2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/sched/signal.h>
20 #include <linux/pagemap.h>
21 #include <linux/writeback.h>
22 #include <linux/blkdev.h>
23 #include <linux/sort.h>
24 #include <linux/rcupdate.h>
25 #include <linux/kthread.h>
26 #include <linux/slab.h>
27 #include <linux/ratelimit.h>
28 #include <linux/percpu_counter.h>
29 #include <linux/lockdep.h>
30 #include <linux/crc32c.h>
33 #include "print-tree.h"
37 #include "free-space-cache.h"
38 #include "free-space-tree.h"
42 #include "ref-verify.h"
44 #undef SCRAMBLE_DELAYED_REFS
47 * control flags for do_chunk_alloc's force field
48 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
49 * if we really need one.
51 * CHUNK_ALLOC_LIMITED means to only try and allocate one
52 * if we have very few chunks already allocated. This is
53 * used as part of the clustering code to help make sure
54 * we have a good pool of storage to cluster in, without
55 * filling the FS with empty chunks
57 * CHUNK_ALLOC_FORCE means it must try to allocate one
61 CHUNK_ALLOC_NO_FORCE = 0,
62 CHUNK_ALLOC_LIMITED = 1,
63 CHUNK_ALLOC_FORCE = 2,
66 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
67 struct btrfs_fs_info *fs_info,
68 struct btrfs_delayed_ref_node *node, u64 parent,
69 u64 root_objectid, u64 owner_objectid,
70 u64 owner_offset, int refs_to_drop,
71 struct btrfs_delayed_extent_op *extra_op);
72 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
73 struct extent_buffer *leaf,
74 struct btrfs_extent_item *ei);
75 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
76 struct btrfs_fs_info *fs_info,
77 u64 parent, u64 root_objectid,
78 u64 flags, u64 owner, u64 offset,
79 struct btrfs_key *ins, int ref_mod);
80 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
81 struct btrfs_fs_info *fs_info,
82 u64 parent, u64 root_objectid,
83 u64 flags, struct btrfs_disk_key *key,
84 int level, struct btrfs_key *ins);
85 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
86 struct btrfs_fs_info *fs_info, u64 flags,
88 static int find_next_key(struct btrfs_path *path, int level,
89 struct btrfs_key *key);
90 static void dump_space_info(struct btrfs_fs_info *fs_info,
91 struct btrfs_space_info *info, u64 bytes,
92 int dump_block_groups);
93 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
95 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
96 struct btrfs_space_info *space_info,
98 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
99 struct btrfs_space_info *space_info,
103 block_group_cache_done(struct btrfs_block_group_cache *cache)
106 return cache->cached == BTRFS_CACHE_FINISHED ||
107 cache->cached == BTRFS_CACHE_ERROR;
110 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
112 return (cache->flags & bits) == bits;
115 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
117 atomic_inc(&cache->count);
120 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
122 if (atomic_dec_and_test(&cache->count)) {
123 WARN_ON(cache->pinned > 0);
124 WARN_ON(cache->reserved > 0);
127 * If not empty, someone is still holding mutex of
128 * full_stripe_lock, which can only be released by caller.
129 * And it will definitely cause use-after-free when caller
130 * tries to release full stripe lock.
132 * No better way to resolve, but only to warn.
134 WARN_ON(!RB_EMPTY_ROOT(&cache->full_stripe_locks_root.root));
135 kfree(cache->free_space_ctl);
141 * this adds the block group to the fs_info rb tree for the block group
144 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
145 struct btrfs_block_group_cache *block_group)
148 struct rb_node *parent = NULL;
149 struct btrfs_block_group_cache *cache;
151 spin_lock(&info->block_group_cache_lock);
152 p = &info->block_group_cache_tree.rb_node;
156 cache = rb_entry(parent, struct btrfs_block_group_cache,
158 if (block_group->key.objectid < cache->key.objectid) {
160 } else if (block_group->key.objectid > cache->key.objectid) {
163 spin_unlock(&info->block_group_cache_lock);
168 rb_link_node(&block_group->cache_node, parent, p);
169 rb_insert_color(&block_group->cache_node,
170 &info->block_group_cache_tree);
172 if (info->first_logical_byte > block_group->key.objectid)
173 info->first_logical_byte = block_group->key.objectid;
175 spin_unlock(&info->block_group_cache_lock);
181 * This will return the block group at or after bytenr if contains is 0, else
182 * it will return the block group that contains the bytenr
184 static struct btrfs_block_group_cache *
185 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
188 struct btrfs_block_group_cache *cache, *ret = NULL;
192 spin_lock(&info->block_group_cache_lock);
193 n = info->block_group_cache_tree.rb_node;
196 cache = rb_entry(n, struct btrfs_block_group_cache,
198 end = cache->key.objectid + cache->key.offset - 1;
199 start = cache->key.objectid;
201 if (bytenr < start) {
202 if (!contains && (!ret || start < ret->key.objectid))
205 } else if (bytenr > start) {
206 if (contains && bytenr <= end) {
217 btrfs_get_block_group(ret);
218 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
219 info->first_logical_byte = ret->key.objectid;
221 spin_unlock(&info->block_group_cache_lock);
226 static int add_excluded_extent(struct btrfs_fs_info *fs_info,
227 u64 start, u64 num_bytes)
229 u64 end = start + num_bytes - 1;
230 set_extent_bits(&fs_info->freed_extents[0],
231 start, end, EXTENT_UPTODATE);
232 set_extent_bits(&fs_info->freed_extents[1],
233 start, end, EXTENT_UPTODATE);
237 static void free_excluded_extents(struct btrfs_fs_info *fs_info,
238 struct btrfs_block_group_cache *cache)
242 start = cache->key.objectid;
243 end = start + cache->key.offset - 1;
245 clear_extent_bits(&fs_info->freed_extents[0],
246 start, end, EXTENT_UPTODATE);
247 clear_extent_bits(&fs_info->freed_extents[1],
248 start, end, EXTENT_UPTODATE);
251 static int exclude_super_stripes(struct btrfs_fs_info *fs_info,
252 struct btrfs_block_group_cache *cache)
259 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
260 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
261 cache->bytes_super += stripe_len;
262 ret = add_excluded_extent(fs_info, cache->key.objectid,
268 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
269 bytenr = btrfs_sb_offset(i);
270 ret = btrfs_rmap_block(fs_info, cache->key.objectid,
271 bytenr, 0, &logical, &nr, &stripe_len);
278 if (logical[nr] > cache->key.objectid +
282 if (logical[nr] + stripe_len <= cache->key.objectid)
286 if (start < cache->key.objectid) {
287 start = cache->key.objectid;
288 len = (logical[nr] + stripe_len) - start;
290 len = min_t(u64, stripe_len,
291 cache->key.objectid +
292 cache->key.offset - start);
295 cache->bytes_super += len;
296 ret = add_excluded_extent(fs_info, start, len);
308 static struct btrfs_caching_control *
309 get_caching_control(struct btrfs_block_group_cache *cache)
311 struct btrfs_caching_control *ctl;
313 spin_lock(&cache->lock);
314 if (!cache->caching_ctl) {
315 spin_unlock(&cache->lock);
319 ctl = cache->caching_ctl;
320 refcount_inc(&ctl->count);
321 spin_unlock(&cache->lock);
325 static void put_caching_control(struct btrfs_caching_control *ctl)
327 if (refcount_dec_and_test(&ctl->count))
331 #ifdef CONFIG_BTRFS_DEBUG
332 static void fragment_free_space(struct btrfs_block_group_cache *block_group)
334 struct btrfs_fs_info *fs_info = block_group->fs_info;
335 u64 start = block_group->key.objectid;
336 u64 len = block_group->key.offset;
337 u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ?
338 fs_info->nodesize : fs_info->sectorsize;
339 u64 step = chunk << 1;
341 while (len > chunk) {
342 btrfs_remove_free_space(block_group, start, chunk);
353 * this is only called by cache_block_group, since we could have freed extents
354 * we need to check the pinned_extents for any extents that can't be used yet
355 * since their free space will be released as soon as the transaction commits.
357 u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
358 struct btrfs_fs_info *info, u64 start, u64 end)
360 u64 extent_start, extent_end, size, total_added = 0;
363 while (start < end) {
364 ret = find_first_extent_bit(info->pinned_extents, start,
365 &extent_start, &extent_end,
366 EXTENT_DIRTY | EXTENT_UPTODATE,
371 if (extent_start <= start) {
372 start = extent_end + 1;
373 } else if (extent_start > start && extent_start < end) {
374 size = extent_start - start;
376 ret = btrfs_add_free_space(block_group, start,
378 BUG_ON(ret); /* -ENOMEM or logic error */
379 start = extent_end + 1;
388 ret = btrfs_add_free_space(block_group, start, size);
389 BUG_ON(ret); /* -ENOMEM or logic error */
395 static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl)
397 struct btrfs_block_group_cache *block_group = caching_ctl->block_group;
398 struct btrfs_fs_info *fs_info = block_group->fs_info;
399 struct btrfs_root *extent_root = fs_info->extent_root;
400 struct btrfs_path *path;
401 struct extent_buffer *leaf;
402 struct btrfs_key key;
409 path = btrfs_alloc_path();
413 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
415 #ifdef CONFIG_BTRFS_DEBUG
417 * If we're fragmenting we don't want to make anybody think we can
418 * allocate from this block group until we've had a chance to fragment
421 if (btrfs_should_fragment_free_space(block_group))
425 * We don't want to deadlock with somebody trying to allocate a new
426 * extent for the extent root while also trying to search the extent
427 * root to add free space. So we skip locking and search the commit
428 * root, since its read-only
430 path->skip_locking = 1;
431 path->search_commit_root = 1;
432 path->reada = READA_FORWARD;
436 key.type = BTRFS_EXTENT_ITEM_KEY;
439 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
443 leaf = path->nodes[0];
444 nritems = btrfs_header_nritems(leaf);
447 if (btrfs_fs_closing(fs_info) > 1) {
452 if (path->slots[0] < nritems) {
453 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
455 ret = find_next_key(path, 0, &key);
459 if (need_resched() ||
460 rwsem_is_contended(&fs_info->commit_root_sem)) {
462 caching_ctl->progress = last;
463 btrfs_release_path(path);
464 up_read(&fs_info->commit_root_sem);
465 mutex_unlock(&caching_ctl->mutex);
467 mutex_lock(&caching_ctl->mutex);
468 down_read(&fs_info->commit_root_sem);
472 ret = btrfs_next_leaf(extent_root, path);
477 leaf = path->nodes[0];
478 nritems = btrfs_header_nritems(leaf);
482 if (key.objectid < last) {
485 key.type = BTRFS_EXTENT_ITEM_KEY;
488 caching_ctl->progress = last;
489 btrfs_release_path(path);
493 if (key.objectid < block_group->key.objectid) {
498 if (key.objectid >= block_group->key.objectid +
499 block_group->key.offset)
502 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
503 key.type == BTRFS_METADATA_ITEM_KEY) {
504 total_found += add_new_free_space(block_group,
507 if (key.type == BTRFS_METADATA_ITEM_KEY)
508 last = key.objectid +
511 last = key.objectid + key.offset;
513 if (total_found > CACHING_CTL_WAKE_UP) {
516 wake_up(&caching_ctl->wait);
523 total_found += add_new_free_space(block_group, fs_info, last,
524 block_group->key.objectid +
525 block_group->key.offset);
526 caching_ctl->progress = (u64)-1;
529 btrfs_free_path(path);
533 static noinline void caching_thread(struct btrfs_work *work)
535 struct btrfs_block_group_cache *block_group;
536 struct btrfs_fs_info *fs_info;
537 struct btrfs_caching_control *caching_ctl;
538 struct btrfs_root *extent_root;
541 caching_ctl = container_of(work, struct btrfs_caching_control, work);
542 block_group = caching_ctl->block_group;
543 fs_info = block_group->fs_info;
544 extent_root = fs_info->extent_root;
546 mutex_lock(&caching_ctl->mutex);
547 down_read(&fs_info->commit_root_sem);
549 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
550 ret = load_free_space_tree(caching_ctl);
552 ret = load_extent_tree_free(caching_ctl);
554 spin_lock(&block_group->lock);
555 block_group->caching_ctl = NULL;
556 block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED;
557 spin_unlock(&block_group->lock);
559 #ifdef CONFIG_BTRFS_DEBUG
560 if (btrfs_should_fragment_free_space(block_group)) {
563 spin_lock(&block_group->space_info->lock);
564 spin_lock(&block_group->lock);
565 bytes_used = block_group->key.offset -
566 btrfs_block_group_used(&block_group->item);
567 block_group->space_info->bytes_used += bytes_used >> 1;
568 spin_unlock(&block_group->lock);
569 spin_unlock(&block_group->space_info->lock);
570 fragment_free_space(block_group);
574 caching_ctl->progress = (u64)-1;
576 up_read(&fs_info->commit_root_sem);
577 free_excluded_extents(fs_info, block_group);
578 mutex_unlock(&caching_ctl->mutex);
580 wake_up(&caching_ctl->wait);
582 put_caching_control(caching_ctl);
583 btrfs_put_block_group(block_group);
586 static int cache_block_group(struct btrfs_block_group_cache *cache,
590 struct btrfs_fs_info *fs_info = cache->fs_info;
591 struct btrfs_caching_control *caching_ctl;
594 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
598 INIT_LIST_HEAD(&caching_ctl->list);
599 mutex_init(&caching_ctl->mutex);
600 init_waitqueue_head(&caching_ctl->wait);
601 caching_ctl->block_group = cache;
602 caching_ctl->progress = cache->key.objectid;
603 refcount_set(&caching_ctl->count, 1);
604 btrfs_init_work(&caching_ctl->work, btrfs_cache_helper,
605 caching_thread, NULL, NULL);
607 spin_lock(&cache->lock);
609 * This should be a rare occasion, but this could happen I think in the
610 * case where one thread starts to load the space cache info, and then
611 * some other thread starts a transaction commit which tries to do an
612 * allocation while the other thread is still loading the space cache
613 * info. The previous loop should have kept us from choosing this block
614 * group, but if we've moved to the state where we will wait on caching
615 * block groups we need to first check if we're doing a fast load here,
616 * so we can wait for it to finish, otherwise we could end up allocating
617 * from a block group who's cache gets evicted for one reason or
620 while (cache->cached == BTRFS_CACHE_FAST) {
621 struct btrfs_caching_control *ctl;
623 ctl = cache->caching_ctl;
624 refcount_inc(&ctl->count);
625 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
626 spin_unlock(&cache->lock);
630 finish_wait(&ctl->wait, &wait);
631 put_caching_control(ctl);
632 spin_lock(&cache->lock);
635 if (cache->cached != BTRFS_CACHE_NO) {
636 spin_unlock(&cache->lock);
640 WARN_ON(cache->caching_ctl);
641 cache->caching_ctl = caching_ctl;
642 cache->cached = BTRFS_CACHE_FAST;
643 spin_unlock(&cache->lock);
645 if (btrfs_test_opt(fs_info, SPACE_CACHE)) {
646 mutex_lock(&caching_ctl->mutex);
647 ret = load_free_space_cache(fs_info, cache);
649 spin_lock(&cache->lock);
651 cache->caching_ctl = NULL;
652 cache->cached = BTRFS_CACHE_FINISHED;
653 cache->last_byte_to_unpin = (u64)-1;
654 caching_ctl->progress = (u64)-1;
656 if (load_cache_only) {
657 cache->caching_ctl = NULL;
658 cache->cached = BTRFS_CACHE_NO;
660 cache->cached = BTRFS_CACHE_STARTED;
661 cache->has_caching_ctl = 1;
664 spin_unlock(&cache->lock);
665 #ifdef CONFIG_BTRFS_DEBUG
667 btrfs_should_fragment_free_space(cache)) {
670 spin_lock(&cache->space_info->lock);
671 spin_lock(&cache->lock);
672 bytes_used = cache->key.offset -
673 btrfs_block_group_used(&cache->item);
674 cache->space_info->bytes_used += bytes_used >> 1;
675 spin_unlock(&cache->lock);
676 spin_unlock(&cache->space_info->lock);
677 fragment_free_space(cache);
680 mutex_unlock(&caching_ctl->mutex);
682 wake_up(&caching_ctl->wait);
684 put_caching_control(caching_ctl);
685 free_excluded_extents(fs_info, cache);
690 * We're either using the free space tree or no caching at all.
691 * Set cached to the appropriate value and wakeup any waiters.
693 spin_lock(&cache->lock);
694 if (load_cache_only) {
695 cache->caching_ctl = NULL;
696 cache->cached = BTRFS_CACHE_NO;
698 cache->cached = BTRFS_CACHE_STARTED;
699 cache->has_caching_ctl = 1;
701 spin_unlock(&cache->lock);
702 wake_up(&caching_ctl->wait);
705 if (load_cache_only) {
706 put_caching_control(caching_ctl);
710 down_write(&fs_info->commit_root_sem);
711 refcount_inc(&caching_ctl->count);
712 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
713 up_write(&fs_info->commit_root_sem);
715 btrfs_get_block_group(cache);
717 btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
723 * return the block group that starts at or after bytenr
725 static struct btrfs_block_group_cache *
726 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
728 return block_group_cache_tree_search(info, bytenr, 0);
732 * return the block group that contains the given bytenr
734 struct btrfs_block_group_cache *btrfs_lookup_block_group(
735 struct btrfs_fs_info *info,
738 return block_group_cache_tree_search(info, bytenr, 1);
741 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
744 struct list_head *head = &info->space_info;
745 struct btrfs_space_info *found;
747 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
750 list_for_each_entry_rcu(found, head, list) {
751 if (found->flags & flags) {
760 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, s64 num_bytes,
761 u64 owner, u64 root_objectid)
763 struct btrfs_space_info *space_info;
766 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
767 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
768 flags = BTRFS_BLOCK_GROUP_SYSTEM;
770 flags = BTRFS_BLOCK_GROUP_METADATA;
772 flags = BTRFS_BLOCK_GROUP_DATA;
775 space_info = __find_space_info(fs_info, flags);
777 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
781 * after adding space to the filesystem, we need to clear the full flags
782 * on all the space infos.
784 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
786 struct list_head *head = &info->space_info;
787 struct btrfs_space_info *found;
790 list_for_each_entry_rcu(found, head, list)
795 /* simple helper to search for an existing data extent at a given offset */
796 int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len)
799 struct btrfs_key key;
800 struct btrfs_path *path;
802 path = btrfs_alloc_path();
806 key.objectid = start;
808 key.type = BTRFS_EXTENT_ITEM_KEY;
809 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
810 btrfs_free_path(path);
815 * helper function to lookup reference count and flags of a tree block.
817 * the head node for delayed ref is used to store the sum of all the
818 * reference count modifications queued up in the rbtree. the head
819 * node may also store the extent flags to set. This way you can check
820 * to see what the reference count and extent flags would be if all of
821 * the delayed refs are not processed.
823 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
824 struct btrfs_fs_info *fs_info, u64 bytenr,
825 u64 offset, int metadata, u64 *refs, u64 *flags)
827 struct btrfs_delayed_ref_head *head;
828 struct btrfs_delayed_ref_root *delayed_refs;
829 struct btrfs_path *path;
830 struct btrfs_extent_item *ei;
831 struct extent_buffer *leaf;
832 struct btrfs_key key;
839 * If we don't have skinny metadata, don't bother doing anything
842 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) {
843 offset = fs_info->nodesize;
847 path = btrfs_alloc_path();
852 path->skip_locking = 1;
853 path->search_commit_root = 1;
857 key.objectid = bytenr;
860 key.type = BTRFS_METADATA_ITEM_KEY;
862 key.type = BTRFS_EXTENT_ITEM_KEY;
864 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
868 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
869 if (path->slots[0]) {
871 btrfs_item_key_to_cpu(path->nodes[0], &key,
873 if (key.objectid == bytenr &&
874 key.type == BTRFS_EXTENT_ITEM_KEY &&
875 key.offset == fs_info->nodesize)
881 leaf = path->nodes[0];
882 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
883 if (item_size >= sizeof(*ei)) {
884 ei = btrfs_item_ptr(leaf, path->slots[0],
885 struct btrfs_extent_item);
886 num_refs = btrfs_extent_refs(leaf, ei);
887 extent_flags = btrfs_extent_flags(leaf, ei);
889 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
890 struct btrfs_extent_item_v0 *ei0;
891 BUG_ON(item_size != sizeof(*ei0));
892 ei0 = btrfs_item_ptr(leaf, path->slots[0],
893 struct btrfs_extent_item_v0);
894 num_refs = btrfs_extent_refs_v0(leaf, ei0);
895 /* FIXME: this isn't correct for data */
896 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
901 BUG_ON(num_refs == 0);
911 delayed_refs = &trans->transaction->delayed_refs;
912 spin_lock(&delayed_refs->lock);
913 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
915 if (!mutex_trylock(&head->mutex)) {
916 refcount_inc(&head->refs);
917 spin_unlock(&delayed_refs->lock);
919 btrfs_release_path(path);
922 * Mutex was contended, block until it's released and try
925 mutex_lock(&head->mutex);
926 mutex_unlock(&head->mutex);
927 btrfs_put_delayed_ref_head(head);
930 spin_lock(&head->lock);
931 if (head->extent_op && head->extent_op->update_flags)
932 extent_flags |= head->extent_op->flags_to_set;
934 BUG_ON(num_refs == 0);
936 num_refs += head->ref_mod;
937 spin_unlock(&head->lock);
938 mutex_unlock(&head->mutex);
940 spin_unlock(&delayed_refs->lock);
942 WARN_ON(num_refs == 0);
946 *flags = extent_flags;
948 btrfs_free_path(path);
953 * Back reference rules. Back refs have three main goals:
955 * 1) differentiate between all holders of references to an extent so that
956 * when a reference is dropped we can make sure it was a valid reference
957 * before freeing the extent.
959 * 2) Provide enough information to quickly find the holders of an extent
960 * if we notice a given block is corrupted or bad.
962 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
963 * maintenance. This is actually the same as #2, but with a slightly
964 * different use case.
966 * There are two kinds of back refs. The implicit back refs is optimized
967 * for pointers in non-shared tree blocks. For a given pointer in a block,
968 * back refs of this kind provide information about the block's owner tree
969 * and the pointer's key. These information allow us to find the block by
970 * b-tree searching. The full back refs is for pointers in tree blocks not
971 * referenced by their owner trees. The location of tree block is recorded
972 * in the back refs. Actually the full back refs is generic, and can be
973 * used in all cases the implicit back refs is used. The major shortcoming
974 * of the full back refs is its overhead. Every time a tree block gets
975 * COWed, we have to update back refs entry for all pointers in it.
977 * For a newly allocated tree block, we use implicit back refs for
978 * pointers in it. This means most tree related operations only involve
979 * implicit back refs. For a tree block created in old transaction, the
980 * only way to drop a reference to it is COW it. So we can detect the
981 * event that tree block loses its owner tree's reference and do the
982 * back refs conversion.
984 * When a tree block is COWed through a tree, there are four cases:
986 * The reference count of the block is one and the tree is the block's
987 * owner tree. Nothing to do in this case.
989 * The reference count of the block is one and the tree is not the
990 * block's owner tree. In this case, full back refs is used for pointers
991 * in the block. Remove these full back refs, add implicit back refs for
992 * every pointers in the new block.
994 * The reference count of the block is greater than one and the tree is
995 * the block's owner tree. In this case, implicit back refs is used for
996 * pointers in the block. Add full back refs for every pointers in the
997 * block, increase lower level extents' reference counts. The original
998 * implicit back refs are entailed to the new block.
1000 * The reference count of the block is greater than one and the tree is
1001 * not the block's owner tree. Add implicit back refs for every pointer in
1002 * the new block, increase lower level extents' reference count.
1004 * Back Reference Key composing:
1006 * The key objectid corresponds to the first byte in the extent,
1007 * The key type is used to differentiate between types of back refs.
1008 * There are different meanings of the key offset for different types
1011 * File extents can be referenced by:
1013 * - multiple snapshots, subvolumes, or different generations in one subvol
1014 * - different files inside a single subvolume
1015 * - different offsets inside a file (bookend extents in file.c)
1017 * The extent ref structure for the implicit back refs has fields for:
1019 * - Objectid of the subvolume root
1020 * - objectid of the file holding the reference
1021 * - original offset in the file
1022 * - how many bookend extents
1024 * The key offset for the implicit back refs is hash of the first
1027 * The extent ref structure for the full back refs has field for:
1029 * - number of pointers in the tree leaf
1031 * The key offset for the implicit back refs is the first byte of
1034 * When a file extent is allocated, The implicit back refs is used.
1035 * the fields are filled in:
1037 * (root_key.objectid, inode objectid, offset in file, 1)
1039 * When a file extent is removed file truncation, we find the
1040 * corresponding implicit back refs and check the following fields:
1042 * (btrfs_header_owner(leaf), inode objectid, offset in file)
1044 * Btree extents can be referenced by:
1046 * - Different subvolumes
1048 * Both the implicit back refs and the full back refs for tree blocks
1049 * only consist of key. The key offset for the implicit back refs is
1050 * objectid of block's owner tree. The key offset for the full back refs
1051 * is the first byte of parent block.
1053 * When implicit back refs is used, information about the lowest key and
1054 * level of the tree block are required. These information are stored in
1055 * tree block info structure.
1058 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1059 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
1060 struct btrfs_fs_info *fs_info,
1061 struct btrfs_path *path,
1062 u64 owner, u32 extra_size)
1064 struct btrfs_root *root = fs_info->extent_root;
1065 struct btrfs_extent_item *item;
1066 struct btrfs_extent_item_v0 *ei0;
1067 struct btrfs_extent_ref_v0 *ref0;
1068 struct btrfs_tree_block_info *bi;
1069 struct extent_buffer *leaf;
1070 struct btrfs_key key;
1071 struct btrfs_key found_key;
1072 u32 new_size = sizeof(*item);
1076 leaf = path->nodes[0];
1077 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
1079 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1080 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1081 struct btrfs_extent_item_v0);
1082 refs = btrfs_extent_refs_v0(leaf, ei0);
1084 if (owner == (u64)-1) {
1086 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1087 ret = btrfs_next_leaf(root, path);
1090 BUG_ON(ret > 0); /* Corruption */
1091 leaf = path->nodes[0];
1093 btrfs_item_key_to_cpu(leaf, &found_key,
1095 BUG_ON(key.objectid != found_key.objectid);
1096 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1100 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1101 struct btrfs_extent_ref_v0);
1102 owner = btrfs_ref_objectid_v0(leaf, ref0);
1106 btrfs_release_path(path);
1108 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1109 new_size += sizeof(*bi);
1111 new_size -= sizeof(*ei0);
1112 ret = btrfs_search_slot(trans, root, &key, path,
1113 new_size + extra_size, 1);
1116 BUG_ON(ret); /* Corruption */
1118 btrfs_extend_item(fs_info, path, new_size);
1120 leaf = path->nodes[0];
1121 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1122 btrfs_set_extent_refs(leaf, item, refs);
1123 /* FIXME: get real generation */
1124 btrfs_set_extent_generation(leaf, item, 0);
1125 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1126 btrfs_set_extent_flags(leaf, item,
1127 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1128 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1129 bi = (struct btrfs_tree_block_info *)(item + 1);
1130 /* FIXME: get first key of the block */
1131 memzero_extent_buffer(leaf, (unsigned long)bi, sizeof(*bi));
1132 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1134 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1136 btrfs_mark_buffer_dirty(leaf);
1142 * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required,
1143 * is_data == BTRFS_REF_TYPE_DATA, data type is requried,
1144 * is_data == BTRFS_REF_TYPE_ANY, either type is OK.
