block_group(block_group
,
7443 block_group
= used_block_group
;
7448 WARN_ON(last_ptr
->block_group
!= used_block_group
);
7450 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7451 * set up a new clusters, so lets just skip it
7452 * and let the allocator find whatever block
7453 * it can find. If we reach this point, we
7454 * will have tried the cluster allocator
7455 * plenty of times and not have found
7456 * anything, so we are likely way too
7457 * fragmented for the clustering stuff to find
7460 * However, if the cluster is taken from the
7461 * current block group, release the cluster
7462 * first, so that we stand a better chance of
7463 * succeeding in the unclustered
7465 if (loop
>= LOOP_NO_EMPTY_SIZE
&&
7466 used_block_group
!= block_group
) {
7467 spin_unlock(&last_ptr
->refill_lock
);
7468 btrfs_release_block_group(used_block_group
,
7470 goto unclustered_alloc
;
7474 * this cluster didn't work out, free it and
7477 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
7479 if (used_block_group
!= block_group
)
7480 btrfs_release_block_group(used_block_group
,
7483 if (loop
>= LOOP_NO_EMPTY_SIZE
) {
7484 spin_unlock(&last_ptr
->refill_lock
);
7485 goto unclustered_alloc
;
7488 aligned_cluster
= max_t(unsigned long,
7489 empty_cluster
+ empty_size
,
7490 block_group
->full_stripe_len
);
7492 /* allocate a cluster in this block group */
7493 ret
= btrfs_find_space_cluster(fs_info
, block_group
,
7494 last_ptr
, search_start
,
7499 * now pull our allocation out of this
7502 offset
= btrfs_alloc_from_cluster(block_group
,
7508 /* we found one, proceed */
7509 spin_unlock(&last_ptr
->refill_lock
);
7510 trace_btrfs_reserve_extent_cluster(
7511 block_group
, search_start
,
7515 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
7516 && !failed_cluster_refill
) {
7517 spin_unlock(&last_ptr
->refill_lock
);
7519 failed_cluster_refill
= true;
7520 wait_block_group_cache_progress(block_group
,
7521 num_bytes
+ empty_cluster
+ empty_size
);
7522 goto have_block_group
;
7526 * at this point we either didn't find a cluster
7527 * or we weren't able to allocate a block from our
7528 * cluster. Free the cluster we've been trying
7529 * to use, and go to the next block group
7531 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
7532 spin_unlock(&last_ptr
->refill_lock
);
7538 * We are doing an unclustered alloc, set the fragmented flag so
7539 * we don't bother trying to setup a cluster again until we get
7542 if (unlikely(last_ptr
)) {
7543 spin_lock(&last_ptr
->lock
);
7544 last_ptr
->fragmented
= 1;
7545 spin_unlock(&last_ptr
->lock
);
7548 struct btrfs_free_space_ctl
*ctl
=
7549 block_group
->free_space_ctl
;
7551 spin_lock(&ctl
->tree_lock
);
7552 if (ctl
->free_space
<
7553 num_bytes
+ empty_cluster
+ empty_size
) {
7554 if (ctl
->free_space
> max_extent_size
)
7555 max_extent_size
= ctl
->free_space
;
7556 spin_unlock(&ctl
->tree_lock
);
7559 spin_unlock(&ctl
->tree_lock
);
7562 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
7563 num_bytes
, empty_size
,
7566 * If we didn't find a chunk, and we haven't failed on this
7567 * block group before, and this block group is in the middle of
7568 * caching and we are ok with waiting, then go ahead and wait
7569 * for progress to be made, and set failed_alloc to true.
7571 * If failed_alloc is true then we've already waited on this
7572 * block group once and should move on to the next block group.
7574 if (!offset
&& !failed_alloc
&& !cached
&&
7575 loop
> LOOP_CACHING_NOWAIT
) {
7576 wait_block_group_cache_progress(block_group
,
7577 num_bytes
+ empty_size
);
7578 failed_alloc
= true;
7579 goto have_block_group
;
7580 } else if (!offset
) {
7584 search_start
= ALIGN(offset
, fs_info
->stripesize
);
7586 /* move on to the next group */
7587 if (search_start
+ num_bytes
>
7588 block_group
->key
.objectid
+ block_group
->key
.offset
) {
7589 btrfs_add_free_space(block_group
, offset
, num_bytes
);
7593 if (offset
< search_start
)
7594 btrfs_add_free_space(block_group
, offset
,
7595 search_start
- offset
);
7596 BUG_ON(offset
> search_start
);
7598 ret
= btrfs_add_reserved_bytes(block_group
, ram_bytes
,
7599 num_bytes
, delalloc
);
7600 if (ret
== -EAGAIN
) {
7601 btrfs_add_free_space(block_group
, offset
, num_bytes
);
7604 btrfs_inc_block_group_reservations(block_group
);
7606 /* we are all good, lets return */
7607 ins
->objectid
= search_start
;
7608 ins
->offset
= num_bytes
;
7610 trace_btrfs_reserve_extent(block_group
, search_start
, num_bytes
);
7611 btrfs_release_block_group(block_group
, delalloc
);
7614 failed_cluster_refill
= false;
7615 failed_alloc
= false;
7616 BUG_ON(btrfs_bg_flags_to_raid_index(block_group
->flags
) !=
7618 btrfs_release_block_group(block_group
, delalloc
);
7621 up_read(&space_info
->groups_sem
);
7623 if ((loop
== LOOP_CACHING_NOWAIT
) && have_caching_bg
7624 && !orig_have_caching_bg
)
7625 orig_have_caching_bg
= true;
7627 if (!ins
->objectid
&& loop
>= LOOP_CACHING_WAIT
&& have_caching_bg
)
7630 if (!ins
->objectid
&& ++index
< BTRFS_NR_RAID_TYPES
)
7634 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7635 * caching kthreads as we move along
7636 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7637 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7638 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7641 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
) {
7643 if (loop
== LOOP_CACHING_NOWAIT
) {
7645 * We want to skip the LOOP_CACHING_WAIT step if we
7646 * don't have any uncached bgs and we've already done a
7647 * full search through.
7649 if (orig_have_caching_bg
|| !full_search
)
7650 loop
= LOOP_CACHING_WAIT
;
7652 loop
= LOOP_ALLOC_CHUNK
;
7657 if (loop
== LOOP_ALLOC_CHUNK
) {
7658 struct btrfs_trans_handle
*trans
;
7661 trans
= current
->journal_info
;
7665 trans
= btrfs_join_transaction(root
);
7667 if (IS_ERR(trans
)) {
7668 ret
= PTR_ERR(trans
);
7672 ret
= do_chunk_alloc(trans
, flags
, CHUNK_ALLOC_FORCE
);
7675 * If we can't allocate a new chunk we've already looped
7676 * through at least once, move on to the NO_EMPTY_SIZE
7680 loop
= LOOP_NO_EMPTY_SIZE
;
7683 * Do not bail out on ENOSPC since we
7684 * can do more things.
7686 if (ret
< 0 && ret
!= -ENOSPC
)
7687 btrfs_abort_transaction(trans
, ret
);
7691 btrfs_end_transaction(trans
);
7696 if (loop
== LOOP_NO_EMPTY_SIZE
) {
7698 * Don't loop again if we already have no empty_size and
7701 if (empty_size
== 0 &&
7702 empty_cluster
== 0) {
7711 } else if (!ins
->objectid
) {
7713 } else if (ins
->objectid
) {
7714 if (!use_cluster
&& last_ptr
) {
7715 spin_lock(&last_ptr
->lock
);
7716 last_ptr
->window_start
= ins
->objectid
;
7717 spin_unlock(&last_ptr
->lock
);
7722 if (ret
== -ENOSPC
) {
7723 spin_lock(&space_info
->lock
);
7724 space_info
->max_extent_size
= max_extent_size
;
7725 spin_unlock(&space_info
->lock
);
7726 ins
->offset
= max_extent_size
;
7731 static void dump_space_info(struct btrfs_fs_info
*fs_info
,
7732 struct btrfs_space_info
*info
, u64 bytes
,
7733 int dump_block_groups
)
7735 struct btrfs_block_group_cache
*cache
;
7738 spin_lock(&info
->lock
);
7739 btrfs_info(fs_info
, "space_info %llu has %llu free, is %sfull",
7741 info
->total_bytes
- btrfs_space_info_used(info
, true),
7742 info
->full
? "" : "not ");
7744 "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu",
7745 info
->total_bytes
, info
->bytes_used
, info
->bytes_pinned
,
7746 info
->bytes_reserved
, info
->bytes_may_use
,
7747 info
->bytes_readonly
);
7748 spin_unlock(&info
->lock
);
7750 if (!dump_block_groups
)
7753 down_read(&info
->groups_sem
);
7755 list_for_each_entry(cache
, &info
->block_groups
[index
], list
) {
7756 spin_lock(&cache
->lock
);
7758 "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s",
7759 cache
->key
.objectid
, cache
->key
.offset
,
7760 btrfs_block_group_used(&cache
->item
), cache
->pinned
,
7761 cache
->reserved
, cache
->ro
? "[readonly]" : "");
7762 btrfs_dump_free_space(cache
, bytes
);
7763 spin_unlock(&cache
->lock
);
7765 if (++index
< BTRFS_NR_RAID_TYPES
)
7767 up_read(&info
->groups_sem
);
7771 * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a
7772 * hole that is at least as big as @num_bytes.
7774 * @root - The root that will contain this extent
7776 * @ram_bytes - The amount of space in ram that @num_bytes take. This
7777 * is used for accounting purposes. This value differs
7778 * from @num_bytes only in the case of compressed extents.
7780 * @num_bytes - Number of bytes to allocate on-disk.
7782 * @min_alloc_size - Indicates the minimum amount of space that the
7783 * allocator should try to satisfy. In some cases
7784 * @num_bytes may be larger than what is required and if
7785 * the filesystem is fragmented then allocation fails.
7786 * However, the presence of @min_alloc_size gives a
7787 * chance to try and satisfy the smaller allocation.
7789 * @empty_size - A hint that you plan on doing more COW. This is the
7790 * size in bytes the allocator should try to find free
7791 * next to the block it returns. This is just a hint and
7792 * may be ignored by the allocator.
7794 * @hint_byte - Hint to the allocator to start searching above the byte
7795 * address passed. It might be ignored.
7797 * @ins - This key is modified to record the found hole. It will
7798 * have the following values:
7799 * ins->objectid == start position
7800 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7801 * ins->offset == the size of the hole.
7803 * @is_data - Boolean flag indicating whether an extent is
7804 * allocated for data (true) or metadata (false)
7806 * @delalloc - Boolean flag indicating whether this allocation is for
7807 * delalloc or not. If 'true' data_rwsem of block groups
7808 * is going to be acquired.
7811 * Returns 0 when an allocation succeeded or < 0 when an error occurred. In
7812 * case -ENOSPC is returned then @ins->offset will contain the size of the
7813 * largest available hole the allocator managed to find.
