1 // SPDX-License-Identifier: GPL-2.0
3 #include <linux/sizes.h>
4 #include <linux/list_sort.h>
7 #include "block-group.h"
8 #include "space-info.h"
10 #include "free-space-cache.h"
11 #include "free-space-tree.h"
13 #include "transaction.h"
14 #include "ref-verify.h"
17 #include "delalloc-space.h"
22 #include "accessors.h"
23 #include "extent-tree.h"
25 #ifdef CONFIG_BTRFS_DEBUG
26 int btrfs_should_fragment_free_space(struct btrfs_block_group
*block_group
)
28 struct btrfs_fs_info
*fs_info
= block_group
->fs_info
;
30 return (btrfs_test_opt(fs_info
, FRAGMENT_METADATA
) &&
31 block_group
->flags
& BTRFS_BLOCK_GROUP_METADATA
) ||
32 (btrfs_test_opt(fs_info
, FRAGMENT_DATA
) &&
33 block_group
->flags
& BTRFS_BLOCK_GROUP_DATA
);
38 * Return target flags in extended format or 0 if restripe for this chunk_type
41 * Should be called with balance_lock held
43 static u64
get_restripe_target(struct btrfs_fs_info
*fs_info
, u64 flags
)
45 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
51 if (flags
& BTRFS_BLOCK_GROUP_DATA
&&
52 bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
53 target
= BTRFS_BLOCK_GROUP_DATA
| bctl
->data
.target
;
54 } else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
&&
55 bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
56 target
= BTRFS_BLOCK_GROUP_SYSTEM
| bctl
->sys
.target
;
57 } else if (flags
& BTRFS_BLOCK_GROUP_METADATA
&&
58 bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
59 target
= BTRFS_BLOCK_GROUP_METADATA
| bctl
->meta
.target
;
66 * @flags: available profiles in extended format (see ctree.h)
68 * Return reduced profile in chunk format. If profile changing is in progress
69 * (either running or paused) picks the target profile (if it's already
70 * available), otherwise falls back to plain reducing.
72 static u64
btrfs_reduce_alloc_profile(struct btrfs_fs_info
*fs_info
, u64 flags
)
74 u64 num_devices
= fs_info
->fs_devices
->rw_devices
;
80 * See if restripe for this chunk_type is in progress, if so try to
81 * reduce to the target profile
83 spin_lock(&fs_info
->balance_lock
);
84 target
= get_restripe_target(fs_info
, flags
);
86 spin_unlock(&fs_info
->balance_lock
);
87 return extended_to_chunk(target
);
89 spin_unlock(&fs_info
->balance_lock
);
91 /* First, mask out the RAID levels which aren't possible */
92 for (raid_type
= 0; raid_type
< BTRFS_NR_RAID_TYPES
; raid_type
++) {
93 if (num_devices
>= btrfs_raid_array
[raid_type
].devs_min
)
94 allowed
|= btrfs_raid_array
[raid_type
].bg_flag
;
98 /* Select the highest-redundancy RAID level. */
99 if (allowed
& BTRFS_BLOCK_GROUP_RAID1C4
)
100 allowed
= BTRFS_BLOCK_GROUP_RAID1C4
;
101 else if (allowed
& BTRFS_BLOCK_GROUP_RAID6
)
102 allowed
= BTRFS_BLOCK_GROUP_RAID6
;
103 else if (allowed
& BTRFS_BLOCK_GROUP_RAID1C3
)
104 allowed
= BTRFS_BLOCK_GROUP_RAID1C3
;
105 else if (allowed
& BTRFS_BLOCK_GROUP_RAID5
)
106 allowed
= BTRFS_BLOCK_GROUP_RAID5
;
107 else if (allowed
& BTRFS_BLOCK_GROUP_RAID10
)
108 allowed
= BTRFS_BLOCK_GROUP_RAID10
;
109 else if (allowed
& BTRFS_BLOCK_GROUP_RAID1
)
110 allowed
= BTRFS_BLOCK_GROUP_RAID1
;
111 else if (allowed
& BTRFS_BLOCK_GROUP_DUP
)
112 allowed
= BTRFS_BLOCK_GROUP_DUP
;
113 else if (allowed
& BTRFS_BLOCK_GROUP_RAID0
)
114 allowed
= BTRFS_BLOCK_GROUP_RAID0
;
116 flags
&= ~BTRFS_BLOCK_GROUP_PROFILE_MASK
;
118 return extended_to_chunk(flags
| allowed
);
121 u64
btrfs_get_alloc_profile(struct btrfs_fs_info
*fs_info
, u64 orig_flags
)
128 seq
= read_seqbegin(&fs_info
->profiles_lock
);
130 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
131 flags
|= fs_info
->avail_data_alloc_bits
;
132 else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
133 flags
|= fs_info
->avail_system_alloc_bits
;
134 else if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
135 flags
|= fs_info
->avail_metadata_alloc_bits
;
136 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
138 return btrfs_reduce_alloc_profile(fs_info
, flags
);
141 void btrfs_get_block_group(struct btrfs_block_group
*cache
)
143 refcount_inc(&cache
->refs
);
146 void btrfs_put_block_group(struct btrfs_block_group
*cache
)
148 if (refcount_dec_and_test(&cache
->refs
)) {
149 WARN_ON(cache
->pinned
> 0);
151 * If there was a failure to cleanup a log tree, very likely due
152 * to an IO failure on a writeback attempt of one or more of its
153 * extent buffers, we could not do proper (and cheap) unaccounting
154 * of their reserved space, so don't warn on reserved > 0 in that
157 if (!(cache
->flags
& BTRFS_BLOCK_GROUP_METADATA
) ||
158 !BTRFS_FS_LOG_CLEANUP_ERROR(cache
->fs_info
))
159 WARN_ON(cache
->reserved
> 0);
162 * A block_group shouldn't be on the discard_list anymore.
163 * Remove the block_group from the discard_list to prevent us
164 * from causing a panic due to NULL pointer dereference.
166 if (WARN_ON(!list_empty(&cache
->discard_list
)))
167 btrfs_discard_cancel_work(&cache
->fs_info
->discard_ctl
,
170 kfree(cache
->free_space_ctl
);
171 kfree(cache
->physical_map
);
177 * This adds the block group to the fs_info rb tree for the block group cache
179 static int btrfs_add_block_group_cache(struct btrfs_fs_info
*info
,
180 struct btrfs_block_group
*block_group
)
183 struct rb_node
*parent
= NULL
;
184 struct btrfs_block_group
*cache
;
185 bool leftmost
= true;
187 ASSERT(block_group
->length
!= 0);
189 write_lock(&info
->block_group_cache_lock
);
190 p
= &info
->block_group_cache_tree
.rb_root
.rb_node
;
194 cache
= rb_entry(parent
, struct btrfs_block_group
, cache_node
);
195 if (block_group
->start
< cache
->start
) {
197 } else if (block_group
->start
> cache
->start
) {
201 write_unlock(&info
->block_group_cache_lock
);
206 rb_link_node(&block_group
->cache_node
, parent
, p
);
207 rb_insert_color_cached(&block_group
->cache_node
,
208 &info
->block_group_cache_tree
, leftmost
);
210 write_unlock(&info
->block_group_cache_lock
);
216 * This will return the block group at or after bytenr if contains is 0, else
217 * it will return the block group that contains the bytenr
219 static struct btrfs_block_group
*block_group_cache_tree_search(
220 struct btrfs_fs_info
*info
, u64 bytenr
, int contains
)
222 struct btrfs_block_group
*cache
, *ret
= NULL
;
226 read_lock(&info
->block_group_cache_lock
);
227 n
= info
->block_group_cache_tree
.rb_root
.rb_node
;
230 cache
= rb_entry(n
, struct btrfs_block_group
, cache_node
);
231 end
= cache
->start
+ cache
->length
- 1;
232 start
= cache
->start
;
234 if (bytenr
< start
) {
235 if (!contains
&& (!ret
|| start
< ret
->start
))
238 } else if (bytenr
> start
) {
239 if (contains
&& bytenr
<= end
) {
250 btrfs_get_block_group(ret
);
251 read_unlock(&info
->block_group_cache_lock
);
257 * Return the block group that starts at or after bytenr
259 struct btrfs_block_group
*btrfs_lookup_first_block_group(
260 struct btrfs_fs_info
*info
, u64 bytenr
)
262 return block_group_cache_tree_search(info
, bytenr
, 0);
266 * Return the block group that contains the given bytenr
268 struct btrfs_block_group
*btrfs_lookup_block_group(
269 struct btrfs_fs_info
*info
, u64 bytenr
)
271 return block_group_cache_tree_search(info
, bytenr
, 1);
274 struct btrfs_block_group
*btrfs_next_block_group(
275 struct btrfs_block_group
*cache
)
277 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
278 struct rb_node
*node
;
280 read_lock(&fs_info
->block_group_cache_lock
);
282 /* If our block group was removed, we need a full search. */
283 if (RB_EMPTY_NODE(&cache
->cache_node
)) {
284 const u64 next_bytenr
= cache
->start
+ cache
->length
;
286 read_unlock(&fs_info
->block_group_cache_lock
);
287 btrfs_put_block_group(cache
);
288 return btrfs_lookup_first_block_group(fs_info
, next_bytenr
);
290 node
= rb_next(&cache
->cache_node
);
291 btrfs_put_block_group(cache
);
293 cache
= rb_entry(node
, struct btrfs_block_group
, cache_node
);
294 btrfs_get_block_group(cache
);
297 read_unlock(&fs_info
->block_group_cache_lock
);
302 * Check if we can do a NOCOW write for a given extent.
304 * @fs_info: The filesystem information object.
305 * @bytenr: Logical start address of the extent.
307 * Check if we can do a NOCOW write for the given extent, and increments the
308 * number of NOCOW writers in the block group that contains the extent, as long
309 * as the block group exists and it's currently not in read-only mode.
311 * Returns: A non-NULL block group pointer if we can do a NOCOW write, the caller
312 * is responsible for calling btrfs_dec_nocow_writers() later.
314 * Or NULL if we can not do a NOCOW write
316 struct btrfs_block_group
*btrfs_inc_nocow_writers(struct btrfs_fs_info
*fs_info
,
319 struct btrfs_block_group
*bg
;
320 bool can_nocow
= true;
322 bg
= btrfs_lookup_block_group(fs_info
, bytenr
);
326 spin_lock(&bg
->lock
);
330 atomic_inc(&bg
->nocow_writers
);
331 spin_unlock(&bg
->lock
);
334 btrfs_put_block_group(bg
);
338 /* No put on block group, done by btrfs_dec_nocow_writers(). */
343 * Decrement the number of NOCOW writers in a block group.
345 * This is meant to be called after a previous call to btrfs_inc_nocow_writers(),
346 * and on the block group returned by that call. Typically this is called after
347 * creating an ordered extent for a NOCOW write, to prevent races with scrub and
350 * After this call, the caller should not use the block group anymore. It it wants
351 * to use it, then it should get a reference on it before calling this function.
353 void btrfs_dec_nocow_writers(struct btrfs_block_group
*bg
)
355 if (atomic_dec_and_test(&bg
->nocow_writers
))
356 wake_up_var(&bg
->nocow_writers
);
358 /* For the lookup done by a previous call to btrfs_inc_nocow_writers(). */
359 btrfs_put_block_group(bg
);
362 void btrfs_wait_nocow_writers(struct btrfs_block_group
*bg
)
364 wait_var_event(&bg
->nocow_writers
, !atomic_read(&bg
->nocow_writers
));
367 void btrfs_dec_block_group_reservations(struct btrfs_fs_info
*fs_info
,
370 struct btrfs_block_group
*bg
;
372 bg
= btrfs_lookup_block_group(fs_info
, start
);
374 if (atomic_dec_and_test(&bg
->reservations
))
375 wake_up_var(&bg
->reservations
);
376 btrfs_put_block_group(bg
);
379 void btrfs_wait_block_group_reservations(struct btrfs_block_group
*bg
)
381 struct btrfs_space_info
*space_info
= bg
->space_info
;
385 if (!(bg
->flags
& BTRFS_BLOCK_GROUP_DATA
))
389 * Our block group is read only but before we set it to read only,
390 * some task might have had allocated an extent from it already, but it
391 * has not yet created a respective ordered extent (and added it to a
392 * root's list of ordered extents).
393 * Therefore wait for any task currently allocating extents, since the
394 * block group's reservations counter is incremented while a read lock
395 * on the groups' semaphore is held and decremented after releasing
396 * the read access on that semaphore and creating the ordered extent.
398 down_write(&space_info
->groups_sem
);
399 up_write(&space_info
->groups_sem
);
401 wait_var_event(&bg
->reservations
, !atomic_read(&bg
->reservations
));
404 struct btrfs_caching_control
*btrfs_get_caching_control(
405 struct btrfs_block_group
*cache
)
407 struct btrfs_caching_control
*ctl
;
409 spin_lock(&cache
->lock
);
410 if (!cache
->caching_ctl
) {
411 spin_unlock(&cache
->lock
);
415 ctl
= cache
->caching_ctl
;
416 refcount_inc(&ctl
->count
);
417 spin_unlock(&cache
->lock
);
421 void btrfs_put_caching_control(struct btrfs_caching_control
*ctl
)
423 if (refcount_dec_and_test(&ctl
->count
))
428 * When we wait for progress in the block group caching, its because our
429 * allocation attempt failed at least once. So, we must sleep and let some
430 * progress happen before we try again.
432 * This function will sleep at least once waiting for new free space to show
433 * up, and then it will check the block group free space numbers for our min
434 * num_bytes. Another option is to have it go ahead and look in the rbtree for
435 * a free extent of a given size, but this is a good start.
437 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
438 * any of the information in this block group.
440 void btrfs_wait_block_group_cache_progress(struct btrfs_block_group
*cache
,
443 struct btrfs_caching_control
*caching_ctl
;
446 caching_ctl
= btrfs_get_caching_control(cache
);
451 * We've already failed to allocate from this block group, so even if
452 * there's enough space in the block group it isn't contiguous enough to
453 * allow for an allocation, so wait for at least the next wakeup tick,
454 * or for the thing to be done.
456 progress
= atomic_read(&caching_ctl
->progress
);
458 wait_event(caching_ctl
->wait
, btrfs_block_group_done(cache
) ||
459 (progress
!= atomic_read(&caching_ctl
->progress
) &&
460 (cache
->free_space_ctl
->free_space
>= num_bytes
)));
462 btrfs_put_caching_control(caching_ctl
);
465 static int btrfs_caching_ctl_wait_done(struct btrfs_block_group
*cache
,
466 struct btrfs_caching_control
*caching_ctl
)
468 wait_event(caching_ctl
->wait
, btrfs_block_group_done(cache
));
469 return cache
->cached
== BTRFS_CACHE_ERROR
? -EIO
: 0;
472 static int btrfs_wait_block_group_cache_done(struct btrfs_block_group
*cache
)
474 struct btrfs_caching_control
*caching_ctl
;
477 caching_ctl
= btrfs_get_caching_control(cache
);
479 return (cache
->cached
== BTRFS_CACHE_ERROR
) ? -EIO
: 0;
480 ret
= btrfs_caching_ctl_wait_done(cache
, caching_ctl
);
481 btrfs_put_caching_control(caching_ctl
);
485 #ifdef CONFIG_BTRFS_DEBUG
486 static void fragment_free_space(struct btrfs_block_group
*block_group
)
488 struct btrfs_fs_info
*fs_info
= block_group
->fs_info
;
489 u64 start
= block_group
->start
;
490 u64 len
= block_group
->length
;
491 u64 chunk
= block_group
->flags
& BTRFS_BLOCK_GROUP_METADATA
?
492 fs_info
->nodesize
: fs_info
->sectorsize
;
493 u64 step
= chunk
<< 1;
495 while (len
> chunk
) {
496 btrfs_remove_free_space(block_group
, start
, chunk
);
507 * Add a free space range to the in memory free space cache of a block group.
508 * This checks if the range contains super block locations and any such
509 * locations are not added to the free space cache.
511 * @block_group: The target block group.
512 * @start: Start offset of the range.
513 * @end: End offset of the range (exclusive).
514 * @total_added_ret: Optional pointer to return the total amount of space
515 * added to the block group's free space cache.
517 * Returns 0 on success or < 0 on error.
519 int btrfs_add_new_free_space(struct btrfs_block_group
*block_group
, u64 start
,
520 u64 end
, u64
*total_added_ret
)
522 struct btrfs_fs_info
*info
= block_group
->fs_info
;
523 u64 extent_start
, extent_end
, size
;
527 *total_added_ret
= 0;
529 while (start
< end
) {
530 if (!find_first_extent_bit(&info
->excluded_extents
, start
,
531 &extent_start
, &extent_end
,
532 EXTENT_DIRTY
| EXTENT_UPTODATE
,
536 if (extent_start
<= start
) {
537 start
= extent_end
+ 1;
538 } else if (extent_start
> start
&& extent_start
< end
) {
539 size
= extent_start
- start
;
540 ret
= btrfs_add_free_space_async_trimmed(block_group
,
545 *total_added_ret
+= size
;
546 start
= extent_end
+ 1;
554 ret
= btrfs_add_free_space_async_trimmed(block_group
, start
,
559 *total_added_ret
+= size
;
566 * Get an arbitrary extent item index / max_index through the block group
568 * @block_group the block group to sample from
569 * @index: the integral step through the block group to grab from
570 * @max_index: the granularity of the sampling
571 * @key: return value parameter for the item we find
573 * Pre-conditions on indices:
574 * 0 <= index <= max_index
577 * Returns: 0 on success, 1 if the search didn't yield a useful item, negative
578 * error code on error.