1146 int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb,
1147 struct btrfs_extent_inline_ref *iref,
1148 enum btrfs_inline_ref_type is_data)
1150 int type = btrfs_extent_inline_ref_type(eb, iref);
1151 u64 offset = btrfs_extent_inline_ref_offset(eb, iref);
1153 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1154 type == BTRFS_SHARED_BLOCK_REF_KEY ||
1155 type == BTRFS_SHARED_DATA_REF_KEY ||
1156 type == BTRFS_EXTENT_DATA_REF_KEY) {
1157 if (is_data == BTRFS_REF_TYPE_BLOCK) {
1158 if (type == BTRFS_TREE_BLOCK_REF_KEY)
1160 if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1161 ASSERT(eb->fs_info);
1163 * Every shared one has parent tree
1164 * block, which must be aligned to
1168 IS_ALIGNED(offset, eb->fs_info->nodesize))
1171 } else if (is_data == BTRFS_REF_TYPE_DATA) {
1172 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1174 if (type == BTRFS_SHARED_DATA_REF_KEY) {
1175 ASSERT(eb->fs_info);
1177 * Every shared one has parent tree
1178 * block, which must be aligned to
1182 IS_ALIGNED(offset, eb->fs_info->nodesize))
1186 ASSERT(is_data == BTRFS_REF_TYPE_ANY);
1191 btrfs_print_leaf((struct extent_buffer *)eb);
1192 btrfs_err(eb->fs_info, "eb %llu invalid extent inline ref type %d",
1196 return BTRFS_REF_TYPE_INVALID;
1199 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1201 u32 high_crc = ~(u32)0;
1202 u32 low_crc = ~(u32)0;
1205 lenum = cpu_to_le64(root_objectid);
1206 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1207 lenum = cpu_to_le64(owner);
1208 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1209 lenum = cpu_to_le64(offset);
1210 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1212 return ((u64)high_crc << 31) ^ (u64)low_crc;
1215 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1216 struct btrfs_extent_data_ref *ref)
1218 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1219 btrfs_extent_data_ref_objectid(leaf, ref),
1220 btrfs_extent_data_ref_offset(leaf, ref));
1223 static int match_extent_data_ref(struct extent_buffer *leaf,
1224 struct btrfs_extent_data_ref *ref,
1225 u64 root_objectid, u64 owner, u64 offset)
1227 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1228 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1229 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1234 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1235 struct btrfs_fs_info *fs_info,
1236 struct btrfs_path *path,
1237 u64 bytenr, u64 parent,
1239 u64 owner, u64 offset)
1241 struct btrfs_root *root = fs_info->extent_root;
1242 struct btrfs_key key;
1243 struct btrfs_extent_data_ref *ref;
1244 struct extent_buffer *leaf;
1250 key.objectid = bytenr;
1252 key.type = BTRFS_SHARED_DATA_REF_KEY;
1253 key.offset = parent;
1255 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1256 key.offset = hash_extent_data_ref(root_objectid,
1261 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1270 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1271 key.type = BTRFS_EXTENT_REF_V0_KEY;
1272 btrfs_release_path(path);
1273 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1284 leaf = path->nodes[0];
1285 nritems = btrfs_header_nritems(leaf);
1287 if (path->slots[0] >= nritems) {
1288 ret = btrfs_next_leaf(root, path);
1294 leaf = path->nodes[0];
1295 nritems = btrfs_header_nritems(leaf);
1299 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1300 if (key.objectid != bytenr ||
1301 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1304 ref = btrfs_item_ptr(leaf, path->slots[0],
1305 struct btrfs_extent_data_ref);
1307 if (match_extent_data_ref(leaf, ref, root_objectid,
1310 btrfs_release_path(path);
1322 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1323 struct btrfs_fs_info *fs_info,
1324 struct btrfs_path *path,
1325 u64 bytenr, u64 parent,
1326 u64 root_objectid, u64 owner,
1327 u64 offset, int refs_to_add)
1329 struct btrfs_root *root = fs_info->extent_root;
1330 struct btrfs_key key;
1331 struct extent_buffer *leaf;
1336 key.objectid = bytenr;
1338 key.type = BTRFS_SHARED_DATA_REF_KEY;
1339 key.offset = parent;
1340 size = sizeof(struct btrfs_shared_data_ref);
1342 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1343 key.offset = hash_extent_data_ref(root_objectid,
1345 size = sizeof(struct btrfs_extent_data_ref);
1348 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1349 if (ret && ret != -EEXIST)
1352 leaf = path->nodes[0];
1354 struct btrfs_shared_data_ref *ref;
1355 ref = btrfs_item_ptr(leaf, path->slots[0],
1356 struct btrfs_shared_data_ref);
1358 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1360 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1361 num_refs += refs_to_add;
1362 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1365 struct btrfs_extent_data_ref *ref;
1366 while (ret == -EEXIST) {
1367 ref = btrfs_item_ptr(leaf, path->slots[0],
1368 struct btrfs_extent_data_ref);
1369 if (match_extent_data_ref(leaf, ref, root_objectid,
1372 btrfs_release_path(path);
1374 ret = btrfs_insert_empty_item(trans, root, path, &key,
1376 if (ret && ret != -EEXIST)
1379 leaf = path->nodes[0];
1381 ref = btrfs_item_ptr(leaf, path->slots[0],
1382 struct btrfs_extent_data_ref);
1384 btrfs_set_extent_data_ref_root(leaf, ref,
1386 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1387 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1388 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1390 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1391 num_refs += refs_to_add;
1392 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1395 btrfs_mark_buffer_dirty(leaf);
1398 btrfs_release_path(path);
1402 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1403 struct btrfs_fs_info *fs_info,
1404 struct btrfs_path *path,
1405 int refs_to_drop, int *last_ref)
1407 struct btrfs_key key;
1408 struct btrfs_extent_data_ref *ref1 = NULL;
1409 struct btrfs_shared_data_ref *ref2 = NULL;
1410 struct extent_buffer *leaf;
1414 leaf = path->nodes[0];
1415 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1417 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1418 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1419 struct btrfs_extent_data_ref);
1420 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1421 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1422 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1423 struct btrfs_shared_data_ref);
1424 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1425 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1426 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1427 struct btrfs_extent_ref_v0 *ref0;
1428 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1429 struct btrfs_extent_ref_v0);
1430 num_refs = btrfs_ref_count_v0(leaf, ref0);
1436 BUG_ON(num_refs < refs_to_drop);
1437 num_refs -= refs_to_drop;
1439 if (num_refs == 0) {
1440 ret = btrfs_del_item(trans, fs_info->extent_root, path);
1443 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1444 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1445 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1446 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1447 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1449 struct btrfs_extent_ref_v0 *ref0;
1450 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1451 struct btrfs_extent_ref_v0);
1452 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1455 btrfs_mark_buffer_dirty(leaf);
1460 static noinline u32 extent_data_ref_count(struct btrfs_path *path,
1461 struct btrfs_extent_inline_ref *iref)
1463 struct btrfs_key key;
1464 struct extent_buffer *leaf;
1465 struct btrfs_extent_data_ref *ref1;
1466 struct btrfs_shared_data_ref *ref2;
1470 leaf = path->nodes[0];
1471 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1474 * If type is invalid, we should have bailed out earlier than
1477 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
1478 ASSERT(type != BTRFS_REF_TYPE_INVALID);
1479 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1480 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1481 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1483 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1484 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1486 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1487 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1488 struct btrfs_extent_data_ref);
1489 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1490 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1491 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1492 struct btrfs_shared_data_ref);
1493 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1494 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1495 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1496 struct btrfs_extent_ref_v0 *ref0;
1497 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1498 struct btrfs_extent_ref_v0);
1499 num_refs = btrfs_ref_count_v0(leaf, ref0);
1507 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1508 struct btrfs_fs_info *fs_info,
1509 struct btrfs_path *path,
1510 u64 bytenr, u64 parent,
1513 struct btrfs_root *root = fs_info->extent_root;
1514 struct btrfs_key key;
1517 key.objectid = bytenr;
1519 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1520 key.offset = parent;
1522 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1523 key.offset = root_objectid;
1526 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1529 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1530 if (ret == -ENOENT && parent) {
1531 btrfs_release_path(path);
1532 key.type = BTRFS_EXTENT_REF_V0_KEY;
1533 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1541 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1542 struct btrfs_fs_info *fs_info,
1543 struct btrfs_path *path,
1544 u64 bytenr, u64 parent,
1547 struct btrfs_key key;
1550 key.objectid = bytenr;
1552 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1553 key.offset = parent;
1555 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1556 key.offset = root_objectid;
1559 ret = btrfs_insert_empty_item(trans, fs_info->extent_root,
1561 btrfs_release_path(path);
1565 static inline int extent_ref_type(u64 parent, u64 owner)
1568 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1570 type = BTRFS_SHARED_BLOCK_REF_KEY;
1572 type = BTRFS_TREE_BLOCK_REF_KEY;
1575 type = BTRFS_SHARED_DATA_REF_KEY;
1577 type = BTRFS_EXTENT_DATA_REF_KEY;
1582 static int find_next_key(struct btrfs_path *path, int level,
1583 struct btrfs_key *key)
1586 for (; level < BTRFS_MAX_LEVEL; level++) {
1587 if (!path->nodes[level])
1589 if (path->slots[level] + 1 >=
1590 btrfs_header_nritems(path->nodes[level]))
1593 btrfs_item_key_to_cpu(path->nodes[level], key,
1594 path->slots[level] + 1);
1596 btrfs_node_key_to_cpu(path->nodes[level], key,
1597 path->slots[level] + 1);
1604 * look for inline back ref. if back ref is found, *ref_ret is set
1605 * to the address of inline back ref, and 0 is returned.
1607 * if back ref isn't found, *ref_ret is set to the address where it
1608 * should be inserted, and -ENOENT is returned.
1610 * if insert is true and there are too many inline back refs, the path
1611 * points to the extent item, and -EAGAIN is returned.
1613 * NOTE: inline back refs are ordered in the same way that back ref
1614 * items in the tree are ordered.
1616 static noinline_for_stack
1617 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1618 struct btrfs_fs_info *fs_info,
1619 struct btrfs_path *path,
1620 struct btrfs_extent_inline_ref **ref_ret,
1621 u64 bytenr, u64 num_bytes,
1622 u64 parent, u64 root_objectid,
1623 u64 owner, u64 offset, int insert)
1625 struct btrfs_root *root = fs_info->extent_root;
1626 struct btrfs_key key;
1627 struct extent_buffer *leaf;
1628 struct btrfs_extent_item *ei;
1629 struct btrfs_extent_inline_ref *iref;
1639 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
1642 key.objectid = bytenr;
1643 key.type = BTRFS_EXTENT_ITEM_KEY;
1644 key.offset = num_bytes;
1646 want = extent_ref_type(parent, owner);
1648 extra_size = btrfs_extent_inline_ref_size(want);
1649 path->keep_locks = 1;
1654 * Owner is our parent level, so we can just add one to get the level
1655 * for the block we are interested in.
1657 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1658 key.type = BTRFS_METADATA_ITEM_KEY;
1663 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1670 * We may be a newly converted file system which still has the old fat
1671 * extent entries for metadata, so try and see if we have one of those.
1673 if (ret > 0 && skinny_metadata) {
1674 skinny_metadata = false;
1675 if (path->slots[0]) {
1677 btrfs_item_key_to_cpu(path->nodes[0], &key,
1679 if (key.objectid == bytenr &&
1680 key.type == BTRFS_EXTENT_ITEM_KEY &&
1681 key.offset == num_bytes)
1685 key.objectid = bytenr;
1686 key.type = BTRFS_EXTENT_ITEM_KEY;
1687 key.offset = num_bytes;
1688 btrfs_release_path(path);
1693 if (ret && !insert) {
1696 } else if (WARN_ON(ret)) {
1701 leaf = path->nodes[0];
1702 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1703 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1704 if (item_size < sizeof(*ei)) {
1709 ret = convert_extent_item_v0(trans, fs_info, path, owner,
1715 leaf = path->nodes[0];
1716 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1719 BUG_ON(item_size < sizeof(*ei));
1721 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1722 flags = btrfs_extent_flags(leaf, ei);
1724 ptr = (unsigned long)(ei + 1);
1725 end = (unsigned long)ei + item_size;
1727 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1728 ptr += sizeof(struct btrfs_tree_block_info);
1732 if (owner >= BTRFS_FIRST_FREE_OBJECTID)
1733 needed = BTRFS_REF_TYPE_DATA;
1735 needed = BTRFS_REF_TYPE_BLOCK;
1743 iref = (struct btrfs_extent_inline_ref *)ptr;
1744 type = btrfs_get_extent_inline_ref_type(leaf, iref, needed);
1745 if (type == BTRFS_REF_TYPE_INVALID) {
1753 ptr += btrfs_extent_inline_ref_size(type);
1757 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1758 struct btrfs_extent_data_ref *dref;
1759 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1760 if (match_extent_data_ref(leaf, dref, root_objectid,
1765 if (hash_extent_data_ref_item(leaf, dref) <
1766 hash_extent_data_ref(root_objectid, owner, offset))
1770 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1772 if (parent == ref_offset) {
1776 if (ref_offset < parent)
1779 if (root_objectid == ref_offset) {
1783 if (ref_offset < root_objectid)
1787 ptr += btrfs_extent_inline_ref_size(type);
1789 if (err == -ENOENT && insert) {
1790 if (item_size + extra_size >=
1791 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1796 * To add new inline back ref, we have to make sure
1797 * there is no corresponding back ref item.
1798 * For simplicity, we just do not add new inline back
1799 * ref if there is any kind of item for this block
1801 if (find_next_key(path, 0, &key) == 0 &&
1802 key.objectid == bytenr &&
1803 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1808 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1811 path->keep_locks = 0;
1812 btrfs_unlock_up_safe(path, 1);
1818 * helper to add new inline back ref
1820 static noinline_for_stack
1821 void setup_inline_extent_backref(struct btrfs_fs_info *fs_info,
1822 struct btrfs_path *path,
1823 struct btrfs_extent_inline_ref *iref,
1824 u64 parent, u64 root_objectid,
1825 u64 owner, u64 offset, int refs_to_add,
1826 struct btrfs_delayed_extent_op *extent_op)
1828 struct extent_buffer *leaf;
1829 struct btrfs_extent_item *ei;
1832 unsigned long item_offset;
1837 leaf = path->nodes[0];
1838 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1839 item_offset = (unsigned long)iref - (unsigned long)ei;
1841 type = extent_ref_type(parent, owner);
1842 size = btrfs_extent_inline_ref_size(type);
1844 btrfs_extend_item(fs_info, path, size);
1846 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1847 refs = btrfs_extent_refs(leaf, ei);
1848 refs += refs_to_add;
1849 btrfs_set_extent_refs(leaf, ei, refs);
1851 __run_delayed_extent_op(extent_op, leaf, ei);
1853 ptr = (unsigned long)ei + item_offset;
1854 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1855 if (ptr < end - size)
1856 memmove_extent_buffer(leaf, ptr + size, ptr,
1859 iref = (struct btrfs_extent_inline_ref *)ptr;
1860 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1861 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1862 struct btrfs_extent_data_ref *dref;
1863 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1864 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1865 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1866 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1867 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1868 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1869 struct btrfs_shared_data_ref *sref;
1870 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1871 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1872 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1873 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1874 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1876 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1878 btrfs_mark_buffer_dirty(leaf);
1881 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1882 struct btrfs_fs_info *fs_info,
1883 struct btrfs_path *path,
1884 struct btrfs_extent_inline_ref **ref_ret,
1885 u64 bytenr, u64 num_bytes, u64 parent,
1886 u64 root_objectid, u64 owner, u64 offset)
1890 ret = lookup_inline_extent_backref(trans, fs_info, path, ref_ret,
1891 bytenr, num_bytes, parent,
1892 root_objectid, owner, offset, 0);
1896 btrfs_release_path(path);
1899 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1900 ret = lookup_tree_block_ref(trans, fs_info, path, bytenr,
1901 parent, root_objectid);
1903 ret = lookup_extent_data_ref(trans, fs_info, path, bytenr,
1904 parent, root_objectid, owner,
1911 * helper to update/remove inline back ref
1913 static noinline_for_stack
1914 void update_inline_extent_backref(struct btrfs_fs_info *fs_info,
1915 struct btrfs_path *path,
1916 struct btrfs_extent_inline_ref *iref,
1918 struct btrfs_delayed_extent_op *extent_op,
1921 struct extent_buffer *leaf;
1922 struct btrfs_extent_item *ei;
1923 struct btrfs_extent_data_ref *dref = NULL;
1924 struct btrfs_shared_data_ref *sref = NULL;
1932 leaf = path->nodes[0];
1933 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1934 refs = btrfs_extent_refs(leaf, ei);
1935 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1936 refs += refs_to_mod;
1937 btrfs_set_extent_refs(leaf, ei, refs);
1939 __run_delayed_extent_op(extent_op, leaf, ei);
1942 * If type is invalid, we should have bailed out after
1943 * lookup_inline_extent_backref().
1945 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY);
1946 ASSERT(type != BTRFS_REF_TYPE_INVALID);
1948 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1949 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1950 refs = btrfs_extent_data_ref_count(leaf, dref);
1951 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1952 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1953 refs = btrfs_shared_data_ref_count(leaf, sref);
1956 BUG_ON(refs_to_mod != -1);
1959 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1960 refs += refs_to_mod;
1963 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1964 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1966 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1969 size = btrfs_extent_inline_ref_size(type);
1970 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1971 ptr = (unsigned long)iref;
1972 end = (unsigned long)ei + item_size;
1973 if (ptr + size < end)
1974 memmove_extent_buffer(leaf, ptr, ptr + size,
1977 btrfs_truncate_item(fs_info, path, item_size, 1);
1979 btrfs_mark_buffer_dirty(leaf);
1982 static noinline_for_stack
1983 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1984 struct btrfs_fs_info *fs_info,
1985 struct btrfs_path *path,
1986 u64 bytenr, u64 num_bytes, u64 parent,
1987 u64 root_objectid, u64 owner,
1988 u64 offset, int refs_to_add,
1989 struct btrfs_delayed_extent_op *extent_op)
1991 struct btrfs_extent_inline_ref *iref;
1994 ret = lookup_inline_extent_backref(trans, fs_info, path, &iref,
1995 bytenr, num_bytes, parent,
1996 root_objectid, owner, offset, 1);
1998 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1999 update_inline_extent_backref(fs_info, path, iref,
2000 refs_to_add, extent_op, NULL);
2001 } else if (ret == -ENOENT) {
2002 setup_inline_extent_backref(fs_info, path, iref, parent,
2003 root_objectid, owner, offset,
2004 refs_to_add, extent_op);
2010 static int insert_extent_backref(struct btrfs_trans_handle *trans,
2011 struct btrfs_fs_info *fs_info,
2012 struct btrfs_path *path,
2013 u64 bytenr, u64 parent, u64 root_objectid,
2014 u64 owner, u64 offset, int refs_to_add)
2017 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2018 BUG_ON(refs_to_add != 1);
2019 ret = insert_tree_block_ref(trans, fs_info, path, bytenr,
2020 parent, root_objectid);
2022 ret = insert_extent_data_ref(trans, fs_info, path, bytenr,
2023 parent, root_objectid,
2024 owner, offset, refs_to_add);
2029 static int remove_extent_backref(struct btrfs_trans_handle *trans,
2030 struct btrfs_fs_info *fs_info,
2031 struct btrfs_path *path,
2032 struct btrfs_extent_inline_ref *iref,
2033 int refs_to_drop, int is_data, int *last_ref)
2037 BUG_ON(!is_data && refs_to_drop != 1);
2039 update_inline_extent_backref(fs_info, path, iref,
2040 -refs_to_drop, NULL, last_ref);
2041 } else if (is_data) {
2042 ret = remove_extent_data_ref(trans, fs_info, path, refs_to_drop,
2046 ret = btrfs_del_item(trans, fs_info->extent_root, path);
2051 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
2052 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
2053 u64 *discarded_bytes)
2056 u64 bytes_left, end;
2057 u64 aligned_start = ALIGN(start, 1 << 9);
2059 if (WARN_ON(start != aligned_start)) {
2060 len -= aligned_start - start;
2061 len = round_down(len, 1 << 9);
2062 start = aligned_start;
2065 *discarded_bytes = 0;
2073 /* Skip any superblocks on this device. */
2074 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
2075 u64 sb_start = btrfs_sb_offset(j);
2076 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
2077 u64 size = sb_start - start;
2079 if (!in_range(sb_start, start, bytes_left) &&
2080 !in_range(sb_end, start, bytes_left) &&
2081 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
2085 * Superblock spans beginning of range. Adjust start and
2088 if (sb_start <= start) {
2089 start += sb_end - start;
2094 bytes_left = end - start;
2099 ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
2102 *discarded_bytes += size;
2103 else if (ret != -EOPNOTSUPP)
2112 bytes_left = end - start;
2116 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
2119 *discarded_bytes += bytes_left;
2124 int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr,
2125 u64 num_bytes, u64 *actual_bytes)
2128 u64 discarded_bytes = 0;
2129 struct btrfs_bio *bbio = NULL;
2133 * Avoid races with device replace and make sure our bbio has devices
2134 * associated to its stripes that don't go away while we are discarding.
2136 btrfs_bio_counter_inc_blocked(fs_info);
2137 /* Tell the block device(s) that the sectors can be discarded */
2138 ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, bytenr, &num_bytes,
2140 /* Error condition is -ENOMEM */
2142 struct btrfs_bio_stripe *stripe = bbio->stripes;
2146 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
2148 struct request_queue *req_q;
2150 if (!stripe->dev->bdev) {
2151 ASSERT(btrfs_test_opt(fs_info, DEGRADED));
2154 req_q = bdev_get_queue(stripe->dev->bdev);
2155 if (!blk_queue_discard(req_q))
2158 ret = btrfs_issue_discard(stripe->dev->bdev,
2163 discarded_bytes += bytes;
2164 else if (ret != -EOPNOTSUPP)
2165 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2168 * Just in case we get back EOPNOTSUPP for some reason,
2169 * just ignore the return value so we don't screw up
2170 * people calling discard_extent.
2174 btrfs_put_bbio(bbio);
2176 btrfs_bio_counter_dec(fs_info);
2179 *actual_bytes = discarded_bytes;
2182 if (ret == -EOPNOTSUPP)
2187 /* Can return -ENOMEM */
2188 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2189 struct btrfs_root *root,
2190 u64 bytenr, u64 num_bytes, u64 parent,
2191 u64 root_objectid, u64 owner, u64 offset)
2193 struct btrfs_fs_info *fs_info = root->fs_info;
2194 int old_ref_mod, new_ref_mod;
2197 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
2198 root_objectid == BTRFS_TREE_LOG_OBJECTID);
2200 btrfs_ref_tree_mod(root, bytenr, num_bytes, parent, root_objectid,
2201 owner, offset, BTRFS_ADD_DELAYED_REF);
2203 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2204 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
2206 root_objectid, (int)owner,
2207 BTRFS_ADD_DELAYED_REF, NULL,
2208 &old_ref_mod, &new_ref_mod);
2210 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
2212 root_objectid, owner, offset,
2213 0, BTRFS_ADD_DELAYED_REF,
2214 &old_ref_mod, &new_ref_mod);
2217 if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0)
2218 add_pinned_bytes(fs_info, -num_bytes, owner, root_objectid);
2223 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2224 struct btrfs_fs_info *fs_info,
2225 struct btrfs_delayed_ref_node *node,
2226 u64 parent, u64 root_objectid,
2227 u64 owner, u64 offset, int refs_to_add,
2228 struct btrfs_delayed_extent_op *extent_op)
2230 struct btrfs_path *path;
2231 struct extent_buffer *leaf;
2232 struct btrfs_extent_item *item;
2233 struct btrfs_key key;
2234 u64 bytenr = node->bytenr;
2235 u64 num_bytes = node->num_bytes;
2239 path = btrfs_alloc_path();
2243 path->reada = READA_FORWARD;
2244 path->leave_spinning = 1;
2245 /* this will setup the path even if it fails to insert the back ref */
2246 ret = insert_inline_extent_backref(trans, fs_info, path, bytenr,
2247 num_bytes, parent, root_objectid,
2249 refs_to_add, extent_op);
2250 if ((ret < 0 && ret != -EAGAIN) || !ret)
2254 * Ok we had -EAGAIN which means we didn't have space to insert and
2255 * inline extent ref, so just update the reference count and add a
2258 leaf = path->nodes[0];
2259 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2260 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2261 refs = btrfs_extent_refs(leaf, item);
2262 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2264 __run_delayed_extent_op(extent_op, leaf, item);
2266 btrfs_mark_buffer_dirty(leaf);
2267 btrfs_release_path(path);
2269 path->reada = READA_FORWARD;
2270 path->leave_spinning = 1;
2271 /* now insert the actual backref */
2272 ret = insert_extent_backref(trans, fs_info, path, bytenr, parent,
2273 root_objectid, owner, offset, refs_to_add);
2275 btrfs_abort_transaction(trans, ret);
2277 btrfs_free_path(path);
2281 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2282 struct btrfs_fs_info *fs_info,
2283 struct btrfs_delayed_ref_node *node,
2284 struct btrfs_delayed_extent_op *extent_op,
2285 int insert_reserved)
2288 struct btrfs_delayed_data_ref *ref;
2289 struct btrfs_key ins;
2294 ins.objectid = node->bytenr;
2295 ins.offset = node->num_bytes;
2296 ins.type = BTRFS_EXTENT_ITEM_KEY;
2298 ref = btrfs_delayed_node_to_data_ref(node);
2299 trace_run_delayed_data_ref(fs_info, node, ref, node->action);
2301 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2302 parent = ref->parent;
2303 ref_root = ref->root;
2305 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2307 flags |= extent_op->flags_to_set;
2308 ret = alloc_reserved_file_extent(trans, fs_info,
2309 parent, ref_root, flags,
2310 ref->objectid, ref->offset,
2311 &ins, node->ref_mod);
2312 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2313 ret = __btrfs_inc_extent_ref(trans, fs_info, node, parent,
2314 ref_root, ref->objectid,
2315 ref->offset, node->ref_mod,
2317 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2318 ret = __btrfs_free_extent(trans, fs_info, node, parent,
2319 ref_root, ref->objectid,
2320 ref->offset, node->ref_mod,
2328 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2329 struct extent_buffer *leaf,
2330 struct btrfs_extent_item *ei)
2332 u64 flags = btrfs_extent_flags(leaf, ei);
2333 if (extent_op->update_flags) {
2334 flags |= extent_op->flags_to_set;
2335 btrfs_set_extent_flags(leaf, ei, flags);
2338 if (extent_op->update_key) {
2339 struct btrfs_tree_block_info *bi;
2340 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2341 bi = (struct btrfs_tree_block_info *)(ei + 1);
2342 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2346 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2347 struct btrfs_fs_info *fs_info,
2348 struct btrfs_delayed_ref_head *head,
2349 struct btrfs_delayed_extent_op *extent_op)
2351 struct btrfs_key key;
2352 struct btrfs_path *path;
2353 struct btrfs_extent_item *ei;
2354 struct extent_buffer *leaf;
2358 int metadata = !extent_op->is_data;
2363 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2366 path = btrfs_alloc_path();
2370 key.objectid = head->bytenr;
2373 key.type = BTRFS_METADATA_ITEM_KEY;
2374 key.offset = extent_op->level;
2376 key.type = BTRFS_EXTENT_ITEM_KEY;
2377 key.offset = head->num_bytes;
2381 path->reada = READA_FORWARD;
2382 path->leave_spinning = 1;
2383 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1);
2390 if (path->slots[0] > 0) {
2392 btrfs_item_key_to_cpu(path->nodes[0], &key,
2394 if (key.objectid == head->bytenr &&
2395 key.type == BTRFS_EXTENT_ITEM_KEY &&
2396 key.offset == head->num_bytes)
2400 btrfs_release_path(path);
2403 key.objectid = head->bytenr;
2404 key.offset = head->num_bytes;
2405 key.type = BTRFS_EXTENT_ITEM_KEY;
2414 leaf = path->nodes[0];
2415 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2416 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2417 if (item_size < sizeof(*ei)) {
2418 ret = convert_extent_item_v0(trans, fs_info, path, (u64)-1, 0);
2423 leaf = path->nodes[0];
2424 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2427 BUG_ON(item_size < sizeof(*ei));
2428 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2429 __run_delayed_extent_op(extent_op, leaf, ei);
2431 btrfs_mark_buffer_dirty(leaf);
2433 btrfs_free_path(path);
2437 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2438 struct btrfs_fs_info *fs_info,
2439 struct btrfs_delayed_ref_node *node,
2440 struct btrfs_delayed_extent_op *extent_op,
2441 int insert_reserved)
2444 struct btrfs_delayed_tree_ref *ref;
2445 struct btrfs_key ins;
2448 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
2450 ref = btrfs_delayed_node_to_tree_ref(node);
2451 trace_run_delayed_tree_ref(fs_info, node, ref, node->action);
2453 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2454 parent = ref->parent;
2455 ref_root = ref->root;
2457 ins.objectid = node->bytenr;
2458 if (skinny_metadata) {
2459 ins.offset = ref->level;
2460 ins.type = BTRFS_METADATA_ITEM_KEY;
2462 ins.offset = node->num_bytes;
2463 ins.type = BTRFS_EXTENT_ITEM_KEY;
2466 if (node->ref_mod != 1) {
2468 "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu",
2469 node->bytenr, node->ref_mod, node->action, ref_root,
2473 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2474 BUG_ON(!extent_op || !extent_op->update_flags);
2475 ret = alloc_reserved_tree_block(trans, fs_info,
2477 extent_op->flags_to_set,
2480 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2481 ret = __btrfs_inc_extent_ref(trans, fs_info, node,
2485 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2486 ret = __btrfs_free_extent(trans, fs_info, node,
2488 ref->level, 0, 1, extent_op);
2495 /* helper function to actually process a single delayed ref entry */
2496 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2497 struct btrfs_fs_info *fs_info,
2498 struct btrfs_delayed_ref_node *node,
2499 struct btrfs_delayed_extent_op *extent_op,
2500 int insert_reserved)
2504 if (trans->aborted) {
2505 if (insert_reserved)
2506 btrfs_pin_extent(fs_info, node->bytenr,
2507 node->num_bytes, 1);
2511 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2512 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2513 ret = run_delayed_tree_ref(trans, fs_info, node, extent_op,
2515 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2516 node->type == BTRFS_SHARED_DATA_REF_KEY)
2517 ret = run_delayed_data_ref(trans, fs_info, node, extent_op,
2524 static inline struct btrfs_delayed_ref_node *
2525 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2527 struct btrfs_delayed_ref_node *ref;
2529 if (RB_EMPTY_ROOT(&head->ref_tree))
2533 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2534 * This is to prevent a ref count from going down to zero, which deletes
2535 * the extent item from the extent tree, when there still are references
2536 * to add, which would fail because they would not find the extent item.