7815 int btrfs_reserve_extent(struct btrfs_root
*root
, u64 ram_bytes
,
7816 u64 num_bytes
, u64 min_alloc_size
,
7817 u64 empty_size
, u64 hint_byte
,
7818 struct btrfs_key
*ins
, int is_data
, int delalloc
)
7820 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7821 bool final_tried
= num_bytes
== min_alloc_size
;
7825 flags
= get_alloc_profile_by_root(root
, is_data
);
7827 WARN_ON(num_bytes
< fs_info
->sectorsize
);
7828 ret
= find_free_extent(fs_info
, ram_bytes
, num_bytes
, empty_size
,
7829 hint_byte
, ins
, flags
, delalloc
);
7830 if (!ret
&& !is_data
) {
7831 btrfs_dec_block_group_reservations(fs_info
, ins
->objectid
);
7832 } else if (ret
== -ENOSPC
) {
7833 if (!final_tried
&& ins
->offset
) {
7834 num_bytes
= min(num_bytes
>> 1, ins
->offset
);
7835 num_bytes
= round_down(num_bytes
,
7836 fs_info
->sectorsize
);
7837 num_bytes
= max(num_bytes
, min_alloc_size
);
7838 ram_bytes
= num_bytes
;
7839 if (num_bytes
== min_alloc_size
)
7842 } else if (btrfs_test_opt(fs_info
, ENOSPC_DEBUG
)) {
7843 struct btrfs_space_info
*sinfo
;
7845 sinfo
= __find_space_info(fs_info
, flags
);
7847 "allocation failed flags %llu, wanted %llu",
7850 dump_space_info(fs_info
, sinfo
, num_bytes
, 1);
7857 static int __btrfs_free_reserved_extent(struct btrfs_fs_info
*fs_info
,
7859 int pin
, int delalloc
)
7861 struct btrfs_block_group_cache
*cache
;
7864 cache
= btrfs_lookup_block_group(fs_info
, start
);
7866 btrfs_err(fs_info
, "Unable to find block group for %llu",
7872 pin_down_extent(fs_info
, cache
, start
, len
, 1);
7874 if (btrfs_test_opt(fs_info
, DISCARD
))
7875 ret
= btrfs_discard_extent(fs_info
, start
, len
, NULL
);
7876 btrfs_add_free_space(cache
, start
, len
);
7877 btrfs_free_reserved_bytes(cache
, len
, delalloc
);
7878 trace_btrfs_reserved_extent_free(fs_info
, start
, len
);
7881 btrfs_put_block_group(cache
);
7885 int btrfs_free_reserved_extent(struct btrfs_fs_info
*fs_info
,
7886 u64 start
, u64 len
, int delalloc
)
7888 return __btrfs_free_reserved_extent(fs_info
, start
, len
, 0, delalloc
);
7891 int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info
*fs_info
,
7894 return __btrfs_free_reserved_extent(fs_info
, start
, len
, 1, 0);
7897 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
7898 u64 parent
, u64 root_objectid
,
7899 u64 flags
, u64 owner
, u64 offset
,
7900 struct btrfs_key
*ins
, int ref_mod
)
7902 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
7904 struct btrfs_extent_item
*extent_item
;
7905 struct btrfs_extent_inline_ref
*iref
;
7906 struct btrfs_path
*path
;
7907 struct extent_buffer
*leaf
;
7912 type
= BTRFS_SHARED_DATA_REF_KEY
;
7914 type
= BTRFS_EXTENT_DATA_REF_KEY
;
7916 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
7918 path
= btrfs_alloc_path();
7922 path
->leave_spinning
= 1;
7923 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
7926 btrfs_free_path(path
);
7930 leaf
= path
->nodes
[0];
7931 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
7932 struct btrfs_extent_item
);
7933 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
7934 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
7935 btrfs_set_extent_flags(leaf
, extent_item
,
7936 flags
| BTRFS_EXTENT_FLAG_DATA
);
7938 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
7939 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
7941 struct btrfs_shared_data_ref
*ref
;
7942 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
7943 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
7944 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
7946 struct btrfs_extent_data_ref
*ref
;
7947 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
7948 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
7949 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
7950 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
7951 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
7954 btrfs_mark_buffer_dirty(path
->nodes
[0]);
7955 btrfs_free_path(path
);
7957 ret
= remove_from_free_space_tree(trans
, ins
->objectid
, ins
->offset
);
7961 ret
= update_block_group(trans
, fs_info
, ins
->objectid
, ins
->offset
, 1);
7962 if (ret
) { /* -ENOENT, logic error */
7963 btrfs_err(fs_info
, "update block group failed for %llu %llu",
7964 ins
->objectid
, ins
->offset
);
7967 trace_btrfs_reserved_extent_alloc(fs_info
, ins
->objectid
, ins
->offset
);
7971 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
7972 struct btrfs_delayed_ref_node
*node
,
7973 struct btrfs_delayed_extent_op
*extent_op
)
7975 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
7977 struct btrfs_extent_item
*extent_item
;
7978 struct btrfs_key extent_key
;
7979 struct btrfs_tree_block_info
*block_info
;
7980 struct btrfs_extent_inline_ref
*iref
;
7981 struct btrfs_path
*path
;
7982 struct extent_buffer
*leaf
;
7983 struct btrfs_delayed_tree_ref
*ref
;
7984 u32 size
= sizeof(*extent_item
) + sizeof(*iref
);
7986 u64 flags
= extent_op
->flags_to_set
;
7987 bool skinny_metadata
= btrfs_fs_incompat(fs_info
, SKINNY_METADATA
);
7989 ref
= btrfs_delayed_node_to_tree_ref(node
);
7991 extent_key
.objectid
= node
->bytenr
;
7992 if (skinny_metadata
) {
7993 extent_key
.offset
= ref
->level
;
7994 extent_key
.type
= BTRFS_METADATA_ITEM_KEY
;
7995 num_bytes
= fs_info
->nodesize
;
7997 extent_key
.offset
= node
->num_bytes
;
7998 extent_key
.type
= BTRFS_EXTENT_ITEM_KEY
;
7999 size
+= sizeof(*block_info
);
8000 num_bytes
= node
->num_bytes
;
8003 path
= btrfs_alloc_path();
8005 btrfs_free_and_pin_reserved_extent(fs_info
,
8006 extent_key
.objectid
,
8011 path
->leave_spinning
= 1;
8012 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
8015 btrfs_free_path(path
);
8016 btrfs_free_and_pin_reserved_extent(fs_info
,
8017 extent_key
.objectid
,
8022 leaf
= path
->nodes
[0];
8023 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
8024 struct btrfs_extent_item
);
8025 btrfs_set_extent_refs(leaf
, extent_item
, 1);
8026 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
8027 btrfs_set_extent_flags(leaf
, extent_item
,
8028 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
8030 if (skinny_metadata
) {
8031 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
8033 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
8034 btrfs_set_tree_block_key(leaf
, block_info
, &extent_op
->key
);
8035 btrfs_set_tree_block_level(leaf
, block_info
, ref
->level
);
8036 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
8039 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
8040 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
8041 btrfs_set_extent_inline_ref_type(leaf
, iref
,
8042 BTRFS_SHARED_BLOCK_REF_KEY
);
8043 btrfs_set_extent_inline_ref_offset(leaf
, iref
, ref
->parent
);
8045 btrfs_set_extent_inline_ref_type(leaf
, iref
,
8046 BTRFS_TREE_BLOCK_REF_KEY
);
8047 btrfs_set_extent_inline_ref_offset(leaf
, iref
, ref
->root
);
8050 btrfs_mark_buffer_dirty(leaf
);
8051 btrfs_free_path(path
);
8053 ret
= remove_from_free_space_tree(trans
, extent_key
.objectid
,
8058 ret
= update_block_group(trans
, fs_info
, extent_key
.objectid
,
8059 fs_info
->nodesize
, 1);
8060 if (ret
) { /* -ENOENT, logic error */
8061 btrfs_err(fs_info
, "update block group failed for %llu %llu",
8062 extent_key
.objectid
, extent_key
.offset
);
8066 trace_btrfs_reserved_extent_alloc(fs_info
, extent_key
.objectid
,
8071 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
8072 struct btrfs_root
*root
, u64 owner
,
8073 u64 offset
, u64 ram_bytes
,
8074 struct btrfs_key
*ins
)
8078 BUG_ON(root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
);
8080 btrfs_ref_tree_mod(root
, ins
->objectid
, ins
->offset
, 0,
8081 root
->root_key
.objectid
, owner
, offset
,
8082 BTRFS_ADD_DELAYED_EXTENT
);
8084 ret
= btrfs_add_delayed_data_ref(trans
, ins
->objectid
,
8086 root
->root_key
.objectid
, owner
,
8088 BTRFS_ADD_DELAYED_EXTENT
, NULL
, NULL
);
8093 * this is used by the tree logging recovery code. It records that
8094 * an extent has been allocated and makes sure to clear the free
8095 * space cache bits as well
8097 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
8098 u64 root_objectid
, u64 owner
, u64 offset
,
8099 struct btrfs_key
*ins
)
8101 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
8103 struct btrfs_block_group_cache
*block_group
;
8104 struct btrfs_space_info
*space_info
;
8107 * Mixed block groups will exclude before processing the log so we only
8108 * need to do the exclude dance if this fs isn't mixed.
8110 if (!btrfs_fs_incompat(fs_info
, MIXED_GROUPS
)) {
8111 ret
= __exclude_logged_extent(fs_info
, ins
->objectid
,
8117 block_group
= btrfs_lookup_block_group(fs_info
, ins
->objectid
);
8121 space_info
= block_group
->space_info
;
8122 spin_lock(&space_info
->lock
);
8123 spin_lock(&block_group
->lock
);
8124 space_info
->bytes_reserved
+= ins
->offset
;
8125 block_group
->reserved
+= ins
->offset
;
8126 spin_unlock(&block_group
->lock
);
8127 spin_unlock(&space_info
->lock
);
8129 ret
= alloc_reserved_file_extent(trans
, 0, root_objectid
, 0, owner
,
8131 btrfs_put_block_group(block_group
);
8135 static struct extent_buffer
*
8136 btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
8137 u64 bytenr
, int level
, u64 owner
)
8139 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8140 struct extent_buffer
*buf
;
8142 buf
= btrfs_find_create_tree_block(fs_info
, bytenr
);
8146 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, buf
, level
);
8147 btrfs_tree_lock(buf
);
8148 clean_tree_block(fs_info
, buf
);
8149 clear_bit(EXTENT_BUFFER_STALE
, &buf
->bflags
);
8151 btrfs_set_lock_blocking(buf
);
8152 set_extent_buffer_uptodate(buf
);
8154 memzero_extent_buffer(buf
, 0, sizeof(struct btrfs_header
));
8155 btrfs_set_header_level(buf
, level
);
8156 btrfs_set_header_bytenr(buf
, buf
->start
);
8157 btrfs_set_header_generation(buf
, trans
->transid
);
8158 btrfs_set_header_backref_rev(buf
, BTRFS_MIXED_BACKREF_REV
);
8159 btrfs_set_header_owner(buf
, owner
);
8160 write_extent_buffer_fsid(buf
, fs_info
->fsid
);
8161 write_extent_buffer_chunk_tree_uuid(buf
, fs_info
->chunk_tree_uuid
);
8162 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
8163 buf
->log_index
= root
->log_transid
% 2;
8165 * we allow two log transactions at a time, use different
8166 * EXENT bit to differentiate dirty pages.
8168 if (buf
->log_index
== 0)
8169 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
8170 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
8172 set_extent_new(&root
->dirty_log_pages
, buf
->start
,
8173 buf
->start
+ buf
->len
- 1);
8175 buf
->log_index
= -1;
8176 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
8177 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
8179 trans
->dirty
= true;
8180 /* this returns a buffer locked for blocking */
8184 static struct btrfs_block_rsv
*
8185 use_block_rsv(struct btrfs_trans_handle
*trans
,
8186 struct btrfs_root
*root
, u32 blocksize
)
8188 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8189 struct btrfs_block_rsv
*block_rsv
;
8190 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
8192 bool global_updated
= false;
8194 block_rsv
= get_block_rsv(trans
, root
);
8196 if (unlikely(block_rsv
->size
== 0))
8199 ret
= block_rsv_use_bytes(block_rsv
, blocksize
);
8203 if (block_rsv
->failfast
)
8204 return ERR_PTR(ret
);
8206 if (block_rsv
->type
== BTRFS_BLOCK_RSV_GLOBAL
&& !global_updated
) {
8207 global_updated
= true;
8208 update_global_block_rsv(fs_info
);
8212 if (btrfs_test_opt(fs_info
, ENOSPC_DEBUG
)) {
8213 static DEFINE_RATELIMIT_STATE(_rs
,
8214 DEFAULT_RATELIMIT_INTERVAL
* 10,
8215 /*DEFAULT_RATELIMIT_BURST*/ 1);
8216 if (__ratelimit(&_rs
))
8218 "BTRFS: block rsv returned %d\n", ret
);
8221 ret
= reserve_metadata_bytes(root
, block_rsv
, blocksize
,
8222 BTRFS_RESERVE_NO_FLUSH
);
8226 * If we couldn't reserve metadata bytes try and use some from
8227 * the global reserve if its space type is the same as the global
8230 if (block_rsv
->type
!= BTRFS_BLOCK_RSV_GLOBAL
&&
8231 block_rsv
->space_info
== global_rsv
->space_info
) {
8232 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
8236 return ERR_PTR(ret
);
8239 static void unuse_block_rsv(struct btrfs_fs_info
*fs_info
,
8240 struct btrfs_block_rsv
*block_rsv
, u32 blocksize
)
8242 block_rsv_add_bytes(block_rsv
, blocksize
, 0);
8243 block_rsv_release_bytes(fs_info
, block_rsv
, NULL
, 0, NULL
);
8247 * finds a free extent and does all the dirty work required for allocation
8248 * returns the tree buffer or an ERR_PTR on error.
