1 // SPDX-License-Identifier: GPL-2.0
3 #include <linux/list_sort.h>
6 #include "block-group.h"
7 #include "space-info.h"
9 #include "free-space-cache.h"
10 #include "free-space-tree.h"
12 #include "transaction.h"
13 #include "ref-verify.h"
16 #include "delalloc-space.h"
22 * Return target flags in extended format or 0 if restripe for this chunk_type
25 * Should be called with balance_lock held
27 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
29 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
35 if (flags & BTRFS_BLOCK_GROUP_DATA &&
36 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
37 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
38 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
39 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
40 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
41 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
42 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
43 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
50 * @flags: available profiles in extended format (see ctree.h)
52 * Return reduced profile in chunk format. If profile changing is in progress
53 * (either running or paused) picks the target profile (if it's already
54 * available), otherwise falls back to plain reducing.
56 static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags)
58 u64 num_devices = fs_info->fs_devices->rw_devices;
64 * See if restripe for this chunk_type is in progress, if so try to
65 * reduce to the target profile
67 spin_lock(&fs_info->balance_lock);
68 target = get_restripe_target(fs_info, flags);
70 spin_unlock(&fs_info->balance_lock);
71 return extended_to_chunk(target);
73 spin_unlock(&fs_info->balance_lock);
75 /* First, mask out the RAID levels which aren't possible */
76 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
77 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
78 allowed |= btrfs_raid_array[raid_type].bg_flag;
82 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
83 allowed = BTRFS_BLOCK_GROUP_RAID6;
84 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
85 allowed = BTRFS_BLOCK_GROUP_RAID5;
86 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
87 allowed = BTRFS_BLOCK_GROUP_RAID10;
88 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
89 allowed = BTRFS_BLOCK_GROUP_RAID1;
90 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
91 allowed = BTRFS_BLOCK_GROUP_RAID0;
93 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
95 return extended_to_chunk(flags | allowed);
98 u64 btrfs_get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags)
105 seq = read_seqbegin(&fs_info->profiles_lock);
107 if (flags & BTRFS_BLOCK_GROUP_DATA)
108 flags |= fs_info->avail_data_alloc_bits;
109 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
110 flags |= fs_info->avail_system_alloc_bits;
111 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
112 flags |= fs_info->avail_metadata_alloc_bits;
113 } while (read_seqretry(&fs_info->profiles_lock, seq));
115 return btrfs_reduce_alloc_profile(fs_info, flags);
118 void btrfs_get_block_group(struct btrfs_block_group *cache)
120 refcount_inc(&cache->refs);
123 void btrfs_put_block_group(struct btrfs_block_group *cache)
125 if (refcount_dec_and_test(&cache->refs)) {
126 WARN_ON(cache->pinned > 0);
128 * If there was a failure to cleanup a log tree, very likely due
129 * to an IO failure on a writeback attempt of one or more of its
130 * extent buffers, we could not do proper (and cheap) unaccounting
131 * of their reserved space, so don't warn on reserved > 0 in that
134 if (!(cache->flags & BTRFS_BLOCK_GROUP_METADATA) ||
135 !BTRFS_FS_LOG_CLEANUP_ERROR(cache->fs_info))
136 WARN_ON(cache->reserved > 0);
139 * A block_group shouldn't be on the discard_list anymore.
140 * Remove the block_group from the discard_list to prevent us
141 * from causing a panic due to NULL pointer dereference.
143 if (WARN_ON(!list_empty(&cache->discard_list)))
144 btrfs_discard_cancel_work(&cache->fs_info->discard_ctl,
148 * If not empty, someone is still holding mutex of
149 * full_stripe_lock, which can only be released by caller.
150 * And it will definitely cause use-after-free when caller
151 * tries to release full stripe lock.
153 * No better way to resolve, but only to warn.
155 WARN_ON(!RB_EMPTY_ROOT(&cache->full_stripe_locks_root.root));
156 kfree(cache->free_space_ctl);
157 kfree(cache->physical_map);
163 * This adds the block group to the fs_info rb tree for the block group cache
165 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
166 struct btrfs_block_group *block_group)
169 struct rb_node *parent = NULL;
170 struct btrfs_block_group *cache;
172 ASSERT(block_group->length != 0);
174 spin_lock(&info->block_group_cache_lock);
175 p = &info->block_group_cache_tree.rb_node;
179 cache = rb_entry(parent, struct btrfs_block_group, cache_node);
180 if (block_group->start < cache->start) {
182 } else if (block_group->start > cache->start) {
185 spin_unlock(&info->block_group_cache_lock);
190 rb_link_node(&block_group->cache_node, parent, p);
191 rb_insert_color(&block_group->cache_node,
192 &info->block_group_cache_tree);
194 if (info->first_logical_byte > block_group->start)
195 info->first_logical_byte = block_group->start;
197 spin_unlock(&info->block_group_cache_lock);
203 * This will return the block group at or after bytenr if contains is 0, else
204 * it will return the block group that contains the bytenr
206 static struct btrfs_block_group *block_group_cache_tree_search(
207 struct btrfs_fs_info *info, u64 bytenr, int contains)
209 struct btrfs_block_group *cache, *ret = NULL;
213 spin_lock(&info->block_group_cache_lock);
214 n = info->block_group_cache_tree.rb_node;
217 cache = rb_entry(n, struct btrfs_block_group, cache_node);
218 end = cache->start + cache->length - 1;
219 start = cache->start;
221 if (bytenr < start) {
222 if (!contains && (!ret || start < ret->start))
225 } else if (bytenr > start) {
226 if (contains && bytenr <= end) {
237 btrfs_get_block_group(ret);
238 if (bytenr == 0 && info->first_logical_byte > ret->start)
239 info->first_logical_byte = ret->start;
241 spin_unlock(&info->block_group_cache_lock);
247 * Return the block group that starts at or after bytenr
249 struct btrfs_block_group *btrfs_lookup_first_block_group(
250 struct btrfs_fs_info *info, u64 bytenr)
252 return block_group_cache_tree_search(info, bytenr, 0);
256 * Return the block group that contains the given bytenr
258 struct btrfs_block_group *btrfs_lookup_block_group(
259 struct btrfs_fs_info *info, u64 bytenr)
261 return block_group_cache_tree_search(info, bytenr, 1);
264 struct btrfs_block_group *btrfs_next_block_group(
265 struct btrfs_block_group *cache)
267 struct btrfs_fs_info *fs_info = cache->fs_info;
268 struct rb_node *node;
270 spin_lock(&fs_info->block_group_cache_lock);
272 /* If our block group was removed, we need a full search. */
273 if (RB_EMPTY_NODE(&cache->cache_node)) {
274 const u64 next_bytenr = cache->start + cache->length;
276 spin_unlock(&fs_info->block_group_cache_lock);
277 btrfs_put_block_group(cache);
278 cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache;
280 node = rb_next(&cache->cache_node);
281 btrfs_put_block_group(cache);
283 cache = rb_entry(node, struct btrfs_block_group, cache_node);
284 btrfs_get_block_group(cache);
287 spin_unlock(&fs_info->block_group_cache_lock);
291 bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
293 struct btrfs_block_group *bg;
296 bg = btrfs_lookup_block_group(fs_info, bytenr);
300 spin_lock(&bg->lock);
304 atomic_inc(&bg->nocow_writers);
305 spin_unlock(&bg->lock);
307 /* No put on block group, done by btrfs_dec_nocow_writers */
309 btrfs_put_block_group(bg);
314 void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
316 struct btrfs_block_group *bg;
318 bg = btrfs_lookup_block_group(fs_info, bytenr);
320 if (atomic_dec_and_test(&bg->nocow_writers))
321 wake_up_var(&bg->nocow_writers);
323 * Once for our lookup and once for the lookup done by a previous call
324 * to btrfs_inc_nocow_writers()
326 btrfs_put_block_group(bg);
327 btrfs_put_block_group(bg);
330 void btrfs_wait_nocow_writers(struct btrfs_block_group *bg)
332 wait_var_event(&bg->nocow_writers, !atomic_read(&bg->nocow_writers));
335 void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info,
338 struct btrfs_block_group *bg;
340 bg = btrfs_lookup_block_group(fs_info, start);
342 if (atomic_dec_and_test(&bg->reservations))
343 wake_up_var(&bg->reservations);
344 btrfs_put_block_group(bg);
347 void btrfs_wait_block_group_reservations(struct btrfs_block_group *bg)
349 struct btrfs_space_info *space_info = bg->space_info;
353 if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA))
357 * Our block group is read only but before we set it to read only,
358 * some task might have had allocated an extent from it already, but it
359 * has not yet created a respective ordered extent (and added it to a
360 * root's list of ordered extents).
361 * Therefore wait for any task currently allocating extents, since the
362 * block group's reservations counter is incremented while a read lock
363 * on the groups' semaphore is held and decremented after releasing
364 * the read access on that semaphore and creating the ordered extent.
366 down_write(&space_info->groups_sem);
367 up_write(&space_info->groups_sem);
369 wait_var_event(&bg->reservations, !atomic_read(&bg->reservations));
372 struct btrfs_caching_control *btrfs_get_caching_control(
373 struct btrfs_block_group *cache)
375 struct btrfs_caching_control *ctl;
377 spin_lock(&cache->lock);
378 if (!cache->caching_ctl) {
379 spin_unlock(&cache->lock);
383 ctl = cache->caching_ctl;
384 refcount_inc(&ctl->count);
385 spin_unlock(&cache->lock);
389 void btrfs_put_caching_control(struct btrfs_caching_control *ctl)
391 if (refcount_dec_and_test(&ctl->count))
396 * When we wait for progress in the block group caching, its because our
397 * allocation attempt failed at least once. So, we must sleep and let some
398 * progress happen before we try again.
400 * This function will sleep at least once waiting for new free space to show
401 * up, and then it will check the block group free space numbers for our min
402 * num_bytes. Another option is to have it go ahead and look in the rbtree for
403 * a free extent of a given size, but this is a good start.
405 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
406 * any of the information in this block group.
408 void btrfs_wait_block_group_cache_progress(struct btrfs_block_group *cache,
411 struct btrfs_caching_control *caching_ctl;
413 caching_ctl = btrfs_get_caching_control(cache);
417 wait_event(caching_ctl->wait, btrfs_block_group_done(cache) ||
418 (cache->free_space_ctl->free_space >= num_bytes));
420 btrfs_put_caching_control(caching_ctl);
423 int btrfs_wait_block_group_cache_done(struct btrfs_block_group *cache)
425 struct btrfs_caching_control *caching_ctl;
428 caching_ctl = btrfs_get_caching_control(cache);
430 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
432 wait_event(caching_ctl->wait, btrfs_block_group_done(cache));
433 if (cache->cached == BTRFS_CACHE_ERROR)
435 btrfs_put_caching_control(caching_ctl);
439 static bool space_cache_v1_done(struct btrfs_block_group *cache)
443 spin_lock(&cache->lock);
444 ret = cache->cached != BTRFS_CACHE_FAST;
445 spin_unlock(&cache->lock);
450 void btrfs_wait_space_cache_v1_finished(struct btrfs_block_group *cache,
451 struct btrfs_caching_control *caching_ctl)
453 wait_event(caching_ctl->wait, space_cache_v1_done(cache));
456 #ifdef CONFIG_BTRFS_DEBUG
457 static void fragment_free_space(struct btrfs_block_group *block_group)
459 struct btrfs_fs_info *fs_info = block_group->fs_info;
460 u64 start = block_group->start;
461 u64 len = block_group->length;
462 u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ?
463 fs_info->nodesize : fs_info->sectorsize;
464 u64 step = chunk << 1;
466 while (len > chunk) {
467 btrfs_remove_free_space(block_group, start, chunk);
478 * This is only called by btrfs_cache_block_group, since we could have freed
479 * extents we need to check the pinned_extents for any extents that can't be
480 * used yet since their free space will be released as soon as the transaction
483 u64 add_new_free_space(struct btrfs_block_group *block_group, u64 start, u64 end)
485 struct btrfs_fs_info *info = block_group->fs_info;
486 u64 extent_start, extent_end, size, total_added = 0;
489 while (start < end) {
490 ret = find_first_extent_bit(&info->excluded_extents, start,
491 &extent_start, &extent_end,
492 EXTENT_DIRTY | EXTENT_UPTODATE,
497 if (extent_start <= start) {
498 start = extent_end + 1;
499 } else if (extent_start > start && extent_start < end) {
500 size = extent_start - start;
502 ret = btrfs_add_free_space_async_trimmed(block_group,
504 BUG_ON(ret); /* -ENOMEM or logic error */
505 start = extent_end + 1;
514 ret = btrfs_add_free_space_async_trimmed(block_group, start,
516 BUG_ON(ret); /* -ENOMEM or logic error */
522 static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl)
524 struct btrfs_block_group *block_group = caching_ctl->block_group;
525 struct btrfs_fs_info *fs_info = block_group->fs_info;
526 struct btrfs_root *extent_root;
527 struct btrfs_path *path;
528 struct extent_buffer *leaf;
529 struct btrfs_key key;
536 path = btrfs_alloc_path();
540 last = max_t(u64, block_group->start, BTRFS_SUPER_INFO_OFFSET);
541 extent_root = btrfs_extent_root(fs_info, last);
543 #ifdef CONFIG_BTRFS_DEBUG
545 * If we're fragmenting we don't want to make anybody think we can
546 * allocate from this block group until we've had a chance to fragment
549 if (btrfs_should_fragment_free_space(block_group))
553 * We don't want to deadlock with somebody trying to allocate a new
554 * extent for the extent root while also trying to search the extent
555 * root to add free space. So we skip locking and search the commit
556 * root, since its read-only
558 path->skip_locking = 1;
559 path->search_commit_root = 1;
560 path->reada = READA_FORWARD;
564 key.type = BTRFS_EXTENT_ITEM_KEY;
567 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
571 leaf = path->nodes[0];
572 nritems = btrfs_header_nritems(leaf);
575 if (btrfs_fs_closing(fs_info) > 1) {
580 if (path->slots[0] < nritems) {
581 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
583 ret = btrfs_find_next_key(extent_root, path, &key, 0, 0);
587 if (need_resched() ||
588 rwsem_is_contended(&fs_info->commit_root_sem)) {
590 caching_ctl->progress = last;
591 btrfs_release_path(path);
592 up_read(&fs_info->commit_root_sem);
593 mutex_unlock(&caching_ctl->mutex);
595 mutex_lock(&caching_ctl->mutex);
596 down_read(&fs_info->commit_root_sem);
600 ret = btrfs_next_leaf(extent_root, path);
605 leaf = path->nodes[0];
606 nritems = btrfs_header_nritems(leaf);
610 if (key.objectid < last) {
613 key.type = BTRFS_EXTENT_ITEM_KEY;
616 caching_ctl->progress = last;
617 btrfs_release_path(path);
621 if (key.objectid < block_group->start) {
626 if (key.objectid >= block_group->start + block_group->length)
629 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
630 key.type == BTRFS_METADATA_ITEM_KEY) {
631 total_found += add_new_free_space(block_group, last,
633 if (key.type == BTRFS_METADATA_ITEM_KEY)
634 last = key.objectid +
637 last = key.objectid + key.offset;
639 if (total_found > CACHING_CTL_WAKE_UP) {
642 wake_up(&caching_ctl->wait);
649 total_found += add_new_free_space(block_group, last,
650 block_group->start + block_group->length);
651 caching_ctl->progress = (u64)-1;
654 btrfs_free_path(path);
658 static noinline void caching_thread(struct btrfs_work *work)
660 struct btrfs_block_group *block_group;
661 struct btrfs_fs_info *fs_info;
662 struct btrfs_caching_control *caching_ctl;
665 caching_ctl = container_of(work, struct btrfs_caching_control, work);
666 block_group = caching_ctl->block_group;
667 fs_info = block_group->fs_info;
669 mutex_lock(&caching_ctl->mutex);
670 down_read(&fs_info->commit_root_sem);
672 if (btrfs_test_opt(fs_info, SPACE_CACHE)) {
673 ret = load_free_space_cache(block_group);
680 * We failed to load the space cache, set ourselves to
681 * CACHE_STARTED and carry on.
