1 // SPDX-License-Identifier: GPL-2.0
3 #include <linux/list_sort.h>
6 #include "block-group.h"
7 #include "space-info.h"
9 #include "free-space-cache.h"
10 #include "free-space-tree.h"
12 #include "transaction.h"
13 #include "ref-verify.h"
16 #include "delalloc-space.h"
22 * Return target flags in extended format or 0 if restripe for this chunk_type
25 * Should be called with balance_lock held
27 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
29 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
35 if (flags & BTRFS_BLOCK_GROUP_DATA &&
36 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
37 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
38 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
39 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
40 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
41 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
42 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
43 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
50 * @flags: available profiles in extended format (see ctree.h)
52 * Return reduced profile in chunk format. If profile changing is in progress
53 * (either running or paused) picks the target profile (if it's already
54 * available), otherwise falls back to plain reducing.
56 static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags)
58 u64 num_devices = fs_info->fs_devices->rw_devices;
64 * See if restripe for this chunk_type is in progress, if so try to
65 * reduce to the target profile
67 spin_lock(&fs_info->balance_lock);
68 target = get_restripe_target(fs_info, flags);
70 spin_unlock(&fs_info->balance_lock);
71 return extended_to_chunk(target);
73 spin_unlock(&fs_info->balance_lock);
75 /* First, mask out the RAID levels which aren't possible */
76 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
77 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
78 allowed |= btrfs_raid_array[raid_type].bg_flag;
82 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
83 allowed = BTRFS_BLOCK_GROUP_RAID6;
84 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
85 allowed = BTRFS_BLOCK_GROUP_RAID5;
86 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
87 allowed = BTRFS_BLOCK_GROUP_RAID10;
88 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
89 allowed = BTRFS_BLOCK_GROUP_RAID1;
90 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
91 allowed = BTRFS_BLOCK_GROUP_RAID0;
93 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
95 return extended_to_chunk(flags | allowed);
98 u64 btrfs_get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags)
105 seq = read_seqbegin(&fs_info->profiles_lock);
107 if (flags & BTRFS_BLOCK_GROUP_DATA)
108 flags |= fs_info->avail_data_alloc_bits;
109 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
110 flags |= fs_info->avail_system_alloc_bits;
111 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
112 flags |= fs_info->avail_metadata_alloc_bits;
113 } while (read_seqretry(&fs_info->profiles_lock, seq));
115 return btrfs_reduce_alloc_profile(fs_info, flags);
118 void btrfs_get_block_group(struct btrfs_block_group *cache)
120 refcount_inc(&cache->refs);
123 void btrfs_put_block_group(struct btrfs_block_group *cache)
125 if (refcount_dec_and_test(&cache->refs)) {
126 WARN_ON(cache->pinned > 0);
128 * If there was a failure to cleanup a log tree, very likely due
129 * to an IO failure on a writeback attempt of one or more of its
130 * extent buffers, we could not do proper (and cheap) unaccounting
131 * of their reserved space, so don't warn on reserved > 0 in that
134 if (!(cache->flags & BTRFS_BLOCK_GROUP_METADATA) ||
135 !BTRFS_FS_LOG_CLEANUP_ERROR(cache->fs_info))
136 WARN_ON(cache->reserved > 0);
139 * A block_group shouldn't be on the discard_list anymore.
140 * Remove the block_group from the discard_list to prevent us
141 * from causing a panic due to NULL pointer dereference.
143 if (WARN_ON(!list_empty(&cache->discard_list)))
144 btrfs_discard_cancel_work(&cache->fs_info->discard_ctl,
148 * If not empty, someone is still holding mutex of
149 * full_stripe_lock, which can only be released by caller.
150 * And it will definitely cause use-after-free when caller
151 * tries to release full stripe lock.
153 * No better way to resolve, but only to warn.
155 WARN_ON(!RB_EMPTY_ROOT(&cache->full_stripe_locks_root.root));
156 kfree(cache->free_space_ctl);
157 kfree(cache->physical_map);
163 * This adds the block group to the fs_info rb tree for the block group cache
165 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
166 struct btrfs_block_group *block_group)
169 struct rb_node *parent = NULL;
170 struct btrfs_block_group *cache;
171 bool leftmost = true;
173 ASSERT(block_group->length != 0);
175 write_lock(&info->block_group_cache_lock);
176 p = &info->block_group_cache_tree.rb_root.rb_node;
180 cache = rb_entry(parent, struct btrfs_block_group, cache_node);
181 if (block_group->start < cache->start) {
183 } else if (block_group->start > cache->start) {
187 write_unlock(&info->block_group_cache_lock);
192 rb_link_node(&block_group->cache_node, parent, p);
193 rb_insert_color_cached(&block_group->cache_node,
194 &info->block_group_cache_tree, leftmost);
196 write_unlock(&info->block_group_cache_lock);
202 * This will return the block group at or after bytenr if contains is 0, else
203 * it will return the block group that contains the bytenr
205 static struct btrfs_block_group *block_group_cache_tree_search(
206 struct btrfs_fs_info *info, u64 bytenr, int contains)
208 struct btrfs_block_group *cache, *ret = NULL;
212 read_lock(&info->block_group_cache_lock);
213 n = info->block_group_cache_tree.rb_root.rb_node;
216 cache = rb_entry(n, struct btrfs_block_group, cache_node);
217 end = cache->start + cache->length - 1;
218 start = cache->start;
220 if (bytenr < start) {
221 if (!contains && (!ret || start < ret->start))
224 } else if (bytenr > start) {
225 if (contains && bytenr <= end) {
236 btrfs_get_block_group(ret);
237 read_unlock(&info->block_group_cache_lock);
243 * Return the block group that starts at or after bytenr
245 struct btrfs_block_group *btrfs_lookup_first_block_group(
246 struct btrfs_fs_info *info, u64 bytenr)
248 return block_group_cache_tree_search(info, bytenr, 0);
252 * Return the block group that contains the given bytenr
254 struct btrfs_block_group *btrfs_lookup_block_group(
255 struct btrfs_fs_info *info, u64 bytenr)
257 return block_group_cache_tree_search(info, bytenr, 1);
260 struct btrfs_block_group *btrfs_next_block_group(
261 struct btrfs_block_group *cache)
263 struct btrfs_fs_info *fs_info = cache->fs_info;
264 struct rb_node *node;
266 read_lock(&fs_info->block_group_cache_lock);
268 /* If our block group was removed, we need a full search. */
269 if (RB_EMPTY_NODE(&cache->cache_node)) {
270 const u64 next_bytenr = cache->start + cache->length;
272 read_unlock(&fs_info->block_group_cache_lock);
273 btrfs_put_block_group(cache);
274 return btrfs_lookup_first_block_group(fs_info, next_bytenr);
276 node = rb_next(&cache->cache_node);
277 btrfs_put_block_group(cache);
279 cache = rb_entry(node, struct btrfs_block_group, cache_node);
280 btrfs_get_block_group(cache);
283 read_unlock(&fs_info->block_group_cache_lock);
288 * Check if we can do a NOCOW write for a given extent.
290 * @fs_info: The filesystem information object.
291 * @bytenr: Logical start address of the extent.
293 * Check if we can do a NOCOW write for the given extent, and increments the
294 * number of NOCOW writers in the block group that contains the extent, as long
295 * as the block group exists and it's currently not in read-only mode.
297 * Returns: A non-NULL block group pointer if we can do a NOCOW write, the caller
298 * is responsible for calling btrfs_dec_nocow_writers() later.
300 * Or NULL if we can not do a NOCOW write
302 struct btrfs_block_group *btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info,
305 struct btrfs_block_group *bg;
306 bool can_nocow = true;
308 bg = btrfs_lookup_block_group(fs_info, bytenr);
312 spin_lock(&bg->lock);
316 atomic_inc(&bg->nocow_writers);
317 spin_unlock(&bg->lock);
320 btrfs_put_block_group(bg);
324 /* No put on block group, done by btrfs_dec_nocow_writers(). */
329 * Decrement the number of NOCOW writers in a block group.
331 * @bg: The block group.
333 * This is meant to be called after a previous call to btrfs_inc_nocow_writers(),
334 * and on the block group returned by that call. Typically this is called after
335 * creating an ordered extent for a NOCOW write, to prevent races with scrub and
338 * After this call, the caller should not use the block group anymore. It it wants
339 * to use it, then it should get a reference on it before calling this function.
341 void btrfs_dec_nocow_writers(struct btrfs_block_group *bg)
343 if (atomic_dec_and_test(&bg->nocow_writers))
344 wake_up_var(&bg->nocow_writers);
346 /* For the lookup done by a previous call to btrfs_inc_nocow_writers(). */
347 btrfs_put_block_group(bg);
350 void btrfs_wait_nocow_writers(struct btrfs_block_group *bg)
352 wait_var_event(&bg->nocow_writers, !atomic_read(&bg->nocow_writers));
355 void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info,
358 struct btrfs_block_group *bg;
360 bg = btrfs_lookup_block_group(fs_info, start);
362 if (atomic_dec_and_test(&bg->reservations))
363 wake_up_var(&bg->reservations);
364 btrfs_put_block_group(bg);
367 void btrfs_wait_block_group_reservations(struct btrfs_block_group *bg)
369 struct btrfs_space_info *space_info = bg->space_info;
373 if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA))
377 * Our block group is read only but before we set it to read only,
378 * some task might have had allocated an extent from it already, but it
379 * has not yet created a respective ordered extent (and added it to a
380 * root's list of ordered extents).
381 * Therefore wait for any task currently allocating extents, since the
382 * block group's reservations counter is incremented while a read lock
383 * on the groups' semaphore is held and decremented after releasing
384 * the read access on that semaphore and creating the ordered extent.
386 down_write(&space_info->groups_sem);
387 up_write(&space_info->groups_sem);
389 wait_var_event(&bg->reservations, !atomic_read(&bg->reservations));
392 struct btrfs_caching_control *btrfs_get_caching_control(
393 struct btrfs_block_group *cache)
395 struct btrfs_caching_control *ctl;
397 spin_lock(&cache->lock);
398 if (!cache->caching_ctl) {
399 spin_unlock(&cache->lock);
403 ctl = cache->caching_ctl;
404 refcount_inc(&ctl->count);
405 spin_unlock(&cache->lock);
409 void btrfs_put_caching_control(struct btrfs_caching_control *ctl)
411 if (refcount_dec_and_test(&ctl->count))
416 * When we wait for progress in the block group caching, its because our
417 * allocation attempt failed at least once. So, we must sleep and let some
418 * progress happen before we try again.
420 * This function will sleep at least once waiting for new free space to show
421 * up, and then it will check the block group free space numbers for our min
422 * num_bytes. Another option is to have it go ahead and look in the rbtree for
423 * a free extent of a given size, but this is a good start.
425 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
426 * any of the information in this block group.
428 void btrfs_wait_block_group_cache_progress(struct btrfs_block_group *cache,
431 struct btrfs_caching_control *caching_ctl;
433 caching_ctl = btrfs_get_caching_control(cache);
437 wait_event(caching_ctl->wait, btrfs_block_group_done(cache) ||
438 (cache->free_space_ctl->free_space >= num_bytes));
440 btrfs_put_caching_control(caching_ctl);
443 static int btrfs_caching_ctl_wait_done(struct btrfs_block_group *cache,
444 struct btrfs_caching_control *caching_ctl)
446 wait_event(caching_ctl->wait, btrfs_block_group_done(cache));
447 return cache->cached == BTRFS_CACHE_ERROR ? -EIO : 0;
450 static int btrfs_wait_block_group_cache_done(struct btrfs_block_group *cache)
452 struct btrfs_caching_control *caching_ctl;
455 caching_ctl = btrfs_get_caching_control(cache);
457 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
458 ret = btrfs_caching_ctl_wait_done(cache, caching_ctl);
459 btrfs_put_caching_control(caching_ctl);
463 #ifdef CONFIG_BTRFS_DEBUG
464 static void fragment_free_space(struct btrfs_block_group *block_group)
466 struct btrfs_fs_info *fs_info = block_group->fs_info;
467 u64 start = block_group->start;
468 u64 len = block_group->length;
469 u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ?
470 fs_info->nodesize : fs_info->sectorsize;
471 u64 step = chunk << 1;
473 while (len > chunk) {
474 btrfs_remove_free_space(block_group, start, chunk);
485 * This is only called by btrfs_cache_block_group, since we could have freed
486 * extents we need to check the pinned_extents for any extents that can't be
487 * used yet since their free space will be released as soon as the transaction
490 u64 add_new_free_space(struct btrfs_block_group *block_group, u64 start, u64 end)
492 struct btrfs_fs_info *info = block_group->fs_info;
493 u64 extent_start, extent_end, size, total_added = 0;
496 while (start < end) {
497 ret = find_first_extent_bit(&info->excluded_extents, start,
498 &extent_start, &extent_end,
499 EXTENT_DIRTY | EXTENT_UPTODATE,
504 if (extent_start <= start) {
505 start = extent_end + 1;
506 } else if (extent_start > start && extent_start < end) {
507 size = extent_start - start;
509 ret = btrfs_add_free_space_async_trimmed(block_group,
511 BUG_ON(ret); /* -ENOMEM or logic error */
512 start = extent_end + 1;
521 ret = btrfs_add_free_space_async_trimmed(block_group, start,
523 BUG_ON(ret); /* -ENOMEM or logic error */
529 static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl)
531 struct btrfs_block_group *block_group = caching_ctl->block_group;
532 struct btrfs_fs_info *fs_info = block_group->fs_info;
533 struct btrfs_root *extent_root;
534 struct btrfs_path *path;
535 struct extent_buffer *leaf;
536 struct btrfs_key key;
543 path = btrfs_alloc_path();
547 last = max_t(u64, block_group->start, BTRFS_SUPER_INFO_OFFSET);
548 extent_root = btrfs_extent_root(fs_info, last);
550 #ifdef CONFIG_BTRFS_DEBUG
552 * If we're fragmenting we don't want to make anybody think we can
553 * allocate from this block group until we've had a chance to fragment
556 if (btrfs_should_fragment_free_space(block_group))
560 * We don't want to deadlock with somebody trying to allocate a new
561 * extent for the extent root while also trying to search the extent
562 * root to add free space. So we skip locking and search the commit
563 * root, since its read-only
565 path->skip_locking = 1;
566 path->search_commit_root = 1;
567 path->reada = READA_FORWARD;
571 key.type = BTRFS_EXTENT_ITEM_KEY;
574 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
578 leaf = path->nodes[0];
579 nritems = btrfs_header_nritems(leaf);
582 if (btrfs_fs_closing(fs_info) > 1) {
587 if (path->slots[0] < nritems) {
588 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
590 ret = btrfs_find_next_key(extent_root, path, &key, 0, 0);
594 if (need_resched() ||
595 rwsem_is_contended(&fs_info->commit_root_sem)) {
597 caching_ctl->progress = last;
598 btrfs_release_path(path);
599 up_read(&fs_info->commit_root_sem);
600 mutex_unlock(&caching_ctl->mutex);
602 mutex_lock(&caching_ctl->mutex);
603 down_read(&fs_info->commit_root_sem);
607 ret = btrfs_next_leaf(extent_root, path);
612 leaf = path->nodes[0];
613 nritems = btrfs_header_nritems(leaf);
617 if (key.objectid < last) {
620 key.type = BTRFS_EXTENT_ITEM_KEY;
623 caching_ctl->progress = last;
624 btrfs_release_path(path);
628 if (key.objectid < block_group->start) {
633 if (key.objectid >= block_group->start + block_group->length)
636 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
637 key.type == BTRFS_METADATA_ITEM_KEY) {
638 total_found += add_new_free_space(block_group, last,
640 if (key.type == BTRFS_METADATA_ITEM_KEY)
641 last = key.objectid +
644 last = key.objectid + key.offset;
646 if (total_found > CACHING_CTL_WAKE_UP) {
649 wake_up(&caching_ctl->wait);
656 total_found += add_new_free_space(block_group, last,
657 block_group->start + block_group->length);
658 caching_ctl->progress = (u64)-1;
661 btrfs_free_path(path);
665 static noinline void caching_thread(struct btrfs_work *work)
667 struct btrfs_block_group *block_group;
668 struct btrfs_fs_info *fs_info;
669 struct btrfs_caching_control *caching_ctl;
672 caching_ctl = container_of(work, struct btrfs_caching_control, work);
673 block_group = caching_ctl->block_group;
674 fs_info = block_group->fs_info;
676 mutex_lock(&caching_ctl->mutex);
677 down_read(&fs_info->commit_root_sem);
679 if (btrfs_test_opt(fs_info, SPACE_CACHE)) {
680 ret = load_free_space_cache(block_group);
687 * We failed to load the space cache, set ourselves to
688 * CACHE_STARTED and carry on.
