1 // SPDX-License-Identifier: GPL-2.0
5 #include "block-group.h"
6 #include "space-info.h"
8 #include "free-space-cache.h"
9 #include "free-space-tree.h"
11 #include "transaction.h"
12 #include "ref-verify.h"
15 #include "delalloc-space.h"
21 * Return target flags in extended format or 0 if restripe for this chunk_type
24 * Should be called with balance_lock held
26 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
28 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
34 if (flags & BTRFS_BLOCK_GROUP_DATA &&
35 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
36 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
37 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
38 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
39 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
40 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
41 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
42 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
49 * @flags: available profiles in extended format (see ctree.h)
51 * Return reduced profile in chunk format. If profile changing is in progress
52 * (either running or paused) picks the target profile (if it's already
53 * available), otherwise falls back to plain reducing.
55 static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags)
57 u64 num_devices = fs_info->fs_devices->rw_devices;
63 * See if restripe for this chunk_type is in progress, if so try to
64 * reduce to the target profile
66 spin_lock(&fs_info->balance_lock);
67 target = get_restripe_target(fs_info, flags);
69 spin_unlock(&fs_info->balance_lock);
70 return extended_to_chunk(target);
72 spin_unlock(&fs_info->balance_lock);
74 /* First, mask out the RAID levels which aren't possible */
75 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
76 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
77 allowed |= btrfs_raid_array[raid_type].bg_flag;
81 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
82 allowed = BTRFS_BLOCK_GROUP_RAID6;
83 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
84 allowed = BTRFS_BLOCK_GROUP_RAID5;
85 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
86 allowed = BTRFS_BLOCK_GROUP_RAID10;
87 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
88 allowed = BTRFS_BLOCK_GROUP_RAID1;
89 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
90 allowed = BTRFS_BLOCK_GROUP_RAID0;
92 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
94 return extended_to_chunk(flags | allowed);
97 u64 btrfs_get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags)
104 seq = read_seqbegin(&fs_info->profiles_lock);
106 if (flags & BTRFS_BLOCK_GROUP_DATA)
107 flags |= fs_info->avail_data_alloc_bits;
108 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
109 flags |= fs_info->avail_system_alloc_bits;
110 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
111 flags |= fs_info->avail_metadata_alloc_bits;
112 } while (read_seqretry(&fs_info->profiles_lock, seq));
114 return btrfs_reduce_alloc_profile(fs_info, flags);
117 void btrfs_get_block_group(struct btrfs_block_group *cache)
119 refcount_inc(&cache->refs);
122 void btrfs_put_block_group(struct btrfs_block_group *cache)
124 if (refcount_dec_and_test(&cache->refs)) {
125 WARN_ON(cache->pinned > 0);
126 WARN_ON(cache->reserved > 0);
129 * A block_group shouldn't be on the discard_list anymore.
130 * Remove the block_group from the discard_list to prevent us
131 * from causing a panic due to NULL pointer dereference.
133 if (WARN_ON(!list_empty(&cache->discard_list)))
134 btrfs_discard_cancel_work(&cache->fs_info->discard_ctl,
138 * If not empty, someone is still holding mutex of
139 * full_stripe_lock, which can only be released by caller.
140 * And it will definitely cause use-after-free when caller
141 * tries to release full stripe lock.
143 * No better way to resolve, but only to warn.
145 WARN_ON(!RB_EMPTY_ROOT(&cache->full_stripe_locks_root.root));
146 kfree(cache->free_space_ctl);
152 * This adds the block group to the fs_info rb tree for the block group cache
154 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
155 struct btrfs_block_group *block_group)
158 struct rb_node *parent = NULL;
159 struct btrfs_block_group *cache;
161 ASSERT(block_group->length != 0);
163 spin_lock(&info->block_group_cache_lock);
164 p = &info->block_group_cache_tree.rb_node;
168 cache = rb_entry(parent, struct btrfs_block_group, cache_node);
169 if (block_group->start < cache->start) {
171 } else if (block_group->start > cache->start) {
174 spin_unlock(&info->block_group_cache_lock);
179 rb_link_node(&block_group->cache_node, parent, p);
180 rb_insert_color(&block_group->cache_node,
181 &info->block_group_cache_tree);
183 if (info->first_logical_byte > block_group->start)
184 info->first_logical_byte = block_group->start;
186 spin_unlock(&info->block_group_cache_lock);
192 * This will return the block group at or after bytenr if contains is 0, else
193 * it will return the block group that contains the bytenr
195 static struct btrfs_block_group *block_group_cache_tree_search(
196 struct btrfs_fs_info *info, u64 bytenr, int contains)
198 struct btrfs_block_group *cache, *ret = NULL;
202 spin_lock(&info->block_group_cache_lock);
203 n = info->block_group_cache_tree.rb_node;
206 cache = rb_entry(n, struct btrfs_block_group, cache_node);
207 end = cache->start + cache->length - 1;
208 start = cache->start;
210 if (bytenr < start) {
211 if (!contains && (!ret || start < ret->start))
214 } else if (bytenr > start) {
215 if (contains && bytenr <= end) {
226 btrfs_get_block_group(ret);
227 if (bytenr == 0 && info->first_logical_byte > ret->start)
228 info->first_logical_byte = ret->start;
230 spin_unlock(&info->block_group_cache_lock);
236 * Return the block group that starts at or after bytenr
238 struct btrfs_block_group *btrfs_lookup_first_block_group(
239 struct btrfs_fs_info *info, u64 bytenr)
241 return block_group_cache_tree_search(info, bytenr, 0);
245 * Return the block group that contains the given bytenr
247 struct btrfs_block_group *btrfs_lookup_block_group(
248 struct btrfs_fs_info *info, u64 bytenr)
250 return block_group_cache_tree_search(info, bytenr, 1);
253 struct btrfs_block_group *btrfs_next_block_group(
254 struct btrfs_block_group *cache)
256 struct btrfs_fs_info *fs_info = cache->fs_info;
257 struct rb_node *node;
259 spin_lock(&fs_info->block_group_cache_lock);
261 /* If our block group was removed, we need a full search. */
262 if (RB_EMPTY_NODE(&cache->cache_node)) {
263 const u64 next_bytenr = cache->start + cache->length;
265 spin_unlock(&fs_info->block_group_cache_lock);
266 btrfs_put_block_group(cache);
267 cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache;
269 node = rb_next(&cache->cache_node);
270 btrfs_put_block_group(cache);
272 cache = rb_entry(node, struct btrfs_block_group, cache_node);
273 btrfs_get_block_group(cache);
276 spin_unlock(&fs_info->block_group_cache_lock);
280 bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
282 struct btrfs_block_group *bg;
285 bg = btrfs_lookup_block_group(fs_info, bytenr);
289 spin_lock(&bg->lock);
293 atomic_inc(&bg->nocow_writers);
294 spin_unlock(&bg->lock);
296 /* No put on block group, done by btrfs_dec_nocow_writers */
298 btrfs_put_block_group(bg);
303 void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
305 struct btrfs_block_group *bg;
307 bg = btrfs_lookup_block_group(fs_info, bytenr);
309 if (atomic_dec_and_test(&bg->nocow_writers))
310 wake_up_var(&bg->nocow_writers);
312 * Once for our lookup and once for the lookup done by a previous call
313 * to btrfs_inc_nocow_writers()
315 btrfs_put_block_group(bg);
316 btrfs_put_block_group(bg);
319 void btrfs_wait_nocow_writers(struct btrfs_block_group *bg)
321 wait_var_event(&bg->nocow_writers, !atomic_read(&bg->nocow_writers));
324 void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info,
327 struct btrfs_block_group *bg;
329 bg = btrfs_lookup_block_group(fs_info, start);
331 if (atomic_dec_and_test(&bg->reservations))
332 wake_up_var(&bg->reservations);
333 btrfs_put_block_group(bg);
336 void btrfs_wait_block_group_reservations(struct btrfs_block_group *bg)
338 struct btrfs_space_info *space_info = bg->space_info;
342 if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA))
346 * Our block group is read only but before we set it to read only,
347 * some task might have had allocated an extent from it already, but it
348 * has not yet created a respective ordered extent (and added it to a
349 * root's list of ordered extents).
350 * Therefore wait for any task currently allocating extents, since the
351 * block group's reservations counter is incremented while a read lock
352 * on the groups' semaphore is held and decremented after releasing
353 * the read access on that semaphore and creating the ordered extent.
355 down_write(&space_info->groups_sem);
356 up_write(&space_info->groups_sem);
358 wait_var_event(&bg->reservations, !atomic_read(&bg->reservations));
361 struct btrfs_caching_control *btrfs_get_caching_control(
362 struct btrfs_block_group *cache)
364 struct btrfs_caching_control *ctl;
366 spin_lock(&cache->lock);
367 if (!cache->caching_ctl) {
368 spin_unlock(&cache->lock);
372 ctl = cache->caching_ctl;
373 refcount_inc(&ctl->count);
374 spin_unlock(&cache->lock);
378 void btrfs_put_caching_control(struct btrfs_caching_control *ctl)
380 if (refcount_dec_and_test(&ctl->count))
385 * When we wait for progress in the block group caching, its because our
386 * allocation attempt failed at least once. So, we must sleep and let some
387 * progress happen before we try again.
389 * This function will sleep at least once waiting for new free space to show
390 * up, and then it will check the block group free space numbers for our min
391 * num_bytes. Another option is to have it go ahead and look in the rbtree for
392 * a free extent of a given size, but this is a good start.
394 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
395 * any of the information in this block group.
397 void btrfs_wait_block_group_cache_progress(struct btrfs_block_group *cache,
400 struct btrfs_caching_control *caching_ctl;
402 caching_ctl = btrfs_get_caching_control(cache);
406 wait_event(caching_ctl->wait, btrfs_block_group_done(cache) ||
407 (cache->free_space_ctl->free_space >= num_bytes));
409 btrfs_put_caching_control(caching_ctl);
412 int btrfs_wait_block_group_cache_done(struct btrfs_block_group *cache)
414 struct btrfs_caching_control *caching_ctl;
417 caching_ctl = btrfs_get_caching_control(cache);
419 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
421 wait_event(caching_ctl->wait, btrfs_block_group_done(cache));
422 if (cache->cached == BTRFS_CACHE_ERROR)
424 btrfs_put_caching_control(caching_ctl);
428 static bool space_cache_v1_done(struct btrfs_block_group *cache)
432 spin_lock(&cache->lock);
433 ret = cache->cached != BTRFS_CACHE_FAST;
434 spin_unlock(&cache->lock);
439 void btrfs_wait_space_cache_v1_finished(struct btrfs_block_group *cache,
440 struct btrfs_caching_control *caching_ctl)
442 wait_event(caching_ctl->wait, space_cache_v1_done(cache));
445 #ifdef CONFIG_BTRFS_DEBUG
446 static void fragment_free_space(struct btrfs_block_group *block_group)
448 struct btrfs_fs_info *fs_info = block_group->fs_info;
449 u64 start = block_group->start;
450 u64 len = block_group->length;
451 u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ?