2538 if (!list_empty(&head->ref_add_list))
2539 return list_first_entry(&head->ref_add_list,
2540 struct btrfs_delayed_ref_node, add_list);
2542 ref = rb_entry(rb_first(&head->ref_tree),
2543 struct btrfs_delayed_ref_node, ref_node);
2544 ASSERT(list_empty(&ref->add_list));
2548 static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
2549 struct btrfs_delayed_ref_head *head)
2551 spin_lock(&delayed_refs->lock);
2552 head->processing = 0;
2553 delayed_refs->num_heads_ready++;
2554 spin_unlock(&delayed_refs->lock);
2555 btrfs_delayed_ref_unlock(head);
2558 static int cleanup_extent_op(struct btrfs_trans_handle *trans,
2559 struct btrfs_fs_info *fs_info,
2560 struct btrfs_delayed_ref_head *head)
2562 struct btrfs_delayed_extent_op *extent_op = head->extent_op;
2567 head->extent_op = NULL;
2568 if (head->must_insert_reserved) {
2569 btrfs_free_delayed_extent_op(extent_op);
2572 spin_unlock(&head->lock);
2573 ret = run_delayed_extent_op(trans, fs_info, head, extent_op);
2574 btrfs_free_delayed_extent_op(extent_op);
2575 return ret ? ret : 1;
2578 static int cleanup_ref_head(struct btrfs_trans_handle *trans,
2579 struct btrfs_fs_info *fs_info,
2580 struct btrfs_delayed_ref_head *head)
2582 struct btrfs_delayed_ref_root *delayed_refs;
2585 delayed_refs = &trans->transaction->delayed_refs;
2587 ret = cleanup_extent_op(trans, fs_info, head);
2589 unselect_delayed_ref_head(delayed_refs, head);
2590 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2597 * Need to drop our head ref lock and re-acquire the delayed ref lock
2598 * and then re-check to make sure nobody got added.
2600 spin_unlock(&head->lock);
2601 spin_lock(&delayed_refs->lock);
2602 spin_lock(&head->lock);
2603 if (!RB_EMPTY_ROOT(&head->ref_tree) || head->extent_op) {
2604 spin_unlock(&head->lock);
2605 spin_unlock(&delayed_refs->lock);
2608 delayed_refs->num_heads--;
2609 rb_erase(&head->href_node, &delayed_refs->href_root);
2610 RB_CLEAR_NODE(&head->href_node);
2611 spin_unlock(&delayed_refs->lock);
2612 spin_unlock(&head->lock);
2613 atomic_dec(&delayed_refs->num_entries);
2615 trace_run_delayed_ref_head(fs_info, head, 0);
2617 if (head->total_ref_mod < 0) {
2618 struct btrfs_block_group_cache *cache;
2620 cache = btrfs_lookup_block_group(fs_info, head->bytenr);
2622 percpu_counter_add(&cache->space_info->total_bytes_pinned,
2624 btrfs_put_block_group(cache);
2626 if (head->is_data) {
2627 spin_lock(&delayed_refs->lock);
2628 delayed_refs->pending_csums -= head->num_bytes;
2629 spin_unlock(&delayed_refs->lock);
2633 if (head->must_insert_reserved) {
2634 btrfs_pin_extent(fs_info, head->bytenr,
2635 head->num_bytes, 1);
2636 if (head->is_data) {
2637 ret = btrfs_del_csums(trans, fs_info, head->bytenr,
2642 /* Also free its reserved qgroup space */
2643 btrfs_qgroup_free_delayed_ref(fs_info, head->qgroup_ref_root,
2644 head->qgroup_reserved);
2645 btrfs_delayed_ref_unlock(head);
2646 btrfs_put_delayed_ref_head(head);
2651 * Returns 0 on success or if called with an already aborted transaction.
2652 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2654 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2655 struct btrfs_fs_info *fs_info,
2658 struct btrfs_delayed_ref_root *delayed_refs;
2659 struct btrfs_delayed_ref_node *ref;
2660 struct btrfs_delayed_ref_head *locked_ref = NULL;
2661 struct btrfs_delayed_extent_op *extent_op;
2662 ktime_t start = ktime_get();
2664 unsigned long count = 0;
2665 unsigned long actual_count = 0;
2666 int must_insert_reserved = 0;
2668 delayed_refs = &trans->transaction->delayed_refs;
2674 spin_lock(&delayed_refs->lock);
2675 locked_ref = btrfs_select_ref_head(trans);
2677 spin_unlock(&delayed_refs->lock);
2681 /* grab the lock that says we are going to process
2682 * all the refs for this head */
2683 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2684 spin_unlock(&delayed_refs->lock);
2686 * we may have dropped the spin lock to get the head
2687 * mutex lock, and that might have given someone else
2688 * time to free the head. If that's true, it has been
2689 * removed from our list and we can move on.
2691 if (ret == -EAGAIN) {
2699 * We need to try and merge add/drops of the same ref since we
2700 * can run into issues with relocate dropping the implicit ref
2701 * and then it being added back again before the drop can
2702 * finish. If we merged anything we need to re-loop so we can
2704 * Or we can get node references of the same type that weren't
2705 * merged when created due to bumps in the tree mod seq, and
2706 * we need to merge them to prevent adding an inline extent
2707 * backref before dropping it (triggering a BUG_ON at
2708 * insert_inline_extent_backref()).
2710 spin_lock(&locked_ref->lock);
2711 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2715 * locked_ref is the head node, so we have to go one
2716 * node back for any delayed ref updates
2718 ref = select_delayed_ref(locked_ref);
2720 if (ref && ref->seq &&
2721 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2722 spin_unlock(&locked_ref->lock);
2723 unselect_delayed_ref_head(delayed_refs, locked_ref);
2731 * We're done processing refs in this ref_head, clean everything
2732 * up and move on to the next ref_head.
2735 ret = cleanup_ref_head(trans, fs_info, locked_ref);
2737 /* We dropped our lock, we need to loop. */
2750 rb_erase(&ref->ref_node, &locked_ref->ref_tree);
2751 RB_CLEAR_NODE(&ref->ref_node);
2752 if (!list_empty(&ref->add_list))
2753 list_del(&ref->add_list);
2755 * When we play the delayed ref, also correct the ref_mod on
2758 switch (ref->action) {
2759 case BTRFS_ADD_DELAYED_REF:
2760 case BTRFS_ADD_DELAYED_EXTENT:
2761 locked_ref->ref_mod -= ref->ref_mod;
2763 case BTRFS_DROP_DELAYED_REF:
2764 locked_ref->ref_mod += ref->ref_mod;
2769 atomic_dec(&delayed_refs->num_entries);
2772 * Record the must-insert_reserved flag before we drop the spin
2775 must_insert_reserved = locked_ref->must_insert_reserved;
2776 locked_ref->must_insert_reserved = 0;
2778 extent_op = locked_ref->extent_op;
2779 locked_ref->extent_op = NULL;
2780 spin_unlock(&locked_ref->lock);
2782 ret = run_one_delayed_ref(trans, fs_info, ref, extent_op,
2783 must_insert_reserved);
2785 btrfs_free_delayed_extent_op(extent_op);
2787 unselect_delayed_ref_head(delayed_refs, locked_ref);
2788 btrfs_put_delayed_ref(ref);
2789 btrfs_debug(fs_info, "run_one_delayed_ref returned %d",
2794 btrfs_put_delayed_ref(ref);
2800 * We don't want to include ref heads since we can have empty ref heads
2801 * and those will drastically skew our runtime down since we just do
2802 * accounting, no actual extent tree updates.
2804 if (actual_count > 0) {
2805 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2809 * We weigh the current average higher than our current runtime
2810 * to avoid large swings in the average.
2812 spin_lock(&delayed_refs->lock);
2813 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2814 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2815 spin_unlock(&delayed_refs->lock);
2820 #ifdef SCRAMBLE_DELAYED_REFS
2822 * Normally delayed refs get processed in ascending bytenr order. This
2823 * correlates in most cases to the order added. To expose dependencies on this
2824 * order, we start to process the tree in the middle instead of the beginning
2826 static u64 find_middle(struct rb_root *root)
2828 struct rb_node *n = root->rb_node;
2829 struct btrfs_delayed_ref_node *entry;
2832 u64 first = 0, last = 0;
2836 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2837 first = entry->bytenr;
2841 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2842 last = entry->bytenr;
2847 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2848 WARN_ON(!entry->in_tree);
2850 middle = entry->bytenr;
2863 static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads)
2867 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2868 sizeof(struct btrfs_extent_inline_ref));
2869 if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2870 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2873 * We don't ever fill up leaves all the way so multiply by 2 just to be
2874 * closer to what we're really going to want to use.
2876 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info));
2880 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2881 * would require to store the csums for that many bytes.
2883 u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes)
2886 u64 num_csums_per_leaf;
2889 csum_size = BTRFS_MAX_ITEM_SIZE(fs_info);
2890 num_csums_per_leaf = div64_u64(csum_size,
2891 (u64)btrfs_super_csum_size(fs_info->super_copy));
2892 num_csums = div64_u64(csum_bytes, fs_info->sectorsize);
2893 num_csums += num_csums_per_leaf - 1;
2894 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2898 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2899 struct btrfs_fs_info *fs_info)
2901 struct btrfs_block_rsv *global_rsv;
2902 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2903 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2904 unsigned int num_dirty_bgs = trans->transaction->num_dirty_bgs;
2905 u64 num_bytes, num_dirty_bgs_bytes;
2908 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
2909 num_heads = heads_to_leaves(fs_info, num_heads);
2911 num_bytes += (num_heads - 1) * fs_info->nodesize;
2913 num_bytes += btrfs_csum_bytes_to_leaves(fs_info, csum_bytes) *
2915 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(fs_info,
2917 global_rsv = &fs_info->global_block_rsv;
2920 * If we can't allocate any more chunks lets make sure we have _lots_ of
2921 * wiggle room since running delayed refs can create more delayed refs.
2923 if (global_rsv->space_info->full) {
2924 num_dirty_bgs_bytes <<= 1;
2928 spin_lock(&global_rsv->lock);
2929 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2931 spin_unlock(&global_rsv->lock);
2935 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2936 struct btrfs_fs_info *fs_info)
2939 atomic_read(&trans->transaction->delayed_refs.num_entries);
2944 avg_runtime = fs_info->avg_delayed_ref_runtime;
2945 val = num_entries * avg_runtime;
2946 if (val >= NSEC_PER_SEC)
2948 if (val >= NSEC_PER_SEC / 2)
2951 return btrfs_check_space_for_delayed_refs(trans, fs_info);
2954 struct async_delayed_refs {
2955 struct btrfs_root *root;
2960 struct completion wait;
2961 struct btrfs_work work;
2964 static inline struct async_delayed_refs *
2965 to_async_delayed_refs(struct btrfs_work *work)
2967 return container_of(work, struct async_delayed_refs, work);
2970 static void delayed_ref_async_start(struct btrfs_work *work)
2972 struct async_delayed_refs *async = to_async_delayed_refs(work);
2973 struct btrfs_trans_handle *trans;
2974 struct btrfs_fs_info *fs_info = async->root->fs_info;
2977 /* if the commit is already started, we don't need to wait here */
2978 if (btrfs_transaction_blocked(fs_info))
2981 trans = btrfs_join_transaction(async->root);
2982 if (IS_ERR(trans)) {
2983 async->error = PTR_ERR(trans);
2988 * trans->sync means that when we call end_transaction, we won't
2989 * wait on delayed refs
2993 /* Don't bother flushing if we got into a different transaction */
2994 if (trans->transid > async->transid)
2997 ret = btrfs_run_delayed_refs(trans, fs_info, async->count);
3001 ret = btrfs_end_transaction(trans);
3002 if (ret && !async->error)
3006 complete(&async->wait);
3011 int btrfs_async_run_delayed_refs(struct btrfs_fs_info *fs_info,
3012 unsigned long count, u64 transid, int wait)
3014 struct async_delayed_refs *async;
3017 async = kmalloc(sizeof(*async), GFP_NOFS);
3021 async->root = fs_info->tree_root;
3022 async->count = count;
3024 async->transid = transid;
3029 init_completion(&async->wait);
3031 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
3032 delayed_ref_async_start, NULL, NULL);
3034 btrfs_queue_work(fs_info->extent_workers, &async->work);
3037 wait_for_completion(&async->wait);
3046 * this starts processing the delayed reference count updates and
3047 * extent insertions we have queued up so far. count can be
3048 * 0, which means to process everything in the tree at the start
3049 * of the run (but not newly added entries), or it can be some target
3050 * number you'd like to process.
3052 * Returns 0 on success or if called with an aborted transaction
3053 * Returns <0 on error and aborts the transaction
3055 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
3056 struct btrfs_fs_info *fs_info, unsigned long count)
3058 struct rb_node *node;
3059 struct btrfs_delayed_ref_root *delayed_refs;
3060 struct btrfs_delayed_ref_head *head;
3062 int run_all = count == (unsigned long)-1;
3063 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
3065 /* We'll clean this up in btrfs_cleanup_transaction */
3069 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
3072 delayed_refs = &trans->transaction->delayed_refs;
3074 count = atomic_read(&delayed_refs->num_entries) * 2;
3077 #ifdef SCRAMBLE_DELAYED_REFS
3078 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
3080 trans->can_flush_pending_bgs = false;
3081 ret = __btrfs_run_delayed_refs(trans, fs_info, count);
3083 btrfs_abort_transaction(trans, ret);
3088 if (!list_empty(&trans->new_bgs))
3089 btrfs_create_pending_block_groups(trans);
3091 spin_lock(&delayed_refs->lock);
3092 node = rb_first(&delayed_refs->href_root);
3094 spin_unlock(&delayed_refs->lock);
3097 head = rb_entry(node, struct btrfs_delayed_ref_head,
3099 refcount_inc(&head->refs);
3100 spin_unlock(&delayed_refs->lock);
3102 /* Mutex was contended, block until it's released and retry. */
3103 mutex_lock(&head->mutex);
3104 mutex_unlock(&head->mutex);
3106 btrfs_put_delayed_ref_head(head);
3111 trans->can_flush_pending_bgs = can_flush_pending_bgs;
3115 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
3116 struct btrfs_fs_info *fs_info,
3117 u64 bytenr, u64 num_bytes, u64 flags,
3118 int level, int is_data)
3120 struct btrfs_delayed_extent_op *extent_op;
3123 extent_op = btrfs_alloc_delayed_extent_op();
3127 extent_op->flags_to_set = flags;
3128 extent_op->update_flags = true;
3129 extent_op->update_key = false;
3130 extent_op->is_data = is_data ? true : false;
3131 extent_op->level = level;
3133 ret = btrfs_add_delayed_extent_op(fs_info, trans, bytenr,
3134 num_bytes, extent_op);
3136 btrfs_free_delayed_extent_op(extent_op);
3140 static noinline int check_delayed_ref(struct btrfs_root *root,
3141 struct btrfs_path *path,
3142 u64 objectid, u64 offset, u64 bytenr)
3144 struct btrfs_delayed_ref_head *head;
3145 struct btrfs_delayed_ref_node *ref;
3146 struct btrfs_delayed_data_ref *data_ref;
3147 struct btrfs_delayed_ref_root *delayed_refs;
3148 struct btrfs_transaction *cur_trans;
3149 struct rb_node *node;
3152 cur_trans = root->fs_info->running_transaction;
3156 delayed_refs = &cur_trans->delayed_refs;
3157 spin_lock(&delayed_refs->lock);
3158 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
3160 spin_unlock(&delayed_refs->lock);
3164 if (!mutex_trylock(&head->mutex)) {
3165 refcount_inc(&head->refs);
3166 spin_unlock(&delayed_refs->lock);
3168 btrfs_release_path(path);
3171 * Mutex was contended, block until it's released and let
3174 mutex_lock(&head->mutex);
3175 mutex_unlock(&head->mutex);
3176 btrfs_put_delayed_ref_head(head);
3179 spin_unlock(&delayed_refs->lock);
3181 spin_lock(&head->lock);
3183 * XXX: We should replace this with a proper search function in the
3186 for (node = rb_first(&head->ref_tree); node; node = rb_next(node)) {
3187 ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node);
3188 /* If it's a shared ref we know a cross reference exists */
3189 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
3194 data_ref = btrfs_delayed_node_to_data_ref(ref);
3197 * If our ref doesn't match the one we're currently looking at
3198 * then we have a cross reference.
3200 if (data_ref->root != root->root_key.objectid ||
3201 data_ref->objectid != objectid ||
3202 data_ref->offset != offset) {
3207 spin_unlock(&head->lock);
3208 mutex_unlock(&head->mutex);
3212 static noinline int check_committed_ref(struct btrfs_root *root,
3213 struct btrfs_path *path,
3214 u64 objectid, u64 offset, u64 bytenr)
3216 struct btrfs_fs_info *fs_info = root->fs_info;
3217 struct btrfs_root *extent_root = fs_info->extent_root;
3218 struct extent_buffer *leaf;
3219 struct btrfs_extent_data_ref *ref;
3220 struct btrfs_extent_inline_ref *iref;
3221 struct btrfs_extent_item *ei;
3222 struct btrfs_key key;
3227 key.objectid = bytenr;
3228 key.offset = (u64)-1;
3229 key.type = BTRFS_EXTENT_ITEM_KEY;
3231 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3234 BUG_ON(ret == 0); /* Corruption */
3237 if (path->slots[0] == 0)
3241 leaf = path->nodes[0];
3242 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3244 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3248 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3249 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3250 if (item_size < sizeof(*ei)) {
3251 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3255 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3257 if (item_size != sizeof(*ei) +
3258 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3261 if (btrfs_extent_generation(leaf, ei) <=
3262 btrfs_root_last_snapshot(&root->root_item))
3265 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3267 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
3268 if (type != BTRFS_EXTENT_DATA_REF_KEY)
3271 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3272 if (btrfs_extent_refs(leaf, ei) !=
3273 btrfs_extent_data_ref_count(leaf, ref) ||
3274 btrfs_extent_data_ref_root(leaf, ref) !=
3275 root->root_key.objectid ||
3276 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3277 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3285 int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
3288 struct btrfs_path *path;
3292 path = btrfs_alloc_path();
3297 ret = check_committed_ref(root, path, objectid,
3299 if (ret && ret != -ENOENT)
3302 ret2 = check_delayed_ref(root, path, objectid,
3304 } while (ret2 == -EAGAIN);
3306 if (ret2 && ret2 != -ENOENT) {
3311 if (ret != -ENOENT || ret2 != -ENOENT)
3314 btrfs_free_path(path);
3315 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3320 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3321 struct btrfs_root *root,
3322 struct extent_buffer *buf,
3323 int full_backref, int inc)
3325 struct btrfs_fs_info *fs_info = root->fs_info;
3331 struct btrfs_key key;
3332 struct btrfs_file_extent_item *fi;
3336 int (*process_func)(struct btrfs_trans_handle *,
3337 struct btrfs_root *,
3338 u64, u64, u64, u64, u64, u64);
3341 if (btrfs_is_testing(fs_info))
3344 ref_root = btrfs_header_owner(buf);
3345 nritems = btrfs_header_nritems(buf);
3346 level = btrfs_header_level(buf);
3348 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3352 process_func = btrfs_inc_extent_ref;
3354 process_func = btrfs_free_extent;
3357 parent = buf->start;
3361 for (i = 0; i < nritems; i++) {
3363 btrfs_item_key_to_cpu(buf, &key, i);
3364 if (key.type != BTRFS_EXTENT_DATA_KEY)
3366 fi = btrfs_item_ptr(buf, i,
3367 struct btrfs_file_extent_item);
3368 if (btrfs_file_extent_type(buf, fi) ==
3369 BTRFS_FILE_EXTENT_INLINE)
3371 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3375 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3376 key.offset -= btrfs_file_extent_offset(buf, fi);
3377 ret = process_func(trans, root, bytenr, num_bytes,
3378 parent, ref_root, key.objectid,
3383 bytenr = btrfs_node_blockptr(buf, i);
3384 num_bytes = fs_info->nodesize;
3385 ret = process_func(trans, root, bytenr, num_bytes,
3386 parent, ref_root, level - 1, 0);
3396 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3397 struct extent_buffer *buf, int full_backref)
3399 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3402 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3403 struct extent_buffer *buf, int full_backref)
3405 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3408 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3409 struct btrfs_fs_info *fs_info,
3410 struct btrfs_path *path,
3411 struct btrfs_block_group_cache *cache)
3414 struct btrfs_root *extent_root = fs_info->extent_root;
3416 struct extent_buffer *leaf;
3418 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3425 leaf = path->nodes[0];
3426 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3427 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3428 btrfs_mark_buffer_dirty(leaf);
3430 btrfs_release_path(path);
3435 static struct btrfs_block_group_cache *
3436 next_block_group(struct btrfs_fs_info *fs_info,
3437 struct btrfs_block_group_cache *cache)
3439 struct rb_node *node;
3441 spin_lock(&fs_info->block_group_cache_lock);
3443 /* If our block group was removed, we need a full search. */
3444 if (RB_EMPTY_NODE(&cache->cache_node)) {
3445 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3447 spin_unlock(&fs_info->block_group_cache_lock);
3448 btrfs_put_block_group(cache);
3449 cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache;
3451 node = rb_next(&cache->cache_node);
3452 btrfs_put_block_group(cache);
3454 cache = rb_entry(node, struct btrfs_block_group_cache,
3456 btrfs_get_block_group(cache);
3459 spin_unlock(&fs_info->block_group_cache_lock);
3463 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3464 struct btrfs_trans_handle *trans,
3465 struct btrfs_path *path)
3467 struct btrfs_fs_info *fs_info = block_group->fs_info;
3468 struct btrfs_root *root = fs_info->tree_root;
3469 struct inode *inode = NULL;
3470 struct extent_changeset *data_reserved = NULL;
3472 int dcs = BTRFS_DC_ERROR;
3478 * If this block group is smaller than 100 megs don't bother caching the
3481 if (block_group->key.offset < (100 * SZ_1M)) {
3482 spin_lock(&block_group->lock);
3483 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3484 spin_unlock(&block_group->lock);
3491 inode = lookup_free_space_inode(fs_info, block_group, path);
3492 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3493 ret = PTR_ERR(inode);
3494 btrfs_release_path(path);
3498 if (IS_ERR(inode)) {
3502 if (block_group->ro)
3505 ret = create_free_space_inode(fs_info, trans, block_group,
3513 * We want to set the generation to 0, that way if anything goes wrong
3514 * from here on out we know not to trust this cache when we load up next
3517 BTRFS_I(inode)->generation = 0;
3518 ret = btrfs_update_inode(trans, root, inode);
3521 * So theoretically we could recover from this, simply set the
3522 * super cache generation to 0 so we know to invalidate the
3523 * cache, but then we'd have to keep track of the block groups
3524 * that fail this way so we know we _have_ to reset this cache
3525 * before the next commit or risk reading stale cache. So to
3526 * limit our exposure to horrible edge cases lets just abort the
3527 * transaction, this only happens in really bad situations
3530 btrfs_abort_transaction(trans, ret);
3535 /* We've already setup this transaction, go ahead and exit */
3536 if (block_group->cache_generation == trans->transid &&
3537 i_size_read(inode)) {
3538 dcs = BTRFS_DC_SETUP;
3542 if (i_size_read(inode) > 0) {
3543 ret = btrfs_check_trunc_cache_free_space(fs_info,
3544 &fs_info->global_block_rsv);
3548 ret = btrfs_truncate_free_space_cache(trans, NULL, inode);
3553 spin_lock(&block_group->lock);
3554 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3555 !btrfs_test_opt(fs_info, SPACE_CACHE)) {
3557 * don't bother trying to write stuff out _if_
3558 * a) we're not cached,
3559 * b) we're with nospace_cache mount option,
3560 * c) we're with v2 space_cache (FREE_SPACE_TREE).
3562 dcs = BTRFS_DC_WRITTEN;
3563 spin_unlock(&block_group->lock);
3566 spin_unlock(&block_group->lock);
3569 * We hit an ENOSPC when setting up the cache in this transaction, just
3570 * skip doing the setup, we've already cleared the cache so we're safe.
3572 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
3578 * Try to preallocate enough space based on how big the block group is.
3579 * Keep in mind this has to include any pinned space which could end up
3580 * taking up quite a bit since it's not folded into the other space
3583 num_pages = div_u64(block_group->key.offset, SZ_256M);
3588 num_pages *= PAGE_SIZE;
3590 ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages);
3594 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3595 num_pages, num_pages,
3598 * Our cache requires contiguous chunks so that we don't modify a bunch
3599 * of metadata or split extents when writing the cache out, which means
3600 * we can enospc if we are heavily fragmented in addition to just normal
3601 * out of space conditions. So if we hit this just skip setting up any
3602 * other block groups for this transaction, maybe we'll unpin enough
3603 * space the next time around.
3606 dcs = BTRFS_DC_SETUP;
3607 else if (ret == -ENOSPC)
3608 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
3613 btrfs_release_path(path);
3615 spin_lock(&block_group->lock);
3616 if (!ret && dcs == BTRFS_DC_SETUP)
3617 block_group->cache_generation = trans->transid;
3618 block_group->disk_cache_state = dcs;
3619 spin_unlock(&block_group->lock);
3621 extent_changeset_free(data_reserved);
3625 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3626 struct btrfs_fs_info *fs_info)
3628 struct btrfs_block_group_cache *cache, *tmp;
3629 struct btrfs_transaction *cur_trans = trans->transaction;
3630 struct btrfs_path *path;
3632 if (list_empty(&cur_trans->dirty_bgs) ||
3633 !btrfs_test_opt(fs_info, SPACE_CACHE))
3636 path = btrfs_alloc_path();
3640 /* Could add new block groups, use _safe just in case */
3641 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3643 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3644 cache_save_setup(cache, trans, path);
3647 btrfs_free_path(path);
3652 * transaction commit does final block group cache writeback during a
3653 * critical section where nothing is allowed to change the FS. This is
3654 * required in order for the cache to actually match the block group,
3655 * but can introduce a lot of latency into the commit.
3657 * So, btrfs_start_dirty_block_groups is here to kick off block group
3658 * cache IO. There's a chance we'll have to redo some of it if the
3659 * block group changes again during the commit, but it greatly reduces
3660 * the commit latency by getting rid of the easy block groups while
3661 * we're still allowing others to join the commit.