8250 struct extent_buffer
*btrfs_alloc_tree_block(struct btrfs_trans_handle
*trans
,
8251 struct btrfs_root
*root
,
8252 u64 parent
, u64 root_objectid
,
8253 const struct btrfs_disk_key
*key
,
8254 int level
, u64 hint
,
8257 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8258 struct btrfs_key ins
;
8259 struct btrfs_block_rsv
*block_rsv
;
8260 struct extent_buffer
*buf
;
8261 struct btrfs_delayed_extent_op
*extent_op
;
8264 u32 blocksize
= fs_info
->nodesize
;
8265 bool skinny_metadata
= btrfs_fs_incompat(fs_info
, SKINNY_METADATA
);
8267 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8268 if (btrfs_is_testing(fs_info
)) {
8269 buf
= btrfs_init_new_buffer(trans
, root
, root
->alloc_bytenr
,
8270 level
, root_objectid
);
8272 root
->alloc_bytenr
+= blocksize
;
8277 block_rsv
= use_block_rsv(trans
, root
, blocksize
);
8278 if (IS_ERR(block_rsv
))
8279 return ERR_CAST(block_rsv
);
8281 ret
= btrfs_reserve_extent(root
, blocksize
, blocksize
, blocksize
,
8282 empty_size
, hint
, &ins
, 0, 0);
8286 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
, level
,
8290 goto out_free_reserved
;
8293 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
8295 parent
= ins
.objectid
;
8296 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
8300 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
8301 extent_op
= btrfs_alloc_delayed_extent_op();
8307 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
8309 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
8310 extent_op
->flags_to_set
= flags
;
8311 extent_op
->update_key
= skinny_metadata
? false : true;
8312 extent_op
->update_flags
= true;
8313 extent_op
->is_data
= false;
8314 extent_op
->level
= level
;
8316 btrfs_ref_tree_mod(root
, ins
.objectid
, ins
.offset
, parent
,
8317 root_objectid
, level
, 0,
8318 BTRFS_ADD_DELAYED_EXTENT
);
8319 ret
= btrfs_add_delayed_tree_ref(trans
, ins
.objectid
,
8321 root_objectid
, level
,
8322 BTRFS_ADD_DELAYED_EXTENT
,
8323 extent_op
, NULL
, NULL
);
8325 goto out_free_delayed
;
8330 btrfs_free_delayed_extent_op(extent_op
);
8332 free_extent_buffer(buf
);
8334 btrfs_free_reserved_extent(fs_info
, ins
.objectid
, ins
.offset
, 0);
8336 unuse_block_rsv(fs_info
, block_rsv
, blocksize
);
8337 return ERR_PTR(ret
);
8340 struct walk_control
{
8341 u64 refs
[BTRFS_MAX_LEVEL
];
8342 u64 flags
[BTRFS_MAX_LEVEL
];
8343 struct btrfs_key update_progress
;
8354 #define DROP_REFERENCE 1
8355 #define UPDATE_BACKREF 2
8357 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
8358 struct btrfs_root
*root
,
8359 struct walk_control
*wc
,
8360 struct btrfs_path
*path
)
8362 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8368 struct btrfs_key key
;
8369 struct extent_buffer
*eb
;
8374 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
8375 wc
->reada_count
= wc
->reada_count
* 2 / 3;
8376 wc
->reada_count
= max(wc
->reada_count
, 2);
8378 wc
->reada_count
= wc
->reada_count
* 3 / 2;
8379 wc
->reada_count
= min_t(int, wc
->reada_count
,
8380 BTRFS_NODEPTRS_PER_BLOCK(fs_info
));
8383 eb
= path
->nodes
[wc
->level
];
8384 nritems
= btrfs_header_nritems(eb
);
8386 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
8387 if (nread
>= wc
->reada_count
)
8391 bytenr
= btrfs_node_blockptr(eb
, slot
);
8392 generation
= btrfs_node_ptr_generation(eb
, slot
);
8394 if (slot
== path
->slots
[wc
->level
])
8397 if (wc
->stage
== UPDATE_BACKREF
&&
8398 generation
<= root
->root_key
.offset
)
8401 /* We don't lock the tree block, it's OK to be racy here */
8402 ret
= btrfs_lookup_extent_info(trans
, fs_info
, bytenr
,
8403 wc
->level
- 1, 1, &refs
,
8405 /* We don't care about errors in readahead. */
8410 if (wc
->stage
== DROP_REFERENCE
) {
8414 if (wc
->level
== 1 &&
8415 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8417 if (!wc
->update_ref
||
8418 generation
<= root
->root_key
.offset
)
8420 btrfs_node_key_to_cpu(eb
, &key
, slot
);
8421 ret
= btrfs_comp_cpu_keys(&key
,
8422 &wc
->update_progress
);
8426 if (wc
->level
== 1 &&
8427 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8431 readahead_tree_block(fs_info
, bytenr
);
8434 wc
->reada_slot
= slot
;
8438 * helper to process tree block while walking down the tree.
8440 * when wc->stage == UPDATE_BACKREF, this function updates
8441 * back refs for pointers in the block.
8443 * NOTE: return value 1 means we should stop walking down.
8445 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
8446 struct btrfs_root
*root
,
8447 struct btrfs_path
*path
,
8448 struct walk_control
*wc
, int lookup_info
)
8450 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8451 int level
= wc
->level
;
8452 struct extent_buffer
*eb
= path
->nodes
[level
];
8453 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
8456 if (wc
->stage
== UPDATE_BACKREF
&&
8457 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
8461 * when reference count of tree block is 1, it won't increase
8462 * again. once full backref flag is set, we never clear it.
8465 ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
8466 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
)))) {
8467 BUG_ON(!path
->locks
[level
]);
8468 ret
= btrfs_lookup_extent_info(trans
, fs_info
,
8469 eb
->start
, level
, 1,
8472 BUG_ON(ret
== -ENOMEM
);
8475 BUG_ON(wc
->refs
[level
] == 0);
8478 if (wc
->stage
== DROP_REFERENCE
) {
8479 if (wc
->refs
[level
] > 1)
8482 if (path
->locks
[level
] && !wc
->keep_locks
) {
8483 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8484 path
->locks
[level
] = 0;
8489 /* wc->stage == UPDATE_BACKREF */
8490 if (!(wc
->flags
[level
] & flag
)) {
8491 BUG_ON(!path
->locks
[level
]);
8492 ret
= btrfs_inc_ref(trans
, root
, eb
, 1);
8493 BUG_ON(ret
); /* -ENOMEM */
8494 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
8495 BUG_ON(ret
); /* -ENOMEM */
8496 ret
= btrfs_set_disk_extent_flags(trans
, fs_info
, eb
->start
,
8498 btrfs_header_level(eb
), 0);
8499 BUG_ON(ret
); /* -ENOMEM */
8500 wc
->flags
[level
] |= flag
;
8504 * the block is shared by multiple trees, so it's not good to
8505 * keep the tree lock
8507 if (path
->locks
[level
] && level
> 0) {
8508 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8509 path
->locks
[level
] = 0;
8515 * helper to process tree block pointer.
8517 * when wc->stage == DROP_REFERENCE, this function checks
8518 * reference count of the block pointed to. if the block
8519 * is shared and we need update back refs for the subtree
8520 * rooted at the block, this function changes wc->stage to
8521 * UPDATE_BACKREF. if the block is shared and there is no
8522 * need to update back, this function drops the reference
8525 * NOTE: return value 1 means we should stop walking down.
8527 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
8528 struct btrfs_root
*root
,
8529 struct btrfs_path
*path
,
8530 struct walk_control
*wc
, int *lookup_info
)
8532 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8537 struct btrfs_key key
;
8538 struct btrfs_key first_key
;
8539 struct extent_buffer
*next
;
8540 int level
= wc
->level
;
8543 bool need_account
= false;
8545 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
8546 path
->slots
[level
]);
8548 * if the lower level block was created before the snapshot
8549 * was created, we know there is no need to update back refs
8552 if (wc
->stage
== UPDATE_BACKREF
&&
8553 generation
<= root
->root_key
.offset
) {
8558 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
8559 btrfs_node_key_to_cpu(path
->nodes
[level
], &first_key
,
8560 path
->slots
[level
]);
8561 blocksize
= fs_info
->nodesize
;
8563 next
= find_extent_buffer(fs_info
, bytenr
);
8565 next
= btrfs_find_create_tree_block(fs_info
, bytenr
);
8567 return PTR_ERR(next
);
8569 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, next
,
8573 btrfs_tree_lock(next
);
8574 btrfs_set_lock_blocking(next
);
8576 ret
= btrfs_lookup_extent_info(trans
, fs_info
, bytenr
, level
- 1, 1,
8577 &wc
->refs
[level
- 1],
8578 &wc
->flags
[level
- 1]);
8582 if (unlikely(wc
->refs
[level
- 1] == 0)) {
8583 btrfs_err(fs_info
, "Missing references.");
8589 if (wc
->stage
== DROP_REFERENCE
) {
8590 if (wc
->refs
[level
- 1] > 1) {
8591 need_account
= true;
8593 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8596 if (!wc
->update_ref
||
8597 generation
<= root
->root_key
.offset
)
8600 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
8601 path
->slots
[level
]);
8602 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
8606 wc
->stage
= UPDATE_BACKREF
;
8607 wc
->shared_level
= level
- 1;
8611 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8615 if (!btrfs_buffer_uptodate(next
, generation
, 0)) {
8616 btrfs_tree_unlock(next
);
8617 free_extent_buffer(next
);
8623 if (reada
&& level
== 1)
8624 reada_walk_down(trans
, root
, wc
, path
);
8625 next
= read_tree_block(fs_info
, bytenr
, generation
, level
- 1,
8628 return PTR_ERR(next
);
8629 } else if (!extent_buffer_uptodate(next
)) {
8630 free_extent_buffer(next
);
8633 btrfs_tree_lock(next
);
8634 btrfs_set_lock_blocking(next
);
8638 ASSERT(level
== btrfs_header_level(next
));
8639 if (level
!= btrfs_header_level(next
)) {
8640 btrfs_err(root
->fs_info
, "mismatched level");
8644 path
->nodes
[level
] = next
;
8645 path
->slots
[level
] = 0;
8646 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8652 wc
->refs
[level
- 1] = 0;
8653 wc
->flags
[level
- 1] = 0;
8654 if (wc
->stage
== DROP_REFERENCE
) {
8655 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
8656 parent
= path
->nodes
[level
]->start
;
8658 ASSERT(root
->root_key
.objectid
==
8659 btrfs_header_owner(path
->nodes
[level
]));
8660 if (root
->root_key
.objectid
!=
8661 btrfs_header_owner(path
->nodes
[level
])) {
8662 btrfs_err(root
->fs_info
,
8663 "mismatched block owner");
8671 ret
= btrfs_qgroup_trace_subtree(trans
, root
, next
,
8672 generation
, level
- 1);
8674 btrfs_err_rl(fs_info
,
8675 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
8679 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
,
8680 parent
, root
->root_key
.objectid
,
8690 btrfs_tree_unlock(next
);
8691 free_extent_buffer(next
);
8697 * helper to process tree block while walking up the tree.
8699 * when wc->stage == DROP_REFERENCE, this function drops
8700 * reference count on the block.
8702 * when wc->stage == UPDATE_BACKREF, this function changes
8703 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8704 * to UPDATE_BACKREF previously while processing the block.
8706 * NOTE: return value 1 means we should stop walking up.
8708 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
8709 struct btrfs_root
*root
,
8710 struct btrfs_path
*path
,
8711 struct walk_control
*wc
)
8713 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8715 int level
= wc
->level
;
8716 struct extent_buffer
*eb
= path
->nodes
[level
];
8719 if (wc
->stage
== UPDATE_BACKREF
) {
8720 BUG_ON(wc
->shared_level
< level
);
8721 if (level
< wc
->shared_level
)
8724 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
8728 wc
->stage
= DROP_REFERENCE
;
8729 wc
->shared_level
= -1;
8730 path
->slots
[level
] = 0;
8733 * check reference count again if the block isn't locked.