452 fs_info->nodesize : fs_info->sectorsize;
453 u64 step = chunk << 1;
455 while (len > chunk) {
456 btrfs_remove_free_space(block_group, start, chunk);
467 * This is only called by btrfs_cache_block_group, since we could have freed
468 * extents we need to check the pinned_extents for any extents that can't be
469 * used yet since their free space will be released as soon as the transaction
472 u64 add_new_free_space(struct btrfs_block_group *block_group, u64 start, u64 end)
474 struct btrfs_fs_info *info = block_group->fs_info;
475 u64 extent_start, extent_end, size, total_added = 0;
478 while (start < end) {
479 ret = find_first_extent_bit(&info->excluded_extents, start,
480 &extent_start, &extent_end,
481 EXTENT_DIRTY | EXTENT_UPTODATE,
486 if (extent_start <= start) {
487 start = extent_end + 1;
488 } else if (extent_start > start && extent_start < end) {
489 size = extent_start - start;
491 ret = btrfs_add_free_space_async_trimmed(block_group,
493 BUG_ON(ret); /* -ENOMEM or logic error */
494 start = extent_end + 1;
503 ret = btrfs_add_free_space_async_trimmed(block_group, start,
505 BUG_ON(ret); /* -ENOMEM or logic error */
511 static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl)
513 struct btrfs_block_group *block_group = caching_ctl->block_group;
514 struct btrfs_fs_info *fs_info = block_group->fs_info;
515 struct btrfs_root *extent_root = fs_info->extent_root;
516 struct btrfs_path *path;
517 struct extent_buffer *leaf;
518 struct btrfs_key key;
525 path = btrfs_alloc_path();
529 last = max_t(u64, block_group->start, BTRFS_SUPER_INFO_OFFSET);
531 #ifdef CONFIG_BTRFS_DEBUG
533 * If we're fragmenting we don't want to make anybody think we can
534 * allocate from this block group until we've had a chance to fragment
537 if (btrfs_should_fragment_free_space(block_group))
541 * We don't want to deadlock with somebody trying to allocate a new
542 * extent for the extent root while also trying to search the extent
543 * root to add free space. So we skip locking and search the commit
544 * root, since its read-only
546 path->skip_locking = 1;
547 path->search_commit_root = 1;
548 path->reada = READA_FORWARD;
552 key.type = BTRFS_EXTENT_ITEM_KEY;
555 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
559 leaf = path->nodes[0];
560 nritems = btrfs_header_nritems(leaf);
563 if (btrfs_fs_closing(fs_info) > 1) {
568 if (path->slots[0] < nritems) {
569 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
571 ret = btrfs_find_next_key(extent_root, path, &key, 0, 0);
575 if (need_resched() ||
576 rwsem_is_contended(&fs_info->commit_root_sem)) {
578 caching_ctl->progress = last;
579 btrfs_release_path(path);
580 up_read(&fs_info->commit_root_sem);
581 mutex_unlock(&caching_ctl->mutex);
583 mutex_lock(&caching_ctl->mutex);
584 down_read(&fs_info->commit_root_sem);
588 ret = btrfs_next_leaf(extent_root, path);
593 leaf = path->nodes[0];
594 nritems = btrfs_header_nritems(leaf);
598 if (key.objectid < last) {
601 key.type = BTRFS_EXTENT_ITEM_KEY;
604 caching_ctl->progress = last;
605 btrfs_release_path(path);
609 if (key.objectid < block_group->start) {
614 if (key.objectid >= block_group->start + block_group->length)
617 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
618 key.type == BTRFS_METADATA_ITEM_KEY) {
619 total_found += add_new_free_space(block_group, last,
621 if (key.type == BTRFS_METADATA_ITEM_KEY)
622 last = key.objectid +
625 last = key.objectid + key.offset;
627 if (total_found > CACHING_CTL_WAKE_UP) {
630 wake_up(&caching_ctl->wait);
637 total_found += add_new_free_space(block_group, last,
638 block_group->start + block_group->length);
639 caching_ctl->progress = (u64)-1;
642 btrfs_free_path(path);
646 static noinline void caching_thread(struct btrfs_work *work)
648 struct btrfs_block_group *block_group;
649 struct btrfs_fs_info *fs_info;
650 struct btrfs_caching_control *caching_ctl;
653 caching_ctl = container_of(work, struct btrfs_caching_control, work);
654 block_group = caching_ctl->block_group;
655 fs_info = block_group->fs_info;
657 mutex_lock(&caching_ctl->mutex);
658 down_read(&fs_info->commit_root_sem);
660 if (btrfs_test_opt(fs_info, SPACE_CACHE)) {
661 ret = load_free_space_cache(block_group);
668 * We failed to load the space cache, set ourselves to
669 * CACHE_STARTED and carry on.
671 spin_lock(&block_group->lock);
672 block_group->cached = BTRFS_CACHE_STARTED;
673 spin_unlock(&block_group->lock);
674 wake_up(&caching_ctl->wait);
678 * If we are in the transaction that populated the free space tree we
679 * can't actually cache from the free space tree as our commit root and
680 * real root are the same, so we could change the contents of the blocks
681 * while caching. Instead do the slow caching in this case, and after
682 * the transaction has committed we will be safe.
684 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
685 !(test_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags)))
686 ret = load_free_space_tree(caching_ctl);
688 ret = load_extent_tree_free(caching_ctl);
690 spin_lock(&block_group->lock);
691 block_group->caching_ctl = NULL;
692 block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED;
693 spin_unlock(&block_group->lock);
695 #ifdef CONFIG_BTRFS_DEBUG
696 if (btrfs_should_fragment_free_space(block_group)) {
699 spin_lock(&block_group->space_info->lock);
700 spin_lock(&block_group->lock);
701 bytes_used = block_group->length - block_group->used;
702 block_group->space_info->bytes_used += bytes_used >> 1;
703 spin_unlock(&block_group->lock);
704 spin_unlock(&block_group->space_info->lock);
705 fragment_free_space(block_group);
709 caching_ctl->progress = (u64)-1;
711 up_read(&fs_info->commit_root_sem);
712 btrfs_free_excluded_extents(block_group);
713 mutex_unlock(&caching_ctl->mutex);
715 wake_up(&caching_ctl->wait);
717 btrfs_put_caching_control(caching_ctl);
718 btrfs_put_block_group(block_group);
721 int btrfs_cache_block_group(struct btrfs_block_group *cache, int load_cache_only)
724 struct btrfs_fs_info *fs_info = cache->fs_info;
725 struct btrfs_caching_control *caching_ctl = NULL;
728 /* Allocator for zoned filesystems does not use the cache at all */
729 if (btrfs_is_zoned(fs_info))
732 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
736 INIT_LIST_HEAD(&caching_ctl->list);
737 mutex_init(&caching_ctl->mutex);
738 init_waitqueue_head(&caching_ctl->wait);
739 caching_ctl->block_group = cache;
740 caching_ctl->progress = cache->start;
741 refcount_set(&caching_ctl->count, 2);
742 btrfs_init_work(&caching_ctl->work, caching_thread, NULL, NULL);
744 spin_lock(&cache->lock);
745 if (cache->cached != BTRFS_CACHE_NO) {
748 caching_ctl = cache->caching_ctl;
750 refcount_inc(&caching_ctl->count);
751 spin_unlock(&cache->lock);
754 WARN_ON(cache->caching_ctl);
755 cache->caching_ctl = caching_ctl;
756 if (btrfs_test_opt(fs_info, SPACE_CACHE))
757 cache->cached = BTRFS_CACHE_FAST;
759 cache->cached = BTRFS_CACHE_STARTED;
760 cache->has_caching_ctl = 1;
761 spin_unlock(&cache->lock);
763 spin_lock(&fs_info->block_group_cache_lock);
764 refcount_inc(&caching_ctl->count);
765 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
766 spin_unlock(&fs_info->block_group_cache_lock);
768 btrfs_get_block_group(cache);
770 btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
772 if (load_cache_only && caching_ctl)
773 btrfs_wait_space_cache_v1_finished(cache, caching_ctl);
775 btrfs_put_caching_control(caching_ctl);
780 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
782 u64 extra_flags = chunk_to_extended(flags) &
783 BTRFS_EXTENDED_PROFILE_MASK;
785 write_seqlock(&fs_info->profiles_lock);
786 if (flags & BTRFS_BLOCK_GROUP_DATA)
787 fs_info->avail_data_alloc_bits &= ~extra_flags;
788 if (flags & BTRFS_BLOCK_GROUP_METADATA)
789 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
790 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
791 fs_info->avail_system_alloc_bits &= ~extra_flags;
792 write_sequnlock(&fs_info->profiles_lock);
796 * Clear incompat bits for the following feature(s):
798 * - RAID56 - in case there's neither RAID5 nor RAID6 profile block group
799 * in the whole filesystem
801 * - RAID1C34 - same as above for RAID1C3 and RAID1C4 block groups
803 static void clear_incompat_bg_bits(struct btrfs_fs_info *fs_info, u64 flags)
805 bool found_raid56 = false;
806 bool found_raid1c34 = false;
808 if ((flags & BTRFS_BLOCK_GROUP_RAID56_MASK) ||
809 (flags & BTRFS_BLOCK_GROUP_RAID1C3) ||
810 (flags & BTRFS_BLOCK_GROUP_RAID1C4)) {
811 struct list_head *head = &fs_info->space_info;
812 struct btrfs_space_info *sinfo;
814 list_for_each_entry_rcu(sinfo, head, list) {
815 down_read(&sinfo->groups_sem);
816 if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID5]))
818 if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID6]))
820 if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID1C3]))
821 found_raid1c34 = true;
822 if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID1C4]))
823 found_raid1c34 = true;
824 up_read(&sinfo->groups_sem);
827 btrfs_clear_fs_incompat(fs_info, RAID56);
829 btrfs_clear_fs_incompat(fs_info, RAID1C34);
833 static int remove_block_group_item(struct btrfs_trans_handle *trans,
834 struct btrfs_path *path,
835 struct btrfs_block_group *block_group)
837 struct btrfs_fs_info *fs_info = trans->fs_info;
838 struct btrfs_root *root;
839 struct btrfs_key key;
842 root = fs_info->extent_root;
843 key.objectid = block_group->start;
844 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
845 key.offset = block_group->length;
847 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
853 ret = btrfs_del_item(trans, root, path);
857 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
858 u64 group_start, struct extent_map *em)
860 struct btrfs_fs_info *fs_info = trans->fs_info;
861 struct btrfs_path *path;
862 struct btrfs_block_group *block_group;
863 struct btrfs_free_cluster *cluster;
865 struct kobject *kobj = NULL;
869 struct btrfs_caching_control *caching_ctl = NULL;
871 bool remove_rsv = false;
873 block_group = btrfs_lookup_block_group(fs_info, group_start);
874 BUG_ON(!block_group);
875 BUG_ON(!block_group->ro);
877 trace_btrfs_remove_block_group(block_group);
879 * Free the reserved super bytes from this block group before
882 btrfs_free_excluded_extents(block_group);
883 btrfs_free_ref_tree_range(fs_info, block_group->start,
884 block_group->length);
886 index = btrfs_bg_flags_to_raid_index(block_group->flags);
887 factor = btrfs_bg_type_to_factor(block_group->flags);
889 /* make sure this block group isn't part of an allocation cluster */
890 cluster = &fs_info->data_alloc_cluster;
891 spin_lock(&cluster->refill_lock);
892 btrfs_return_cluster_to_free_space(block_group, cluster);
893 spin_unlock(&cluster->refill_lock);
896 * make sure this block group isn't part of a metadata
899 cluster = &fs_info->meta_alloc_cluster;
900 spin_lock(&cluster->refill_lock);
901 btrfs_return_cluster_to_free_space(block_group, cluster);
902 spin_unlock(&cluster->refill_lock);
904 btrfs_clear_treelog_bg(block_group);
906 path = btrfs_alloc_path();
913 * get the inode first so any iput calls done for the io_list
914 * aren't the final iput (no unlinks allowed now)
916 inode = lookup_free_space_inode(block_group, path);
918 mutex_lock(&trans->transaction->cache_write_mutex);
920 * Make sure our free space cache IO is done before removing the
923 spin_lock(&trans->transaction->dirty_bgs_lock);
924 if (!list_empty(&block_group->io_list)) {
925 list_del_init(&block_group->io_list);
927 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
929 spin_unlock(&trans->transaction->dirty_bgs_lock);
930 btrfs_wait_cache_io(trans, block_group, path);
931 btrfs_put_block_group(block_group);
932 spin_lock(&trans->transaction->dirty_bgs_lock);
935 if (!list_empty(&block_group->dirty_list)) {
936 list_del_init(&block_group->dirty_list);
938 btrfs_put_block_group(block_group);
940 spin_unlock(&trans->transaction->dirty_bgs_lock);
941 mutex_unlock(&trans->transaction->cache_write_mutex);
943 ret = btrfs_remove_free_space_inode(trans, inode, block_group);
947 spin_lock(&fs_info->block_group_cache_lock);
948 rb_erase(&block_group->cache_node,
949 &fs_info->block_group_cache_tree);
950 RB_CLEAR_NODE(&block_group->cache_node);
952 /* Once for the block groups rbtree */
953 btrfs_put_block_group(block_group);
955 if (fs_info->first_logical_byte == block_group->start)
956 fs_info->first_logical_byte = (u64)-1;
957 spin_unlock(&fs_info->block_group_cache_lock);
959 down_write(&block_group->space_info->groups_sem);
961 * we must use list_del_init so people can check to see if they
962 * are still on the list after taking the semaphore
964 list_del_init(&block_group->list);
965 if (list_empty(&block_group->space_info->block_groups[index])) {
966 kobj = block_group->space_info->block_group_kobjs[index];
967 block_group->space_info->block_group_kobjs[index] = NULL;
968 clear_avail_alloc_bits(fs_info, block_group->flags);
970 up_write(&block_group->space_info->groups_sem);
971 clear_incompat_bg_bits(fs_info, block_group->flags);
977 if (block_group->has_caching_ctl)
978 caching_ctl = btrfs_get_caching_control(block_group);
979 if (block_group->cached == BTRFS_CACHE_STARTED)
980 btrfs_wait_block_group_cache_done(block_group);
981 if (block_group->has_caching_ctl) {
982 spin_lock(&fs_info->block_group_cache_lock);
984 struct btrfs_caching_control *ctl;
986 list_for_each_entry(ctl,
987 &fs_info->caching_block_groups, list)
988 if (ctl->block_group == block_group) {
990 refcount_inc(&caching_ctl->count);
995 list_del_init(&caching_ctl->list);
996 spin_unlock(&fs_info->block_group_cache_lock);
998 /* Once for the caching bgs list and once for us. */
999 btrfs_put_caching_control(caching_ctl);
1000 btrfs_put_caching_control(caching_ctl);
1004 spin_lock(&trans->transaction->dirty_bgs_lock);
1005 WARN_ON(!list_empty(&block_group->dirty_list));
1006 WARN_ON(!list_empty(&block_group->io_list));
1007 spin_unlock(&trans->transaction->dirty_bgs_lock);
1009 btrfs_remove_free_space_cache(block_group);
1011 spin_lock(&block_group->space_info->lock);
1012 list_del_init(&block_group->ro_list);
1014 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
1015 WARN_ON(block_group->space_info->total_bytes
1016 < block_group->length);
1017 WARN_ON(block_group->space_info->bytes_readonly
1018 < block_group->length - block_group->zone_unusable);
1019 WARN_ON(block_group->space_info->bytes_zone_unusable
1020 < block_group->zone_unusable);
1021 WARN_ON(block_group->space_info->disk_total
1022 < block_group->length * factor);
1024 block_group->space_info->total_bytes -= block_group->length;
1025 block_group->space_info->bytes_readonly -=
1026 (block_group->length - block_group->zone_unusable);
1027 block_group->space_info->bytes_zone_unusable -=
1028 block_group->zone_unusable;
1029 block_group->space_info->disk_total -= block_group->length * factor;
1031 spin_unlock(&block_group->space_info->lock);
1034 * Remove the free space for the block group from the free space tree
1035 * and the block group's item from the extent tree before marking the
1036 * block group as removed. This is to prevent races with tasks that
1037 * freeze and unfreeze a block group, this task and another task
1038 * allocating a new block group - the unfreeze task ends up removing
1039 * the block group's extent map before the task calling this function
1040 * deletes the block group item from the extent tree, allowing for
1041 * another task to attempt to create another block group with the same
1042 * item key (and failing with -EEXIST and a transaction abort).
