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_delayed_ref_node *node,
70 struct btrfs_delayed_extent_op *extent_op);
71 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
72 struct btrfs_fs_info *fs_info, u64 flags,
74 static int find_next_key(struct btrfs_path *path, int level,
75 struct btrfs_key *key);
76 static void dump_space_info(struct btrfs_fs_info *fs_info,
77 struct btrfs_space_info *info, u64 bytes,
78 int dump_block_groups);
79 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
81 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
82 struct btrfs_space_info *space_info,
84 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
85 struct btrfs_space_info *space_info,
89 block_group_cache_done(struct btrfs_block_group_cache *cache)
92 return cache->cached == BTRFS_CACHE_FINISHED ||
93 cache->cached == BTRFS_CACHE_ERROR;
96 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
98 return (cache->flags & bits) == bits;
101 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
103 atomic_inc(&cache->count);
106 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
108 if (atomic_dec_and_test(&cache->count)) {
109 WARN_ON(cache->pinned > 0);
110 WARN_ON(cache->reserved > 0);
113 * If not empty, someone is still holding mutex of
114 * full_stripe_lock, which can only be released by caller.
115 * And it will definitely cause use-after-free when caller
116 * tries to release full stripe lock.
118 * No better way to resolve, but only to warn.
120 WARN_ON(!RB_EMPTY_ROOT(&cache->full_stripe_locks_root.root));
121 kfree(cache->free_space_ctl);
127 * this adds the block group to the fs_info rb tree for the block group
130 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
131 struct btrfs_block_group_cache *block_group)
134 struct rb_node *parent = NULL;
135 struct btrfs_block_group_cache *cache;
137 spin_lock(&info->block_group_cache_lock);
138 p = &info->block_group_cache_tree.rb_node;
142 cache = rb_entry(parent, struct btrfs_block_group_cache,
144 if (block_group->key.objectid < cache->key.objectid) {
146 } else if (block_group->key.objectid > cache->key.objectid) {
149 spin_unlock(&info->block_group_cache_lock);
154 rb_link_node(&block_group->cache_node, parent, p);
155 rb_insert_color(&block_group->cache_node,
156 &info->block_group_cache_tree);
158 if (info->first_logical_byte > block_group->key.objectid)
159 info->first_logical_byte = block_group->key.objectid;
161 spin_unlock(&info->block_group_cache_lock);
167 * This will return the block group at or after bytenr if contains is 0, else
168 * it will return the block group that contains the bytenr
170 static struct btrfs_block_group_cache *
171 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
174 struct btrfs_block_group_cache *cache, *ret = NULL;
178 spin_lock(&info->block_group_cache_lock);
179 n = info->block_group_cache_tree.rb_node;
182 cache = rb_entry(n, struct btrfs_block_group_cache,
184 end = cache->key.objectid + cache->key.offset - 1;
185 start = cache->key.objectid;
187 if (bytenr < start) {
188 if (!contains && (!ret || start < ret->key.objectid))
191 } else if (bytenr > start) {
192 if (contains && bytenr <= end) {
203 btrfs_get_block_group(ret);
204 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
205 info->first_logical_byte = ret->key.objectid;
207 spin_unlock(&info->block_group_cache_lock);
212 static int add_excluded_extent(struct btrfs_fs_info *fs_info,
213 u64 start, u64 num_bytes)
215 u64 end = start + num_bytes - 1;
216 set_extent_bits(&fs_info->freed_extents[0],
217 start, end, EXTENT_UPTODATE);
218 set_extent_bits(&fs_info->freed_extents[1],
219 start, end, EXTENT_UPTODATE);
223 static void free_excluded_extents(struct btrfs_fs_info *fs_info,
224 struct btrfs_block_group_cache *cache)
228 start = cache->key.objectid;
229 end = start + cache->key.offset - 1;
231 clear_extent_bits(&fs_info->freed_extents[0],
232 start, end, EXTENT_UPTODATE);
233 clear_extent_bits(&fs_info->freed_extents[1],
234 start, end, EXTENT_UPTODATE);
237 static int exclude_super_stripes(struct btrfs_fs_info *fs_info,
238 struct btrfs_block_group_cache *cache)
245 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
246 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
247 cache->bytes_super += stripe_len;
248 ret = add_excluded_extent(fs_info, cache->key.objectid,
254 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
255 bytenr = btrfs_sb_offset(i);
256 ret = btrfs_rmap_block(fs_info, cache->key.objectid,
257 bytenr, &logical, &nr, &stripe_len);
264 if (logical[nr] > cache->key.objectid +
268 if (logical[nr] + stripe_len <= cache->key.objectid)
272 if (start < cache->key.objectid) {
273 start = cache->key.objectid;
274 len = (logical[nr] + stripe_len) - start;
276 len = min_t(u64, stripe_len,
277 cache->key.objectid +
278 cache->key.offset - start);
281 cache->bytes_super += len;
282 ret = add_excluded_extent(fs_info, start, len);
294 static struct btrfs_caching_control *
295 get_caching_control(struct btrfs_block_group_cache *cache)
297 struct btrfs_caching_control *ctl;
299 spin_lock(&cache->lock);
300 if (!cache->caching_ctl) {
301 spin_unlock(&cache->lock);
305 ctl = cache->caching_ctl;
306 refcount_inc(&ctl->count);
307 spin_unlock(&cache->lock);
311 static void put_caching_control(struct btrfs_caching_control *ctl)
313 if (refcount_dec_and_test(&ctl->count))
317 #ifdef CONFIG_BTRFS_DEBUG
318 static void fragment_free_space(struct btrfs_block_group_cache *block_group)
320 struct btrfs_fs_info *fs_info = block_group->fs_info;
321 u64 start = block_group->key.objectid;
322 u64 len = block_group->key.offset;
323 u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ?
324 fs_info->nodesize : fs_info->sectorsize;
325 u64 step = chunk << 1;
327 while (len > chunk) {
328 btrfs_remove_free_space(block_group, start, chunk);
339 * this is only called by cache_block_group, since we could have freed extents
340 * we need to check the pinned_extents for any extents that can't be used yet
341 * since their free space will be released as soon as the transaction commits.
343 u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
346 struct btrfs_fs_info *info = block_group->fs_info;
347 u64 extent_start, extent_end, size, total_added = 0;
350 while (start < end) {
351 ret = find_first_extent_bit(info->pinned_extents, start,
352 &extent_start, &extent_end,
353 EXTENT_DIRTY | EXTENT_UPTODATE,
358 if (extent_start <= start) {
359 start = extent_end + 1;
360 } else if (extent_start > start && extent_start < end) {
361 size = extent_start - start;
363 ret = btrfs_add_free_space(block_group, start,
365 BUG_ON(ret); /* -ENOMEM or logic error */
366 start = extent_end + 1;
375 ret = btrfs_add_free_space(block_group, start, size);
376 BUG_ON(ret); /* -ENOMEM or logic error */
382 static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl)
384 struct btrfs_block_group_cache *block_group = caching_ctl->block_group;
385 struct btrfs_fs_info *fs_info = block_group->fs_info;
386 struct btrfs_root *extent_root = fs_info->extent_root;
387 struct btrfs_path *path;
388 struct extent_buffer *leaf;
389 struct btrfs_key key;
396 path = btrfs_alloc_path();
400 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
402 #ifdef CONFIG_BTRFS_DEBUG
404 * If we're fragmenting we don't want to make anybody think we can
405 * allocate from this block group until we've had a chance to fragment
408 if (btrfs_should_fragment_free_space(block_group))
412 * We don't want to deadlock with somebody trying to allocate a new
413 * extent for the extent root while also trying to search the extent
414 * root to add free space. So we skip locking and search the commit
415 * root, since its read-only
417 path->skip_locking = 1;
418 path->search_commit_root = 1;
419 path->reada = READA_FORWARD;
423 key.type = BTRFS_EXTENT_ITEM_KEY;
426 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
430 leaf = path->nodes[0];
431 nritems = btrfs_header_nritems(leaf);
434 if (btrfs_fs_closing(fs_info) > 1) {
439 if (path->slots[0] < nritems) {
440 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
442 ret = find_next_key(path, 0, &key);
446 if (need_resched() ||
447 rwsem_is_contended(&fs_info->commit_root_sem)) {
449 caching_ctl->progress = last;
450 btrfs_release_path(path);
451 up_read(&fs_info->commit_root_sem);
452 mutex_unlock(&caching_ctl->mutex);
454 mutex_lock(&caching_ctl->mutex);
455 down_read(&fs_info->commit_root_sem);
459 ret = btrfs_next_leaf(extent_root, path);
464 leaf = path->nodes[0];
465 nritems = btrfs_header_nritems(leaf);
469 if (key.objectid < last) {
472 key.type = BTRFS_EXTENT_ITEM_KEY;
475 caching_ctl->progress = last;
476 btrfs_release_path(path);
480 if (key.objectid < block_group->key.objectid) {
485 if (key.objectid >= block_group->key.objectid +
486 block_group->key.offset)
489 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
490 key.type == BTRFS_METADATA_ITEM_KEY) {
491 total_found += add_new_free_space(block_group, last,
493 if (key.type == BTRFS_METADATA_ITEM_KEY)
494 last = key.objectid +
497 last = key.objectid + key.offset;
499 if (total_found > CACHING_CTL_WAKE_UP) {
502 wake_up(&caching_ctl->wait);
509 total_found += add_new_free_space(block_group, last,
510 block_group->key.objectid +
511 block_group->key.offset);
512 caching_ctl->progress = (u64)-1;
515 btrfs_free_path(path);
519 static noinline void caching_thread(struct btrfs_work *work)
521 struct btrfs_block_group_cache *block_group;
522 struct btrfs_fs_info *fs_info;
523 struct btrfs_caching_control *caching_ctl;
526 caching_ctl = container_of(work, struct btrfs_caching_control, work);
527 block_group = caching_ctl->block_group;
528 fs_info = block_group->fs_info;
530 mutex_lock(&caching_ctl->mutex);
531 down_read(&fs_info->commit_root_sem);
533 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
534 ret = load_free_space_tree(caching_ctl);
536 ret = load_extent_tree_free(caching_ctl);
538 spin_lock(&block_group->lock);
539 block_group->caching_ctl = NULL;
540 block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED;
541 spin_unlock(&block_group->lock);
543 #ifdef CONFIG_BTRFS_DEBUG
544 if (btrfs_should_fragment_free_space(block_group)) {
547 spin_lock(&block_group->space_info->lock);
548 spin_lock(&block_group->lock);
549 bytes_used = block_group->key.offset -
550 btrfs_block_group_used(&block_group->item);
551 block_group->space_info->bytes_used += bytes_used >> 1;
552 spin_unlock(&block_group->lock);
553 spin_unlock(&block_group->space_info->lock);
554 fragment_free_space(block_group);
558 caching_ctl->progress = (u64)-1;
560 up_read(&fs_info->commit_root_sem);
561 free_excluded_extents(fs_info, block_group);
562 mutex_unlock(&caching_ctl->mutex);
564 wake_up(&caching_ctl->wait);
566 put_caching_control(caching_ctl);
567 btrfs_put_block_group(block_group);
570 static int cache_block_group(struct btrfs_block_group_cache *cache,
574 struct btrfs_fs_info *fs_info = cache->fs_info;
575 struct btrfs_caching_control *caching_ctl;
578 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
582 INIT_LIST_HEAD(&caching_ctl->list);
583 mutex_init(&caching_ctl->mutex);
584 init_waitqueue_head(&caching_ctl->wait);
585 caching_ctl->block_group = cache;
586 caching_ctl->progress = cache->key.objectid;
587 refcount_set(&caching_ctl->count, 1);
588 btrfs_init_work(&caching_ctl->work, btrfs_cache_helper,
589 caching_thread, NULL, NULL);
591 spin_lock(&cache->lock);
593 * This should be a rare occasion, but this could happen I think in the
594 * case where one thread starts to load the space cache info, and then
595 * some other thread starts a transaction commit which tries to do an
596 * allocation while the other thread is still loading the space cache
597 * info. The previous loop should have kept us from choosing this block
598 * group, but if we've moved to the state where we will wait on caching
599 * block groups we need to first check if we're doing a fast load here,
600 * so we can wait for it to finish, otherwise we could end up allocating
601 * from a block group who's cache gets evicted for one reason or
604 while (cache->cached == BTRFS_CACHE_FAST) {
605 struct btrfs_caching_control *ctl;
607 ctl = cache->caching_ctl;
608 refcount_inc(&ctl->count);
609 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
610 spin_unlock(&cache->lock);
614 finish_wait(&ctl->wait, &wait);
615 put_caching_control(ctl);
616 spin_lock(&cache->lock);
619 if (cache->cached != BTRFS_CACHE_NO) {
620 spin_unlock(&cache->lock);
624 WARN_ON(cache->caching_ctl);
625 cache->caching_ctl = caching_ctl;
626 cache->cached = BTRFS_CACHE_FAST;
627 spin_unlock(&cache->lock);
629 if (btrfs_test_opt(fs_info, SPACE_CACHE)) {
630 mutex_lock(&caching_ctl->mutex);
631 ret = load_free_space_cache(fs_info, cache);
633 spin_lock(&cache->lock);
635 cache->caching_ctl = NULL;
636 cache->cached = BTRFS_CACHE_FINISHED;
637 cache->last_byte_to_unpin = (u64)-1;
638 caching_ctl->progress = (u64)-1;
640 if (load_cache_only) {
641 cache->caching_ctl = NULL;
642 cache->cached = BTRFS_CACHE_NO;
644 cache->cached = BTRFS_CACHE_STARTED;
645 cache->has_caching_ctl = 1;
648 spin_unlock(&cache->lock);
649 #ifdef CONFIG_BTRFS_DEBUG
651 btrfs_should_fragment_free_space(cache)) {
654 spin_lock(&cache->space_info->lock);
655 spin_lock(&cache->lock);
656 bytes_used = cache->key.offset -
657 btrfs_block_group_used(&cache->item);
658 cache->space_info->bytes_used += bytes_used >> 1;
659 spin_unlock(&cache->lock);
660 spin_unlock(&cache->space_info->lock);
661 fragment_free_space(cache);
664 mutex_unlock(&caching_ctl->mutex);
666 wake_up(&caching_ctl->wait);
668 put_caching_control(caching_ctl);
669 free_excluded_extents(fs_info, cache);
674 * We're either using the free space tree or no caching at all.
675 * Set cached to the appropriate value and wakeup any waiters.
677 spin_lock(&cache->lock);
678 if (load_cache_only) {
679 cache->caching_ctl = NULL;
680 cache->cached = BTRFS_CACHE_NO;
682 cache->cached = BTRFS_CACHE_STARTED;
683 cache->has_caching_ctl = 1;
685 spin_unlock(&cache->lock);
686 wake_up(&caching_ctl->wait);
689 if (load_cache_only) {
690 put_caching_control(caching_ctl);
694 down_write(&fs_info->commit_root_sem);
695 refcount_inc(&caching_ctl->count);
696 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
697 up_write(&fs_info->commit_root_sem);
699 btrfs_get_block_group(cache);
701 btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
707 * return the block group that starts at or after bytenr
709 static struct btrfs_block_group_cache *
710 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
712 return block_group_cache_tree_search(info, bytenr, 0);
716 * return the block group that contains the given bytenr
718 struct btrfs_block_group_cache *btrfs_lookup_block_group(
719 struct btrfs_fs_info *info,
722 return block_group_cache_tree_search(info, bytenr, 1);
725 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
728 struct list_head *head = &info->space_info;
729 struct btrfs_space_info *found;
731 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
734 list_for_each_entry_rcu(found, head, list) {
735 if (found->flags & flags) {
744 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, s64 num_bytes,
745 bool metadata, u64 root_objectid)
747 struct btrfs_space_info *space_info;
751 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
752 flags = BTRFS_BLOCK_GROUP_SYSTEM;
754 flags = BTRFS_BLOCK_GROUP_METADATA;
756 flags = BTRFS_BLOCK_GROUP_DATA;
759 space_info = __find_space_info(fs_info, flags);
761 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
765 * after adding space to the filesystem, we need to clear the full flags
766 * on all the space infos.
768 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
770 struct list_head *head = &info->space_info;
771 struct btrfs_space_info *found;
774 list_for_each_entry_rcu(found, head, list)
779 /* simple helper to search for an existing data extent at a given offset */
780 int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len)
783 struct btrfs_key key;
784 struct btrfs_path *path;
786 path = btrfs_alloc_path();
790 key.objectid = start;
792 key.type = BTRFS_EXTENT_ITEM_KEY;
793 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
794 btrfs_free_path(path);
799 * helper function to lookup reference count and flags of a tree block.
801 * the head node for delayed ref is used to store the sum of all the
802 * reference count modifications queued up in the rbtree. the head
803 * node may also store the extent flags to set. This way you can check
804 * to see what the reference count and extent flags would be if all of
805 * the delayed refs are not processed.
807 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
808 struct btrfs_fs_info *fs_info, u64 bytenr,
809 u64 offset, int metadata, u64 *refs, u64 *flags)
811 struct btrfs_delayed_ref_head *head;
812 struct btrfs_delayed_ref_root *delayed_refs;
813 struct btrfs_path *path;
814 struct btrfs_extent_item *ei;
815 struct extent_buffer *leaf;
816 struct btrfs_key key;
823 * If we don't have skinny metadata, don't bother doing anything
826 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) {
827 offset = fs_info->nodesize;
831 path = btrfs_alloc_path();
836 path->skip_locking = 1;
837 path->search_commit_root = 1;
841 key.objectid = bytenr;
844 key.type = BTRFS_METADATA_ITEM_KEY;
846 key.type = BTRFS_EXTENT_ITEM_KEY;
848 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
852 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
853 if (path->slots[0]) {
855 btrfs_item_key_to_cpu(path->nodes[0], &key,
857 if (key.objectid == bytenr &&
858 key.type == BTRFS_EXTENT_ITEM_KEY &&
859 key.offset == fs_info->nodesize)
865 leaf = path->nodes[0];
866 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
867 if (item_size >= sizeof(*ei)) {
868 ei = btrfs_item_ptr(leaf, path->slots[0],
869 struct btrfs_extent_item);
870 num_refs = btrfs_extent_refs(leaf, ei);
871 extent_flags = btrfs_extent_flags(leaf, ei);
873 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
874 struct btrfs_extent_item_v0 *ei0;
875 BUG_ON(item_size != sizeof(*ei0));
876 ei0 = btrfs_item_ptr(leaf, path->slots[0],
877 struct btrfs_extent_item_v0);
878 num_refs = btrfs_extent_refs_v0(leaf, ei0);
879 /* FIXME: this isn't correct for data */
880 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
885 BUG_ON(num_refs == 0);
895 delayed_refs = &trans->transaction->delayed_refs;
896 spin_lock(&delayed_refs->lock);
897 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
899 if (!mutex_trylock(&head->mutex)) {
900 refcount_inc(&head->refs);
901 spin_unlock(&delayed_refs->lock);
903 btrfs_release_path(path);
906 * Mutex was contended, block until it's released and try
909 mutex_lock(&head->mutex);
910 mutex_unlock(&head->mutex);
911 btrfs_put_delayed_ref_head(head);
914 spin_lock(&head->lock);
915 if (head->extent_op && head->extent_op->update_flags)
916 extent_flags |= head->extent_op->flags_to_set;
918 BUG_ON(num_refs == 0);
920 num_refs += head->ref_mod;
921 spin_unlock(&head->lock);
922 mutex_unlock(&head->mutex);
924 spin_unlock(&delayed_refs->lock);
926 WARN_ON(num_refs == 0);
930 *flags = extent_flags;
932 btrfs_free_path(path);
937 * Back reference rules. Back refs have three main goals:
939 * 1) differentiate between all holders of references to an extent so that
940 * when a reference is dropped we can make sure it was a valid reference
941 * before freeing the extent.
943 * 2) Provide enough information to quickly find the holders of an extent
944 * if we notice a given block is corrupted or bad.
946 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
947 * maintenance. This is actually the same as #2, but with a slightly
948 * different use case.
950 * There are two kinds of back refs. The implicit back refs is optimized
951 * for pointers in non-shared tree blocks. For a given pointer in a block,
952 * back refs of this kind provide information about the block's owner tree
953 * and the pointer's key. These information allow us to find the block by
954 * b-tree searching. The full back refs is for pointers in tree blocks not
955 * referenced by their owner trees. The location of tree block is recorded
956 * in the back refs. Actually the full back refs is generic, and can be
957 * used in all cases the implicit back refs is used. The major shortcoming
958 * of the full back refs is its overhead. Every time a tree block gets
959 * COWed, we have to update back refs entry for all pointers in it.
961 * For a newly allocated tree block, we use implicit back refs for
962 * pointers in it. This means most tree related operations only involve
963 * implicit back refs. For a tree block created in old transaction, the
964 * only way to drop a reference to it is COW it. So we can detect the
965 * event that tree block loses its owner tree's reference and do the
966 * back refs conversion.
968 * When a tree block is COWed through a tree, there are four cases:
970 * The reference count of the block is one and the tree is the block's
971 * owner tree. Nothing to do in this case.
973 * The reference count of the block is one and the tree is not the
974 * block's owner tree. In this case, full back refs is used for pointers
975 * in the block. Remove these full back refs, add implicit back refs for
976 * every pointers in the new block.
978 * The reference count of the block is greater than one and the tree is
979 * the block's owner tree. In this case, implicit back refs is used for
980 * pointers in the block. Add full back refs for every pointers in the
981 * block, increase lower level extents' reference counts. The original
982 * implicit back refs are entailed to the new block.
984 * The reference count of the block is greater than one and the tree is
985 * not the block's owner tree. Add implicit back refs for every pointer in
986 * the new block, increase lower level extents' reference count.
988 * Back Reference Key composing:
990 * The key objectid corresponds to the first byte in the extent,
991 * The key type is used to differentiate between types of back refs.
992 * There are different meanings of the key offset for different types
995 * File extents can be referenced by:
997 * - multiple snapshots, subvolumes, or different generations in one subvol
998 * - different files inside a single subvolume
999 * - different offsets inside a file (bookend extents in file.c)
1001 * The extent ref structure for the implicit back refs has fields for:
1003 * - Objectid of the subvolume root
1004 * - objectid of the file holding the reference
1005 * - original offset in the file
1006 * - how many bookend extents
1008 * The key offset for the implicit back refs is hash of the first
1011 * The extent ref structure for the full back refs has field for:
1013 * - number of pointers in the tree leaf
1015 * The key offset for the implicit back refs is the first byte of
1018 * When a file extent is allocated, The implicit back refs is used.
1019 * the fields are filled in:
1021 * (root_key.objectid, inode objectid, offset in file, 1)
1023 * When a file extent is removed file truncation, we find the
1024 * corresponding implicit back refs and check the following fields:
1026 * (btrfs_header_owner(leaf), inode objectid, offset in file)
1028 * Btree extents can be referenced by:
1030 * - Different subvolumes
1032 * Both the implicit back refs and the full back refs for tree blocks
1033 * only consist of key. The key offset for the implicit back refs is
1034 * objectid of block's owner tree. The key offset for the full back refs
1035 * is the first byte of parent block.
1037 * When implicit back refs is used, information about the lowest key and
1038 * level of the tree block are required. These information are stored in
1039 * tree block info structure.
1042 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1043 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
1044 struct btrfs_fs_info *fs_info,
1045 struct btrfs_path *path,
1046 u64 owner, u32 extra_size)
1048 struct btrfs_root *root = fs_info->extent_root;
1049 struct btrfs_extent_item *item;
1050 struct btrfs_extent_item_v0 *ei0;
1051 struct btrfs_extent_ref_v0 *ref0;
1052 struct btrfs_tree_block_info *bi;
1053 struct extent_buffer *leaf;
1054 struct btrfs_key key;
1055 struct btrfs_key found_key;
1056 u32 new_size = sizeof(*item);
1060 leaf = path->nodes[0];
1061 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
1063 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1064 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1065 struct btrfs_extent_item_v0);
1066 refs = btrfs_extent_refs_v0(leaf, ei0);
1068 if (owner == (u64)-1) {
1070 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1071 ret = btrfs_next_leaf(root, path);
1074 BUG_ON(ret > 0); /* Corruption */
1075 leaf = path->nodes[0];
1077 btrfs_item_key_to_cpu(leaf, &found_key,
1079 BUG_ON(key.objectid != found_key.objectid);
1080 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1084 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1085 struct btrfs_extent_ref_v0);
1086 owner = btrfs_ref_objectid_v0(leaf, ref0);
1090 btrfs_release_path(path);
1092 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1093 new_size += sizeof(*bi);
1095 new_size -= sizeof(*ei0);
1096 ret = btrfs_search_slot(trans, root, &key, path,
1097 new_size + extra_size, 1);
1100 BUG_ON(ret); /* Corruption */
1102 btrfs_extend_item(fs_info, path, new_size);
1104 leaf = path->nodes[0];
1105 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1106 btrfs_set_extent_refs(leaf, item, refs);
1107 /* FIXME: get real generation */
1108 btrfs_set_extent_generation(leaf, item, 0);
1109 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1110 btrfs_set_extent_flags(leaf, item,
1111 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1112 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1113 bi = (struct btrfs_tree_block_info *)(item + 1);
1114 /* FIXME: get first key of the block */
1115 memzero_extent_buffer(leaf, (unsigned long)bi, sizeof(*bi));
1116 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1118 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1120 btrfs_mark_buffer_dirty(leaf);
1126 * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required,
1127 * is_data == BTRFS_REF_TYPE_DATA, data type is requried,
1128 * is_data == BTRFS_REF_TYPE_ANY, either type is OK.
1130 int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb,
1131 struct btrfs_extent_inline_ref *iref,
1132 enum btrfs_inline_ref_type is_data)
1134 int type = btrfs_extent_inline_ref_type(eb, iref);
1135 u64 offset = btrfs_extent_inline_ref_offset(eb, iref);
1137 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1138 type == BTRFS_SHARED_BLOCK_REF_KEY ||
1139 type == BTRFS_SHARED_DATA_REF_KEY ||
1140 type == BTRFS_EXTENT_DATA_REF_KEY) {
1141 if (is_data == BTRFS_REF_TYPE_BLOCK) {
1142 if (type == BTRFS_TREE_BLOCK_REF_KEY)
1144 if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1145 ASSERT(eb->fs_info);
1147 * Every shared one has parent tree
1148 * block, which must be aligned to
1152 IS_ALIGNED(offset, eb->fs_info->nodesize))
1155 } else if (is_data == BTRFS_REF_TYPE_DATA) {
1156 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1158 if (type == BTRFS_SHARED_DATA_REF_KEY) {
1159 ASSERT(eb->fs_info);
1161 * Every shared one has parent tree
1162 * block, which must be aligned to
1166 IS_ALIGNED(offset, eb->fs_info->nodesize))
1170 ASSERT(is_data == BTRFS_REF_TYPE_ANY);
1175 btrfs_print_leaf((struct extent_buffer *)eb);
1176 btrfs_err(eb->fs_info, "eb %llu invalid extent inline ref type %d",
1180 return BTRFS_REF_TYPE_INVALID;
1183 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1185 u32 high_crc = ~(u32)0;
1186 u32 low_crc = ~(u32)0;
1189 lenum = cpu_to_le64(root_objectid);
1190 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1191 lenum = cpu_to_le64(owner);
1192 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1193 lenum = cpu_to_le64(offset);
1194 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1196 return ((u64)high_crc << 31) ^ (u64)low_crc;
1199 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1200 struct btrfs_extent_data_ref *ref)
1202 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1203 btrfs_extent_data_ref_objectid(leaf, ref),
1204 btrfs_extent_data_ref_offset(leaf, ref));
1207 static int match_extent_data_ref(struct extent_buffer *leaf,
1208 struct btrfs_extent_data_ref *ref,
1209 u64 root_objectid, u64 owner, u64 offset)
1211 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1212 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1213 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1218 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1219 struct btrfs_fs_info *fs_info,
1220 struct btrfs_path *path,
1221 u64 bytenr, u64 parent,
1223 u64 owner, u64 offset)
1225 struct btrfs_root *root = fs_info->extent_root;
1226 struct btrfs_key key;
1227 struct btrfs_extent_data_ref *ref;
1228 struct extent_buffer *leaf;
1234 key.objectid = bytenr;
1236 key.type = BTRFS_SHARED_DATA_REF_KEY;
1237 key.offset = parent;
1239 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1240 key.offset = hash_extent_data_ref(root_objectid,
1245 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1254 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1255 key.type = BTRFS_EXTENT_REF_V0_KEY;
1256 btrfs_release_path(path);
1257 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1268 leaf = path->nodes[0];
1269 nritems = btrfs_header_nritems(leaf);
1271 if (path->slots[0] >= nritems) {
1272 ret = btrfs_next_leaf(root, path);
1278 leaf = path->nodes[0];
1279 nritems = btrfs_header_nritems(leaf);
1283 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1284 if (key.objectid != bytenr ||
1285 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1288 ref = btrfs_item_ptr(leaf, path->slots[0],
1289 struct btrfs_extent_data_ref);
1291 if (match_extent_data_ref(leaf, ref, root_objectid,
1294 btrfs_release_path(path);
1306 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1307 struct btrfs_fs_info *fs_info,
1308 struct btrfs_path *path,
1309 u64 bytenr, u64 parent,
1310 u64 root_objectid, u64 owner,
1311 u64 offset, int refs_to_add)
1313 struct btrfs_root *root = fs_info->extent_root;
1314 struct btrfs_key key;
1315 struct extent_buffer *leaf;
1320 key.objectid = bytenr;
1322 key.type = BTRFS_SHARED_DATA_REF_KEY;
1323 key.offset = parent;
1324 size = sizeof(struct btrfs_shared_data_ref);
1326 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1327 key.offset = hash_extent_data_ref(root_objectid,
1329 size = sizeof(struct btrfs_extent_data_ref);
1332 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1333 if (ret && ret != -EEXIST)
1336 leaf = path->nodes[0];
1338 struct btrfs_shared_data_ref *ref;
1339 ref = btrfs_item_ptr(leaf, path->slots[0],
1340 struct btrfs_shared_data_ref);
1342 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1344 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1345 num_refs += refs_to_add;
1346 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1349 struct btrfs_extent_data_ref *ref;
1350 while (ret == -EEXIST) {
1351 ref = btrfs_item_ptr(leaf, path->slots[0],
1352 struct btrfs_extent_data_ref);
1353 if (match_extent_data_ref(leaf, ref, root_objectid,
1356 btrfs_release_path(path);
1358 ret = btrfs_insert_empty_item(trans, root, path, &key,
1360 if (ret && ret != -EEXIST)
1363 leaf = path->nodes[0];
1365 ref = btrfs_item_ptr(leaf, path->slots[0],
1366 struct btrfs_extent_data_ref);
1368 btrfs_set_extent_data_ref_root(leaf, ref,
1370 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1371 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1372 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1374 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1375 num_refs += refs_to_add;
1376 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1379 btrfs_mark_buffer_dirty(leaf);
1382 btrfs_release_path(path);
1386 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1387 struct btrfs_fs_info *fs_info,
1388 struct btrfs_path *path,
1389 int refs_to_drop, int *last_ref)
1391 struct btrfs_key key;
1392 struct btrfs_extent_data_ref *ref1 = NULL;
1393 struct btrfs_shared_data_ref *ref2 = NULL;
1394 struct extent_buffer *leaf;
1398 leaf = path->nodes[0];
1399 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1401 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1402 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1403 struct btrfs_extent_data_ref);
1404 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1405 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1406 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1407 struct btrfs_shared_data_ref);
1408 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1409 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1410 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1411 struct btrfs_extent_ref_v0 *ref0;
1412 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1413 struct btrfs_extent_ref_v0);
1414 num_refs = btrfs_ref_count_v0(leaf, ref0);
1420 BUG_ON(num_refs < refs_to_drop);
1421 num_refs -= refs_to_drop;
1423 if (num_refs == 0) {
1424 ret = btrfs_del_item(trans, fs_info->extent_root, path);
1427 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1428 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1429 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1430 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1431 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1433 struct btrfs_extent_ref_v0 *ref0;
1434 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1435 struct btrfs_extent_ref_v0);
1436 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1439 btrfs_mark_buffer_dirty(leaf);
1444 static noinline u32 extent_data_ref_count(struct btrfs_path *path,
1445 struct btrfs_extent_inline_ref *iref)
1447 struct btrfs_key key;
1448 struct extent_buffer *leaf;
1449 struct btrfs_extent_data_ref *ref1;
1450 struct btrfs_shared_data_ref *ref2;
1454 leaf = path->nodes[0];
1455 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1458 * If type is invalid, we should have bailed out earlier than
1461 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
1462 ASSERT(type != BTRFS_REF_TYPE_INVALID);
1463 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1464 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1465 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1467 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1468 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1470 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1471 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1472 struct btrfs_extent_data_ref);
1473 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1474 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1475 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1476 struct btrfs_shared_data_ref);
1477 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1478 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1479 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1480 struct btrfs_extent_ref_v0 *ref0;
1481 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1482 struct btrfs_extent_ref_v0);
1483 num_refs = btrfs_ref_count_v0(leaf, ref0);
1491 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1492 struct btrfs_fs_info *fs_info,
1493 struct btrfs_path *path,
1494 u64 bytenr, u64 parent,
1497 struct btrfs_root *root = fs_info->extent_root;
1498 struct btrfs_key key;
1501 key.objectid = bytenr;
1503 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1504 key.offset = parent;
1506 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1507 key.offset = root_objectid;
1510 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1513 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1514 if (ret == -ENOENT && parent) {
1515 btrfs_release_path(path);
1516 key.type = BTRFS_EXTENT_REF_V0_KEY;
1517 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1525 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1526 struct btrfs_fs_info *fs_info,
1527 struct btrfs_path *path,
1528 u64 bytenr, u64 parent,
1531 struct btrfs_key key;
1534 key.objectid = bytenr;
1536 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1537 key.offset = parent;
1539 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1540 key.offset = root_objectid;
1543 ret = btrfs_insert_empty_item(trans, fs_info->extent_root,
1545 btrfs_release_path(path);
1549 static inline int extent_ref_type(u64 parent, u64 owner)
1552 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1554 type = BTRFS_SHARED_BLOCK_REF_KEY;
1556 type = BTRFS_TREE_BLOCK_REF_KEY;
1559 type = BTRFS_SHARED_DATA_REF_KEY;
1561 type = BTRFS_EXTENT_DATA_REF_KEY;
1566 static int find_next_key(struct btrfs_path *path, int level,
1567 struct btrfs_key *key)
1570 for (; level < BTRFS_MAX_LEVEL; level++) {
1571 if (!path->nodes[level])
1573 if (path->slots[level] + 1 >=
1574 btrfs_header_nritems(path->nodes[level]))
1577 btrfs_item_key_to_cpu(path->nodes[level], key,
1578 path->slots[level] + 1);
1580 btrfs_node_key_to_cpu(path->nodes[level], key,
1581 path->slots[level] + 1);
1588 * look for inline back ref. if back ref is found, *ref_ret is set
1589 * to the address of inline back ref, and 0 is returned.
