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);
1171 num_bytes = cache->length - cache->reserved - cache->pinned -
1172 cache->bytes_super - cache->zone_unusable - cache->used;
1175 * Data never overcommits, even in mixed mode, so do just the straight
1176 * check of left over space in how much we have allocated.
1180 } else if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA) {
1181 u64 sinfo_used = btrfs_space_info_used(sinfo, true);
1184 * Here we make sure if we mark this bg RO, we still have enough
1185 * free space as buffer.
1187 if (sinfo_used + num_bytes <= sinfo->total_bytes)
1191 * We overcommit metadata, so we need to do the
1192 * btrfs_can_overcommit check here, and we need to pass in
1193 * BTRFS_RESERVE_NO_FLUSH to give ourselves the most amount of
1194 * leeway to allow us to mark this block group as read only.
1196 if (btrfs_can_overcommit(cache->fs_info, sinfo, num_bytes,
1197 BTRFS_RESERVE_NO_FLUSH))
1202 sinfo->bytes_readonly += num_bytes;
1203 if (btrfs_is_zoned(cache->fs_info)) {
1204 /* Migrate zone_unusable bytes to readonly */
1205 sinfo->bytes_readonly += cache->zone_unusable;
1206 sinfo->bytes_zone_unusable -= cache->zone_unusable;
1207 cache->zone_unusable = 0;
1210 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
1213 spin_unlock(&cache->lock);
1214 spin_unlock(&sinfo->lock);
1215 if (ret == -ENOSPC && btrfs_test_opt(cache->fs_info, ENOSPC_DEBUG)) {
1216 btrfs_info(cache->fs_info,
1217 "unable to make block group %llu ro", cache->start);
1218 btrfs_dump_space_info(cache->fs_info, cache->space_info, 0, 0);
1223 static bool clean_pinned_extents(struct btrfs_trans_handle *trans,
1224 struct btrfs_block_group *bg)
1226 struct btrfs_fs_info *fs_info = bg->fs_info;
1227 struct btrfs_transaction *prev_trans = NULL;
1228 const u64 start = bg->start;
1229 const u64 end = start + bg->length - 1;
1232 spin_lock(&fs_info->trans_lock);
1233 if (trans->transaction->list.prev != &fs_info->trans_list) {
1234 prev_trans = list_last_entry(&trans->transaction->list,
1235 struct btrfs_transaction, list);
1236 refcount_inc(&prev_trans->use_count);
1238 spin_unlock(&fs_info->trans_lock);
1241 * Hold the unused_bg_unpin_mutex lock to avoid racing with
1242 * btrfs_finish_extent_commit(). If we are at transaction N, another
1243 * task might be running finish_extent_commit() for the previous
1244 * transaction N - 1, and have seen a range belonging to the block
1245 * group in pinned_extents before we were able to clear the whole block
1246 * group range from pinned_extents. This means that task can lookup for
1247 * the block group after we unpinned it from pinned_extents and removed
1248 * it, leading to a BUG_ON() at unpin_extent_range().
1250 mutex_lock(&fs_info->unused_bg_unpin_mutex);
1252 ret = clear_extent_bits(&prev_trans->pinned_extents, start, end,
1258 ret = clear_extent_bits(&trans->transaction->pinned_extents, start, end,
1261 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
1263 btrfs_put_transaction(prev_trans);
1269 * Process the unused_bgs list and remove any that don't have any allocated
1270 * space inside of them.
1272 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
1274 struct btrfs_block_group *block_group;
1275 struct btrfs_space_info *space_info;
1276 struct btrfs_trans_handle *trans;
1277 const bool async_trim_enabled = btrfs_test_opt(fs_info, DISCARD_ASYNC);
1280 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1284 * Long running balances can keep us blocked here for eternity, so
1285 * simply skip deletion if we're unable to get the mutex.
1287 if (!mutex_trylock(&fs_info->delete_unused_bgs_mutex))
1290 spin_lock(&fs_info->unused_bgs_lock);
1291 while (!list_empty(&fs_info->unused_bgs)) {
1294 block_group = list_first_entry(&fs_info->unused_bgs,
1295 struct btrfs_block_group,
1297 list_del_init(&block_group->bg_list);
1299 space_info = block_group->space_info;
1301 if (ret || btrfs_mixed_space_info(space_info)) {
1302 btrfs_put_block_group(block_group);
1305 spin_unlock(&fs_info->unused_bgs_lock);
1307 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group);
1309 /* Don't want to race with allocators so take the groups_sem */
1310 down_write(&space_info->groups_sem);
1313 * Async discard moves the final block group discard to be prior
1314 * to the unused_bgs code path. Therefore, if it's not fully
1315 * trimmed, punt it back to the async discard lists.
1317 if (btrfs_test_opt(fs_info, DISCARD_ASYNC) &&
1318 !btrfs_is_free_space_trimmed(block_group)) {
1319 trace_btrfs_skip_unused_block_group(block_group);
1320 up_write(&space_info->groups_sem);
1321 /* Requeue if we failed because of async discard */
1322 btrfs_discard_queue_work(&fs_info->discard_ctl,
1327 spin_lock(&block_group->lock);
1328 if (block_group->reserved || block_group->pinned ||
1329 block_group->used || block_group->ro ||
1330 list_is_singular(&block_group->list)) {
1332 * We want to bail if we made new allocations or have
1333 * outstanding allocations in this block group. We do
1334 * the ro check in case balance is currently acting on
1337 trace_btrfs_skip_unused_block_group(block_group);
1338 spin_unlock(&block_group->lock);
1339 up_write(&space_info->groups_sem);
1342 spin_unlock(&block_group->lock);
1344 /* We don't want to force the issue, only flip if it's ok. */
1345 ret = inc_block_group_ro(block_group, 0);
1346 up_write(&space_info->groups_sem);
1353 * Want to do this before we do anything else so we can recover
1354 * properly if we fail to join the transaction.
1356 trans = btrfs_start_trans_remove_block_group(fs_info,
1357 block_group->start);
1358 if (IS_ERR(trans)) {
1359 btrfs_dec_block_group_ro(block_group);
1360 ret = PTR_ERR(trans);
1365 * We could have pending pinned extents for this block group,
1366 * just delete them, we don't care about them anymore.
1368 if (!clean_pinned_extents(trans, block_group)) {
1369 btrfs_dec_block_group_ro(block_group);
1374 * At this point, the block_group is read only and should fail
1375 * new allocations. However, btrfs_finish_extent_commit() can
1376 * cause this block_group to be placed back on the discard
1377 * lists because now the block_group isn't fully discarded.
1378 * Bail here and try again later after discarding everything.
1380 spin_lock(&fs_info->discard_ctl.lock);
1381 if (!list_empty(&block_group->discard_list)) {
1382 spin_unlock(&fs_info->discard_ctl.lock);
1383 btrfs_dec_block_group_ro(block_group);
1384 btrfs_discard_queue_work(&fs_info->discard_ctl,
1388 spin_unlock(&fs_info->discard_ctl.lock);
1390 /* Reset pinned so btrfs_put_block_group doesn't complain */
1391 spin_lock(&space_info->lock);
1392 spin_lock(&block_group->lock);
1394 btrfs_space_info_update_bytes_pinned(fs_info, space_info,
1395 -block_group->pinned);
1396 space_info->bytes_readonly += block_group->pinned;
1397 __btrfs_mod_total_bytes_pinned(space_info, -block_group->pinned);
1398 block_group->pinned = 0;
1400 spin_unlock(&block_group->lock);
1401 spin_unlock(&space_info->lock);
1404 * The normal path here is an unused block group is passed here,
1405 * then trimming is handled in the transaction commit path.
1406 * Async discard interposes before this to do the trimming
1407 * before coming down the unused block group path as trimming
1408 * will no longer be done later in the transaction commit path.
1410 if (!async_trim_enabled && btrfs_test_opt(fs_info, DISCARD_ASYNC))
1414 * DISCARD can flip during remount. On zoned filesystems, we
1415 * need to reset sequential-required zones.
1417 trimming = btrfs_test_opt(fs_info, DISCARD_SYNC) ||
1418 btrfs_is_zoned(fs_info);
1420 /* Implicit trim during transaction commit. */
1422 btrfs_freeze_block_group(block_group);
1425 * Btrfs_remove_chunk will abort the transaction if things go
1428 ret = btrfs_remove_chunk(trans, block_group->start);
1432 btrfs_unfreeze_block_group(block_group);
1437 * If we're not mounted with -odiscard, we can just forget
1438 * about this block group. Otherwise we'll need to wait
1439 * until transaction commit to do the actual discard.