1044 ret = remove_block_group_free_space(trans, block_group);
1048 ret = remove_block_group_item(trans, path, block_group);
1052 spin_lock(&block_group->lock);
1053 block_group->removed = 1;
1055 * At this point trimming or scrub can't start on this block group,
1056 * because we removed the block group from the rbtree
1057 * fs_info->block_group_cache_tree so no one can't find it anymore and
1058 * even if someone already got this block group before we removed it
1059 * from the rbtree, they have already incremented block_group->frozen -
1060 * if they didn't, for the trimming case they won't find any free space
1061 * entries because we already removed them all when we called
1062 * btrfs_remove_free_space_cache().
1064 * And we must not remove the extent map from the fs_info->mapping_tree
1065 * to prevent the same logical address range and physical device space
1066 * ranges from being reused for a new block group. This is needed to
1067 * avoid races with trimming and scrub.
1069 * An fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
1070 * completely transactionless, so while it is trimming a range the
1071 * currently running transaction might finish and a new one start,
1072 * allowing for new block groups to be created that can reuse the same
1073 * physical device locations unless we take this special care.
1075 * There may also be an implicit trim operation if the file system
1076 * is mounted with -odiscard. The same protections must remain
1077 * in place until the extents have been discarded completely when
1078 * the transaction commit has completed.
1080 remove_em = (atomic_read(&block_group->frozen) == 0);
1081 spin_unlock(&block_group->lock);
1084 struct extent_map_tree *em_tree;
1086 em_tree = &fs_info->mapping_tree;
1087 write_lock(&em_tree->lock);
1088 remove_extent_mapping(em_tree, em);
1089 write_unlock(&em_tree->lock);
1090 /* once for the tree */
1091 free_extent_map(em);
1095 /* Once for the lookup reference */
1096 btrfs_put_block_group(block_group);
1098 btrfs_delayed_refs_rsv_release(fs_info, 1);
1099 btrfs_free_path(path);
1103 struct btrfs_trans_handle *btrfs_start_trans_remove_block_group(
1104 struct btrfs_fs_info *fs_info, const u64 chunk_offset)
1106 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
1107 struct extent_map *em;
1108 struct map_lookup *map;
1109 unsigned int num_items;
1111 read_lock(&em_tree->lock);
1112 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1113 read_unlock(&em_tree->lock);
1114 ASSERT(em && em->start == chunk_offset);
1117 * We need to reserve 3 + N units from the metadata space info in order
1118 * to remove a block group (done at btrfs_remove_chunk() and at
1119 * btrfs_remove_block_group()), which are used for:
1121 * 1 unit for adding the free space inode's orphan (located in the tree
1123 * 1 unit for deleting the block group item (located in the extent
1125 * 1 unit for deleting the free space item (located in tree of tree
1127 * N units for deleting N device extent items corresponding to each
1128 * stripe (located in the device tree).
1130 * In order to remove a block group we also need to reserve units in the
1131 * system space info in order to update the chunk tree (update one or
1132 * more device items and remove one chunk item), but this is done at
1133 * btrfs_remove_chunk() through a call to check_system_chunk().
1135 map = em->map_lookup;
1136 num_items = 3 + map->num_stripes;
1137 free_extent_map(em);
1139 return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root,
1144 * Mark block group @cache read-only, so later write won't happen to block
1147 * If @force is not set, this function will only mark the block group readonly
1148 * if we have enough free space (1M) in other metadata/system block groups.
1149 * If @force is not set, this function will mark the block group readonly
1150 * without checking free space.
1152 * NOTE: This function doesn't care if other block groups can contain all the
1153 * data in this block group. That check should be done by relocation routine,
1154 * not this function.
1156 static int inc_block_group_ro(struct btrfs_block_group *cache, int force)
1158 struct btrfs_space_info *sinfo = cache->space_info;
1162 spin_lock(&sinfo->lock);
1163 spin_lock(&cache->lock);
1165 if (cache->swap_extents) {
1176 num_bytes = cache->length - cache->reserved - cache->pinned -
1177 cache->bytes_super - cache->zone_unusable - cache->used;
1180 * Data never overcommits, even in mixed mode, so do just the straight
1181 * check of left over space in how much we have allocated.
1185 } else if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA) {
1186 u64 sinfo_used = btrfs_space_info_used(sinfo, true);
1189 * Here we make sure if we mark this bg RO, we still have enough
1190 * free space as buffer.
1192 if (sinfo_used + num_bytes <= sinfo->total_bytes)
1196 * We overcommit metadata, so we need to do the
1197 * btrfs_can_overcommit check here, and we need to pass in
1198 * BTRFS_RESERVE_NO_FLUSH to give ourselves the most amount of
1199 * leeway to allow us to mark this block group as read only.
1201 if (btrfs_can_overcommit(cache->fs_info, sinfo, num_bytes,
1202 BTRFS_RESERVE_NO_FLUSH))
1207 sinfo->bytes_readonly += num_bytes;
1208 if (btrfs_is_zoned(cache->fs_info)) {
1209 /* Migrate zone_unusable bytes to readonly */
1210 sinfo->bytes_readonly += cache->zone_unusable;
1211 sinfo->bytes_zone_unusable -= cache->zone_unusable;
1212 cache->zone_unusable = 0;
1215 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
1218 spin_unlock(&cache->lock);
1219 spin_unlock(&sinfo->lock);
1220 if (ret == -ENOSPC && btrfs_test_opt(cache->fs_info, ENOSPC_DEBUG)) {
1221 btrfs_info(cache->fs_info,
1222 "unable to make block group %llu ro", cache->start);
1223 btrfs_dump_space_info(cache->fs_info, cache->space_info, 0, 0);
1228 static bool clean_pinned_extents(struct btrfs_trans_handle *trans,
1229 struct btrfs_block_group *bg)
1231 struct btrfs_fs_info *fs_info = bg->fs_info;
1232 struct btrfs_transaction *prev_trans = NULL;
1233 const u64 start = bg->start;
1234 const u64 end = start + bg->length - 1;
1237 spin_lock(&fs_info->trans_lock);
1238 if (trans->transaction->list.prev != &fs_info->trans_list) {
1239 prev_trans = list_last_entry(&trans->transaction->list,
1240 struct btrfs_transaction, list);
1241 refcount_inc(&prev_trans->use_count);
1243 spin_unlock(&fs_info->trans_lock);
1246 * Hold the unused_bg_unpin_mutex lock to avoid racing with
1247 * btrfs_finish_extent_commit(). If we are at transaction N, another
1248 * task might be running finish_extent_commit() for the previous
1249 * transaction N - 1, and have seen a range belonging to the block
1250 * group in pinned_extents before we were able to clear the whole block
1251 * group range from pinned_extents. This means that task can lookup for
1252 * the block group after we unpinned it from pinned_extents and removed
1253 * it, leading to a BUG_ON() at unpin_extent_range().
1255 mutex_lock(&fs_info->unused_bg_unpin_mutex);
1257 ret = clear_extent_bits(&prev_trans->pinned_extents, start, end,
1263 ret = clear_extent_bits(&trans->transaction->pinned_extents, start, end,
1266 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
1268 btrfs_put_transaction(prev_trans);
1274 * Process the unused_bgs list and remove any that don't have any allocated
1275 * space inside of them.
1277 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
1279 struct btrfs_block_group *block_group;
1280 struct btrfs_space_info *space_info;
1281 struct btrfs_trans_handle *trans;
1282 const bool async_trim_enabled = btrfs_test_opt(fs_info, DISCARD_ASYNC);
1285 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1289 * Long running balances can keep us blocked here for eternity, so
1290 * simply skip deletion if we're unable to get the mutex.
1292 if (!mutex_trylock(&fs_info->reclaim_bgs_lock))
1295 spin_lock(&fs_info->unused_bgs_lock);
1296 while (!list_empty(&fs_info->unused_bgs)) {
1299 block_group = list_first_entry(&fs_info->unused_bgs,
1300 struct btrfs_block_group,
1302 list_del_init(&block_group->bg_list);
1304 space_info = block_group->space_info;
1306 if (ret || btrfs_mixed_space_info(space_info)) {
1307 btrfs_put_block_group(block_group);
1310 spin_unlock(&fs_info->unused_bgs_lock);
1312 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group);
1314 /* Don't want to race with allocators so take the groups_sem */
1315 down_write(&space_info->groups_sem);
1318 * Async discard moves the final block group discard to be prior
1319 * to the unused_bgs code path. Therefore, if it's not fully
1320 * trimmed, punt it back to the async discard lists.
1322 if (btrfs_test_opt(fs_info, DISCARD_ASYNC) &&
1323 !btrfs_is_free_space_trimmed(block_group)) {
1324 trace_btrfs_skip_unused_block_group(block_group);
1325 up_write(&space_info->groups_sem);
1326 /* Requeue if we failed because of async discard */
1327 btrfs_discard_queue_work(&fs_info->discard_ctl,
1332 spin_lock(&block_group->lock);
1333 if (block_group->reserved || block_group->pinned ||
1334 block_group->used || block_group->ro ||
1335 list_is_singular(&block_group->list)) {
1337 * We want to bail if we made new allocations or have
1338 * outstanding allocations in this block group. We do
1339 * the ro check in case balance is currently acting on
1342 trace_btrfs_skip_unused_block_group(block_group);
1343 spin_unlock(&block_group->lock);
1344 up_write(&space_info->groups_sem);
1347 spin_unlock(&block_group->lock);
1349 /* We don't want to force the issue, only flip if it's ok. */
1350 ret = inc_block_group_ro(block_group, 0);
1351 up_write(&space_info->groups_sem);
1358 * Want to do this before we do anything else so we can recover
1359 * properly if we fail to join the transaction.
1361 trans = btrfs_start_trans_remove_block_group(fs_info,
1362 block_group->start);
1363 if (IS_ERR(trans)) {
1364 btrfs_dec_block_group_ro(block_group);
1365 ret = PTR_ERR(trans);
1370 * We could have pending pinned extents for this block group,
1371 * just delete them, we don't care about them anymore.
1373 if (!clean_pinned_extents(trans, block_group)) {
1374 btrfs_dec_block_group_ro(block_group);
1379 * At this point, the block_group is read only and should fail
1380 * new allocations. However, btrfs_finish_extent_commit() can
1381 * cause this block_group to be placed back on the discard
1382 * lists because now the block_group isn't fully discarded.
1383 * Bail here and try again later after discarding everything.
1385 spin_lock(&fs_info->discard_ctl.lock);
1386 if (!list_empty(&block_group->discard_list)) {
1387 spin_unlock(&fs_info->discard_ctl.lock);
1388 btrfs_dec_block_group_ro(block_group);
1389 btrfs_discard_queue_work(&fs_info->discard_ctl,
1393 spin_unlock(&fs_info->discard_ctl.lock);
1395 /* Reset pinned so btrfs_put_block_group doesn't complain */
1396 spin_lock(&space_info->lock);
1397 spin_lock(&block_group->lock);
1399 btrfs_space_info_update_bytes_pinned(fs_info, space_info,
1400 -block_group->pinned);
1401 space_info->bytes_readonly += block_group->pinned;
1402 block_group->pinned = 0;
1404 spin_unlock(&block_group->lock);
1405 spin_unlock(&space_info->lock);
1408 * The normal path here is an unused block group is passed here,
1409 * then trimming is handled in the transaction commit path.
1410 * Async discard interposes before this to do the trimming
1411 * before coming down the unused block group path as trimming
1412 * will no longer be done later in the transaction commit path.
1414 if (!async_trim_enabled && btrfs_test_opt(fs_info, DISCARD_ASYNC))
1418 * DISCARD can flip during remount. On zoned filesystems, we
1419 * need to reset sequential-required zones.