3663 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans)
3665 struct btrfs_fs_info *fs_info = trans->fs_info;
3666 struct btrfs_block_group_cache *cache;
3667 struct btrfs_transaction *cur_trans = trans->transaction;
3670 struct btrfs_path *path = NULL;
3672 struct list_head *io = &cur_trans->io_bgs;
3673 int num_started = 0;
3676 spin_lock(&cur_trans->dirty_bgs_lock);
3677 if (list_empty(&cur_trans->dirty_bgs)) {
3678 spin_unlock(&cur_trans->dirty_bgs_lock);
3681 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3682 spin_unlock(&cur_trans->dirty_bgs_lock);
3686 * make sure all the block groups on our dirty list actually
3689 btrfs_create_pending_block_groups(trans);
3692 path = btrfs_alloc_path();
3698 * cache_write_mutex is here only to save us from balance or automatic
3699 * removal of empty block groups deleting this block group while we are
3700 * writing out the cache
3702 mutex_lock(&trans->transaction->cache_write_mutex);
3703 while (!list_empty(&dirty)) {
3704 cache = list_first_entry(&dirty,
3705 struct btrfs_block_group_cache,
3708 * this can happen if something re-dirties a block
3709 * group that is already under IO. Just wait for it to
3710 * finish and then do it all again
3712 if (!list_empty(&cache->io_list)) {
3713 list_del_init(&cache->io_list);
3714 btrfs_wait_cache_io(trans, cache, path);
3715 btrfs_put_block_group(cache);
3720 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3721 * if it should update the cache_state. Don't delete
3722 * until after we wait.
3724 * Since we're not running in the commit critical section
3725 * we need the dirty_bgs_lock to protect from update_block_group
3727 spin_lock(&cur_trans->dirty_bgs_lock);
3728 list_del_init(&cache->dirty_list);
3729 spin_unlock(&cur_trans->dirty_bgs_lock);
3733 cache_save_setup(cache, trans, path);
3735 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3736 cache->io_ctl.inode = NULL;
3737 ret = btrfs_write_out_cache(fs_info, trans,
3739 if (ret == 0 && cache->io_ctl.inode) {
3744 * The cache_write_mutex is protecting the
3745 * io_list, also refer to the definition of
3746 * btrfs_transaction::io_bgs for more details
3748 list_add_tail(&cache->io_list, io);
3751 * if we failed to write the cache, the
3752 * generation will be bad and life goes on
3758 ret = write_one_cache_group(trans, fs_info,
3761 * Our block group might still be attached to the list
3762 * of new block groups in the transaction handle of some
3763 * other task (struct btrfs_trans_handle->new_bgs). This
3764 * means its block group item isn't yet in the extent
3765 * tree. If this happens ignore the error, as we will
3766 * try again later in the critical section of the
3767 * transaction commit.
3769 if (ret == -ENOENT) {
3771 spin_lock(&cur_trans->dirty_bgs_lock);
3772 if (list_empty(&cache->dirty_list)) {
3773 list_add_tail(&cache->dirty_list,
3774 &cur_trans->dirty_bgs);
3775 btrfs_get_block_group(cache);
3777 spin_unlock(&cur_trans->dirty_bgs_lock);
3779 btrfs_abort_transaction(trans, ret);
3783 /* if its not on the io list, we need to put the block group */
3785 btrfs_put_block_group(cache);
3791 * Avoid blocking other tasks for too long. It might even save
3792 * us from writing caches for block groups that are going to be
3795 mutex_unlock(&trans->transaction->cache_write_mutex);
3796 mutex_lock(&trans->transaction->cache_write_mutex);
3798 mutex_unlock(&trans->transaction->cache_write_mutex);
3801 * go through delayed refs for all the stuff we've just kicked off
3802 * and then loop back (just once)
3804 ret = btrfs_run_delayed_refs(trans, fs_info, 0);
3805 if (!ret && loops == 0) {
3807 spin_lock(&cur_trans->dirty_bgs_lock);
3808 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3810 * dirty_bgs_lock protects us from concurrent block group
3811 * deletes too (not just cache_write_mutex).
3813 if (!list_empty(&dirty)) {
3814 spin_unlock(&cur_trans->dirty_bgs_lock);
3817 spin_unlock(&cur_trans->dirty_bgs_lock);
3818 } else if (ret < 0) {
3819 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
3822 btrfs_free_path(path);
3826 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3827 struct btrfs_fs_info *fs_info)
3829 struct btrfs_block_group_cache *cache;
3830 struct btrfs_transaction *cur_trans = trans->transaction;
3833 struct btrfs_path *path;
3834 struct list_head *io = &cur_trans->io_bgs;
3835 int num_started = 0;
3837 path = btrfs_alloc_path();
3842 * Even though we are in the critical section of the transaction commit,
3843 * we can still have concurrent tasks adding elements to this
3844 * transaction's list of dirty block groups. These tasks correspond to
3845 * endio free space workers started when writeback finishes for a
3846 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3847 * allocate new block groups as a result of COWing nodes of the root
3848 * tree when updating the free space inode. The writeback for the space
3849 * caches is triggered by an earlier call to
3850 * btrfs_start_dirty_block_groups() and iterations of the following
3852 * Also we want to do the cache_save_setup first and then run the
3853 * delayed refs to make sure we have the best chance at doing this all
3856 spin_lock(&cur_trans->dirty_bgs_lock);
3857 while (!list_empty(&cur_trans->dirty_bgs)) {
3858 cache = list_first_entry(&cur_trans->dirty_bgs,
3859 struct btrfs_block_group_cache,
3863 * this can happen if cache_save_setup re-dirties a block
3864 * group that is already under IO. Just wait for it to
3865 * finish and then do it all again
3867 if (!list_empty(&cache->io_list)) {
3868 spin_unlock(&cur_trans->dirty_bgs_lock);
3869 list_del_init(&cache->io_list);
3870 btrfs_wait_cache_io(trans, cache, path);
3871 btrfs_put_block_group(cache);
3872 spin_lock(&cur_trans->dirty_bgs_lock);
3876 * don't remove from the dirty list until after we've waited
3879 list_del_init(&cache->dirty_list);
3880 spin_unlock(&cur_trans->dirty_bgs_lock);
3883 cache_save_setup(cache, trans, path);
3886 ret = btrfs_run_delayed_refs(trans, fs_info,
3887 (unsigned long) -1);
3889 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3890 cache->io_ctl.inode = NULL;
3891 ret = btrfs_write_out_cache(fs_info, trans,
3893 if (ret == 0 && cache->io_ctl.inode) {
3896 list_add_tail(&cache->io_list, io);
3899 * if we failed to write the cache, the
3900 * generation will be bad and life goes on
3906 ret = write_one_cache_group(trans, fs_info,
3909 * One of the free space endio workers might have
3910 * created a new block group while updating a free space
3911 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3912 * and hasn't released its transaction handle yet, in
3913 * which case the new block group is still attached to
3914 * its transaction handle and its creation has not
3915 * finished yet (no block group item in the extent tree
3916 * yet, etc). If this is the case, wait for all free
3917 * space endio workers to finish and retry. This is a
3918 * a very rare case so no need for a more efficient and
3921 if (ret == -ENOENT) {
3922 wait_event(cur_trans->writer_wait,
3923 atomic_read(&cur_trans->num_writers) == 1);
3924 ret = write_one_cache_group(trans, fs_info,
3928 btrfs_abort_transaction(trans, ret);
3931 /* if its not on the io list, we need to put the block group */
3933 btrfs_put_block_group(cache);
3934 spin_lock(&cur_trans->dirty_bgs_lock);
3936 spin_unlock(&cur_trans->dirty_bgs_lock);
3939 * Refer to the definition of io_bgs member for details why it's safe
3940 * to use it without any locking
3942 while (!list_empty(io)) {
3943 cache = list_first_entry(io, struct btrfs_block_group_cache,
3945 list_del_init(&cache->io_list);
3946 btrfs_wait_cache_io(trans, cache, path);
3947 btrfs_put_block_group(cache);
3950 btrfs_free_path(path);
3954 int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
3956 struct btrfs_block_group_cache *block_group;
3959 block_group = btrfs_lookup_block_group(fs_info, bytenr);
3960 if (!block_group || block_group->ro)
3963 btrfs_put_block_group(block_group);
3967 bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3969 struct btrfs_block_group_cache *bg;
3972 bg = btrfs_lookup_block_group(fs_info, bytenr);
3976 spin_lock(&bg->lock);
3980 atomic_inc(&bg->nocow_writers);
3981 spin_unlock(&bg->lock);
3983 /* no put on block group, done by btrfs_dec_nocow_writers */
3985 btrfs_put_block_group(bg);
3991 void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3993 struct btrfs_block_group_cache *bg;
3995 bg = btrfs_lookup_block_group(fs_info, bytenr);
3997 if (atomic_dec_and_test(&bg->nocow_writers))
3998 wake_up_atomic_t(&bg->nocow_writers);
4000 * Once for our lookup and once for the lookup done by a previous call
4001 * to btrfs_inc_nocow_writers()
4003 btrfs_put_block_group(bg);
4004 btrfs_put_block_group(bg);
4007 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg)
4009 wait_on_atomic_t(&bg->nocow_writers, atomic_t_wait,
4010 TASK_UNINTERRUPTIBLE);
4013 static const char *alloc_name(u64 flags)
4016 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
4018 case BTRFS_BLOCK_GROUP_METADATA:
4020 case BTRFS_BLOCK_GROUP_DATA:
4022 case BTRFS_BLOCK_GROUP_SYSTEM:
4026 return "invalid-combination";
4030 static int create_space_info(struct btrfs_fs_info *info, u64 flags,
4031 struct btrfs_space_info **new)
4034 struct btrfs_space_info *space_info;
4038 space_info = kzalloc(sizeof(*space_info), GFP_NOFS);
4042 ret = percpu_counter_init(&space_info->total_bytes_pinned, 0,
4049 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
4050 INIT_LIST_HEAD(&space_info->block_groups[i]);
4051 init_rwsem(&space_info->groups_sem);
4052 spin_lock_init(&space_info->lock);
4053 space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
4054 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4055 init_waitqueue_head(&space_info->wait);
4056 INIT_LIST_HEAD(&space_info->ro_bgs);
4057 INIT_LIST_HEAD(&space_info->tickets);
4058 INIT_LIST_HEAD(&space_info->priority_tickets);
4060 ret = kobject_init_and_add(&space_info->kobj, &space_info_ktype,
4061 info->space_info_kobj, "%s",
4062 alloc_name(space_info->flags));
4064 percpu_counter_destroy(&space_info->total_bytes_pinned);
4070 list_add_rcu(&space_info->list, &info->space_info);
4071 if (flags & BTRFS_BLOCK_GROUP_DATA)
4072 info->data_sinfo = space_info;
4077 static void update_space_info(struct btrfs_fs_info *info, u64 flags,
4078 u64 total_bytes, u64 bytes_used,
4080 struct btrfs_space_info **space_info)
4082 struct btrfs_space_info *found;
4085 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
4086 BTRFS_BLOCK_GROUP_RAID10))
4091 found = __find_space_info(info, flags);
4093 spin_lock(&found->lock);
4094 found->total_bytes += total_bytes;
4095 found->disk_total += total_bytes * factor;
4096 found->bytes_used += bytes_used;
4097 found->disk_used += bytes_used * factor;
4098 found->bytes_readonly += bytes_readonly;
4099 if (total_bytes > 0)
4101 space_info_add_new_bytes(info, found, total_bytes -
4102 bytes_used - bytes_readonly);
4103 spin_unlock(&found->lock);
4104 *space_info = found;
4107 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
4109 u64 extra_flags = chunk_to_extended(flags) &
4110 BTRFS_EXTENDED_PROFILE_MASK;
4112 write_seqlock(&fs_info->profiles_lock);
4113 if (flags & BTRFS_BLOCK_GROUP_DATA)
4114 fs_info->avail_data_alloc_bits |= extra_flags;
4115 if (flags & BTRFS_BLOCK_GROUP_METADATA)
4116 fs_info->avail_metadata_alloc_bits |= extra_flags;
4117 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4118 fs_info->avail_system_alloc_bits |= extra_flags;
4119 write_sequnlock(&fs_info->profiles_lock);
4123 * returns target flags in extended format or 0 if restripe for this
4124 * chunk_type is not in progress
4126 * should be called with either volume_mutex or balance_lock held
4128 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
4130 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4136 if (flags & BTRFS_BLOCK_GROUP_DATA &&
4137 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4138 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
4139 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
4140 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4141 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
4142 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
4143 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4144 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
4151 * @flags: available profiles in extended format (see ctree.h)
4153 * Returns reduced profile in chunk format. If profile changing is in
4154 * progress (either running or paused) picks the target profile (if it's
4155 * already available), otherwise falls back to plain reducing.
4157 static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags)
4159 u64 num_devices = fs_info->fs_devices->rw_devices;
4165 * see if restripe for this chunk_type is in progress, if so
4166 * try to reduce to the target profile
4168 spin_lock(&fs_info->balance_lock);
4169 target = get_restripe_target(fs_info, flags);
4171 /* pick target profile only if it's already available */
4172 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
4173 spin_unlock(&fs_info->balance_lock);
4174 return extended_to_chunk(target);
4177 spin_unlock(&fs_info->balance_lock);
4179 /* First, mask out the RAID levels which aren't possible */
4180 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4181 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
4182 allowed |= btrfs_raid_group[raid_type];
4186 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
4187 allowed = BTRFS_BLOCK_GROUP_RAID6;
4188 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
4189 allowed = BTRFS_BLOCK_GROUP_RAID5;
4190 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
4191 allowed = BTRFS_BLOCK_GROUP_RAID10;
4192 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
4193 allowed = BTRFS_BLOCK_GROUP_RAID1;
4194 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
4195 allowed = BTRFS_BLOCK_GROUP_RAID0;
4197 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
4199 return extended_to_chunk(flags | allowed);
4202 static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags)
4209 seq = read_seqbegin(&fs_info->profiles_lock);
4211 if (flags & BTRFS_BLOCK_GROUP_DATA)
4212 flags |= fs_info->avail_data_alloc_bits;
4213 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4214 flags |= fs_info->avail_system_alloc_bits;
4215 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
4216 flags |= fs_info->avail_metadata_alloc_bits;
4217 } while (read_seqretry(&fs_info->profiles_lock, seq));
4219 return btrfs_reduce_alloc_profile(fs_info, flags);
4222 static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
4224 struct btrfs_fs_info *fs_info = root->fs_info;
4229 flags = BTRFS_BLOCK_GROUP_DATA;
4230 else if (root == fs_info->chunk_root)
4231 flags = BTRFS_BLOCK_GROUP_SYSTEM;
4233 flags = BTRFS_BLOCK_GROUP_METADATA;
4235 ret = get_alloc_profile(fs_info, flags);
4239 u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info)
4241 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA);
4244 u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info)
4246 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4249 u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info)
4251 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4254 static u64 btrfs_space_info_used(struct btrfs_space_info *s_info,
4255 bool may_use_included)
4258 return s_info->bytes_used + s_info->bytes_reserved +
4259 s_info->bytes_pinned + s_info->bytes_readonly +
4260 (may_use_included ? s_info->bytes_may_use : 0);
4263 int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes)
4265 struct btrfs_root *root = inode->root;
4266 struct btrfs_fs_info *fs_info = root->fs_info;
4267 struct btrfs_space_info *data_sinfo = fs_info->data_sinfo;
4270 int need_commit = 2;
4271 int have_pinned_space;
4273 /* make sure bytes are sectorsize aligned */
4274 bytes = ALIGN(bytes, fs_info->sectorsize);
4276 if (btrfs_is_free_space_inode(inode)) {
4278 ASSERT(current->journal_info);
4282 /* make sure we have enough space to handle the data first */
4283 spin_lock(&data_sinfo->lock);
4284 used = btrfs_space_info_used(data_sinfo, true);
4286 if (used + bytes > data_sinfo->total_bytes) {
4287 struct btrfs_trans_handle *trans;
4290 * if we don't have enough free bytes in this space then we need
4291 * to alloc a new chunk.
4293 if (!data_sinfo->full) {
4296 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4297 spin_unlock(&data_sinfo->lock);
4299 alloc_target = btrfs_data_alloc_profile(fs_info);
4301 * It is ugly that we don't call nolock join
4302 * transaction for the free space inode case here.
4303 * But it is safe because we only do the data space
4304 * reservation for the free space cache in the
4305 * transaction context, the common join transaction
4306 * just increase the counter of the current transaction
4307 * handler, doesn't try to acquire the trans_lock of
4310 trans = btrfs_join_transaction(root);
4312 return PTR_ERR(trans);
4314 ret = do_chunk_alloc(trans, fs_info, alloc_target,
4315 CHUNK_ALLOC_NO_FORCE);
4316 btrfs_end_transaction(trans);
4321 have_pinned_space = 1;
4330 * If we don't have enough pinned space to deal with this
4331 * allocation, and no removed chunk in current transaction,
4332 * don't bother committing the transaction.
4334 have_pinned_space = percpu_counter_compare(
4335 &data_sinfo->total_bytes_pinned,
4336 used + bytes - data_sinfo->total_bytes);
4337 spin_unlock(&data_sinfo->lock);
4339 /* commit the current transaction and try again */
4344 if (need_commit > 0) {
4345 btrfs_start_delalloc_roots(fs_info, 0, -1);
4346 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0,
4350 trans = btrfs_join_transaction(root);
4352 return PTR_ERR(trans);
4353 if (have_pinned_space >= 0 ||
4354 test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4355 &trans->transaction->flags) ||
4357 ret = btrfs_commit_transaction(trans);
4361 * The cleaner kthread might still be doing iput
4362 * operations. Wait for it to finish so that
4363 * more space is released.
4365 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
4366 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
4369 btrfs_end_transaction(trans);
4373 trace_btrfs_space_reservation(fs_info,
4374 "space_info:enospc",
4375 data_sinfo->flags, bytes, 1);
4378 data_sinfo->bytes_may_use += bytes;
4379 trace_btrfs_space_reservation(fs_info, "space_info",
4380 data_sinfo->flags, bytes, 1);
4381 spin_unlock(&data_sinfo->lock);
4386 int btrfs_check_data_free_space(struct inode *inode,
4387 struct extent_changeset **reserved, u64 start, u64 len)
4389 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4392 /* align the range */
4393 len = round_up(start + len, fs_info->sectorsize) -
4394 round_down(start, fs_info->sectorsize);
4395 start = round_down(start, fs_info->sectorsize);
4397 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len);
4401 /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */
4402 ret = btrfs_qgroup_reserve_data(inode, reserved, start, len);
4404 btrfs_free_reserved_data_space_noquota(inode, start, len);
4411 * Called if we need to clear a data reservation for this inode
4412 * Normally in a error case.
4414 * This one will *NOT* use accurate qgroup reserved space API, just for case
4415 * which we can't sleep and is sure it won't affect qgroup reserved space.
4416 * Like clear_bit_hook().
4418 void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4421 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4422 struct btrfs_space_info *data_sinfo;
4424 /* Make sure the range is aligned to sectorsize */
4425 len = round_up(start + len, fs_info->sectorsize) -
4426 round_down(start, fs_info->sectorsize);
4427 start = round_down(start, fs_info->sectorsize);
4429 data_sinfo = fs_info->data_sinfo;
4430 spin_lock(&data_sinfo->lock);
4431 if (WARN_ON(data_sinfo->bytes_may_use < len))
4432 data_sinfo->bytes_may_use = 0;
4434 data_sinfo->bytes_may_use -= len;
4435 trace_btrfs_space_reservation(fs_info, "space_info",
4436 data_sinfo->flags, len, 0);
4437 spin_unlock(&data_sinfo->lock);
4441 * Called if we need to clear a data reservation for this inode
4442 * Normally in a error case.
4444 * This one will handle the per-inode data rsv map for accurate reserved
4447 void btrfs_free_reserved_data_space(struct inode *inode,
4448 struct extent_changeset *reserved, u64 start, u64 len)
4450 struct btrfs_root *root = BTRFS_I(inode)->root;
4452 /* Make sure the range is aligned to sectorsize */
4453 len = round_up(start + len, root->fs_info->sectorsize) -
4454 round_down(start, root->fs_info->sectorsize);
4455 start = round_down(start, root->fs_info->sectorsize);
4457 btrfs_free_reserved_data_space_noquota(inode, start, len);
4458 btrfs_qgroup_free_data(inode, reserved, start, len);
4461 static void force_metadata_allocation(struct btrfs_fs_info *info)
4463 struct list_head *head = &info->space_info;
4464 struct btrfs_space_info *found;
4467 list_for_each_entry_rcu(found, head, list) {
4468 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4469 found->force_alloc = CHUNK_ALLOC_FORCE;
4474 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4476 return (global->size << 1);
4479 static int should_alloc_chunk(struct btrfs_fs_info *fs_info,
4480 struct btrfs_space_info *sinfo, int force)
4482 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4483 u64 bytes_used = btrfs_space_info_used(sinfo, false);
4486 if (force == CHUNK_ALLOC_FORCE)
4490 * We need to take into account the global rsv because for all intents
4491 * and purposes it's used space. Don't worry about locking the
4492 * global_rsv, it doesn't change except when the transaction commits.
4494 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4495 bytes_used += calc_global_rsv_need_space(global_rsv);
4498 * in limited mode, we want to have some free space up to
4499 * about 1% of the FS size.
4501 if (force == CHUNK_ALLOC_LIMITED) {
4502 thresh = btrfs_super_total_bytes(fs_info->super_copy);
4503 thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1));
4505 if (sinfo->total_bytes - bytes_used < thresh)
4509 if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8))
4514 static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type)
4518 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4519 BTRFS_BLOCK_GROUP_RAID0 |
4520 BTRFS_BLOCK_GROUP_RAID5 |
4521 BTRFS_BLOCK_GROUP_RAID6))
4522 num_dev = fs_info->fs_devices->rw_devices;
4523 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4526 num_dev = 1; /* DUP or single */
4532 * If @is_allocation is true, reserve space in the system space info necessary
4533 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4536 void check_system_chunk(struct btrfs_trans_handle *trans,
4537 struct btrfs_fs_info *fs_info, u64 type)
4539 struct btrfs_space_info *info;
4546 * Needed because we can end up allocating a system chunk and for an
4547 * atomic and race free space reservation in the chunk block reserve.
4549 ASSERT(mutex_is_locked(&fs_info->chunk_mutex));
4551 info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4552 spin_lock(&info->lock);
4553 left = info->total_bytes - btrfs_space_info_used(info, true);
4554 spin_unlock(&info->lock);
4556 num_devs = get_profile_num_devs(fs_info, type);
4558 /* num_devs device items to update and 1 chunk item to add or remove */
4559 thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) +
4560 btrfs_calc_trans_metadata_size(fs_info, 1);
4562 if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4563 btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu",
4564 left, thresh, type);
4565 dump_space_info(fs_info, info, 0, 0);
4568 if (left < thresh) {
4569 u64 flags = btrfs_system_alloc_profile(fs_info);
4572 * Ignore failure to create system chunk. We might end up not
4573 * needing it, as we might not need to COW all nodes/leafs from
4574 * the paths we visit in the chunk tree (they were already COWed
4575 * or created in the current transaction for example).
4577 ret = btrfs_alloc_chunk(trans, fs_info, flags);
4581 ret = btrfs_block_rsv_add(fs_info->chunk_root,
4582 &fs_info->chunk_block_rsv,
4583 thresh, BTRFS_RESERVE_NO_FLUSH);
4585 trans->chunk_bytes_reserved += thresh;
4590 * If force is CHUNK_ALLOC_FORCE:
4591 * - return 1 if it successfully allocates a chunk,
4592 * - return errors including -ENOSPC otherwise.
4593 * If force is NOT CHUNK_ALLOC_FORCE:
4594 * - return 0 if it doesn't need to allocate a new chunk,
4595 * - return 1 if it successfully allocates a chunk,
4596 * - return errors including -ENOSPC otherwise.
4598 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4599 struct btrfs_fs_info *fs_info, u64 flags, int force)
4601 struct btrfs_space_info *space_info;
4602 int wait_for_alloc = 0;
4605 /* Don't re-enter if we're already allocating a chunk */
4606 if (trans->allocating_chunk)
4609 space_info = __find_space_info(fs_info, flags);
4611 ret = create_space_info(fs_info, flags, &space_info);
4617 spin_lock(&space_info->lock);
4618 if (force < space_info->force_alloc)
4619 force = space_info->force_alloc;
4620 if (space_info->full) {
4621 if (should_alloc_chunk(fs_info, space_info, force))
4625 spin_unlock(&space_info->lock);
4629 if (!should_alloc_chunk(fs_info, space_info, force)) {
4630 spin_unlock(&space_info->lock);
4632 } else if (space_info->chunk_alloc) {
4635 space_info->chunk_alloc = 1;
4638 spin_unlock(&space_info->lock);
4640 mutex_lock(&fs_info->chunk_mutex);
4643 * The chunk_mutex is held throughout the entirety of a chunk
4644 * allocation, so once we've acquired the chunk_mutex we know that the
4645 * other guy is done and we need to recheck and see if we should
4648 if (wait_for_alloc) {
4649 mutex_unlock(&fs_info->chunk_mutex);
4654 trans->allocating_chunk = true;
4657 * If we have mixed data/metadata chunks we want to make sure we keep
4658 * allocating mixed chunks instead of individual chunks.
4660 if (btrfs_mixed_space_info(space_info))
4661 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4664 * if we're doing a data chunk, go ahead and make sure that
4665 * we keep a reasonable number of metadata chunks allocated in the
4668 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4669 fs_info->data_chunk_allocations++;
4670 if (!(fs_info->data_chunk_allocations %
4671 fs_info->metadata_ratio))
4672 force_metadata_allocation(fs_info);
4676 * Check if we have enough space in SYSTEM chunk because we may need
4677 * to update devices.
4679 check_system_chunk(trans, fs_info, flags);
4681 ret = btrfs_alloc_chunk(trans, fs_info, flags);
4682 trans->allocating_chunk = false;
4684 spin_lock(&space_info->lock);
4685 if (ret < 0 && ret != -ENOSPC)
4688 space_info->full = 1;
4692 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4694 space_info->chunk_alloc = 0;
4695 spin_unlock(&space_info->lock);
4696 mutex_unlock(&fs_info->chunk_mutex);
4698 * When we allocate a new chunk we reserve space in the chunk block
4699 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4700 * add new nodes/leafs to it if we end up needing to do it when
4701 * inserting the chunk item and updating device items as part of the
4702 * second phase of chunk allocation, performed by
4703 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4704 * large number of new block groups to create in our transaction
4705 * handle's new_bgs list to avoid exhausting the chunk block reserve
4706 * in extreme cases - like having a single transaction create many new
4707 * block groups when starting to write out the free space caches of all
4708 * the block groups that were made dirty during the lifetime of the
4711 if (trans->can_flush_pending_bgs &&
4712 trans->chunk_bytes_reserved >= (u64)SZ_2M) {
4713 btrfs_create_pending_block_groups(trans);
4714 btrfs_trans_release_chunk_metadata(trans);
4719 static int can_overcommit(struct btrfs_fs_info *fs_info,
4720 struct btrfs_space_info *space_info, u64 bytes,
4721 enum btrfs_reserve_flush_enum flush,
4724 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4730 /* Don't overcommit when in mixed mode. */
4731 if (space_info->flags & BTRFS_BLOCK_GROUP_DATA)
4735 profile = btrfs_system_alloc_profile(fs_info);
4737 profile = btrfs_metadata_alloc_profile(fs_info);
4739 used = btrfs_space_info_used(space_info, false);
4742 * We only want to allow over committing if we have lots of actual space
4743 * free, but if we don't have enough space to handle the global reserve
4744 * space then we could end up having a real enospc problem when trying
4745 * to allocate a chunk or some other such important allocation.
4747 spin_lock(&global_rsv->lock);
4748 space_size = calc_global_rsv_need_space(global_rsv);
4749 spin_unlock(&global_rsv->lock);
4750 if (used + space_size >= space_info->total_bytes)
4753 used += space_info->bytes_may_use;
4755 avail = atomic64_read(&fs_info->free_chunk_space);
4758 * If we have dup, raid1 or raid10 then only half of the free
4759 * space is actually useable. For raid56, the space info used
4760 * doesn't include the parity drive, so we don't have to
4763 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4764 BTRFS_BLOCK_GROUP_RAID1 |
4765 BTRFS_BLOCK_GROUP_RAID10))
4769 * If we aren't flushing all things, let us overcommit up to
4770 * 1/2th of the space. If we can flush, don't let us overcommit
4771 * too much, let it overcommit up to 1/8 of the space.