683 spin_lock(&block_group->lock);
684 block_group->cached = BTRFS_CACHE_STARTED;
685 spin_unlock(&block_group->lock);
686 wake_up(&caching_ctl->wait);
690 * If we are in the transaction that populated the free space tree we
691 * can't actually cache from the free space tree as our commit root and
692 * real root are the same, so we could change the contents of the blocks
693 * while caching. Instead do the slow caching in this case, and after
694 * the transaction has committed we will be safe.
696 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
697 !(test_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags)))
698 ret = load_free_space_tree(caching_ctl);
700 ret = load_extent_tree_free(caching_ctl);
702 spin_lock(&block_group->lock);
703 block_group->caching_ctl = NULL;
704 block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED;
705 spin_unlock(&block_group->lock);
707 #ifdef CONFIG_BTRFS_DEBUG
708 if (btrfs_should_fragment_free_space(block_group)) {
711 spin_lock(&block_group->space_info->lock);
712 spin_lock(&block_group->lock);
713 bytes_used = block_group->length - block_group->used;
714 block_group->space_info->bytes_used += bytes_used >> 1;
715 spin_unlock(&block_group->lock);
716 spin_unlock(&block_group->space_info->lock);
717 fragment_free_space(block_group);
721 caching_ctl->progress = (u64)-1;
723 up_read(&fs_info->commit_root_sem);
724 btrfs_free_excluded_extents(block_group);
725 mutex_unlock(&caching_ctl->mutex);
727 wake_up(&caching_ctl->wait);
729 btrfs_put_caching_control(caching_ctl);
730 btrfs_put_block_group(block_group);
733 int btrfs_cache_block_group(struct btrfs_block_group *cache, int load_cache_only)
736 struct btrfs_fs_info *fs_info = cache->fs_info;
737 struct btrfs_caching_control *caching_ctl = NULL;
740 /* Allocator for zoned filesystems does not use the cache at all */
741 if (btrfs_is_zoned(fs_info))
744 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
748 INIT_LIST_HEAD(&caching_ctl->list);
749 mutex_init(&caching_ctl->mutex);
750 init_waitqueue_head(&caching_ctl->wait);
751 caching_ctl->block_group = cache;
752 caching_ctl->progress = cache->start;
753 refcount_set(&caching_ctl->count, 2);
754 btrfs_init_work(&caching_ctl->work, caching_thread, NULL, NULL);
756 spin_lock(&cache->lock);
757 if (cache->cached != BTRFS_CACHE_NO) {
760 caching_ctl = cache->caching_ctl;
762 refcount_inc(&caching_ctl->count);
763 spin_unlock(&cache->lock);
766 WARN_ON(cache->caching_ctl);
767 cache->caching_ctl = caching_ctl;
768 if (btrfs_test_opt(fs_info, SPACE_CACHE))
769 cache->cached = BTRFS_CACHE_FAST;
771 cache->cached = BTRFS_CACHE_STARTED;
772 cache->has_caching_ctl = 1;
773 spin_unlock(&cache->lock);
775 spin_lock(&fs_info->block_group_cache_lock);
776 refcount_inc(&caching_ctl->count);
777 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
778 spin_unlock(&fs_info->block_group_cache_lock);
780 btrfs_get_block_group(cache);
782 btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
784 if (load_cache_only && caching_ctl)
785 btrfs_wait_space_cache_v1_finished(cache, caching_ctl);
787 btrfs_put_caching_control(caching_ctl);
792 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
794 u64 extra_flags = chunk_to_extended(flags) &
795 BTRFS_EXTENDED_PROFILE_MASK;
797 write_seqlock(&fs_info->profiles_lock);
798 if (flags & BTRFS_BLOCK_GROUP_DATA)
799 fs_info->avail_data_alloc_bits &= ~extra_flags;
800 if (flags & BTRFS_BLOCK_GROUP_METADATA)
801 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
802 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
803 fs_info->avail_system_alloc_bits &= ~extra_flags;
804 write_sequnlock(&fs_info->profiles_lock);
808 * Clear incompat bits for the following feature(s):
810 * - RAID56 - in case there's neither RAID5 nor RAID6 profile block group
811 * in the whole filesystem
813 * - RAID1C34 - same as above for RAID1C3 and RAID1C4 block groups
815 static void clear_incompat_bg_bits(struct btrfs_fs_info *fs_info, u64 flags)
817 bool found_raid56 = false;
818 bool found_raid1c34 = false;
820 if ((flags & BTRFS_BLOCK_GROUP_RAID56_MASK) ||
821 (flags & BTRFS_BLOCK_GROUP_RAID1C3) ||
822 (flags & BTRFS_BLOCK_GROUP_RAID1C4)) {
823 struct list_head *head = &fs_info->space_info;
824 struct btrfs_space_info *sinfo;
826 list_for_each_entry_rcu(sinfo, head, list) {
827 down_read(&sinfo->groups_sem);
828 if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID5]))
830 if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID6]))
832 if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID1C3]))
833 found_raid1c34 = true;
834 if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID1C4]))
835 found_raid1c34 = true;
836 up_read(&sinfo->groups_sem);
839 btrfs_clear_fs_incompat(fs_info, RAID56);
841 btrfs_clear_fs_incompat(fs_info, RAID1C34);
845 static int remove_block_group_item(struct btrfs_trans_handle *trans,
846 struct btrfs_path *path,
847 struct btrfs_block_group *block_group)
849 struct btrfs_fs_info *fs_info = trans->fs_info;
850 struct btrfs_root *root;
851 struct btrfs_key key;
854 root = btrfs_block_group_root(fs_info);
855 key.objectid = block_group->start;
856 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
857 key.offset = block_group->length;
859 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
865 ret = btrfs_del_item(trans, root, path);
869 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
870 u64 group_start, struct extent_map *em)
872 struct btrfs_fs_info *fs_info = trans->fs_info;
873 struct btrfs_path *path;
874 struct btrfs_block_group *block_group;
875 struct btrfs_free_cluster *cluster;
877 struct kobject *kobj = NULL;
881 struct btrfs_caching_control *caching_ctl = NULL;
883 bool remove_rsv = false;
885 block_group = btrfs_lookup_block_group(fs_info, group_start);
886 BUG_ON(!block_group);
887 BUG_ON(!block_group->ro);
889 trace_btrfs_remove_block_group(block_group);
891 * Free the reserved super bytes from this block group before
894 btrfs_free_excluded_extents(block_group);
895 btrfs_free_ref_tree_range(fs_info, block_group->start,
896 block_group->length);
898 index = btrfs_bg_flags_to_raid_index(block_group->flags);
899 factor = btrfs_bg_type_to_factor(block_group->flags);
901 /* make sure this block group isn't part of an allocation cluster */
902 cluster = &fs_info->data_alloc_cluster;
903 spin_lock(&cluster->refill_lock);
904 btrfs_return_cluster_to_free_space(block_group, cluster);
905 spin_unlock(&cluster->refill_lock);
908 * make sure this block group isn't part of a metadata
911 cluster = &fs_info->meta_alloc_cluster;
912 spin_lock(&cluster->refill_lock);
913 btrfs_return_cluster_to_free_space(block_group, cluster);
914 spin_unlock(&cluster->refill_lock);
916 btrfs_clear_treelog_bg(block_group);
917 btrfs_clear_data_reloc_bg(block_group);
919 path = btrfs_alloc_path();
926 * get the inode first so any iput calls done for the io_list
927 * aren't the final iput (no unlinks allowed now)
929 inode = lookup_free_space_inode(block_group, path);
931 mutex_lock(&trans->transaction->cache_write_mutex);
933 * Make sure our free space cache IO is done before removing the
936 spin_lock(&trans->transaction->dirty_bgs_lock);
937 if (!list_empty(&block_group->io_list)) {
938 list_del_init(&block_group->io_list);
940 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
942 spin_unlock(&trans->transaction->dirty_bgs_lock);
943 btrfs_wait_cache_io(trans, block_group, path);
944 btrfs_put_block_group(block_group);
945 spin_lock(&trans->transaction->dirty_bgs_lock);
948 if (!list_empty(&block_group->dirty_list)) {
949 list_del_init(&block_group->dirty_list);
951 btrfs_put_block_group(block_group);
953 spin_unlock(&trans->transaction->dirty_bgs_lock);
954 mutex_unlock(&trans->transaction->cache_write_mutex);
956 ret = btrfs_remove_free_space_inode(trans, inode, block_group);
960 spin_lock(&fs_info->block_group_cache_lock);
961 rb_erase(&block_group->cache_node,
962 &fs_info->block_group_cache_tree);
963 RB_CLEAR_NODE(&block_group->cache_node);
965 /* Once for the block groups rbtree */
966 btrfs_put_block_group(block_group);
968 if (fs_info->first_logical_byte == block_group->start)
969 fs_info->first_logical_byte = (u64)-1;
970 spin_unlock(&fs_info->block_group_cache_lock);
972 down_write(&block_group->space_info->groups_sem);
974 * we must use list_del_init so people can check to see if they
975 * are still on the list after taking the semaphore
977 list_del_init(&block_group->list);
978 if (list_empty(&block_group->space_info->block_groups[index])) {
979 kobj = block_group->space_info->block_group_kobjs[index];
980 block_group->space_info->block_group_kobjs[index] = NULL;
981 clear_avail_alloc_bits(fs_info, block_group->flags);
983 up_write(&block_group->space_info->groups_sem);
984 clear_incompat_bg_bits(fs_info, block_group->flags);
990 if (block_group->has_caching_ctl)
991 caching_ctl = btrfs_get_caching_control(block_group);
992 if (block_group->cached == BTRFS_CACHE_STARTED)
993 btrfs_wait_block_group_cache_done(block_group);
994 if (block_group->has_caching_ctl) {
995 spin_lock(&fs_info->block_group_cache_lock);
997 struct btrfs_caching_control *ctl;
999 list_for_each_entry(ctl,
1000 &fs_info->caching_block_groups, list)
1001 if (ctl->block_group == block_group) {
1003 refcount_inc(&caching_ctl->count);
1008 list_del_init(&caching_ctl->list);
1009 spin_unlock(&fs_info->block_group_cache_lock);
1011 /* Once for the caching bgs list and once for us. */
1012 btrfs_put_caching_control(caching_ctl);
1013 btrfs_put_caching_control(caching_ctl);
1017 spin_lock(&trans->transaction->dirty_bgs_lock);
1018 WARN_ON(!list_empty(&block_group->dirty_list));
1019 WARN_ON(!list_empty(&block_group->io_list));
1020 spin_unlock(&trans->transaction->dirty_bgs_lock);
1022 btrfs_remove_free_space_cache(block_group);
1024 spin_lock(&block_group->space_info->lock);
1025 list_del_init(&block_group->ro_list);
1027 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
1028 WARN_ON(block_group->space_info->total_bytes
1029 < block_group->length);
1030 WARN_ON(block_group->space_info->bytes_readonly
1031 < block_group->length - block_group->zone_unusable);
1032 WARN_ON(block_group->space_info->bytes_zone_unusable
1033 < block_group->zone_unusable);
1034 WARN_ON(block_group->space_info->disk_total
1035 < block_group->length * factor);
1037 block_group->space_info->total_bytes -= block_group->length;
1038 block_group->space_info->bytes_readonly -=
1039 (block_group->length - block_group->zone_unusable);
1040 block_group->space_info->bytes_zone_unusable -=
1041 block_group->zone_unusable;
1042 block_group->space_info->disk_total -= block_group->length * factor;
1044 spin_unlock(&block_group->space_info->lock);
1047 * Remove the free space for the block group from the free space tree
1048 * and the block group's item from the extent tree before marking the
1049 * block group as removed. This is to prevent races with tasks that
1050 * freeze and unfreeze a block group, this task and another task
1051 * allocating a new block group - the unfreeze task ends up removing
1052 * the block group's extent map before the task calling this function
1053 * deletes the block group item from the extent tree, allowing for
1054 * another task to attempt to create another block group with the same
1055 * item key (and failing with -EEXIST and a transaction abort).
1057 ret = remove_block_group_free_space(trans, block_group);
1061 ret = remove_block_group_item(trans, path, block_group);
1065 spin_lock(&block_group->lock);
1066 block_group->removed = 1;
1068 * At this point trimming or scrub can't start on this block group,
1069 * because we removed the block group from the rbtree
1070 * fs_info->block_group_cache_tree so no one can't find it anymore and
1071 * even if someone already got this block group before we removed it
1072 * from the rbtree, they have already incremented block_group->frozen -
1073 * if they didn't, for the trimming case they won't find any free space
1074 * entries because we already removed them all when we called
1075 * btrfs_remove_free_space_cache().
1077 * And we must not remove the extent map from the fs_info->mapping_tree
1078 * to prevent the same logical address range and physical device space
1079 * ranges from being reused for a new block group. This is needed to
1080 * avoid races with trimming and scrub.
1082 * An fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
1083 * completely transactionless, so while it is trimming a range the
1084 * currently running transaction might finish and a new one start,
1085 * allowing for new block groups to be created that can reuse the same
1086 * physical device locations unless we take this special care.
1088 * There may also be an implicit trim operation if the file system
1089 * is mounted with -odiscard. The same protections must remain
1090 * in place until the extents have been discarded completely when
1091 * the transaction commit has completed.
1093 remove_em = (atomic_read(&block_group->frozen) == 0);
1094 spin_unlock(&block_group->lock);
1097 struct extent_map_tree *em_tree;
1099 em_tree = &fs_info->mapping_tree;
1100 write_lock(&em_tree->lock);
1101 remove_extent_mapping(em_tree, em);
1102 write_unlock(&em_tree->lock);
1103 /* once for the tree */
1104 free_extent_map(em);
1108 /* Once for the lookup reference */
1109 btrfs_put_block_group(block_group);
1111 btrfs_delayed_refs_rsv_release(fs_info, 1);
1112 btrfs_free_path(path);
1116 struct btrfs_trans_handle *btrfs_start_trans_remove_block_group(
1117 struct btrfs_fs_info *fs_info, const u64 chunk_offset)
1119 struct btrfs_root *root = btrfs_block_group_root(fs_info);
1120 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
1121 struct extent_map *em;
1122 struct map_lookup *map;
1123 unsigned int num_items;
1125 read_lock(&em_tree->lock);
1126 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1127 read_unlock(&em_tree->lock);
1128 ASSERT(em && em->start == chunk_offset);
1131 * We need to reserve 3 + N units from the metadata space info in order
1132 * to remove a block group (done at btrfs_remove_chunk() and at
1133 * btrfs_remove_block_group()), which are used for:
1135 * 1 unit for adding the free space inode's orphan (located in the tree
1137 * 1 unit for deleting the block group item (located in the extent
1139 * 1 unit for deleting the free space item (located in tree of tree
1141 * N units for deleting N device extent items corresponding to each
1142 * stripe (located in the device tree).