1421 trimming = btrfs_test_opt(fs_info, DISCARD_SYNC) ||
1422 btrfs_is_zoned(fs_info);
1424 /* Implicit trim during transaction commit. */
1426 btrfs_freeze_block_group(block_group);
1429 * Btrfs_remove_chunk will abort the transaction if things go
1432 ret = btrfs_remove_chunk(trans, block_group->start);
1436 btrfs_unfreeze_block_group(block_group);
1441 * If we're not mounted with -odiscard, we can just forget
1442 * about this block group. Otherwise we'll need to wait
1443 * until transaction commit to do the actual discard.
1446 spin_lock(&fs_info->unused_bgs_lock);
1448 * A concurrent scrub might have added us to the list
1449 * fs_info->unused_bgs, so use a list_move operation
1450 * to add the block group to the deleted_bgs list.
1452 list_move(&block_group->bg_list,
1453 &trans->transaction->deleted_bgs);
1454 spin_unlock(&fs_info->unused_bgs_lock);
1455 btrfs_get_block_group(block_group);
1458 btrfs_end_transaction(trans);
1460 btrfs_put_block_group(block_group);
1461 spin_lock(&fs_info->unused_bgs_lock);
1463 spin_unlock(&fs_info->unused_bgs_lock);
1464 mutex_unlock(&fs_info->reclaim_bgs_lock);
1468 btrfs_end_transaction(trans);
1469 mutex_unlock(&fs_info->reclaim_bgs_lock);
1470 btrfs_put_block_group(block_group);
1471 btrfs_discard_punt_unused_bgs_list(fs_info);
1474 void btrfs_mark_bg_unused(struct btrfs_block_group *bg)
1476 struct btrfs_fs_info *fs_info = bg->fs_info;
1478 spin_lock(&fs_info->unused_bgs_lock);
1479 if (list_empty(&bg->bg_list)) {
1480 btrfs_get_block_group(bg);
1481 trace_btrfs_add_unused_block_group(bg);
1482 list_add_tail(&bg->bg_list, &fs_info->unused_bgs);
1484 spin_unlock(&fs_info->unused_bgs_lock);
1487 void btrfs_reclaim_bgs_work(struct work_struct *work)
1489 struct btrfs_fs_info *fs_info =
1490 container_of(work, struct btrfs_fs_info, reclaim_bgs_work);
1491 struct btrfs_block_group *bg;
1492 struct btrfs_space_info *space_info;
1493 LIST_HEAD(again_list);
1495 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1498 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE))
1501 mutex_lock(&fs_info->reclaim_bgs_lock);
1502 spin_lock(&fs_info->unused_bgs_lock);
1503 while (!list_empty(&fs_info->reclaim_bgs)) {
1507 bg = list_first_entry(&fs_info->reclaim_bgs,
1508 struct btrfs_block_group,
1510 list_del_init(&bg->bg_list);
1512 space_info = bg->space_info;
1513 spin_unlock(&fs_info->unused_bgs_lock);
1515 /* Don't race with allocators so take the groups_sem */
1516 down_write(&space_info->groups_sem);
1518 spin_lock(&bg->lock);
1519 if (bg->reserved || bg->pinned || bg->ro) {
1521 * We want to bail if we made new allocations or have
1522 * outstanding allocations in this block group. We do
1523 * the ro check in case balance is currently acting on
1526 spin_unlock(&bg->lock);
1527 up_write(&space_info->groups_sem);
1530 spin_unlock(&bg->lock);
1532 /* Get out fast, in case we're unmounting the filesystem */
1533 if (btrfs_fs_closing(fs_info)) {
1534 up_write(&space_info->groups_sem);
1539 * Cache the zone_unusable value before turning the block group
1540 * to read only. As soon as the blog group is read only it's
1541 * zone_unusable value gets moved to the block group's read-only
1542 * bytes and isn't available for calculations anymore.
1544 zone_unusable = bg->zone_unusable;
1545 ret = inc_block_group_ro(bg, 0);
1546 up_write(&space_info->groups_sem);
1551 "reclaiming chunk %llu with %llu%% used %llu%% unusable",
1552 bg->start, div_u64(bg->used * 100, bg->length),
1553 div64_u64(zone_unusable * 100, bg->length));
1554 trace_btrfs_reclaim_block_group(bg);
1555 ret = btrfs_relocate_chunk(fs_info, bg->start);
1557 btrfs_err(fs_info, "error relocating chunk %llu",
1561 spin_lock(&fs_info->unused_bgs_lock);
1562 if (ret == -EAGAIN && list_empty(&bg->bg_list))
1563 list_add_tail(&bg->bg_list, &again_list);
1565 btrfs_put_block_group(bg);
1567 list_splice_tail(&again_list, &fs_info->reclaim_bgs);
1568 spin_unlock(&fs_info->unused_bgs_lock);
1569 mutex_unlock(&fs_info->reclaim_bgs_lock);
1570 btrfs_exclop_finish(fs_info);
1573 void btrfs_reclaim_bgs(struct btrfs_fs_info *fs_info)
1575 spin_lock(&fs_info->unused_bgs_lock);
1576 if (!list_empty(&fs_info->reclaim_bgs))
1577 queue_work(system_unbound_wq, &fs_info->reclaim_bgs_work);
1578 spin_unlock(&fs_info->unused_bgs_lock);
1581 void btrfs_mark_bg_to_reclaim(struct btrfs_block_group *bg)
1583 struct btrfs_fs_info *fs_info = bg->fs_info;
1585 spin_lock(&fs_info->unused_bgs_lock);
1586 if (list_empty(&bg->bg_list)) {
1587 btrfs_get_block_group(bg);
1588 trace_btrfs_add_reclaim_block_group(bg);
1589 list_add_tail(&bg->bg_list, &fs_info->reclaim_bgs);
1591 spin_unlock(&fs_info->unused_bgs_lock);
1594 static int read_bg_from_eb(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
1595 struct btrfs_path *path)
1597 struct extent_map_tree *em_tree;
1598 struct extent_map *em;
1599 struct btrfs_block_group_item bg;
1600 struct extent_buffer *leaf;
1605 slot = path->slots[0];
1606 leaf = path->nodes[0];
1608 em_tree = &fs_info->mapping_tree;
1609 read_lock(&em_tree->lock);
1610 em = lookup_extent_mapping(em_tree, key->objectid, key->offset);
1611 read_unlock(&em_tree->lock);
1614 "logical %llu len %llu found bg but no related chunk",
1615 key->objectid, key->offset);
1619 if (em->start != key->objectid || em->len != key->offset) {
1621 "block group %llu len %llu mismatch with chunk %llu len %llu",
1622 key->objectid, key->offset, em->start, em->len);
1627 read_extent_buffer(leaf, &bg, btrfs_item_ptr_offset(leaf, slot),
1629 flags = btrfs_stack_block_group_flags(&bg) &
1630 BTRFS_BLOCK_GROUP_TYPE_MASK;
1632 if (flags != (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
1634 "block group %llu len %llu type flags 0x%llx mismatch with chunk type flags 0x%llx",
1635 key->objectid, key->offset, flags,
1636 (BTRFS_BLOCK_GROUP_TYPE_MASK & em->map_lookup->type));
1641 free_extent_map(em);
1645 static int find_first_block_group(struct btrfs_fs_info *fs_info,
1646 struct btrfs_path *path,
1647 struct btrfs_key *key)
1649 struct btrfs_root *root = fs_info->extent_root;
1651 struct btrfs_key found_key;
1652 struct extent_buffer *leaf;
1655 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
1660 slot = path->slots[0];
1661 leaf = path->nodes[0];
1662 if (slot >= btrfs_header_nritems(leaf)) {
1663 ret = btrfs_next_leaf(root, path);
1670 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1672 if (found_key.objectid >= key->objectid &&
1673 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
1674 ret = read_bg_from_eb(fs_info, &found_key, path);
1684 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1686 u64 extra_flags = chunk_to_extended(flags) &
1687 BTRFS_EXTENDED_PROFILE_MASK;
1689 write_seqlock(&fs_info->profiles_lock);
1690 if (flags & BTRFS_BLOCK_GROUP_DATA)
1691 fs_info->avail_data_alloc_bits |= extra_flags;
1692 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1693 fs_info->avail_metadata_alloc_bits |= extra_flags;
1694 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1695 fs_info->avail_system_alloc_bits |= extra_flags;
1696 write_sequnlock(&fs_info->profiles_lock);
1700 * Map a physical disk address to a list of logical addresses
1702 * @fs_info: the filesystem
1703 * @chunk_start: logical address of block group
1704 * @bdev: physical device to resolve, can be NULL to indicate any device
1705 * @physical: physical address to map to logical addresses
1706 * @logical: return array of logical addresses which map to @physical
1707 * @naddrs: length of @logical
1708 * @stripe_len: size of IO stripe for the given block group
1710 * Maps a particular @physical disk address to a list of @logical addresses.
1711 * Used primarily to exclude those portions of a block group that contain super
1714 int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start,
1715 struct block_device *bdev, u64 physical, u64 **logical,
1716 int *naddrs, int *stripe_len)
1718 struct extent_map *em;
1719 struct map_lookup *map;
1722 u64 data_stripe_length;
1727 em = btrfs_get_chunk_map(fs_info, chunk_start, 1);
1731 map = em->map_lookup;
1732 data_stripe_length = em->orig_block_len;
1733 io_stripe_size = map->stripe_len;
1734 chunk_start = em->start;
1736 /* For RAID5/6 adjust to a full IO stripe length */
1737 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
1738 io_stripe_size = map->stripe_len * nr_data_stripes(map);
1740 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
1746 for (i = 0; i < map->num_stripes; i++) {
1747 bool already_inserted = false;
1752 if (!in_range(physical, map->stripes[i].physical,
1753 data_stripe_length))
1756 if (bdev && map->stripes[i].dev->bdev != bdev)
1759 stripe_nr = physical - map->stripes[i].physical;
1760 stripe_nr = div64_u64_rem(stripe_nr, map->stripe_len, &offset);
1762 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1763 stripe_nr = stripe_nr * map->num_stripes + i;
1764 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
1765 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
1766 stripe_nr = stripe_nr * map->num_stripes + i;
1769 * The remaining case would be for RAID56, multiply by
1770 * nr_data_stripes(). Alternatively, just use rmap_len below
1771 * instead of map->stripe_len
1774 bytenr = chunk_start + stripe_nr * io_stripe_size + offset;
1776 /* Ensure we don't add duplicate addresses */
1777 for (j = 0; j < nr; j++) {
1778 if (buf[j] == bytenr) {
1779 already_inserted = true;
1784 if (!already_inserted)
1790 *stripe_len = io_stripe_size;
1792 free_extent_map(em);
1796 static int exclude_super_stripes(struct btrfs_block_group *cache)
1798 struct btrfs_fs_info *fs_info = cache->fs_info;
1799 const bool zoned = btrfs_is_zoned(fs_info);
1805 if (cache->start < BTRFS_SUPER_INFO_OFFSET) {
1806 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->start;
1807 cache->bytes_super += stripe_len;
1808 ret = btrfs_add_excluded_extent(fs_info, cache->start,
1814 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1815 bytenr = btrfs_sb_offset(i);
1816 ret = btrfs_rmap_block(fs_info, cache->start, NULL,
1817 bytenr, &logical, &nr, &stripe_len);
1821 /* Shouldn't have super stripes in sequential zones */
1824 "zoned: block group %llu must not contain super block",
1830 u64 len = min_t(u64, stripe_len,
1831 cache->start + cache->length - logical[nr]);
1833 cache->bytes_super += len;
1834 ret = btrfs_add_excluded_extent(fs_info, logical[nr],
1847 static void link_block_group(struct btrfs_block_group *cache)
1849 struct btrfs_space_info *space_info = cache->space_info;
1850 int index = btrfs_bg_flags_to_raid_index(cache->flags);
1852 down_write(&space_info->groups_sem);
1853 list_add_tail(&cache->list, &space_info->block_groups[index]);
1854 up_write(&space_info->groups_sem);
1857 static struct btrfs_block_group *btrfs_create_block_group_cache(
1858 struct btrfs_fs_info *fs_info, u64 start)
1860 struct btrfs_block_group *cache;
1862 cache = kzalloc(sizeof(*cache), GFP_NOFS);
1866 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
1868 if (!cache->free_space_ctl) {
1873 cache->start = start;
1875 cache->fs_info = fs_info;
1876 cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start);
1878 cache->discard_index = BTRFS_DISCARD_INDEX_UNUSED;
1880 refcount_set(&cache->refs, 1);
1881 spin_lock_init(&cache->lock);
1882 init_rwsem(&cache->data_rwsem);
1883 INIT_LIST_HEAD(&cache->list);
1884 INIT_LIST_HEAD(&cache->cluster_list);
1885 INIT_LIST_HEAD(&cache->bg_list);
1886 INIT_LIST_HEAD(&cache->ro_list);
1887 INIT_LIST_HEAD(&cache->discard_list);
1888 INIT_LIST_HEAD(&cache->dirty_list);
1889 INIT_LIST_HEAD(&cache->io_list);
1890 btrfs_init_free_space_ctl(cache, cache->free_space_ctl);
1891 atomic_set(&cache->frozen, 0);
1892 mutex_init(&cache->free_space_lock);
1893 btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root);
1899 * Iterate all chunks and verify that each of them has the corresponding block
1902 static int check_chunk_block_group_mappings(struct btrfs_fs_info *fs_info)
1904 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
1905 struct extent_map *em;
1906 struct btrfs_block_group *bg;
1911 read_lock(&map_tree->lock);
1913 * lookup_extent_mapping will return the first extent map
1914 * intersecting the range, so setting @len to 1 is enough to
1915 * get the first chunk.