690 spin_lock(&block_group->lock);
691 block_group->cached = BTRFS_CACHE_STARTED;
692 spin_unlock(&block_group->lock);
693 wake_up(&caching_ctl->wait);
697 * If we are in the transaction that populated the free space tree we
698 * can't actually cache from the free space tree as our commit root and
699 * real root are the same, so we could change the contents of the blocks
700 * while caching. Instead do the slow caching in this case, and after
701 * the transaction has committed we will be safe.
703 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
704 !(test_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags)))
705 ret = load_free_space_tree(caching_ctl);
707 ret = load_extent_tree_free(caching_ctl);
709 spin_lock(&block_group->lock);
710 block_group->caching_ctl = NULL;
711 block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED;
712 spin_unlock(&block_group->lock);
714 #ifdef CONFIG_BTRFS_DEBUG
715 if (btrfs_should_fragment_free_space(block_group)) {
718 spin_lock(&block_group->space_info->lock);
719 spin_lock(&block_group->lock);
720 bytes_used = block_group->length - block_group->used;
721 block_group->space_info->bytes_used += bytes_used >> 1;
722 spin_unlock(&block_group->lock);
723 spin_unlock(&block_group->space_info->lock);
724 fragment_free_space(block_group);
728 caching_ctl->progress = (u64)-1;
730 up_read(&fs_info->commit_root_sem);
731 btrfs_free_excluded_extents(block_group);
732 mutex_unlock(&caching_ctl->mutex);
734 wake_up(&caching_ctl->wait);
736 btrfs_put_caching_control(caching_ctl);
737 btrfs_put_block_group(block_group);
740 int btrfs_cache_block_group(struct btrfs_block_group *cache, bool wait)
742 struct btrfs_fs_info *fs_info = cache->fs_info;
743 struct btrfs_caching_control *caching_ctl = NULL;
746 /* Allocator for zoned filesystems does not use the cache at all */
747 if (btrfs_is_zoned(fs_info))
750 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
754 INIT_LIST_HEAD(&caching_ctl->list);
755 mutex_init(&caching_ctl->mutex);
756 init_waitqueue_head(&caching_ctl->wait);
757 caching_ctl->block_group = cache;
758 caching_ctl->progress = cache->start;
759 refcount_set(&caching_ctl->count, 2);
760 btrfs_init_work(&caching_ctl->work, caching_thread, NULL, NULL);
762 spin_lock(&cache->lock);
763 if (cache->cached != BTRFS_CACHE_NO) {
766 caching_ctl = cache->caching_ctl;
768 refcount_inc(&caching_ctl->count);
769 spin_unlock(&cache->lock);
772 WARN_ON(cache->caching_ctl);
773 cache->caching_ctl = caching_ctl;
774 cache->cached = BTRFS_CACHE_STARTED;
775 cache->has_caching_ctl = 1;
776 spin_unlock(&cache->lock);
778 write_lock(&fs_info->block_group_cache_lock);
779 refcount_inc(&caching_ctl->count);
780 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
781 write_unlock(&fs_info->block_group_cache_lock);
783 btrfs_get_block_group(cache);
785 btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
787 if (wait && caching_ctl)
788 ret = btrfs_caching_ctl_wait_done(cache, caching_ctl);
790 btrfs_put_caching_control(caching_ctl);
795 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
797 u64 extra_flags = chunk_to_extended(flags) &
798 BTRFS_EXTENDED_PROFILE_MASK;
800 write_seqlock(&fs_info->profiles_lock);
801 if (flags & BTRFS_BLOCK_GROUP_DATA)
802 fs_info->avail_data_alloc_bits &= ~extra_flags;
803 if (flags & BTRFS_BLOCK_GROUP_METADATA)
804 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
805 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
806 fs_info->avail_system_alloc_bits &= ~extra_flags;
807 write_sequnlock(&fs_info->profiles_lock);
811 * Clear incompat bits for the following feature(s):
813 * - RAID56 - in case there's neither RAID5 nor RAID6 profile block group
814 * in the whole filesystem
816 * - RAID1C34 - same as above for RAID1C3 and RAID1C4 block groups
818 static void clear_incompat_bg_bits(struct btrfs_fs_info *fs_info, u64 flags)
820 bool found_raid56 = false;
821 bool found_raid1c34 = false;
823 if ((flags & BTRFS_BLOCK_GROUP_RAID56_MASK) ||
824 (flags & BTRFS_BLOCK_GROUP_RAID1C3) ||
825 (flags & BTRFS_BLOCK_GROUP_RAID1C4)) {
826 struct list_head *head = &fs_info->space_info;
827 struct btrfs_space_info *sinfo;
829 list_for_each_entry_rcu(sinfo, head, list) {
830 down_read(&sinfo->groups_sem);
831 if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID5]))
833 if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID6]))
835 if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID1C3]))
836 found_raid1c34 = true;
837 if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID1C4]))
838 found_raid1c34 = true;
839 up_read(&sinfo->groups_sem);
842 btrfs_clear_fs_incompat(fs_info, RAID56);
844 btrfs_clear_fs_incompat(fs_info, RAID1C34);
848 static int remove_block_group_item(struct btrfs_trans_handle *trans,
849 struct btrfs_path *path,
850 struct btrfs_block_group *block_group)
852 struct btrfs_fs_info *fs_info = trans->fs_info;
853 struct btrfs_root *root;
854 struct btrfs_key key;
857 root = btrfs_block_group_root(fs_info);
858 key.objectid = block_group->start;
859 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
860 key.offset = block_group->length;
862 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
868 ret = btrfs_del_item(trans, root, path);
872 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
873 u64 group_start, struct extent_map *em)
875 struct btrfs_fs_info *fs_info = trans->fs_info;
876 struct btrfs_path *path;
877 struct btrfs_block_group *block_group;
878 struct btrfs_free_cluster *cluster;
880 struct kobject *kobj = NULL;
884 struct btrfs_caching_control *caching_ctl = NULL;
886 bool remove_rsv = false;
888 block_group = btrfs_lookup_block_group(fs_info, group_start);
889 BUG_ON(!block_group);
890 BUG_ON(!block_group->ro);
892 trace_btrfs_remove_block_group(block_group);
894 * Free the reserved super bytes from this block group before
897 btrfs_free_excluded_extents(block_group);
898 btrfs_free_ref_tree_range(fs_info, block_group->start,
899 block_group->length);
901 index = btrfs_bg_flags_to_raid_index(block_group->flags);
902 factor = btrfs_bg_type_to_factor(block_group->flags);
904 /* make sure this block group isn't part of an allocation cluster */
905 cluster = &fs_info->data_alloc_cluster;
906 spin_lock(&cluster->refill_lock);
907 btrfs_return_cluster_to_free_space(block_group, cluster);
908 spin_unlock(&cluster->refill_lock);
911 * make sure this block group isn't part of a metadata
914 cluster = &fs_info->meta_alloc_cluster;
915 spin_lock(&cluster->refill_lock);
916 btrfs_return_cluster_to_free_space(block_group, cluster);
917 spin_unlock(&cluster->refill_lock);
919 btrfs_clear_treelog_bg(block_group);
920 btrfs_clear_data_reloc_bg(block_group);
922 path = btrfs_alloc_path();
929 * get the inode first so any iput calls done for the io_list
930 * aren't the final iput (no unlinks allowed now)
932 inode = lookup_free_space_inode(block_group, path);
934 mutex_lock(&trans->transaction->cache_write_mutex);
936 * Make sure our free space cache IO is done before removing the
939 spin_lock(&trans->transaction->dirty_bgs_lock);
940 if (!list_empty(&block_group->io_list)) {
941 list_del_init(&block_group->io_list);
943 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
945 spin_unlock(&trans->transaction->dirty_bgs_lock);
946 btrfs_wait_cache_io(trans, block_group, path);
947 btrfs_put_block_group(block_group);
948 spin_lock(&trans->transaction->dirty_bgs_lock);
951 if (!list_empty(&block_group->dirty_list)) {
952 list_del_init(&block_group->dirty_list);
954 btrfs_put_block_group(block_group);
956 spin_unlock(&trans->transaction->dirty_bgs_lock);
957 mutex_unlock(&trans->transaction->cache_write_mutex);
959 ret = btrfs_remove_free_space_inode(trans, inode, block_group);
963 write_lock(&fs_info->block_group_cache_lock);
964 rb_erase_cached(&block_group->cache_node,
965 &fs_info->block_group_cache_tree);
966 RB_CLEAR_NODE(&block_group->cache_node);
968 /* Once for the block groups rbtree */
969 btrfs_put_block_group(block_group);
971 write_unlock(&fs_info->block_group_cache_lock);
973 down_write(&block_group->space_info->groups_sem);
975 * we must use list_del_init so people can check to see if they
976 * are still on the list after taking the semaphore
978 list_del_init(&block_group->list);
979 if (list_empty(&block_group->space_info->block_groups[index])) {
980 kobj = block_group->space_info->block_group_kobjs[index];
981 block_group->space_info->block_group_kobjs[index] = NULL;
982 clear_avail_alloc_bits(fs_info, block_group->flags);
984 up_write(&block_group->space_info->groups_sem);
985 clear_incompat_bg_bits(fs_info, block_group->flags);
991 if (block_group->has_caching_ctl)
992 caching_ctl = btrfs_get_caching_control(block_group);
993 if (block_group->cached == BTRFS_CACHE_STARTED)
994 btrfs_wait_block_group_cache_done(block_group);
995 if (block_group->has_caching_ctl) {
996 write_lock(&fs_info->block_group_cache_lock);
998 struct btrfs_caching_control *ctl;
1000 list_for_each_entry(ctl,
1001 &fs_info->caching_block_groups, list)
1002 if (ctl->block_group == block_group) {
1004 refcount_inc(&caching_ctl->count);
1009 list_del_init(&caching_ctl->list);
1010 write_unlock(&fs_info->block_group_cache_lock);
1012 /* Once for the caching bgs list and once for us. */
1013 btrfs_put_caching_control(caching_ctl);
1014 btrfs_put_caching_control(caching_ctl);
1018 spin_lock(&trans->transaction->dirty_bgs_lock);
1019 WARN_ON(!list_empty(&block_group->dirty_list));
1020 WARN_ON(!list_empty(&block_group->io_list));
1021 spin_unlock(&trans->transaction->dirty_bgs_lock);
1023 btrfs_remove_free_space_cache(block_group);
1025 spin_lock(&block_group->space_info->lock);
1026 list_del_init(&block_group->ro_list);
1028 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
1029 WARN_ON(block_group->space_info->total_bytes
1030 < block_group->length);
1031 WARN_ON(block_group->space_info->bytes_readonly
1032 < block_group->length - block_group->zone_unusable);
1033 WARN_ON(block_group->space_info->bytes_zone_unusable
1034 < block_group->zone_unusable);
1035 WARN_ON(block_group->space_info->disk_total
1036 < block_group->length * factor);
1037 WARN_ON(block_group->zone_is_active &&
1038 block_group->space_info->active_total_bytes
1039 < block_group->length);
1041 block_group->space_info->total_bytes -= block_group->length;
1042 if (block_group->zone_is_active)
1043 block_group->space_info->active_total_bytes -= block_group->length;
1044 block_group->space_info->bytes_readonly -=
1045 (block_group->length - block_group->zone_unusable);
1046 block_group->space_info->bytes_zone_unusable -=
1047 block_group->zone_unusable;
1048 block_group->space_info->disk_total -= block_group->length * factor;
1050 spin_unlock(&block_group->space_info->lock);
1053 * Remove the free space for the block group from the free space tree
1054 * and the block group's item from the extent tree before marking the
1055 * block group as removed. This is to prevent races with tasks that
1056 * freeze and unfreeze a block group, this task and another task
1057 * allocating a new block group - the unfreeze task ends up removing
1058 * the block group's extent map before the task calling this function
1059 * deletes the block group item from the extent tree, allowing for
1060 * another task to attempt to create another block group with the same
1061 * item key (and failing with -EEXIST and a transaction abort).
1063 ret = remove_block_group_free_space(trans, block_group);
1067 ret = remove_block_group_item(trans, path, block_group);
1071 spin_lock(&block_group->lock);
1072 block_group->removed = 1;
1074 * At this point trimming or scrub can't start on this block group,
1075 * because we removed the block group from the rbtree
1076 * fs_info->block_group_cache_tree so no one can't find it anymore and
1077 * even if someone already got this block group before we removed it
1078 * from the rbtree, they have already incremented block_group->frozen -
1079 * if they didn't, for the trimming case they won't find any free space
1080 * entries because we already removed them all when we called
1081 * btrfs_remove_free_space_cache().
1083 * And we must not remove the extent map from the fs_info->mapping_tree
1084 * to prevent the same logical address range and physical device space
1085 * ranges from being reused for a new block group. This is needed to
1086 * avoid races with trimming and scrub.
1088 * An fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
1089 * completely transactionless, so while it is trimming a range the
1090 * currently running transaction might finish and a new one start,
1091 * allowing for new block groups to be created that can reuse the same
1092 * physical device locations unless we take this special care.
1094 * There may also be an implicit trim operation if the file system
1095 * is mounted with -odiscard. The same protections must remain
1096 * in place until the extents have been discarded completely when
1097 * the transaction commit has completed.
1099 remove_em = (atomic_read(&block_group->frozen) == 0);
1100 spin_unlock(&block_group->lock);
1103 struct extent_map_tree *em_tree;
1105 em_tree = &fs_info->mapping_tree;
1106 write_lock(&em_tree->lock);
1107 remove_extent_mapping(em_tree, em);
1108 write_unlock(&em_tree->lock);
1109 /* once for the tree */
1110 free_extent_map(em);
1114 /* Once for the lookup reference */
1115 btrfs_put_block_group(block_group);
1117 btrfs_delayed_refs_rsv_release(fs_info, 1);
1118 btrfs_free_path(path);
1122 struct btrfs_trans_handle *btrfs_start_trans_remove_block_group(
1123 struct btrfs_fs_info *fs_info, const u64 chunk_offset)
1125 struct btrfs_root *root = btrfs_block_group_root(fs_info);
1126 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
1127 struct extent_map *em;
1128 struct map_lookup *map;
1129 unsigned int num_items;
1131 read_lock(&em_tree->lock);
1132 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1133 read_unlock(&em_tree->lock);
1134 ASSERT(em && em->start == chunk_offset);
1137 * We need to reserve 3 + N units from the metadata space info in order
1138 * to remove a block group (done at btrfs_remove_chunk() and at
1139 * btrfs_remove_block_group()), which are used for:
1141 * 1 unit for adding the free space inode's orphan (located in the tree
1143 * 1 unit for deleting the block group item (located in the extent
1145 * 1 unit for deleting the free space item (located in tree of tree
1147 * N units for deleting N device extent items corresponding to each
1148 * stripe (located in the device tree).