1442 spin_lock(&fs_info->unused_bgs_lock);
1444 * A concurrent scrub might have added us to the list
1445 * fs_info->unused_bgs, so use a list_move operation
1446 * to add the block group to the deleted_bgs list.
1448 list_move(&block_group->bg_list,
1449 &trans->transaction->deleted_bgs);
1450 spin_unlock(&fs_info->unused_bgs_lock);
1451 btrfs_get_block_group(block_group);
1454 btrfs_end_transaction(trans);
1456 btrfs_put_block_group(block_group);
1457 spin_lock(&fs_info->unused_bgs_lock);
1459 spin_unlock(&fs_info->unused_bgs_lock);
1460 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
1464 btrfs_end_transaction(trans);
1465 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
1466 btrfs_put_block_group(block_group);
1467 btrfs_discard_punt_unused_bgs_list(fs_info);
1470 void btrfs_mark_bg_unused(struct btrfs_block_group *bg)
1472 struct btrfs_fs_info *fs_info = bg->fs_info;
1474 spin_lock(&fs_info->unused_bgs_lock);
1475 if (list_empty(&bg->bg_list)) {
1476 btrfs_get_block_group(bg);
1477 trace_btrfs_add_unused_block_group(bg);
1478 list_add_tail(&bg->bg_list, &fs_info->unused_bgs);
1480 spin_unlock(&fs_info->unused_bgs_lock);
1483 static int read_bg_from_eb(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
1484 struct btrfs_path *path)
1486 struct extent_map_tree *em_tree;
1487 struct extent_map *em;
1488 struct btrfs_block_group_item bg;
1489 struct extent_buffer *leaf;
1494 slot = path->slots[0];
1495 leaf = path->nodes[0];
1497 em_tree = &fs_info->mapping_tree;
1498 read_lock(&em_tree->lock);
1499 em = lookup_extent_mapping(em_tree, key->objectid, key->offset);
1500 read_unlock(&em_tree->lock);
1503 "logical %llu len %llu found bg but no related chunk",
1504 key->objectid, key->offset);
1508 if (em->start != key->objectid || em->len != key->offset) {
1510 "block group %llu len %llu mismatch with chunk %llu len %llu",
1511 key->objectid, key->offset, em->start, em->len);
1516 read_extent_buffer(leaf, &bg, btrfs_item_ptr_offset(leaf, slot),
1518 flags = btrfs_stack_block_group_flags(&bg) &
1519 BTRFS_BLOCK_GROUP_TYPE_MASK;
1521 if (flags != (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
1523 "block group %llu len %llu type flags 0x%llx mismatch with chunk type flags 0x%llx",
1524 key->objectid, key->offset, flags,
1525 (BTRFS_BLOCK_GROUP_TYPE_MASK & em->map_lookup->type));
1530 free_extent_map(em);
1534 static int find_first_block_group(struct btrfs_fs_info *fs_info,
1535 struct btrfs_path *path,
1536 struct btrfs_key *key)
1538 struct btrfs_root *root = fs_info->extent_root;
1540 struct btrfs_key found_key;
1541 struct extent_buffer *leaf;
1544 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
1549 slot = path->slots[0];
1550 leaf = path->nodes[0];
1551 if (slot >= btrfs_header_nritems(leaf)) {
1552 ret = btrfs_next_leaf(root, path);
1559 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1561 if (found_key.objectid >= key->objectid &&
1562 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
1563 ret = read_bg_from_eb(fs_info, &found_key, path);
1573 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1575 u64 extra_flags = chunk_to_extended(flags) &
1576 BTRFS_EXTENDED_PROFILE_MASK;
1578 write_seqlock(&fs_info->profiles_lock);
1579 if (flags & BTRFS_BLOCK_GROUP_DATA)
1580 fs_info->avail_data_alloc_bits |= extra_flags;
1581 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1582 fs_info->avail_metadata_alloc_bits |= extra_flags;
1583 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1584 fs_info->avail_system_alloc_bits |= extra_flags;
1585 write_sequnlock(&fs_info->profiles_lock);
1589 * Map a physical disk address to a list of logical addresses
1591 * @fs_info: the filesystem
1592 * @chunk_start: logical address of block group
1593 * @bdev: physical device to resolve, can be NULL to indicate any device
1594 * @physical: physical address to map to logical addresses
1595 * @logical: return array of logical addresses which map to @physical
1596 * @naddrs: length of @logical
1597 * @stripe_len: size of IO stripe for the given block group
1599 * Maps a particular @physical disk address to a list of @logical addresses.
1600 * Used primarily to exclude those portions of a block group that contain super
1603 int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start,
1604 struct block_device *bdev, u64 physical, u64 **logical,
1605 int *naddrs, int *stripe_len)
1607 struct extent_map *em;
1608 struct map_lookup *map;
1611 u64 data_stripe_length;
1616 em = btrfs_get_chunk_map(fs_info, chunk_start, 1);
1620 map = em->map_lookup;
1621 data_stripe_length = em->orig_block_len;
1622 io_stripe_size = map->stripe_len;
1623 chunk_start = em->start;
1625 /* For RAID5/6 adjust to a full IO stripe length */
1626 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
1627 io_stripe_size = map->stripe_len * nr_data_stripes(map);
1629 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
1635 for (i = 0; i < map->num_stripes; i++) {
1636 bool already_inserted = false;
1641 if (!in_range(physical, map->stripes[i].physical,
1642 data_stripe_length))
1645 if (bdev && map->stripes[i].dev->bdev != bdev)
1648 stripe_nr = physical - map->stripes[i].physical;
1649 stripe_nr = div64_u64_rem(stripe_nr, map->stripe_len, &offset);
1651 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1652 stripe_nr = stripe_nr * map->num_stripes + i;
1653 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
1654 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
1655 stripe_nr = stripe_nr * map->num_stripes + i;
1658 * The remaining case would be for RAID56, multiply by
1659 * nr_data_stripes(). Alternatively, just use rmap_len below
1660 * instead of map->stripe_len
1663 bytenr = chunk_start + stripe_nr * io_stripe_size + offset;
1665 /* Ensure we don't add duplicate addresses */
1666 for (j = 0; j < nr; j++) {
1667 if (buf[j] == bytenr) {
1668 already_inserted = true;
1673 if (!already_inserted)
1679 *stripe_len = io_stripe_size;
1681 free_extent_map(em);
1685 static int exclude_super_stripes(struct btrfs_block_group *cache)
1687 struct btrfs_fs_info *fs_info = cache->fs_info;
1688 const bool zoned = btrfs_is_zoned(fs_info);
1694 if (cache->start < BTRFS_SUPER_INFO_OFFSET) {
1695 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->start;
1696 cache->bytes_super += stripe_len;
1697 ret = btrfs_add_excluded_extent(fs_info, cache->start,
1703 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1704 bytenr = btrfs_sb_offset(i);
1705 ret = btrfs_rmap_block(fs_info, cache->start, NULL,
1706 bytenr, &logical, &nr, &stripe_len);
1710 /* Shouldn't have super stripes in sequential zones */
1713 "zoned: block group %llu must not contain super block",
1719 u64 len = min_t(u64, stripe_len,
1720 cache->start + cache->length - logical[nr]);
1722 cache->bytes_super += len;
1723 ret = btrfs_add_excluded_extent(fs_info, logical[nr],
1736 static void link_block_group(struct btrfs_block_group *cache)
1738 struct btrfs_space_info *space_info = cache->space_info;
1739 int index = btrfs_bg_flags_to_raid_index(cache->flags);
1741 down_write(&space_info->groups_sem);
1742 list_add_tail(&cache->list, &space_info->block_groups[index]);
1743 up_write(&space_info->groups_sem);
1746 static struct btrfs_block_group *btrfs_create_block_group_cache(
1747 struct btrfs_fs_info *fs_info, u64 start)
1749 struct btrfs_block_group *cache;
1751 cache = kzalloc(sizeof(*cache), GFP_NOFS);
1755 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
1757 if (!cache->free_space_ctl) {
1762 cache->start = start;
1764 cache->fs_info = fs_info;
1765 cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start);
1767 cache->discard_index = BTRFS_DISCARD_INDEX_UNUSED;
1769 refcount_set(&cache->refs, 1);
1770 spin_lock_init(&cache->lock);
1771 init_rwsem(&cache->data_rwsem);
1772 INIT_LIST_HEAD(&cache->list);
1773 INIT_LIST_HEAD(&cache->cluster_list);
1774 INIT_LIST_HEAD(&cache->bg_list);
1775 INIT_LIST_HEAD(&cache->ro_list);
1776 INIT_LIST_HEAD(&cache->discard_list);
1777 INIT_LIST_HEAD(&cache->dirty_list);
1778 INIT_LIST_HEAD(&cache->io_list);
1779 btrfs_init_free_space_ctl(cache, cache->free_space_ctl);
1780 atomic_set(&cache->frozen, 0);
1781 mutex_init(&cache->free_space_lock);
1782 btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root);
1788 * Iterate all chunks and verify that each of them has the corresponding block
1791 static int check_chunk_block_group_mappings(struct btrfs_fs_info *fs_info)
1793 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
1794 struct extent_map *em;
1795 struct btrfs_block_group *bg;
1800 read_lock(&map_tree->lock);
1802 * lookup_extent_mapping will return the first extent map
1803 * intersecting the range, so setting @len to 1 is enough to
1804 * get the first chunk.