580 static int sample_block_group_extent_item(struct btrfs_caching_control
*caching_ctl
,
581 struct btrfs_block_group
*block_group
,
582 int index
, int max_index
,
583 struct btrfs_key
*found_key
)
585 struct btrfs_fs_info
*fs_info
= block_group
->fs_info
;
586 struct btrfs_root
*extent_root
;
588 u64 search_end
= block_group
->start
+ block_group
->length
;
589 struct btrfs_path
*path
;
590 struct btrfs_key search_key
;
594 ASSERT(index
<= max_index
);
595 ASSERT(max_index
> 0);
596 lockdep_assert_held(&caching_ctl
->mutex
);
597 lockdep_assert_held_read(&fs_info
->commit_root_sem
);
599 path
= btrfs_alloc_path();
603 extent_root
= btrfs_extent_root(fs_info
, max_t(u64
, block_group
->start
,
604 BTRFS_SUPER_INFO_OFFSET
));
606 path
->skip_locking
= 1;
607 path
->search_commit_root
= 1;
608 path
->reada
= READA_FORWARD
;
610 search_offset
= index
* div_u64(block_group
->length
, max_index
);
611 search_key
.objectid
= block_group
->start
+ search_offset
;
612 search_key
.type
= BTRFS_EXTENT_ITEM_KEY
;
613 search_key
.offset
= 0;
615 btrfs_for_each_slot(extent_root
, &search_key
, found_key
, path
, ret
) {
616 /* Success; sampled an extent item in the block group */
617 if (found_key
->type
== BTRFS_EXTENT_ITEM_KEY
&&
618 found_key
->objectid
>= block_group
->start
&&
619 found_key
->objectid
+ found_key
->offset
<= search_end
)
622 /* We can't possibly find a valid extent item anymore */
623 if (found_key
->objectid
>= search_end
) {
629 lockdep_assert_held(&caching_ctl
->mutex
);
630 lockdep_assert_held_read(&fs_info
->commit_root_sem
);
631 btrfs_free_path(path
);
636 * Best effort attempt to compute a block group's size class while caching it.
638 * @block_group: the block group we are caching
640 * We cannot infer the size class while adding free space extents, because that
641 * logic doesn't care about contiguous file extents (it doesn't differentiate
642 * between a 100M extent and 100 contiguous 1M extents). So we need to read the
643 * file extent items. Reading all of them is quite wasteful, because usually
644 * only a handful are enough to give a good answer. Therefore, we just grab 5 of
645 * them at even steps through the block group and pick the smallest size class
646 * we see. Since size class is best effort, and not guaranteed in general,
647 * inaccuracy is acceptable.
649 * To be more explicit about why this algorithm makes sense:
651 * If we are caching in a block group from disk, then there are three major cases
653 * 1. the block group is well behaved and all extents in it are the same size
655 * 2. the block group is mostly one size class with rare exceptions for last
657 * 3. the block group was populated before size classes and can have a totally
658 * arbitrary mix of size classes.
660 * In case 1, looking at any extent in the block group will yield the correct
661 * result. For the mixed cases, taking the minimum size class seems like a good
662 * approximation, since gaps from frees will be usable to the size class. For
663 * 2., a small handful of file extents is likely to yield the right answer. For
664 * 3, we can either read every file extent, or admit that this is best effort
665 * anyway and try to stay fast.
667 * Returns: 0 on success, negative error code on error.
669 static int load_block_group_size_class(struct btrfs_caching_control
*caching_ctl
,
670 struct btrfs_block_group
*block_group
)
672 struct btrfs_fs_info
*fs_info
= block_group
->fs_info
;
673 struct btrfs_key key
;
675 u64 min_size
= block_group
->length
;
676 enum btrfs_block_group_size_class size_class
= BTRFS_BG_SZ_NONE
;
679 if (!btrfs_block_group_should_use_size_class(block_group
))
682 lockdep_assert_held(&caching_ctl
->mutex
);
683 lockdep_assert_held_read(&fs_info
->commit_root_sem
);
684 for (i
= 0; i
< 5; ++i
) {
685 ret
= sample_block_group_extent_item(caching_ctl
, block_group
, i
, 5, &key
);
690 min_size
= min_t(u64
, min_size
, key
.offset
);
691 size_class
= btrfs_calc_block_group_size_class(min_size
);
693 if (size_class
!= BTRFS_BG_SZ_NONE
) {
694 spin_lock(&block_group
->lock
);
695 block_group
->size_class
= size_class
;
696 spin_unlock(&block_group
->lock
);
702 static int load_extent_tree_free(struct btrfs_caching_control
*caching_ctl
)
704 struct btrfs_block_group
*block_group
= caching_ctl
->block_group
;
705 struct btrfs_fs_info
*fs_info
= block_group
->fs_info
;
706 struct btrfs_root
*extent_root
;
707 struct btrfs_path
*path
;
708 struct extent_buffer
*leaf
;
709 struct btrfs_key key
;
716 path
= btrfs_alloc_path();
720 last
= max_t(u64
, block_group
->start
, BTRFS_SUPER_INFO_OFFSET
);
721 extent_root
= btrfs_extent_root(fs_info
, last
);
723 #ifdef CONFIG_BTRFS_DEBUG
725 * If we're fragmenting we don't want to make anybody think we can
726 * allocate from this block group until we've had a chance to fragment
729 if (btrfs_should_fragment_free_space(block_group
))
733 * We don't want to deadlock with somebody trying to allocate a new
734 * extent for the extent root while also trying to search the extent
735 * root to add free space. So we skip locking and search the commit
736 * root, since its read-only
738 path
->skip_locking
= 1;
739 path
->search_commit_root
= 1;
740 path
->reada
= READA_FORWARD
;
744 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
747 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
751 leaf
= path
->nodes
[0];
752 nritems
= btrfs_header_nritems(leaf
);
755 if (btrfs_fs_closing(fs_info
) > 1) {
760 if (path
->slots
[0] < nritems
) {
761 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
763 ret
= btrfs_find_next_key(extent_root
, path
, &key
, 0, 0);
767 if (need_resched() ||
768 rwsem_is_contended(&fs_info
->commit_root_sem
)) {
769 btrfs_release_path(path
);
770 up_read(&fs_info
->commit_root_sem
);
771 mutex_unlock(&caching_ctl
->mutex
);
773 mutex_lock(&caching_ctl
->mutex
);
774 down_read(&fs_info
->commit_root_sem
);
778 ret
= btrfs_next_leaf(extent_root
, path
);
783 leaf
= path
->nodes
[0];
784 nritems
= btrfs_header_nritems(leaf
);
788 if (key
.objectid
< last
) {
791 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
792 btrfs_release_path(path
);
796 if (key
.objectid
< block_group
->start
) {
801 if (key
.objectid
>= block_group
->start
+ block_group
->length
)
804 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
||
805 key
.type
== BTRFS_METADATA_ITEM_KEY
) {
808 ret
= btrfs_add_new_free_space(block_group
, last
,
809 key
.objectid
, &space_added
);
812 total_found
+= space_added
;
813 if (key
.type
== BTRFS_METADATA_ITEM_KEY
)
814 last
= key
.objectid
+
817 last
= key
.objectid
+ key
.offset
;
819 if (total_found
> CACHING_CTL_WAKE_UP
) {
822 atomic_inc(&caching_ctl
->progress
);
823 wake_up(&caching_ctl
->wait
);
830 ret
= btrfs_add_new_free_space(block_group
, last
,
831 block_group
->start
+ block_group
->length
,
834 btrfs_free_path(path
);
838 static inline void btrfs_free_excluded_extents(const struct btrfs_block_group
*bg
)
840 clear_extent_bits(&bg
->fs_info
->excluded_extents
, bg
->start
,
841 bg
->start
+ bg
->length
- 1, EXTENT_UPTODATE
);
844 static noinline
void caching_thread(struct btrfs_work
*work
)
846 struct btrfs_block_group
*block_group
;
847 struct btrfs_fs_info
*fs_info
;
848 struct btrfs_caching_control
*caching_ctl
;
851 caching_ctl
= container_of(work
, struct btrfs_caching_control
, work
);
852 block_group
= caching_ctl
->block_group
;
853 fs_info
= block_group
->fs_info
;
855 mutex_lock(&caching_ctl
->mutex
);
856 down_read(&fs_info
->commit_root_sem
);
858 load_block_group_size_class(caching_ctl
, block_group
);
859 if (btrfs_test_opt(fs_info
, SPACE_CACHE
)) {
860 ret
= load_free_space_cache(block_group
);
867 * We failed to load the space cache, set ourselves to
868 * CACHE_STARTED and carry on.
870 spin_lock(&block_group
->lock
);
871 block_group
->cached
= BTRFS_CACHE_STARTED
;
872 spin_unlock(&block_group
->lock
);
873 wake_up(&caching_ctl
->wait
);
877 * If we are in the transaction that populated the free space tree we
878 * can't actually cache from the free space tree as our commit root and
879 * real root are the same, so we could change the contents of the blocks
880 * while caching. Instead do the slow caching in this case, and after
881 * the transaction has committed we will be safe.
883 if (btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
) &&
884 !(test_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED
, &fs_info
->flags
)))
885 ret
= load_free_space_tree(caching_ctl
);
887 ret
= load_extent_tree_free(caching_ctl
);
889 spin_lock(&block_group
->lock
);
890 block_group
->caching_ctl
= NULL
;
891 block_group
->cached
= ret
? BTRFS_CACHE_ERROR
: BTRFS_CACHE_FINISHED
;
892 spin_unlock(&block_group
->lock
);
894 #ifdef CONFIG_BTRFS_DEBUG
895 if (btrfs_should_fragment_free_space(block_group
)) {
898 spin_lock(&block_group
->space_info
->lock
);
899 spin_lock(&block_group
->lock
);
900 bytes_used
= block_group
->length
- block_group
->used
;
901 block_group
->space_info
->bytes_used
+= bytes_used
>> 1;
902 spin_unlock(&block_group
->lock
);
903 spin_unlock(&block_group
->space_info
->lock
);
904 fragment_free_space(block_group
);
908 up_read(&fs_info
->commit_root_sem
);
909 btrfs_free_excluded_extents(block_group
);
910 mutex_unlock(&caching_ctl
->mutex
);
912 wake_up(&caching_ctl
->wait
);
914 btrfs_put_caching_control(caching_ctl
);
915 btrfs_put_block_group(block_group
);
918 int btrfs_cache_block_group(struct btrfs_block_group
*cache
, bool wait
)
920 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
921 struct btrfs_caching_control
*caching_ctl
= NULL
;
924 /* Allocator for zoned filesystems does not use the cache at all */
925 if (btrfs_is_zoned(fs_info
))
928 caching_ctl
= kzalloc(sizeof(*caching_ctl
), GFP_NOFS
);
932 INIT_LIST_HEAD(&caching_ctl
->list
);
933 mutex_init(&caching_ctl
->mutex
);
934 init_waitqueue_head(&caching_ctl
->wait
);
935 caching_ctl
->block_group
= cache
;
936 refcount_set(&caching_ctl
->count
, 2);
937 atomic_set(&caching_ctl
->progress
, 0);
938 btrfs_init_work(&caching_ctl
->work
, caching_thread
, NULL
);
940 spin_lock(&cache
->lock
);
941 if (cache
->cached
!= BTRFS_CACHE_NO
) {
944 caching_ctl
= cache
->caching_ctl
;
946 refcount_inc(&caching_ctl
->count
);
947 spin_unlock(&cache
->lock
);
950 WARN_ON(cache
->caching_ctl
);
951 cache
->caching_ctl
= caching_ctl
;
952 cache
->cached
= BTRFS_CACHE_STARTED
;
953 spin_unlock(&cache
->lock
);
955 write_lock(&fs_info
->block_group_cache_lock
);
956 refcount_inc(&caching_ctl
->count
);
957 list_add_tail(&caching_ctl
->list
, &fs_info
->caching_block_groups
);
958 write_unlock(&fs_info
->block_group_cache_lock
);
960 btrfs_get_block_group(cache
);
962 btrfs_queue_work(fs_info
->caching_workers
, &caching_ctl
->work
);
964 if (wait
&& caching_ctl
)
965 ret
= btrfs_caching_ctl_wait_done(cache
, caching_ctl
);
967 btrfs_put_caching_control(caching_ctl
);
972 static void clear_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
974 u64 extra_flags
= chunk_to_extended(flags
) &
975 BTRFS_EXTENDED_PROFILE_MASK
;
977 write_seqlock(&fs_info
->profiles_lock
);
978 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
979 fs_info
->avail_data_alloc_bits
&= ~extra_flags
;
980 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
981 fs_info
->avail_metadata_alloc_bits
&= ~extra_flags
;
982 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
983 fs_info
->avail_system_alloc_bits
&= ~extra_flags
;
984 write_sequnlock(&fs_info
->profiles_lock
);
988 * Clear incompat bits for the following feature(s):
990 * - RAID56 - in case there's neither RAID5 nor RAID6 profile block group
991 * in the whole filesystem
993 * - RAID1C34 - same as above for RAID1C3 and RAID1C4 block groups
995 static void clear_incompat_bg_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
997 bool found_raid56
= false;
998 bool found_raid1c34
= false;
1000 if ((flags
& BTRFS_BLOCK_GROUP_RAID56_MASK
) ||
1001 (flags
& BTRFS_BLOCK_GROUP_RAID1C3
) ||
1002 (flags
& BTRFS_BLOCK_GROUP_RAID1C4
)) {
1003 struct list_head
*head
= &fs_info
->space_info
;
1004 struct btrfs_space_info
*sinfo
;
1006 list_for_each_entry_rcu(sinfo
, head
, list
) {
1007 down_read(&sinfo
->groups_sem
);
1008 if (!list_empty(&sinfo
->block_groups
[BTRFS_RAID_RAID5
]))
1009 found_raid56
= true;
1010 if (!list_empty(&sinfo
->block_groups
[BTRFS_RAID_RAID6
]))
1011 found_raid56
= true;
1012 if (!list_empty(&sinfo
->block_groups
[BTRFS_RAID_RAID1C3
]))
1013 found_raid1c34
= true;
1014 if (!list_empty(&sinfo
->block_groups
[BTRFS_RAID_RAID1C4
]))
1015 found_raid1c34
= true;
1016 up_read(&sinfo
->groups_sem
);
1019 btrfs_clear_fs_incompat(fs_info
, RAID56
);
1020 if (!found_raid1c34
)
1021 btrfs_clear_fs_incompat(fs_info
, RAID1C34
);
1025 static int remove_block_group_item(struct btrfs_trans_handle
*trans
,
1026 struct btrfs_path
*path
,
1027 struct btrfs_block_group
*block_group
)
1029 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
1030 struct btrfs_root
*root
;
1031 struct btrfs_key key
;
1034 root
= btrfs_block_group_root(fs_info
);
1035 key
.objectid
= block_group
->start
;
1036 key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
1037 key
.offset
= block_group
->length
;
1039 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1045 ret
= btrfs_del_item(trans
, root
, path
);
1049 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
1050 u64 group_start
, struct extent_map
*em
)
1052 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
1053 struct btrfs_path
*path
;
1054 struct btrfs_block_group
*block_group
;
1055 struct btrfs_free_cluster
*cluster
;
1056 struct inode
*inode
;
1057 struct kobject
*kobj
= NULL
;
1061 struct btrfs_caching_control
*caching_ctl
= NULL
;
1063 bool remove_rsv
= false;
1065 block_group
= btrfs_lookup_block_group(fs_info
, group_start
);
1066 BUG_ON(!block_group
);
1067 BUG_ON(!block_group
->ro
);
1069 trace_btrfs_remove_block_group(block_group
);
1071 * Free the reserved super bytes from this block group before
1074 btrfs_free_excluded_extents(block_group
);
1075 btrfs_free_ref_tree_range(fs_info
, block_group
->start
,
1076 block_group
->length
);
1078 index
= btrfs_bg_flags_to_raid_index(block_group
->flags
);
1079 factor
= btrfs_bg_type_to_factor(block_group
->flags
);
1081 /* make sure this block group isn't part of an allocation cluster */
1082 cluster
= &fs_info
->data_alloc_cluster
;
1083 spin_lock(&cluster
->refill_lock
);
1084 btrfs_return_cluster_to_free_space(block_group
, cluster
);
1085 spin_unlock(&cluster
->refill_lock
);
1088 * make sure this block group isn't part of a metadata
1089 * allocation cluster
1091 cluster
= &fs_info
->meta_alloc_cluster
;
1092 spin_lock(&cluster
->refill_lock
);
1093 btrfs_return_cluster_to_free_space(block_group
, cluster
);
1094 spin_unlock(&cluster
->refill_lock
);
1096 btrfs_clear_treelog_bg(block_group
);
1097 btrfs_clear_data_reloc_bg(block_group
);
1099 path
= btrfs_alloc_path();
1106 * get the inode first so any iput calls done for the io_list
1107 * aren't the final iput (no unlinks allowed now)
1109 inode
= lookup_free_space_inode(block_group
, path
);
1111 mutex_lock(&trans
->transaction
->cache_write_mutex
);
1113 * Make sure our free space cache IO is done before removing the
1116 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
1117 if (!list_empty(&block_group
->io_list
)) {
1118 list_del_init(&block_group
->io_list
);
1120 WARN_ON(!IS_ERR(inode
) && inode
!= block_group
->io_ctl
.inode
);
1122 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
1123 btrfs_wait_cache_io(trans
, block_group
, path
);
1124 btrfs_put_block_group(block_group
);
1125 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
1128 if (!list_empty(&block_group
->dirty_list
)) {
1129 list_del_init(&block_group
->dirty_list
);
1131 btrfs_put_block_group(block_group
);
1133 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
1134 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
1136 ret
= btrfs_remove_free_space_inode(trans
, inode
, block_group
);
1140 write_lock(&fs_info
->block_group_cache_lock
);
1141 rb_erase_cached(&block_group
->cache_node
,
1142 &fs_info
->block_group_cache_tree
);
1143 RB_CLEAR_NODE(&block_group
->cache_node
);
1145 /* Once for the block groups rbtree */
1146 btrfs_put_block_group(block_group
);
1148 write_unlock(&fs_info
->block_group_cache_lock
);
1150 down_write(&block_group
->space_info
->groups_sem
);
1152 * we must use list_del_init so people can check to see if they
1153 * are still on the list after taking the semaphore
1155 list_del_init(&block_group
->list
);
1156 if (list_empty(&block_group
->space_info
->block_groups
[index
])) {
1157 kobj
= block_group
->space_info
->block_group_kobjs
[index
];
1158 block_group
->space_info
->block_group_kobjs
[index
] = NULL
;
1159 clear_avail_alloc_bits(fs_info
, block_group
->flags
);
1161 up_write(&block_group
->space_info
->groups_sem
);
1162 clear_incompat_bg_bits(fs_info
, block_group
->flags
);
1168 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
1169 btrfs_wait_block_group_cache_done(block_group
);
1171 write_lock(&fs_info
->block_group_cache_lock
);
1172 caching_ctl
= btrfs_get_caching_control(block_group
);
1174 struct btrfs_caching_control
*ctl
;
1176 list_for_each_entry(ctl
, &fs_info
->caching_block_groups
, list
) {
1177 if (ctl
->block_group
== block_group
) {
1179 refcount_inc(&caching_ctl
->count
);
1185 list_del_init(&caching_ctl
->list
);
1186 write_unlock(&fs_info
->block_group_cache_lock
);
1189 /* Once for the caching bgs list and once for us. */
1190 btrfs_put_caching_control(caching_ctl
);
1191 btrfs_put_caching_control(caching_ctl
);
1194 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
1195 WARN_ON(!list_empty(&block_group
->dirty_list
));
1196 WARN_ON(!list_empty(&block_group
->io_list
));
1197 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
1199 btrfs_remove_free_space_cache(block_group
);
1201 spin_lock(&block_group
->space_info
->lock
);
1202 list_del_init(&block_group
->ro_list
);
1204 if (btrfs_test_opt(fs_info
, ENOSPC_DEBUG
)) {
1205 WARN_ON(block_group
->space_info
->total_bytes
1206 < block_group
->length
);
1207 WARN_ON(block_group
->space_info
->bytes_readonly
1208 < block_group
->length
- block_group
->zone_unusable
);
1209 WARN_ON(block_group
->space_info
->bytes_zone_unusable
1210 < block_group
->zone_unusable
);
1211 WARN_ON(block_group
->space_info
->disk_total
1212 < block_group
->length
* factor
);
1214 block_group
->space_info
->total_bytes
-= block_group
->length
;
1215 block_group
->space_info
->bytes_readonly
-=
1216 (block_group
->length
- block_group
->zone_unusable
);
1217 block_group
->space_info
->bytes_zone_unusable
-=
1218 block_group
->zone_unusable
;
1219 block_group
->space_info
->disk_total
-= block_group
->length
* factor
;
1221 spin_unlock(&block_group
->space_info
->lock
);
1224 * Remove the free space for the block group from the free space tree
1225 * and the block group's item from the extent tree before marking the
1226 * block group as removed. This is to prevent races with tasks that
1227 * freeze and unfreeze a block group, this task and another task
1228 * allocating a new block group - the unfreeze task ends up removing
1229 * the block group's extent map before the task calling this function
1230 * deletes the block group item from the extent tree, allowing for
1231 * another task to attempt to create another block group with the same
1232 * item key (and failing with -EEXIST and a transaction abort).