8734 * we should start walking down the tree again if reference
8737 if (!path
->locks
[level
]) {
8739 btrfs_tree_lock(eb
);
8740 btrfs_set_lock_blocking(eb
);
8741 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8743 ret
= btrfs_lookup_extent_info(trans
, fs_info
,
8744 eb
->start
, level
, 1,
8748 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8749 path
->locks
[level
] = 0;
8752 BUG_ON(wc
->refs
[level
] == 0);
8753 if (wc
->refs
[level
] == 1) {
8754 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8755 path
->locks
[level
] = 0;
8761 /* wc->stage == DROP_REFERENCE */
8762 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
8764 if (wc
->refs
[level
] == 1) {
8766 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8767 ret
= btrfs_dec_ref(trans
, root
, eb
, 1);
8769 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
8770 BUG_ON(ret
); /* -ENOMEM */
8771 ret
= btrfs_qgroup_trace_leaf_items(trans
, fs_info
, eb
);
8773 btrfs_err_rl(fs_info
,
8774 "error %d accounting leaf items. Quota is out of sync, rescan required.",
8778 /* make block locked assertion in clean_tree_block happy */
8779 if (!path
->locks
[level
] &&
8780 btrfs_header_generation(eb
) == trans
->transid
) {
8781 btrfs_tree_lock(eb
);
8782 btrfs_set_lock_blocking(eb
);
8783 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8785 clean_tree_block(fs_info
, eb
);
8788 if (eb
== root
->node
) {
8789 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8792 BUG_ON(root
->root_key
.objectid
!=
8793 btrfs_header_owner(eb
));
8795 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8796 parent
= path
->nodes
[level
+ 1]->start
;
8798 BUG_ON(root
->root_key
.objectid
!=
8799 btrfs_header_owner(path
->nodes
[level
+ 1]));
8802 btrfs_free_tree_block(trans
, root
, eb
, parent
, wc
->refs
[level
] == 1);
8804 wc
->refs
[level
] = 0;
8805 wc
->flags
[level
] = 0;
8809 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
8810 struct btrfs_root
*root
,
8811 struct btrfs_path
*path
,
8812 struct walk_control
*wc
)
8814 int level
= wc
->level
;
8815 int lookup_info
= 1;
8818 while (level
>= 0) {
8819 ret
= walk_down_proc(trans
, root
, path
, wc
, lookup_info
);
8826 if (path
->slots
[level
] >=
8827 btrfs_header_nritems(path
->nodes
[level
]))
8830 ret
= do_walk_down(trans
, root
, path
, wc
, &lookup_info
);
8832 path
->slots
[level
]++;
8841 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
8842 struct btrfs_root
*root
,
8843 struct btrfs_path
*path
,
8844 struct walk_control
*wc
, int max_level
)
8846 int level
= wc
->level
;
8849 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
8850 while (level
< max_level
&& path
->nodes
[level
]) {
8852 if (path
->slots
[level
] + 1 <
8853 btrfs_header_nritems(path
->nodes
[level
])) {
8854 path
->slots
[level
]++;
8857 ret
= walk_up_proc(trans
, root
, path
, wc
);
8861 if (path
->locks
[level
]) {
8862 btrfs_tree_unlock_rw(path
->nodes
[level
],
8863 path
->locks
[level
]);
8864 path
->locks
[level
] = 0;
8866 free_extent_buffer(path
->nodes
[level
]);
8867 path
->nodes
[level
] = NULL
;
8875 * drop a subvolume tree.
8877 * this function traverses the tree freeing any blocks that only
8878 * referenced by the tree.
8880 * when a shared tree block is found. this function decreases its
8881 * reference count by one. if update_ref is true, this function
8882 * also make sure backrefs for the shared block and all lower level
8883 * blocks are properly updated.
8885 * If called with for_reloc == 0, may exit early with -EAGAIN
8887 int btrfs_drop_snapshot(struct btrfs_root
*root
,
8888 struct btrfs_block_rsv
*block_rsv
, int update_ref
,
8891 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8892 struct btrfs_path
*path
;
8893 struct btrfs_trans_handle
*trans
;
8894 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
8895 struct btrfs_root_item
*root_item
= &root
->root_item
;
8896 struct walk_control
*wc
;
8897 struct btrfs_key key
;
8901 bool root_dropped
= false;
8903 btrfs_debug(fs_info
, "Drop subvolume %llu", root
->objectid
);
8905 path
= btrfs_alloc_path();
8911 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
8913 btrfs_free_path(path
);
8918 trans
= btrfs_start_transaction(tree_root
, 0);
8919 if (IS_ERR(trans
)) {
8920 err
= PTR_ERR(trans
);
8925 trans
->block_rsv
= block_rsv
;
8927 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
8928 level
= btrfs_header_level(root
->node
);
8929 path
->nodes
[level
] = btrfs_lock_root_node(root
);
8930 btrfs_set_lock_blocking(path
->nodes
[level
]);
8931 path
->slots
[level
] = 0;
8932 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8933 memset(&wc
->update_progress
, 0,
8934 sizeof(wc
->update_progress
));
8936 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
8937 memcpy(&wc
->update_progress
, &key
,
8938 sizeof(wc
->update_progress
));
8940 level
= root_item
->drop_level
;
8942 path
->lowest_level
= level
;
8943 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
8944 path
->lowest_level
= 0;
8952 * unlock our path, this is safe because only this
8953 * function is allowed to delete this snapshot
8955 btrfs_unlock_up_safe(path
, 0);
8957 level
= btrfs_header_level(root
->node
);
8959 btrfs_tree_lock(path
->nodes
[level
]);
8960 btrfs_set_lock_blocking(path
->nodes
[level
]);
8961 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8963 ret
= btrfs_lookup_extent_info(trans
, fs_info
,
8964 path
->nodes
[level
]->start
,
8965 level
, 1, &wc
->refs
[level
],
8971 BUG_ON(wc
->refs
[level
] == 0);
8973 if (level
== root_item
->drop_level
)
8976 btrfs_tree_unlock(path
->nodes
[level
]);
8977 path
->locks
[level
] = 0;
8978 WARN_ON(wc
->refs
[level
] != 1);
8984 wc
->shared_level
= -1;
8985 wc
->stage
= DROP_REFERENCE
;
8986 wc
->update_ref
= update_ref
;
8988 wc
->for_reloc
= for_reloc
;
8989 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(fs_info
);
8993 ret
= walk_down_tree(trans
, root
, path
, wc
);
8999 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
9006 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
9010 if (wc
->stage
== DROP_REFERENCE
) {
9012 btrfs_node_key(path
->nodes
[level
],
9013 &root_item
->drop_progress
,
9014 path
->slots
[level
]);
9015 root_item
->drop_level
= level
;
9018 BUG_ON(wc
->level
== 0);
9019 if (btrfs_should_end_transaction(trans
) ||
9020 (!for_reloc
&& btrfs_need_cleaner_sleep(fs_info
))) {
9021 ret
= btrfs_update_root(trans
, tree_root
,
9025 btrfs_abort_transaction(trans
, ret
);
9030 btrfs_end_transaction_throttle(trans
);
9031 if (!for_reloc
&& btrfs_need_cleaner_sleep(fs_info
)) {
9032 btrfs_debug(fs_info
,
9033 "drop snapshot early exit");
9038 trans
= btrfs_start_transaction(tree_root
, 0);
9039 if (IS_ERR(trans
)) {
9040 err
= PTR_ERR(trans
);
9044 trans
->block_rsv
= block_rsv
;
9047 btrfs_release_path(path
);
9051 ret
= btrfs_del_root(trans
, fs_info
, &root
->root_key
);
9053 btrfs_abort_transaction(trans
, ret
);
9058 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
9059 ret
= btrfs_find_root(tree_root
, &root
->root_key
, path
,
9062 btrfs_abort_transaction(trans
, ret
);
9065 } else if (ret
> 0) {
9066 /* if we fail to delete the orphan item this time
9067 * around, it'll get picked up the next time.
9069 * The most common failure here is just -ENOENT.
9071 btrfs_del_orphan_item(trans
, tree_root
,
9072 root
->root_key
.objectid
);
9076 if (test_bit(BTRFS_ROOT_IN_RADIX
, &root
->state
)) {
9077 btrfs_add_dropped_root(trans
, root
);
9079 free_extent_buffer(root
->node
);
9080 free_extent_buffer(root
->commit_root
);
9081 btrfs_put_fs_root(root
);
9083 root_dropped
= true;
9085 btrfs_end_transaction_throttle(trans
);
9088 btrfs_free_path(path
);
9091 * So if we need to stop dropping the snapshot for whatever reason we
9092 * need to make sure to add it back to the dead root list so that we
9093 * keep trying to do the work later. This also cleans up roots if we
9094 * don't have it in the radix (like when we recover after a power fail
9095 * or unmount) so we don't leak memory.
9097 if (!for_reloc
&& !root_dropped
)
9098 btrfs_add_dead_root(root
);
9099 if (err
&& err
!= -EAGAIN
)
9100 btrfs_handle_fs_error(fs_info
, err
, NULL
);
9105 * drop subtree rooted at tree block 'node'.
9107 * NOTE: this function will unlock and release tree block 'node'
9108 * only used by relocation code
9110 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
9111 struct btrfs_root
*root
,
9112 struct extent_buffer
*node
,
9113 struct extent_buffer
*parent
)
9115 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
9116 struct btrfs_path
*path
;
9117 struct walk_control
*wc
;
9123 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
9125 path
= btrfs_alloc_path();
9129 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
9131 btrfs_free_path(path
);
9135 btrfs_assert_tree_locked(parent
);
9136 parent_level
= btrfs_header_level(parent
);
9137 extent_buffer_get(parent
);
9138 path
->nodes
[parent_level
] = parent
;
9139 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
9141 btrfs_assert_tree_locked(node
);
9142 level
= btrfs_header_level(node
);
9143 path
->nodes
[level
] = node
;
9144 path
->slots
[level
] = 0;
9145 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
9147 wc
->refs
[parent_level
] = 1;
9148 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
9150 wc
->shared_level
= -1;
9151 wc
->stage
= DROP_REFERENCE
;
9155 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(fs_info
);
9158 wret
= walk_down_tree(trans
, root
, path
, wc
);
9164 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
9172 btrfs_free_path(path
);
9176 static u64
update_block_group_flags(struct btrfs_fs_info
*fs_info
, u64 flags
)
9182 * if restripe for this chunk_type is on pick target profile and
9183 * return, otherwise do the usual balance
9185 stripped
= get_restripe_target(fs_info
, flags
);
9187 return extended_to_chunk(stripped
);
9189 num_devices
= fs_info
->fs_devices
->rw_devices
;
9191 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
9192 BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
|
9193 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
9195 if (num_devices
== 1) {
9196 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
9197 stripped
= flags
& ~stripped
;
9199 /* turn raid0 into single device chunks */
9200 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
9203 /* turn mirroring into duplication */
9204 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
9205 BTRFS_BLOCK_GROUP_RAID10
))
9206 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
9208 /* they already had raid on here, just return */
9209 if (flags
& stripped
)
9212 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
9213 stripped
= flags
& ~stripped
;
9215 /* switch duplicated blocks with raid1 */
9216 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
9217 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
9219 /* this is drive concat, leave it alone */
9225 static int inc_block_group_ro(struct btrfs_block_group_cache
*cache
, int force
)
9227 struct btrfs_space_info
*sinfo
= cache
->space_info
;
9229 u64 min_allocable_bytes
;
9233 * We need some metadata space and system metadata space for
9234 * allocating chunks in some corner cases until we force to set
9235 * it to be readonly.