4773 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4778 if (used + bytes < space_info->total_bytes + avail)
4783 static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info,
4784 unsigned long nr_pages, int nr_items)
4786 struct super_block *sb = fs_info->sb;
4788 if (down_read_trylock(&sb->s_umount)) {
4789 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4790 up_read(&sb->s_umount);
4793 * We needn't worry the filesystem going from r/w to r/o though
4794 * we don't acquire ->s_umount mutex, because the filesystem
4795 * should guarantee the delalloc inodes list be empty after
4796 * the filesystem is readonly(all dirty pages are written to
4799 btrfs_start_delalloc_roots(fs_info, 0, nr_items);
4800 if (!current->journal_info)
4801 btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1);
4805 static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info,
4811 bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
4812 nr = div64_u64(to_reclaim, bytes);
4818 #define EXTENT_SIZE_PER_ITEM SZ_256K
4821 * shrink metadata reservation for delalloc
4823 static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim,
4824 u64 orig, bool wait_ordered)
4826 struct btrfs_space_info *space_info;
4827 struct btrfs_trans_handle *trans;
4832 unsigned long nr_pages;
4834 enum btrfs_reserve_flush_enum flush;
4836 /* Calc the number of the pages we need flush for space reservation */
4837 items = calc_reclaim_items_nr(fs_info, to_reclaim);
4838 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4840 trans = (struct btrfs_trans_handle *)current->journal_info;
4841 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4843 delalloc_bytes = percpu_counter_sum_positive(
4844 &fs_info->delalloc_bytes);
4845 if (delalloc_bytes == 0) {
4849 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4854 while (delalloc_bytes && loops < 3) {
4855 max_reclaim = min(delalloc_bytes, to_reclaim);
4856 nr_pages = max_reclaim >> PAGE_SHIFT;
4857 btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items);
4859 * We need to wait for the async pages to actually start before
4862 max_reclaim = atomic_read(&fs_info->async_delalloc_pages);
4866 if (max_reclaim <= nr_pages)
4869 max_reclaim -= nr_pages;
4871 wait_event(fs_info->async_submit_wait,
4872 atomic_read(&fs_info->async_delalloc_pages) <=
4876 flush = BTRFS_RESERVE_FLUSH_ALL;
4878 flush = BTRFS_RESERVE_NO_FLUSH;
4879 spin_lock(&space_info->lock);
4880 if (list_empty(&space_info->tickets) &&
4881 list_empty(&space_info->priority_tickets)) {
4882 spin_unlock(&space_info->lock);
4885 spin_unlock(&space_info->lock);
4888 if (wait_ordered && !trans) {
4889 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4891 time_left = schedule_timeout_killable(1);
4895 delalloc_bytes = percpu_counter_sum_positive(
4896 &fs_info->delalloc_bytes);
4900 struct reserve_ticket {
4903 struct list_head list;
4904 wait_queue_head_t wait;
4908 * maybe_commit_transaction - possibly commit the transaction if its ok to
4909 * @root - the root we're allocating for
4910 * @bytes - the number of bytes we want to reserve
4911 * @force - force the commit
4913 * This will check to make sure that committing the transaction will actually
4914 * get us somewhere and then commit the transaction if it does. Otherwise it
4915 * will return -ENOSPC.
4917 static int may_commit_transaction(struct btrfs_fs_info *fs_info,
4918 struct btrfs_space_info *space_info)
4920 struct reserve_ticket *ticket = NULL;
4921 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv;
4922 struct btrfs_trans_handle *trans;
4925 trans = (struct btrfs_trans_handle *)current->journal_info;
4929 spin_lock(&space_info->lock);
4930 if (!list_empty(&space_info->priority_tickets))
4931 ticket = list_first_entry(&space_info->priority_tickets,
4932 struct reserve_ticket, list);
4933 else if (!list_empty(&space_info->tickets))
4934 ticket = list_first_entry(&space_info->tickets,
4935 struct reserve_ticket, list);
4936 bytes = (ticket) ? ticket->bytes : 0;
4937 spin_unlock(&space_info->lock);
4942 /* See if there is enough pinned space to make this reservation */
4943 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4948 * See if there is some space in the delayed insertion reservation for
4951 if (space_info != delayed_rsv->space_info)
4954 spin_lock(&delayed_rsv->lock);
4955 if (delayed_rsv->size > bytes)
4958 bytes -= delayed_rsv->size;
4959 spin_unlock(&delayed_rsv->lock);
4961 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4967 trans = btrfs_join_transaction(fs_info->extent_root);
4971 return btrfs_commit_transaction(trans);
4975 * Try to flush some data based on policy set by @state. This is only advisory
4976 * and may fail for various reasons. The caller is supposed to examine the
4977 * state of @space_info to detect the outcome.
4979 static void flush_space(struct btrfs_fs_info *fs_info,
4980 struct btrfs_space_info *space_info, u64 num_bytes,
4983 struct btrfs_root *root = fs_info->extent_root;
4984 struct btrfs_trans_handle *trans;
4989 case FLUSH_DELAYED_ITEMS_NR:
4990 case FLUSH_DELAYED_ITEMS:
4991 if (state == FLUSH_DELAYED_ITEMS_NR)
4992 nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2;
4996 trans = btrfs_join_transaction(root);
4997 if (IS_ERR(trans)) {
4998 ret = PTR_ERR(trans);
5001 ret = btrfs_run_delayed_items_nr(trans, nr);
5002 btrfs_end_transaction(trans);
5004 case FLUSH_DELALLOC:
5005 case FLUSH_DELALLOC_WAIT:
5006 shrink_delalloc(fs_info, num_bytes * 2, num_bytes,
5007 state == FLUSH_DELALLOC_WAIT);
5010 trans = btrfs_join_transaction(root);
5011 if (IS_ERR(trans)) {
5012 ret = PTR_ERR(trans);
5015 ret = do_chunk_alloc(trans, fs_info,
5016 btrfs_metadata_alloc_profile(fs_info),
5017 CHUNK_ALLOC_NO_FORCE);
5018 btrfs_end_transaction(trans);
5019 if (ret > 0 || ret == -ENOSPC)
5023 ret = may_commit_transaction(fs_info, space_info);
5030 trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, state,
5036 btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info,
5037 struct btrfs_space_info *space_info,
5040 struct reserve_ticket *ticket;
5045 list_for_each_entry(ticket, &space_info->tickets, list)
5046 to_reclaim += ticket->bytes;
5047 list_for_each_entry(ticket, &space_info->priority_tickets, list)
5048 to_reclaim += ticket->bytes;
5052 to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M);
5053 if (can_overcommit(fs_info, space_info, to_reclaim,
5054 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
5057 used = btrfs_space_info_used(space_info, true);
5059 if (can_overcommit(fs_info, space_info, SZ_1M,
5060 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
5061 expected = div_factor_fine(space_info->total_bytes, 95);
5063 expected = div_factor_fine(space_info->total_bytes, 90);
5065 if (used > expected)
5066 to_reclaim = used - expected;
5069 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
5070 space_info->bytes_reserved);
5074 static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info,
5075 struct btrfs_space_info *space_info,
5076 u64 used, bool system_chunk)
5078 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
5080 /* If we're just plain full then async reclaim just slows us down. */
5081 if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh)
5084 if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5088 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
5089 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
5092 static void wake_all_tickets(struct list_head *head)
5094 struct reserve_ticket *ticket;
5096 while (!list_empty(head)) {
5097 ticket = list_first_entry(head, struct reserve_ticket, list);
5098 list_del_init(&ticket->list);
5099 ticket->error = -ENOSPC;
5100 wake_up(&ticket->wait);
5105 * This is for normal flushers, we can wait all goddamned day if we want to. We
5106 * will loop and continuously try to flush as long as we are making progress.
5107 * We count progress as clearing off tickets each time we have to loop.
5109 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
5111 struct btrfs_fs_info *fs_info;
5112 struct btrfs_space_info *space_info;
5115 int commit_cycles = 0;
5116 u64 last_tickets_id;
5118 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
5119 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5121 spin_lock(&space_info->lock);
5122 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5125 space_info->flush = 0;
5126 spin_unlock(&space_info->lock);
5129 last_tickets_id = space_info->tickets_id;
5130 spin_unlock(&space_info->lock);
5132 flush_state = FLUSH_DELAYED_ITEMS_NR;
5134 flush_space(fs_info, space_info, to_reclaim, flush_state);
5135 spin_lock(&space_info->lock);
5136 if (list_empty(&space_info->tickets)) {
5137 space_info->flush = 0;
5138 spin_unlock(&space_info->lock);
5141 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info,
5144 if (last_tickets_id == space_info->tickets_id) {
5147 last_tickets_id = space_info->tickets_id;
5148 flush_state = FLUSH_DELAYED_ITEMS_NR;
5153 if (flush_state > COMMIT_TRANS) {
5155 if (commit_cycles > 2) {
5156 wake_all_tickets(&space_info->tickets);
5157 space_info->flush = 0;
5159 flush_state = FLUSH_DELAYED_ITEMS_NR;
5162 spin_unlock(&space_info->lock);
5163 } while (flush_state <= COMMIT_TRANS);
5166 void btrfs_init_async_reclaim_work(struct work_struct *work)
5168 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
5171 static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
5172 struct btrfs_space_info *space_info,
5173 struct reserve_ticket *ticket)
5176 int flush_state = FLUSH_DELAYED_ITEMS_NR;
5178 spin_lock(&space_info->lock);
5179 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5182 spin_unlock(&space_info->lock);
5185 spin_unlock(&space_info->lock);
5188 flush_space(fs_info, space_info, to_reclaim, flush_state);
5190 spin_lock(&space_info->lock);
5191 if (ticket->bytes == 0) {
5192 spin_unlock(&space_info->lock);
5195 spin_unlock(&space_info->lock);
5198 * Priority flushers can't wait on delalloc without
5201 if (flush_state == FLUSH_DELALLOC ||
5202 flush_state == FLUSH_DELALLOC_WAIT)
5203 flush_state = ALLOC_CHUNK;
5204 } while (flush_state < COMMIT_TRANS);
5207 static int wait_reserve_ticket(struct btrfs_fs_info *fs_info,
5208 struct btrfs_space_info *space_info,
5209 struct reserve_ticket *ticket, u64 orig_bytes)
5215 spin_lock(&space_info->lock);
5216 while (ticket->bytes > 0 && ticket->error == 0) {
5217 ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE);
5222 spin_unlock(&space_info->lock);
5226 finish_wait(&ticket->wait, &wait);
5227 spin_lock(&space_info->lock);
5230 ret = ticket->error;
5231 if (!list_empty(&ticket->list))
5232 list_del_init(&ticket->list);
5233 if (ticket->bytes && ticket->bytes < orig_bytes) {
5234 u64 num_bytes = orig_bytes - ticket->bytes;
5235 space_info->bytes_may_use -= num_bytes;
5236 trace_btrfs_space_reservation(fs_info, "space_info",
5237 space_info->flags, num_bytes, 0);
5239 spin_unlock(&space_info->lock);
5245 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5246 * @root - the root we're allocating for
5247 * @space_info - the space info we want to allocate from
5248 * @orig_bytes - the number of bytes we want
5249 * @flush - whether or not we can flush to make our reservation
5251 * This will reserve orig_bytes number of bytes from the space info associated
5252 * with the block_rsv. If there is not enough space it will make an attempt to
5253 * flush out space to make room. It will do this by flushing delalloc if
5254 * possible or committing the transaction. If flush is 0 then no attempts to
5255 * regain reservations will be made and this will fail if there is not enough
5258 static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info,
5259 struct btrfs_space_info *space_info,
5261 enum btrfs_reserve_flush_enum flush,
5264 struct reserve_ticket ticket;
5269 ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL);
5271 spin_lock(&space_info->lock);
5273 used = btrfs_space_info_used(space_info, true);
5276 * If we have enough space then hooray, make our reservation and carry
5277 * on. If not see if we can overcommit, and if we can, hooray carry on.
5278 * If not things get more complicated.
5280 if (used + orig_bytes <= space_info->total_bytes) {
5281 space_info->bytes_may_use += orig_bytes;
5282 trace_btrfs_space_reservation(fs_info, "space_info",
5283 space_info->flags, orig_bytes, 1);
5285 } else if (can_overcommit(fs_info, space_info, orig_bytes, flush,
5287 space_info->bytes_may_use += orig_bytes;
5288 trace_btrfs_space_reservation(fs_info, "space_info",
5289 space_info->flags, orig_bytes, 1);
5294 * If we couldn't make a reservation then setup our reservation ticket
5295 * and kick the async worker if it's not already running.
5297 * If we are a priority flusher then we just need to add our ticket to
5298 * the list and we will do our own flushing further down.
5300 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
5301 ticket.bytes = orig_bytes;
5303 init_waitqueue_head(&ticket.wait);
5304 if (flush == BTRFS_RESERVE_FLUSH_ALL) {
5305 list_add_tail(&ticket.list, &space_info->tickets);
5306 if (!space_info->flush) {
5307 space_info->flush = 1;
5308 trace_btrfs_trigger_flush(fs_info,
5312 queue_work(system_unbound_wq,
5313 &fs_info->async_reclaim_work);
5316 list_add_tail(&ticket.list,
5317 &space_info->priority_tickets);
5319 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
5322 * We will do the space reservation dance during log replay,
5323 * which means we won't have fs_info->fs_root set, so don't do
5324 * the async reclaim as we will panic.
5326 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) &&
5327 need_do_async_reclaim(fs_info, space_info,
5328 used, system_chunk) &&
5329 !work_busy(&fs_info->async_reclaim_work)) {
5330 trace_btrfs_trigger_flush(fs_info, space_info->flags,
5331 orig_bytes, flush, "preempt");
5332 queue_work(system_unbound_wq,
5333 &fs_info->async_reclaim_work);
5336 spin_unlock(&space_info->lock);
5337 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
5340 if (flush == BTRFS_RESERVE_FLUSH_ALL)
5341 return wait_reserve_ticket(fs_info, space_info, &ticket,
5345 priority_reclaim_metadata_space(fs_info, space_info, &ticket);
5346 spin_lock(&space_info->lock);
5348 if (ticket.bytes < orig_bytes) {
5349 u64 num_bytes = orig_bytes - ticket.bytes;
5350 space_info->bytes_may_use -= num_bytes;
5351 trace_btrfs_space_reservation(fs_info, "space_info",
5356 list_del_init(&ticket.list);
5359 spin_unlock(&space_info->lock);
5360 ASSERT(list_empty(&ticket.list));
5365 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5366 * @root - the root we're allocating for
5367 * @block_rsv - the block_rsv we're allocating for
5368 * @orig_bytes - the number of bytes we want
5369 * @flush - whether or not we can flush to make our reservation
5371 * This will reserve orgi_bytes number of bytes from the space info associated
5372 * with the block_rsv. If there is not enough space it will make an attempt to
5373 * flush out space to make room. It will do this by flushing delalloc if
5374 * possible or committing the transaction. If flush is 0 then no attempts to
5375 * regain reservations will be made and this will fail if there is not enough
5378 static int reserve_metadata_bytes(struct btrfs_root *root,
5379 struct btrfs_block_rsv *block_rsv,
5381 enum btrfs_reserve_flush_enum flush)
5383 struct btrfs_fs_info *fs_info = root->fs_info;
5384 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5386 bool system_chunk = (root == fs_info->chunk_root);
5388 ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info,
5389 orig_bytes, flush, system_chunk);
5390 if (ret == -ENOSPC &&
5391 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5392 if (block_rsv != global_rsv &&
5393 !block_rsv_use_bytes(global_rsv, orig_bytes))
5396 if (ret == -ENOSPC) {
5397 trace_btrfs_space_reservation(fs_info, "space_info:enospc",
5398 block_rsv->space_info->flags,
5401 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG))
5402 dump_space_info(fs_info, block_rsv->space_info,
5408 static struct btrfs_block_rsv *get_block_rsv(
5409 const struct btrfs_trans_handle *trans,
5410 const struct btrfs_root *root)
5412 struct btrfs_fs_info *fs_info = root->fs_info;
5413 struct btrfs_block_rsv *block_rsv = NULL;
5415 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5416 (root == fs_info->csum_root && trans->adding_csums) ||
5417 (root == fs_info->uuid_root))
5418 block_rsv = trans->block_rsv;
5421 block_rsv = root->block_rsv;
5424 block_rsv = &fs_info->empty_block_rsv;
5429 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5433 spin_lock(&block_rsv->lock);
5434 if (block_rsv->reserved >= num_bytes) {
5435 block_rsv->reserved -= num_bytes;
5436 if (block_rsv->reserved < block_rsv->size)
5437 block_rsv->full = 0;
5440 spin_unlock(&block_rsv->lock);
5444 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5445 u64 num_bytes, int update_size)
5447 spin_lock(&block_rsv->lock);
5448 block_rsv->reserved += num_bytes;
5450 block_rsv->size += num_bytes;
5451 else if (block_rsv->reserved >= block_rsv->size)
5452 block_rsv->full = 1;
5453 spin_unlock(&block_rsv->lock);
5456 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5457 struct btrfs_block_rsv *dest, u64 num_bytes,
5460 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5463 if (global_rsv->space_info != dest->space_info)
5466 spin_lock(&global_rsv->lock);
5467 min_bytes = div_factor(global_rsv->size, min_factor);
5468 if (global_rsv->reserved < min_bytes + num_bytes) {
5469 spin_unlock(&global_rsv->lock);
5472 global_rsv->reserved -= num_bytes;
5473 if (global_rsv->reserved < global_rsv->size)
5474 global_rsv->full = 0;
5475 spin_unlock(&global_rsv->lock);
5477 block_rsv_add_bytes(dest, num_bytes, 1);
5482 * This is for space we already have accounted in space_info->bytes_may_use, so
5483 * basically when we're returning space from block_rsv's.
5485 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
5486 struct btrfs_space_info *space_info,
5489 struct reserve_ticket *ticket;
5490 struct list_head *head;
5492 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH;
5493 bool check_overcommit = false;
5495 spin_lock(&space_info->lock);
5496 head = &space_info->priority_tickets;
5499 * If we are over our limit then we need to check and see if we can
5500 * overcommit, and if we can't then we just need to free up our space
5501 * and not satisfy any requests.
5503 used = btrfs_space_info_used(space_info, true);
5504 if (used - num_bytes >= space_info->total_bytes)
5505 check_overcommit = true;
5507 while (!list_empty(head) && num_bytes) {
5508 ticket = list_first_entry(head, struct reserve_ticket,
5511 * We use 0 bytes because this space is already reserved, so
5512 * adding the ticket space would be a double count.
5514 if (check_overcommit &&
5515 !can_overcommit(fs_info, space_info, 0, flush, false))
5517 if (num_bytes >= ticket->bytes) {
5518 list_del_init(&ticket->list);
5519 num_bytes -= ticket->bytes;
5521 space_info->tickets_id++;
5522 wake_up(&ticket->wait);
5524 ticket->bytes -= num_bytes;
5529 if (num_bytes && head == &space_info->priority_tickets) {
5530 head = &space_info->tickets;
5531 flush = BTRFS_RESERVE_FLUSH_ALL;
5534 space_info->bytes_may_use -= num_bytes;
5535 trace_btrfs_space_reservation(fs_info, "space_info",
5536 space_info->flags, num_bytes, 0);
5537 spin_unlock(&space_info->lock);
5541 * This is for newly allocated space that isn't accounted in
5542 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5543 * we use this helper.
5545 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
5546 struct btrfs_space_info *space_info,
5549 struct reserve_ticket *ticket;
5550 struct list_head *head = &space_info->priority_tickets;
5553 while (!list_empty(head) && num_bytes) {
5554 ticket = list_first_entry(head, struct reserve_ticket,
5556 if (num_bytes >= ticket->bytes) {
5557 trace_btrfs_space_reservation(fs_info, "space_info",
5560 list_del_init(&ticket->list);
5561 num_bytes -= ticket->bytes;
5562 space_info->bytes_may_use += ticket->bytes;
5564 space_info->tickets_id++;
5565 wake_up(&ticket->wait);
5567 trace_btrfs_space_reservation(fs_info, "space_info",
5570 space_info->bytes_may_use += num_bytes;
5571 ticket->bytes -= num_bytes;
5576 if (num_bytes && head == &space_info->priority_tickets) {
5577 head = &space_info->tickets;
5582 static u64 block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5583 struct btrfs_block_rsv *block_rsv,
5584 struct btrfs_block_rsv *dest, u64 num_bytes)
5586 struct btrfs_space_info *space_info = block_rsv->space_info;
5589 spin_lock(&block_rsv->lock);
5590 if (num_bytes == (u64)-1)
5591 num_bytes = block_rsv->size;
5592 block_rsv->size -= num_bytes;
5593 if (block_rsv->reserved >= block_rsv->size) {
5594 num_bytes = block_rsv->reserved - block_rsv->size;
5595 block_rsv->reserved = block_rsv->size;
5596 block_rsv->full = 1;
5600 spin_unlock(&block_rsv->lock);
5603 if (num_bytes > 0) {
5605 spin_lock(&dest->lock);
5609 bytes_to_add = dest->size - dest->reserved;
5610 bytes_to_add = min(num_bytes, bytes_to_add);
5611 dest->reserved += bytes_to_add;
5612 if (dest->reserved >= dest->size)
5614 num_bytes -= bytes_to_add;
5616 spin_unlock(&dest->lock);
5619 space_info_add_old_bytes(fs_info, space_info,
5625 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
5626 struct btrfs_block_rsv *dst, u64 num_bytes,
5631 ret = block_rsv_use_bytes(src, num_bytes);
5635 block_rsv_add_bytes(dst, num_bytes, update_size);
5639 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5641 memset(rsv, 0, sizeof(*rsv));
5642 spin_lock_init(&rsv->lock);
5646 void btrfs_init_metadata_block_rsv(struct btrfs_fs_info *fs_info,
5647 struct btrfs_block_rsv *rsv,
5648 unsigned short type)
5650 btrfs_init_block_rsv(rsv, type);
5651 rsv->space_info = __find_space_info(fs_info,
5652 BTRFS_BLOCK_GROUP_METADATA);
5655 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info,
5656 unsigned short type)
5658 struct btrfs_block_rsv *block_rsv;
5660 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5664 btrfs_init_metadata_block_rsv(fs_info, block_rsv, type);
5668 void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info,
5669 struct btrfs_block_rsv *rsv)
5673 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
5677 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
5682 int btrfs_block_rsv_add(struct btrfs_root *root,
5683 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5684 enum btrfs_reserve_flush_enum flush)
5691 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5693 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5700 int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor)
5708 spin_lock(&block_rsv->lock);
5709 num_bytes = div_factor(block_rsv->size, min_factor);
5710 if (block_rsv->reserved >= num_bytes)
5712 spin_unlock(&block_rsv->lock);
5717 int btrfs_block_rsv_refill(struct btrfs_root *root,
5718 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5719 enum btrfs_reserve_flush_enum flush)
5727 spin_lock(&block_rsv->lock);
5728 num_bytes = min_reserved;
5729 if (block_rsv->reserved >= num_bytes)
5732 num_bytes -= block_rsv->reserved;
5733 spin_unlock(&block_rsv->lock);
5738 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5740 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5748 * btrfs_inode_rsv_refill - refill the inode block rsv.
5749 * @inode - the inode we are refilling.
5750 * @flush - the flusing restriction.
5752 * Essentially the same as btrfs_block_rsv_refill, except it uses the
5753 * block_rsv->size as the minimum size. We'll either refill the missing amount
5754 * or return if we already have enough space. This will also handle the resreve
5755 * tracepoint for the reserved amount.
5757 static int btrfs_inode_rsv_refill(struct btrfs_inode *inode,
5758 enum btrfs_reserve_flush_enum flush)
5760 struct btrfs_root *root = inode->root;
5761 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5765 spin_lock(&block_rsv->lock);
5766 if (block_rsv->reserved < block_rsv->size)
5767 num_bytes = block_rsv->size - block_rsv->reserved;
5768 spin_unlock(&block_rsv->lock);
5773 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5775 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5776 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5777 btrfs_ino(inode), num_bytes, 1);
5783 * btrfs_inode_rsv_release - release any excessive reservation.
5784 * @inode - the inode we need to release from.
5786 * This is the same as btrfs_block_rsv_release, except that it handles the
5787 * tracepoint for the reservation.
5789 static void btrfs_inode_rsv_release(struct btrfs_inode *inode)
5791 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5792 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5793 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5797 * Since we statically set the block_rsv->size we just want to say we
5798 * are releasing 0 bytes, and then we'll just get the reservation over
5801 released = block_rsv_release_bytes(fs_info, block_rsv, global_rsv, 0);
5803 trace_btrfs_space_reservation(fs_info, "delalloc",
5804 btrfs_ino(inode), released, 0);
5807 void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
5808 struct btrfs_block_rsv *block_rsv,
5811 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5813 if (global_rsv == block_rsv ||
5814 block_rsv->space_info != global_rsv->space_info)
5816 block_rsv_release_bytes(fs_info, block_rsv, global_rsv, num_bytes);
5819 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5821 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5822 struct btrfs_space_info *sinfo = block_rsv->space_info;
5826 * The global block rsv is based on the size of the extent tree, the
5827 * checksum tree and the root tree. If the fs is empty we want to set
5828 * it to a minimal amount for safety.
5830 num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) +
5831 btrfs_root_used(&fs_info->csum_root->root_item) +
5832 btrfs_root_used(&fs_info->tree_root->root_item);
5833 num_bytes = max_t(u64, num_bytes, SZ_16M);
5835 spin_lock(&sinfo->lock);
5836 spin_lock(&block_rsv->lock);
5838 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
5840 if (block_rsv->reserved < block_rsv->size) {
5841 num_bytes = btrfs_space_info_used(sinfo, true);
5842 if (sinfo->total_bytes > num_bytes) {
5843 num_bytes = sinfo->total_bytes - num_bytes;
5844 num_bytes = min(num_bytes,
5845 block_rsv->size - block_rsv->reserved);
5846 block_rsv->reserved += num_bytes;
5847 sinfo->bytes_may_use += num_bytes;
5848 trace_btrfs_space_reservation(fs_info, "space_info",
5849 sinfo->flags, num_bytes,
5852 } else if (block_rsv->reserved > block_rsv->size) {
5853 num_bytes = block_rsv->reserved - block_rsv->size;
5854 sinfo->bytes_may_use -= num_bytes;
5855 trace_btrfs_space_reservation(fs_info, "space_info",
5856 sinfo->flags, num_bytes, 0);
5857 block_rsv->reserved = block_rsv->size;
5860 if (block_rsv->reserved == block_rsv->size)
5861 block_rsv->full = 1;
5863 block_rsv->full = 0;
5865 spin_unlock(&block_rsv->lock);
5866 spin_unlock(&sinfo->lock);
5869 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5871 struct btrfs_space_info *space_info;
5873 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5874 fs_info->chunk_block_rsv.space_info = space_info;
5876 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5877 fs_info->global_block_rsv.space_info = space_info;
5878 fs_info->trans_block_rsv.space_info = space_info;
5879 fs_info->empty_block_rsv.space_info = space_info;
5880 fs_info->delayed_block_rsv.space_info = space_info;
5882 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5883 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5884 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5885 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5886 if (fs_info->quota_root)
5887 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5888 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5890 update_global_block_rsv(fs_info);
5893 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5895 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5897 WARN_ON(fs_info->trans_block_rsv.size > 0);
5898 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5899 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5900 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5901 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5902 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5907 * To be called after all the new block groups attached to the transaction
5908 * handle have been created (btrfs_create_pending_block_groups()).
5910 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5912 struct btrfs_fs_info *fs_info = trans->fs_info;
5914 if (!trans->chunk_bytes_reserved)
5917 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5919 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5920 trans->chunk_bytes_reserved);
5921 trans->chunk_bytes_reserved = 0;
5924 /* Can only return 0 or -ENOSPC */
5925 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5926 struct btrfs_inode *inode)
5928 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5929 struct btrfs_root *root = inode->root;
5931 * We always use trans->block_rsv here as we will have reserved space
5932 * for our orphan when starting the transaction, using get_block_rsv()
5933 * here will sometimes make us choose the wrong block rsv as we could be
5934 * doing a reloc inode for a non refcounted root.
5936 struct btrfs_block_rsv *src_rsv = trans->block_rsv;
5937 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5940 * We need to hold space in order to delete our orphan item once we've
5941 * added it, so this takes the reservation so we can release it later
5942 * when we are truly done with the orphan item.
5944 u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
5946 trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode),
5948 return btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1);
5951 void btrfs_orphan_release_metadata(struct btrfs_inode *inode)
5953 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5954 struct btrfs_root *root = inode->root;
5955 u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
5957 trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode),
5959 btrfs_block_rsv_release(fs_info, root->orphan_block_rsv, num_bytes);
5963 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5964 * root: the root of the parent directory
5965 * rsv: block reservation
5966 * items: the number of items that we need do reservation
5967 * qgroup_reserved: used to return the reserved size in qgroup
5969 * This function is used to reserve the space for snapshot/subvolume
5970 * creation and deletion. Those operations are different with the
5971 * common file/directory operations, they change two fs/file trees
5972 * and root tree, the number of items that the qgroup reserves is
5973 * different with the free space reservation. So we can not use
5974 * the space reservation mechanism in start_transaction().