1144 * In order to remove a block group we also need to reserve units in the
1145 * system space info in order to update the chunk tree (update one or
1146 * more device items and remove one chunk item), but this is done at
1147 * btrfs_remove_chunk() through a call to check_system_chunk().
1149 map = em->map_lookup;
1150 num_items = 3 + map->num_stripes;
1151 free_extent_map(em);
1153 return btrfs_start_transaction_fallback_global_rsv(root, num_items);
1157 * Mark block group @cache read-only, so later write won't happen to block
1160 * If @force is not set, this function will only mark the block group readonly
1161 * if we have enough free space (1M) in other metadata/system block groups.
1162 * If @force is not set, this function will mark the block group readonly
1163 * without checking free space.
1165 * NOTE: This function doesn't care if other block groups can contain all the
1166 * data in this block group. That check should be done by relocation routine,
1167 * not this function.
1169 static int inc_block_group_ro(struct btrfs_block_group *cache, int force)
1171 struct btrfs_space_info *sinfo = cache->space_info;
1175 spin_lock(&sinfo->lock);
1176 spin_lock(&cache->lock);
1178 if (cache->swap_extents) {
1189 num_bytes = cache->length - cache->reserved - cache->pinned -
1190 cache->bytes_super - cache->zone_unusable - cache->used;
1193 * Data never overcommits, even in mixed mode, so do just the straight
1194 * check of left over space in how much we have allocated.
1198 } else if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA) {
1199 u64 sinfo_used = btrfs_space_info_used(sinfo, true);
1202 * Here we make sure if we mark this bg RO, we still have enough
1203 * free space as buffer.
1205 if (sinfo_used + num_bytes <= sinfo->total_bytes)
1209 * We overcommit metadata, so we need to do the
1210 * btrfs_can_overcommit check here, and we need to pass in
1211 * BTRFS_RESERVE_NO_FLUSH to give ourselves the most amount of
1212 * leeway to allow us to mark this block group as read only.
1214 if (btrfs_can_overcommit(cache->fs_info, sinfo, num_bytes,
1215 BTRFS_RESERVE_NO_FLUSH))
1220 sinfo->bytes_readonly += num_bytes;
1221 if (btrfs_is_zoned(cache->fs_info)) {
1222 /* Migrate zone_unusable bytes to readonly */
1223 sinfo->bytes_readonly += cache->zone_unusable;
1224 sinfo->bytes_zone_unusable -= cache->zone_unusable;
1225 cache->zone_unusable = 0;
1228 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
1231 spin_unlock(&cache->lock);
1232 spin_unlock(&sinfo->lock);
1233 if (ret == -ENOSPC && btrfs_test_opt(cache->fs_info, ENOSPC_DEBUG)) {
1234 btrfs_info(cache->fs_info,
1235 "unable to make block group %llu ro", cache->start);
1236 btrfs_dump_space_info(cache->fs_info, cache->space_info, 0, 0);
1241 static bool clean_pinned_extents(struct btrfs_trans_handle *trans,
1242 struct btrfs_block_group *bg)
1244 struct btrfs_fs_info *fs_info = bg->fs_info;
1245 struct btrfs_transaction *prev_trans = NULL;
1246 const u64 start = bg->start;
1247 const u64 end = start + bg->length - 1;
1250 spin_lock(&fs_info->trans_lock);
1251 if (trans->transaction->list.prev != &fs_info->trans_list) {
1252 prev_trans = list_last_entry(&trans->transaction->list,
1253 struct btrfs_transaction, list);
1254 refcount_inc(&prev_trans->use_count);
1256 spin_unlock(&fs_info->trans_lock);
1259 * Hold the unused_bg_unpin_mutex lock to avoid racing with
1260 * btrfs_finish_extent_commit(). If we are at transaction N, another
1261 * task might be running finish_extent_commit() for the previous
1262 * transaction N - 1, and have seen a range belonging to the block
1263 * group in pinned_extents before we were able to clear the whole block
1264 * group range from pinned_extents. This means that task can lookup for
1265 * the block group after we unpinned it from pinned_extents and removed
1266 * it, leading to a BUG_ON() at unpin_extent_range().
1268 mutex_lock(&fs_info->unused_bg_unpin_mutex);
1270 ret = clear_extent_bits(&prev_trans->pinned_extents, start, end,
1276 ret = clear_extent_bits(&trans->transaction->pinned_extents, start, end,
1279 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
1281 btrfs_put_transaction(prev_trans);
1287 * Process the unused_bgs list and remove any that don't have any allocated
1288 * space inside of them.
1290 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
1292 struct btrfs_block_group *block_group;
1293 struct btrfs_space_info *space_info;
1294 struct btrfs_trans_handle *trans;
1295 const bool async_trim_enabled = btrfs_test_opt(fs_info, DISCARD_ASYNC);
1298 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1302 * Long running balances can keep us blocked here for eternity, so
1303 * simply skip deletion if we're unable to get the mutex.
1305 if (!mutex_trylock(&fs_info->reclaim_bgs_lock))
1308 spin_lock(&fs_info->unused_bgs_lock);
1309 while (!list_empty(&fs_info->unused_bgs)) {
1312 block_group = list_first_entry(&fs_info->unused_bgs,
1313 struct btrfs_block_group,
1315 list_del_init(&block_group->bg_list);
1317 space_info = block_group->space_info;
1319 if (ret || btrfs_mixed_space_info(space_info)) {
1320 btrfs_put_block_group(block_group);
1323 spin_unlock(&fs_info->unused_bgs_lock);
1325 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group);
1327 /* Don't want to race with allocators so take the groups_sem */
1328 down_write(&space_info->groups_sem);
1331 * Async discard moves the final block group discard to be prior
1332 * to the unused_bgs code path. Therefore, if it's not fully
1333 * trimmed, punt it back to the async discard lists.
1335 if (btrfs_test_opt(fs_info, DISCARD_ASYNC) &&
1336 !btrfs_is_free_space_trimmed(block_group)) {
1337 trace_btrfs_skip_unused_block_group(block_group);
1338 up_write(&space_info->groups_sem);
1339 /* Requeue if we failed because of async discard */
1340 btrfs_discard_queue_work(&fs_info->discard_ctl,
1345 spin_lock(&block_group->lock);
1346 if (block_group->reserved || block_group->pinned ||
1347 block_group->used || block_group->ro ||
1348 list_is_singular(&block_group->list)) {
1350 * We want to bail if we made new allocations or have
1351 * outstanding allocations in this block group. We do
1352 * the ro check in case balance is currently acting on
1355 trace_btrfs_skip_unused_block_group(block_group);
1356 spin_unlock(&block_group->lock);
1357 up_write(&space_info->groups_sem);
1360 spin_unlock(&block_group->lock);
1362 /* We don't want to force the issue, only flip if it's ok. */
1363 ret = inc_block_group_ro(block_group, 0);
1364 up_write(&space_info->groups_sem);
1371 * Want to do this before we do anything else so we can recover
1372 * properly if we fail to join the transaction.
1374 trans = btrfs_start_trans_remove_block_group(fs_info,
1375 block_group->start);
1376 if (IS_ERR(trans)) {
1377 btrfs_dec_block_group_ro(block_group);
1378 ret = PTR_ERR(trans);
1383 * We could have pending pinned extents for this block group,
1384 * just delete them, we don't care about them anymore.
1386 if (!clean_pinned_extents(trans, block_group)) {
1387 btrfs_dec_block_group_ro(block_group);
1392 * At this point, the block_group is read only and should fail
1393 * new allocations. However, btrfs_finish_extent_commit() can
1394 * cause this block_group to be placed back on the discard
1395 * lists because now the block_group isn't fully discarded.
1396 * Bail here and try again later after discarding everything.
1398 spin_lock(&fs_info->discard_ctl.lock);
1399 if (!list_empty(&block_group->discard_list)) {
1400 spin_unlock(&fs_info->discard_ctl.lock);
1401 btrfs_dec_block_group_ro(block_group);
1402 btrfs_discard_queue_work(&fs_info->discard_ctl,
1406 spin_unlock(&fs_info->discard_ctl.lock);
1408 /* Reset pinned so btrfs_put_block_group doesn't complain */
1409 spin_lock(&space_info->lock);
1410 spin_lock(&block_group->lock);
1412 btrfs_space_info_update_bytes_pinned(fs_info, space_info,
1413 -block_group->pinned);
1414 space_info->bytes_readonly += block_group->pinned;
1415 block_group->pinned = 0;
1417 spin_unlock(&block_group->lock);
1418 spin_unlock(&space_info->lock);
1421 * The normal path here is an unused block group is passed here,
1422 * then trimming is handled in the transaction commit path.
1423 * Async discard interposes before this to do the trimming
1424 * before coming down the unused block group path as trimming
1425 * will no longer be done later in the transaction commit path.
1427 if (!async_trim_enabled && btrfs_test_opt(fs_info, DISCARD_ASYNC))
1431 * DISCARD can flip during remount. On zoned filesystems, we
1432 * need to reset sequential-required zones.
1434 trimming = btrfs_test_opt(fs_info, DISCARD_SYNC) ||
1435 btrfs_is_zoned(fs_info);
1437 /* Implicit trim during transaction commit. */
1439 btrfs_freeze_block_group(block_group);
1442 * Btrfs_remove_chunk will abort the transaction if things go
1445 ret = btrfs_remove_chunk(trans, block_group->start);
1449 btrfs_unfreeze_block_group(block_group);
1454 * If we're not mounted with -odiscard, we can just forget
1455 * about this block group. Otherwise we'll need to wait
1456 * until transaction commit to do the actual discard.
1459 spin_lock(&fs_info->unused_bgs_lock);
1461 * A concurrent scrub might have added us to the list
1462 * fs_info->unused_bgs, so use a list_move operation
1463 * to add the block group to the deleted_bgs list.
1465 list_move(&block_group->bg_list,
1466 &trans->transaction->deleted_bgs);
1467 spin_unlock(&fs_info->unused_bgs_lock);
1468 btrfs_get_block_group(block_group);
1471 btrfs_end_transaction(trans);
1473 btrfs_put_block_group(block_group);
1474 spin_lock(&fs_info->unused_bgs_lock);
1476 spin_unlock(&fs_info->unused_bgs_lock);
1477 mutex_unlock(&fs_info->reclaim_bgs_lock);
1481 btrfs_end_transaction(trans);
1482 mutex_unlock(&fs_info->reclaim_bgs_lock);
1483 btrfs_put_block_group(block_group);
1484 btrfs_discard_punt_unused_bgs_list(fs_info);
1487 void btrfs_mark_bg_unused(struct btrfs_block_group *bg)
1489 struct btrfs_fs_info *fs_info = bg->fs_info;
1491 spin_lock(&fs_info->unused_bgs_lock);
1492 if (list_empty(&bg->bg_list)) {
1493 btrfs_get_block_group(bg);
1494 trace_btrfs_add_unused_block_group(bg);
1495 list_add_tail(&bg->bg_list, &fs_info->unused_bgs);
1497 spin_unlock(&fs_info->unused_bgs_lock);
1501 * We want block groups with a low number of used bytes to be in the beginning
1502 * of the list, so they will get reclaimed first.
1504 static int reclaim_bgs_cmp(void *unused, const struct list_head *a,
1505 const struct list_head *b)
1507 const struct btrfs_block_group *bg1, *bg2;
1509 bg1 = list_entry(a, struct btrfs_block_group, bg_list);
1510 bg2 = list_entry(b, struct btrfs_block_group, bg_list);
1512 return bg1->used > bg2->used;
1515 void btrfs_reclaim_bgs_work(struct work_struct *work)
1517 struct btrfs_fs_info *fs_info =
1518 container_of(work, struct btrfs_fs_info, reclaim_bgs_work);
1519 struct btrfs_block_group *bg;
1520 struct btrfs_space_info *space_info;
1522 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1525 sb_start_write(fs_info->sb);
1527 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
1528 sb_end_write(fs_info->sb);
1533 * Long running balances can keep us blocked here for eternity, so
1534 * simply skip reclaim if we're unable to get the mutex.
1536 if (!mutex_trylock(&fs_info->reclaim_bgs_lock)) {
1537 btrfs_exclop_finish(fs_info);
1538 sb_end_write(fs_info->sb);
1542 spin_lock(&fs_info->unused_bgs_lock);
1544 * Sort happens under lock because we can't simply splice it and sort.
1545 * The block groups might still be in use and reachable via bg_list,
1546 * and their presence in the reclaim_bgs list must be preserved.
1548 list_sort(NULL, &fs_info->reclaim_bgs, reclaim_bgs_cmp);
1549 while (!list_empty(&fs_info->reclaim_bgs)) {
1553 bg = list_first_entry(&fs_info->reclaim_bgs,
1554 struct btrfs_block_group,
1556 list_del_init(&bg->bg_list);
1558 space_info = bg->space_info;
1559 spin_unlock(&fs_info->unused_bgs_lock);
1561 /* Don't race with allocators so take the groups_sem */
1562 down_write(&space_info->groups_sem);
1564 spin_lock(&bg->lock);
1565 if (bg->reserved || bg->pinned || bg->ro) {
1567 * We want to bail if we made new allocations or have
1568 * outstanding allocations in this block group. We do
1569 * the ro check in case balance is currently acting on
1572 spin_unlock(&bg->lock);
1573 up_write(&space_info->groups_sem);
1576 spin_unlock(&bg->lock);
1578 /* Get out fast, in case we're unmounting the filesystem */
1579 if (btrfs_fs_closing(fs_info)) {
1580 up_write(&space_info->groups_sem);
1585 * Cache the zone_unusable value before turning the block group
1586 * to read only. As soon as the blog group is read only it's
1587 * zone_unusable value gets moved to the block group's read-only
1588 * bytes and isn't available for calculations anymore.