9238 (BTRFS_BLOCK_GROUP_SYSTEM
| BTRFS_BLOCK_GROUP_METADATA
)) &&
9240 min_allocable_bytes
= SZ_1M
;
9242 min_allocable_bytes
= 0;
9244 spin_lock(&sinfo
->lock
);
9245 spin_lock(&cache
->lock
);
9253 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
9254 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
9256 if (btrfs_space_info_used(sinfo
, true) + num_bytes
+
9257 min_allocable_bytes
<= sinfo
->total_bytes
) {
9258 sinfo
->bytes_readonly
+= num_bytes
;
9260 list_add_tail(&cache
->ro_list
, &sinfo
->ro_bgs
);
9264 spin_unlock(&cache
->lock
);
9265 spin_unlock(&sinfo
->lock
);
9269 int btrfs_inc_block_group_ro(struct btrfs_block_group_cache
*cache
)
9272 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
9273 struct btrfs_trans_handle
*trans
;
9278 trans
= btrfs_join_transaction(fs_info
->extent_root
);
9280 return PTR_ERR(trans
);
9283 * we're not allowed to set block groups readonly after the dirty
9284 * block groups cache has started writing. If it already started,
9285 * back off and let this transaction commit
9287 mutex_lock(&fs_info
->ro_block_group_mutex
);
9288 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN
, &trans
->transaction
->flags
)) {
9289 u64 transid
= trans
->transid
;
9291 mutex_unlock(&fs_info
->ro_block_group_mutex
);
9292 btrfs_end_transaction(trans
);
9294 ret
= btrfs_wait_for_commit(fs_info
, transid
);
9301 * if we are changing raid levels, try to allocate a corresponding
9302 * block group with the new raid level.
9304 alloc_flags
= update_block_group_flags(fs_info
, cache
->flags
);
9305 if (alloc_flags
!= cache
->flags
) {
9306 ret
= do_chunk_alloc(trans
, alloc_flags
,
9309 * ENOSPC is allowed here, we may have enough space
9310 * already allocated at the new raid level to
9319 ret
= inc_block_group_ro(cache
, 0);
9322 alloc_flags
= get_alloc_profile(fs_info
, cache
->space_info
->flags
);
9323 ret
= do_chunk_alloc(trans
, alloc_flags
, CHUNK_ALLOC_FORCE
);
9326 ret
= inc_block_group_ro(cache
, 0);
9328 if (cache
->flags
& BTRFS_BLOCK_GROUP_SYSTEM
) {
9329 alloc_flags
= update_block_group_flags(fs_info
, cache
->flags
);
9330 mutex_lock(&fs_info
->chunk_mutex
);
9331 check_system_chunk(trans
, alloc_flags
);
9332 mutex_unlock(&fs_info
->chunk_mutex
);
9334 mutex_unlock(&fs_info
->ro_block_group_mutex
);
9336 btrfs_end_transaction(trans
);
9340 int btrfs_force_chunk_alloc(struct btrfs_trans_handle
*trans
, u64 type
)
9342 u64 alloc_flags
= get_alloc_profile(trans
->fs_info
, type
);
9344 return do_chunk_alloc(trans
, alloc_flags
, CHUNK_ALLOC_FORCE
);
9348 * helper to account the unused space of all the readonly block group in the
9349 * space_info. takes mirrors into account.
9351 u64
btrfs_account_ro_block_groups_free_space(struct btrfs_space_info
*sinfo
)
9353 struct btrfs_block_group_cache
*block_group
;
9357 /* It's df, we don't care if it's racy */
9358 if (list_empty(&sinfo
->ro_bgs
))
9361 spin_lock(&sinfo
->lock
);
9362 list_for_each_entry(block_group
, &sinfo
->ro_bgs
, ro_list
) {
9363 spin_lock(&block_group
->lock
);
9365 if (!block_group
->ro
) {
9366 spin_unlock(&block_group
->lock
);
9370 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_RAID1
|
9371 BTRFS_BLOCK_GROUP_RAID10
|
9372 BTRFS_BLOCK_GROUP_DUP
))
9377 free_bytes
+= (block_group
->key
.offset
-
9378 btrfs_block_group_used(&block_group
->item
)) *
9381 spin_unlock(&block_group
->lock
);
9383 spin_unlock(&sinfo
->lock
);
9388 void btrfs_dec_block_group_ro(struct btrfs_block_group_cache
*cache
)
9390 struct btrfs_space_info
*sinfo
= cache
->space_info
;
9395 spin_lock(&sinfo
->lock
);
9396 spin_lock(&cache
->lock
);
9398 num_bytes
= cache
->key
.offset
- cache
->reserved
-
9399 cache
->pinned
- cache
->bytes_super
-
9400 btrfs_block_group_used(&cache
->item
);
9401 sinfo
->bytes_readonly
-= num_bytes
;
9402 list_del_init(&cache
->ro_list
);
9404 spin_unlock(&cache
->lock
);
9405 spin_unlock(&sinfo
->lock
);
9409 * checks to see if its even possible to relocate this block group.
9411 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9412 * ok to go ahead and try.
9414 int btrfs_can_relocate(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
9416 struct btrfs_root
*root
= fs_info
->extent_root
;
9417 struct btrfs_block_group_cache
*block_group
;
9418 struct btrfs_space_info
*space_info
;
9419 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
9420 struct btrfs_device
*device
;
9421 struct btrfs_trans_handle
*trans
;
9431 debug
= btrfs_test_opt(fs_info
, ENOSPC_DEBUG
);
9433 block_group
= btrfs_lookup_block_group(fs_info
, bytenr
);
9435 /* odd, couldn't find the block group, leave it alone */
9439 "can't find block group for bytenr %llu",
9444 min_free
= btrfs_block_group_used(&block_group
->item
);
9446 /* no bytes used, we're good */
9450 space_info
= block_group
->space_info
;
9451 spin_lock(&space_info
->lock
);
9453 full
= space_info
->full
;
9456 * if this is the last block group we have in this space, we can't
9457 * relocate it unless we're able to allocate a new chunk below.
9459 * Otherwise, we need to make sure we have room in the space to handle
9460 * all of the extents from this block group. If we can, we're good
9462 if ((space_info
->total_bytes
!= block_group
->key
.offset
) &&
9463 (btrfs_space_info_used(space_info
, false) + min_free
<
9464 space_info
->total_bytes
)) {
9465 spin_unlock(&space_info
->lock
);
9468 spin_unlock(&space_info
->lock
);
9471 * ok we don't have enough space, but maybe we have free space on our
9472 * devices to allocate new chunks for relocation, so loop through our
9473 * alloc devices and guess if we have enough space. if this block
9474 * group is going to be restriped, run checks against the target
9475 * profile instead of the current one.
9487 target
= get_restripe_target(fs_info
, block_group
->flags
);
9489 index
= btrfs_bg_flags_to_raid_index(extended_to_chunk(target
));
9492 * this is just a balance, so if we were marked as full
9493 * we know there is no space for a new chunk
9498 "no space to alloc new chunk for block group %llu",
9499 block_group
->key
.objectid
);
9503 index
= btrfs_bg_flags_to_raid_index(block_group
->flags
);
9506 if (index
== BTRFS_RAID_RAID10
) {
9510 } else if (index
== BTRFS_RAID_RAID1
) {
9512 } else if (index
== BTRFS_RAID_DUP
) {
9515 } else if (index
== BTRFS_RAID_RAID0
) {
9516 dev_min
= fs_devices
->rw_devices
;
9517 min_free
= div64_u64(min_free
, dev_min
);
9520 /* We need to do this so that we can look at pending chunks */
9521 trans
= btrfs_join_transaction(root
);
9522 if (IS_ERR(trans
)) {
9523 ret
= PTR_ERR(trans
);
9527 mutex_lock(&fs_info
->chunk_mutex
);
9528 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
9532 * check to make sure we can actually find a chunk with enough
9533 * space to fit our block group in.
9535 if (device
->total_bytes
> device
->bytes_used
+ min_free
&&
9536 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT
, &device
->dev_state
)) {
9537 ret
= find_free_dev_extent(trans
, device
, min_free
,
9542 if (dev_nr
>= dev_min
)
9548 if (debug
&& ret
== -1)
9550 "no space to allocate a new chunk for block group %llu",
9551 block_group
->key
.objectid
);
9552 mutex_unlock(&fs_info
->chunk_mutex
);
9553 btrfs_end_transaction(trans
);
9555 btrfs_put_block_group(block_group
);
9559 static int find_first_block_group(struct btrfs_fs_info
*fs_info
,
9560 struct btrfs_path
*path
,
9561 struct btrfs_key
*key
)
9563 struct btrfs_root
*root
= fs_info
->extent_root
;
9565 struct btrfs_key found_key
;
9566 struct extent_buffer
*leaf
;
9569 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
9574 slot
= path
->slots
[0];
9575 leaf
= path
->nodes
[0];
9576 if (slot
>= btrfs_header_nritems(leaf
)) {
9577 ret
= btrfs_next_leaf(root
, path
);
9584 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
9586 if (found_key
.objectid
>= key
->objectid
&&
9587 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
9588 struct extent_map_tree
*em_tree
;
9589 struct extent_map
*em
;
9591 em_tree
= &root
->fs_info
->mapping_tree
.map_tree
;
9592 read_lock(&em_tree
->lock
);
9593 em
= lookup_extent_mapping(em_tree
, found_key
.objectid
,
9595 read_unlock(&em_tree
->lock
);
9598 "logical %llu len %llu found bg but no related chunk",
9599 found_key
.objectid
, found_key
.offset
);
9604 free_extent_map(em
);
9613 void btrfs_put_block_group_cache(struct btrfs_fs_info
*info
)
9615 struct btrfs_block_group_cache
*block_group
;
9619 struct inode
*inode
;
9621 block_group
= btrfs_lookup_first_block_group(info
, last
);
9622 while (block_group
) {
9623 spin_lock(&block_group
->lock
);
9624 if (block_group
->iref
)
9626 spin_unlock(&block_group
->lock
);
9627 block_group
= next_block_group(info
, block_group
);
9636 inode
= block_group
->inode
;
9637 block_group
->iref
= 0;
9638 block_group
->inode
= NULL
;
9639 spin_unlock(&block_group
->lock
);
9640 ASSERT(block_group
->io_ctl
.inode
== NULL
);
9642 last
= block_group
->key
.objectid
+ block_group
->key
.offset
;
9643 btrfs_put_block_group(block_group
);
9648 * Must be called only after stopping all workers, since we could have block
9649 * group caching kthreads running, and therefore they could race with us if we
9650 * freed the block groups before stopping them.
9652 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
9654 struct btrfs_block_group_cache
*block_group
;
9655 struct btrfs_space_info
*space_info
;
9656 struct btrfs_caching_control
*caching_ctl
;
9659 down_write(&info
->commit_root_sem
);
9660 while (!list_empty(&info
->caching_block_groups
)) {
9661 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
9662 struct btrfs_caching_control
, list
);
9663 list_del(&caching_ctl
->list
);
9664 put_caching_control(caching_ctl
);
9666 up_write(&info
->commit_root_sem
);
9668 spin_lock(&info
->unused_bgs_lock
);
9669 while (!list_empty(&info
->unused_bgs
)) {
9670 block_group
= list_first_entry(&info
->unused_bgs
,
9671 struct btrfs_block_group_cache
,
9673 list_del_init(&block_group
->bg_list
);
9674 btrfs_put_block_group(block_group
);
9676 spin_unlock(&info
->unused_bgs_lock
);
9678 spin_lock(&info
->block_group_cache_lock
);
9679 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
9680 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
9682 rb_erase(&block_group
->cache_node
,
9683 &info
->block_group_cache_tree
);
9684 RB_CLEAR_NODE(&block_group
->cache_node
);
9685 spin_unlock(&info
->block_group_cache_lock
);
9687 down_write(&block_group
->space_info
->groups_sem
);
9688 list_del(&block_group
->list
);
9689 up_write(&block_group
->space_info
->groups_sem
);
9692 * We haven't cached this block group, which means we could
9693 * possibly have excluded extents on this block group.
9695 if (block_group
->cached
== BTRFS_CACHE_NO
||
9696 block_group
->cached
== BTRFS_CACHE_ERROR
)
9697 free_excluded_extents(block_group
);
9699 btrfs_remove_free_space_cache(block_group
);
9700 ASSERT(block_group
->cached
!= BTRFS_CACHE_STARTED
);
9701 ASSERT(list_empty(&block_group
->dirty_list
));
9702 ASSERT(list_empty(&block_group
->io_list
));
9703 ASSERT(list_empty(&block_group
->bg_list
));
9704 ASSERT(atomic_read(&block_group
->count
) == 1);
9705 btrfs_put_block_group(block_group
);
9707 spin_lock(&info
->block_group_cache_lock
);
9709 spin_unlock(&info
->block_group_cache_lock
);
9711 /* now that all the block groups are freed, go through and
9712 * free all the space_info structs. This is only called during
9713 * the final stages of unmount, and so we know nobody is
9714 * using them. We call synchronize_rcu() once before we start,
9715 * just to be on the safe side.