1591 * if back ref isn't found, *ref_ret is set to the address where it
1592 * should be inserted, and -ENOENT is returned.
1594 * if insert is true and there are too many inline back refs, the path
1595 * points to the extent item, and -EAGAIN is returned.
1597 * NOTE: inline back refs are ordered in the same way that back ref
1598 * items in the tree are ordered.
1600 static noinline_for_stack
1601 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1602 struct btrfs_fs_info *fs_info,
1603 struct btrfs_path *path,
1604 struct btrfs_extent_inline_ref **ref_ret,
1605 u64 bytenr, u64 num_bytes,
1606 u64 parent, u64 root_objectid,
1607 u64 owner, u64 offset, int insert)
1609 struct btrfs_root *root = fs_info->extent_root;
1610 struct btrfs_key key;
1611 struct extent_buffer *leaf;
1612 struct btrfs_extent_item *ei;
1613 struct btrfs_extent_inline_ref *iref;
1623 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
1626 key.objectid = bytenr;
1627 key.type = BTRFS_EXTENT_ITEM_KEY;
1628 key.offset = num_bytes;
1630 want = extent_ref_type(parent, owner);
1632 extra_size = btrfs_extent_inline_ref_size(want);
1633 path->keep_locks = 1;
1638 * Owner is our parent level, so we can just add one to get the level
1639 * for the block we are interested in.
1641 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1642 key.type = BTRFS_METADATA_ITEM_KEY;
1647 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1654 * We may be a newly converted file system which still has the old fat
1655 * extent entries for metadata, so try and see if we have one of those.
1657 if (ret > 0 && skinny_metadata) {
1658 skinny_metadata = false;
1659 if (path->slots[0]) {
1661 btrfs_item_key_to_cpu(path->nodes[0], &key,
1663 if (key.objectid == bytenr &&
1664 key.type == BTRFS_EXTENT_ITEM_KEY &&
1665 key.offset == num_bytes)
1669 key.objectid = bytenr;
1670 key.type = BTRFS_EXTENT_ITEM_KEY;
1671 key.offset = num_bytes;
1672 btrfs_release_path(path);
1677 if (ret && !insert) {
1680 } else if (WARN_ON(ret)) {
1685 leaf = path->nodes[0];
1686 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1687 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1688 if (item_size < sizeof(*ei)) {
1693 ret = convert_extent_item_v0(trans, fs_info, path, owner,
1699 leaf = path->nodes[0];
1700 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1703 BUG_ON(item_size < sizeof(*ei));
1705 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1706 flags = btrfs_extent_flags(leaf, ei);
1708 ptr = (unsigned long)(ei + 1);
1709 end = (unsigned long)ei + item_size;
1711 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1712 ptr += sizeof(struct btrfs_tree_block_info);
1716 if (owner >= BTRFS_FIRST_FREE_OBJECTID)
1717 needed = BTRFS_REF_TYPE_DATA;
1719 needed = BTRFS_REF_TYPE_BLOCK;
1727 iref = (struct btrfs_extent_inline_ref *)ptr;
1728 type = btrfs_get_extent_inline_ref_type(leaf, iref, needed);
1729 if (type == BTRFS_REF_TYPE_INVALID) {
1737 ptr += btrfs_extent_inline_ref_size(type);
1741 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1742 struct btrfs_extent_data_ref *dref;
1743 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1744 if (match_extent_data_ref(leaf, dref, root_objectid,
1749 if (hash_extent_data_ref_item(leaf, dref) <
1750 hash_extent_data_ref(root_objectid, owner, offset))
1754 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1756 if (parent == ref_offset) {
1760 if (ref_offset < parent)
1763 if (root_objectid == ref_offset) {
1767 if (ref_offset < root_objectid)
1771 ptr += btrfs_extent_inline_ref_size(type);
1773 if (err == -ENOENT && insert) {
1774 if (item_size + extra_size >=
1775 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1780 * To add new inline back ref, we have to make sure
1781 * there is no corresponding back ref item.
1782 * For simplicity, we just do not add new inline back
1783 * ref if there is any kind of item for this block
1785 if (find_next_key(path, 0, &key) == 0 &&
1786 key.objectid == bytenr &&
1787 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1792 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1795 path->keep_locks = 0;
1796 btrfs_unlock_up_safe(path, 1);
1802 * helper to add new inline back ref
1804 static noinline_for_stack
1805 void setup_inline_extent_backref(struct btrfs_fs_info *fs_info,
1806 struct btrfs_path *path,
1807 struct btrfs_extent_inline_ref *iref,
1808 u64 parent, u64 root_objectid,
1809 u64 owner, u64 offset, int refs_to_add,
1810 struct btrfs_delayed_extent_op *extent_op)
1812 struct extent_buffer *leaf;
1813 struct btrfs_extent_item *ei;
1816 unsigned long item_offset;
1821 leaf = path->nodes[0];
1822 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1823 item_offset = (unsigned long)iref - (unsigned long)ei;
1825 type = extent_ref_type(parent, owner);
1826 size = btrfs_extent_inline_ref_size(type);
1828 btrfs_extend_item(fs_info, path, size);
1830 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1831 refs = btrfs_extent_refs(leaf, ei);
1832 refs += refs_to_add;
1833 btrfs_set_extent_refs(leaf, ei, refs);
1835 __run_delayed_extent_op(extent_op, leaf, ei);
1837 ptr = (unsigned long)ei + item_offset;
1838 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1839 if (ptr < end - size)
1840 memmove_extent_buffer(leaf, ptr + size, ptr,
1843 iref = (struct btrfs_extent_inline_ref *)ptr;
1844 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1845 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1846 struct btrfs_extent_data_ref *dref;
1847 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1848 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1849 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1850 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1851 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1852 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1853 struct btrfs_shared_data_ref *sref;
1854 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1855 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1856 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1857 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1858 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1860 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1862 btrfs_mark_buffer_dirty(leaf);
1865 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1866 struct btrfs_fs_info *fs_info,
1867 struct btrfs_path *path,
1868 struct btrfs_extent_inline_ref **ref_ret,
1869 u64 bytenr, u64 num_bytes, u64 parent,
1870 u64 root_objectid, u64 owner, u64 offset)
1874 ret = lookup_inline_extent_backref(trans, fs_info, path, ref_ret,
1875 bytenr, num_bytes, parent,
1876 root_objectid, owner, offset, 0);
1880 btrfs_release_path(path);
1883 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1884 ret = lookup_tree_block_ref(trans, fs_info, path, bytenr,
1885 parent, root_objectid);
1887 ret = lookup_extent_data_ref(trans, fs_info, path, bytenr,
1888 parent, root_objectid, owner,
1895 * helper to update/remove inline back ref
1897 static noinline_for_stack
1898 void update_inline_extent_backref(struct btrfs_fs_info *fs_info,
1899 struct btrfs_path *path,
1900 struct btrfs_extent_inline_ref *iref,
1902 struct btrfs_delayed_extent_op *extent_op,
1905 struct extent_buffer *leaf;
1906 struct btrfs_extent_item *ei;
1907 struct btrfs_extent_data_ref *dref = NULL;
1908 struct btrfs_shared_data_ref *sref = NULL;
1916 leaf = path->nodes[0];
1917 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1918 refs = btrfs_extent_refs(leaf, ei);
1919 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1920 refs += refs_to_mod;
1921 btrfs_set_extent_refs(leaf, ei, refs);
1923 __run_delayed_extent_op(extent_op, leaf, ei);
1926 * If type is invalid, we should have bailed out after
1927 * lookup_inline_extent_backref().
1929 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY);
1930 ASSERT(type != BTRFS_REF_TYPE_INVALID);
1932 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1933 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1934 refs = btrfs_extent_data_ref_count(leaf, dref);
1935 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1936 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1937 refs = btrfs_shared_data_ref_count(leaf, sref);
1940 BUG_ON(refs_to_mod != -1);
1943 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1944 refs += refs_to_mod;
1947 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1948 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1950 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1953 size = btrfs_extent_inline_ref_size(type);
1954 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1955 ptr = (unsigned long)iref;
1956 end = (unsigned long)ei + item_size;
1957 if (ptr + size < end)
1958 memmove_extent_buffer(leaf, ptr, ptr + size,
1961 btrfs_truncate_item(fs_info, path, item_size, 1);
1963 btrfs_mark_buffer_dirty(leaf);
1966 static noinline_for_stack
1967 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1968 struct btrfs_fs_info *fs_info,
1969 struct btrfs_path *path,
1970 u64 bytenr, u64 num_bytes, u64 parent,
1971 u64 root_objectid, u64 owner,
1972 u64 offset, int refs_to_add,
1973 struct btrfs_delayed_extent_op *extent_op)
1975 struct btrfs_extent_inline_ref *iref;
1978 ret = lookup_inline_extent_backref(trans, fs_info, path, &iref,
1979 bytenr, num_bytes, parent,
1980 root_objectid, owner, offset, 1);
1982 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1983 update_inline_extent_backref(fs_info, path, iref,
1984 refs_to_add, extent_op, NULL);
1985 } else if (ret == -ENOENT) {
1986 setup_inline_extent_backref(fs_info, path, iref, parent,
1987 root_objectid, owner, offset,
1988 refs_to_add, extent_op);
1994 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1995 struct btrfs_fs_info *fs_info,
1996 struct btrfs_path *path,
1997 u64 bytenr, u64 parent, u64 root_objectid,
1998 u64 owner, u64 offset, int refs_to_add)
2001 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2002 BUG_ON(refs_to_add != 1);
2003 ret = insert_tree_block_ref(trans, fs_info, path, bytenr,
2004 parent, root_objectid);
2006 ret = insert_extent_data_ref(trans, fs_info, path, bytenr,
2007 parent, root_objectid,
2008 owner, offset, refs_to_add);
2013 static int remove_extent_backref(struct btrfs_trans_handle *trans,
2014 struct btrfs_fs_info *fs_info,
2015 struct btrfs_path *path,
2016 struct btrfs_extent_inline_ref *iref,
2017 int refs_to_drop, int is_data, int *last_ref)
2021 BUG_ON(!is_data && refs_to_drop != 1);
2023 update_inline_extent_backref(fs_info, path, iref,
2024 -refs_to_drop, NULL, last_ref);
2025 } else if (is_data) {
2026 ret = remove_extent_data_ref(trans, fs_info, path, refs_to_drop,
2030 ret = btrfs_del_item(trans, fs_info->extent_root, path);
2035 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
2036 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
2037 u64 *discarded_bytes)
2040 u64 bytes_left, end;
2041 u64 aligned_start = ALIGN(start, 1 << 9);
2043 if (WARN_ON(start != aligned_start)) {
2044 len -= aligned_start - start;
2045 len = round_down(len, 1 << 9);
2046 start = aligned_start;
2049 *discarded_bytes = 0;
2057 /* Skip any superblocks on this device. */
2058 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
2059 u64 sb_start = btrfs_sb_offset(j);
2060 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
2061 u64 size = sb_start - start;
2063 if (!in_range(sb_start, start, bytes_left) &&
2064 !in_range(sb_end, start, bytes_left) &&
2065 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
2069 * Superblock spans beginning of range. Adjust start and
2072 if (sb_start <= start) {
2073 start += sb_end - start;
2078 bytes_left = end - start;
2083 ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
2086 *discarded_bytes += size;
2087 else if (ret != -EOPNOTSUPP)
2096 bytes_left = end - start;
2100 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
2103 *discarded_bytes += bytes_left;
2108 int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr,
2109 u64 num_bytes, u64 *actual_bytes)
2112 u64 discarded_bytes = 0;
2113 struct btrfs_bio *bbio = NULL;
2117 * Avoid races with device replace and make sure our bbio has devices
2118 * associated to its stripes that don't go away while we are discarding.
2120 btrfs_bio_counter_inc_blocked(fs_info);
2121 /* Tell the block device(s) that the sectors can be discarded */
2122 ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, bytenr, &num_bytes,
2124 /* Error condition is -ENOMEM */
2126 struct btrfs_bio_stripe *stripe = bbio->stripes;
2130 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
2132 struct request_queue *req_q;
2134 if (!stripe->dev->bdev) {
2135 ASSERT(btrfs_test_opt(fs_info, DEGRADED));
2138 req_q = bdev_get_queue(stripe->dev->bdev);
2139 if (!blk_queue_discard(req_q))
2142 ret = btrfs_issue_discard(stripe->dev->bdev,
2147 discarded_bytes += bytes;
2148 else if (ret != -EOPNOTSUPP)
2149 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2152 * Just in case we get back EOPNOTSUPP for some reason,
2153 * just ignore the return value so we don't screw up
2154 * people calling discard_extent.
2158 btrfs_put_bbio(bbio);
2160 btrfs_bio_counter_dec(fs_info);
2163 *actual_bytes = discarded_bytes;
2166 if (ret == -EOPNOTSUPP)
2171 /* Can return -ENOMEM */
2172 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2173 struct btrfs_root *root,
2174 u64 bytenr, u64 num_bytes, u64 parent,
2175 u64 root_objectid, u64 owner, u64 offset)
2177 struct btrfs_fs_info *fs_info = root->fs_info;
2178 int old_ref_mod, new_ref_mod;
2181 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
2182 root_objectid == BTRFS_TREE_LOG_OBJECTID);
2184 btrfs_ref_tree_mod(root, bytenr, num_bytes, parent, root_objectid,
2185 owner, offset, BTRFS_ADD_DELAYED_REF);
2187 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2188 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
2190 root_objectid, (int)owner,
2191 BTRFS_ADD_DELAYED_REF, NULL,
2192 &old_ref_mod, &new_ref_mod);
2194 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
2196 root_objectid, owner, offset,
2197 0, BTRFS_ADD_DELAYED_REF,
2198 &old_ref_mod, &new_ref_mod);
2201 if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0) {
2202 bool metadata = owner < BTRFS_FIRST_FREE_OBJECTID;
2204 add_pinned_bytes(fs_info, -num_bytes, metadata, root_objectid);
2210 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2211 struct btrfs_fs_info *fs_info,
2212 struct btrfs_delayed_ref_node *node,
2213 u64 parent, u64 root_objectid,
2214 u64 owner, u64 offset, int refs_to_add,
2215 struct btrfs_delayed_extent_op *extent_op)
2217 struct btrfs_path *path;
2218 struct extent_buffer *leaf;
2219 struct btrfs_extent_item *item;
2220 struct btrfs_key key;
2221 u64 bytenr = node->bytenr;
2222 u64 num_bytes = node->num_bytes;
2226 path = btrfs_alloc_path();
2230 path->reada = READA_FORWARD;
2231 path->leave_spinning = 1;
2232 /* this will setup the path even if it fails to insert the back ref */
2233 ret = insert_inline_extent_backref(trans, fs_info, path, bytenr,
2234 num_bytes, parent, root_objectid,
2236 refs_to_add, extent_op);
2237 if ((ret < 0 && ret != -EAGAIN) || !ret)
2241 * Ok we had -EAGAIN which means we didn't have space to insert and
2242 * inline extent ref, so just update the reference count and add a
2245 leaf = path->nodes[0];
2246 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2247 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2248 refs = btrfs_extent_refs(leaf, item);
2249 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2251 __run_delayed_extent_op(extent_op, leaf, item);
2253 btrfs_mark_buffer_dirty(leaf);
2254 btrfs_release_path(path);
2256 path->reada = READA_FORWARD;
2257 path->leave_spinning = 1;
2258 /* now insert the actual backref */
2259 ret = insert_extent_backref(trans, fs_info, path, bytenr, parent,
2260 root_objectid, owner, offset, refs_to_add);
2262 btrfs_abort_transaction(trans, ret);
2264 btrfs_free_path(path);
2268 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2269 struct btrfs_fs_info *fs_info,
2270 struct btrfs_delayed_ref_node *node,
2271 struct btrfs_delayed_extent_op *extent_op,
2272 int insert_reserved)
2275 struct btrfs_delayed_data_ref *ref;
2276 struct btrfs_key ins;
2281 ins.objectid = node->bytenr;
2282 ins.offset = node->num_bytes;
2283 ins.type = BTRFS_EXTENT_ITEM_KEY;
2285 ref = btrfs_delayed_node_to_data_ref(node);
2286 trace_run_delayed_data_ref(fs_info, node, ref, node->action);
2288 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2289 parent = ref->parent;
2290 ref_root = ref->root;
2292 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2294 flags |= extent_op->flags_to_set;
2295 ret = alloc_reserved_file_extent(trans, fs_info,
2296 parent, ref_root, flags,
2297 ref->objectid, ref->offset,
2298 &ins, node->ref_mod);
2299 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2300 ret = __btrfs_inc_extent_ref(trans, fs_info, node, parent,
2301 ref_root, ref->objectid,
2302 ref->offset, node->ref_mod,
2304 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2305 ret = __btrfs_free_extent(trans, fs_info, node, parent,
2306 ref_root, ref->objectid,
2307 ref->offset, node->ref_mod,
2315 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2316 struct extent_buffer *leaf,
2317 struct btrfs_extent_item *ei)
2319 u64 flags = btrfs_extent_flags(leaf, ei);
2320 if (extent_op->update_flags) {
2321 flags |= extent_op->flags_to_set;
2322 btrfs_set_extent_flags(leaf, ei, flags);
2325 if (extent_op->update_key) {
2326 struct btrfs_tree_block_info *bi;
2327 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2328 bi = (struct btrfs_tree_block_info *)(ei + 1);
2329 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2333 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2334 struct btrfs_fs_info *fs_info,
2335 struct btrfs_delayed_ref_head *head,
2336 struct btrfs_delayed_extent_op *extent_op)
2338 struct btrfs_key key;
2339 struct btrfs_path *path;
2340 struct btrfs_extent_item *ei;
2341 struct extent_buffer *leaf;
2345 int metadata = !extent_op->is_data;
2350 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2353 path = btrfs_alloc_path();
2357 key.objectid = head->bytenr;
2360 key.type = BTRFS_METADATA_ITEM_KEY;
2361 key.offset = extent_op->level;
2363 key.type = BTRFS_EXTENT_ITEM_KEY;
2364 key.offset = head->num_bytes;
2368 path->reada = READA_FORWARD;
2369 path->leave_spinning = 1;
2370 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1);
2377 if (path->slots[0] > 0) {
2379 btrfs_item_key_to_cpu(path->nodes[0], &key,
2381 if (key.objectid == head->bytenr &&
2382 key.type == BTRFS_EXTENT_ITEM_KEY &&
2383 key.offset == head->num_bytes)
2387 btrfs_release_path(path);
2390 key.objectid = head->bytenr;
2391 key.offset = head->num_bytes;
2392 key.type = BTRFS_EXTENT_ITEM_KEY;
2401 leaf = path->nodes[0];
2402 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2403 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2404 if (item_size < sizeof(*ei)) {
2405 ret = convert_extent_item_v0(trans, fs_info, path, (u64)-1, 0);
2410 leaf = path->nodes[0];
2411 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2414 BUG_ON(item_size < sizeof(*ei));
2415 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2416 __run_delayed_extent_op(extent_op, leaf, ei);
2418 btrfs_mark_buffer_dirty(leaf);
2420 btrfs_free_path(path);
2424 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2425 struct btrfs_fs_info *fs_info,
2426 struct btrfs_delayed_ref_node *node,
2427 struct btrfs_delayed_extent_op *extent_op,
2428 int insert_reserved)
2431 struct btrfs_delayed_tree_ref *ref;
2435 ref = btrfs_delayed_node_to_tree_ref(node);
2436 trace_run_delayed_tree_ref(fs_info, node, ref, node->action);
2438 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2439 parent = ref->parent;
2440 ref_root = ref->root;
2442 if (node->ref_mod != 1) {
2444 "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu",
2445 node->bytenr, node->ref_mod, node->action, ref_root,
2449 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2450 BUG_ON(!extent_op || !extent_op->update_flags);
2451 ret = alloc_reserved_tree_block(trans, node, extent_op);
2452 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2453 ret = __btrfs_inc_extent_ref(trans, fs_info, node,
2457 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2458 ret = __btrfs_free_extent(trans, fs_info, node,
2460 ref->level, 0, 1, extent_op);
2467 /* helper function to actually process a single delayed ref entry */
2468 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2469 struct btrfs_fs_info *fs_info,
2470 struct btrfs_delayed_ref_node *node,
2471 struct btrfs_delayed_extent_op *extent_op,
2472 int insert_reserved)
2476 if (trans->aborted) {
2477 if (insert_reserved)
2478 btrfs_pin_extent(fs_info, node->bytenr,
2479 node->num_bytes, 1);
2483 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2484 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2485 ret = run_delayed_tree_ref(trans, fs_info, node, extent_op,
2487 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2488 node->type == BTRFS_SHARED_DATA_REF_KEY)
2489 ret = run_delayed_data_ref(trans, fs_info, node, extent_op,
2496 static inline struct btrfs_delayed_ref_node *
2497 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2499 struct btrfs_delayed_ref_node *ref;
2501 if (RB_EMPTY_ROOT(&head->ref_tree))
2505 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2506 * This is to prevent a ref count from going down to zero, which deletes
2507 * the extent item from the extent tree, when there still are references
2508 * to add, which would fail because they would not find the extent item.
2510 if (!list_empty(&head->ref_add_list))
2511 return list_first_entry(&head->ref_add_list,
2512 struct btrfs_delayed_ref_node, add_list);
2514 ref = rb_entry(rb_first(&head->ref_tree),
2515 struct btrfs_delayed_ref_node, ref_node);
2516 ASSERT(list_empty(&ref->add_list));
2520 static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
2521 struct btrfs_delayed_ref_head *head)
2523 spin_lock(&delayed_refs->lock);
2524 head->processing = 0;
2525 delayed_refs->num_heads_ready++;
2526 spin_unlock(&delayed_refs->lock);
2527 btrfs_delayed_ref_unlock(head);
2530 static int cleanup_extent_op(struct btrfs_trans_handle *trans,
2531 struct btrfs_fs_info *fs_info,
2532 struct btrfs_delayed_ref_head *head)
2534 struct btrfs_delayed_extent_op *extent_op = head->extent_op;
2539 head->extent_op = NULL;
2540 if (head->must_insert_reserved) {
2541 btrfs_free_delayed_extent_op(extent_op);
2544 spin_unlock(&head->lock);
2545 ret = run_delayed_extent_op(trans, fs_info, head, extent_op);
2546 btrfs_free_delayed_extent_op(extent_op);
2547 return ret ? ret : 1;
2550 static int cleanup_ref_head(struct btrfs_trans_handle *trans,
2551 struct btrfs_fs_info *fs_info,
2552 struct btrfs_delayed_ref_head *head)
2554 struct btrfs_delayed_ref_root *delayed_refs;
2557 delayed_refs = &trans->transaction->delayed_refs;
2559 ret = cleanup_extent_op(trans, fs_info, head);
2561 unselect_delayed_ref_head(delayed_refs, head);
2562 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2569 * Need to drop our head ref lock and re-acquire the delayed ref lock
2570 * and then re-check to make sure nobody got added.
2572 spin_unlock(&head->lock);
2573 spin_lock(&delayed_refs->lock);
2574 spin_lock(&head->lock);
2575 if (!RB_EMPTY_ROOT(&head->ref_tree) || head->extent_op) {
2576 spin_unlock(&head->lock);
2577 spin_unlock(&delayed_refs->lock);
2580 delayed_refs->num_heads--;
2581 rb_erase(&head->href_node, &delayed_refs->href_root);
2582 RB_CLEAR_NODE(&head->href_node);
2583 spin_unlock(&head->lock);
2584 spin_unlock(&delayed_refs->lock);
2585 atomic_dec(&delayed_refs->num_entries);
2587 trace_run_delayed_ref_head(fs_info, head, 0);
2589 if (head->total_ref_mod < 0) {
2590 struct btrfs_space_info *space_info;
2594 flags = BTRFS_BLOCK_GROUP_DATA;
2595 else if (head->is_system)
2596 flags = BTRFS_BLOCK_GROUP_SYSTEM;
2598 flags = BTRFS_BLOCK_GROUP_METADATA;
2599 space_info = __find_space_info(fs_info, flags);
2601 percpu_counter_add(&space_info->total_bytes_pinned,
2604 if (head->is_data) {
2605 spin_lock(&delayed_refs->lock);
2606 delayed_refs->pending_csums -= head->num_bytes;
2607 spin_unlock(&delayed_refs->lock);
2611 if (head->must_insert_reserved) {
2612 btrfs_pin_extent(fs_info, head->bytenr,
2613 head->num_bytes, 1);
2614 if (head->is_data) {
2615 ret = btrfs_del_csums(trans, fs_info, head->bytenr,
2620 /* Also free its reserved qgroup space */
2621 btrfs_qgroup_free_delayed_ref(fs_info, head->qgroup_ref_root,
2622 head->qgroup_reserved);
2623 btrfs_delayed_ref_unlock(head);
2624 btrfs_put_delayed_ref_head(head);
2629 * Returns 0 on success or if called with an already aborted transaction.
2630 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2632 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2635 struct btrfs_fs_info *fs_info = trans->fs_info;
2636 struct btrfs_delayed_ref_root *delayed_refs;
2637 struct btrfs_delayed_ref_node *ref;
2638 struct btrfs_delayed_ref_head *locked_ref = NULL;
2639 struct btrfs_delayed_extent_op *extent_op;
2640 ktime_t start = ktime_get();
2642 unsigned long count = 0;
2643 unsigned long actual_count = 0;
2644 int must_insert_reserved = 0;
2646 delayed_refs = &trans->transaction->delayed_refs;
2652 spin_lock(&delayed_refs->lock);
2653 locked_ref = btrfs_select_ref_head(trans);
2655 spin_unlock(&delayed_refs->lock);
2659 /* grab the lock that says we are going to process
2660 * all the refs for this head */
2661 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2662 spin_unlock(&delayed_refs->lock);
2664 * we may have dropped the spin lock to get the head
2665 * mutex lock, and that might have given someone else
2666 * time to free the head. If that's true, it has been
2667 * removed from our list and we can move on.
2669 if (ret == -EAGAIN) {
2677 * We need to try and merge add/drops of the same ref since we
2678 * can run into issues with relocate dropping the implicit ref
2679 * and then it being added back again before the drop can
2680 * finish. If we merged anything we need to re-loop so we can
2682 * Or we can get node references of the same type that weren't
2683 * merged when created due to bumps in the tree mod seq, and
2684 * we need to merge them to prevent adding an inline extent
2685 * backref before dropping it (triggering a BUG_ON at
2686 * insert_inline_extent_backref()).
2688 spin_lock(&locked_ref->lock);
2689 btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref);
2691 ref = select_delayed_ref(locked_ref);
2693 if (ref && ref->seq &&
2694 btrfs_check_delayed_seq(fs_info, ref->seq)) {
2695 spin_unlock(&locked_ref->lock);
2696 unselect_delayed_ref_head(delayed_refs, locked_ref);
2704 * We're done processing refs in this ref_head, clean everything
2705 * up and move on to the next ref_head.
2708 ret = cleanup_ref_head(trans, fs_info, locked_ref);
2710 /* We dropped our lock, we need to loop. */
2723 rb_erase(&ref->ref_node, &locked_ref->ref_tree);
2724 RB_CLEAR_NODE(&ref->ref_node);
2725 if (!list_empty(&ref->add_list))
2726 list_del(&ref->add_list);
2728 * When we play the delayed ref, also correct the ref_mod on
2731 switch (ref->action) {
2732 case BTRFS_ADD_DELAYED_REF:
2733 case BTRFS_ADD_DELAYED_EXTENT:
2734 locked_ref->ref_mod -= ref->ref_mod;
2736 case BTRFS_DROP_DELAYED_REF:
2737 locked_ref->ref_mod += ref->ref_mod;
2742 atomic_dec(&delayed_refs->num_entries);
2745 * Record the must-insert_reserved flag before we drop the spin
2748 must_insert_reserved = locked_ref->must_insert_reserved;
2749 locked_ref->must_insert_reserved = 0;
2751 extent_op = locked_ref->extent_op;
2752 locked_ref->extent_op = NULL;
2753 spin_unlock(&locked_ref->lock);
2755 ret = run_one_delayed_ref(trans, fs_info, ref, extent_op,
2756 must_insert_reserved);
2758 btrfs_free_delayed_extent_op(extent_op);
2760 unselect_delayed_ref_head(delayed_refs, locked_ref);
2761 btrfs_put_delayed_ref(ref);
2762 btrfs_debug(fs_info, "run_one_delayed_ref returned %d",
2767 btrfs_put_delayed_ref(ref);
2773 * We don't want to include ref heads since we can have empty ref heads
2774 * and those will drastically skew our runtime down since we just do
2775 * accounting, no actual extent tree updates.
2777 if (actual_count > 0) {
2778 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2782 * We weigh the current average higher than our current runtime
2783 * to avoid large swings in the average.
2785 spin_lock(&delayed_refs->lock);
2786 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2787 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2788 spin_unlock(&delayed_refs->lock);
2793 #ifdef SCRAMBLE_DELAYED_REFS
2795 * Normally delayed refs get processed in ascending bytenr order. This
2796 * correlates in most cases to the order added. To expose dependencies on this
2797 * order, we start to process the tree in the middle instead of the beginning
2799 static u64 find_middle(struct rb_root *root)
2801 struct rb_node *n = root->rb_node;
2802 struct btrfs_delayed_ref_node *entry;
2805 u64 first = 0, last = 0;
2809 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2810 first = entry->bytenr;
2814 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2815 last = entry->bytenr;
2820 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2821 WARN_ON(!entry->in_tree);
2823 middle = entry->bytenr;
2836 static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads)
2840 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2841 sizeof(struct btrfs_extent_inline_ref));
2842 if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2843 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2846 * We don't ever fill up leaves all the way so multiply by 2 just to be
2847 * closer to what we're really going to want to use.
2849 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info));
2853 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2854 * would require to store the csums for that many bytes.
2856 u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes)
2859 u64 num_csums_per_leaf;
2862 csum_size = BTRFS_MAX_ITEM_SIZE(fs_info);
2863 num_csums_per_leaf = div64_u64(csum_size,
2864 (u64)btrfs_super_csum_size(fs_info->super_copy));
2865 num_csums = div64_u64(csum_bytes, fs_info->sectorsize);
2866 num_csums += num_csums_per_leaf - 1;
2867 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2871 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2872 struct btrfs_fs_info *fs_info)
2874 struct btrfs_block_rsv *global_rsv;
2875 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2876 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2877 unsigned int num_dirty_bgs = trans->transaction->num_dirty_bgs;
2878 u64 num_bytes, num_dirty_bgs_bytes;
2881 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
2882 num_heads = heads_to_leaves(fs_info, num_heads);
2884 num_bytes += (num_heads - 1) * fs_info->nodesize;
2886 num_bytes += btrfs_csum_bytes_to_leaves(fs_info, csum_bytes) *
2888 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(fs_info,
2890 global_rsv = &fs_info->global_block_rsv;
2893 * If we can't allocate any more chunks lets make sure we have _lots_ of
2894 * wiggle room since running delayed refs can create more delayed refs.
2896 if (global_rsv->space_info->full) {
2897 num_dirty_bgs_bytes <<= 1;
2901 spin_lock(&global_rsv->lock);
2902 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2904 spin_unlock(&global_rsv->lock);
2908 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2909 struct btrfs_fs_info *fs_info)
2912 atomic_read(&trans->transaction->delayed_refs.num_entries);
2917 avg_runtime = fs_info->avg_delayed_ref_runtime;
2918 val = num_entries * avg_runtime;
2919 if (val >= NSEC_PER_SEC)
2921 if (val >= NSEC_PER_SEC / 2)
2924 return btrfs_check_space_for_delayed_refs(trans, fs_info);
2927 struct async_delayed_refs {
2928 struct btrfs_root *root;
2933 struct completion wait;
2934 struct btrfs_work work;
2937 static inline struct async_delayed_refs *
2938 to_async_delayed_refs(struct btrfs_work *work)
2940 return container_of(work, struct async_delayed_refs, work);
2943 static void delayed_ref_async_start(struct btrfs_work *work)
2945 struct async_delayed_refs *async = to_async_delayed_refs(work);
2946 struct btrfs_trans_handle *trans;
2947 struct btrfs_fs_info *fs_info = async->root->fs_info;
2950 /* if the commit is already started, we don't need to wait here */
2951 if (btrfs_transaction_blocked(fs_info))
2954 trans = btrfs_join_transaction(async->root);
2955 if (IS_ERR(trans)) {
2956 async->error = PTR_ERR(trans);
2961 * trans->sync means that when we call end_transaction, we won't
2962 * wait on delayed refs
2966 /* Don't bother flushing if we got into a different transaction */
2967 if (trans->transid > async->transid)
2970 ret = btrfs_run_delayed_refs(trans, async->count);
2974 ret = btrfs_end_transaction(trans);
2975 if (ret && !async->error)
2979 complete(&async->wait);
2984 int btrfs_async_run_delayed_refs(struct btrfs_fs_info *fs_info,
2985 unsigned long count, u64 transid, int wait)
2987 struct async_delayed_refs *async;
2990 async = kmalloc(sizeof(*async), GFP_NOFS);
2994 async->root = fs_info->tree_root;
2995 async->count = count;
2997 async->transid = transid;
3002 init_completion(&async->wait);
3004 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
3005 delayed_ref_async_start, NULL, NULL);
3007 btrfs_queue_work(fs_info->extent_workers, &async->work);
3010 wait_for_completion(&async->wait);
3019 * this starts processing the delayed reference count updates and
3020 * extent insertions we have queued up so far. count can be
3021 * 0, which means to process everything in the tree at the start
3022 * of the run (but not newly added entries), or it can be some target
3023 * number you'd like to process.