1234 ret
= remove_block_group_free_space(trans
, block_group
);
1238 ret
= remove_block_group_item(trans
, path
, block_group
);
1242 spin_lock(&block_group
->lock
);
1243 set_bit(BLOCK_GROUP_FLAG_REMOVED
, &block_group
->runtime_flags
);
1246 * At this point trimming or scrub can't start on this block group,
1247 * because we removed the block group from the rbtree
1248 * fs_info->block_group_cache_tree so no one can't find it anymore and
1249 * even if someone already got this block group before we removed it
1250 * from the rbtree, they have already incremented block_group->frozen -
1251 * if they didn't, for the trimming case they won't find any free space
1252 * entries because we already removed them all when we called
1253 * btrfs_remove_free_space_cache().
1255 * And we must not remove the extent map from the fs_info->mapping_tree
1256 * to prevent the same logical address range and physical device space
1257 * ranges from being reused for a new block group. This is needed to
1258 * avoid races with trimming and scrub.
1260 * An fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
1261 * completely transactionless, so while it is trimming a range the
1262 * currently running transaction might finish and a new one start,
1263 * allowing for new block groups to be created that can reuse the same
1264 * physical device locations unless we take this special care.
1266 * There may also be an implicit trim operation if the file system
1267 * is mounted with -odiscard. The same protections must remain
1268 * in place until the extents have been discarded completely when
1269 * the transaction commit has completed.
1271 remove_em
= (atomic_read(&block_group
->frozen
) == 0);
1272 spin_unlock(&block_group
->lock
);
1275 struct extent_map_tree
*em_tree
;
1277 em_tree
= &fs_info
->mapping_tree
;
1278 write_lock(&em_tree
->lock
);
1279 remove_extent_mapping(em_tree
, em
);
1280 write_unlock(&em_tree
->lock
);
1281 /* once for the tree */
1282 free_extent_map(em
);
1286 /* Once for the lookup reference */
1287 btrfs_put_block_group(block_group
);
1289 btrfs_dec_delayed_refs_rsv_bg_updates(fs_info
);
1290 btrfs_free_path(path
);
1294 struct btrfs_trans_handle
*btrfs_start_trans_remove_block_group(
1295 struct btrfs_fs_info
*fs_info
, const u64 chunk_offset
)
1297 struct btrfs_root
*root
= btrfs_block_group_root(fs_info
);
1298 struct extent_map_tree
*em_tree
= &fs_info
->mapping_tree
;
1299 struct extent_map
*em
;
1300 struct map_lookup
*map
;
1301 unsigned int num_items
;
1303 read_lock(&em_tree
->lock
);
1304 em
= lookup_extent_mapping(em_tree
, chunk_offset
, 1);
1305 read_unlock(&em_tree
->lock
);
1306 ASSERT(em
&& em
->start
== chunk_offset
);
1309 * We need to reserve 3 + N units from the metadata space info in order
1310 * to remove a block group (done at btrfs_remove_chunk() and at
1311 * btrfs_remove_block_group()), which are used for:
1313 * 1 unit for adding the free space inode's orphan (located in the tree
1315 * 1 unit for deleting the block group item (located in the extent
1317 * 1 unit for deleting the free space item (located in tree of tree
1319 * N units for deleting N device extent items corresponding to each
1320 * stripe (located in the device tree).
1322 * In order to remove a block group we also need to reserve units in the
1323 * system space info in order to update the chunk tree (update one or
1324 * more device items and remove one chunk item), but this is done at
1325 * btrfs_remove_chunk() through a call to check_system_chunk().
1327 map
= em
->map_lookup
;
1328 num_items
= 3 + map
->num_stripes
;
1329 free_extent_map(em
);
1331 return btrfs_start_transaction_fallback_global_rsv(root
, num_items
);
1335 * Mark block group @cache read-only, so later write won't happen to block
1338 * If @force is not set, this function will only mark the block group readonly
1339 * if we have enough free space (1M) in other metadata/system block groups.
1340 * If @force is not set, this function will mark the block group readonly
1341 * without checking free space.
1343 * NOTE: This function doesn't care if other block groups can contain all the
1344 * data in this block group. That check should be done by relocation routine,
1345 * not this function.
1347 static int inc_block_group_ro(struct btrfs_block_group
*cache
, int force
)
1349 struct btrfs_space_info
*sinfo
= cache
->space_info
;
1353 spin_lock(&sinfo
->lock
);
1354 spin_lock(&cache
->lock
);
1356 if (cache
->swap_extents
) {
1367 num_bytes
= cache
->length
- cache
->reserved
- cache
->pinned
-
1368 cache
->bytes_super
- cache
->zone_unusable
- cache
->used
;
1371 * Data never overcommits, even in mixed mode, so do just the straight
1372 * check of left over space in how much we have allocated.
1376 } else if (sinfo
->flags
& BTRFS_BLOCK_GROUP_DATA
) {
1377 u64 sinfo_used
= btrfs_space_info_used(sinfo
, true);
1380 * Here we make sure if we mark this bg RO, we still have enough
1381 * free space as buffer.
1383 if (sinfo_used
+ num_bytes
<= sinfo
->total_bytes
)
1387 * We overcommit metadata, so we need to do the
1388 * btrfs_can_overcommit check here, and we need to pass in
1389 * BTRFS_RESERVE_NO_FLUSH to give ourselves the most amount of
1390 * leeway to allow us to mark this block group as read only.
1392 if (btrfs_can_overcommit(cache
->fs_info
, sinfo
, num_bytes
,
1393 BTRFS_RESERVE_NO_FLUSH
))
1398 sinfo
->bytes_readonly
+= num_bytes
;
1399 if (btrfs_is_zoned(cache
->fs_info
)) {
1400 /* Migrate zone_unusable bytes to readonly */
1401 sinfo
->bytes_readonly
+= cache
->zone_unusable
;
1402 sinfo
->bytes_zone_unusable
-= cache
->zone_unusable
;
1403 cache
->zone_unusable
= 0;
1406 list_add_tail(&cache
->ro_list
, &sinfo
->ro_bgs
);
1409 spin_unlock(&cache
->lock
);
1410 spin_unlock(&sinfo
->lock
);
1411 if (ret
== -ENOSPC
&& btrfs_test_opt(cache
->fs_info
, ENOSPC_DEBUG
)) {
1412 btrfs_info(cache
->fs_info
,
1413 "unable to make block group %llu ro", cache
->start
);
1414 btrfs_dump_space_info(cache
->fs_info
, cache
->space_info
, 0, 0);
1419 static bool clean_pinned_extents(struct btrfs_trans_handle
*trans
,
1420 struct btrfs_block_group
*bg
)
1422 struct btrfs_fs_info
*fs_info
= bg
->fs_info
;
1423 struct btrfs_transaction
*prev_trans
= NULL
;
1424 const u64 start
= bg
->start
;
1425 const u64 end
= start
+ bg
->length
- 1;
1428 spin_lock(&fs_info
->trans_lock
);
1429 if (trans
->transaction
->list
.prev
!= &fs_info
->trans_list
) {
1430 prev_trans
= list_last_entry(&trans
->transaction
->list
,
1431 struct btrfs_transaction
, list
);
1432 refcount_inc(&prev_trans
->use_count
);
1434 spin_unlock(&fs_info
->trans_lock
);
1437 * Hold the unused_bg_unpin_mutex lock to avoid racing with
1438 * btrfs_finish_extent_commit(). If we are at transaction N, another
1439 * task might be running finish_extent_commit() for the previous
1440 * transaction N - 1, and have seen a range belonging to the block
1441 * group in pinned_extents before we were able to clear the whole block
1442 * group range from pinned_extents. This means that task can lookup for
1443 * the block group after we unpinned it from pinned_extents and removed
1444 * it, leading to a BUG_ON() at unpin_extent_range().
1446 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
1448 ret
= clear_extent_bits(&prev_trans
->pinned_extents
, start
, end
,
1454 ret
= clear_extent_bits(&trans
->transaction
->pinned_extents
, start
, end
,
1457 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
1459 btrfs_put_transaction(prev_trans
);
1465 * Process the unused_bgs list and remove any that don't have any allocated
1466 * space inside of them.
1468 void btrfs_delete_unused_bgs(struct btrfs_fs_info
*fs_info
)
1470 struct btrfs_block_group
*block_group
;
1471 struct btrfs_space_info
*space_info
;
1472 struct btrfs_trans_handle
*trans
;
1473 const bool async_trim_enabled
= btrfs_test_opt(fs_info
, DISCARD_ASYNC
);
1476 if (!test_bit(BTRFS_FS_OPEN
, &fs_info
->flags
))
1479 if (btrfs_fs_closing(fs_info
))
1483 * Long running balances can keep us blocked here for eternity, so
1484 * simply skip deletion if we're unable to get the mutex.
1486 if (!mutex_trylock(&fs_info
->reclaim_bgs_lock
))
1489 spin_lock(&fs_info
->unused_bgs_lock
);
1490 while (!list_empty(&fs_info
->unused_bgs
)) {
1493 block_group
= list_first_entry(&fs_info
->unused_bgs
,
1494 struct btrfs_block_group
,
1496 list_del_init(&block_group
->bg_list
);
1498 space_info
= block_group
->space_info
;
1500 if (ret
|| btrfs_mixed_space_info(space_info
)) {
1501 btrfs_put_block_group(block_group
);
1504 spin_unlock(&fs_info
->unused_bgs_lock
);
1506 btrfs_discard_cancel_work(&fs_info
->discard_ctl
, block_group
);
1508 /* Don't want to race with allocators so take the groups_sem */
1509 down_write(&space_info
->groups_sem
);
1512 * Async discard moves the final block group discard to be prior
1513 * to the unused_bgs code path. Therefore, if it's not fully
1514 * trimmed, punt it back to the async discard lists.
1516 if (btrfs_test_opt(fs_info
, DISCARD_ASYNC
) &&
1517 !btrfs_is_free_space_trimmed(block_group
)) {
1518 trace_btrfs_skip_unused_block_group(block_group
);
1519 up_write(&space_info
->groups_sem
);
1520 /* Requeue if we failed because of async discard */
1521 btrfs_discard_queue_work(&fs_info
->discard_ctl
,
1526 spin_lock(&block_group
->lock
);
1527 if (block_group
->reserved
|| block_group
->pinned
||
1528 block_group
->used
|| block_group
->ro
||
1529 list_is_singular(&block_group
->list
)) {
1531 * We want to bail if we made new allocations or have
1532 * outstanding allocations in this block group. We do
1533 * the ro check in case balance is currently acting on
1536 trace_btrfs_skip_unused_block_group(block_group
);
1537 spin_unlock(&block_group
->lock
);
1538 up_write(&space_info
->groups_sem
);
1541 spin_unlock(&block_group
->lock
);
1543 /* We don't want to force the issue, only flip if it's ok. */
1544 ret
= inc_block_group_ro(block_group
, 0);
1545 up_write(&space_info
->groups_sem
);
1551 ret
= btrfs_zone_finish(block_group
);
1553 btrfs_dec_block_group_ro(block_group
);
1560 * Want to do this before we do anything else so we can recover
1561 * properly if we fail to join the transaction.
1563 trans
= btrfs_start_trans_remove_block_group(fs_info
,
1564 block_group
->start
);
1565 if (IS_ERR(trans
)) {
1566 btrfs_dec_block_group_ro(block_group
);
1567 ret
= PTR_ERR(trans
);
1572 * We could have pending pinned extents for this block group,
1573 * just delete them, we don't care about them anymore.
1575 if (!clean_pinned_extents(trans
, block_group
)) {
1576 btrfs_dec_block_group_ro(block_group
);
1581 * At this point, the block_group is read only and should fail
1582 * new allocations. However, btrfs_finish_extent_commit() can
1583 * cause this block_group to be placed back on the discard
1584 * lists because now the block_group isn't fully discarded.
1585 * Bail here and try again later after discarding everything.
1587 spin_lock(&fs_info
->discard_ctl
.lock
);
1588 if (!list_empty(&block_group
->discard_list
)) {
1589 spin_unlock(&fs_info
->discard_ctl
.lock
);
1590 btrfs_dec_block_group_ro(block_group
);
1591 btrfs_discard_queue_work(&fs_info
->discard_ctl
,
1595 spin_unlock(&fs_info
->discard_ctl
.lock
);
1597 /* Reset pinned so btrfs_put_block_group doesn't complain */
1598 spin_lock(&space_info
->lock
);
1599 spin_lock(&block_group
->lock
);
1601 btrfs_space_info_update_bytes_pinned(fs_info
, space_info
,
1602 -block_group
->pinned
);
1603 space_info
->bytes_readonly
+= block_group
->pinned
;
1604 block_group
->pinned
= 0;
1606 spin_unlock(&block_group
->lock
);
1607 spin_unlock(&space_info
->lock
);
1610 * The normal path here is an unused block group is passed here,
1611 * then trimming is handled in the transaction commit path.
1612 * Async discard interposes before this to do the trimming
1613 * before coming down the unused block group path as trimming
1614 * will no longer be done later in the transaction commit path.
1616 if (!async_trim_enabled
&& btrfs_test_opt(fs_info
, DISCARD_ASYNC
))
1620 * DISCARD can flip during remount. On zoned filesystems, we
1621 * need to reset sequential-required zones.
1623 trimming
= btrfs_test_opt(fs_info
, DISCARD_SYNC
) ||
1624 btrfs_is_zoned(fs_info
);
1626 /* Implicit trim during transaction commit. */
1628 btrfs_freeze_block_group(block_group
);
1631 * Btrfs_remove_chunk will abort the transaction if things go
1634 ret
= btrfs_remove_chunk(trans
, block_group
->start
);
1638 btrfs_unfreeze_block_group(block_group
);
1643 * If we're not mounted with -odiscard, we can just forget
1644 * about this block group. Otherwise we'll need to wait
1645 * until transaction commit to do the actual discard.
1648 spin_lock(&fs_info
->unused_bgs_lock
);
1650 * A concurrent scrub might have added us to the list
1651 * fs_info->unused_bgs, so use a list_move operation
1652 * to add the block group to the deleted_bgs list.
1654 list_move(&block_group
->bg_list
,
1655 &trans
->transaction
->deleted_bgs
);
1656 spin_unlock(&fs_info
->unused_bgs_lock
);
1657 btrfs_get_block_group(block_group
);
1660 btrfs_end_transaction(trans
);
1662 btrfs_put_block_group(block_group
);
1663 spin_lock(&fs_info
->unused_bgs_lock
);
1665 spin_unlock(&fs_info
->unused_bgs_lock
);
1666 mutex_unlock(&fs_info
->reclaim_bgs_lock
);
1670 btrfs_end_transaction(trans
);
1671 mutex_unlock(&fs_info
->reclaim_bgs_lock
);
1672 btrfs_put_block_group(block_group
);
1673 btrfs_discard_punt_unused_bgs_list(fs_info
);
1676 void btrfs_mark_bg_unused(struct btrfs_block_group
*bg
)
1678 struct btrfs_fs_info
*fs_info
= bg
->fs_info
;
1680 spin_lock(&fs_info
->unused_bgs_lock
);
1681 if (list_empty(&bg
->bg_list
)) {
1682 btrfs_get_block_group(bg
);
1683 trace_btrfs_add_unused_block_group(bg
);
1684 list_add_tail(&bg
->bg_list
, &fs_info
->unused_bgs
);
1685 } else if (!test_bit(BLOCK_GROUP_FLAG_NEW
, &bg
->runtime_flags
)) {
1686 /* Pull out the block group from the reclaim_bgs list. */
1687 trace_btrfs_add_unused_block_group(bg
);
1688 list_move_tail(&bg
->bg_list
, &fs_info
->unused_bgs
);
1690 spin_unlock(&fs_info
->unused_bgs_lock
);
1694 * We want block groups with a low number of used bytes to be in the beginning
1695 * of the list, so they will get reclaimed first.
1697 static int reclaim_bgs_cmp(void *unused
, const struct list_head
*a
,
1698 const struct list_head
*b
)
1700 const struct btrfs_block_group
*bg1
, *bg2
;
1702 bg1
= list_entry(a
, struct btrfs_block_group
, bg_list
);
1703 bg2
= list_entry(b
, struct btrfs_block_group
, bg_list
);
1705 return bg1
->used
> bg2
->used
;
1708 static inline bool btrfs_should_reclaim(struct btrfs_fs_info
*fs_info
)
1710 if (btrfs_is_zoned(fs_info
))
1711 return btrfs_zoned_should_reclaim(fs_info
);
1715 static bool should_reclaim_block_group(struct btrfs_block_group
*bg
, u64 bytes_freed
)
1717 const struct btrfs_space_info
*space_info
= bg
->space_info
;
1718 const int reclaim_thresh
= READ_ONCE(space_info
->bg_reclaim_threshold
);
1719 const u64 new_val
= bg
->used
;
1720 const u64 old_val
= new_val
+ bytes_freed
;
1723 if (reclaim_thresh
== 0)
1726 thresh
= mult_perc(bg
->length
, reclaim_thresh
);
1729 * If we were below the threshold before don't reclaim, we are likely a
1730 * brand new block group and we don't want to relocate new block groups.