9719 release_global_block_rsv(info
);
9721 while (!list_empty(&info
->space_info
)) {
9724 space_info
= list_entry(info
->space_info
.next
,
9725 struct btrfs_space_info
,
9729 * Do not hide this behind enospc_debug, this is actually
9730 * important and indicates a real bug if this happens.
9732 if (WARN_ON(space_info
->bytes_pinned
> 0 ||
9733 space_info
->bytes_reserved
> 0 ||
9734 space_info
->bytes_may_use
> 0))
9735 dump_space_info(info
, space_info
, 0, 0);
9736 list_del(&space_info
->list
);
9737 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++) {
9738 struct kobject
*kobj
;
9739 kobj
= space_info
->block_group_kobjs
[i
];
9740 space_info
->block_group_kobjs
[i
] = NULL
;
9746 kobject_del(&space_info
->kobj
);
9747 kobject_put(&space_info
->kobj
);
9752 /* link_block_group will queue up kobjects to add when we're reclaim-safe */
9753 void btrfs_add_raid_kobjects(struct btrfs_fs_info
*fs_info
)
9755 struct btrfs_space_info
*space_info
;
9756 struct raid_kobject
*rkobj
;
9761 spin_lock(&fs_info
->pending_raid_kobjs_lock
);
9762 list_splice_init(&fs_info
->pending_raid_kobjs
, &list
);
9763 spin_unlock(&fs_info
->pending_raid_kobjs_lock
);
9765 list_for_each_entry(rkobj
, &list
, list
) {
9766 space_info
= __find_space_info(fs_info
, rkobj
->flags
);
9767 index
= btrfs_bg_flags_to_raid_index(rkobj
->flags
);
9769 ret
= kobject_add(&rkobj
->kobj
, &space_info
->kobj
,
9770 "%s", get_raid_name(index
));
9772 kobject_put(&rkobj
->kobj
);
9778 "failed to add kobject for block cache, ignoring");
9781 static void link_block_group(struct btrfs_block_group_cache
*cache
)
9783 struct btrfs_space_info
*space_info
= cache
->space_info
;
9784 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
9785 int index
= btrfs_bg_flags_to_raid_index(cache
->flags
);
9788 down_write(&space_info
->groups_sem
);
9789 if (list_empty(&space_info
->block_groups
[index
]))
9791 list_add_tail(&cache
->list
, &space_info
->block_groups
[index
]);
9792 up_write(&space_info
->groups_sem
);
9795 struct raid_kobject
*rkobj
= kzalloc(sizeof(*rkobj
), GFP_NOFS
);
9797 btrfs_warn(cache
->fs_info
,
9798 "couldn't alloc memory for raid level kobject");
9801 rkobj
->flags
= cache
->flags
;
9802 kobject_init(&rkobj
->kobj
, &btrfs_raid_ktype
);
9804 spin_lock(&fs_info
->pending_raid_kobjs_lock
);
9805 list_add_tail(&rkobj
->list
, &fs_info
->pending_raid_kobjs
);
9806 spin_unlock(&fs_info
->pending_raid_kobjs_lock
);
9807 space_info
->block_group_kobjs
[index
] = &rkobj
->kobj
;
9811 static struct btrfs_block_group_cache
*
9812 btrfs_create_block_group_cache(struct btrfs_fs_info
*fs_info
,
9813 u64 start
, u64 size
)
9815 struct btrfs_block_group_cache
*cache
;
9817 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
9821 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
9823 if (!cache
->free_space_ctl
) {
9828 cache
->key
.objectid
= start
;
9829 cache
->key
.offset
= size
;
9830 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
9832 cache
->fs_info
= fs_info
;
9833 cache
->full_stripe_len
= btrfs_full_stripe_len(fs_info
, start
);
9834 set_free_space_tree_thresholds(cache
);
9836 atomic_set(&cache
->count
, 1);
9837 spin_lock_init(&cache
->lock
);
9838 init_rwsem(&cache
->data_rwsem
);
9839 INIT_LIST_HEAD(&cache
->list
);
9840 INIT_LIST_HEAD(&cache
->cluster_list
);
9841 INIT_LIST_HEAD(&cache
->bg_list
);
9842 INIT_LIST_HEAD(&cache
->ro_list
);
9843 INIT_LIST_HEAD(&cache
->dirty_list
);
9844 INIT_LIST_HEAD(&cache
->io_list
);
9845 btrfs_init_free_space_ctl(cache
);
9846 atomic_set(&cache
->trimming
, 0);
9847 mutex_init(&cache
->free_space_lock
);
9848 btrfs_init_full_stripe_locks_tree(&cache
->full_stripe_locks_root
);
9853 int btrfs_read_block_groups(struct btrfs_fs_info
*info
)
9855 struct btrfs_path
*path
;
9857 struct btrfs_block_group_cache
*cache
;
9858 struct btrfs_space_info
*space_info
;
9859 struct btrfs_key key
;
9860 struct btrfs_key found_key
;
9861 struct extent_buffer
*leaf
;
9867 feature
= btrfs_super_incompat_flags(info
->super_copy
);
9868 mixed
= !!(feature
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
);
9872 key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
9873 path
= btrfs_alloc_path();
9876 path
->reada
= READA_FORWARD
;
9878 cache_gen
= btrfs_super_cache_generation(info
->super_copy
);
9879 if (btrfs_test_opt(info
, SPACE_CACHE
) &&
9880 btrfs_super_generation(info
->super_copy
) != cache_gen
)
9882 if (btrfs_test_opt(info
, CLEAR_CACHE
))
9886 ret
= find_first_block_group(info
, path
, &key
);
9892 leaf
= path
->nodes
[0];
9893 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
9895 cache
= btrfs_create_block_group_cache(info
, found_key
.objectid
,
9904 * When we mount with old space cache, we need to
9905 * set BTRFS_DC_CLEAR and set dirty flag.
9907 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9908 * truncate the old free space cache inode and
9910 * b) Setting 'dirty flag' makes sure that we flush
9911 * the new space cache info onto disk.
9913 if (btrfs_test_opt(info
, SPACE_CACHE
))
9914 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
9917 read_extent_buffer(leaf
, &cache
->item
,
9918 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
9919 sizeof(cache
->item
));
9920 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
9922 ((cache
->flags
& BTRFS_BLOCK_GROUP_METADATA
) &&
9923 (cache
->flags
& BTRFS_BLOCK_GROUP_DATA
))) {
9925 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
9926 cache
->key
.objectid
);
9931 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
9932 btrfs_release_path(path
);
9935 * We need to exclude the super stripes now so that the space
9936 * info has super bytes accounted for, otherwise we'll think
9937 * we have more space than we actually do.
9939 ret
= exclude_super_stripes(cache
);
9942 * We may have excluded something, so call this just in
9945 free_excluded_extents(cache
);
9946 btrfs_put_block_group(cache
);
9951 * check for two cases, either we are full, and therefore
9952 * don't need to bother with the caching work since we won't
9953 * find any space, or we are empty, and we can just add all
9954 * the space in and be done with it. This saves us _alot_ of
9955 * time, particularly in the full case.
9957 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
9958 cache
->last_byte_to_unpin
= (u64
)-1;
9959 cache
->cached
= BTRFS_CACHE_FINISHED
;
9960 free_excluded_extents(cache
);
9961 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
9962 cache
->last_byte_to_unpin
= (u64
)-1;
9963 cache
->cached
= BTRFS_CACHE_FINISHED
;
9964 add_new_free_space(cache
, found_key
.objectid
,
9965 found_key
.objectid
+
9967 free_excluded_extents(cache
);
9970 ret
= btrfs_add_block_group_cache(info
, cache
);
9972 btrfs_remove_free_space_cache(cache
);
9973 btrfs_put_block_group(cache
);
9977 trace_btrfs_add_block_group(info
, cache
, 0);
9978 update_space_info(info
, cache
->flags
, found_key
.offset
,
9979 btrfs_block_group_used(&cache
->item
),
9980 cache
->bytes_super
, &space_info
);
9982 cache
->space_info
= space_info
;
9984 link_block_group(cache
);
9986 set_avail_alloc_bits(info
, cache
->flags
);
9987 if (btrfs_chunk_readonly(info
, cache
->key
.objectid
)) {
9988 inc_block_group_ro(cache
, 1);
9989 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
9990 spin_lock(&info
->unused_bgs_lock
);
9991 /* Should always be true but just in case. */
9992 if (list_empty(&cache
->bg_list
)) {
9993 btrfs_get_block_group(cache
);
9994 trace_btrfs_add_unused_block_group(cache
);
9995 list_add_tail(&cache
->bg_list
,
9998 spin_unlock(&info
->unused_bgs_lock
);
10002 list_for_each_entry_rcu(space_info
, &info
->space_info
, list
) {
10003 if (!(get_alloc_profile(info
, space_info
->flags
) &
10004 (BTRFS_BLOCK_GROUP_RAID10
|
10005 BTRFS_BLOCK_GROUP_RAID1
|
10006 BTRFS_BLOCK_GROUP_RAID5
|
10007 BTRFS_BLOCK_GROUP_RAID6
|
10008 BTRFS_BLOCK_GROUP_DUP
)))
10011 * avoid allocating from un-mirrored block group if there are
10012 * mirrored block groups.
10014 list_for_each_entry(cache
,
10015 &space_info
->block_groups
[BTRFS_RAID_RAID0
],
10017 inc_block_group_ro(cache
, 1);
10018 list_for_each_entry(cache
,
10019 &space_info
->block_groups
[BTRFS_RAID_SINGLE
],
10021 inc_block_group_ro(cache
, 1);
10024 btrfs_add_raid_kobjects(info
);
10025 init_global_block_rsv(info
);
10028 btrfs_free_path(path
);
10032 void btrfs_create_pending_block_groups(struct btrfs_trans_handle
*trans
)
10034 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
10035 struct btrfs_block_group_cache
*block_group
, *tmp
;
10036 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
10037 struct btrfs_block_group_item item
;
10038 struct btrfs_key key
;
10040 bool can_flush_pending_bgs
= trans
->can_flush_pending_bgs
;
10042 trans
->can_flush_pending_bgs
= false;
10043 list_for_each_entry_safe(block_group
, tmp
, &trans
->new_bgs
, bg_list
) {
10047 spin_lock(&block_group
->lock
);
10048 memcpy(&item
, &block_group
->item
, sizeof(item
));
10049 memcpy(&key
, &block_group
->key
, sizeof(key
));
10050 spin_unlock(&block_group
->lock
);
10052 ret
= btrfs_insert_item(trans
, extent_root
, &key
, &item
,
10055 btrfs_abort_transaction(trans
, ret
);
10056 ret
= btrfs_finish_chunk_alloc(trans
, fs_info
, key
.objectid
,
10059 btrfs_abort_transaction(trans
, ret
);
10060 add_block_group_free_space(trans
, block_group
);
10061 /* already aborted the transaction if it failed. */
10063 list_del_init(&block_group
->bg_list
);
10065 trans
->can_flush_pending_bgs
= can_flush_pending_bgs
;
10068 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
, u64 bytes_used
,
10069 u64 type
, u64 chunk_offset
, u64 size
)
10071 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
10072 struct btrfs_block_group_cache
*cache
;
10075 btrfs_set_log_full_commit(fs_info
, trans
);
10077 cache
= btrfs_create_block_group_cache(fs_info
, chunk_offset
, size
);
10081 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
10082 btrfs_set_block_group_chunk_objectid(&cache
->item
,
10083 BTRFS_FIRST_CHUNK_TREE_OBJECTID
);
10084 btrfs_set_block_group_flags(&cache
->item
, type
);
10086 cache
->flags
= type
;
10087 cache
->last_byte_to_unpin
= (u64
)-1;
10088 cache
->cached
= BTRFS_CACHE_FINISHED
;
10089 cache
->needs_free_space
= 1;
10090 ret
= exclude_super_stripes(cache
);
10093 * We may have excluded something, so call this just in
10096 free_excluded_extents(cache
);
10097 btrfs_put_block_group(cache
);
10101 add_new_free_space(cache
, chunk_offset
, chunk_offset
+ size
);
10103 free_excluded_extents(cache
);
10105 #ifdef CONFIG_BTRFS_DEBUG
10106 if (btrfs_should_fragment_free_space(cache
)) {
10107 u64 new_bytes_used
= size
- bytes_used
;
10109 bytes_used
+= new_bytes_used
>> 1;
10110 fragment_free_space(cache
);
10114 * Ensure the corresponding space_info object is created and
10115 * assigned to our block group. We want our bg to be added to the rbtree
10116 * with its ->space_info set.