1806 em = lookup_extent_mapping(map_tree, start, 1);
1807 read_unlock(&map_tree->lock);
1811 bg = btrfs_lookup_block_group(fs_info, em->start);
1814 "chunk start=%llu len=%llu doesn't have corresponding block group",
1815 em->start, em->len);
1817 free_extent_map(em);
1820 if (bg->start != em->start || bg->length != em->len ||
1821 (bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK) !=
1822 (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
1824 "chunk start=%llu len=%llu flags=0x%llx doesn't match block group start=%llu len=%llu flags=0x%llx",
1826 em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK,
1827 bg->start, bg->length,
1828 bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK);
1830 free_extent_map(em);
1831 btrfs_put_block_group(bg);
1834 start = em->start + em->len;
1835 free_extent_map(em);
1836 btrfs_put_block_group(bg);
1841 static int read_one_block_group(struct btrfs_fs_info *info,
1842 struct btrfs_block_group_item *bgi,
1843 const struct btrfs_key *key,
1846 struct btrfs_block_group *cache;
1847 struct btrfs_space_info *space_info;
1848 const bool mixed = btrfs_fs_incompat(info, MIXED_GROUPS);
1851 ASSERT(key->type == BTRFS_BLOCK_GROUP_ITEM_KEY);
1853 cache = btrfs_create_block_group_cache(info, key->objectid);
1857 cache->length = key->offset;
1858 cache->used = btrfs_stack_block_group_used(bgi);
1859 cache->flags = btrfs_stack_block_group_flags(bgi);
1861 set_free_space_tree_thresholds(cache);
1865 * When we mount with old space cache, we need to
1866 * set BTRFS_DC_CLEAR and set dirty flag.
1868 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
1869 * truncate the old free space cache inode and
1871 * b) Setting 'dirty flag' makes sure that we flush
1872 * the new space cache info onto disk.
1874 if (btrfs_test_opt(info, SPACE_CACHE))
1875 cache->disk_cache_state = BTRFS_DC_CLEAR;
1877 if (!mixed && ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) &&
1878 (cache->flags & BTRFS_BLOCK_GROUP_DATA))) {
1880 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
1886 ret = btrfs_load_block_group_zone_info(cache, false);
1888 btrfs_err(info, "zoned: failed to load zone info of bg %llu",
1894 * We need to exclude the super stripes now so that the space info has
1895 * super bytes accounted for, otherwise we'll think we have more space
1896 * than we actually do.
1898 ret = exclude_super_stripes(cache);
1900 /* We may have excluded something, so call this just in case. */
1901 btrfs_free_excluded_extents(cache);
1906 * For zoned filesystem, space after the allocation offset is the only
1907 * free space for a block group. So, we don't need any caching work.
1908 * btrfs_calc_zone_unusable() will set the amount of free space and
1909 * zone_unusable space.
1911 * For regular filesystem, check for two cases, either we are full, and
1912 * therefore don't need to bother with the caching work since we won't
1913 * find any space, or we are empty, and we can just add all the space
1914 * in and be done with it. This saves us _a_lot_ of time, particularly
1917 if (btrfs_is_zoned(info)) {
1918 btrfs_calc_zone_unusable(cache);
1919 } else if (cache->length == cache->used) {
1920 cache->last_byte_to_unpin = (u64)-1;
1921 cache->cached = BTRFS_CACHE_FINISHED;
1922 btrfs_free_excluded_extents(cache);
1923 } else if (cache->used == 0) {
1924 cache->last_byte_to_unpin = (u64)-1;
1925 cache->cached = BTRFS_CACHE_FINISHED;
1926 add_new_free_space(cache, cache->start,
1927 cache->start + cache->length);
1928 btrfs_free_excluded_extents(cache);
1931 ret = btrfs_add_block_group_cache(info, cache);
1933 btrfs_remove_free_space_cache(cache);
1936 trace_btrfs_add_block_group(info, cache, 0);
1937 btrfs_update_space_info(info, cache->flags, cache->length,
1938 cache->used, cache->bytes_super,
1939 cache->zone_unusable, &space_info);
1941 cache->space_info = space_info;
1943 link_block_group(cache);
1945 set_avail_alloc_bits(info, cache->flags);
1946 if (btrfs_chunk_readonly(info, cache->start)) {
1947 inc_block_group_ro(cache, 1);
1948 } else if (cache->used == 0) {
1949 ASSERT(list_empty(&cache->bg_list));
1950 if (btrfs_test_opt(info, DISCARD_ASYNC))
1951 btrfs_discard_queue_work(&info->discard_ctl, cache);
1953 btrfs_mark_bg_unused(cache);
1957 btrfs_put_block_group(cache);
1961 static int fill_dummy_bgs(struct btrfs_fs_info *fs_info)
1963 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
1964 struct btrfs_space_info *space_info;
1965 struct rb_node *node;
1968 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) {
1969 struct extent_map *em;
1970 struct map_lookup *map;
1971 struct btrfs_block_group *bg;
1973 em = rb_entry(node, struct extent_map, rb_node);
1974 map = em->map_lookup;
1975 bg = btrfs_create_block_group_cache(fs_info, em->start);
1981 /* Fill dummy cache as FULL */
1982 bg->length = em->len;
1983 bg->flags = map->type;
1984 bg->last_byte_to_unpin = (u64)-1;
1985 bg->cached = BTRFS_CACHE_FINISHED;
1987 bg->flags = map->type;
1988 ret = btrfs_add_block_group_cache(fs_info, bg);
1990 btrfs_remove_free_space_cache(bg);
1991 btrfs_put_block_group(bg);
1994 btrfs_update_space_info(fs_info, bg->flags, em->len, em->len,
1996 bg->space_info = space_info;
1997 link_block_group(bg);
1999 set_avail_alloc_bits(fs_info, bg->flags);
2002 btrfs_init_global_block_rsv(fs_info);
2006 int btrfs_read_block_groups(struct btrfs_fs_info *info)
2008 struct btrfs_path *path;
2010 struct btrfs_block_group *cache;
2011 struct btrfs_space_info *space_info;
2012 struct btrfs_key key;
2016 if (!info->extent_root)
2017 return fill_dummy_bgs(info);
2021 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
2022 path = btrfs_alloc_path();
2026 cache_gen = btrfs_super_cache_generation(info->super_copy);
2027 if (btrfs_test_opt(info, SPACE_CACHE) &&
2028 btrfs_super_generation(info->super_copy) != cache_gen)
2030 if (btrfs_test_opt(info, CLEAR_CACHE))
2034 struct btrfs_block_group_item bgi;
2035 struct extent_buffer *leaf;
2038 ret = find_first_block_group(info, path, &key);
2044 leaf = path->nodes[0];
2045 slot = path->slots[0];
2047 read_extent_buffer(leaf, &bgi, btrfs_item_ptr_offset(leaf, slot),
2050 btrfs_item_key_to_cpu(leaf, &key, slot);
2051 btrfs_release_path(path);
2052 ret = read_one_block_group(info, &bgi, &key, need_clear);
2055 key.objectid += key.offset;
2058 btrfs_release_path(path);
2060 list_for_each_entry(space_info, &info->space_info, list) {
2063 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
2064 if (list_empty(&space_info->block_groups[i]))
2066 cache = list_first_entry(&space_info->block_groups[i],
2067 struct btrfs_block_group,
2069 btrfs_sysfs_add_block_group_type(cache);
2072 if (!(btrfs_get_alloc_profile(info, space_info->flags) &
2073 (BTRFS_BLOCK_GROUP_RAID10 |
2074 BTRFS_BLOCK_GROUP_RAID1_MASK |
2075 BTRFS_BLOCK_GROUP_RAID56_MASK |
2076 BTRFS_BLOCK_GROUP_DUP)))
2079 * Avoid allocating from un-mirrored block group if there are
2080 * mirrored block groups.
