1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
6 #include <linux/sched.h>
7 #include <linux/sched/signal.h>
8 #include <linux/pagemap.h>
9 #include <linux/writeback.h>
10 #include <linux/blkdev.h>
11 #include <linux/sort.h>
12 #include <linux/rcupdate.h>
13 #include <linux/kthread.h>
14 #include <linux/slab.h>
15 #include <linux/ratelimit.h>
16 #include <linux/percpu_counter.h>
17 #include <linux/lockdep.h>
18 #include <linux/crc32c.h>
21 #include "print-tree.h"
25 #include "free-space-cache.h"
26 #include "free-space-tree.h"
30 #include "ref-verify.h"
32 #undef SCRAMBLE_DELAYED_REFS
35 * control flags for do_chunk_alloc's force field
36 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
37 * if we really need one.
39 * CHUNK_ALLOC_LIMITED means to only try and allocate one
40 * if we have very few chunks already allocated. This is
41 * used as part of the clustering code to help make sure
42 * we have a good pool of storage to cluster in, without
43 * filling the FS with empty chunks
45 * CHUNK_ALLOC_FORCE means it must try to allocate one
49 CHUNK_ALLOC_NO_FORCE = 0,
50 CHUNK_ALLOC_LIMITED = 1,
51 CHUNK_ALLOC_FORCE = 2,
54 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
55 struct btrfs_fs_info *fs_info,
56 struct btrfs_delayed_ref_node *node, u64 parent,
57 u64 root_objectid, u64 owner_objectid,
58 u64 owner_offset, int refs_to_drop,
59 struct btrfs_delayed_extent_op *extra_op);
60 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
61 struct extent_buffer *leaf,
62 struct btrfs_extent_item *ei);
63 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
64 struct btrfs_fs_info *fs_info,
65 u64 parent, u64 root_objectid,
66 u64 flags, u64 owner, u64 offset,
67 struct btrfs_key *ins, int ref_mod);
68 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
69 struct btrfs_fs_info *fs_info,
70 u64 parent, u64 root_objectid,
71 u64 flags, struct btrfs_disk_key *key,
72 int level, struct btrfs_key *ins);
73 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
74 struct btrfs_fs_info *fs_info, u64 flags,
76 static int find_next_key(struct btrfs_path *path, int level,
77 struct btrfs_key *key);
78 static void dump_space_info(struct btrfs_fs_info *fs_info,
79 struct btrfs_space_info *info, u64 bytes,
80 int dump_block_groups);
81 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
83 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
84 struct btrfs_space_info *space_info,
86 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
87 struct btrfs_space_info *space_info,
91 block_group_cache_done(struct btrfs_block_group_cache *cache)
94 return cache->cached == BTRFS_CACHE_FINISHED ||
95 cache->cached == BTRFS_CACHE_ERROR;
98 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
100 return (cache->flags & bits) == bits;
103 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
105 atomic_inc(&cache->count);
108 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
110 if (atomic_dec_and_test(&cache->count)) {
111 WARN_ON(cache->pinned > 0);
112 WARN_ON(cache->reserved > 0);
115 * If not empty, someone is still holding mutex of
116 * full_stripe_lock, which can only be released by caller.
117 * And it will definitely cause use-after-free when caller
118 * tries to release full stripe lock.
120 * No better way to resolve, but only to warn.
122 WARN_ON(!RB_EMPTY_ROOT(&cache->full_stripe_locks_root.root));
123 kfree(cache->free_space_ctl);
129 * this adds the block group to the fs_info rb tree for the block group
132 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
133 struct btrfs_block_group_cache *block_group)
136 struct rb_node *parent = NULL;
137 struct btrfs_block_group_cache *cache;
139 spin_lock(&info->block_group_cache_lock);
140 p = &info->block_group_cache_tree.rb_node;
144 cache = rb_entry(parent, struct btrfs_block_group_cache,
146 if (block_group->key.objectid < cache->key.objectid) {
148 } else if (block_group->key.objectid > cache->key.objectid) {
151 spin_unlock(&info->block_group_cache_lock);
156 rb_link_node(&block_group->cache_node, parent, p);
157 rb_insert_color(&block_group->cache_node,
158 &info->block_group_cache_tree);
160 if (info->first_logical_byte > block_group->key.objectid)
161 info->first_logical_byte = block_group->key.objectid;
163 spin_unlock(&info->block_group_cache_lock);
169 * This will return the block group at or after bytenr if contains is 0, else
170 * it will return the block group that contains the bytenr
172 static struct btrfs_block_group_cache *
173 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
176 struct btrfs_block_group_cache *cache, *ret = NULL;
180 spin_lock(&info->block_group_cache_lock);
181 n = info->block_group_cache_tree.rb_node;
184 cache = rb_entry(n, struct btrfs_block_group_cache,
186 end = cache->key.objectid + cache->key.offset - 1;
187 start = cache->key.objectid;
189 if (bytenr < start) {
190 if (!contains && (!ret || start < ret->key.objectid))
193 } else if (bytenr > start) {
194 if (contains && bytenr <= end) {
205 btrfs_get_block_group(ret);
206 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
207 info->first_logical_byte = ret->key.objectid;
209 spin_unlock(&info->block_group_cache_lock);
214 static int add_excluded_extent(struct btrfs_fs_info *fs_info,
215 u64 start, u64 num_bytes)
217 u64 end = start + num_bytes - 1;
218 set_extent_bits(&fs_info->freed_extents[0],
219 start, end, EXTENT_UPTODATE);
220 set_extent_bits(&fs_info->freed_extents[1],
221 start, end, EXTENT_UPTODATE);
225 static void free_excluded_extents(struct btrfs_fs_info *fs_info,
226 struct btrfs_block_group_cache *cache)
230 start = cache->key.objectid;
231 end = start + cache->key.offset - 1;
233 clear_extent_bits(&fs_info->freed_extents[0],
234 start, end, EXTENT_UPTODATE);
235 clear_extent_bits(&fs_info->freed_extents[1],
236 start, end, EXTENT_UPTODATE);
239 static int exclude_super_stripes(struct btrfs_fs_info *fs_info,
240 struct btrfs_block_group_cache *cache)
247 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
248 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
249 cache->bytes_super += stripe_len;
250 ret = add_excluded_extent(fs_info, cache->key.objectid,
256 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
257 bytenr = btrfs_sb_offset(i);
258 ret = btrfs_rmap_block(fs_info, cache->key.objectid,
259 bytenr, 0, &logical, &nr, &stripe_len);
266 if (logical[nr] > cache->key.objectid +
270 if (logical[nr] + stripe_len <= cache->key.objectid)
274 if (start < cache->key.objectid) {
275 start = cache->key.objectid;
276 len = (logical[nr] + stripe_len) - start;
278 len = min_t(u64, stripe_len,
279 cache->key.objectid +
280 cache->key.offset - start);
283 cache->bytes_super += len;
284 ret = add_excluded_extent(fs_info, start, len);
296 static struct btrfs_caching_control *
297 get_caching_control(struct btrfs_block_group_cache *cache)
299 struct btrfs_caching_control *ctl;
301 spin_lock(&cache->lock);
302 if (!cache->caching_ctl) {
303 spin_unlock(&cache->lock);
307 ctl = cache->caching_ctl;
308 refcount_inc(&ctl->count);
309 spin_unlock(&cache->lock);
313 static void put_caching_control(struct btrfs_caching_control *ctl)
315 if (refcount_dec_and_test(&ctl->count))
319 #ifdef CONFIG_BTRFS_DEBUG
320 static void fragment_free_space(struct btrfs_block_group_cache *block_group)
322 struct btrfs_fs_info *fs_info = block_group->fs_info;
323 u64 start = block_group->key.objectid;
324 u64 len = block_group->key.offset;
325 u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ?
326 fs_info->nodesize : fs_info->sectorsize;
327 u64 step = chunk << 1;
329 while (len > chunk) {
330 btrfs_remove_free_space(block_group, start, chunk);
341 * this is only called by cache_block_group, since we could have freed extents
342 * we need to check the pinned_extents for any extents that can't be used yet
343 * since their free space will be released as soon as the transaction commits.
345 u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
346 struct btrfs_fs_info *info, u64 start, u64 end)
348 u64 extent_start, extent_end, size, total_added = 0;
351 while (start < end) {
352 ret = find_first_extent_bit(info->pinned_extents, start,
353 &extent_start, &extent_end,
354 EXTENT_DIRTY | EXTENT_UPTODATE,
359 if (extent_start <= start) {
360 start = extent_end + 1;
361 } else if (extent_start > start && extent_start < end) {
362 size = extent_start - start;
364 ret = btrfs_add_free_space(block_group, start,
366 BUG_ON(ret); /* -ENOMEM or logic error */
367 start = extent_end + 1;
376 ret = btrfs_add_free_space(block_group, start, size);
377 BUG_ON(ret); /* -ENOMEM or logic error */
383 static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl)
385 struct btrfs_block_group_cache *block_group = caching_ctl->block_group;
386 struct btrfs_fs_info *fs_info = block_group->fs_info;
387 struct btrfs_root *extent_root = fs_info->extent_root;
388 struct btrfs_path *path;
389 struct extent_buffer *leaf;
390 struct btrfs_key key;
397 path = btrfs_alloc_path();
401 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
403 #ifdef CONFIG_BTRFS_DEBUG
405 * If we're fragmenting we don't want to make anybody think we can
406 * allocate from this block group until we've had a chance to fragment
409 if (btrfs_should_fragment_free_space(block_group))
413 * We don't want to deadlock with somebody trying to allocate a new
414 * extent for the extent root while also trying to search the extent
415 * root to add free space. So we skip locking and search the commit
416 * root, since its read-only
418 path->skip_locking = 1;
419 path->search_commit_root = 1;
420 path->reada = READA_FORWARD;
424 key.type = BTRFS_EXTENT_ITEM_KEY;
427 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
431 leaf = path->nodes[0];
432 nritems = btrfs_header_nritems(leaf);
435 if (btrfs_fs_closing(fs_info) > 1) {
440 if (path->slots[0] < nritems) {
441 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
443 ret = find_next_key(path, 0, &key);
447 if (need_resched() ||
448 rwsem_is_contended(&fs_info->commit_root_sem)) {
450 caching_ctl->progress = last;
451 btrfs_release_path(path);
452 up_read(&fs_info->commit_root_sem);
453 mutex_unlock(&caching_ctl->mutex);
455 mutex_lock(&caching_ctl->mutex);
456 down_read(&fs_info->commit_root_sem);
460 ret = btrfs_next_leaf(extent_root, path);
465 leaf = path->nodes[0];
466 nritems = btrfs_header_nritems(leaf);
470 if (key.objectid < last) {
473 key.type = BTRFS_EXTENT_ITEM_KEY;
476 caching_ctl->progress = last;
477 btrfs_release_path(path);
481 if (key.objectid < block_group->key.objectid) {
486 if (key.objectid >= block_group->key.objectid +
487 block_group->key.offset)
490 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
491 key.type == BTRFS_METADATA_ITEM_KEY) {
492 total_found += add_new_free_space(block_group,
495 if (key.type == BTRFS_METADATA_ITEM_KEY)
496 last = key.objectid +
499 last = key.objectid + key.offset;
501 if (total_found > CACHING_CTL_WAKE_UP) {
504 wake_up(&caching_ctl->wait);
511 total_found += add_new_free_space(block_group, fs_info, last,
512 block_group->key.objectid +
513 block_group->key.offset);
514 caching_ctl->progress = (u64)-1;
517 btrfs_free_path(path);
521 static noinline void caching_thread(struct btrfs_work *work)
523 struct btrfs_block_group_cache *block_group;
524 struct btrfs_fs_info *fs_info;
525 struct btrfs_caching_control *caching_ctl;
528 caching_ctl = container_of(work, struct btrfs_caching_control, work);
529 block_group = caching_ctl->block_group;
530 fs_info = block_group->fs_info;
532 mutex_lock(&caching_ctl->mutex);
533 down_read(&fs_info->commit_root_sem);
535 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
536 ret = load_free_space_tree(caching_ctl);
538 ret = load_extent_tree_free(caching_ctl);
540 spin_lock(&block_group->lock);
541 block_group->caching_ctl = NULL;
542 block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED;
543 spin_unlock(&block_group->lock);
545 #ifdef CONFIG_BTRFS_DEBUG
546 if (btrfs_should_fragment_free_space(block_group)) {
549 spin_lock(&block_group->space_info->lock);
550 spin_lock(&block_group->lock);
551 bytes_used = block_group->key.offset -
552 btrfs_block_group_used(&block_group->item);
553 block_group->space_info->bytes_used += bytes_used >> 1;
554 spin_unlock(&block_group->lock);
555 spin_unlock(&block_group->space_info->lock);
556 fragment_free_space(block_group);
560 caching_ctl->progress = (u64)-1;
562 up_read(&fs_info->commit_root_sem);
563 free_excluded_extents(fs_info, block_group);
564 mutex_unlock(&caching_ctl->mutex);
566 wake_up(&caching_ctl->wait);
568 put_caching_control(caching_ctl);
569 btrfs_put_block_group(block_group);
572 static int cache_block_group(struct btrfs_block_group_cache *cache,
576 struct btrfs_fs_info *fs_info = cache->fs_info;
577 struct btrfs_caching_control *caching_ctl;
580 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
584 INIT_LIST_HEAD(&caching_ctl->list);
585 mutex_init(&caching_ctl->mutex);
586 init_waitqueue_head(&caching_ctl->wait);
587 caching_ctl->block_group = cache;
588 caching_ctl->progress = cache->key.objectid;
589 refcount_set(&caching_ctl->count, 1);
590 btrfs_init_work(&caching_ctl->work, btrfs_cache_helper,
591 caching_thread, NULL, NULL);
593 spin_lock(&cache->lock);
595 * This should be a rare occasion, but this could happen I think in the
596 * case where one thread starts to load the space cache info, and then
597 * some other thread starts a transaction commit which tries to do an
598 * allocation while the other thread is still loading the space cache
599 * info. The previous loop should have kept us from choosing this block
600 * group, but if we've moved to the state where we will wait on caching
601 * block groups we need to first check if we're doing a fast load here,
602 * so we can wait for it to finish, otherwise we could end up allocating
603 * from a block group who's cache gets evicted for one reason or
606 while (cache->cached == BTRFS_CACHE_FAST) {
607 struct btrfs_caching_control *ctl;
609 ctl = cache->caching_ctl;
610 refcount_inc(&ctl->count);
611 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
612 spin_unlock(&cache->lock);
616 finish_wait(&ctl->wait, &wait);
617 put_caching_control(ctl);
618 spin_lock(&cache->lock);
621 if (cache->cached != BTRFS_CACHE_NO) {
622 spin_unlock(&cache->lock);
626 WARN_ON(cache->caching_ctl);
627 cache->caching_ctl = caching_ctl;
628 cache->cached = BTRFS_CACHE_FAST;
629 spin_unlock(&cache->lock);
631 if (btrfs_test_opt(fs_info, SPACE_CACHE)) {
632 mutex_lock(&caching_ctl->mutex);
633 ret = load_free_space_cache(fs_info, cache);
635 spin_lock(&cache->lock);
637 cache->caching_ctl = NULL;
638 cache->cached = BTRFS_CACHE_FINISHED;
639 cache->last_byte_to_unpin = (u64)-1;
640 caching_ctl->progress = (u64)-1;
642 if (load_cache_only) {
643 cache->caching_ctl = NULL;
644 cache->cached = BTRFS_CACHE_NO;
646 cache->cached = BTRFS_CACHE_STARTED;
647 cache->has_caching_ctl = 1;
650 spin_unlock(&cache->lock);
651 #ifdef CONFIG_BTRFS_DEBUG
653 btrfs_should_fragment_free_space(cache)) {
656 spin_lock(&cache->space_info->lock);
657 spin_lock(&cache->lock);
658 bytes_used = cache->key.offset -
659 btrfs_block_group_used(&cache->item);
660 cache->space_info->bytes_used += bytes_used >> 1;
661 spin_unlock(&cache->lock);
662 spin_unlock(&cache->space_info->lock);
663 fragment_free_space(cache);
666 mutex_unlock(&caching_ctl->mutex);
668 wake_up(&caching_ctl->wait);
670 put_caching_control(caching_ctl);
671 free_excluded_extents(fs_info, cache);
676 * We're either using the free space tree or no caching at all.
677 * Set cached to the appropriate value and wakeup any waiters.
679 spin_lock(&cache->lock);
680 if (load_cache_only) {
681 cache->caching_ctl = NULL;
682 cache->cached = BTRFS_CACHE_NO;
684 cache->cached = BTRFS_CACHE_STARTED;
685 cache->has_caching_ctl = 1;
687 spin_unlock(&cache->lock);
688 wake_up(&caching_ctl->wait);
691 if (load_cache_only) {
692 put_caching_control(caching_ctl);
696 down_write(&fs_info->commit_root_sem);
697 refcount_inc(&caching_ctl->count);
698 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
699 up_write(&fs_info->commit_root_sem);
701 btrfs_get_block_group(cache);
703 btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
709 * return the block group that starts at or after bytenr
711 static struct btrfs_block_group_cache *
712 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
714 return block_group_cache_tree_search(info, bytenr, 0);
718 * return the block group that contains the given bytenr
720 struct btrfs_block_group_cache *btrfs_lookup_block_group(
721 struct btrfs_fs_info *info,
724 return block_group_cache_tree_search(info, bytenr, 1);
727 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
730 struct list_head *head = &info->space_info;
731 struct btrfs_space_info *found;
733 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
736 list_for_each_entry_rcu(found, head, list) {
737 if (found->flags & flags) {
746 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, s64 num_bytes,
747 bool metadata, u64 root_objectid)
749 struct btrfs_space_info *space_info;
753 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
754 flags = BTRFS_BLOCK_GROUP_SYSTEM;
756 flags = BTRFS_BLOCK_GROUP_METADATA;
758 flags = BTRFS_BLOCK_GROUP_DATA;
761 space_info = __find_space_info(fs_info, flags);
763 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
767 * after adding space to the filesystem, we need to clear the full flags
768 * on all the space infos.
770 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
772 struct list_head *head = &info->space_info;
773 struct btrfs_space_info *found;
776 list_for_each_entry_rcu(found, head, list)
781 /* simple helper to search for an existing data extent at a given offset */
782 int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len)
785 struct btrfs_key key;
786 struct btrfs_path *path;
788 path = btrfs_alloc_path();
792 key.objectid = start;
794 key.type = BTRFS_EXTENT_ITEM_KEY;
795 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
796 btrfs_free_path(path);
801 * helper function to lookup reference count and flags of a tree block.
803 * the head node for delayed ref is used to store the sum of all the
804 * reference count modifications queued up in the rbtree. the head
805 * node may also store the extent flags to set. This way you can check
806 * to see what the reference count and extent flags would be if all of
807 * the delayed refs are not processed.
809 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
810 struct btrfs_fs_info *fs_info, u64 bytenr,
811 u64 offset, int metadata, u64 *refs, u64 *flags)
813 struct btrfs_delayed_ref_head *head;
814 struct btrfs_delayed_ref_root *delayed_refs;
815 struct btrfs_path *path;
816 struct btrfs_extent_item *ei;
817 struct extent_buffer *leaf;
818 struct btrfs_key key;
825 * If we don't have skinny metadata, don't bother doing anything
828 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) {
829 offset = fs_info->nodesize;
833 path = btrfs_alloc_path();
838 path->skip_locking = 1;
839 path->search_commit_root = 1;
843 key.objectid = bytenr;
846 key.type = BTRFS_METADATA_ITEM_KEY;
848 key.type = BTRFS_EXTENT_ITEM_KEY;
850 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
854 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
855 if (path->slots[0]) {
857 btrfs_item_key_to_cpu(path->nodes[0], &key,
859 if (key.objectid == bytenr &&
860 key.type == BTRFS_EXTENT_ITEM_KEY &&
861 key.offset == fs_info->nodesize)
867 leaf = path->nodes[0];
868 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
869 if (item_size >= sizeof(*ei)) {
870 ei = btrfs_item_ptr(leaf, path->slots[0],
871 struct btrfs_extent_item);
872 num_refs = btrfs_extent_refs(leaf, ei);
873 extent_flags = btrfs_extent_flags(leaf, ei);
875 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
876 struct btrfs_extent_item_v0 *ei0;
877 BUG_ON(item_size != sizeof(*ei0));
878 ei0 = btrfs_item_ptr(leaf, path->slots[0],
879 struct btrfs_extent_item_v0);
880 num_refs = btrfs_extent_refs_v0(leaf, ei0);
881 /* FIXME: this isn't correct for data */
882 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
887 BUG_ON(num_refs == 0);
897 delayed_refs = &trans->transaction->delayed_refs;
898 spin_lock(&delayed_refs->lock);
899 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
901 if (!mutex_trylock(&head->mutex)) {
902 refcount_inc(&head->refs);
903 spin_unlock(&delayed_refs->lock);
905 btrfs_release_path(path);
908 * Mutex was contended, block until it's released and try
911 mutex_lock(&head->mutex);
912 mutex_unlock(&head->mutex);
913 btrfs_put_delayed_ref_head(head);
916 spin_lock(&head->lock);
917 if (head->extent_op && head->extent_op->update_flags)
918 extent_flags |= head->extent_op->flags_to_set;
920 BUG_ON(num_refs == 0);
922 num_refs += head->ref_mod;
923 spin_unlock(&head->lock);
924 mutex_unlock(&head->mutex);
926 spin_unlock(&delayed_refs->lock);
928 WARN_ON(num_refs == 0);
932 *flags = extent_flags;
934 btrfs_free_path(path);
939 * Back reference rules. Back refs have three main goals:
941 * 1) differentiate between all holders of references to an extent so that
942 * when a reference is dropped we can make sure it was a valid reference
943 * before freeing the extent.
945 * 2) Provide enough information to quickly find the holders of an extent
946 * if we notice a given block is corrupted or bad.
948 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
949 * maintenance. This is actually the same as #2, but with a slightly
950 * different use case.
952 * There are two kinds of back refs. The implicit back refs is optimized
953 * for pointers in non-shared tree blocks. For a given pointer in a block,
954 * back refs of this kind provide information about the block's owner tree
955 * and the pointer's key. These information allow us to find the block by
956 * b-tree searching. The full back refs is for pointers in tree blocks not
957 * referenced by their owner trees. The location of tree block is recorded
958 * in the back refs. Actually the full back refs is generic, and can be
959 * used in all cases the implicit back refs is used. The major shortcoming
960 * of the full back refs is its overhead. Every time a tree block gets
961 * COWed, we have to update back refs entry for all pointers in it.
963 * For a newly allocated tree block, we use implicit back refs for
964 * pointers in it. This means most tree related operations only involve
965 * implicit back refs. For a tree block created in old transaction, the
966 * only way to drop a reference to it is COW it. So we can detect the
967 * event that tree block loses its owner tree's reference and do the
968 * back refs conversion.
970 * When a tree block is COWed through a tree, there are four cases:
972 * The reference count of the block is one and the tree is the block's
973 * owner tree. Nothing to do in this case.
975 * The reference count of the block is one and the tree is not the
976 * block's owner tree. In this case, full back refs is used for pointers
977 * in the block. Remove these full back refs, add implicit back refs for
978 * every pointers in the new block.
980 * The reference count of the block is greater than one and the tree is
981 * the block's owner tree. In this case, implicit back refs is used for
982 * pointers in the block. Add full back refs for every pointers in the
983 * block, increase lower level extents' reference counts. The original
984 * implicit back refs are entailed to the new block.
986 * The reference count of the block is greater than one and the tree is
987 * not the block's owner tree. Add implicit back refs for every pointer in
988 * the new block, increase lower level extents' reference count.
990 * Back Reference Key composing:
992 * The key objectid corresponds to the first byte in the extent,
993 * The key type is used to differentiate between types of back refs.
994 * There are different meanings of the key offset for different types
997 * File extents can be referenced by:
999 * - multiple snapshots, subvolumes, or different generations in one subvol
1000 * - different files inside a single subvolume
1001 * - different offsets inside a file (bookend extents in file.c)
1003 * The extent ref structure for the implicit back refs has fields for:
1005 * - Objectid of the subvolume root
1006 * - objectid of the file holding the reference
1007 * - original offset in the file
1008 * - how many bookend extents
1010 * The key offset for the implicit back refs is hash of the first
1013 * The extent ref structure for the full back refs has field for:
1015 * - number of pointers in the tree leaf
1017 * The key offset for the implicit back refs is the first byte of
1020 * When a file extent is allocated, The implicit back refs is used.
1021 * the fields are filled in:
1023 * (root_key.objectid, inode objectid, offset in file, 1)
1025 * When a file extent is removed file truncation, we find the
1026 * corresponding implicit back refs and check the following fields:
1028 * (btrfs_header_owner(leaf), inode objectid, offset in file)
1030 * Btree extents can be referenced by:
1032 * - Different subvolumes
1034 * Both the implicit back refs and the full back refs for tree blocks
1035 * only consist of key. The key offset for the implicit back refs is
1036 * objectid of block's owner tree. The key offset for the full back refs
1037 * is the first byte of parent block.
1039 * When implicit back refs is used, information about the lowest key and
1040 * level of the tree block are required. These information are stored in
1041 * tree block info structure.
1044 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1045 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
1046 struct btrfs_fs_info *fs_info,
1047 struct btrfs_path *path,
1048 u64 owner, u32 extra_size)
1050 struct btrfs_root *root = fs_info->extent_root;
1051 struct btrfs_extent_item *item;
1052 struct btrfs_extent_item_v0 *ei0;
1053 struct btrfs_extent_ref_v0 *ref0;
1054 struct btrfs_tree_block_info *bi;
1055 struct extent_buffer *leaf;
1056 struct btrfs_key key;
1057 struct btrfs_key found_key;
1058 u32 new_size = sizeof(*item);
1062 leaf = path->nodes[0];
1063 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
1065 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1066 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1067 struct btrfs_extent_item_v0);
1068 refs = btrfs_extent_refs_v0(leaf, ei0);
1070 if (owner == (u64)-1) {
1072 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1073 ret = btrfs_next_leaf(root, path);
1076 BUG_ON(ret > 0); /* Corruption */
1077 leaf = path->nodes[0];
1079 btrfs_item_key_to_cpu(leaf, &found_key,
1081 BUG_ON(key.objectid != found_key.objectid);
1082 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1086 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1087 struct btrfs_extent_ref_v0);
1088 owner = btrfs_ref_objectid_v0(leaf, ref0);
1092 btrfs_release_path(path);
1094 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1095 new_size += sizeof(*bi);
1097 new_size -= sizeof(*ei0);
1098 ret = btrfs_search_slot(trans, root, &key, path,
1099 new_size + extra_size, 1);
1102 BUG_ON(ret); /* Corruption */
1104 btrfs_extend_item(fs_info, path, new_size);
1106 leaf = path->nodes[0];
1107 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1108 btrfs_set_extent_refs(leaf, item, refs);
1109 /* FIXME: get real generation */
1110 btrfs_set_extent_generation(leaf, item, 0);
1111 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1112 btrfs_set_extent_flags(leaf, item,
1113 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1114 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1115 bi = (struct btrfs_tree_block_info *)(item + 1);
1116 /* FIXME: get first key of the block */
1117 memzero_extent_buffer(leaf, (unsigned long)bi, sizeof(*bi));
1118 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1120 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1122 btrfs_mark_buffer_dirty(leaf);
1128 * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required,
1129 * is_data == BTRFS_REF_TYPE_DATA, data type is requried,
1130 * is_data == BTRFS_REF_TYPE_ANY, either type is OK.
1132 int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb,
1133 struct btrfs_extent_inline_ref *iref,
1134 enum btrfs_inline_ref_type is_data)
1136 int type = btrfs_extent_inline_ref_type(eb, iref);
1137 u64 offset = btrfs_extent_inline_ref_offset(eb, iref);
1139 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1140 type == BTRFS_SHARED_BLOCK_REF_KEY ||
1141 type == BTRFS_SHARED_DATA_REF_KEY ||
1142 type == BTRFS_EXTENT_DATA_REF_KEY) {
1143 if (is_data == BTRFS_REF_TYPE_BLOCK) {
1144 if (type == BTRFS_TREE_BLOCK_REF_KEY)
1146 if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1147 ASSERT(eb->fs_info);
1149 * Every shared one has parent tree
1150 * block, which must be aligned to
1154 IS_ALIGNED(offset, eb->fs_info->nodesize))
1157 } else if (is_data == BTRFS_REF_TYPE_DATA) {
1158 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1160 if (type == BTRFS_SHARED_DATA_REF_KEY) {
1161 ASSERT(eb->fs_info);
1163 * Every shared one has parent tree
1164 * block, which must be aligned to
1168 IS_ALIGNED(offset, eb->fs_info->nodesize))
1172 ASSERT(is_data == BTRFS_REF_TYPE_ANY);
1177 btrfs_print_leaf((struct extent_buffer *)eb);
1178 btrfs_err(eb->fs_info, "eb %llu invalid extent inline ref type %d",
1182 return BTRFS_REF_TYPE_INVALID;
1185 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1187 u32 high_crc = ~(u32)0;
1188 u32 low_crc = ~(u32)0;
1191 lenum = cpu_to_le64(root_objectid);
1192 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1193 lenum = cpu_to_le64(owner);
1194 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1195 lenum = cpu_to_le64(offset);
1196 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1198 return ((u64)high_crc << 31) ^ (u64)low_crc;
1201 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1202 struct btrfs_extent_data_ref *ref)
1204 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1205 btrfs_extent_data_ref_objectid(leaf, ref),
1206 btrfs_extent_data_ref_offset(leaf, ref));
1209 static int match_extent_data_ref(struct extent_buffer *leaf,
1210 struct btrfs_extent_data_ref *ref,
1211 u64 root_objectid, u64 owner, u64 offset)
1213 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1214 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1215 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1220 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1221 struct btrfs_fs_info *fs_info,
1222 struct btrfs_path *path,
1223 u64 bytenr, u64 parent,
1225 u64 owner, u64 offset)
1227 struct btrfs_root *root = fs_info->extent_root;
1228 struct btrfs_key key;
1229 struct btrfs_extent_data_ref *ref;
1230 struct extent_buffer *leaf;
1236 key.objectid = bytenr;
1238 key.type = BTRFS_SHARED_DATA_REF_KEY;
1239 key.offset = parent;
1241 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1242 key.offset = hash_extent_data_ref(root_objectid,
1247 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1256 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1257 key.type = BTRFS_EXTENT_REF_V0_KEY;
1258 btrfs_release_path(path);
1259 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1270 leaf = path->nodes[0];
1271 nritems = btrfs_header_nritems(leaf);
1273 if (path->slots[0] >= nritems) {
1274 ret = btrfs_next_leaf(root, path);
1280 leaf = path->nodes[0];
1281 nritems = btrfs_header_nritems(leaf);
1285 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1286 if (key.objectid != bytenr ||
1287 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1290 ref = btrfs_item_ptr(leaf, path->slots[0],
1291 struct btrfs_extent_data_ref);
1293 if (match_extent_data_ref(leaf, ref, root_objectid,
1296 btrfs_release_path(path);
1308 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1309 struct btrfs_fs_info *fs_info,
1310 struct btrfs_path *path,
1311 u64 bytenr, u64 parent,
1312 u64 root_objectid, u64 owner,
1313 u64 offset, int refs_to_add)
1315 struct btrfs_root *root = fs_info->extent_root;
1316 struct btrfs_key key;
1317 struct extent_buffer *leaf;
1322 key.objectid = bytenr;
1324 key.type = BTRFS_SHARED_DATA_REF_KEY;
1325 key.offset = parent;
1326 size = sizeof(struct btrfs_shared_data_ref);
1328 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1329 key.offset = hash_extent_data_ref(root_objectid,
1331 size = sizeof(struct btrfs_extent_data_ref);
1334 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1335 if (ret && ret != -EEXIST)
1338 leaf = path->nodes[0];
1340 struct btrfs_shared_data_ref *ref;
1341 ref = btrfs_item_ptr(leaf, path->slots[0],
1342 struct btrfs_shared_data_ref);
1344 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1346 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1347 num_refs += refs_to_add;
1348 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1351 struct btrfs_extent_data_ref *ref;
1352 while (ret == -EEXIST) {
1353 ref = btrfs_item_ptr(leaf, path->slots[0],
1354 struct btrfs_extent_data_ref);
1355 if (match_extent_data_ref(leaf, ref, root_objectid,
1358 btrfs_release_path(path);
1360 ret = btrfs_insert_empty_item(trans, root, path, &key,
1362 if (ret && ret != -EEXIST)
1365 leaf = path->nodes[0];
1367 ref = btrfs_item_ptr(leaf, path->slots[0],
1368 struct btrfs_extent_data_ref);
1370 btrfs_set_extent_data_ref_root(leaf, ref,
1372 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1373 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1374 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1376 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1377 num_refs += refs_to_add;
1378 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1381 btrfs_mark_buffer_dirty(leaf);
1384 btrfs_release_path(path);
1388 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1389 struct btrfs_fs_info *fs_info,
1390 struct btrfs_path *path,
1391 int refs_to_drop, int *last_ref)
1393 struct btrfs_key key;
1394 struct btrfs_extent_data_ref *ref1 = NULL;
1395 struct btrfs_shared_data_ref *ref2 = NULL;
1396 struct extent_buffer *leaf;
1400 leaf = path->nodes[0];
1401 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1403 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1404 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1405 struct btrfs_extent_data_ref);
1406 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1407 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1408 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1409 struct btrfs_shared_data_ref);
1410 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1411 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1412 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1413 struct btrfs_extent_ref_v0 *ref0;
1414 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1415 struct btrfs_extent_ref_v0);
1416 num_refs = btrfs_ref_count_v0(leaf, ref0);
1422 BUG_ON(num_refs < refs_to_drop);
1423 num_refs -= refs_to_drop;
1425 if (num_refs == 0) {
1426 ret = btrfs_del_item(trans, fs_info->extent_root, path);
1429 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1430 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1431 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1432 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1433 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1435 struct btrfs_extent_ref_v0 *ref0;
1436 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1437 struct btrfs_extent_ref_v0);
1438 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1441 btrfs_mark_buffer_dirty(leaf);
1446 static noinline u32 extent_data_ref_count(struct btrfs_path *path,
1447 struct btrfs_extent_inline_ref *iref)
1449 struct btrfs_key key;
1450 struct extent_buffer *leaf;
1451 struct btrfs_extent_data_ref *ref1;
1452 struct btrfs_shared_data_ref *ref2;
1456 leaf = path->nodes[0];
1457 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1460 * If type is invalid, we should have bailed out earlier than
1463 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
1464 ASSERT(type != BTRFS_REF_TYPE_INVALID);
1465 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1466 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1467 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1469 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1470 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1472 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1473 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1474 struct btrfs_extent_data_ref);
1475 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1476 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1477 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1478 struct btrfs_shared_data_ref);
1479 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1480 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1481 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1482 struct btrfs_extent_ref_v0 *ref0;
1483 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1484 struct btrfs_extent_ref_v0);
1485 num_refs = btrfs_ref_count_v0(leaf, ref0);
1493 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1494 struct btrfs_fs_info *fs_info,
1495 struct btrfs_path *path,
1496 u64 bytenr, u64 parent,
1499 struct btrfs_root *root = fs_info->extent_root;
1500 struct btrfs_key key;
1503 key.objectid = bytenr;
1505 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1506 key.offset = parent;
1508 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1509 key.offset = root_objectid;
1512 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1515 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1516 if (ret == -ENOENT && parent) {
1517 btrfs_release_path(path);
1518 key.type = BTRFS_EXTENT_REF_V0_KEY;
1519 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1527 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1528 struct btrfs_fs_info *fs_info,
1529 struct btrfs_path *path,
1530 u64 bytenr, u64 parent,
1533 struct btrfs_key key;
1536 key.objectid = bytenr;
1538 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1539 key.offset = parent;
1541 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1542 key.offset = root_objectid;
1545 ret = btrfs_insert_empty_item(trans, fs_info->extent_root,
1547 btrfs_release_path(path);
1551 static inline int extent_ref_type(u64 parent, u64 owner)
1554 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1556 type = BTRFS_SHARED_BLOCK_REF_KEY;
1558 type = BTRFS_TREE_BLOCK_REF_KEY;
1561 type = BTRFS_SHARED_DATA_REF_KEY;
1563 type = BTRFS_EXTENT_DATA_REF_KEY;
1568 static int find_next_key(struct btrfs_path *path, int level,
1569 struct btrfs_key *key)
1572 for (; level < BTRFS_MAX_LEVEL; level++) {
1573 if (!path->nodes[level])
1575 if (path->slots[level] + 1 >=
1576 btrfs_header_nritems(path->nodes[level]))
1579 btrfs_item_key_to_cpu(path->nodes[level], key,
1580 path->slots[level] + 1);
1582 btrfs_node_key_to_cpu(path->nodes[level], key,
1583 path->slots[level] + 1);
1590 * look for inline back ref. if back ref is found, *ref_ret is set
1591 * to the address of inline back ref, and 0 is returned.