10118 cache
->space_info
= __find_space_info(fs_info
, cache
->flags
);
10119 ASSERT(cache
->space_info
);
10121 ret
= btrfs_add_block_group_cache(fs_info
, cache
);
10123 btrfs_remove_free_space_cache(cache
);
10124 btrfs_put_block_group(cache
);
10129 * Now that our block group has its ->space_info set and is inserted in
10130 * the rbtree, update the space info's counters.
10132 trace_btrfs_add_block_group(fs_info
, cache
, 1);
10133 update_space_info(fs_info
, cache
->flags
, size
, bytes_used
,
10134 cache
->bytes_super
, &cache
->space_info
);
10135 update_global_block_rsv(fs_info
);
10137 link_block_group(cache
);
10139 list_add_tail(&cache
->bg_list
, &trans
->new_bgs
);
10141 set_avail_alloc_bits(fs_info
, type
);
10145 static void clear_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
10147 u64 extra_flags
= chunk_to_extended(flags
) &
10148 BTRFS_EXTENDED_PROFILE_MASK
;
10150 write_seqlock(&fs_info
->profiles_lock
);
10151 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
10152 fs_info
->avail_data_alloc_bits
&= ~extra_flags
;
10153 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
10154 fs_info
->avail_metadata_alloc_bits
&= ~extra_flags
;
10155 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
10156 fs_info
->avail_system_alloc_bits
&= ~extra_flags
;
10157 write_sequnlock(&fs_info
->profiles_lock
);
10160 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
10161 u64 group_start
, struct extent_map
*em
)
10163 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
10164 struct btrfs_root
*root
= fs_info
->extent_root
;
10165 struct btrfs_path
*path
;
10166 struct btrfs_block_group_cache
*block_group
;
10167 struct btrfs_free_cluster
*cluster
;
10168 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
10169 struct btrfs_key key
;
10170 struct inode
*inode
;
10171 struct kobject
*kobj
= NULL
;
10175 struct btrfs_caching_control
*caching_ctl
= NULL
;
10178 block_group
= btrfs_lookup_block_group(fs_info
, group_start
);
10179 BUG_ON(!block_group
);
10180 BUG_ON(!block_group
->ro
);
10182 trace_btrfs_remove_block_group(block_group
);
10184 * Free the reserved super bytes from this block group before
10187 free_excluded_extents(block_group
);
10188 btrfs_free_ref_tree_range(fs_info
, block_group
->key
.objectid
,
10189 block_group
->key
.offset
);
10191 memcpy(&key
, &block_group
->key
, sizeof(key
));
10192 index
= btrfs_bg_flags_to_raid_index(block_group
->flags
);
10193 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
10194 BTRFS_BLOCK_GROUP_RAID1
|
10195 BTRFS_BLOCK_GROUP_RAID10
))
10200 /* make sure this block group isn't part of an allocation cluster */
10201 cluster
= &fs_info
->data_alloc_cluster
;
10202 spin_lock(&cluster
->refill_lock
);
10203 btrfs_return_cluster_to_free_space(block_group
, cluster
);
10204 spin_unlock(&cluster
->refill_lock
);
10207 * make sure this block group isn't part of a metadata
10208 * allocation cluster
10210 cluster
= &fs_info
->meta_alloc_cluster
;
10211 spin_lock(&cluster
->refill_lock
);
10212 btrfs_return_cluster_to_free_space(block_group
, cluster
);
10213 spin_unlock(&cluster
->refill_lock
);
10215 path
= btrfs_alloc_path();
10222 * get the inode first so any iput calls done for the io_list
10223 * aren't the final iput (no unlinks allowed now)
10225 inode
= lookup_free_space_inode(fs_info
, block_group
, path
);
10227 mutex_lock(&trans
->transaction
->cache_write_mutex
);
10229 * make sure our free spache cache IO is done before remove the
10232 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
10233 if (!list_empty(&block_group
->io_list
)) {
10234 list_del_init(&block_group
->io_list
);
10236 WARN_ON(!IS_ERR(inode
) && inode
!= block_group
->io_ctl
.inode
);
10238 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
10239 btrfs_wait_cache_io(trans
, block_group
, path
);
10240 btrfs_put_block_group(block_group
);
10241 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
10244 if (!list_empty(&block_group
->dirty_list
)) {
10245 list_del_init(&block_group
->dirty_list
);
10246 btrfs_put_block_group(block_group
);
10248 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
10249 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
10251 if (!IS_ERR(inode
)) {
10252 ret
= btrfs_orphan_add(trans
, BTRFS_I(inode
));
10254 btrfs_add_delayed_iput(inode
);
10257 clear_nlink(inode
);
10258 /* One for the block groups ref */
10259 spin_lock(&block_group
->lock
);
10260 if (block_group
->iref
) {
10261 block_group
->iref
= 0;
10262 block_group
->inode
= NULL
;
10263 spin_unlock(&block_group
->lock
);
10266 spin_unlock(&block_group
->lock
);
10268 /* One for our lookup ref */
10269 btrfs_add_delayed_iput(inode
);
10272 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
10273 key
.offset
= block_group
->key
.objectid
;
10276 ret
= btrfs_search_slot(trans
, tree_root
, &key
, path
, -1, 1);
10280 btrfs_release_path(path
);
10282 ret
= btrfs_del_item(trans
, tree_root
, path
);
10285 btrfs_release_path(path
);
10288 spin_lock(&fs_info
->block_group_cache_lock
);
10289 rb_erase(&block_group
->cache_node
,
10290 &fs_info
->block_group_cache_tree
);
10291 RB_CLEAR_NODE(&block_group
->cache_node
);
10293 if (fs_info
->first_logical_byte
== block_group
->key
.objectid
)
10294 fs_info
->first_logical_byte
= (u64
)-1;
10295 spin_unlock(&fs_info
->block_group_cache_lock
);
10297 down_write(&block_group
->space_info
->groups_sem
);
10299 * we must use list_del_init so people can check to see if they
10300 * are still on the list after taking the semaphore
10302 list_del_init(&block_group
->list
);
10303 if (list_empty(&block_group
->space_info
->block_groups
[index
])) {
10304 kobj
= block_group
->space_info
->block_group_kobjs
[index
];
10305 block_group
->space_info
->block_group_kobjs
[index
] = NULL
;
10306 clear_avail_alloc_bits(fs_info
, block_group
->flags
);
10308 up_write(&block_group
->space_info
->groups_sem
);
10314 if (block_group
->has_caching_ctl
)
10315 caching_ctl
= get_caching_control(block_group
);
10316 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
10317 wait_block_group_cache_done(block_group
);
10318 if (block_group
->has_caching_ctl
) {
10319 down_write(&fs_info
->commit_root_sem
);
10320 if (!caching_ctl
) {
10321 struct btrfs_caching_control
*ctl
;
10323 list_for_each_entry(ctl
,
10324 &fs_info
->caching_block_groups
, list
)
10325 if (ctl
->block_group
== block_group
) {
10327 refcount_inc(&caching_ctl
->count
);
10332 list_del_init(&caching_ctl
->list
);
10333 up_write(&fs_info
->commit_root_sem
);
10335 /* Once for the caching bgs list and once for us. */
10336 put_caching_control(caching_ctl
);
10337 put_caching_control(caching_ctl
);
10341 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
10342 if (!list_empty(&block_group
->dirty_list
)) {
10345 if (!list_empty(&block_group
->io_list
)) {
10348 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
10349 btrfs_remove_free_space_cache(block_group
);
10351 spin_lock(&block_group
->space_info
->lock
);
10352 list_del_init(&block_group
->ro_list
);
10354 if (btrfs_test_opt(fs_info
, ENOSPC_DEBUG
)) {
10355 WARN_ON(block_group
->space_info
->total_bytes
10356 < block_group
->key
.offset
);
10357 WARN_ON(block_group
->space_info
->bytes_readonly
10358 < block_group
->key
.offset
);
10359 WARN_ON(block_group
->space_info
->disk_total
10360 < block_group
->key
.offset
* factor
);
10362 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
10363 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
10364 block_group
->space_info
->disk_total
-= block_group
->key
.offset
* factor
;
10366 spin_unlock(&block_group
->space_info
->lock
);
10368 memcpy(&key
, &block_group
->key
, sizeof(key
));
10370 mutex_lock(&fs_info
->chunk_mutex
);
10371 if (!list_empty(&em
->list
)) {
10372 /* We're in the transaction->pending_chunks list. */
10373 free_extent_map(em
);
10375 spin_lock(&block_group
->lock
);
10376 block_group
->removed
= 1;
10378 * At this point trimming can't start on this block group, because we
10379 * removed the block group from the tree fs_info->block_group_cache_tree
10380 * so no one can't find it anymore and even if someone already got this
10381 * block group before we removed it from the rbtree, they have already
10382 * incremented block_group->trimming - if they didn't, they won't find
10383 * any free space entries because we already removed them all when we
10384 * called btrfs_remove_free_space_cache().
10386 * And we must not remove the extent map from the fs_info->mapping_tree
10387 * to prevent the same logical address range and physical device space
10388 * ranges from being reused for a new block group. This is because our
10389 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10390 * completely transactionless, so while it is trimming a range the
10391 * currently running transaction might finish and a new one start,
10392 * allowing for new block groups to be created that can reuse the same
10393 * physical device locations unless we take this special care.
10395 * There may also be an implicit trim operation if the file system
10396 * is mounted with -odiscard. The same protections must remain
10397 * in place until the extents have been discarded completely when
10398 * the transaction commit has completed.
10400 remove_em
= (atomic_read(&block_group
->trimming
) == 0);
10402 * Make sure a trimmer task always sees the em in the pinned_chunks list
10403 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10404 * before checking block_group->removed).
10408 * Our em might be in trans->transaction->pending_chunks which
10409 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10410 * and so is the fs_info->pinned_chunks list.
10412 * So at this point we must be holding the chunk_mutex to avoid
10413 * any races with chunk allocation (more specifically at
10414 * volumes.c:contains_pending_extent()), to ensure it always
10415 * sees the em, either in the pending_chunks list or in the
10416 * pinned_chunks list.
10418 list_move_tail(&em
->list
, &fs_info
->pinned_chunks
);
10420 spin_unlock(&block_group
->lock
);
10423 struct extent_map_tree
*em_tree
;
10425 em_tree
= &fs_info
->mapping_tree
.map_tree
;
10426 write_lock(&em_tree
->lock
);
10428 * The em might be in the pending_chunks list, so make sure the
10429 * chunk mutex is locked, since remove_extent_mapping() will
10430 * delete us from that list.
10432 remove_extent_mapping(em_tree
, em
);
10433 write_unlock(&em_tree
->lock
);
10434 /* once for the tree */
10435 free_extent_map(em
);
10438 mutex_unlock(&fs_info
->chunk_mutex
);
10440 ret
= remove_block_group_free_space(trans
, block_group
);
10444 btrfs_put_block_group(block_group
);
10445 btrfs_put_block_group(block_group
);
10447 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
10453 ret
= btrfs_del_item(trans
, root
, path
);
10455 btrfs_free_path(path
);
10459 struct btrfs_trans_handle
*
10460 btrfs_start_trans_remove_block_group(struct btrfs_fs_info
*fs_info
,
10461 const u64 chunk_offset
)
10463 struct extent_map_tree
*em_tree
= &fs_info
->mapping_tree
.map_tree
;
10464 struct extent_map
*em
;
10465 struct map_lookup
*map
;
10466 unsigned int num_items
;
10468 read_lock(&em_tree
->lock
);
10469 em
= lookup_extent_mapping(em_tree
, chunk_offset
, 1);
10470 read_unlock(&em_tree
->lock
);
10471 ASSERT(em
&& em
->start
== chunk_offset
);
10474 * We need to reserve 3 + N units from the metadata space info in order
10475 * to remove a block group (done at btrfs_remove_chunk() and at
10476 * btrfs_remove_block_group()), which are used for:
10478 * 1 unit for adding the free space inode's orphan (located in the tree
10480 * 1 unit for deleting the block group item (located in the extent
10482 * 1 unit for deleting the free space item (located in tree of tree
10484 * N units for deleting N device extent items corresponding to each
10485 * stripe (located in the device tree).