1917 em = lookup_extent_mapping(map_tree, start, 1);
1918 read_unlock(&map_tree->lock);
1922 bg = btrfs_lookup_block_group(fs_info, em->start);
1925 "chunk start=%llu len=%llu doesn't have corresponding block group",
1926 em->start, em->len);
1928 free_extent_map(em);
1931 if (bg->start != em->start || bg->length != em->len ||
1932 (bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK) !=
1933 (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
1935 "chunk start=%llu len=%llu flags=0x%llx doesn't match block group start=%llu len=%llu flags=0x%llx",
1937 em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK,
1938 bg->start, bg->length,
1939 bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK);
1941 free_extent_map(em);
1942 btrfs_put_block_group(bg);
1945 start = em->start + em->len;
1946 free_extent_map(em);
1947 btrfs_put_block_group(bg);
1952 static int read_one_block_group(struct btrfs_fs_info *info,
1953 struct btrfs_block_group_item *bgi,
1954 const struct btrfs_key *key,
1957 struct btrfs_block_group *cache;
1958 struct btrfs_space_info *space_info;
1959 const bool mixed = btrfs_fs_incompat(info, MIXED_GROUPS);
1962 ASSERT(key->type == BTRFS_BLOCK_GROUP_ITEM_KEY);
1964 cache = btrfs_create_block_group_cache(info, key->objectid);
1968 cache->length = key->offset;
1969 cache->used = btrfs_stack_block_group_used(bgi);
1970 cache->flags = btrfs_stack_block_group_flags(bgi);
1972 set_free_space_tree_thresholds(cache);
1976 * When we mount with old space cache, we need to
1977 * set BTRFS_DC_CLEAR and set dirty flag.
1979 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
1980 * truncate the old free space cache inode and
1982 * b) Setting 'dirty flag' makes sure that we flush
1983 * the new space cache info onto disk.
1985 if (btrfs_test_opt(info, SPACE_CACHE))
1986 cache->disk_cache_state = BTRFS_DC_CLEAR;
1988 if (!mixed && ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) &&
1989 (cache->flags & BTRFS_BLOCK_GROUP_DATA))) {
1991 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
1997 ret = btrfs_load_block_group_zone_info(cache, false);
1999 btrfs_err(info, "zoned: failed to load zone info of bg %llu",
2005 * We need to exclude the super stripes now so that the space info has
2006 * super bytes accounted for, otherwise we'll think we have more space
2007 * than we actually do.
2009 ret = exclude_super_stripes(cache);
2011 /* We may have excluded something, so call this just in case. */
2012 btrfs_free_excluded_extents(cache);
2017 * For zoned filesystem, space after the allocation offset is the only
2018 * free space for a block group. So, we don't need any caching work.
2019 * btrfs_calc_zone_unusable() will set the amount of free space and
2020 * zone_unusable space.
2022 * For regular filesystem, check for two cases, either we are full, and
2023 * therefore don't need to bother with the caching work since we won't
2024 * find any space, or we are empty, and we can just add all the space
2025 * in and be done with it. This saves us _a_lot_ of time, particularly
2028 if (btrfs_is_zoned(info)) {
2029 btrfs_calc_zone_unusable(cache);
2030 } else if (cache->length == cache->used) {
2031 cache->last_byte_to_unpin = (u64)-1;
2032 cache->cached = BTRFS_CACHE_FINISHED;
2033 btrfs_free_excluded_extents(cache);
2034 } else if (cache->used == 0) {
2035 cache->last_byte_to_unpin = (u64)-1;
2036 cache->cached = BTRFS_CACHE_FINISHED;
2037 add_new_free_space(cache, cache->start,
2038 cache->start + cache->length);
2039 btrfs_free_excluded_extents(cache);
2042 ret = btrfs_add_block_group_cache(info, cache);
2044 btrfs_remove_free_space_cache(cache);
2047 trace_btrfs_add_block_group(info, cache, 0);
2048 btrfs_update_space_info(info, cache->flags, cache->length,
2049 cache->used, cache->bytes_super,
2050 cache->zone_unusable, &space_info);
2052 cache->space_info = space_info;
2054 link_block_group(cache);
2056 set_avail_alloc_bits(info, cache->flags);
2057 if (btrfs_chunk_readonly(info, cache->start)) {
2058 inc_block_group_ro(cache, 1);
2059 } else if (cache->used == 0) {
2060 ASSERT(list_empty(&cache->bg_list));
2061 if (btrfs_test_opt(info, DISCARD_ASYNC))
2062 btrfs_discard_queue_work(&info->discard_ctl, cache);
2064 btrfs_mark_bg_unused(cache);
2068 btrfs_put_block_group(cache);
2072 static int fill_dummy_bgs(struct btrfs_fs_info *fs_info)
2074 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
2075 struct btrfs_space_info *space_info;
2076 struct rb_node *node;
2079 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) {
2080 struct extent_map *em;
2081 struct map_lookup *map;
2082 struct btrfs_block_group *bg;
2084 em = rb_entry(node, struct extent_map, rb_node);
2085 map = em->map_lookup;
2086 bg = btrfs_create_block_group_cache(fs_info, em->start);
2092 /* Fill dummy cache as FULL */
2093 bg->length = em->len;
2094 bg->flags = map->type;
2095 bg->last_byte_to_unpin = (u64)-1;
2096 bg->cached = BTRFS_CACHE_FINISHED;
2098 bg->flags = map->type;
2099 ret = btrfs_add_block_group_cache(fs_info, bg);
2101 btrfs_remove_free_space_cache(bg);
2102 btrfs_put_block_group(bg);
2105 btrfs_update_space_info(fs_info, bg->flags, em->len, em->len,
2107 bg->space_info = space_info;
2108 link_block_group(bg);
2110 set_avail_alloc_bits(fs_info, bg->flags);
2113 btrfs_init_global_block_rsv(fs_info);
2117 int btrfs_read_block_groups(struct btrfs_fs_info *info)
2119 struct btrfs_path *path;
2121 struct btrfs_block_group *cache;
2122 struct btrfs_space_info *space_info;
2123 struct btrfs_key key;
2127 if (!info->extent_root)
2128 return fill_dummy_bgs(info);
2132 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
2133 path = btrfs_alloc_path();
2137 cache_gen = btrfs_super_cache_generation(info->super_copy);
2138 if (btrfs_test_opt(info, SPACE_CACHE) &&
2139 btrfs_super_generation(info->super_copy) != cache_gen)
2141 if (btrfs_test_opt(info, CLEAR_CACHE))
2145 struct btrfs_block_group_item bgi;
2146 struct extent_buffer *leaf;
2149 ret = find_first_block_group(info, path, &key);
2155 leaf = path->nodes[0];
2156 slot = path->slots[0];
2158 read_extent_buffer(leaf, &bgi, btrfs_item_ptr_offset(leaf, slot),
2161 btrfs_item_key_to_cpu(leaf, &key, slot);
2162 btrfs_release_path(path);
2163 ret = read_one_block_group(info, &bgi, &key, need_clear);
2166 key.objectid += key.offset;
2169 btrfs_release_path(path);
2171 list_for_each_entry(space_info, &info->space_info, list) {
2174 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
2175 if (list_empty(&space_info->block_groups[i]))
2177 cache = list_first_entry(&space_info->block_groups[i],
2178 struct btrfs_block_group,
2180 btrfs_sysfs_add_block_group_type(cache);
2183 if (!(btrfs_get_alloc_profile(info, space_info->flags) &
2184 (BTRFS_BLOCK_GROUP_RAID10 |
2185 BTRFS_BLOCK_GROUP_RAID1_MASK |
2186 BTRFS_BLOCK_GROUP_RAID56_MASK |
2187 BTRFS_BLOCK_GROUP_DUP)))
2190 * Avoid allocating from un-mirrored block group if there are
2191 * mirrored block groups.
2193 list_for_each_entry(cache,
2194 &space_info->block_groups[BTRFS_RAID_RAID0],
2196 inc_block_group_ro(cache, 1);
2197 list_for_each_entry(cache,
2198 &space_info->block_groups[BTRFS_RAID_SINGLE],
2200 inc_block_group_ro(cache, 1);
2203 btrfs_init_global_block_rsv(info);
2204 ret = check_chunk_block_group_mappings(info);
2206 btrfs_free_path(path);
2210 static int insert_block_group_item(struct btrfs_trans_handle *trans,
2211 struct btrfs_block_group *block_group)
2213 struct btrfs_fs_info *fs_info = trans->fs_info;
2214 struct btrfs_block_group_item bgi;
2215 struct btrfs_root *root;
2216 struct btrfs_key key;
2218 spin_lock(&block_group->lock);
2219 btrfs_set_stack_block_group_used(&bgi, block_group->used);
2220 btrfs_set_stack_block_group_chunk_objectid(&bgi,
2221 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
2222 btrfs_set_stack_block_group_flags(&bgi, block_group->flags);
2223 key.objectid = block_group->start;
2224 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
2225 key.offset = block_group->length;
2226 spin_unlock(&block_group->lock);
2228 root = fs_info->extent_root;
2229 return btrfs_insert_item(trans, root, &key, &bgi, sizeof(bgi));
2232 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans)
2234 struct btrfs_fs_info *fs_info = trans->fs_info;
2235 struct btrfs_block_group *block_group;
2238 if (!trans->can_flush_pending_bgs)
2241 while (!list_empty(&trans->new_bgs)) {
2244 block_group = list_first_entry(&trans->new_bgs,
2245 struct btrfs_block_group,
2250 index = btrfs_bg_flags_to_raid_index(block_group->flags);
2252 ret = insert_block_group_item(trans, block_group);
2254 btrfs_abort_transaction(trans, ret);
2255 ret = btrfs_finish_chunk_alloc(trans, block_group->start,
2256 block_group->length);
2258 btrfs_abort_transaction(trans, ret);
2259 add_block_group_free_space(trans, block_group);
2262 * If we restriped during balance, we may have added a new raid
2263 * type, so now add the sysfs entries when it is safe to do so.
2264 * We don't have to worry about locking here as it's handled in
2265 * btrfs_sysfs_add_block_group_type.
2267 if (block_group->space_info->block_group_kobjs[index] == NULL)
2268 btrfs_sysfs_add_block_group_type(block_group);
2270 /* Already aborted the transaction if it failed. */
2272 btrfs_delayed_refs_rsv_release(fs_info, 1);
2273 list_del_init(&block_group->bg_list);
2275 btrfs_trans_release_chunk_metadata(trans);
2278 int btrfs_make_block_group(struct btrfs_trans_handle *trans, u64 bytes_used,
2279 u64 type, u64 chunk_offset, u64 size)
2281 struct btrfs_fs_info *fs_info = trans->fs_info;
2282 struct btrfs_block_group *cache;
2285 btrfs_set_log_full_commit(trans);
2287 cache = btrfs_create_block_group_cache(fs_info, chunk_offset);
2291 cache->length = size;
2292 set_free_space_tree_thresholds(cache);
2293 cache->used = bytes_used;
2294 cache->flags = type;
2295 cache->last_byte_to_unpin = (u64)-1;
2296 cache->cached = BTRFS_CACHE_FINISHED;
2297 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
2298 cache->needs_free_space = 1;
2300 ret = btrfs_load_block_group_zone_info(cache, true);
2302 btrfs_put_block_group(cache);
2306 ret = exclude_super_stripes(cache);
2308 /* We may have excluded something, so call this just in case */
2309 btrfs_free_excluded_extents(cache);
2310 btrfs_put_block_group(cache);
2314 add_new_free_space(cache, chunk_offset, chunk_offset + size);
2316 btrfs_free_excluded_extents(cache);
2318 #ifdef CONFIG_BTRFS_DEBUG
2319 if (btrfs_should_fragment_free_space(cache)) {
2320 u64 new_bytes_used = size - bytes_used;
2322 bytes_used += new_bytes_used >> 1;
2323 fragment_free_space(cache);
2327 * Ensure the corresponding space_info object is created and
2328 * assigned to our block group. We want our bg to be added to the rbtree
2329 * with its ->space_info set.
2331 cache->space_info = btrfs_find_space_info(fs_info, cache->flags);
2332 ASSERT(cache->space_info);
2334 ret = btrfs_add_block_group_cache(fs_info, cache);
2336 btrfs_remove_free_space_cache(cache);
2337 btrfs_put_block_group(cache);
2342 * Now that our block group has its ->space_info set and is inserted in
2343 * the rbtree, update the space info's counters.