1593 * if back ref isn't found, *ref_ret is set to the address where it
1594 * should be inserted, and -ENOENT is returned.
1596 * if insert is true and there are too many inline back refs, the path
1597 * points to the extent item, and -EAGAIN is returned.
1599 * NOTE: inline back refs are ordered in the same way that back ref
1600 * items in the tree are ordered.
1602 static noinline_for_stack
1603 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1604 struct btrfs_fs_info *fs_info,
1605 struct btrfs_path *path,
1606 struct btrfs_extent_inline_ref **ref_ret,
1607 u64 bytenr, u64 num_bytes,
1608 u64 parent, u64 root_objectid,
1609 u64 owner, u64 offset, int insert)
1611 struct btrfs_root *root = fs_info->extent_root;
1612 struct btrfs_key key;
1613 struct extent_buffer *leaf;
1614 struct btrfs_extent_item *ei;
1615 struct btrfs_extent_inline_ref *iref;
1625 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
1628 key.objectid = bytenr;
1629 key.type = BTRFS_EXTENT_ITEM_KEY;
1630 key.offset = num_bytes;
1632 want = extent_ref_type(parent, owner);
1634 extra_size = btrfs_extent_inline_ref_size(want);
1635 path->keep_locks = 1;
1640 * Owner is our parent level, so we can just add one to get the level
1641 * for the block we are interested in.
1643 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1644 key.type = BTRFS_METADATA_ITEM_KEY;
1649 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1656 * We may be a newly converted file system which still has the old fat
1657 * extent entries for metadata, so try and see if we have one of those.
1659 if (ret > 0 && skinny_metadata) {
1660 skinny_metadata = false;
1661 if (path->slots[0]) {
1663 btrfs_item_key_to_cpu(path->nodes[0], &key,
1665 if (key.objectid == bytenr &&
1666 key.type == BTRFS_EXTENT_ITEM_KEY &&
1667 key.offset == num_bytes)
1671 key.objectid = bytenr;
1672 key.type = BTRFS_EXTENT_ITEM_KEY;
1673 key.offset = num_bytes;
1674 btrfs_release_path(path);
1679 if (ret && !insert) {
1682 } else if (WARN_ON(ret)) {
1687 leaf = path->nodes[0];
1688 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1689 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1690 if (item_size < sizeof(*ei)) {
1695 ret = convert_extent_item_v0(trans, fs_info, path, owner,
1701 leaf = path->nodes[0];
1702 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1705 BUG_ON(item_size < sizeof(*ei));
1707 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1708 flags = btrfs_extent_flags(leaf, ei);
1710 ptr = (unsigned long)(ei + 1);
1711 end = (unsigned long)ei + item_size;
1713 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1714 ptr += sizeof(struct btrfs_tree_block_info);
1718 if (owner >= BTRFS_FIRST_FREE_OBJECTID)
1719 needed = BTRFS_REF_TYPE_DATA;
1721 needed = BTRFS_REF_TYPE_BLOCK;
1729 iref = (struct btrfs_extent_inline_ref *)ptr;
1730 type = btrfs_get_extent_inline_ref_type(leaf, iref, needed);
1731 if (type == BTRFS_REF_TYPE_INVALID) {
1739 ptr += btrfs_extent_inline_ref_size(type);
1743 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1744 struct btrfs_extent_data_ref *dref;
1745 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1746 if (match_extent_data_ref(leaf, dref, root_objectid,
1751 if (hash_extent_data_ref_item(leaf, dref) <
1752 hash_extent_data_ref(root_objectid, owner, offset))
1756 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1758 if (parent == ref_offset) {
1762 if (ref_offset < parent)
1765 if (root_objectid == ref_offset) {
1769 if (ref_offset < root_objectid)
1773 ptr += btrfs_extent_inline_ref_size(type);
1775 if (err == -ENOENT && insert) {
1776 if (item_size + extra_size >=
1777 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1782 * To add new inline back ref, we have to make sure
1783 * there is no corresponding back ref item.
1784 * For simplicity, we just do not add new inline back
1785 * ref if there is any kind of item for this block
1787 if (find_next_key(path, 0, &key) == 0 &&
1788 key.objectid == bytenr &&
1789 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1794 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1797 path->keep_locks = 0;
1798 btrfs_unlock_up_safe(path, 1);
1804 * helper to add new inline back ref
1806 static noinline_for_stack
1807 void setup_inline_extent_backref(struct btrfs_fs_info *fs_info,
1808 struct btrfs_path *path,
1809 struct btrfs_extent_inline_ref *iref,
1810 u64 parent, u64 root_objectid,
1811 u64 owner, u64 offset, int refs_to_add,
1812 struct btrfs_delayed_extent_op *extent_op)
1814 struct extent_buffer *leaf;
1815 struct btrfs_extent_item *ei;
1818 unsigned long item_offset;
1823 leaf = path->nodes[0];
1824 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1825 item_offset = (unsigned long)iref - (unsigned long)ei;
1827 type = extent_ref_type(parent, owner);
1828 size = btrfs_extent_inline_ref_size(type);
1830 btrfs_extend_item(fs_info, path, size);
1832 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1833 refs = btrfs_extent_refs(leaf, ei);
1834 refs += refs_to_add;
1835 btrfs_set_extent_refs(leaf, ei, refs);
1837 __run_delayed_extent_op(extent_op, leaf, ei);
1839 ptr = (unsigned long)ei + item_offset;
1840 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1841 if (ptr < end - size)
1842 memmove_extent_buffer(leaf, ptr + size, ptr,
1845 iref = (struct btrfs_extent_inline_ref *)ptr;
1846 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1847 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1848 struct btrfs_extent_data_ref *dref;
1849 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1850 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1851 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1852 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1853 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1854 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1855 struct btrfs_shared_data_ref *sref;
1856 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1857 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1858 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1859 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1860 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1862 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1864 btrfs_mark_buffer_dirty(leaf);
1867 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1868 struct btrfs_fs_info *fs_info,
1869 struct btrfs_path *path,
1870 struct btrfs_extent_inline_ref **ref_ret,
1871 u64 bytenr, u64 num_bytes, u64 parent,
1872 u64 root_objectid, u64 owner, u64 offset)
1876 ret = lookup_inline_extent_backref(trans, fs_info, path, ref_ret,
1877 bytenr, num_bytes, parent,
1878 root_objectid, owner, offset, 0);
1882 btrfs_release_path(path);
1885 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1886 ret = lookup_tree_block_ref(trans, fs_info, path, bytenr,
1887 parent, root_objectid);
1889 ret = lookup_extent_data_ref(trans, fs_info, path, bytenr,
1890 parent, root_objectid, owner,
1897 * helper to update/remove inline back ref
1899 static noinline_for_stack
1900 void update_inline_extent_backref(struct btrfs_fs_info *fs_info,
1901 struct btrfs_path *path,
1902 struct btrfs_extent_inline_ref *iref,
1904 struct btrfs_delayed_extent_op *extent_op,
1907 struct extent_buffer *leaf;
1908 struct btrfs_extent_item *ei;
1909 struct btrfs_extent_data_ref *dref = NULL;
1910 struct btrfs_shared_data_ref *sref = NULL;
1918 leaf = path->nodes[0];
1919 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1920 refs = btrfs_extent_refs(leaf, ei);
1921 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1922 refs += refs_to_mod;
1923 btrfs_set_extent_refs(leaf, ei, refs);
1925 __run_delayed_extent_op(extent_op, leaf, ei);
1928 * If type is invalid, we should have bailed out after
1929 * lookup_inline_extent_backref().
1931 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY);
1932 ASSERT(type != BTRFS_REF_TYPE_INVALID);
1934 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1935 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1936 refs = btrfs_extent_data_ref_count(leaf, dref);
1937 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1938 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1939 refs = btrfs_shared_data_ref_count(leaf, sref);
1942 BUG_ON(refs_to_mod != -1);
1945 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1946 refs += refs_to_mod;
1949 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1950 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1952 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1955 size = btrfs_extent_inline_ref_size(type);
1956 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1957 ptr = (unsigned long)iref;
1958 end = (unsigned long)ei + item_size;
1959 if (ptr + size < end)
1960 memmove_extent_buffer(leaf, ptr, ptr + size,
1963 btrfs_truncate_item(fs_info, path, item_size, 1);
1965 btrfs_mark_buffer_dirty(leaf);
1968 static noinline_for_stack
1969 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1970 struct btrfs_fs_info *fs_info,
1971 struct btrfs_path *path,
1972 u64 bytenr, u64 num_bytes, u64 parent,
1973 u64 root_objectid, u64 owner,
1974 u64 offset, int refs_to_add,
1975 struct btrfs_delayed_extent_op *extent_op)
1977 struct btrfs_extent_inline_ref *iref;
1980 ret = lookup_inline_extent_backref(trans, fs_info, path, &iref,
1981 bytenr, num_bytes, parent,
1982 root_objectid, owner, offset, 1);
1984 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1985 update_inline_extent_backref(fs_info, path, iref,
1986 refs_to_add, extent_op, NULL);
1987 } else if (ret == -ENOENT) {
1988 setup_inline_extent_backref(fs_info, path, iref, parent,
1989 root_objectid, owner, offset,
1990 refs_to_add, extent_op);
1996 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1997 struct btrfs_fs_info *fs_info,
1998 struct btrfs_path *path,
1999 u64 bytenr, u64 parent, u64 root_objectid,
2000 u64 owner, u64 offset, int refs_to_add)
2003 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2004 BUG_ON(refs_to_add != 1);
2005 ret = insert_tree_block_ref(trans, fs_info, path, bytenr,
2006 parent, root_objectid);
2008 ret = insert_extent_data_ref(trans, fs_info, path, bytenr,
2009 parent, root_objectid,
2010 owner, offset, refs_to_add);
2015 static int remove_extent_backref(struct btrfs_trans_handle *trans,
2016 struct btrfs_fs_info *fs_info,
2017 struct btrfs_path *path,
2018 struct btrfs_extent_inline_ref *iref,
2019 int refs_to_drop, int is_data, int *last_ref)
2023 BUG_ON(!is_data && refs_to_drop != 1);
2025 update_inline_extent_backref(fs_info, path, iref,
2026 -refs_to_drop, NULL, last_ref);
2027 } else if (is_data) {
2028 ret = remove_extent_data_ref(trans, fs_info, path, refs_to_drop,
2032 ret = btrfs_del_item(trans, fs_info->extent_root, path);
2037 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
2038 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
2039 u64 *discarded_bytes)
2042 u64 bytes_left, end;
2043 u64 aligned_start = ALIGN(start, 1 << 9);
2045 if (WARN_ON(start != aligned_start)) {
2046 len -= aligned_start - start;
2047 len = round_down(len, 1 << 9);
2048 start = aligned_start;
2051 *discarded_bytes = 0;
2059 /* Skip any superblocks on this device. */
2060 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
2061 u64 sb_start = btrfs_sb_offset(j);
2062 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
2063 u64 size = sb_start - start;
2065 if (!in_range(sb_start, start, bytes_left) &&
2066 !in_range(sb_end, start, bytes_left) &&
2067 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
2071 * Superblock spans beginning of range. Adjust start and
2074 if (sb_start <= start) {
2075 start += sb_end - start;
2080 bytes_left = end - start;
2085 ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
2088 *discarded_bytes += size;
2089 else if (ret != -EOPNOTSUPP)
2098 bytes_left = end - start;
2102 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
2105 *discarded_bytes += bytes_left;
2110 int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr,
2111 u64 num_bytes, u64 *actual_bytes)
2114 u64 discarded_bytes = 0;
2115 struct btrfs_bio *bbio = NULL;
2119 * Avoid races with device replace and make sure our bbio has devices
2120 * associated to its stripes that don't go away while we are discarding.
2122 btrfs_bio_counter_inc_blocked(fs_info);
2123 /* Tell the block device(s) that the sectors can be discarded */
2124 ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, bytenr, &num_bytes,
2126 /* Error condition is -ENOMEM */
2128 struct btrfs_bio_stripe *stripe = bbio->stripes;
2132 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
2134 struct request_queue *req_q;
2136 if (!stripe->dev->bdev) {
2137 ASSERT(btrfs_test_opt(fs_info, DEGRADED));
2140 req_q = bdev_get_queue(stripe->dev->bdev);
2141 if (!blk_queue_discard(req_q))
2144 ret = btrfs_issue_discard(stripe->dev->bdev,
2149 discarded_bytes += bytes;
2150 else if (ret != -EOPNOTSUPP)
2151 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2154 * Just in case we get back EOPNOTSUPP for some reason,
2155 * just ignore the return value so we don't screw up
2156 * people calling discard_extent.
2160 btrfs_put_bbio(bbio);
2162 btrfs_bio_counter_dec(fs_info);
2165 *actual_bytes = discarded_bytes;
2168 if (ret == -EOPNOTSUPP)
2173 /* Can return -ENOMEM */
2174 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2175 struct btrfs_root *root,
2176 u64 bytenr, u64 num_bytes, u64 parent,
2177 u64 root_objectid, u64 owner, u64 offset)
2179 struct btrfs_fs_info *fs_info = root->fs_info;
2180 int old_ref_mod, new_ref_mod;
2183 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
2184 root_objectid == BTRFS_TREE_LOG_OBJECTID);
2186 btrfs_ref_tree_mod(root, bytenr, num_bytes, parent, root_objectid,
2187 owner, offset, BTRFS_ADD_DELAYED_REF);
2189 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2190 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
2192 root_objectid, (int)owner,
2193 BTRFS_ADD_DELAYED_REF, NULL,
2194 &old_ref_mod, &new_ref_mod);
2196 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
2198 root_objectid, owner, offset,
2199 0, BTRFS_ADD_DELAYED_REF,
2200 &old_ref_mod, &new_ref_mod);
2203 if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0) {
2204 bool metadata = owner < BTRFS_FIRST_FREE_OBJECTID;
2206 add_pinned_bytes(fs_info, -num_bytes, metadata, root_objectid);
2212 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2213 struct btrfs_fs_info *fs_info,
2214 struct btrfs_delayed_ref_node *node,
2215 u64 parent, u64 root_objectid,
2216 u64 owner, u64 offset, int refs_to_add,
2217 struct btrfs_delayed_extent_op *extent_op)
2219 struct btrfs_path *path;
2220 struct extent_buffer *leaf;
2221 struct btrfs_extent_item *item;
2222 struct btrfs_key key;
2223 u64 bytenr = node->bytenr;
2224 u64 num_bytes = node->num_bytes;
2228 path = btrfs_alloc_path();
2232 path->reada = READA_FORWARD;
2233 path->leave_spinning = 1;
2234 /* this will setup the path even if it fails to insert the back ref */
2235 ret = insert_inline_extent_backref(trans, fs_info, path, bytenr,
2236 num_bytes, parent, root_objectid,
2238 refs_to_add, extent_op);
2239 if ((ret < 0 && ret != -EAGAIN) || !ret)
2243 * Ok we had -EAGAIN which means we didn't have space to insert and
2244 * inline extent ref, so just update the reference count and add a
2247 leaf = path->nodes[0];
2248 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2249 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2250 refs = btrfs_extent_refs(leaf, item);
2251 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2253 __run_delayed_extent_op(extent_op, leaf, item);
2255 btrfs_mark_buffer_dirty(leaf);
2256 btrfs_release_path(path);
2258 path->reada = READA_FORWARD;
2259 path->leave_spinning = 1;
2260 /* now insert the actual backref */
2261 ret = insert_extent_backref(trans, fs_info, path, bytenr, parent,
2262 root_objectid, owner, offset, refs_to_add);
2264 btrfs_abort_transaction(trans, ret);
2266 btrfs_free_path(path);
2270 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2271 struct btrfs_fs_info *fs_info,
2272 struct btrfs_delayed_ref_node *node,
2273 struct btrfs_delayed_extent_op *extent_op,
2274 int insert_reserved)
2277 struct btrfs_delayed_data_ref *ref;
2278 struct btrfs_key ins;
2283 ins.objectid = node->bytenr;
2284 ins.offset = node->num_bytes;
2285 ins.type = BTRFS_EXTENT_ITEM_KEY;
2287 ref = btrfs_delayed_node_to_data_ref(node);
2288 trace_run_delayed_data_ref(fs_info, node, ref, node->action);
2290 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2291 parent = ref->parent;
2292 ref_root = ref->root;
2294 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2296 flags |= extent_op->flags_to_set;
2297 ret = alloc_reserved_file_extent(trans, fs_info,
2298 parent, ref_root, flags,
2299 ref->objectid, ref->offset,
2300 &ins, node->ref_mod);
2301 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2302 ret = __btrfs_inc_extent_ref(trans, fs_info, node, parent,
2303 ref_root, ref->objectid,
2304 ref->offset, node->ref_mod,
2306 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2307 ret = __btrfs_free_extent(trans, fs_info, node, parent,
2308 ref_root, ref->objectid,
2309 ref->offset, node->ref_mod,
2317 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2318 struct extent_buffer *leaf,
2319 struct btrfs_extent_item *ei)
2321 u64 flags = btrfs_extent_flags(leaf, ei);
2322 if (extent_op->update_flags) {
2323 flags |= extent_op->flags_to_set;
2324 btrfs_set_extent_flags(leaf, ei, flags);
2327 if (extent_op->update_key) {
2328 struct btrfs_tree_block_info *bi;
2329 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2330 bi = (struct btrfs_tree_block_info *)(ei + 1);
2331 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2335 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2336 struct btrfs_fs_info *fs_info,
2337 struct btrfs_delayed_ref_head *head,
2338 struct btrfs_delayed_extent_op *extent_op)
2340 struct btrfs_key key;
2341 struct btrfs_path *path;
2342 struct btrfs_extent_item *ei;
2343 struct extent_buffer *leaf;
2347 int metadata = !extent_op->is_data;
2352 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2355 path = btrfs_alloc_path();
2359 key.objectid = head->bytenr;
2362 key.type = BTRFS_METADATA_ITEM_KEY;
2363 key.offset = extent_op->level;
2365 key.type = BTRFS_EXTENT_ITEM_KEY;
2366 key.offset = head->num_bytes;
2370 path->reada = READA_FORWARD;
2371 path->leave_spinning = 1;
2372 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1);
2379 if (path->slots[0] > 0) {
2381 btrfs_item_key_to_cpu(path->nodes[0], &key,
2383 if (key.objectid == head->bytenr &&
2384 key.type == BTRFS_EXTENT_ITEM_KEY &&
2385 key.offset == head->num_bytes)
2389 btrfs_release_path(path);
2392 key.objectid = head->bytenr;
2393 key.offset = head->num_bytes;
2394 key.type = BTRFS_EXTENT_ITEM_KEY;
2403 leaf = path->nodes[0];
2404 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2405 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2406 if (item_size < sizeof(*ei)) {
2407 ret = convert_extent_item_v0(trans, fs_info, path, (u64)-1, 0);
2412 leaf = path->nodes[0];
2413 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2416 BUG_ON(item_size < sizeof(*ei));
2417 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2418 __run_delayed_extent_op(extent_op, leaf, ei);
2420 btrfs_mark_buffer_dirty(leaf);
2422 btrfs_free_path(path);
2426 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2427 struct btrfs_fs_info *fs_info,
2428 struct btrfs_delayed_ref_node *node,
2429 struct btrfs_delayed_extent_op *extent_op,
2430 int insert_reserved)
2433 struct btrfs_delayed_tree_ref *ref;
2434 struct btrfs_key ins;
2437 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
2439 ref = btrfs_delayed_node_to_tree_ref(node);
2440 trace_run_delayed_tree_ref(fs_info, node, ref, node->action);
2442 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2443 parent = ref->parent;
2444 ref_root = ref->root;
2446 ins.objectid = node->bytenr;
2447 if (skinny_metadata) {
2448 ins.offset = ref->level;
2449 ins.type = BTRFS_METADATA_ITEM_KEY;
2451 ins.offset = node->num_bytes;
2452 ins.type = BTRFS_EXTENT_ITEM_KEY;
2455 if (node->ref_mod != 1) {
2457 "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu",
2458 node->bytenr, node->ref_mod, node->action, ref_root,
2462 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2463 BUG_ON(!extent_op || !extent_op->update_flags);
2464 ret = alloc_reserved_tree_block(trans, fs_info,
2466 extent_op->flags_to_set,
2469 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2470 ret = __btrfs_inc_extent_ref(trans, fs_info, node,
2474 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2475 ret = __btrfs_free_extent(trans, fs_info, node,
2477 ref->level, 0, 1, extent_op);
2484 /* helper function to actually process a single delayed ref entry */
2485 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2486 struct btrfs_fs_info *fs_info,
2487 struct btrfs_delayed_ref_node *node,
2488 struct btrfs_delayed_extent_op *extent_op,
2489 int insert_reserved)
2493 if (trans->aborted) {
2494 if (insert_reserved)
2495 btrfs_pin_extent(fs_info, node->bytenr,
2496 node->num_bytes, 1);
2500 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2501 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2502 ret = run_delayed_tree_ref(trans, fs_info, node, extent_op,
2504 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2505 node->type == BTRFS_SHARED_DATA_REF_KEY)
2506 ret = run_delayed_data_ref(trans, fs_info, node, extent_op,
2513 static inline struct btrfs_delayed_ref_node *
2514 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2516 struct btrfs_delayed_ref_node *ref;
2518 if (RB_EMPTY_ROOT(&head->ref_tree))
2522 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2523 * This is to prevent a ref count from going down to zero, which deletes
2524 * the extent item from the extent tree, when there still are references
2525 * to add, which would fail because they would not find the extent item.
2527 if (!list_empty(&head->ref_add_list))
2528 return list_first_entry(&head->ref_add_list,
2529 struct btrfs_delayed_ref_node, add_list);
2531 ref = rb_entry(rb_first(&head->ref_tree),
2532 struct btrfs_delayed_ref_node, ref_node);
2533 ASSERT(list_empty(&ref->add_list));
2537 static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
2538 struct btrfs_delayed_ref_head *head)
2540 spin_lock(&delayed_refs->lock);
2541 head->processing = 0;
2542 delayed_refs->num_heads_ready++;
2543 spin_unlock(&delayed_refs->lock);
2544 btrfs_delayed_ref_unlock(head);
2547 static int cleanup_extent_op(struct btrfs_trans_handle *trans,
2548 struct btrfs_fs_info *fs_info,
2549 struct btrfs_delayed_ref_head *head)
2551 struct btrfs_delayed_extent_op *extent_op = head->extent_op;
2556 head->extent_op = NULL;
2557 if (head->must_insert_reserved) {
2558 btrfs_free_delayed_extent_op(extent_op);
2561 spin_unlock(&head->lock);
2562 ret = run_delayed_extent_op(trans, fs_info, head, extent_op);
2563 btrfs_free_delayed_extent_op(extent_op);
2564 return ret ? ret : 1;
2567 static int cleanup_ref_head(struct btrfs_trans_handle *trans,
2568 struct btrfs_fs_info *fs_info,
2569 struct btrfs_delayed_ref_head *head)
2571 struct btrfs_delayed_ref_root *delayed_refs;
2574 delayed_refs = &trans->transaction->delayed_refs;
2576 ret = cleanup_extent_op(trans, fs_info, head);
2578 unselect_delayed_ref_head(delayed_refs, head);
2579 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2586 * Need to drop our head ref lock and re-acquire the delayed ref lock
2587 * and then re-check to make sure nobody got added.
2589 spin_unlock(&head->lock);
2590 spin_lock(&delayed_refs->lock);
2591 spin_lock(&head->lock);
2592 if (!RB_EMPTY_ROOT(&head->ref_tree) || head->extent_op) {
2593 spin_unlock(&head->lock);
2594 spin_unlock(&delayed_refs->lock);
2597 delayed_refs->num_heads--;
2598 rb_erase(&head->href_node, &delayed_refs->href_root);
2599 RB_CLEAR_NODE(&head->href_node);
2600 spin_unlock(&head->lock);
2601 spin_unlock(&delayed_refs->lock);
2602 atomic_dec(&delayed_refs->num_entries);
2604 trace_run_delayed_ref_head(fs_info, head, 0);
2606 if (head->total_ref_mod < 0) {
2607 struct btrfs_space_info *space_info;
2611 flags = BTRFS_BLOCK_GROUP_DATA;
2612 else if (head->is_system)
2613 flags = BTRFS_BLOCK_GROUP_SYSTEM;
2615 flags = BTRFS_BLOCK_GROUP_METADATA;
2616 space_info = __find_space_info(fs_info, flags);
2618 percpu_counter_add(&space_info->total_bytes_pinned,
2621 if (head->is_data) {
2622 spin_lock(&delayed_refs->lock);
2623 delayed_refs->pending_csums -= head->num_bytes;
2624 spin_unlock(&delayed_refs->lock);
2628 if (head->must_insert_reserved) {
2629 btrfs_pin_extent(fs_info, head->bytenr,
2630 head->num_bytes, 1);
2631 if (head->is_data) {
2632 ret = btrfs_del_csums(trans, fs_info, head->bytenr,
2637 /* Also free its reserved qgroup space */
2638 btrfs_qgroup_free_delayed_ref(fs_info, head->qgroup_ref_root,
2639 head->qgroup_reserved);
2640 btrfs_delayed_ref_unlock(head);
2641 btrfs_put_delayed_ref_head(head);
2646 * Returns 0 on success or if called with an already aborted transaction.
2647 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2649 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2652 struct btrfs_fs_info *fs_info = trans->fs_info;
2653 struct btrfs_delayed_ref_root *delayed_refs;
2654 struct btrfs_delayed_ref_node *ref;
2655 struct btrfs_delayed_ref_head *locked_ref = NULL;
2656 struct btrfs_delayed_extent_op *extent_op;
2657 ktime_t start = ktime_get();
2659 unsigned long count = 0;
2660 unsigned long actual_count = 0;
2661 int must_insert_reserved = 0;
2663 delayed_refs = &trans->transaction->delayed_refs;
2669 spin_lock(&delayed_refs->lock);
2670 locked_ref = btrfs_select_ref_head(trans);
2672 spin_unlock(&delayed_refs->lock);
2676 /* grab the lock that says we are going to process
2677 * all the refs for this head */
2678 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2679 spin_unlock(&delayed_refs->lock);
2681 * we may have dropped the spin lock to get the head
2682 * mutex lock, and that might have given someone else
2683 * time to free the head. If that's true, it has been
2684 * removed from our list and we can move on.
2686 if (ret == -EAGAIN) {
2694 * We need to try and merge add/drops of the same ref since we
2695 * can run into issues with relocate dropping the implicit ref
2696 * and then it being added back again before the drop can
2697 * finish. If we merged anything we need to re-loop so we can
2699 * Or we can get node references of the same type that weren't
2700 * merged when created due to bumps in the tree mod seq, and
2701 * we need to merge them to prevent adding an inline extent
2702 * backref before dropping it (triggering a BUG_ON at
2703 * insert_inline_extent_backref()).
2705 spin_lock(&locked_ref->lock);
2706 btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref);
2709 * locked_ref is the head node, so we have to go one
2710 * node back for any delayed ref updates
2712 ref = select_delayed_ref(locked_ref);
2714 if (ref && ref->seq &&
2715 btrfs_check_delayed_seq(fs_info, ref->seq)) {
2716 spin_unlock(&locked_ref->lock);
2717 unselect_delayed_ref_head(delayed_refs, locked_ref);
2725 * We're done processing refs in this ref_head, clean everything
2726 * up and move on to the next ref_head.
2729 ret = cleanup_ref_head(trans, fs_info, locked_ref);
2731 /* We dropped our lock, we need to loop. */
2744 rb_erase(&ref->ref_node, &locked_ref->ref_tree);
2745 RB_CLEAR_NODE(&ref->ref_node);
2746 if (!list_empty(&ref->add_list))
2747 list_del(&ref->add_list);
2749 * When we play the delayed ref, also correct the ref_mod on
2752 switch (ref->action) {
2753 case BTRFS_ADD_DELAYED_REF:
2754 case BTRFS_ADD_DELAYED_EXTENT:
2755 locked_ref->ref_mod -= ref->ref_mod;
2757 case BTRFS_DROP_DELAYED_REF:
2758 locked_ref->ref_mod += ref->ref_mod;
2763 atomic_dec(&delayed_refs->num_entries);
2766 * Record the must-insert_reserved flag before we drop the spin
2769 must_insert_reserved = locked_ref->must_insert_reserved;
2770 locked_ref->must_insert_reserved = 0;
2772 extent_op = locked_ref->extent_op;
2773 locked_ref->extent_op = NULL;
2774 spin_unlock(&locked_ref->lock);
2776 ret = run_one_delayed_ref(trans, fs_info, ref, extent_op,
2777 must_insert_reserved);
2779 btrfs_free_delayed_extent_op(extent_op);
2781 unselect_delayed_ref_head(delayed_refs, locked_ref);
2782 btrfs_put_delayed_ref(ref);
2783 btrfs_debug(fs_info, "run_one_delayed_ref returned %d",
2788 btrfs_put_delayed_ref(ref);
2794 * We don't want to include ref heads since we can have empty ref heads
2795 * and those will drastically skew our runtime down since we just do
2796 * accounting, no actual extent tree updates.
2798 if (actual_count > 0) {
2799 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2803 * We weigh the current average higher than our current runtime
2804 * to avoid large swings in the average.
2806 spin_lock(&delayed_refs->lock);
2807 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2808 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2809 spin_unlock(&delayed_refs->lock);
2814 #ifdef SCRAMBLE_DELAYED_REFS
2816 * Normally delayed refs get processed in ascending bytenr order. This
2817 * correlates in most cases to the order added. To expose dependencies on this
2818 * order, we start to process the tree in the middle instead of the beginning
2820 static u64 find_middle(struct rb_root *root)
2822 struct rb_node *n = root->rb_node;
2823 struct btrfs_delayed_ref_node *entry;
2826 u64 first = 0, last = 0;
2830 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2831 first = entry->bytenr;
2835 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2836 last = entry->bytenr;
2841 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2842 WARN_ON(!entry->in_tree);
2844 middle = entry->bytenr;
2857 static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads)
2861 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2862 sizeof(struct btrfs_extent_inline_ref));
2863 if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2864 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2867 * We don't ever fill up leaves all the way so multiply by 2 just to be
2868 * closer to what we're really going to want to use.
2870 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info));
2874 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2875 * would require to store the csums for that many bytes.
2877 u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes)
2880 u64 num_csums_per_leaf;
2883 csum_size = BTRFS_MAX_ITEM_SIZE(fs_info);
2884 num_csums_per_leaf = div64_u64(csum_size,
2885 (u64)btrfs_super_csum_size(fs_info->super_copy));
2886 num_csums = div64_u64(csum_bytes, fs_info->sectorsize);
2887 num_csums += num_csums_per_leaf - 1;
2888 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2892 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2893 struct btrfs_fs_info *fs_info)
2895 struct btrfs_block_rsv *global_rsv;
2896 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2897 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2898 unsigned int num_dirty_bgs = trans->transaction->num_dirty_bgs;
2899 u64 num_bytes, num_dirty_bgs_bytes;
2902 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
2903 num_heads = heads_to_leaves(fs_info, num_heads);
2905 num_bytes += (num_heads - 1) * fs_info->nodesize;
2907 num_bytes += btrfs_csum_bytes_to_leaves(fs_info, csum_bytes) *
2909 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(fs_info,
2911 global_rsv = &fs_info->global_block_rsv;
2914 * If we can't allocate any more chunks lets make sure we have _lots_ of
2915 * wiggle room since running delayed refs can create more delayed refs.