1732 if (old_val
< thresh
)
1734 if (new_val
>= thresh
)
1739 void btrfs_reclaim_bgs_work(struct work_struct
*work
)
1741 struct btrfs_fs_info
*fs_info
=
1742 container_of(work
, struct btrfs_fs_info
, reclaim_bgs_work
);
1743 struct btrfs_block_group
*bg
;
1744 struct btrfs_space_info
*space_info
;
1746 if (!test_bit(BTRFS_FS_OPEN
, &fs_info
->flags
))
1749 if (btrfs_fs_closing(fs_info
))
1752 if (!btrfs_should_reclaim(fs_info
))
1755 sb_start_write(fs_info
->sb
);
1757 if (!btrfs_exclop_start(fs_info
, BTRFS_EXCLOP_BALANCE
)) {
1758 sb_end_write(fs_info
->sb
);
1763 * Long running balances can keep us blocked here for eternity, so
1764 * simply skip reclaim if we're unable to get the mutex.
1766 if (!mutex_trylock(&fs_info
->reclaim_bgs_lock
)) {
1767 btrfs_exclop_finish(fs_info
);
1768 sb_end_write(fs_info
->sb
);
1772 spin_lock(&fs_info
->unused_bgs_lock
);
1774 * Sort happens under lock because we can't simply splice it and sort.
1775 * The block groups might still be in use and reachable via bg_list,
1776 * and their presence in the reclaim_bgs list must be preserved.
1778 list_sort(NULL
, &fs_info
->reclaim_bgs
, reclaim_bgs_cmp
);
1779 while (!list_empty(&fs_info
->reclaim_bgs
)) {
1783 bg
= list_first_entry(&fs_info
->reclaim_bgs
,
1784 struct btrfs_block_group
,
1786 list_del_init(&bg
->bg_list
);
1788 space_info
= bg
->space_info
;
1789 spin_unlock(&fs_info
->unused_bgs_lock
);
1791 /* Don't race with allocators so take the groups_sem */
1792 down_write(&space_info
->groups_sem
);
1794 spin_lock(&bg
->lock
);
1795 if (bg
->reserved
|| bg
->pinned
|| bg
->ro
) {
1797 * We want to bail if we made new allocations or have
1798 * outstanding allocations in this block group. We do
1799 * the ro check in case balance is currently acting on
1802 spin_unlock(&bg
->lock
);
1803 up_write(&space_info
->groups_sem
);
1806 if (bg
->used
== 0) {
1808 * It is possible that we trigger relocation on a block
1809 * group as its extents are deleted and it first goes
1810 * below the threshold, then shortly after goes empty.
1812 * In this case, relocating it does delete it, but has
1813 * some overhead in relocation specific metadata, looking
1814 * for the non-existent extents and running some extra
1815 * transactions, which we can avoid by using one of the
1816 * other mechanisms for dealing with empty block groups.
1818 if (!btrfs_test_opt(fs_info
, DISCARD_ASYNC
))
1819 btrfs_mark_bg_unused(bg
);
1820 spin_unlock(&bg
->lock
);
1821 up_write(&space_info
->groups_sem
);
1826 * The block group might no longer meet the reclaim condition by
1827 * the time we get around to reclaiming it, so to avoid
1828 * reclaiming overly full block_groups, skip reclaiming them.
1830 * Since the decision making process also depends on the amount
1831 * being freed, pass in a fake giant value to skip that extra
1832 * check, which is more meaningful when adding to the list in
1835 if (!should_reclaim_block_group(bg
, bg
->length
)) {
1836 spin_unlock(&bg
->lock
);
1837 up_write(&space_info
->groups_sem
);
1840 spin_unlock(&bg
->lock
);
1843 * Get out fast, in case we're read-only or unmounting the
1844 * filesystem. It is OK to drop block groups from the list even
1845 * for the read-only case. As we did sb_start_write(),
1846 * "mount -o remount,ro" won't happen and read-only filesystem
1847 * means it is forced read-only due to a fatal error. So, it
1848 * never gets back to read-write to let us reclaim again.
1850 if (btrfs_need_cleaner_sleep(fs_info
)) {
1851 up_write(&space_info
->groups_sem
);
1856 * Cache the zone_unusable value before turning the block group
1857 * to read only. As soon as the blog group is read only it's
1858 * zone_unusable value gets moved to the block group's read-only
1859 * bytes and isn't available for calculations anymore.
1861 zone_unusable
= bg
->zone_unusable
;
1862 ret
= inc_block_group_ro(bg
, 0);
1863 up_write(&space_info
->groups_sem
);
1868 "reclaiming chunk %llu with %llu%% used %llu%% unusable",
1870 div64_u64(bg
->used
* 100, bg
->length
),
1871 div64_u64(zone_unusable
* 100, bg
->length
));
1872 trace_btrfs_reclaim_block_group(bg
);
1873 ret
= btrfs_relocate_chunk(fs_info
, bg
->start
);
1875 btrfs_dec_block_group_ro(bg
);
1876 btrfs_err(fs_info
, "error relocating chunk %llu",
1882 btrfs_mark_bg_to_reclaim(bg
);
1883 btrfs_put_block_group(bg
);
1885 mutex_unlock(&fs_info
->reclaim_bgs_lock
);
1887 * Reclaiming all the block groups in the list can take really
1888 * long. Prioritize cleaning up unused block groups.
1890 btrfs_delete_unused_bgs(fs_info
);
1892 * If we are interrupted by a balance, we can just bail out. The
1893 * cleaner thread restart again if necessary.
1895 if (!mutex_trylock(&fs_info
->reclaim_bgs_lock
))
1897 spin_lock(&fs_info
->unused_bgs_lock
);
1899 spin_unlock(&fs_info
->unused_bgs_lock
);
1900 mutex_unlock(&fs_info
->reclaim_bgs_lock
);
1902 btrfs_exclop_finish(fs_info
);
1903 sb_end_write(fs_info
->sb
);
1906 void btrfs_reclaim_bgs(struct btrfs_fs_info
*fs_info
)
1908 spin_lock(&fs_info
->unused_bgs_lock
);
1909 if (!list_empty(&fs_info
->reclaim_bgs
))
1910 queue_work(system_unbound_wq
, &fs_info
->reclaim_bgs_work
);
1911 spin_unlock(&fs_info
->unused_bgs_lock
);
1914 void btrfs_mark_bg_to_reclaim(struct btrfs_block_group
*bg
)
1916 struct btrfs_fs_info
*fs_info
= bg
->fs_info
;
1918 spin_lock(&fs_info
->unused_bgs_lock
);
1919 if (list_empty(&bg
->bg_list
)) {
1920 btrfs_get_block_group(bg
);
1921 trace_btrfs_add_reclaim_block_group(bg
);
1922 list_add_tail(&bg
->bg_list
, &fs_info
->reclaim_bgs
);
1924 spin_unlock(&fs_info
->unused_bgs_lock
);
1927 static int read_bg_from_eb(struct btrfs_fs_info
*fs_info
, struct btrfs_key
*key
,
1928 struct btrfs_path
*path
)
1930 struct extent_map_tree
*em_tree
;
1931 struct extent_map
*em
;
1932 struct btrfs_block_group_item bg
;
1933 struct extent_buffer
*leaf
;
1938 slot
= path
->slots
[0];
1939 leaf
= path
->nodes
[0];
1941 em_tree
= &fs_info
->mapping_tree
;
1942 read_lock(&em_tree
->lock
);
1943 em
= lookup_extent_mapping(em_tree
, key
->objectid
, key
->offset
);
1944 read_unlock(&em_tree
->lock
);
1947 "logical %llu len %llu found bg but no related chunk",
1948 key
->objectid
, key
->offset
);
1952 if (em
->start
!= key
->objectid
|| em
->len
!= key
->offset
) {
1954 "block group %llu len %llu mismatch with chunk %llu len %llu",
1955 key
->objectid
, key
->offset
, em
->start
, em
->len
);
1960 read_extent_buffer(leaf
, &bg
, btrfs_item_ptr_offset(leaf
, slot
),
1962 flags
= btrfs_stack_block_group_flags(&bg
) &
1963 BTRFS_BLOCK_GROUP_TYPE_MASK
;
1965 if (flags
!= (em
->map_lookup
->type
& BTRFS_BLOCK_GROUP_TYPE_MASK
)) {
1967 "block group %llu len %llu type flags 0x%llx mismatch with chunk type flags 0x%llx",
1968 key
->objectid
, key
->offset
, flags
,
1969 (BTRFS_BLOCK_GROUP_TYPE_MASK
& em
->map_lookup
->type
));
1974 free_extent_map(em
);
1978 static int find_first_block_group(struct btrfs_fs_info
*fs_info
,
1979 struct btrfs_path
*path
,
1980 struct btrfs_key
*key
)
1982 struct btrfs_root
*root
= btrfs_block_group_root(fs_info
);
1984 struct btrfs_key found_key
;
1986 btrfs_for_each_slot(root
, key
, &found_key
, path
, ret
) {
1987 if (found_key
.objectid
>= key
->objectid
&&
1988 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
1989 return read_bg_from_eb(fs_info
, &found_key
, path
);
1995 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
1997 u64 extra_flags
= chunk_to_extended(flags
) &
1998 BTRFS_EXTENDED_PROFILE_MASK
;
2000 write_seqlock(&fs_info
->profiles_lock
);
2001 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
2002 fs_info
->avail_data_alloc_bits
|= extra_flags
;
2003 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
2004 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
2005 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
2006 fs_info
->avail_system_alloc_bits
|= extra_flags
;
2007 write_sequnlock(&fs_info
->profiles_lock
);
2011 * Map a physical disk address to a list of logical addresses.
2013 * @fs_info: the filesystem
2014 * @chunk_start: logical address of block group
2015 * @physical: physical address to map to logical addresses
2016 * @logical: return array of logical addresses which map to @physical
2017 * @naddrs: length of @logical
2018 * @stripe_len: size of IO stripe for the given block group
2020 * Maps a particular @physical disk address to a list of @logical addresses.
2021 * Used primarily to exclude those portions of a block group that contain super
2024 int btrfs_rmap_block(struct btrfs_fs_info
*fs_info
, u64 chunk_start
,
2025 u64 physical
, u64
**logical
, int *naddrs
, int *stripe_len
)
2027 struct extent_map
*em
;
2028 struct map_lookup
*map
;
2031 u64 data_stripe_length
;
2036 em
= btrfs_get_chunk_map(fs_info
, chunk_start
, 1);
2040 map
= em
->map_lookup
;
2041 data_stripe_length
= em
->orig_block_len
;
2042 io_stripe_size
= BTRFS_STRIPE_LEN
;
2043 chunk_start
= em
->start
;
2045 /* For RAID5/6 adjust to a full IO stripe length */
2046 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
)
2047 io_stripe_size
= btrfs_stripe_nr_to_offset(nr_data_stripes(map
));
2049 buf
= kcalloc(map
->num_stripes
, sizeof(u64
), GFP_NOFS
);
2055 for (i
= 0; i
< map
->num_stripes
; i
++) {
2056 bool already_inserted
= false;
2061 if (!in_range(physical
, map
->stripes
[i
].physical
,
2062 data_stripe_length
))
2065 stripe_nr
= (physical
- map
->stripes
[i
].physical
) >>
2066 BTRFS_STRIPE_LEN_SHIFT
;
2067 offset
= (physical
- map
->stripes
[i
].physical
) &
2068 BTRFS_STRIPE_LEN_MASK
;
2070 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID0
|
2071 BTRFS_BLOCK_GROUP_RAID10
))
2072 stripe_nr
= div_u64(stripe_nr
* map
->num_stripes
+ i
,
2075 * The remaining case would be for RAID56, multiply by
2076 * nr_data_stripes(). Alternatively, just use rmap_len below
2077 * instead of map->stripe_len
2079 bytenr
= chunk_start
+ stripe_nr
* io_stripe_size
+ offset
;
2081 /* Ensure we don't add duplicate addresses */
2082 for (j
= 0; j
< nr
; j
++) {
2083 if (buf
[j
] == bytenr
) {
2084 already_inserted
= true;
2089 if (!already_inserted
)
2095 *stripe_len
= io_stripe_size
;
2097 free_extent_map(em
);
2101 static int exclude_super_stripes(struct btrfs_block_group
*cache
)
2103 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
2104 const bool zoned
= btrfs_is_zoned(fs_info
);
2110 if (cache
->start
< BTRFS_SUPER_INFO_OFFSET
) {
2111 stripe_len
= BTRFS_SUPER_INFO_OFFSET
- cache
->start
;
2112 cache
->bytes_super
+= stripe_len
;
2113 ret
= set_extent_bit(&fs_info
->excluded_extents
, cache
->start
,
2114 cache
->start
+ stripe_len
- 1,
2115 EXTENT_UPTODATE
, NULL
);
2120 for (i
= 0; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
2121 bytenr
= btrfs_sb_offset(i
);
2122 ret
= btrfs_rmap_block(fs_info
, cache
->start
,
2123 bytenr
, &logical
, &nr
, &stripe_len
);
2127 /* Shouldn't have super stripes in sequential zones */
2131 "zoned: block group %llu must not contain super block",
2137 u64 len
= min_t(u64
, stripe_len
,
2138 cache
->start
+ cache
->length
- logical
[nr
]);
2140 cache
->bytes_super
+= len
;
2141 ret
= set_extent_bit(&fs_info
->excluded_extents
, logical
[nr
],
2142 logical
[nr
] + len
- 1,
2143 EXTENT_UPTODATE
, NULL
);
2155 static struct btrfs_block_group
*btrfs_create_block_group_cache(
2156 struct btrfs_fs_info
*fs_info
, u64 start
)
2158 struct btrfs_block_group
*cache
;
2160 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
2164 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
2166 if (!cache
->free_space_ctl
) {
2171 cache
->start
= start
;
2173 cache
->fs_info
= fs_info
;
2174 cache
->full_stripe_len
= btrfs_full_stripe_len(fs_info
, start
);
2176 cache
->discard_index
= BTRFS_DISCARD_INDEX_UNUSED
;
2178 refcount_set(&cache
->refs
, 1);
2179 spin_lock_init(&cache
->lock
);
2180 init_rwsem(&cache
->data_rwsem
);
2181 INIT_LIST_HEAD(&cache
->list
);
2182 INIT_LIST_HEAD(&cache
->cluster_list
);
2183 INIT_LIST_HEAD(&cache
->bg_list
);
2184 INIT_LIST_HEAD(&cache
->ro_list
);
2185 INIT_LIST_HEAD(&cache
->discard_list
);
2186 INIT_LIST_HEAD(&cache
->dirty_list
);
2187 INIT_LIST_HEAD(&cache
->io_list
);
2188 INIT_LIST_HEAD(&cache
->active_bg_list
);
2189 btrfs_init_free_space_ctl(cache
, cache
->free_space_ctl
);
2190 atomic_set(&cache
->frozen
, 0);
2191 mutex_init(&cache
->free_space_lock
);
2197 * Iterate all chunks and verify that each of them has the corresponding block
2200 static int check_chunk_block_group_mappings(struct btrfs_fs_info
*fs_info
)
2202 struct extent_map_tree
*map_tree
= &fs_info
->mapping_tree
;
2203 struct extent_map
*em
;
2204 struct btrfs_block_group
*bg
;
2209 read_lock(&map_tree
->lock
);
2211 * lookup_extent_mapping will return the first extent map
2212 * intersecting the range, so setting @len to 1 is enough to
2213 * get the first chunk.
2215 em
= lookup_extent_mapping(map_tree
, start
, 1);
2216 read_unlock(&map_tree
->lock
);
2220 bg
= btrfs_lookup_block_group(fs_info
, em
->start
);
2223 "chunk start=%llu len=%llu doesn't have corresponding block group",
2224 em
->start
, em
->len
);
2226 free_extent_map(em
);
2229 if (bg
->start
!= em
->start
|| bg
->length
!= em
->len
||
2230 (bg
->flags
& BTRFS_BLOCK_GROUP_TYPE_MASK
) !=
2231 (em
->map_lookup
->type
& BTRFS_BLOCK_GROUP_TYPE_MASK
)) {
2233 "chunk start=%llu len=%llu flags=0x%llx doesn't match block group start=%llu len=%llu flags=0x%llx",
2235 em
->map_lookup
->type
& BTRFS_BLOCK_GROUP_TYPE_MASK
,
2236 bg
->start
, bg
->length
,
2237 bg
->flags
& BTRFS_BLOCK_GROUP_TYPE_MASK
);
2239 free_extent_map(em
);
2240 btrfs_put_block_group(bg
);
2243 start
= em
->start
+ em
->len
;
2244 free_extent_map(em
);
2245 btrfs_put_block_group(bg
);
2250 static int read_one_block_group(struct btrfs_fs_info
*info
,
2251 struct btrfs_block_group_item
*bgi
,
2252 const struct btrfs_key
*key
,
2255 struct btrfs_block_group
*cache
;
2256 const bool mixed
= btrfs_fs_incompat(info
, MIXED_GROUPS
);
2259 ASSERT(key
->type
== BTRFS_BLOCK_GROUP_ITEM_KEY
);
2261 cache
= btrfs_create_block_group_cache(info
, key
->objectid
);
2265 cache
->length
= key
->offset
;
2266 cache
->used
= btrfs_stack_block_group_used(bgi
);
2267 cache
->commit_used
= cache
->used
;
2268 cache
->flags
= btrfs_stack_block_group_flags(bgi
);
2269 cache
->global_root_id
= btrfs_stack_block_group_chunk_objectid(bgi
);
2271 set_free_space_tree_thresholds(cache
);
2275 * When we mount with old space cache, we need to
2276 * set BTRFS_DC_CLEAR and set dirty flag.
2278 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
2279 * truncate the old free space cache inode and
2281 * b) Setting 'dirty flag' makes sure that we flush
2282 * the new space cache info onto disk.
2284 if (btrfs_test_opt(info
, SPACE_CACHE
))
2285 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
2287 if (!mixed
&& ((cache
->flags
& BTRFS_BLOCK_GROUP_METADATA
) &&
2288 (cache
->flags
& BTRFS_BLOCK_GROUP_DATA
))) {
2290 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
2296 ret
= btrfs_load_block_group_zone_info(cache
, false);
2298 btrfs_err(info
, "zoned: failed to load zone info of bg %llu",
2304 * We need to exclude the super stripes now so that the space info has
2305 * super bytes accounted for, otherwise we'll think we have more space
2306 * than we actually do.