1150 * In order to remove a block group we also need to reserve units in the
1151 * system space info in order to update the chunk tree (update one or
1152 * more device items and remove one chunk item), but this is done at
1153 * btrfs_remove_chunk() through a call to check_system_chunk().
1155 map = em->map_lookup;
1156 num_items = 3 + map->num_stripes;
1157 free_extent_map(em);
1159 return btrfs_start_transaction_fallback_global_rsv(root, num_items);
1163 * Mark block group @cache read-only, so later write won't happen to block
1166 * If @force is not set, this function will only mark the block group readonly
1167 * if we have enough free space (1M) in other metadata/system block groups.
1168 * If @force is not set, this function will mark the block group readonly
1169 * without checking free space.
1171 * NOTE: This function doesn't care if other block groups can contain all the
1172 * data in this block group. That check should be done by relocation routine,
1173 * not this function.
1175 static int inc_block_group_ro(struct btrfs_block_group *cache, int force)
1177 struct btrfs_space_info *sinfo = cache->space_info;
1181 spin_lock(&sinfo->lock);
1182 spin_lock(&cache->lock);
1184 if (cache->swap_extents) {
1195 num_bytes = cache->length - cache->reserved - cache->pinned -
1196 cache->bytes_super - cache->zone_unusable - cache->used;
1199 * Data never overcommits, even in mixed mode, so do just the straight
1200 * check of left over space in how much we have allocated.
1204 } else if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA) {
1205 u64 sinfo_used = btrfs_space_info_used(sinfo, true);
1208 * Here we make sure if we mark this bg RO, we still have enough
1209 * free space as buffer.
1211 if (sinfo_used + num_bytes <= sinfo->total_bytes)
1215 * We overcommit metadata, so we need to do the
1216 * btrfs_can_overcommit check here, and we need to pass in
1217 * BTRFS_RESERVE_NO_FLUSH to give ourselves the most amount of
1218 * leeway to allow us to mark this block group as read only.
1220 if (btrfs_can_overcommit(cache->fs_info, sinfo, num_bytes,
1221 BTRFS_RESERVE_NO_FLUSH))
1226 sinfo->bytes_readonly += num_bytes;
1227 if (btrfs_is_zoned(cache->fs_info)) {
1228 /* Migrate zone_unusable bytes to readonly */
1229 sinfo->bytes_readonly += cache->zone_unusable;
1230 sinfo->bytes_zone_unusable -= cache->zone_unusable;
1231 cache->zone_unusable = 0;
1234 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
1237 spin_unlock(&cache->lock);
1238 spin_unlock(&sinfo->lock);
1239 if (ret == -ENOSPC && btrfs_test_opt(cache->fs_info, ENOSPC_DEBUG)) {
1240 btrfs_info(cache->fs_info,
1241 "unable to make block group %llu ro", cache->start);
1242 btrfs_dump_space_info(cache->fs_info, cache->space_info, 0, 0);
1247 static bool clean_pinned_extents(struct btrfs_trans_handle *trans,
1248 struct btrfs_block_group *bg)
1250 struct btrfs_fs_info *fs_info = bg->fs_info;
1251 struct btrfs_transaction *prev_trans = NULL;
1252 const u64 start = bg->start;
1253 const u64 end = start + bg->length - 1;
1256 spin_lock(&fs_info->trans_lock);
1257 if (trans->transaction->list.prev != &fs_info->trans_list) {
1258 prev_trans = list_last_entry(&trans->transaction->list,
1259 struct btrfs_transaction, list);
1260 refcount_inc(&prev_trans->use_count);
1262 spin_unlock(&fs_info->trans_lock);
1265 * Hold the unused_bg_unpin_mutex lock to avoid racing with
1266 * btrfs_finish_extent_commit(). If we are at transaction N, another
1267 * task might be running finish_extent_commit() for the previous
1268 * transaction N - 1, and have seen a range belonging to the block
1269 * group in pinned_extents before we were able to clear the whole block
1270 * group range from pinned_extents. This means that task can lookup for
1271 * the block group after we unpinned it from pinned_extents and removed
1272 * it, leading to a BUG_ON() at unpin_extent_range().
1274 mutex_lock(&fs_info->unused_bg_unpin_mutex);
1276 ret = clear_extent_bits(&prev_trans->pinned_extents, start, end,
1282 ret = clear_extent_bits(&trans->transaction->pinned_extents, start, end,
1285 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
1287 btrfs_put_transaction(prev_trans);
1293 * Process the unused_bgs list and remove any that don't have any allocated
1294 * space inside of them.
1296 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
1298 struct btrfs_block_group *block_group;
1299 struct btrfs_space_info *space_info;
1300 struct btrfs_trans_handle *trans;
1301 const bool async_trim_enabled = btrfs_test_opt(fs_info, DISCARD_ASYNC);
1304 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1308 * Long running balances can keep us blocked here for eternity, so
1309 * simply skip deletion if we're unable to get the mutex.
1311 if (!mutex_trylock(&fs_info->reclaim_bgs_lock))
1314 spin_lock(&fs_info->unused_bgs_lock);
1315 while (!list_empty(&fs_info->unused_bgs)) {
1318 block_group = list_first_entry(&fs_info->unused_bgs,
1319 struct btrfs_block_group,
1321 list_del_init(&block_group->bg_list);
1323 space_info = block_group->space_info;
1325 if (ret || btrfs_mixed_space_info(space_info)) {
1326 btrfs_put_block_group(block_group);
1329 spin_unlock(&fs_info->unused_bgs_lock);
1331 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group);
1333 /* Don't want to race with allocators so take the groups_sem */
1334 down_write(&space_info->groups_sem);
1337 * Async discard moves the final block group discard to be prior
1338 * to the unused_bgs code path. Therefore, if it's not fully
1339 * trimmed, punt it back to the async discard lists.
1341 if (btrfs_test_opt(fs_info, DISCARD_ASYNC) &&
1342 !btrfs_is_free_space_trimmed(block_group)) {
1343 trace_btrfs_skip_unused_block_group(block_group);
1344 up_write(&space_info->groups_sem);
1345 /* Requeue if we failed because of async discard */
1346 btrfs_discard_queue_work(&fs_info->discard_ctl,
1351 spin_lock(&block_group->lock);
1352 if (block_group->reserved || block_group->pinned ||
1353 block_group->used || block_group->ro ||
1354 list_is_singular(&block_group->list)) {
1356 * We want to bail if we made new allocations or have
1357 * outstanding allocations in this block group. We do
1358 * the ro check in case balance is currently acting on
1361 trace_btrfs_skip_unused_block_group(block_group);
1362 spin_unlock(&block_group->lock);
1363 up_write(&space_info->groups_sem);
1366 spin_unlock(&block_group->lock);
1368 /* We don't want to force the issue, only flip if it's ok. */
1369 ret = inc_block_group_ro(block_group, 0);
1370 up_write(&space_info->groups_sem);
1376 ret = btrfs_zone_finish(block_group);
1378 btrfs_dec_block_group_ro(block_group);
1385 * Want to do this before we do anything else so we can recover
1386 * properly if we fail to join the transaction.
1388 trans = btrfs_start_trans_remove_block_group(fs_info,
1389 block_group->start);
1390 if (IS_ERR(trans)) {
1391 btrfs_dec_block_group_ro(block_group);
1392 ret = PTR_ERR(trans);
1397 * We could have pending pinned extents for this block group,
1398 * just delete them, we don't care about them anymore.
1400 if (!clean_pinned_extents(trans, block_group)) {
1401 btrfs_dec_block_group_ro(block_group);
1406 * At this point, the block_group is read only and should fail
1407 * new allocations. However, btrfs_finish_extent_commit() can
1408 * cause this block_group to be placed back on the discard
1409 * lists because now the block_group isn't fully discarded.
1410 * Bail here and try again later after discarding everything.
1412 spin_lock(&fs_info->discard_ctl.lock);
1413 if (!list_empty(&block_group->discard_list)) {
1414 spin_unlock(&fs_info->discard_ctl.lock);
1415 btrfs_dec_block_group_ro(block_group);
1416 btrfs_discard_queue_work(&fs_info->discard_ctl,
1420 spin_unlock(&fs_info->discard_ctl.lock);
1422 /* Reset pinned so btrfs_put_block_group doesn't complain */
1423 spin_lock(&space_info->lock);
1424 spin_lock(&block_group->lock);
1426 btrfs_space_info_update_bytes_pinned(fs_info, space_info,
1427 -block_group->pinned);
1428 space_info->bytes_readonly += block_group->pinned;
1429 block_group->pinned = 0;
1431 spin_unlock(&block_group->lock);
1432 spin_unlock(&space_info->lock);
1435 * The normal path here is an unused block group is passed here,
1436 * then trimming is handled in the transaction commit path.
1437 * Async discard interposes before this to do the trimming
1438 * before coming down the unused block group path as trimming
1439 * will no longer be done later in the transaction commit path.
1441 if (!async_trim_enabled && btrfs_test_opt(fs_info, DISCARD_ASYNC))
1445 * DISCARD can flip during remount. On zoned filesystems, we
1446 * need to reset sequential-required zones.
1448 trimming = btrfs_test_opt(fs_info, DISCARD_SYNC) ||
1449 btrfs_is_zoned(fs_info);
1451 /* Implicit trim during transaction commit. */
1453 btrfs_freeze_block_group(block_group);
1456 * Btrfs_remove_chunk will abort the transaction if things go
1459 ret = btrfs_remove_chunk(trans, block_group->start);
1463 btrfs_unfreeze_block_group(block_group);
1468 * If we're not mounted with -odiscard, we can just forget
1469 * about this block group. Otherwise we'll need to wait
1470 * until transaction commit to do the actual discard.
1473 spin_lock(&fs_info->unused_bgs_lock);
1475 * A concurrent scrub might have added us to the list
1476 * fs_info->unused_bgs, so use a list_move operation
1477 * to add the block group to the deleted_bgs list.
1479 list_move(&block_group->bg_list,
1480 &trans->transaction->deleted_bgs);
1481 spin_unlock(&fs_info->unused_bgs_lock);
1482 btrfs_get_block_group(block_group);
1485 btrfs_end_transaction(trans);
1487 btrfs_put_block_group(block_group);
1488 spin_lock(&fs_info->unused_bgs_lock);
1490 spin_unlock(&fs_info->unused_bgs_lock);
1491 mutex_unlock(&fs_info->reclaim_bgs_lock);
1495 btrfs_end_transaction(trans);
1496 mutex_unlock(&fs_info->reclaim_bgs_lock);
1497 btrfs_put_block_group(block_group);
1498 btrfs_discard_punt_unused_bgs_list(fs_info);
1501 void btrfs_mark_bg_unused(struct btrfs_block_group *bg)
1503 struct btrfs_fs_info *fs_info = bg->fs_info;
1505 spin_lock(&fs_info->unused_bgs_lock);
1506 if (list_empty(&bg->bg_list)) {
1507 btrfs_get_block_group(bg);
1508 trace_btrfs_add_unused_block_group(bg);
1509 list_add_tail(&bg->bg_list, &fs_info->unused_bgs);
1511 spin_unlock(&fs_info->unused_bgs_lock);
1515 * We want block groups with a low number of used bytes to be in the beginning
1516 * of the list, so they will get reclaimed first.
1518 static int reclaim_bgs_cmp(void *unused, const struct list_head *a,
1519 const struct list_head *b)
1521 const struct btrfs_block_group *bg1, *bg2;
1523 bg1 = list_entry(a, struct btrfs_block_group, bg_list);
1524 bg2 = list_entry(b, struct btrfs_block_group, bg_list);
1526 return bg1->used > bg2->used;
1529 static inline bool btrfs_should_reclaim(struct btrfs_fs_info *fs_info)
1531 if (btrfs_is_zoned(fs_info))
1532 return btrfs_zoned_should_reclaim(fs_info);
1536 void btrfs_reclaim_bgs_work(struct work_struct *work)
1538 struct btrfs_fs_info *fs_info =
1539 container_of(work, struct btrfs_fs_info, reclaim_bgs_work);
1540 struct btrfs_block_group *bg;
1541 struct btrfs_space_info *space_info;
1543 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1546 if (!btrfs_should_reclaim(fs_info))
1549 sb_start_write(fs_info->sb);
1551 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
1552 sb_end_write(fs_info->sb);
1557 * Long running balances can keep us blocked here for eternity, so
1558 * simply skip reclaim if we're unable to get the mutex.
1560 if (!mutex_trylock(&fs_info->reclaim_bgs_lock)) {
1561 btrfs_exclop_finish(fs_info);
1562 sb_end_write(fs_info->sb);
1566 spin_lock(&fs_info->unused_bgs_lock);
1568 * Sort happens under lock because we can't simply splice it and sort.
1569 * The block groups might still be in use and reachable via bg_list,
1570 * and their presence in the reclaim_bgs list must be preserved.
1572 list_sort(NULL, &fs_info->reclaim_bgs, reclaim_bgs_cmp);
1573 while (!list_empty(&fs_info->reclaim_bgs)) {
1577 bg = list_first_entry(&fs_info->reclaim_bgs,
1578 struct btrfs_block_group,
1580 list_del_init(&bg->bg_list);
1582 space_info = bg->space_info;
1583 spin_unlock(&fs_info->unused_bgs_lock);
1585 /* Don't race with allocators so take the groups_sem */
1586 down_write(&space_info->groups_sem);
1588 spin_lock(&bg->lock);
1589 if (bg->reserved || bg->pinned || bg->ro) {
1591 * We want to bail if we made new allocations or have
1592 * outstanding allocations in this block group. We do
1593 * the ro check in case balance is currently acting on
1596 spin_unlock(&bg->lock);
1597 up_write(&space_info->groups_sem);
1600 spin_unlock(&bg->lock);
1602 /* Get out fast, in case we're unmounting the filesystem */
1603 if (btrfs_fs_closing(fs_info)) {
1604 up_write(&space_info->groups_sem);
1609 * Cache the zone_unusable value before turning the block group
1610 * to read only. As soon as the blog group is read only it's
1611 * zone_unusable value gets moved to the block group's read-only
1612 * bytes and isn't available for calculations anymore.