1590 zone_unusable = bg->zone_unusable;
1591 ret = inc_block_group_ro(bg, 0);
1592 up_write(&space_info->groups_sem);
1597 "reclaiming chunk %llu with %llu%% used %llu%% unusable",
1598 bg->start, div_u64(bg->used * 100, bg->length),
1599 div64_u64(zone_unusable * 100, bg->length));
1600 trace_btrfs_reclaim_block_group(bg);
1601 ret = btrfs_relocate_chunk(fs_info, bg->start);
1603 btrfs_err(fs_info, "error relocating chunk %llu",
1607 btrfs_put_block_group(bg);
1608 spin_lock(&fs_info->unused_bgs_lock);
1610 spin_unlock(&fs_info->unused_bgs_lock);
1611 mutex_unlock(&fs_info->reclaim_bgs_lock);
1612 btrfs_exclop_finish(fs_info);
1613 sb_end_write(fs_info->sb);
1616 void btrfs_reclaim_bgs(struct btrfs_fs_info *fs_info)
1618 spin_lock(&fs_info->unused_bgs_lock);
1619 if (!list_empty(&fs_info->reclaim_bgs))
1620 queue_work(system_unbound_wq, &fs_info->reclaim_bgs_work);
1621 spin_unlock(&fs_info->unused_bgs_lock);
1624 void btrfs_mark_bg_to_reclaim(struct btrfs_block_group *bg)
1626 struct btrfs_fs_info *fs_info = bg->fs_info;
1628 spin_lock(&fs_info->unused_bgs_lock);
1629 if (list_empty(&bg->bg_list)) {
1630 btrfs_get_block_group(bg);
1631 trace_btrfs_add_reclaim_block_group(bg);
1632 list_add_tail(&bg->bg_list, &fs_info->reclaim_bgs);
1634 spin_unlock(&fs_info->unused_bgs_lock);
1637 static int read_bg_from_eb(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
1638 struct btrfs_path *path)
1640 struct extent_map_tree *em_tree;
1641 struct extent_map *em;
1642 struct btrfs_block_group_item bg;
1643 struct extent_buffer *leaf;
1648 slot = path->slots[0];
1649 leaf = path->nodes[0];
1651 em_tree = &fs_info->mapping_tree;
1652 read_lock(&em_tree->lock);
1653 em = lookup_extent_mapping(em_tree, key->objectid, key->offset);
1654 read_unlock(&em_tree->lock);
1657 "logical %llu len %llu found bg but no related chunk",
1658 key->objectid, key->offset);
1662 if (em->start != key->objectid || em->len != key->offset) {
1664 "block group %llu len %llu mismatch with chunk %llu len %llu",
1665 key->objectid, key->offset, em->start, em->len);
1670 read_extent_buffer(leaf, &bg, btrfs_item_ptr_offset(leaf, slot),
1672 flags = btrfs_stack_block_group_flags(&bg) &
1673 BTRFS_BLOCK_GROUP_TYPE_MASK;
1675 if (flags != (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
1677 "block group %llu len %llu type flags 0x%llx mismatch with chunk type flags 0x%llx",
1678 key->objectid, key->offset, flags,
1679 (BTRFS_BLOCK_GROUP_TYPE_MASK & em->map_lookup->type));
1684 free_extent_map(em);
1688 static int find_first_block_group(struct btrfs_fs_info *fs_info,
1689 struct btrfs_path *path,
1690 struct btrfs_key *key)
1692 struct btrfs_root *root = btrfs_block_group_root(fs_info);
1694 struct btrfs_key found_key;
1695 struct extent_buffer *leaf;
1698 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
1703 slot = path->slots[0];
1704 leaf = path->nodes[0];
1705 if (slot >= btrfs_header_nritems(leaf)) {
1706 ret = btrfs_next_leaf(root, path);
1713 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1715 if (found_key.objectid >= key->objectid &&
1716 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
1717 ret = read_bg_from_eb(fs_info, &found_key, path);
1727 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1729 u64 extra_flags = chunk_to_extended(flags) &
1730 BTRFS_EXTENDED_PROFILE_MASK;
1732 write_seqlock(&fs_info->profiles_lock);
1733 if (flags & BTRFS_BLOCK_GROUP_DATA)
1734 fs_info->avail_data_alloc_bits |= extra_flags;
1735 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1736 fs_info->avail_metadata_alloc_bits |= extra_flags;
1737 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1738 fs_info->avail_system_alloc_bits |= extra_flags;
1739 write_sequnlock(&fs_info->profiles_lock);
1743 * Map a physical disk address to a list of logical addresses
1745 * @fs_info: the filesystem
1746 * @chunk_start: logical address of block group
1747 * @bdev: physical device to resolve, can be NULL to indicate any device
1748 * @physical: physical address to map to logical addresses
1749 * @logical: return array of logical addresses which map to @physical
1750 * @naddrs: length of @logical
1751 * @stripe_len: size of IO stripe for the given block group
1753 * Maps a particular @physical disk address to a list of @logical addresses.
1754 * Used primarily to exclude those portions of a block group that contain super
1757 int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start,
1758 struct block_device *bdev, u64 physical, u64 **logical,
1759 int *naddrs, int *stripe_len)
1761 struct extent_map *em;
1762 struct map_lookup *map;
1765 u64 data_stripe_length;
1770 em = btrfs_get_chunk_map(fs_info, chunk_start, 1);
1774 map = em->map_lookup;
1775 data_stripe_length = em->orig_block_len;
1776 io_stripe_size = map->stripe_len;
1777 chunk_start = em->start;
1779 /* For RAID5/6 adjust to a full IO stripe length */
1780 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
1781 io_stripe_size = map->stripe_len * nr_data_stripes(map);
1783 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
1789 for (i = 0; i < map->num_stripes; i++) {
1790 bool already_inserted = false;
1795 if (!in_range(physical, map->stripes[i].physical,
1796 data_stripe_length))
1799 if (bdev && map->stripes[i].dev->bdev != bdev)
1802 stripe_nr = physical - map->stripes[i].physical;
1803 stripe_nr = div64_u64_rem(stripe_nr, map->stripe_len, &offset);
1805 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1806 stripe_nr = stripe_nr * map->num_stripes + i;
1807 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
1808 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
1809 stripe_nr = stripe_nr * map->num_stripes + i;
1812 * The remaining case would be for RAID56, multiply by
1813 * nr_data_stripes(). Alternatively, just use rmap_len below
1814 * instead of map->stripe_len
1817 bytenr = chunk_start + stripe_nr * io_stripe_size + offset;
1819 /* Ensure we don't add duplicate addresses */
1820 for (j = 0; j < nr; j++) {
1821 if (buf[j] == bytenr) {
1822 already_inserted = true;
1827 if (!already_inserted)
1833 *stripe_len = io_stripe_size;
1835 free_extent_map(em);
1839 static int exclude_super_stripes(struct btrfs_block_group *cache)
1841 struct btrfs_fs_info *fs_info = cache->fs_info;
1842 const bool zoned = btrfs_is_zoned(fs_info);
1848 if (cache->start < BTRFS_SUPER_INFO_OFFSET) {
1849 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->start;
1850 cache->bytes_super += stripe_len;
1851 ret = btrfs_add_excluded_extent(fs_info, cache->start,
1857 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1858 bytenr = btrfs_sb_offset(i);
1859 ret = btrfs_rmap_block(fs_info, cache->start, NULL,
1860 bytenr, &logical, &nr, &stripe_len);
1864 /* Shouldn't have super stripes in sequential zones */
1867 "zoned: block group %llu must not contain super block",
1873 u64 len = min_t(u64, stripe_len,
1874 cache->start + cache->length - logical[nr]);
1876 cache->bytes_super += len;
1877 ret = btrfs_add_excluded_extent(fs_info, logical[nr],
1890 static void link_block_group(struct btrfs_block_group *cache)
1892 struct btrfs_space_info *space_info = cache->space_info;
1893 int index = btrfs_bg_flags_to_raid_index(cache->flags);
1895 down_write(&space_info->groups_sem);
1896 list_add_tail(&cache->list, &space_info->block_groups[index]);
1897 up_write(&space_info->groups_sem);
1900 static struct btrfs_block_group *btrfs_create_block_group_cache(
1901 struct btrfs_fs_info *fs_info, u64 start)
1903 struct btrfs_block_group *cache;
1905 cache = kzalloc(sizeof(*cache), GFP_NOFS);
1909 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
1911 if (!cache->free_space_ctl) {
1916 cache->start = start;
1918 cache->fs_info = fs_info;
1919 cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start);
1921 cache->discard_index = BTRFS_DISCARD_INDEX_UNUSED;
1923 refcount_set(&cache->refs, 1);
1924 spin_lock_init(&cache->lock);
1925 init_rwsem(&cache->data_rwsem);
1926 INIT_LIST_HEAD(&cache->list);
1927 INIT_LIST_HEAD(&cache->cluster_list);
1928 INIT_LIST_HEAD(&cache->bg_list);
1929 INIT_LIST_HEAD(&cache->ro_list);
1930 INIT_LIST_HEAD(&cache->discard_list);
1931 INIT_LIST_HEAD(&cache->dirty_list);
1932 INIT_LIST_HEAD(&cache->io_list);
1933 INIT_LIST_HEAD(&cache->active_bg_list);
1934 btrfs_init_free_space_ctl(cache, cache->free_space_ctl);
1935 atomic_set(&cache->frozen, 0);
1936 mutex_init(&cache->free_space_lock);
1937 btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root);
1943 * Iterate all chunks and verify that each of them has the corresponding block
1946 static int check_chunk_block_group_mappings(struct btrfs_fs_info *fs_info)
1948 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
1949 struct extent_map *em;
1950 struct btrfs_block_group *bg;
1955 read_lock(&map_tree->lock);
1957 * lookup_extent_mapping will return the first extent map
1958 * intersecting the range, so setting @len to 1 is enough to
1959 * get the first chunk.
1961 em = lookup_extent_mapping(map_tree, start, 1);
1962 read_unlock(&map_tree->lock);
1966 bg = btrfs_lookup_block_group(fs_info, em->start);
1969 "chunk start=%llu len=%llu doesn't have corresponding block group",
1970 em->start, em->len);
1972 free_extent_map(em);
1975 if (bg->start != em->start || bg->length != em->len ||
1976 (bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK) !=
1977 (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
1979 "chunk start=%llu len=%llu flags=0x%llx doesn't match block group start=%llu len=%llu flags=0x%llx",
1981 em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK,
1982 bg->start, bg->length,
1983 bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK);
1985 free_extent_map(em);
1986 btrfs_put_block_group(bg);
1989 start = em->start + em->len;
1990 free_extent_map(em);
1991 btrfs_put_block_group(bg);
1996 static int read_one_block_group(struct btrfs_fs_info *info,
1997 struct btrfs_block_group_item *bgi,
1998 const struct btrfs_key *key,
2001 struct btrfs_block_group *cache;
2002 struct btrfs_space_info *space_info;
2003 const bool mixed = btrfs_fs_incompat(info, MIXED_GROUPS);
2006 ASSERT(key->type == BTRFS_BLOCK_GROUP_ITEM_KEY);
2008 cache = btrfs_create_block_group_cache(info, key->objectid);
2012 cache->length = key->offset;
2013 cache->used = btrfs_stack_block_group_used(bgi);
2014 cache->flags = btrfs_stack_block_group_flags(bgi);
2015 cache->global_root_id = btrfs_stack_block_group_chunk_objectid(bgi);
2017 set_free_space_tree_thresholds(cache);
2021 * When we mount with old space cache, we need to
2022 * set BTRFS_DC_CLEAR and set dirty flag.
2024 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
2025 * truncate the old free space cache inode and
2027 * b) Setting 'dirty flag' makes sure that we flush
2028 * the new space cache info onto disk.
2030 if (btrfs_test_opt(info, SPACE_CACHE))
2031 cache->disk_cache_state = BTRFS_DC_CLEAR;
2033 if (!mixed && ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) &&
2034 (cache->flags & BTRFS_BLOCK_GROUP_DATA))) {
2036 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
2042 ret = btrfs_load_block_group_zone_info(cache, false);
2044 btrfs_err(info, "zoned: failed to load zone info of bg %llu",
2050 * We need to exclude the super stripes now so that the space info has
2051 * super bytes accounted for, otherwise we'll think we have more space
2052 * than we actually do.
2054 ret = exclude_super_stripes(cache);
2056 /* We may have excluded something, so call this just in case. */
2057 btrfs_free_excluded_extents(cache);
2062 * For zoned filesystem, space after the allocation offset is the only
2063 * free space for a block group. So, we don't need any caching work.
2064 * btrfs_calc_zone_unusable() will set the amount of free space and
2065 * zone_unusable space.
2067 * For regular filesystem, check for two cases, either we are full, and
2068 * therefore don't need to bother with the caching work since we won't
2069 * find any space, or we are empty, and we can just add all the space
2070 * in and be done with it. This saves us _a_lot_ of time, particularly
2073 if (btrfs_is_zoned(info)) {
2074 btrfs_calc_zone_unusable(cache);
2075 /* Should not have any excluded extents. Just in case, though. */
2076 btrfs_free_excluded_extents(cache);
2077 } else if (cache->length == cache->used) {
2078 cache->last_byte_to_unpin = (u64)-1;
2079 cache->cached = BTRFS_CACHE_FINISHED;
2080 btrfs_free_excluded_extents(cache);
2081 } else if (cache->used == 0) {
2082 cache->last_byte_to_unpin = (u64)-1;
2083 cache->cached = BTRFS_CACHE_FINISHED;
2084 add_new_free_space(cache, cache->start,
2085 cache->start + cache->length);
2086 btrfs_free_excluded_extents(cache);
2089 ret = btrfs_add_block_group_cache(info, cache);
2091 btrfs_remove_free_space_cache(cache);
2094 trace_btrfs_add_block_group(info, cache, 0);
2095 btrfs_update_space_info(info, cache->flags, cache->length,
2096 cache->used, cache->bytes_super,
2097 cache->zone_unusable, &space_info);
2099 cache->space_info = space_info;
2101 link_block_group(cache);
2103 set_avail_alloc_bits(info, cache->flags);
2104 if (btrfs_chunk_writeable(info, cache->start)) {
2105 if (cache->used == 0) {
2106 ASSERT(list_empty(&cache->bg_list));
2107 if (btrfs_test_opt(info, DISCARD_ASYNC))
2108 btrfs_discard_queue_work(&info->discard_ctl, cache);
2110 btrfs_mark_bg_unused(cache);
2113 inc_block_group_ro(cache, 1);
2118 btrfs_put_block_group(cache);
2122 static int fill_dummy_bgs(struct btrfs_fs_info *fs_info)
2124 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
2125 struct btrfs_space_info *space_info;
2126 struct rb_node *node;
2129 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) {
2130 struct extent_map *em;
2131 struct map_lookup *map;
2132 struct btrfs_block_group *bg;
2134 em = rb_entry(node, struct extent_map, rb_node);
2135 map = em->map_lookup;
2136 bg = btrfs_create_block_group_cache(fs_info, em->start);
2142 /* Fill dummy cache as FULL */
2143 bg->length = em->len;
2144 bg->flags = map->type;
2145 bg->last_byte_to_unpin = (u64)-1;
2146 bg->cached = BTRFS_CACHE_FINISHED;
2148 bg->flags = map->type;
2149 ret = btrfs_add_block_group_cache(fs_info, bg);
2151 * We may have some valid block group cache added already, in
2152 * that case we skip to the next one.