2082 list_for_each_entry(cache,
2083 &space_info->block_groups[BTRFS_RAID_RAID0],
2085 inc_block_group_ro(cache, 1);
2086 list_for_each_entry(cache,
2087 &space_info->block_groups[BTRFS_RAID_SINGLE],
2089 inc_block_group_ro(cache, 1);
2092 btrfs_init_global_block_rsv(info);
2093 ret = check_chunk_block_group_mappings(info);
2095 btrfs_free_path(path);
2099 static int insert_block_group_item(struct btrfs_trans_handle *trans,
2100 struct btrfs_block_group *block_group)
2102 struct btrfs_fs_info *fs_info = trans->fs_info;
2103 struct btrfs_block_group_item bgi;
2104 struct btrfs_root *root;
2105 struct btrfs_key key;
2107 spin_lock(&block_group->lock);
2108 btrfs_set_stack_block_group_used(&bgi, block_group->used);
2109 btrfs_set_stack_block_group_chunk_objectid(&bgi,
2110 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
2111 btrfs_set_stack_block_group_flags(&bgi, block_group->flags);
2112 key.objectid = block_group->start;
2113 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
2114 key.offset = block_group->length;
2115 spin_unlock(&block_group->lock);
2117 root = fs_info->extent_root;
2118 return btrfs_insert_item(trans, root, &key, &bgi, sizeof(bgi));
2121 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans)
2123 struct btrfs_fs_info *fs_info = trans->fs_info;
2124 struct btrfs_block_group *block_group;
2127 if (!trans->can_flush_pending_bgs)
2130 while (!list_empty(&trans->new_bgs)) {
2133 block_group = list_first_entry(&trans->new_bgs,
2134 struct btrfs_block_group,
2139 index = btrfs_bg_flags_to_raid_index(block_group->flags);
2141 ret = insert_block_group_item(trans, block_group);
2143 btrfs_abort_transaction(trans, ret);
2144 ret = btrfs_finish_chunk_alloc(trans, block_group->start,
2145 block_group->length);
2147 btrfs_abort_transaction(trans, ret);
2148 add_block_group_free_space(trans, block_group);
2151 * If we restriped during balance, we may have added a new raid
2152 * type, so now add the sysfs entries when it is safe to do so.
2153 * We don't have to worry about locking here as it's handled in
2154 * btrfs_sysfs_add_block_group_type.
2156 if (block_group->space_info->block_group_kobjs[index] == NULL)
2157 btrfs_sysfs_add_block_group_type(block_group);
2159 /* Already aborted the transaction if it failed. */
2161 btrfs_delayed_refs_rsv_release(fs_info, 1);
2162 list_del_init(&block_group->bg_list);
2164 btrfs_trans_release_chunk_metadata(trans);
2167 int btrfs_make_block_group(struct btrfs_trans_handle *trans, u64 bytes_used,
2168 u64 type, u64 chunk_offset, u64 size)
2170 struct btrfs_fs_info *fs_info = trans->fs_info;
2171 struct btrfs_block_group *cache;
2174 btrfs_set_log_full_commit(trans);
2176 cache = btrfs_create_block_group_cache(fs_info, chunk_offset);
2180 cache->length = size;
2181 set_free_space_tree_thresholds(cache);
2182 cache->used = bytes_used;
2183 cache->flags = type;
2184 cache->last_byte_to_unpin = (u64)-1;
2185 cache->cached = BTRFS_CACHE_FINISHED;
2186 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
2187 cache->needs_free_space = 1;
2189 ret = btrfs_load_block_group_zone_info(cache, true);
2191 btrfs_put_block_group(cache);
2195 ret = exclude_super_stripes(cache);
2197 /* We may have excluded something, so call this just in case */
2198 btrfs_free_excluded_extents(cache);
2199 btrfs_put_block_group(cache);
2203 add_new_free_space(cache, chunk_offset, chunk_offset + size);
2205 btrfs_free_excluded_extents(cache);
2207 #ifdef CONFIG_BTRFS_DEBUG
2208 if (btrfs_should_fragment_free_space(cache)) {
2209 u64 new_bytes_used = size - bytes_used;
2211 bytes_used += new_bytes_used >> 1;
2212 fragment_free_space(cache);
2216 * Ensure the corresponding space_info object is created and
2217 * assigned to our block group. We want our bg to be added to the rbtree
2218 * with its ->space_info set.
2220 cache->space_info = btrfs_find_space_info(fs_info, cache->flags);
2221 ASSERT(cache->space_info);
2223 ret = btrfs_add_block_group_cache(fs_info, cache);
2225 btrfs_remove_free_space_cache(cache);
2226 btrfs_put_block_group(cache);
2231 * Now that our block group has its ->space_info set and is inserted in
2232 * the rbtree, update the space info's counters.
2234 trace_btrfs_add_block_group(fs_info, cache, 1);
2235 btrfs_update_space_info(fs_info, cache->flags, size, bytes_used,
2236 cache->bytes_super, 0, &cache->space_info);
2237 btrfs_update_global_block_rsv(fs_info);
2239 link_block_group(cache);
2241 list_add_tail(&cache->bg_list, &trans->new_bgs);
2242 trans->delayed_ref_updates++;
2243 btrfs_update_delayed_refs_rsv(trans);
2245 set_avail_alloc_bits(fs_info, type);
2250 * Mark one block group RO, can be called several times for the same block
2253 * @cache: the destination block group
2254 * @do_chunk_alloc: whether need to do chunk pre-allocation, this is to
2255 * ensure we still have some free space after marking this
2258 int btrfs_inc_block_group_ro(struct btrfs_block_group *cache,
2259 bool do_chunk_alloc)
2261 struct btrfs_fs_info *fs_info = cache->fs_info;
2262 struct btrfs_trans_handle *trans;
2267 trans = btrfs_join_transaction(fs_info->extent_root);
2269 return PTR_ERR(trans);
2272 * we're not allowed to set block groups readonly after the dirty
2273 * block groups cache has started writing. If it already started,
2274 * back off and let this transaction commit
2276 mutex_lock(&fs_info->ro_block_group_mutex);
2277 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
2278 u64 transid = trans->transid;
2280 mutex_unlock(&fs_info->ro_block_group_mutex);
2281 btrfs_end_transaction(trans);
2283 ret = btrfs_wait_for_commit(fs_info, transid);
2289 if (do_chunk_alloc) {
2291 * If we are changing raid levels, try to allocate a
2292 * corresponding block group with the new raid level.