2308 ret
= exclude_super_stripes(cache
);
2310 /* We may have excluded something, so call this just in case. */
2311 btrfs_free_excluded_extents(cache
);
2316 * For zoned filesystem, space after the allocation offset is the only
2317 * free space for a block group. So, we don't need any caching work.
2318 * btrfs_calc_zone_unusable() will set the amount of free space and
2319 * zone_unusable space.
2321 * For regular filesystem, check for two cases, either we are full, and
2322 * therefore don't need to bother with the caching work since we won't
2323 * find any space, or we are empty, and we can just add all the space
2324 * in and be done with it. This saves us _a_lot_ of time, particularly
2327 if (btrfs_is_zoned(info
)) {
2328 btrfs_calc_zone_unusable(cache
);
2329 /* Should not have any excluded extents. Just in case, though. */
2330 btrfs_free_excluded_extents(cache
);
2331 } else if (cache
->length
== cache
->used
) {
2332 cache
->cached
= BTRFS_CACHE_FINISHED
;
2333 btrfs_free_excluded_extents(cache
);
2334 } else if (cache
->used
== 0) {
2335 cache
->cached
= BTRFS_CACHE_FINISHED
;
2336 ret
= btrfs_add_new_free_space(cache
, cache
->start
,
2337 cache
->start
+ cache
->length
, NULL
);
2338 btrfs_free_excluded_extents(cache
);
2343 ret
= btrfs_add_block_group_cache(info
, cache
);
2345 btrfs_remove_free_space_cache(cache
);
2348 trace_btrfs_add_block_group(info
, cache
, 0);
2349 btrfs_add_bg_to_space_info(info
, cache
);
2351 set_avail_alloc_bits(info
, cache
->flags
);
2352 if (btrfs_chunk_writeable(info
, cache
->start
)) {
2353 if (cache
->used
== 0) {
2354 ASSERT(list_empty(&cache
->bg_list
));
2355 if (btrfs_test_opt(info
, DISCARD_ASYNC
))
2356 btrfs_discard_queue_work(&info
->discard_ctl
, cache
);
2358 btrfs_mark_bg_unused(cache
);
2361 inc_block_group_ro(cache
, 1);
2366 btrfs_put_block_group(cache
);
2370 static int fill_dummy_bgs(struct btrfs_fs_info
*fs_info
)
2372 struct extent_map_tree
*em_tree
= &fs_info
->mapping_tree
;
2373 struct rb_node
*node
;
2376 for (node
= rb_first_cached(&em_tree
->map
); node
; node
= rb_next(node
)) {
2377 struct extent_map
*em
;
2378 struct map_lookup
*map
;
2379 struct btrfs_block_group
*bg
;
2381 em
= rb_entry(node
, struct extent_map
, rb_node
);
2382 map
= em
->map_lookup
;
2383 bg
= btrfs_create_block_group_cache(fs_info
, em
->start
);
2389 /* Fill dummy cache as FULL */
2390 bg
->length
= em
->len
;
2391 bg
->flags
= map
->type
;
2392 bg
->cached
= BTRFS_CACHE_FINISHED
;
2394 bg
->flags
= map
->type
;
2395 ret
= btrfs_add_block_group_cache(fs_info
, bg
);
2397 * We may have some valid block group cache added already, in
2398 * that case we skip to the next one.
2400 if (ret
== -EEXIST
) {
2402 btrfs_put_block_group(bg
);
2407 btrfs_remove_free_space_cache(bg
);
2408 btrfs_put_block_group(bg
);
2412 btrfs_add_bg_to_space_info(fs_info
, bg
);
2414 set_avail_alloc_bits(fs_info
, bg
->flags
);
2417 btrfs_init_global_block_rsv(fs_info
);
2421 int btrfs_read_block_groups(struct btrfs_fs_info
*info
)
2423 struct btrfs_root
*root
= btrfs_block_group_root(info
);
2424 struct btrfs_path
*path
;
2426 struct btrfs_block_group
*cache
;
2427 struct btrfs_space_info
*space_info
;
2428 struct btrfs_key key
;
2433 * Either no extent root (with ibadroots rescue option) or we have
2434 * unsupported RO options. The fs can never be mounted read-write, so no
2435 * need to waste time searching block group items.
2437 * This also allows new extent tree related changes to be RO compat,
2438 * no need for a full incompat flag.
2440 if (!root
|| (btrfs_super_compat_ro_flags(info
->super_copy
) &
2441 ~BTRFS_FEATURE_COMPAT_RO_SUPP
))
2442 return fill_dummy_bgs(info
);
2446 key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
2447 path
= btrfs_alloc_path();
2451 cache_gen
= btrfs_super_cache_generation(info
->super_copy
);
2452 if (btrfs_test_opt(info
, SPACE_CACHE
) &&
2453 btrfs_super_generation(info
->super_copy
) != cache_gen
)
2455 if (btrfs_test_opt(info
, CLEAR_CACHE
))
2459 struct btrfs_block_group_item bgi
;
2460 struct extent_buffer
*leaf
;
2463 ret
= find_first_block_group(info
, path
, &key
);
2469 leaf
= path
->nodes
[0];
2470 slot
= path
->slots
[0];
2472 read_extent_buffer(leaf
, &bgi
, btrfs_item_ptr_offset(leaf
, slot
),
2475 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
2476 btrfs_release_path(path
);
2477 ret
= read_one_block_group(info
, &bgi
, &key
, need_clear
);
2480 key
.objectid
+= key
.offset
;
2483 btrfs_release_path(path
);
2485 list_for_each_entry(space_info
, &info
->space_info
, list
) {
2488 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++) {
2489 if (list_empty(&space_info
->block_groups
[i
]))
2491 cache
= list_first_entry(&space_info
->block_groups
[i
],
2492 struct btrfs_block_group
,
2494 btrfs_sysfs_add_block_group_type(cache
);
2497 if (!(btrfs_get_alloc_profile(info
, space_info
->flags
) &
2498 (BTRFS_BLOCK_GROUP_RAID10
|
2499 BTRFS_BLOCK_GROUP_RAID1_MASK
|
2500 BTRFS_BLOCK_GROUP_RAID56_MASK
|
2501 BTRFS_BLOCK_GROUP_DUP
)))
2504 * Avoid allocating from un-mirrored block group if there are
2505 * mirrored block groups.
2507 list_for_each_entry(cache
,
2508 &space_info
->block_groups
[BTRFS_RAID_RAID0
],
2510 inc_block_group_ro(cache
, 1);
2511 list_for_each_entry(cache
,
2512 &space_info
->block_groups
[BTRFS_RAID_SINGLE
],
2514 inc_block_group_ro(cache
, 1);
2517 btrfs_init_global_block_rsv(info
);
2518 ret
= check_chunk_block_group_mappings(info
);
2520 btrfs_free_path(path
);
2522 * We've hit some error while reading the extent tree, and have
2523 * rescue=ibadroots mount option.
2524 * Try to fill the tree using dummy block groups so that the user can
2525 * continue to mount and grab their data.
2527 if (ret
&& btrfs_test_opt(info
, IGNOREBADROOTS
))
2528 ret
= fill_dummy_bgs(info
);
2533 * This function, insert_block_group_item(), belongs to the phase 2 of chunk
2536 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation
2539 static int insert_block_group_item(struct btrfs_trans_handle
*trans
,
2540 struct btrfs_block_group
*block_group
)
2542 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2543 struct btrfs_block_group_item bgi
;
2544 struct btrfs_root
*root
= btrfs_block_group_root(fs_info
);
2545 struct btrfs_key key
;
2546 u64 old_commit_used
;
2549 spin_lock(&block_group
->lock
);
2550 btrfs_set_stack_block_group_used(&bgi
, block_group
->used
);
2551 btrfs_set_stack_block_group_chunk_objectid(&bgi
,
2552 block_group
->global_root_id
);
2553 btrfs_set_stack_block_group_flags(&bgi
, block_group
->flags
);
2554 old_commit_used
= block_group
->commit_used
;
2555 block_group
->commit_used
= block_group
->used
;
2556 key
.objectid
= block_group
->start
;
2557 key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
2558 key
.offset
= block_group
->length
;
2559 spin_unlock(&block_group
->lock
);
2561 ret
= btrfs_insert_item(trans
, root
, &key
, &bgi
, sizeof(bgi
));
2563 spin_lock(&block_group
->lock
);
2564 block_group
->commit_used
= old_commit_used
;
2565 spin_unlock(&block_group
->lock
);
2571 static int insert_dev_extent(struct btrfs_trans_handle
*trans
,
2572 struct btrfs_device
*device
, u64 chunk_offset
,
2573 u64 start
, u64 num_bytes
)
2575 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
2576 struct btrfs_root
*root
= fs_info
->dev_root
;
2577 struct btrfs_path
*path
;
2578 struct btrfs_dev_extent
*extent
;
2579 struct extent_buffer
*leaf
;
2580 struct btrfs_key key
;
2583 WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA
, &device
->dev_state
));
2584 WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT
, &device
->dev_state
));
2585 path
= btrfs_alloc_path();
2589 key
.objectid
= device
->devid
;
2590 key
.type
= BTRFS_DEV_EXTENT_KEY
;
2592 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, sizeof(*extent
));
2596 leaf
= path
->nodes
[0];
2597 extent
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_extent
);
2598 btrfs_set_dev_extent_chunk_tree(leaf
, extent
, BTRFS_CHUNK_TREE_OBJECTID
);
2599 btrfs_set_dev_extent_chunk_objectid(leaf
, extent
,
2600 BTRFS_FIRST_CHUNK_TREE_OBJECTID
);
2601 btrfs_set_dev_extent_chunk_offset(leaf
, extent
, chunk_offset
);
2603 btrfs_set_dev_extent_length(leaf
, extent
, num_bytes
);
2604 btrfs_mark_buffer_dirty(trans
, leaf
);
2606 btrfs_free_path(path
);
2611 * This function belongs to phase 2.
2613 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation
2616 static int insert_dev_extents(struct btrfs_trans_handle
*trans
,
2617 u64 chunk_offset
, u64 chunk_size
)
2619 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2620 struct btrfs_device
*device
;
2621 struct extent_map
*em
;
2622 struct map_lookup
*map
;
2628 em
= btrfs_get_chunk_map(fs_info
, chunk_offset
, chunk_size
);
2632 map
= em
->map_lookup
;
2633 stripe_size
= em
->orig_block_len
;
2636 * Take the device list mutex to prevent races with the final phase of
2637 * a device replace operation that replaces the device object associated
2638 * with the map's stripes, because the device object's id can change
2639 * at any time during that final phase of the device replace operation
2640 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
2641 * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID,
2642 * resulting in persisting a device extent item with such ID.
2644 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2645 for (i
= 0; i
< map
->num_stripes
; i
++) {
2646 device
= map
->stripes
[i
].dev
;
2647 dev_offset
= map
->stripes
[i
].physical
;
2649 ret
= insert_dev_extent(trans
, device
, chunk_offset
, dev_offset
,
2654 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2656 free_extent_map(em
);
2661 * This function, btrfs_create_pending_block_groups(), belongs to the phase 2 of
2664 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation
2667 void btrfs_create_pending_block_groups(struct btrfs_trans_handle
*trans
)
2669 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2670 struct btrfs_block_group
*block_group
;
2673 while (!list_empty(&trans
->new_bgs
)) {
2676 block_group
= list_first_entry(&trans
->new_bgs
,
2677 struct btrfs_block_group
,
2682 index
= btrfs_bg_flags_to_raid_index(block_group
->flags
);
2684 ret
= insert_block_group_item(trans
, block_group
);
2686 btrfs_abort_transaction(trans
, ret
);
2687 if (!test_bit(BLOCK_GROUP_FLAG_CHUNK_ITEM_INSERTED
,
2688 &block_group
->runtime_flags
)) {
2689 mutex_lock(&fs_info
->chunk_mutex
);
2690 ret
= btrfs_chunk_alloc_add_chunk_item(trans
, block_group
);
2691 mutex_unlock(&fs_info
->chunk_mutex
);
2693 btrfs_abort_transaction(trans
, ret
);
2695 ret
= insert_dev_extents(trans
, block_group
->start
,
2696 block_group
->length
);
2698 btrfs_abort_transaction(trans
, ret
);
2699 add_block_group_free_space(trans
, block_group
);
2702 * If we restriped during balance, we may have added a new raid
2703 * type, so now add the sysfs entries when it is safe to do so.
2704 * We don't have to worry about locking here as it's handled in
2705 * btrfs_sysfs_add_block_group_type.
2707 if (block_group
->space_info
->block_group_kobjs
[index
] == NULL
)
2708 btrfs_sysfs_add_block_group_type(block_group
);
2710 /* Already aborted the transaction if it failed. */
2712 btrfs_dec_delayed_refs_rsv_bg_inserts(fs_info
);
2713 list_del_init(&block_group
->bg_list
);
2714 clear_bit(BLOCK_GROUP_FLAG_NEW
, &block_group
->runtime_flags
);
2716 btrfs_trans_release_chunk_metadata(trans
);
2720 * For extent tree v2 we use the block_group_item->chunk_offset to point at our
2721 * global root id. For v1 it's always set to BTRFS_FIRST_CHUNK_TREE_OBJECTID.
2723 static u64
calculate_global_root_id(struct btrfs_fs_info
*fs_info
, u64 offset
)
2728 if (!btrfs_fs_incompat(fs_info
, EXTENT_TREE_V2
))
2729 return BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2731 /* If we have a smaller fs index based on 128MiB. */
2732 if (btrfs_super_total_bytes(fs_info
->super_copy
) <= (SZ_1G
* 10ULL))
2735 offset
= div64_u64(offset
, div
);
2736 div64_u64_rem(offset
, fs_info
->nr_global_roots
, &index
);
2740 struct btrfs_block_group
*btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
2742 u64 chunk_offset
, u64 size
)
2744 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2745 struct btrfs_block_group
*cache
;
2748 btrfs_set_log_full_commit(trans
);
2750 cache
= btrfs_create_block_group_cache(fs_info
, chunk_offset
);
2752 return ERR_PTR(-ENOMEM
);
2755 * Mark it as new before adding it to the rbtree of block groups or any
2756 * list, so that no other task finds it and calls btrfs_mark_bg_unused()
2757 * before the new flag is set.
2759 set_bit(BLOCK_GROUP_FLAG_NEW
, &cache
->runtime_flags
);
2761 cache
->length
= size
;
2762 set_free_space_tree_thresholds(cache
);
2763 cache
->flags
= type
;
2764 cache
->cached
= BTRFS_CACHE_FINISHED
;
2765 cache
->global_root_id
= calculate_global_root_id(fs_info
, cache
->start
);
2767 if (btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
))
2768 set_bit(BLOCK_GROUP_FLAG_NEEDS_FREE_SPACE
, &cache
->runtime_flags
);
2770 ret
= btrfs_load_block_group_zone_info(cache
, true);
2772 btrfs_put_block_group(cache
);
2773 return ERR_PTR(ret
);
2776 ret
= exclude_super_stripes(cache
);
2778 /* We may have excluded something, so call this just in case */
2779 btrfs_free_excluded_extents(cache
);
2780 btrfs_put_block_group(cache
);
2781 return ERR_PTR(ret
);
2784 ret
= btrfs_add_new_free_space(cache
, chunk_offset
, chunk_offset
+ size
, NULL
);
2785 btrfs_free_excluded_extents(cache
);
2787 btrfs_put_block_group(cache
);
2788 return ERR_PTR(ret
);
2792 * Ensure the corresponding space_info object is created and
2793 * assigned to our block group. We want our bg to be added to the rbtree
2794 * with its ->space_info set.
2796 cache
->space_info
= btrfs_find_space_info(fs_info
, cache
->flags
);
2797 ASSERT(cache
->space_info
);
2799 ret
= btrfs_add_block_group_cache(fs_info
, cache
);
2801 btrfs_remove_free_space_cache(cache
);
2802 btrfs_put_block_group(cache
);
2803 return ERR_PTR(ret
);
2807 * Now that our block group has its ->space_info set and is inserted in
2808 * the rbtree, update the space info's counters.
2810 trace_btrfs_add_block_group(fs_info
, cache
, 1);
2811 btrfs_add_bg_to_space_info(fs_info
, cache
);
2812 btrfs_update_global_block_rsv(fs_info
);
2814 #ifdef CONFIG_BTRFS_DEBUG
2815 if (btrfs_should_fragment_free_space(cache
)) {
2816 cache
->space_info
->bytes_used
+= size
>> 1;
2817 fragment_free_space(cache
);
2821 list_add_tail(&cache
->bg_list
, &trans
->new_bgs
);
2822 btrfs_inc_delayed_refs_rsv_bg_inserts(fs_info
);
2824 set_avail_alloc_bits(fs_info
, type
);
2829 * Mark one block group RO, can be called several times for the same block
2832 * @cache: the destination block group
2833 * @do_chunk_alloc: whether need to do chunk pre-allocation, this is to
2834 * ensure we still have some free space after marking this
2837 int btrfs_inc_block_group_ro(struct btrfs_block_group
*cache
,
2838 bool do_chunk_alloc
)
2840 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
2841 struct btrfs_trans_handle
*trans
;
2842 struct btrfs_root
*root
= btrfs_block_group_root(fs_info
);
2845 bool dirty_bg_running
;
2848 * This can only happen when we are doing read-only scrub on read-only
2850 * In that case we should not start a new transaction on read-only fs.
2851 * Thus here we skip all chunk allocations.
2853 if (sb_rdonly(fs_info
->sb
)) {
2854 mutex_lock(&fs_info
->ro_block_group_mutex
);
2855 ret
= inc_block_group_ro(cache
, 0);
2856 mutex_unlock(&fs_info
->ro_block_group_mutex
);
2861 trans
= btrfs_join_transaction(root
);
2863 return PTR_ERR(trans
);
2865 dirty_bg_running
= false;
2868 * We're not allowed to set block groups readonly after the dirty
2869 * block group cache has started writing. If it already started,
2870 * back off and let this transaction commit.
2872 mutex_lock(&fs_info
->ro_block_group_mutex
);
2873 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN
, &trans
->transaction
->flags
)) {
2874 u64 transid
= trans
->transid
;
2876 mutex_unlock(&fs_info
->ro_block_group_mutex
);
2877 btrfs_end_transaction(trans
);
2879 ret
= btrfs_wait_for_commit(fs_info
, transid
);
2882 dirty_bg_running
= true;
2884 } while (dirty_bg_running
);
2886 if (do_chunk_alloc
) {
2888 * If we are changing raid levels, try to allocate a
2889 * corresponding block group with the new raid level.