2917 if (global_rsv->space_info->full) {
2918 num_dirty_bgs_bytes <<= 1;
2922 spin_lock(&global_rsv->lock);
2923 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2925 spin_unlock(&global_rsv->lock);
2929 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2930 struct btrfs_fs_info *fs_info)
2933 atomic_read(&trans->transaction->delayed_refs.num_entries);
2938 avg_runtime = fs_info->avg_delayed_ref_runtime;
2939 val = num_entries * avg_runtime;
2940 if (val >= NSEC_PER_SEC)
2942 if (val >= NSEC_PER_SEC / 2)
2945 return btrfs_check_space_for_delayed_refs(trans, fs_info);
2948 struct async_delayed_refs {
2949 struct btrfs_root *root;
2954 struct completion wait;
2955 struct btrfs_work work;
2958 static inline struct async_delayed_refs *
2959 to_async_delayed_refs(struct btrfs_work *work)
2961 return container_of(work, struct async_delayed_refs, work);
2964 static void delayed_ref_async_start(struct btrfs_work *work)
2966 struct async_delayed_refs *async = to_async_delayed_refs(work);
2967 struct btrfs_trans_handle *trans;
2968 struct btrfs_fs_info *fs_info = async->root->fs_info;
2971 /* if the commit is already started, we don't need to wait here */
2972 if (btrfs_transaction_blocked(fs_info))
2975 trans = btrfs_join_transaction(async->root);
2976 if (IS_ERR(trans)) {
2977 async->error = PTR_ERR(trans);
2982 * trans->sync means that when we call end_transaction, we won't
2983 * wait on delayed refs
2987 /* Don't bother flushing if we got into a different transaction */
2988 if (trans->transid > async->transid)
2991 ret = btrfs_run_delayed_refs(trans, async->count);
2995 ret = btrfs_end_transaction(trans);
2996 if (ret && !async->error)
3000 complete(&async->wait);
3005 int btrfs_async_run_delayed_refs(struct btrfs_fs_info *fs_info,
3006 unsigned long count, u64 transid, int wait)
3008 struct async_delayed_refs *async;
3011 async = kmalloc(sizeof(*async), GFP_NOFS);
3015 async->root = fs_info->tree_root;
3016 async->count = count;
3018 async->transid = transid;
3023 init_completion(&async->wait);
3025 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
3026 delayed_ref_async_start, NULL, NULL);
3028 btrfs_queue_work(fs_info->extent_workers, &async->work);
3031 wait_for_completion(&async->wait);
3040 * this starts processing the delayed reference count updates and
3041 * extent insertions we have queued up so far. count can be
3042 * 0, which means to process everything in the tree at the start
3043 * of the run (but not newly added entries), or it can be some target
3044 * number you'd like to process.
3046 * Returns 0 on success or if called with an aborted transaction
3047 * Returns <0 on error and aborts the transaction
3049 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
3050 unsigned long count)
3052 struct btrfs_fs_info *fs_info = trans->fs_info;
3053 struct rb_node *node;
3054 struct btrfs_delayed_ref_root *delayed_refs;
3055 struct btrfs_delayed_ref_head *head;
3057 int run_all = count == (unsigned long)-1;
3058 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
3060 /* We'll clean this up in btrfs_cleanup_transaction */
3064 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
3067 delayed_refs = &trans->transaction->delayed_refs;
3069 count = atomic_read(&delayed_refs->num_entries) * 2;
3072 #ifdef SCRAMBLE_DELAYED_REFS
3073 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
3075 trans->can_flush_pending_bgs = false;
3076 ret = __btrfs_run_delayed_refs(trans, count);
3078 btrfs_abort_transaction(trans, ret);
3083 if (!list_empty(&trans->new_bgs))
3084 btrfs_create_pending_block_groups(trans);
3086 spin_lock(&delayed_refs->lock);
3087 node = rb_first(&delayed_refs->href_root);
3089 spin_unlock(&delayed_refs->lock);
3092 head = rb_entry(node, struct btrfs_delayed_ref_head,
3094 refcount_inc(&head->refs);
3095 spin_unlock(&delayed_refs->lock);
3097 /* Mutex was contended, block until it's released and retry. */
3098 mutex_lock(&head->mutex);
3099 mutex_unlock(&head->mutex);
3101 btrfs_put_delayed_ref_head(head);
3106 trans->can_flush_pending_bgs = can_flush_pending_bgs;
3110 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
3111 struct btrfs_fs_info *fs_info,
3112 u64 bytenr, u64 num_bytes, u64 flags,
3113 int level, int is_data)
3115 struct btrfs_delayed_extent_op *extent_op;
3118 extent_op = btrfs_alloc_delayed_extent_op();
3122 extent_op->flags_to_set = flags;
3123 extent_op->update_flags = true;
3124 extent_op->update_key = false;
3125 extent_op->is_data = is_data ? true : false;
3126 extent_op->level = level;
3128 ret = btrfs_add_delayed_extent_op(fs_info, trans, bytenr,
3129 num_bytes, extent_op);
3131 btrfs_free_delayed_extent_op(extent_op);
3135 static noinline int check_delayed_ref(struct btrfs_root *root,
3136 struct btrfs_path *path,
3137 u64 objectid, u64 offset, u64 bytenr)
3139 struct btrfs_delayed_ref_head *head;
3140 struct btrfs_delayed_ref_node *ref;
3141 struct btrfs_delayed_data_ref *data_ref;
3142 struct btrfs_delayed_ref_root *delayed_refs;
3143 struct btrfs_transaction *cur_trans;
3144 struct rb_node *node;
3147 spin_lock(&root->fs_info->trans_lock);
3148 cur_trans = root->fs_info->running_transaction;
3150 refcount_inc(&cur_trans->use_count);
3151 spin_unlock(&root->fs_info->trans_lock);
3155 delayed_refs = &cur_trans->delayed_refs;
3156 spin_lock(&delayed_refs->lock);
3157 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
3159 spin_unlock(&delayed_refs->lock);
3160 btrfs_put_transaction(cur_trans);
3164 if (!mutex_trylock(&head->mutex)) {
3165 refcount_inc(&head->refs);
3166 spin_unlock(&delayed_refs->lock);
3168 btrfs_release_path(path);
3171 * Mutex was contended, block until it's released and let
3174 mutex_lock(&head->mutex);
3175 mutex_unlock(&head->mutex);
3176 btrfs_put_delayed_ref_head(head);
3177 btrfs_put_transaction(cur_trans);
3180 spin_unlock(&delayed_refs->lock);
3182 spin_lock(&head->lock);
3184 * XXX: We should replace this with a proper search function in the
3187 for (node = rb_first(&head->ref_tree); node; node = rb_next(node)) {
3188 ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node);
3189 /* If it's a shared ref we know a cross reference exists */
3190 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
3195 data_ref = btrfs_delayed_node_to_data_ref(ref);
3198 * If our ref doesn't match the one we're currently looking at
3199 * then we have a cross reference.
3201 if (data_ref->root != root->root_key.objectid ||
3202 data_ref->objectid != objectid ||
3203 data_ref->offset != offset) {
3208 spin_unlock(&head->lock);
3209 mutex_unlock(&head->mutex);
3210 btrfs_put_transaction(cur_trans);
3214 static noinline int check_committed_ref(struct btrfs_root *root,
3215 struct btrfs_path *path,
3216 u64 objectid, u64 offset, u64 bytenr)
3218 struct btrfs_fs_info *fs_info = root->fs_info;
3219 struct btrfs_root *extent_root = fs_info->extent_root;
3220 struct extent_buffer *leaf;
3221 struct btrfs_extent_data_ref *ref;
3222 struct btrfs_extent_inline_ref *iref;
3223 struct btrfs_extent_item *ei;
3224 struct btrfs_key key;
3229 key.objectid = bytenr;
3230 key.offset = (u64)-1;
3231 key.type = BTRFS_EXTENT_ITEM_KEY;
3233 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3236 BUG_ON(ret == 0); /* Corruption */
3239 if (path->slots[0] == 0)
3243 leaf = path->nodes[0];
3244 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3246 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3250 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3251 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3252 if (item_size < sizeof(*ei)) {
3253 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3257 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3259 if (item_size != sizeof(*ei) +
3260 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3263 if (btrfs_extent_generation(leaf, ei) <=
3264 btrfs_root_last_snapshot(&root->root_item))
3267 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3269 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
3270 if (type != BTRFS_EXTENT_DATA_REF_KEY)
3273 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3274 if (btrfs_extent_refs(leaf, ei) !=
3275 btrfs_extent_data_ref_count(leaf, ref) ||
3276 btrfs_extent_data_ref_root(leaf, ref) !=
3277 root->root_key.objectid ||
3278 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3279 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3287 int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
3290 struct btrfs_path *path;
3294 path = btrfs_alloc_path();
3299 ret = check_committed_ref(root, path, objectid,
3301 if (ret && ret != -ENOENT)
3304 ret2 = check_delayed_ref(root, path, objectid,
3306 } while (ret2 == -EAGAIN);
3308 if (ret2 && ret2 != -ENOENT) {
3313 if (ret != -ENOENT || ret2 != -ENOENT)
3316 btrfs_free_path(path);
3317 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3322 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3323 struct btrfs_root *root,
3324 struct extent_buffer *buf,
3325 int full_backref, int inc)
3327 struct btrfs_fs_info *fs_info = root->fs_info;
3333 struct btrfs_key key;
3334 struct btrfs_file_extent_item *fi;
3338 int (*process_func)(struct btrfs_trans_handle *,
3339 struct btrfs_root *,
3340 u64, u64, u64, u64, u64, u64);
3343 if (btrfs_is_testing(fs_info))
3346 ref_root = btrfs_header_owner(buf);
3347 nritems = btrfs_header_nritems(buf);
3348 level = btrfs_header_level(buf);
3350 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3354 process_func = btrfs_inc_extent_ref;
3356 process_func = btrfs_free_extent;
3359 parent = buf->start;
3363 for (i = 0; i < nritems; i++) {
3365 btrfs_item_key_to_cpu(buf, &key, i);
3366 if (key.type != BTRFS_EXTENT_DATA_KEY)
3368 fi = btrfs_item_ptr(buf, i,
3369 struct btrfs_file_extent_item);
3370 if (btrfs_file_extent_type(buf, fi) ==
3371 BTRFS_FILE_EXTENT_INLINE)
3373 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3377 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3378 key.offset -= btrfs_file_extent_offset(buf, fi);
3379 ret = process_func(trans, root, bytenr, num_bytes,
3380 parent, ref_root, key.objectid,
3385 bytenr = btrfs_node_blockptr(buf, i);
3386 num_bytes = fs_info->nodesize;
3387 ret = process_func(trans, root, bytenr, num_bytes,
3388 parent, ref_root, level - 1, 0);
3398 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3399 struct extent_buffer *buf, int full_backref)
3401 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3404 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3405 struct extent_buffer *buf, int full_backref)
3407 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3410 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3411 struct btrfs_fs_info *fs_info,
3412 struct btrfs_path *path,
3413 struct btrfs_block_group_cache *cache)
3416 struct btrfs_root *extent_root = fs_info->extent_root;
3418 struct extent_buffer *leaf;
3420 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3427 leaf = path->nodes[0];
3428 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3429 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3430 btrfs_mark_buffer_dirty(leaf);
3432 btrfs_release_path(path);
3437 static struct btrfs_block_group_cache *
3438 next_block_group(struct btrfs_fs_info *fs_info,
3439 struct btrfs_block_group_cache *cache)
3441 struct rb_node *node;
3443 spin_lock(&fs_info->block_group_cache_lock);
3445 /* If our block group was removed, we need a full search. */
3446 if (RB_EMPTY_NODE(&cache->cache_node)) {
3447 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3449 spin_unlock(&fs_info->block_group_cache_lock);
3450 btrfs_put_block_group(cache);
3451 cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache;
3453 node = rb_next(&cache->cache_node);
3454 btrfs_put_block_group(cache);
3456 cache = rb_entry(node, struct btrfs_block_group_cache,
3458 btrfs_get_block_group(cache);
3461 spin_unlock(&fs_info->block_group_cache_lock);
3465 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3466 struct btrfs_trans_handle *trans,
3467 struct btrfs_path *path)
3469 struct btrfs_fs_info *fs_info = block_group->fs_info;
3470 struct btrfs_root *root = fs_info->tree_root;
3471 struct inode *inode = NULL;
3472 struct extent_changeset *data_reserved = NULL;
3474 int dcs = BTRFS_DC_ERROR;
3480 * If this block group is smaller than 100 megs don't bother caching the
3483 if (block_group->key.offset < (100 * SZ_1M)) {
3484 spin_lock(&block_group->lock);
3485 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3486 spin_unlock(&block_group->lock);
3493 inode = lookup_free_space_inode(fs_info, block_group, path);
3494 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3495 ret = PTR_ERR(inode);
3496 btrfs_release_path(path);
3500 if (IS_ERR(inode)) {
3504 if (block_group->ro)
3507 ret = create_free_space_inode(fs_info, trans, block_group,
3515 * We want to set the generation to 0, that way if anything goes wrong
3516 * from here on out we know not to trust this cache when we load up next
3519 BTRFS_I(inode)->generation = 0;
3520 ret = btrfs_update_inode(trans, root, inode);
3523 * So theoretically we could recover from this, simply set the
3524 * super cache generation to 0 so we know to invalidate the
3525 * cache, but then we'd have to keep track of the block groups
3526 * that fail this way so we know we _have_ to reset this cache
3527 * before the next commit or risk reading stale cache. So to
3528 * limit our exposure to horrible edge cases lets just abort the
3529 * transaction, this only happens in really bad situations
3532 btrfs_abort_transaction(trans, ret);
3537 /* We've already setup this transaction, go ahead and exit */
3538 if (block_group->cache_generation == trans->transid &&
3539 i_size_read(inode)) {
3540 dcs = BTRFS_DC_SETUP;
3544 if (i_size_read(inode) > 0) {
3545 ret = btrfs_check_trunc_cache_free_space(fs_info,
3546 &fs_info->global_block_rsv);
3550 ret = btrfs_truncate_free_space_cache(trans, NULL, inode);
3555 spin_lock(&block_group->lock);
3556 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3557 !btrfs_test_opt(fs_info, SPACE_CACHE)) {
3559 * don't bother trying to write stuff out _if_
3560 * a) we're not cached,
3561 * b) we're with nospace_cache mount option,
3562 * c) we're with v2 space_cache (FREE_SPACE_TREE).
3564 dcs = BTRFS_DC_WRITTEN;
3565 spin_unlock(&block_group->lock);
3568 spin_unlock(&block_group->lock);
3571 * We hit an ENOSPC when setting up the cache in this transaction, just
3572 * skip doing the setup, we've already cleared the cache so we're safe.
3574 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
3580 * Try to preallocate enough space based on how big the block group is.
3581 * Keep in mind this has to include any pinned space which could end up
3582 * taking up quite a bit since it's not folded into the other space
3585 num_pages = div_u64(block_group->key.offset, SZ_256M);
3590 num_pages *= PAGE_SIZE;
3592 ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages);
3596 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3597 num_pages, num_pages,
3600 * Our cache requires contiguous chunks so that we don't modify a bunch
3601 * of metadata or split extents when writing the cache out, which means
3602 * we can enospc if we are heavily fragmented in addition to just normal
3603 * out of space conditions. So if we hit this just skip setting up any
3604 * other block groups for this transaction, maybe we'll unpin enough
3605 * space the next time around.
3608 dcs = BTRFS_DC_SETUP;
3609 else if (ret == -ENOSPC)
3610 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
3615 btrfs_release_path(path);
3617 spin_lock(&block_group->lock);
3618 if (!ret && dcs == BTRFS_DC_SETUP)
3619 block_group->cache_generation = trans->transid;
3620 block_group->disk_cache_state = dcs;
3621 spin_unlock(&block_group->lock);
3623 extent_changeset_free(data_reserved);
3627 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3628 struct btrfs_fs_info *fs_info)
3630 struct btrfs_block_group_cache *cache, *tmp;
3631 struct btrfs_transaction *cur_trans = trans->transaction;
3632 struct btrfs_path *path;
3634 if (list_empty(&cur_trans->dirty_bgs) ||
3635 !btrfs_test_opt(fs_info, SPACE_CACHE))
3638 path = btrfs_alloc_path();
3642 /* Could add new block groups, use _safe just in case */
3643 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3645 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3646 cache_save_setup(cache, trans, path);
3649 btrfs_free_path(path);
3654 * transaction commit does final block group cache writeback during a
3655 * critical section where nothing is allowed to change the FS. This is
3656 * required in order for the cache to actually match the block group,
3657 * but can introduce a lot of latency into the commit.
3659 * So, btrfs_start_dirty_block_groups is here to kick off block group
3660 * cache IO. There's a chance we'll have to redo some of it if the
3661 * block group changes again during the commit, but it greatly reduces
3662 * the commit latency by getting rid of the easy block groups while
3663 * we're still allowing others to join the commit.
3665 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans)
3667 struct btrfs_fs_info *fs_info = trans->fs_info;
3668 struct btrfs_block_group_cache *cache;
3669 struct btrfs_transaction *cur_trans = trans->transaction;
3672 struct btrfs_path *path = NULL;
3674 struct list_head *io = &cur_trans->io_bgs;
3675 int num_started = 0;
3678 spin_lock(&cur_trans->dirty_bgs_lock);
3679 if (list_empty(&cur_trans->dirty_bgs)) {
3680 spin_unlock(&cur_trans->dirty_bgs_lock);
3683 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3684 spin_unlock(&cur_trans->dirty_bgs_lock);
3688 * make sure all the block groups on our dirty list actually
3691 btrfs_create_pending_block_groups(trans);
3694 path = btrfs_alloc_path();
3700 * cache_write_mutex is here only to save us from balance or automatic
3701 * removal of empty block groups deleting this block group while we are
3702 * writing out the cache
3704 mutex_lock(&trans->transaction->cache_write_mutex);
3705 while (!list_empty(&dirty)) {
3706 cache = list_first_entry(&dirty,
3707 struct btrfs_block_group_cache,
3710 * this can happen if something re-dirties a block
3711 * group that is already under IO. Just wait for it to
3712 * finish and then do it all again
3714 if (!list_empty(&cache->io_list)) {
3715 list_del_init(&cache->io_list);
3716 btrfs_wait_cache_io(trans, cache, path);
3717 btrfs_put_block_group(cache);
3722 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3723 * if it should update the cache_state. Don't delete
3724 * until after we wait.
3726 * Since we're not running in the commit critical section
3727 * we need the dirty_bgs_lock to protect from update_block_group
3729 spin_lock(&cur_trans->dirty_bgs_lock);
3730 list_del_init(&cache->dirty_list);
3731 spin_unlock(&cur_trans->dirty_bgs_lock);
3735 cache_save_setup(cache, trans, path);
3737 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3738 cache->io_ctl.inode = NULL;
3739 ret = btrfs_write_out_cache(fs_info, trans,
3741 if (ret == 0 && cache->io_ctl.inode) {
3746 * The cache_write_mutex is protecting the
3747 * io_list, also refer to the definition of
3748 * btrfs_transaction::io_bgs for more details
3750 list_add_tail(&cache->io_list, io);
3753 * if we failed to write the cache, the
3754 * generation will be bad and life goes on
3760 ret = write_one_cache_group(trans, fs_info,
3763 * Our block group might still be attached to the list
3764 * of new block groups in the transaction handle of some
3765 * other task (struct btrfs_trans_handle->new_bgs). This
3766 * means its block group item isn't yet in the extent
3767 * tree. If this happens ignore the error, as we will
3768 * try again later in the critical section of the
3769 * transaction commit.
3771 if (ret == -ENOENT) {
3773 spin_lock(&cur_trans->dirty_bgs_lock);
3774 if (list_empty(&cache->dirty_list)) {
3775 list_add_tail(&cache->dirty_list,
3776 &cur_trans->dirty_bgs);
3777 btrfs_get_block_group(cache);
3779 spin_unlock(&cur_trans->dirty_bgs_lock);
3781 btrfs_abort_transaction(trans, ret);
3785 /* if its not on the io list, we need to put the block group */
3787 btrfs_put_block_group(cache);
3793 * Avoid blocking other tasks for too long. It might even save
3794 * us from writing caches for block groups that are going to be
3797 mutex_unlock(&trans->transaction->cache_write_mutex);
3798 mutex_lock(&trans->transaction->cache_write_mutex);
3800 mutex_unlock(&trans->transaction->cache_write_mutex);
3803 * go through delayed refs for all the stuff we've just kicked off
3804 * and then loop back (just once)
3806 ret = btrfs_run_delayed_refs(trans, 0);
3807 if (!ret && loops == 0) {
3809 spin_lock(&cur_trans->dirty_bgs_lock);
3810 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3812 * dirty_bgs_lock protects us from concurrent block group
3813 * deletes too (not just cache_write_mutex).
3815 if (!list_empty(&dirty)) {
3816 spin_unlock(&cur_trans->dirty_bgs_lock);
3819 spin_unlock(&cur_trans->dirty_bgs_lock);
3820 } else if (ret < 0) {
3821 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
3824 btrfs_free_path(path);
3828 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3829 struct btrfs_fs_info *fs_info)
3831 struct btrfs_block_group_cache *cache;
3832 struct btrfs_transaction *cur_trans = trans->transaction;
3835 struct btrfs_path *path;
3836 struct list_head *io = &cur_trans->io_bgs;
3837 int num_started = 0;
3839 path = btrfs_alloc_path();
3844 * Even though we are in the critical section of the transaction commit,
3845 * we can still have concurrent tasks adding elements to this
3846 * transaction's list of dirty block groups. These tasks correspond to
3847 * endio free space workers started when writeback finishes for a
3848 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3849 * allocate new block groups as a result of COWing nodes of the root
3850 * tree when updating the free space inode. The writeback for the space
3851 * caches is triggered by an earlier call to
3852 * btrfs_start_dirty_block_groups() and iterations of the following
3854 * Also we want to do the cache_save_setup first and then run the
3855 * delayed refs to make sure we have the best chance at doing this all
3858 spin_lock(&cur_trans->dirty_bgs_lock);
3859 while (!list_empty(&cur_trans->dirty_bgs)) {
3860 cache = list_first_entry(&cur_trans->dirty_bgs,
3861 struct btrfs_block_group_cache,
3865 * this can happen if cache_save_setup re-dirties a block
3866 * group that is already under IO. Just wait for it to
3867 * finish and then do it all again
3869 if (!list_empty(&cache->io_list)) {
3870 spin_unlock(&cur_trans->dirty_bgs_lock);
3871 list_del_init(&cache->io_list);
3872 btrfs_wait_cache_io(trans, cache, path);
3873 btrfs_put_block_group(cache);
3874 spin_lock(&cur_trans->dirty_bgs_lock);
3878 * don't remove from the dirty list until after we've waited
3881 list_del_init(&cache->dirty_list);
3882 spin_unlock(&cur_trans->dirty_bgs_lock);
3885 cache_save_setup(cache, trans, path);
3888 ret = btrfs_run_delayed_refs(trans,
3889 (unsigned long) -1);
3891 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3892 cache->io_ctl.inode = NULL;
3893 ret = btrfs_write_out_cache(fs_info, trans,
3895 if (ret == 0 && cache->io_ctl.inode) {
3898 list_add_tail(&cache->io_list, io);
3901 * if we failed to write the cache, the
3902 * generation will be bad and life goes on
3908 ret = write_one_cache_group(trans, fs_info,
3911 * One of the free space endio workers might have
3912 * created a new block group while updating a free space
3913 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3914 * and hasn't released its transaction handle yet, in
3915 * which case the new block group is still attached to
3916 * its transaction handle and its creation has not
3917 * finished yet (no block group item in the extent tree
3918 * yet, etc). If this is the case, wait for all free
3919 * space endio workers to finish and retry. This is a
3920 * a very rare case so no need for a more efficient and
3923 if (ret == -ENOENT) {
3924 wait_event(cur_trans->writer_wait,
3925 atomic_read(&cur_trans->num_writers) == 1);
3926 ret = write_one_cache_group(trans, fs_info,
3930 btrfs_abort_transaction(trans, ret);
3933 /* if its not on the io list, we need to put the block group */
3935 btrfs_put_block_group(cache);
3936 spin_lock(&cur_trans->dirty_bgs_lock);
3938 spin_unlock(&cur_trans->dirty_bgs_lock);
3941 * Refer to the definition of io_bgs member for details why it's safe
3942 * to use it without any locking
3944 while (!list_empty(io)) {
3945 cache = list_first_entry(io, struct btrfs_block_group_cache,
3947 list_del_init(&cache->io_list);
3948 btrfs_wait_cache_io(trans, cache, path);
3949 btrfs_put_block_group(cache);
3952 btrfs_free_path(path);
3956 int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
3958 struct btrfs_block_group_cache *block_group;
3961 block_group = btrfs_lookup_block_group(fs_info, bytenr);
3962 if (!block_group || block_group->ro)
3965 btrfs_put_block_group(block_group);
3969 bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3971 struct btrfs_block_group_cache *bg;
3974 bg = btrfs_lookup_block_group(fs_info, bytenr);
3978 spin_lock(&bg->lock);
3982 atomic_inc(&bg->nocow_writers);
3983 spin_unlock(&bg->lock);
3985 /* no put on block group, done by btrfs_dec_nocow_writers */
3987 btrfs_put_block_group(bg);
3993 void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3995 struct btrfs_block_group_cache *bg;
3997 bg = btrfs_lookup_block_group(fs_info, bytenr);
3999 if (atomic_dec_and_test(&bg->nocow_writers))
4000 wake_up_var(&bg->nocow_writers);
4002 * Once for our lookup and once for the lookup done by a previous call
4003 * to btrfs_inc_nocow_writers()
4005 btrfs_put_block_group(bg);
4006 btrfs_put_block_group(bg);
4009 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg)
4011 wait_var_event(&bg->nocow_writers, !atomic_read(&bg->nocow_writers));
4014 static const char *alloc_name(u64 flags)
4017 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
4019 case BTRFS_BLOCK_GROUP_METADATA:
4021 case BTRFS_BLOCK_GROUP_DATA:
4023 case BTRFS_BLOCK_GROUP_SYSTEM:
4027 return "invalid-combination";
4031 static int create_space_info(struct btrfs_fs_info *info, u64 flags,
4032 struct btrfs_space_info **new)
4035 struct btrfs_space_info *space_info;
4039 space_info = kzalloc(sizeof(*space_info), GFP_NOFS);
4043 ret = percpu_counter_init(&space_info->total_bytes_pinned, 0,
4050 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
4051 INIT_LIST_HEAD(&space_info->block_groups[i]);
4052 init_rwsem(&space_info->groups_sem);
4053 spin_lock_init(&space_info->lock);
4054 space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
4055 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4056 init_waitqueue_head(&space_info->wait);
4057 INIT_LIST_HEAD(&space_info->ro_bgs);
4058 INIT_LIST_HEAD(&space_info->tickets);
4059 INIT_LIST_HEAD(&space_info->priority_tickets);
4061 ret = kobject_init_and_add(&space_info->kobj, &space_info_ktype,
4062 info->space_info_kobj, "%s",
4063 alloc_name(space_info->flags));
4065 percpu_counter_destroy(&space_info->total_bytes_pinned);
4071 list_add_rcu(&space_info->list, &info->space_info);
4072 if (flags & BTRFS_BLOCK_GROUP_DATA)
4073 info->data_sinfo = space_info;
4078 static void update_space_info(struct btrfs_fs_info *info, u64 flags,
4079 u64 total_bytes, u64 bytes_used,
4081 struct btrfs_space_info **space_info)
4083 struct btrfs_space_info *found;
4086 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
4087 BTRFS_BLOCK_GROUP_RAID10))
4092 found = __find_space_info(info, flags);
4094 spin_lock(&found->lock);
4095 found->total_bytes += total_bytes;
4096 found->disk_total += total_bytes * factor;
4097 found->bytes_used += bytes_used;
4098 found->disk_used += bytes_used * factor;
4099 found->bytes_readonly += bytes_readonly;
4100 if (total_bytes > 0)
4102 space_info_add_new_bytes(info, found, total_bytes -
4103 bytes_used - bytes_readonly);
4104 spin_unlock(&found->lock);
4105 *space_info = found;
4108 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
4110 u64 extra_flags = chunk_to_extended(flags) &
4111 BTRFS_EXTENDED_PROFILE_MASK;
4113 write_seqlock(&fs_info->profiles_lock);
4114 if (flags & BTRFS_BLOCK_GROUP_DATA)
4115 fs_info->avail_data_alloc_bits |= extra_flags;
4116 if (flags & BTRFS_BLOCK_GROUP_METADATA)
4117 fs_info->avail_metadata_alloc_bits |= extra_flags;
4118 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4119 fs_info->avail_system_alloc_bits |= extra_flags;
4120 write_sequnlock(&fs_info->profiles_lock);
4124 * returns target flags in extended format or 0 if restripe for this
4125 * chunk_type is not in progress
4127 * should be called with balance_lock held
4129 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
4131 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4137 if (flags & BTRFS_BLOCK_GROUP_DATA &&
4138 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4139 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
4140 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
4141 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4142 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
4143 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
4144 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4145 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
4152 * @flags: available profiles in extended format (see ctree.h)
4154 * Returns reduced profile in chunk format. If profile changing is in
4155 * progress (either running or paused) picks the target profile (if it's
4156 * already available), otherwise falls back to plain reducing.
4158 static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags)
4160 u64 num_devices = fs_info->fs_devices->rw_devices;
4166 * see if restripe for this chunk_type is in progress, if so
4167 * try to reduce to the target profile
4169 spin_lock(&fs_info->balance_lock);
4170 target = get_restripe_target(fs_info, flags);
4172 /* pick target profile only if it's already available */
4173 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
4174 spin_unlock(&fs_info->balance_lock);
4175 return extended_to_chunk(target);
4178 spin_unlock(&fs_info->balance_lock);
4180 /* First, mask out the RAID levels which aren't possible */
4181 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4182 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
4183 allowed |= btrfs_raid_array[raid_type].bg_flag;
4187 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
4188 allowed = BTRFS_BLOCK_GROUP_RAID6;
4189 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
4190 allowed = BTRFS_BLOCK_GROUP_RAID5;
4191 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
4192 allowed = BTRFS_BLOCK_GROUP_RAID10;
4193 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
4194 allowed = BTRFS_BLOCK_GROUP_RAID1;
4195 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
4196 allowed = BTRFS_BLOCK_GROUP_RAID0;
4198 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
4200 return extended_to_chunk(flags | allowed);
4203 static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags)
4210 seq = read_seqbegin(&fs_info->profiles_lock);
4212 if (flags & BTRFS_BLOCK_GROUP_DATA)
4213 flags |= fs_info->avail_data_alloc_bits;
4214 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4215 flags |= fs_info->avail_system_alloc_bits;
4216 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
4217 flags |= fs_info->avail_metadata_alloc_bits;
4218 } while (read_seqretry(&fs_info->profiles_lock, seq));
4220 return btrfs_reduce_alloc_profile(fs_info, flags);
4223 static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
4225 struct btrfs_fs_info *fs_info = root->fs_info;
4230 flags = BTRFS_BLOCK_GROUP_DATA;
4231 else if (root == fs_info->chunk_root)
4232 flags = BTRFS_BLOCK_GROUP_SYSTEM;
4234 flags = BTRFS_BLOCK_GROUP_METADATA;
4236 ret = get_alloc_profile(fs_info, flags);
4240 u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info)
4242 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA);
4245 u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info)
4247 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4250 u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info)
4252 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4255 static u64 btrfs_space_info_used(struct btrfs_space_info *s_info,
4256 bool may_use_included)
4259 return s_info->bytes_used + s_info->bytes_reserved +
4260 s_info->bytes_pinned + s_info->bytes_readonly +
4261 (may_use_included ? s_info->bytes_may_use : 0);
4264 int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes)
4266 struct btrfs_root *root = inode->root;
4267 struct btrfs_fs_info *fs_info = root->fs_info;
4268 struct btrfs_space_info *data_sinfo = fs_info->data_sinfo;
4271 int need_commit = 2;
4272 int have_pinned_space;
4274 /* make sure bytes are sectorsize aligned */
4275 bytes = ALIGN(bytes, fs_info->sectorsize);
4277 if (btrfs_is_free_space_inode(inode)) {
4279 ASSERT(current->journal_info);
4283 /* make sure we have enough space to handle the data first */
4284 spin_lock(&data_sinfo->lock);
4285 used = btrfs_space_info_used(data_sinfo, true);
4287 if (used + bytes > data_sinfo->total_bytes) {
4288 struct btrfs_trans_handle *trans;
4291 * if we don't have enough free bytes in this space then we need
4292 * to alloc a new chunk.
4294 if (!data_sinfo->full) {
4297 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4298 spin_unlock(&data_sinfo->lock);
4300 alloc_target = btrfs_data_alloc_profile(fs_info);
4302 * It is ugly that we don't call nolock join
4303 * transaction for the free space inode case here.
4304 * But it is safe because we only do the data space
4305 * reservation for the free space cache in the
4306 * transaction context, the common join transaction
4307 * just increase the counter of the current transaction
4308 * handler, doesn't try to acquire the trans_lock of
4311 trans = btrfs_join_transaction(root);
4313 return PTR_ERR(trans);
4315 ret = do_chunk_alloc(trans, fs_info, alloc_target,
4316 CHUNK_ALLOC_NO_FORCE);
4317 btrfs_end_transaction(trans);
4322 have_pinned_space = 1;
4331 * If we don't have enough pinned space to deal with this
4332 * allocation, and no removed chunk in current transaction,
4333 * don't bother committing the transaction.
4335 have_pinned_space = percpu_counter_compare(
4336 &data_sinfo->total_bytes_pinned,
4337 used + bytes - data_sinfo->total_bytes);
4338 spin_unlock(&data_sinfo->lock);
4340 /* commit the current transaction and try again */
4345 if (need_commit > 0) {
4346 btrfs_start_delalloc_roots(fs_info, 0, -1);
4347 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0,
4351 trans = btrfs_join_transaction(root);
4353 return PTR_ERR(trans);
4354 if (have_pinned_space >= 0 ||
4355 test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4356 &trans->transaction->flags) ||
4358 ret = btrfs_commit_transaction(trans);
4362 * The cleaner kthread might still be doing iput
4363 * operations. Wait for it to finish so that
4364 * more space is released.