2345 trace_btrfs_add_block_group(fs_info, cache, 1);
2346 btrfs_update_space_info(fs_info, cache->flags, size, bytes_used,
2347 cache->bytes_super, 0, &cache->space_info);
2348 btrfs_update_global_block_rsv(fs_info);
2350 link_block_group(cache);
2352 list_add_tail(&cache->bg_list, &trans->new_bgs);
2353 trans->delayed_ref_updates++;
2354 btrfs_update_delayed_refs_rsv(trans);
2356 set_avail_alloc_bits(fs_info, type);
2361 * Mark one block group RO, can be called several times for the same block
2364 * @cache: the destination block group
2365 * @do_chunk_alloc: whether need to do chunk pre-allocation, this is to
2366 * ensure we still have some free space after marking this
2369 int btrfs_inc_block_group_ro(struct btrfs_block_group *cache,
2370 bool do_chunk_alloc)
2372 struct btrfs_fs_info *fs_info = cache->fs_info;
2373 struct btrfs_trans_handle *trans;
2376 bool dirty_bg_running;
2379 trans = btrfs_join_transaction(fs_info->extent_root);
2381 return PTR_ERR(trans);
2383 dirty_bg_running = false;
2386 * We're not allowed to set block groups readonly after the dirty
2387 * block group cache has started writing. If it already started,
2388 * back off and let this transaction commit.
2390 mutex_lock(&fs_info->ro_block_group_mutex);
2391 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
2392 u64 transid = trans->transid;
2394 mutex_unlock(&fs_info->ro_block_group_mutex);
2395 btrfs_end_transaction(trans);
2397 ret = btrfs_wait_for_commit(fs_info, transid);
2400 dirty_bg_running = true;
2402 } while (dirty_bg_running);
2404 if (do_chunk_alloc) {
2406 * If we are changing raid levels, try to allocate a
2407 * corresponding block group with the new raid level.
2409 alloc_flags = btrfs_get_alloc_profile(fs_info, cache->flags);
2410 if (alloc_flags != cache->flags) {
2411 ret = btrfs_chunk_alloc(trans, alloc_flags,
2414 * ENOSPC is allowed here, we may have enough space
2415 * already allocated at the new raid level to carry on
2424 ret = inc_block_group_ro(cache, 0);
2425 if (!do_chunk_alloc || ret == -ETXTBSY)
2429 alloc_flags = btrfs_get_alloc_profile(fs_info, cache->space_info->flags);
2430 ret = btrfs_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE);
2433 ret = inc_block_group_ro(cache, 0);
2434 if (ret == -ETXTBSY)
2437 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
2438 alloc_flags = btrfs_get_alloc_profile(fs_info, cache->flags);
2439 mutex_lock(&fs_info->chunk_mutex);
2440 check_system_chunk(trans, alloc_flags);
2441 mutex_unlock(&fs_info->chunk_mutex);
2444 mutex_unlock(&fs_info->ro_block_group_mutex);
2446 btrfs_end_transaction(trans);
2450 void btrfs_dec_block_group_ro(struct btrfs_block_group *cache)
2452 struct btrfs_space_info *sinfo = cache->space_info;
2457 spin_lock(&sinfo->lock);
2458 spin_lock(&cache->lock);
2460 if (btrfs_is_zoned(cache->fs_info)) {
2461 /* Migrate zone_unusable bytes back */
2462 cache->zone_unusable = cache->alloc_offset - cache->used;
2463 sinfo->bytes_zone_unusable += cache->zone_unusable;
2464 sinfo->bytes_readonly -= cache->zone_unusable;
2466 num_bytes = cache->length - cache->reserved -
2467 cache->pinned - cache->bytes_super -
2468 cache->zone_unusable - cache->used;
2469 sinfo->bytes_readonly -= num_bytes;
2470 list_del_init(&cache->ro_list);
2472 spin_unlock(&cache->lock);
2473 spin_unlock(&sinfo->lock);
2476 static int update_block_group_item(struct btrfs_trans_handle *trans,
2477 struct btrfs_path *path,
2478 struct btrfs_block_group *cache)
2480 struct btrfs_fs_info *fs_info = trans->fs_info;
2482 struct btrfs_root *root = fs_info->extent_root;
2484 struct extent_buffer *leaf;
2485 struct btrfs_block_group_item bgi;
2486 struct btrfs_key key;
2488 key.objectid = cache->start;
2489 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
2490 key.offset = cache->length;
2492 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2499 leaf = path->nodes[0];
2500 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2501 btrfs_set_stack_block_group_used(&bgi, cache->used);
2502 btrfs_set_stack_block_group_chunk_objectid(&bgi,
2503 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
2504 btrfs_set_stack_block_group_flags(&bgi, cache->flags);
2505 write_extent_buffer(leaf, &bgi, bi, sizeof(bgi));
2506 btrfs_mark_buffer_dirty(leaf);
2508 btrfs_release_path(path);
2513 static int cache_save_setup(struct btrfs_block_group *block_group,
2514 struct btrfs_trans_handle *trans,
2515 struct btrfs_path *path)
2517 struct btrfs_fs_info *fs_info = block_group->fs_info;
2518 struct btrfs_root *root = fs_info->tree_root;
2519 struct inode *inode = NULL;
2520 struct extent_changeset *data_reserved = NULL;
2522 int dcs = BTRFS_DC_ERROR;
2527 if (!btrfs_test_opt(fs_info, SPACE_CACHE))
2531 * If this block group is smaller than 100 megs don't bother caching the
2534 if (block_group->length < (100 * SZ_1M)) {
2535 spin_lock(&block_group->lock);
2536 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2537 spin_unlock(&block_group->lock);
2541 if (TRANS_ABORTED(trans))
2544 inode = lookup_free_space_inode(block_group, path);
2545 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2546 ret = PTR_ERR(inode);
2547 btrfs_release_path(path);
2551 if (IS_ERR(inode)) {
2555 if (block_group->ro)
2558 ret = create_free_space_inode(trans, block_group, path);
2565 * We want to set the generation to 0, that way if anything goes wrong
2566 * from here on out we know not to trust this cache when we load up next
2569 BTRFS_I(inode)->generation = 0;
2570 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
2573 * So theoretically we could recover from this, simply set the
2574 * super cache generation to 0 so we know to invalidate the
2575 * cache, but then we'd have to keep track of the block groups
2576 * that fail this way so we know we _have_ to reset this cache
2577 * before the next commit or risk reading stale cache. So to
2578 * limit our exposure to horrible edge cases lets just abort the
2579 * transaction, this only happens in really bad situations
2582 btrfs_abort_transaction(trans, ret);
2587 /* We've already setup this transaction, go ahead and exit */
2588 if (block_group->cache_generation == trans->transid &&
2589 i_size_read(inode)) {
2590 dcs = BTRFS_DC_SETUP;
2594 if (i_size_read(inode) > 0) {
2595 ret = btrfs_check_trunc_cache_free_space(fs_info,
2596 &fs_info->global_block_rsv);
2600 ret = btrfs_truncate_free_space_cache(trans, NULL, inode);
2605 spin_lock(&block_group->lock);
2606 if (block_group->cached != BTRFS_CACHE_FINISHED ||
2607 !btrfs_test_opt(fs_info, SPACE_CACHE)) {
2609 * don't bother trying to write stuff out _if_
2610 * a) we're not cached,
2611 * b) we're with nospace_cache mount option,
2612 * c) we're with v2 space_cache (FREE_SPACE_TREE).
2614 dcs = BTRFS_DC_WRITTEN;
2615 spin_unlock(&block_group->lock);
2618 spin_unlock(&block_group->lock);
2621 * We hit an ENOSPC when setting up the cache in this transaction, just
2622 * skip doing the setup, we've already cleared the cache so we're safe.
2624 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
2630 * Try to preallocate enough space based on how big the block group is.
2631 * Keep in mind this has to include any pinned space which could end up
2632 * taking up quite a bit since it's not folded into the other space
2635 cache_size = div_u64(block_group->length, SZ_256M);
2640 cache_size *= fs_info->sectorsize;
2642 ret = btrfs_check_data_free_space(BTRFS_I(inode), &data_reserved, 0,
2647 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, cache_size,
2648 cache_size, cache_size,
2651 * Our cache requires contiguous chunks so that we don't modify a bunch
2652 * of metadata or split extents when writing the cache out, which means
2653 * we can enospc if we are heavily fragmented in addition to just normal
2654 * out of space conditions. So if we hit this just skip setting up any
2655 * other block groups for this transaction, maybe we'll unpin enough
2656 * space the next time around.
2659 dcs = BTRFS_DC_SETUP;
2660 else if (ret == -ENOSPC)
2661 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
2666 btrfs_release_path(path);
2668 spin_lock(&block_group->lock);
2669 if (!ret && dcs == BTRFS_DC_SETUP)
2670 block_group->cache_generation = trans->transid;
2671 block_group->disk_cache_state = dcs;
2672 spin_unlock(&block_group->lock);
2674 extent_changeset_free(data_reserved);
2678 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans)
2680 struct btrfs_fs_info *fs_info = trans->fs_info;
2681 struct btrfs_block_group *cache, *tmp;
2682 struct btrfs_transaction *cur_trans = trans->transaction;
2683 struct btrfs_path *path;
2685 if (list_empty(&cur_trans->dirty_bgs) ||
2686 !btrfs_test_opt(fs_info, SPACE_CACHE))
2689 path = btrfs_alloc_path();
2693 /* Could add new block groups, use _safe just in case */
2694 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
2696 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2697 cache_save_setup(cache, trans, path);
2700 btrfs_free_path(path);
2705 * Transaction commit does final block group cache writeback during a critical
2706 * section where nothing is allowed to change the FS. This is required in
2707 * order for the cache to actually match the block group, but can introduce a
2708 * lot of latency into the commit.
2710 * So, btrfs_start_dirty_block_groups is here to kick off block group cache IO.
2711 * There's a chance we'll have to redo some of it if the block group changes
2712 * again during the commit, but it greatly reduces the commit latency by
2713 * getting rid of the easy block groups while we're still allowing others to
2716 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans)
2718 struct btrfs_fs_info *fs_info = trans->fs_info;
2719 struct btrfs_block_group *cache;
2720 struct btrfs_transaction *cur_trans = trans->transaction;
2723 struct btrfs_path *path = NULL;
2725 struct list_head *io = &cur_trans->io_bgs;
2726 int num_started = 0;
2729 spin_lock(&cur_trans->dirty_bgs_lock);
2730 if (list_empty(&cur_trans->dirty_bgs)) {
2731 spin_unlock(&cur_trans->dirty_bgs_lock);
2734 list_splice_init(&cur_trans->dirty_bgs, &dirty);
2735 spin_unlock(&cur_trans->dirty_bgs_lock);
2738 /* Make sure all the block groups on our dirty list actually exist */
2739 btrfs_create_pending_block_groups(trans);
2742 path = btrfs_alloc_path();
2750 * cache_write_mutex is here only to save us from balance or automatic
2751 * removal of empty block groups deleting this block group while we are
2752 * writing out the cache
2754 mutex_lock(&trans->transaction->cache_write_mutex);
2755 while (!list_empty(&dirty)) {
2756 bool drop_reserve = true;
2758 cache = list_first_entry(&dirty, struct btrfs_block_group,
2761 * This can happen if something re-dirties a block group that
2762 * is already under IO. Just wait for it to finish and then do
2765 if (!list_empty(&cache->io_list)) {
2766 list_del_init(&cache->io_list);
2767 btrfs_wait_cache_io(trans, cache, path);
2768 btrfs_put_block_group(cache);
2773 * btrfs_wait_cache_io uses the cache->dirty_list to decide if
2774 * it should update the cache_state. Don't delete until after
2777 * Since we're not running in the commit critical section
2778 * we need the dirty_bgs_lock to protect from update_block_group
2780 spin_lock(&cur_trans->dirty_bgs_lock);
2781 list_del_init(&cache->dirty_list);
2782 spin_unlock(&cur_trans->dirty_bgs_lock);
2786 cache_save_setup(cache, trans, path);
2788 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
2789 cache->io_ctl.inode = NULL;
2790 ret = btrfs_write_out_cache(trans, cache, path);
2791 if (ret == 0 && cache->io_ctl.inode) {
2796 * The cache_write_mutex is protecting the
2797 * io_list, also refer to the definition of
2798 * btrfs_transaction::io_bgs for more details
2800 list_add_tail(&cache->io_list, io);
2803 * If we failed to write the cache, the
2804 * generation will be bad and life goes on
2810 ret = update_block_group_item(trans, path, cache);
2812 * Our block group might still be attached to the list
2813 * of new block groups in the transaction handle of some
2814 * other task (struct btrfs_trans_handle->new_bgs). This
2815 * means its block group item isn't yet in the extent
2816 * tree. If this happens ignore the error, as we will
2817 * try again later in the critical section of the
2818 * transaction commit.