1614 zone_unusable = bg->zone_unusable;
1615 ret = inc_block_group_ro(bg, 0);
1616 up_write(&space_info->groups_sem);
1621 "reclaiming chunk %llu with %llu%% used %llu%% unusable",
1622 bg->start, div_u64(bg->used * 100, bg->length),
1623 div64_u64(zone_unusable * 100, bg->length));
1624 trace_btrfs_reclaim_block_group(bg);
1625 ret = btrfs_relocate_chunk(fs_info, bg->start);
1627 btrfs_dec_block_group_ro(bg);
1628 btrfs_err(fs_info, "error relocating chunk %llu",
1633 btrfs_put_block_group(bg);
1634 spin_lock(&fs_info->unused_bgs_lock);
1636 spin_unlock(&fs_info->unused_bgs_lock);
1637 mutex_unlock(&fs_info->reclaim_bgs_lock);
1638 btrfs_exclop_finish(fs_info);
1639 sb_end_write(fs_info->sb);
1642 void btrfs_reclaim_bgs(struct btrfs_fs_info *fs_info)
1644 spin_lock(&fs_info->unused_bgs_lock);
1645 if (!list_empty(&fs_info->reclaim_bgs))
1646 queue_work(system_unbound_wq, &fs_info->reclaim_bgs_work);
1647 spin_unlock(&fs_info->unused_bgs_lock);
1650 void btrfs_mark_bg_to_reclaim(struct btrfs_block_group *bg)
1652 struct btrfs_fs_info *fs_info = bg->fs_info;
1654 spin_lock(&fs_info->unused_bgs_lock);
1655 if (list_empty(&bg->bg_list)) {
1656 btrfs_get_block_group(bg);
1657 trace_btrfs_add_reclaim_block_group(bg);
1658 list_add_tail(&bg->bg_list, &fs_info->reclaim_bgs);
1660 spin_unlock(&fs_info->unused_bgs_lock);
1663 static int read_bg_from_eb(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
1664 struct btrfs_path *path)
1666 struct extent_map_tree *em_tree;
1667 struct extent_map *em;
1668 struct btrfs_block_group_item bg;
1669 struct extent_buffer *leaf;
1674 slot = path->slots[0];
1675 leaf = path->nodes[0];
1677 em_tree = &fs_info->mapping_tree;
1678 read_lock(&em_tree->lock);
1679 em = lookup_extent_mapping(em_tree, key->objectid, key->offset);
1680 read_unlock(&em_tree->lock);
1683 "logical %llu len %llu found bg but no related chunk",
1684 key->objectid, key->offset);
1688 if (em->start != key->objectid || em->len != key->offset) {
1690 "block group %llu len %llu mismatch with chunk %llu len %llu",
1691 key->objectid, key->offset, em->start, em->len);
1696 read_extent_buffer(leaf, &bg, btrfs_item_ptr_offset(leaf, slot),
1698 flags = btrfs_stack_block_group_flags(&bg) &
1699 BTRFS_BLOCK_GROUP_TYPE_MASK;
1701 if (flags != (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
1703 "block group %llu len %llu type flags 0x%llx mismatch with chunk type flags 0x%llx",
1704 key->objectid, key->offset, flags,
1705 (BTRFS_BLOCK_GROUP_TYPE_MASK & em->map_lookup->type));
1710 free_extent_map(em);
1714 static int find_first_block_group(struct btrfs_fs_info *fs_info,
1715 struct btrfs_path *path,
1716 struct btrfs_key *key)
1718 struct btrfs_root *root = btrfs_block_group_root(fs_info);
1720 struct btrfs_key found_key;
1722 btrfs_for_each_slot(root, key, &found_key, path, ret) {
1723 if (found_key.objectid >= key->objectid &&
1724 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
1725 return read_bg_from_eb(fs_info, &found_key, path);
1731 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1733 u64 extra_flags = chunk_to_extended(flags) &
1734 BTRFS_EXTENDED_PROFILE_MASK;
1736 write_seqlock(&fs_info->profiles_lock);
1737 if (flags & BTRFS_BLOCK_GROUP_DATA)
1738 fs_info->avail_data_alloc_bits |= extra_flags;
1739 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1740 fs_info->avail_metadata_alloc_bits |= extra_flags;
1741 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1742 fs_info->avail_system_alloc_bits |= extra_flags;
1743 write_sequnlock(&fs_info->profiles_lock);
1747 * Map a physical disk address to a list of logical addresses
1749 * @fs_info: the filesystem
1750 * @chunk_start: logical address of block group
1751 * @bdev: physical device to resolve, can be NULL to indicate any device
1752 * @physical: physical address to map to logical addresses
1753 * @logical: return array of logical addresses which map to @physical
1754 * @naddrs: length of @logical
1755 * @stripe_len: size of IO stripe for the given block group
1757 * Maps a particular @physical disk address to a list of @logical addresses.
1758 * Used primarily to exclude those portions of a block group that contain super
1761 int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start,
1762 struct block_device *bdev, u64 physical, u64 **logical,
1763 int *naddrs, int *stripe_len)
1765 struct extent_map *em;
1766 struct map_lookup *map;
1769 u64 data_stripe_length;
1774 em = btrfs_get_chunk_map(fs_info, chunk_start, 1);
1778 map = em->map_lookup;
1779 data_stripe_length = em->orig_block_len;
1780 io_stripe_size = map->stripe_len;
1781 chunk_start = em->start;
1783 /* For RAID5/6 adjust to a full IO stripe length */
1784 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
1785 io_stripe_size = map->stripe_len * nr_data_stripes(map);
1787 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
1793 for (i = 0; i < map->num_stripes; i++) {
1794 bool already_inserted = false;
1799 if (!in_range(physical, map->stripes[i].physical,
1800 data_stripe_length))
1803 if (bdev && map->stripes[i].dev->bdev != bdev)
1806 stripe_nr = physical - map->stripes[i].physical;
1807 stripe_nr = div64_u64_rem(stripe_nr, map->stripe_len, &offset);
1809 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
1810 BTRFS_BLOCK_GROUP_RAID10)) {
1811 stripe_nr = stripe_nr * map->num_stripes + i;
1812 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
1815 * The remaining case would be for RAID56, multiply by
1816 * nr_data_stripes(). Alternatively, just use rmap_len below
1817 * instead of map->stripe_len
1820 bytenr = chunk_start + stripe_nr * io_stripe_size + offset;
1822 /* Ensure we don't add duplicate addresses */
1823 for (j = 0; j < nr; j++) {
1824 if (buf[j] == bytenr) {
1825 already_inserted = true;
1830 if (!already_inserted)
1836 *stripe_len = io_stripe_size;
1838 free_extent_map(em);
1842 static int exclude_super_stripes(struct btrfs_block_group *cache)
1844 struct btrfs_fs_info *fs_info = cache->fs_info;
1845 const bool zoned = btrfs_is_zoned(fs_info);
1851 if (cache->start < BTRFS_SUPER_INFO_OFFSET) {
1852 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->start;
1853 cache->bytes_super += stripe_len;
1854 ret = btrfs_add_excluded_extent(fs_info, cache->start,
1860 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1861 bytenr = btrfs_sb_offset(i);
1862 ret = btrfs_rmap_block(fs_info, cache->start, NULL,
1863 bytenr, &logical, &nr, &stripe_len);
1867 /* Shouldn't have super stripes in sequential zones */
1870 "zoned: block group %llu must not contain super block",
1876 u64 len = min_t(u64, stripe_len,
1877 cache->start + cache->length - logical[nr]);
1879 cache->bytes_super += len;
1880 ret = btrfs_add_excluded_extent(fs_info, logical[nr],
1893 static void link_block_group(struct btrfs_block_group *cache)
1895 struct btrfs_space_info *space_info = cache->space_info;
1896 int index = btrfs_bg_flags_to_raid_index(cache->flags);
1898 down_write(&space_info->groups_sem);
1899 list_add_tail(&cache->list, &space_info->block_groups[index]);
1900 up_write(&space_info->groups_sem);
1903 static struct btrfs_block_group *btrfs_create_block_group_cache(
1904 struct btrfs_fs_info *fs_info, u64 start)
1906 struct btrfs_block_group *cache;
1908 cache = kzalloc(sizeof(*cache), GFP_NOFS);
1912 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
1914 if (!cache->free_space_ctl) {
1919 cache->start = start;
1921 cache->fs_info = fs_info;
1922 cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start);
1924 cache->discard_index = BTRFS_DISCARD_INDEX_UNUSED;
1926 refcount_set(&cache->refs, 1);
1927 spin_lock_init(&cache->lock);
1928 init_rwsem(&cache->data_rwsem);
1929 INIT_LIST_HEAD(&cache->list);
1930 INIT_LIST_HEAD(&cache->cluster_list);
1931 INIT_LIST_HEAD(&cache->bg_list);
1932 INIT_LIST_HEAD(&cache->ro_list);
1933 INIT_LIST_HEAD(&cache->discard_list);
1934 INIT_LIST_HEAD(&cache->dirty_list);
1935 INIT_LIST_HEAD(&cache->io_list);
1936 INIT_LIST_HEAD(&cache->active_bg_list);
1937 btrfs_init_free_space_ctl(cache, cache->free_space_ctl);
1938 atomic_set(&cache->frozen, 0);
1939 mutex_init(&cache->free_space_lock);
1940 btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root);
1946 * Iterate all chunks and verify that each of them has the corresponding block
1949 static int check_chunk_block_group_mappings(struct btrfs_fs_info *fs_info)
1951 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
1952 struct extent_map *em;
1953 struct btrfs_block_group *bg;
1958 read_lock(&map_tree->lock);
1960 * lookup_extent_mapping will return the first extent map
1961 * intersecting the range, so setting @len to 1 is enough to
1962 * get the first chunk.
1964 em = lookup_extent_mapping(map_tree, start, 1);
1965 read_unlock(&map_tree->lock);
1969 bg = btrfs_lookup_block_group(fs_info, em->start);
1972 "chunk start=%llu len=%llu doesn't have corresponding block group",
1973 em->start, em->len);
1975 free_extent_map(em);
1978 if (bg->start != em->start || bg->length != em->len ||
1979 (bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK) !=
1980 (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
1982 "chunk start=%llu len=%llu flags=0x%llx doesn't match block group start=%llu len=%llu flags=0x%llx",
1984 em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK,
1985 bg->start, bg->length,
1986 bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK);
1988 free_extent_map(em);
1989 btrfs_put_block_group(bg);
1992 start = em->start + em->len;
1993 free_extent_map(em);
1994 btrfs_put_block_group(bg);
1999 static int read_one_block_group(struct btrfs_fs_info *info,
2000 struct btrfs_block_group_item *bgi,
2001 const struct btrfs_key *key,
2004 struct btrfs_block_group *cache;
2005 struct btrfs_space_info *space_info;
2006 const bool mixed = btrfs_fs_incompat(info, MIXED_GROUPS);
2009 ASSERT(key->type == BTRFS_BLOCK_GROUP_ITEM_KEY);
2011 cache = btrfs_create_block_group_cache(info, key->objectid);
2015 cache->length = key->offset;
2016 cache->used = btrfs_stack_block_group_used(bgi);
2017 cache->flags = btrfs_stack_block_group_flags(bgi);
2018 cache->global_root_id = btrfs_stack_block_group_chunk_objectid(bgi);
2020 set_free_space_tree_thresholds(cache);
2024 * When we mount with old space cache, we need to
2025 * set BTRFS_DC_CLEAR and set dirty flag.
2027 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
2028 * truncate the old free space cache inode and
2030 * b) Setting 'dirty flag' makes sure that we flush
2031 * the new space cache info onto disk.
2033 if (btrfs_test_opt(info, SPACE_CACHE))
2034 cache->disk_cache_state = BTRFS_DC_CLEAR;
2036 if (!mixed && ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) &&
2037 (cache->flags & BTRFS_BLOCK_GROUP_DATA))) {
2039 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
2045 ret = btrfs_load_block_group_zone_info(cache, false);
2047 btrfs_err(info, "zoned: failed to load zone info of bg %llu",
2053 * We need to exclude the super stripes now so that the space info has
2054 * super bytes accounted for, otherwise we'll think we have more space
2055 * than we actually do.
2057 ret = exclude_super_stripes(cache);
2059 /* We may have excluded something, so call this just in case. */
2060 btrfs_free_excluded_extents(cache);
2065 * For zoned filesystem, space after the allocation offset is the only
2066 * free space for a block group. So, we don't need any caching work.
2067 * btrfs_calc_zone_unusable() will set the amount of free space and
2068 * zone_unusable space.
2070 * For regular filesystem, check for two cases, either we are full, and
2071 * therefore don't need to bother with the caching work since we won't
2072 * find any space, or we are empty, and we can just add all the space
2073 * in and be done with it. This saves us _a_lot_ of time, particularly
2076 if (btrfs_is_zoned(info)) {
2077 btrfs_calc_zone_unusable(cache);
2078 /* Should not have any excluded extents. Just in case, though. */
2079 btrfs_free_excluded_extents(cache);
2080 } else if (cache->length == cache->used) {
2081 cache->last_byte_to_unpin = (u64)-1;
2082 cache->cached = BTRFS_CACHE_FINISHED;
2083 btrfs_free_excluded_extents(cache);
2084 } else if (cache->used == 0) {
2085 cache->last_byte_to_unpin = (u64)-1;
2086 cache->cached = BTRFS_CACHE_FINISHED;
2087 add_new_free_space(cache, cache->start,
2088 cache->start + cache->length);
2089 btrfs_free_excluded_extents(cache);
2092 ret = btrfs_add_block_group_cache(info, cache);
2094 btrfs_remove_free_space_cache(cache);
2097 trace_btrfs_add_block_group(info, cache, 0);
2098 btrfs_update_space_info(info, cache->flags, cache->length,
2099 cache->used, cache->bytes_super,
2100 cache->zone_unusable, cache->zone_is_active,
2103 cache->space_info = space_info;
2105 link_block_group(cache);
2107 set_avail_alloc_bits(info, cache->flags);
2108 if (btrfs_chunk_writeable(info, cache->start)) {
2109 if (cache->used == 0) {
2110 ASSERT(list_empty(&cache->bg_list));
2111 if (btrfs_test_opt(info, DISCARD_ASYNC))
2112 btrfs_discard_queue_work(&info->discard_ctl, cache);
2114 btrfs_mark_bg_unused(cache);
2117 inc_block_group_ro(cache, 1);
2122 btrfs_put_block_group(cache);
2126 static int fill_dummy_bgs(struct btrfs_fs_info *fs_info)
2128 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
2129 struct btrfs_space_info *space_info;
2130 struct rb_node *node;
2133 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) {
2134 struct extent_map *em;
2135 struct map_lookup *map;
2136 struct btrfs_block_group *bg;
2138 em = rb_entry(node, struct extent_map, rb_node);
2139 map = em->map_lookup;
2140 bg = btrfs_create_block_group_cache(fs_info, em->start);
2146 /* Fill dummy cache as FULL */
2147 bg->length = em->len;
2148 bg->flags = map->type;
2149 bg->last_byte_to_unpin = (u64)-1;
2150 bg->cached = BTRFS_CACHE_FINISHED;
2152 bg->flags = map->type;
2153 ret = btrfs_add_block_group_cache(fs_info, bg);
2155 * We may have some valid block group cache added already, in
2156 * that case we skip to the next one.