4366 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
4367 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
4370 btrfs_end_transaction(trans);
4374 trace_btrfs_space_reservation(fs_info,
4375 "space_info:enospc",
4376 data_sinfo->flags, bytes, 1);
4379 data_sinfo->bytes_may_use += bytes;
4380 trace_btrfs_space_reservation(fs_info, "space_info",
4381 data_sinfo->flags, bytes, 1);
4382 spin_unlock(&data_sinfo->lock);
4387 int btrfs_check_data_free_space(struct inode *inode,
4388 struct extent_changeset **reserved, u64 start, u64 len)
4390 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4393 /* align the range */
4394 len = round_up(start + len, fs_info->sectorsize) -
4395 round_down(start, fs_info->sectorsize);
4396 start = round_down(start, fs_info->sectorsize);
4398 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len);
4402 /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */
4403 ret = btrfs_qgroup_reserve_data(inode, reserved, start, len);
4405 btrfs_free_reserved_data_space_noquota(inode, start, len);
4412 * Called if we need to clear a data reservation for this inode
4413 * Normally in a error case.
4415 * This one will *NOT* use accurate qgroup reserved space API, just for case
4416 * which we can't sleep and is sure it won't affect qgroup reserved space.
4417 * Like clear_bit_hook().
4419 void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4422 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4423 struct btrfs_space_info *data_sinfo;
4425 /* Make sure the range is aligned to sectorsize */
4426 len = round_up(start + len, fs_info->sectorsize) -
4427 round_down(start, fs_info->sectorsize);
4428 start = round_down(start, fs_info->sectorsize);
4430 data_sinfo = fs_info->data_sinfo;
4431 spin_lock(&data_sinfo->lock);
4432 if (WARN_ON(data_sinfo->bytes_may_use < len))
4433 data_sinfo->bytes_may_use = 0;
4435 data_sinfo->bytes_may_use -= len;
4436 trace_btrfs_space_reservation(fs_info, "space_info",
4437 data_sinfo->flags, len, 0);
4438 spin_unlock(&data_sinfo->lock);
4442 * Called if we need to clear a data reservation for this inode
4443 * Normally in a error case.
4445 * This one will handle the per-inode data rsv map for accurate reserved
4448 void btrfs_free_reserved_data_space(struct inode *inode,
4449 struct extent_changeset *reserved, u64 start, u64 len)
4451 struct btrfs_root *root = BTRFS_I(inode)->root;
4453 /* Make sure the range is aligned to sectorsize */
4454 len = round_up(start + len, root->fs_info->sectorsize) -
4455 round_down(start, root->fs_info->sectorsize);
4456 start = round_down(start, root->fs_info->sectorsize);
4458 btrfs_free_reserved_data_space_noquota(inode, start, len);
4459 btrfs_qgroup_free_data(inode, reserved, start, len);
4462 static void force_metadata_allocation(struct btrfs_fs_info *info)
4464 struct list_head *head = &info->space_info;
4465 struct btrfs_space_info *found;
4468 list_for_each_entry_rcu(found, head, list) {
4469 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4470 found->force_alloc = CHUNK_ALLOC_FORCE;
4475 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4477 return (global->size << 1);
4480 static int should_alloc_chunk(struct btrfs_fs_info *fs_info,
4481 struct btrfs_space_info *sinfo, int force)
4483 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4484 u64 bytes_used = btrfs_space_info_used(sinfo, false);
4487 if (force == CHUNK_ALLOC_FORCE)
4491 * We need to take into account the global rsv because for all intents
4492 * and purposes it's used space. Don't worry about locking the
4493 * global_rsv, it doesn't change except when the transaction commits.
4495 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4496 bytes_used += calc_global_rsv_need_space(global_rsv);
4499 * in limited mode, we want to have some free space up to
4500 * about 1% of the FS size.
4502 if (force == CHUNK_ALLOC_LIMITED) {
4503 thresh = btrfs_super_total_bytes(fs_info->super_copy);
4504 thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1));
4506 if (sinfo->total_bytes - bytes_used < thresh)
4510 if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8))
4515 static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type)
4519 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4520 BTRFS_BLOCK_GROUP_RAID0 |
4521 BTRFS_BLOCK_GROUP_RAID5 |
4522 BTRFS_BLOCK_GROUP_RAID6))
4523 num_dev = fs_info->fs_devices->rw_devices;
4524 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4527 num_dev = 1; /* DUP or single */
4533 * If @is_allocation is true, reserve space in the system space info necessary
4534 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4537 void check_system_chunk(struct btrfs_trans_handle *trans,
4538 struct btrfs_fs_info *fs_info, u64 type)
4540 struct btrfs_space_info *info;
4547 * Needed because we can end up allocating a system chunk and for an
4548 * atomic and race free space reservation in the chunk block reserve.
4550 lockdep_assert_held(&fs_info->chunk_mutex);
4552 info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4553 spin_lock(&info->lock);
4554 left = info->total_bytes - btrfs_space_info_used(info, true);
4555 spin_unlock(&info->lock);
4557 num_devs = get_profile_num_devs(fs_info, type);
4559 /* num_devs device items to update and 1 chunk item to add or remove */
4560 thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) +
4561 btrfs_calc_trans_metadata_size(fs_info, 1);
4563 if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4564 btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu",
4565 left, thresh, type);
4566 dump_space_info(fs_info, info, 0, 0);
4569 if (left < thresh) {
4570 u64 flags = btrfs_system_alloc_profile(fs_info);
4573 * Ignore failure to create system chunk. We might end up not
4574 * needing it, as we might not need to COW all nodes/leafs from
4575 * the paths we visit in the chunk tree (they were already COWed
4576 * or created in the current transaction for example).
4578 ret = btrfs_alloc_chunk(trans, fs_info, flags);
4582 ret = btrfs_block_rsv_add(fs_info->chunk_root,
4583 &fs_info->chunk_block_rsv,
4584 thresh, BTRFS_RESERVE_NO_FLUSH);
4586 trans->chunk_bytes_reserved += thresh;
4591 * If force is CHUNK_ALLOC_FORCE:
4592 * - return 1 if it successfully allocates a chunk,
4593 * - return errors including -ENOSPC otherwise.
4594 * If force is NOT CHUNK_ALLOC_FORCE:
4595 * - return 0 if it doesn't need to allocate a new chunk,
4596 * - return 1 if it successfully allocates a chunk,
4597 * - return errors including -ENOSPC otherwise.
4599 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4600 struct btrfs_fs_info *fs_info, u64 flags, int force)
4602 struct btrfs_space_info *space_info;
4603 int wait_for_alloc = 0;
4606 /* Don't re-enter if we're already allocating a chunk */
4607 if (trans->allocating_chunk)
4610 space_info = __find_space_info(fs_info, flags);
4614 spin_lock(&space_info->lock);
4615 if (force < space_info->force_alloc)
4616 force = space_info->force_alloc;
4617 if (space_info->full) {
4618 if (should_alloc_chunk(fs_info, space_info, force))
4622 spin_unlock(&space_info->lock);
4626 if (!should_alloc_chunk(fs_info, space_info, force)) {
4627 spin_unlock(&space_info->lock);
4629 } else if (space_info->chunk_alloc) {
4632 space_info->chunk_alloc = 1;
4635 spin_unlock(&space_info->lock);
4637 mutex_lock(&fs_info->chunk_mutex);
4640 * The chunk_mutex is held throughout the entirety of a chunk
4641 * allocation, so once we've acquired the chunk_mutex we know that the
4642 * other guy is done and we need to recheck and see if we should
4645 if (wait_for_alloc) {
4646 mutex_unlock(&fs_info->chunk_mutex);
4652 trans->allocating_chunk = true;
4655 * If we have mixed data/metadata chunks we want to make sure we keep
4656 * allocating mixed chunks instead of individual chunks.
4658 if (btrfs_mixed_space_info(space_info))
4659 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4662 * if we're doing a data chunk, go ahead and make sure that
4663 * we keep a reasonable number of metadata chunks allocated in the
4666 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4667 fs_info->data_chunk_allocations++;
4668 if (!(fs_info->data_chunk_allocations %
4669 fs_info->metadata_ratio))
4670 force_metadata_allocation(fs_info);
4674 * Check if we have enough space in SYSTEM chunk because we may need
4675 * to update devices.
4677 check_system_chunk(trans, fs_info, flags);
4679 ret = btrfs_alloc_chunk(trans, fs_info, flags);
4680 trans->allocating_chunk = false;
4682 spin_lock(&space_info->lock);
4685 space_info->full = 1;
4692 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4694 space_info->chunk_alloc = 0;
4695 spin_unlock(&space_info->lock);
4696 mutex_unlock(&fs_info->chunk_mutex);
4698 * When we allocate a new chunk we reserve space in the chunk block
4699 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4700 * add new nodes/leafs to it if we end up needing to do it when
4701 * inserting the chunk item and updating device items as part of the
4702 * second phase of chunk allocation, performed by
4703 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4704 * large number of new block groups to create in our transaction
4705 * handle's new_bgs list to avoid exhausting the chunk block reserve
4706 * in extreme cases - like having a single transaction create many new
4707 * block groups when starting to write out the free space caches of all
4708 * the block groups that were made dirty during the lifetime of the
4711 if (trans->can_flush_pending_bgs &&
4712 trans->chunk_bytes_reserved >= (u64)SZ_2M) {
4713 btrfs_create_pending_block_groups(trans);
4714 btrfs_trans_release_chunk_metadata(trans);
4719 static int can_overcommit(struct btrfs_fs_info *fs_info,
4720 struct btrfs_space_info *space_info, u64 bytes,
4721 enum btrfs_reserve_flush_enum flush,
4724 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4730 /* Don't overcommit when in mixed mode. */
4731 if (space_info->flags & BTRFS_BLOCK_GROUP_DATA)
4735 profile = btrfs_system_alloc_profile(fs_info);
4737 profile = btrfs_metadata_alloc_profile(fs_info);
4739 used = btrfs_space_info_used(space_info, false);
4742 * We only want to allow over committing if we have lots of actual space
4743 * free, but if we don't have enough space to handle the global reserve
4744 * space then we could end up having a real enospc problem when trying
4745 * to allocate a chunk or some other such important allocation.
4747 spin_lock(&global_rsv->lock);
4748 space_size = calc_global_rsv_need_space(global_rsv);
4749 spin_unlock(&global_rsv->lock);
4750 if (used + space_size >= space_info->total_bytes)
4753 used += space_info->bytes_may_use;
4755 avail = atomic64_read(&fs_info->free_chunk_space);
4758 * If we have dup, raid1 or raid10 then only half of the free
4759 * space is actually useable. For raid56, the space info used
4760 * doesn't include the parity drive, so we don't have to
4763 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4764 BTRFS_BLOCK_GROUP_RAID1 |
4765 BTRFS_BLOCK_GROUP_RAID10))
4769 * If we aren't flushing all things, let us overcommit up to
4770 * 1/2th of the space. If we can flush, don't let us overcommit
4771 * too much, let it overcommit up to 1/8 of the space.
4773 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4778 if (used + bytes < space_info->total_bytes + avail)
4783 static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info,
4784 unsigned long nr_pages, int nr_items)
4786 struct super_block *sb = fs_info->sb;
4788 if (down_read_trylock(&sb->s_umount)) {
4789 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4790 up_read(&sb->s_umount);
4793 * We needn't worry the filesystem going from r/w to r/o though
4794 * we don't acquire ->s_umount mutex, because the filesystem
4795 * should guarantee the delalloc inodes list be empty after
4796 * the filesystem is readonly(all dirty pages are written to
4799 btrfs_start_delalloc_roots(fs_info, 0, nr_items);
4800 if (!current->journal_info)
4801 btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1);
4805 static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info,
4811 bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
4812 nr = div64_u64(to_reclaim, bytes);
4818 #define EXTENT_SIZE_PER_ITEM SZ_256K
4821 * shrink metadata reservation for delalloc
4823 static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim,
4824 u64 orig, bool wait_ordered)
4826 struct btrfs_space_info *space_info;
4827 struct btrfs_trans_handle *trans;
4832 unsigned long nr_pages;
4835 /* Calc the number of the pages we need flush for space reservation */
4836 items = calc_reclaim_items_nr(fs_info, to_reclaim);
4837 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4839 trans = (struct btrfs_trans_handle *)current->journal_info;
4840 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4842 delalloc_bytes = percpu_counter_sum_positive(
4843 &fs_info->delalloc_bytes);
4844 if (delalloc_bytes == 0) {
4848 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4853 while (delalloc_bytes && loops < 3) {
4854 max_reclaim = min(delalloc_bytes, to_reclaim);
4855 nr_pages = max_reclaim >> PAGE_SHIFT;
4856 btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items);
4858 * We need to wait for the async pages to actually start before
4861 max_reclaim = atomic_read(&fs_info->async_delalloc_pages);
4865 if (max_reclaim <= nr_pages)
4868 max_reclaim -= nr_pages;
4870 wait_event(fs_info->async_submit_wait,
4871 atomic_read(&fs_info->async_delalloc_pages) <=
4874 spin_lock(&space_info->lock);
4875 if (list_empty(&space_info->tickets) &&
4876 list_empty(&space_info->priority_tickets)) {
4877 spin_unlock(&space_info->lock);
4880 spin_unlock(&space_info->lock);
4883 if (wait_ordered && !trans) {
4884 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4886 time_left = schedule_timeout_killable(1);
4890 delalloc_bytes = percpu_counter_sum_positive(
4891 &fs_info->delalloc_bytes);
4895 struct reserve_ticket {
4898 struct list_head list;
4899 wait_queue_head_t wait;
4903 * maybe_commit_transaction - possibly commit the transaction if its ok to
4904 * @root - the root we're allocating for
4905 * @bytes - the number of bytes we want to reserve
4906 * @force - force the commit
4908 * This will check to make sure that committing the transaction will actually
4909 * get us somewhere and then commit the transaction if it does. Otherwise it
4910 * will return -ENOSPC.
4912 static int may_commit_transaction(struct btrfs_fs_info *fs_info,
4913 struct btrfs_space_info *space_info)
4915 struct reserve_ticket *ticket = NULL;
4916 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv;
4917 struct btrfs_trans_handle *trans;
4920 trans = (struct btrfs_trans_handle *)current->journal_info;
4924 spin_lock(&space_info->lock);
4925 if (!list_empty(&space_info->priority_tickets))
4926 ticket = list_first_entry(&space_info->priority_tickets,
4927 struct reserve_ticket, list);
4928 else if (!list_empty(&space_info->tickets))
4929 ticket = list_first_entry(&space_info->tickets,
4930 struct reserve_ticket, list);
4931 bytes = (ticket) ? ticket->bytes : 0;
4932 spin_unlock(&space_info->lock);
4937 /* See if there is enough pinned space to make this reservation */
4938 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4943 * See if there is some space in the delayed insertion reservation for
4946 if (space_info != delayed_rsv->space_info)
4949 spin_lock(&delayed_rsv->lock);
4950 if (delayed_rsv->size > bytes)
4953 bytes -= delayed_rsv->size;
4954 spin_unlock(&delayed_rsv->lock);
4956 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4962 trans = btrfs_join_transaction(fs_info->extent_root);
4966 return btrfs_commit_transaction(trans);
4970 * Try to flush some data based on policy set by @state. This is only advisory
4971 * and may fail for various reasons. The caller is supposed to examine the
4972 * state of @space_info to detect the outcome.
4974 static void flush_space(struct btrfs_fs_info *fs_info,
4975 struct btrfs_space_info *space_info, u64 num_bytes,
4978 struct btrfs_root *root = fs_info->extent_root;
4979 struct btrfs_trans_handle *trans;
4984 case FLUSH_DELAYED_ITEMS_NR:
4985 case FLUSH_DELAYED_ITEMS:
4986 if (state == FLUSH_DELAYED_ITEMS_NR)
4987 nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2;
4991 trans = btrfs_join_transaction(root);
4992 if (IS_ERR(trans)) {
4993 ret = PTR_ERR(trans);
4996 ret = btrfs_run_delayed_items_nr(trans, nr);
4997 btrfs_end_transaction(trans);
4999 case FLUSH_DELALLOC:
5000 case FLUSH_DELALLOC_WAIT:
5001 shrink_delalloc(fs_info, num_bytes * 2, num_bytes,
5002 state == FLUSH_DELALLOC_WAIT);
5005 trans = btrfs_join_transaction(root);
5006 if (IS_ERR(trans)) {
5007 ret = PTR_ERR(trans);
5010 ret = do_chunk_alloc(trans, fs_info,
5011 btrfs_metadata_alloc_profile(fs_info),
5012 CHUNK_ALLOC_NO_FORCE);
5013 btrfs_end_transaction(trans);
5014 if (ret > 0 || ret == -ENOSPC)
5018 ret = may_commit_transaction(fs_info, space_info);
5025 trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, state,
5031 btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info,
5032 struct btrfs_space_info *space_info,
5035 struct reserve_ticket *ticket;
5040 list_for_each_entry(ticket, &space_info->tickets, list)
5041 to_reclaim += ticket->bytes;
5042 list_for_each_entry(ticket, &space_info->priority_tickets, list)
5043 to_reclaim += ticket->bytes;
5047 to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M);
5048 if (can_overcommit(fs_info, space_info, to_reclaim,
5049 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
5052 used = btrfs_space_info_used(space_info, true);
5054 if (can_overcommit(fs_info, space_info, SZ_1M,
5055 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
5056 expected = div_factor_fine(space_info->total_bytes, 95);
5058 expected = div_factor_fine(space_info->total_bytes, 90);
5060 if (used > expected)
5061 to_reclaim = used - expected;
5064 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
5065 space_info->bytes_reserved);
5069 static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info,
5070 struct btrfs_space_info *space_info,
5071 u64 used, bool system_chunk)
5073 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
5075 /* If we're just plain full then async reclaim just slows us down. */
5076 if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh)
5079 if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5083 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
5084 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
5087 static void wake_all_tickets(struct list_head *head)
5089 struct reserve_ticket *ticket;
5091 while (!list_empty(head)) {
5092 ticket = list_first_entry(head, struct reserve_ticket, list);
5093 list_del_init(&ticket->list);
5094 ticket->error = -ENOSPC;
5095 wake_up(&ticket->wait);
5100 * This is for normal flushers, we can wait all goddamned day if we want to. We
5101 * will loop and continuously try to flush as long as we are making progress.
5102 * We count progress as clearing off tickets each time we have to loop.
5104 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
5106 struct btrfs_fs_info *fs_info;
5107 struct btrfs_space_info *space_info;
5110 int commit_cycles = 0;
5111 u64 last_tickets_id;
5113 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
5114 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5116 spin_lock(&space_info->lock);
5117 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5120 space_info->flush = 0;
5121 spin_unlock(&space_info->lock);
5124 last_tickets_id = space_info->tickets_id;
5125 spin_unlock(&space_info->lock);
5127 flush_state = FLUSH_DELAYED_ITEMS_NR;
5129 flush_space(fs_info, space_info, to_reclaim, flush_state);
5130 spin_lock(&space_info->lock);
5131 if (list_empty(&space_info->tickets)) {
5132 space_info->flush = 0;
5133 spin_unlock(&space_info->lock);
5136 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info,
5139 if (last_tickets_id == space_info->tickets_id) {
5142 last_tickets_id = space_info->tickets_id;
5143 flush_state = FLUSH_DELAYED_ITEMS_NR;
5148 if (flush_state > COMMIT_TRANS) {
5150 if (commit_cycles > 2) {
5151 wake_all_tickets(&space_info->tickets);
5152 space_info->flush = 0;
5154 flush_state = FLUSH_DELAYED_ITEMS_NR;
5157 spin_unlock(&space_info->lock);
5158 } while (flush_state <= COMMIT_TRANS);
5161 void btrfs_init_async_reclaim_work(struct work_struct *work)
5163 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
5166 static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
5167 struct btrfs_space_info *space_info,
5168 struct reserve_ticket *ticket)
5171 int flush_state = FLUSH_DELAYED_ITEMS_NR;
5173 spin_lock(&space_info->lock);
5174 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5177 spin_unlock(&space_info->lock);
5180 spin_unlock(&space_info->lock);
5183 flush_space(fs_info, space_info, to_reclaim, flush_state);
5185 spin_lock(&space_info->lock);
5186 if (ticket->bytes == 0) {
5187 spin_unlock(&space_info->lock);
5190 spin_unlock(&space_info->lock);
5193 * Priority flushers can't wait on delalloc without
5196 if (flush_state == FLUSH_DELALLOC ||
5197 flush_state == FLUSH_DELALLOC_WAIT)
5198 flush_state = ALLOC_CHUNK;
5199 } while (flush_state < COMMIT_TRANS);
5202 static int wait_reserve_ticket(struct btrfs_fs_info *fs_info,
5203 struct btrfs_space_info *space_info,
5204 struct reserve_ticket *ticket, u64 orig_bytes)
5210 spin_lock(&space_info->lock);
5211 while (ticket->bytes > 0 && ticket->error == 0) {
5212 ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE);
5217 spin_unlock(&space_info->lock);
5221 finish_wait(&ticket->wait, &wait);
5222 spin_lock(&space_info->lock);
5225 ret = ticket->error;
5226 if (!list_empty(&ticket->list))
5227 list_del_init(&ticket->list);
5228 if (ticket->bytes && ticket->bytes < orig_bytes) {
5229 u64 num_bytes = orig_bytes - ticket->bytes;
5230 space_info->bytes_may_use -= num_bytes;
5231 trace_btrfs_space_reservation(fs_info, "space_info",
5232 space_info->flags, num_bytes, 0);
5234 spin_unlock(&space_info->lock);
5240 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5241 * @root - the root we're allocating for
5242 * @space_info - the space info we want to allocate from
5243 * @orig_bytes - the number of bytes we want
5244 * @flush - whether or not we can flush to make our reservation
5246 * This will reserve orig_bytes number of bytes from the space info associated
5247 * with the block_rsv. If there is not enough space it will make an attempt to
5248 * flush out space to make room. It will do this by flushing delalloc if
5249 * possible or committing the transaction. If flush is 0 then no attempts to
5250 * regain reservations will be made and this will fail if there is not enough
5253 static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info,
5254 struct btrfs_space_info *space_info,
5256 enum btrfs_reserve_flush_enum flush,
5259 struct reserve_ticket ticket;
5264 ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL);
5266 spin_lock(&space_info->lock);
5268 used = btrfs_space_info_used(space_info, true);
5271 * If we have enough space then hooray, make our reservation and carry
5272 * on. If not see if we can overcommit, and if we can, hooray carry on.
5273 * If not things get more complicated.
5275 if (used + orig_bytes <= space_info->total_bytes) {
5276 space_info->bytes_may_use += orig_bytes;
5277 trace_btrfs_space_reservation(fs_info, "space_info",
5278 space_info->flags, orig_bytes, 1);
5280 } else if (can_overcommit(fs_info, space_info, orig_bytes, flush,
5282 space_info->bytes_may_use += orig_bytes;
5283 trace_btrfs_space_reservation(fs_info, "space_info",
5284 space_info->flags, orig_bytes, 1);
5289 * If we couldn't make a reservation then setup our reservation ticket
5290 * and kick the async worker if it's not already running.
5292 * If we are a priority flusher then we just need to add our ticket to
5293 * the list and we will do our own flushing further down.
5295 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
5296 ticket.bytes = orig_bytes;
5298 init_waitqueue_head(&ticket.wait);
5299 if (flush == BTRFS_RESERVE_FLUSH_ALL) {
5300 list_add_tail(&ticket.list, &space_info->tickets);
5301 if (!space_info->flush) {
5302 space_info->flush = 1;
5303 trace_btrfs_trigger_flush(fs_info,
5307 queue_work(system_unbound_wq,
5308 &fs_info->async_reclaim_work);
5311 list_add_tail(&ticket.list,
5312 &space_info->priority_tickets);
5314 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
5317 * We will do the space reservation dance during log replay,
5318 * which means we won't have fs_info->fs_root set, so don't do
5319 * the async reclaim as we will panic.
5321 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) &&
5322 need_do_async_reclaim(fs_info, space_info,
5323 used, system_chunk) &&
5324 !work_busy(&fs_info->async_reclaim_work)) {
5325 trace_btrfs_trigger_flush(fs_info, space_info->flags,
5326 orig_bytes, flush, "preempt");
5327 queue_work(system_unbound_wq,
5328 &fs_info->async_reclaim_work);
5331 spin_unlock(&space_info->lock);
5332 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
5335 if (flush == BTRFS_RESERVE_FLUSH_ALL)
5336 return wait_reserve_ticket(fs_info, space_info, &ticket,
5340 priority_reclaim_metadata_space(fs_info, space_info, &ticket);
5341 spin_lock(&space_info->lock);
5343 if (ticket.bytes < orig_bytes) {
5344 u64 num_bytes = orig_bytes - ticket.bytes;
5345 space_info->bytes_may_use -= num_bytes;
5346 trace_btrfs_space_reservation(fs_info, "space_info",
5351 list_del_init(&ticket.list);
5354 spin_unlock(&space_info->lock);
5355 ASSERT(list_empty(&ticket.list));
5360 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5361 * @root - the root we're allocating for
5362 * @block_rsv - the block_rsv we're allocating for
5363 * @orig_bytes - the number of bytes we want
5364 * @flush - whether or not we can flush to make our reservation
5366 * This will reserve orgi_bytes number of bytes from the space info associated
5367 * with the block_rsv. If there is not enough space it will make an attempt to
5368 * flush out space to make room. It will do this by flushing delalloc if
5369 * possible or committing the transaction. If flush is 0 then no attempts to
5370 * regain reservations will be made and this will fail if there is not enough
5373 static int reserve_metadata_bytes(struct btrfs_root *root,
5374 struct btrfs_block_rsv *block_rsv,
5376 enum btrfs_reserve_flush_enum flush)
5378 struct btrfs_fs_info *fs_info = root->fs_info;
5379 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5381 bool system_chunk = (root == fs_info->chunk_root);
5383 ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info,
5384 orig_bytes, flush, system_chunk);
5385 if (ret == -ENOSPC &&
5386 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5387 if (block_rsv != global_rsv &&
5388 !block_rsv_use_bytes(global_rsv, orig_bytes))
5391 if (ret == -ENOSPC) {
5392 trace_btrfs_space_reservation(fs_info, "space_info:enospc",
5393 block_rsv->space_info->flags,
5396 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG))
5397 dump_space_info(fs_info, block_rsv->space_info,
5403 static struct btrfs_block_rsv *get_block_rsv(
5404 const struct btrfs_trans_handle *trans,
5405 const struct btrfs_root *root)
5407 struct btrfs_fs_info *fs_info = root->fs_info;
5408 struct btrfs_block_rsv *block_rsv = NULL;
5410 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5411 (root == fs_info->csum_root && trans->adding_csums) ||
5412 (root == fs_info->uuid_root))
5413 block_rsv = trans->block_rsv;
5416 block_rsv = root->block_rsv;
5419 block_rsv = &fs_info->empty_block_rsv;
5424 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5428 spin_lock(&block_rsv->lock);
5429 if (block_rsv->reserved >= num_bytes) {
5430 block_rsv->reserved -= num_bytes;
5431 if (block_rsv->reserved < block_rsv->size)
5432 block_rsv->full = 0;
5435 spin_unlock(&block_rsv->lock);
5439 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5440 u64 num_bytes, int update_size)
5442 spin_lock(&block_rsv->lock);
5443 block_rsv->reserved += num_bytes;
5445 block_rsv->size += num_bytes;
5446 else if (block_rsv->reserved >= block_rsv->size)
5447 block_rsv->full = 1;
5448 spin_unlock(&block_rsv->lock);
5451 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5452 struct btrfs_block_rsv *dest, u64 num_bytes,
5455 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5458 if (global_rsv->space_info != dest->space_info)
5461 spin_lock(&global_rsv->lock);
5462 min_bytes = div_factor(global_rsv->size, min_factor);
5463 if (global_rsv->reserved < min_bytes + num_bytes) {
5464 spin_unlock(&global_rsv->lock);
5467 global_rsv->reserved -= num_bytes;
5468 if (global_rsv->reserved < global_rsv->size)
5469 global_rsv->full = 0;
5470 spin_unlock(&global_rsv->lock);
5472 block_rsv_add_bytes(dest, num_bytes, 1);
5477 * This is for space we already have accounted in space_info->bytes_may_use, so
5478 * basically when we're returning space from block_rsv's.
5480 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
5481 struct btrfs_space_info *space_info,
5484 struct reserve_ticket *ticket;
5485 struct list_head *head;
5487 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH;
5488 bool check_overcommit = false;
5490 spin_lock(&space_info->lock);
5491 head = &space_info->priority_tickets;
5494 * If we are over our limit then we need to check and see if we can
5495 * overcommit, and if we can't then we just need to free up our space
5496 * and not satisfy any requests.
5498 used = btrfs_space_info_used(space_info, true);
5499 if (used - num_bytes >= space_info->total_bytes)
5500 check_overcommit = true;
5502 while (!list_empty(head) && num_bytes) {
5503 ticket = list_first_entry(head, struct reserve_ticket,
5506 * We use 0 bytes because this space is already reserved, so
5507 * adding the ticket space would be a double count.
5509 if (check_overcommit &&
5510 !can_overcommit(fs_info, space_info, 0, flush, false))
5512 if (num_bytes >= ticket->bytes) {
5513 list_del_init(&ticket->list);
5514 num_bytes -= ticket->bytes;
5516 space_info->tickets_id++;
5517 wake_up(&ticket->wait);
5519 ticket->bytes -= num_bytes;
5524 if (num_bytes && head == &space_info->priority_tickets) {
5525 head = &space_info->tickets;
5526 flush = BTRFS_RESERVE_FLUSH_ALL;
5529 space_info->bytes_may_use -= num_bytes;
5530 trace_btrfs_space_reservation(fs_info, "space_info",
5531 space_info->flags, num_bytes, 0);
5532 spin_unlock(&space_info->lock);
5536 * This is for newly allocated space that isn't accounted in
5537 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5538 * we use this helper.
5540 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
5541 struct btrfs_space_info *space_info,
5544 struct reserve_ticket *ticket;
5545 struct list_head *head = &space_info->priority_tickets;
5548 while (!list_empty(head) && num_bytes) {
5549 ticket = list_first_entry(head, struct reserve_ticket,
5551 if (num_bytes >= ticket->bytes) {
5552 trace_btrfs_space_reservation(fs_info, "space_info",
5555 list_del_init(&ticket->list);
5556 num_bytes -= ticket->bytes;
5557 space_info->bytes_may_use += ticket->bytes;
5559 space_info->tickets_id++;
5560 wake_up(&ticket->wait);
5562 trace_btrfs_space_reservation(fs_info, "space_info",
5565 space_info->bytes_may_use += num_bytes;
5566 ticket->bytes -= num_bytes;
5571 if (num_bytes && head == &space_info->priority_tickets) {
5572 head = &space_info->tickets;
5577 static u64 block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5578 struct btrfs_block_rsv *block_rsv,
5579 struct btrfs_block_rsv *dest, u64 num_bytes,
5580 u64 *qgroup_to_release_ret)
5582 struct btrfs_space_info *space_info = block_rsv->space_info;
5583 u64 qgroup_to_release = 0;
5586 spin_lock(&block_rsv->lock);
5587 if (num_bytes == (u64)-1) {
5588 num_bytes = block_rsv->size;
5589 qgroup_to_release = block_rsv->qgroup_rsv_size;
5591 block_rsv->size -= num_bytes;
5592 if (block_rsv->reserved >= block_rsv->size) {
5593 num_bytes = block_rsv->reserved - block_rsv->size;
5594 block_rsv->reserved = block_rsv->size;
5595 block_rsv->full = 1;
5599 if (block_rsv->qgroup_rsv_reserved >= block_rsv->qgroup_rsv_size) {
5600 qgroup_to_release = block_rsv->qgroup_rsv_reserved -
5601 block_rsv->qgroup_rsv_size;
5602 block_rsv->qgroup_rsv_reserved = block_rsv->qgroup_rsv_size;
5604 qgroup_to_release = 0;
5606 spin_unlock(&block_rsv->lock);
5609 if (num_bytes > 0) {
5611 spin_lock(&dest->lock);
5615 bytes_to_add = dest->size - dest->reserved;
5616 bytes_to_add = min(num_bytes, bytes_to_add);
5617 dest->reserved += bytes_to_add;
5618 if (dest->reserved >= dest->size)
5620 num_bytes -= bytes_to_add;
5622 spin_unlock(&dest->lock);
5625 space_info_add_old_bytes(fs_info, space_info,
5628 if (qgroup_to_release_ret)
5629 *qgroup_to_release_ret = qgroup_to_release;
5633 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
5634 struct btrfs_block_rsv *dst, u64 num_bytes,
5639 ret = block_rsv_use_bytes(src, num_bytes);
5643 block_rsv_add_bytes(dst, num_bytes, update_size);
5647 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5649 memset(rsv, 0, sizeof(*rsv));
5650 spin_lock_init(&rsv->lock);
5654 void btrfs_init_metadata_block_rsv(struct btrfs_fs_info *fs_info,
5655 struct btrfs_block_rsv *rsv,
5656 unsigned short type)
5658 btrfs_init_block_rsv(rsv, type);
5659 rsv->space_info = __find_space_info(fs_info,
5660 BTRFS_BLOCK_GROUP_METADATA);
5663 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info,
5664 unsigned short type)
5666 struct btrfs_block_rsv *block_rsv;
5668 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5672 btrfs_init_metadata_block_rsv(fs_info, block_rsv, type);
5676 void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info,
5677 struct btrfs_block_rsv *rsv)
5681 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
5685 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
5690 int btrfs_block_rsv_add(struct btrfs_root *root,
5691 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5692 enum btrfs_reserve_flush_enum flush)
5699 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5701 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5708 int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor)
5716 spin_lock(&block_rsv->lock);
5717 num_bytes = div_factor(block_rsv->size, min_factor);
5718 if (block_rsv->reserved >= num_bytes)
5720 spin_unlock(&block_rsv->lock);
5725 int btrfs_block_rsv_refill(struct btrfs_root *root,
5726 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5727 enum btrfs_reserve_flush_enum flush)
5735 spin_lock(&block_rsv->lock);
5736 num_bytes = min_reserved;
5737 if (block_rsv->reserved >= num_bytes)
5740 num_bytes -= block_rsv->reserved;
5741 spin_unlock(&block_rsv->lock);
5746 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5748 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5756 * btrfs_inode_rsv_refill - refill the inode block rsv.
5757 * @inode - the inode we are refilling.
5758 * @flush - the flusing restriction.