2820 if (ret == -ENOENT) {
2822 spin_lock(&cur_trans->dirty_bgs_lock);
2823 if (list_empty(&cache->dirty_list)) {
2824 list_add_tail(&cache->dirty_list,
2825 &cur_trans->dirty_bgs);
2826 btrfs_get_block_group(cache);
2827 drop_reserve = false;
2829 spin_unlock(&cur_trans->dirty_bgs_lock);
2831 btrfs_abort_transaction(trans, ret);
2835 /* If it's not on the io list, we need to put the block group */
2837 btrfs_put_block_group(cache);
2839 btrfs_delayed_refs_rsv_release(fs_info, 1);
2841 * Avoid blocking other tasks for too long. It might even save
2842 * us from writing caches for block groups that are going to be
2845 mutex_unlock(&trans->transaction->cache_write_mutex);
2848 mutex_lock(&trans->transaction->cache_write_mutex);
2850 mutex_unlock(&trans->transaction->cache_write_mutex);
2853 * Go through delayed refs for all the stuff we've just kicked off
2854 * and then loop back (just once)
2857 ret = btrfs_run_delayed_refs(trans, 0);
2858 if (!ret && loops == 0) {
2860 spin_lock(&cur_trans->dirty_bgs_lock);
2861 list_splice_init(&cur_trans->dirty_bgs, &dirty);
2863 * dirty_bgs_lock protects us from concurrent block group
2864 * deletes too (not just cache_write_mutex).
2866 if (!list_empty(&dirty)) {
2867 spin_unlock(&cur_trans->dirty_bgs_lock);
2870 spin_unlock(&cur_trans->dirty_bgs_lock);
2874 spin_lock(&cur_trans->dirty_bgs_lock);
2875 list_splice_init(&dirty, &cur_trans->dirty_bgs);
2876 spin_unlock(&cur_trans->dirty_bgs_lock);
2877 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
2880 btrfs_free_path(path);
2884 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans)
2886 struct btrfs_fs_info *fs_info = trans->fs_info;
2887 struct btrfs_block_group *cache;
2888 struct btrfs_transaction *cur_trans = trans->transaction;
2891 struct btrfs_path *path;
2892 struct list_head *io = &cur_trans->io_bgs;
2893 int num_started = 0;
2895 path = btrfs_alloc_path();
2900 * Even though we are in the critical section of the transaction commit,
2901 * we can still have concurrent tasks adding elements to this
2902 * transaction's list of dirty block groups. These tasks correspond to
2903 * endio free space workers started when writeback finishes for a
2904 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
2905 * allocate new block groups as a result of COWing nodes of the root
2906 * tree when updating the free space inode. The writeback for the space
2907 * caches is triggered by an earlier call to
2908 * btrfs_start_dirty_block_groups() and iterations of the following
2910 * Also we want to do the cache_save_setup first and then run the
2911 * delayed refs to make sure we have the best chance at doing this all
2914 spin_lock(&cur_trans->dirty_bgs_lock);
2915 while (!list_empty(&cur_trans->dirty_bgs)) {
2916 cache = list_first_entry(&cur_trans->dirty_bgs,
2917 struct btrfs_block_group,
2921 * This can happen if cache_save_setup re-dirties a block group
2922 * that is already under IO. Just wait for it to finish and
2923 * then do it all again
2925 if (!list_empty(&cache->io_list)) {
2926 spin_unlock(&cur_trans->dirty_bgs_lock);
2927 list_del_init(&cache->io_list);
2928 btrfs_wait_cache_io(trans, cache, path);
2929 btrfs_put_block_group(cache);
2930 spin_lock(&cur_trans->dirty_bgs_lock);
2934 * Don't remove from the dirty list until after we've waited on
2937 list_del_init(&cache->dirty_list);
2938 spin_unlock(&cur_trans->dirty_bgs_lock);
2941 cache_save_setup(cache, trans, path);
2944 ret = btrfs_run_delayed_refs(trans,
2945 (unsigned long) -1);
2947 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
2948 cache->io_ctl.inode = NULL;
2949 ret = btrfs_write_out_cache(trans, cache, path);
2950 if (ret == 0 && cache->io_ctl.inode) {
2953 list_add_tail(&cache->io_list, io);
2956 * If we failed to write the cache, the
2957 * generation will be bad and life goes on
2963 ret = update_block_group_item(trans, path, cache);
2965 * One of the free space endio workers might have
2966 * created a new block group while updating a free space
2967 * cache's inode (at inode.c:btrfs_finish_ordered_io())
2968 * and hasn't released its transaction handle yet, in
2969 * which case the new block group is still attached to
2970 * its transaction handle and its creation has not
2971 * finished yet (no block group item in the extent tree
2972 * yet, etc). If this is the case, wait for all free
2973 * space endio workers to finish and retry. This is a
2974 * very rare case so no need for a more efficient and
2977 if (ret == -ENOENT) {
2978 wait_event(cur_trans->writer_wait,
2979 atomic_read(&cur_trans->num_writers) == 1);
2980 ret = update_block_group_item(trans, path, cache);
2983 btrfs_abort_transaction(trans, ret);
2986 /* If its not on the io list, we need to put the block group */
2988 btrfs_put_block_group(cache);
2989 btrfs_delayed_refs_rsv_release(fs_info, 1);
2990 spin_lock(&cur_trans->dirty_bgs_lock);
2992 spin_unlock(&cur_trans->dirty_bgs_lock);
2995 * Refer to the definition of io_bgs member for details why it's safe
2996 * to use it without any locking
2998 while (!list_empty(io)) {
2999 cache = list_first_entry(io, struct btrfs_block_group,
3001 list_del_init(&cache->io_list);
3002 btrfs_wait_cache_io(trans, cache, path);
3003 btrfs_put_block_group(cache);
3006 btrfs_free_path(path);
3010 int btrfs_update_block_group(struct btrfs_trans_handle *trans,
3011 u64 bytenr, u64 num_bytes, int alloc)
3013 struct btrfs_fs_info *info = trans->fs_info;
3014 struct btrfs_block_group *cache = NULL;
3015 u64 total = num_bytes;
3021 /* Block accounting for super block */
3022 spin_lock(&info->delalloc_root_lock);
3023 old_val = btrfs_super_bytes_used(info->super_copy);
3025 old_val += num_bytes;
3027 old_val -= num_bytes;
3028 btrfs_set_super_bytes_used(info->super_copy, old_val);
3029 spin_unlock(&info->delalloc_root_lock);
3032 cache = btrfs_lookup_block_group(info, bytenr);
3037 factor = btrfs_bg_type_to_factor(cache->flags);
3040 * If this block group has free space cache written out, we
3041 * need to make sure to load it if we are removing space. This
3042 * is because we need the unpinning stage to actually add the
3043 * space back to the block group, otherwise we will leak space.
3045 if (!alloc && !btrfs_block_group_done(cache))
3046 btrfs_cache_block_group(cache, 1);
3048 byte_in_group = bytenr - cache->start;
3049 WARN_ON(byte_in_group > cache->length);
3051 spin_lock(&cache->space_info->lock);
3052 spin_lock(&cache->lock);
3054 if (btrfs_test_opt(info, SPACE_CACHE) &&
3055 cache->disk_cache_state < BTRFS_DC_CLEAR)
3056 cache->disk_cache_state = BTRFS_DC_CLEAR;
3058 old_val = cache->used;
3059 num_bytes = min(total, cache->length - byte_in_group);
3061 old_val += num_bytes;
3062 cache->used = old_val;
3063 cache->reserved -= num_bytes;
3064 cache->space_info->bytes_reserved -= num_bytes;
3065 cache->space_info->bytes_used += num_bytes;
3066 cache->space_info->disk_used += num_bytes * factor;
3067 spin_unlock(&cache->lock);
3068 spin_unlock(&cache->space_info->lock);
3070 old_val -= num_bytes;
3071 cache->used = old_val;
3072 cache->pinned += num_bytes;
3073 btrfs_space_info_update_bytes_pinned(info,
3074 cache->space_info, num_bytes);
3075 cache->space_info->bytes_used -= num_bytes;
3076 cache->space_info->disk_used -= num_bytes * factor;
3077 spin_unlock(&cache->lock);
3078 spin_unlock(&cache->space_info->lock);
3080 set_extent_dirty(&trans->transaction->pinned_extents,
3081 bytenr, bytenr + num_bytes - 1,
3082 GFP_NOFS | __GFP_NOFAIL);
3085 spin_lock(&trans->transaction->dirty_bgs_lock);
3086 if (list_empty(&cache->dirty_list)) {
3087 list_add_tail(&cache->dirty_list,
3088 &trans->transaction->dirty_bgs);
3089 trans->delayed_ref_updates++;
3090 btrfs_get_block_group(cache);
3092 spin_unlock(&trans->transaction->dirty_bgs_lock);
3095 * No longer have used bytes in this block group, queue it for
3096 * deletion. We do this after adding the block group to the
3097 * dirty list to avoid races between cleaner kthread and space
3100 if (!alloc && old_val == 0) {
3101 if (!btrfs_test_opt(info, DISCARD_ASYNC))
3102 btrfs_mark_bg_unused(cache);
3105 btrfs_put_block_group(cache);
3107 bytenr += num_bytes;
3110 /* Modified block groups are accounted for in the delayed_refs_rsv. */
3111 btrfs_update_delayed_refs_rsv(trans);
3116 * btrfs_add_reserved_bytes - update the block_group and space info counters
3117 * @cache: The cache we are manipulating
3118 * @ram_bytes: The number of bytes of file content, and will be same to
3119 * @num_bytes except for the compress path.
3120 * @num_bytes: The number of bytes in question
3121 * @delalloc: The blocks are allocated for the delalloc write
3123 * This is called by the allocator when it reserves space. If this is a
3124 * reservation and the block group has become read only we cannot make the
3125 * reservation and return -EAGAIN, otherwise this function always succeeds.
3127 int btrfs_add_reserved_bytes(struct btrfs_block_group *cache,
3128 u64 ram_bytes, u64 num_bytes, int delalloc)
3130 struct btrfs_space_info *space_info = cache->space_info;
3133 spin_lock(&space_info->lock);
3134 spin_lock(&cache->lock);
3138 cache->reserved += num_bytes;
3139 space_info->bytes_reserved += num_bytes;
3140 trace_btrfs_space_reservation(cache->fs_info, "space_info",
3141 space_info->flags, num_bytes, 1);
3142 btrfs_space_info_update_bytes_may_use(cache->fs_info,
3143 space_info, -ram_bytes);
3145 cache->delalloc_bytes += num_bytes;
3148 * Compression can use less space than we reserved, so wake
3149 * tickets if that happens
3151 if (num_bytes < ram_bytes)
3152 btrfs_try_granting_tickets(cache->fs_info, space_info);
3154 spin_unlock(&cache->lock);
3155 spin_unlock(&space_info->lock);
3160 * btrfs_free_reserved_bytes - update the block_group and space info counters
3161 * @cache: The cache we are manipulating
3162 * @num_bytes: The number of bytes in question
3163 * @delalloc: The blocks are allocated for the delalloc write
3165 * This is called by somebody who is freeing space that was never actually used
3166 * on disk. For example if you reserve some space for a new leaf in transaction
3167 * A and before transaction A commits you free that leaf, you call this with
3168 * reserve set to 0 in order to clear the reservation.
3170 void btrfs_free_reserved_bytes(struct btrfs_block_group *cache,
3171 u64 num_bytes, int delalloc)
3173 struct btrfs_space_info *space_info = cache->space_info;
3175 spin_lock(&space_info->lock);
3176 spin_lock(&cache->lock);
3178 space_info->bytes_readonly += num_bytes;
3179 cache->reserved -= num_bytes;
3180 space_info->bytes_reserved -= num_bytes;
3181 space_info->max_extent_size = 0;
3184 cache->delalloc_bytes -= num_bytes;
3185 spin_unlock(&cache->lock);
3187 btrfs_try_granting_tickets(cache->fs_info, space_info);
3188 spin_unlock(&space_info->lock);
3191 static void force_metadata_allocation(struct btrfs_fs_info *info)
3193 struct list_head *head = &info->space_info;
3194 struct btrfs_space_info *found;
3196 list_for_each_entry(found, head, list) {
3197 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3198 found->force_alloc = CHUNK_ALLOC_FORCE;
3202 static int should_alloc_chunk(struct btrfs_fs_info *fs_info,
3203 struct btrfs_space_info *sinfo, int force)
3205 u64 bytes_used = btrfs_space_info_used(sinfo, false);
3208 if (force == CHUNK_ALLOC_FORCE)
3212 * in limited mode, we want to have some free space up to
3213 * about 1% of the FS size.
3215 if (force == CHUNK_ALLOC_LIMITED) {
3216 thresh = btrfs_super_total_bytes(fs_info->super_copy);
3217 thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1));
3219 if (sinfo->total_bytes - bytes_used < thresh)
3223 if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8))
3228 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, u64 type)
3230 u64 alloc_flags = btrfs_get_alloc_profile(trans->fs_info, type);
3232 return btrfs_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE);
3236 * If force is CHUNK_ALLOC_FORCE:
3237 * - return 1 if it successfully allocates a chunk,
3238 * - return errors including -ENOSPC otherwise.
3239 * If force is NOT CHUNK_ALLOC_FORCE:
3240 * - return 0 if it doesn't need to allocate a new chunk,
3241 * - return 1 if it successfully allocates a chunk,
3242 * - return errors including -ENOSPC otherwise.