2158 if (ret == -EEXIST) {
2160 btrfs_put_block_group(bg);
2165 btrfs_remove_free_space_cache(bg);
2166 btrfs_put_block_group(bg);
2170 btrfs_update_space_info(fs_info, bg->flags, em->len, em->len,
2171 0, 0, false, &space_info);
2172 bg->space_info = space_info;
2173 link_block_group(bg);
2175 set_avail_alloc_bits(fs_info, bg->flags);
2178 btrfs_init_global_block_rsv(fs_info);
2182 int btrfs_read_block_groups(struct btrfs_fs_info *info)
2184 struct btrfs_root *root = btrfs_block_group_root(info);
2185 struct btrfs_path *path;
2187 struct btrfs_block_group *cache;
2188 struct btrfs_space_info *space_info;
2189 struct btrfs_key key;
2194 return fill_dummy_bgs(info);
2198 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
2199 path = btrfs_alloc_path();
2203 cache_gen = btrfs_super_cache_generation(info->super_copy);
2204 if (btrfs_test_opt(info, SPACE_CACHE) &&
2205 btrfs_super_generation(info->super_copy) != cache_gen)
2207 if (btrfs_test_opt(info, CLEAR_CACHE))
2211 struct btrfs_block_group_item bgi;
2212 struct extent_buffer *leaf;
2215 ret = find_first_block_group(info, path, &key);
2221 leaf = path->nodes[0];
2222 slot = path->slots[0];
2224 read_extent_buffer(leaf, &bgi, btrfs_item_ptr_offset(leaf, slot),
2227 btrfs_item_key_to_cpu(leaf, &key, slot);
2228 btrfs_release_path(path);
2229 ret = read_one_block_group(info, &bgi, &key, need_clear);
2232 key.objectid += key.offset;
2235 btrfs_release_path(path);
2237 list_for_each_entry(space_info, &info->space_info, list) {
2240 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
2241 if (list_empty(&space_info->block_groups[i]))
2243 cache = list_first_entry(&space_info->block_groups[i],
2244 struct btrfs_block_group,
2246 btrfs_sysfs_add_block_group_type(cache);
2249 if (!(btrfs_get_alloc_profile(info, space_info->flags) &
2250 (BTRFS_BLOCK_GROUP_RAID10 |
2251 BTRFS_BLOCK_GROUP_RAID1_MASK |
2252 BTRFS_BLOCK_GROUP_RAID56_MASK |
2253 BTRFS_BLOCK_GROUP_DUP)))
2256 * Avoid allocating from un-mirrored block group if there are
2257 * mirrored block groups.
2259 list_for_each_entry(cache,
2260 &space_info->block_groups[BTRFS_RAID_RAID0],
2262 inc_block_group_ro(cache, 1);
2263 list_for_each_entry(cache,
2264 &space_info->block_groups[BTRFS_RAID_SINGLE],
2266 inc_block_group_ro(cache, 1);
2269 btrfs_init_global_block_rsv(info);
2270 ret = check_chunk_block_group_mappings(info);
2272 btrfs_free_path(path);
2274 * We've hit some error while reading the extent tree, and have
2275 * rescue=ibadroots mount option.
2276 * Try to fill the tree using dummy block groups so that the user can
2277 * continue to mount and grab their data.
2279 if (ret && btrfs_test_opt(info, IGNOREBADROOTS))
2280 ret = fill_dummy_bgs(info);
2285 * This function, insert_block_group_item(), belongs to the phase 2 of chunk
2288 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation
2291 static int insert_block_group_item(struct btrfs_trans_handle *trans,
2292 struct btrfs_block_group *block_group)
2294 struct btrfs_fs_info *fs_info = trans->fs_info;
2295 struct btrfs_block_group_item bgi;
2296 struct btrfs_root *root = btrfs_block_group_root(fs_info);
2297 struct btrfs_key key;
2299 spin_lock(&block_group->lock);
2300 btrfs_set_stack_block_group_used(&bgi, block_group->used);
2301 btrfs_set_stack_block_group_chunk_objectid(&bgi,
2302 block_group->global_root_id);
2303 btrfs_set_stack_block_group_flags(&bgi, block_group->flags);
2304 key.objectid = block_group->start;
2305 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
2306 key.offset = block_group->length;
2307 spin_unlock(&block_group->lock);
2309 return btrfs_insert_item(trans, root, &key, &bgi, sizeof(bgi));
2312 static int insert_dev_extent(struct btrfs_trans_handle *trans,
2313 struct btrfs_device *device, u64 chunk_offset,
2314 u64 start, u64 num_bytes)
2316 struct btrfs_fs_info *fs_info = device->fs_info;
2317 struct btrfs_root *root = fs_info->dev_root;
2318 struct btrfs_path *path;
2319 struct btrfs_dev_extent *extent;
2320 struct extent_buffer *leaf;
2321 struct btrfs_key key;
2324 WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state));
2325 WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
2326 path = btrfs_alloc_path();
2330 key.objectid = device->devid;
2331 key.type = BTRFS_DEV_EXTENT_KEY;
2333 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*extent));
2337 leaf = path->nodes[0];
2338 extent = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_extent);
2339 btrfs_set_dev_extent_chunk_tree(leaf, extent, BTRFS_CHUNK_TREE_OBJECTID);
2340 btrfs_set_dev_extent_chunk_objectid(leaf, extent,
2341 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
2342 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
2344 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
2345 btrfs_mark_buffer_dirty(leaf);
2347 btrfs_free_path(path);
2352 * This function belongs to phase 2.
2354 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation
2357 static int insert_dev_extents(struct btrfs_trans_handle *trans,
2358 u64 chunk_offset, u64 chunk_size)
2360 struct btrfs_fs_info *fs_info = trans->fs_info;
2361 struct btrfs_device *device;
2362 struct extent_map *em;
2363 struct map_lookup *map;
2369 em = btrfs_get_chunk_map(fs_info, chunk_offset, chunk_size);
2373 map = em->map_lookup;
2374 stripe_size = em->orig_block_len;
2377 * Take the device list mutex to prevent races with the final phase of
2378 * a device replace operation that replaces the device object associated
2379 * with the map's stripes, because the device object's id can change
2380 * at any time during that final phase of the device replace operation
2381 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
2382 * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID,
2383 * resulting in persisting a device extent item with such ID.
2385 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2386 for (i = 0; i < map->num_stripes; i++) {
2387 device = map->stripes[i].dev;
2388 dev_offset = map->stripes[i].physical;
2390 ret = insert_dev_extent(trans, device, chunk_offset, dev_offset,
2395 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2397 free_extent_map(em);
2402 * This function, btrfs_create_pending_block_groups(), belongs to the phase 2 of
2405 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation
2408 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans)
2410 struct btrfs_fs_info *fs_info = trans->fs_info;
2411 struct btrfs_block_group *block_group;
2414 while (!list_empty(&trans->new_bgs)) {
2417 block_group = list_first_entry(&trans->new_bgs,
2418 struct btrfs_block_group,
2423 index = btrfs_bg_flags_to_raid_index(block_group->flags);
2425 ret = insert_block_group_item(trans, block_group);
2427 btrfs_abort_transaction(trans, ret);
2428 if (!block_group->chunk_item_inserted) {
2429 mutex_lock(&fs_info->chunk_mutex);
2430 ret = btrfs_chunk_alloc_add_chunk_item(trans, block_group);
2431 mutex_unlock(&fs_info->chunk_mutex);
2433 btrfs_abort_transaction(trans, ret);
2435 ret = insert_dev_extents(trans, block_group->start,
2436 block_group->length);
2438 btrfs_abort_transaction(trans, ret);
2439 add_block_group_free_space(trans, block_group);
2442 * If we restriped during balance, we may have added a new raid
2443 * type, so now add the sysfs entries when it is safe to do so.
2444 * We don't have to worry about locking here as it's handled in
2445 * btrfs_sysfs_add_block_group_type.
2447 if (block_group->space_info->block_group_kobjs[index] == NULL)
2448 btrfs_sysfs_add_block_group_type(block_group);
2450 /* Already aborted the transaction if it failed. */
2452 btrfs_delayed_refs_rsv_release(fs_info, 1);
2453 list_del_init(&block_group->bg_list);
2455 btrfs_trans_release_chunk_metadata(trans);
2459 * For extent tree v2 we use the block_group_item->chunk_offset to point at our
2460 * global root id. For v1 it's always set to BTRFS_FIRST_CHUNK_TREE_OBJECTID.
2462 static u64 calculate_global_root_id(struct btrfs_fs_info *fs_info, u64 offset)
2467 if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
2468 return BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2470 /* If we have a smaller fs index based on 128MiB. */
2471 if (btrfs_super_total_bytes(fs_info->super_copy) <= (SZ_1G * 10ULL))
2474 offset = div64_u64(offset, div);
2475 div64_u64_rem(offset, fs_info->nr_global_roots, &index);
2479 struct btrfs_block_group *btrfs_make_block_group(struct btrfs_trans_handle *trans,
2480 u64 bytes_used, u64 type,
2481 u64 chunk_offset, u64 size)
2483 struct btrfs_fs_info *fs_info = trans->fs_info;
2484 struct btrfs_block_group *cache;
2487 btrfs_set_log_full_commit(trans);
2489 cache = btrfs_create_block_group_cache(fs_info, chunk_offset);
2491 return ERR_PTR(-ENOMEM);
2493 cache->length = size;
2494 set_free_space_tree_thresholds(cache);
2495 cache->used = bytes_used;
2496 cache->flags = type;
2497 cache->last_byte_to_unpin = (u64)-1;
2498 cache->cached = BTRFS_CACHE_FINISHED;
2499 cache->global_root_id = calculate_global_root_id(fs_info, cache->start);
2501 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
2502 cache->needs_free_space = 1;
2504 ret = btrfs_load_block_group_zone_info(cache, true);
2506 btrfs_put_block_group(cache);
2507 return ERR_PTR(ret);
2510 ret = exclude_super_stripes(cache);
2512 /* We may have excluded something, so call this just in case */
2513 btrfs_free_excluded_extents(cache);
2514 btrfs_put_block_group(cache);
2515 return ERR_PTR(ret);
2518 add_new_free_space(cache, chunk_offset, chunk_offset + size);
2520 btrfs_free_excluded_extents(cache);
2522 #ifdef CONFIG_BTRFS_DEBUG
2523 if (btrfs_should_fragment_free_space(cache)) {
2524 u64 new_bytes_used = size - bytes_used;
2526 bytes_used += new_bytes_used >> 1;
2527 fragment_free_space(cache);
2531 * Ensure the corresponding space_info object is created and
2532 * assigned to our block group. We want our bg to be added to the rbtree
2533 * with its ->space_info set.
2535 cache->space_info = btrfs_find_space_info(fs_info, cache->flags);
2536 ASSERT(cache->space_info);
2538 ret = btrfs_add_block_group_cache(fs_info, cache);
2540 btrfs_remove_free_space_cache(cache);
2541 btrfs_put_block_group(cache);
2542 return ERR_PTR(ret);
2546 * Now that our block group has its ->space_info set and is inserted in
2547 * the rbtree, update the space info's counters.
2549 trace_btrfs_add_block_group(fs_info, cache, 1);
2550 btrfs_update_space_info(fs_info, cache->flags, size, bytes_used,
2551 cache->bytes_super, cache->zone_unusable,
2552 cache->zone_is_active, &cache->space_info);
2553 btrfs_update_global_block_rsv(fs_info);
2555 link_block_group(cache);
2557 list_add_tail(&cache->bg_list, &trans->new_bgs);
2558 trans->delayed_ref_updates++;
2559 btrfs_update_delayed_refs_rsv(trans);
2561 set_avail_alloc_bits(fs_info, type);
2566 * Mark one block group RO, can be called several times for the same block
2569 * @cache: the destination block group
2570 * @do_chunk_alloc: whether need to do chunk pre-allocation, this is to
2571 * ensure we still have some free space after marking this
2574 int btrfs_inc_block_group_ro(struct btrfs_block_group *cache,
2575 bool do_chunk_alloc)
2577 struct btrfs_fs_info *fs_info = cache->fs_info;
2578 struct btrfs_trans_handle *trans;
2579 struct btrfs_root *root = btrfs_block_group_root(fs_info);
2582 bool dirty_bg_running;
2585 * This can only happen when we are doing read-only scrub on read-only
2587 * In that case we should not start a new transaction on read-only fs.
2588 * Thus here we skip all chunk allocations.
2590 if (sb_rdonly(fs_info->sb)) {
2591 mutex_lock(&fs_info->ro_block_group_mutex);
2592 ret = inc_block_group_ro(cache, 0);
2593 mutex_unlock(&fs_info->ro_block_group_mutex);
2598 trans = btrfs_join_transaction(root);
2600 return PTR_ERR(trans);
2602 dirty_bg_running = false;
2605 * We're not allowed to set block groups readonly after the dirty
2606 * block group cache has started writing. If it already started,
2607 * back off and let this transaction commit.
2609 mutex_lock(&fs_info->ro_block_group_mutex);
2610 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
2611 u64 transid = trans->transid;
2613 mutex_unlock(&fs_info->ro_block_group_mutex);
2614 btrfs_end_transaction(trans);
2616 ret = btrfs_wait_for_commit(fs_info, transid);
2619 dirty_bg_running = true;
2621 } while (dirty_bg_running);
2623 if (do_chunk_alloc) {
2625 * If we are changing raid levels, try to allocate a
2626 * corresponding block group with the new raid level.
2628 alloc_flags = btrfs_get_alloc_profile(fs_info, cache->flags);
2629 if (alloc_flags != cache->flags) {
2630 ret = btrfs_chunk_alloc(trans, alloc_flags,
2633 * ENOSPC is allowed here, we may have enough space
2634 * already allocated at the new raid level to carry on
2643 ret = inc_block_group_ro(cache, 0);
2644 if (!do_chunk_alloc || ret == -ETXTBSY)
2648 alloc_flags = btrfs_get_alloc_profile(fs_info, cache->space_info->flags);
2649 ret = btrfs_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE);
2653 * We have allocated a new chunk. We also need to activate that chunk to
2654 * grant metadata tickets for zoned filesystem.