2891 alloc_flags
= btrfs_get_alloc_profile(fs_info
, cache
->flags
);
2892 if (alloc_flags
!= cache
->flags
) {
2893 ret
= btrfs_chunk_alloc(trans
, alloc_flags
,
2896 * ENOSPC is allowed here, we may have enough space
2897 * already allocated at the new raid level to carry on
2906 ret
= inc_block_group_ro(cache
, 0);
2909 if (ret
== -ETXTBSY
)
2913 * Skip chunk alloction if the bg is SYSTEM, this is to avoid system
2914 * chunk allocation storm to exhaust the system chunk array. Otherwise
2915 * we still want to try our best to mark the block group read-only.
2917 if (!do_chunk_alloc
&& ret
== -ENOSPC
&&
2918 (cache
->flags
& BTRFS_BLOCK_GROUP_SYSTEM
))
2921 alloc_flags
= btrfs_get_alloc_profile(fs_info
, cache
->space_info
->flags
);
2922 ret
= btrfs_chunk_alloc(trans
, alloc_flags
, CHUNK_ALLOC_FORCE
);
2926 * We have allocated a new chunk. We also need to activate that chunk to
2927 * grant metadata tickets for zoned filesystem.
2929 ret
= btrfs_zoned_activate_one_bg(fs_info
, cache
->space_info
, true);
2933 ret
= inc_block_group_ro(cache
, 0);
2934 if (ret
== -ETXTBSY
)
2937 if (cache
->flags
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2938 alloc_flags
= btrfs_get_alloc_profile(fs_info
, cache
->flags
);
2939 mutex_lock(&fs_info
->chunk_mutex
);
2940 check_system_chunk(trans
, alloc_flags
);
2941 mutex_unlock(&fs_info
->chunk_mutex
);
2944 mutex_unlock(&fs_info
->ro_block_group_mutex
);
2946 btrfs_end_transaction(trans
);
2950 void btrfs_dec_block_group_ro(struct btrfs_block_group
*cache
)
2952 struct btrfs_space_info
*sinfo
= cache
->space_info
;
2957 spin_lock(&sinfo
->lock
);
2958 spin_lock(&cache
->lock
);
2960 if (btrfs_is_zoned(cache
->fs_info
)) {
2961 /* Migrate zone_unusable bytes back */
2962 cache
->zone_unusable
=
2963 (cache
->alloc_offset
- cache
->used
) +
2964 (cache
->length
- cache
->zone_capacity
);
2965 sinfo
->bytes_zone_unusable
+= cache
->zone_unusable
;
2966 sinfo
->bytes_readonly
-= cache
->zone_unusable
;
2968 num_bytes
= cache
->length
- cache
->reserved
-
2969 cache
->pinned
- cache
->bytes_super
-
2970 cache
->zone_unusable
- cache
->used
;
2971 sinfo
->bytes_readonly
-= num_bytes
;
2972 list_del_init(&cache
->ro_list
);
2974 spin_unlock(&cache
->lock
);
2975 spin_unlock(&sinfo
->lock
);
2978 static int update_block_group_item(struct btrfs_trans_handle
*trans
,
2979 struct btrfs_path
*path
,
2980 struct btrfs_block_group
*cache
)
2982 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2984 struct btrfs_root
*root
= btrfs_block_group_root(fs_info
);
2986 struct extent_buffer
*leaf
;
2987 struct btrfs_block_group_item bgi
;
2988 struct btrfs_key key
;
2989 u64 old_commit_used
;
2993 * Block group items update can be triggered out of commit transaction
2994 * critical section, thus we need a consistent view of used bytes.
2995 * We cannot use cache->used directly outside of the spin lock, as it
2998 spin_lock(&cache
->lock
);
2999 old_commit_used
= cache
->commit_used
;
3001 /* No change in used bytes, can safely skip it. */
3002 if (cache
->commit_used
== used
) {
3003 spin_unlock(&cache
->lock
);
3006 cache
->commit_used
= used
;
3007 spin_unlock(&cache
->lock
);
3009 key
.objectid
= cache
->start
;
3010 key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
3011 key
.offset
= cache
->length
;
3013 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
3020 leaf
= path
->nodes
[0];
3021 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
3022 btrfs_set_stack_block_group_used(&bgi
, used
);
3023 btrfs_set_stack_block_group_chunk_objectid(&bgi
,
3024 cache
->global_root_id
);
3025 btrfs_set_stack_block_group_flags(&bgi
, cache
->flags
);
3026 write_extent_buffer(leaf
, &bgi
, bi
, sizeof(bgi
));
3027 btrfs_mark_buffer_dirty(trans
, leaf
);
3029 btrfs_release_path(path
);
3031 * We didn't update the block group item, need to revert commit_used
3032 * unless the block group item didn't exist yet - this is to prevent a
3033 * race with a concurrent insertion of the block group item, with
3034 * insert_block_group_item(), that happened just after we attempted to
3035 * update. In that case we would reset commit_used to 0 just after the
3036 * insertion set it to a value greater than 0 - if the block group later
3037 * becomes with 0 used bytes, we would incorrectly skip its update.
3039 if (ret
< 0 && ret
!= -ENOENT
) {
3040 spin_lock(&cache
->lock
);
3041 cache
->commit_used
= old_commit_used
;
3042 spin_unlock(&cache
->lock
);
3048 static int cache_save_setup(struct btrfs_block_group
*block_group
,
3049 struct btrfs_trans_handle
*trans
,
3050 struct btrfs_path
*path
)
3052 struct btrfs_fs_info
*fs_info
= block_group
->fs_info
;
3053 struct inode
*inode
= NULL
;
3054 struct extent_changeset
*data_reserved
= NULL
;
3056 int dcs
= BTRFS_DC_ERROR
;
3061 if (!btrfs_test_opt(fs_info
, SPACE_CACHE
))
3065 * If this block group is smaller than 100 megs don't bother caching the
3068 if (block_group
->length
< (100 * SZ_1M
)) {
3069 spin_lock(&block_group
->lock
);
3070 block_group
->disk_cache_state
= BTRFS_DC_WRITTEN
;
3071 spin_unlock(&block_group
->lock
);
3075 if (TRANS_ABORTED(trans
))
3078 inode
= lookup_free_space_inode(block_group
, path
);
3079 if (IS_ERR(inode
) && PTR_ERR(inode
) != -ENOENT
) {
3080 ret
= PTR_ERR(inode
);
3081 btrfs_release_path(path
);
3085 if (IS_ERR(inode
)) {
3089 if (block_group
->ro
)
3092 ret
= create_free_space_inode(trans
, block_group
, path
);
3099 * We want to set the generation to 0, that way if anything goes wrong
3100 * from here on out we know not to trust this cache when we load up next
3103 BTRFS_I(inode
)->generation
= 0;
3104 ret
= btrfs_update_inode(trans
, BTRFS_I(inode
));
3107 * So theoretically we could recover from this, simply set the
3108 * super cache generation to 0 so we know to invalidate the
3109 * cache, but then we'd have to keep track of the block groups
3110 * that fail this way so we know we _have_ to reset this cache
3111 * before the next commit or risk reading stale cache. So to
3112 * limit our exposure to horrible edge cases lets just abort the
3113 * transaction, this only happens in really bad situations
3116 btrfs_abort_transaction(trans
, ret
);
3121 /* We've already setup this transaction, go ahead and exit */
3122 if (block_group
->cache_generation
== trans
->transid
&&
3123 i_size_read(inode
)) {
3124 dcs
= BTRFS_DC_SETUP
;
3128 if (i_size_read(inode
) > 0) {
3129 ret
= btrfs_check_trunc_cache_free_space(fs_info
,
3130 &fs_info
->global_block_rsv
);
3134 ret
= btrfs_truncate_free_space_cache(trans
, NULL
, inode
);
3139 spin_lock(&block_group
->lock
);
3140 if (block_group
->cached
!= BTRFS_CACHE_FINISHED
||
3141 !btrfs_test_opt(fs_info
, SPACE_CACHE
)) {
3143 * don't bother trying to write stuff out _if_
3144 * a) we're not cached,
3145 * b) we're with nospace_cache mount option,
3146 * c) we're with v2 space_cache (FREE_SPACE_TREE).
3148 dcs
= BTRFS_DC_WRITTEN
;
3149 spin_unlock(&block_group
->lock
);
3152 spin_unlock(&block_group
->lock
);
3155 * We hit an ENOSPC when setting up the cache in this transaction, just
3156 * skip doing the setup, we've already cleared the cache so we're safe.
3158 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC
, &trans
->transaction
->flags
)) {
3164 * Try to preallocate enough space based on how big the block group is.
3165 * Keep in mind this has to include any pinned space which could end up
3166 * taking up quite a bit since it's not folded into the other space
3169 cache_size
= div_u64(block_group
->length
, SZ_256M
);
3174 cache_size
*= fs_info
->sectorsize
;
3176 ret
= btrfs_check_data_free_space(BTRFS_I(inode
), &data_reserved
, 0,
3181 ret
= btrfs_prealloc_file_range_trans(inode
, trans
, 0, 0, cache_size
,
3182 cache_size
, cache_size
,
3185 * Our cache requires contiguous chunks so that we don't modify a bunch
3186 * of metadata or split extents when writing the cache out, which means
3187 * we can enospc if we are heavily fragmented in addition to just normal
3188 * out of space conditions. So if we hit this just skip setting up any
3189 * other block groups for this transaction, maybe we'll unpin enough
3190 * space the next time around.
3193 dcs
= BTRFS_DC_SETUP
;
3194 else if (ret
== -ENOSPC
)
3195 set_bit(BTRFS_TRANS_CACHE_ENOSPC
, &trans
->transaction
->flags
);
3200 btrfs_release_path(path
);
3202 spin_lock(&block_group
->lock
);
3203 if (!ret
&& dcs
== BTRFS_DC_SETUP
)
3204 block_group
->cache_generation
= trans
->transid
;
3205 block_group
->disk_cache_state
= dcs
;
3206 spin_unlock(&block_group
->lock
);
3208 extent_changeset_free(data_reserved
);
3212 int btrfs_setup_space_cache(struct btrfs_trans_handle
*trans
)
3214 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
3215 struct btrfs_block_group
*cache
, *tmp
;
3216 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3217 struct btrfs_path
*path
;
3219 if (list_empty(&cur_trans
->dirty_bgs
) ||
3220 !btrfs_test_opt(fs_info
, SPACE_CACHE
))
3223 path
= btrfs_alloc_path();
3227 /* Could add new block groups, use _safe just in case */
3228 list_for_each_entry_safe(cache
, tmp
, &cur_trans
->dirty_bgs
,
3230 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
)
3231 cache_save_setup(cache
, trans
, path
);
3234 btrfs_free_path(path
);
3239 * Transaction commit does final block group cache writeback during a critical
3240 * section where nothing is allowed to change the FS. This is required in
3241 * order for the cache to actually match the block group, but can introduce a
3242 * lot of latency into the commit.
3244 * So, btrfs_start_dirty_block_groups is here to kick off block group cache IO.
3245 * There's a chance we'll have to redo some of it if the block group changes
3246 * again during the commit, but it greatly reduces the commit latency by
3247 * getting rid of the easy block groups while we're still allowing others to
3250 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle
*trans
)
3252 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
3253 struct btrfs_block_group
*cache
;
3254 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3257 struct btrfs_path
*path
= NULL
;
3259 struct list_head
*io
= &cur_trans
->io_bgs
;
3262 spin_lock(&cur_trans
->dirty_bgs_lock
);
3263 if (list_empty(&cur_trans
->dirty_bgs
)) {
3264 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3267 list_splice_init(&cur_trans
->dirty_bgs
, &dirty
);
3268 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3271 /* Make sure all the block groups on our dirty list actually exist */
3272 btrfs_create_pending_block_groups(trans
);
3275 path
= btrfs_alloc_path();
3283 * cache_write_mutex is here only to save us from balance or automatic
3284 * removal of empty block groups deleting this block group while we are
3285 * writing out the cache
3287 mutex_lock(&trans
->transaction
->cache_write_mutex
);
3288 while (!list_empty(&dirty
)) {
3289 bool drop_reserve
= true;
3291 cache
= list_first_entry(&dirty
, struct btrfs_block_group
,
3294 * This can happen if something re-dirties a block group that
3295 * is already under IO. Just wait for it to finish and then do
3298 if (!list_empty(&cache
->io_list
)) {
3299 list_del_init(&cache
->io_list
);
3300 btrfs_wait_cache_io(trans
, cache
, path
);
3301 btrfs_put_block_group(cache
);
3306 * btrfs_wait_cache_io uses the cache->dirty_list to decide if
3307 * it should update the cache_state. Don't delete until after
3310 * Since we're not running in the commit critical section
3311 * we need the dirty_bgs_lock to protect from update_block_group
3313 spin_lock(&cur_trans
->dirty_bgs_lock
);
3314 list_del_init(&cache
->dirty_list
);
3315 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3319 cache_save_setup(cache
, trans
, path
);
3321 if (cache
->disk_cache_state
== BTRFS_DC_SETUP
) {
3322 cache
->io_ctl
.inode
= NULL
;
3323 ret
= btrfs_write_out_cache(trans
, cache
, path
);
3324 if (ret
== 0 && cache
->io_ctl
.inode
) {
3328 * The cache_write_mutex is protecting the
3329 * io_list, also refer to the definition of
3330 * btrfs_transaction::io_bgs for more details
3332 list_add_tail(&cache
->io_list
, io
);
3335 * If we failed to write the cache, the
3336 * generation will be bad and life goes on
3342 ret
= update_block_group_item(trans
, path
, cache
);
3344 * Our block group might still be attached to the list
3345 * of new block groups in the transaction handle of some
3346 * other task (struct btrfs_trans_handle->new_bgs). This
3347 * means its block group item isn't yet in the extent
3348 * tree. If this happens ignore the error, as we will
3349 * try again later in the critical section of the
3350 * transaction commit.
3352 if (ret
== -ENOENT
) {
3354 spin_lock(&cur_trans
->dirty_bgs_lock
);
3355 if (list_empty(&cache
->dirty_list
)) {
3356 list_add_tail(&cache
->dirty_list
,
3357 &cur_trans
->dirty_bgs
);
3358 btrfs_get_block_group(cache
);
3359 drop_reserve
= false;
3361 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3363 btrfs_abort_transaction(trans
, ret
);
3367 /* If it's not on the io list, we need to put the block group */
3369 btrfs_put_block_group(cache
);
3371 btrfs_dec_delayed_refs_rsv_bg_updates(fs_info
);
3373 * Avoid blocking other tasks for too long. It might even save
3374 * us from writing caches for block groups that are going to be
3377 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
3380 mutex_lock(&trans
->transaction
->cache_write_mutex
);
3382 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
3385 * Go through delayed refs for all the stuff we've just kicked off
3386 * and then loop back (just once)
3389 ret
= btrfs_run_delayed_refs(trans
, 0);
3390 if (!ret
&& loops
== 0) {
3392 spin_lock(&cur_trans
->dirty_bgs_lock
);
3393 list_splice_init(&cur_trans
->dirty_bgs
, &dirty
);
3395 * dirty_bgs_lock protects us from concurrent block group
3396 * deletes too (not just cache_write_mutex).
3398 if (!list_empty(&dirty
)) {
3399 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3402 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3406 spin_lock(&cur_trans
->dirty_bgs_lock
);
3407 list_splice_init(&dirty
, &cur_trans
->dirty_bgs
);
3408 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3409 btrfs_cleanup_dirty_bgs(cur_trans
, fs_info
);
3412 btrfs_free_path(path
);
3416 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
)
3418 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
3419 struct btrfs_block_group
*cache
;
3420 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3423 struct btrfs_path
*path
;
3424 struct list_head
*io
= &cur_trans
->io_bgs
;
3426 path
= btrfs_alloc_path();
3431 * Even though we are in the critical section of the transaction commit,
3432 * we can still have concurrent tasks adding elements to this
3433 * transaction's list of dirty block groups. These tasks correspond to
3434 * endio free space workers started when writeback finishes for a
3435 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3436 * allocate new block groups as a result of COWing nodes of the root
3437 * tree when updating the free space inode. The writeback for the space
3438 * caches is triggered by an earlier call to
3439 * btrfs_start_dirty_block_groups() and iterations of the following
3441 * Also we want to do the cache_save_setup first and then run the
3442 * delayed refs to make sure we have the best chance at doing this all
3445 spin_lock(&cur_trans
->dirty_bgs_lock
);
3446 while (!list_empty(&cur_trans
->dirty_bgs
)) {
3447 cache
= list_first_entry(&cur_trans
->dirty_bgs
,
3448 struct btrfs_block_group
,
3452 * This can happen if cache_save_setup re-dirties a block group
3453 * that is already under IO. Just wait for it to finish and
3454 * then do it all again
3456 if (!list_empty(&cache
->io_list
)) {
3457 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3458 list_del_init(&cache
->io_list
);
3459 btrfs_wait_cache_io(trans
, cache
, path
);
3460 btrfs_put_block_group(cache
);
3461 spin_lock(&cur_trans
->dirty_bgs_lock
);
3465 * Don't remove from the dirty list until after we've waited on
3468 list_del_init(&cache
->dirty_list
);
3469 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3472 cache_save_setup(cache
, trans
, path
);
3475 ret
= btrfs_run_delayed_refs(trans
, U64_MAX
);
3477 if (!ret
&& cache
->disk_cache_state
== BTRFS_DC_SETUP
) {
3478 cache
->io_ctl
.inode
= NULL
;
3479 ret
= btrfs_write_out_cache(trans
, cache
, path
);
3480 if (ret
== 0 && cache
->io_ctl
.inode
) {
3482 list_add_tail(&cache
->io_list
, io
);
3485 * If we failed to write the cache, the
3486 * generation will be bad and life goes on
3492 ret
= update_block_group_item(trans
, path
, cache
);
3494 * One of the free space endio workers might have
3495 * created a new block group while updating a free space
3496 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3497 * and hasn't released its transaction handle yet, in
3498 * which case the new block group is still attached to
3499 * its transaction handle and its creation has not
3500 * finished yet (no block group item in the extent tree
3501 * yet, etc). If this is the case, wait for all free
3502 * space endio workers to finish and retry. This is a
3503 * very rare case so no need for a more efficient and
3506 if (ret
== -ENOENT
) {
3507 wait_event(cur_trans
->writer_wait
,
3508 atomic_read(&cur_trans
->num_writers
) == 1);
3509 ret
= update_block_group_item(trans
, path
, cache
);
3512 btrfs_abort_transaction(trans
, ret
);
3515 /* If its not on the io list, we need to put the block group */
3517 btrfs_put_block_group(cache
);
3518 btrfs_dec_delayed_refs_rsv_bg_updates(fs_info
);
3519 spin_lock(&cur_trans
->dirty_bgs_lock
);
3521 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3524 * Refer to the definition of io_bgs member for details why it's safe
3525 * to use it without any locking
3527 while (!list_empty(io
)) {
3528 cache
= list_first_entry(io
, struct btrfs_block_group
,
3530 list_del_init(&cache
->io_list
);
3531 btrfs_wait_cache_io(trans
, cache
, path
);
3532 btrfs_put_block_group(cache
);
3535 btrfs_free_path(path
);
3539 int btrfs_update_block_group(struct btrfs_trans_handle
*trans
,
3540 u64 bytenr
, u64 num_bytes
, bool alloc
)
3542 struct btrfs_fs_info
*info
= trans
->fs_info
;
3543 struct btrfs_space_info
*space_info
;
3544 struct btrfs_block_group
*cache
;
3546 bool reclaim
= false;
3547 bool bg_already_dirty
= true;
3550 /* Block accounting for super block */
3551 spin_lock(&info
->delalloc_root_lock
);
3552 old_val
= btrfs_super_bytes_used(info
->super_copy
);
3554 old_val
+= num_bytes
;
3556 old_val
-= num_bytes
;
3557 btrfs_set_super_bytes_used(info
->super_copy
, old_val
);
3558 spin_unlock(&info
->delalloc_root_lock
);
3560 cache
= btrfs_lookup_block_group(info
, bytenr
);
3564 /* An extent can not span multiple block groups. */
3565 ASSERT(bytenr
+ num_bytes
<= cache
->start
+ cache
->length
);
3567 space_info
= cache
->space_info
;
3568 factor
= btrfs_bg_type_to_factor(cache
->flags
);
3571 * If this block group has free space cache written out, we need to make
3572 * sure to load it if we are removing space. This is because we need
3573 * the unpinning stage to actually add the space back to the block group,
3574 * otherwise we will leak space.