5760 * Essentially the same as btrfs_block_rsv_refill, except it uses the
5761 * block_rsv->size as the minimum size. We'll either refill the missing amount
5762 * or return if we already have enough space. This will also handle the resreve
5763 * tracepoint for the reserved amount.
5765 static int btrfs_inode_rsv_refill(struct btrfs_inode *inode,
5766 enum btrfs_reserve_flush_enum flush)
5768 struct btrfs_root *root = inode->root;
5769 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5771 u64 qgroup_num_bytes = 0;
5774 spin_lock(&block_rsv->lock);
5775 if (block_rsv->reserved < block_rsv->size)
5776 num_bytes = block_rsv->size - block_rsv->reserved;
5777 if (block_rsv->qgroup_rsv_reserved < block_rsv->qgroup_rsv_size)
5778 qgroup_num_bytes = block_rsv->qgroup_rsv_size -
5779 block_rsv->qgroup_rsv_reserved;
5780 spin_unlock(&block_rsv->lock);
5785 ret = btrfs_qgroup_reserve_meta_prealloc(root, qgroup_num_bytes, true);
5788 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5790 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5791 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5792 btrfs_ino(inode), num_bytes, 1);
5794 /* Don't forget to increase qgroup_rsv_reserved */
5795 spin_lock(&block_rsv->lock);
5796 block_rsv->qgroup_rsv_reserved += qgroup_num_bytes;
5797 spin_unlock(&block_rsv->lock);
5799 btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes);
5804 * btrfs_inode_rsv_release - release any excessive reservation.
5805 * @inode - the inode we need to release from.
5806 * @qgroup_free - free or convert qgroup meta.
5807 * Unlike normal operation, qgroup meta reservation needs to know if we are
5808 * freeing qgroup reservation or just converting it into per-trans. Normally
5809 * @qgroup_free is true for error handling, and false for normal release.
5811 * This is the same as btrfs_block_rsv_release, except that it handles the
5812 * tracepoint for the reservation.
5814 static void btrfs_inode_rsv_release(struct btrfs_inode *inode, bool qgroup_free)
5816 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5817 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5818 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5820 u64 qgroup_to_release = 0;
5823 * Since we statically set the block_rsv->size we just want to say we
5824 * are releasing 0 bytes, and then we'll just get the reservation over
5827 released = block_rsv_release_bytes(fs_info, block_rsv, global_rsv, 0,
5828 &qgroup_to_release);
5830 trace_btrfs_space_reservation(fs_info, "delalloc",
5831 btrfs_ino(inode), released, 0);
5833 btrfs_qgroup_free_meta_prealloc(inode->root, qgroup_to_release);
5835 btrfs_qgroup_convert_reserved_meta(inode->root,
5839 void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
5840 struct btrfs_block_rsv *block_rsv,
5843 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5845 if (global_rsv == block_rsv ||
5846 block_rsv->space_info != global_rsv->space_info)
5848 block_rsv_release_bytes(fs_info, block_rsv, global_rsv, num_bytes, NULL);
5851 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5853 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5854 struct btrfs_space_info *sinfo = block_rsv->space_info;
5858 * The global block rsv is based on the size of the extent tree, the
5859 * checksum tree and the root tree. If the fs is empty we want to set
5860 * it to a minimal amount for safety.
5862 num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) +
5863 btrfs_root_used(&fs_info->csum_root->root_item) +
5864 btrfs_root_used(&fs_info->tree_root->root_item);
5865 num_bytes = max_t(u64, num_bytes, SZ_16M);
5867 spin_lock(&sinfo->lock);
5868 spin_lock(&block_rsv->lock);
5870 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
5872 if (block_rsv->reserved < block_rsv->size) {
5873 num_bytes = btrfs_space_info_used(sinfo, true);
5874 if (sinfo->total_bytes > num_bytes) {
5875 num_bytes = sinfo->total_bytes - num_bytes;
5876 num_bytes = min(num_bytes,
5877 block_rsv->size - block_rsv->reserved);
5878 block_rsv->reserved += num_bytes;
5879 sinfo->bytes_may_use += num_bytes;
5880 trace_btrfs_space_reservation(fs_info, "space_info",
5881 sinfo->flags, num_bytes,
5884 } else if (block_rsv->reserved > block_rsv->size) {
5885 num_bytes = block_rsv->reserved - block_rsv->size;
5886 sinfo->bytes_may_use -= num_bytes;
5887 trace_btrfs_space_reservation(fs_info, "space_info",
5888 sinfo->flags, num_bytes, 0);
5889 block_rsv->reserved = block_rsv->size;
5892 if (block_rsv->reserved == block_rsv->size)
5893 block_rsv->full = 1;
5895 block_rsv->full = 0;
5897 spin_unlock(&block_rsv->lock);
5898 spin_unlock(&sinfo->lock);
5901 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5903 struct btrfs_space_info *space_info;
5905 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5906 fs_info->chunk_block_rsv.space_info = space_info;
5908 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5909 fs_info->global_block_rsv.space_info = space_info;
5910 fs_info->trans_block_rsv.space_info = space_info;
5911 fs_info->empty_block_rsv.space_info = space_info;
5912 fs_info->delayed_block_rsv.space_info = space_info;
5914 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5915 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5916 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5917 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5918 if (fs_info->quota_root)
5919 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5920 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5922 update_global_block_rsv(fs_info);
5925 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5927 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5929 WARN_ON(fs_info->trans_block_rsv.size > 0);
5930 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5931 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5932 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5933 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5934 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5939 * To be called after all the new block groups attached to the transaction
5940 * handle have been created (btrfs_create_pending_block_groups()).
5942 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5944 struct btrfs_fs_info *fs_info = trans->fs_info;
5946 if (!trans->chunk_bytes_reserved)
5949 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5951 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5952 trans->chunk_bytes_reserved, NULL);
5953 trans->chunk_bytes_reserved = 0;
5956 /* Can only return 0 or -ENOSPC */
5957 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5958 struct btrfs_inode *inode)
5960 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5961 struct btrfs_root *root = inode->root;
5963 * We always use trans->block_rsv here as we will have reserved space
5964 * for our orphan when starting the transaction, using get_block_rsv()
5965 * here will sometimes make us choose the wrong block rsv as we could be
5966 * doing a reloc inode for a non refcounted root.
5968 struct btrfs_block_rsv *src_rsv = trans->block_rsv;
5969 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5972 * We need to hold space in order to delete our orphan item once we've
5973 * added it, so this takes the reservation so we can release it later
5974 * when we are truly done with the orphan item.
5976 u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
5978 trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode),
5980 return btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1);
5983 void btrfs_orphan_release_metadata(struct btrfs_inode *inode)
5985 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5986 struct btrfs_root *root = inode->root;
5987 u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
5989 trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode),
5991 btrfs_block_rsv_release(fs_info, root->orphan_block_rsv, num_bytes);
5995 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5996 * root: the root of the parent directory
5997 * rsv: block reservation
5998 * items: the number of items that we need do reservation
5999 * qgroup_reserved: used to return the reserved size in qgroup
6001 * This function is used to reserve the space for snapshot/subvolume
6002 * creation and deletion. Those operations are different with the
6003 * common file/directory operations, they change two fs/file trees
6004 * and root tree, the number of items that the qgroup reserves is
6005 * different with the free space reservation. So we can not use
6006 * the space reservation mechanism in start_transaction().
6008 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
6009 struct btrfs_block_rsv *rsv,
6011 u64 *qgroup_reserved,
6012 bool use_global_rsv)
6016 struct btrfs_fs_info *fs_info = root->fs_info;
6017 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6019 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
6020 /* One for parent inode, two for dir entries */
6021 num_bytes = 3 * fs_info->nodesize;
6022 ret = btrfs_qgroup_reserve_meta_prealloc(root, num_bytes, true);
6029 *qgroup_reserved = num_bytes;
6031 num_bytes = btrfs_calc_trans_metadata_size(fs_info, items);
6032 rsv->space_info = __find_space_info(fs_info,
6033 BTRFS_BLOCK_GROUP_METADATA);
6034 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
6035 BTRFS_RESERVE_FLUSH_ALL);
6037 if (ret == -ENOSPC && use_global_rsv)
6038 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, 1);
6040 if (ret && *qgroup_reserved)
6041 btrfs_qgroup_free_meta_prealloc(root, *qgroup_reserved);
6046 void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info,
6047 struct btrfs_block_rsv *rsv)
6049 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
6052 static void btrfs_calculate_inode_block_rsv_size(struct btrfs_fs_info *fs_info,
6053 struct btrfs_inode *inode)
6055 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
6056 u64 reserve_size = 0;
6057 u64 qgroup_rsv_size = 0;
6059 unsigned outstanding_extents;
6061 lockdep_assert_held(&inode->lock);
6062 outstanding_extents = inode->outstanding_extents;
6063 if (outstanding_extents)
6064 reserve_size = btrfs_calc_trans_metadata_size(fs_info,
6065 outstanding_extents + 1);
6066 csum_leaves = btrfs_csum_bytes_to_leaves(fs_info,
6068 reserve_size += btrfs_calc_trans_metadata_size(fs_info,
6071 * For qgroup rsv, the calculation is very simple:
6072 * account one nodesize for each outstanding extent
6074 * This is overestimating in most cases.
6076 qgroup_rsv_size = outstanding_extents * fs_info->nodesize;
6078 spin_lock(&block_rsv->lock);
6079 block_rsv->size = reserve_size;
6080 block_rsv->qgroup_rsv_size = qgroup_rsv_size;
6081 spin_unlock(&block_rsv->lock);
6084 int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes)
6086 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6087 unsigned nr_extents;
6088 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
6090 bool delalloc_lock = true;
6092 /* If we are a free space inode we need to not flush since we will be in
6093 * the middle of a transaction commit. We also don't need the delalloc
6094 * mutex since we won't race with anybody. We need this mostly to make
6095 * lockdep shut its filthy mouth.
6097 * If we have a transaction open (can happen if we call truncate_block
6098 * from truncate), then we need FLUSH_LIMIT so we don't deadlock.
6100 if (btrfs_is_free_space_inode(inode)) {
6101 flush = BTRFS_RESERVE_NO_FLUSH;
6102 delalloc_lock = false;
6104 if (current->journal_info)
6105 flush = BTRFS_RESERVE_FLUSH_LIMIT;
6107 if (btrfs_transaction_in_commit(fs_info))
6108 schedule_timeout(1);
6112 mutex_lock(&inode->delalloc_mutex);
6114 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6116 /* Add our new extents and calculate the new rsv size. */
6117 spin_lock(&inode->lock);
6118 nr_extents = count_max_extents(num_bytes);
6119 btrfs_mod_outstanding_extents(inode, nr_extents);
6120 inode->csum_bytes += num_bytes;
6121 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6122 spin_unlock(&inode->lock);
6124 ret = btrfs_inode_rsv_refill(inode, flush);
6129 mutex_unlock(&inode->delalloc_mutex);
6133 spin_lock(&inode->lock);
6134 nr_extents = count_max_extents(num_bytes);
6135 btrfs_mod_outstanding_extents(inode, -nr_extents);
6136 inode->csum_bytes -= num_bytes;
6137 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6138 spin_unlock(&inode->lock);
6140 btrfs_inode_rsv_release(inode, true);
6142 mutex_unlock(&inode->delalloc_mutex);
6147 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
6148 * @inode: the inode to release the reservation for.
6149 * @num_bytes: the number of bytes we are releasing.
6150 * @qgroup_free: free qgroup reservation or convert it to per-trans reservation
6152 * This will release the metadata reservation for an inode. This can be called
6153 * once we complete IO for a given set of bytes to release their metadata
6154 * reservations, or on error for the same reason.
6156 void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes,
6159 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6161 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6162 spin_lock(&inode->lock);
6163 inode->csum_bytes -= num_bytes;
6164 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6165 spin_unlock(&inode->lock);
6167 if (btrfs_is_testing(fs_info))
6170 btrfs_inode_rsv_release(inode, qgroup_free);
6174 * btrfs_delalloc_release_extents - release our outstanding_extents
6175 * @inode: the inode to balance the reservation for.
6176 * @num_bytes: the number of bytes we originally reserved with
6177 * @qgroup_free: do we need to free qgroup meta reservation or convert them.
6179 * When we reserve space we increase outstanding_extents for the extents we may
6180 * add. Once we've set the range as delalloc or created our ordered extents we
6181 * have outstanding_extents to track the real usage, so we use this to free our
6182 * temporarily tracked outstanding_extents. This _must_ be used in conjunction
6183 * with btrfs_delalloc_reserve_metadata.
6185 void btrfs_delalloc_release_extents(struct btrfs_inode *inode, u64 num_bytes,
6188 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6189 unsigned num_extents;
6191 spin_lock(&inode->lock);
6192 num_extents = count_max_extents(num_bytes);
6193 btrfs_mod_outstanding_extents(inode, -num_extents);
6194 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6195 spin_unlock(&inode->lock);
6197 if (btrfs_is_testing(fs_info))
6200 btrfs_inode_rsv_release(inode, qgroup_free);
6204 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6206 * @inode: inode we're writing to
6207 * @start: start range we are writing to
6208 * @len: how long the range we are writing to
6209 * @reserved: mandatory parameter, record actually reserved qgroup ranges of
6210 * current reservation.
6212 * This will do the following things
6214 * o reserve space in data space info for num bytes
6215 * and reserve precious corresponding qgroup space
6216 * (Done in check_data_free_space)
6218 * o reserve space for metadata space, based on the number of outstanding
6219 * extents and how much csums will be needed
6220 * also reserve metadata space in a per root over-reserve method.
6221 * o add to the inodes->delalloc_bytes
6222 * o add it to the fs_info's delalloc inodes list.
6223 * (Above 3 all done in delalloc_reserve_metadata)
6225 * Return 0 for success
6226 * Return <0 for error(-ENOSPC or -EQUOT)
6228 int btrfs_delalloc_reserve_space(struct inode *inode,
6229 struct extent_changeset **reserved, u64 start, u64 len)
6233 ret = btrfs_check_data_free_space(inode, reserved, start, len);
6236 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len);
6238 btrfs_free_reserved_data_space(inode, *reserved, start, len);
6243 * btrfs_delalloc_release_space - release data and metadata space for delalloc
6244 * @inode: inode we're releasing space for
6245 * @start: start position of the space already reserved
6246 * @len: the len of the space already reserved
6247 * @release_bytes: the len of the space we consumed or didn't use
6249 * This function will release the metadata space that was not used and will
6250 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
6251 * list if there are no delalloc bytes left.
6252 * Also it will handle the qgroup reserved space.
6254 void btrfs_delalloc_release_space(struct inode *inode,
6255 struct extent_changeset *reserved,
6256 u64 start, u64 len, bool qgroup_free)
6258 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, qgroup_free);
6259 btrfs_free_reserved_data_space(inode, reserved, start, len);
6262 static int update_block_group(struct btrfs_trans_handle *trans,
6263 struct btrfs_fs_info *info, u64 bytenr,
6264 u64 num_bytes, int alloc)
6266 struct btrfs_block_group_cache *cache = NULL;
6267 u64 total = num_bytes;
6272 /* block accounting for super block */
6273 spin_lock(&info->delalloc_root_lock);
6274 old_val = btrfs_super_bytes_used(info->super_copy);
6276 old_val += num_bytes;
6278 old_val -= num_bytes;
6279 btrfs_set_super_bytes_used(info->super_copy, old_val);
6280 spin_unlock(&info->delalloc_root_lock);
6283 cache = btrfs_lookup_block_group(info, bytenr);
6286 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
6287 BTRFS_BLOCK_GROUP_RAID1 |
6288 BTRFS_BLOCK_GROUP_RAID10))
6293 * If this block group has free space cache written out, we
6294 * need to make sure to load it if we are removing space. This
6295 * is because we need the unpinning stage to actually add the
6296 * space back to the block group, otherwise we will leak space.
6298 if (!alloc && cache->cached == BTRFS_CACHE_NO)
6299 cache_block_group(cache, 1);
6301 byte_in_group = bytenr - cache->key.objectid;
6302 WARN_ON(byte_in_group > cache->key.offset);
6304 spin_lock(&cache->space_info->lock);
6305 spin_lock(&cache->lock);
6307 if (btrfs_test_opt(info, SPACE_CACHE) &&
6308 cache->disk_cache_state < BTRFS_DC_CLEAR)
6309 cache->disk_cache_state = BTRFS_DC_CLEAR;
6311 old_val = btrfs_block_group_used(&cache->item);
6312 num_bytes = min(total, cache->key.offset - byte_in_group);
6314 old_val += num_bytes;
6315 btrfs_set_block_group_used(&cache->item, old_val);
6316 cache->reserved -= num_bytes;
6317 cache->space_info->bytes_reserved -= num_bytes;
6318 cache->space_info->bytes_used += num_bytes;
6319 cache->space_info->disk_used += num_bytes * factor;
6320 spin_unlock(&cache->lock);
6321 spin_unlock(&cache->space_info->lock);
6323 old_val -= num_bytes;
6324 btrfs_set_block_group_used(&cache->item, old_val);
6325 cache->pinned += num_bytes;
6326 cache->space_info->bytes_pinned += num_bytes;
6327 cache->space_info->bytes_used -= num_bytes;
6328 cache->space_info->disk_used -= num_bytes * factor;
6329 spin_unlock(&cache->lock);
6330 spin_unlock(&cache->space_info->lock);
6332 trace_btrfs_space_reservation(info, "pinned",
6333 cache->space_info->flags,
6335 percpu_counter_add(&cache->space_info->total_bytes_pinned,
6337 set_extent_dirty(info->pinned_extents,
6338 bytenr, bytenr + num_bytes - 1,
6339 GFP_NOFS | __GFP_NOFAIL);
6342 spin_lock(&trans->transaction->dirty_bgs_lock);
6343 if (list_empty(&cache->dirty_list)) {
6344 list_add_tail(&cache->dirty_list,
6345 &trans->transaction->dirty_bgs);
6346 trans->transaction->num_dirty_bgs++;
6347 btrfs_get_block_group(cache);
6349 spin_unlock(&trans->transaction->dirty_bgs_lock);
6352 * No longer have used bytes in this block group, queue it for
6353 * deletion. We do this after adding the block group to the
6354 * dirty list to avoid races between cleaner kthread and space
6357 if (!alloc && old_val == 0) {
6358 spin_lock(&info->unused_bgs_lock);
6359 if (list_empty(&cache->bg_list)) {
6360 btrfs_get_block_group(cache);
6361 list_add_tail(&cache->bg_list,
6364 spin_unlock(&info->unused_bgs_lock);
6367 btrfs_put_block_group(cache);
6369 bytenr += num_bytes;
6374 static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start)
6376 struct btrfs_block_group_cache *cache;
6379 spin_lock(&fs_info->block_group_cache_lock);
6380 bytenr = fs_info->first_logical_byte;
6381 spin_unlock(&fs_info->block_group_cache_lock);
6383 if (bytenr < (u64)-1)
6386 cache = btrfs_lookup_first_block_group(fs_info, search_start);
6390 bytenr = cache->key.objectid;
6391 btrfs_put_block_group(cache);
6396 static int pin_down_extent(struct btrfs_fs_info *fs_info,
6397 struct btrfs_block_group_cache *cache,
6398 u64 bytenr, u64 num_bytes, int reserved)
6400 spin_lock(&cache->space_info->lock);
6401 spin_lock(&cache->lock);
6402 cache->pinned += num_bytes;
6403 cache->space_info->bytes_pinned += num_bytes;
6405 cache->reserved -= num_bytes;
6406 cache->space_info->bytes_reserved -= num_bytes;
6408 spin_unlock(&cache->lock);
6409 spin_unlock(&cache->space_info->lock);
6411 trace_btrfs_space_reservation(fs_info, "pinned",
6412 cache->space_info->flags, num_bytes, 1);
6413 percpu_counter_add(&cache->space_info->total_bytes_pinned, num_bytes);
6414 set_extent_dirty(fs_info->pinned_extents, bytenr,
6415 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
6420 * this function must be called within transaction
6422 int btrfs_pin_extent(struct btrfs_fs_info *fs_info,
6423 u64 bytenr, u64 num_bytes, int reserved)
6425 struct btrfs_block_group_cache *cache;
6427 cache = btrfs_lookup_block_group(fs_info, bytenr);
6428 BUG_ON(!cache); /* Logic error */
6430 pin_down_extent(fs_info, cache, bytenr, num_bytes, reserved);
6432 btrfs_put_block_group(cache);
6437 * this function must be called within transaction
6439 int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info,
6440 u64 bytenr, u64 num_bytes)
6442 struct btrfs_block_group_cache *cache;
6445 cache = btrfs_lookup_block_group(fs_info, bytenr);
6450 * pull in the free space cache (if any) so that our pin
6451 * removes the free space from the cache. We have load_only set
6452 * to one because the slow code to read in the free extents does check
6453 * the pinned extents.
6455 cache_block_group(cache, 1);
6457 pin_down_extent(fs_info, cache, bytenr, num_bytes, 0);
6459 /* remove us from the free space cache (if we're there at all) */
6460 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6461 btrfs_put_block_group(cache);
6465 static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
6466 u64 start, u64 num_bytes)
6469 struct btrfs_block_group_cache *block_group;
6470 struct btrfs_caching_control *caching_ctl;
6472 block_group = btrfs_lookup_block_group(fs_info, start);
6476 cache_block_group(block_group, 0);
6477 caching_ctl = get_caching_control(block_group);
6481 BUG_ON(!block_group_cache_done(block_group));
6482 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6484 mutex_lock(&caching_ctl->mutex);
6486 if (start >= caching_ctl->progress) {
6487 ret = add_excluded_extent(fs_info, start, num_bytes);
6488 } else if (start + num_bytes <= caching_ctl->progress) {
6489 ret = btrfs_remove_free_space(block_group,
6492 num_bytes = caching_ctl->progress - start;
6493 ret = btrfs_remove_free_space(block_group,
6498 num_bytes = (start + num_bytes) -
6499 caching_ctl->progress;
6500 start = caching_ctl->progress;
6501 ret = add_excluded_extent(fs_info, start, num_bytes);
6504 mutex_unlock(&caching_ctl->mutex);
6505 put_caching_control(caching_ctl);
6507 btrfs_put_block_group(block_group);
6511 int btrfs_exclude_logged_extents(struct btrfs_fs_info *fs_info,
6512 struct extent_buffer *eb)
6514 struct btrfs_file_extent_item *item;
6515 struct btrfs_key key;
6519 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
6522 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6523 btrfs_item_key_to_cpu(eb, &key, i);
6524 if (key.type != BTRFS_EXTENT_DATA_KEY)
6526 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6527 found_type = btrfs_file_extent_type(eb, item);
6528 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6530 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6532 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6533 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6534 __exclude_logged_extent(fs_info, key.objectid, key.offset);
6541 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg)
6543 atomic_inc(&bg->reservations);
6546 void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info,
6549 struct btrfs_block_group_cache *bg;
6551 bg = btrfs_lookup_block_group(fs_info, start);
6553 if (atomic_dec_and_test(&bg->reservations))
6554 wake_up_var(&bg->reservations);
6555 btrfs_put_block_group(bg);
6558 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg)
6560 struct btrfs_space_info *space_info = bg->space_info;
6564 if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA))
6568 * Our block group is read only but before we set it to read only,
6569 * some task might have had allocated an extent from it already, but it
6570 * has not yet created a respective ordered extent (and added it to a
6571 * root's list of ordered extents).
6572 * Therefore wait for any task currently allocating extents, since the
6573 * block group's reservations counter is incremented while a read lock
6574 * on the groups' semaphore is held and decremented after releasing
6575 * the read access on that semaphore and creating the ordered extent.
6577 down_write(&space_info->groups_sem);
6578 up_write(&space_info->groups_sem);
6580 wait_var_event(&bg->reservations, !atomic_read(&bg->reservations));
6584 * btrfs_add_reserved_bytes - update the block_group and space info counters
6585 * @cache: The cache we are manipulating
6586 * @ram_bytes: The number of bytes of file content, and will be same to
6587 * @num_bytes except for the compress path.
6588 * @num_bytes: The number of bytes in question
6589 * @delalloc: The blocks are allocated for the delalloc write
6591 * This is called by the allocator when it reserves space. If this is a
6592 * reservation and the block group has become read only we cannot make the
6593 * reservation and return -EAGAIN, otherwise this function always succeeds.
6595 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache,
6596 u64 ram_bytes, u64 num_bytes, int delalloc)
6598 struct btrfs_space_info *space_info = cache->space_info;
6601 spin_lock(&space_info->lock);
6602 spin_lock(&cache->lock);
6606 cache->reserved += num_bytes;
6607 space_info->bytes_reserved += num_bytes;
6609 trace_btrfs_space_reservation(cache->fs_info,
6610 "space_info", space_info->flags,
6612 space_info->bytes_may_use -= ram_bytes;
6614 cache->delalloc_bytes += num_bytes;
6616 spin_unlock(&cache->lock);
6617 spin_unlock(&space_info->lock);
6622 * btrfs_free_reserved_bytes - update the block_group and space info counters
6623 * @cache: The cache we are manipulating
6624 * @num_bytes: The number of bytes in question
6625 * @delalloc: The blocks are allocated for the delalloc write
6627 * This is called by somebody who is freeing space that was never actually used
6628 * on disk. For example if you reserve some space for a new leaf in transaction
6629 * A and before transaction A commits you free that leaf, you call this with
6630 * reserve set to 0 in order to clear the reservation.
6633 static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache,
6634 u64 num_bytes, int delalloc)
6636 struct btrfs_space_info *space_info = cache->space_info;
6639 spin_lock(&space_info->lock);
6640 spin_lock(&cache->lock);
6642 space_info->bytes_readonly += num_bytes;
6643 cache->reserved -= num_bytes;
6644 space_info->bytes_reserved -= num_bytes;
6647 cache->delalloc_bytes -= num_bytes;
6648 spin_unlock(&cache->lock);
6649 spin_unlock(&space_info->lock);
6652 void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info)
6654 struct btrfs_caching_control *next;
6655 struct btrfs_caching_control *caching_ctl;
6656 struct btrfs_block_group_cache *cache;
6658 down_write(&fs_info->commit_root_sem);
6660 list_for_each_entry_safe(caching_ctl, next,
6661 &fs_info->caching_block_groups, list) {
6662 cache = caching_ctl->block_group;
6663 if (block_group_cache_done(cache)) {
6664 cache->last_byte_to_unpin = (u64)-1;
6665 list_del_init(&caching_ctl->list);
6666 put_caching_control(caching_ctl);
6668 cache->last_byte_to_unpin = caching_ctl->progress;
6672 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6673 fs_info->pinned_extents = &fs_info->freed_extents[1];
6675 fs_info->pinned_extents = &fs_info->freed_extents[0];
6677 up_write(&fs_info->commit_root_sem);
6679 update_global_block_rsv(fs_info);
6683 * Returns the free cluster for the given space info and sets empty_cluster to
6684 * what it should be based on the mount options.
6686 static struct btrfs_free_cluster *
6687 fetch_cluster_info(struct btrfs_fs_info *fs_info,
6688 struct btrfs_space_info *space_info, u64 *empty_cluster)
6690 struct btrfs_free_cluster *ret = NULL;
6693 if (btrfs_mixed_space_info(space_info))
6696 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
6697 ret = &fs_info->meta_alloc_cluster;
6698 if (btrfs_test_opt(fs_info, SSD))
6699 *empty_cluster = SZ_2M;
6701 *empty_cluster = SZ_64K;
6702 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
6703 btrfs_test_opt(fs_info, SSD_SPREAD)) {
6704 *empty_cluster = SZ_2M;
6705 ret = &fs_info->data_alloc_cluster;
6711 static int unpin_extent_range(struct btrfs_fs_info *fs_info,
6713 const bool return_free_space)
6715 struct btrfs_block_group_cache *cache = NULL;
6716 struct btrfs_space_info *space_info;
6717 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6718 struct btrfs_free_cluster *cluster = NULL;
6720 u64 total_unpinned = 0;
6721 u64 empty_cluster = 0;
6724 while (start <= end) {
6727 start >= cache->key.objectid + cache->key.offset) {
6729 btrfs_put_block_group(cache);
6731 cache = btrfs_lookup_block_group(fs_info, start);
6732 BUG_ON(!cache); /* Logic error */
6734 cluster = fetch_cluster_info(fs_info,
6737 empty_cluster <<= 1;
6740 len = cache->key.objectid + cache->key.offset - start;
6741 len = min(len, end + 1 - start);
6743 if (start < cache->last_byte_to_unpin) {
6744 len = min(len, cache->last_byte_to_unpin - start);
6745 if (return_free_space)
6746 btrfs_add_free_space(cache, start, len);
6750 total_unpinned += len;
6751 space_info = cache->space_info;
6754 * If this space cluster has been marked as fragmented and we've
6755 * unpinned enough in this block group to potentially allow a
6756 * cluster to be created inside of it go ahead and clear the
6759 if (cluster && cluster->fragmented &&
6760 total_unpinned > empty_cluster) {
6761 spin_lock(&cluster->lock);
6762 cluster->fragmented = 0;
6763 spin_unlock(&cluster->lock);
6766 spin_lock(&space_info->lock);
6767 spin_lock(&cache->lock);
6768 cache->pinned -= len;
6769 space_info->bytes_pinned -= len;
6771 trace_btrfs_space_reservation(fs_info, "pinned",
6772 space_info->flags, len, 0);
6773 space_info->max_extent_size = 0;
6774 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6776 space_info->bytes_readonly += len;
6779 spin_unlock(&cache->lock);
6780 if (!readonly && return_free_space &&
6781 global_rsv->space_info == space_info) {
6784 spin_lock(&global_rsv->lock);
6785 if (!global_rsv->full) {
6786 to_add = min(len, global_rsv->size -
6787 global_rsv->reserved);
6788 global_rsv->reserved += to_add;
6789 space_info->bytes_may_use += to_add;
6790 if (global_rsv->reserved >= global_rsv->size)
6791 global_rsv->full = 1;
6792 trace_btrfs_space_reservation(fs_info,
6798 spin_unlock(&global_rsv->lock);
6799 /* Add to any tickets we may have */
6801 space_info_add_new_bytes(fs_info, space_info,
6804 spin_unlock(&space_info->lock);
6808 btrfs_put_block_group(cache);
6812 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans)
6814 struct btrfs_fs_info *fs_info = trans->fs_info;
6815 struct btrfs_block_group_cache *block_group, *tmp;
6816 struct list_head *deleted_bgs;
6817 struct extent_io_tree *unpin;
6822 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6823 unpin = &fs_info->freed_extents[1];
6825 unpin = &fs_info->freed_extents[0];
6827 while (!trans->aborted) {
6828 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6829 ret = find_first_extent_bit(unpin, 0, &start, &end,
6830 EXTENT_DIRTY, NULL);
6832 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6836 if (btrfs_test_opt(fs_info, DISCARD))
6837 ret = btrfs_discard_extent(fs_info, start,
6838 end + 1 - start, NULL);
6840 clear_extent_dirty(unpin, start, end);
6841 unpin_extent_range(fs_info, start, end, true);
6842 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6847 * Transaction is finished. We don't need the lock anymore. We
6848 * do need to clean up the block groups in case of a transaction
6851 deleted_bgs = &trans->transaction->deleted_bgs;
6852 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6856 if (!trans->aborted)
6857 ret = btrfs_discard_extent(fs_info,
6858 block_group->key.objectid,
6859 block_group->key.offset,
6862 list_del_init(&block_group->bg_list);
6863 btrfs_put_block_group_trimming(block_group);
6864 btrfs_put_block_group(block_group);
6867 const char *errstr = btrfs_decode_error(ret);
6869 "discard failed while removing blockgroup: errno=%d %s",
6877 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6878 struct btrfs_fs_info *info,
6879 struct btrfs_delayed_ref_node *node, u64 parent,
6880 u64 root_objectid, u64 owner_objectid,
6881 u64 owner_offset, int refs_to_drop,
6882 struct btrfs_delayed_extent_op *extent_op)
6884 struct btrfs_key key;
6885 struct btrfs_path *path;
6886 struct btrfs_root *extent_root = info->extent_root;
6887 struct extent_buffer *leaf;
6888 struct btrfs_extent_item *ei;
6889 struct btrfs_extent_inline_ref *iref;
6892 int extent_slot = 0;
6893 int found_extent = 0;
6897 u64 bytenr = node->bytenr;
6898 u64 num_bytes = node->num_bytes;
6900 bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
6902 path = btrfs_alloc_path();
6906 path->reada = READA_FORWARD;
6907 path->leave_spinning = 1;
6909 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6910 BUG_ON(!is_data && refs_to_drop != 1);
6913 skinny_metadata = false;
6915 ret = lookup_extent_backref(trans, info, path, &iref,
6916 bytenr, num_bytes, parent,
6917 root_objectid, owner_objectid,
6920 extent_slot = path->slots[0];
6921 while (extent_slot >= 0) {
6922 btrfs_item_key_to_cpu(path->nodes[0], &key,
6924 if (key.objectid != bytenr)
6926 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6927 key.offset == num_bytes) {
6931 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6932 key.offset == owner_objectid) {
6936 if (path->slots[0] - extent_slot > 5)
6940 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6941 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6942 if (found_extent && item_size < sizeof(*ei))
6945 if (!found_extent) {
6947 ret = remove_extent_backref(trans, info, path, NULL,
6949 is_data, &last_ref);
6951 btrfs_abort_transaction(trans, ret);
6954 btrfs_release_path(path);
6955 path->leave_spinning = 1;
6957 key.objectid = bytenr;
6958 key.type = BTRFS_EXTENT_ITEM_KEY;
6959 key.offset = num_bytes;
6961 if (!is_data && skinny_metadata) {
6962 key.type = BTRFS_METADATA_ITEM_KEY;
6963 key.offset = owner_objectid;
6966 ret = btrfs_search_slot(trans, extent_root,
6968 if (ret > 0 && skinny_metadata && path->slots[0]) {
6970 * Couldn't find our skinny metadata item,
6971 * see if we have ye olde extent item.