5976 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5977 struct btrfs_block_rsv *rsv,
5979 u64 *qgroup_reserved,
5980 bool use_global_rsv)
5984 struct btrfs_fs_info *fs_info = root->fs_info;
5985 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5987 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
5988 /* One for parent inode, two for dir entries */
5989 num_bytes = 3 * fs_info->nodesize;
5990 ret = btrfs_qgroup_reserve_meta(root, num_bytes, true);
5997 *qgroup_reserved = num_bytes;
5999 num_bytes = btrfs_calc_trans_metadata_size(fs_info, items);
6000 rsv->space_info = __find_space_info(fs_info,
6001 BTRFS_BLOCK_GROUP_METADATA);
6002 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
6003 BTRFS_RESERVE_FLUSH_ALL);
6005 if (ret == -ENOSPC && use_global_rsv)
6006 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, 1);
6008 if (ret && *qgroup_reserved)
6009 btrfs_qgroup_free_meta(root, *qgroup_reserved);
6014 void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info,
6015 struct btrfs_block_rsv *rsv)
6017 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
6020 static void btrfs_calculate_inode_block_rsv_size(struct btrfs_fs_info *fs_info,
6021 struct btrfs_inode *inode)
6023 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
6024 u64 reserve_size = 0;
6026 unsigned outstanding_extents;
6028 lockdep_assert_held(&inode->lock);
6029 outstanding_extents = inode->outstanding_extents;
6030 if (outstanding_extents)
6031 reserve_size = btrfs_calc_trans_metadata_size(fs_info,
6032 outstanding_extents + 1);
6033 csum_leaves = btrfs_csum_bytes_to_leaves(fs_info,
6035 reserve_size += btrfs_calc_trans_metadata_size(fs_info,
6038 spin_lock(&block_rsv->lock);
6039 block_rsv->size = reserve_size;
6040 spin_unlock(&block_rsv->lock);
6043 int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes)
6045 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6046 struct btrfs_root *root = inode->root;
6047 unsigned nr_extents;
6048 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
6050 bool delalloc_lock = true;
6052 /* If we are a free space inode we need to not flush since we will be in
6053 * the middle of a transaction commit. We also don't need the delalloc
6054 * mutex since we won't race with anybody. We need this mostly to make
6055 * lockdep shut its filthy mouth.
6057 * If we have a transaction open (can happen if we call truncate_block
6058 * from truncate), then we need FLUSH_LIMIT so we don't deadlock.
6060 if (btrfs_is_free_space_inode(inode)) {
6061 flush = BTRFS_RESERVE_NO_FLUSH;
6062 delalloc_lock = false;
6064 if (current->journal_info)
6065 flush = BTRFS_RESERVE_FLUSH_LIMIT;
6067 if (btrfs_transaction_in_commit(fs_info))
6068 schedule_timeout(1);
6072 mutex_lock(&inode->delalloc_mutex);
6074 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6076 /* Add our new extents and calculate the new rsv size. */
6077 spin_lock(&inode->lock);
6078 nr_extents = count_max_extents(num_bytes);
6079 btrfs_mod_outstanding_extents(inode, nr_extents);
6080 inode->csum_bytes += num_bytes;
6081 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6082 spin_unlock(&inode->lock);
6084 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
6085 ret = btrfs_qgroup_reserve_meta(root,
6086 nr_extents * fs_info->nodesize, true);
6091 ret = btrfs_inode_rsv_refill(inode, flush);
6092 if (unlikely(ret)) {
6093 btrfs_qgroup_free_meta(root,
6094 nr_extents * fs_info->nodesize);
6099 mutex_unlock(&inode->delalloc_mutex);
6103 spin_lock(&inode->lock);
6104 nr_extents = count_max_extents(num_bytes);
6105 btrfs_mod_outstanding_extents(inode, -nr_extents);
6106 inode->csum_bytes -= num_bytes;
6107 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6108 spin_unlock(&inode->lock);
6110 btrfs_inode_rsv_release(inode);
6112 mutex_unlock(&inode->delalloc_mutex);
6117 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
6118 * @inode: the inode to release the reservation for.
6119 * @num_bytes: the number of bytes we are releasing.
6121 * This will release the metadata reservation for an inode. This can be called
6122 * once we complete IO for a given set of bytes to release their metadata
6123 * reservations, or on error for the same reason.
6125 void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes)
6127 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6129 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6130 spin_lock(&inode->lock);
6131 inode->csum_bytes -= num_bytes;
6132 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6133 spin_unlock(&inode->lock);
6135 if (btrfs_is_testing(fs_info))
6138 btrfs_inode_rsv_release(inode);
6142 * btrfs_delalloc_release_extents - release our outstanding_extents
6143 * @inode: the inode to balance the reservation for.
6144 * @num_bytes: the number of bytes we originally reserved with
6146 * When we reserve space we increase outstanding_extents for the extents we may
6147 * add. Once we've set the range as delalloc or created our ordered extents we
6148 * have outstanding_extents to track the real usage, so we use this to free our
6149 * temporarily tracked outstanding_extents. This _must_ be used in conjunction
6150 * with btrfs_delalloc_reserve_metadata.
6152 void btrfs_delalloc_release_extents(struct btrfs_inode *inode, u64 num_bytes)
6154 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6155 unsigned num_extents;
6157 spin_lock(&inode->lock);
6158 num_extents = count_max_extents(num_bytes);
6159 btrfs_mod_outstanding_extents(inode, -num_extents);
6160 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6161 spin_unlock(&inode->lock);
6163 if (btrfs_is_testing(fs_info))
6166 btrfs_inode_rsv_release(inode);
6170 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6172 * @inode: inode we're writing to
6173 * @start: start range we are writing to
6174 * @len: how long the range we are writing to
6175 * @reserved: mandatory parameter, record actually reserved qgroup ranges of
6176 * current reservation.
6178 * This will do the following things
6180 * o reserve space in data space info for num bytes
6181 * and reserve precious corresponding qgroup space
6182 * (Done in check_data_free_space)
6184 * o reserve space for metadata space, based on the number of outstanding
6185 * extents and how much csums will be needed
6186 * also reserve metadata space in a per root over-reserve method.
6187 * o add to the inodes->delalloc_bytes
6188 * o add it to the fs_info's delalloc inodes list.
6189 * (Above 3 all done in delalloc_reserve_metadata)
6191 * Return 0 for success
6192 * Return <0 for error(-ENOSPC or -EQUOT)
6194 int btrfs_delalloc_reserve_space(struct inode *inode,
6195 struct extent_changeset **reserved, u64 start, u64 len)
6199 ret = btrfs_check_data_free_space(inode, reserved, start, len);
6202 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len);
6204 btrfs_free_reserved_data_space(inode, *reserved, start, len);
6209 * btrfs_delalloc_release_space - release data and metadata space for delalloc
6210 * @inode: inode we're releasing space for
6211 * @start: start position of the space already reserved
6212 * @len: the len of the space already reserved
6213 * @release_bytes: the len of the space we consumed or didn't use
6215 * This function will release the metadata space that was not used and will
6216 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
6217 * list if there are no delalloc bytes left.
6218 * Also it will handle the qgroup reserved space.
6220 void btrfs_delalloc_release_space(struct inode *inode,
6221 struct extent_changeset *reserved,
6224 btrfs_delalloc_release_metadata(BTRFS_I(inode), len);
6225 btrfs_free_reserved_data_space(inode, reserved, start, len);
6228 static int update_block_group(struct btrfs_trans_handle *trans,
6229 struct btrfs_fs_info *info, u64 bytenr,
6230 u64 num_bytes, int alloc)
6232 struct btrfs_block_group_cache *cache = NULL;
6233 u64 total = num_bytes;
6238 /* block accounting for super block */
6239 spin_lock(&info->delalloc_root_lock);
6240 old_val = btrfs_super_bytes_used(info->super_copy);
6242 old_val += num_bytes;
6244 old_val -= num_bytes;
6245 btrfs_set_super_bytes_used(info->super_copy, old_val);
6246 spin_unlock(&info->delalloc_root_lock);
6249 cache = btrfs_lookup_block_group(info, bytenr);
6252 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
6253 BTRFS_BLOCK_GROUP_RAID1 |
6254 BTRFS_BLOCK_GROUP_RAID10))
6259 * If this block group has free space cache written out, we
6260 * need to make sure to load it if we are removing space. This
6261 * is because we need the unpinning stage to actually add the
6262 * space back to the block group, otherwise we will leak space.
6264 if (!alloc && cache->cached == BTRFS_CACHE_NO)
6265 cache_block_group(cache, 1);
6267 byte_in_group = bytenr - cache->key.objectid;
6268 WARN_ON(byte_in_group > cache->key.offset);
6270 spin_lock(&cache->space_info->lock);
6271 spin_lock(&cache->lock);
6273 if (btrfs_test_opt(info, SPACE_CACHE) &&
6274 cache->disk_cache_state < BTRFS_DC_CLEAR)
6275 cache->disk_cache_state = BTRFS_DC_CLEAR;
6277 old_val = btrfs_block_group_used(&cache->item);
6278 num_bytes = min(total, cache->key.offset - byte_in_group);
6280 old_val += num_bytes;
6281 btrfs_set_block_group_used(&cache->item, old_val);
6282 cache->reserved -= num_bytes;
6283 cache->space_info->bytes_reserved -= num_bytes;
6284 cache->space_info->bytes_used += num_bytes;
6285 cache->space_info->disk_used += num_bytes * factor;
6286 spin_unlock(&cache->lock);
6287 spin_unlock(&cache->space_info->lock);
6289 old_val -= num_bytes;
6290 btrfs_set_block_group_used(&cache->item, old_val);
6291 cache->pinned += num_bytes;
6292 cache->space_info->bytes_pinned += num_bytes;
6293 cache->space_info->bytes_used -= num_bytes;
6294 cache->space_info->disk_used -= num_bytes * factor;
6295 spin_unlock(&cache->lock);
6296 spin_unlock(&cache->space_info->lock);
6298 trace_btrfs_space_reservation(info, "pinned",
6299 cache->space_info->flags,
6301 percpu_counter_add(&cache->space_info->total_bytes_pinned,
6303 set_extent_dirty(info->pinned_extents,
6304 bytenr, bytenr + num_bytes - 1,
6305 GFP_NOFS | __GFP_NOFAIL);
6308 spin_lock(&trans->transaction->dirty_bgs_lock);
6309 if (list_empty(&cache->dirty_list)) {
6310 list_add_tail(&cache->dirty_list,
6311 &trans->transaction->dirty_bgs);
6312 trans->transaction->num_dirty_bgs++;
6313 btrfs_get_block_group(cache);
6315 spin_unlock(&trans->transaction->dirty_bgs_lock);
6318 * No longer have used bytes in this block group, queue it for
6319 * deletion. We do this after adding the block group to the
6320 * dirty list to avoid races between cleaner kthread and space
6323 if (!alloc && old_val == 0) {
6324 spin_lock(&info->unused_bgs_lock);
6325 if (list_empty(&cache->bg_list)) {
6326 btrfs_get_block_group(cache);
6327 list_add_tail(&cache->bg_list,
6330 spin_unlock(&info->unused_bgs_lock);
6333 btrfs_put_block_group(cache);
6335 bytenr += num_bytes;
6340 static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start)
6342 struct btrfs_block_group_cache *cache;
6345 spin_lock(&fs_info->block_group_cache_lock);
6346 bytenr = fs_info->first_logical_byte;
6347 spin_unlock(&fs_info->block_group_cache_lock);
6349 if (bytenr < (u64)-1)
6352 cache = btrfs_lookup_first_block_group(fs_info, search_start);
6356 bytenr = cache->key.objectid;
6357 btrfs_put_block_group(cache);
6362 static int pin_down_extent(struct btrfs_fs_info *fs_info,
6363 struct btrfs_block_group_cache *cache,
6364 u64 bytenr, u64 num_bytes, int reserved)
6366 spin_lock(&cache->space_info->lock);
6367 spin_lock(&cache->lock);
6368 cache->pinned += num_bytes;
6369 cache->space_info->bytes_pinned += num_bytes;
6371 cache->reserved -= num_bytes;
6372 cache->space_info->bytes_reserved -= num_bytes;
6374 spin_unlock(&cache->lock);
6375 spin_unlock(&cache->space_info->lock);
6377 trace_btrfs_space_reservation(fs_info, "pinned",
6378 cache->space_info->flags, num_bytes, 1);
6379 percpu_counter_add(&cache->space_info->total_bytes_pinned, num_bytes);
6380 set_extent_dirty(fs_info->pinned_extents, bytenr,
6381 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
6386 * this function must be called within transaction
6388 int btrfs_pin_extent(struct btrfs_fs_info *fs_info,
6389 u64 bytenr, u64 num_bytes, int reserved)
6391 struct btrfs_block_group_cache *cache;
6393 cache = btrfs_lookup_block_group(fs_info, bytenr);
6394 BUG_ON(!cache); /* Logic error */
6396 pin_down_extent(fs_info, cache, bytenr, num_bytes, reserved);
6398 btrfs_put_block_group(cache);
6403 * this function must be called within transaction
6405 int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info,
6406 u64 bytenr, u64 num_bytes)
6408 struct btrfs_block_group_cache *cache;
6411 cache = btrfs_lookup_block_group(fs_info, bytenr);
6416 * pull in the free space cache (if any) so that our pin
6417 * removes the free space from the cache. We have load_only set
6418 * to one because the slow code to read in the free extents does check
6419 * the pinned extents.
6421 cache_block_group(cache, 1);
6423 pin_down_extent(fs_info, cache, bytenr, num_bytes, 0);
6425 /* remove us from the free space cache (if we're there at all) */
6426 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6427 btrfs_put_block_group(cache);
6431 static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
6432 u64 start, u64 num_bytes)
6435 struct btrfs_block_group_cache *block_group;
6436 struct btrfs_caching_control *caching_ctl;
6438 block_group = btrfs_lookup_block_group(fs_info, start);
6442 cache_block_group(block_group, 0);
6443 caching_ctl = get_caching_control(block_group);
6447 BUG_ON(!block_group_cache_done(block_group));
6448 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6450 mutex_lock(&caching_ctl->mutex);
6452 if (start >= caching_ctl->progress) {
6453 ret = add_excluded_extent(fs_info, start, num_bytes);
6454 } else if (start + num_bytes <= caching_ctl->progress) {
6455 ret = btrfs_remove_free_space(block_group,
6458 num_bytes = caching_ctl->progress - start;
6459 ret = btrfs_remove_free_space(block_group,
6464 num_bytes = (start + num_bytes) -
6465 caching_ctl->progress;
6466 start = caching_ctl->progress;
6467 ret = add_excluded_extent(fs_info, start, num_bytes);
6470 mutex_unlock(&caching_ctl->mutex);
6471 put_caching_control(caching_ctl);
6473 btrfs_put_block_group(block_group);
6477 int btrfs_exclude_logged_extents(struct btrfs_fs_info *fs_info,
6478 struct extent_buffer *eb)
6480 struct btrfs_file_extent_item *item;
6481 struct btrfs_key key;
6485 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
6488 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6489 btrfs_item_key_to_cpu(eb, &key, i);
6490 if (key.type != BTRFS_EXTENT_DATA_KEY)
6492 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6493 found_type = btrfs_file_extent_type(eb, item);
6494 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6496 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6498 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6499 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6500 __exclude_logged_extent(fs_info, key.objectid, key.offset);
6507 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg)
6509 atomic_inc(&bg->reservations);
6512 void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info,
6515 struct btrfs_block_group_cache *bg;
6517 bg = btrfs_lookup_block_group(fs_info, start);
6519 if (atomic_dec_and_test(&bg->reservations))
6520 wake_up_atomic_t(&bg->reservations);
6521 btrfs_put_block_group(bg);
6524 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg)
6526 struct btrfs_space_info *space_info = bg->space_info;
6530 if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA))
6534 * Our block group is read only but before we set it to read only,
6535 * some task might have had allocated an extent from it already, but it
6536 * has not yet created a respective ordered extent (and added it to a
6537 * root's list of ordered extents).
6538 * Therefore wait for any task currently allocating extents, since the
6539 * block group's reservations counter is incremented while a read lock
6540 * on the groups' semaphore is held and decremented after releasing
6541 * the read access on that semaphore and creating the ordered extent.
6543 down_write(&space_info->groups_sem);
6544 up_write(&space_info->groups_sem);
6546 wait_on_atomic_t(&bg->reservations, atomic_t_wait,
6547 TASK_UNINTERRUPTIBLE);
6551 * btrfs_add_reserved_bytes - update the block_group and space info counters
6552 * @cache: The cache we are manipulating
6553 * @ram_bytes: The number of bytes of file content, and will be same to
6554 * @num_bytes except for the compress path.
6555 * @num_bytes: The number of bytes in question
6556 * @delalloc: The blocks are allocated for the delalloc write
6558 * This is called by the allocator when it reserves space. If this is a
6559 * reservation and the block group has become read only we cannot make the
6560 * reservation and return -EAGAIN, otherwise this function always succeeds.
6562 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache,
6563 u64 ram_bytes, u64 num_bytes, int delalloc)
6565 struct btrfs_space_info *space_info = cache->space_info;
6568 spin_lock(&space_info->lock);
6569 spin_lock(&cache->lock);
6573 cache->reserved += num_bytes;
6574 space_info->bytes_reserved += num_bytes;
6576 trace_btrfs_space_reservation(cache->fs_info,
6577 "space_info", space_info->flags,
6579 space_info->bytes_may_use -= ram_bytes;
6581 cache->delalloc_bytes += num_bytes;
6583 spin_unlock(&cache->lock);
6584 spin_unlock(&space_info->lock);
6589 * btrfs_free_reserved_bytes - update the block_group and space info counters
6590 * @cache: The cache we are manipulating
6591 * @num_bytes: The number of bytes in question
6592 * @delalloc: The blocks are allocated for the delalloc write
6594 * This is called by somebody who is freeing space that was never actually used
6595 * on disk. For example if you reserve some space for a new leaf in transaction
6596 * A and before transaction A commits you free that leaf, you call this with
6597 * reserve set to 0 in order to clear the reservation.
6600 static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache,
6601 u64 num_bytes, int delalloc)
6603 struct btrfs_space_info *space_info = cache->space_info;
6606 spin_lock(&space_info->lock);
6607 spin_lock(&cache->lock);
6609 space_info->bytes_readonly += num_bytes;
6610 cache->reserved -= num_bytes;
6611 space_info->bytes_reserved -= num_bytes;
6614 cache->delalloc_bytes -= num_bytes;
6615 spin_unlock(&cache->lock);
6616 spin_unlock(&space_info->lock);
6619 void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info)
6621 struct btrfs_caching_control *next;
6622 struct btrfs_caching_control *caching_ctl;
6623 struct btrfs_block_group_cache *cache;
6625 down_write(&fs_info->commit_root_sem);
6627 list_for_each_entry_safe(caching_ctl, next,
6628 &fs_info->caching_block_groups, list) {
6629 cache = caching_ctl->block_group;
6630 if (block_group_cache_done(cache)) {
6631 cache->last_byte_to_unpin = (u64)-1;
6632 list_del_init(&caching_ctl->list);
6633 put_caching_control(caching_ctl);
6635 cache->last_byte_to_unpin = caching_ctl->progress;
6639 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6640 fs_info->pinned_extents = &fs_info->freed_extents[1];
6642 fs_info->pinned_extents = &fs_info->freed_extents[0];
6644 up_write(&fs_info->commit_root_sem);
6646 update_global_block_rsv(fs_info);
6650 * Returns the free cluster for the given space info and sets empty_cluster to
6651 * what it should be based on the mount options.
6653 static struct btrfs_free_cluster *
6654 fetch_cluster_info(struct btrfs_fs_info *fs_info,
6655 struct btrfs_space_info *space_info, u64 *empty_cluster)
6657 struct btrfs_free_cluster *ret = NULL;
6660 if (btrfs_mixed_space_info(space_info))
6663 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
6664 ret = &fs_info->meta_alloc_cluster;
6665 if (btrfs_test_opt(fs_info, SSD))
6666 *empty_cluster = SZ_2M;
6668 *empty_cluster = SZ_64K;
6669 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
6670 btrfs_test_opt(fs_info, SSD_SPREAD)) {
6671 *empty_cluster = SZ_2M;
6672 ret = &fs_info->data_alloc_cluster;
6678 static int unpin_extent_range(struct btrfs_fs_info *fs_info,
6680 const bool return_free_space)
6682 struct btrfs_block_group_cache *cache = NULL;
6683 struct btrfs_space_info *space_info;
6684 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6685 struct btrfs_free_cluster *cluster = NULL;
6687 u64 total_unpinned = 0;
6688 u64 empty_cluster = 0;
6691 while (start <= end) {
6694 start >= cache->key.objectid + cache->key.offset) {
6696 btrfs_put_block_group(cache);
6698 cache = btrfs_lookup_block_group(fs_info, start);
6699 BUG_ON(!cache); /* Logic error */
6701 cluster = fetch_cluster_info(fs_info,
6704 empty_cluster <<= 1;
6707 len = cache->key.objectid + cache->key.offset - start;
6708 len = min(len, end + 1 - start);
6710 if (start < cache->last_byte_to_unpin) {
6711 len = min(len, cache->last_byte_to_unpin - start);
6712 if (return_free_space)
6713 btrfs_add_free_space(cache, start, len);
6717 total_unpinned += len;
6718 space_info = cache->space_info;
6721 * If this space cluster has been marked as fragmented and we've
6722 * unpinned enough in this block group to potentially allow a
6723 * cluster to be created inside of it go ahead and clear the
6726 if (cluster && cluster->fragmented &&
6727 total_unpinned > empty_cluster) {
6728 spin_lock(&cluster->lock);
6729 cluster->fragmented = 0;
6730 spin_unlock(&cluster->lock);
6733 spin_lock(&space_info->lock);
6734 spin_lock(&cache->lock);
6735 cache->pinned -= len;
6736 space_info->bytes_pinned -= len;
6738 trace_btrfs_space_reservation(fs_info, "pinned",
6739 space_info->flags, len, 0);
6740 space_info->max_extent_size = 0;
6741 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6743 space_info->bytes_readonly += len;
6746 spin_unlock(&cache->lock);
6747 if (!readonly && return_free_space &&
6748 global_rsv->space_info == space_info) {
6751 spin_lock(&global_rsv->lock);
6752 if (!global_rsv->full) {
6753 to_add = min(len, global_rsv->size -
6754 global_rsv->reserved);
6755 global_rsv->reserved += to_add;
6756 space_info->bytes_may_use += to_add;
6757 if (global_rsv->reserved >= global_rsv->size)
6758 global_rsv->full = 1;
6759 trace_btrfs_space_reservation(fs_info,
6765 spin_unlock(&global_rsv->lock);
6766 /* Add to any tickets we may have */
6768 space_info_add_new_bytes(fs_info, space_info,
6771 spin_unlock(&space_info->lock);
6775 btrfs_put_block_group(cache);
6779 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
6780 struct btrfs_fs_info *fs_info)
6782 struct btrfs_block_group_cache *block_group, *tmp;
6783 struct list_head *deleted_bgs;
6784 struct extent_io_tree *unpin;
6789 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6790 unpin = &fs_info->freed_extents[1];
6792 unpin = &fs_info->freed_extents[0];
6794 while (!trans->aborted) {
6795 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6796 ret = find_first_extent_bit(unpin, 0, &start, &end,
6797 EXTENT_DIRTY, NULL);
6799 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6803 if (btrfs_test_opt(fs_info, DISCARD))
6804 ret = btrfs_discard_extent(fs_info, start,
6805 end + 1 - start, NULL);
6807 clear_extent_dirty(unpin, start, end);
6808 unpin_extent_range(fs_info, start, end, true);
6809 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6814 * Transaction is finished. We don't need the lock anymore. We
6815 * do need to clean up the block groups in case of a transaction
6818 deleted_bgs = &trans->transaction->deleted_bgs;
6819 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6823 if (!trans->aborted)
6824 ret = btrfs_discard_extent(fs_info,
6825 block_group->key.objectid,
6826 block_group->key.offset,
6829 list_del_init(&block_group->bg_list);
6830 btrfs_put_block_group_trimming(block_group);
6831 btrfs_put_block_group(block_group);
6834 const char *errstr = btrfs_decode_error(ret);
6836 "discard failed while removing blockgroup: errno=%d %s",
6844 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6845 struct btrfs_fs_info *info,
6846 struct btrfs_delayed_ref_node *node, u64 parent,
6847 u64 root_objectid, u64 owner_objectid,
6848 u64 owner_offset, int refs_to_drop,
6849 struct btrfs_delayed_extent_op *extent_op)
6851 struct btrfs_key key;
6852 struct btrfs_path *path;
6853 struct btrfs_root *extent_root = info->extent_root;
6854 struct extent_buffer *leaf;
6855 struct btrfs_extent_item *ei;
6856 struct btrfs_extent_inline_ref *iref;
6859 int extent_slot = 0;
6860 int found_extent = 0;
6864 u64 bytenr = node->bytenr;
6865 u64 num_bytes = node->num_bytes;
6867 bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
6869 path = btrfs_alloc_path();
6873 path->reada = READA_FORWARD;
6874 path->leave_spinning = 1;
6876 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6877 BUG_ON(!is_data && refs_to_drop != 1);
6880 skinny_metadata = false;
6882 ret = lookup_extent_backref(trans, info, path, &iref,
6883 bytenr, num_bytes, parent,
6884 root_objectid, owner_objectid,
6887 extent_slot = path->slots[0];
6888 while (extent_slot >= 0) {
6889 btrfs_item_key_to_cpu(path->nodes[0], &key,
6891 if (key.objectid != bytenr)
6893 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6894 key.offset == num_bytes) {
6898 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6899 key.offset == owner_objectid) {
6903 if (path->slots[0] - extent_slot > 5)
6907 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6908 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6909 if (found_extent && item_size < sizeof(*ei))
6912 if (!found_extent) {
6914 ret = remove_extent_backref(trans, info, path, NULL,
6916 is_data, &last_ref);
6918 btrfs_abort_transaction(trans, ret);
6921 btrfs_release_path(path);
6922 path->leave_spinning = 1;
6924 key.objectid = bytenr;
6925 key.type = BTRFS_EXTENT_ITEM_KEY;
6926 key.offset = num_bytes;
6928 if (!is_data && skinny_metadata) {
6929 key.type = BTRFS_METADATA_ITEM_KEY;
6930 key.offset = owner_objectid;
6933 ret = btrfs_search_slot(trans, extent_root,
6935 if (ret > 0 && skinny_metadata && path->slots[0]) {
6937 * Couldn't find our skinny metadata item,
6938 * see if we have ye olde extent item.