3025 * Returns 0 on success or if called with an aborted transaction
3026 * Returns <0 on error and aborts the transaction
3028 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
3029 unsigned long count)
3031 struct btrfs_fs_info *fs_info = trans->fs_info;
3032 struct rb_node *node;
3033 struct btrfs_delayed_ref_root *delayed_refs;
3034 struct btrfs_delayed_ref_head *head;
3036 int run_all = count == (unsigned long)-1;
3037 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
3039 /* We'll clean this up in btrfs_cleanup_transaction */
3043 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
3046 delayed_refs = &trans->transaction->delayed_refs;
3048 count = atomic_read(&delayed_refs->num_entries) * 2;
3051 #ifdef SCRAMBLE_DELAYED_REFS
3052 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
3054 trans->can_flush_pending_bgs = false;
3055 ret = __btrfs_run_delayed_refs(trans, count);
3057 btrfs_abort_transaction(trans, ret);
3062 if (!list_empty(&trans->new_bgs))
3063 btrfs_create_pending_block_groups(trans);
3065 spin_lock(&delayed_refs->lock);
3066 node = rb_first(&delayed_refs->href_root);
3068 spin_unlock(&delayed_refs->lock);
3071 head = rb_entry(node, struct btrfs_delayed_ref_head,
3073 refcount_inc(&head->refs);
3074 spin_unlock(&delayed_refs->lock);
3076 /* Mutex was contended, block until it's released and retry. */
3077 mutex_lock(&head->mutex);
3078 mutex_unlock(&head->mutex);
3080 btrfs_put_delayed_ref_head(head);
3085 trans->can_flush_pending_bgs = can_flush_pending_bgs;
3089 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
3090 struct btrfs_fs_info *fs_info,
3091 u64 bytenr, u64 num_bytes, u64 flags,
3092 int level, int is_data)
3094 struct btrfs_delayed_extent_op *extent_op;
3097 extent_op = btrfs_alloc_delayed_extent_op();
3101 extent_op->flags_to_set = flags;
3102 extent_op->update_flags = true;
3103 extent_op->update_key = false;
3104 extent_op->is_data = is_data ? true : false;
3105 extent_op->level = level;
3107 ret = btrfs_add_delayed_extent_op(fs_info, trans, bytenr,
3108 num_bytes, extent_op);
3110 btrfs_free_delayed_extent_op(extent_op);
3114 static noinline int check_delayed_ref(struct btrfs_root *root,
3115 struct btrfs_path *path,
3116 u64 objectid, u64 offset, u64 bytenr)
3118 struct btrfs_delayed_ref_head *head;
3119 struct btrfs_delayed_ref_node *ref;
3120 struct btrfs_delayed_data_ref *data_ref;
3121 struct btrfs_delayed_ref_root *delayed_refs;
3122 struct btrfs_transaction *cur_trans;
3123 struct rb_node *node;
3126 spin_lock(&root->fs_info->trans_lock);
3127 cur_trans = root->fs_info->running_transaction;
3129 refcount_inc(&cur_trans->use_count);
3130 spin_unlock(&root->fs_info->trans_lock);
3134 delayed_refs = &cur_trans->delayed_refs;
3135 spin_lock(&delayed_refs->lock);
3136 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
3138 spin_unlock(&delayed_refs->lock);
3139 btrfs_put_transaction(cur_trans);
3143 if (!mutex_trylock(&head->mutex)) {
3144 refcount_inc(&head->refs);
3145 spin_unlock(&delayed_refs->lock);
3147 btrfs_release_path(path);
3150 * Mutex was contended, block until it's released and let
3153 mutex_lock(&head->mutex);
3154 mutex_unlock(&head->mutex);
3155 btrfs_put_delayed_ref_head(head);
3156 btrfs_put_transaction(cur_trans);
3159 spin_unlock(&delayed_refs->lock);
3161 spin_lock(&head->lock);
3163 * XXX: We should replace this with a proper search function in the
3166 for (node = rb_first(&head->ref_tree); node; node = rb_next(node)) {
3167 ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node);
3168 /* If it's a shared ref we know a cross reference exists */
3169 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
3174 data_ref = btrfs_delayed_node_to_data_ref(ref);
3177 * If our ref doesn't match the one we're currently looking at
3178 * then we have a cross reference.
3180 if (data_ref->root != root->root_key.objectid ||
3181 data_ref->objectid != objectid ||
3182 data_ref->offset != offset) {
3187 spin_unlock(&head->lock);
3188 mutex_unlock(&head->mutex);
3189 btrfs_put_transaction(cur_trans);
3193 static noinline int check_committed_ref(struct btrfs_root *root,
3194 struct btrfs_path *path,
3195 u64 objectid, u64 offset, u64 bytenr)
3197 struct btrfs_fs_info *fs_info = root->fs_info;
3198 struct btrfs_root *extent_root = fs_info->extent_root;
3199 struct extent_buffer *leaf;
3200 struct btrfs_extent_data_ref *ref;
3201 struct btrfs_extent_inline_ref *iref;
3202 struct btrfs_extent_item *ei;
3203 struct btrfs_key key;
3208 key.objectid = bytenr;
3209 key.offset = (u64)-1;
3210 key.type = BTRFS_EXTENT_ITEM_KEY;
3212 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3215 BUG_ON(ret == 0); /* Corruption */
3218 if (path->slots[0] == 0)
3222 leaf = path->nodes[0];
3223 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3225 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3229 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3230 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3231 if (item_size < sizeof(*ei)) {
3232 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3236 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3238 if (item_size != sizeof(*ei) +
3239 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3242 if (btrfs_extent_generation(leaf, ei) <=
3243 btrfs_root_last_snapshot(&root->root_item))
3246 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3248 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
3249 if (type != BTRFS_EXTENT_DATA_REF_KEY)
3252 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3253 if (btrfs_extent_refs(leaf, ei) !=
3254 btrfs_extent_data_ref_count(leaf, ref) ||
3255 btrfs_extent_data_ref_root(leaf, ref) !=
3256 root->root_key.objectid ||
3257 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3258 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3266 int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
3269 struct btrfs_path *path;
3273 path = btrfs_alloc_path();
3278 ret = check_committed_ref(root, path, objectid,
3280 if (ret && ret != -ENOENT)
3283 ret2 = check_delayed_ref(root, path, objectid,
3285 } while (ret2 == -EAGAIN);
3287 if (ret2 && ret2 != -ENOENT) {
3292 if (ret != -ENOENT || ret2 != -ENOENT)
3295 btrfs_free_path(path);
3296 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3301 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3302 struct btrfs_root *root,
3303 struct extent_buffer *buf,
3304 int full_backref, int inc)
3306 struct btrfs_fs_info *fs_info = root->fs_info;
3312 struct btrfs_key key;
3313 struct btrfs_file_extent_item *fi;
3317 int (*process_func)(struct btrfs_trans_handle *,
3318 struct btrfs_root *,
3319 u64, u64, u64, u64, u64, u64);
3322 if (btrfs_is_testing(fs_info))
3325 ref_root = btrfs_header_owner(buf);
3326 nritems = btrfs_header_nritems(buf);
3327 level = btrfs_header_level(buf);
3329 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3333 process_func = btrfs_inc_extent_ref;
3335 process_func = btrfs_free_extent;
3338 parent = buf->start;
3342 for (i = 0; i < nritems; i++) {
3344 btrfs_item_key_to_cpu(buf, &key, i);
3345 if (key.type != BTRFS_EXTENT_DATA_KEY)
3347 fi = btrfs_item_ptr(buf, i,
3348 struct btrfs_file_extent_item);
3349 if (btrfs_file_extent_type(buf, fi) ==
3350 BTRFS_FILE_EXTENT_INLINE)
3352 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3356 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3357 key.offset -= btrfs_file_extent_offset(buf, fi);
3358 ret = process_func(trans, root, bytenr, num_bytes,
3359 parent, ref_root, key.objectid,
3364 bytenr = btrfs_node_blockptr(buf, i);
3365 num_bytes = fs_info->nodesize;
3366 ret = process_func(trans, root, bytenr, num_bytes,
3367 parent, ref_root, level - 1, 0);
3377 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3378 struct extent_buffer *buf, int full_backref)
3380 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3383 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3384 struct extent_buffer *buf, int full_backref)
3386 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3389 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3390 struct btrfs_fs_info *fs_info,
3391 struct btrfs_path *path,
3392 struct btrfs_block_group_cache *cache)
3395 struct btrfs_root *extent_root = fs_info->extent_root;
3397 struct extent_buffer *leaf;
3399 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3406 leaf = path->nodes[0];
3407 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3408 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3409 btrfs_mark_buffer_dirty(leaf);
3411 btrfs_release_path(path);
3416 static struct btrfs_block_group_cache *
3417 next_block_group(struct btrfs_fs_info *fs_info,
3418 struct btrfs_block_group_cache *cache)
3420 struct rb_node *node;
3422 spin_lock(&fs_info->block_group_cache_lock);
3424 /* If our block group was removed, we need a full search. */
3425 if (RB_EMPTY_NODE(&cache->cache_node)) {
3426 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3428 spin_unlock(&fs_info->block_group_cache_lock);
3429 btrfs_put_block_group(cache);
3430 cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache;
3432 node = rb_next(&cache->cache_node);
3433 btrfs_put_block_group(cache);
3435 cache = rb_entry(node, struct btrfs_block_group_cache,
3437 btrfs_get_block_group(cache);
3440 spin_unlock(&fs_info->block_group_cache_lock);
3444 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3445 struct btrfs_trans_handle *trans,
3446 struct btrfs_path *path)
3448 struct btrfs_fs_info *fs_info = block_group->fs_info;
3449 struct btrfs_root *root = fs_info->tree_root;
3450 struct inode *inode = NULL;
3451 struct extent_changeset *data_reserved = NULL;
3453 int dcs = BTRFS_DC_ERROR;
3459 * If this block group is smaller than 100 megs don't bother caching the
3462 if (block_group->key.offset < (100 * SZ_1M)) {
3463 spin_lock(&block_group->lock);
3464 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3465 spin_unlock(&block_group->lock);
3472 inode = lookup_free_space_inode(fs_info, block_group, path);
3473 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3474 ret = PTR_ERR(inode);
3475 btrfs_release_path(path);
3479 if (IS_ERR(inode)) {
3483 if (block_group->ro)
3486 ret = create_free_space_inode(fs_info, trans, block_group,
3494 * We want to set the generation to 0, that way if anything goes wrong
3495 * from here on out we know not to trust this cache when we load up next
3498 BTRFS_I(inode)->generation = 0;
3499 ret = btrfs_update_inode(trans, root, inode);
3502 * So theoretically we could recover from this, simply set the
3503 * super cache generation to 0 so we know to invalidate the
3504 * cache, but then we'd have to keep track of the block groups
3505 * that fail this way so we know we _have_ to reset this cache
3506 * before the next commit or risk reading stale cache. So to
3507 * limit our exposure to horrible edge cases lets just abort the
3508 * transaction, this only happens in really bad situations
3511 btrfs_abort_transaction(trans, ret);
3516 /* We've already setup this transaction, go ahead and exit */
3517 if (block_group->cache_generation == trans->transid &&
3518 i_size_read(inode)) {
3519 dcs = BTRFS_DC_SETUP;
3523 if (i_size_read(inode) > 0) {
3524 ret = btrfs_check_trunc_cache_free_space(fs_info,
3525 &fs_info->global_block_rsv);
3529 ret = btrfs_truncate_free_space_cache(trans, NULL, inode);
3534 spin_lock(&block_group->lock);
3535 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3536 !btrfs_test_opt(fs_info, SPACE_CACHE)) {
3538 * don't bother trying to write stuff out _if_
3539 * a) we're not cached,
3540 * b) we're with nospace_cache mount option,
3541 * c) we're with v2 space_cache (FREE_SPACE_TREE).
3543 dcs = BTRFS_DC_WRITTEN;
3544 spin_unlock(&block_group->lock);
3547 spin_unlock(&block_group->lock);
3550 * We hit an ENOSPC when setting up the cache in this transaction, just
3551 * skip doing the setup, we've already cleared the cache so we're safe.
3553 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
3559 * Try to preallocate enough space based on how big the block group is.
3560 * Keep in mind this has to include any pinned space which could end up
3561 * taking up quite a bit since it's not folded into the other space
3564 num_pages = div_u64(block_group->key.offset, SZ_256M);
3569 num_pages *= PAGE_SIZE;
3571 ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages);
3575 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3576 num_pages, num_pages,
3579 * Our cache requires contiguous chunks so that we don't modify a bunch
3580 * of metadata or split extents when writing the cache out, which means
3581 * we can enospc if we are heavily fragmented in addition to just normal
3582 * out of space conditions. So if we hit this just skip setting up any
3583 * other block groups for this transaction, maybe we'll unpin enough
3584 * space the next time around.
3587 dcs = BTRFS_DC_SETUP;
3588 else if (ret == -ENOSPC)
3589 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
3594 btrfs_release_path(path);
3596 spin_lock(&block_group->lock);
3597 if (!ret && dcs == BTRFS_DC_SETUP)
3598 block_group->cache_generation = trans->transid;
3599 block_group->disk_cache_state = dcs;
3600 spin_unlock(&block_group->lock);
3602 extent_changeset_free(data_reserved);
3606 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3607 struct btrfs_fs_info *fs_info)
3609 struct btrfs_block_group_cache *cache, *tmp;
3610 struct btrfs_transaction *cur_trans = trans->transaction;
3611 struct btrfs_path *path;
3613 if (list_empty(&cur_trans->dirty_bgs) ||
3614 !btrfs_test_opt(fs_info, SPACE_CACHE))
3617 path = btrfs_alloc_path();
3621 /* Could add new block groups, use _safe just in case */
3622 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3624 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3625 cache_save_setup(cache, trans, path);
3628 btrfs_free_path(path);
3633 * transaction commit does final block group cache writeback during a
3634 * critical section where nothing is allowed to change the FS. This is
3635 * required in order for the cache to actually match the block group,
3636 * but can introduce a lot of latency into the commit.
3638 * So, btrfs_start_dirty_block_groups is here to kick off block group
3639 * cache IO. There's a chance we'll have to redo some of it if the
3640 * block group changes again during the commit, but it greatly reduces
3641 * the commit latency by getting rid of the easy block groups while
3642 * we're still allowing others to join the commit.
3644 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans)
3646 struct btrfs_fs_info *fs_info = trans->fs_info;
3647 struct btrfs_block_group_cache *cache;
3648 struct btrfs_transaction *cur_trans = trans->transaction;
3651 struct btrfs_path *path = NULL;
3653 struct list_head *io = &cur_trans->io_bgs;
3654 int num_started = 0;
3657 spin_lock(&cur_trans->dirty_bgs_lock);
3658 if (list_empty(&cur_trans->dirty_bgs)) {
3659 spin_unlock(&cur_trans->dirty_bgs_lock);
3662 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3663 spin_unlock(&cur_trans->dirty_bgs_lock);
3667 * make sure all the block groups on our dirty list actually
3670 btrfs_create_pending_block_groups(trans);
3673 path = btrfs_alloc_path();
3679 * cache_write_mutex is here only to save us from balance or automatic
3680 * removal of empty block groups deleting this block group while we are
3681 * writing out the cache
3683 mutex_lock(&trans->transaction->cache_write_mutex);
3684 while (!list_empty(&dirty)) {
3685 cache = list_first_entry(&dirty,
3686 struct btrfs_block_group_cache,
3689 * this can happen if something re-dirties a block
3690 * group that is already under IO. Just wait for it to
3691 * finish and then do it all again
3693 if (!list_empty(&cache->io_list)) {
3694 list_del_init(&cache->io_list);
3695 btrfs_wait_cache_io(trans, cache, path);
3696 btrfs_put_block_group(cache);
3701 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3702 * if it should update the cache_state. Don't delete
3703 * until after we wait.
3705 * Since we're not running in the commit critical section
3706 * we need the dirty_bgs_lock to protect from update_block_group
3708 spin_lock(&cur_trans->dirty_bgs_lock);
3709 list_del_init(&cache->dirty_list);
3710 spin_unlock(&cur_trans->dirty_bgs_lock);
3714 cache_save_setup(cache, trans, path);
3716 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3717 cache->io_ctl.inode = NULL;
3718 ret = btrfs_write_out_cache(fs_info, trans,
3720 if (ret == 0 && cache->io_ctl.inode) {
3725 * The cache_write_mutex is protecting the
3726 * io_list, also refer to the definition of
3727 * btrfs_transaction::io_bgs for more details
3729 list_add_tail(&cache->io_list, io);
3732 * if we failed to write the cache, the
3733 * generation will be bad and life goes on
3739 ret = write_one_cache_group(trans, fs_info,
3742 * Our block group might still be attached to the list
3743 * of new block groups in the transaction handle of some
3744 * other task (struct btrfs_trans_handle->new_bgs). This
3745 * means its block group item isn't yet in the extent
3746 * tree. If this happens ignore the error, as we will
3747 * try again later in the critical section of the
3748 * transaction commit.
3750 if (ret == -ENOENT) {
3752 spin_lock(&cur_trans->dirty_bgs_lock);
3753 if (list_empty(&cache->dirty_list)) {
3754 list_add_tail(&cache->dirty_list,
3755 &cur_trans->dirty_bgs);
3756 btrfs_get_block_group(cache);
3758 spin_unlock(&cur_trans->dirty_bgs_lock);
3760 btrfs_abort_transaction(trans, ret);
3764 /* if its not on the io list, we need to put the block group */
3766 btrfs_put_block_group(cache);
3772 * Avoid blocking other tasks for too long. It might even save
3773 * us from writing caches for block groups that are going to be
3776 mutex_unlock(&trans->transaction->cache_write_mutex);
3777 mutex_lock(&trans->transaction->cache_write_mutex);
3779 mutex_unlock(&trans->transaction->cache_write_mutex);
3782 * go through delayed refs for all the stuff we've just kicked off
3783 * and then loop back (just once)
3785 ret = btrfs_run_delayed_refs(trans, 0);
3786 if (!ret && loops == 0) {
3788 spin_lock(&cur_trans->dirty_bgs_lock);
3789 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3791 * dirty_bgs_lock protects us from concurrent block group
3792 * deletes too (not just cache_write_mutex).
3794 if (!list_empty(&dirty)) {
3795 spin_unlock(&cur_trans->dirty_bgs_lock);
3798 spin_unlock(&cur_trans->dirty_bgs_lock);
3799 } else if (ret < 0) {
3800 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
3803 btrfs_free_path(path);
3807 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3808 struct btrfs_fs_info *fs_info)
3810 struct btrfs_block_group_cache *cache;
3811 struct btrfs_transaction *cur_trans = trans->transaction;
3814 struct btrfs_path *path;
3815 struct list_head *io = &cur_trans->io_bgs;
3816 int num_started = 0;
3818 path = btrfs_alloc_path();
3823 * Even though we are in the critical section of the transaction commit,
3824 * we can still have concurrent tasks adding elements to this
3825 * transaction's list of dirty block groups. These tasks correspond to
3826 * endio free space workers started when writeback finishes for a
3827 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3828 * allocate new block groups as a result of COWing nodes of the root
3829 * tree when updating the free space inode. The writeback for the space
3830 * caches is triggered by an earlier call to
3831 * btrfs_start_dirty_block_groups() and iterations of the following
3833 * Also we want to do the cache_save_setup first and then run the
3834 * delayed refs to make sure we have the best chance at doing this all
3837 spin_lock(&cur_trans->dirty_bgs_lock);
3838 while (!list_empty(&cur_trans->dirty_bgs)) {
3839 cache = list_first_entry(&cur_trans->dirty_bgs,
3840 struct btrfs_block_group_cache,
3844 * this can happen if cache_save_setup re-dirties a block
3845 * group that is already under IO. Just wait for it to
3846 * finish and then do it all again
3848 if (!list_empty(&cache->io_list)) {
3849 spin_unlock(&cur_trans->dirty_bgs_lock);
3850 list_del_init(&cache->io_list);
3851 btrfs_wait_cache_io(trans, cache, path);
3852 btrfs_put_block_group(cache);
3853 spin_lock(&cur_trans->dirty_bgs_lock);
3857 * don't remove from the dirty list until after we've waited
3860 list_del_init(&cache->dirty_list);
3861 spin_unlock(&cur_trans->dirty_bgs_lock);
3864 cache_save_setup(cache, trans, path);
3867 ret = btrfs_run_delayed_refs(trans,
3868 (unsigned long) -1);
3870 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3871 cache->io_ctl.inode = NULL;
3872 ret = btrfs_write_out_cache(fs_info, trans,
3874 if (ret == 0 && cache->io_ctl.inode) {
3877 list_add_tail(&cache->io_list, io);
3880 * if we failed to write the cache, the
3881 * generation will be bad and life goes on
3887 ret = write_one_cache_group(trans, fs_info,
3890 * One of the free space endio workers might have
3891 * created a new block group while updating a free space
3892 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3893 * and hasn't released its transaction handle yet, in
3894 * which case the new block group is still attached to
3895 * its transaction handle and its creation has not
3896 * finished yet (no block group item in the extent tree
3897 * yet, etc). If this is the case, wait for all free
3898 * space endio workers to finish and retry. This is a
3899 * a very rare case so no need for a more efficient and
3902 if (ret == -ENOENT) {
3903 wait_event(cur_trans->writer_wait,
3904 atomic_read(&cur_trans->num_writers) == 1);
3905 ret = write_one_cache_group(trans, fs_info,
3909 btrfs_abort_transaction(trans, ret);
3912 /* if its not on the io list, we need to put the block group */
3914 btrfs_put_block_group(cache);
3915 spin_lock(&cur_trans->dirty_bgs_lock);
3917 spin_unlock(&cur_trans->dirty_bgs_lock);
3920 * Refer to the definition of io_bgs member for details why it's safe
3921 * to use it without any locking
3923 while (!list_empty(io)) {
3924 cache = list_first_entry(io, struct btrfs_block_group_cache,
3926 list_del_init(&cache->io_list);
3927 btrfs_wait_cache_io(trans, cache, path);
3928 btrfs_put_block_group(cache);
3931 btrfs_free_path(path);
3935 int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
3937 struct btrfs_block_group_cache *block_group;
3940 block_group = btrfs_lookup_block_group(fs_info, bytenr);
3941 if (!block_group || block_group->ro)
3944 btrfs_put_block_group(block_group);
3948 bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3950 struct btrfs_block_group_cache *bg;
3953 bg = btrfs_lookup_block_group(fs_info, bytenr);
3957 spin_lock(&bg->lock);
3961 atomic_inc(&bg->nocow_writers);
3962 spin_unlock(&bg->lock);
3964 /* no put on block group, done by btrfs_dec_nocow_writers */
3966 btrfs_put_block_group(bg);
3972 void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3974 struct btrfs_block_group_cache *bg;
3976 bg = btrfs_lookup_block_group(fs_info, bytenr);
3978 if (atomic_dec_and_test(&bg->nocow_writers))
3979 wake_up_var(&bg->nocow_writers);
3981 * Once for our lookup and once for the lookup done by a previous call
3982 * to btrfs_inc_nocow_writers()
3984 btrfs_put_block_group(bg);
3985 btrfs_put_block_group(bg);
3988 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg)
3990 wait_var_event(&bg->nocow_writers, !atomic_read(&bg->nocow_writers));
3993 static const char *alloc_name(u64 flags)
3996 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3998 case BTRFS_BLOCK_GROUP_METADATA:
4000 case BTRFS_BLOCK_GROUP_DATA:
4002 case BTRFS_BLOCK_GROUP_SYSTEM:
4006 return "invalid-combination";
4010 static int create_space_info(struct btrfs_fs_info *info, u64 flags)
4013 struct btrfs_space_info *space_info;
4017 space_info = kzalloc(sizeof(*space_info), GFP_NOFS);
4021 ret = percpu_counter_init(&space_info->total_bytes_pinned, 0,
4028 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
4029 INIT_LIST_HEAD(&space_info->block_groups[i]);
4030 init_rwsem(&space_info->groups_sem);
4031 spin_lock_init(&space_info->lock);
4032 space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
4033 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4034 init_waitqueue_head(&space_info->wait);
4035 INIT_LIST_HEAD(&space_info->ro_bgs);
4036 INIT_LIST_HEAD(&space_info->tickets);
4037 INIT_LIST_HEAD(&space_info->priority_tickets);
4039 ret = kobject_init_and_add(&space_info->kobj, &space_info_ktype,
4040 info->space_info_kobj, "%s",
4041 alloc_name(space_info->flags));
4043 percpu_counter_destroy(&space_info->total_bytes_pinned);
4048 list_add_rcu(&space_info->list, &info->space_info);
4049 if (flags & BTRFS_BLOCK_GROUP_DATA)
4050 info->data_sinfo = space_info;
4055 static void update_space_info(struct btrfs_fs_info *info, u64 flags,
4056 u64 total_bytes, u64 bytes_used,
4058 struct btrfs_space_info **space_info)
4060 struct btrfs_space_info *found;
4063 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
4064 BTRFS_BLOCK_GROUP_RAID10))
4069 found = __find_space_info(info, flags);
4071 spin_lock(&found->lock);
4072 found->total_bytes += total_bytes;
4073 found->disk_total += total_bytes * factor;
4074 found->bytes_used += bytes_used;
4075 found->disk_used += bytes_used * factor;
4076 found->bytes_readonly += bytes_readonly;
4077 if (total_bytes > 0)
4079 space_info_add_new_bytes(info, found, total_bytes -
4080 bytes_used - bytes_readonly);
4081 spin_unlock(&found->lock);
4082 *space_info = found;
4085 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
4087 u64 extra_flags = chunk_to_extended(flags) &
4088 BTRFS_EXTENDED_PROFILE_MASK;
4090 write_seqlock(&fs_info->profiles_lock);
4091 if (flags & BTRFS_BLOCK_GROUP_DATA)
4092 fs_info->avail_data_alloc_bits |= extra_flags;
4093 if (flags & BTRFS_BLOCK_GROUP_METADATA)
4094 fs_info->avail_metadata_alloc_bits |= extra_flags;
4095 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4096 fs_info->avail_system_alloc_bits |= extra_flags;
4097 write_sequnlock(&fs_info->profiles_lock);
4101 * returns target flags in extended format or 0 if restripe for this
4102 * chunk_type is not in progress
4104 * should be called with balance_lock held
4106 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
4108 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4114 if (flags & BTRFS_BLOCK_GROUP_DATA &&
4115 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4116 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
4117 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
4118 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4119 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
4120 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
4121 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4122 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
4129 * @flags: available profiles in extended format (see ctree.h)
4131 * Returns reduced profile in chunk format. If profile changing is in
4132 * progress (either running or paused) picks the target profile (if it's
4133 * already available), otherwise falls back to plain reducing.
4135 static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags)
4137 u64 num_devices = fs_info->fs_devices->rw_devices;
4143 * see if restripe for this chunk_type is in progress, if so
4144 * try to reduce to the target profile
4146 spin_lock(&fs_info->balance_lock);
4147 target = get_restripe_target(fs_info, flags);
4149 /* pick target profile only if it's already available */
4150 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
4151 spin_unlock(&fs_info->balance_lock);
4152 return extended_to_chunk(target);
4155 spin_unlock(&fs_info->balance_lock);
4157 /* First, mask out the RAID levels which aren't possible */
4158 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4159 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
4160 allowed |= btrfs_raid_array[raid_type].bg_flag;
4164 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
4165 allowed = BTRFS_BLOCK_GROUP_RAID6;
4166 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
4167 allowed = BTRFS_BLOCK_GROUP_RAID5;
4168 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
4169 allowed = BTRFS_BLOCK_GROUP_RAID10;
4170 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
4171 allowed = BTRFS_BLOCK_GROUP_RAID1;
4172 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
4173 allowed = BTRFS_BLOCK_GROUP_RAID0;
4175 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
4177 return extended_to_chunk(flags | allowed);
4180 static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags)
4187 seq = read_seqbegin(&fs_info->profiles_lock);
4189 if (flags & BTRFS_BLOCK_GROUP_DATA)
4190 flags |= fs_info->avail_data_alloc_bits;
4191 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4192 flags |= fs_info->avail_system_alloc_bits;
4193 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
4194 flags |= fs_info->avail_metadata_alloc_bits;
4195 } while (read_seqretry(&fs_info->profiles_lock, seq));
4197 return btrfs_reduce_alloc_profile(fs_info, flags);
4200 static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
4202 struct btrfs_fs_info *fs_info = root->fs_info;
4207 flags = BTRFS_BLOCK_GROUP_DATA;
4208 else if (root == fs_info->chunk_root)
4209 flags = BTRFS_BLOCK_GROUP_SYSTEM;
4211 flags = BTRFS_BLOCK_GROUP_METADATA;
4213 ret = get_alloc_profile(fs_info, flags);
4217 u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info)
4219 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA);
4222 u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info)
4224 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4227 u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info)
4229 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4232 static u64 btrfs_space_info_used(struct btrfs_space_info *s_info,
4233 bool may_use_included)
4236 return s_info->bytes_used + s_info->bytes_reserved +
4237 s_info->bytes_pinned + s_info->bytes_readonly +
4238 (may_use_included ? s_info->bytes_may_use : 0);
4241 int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes)
4243 struct btrfs_root *root = inode->root;
4244 struct btrfs_fs_info *fs_info = root->fs_info;
4245 struct btrfs_space_info *data_sinfo = fs_info->data_sinfo;
4248 int need_commit = 2;
4249 int have_pinned_space;
4251 /* make sure bytes are sectorsize aligned */
4252 bytes = ALIGN(bytes, fs_info->sectorsize);
4254 if (btrfs_is_free_space_inode(inode)) {
4256 ASSERT(current->journal_info);
4260 /* make sure we have enough space to handle the data first */
4261 spin_lock(&data_sinfo->lock);
4262 used = btrfs_space_info_used(data_sinfo, true);
4264 if (used + bytes > data_sinfo->total_bytes) {
4265 struct btrfs_trans_handle *trans;
4268 * if we don't have enough free bytes in this space then we need
4269 * to alloc a new chunk.
4271 if (!data_sinfo->full) {
4274 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4275 spin_unlock(&data_sinfo->lock);
4277 alloc_target = btrfs_data_alloc_profile(fs_info);
4279 * It is ugly that we don't call nolock join
4280 * transaction for the free space inode case here.
4281 * But it is safe because we only do the data space
4282 * reservation for the free space cache in the
4283 * transaction context, the common join transaction
4284 * just increase the counter of the current transaction
4285 * handler, doesn't try to acquire the trans_lock of
4288 trans = btrfs_join_transaction(root);
4290 return PTR_ERR(trans);
4292 ret = do_chunk_alloc(trans, fs_info, alloc_target,
4293 CHUNK_ALLOC_NO_FORCE);
4294 btrfs_end_transaction(trans);
4299 have_pinned_space = 1;
4308 * If we don't have enough pinned space to deal with this
4309 * allocation, and no removed chunk in current transaction,
4310 * don't bother committing the transaction.
4312 have_pinned_space = percpu_counter_compare(
4313 &data_sinfo->total_bytes_pinned,
4314 used + bytes - data_sinfo->total_bytes);
4315 spin_unlock(&data_sinfo->lock);
4317 /* commit the current transaction and try again */
4322 if (need_commit > 0) {
4323 btrfs_start_delalloc_roots(fs_info, -1);
4324 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0,
4328 trans = btrfs_join_transaction(root);
4330 return PTR_ERR(trans);
4331 if (have_pinned_space >= 0 ||
4332 test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4333 &trans->transaction->flags) ||
4335 ret = btrfs_commit_transaction(trans);
4339 * The cleaner kthread might still be doing iput
4340 * operations. Wait for it to finish so that
4341 * more space is released.
4343 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
4344 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
4347 btrfs_end_transaction(trans);
4351 trace_btrfs_space_reservation(fs_info,
4352 "space_info:enospc",
4353 data_sinfo->flags, bytes, 1);
4356 data_sinfo->bytes_may_use += bytes;
4357 trace_btrfs_space_reservation(fs_info, "space_info",
4358 data_sinfo->flags, bytes, 1);
4359 spin_unlock(&data_sinfo->lock);
4364 int btrfs_check_data_free_space(struct inode *inode,
4365 struct extent_changeset **reserved, u64 start, u64 len)
4367 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4370 /* align the range */
4371 len = round_up(start + len, fs_info->sectorsize) -
4372 round_down(start, fs_info->sectorsize);
4373 start = round_down(start, fs_info->sectorsize);
4375 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len);
4379 /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */
4380 ret = btrfs_qgroup_reserve_data(inode, reserved, start, len);
4382 btrfs_free_reserved_data_space_noquota(inode, start, len);
4389 * Called if we need to clear a data reservation for this inode
4390 * Normally in a error case.
4392 * This one will *NOT* use accurate qgroup reserved space API, just for case
4393 * which we can't sleep and is sure it won't affect qgroup reserved space.
4394 * Like clear_bit_hook().