2656 ret = btrfs_zoned_activate_one_bg(fs_info, cache->space_info, true);
2660 ret = inc_block_group_ro(cache, 0);
2661 if (ret == -ETXTBSY)
2664 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
2665 alloc_flags = btrfs_get_alloc_profile(fs_info, cache->flags);
2666 mutex_lock(&fs_info->chunk_mutex);
2667 check_system_chunk(trans, alloc_flags);
2668 mutex_unlock(&fs_info->chunk_mutex);
2671 mutex_unlock(&fs_info->ro_block_group_mutex);
2673 btrfs_end_transaction(trans);
2677 void btrfs_dec_block_group_ro(struct btrfs_block_group *cache)
2679 struct btrfs_space_info *sinfo = cache->space_info;
2684 spin_lock(&sinfo->lock);
2685 spin_lock(&cache->lock);
2687 if (btrfs_is_zoned(cache->fs_info)) {
2688 /* Migrate zone_unusable bytes back */
2689 cache->zone_unusable =
2690 (cache->alloc_offset - cache->used) +
2691 (cache->length - cache->zone_capacity);
2692 sinfo->bytes_zone_unusable += cache->zone_unusable;
2693 sinfo->bytes_readonly -= cache->zone_unusable;
2695 num_bytes = cache->length - cache->reserved -
2696 cache->pinned - cache->bytes_super -
2697 cache->zone_unusable - cache->used;
2698 sinfo->bytes_readonly -= num_bytes;
2699 list_del_init(&cache->ro_list);
2701 spin_unlock(&cache->lock);
2702 spin_unlock(&sinfo->lock);
2705 static int update_block_group_item(struct btrfs_trans_handle *trans,
2706 struct btrfs_path *path,
2707 struct btrfs_block_group *cache)
2709 struct btrfs_fs_info *fs_info = trans->fs_info;
2711 struct btrfs_root *root = btrfs_block_group_root(fs_info);
2713 struct extent_buffer *leaf;
2714 struct btrfs_block_group_item bgi;
2715 struct btrfs_key key;
2717 key.objectid = cache->start;
2718 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
2719 key.offset = cache->length;
2721 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2728 leaf = path->nodes[0];
2729 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2730 btrfs_set_stack_block_group_used(&bgi, cache->used);
2731 btrfs_set_stack_block_group_chunk_objectid(&bgi,
2732 cache->global_root_id);
2733 btrfs_set_stack_block_group_flags(&bgi, cache->flags);
2734 write_extent_buffer(leaf, &bgi, bi, sizeof(bgi));
2735 btrfs_mark_buffer_dirty(leaf);
2737 btrfs_release_path(path);
2742 static int cache_save_setup(struct btrfs_block_group *block_group,
2743 struct btrfs_trans_handle *trans,
2744 struct btrfs_path *path)
2746 struct btrfs_fs_info *fs_info = block_group->fs_info;
2747 struct btrfs_root *root = fs_info->tree_root;
2748 struct inode *inode = NULL;
2749 struct extent_changeset *data_reserved = NULL;
2751 int dcs = BTRFS_DC_ERROR;
2756 if (!btrfs_test_opt(fs_info, SPACE_CACHE))
2760 * If this block group is smaller than 100 megs don't bother caching the
2763 if (block_group->length < (100 * SZ_1M)) {
2764 spin_lock(&block_group->lock);
2765 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2766 spin_unlock(&block_group->lock);
2770 if (TRANS_ABORTED(trans))
2773 inode = lookup_free_space_inode(block_group, path);
2774 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2775 ret = PTR_ERR(inode);
2776 btrfs_release_path(path);
2780 if (IS_ERR(inode)) {
2784 if (block_group->ro)
2787 ret = create_free_space_inode(trans, block_group, path);
2794 * We want to set the generation to 0, that way if anything goes wrong
2795 * from here on out we know not to trust this cache when we load up next
2798 BTRFS_I(inode)->generation = 0;
2799 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
2802 * So theoretically we could recover from this, simply set the
2803 * super cache generation to 0 so we know to invalidate the
2804 * cache, but then we'd have to keep track of the block groups
2805 * that fail this way so we know we _have_ to reset this cache
2806 * before the next commit or risk reading stale cache. So to
2807 * limit our exposure to horrible edge cases lets just abort the
2808 * transaction, this only happens in really bad situations
2811 btrfs_abort_transaction(trans, ret);
2816 /* We've already setup this transaction, go ahead and exit */
2817 if (block_group->cache_generation == trans->transid &&
2818 i_size_read(inode)) {
2819 dcs = BTRFS_DC_SETUP;
2823 if (i_size_read(inode) > 0) {
2824 ret = btrfs_check_trunc_cache_free_space(fs_info,
2825 &fs_info->global_block_rsv);
2829 ret = btrfs_truncate_free_space_cache(trans, NULL, inode);
2834 spin_lock(&block_group->lock);
2835 if (block_group->cached != BTRFS_CACHE_FINISHED ||
2836 !btrfs_test_opt(fs_info, SPACE_CACHE)) {
2838 * don't bother trying to write stuff out _if_
2839 * a) we're not cached,
2840 * b) we're with nospace_cache mount option,
2841 * c) we're with v2 space_cache (FREE_SPACE_TREE).
2843 dcs = BTRFS_DC_WRITTEN;
2844 spin_unlock(&block_group->lock);
2847 spin_unlock(&block_group->lock);
2850 * We hit an ENOSPC when setting up the cache in this transaction, just
2851 * skip doing the setup, we've already cleared the cache so we're safe.
2853 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
2859 * Try to preallocate enough space based on how big the block group is.
2860 * Keep in mind this has to include any pinned space which could end up
2861 * taking up quite a bit since it's not folded into the other space
2864 cache_size = div_u64(block_group->length, SZ_256M);
2869 cache_size *= fs_info->sectorsize;
2871 ret = btrfs_check_data_free_space(BTRFS_I(inode), &data_reserved, 0,
2876 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, cache_size,
2877 cache_size, cache_size,
2880 * Our cache requires contiguous chunks so that we don't modify a bunch
2881 * of metadata or split extents when writing the cache out, which means
2882 * we can enospc if we are heavily fragmented in addition to just normal
2883 * out of space conditions. So if we hit this just skip setting up any
2884 * other block groups for this transaction, maybe we'll unpin enough
2885 * space the next time around.
2888 dcs = BTRFS_DC_SETUP;
2889 else if (ret == -ENOSPC)
2890 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
2895 btrfs_release_path(path);
2897 spin_lock(&block_group->lock);
2898 if (!ret && dcs == BTRFS_DC_SETUP)
2899 block_group->cache_generation = trans->transid;
2900 block_group->disk_cache_state = dcs;
2901 spin_unlock(&block_group->lock);
2903 extent_changeset_free(data_reserved);
2907 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans)
2909 struct btrfs_fs_info *fs_info = trans->fs_info;
2910 struct btrfs_block_group *cache, *tmp;
2911 struct btrfs_transaction *cur_trans = trans->transaction;
2912 struct btrfs_path *path;
2914 if (list_empty(&cur_trans->dirty_bgs) ||
2915 !btrfs_test_opt(fs_info, SPACE_CACHE))
2918 path = btrfs_alloc_path();
2922 /* Could add new block groups, use _safe just in case */
2923 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
2925 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2926 cache_save_setup(cache, trans, path);
2929 btrfs_free_path(path);
2934 * Transaction commit does final block group cache writeback during a critical
2935 * section where nothing is allowed to change the FS. This is required in
2936 * order for the cache to actually match the block group, but can introduce a
2937 * lot of latency into the commit.
2939 * So, btrfs_start_dirty_block_groups is here to kick off block group cache IO.
2940 * There's a chance we'll have to redo some of it if the block group changes
2941 * again during the commit, but it greatly reduces the commit latency by
2942 * getting rid of the easy block groups while we're still allowing others to
2945 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans)
2947 struct btrfs_fs_info *fs_info = trans->fs_info;
2948 struct btrfs_block_group *cache;
2949 struct btrfs_transaction *cur_trans = trans->transaction;
2952 struct btrfs_path *path = NULL;
2954 struct list_head *io = &cur_trans->io_bgs;
2957 spin_lock(&cur_trans->dirty_bgs_lock);
2958 if (list_empty(&cur_trans->dirty_bgs)) {
2959 spin_unlock(&cur_trans->dirty_bgs_lock);
2962 list_splice_init(&cur_trans->dirty_bgs, &dirty);
2963 spin_unlock(&cur_trans->dirty_bgs_lock);
2966 /* Make sure all the block groups on our dirty list actually exist */
2967 btrfs_create_pending_block_groups(trans);
2970 path = btrfs_alloc_path();
2978 * cache_write_mutex is here only to save us from balance or automatic
2979 * removal of empty block groups deleting this block group while we are
2980 * writing out the cache
2982 mutex_lock(&trans->transaction->cache_write_mutex);
2983 while (!list_empty(&dirty)) {
2984 bool drop_reserve = true;
2986 cache = list_first_entry(&dirty, struct btrfs_block_group,
2989 * This can happen if something re-dirties a block group that
2990 * is already under IO. Just wait for it to finish and then do
2993 if (!list_empty(&cache->io_list)) {
2994 list_del_init(&cache->io_list);
2995 btrfs_wait_cache_io(trans, cache, path);
2996 btrfs_put_block_group(cache);
3001 * btrfs_wait_cache_io uses the cache->dirty_list to decide if
3002 * it should update the cache_state. Don't delete until after
3005 * Since we're not running in the commit critical section
3006 * we need the dirty_bgs_lock to protect from update_block_group
3008 spin_lock(&cur_trans->dirty_bgs_lock);
3009 list_del_init(&cache->dirty_list);
3010 spin_unlock(&cur_trans->dirty_bgs_lock);
3014 cache_save_setup(cache, trans, path);
3016 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3017 cache->io_ctl.inode = NULL;
3018 ret = btrfs_write_out_cache(trans, cache, path);
3019 if (ret == 0 && cache->io_ctl.inode) {
3023 * The cache_write_mutex is protecting the
3024 * io_list, also refer to the definition of
3025 * btrfs_transaction::io_bgs for more details
3027 list_add_tail(&cache->io_list, io);
3030 * If we failed to write the cache, the
3031 * generation will be bad and life goes on
3037 ret = update_block_group_item(trans, path, cache);
3039 * Our block group might still be attached to the list
3040 * of new block groups in the transaction handle of some
3041 * other task (struct btrfs_trans_handle->new_bgs). This
3042 * means its block group item isn't yet in the extent
3043 * tree. If this happens ignore the error, as we will
3044 * try again later in the critical section of the
3045 * transaction commit.
3047 if (ret == -ENOENT) {
3049 spin_lock(&cur_trans->dirty_bgs_lock);
3050 if (list_empty(&cache->dirty_list)) {
3051 list_add_tail(&cache->dirty_list,
3052 &cur_trans->dirty_bgs);
3053 btrfs_get_block_group(cache);
3054 drop_reserve = false;
3056 spin_unlock(&cur_trans->dirty_bgs_lock);
3058 btrfs_abort_transaction(trans, ret);
3062 /* If it's not on the io list, we need to put the block group */
3064 btrfs_put_block_group(cache);
3066 btrfs_delayed_refs_rsv_release(fs_info, 1);
3068 * Avoid blocking other tasks for too long. It might even save
3069 * us from writing caches for block groups that are going to be
3072 mutex_unlock(&trans->transaction->cache_write_mutex);
3075 mutex_lock(&trans->transaction->cache_write_mutex);
3077 mutex_unlock(&trans->transaction->cache_write_mutex);
3080 * Go through delayed refs for all the stuff we've just kicked off
3081 * and then loop back (just once)
3084 ret = btrfs_run_delayed_refs(trans, 0);
3085 if (!ret && loops == 0) {
3087 spin_lock(&cur_trans->dirty_bgs_lock);
3088 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3090 * dirty_bgs_lock protects us from concurrent block group
3091 * deletes too (not just cache_write_mutex).
3093 if (!list_empty(&dirty)) {
3094 spin_unlock(&cur_trans->dirty_bgs_lock);
3097 spin_unlock(&cur_trans->dirty_bgs_lock);
3101 spin_lock(&cur_trans->dirty_bgs_lock);
3102 list_splice_init(&dirty, &cur_trans->dirty_bgs);
3103 spin_unlock(&cur_trans->dirty_bgs_lock);
3104 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
3107 btrfs_free_path(path);
3111 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans)
3113 struct btrfs_fs_info *fs_info = trans->fs_info;
3114 struct btrfs_block_group *cache;
3115 struct btrfs_transaction *cur_trans = trans->transaction;
3118 struct btrfs_path *path;
3119 struct list_head *io = &cur_trans->io_bgs;
3121 path = btrfs_alloc_path();
3126 * Even though we are in the critical section of the transaction commit,
3127 * we can still have concurrent tasks adding elements to this
3128 * transaction's list of dirty block groups. These tasks correspond to
3129 * endio free space workers started when writeback finishes for a
3130 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3131 * allocate new block groups as a result of COWing nodes of the root
3132 * tree when updating the free space inode. The writeback for the space
3133 * caches is triggered by an earlier call to
3134 * btrfs_start_dirty_block_groups() and iterations of the following
3136 * Also we want to do the cache_save_setup first and then run the
3137 * delayed refs to make sure we have the best chance at doing this all
3140 spin_lock(&cur_trans->dirty_bgs_lock);
3141 while (!list_empty(&cur_trans->dirty_bgs)) {
3142 cache = list_first_entry(&cur_trans->dirty_bgs,
3143 struct btrfs_block_group,
3147 * This can happen if cache_save_setup re-dirties a block group
3148 * that is already under IO. Just wait for it to finish and
3149 * then do it all again
3151 if (!list_empty(&cache->io_list)) {
3152 spin_unlock(&cur_trans->dirty_bgs_lock);
3153 list_del_init(&cache->io_list);
3154 btrfs_wait_cache_io(trans, cache, path);
3155 btrfs_put_block_group(cache);
3156 spin_lock(&cur_trans->dirty_bgs_lock);
3160 * Don't remove from the dirty list until after we've waited on
3163 list_del_init(&cache->dirty_list);
3164 spin_unlock(&cur_trans->dirty_bgs_lock);
3167 cache_save_setup(cache, trans, path);
3170 ret = btrfs_run_delayed_refs(trans,
3171 (unsigned long) -1);
3173 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3174 cache->io_ctl.inode = NULL;
3175 ret = btrfs_write_out_cache(trans, cache, path);
3176 if (ret == 0 && cache->io_ctl.inode) {
3178 list_add_tail(&cache->io_list, io);
3181 * If we failed to write the cache, the
3182 * generation will be bad and life goes on
3188 ret = update_block_group_item(trans, path, cache);
3190 * One of the free space endio workers might have
3191 * created a new block group while updating a free space
3192 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3193 * and hasn't released its transaction handle yet, in
3194 * which case the new block group is still attached to
3195 * its transaction handle and its creation has not
3196 * finished yet (no block group item in the extent tree
3197 * yet, etc). If this is the case, wait for all free
3198 * space endio workers to finish and retry. This is a
3199 * very rare case so no need for a more efficient and
3202 if (ret == -ENOENT) {
3203 wait_event(cur_trans->writer_wait,
3204 atomic_read(&cur_trans->num_writers) == 1);
3205 ret = update_block_group_item(trans, path, cache);
3208 btrfs_abort_transaction(trans, ret);
3211 /* If its not on the io list, we need to put the block group */
3213 btrfs_put_block_group(cache);
3214 btrfs_delayed_refs_rsv_release(fs_info, 1);
3215 spin_lock(&cur_trans->dirty_bgs_lock);
3217 spin_unlock(&cur_trans->dirty_bgs_lock);
3220 * Refer to the definition of io_bgs member for details why it's safe
3221 * to use it without any locking
3223 while (!list_empty(io)) {
3224 cache = list_first_entry(io, struct btrfs_block_group,
3226 list_del_init(&cache->io_list);
3227 btrfs_wait_cache_io(trans, cache, path);
3228 btrfs_put_block_group(cache);
3231 btrfs_free_path(path);
3235 static inline bool should_reclaim_block_group(struct btrfs_block_group *bg,
3238 const struct btrfs_space_info *space_info = bg->space_info;
3239 const int reclaim_thresh = READ_ONCE(space_info->bg_reclaim_threshold);
3240 const u64 new_val = bg->used;
3241 const u64 old_val = new_val + bytes_freed;
3244 if (reclaim_thresh == 0)
3247 thresh = div_factor_fine(bg->length, reclaim_thresh);
3250 * If we were below the threshold before don't reclaim, we are likely a
3251 * brand new block group and we don't want to relocate new block groups.
3253 if (old_val < thresh)
3255 if (new_val >= thresh)
3260 int btrfs_update_block_group(struct btrfs_trans_handle *trans,
3261 u64 bytenr, u64 num_bytes, bool alloc)
3263 struct btrfs_fs_info *info = trans->fs_info;
3264 struct btrfs_block_group *cache = NULL;
3265 u64 total = num_bytes;
3271 /* Block accounting for super block */
3272 spin_lock(&info->delalloc_root_lock);
3273 old_val = btrfs_super_bytes_used(info->super_copy);
3275 old_val += num_bytes;
3277 old_val -= num_bytes;
3278 btrfs_set_super_bytes_used(info->super_copy, old_val);
3279 spin_unlock(&info->delalloc_root_lock);
3284 cache = btrfs_lookup_block_group(info, bytenr);
3289 factor = btrfs_bg_type_to_factor(cache->flags);
3292 * If this block group has free space cache written out, we
3293 * need to make sure to load it if we are removing space. This
3294 * is because we need the unpinning stage to actually add the
3295 * space back to the block group, otherwise we will leak space.