10487 * In order to remove a block group we also need to reserve units in the
10488 * system space info in order to update the chunk tree (update one or
10489 * more device items and remove one chunk item), but this is done at
10490 * btrfs_remove_chunk() through a call to check_system_chunk().
10492 map
= em
->map_lookup
;
10493 num_items
= 3 + map
->num_stripes
;
10494 free_extent_map(em
);
10496 return btrfs_start_transaction_fallback_global_rsv(fs_info
->extent_root
,
10501 * Process the unused_bgs list and remove any that don't have any allocated
10502 * space inside of them.
10504 void btrfs_delete_unused_bgs(struct btrfs_fs_info
*fs_info
)
10506 struct btrfs_block_group_cache
*block_group
;
10507 struct btrfs_space_info
*space_info
;
10508 struct btrfs_trans_handle
*trans
;
10511 if (!test_bit(BTRFS_FS_OPEN
, &fs_info
->flags
))
10514 spin_lock(&fs_info
->unused_bgs_lock
);
10515 while (!list_empty(&fs_info
->unused_bgs
)) {
10519 block_group
= list_first_entry(&fs_info
->unused_bgs
,
10520 struct btrfs_block_group_cache
,
10522 list_del_init(&block_group
->bg_list
);
10524 space_info
= block_group
->space_info
;
10526 if (ret
|| btrfs_mixed_space_info(space_info
)) {
10527 btrfs_put_block_group(block_group
);
10530 spin_unlock(&fs_info
->unused_bgs_lock
);
10532 mutex_lock(&fs_info
->delete_unused_bgs_mutex
);
10534 /* Don't want to race with allocators so take the groups_sem */
10535 down_write(&space_info
->groups_sem
);
10536 spin_lock(&block_group
->lock
);
10537 if (block_group
->reserved
|| block_group
->pinned
||
10538 btrfs_block_group_used(&block_group
->item
) ||
10540 list_is_singular(&block_group
->list
)) {
10542 * We want to bail if we made new allocations or have
10543 * outstanding allocations in this block group. We do
10544 * the ro check in case balance is currently acting on
10545 * this block group.
10547 trace_btrfs_skip_unused_block_group(block_group
);
10548 spin_unlock(&block_group
->lock
);
10549 up_write(&space_info
->groups_sem
);
10552 spin_unlock(&block_group
->lock
);
10554 /* We don't want to force the issue, only flip if it's ok. */
10555 ret
= inc_block_group_ro(block_group
, 0);
10556 up_write(&space_info
->groups_sem
);
10563 * Want to do this before we do anything else so we can recover
10564 * properly if we fail to join the transaction.
10566 trans
= btrfs_start_trans_remove_block_group(fs_info
,
10567 block_group
->key
.objectid
);
10568 if (IS_ERR(trans
)) {
10569 btrfs_dec_block_group_ro(block_group
);
10570 ret
= PTR_ERR(trans
);
10575 * We could have pending pinned extents for this block group,
10576 * just delete them, we don't care about them anymore.
10578 start
= block_group
->key
.objectid
;
10579 end
= start
+ block_group
->key
.offset
- 1;
10581 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10582 * btrfs_finish_extent_commit(). If we are at transaction N,
10583 * another task might be running finish_extent_commit() for the
10584 * previous transaction N - 1, and have seen a range belonging
10585 * to the block group in freed_extents[] before we were able to
10586 * clear the whole block group range from freed_extents[]. This
10587 * means that task can lookup for the block group after we
10588 * unpinned it from freed_extents[] and removed it, leading to
10589 * a BUG_ON() at btrfs_unpin_extent_range().
10591 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
10592 ret
= clear_extent_bits(&fs_info
->freed_extents
[0], start
, end
,
10595 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10596 btrfs_dec_block_group_ro(block_group
);
10599 ret
= clear_extent_bits(&fs_info
->freed_extents
[1], start
, end
,
10602 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10603 btrfs_dec_block_group_ro(block_group
);
10606 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10608 /* Reset pinned so btrfs_put_block_group doesn't complain */
10609 spin_lock(&space_info
->lock
);
10610 spin_lock(&block_group
->lock
);
10612 space_info
->bytes_pinned
-= block_group
->pinned
;
10613 space_info
->bytes_readonly
+= block_group
->pinned
;
10614 percpu_counter_add_batch(&space_info
->total_bytes_pinned
,
10615 -block_group
->pinned
,
10616 BTRFS_TOTAL_BYTES_PINNED_BATCH
);
10617 block_group
->pinned
= 0;
10619 spin_unlock(&block_group
->lock
);
10620 spin_unlock(&space_info
->lock
);
10622 /* DISCARD can flip during remount */
10623 trimming
= btrfs_test_opt(fs_info
, DISCARD
);
10625 /* Implicit trim during transaction commit. */
10627 btrfs_get_block_group_trimming(block_group
);
10630 * Btrfs_remove_chunk will abort the transaction if things go
10633 ret
= btrfs_remove_chunk(trans
, fs_info
,
10634 block_group
->key
.objectid
);
10638 btrfs_put_block_group_trimming(block_group
);
10643 * If we're not mounted with -odiscard, we can just forget
10644 * about this block group. Otherwise we'll need to wait
10645 * until transaction commit to do the actual discard.
10648 spin_lock(&fs_info
->unused_bgs_lock
);
10650 * A concurrent scrub might have added us to the list
10651 * fs_info->unused_bgs, so use a list_move operation
10652 * to add the block group to the deleted_bgs list.
10654 list_move(&block_group
->bg_list
,
10655 &trans
->transaction
->deleted_bgs
);
10656 spin_unlock(&fs_info
->unused_bgs_lock
);
10657 btrfs_get_block_group(block_group
);
10660 btrfs_end_transaction(trans
);
10662 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
10663 btrfs_put_block_group(block_group
);
10664 spin_lock(&fs_info
->unused_bgs_lock
);
10666 spin_unlock(&fs_info
->unused_bgs_lock
);
10669 int btrfs_init_space_info(struct btrfs_fs_info
*fs_info
)
10671 struct btrfs_super_block
*disk_super
;
10677 disk_super
= fs_info
->super_copy
;
10678 if (!btrfs_super_root(disk_super
))
10681 features
= btrfs_super_incompat_flags(disk_super
);
10682 if (features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
10685 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
10686 ret
= create_space_info(fs_info
, flags
);
10691 flags
= BTRFS_BLOCK_GROUP_METADATA
| BTRFS_BLOCK_GROUP_DATA
;
10692 ret
= create_space_info(fs_info
, flags
);
10694 flags
= BTRFS_BLOCK_GROUP_METADATA
;
10695 ret
= create_space_info(fs_info
, flags
);
10699 flags
= BTRFS_BLOCK_GROUP_DATA
;
10700 ret
= create_space_info(fs_info
, flags
);
10706 int btrfs_error_unpin_extent_range(struct btrfs_fs_info
*fs_info
,
10707 u64 start
, u64 end
)
10709 return unpin_extent_range(fs_info
, start
, end
, false);
10713 * It used to be that old block groups would be left around forever.
10714 * Iterating over them would be enough to trim unused space. Since we
10715 * now automatically remove them, we also need to iterate over unallocated
10718 * We don't want a transaction for this since the discard may take a
10719 * substantial amount of time. We don't require that a transaction be
10720 * running, but we do need to take a running transaction into account
10721 * to ensure that we're not discarding chunks that were released in
10722 * the current transaction.
10724 * Holding the chunks lock will prevent other threads from allocating
10725 * or releasing chunks, but it won't prevent a running transaction
10726 * from committing and releasing the memory that the pending chunks
10727 * list head uses. For that, we need to take a reference to the
10730 static int btrfs_trim_free_extents(struct btrfs_device
*device
,
10731 u64 minlen
, u64
*trimmed
)
10733 u64 start
= 0, len
= 0;
10738 /* Not writeable = nothing to do. */
10739 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
))
10742 /* No free space = nothing to do. */
10743 if (device
->total_bytes
<= device
->bytes_used
)
10749 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
10750 struct btrfs_transaction
*trans
;
10753 ret
= mutex_lock_interruptible(&fs_info
->chunk_mutex
);
10757 down_read(&fs_info
->commit_root_sem
);
10759 spin_lock(&fs_info
->trans_lock
);
10760 trans
= fs_info
->running_transaction
;
10762 refcount_inc(&trans
->use_count
);
10763 spin_unlock(&fs_info
->trans_lock
);
10765 ret
= find_free_dev_extent_start(trans
, device
, minlen
, start
,
10768 btrfs_put_transaction(trans
);
10771 up_read(&fs_info
->commit_root_sem
);
10772 mutex_unlock(&fs_info
->chunk_mutex
);
10773 if (ret
== -ENOSPC
)
10778 ret
= btrfs_issue_discard(device
->bdev
, start
, len
, &bytes
);
10779 up_read(&fs_info
->commit_root_sem
);
10780 mutex_unlock(&fs_info
->chunk_mutex
);
10788 if (fatal_signal_pending(current
)) {
10789 ret
= -ERESTARTSYS
;
10799 int btrfs_trim_fs(struct btrfs_fs_info
*fs_info
, struct fstrim_range
*range
)
10801 struct btrfs_block_group_cache
*cache
= NULL
;
10802 struct btrfs_device
*device
;
10803 struct list_head
*devices
;
10808 u64 total_bytes
= btrfs_super_total_bytes(fs_info
->super_copy
);
10812 * try to trim all FS space, our block group may start from non-zero.
10814 if (range
->len
== total_bytes
)
10815 cache
= btrfs_lookup_first_block_group(fs_info
, range
->start
);
10817 cache
= btrfs_lookup_block_group(fs_info
, range
->start
);
10820 if (cache
->key
.objectid
>= (range
->start
+ range
->len
)) {
10821 btrfs_put_block_group(cache
);
10825 start
= max(range
->start
, cache
->key
.objectid
);
10826 end
= min(range
->start
+ range
->len
,
10827 cache
->key
.objectid
+ cache
->key
.offset
);
10829 if (end
- start
>= range
->minlen
) {
10830 if (!block_group_cache_done(cache
)) {
10831 ret
= cache_block_group(cache
, 0);
10833 btrfs_put_block_group(cache
);
10836 ret
= wait_block_group_cache_done(cache
);
10838 btrfs_put_block_group(cache
);
10842 ret
= btrfs_trim_block_group(cache
,
10848 trimmed
+= group_trimmed
;
10850 btrfs_put_block_group(cache
);
10855 cache
= next_block_group(fs_info
, cache
);
10858 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
10859 devices
= &fs_info
->fs_devices
->alloc_list
;
10860 list_for_each_entry(device
, devices
, dev_alloc_list
) {
10861 ret
= btrfs_trim_free_extents(device
, range
->minlen
,
10866 trimmed
+= group_trimmed
;
10868 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
10870 range
->len
= trimmed
;
10875 * btrfs_{start,end}_write_no_snapshotting() are similar to
10876 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10877 * data into the page cache through nocow before the subvolume is snapshoted,
10878 * but flush the data into disk after the snapshot creation, or to prevent
10879 * operations while snapshotting is ongoing and that cause the snapshot to be
10880 * inconsistent (writes followed by expanding truncates for example).
10882 void btrfs_end_write_no_snapshotting(struct btrfs_root
*root
)
10884 percpu_counter_dec(&root
->subv_writers
->counter
);
10885 cond_wake_up(&root
->subv_writers
->wait
);
10888 int btrfs_start_write_no_snapshotting(struct btrfs_root
*root
)
10890 if (atomic_read(&root
->will_be_snapshotted
))
10893 percpu_counter_inc(&root
->subv_writers
->counter
);
10895 * Make sure counter is updated before we check for snapshot creation.
10898 if (atomic_read(&root
->will_be_snapshotted
)) {
10899 btrfs_end_write_no_snapshotting(root
);
10905 void btrfs_wait_for_snapshot_creation(struct btrfs_root
*root
)
10910 ret
= btrfs_start_write_no_snapshotting(root
);
10913 wait_var_event(&root
->will_be_snapshotted
,
10914 !atomic_read(&root
->will_be_snapshotted
));