4396 void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4399 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4400 struct btrfs_space_info *data_sinfo;
4402 /* Make sure the range is aligned to sectorsize */
4403 len = round_up(start + len, fs_info->sectorsize) -
4404 round_down(start, fs_info->sectorsize);
4405 start = round_down(start, fs_info->sectorsize);
4407 data_sinfo = fs_info->data_sinfo;
4408 spin_lock(&data_sinfo->lock);
4409 if (WARN_ON(data_sinfo->bytes_may_use < len))
4410 data_sinfo->bytes_may_use = 0;
4412 data_sinfo->bytes_may_use -= len;
4413 trace_btrfs_space_reservation(fs_info, "space_info",
4414 data_sinfo->flags, len, 0);
4415 spin_unlock(&data_sinfo->lock);
4419 * Called if we need to clear a data reservation for this inode
4420 * Normally in a error case.
4422 * This one will handle the per-inode data rsv map for accurate reserved
4425 void btrfs_free_reserved_data_space(struct inode *inode,
4426 struct extent_changeset *reserved, u64 start, u64 len)
4428 struct btrfs_root *root = BTRFS_I(inode)->root;
4430 /* Make sure the range is aligned to sectorsize */
4431 len = round_up(start + len, root->fs_info->sectorsize) -
4432 round_down(start, root->fs_info->sectorsize);
4433 start = round_down(start, root->fs_info->sectorsize);
4435 btrfs_free_reserved_data_space_noquota(inode, start, len);
4436 btrfs_qgroup_free_data(inode, reserved, start, len);
4439 static void force_metadata_allocation(struct btrfs_fs_info *info)
4441 struct list_head *head = &info->space_info;
4442 struct btrfs_space_info *found;
4445 list_for_each_entry_rcu(found, head, list) {
4446 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4447 found->force_alloc = CHUNK_ALLOC_FORCE;
4452 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4454 return (global->size << 1);
4457 static int should_alloc_chunk(struct btrfs_fs_info *fs_info,
4458 struct btrfs_space_info *sinfo, int force)
4460 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4461 u64 bytes_used = btrfs_space_info_used(sinfo, false);
4464 if (force == CHUNK_ALLOC_FORCE)
4468 * We need to take into account the global rsv because for all intents
4469 * and purposes it's used space. Don't worry about locking the
4470 * global_rsv, it doesn't change except when the transaction commits.
4472 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4473 bytes_used += calc_global_rsv_need_space(global_rsv);
4476 * in limited mode, we want to have some free space up to
4477 * about 1% of the FS size.
4479 if (force == CHUNK_ALLOC_LIMITED) {
4480 thresh = btrfs_super_total_bytes(fs_info->super_copy);
4481 thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1));
4483 if (sinfo->total_bytes - bytes_used < thresh)
4487 if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8))
4492 static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type)
4496 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4497 BTRFS_BLOCK_GROUP_RAID0 |
4498 BTRFS_BLOCK_GROUP_RAID5 |
4499 BTRFS_BLOCK_GROUP_RAID6))
4500 num_dev = fs_info->fs_devices->rw_devices;
4501 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4504 num_dev = 1; /* DUP or single */
4510 * If @is_allocation is true, reserve space in the system space info necessary
4511 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4514 void check_system_chunk(struct btrfs_trans_handle *trans,
4515 struct btrfs_fs_info *fs_info, u64 type)
4517 struct btrfs_space_info *info;
4524 * Needed because we can end up allocating a system chunk and for an
4525 * atomic and race free space reservation in the chunk block reserve.
4527 lockdep_assert_held(&fs_info->chunk_mutex);
4529 info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4530 spin_lock(&info->lock);
4531 left = info->total_bytes - btrfs_space_info_used(info, true);
4532 spin_unlock(&info->lock);
4534 num_devs = get_profile_num_devs(fs_info, type);
4536 /* num_devs device items to update and 1 chunk item to add or remove */
4537 thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) +
4538 btrfs_calc_trans_metadata_size(fs_info, 1);
4540 if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4541 btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu",
4542 left, thresh, type);
4543 dump_space_info(fs_info, info, 0, 0);
4546 if (left < thresh) {
4547 u64 flags = btrfs_system_alloc_profile(fs_info);
4550 * Ignore failure to create system chunk. We might end up not
4551 * needing it, as we might not need to COW all nodes/leafs from
4552 * the paths we visit in the chunk tree (they were already COWed
4553 * or created in the current transaction for example).
4555 ret = btrfs_alloc_chunk(trans, fs_info, flags);
4559 ret = btrfs_block_rsv_add(fs_info->chunk_root,
4560 &fs_info->chunk_block_rsv,
4561 thresh, BTRFS_RESERVE_NO_FLUSH);
4563 trans->chunk_bytes_reserved += thresh;
4568 * If force is CHUNK_ALLOC_FORCE:
4569 * - return 1 if it successfully allocates a chunk,
4570 * - return errors including -ENOSPC otherwise.
4571 * If force is NOT CHUNK_ALLOC_FORCE:
4572 * - return 0 if it doesn't need to allocate a new chunk,
4573 * - return 1 if it successfully allocates a chunk,
4574 * - return errors including -ENOSPC otherwise.
4576 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4577 struct btrfs_fs_info *fs_info, u64 flags, int force)
4579 struct btrfs_space_info *space_info;
4580 int wait_for_alloc = 0;
4583 /* Don't re-enter if we're already allocating a chunk */
4584 if (trans->allocating_chunk)
4587 space_info = __find_space_info(fs_info, flags);
4591 spin_lock(&space_info->lock);
4592 if (force < space_info->force_alloc)
4593 force = space_info->force_alloc;
4594 if (space_info->full) {
4595 if (should_alloc_chunk(fs_info, space_info, force))
4599 spin_unlock(&space_info->lock);
4603 if (!should_alloc_chunk(fs_info, space_info, force)) {
4604 spin_unlock(&space_info->lock);
4606 } else if (space_info->chunk_alloc) {
4609 space_info->chunk_alloc = 1;
4612 spin_unlock(&space_info->lock);
4614 mutex_lock(&fs_info->chunk_mutex);
4617 * The chunk_mutex is held throughout the entirety of a chunk
4618 * allocation, so once we've acquired the chunk_mutex we know that the
4619 * other guy is done and we need to recheck and see if we should
4622 if (wait_for_alloc) {
4623 mutex_unlock(&fs_info->chunk_mutex);
4629 trans->allocating_chunk = true;
4632 * If we have mixed data/metadata chunks we want to make sure we keep
4633 * allocating mixed chunks instead of individual chunks.
4635 if (btrfs_mixed_space_info(space_info))
4636 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4639 * if we're doing a data chunk, go ahead and make sure that
4640 * we keep a reasonable number of metadata chunks allocated in the
4643 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4644 fs_info->data_chunk_allocations++;
4645 if (!(fs_info->data_chunk_allocations %
4646 fs_info->metadata_ratio))
4647 force_metadata_allocation(fs_info);
4651 * Check if we have enough space in SYSTEM chunk because we may need
4652 * to update devices.
4654 check_system_chunk(trans, fs_info, flags);
4656 ret = btrfs_alloc_chunk(trans, fs_info, flags);
4657 trans->allocating_chunk = false;
4659 spin_lock(&space_info->lock);
4662 space_info->full = 1;
4669 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4671 space_info->chunk_alloc = 0;
4672 spin_unlock(&space_info->lock);
4673 mutex_unlock(&fs_info->chunk_mutex);
4675 * When we allocate a new chunk we reserve space in the chunk block
4676 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4677 * add new nodes/leafs to it if we end up needing to do it when
4678 * inserting the chunk item and updating device items as part of the
4679 * second phase of chunk allocation, performed by
4680 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4681 * large number of new block groups to create in our transaction
4682 * handle's new_bgs list to avoid exhausting the chunk block reserve
4683 * in extreme cases - like having a single transaction create many new
4684 * block groups when starting to write out the free space caches of all
4685 * the block groups that were made dirty during the lifetime of the
4688 if (trans->can_flush_pending_bgs &&
4689 trans->chunk_bytes_reserved >= (u64)SZ_2M) {
4690 btrfs_create_pending_block_groups(trans);
4691 btrfs_trans_release_chunk_metadata(trans);
4696 static int can_overcommit(struct btrfs_fs_info *fs_info,
4697 struct btrfs_space_info *space_info, u64 bytes,
4698 enum btrfs_reserve_flush_enum flush,
4701 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4707 /* Don't overcommit when in mixed mode. */
4708 if (space_info->flags & BTRFS_BLOCK_GROUP_DATA)
4712 profile = btrfs_system_alloc_profile(fs_info);
4714 profile = btrfs_metadata_alloc_profile(fs_info);
4716 used = btrfs_space_info_used(space_info, false);
4719 * We only want to allow over committing if we have lots of actual space
4720 * free, but if we don't have enough space to handle the global reserve
4721 * space then we could end up having a real enospc problem when trying
4722 * to allocate a chunk or some other such important allocation.
4724 spin_lock(&global_rsv->lock);
4725 space_size = calc_global_rsv_need_space(global_rsv);
4726 spin_unlock(&global_rsv->lock);
4727 if (used + space_size >= space_info->total_bytes)
4730 used += space_info->bytes_may_use;
4732 avail = atomic64_read(&fs_info->free_chunk_space);
4735 * If we have dup, raid1 or raid10 then only half of the free
4736 * space is actually useable. For raid56, the space info used
4737 * doesn't include the parity drive, so we don't have to
4740 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4741 BTRFS_BLOCK_GROUP_RAID1 |
4742 BTRFS_BLOCK_GROUP_RAID10))
4746 * If we aren't flushing all things, let us overcommit up to
4747 * 1/2th of the space. If we can flush, don't let us overcommit
4748 * too much, let it overcommit up to 1/8 of the space.
4750 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4755 if (used + bytes < space_info->total_bytes + avail)
4760 static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info,
4761 unsigned long nr_pages, int nr_items)
4763 struct super_block *sb = fs_info->sb;
4765 if (down_read_trylock(&sb->s_umount)) {
4766 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4767 up_read(&sb->s_umount);
4770 * We needn't worry the filesystem going from r/w to r/o though
4771 * we don't acquire ->s_umount mutex, because the filesystem
4772 * should guarantee the delalloc inodes list be empty after
4773 * the filesystem is readonly(all dirty pages are written to
4776 btrfs_start_delalloc_roots(fs_info, nr_items);
4777 if (!current->journal_info)
4778 btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1);
4782 static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info,
4788 bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
4789 nr = div64_u64(to_reclaim, bytes);
4795 #define EXTENT_SIZE_PER_ITEM SZ_256K
4798 * shrink metadata reservation for delalloc
4800 static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim,
4801 u64 orig, bool wait_ordered)
4803 struct btrfs_space_info *space_info;
4804 struct btrfs_trans_handle *trans;
4809 unsigned long nr_pages;
4812 /* Calc the number of the pages we need flush for space reservation */
4813 items = calc_reclaim_items_nr(fs_info, to_reclaim);
4814 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4816 trans = (struct btrfs_trans_handle *)current->journal_info;
4817 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4819 delalloc_bytes = percpu_counter_sum_positive(
4820 &fs_info->delalloc_bytes);
4821 if (delalloc_bytes == 0) {
4825 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4830 while (delalloc_bytes && loops < 3) {
4831 max_reclaim = min(delalloc_bytes, to_reclaim);
4832 nr_pages = max_reclaim >> PAGE_SHIFT;
4833 btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items);
4835 * We need to wait for the async pages to actually start before
4838 max_reclaim = atomic_read(&fs_info->async_delalloc_pages);
4842 if (max_reclaim <= nr_pages)
4845 max_reclaim -= nr_pages;
4847 wait_event(fs_info->async_submit_wait,
4848 atomic_read(&fs_info->async_delalloc_pages) <=
4851 spin_lock(&space_info->lock);
4852 if (list_empty(&space_info->tickets) &&
4853 list_empty(&space_info->priority_tickets)) {
4854 spin_unlock(&space_info->lock);
4857 spin_unlock(&space_info->lock);
4860 if (wait_ordered && !trans) {
4861 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4863 time_left = schedule_timeout_killable(1);
4867 delalloc_bytes = percpu_counter_sum_positive(
4868 &fs_info->delalloc_bytes);
4872 struct reserve_ticket {
4875 struct list_head list;
4876 wait_queue_head_t wait;
4880 * maybe_commit_transaction - possibly commit the transaction if its ok to
4881 * @root - the root we're allocating for
4882 * @bytes - the number of bytes we want to reserve
4883 * @force - force the commit
4885 * This will check to make sure that committing the transaction will actually
4886 * get us somewhere and then commit the transaction if it does. Otherwise it
4887 * will return -ENOSPC.
4889 static int may_commit_transaction(struct btrfs_fs_info *fs_info,
4890 struct btrfs_space_info *space_info)
4892 struct reserve_ticket *ticket = NULL;
4893 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv;
4894 struct btrfs_trans_handle *trans;
4897 trans = (struct btrfs_trans_handle *)current->journal_info;
4901 spin_lock(&space_info->lock);
4902 if (!list_empty(&space_info->priority_tickets))
4903 ticket = list_first_entry(&space_info->priority_tickets,
4904 struct reserve_ticket, list);
4905 else if (!list_empty(&space_info->tickets))
4906 ticket = list_first_entry(&space_info->tickets,
4907 struct reserve_ticket, list);
4908 bytes = (ticket) ? ticket->bytes : 0;
4909 spin_unlock(&space_info->lock);
4914 /* See if there is enough pinned space to make this reservation */
4915 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4920 * See if there is some space in the delayed insertion reservation for
4923 if (space_info != delayed_rsv->space_info)
4926 spin_lock(&delayed_rsv->lock);
4927 if (delayed_rsv->size > bytes)
4930 bytes -= delayed_rsv->size;
4931 spin_unlock(&delayed_rsv->lock);
4933 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4939 trans = btrfs_join_transaction(fs_info->extent_root);
4943 return btrfs_commit_transaction(trans);
4947 * Try to flush some data based on policy set by @state. This is only advisory
4948 * and may fail for various reasons. The caller is supposed to examine the
4949 * state of @space_info to detect the outcome.
4951 static void flush_space(struct btrfs_fs_info *fs_info,
4952 struct btrfs_space_info *space_info, u64 num_bytes,
4955 struct btrfs_root *root = fs_info->extent_root;
4956 struct btrfs_trans_handle *trans;
4961 case FLUSH_DELAYED_ITEMS_NR:
4962 case FLUSH_DELAYED_ITEMS:
4963 if (state == FLUSH_DELAYED_ITEMS_NR)
4964 nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2;
4968 trans = btrfs_join_transaction(root);
4969 if (IS_ERR(trans)) {
4970 ret = PTR_ERR(trans);
4973 ret = btrfs_run_delayed_items_nr(trans, nr);
4974 btrfs_end_transaction(trans);
4976 case FLUSH_DELALLOC:
4977 case FLUSH_DELALLOC_WAIT:
4978 shrink_delalloc(fs_info, num_bytes * 2, num_bytes,
4979 state == FLUSH_DELALLOC_WAIT);
4982 trans = btrfs_join_transaction(root);
4983 if (IS_ERR(trans)) {
4984 ret = PTR_ERR(trans);
4987 ret = do_chunk_alloc(trans, fs_info,
4988 btrfs_metadata_alloc_profile(fs_info),
4989 CHUNK_ALLOC_NO_FORCE);
4990 btrfs_end_transaction(trans);
4991 if (ret > 0 || ret == -ENOSPC)
4995 ret = may_commit_transaction(fs_info, space_info);
5002 trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, state,
5008 btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info,
5009 struct btrfs_space_info *space_info,
5012 struct reserve_ticket *ticket;
5017 list_for_each_entry(ticket, &space_info->tickets, list)
5018 to_reclaim += ticket->bytes;
5019 list_for_each_entry(ticket, &space_info->priority_tickets, list)
5020 to_reclaim += ticket->bytes;
5024 to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M);
5025 if (can_overcommit(fs_info, space_info, to_reclaim,
5026 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
5029 used = btrfs_space_info_used(space_info, true);
5031 if (can_overcommit(fs_info, space_info, SZ_1M,
5032 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
5033 expected = div_factor_fine(space_info->total_bytes, 95);
5035 expected = div_factor_fine(space_info->total_bytes, 90);
5037 if (used > expected)
5038 to_reclaim = used - expected;
5041 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
5042 space_info->bytes_reserved);
5046 static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info,
5047 struct btrfs_space_info *space_info,
5048 u64 used, bool system_chunk)
5050 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
5052 /* If we're just plain full then async reclaim just slows us down. */
5053 if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh)
5056 if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5060 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
5061 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
5064 static void wake_all_tickets(struct list_head *head)
5066 struct reserve_ticket *ticket;
5068 while (!list_empty(head)) {
5069 ticket = list_first_entry(head, struct reserve_ticket, list);
5070 list_del_init(&ticket->list);
5071 ticket->error = -ENOSPC;
5072 wake_up(&ticket->wait);
5077 * This is for normal flushers, we can wait all goddamned day if we want to. We
5078 * will loop and continuously try to flush as long as we are making progress.
5079 * We count progress as clearing off tickets each time we have to loop.
5081 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
5083 struct btrfs_fs_info *fs_info;
5084 struct btrfs_space_info *space_info;
5087 int commit_cycles = 0;
5088 u64 last_tickets_id;
5090 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
5091 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5093 spin_lock(&space_info->lock);
5094 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5097 space_info->flush = 0;
5098 spin_unlock(&space_info->lock);
5101 last_tickets_id = space_info->tickets_id;
5102 spin_unlock(&space_info->lock);
5104 flush_state = FLUSH_DELAYED_ITEMS_NR;
5106 flush_space(fs_info, space_info, to_reclaim, flush_state);
5107 spin_lock(&space_info->lock);
5108 if (list_empty(&space_info->tickets)) {
5109 space_info->flush = 0;
5110 spin_unlock(&space_info->lock);
5113 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info,
5116 if (last_tickets_id == space_info->tickets_id) {
5119 last_tickets_id = space_info->tickets_id;
5120 flush_state = FLUSH_DELAYED_ITEMS_NR;
5125 if (flush_state > COMMIT_TRANS) {
5127 if (commit_cycles > 2) {
5128 wake_all_tickets(&space_info->tickets);
5129 space_info->flush = 0;
5131 flush_state = FLUSH_DELAYED_ITEMS_NR;
5134 spin_unlock(&space_info->lock);
5135 } while (flush_state <= COMMIT_TRANS);
5138 void btrfs_init_async_reclaim_work(struct work_struct *work)
5140 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
5143 static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
5144 struct btrfs_space_info *space_info,
5145 struct reserve_ticket *ticket)
5148 int flush_state = FLUSH_DELAYED_ITEMS_NR;
5150 spin_lock(&space_info->lock);
5151 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5154 spin_unlock(&space_info->lock);
5157 spin_unlock(&space_info->lock);
5160 flush_space(fs_info, space_info, to_reclaim, flush_state);
5162 spin_lock(&space_info->lock);
5163 if (ticket->bytes == 0) {
5164 spin_unlock(&space_info->lock);
5167 spin_unlock(&space_info->lock);
5170 * Priority flushers can't wait on delalloc without
5173 if (flush_state == FLUSH_DELALLOC ||
5174 flush_state == FLUSH_DELALLOC_WAIT)
5175 flush_state = ALLOC_CHUNK;
5176 } while (flush_state < COMMIT_TRANS);
5179 static int wait_reserve_ticket(struct btrfs_fs_info *fs_info,
5180 struct btrfs_space_info *space_info,
5181 struct reserve_ticket *ticket, u64 orig_bytes)
5187 spin_lock(&space_info->lock);
5188 while (ticket->bytes > 0 && ticket->error == 0) {
5189 ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE);
5194 spin_unlock(&space_info->lock);
5198 finish_wait(&ticket->wait, &wait);
5199 spin_lock(&space_info->lock);
5202 ret = ticket->error;
5203 if (!list_empty(&ticket->list))
5204 list_del_init(&ticket->list);
5205 if (ticket->bytes && ticket->bytes < orig_bytes) {
5206 u64 num_bytes = orig_bytes - ticket->bytes;
5207 space_info->bytes_may_use -= num_bytes;
5208 trace_btrfs_space_reservation(fs_info, "space_info",
5209 space_info->flags, num_bytes, 0);
5211 spin_unlock(&space_info->lock);
5217 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5218 * @root - the root we're allocating for
5219 * @space_info - the space info we want to allocate from
5220 * @orig_bytes - the number of bytes we want
5221 * @flush - whether or not we can flush to make our reservation
5223 * This will reserve orig_bytes number of bytes from the space info associated
5224 * with the block_rsv. If there is not enough space it will make an attempt to
5225 * flush out space to make room. It will do this by flushing delalloc if
5226 * possible or committing the transaction. If flush is 0 then no attempts to
5227 * regain reservations will be made and this will fail if there is not enough
5230 static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info,
5231 struct btrfs_space_info *space_info,
5233 enum btrfs_reserve_flush_enum flush,
5236 struct reserve_ticket ticket;
5241 ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL);
5243 spin_lock(&space_info->lock);
5245 used = btrfs_space_info_used(space_info, true);
5248 * If we have enough space then hooray, make our reservation and carry
5249 * on. If not see if we can overcommit, and if we can, hooray carry on.
5250 * If not things get more complicated.
5252 if (used + orig_bytes <= space_info->total_bytes) {
5253 space_info->bytes_may_use += orig_bytes;
5254 trace_btrfs_space_reservation(fs_info, "space_info",
5255 space_info->flags, orig_bytes, 1);
5257 } else if (can_overcommit(fs_info, space_info, orig_bytes, flush,
5259 space_info->bytes_may_use += orig_bytes;
5260 trace_btrfs_space_reservation(fs_info, "space_info",
5261 space_info->flags, orig_bytes, 1);
5266 * If we couldn't make a reservation then setup our reservation ticket
5267 * and kick the async worker if it's not already running.
5269 * If we are a priority flusher then we just need to add our ticket to
5270 * the list and we will do our own flushing further down.
5272 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
5273 ticket.bytes = orig_bytes;
5275 init_waitqueue_head(&ticket.wait);
5276 if (flush == BTRFS_RESERVE_FLUSH_ALL) {
5277 list_add_tail(&ticket.list, &space_info->tickets);
5278 if (!space_info->flush) {
5279 space_info->flush = 1;
5280 trace_btrfs_trigger_flush(fs_info,
5284 queue_work(system_unbound_wq,
5285 &fs_info->async_reclaim_work);
5288 list_add_tail(&ticket.list,
5289 &space_info->priority_tickets);
5291 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
5294 * We will do the space reservation dance during log replay,
5295 * which means we won't have fs_info->fs_root set, so don't do
5296 * the async reclaim as we will panic.
5298 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) &&
5299 need_do_async_reclaim(fs_info, space_info,
5300 used, system_chunk) &&
5301 !work_busy(&fs_info->async_reclaim_work)) {
5302 trace_btrfs_trigger_flush(fs_info, space_info->flags,
5303 orig_bytes, flush, "preempt");
5304 queue_work(system_unbound_wq,
5305 &fs_info->async_reclaim_work);
5308 spin_unlock(&space_info->lock);
5309 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
5312 if (flush == BTRFS_RESERVE_FLUSH_ALL)
5313 return wait_reserve_ticket(fs_info, space_info, &ticket,
5317 priority_reclaim_metadata_space(fs_info, space_info, &ticket);
5318 spin_lock(&space_info->lock);
5320 if (ticket.bytes < orig_bytes) {
5321 u64 num_bytes = orig_bytes - ticket.bytes;
5322 space_info->bytes_may_use -= num_bytes;
5323 trace_btrfs_space_reservation(fs_info, "space_info",
5328 list_del_init(&ticket.list);
5331 spin_unlock(&space_info->lock);
5332 ASSERT(list_empty(&ticket.list));
5337 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5338 * @root - the root we're allocating for
5339 * @block_rsv - the block_rsv we're allocating for
5340 * @orig_bytes - the number of bytes we want
5341 * @flush - whether or not we can flush to make our reservation
5343 * This will reserve orgi_bytes number of bytes from the space info associated
5344 * with the block_rsv. If there is not enough space it will make an attempt to
5345 * flush out space to make room. It will do this by flushing delalloc if
5346 * possible or committing the transaction. If flush is 0 then no attempts to
5347 * regain reservations will be made and this will fail if there is not enough
5350 static int reserve_metadata_bytes(struct btrfs_root *root,
5351 struct btrfs_block_rsv *block_rsv,
5353 enum btrfs_reserve_flush_enum flush)
5355 struct btrfs_fs_info *fs_info = root->fs_info;
5356 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5358 bool system_chunk = (root == fs_info->chunk_root);
5360 ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info,
5361 orig_bytes, flush, system_chunk);
5362 if (ret == -ENOSPC &&
5363 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5364 if (block_rsv != global_rsv &&
5365 !block_rsv_use_bytes(global_rsv, orig_bytes))
5368 if (ret == -ENOSPC) {
5369 trace_btrfs_space_reservation(fs_info, "space_info:enospc",
5370 block_rsv->space_info->flags,
5373 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG))
5374 dump_space_info(fs_info, block_rsv->space_info,
5380 static struct btrfs_block_rsv *get_block_rsv(
5381 const struct btrfs_trans_handle *trans,
5382 const struct btrfs_root *root)
5384 struct btrfs_fs_info *fs_info = root->fs_info;
5385 struct btrfs_block_rsv *block_rsv = NULL;
5387 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5388 (root == fs_info->csum_root && trans->adding_csums) ||
5389 (root == fs_info->uuid_root))
5390 block_rsv = trans->block_rsv;
5393 block_rsv = root->block_rsv;
5396 block_rsv = &fs_info->empty_block_rsv;
5401 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5405 spin_lock(&block_rsv->lock);
5406 if (block_rsv->reserved >= num_bytes) {
5407 block_rsv->reserved -= num_bytes;
5408 if (block_rsv->reserved < block_rsv->size)
5409 block_rsv->full = 0;
5412 spin_unlock(&block_rsv->lock);
5416 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5417 u64 num_bytes, int update_size)
5419 spin_lock(&block_rsv->lock);
5420 block_rsv->reserved += num_bytes;
5422 block_rsv->size += num_bytes;
5423 else if (block_rsv->reserved >= block_rsv->size)
5424 block_rsv->full = 1;
5425 spin_unlock(&block_rsv->lock);
5428 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5429 struct btrfs_block_rsv *dest, u64 num_bytes,
5432 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5435 if (global_rsv->space_info != dest->space_info)
5438 spin_lock(&global_rsv->lock);
5439 min_bytes = div_factor(global_rsv->size, min_factor);
5440 if (global_rsv->reserved < min_bytes + num_bytes) {
5441 spin_unlock(&global_rsv->lock);
5444 global_rsv->reserved -= num_bytes;
5445 if (global_rsv->reserved < global_rsv->size)
5446 global_rsv->full = 0;
5447 spin_unlock(&global_rsv->lock);
5449 block_rsv_add_bytes(dest, num_bytes, 1);
5454 * This is for space we already have accounted in space_info->bytes_may_use, so
5455 * basically when we're returning space from block_rsv's.
5457 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
5458 struct btrfs_space_info *space_info,
5461 struct reserve_ticket *ticket;
5462 struct list_head *head;
5464 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH;
5465 bool check_overcommit = false;
5467 spin_lock(&space_info->lock);
5468 head = &space_info->priority_tickets;
5471 * If we are over our limit then we need to check and see if we can
5472 * overcommit, and if we can't then we just need to free up our space
5473 * and not satisfy any requests.
5475 used = btrfs_space_info_used(space_info, true);
5476 if (used - num_bytes >= space_info->total_bytes)
5477 check_overcommit = true;
5479 while (!list_empty(head) && num_bytes) {
5480 ticket = list_first_entry(head, struct reserve_ticket,
5483 * We use 0 bytes because this space is already reserved, so
5484 * adding the ticket space would be a double count.
5486 if (check_overcommit &&
5487 !can_overcommit(fs_info, space_info, 0, flush, false))
5489 if (num_bytes >= ticket->bytes) {
5490 list_del_init(&ticket->list);
5491 num_bytes -= ticket->bytes;
5493 space_info->tickets_id++;
5494 wake_up(&ticket->wait);
5496 ticket->bytes -= num_bytes;
5501 if (num_bytes && head == &space_info->priority_tickets) {
5502 head = &space_info->tickets;
5503 flush = BTRFS_RESERVE_FLUSH_ALL;
5506 space_info->bytes_may_use -= num_bytes;
5507 trace_btrfs_space_reservation(fs_info, "space_info",
5508 space_info->flags, num_bytes, 0);
5509 spin_unlock(&space_info->lock);
5513 * This is for newly allocated space that isn't accounted in
5514 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5515 * we use this helper.
5517 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
5518 struct btrfs_space_info *space_info,
5521 struct reserve_ticket *ticket;
5522 struct list_head *head = &space_info->priority_tickets;
5525 while (!list_empty(head) && num_bytes) {
5526 ticket = list_first_entry(head, struct reserve_ticket,
5528 if (num_bytes >= ticket->bytes) {
5529 trace_btrfs_space_reservation(fs_info, "space_info",
5532 list_del_init(&ticket->list);
5533 num_bytes -= ticket->bytes;
5534 space_info->bytes_may_use += ticket->bytes;
5536 space_info->tickets_id++;
5537 wake_up(&ticket->wait);
5539 trace_btrfs_space_reservation(fs_info, "space_info",
5542 space_info->bytes_may_use += num_bytes;
5543 ticket->bytes -= num_bytes;
5548 if (num_bytes && head == &space_info->priority_tickets) {
5549 head = &space_info->tickets;
5554 static u64 block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5555 struct btrfs_block_rsv *block_rsv,
5556 struct btrfs_block_rsv *dest, u64 num_bytes,
5557 u64 *qgroup_to_release_ret)
5559 struct btrfs_space_info *space_info = block_rsv->space_info;
5560 u64 qgroup_to_release = 0;
5563 spin_lock(&block_rsv->lock);
5564 if (num_bytes == (u64)-1) {
5565 num_bytes = block_rsv->size;
5566 qgroup_to_release = block_rsv->qgroup_rsv_size;
5568 block_rsv->size -= num_bytes;
5569 if (block_rsv->reserved >= block_rsv->size) {
5570 num_bytes = block_rsv->reserved - block_rsv->size;
5571 block_rsv->reserved = block_rsv->size;
5572 block_rsv->full = 1;
5576 if (block_rsv->qgroup_rsv_reserved >= block_rsv->qgroup_rsv_size) {
5577 qgroup_to_release = block_rsv->qgroup_rsv_reserved -
5578 block_rsv->qgroup_rsv_size;
5579 block_rsv->qgroup_rsv_reserved = block_rsv->qgroup_rsv_size;
5581 qgroup_to_release = 0;
5583 spin_unlock(&block_rsv->lock);
5586 if (num_bytes > 0) {
5588 spin_lock(&dest->lock);
5592 bytes_to_add = dest->size - dest->reserved;
5593 bytes_to_add = min(num_bytes, bytes_to_add);
5594 dest->reserved += bytes_to_add;
5595 if (dest->reserved >= dest->size)
5597 num_bytes -= bytes_to_add;
5599 spin_unlock(&dest->lock);
5602 space_info_add_old_bytes(fs_info, space_info,
5605 if (qgroup_to_release_ret)
5606 *qgroup_to_release_ret = qgroup_to_release;
5610 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
5611 struct btrfs_block_rsv *dst, u64 num_bytes,
5616 ret = block_rsv_use_bytes(src, num_bytes);
5620 block_rsv_add_bytes(dst, num_bytes, update_size);
5624 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5626 memset(rsv, 0, sizeof(*rsv));
5627 spin_lock_init(&rsv->lock);
5631 void btrfs_init_metadata_block_rsv(struct btrfs_fs_info *fs_info,
5632 struct btrfs_block_rsv *rsv,
5633 unsigned short type)
5635 btrfs_init_block_rsv(rsv, type);
5636 rsv->space_info = __find_space_info(fs_info,
5637 BTRFS_BLOCK_GROUP_METADATA);
5640 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info,
5641 unsigned short type)
5643 struct btrfs_block_rsv *block_rsv;
5645 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5649 btrfs_init_metadata_block_rsv(fs_info, block_rsv, type);
5653 void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info,
5654 struct btrfs_block_rsv *rsv)
5658 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
5662 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
5667 int btrfs_block_rsv_add(struct btrfs_root *root,
5668 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5669 enum btrfs_reserve_flush_enum flush)
5676 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5678 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5685 int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor)
5693 spin_lock(&block_rsv->lock);
5694 num_bytes = div_factor(block_rsv->size, min_factor);
5695 if (block_rsv->reserved >= num_bytes)
5697 spin_unlock(&block_rsv->lock);
5702 int btrfs_block_rsv_refill(struct btrfs_root *root,
5703 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5704 enum btrfs_reserve_flush_enum flush)
5712 spin_lock(&block_rsv->lock);
5713 num_bytes = min_reserved;
5714 if (block_rsv->reserved >= num_bytes)
5717 num_bytes -= block_rsv->reserved;
5718 spin_unlock(&block_rsv->lock);
5723 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5725 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5733 * btrfs_inode_rsv_refill - refill the inode block rsv.
5734 * @inode - the inode we are refilling.
5735 * @flush - the flusing restriction.
5737 * Essentially the same as btrfs_block_rsv_refill, except it uses the
5738 * block_rsv->size as the minimum size. We'll either refill the missing amount
5739 * or return if we already have enough space. This will also handle the resreve
5740 * tracepoint for the reserved amount.
5742 static int btrfs_inode_rsv_refill(struct btrfs_inode *inode,
5743 enum btrfs_reserve_flush_enum flush)
5745 struct btrfs_root *root = inode->root;
5746 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5748 u64 qgroup_num_bytes = 0;
5751 spin_lock(&block_rsv->lock);
5752 if (block_rsv->reserved < block_rsv->size)
5753 num_bytes = block_rsv->size - block_rsv->reserved;
5754 if (block_rsv->qgroup_rsv_reserved < block_rsv->qgroup_rsv_size)
5755 qgroup_num_bytes = block_rsv->qgroup_rsv_size -
5756 block_rsv->qgroup_rsv_reserved;
5757 spin_unlock(&block_rsv->lock);
5762 ret = btrfs_qgroup_reserve_meta_prealloc(root, qgroup_num_bytes, true);
5765 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5767 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5768 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5769 btrfs_ino(inode), num_bytes, 1);
5771 /* Don't forget to increase qgroup_rsv_reserved */
5772 spin_lock(&block_rsv->lock);
5773 block_rsv->qgroup_rsv_reserved += qgroup_num_bytes;
5774 spin_unlock(&block_rsv->lock);
5776 btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes);
5781 * btrfs_inode_rsv_release - release any excessive reservation.