3576 if (!alloc
&& !btrfs_block_group_done(cache
))
3577 btrfs_cache_block_group(cache
, true);
3579 spin_lock(&space_info
->lock
);
3580 spin_lock(&cache
->lock
);
3582 if (btrfs_test_opt(info
, SPACE_CACHE
) &&
3583 cache
->disk_cache_state
< BTRFS_DC_CLEAR
)
3584 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
3586 old_val
= cache
->used
;
3588 old_val
+= num_bytes
;
3589 cache
->used
= old_val
;
3590 cache
->reserved
-= num_bytes
;
3591 space_info
->bytes_reserved
-= num_bytes
;
3592 space_info
->bytes_used
+= num_bytes
;
3593 space_info
->disk_used
+= num_bytes
* factor
;
3594 spin_unlock(&cache
->lock
);
3595 spin_unlock(&space_info
->lock
);
3597 old_val
-= num_bytes
;
3598 cache
->used
= old_val
;
3599 cache
->pinned
+= num_bytes
;
3600 btrfs_space_info_update_bytes_pinned(info
, space_info
, num_bytes
);
3601 space_info
->bytes_used
-= num_bytes
;
3602 space_info
->disk_used
-= num_bytes
* factor
;
3604 reclaim
= should_reclaim_block_group(cache
, num_bytes
);
3606 spin_unlock(&cache
->lock
);
3607 spin_unlock(&space_info
->lock
);
3609 set_extent_bit(&trans
->transaction
->pinned_extents
, bytenr
,
3610 bytenr
+ num_bytes
- 1, EXTENT_DIRTY
, NULL
);
3613 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
3614 if (list_empty(&cache
->dirty_list
)) {
3615 list_add_tail(&cache
->dirty_list
, &trans
->transaction
->dirty_bgs
);
3616 bg_already_dirty
= false;
3617 btrfs_get_block_group(cache
);
3619 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
3622 * No longer have used bytes in this block group, queue it for deletion.
3623 * We do this after adding the block group to the dirty list to avoid
3624 * races between cleaner kthread and space cache writeout.
3626 if (!alloc
&& old_val
== 0) {
3627 if (!btrfs_test_opt(info
, DISCARD_ASYNC
))
3628 btrfs_mark_bg_unused(cache
);
3629 } else if (!alloc
&& reclaim
) {
3630 btrfs_mark_bg_to_reclaim(cache
);
3633 btrfs_put_block_group(cache
);
3635 /* Modified block groups are accounted for in the delayed_refs_rsv. */
3636 if (!bg_already_dirty
)
3637 btrfs_inc_delayed_refs_rsv_bg_updates(info
);
3643 * Update the block_group and space info counters.
3645 * @cache: The cache we are manipulating
3646 * @ram_bytes: The number of bytes of file content, and will be same to
3647 * @num_bytes except for the compress path.
3648 * @num_bytes: The number of bytes in question
3649 * @delalloc: The blocks are allocated for the delalloc write
3651 * This is called by the allocator when it reserves space. If this is a
3652 * reservation and the block group has become read only we cannot make the
3653 * reservation and return -EAGAIN, otherwise this function always succeeds.
3655 int btrfs_add_reserved_bytes(struct btrfs_block_group
*cache
,
3656 u64 ram_bytes
, u64 num_bytes
, int delalloc
,
3657 bool force_wrong_size_class
)
3659 struct btrfs_space_info
*space_info
= cache
->space_info
;
3660 enum btrfs_block_group_size_class size_class
;
3663 spin_lock(&space_info
->lock
);
3664 spin_lock(&cache
->lock
);
3670 if (btrfs_block_group_should_use_size_class(cache
)) {
3671 size_class
= btrfs_calc_block_group_size_class(num_bytes
);
3672 ret
= btrfs_use_block_group_size_class(cache
, size_class
, force_wrong_size_class
);
3676 cache
->reserved
+= num_bytes
;
3677 space_info
->bytes_reserved
+= num_bytes
;
3678 trace_btrfs_space_reservation(cache
->fs_info
, "space_info",
3679 space_info
->flags
, num_bytes
, 1);
3680 btrfs_space_info_update_bytes_may_use(cache
->fs_info
,
3681 space_info
, -ram_bytes
);
3683 cache
->delalloc_bytes
+= num_bytes
;
3686 * Compression can use less space than we reserved, so wake tickets if
3689 if (num_bytes
< ram_bytes
)
3690 btrfs_try_granting_tickets(cache
->fs_info
, space_info
);
3692 spin_unlock(&cache
->lock
);
3693 spin_unlock(&space_info
->lock
);
3698 * Update the block_group and space info counters.
3700 * @cache: The cache we are manipulating
3701 * @num_bytes: The number of bytes in question
3702 * @delalloc: The blocks are allocated for the delalloc write
3704 * This is called by somebody who is freeing space that was never actually used
3705 * on disk. For example if you reserve some space for a new leaf in transaction
3706 * A and before transaction A commits you free that leaf, you call this with
3707 * reserve set to 0 in order to clear the reservation.
3709 void btrfs_free_reserved_bytes(struct btrfs_block_group
*cache
,
3710 u64 num_bytes
, int delalloc
)
3712 struct btrfs_space_info
*space_info
= cache
->space_info
;
3714 spin_lock(&space_info
->lock
);
3715 spin_lock(&cache
->lock
);
3717 space_info
->bytes_readonly
+= num_bytes
;
3718 cache
->reserved
-= num_bytes
;
3719 space_info
->bytes_reserved
-= num_bytes
;
3720 space_info
->max_extent_size
= 0;
3723 cache
->delalloc_bytes
-= num_bytes
;
3724 spin_unlock(&cache
->lock
);
3726 btrfs_try_granting_tickets(cache
->fs_info
, space_info
);
3727 spin_unlock(&space_info
->lock
);
3730 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
3732 struct list_head
*head
= &info
->space_info
;
3733 struct btrfs_space_info
*found
;
3735 list_for_each_entry(found
, head
, list
) {
3736 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
3737 found
->force_alloc
= CHUNK_ALLOC_FORCE
;
3741 static int should_alloc_chunk(struct btrfs_fs_info
*fs_info
,
3742 struct btrfs_space_info
*sinfo
, int force
)
3744 u64 bytes_used
= btrfs_space_info_used(sinfo
, false);
3747 if (force
== CHUNK_ALLOC_FORCE
)
3751 * in limited mode, we want to have some free space up to
3752 * about 1% of the FS size.
3754 if (force
== CHUNK_ALLOC_LIMITED
) {
3755 thresh
= btrfs_super_total_bytes(fs_info
->super_copy
);
3756 thresh
= max_t(u64
, SZ_64M
, mult_perc(thresh
, 1));
3758 if (sinfo
->total_bytes
- bytes_used
< thresh
)
3762 if (bytes_used
+ SZ_2M
< mult_perc(sinfo
->total_bytes
, 80))
3767 int btrfs_force_chunk_alloc(struct btrfs_trans_handle
*trans
, u64 type
)
3769 u64 alloc_flags
= btrfs_get_alloc_profile(trans
->fs_info
, type
);
3771 return btrfs_chunk_alloc(trans
, alloc_flags
, CHUNK_ALLOC_FORCE
);
3774 static struct btrfs_block_group
*do_chunk_alloc(struct btrfs_trans_handle
*trans
, u64 flags
)
3776 struct btrfs_block_group
*bg
;
3780 * Check if we have enough space in the system space info because we
3781 * will need to update device items in the chunk btree and insert a new
3782 * chunk item in the chunk btree as well. This will allocate a new
3783 * system block group if needed.
3785 check_system_chunk(trans
, flags
);
3787 bg
= btrfs_create_chunk(trans
, flags
);
3793 ret
= btrfs_chunk_alloc_add_chunk_item(trans
, bg
);
3795 * Normally we are not expected to fail with -ENOSPC here, since we have
3796 * previously reserved space in the system space_info and allocated one
3797 * new system chunk if necessary. However there are three exceptions:
3799 * 1) We may have enough free space in the system space_info but all the
3800 * existing system block groups have a profile which can not be used
3801 * for extent allocation.
3803 * This happens when mounting in degraded mode. For example we have a
3804 * RAID1 filesystem with 2 devices, lose one device and mount the fs
3805 * using the other device in degraded mode. If we then allocate a chunk,
3806 * we may have enough free space in the existing system space_info, but
3807 * none of the block groups can be used for extent allocation since they
3808 * have a RAID1 profile, and because we are in degraded mode with a
3809 * single device, we are forced to allocate a new system chunk with a
3810 * SINGLE profile. Making check_system_chunk() iterate over all system
3811 * block groups and check if they have a usable profile and enough space
3812 * can be slow on very large filesystems, so we tolerate the -ENOSPC and
3813 * try again after forcing allocation of a new system chunk. Like this
3814 * we avoid paying the cost of that search in normal circumstances, when
3815 * we were not mounted in degraded mode;
3817 * 2) We had enough free space info the system space_info, and one suitable
3818 * block group to allocate from when we called check_system_chunk()
3819 * above. However right after we called it, the only system block group
3820 * with enough free space got turned into RO mode by a running scrub,
3821 * and in this case we have to allocate a new one and retry. We only
3822 * need do this allocate and retry once, since we have a transaction
3823 * handle and scrub uses the commit root to search for block groups;
3825 * 3) We had one system block group with enough free space when we called
3826 * check_system_chunk(), but after that, right before we tried to
3827 * allocate the last extent buffer we needed, a discard operation came
3828 * in and it temporarily removed the last free space entry from the
3829 * block group (discard removes a free space entry, discards it, and
3830 * then adds back the entry to the block group cache).
3832 if (ret
== -ENOSPC
) {
3833 const u64 sys_flags
= btrfs_system_alloc_profile(trans
->fs_info
);
3834 struct btrfs_block_group
*sys_bg
;
3836 sys_bg
= btrfs_create_chunk(trans
, sys_flags
);
3837 if (IS_ERR(sys_bg
)) {
3838 ret
= PTR_ERR(sys_bg
);
3839 btrfs_abort_transaction(trans
, ret
);
3843 ret
= btrfs_chunk_alloc_add_chunk_item(trans
, sys_bg
);
3845 btrfs_abort_transaction(trans
, ret
);
3849 ret
= btrfs_chunk_alloc_add_chunk_item(trans
, bg
);
3851 btrfs_abort_transaction(trans
, ret
);
3855 btrfs_abort_transaction(trans
, ret
);
3859 btrfs_trans_release_chunk_metadata(trans
);
3862 return ERR_PTR(ret
);
3864 btrfs_get_block_group(bg
);
3869 * Chunk allocation is done in 2 phases:
3871 * 1) Phase 1 - through btrfs_chunk_alloc() we allocate device extents for
3872 * the chunk, the chunk mapping, create its block group and add the items
3873 * that belong in the chunk btree to it - more specifically, we need to
3874 * update device items in the chunk btree and add a new chunk item to it.
3876 * 2) Phase 2 - through btrfs_create_pending_block_groups(), we add the block
3877 * group item to the extent btree and the device extent items to the devices
3880 * This is done to prevent deadlocks. For example when COWing a node from the
3881 * extent btree we are holding a write lock on the node's parent and if we
3882 * trigger chunk allocation and attempted to insert the new block group item
3883 * in the extent btree right way, we could deadlock because the path for the
3884 * insertion can include that parent node. At first glance it seems impossible
3885 * to trigger chunk allocation after starting a transaction since tasks should
3886 * reserve enough transaction units (metadata space), however while that is true
3887 * most of the time, chunk allocation may still be triggered for several reasons:
3889 * 1) When reserving metadata, we check if there is enough free space in the
3890 * metadata space_info and therefore don't trigger allocation of a new chunk.
3891 * However later when the task actually tries to COW an extent buffer from
3892 * the extent btree or from the device btree for example, it is forced to
3893 * allocate a new block group (chunk) because the only one that had enough
3894 * free space was just turned to RO mode by a running scrub for example (or
3895 * device replace, block group reclaim thread, etc), so we can not use it
3896 * for allocating an extent and end up being forced to allocate a new one;
3898 * 2) Because we only check that the metadata space_info has enough free bytes,
3899 * we end up not allocating a new metadata chunk in that case. However if
3900 * the filesystem was mounted in degraded mode, none of the existing block
3901 * groups might be suitable for extent allocation due to their incompatible
3902 * profile (for e.g. mounting a 2 devices filesystem, where all block groups
3903 * use a RAID1 profile, in degraded mode using a single device). In this case
3904 * when the task attempts to COW some extent buffer of the extent btree for
3905 * example, it will trigger allocation of a new metadata block group with a
3906 * suitable profile (SINGLE profile in the example of the degraded mount of
3907 * the RAID1 filesystem);
3909 * 3) The task has reserved enough transaction units / metadata space, but when
3910 * it attempts to COW an extent buffer from the extent or device btree for
3911 * example, it does not find any free extent in any metadata block group,
3912 * therefore forced to try to allocate a new metadata block group.
3913 * This is because some other task allocated all available extents in the
3914 * meanwhile - this typically happens with tasks that don't reserve space
3915 * properly, either intentionally or as a bug. One example where this is
3916 * done intentionally is fsync, as it does not reserve any transaction units
3917 * and ends up allocating a variable number of metadata extents for log
3918 * tree extent buffers;
3920 * 4) The task has reserved enough transaction units / metadata space, but right
3921 * before it tries to allocate the last extent buffer it needs, a discard
3922 * operation comes in and, temporarily, removes the last free space entry from
3923 * the only metadata block group that had free space (discard starts by
3924 * removing a free space entry from a block group, then does the discard
3925 * operation and, once it's done, it adds back the free space entry to the
3928 * We also need this 2 phases setup when adding a device to a filesystem with
3929 * a seed device - we must create new metadata and system chunks without adding
3930 * any of the block group items to the chunk, extent and device btrees. If we
3931 * did not do it this way, we would get ENOSPC when attempting to update those
3932 * btrees, since all the chunks from the seed device are read-only.
3934 * Phase 1 does the updates and insertions to the chunk btree because if we had
3935 * it done in phase 2 and have a thundering herd of tasks allocating chunks in
3936 * parallel, we risk having too many system chunks allocated by many tasks if
3937 * many tasks reach phase 1 without the previous ones completing phase 2. In the
3938 * extreme case this leads to exhaustion of the system chunk array in the
3939 * superblock. This is easier to trigger if using a btree node/leaf size of 64K
3940 * and with RAID filesystems (so we have more device items in the chunk btree).
3941 * This has happened before and commit eafa4fd0ad0607 ("btrfs: fix exhaustion of
3942 * the system chunk array due to concurrent allocations") provides more details.
3944 * Allocation of system chunks does not happen through this function. A task that
3945 * needs to update the chunk btree (the only btree that uses system chunks), must
3946 * preallocate chunk space by calling either check_system_chunk() or
3947 * btrfs_reserve_chunk_metadata() - the former is used when allocating a data or
3948 * metadata chunk or when removing a chunk, while the later is used before doing
3949 * a modification to the chunk btree - use cases for the later are adding,
3950 * removing and resizing a device as well as relocation of a system chunk.
3951 * See the comment below for more details.
3953 * The reservation of system space, done through check_system_chunk(), as well
3954 * as all the updates and insertions into the chunk btree must be done while
3955 * holding fs_info->chunk_mutex. This is important to guarantee that while COWing
3956 * an extent buffer from the chunks btree we never trigger allocation of a new
3957 * system chunk, which would result in a deadlock (trying to lock twice an
3958 * extent buffer of the chunk btree, first time before triggering the chunk
3959 * allocation and the second time during chunk allocation while attempting to
3960 * update the chunks btree). The system chunk array is also updated while holding
3961 * that mutex. The same logic applies to removing chunks - we must reserve system
3962 * space, update the chunk btree and the system chunk array in the superblock
3963 * while holding fs_info->chunk_mutex.
3965 * This function, btrfs_chunk_alloc(), belongs to phase 1.
3967 * If @force is CHUNK_ALLOC_FORCE:
3968 * - return 1 if it successfully allocates a chunk,
3969 * - return errors including -ENOSPC otherwise.
3970 * If @force is NOT CHUNK_ALLOC_FORCE:
3971 * - return 0 if it doesn't need to allocate a new chunk,
3972 * - return 1 if it successfully allocates a chunk,
3973 * - return errors including -ENOSPC otherwise.