2154 if (ret == -EEXIST) {
2156 btrfs_put_block_group(bg);
2161 btrfs_remove_free_space_cache(bg);
2162 btrfs_put_block_group(bg);
2166 btrfs_update_space_info(fs_info, bg->flags, em->len, em->len,
2168 bg->space_info = space_info;
2169 link_block_group(bg);
2171 set_avail_alloc_bits(fs_info, bg->flags);
2174 btrfs_init_global_block_rsv(fs_info);
2178 int btrfs_read_block_groups(struct btrfs_fs_info *info)
2180 struct btrfs_root *root = btrfs_block_group_root(info);
2181 struct btrfs_path *path;
2183 struct btrfs_block_group *cache;
2184 struct btrfs_space_info *space_info;
2185 struct btrfs_key key;
2190 return fill_dummy_bgs(info);
2194 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
2195 path = btrfs_alloc_path();
2199 cache_gen = btrfs_super_cache_generation(info->super_copy);
2200 if (btrfs_test_opt(info, SPACE_CACHE) &&
2201 btrfs_super_generation(info->super_copy) != cache_gen)
2203 if (btrfs_test_opt(info, CLEAR_CACHE))
2207 struct btrfs_block_group_item bgi;
2208 struct extent_buffer *leaf;
2211 ret = find_first_block_group(info, path, &key);
2217 leaf = path->nodes[0];
2218 slot = path->slots[0];
2220 read_extent_buffer(leaf, &bgi, btrfs_item_ptr_offset(leaf, slot),
2223 btrfs_item_key_to_cpu(leaf, &key, slot);
2224 btrfs_release_path(path);
2225 ret = read_one_block_group(info, &bgi, &key, need_clear);
2228 key.objectid += key.offset;
2231 btrfs_release_path(path);
2233 list_for_each_entry(space_info, &info->space_info, list) {
2236 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
2237 if (list_empty(&space_info->block_groups[i]))
2239 cache = list_first_entry(&space_info->block_groups[i],
2240 struct btrfs_block_group,
2242 btrfs_sysfs_add_block_group_type(cache);
2245 if (!(btrfs_get_alloc_profile(info, space_info->flags) &
2246 (BTRFS_BLOCK_GROUP_RAID10 |
2247 BTRFS_BLOCK_GROUP_RAID1_MASK |
2248 BTRFS_BLOCK_GROUP_RAID56_MASK |
2249 BTRFS_BLOCK_GROUP_DUP)))
2252 * Avoid allocating from un-mirrored block group if there are
2253 * mirrored block groups.
2255 list_for_each_entry(cache,
2256 &space_info->block_groups[BTRFS_RAID_RAID0],
2258 inc_block_group_ro(cache, 1);
2259 list_for_each_entry(cache,
2260 &space_info->block_groups[BTRFS_RAID_SINGLE],
2262 inc_block_group_ro(cache, 1);
2265 btrfs_init_global_block_rsv(info);
2266 ret = check_chunk_block_group_mappings(info);
2268 btrfs_free_path(path);
2270 * We've hit some error while reading the extent tree, and have
2271 * rescue=ibadroots mount option.
2272 * Try to fill the tree using dummy block groups so that the user can
2273 * continue to mount and grab their data.
2275 if (ret && btrfs_test_opt(info, IGNOREBADROOTS))
2276 ret = fill_dummy_bgs(info);
2281 * This function, insert_block_group_item(), belongs to the phase 2 of chunk
2284 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation
2287 static int insert_block_group_item(struct btrfs_trans_handle *trans,
2288 struct btrfs_block_group *block_group)
2290 struct btrfs_fs_info *fs_info = trans->fs_info;
2291 struct btrfs_block_group_item bgi;
2292 struct btrfs_root *root = btrfs_block_group_root(fs_info);
2293 struct btrfs_key key;
2295 spin_lock(&block_group->lock);
2296 btrfs_set_stack_block_group_used(&bgi, block_group->used);
2297 btrfs_set_stack_block_group_chunk_objectid(&bgi,
2298 block_group->global_root_id);
2299 btrfs_set_stack_block_group_flags(&bgi, block_group->flags);
2300 key.objectid = block_group->start;
2301 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
2302 key.offset = block_group->length;
2303 spin_unlock(&block_group->lock);
2305 return btrfs_insert_item(trans, root, &key, &bgi, sizeof(bgi));
2308 static int insert_dev_extent(struct btrfs_trans_handle *trans,
2309 struct btrfs_device *device, u64 chunk_offset,
2310 u64 start, u64 num_bytes)
2312 struct btrfs_fs_info *fs_info = device->fs_info;
2313 struct btrfs_root *root = fs_info->dev_root;
2314 struct btrfs_path *path;
2315 struct btrfs_dev_extent *extent;
2316 struct extent_buffer *leaf;
2317 struct btrfs_key key;
2320 WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state));
2321 WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
2322 path = btrfs_alloc_path();
2326 key.objectid = device->devid;
2327 key.type = BTRFS_DEV_EXTENT_KEY;
2329 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*extent));
2333 leaf = path->nodes[0];
2334 extent = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_extent);
2335 btrfs_set_dev_extent_chunk_tree(leaf, extent, BTRFS_CHUNK_TREE_OBJECTID);
2336 btrfs_set_dev_extent_chunk_objectid(leaf, extent,
2337 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
2338 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
2340 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
2341 btrfs_mark_buffer_dirty(leaf);
2343 btrfs_free_path(path);
2348 * This function belongs to phase 2.
2350 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation
2353 static int insert_dev_extents(struct btrfs_trans_handle *trans,
2354 u64 chunk_offset, u64 chunk_size)
2356 struct btrfs_fs_info *fs_info = trans->fs_info;
2357 struct btrfs_device *device;
2358 struct extent_map *em;
2359 struct map_lookup *map;
2365 em = btrfs_get_chunk_map(fs_info, chunk_offset, chunk_size);
2369 map = em->map_lookup;
2370 stripe_size = em->orig_block_len;
2373 * Take the device list mutex to prevent races with the final phase of
2374 * a device replace operation that replaces the device object associated
2375 * with the map's stripes, because the device object's id can change
2376 * at any time during that final phase of the device replace operation
2377 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
2378 * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID,
2379 * resulting in persisting a device extent item with such ID.
2381 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2382 for (i = 0; i < map->num_stripes; i++) {
2383 device = map->stripes[i].dev;
2384 dev_offset = map->stripes[i].physical;
2386 ret = insert_dev_extent(trans, device, chunk_offset, dev_offset,
2391 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2393 free_extent_map(em);
2398 * This function, btrfs_create_pending_block_groups(), belongs to the phase 2 of
2401 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation
2404 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans)
2406 struct btrfs_fs_info *fs_info = trans->fs_info;
2407 struct btrfs_block_group *block_group;
2410 while (!list_empty(&trans->new_bgs)) {
2413 block_group = list_first_entry(&trans->new_bgs,
2414 struct btrfs_block_group,
2419 index = btrfs_bg_flags_to_raid_index(block_group->flags);
2421 ret = insert_block_group_item(trans, block_group);
2423 btrfs_abort_transaction(trans, ret);
2424 if (!block_group->chunk_item_inserted) {
2425 mutex_lock(&fs_info->chunk_mutex);
2426 ret = btrfs_chunk_alloc_add_chunk_item(trans, block_group);
2427 mutex_unlock(&fs_info->chunk_mutex);
2429 btrfs_abort_transaction(trans, ret);
2431 ret = insert_dev_extents(trans, block_group->start,
2432 block_group->length);
2434 btrfs_abort_transaction(trans, ret);
2435 add_block_group_free_space(trans, block_group);
2438 * If we restriped during balance, we may have added a new raid
2439 * type, so now add the sysfs entries when it is safe to do so.
2440 * We don't have to worry about locking here as it's handled in
2441 * btrfs_sysfs_add_block_group_type.
2443 if (block_group->space_info->block_group_kobjs[index] == NULL)
2444 btrfs_sysfs_add_block_group_type(block_group);
2446 /* Already aborted the transaction if it failed. */
2448 btrfs_delayed_refs_rsv_release(fs_info, 1);
2449 list_del_init(&block_group->bg_list);
2451 btrfs_trans_release_chunk_metadata(trans);
2455 * For extent tree v2 we use the block_group_item->chunk_offset to point at our
2456 * global root id. For v1 it's always set to BTRFS_FIRST_CHUNK_TREE_OBJECTID.
2458 static u64 calculate_global_root_id(struct btrfs_fs_info *fs_info, u64 offset)
2463 if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
2464 return BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2466 /* If we have a smaller fs index based on 128MiB. */
2467 if (btrfs_super_total_bytes(fs_info->super_copy) <= (SZ_1G * 10ULL))
2470 offset = div64_u64(offset, div);
2471 div64_u64_rem(offset, fs_info->nr_global_roots, &index);
2475 struct btrfs_block_group *btrfs_make_block_group(struct btrfs_trans_handle *trans,
2476 u64 bytes_used, u64 type,
2477 u64 chunk_offset, u64 size)
2479 struct btrfs_fs_info *fs_info = trans->fs_info;
2480 struct btrfs_block_group *cache;
2483 btrfs_set_log_full_commit(trans);
2485 cache = btrfs_create_block_group_cache(fs_info, chunk_offset);
2487 return ERR_PTR(-ENOMEM);
2489 cache->length = size;
2490 set_free_space_tree_thresholds(cache);
2491 cache->used = bytes_used;
2492 cache->flags = type;
2493 cache->last_byte_to_unpin = (u64)-1;
2494 cache->cached = BTRFS_CACHE_FINISHED;
2495 cache->global_root_id = calculate_global_root_id(fs_info, cache->start);
2497 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
2498 cache->needs_free_space = 1;
2500 ret = btrfs_load_block_group_zone_info(cache, true);
2502 btrfs_put_block_group(cache);
2503 return ERR_PTR(ret);
2506 ret = exclude_super_stripes(cache);
2508 /* We may have excluded something, so call this just in case */
2509 btrfs_free_excluded_extents(cache);
2510 btrfs_put_block_group(cache);
2511 return ERR_PTR(ret);
2514 add_new_free_space(cache, chunk_offset, chunk_offset + size);
2516 btrfs_free_excluded_extents(cache);
2518 #ifdef CONFIG_BTRFS_DEBUG
2519 if (btrfs_should_fragment_free_space(cache)) {
2520 u64 new_bytes_used = size - bytes_used;
2522 bytes_used += new_bytes_used >> 1;
2523 fragment_free_space(cache);
2527 * Ensure the corresponding space_info object is created and
2528 * assigned to our block group. We want our bg to be added to the rbtree
2529 * with its ->space_info set.
2531 cache->space_info = btrfs_find_space_info(fs_info, cache->flags);
2532 ASSERT(cache->space_info);
2534 ret = btrfs_add_block_group_cache(fs_info, cache);
2536 btrfs_remove_free_space_cache(cache);
2537 btrfs_put_block_group(cache);
2538 return ERR_PTR(ret);
2542 * Now that our block group has its ->space_info set and is inserted in
2543 * the rbtree, update the space info's counters.
2545 trace_btrfs_add_block_group(fs_info, cache, 1);
2546 btrfs_update_space_info(fs_info, cache->flags, size, bytes_used,
2547 cache->bytes_super, cache->zone_unusable,
2548 &cache->space_info);
2549 btrfs_update_global_block_rsv(fs_info);
2551 link_block_group(cache);
2553 list_add_tail(&cache->bg_list, &trans->new_bgs);
2554 trans->delayed_ref_updates++;
2555 btrfs_update_delayed_refs_rsv(trans);
2557 set_avail_alloc_bits(fs_info, type);
2562 * Mark one block group RO, can be called several times for the same block
2565 * @cache: the destination block group
2566 * @do_chunk_alloc: whether need to do chunk pre-allocation, this is to
2567 * ensure we still have some free space after marking this
2570 int btrfs_inc_block_group_ro(struct btrfs_block_group *cache,
2571 bool do_chunk_alloc)
2573 struct btrfs_fs_info *fs_info = cache->fs_info;
2574 struct btrfs_trans_handle *trans;
2575 struct btrfs_root *root = btrfs_block_group_root(fs_info);
2578 bool dirty_bg_running;
2581 * This can only happen when we are doing read-only scrub on read-only
2583 * In that case we should not start a new transaction on read-only fs.
2584 * Thus here we skip all chunk allocations.
2586 if (sb_rdonly(fs_info->sb)) {
2587 mutex_lock(&fs_info->ro_block_group_mutex);
2588 ret = inc_block_group_ro(cache, 0);
2589 mutex_unlock(&fs_info->ro_block_group_mutex);
2594 trans = btrfs_join_transaction(root);
2596 return PTR_ERR(trans);
2598 dirty_bg_running = false;
2601 * We're not allowed to set block groups readonly after the dirty
2602 * block group cache has started writing. If it already started,
2603 * back off and let this transaction commit.
2605 mutex_lock(&fs_info->ro_block_group_mutex);
2606 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
2607 u64 transid = trans->transid;
2609 mutex_unlock(&fs_info->ro_block_group_mutex);
2610 btrfs_end_transaction(trans);
2612 ret = btrfs_wait_for_commit(fs_info, transid);
2615 dirty_bg_running = true;
2617 } while (dirty_bg_running);
2619 if (do_chunk_alloc) {
2621 * If we are changing raid levels, try to allocate a
2622 * corresponding block group with the new raid level.