6974 btrfs_item_key_to_cpu(path->nodes[0], &key,
6976 if (key.objectid == bytenr &&
6977 key.type == BTRFS_EXTENT_ITEM_KEY &&
6978 key.offset == num_bytes)
6982 if (ret > 0 && skinny_metadata) {
6983 skinny_metadata = false;
6984 key.objectid = bytenr;
6985 key.type = BTRFS_EXTENT_ITEM_KEY;
6986 key.offset = num_bytes;
6987 btrfs_release_path(path);
6988 ret = btrfs_search_slot(trans, extent_root,
6994 "umm, got %d back from search, was looking for %llu",
6997 btrfs_print_leaf(path->nodes[0]);
7000 btrfs_abort_transaction(trans, ret);
7003 extent_slot = path->slots[0];
7005 } else if (WARN_ON(ret == -ENOENT)) {
7006 btrfs_print_leaf(path->nodes[0]);
7008 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
7009 bytenr, parent, root_objectid, owner_objectid,
7011 btrfs_abort_transaction(trans, ret);
7014 btrfs_abort_transaction(trans, ret);
7018 leaf = path->nodes[0];
7019 item_size = btrfs_item_size_nr(leaf, extent_slot);
7020 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
7021 if (item_size < sizeof(*ei)) {
7022 BUG_ON(found_extent || extent_slot != path->slots[0]);
7023 ret = convert_extent_item_v0(trans, info, path, owner_objectid,
7026 btrfs_abort_transaction(trans, ret);
7030 btrfs_release_path(path);
7031 path->leave_spinning = 1;
7033 key.objectid = bytenr;
7034 key.type = BTRFS_EXTENT_ITEM_KEY;
7035 key.offset = num_bytes;
7037 ret = btrfs_search_slot(trans, extent_root, &key, path,
7041 "umm, got %d back from search, was looking for %llu",
7043 btrfs_print_leaf(path->nodes[0]);
7046 btrfs_abort_transaction(trans, ret);
7050 extent_slot = path->slots[0];
7051 leaf = path->nodes[0];
7052 item_size = btrfs_item_size_nr(leaf, extent_slot);
7055 BUG_ON(item_size < sizeof(*ei));
7056 ei = btrfs_item_ptr(leaf, extent_slot,
7057 struct btrfs_extent_item);
7058 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
7059 key.type == BTRFS_EXTENT_ITEM_KEY) {
7060 struct btrfs_tree_block_info *bi;
7061 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
7062 bi = (struct btrfs_tree_block_info *)(ei + 1);
7063 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
7066 refs = btrfs_extent_refs(leaf, ei);
7067 if (refs < refs_to_drop) {
7069 "trying to drop %d refs but we only have %Lu for bytenr %Lu",
7070 refs_to_drop, refs, bytenr);
7072 btrfs_abort_transaction(trans, ret);
7075 refs -= refs_to_drop;
7079 __run_delayed_extent_op(extent_op, leaf, ei);
7081 * In the case of inline back ref, reference count will
7082 * be updated by remove_extent_backref
7085 BUG_ON(!found_extent);
7087 btrfs_set_extent_refs(leaf, ei, refs);
7088 btrfs_mark_buffer_dirty(leaf);
7091 ret = remove_extent_backref(trans, info, path,
7093 is_data, &last_ref);
7095 btrfs_abort_transaction(trans, ret);
7101 BUG_ON(is_data && refs_to_drop !=
7102 extent_data_ref_count(path, iref));
7104 BUG_ON(path->slots[0] != extent_slot);
7106 BUG_ON(path->slots[0] != extent_slot + 1);
7107 path->slots[0] = extent_slot;
7113 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
7116 btrfs_abort_transaction(trans, ret);
7119 btrfs_release_path(path);
7122 ret = btrfs_del_csums(trans, info, bytenr, num_bytes);
7124 btrfs_abort_transaction(trans, ret);
7129 ret = add_to_free_space_tree(trans, info, bytenr, num_bytes);
7131 btrfs_abort_transaction(trans, ret);
7135 ret = update_block_group(trans, info, bytenr, num_bytes, 0);
7137 btrfs_abort_transaction(trans, ret);
7141 btrfs_release_path(path);
7144 btrfs_free_path(path);
7149 * when we free an block, it is possible (and likely) that we free the last
7150 * delayed ref for that extent as well. This searches the delayed ref tree for
7151 * a given extent, and if there are no other delayed refs to be processed, it
7152 * removes it from the tree.
7154 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
7157 struct btrfs_delayed_ref_head *head;
7158 struct btrfs_delayed_ref_root *delayed_refs;
7161 delayed_refs = &trans->transaction->delayed_refs;
7162 spin_lock(&delayed_refs->lock);
7163 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
7165 goto out_delayed_unlock;
7167 spin_lock(&head->lock);
7168 if (!RB_EMPTY_ROOT(&head->ref_tree))
7171 if (head->extent_op) {
7172 if (!head->must_insert_reserved)
7174 btrfs_free_delayed_extent_op(head->extent_op);
7175 head->extent_op = NULL;
7179 * waiting for the lock here would deadlock. If someone else has it
7180 * locked they are already in the process of dropping it anyway
7182 if (!mutex_trylock(&head->mutex))
7186 * at this point we have a head with no other entries. Go
7187 * ahead and process it.
7189 rb_erase(&head->href_node, &delayed_refs->href_root);
7190 RB_CLEAR_NODE(&head->href_node);
7191 atomic_dec(&delayed_refs->num_entries);
7194 * we don't take a ref on the node because we're removing it from the
7195 * tree, so we just steal the ref the tree was holding.
7197 delayed_refs->num_heads--;
7198 if (head->processing == 0)
7199 delayed_refs->num_heads_ready--;
7200 head->processing = 0;
7201 spin_unlock(&head->lock);
7202 spin_unlock(&delayed_refs->lock);
7204 BUG_ON(head->extent_op);
7205 if (head->must_insert_reserved)
7208 mutex_unlock(&head->mutex);
7209 btrfs_put_delayed_ref_head(head);
7212 spin_unlock(&head->lock);
7215 spin_unlock(&delayed_refs->lock);
7219 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
7220 struct btrfs_root *root,
7221 struct extent_buffer *buf,
7222 u64 parent, int last_ref)
7224 struct btrfs_fs_info *fs_info = root->fs_info;
7228 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7229 int old_ref_mod, new_ref_mod;
7231 btrfs_ref_tree_mod(root, buf->start, buf->len, parent,
7232 root->root_key.objectid,
7233 btrfs_header_level(buf), 0,
7234 BTRFS_DROP_DELAYED_REF);
7235 ret = btrfs_add_delayed_tree_ref(fs_info, trans, buf->start,
7237 root->root_key.objectid,
7238 btrfs_header_level(buf),
7239 BTRFS_DROP_DELAYED_REF, NULL,
7240 &old_ref_mod, &new_ref_mod);
7241 BUG_ON(ret); /* -ENOMEM */
7242 pin = old_ref_mod >= 0 && new_ref_mod < 0;
7245 if (last_ref && btrfs_header_generation(buf) == trans->transid) {
7246 struct btrfs_block_group_cache *cache;
7248 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7249 ret = check_ref_cleanup(trans, buf->start);
7255 cache = btrfs_lookup_block_group(fs_info, buf->start);
7257 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
7258 pin_down_extent(fs_info, cache, buf->start,
7260 btrfs_put_block_group(cache);
7264 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
7266 btrfs_add_free_space(cache, buf->start, buf->len);
7267 btrfs_free_reserved_bytes(cache, buf->len, 0);
7268 btrfs_put_block_group(cache);
7269 trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
7273 add_pinned_bytes(fs_info, buf->len, true,
7274 root->root_key.objectid);
7278 * Deleting the buffer, clear the corrupt flag since it doesn't
7281 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
7285 /* Can return -ENOMEM */
7286 int btrfs_free_extent(struct btrfs_trans_handle *trans,
7287 struct btrfs_root *root,
7288 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
7289 u64 owner, u64 offset)
7291 struct btrfs_fs_info *fs_info = root->fs_info;
7292 int old_ref_mod, new_ref_mod;
7295 if (btrfs_is_testing(fs_info))
7298 if (root_objectid != BTRFS_TREE_LOG_OBJECTID)
7299 btrfs_ref_tree_mod(root, bytenr, num_bytes, parent,
7300 root_objectid, owner, offset,
7301 BTRFS_DROP_DELAYED_REF);
7304 * tree log blocks never actually go into the extent allocation
7305 * tree, just update pinning info and exit early.
7307 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
7308 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
7309 /* unlocks the pinned mutex */
7310 btrfs_pin_extent(fs_info, bytenr, num_bytes, 1);
7311 old_ref_mod = new_ref_mod = 0;
7313 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
7314 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
7316 root_objectid, (int)owner,
7317 BTRFS_DROP_DELAYED_REF, NULL,
7318 &old_ref_mod, &new_ref_mod);
7320 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
7322 root_objectid, owner, offset,
7323 0, BTRFS_DROP_DELAYED_REF,
7324 &old_ref_mod, &new_ref_mod);
7327 if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0) {
7328 bool metadata = owner < BTRFS_FIRST_FREE_OBJECTID;
7330 add_pinned_bytes(fs_info, num_bytes, metadata, root_objectid);
7337 * when we wait for progress in the block group caching, its because
7338 * our allocation attempt failed at least once. So, we must sleep
7339 * and let some progress happen before we try again.
7341 * This function will sleep at least once waiting for new free space to
7342 * show up, and then it will check the block group free space numbers
7343 * for our min num_bytes. Another option is to have it go ahead
7344 * and look in the rbtree for a free extent of a given size, but this
7347 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7348 * any of the information in this block group.
7350 static noinline void
7351 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
7354 struct btrfs_caching_control *caching_ctl;
7356 caching_ctl = get_caching_control(cache);
7360 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
7361 (cache->free_space_ctl->free_space >= num_bytes));
7363 put_caching_control(caching_ctl);
7367 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
7369 struct btrfs_caching_control *caching_ctl;
7372 caching_ctl = get_caching_control(cache);
7374 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
7376 wait_event(caching_ctl->wait, block_group_cache_done(cache));
7377 if (cache->cached == BTRFS_CACHE_ERROR)
7379 put_caching_control(caching_ctl);
7383 enum btrfs_loop_type {
7384 LOOP_CACHING_NOWAIT = 0,
7385 LOOP_CACHING_WAIT = 1,
7386 LOOP_ALLOC_CHUNK = 2,
7387 LOOP_NO_EMPTY_SIZE = 3,
7391 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
7395 down_read(&cache->data_rwsem);
7399 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
7402 btrfs_get_block_group(cache);
7404 down_read(&cache->data_rwsem);
7407 static struct btrfs_block_group_cache *
7408 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
7409 struct btrfs_free_cluster *cluster,
7412 struct btrfs_block_group_cache *used_bg = NULL;
7414 spin_lock(&cluster->refill_lock);
7416 used_bg = cluster->block_group;
7420 if (used_bg == block_group)
7423 btrfs_get_block_group(used_bg);
7428 if (down_read_trylock(&used_bg->data_rwsem))
7431 spin_unlock(&cluster->refill_lock);
7433 /* We should only have one-level nested. */
7434 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
7436 spin_lock(&cluster->refill_lock);
7437 if (used_bg == cluster->block_group)
7440 up_read(&used_bg->data_rwsem);
7441 btrfs_put_block_group(used_bg);
7446 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
7450 up_read(&cache->data_rwsem);
7451 btrfs_put_block_group(cache);
7455 * walks the btree of allocated extents and find a hole of a given size.
7456 * The key ins is changed to record the hole:
7457 * ins->objectid == start position
7458 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7459 * ins->offset == the size of the hole.
7460 * Any available blocks before search_start are skipped.
7462 * If there is no suitable free space, we will record the max size of
7463 * the free space extent currently.
7465 static noinline int find_free_extent(struct btrfs_fs_info *fs_info,
7466 u64 ram_bytes, u64 num_bytes, u64 empty_size,
7467 u64 hint_byte, struct btrfs_key *ins,
7468 u64 flags, int delalloc)
7471 struct btrfs_root *root = fs_info->extent_root;
7472 struct btrfs_free_cluster *last_ptr = NULL;
7473 struct btrfs_block_group_cache *block_group = NULL;
7474 u64 search_start = 0;
7475 u64 max_extent_size = 0;
7476 u64 empty_cluster = 0;
7477 struct btrfs_space_info *space_info;
7479 int index = btrfs_bg_flags_to_raid_index(flags);
7480 bool failed_cluster_refill = false;
7481 bool failed_alloc = false;
7482 bool use_cluster = true;
7483 bool have_caching_bg = false;
7484 bool orig_have_caching_bg = false;
7485 bool full_search = false;
7487 WARN_ON(num_bytes < fs_info->sectorsize);
7488 ins->type = BTRFS_EXTENT_ITEM_KEY;
7492 trace_find_free_extent(fs_info, num_bytes, empty_size, flags);
7494 space_info = __find_space_info(fs_info, flags);
7496 btrfs_err(fs_info, "No space info for %llu", flags);
7501 * If our free space is heavily fragmented we may not be able to make
7502 * big contiguous allocations, so instead of doing the expensive search
7503 * for free space, simply return ENOSPC with our max_extent_size so we
7504 * can go ahead and search for a more manageable chunk.
7506 * If our max_extent_size is large enough for our allocation simply
7507 * disable clustering since we will likely not be able to find enough
7508 * space to create a cluster and induce latency trying.
7510 if (unlikely(space_info->max_extent_size)) {
7511 spin_lock(&space_info->lock);
7512 if (space_info->max_extent_size &&
7513 num_bytes > space_info->max_extent_size) {
7514 ins->offset = space_info->max_extent_size;
7515 spin_unlock(&space_info->lock);
7517 } else if (space_info->max_extent_size) {
7518 use_cluster = false;
7520 spin_unlock(&space_info->lock);
7523 last_ptr = fetch_cluster_info(fs_info, space_info, &empty_cluster);
7525 spin_lock(&last_ptr->lock);
7526 if (last_ptr->block_group)
7527 hint_byte = last_ptr->window_start;
7528 if (last_ptr->fragmented) {
7530 * We still set window_start so we can keep track of the
7531 * last place we found an allocation to try and save
7534 hint_byte = last_ptr->window_start;
7535 use_cluster = false;
7537 spin_unlock(&last_ptr->lock);
7540 search_start = max(search_start, first_logical_byte(fs_info, 0));
7541 search_start = max(search_start, hint_byte);
7542 if (search_start == hint_byte) {
7543 block_group = btrfs_lookup_block_group(fs_info, search_start);
7545 * we don't want to use the block group if it doesn't match our
7546 * allocation bits, or if its not cached.
7548 * However if we are re-searching with an ideal block group
7549 * picked out then we don't care that the block group is cached.
7551 if (block_group && block_group_bits(block_group, flags) &&
7552 block_group->cached != BTRFS_CACHE_NO) {
7553 down_read(&space_info->groups_sem);
7554 if (list_empty(&block_group->list) ||
7557 * someone is removing this block group,
7558 * we can't jump into the have_block_group
7559 * target because our list pointers are not
7562 btrfs_put_block_group(block_group);
7563 up_read(&space_info->groups_sem);
7565 index = btrfs_bg_flags_to_raid_index(
7566 block_group->flags);
7567 btrfs_lock_block_group(block_group, delalloc);
7568 goto have_block_group;
7570 } else if (block_group) {
7571 btrfs_put_block_group(block_group);
7575 have_caching_bg = false;
7576 if (index == 0 || index == btrfs_bg_flags_to_raid_index(flags))
7578 down_read(&space_info->groups_sem);
7579 list_for_each_entry(block_group, &space_info->block_groups[index],
7584 /* If the block group is read-only, we can skip it entirely. */
7585 if (unlikely(block_group->ro))
7588 btrfs_grab_block_group(block_group, delalloc);
7589 search_start = block_group->key.objectid;
7592 * this can happen if we end up cycling through all the
7593 * raid types, but we want to make sure we only allocate
7594 * for the proper type.
7596 if (!block_group_bits(block_group, flags)) {
7597 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7598 BTRFS_BLOCK_GROUP_RAID1 |
7599 BTRFS_BLOCK_GROUP_RAID5 |
7600 BTRFS_BLOCK_GROUP_RAID6 |
7601 BTRFS_BLOCK_GROUP_RAID10;
7604 * if they asked for extra copies and this block group
7605 * doesn't provide them, bail. This does allow us to
7606 * fill raid0 from raid1.
7608 if ((flags & extra) && !(block_group->flags & extra))
7613 cached = block_group_cache_done(block_group);
7614 if (unlikely(!cached)) {
7615 have_caching_bg = true;
7616 ret = cache_block_group(block_group, 0);
7621 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7625 * Ok we want to try and use the cluster allocator, so
7628 if (last_ptr && use_cluster) {
7629 struct btrfs_block_group_cache *used_block_group;
7630 unsigned long aligned_cluster;
7632 * the refill lock keeps out other
7633 * people trying to start a new cluster
7635 used_block_group = btrfs_lock_cluster(block_group,
7638 if (!used_block_group)
7639 goto refill_cluster;
7641 if (used_block_group != block_group &&
7642 (used_block_group->ro ||
7643 !block_group_bits(used_block_group, flags)))
7644 goto release_cluster;
7646 offset = btrfs_alloc_from_cluster(used_block_group,
7649 used_block_group->key.objectid,
7652 /* we have a block, we're done */
7653 spin_unlock(&last_ptr->refill_lock);
7654 trace_btrfs_reserve_extent_cluster(fs_info,
7656 search_start, num_bytes);
7657 if (used_block_group != block_group) {
7658 btrfs_release_block_group(block_group,
7660 block_group = used_block_group;
7665 WARN_ON(last_ptr->block_group != used_block_group);
7667 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7668 * set up a new clusters, so lets just skip it
7669 * and let the allocator find whatever block
7670 * it can find. If we reach this point, we
7671 * will have tried the cluster allocator
7672 * plenty of times and not have found
7673 * anything, so we are likely way too
7674 * fragmented for the clustering stuff to find
7677 * However, if the cluster is taken from the
7678 * current block group, release the cluster
7679 * first, so that we stand a better chance of
7680 * succeeding in the unclustered
7682 if (loop >= LOOP_NO_EMPTY_SIZE &&
7683 used_block_group != block_group) {
7684 spin_unlock(&last_ptr->refill_lock);
7685 btrfs_release_block_group(used_block_group,
7687 goto unclustered_alloc;
7691 * this cluster didn't work out, free it and
7694 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7696 if (used_block_group != block_group)
7697 btrfs_release_block_group(used_block_group,
7700 if (loop >= LOOP_NO_EMPTY_SIZE) {
7701 spin_unlock(&last_ptr->refill_lock);
7702 goto unclustered_alloc;
7705 aligned_cluster = max_t(unsigned long,
7706 empty_cluster + empty_size,
7707 block_group->full_stripe_len);
7709 /* allocate a cluster in this block group */
7710 ret = btrfs_find_space_cluster(fs_info, block_group,
7711 last_ptr, search_start,
7716 * now pull our allocation out of this
7719 offset = btrfs_alloc_from_cluster(block_group,
7725 /* we found one, proceed */
7726 spin_unlock(&last_ptr->refill_lock);
7727 trace_btrfs_reserve_extent_cluster(fs_info,
7728 block_group, search_start,
7732 } else if (!cached && loop > LOOP_CACHING_NOWAIT
7733 && !failed_cluster_refill) {
7734 spin_unlock(&last_ptr->refill_lock);
7736 failed_cluster_refill = true;
7737 wait_block_group_cache_progress(block_group,
7738 num_bytes + empty_cluster + empty_size);
7739 goto have_block_group;
7743 * at this point we either didn't find a cluster
7744 * or we weren't able to allocate a block from our
7745 * cluster. Free the cluster we've been trying
7746 * to use, and go to the next block group
7748 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7749 spin_unlock(&last_ptr->refill_lock);
7755 * We are doing an unclustered alloc, set the fragmented flag so
7756 * we don't bother trying to setup a cluster again until we get
7759 if (unlikely(last_ptr)) {
7760 spin_lock(&last_ptr->lock);
7761 last_ptr->fragmented = 1;
7762 spin_unlock(&last_ptr->lock);
7765 struct btrfs_free_space_ctl *ctl =
7766 block_group->free_space_ctl;
7768 spin_lock(&ctl->tree_lock);
7769 if (ctl->free_space <
7770 num_bytes + empty_cluster + empty_size) {
7771 if (ctl->free_space > max_extent_size)
7772 max_extent_size = ctl->free_space;
7773 spin_unlock(&ctl->tree_lock);
7776 spin_unlock(&ctl->tree_lock);
7779 offset = btrfs_find_space_for_alloc(block_group, search_start,
7780 num_bytes, empty_size,
7783 * If we didn't find a chunk, and we haven't failed on this
7784 * block group before, and this block group is in the middle of
7785 * caching and we are ok with waiting, then go ahead and wait
7786 * for progress to be made, and set failed_alloc to true.
7788 * If failed_alloc is true then we've already waited on this
7789 * block group once and should move on to the next block group.
7791 if (!offset && !failed_alloc && !cached &&
7792 loop > LOOP_CACHING_NOWAIT) {
7793 wait_block_group_cache_progress(block_group,
7794 num_bytes + empty_size);
7795 failed_alloc = true;
7796 goto have_block_group;
7797 } else if (!offset) {
7801 search_start = ALIGN(offset, fs_info->stripesize);
7803 /* move on to the next group */
7804 if (search_start + num_bytes >
7805 block_group->key.objectid + block_group->key.offset) {
7806 btrfs_add_free_space(block_group, offset, num_bytes);
7810 if (offset < search_start)
7811 btrfs_add_free_space(block_group, offset,
7812 search_start - offset);
7813 BUG_ON(offset > search_start);
7815 ret = btrfs_add_reserved_bytes(block_group, ram_bytes,
7816 num_bytes, delalloc);
7817 if (ret == -EAGAIN) {
7818 btrfs_add_free_space(block_group, offset, num_bytes);
7821 btrfs_inc_block_group_reservations(block_group);
7823 /* we are all good, lets return */
7824 ins->objectid = search_start;
7825 ins->offset = num_bytes;
7827 trace_btrfs_reserve_extent(fs_info, block_group,
7828 search_start, num_bytes);
7829 btrfs_release_block_group(block_group, delalloc);
7832 failed_cluster_refill = false;
7833 failed_alloc = false;
7834 BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) !=
7836 btrfs_release_block_group(block_group, delalloc);
7839 up_read(&space_info->groups_sem);
7841 if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg
7842 && !orig_have_caching_bg)
7843 orig_have_caching_bg = true;
7845 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7848 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7852 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7853 * caching kthreads as we move along
7854 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7855 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7856 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7859 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7861 if (loop == LOOP_CACHING_NOWAIT) {
7863 * We want to skip the LOOP_CACHING_WAIT step if we
7864 * don't have any uncached bgs and we've already done a
7865 * full search through.
7867 if (orig_have_caching_bg || !full_search)
7868 loop = LOOP_CACHING_WAIT;
7870 loop = LOOP_ALLOC_CHUNK;
7875 if (loop == LOOP_ALLOC_CHUNK) {
7876 struct btrfs_trans_handle *trans;
7879 trans = current->journal_info;
7883 trans = btrfs_join_transaction(root);
7885 if (IS_ERR(trans)) {
7886 ret = PTR_ERR(trans);
7890 ret = do_chunk_alloc(trans, fs_info, flags,
7894 * If we can't allocate a new chunk we've already looped
7895 * through at least once, move on to the NO_EMPTY_SIZE
7899 loop = LOOP_NO_EMPTY_SIZE;
7902 * Do not bail out on ENOSPC since we
7903 * can do more things.
7905 if (ret < 0 && ret != -ENOSPC)
7906 btrfs_abort_transaction(trans, ret);
7910 btrfs_end_transaction(trans);
7915 if (loop == LOOP_NO_EMPTY_SIZE) {
7917 * Don't loop again if we already have no empty_size and
7920 if (empty_size == 0 &&
7921 empty_cluster == 0) {
7930 } else if (!ins->objectid) {
7932 } else if (ins->objectid) {
7933 if (!use_cluster && last_ptr) {
7934 spin_lock(&last_ptr->lock);
7935 last_ptr->window_start = ins->objectid;
7936 spin_unlock(&last_ptr->lock);
7941 if (ret == -ENOSPC) {
7942 spin_lock(&space_info->lock);
7943 space_info->max_extent_size = max_extent_size;
7944 spin_unlock(&space_info->lock);
7945 ins->offset = max_extent_size;
7950 static void dump_space_info(struct btrfs_fs_info *fs_info,
7951 struct btrfs_space_info *info, u64 bytes,
7952 int dump_block_groups)
7954 struct btrfs_block_group_cache *cache;
7957 spin_lock(&info->lock);
7958 btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull",
7960 info->total_bytes - btrfs_space_info_used(info, true),
7961 info->full ? "" : "not ");
7963 "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu",
7964 info->total_bytes, info->bytes_used, info->bytes_pinned,
7965 info->bytes_reserved, info->bytes_may_use,
7966 info->bytes_readonly);
7967 spin_unlock(&info->lock);
7969 if (!dump_block_groups)
7972 down_read(&info->groups_sem);
7974 list_for_each_entry(cache, &info->block_groups[index], list) {
7975 spin_lock(&cache->lock);
7977 "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s",
7978 cache->key.objectid, cache->key.offset,
7979 btrfs_block_group_used(&cache->item), cache->pinned,
7980 cache->reserved, cache->ro ? "[readonly]" : "");
7981 btrfs_dump_free_space(cache, bytes);
7982 spin_unlock(&cache->lock);
7984 if (++index < BTRFS_NR_RAID_TYPES)
7986 up_read(&info->groups_sem);
7990 * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a
7991 * hole that is at least as big as @num_bytes.
7993 * @root - The root that will contain this extent
7995 * @ram_bytes - The amount of space in ram that @num_bytes take. This
7996 * is used for accounting purposes. This value differs
7997 * from @num_bytes only in the case of compressed extents.
7999 * @num_bytes - Number of bytes to allocate on-disk.
8001 * @min_alloc_size - Indicates the minimum amount of space that the
8002 * allocator should try to satisfy. In some cases
8003 * @num_bytes may be larger than what is required and if
8004 * the filesystem is fragmented then allocation fails.
8005 * However, the presence of @min_alloc_size gives a
8006 * chance to try and satisfy the smaller allocation.
8008 * @empty_size - A hint that you plan on doing more COW. This is the
8009 * size in bytes the allocator should try to find free
8010 * next to the block it returns. This is just a hint and
8011 * may be ignored by the allocator.
8013 * @hint_byte - Hint to the allocator to start searching above the byte
8014 * address passed. It might be ignored.
8016 * @ins - This key is modified to record the found hole. It will
8017 * have the following values:
8018 * ins->objectid == start position
8019 * ins->flags = BTRFS_EXTENT_ITEM_KEY
8020 * ins->offset == the size of the hole.
8022 * @is_data - Boolean flag indicating whether an extent is
8023 * allocated for data (true) or metadata (false)
8025 * @delalloc - Boolean flag indicating whether this allocation is for
8026 * delalloc or not. If 'true' data_rwsem of block groups
8027 * is going to be acquired.
8030 * Returns 0 when an allocation succeeded or < 0 when an error occurred. In
8031 * case -ENOSPC is returned then @ins->offset will contain the size of the
8032 * largest available hole the allocator managed to find.
8034 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
8035 u64 num_bytes, u64 min_alloc_size,
8036 u64 empty_size, u64 hint_byte,
8037 struct btrfs_key *ins, int is_data, int delalloc)
8039 struct btrfs_fs_info *fs_info = root->fs_info;
8040 bool final_tried = num_bytes == min_alloc_size;
8044 flags = get_alloc_profile_by_root(root, is_data);
8046 WARN_ON(num_bytes < fs_info->sectorsize);
8047 ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size,
8048 hint_byte, ins, flags, delalloc);
8049 if (!ret && !is_data) {
8050 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
8051 } else if (ret == -ENOSPC) {
8052 if (!final_tried && ins->offset) {
8053 num_bytes = min(num_bytes >> 1, ins->offset);
8054 num_bytes = round_down(num_bytes,
8055 fs_info->sectorsize);
8056 num_bytes = max(num_bytes, min_alloc_size);
8057 ram_bytes = num_bytes;
8058 if (num_bytes == min_alloc_size)
8061 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8062 struct btrfs_space_info *sinfo;
8064 sinfo = __find_space_info(fs_info, flags);
8066 "allocation failed flags %llu, wanted %llu",
8069 dump_space_info(fs_info, sinfo, num_bytes, 1);
8076 static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8078 int pin, int delalloc)
8080 struct btrfs_block_group_cache *cache;
8083 cache = btrfs_lookup_block_group(fs_info, start);
8085 btrfs_err(fs_info, "Unable to find block group for %llu",
8091 pin_down_extent(fs_info, cache, start, len, 1);
8093 if (btrfs_test_opt(fs_info, DISCARD))
8094 ret = btrfs_discard_extent(fs_info, start, len, NULL);
8095 btrfs_add_free_space(cache, start, len);
8096 btrfs_free_reserved_bytes(cache, len, delalloc);
8097 trace_btrfs_reserved_extent_free(fs_info, start, len);
8100 btrfs_put_block_group(cache);
8104 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8105 u64 start, u64 len, int delalloc)
8107 return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc);
8110 int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info,
8113 return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0);
8116 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8117 struct btrfs_fs_info *fs_info,
8118 u64 parent, u64 root_objectid,
8119 u64 flags, u64 owner, u64 offset,
8120 struct btrfs_key *ins, int ref_mod)
8123 struct btrfs_extent_item *extent_item;
8124 struct btrfs_extent_inline_ref *iref;
8125 struct btrfs_path *path;
8126 struct extent_buffer *leaf;
8131 type = BTRFS_SHARED_DATA_REF_KEY;
8133 type = BTRFS_EXTENT_DATA_REF_KEY;
8135 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
8137 path = btrfs_alloc_path();
8141 path->leave_spinning = 1;
8142 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8145 btrfs_free_path(path);
8149 leaf = path->nodes[0];
8150 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8151 struct btrfs_extent_item);
8152 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
8153 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8154 btrfs_set_extent_flags(leaf, extent_item,
8155 flags | BTRFS_EXTENT_FLAG_DATA);
8157 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8158 btrfs_set_extent_inline_ref_type(leaf, iref, type);
8160 struct btrfs_shared_data_ref *ref;
8161 ref = (struct btrfs_shared_data_ref *)(iref + 1);
8162 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8163 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
8165 struct btrfs_extent_data_ref *ref;
8166 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
8167 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
8168 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
8169 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
8170 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
8173 btrfs_mark_buffer_dirty(path->nodes[0]);
8174 btrfs_free_path(path);
8176 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
8181 ret = update_block_group(trans, fs_info, ins->objectid, ins->offset, 1);
8182 if (ret) { /* -ENOENT, logic error */
8183 btrfs_err(fs_info, "update block group failed for %llu %llu",
8184 ins->objectid, ins->offset);
8187 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset);
8191 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
8192 struct btrfs_fs_info *fs_info,
8193 u64 parent, u64 root_objectid,
8194 u64 flags, struct btrfs_disk_key *key,
8195 int level, struct btrfs_key *ins)
8198 struct btrfs_extent_item *extent_item;
8199 struct btrfs_tree_block_info *block_info;
8200 struct btrfs_extent_inline_ref *iref;
8201 struct btrfs_path *path;
8202 struct extent_buffer *leaf;
8203 u32 size = sizeof(*extent_item) + sizeof(*iref);
8204 u64 num_bytes = ins->offset;
8205 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8207 if (!skinny_metadata)
8208 size += sizeof(*block_info);
8210 path = btrfs_alloc_path();
8212 btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid,
8217 path->leave_spinning = 1;
8218 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8221 btrfs_free_path(path);
8222 btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid,
8227 leaf = path->nodes[0];
8228 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8229 struct btrfs_extent_item);
8230 btrfs_set_extent_refs(leaf, extent_item, 1);
8231 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8232 btrfs_set_extent_flags(leaf, extent_item,
8233 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
8235 if (skinny_metadata) {
8236 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8237 num_bytes = fs_info->nodesize;
8239 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
8240 btrfs_set_tree_block_key(leaf, block_info, key);
8241 btrfs_set_tree_block_level(leaf, block_info, level);
8242 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
8246 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
8247 btrfs_set_extent_inline_ref_type(leaf, iref,
8248 BTRFS_SHARED_BLOCK_REF_KEY);
8249 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8251 btrfs_set_extent_inline_ref_type(leaf, iref,
8252 BTRFS_TREE_BLOCK_REF_KEY);
8253 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
8256 btrfs_mark_buffer_dirty(leaf);
8257 btrfs_free_path(path);
8259 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
8264 ret = update_block_group(trans, fs_info, ins->objectid,
8265 fs_info->nodesize, 1);
8266 if (ret) { /* -ENOENT, logic error */
8267 btrfs_err(fs_info, "update block group failed for %llu %llu",
8268 ins->objectid, ins->offset);
8272 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid,
8277 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8278 struct btrfs_root *root, u64 owner,
8279 u64 offset, u64 ram_bytes,
8280 struct btrfs_key *ins)
8282 struct btrfs_fs_info *fs_info = root->fs_info;
8285 BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
8287 btrfs_ref_tree_mod(root, ins->objectid, ins->offset, 0,
8288 root->root_key.objectid, owner, offset,
8289 BTRFS_ADD_DELAYED_EXTENT);
8291 ret = btrfs_add_delayed_data_ref(fs_info, trans, ins->objectid,
8293 root->root_key.objectid, owner,
8295 BTRFS_ADD_DELAYED_EXTENT, NULL, NULL);
8300 * this is used by the tree logging recovery code. It records that
8301 * an extent has been allocated and makes sure to clear the free
8302 * space cache bits as well
8304 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
8305 struct btrfs_fs_info *fs_info,
8306 u64 root_objectid, u64 owner, u64 offset,
8307 struct btrfs_key *ins)
8310 struct btrfs_block_group_cache *block_group;
8311 struct btrfs_space_info *space_info;
8314 * Mixed block groups will exclude before processing the log so we only
8315 * need to do the exclude dance if this fs isn't mixed.