3297 if (!alloc && !btrfs_block_group_done(cache))
3298 btrfs_cache_block_group(cache, true);
3300 byte_in_group = bytenr - cache->start;
3301 WARN_ON(byte_in_group > cache->length);
3303 spin_lock(&cache->space_info->lock);
3304 spin_lock(&cache->lock);
3306 if (btrfs_test_opt(info, SPACE_CACHE) &&
3307 cache->disk_cache_state < BTRFS_DC_CLEAR)
3308 cache->disk_cache_state = BTRFS_DC_CLEAR;
3310 old_val = cache->used;
3311 num_bytes = min(total, cache->length - byte_in_group);
3313 old_val += num_bytes;
3314 cache->used = old_val;
3315 cache->reserved -= num_bytes;
3316 cache->space_info->bytes_reserved -= num_bytes;
3317 cache->space_info->bytes_used += num_bytes;
3318 cache->space_info->disk_used += num_bytes * factor;
3319 spin_unlock(&cache->lock);
3320 spin_unlock(&cache->space_info->lock);
3322 old_val -= num_bytes;
3323 cache->used = old_val;
3324 cache->pinned += num_bytes;
3325 btrfs_space_info_update_bytes_pinned(info,
3326 cache->space_info, num_bytes);
3327 cache->space_info->bytes_used -= num_bytes;
3328 cache->space_info->disk_used -= num_bytes * factor;
3330 reclaim = should_reclaim_block_group(cache, num_bytes);
3331 spin_unlock(&cache->lock);
3332 spin_unlock(&cache->space_info->lock);
3334 set_extent_dirty(&trans->transaction->pinned_extents,
3335 bytenr, bytenr + num_bytes - 1,
3336 GFP_NOFS | __GFP_NOFAIL);
3339 spin_lock(&trans->transaction->dirty_bgs_lock);
3340 if (list_empty(&cache->dirty_list)) {
3341 list_add_tail(&cache->dirty_list,
3342 &trans->transaction->dirty_bgs);
3343 trans->delayed_ref_updates++;
3344 btrfs_get_block_group(cache);
3346 spin_unlock(&trans->transaction->dirty_bgs_lock);
3349 * No longer have used bytes in this block group, queue it for
3350 * deletion. We do this after adding the block group to the
3351 * dirty list to avoid races between cleaner kthread and space
3354 if (!alloc && old_val == 0) {
3355 if (!btrfs_test_opt(info, DISCARD_ASYNC))
3356 btrfs_mark_bg_unused(cache);
3357 } else if (!alloc && reclaim) {
3358 btrfs_mark_bg_to_reclaim(cache);
3361 btrfs_put_block_group(cache);
3363 bytenr += num_bytes;
3366 /* Modified block groups are accounted for in the delayed_refs_rsv. */
3367 btrfs_update_delayed_refs_rsv(trans);
3372 * btrfs_add_reserved_bytes - update the block_group and space info counters
3373 * @cache: The cache we are manipulating
3374 * @ram_bytes: The number of bytes of file content, and will be same to
3375 * @num_bytes except for the compress path.
3376 * @num_bytes: The number of bytes in question
3377 * @delalloc: The blocks are allocated for the delalloc write
3379 * This is called by the allocator when it reserves space. If this is a
3380 * reservation and the block group has become read only we cannot make the
3381 * reservation and return -EAGAIN, otherwise this function always succeeds.
3383 int btrfs_add_reserved_bytes(struct btrfs_block_group *cache,
3384 u64 ram_bytes, u64 num_bytes, int delalloc)
3386 struct btrfs_space_info *space_info = cache->space_info;
3389 spin_lock(&space_info->lock);
3390 spin_lock(&cache->lock);
3394 cache->reserved += num_bytes;
3395 space_info->bytes_reserved += num_bytes;
3396 trace_btrfs_space_reservation(cache->fs_info, "space_info",
3397 space_info->flags, num_bytes, 1);
3398 btrfs_space_info_update_bytes_may_use(cache->fs_info,
3399 space_info, -ram_bytes);
3401 cache->delalloc_bytes += num_bytes;
3404 * Compression can use less space than we reserved, so wake
3405 * tickets if that happens
3407 if (num_bytes < ram_bytes)
3408 btrfs_try_granting_tickets(cache->fs_info, space_info);
3410 spin_unlock(&cache->lock);
3411 spin_unlock(&space_info->lock);
3416 * btrfs_free_reserved_bytes - update the block_group and space info counters
3417 * @cache: The cache we are manipulating
3418 * @num_bytes: The number of bytes in question
3419 * @delalloc: The blocks are allocated for the delalloc write
3421 * This is called by somebody who is freeing space that was never actually used
3422 * on disk. For example if you reserve some space for a new leaf in transaction
3423 * A and before transaction A commits you free that leaf, you call this with
3424 * reserve set to 0 in order to clear the reservation.
3426 void btrfs_free_reserved_bytes(struct btrfs_block_group *cache,
3427 u64 num_bytes, int delalloc)
3429 struct btrfs_space_info *space_info = cache->space_info;
3431 spin_lock(&space_info->lock);
3432 spin_lock(&cache->lock);
3434 space_info->bytes_readonly += num_bytes;
3435 cache->reserved -= num_bytes;
3436 space_info->bytes_reserved -= num_bytes;
3437 space_info->max_extent_size = 0;
3440 cache->delalloc_bytes -= num_bytes;
3441 spin_unlock(&cache->lock);
3443 btrfs_try_granting_tickets(cache->fs_info, space_info);
3444 spin_unlock(&space_info->lock);
3447 static void force_metadata_allocation(struct btrfs_fs_info *info)
3449 struct list_head *head = &info->space_info;
3450 struct btrfs_space_info *found;
3452 list_for_each_entry(found, head, list) {
3453 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3454 found->force_alloc = CHUNK_ALLOC_FORCE;
3458 static int should_alloc_chunk(struct btrfs_fs_info *fs_info,
3459 struct btrfs_space_info *sinfo, int force)
3461 u64 bytes_used = btrfs_space_info_used(sinfo, false);
3464 if (force == CHUNK_ALLOC_FORCE)
3468 * in limited mode, we want to have some free space up to
3469 * about 1% of the FS size.
3471 if (force == CHUNK_ALLOC_LIMITED) {
3472 thresh = btrfs_super_total_bytes(fs_info->super_copy);
3473 thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1));
3475 if (sinfo->total_bytes - bytes_used < thresh)
3479 if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8))
3484 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, u64 type)
3486 u64 alloc_flags = btrfs_get_alloc_profile(trans->fs_info, type);
3488 return btrfs_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE);
3491 static struct btrfs_block_group *do_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags)
3493 struct btrfs_block_group *bg;
3497 * Check if we have enough space in the system space info because we
3498 * will need to update device items in the chunk btree and insert a new
3499 * chunk item in the chunk btree as well. This will allocate a new
3500 * system block group if needed.
3502 check_system_chunk(trans, flags);
3504 bg = btrfs_create_chunk(trans, flags);
3510 ret = btrfs_chunk_alloc_add_chunk_item(trans, bg);
3512 * Normally we are not expected to fail with -ENOSPC here, since we have
3513 * previously reserved space in the system space_info and allocated one
3514 * new system chunk if necessary. However there are three exceptions:
3516 * 1) We may have enough free space in the system space_info but all the
3517 * existing system block groups have a profile which can not be used
3518 * for extent allocation.
3520 * This happens when mounting in degraded mode. For example we have a
3521 * RAID1 filesystem with 2 devices, lose one device and mount the fs
3522 * using the other device in degraded mode. If we then allocate a chunk,
3523 * we may have enough free space in the existing system space_info, but
3524 * none of the block groups can be used for extent allocation since they
3525 * have a RAID1 profile, and because we are in degraded mode with a
3526 * single device, we are forced to allocate a new system chunk with a
3527 * SINGLE profile. Making check_system_chunk() iterate over all system
3528 * block groups and check if they have a usable profile and enough space
3529 * can be slow on very large filesystems, so we tolerate the -ENOSPC and
3530 * try again after forcing allocation of a new system chunk. Like this
3531 * we avoid paying the cost of that search in normal circumstances, when
3532 * we were not mounted in degraded mode;
3534 * 2) We had enough free space info the system space_info, and one suitable
3535 * block group to allocate from when we called check_system_chunk()
3536 * above. However right after we called it, the only system block group
3537 * with enough free space got turned into RO mode by a running scrub,
3538 * and in this case we have to allocate a new one and retry. We only
3539 * need do this allocate and retry once, since we have a transaction
3540 * handle and scrub uses the commit root to search for block groups;
3542 * 3) We had one system block group with enough free space when we called
3543 * check_system_chunk(), but after that, right before we tried to
3544 * allocate the last extent buffer we needed, a discard operation came
3545 * in and it temporarily removed the last free space entry from the
3546 * block group (discard removes a free space entry, discards it, and
3547 * then adds back the entry to the block group cache).
3549 if (ret == -ENOSPC) {
3550 const u64 sys_flags = btrfs_system_alloc_profile(trans->fs_info);
3551 struct btrfs_block_group *sys_bg;
3553 sys_bg = btrfs_create_chunk(trans, sys_flags);
3554 if (IS_ERR(sys_bg)) {
3555 ret = PTR_ERR(sys_bg);
3556 btrfs_abort_transaction(trans, ret);
3560 ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg);
3562 btrfs_abort_transaction(trans, ret);
3566 ret = btrfs_chunk_alloc_add_chunk_item(trans, bg);
3568 btrfs_abort_transaction(trans, ret);
3572 btrfs_abort_transaction(trans, ret);
3576 btrfs_trans_release_chunk_metadata(trans);
3579 return ERR_PTR(ret);
3581 btrfs_get_block_group(bg);
3586 * Chunk allocation is done in 2 phases:
3588 * 1) Phase 1 - through btrfs_chunk_alloc() we allocate device extents for
3589 * the chunk, the chunk mapping, create its block group and add the items
3590 * that belong in the chunk btree to it - more specifically, we need to
3591 * update device items in the chunk btree and add a new chunk item to it.
3593 * 2) Phase 2 - through btrfs_create_pending_block_groups(), we add the block
3594 * group item to the extent btree and the device extent items to the devices
3597 * This is done to prevent deadlocks. For example when COWing a node from the
3598 * extent btree we are holding a write lock on the node's parent and if we
3599 * trigger chunk allocation and attempted to insert the new block group item
3600 * in the extent btree right way, we could deadlock because the path for the
3601 * insertion can include that parent node. At first glance it seems impossible
3602 * to trigger chunk allocation after starting a transaction since tasks should
3603 * reserve enough transaction units (metadata space), however while that is true
3604 * most of the time, chunk allocation may still be triggered for several reasons:
3606 * 1) When reserving metadata, we check if there is enough free space in the
3607 * metadata space_info and therefore don't trigger allocation of a new chunk.
3608 * However later when the task actually tries to COW an extent buffer from
3609 * the extent btree or from the device btree for example, it is forced to
3610 * allocate a new block group (chunk) because the only one that had enough
3611 * free space was just turned to RO mode by a running scrub for example (or
3612 * device replace, block group reclaim thread, etc), so we can not use it
3613 * for allocating an extent and end up being forced to allocate a new one;
3615 * 2) Because we only check that the metadata space_info has enough free bytes,
3616 * we end up not allocating a new metadata chunk in that case. However if
3617 * the filesystem was mounted in degraded mode, none of the existing block
3618 * groups might be suitable for extent allocation due to their incompatible
3619 * profile (for e.g. mounting a 2 devices filesystem, where all block groups
3620 * use a RAID1 profile, in degraded mode using a single device). In this case
3621 * when the task attempts to COW some extent buffer of the extent btree for
3622 * example, it will trigger allocation of a new metadata block group with a
3623 * suitable profile (SINGLE profile in the example of the degraded mount of
3624 * the RAID1 filesystem);
3626 * 3) The task has reserved enough transaction units / metadata space, but when
3627 * it attempts to COW an extent buffer from the extent or device btree for
3628 * example, it does not find any free extent in any metadata block group,
3629 * therefore forced to try to allocate a new metadata block group.
3630 * This is because some other task allocated all available extents in the
3631 * meanwhile - this typically happens with tasks that don't reserve space
3632 * properly, either intentionally or as a bug. One example where this is
3633 * done intentionally is fsync, as it does not reserve any transaction units
3634 * and ends up allocating a variable number of metadata extents for log
3635 * tree extent buffers;
3637 * 4) The task has reserved enough transaction units / metadata space, but right
3638 * before it tries to allocate the last extent buffer it needs, a discard
3639 * operation comes in and, temporarily, removes the last free space entry from
3640 * the only metadata block group that had free space (discard starts by
3641 * removing a free space entry from a block group, then does the discard
3642 * operation and, once it's done, it adds back the free space entry to the
3645 * We also need this 2 phases setup when adding a device to a filesystem with
3646 * a seed device - we must create new metadata and system chunks without adding
3647 * any of the block group items to the chunk, extent and device btrees. If we
3648 * did not do it this way, we would get ENOSPC when attempting to update those
3649 * btrees, since all the chunks from the seed device are read-only.
3651 * Phase 1 does the updates and insertions to the chunk btree because if we had
3652 * it done in phase 2 and have a thundering herd of tasks allocating chunks in
3653 * parallel, we risk having too many system chunks allocated by many tasks if
3654 * many tasks reach phase 1 without the previous ones completing phase 2. In the
3655 * extreme case this leads to exhaustion of the system chunk array in the
3656 * superblock. This is easier to trigger if using a btree node/leaf size of 64K
3657 * and with RAID filesystems (so we have more device items in the chunk btree).
3658 * This has happened before and commit eafa4fd0ad0607 ("btrfs: fix exhaustion of
3659 * the system chunk array due to concurrent allocations") provides more details.
3661 * Allocation of system chunks does not happen through this function. A task that
3662 * needs to update the chunk btree (the only btree that uses system chunks), must
3663 * preallocate chunk space by calling either check_system_chunk() or
3664 * btrfs_reserve_chunk_metadata() - the former is used when allocating a data or
3665 * metadata chunk or when removing a chunk, while the later is used before doing
3666 * a modification to the chunk btree - use cases for the later are adding,
3667 * removing and resizing a device as well as relocation of a system chunk.
3668 * See the comment below for more details.
3670 * The reservation of system space, done through check_system_chunk(), as well
3671 * as all the updates and insertions into the chunk btree must be done while
3672 * holding fs_info->chunk_mutex. This is important to guarantee that while COWing
3673 * an extent buffer from the chunks btree we never trigger allocation of a new
3674 * system chunk, which would result in a deadlock (trying to lock twice an
3675 * extent buffer of the chunk btree, first time before triggering the chunk
3676 * allocation and the second time during chunk allocation while attempting to
3677 * update the chunks btree). The system chunk array is also updated while holding
3678 * that mutex. The same logic applies to removing chunks - we must reserve system
3679 * space, update the chunk btree and the system chunk array in the superblock
3680 * while holding fs_info->chunk_mutex.
3682 * This function, btrfs_chunk_alloc(), belongs to phase 1.