5782 * @inode - the inode we need to release from.
5783 * @qgroup_free - free or convert qgroup meta.
5784 * Unlike normal operation, qgroup meta reservation needs to know if we are
5785 * freeing qgroup reservation or just converting it into per-trans. Normally
5786 * @qgroup_free is true for error handling, and false for normal release.
5788 * This is the same as btrfs_block_rsv_release, except that it handles the
5789 * tracepoint for the reservation.
5791 static void btrfs_inode_rsv_release(struct btrfs_inode *inode, bool qgroup_free)
5793 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5794 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5795 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5797 u64 qgroup_to_release = 0;
5800 * Since we statically set the block_rsv->size we just want to say we
5801 * are releasing 0 bytes, and then we'll just get the reservation over
5804 released = block_rsv_release_bytes(fs_info, block_rsv, global_rsv, 0,
5805 &qgroup_to_release);
5807 trace_btrfs_space_reservation(fs_info, "delalloc",
5808 btrfs_ino(inode), released, 0);
5810 btrfs_qgroup_free_meta_prealloc(inode->root, qgroup_to_release);
5812 btrfs_qgroup_convert_reserved_meta(inode->root,
5816 void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
5817 struct btrfs_block_rsv *block_rsv,
5820 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5822 if (global_rsv == block_rsv ||
5823 block_rsv->space_info != global_rsv->space_info)
5825 block_rsv_release_bytes(fs_info, block_rsv, global_rsv, num_bytes, NULL);
5828 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5830 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5831 struct btrfs_space_info *sinfo = block_rsv->space_info;
5835 * The global block rsv is based on the size of the extent tree, the
5836 * checksum tree and the root tree. If the fs is empty we want to set
5837 * it to a minimal amount for safety.
5839 num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) +
5840 btrfs_root_used(&fs_info->csum_root->root_item) +
5841 btrfs_root_used(&fs_info->tree_root->root_item);
5842 num_bytes = max_t(u64, num_bytes, SZ_16M);
5844 spin_lock(&sinfo->lock);
5845 spin_lock(&block_rsv->lock);
5847 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
5849 if (block_rsv->reserved < block_rsv->size) {
5850 num_bytes = btrfs_space_info_used(sinfo, true);
5851 if (sinfo->total_bytes > num_bytes) {
5852 num_bytes = sinfo->total_bytes - num_bytes;
5853 num_bytes = min(num_bytes,
5854 block_rsv->size - block_rsv->reserved);
5855 block_rsv->reserved += num_bytes;
5856 sinfo->bytes_may_use += num_bytes;
5857 trace_btrfs_space_reservation(fs_info, "space_info",
5858 sinfo->flags, num_bytes,
5861 } else if (block_rsv->reserved > block_rsv->size) {
5862 num_bytes = block_rsv->reserved - block_rsv->size;
5863 sinfo->bytes_may_use -= num_bytes;
5864 trace_btrfs_space_reservation(fs_info, "space_info",
5865 sinfo->flags, num_bytes, 0);
5866 block_rsv->reserved = block_rsv->size;
5869 if (block_rsv->reserved == block_rsv->size)
5870 block_rsv->full = 1;
5872 block_rsv->full = 0;
5874 spin_unlock(&block_rsv->lock);
5875 spin_unlock(&sinfo->lock);
5878 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5880 struct btrfs_space_info *space_info;
5882 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5883 fs_info->chunk_block_rsv.space_info = space_info;
5885 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5886 fs_info->global_block_rsv.space_info = space_info;
5887 fs_info->trans_block_rsv.space_info = space_info;
5888 fs_info->empty_block_rsv.space_info = space_info;
5889 fs_info->delayed_block_rsv.space_info = space_info;
5891 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5892 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5893 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5894 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5895 if (fs_info->quota_root)
5896 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5897 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5899 update_global_block_rsv(fs_info);
5902 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5904 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5906 WARN_ON(fs_info->trans_block_rsv.size > 0);
5907 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5908 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5909 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5910 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5911 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5916 * To be called after all the new block groups attached to the transaction
5917 * handle have been created (btrfs_create_pending_block_groups()).
5919 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5921 struct btrfs_fs_info *fs_info = trans->fs_info;
5923 if (!trans->chunk_bytes_reserved)
5926 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5928 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5929 trans->chunk_bytes_reserved, NULL);
5930 trans->chunk_bytes_reserved = 0;
5934 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5935 * root: the root of the parent directory
5936 * rsv: block reservation
5937 * items: the number of items that we need do reservation
5938 * qgroup_reserved: used to return the reserved size in qgroup
5940 * This function is used to reserve the space for snapshot/subvolume
5941 * creation and deletion. Those operations are different with the
5942 * common file/directory operations, they change two fs/file trees
5943 * and root tree, the number of items that the qgroup reserves is
5944 * different with the free space reservation. So we can not use
5945 * the space reservation mechanism in start_transaction().
5947 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5948 struct btrfs_block_rsv *rsv,
5950 bool use_global_rsv)
5954 struct btrfs_fs_info *fs_info = root->fs_info;
5955 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5957 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
5958 /* One for parent inode, two for dir entries */
5959 num_bytes = 3 * fs_info->nodesize;
5960 ret = btrfs_qgroup_reserve_meta_prealloc(root, num_bytes, true);
5967 num_bytes = btrfs_calc_trans_metadata_size(fs_info, items);
5968 rsv->space_info = __find_space_info(fs_info,
5969 BTRFS_BLOCK_GROUP_METADATA);
5970 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5971 BTRFS_RESERVE_FLUSH_ALL);
5973 if (ret == -ENOSPC && use_global_rsv)
5974 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, 1);
5976 if (ret && num_bytes)
5977 btrfs_qgroup_free_meta_prealloc(root, num_bytes);
5982 void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info,
5983 struct btrfs_block_rsv *rsv)
5985 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
5988 static void btrfs_calculate_inode_block_rsv_size(struct btrfs_fs_info *fs_info,
5989 struct btrfs_inode *inode)
5991 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5992 u64 reserve_size = 0;
5993 u64 qgroup_rsv_size = 0;
5995 unsigned outstanding_extents;
5997 lockdep_assert_held(&inode->lock);
5998 outstanding_extents = inode->outstanding_extents;
5999 if (outstanding_extents)
6000 reserve_size = btrfs_calc_trans_metadata_size(fs_info,
6001 outstanding_extents + 1);
6002 csum_leaves = btrfs_csum_bytes_to_leaves(fs_info,
6004 reserve_size += btrfs_calc_trans_metadata_size(fs_info,
6007 * For qgroup rsv, the calculation is very simple:
6008 * account one nodesize for each outstanding extent
6010 * This is overestimating in most cases.
6012 qgroup_rsv_size = outstanding_extents * fs_info->nodesize;
6014 spin_lock(&block_rsv->lock);
6015 block_rsv->size = reserve_size;
6016 block_rsv->qgroup_rsv_size = qgroup_rsv_size;
6017 spin_unlock(&block_rsv->lock);
6020 int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes)
6022 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6023 unsigned nr_extents;
6024 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
6026 bool delalloc_lock = true;
6028 /* If we are a free space inode we need to not flush since we will be in
6029 * the middle of a transaction commit. We also don't need the delalloc
6030 * mutex since we won't race with anybody. We need this mostly to make
6031 * lockdep shut its filthy mouth.
6033 * If we have a transaction open (can happen if we call truncate_block
6034 * from truncate), then we need FLUSH_LIMIT so we don't deadlock.
6036 if (btrfs_is_free_space_inode(inode)) {
6037 flush = BTRFS_RESERVE_NO_FLUSH;
6038 delalloc_lock = false;
6040 if (current->journal_info)
6041 flush = BTRFS_RESERVE_FLUSH_LIMIT;
6043 if (btrfs_transaction_in_commit(fs_info))
6044 schedule_timeout(1);
6048 mutex_lock(&inode->delalloc_mutex);
6050 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6052 /* Add our new extents and calculate the new rsv size. */
6053 spin_lock(&inode->lock);
6054 nr_extents = count_max_extents(num_bytes);
6055 btrfs_mod_outstanding_extents(inode, nr_extents);
6056 inode->csum_bytes += num_bytes;
6057 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6058 spin_unlock(&inode->lock);
6060 ret = btrfs_inode_rsv_refill(inode, flush);
6065 mutex_unlock(&inode->delalloc_mutex);
6069 spin_lock(&inode->lock);
6070 nr_extents = count_max_extents(num_bytes);
6071 btrfs_mod_outstanding_extents(inode, -nr_extents);
6072 inode->csum_bytes -= num_bytes;
6073 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6074 spin_unlock(&inode->lock);
6076 btrfs_inode_rsv_release(inode, true);
6078 mutex_unlock(&inode->delalloc_mutex);
6083 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
6084 * @inode: the inode to release the reservation for.
6085 * @num_bytes: the number of bytes we are releasing.
6086 * @qgroup_free: free qgroup reservation or convert it to per-trans reservation
6088 * This will release the metadata reservation for an inode. This can be called
6089 * once we complete IO for a given set of bytes to release their metadata
6090 * reservations, or on error for the same reason.
6092 void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes,
6095 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6097 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6098 spin_lock(&inode->lock);
6099 inode->csum_bytes -= num_bytes;
6100 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6101 spin_unlock(&inode->lock);
6103 if (btrfs_is_testing(fs_info))
6106 btrfs_inode_rsv_release(inode, qgroup_free);
6110 * btrfs_delalloc_release_extents - release our outstanding_extents
6111 * @inode: the inode to balance the reservation for.
6112 * @num_bytes: the number of bytes we originally reserved with
6113 * @qgroup_free: do we need to free qgroup meta reservation or convert them.
6115 * When we reserve space we increase outstanding_extents for the extents we may
6116 * add. Once we've set the range as delalloc or created our ordered extents we
6117 * have outstanding_extents to track the real usage, so we use this to free our
6118 * temporarily tracked outstanding_extents. This _must_ be used in conjunction
6119 * with btrfs_delalloc_reserve_metadata.
6121 void btrfs_delalloc_release_extents(struct btrfs_inode *inode, u64 num_bytes,
6124 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6125 unsigned num_extents;
6127 spin_lock(&inode->lock);
6128 num_extents = count_max_extents(num_bytes);
6129 btrfs_mod_outstanding_extents(inode, -num_extents);
6130 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6131 spin_unlock(&inode->lock);
6133 if (btrfs_is_testing(fs_info))
6136 btrfs_inode_rsv_release(inode, qgroup_free);
6140 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6142 * @inode: inode we're writing to
6143 * @start: start range we are writing to
6144 * @len: how long the range we are writing to
6145 * @reserved: mandatory parameter, record actually reserved qgroup ranges of
6146 * current reservation.
6148 * This will do the following things
6150 * o reserve space in data space info for num bytes
6151 * and reserve precious corresponding qgroup space
6152 * (Done in check_data_free_space)
6154 * o reserve space for metadata space, based on the number of outstanding
6155 * extents and how much csums will be needed
6156 * also reserve metadata space in a per root over-reserve method.
6157 * o add to the inodes->delalloc_bytes
6158 * o add it to the fs_info's delalloc inodes list.
6159 * (Above 3 all done in delalloc_reserve_metadata)
6161 * Return 0 for success
6162 * Return <0 for error(-ENOSPC or -EQUOT)
6164 int btrfs_delalloc_reserve_space(struct inode *inode,
6165 struct extent_changeset **reserved, u64 start, u64 len)
6169 ret = btrfs_check_data_free_space(inode, reserved, start, len);
6172 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len);
6174 btrfs_free_reserved_data_space(inode, *reserved, start, len);
6179 * btrfs_delalloc_release_space - release data and metadata space for delalloc
6180 * @inode: inode we're releasing space for
6181 * @start: start position of the space already reserved
6182 * @len: the len of the space already reserved
6183 * @release_bytes: the len of the space we consumed or didn't use
6185 * This function will release the metadata space that was not used and will
6186 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
6187 * list if there are no delalloc bytes left.
6188 * Also it will handle the qgroup reserved space.
6190 void btrfs_delalloc_release_space(struct inode *inode,
6191 struct extent_changeset *reserved,
6192 u64 start, u64 len, bool qgroup_free)
6194 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, qgroup_free);
6195 btrfs_free_reserved_data_space(inode, reserved, start, len);
6198 static int update_block_group(struct btrfs_trans_handle *trans,
6199 struct btrfs_fs_info *info, u64 bytenr,
6200 u64 num_bytes, int alloc)
6202 struct btrfs_block_group_cache *cache = NULL;
6203 u64 total = num_bytes;
6208 /* block accounting for super block */
6209 spin_lock(&info->delalloc_root_lock);
6210 old_val = btrfs_super_bytes_used(info->super_copy);
6212 old_val += num_bytes;
6214 old_val -= num_bytes;
6215 btrfs_set_super_bytes_used(info->super_copy, old_val);
6216 spin_unlock(&info->delalloc_root_lock);
6219 cache = btrfs_lookup_block_group(info, bytenr);
6222 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
6223 BTRFS_BLOCK_GROUP_RAID1 |
6224 BTRFS_BLOCK_GROUP_RAID10))
6229 * If this block group has free space cache written out, we
6230 * need to make sure to load it if we are removing space. This
6231 * is because we need the unpinning stage to actually add the
6232 * space back to the block group, otherwise we will leak space.
6234 if (!alloc && cache->cached == BTRFS_CACHE_NO)
6235 cache_block_group(cache, 1);
6237 byte_in_group = bytenr - cache->key.objectid;
6238 WARN_ON(byte_in_group > cache->key.offset);
6240 spin_lock(&cache->space_info->lock);
6241 spin_lock(&cache->lock);
6243 if (btrfs_test_opt(info, SPACE_CACHE) &&
6244 cache->disk_cache_state < BTRFS_DC_CLEAR)
6245 cache->disk_cache_state = BTRFS_DC_CLEAR;
6247 old_val = btrfs_block_group_used(&cache->item);
6248 num_bytes = min(total, cache->key.offset - byte_in_group);
6250 old_val += num_bytes;
6251 btrfs_set_block_group_used(&cache->item, old_val);
6252 cache->reserved -= num_bytes;
6253 cache->space_info->bytes_reserved -= num_bytes;
6254 cache->space_info->bytes_used += num_bytes;
6255 cache->space_info->disk_used += num_bytes * factor;
6256 spin_unlock(&cache->lock);
6257 spin_unlock(&cache->space_info->lock);
6259 old_val -= num_bytes;
6260 btrfs_set_block_group_used(&cache->item, old_val);
6261 cache->pinned += num_bytes;
6262 cache->space_info->bytes_pinned += num_bytes;
6263 cache->space_info->bytes_used -= num_bytes;
6264 cache->space_info->disk_used -= num_bytes * factor;
6265 spin_unlock(&cache->lock);
6266 spin_unlock(&cache->space_info->lock);
6268 trace_btrfs_space_reservation(info, "pinned",
6269 cache->space_info->flags,
6271 percpu_counter_add(&cache->space_info->total_bytes_pinned,
6273 set_extent_dirty(info->pinned_extents,
6274 bytenr, bytenr + num_bytes - 1,
6275 GFP_NOFS | __GFP_NOFAIL);
6278 spin_lock(&trans->transaction->dirty_bgs_lock);
6279 if (list_empty(&cache->dirty_list)) {
6280 list_add_tail(&cache->dirty_list,
6281 &trans->transaction->dirty_bgs);
6282 trans->transaction->num_dirty_bgs++;
6283 btrfs_get_block_group(cache);
6285 spin_unlock(&trans->transaction->dirty_bgs_lock);
6288 * No longer have used bytes in this block group, queue it for
6289 * deletion. We do this after adding the block group to the
6290 * dirty list to avoid races between cleaner kthread and space
6293 if (!alloc && old_val == 0) {
6294 spin_lock(&info->unused_bgs_lock);
6295 if (list_empty(&cache->bg_list)) {
6296 btrfs_get_block_group(cache);
6297 trace_btrfs_add_unused_block_group(cache);
6298 list_add_tail(&cache->bg_list,
6301 spin_unlock(&info->unused_bgs_lock);
6304 btrfs_put_block_group(cache);
6306 bytenr += num_bytes;
6311 static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start)
6313 struct btrfs_block_group_cache *cache;
6316 spin_lock(&fs_info->block_group_cache_lock);
6317 bytenr = fs_info->first_logical_byte;
6318 spin_unlock(&fs_info->block_group_cache_lock);
6320 if (bytenr < (u64)-1)
6323 cache = btrfs_lookup_first_block_group(fs_info, search_start);
6327 bytenr = cache->key.objectid;
6328 btrfs_put_block_group(cache);
6333 static int pin_down_extent(struct btrfs_fs_info *fs_info,
6334 struct btrfs_block_group_cache *cache,
6335 u64 bytenr, u64 num_bytes, int reserved)
6337 spin_lock(&cache->space_info->lock);
6338 spin_lock(&cache->lock);
6339 cache->pinned += num_bytes;
6340 cache->space_info->bytes_pinned += num_bytes;
6342 cache->reserved -= num_bytes;
6343 cache->space_info->bytes_reserved -= num_bytes;
6345 spin_unlock(&cache->lock);
6346 spin_unlock(&cache->space_info->lock);
6348 trace_btrfs_space_reservation(fs_info, "pinned",
6349 cache->space_info->flags, num_bytes, 1);
6350 percpu_counter_add(&cache->space_info->total_bytes_pinned, num_bytes);
6351 set_extent_dirty(fs_info->pinned_extents, bytenr,
6352 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
6357 * this function must be called within transaction
6359 int btrfs_pin_extent(struct btrfs_fs_info *fs_info,
6360 u64 bytenr, u64 num_bytes, int reserved)
6362 struct btrfs_block_group_cache *cache;
6364 cache = btrfs_lookup_block_group(fs_info, bytenr);
6365 BUG_ON(!cache); /* Logic error */
6367 pin_down_extent(fs_info, cache, bytenr, num_bytes, reserved);
6369 btrfs_put_block_group(cache);
6374 * this function must be called within transaction
6376 int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info,
6377 u64 bytenr, u64 num_bytes)
6379 struct btrfs_block_group_cache *cache;
6382 cache = btrfs_lookup_block_group(fs_info, bytenr);
6387 * pull in the free space cache (if any) so that our pin
6388 * removes the free space from the cache. We have load_only set
6389 * to one because the slow code to read in the free extents does check
6390 * the pinned extents.
6392 cache_block_group(cache, 1);
6394 pin_down_extent(fs_info, cache, bytenr, num_bytes, 0);
6396 /* remove us from the free space cache (if we're there at all) */
6397 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6398 btrfs_put_block_group(cache);
6402 static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
6403 u64 start, u64 num_bytes)
6406 struct btrfs_block_group_cache *block_group;
6407 struct btrfs_caching_control *caching_ctl;
6409 block_group = btrfs_lookup_block_group(fs_info, start);
6413 cache_block_group(block_group, 0);
6414 caching_ctl = get_caching_control(block_group);
6418 BUG_ON(!block_group_cache_done(block_group));
6419 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6421 mutex_lock(&caching_ctl->mutex);
6423 if (start >= caching_ctl->progress) {
6424 ret = add_excluded_extent(fs_info, start, num_bytes);
6425 } else if (start + num_bytes <= caching_ctl->progress) {
6426 ret = btrfs_remove_free_space(block_group,
6429 num_bytes = caching_ctl->progress - start;
6430 ret = btrfs_remove_free_space(block_group,
6435 num_bytes = (start + num_bytes) -
6436 caching_ctl->progress;
6437 start = caching_ctl->progress;
6438 ret = add_excluded_extent(fs_info, start, num_bytes);
6441 mutex_unlock(&caching_ctl->mutex);
6442 put_caching_control(caching_ctl);
6444 btrfs_put_block_group(block_group);
6448 int btrfs_exclude_logged_extents(struct btrfs_fs_info *fs_info,
6449 struct extent_buffer *eb)
6451 struct btrfs_file_extent_item *item;
6452 struct btrfs_key key;
6457 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
6460 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6461 btrfs_item_key_to_cpu(eb, &key, i);
6462 if (key.type != BTRFS_EXTENT_DATA_KEY)
6464 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6465 found_type = btrfs_file_extent_type(eb, item);
6466 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6468 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6470 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6471 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6472 ret = __exclude_logged_extent(fs_info, key.objectid, key.offset);
6481 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg)
6483 atomic_inc(&bg->reservations);
6486 void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info,
6489 struct btrfs_block_group_cache *bg;
6491 bg = btrfs_lookup_block_group(fs_info, start);
6493 if (atomic_dec_and_test(&bg->reservations))
6494 wake_up_var(&bg->reservations);
6495 btrfs_put_block_group(bg);
6498 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg)
6500 struct btrfs_space_info *space_info = bg->space_info;
6504 if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA))
6508 * Our block group is read only but before we set it to read only,
6509 * some task might have had allocated an extent from it already, but it
6510 * has not yet created a respective ordered extent (and added it to a
6511 * root's list of ordered extents).
6512 * Therefore wait for any task currently allocating extents, since the
6513 * block group's reservations counter is incremented while a read lock
6514 * on the groups' semaphore is held and decremented after releasing
6515 * the read access on that semaphore and creating the ordered extent.
6517 down_write(&space_info->groups_sem);
6518 up_write(&space_info->groups_sem);
6520 wait_var_event(&bg->reservations, !atomic_read(&bg->reservations));
6524 * btrfs_add_reserved_bytes - update the block_group and space info counters
6525 * @cache: The cache we are manipulating
6526 * @ram_bytes: The number of bytes of file content, and will be same to
6527 * @num_bytes except for the compress path.
6528 * @num_bytes: The number of bytes in question
6529 * @delalloc: The blocks are allocated for the delalloc write
6531 * This is called by the allocator when it reserves space. If this is a
6532 * reservation and the block group has become read only we cannot make the
6533 * reservation and return -EAGAIN, otherwise this function always succeeds.
6535 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache,
6536 u64 ram_bytes, u64 num_bytes, int delalloc)
6538 struct btrfs_space_info *space_info = cache->space_info;
6541 spin_lock(&space_info->lock);
6542 spin_lock(&cache->lock);
6546 cache->reserved += num_bytes;
6547 space_info->bytes_reserved += num_bytes;
6549 trace_btrfs_space_reservation(cache->fs_info,
6550 "space_info", space_info->flags,
6552 space_info->bytes_may_use -= ram_bytes;
6554 cache->delalloc_bytes += num_bytes;
6556 spin_unlock(&cache->lock);
6557 spin_unlock(&space_info->lock);
6562 * btrfs_free_reserved_bytes - update the block_group and space info counters
6563 * @cache: The cache we are manipulating
6564 * @num_bytes: The number of bytes in question
6565 * @delalloc: The blocks are allocated for the delalloc write
6567 * This is called by somebody who is freeing space that was never actually used
6568 * on disk. For example if you reserve some space for a new leaf in transaction
6569 * A and before transaction A commits you free that leaf, you call this with
6570 * reserve set to 0 in order to clear the reservation.
6573 static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache,
6574 u64 num_bytes, int delalloc)
6576 struct btrfs_space_info *space_info = cache->space_info;
6579 spin_lock(&space_info->lock);
6580 spin_lock(&cache->lock);
6582 space_info->bytes_readonly += num_bytes;
6583 cache->reserved -= num_bytes;
6584 space_info->bytes_reserved -= num_bytes;
6587 cache->delalloc_bytes -= num_bytes;
6588 spin_unlock(&cache->lock);
6589 spin_unlock(&space_info->lock);
6592 void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info)
6594 struct btrfs_caching_control *next;
6595 struct btrfs_caching_control *caching_ctl;
6596 struct btrfs_block_group_cache *cache;
6598 down_write(&fs_info->commit_root_sem);
6600 list_for_each_entry_safe(caching_ctl, next,
6601 &fs_info->caching_block_groups, list) {
6602 cache = caching_ctl->block_group;
6603 if (block_group_cache_done(cache)) {
6604 cache->last_byte_to_unpin = (u64)-1;
6605 list_del_init(&caching_ctl->list);
6606 put_caching_control(caching_ctl);
6608 cache->last_byte_to_unpin = caching_ctl->progress;
6612 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6613 fs_info->pinned_extents = &fs_info->freed_extents[1];
6615 fs_info->pinned_extents = &fs_info->freed_extents[0];
6617 up_write(&fs_info->commit_root_sem);
6619 update_global_block_rsv(fs_info);
6623 * Returns the free cluster for the given space info and sets empty_cluster to
6624 * what it should be based on the mount options.
6626 static struct btrfs_free_cluster *
6627 fetch_cluster_info(struct btrfs_fs_info *fs_info,
6628 struct btrfs_space_info *space_info, u64 *empty_cluster)
6630 struct btrfs_free_cluster *ret = NULL;
6633 if (btrfs_mixed_space_info(space_info))
6636 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
6637 ret = &fs_info->meta_alloc_cluster;
6638 if (btrfs_test_opt(fs_info, SSD))
6639 *empty_cluster = SZ_2M;
6641 *empty_cluster = SZ_64K;
6642 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
6643 btrfs_test_opt(fs_info, SSD_SPREAD)) {
6644 *empty_cluster = SZ_2M;
6645 ret = &fs_info->data_alloc_cluster;
6651 static int unpin_extent_range(struct btrfs_fs_info *fs_info,
6653 const bool return_free_space)
6655 struct btrfs_block_group_cache *cache = NULL;
6656 struct btrfs_space_info *space_info;
6657 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6658 struct btrfs_free_cluster *cluster = NULL;
6660 u64 total_unpinned = 0;
6661 u64 empty_cluster = 0;
6664 while (start <= end) {
6667 start >= cache->key.objectid + cache->key.offset) {
6669 btrfs_put_block_group(cache);
6671 cache = btrfs_lookup_block_group(fs_info, start);
6672 BUG_ON(!cache); /* Logic error */
6674 cluster = fetch_cluster_info(fs_info,
6677 empty_cluster <<= 1;
6680 len = cache->key.objectid + cache->key.offset - start;
6681 len = min(len, end + 1 - start);
6683 if (start < cache->last_byte_to_unpin) {
6684 len = min(len, cache->last_byte_to_unpin - start);
6685 if (return_free_space)
6686 btrfs_add_free_space(cache, start, len);
6690 total_unpinned += len;
6691 space_info = cache->space_info;
6694 * If this space cluster has been marked as fragmented and we've
6695 * unpinned enough in this block group to potentially allow a
6696 * cluster to be created inside of it go ahead and clear the
6699 if (cluster && cluster->fragmented &&
6700 total_unpinned > empty_cluster) {
6701 spin_lock(&cluster->lock);
6702 cluster->fragmented = 0;
6703 spin_unlock(&cluster->lock);
6706 spin_lock(&space_info->lock);
6707 spin_lock(&cache->lock);
6708 cache->pinned -= len;
6709 space_info->bytes_pinned -= len;
6711 trace_btrfs_space_reservation(fs_info, "pinned",
6712 space_info->flags, len, 0);
6713 space_info->max_extent_size = 0;
6714 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6716 space_info->bytes_readonly += len;
6719 spin_unlock(&cache->lock);
6720 if (!readonly && return_free_space &&
6721 global_rsv->space_info == space_info) {
6724 spin_lock(&global_rsv->lock);
6725 if (!global_rsv->full) {
6726 to_add = min(len, global_rsv->size -
6727 global_rsv->reserved);
6728 global_rsv->reserved += to_add;
6729 space_info->bytes_may_use += to_add;
6730 if (global_rsv->reserved >= global_rsv->size)
6731 global_rsv->full = 1;
6732 trace_btrfs_space_reservation(fs_info,
6738 spin_unlock(&global_rsv->lock);
6739 /* Add to any tickets we may have */
6741 space_info_add_new_bytes(fs_info, space_info,
6744 spin_unlock(&space_info->lock);
6748 btrfs_put_block_group(cache);
6752 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans)
6754 struct btrfs_fs_info *fs_info = trans->fs_info;
6755 struct btrfs_block_group_cache *block_group, *tmp;
6756 struct list_head *deleted_bgs;
6757 struct extent_io_tree *unpin;
6762 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6763 unpin = &fs_info->freed_extents[1];
6765 unpin = &fs_info->freed_extents[0];
6767 while (!trans->aborted) {
6768 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6769 ret = find_first_extent_bit(unpin, 0, &start, &end,
6770 EXTENT_DIRTY, NULL);
6772 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6776 if (btrfs_test_opt(fs_info, DISCARD))
6777 ret = btrfs_discard_extent(fs_info, start,
6778 end + 1 - start, NULL);
6780 clear_extent_dirty(unpin, start, end);
6781 unpin_extent_range(fs_info, start, end, true);
6782 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6787 * Transaction is finished. We don't need the lock anymore. We
6788 * do need to clean up the block groups in case of a transaction
6791 deleted_bgs = &trans->transaction->deleted_bgs;
6792 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6796 if (!trans->aborted)
6797 ret = btrfs_discard_extent(fs_info,
6798 block_group->key.objectid,
6799 block_group->key.offset,
6802 list_del_init(&block_group->bg_list);
6803 btrfs_put_block_group_trimming(block_group);
6804 btrfs_put_block_group(block_group);
6807 const char *errstr = btrfs_decode_error(ret);
6809 "discard failed while removing blockgroup: errno=%d %s",
6817 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6818 struct btrfs_fs_info *info,
6819 struct btrfs_delayed_ref_node *node, u64 parent,
6820 u64 root_objectid, u64 owner_objectid,
6821 u64 owner_offset, int refs_to_drop,
6822 struct btrfs_delayed_extent_op *extent_op)
6824 struct btrfs_key key;
6825 struct btrfs_path *path;
6826 struct btrfs_root *extent_root = info->extent_root;
6827 struct extent_buffer *leaf;
6828 struct btrfs_extent_item *ei;
6829 struct btrfs_extent_inline_ref *iref;
6832 int extent_slot = 0;
6833 int found_extent = 0;
6837 u64 bytenr = node->bytenr;
6838 u64 num_bytes = node->num_bytes;
6840 bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
6842 path = btrfs_alloc_path();
6846 path->reada = READA_FORWARD;
6847 path->leave_spinning = 1;
6849 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6850 BUG_ON(!is_data && refs_to_drop != 1);
6853 skinny_metadata = false;
6855 ret = lookup_extent_backref(trans, info, path, &iref,
6856 bytenr, num_bytes, parent,
6857 root_objectid, owner_objectid,
6860 extent_slot = path->slots[0];
6861 while (extent_slot >= 0) {
6862 btrfs_item_key_to_cpu(path->nodes[0], &key,
6864 if (key.objectid != bytenr)
6866 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6867 key.offset == num_bytes) {
6871 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6872 key.offset == owner_objectid) {
6876 if (path->slots[0] - extent_slot > 5)
6880 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6881 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6882 if (found_extent && item_size < sizeof(*ei))
6885 if (!found_extent) {
6887 ret = remove_extent_backref(trans, info, path, NULL,
6889 is_data, &last_ref);
6891 btrfs_abort_transaction(trans, ret);
6894 btrfs_release_path(path);
6895 path->leave_spinning = 1;
6897 key.objectid = bytenr;
6898 key.type = BTRFS_EXTENT_ITEM_KEY;
6899 key.offset = num_bytes;
6901 if (!is_data && skinny_metadata) {
6902 key.type = BTRFS_METADATA_ITEM_KEY;
6903 key.offset = owner_objectid;
6906 ret = btrfs_search_slot(trans, extent_root,
6908 if (ret > 0 && skinny_metadata && path->slots[0]) {
6910 * Couldn't find our skinny metadata item,
6911 * see if we have ye olde extent item.