2624 alloc_flags = btrfs_get_alloc_profile(fs_info, cache->flags);
2625 if (alloc_flags != cache->flags) {
2626 ret = btrfs_chunk_alloc(trans, alloc_flags,
2629 * ENOSPC is allowed here, we may have enough space
2630 * already allocated at the new raid level to carry on
2639 ret = inc_block_group_ro(cache, 0);
2640 if (!do_chunk_alloc || ret == -ETXTBSY)
2644 alloc_flags = btrfs_get_alloc_profile(fs_info, cache->space_info->flags);
2645 ret = btrfs_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE);
2648 ret = inc_block_group_ro(cache, 0);
2649 if (ret == -ETXTBSY)
2652 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
2653 alloc_flags = btrfs_get_alloc_profile(fs_info, cache->flags);
2654 mutex_lock(&fs_info->chunk_mutex);
2655 check_system_chunk(trans, alloc_flags);
2656 mutex_unlock(&fs_info->chunk_mutex);
2659 mutex_unlock(&fs_info->ro_block_group_mutex);
2661 btrfs_end_transaction(trans);
2665 void btrfs_dec_block_group_ro(struct btrfs_block_group *cache)
2667 struct btrfs_space_info *sinfo = cache->space_info;
2672 spin_lock(&sinfo->lock);
2673 spin_lock(&cache->lock);
2675 if (btrfs_is_zoned(cache->fs_info)) {
2676 /* Migrate zone_unusable bytes back */
2677 cache->zone_unusable =
2678 (cache->alloc_offset - cache->used) +
2679 (cache->length - cache->zone_capacity);
2680 sinfo->bytes_zone_unusable += cache->zone_unusable;
2681 sinfo->bytes_readonly -= cache->zone_unusable;
2683 num_bytes = cache->length - cache->reserved -
2684 cache->pinned - cache->bytes_super -
2685 cache->zone_unusable - cache->used;
2686 sinfo->bytes_readonly -= num_bytes;
2687 list_del_init(&cache->ro_list);
2689 spin_unlock(&cache->lock);
2690 spin_unlock(&sinfo->lock);
2693 static int update_block_group_item(struct btrfs_trans_handle *trans,
2694 struct btrfs_path *path,
2695 struct btrfs_block_group *cache)
2697 struct btrfs_fs_info *fs_info = trans->fs_info;
2699 struct btrfs_root *root = btrfs_block_group_root(fs_info);
2701 struct extent_buffer *leaf;
2702 struct btrfs_block_group_item bgi;
2703 struct btrfs_key key;
2705 key.objectid = cache->start;
2706 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
2707 key.offset = cache->length;
2709 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2716 leaf = path->nodes[0];
2717 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2718 btrfs_set_stack_block_group_used(&bgi, cache->used);
2719 btrfs_set_stack_block_group_chunk_objectid(&bgi,
2720 cache->global_root_id);
2721 btrfs_set_stack_block_group_flags(&bgi, cache->flags);
2722 write_extent_buffer(leaf, &bgi, bi, sizeof(bgi));
2723 btrfs_mark_buffer_dirty(leaf);
2725 btrfs_release_path(path);
2730 static int cache_save_setup(struct btrfs_block_group *block_group,
2731 struct btrfs_trans_handle *trans,
2732 struct btrfs_path *path)
2734 struct btrfs_fs_info *fs_info = block_group->fs_info;
2735 struct btrfs_root *root = fs_info->tree_root;
2736 struct inode *inode = NULL;
2737 struct extent_changeset *data_reserved = NULL;
2739 int dcs = BTRFS_DC_ERROR;
2744 if (!btrfs_test_opt(fs_info, SPACE_CACHE))
2748 * If this block group is smaller than 100 megs don't bother caching the
2751 if (block_group->length < (100 * SZ_1M)) {
2752 spin_lock(&block_group->lock);
2753 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2754 spin_unlock(&block_group->lock);
2758 if (TRANS_ABORTED(trans))
2761 inode = lookup_free_space_inode(block_group, path);
2762 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2763 ret = PTR_ERR(inode);
2764 btrfs_release_path(path);
2768 if (IS_ERR(inode)) {
2772 if (block_group->ro)
2775 ret = create_free_space_inode(trans, block_group, path);
2782 * We want to set the generation to 0, that way if anything goes wrong
2783 * from here on out we know not to trust this cache when we load up next
2786 BTRFS_I(inode)->generation = 0;
2787 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
2790 * So theoretically we could recover from this, simply set the
2791 * super cache generation to 0 so we know to invalidate the
2792 * cache, but then we'd have to keep track of the block groups
2793 * that fail this way so we know we _have_ to reset this cache
2794 * before the next commit or risk reading stale cache. So to
2795 * limit our exposure to horrible edge cases lets just abort the
2796 * transaction, this only happens in really bad situations
2799 btrfs_abort_transaction(trans, ret);
2804 /* We've already setup this transaction, go ahead and exit */
2805 if (block_group->cache_generation == trans->transid &&
2806 i_size_read(inode)) {
2807 dcs = BTRFS_DC_SETUP;
2811 if (i_size_read(inode) > 0) {
2812 ret = btrfs_check_trunc_cache_free_space(fs_info,
2813 &fs_info->global_block_rsv);
2817 ret = btrfs_truncate_free_space_cache(trans, NULL, inode);
2822 spin_lock(&block_group->lock);
2823 if (block_group->cached != BTRFS_CACHE_FINISHED ||
2824 !btrfs_test_opt(fs_info, SPACE_CACHE)) {
2826 * don't bother trying to write stuff out _if_
2827 * a) we're not cached,
2828 * b) we're with nospace_cache mount option,
2829 * c) we're with v2 space_cache (FREE_SPACE_TREE).
2831 dcs = BTRFS_DC_WRITTEN;
2832 spin_unlock(&block_group->lock);
2835 spin_unlock(&block_group->lock);
2838 * We hit an ENOSPC when setting up the cache in this transaction, just
2839 * skip doing the setup, we've already cleared the cache so we're safe.
2841 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
2847 * Try to preallocate enough space based on how big the block group is.
2848 * Keep in mind this has to include any pinned space which could end up
2849 * taking up quite a bit since it's not folded into the other space
2852 cache_size = div_u64(block_group->length, SZ_256M);
2857 cache_size *= fs_info->sectorsize;
2859 ret = btrfs_check_data_free_space(BTRFS_I(inode), &data_reserved, 0,
2864 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, cache_size,
2865 cache_size, cache_size,
2868 * Our cache requires contiguous chunks so that we don't modify a bunch
2869 * of metadata or split extents when writing the cache out, which means
2870 * we can enospc if we are heavily fragmented in addition to just normal
2871 * out of space conditions. So if we hit this just skip setting up any
2872 * other block groups for this transaction, maybe we'll unpin enough
2873 * space the next time around.
2876 dcs = BTRFS_DC_SETUP;
2877 else if (ret == -ENOSPC)
2878 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
2883 btrfs_release_path(path);
2885 spin_lock(&block_group->lock);
2886 if (!ret && dcs == BTRFS_DC_SETUP)
2887 block_group->cache_generation = trans->transid;
2888 block_group->disk_cache_state = dcs;
2889 spin_unlock(&block_group->lock);
2891 extent_changeset_free(data_reserved);
2895 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans)
2897 struct btrfs_fs_info *fs_info = trans->fs_info;
2898 struct btrfs_block_group *cache, *tmp;
2899 struct btrfs_transaction *cur_trans = trans->transaction;
2900 struct btrfs_path *path;
2902 if (list_empty(&cur_trans->dirty_bgs) ||
2903 !btrfs_test_opt(fs_info, SPACE_CACHE))
2906 path = btrfs_alloc_path();
2910 /* Could add new block groups, use _safe just in case */
2911 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
2913 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2914 cache_save_setup(cache, trans, path);
2917 btrfs_free_path(path);
2922 * Transaction commit does final block group cache writeback during a critical
2923 * section where nothing is allowed to change the FS. This is required in
2924 * order for the cache to actually match the block group, but can introduce a
2925 * lot of latency into the commit.
2927 * So, btrfs_start_dirty_block_groups is here to kick off block group cache IO.
2928 * There's a chance we'll have to redo some of it if the block group changes
2929 * again during the commit, but it greatly reduces the commit latency by
2930 * getting rid of the easy block groups while we're still allowing others to
2933 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans)
2935 struct btrfs_fs_info *fs_info = trans->fs_info;
2936 struct btrfs_block_group *cache;
2937 struct btrfs_transaction *cur_trans = trans->transaction;
2940 struct btrfs_path *path = NULL;
2942 struct list_head *io = &cur_trans->io_bgs;
2945 spin_lock(&cur_trans->dirty_bgs_lock);
2946 if (list_empty(&cur_trans->dirty_bgs)) {
2947 spin_unlock(&cur_trans->dirty_bgs_lock);
2950 list_splice_init(&cur_trans->dirty_bgs, &dirty);
2951 spin_unlock(&cur_trans->dirty_bgs_lock);
2954 /* Make sure all the block groups on our dirty list actually exist */
2955 btrfs_create_pending_block_groups(trans);
2958 path = btrfs_alloc_path();
2966 * cache_write_mutex is here only to save us from balance or automatic
2967 * removal of empty block groups deleting this block group while we are
2968 * writing out the cache
2970 mutex_lock(&trans->transaction->cache_write_mutex);
2971 while (!list_empty(&dirty)) {
2972 bool drop_reserve = true;
2974 cache = list_first_entry(&dirty, struct btrfs_block_group,
2977 * This can happen if something re-dirties a block group that
2978 * is already under IO. Just wait for it to finish and then do
2981 if (!list_empty(&cache->io_list)) {
2982 list_del_init(&cache->io_list);
2983 btrfs_wait_cache_io(trans, cache, path);
2984 btrfs_put_block_group(cache);
2989 * btrfs_wait_cache_io uses the cache->dirty_list to decide if
2990 * it should update the cache_state. Don't delete until after
2993 * Since we're not running in the commit critical section
2994 * we need the dirty_bgs_lock to protect from update_block_group
2996 spin_lock(&cur_trans->dirty_bgs_lock);
2997 list_del_init(&cache->dirty_list);
2998 spin_unlock(&cur_trans->dirty_bgs_lock);
3002 cache_save_setup(cache, trans, path);
3004 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3005 cache->io_ctl.inode = NULL;
3006 ret = btrfs_write_out_cache(trans, cache, path);
3007 if (ret == 0 && cache->io_ctl.inode) {
3011 * The cache_write_mutex is protecting the
3012 * io_list, also refer to the definition of
3013 * btrfs_transaction::io_bgs for more details
3015 list_add_tail(&cache->io_list, io);
3018 * If we failed to write the cache, the
3019 * generation will be bad and life goes on
3025 ret = update_block_group_item(trans, path, cache);
3027 * Our block group might still be attached to the list
3028 * of new block groups in the transaction handle of some
3029 * other task (struct btrfs_trans_handle->new_bgs). This
3030 * means its block group item isn't yet in the extent
3031 * tree. If this happens ignore the error, as we will
3032 * try again later in the critical section of the
3033 * transaction commit.
3035 if (ret == -ENOENT) {
3037 spin_lock(&cur_trans->dirty_bgs_lock);
3038 if (list_empty(&cache->dirty_list)) {
3039 list_add_tail(&cache->dirty_list,
3040 &cur_trans->dirty_bgs);
3041 btrfs_get_block_group(cache);
3042 drop_reserve = false;
3044 spin_unlock(&cur_trans->dirty_bgs_lock);
3046 btrfs_abort_transaction(trans, ret);
3050 /* If it's not on the io list, we need to put the block group */
3052 btrfs_put_block_group(cache);
3054 btrfs_delayed_refs_rsv_release(fs_info, 1);
3056 * Avoid blocking other tasks for too long. It might even save
3057 * us from writing caches for block groups that are going to be
3060 mutex_unlock(&trans->transaction->cache_write_mutex);
3063 mutex_lock(&trans->transaction->cache_write_mutex);
3065 mutex_unlock(&trans->transaction->cache_write_mutex);
3068 * Go through delayed refs for all the stuff we've just kicked off
3069 * and then loop back (just once)
3072 ret = btrfs_run_delayed_refs(trans, 0);
3073 if (!ret && loops == 0) {
3075 spin_lock(&cur_trans->dirty_bgs_lock);
3076 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3078 * dirty_bgs_lock protects us from concurrent block group
3079 * deletes too (not just cache_write_mutex).
3081 if (!list_empty(&dirty)) {
3082 spin_unlock(&cur_trans->dirty_bgs_lock);
3085 spin_unlock(&cur_trans->dirty_bgs_lock);
3089 spin_lock(&cur_trans->dirty_bgs_lock);
3090 list_splice_init(&dirty, &cur_trans->dirty_bgs);
3091 spin_unlock(&cur_trans->dirty_bgs_lock);
3092 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
3095 btrfs_free_path(path);
3099 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans)
3101 struct btrfs_fs_info *fs_info = trans->fs_info;
3102 struct btrfs_block_group *cache;
3103 struct btrfs_transaction *cur_trans = trans->transaction;
3106 struct btrfs_path *path;
3107 struct list_head *io = &cur_trans->io_bgs;
3109 path = btrfs_alloc_path();
3114 * Even though we are in the critical section of the transaction commit,
3115 * we can still have concurrent tasks adding elements to this
3116 * transaction's list of dirty block groups. These tasks correspond to
3117 * endio free space workers started when writeback finishes for a
3118 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3119 * allocate new block groups as a result of COWing nodes of the root
3120 * tree when updating the free space inode. The writeback for the space
3121 * caches is triggered by an earlier call to
3122 * btrfs_start_dirty_block_groups() and iterations of the following
3124 * Also we want to do the cache_save_setup first and then run the
3125 * delayed refs to make sure we have the best chance at doing this all
3128 spin_lock(&cur_trans->dirty_bgs_lock);
3129 while (!list_empty(&cur_trans->dirty_bgs)) {
3130 cache = list_first_entry(&cur_trans->dirty_bgs,
3131 struct btrfs_block_group,
3135 * This can happen if cache_save_setup re-dirties a block group
3136 * that is already under IO. Just wait for it to finish and
3137 * then do it all again
3139 if (!list_empty(&cache->io_list)) {
3140 spin_unlock(&cur_trans->dirty_bgs_lock);
3141 list_del_init(&cache->io_list);
3142 btrfs_wait_cache_io(trans, cache, path);
3143 btrfs_put_block_group(cache);
3144 spin_lock(&cur_trans->dirty_bgs_lock);
3148 * Don't remove from the dirty list until after we've waited on
3151 list_del_init(&cache->dirty_list);
3152 spin_unlock(&cur_trans->dirty_bgs_lock);
3155 cache_save_setup(cache, trans, path);
3158 ret = btrfs_run_delayed_refs(trans,
3159 (unsigned long) -1);
3161 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3162 cache->io_ctl.inode = NULL;
3163 ret = btrfs_write_out_cache(trans, cache, path);
3164 if (ret == 0 && cache->io_ctl.inode) {
3166 list_add_tail(&cache->io_list, io);
3169 * If we failed to write the cache, the
3170 * generation will be bad and life goes on
3176 ret = update_block_group_item(trans, path, cache);
3178 * One of the free space endio workers might have
3179 * created a new block group while updating a free space
3180 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3181 * and hasn't released its transaction handle yet, in
3182 * which case the new block group is still attached to
3183 * its transaction handle and its creation has not
3184 * finished yet (no block group item in the extent tree
3185 * yet, etc). If this is the case, wait for all free
3186 * space endio workers to finish and retry. This is a
3187 * very rare case so no need for a more efficient and
3190 if (ret == -ENOENT) {
3191 wait_event(cur_trans->writer_wait,
3192 atomic_read(&cur_trans->num_writers) == 1);
3193 ret = update_block_group_item(trans, path, cache);
3196 btrfs_abort_transaction(trans, ret);
3199 /* If its not on the io list, we need to put the block group */
3201 btrfs_put_block_group(cache);
3202 btrfs_delayed_refs_rsv_release(fs_info, 1);
3203 spin_lock(&cur_trans->dirty_bgs_lock);
3205 spin_unlock(&cur_trans->dirty_bgs_lock);
3208 * Refer to the definition of io_bgs member for details why it's safe
3209 * to use it without any locking
3211 while (!list_empty(io)) {
3212 cache = list_first_entry(io, struct btrfs_block_group,
3214 list_del_init(&cache->io_list);
3215 btrfs_wait_cache_io(trans, cache, path);
3216 btrfs_put_block_group(cache);
3219 btrfs_free_path(path);
3223 int btrfs_update_block_group(struct btrfs_trans_handle *trans,
3224 u64 bytenr, u64 num_bytes, bool alloc)
3226 struct btrfs_fs_info *info = trans->fs_info;
3227 struct btrfs_block_group *cache = NULL;
3228 u64 total = num_bytes;
3234 /* Block accounting for super block */
3235 spin_lock(&info->delalloc_root_lock);
3236 old_val = btrfs_super_bytes_used(info->super_copy);
3238 old_val += num_bytes;
3240 old_val -= num_bytes;
3241 btrfs_set_super_bytes_used(info->super_copy, old_val);
3242 spin_unlock(&info->delalloc_root_lock);
3245 cache = btrfs_lookup_block_group(info, bytenr);
3250 factor = btrfs_bg_type_to_factor(cache->flags);
3253 * If this block group has free space cache written out, we
3254 * need to make sure to load it if we are removing space. This
3255 * is because we need the unpinning stage to actually add the
3256 * space back to the block group, otherwise we will leak space.