3684 * If @force is CHUNK_ALLOC_FORCE:
3685 * - return 1 if it successfully allocates a chunk,
3686 * - return errors including -ENOSPC otherwise.
3687 * If @force is NOT CHUNK_ALLOC_FORCE:
3688 * - return 0 if it doesn't need to allocate a new chunk,
3689 * - return 1 if it successfully allocates a chunk,
3690 * - return errors including -ENOSPC otherwise.
3692 int btrfs_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags,
3693 enum btrfs_chunk_alloc_enum force)
3695 struct btrfs_fs_info *fs_info = trans->fs_info;
3696 struct btrfs_space_info *space_info;
3697 struct btrfs_block_group *ret_bg;
3698 bool wait_for_alloc = false;
3699 bool should_alloc = false;
3700 bool from_extent_allocation = false;
3703 if (force == CHUNK_ALLOC_FORCE_FOR_EXTENT) {
3704 from_extent_allocation = true;
3705 force = CHUNK_ALLOC_FORCE;
3708 /* Don't re-enter if we're already allocating a chunk */
3709 if (trans->allocating_chunk)
3712 * Allocation of system chunks can not happen through this path, as we
3713 * could end up in a deadlock if we are allocating a data or metadata
3714 * chunk and there is another task modifying the chunk btree.
3716 * This is because while we are holding the chunk mutex, we will attempt
3717 * to add the new chunk item to the chunk btree or update an existing
3718 * device item in the chunk btree, while the other task that is modifying
3719 * the chunk btree is attempting to COW an extent buffer while holding a
3720 * lock on it and on its parent - if the COW operation triggers a system
3721 * chunk allocation, then we can deadlock because we are holding the
3722 * chunk mutex and we may need to access that extent buffer or its parent
3723 * in order to add the chunk item or update a device item.
3725 * Tasks that want to modify the chunk tree should reserve system space
3726 * before updating the chunk btree, by calling either
3727 * btrfs_reserve_chunk_metadata() or check_system_chunk().
3728 * It's possible that after a task reserves the space, it still ends up
3729 * here - this happens in the cases described above at do_chunk_alloc().
3730 * The task will have to either retry or fail.
3732 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3735 space_info = btrfs_find_space_info(fs_info, flags);
3739 spin_lock(&space_info->lock);
3740 if (force < space_info->force_alloc)
3741 force = space_info->force_alloc;
3742 should_alloc = should_alloc_chunk(fs_info, space_info, force);
3743 if (space_info->full) {
3744 /* No more free physical space */
3749 spin_unlock(&space_info->lock);
3751 } else if (!should_alloc) {
3752 spin_unlock(&space_info->lock);
3754 } else if (space_info->chunk_alloc) {
3756 * Someone is already allocating, so we need to block
3757 * until this someone is finished and then loop to
3758 * recheck if we should continue with our allocation
3761 wait_for_alloc = true;
3762 force = CHUNK_ALLOC_NO_FORCE;
3763 spin_unlock(&space_info->lock);
3764 mutex_lock(&fs_info->chunk_mutex);
3765 mutex_unlock(&fs_info->chunk_mutex);
3767 /* Proceed with allocation */
3768 space_info->chunk_alloc = 1;
3769 wait_for_alloc = false;
3770 spin_unlock(&space_info->lock);
3774 } while (wait_for_alloc);
3776 mutex_lock(&fs_info->chunk_mutex);
3777 trans->allocating_chunk = true;
3780 * If we have mixed data/metadata chunks we want to make sure we keep
3781 * allocating mixed chunks instead of individual chunks.
3783 if (btrfs_mixed_space_info(space_info))
3784 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3787 * if we're doing a data chunk, go ahead and make sure that
3788 * we keep a reasonable number of metadata chunks allocated in the
3791 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3792 fs_info->data_chunk_allocations++;
3793 if (!(fs_info->data_chunk_allocations %
3794 fs_info->metadata_ratio))
3795 force_metadata_allocation(fs_info);
3798 ret_bg = do_chunk_alloc(trans, flags);
3799 trans->allocating_chunk = false;
3801 if (IS_ERR(ret_bg)) {
3802 ret = PTR_ERR(ret_bg);
3803 } else if (from_extent_allocation) {
3805 * New block group is likely to be used soon. Try to activate
3806 * it now. Failure is OK for now.
3808 btrfs_zone_activate(ret_bg);
3812 btrfs_put_block_group(ret_bg);
3814 spin_lock(&space_info->lock);
3817 space_info->full = 1;
3822 space_info->max_extent_size = 0;
3825 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3827 space_info->chunk_alloc = 0;
3828 spin_unlock(&space_info->lock);
3829 mutex_unlock(&fs_info->chunk_mutex);
3834 static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type)
3838 num_dev = btrfs_raid_array[btrfs_bg_flags_to_raid_index(type)].devs_max;
3840 num_dev = fs_info->fs_devices->rw_devices;
3845 static void reserve_chunk_space(struct btrfs_trans_handle *trans,
3849 struct btrfs_fs_info *fs_info = trans->fs_info;
3850 struct btrfs_space_info *info;
3855 * Needed because we can end up allocating a system chunk and for an
3856 * atomic and race free space reservation in the chunk block reserve.
3858 lockdep_assert_held(&fs_info->chunk_mutex);
3860 info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3861 spin_lock(&info->lock);
3862 left = info->total_bytes - btrfs_space_info_used(info, true);
3863 spin_unlock(&info->lock);
3865 if (left < bytes && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
3866 btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu",
3868 btrfs_dump_space_info(fs_info, info, 0, 0);
3872 u64 flags = btrfs_system_alloc_profile(fs_info);
3873 struct btrfs_block_group *bg;
3876 * Ignore failure to create system chunk. We might end up not
3877 * needing it, as we might not need to COW all nodes/leafs from
3878 * the paths we visit in the chunk tree (they were already COWed
3879 * or created in the current transaction for example).
3881 bg = btrfs_create_chunk(trans, flags);
3886 * We have a new chunk. We also need to activate it for
3889 ret = btrfs_zoned_activate_one_bg(fs_info, info, true);
3894 * If we fail to add the chunk item here, we end up
3895 * trying again at phase 2 of chunk allocation, at
3896 * btrfs_create_pending_block_groups(). So ignore
3897 * any error here. An ENOSPC here could happen, due to
3898 * the cases described at do_chunk_alloc() - the system
3899 * block group we just created was just turned into RO
3900 * mode by a scrub for example, or a running discard
3901 * temporarily removed its free space entries, etc.
3903 btrfs_chunk_alloc_add_chunk_item(trans, bg);
3908 ret = btrfs_block_rsv_add(fs_info,
3909 &fs_info->chunk_block_rsv,
3910 bytes, BTRFS_RESERVE_NO_FLUSH);
3912 trans->chunk_bytes_reserved += bytes;
3917 * Reserve space in the system space for allocating or removing a chunk.
3918 * The caller must be holding fs_info->chunk_mutex.
3920 void check_system_chunk(struct btrfs_trans_handle *trans, u64 type)
3922 struct btrfs_fs_info *fs_info = trans->fs_info;
3923 const u64 num_devs = get_profile_num_devs(fs_info, type);
3926 /* num_devs device items to update and 1 chunk item to add or remove. */
3927 bytes = btrfs_calc_metadata_size(fs_info, num_devs) +
3928 btrfs_calc_insert_metadata_size(fs_info, 1);
3930 reserve_chunk_space(trans, bytes, type);
3934 * Reserve space in the system space, if needed, for doing a modification to the
3937 * @trans: A transaction handle.
3938 * @is_item_insertion: Indicate if the modification is for inserting a new item
3939 * in the chunk btree or if it's for the deletion or update
3940 * of an existing item.
3942 * This is used in a context where we need to update the chunk btree outside
3943 * block group allocation and removal, to avoid a deadlock with a concurrent
3944 * task that is allocating a metadata or data block group and therefore needs to
3945 * update the chunk btree while holding the chunk mutex. After the update to the
3946 * chunk btree is done, btrfs_trans_release_chunk_metadata() should be called.
3949 void btrfs_reserve_chunk_metadata(struct btrfs_trans_handle *trans,
3950 bool is_item_insertion)
3952 struct btrfs_fs_info *fs_info = trans->fs_info;
3955 if (is_item_insertion)
3956 bytes = btrfs_calc_insert_metadata_size(fs_info, 1);
3958 bytes = btrfs_calc_metadata_size(fs_info, 1);
3960 mutex_lock(&fs_info->chunk_mutex);
3961 reserve_chunk_space(trans, bytes, BTRFS_BLOCK_GROUP_SYSTEM);
3962 mutex_unlock(&fs_info->chunk_mutex);
3965 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
3967 struct btrfs_block_group *block_group;
3971 struct inode *inode;
3973 block_group = btrfs_lookup_first_block_group(info, last);
3974 while (block_group) {
3975 btrfs_wait_block_group_cache_done(block_group);
3976 spin_lock(&block_group->lock);
3977 if (block_group->iref)
3979 spin_unlock(&block_group->lock);
3980 block_group = btrfs_next_block_group(block_group);
3989 inode = block_group->inode;
3990 block_group->iref = 0;
3991 block_group->inode = NULL;
3992 spin_unlock(&block_group->lock);
3993 ASSERT(block_group->io_ctl.inode == NULL);
3995 last = block_group->start + block_group->length;
3996 btrfs_put_block_group(block_group);
4001 * Must be called only after stopping all workers, since we could have block
4002 * group caching kthreads running, and therefore they could race with us if we
4003 * freed the block groups before stopping them.
4005 int btrfs_free_block_groups(struct btrfs_fs_info *info)
4007 struct btrfs_block_group *block_group;
4008 struct btrfs_space_info *space_info;
4009 struct btrfs_caching_control *caching_ctl;
4012 write_lock(&info->block_group_cache_lock);
4013 while (!list_empty(&info->caching_block_groups)) {
4014 caching_ctl = list_entry(info->caching_block_groups.next,
4015 struct btrfs_caching_control, list);
4016 list_del(&caching_ctl->list);
4017 btrfs_put_caching_control(caching_ctl);
4019 write_unlock(&info->block_group_cache_lock);
4021 spin_lock(&info->unused_bgs_lock);
4022 while (!list_empty(&info->unused_bgs)) {
4023 block_group = list_first_entry(&info->unused_bgs,
4024 struct btrfs_block_group,
4026 list_del_init(&block_group->bg_list);
4027 btrfs_put_block_group(block_group);
4030 while (!list_empty(&info->reclaim_bgs)) {
4031 block_group = list_first_entry(&info->reclaim_bgs,
4032 struct btrfs_block_group,
4034 list_del_init(&block_group->bg_list);
4035 btrfs_put_block_group(block_group);
4037 spin_unlock(&info->unused_bgs_lock);
4039 spin_lock(&info->zone_active_bgs_lock);
4040 while (!list_empty(&info->zone_active_bgs)) {
4041 block_group = list_first_entry(&info->zone_active_bgs,
4042 struct btrfs_block_group,
4044 list_del_init(&block_group->active_bg_list);
4045 btrfs_put_block_group(block_group);
4047 spin_unlock(&info->zone_active_bgs_lock);
4049 write_lock(&info->block_group_cache_lock);
4050 while ((n = rb_last(&info->block_group_cache_tree.rb_root)) != NULL) {
4051 block_group = rb_entry(n, struct btrfs_block_group,
4053 rb_erase_cached(&block_group->cache_node,
4054 &info->block_group_cache_tree);
4055 RB_CLEAR_NODE(&block_group->cache_node);
4056 write_unlock(&info->block_group_cache_lock);
4058 down_write(&block_group->space_info->groups_sem);
4059 list_del(&block_group->list);
4060 up_write(&block_group->space_info->groups_sem);
4063 * We haven't cached this block group, which means we could
4064 * possibly have excluded extents on this block group.
4066 if (block_group->cached == BTRFS_CACHE_NO ||
4067 block_group->cached == BTRFS_CACHE_ERROR)
4068 btrfs_free_excluded_extents(block_group);
4070 btrfs_remove_free_space_cache(block_group);
4071 ASSERT(block_group->cached != BTRFS_CACHE_STARTED);
4072 ASSERT(list_empty(&block_group->dirty_list));
4073 ASSERT(list_empty(&block_group->io_list));
4074 ASSERT(list_empty(&block_group->bg_list));
4075 ASSERT(refcount_read(&block_group->refs) == 1);
4076 ASSERT(block_group->swap_extents == 0);
4077 btrfs_put_block_group(block_group);
4079 write_lock(&info->block_group_cache_lock);
4081 write_unlock(&info->block_group_cache_lock);
4083 btrfs_release_global_block_rsv(info);
4085 while (!list_empty(&info->space_info)) {
4086 space_info = list_entry(info->space_info.next,
4087 struct btrfs_space_info,
4091 * Do not hide this behind enospc_debug, this is actually
4092 * important and indicates a real bug if this happens.
4094 if (WARN_ON(space_info->bytes_pinned > 0 ||
4095 space_info->bytes_may_use > 0))
4096 btrfs_dump_space_info(info, space_info, 0, 0);
4099 * If there was a failure to cleanup a log tree, very likely due
4100 * to an IO failure on a writeback attempt of one or more of its
4101 * extent buffers, we could not do proper (and cheap) unaccounting
4102 * of their reserved space, so don't warn on bytes_reserved > 0 in
4105 if (!(space_info->flags & BTRFS_BLOCK_GROUP_METADATA) ||
4106 !BTRFS_FS_LOG_CLEANUP_ERROR(info)) {
4107 if (WARN_ON(space_info->bytes_reserved > 0))
4108 btrfs_dump_space_info(info, space_info, 0, 0);
4111 WARN_ON(space_info->reclaim_size > 0);
4112 list_del(&space_info->list);
4113 btrfs_sysfs_remove_space_info(space_info);
4118 void btrfs_freeze_block_group(struct btrfs_block_group *cache)
4120 atomic_inc(&cache->frozen);
4123 void btrfs_unfreeze_block_group(struct btrfs_block_group *block_group)
4125 struct btrfs_fs_info *fs_info = block_group->fs_info;
4126 struct extent_map_tree *em_tree;
4127 struct extent_map *em;
4130 spin_lock(&block_group->lock);
4131 cleanup = (atomic_dec_and_test(&block_group->frozen) &&
4132 block_group->removed);
4133 spin_unlock(&block_group->lock);
4136 em_tree = &fs_info->mapping_tree;
4137 write_lock(&em_tree->lock);
4138 em = lookup_extent_mapping(em_tree, block_group->start,
4140 BUG_ON(!em); /* logic error, can't happen */
4141 remove_extent_mapping(em_tree, em);
4142 write_unlock(&em_tree->lock);
4144 /* once for us and once for the tree */
4145 free_extent_map(em);
4146 free_extent_map(em);
4149 * We may have left one free space entry and other possible
4150 * tasks trimming this block group have left 1 entry each one.
4153 __btrfs_remove_free_space_cache(block_group->free_space_ctl);
4157 bool btrfs_inc_block_group_swap_extents(struct btrfs_block_group *bg)
4161 spin_lock(&bg->lock);
4166 spin_unlock(&bg->lock);
4171 void btrfs_dec_block_group_swap_extents(struct btrfs_block_group *bg, int amount)
4173 spin_lock(&bg->lock);
4175 ASSERT(bg->swap_extents >= amount);
4176 bg->swap_extents -= amount;
4177 spin_unlock(&bg->lock);