6914 btrfs_item_key_to_cpu(path->nodes[0], &key,
6916 if (key.objectid == bytenr &&
6917 key.type == BTRFS_EXTENT_ITEM_KEY &&
6918 key.offset == num_bytes)
6922 if (ret > 0 && skinny_metadata) {
6923 skinny_metadata = false;
6924 key.objectid = bytenr;
6925 key.type = BTRFS_EXTENT_ITEM_KEY;
6926 key.offset = num_bytes;
6927 btrfs_release_path(path);
6928 ret = btrfs_search_slot(trans, extent_root,
6934 "umm, got %d back from search, was looking for %llu",
6937 btrfs_print_leaf(path->nodes[0]);
6940 btrfs_abort_transaction(trans, ret);
6943 extent_slot = path->slots[0];
6945 } else if (WARN_ON(ret == -ENOENT)) {
6946 btrfs_print_leaf(path->nodes[0]);
6948 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6949 bytenr, parent, root_objectid, owner_objectid,
6951 btrfs_abort_transaction(trans, ret);
6954 btrfs_abort_transaction(trans, ret);
6958 leaf = path->nodes[0];
6959 item_size = btrfs_item_size_nr(leaf, extent_slot);
6960 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6961 if (item_size < sizeof(*ei)) {
6962 BUG_ON(found_extent || extent_slot != path->slots[0]);
6963 ret = convert_extent_item_v0(trans, info, path, owner_objectid,
6966 btrfs_abort_transaction(trans, ret);
6970 btrfs_release_path(path);
6971 path->leave_spinning = 1;
6973 key.objectid = bytenr;
6974 key.type = BTRFS_EXTENT_ITEM_KEY;
6975 key.offset = num_bytes;
6977 ret = btrfs_search_slot(trans, extent_root, &key, path,
6981 "umm, got %d back from search, was looking for %llu",
6983 btrfs_print_leaf(path->nodes[0]);
6986 btrfs_abort_transaction(trans, ret);
6990 extent_slot = path->slots[0];
6991 leaf = path->nodes[0];
6992 item_size = btrfs_item_size_nr(leaf, extent_slot);
6995 BUG_ON(item_size < sizeof(*ei));
6996 ei = btrfs_item_ptr(leaf, extent_slot,
6997 struct btrfs_extent_item);
6998 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
6999 key.type == BTRFS_EXTENT_ITEM_KEY) {
7000 struct btrfs_tree_block_info *bi;
7001 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
7002 bi = (struct btrfs_tree_block_info *)(ei + 1);
7003 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
7006 refs = btrfs_extent_refs(leaf, ei);
7007 if (refs < refs_to_drop) {
7009 "trying to drop %d refs but we only have %Lu for bytenr %Lu",
7010 refs_to_drop, refs, bytenr);
7012 btrfs_abort_transaction(trans, ret);
7015 refs -= refs_to_drop;
7019 __run_delayed_extent_op(extent_op, leaf, ei);
7021 * In the case of inline back ref, reference count will
7022 * be updated by remove_extent_backref
7025 BUG_ON(!found_extent);
7027 btrfs_set_extent_refs(leaf, ei, refs);
7028 btrfs_mark_buffer_dirty(leaf);
7031 ret = remove_extent_backref(trans, info, path,
7033 is_data, &last_ref);
7035 btrfs_abort_transaction(trans, ret);
7041 BUG_ON(is_data && refs_to_drop !=
7042 extent_data_ref_count(path, iref));
7044 BUG_ON(path->slots[0] != extent_slot);
7046 BUG_ON(path->slots[0] != extent_slot + 1);
7047 path->slots[0] = extent_slot;
7053 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
7056 btrfs_abort_transaction(trans, ret);
7059 btrfs_release_path(path);
7062 ret = btrfs_del_csums(trans, info, bytenr, num_bytes);
7064 btrfs_abort_transaction(trans, ret);
7069 ret = add_to_free_space_tree(trans, bytenr, num_bytes);
7071 btrfs_abort_transaction(trans, ret);
7075 ret = update_block_group(trans, info, bytenr, num_bytes, 0);
7077 btrfs_abort_transaction(trans, ret);
7081 btrfs_release_path(path);
7084 btrfs_free_path(path);
7089 * when we free an block, it is possible (and likely) that we free the last
7090 * delayed ref for that extent as well. This searches the delayed ref tree for
7091 * a given extent, and if there are no other delayed refs to be processed, it
7092 * removes it from the tree.
7094 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
7097 struct btrfs_delayed_ref_head *head;
7098 struct btrfs_delayed_ref_root *delayed_refs;
7101 delayed_refs = &trans->transaction->delayed_refs;
7102 spin_lock(&delayed_refs->lock);
7103 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
7105 goto out_delayed_unlock;
7107 spin_lock(&head->lock);
7108 if (!RB_EMPTY_ROOT(&head->ref_tree))
7111 if (head->extent_op) {
7112 if (!head->must_insert_reserved)
7114 btrfs_free_delayed_extent_op(head->extent_op);
7115 head->extent_op = NULL;
7119 * waiting for the lock here would deadlock. If someone else has it
7120 * locked they are already in the process of dropping it anyway
7122 if (!mutex_trylock(&head->mutex))
7126 * at this point we have a head with no other entries. Go
7127 * ahead and process it.
7129 rb_erase(&head->href_node, &delayed_refs->href_root);
7130 RB_CLEAR_NODE(&head->href_node);
7131 atomic_dec(&delayed_refs->num_entries);
7134 * we don't take a ref on the node because we're removing it from the
7135 * tree, so we just steal the ref the tree was holding.
7137 delayed_refs->num_heads--;
7138 if (head->processing == 0)
7139 delayed_refs->num_heads_ready--;
7140 head->processing = 0;
7141 spin_unlock(&head->lock);
7142 spin_unlock(&delayed_refs->lock);
7144 BUG_ON(head->extent_op);
7145 if (head->must_insert_reserved)
7148 mutex_unlock(&head->mutex);
7149 btrfs_put_delayed_ref_head(head);
7152 spin_unlock(&head->lock);
7155 spin_unlock(&delayed_refs->lock);
7159 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
7160 struct btrfs_root *root,
7161 struct extent_buffer *buf,
7162 u64 parent, int last_ref)
7164 struct btrfs_fs_info *fs_info = root->fs_info;
7168 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7169 int old_ref_mod, new_ref_mod;
7171 btrfs_ref_tree_mod(root, buf->start, buf->len, parent,
7172 root->root_key.objectid,
7173 btrfs_header_level(buf), 0,
7174 BTRFS_DROP_DELAYED_REF);
7175 ret = btrfs_add_delayed_tree_ref(fs_info, trans, buf->start,
7177 root->root_key.objectid,
7178 btrfs_header_level(buf),
7179 BTRFS_DROP_DELAYED_REF, NULL,
7180 &old_ref_mod, &new_ref_mod);
7181 BUG_ON(ret); /* -ENOMEM */
7182 pin = old_ref_mod >= 0 && new_ref_mod < 0;
7185 if (last_ref && btrfs_header_generation(buf) == trans->transid) {
7186 struct btrfs_block_group_cache *cache;
7188 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7189 ret = check_ref_cleanup(trans, buf->start);
7195 cache = btrfs_lookup_block_group(fs_info, buf->start);
7197 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
7198 pin_down_extent(fs_info, cache, buf->start,
7200 btrfs_put_block_group(cache);
7204 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
7206 btrfs_add_free_space(cache, buf->start, buf->len);
7207 btrfs_free_reserved_bytes(cache, buf->len, 0);
7208 btrfs_put_block_group(cache);
7209 trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
7213 add_pinned_bytes(fs_info, buf->len, true,
7214 root->root_key.objectid);
7218 * Deleting the buffer, clear the corrupt flag since it doesn't
7221 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
7225 /* Can return -ENOMEM */
7226 int btrfs_free_extent(struct btrfs_trans_handle *trans,
7227 struct btrfs_root *root,
7228 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
7229 u64 owner, u64 offset)
7231 struct btrfs_fs_info *fs_info = root->fs_info;
7232 int old_ref_mod, new_ref_mod;
7235 if (btrfs_is_testing(fs_info))
7238 if (root_objectid != BTRFS_TREE_LOG_OBJECTID)
7239 btrfs_ref_tree_mod(root, bytenr, num_bytes, parent,
7240 root_objectid, owner, offset,
7241 BTRFS_DROP_DELAYED_REF);
7244 * tree log blocks never actually go into the extent allocation
7245 * tree, just update pinning info and exit early.
7247 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
7248 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
7249 /* unlocks the pinned mutex */
7250 btrfs_pin_extent(fs_info, bytenr, num_bytes, 1);
7251 old_ref_mod = new_ref_mod = 0;
7253 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
7254 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
7256 root_objectid, (int)owner,
7257 BTRFS_DROP_DELAYED_REF, NULL,
7258 &old_ref_mod, &new_ref_mod);
7260 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
7262 root_objectid, owner, offset,
7263 0, BTRFS_DROP_DELAYED_REF,
7264 &old_ref_mod, &new_ref_mod);
7267 if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0) {
7268 bool metadata = owner < BTRFS_FIRST_FREE_OBJECTID;
7270 add_pinned_bytes(fs_info, num_bytes, metadata, root_objectid);
7277 * when we wait for progress in the block group caching, its because
7278 * our allocation attempt failed at least once. So, we must sleep
7279 * and let some progress happen before we try again.
7281 * This function will sleep at least once waiting for new free space to
7282 * show up, and then it will check the block group free space numbers
7283 * for our min num_bytes. Another option is to have it go ahead
7284 * and look in the rbtree for a free extent of a given size, but this
7287 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7288 * any of the information in this block group.
7290 static noinline void
7291 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
7294 struct btrfs_caching_control *caching_ctl;
7296 caching_ctl = get_caching_control(cache);
7300 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
7301 (cache->free_space_ctl->free_space >= num_bytes));
7303 put_caching_control(caching_ctl);
7307 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
7309 struct btrfs_caching_control *caching_ctl;
7312 caching_ctl = get_caching_control(cache);
7314 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
7316 wait_event(caching_ctl->wait, block_group_cache_done(cache));
7317 if (cache->cached == BTRFS_CACHE_ERROR)
7319 put_caching_control(caching_ctl);
7323 enum btrfs_loop_type {
7324 LOOP_CACHING_NOWAIT = 0,
7325 LOOP_CACHING_WAIT = 1,
7326 LOOP_ALLOC_CHUNK = 2,
7327 LOOP_NO_EMPTY_SIZE = 3,
7331 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
7335 down_read(&cache->data_rwsem);
7339 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
7342 btrfs_get_block_group(cache);
7344 down_read(&cache->data_rwsem);
7347 static struct btrfs_block_group_cache *
7348 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
7349 struct btrfs_free_cluster *cluster,
7352 struct btrfs_block_group_cache *used_bg = NULL;
7354 spin_lock(&cluster->refill_lock);
7356 used_bg = cluster->block_group;
7360 if (used_bg == block_group)
7363 btrfs_get_block_group(used_bg);
7368 if (down_read_trylock(&used_bg->data_rwsem))
7371 spin_unlock(&cluster->refill_lock);
7373 /* We should only have one-level nested. */
7374 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
7376 spin_lock(&cluster->refill_lock);
7377 if (used_bg == cluster->block_group)
7380 up_read(&used_bg->data_rwsem);
7381 btrfs_put_block_group(used_bg);
7386 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
7390 up_read(&cache->data_rwsem);
7391 btrfs_put_block_group(cache);
7395 * walks the btree of allocated extents and find a hole of a given size.
7396 * The key ins is changed to record the hole:
7397 * ins->objectid == start position
7398 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7399 * ins->offset == the size of the hole.
7400 * Any available blocks before search_start are skipped.
7402 * If there is no suitable free space, we will record the max size of
7403 * the free space extent currently.
7405 static noinline int find_free_extent(struct btrfs_fs_info *fs_info,
7406 u64 ram_bytes, u64 num_bytes, u64 empty_size,
7407 u64 hint_byte, struct btrfs_key *ins,
7408 u64 flags, int delalloc)
7411 struct btrfs_root *root = fs_info->extent_root;
7412 struct btrfs_free_cluster *last_ptr = NULL;
7413 struct btrfs_block_group_cache *block_group = NULL;
7414 u64 search_start = 0;
7415 u64 max_extent_size = 0;
7416 u64 empty_cluster = 0;
7417 struct btrfs_space_info *space_info;
7419 int index = btrfs_bg_flags_to_raid_index(flags);
7420 bool failed_cluster_refill = false;
7421 bool failed_alloc = false;
7422 bool use_cluster = true;
7423 bool have_caching_bg = false;
7424 bool orig_have_caching_bg = false;
7425 bool full_search = false;
7427 WARN_ON(num_bytes < fs_info->sectorsize);
7428 ins->type = BTRFS_EXTENT_ITEM_KEY;
7432 trace_find_free_extent(fs_info, num_bytes, empty_size, flags);
7434 space_info = __find_space_info(fs_info, flags);
7436 btrfs_err(fs_info, "No space info for %llu", flags);
7441 * If our free space is heavily fragmented we may not be able to make
7442 * big contiguous allocations, so instead of doing the expensive search
7443 * for free space, simply return ENOSPC with our max_extent_size so we
7444 * can go ahead and search for a more manageable chunk.
7446 * If our max_extent_size is large enough for our allocation simply
7447 * disable clustering since we will likely not be able to find enough
7448 * space to create a cluster and induce latency trying.
7450 if (unlikely(space_info->max_extent_size)) {
7451 spin_lock(&space_info->lock);
7452 if (space_info->max_extent_size &&
7453 num_bytes > space_info->max_extent_size) {
7454 ins->offset = space_info->max_extent_size;
7455 spin_unlock(&space_info->lock);
7457 } else if (space_info->max_extent_size) {
7458 use_cluster = false;
7460 spin_unlock(&space_info->lock);
7463 last_ptr = fetch_cluster_info(fs_info, space_info, &empty_cluster);
7465 spin_lock(&last_ptr->lock);
7466 if (last_ptr->block_group)
7467 hint_byte = last_ptr->window_start;
7468 if (last_ptr->fragmented) {
7470 * We still set window_start so we can keep track of the
7471 * last place we found an allocation to try and save
7474 hint_byte = last_ptr->window_start;
7475 use_cluster = false;
7477 spin_unlock(&last_ptr->lock);
7480 search_start = max(search_start, first_logical_byte(fs_info, 0));
7481 search_start = max(search_start, hint_byte);
7482 if (search_start == hint_byte) {
7483 block_group = btrfs_lookup_block_group(fs_info, search_start);
7485 * we don't want to use the block group if it doesn't match our
7486 * allocation bits, or if its not cached.
7488 * However if we are re-searching with an ideal block group
7489 * picked out then we don't care that the block group is cached.
7491 if (block_group && block_group_bits(block_group, flags) &&
7492 block_group->cached != BTRFS_CACHE_NO) {
7493 down_read(&space_info->groups_sem);
7494 if (list_empty(&block_group->list) ||
7497 * someone is removing this block group,
7498 * we can't jump into the have_block_group
7499 * target because our list pointers are not
7502 btrfs_put_block_group(block_group);
7503 up_read(&space_info->groups_sem);
7505 index = btrfs_bg_flags_to_raid_index(
7506 block_group->flags);
7507 btrfs_lock_block_group(block_group, delalloc);
7508 goto have_block_group;
7510 } else if (block_group) {
7511 btrfs_put_block_group(block_group);
7515 have_caching_bg = false;
7516 if (index == 0 || index == btrfs_bg_flags_to_raid_index(flags))
7518 down_read(&space_info->groups_sem);
7519 list_for_each_entry(block_group, &space_info->block_groups[index],
7524 /* If the block group is read-only, we can skip it entirely. */
7525 if (unlikely(block_group->ro))
7528 btrfs_grab_block_group(block_group, delalloc);
7529 search_start = block_group->key.objectid;
7532 * this can happen if we end up cycling through all the
7533 * raid types, but we want to make sure we only allocate
7534 * for the proper type.
7536 if (!block_group_bits(block_group, flags)) {
7537 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7538 BTRFS_BLOCK_GROUP_RAID1 |
7539 BTRFS_BLOCK_GROUP_RAID5 |
7540 BTRFS_BLOCK_GROUP_RAID6 |
7541 BTRFS_BLOCK_GROUP_RAID10;
7544 * if they asked for extra copies and this block group
7545 * doesn't provide them, bail. This does allow us to
7546 * fill raid0 from raid1.
7548 if ((flags & extra) && !(block_group->flags & extra))
7553 cached = block_group_cache_done(block_group);
7554 if (unlikely(!cached)) {
7555 have_caching_bg = true;
7556 ret = cache_block_group(block_group, 0);
7561 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7565 * Ok we want to try and use the cluster allocator, so
7568 if (last_ptr && use_cluster) {
7569 struct btrfs_block_group_cache *used_block_group;
7570 unsigned long aligned_cluster;
7572 * the refill lock keeps out other
7573 * people trying to start a new cluster
7575 used_block_group = btrfs_lock_cluster(block_group,
7578 if (!used_block_group)
7579 goto refill_cluster;
7581 if (used_block_group != block_group &&
7582 (used_block_group->ro ||
7583 !block_group_bits(used_block_group, flags)))
7584 goto release_cluster;
7586 offset = btrfs_alloc_from_cluster(used_block_group,
7589 used_block_group->key.objectid,
7592 /* we have a block, we're done */
7593 spin_unlock(&last_ptr->refill_lock);
7594 trace_btrfs_reserve_extent_cluster(
7596 search_start, num_bytes);
7597 if (used_block_group != block_group) {
7598 btrfs_release_block_group(block_group,
7600 block_group = used_block_group;
7605 WARN_ON(last_ptr->block_group != used_block_group);
7607 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7608 * set up a new clusters, so lets just skip it
7609 * and let the allocator find whatever block
7610 * it can find. If we reach this point, we
7611 * will have tried the cluster allocator
7612 * plenty of times and not have found
7613 * anything, so we are likely way too
7614 * fragmented for the clustering stuff to find
7617 * However, if the cluster is taken from the
7618 * current block group, release the cluster
7619 * first, so that we stand a better chance of
7620 * succeeding in the unclustered
7622 if (loop >= LOOP_NO_EMPTY_SIZE &&
7623 used_block_group != block_group) {
7624 spin_unlock(&last_ptr->refill_lock);
7625 btrfs_release_block_group(used_block_group,
7627 goto unclustered_alloc;
7631 * this cluster didn't work out, free it and
7634 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7636 if (used_block_group != block_group)
7637 btrfs_release_block_group(used_block_group,
7640 if (loop >= LOOP_NO_EMPTY_SIZE) {
7641 spin_unlock(&last_ptr->refill_lock);
7642 goto unclustered_alloc;
7645 aligned_cluster = max_t(unsigned long,
7646 empty_cluster + empty_size,
7647 block_group->full_stripe_len);
7649 /* allocate a cluster in this block group */
7650 ret = btrfs_find_space_cluster(fs_info, block_group,
7651 last_ptr, search_start,
7656 * now pull our allocation out of this
7659 offset = btrfs_alloc_from_cluster(block_group,
7665 /* we found one, proceed */
7666 spin_unlock(&last_ptr->refill_lock);
7667 trace_btrfs_reserve_extent_cluster(
7668 block_group, search_start,
7672 } else if (!cached && loop > LOOP_CACHING_NOWAIT
7673 && !failed_cluster_refill) {
7674 spin_unlock(&last_ptr->refill_lock);
7676 failed_cluster_refill = true;
7677 wait_block_group_cache_progress(block_group,
7678 num_bytes + empty_cluster + empty_size);
7679 goto have_block_group;
7683 * at this point we either didn't find a cluster
7684 * or we weren't able to allocate a block from our
7685 * cluster. Free the cluster we've been trying
7686 * to use, and go to the next block group
7688 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7689 spin_unlock(&last_ptr->refill_lock);
7695 * We are doing an unclustered alloc, set the fragmented flag so
7696 * we don't bother trying to setup a cluster again until we get
7699 if (unlikely(last_ptr)) {
7700 spin_lock(&last_ptr->lock);
7701 last_ptr->fragmented = 1;
7702 spin_unlock(&last_ptr->lock);
7705 struct btrfs_free_space_ctl *ctl =
7706 block_group->free_space_ctl;
7708 spin_lock(&ctl->tree_lock);
7709 if (ctl->free_space <
7710 num_bytes + empty_cluster + empty_size) {
7711 if (ctl->free_space > max_extent_size)
7712 max_extent_size = ctl->free_space;
7713 spin_unlock(&ctl->tree_lock);
7716 spin_unlock(&ctl->tree_lock);
7719 offset = btrfs_find_space_for_alloc(block_group, search_start,
7720 num_bytes, empty_size,
7723 * If we didn't find a chunk, and we haven't failed on this
7724 * block group before, and this block group is in the middle of
7725 * caching and we are ok with waiting, then go ahead and wait
7726 * for progress to be made, and set failed_alloc to true.
7728 * If failed_alloc is true then we've already waited on this
7729 * block group once and should move on to the next block group.
7731 if (!offset && !failed_alloc && !cached &&
7732 loop > LOOP_CACHING_NOWAIT) {
7733 wait_block_group_cache_progress(block_group,
7734 num_bytes + empty_size);
7735 failed_alloc = true;
7736 goto have_block_group;
7737 } else if (!offset) {
7741 search_start = ALIGN(offset, fs_info->stripesize);
7743 /* move on to the next group */
7744 if (search_start + num_bytes >
7745 block_group->key.objectid + block_group->key.offset) {
7746 btrfs_add_free_space(block_group, offset, num_bytes);
7750 if (offset < search_start)
7751 btrfs_add_free_space(block_group, offset,
7752 search_start - offset);
7753 BUG_ON(offset > search_start);
7755 ret = btrfs_add_reserved_bytes(block_group, ram_bytes,
7756 num_bytes, delalloc);
7757 if (ret == -EAGAIN) {
7758 btrfs_add_free_space(block_group, offset, num_bytes);
7761 btrfs_inc_block_group_reservations(block_group);
7763 /* we are all good, lets return */
7764 ins->objectid = search_start;
7765 ins->offset = num_bytes;
7767 trace_btrfs_reserve_extent(block_group, search_start, num_bytes);
7768 btrfs_release_block_group(block_group, delalloc);
7771 failed_cluster_refill = false;
7772 failed_alloc = false;
7773 BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) !=
7775 btrfs_release_block_group(block_group, delalloc);
7778 up_read(&space_info->groups_sem);
7780 if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg
7781 && !orig_have_caching_bg)
7782 orig_have_caching_bg = true;
7784 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7787 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7791 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7792 * caching kthreads as we move along
7793 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7794 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7795 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7798 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7800 if (loop == LOOP_CACHING_NOWAIT) {
7802 * We want to skip the LOOP_CACHING_WAIT step if we
7803 * don't have any uncached bgs and we've already done a
7804 * full search through.
7806 if (orig_have_caching_bg || !full_search)
7807 loop = LOOP_CACHING_WAIT;
7809 loop = LOOP_ALLOC_CHUNK;
7814 if (loop == LOOP_ALLOC_CHUNK) {
7815 struct btrfs_trans_handle *trans;
7818 trans = current->journal_info;
7822 trans = btrfs_join_transaction(root);
7824 if (IS_ERR(trans)) {
7825 ret = PTR_ERR(trans);
7829 ret = do_chunk_alloc(trans, fs_info, flags,
7833 * If we can't allocate a new chunk we've already looped
7834 * through at least once, move on to the NO_EMPTY_SIZE
7838 loop = LOOP_NO_EMPTY_SIZE;
7841 * Do not bail out on ENOSPC since we
7842 * can do more things.
7844 if (ret < 0 && ret != -ENOSPC)
7845 btrfs_abort_transaction(trans, ret);
7849 btrfs_end_transaction(trans);
7854 if (loop == LOOP_NO_EMPTY_SIZE) {
7856 * Don't loop again if we already have no empty_size and
7859 if (empty_size == 0 &&
7860 empty_cluster == 0) {
7869 } else if (!ins->objectid) {
7871 } else if (ins->objectid) {
7872 if (!use_cluster && last_ptr) {
7873 spin_lock(&last_ptr->lock);
7874 last_ptr->window_start = ins->objectid;
7875 spin_unlock(&last_ptr->lock);
7880 if (ret == -ENOSPC) {
7881 spin_lock(&space_info->lock);
7882 space_info->max_extent_size = max_extent_size;
7883 spin_unlock(&space_info->lock);
7884 ins->offset = max_extent_size;
7889 static void dump_space_info(struct btrfs_fs_info *fs_info,
7890 struct btrfs_space_info *info, u64 bytes,
7891 int dump_block_groups)
7893 struct btrfs_block_group_cache *cache;
7896 spin_lock(&info->lock);
7897 btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull",
7899 info->total_bytes - btrfs_space_info_used(info, true),
7900 info->full ? "" : "not ");
7902 "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu",
7903 info->total_bytes, info->bytes_used, info->bytes_pinned,
7904 info->bytes_reserved, info->bytes_may_use,
7905 info->bytes_readonly);
7906 spin_unlock(&info->lock);
7908 if (!dump_block_groups)
7911 down_read(&info->groups_sem);
7913 list_for_each_entry(cache, &info->block_groups[index], list) {
7914 spin_lock(&cache->lock);
7916 "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s",
7917 cache->key.objectid, cache->key.offset,
7918 btrfs_block_group_used(&cache->item), cache->pinned,
7919 cache->reserved, cache->ro ? "[readonly]" : "");
7920 btrfs_dump_free_space(cache, bytes);
7921 spin_unlock(&cache->lock);
7923 if (++index < BTRFS_NR_RAID_TYPES)
7925 up_read(&info->groups_sem);
7929 * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a
7930 * hole that is at least as big as @num_bytes.
7932 * @root - The root that will contain this extent
7934 * @ram_bytes - The amount of space in ram that @num_bytes take. This
7935 * is used for accounting purposes. This value differs
7936 * from @num_bytes only in the case of compressed extents.
7938 * @num_bytes - Number of bytes to allocate on-disk.
7940 * @min_alloc_size - Indicates the minimum amount of space that the
7941 * allocator should try to satisfy. In some cases
7942 * @num_bytes may be larger than what is required and if
7943 * the filesystem is fragmented then allocation fails.
7944 * However, the presence of @min_alloc_size gives a
7945 * chance to try and satisfy the smaller allocation.
7947 * @empty_size - A hint that you plan on doing more COW. This is the
7948 * size in bytes the allocator should try to find free
7949 * next to the block it returns. This is just a hint and
7950 * may be ignored by the allocator.
7952 * @hint_byte - Hint to the allocator to start searching above the byte
7953 * address passed. It might be ignored.
7955 * @ins - This key is modified to record the found hole. It will
7956 * have the following values:
7957 * ins->objectid == start position
7958 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7959 * ins->offset == the size of the hole.
7961 * @is_data - Boolean flag indicating whether an extent is
7962 * allocated for data (true) or metadata (false)
7964 * @delalloc - Boolean flag indicating whether this allocation is for
7965 * delalloc or not. If 'true' data_rwsem of block groups
7966 * is going to be acquired.
7969 * Returns 0 when an allocation succeeded or < 0 when an error occurred. In
7970 * case -ENOSPC is returned then @ins->offset will contain the size of the
7971 * largest available hole the allocator managed to find.
7973 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
7974 u64 num_bytes, u64 min_alloc_size,
7975 u64 empty_size, u64 hint_byte,
7976 struct btrfs_key *ins, int is_data, int delalloc)
7978 struct btrfs_fs_info *fs_info = root->fs_info;
7979 bool final_tried = num_bytes == min_alloc_size;
7983 flags = get_alloc_profile_by_root(root, is_data);
7985 WARN_ON(num_bytes < fs_info->sectorsize);
7986 ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size,
7987 hint_byte, ins, flags, delalloc);
7988 if (!ret && !is_data) {
7989 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
7990 } else if (ret == -ENOSPC) {
7991 if (!final_tried && ins->offset) {
7992 num_bytes = min(num_bytes >> 1, ins->offset);
7993 num_bytes = round_down(num_bytes,
7994 fs_info->sectorsize);
7995 num_bytes = max(num_bytes, min_alloc_size);
7996 ram_bytes = num_bytes;
7997 if (num_bytes == min_alloc_size)
8000 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8001 struct btrfs_space_info *sinfo;
8003 sinfo = __find_space_info(fs_info, flags);
8005 "allocation failed flags %llu, wanted %llu",
8008 dump_space_info(fs_info, sinfo, num_bytes, 1);
8015 static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8017 int pin, int delalloc)
8019 struct btrfs_block_group_cache *cache;
8022 cache = btrfs_lookup_block_group(fs_info, start);
8024 btrfs_err(fs_info, "Unable to find block group for %llu",
8030 pin_down_extent(fs_info, cache, start, len, 1);
8032 if (btrfs_test_opt(fs_info, DISCARD))
8033 ret = btrfs_discard_extent(fs_info, start, len, NULL);
8034 btrfs_add_free_space(cache, start, len);
8035 btrfs_free_reserved_bytes(cache, len, delalloc);
8036 trace_btrfs_reserved_extent_free(fs_info, start, len);
8039 btrfs_put_block_group(cache);
8043 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8044 u64 start, u64 len, int delalloc)
8046 return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc);
8049 int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info,
8052 return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0);
8055 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8056 struct btrfs_fs_info *fs_info,
8057 u64 parent, u64 root_objectid,
8058 u64 flags, u64 owner, u64 offset,
8059 struct btrfs_key *ins, int ref_mod)
8062 struct btrfs_extent_item *extent_item;
8063 struct btrfs_extent_inline_ref *iref;
8064 struct btrfs_path *path;
8065 struct extent_buffer *leaf;
8070 type = BTRFS_SHARED_DATA_REF_KEY;
8072 type = BTRFS_EXTENT_DATA_REF_KEY;
8074 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
8076 path = btrfs_alloc_path();
8080 path->leave_spinning = 1;
8081 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8084 btrfs_free_path(path);
8088 leaf = path->nodes[0];
8089 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8090 struct btrfs_extent_item);
8091 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
8092 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8093 btrfs_set_extent_flags(leaf, extent_item,
8094 flags | BTRFS_EXTENT_FLAG_DATA);
8096 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8097 btrfs_set_extent_inline_ref_type(leaf, iref, type);
8099 struct btrfs_shared_data_ref *ref;
8100 ref = (struct btrfs_shared_data_ref *)(iref + 1);
8101 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8102 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
8104 struct btrfs_extent_data_ref *ref;
8105 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
8106 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
8107 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
8108 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
8109 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
8112 btrfs_mark_buffer_dirty(path->nodes[0]);
8113 btrfs_free_path(path);
8115 ret = remove_from_free_space_tree(trans, ins->objectid, ins->offset);
8119 ret = update_block_group(trans, fs_info, ins->objectid, ins->offset, 1);
8120 if (ret) { /* -ENOENT, logic error */
8121 btrfs_err(fs_info, "update block group failed for %llu %llu",
8122 ins->objectid, ins->offset);
8125 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset);
8129 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
8130 struct btrfs_delayed_ref_node *node,
8131 struct btrfs_delayed_extent_op *extent_op)
8133 struct btrfs_fs_info *fs_info = trans->fs_info;
8135 struct btrfs_extent_item *extent_item;
8136 struct btrfs_key extent_key;
8137 struct btrfs_tree_block_info *block_info;
8138 struct btrfs_extent_inline_ref *iref;
8139 struct btrfs_path *path;
8140 struct extent_buffer *leaf;
8141 struct btrfs_delayed_tree_ref *ref;
8142 u32 size = sizeof(*extent_item) + sizeof(*iref);
8144 u64 flags = extent_op->flags_to_set;
8145 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8147 ref = btrfs_delayed_node_to_tree_ref(node);
8149 extent_key.objectid = node->bytenr;
8150 if (skinny_metadata) {
8151 extent_key.offset = ref->level;
8152 extent_key.type = BTRFS_METADATA_ITEM_KEY;
8153 num_bytes = fs_info->nodesize;
8155 extent_key.offset = node->num_bytes;
8156 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
8157 size += sizeof(*block_info);
8158 num_bytes = node->num_bytes;
8161 path = btrfs_alloc_path();
8163 btrfs_free_and_pin_reserved_extent(fs_info,
8164 extent_key.objectid,
8169 path->leave_spinning = 1;
8170 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8173 btrfs_free_path(path);
8174 btrfs_free_and_pin_reserved_extent(fs_info,
8175 extent_key.objectid,
8180 leaf = path->nodes[0];
8181 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8182 struct btrfs_extent_item);
8183 btrfs_set_extent_refs(leaf, extent_item, 1);
8184 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8185 btrfs_set_extent_flags(leaf, extent_item,
8186 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
8188 if (skinny_metadata) {
8189 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8191 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
8192 btrfs_set_tree_block_key(leaf, block_info, &extent_op->key);
8193 btrfs_set_tree_block_level(leaf, block_info, ref->level);
8194 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
8197 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) {
8198 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
8199 btrfs_set_extent_inline_ref_type(leaf, iref,
8200 BTRFS_SHARED_BLOCK_REF_KEY);
8201 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent);
8203 btrfs_set_extent_inline_ref_type(leaf, iref,
8204 BTRFS_TREE_BLOCK_REF_KEY);
8205 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root);
8208 btrfs_mark_buffer_dirty(leaf);
8209 btrfs_free_path(path);
8211 ret = remove_from_free_space_tree(trans, extent_key.objectid,
8216 ret = update_block_group(trans, fs_info, extent_key.objectid,
8217 fs_info->nodesize, 1);
8218 if (ret) { /* -ENOENT, logic error */
8219 btrfs_err(fs_info, "update block group failed for %llu %llu",
8220 extent_key.objectid, extent_key.offset);
8224 trace_btrfs_reserved_extent_alloc(fs_info, extent_key.objectid,
8229 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8230 struct btrfs_root *root, u64 owner,
8231 u64 offset, u64 ram_bytes,
8232 struct btrfs_key *ins)
8234 struct btrfs_fs_info *fs_info = root->fs_info;
8237 BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
8239 btrfs_ref_tree_mod(root, ins->objectid, ins->offset, 0,
8240 root->root_key.objectid, owner, offset,
8241 BTRFS_ADD_DELAYED_EXTENT);
8243 ret = btrfs_add_delayed_data_ref(fs_info, trans, ins->objectid,
8245 root->root_key.objectid, owner,
8247 BTRFS_ADD_DELAYED_EXTENT, NULL, NULL);
8252 * this is used by the tree logging recovery code. It records that
8253 * an extent has been allocated and makes sure to clear the free
8254 * space cache bits as well
8256 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
8257 struct btrfs_fs_info *fs_info,
8258 u64 root_objectid, u64 owner, u64 offset,
8259 struct btrfs_key *ins)
8262 struct btrfs_block_group_cache *block_group;
8263 struct btrfs_space_info *space_info;
8266 * Mixed block groups will exclude before processing the log so we only
8267 * need to do the exclude dance if this fs isn't mixed.
8269 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
8270 ret = __exclude_logged_extent(fs_info, ins->objectid,
8276 block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
8280 space_info = block_group->space_info;
8281 spin_lock(&space_info->lock);
8282 spin_lock(&block_group->lock);
8283 space_info->bytes_reserved += ins->offset;
8284 block_group->reserved += ins->offset;
8285 spin_unlock(&block_group->lock);
8286 spin_unlock(&space_info->lock);
8288 ret = alloc_reserved_file_extent(trans, fs_info, 0, root_objectid,
8289 0, owner, offset, ins, 1);
8290 btrfs_put_block_group(block_group);
8294 static struct extent_buffer *
8295 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
8296 u64 bytenr, int level)
8298 struct btrfs_fs_info *fs_info = root->fs_info;
8299 struct extent_buffer *buf;
8301 buf = btrfs_find_create_tree_block(fs_info, bytenr);
8305 btrfs_set_header_generation(buf, trans->transid);
8306 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
8307 btrfs_tree_lock(buf);
8308 clean_tree_block(fs_info, buf);
8309 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
8311 btrfs_set_lock_blocking(buf);
8312 set_extent_buffer_uptodate(buf);
8314 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
8315 buf->log_index = root->log_transid % 2;
8317 * we allow two log transactions at a time, use different
8318 * EXENT bit to differentiate dirty pages.