3258 if (!alloc && !btrfs_block_group_done(cache))
3259 btrfs_cache_block_group(cache, 1);
3261 byte_in_group = bytenr - cache->start;
3262 WARN_ON(byte_in_group > cache->length);
3264 spin_lock(&cache->space_info->lock);
3265 spin_lock(&cache->lock);
3267 if (btrfs_test_opt(info, SPACE_CACHE) &&
3268 cache->disk_cache_state < BTRFS_DC_CLEAR)
3269 cache->disk_cache_state = BTRFS_DC_CLEAR;
3271 old_val = cache->used;
3272 num_bytes = min(total, cache->length - byte_in_group);
3274 old_val += num_bytes;
3275 cache->used = old_val;
3276 cache->reserved -= num_bytes;
3277 cache->space_info->bytes_reserved -= num_bytes;
3278 cache->space_info->bytes_used += num_bytes;
3279 cache->space_info->disk_used += num_bytes * factor;
3280 spin_unlock(&cache->lock);
3281 spin_unlock(&cache->space_info->lock);
3283 old_val -= num_bytes;
3284 cache->used = old_val;
3285 cache->pinned += num_bytes;
3286 btrfs_space_info_update_bytes_pinned(info,
3287 cache->space_info, num_bytes);
3288 cache->space_info->bytes_used -= num_bytes;
3289 cache->space_info->disk_used -= num_bytes * factor;
3290 spin_unlock(&cache->lock);
3291 spin_unlock(&cache->space_info->lock);
3293 set_extent_dirty(&trans->transaction->pinned_extents,
3294 bytenr, bytenr + num_bytes - 1,
3295 GFP_NOFS | __GFP_NOFAIL);
3298 spin_lock(&trans->transaction->dirty_bgs_lock);
3299 if (list_empty(&cache->dirty_list)) {
3300 list_add_tail(&cache->dirty_list,
3301 &trans->transaction->dirty_bgs);
3302 trans->delayed_ref_updates++;
3303 btrfs_get_block_group(cache);
3305 spin_unlock(&trans->transaction->dirty_bgs_lock);
3308 * No longer have used bytes in this block group, queue it for
3309 * deletion. We do this after adding the block group to the
3310 * dirty list to avoid races between cleaner kthread and space
3313 if (!alloc && old_val == 0) {
3314 if (!btrfs_test_opt(info, DISCARD_ASYNC))
3315 btrfs_mark_bg_unused(cache);
3318 btrfs_put_block_group(cache);
3320 bytenr += num_bytes;
3323 /* Modified block groups are accounted for in the delayed_refs_rsv. */
3324 btrfs_update_delayed_refs_rsv(trans);
3329 * btrfs_add_reserved_bytes - update the block_group and space info counters
3330 * @cache: The cache we are manipulating
3331 * @ram_bytes: The number of bytes of file content, and will be same to
3332 * @num_bytes except for the compress path.
3333 * @num_bytes: The number of bytes in question
3334 * @delalloc: The blocks are allocated for the delalloc write
3336 * This is called by the allocator when it reserves space. If this is a
3337 * reservation and the block group has become read only we cannot make the
3338 * reservation and return -EAGAIN, otherwise this function always succeeds.
3340 int btrfs_add_reserved_bytes(struct btrfs_block_group *cache,
3341 u64 ram_bytes, u64 num_bytes, int delalloc)
3343 struct btrfs_space_info *space_info = cache->space_info;
3346 spin_lock(&space_info->lock);
3347 spin_lock(&cache->lock);
3351 cache->reserved += num_bytes;
3352 space_info->bytes_reserved += num_bytes;
3353 trace_btrfs_space_reservation(cache->fs_info, "space_info",
3354 space_info->flags, num_bytes, 1);
3355 btrfs_space_info_update_bytes_may_use(cache->fs_info,
3356 space_info, -ram_bytes);
3358 cache->delalloc_bytes += num_bytes;
3361 * Compression can use less space than we reserved, so wake
3362 * tickets if that happens
3364 if (num_bytes < ram_bytes)
3365 btrfs_try_granting_tickets(cache->fs_info, space_info);
3367 spin_unlock(&cache->lock);
3368 spin_unlock(&space_info->lock);
3373 * btrfs_free_reserved_bytes - update the block_group and space info counters
3374 * @cache: The cache we are manipulating
3375 * @num_bytes: The number of bytes in question
3376 * @delalloc: The blocks are allocated for the delalloc write
3378 * This is called by somebody who is freeing space that was never actually used
3379 * on disk. For example if you reserve some space for a new leaf in transaction
3380 * A and before transaction A commits you free that leaf, you call this with
3381 * reserve set to 0 in order to clear the reservation.
3383 void btrfs_free_reserved_bytes(struct btrfs_block_group *cache,
3384 u64 num_bytes, int delalloc)
3386 struct btrfs_space_info *space_info = cache->space_info;
3388 spin_lock(&space_info->lock);
3389 spin_lock(&cache->lock);
3391 space_info->bytes_readonly += num_bytes;
3392 cache->reserved -= num_bytes;
3393 space_info->bytes_reserved -= num_bytes;
3394 space_info->max_extent_size = 0;
3397 cache->delalloc_bytes -= num_bytes;
3398 spin_unlock(&cache->lock);
3400 btrfs_try_granting_tickets(cache->fs_info, space_info);
3401 spin_unlock(&space_info->lock);
3404 static void force_metadata_allocation(struct btrfs_fs_info *info)
3406 struct list_head *head = &info->space_info;
3407 struct btrfs_space_info *found;
3409 list_for_each_entry(found, head, list) {
3410 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3411 found->force_alloc = CHUNK_ALLOC_FORCE;
3415 static int should_alloc_chunk(struct btrfs_fs_info *fs_info,
3416 struct btrfs_space_info *sinfo, int force)
3418 u64 bytes_used = btrfs_space_info_used(sinfo, false);
3421 if (force == CHUNK_ALLOC_FORCE)
3425 * in limited mode, we want to have some free space up to
3426 * about 1% of the FS size.
3428 if (force == CHUNK_ALLOC_LIMITED) {
3429 thresh = btrfs_super_total_bytes(fs_info->super_copy);
3430 thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1));
3432 if (sinfo->total_bytes - bytes_used < thresh)
3436 if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8))
3441 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, u64 type)
3443 u64 alloc_flags = btrfs_get_alloc_profile(trans->fs_info, type);
3445 return btrfs_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE);
3448 static struct btrfs_block_group *do_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags)
3450 struct btrfs_block_group *bg;
3454 * Check if we have enough space in the system space info because we
3455 * will need to update device items in the chunk btree and insert a new
3456 * chunk item in the chunk btree as well. This will allocate a new
3457 * system block group if needed.
3459 check_system_chunk(trans, flags);
3461 bg = btrfs_create_chunk(trans, flags);
3467 ret = btrfs_chunk_alloc_add_chunk_item(trans, bg);
3469 * Normally we are not expected to fail with -ENOSPC here, since we have
3470 * previously reserved space in the system space_info and allocated one
3471 * new system chunk if necessary. However there are three exceptions:
3473 * 1) We may have enough free space in the system space_info but all the
3474 * existing system block groups have a profile which can not be used
3475 * for extent allocation.
3477 * This happens when mounting in degraded mode. For example we have a
3478 * RAID1 filesystem with 2 devices, lose one device and mount the fs
3479 * using the other device in degraded mode. If we then allocate a chunk,
3480 * we may have enough free space in the existing system space_info, but
3481 * none of the block groups can be used for extent allocation since they
3482 * have a RAID1 profile, and because we are in degraded mode with a
3483 * single device, we are forced to allocate a new system chunk with a
3484 * SINGLE profile. Making check_system_chunk() iterate over all system
3485 * block groups and check if they have a usable profile and enough space
3486 * can be slow on very large filesystems, so we tolerate the -ENOSPC and
3487 * try again after forcing allocation of a new system chunk. Like this
3488 * we avoid paying the cost of that search in normal circumstances, when
3489 * we were not mounted in degraded mode;
3491 * 2) We had enough free space info the system space_info, and one suitable
3492 * block group to allocate from when we called check_system_chunk()
3493 * above. However right after we called it, the only system block group
3494 * with enough free space got turned into RO mode by a running scrub,
3495 * and in this case we have to allocate a new one and retry. We only
3496 * need do this allocate and retry once, since we have a transaction
3497 * handle and scrub uses the commit root to search for block groups;
3499 * 3) We had one system block group with enough free space when we called
3500 * check_system_chunk(), but after that, right before we tried to
3501 * allocate the last extent buffer we needed, a discard operation came
3502 * in and it temporarily removed the last free space entry from the
3503 * block group (discard removes a free space entry, discards it, and
3504 * then adds back the entry to the block group cache).
3506 if (ret == -ENOSPC) {
3507 const u64 sys_flags = btrfs_system_alloc_profile(trans->fs_info);
3508 struct btrfs_block_group *sys_bg;
3510 sys_bg = btrfs_create_chunk(trans, sys_flags);
3511 if (IS_ERR(sys_bg)) {
3512 ret = PTR_ERR(sys_bg);
3513 btrfs_abort_transaction(trans, ret);
3517 ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg);
3519 btrfs_abort_transaction(trans, ret);
3523 ret = btrfs_chunk_alloc_add_chunk_item(trans, bg);
3525 btrfs_abort_transaction(trans, ret);
3529 btrfs_abort_transaction(trans, ret);
3533 btrfs_trans_release_chunk_metadata(trans);
3536 return ERR_PTR(ret);
3538 btrfs_get_block_group(bg);
3543 * Chunk allocation is done in 2 phases:
3545 * 1) Phase 1 - through btrfs_chunk_alloc() we allocate device extents for
3546 * the chunk, the chunk mapping, create its block group and add the items
3547 * that belong in the chunk btree to it - more specifically, we need to
3548 * update device items in the chunk btree and add a new chunk item to it.
3550 * 2) Phase 2 - through btrfs_create_pending_block_groups(), we add the block
3551 * group item to the extent btree and the device extent items to the devices
3554 * This is done to prevent deadlocks. For example when COWing a node from the
3555 * extent btree we are holding a write lock on the node's parent and if we
3556 * trigger chunk allocation and attempted to insert the new block group item
3557 * in the extent btree right way, we could deadlock because the path for the
3558 * insertion can include that parent node. At first glance it seems impossible
3559 * to trigger chunk allocation after starting a transaction since tasks should
3560 * reserve enough transaction units (metadata space), however while that is true
3561 * most of the time, chunk allocation may still be triggered for several reasons:
3563 * 1) When reserving metadata, we check if there is enough free space in the
3564 * metadata space_info and therefore don't trigger allocation of a new chunk.
3565 * However later when the task actually tries to COW an extent buffer from
3566 * the extent btree or from the device btree for example, it is forced to
3567 * allocate a new block group (chunk) because the only one that had enough
3568 * free space was just turned to RO mode by a running scrub for example (or
3569 * device replace, block group reclaim thread, etc), so we can not use it
3570 * for allocating an extent and end up being forced to allocate a new one;
3572 * 2) Because we only check that the metadata space_info has enough free bytes,
3573 * we end up not allocating a new metadata chunk in that case. However if
3574 * the filesystem was mounted in degraded mode, none of the existing block
3575 * groups might be suitable for extent allocation due to their incompatible
3576 * profile (for e.g. mounting a 2 devices filesystem, where all block groups
3577 * use a RAID1 profile, in degraded mode using a single device). In this case
3578 * when the task attempts to COW some extent buffer of the extent btree for
3579 * example, it will trigger allocation of a new metadata block group with a
3580 * suitable profile (SINGLE profile in the example of the degraded mount of
3581 * the RAID1 filesystem);
3583 * 3) The task has reserved enough transaction units / metadata space, but when
3584 * it attempts to COW an extent buffer from the extent or device btree for
3585 * example, it does not find any free extent in any metadata block group,
3586 * therefore forced to try to allocate a new metadata block group.
3587 * This is because some other task allocated all available extents in the
3588 * meanwhile - this typically happens with tasks that don't reserve space
3589 * properly, either intentionally or as a bug. One example where this is
3590 * done intentionally is fsync, as it does not reserve any transaction units
3591 * and ends up allocating a variable number of metadata extents for log
3592 * tree extent buffers;
3594 * 4) The task has reserved enough transaction units / metadata space, but right
3595 * before it tries to allocate the last extent buffer it needs, a discard
3596 * operation comes in and, temporarily, removes the last free space entry from
3597 * the only metadata block group that had free space (discard starts by
3598 * removing a free space entry from a block group, then does the discard
3599 * operation and, once it's done, it adds back the free space entry to the
3602 * We also need this 2 phases setup when adding a device to a filesystem with
3603 * a seed device - we must create new metadata and system chunks without adding
3604 * any of the block group items to the chunk, extent and device btrees. If we
3605 * did not do it this way, we would get ENOSPC when attempting to update those
3606 * btrees, since all the chunks from the seed device are read-only.
3608 * Phase 1 does the updates and insertions to the chunk btree because if we had
3609 * it done in phase 2 and have a thundering herd of tasks allocating chunks in
3610 * parallel, we risk having too many system chunks allocated by many tasks if
3611 * many tasks reach phase 1 without the previous ones completing phase 2. In the
3612 * extreme case this leads to exhaustion of the system chunk array in the
3613 * superblock. This is easier to trigger if using a btree node/leaf size of 64K
3614 * and with RAID filesystems (so we have more device items in the chunk btree).
3615 * This has happened before and commit eafa4fd0ad0607 ("btrfs: fix exhaustion of
3616 * the system chunk array due to concurrent allocations") provides more details.
3618 * Allocation of system chunks does not happen through this function. A task that
3619 * needs to update the chunk btree (the only btree that uses system chunks), must
3620 * preallocate chunk space by calling either check_system_chunk() or
3621 * btrfs_reserve_chunk_metadata() - the former is used when allocating a data or
3622 * metadata chunk or when removing a chunk, while the later is used before doing
3623 * a modification to the chunk btree - use cases for the later are adding,
3624 * removing and resizing a device as well as relocation of a system chunk.
3625 * See the comment below for more details.
3627 * The reservation of system space, done through check_system_chunk(), as well
3628 * as all the updates and insertions into the chunk btree must be done while
3629 * holding fs_info->chunk_mutex. This is important to guarantee that while COWing
3630 * an extent buffer from the chunks btree we never trigger allocation of a new
3631 * system chunk, which would result in a deadlock (trying to lock twice an
3632 * extent buffer of the chunk btree, first time before triggering the chunk
3633 * allocation and the second time during chunk allocation while attempting to
3634 * update the chunks btree). The system chunk array is also updated while holding
3635 * that mutex. The same logic applies to removing chunks - we must reserve system
3636 * space, update the chunk btree and the system chunk array in the superblock
3637 * while holding fs_info->chunk_mutex.
3639 * This function, btrfs_chunk_alloc(), belongs to phase 1.