2294 alloc_flags = btrfs_get_alloc_profile(fs_info, cache->flags);
2295 if (alloc_flags != cache->flags) {
2296 ret = btrfs_chunk_alloc(trans, alloc_flags,
2299 * ENOSPC is allowed here, we may have enough space
2300 * already allocated at the new raid level to carry on
2309 ret = inc_block_group_ro(cache, 0);
2310 if (!do_chunk_alloc)
2314 alloc_flags = btrfs_get_alloc_profile(fs_info, cache->space_info->flags);
2315 ret = btrfs_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE);
2318 ret = inc_block_group_ro(cache, 0);
2320 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
2321 alloc_flags = btrfs_get_alloc_profile(fs_info, cache->flags);
2322 mutex_lock(&fs_info->chunk_mutex);
2323 check_system_chunk(trans, alloc_flags);
2324 mutex_unlock(&fs_info->chunk_mutex);
2327 mutex_unlock(&fs_info->ro_block_group_mutex);
2329 btrfs_end_transaction(trans);
2333 void btrfs_dec_block_group_ro(struct btrfs_block_group *cache)
2335 struct btrfs_space_info *sinfo = cache->space_info;
2340 spin_lock(&sinfo->lock);
2341 spin_lock(&cache->lock);
2343 num_bytes = cache->length - cache->reserved -
2344 cache->pinned - cache->bytes_super -
2345 cache->zone_unusable - cache->used;
2346 sinfo->bytes_readonly -= num_bytes;
2347 if (btrfs_is_zoned(cache->fs_info)) {
2348 /* Migrate zone_unusable bytes back */
2349 cache->zone_unusable = cache->alloc_offset - cache->used;
2350 sinfo->bytes_zone_unusable += cache->zone_unusable;
2351 sinfo->bytes_readonly -= cache->zone_unusable;
2353 list_del_init(&cache->ro_list);
2355 spin_unlock(&cache->lock);
2356 spin_unlock(&sinfo->lock);
2359 static int update_block_group_item(struct btrfs_trans_handle *trans,
2360 struct btrfs_path *path,
2361 struct btrfs_block_group *cache)
2363 struct btrfs_fs_info *fs_info = trans->fs_info;
2365 struct btrfs_root *root = fs_info->extent_root;
2367 struct extent_buffer *leaf;
2368 struct btrfs_block_group_item bgi;
2369 struct btrfs_key key;
2371 key.objectid = cache->start;
2372 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
2373 key.offset = cache->length;
2375 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2382 leaf = path->nodes[0];
2383 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2384 btrfs_set_stack_block_group_used(&bgi, cache->used);
2385 btrfs_set_stack_block_group_chunk_objectid(&bgi,
2386 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
2387 btrfs_set_stack_block_group_flags(&bgi, cache->flags);
2388 write_extent_buffer(leaf, &bgi, bi, sizeof(bgi));
2389 btrfs_mark_buffer_dirty(leaf);
2391 btrfs_release_path(path);
2396 static int cache_save_setup(struct btrfs_block_group *block_group,
2397 struct btrfs_trans_handle *trans,
2398 struct btrfs_path *path)
2400 struct btrfs_fs_info *fs_info = block_group->fs_info;
2401 struct btrfs_root *root = fs_info->tree_root;
2402 struct inode *inode = NULL;
2403 struct extent_changeset *data_reserved = NULL;
2405 int dcs = BTRFS_DC_ERROR;
2410 if (!btrfs_test_opt(fs_info, SPACE_CACHE))
2414 * If this block group is smaller than 100 megs don't bother caching the
2417 if (block_group->length < (100 * SZ_1M)) {
2418 spin_lock(&block_group->lock);
2419 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2420 spin_unlock(&block_group->lock);
2424 if (TRANS_ABORTED(trans))
2427 inode = lookup_free_space_inode(block_group, path);
2428 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2429 ret = PTR_ERR(inode);
2430 btrfs_release_path(path);
2434 if (IS_ERR(inode)) {
2438 if (block_group->ro)
2441 ret = create_free_space_inode(trans, block_group, path);
2448 * We want to set the generation to 0, that way if anything goes wrong
2449 * from here on out we know not to trust this cache when we load up next
2452 BTRFS_I(inode)->generation = 0;
2453 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
2456 * So theoretically we could recover from this, simply set the
2457 * super cache generation to 0 so we know to invalidate the
2458 * cache, but then we'd have to keep track of the block groups
2459 * that fail this way so we know we _have_ to reset this cache
2460 * before the next commit or risk reading stale cache. So to
2461 * limit our exposure to horrible edge cases lets just abort the
2462 * transaction, this only happens in really bad situations
2465 btrfs_abort_transaction(trans, ret);
2470 /* We've already setup this transaction, go ahead and exit */
2471 if (block_group->cache_generation == trans->transid &&
2472 i_size_read(inode)) {
2473 dcs = BTRFS_DC_SETUP;
2477 if (i_size_read(inode) > 0) {
2478 ret = btrfs_check_trunc_cache_free_space(fs_info,
2479 &fs_info->global_block_rsv);
2483 ret = btrfs_truncate_free_space_cache(trans, NULL, inode);
2488 spin_lock(&block_group->lock);
2489 if (block_group->cached != BTRFS_CACHE_FINISHED ||
2490 !btrfs_test_opt(fs_info, SPACE_CACHE)) {
2492 * don't bother trying to write stuff out _if_
2493 * a) we're not cached,
2494 * b) we're with nospace_cache mount option,
2495 * c) we're with v2 space_cache (FREE_SPACE_TREE).
2497 dcs = BTRFS_DC_WRITTEN;
2498 spin_unlock(&block_group->lock);
2501 spin_unlock(&block_group->lock);
2504 * We hit an ENOSPC when setting up the cache in this transaction, just
2505 * skip doing the setup, we've already cleared the cache so we're safe.
2507 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
2513 * Try to preallocate enough space based on how big the block group is.
2514 * Keep in mind this has to include any pinned space which could end up
2515 * taking up quite a bit since it's not folded into the other space
2518 num_pages = div_u64(block_group->length, SZ_256M);
2523 num_pages *= PAGE_SIZE;
2525 ret = btrfs_check_data_free_space(BTRFS_I(inode), &data_reserved, 0,
2530 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2531 num_pages, num_pages,
2534 * Our cache requires contiguous chunks so that we don't modify a bunch
2535 * of metadata or split extents when writing the cache out, which means
2536 * we can enospc if we are heavily fragmented in addition to just normal
2537 * out of space conditions. So if we hit this just skip setting up any
2538 * other block groups for this transaction, maybe we'll unpin enough
2539 * space the next time around.
2542 dcs = BTRFS_DC_SETUP;
2543 else if (ret == -ENOSPC)
2544 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
2549 btrfs_release_path(path);
2551 spin_lock(&block_group->lock);
2552 if (!ret && dcs == BTRFS_DC_SETUP)
2553 block_group->cache_generation = trans->transid;
2554 block_group->disk_cache_state = dcs;
2555 spin_unlock(&block_group->lock);
2557 extent_changeset_free(data_reserved);
2561 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans)
2563 struct btrfs_fs_info *fs_info = trans->fs_info;
2564 struct btrfs_block_group *cache, *tmp;
2565 struct btrfs_transaction *cur_trans = trans->transaction;
2566 struct btrfs_path *path;
2568 if (list_empty(&cur_trans->dirty_bgs) ||
2569 !btrfs_test_opt(fs_info, SPACE_CACHE))
2572 path = btrfs_alloc_path();
2576 /* Could add new block groups, use _safe just in case */
2577 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
2579 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2580 cache_save_setup(cache, trans, path);
2583 btrfs_free_path(path);
2588 * Transaction commit does final block group cache writeback during a critical
2589 * section where nothing is allowed to change the FS. This is required in
2590 * order for the cache to actually match the block group, but can introduce a
2591 * lot of latency into the commit.
2593 * So, btrfs_start_dirty_block_groups is here to kick off block group cache IO.
2594 * There's a chance we'll have to redo some of it if the block group changes
2595 * again during the commit, but it greatly reduces the commit latency by
2596 * getting rid of the easy block groups while we're still allowing others to
2599 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans)
2601 struct btrfs_fs_info *fs_info = trans->fs_info;
2602 struct btrfs_block_group *cache;
2603 struct btrfs_transaction *cur_trans = trans->transaction;
2606 struct btrfs_path *path = NULL;
2608 struct list_head *io = &cur_trans->io_bgs;
2609 int num_started = 0;
2612 spin_lock(&cur_trans->dirty_bgs_lock);
2613 if (list_empty(&cur_trans->dirty_bgs)) {
2614 spin_unlock(&cur_trans->dirty_bgs_lock);
2617 list_splice_init(&cur_trans->dirty_bgs, &dirty);
2618 spin_unlock(&cur_trans->dirty_bgs_lock);
2621 /* Make sure all the block groups on our dirty list actually exist */
2622 btrfs_create_pending_block_groups(trans);
2625 path = btrfs_alloc_path();
2633 * cache_write_mutex is here only to save us from balance or automatic
2634 * removal of empty block groups deleting this block group while we are
2635 * writing out the cache
2637 mutex_lock(&trans->transaction->cache_write_mutex);
2638 while (!list_empty(&dirty)) {
2639 bool drop_reserve = true;
2641 cache = list_first_entry(&dirty, struct btrfs_block_group,
2644 * This can happen if something re-dirties a block group that
2645 * is already under IO. Just wait for it to finish and then do
2648 if (!list_empty(&cache->io_list)) {
2649 list_del_init(&cache->io_list);
2650 btrfs_wait_cache_io(trans, cache, path);
2651 btrfs_put_block_group(cache);
2656 * btrfs_wait_cache_io uses the cache->dirty_list to decide if
2657 * it should update the cache_state. Don't delete until after
2660 * Since we're not running in the commit critical section
2661 * we need the dirty_bgs_lock to protect from update_block_group
2663 spin_lock(&cur_trans->dirty_bgs_lock);
2664 list_del_init(&cache->dirty_list);
2665 spin_unlock(&cur_trans->dirty_bgs_lock);
2669 cache_save_setup(cache, trans, path);
2671 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
2672 cache->io_ctl.inode = NULL;
2673 ret = btrfs_write_out_cache(trans, cache, path);
2674 if (ret == 0 && cache->io_ctl.inode) {
2679 * The cache_write_mutex is protecting the
2680 * io_list, also refer to the definition of
2681 * btrfs_transaction::io_bgs for more details
2683 list_add_tail(&cache->io_list, io);
2686 * If we failed to write the cache, the
2687 * generation will be bad and life goes on
2693 ret = update_block_group_item(trans, path, cache);
2695 * Our block group might still be attached to the list
2696 * of new block groups in the transaction handle of some
2697 * other task (struct btrfs_trans_handle->new_bgs). This
2698 * means its block group item isn't yet in the extent
2699 * tree. If this happens ignore the error, as we will
2700 * try again later in the critical section of the
2701 * transaction commit.