8320 if (buf->log_index == 0)
8321 set_extent_dirty(&root->dirty_log_pages, buf->start,
8322 buf->start + buf->len - 1, GFP_NOFS);
8324 set_extent_new(&root->dirty_log_pages, buf->start,
8325 buf->start + buf->len - 1);
8327 buf->log_index = -1;
8328 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
8329 buf->start + buf->len - 1, GFP_NOFS);
8331 trans->dirty = true;
8332 /* this returns a buffer locked for blocking */
8336 static struct btrfs_block_rsv *
8337 use_block_rsv(struct btrfs_trans_handle *trans,
8338 struct btrfs_root *root, u32 blocksize)
8340 struct btrfs_fs_info *fs_info = root->fs_info;
8341 struct btrfs_block_rsv *block_rsv;
8342 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
8344 bool global_updated = false;
8346 block_rsv = get_block_rsv(trans, root);
8348 if (unlikely(block_rsv->size == 0))
8351 ret = block_rsv_use_bytes(block_rsv, blocksize);
8355 if (block_rsv->failfast)
8356 return ERR_PTR(ret);
8358 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
8359 global_updated = true;
8360 update_global_block_rsv(fs_info);
8364 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8365 static DEFINE_RATELIMIT_STATE(_rs,
8366 DEFAULT_RATELIMIT_INTERVAL * 10,
8367 /*DEFAULT_RATELIMIT_BURST*/ 1);
8368 if (__ratelimit(&_rs))
8370 "BTRFS: block rsv returned %d\n", ret);
8373 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
8374 BTRFS_RESERVE_NO_FLUSH);
8378 * If we couldn't reserve metadata bytes try and use some from
8379 * the global reserve if its space type is the same as the global
8382 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
8383 block_rsv->space_info == global_rsv->space_info) {
8384 ret = block_rsv_use_bytes(global_rsv, blocksize);
8388 return ERR_PTR(ret);
8391 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
8392 struct btrfs_block_rsv *block_rsv, u32 blocksize)
8394 block_rsv_add_bytes(block_rsv, blocksize, 0);
8395 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0, NULL);
8399 * finds a free extent and does all the dirty work required for allocation
8400 * returns the tree buffer or an ERR_PTR on error.
8402 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
8403 struct btrfs_root *root,
8404 u64 parent, u64 root_objectid,
8405 const struct btrfs_disk_key *key,
8406 int level, u64 hint,
8409 struct btrfs_fs_info *fs_info = root->fs_info;
8410 struct btrfs_key ins;
8411 struct btrfs_block_rsv *block_rsv;
8412 struct extent_buffer *buf;
8413 struct btrfs_delayed_extent_op *extent_op;
8416 u32 blocksize = fs_info->nodesize;
8417 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8419 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8420 if (btrfs_is_testing(fs_info)) {
8421 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
8424 root->alloc_bytenr += blocksize;
8429 block_rsv = use_block_rsv(trans, root, blocksize);
8430 if (IS_ERR(block_rsv))
8431 return ERR_CAST(block_rsv);
8433 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
8434 empty_size, hint, &ins, 0, 0);
8438 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
8441 goto out_free_reserved;
8444 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
8446 parent = ins.objectid;
8447 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
8451 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
8452 extent_op = btrfs_alloc_delayed_extent_op();
8458 memcpy(&extent_op->key, key, sizeof(extent_op->key));
8460 memset(&extent_op->key, 0, sizeof(extent_op->key));
8461 extent_op->flags_to_set = flags;
8462 extent_op->update_key = skinny_metadata ? false : true;
8463 extent_op->update_flags = true;
8464 extent_op->is_data = false;
8465 extent_op->level = level;
8467 btrfs_ref_tree_mod(root, ins.objectid, ins.offset, parent,
8468 root_objectid, level, 0,
8469 BTRFS_ADD_DELAYED_EXTENT);
8470 ret = btrfs_add_delayed_tree_ref(fs_info, trans, ins.objectid,
8472 root_objectid, level,
8473 BTRFS_ADD_DELAYED_EXTENT,
8474 extent_op, NULL, NULL);
8476 goto out_free_delayed;
8481 btrfs_free_delayed_extent_op(extent_op);
8483 free_extent_buffer(buf);
8485 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
8487 unuse_block_rsv(fs_info, block_rsv, blocksize);
8488 return ERR_PTR(ret);
8491 struct walk_control {
8492 u64 refs[BTRFS_MAX_LEVEL];
8493 u64 flags[BTRFS_MAX_LEVEL];
8494 struct btrfs_key update_progress;
8505 #define DROP_REFERENCE 1
8506 #define UPDATE_BACKREF 2
8508 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
8509 struct btrfs_root *root,
8510 struct walk_control *wc,
8511 struct btrfs_path *path)
8513 struct btrfs_fs_info *fs_info = root->fs_info;
8519 struct btrfs_key key;
8520 struct extent_buffer *eb;
8525 if (path->slots[wc->level] < wc->reada_slot) {
8526 wc->reada_count = wc->reada_count * 2 / 3;
8527 wc->reada_count = max(wc->reada_count, 2);
8529 wc->reada_count = wc->reada_count * 3 / 2;
8530 wc->reada_count = min_t(int, wc->reada_count,
8531 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
8534 eb = path->nodes[wc->level];
8535 nritems = btrfs_header_nritems(eb);
8537 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
8538 if (nread >= wc->reada_count)
8542 bytenr = btrfs_node_blockptr(eb, slot);
8543 generation = btrfs_node_ptr_generation(eb, slot);
8545 if (slot == path->slots[wc->level])
8548 if (wc->stage == UPDATE_BACKREF &&
8549 generation <= root->root_key.offset)
8552 /* We don't lock the tree block, it's OK to be racy here */
8553 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
8554 wc->level - 1, 1, &refs,
8556 /* We don't care about errors in readahead. */
8561 if (wc->stage == DROP_REFERENCE) {
8565 if (wc->level == 1 &&
8566 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8568 if (!wc->update_ref ||
8569 generation <= root->root_key.offset)
8571 btrfs_node_key_to_cpu(eb, &key, slot);
8572 ret = btrfs_comp_cpu_keys(&key,
8573 &wc->update_progress);
8577 if (wc->level == 1 &&
8578 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8582 readahead_tree_block(fs_info, bytenr);
8585 wc->reada_slot = slot;
8589 * helper to process tree block while walking down the tree.
8591 * when wc->stage == UPDATE_BACKREF, this function updates
8592 * back refs for pointers in the block.
8594 * NOTE: return value 1 means we should stop walking down.
8596 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8597 struct btrfs_root *root,
8598 struct btrfs_path *path,
8599 struct walk_control *wc, int lookup_info)
8601 struct btrfs_fs_info *fs_info = root->fs_info;
8602 int level = wc->level;
8603 struct extent_buffer *eb = path->nodes[level];
8604 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8607 if (wc->stage == UPDATE_BACKREF &&
8608 btrfs_header_owner(eb) != root->root_key.objectid)
8612 * when reference count of tree block is 1, it won't increase
8613 * again. once full backref flag is set, we never clear it.
8616 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8617 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8618 BUG_ON(!path->locks[level]);
8619 ret = btrfs_lookup_extent_info(trans, fs_info,
8620 eb->start, level, 1,
8623 BUG_ON(ret == -ENOMEM);
8626 BUG_ON(wc->refs[level] == 0);
8629 if (wc->stage == DROP_REFERENCE) {
8630 if (wc->refs[level] > 1)
8633 if (path->locks[level] && !wc->keep_locks) {
8634 btrfs_tree_unlock_rw(eb, path->locks[level]);
8635 path->locks[level] = 0;
8640 /* wc->stage == UPDATE_BACKREF */
8641 if (!(wc->flags[level] & flag)) {
8642 BUG_ON(!path->locks[level]);
8643 ret = btrfs_inc_ref(trans, root, eb, 1);
8644 BUG_ON(ret); /* -ENOMEM */
8645 ret = btrfs_dec_ref(trans, root, eb, 0);
8646 BUG_ON(ret); /* -ENOMEM */
8647 ret = btrfs_set_disk_extent_flags(trans, fs_info, eb->start,
8649 btrfs_header_level(eb), 0);
8650 BUG_ON(ret); /* -ENOMEM */
8651 wc->flags[level] |= flag;
8655 * the block is shared by multiple trees, so it's not good to
8656 * keep the tree lock
8658 if (path->locks[level] && level > 0) {
8659 btrfs_tree_unlock_rw(eb, path->locks[level]);
8660 path->locks[level] = 0;
8666 * helper to process tree block pointer.
8668 * when wc->stage == DROP_REFERENCE, this function checks
8669 * reference count of the block pointed to. if the block
8670 * is shared and we need update back refs for the subtree
8671 * rooted at the block, this function changes wc->stage to
8672 * UPDATE_BACKREF. if the block is shared and there is no
8673 * need to update back, this function drops the reference
8676 * NOTE: return value 1 means we should stop walking down.
8678 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8679 struct btrfs_root *root,
8680 struct btrfs_path *path,
8681 struct walk_control *wc, int *lookup_info)
8683 struct btrfs_fs_info *fs_info = root->fs_info;
8688 struct btrfs_key key;
8689 struct btrfs_key first_key;
8690 struct extent_buffer *next;
8691 int level = wc->level;
8694 bool need_account = false;
8696 generation = btrfs_node_ptr_generation(path->nodes[level],
8697 path->slots[level]);
8699 * if the lower level block was created before the snapshot
8700 * was created, we know there is no need to update back refs
8703 if (wc->stage == UPDATE_BACKREF &&
8704 generation <= root->root_key.offset) {
8709 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8710 btrfs_node_key_to_cpu(path->nodes[level], &first_key,
8711 path->slots[level]);
8712 blocksize = fs_info->nodesize;
8714 next = find_extent_buffer(fs_info, bytenr);
8716 next = btrfs_find_create_tree_block(fs_info, bytenr);
8718 return PTR_ERR(next);
8720 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8724 btrfs_tree_lock(next);
8725 btrfs_set_lock_blocking(next);
8727 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
8728 &wc->refs[level - 1],
8729 &wc->flags[level - 1]);
8733 if (unlikely(wc->refs[level - 1] == 0)) {
8734 btrfs_err(fs_info, "Missing references.");
8740 if (wc->stage == DROP_REFERENCE) {
8741 if (wc->refs[level - 1] > 1) {
8742 need_account = true;
8744 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8747 if (!wc->update_ref ||
8748 generation <= root->root_key.offset)
8751 btrfs_node_key_to_cpu(path->nodes[level], &key,
8752 path->slots[level]);
8753 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8757 wc->stage = UPDATE_BACKREF;
8758 wc->shared_level = level - 1;
8762 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8766 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8767 btrfs_tree_unlock(next);
8768 free_extent_buffer(next);
8774 if (reada && level == 1)
8775 reada_walk_down(trans, root, wc, path);
8776 next = read_tree_block(fs_info, bytenr, generation, level - 1,
8779 return PTR_ERR(next);
8780 } else if (!extent_buffer_uptodate(next)) {
8781 free_extent_buffer(next);
8784 btrfs_tree_lock(next);
8785 btrfs_set_lock_blocking(next);
8789 ASSERT(level == btrfs_header_level(next));
8790 if (level != btrfs_header_level(next)) {
8791 btrfs_err(root->fs_info, "mismatched level");
8795 path->nodes[level] = next;
8796 path->slots[level] = 0;
8797 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8803 wc->refs[level - 1] = 0;
8804 wc->flags[level - 1] = 0;
8805 if (wc->stage == DROP_REFERENCE) {
8806 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8807 parent = path->nodes[level]->start;
8809 ASSERT(root->root_key.objectid ==
8810 btrfs_header_owner(path->nodes[level]));
8811 if (root->root_key.objectid !=
8812 btrfs_header_owner(path->nodes[level])) {
8813 btrfs_err(root->fs_info,
8814 "mismatched block owner");
8822 ret = btrfs_qgroup_trace_subtree(trans, root, next,
8823 generation, level - 1);
8825 btrfs_err_rl(fs_info,
8826 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
8830 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
8831 parent, root->root_key.objectid,
8841 btrfs_tree_unlock(next);
8842 free_extent_buffer(next);
8848 * helper to process tree block while walking up the tree.
8850 * when wc->stage == DROP_REFERENCE, this function drops
8851 * reference count on the block.
8853 * when wc->stage == UPDATE_BACKREF, this function changes
8854 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8855 * to UPDATE_BACKREF previously while processing the block.
8857 * NOTE: return value 1 means we should stop walking up.
8859 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8860 struct btrfs_root *root,
8861 struct btrfs_path *path,
8862 struct walk_control *wc)
8864 struct btrfs_fs_info *fs_info = root->fs_info;
8866 int level = wc->level;
8867 struct extent_buffer *eb = path->nodes[level];
8870 if (wc->stage == UPDATE_BACKREF) {
8871 BUG_ON(wc->shared_level < level);
8872 if (level < wc->shared_level)
8875 ret = find_next_key(path, level + 1, &wc->update_progress);
8879 wc->stage = DROP_REFERENCE;
8880 wc->shared_level = -1;
8881 path->slots[level] = 0;
8884 * check reference count again if the block isn't locked.
8885 * we should start walking down the tree again if reference
8888 if (!path->locks[level]) {
8890 btrfs_tree_lock(eb);
8891 btrfs_set_lock_blocking(eb);
8892 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8894 ret = btrfs_lookup_extent_info(trans, fs_info,
8895 eb->start, level, 1,
8899 btrfs_tree_unlock_rw(eb, path->locks[level]);
8900 path->locks[level] = 0;
8903 BUG_ON(wc->refs[level] == 0);
8904 if (wc->refs[level] == 1) {
8905 btrfs_tree_unlock_rw(eb, path->locks[level]);
8906 path->locks[level] = 0;
8912 /* wc->stage == DROP_REFERENCE */
8913 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8915 if (wc->refs[level] == 1) {
8917 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8918 ret = btrfs_dec_ref(trans, root, eb, 1);
8920 ret = btrfs_dec_ref(trans, root, eb, 0);
8921 BUG_ON(ret); /* -ENOMEM */
8922 ret = btrfs_qgroup_trace_leaf_items(trans, fs_info, eb);
8924 btrfs_err_rl(fs_info,
8925 "error %d accounting leaf items. Quota is out of sync, rescan required.",
8929 /* make block locked assertion in clean_tree_block happy */
8930 if (!path->locks[level] &&
8931 btrfs_header_generation(eb) == trans->transid) {
8932 btrfs_tree_lock(eb);
8933 btrfs_set_lock_blocking(eb);
8934 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8936 clean_tree_block(fs_info, eb);
8939 if (eb == root->node) {
8940 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8943 BUG_ON(root->root_key.objectid !=
8944 btrfs_header_owner(eb));
8946 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8947 parent = path->nodes[level + 1]->start;
8949 BUG_ON(root->root_key.objectid !=
8950 btrfs_header_owner(path->nodes[level + 1]));
8953 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8955 wc->refs[level] = 0;
8956 wc->flags[level] = 0;
8960 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8961 struct btrfs_root *root,
8962 struct btrfs_path *path,
8963 struct walk_control *wc)
8965 int level = wc->level;
8966 int lookup_info = 1;
8969 while (level >= 0) {
8970 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8977 if (path->slots[level] >=
8978 btrfs_header_nritems(path->nodes[level]))
8981 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8983 path->slots[level]++;
8992 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
8993 struct btrfs_root *root,
8994 struct btrfs_path *path,
8995 struct walk_control *wc, int max_level)
8997 int level = wc->level;
9000 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
9001 while (level < max_level && path->nodes[level]) {
9003 if (path->slots[level] + 1 <
9004 btrfs_header_nritems(path->nodes[level])) {
9005 path->slots[level]++;
9008 ret = walk_up_proc(trans, root, path, wc);
9012 if (path->locks[level]) {
9013 btrfs_tree_unlock_rw(path->nodes[level],
9014 path->locks[level]);
9015 path->locks[level] = 0;
9017 free_extent_buffer(path->nodes[level]);
9018 path->nodes[level] = NULL;
9026 * drop a subvolume tree.
9028 * this function traverses the tree freeing any blocks that only
9029 * referenced by the tree.
9031 * when a shared tree block is found. this function decreases its
9032 * reference count by one. if update_ref is true, this function
9033 * also make sure backrefs for the shared block and all lower level
9034 * blocks are properly updated.
9036 * If called with for_reloc == 0, may exit early with -EAGAIN
9038 int btrfs_drop_snapshot(struct btrfs_root *root,
9039 struct btrfs_block_rsv *block_rsv, int update_ref,
9042 struct btrfs_fs_info *fs_info = root->fs_info;
9043 struct btrfs_path *path;
9044 struct btrfs_trans_handle *trans;
9045 struct btrfs_root *tree_root = fs_info->tree_root;
9046 struct btrfs_root_item *root_item = &root->root_item;
9047 struct walk_control *wc;
9048 struct btrfs_key key;
9052 bool root_dropped = false;
9054 btrfs_debug(fs_info, "Drop subvolume %llu", root->objectid);
9056 path = btrfs_alloc_path();
9062 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9064 btrfs_free_path(path);
9069 trans = btrfs_start_transaction(tree_root, 0);
9070 if (IS_ERR(trans)) {
9071 err = PTR_ERR(trans);
9076 trans->block_rsv = block_rsv;
9078 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
9079 level = btrfs_header_level(root->node);
9080 path->nodes[level] = btrfs_lock_root_node(root);
9081 btrfs_set_lock_blocking(path->nodes[level]);
9082 path->slots[level] = 0;
9083 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9084 memset(&wc->update_progress, 0,
9085 sizeof(wc->update_progress));
9087 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
9088 memcpy(&wc->update_progress, &key,
9089 sizeof(wc->update_progress));
9091 level = root_item->drop_level;
9093 path->lowest_level = level;
9094 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
9095 path->lowest_level = 0;
9103 * unlock our path, this is safe because only this
9104 * function is allowed to delete this snapshot
9106 btrfs_unlock_up_safe(path, 0);
9108 level = btrfs_header_level(root->node);
9110 btrfs_tree_lock(path->nodes[level]);
9111 btrfs_set_lock_blocking(path->nodes[level]);
9112 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9114 ret = btrfs_lookup_extent_info(trans, fs_info,
9115 path->nodes[level]->start,
9116 level, 1, &wc->refs[level],
9122 BUG_ON(wc->refs[level] == 0);
9124 if (level == root_item->drop_level)
9127 btrfs_tree_unlock(path->nodes[level]);
9128 path->locks[level] = 0;
9129 WARN_ON(wc->refs[level] != 1);
9135 wc->shared_level = -1;
9136 wc->stage = DROP_REFERENCE;
9137 wc->update_ref = update_ref;
9139 wc->for_reloc = for_reloc;
9140 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9144 ret = walk_down_tree(trans, root, path, wc);
9150 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
9157 BUG_ON(wc->stage != DROP_REFERENCE);
9161 if (wc->stage == DROP_REFERENCE) {
9163 btrfs_node_key(path->nodes[level],
9164 &root_item->drop_progress,
9165 path->slots[level]);
9166 root_item->drop_level = level;
9169 BUG_ON(wc->level == 0);
9170 if (btrfs_should_end_transaction(trans) ||
9171 (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
9172 ret = btrfs_update_root(trans, tree_root,
9176 btrfs_abort_transaction(trans, ret);
9181 btrfs_end_transaction_throttle(trans);
9182 if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
9183 btrfs_debug(fs_info,
9184 "drop snapshot early exit");
9189 trans = btrfs_start_transaction(tree_root, 0);
9190 if (IS_ERR(trans)) {
9191 err = PTR_ERR(trans);
9195 trans->block_rsv = block_rsv;
9198 btrfs_release_path(path);
9202 ret = btrfs_del_root(trans, fs_info, &root->root_key);
9204 btrfs_abort_transaction(trans, ret);
9209 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
9210 ret = btrfs_find_root(tree_root, &root->root_key, path,
9213 btrfs_abort_transaction(trans, ret);
9216 } else if (ret > 0) {
9217 /* if we fail to delete the orphan item this time
9218 * around, it'll get picked up the next time.
9220 * The most common failure here is just -ENOENT.
9222 btrfs_del_orphan_item(trans, tree_root,
9223 root->root_key.objectid);
9227 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
9228 btrfs_add_dropped_root(trans, root);
9230 free_extent_buffer(root->node);
9231 free_extent_buffer(root->commit_root);
9232 btrfs_put_fs_root(root);
9234 root_dropped = true;
9236 btrfs_end_transaction_throttle(trans);
9239 btrfs_free_path(path);
9242 * So if we need to stop dropping the snapshot for whatever reason we
9243 * need to make sure to add it back to the dead root list so that we
9244 * keep trying to do the work later. This also cleans up roots if we
9245 * don't have it in the radix (like when we recover after a power fail
9246 * or unmount) so we don't leak memory.
9248 if (!for_reloc && !root_dropped)
9249 btrfs_add_dead_root(root);
9250 if (err && err != -EAGAIN)
9251 btrfs_handle_fs_error(fs_info, err, NULL);
9256 * drop subtree rooted at tree block 'node'.
9258 * NOTE: this function will unlock and release tree block 'node'
9259 * only used by relocation code
9261 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
9262 struct btrfs_root *root,
9263 struct extent_buffer *node,
9264 struct extent_buffer *parent)
9266 struct btrfs_fs_info *fs_info = root->fs_info;
9267 struct btrfs_path *path;
9268 struct walk_control *wc;
9274 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
9276 path = btrfs_alloc_path();
9280 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9282 btrfs_free_path(path);
9286 btrfs_assert_tree_locked(parent);
9287 parent_level = btrfs_header_level(parent);
9288 extent_buffer_get(parent);
9289 path->nodes[parent_level] = parent;
9290 path->slots[parent_level] = btrfs_header_nritems(parent);
9292 btrfs_assert_tree_locked(node);
9293 level = btrfs_header_level(node);
9294 path->nodes[level] = node;
9295 path->slots[level] = 0;
9296 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9298 wc->refs[parent_level] = 1;
9299 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
9301 wc->shared_level = -1;
9302 wc->stage = DROP_REFERENCE;
9306 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9309 wret = walk_down_tree(trans, root, path, wc);
9315 wret = walk_up_tree(trans, root, path, wc, parent_level);
9323 btrfs_free_path(path);
9327 static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags)
9333 * if restripe for this chunk_type is on pick target profile and
9334 * return, otherwise do the usual balance
9336 stripped = get_restripe_target(fs_info, flags);
9338 return extended_to_chunk(stripped);
9340 num_devices = fs_info->fs_devices->rw_devices;
9342 stripped = BTRFS_BLOCK_GROUP_RAID0 |
9343 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
9344 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
9346 if (num_devices == 1) {
9347 stripped |= BTRFS_BLOCK_GROUP_DUP;
9348 stripped = flags & ~stripped;
9350 /* turn raid0 into single device chunks */
9351 if (flags & BTRFS_BLOCK_GROUP_RAID0)
9354 /* turn mirroring into duplication */
9355 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
9356 BTRFS_BLOCK_GROUP_RAID10))
9357 return stripped | BTRFS_BLOCK_GROUP_DUP;
9359 /* they already had raid on here, just return */
9360 if (flags & stripped)
9363 stripped |= BTRFS_BLOCK_GROUP_DUP;
9364 stripped = flags & ~stripped;
9366 /* switch duplicated blocks with raid1 */
9367 if (flags & BTRFS_BLOCK_GROUP_DUP)
9368 return stripped | BTRFS_BLOCK_GROUP_RAID1;
9370 /* this is drive concat, leave it alone */
9376 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
9378 struct btrfs_space_info *sinfo = cache->space_info;
9380 u64 min_allocable_bytes;
9384 * We need some metadata space and system metadata space for
9385 * allocating chunks in some corner cases until we force to set
9386 * it to be readonly.
9389 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
9391 min_allocable_bytes = SZ_1M;
9393 min_allocable_bytes = 0;
9395 spin_lock(&sinfo->lock);
9396 spin_lock(&cache->lock);
9404 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
9405 cache->bytes_super - btrfs_block_group_used(&cache->item);
9407 if (btrfs_space_info_used(sinfo, true) + num_bytes +
9408 min_allocable_bytes <= sinfo->total_bytes) {
9409 sinfo->bytes_readonly += num_bytes;
9411 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
9415 spin_unlock(&cache->lock);
9416 spin_unlock(&sinfo->lock);
9420 int btrfs_inc_block_group_ro(struct btrfs_fs_info *fs_info,
9421 struct btrfs_block_group_cache *cache)
9424 struct btrfs_trans_handle *trans;
9429 trans = btrfs_join_transaction(fs_info->extent_root);
9431 return PTR_ERR(trans);
9434 * we're not allowed to set block groups readonly after the dirty
9435 * block groups cache has started writing. If it already started,
9436 * back off and let this transaction commit
9438 mutex_lock(&fs_info->ro_block_group_mutex);
9439 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9440 u64 transid = trans->transid;
9442 mutex_unlock(&fs_info->ro_block_group_mutex);
9443 btrfs_end_transaction(trans);
9445 ret = btrfs_wait_for_commit(fs_info, transid);
9452 * if we are changing raid levels, try to allocate a corresponding
9453 * block group with the new raid level.
9455 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9456 if (alloc_flags != cache->flags) {
9457 ret = do_chunk_alloc(trans, fs_info, alloc_flags,
9460 * ENOSPC is allowed here, we may have enough space
9461 * already allocated at the new raid level to
9470 ret = inc_block_group_ro(cache, 0);
9473 alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags);
9474 ret = do_chunk_alloc(trans, fs_info, alloc_flags,
9478 ret = inc_block_group_ro(cache, 0);
9480 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9481 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9482 mutex_lock(&fs_info->chunk_mutex);
9483 check_system_chunk(trans, fs_info, alloc_flags);
9484 mutex_unlock(&fs_info->chunk_mutex);
9486 mutex_unlock(&fs_info->ro_block_group_mutex);
9488 btrfs_end_transaction(trans);
9492 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
9493 struct btrfs_fs_info *fs_info, u64 type)
9495 u64 alloc_flags = get_alloc_profile(fs_info, type);
9497 return do_chunk_alloc(trans, fs_info, alloc_flags, CHUNK_ALLOC_FORCE);
9501 * helper to account the unused space of all the readonly block group in the
9502 * space_info. takes mirrors into account.
9504 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9506 struct btrfs_block_group_cache *block_group;
9510 /* It's df, we don't care if it's racy */
9511 if (list_empty(&sinfo->ro_bgs))
9514 spin_lock(&sinfo->lock);
9515 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9516 spin_lock(&block_group->lock);
9518 if (!block_group->ro) {
9519 spin_unlock(&block_group->lock);
9523 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
9524 BTRFS_BLOCK_GROUP_RAID10 |
9525 BTRFS_BLOCK_GROUP_DUP))
9530 free_bytes += (block_group->key.offset -
9531 btrfs_block_group_used(&block_group->item)) *
9534 spin_unlock(&block_group->lock);
9536 spin_unlock(&sinfo->lock);
9541 void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache)
9543 struct btrfs_space_info *sinfo = cache->space_info;
9548 spin_lock(&sinfo->lock);
9549 spin_lock(&cache->lock);
9551 num_bytes = cache->key.offset - cache->reserved -
9552 cache->pinned - cache->bytes_super -
9553 btrfs_block_group_used(&cache->item);
9554 sinfo->bytes_readonly -= num_bytes;
9555 list_del_init(&cache->ro_list);
9557 spin_unlock(&cache->lock);
9558 spin_unlock(&sinfo->lock);
9562 * checks to see if its even possible to relocate this block group.
9564 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9565 * ok to go ahead and try.
9567 int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr)
9569 struct btrfs_root *root = fs_info->extent_root;
9570 struct btrfs_block_group_cache *block_group;
9571 struct btrfs_space_info *space_info;
9572 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
9573 struct btrfs_device *device;
9574 struct btrfs_trans_handle *trans;
9584 debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG);
9586 block_group = btrfs_lookup_block_group(fs_info, bytenr);
9588 /* odd, couldn't find the block group, leave it alone */
9592 "can't find block group for bytenr %llu",
9597 min_free = btrfs_block_group_used(&block_group->item);
9599 /* no bytes used, we're good */
9603 space_info = block_group->space_info;
9604 spin_lock(&space_info->lock);
9606 full = space_info->full;
9609 * if this is the last block group we have in this space, we can't
9610 * relocate it unless we're able to allocate a new chunk below.
9612 * Otherwise, we need to make sure we have room in the space to handle
9613 * all of the extents from this block group. If we can, we're good
9615 if ((space_info->total_bytes != block_group->key.offset) &&
9616 (btrfs_space_info_used(space_info, false) + min_free <
9617 space_info->total_bytes)) {
9618 spin_unlock(&space_info->lock);
9621 spin_unlock(&space_info->lock);
9624 * ok we don't have enough space, but maybe we have free space on our
9625 * devices to allocate new chunks for relocation, so loop through our
9626 * alloc devices and guess if we have enough space. if this block
9627 * group is going to be restriped, run checks against the target
9628 * profile instead of the current one.
9640 target = get_restripe_target(fs_info, block_group->flags);
9642 index = btrfs_bg_flags_to_raid_index(extended_to_chunk(target));
9645 * this is just a balance, so if we were marked as full
9646 * we know there is no space for a new chunk
9651 "no space to alloc new chunk for block group %llu",
9652 block_group->key.objectid);
9656 index = btrfs_bg_flags_to_raid_index(block_group->flags);
9659 if (index == BTRFS_RAID_RAID10) {
9663 } else if (index == BTRFS_RAID_RAID1) {
9665 } else if (index == BTRFS_RAID_DUP) {
9668 } else if (index == BTRFS_RAID_RAID0) {
9669 dev_min = fs_devices->rw_devices;
9670 min_free = div64_u64(min_free, dev_min);
9673 /* We need to do this so that we can look at pending chunks */
9674 trans = btrfs_join_transaction(root);
9675 if (IS_ERR(trans)) {
9676 ret = PTR_ERR(trans);
9680 mutex_lock(&fs_info->chunk_mutex);
9681 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9685 * check to make sure we can actually find a chunk with enough
9686 * space to fit our block group in.
9688 if (device->total_bytes > device->bytes_used + min_free &&
9689 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
9690 ret = find_free_dev_extent(trans, device, min_free,
9695 if (dev_nr >= dev_min)
9701 if (debug && ret == -1)
9703 "no space to allocate a new chunk for block group %llu",
9704 block_group->key.objectid);
9705 mutex_unlock(&fs_info->chunk_mutex);
9706 btrfs_end_transaction(trans);
9708 btrfs_put_block_group(block_group);
9712 static int find_first_block_group(struct btrfs_fs_info *fs_info,
9713 struct btrfs_path *path,
9714 struct btrfs_key *key)
9716 struct btrfs_root *root = fs_info->extent_root;
9718 struct btrfs_key found_key;
9719 struct extent_buffer *leaf;
9722 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9727 slot = path->slots[0];
9728 leaf = path->nodes[0];
9729 if (slot >= btrfs_header_nritems(leaf)) {
9730 ret = btrfs_next_leaf(root, path);
9737 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9739 if (found_key.objectid >= key->objectid &&
9740 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9741 struct extent_map_tree *em_tree;
9742 struct extent_map *em;
9744 em_tree = &root->fs_info->mapping_tree.map_tree;
9745 read_lock(&em_tree->lock);
9746 em = lookup_extent_mapping(em_tree, found_key.objectid,
9748 read_unlock(&em_tree->lock);
9751 "logical %llu len %llu found bg but no related chunk",
9752 found_key.objectid, found_key.offset);
9757 free_extent_map(em);
9766 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9768 struct btrfs_block_group_cache *block_group;
9772 struct inode *inode;
9774 block_group = btrfs_lookup_first_block_group(info, last);
9775 while (block_group) {
9776 spin_lock(&block_group->lock);
9777 if (block_group->iref)
9779 spin_unlock(&block_group->lock);
9780 block_group = next_block_group(info, block_group);
9789 inode = block_group->inode;
9790 block_group->iref = 0;
9791 block_group->inode = NULL;
9792 spin_unlock(&block_group->lock);
9793 ASSERT(block_group->io_ctl.inode == NULL);
9795 last = block_group->key.objectid + block_group->key.offset;
9796 btrfs_put_block_group(block_group);
9801 * Must be called only after stopping all workers, since we could have block
9802 * group caching kthreads running, and therefore they could race with us if we
9803 * freed the block groups before stopping them.
9805 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9807 struct btrfs_block_group_cache *block_group;
9808 struct btrfs_space_info *space_info;
9809 struct btrfs_caching_control *caching_ctl;
9812 down_write(&info->commit_root_sem);
9813 while (!list_empty(&info->caching_block_groups)) {
9814 caching_ctl = list_entry(info->caching_block_groups.next,
9815 struct btrfs_caching_control, list);
9816 list_del(&caching_ctl->list);
9817 put_caching_control(caching_ctl);
9819 up_write(&info->commit_root_sem);
9821 spin_lock(&info->unused_bgs_lock);
9822 while (!list_empty(&info->unused_bgs)) {
9823 block_group = list_first_entry(&info->unused_bgs,
9824 struct btrfs_block_group_cache,
9826 list_del_init(&block_group->bg_list);
9827 btrfs_put_block_group(block_group);
9829 spin_unlock(&info->unused_bgs_lock);
9831 spin_lock(&info->block_group_cache_lock);
9832 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9833 block_group = rb_entry(n, struct btrfs_block_group_cache,
9835 rb_erase(&block_group->cache_node,
9836 &info->block_group_cache_tree);
9837 RB_CLEAR_NODE(&block_group->cache_node);
9838 spin_unlock(&info->block_group_cache_lock);
9840 down_write(&block_group->space_info->groups_sem);
9841 list_del(&block_group->list);
9842 up_write(&block_group->space_info->groups_sem);
9845 * We haven't cached this block group, which means we could
9846 * possibly have excluded extents on this block group.