3641 * If @force is CHUNK_ALLOC_FORCE:
3642 * - return 1 if it successfully allocates a chunk,
3643 * - return errors including -ENOSPC otherwise.
3644 * If @force is NOT CHUNK_ALLOC_FORCE:
3645 * - return 0 if it doesn't need to allocate a new chunk,
3646 * - return 1 if it successfully allocates a chunk,
3647 * - return errors including -ENOSPC otherwise.
3649 int btrfs_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags,
3650 enum btrfs_chunk_alloc_enum force)
3652 struct btrfs_fs_info *fs_info = trans->fs_info;
3653 struct btrfs_space_info *space_info;
3654 struct btrfs_block_group *ret_bg;
3655 bool wait_for_alloc = false;
3656 bool should_alloc = false;
3657 bool from_extent_allocation = false;
3660 if (force == CHUNK_ALLOC_FORCE_FOR_EXTENT) {
3661 from_extent_allocation = true;
3662 force = CHUNK_ALLOC_FORCE;
3665 /* Don't re-enter if we're already allocating a chunk */
3666 if (trans->allocating_chunk)
3669 * Allocation of system chunks can not happen through this path, as we
3670 * could end up in a deadlock if we are allocating a data or metadata
3671 * chunk and there is another task modifying the chunk btree.
3673 * This is because while we are holding the chunk mutex, we will attempt
3674 * to add the new chunk item to the chunk btree or update an existing
3675 * device item in the chunk btree, while the other task that is modifying
3676 * the chunk btree is attempting to COW an extent buffer while holding a
3677 * lock on it and on its parent - if the COW operation triggers a system
3678 * chunk allocation, then we can deadlock because we are holding the
3679 * chunk mutex and we may need to access that extent buffer or its parent
3680 * in order to add the chunk item or update a device item.
3682 * Tasks that want to modify the chunk tree should reserve system space
3683 * before updating the chunk btree, by calling either
3684 * btrfs_reserve_chunk_metadata() or check_system_chunk().
3685 * It's possible that after a task reserves the space, it still ends up
3686 * here - this happens in the cases described above at do_chunk_alloc().
3687 * The task will have to either retry or fail.
3689 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3692 space_info = btrfs_find_space_info(fs_info, flags);
3696 spin_lock(&space_info->lock);
3697 if (force < space_info->force_alloc)
3698 force = space_info->force_alloc;
3699 should_alloc = should_alloc_chunk(fs_info, space_info, force);
3700 if (space_info->full) {
3701 /* No more free physical space */
3706 spin_unlock(&space_info->lock);
3708 } else if (!should_alloc) {
3709 spin_unlock(&space_info->lock);
3711 } else if (space_info->chunk_alloc) {
3713 * Someone is already allocating, so we need to block
3714 * until this someone is finished and then loop to
3715 * recheck if we should continue with our allocation
3718 wait_for_alloc = true;
3719 spin_unlock(&space_info->lock);
3720 mutex_lock(&fs_info->chunk_mutex);
3721 mutex_unlock(&fs_info->chunk_mutex);
3723 /* Proceed with allocation */
3724 space_info->chunk_alloc = 1;
3725 wait_for_alloc = false;
3726 spin_unlock(&space_info->lock);
3730 } while (wait_for_alloc);
3732 mutex_lock(&fs_info->chunk_mutex);
3733 trans->allocating_chunk = true;
3736 * If we have mixed data/metadata chunks we want to make sure we keep
3737 * allocating mixed chunks instead of individual chunks.
3739 if (btrfs_mixed_space_info(space_info))
3740 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3743 * if we're doing a data chunk, go ahead and make sure that
3744 * we keep a reasonable number of metadata chunks allocated in the
3747 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3748 fs_info->data_chunk_allocations++;
3749 if (!(fs_info->data_chunk_allocations %
3750 fs_info->metadata_ratio))
3751 force_metadata_allocation(fs_info);
3754 ret_bg = do_chunk_alloc(trans, flags);
3755 trans->allocating_chunk = false;
3757 if (IS_ERR(ret_bg)) {
3758 ret = PTR_ERR(ret_bg);
3759 } else if (from_extent_allocation) {
3761 * New block group is likely to be used soon. Try to activate
3762 * it now. Failure is OK for now.
3764 btrfs_zone_activate(ret_bg);
3768 btrfs_put_block_group(ret_bg);
3770 spin_lock(&space_info->lock);
3773 space_info->full = 1;
3778 space_info->max_extent_size = 0;
3781 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3783 space_info->chunk_alloc = 0;
3784 spin_unlock(&space_info->lock);
3785 mutex_unlock(&fs_info->chunk_mutex);
3790 static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type)
3794 num_dev = btrfs_raid_array[btrfs_bg_flags_to_raid_index(type)].devs_max;
3796 num_dev = fs_info->fs_devices->rw_devices;
3801 static void reserve_chunk_space(struct btrfs_trans_handle *trans,
3805 struct btrfs_fs_info *fs_info = trans->fs_info;
3806 struct btrfs_space_info *info;
3811 * Needed because we can end up allocating a system chunk and for an
3812 * atomic and race free space reservation in the chunk block reserve.
3814 lockdep_assert_held(&fs_info->chunk_mutex);
3816 info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3817 spin_lock(&info->lock);
3818 left = info->total_bytes - btrfs_space_info_used(info, true);
3819 spin_unlock(&info->lock);
3821 if (left < bytes && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
3822 btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu",
3824 btrfs_dump_space_info(fs_info, info, 0, 0);
3828 u64 flags = btrfs_system_alloc_profile(fs_info);
3829 struct btrfs_block_group *bg;
3832 * Ignore failure to create system chunk. We might end up not
3833 * needing it, as we might not need to COW all nodes/leafs from
3834 * the paths we visit in the chunk tree (they were already COWed
3835 * or created in the current transaction for example).
3837 bg = btrfs_create_chunk(trans, flags);
3842 * If we fail to add the chunk item here, we end up
3843 * trying again at phase 2 of chunk allocation, at
3844 * btrfs_create_pending_block_groups(). So ignore
3845 * any error here. An ENOSPC here could happen, due to
3846 * the cases described at do_chunk_alloc() - the system
3847 * block group we just created was just turned into RO
3848 * mode by a scrub for example, or a running discard
3849 * temporarily removed its free space entries, etc.
3851 btrfs_chunk_alloc_add_chunk_item(trans, bg);
3856 ret = btrfs_block_rsv_add(fs_info,
3857 &fs_info->chunk_block_rsv,
3858 bytes, BTRFS_RESERVE_NO_FLUSH);
3860 trans->chunk_bytes_reserved += bytes;
3865 * Reserve space in the system space for allocating or removing a chunk.
3866 * The caller must be holding fs_info->chunk_mutex.
3868 void check_system_chunk(struct btrfs_trans_handle *trans, u64 type)
3870 struct btrfs_fs_info *fs_info = trans->fs_info;
3871 const u64 num_devs = get_profile_num_devs(fs_info, type);
3874 /* num_devs device items to update and 1 chunk item to add or remove. */
3875 bytes = btrfs_calc_metadata_size(fs_info, num_devs) +
3876 btrfs_calc_insert_metadata_size(fs_info, 1);
3878 reserve_chunk_space(trans, bytes, type);
3882 * Reserve space in the system space, if needed, for doing a modification to the
3885 * @trans: A transaction handle.
3886 * @is_item_insertion: Indicate if the modification is for inserting a new item
3887 * in the chunk btree or if it's for the deletion or update
3888 * of an existing item.
3890 * This is used in a context where we need to update the chunk btree outside
3891 * block group allocation and removal, to avoid a deadlock with a concurrent
3892 * task that is allocating a metadata or data block group and therefore needs to
3893 * update the chunk btree while holding the chunk mutex. After the update to the
3894 * chunk btree is done, btrfs_trans_release_chunk_metadata() should be called.
3897 void btrfs_reserve_chunk_metadata(struct btrfs_trans_handle *trans,
3898 bool is_item_insertion)
3900 struct btrfs_fs_info *fs_info = trans->fs_info;
3903 if (is_item_insertion)
3904 bytes = btrfs_calc_insert_metadata_size(fs_info, 1);
3906 bytes = btrfs_calc_metadata_size(fs_info, 1);
3908 mutex_lock(&fs_info->chunk_mutex);
3909 reserve_chunk_space(trans, bytes, BTRFS_BLOCK_GROUP_SYSTEM);
3910 mutex_unlock(&fs_info->chunk_mutex);
3913 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
3915 struct btrfs_block_group *block_group;
3919 struct inode *inode;
3921 block_group = btrfs_lookup_first_block_group(info, last);
3922 while (block_group) {
3923 btrfs_wait_block_group_cache_done(block_group);
3924 spin_lock(&block_group->lock);
3925 if (block_group->iref)
3927 spin_unlock(&block_group->lock);
3928 block_group = btrfs_next_block_group(block_group);
3937 inode = block_group->inode;
3938 block_group->iref = 0;
3939 block_group->inode = NULL;
3940 spin_unlock(&block_group->lock);
3941 ASSERT(block_group->io_ctl.inode == NULL);
3943 last = block_group->start + block_group->length;
3944 btrfs_put_block_group(block_group);
3949 * Must be called only after stopping all workers, since we could have block
3950 * group caching kthreads running, and therefore they could race with us if we
3951 * freed the block groups before stopping them.
3953 int btrfs_free_block_groups(struct btrfs_fs_info *info)
3955 struct btrfs_block_group *block_group;
3956 struct btrfs_space_info *space_info;
3957 struct btrfs_caching_control *caching_ctl;
3960 spin_lock(&info->block_group_cache_lock);
3961 while (!list_empty(&info->caching_block_groups)) {
3962 caching_ctl = list_entry(info->caching_block_groups.next,
3963 struct btrfs_caching_control, list);
3964 list_del(&caching_ctl->list);
3965 btrfs_put_caching_control(caching_ctl);
3967 spin_unlock(&info->block_group_cache_lock);
3969 spin_lock(&info->unused_bgs_lock);
3970 while (!list_empty(&info->unused_bgs)) {
3971 block_group = list_first_entry(&info->unused_bgs,
3972 struct btrfs_block_group,
3974 list_del_init(&block_group->bg_list);
3975 btrfs_put_block_group(block_group);
3978 while (!list_empty(&info->reclaim_bgs)) {
3979 block_group = list_first_entry(&info->reclaim_bgs,
3980 struct btrfs_block_group,
3982 list_del_init(&block_group->bg_list);
3983 btrfs_put_block_group(block_group);
3985 spin_unlock(&info->unused_bgs_lock);
3987 spin_lock(&info->zone_active_bgs_lock);
3988 while (!list_empty(&info->zone_active_bgs)) {
3989 block_group = list_first_entry(&info->zone_active_bgs,
3990 struct btrfs_block_group,
3992 list_del_init(&block_group->active_bg_list);
3993 btrfs_put_block_group(block_group);
3995 spin_unlock(&info->zone_active_bgs_lock);
3997 spin_lock(&info->block_group_cache_lock);
3998 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
3999 block_group = rb_entry(n, struct btrfs_block_group,
4001 rb_erase(&block_group->cache_node,
4002 &info->block_group_cache_tree);
4003 RB_CLEAR_NODE(&block_group->cache_node);
4004 spin_unlock(&info->block_group_cache_lock);
4006 down_write(&block_group->space_info->groups_sem);
4007 list_del(&block_group->list);
4008 up_write(&block_group->space_info->groups_sem);
4011 * We haven't cached this block group, which means we could
4012 * possibly have excluded extents on this block group.
4014 if (block_group->cached == BTRFS_CACHE_NO ||
4015 block_group->cached == BTRFS_CACHE_ERROR)
4016 btrfs_free_excluded_extents(block_group);
4018 btrfs_remove_free_space_cache(block_group);
4019 ASSERT(block_group->cached != BTRFS_CACHE_STARTED);
4020 ASSERT(list_empty(&block_group->dirty_list));
4021 ASSERT(list_empty(&block_group->io_list));
4022 ASSERT(list_empty(&block_group->bg_list));
4023 ASSERT(refcount_read(&block_group->refs) == 1);
4024 ASSERT(block_group->swap_extents == 0);
4025 btrfs_put_block_group(block_group);
4027 spin_lock(&info->block_group_cache_lock);
4029 spin_unlock(&info->block_group_cache_lock);
4031 btrfs_release_global_block_rsv(info);
4033 while (!list_empty(&info->space_info)) {
4034 space_info = list_entry(info->space_info.next,
4035 struct btrfs_space_info,
4039 * Do not hide this behind enospc_debug, this is actually
4040 * important and indicates a real bug if this happens.
4042 if (WARN_ON(space_info->bytes_pinned > 0 ||
4043 space_info->bytes_may_use > 0))
4044 btrfs_dump_space_info(info, space_info, 0, 0);
4047 * If there was a failure to cleanup a log tree, very likely due
4048 * to an IO failure on a writeback attempt of one or more of its
4049 * extent buffers, we could not do proper (and cheap) unaccounting
4050 * of their reserved space, so don't warn on bytes_reserved > 0 in
4053 if (!(space_info->flags & BTRFS_BLOCK_GROUP_METADATA) ||
4054 !BTRFS_FS_LOG_CLEANUP_ERROR(info)) {
4055 if (WARN_ON(space_info->bytes_reserved > 0))
4056 btrfs_dump_space_info(info, space_info, 0, 0);
4059 WARN_ON(space_info->reclaim_size > 0);
4060 list_del(&space_info->list);
4061 btrfs_sysfs_remove_space_info(space_info);
4066 void btrfs_freeze_block_group(struct btrfs_block_group *cache)
4068 atomic_inc(&cache->frozen);
4071 void btrfs_unfreeze_block_group(struct btrfs_block_group *block_group)
4073 struct btrfs_fs_info *fs_info = block_group->fs_info;
4074 struct extent_map_tree *em_tree;
4075 struct extent_map *em;
4078 spin_lock(&block_group->lock);
4079 cleanup = (atomic_dec_and_test(&block_group->frozen) &&
4080 block_group->removed);
4081 spin_unlock(&block_group->lock);
4084 em_tree = &fs_info->mapping_tree;
4085 write_lock(&em_tree->lock);
4086 em = lookup_extent_mapping(em_tree, block_group->start,
4088 BUG_ON(!em); /* logic error, can't happen */
4089 remove_extent_mapping(em_tree, em);
4090 write_unlock(&em_tree->lock);
4092 /* once for us and once for the tree */
4093 free_extent_map(em);
4094 free_extent_map(em);
4097 * We may have left one free space entry and other possible
4098 * tasks trimming this block group have left 1 entry each one.
4101 __btrfs_remove_free_space_cache(block_group->free_space_ctl);
4105 bool btrfs_inc_block_group_swap_extents(struct btrfs_block_group *bg)
4109 spin_lock(&bg->lock);
4114 spin_unlock(&bg->lock);
4119 void btrfs_dec_block_group_swap_extents(struct btrfs_block_group *bg, int amount)
4121 spin_lock(&bg->lock);
4123 ASSERT(bg->swap_extents >= amount);
4124 bg->swap_extents -= amount;
4125 spin_unlock(&bg->lock);