2703 if (ret == -ENOENT) {
2705 spin_lock(&cur_trans->dirty_bgs_lock);
2706 if (list_empty(&cache->dirty_list)) {
2707 list_add_tail(&cache->dirty_list,
2708 &cur_trans->dirty_bgs);
2709 btrfs_get_block_group(cache);
2710 drop_reserve = false;
2712 spin_unlock(&cur_trans->dirty_bgs_lock);
2714 btrfs_abort_transaction(trans, ret);
2718 /* If it's not on the io list, we need to put the block group */
2720 btrfs_put_block_group(cache);
2722 btrfs_delayed_refs_rsv_release(fs_info, 1);
2724 * Avoid blocking other tasks for too long. It might even save
2725 * us from writing caches for block groups that are going to be
2728 mutex_unlock(&trans->transaction->cache_write_mutex);
2731 mutex_lock(&trans->transaction->cache_write_mutex);
2733 mutex_unlock(&trans->transaction->cache_write_mutex);
2736 * Go through delayed refs for all the stuff we've just kicked off
2737 * and then loop back (just once)
2740 ret = btrfs_run_delayed_refs(trans, 0);
2741 if (!ret && loops == 0) {
2743 spin_lock(&cur_trans->dirty_bgs_lock);
2744 list_splice_init(&cur_trans->dirty_bgs, &dirty);
2746 * dirty_bgs_lock protects us from concurrent block group
2747 * deletes too (not just cache_write_mutex).
2749 if (!list_empty(&dirty)) {
2750 spin_unlock(&cur_trans->dirty_bgs_lock);
2753 spin_unlock(&cur_trans->dirty_bgs_lock);
2757 spin_lock(&cur_trans->dirty_bgs_lock);
2758 list_splice_init(&dirty, &cur_trans->dirty_bgs);
2759 spin_unlock(&cur_trans->dirty_bgs_lock);
2760 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
2763 btrfs_free_path(path);
2767 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans)
2769 struct btrfs_fs_info *fs_info = trans->fs_info;
2770 struct btrfs_block_group *cache;
2771 struct btrfs_transaction *cur_trans = trans->transaction;
2774 struct btrfs_path *path;
2775 struct list_head *io = &cur_trans->io_bgs;
2776 int num_started = 0;
2778 path = btrfs_alloc_path();
2783 * Even though we are in the critical section of the transaction commit,
2784 * we can still have concurrent tasks adding elements to this
2785 * transaction's list of dirty block groups. These tasks correspond to
2786 * endio free space workers started when writeback finishes for a
2787 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
2788 * allocate new block groups as a result of COWing nodes of the root
2789 * tree when updating the free space inode. The writeback for the space
2790 * caches is triggered by an earlier call to
2791 * btrfs_start_dirty_block_groups() and iterations of the following
2793 * Also we want to do the cache_save_setup first and then run the
2794 * delayed refs to make sure we have the best chance at doing this all
2797 spin_lock(&cur_trans->dirty_bgs_lock);
2798 while (!list_empty(&cur_trans->dirty_bgs)) {
2799 cache = list_first_entry(&cur_trans->dirty_bgs,
2800 struct btrfs_block_group,
2804 * This can happen if cache_save_setup re-dirties a block group
2805 * that is already under IO. Just wait for it to finish and
2806 * then do it all again
2808 if (!list_empty(&cache->io_list)) {
2809 spin_unlock(&cur_trans->dirty_bgs_lock);
2810 list_del_init(&cache->io_list);
2811 btrfs_wait_cache_io(trans, cache, path);
2812 btrfs_put_block_group(cache);
2813 spin_lock(&cur_trans->dirty_bgs_lock);
2817 * Don't remove from the dirty list until after we've waited on
2820 list_del_init(&cache->dirty_list);
2821 spin_unlock(&cur_trans->dirty_bgs_lock);
2824 cache_save_setup(cache, trans, path);
2827 ret = btrfs_run_delayed_refs(trans,
2828 (unsigned long) -1);
2830 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
2831 cache->io_ctl.inode = NULL;
2832 ret = btrfs_write_out_cache(trans, cache, path);
2833 if (ret == 0 && cache->io_ctl.inode) {
2836 list_add_tail(&cache->io_list, io);
2839 * If we failed to write the cache, the
2840 * generation will be bad and life goes on
2846 ret = update_block_group_item(trans, path, cache);
2848 * One of the free space endio workers might have
2849 * created a new block group while updating a free space
2850 * cache's inode (at inode.c:btrfs_finish_ordered_io())
2851 * and hasn't released its transaction handle yet, in
2852 * which case the new block group is still attached to
2853 * its transaction handle and its creation has not
2854 * finished yet (no block group item in the extent tree
2855 * yet, etc). If this is the case, wait for all free
2856 * space endio workers to finish and retry. This is a
2857 * very rare case so no need for a more efficient and
2860 if (ret == -ENOENT) {
2861 wait_event(cur_trans->writer_wait,
2862 atomic_read(&cur_trans->num_writers) == 1);
2863 ret = update_block_group_item(trans, path, cache);
2866 btrfs_abort_transaction(trans, ret);
2869 /* If its not on the io list, we need to put the block group */
2871 btrfs_put_block_group(cache);
2872 btrfs_delayed_refs_rsv_release(fs_info, 1);
2873 spin_lock(&cur_trans->dirty_bgs_lock);
2875 spin_unlock(&cur_trans->dirty_bgs_lock);
2878 * Refer to the definition of io_bgs member for details why it's safe
2879 * to use it without any locking
2881 while (!list_empty(io)) {
2882 cache = list_first_entry(io, struct btrfs_block_group,
2884 list_del_init(&cache->io_list);
2885 btrfs_wait_cache_io(trans, cache, path);
2886 btrfs_put_block_group(cache);
2889 btrfs_free_path(path);
2893 int btrfs_update_block_group(struct btrfs_trans_handle *trans,
2894 u64 bytenr, u64 num_bytes, int alloc)
2896 struct btrfs_fs_info *info = trans->fs_info;
2897 struct btrfs_block_group *cache = NULL;
2898 u64 total = num_bytes;
2904 /* Block accounting for super block */
2905 spin_lock(&info->delalloc_root_lock);
2906 old_val = btrfs_super_bytes_used(info->super_copy);
2908 old_val += num_bytes;
2910 old_val -= num_bytes;
2911 btrfs_set_super_bytes_used(info->super_copy, old_val);
2912 spin_unlock(&info->delalloc_root_lock);
2915 cache = btrfs_lookup_block_group(info, bytenr);
2920 factor = btrfs_bg_type_to_factor(cache->flags);
2923 * If this block group has free space cache written out, we
2924 * need to make sure to load it if we are removing space. This
2925 * is because we need the unpinning stage to actually add the
2926 * space back to the block group, otherwise we will leak space.