9848 if (block_group->cached == BTRFS_CACHE_NO ||
9849 block_group->cached == BTRFS_CACHE_ERROR)
9850 free_excluded_extents(info, block_group);
9852 btrfs_remove_free_space_cache(block_group);
9853 ASSERT(block_group->cached != BTRFS_CACHE_STARTED);
9854 ASSERT(list_empty(&block_group->dirty_list));
9855 ASSERT(list_empty(&block_group->io_list));
9856 ASSERT(list_empty(&block_group->bg_list));
9857 ASSERT(atomic_read(&block_group->count) == 1);
9858 btrfs_put_block_group(block_group);
9860 spin_lock(&info->block_group_cache_lock);
9862 spin_unlock(&info->block_group_cache_lock);
9864 /* now that all the block groups are freed, go through and
9865 * free all the space_info structs. This is only called during
9866 * the final stages of unmount, and so we know nobody is
9867 * using them. We call synchronize_rcu() once before we start,
9868 * just to be on the safe side.
9872 release_global_block_rsv(info);
9874 while (!list_empty(&info->space_info)) {
9877 space_info = list_entry(info->space_info.next,
9878 struct btrfs_space_info,
9882 * Do not hide this behind enospc_debug, this is actually
9883 * important and indicates a real bug if this happens.
9885 if (WARN_ON(space_info->bytes_pinned > 0 ||
9886 space_info->bytes_reserved > 0 ||
9887 space_info->bytes_may_use > 0))
9888 dump_space_info(info, space_info, 0, 0);
9889 list_del(&space_info->list);
9890 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9891 struct kobject *kobj;
9892 kobj = space_info->block_group_kobjs[i];
9893 space_info->block_group_kobjs[i] = NULL;
9899 kobject_del(&space_info->kobj);
9900 kobject_put(&space_info->kobj);
9905 /* link_block_group will queue up kobjects to add when we're reclaim-safe */
9906 void btrfs_add_raid_kobjects(struct btrfs_fs_info *fs_info)
9908 struct btrfs_space_info *space_info;
9909 struct raid_kobject *rkobj;
9914 spin_lock(&fs_info->pending_raid_kobjs_lock);
9915 list_splice_init(&fs_info->pending_raid_kobjs, &list);
9916 spin_unlock(&fs_info->pending_raid_kobjs_lock);
9918 list_for_each_entry(rkobj, &list, list) {
9919 space_info = __find_space_info(fs_info, rkobj->flags);
9920 index = btrfs_bg_flags_to_raid_index(rkobj->flags);
9922 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9923 "%s", get_raid_name(index));
9925 kobject_put(&rkobj->kobj);
9931 "failed to add kobject for block cache, ignoring");
9934 static void link_block_group(struct btrfs_block_group_cache *cache)
9936 struct btrfs_space_info *space_info = cache->space_info;
9937 struct btrfs_fs_info *fs_info = cache->fs_info;
9938 int index = btrfs_bg_flags_to_raid_index(cache->flags);
9941 down_write(&space_info->groups_sem);
9942 if (list_empty(&space_info->block_groups[index]))
9944 list_add_tail(&cache->list, &space_info->block_groups[index]);
9945 up_write(&space_info->groups_sem);
9948 struct raid_kobject *rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9950 btrfs_warn(cache->fs_info,
9951 "couldn't alloc memory for raid level kobject");
9954 rkobj->flags = cache->flags;
9955 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9957 spin_lock(&fs_info->pending_raid_kobjs_lock);
9958 list_add_tail(&rkobj->list, &fs_info->pending_raid_kobjs);
9959 spin_unlock(&fs_info->pending_raid_kobjs_lock);
9960 space_info->block_group_kobjs[index] = &rkobj->kobj;
9964 static struct btrfs_block_group_cache *
9965 btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info,
9966 u64 start, u64 size)
9968 struct btrfs_block_group_cache *cache;
9970 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9974 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
9976 if (!cache->free_space_ctl) {
9981 cache->key.objectid = start;
9982 cache->key.offset = size;
9983 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9985 cache->fs_info = fs_info;
9986 cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start);
9987 set_free_space_tree_thresholds(cache);
9989 atomic_set(&cache->count, 1);
9990 spin_lock_init(&cache->lock);
9991 init_rwsem(&cache->data_rwsem);
9992 INIT_LIST_HEAD(&cache->list);
9993 INIT_LIST_HEAD(&cache->cluster_list);
9994 INIT_LIST_HEAD(&cache->bg_list);
9995 INIT_LIST_HEAD(&cache->ro_list);
9996 INIT_LIST_HEAD(&cache->dirty_list);
9997 INIT_LIST_HEAD(&cache->io_list);
9998 btrfs_init_free_space_ctl(cache);
9999 atomic_set(&cache->trimming, 0);
10000 mutex_init(&cache->free_space_lock);
10001 btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root);
10006 int btrfs_read_block_groups(struct btrfs_fs_info *info)
10008 struct btrfs_path *path;
10010 struct btrfs_block_group_cache *cache;
10011 struct btrfs_space_info *space_info;
10012 struct btrfs_key key;
10013 struct btrfs_key found_key;
10014 struct extent_buffer *leaf;
10015 int need_clear = 0;
10020 feature = btrfs_super_incompat_flags(info->super_copy);
10021 mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS);
10025 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10026 path = btrfs_alloc_path();
10029 path->reada = READA_FORWARD;
10031 cache_gen = btrfs_super_cache_generation(info->super_copy);
10032 if (btrfs_test_opt(info, SPACE_CACHE) &&
10033 btrfs_super_generation(info->super_copy) != cache_gen)
10035 if (btrfs_test_opt(info, CLEAR_CACHE))
10039 ret = find_first_block_group(info, path, &key);
10045 leaf = path->nodes[0];
10046 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
10048 cache = btrfs_create_block_group_cache(info, found_key.objectid,
10057 * When we mount with old space cache, we need to
10058 * set BTRFS_DC_CLEAR and set dirty flag.
10060 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
10061 * truncate the old free space cache inode and
10063 * b) Setting 'dirty flag' makes sure that we flush
10064 * the new space cache info onto disk.
10066 if (btrfs_test_opt(info, SPACE_CACHE))
10067 cache->disk_cache_state = BTRFS_DC_CLEAR;
10070 read_extent_buffer(leaf, &cache->item,
10071 btrfs_item_ptr_offset(leaf, path->slots[0]),
10072 sizeof(cache->item));
10073 cache->flags = btrfs_block_group_flags(&cache->item);
10075 ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) &&
10076 (cache->flags & BTRFS_BLOCK_GROUP_DATA))) {
10078 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
10079 cache->key.objectid);
10084 key.objectid = found_key.objectid + found_key.offset;
10085 btrfs_release_path(path);
10088 * We need to exclude the super stripes now so that the space
10089 * info has super bytes accounted for, otherwise we'll think
10090 * we have more space than we actually do.
10092 ret = exclude_super_stripes(info, cache);
10095 * We may have excluded something, so call this just in
10098 free_excluded_extents(info, cache);
10099 btrfs_put_block_group(cache);
10104 * check for two cases, either we are full, and therefore
10105 * don't need to bother with the caching work since we won't
10106 * find any space, or we are empty, and we can just add all
10107 * the space in and be done with it. This saves us _alot_ of
10108 * time, particularly in the full case.
10110 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
10111 cache->last_byte_to_unpin = (u64)-1;
10112 cache->cached = BTRFS_CACHE_FINISHED;
10113 free_excluded_extents(info, cache);
10114 } else if (btrfs_block_group_used(&cache->item) == 0) {
10115 cache->last_byte_to_unpin = (u64)-1;
10116 cache->cached = BTRFS_CACHE_FINISHED;
10117 add_new_free_space(cache, found_key.objectid,
10118 found_key.objectid +
10120 free_excluded_extents(info, cache);
10123 ret = btrfs_add_block_group_cache(info, cache);
10125 btrfs_remove_free_space_cache(cache);
10126 btrfs_put_block_group(cache);
10130 trace_btrfs_add_block_group(info, cache, 0);
10131 update_space_info(info, cache->flags, found_key.offset,
10132 btrfs_block_group_used(&cache->item),
10133 cache->bytes_super, &space_info);
10135 cache->space_info = space_info;
10137 link_block_group(cache);
10139 set_avail_alloc_bits(info, cache->flags);
10140 if (btrfs_chunk_readonly(info, cache->key.objectid)) {
10141 inc_block_group_ro(cache, 1);
10142 } else if (btrfs_block_group_used(&cache->item) == 0) {
10143 spin_lock(&info->unused_bgs_lock);
10144 /* Should always be true but just in case. */
10145 if (list_empty(&cache->bg_list)) {
10146 btrfs_get_block_group(cache);
10147 trace_btrfs_add_unused_block_group(cache);
10148 list_add_tail(&cache->bg_list,
10149 &info->unused_bgs);
10151 spin_unlock(&info->unused_bgs_lock);
10155 list_for_each_entry_rcu(space_info, &info->space_info, list) {
10156 if (!(get_alloc_profile(info, space_info->flags) &
10157 (BTRFS_BLOCK_GROUP_RAID10 |
10158 BTRFS_BLOCK_GROUP_RAID1 |
10159 BTRFS_BLOCK_GROUP_RAID5 |
10160 BTRFS_BLOCK_GROUP_RAID6 |
10161 BTRFS_BLOCK_GROUP_DUP)))
10164 * avoid allocating from un-mirrored block group if there are
10165 * mirrored block groups.
10167 list_for_each_entry(cache,
10168 &space_info->block_groups[BTRFS_RAID_RAID0],
10170 inc_block_group_ro(cache, 1);
10171 list_for_each_entry(cache,
10172 &space_info->block_groups[BTRFS_RAID_SINGLE],
10174 inc_block_group_ro(cache, 1);
10177 btrfs_add_raid_kobjects(info);
10178 init_global_block_rsv(info);
10181 btrfs_free_path(path);
10185 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans)
10187 struct btrfs_fs_info *fs_info = trans->fs_info;
10188 struct btrfs_block_group_cache *block_group, *tmp;
10189 struct btrfs_root *extent_root = fs_info->extent_root;
10190 struct btrfs_block_group_item item;
10191 struct btrfs_key key;
10193 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
10195 trans->can_flush_pending_bgs = false;
10196 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
10200 spin_lock(&block_group->lock);
10201 memcpy(&item, &block_group->item, sizeof(item));
10202 memcpy(&key, &block_group->key, sizeof(key));
10203 spin_unlock(&block_group->lock);
10205 ret = btrfs_insert_item(trans, extent_root, &key, &item,
10208 btrfs_abort_transaction(trans, ret);
10209 ret = btrfs_finish_chunk_alloc(trans, fs_info, key.objectid,
10212 btrfs_abort_transaction(trans, ret);
10213 add_block_group_free_space(trans, block_group);
10214 /* already aborted the transaction if it failed. */
10216 list_del_init(&block_group->bg_list);
10218 trans->can_flush_pending_bgs = can_flush_pending_bgs;
10221 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
10222 struct btrfs_fs_info *fs_info, u64 bytes_used,
10223 u64 type, u64 chunk_offset, u64 size)
10225 struct btrfs_block_group_cache *cache;
10228 btrfs_set_log_full_commit(fs_info, trans);
10230 cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size);
10234 btrfs_set_block_group_used(&cache->item, bytes_used);
10235 btrfs_set_block_group_chunk_objectid(&cache->item,
10236 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
10237 btrfs_set_block_group_flags(&cache->item, type);
10239 cache->flags = type;
10240 cache->last_byte_to_unpin = (u64)-1;
10241 cache->cached = BTRFS_CACHE_FINISHED;
10242 cache->needs_free_space = 1;
10243 ret = exclude_super_stripes(fs_info, cache);
10246 * We may have excluded something, so call this just in
10249 free_excluded_extents(fs_info, cache);
10250 btrfs_put_block_group(cache);
10254 add_new_free_space(cache, chunk_offset, chunk_offset + size);
10256 free_excluded_extents(fs_info, cache);
10258 #ifdef CONFIG_BTRFS_DEBUG
10259 if (btrfs_should_fragment_free_space(cache)) {
10260 u64 new_bytes_used = size - bytes_used;
10262 bytes_used += new_bytes_used >> 1;
10263 fragment_free_space(cache);
10267 * Ensure the corresponding space_info object is created and
10268 * assigned to our block group. We want our bg to be added to the rbtree
10269 * with its ->space_info set.
10271 cache->space_info = __find_space_info(fs_info, cache->flags);
10272 ASSERT(cache->space_info);
10274 ret = btrfs_add_block_group_cache(fs_info, cache);
10276 btrfs_remove_free_space_cache(cache);
10277 btrfs_put_block_group(cache);
10282 * Now that our block group has its ->space_info set and is inserted in
10283 * the rbtree, update the space info's counters.
10285 trace_btrfs_add_block_group(fs_info, cache, 1);
10286 update_space_info(fs_info, cache->flags, size, bytes_used,
10287 cache->bytes_super, &cache->space_info);
10288 update_global_block_rsv(fs_info);
10290 link_block_group(cache);
10292 list_add_tail(&cache->bg_list, &trans->new_bgs);
10294 set_avail_alloc_bits(fs_info, type);
10298 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
10300 u64 extra_flags = chunk_to_extended(flags) &
10301 BTRFS_EXTENDED_PROFILE_MASK;
10303 write_seqlock(&fs_info->profiles_lock);
10304 if (flags & BTRFS_BLOCK_GROUP_DATA)
10305 fs_info->avail_data_alloc_bits &= ~extra_flags;
10306 if (flags & BTRFS_BLOCK_GROUP_METADATA)
10307 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
10308 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
10309 fs_info->avail_system_alloc_bits &= ~extra_flags;
10310 write_sequnlock(&fs_info->profiles_lock);
10313 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
10314 struct btrfs_fs_info *fs_info, u64 group_start,
10315 struct extent_map *em)
10317 struct btrfs_root *root = fs_info->extent_root;
10318 struct btrfs_path *path;
10319 struct btrfs_block_group_cache *block_group;
10320 struct btrfs_free_cluster *cluster;
10321 struct btrfs_root *tree_root = fs_info->tree_root;
10322 struct btrfs_key key;
10323 struct inode *inode;
10324 struct kobject *kobj = NULL;
10328 struct btrfs_caching_control *caching_ctl = NULL;
10331 block_group = btrfs_lookup_block_group(fs_info, group_start);
10332 BUG_ON(!block_group);
10333 BUG_ON(!block_group->ro);
10335 trace_btrfs_remove_block_group(block_group);
10337 * Free the reserved super bytes from this block group before
10340 free_excluded_extents(fs_info, block_group);
10341 btrfs_free_ref_tree_range(fs_info, block_group->key.objectid,
10342 block_group->key.offset);
10344 memcpy(&key, &block_group->key, sizeof(key));
10345 index = btrfs_bg_flags_to_raid_index(block_group->flags);
10346 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
10347 BTRFS_BLOCK_GROUP_RAID1 |
10348 BTRFS_BLOCK_GROUP_RAID10))
10353 /* make sure this block group isn't part of an allocation cluster */
10354 cluster = &fs_info->data_alloc_cluster;
10355 spin_lock(&cluster->refill_lock);
10356 btrfs_return_cluster_to_free_space(block_group, cluster);
10357 spin_unlock(&cluster->refill_lock);
10360 * make sure this block group isn't part of a metadata
10361 * allocation cluster
10363 cluster = &fs_info->meta_alloc_cluster;
10364 spin_lock(&cluster->refill_lock);
10365 btrfs_return_cluster_to_free_space(block_group, cluster);
10366 spin_unlock(&cluster->refill_lock);
10368 path = btrfs_alloc_path();
10375 * get the inode first so any iput calls done for the io_list
10376 * aren't the final iput (no unlinks allowed now)
10378 inode = lookup_free_space_inode(fs_info, block_group, path);
10380 mutex_lock(&trans->transaction->cache_write_mutex);
10382 * make sure our free spache cache IO is done before remove the
10385 spin_lock(&trans->transaction->dirty_bgs_lock);
10386 if (!list_empty(&block_group->io_list)) {
10387 list_del_init(&block_group->io_list);
10389 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
10391 spin_unlock(&trans->transaction->dirty_bgs_lock);
10392 btrfs_wait_cache_io(trans, block_group, path);
10393 btrfs_put_block_group(block_group);
10394 spin_lock(&trans->transaction->dirty_bgs_lock);
10397 if (!list_empty(&block_group->dirty_list)) {
10398 list_del_init(&block_group->dirty_list);
10399 btrfs_put_block_group(block_group);
10401 spin_unlock(&trans->transaction->dirty_bgs_lock);
10402 mutex_unlock(&trans->transaction->cache_write_mutex);
10404 if (!IS_ERR(inode)) {
10405 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10407 btrfs_add_delayed_iput(inode);
10410 clear_nlink(inode);
10411 /* One for the block groups ref */
10412 spin_lock(&block_group->lock);
10413 if (block_group->iref) {
10414 block_group->iref = 0;
10415 block_group->inode = NULL;
10416 spin_unlock(&block_group->lock);
10419 spin_unlock(&block_group->lock);
10421 /* One for our lookup ref */
10422 btrfs_add_delayed_iput(inode);
10425 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
10426 key.offset = block_group->key.objectid;
10429 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
10433 btrfs_release_path(path);
10435 ret = btrfs_del_item(trans, tree_root, path);
10438 btrfs_release_path(path);
10441 spin_lock(&fs_info->block_group_cache_lock);
10442 rb_erase(&block_group->cache_node,
10443 &fs_info->block_group_cache_tree);
10444 RB_CLEAR_NODE(&block_group->cache_node);
10446 if (fs_info->first_logical_byte == block_group->key.objectid)
10447 fs_info->first_logical_byte = (u64)-1;
10448 spin_unlock(&fs_info->block_group_cache_lock);
10450 down_write(&block_group->space_info->groups_sem);
10452 * we must use list_del_init so people can check to see if they
10453 * are still on the list after taking the semaphore
10455 list_del_init(&block_group->list);
10456 if (list_empty(&block_group->space_info->block_groups[index])) {
10457 kobj = block_group->space_info->block_group_kobjs[index];
10458 block_group->space_info->block_group_kobjs[index] = NULL;
10459 clear_avail_alloc_bits(fs_info, block_group->flags);
10461 up_write(&block_group->space_info->groups_sem);
10467 if (block_group->has_caching_ctl)
10468 caching_ctl = get_caching_control(block_group);
10469 if (block_group->cached == BTRFS_CACHE_STARTED)
10470 wait_block_group_cache_done(block_group);
10471 if (block_group->has_caching_ctl) {
10472 down_write(&fs_info->commit_root_sem);
10473 if (!caching_ctl) {
10474 struct btrfs_caching_control *ctl;
10476 list_for_each_entry(ctl,
10477 &fs_info->caching_block_groups, list)
10478 if (ctl->block_group == block_group) {
10480 refcount_inc(&caching_ctl->count);
10485 list_del_init(&caching_ctl->list);
10486 up_write(&fs_info->commit_root_sem);
10488 /* Once for the caching bgs list and once for us. */
10489 put_caching_control(caching_ctl);
10490 put_caching_control(caching_ctl);
10494 spin_lock(&trans->transaction->dirty_bgs_lock);
10495 if (!list_empty(&block_group->dirty_list)) {
10498 if (!list_empty(&block_group->io_list)) {
10501 spin_unlock(&trans->transaction->dirty_bgs_lock);
10502 btrfs_remove_free_space_cache(block_group);
10504 spin_lock(&block_group->space_info->lock);
10505 list_del_init(&block_group->ro_list);
10507 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
10508 WARN_ON(block_group->space_info->total_bytes
10509 < block_group->key.offset);
10510 WARN_ON(block_group->space_info->bytes_readonly
10511 < block_group->key.offset);
10512 WARN_ON(block_group->space_info->disk_total
10513 < block_group->key.offset * factor);
10515 block_group->space_info->total_bytes -= block_group->key.offset;
10516 block_group->space_info->bytes_readonly -= block_group->key.offset;
10517 block_group->space_info->disk_total -= block_group->key.offset * factor;
10519 spin_unlock(&block_group->space_info->lock);
10521 memcpy(&key, &block_group->key, sizeof(key));
10523 mutex_lock(&fs_info->chunk_mutex);
10524 if (!list_empty(&em->list)) {
10525 /* We're in the transaction->pending_chunks list. */
10526 free_extent_map(em);
10528 spin_lock(&block_group->lock);
10529 block_group->removed = 1;
10531 * At this point trimming can't start on this block group, because we
10532 * removed the block group from the tree fs_info->block_group_cache_tree
10533 * so no one can't find it anymore and even if someone already got this
10534 * block group before we removed it from the rbtree, they have already
10535 * incremented block_group->trimming - if they didn't, they won't find
10536 * any free space entries because we already removed them all when we
10537 * called btrfs_remove_free_space_cache().
10539 * And we must not remove the extent map from the fs_info->mapping_tree
10540 * to prevent the same logical address range and physical device space
10541 * ranges from being reused for a new block group. This is because our
10542 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10543 * completely transactionless, so while it is trimming a range the
10544 * currently running transaction might finish and a new one start,
10545 * allowing for new block groups to be created that can reuse the same
10546 * physical device locations unless we take this special care.
10548 * There may also be an implicit trim operation if the file system
10549 * is mounted with -odiscard. The same protections must remain
10550 * in place until the extents have been discarded completely when
10551 * the transaction commit has completed.
10553 remove_em = (atomic_read(&block_group->trimming) == 0);
10555 * Make sure a trimmer task always sees the em in the pinned_chunks list
10556 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10557 * before checking block_group->removed).
10561 * Our em might be in trans->transaction->pending_chunks which
10562 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10563 * and so is the fs_info->pinned_chunks list.
10565 * So at this point we must be holding the chunk_mutex to avoid
10566 * any races with chunk allocation (more specifically at
10567 * volumes.c:contains_pending_extent()), to ensure it always
10568 * sees the em, either in the pending_chunks list or in the
10569 * pinned_chunks list.
10571 list_move_tail(&em->list, &fs_info->pinned_chunks);
10573 spin_unlock(&block_group->lock);
10576 struct extent_map_tree *em_tree;
10578 em_tree = &fs_info->mapping_tree.map_tree;
10579 write_lock(&em_tree->lock);
10581 * The em might be in the pending_chunks list, so make sure the
10582 * chunk mutex is locked, since remove_extent_mapping() will
10583 * delete us from that list.
10585 remove_extent_mapping(em_tree, em);
10586 write_unlock(&em_tree->lock);
10587 /* once for the tree */
10588 free_extent_map(em);
10591 mutex_unlock(&fs_info->chunk_mutex);
10593 ret = remove_block_group_free_space(trans, block_group);
10597 btrfs_put_block_group(block_group);
10598 btrfs_put_block_group(block_group);
10600 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10606 ret = btrfs_del_item(trans, root, path);
10608 btrfs_free_path(path);
10612 struct btrfs_trans_handle *
10613 btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info,
10614 const u64 chunk_offset)
10616 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
10617 struct extent_map *em;
10618 struct map_lookup *map;
10619 unsigned int num_items;
10621 read_lock(&em_tree->lock);
10622 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
10623 read_unlock(&em_tree->lock);
10624 ASSERT(em && em->start == chunk_offset);
10627 * We need to reserve 3 + N units from the metadata space info in order
10628 * to remove a block group (done at btrfs_remove_chunk() and at
10629 * btrfs_remove_block_group()), which are used for:
10631 * 1 unit for adding the free space inode's orphan (located in the tree
10633 * 1 unit for deleting the block group item (located in the extent
10635 * 1 unit for deleting the free space item (located in tree of tree
10637 * N units for deleting N device extent items corresponding to each
10638 * stripe (located in the device tree).
10640 * In order to remove a block group we also need to reserve units in the
10641 * system space info in order to update the chunk tree (update one or
10642 * more device items and remove one chunk item), but this is done at
10643 * btrfs_remove_chunk() through a call to check_system_chunk().
10645 map = em->map_lookup;
10646 num_items = 3 + map->num_stripes;
10647 free_extent_map(em);
10649 return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root,
10654 * Process the unused_bgs list and remove any that don't have any allocated
10655 * space inside of them.
10657 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10659 struct btrfs_block_group_cache *block_group;
10660 struct btrfs_space_info *space_info;
10661 struct btrfs_trans_handle *trans;
10664 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
10667 spin_lock(&fs_info->unused_bgs_lock);
10668 while (!list_empty(&fs_info->unused_bgs)) {
10672 block_group = list_first_entry(&fs_info->unused_bgs,
10673 struct btrfs_block_group_cache,
10675 list_del_init(&block_group->bg_list);
10677 space_info = block_group->space_info;
10679 if (ret || btrfs_mixed_space_info(space_info)) {
10680 btrfs_put_block_group(block_group);
10683 spin_unlock(&fs_info->unused_bgs_lock);
10685 mutex_lock(&fs_info->delete_unused_bgs_mutex);
10687 /* Don't want to race with allocators so take the groups_sem */
10688 down_write(&space_info->groups_sem);
10689 spin_lock(&block_group->lock);
10690 if (block_group->reserved ||
10691 btrfs_block_group_used(&block_group->item) ||
10693 list_is_singular(&block_group->list)) {
10695 * We want to bail if we made new allocations or have
10696 * outstanding allocations in this block group. We do
10697 * the ro check in case balance is currently acting on
10698 * this block group.
10700 trace_btrfs_skip_unused_block_group(block_group);
10701 spin_unlock(&block_group->lock);
10702 up_write(&space_info->groups_sem);
10705 spin_unlock(&block_group->lock);
10707 /* We don't want to force the issue, only flip if it's ok. */
10708 ret = inc_block_group_ro(block_group, 0);
10709 up_write(&space_info->groups_sem);
10716 * Want to do this before we do anything else so we can recover
10717 * properly if we fail to join the transaction.
10719 trans = btrfs_start_trans_remove_block_group(fs_info,
10720 block_group->key.objectid);
10721 if (IS_ERR(trans)) {
10722 btrfs_dec_block_group_ro(block_group);
10723 ret = PTR_ERR(trans);
10728 * We could have pending pinned extents for this block group,
10729 * just delete them, we don't care about them anymore.
10731 start = block_group->key.objectid;
10732 end = start + block_group->key.offset - 1;
10734 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10735 * btrfs_finish_extent_commit(). If we are at transaction N,
10736 * another task might be running finish_extent_commit() for the
10737 * previous transaction N - 1, and have seen a range belonging
10738 * to the block group in freed_extents[] before we were able to
10739 * clear the whole block group range from freed_extents[]. This
10740 * means that task can lookup for the block group after we
10741 * unpinned it from freed_extents[] and removed it, leading to
10742 * a BUG_ON() at btrfs_unpin_extent_range().
10744 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10745 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10748 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10749 btrfs_dec_block_group_ro(block_group);
10752 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10755 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10756 btrfs_dec_block_group_ro(block_group);
10759 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10761 /* Reset pinned so btrfs_put_block_group doesn't complain */
10762 spin_lock(&space_info->lock);
10763 spin_lock(&block_group->lock);
10765 space_info->bytes_pinned -= block_group->pinned;
10766 space_info->bytes_readonly += block_group->pinned;
10767 percpu_counter_add(&space_info->total_bytes_pinned,
10768 -block_group->pinned);
10769 block_group->pinned = 0;
10771 spin_unlock(&block_group->lock);
10772 spin_unlock(&space_info->lock);
10774 /* DISCARD can flip during remount */
10775 trimming = btrfs_test_opt(fs_info, DISCARD);
10777 /* Implicit trim during transaction commit. */
10779 btrfs_get_block_group_trimming(block_group);
10782 * Btrfs_remove_chunk will abort the transaction if things go
10785 ret = btrfs_remove_chunk(trans, fs_info,
10786 block_group->key.objectid);
10790 btrfs_put_block_group_trimming(block_group);
10795 * If we're not mounted with -odiscard, we can just forget
10796 * about this block group. Otherwise we'll need to wait
10797 * until transaction commit to do the actual discard.
10800 spin_lock(&fs_info->unused_bgs_lock);
10802 * A concurrent scrub might have added us to the list
10803 * fs_info->unused_bgs, so use a list_move operation
10804 * to add the block group to the deleted_bgs list.
10806 list_move(&block_group->bg_list,
10807 &trans->transaction->deleted_bgs);
10808 spin_unlock(&fs_info->unused_bgs_lock);
10809 btrfs_get_block_group(block_group);
10812 btrfs_end_transaction(trans);
10814 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
10815 btrfs_put_block_group(block_group);
10816 spin_lock(&fs_info->unused_bgs_lock);
10818 spin_unlock(&fs_info->unused_bgs_lock);
10821 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10823 struct btrfs_super_block *disk_super;
10829 disk_super = fs_info->super_copy;
10830 if (!btrfs_super_root(disk_super))
10833 features = btrfs_super_incompat_flags(disk_super);
10834 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10837 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10838 ret = create_space_info(fs_info, flags);
10843 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10844 ret = create_space_info(fs_info, flags);
10846 flags = BTRFS_BLOCK_GROUP_METADATA;
10847 ret = create_space_info(fs_info, flags);
10851 flags = BTRFS_BLOCK_GROUP_DATA;
10852 ret = create_space_info(fs_info, flags);
10858 int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
10859 u64 start, u64 end)
10861 return unpin_extent_range(fs_info, start, end, false);
10865 * It used to be that old block groups would be left around forever.
10866 * Iterating over them would be enough to trim unused space. Since we
10867 * now automatically remove them, we also need to iterate over unallocated
10870 * We don't want a transaction for this since the discard may take a
10871 * substantial amount of time. We don't require that a transaction be
10872 * running, but we do need to take a running transaction into account
10873 * to ensure that we're not discarding chunks that were released in
10874 * the current transaction.
10876 * Holding the chunks lock will prevent other threads from allocating
10877 * or releasing chunks, but it won't prevent a running transaction
10878 * from committing and releasing the memory that the pending chunks
10879 * list head uses. For that, we need to take a reference to the
10882 static int btrfs_trim_free_extents(struct btrfs_device *device,
10883 u64 minlen, u64 *trimmed)
10885 u64 start = 0, len = 0;
10890 /* Not writeable = nothing to do. */
10891 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
10894 /* No free space = nothing to do. */
10895 if (device->total_bytes <= device->bytes_used)
10901 struct btrfs_fs_info *fs_info = device->fs_info;
10902 struct btrfs_transaction *trans;
10905 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
10909 down_read(&fs_info->commit_root_sem);
10911 spin_lock(&fs_info->trans_lock);
10912 trans = fs_info->running_transaction;
10914 refcount_inc(&trans->use_count);
10915 spin_unlock(&fs_info->trans_lock);
10917 ret = find_free_dev_extent_start(trans, device, minlen, start,
10920 btrfs_put_transaction(trans);
10923 up_read(&fs_info->commit_root_sem);
10924 mutex_unlock(&fs_info->chunk_mutex);
10925 if (ret == -ENOSPC)
10930 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
10931 up_read(&fs_info->commit_root_sem);
10932 mutex_unlock(&fs_info->chunk_mutex);
10940 if (fatal_signal_pending(current)) {
10941 ret = -ERESTARTSYS;
10951 int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
10953 struct btrfs_block_group_cache *cache = NULL;
10954 struct btrfs_device *device;
10955 struct list_head *devices;
10960 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
10964 * try to trim all FS space, our block group may start from non-zero.
10966 if (range->len == total_bytes)
10967 cache = btrfs_lookup_first_block_group(fs_info, range->start);
10969 cache = btrfs_lookup_block_group(fs_info, range->start);
10972 if (cache->key.objectid >= (range->start + range->len)) {
10973 btrfs_put_block_group(cache);
10977 start = max(range->start, cache->key.objectid);
10978 end = min(range->start + range->len,
10979 cache->key.objectid + cache->key.offset);
10981 if (end - start >= range->minlen) {
10982 if (!block_group_cache_done(cache)) {
10983 ret = cache_block_group(cache, 0);
10985 btrfs_put_block_group(cache);
10988 ret = wait_block_group_cache_done(cache);
10990 btrfs_put_block_group(cache);
10994 ret = btrfs_trim_block_group(cache,
11000 trimmed += group_trimmed;
11002 btrfs_put_block_group(cache);
11007 cache = next_block_group(fs_info, cache);
11010 mutex_lock(&fs_info->fs_devices->device_list_mutex);
11011 devices = &fs_info->fs_devices->alloc_list;
11012 list_for_each_entry(device, devices, dev_alloc_list) {
11013 ret = btrfs_trim_free_extents(device, range->minlen,
11018 trimmed += group_trimmed;
11020 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
11022 range->len = trimmed;
11027 * btrfs_{start,end}_write_no_snapshotting() are similar to
11028 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
11029 * data into the page cache through nocow before the subvolume is snapshoted,
11030 * but flush the data into disk after the snapshot creation, or to prevent
11031 * operations while snapshotting is ongoing and that cause the snapshot to be
11032 * inconsistent (writes followed by expanding truncates for example).
11034 void btrfs_end_write_no_snapshotting(struct btrfs_root *root)
11036 percpu_counter_dec(&root->subv_writers->counter);
11037 cond_wake_up(&root->subv_writers->wait);
11040 int btrfs_start_write_no_snapshotting(struct btrfs_root *root)
11042 if (atomic_read(&root->will_be_snapshotted))
11045 percpu_counter_inc(&root->subv_writers->counter);
11047 * Make sure counter is updated before we check for snapshot creation.
11050 if (atomic_read(&root->will_be_snapshotted)) {
11051 btrfs_end_write_no_snapshotting(root);
11057 void btrfs_wait_for_snapshot_creation(struct btrfs_root *root)
11062 ret = btrfs_start_write_no_snapshotting(root);
11065 wait_var_event(&root->will_be_snapshotted,
11066 !atomic_read(&root->will_be_snapshotted));