3975 int btrfs_chunk_alloc(struct btrfs_trans_handle
*trans
, u64 flags
,
3976 enum btrfs_chunk_alloc_enum force
)
3978 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
3979 struct btrfs_space_info
*space_info
;
3980 struct btrfs_block_group
*ret_bg
;
3981 bool wait_for_alloc
= false;
3982 bool should_alloc
= false;
3983 bool from_extent_allocation
= false;
3986 if (force
== CHUNK_ALLOC_FORCE_FOR_EXTENT
) {
3987 from_extent_allocation
= true;
3988 force
= CHUNK_ALLOC_FORCE
;
3991 /* Don't re-enter if we're already allocating a chunk */
3992 if (trans
->allocating_chunk
)
3995 * Allocation of system chunks can not happen through this path, as we
3996 * could end up in a deadlock if we are allocating a data or metadata
3997 * chunk and there is another task modifying the chunk btree.
3999 * This is because while we are holding the chunk mutex, we will attempt
4000 * to add the new chunk item to the chunk btree or update an existing
4001 * device item in the chunk btree, while the other task that is modifying
4002 * the chunk btree is attempting to COW an extent buffer while holding a
4003 * lock on it and on its parent - if the COW operation triggers a system
4004 * chunk allocation, then we can deadlock because we are holding the
4005 * chunk mutex and we may need to access that extent buffer or its parent
4006 * in order to add the chunk item or update a device item.
4008 * Tasks that want to modify the chunk tree should reserve system space
4009 * before updating the chunk btree, by calling either
4010 * btrfs_reserve_chunk_metadata() or check_system_chunk().
4011 * It's possible that after a task reserves the space, it still ends up
4012 * here - this happens in the cases described above at do_chunk_alloc().
4013 * The task will have to either retry or fail.
4015 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
4018 space_info
= btrfs_find_space_info(fs_info
, flags
);
4022 spin_lock(&space_info
->lock
);
4023 if (force
< space_info
->force_alloc
)
4024 force
= space_info
->force_alloc
;
4025 should_alloc
= should_alloc_chunk(fs_info
, space_info
, force
);
4026 if (space_info
->full
) {
4027 /* No more free physical space */
4032 spin_unlock(&space_info
->lock
);
4034 } else if (!should_alloc
) {
4035 spin_unlock(&space_info
->lock
);
4037 } else if (space_info
->chunk_alloc
) {
4039 * Someone is already allocating, so we need to block
4040 * until this someone is finished and then loop to
4041 * recheck if we should continue with our allocation
4044 wait_for_alloc
= true;
4045 force
= CHUNK_ALLOC_NO_FORCE
;
4046 spin_unlock(&space_info
->lock
);
4047 mutex_lock(&fs_info
->chunk_mutex
);
4048 mutex_unlock(&fs_info
->chunk_mutex
);
4050 /* Proceed with allocation */
4051 space_info
->chunk_alloc
= 1;
4052 wait_for_alloc
= false;
4053 spin_unlock(&space_info
->lock
);
4057 } while (wait_for_alloc
);
4059 mutex_lock(&fs_info
->chunk_mutex
);
4060 trans
->allocating_chunk
= true;
4063 * If we have mixed data/metadata chunks we want to make sure we keep
4064 * allocating mixed chunks instead of individual chunks.
4066 if (btrfs_mixed_space_info(space_info
))
4067 flags
|= (BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
);
4070 * if we're doing a data chunk, go ahead and make sure that
4071 * we keep a reasonable number of metadata chunks allocated in the
4074 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
4075 fs_info
->data_chunk_allocations
++;
4076 if (!(fs_info
->data_chunk_allocations
%
4077 fs_info
->metadata_ratio
))
4078 force_metadata_allocation(fs_info
);
4081 ret_bg
= do_chunk_alloc(trans
, flags
);
4082 trans
->allocating_chunk
= false;
4084 if (IS_ERR(ret_bg
)) {
4085 ret
= PTR_ERR(ret_bg
);
4086 } else if (from_extent_allocation
&& (flags
& BTRFS_BLOCK_GROUP_DATA
)) {
4088 * New block group is likely to be used soon. Try to activate
4089 * it now. Failure is OK for now.
4091 btrfs_zone_activate(ret_bg
);
4095 btrfs_put_block_group(ret_bg
);
4097 spin_lock(&space_info
->lock
);
4100 space_info
->full
= 1;
4105 space_info
->max_extent_size
= 0;
4108 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
4110 space_info
->chunk_alloc
= 0;
4111 spin_unlock(&space_info
->lock
);
4112 mutex_unlock(&fs_info
->chunk_mutex
);
4117 static u64
get_profile_num_devs(struct btrfs_fs_info
*fs_info
, u64 type
)
4121 num_dev
= btrfs_raid_array
[btrfs_bg_flags_to_raid_index(type
)].devs_max
;
4123 num_dev
= fs_info
->fs_devices
->rw_devices
;
4128 static void reserve_chunk_space(struct btrfs_trans_handle
*trans
,
4132 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
4133 struct btrfs_space_info
*info
;
4138 * Needed because we can end up allocating a system chunk and for an
4139 * atomic and race free space reservation in the chunk block reserve.
4141 lockdep_assert_held(&fs_info
->chunk_mutex
);
4143 info
= btrfs_find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
4144 spin_lock(&info
->lock
);
4145 left
= info
->total_bytes
- btrfs_space_info_used(info
, true);
4146 spin_unlock(&info
->lock
);
4148 if (left
< bytes
&& btrfs_test_opt(fs_info
, ENOSPC_DEBUG
)) {
4149 btrfs_info(fs_info
, "left=%llu, need=%llu, flags=%llu",
4151 btrfs_dump_space_info(fs_info
, info
, 0, 0);
4155 u64 flags
= btrfs_system_alloc_profile(fs_info
);
4156 struct btrfs_block_group
*bg
;
4159 * Ignore failure to create system chunk. We might end up not
4160 * needing it, as we might not need to COW all nodes/leafs from
4161 * the paths we visit in the chunk tree (they were already COWed
4162 * or created in the current transaction for example).
4164 bg
= btrfs_create_chunk(trans
, flags
);
4169 * We have a new chunk. We also need to activate it for
4172 ret
= btrfs_zoned_activate_one_bg(fs_info
, info
, true);
4177 * If we fail to add the chunk item here, we end up
4178 * trying again at phase 2 of chunk allocation, at
4179 * btrfs_create_pending_block_groups(). So ignore
4180 * any error here. An ENOSPC here could happen, due to
4181 * the cases described at do_chunk_alloc() - the system
4182 * block group we just created was just turned into RO
4183 * mode by a scrub for example, or a running discard
4184 * temporarily removed its free space entries, etc.
4186 btrfs_chunk_alloc_add_chunk_item(trans
, bg
);
4191 ret
= btrfs_block_rsv_add(fs_info
,
4192 &fs_info
->chunk_block_rsv
,
4193 bytes
, BTRFS_RESERVE_NO_FLUSH
);
4195 trans
->chunk_bytes_reserved
+= bytes
;
4200 * Reserve space in the system space for allocating or removing a chunk.
4201 * The caller must be holding fs_info->chunk_mutex.
4203 void check_system_chunk(struct btrfs_trans_handle
*trans
, u64 type
)
4205 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
4206 const u64 num_devs
= get_profile_num_devs(fs_info
, type
);
4209 /* num_devs device items to update and 1 chunk item to add or remove. */
4210 bytes
= btrfs_calc_metadata_size(fs_info
, num_devs
) +
4211 btrfs_calc_insert_metadata_size(fs_info
, 1);
4213 reserve_chunk_space(trans
, bytes
, type
);
4217 * Reserve space in the system space, if needed, for doing a modification to the
4220 * @trans: A transaction handle.
4221 * @is_item_insertion: Indicate if the modification is for inserting a new item
4222 * in the chunk btree or if it's for the deletion or update
4223 * of an existing item.
4225 * This is used in a context where we need to update the chunk btree outside
4226 * block group allocation and removal, to avoid a deadlock with a concurrent
4227 * task that is allocating a metadata or data block group and therefore needs to
4228 * update the chunk btree while holding the chunk mutex. After the update to the
4229 * chunk btree is done, btrfs_trans_release_chunk_metadata() should be called.
4232 void btrfs_reserve_chunk_metadata(struct btrfs_trans_handle
*trans
,
4233 bool is_item_insertion
)
4235 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
4238 if (is_item_insertion
)
4239 bytes
= btrfs_calc_insert_metadata_size(fs_info
, 1);
4241 bytes
= btrfs_calc_metadata_size(fs_info
, 1);
4243 mutex_lock(&fs_info
->chunk_mutex
);
4244 reserve_chunk_space(trans
, bytes
, BTRFS_BLOCK_GROUP_SYSTEM
);
4245 mutex_unlock(&fs_info
->chunk_mutex
);
4248 void btrfs_put_block_group_cache(struct btrfs_fs_info
*info
)
4250 struct btrfs_block_group
*block_group
;
4252 block_group
= btrfs_lookup_first_block_group(info
, 0);
4253 while (block_group
) {
4254 btrfs_wait_block_group_cache_done(block_group
);
4255 spin_lock(&block_group
->lock
);
4256 if (test_and_clear_bit(BLOCK_GROUP_FLAG_IREF
,
4257 &block_group
->runtime_flags
)) {
4258 struct inode
*inode
= block_group
->inode
;
4260 block_group
->inode
= NULL
;
4261 spin_unlock(&block_group
->lock
);
4263 ASSERT(block_group
->io_ctl
.inode
== NULL
);
4266 spin_unlock(&block_group
->lock
);
4268 block_group
= btrfs_next_block_group(block_group
);
4273 * Must be called only after stopping all workers, since we could have block
4274 * group caching kthreads running, and therefore they could race with us if we
4275 * freed the block groups before stopping them.
4277 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
4279 struct btrfs_block_group
*block_group
;
4280 struct btrfs_space_info
*space_info
;
4281 struct btrfs_caching_control
*caching_ctl
;
4284 if (btrfs_is_zoned(info
)) {
4285 if (info
->active_meta_bg
) {
4286 btrfs_put_block_group(info
->active_meta_bg
);
4287 info
->active_meta_bg
= NULL
;
4289 if (info
->active_system_bg
) {
4290 btrfs_put_block_group(info
->active_system_bg
);
4291 info
->active_system_bg
= NULL
;
4295 write_lock(&info
->block_group_cache_lock
);
4296 while (!list_empty(&info
->caching_block_groups
)) {
4297 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
4298 struct btrfs_caching_control
, list
);
4299 list_del(&caching_ctl
->list
);
4300 btrfs_put_caching_control(caching_ctl
);
4302 write_unlock(&info
->block_group_cache_lock
);
4304 spin_lock(&info
->unused_bgs_lock
);
4305 while (!list_empty(&info
->unused_bgs
)) {
4306 block_group
= list_first_entry(&info
->unused_bgs
,
4307 struct btrfs_block_group
,
4309 list_del_init(&block_group
->bg_list
);
4310 btrfs_put_block_group(block_group
);
4313 while (!list_empty(&info
->reclaim_bgs
)) {
4314 block_group
= list_first_entry(&info
->reclaim_bgs
,
4315 struct btrfs_block_group
,
4317 list_del_init(&block_group
->bg_list
);
4318 btrfs_put_block_group(block_group
);
4320 spin_unlock(&info
->unused_bgs_lock
);
4322 spin_lock(&info
->zone_active_bgs_lock
);
4323 while (!list_empty(&info
->zone_active_bgs
)) {
4324 block_group
= list_first_entry(&info
->zone_active_bgs
,
4325 struct btrfs_block_group
,
4327 list_del_init(&block_group
->active_bg_list
);
4328 btrfs_put_block_group(block_group
);
4330 spin_unlock(&info
->zone_active_bgs_lock
);
4332 write_lock(&info
->block_group_cache_lock
);
4333 while ((n
= rb_last(&info
->block_group_cache_tree
.rb_root
)) != NULL
) {
4334 block_group
= rb_entry(n
, struct btrfs_block_group
,
4336 rb_erase_cached(&block_group
->cache_node
,
4337 &info
->block_group_cache_tree
);
4338 RB_CLEAR_NODE(&block_group
->cache_node
);
4339 write_unlock(&info
->block_group_cache_lock
);
4341 down_write(&block_group
->space_info
->groups_sem
);
4342 list_del(&block_group
->list
);
4343 up_write(&block_group
->space_info
->groups_sem
);
4346 * We haven't cached this block group, which means we could
4347 * possibly have excluded extents on this block group.
4349 if (block_group
->cached
== BTRFS_CACHE_NO
||
4350 block_group
->cached
== BTRFS_CACHE_ERROR
)
4351 btrfs_free_excluded_extents(block_group
);
4353 btrfs_remove_free_space_cache(block_group
);
4354 ASSERT(block_group
->cached
!= BTRFS_CACHE_STARTED
);
4355 ASSERT(list_empty(&block_group
->dirty_list
));
4356 ASSERT(list_empty(&block_group
->io_list
));
4357 ASSERT(list_empty(&block_group
->bg_list
));
4358 ASSERT(refcount_read(&block_group
->refs
) == 1);
4359 ASSERT(block_group
->swap_extents
== 0);
4360 btrfs_put_block_group(block_group
);
4362 write_lock(&info
->block_group_cache_lock
);
4364 write_unlock(&info
->block_group_cache_lock
);
4366 btrfs_release_global_block_rsv(info
);
4368 while (!list_empty(&info
->space_info
)) {
4369 space_info
= list_entry(info
->space_info
.next
,
4370 struct btrfs_space_info
,
4374 * Do not hide this behind enospc_debug, this is actually
4375 * important and indicates a real bug if this happens.
4377 if (WARN_ON(space_info
->bytes_pinned
> 0 ||
4378 space_info
->bytes_may_use
> 0))
4379 btrfs_dump_space_info(info
, space_info
, 0, 0);
4382 * If there was a failure to cleanup a log tree, very likely due
4383 * to an IO failure on a writeback attempt of one or more of its
4384 * extent buffers, we could not do proper (and cheap) unaccounting
4385 * of their reserved space, so don't warn on bytes_reserved > 0 in
4388 if (!(space_info
->flags
& BTRFS_BLOCK_GROUP_METADATA
) ||
4389 !BTRFS_FS_LOG_CLEANUP_ERROR(info
)) {
4390 if (WARN_ON(space_info
->bytes_reserved
> 0))
4391 btrfs_dump_space_info(info
, space_info
, 0, 0);
4394 WARN_ON(space_info
->reclaim_size
> 0);
4395 list_del(&space_info
->list
);
4396 btrfs_sysfs_remove_space_info(space_info
);
4401 void btrfs_freeze_block_group(struct btrfs_block_group
*cache
)
4403 atomic_inc(&cache
->frozen
);
4406 void btrfs_unfreeze_block_group(struct btrfs_block_group
*block_group
)
4408 struct btrfs_fs_info
*fs_info
= block_group
->fs_info
;
4409 struct extent_map_tree
*em_tree
;
4410 struct extent_map
*em
;
4413 spin_lock(&block_group
->lock
);
4414 cleanup
= (atomic_dec_and_test(&block_group
->frozen
) &&
4415 test_bit(BLOCK_GROUP_FLAG_REMOVED
, &block_group
->runtime_flags
));
4416 spin_unlock(&block_group
->lock
);
4419 em_tree
= &fs_info
->mapping_tree
;
4420 write_lock(&em_tree
->lock
);
4421 em
= lookup_extent_mapping(em_tree
, block_group
->start
,
4423 BUG_ON(!em
); /* logic error, can't happen */
4424 remove_extent_mapping(em_tree
, em
);
4425 write_unlock(&em_tree
->lock
);
4427 /* once for us and once for the tree */
4428 free_extent_map(em
);
4429 free_extent_map(em
);
4432 * We may have left one free space entry and other possible
4433 * tasks trimming this block group have left 1 entry each one.
4436 btrfs_remove_free_space_cache(block_group
);
4440 bool btrfs_inc_block_group_swap_extents(struct btrfs_block_group
*bg
)
4444 spin_lock(&bg
->lock
);
4449 spin_unlock(&bg
->lock
);
4454 void btrfs_dec_block_group_swap_extents(struct btrfs_block_group
*bg
, int amount
)
4456 spin_lock(&bg
->lock
);
4458 ASSERT(bg
->swap_extents
>= amount
);
4459 bg
->swap_extents
-= amount
;
4460 spin_unlock(&bg
->lock
);
4463 enum btrfs_block_group_size_class
btrfs_calc_block_group_size_class(u64 size
)
4465 if (size
<= SZ_128K
)
4466 return BTRFS_BG_SZ_SMALL
;
4468 return BTRFS_BG_SZ_MEDIUM
;
4469 return BTRFS_BG_SZ_LARGE
;
4473 * Handle a block group allocating an extent in a size class
4475 * @bg: The block group we allocated in.
4476 * @size_class: The size class of the allocation.
4477 * @force_wrong_size_class: Whether we are desperate enough to allow
4478 * mismatched size classes.
4480 * Returns: 0 if the size class was valid for this block_group, -EAGAIN in the
4481 * case of a race that leads to the wrong size class without
4482 * force_wrong_size_class set.
4484 * find_free_extent will skip block groups with a mismatched size class until
4485 * it really needs to avoid ENOSPC. In that case it will set
4486 * force_wrong_size_class. However, if a block group is newly allocated and
4487 * doesn't yet have a size class, then it is possible for two allocations of
4488 * different sizes to race and both try to use it. The loser is caught here and
4491 int btrfs_use_block_group_size_class(struct btrfs_block_group
*bg
,
4492 enum btrfs_block_group_size_class size_class
,
4493 bool force_wrong_size_class
)
4495 ASSERT(size_class
!= BTRFS_BG_SZ_NONE
);
4497 /* The new allocation is in the right size class, do nothing */
4498 if (bg
->size_class
== size_class
)
4501 * The new allocation is in a mismatched size class.
4502 * This means one of two things:
4504 * 1. Two tasks in find_free_extent for different size_classes raced
4505 * and hit the same empty block_group. Make the loser try again.
4506 * 2. A call to find_free_extent got desperate enough to set
4507 * 'force_wrong_slab'. Don't change the size_class, but allow the
4510 if (bg
->size_class
!= BTRFS_BG_SZ_NONE
) {
4511 if (force_wrong_size_class
)
4516 * The happy new block group case: the new allocation is the first
4517 * one in the block_group so we set size_class.
4519 bg
->size_class
= size_class
;
4524 bool btrfs_block_group_should_use_size_class(struct btrfs_block_group
*bg
)
4526 if (btrfs_is_zoned(bg
->fs_info
))
4528 if (!btrfs_is_block_group_data_only(bg
))