8317 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
8318 ret = __exclude_logged_extent(fs_info, ins->objectid,
8324 block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
8328 space_info = block_group->space_info;
8329 spin_lock(&space_info->lock);
8330 spin_lock(&block_group->lock);
8331 space_info->bytes_reserved += ins->offset;
8332 block_group->reserved += ins->offset;
8333 spin_unlock(&block_group->lock);
8334 spin_unlock(&space_info->lock);
8336 ret = alloc_reserved_file_extent(trans, fs_info, 0, root_objectid,
8337 0, owner, offset, ins, 1);
8338 btrfs_put_block_group(block_group);
8342 static struct extent_buffer *
8343 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
8344 u64 bytenr, int level)
8346 struct btrfs_fs_info *fs_info = root->fs_info;
8347 struct extent_buffer *buf;
8349 buf = btrfs_find_create_tree_block(fs_info, bytenr);
8353 btrfs_set_header_generation(buf, trans->transid);
8354 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
8355 btrfs_tree_lock(buf);
8356 clean_tree_block(fs_info, buf);
8357 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
8359 btrfs_set_lock_blocking(buf);
8360 set_extent_buffer_uptodate(buf);
8362 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
8363 buf->log_index = root->log_transid % 2;
8365 * we allow two log transactions at a time, use different
8366 * EXENT bit to differentiate dirty pages.
8368 if (buf->log_index == 0)
8369 set_extent_dirty(&root->dirty_log_pages, buf->start,
8370 buf->start + buf->len - 1, GFP_NOFS);
8372 set_extent_new(&root->dirty_log_pages, buf->start,
8373 buf->start + buf->len - 1);
8375 buf->log_index = -1;
8376 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
8377 buf->start + buf->len - 1, GFP_NOFS);
8379 trans->dirty = true;
8380 /* this returns a buffer locked for blocking */
8384 static struct btrfs_block_rsv *
8385 use_block_rsv(struct btrfs_trans_handle *trans,
8386 struct btrfs_root *root, u32 blocksize)
8388 struct btrfs_fs_info *fs_info = root->fs_info;
8389 struct btrfs_block_rsv *block_rsv;
8390 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
8392 bool global_updated = false;
8394 block_rsv = get_block_rsv(trans, root);
8396 if (unlikely(block_rsv->size == 0))
8399 ret = block_rsv_use_bytes(block_rsv, blocksize);
8403 if (block_rsv->failfast)
8404 return ERR_PTR(ret);
8406 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
8407 global_updated = true;
8408 update_global_block_rsv(fs_info);
8412 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8413 static DEFINE_RATELIMIT_STATE(_rs,
8414 DEFAULT_RATELIMIT_INTERVAL * 10,
8415 /*DEFAULT_RATELIMIT_BURST*/ 1);
8416 if (__ratelimit(&_rs))
8418 "BTRFS: block rsv returned %d\n", ret);
8421 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
8422 BTRFS_RESERVE_NO_FLUSH);
8426 * If we couldn't reserve metadata bytes try and use some from
8427 * the global reserve if its space type is the same as the global
8430 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
8431 block_rsv->space_info == global_rsv->space_info) {
8432 ret = block_rsv_use_bytes(global_rsv, blocksize);
8436 return ERR_PTR(ret);
8439 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
8440 struct btrfs_block_rsv *block_rsv, u32 blocksize)
8442 block_rsv_add_bytes(block_rsv, blocksize, 0);
8443 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0, NULL);
8447 * finds a free extent and does all the dirty work required for allocation
8448 * returns the tree buffer or an ERR_PTR on error.
8450 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
8451 struct btrfs_root *root,
8452 u64 parent, u64 root_objectid,
8453 const struct btrfs_disk_key *key,
8454 int level, u64 hint,
8457 struct btrfs_fs_info *fs_info = root->fs_info;
8458 struct btrfs_key ins;
8459 struct btrfs_block_rsv *block_rsv;
8460 struct extent_buffer *buf;
8461 struct btrfs_delayed_extent_op *extent_op;
8464 u32 blocksize = fs_info->nodesize;
8465 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8467 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8468 if (btrfs_is_testing(fs_info)) {
8469 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
8472 root->alloc_bytenr += blocksize;
8477 block_rsv = use_block_rsv(trans, root, blocksize);
8478 if (IS_ERR(block_rsv))
8479 return ERR_CAST(block_rsv);
8481 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
8482 empty_size, hint, &ins, 0, 0);
8486 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
8489 goto out_free_reserved;
8492 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
8494 parent = ins.objectid;
8495 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
8499 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
8500 extent_op = btrfs_alloc_delayed_extent_op();
8506 memcpy(&extent_op->key, key, sizeof(extent_op->key));
8508 memset(&extent_op->key, 0, sizeof(extent_op->key));
8509 extent_op->flags_to_set = flags;
8510 extent_op->update_key = skinny_metadata ? false : true;
8511 extent_op->update_flags = true;
8512 extent_op->is_data = false;
8513 extent_op->level = level;
8515 btrfs_ref_tree_mod(root, ins.objectid, ins.offset, parent,
8516 root_objectid, level, 0,
8517 BTRFS_ADD_DELAYED_EXTENT);
8518 ret = btrfs_add_delayed_tree_ref(fs_info, trans, ins.objectid,
8520 root_objectid, level,
8521 BTRFS_ADD_DELAYED_EXTENT,
8522 extent_op, NULL, NULL);
8524 goto out_free_delayed;
8529 btrfs_free_delayed_extent_op(extent_op);
8531 free_extent_buffer(buf);
8533 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
8535 unuse_block_rsv(fs_info, block_rsv, blocksize);
8536 return ERR_PTR(ret);
8539 struct walk_control {
8540 u64 refs[BTRFS_MAX_LEVEL];
8541 u64 flags[BTRFS_MAX_LEVEL];
8542 struct btrfs_key update_progress;
8553 #define DROP_REFERENCE 1
8554 #define UPDATE_BACKREF 2
8556 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
8557 struct btrfs_root *root,
8558 struct walk_control *wc,
8559 struct btrfs_path *path)
8561 struct btrfs_fs_info *fs_info = root->fs_info;
8567 struct btrfs_key key;
8568 struct extent_buffer *eb;
8573 if (path->slots[wc->level] < wc->reada_slot) {
8574 wc->reada_count = wc->reada_count * 2 / 3;
8575 wc->reada_count = max(wc->reada_count, 2);
8577 wc->reada_count = wc->reada_count * 3 / 2;
8578 wc->reada_count = min_t(int, wc->reada_count,
8579 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
8582 eb = path->nodes[wc->level];
8583 nritems = btrfs_header_nritems(eb);
8585 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
8586 if (nread >= wc->reada_count)
8590 bytenr = btrfs_node_blockptr(eb, slot);
8591 generation = btrfs_node_ptr_generation(eb, slot);
8593 if (slot == path->slots[wc->level])
8596 if (wc->stage == UPDATE_BACKREF &&
8597 generation <= root->root_key.offset)
8600 /* We don't lock the tree block, it's OK to be racy here */
8601 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
8602 wc->level - 1, 1, &refs,
8604 /* We don't care about errors in readahead. */
8609 if (wc->stage == DROP_REFERENCE) {
8613 if (wc->level == 1 &&
8614 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8616 if (!wc->update_ref ||
8617 generation <= root->root_key.offset)
8619 btrfs_node_key_to_cpu(eb, &key, slot);
8620 ret = btrfs_comp_cpu_keys(&key,
8621 &wc->update_progress);
8625 if (wc->level == 1 &&
8626 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8630 readahead_tree_block(fs_info, bytenr);
8633 wc->reada_slot = slot;
8637 * helper to process tree block while walking down the tree.
8639 * when wc->stage == UPDATE_BACKREF, this function updates
8640 * back refs for pointers in the block.
8642 * NOTE: return value 1 means we should stop walking down.
8644 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8645 struct btrfs_root *root,
8646 struct btrfs_path *path,
8647 struct walk_control *wc, int lookup_info)
8649 struct btrfs_fs_info *fs_info = root->fs_info;
8650 int level = wc->level;
8651 struct extent_buffer *eb = path->nodes[level];
8652 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8655 if (wc->stage == UPDATE_BACKREF &&
8656 btrfs_header_owner(eb) != root->root_key.objectid)
8660 * when reference count of tree block is 1, it won't increase
8661 * again. once full backref flag is set, we never clear it.
8664 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8665 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8666 BUG_ON(!path->locks[level]);
8667 ret = btrfs_lookup_extent_info(trans, fs_info,
8668 eb->start, level, 1,
8671 BUG_ON(ret == -ENOMEM);
8674 BUG_ON(wc->refs[level] == 0);
8677 if (wc->stage == DROP_REFERENCE) {
8678 if (wc->refs[level] > 1)
8681 if (path->locks[level] && !wc->keep_locks) {
8682 btrfs_tree_unlock_rw(eb, path->locks[level]);
8683 path->locks[level] = 0;
8688 /* wc->stage == UPDATE_BACKREF */
8689 if (!(wc->flags[level] & flag)) {
8690 BUG_ON(!path->locks[level]);
8691 ret = btrfs_inc_ref(trans, root, eb, 1);
8692 BUG_ON(ret); /* -ENOMEM */
8693 ret = btrfs_dec_ref(trans, root, eb, 0);
8694 BUG_ON(ret); /* -ENOMEM */
8695 ret = btrfs_set_disk_extent_flags(trans, fs_info, eb->start,
8697 btrfs_header_level(eb), 0);
8698 BUG_ON(ret); /* -ENOMEM */
8699 wc->flags[level] |= flag;
8703 * the block is shared by multiple trees, so it's not good to
8704 * keep the tree lock
8706 if (path->locks[level] && level > 0) {
8707 btrfs_tree_unlock_rw(eb, path->locks[level]);
8708 path->locks[level] = 0;
8714 * helper to process tree block pointer.
8716 * when wc->stage == DROP_REFERENCE, this function checks
8717 * reference count of the block pointed to. if the block
8718 * is shared and we need update back refs for the subtree
8719 * rooted at the block, this function changes wc->stage to
8720 * UPDATE_BACKREF. if the block is shared and there is no
8721 * need to update back, this function drops the reference
8724 * NOTE: return value 1 means we should stop walking down.
8726 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8727 struct btrfs_root *root,
8728 struct btrfs_path *path,
8729 struct walk_control *wc, int *lookup_info)
8731 struct btrfs_fs_info *fs_info = root->fs_info;
8736 struct btrfs_key key;
8737 struct btrfs_key first_key;
8738 struct extent_buffer *next;
8739 int level = wc->level;
8742 bool need_account = false;
8744 generation = btrfs_node_ptr_generation(path->nodes[level],
8745 path->slots[level]);
8747 * if the lower level block was created before the snapshot
8748 * was created, we know there is no need to update back refs
8751 if (wc->stage == UPDATE_BACKREF &&
8752 generation <= root->root_key.offset) {
8757 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8758 btrfs_node_key_to_cpu(path->nodes[level], &first_key,
8759 path->slots[level]);
8760 blocksize = fs_info->nodesize;
8762 next = find_extent_buffer(fs_info, bytenr);
8764 next = btrfs_find_create_tree_block(fs_info, bytenr);
8766 return PTR_ERR(next);
8768 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8772 btrfs_tree_lock(next);
8773 btrfs_set_lock_blocking(next);
8775 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
8776 &wc->refs[level - 1],
8777 &wc->flags[level - 1]);
8781 if (unlikely(wc->refs[level - 1] == 0)) {
8782 btrfs_err(fs_info, "Missing references.");
8788 if (wc->stage == DROP_REFERENCE) {
8789 if (wc->refs[level - 1] > 1) {
8790 need_account = true;
8792 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8795 if (!wc->update_ref ||
8796 generation <= root->root_key.offset)
8799 btrfs_node_key_to_cpu(path->nodes[level], &key,
8800 path->slots[level]);
8801 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8805 wc->stage = UPDATE_BACKREF;
8806 wc->shared_level = level - 1;
8810 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8814 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8815 btrfs_tree_unlock(next);
8816 free_extent_buffer(next);
8822 if (reada && level == 1)
8823 reada_walk_down(trans, root, wc, path);
8824 next = read_tree_block(fs_info, bytenr, generation, level - 1,
8827 return PTR_ERR(next);
8828 } else if (!extent_buffer_uptodate(next)) {
8829 free_extent_buffer(next);
8832 btrfs_tree_lock(next);
8833 btrfs_set_lock_blocking(next);
8837 ASSERT(level == btrfs_header_level(next));
8838 if (level != btrfs_header_level(next)) {
8839 btrfs_err(root->fs_info, "mismatched level");
8843 path->nodes[level] = next;
8844 path->slots[level] = 0;
8845 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8851 wc->refs[level - 1] = 0;
8852 wc->flags[level - 1] = 0;
8853 if (wc->stage == DROP_REFERENCE) {
8854 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8855 parent = path->nodes[level]->start;
8857 ASSERT(root->root_key.objectid ==
8858 btrfs_header_owner(path->nodes[level]));
8859 if (root->root_key.objectid !=
8860 btrfs_header_owner(path->nodes[level])) {
8861 btrfs_err(root->fs_info,
8862 "mismatched block owner");
8870 ret = btrfs_qgroup_trace_subtree(trans, root, next,
8871 generation, level - 1);
8873 btrfs_err_rl(fs_info,
8874 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
8878 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
8879 parent, root->root_key.objectid,
8889 btrfs_tree_unlock(next);
8890 free_extent_buffer(next);
8896 * helper to process tree block while walking up the tree.
8898 * when wc->stage == DROP_REFERENCE, this function drops
8899 * reference count on the block.
8901 * when wc->stage == UPDATE_BACKREF, this function changes
8902 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8903 * to UPDATE_BACKREF previously while processing the block.
8905 * NOTE: return value 1 means we should stop walking up.
8907 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8908 struct btrfs_root *root,
8909 struct btrfs_path *path,
8910 struct walk_control *wc)
8912 struct btrfs_fs_info *fs_info = root->fs_info;
8914 int level = wc->level;
8915 struct extent_buffer *eb = path->nodes[level];
8918 if (wc->stage == UPDATE_BACKREF) {
8919 BUG_ON(wc->shared_level < level);
8920 if (level < wc->shared_level)
8923 ret = find_next_key(path, level + 1, &wc->update_progress);
8927 wc->stage = DROP_REFERENCE;
8928 wc->shared_level = -1;
8929 path->slots[level] = 0;
8932 * check reference count again if the block isn't locked.
8933 * we should start walking down the tree again if reference
8936 if (!path->locks[level]) {
8938 btrfs_tree_lock(eb);
8939 btrfs_set_lock_blocking(eb);
8940 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8942 ret = btrfs_lookup_extent_info(trans, fs_info,
8943 eb->start, level, 1,
8947 btrfs_tree_unlock_rw(eb, path->locks[level]);
8948 path->locks[level] = 0;
8951 BUG_ON(wc->refs[level] == 0);
8952 if (wc->refs[level] == 1) {
8953 btrfs_tree_unlock_rw(eb, path->locks[level]);
8954 path->locks[level] = 0;
8960 /* wc->stage == DROP_REFERENCE */
8961 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8963 if (wc->refs[level] == 1) {
8965 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8966 ret = btrfs_dec_ref(trans, root, eb, 1);
8968 ret = btrfs_dec_ref(trans, root, eb, 0);
8969 BUG_ON(ret); /* -ENOMEM */
8970 ret = btrfs_qgroup_trace_leaf_items(trans, fs_info, eb);
8972 btrfs_err_rl(fs_info,
8973 "error %d accounting leaf items. Quota is out of sync, rescan required.",
8977 /* make block locked assertion in clean_tree_block happy */
8978 if (!path->locks[level] &&
8979 btrfs_header_generation(eb) == trans->transid) {
8980 btrfs_tree_lock(eb);
8981 btrfs_set_lock_blocking(eb);
8982 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8984 clean_tree_block(fs_info, eb);
8987 if (eb == root->node) {
8988 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8991 BUG_ON(root->root_key.objectid !=
8992 btrfs_header_owner(eb));
8994 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8995 parent = path->nodes[level + 1]->start;
8997 BUG_ON(root->root_key.objectid !=
8998 btrfs_header_owner(path->nodes[level + 1]));
9001 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
9003 wc->refs[level] = 0;
9004 wc->flags[level] = 0;
9008 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
9009 struct btrfs_root *root,
9010 struct btrfs_path *path,
9011 struct walk_control *wc)
9013 int level = wc->level;
9014 int lookup_info = 1;
9017 while (level >= 0) {
9018 ret = walk_down_proc(trans, root, path, wc, lookup_info);
9025 if (path->slots[level] >=
9026 btrfs_header_nritems(path->nodes[level]))
9029 ret = do_walk_down(trans, root, path, wc, &lookup_info);
9031 path->slots[level]++;
9040 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
9041 struct btrfs_root *root,
9042 struct btrfs_path *path,
9043 struct walk_control *wc, int max_level)
9045 int level = wc->level;
9048 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
9049 while (level < max_level && path->nodes[level]) {
9051 if (path->slots[level] + 1 <
9052 btrfs_header_nritems(path->nodes[level])) {
9053 path->slots[level]++;
9056 ret = walk_up_proc(trans, root, path, wc);
9060 if (path->locks[level]) {
9061 btrfs_tree_unlock_rw(path->nodes[level],
9062 path->locks[level]);
9063 path->locks[level] = 0;
9065 free_extent_buffer(path->nodes[level]);
9066 path->nodes[level] = NULL;
9074 * drop a subvolume tree.
9076 * this function traverses the tree freeing any blocks that only
9077 * referenced by the tree.
9079 * when a shared tree block is found. this function decreases its
9080 * reference count by one. if update_ref is true, this function
9081 * also make sure backrefs for the shared block and all lower level
9082 * blocks are properly updated.
9084 * If called with for_reloc == 0, may exit early with -EAGAIN
9086 int btrfs_drop_snapshot(struct btrfs_root *root,
9087 struct btrfs_block_rsv *block_rsv, int update_ref,
9090 struct btrfs_fs_info *fs_info = root->fs_info;
9091 struct btrfs_path *path;
9092 struct btrfs_trans_handle *trans;
9093 struct btrfs_root *tree_root = fs_info->tree_root;
9094 struct btrfs_root_item *root_item = &root->root_item;
9095 struct walk_control *wc;
9096 struct btrfs_key key;
9100 bool root_dropped = false;
9102 btrfs_debug(fs_info, "Drop subvolume %llu", root->objectid);
9104 path = btrfs_alloc_path();
9110 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9112 btrfs_free_path(path);
9117 trans = btrfs_start_transaction(tree_root, 0);
9118 if (IS_ERR(trans)) {
9119 err = PTR_ERR(trans);
9124 trans->block_rsv = block_rsv;
9126 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
9127 level = btrfs_header_level(root->node);
9128 path->nodes[level] = btrfs_lock_root_node(root);
9129 btrfs_set_lock_blocking(path->nodes[level]);
9130 path->slots[level] = 0;
9131 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9132 memset(&wc->update_progress, 0,
9133 sizeof(wc->update_progress));
9135 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
9136 memcpy(&wc->update_progress, &key,
9137 sizeof(wc->update_progress));
9139 level = root_item->drop_level;
9141 path->lowest_level = level;
9142 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
9143 path->lowest_level = 0;
9151 * unlock our path, this is safe because only this
9152 * function is allowed to delete this snapshot
9154 btrfs_unlock_up_safe(path, 0);
9156 level = btrfs_header_level(root->node);
9158 btrfs_tree_lock(path->nodes[level]);
9159 btrfs_set_lock_blocking(path->nodes[level]);
9160 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9162 ret = btrfs_lookup_extent_info(trans, fs_info,
9163 path->nodes[level]->start,
9164 level, 1, &wc->refs[level],
9170 BUG_ON(wc->refs[level] == 0);
9172 if (level == root_item->drop_level)
9175 btrfs_tree_unlock(path->nodes[level]);
9176 path->locks[level] = 0;
9177 WARN_ON(wc->refs[level] != 1);
9183 wc->shared_level = -1;
9184 wc->stage = DROP_REFERENCE;
9185 wc->update_ref = update_ref;
9187 wc->for_reloc = for_reloc;
9188 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9192 ret = walk_down_tree(trans, root, path, wc);
9198 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
9205 BUG_ON(wc->stage != DROP_REFERENCE);
9209 if (wc->stage == DROP_REFERENCE) {
9211 btrfs_node_key(path->nodes[level],
9212 &root_item->drop_progress,
9213 path->slots[level]);
9214 root_item->drop_level = level;
9217 BUG_ON(wc->level == 0);
9218 if (btrfs_should_end_transaction(trans) ||
9219 (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
9220 ret = btrfs_update_root(trans, tree_root,
9224 btrfs_abort_transaction(trans, ret);
9229 btrfs_end_transaction_throttle(trans);
9230 if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
9231 btrfs_debug(fs_info,
9232 "drop snapshot early exit");
9237 trans = btrfs_start_transaction(tree_root, 0);
9238 if (IS_ERR(trans)) {
9239 err = PTR_ERR(trans);
9243 trans->block_rsv = block_rsv;
9246 btrfs_release_path(path);
9250 ret = btrfs_del_root(trans, fs_info, &root->root_key);
9252 btrfs_abort_transaction(trans, ret);
9257 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
9258 ret = btrfs_find_root(tree_root, &root->root_key, path,
9261 btrfs_abort_transaction(trans, ret);
9264 } else if (ret > 0) {
9265 /* if we fail to delete the orphan item this time
9266 * around, it'll get picked up the next time.
9268 * The most common failure here is just -ENOENT.
9270 btrfs_del_orphan_item(trans, tree_root,
9271 root->root_key.objectid);
9275 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
9276 btrfs_add_dropped_root(trans, root);
9278 free_extent_buffer(root->node);
9279 free_extent_buffer(root->commit_root);
9280 btrfs_put_fs_root(root);
9282 root_dropped = true;
9284 btrfs_end_transaction_throttle(trans);
9287 btrfs_free_path(path);
9290 * So if we need to stop dropping the snapshot for whatever reason we
9291 * need to make sure to add it back to the dead root list so that we
9292 * keep trying to do the work later. This also cleans up roots if we
9293 * don't have it in the radix (like when we recover after a power fail
9294 * or unmount) so we don't leak memory.
9296 if (!for_reloc && !root_dropped)
9297 btrfs_add_dead_root(root);
9298 if (err && err != -EAGAIN)
9299 btrfs_handle_fs_error(fs_info, err, NULL);
9304 * drop subtree rooted at tree block 'node'.
9306 * NOTE: this function will unlock and release tree block 'node'
9307 * only used by relocation code
9309 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
9310 struct btrfs_root *root,
9311 struct extent_buffer *node,
9312 struct extent_buffer *parent)
9314 struct btrfs_fs_info *fs_info = root->fs_info;
9315 struct btrfs_path *path;
9316 struct walk_control *wc;
9322 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
9324 path = btrfs_alloc_path();
9328 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9330 btrfs_free_path(path);
9334 btrfs_assert_tree_locked(parent);
9335 parent_level = btrfs_header_level(parent);
9336 extent_buffer_get(parent);
9337 path->nodes[parent_level] = parent;
9338 path->slots[parent_level] = btrfs_header_nritems(parent);
9340 btrfs_assert_tree_locked(node);
9341 level = btrfs_header_level(node);
9342 path->nodes[level] = node;
9343 path->slots[level] = 0;
9344 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9346 wc->refs[parent_level] = 1;
9347 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
9349 wc->shared_level = -1;
9350 wc->stage = DROP_REFERENCE;
9354 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9357 wret = walk_down_tree(trans, root, path, wc);
9363 wret = walk_up_tree(trans, root, path, wc, parent_level);
9371 btrfs_free_path(path);
9375 static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags)
9381 * if restripe for this chunk_type is on pick target profile and
9382 * return, otherwise do the usual balance
9384 stripped = get_restripe_target(fs_info, flags);
9386 return extended_to_chunk(stripped);
9388 num_devices = fs_info->fs_devices->rw_devices;
9390 stripped = BTRFS_BLOCK_GROUP_RAID0 |
9391 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
9392 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
9394 if (num_devices == 1) {
9395 stripped |= BTRFS_BLOCK_GROUP_DUP;
9396 stripped = flags & ~stripped;
9398 /* turn raid0 into single device chunks */
9399 if (flags & BTRFS_BLOCK_GROUP_RAID0)
9402 /* turn mirroring into duplication */
9403 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
9404 BTRFS_BLOCK_GROUP_RAID10))
9405 return stripped | BTRFS_BLOCK_GROUP_DUP;
9407 /* they already had raid on here, just return */
9408 if (flags & stripped)
9411 stripped |= BTRFS_BLOCK_GROUP_DUP;
9412 stripped = flags & ~stripped;
9414 /* switch duplicated blocks with raid1 */
9415 if (flags & BTRFS_BLOCK_GROUP_DUP)
9416 return stripped | BTRFS_BLOCK_GROUP_RAID1;
9418 /* this is drive concat, leave it alone */
9424 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
9426 struct btrfs_space_info *sinfo = cache->space_info;
9428 u64 min_allocable_bytes;
9432 * We need some metadata space and system metadata space for
9433 * allocating chunks in some corner cases until we force to set
9434 * it to be readonly.
9437 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
9439 min_allocable_bytes = SZ_1M;
9441 min_allocable_bytes = 0;
9443 spin_lock(&sinfo->lock);
9444 spin_lock(&cache->lock);
9452 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
9453 cache->bytes_super - btrfs_block_group_used(&cache->item);
9455 if (btrfs_space_info_used(sinfo, true) + num_bytes +
9456 min_allocable_bytes <= sinfo->total_bytes) {
9457 sinfo->bytes_readonly += num_bytes;
9459 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
9463 spin_unlock(&cache->lock);
9464 spin_unlock(&sinfo->lock);
9468 int btrfs_inc_block_group_ro(struct btrfs_fs_info *fs_info,
9469 struct btrfs_block_group_cache *cache)
9472 struct btrfs_trans_handle *trans;
9477 trans = btrfs_join_transaction(fs_info->extent_root);
9479 return PTR_ERR(trans);
9482 * we're not allowed to set block groups readonly after the dirty
9483 * block groups cache has started writing. If it already started,
9484 * back off and let this transaction commit
9486 mutex_lock(&fs_info->ro_block_group_mutex);
9487 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9488 u64 transid = trans->transid;
9490 mutex_unlock(&fs_info->ro_block_group_mutex);
9491 btrfs_end_transaction(trans);
9493 ret = btrfs_wait_for_commit(fs_info, transid);
9500 * if we are changing raid levels, try to allocate a corresponding
9501 * block group with the new raid level.
9503 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9504 if (alloc_flags != cache->flags) {
9505 ret = do_chunk_alloc(trans, fs_info, alloc_flags,
9508 * ENOSPC is allowed here, we may have enough space
9509 * already allocated at the new raid level to
9518 ret = inc_block_group_ro(cache, 0);
9521 alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags);
9522 ret = do_chunk_alloc(trans, fs_info, alloc_flags,
9526 ret = inc_block_group_ro(cache, 0);
9528 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9529 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9530 mutex_lock(&fs_info->chunk_mutex);
9531 check_system_chunk(trans, fs_info, alloc_flags);
9532 mutex_unlock(&fs_info->chunk_mutex);
9534 mutex_unlock(&fs_info->ro_block_group_mutex);
9536 btrfs_end_transaction(trans);
9540 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
9541 struct btrfs_fs_info *fs_info, u64 type)
9543 u64 alloc_flags = get_alloc_profile(fs_info, type);
9545 return do_chunk_alloc(trans, fs_info, alloc_flags, CHUNK_ALLOC_FORCE);
9549 * helper to account the unused space of all the readonly block group in the
9550 * space_info. takes mirrors into account.
9552 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9554 struct btrfs_block_group_cache *block_group;
9558 /* It's df, we don't care if it's racy */
9559 if (list_empty(&sinfo->ro_bgs))
9562 spin_lock(&sinfo->lock);
9563 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9564 spin_lock(&block_group->lock);
9566 if (!block_group->ro) {
9567 spin_unlock(&block_group->lock);
9571 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
9572 BTRFS_BLOCK_GROUP_RAID10 |
9573 BTRFS_BLOCK_GROUP_DUP))
9578 free_bytes += (block_group->key.offset -
9579 btrfs_block_group_used(&block_group->item)) *
9582 spin_unlock(&block_group->lock);
9584 spin_unlock(&sinfo->lock);
9589 void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache)
9591 struct btrfs_space_info *sinfo = cache->space_info;
9596 spin_lock(&sinfo->lock);
9597 spin_lock(&cache->lock);
9599 num_bytes = cache->key.offset - cache->reserved -
9600 cache->pinned - cache->bytes_super -
9601 btrfs_block_group_used(&cache->item);
9602 sinfo->bytes_readonly -= num_bytes;
9603 list_del_init(&cache->ro_list);
9605 spin_unlock(&cache->lock);
9606 spin_unlock(&sinfo->lock);
9610 * checks to see if its even possible to relocate this block group.
9612 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9613 * ok to go ahead and try.
9615 int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr)
9617 struct btrfs_root *root = fs_info->extent_root;
9618 struct btrfs_block_group_cache *block_group;
9619 struct btrfs_space_info *space_info;
9620 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
9621 struct btrfs_device *device;
9622 struct btrfs_trans_handle *trans;
9632 debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG);
9634 block_group = btrfs_lookup_block_group(fs_info, bytenr);
9636 /* odd, couldn't find the block group, leave it alone */
9640 "can't find block group for bytenr %llu",
9645 min_free = btrfs_block_group_used(&block_group->item);
9647 /* no bytes used, we're good */
9651 space_info = block_group->space_info;
9652 spin_lock(&space_info->lock);
9654 full = space_info->full;
9657 * if this is the last block group we have in this space, we can't
9658 * relocate it unless we're able to allocate a new chunk below.
9660 * Otherwise, we need to make sure we have room in the space to handle
9661 * all of the extents from this block group. If we can, we're good
9663 if ((space_info->total_bytes != block_group->key.offset) &&
9664 (btrfs_space_info_used(space_info, false) + min_free <
9665 space_info->total_bytes)) {
9666 spin_unlock(&space_info->lock);
9669 spin_unlock(&space_info->lock);
9672 * ok we don't have enough space, but maybe we have free space on our
9673 * devices to allocate new chunks for relocation, so loop through our
9674 * alloc devices and guess if we have enough space. if this block
9675 * group is going to be restriped, run checks against the target
9676 * profile instead of the current one.
9688 target = get_restripe_target(fs_info, block_group->flags);
9690 index = btrfs_bg_flags_to_raid_index(extended_to_chunk(target));
9693 * this is just a balance, so if we were marked as full
9694 * we know there is no space for a new chunk
9699 "no space to alloc new chunk for block group %llu",
9700 block_group->key.objectid);
9704 index = btrfs_bg_flags_to_raid_index(block_group->flags);
9707 if (index == BTRFS_RAID_RAID10) {
9711 } else if (index == BTRFS_RAID_RAID1) {
9713 } else if (index == BTRFS_RAID_DUP) {
9716 } else if (index == BTRFS_RAID_RAID0) {
9717 dev_min = fs_devices->rw_devices;
9718 min_free = div64_u64(min_free, dev_min);
9721 /* We need to do this so that we can look at pending chunks */
9722 trans = btrfs_join_transaction(root);
9723 if (IS_ERR(trans)) {
9724 ret = PTR_ERR(trans);
9728 mutex_lock(&fs_info->chunk_mutex);
9729 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9733 * check to make sure we can actually find a chunk with enough
9734 * space to fit our block group in.
9736 if (device->total_bytes > device->bytes_used + min_free &&
9737 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
9738 ret = find_free_dev_extent(trans, device, min_free,
9743 if (dev_nr >= dev_min)
9749 if (debug && ret == -1)
9751 "no space to allocate a new chunk for block group %llu",
9752 block_group->key.objectid);
9753 mutex_unlock(&fs_info->chunk_mutex);
9754 btrfs_end_transaction(trans);
9756 btrfs_put_block_group(block_group);
9760 static int find_first_block_group(struct btrfs_fs_info *fs_info,
9761 struct btrfs_path *path,
9762 struct btrfs_key *key)
9764 struct btrfs_root *root = fs_info->extent_root;
9766 struct btrfs_key found_key;
9767 struct extent_buffer *leaf;
9770 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9775 slot = path->slots[0];
9776 leaf = path->nodes[0];
9777 if (slot >= btrfs_header_nritems(leaf)) {
9778 ret = btrfs_next_leaf(root, path);
9785 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9787 if (found_key.objectid >= key->objectid &&
9788 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9789 struct extent_map_tree *em_tree;
9790 struct extent_map *em;
9792 em_tree = &root->fs_info->mapping_tree.map_tree;
9793 read_lock(&em_tree->lock);
9794 em = lookup_extent_mapping(em_tree, found_key.objectid,
9796 read_unlock(&em_tree->lock);
9799 "logical %llu len %llu found bg but no related chunk",
9800 found_key.objectid, found_key.offset);
9805 free_extent_map(em);
9814 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9816 struct btrfs_block_group_cache *block_group;
9820 struct inode *inode;
9822 block_group = btrfs_lookup_first_block_group(info, last);
9823 while (block_group) {
9824 spin_lock(&block_group->lock);
9825 if (block_group->iref)
9827 spin_unlock(&block_group->lock);
9828 block_group = next_block_group(info, block_group);
9837 inode = block_group->inode;
9838 block_group->iref = 0;
9839 block_group->inode = NULL;
9840 spin_unlock(&block_group->lock);
9841 ASSERT(block_group->io_ctl.inode == NULL);
9843 last = block_group->key.objectid + block_group->key.offset;
9844 btrfs_put_block_group(block_group);
9849 * Must be called only after stopping all workers, since we could have block
9850 * group caching kthreads running, and therefore they could race with us if we
9851 * freed the block groups before stopping them.
9853 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9855 struct btrfs_block_group_cache *block_group;
9856 struct btrfs_space_info *space_info;
9857 struct btrfs_caching_control *caching_ctl;
9860 down_write(&info->commit_root_sem);
9861 while (!list_empty(&info->caching_block_groups)) {
9862 caching_ctl = list_entry(info->caching_block_groups.next,
9863 struct btrfs_caching_control, list);
9864 list_del(&caching_ctl->list);
9865 put_caching_control(caching_ctl);
9867 up_write(&info->commit_root_sem);
9869 spin_lock(&info->unused_bgs_lock);
9870 while (!list_empty(&info->unused_bgs)) {
9871 block_group = list_first_entry(&info->unused_bgs,
9872 struct btrfs_block_group_cache,
9874 list_del_init(&block_group->bg_list);
9875 btrfs_put_block_group(block_group);
9877 spin_unlock(&info->unused_bgs_lock);
9879 spin_lock(&info->block_group_cache_lock);
9880 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9881 block_group = rb_entry(n, struct btrfs_block_group_cache,
9883 rb_erase(&block_group->cache_node,
9884 &info->block_group_cache_tree);
9885 RB_CLEAR_NODE(&block_group->cache_node);
9886 spin_unlock(&info->block_group_cache_lock);
9888 down_write(&block_group->space_info->groups_sem);
9889 list_del(&block_group->list);
9890 up_write(&block_group->space_info->groups_sem);
9893 * We haven't cached this block group, which means we could
9894 * possibly have excluded extents on this block group.