3244 int btrfs_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags,
3245 enum btrfs_chunk_alloc_enum force)
3247 struct btrfs_fs_info *fs_info = trans->fs_info;
3248 struct btrfs_space_info *space_info;
3249 bool wait_for_alloc = false;
3250 bool should_alloc = false;
3253 /* Don't re-enter if we're already allocating a chunk */
3254 if (trans->allocating_chunk)
3257 space_info = btrfs_find_space_info(fs_info, flags);
3261 spin_lock(&space_info->lock);
3262 if (force < space_info->force_alloc)
3263 force = space_info->force_alloc;
3264 should_alloc = should_alloc_chunk(fs_info, space_info, force);
3265 if (space_info->full) {
3266 /* No more free physical space */
3271 spin_unlock(&space_info->lock);
3273 } else if (!should_alloc) {
3274 spin_unlock(&space_info->lock);
3276 } else if (space_info->chunk_alloc) {
3278 * Someone is already allocating, so we need to block
3279 * until this someone is finished and then loop to
3280 * recheck if we should continue with our allocation
3283 wait_for_alloc = true;
3284 spin_unlock(&space_info->lock);
3285 mutex_lock(&fs_info->chunk_mutex);
3286 mutex_unlock(&fs_info->chunk_mutex);
3288 /* Proceed with allocation */
3289 space_info->chunk_alloc = 1;
3290 wait_for_alloc = false;
3291 spin_unlock(&space_info->lock);
3295 } while (wait_for_alloc);
3297 mutex_lock(&fs_info->chunk_mutex);
3298 trans->allocating_chunk = true;
3301 * If we have mixed data/metadata chunks we want to make sure we keep
3302 * allocating mixed chunks instead of individual chunks.
3304 if (btrfs_mixed_space_info(space_info))
3305 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3308 * if we're doing a data chunk, go ahead and make sure that
3309 * we keep a reasonable number of metadata chunks allocated in the
3312 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3313 fs_info->data_chunk_allocations++;
3314 if (!(fs_info->data_chunk_allocations %
3315 fs_info->metadata_ratio))
3316 force_metadata_allocation(fs_info);
3320 * Check if we have enough space in SYSTEM chunk because we may need
3321 * to update devices.
3323 check_system_chunk(trans, flags);
3325 ret = btrfs_alloc_chunk(trans, flags);
3326 trans->allocating_chunk = false;
3328 spin_lock(&space_info->lock);
3331 space_info->full = 1;
3336 space_info->max_extent_size = 0;
3339 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3341 space_info->chunk_alloc = 0;
3342 spin_unlock(&space_info->lock);
3343 mutex_unlock(&fs_info->chunk_mutex);
3345 * When we allocate a new chunk we reserve space in the chunk block
3346 * reserve to make sure we can COW nodes/leafs in the chunk tree or
3347 * add new nodes/leafs to it if we end up needing to do it when
3348 * inserting the chunk item and updating device items as part of the
3349 * second phase of chunk allocation, performed by
3350 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
3351 * large number of new block groups to create in our transaction
3352 * handle's new_bgs list to avoid exhausting the chunk block reserve
3353 * in extreme cases - like having a single transaction create many new
3354 * block groups when starting to write out the free space caches of all
3355 * the block groups that were made dirty during the lifetime of the
3358 if (trans->chunk_bytes_reserved >= (u64)SZ_2M)
3359 btrfs_create_pending_block_groups(trans);
3364 static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type)
3368 num_dev = btrfs_raid_array[btrfs_bg_flags_to_raid_index(type)].devs_max;
3370 num_dev = fs_info->fs_devices->rw_devices;
3376 * Reserve space in the system space for allocating or removing a chunk
3378 void check_system_chunk(struct btrfs_trans_handle *trans, u64 type)
3380 struct btrfs_transaction *cur_trans = trans->transaction;
3381 struct btrfs_fs_info *fs_info = trans->fs_info;
3382 struct btrfs_space_info *info;
3389 * Needed because we can end up allocating a system chunk and for an
3390 * atomic and race free space reservation in the chunk block reserve.
3392 lockdep_assert_held(&fs_info->chunk_mutex);
3394 info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3396 spin_lock(&info->lock);
3397 left = info->total_bytes - btrfs_space_info_used(info, true);
3398 spin_unlock(&info->lock);
3400 num_devs = get_profile_num_devs(fs_info, type);
3402 /* num_devs device items to update and 1 chunk item to add or remove */
3403 thresh = btrfs_calc_metadata_size(fs_info, num_devs) +
3404 btrfs_calc_insert_metadata_size(fs_info, 1);
3406 if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
3407 btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu",
3408 left, thresh, type);
3409 btrfs_dump_space_info(fs_info, info, 0, 0);
3412 if (left < thresh) {
3413 u64 flags = btrfs_system_alloc_profile(fs_info);
3414 u64 reserved = atomic64_read(&cur_trans->chunk_bytes_reserved);
3417 * If there's not available space for the chunk tree (system
3418 * space) and there are other tasks that reserved space for
3419 * creating a new system block group, wait for them to complete
3420 * the creation of their system block group and release excess
3421 * reserved space. We do this because:
3423 * *) We can end up allocating more system chunks than necessary
3424 * when there are multiple tasks that are concurrently
3425 * allocating block groups, which can lead to exhaustion of
3426 * the system array in the superblock;
3428 * *) If we allocate extra and unnecessary system block groups,
3429 * despite being empty for a long time, and possibly forever,
3430 * they end not being added to the list of unused block groups
3431 * because that typically happens only when deallocating the
3432 * last extent from a block group - which never happens since
3433 * we never allocate from them in the first place. The few
3434 * exceptions are when mounting a filesystem or running scrub,
3435 * which add unused block groups to the list of unused block
3436 * groups, to be deleted by the cleaner kthread.
3437 * And even when they are added to the list of unused block
3438 * groups, it can take a long time until they get deleted,
3439 * since the cleaner kthread might be sleeping or busy with
3440 * other work (deleting subvolumes, running delayed iputs,
3441 * defrag scheduling, etc);
3443 * This is rare in practice, but can happen when too many tasks
3444 * are allocating blocks groups in parallel (via fallocate())
3445 * and before the one that reserved space for a new system block
3446 * group finishes the block group creation and releases the space
3447 * reserved in excess (at btrfs_create_pending_block_groups()),
3448 * other tasks end up here and see free system space temporarily
3449 * not enough for updating the chunk tree.
3451 * We unlock the chunk mutex before waiting for such tasks and
3452 * lock it again after the wait, otherwise we would deadlock.
3453 * It is safe to do so because allocating a system chunk is the
3454 * first thing done while allocating a new block group.
3456 if (reserved > trans->chunk_bytes_reserved) {
3457 const u64 min_needed = reserved - thresh;
3459 mutex_unlock(&fs_info->chunk_mutex);
3460 wait_event(cur_trans->chunk_reserve_wait,
3461 atomic64_read(&cur_trans->chunk_bytes_reserved) <=
3463 mutex_lock(&fs_info->chunk_mutex);
3468 * Ignore failure to create system chunk. We might end up not
3469 * needing it, as we might not need to COW all nodes/leafs from
3470 * the paths we visit in the chunk tree (they were already COWed
3471 * or created in the current transaction for example).
3473 ret = btrfs_alloc_chunk(trans, flags);
3477 ret = btrfs_block_rsv_add(fs_info->chunk_root,
3478 &fs_info->chunk_block_rsv,
3479 thresh, BTRFS_RESERVE_NO_FLUSH);
3481 atomic64_add(thresh, &cur_trans->chunk_bytes_reserved);
3482 trans->chunk_bytes_reserved += thresh;
3487 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
3489 struct btrfs_block_group *block_group;
3493 struct inode *inode;
3495 block_group = btrfs_lookup_first_block_group(info, last);
3496 while (block_group) {
3497 btrfs_wait_block_group_cache_done(block_group);
3498 spin_lock(&block_group->lock);
3499 if (block_group->iref)
3501 spin_unlock(&block_group->lock);
3502 block_group = btrfs_next_block_group(block_group);
3511 inode = block_group->inode;
3512 block_group->iref = 0;
3513 block_group->inode = NULL;
3514 spin_unlock(&block_group->lock);
3515 ASSERT(block_group->io_ctl.inode == NULL);
3517 last = block_group->start + block_group->length;
3518 btrfs_put_block_group(block_group);
3523 * Must be called only after stopping all workers, since we could have block
3524 * group caching kthreads running, and therefore they could race with us if we
3525 * freed the block groups before stopping them.
3527 int btrfs_free_block_groups(struct btrfs_fs_info *info)
3529 struct btrfs_block_group *block_group;
3530 struct btrfs_space_info *space_info;
3531 struct btrfs_caching_control *caching_ctl;
3534 spin_lock(&info->block_group_cache_lock);
3535 while (!list_empty(&info->caching_block_groups)) {
3536 caching_ctl = list_entry(info->caching_block_groups.next,
3537 struct btrfs_caching_control, list);
3538 list_del(&caching_ctl->list);
3539 btrfs_put_caching_control(caching_ctl);
3541 spin_unlock(&info->block_group_cache_lock);
3543 spin_lock(&info->unused_bgs_lock);
3544 while (!list_empty(&info->unused_bgs)) {
3545 block_group = list_first_entry(&info->unused_bgs,
3546 struct btrfs_block_group,
3548 list_del_init(&block_group->bg_list);
3549 btrfs_put_block_group(block_group);
3551 spin_unlock(&info->unused_bgs_lock);
3553 spin_lock(&info->unused_bgs_lock);
3554 while (!list_empty(&info->reclaim_bgs)) {
3555 block_group = list_first_entry(&info->reclaim_bgs,
3556 struct btrfs_block_group,
3558 list_del_init(&block_group->bg_list);
3559 btrfs_put_block_group(block_group);
3561 spin_unlock(&info->unused_bgs_lock);
3563 spin_lock(&info->block_group_cache_lock);
3564 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
3565 block_group = rb_entry(n, struct btrfs_block_group,
3567 rb_erase(&block_group->cache_node,
3568 &info->block_group_cache_tree);
3569 RB_CLEAR_NODE(&block_group->cache_node);
3570 spin_unlock(&info->block_group_cache_lock);
3572 down_write(&block_group->space_info->groups_sem);
3573 list_del(&block_group->list);
3574 up_write(&block_group->space_info->groups_sem);
3577 * We haven't cached this block group, which means we could
3578 * possibly have excluded extents on this block group.
3580 if (block_group->cached == BTRFS_CACHE_NO ||
3581 block_group->cached == BTRFS_CACHE_ERROR)
3582 btrfs_free_excluded_extents(block_group);
3584 btrfs_remove_free_space_cache(block_group);
3585 ASSERT(block_group->cached != BTRFS_CACHE_STARTED);
3586 ASSERT(list_empty(&block_group->dirty_list));
3587 ASSERT(list_empty(&block_group->io_list));
3588 ASSERT(list_empty(&block_group->bg_list));
3589 ASSERT(refcount_read(&block_group->refs) == 1);
3590 ASSERT(block_group->swap_extents == 0);
3591 btrfs_put_block_group(block_group);
3593 spin_lock(&info->block_group_cache_lock);
3595 spin_unlock(&info->block_group_cache_lock);
3597 btrfs_release_global_block_rsv(info);
3599 while (!list_empty(&info->space_info)) {
3600 space_info = list_entry(info->space_info.next,
3601 struct btrfs_space_info,
3605 * Do not hide this behind enospc_debug, this is actually
3606 * important and indicates a real bug if this happens.
3608 if (WARN_ON(space_info->bytes_pinned > 0 ||
3609 space_info->bytes_reserved > 0 ||
3610 space_info->bytes_may_use > 0))
3611 btrfs_dump_space_info(info, space_info, 0, 0);
3612 WARN_ON(space_info->reclaim_size > 0);
3613 list_del(&space_info->list);
3614 btrfs_sysfs_remove_space_info(space_info);
3619 void btrfs_freeze_block_group(struct btrfs_block_group *cache)
3621 atomic_inc(&cache->frozen);
3624 void btrfs_unfreeze_block_group(struct btrfs_block_group *block_group)
3626 struct btrfs_fs_info *fs_info = block_group->fs_info;
3627 struct extent_map_tree *em_tree;
3628 struct extent_map *em;
3631 spin_lock(&block_group->lock);
3632 cleanup = (atomic_dec_and_test(&block_group->frozen) &&
3633 block_group->removed);
3634 spin_unlock(&block_group->lock);
3637 em_tree = &fs_info->mapping_tree;
3638 write_lock(&em_tree->lock);
3639 em = lookup_extent_mapping(em_tree, block_group->start,
3641 BUG_ON(!em); /* logic error, can't happen */
3642 remove_extent_mapping(em_tree, em);
3643 write_unlock(&em_tree->lock);
3645 /* once for us and once for the tree */
3646 free_extent_map(em);
3647 free_extent_map(em);
3650 * We may have left one free space entry and other possible
3651 * tasks trimming this block group have left 1 entry each one.
3654 __btrfs_remove_free_space_cache(block_group->free_space_ctl);
3658 bool btrfs_inc_block_group_swap_extents(struct btrfs_block_group *bg)
3662 spin_lock(&bg->lock);
3667 spin_unlock(&bg->lock);
3672 void btrfs_dec_block_group_swap_extents(struct btrfs_block_group *bg, int amount)
3674 spin_lock(&bg->lock);
3676 ASSERT(bg->swap_extents >= amount);
3677 bg->swap_extents -= amount;
3678 spin_unlock(&bg->lock);