6941 btrfs_item_key_to_cpu(path->nodes[0], &key,
6943 if (key.objectid == bytenr &&
6944 key.type == BTRFS_EXTENT_ITEM_KEY &&
6945 key.offset == num_bytes)
6949 if (ret > 0 && skinny_metadata) {
6950 skinny_metadata = false;
6951 key.objectid = bytenr;
6952 key.type = BTRFS_EXTENT_ITEM_KEY;
6953 key.offset = num_bytes;
6954 btrfs_release_path(path);
6955 ret = btrfs_search_slot(trans, extent_root,
6961 "umm, got %d back from search, was looking for %llu",
6964 btrfs_print_leaf(path->nodes[0]);
6967 btrfs_abort_transaction(trans, ret);
6970 extent_slot = path->slots[0];
6972 } else if (WARN_ON(ret == -ENOENT)) {
6973 btrfs_print_leaf(path->nodes[0]);
6975 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6976 bytenr, parent, root_objectid, owner_objectid,
6978 btrfs_abort_transaction(trans, ret);
6981 btrfs_abort_transaction(trans, ret);
6985 leaf = path->nodes[0];
6986 item_size = btrfs_item_size_nr(leaf, extent_slot);
6987 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6988 if (item_size < sizeof(*ei)) {
6989 BUG_ON(found_extent || extent_slot != path->slots[0]);
6990 ret = convert_extent_item_v0(trans, info, path, owner_objectid,
6993 btrfs_abort_transaction(trans, ret);
6997 btrfs_release_path(path);
6998 path->leave_spinning = 1;
7000 key.objectid = bytenr;
7001 key.type = BTRFS_EXTENT_ITEM_KEY;
7002 key.offset = num_bytes;
7004 ret = btrfs_search_slot(trans, extent_root, &key, path,
7008 "umm, got %d back from search, was looking for %llu",
7010 btrfs_print_leaf(path->nodes[0]);
7013 btrfs_abort_transaction(trans, ret);
7017 extent_slot = path->slots[0];
7018 leaf = path->nodes[0];
7019 item_size = btrfs_item_size_nr(leaf, extent_slot);
7022 BUG_ON(item_size < sizeof(*ei));
7023 ei = btrfs_item_ptr(leaf, extent_slot,
7024 struct btrfs_extent_item);
7025 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
7026 key.type == BTRFS_EXTENT_ITEM_KEY) {
7027 struct btrfs_tree_block_info *bi;
7028 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
7029 bi = (struct btrfs_tree_block_info *)(ei + 1);
7030 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
7033 refs = btrfs_extent_refs(leaf, ei);
7034 if (refs < refs_to_drop) {
7036 "trying to drop %d refs but we only have %Lu for bytenr %Lu",
7037 refs_to_drop, refs, bytenr);
7039 btrfs_abort_transaction(trans, ret);
7042 refs -= refs_to_drop;
7046 __run_delayed_extent_op(extent_op, leaf, ei);
7048 * In the case of inline back ref, reference count will
7049 * be updated by remove_extent_backref
7052 BUG_ON(!found_extent);
7054 btrfs_set_extent_refs(leaf, ei, refs);
7055 btrfs_mark_buffer_dirty(leaf);
7058 ret = remove_extent_backref(trans, info, path,
7060 is_data, &last_ref);
7062 btrfs_abort_transaction(trans, ret);
7068 BUG_ON(is_data && refs_to_drop !=
7069 extent_data_ref_count(path, iref));
7071 BUG_ON(path->slots[0] != extent_slot);
7073 BUG_ON(path->slots[0] != extent_slot + 1);
7074 path->slots[0] = extent_slot;
7080 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
7083 btrfs_abort_transaction(trans, ret);
7086 btrfs_release_path(path);
7089 ret = btrfs_del_csums(trans, info, bytenr, num_bytes);
7091 btrfs_abort_transaction(trans, ret);
7096 ret = add_to_free_space_tree(trans, info, bytenr, num_bytes);
7098 btrfs_abort_transaction(trans, ret);
7102 ret = update_block_group(trans, info, bytenr, num_bytes, 0);
7104 btrfs_abort_transaction(trans, ret);
7108 btrfs_release_path(path);
7111 btrfs_free_path(path);
7116 * when we free an block, it is possible (and likely) that we free the last
7117 * delayed ref for that extent as well. This searches the delayed ref tree for
7118 * a given extent, and if there are no other delayed refs to be processed, it
7119 * removes it from the tree.
7121 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
7124 struct btrfs_delayed_ref_head *head;
7125 struct btrfs_delayed_ref_root *delayed_refs;
7128 delayed_refs = &trans->transaction->delayed_refs;
7129 spin_lock(&delayed_refs->lock);
7130 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
7132 goto out_delayed_unlock;
7134 spin_lock(&head->lock);
7135 if (!RB_EMPTY_ROOT(&head->ref_tree))
7138 if (head->extent_op) {
7139 if (!head->must_insert_reserved)
7141 btrfs_free_delayed_extent_op(head->extent_op);
7142 head->extent_op = NULL;
7146 * waiting for the lock here would deadlock. If someone else has it
7147 * locked they are already in the process of dropping it anyway
7149 if (!mutex_trylock(&head->mutex))
7153 * at this point we have a head with no other entries. Go
7154 * ahead and process it.
7156 rb_erase(&head->href_node, &delayed_refs->href_root);
7157 RB_CLEAR_NODE(&head->href_node);
7158 atomic_dec(&delayed_refs->num_entries);
7161 * we don't take a ref on the node because we're removing it from the
7162 * tree, so we just steal the ref the tree was holding.
7164 delayed_refs->num_heads--;
7165 if (head->processing == 0)
7166 delayed_refs->num_heads_ready--;
7167 head->processing = 0;
7168 spin_unlock(&head->lock);
7169 spin_unlock(&delayed_refs->lock);
7171 BUG_ON(head->extent_op);
7172 if (head->must_insert_reserved)
7175 mutex_unlock(&head->mutex);
7176 btrfs_put_delayed_ref_head(head);
7179 spin_unlock(&head->lock);
7182 spin_unlock(&delayed_refs->lock);
7186 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
7187 struct btrfs_root *root,
7188 struct extent_buffer *buf,
7189 u64 parent, int last_ref)
7191 struct btrfs_fs_info *fs_info = root->fs_info;
7195 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7196 int old_ref_mod, new_ref_mod;
7198 btrfs_ref_tree_mod(root, buf->start, buf->len, parent,
7199 root->root_key.objectid,
7200 btrfs_header_level(buf), 0,
7201 BTRFS_DROP_DELAYED_REF);
7202 ret = btrfs_add_delayed_tree_ref(fs_info, trans, buf->start,
7204 root->root_key.objectid,
7205 btrfs_header_level(buf),
7206 BTRFS_DROP_DELAYED_REF, NULL,
7207 &old_ref_mod, &new_ref_mod);
7208 BUG_ON(ret); /* -ENOMEM */
7209 pin = old_ref_mod >= 0 && new_ref_mod < 0;
7212 if (last_ref && btrfs_header_generation(buf) == trans->transid) {
7213 struct btrfs_block_group_cache *cache;
7215 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7216 ret = check_ref_cleanup(trans, buf->start);
7222 cache = btrfs_lookup_block_group(fs_info, buf->start);
7224 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
7225 pin_down_extent(fs_info, cache, buf->start,
7227 btrfs_put_block_group(cache);
7231 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
7233 btrfs_add_free_space(cache, buf->start, buf->len);
7234 btrfs_free_reserved_bytes(cache, buf->len, 0);
7235 btrfs_put_block_group(cache);
7236 trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
7240 add_pinned_bytes(fs_info, buf->len, btrfs_header_level(buf),
7241 root->root_key.objectid);
7245 * Deleting the buffer, clear the corrupt flag since it doesn't
7248 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
7252 /* Can return -ENOMEM */
7253 int btrfs_free_extent(struct btrfs_trans_handle *trans,
7254 struct btrfs_root *root,
7255 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
7256 u64 owner, u64 offset)
7258 struct btrfs_fs_info *fs_info = root->fs_info;
7259 int old_ref_mod, new_ref_mod;
7262 if (btrfs_is_testing(fs_info))
7265 if (root_objectid != BTRFS_TREE_LOG_OBJECTID)
7266 btrfs_ref_tree_mod(root, bytenr, num_bytes, parent,
7267 root_objectid, owner, offset,
7268 BTRFS_DROP_DELAYED_REF);
7271 * tree log blocks never actually go into the extent allocation
7272 * tree, just update pinning info and exit early.
7274 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
7275 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
7276 /* unlocks the pinned mutex */
7277 btrfs_pin_extent(fs_info, bytenr, num_bytes, 1);
7278 old_ref_mod = new_ref_mod = 0;
7280 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
7281 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
7283 root_objectid, (int)owner,
7284 BTRFS_DROP_DELAYED_REF, NULL,
7285 &old_ref_mod, &new_ref_mod);
7287 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
7289 root_objectid, owner, offset,
7290 0, BTRFS_DROP_DELAYED_REF,
7291 &old_ref_mod, &new_ref_mod);
7294 if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0)
7295 add_pinned_bytes(fs_info, num_bytes, owner, root_objectid);
7301 * when we wait for progress in the block group caching, its because
7302 * our allocation attempt failed at least once. So, we must sleep
7303 * and let some progress happen before we try again.
7305 * This function will sleep at least once waiting for new free space to
7306 * show up, and then it will check the block group free space numbers
7307 * for our min num_bytes. Another option is to have it go ahead
7308 * and look in the rbtree for a free extent of a given size, but this
7311 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7312 * any of the information in this block group.
7314 static noinline void
7315 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
7318 struct btrfs_caching_control *caching_ctl;
7320 caching_ctl = get_caching_control(cache);
7324 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
7325 (cache->free_space_ctl->free_space >= num_bytes));
7327 put_caching_control(caching_ctl);
7331 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
7333 struct btrfs_caching_control *caching_ctl;
7336 caching_ctl = get_caching_control(cache);
7338 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
7340 wait_event(caching_ctl->wait, block_group_cache_done(cache));
7341 if (cache->cached == BTRFS_CACHE_ERROR)
7343 put_caching_control(caching_ctl);
7347 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
7348 [BTRFS_RAID_RAID10] = "raid10",
7349 [BTRFS_RAID_RAID1] = "raid1",
7350 [BTRFS_RAID_DUP] = "dup",
7351 [BTRFS_RAID_RAID0] = "raid0",
7352 [BTRFS_RAID_SINGLE] = "single",
7353 [BTRFS_RAID_RAID5] = "raid5",
7354 [BTRFS_RAID_RAID6] = "raid6",
7357 static const char *get_raid_name(enum btrfs_raid_types type)
7359 if (type >= BTRFS_NR_RAID_TYPES)
7362 return btrfs_raid_type_names[type];
7365 enum btrfs_loop_type {
7366 LOOP_CACHING_NOWAIT = 0,
7367 LOOP_CACHING_WAIT = 1,
7368 LOOP_ALLOC_CHUNK = 2,
7369 LOOP_NO_EMPTY_SIZE = 3,
7373 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
7377 down_read(&cache->data_rwsem);
7381 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
7384 btrfs_get_block_group(cache);
7386 down_read(&cache->data_rwsem);
7389 static struct btrfs_block_group_cache *
7390 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
7391 struct btrfs_free_cluster *cluster,
7394 struct btrfs_block_group_cache *used_bg = NULL;
7396 spin_lock(&cluster->refill_lock);
7398 used_bg = cluster->block_group;
7402 if (used_bg == block_group)
7405 btrfs_get_block_group(used_bg);
7410 if (down_read_trylock(&used_bg->data_rwsem))
7413 spin_unlock(&cluster->refill_lock);
7415 /* We should only have one-level nested. */
7416 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
7418 spin_lock(&cluster->refill_lock);
7419 if (used_bg == cluster->block_group)
7422 up_read(&used_bg->data_rwsem);
7423 btrfs_put_block_group(used_bg);
7428 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
7432 up_read(&cache->data_rwsem);
7433 btrfs_put_block_group(cache);
7437 * walks the btree of allocated extents and find a hole of a given size.
7438 * The key ins is changed to record the hole:
7439 * ins->objectid == start position
7440 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7441 * ins->offset == the size of the hole.
7442 * Any available blocks before search_start are skipped.
7444 * If there is no suitable free space, we will record the max size of
7445 * the free space extent currently.
7447 static noinline int find_free_extent(struct btrfs_fs_info *fs_info,
7448 u64 ram_bytes, u64 num_bytes, u64 empty_size,
7449 u64 hint_byte, struct btrfs_key *ins,
7450 u64 flags, int delalloc)
7453 struct btrfs_root *root = fs_info->extent_root;
7454 struct btrfs_free_cluster *last_ptr = NULL;
7455 struct btrfs_block_group_cache *block_group = NULL;
7456 u64 search_start = 0;
7457 u64 max_extent_size = 0;
7458 u64 empty_cluster = 0;
7459 struct btrfs_space_info *space_info;
7461 int index = btrfs_bg_flags_to_raid_index(flags);
7462 bool failed_cluster_refill = false;
7463 bool failed_alloc = false;
7464 bool use_cluster = true;
7465 bool have_caching_bg = false;
7466 bool orig_have_caching_bg = false;
7467 bool full_search = false;
7469 WARN_ON(num_bytes < fs_info->sectorsize);
7470 ins->type = BTRFS_EXTENT_ITEM_KEY;
7474 trace_find_free_extent(fs_info, num_bytes, empty_size, flags);
7476 space_info = __find_space_info(fs_info, flags);
7478 btrfs_err(fs_info, "No space info for %llu", flags);
7483 * If our free space is heavily fragmented we may not be able to make
7484 * big contiguous allocations, so instead of doing the expensive search
7485 * for free space, simply return ENOSPC with our max_extent_size so we
7486 * can go ahead and search for a more manageable chunk.
7488 * If our max_extent_size is large enough for our allocation simply
7489 * disable clustering since we will likely not be able to find enough
7490 * space to create a cluster and induce latency trying.
7492 if (unlikely(space_info->max_extent_size)) {
7493 spin_lock(&space_info->lock);
7494 if (space_info->max_extent_size &&
7495 num_bytes > space_info->max_extent_size) {
7496 ins->offset = space_info->max_extent_size;
7497 spin_unlock(&space_info->lock);
7499 } else if (space_info->max_extent_size) {
7500 use_cluster = false;
7502 spin_unlock(&space_info->lock);
7505 last_ptr = fetch_cluster_info(fs_info, space_info, &empty_cluster);
7507 spin_lock(&last_ptr->lock);
7508 if (last_ptr->block_group)
7509 hint_byte = last_ptr->window_start;
7510 if (last_ptr->fragmented) {
7512 * We still set window_start so we can keep track of the
7513 * last place we found an allocation to try and save
7516 hint_byte = last_ptr->window_start;
7517 use_cluster = false;
7519 spin_unlock(&last_ptr->lock);
7522 search_start = max(search_start, first_logical_byte(fs_info, 0));
7523 search_start = max(search_start, hint_byte);
7524 if (search_start == hint_byte) {
7525 block_group = btrfs_lookup_block_group(fs_info, search_start);
7527 * we don't want to use the block group if it doesn't match our
7528 * allocation bits, or if its not cached.
7530 * However if we are re-searching with an ideal block group
7531 * picked out then we don't care that the block group is cached.
7533 if (block_group && block_group_bits(block_group, flags) &&
7534 block_group->cached != BTRFS_CACHE_NO) {
7535 down_read(&space_info->groups_sem);
7536 if (list_empty(&block_group->list) ||
7539 * someone is removing this block group,
7540 * we can't jump into the have_block_group
7541 * target because our list pointers are not
7544 btrfs_put_block_group(block_group);
7545 up_read(&space_info->groups_sem);
7547 index = btrfs_bg_flags_to_raid_index(
7548 block_group->flags);
7549 btrfs_lock_block_group(block_group, delalloc);
7550 goto have_block_group;
7552 } else if (block_group) {
7553 btrfs_put_block_group(block_group);
7557 have_caching_bg = false;
7558 if (index == 0 || index == btrfs_bg_flags_to_raid_index(flags))
7560 down_read(&space_info->groups_sem);
7561 list_for_each_entry(block_group, &space_info->block_groups[index],
7566 /* If the block group is read-only, we can skip it entirely. */
7567 if (unlikely(block_group->ro))
7570 btrfs_grab_block_group(block_group, delalloc);
7571 search_start = block_group->key.objectid;
7574 * this can happen if we end up cycling through all the
7575 * raid types, but we want to make sure we only allocate
7576 * for the proper type.
7578 if (!block_group_bits(block_group, flags)) {
7579 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7580 BTRFS_BLOCK_GROUP_RAID1 |
7581 BTRFS_BLOCK_GROUP_RAID5 |
7582 BTRFS_BLOCK_GROUP_RAID6 |
7583 BTRFS_BLOCK_GROUP_RAID10;
7586 * if they asked for extra copies and this block group
7587 * doesn't provide them, bail. This does allow us to
7588 * fill raid0 from raid1.
7590 if ((flags & extra) && !(block_group->flags & extra))
7595 cached = block_group_cache_done(block_group);
7596 if (unlikely(!cached)) {
7597 have_caching_bg = true;
7598 ret = cache_block_group(block_group, 0);
7603 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7607 * Ok we want to try and use the cluster allocator, so
7610 if (last_ptr && use_cluster) {
7611 struct btrfs_block_group_cache *used_block_group;
7612 unsigned long aligned_cluster;
7614 * the refill lock keeps out other
7615 * people trying to start a new cluster
7617 used_block_group = btrfs_lock_cluster(block_group,
7620 if (!used_block_group)
7621 goto refill_cluster;
7623 if (used_block_group != block_group &&
7624 (used_block_group->ro ||
7625 !block_group_bits(used_block_group, flags)))
7626 goto release_cluster;
7628 offset = btrfs_alloc_from_cluster(used_block_group,
7631 used_block_group->key.objectid,
7634 /* we have a block, we're done */
7635 spin_unlock(&last_ptr->refill_lock);
7636 trace_btrfs_reserve_extent_cluster(fs_info,
7638 search_start, num_bytes);
7639 if (used_block_group != block_group) {
7640 btrfs_release_block_group(block_group,
7642 block_group = used_block_group;
7647 WARN_ON(last_ptr->block_group != used_block_group);
7649 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7650 * set up a new clusters, so lets just skip it
7651 * and let the allocator find whatever block
7652 * it can find. If we reach this point, we
7653 * will have tried the cluster allocator
7654 * plenty of times and not have found
7655 * anything, so we are likely way too
7656 * fragmented for the clustering stuff to find
7659 * However, if the cluster is taken from the
7660 * current block group, release the cluster
7661 * first, so that we stand a better chance of
7662 * succeeding in the unclustered
7664 if (loop >= LOOP_NO_EMPTY_SIZE &&
7665 used_block_group != block_group) {
7666 spin_unlock(&last_ptr->refill_lock);
7667 btrfs_release_block_group(used_block_group,
7669 goto unclustered_alloc;
7673 * this cluster didn't work out, free it and
7676 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7678 if (used_block_group != block_group)
7679 btrfs_release_block_group(used_block_group,
7682 if (loop >= LOOP_NO_EMPTY_SIZE) {
7683 spin_unlock(&last_ptr->refill_lock);
7684 goto unclustered_alloc;
7687 aligned_cluster = max_t(unsigned long,
7688 empty_cluster + empty_size,
7689 block_group->full_stripe_len);
7691 /* allocate a cluster in this block group */
7692 ret = btrfs_find_space_cluster(fs_info, block_group,
7693 last_ptr, search_start,
7698 * now pull our allocation out of this
7701 offset = btrfs_alloc_from_cluster(block_group,
7707 /* we found one, proceed */
7708 spin_unlock(&last_ptr->refill_lock);
7709 trace_btrfs_reserve_extent_cluster(fs_info,
7710 block_group, search_start,
7714 } else if (!cached && loop > LOOP_CACHING_NOWAIT
7715 && !failed_cluster_refill) {
7716 spin_unlock(&last_ptr->refill_lock);
7718 failed_cluster_refill = true;
7719 wait_block_group_cache_progress(block_group,
7720 num_bytes + empty_cluster + empty_size);
7721 goto have_block_group;
7725 * at this point we either didn't find a cluster
7726 * or we weren't able to allocate a block from our
7727 * cluster. Free the cluster we've been trying
7728 * to use, and go to the next block group
7730 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7731 spin_unlock(&last_ptr->refill_lock);
7737 * We are doing an unclustered alloc, set the fragmented flag so
7738 * we don't bother trying to setup a cluster again until we get
7741 if (unlikely(last_ptr)) {
7742 spin_lock(&last_ptr->lock);
7743 last_ptr->fragmented = 1;
7744 spin_unlock(&last_ptr->lock);
7747 struct btrfs_free_space_ctl *ctl =
7748 block_group->free_space_ctl;
7750 spin_lock(&ctl->tree_lock);
7751 if (ctl->free_space <
7752 num_bytes + empty_cluster + empty_size) {
7753 if (ctl->free_space > max_extent_size)
7754 max_extent_size = ctl->free_space;
7755 spin_unlock(&ctl->tree_lock);
7758 spin_unlock(&ctl->tree_lock);
7761 offset = btrfs_find_space_for_alloc(block_group, search_start,
7762 num_bytes, empty_size,
7765 * If we didn't find a chunk, and we haven't failed on this
7766 * block group before, and this block group is in the middle of
7767 * caching and we are ok with waiting, then go ahead and wait
7768 * for progress to be made, and set failed_alloc to true.
7770 * If failed_alloc is true then we've already waited on this
7771 * block group once and should move on to the next block group.
7773 if (!offset && !failed_alloc && !cached &&
7774 loop > LOOP_CACHING_NOWAIT) {
7775 wait_block_group_cache_progress(block_group,
7776 num_bytes + empty_size);
7777 failed_alloc = true;
7778 goto have_block_group;
7779 } else if (!offset) {
7783 search_start = ALIGN(offset, fs_info->stripesize);
7785 /* move on to the next group */
7786 if (search_start + num_bytes >
7787 block_group->key.objectid + block_group->key.offset) {
7788 btrfs_add_free_space(block_group, offset, num_bytes);
7792 if (offset < search_start)
7793 btrfs_add_free_space(block_group, offset,
7794 search_start - offset);
7795 BUG_ON(offset > search_start);
7797 ret = btrfs_add_reserved_bytes(block_group, ram_bytes,
7798 num_bytes, delalloc);
7799 if (ret == -EAGAIN) {
7800 btrfs_add_free_space(block_group, offset, num_bytes);
7803 btrfs_inc_block_group_reservations(block_group);
7805 /* we are all good, lets return */
7806 ins->objectid = search_start;
7807 ins->offset = num_bytes;
7809 trace_btrfs_reserve_extent(fs_info, block_group,
7810 search_start, num_bytes);
7811 btrfs_release_block_group(block_group, delalloc);
7814 failed_cluster_refill = false;
7815 failed_alloc = false;
7816 BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) !=
7818 btrfs_release_block_group(block_group, delalloc);
7821 up_read(&space_info->groups_sem);
7823 if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg
7824 && !orig_have_caching_bg)
7825 orig_have_caching_bg = true;
7827 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7830 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7834 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7835 * caching kthreads as we move along
7836 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7837 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7838 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7841 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7843 if (loop == LOOP_CACHING_NOWAIT) {
7845 * We want to skip the LOOP_CACHING_WAIT step if we
7846 * don't have any uncached bgs and we've already done a
7847 * full search through.
7849 if (orig_have_caching_bg || !full_search)
7850 loop = LOOP_CACHING_WAIT;
7852 loop = LOOP_ALLOC_CHUNK;
7857 if (loop == LOOP_ALLOC_CHUNK) {
7858 struct btrfs_trans_handle *trans;
7861 trans = current->journal_info;
7865 trans = btrfs_join_transaction(root);
7867 if (IS_ERR(trans)) {
7868 ret = PTR_ERR(trans);
7872 ret = do_chunk_alloc(trans, fs_info, flags,
7876 * If we can't allocate a new chunk we've already looped
7877 * through at least once, move on to the NO_EMPTY_SIZE
7881 loop = LOOP_NO_EMPTY_SIZE;
7884 * Do not bail out on ENOSPC since we
7885 * can do more things.
7887 if (ret < 0 && ret != -ENOSPC)
7888 btrfs_abort_transaction(trans, ret);
7892 btrfs_end_transaction(trans);
7897 if (loop == LOOP_NO_EMPTY_SIZE) {
7899 * Don't loop again if we already have no empty_size and
7902 if (empty_size == 0 &&
7903 empty_cluster == 0) {
7912 } else if (!ins->objectid) {
7914 } else if (ins->objectid) {
7915 if (!use_cluster && last_ptr) {
7916 spin_lock(&last_ptr->lock);
7917 last_ptr->window_start = ins->objectid;
7918 spin_unlock(&last_ptr->lock);
7923 if (ret == -ENOSPC) {
7924 spin_lock(&space_info->lock);
7925 space_info->max_extent_size = max_extent_size;
7926 spin_unlock(&space_info->lock);
7927 ins->offset = max_extent_size;
7932 static void dump_space_info(struct btrfs_fs_info *fs_info,
7933 struct btrfs_space_info *info, u64 bytes,
7934 int dump_block_groups)
7936 struct btrfs_block_group_cache *cache;
7939 spin_lock(&info->lock);
7940 btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull",
7942 info->total_bytes - btrfs_space_info_used(info, true),
7943 info->full ? "" : "not ");
7945 "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu",
7946 info->total_bytes, info->bytes_used, info->bytes_pinned,
7947 info->bytes_reserved, info->bytes_may_use,
7948 info->bytes_readonly);
7949 spin_unlock(&info->lock);
7951 if (!dump_block_groups)
7954 down_read(&info->groups_sem);
7956 list_for_each_entry(cache, &info->block_groups[index], list) {
7957 spin_lock(&cache->lock);
7959 "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s",
7960 cache->key.objectid, cache->key.offset,
7961 btrfs_block_group_used(&cache->item), cache->pinned,
7962 cache->reserved, cache->ro ? "[readonly]" : "");
7963 btrfs_dump_free_space(cache, bytes);
7964 spin_unlock(&cache->lock);
7966 if (++index < BTRFS_NR_RAID_TYPES)
7968 up_read(&info->groups_sem);
7971 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
7972 u64 num_bytes, u64 min_alloc_size,
7973 u64 empty_size, u64 hint_byte,
7974 struct btrfs_key *ins, int is_data, int delalloc)
7976 struct btrfs_fs_info *fs_info = root->fs_info;
7977 bool final_tried = num_bytes == min_alloc_size;
7981 flags = get_alloc_profile_by_root(root, is_data);
7983 WARN_ON(num_bytes < fs_info->sectorsize);
7984 ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size,
7985 hint_byte, ins, flags, delalloc);
7986 if (!ret && !is_data) {
7987 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
7988 } else if (ret == -ENOSPC) {
7989 if (!final_tried && ins->offset) {
7990 num_bytes = min(num_bytes >> 1, ins->offset);
7991 num_bytes = round_down(num_bytes,
7992 fs_info->sectorsize);
7993 num_bytes = max(num_bytes, min_alloc_size);
7994 ram_bytes = num_bytes;
7995 if (num_bytes == min_alloc_size)
7998 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
7999 struct btrfs_space_info *sinfo;
8001 sinfo = __find_space_info(fs_info, flags);
8003 "allocation failed flags %llu, wanted %llu",
8006 dump_space_info(fs_info, sinfo, num_bytes, 1);
8013 static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8015 int pin, int delalloc)
8017 struct btrfs_block_group_cache *cache;
8020 cache = btrfs_lookup_block_group(fs_info, start);
8022 btrfs_err(fs_info, "Unable to find block group for %llu",
8028 pin_down_extent(fs_info, cache, start, len, 1);
8030 if (btrfs_test_opt(fs_info, DISCARD))
8031 ret = btrfs_discard_extent(fs_info, start, len, NULL);
8032 btrfs_add_free_space(cache, start, len);
8033 btrfs_free_reserved_bytes(cache, len, delalloc);
8034 trace_btrfs_reserved_extent_free(fs_info, start, len);
8037 btrfs_put_block_group(cache);
8041 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8042 u64 start, u64 len, int delalloc)
8044 return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc);
8047 int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info,
8050 return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0);
8053 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8054 struct btrfs_fs_info *fs_info,
8055 u64 parent, u64 root_objectid,
8056 u64 flags, u64 owner, u64 offset,
8057 struct btrfs_key *ins, int ref_mod)
8060 struct btrfs_extent_item *extent_item;
8061 struct btrfs_extent_inline_ref *iref;
8062 struct btrfs_path *path;
8063 struct extent_buffer *leaf;
8068 type = BTRFS_SHARED_DATA_REF_KEY;
8070 type = BTRFS_EXTENT_DATA_REF_KEY;
8072 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
8074 path = btrfs_alloc_path();
8078 path->leave_spinning = 1;
8079 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8082 btrfs_free_path(path);
8086 leaf = path->nodes[0];
8087 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8088 struct btrfs_extent_item);
8089 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
8090 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8091 btrfs_set_extent_flags(leaf, extent_item,
8092 flags | BTRFS_EXTENT_FLAG_DATA);
8094 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8095 btrfs_set_extent_inline_ref_type(leaf, iref, type);
8097 struct btrfs_shared_data_ref *ref;
8098 ref = (struct btrfs_shared_data_ref *)(iref + 1);
8099 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8100 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
8102 struct btrfs_extent_data_ref *ref;
8103 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
8104 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
8105 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
8106 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
8107 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
8110 btrfs_mark_buffer_dirty(path->nodes[0]);
8111 btrfs_free_path(path);
8113 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
8118 ret = update_block_group(trans, fs_info, ins->objectid, ins->offset, 1);
8119 if (ret) { /* -ENOENT, logic error */
8120 btrfs_err(fs_info, "update block group failed for %llu %llu",
8121 ins->objectid, ins->offset);
8124 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset);
8128 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
8129 struct btrfs_fs_info *fs_info,
8130 u64 parent, u64 root_objectid,
8131 u64 flags, struct btrfs_disk_key *key,
8132 int level, struct btrfs_key *ins)
8135 struct btrfs_extent_item *extent_item;
8136 struct btrfs_tree_block_info *block_info;
8137 struct btrfs_extent_inline_ref *iref;
8138 struct btrfs_path *path;
8139 struct extent_buffer *leaf;
8140 u32 size = sizeof(*extent_item) + sizeof(*iref);
8141 u64 num_bytes = ins->offset;
8142 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8144 if (!skinny_metadata)
8145 size += sizeof(*block_info);
8147 path = btrfs_alloc_path();
8149 btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid,
8154 path->leave_spinning = 1;
8155 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8158 btrfs_free_path(path);
8159 btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid,
8164 leaf = path->nodes[0];
8165 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8166 struct btrfs_extent_item);
8167 btrfs_set_extent_refs(leaf, extent_item, 1);
8168 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8169 btrfs_set_extent_flags(leaf, extent_item,
8170 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
8172 if (skinny_metadata) {
8173 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8174 num_bytes = fs_info->nodesize;
8176 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
8177 btrfs_set_tree_block_key(leaf, block_info, key);
8178 btrfs_set_tree_block_level(leaf, block_info, level);
8179 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
8183 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
8184 btrfs_set_extent_inline_ref_type(leaf, iref,
8185 BTRFS_SHARED_BLOCK_REF_KEY);
8186 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8188 btrfs_set_extent_inline_ref_type(leaf, iref,
8189 BTRFS_TREE_BLOCK_REF_KEY);
8190 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
8193 btrfs_mark_buffer_dirty(leaf);
8194 btrfs_free_path(path);
8196 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
8201 ret = update_block_group(trans, fs_info, ins->objectid,
8202 fs_info->nodesize, 1);
8203 if (ret) { /* -ENOENT, logic error */
8204 btrfs_err(fs_info, "update block group failed for %llu %llu",
8205 ins->objectid, ins->offset);
8209 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid,
8214 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8215 struct btrfs_root *root, u64 owner,
8216 u64 offset, u64 ram_bytes,
8217 struct btrfs_key *ins)
8219 struct btrfs_fs_info *fs_info = root->fs_info;
8222 BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
8224 btrfs_ref_tree_mod(root, ins->objectid, ins->offset, 0,
8225 root->root_key.objectid, owner, offset,
8226 BTRFS_ADD_DELAYED_EXTENT);
8228 ret = btrfs_add_delayed_data_ref(fs_info, trans, ins->objectid,
8230 root->root_key.objectid, owner,
8232 BTRFS_ADD_DELAYED_EXTENT, NULL, NULL);
8237 * this is used by the tree logging recovery code. It records that
8238 * an extent has been allocated and makes sure to clear the free
8239 * space cache bits as well
8241 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
8242 struct btrfs_fs_info *fs_info,
8243 u64 root_objectid, u64 owner, u64 offset,
8244 struct btrfs_key *ins)
8247 struct btrfs_block_group_cache *block_group;
8248 struct btrfs_space_info *space_info;
8251 * Mixed block groups will exclude before processing the log so we only
8252 * need to do the exclude dance if this fs isn't mixed.