2928 if (!alloc && !btrfs_block_group_done(cache))
2929 btrfs_cache_block_group(cache, 1);
2931 byte_in_group = bytenr - cache->start;
2932 WARN_ON(byte_in_group > cache->length);
2934 spin_lock(&cache->space_info->lock);
2935 spin_lock(&cache->lock);
2937 if (btrfs_test_opt(info, SPACE_CACHE) &&
2938 cache->disk_cache_state < BTRFS_DC_CLEAR)
2939 cache->disk_cache_state = BTRFS_DC_CLEAR;
2941 old_val = cache->used;
2942 num_bytes = min(total, cache->length - byte_in_group);
2944 old_val += num_bytes;
2945 cache->used = old_val;
2946 cache->reserved -= num_bytes;
2947 cache->space_info->bytes_reserved -= num_bytes;
2948 cache->space_info->bytes_used += num_bytes;
2949 cache->space_info->disk_used += num_bytes * factor;
2950 spin_unlock(&cache->lock);
2951 spin_unlock(&cache->space_info->lock);
2953 old_val -= num_bytes;
2954 cache->used = old_val;
2955 cache->pinned += num_bytes;
2956 btrfs_space_info_update_bytes_pinned(info,
2957 cache->space_info, num_bytes);
2958 cache->space_info->bytes_used -= num_bytes;
2959 cache->space_info->disk_used -= num_bytes * factor;
2960 spin_unlock(&cache->lock);
2961 spin_unlock(&cache->space_info->lock);
2963 __btrfs_mod_total_bytes_pinned(cache->space_info,
2965 set_extent_dirty(&trans->transaction->pinned_extents,
2966 bytenr, bytenr + num_bytes - 1,
2967 GFP_NOFS | __GFP_NOFAIL);
2970 spin_lock(&trans->transaction->dirty_bgs_lock);
2971 if (list_empty(&cache->dirty_list)) {
2972 list_add_tail(&cache->dirty_list,
2973 &trans->transaction->dirty_bgs);
2974 trans->delayed_ref_updates++;
2975 btrfs_get_block_group(cache);
2977 spin_unlock(&trans->transaction->dirty_bgs_lock);
2980 * No longer have used bytes in this block group, queue it for
2981 * deletion. We do this after adding the block group to the
2982 * dirty list to avoid races between cleaner kthread and space
2985 if (!alloc && old_val == 0) {
2986 if (!btrfs_test_opt(info, DISCARD_ASYNC))
2987 btrfs_mark_bg_unused(cache);
2990 btrfs_put_block_group(cache);
2992 bytenr += num_bytes;
2995 /* Modified block groups are accounted for in the delayed_refs_rsv. */
2996 btrfs_update_delayed_refs_rsv(trans);
3001 * btrfs_add_reserved_bytes - update the block_group and space info counters
3002 * @cache: The cache we are manipulating
3003 * @ram_bytes: The number of bytes of file content, and will be same to
3004 * @num_bytes except for the compress path.
3005 * @num_bytes: The number of bytes in question
3006 * @delalloc: The blocks are allocated for the delalloc write
3008 * This is called by the allocator when it reserves space. If this is a
3009 * reservation and the block group has become read only we cannot make the
3010 * reservation and return -EAGAIN, otherwise this function always succeeds.
3012 int btrfs_add_reserved_bytes(struct btrfs_block_group *cache,
3013 u64 ram_bytes, u64 num_bytes, int delalloc)
3015 struct btrfs_space_info *space_info = cache->space_info;
3018 spin_lock(&space_info->lock);
3019 spin_lock(&cache->lock);
3023 cache->reserved += num_bytes;
3024 space_info->bytes_reserved += num_bytes;
3025 trace_btrfs_space_reservation(cache->fs_info, "space_info",
3026 space_info->flags, num_bytes, 1);
3027 btrfs_space_info_update_bytes_may_use(cache->fs_info,
3028 space_info, -ram_bytes);
3030 cache->delalloc_bytes += num_bytes;
3033 * Compression can use less space than we reserved, so wake
3034 * tickets if that happens
3036 if (num_bytes < ram_bytes)
3037 btrfs_try_granting_tickets(cache->fs_info, space_info);
3039 spin_unlock(&cache->lock);
3040 spin_unlock(&space_info->lock);
3045 * btrfs_free_reserved_bytes - update the block_group and space info counters
3046 * @cache: The cache we are manipulating
3047 * @num_bytes: The number of bytes in question
3048 * @delalloc: The blocks are allocated for the delalloc write
3050 * This is called by somebody who is freeing space that was never actually used
3051 * on disk. For example if you reserve some space for a new leaf in transaction
3052 * A and before transaction A commits you free that leaf, you call this with
3053 * reserve set to 0 in order to clear the reservation.
3055 void btrfs_free_reserved_bytes(struct btrfs_block_group *cache,
3056 u64 num_bytes, int delalloc)
3058 struct btrfs_space_info *space_info = cache->space_info;
3060 spin_lock(&space_info->lock);
3061 spin_lock(&cache->lock);
3063 space_info->bytes_readonly += num_bytes;
3064 cache->reserved -= num_bytes;
3065 space_info->bytes_reserved -= num_bytes;
3066 space_info->max_extent_size = 0;
3069 cache->delalloc_bytes -= num_bytes;
3070 spin_unlock(&cache->lock);
3072 btrfs_try_granting_tickets(cache->fs_info, space_info);
3073 spin_unlock(&space_info->lock);
3076 static void force_metadata_allocation(struct btrfs_fs_info *info)
3078 struct list_head *head = &info->space_info;
3079 struct btrfs_space_info *found;
3081 list_for_each_entry(found, head, list) {
3082 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3083 found->force_alloc = CHUNK_ALLOC_FORCE;
3087 static int should_alloc_chunk(struct btrfs_fs_info *fs_info,
3088 struct btrfs_space_info *sinfo, int force)
3090 u64 bytes_used = btrfs_space_info_used(sinfo, false);
3093 if (force == CHUNK_ALLOC_FORCE)
3097 * in limited mode, we want to have some free space up to
3098 * about 1% of the FS size.
3100 if (force == CHUNK_ALLOC_LIMITED) {
3101 thresh = btrfs_super_total_bytes(fs_info->super_copy);
3102 thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1));
3104 if (sinfo->total_bytes - bytes_used < thresh)
3108 if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8))
3113 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, u64 type)
3115 u64 alloc_flags = btrfs_get_alloc_profile(trans->fs_info, type);
3117 return btrfs_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE);
3121 * If force is CHUNK_ALLOC_FORCE:
3122 * - return 1 if it successfully allocates a chunk,
3123 * - return errors including -ENOSPC otherwise.
3124 * If force is NOT CHUNK_ALLOC_FORCE:
3125 * - return 0 if it doesn't need to allocate a new chunk,
3126 * - return 1 if it successfully allocates a chunk,
3127 * - return errors including -ENOSPC otherwise.
3129 int btrfs_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags,
3130 enum btrfs_chunk_alloc_enum force)
3132 struct btrfs_fs_info *fs_info = trans->fs_info;
3133 struct btrfs_space_info *space_info;
3134 bool wait_for_alloc = false;
3135 bool should_alloc = false;
3138 /* Don't re-enter if we're already allocating a chunk */
3139 if (trans->allocating_chunk)
3142 space_info = btrfs_find_space_info(fs_info, flags);
3146 spin_lock(&space_info->lock);
3147 if (force < space_info->force_alloc)
3148 force = space_info->force_alloc;
3149 should_alloc = should_alloc_chunk(fs_info, space_info, force);
3150 if (space_info->full) {
3151 /* No more free physical space */
3156 spin_unlock(&space_info->lock);
3158 } else if (!should_alloc) {
3159 spin_unlock(&space_info->lock);
3161 } else if (space_info->chunk_alloc) {
3163 * Someone is already allocating, so we need to block
3164 * until this someone is finished and then loop to
3165 * recheck if we should continue with our allocation
3168 wait_for_alloc = true;
3169 spin_unlock(&space_info->lock);
3170 mutex_lock(&fs_info->chunk_mutex);
3171 mutex_unlock(&fs_info->chunk_mutex);
3173 /* Proceed with allocation */
3174 space_info->chunk_alloc = 1;
3175 wait_for_alloc = false;
3176 spin_unlock(&space_info->lock);
3180 } while (wait_for_alloc);
3182 mutex_lock(&fs_info->chunk_mutex);
3183 trans->allocating_chunk = true;
3186 * If we have mixed data/metadata chunks we want to make sure we keep
3187 * allocating mixed chunks instead of individual chunks.
3189 if (btrfs_mixed_space_info(space_info))
3190 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3193 * if we're doing a data chunk, go ahead and make sure that
3194 * we keep a reasonable number of metadata chunks allocated in the
3197 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3198 fs_info->data_chunk_allocations++;
3199 if (!(fs_info->data_chunk_allocations %
3200 fs_info->metadata_ratio))
3201 force_metadata_allocation(fs_info);
3205 * Check if we have enough space in SYSTEM chunk because we may need
3206 * to update devices.