9896 if (block_group->cached == BTRFS_CACHE_NO ||
9897 block_group->cached == BTRFS_CACHE_ERROR)
9898 free_excluded_extents(info, block_group);
9900 btrfs_remove_free_space_cache(block_group);
9901 ASSERT(block_group->cached != BTRFS_CACHE_STARTED);
9902 ASSERT(list_empty(&block_group->dirty_list));
9903 ASSERT(list_empty(&block_group->io_list));
9904 ASSERT(list_empty(&block_group->bg_list));
9905 ASSERT(atomic_read(&block_group->count) == 1);
9906 btrfs_put_block_group(block_group);
9908 spin_lock(&info->block_group_cache_lock);
9910 spin_unlock(&info->block_group_cache_lock);
9912 /* now that all the block groups are freed, go through and
9913 * free all the space_info structs. This is only called during
9914 * the final stages of unmount, and so we know nobody is
9915 * using them. We call synchronize_rcu() once before we start,
9916 * just to be on the safe side.
9920 release_global_block_rsv(info);
9922 while (!list_empty(&info->space_info)) {
9925 space_info = list_entry(info->space_info.next,
9926 struct btrfs_space_info,
9930 * Do not hide this behind enospc_debug, this is actually
9931 * important and indicates a real bug if this happens.
9933 if (WARN_ON(space_info->bytes_pinned > 0 ||
9934 space_info->bytes_reserved > 0 ||
9935 space_info->bytes_may_use > 0))
9936 dump_space_info(info, space_info, 0, 0);
9937 list_del(&space_info->list);
9938 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9939 struct kobject *kobj;
9940 kobj = space_info->block_group_kobjs[i];
9941 space_info->block_group_kobjs[i] = NULL;
9947 kobject_del(&space_info->kobj);
9948 kobject_put(&space_info->kobj);
9953 /* link_block_group will queue up kobjects to add when we're reclaim-safe */
9954 void btrfs_add_raid_kobjects(struct btrfs_fs_info *fs_info)
9956 struct btrfs_space_info *space_info;
9957 struct raid_kobject *rkobj;
9962 spin_lock(&fs_info->pending_raid_kobjs_lock);
9963 list_splice_init(&fs_info->pending_raid_kobjs, &list);
9964 spin_unlock(&fs_info->pending_raid_kobjs_lock);
9966 list_for_each_entry(rkobj, &list, list) {
9967 space_info = __find_space_info(fs_info, rkobj->flags);
9968 index = btrfs_bg_flags_to_raid_index(rkobj->flags);
9970 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9971 "%s", get_raid_name(index));
9973 kobject_put(&rkobj->kobj);
9979 "failed to add kobject for block cache, ignoring");
9982 static void link_block_group(struct btrfs_block_group_cache *cache)
9984 struct btrfs_space_info *space_info = cache->space_info;
9985 struct btrfs_fs_info *fs_info = cache->fs_info;
9986 int index = btrfs_bg_flags_to_raid_index(cache->flags);
9989 down_write(&space_info->groups_sem);
9990 if (list_empty(&space_info->block_groups[index]))
9992 list_add_tail(&cache->list, &space_info->block_groups[index]);
9993 up_write(&space_info->groups_sem);
9996 struct raid_kobject *rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9998 btrfs_warn(cache->fs_info,
9999 "couldn't alloc memory for raid level kobject");
10002 rkobj->flags = cache->flags;
10003 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
10005 spin_lock(&fs_info->pending_raid_kobjs_lock);
10006 list_add_tail(&rkobj->list, &fs_info->pending_raid_kobjs);
10007 spin_unlock(&fs_info->pending_raid_kobjs_lock);
10008 space_info->block_group_kobjs[index] = &rkobj->kobj;
10012 static struct btrfs_block_group_cache *
10013 btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info,
10014 u64 start, u64 size)
10016 struct btrfs_block_group_cache *cache;
10018 cache = kzalloc(sizeof(*cache), GFP_NOFS);
10022 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
10024 if (!cache->free_space_ctl) {
10029 cache->key.objectid = start;
10030 cache->key.offset = size;
10031 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10033 cache->fs_info = fs_info;
10034 cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start);
10035 set_free_space_tree_thresholds(cache);
10037 atomic_set(&cache->count, 1);
10038 spin_lock_init(&cache->lock);
10039 init_rwsem(&cache->data_rwsem);
10040 INIT_LIST_HEAD(&cache->list);
10041 INIT_LIST_HEAD(&cache->cluster_list);
10042 INIT_LIST_HEAD(&cache->bg_list);
10043 INIT_LIST_HEAD(&cache->ro_list);
10044 INIT_LIST_HEAD(&cache->dirty_list);
10045 INIT_LIST_HEAD(&cache->io_list);
10046 btrfs_init_free_space_ctl(cache);
10047 atomic_set(&cache->trimming, 0);
10048 mutex_init(&cache->free_space_lock);
10049 btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root);
10054 int btrfs_read_block_groups(struct btrfs_fs_info *info)
10056 struct btrfs_path *path;
10058 struct btrfs_block_group_cache *cache;
10059 struct btrfs_space_info *space_info;
10060 struct btrfs_key key;
10061 struct btrfs_key found_key;
10062 struct extent_buffer *leaf;
10063 int need_clear = 0;
10068 feature = btrfs_super_incompat_flags(info->super_copy);
10069 mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS);
10073 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10074 path = btrfs_alloc_path();
10077 path->reada = READA_FORWARD;
10079 cache_gen = btrfs_super_cache_generation(info->super_copy);
10080 if (btrfs_test_opt(info, SPACE_CACHE) &&
10081 btrfs_super_generation(info->super_copy) != cache_gen)
10083 if (btrfs_test_opt(info, CLEAR_CACHE))
10087 ret = find_first_block_group(info, path, &key);
10093 leaf = path->nodes[0];
10094 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
10096 cache = btrfs_create_block_group_cache(info, found_key.objectid,
10105 * When we mount with old space cache, we need to
10106 * set BTRFS_DC_CLEAR and set dirty flag.
10108 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
10109 * truncate the old free space cache inode and
10111 * b) Setting 'dirty flag' makes sure that we flush
10112 * the new space cache info onto disk.
10114 if (btrfs_test_opt(info, SPACE_CACHE))
10115 cache->disk_cache_state = BTRFS_DC_CLEAR;
10118 read_extent_buffer(leaf, &cache->item,
10119 btrfs_item_ptr_offset(leaf, path->slots[0]),
10120 sizeof(cache->item));
10121 cache->flags = btrfs_block_group_flags(&cache->item);
10123 ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) &&
10124 (cache->flags & BTRFS_BLOCK_GROUP_DATA))) {
10126 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
10127 cache->key.objectid);
10132 key.objectid = found_key.objectid + found_key.offset;
10133 btrfs_release_path(path);
10136 * We need to exclude the super stripes now so that the space
10137 * info has super bytes accounted for, otherwise we'll think
10138 * we have more space than we actually do.
10140 ret = exclude_super_stripes(info, cache);
10143 * We may have excluded something, so call this just in
10146 free_excluded_extents(info, cache);
10147 btrfs_put_block_group(cache);
10152 * check for two cases, either we are full, and therefore
10153 * don't need to bother with the caching work since we won't
10154 * find any space, or we are empty, and we can just add all
10155 * the space in and be done with it. This saves us _alot_ of
10156 * time, particularly in the full case.
10158 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
10159 cache->last_byte_to_unpin = (u64)-1;
10160 cache->cached = BTRFS_CACHE_FINISHED;
10161 free_excluded_extents(info, cache);
10162 } else if (btrfs_block_group_used(&cache->item) == 0) {
10163 cache->last_byte_to_unpin = (u64)-1;
10164 cache->cached = BTRFS_CACHE_FINISHED;
10165 add_new_free_space(cache, info,
10166 found_key.objectid,
10167 found_key.objectid +
10169 free_excluded_extents(info, cache);
10172 ret = btrfs_add_block_group_cache(info, cache);
10174 btrfs_remove_free_space_cache(cache);
10175 btrfs_put_block_group(cache);
10179 trace_btrfs_add_block_group(info, cache, 0);
10180 update_space_info(info, cache->flags, found_key.offset,
10181 btrfs_block_group_used(&cache->item),
10182 cache->bytes_super, &space_info);
10184 cache->space_info = space_info;
10186 link_block_group(cache);
10188 set_avail_alloc_bits(info, cache->flags);
10189 if (btrfs_chunk_readonly(info, cache->key.objectid)) {
10190 inc_block_group_ro(cache, 1);
10191 } else if (btrfs_block_group_used(&cache->item) == 0) {
10192 spin_lock(&info->unused_bgs_lock);
10193 /* Should always be true but just in case. */
10194 if (list_empty(&cache->bg_list)) {
10195 btrfs_get_block_group(cache);
10196 list_add_tail(&cache->bg_list,
10197 &info->unused_bgs);
10199 spin_unlock(&info->unused_bgs_lock);
10203 list_for_each_entry_rcu(space_info, &info->space_info, list) {
10204 if (!(get_alloc_profile(info, space_info->flags) &
10205 (BTRFS_BLOCK_GROUP_RAID10 |
10206 BTRFS_BLOCK_GROUP_RAID1 |
10207 BTRFS_BLOCK_GROUP_RAID5 |
10208 BTRFS_BLOCK_GROUP_RAID6 |
10209 BTRFS_BLOCK_GROUP_DUP)))
10212 * avoid allocating from un-mirrored block group if there are
10213 * mirrored block groups.
10215 list_for_each_entry(cache,
10216 &space_info->block_groups[BTRFS_RAID_RAID0],
10218 inc_block_group_ro(cache, 1);
10219 list_for_each_entry(cache,
10220 &space_info->block_groups[BTRFS_RAID_SINGLE],
10222 inc_block_group_ro(cache, 1);
10225 btrfs_add_raid_kobjects(info);
10226 init_global_block_rsv(info);
10229 btrfs_free_path(path);
10233 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans)
10235 struct btrfs_fs_info *fs_info = trans->fs_info;
10236 struct btrfs_block_group_cache *block_group, *tmp;
10237 struct btrfs_root *extent_root = fs_info->extent_root;
10238 struct btrfs_block_group_item item;
10239 struct btrfs_key key;
10241 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
10243 trans->can_flush_pending_bgs = false;
10244 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
10248 spin_lock(&block_group->lock);
10249 memcpy(&item, &block_group->item, sizeof(item));
10250 memcpy(&key, &block_group->key, sizeof(key));
10251 spin_unlock(&block_group->lock);
10253 ret = btrfs_insert_item(trans, extent_root, &key, &item,
10256 btrfs_abort_transaction(trans, ret);
10257 ret = btrfs_finish_chunk_alloc(trans, fs_info, key.objectid,
10260 btrfs_abort_transaction(trans, ret);
10261 add_block_group_free_space(trans, fs_info, block_group);
10262 /* already aborted the transaction if it failed. */
10264 list_del_init(&block_group->bg_list);
10266 trans->can_flush_pending_bgs = can_flush_pending_bgs;
10269 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
10270 struct btrfs_fs_info *fs_info, u64 bytes_used,
10271 u64 type, u64 chunk_offset, u64 size)
10273 struct btrfs_block_group_cache *cache;
10276 btrfs_set_log_full_commit(fs_info, trans);
10278 cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size);
10282 btrfs_set_block_group_used(&cache->item, bytes_used);
10283 btrfs_set_block_group_chunk_objectid(&cache->item,
10284 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
10285 btrfs_set_block_group_flags(&cache->item, type);
10287 cache->flags = type;
10288 cache->last_byte_to_unpin = (u64)-1;
10289 cache->cached = BTRFS_CACHE_FINISHED;
10290 cache->needs_free_space = 1;
10291 ret = exclude_super_stripes(fs_info, cache);
10294 * We may have excluded something, so call this just in
10297 free_excluded_extents(fs_info, cache);
10298 btrfs_put_block_group(cache);
10302 add_new_free_space(cache, fs_info, chunk_offset, chunk_offset + size);
10304 free_excluded_extents(fs_info, cache);
10306 #ifdef CONFIG_BTRFS_DEBUG
10307 if (btrfs_should_fragment_free_space(cache)) {
10308 u64 new_bytes_used = size - bytes_used;
10310 bytes_used += new_bytes_used >> 1;
10311 fragment_free_space(cache);
10315 * Ensure the corresponding space_info object is created and
10316 * assigned to our block group. We want our bg to be added to the rbtree
10317 * with its ->space_info set.
10319 cache->space_info = __find_space_info(fs_info, cache->flags);
10320 ASSERT(cache->space_info);
10322 ret = btrfs_add_block_group_cache(fs_info, cache);
10324 btrfs_remove_free_space_cache(cache);
10325 btrfs_put_block_group(cache);
10330 * Now that our block group has its ->space_info set and is inserted in
10331 * the rbtree, update the space info's counters.
10333 trace_btrfs_add_block_group(fs_info, cache, 1);
10334 update_space_info(fs_info, cache->flags, size, bytes_used,
10335 cache->bytes_super, &cache->space_info);
10336 update_global_block_rsv(fs_info);
10338 link_block_group(cache);
10340 list_add_tail(&cache->bg_list, &trans->new_bgs);
10342 set_avail_alloc_bits(fs_info, type);
10346 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
10348 u64 extra_flags = chunk_to_extended(flags) &
10349 BTRFS_EXTENDED_PROFILE_MASK;
10351 write_seqlock(&fs_info->profiles_lock);
10352 if (flags & BTRFS_BLOCK_GROUP_DATA)
10353 fs_info->avail_data_alloc_bits &= ~extra_flags;
10354 if (flags & BTRFS_BLOCK_GROUP_METADATA)
10355 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
10356 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
10357 fs_info->avail_system_alloc_bits &= ~extra_flags;
10358 write_sequnlock(&fs_info->profiles_lock);
10361 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
10362 struct btrfs_fs_info *fs_info, u64 group_start,
10363 struct extent_map *em)
10365 struct btrfs_root *root = fs_info->extent_root;
10366 struct btrfs_path *path;
10367 struct btrfs_block_group_cache *block_group;
10368 struct btrfs_free_cluster *cluster;
10369 struct btrfs_root *tree_root = fs_info->tree_root;
10370 struct btrfs_key key;
10371 struct inode *inode;
10372 struct kobject *kobj = NULL;
10376 struct btrfs_caching_control *caching_ctl = NULL;
10379 block_group = btrfs_lookup_block_group(fs_info, group_start);
10380 BUG_ON(!block_group);
10381 BUG_ON(!block_group->ro);
10384 * Free the reserved super bytes from this block group before
10387 free_excluded_extents(fs_info, block_group);
10388 btrfs_free_ref_tree_range(fs_info, block_group->key.objectid,
10389 block_group->key.offset);
10391 memcpy(&key, &block_group->key, sizeof(key));
10392 index = btrfs_bg_flags_to_raid_index(block_group->flags);
10393 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
10394 BTRFS_BLOCK_GROUP_RAID1 |
10395 BTRFS_BLOCK_GROUP_RAID10))
10400 /* make sure this block group isn't part of an allocation cluster */
10401 cluster = &fs_info->data_alloc_cluster;
10402 spin_lock(&cluster->refill_lock);
10403 btrfs_return_cluster_to_free_space(block_group, cluster);
10404 spin_unlock(&cluster->refill_lock);
10407 * make sure this block group isn't part of a metadata
10408 * allocation cluster
10410 cluster = &fs_info->meta_alloc_cluster;
10411 spin_lock(&cluster->refill_lock);
10412 btrfs_return_cluster_to_free_space(block_group, cluster);
10413 spin_unlock(&cluster->refill_lock);
10415 path = btrfs_alloc_path();
10422 * get the inode first so any iput calls done for the io_list
10423 * aren't the final iput (no unlinks allowed now)
10425 inode = lookup_free_space_inode(fs_info, block_group, path);
10427 mutex_lock(&trans->transaction->cache_write_mutex);
10429 * make sure our free spache cache IO is done before remove the
10432 spin_lock(&trans->transaction->dirty_bgs_lock);
10433 if (!list_empty(&block_group->io_list)) {
10434 list_del_init(&block_group->io_list);
10436 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
10438 spin_unlock(&trans->transaction->dirty_bgs_lock);
10439 btrfs_wait_cache_io(trans, block_group, path);
10440 btrfs_put_block_group(block_group);
10441 spin_lock(&trans->transaction->dirty_bgs_lock);
10444 if (!list_empty(&block_group->dirty_list)) {
10445 list_del_init(&block_group->dirty_list);
10446 btrfs_put_block_group(block_group);
10448 spin_unlock(&trans->transaction->dirty_bgs_lock);
10449 mutex_unlock(&trans->transaction->cache_write_mutex);
10451 if (!IS_ERR(inode)) {
10452 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10454 btrfs_add_delayed_iput(inode);
10457 clear_nlink(inode);
10458 /* One for the block groups ref */
10459 spin_lock(&block_group->lock);
10460 if (block_group->iref) {
10461 block_group->iref = 0;
10462 block_group->inode = NULL;
10463 spin_unlock(&block_group->lock);
10466 spin_unlock(&block_group->lock);
10468 /* One for our lookup ref */
10469 btrfs_add_delayed_iput(inode);
10472 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
10473 key.offset = block_group->key.objectid;
10476 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
10480 btrfs_release_path(path);
10482 ret = btrfs_del_item(trans, tree_root, path);
10485 btrfs_release_path(path);
10488 spin_lock(&fs_info->block_group_cache_lock);
10489 rb_erase(&block_group->cache_node,
10490 &fs_info->block_group_cache_tree);
10491 RB_CLEAR_NODE(&block_group->cache_node);
10493 if (fs_info->first_logical_byte == block_group->key.objectid)
10494 fs_info->first_logical_byte = (u64)-1;
10495 spin_unlock(&fs_info->block_group_cache_lock);
10497 down_write(&block_group->space_info->groups_sem);
10499 * we must use list_del_init so people can check to see if they
10500 * are still on the list after taking the semaphore
10502 list_del_init(&block_group->list);
10503 if (list_empty(&block_group->space_info->block_groups[index])) {
10504 kobj = block_group->space_info->block_group_kobjs[index];
10505 block_group->space_info->block_group_kobjs[index] = NULL;
10506 clear_avail_alloc_bits(fs_info, block_group->flags);
10508 up_write(&block_group->space_info->groups_sem);
10514 if (block_group->has_caching_ctl)
10515 caching_ctl = get_caching_control(block_group);
10516 if (block_group->cached == BTRFS_CACHE_STARTED)
10517 wait_block_group_cache_done(block_group);
10518 if (block_group->has_caching_ctl) {
10519 down_write(&fs_info->commit_root_sem);
10520 if (!caching_ctl) {
10521 struct btrfs_caching_control *ctl;
10523 list_for_each_entry(ctl,
10524 &fs_info->caching_block_groups, list)
10525 if (ctl->block_group == block_group) {
10527 refcount_inc(&caching_ctl->count);
10532 list_del_init(&caching_ctl->list);
10533 up_write(&fs_info->commit_root_sem);
10535 /* Once for the caching bgs list and once for us. */
10536 put_caching_control(caching_ctl);
10537 put_caching_control(caching_ctl);
10541 spin_lock(&trans->transaction->dirty_bgs_lock);
10542 if (!list_empty(&block_group->dirty_list)) {
10545 if (!list_empty(&block_group->io_list)) {
10548 spin_unlock(&trans->transaction->dirty_bgs_lock);
10549 btrfs_remove_free_space_cache(block_group);
10551 spin_lock(&block_group->space_info->lock);
10552 list_del_init(&block_group->ro_list);
10554 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
10555 WARN_ON(block_group->space_info->total_bytes
10556 < block_group->key.offset);
10557 WARN_ON(block_group->space_info->bytes_readonly
10558 < block_group->key.offset);
10559 WARN_ON(block_group->space_info->disk_total
10560 < block_group->key.offset * factor);
10562 block_group->space_info->total_bytes -= block_group->key.offset;
10563 block_group->space_info->bytes_readonly -= block_group->key.offset;
10564 block_group->space_info->disk_total -= block_group->key.offset * factor;
10566 spin_unlock(&block_group->space_info->lock);
10568 memcpy(&key, &block_group->key, sizeof(key));
10570 mutex_lock(&fs_info->chunk_mutex);
10571 if (!list_empty(&em->list)) {
10572 /* We're in the transaction->pending_chunks list. */
10573 free_extent_map(em);
10575 spin_lock(&block_group->lock);
10576 block_group->removed = 1;
10578 * At this point trimming can't start on this block group, because we
10579 * removed the block group from the tree fs_info->block_group_cache_tree
10580 * so no one can't find it anymore and even if someone already got this
10581 * block group before we removed it from the rbtree, they have already
10582 * incremented block_group->trimming - if they didn't, they won't find
10583 * any free space entries because we already removed them all when we
10584 * called btrfs_remove_free_space_cache().
10586 * And we must not remove the extent map from the fs_info->mapping_tree
10587 * to prevent the same logical address range and physical device space
10588 * ranges from being reused for a new block group. This is because our
10589 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10590 * completely transactionless, so while it is trimming a range the
10591 * currently running transaction might finish and a new one start,
10592 * allowing for new block groups to be created that can reuse the same
10593 * physical device locations unless we take this special care.
10595 * There may also be an implicit trim operation if the file system
10596 * is mounted with -odiscard. The same protections must remain
10597 * in place until the extents have been discarded completely when
10598 * the transaction commit has completed.
10600 remove_em = (atomic_read(&block_group->trimming) == 0);
10602 * Make sure a trimmer task always sees the em in the pinned_chunks list
10603 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10604 * before checking block_group->removed).
10608 * Our em might be in trans->transaction->pending_chunks which
10609 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10610 * and so is the fs_info->pinned_chunks list.
10612 * So at this point we must be holding the chunk_mutex to avoid
10613 * any races with chunk allocation (more specifically at
10614 * volumes.c:contains_pending_extent()), to ensure it always
10615 * sees the em, either in the pending_chunks list or in the
10616 * pinned_chunks list.
10618 list_move_tail(&em->list, &fs_info->pinned_chunks);
10620 spin_unlock(&block_group->lock);
10623 struct extent_map_tree *em_tree;
10625 em_tree = &fs_info->mapping_tree.map_tree;
10626 write_lock(&em_tree->lock);
10628 * The em might be in the pending_chunks list, so make sure the
10629 * chunk mutex is locked, since remove_extent_mapping() will
10630 * delete us from that list.
10632 remove_extent_mapping(em_tree, em);
10633 write_unlock(&em_tree->lock);
10634 /* once for the tree */
10635 free_extent_map(em);
10638 mutex_unlock(&fs_info->chunk_mutex);
10640 ret = remove_block_group_free_space(trans, fs_info, block_group);
10644 btrfs_put_block_group(block_group);
10645 btrfs_put_block_group(block_group);
10647 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10653 ret = btrfs_del_item(trans, root, path);
10655 btrfs_free_path(path);
10659 struct btrfs_trans_handle *
10660 btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info,
10661 const u64 chunk_offset)
10663 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
10664 struct extent_map *em;
10665 struct map_lookup *map;
10666 unsigned int num_items;
10668 read_lock(&em_tree->lock);
10669 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
10670 read_unlock(&em_tree->lock);
10671 ASSERT(em && em->start == chunk_offset);
10674 * We need to reserve 3 + N units from the metadata space info in order
10675 * to remove a block group (done at btrfs_remove_chunk() and at
10676 * btrfs_remove_block_group()), which are used for:
10678 * 1 unit for adding the free space inode's orphan (located in the tree
10680 * 1 unit for deleting the block group item (located in the extent
10682 * 1 unit for deleting the free space item (located in tree of tree
10684 * N units for deleting N device extent items corresponding to each
10685 * stripe (located in the device tree).
10687 * In order to remove a block group we also need to reserve units in the
10688 * system space info in order to update the chunk tree (update one or
10689 * more device items and remove one chunk item), but this is done at
10690 * btrfs_remove_chunk() through a call to check_system_chunk().
10692 map = em->map_lookup;
10693 num_items = 3 + map->num_stripes;
10694 free_extent_map(em);
10696 return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root,
10701 * Process the unused_bgs list and remove any that don't have any allocated
10702 * space inside of them.
10704 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10706 struct btrfs_block_group_cache *block_group;
10707 struct btrfs_space_info *space_info;
10708 struct btrfs_trans_handle *trans;
10711 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
10714 spin_lock(&fs_info->unused_bgs_lock);
10715 while (!list_empty(&fs_info->unused_bgs)) {
10719 block_group = list_first_entry(&fs_info->unused_bgs,
10720 struct btrfs_block_group_cache,
10722 list_del_init(&block_group->bg_list);
10724 space_info = block_group->space_info;
10726 if (ret || btrfs_mixed_space_info(space_info)) {
10727 btrfs_put_block_group(block_group);
10730 spin_unlock(&fs_info->unused_bgs_lock);
10732 mutex_lock(&fs_info->delete_unused_bgs_mutex);
10734 /* Don't want to race with allocators so take the groups_sem */
10735 down_write(&space_info->groups_sem);
10736 spin_lock(&block_group->lock);
10737 if (block_group->reserved ||
10738 btrfs_block_group_used(&block_group->item) ||
10740 list_is_singular(&block_group->list)) {
10742 * We want to bail if we made new allocations or have
10743 * outstanding allocations in this block group. We do
10744 * the ro check in case balance is currently acting on
10745 * this block group.
10747 spin_unlock(&block_group->lock);
10748 up_write(&space_info->groups_sem);
10751 spin_unlock(&block_group->lock);
10753 /* We don't want to force the issue, only flip if it's ok. */
10754 ret = inc_block_group_ro(block_group, 0);
10755 up_write(&space_info->groups_sem);
10762 * Want to do this before we do anything else so we can recover
10763 * properly if we fail to join the transaction.
10765 trans = btrfs_start_trans_remove_block_group(fs_info,
10766 block_group->key.objectid);
10767 if (IS_ERR(trans)) {
10768 btrfs_dec_block_group_ro(block_group);
10769 ret = PTR_ERR(trans);
10774 * We could have pending pinned extents for this block group,
10775 * just delete them, we don't care about them anymore.
10777 start = block_group->key.objectid;
10778 end = start + block_group->key.offset - 1;
10780 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10781 * btrfs_finish_extent_commit(). If we are at transaction N,
10782 * another task might be running finish_extent_commit() for the
10783 * previous transaction N - 1, and have seen a range belonging
10784 * to the block group in freed_extents[] before we were able to
10785 * clear the whole block group range from freed_extents[]. This
10786 * means that task can lookup for the block group after we
10787 * unpinned it from freed_extents[] and removed it, leading to
10788 * a BUG_ON() at btrfs_unpin_extent_range().
10790 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10791 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10794 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10795 btrfs_dec_block_group_ro(block_group);
10798 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10801 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10802 btrfs_dec_block_group_ro(block_group);
10805 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10807 /* Reset pinned so btrfs_put_block_group doesn't complain */
10808 spin_lock(&space_info->lock);
10809 spin_lock(&block_group->lock);
10811 space_info->bytes_pinned -= block_group->pinned;
10812 space_info->bytes_readonly += block_group->pinned;
10813 percpu_counter_add(&space_info->total_bytes_pinned,
10814 -block_group->pinned);
10815 block_group->pinned = 0;
10817 spin_unlock(&block_group->lock);
10818 spin_unlock(&space_info->lock);
10820 /* DISCARD can flip during remount */
10821 trimming = btrfs_test_opt(fs_info, DISCARD);
10823 /* Implicit trim during transaction commit. */
10825 btrfs_get_block_group_trimming(block_group);
10828 * Btrfs_remove_chunk will abort the transaction if things go
10831 ret = btrfs_remove_chunk(trans, fs_info,
10832 block_group->key.objectid);
10836 btrfs_put_block_group_trimming(block_group);
10841 * If we're not mounted with -odiscard, we can just forget
10842 * about this block group. Otherwise we'll need to wait
10843 * until transaction commit to do the actual discard.
10846 spin_lock(&fs_info->unused_bgs_lock);
10848 * A concurrent scrub might have added us to the list
10849 * fs_info->unused_bgs, so use a list_move operation
10850 * to add the block group to the deleted_bgs list.
10852 list_move(&block_group->bg_list,
10853 &trans->transaction->deleted_bgs);
10854 spin_unlock(&fs_info->unused_bgs_lock);
10855 btrfs_get_block_group(block_group);
10858 btrfs_end_transaction(trans);
10860 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
10861 btrfs_put_block_group(block_group);
10862 spin_lock(&fs_info->unused_bgs_lock);
10864 spin_unlock(&fs_info->unused_bgs_lock);
10867 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10869 struct btrfs_space_info *space_info;
10870 struct btrfs_super_block *disk_super;
10876 disk_super = fs_info->super_copy;
10877 if (!btrfs_super_root(disk_super))
10880 features = btrfs_super_incompat_flags(disk_super);
10881 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10884 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10885 ret = create_space_info(fs_info, flags, &space_info);
10890 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10891 ret = create_space_info(fs_info, flags, &space_info);
10893 flags = BTRFS_BLOCK_GROUP_METADATA;
10894 ret = create_space_info(fs_info, flags, &space_info);
10898 flags = BTRFS_BLOCK_GROUP_DATA;
10899 ret = create_space_info(fs_info, flags, &space_info);
10905 int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
10906 u64 start, u64 end)
10908 return unpin_extent_range(fs_info, start, end, false);
10912 * It used to be that old block groups would be left around forever.
10913 * Iterating over them would be enough to trim unused space. Since we
10914 * now automatically remove them, we also need to iterate over unallocated
10917 * We don't want a transaction for this since the discard may take a
10918 * substantial amount of time. We don't require that a transaction be
10919 * running, but we do need to take a running transaction into account
10920 * to ensure that we're not discarding chunks that were released in
10921 * the current transaction.
10923 * Holding the chunks lock will prevent other threads from allocating
10924 * or releasing chunks, but it won't prevent a running transaction
10925 * from committing and releasing the memory that the pending chunks
10926 * list head uses. For that, we need to take a reference to the
10929 static int btrfs_trim_free_extents(struct btrfs_device *device,
10930 u64 minlen, u64 *trimmed)
10932 u64 start = 0, len = 0;
10937 /* Not writeable = nothing to do. */
10938 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
10941 /* No free space = nothing to do. */
10942 if (device->total_bytes <= device->bytes_used)
10948 struct btrfs_fs_info *fs_info = device->fs_info;
10949 struct btrfs_transaction *trans;
10952 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
10956 down_read(&fs_info->commit_root_sem);
10958 spin_lock(&fs_info->trans_lock);
10959 trans = fs_info->running_transaction;
10961 refcount_inc(&trans->use_count);
10962 spin_unlock(&fs_info->trans_lock);
10964 ret = find_free_dev_extent_start(trans, device, minlen, start,
10967 btrfs_put_transaction(trans);
10970 up_read(&fs_info->commit_root_sem);
10971 mutex_unlock(&fs_info->chunk_mutex);
10972 if (ret == -ENOSPC)
10977 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
10978 up_read(&fs_info->commit_root_sem);
10979 mutex_unlock(&fs_info->chunk_mutex);
10987 if (fatal_signal_pending(current)) {
10988 ret = -ERESTARTSYS;
10998 int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
11000 struct btrfs_block_group_cache *cache = NULL;
11001 struct btrfs_device *device;
11002 struct list_head *devices;
11007 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
11011 * try to trim all FS space, our block group may start from non-zero.
11013 if (range->len == total_bytes)
11014 cache = btrfs_lookup_first_block_group(fs_info, range->start);
11016 cache = btrfs_lookup_block_group(fs_info, range->start);
11019 if (cache->key.objectid >= (range->start + range->len)) {
11020 btrfs_put_block_group(cache);
11024 start = max(range->start, cache->key.objectid);
11025 end = min(range->start + range->len,
11026 cache->key.objectid + cache->key.offset);
11028 if (end - start >= range->minlen) {
11029 if (!block_group_cache_done(cache)) {
11030 ret = cache_block_group(cache, 0);
11032 btrfs_put_block_group(cache);
11035 ret = wait_block_group_cache_done(cache);
11037 btrfs_put_block_group(cache);
11041 ret = btrfs_trim_block_group(cache,
11047 trimmed += group_trimmed;
11049 btrfs_put_block_group(cache);
11054 cache = next_block_group(fs_info, cache);
11057 mutex_lock(&fs_info->fs_devices->device_list_mutex);
11058 devices = &fs_info->fs_devices->alloc_list;
11059 list_for_each_entry(device, devices, dev_alloc_list) {
11060 ret = btrfs_trim_free_extents(device, range->minlen,
11065 trimmed += group_trimmed;
11067 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
11069 range->len = trimmed;
11074 * btrfs_{start,end}_write_no_snapshotting() are similar to
11075 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
11076 * data into the page cache through nocow before the subvolume is snapshoted,
11077 * but flush the data into disk after the snapshot creation, or to prevent
11078 * operations while snapshotting is ongoing and that cause the snapshot to be
11079 * inconsistent (writes followed by expanding truncates for example).
11081 void btrfs_end_write_no_snapshotting(struct btrfs_root *root)
11083 percpu_counter_dec(&root->subv_writers->counter);
11085 * Make sure counter is updated before we wake up waiters.
11088 if (waitqueue_active(&root->subv_writers->wait))
11089 wake_up(&root->subv_writers->wait);
11092 int btrfs_start_write_no_snapshotting(struct btrfs_root *root)
11094 if (atomic_read(&root->will_be_snapshotted))
11097 percpu_counter_inc(&root->subv_writers->counter);
11099 * Make sure counter is updated before we check for snapshot creation.
11102 if (atomic_read(&root->will_be_snapshotted)) {
11103 btrfs_end_write_no_snapshotting(root);
11109 void btrfs_wait_for_snapshot_creation(struct btrfs_root *root)
11114 ret = btrfs_start_write_no_snapshotting(root);
11117 wait_var_event(&root->will_be_snapshotted,
11118 !atomic_read(&root->will_be_snapshotted));