3208 check_system_chunk(trans, flags);
3210 ret = btrfs_alloc_chunk(trans, flags);
3211 trans->allocating_chunk = false;
3213 spin_lock(&space_info->lock);
3216 space_info->full = 1;
3221 space_info->max_extent_size = 0;
3224 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3226 space_info->chunk_alloc = 0;
3227 spin_unlock(&space_info->lock);
3228 mutex_unlock(&fs_info->chunk_mutex);
3230 * When we allocate a new chunk we reserve space in the chunk block
3231 * reserve to make sure we can COW nodes/leafs in the chunk tree or
3232 * add new nodes/leafs to it if we end up needing to do it when
3233 * inserting the chunk item and updating device items as part of the
3234 * second phase of chunk allocation, performed by
3235 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
3236 * large number of new block groups to create in our transaction
3237 * handle's new_bgs list to avoid exhausting the chunk block reserve
3238 * in extreme cases - like having a single transaction create many new
3239 * block groups when starting to write out the free space caches of all
3240 * the block groups that were made dirty during the lifetime of the
3243 if (trans->chunk_bytes_reserved >= (u64)SZ_2M)
3244 btrfs_create_pending_block_groups(trans);
3249 static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type)
3253 num_dev = btrfs_raid_array[btrfs_bg_flags_to_raid_index(type)].devs_max;
3255 num_dev = fs_info->fs_devices->rw_devices;
3261 * Reserve space in the system space for allocating or removing a chunk
3263 void check_system_chunk(struct btrfs_trans_handle *trans, u64 type)
3265 struct btrfs_fs_info *fs_info = trans->fs_info;
3266 struct btrfs_space_info *info;
3273 * Needed because we can end up allocating a system chunk and for an
3274 * atomic and race free space reservation in the chunk block reserve.
3276 lockdep_assert_held(&fs_info->chunk_mutex);
3278 info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3279 spin_lock(&info->lock);
3280 left = info->total_bytes - btrfs_space_info_used(info, true);
3281 spin_unlock(&info->lock);
3283 num_devs = get_profile_num_devs(fs_info, type);
3285 /* num_devs device items to update and 1 chunk item to add or remove */
3286 thresh = btrfs_calc_metadata_size(fs_info, num_devs) +
3287 btrfs_calc_insert_metadata_size(fs_info, 1);
3289 if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
3290 btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu",
3291 left, thresh, type);
3292 btrfs_dump_space_info(fs_info, info, 0, 0);
3295 if (left < thresh) {
3296 u64 flags = btrfs_system_alloc_profile(fs_info);
3299 * Ignore failure to create system chunk. We might end up not
3300 * needing it, as we might not need to COW all nodes/leafs from
3301 * the paths we visit in the chunk tree (they were already COWed
3302 * or created in the current transaction for example).
3304 ret = btrfs_alloc_chunk(trans, flags);
3308 ret = btrfs_block_rsv_add(fs_info->chunk_root,
3309 &fs_info->chunk_block_rsv,
3310 thresh, BTRFS_RESERVE_NO_FLUSH);
3312 trans->chunk_bytes_reserved += thresh;
3316 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
3318 struct btrfs_block_group *block_group;
3322 struct inode *inode;
3324 block_group = btrfs_lookup_first_block_group(info, last);
3325 while (block_group) {
3326 btrfs_wait_block_group_cache_done(block_group);
3327 spin_lock(&block_group->lock);
3328 if (block_group->iref)
3330 spin_unlock(&block_group->lock);
3331 block_group = btrfs_next_block_group(block_group);
3340 inode = block_group->inode;
3341 block_group->iref = 0;
3342 block_group->inode = NULL;
3343 spin_unlock(&block_group->lock);
3344 ASSERT(block_group->io_ctl.inode == NULL);
3346 last = block_group->start + block_group->length;
3347 btrfs_put_block_group(block_group);
3352 * Must be called only after stopping all workers, since we could have block
3353 * group caching kthreads running, and therefore they could race with us if we
3354 * freed the block groups before stopping them.
3356 int btrfs_free_block_groups(struct btrfs_fs_info *info)
3358 struct btrfs_block_group *block_group;
3359 struct btrfs_space_info *space_info;
3360 struct btrfs_caching_control *caching_ctl;
3363 spin_lock(&info->block_group_cache_lock);
3364 while (!list_empty(&info->caching_block_groups)) {
3365 caching_ctl = list_entry(info->caching_block_groups.next,
3366 struct btrfs_caching_control, list);
3367 list_del(&caching_ctl->list);
3368 btrfs_put_caching_control(caching_ctl);
3370 spin_unlock(&info->block_group_cache_lock);
3372 spin_lock(&info->unused_bgs_lock);
3373 while (!list_empty(&info->unused_bgs)) {
3374 block_group = list_first_entry(&info->unused_bgs,
3375 struct btrfs_block_group,
3377 list_del_init(&block_group->bg_list);
3378 btrfs_put_block_group(block_group);
3380 spin_unlock(&info->unused_bgs_lock);
3382 spin_lock(&info->block_group_cache_lock);
3383 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
3384 block_group = rb_entry(n, struct btrfs_block_group,
3386 rb_erase(&block_group->cache_node,
3387 &info->block_group_cache_tree);
3388 RB_CLEAR_NODE(&block_group->cache_node);
3389 spin_unlock(&info->block_group_cache_lock);
3391 down_write(&block_group->space_info->groups_sem);
3392 list_del(&block_group->list);
3393 up_write(&block_group->space_info->groups_sem);
3396 * We haven't cached this block group, which means we could
3397 * possibly have excluded extents on this block group.
3399 if (block_group->cached == BTRFS_CACHE_NO ||
3400 block_group->cached == BTRFS_CACHE_ERROR)
3401 btrfs_free_excluded_extents(block_group);
3403 btrfs_remove_free_space_cache(block_group);
3404 ASSERT(block_group->cached != BTRFS_CACHE_STARTED);
3405 ASSERT(list_empty(&block_group->dirty_list));
3406 ASSERT(list_empty(&block_group->io_list));
3407 ASSERT(list_empty(&block_group->bg_list));
3408 ASSERT(refcount_read(&block_group->refs) == 1);
3409 btrfs_put_block_group(block_group);
3411 spin_lock(&info->block_group_cache_lock);
3413 spin_unlock(&info->block_group_cache_lock);
3415 btrfs_release_global_block_rsv(info);
3417 while (!list_empty(&info->space_info)) {
3418 space_info = list_entry(info->space_info.next,
3419 struct btrfs_space_info,
3423 * Do not hide this behind enospc_debug, this is actually
3424 * important and indicates a real bug if this happens.
3426 if (WARN_ON(space_info->bytes_pinned > 0 ||
3427 space_info->bytes_reserved > 0 ||
3428 space_info->bytes_may_use > 0))
3429 btrfs_dump_space_info(info, space_info, 0, 0);
3430 WARN_ON(space_info->reclaim_size > 0);
3431 list_del(&space_info->list);
3432 btrfs_sysfs_remove_space_info(space_info);
3437 void btrfs_freeze_block_group(struct btrfs_block_group *cache)
3439 atomic_inc(&cache->frozen);
3442 void btrfs_unfreeze_block_group(struct btrfs_block_group *block_group)
3444 struct btrfs_fs_info *fs_info = block_group->fs_info;
3445 struct extent_map_tree *em_tree;
3446 struct extent_map *em;
3449 spin_lock(&block_group->lock);
3450 cleanup = (atomic_dec_and_test(&block_group->frozen) &&
3451 block_group->removed);
3452 spin_unlock(&block_group->lock);
3455 em_tree = &fs_info->mapping_tree;
3456 write_lock(&em_tree->lock);
3457 em = lookup_extent_mapping(em_tree, block_group->start,
3459 BUG_ON(!em); /* logic error, can't happen */
3460 remove_extent_mapping(em_tree, em);
3461 write_unlock(&em_tree->lock);
3463 /* once for us and once for the tree */
3464 free_extent_map(em);
3465 free_extent_map(em);
3468 * We may have left one free space entry and other possible
3469 * tasks trimming this block group have left 1 entry each one.
3472 __btrfs_remove_free_space_cache(block_group->free_space_ctl);