8254 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
8255 ret = __exclude_logged_extent(fs_info, ins->objectid,
8261 block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
8265 space_info = block_group->space_info;
8266 spin_lock(&space_info->lock);
8267 spin_lock(&block_group->lock);
8268 space_info->bytes_reserved += ins->offset;
8269 block_group->reserved += ins->offset;
8270 spin_unlock(&block_group->lock);
8271 spin_unlock(&space_info->lock);
8273 ret = alloc_reserved_file_extent(trans, fs_info, 0, root_objectid,
8274 0, owner, offset, ins, 1);
8275 btrfs_put_block_group(block_group);
8279 static struct extent_buffer *
8280 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
8281 u64 bytenr, int level)
8283 struct btrfs_fs_info *fs_info = root->fs_info;
8284 struct extent_buffer *buf;
8286 buf = btrfs_find_create_tree_block(fs_info, bytenr);
8290 btrfs_set_header_generation(buf, trans->transid);
8291 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
8292 btrfs_tree_lock(buf);
8293 clean_tree_block(fs_info, buf);
8294 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
8296 btrfs_set_lock_blocking(buf);
8297 set_extent_buffer_uptodate(buf);
8299 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
8300 buf->log_index = root->log_transid % 2;
8302 * we allow two log transactions at a time, use different
8303 * EXENT bit to differentiate dirty pages.
8305 if (buf->log_index == 0)
8306 set_extent_dirty(&root->dirty_log_pages, buf->start,
8307 buf->start + buf->len - 1, GFP_NOFS);
8309 set_extent_new(&root->dirty_log_pages, buf->start,
8310 buf->start + buf->len - 1);
8312 buf->log_index = -1;
8313 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
8314 buf->start + buf->len - 1, GFP_NOFS);
8316 trans->dirty = true;
8317 /* this returns a buffer locked for blocking */
8321 static struct btrfs_block_rsv *
8322 use_block_rsv(struct btrfs_trans_handle *trans,
8323 struct btrfs_root *root, u32 blocksize)
8325 struct btrfs_fs_info *fs_info = root->fs_info;
8326 struct btrfs_block_rsv *block_rsv;
8327 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
8329 bool global_updated = false;
8331 block_rsv = get_block_rsv(trans, root);
8333 if (unlikely(block_rsv->size == 0))
8336 ret = block_rsv_use_bytes(block_rsv, blocksize);
8340 if (block_rsv->failfast)
8341 return ERR_PTR(ret);
8343 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
8344 global_updated = true;
8345 update_global_block_rsv(fs_info);
8349 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8350 static DEFINE_RATELIMIT_STATE(_rs,
8351 DEFAULT_RATELIMIT_INTERVAL * 10,
8352 /*DEFAULT_RATELIMIT_BURST*/ 1);
8353 if (__ratelimit(&_rs))
8355 "BTRFS: block rsv returned %d\n", ret);
8358 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
8359 BTRFS_RESERVE_NO_FLUSH);
8363 * If we couldn't reserve metadata bytes try and use some from
8364 * the global reserve if its space type is the same as the global
8367 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
8368 block_rsv->space_info == global_rsv->space_info) {
8369 ret = block_rsv_use_bytes(global_rsv, blocksize);
8373 return ERR_PTR(ret);
8376 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
8377 struct btrfs_block_rsv *block_rsv, u32 blocksize)
8379 block_rsv_add_bytes(block_rsv, blocksize, 0);
8380 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
8384 * finds a free extent and does all the dirty work required for allocation
8385 * returns the tree buffer or an ERR_PTR on error.
8387 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
8388 struct btrfs_root *root,
8389 u64 parent, u64 root_objectid,
8390 const struct btrfs_disk_key *key,
8391 int level, u64 hint,
8394 struct btrfs_fs_info *fs_info = root->fs_info;
8395 struct btrfs_key ins;
8396 struct btrfs_block_rsv *block_rsv;
8397 struct extent_buffer *buf;
8398 struct btrfs_delayed_extent_op *extent_op;
8401 u32 blocksize = fs_info->nodesize;
8402 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8404 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8405 if (btrfs_is_testing(fs_info)) {
8406 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
8409 root->alloc_bytenr += blocksize;
8414 block_rsv = use_block_rsv(trans, root, blocksize);
8415 if (IS_ERR(block_rsv))
8416 return ERR_CAST(block_rsv);
8418 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
8419 empty_size, hint, &ins, 0, 0);
8423 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
8426 goto out_free_reserved;
8429 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
8431 parent = ins.objectid;
8432 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
8436 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
8437 extent_op = btrfs_alloc_delayed_extent_op();
8443 memcpy(&extent_op->key, key, sizeof(extent_op->key));
8445 memset(&extent_op->key, 0, sizeof(extent_op->key));
8446 extent_op->flags_to_set = flags;
8447 extent_op->update_key = skinny_metadata ? false : true;
8448 extent_op->update_flags = true;
8449 extent_op->is_data = false;
8450 extent_op->level = level;
8452 btrfs_ref_tree_mod(root, ins.objectid, ins.offset, parent,
8453 root_objectid, level, 0,
8454 BTRFS_ADD_DELAYED_EXTENT);
8455 ret = btrfs_add_delayed_tree_ref(fs_info, trans, ins.objectid,
8457 root_objectid, level,
8458 BTRFS_ADD_DELAYED_EXTENT,
8459 extent_op, NULL, NULL);
8461 goto out_free_delayed;
8466 btrfs_free_delayed_extent_op(extent_op);
8468 free_extent_buffer(buf);
8470 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
8472 unuse_block_rsv(fs_info, block_rsv, blocksize);
8473 return ERR_PTR(ret);
8476 struct walk_control {
8477 u64 refs[BTRFS_MAX_LEVEL];
8478 u64 flags[BTRFS_MAX_LEVEL];
8479 struct btrfs_key update_progress;
8490 #define DROP_REFERENCE 1
8491 #define UPDATE_BACKREF 2
8493 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
8494 struct btrfs_root *root,
8495 struct walk_control *wc,
8496 struct btrfs_path *path)
8498 struct btrfs_fs_info *fs_info = root->fs_info;
8504 struct btrfs_key key;
8505 struct extent_buffer *eb;
8510 if (path->slots[wc->level] < wc->reada_slot) {
8511 wc->reada_count = wc->reada_count * 2 / 3;
8512 wc->reada_count = max(wc->reada_count, 2);
8514 wc->reada_count = wc->reada_count * 3 / 2;
8515 wc->reada_count = min_t(int, wc->reada_count,
8516 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
8519 eb = path->nodes[wc->level];
8520 nritems = btrfs_header_nritems(eb);
8522 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
8523 if (nread >= wc->reada_count)
8527 bytenr = btrfs_node_blockptr(eb, slot);
8528 generation = btrfs_node_ptr_generation(eb, slot);
8530 if (slot == path->slots[wc->level])
8533 if (wc->stage == UPDATE_BACKREF &&
8534 generation <= root->root_key.offset)
8537 /* We don't lock the tree block, it's OK to be racy here */
8538 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
8539 wc->level - 1, 1, &refs,
8541 /* We don't care about errors in readahead. */
8546 if (wc->stage == DROP_REFERENCE) {
8550 if (wc->level == 1 &&
8551 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8553 if (!wc->update_ref ||
8554 generation <= root->root_key.offset)
8556 btrfs_node_key_to_cpu(eb, &key, slot);
8557 ret = btrfs_comp_cpu_keys(&key,
8558 &wc->update_progress);
8562 if (wc->level == 1 &&
8563 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8567 readahead_tree_block(fs_info, bytenr);
8570 wc->reada_slot = slot;
8574 * helper to process tree block while walking down the tree.
8576 * when wc->stage == UPDATE_BACKREF, this function updates
8577 * back refs for pointers in the block.
8579 * NOTE: return value 1 means we should stop walking down.
8581 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8582 struct btrfs_root *root,
8583 struct btrfs_path *path,
8584 struct walk_control *wc, int lookup_info)
8586 struct btrfs_fs_info *fs_info = root->fs_info;
8587 int level = wc->level;
8588 struct extent_buffer *eb = path->nodes[level];
8589 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8592 if (wc->stage == UPDATE_BACKREF &&
8593 btrfs_header_owner(eb) != root->root_key.objectid)
8597 * when reference count of tree block is 1, it won't increase
8598 * again. once full backref flag is set, we never clear it.
8601 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8602 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8603 BUG_ON(!path->locks[level]);
8604 ret = btrfs_lookup_extent_info(trans, fs_info,
8605 eb->start, level, 1,
8608 BUG_ON(ret == -ENOMEM);
8611 BUG_ON(wc->refs[level] == 0);
8614 if (wc->stage == DROP_REFERENCE) {
8615 if (wc->refs[level] > 1)
8618 if (path->locks[level] && !wc->keep_locks) {
8619 btrfs_tree_unlock_rw(eb, path->locks[level]);
8620 path->locks[level] = 0;
8625 /* wc->stage == UPDATE_BACKREF */
8626 if (!(wc->flags[level] & flag)) {
8627 BUG_ON(!path->locks[level]);
8628 ret = btrfs_inc_ref(trans, root, eb, 1);
8629 BUG_ON(ret); /* -ENOMEM */
8630 ret = btrfs_dec_ref(trans, root, eb, 0);
8631 BUG_ON(ret); /* -ENOMEM */
8632 ret = btrfs_set_disk_extent_flags(trans, fs_info, eb->start,
8634 btrfs_header_level(eb), 0);
8635 BUG_ON(ret); /* -ENOMEM */
8636 wc->flags[level] |= flag;
8640 * the block is shared by multiple trees, so it's not good to
8641 * keep the tree lock
8643 if (path->locks[level] && level > 0) {
8644 btrfs_tree_unlock_rw(eb, path->locks[level]);
8645 path->locks[level] = 0;
8651 * helper to process tree block pointer.
8653 * when wc->stage == DROP_REFERENCE, this function checks
8654 * reference count of the block pointed to. if the block
8655 * is shared and we need update back refs for the subtree
8656 * rooted at the block, this function changes wc->stage to
8657 * UPDATE_BACKREF. if the block is shared and there is no
8658 * need to update back, this function drops the reference
8661 * NOTE: return value 1 means we should stop walking down.
8663 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8664 struct btrfs_root *root,
8665 struct btrfs_path *path,
8666 struct walk_control *wc, int *lookup_info)
8668 struct btrfs_fs_info *fs_info = root->fs_info;
8673 struct btrfs_key key;
8674 struct extent_buffer *next;
8675 int level = wc->level;
8678 bool need_account = false;
8680 generation = btrfs_node_ptr_generation(path->nodes[level],
8681 path->slots[level]);
8683 * if the lower level block was created before the snapshot
8684 * was created, we know there is no need to update back refs
8687 if (wc->stage == UPDATE_BACKREF &&
8688 generation <= root->root_key.offset) {
8693 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8694 blocksize = fs_info->nodesize;
8696 next = find_extent_buffer(fs_info, bytenr);
8698 next = btrfs_find_create_tree_block(fs_info, bytenr);
8700 return PTR_ERR(next);
8702 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8706 btrfs_tree_lock(next);
8707 btrfs_set_lock_blocking(next);
8709 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
8710 &wc->refs[level - 1],
8711 &wc->flags[level - 1]);
8715 if (unlikely(wc->refs[level - 1] == 0)) {
8716 btrfs_err(fs_info, "Missing references.");
8722 if (wc->stage == DROP_REFERENCE) {
8723 if (wc->refs[level - 1] > 1) {
8724 need_account = true;
8726 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8729 if (!wc->update_ref ||
8730 generation <= root->root_key.offset)
8733 btrfs_node_key_to_cpu(path->nodes[level], &key,
8734 path->slots[level]);
8735 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8739 wc->stage = UPDATE_BACKREF;
8740 wc->shared_level = level - 1;
8744 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8748 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8749 btrfs_tree_unlock(next);
8750 free_extent_buffer(next);
8756 if (reada && level == 1)
8757 reada_walk_down(trans, root, wc, path);
8758 next = read_tree_block(fs_info, bytenr, generation);
8760 return PTR_ERR(next);
8761 } else if (!extent_buffer_uptodate(next)) {
8762 free_extent_buffer(next);
8765 btrfs_tree_lock(next);
8766 btrfs_set_lock_blocking(next);
8770 ASSERT(level == btrfs_header_level(next));
8771 if (level != btrfs_header_level(next)) {
8772 btrfs_err(root->fs_info, "mismatched level");
8776 path->nodes[level] = next;
8777 path->slots[level] = 0;
8778 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8784 wc->refs[level - 1] = 0;
8785 wc->flags[level - 1] = 0;
8786 if (wc->stage == DROP_REFERENCE) {
8787 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8788 parent = path->nodes[level]->start;
8790 ASSERT(root->root_key.objectid ==
8791 btrfs_header_owner(path->nodes[level]));
8792 if (root->root_key.objectid !=
8793 btrfs_header_owner(path->nodes[level])) {
8794 btrfs_err(root->fs_info,
8795 "mismatched block owner");
8803 ret = btrfs_qgroup_trace_subtree(trans, root, next,
8804 generation, level - 1);
8806 btrfs_err_rl(fs_info,
8807 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
8811 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
8812 parent, root->root_key.objectid,
8822 btrfs_tree_unlock(next);
8823 free_extent_buffer(next);
8829 * helper to process tree block while walking up the tree.
8831 * when wc->stage == DROP_REFERENCE, this function drops
8832 * reference count on the block.
8834 * when wc->stage == UPDATE_BACKREF, this function changes
8835 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8836 * to UPDATE_BACKREF previously while processing the block.
8838 * NOTE: return value 1 means we should stop walking up.
8840 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8841 struct btrfs_root *root,
8842 struct btrfs_path *path,
8843 struct walk_control *wc)
8845 struct btrfs_fs_info *fs_info = root->fs_info;
8847 int level = wc->level;
8848 struct extent_buffer *eb = path->nodes[level];
8851 if (wc->stage == UPDATE_BACKREF) {
8852 BUG_ON(wc->shared_level < level);
8853 if (level < wc->shared_level)
8856 ret = find_next_key(path, level + 1, &wc->update_progress);
8860 wc->stage = DROP_REFERENCE;
8861 wc->shared_level = -1;
8862 path->slots[level] = 0;
8865 * check reference count again if the block isn't locked.
8866 * we should start walking down the tree again if reference
8869 if (!path->locks[level]) {
8871 btrfs_tree_lock(eb);
8872 btrfs_set_lock_blocking(eb);
8873 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8875 ret = btrfs_lookup_extent_info(trans, fs_info,
8876 eb->start, level, 1,
8880 btrfs_tree_unlock_rw(eb, path->locks[level]);
8881 path->locks[level] = 0;
8884 BUG_ON(wc->refs[level] == 0);
8885 if (wc->refs[level] == 1) {
8886 btrfs_tree_unlock_rw(eb, path->locks[level]);
8887 path->locks[level] = 0;
8893 /* wc->stage == DROP_REFERENCE */
8894 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8896 if (wc->refs[level] == 1) {
8898 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8899 ret = btrfs_dec_ref(trans, root, eb, 1);
8901 ret = btrfs_dec_ref(trans, root, eb, 0);
8902 BUG_ON(ret); /* -ENOMEM */
8903 ret = btrfs_qgroup_trace_leaf_items(trans, fs_info, eb);
8905 btrfs_err_rl(fs_info,
8906 "error %d accounting leaf items. Quota is out of sync, rescan required.",
8910 /* make block locked assertion in clean_tree_block happy */
8911 if (!path->locks[level] &&
8912 btrfs_header_generation(eb) == trans->transid) {
8913 btrfs_tree_lock(eb);
8914 btrfs_set_lock_blocking(eb);
8915 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8917 clean_tree_block(fs_info, eb);
8920 if (eb == root->node) {
8921 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8924 BUG_ON(root->root_key.objectid !=
8925 btrfs_header_owner(eb));
8927 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8928 parent = path->nodes[level + 1]->start;
8930 BUG_ON(root->root_key.objectid !=
8931 btrfs_header_owner(path->nodes[level + 1]));
8934 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8936 wc->refs[level] = 0;
8937 wc->flags[level] = 0;
8941 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8942 struct btrfs_root *root,
8943 struct btrfs_path *path,
8944 struct walk_control *wc)
8946 int level = wc->level;
8947 int lookup_info = 1;
8950 while (level >= 0) {
8951 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8958 if (path->slots[level] >=
8959 btrfs_header_nritems(path->nodes[level]))
8962 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8964 path->slots[level]++;
8973 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
8974 struct btrfs_root *root,
8975 struct btrfs_path *path,
8976 struct walk_control *wc, int max_level)
8978 int level = wc->level;
8981 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
8982 while (level < max_level && path->nodes[level]) {
8984 if (path->slots[level] + 1 <
8985 btrfs_header_nritems(path->nodes[level])) {
8986 path->slots[level]++;
8989 ret = walk_up_proc(trans, root, path, wc);
8993 if (path->locks[level]) {
8994 btrfs_tree_unlock_rw(path->nodes[level],
8995 path->locks[level]);
8996 path->locks[level] = 0;
8998 free_extent_buffer(path->nodes[level]);
8999 path->nodes[level] = NULL;
9007 * drop a subvolume tree.
9009 * this function traverses the tree freeing any blocks that only
9010 * referenced by the tree.
9012 * when a shared tree block is found. this function decreases its
9013 * reference count by one. if update_ref is true, this function
9014 * also make sure backrefs for the shared block and all lower level
9015 * blocks are properly updated.
9017 * If called with for_reloc == 0, may exit early with -EAGAIN
9019 int btrfs_drop_snapshot(struct btrfs_root *root,
9020 struct btrfs_block_rsv *block_rsv, int update_ref,
9023 struct btrfs_fs_info *fs_info = root->fs_info;
9024 struct btrfs_path *path;
9025 struct btrfs_trans_handle *trans;
9026 struct btrfs_root *tree_root = fs_info->tree_root;
9027 struct btrfs_root_item *root_item = &root->root_item;
9028 struct walk_control *wc;
9029 struct btrfs_key key;
9033 bool root_dropped = false;
9035 btrfs_debug(fs_info, "Drop subvolume %llu", root->objectid);
9037 path = btrfs_alloc_path();
9043 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9045 btrfs_free_path(path);
9050 trans = btrfs_start_transaction(tree_root, 0);
9051 if (IS_ERR(trans)) {
9052 err = PTR_ERR(trans);
9057 trans->block_rsv = block_rsv;
9059 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
9060 level = btrfs_header_level(root->node);
9061 path->nodes[level] = btrfs_lock_root_node(root);
9062 btrfs_set_lock_blocking(path->nodes[level]);
9063 path->slots[level] = 0;
9064 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9065 memset(&wc->update_progress, 0,
9066 sizeof(wc->update_progress));
9068 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
9069 memcpy(&wc->update_progress, &key,
9070 sizeof(wc->update_progress));
9072 level = root_item->drop_level;
9074 path->lowest_level = level;
9075 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
9076 path->lowest_level = 0;
9084 * unlock our path, this is safe because only this
9085 * function is allowed to delete this snapshot
9087 btrfs_unlock_up_safe(path, 0);
9089 level = btrfs_header_level(root->node);
9091 btrfs_tree_lock(path->nodes[level]);
9092 btrfs_set_lock_blocking(path->nodes[level]);
9093 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9095 ret = btrfs_lookup_extent_info(trans, fs_info,
9096 path->nodes[level]->start,
9097 level, 1, &wc->refs[level],
9103 BUG_ON(wc->refs[level] == 0);
9105 if (level == root_item->drop_level)
9108 btrfs_tree_unlock(path->nodes[level]);
9109 path->locks[level] = 0;
9110 WARN_ON(wc->refs[level] != 1);
9116 wc->shared_level = -1;
9117 wc->stage = DROP_REFERENCE;
9118 wc->update_ref = update_ref;
9120 wc->for_reloc = for_reloc;
9121 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9125 ret = walk_down_tree(trans, root, path, wc);
9131 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
9138 BUG_ON(wc->stage != DROP_REFERENCE);
9142 if (wc->stage == DROP_REFERENCE) {
9144 btrfs_node_key(path->nodes[level],
9145 &root_item->drop_progress,
9146 path->slots[level]);
9147 root_item->drop_level = level;
9150 BUG_ON(wc->level == 0);
9151 if (btrfs_should_end_transaction(trans) ||
9152 (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
9153 ret = btrfs_update_root(trans, tree_root,
9157 btrfs_abort_transaction(trans, ret);
9162 btrfs_end_transaction_throttle(trans);
9163 if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
9164 btrfs_debug(fs_info,
9165 "drop snapshot early exit");
9170 trans = btrfs_start_transaction(tree_root, 0);
9171 if (IS_ERR(trans)) {
9172 err = PTR_ERR(trans);
9176 trans->block_rsv = block_rsv;
9179 btrfs_release_path(path);
9183 ret = btrfs_del_root(trans, fs_info, &root->root_key);
9185 btrfs_abort_transaction(trans, ret);
9190 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
9191 ret = btrfs_find_root(tree_root, &root->root_key, path,
9194 btrfs_abort_transaction(trans, ret);
9197 } else if (ret > 0) {
9198 /* if we fail to delete the orphan item this time
9199 * around, it'll get picked up the next time.
9201 * The most common failure here is just -ENOENT.
9203 btrfs_del_orphan_item(trans, tree_root,
9204 root->root_key.objectid);
9208 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
9209 btrfs_add_dropped_root(trans, root);
9211 free_extent_buffer(root->node);
9212 free_extent_buffer(root->commit_root);
9213 btrfs_put_fs_root(root);
9215 root_dropped = true;
9217 btrfs_end_transaction_throttle(trans);
9220 btrfs_free_path(path);
9223 * So if we need to stop dropping the snapshot for whatever reason we
9224 * need to make sure to add it back to the dead root list so that we
9225 * keep trying to do the work later. This also cleans up roots if we
9226 * don't have it in the radix (like when we recover after a power fail
9227 * or unmount) so we don't leak memory.
9229 if (!for_reloc && !root_dropped)
9230 btrfs_add_dead_root(root);
9231 if (err && err != -EAGAIN)
9232 btrfs_handle_fs_error(fs_info, err, NULL);
9237 * drop subtree rooted at tree block 'node'.
9239 * NOTE: this function will unlock and release tree block 'node'
9240 * only used by relocation code
9242 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
9243 struct btrfs_root *root,
9244 struct extent_buffer *node,
9245 struct extent_buffer *parent)
9247 struct btrfs_fs_info *fs_info = root->fs_info;
9248 struct btrfs_path *path;
9249 struct walk_control *wc;
9255 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
9257 path = btrfs_alloc_path();
9261 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9263 btrfs_free_path(path);
9267 btrfs_assert_tree_locked(parent);
9268 parent_level = btrfs_header_level(parent);
9269 extent_buffer_get(parent);
9270 path->nodes[parent_level] = parent;
9271 path->slots[parent_level] = btrfs_header_nritems(parent);
9273 btrfs_assert_tree_locked(node);
9274 level = btrfs_header_level(node);
9275 path->nodes[level] = node;
9276 path->slots[level] = 0;
9277 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9279 wc->refs[parent_level] = 1;
9280 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
9282 wc->shared_level = -1;
9283 wc->stage = DROP_REFERENCE;
9287 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9290 wret = walk_down_tree(trans, root, path, wc);
9296 wret = walk_up_tree(trans, root, path, wc, parent_level);
9304 btrfs_free_path(path);
9308 static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags)
9314 * if restripe for this chunk_type is on pick target profile and
9315 * return, otherwise do the usual balance
9317 stripped = get_restripe_target(fs_info, flags);
9319 return extended_to_chunk(stripped);
9321 num_devices = fs_info->fs_devices->rw_devices;
9323 stripped = BTRFS_BLOCK_GROUP_RAID0 |
9324 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
9325 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
9327 if (num_devices == 1) {
9328 stripped |= BTRFS_BLOCK_GROUP_DUP;
9329 stripped = flags & ~stripped;
9331 /* turn raid0 into single device chunks */
9332 if (flags & BTRFS_BLOCK_GROUP_RAID0)
9335 /* turn mirroring into duplication */
9336 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
9337 BTRFS_BLOCK_GROUP_RAID10))
9338 return stripped | BTRFS_BLOCK_GROUP_DUP;
9340 /* they already had raid on here, just return */
9341 if (flags & stripped)
9344 stripped |= BTRFS_BLOCK_GROUP_DUP;
9345 stripped = flags & ~stripped;
9347 /* switch duplicated blocks with raid1 */
9348 if (flags & BTRFS_BLOCK_GROUP_DUP)
9349 return stripped | BTRFS_BLOCK_GROUP_RAID1;
9351 /* this is drive concat, leave it alone */
9357 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
9359 struct btrfs_space_info *sinfo = cache->space_info;
9361 u64 min_allocable_bytes;
9365 * We need some metadata space and system metadata space for
9366 * allocating chunks in some corner cases until we force to set
9367 * it to be readonly.
9370 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
9372 min_allocable_bytes = SZ_1M;
9374 min_allocable_bytes = 0;
9376 spin_lock(&sinfo->lock);
9377 spin_lock(&cache->lock);
9385 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
9386 cache->bytes_super - btrfs_block_group_used(&cache->item);
9388 if (btrfs_space_info_used(sinfo, true) + num_bytes +
9389 min_allocable_bytes <= sinfo->total_bytes) {
9390 sinfo->bytes_readonly += num_bytes;
9392 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
9396 spin_unlock(&cache->lock);
9397 spin_unlock(&sinfo->lock);
9401 int btrfs_inc_block_group_ro(struct btrfs_fs_info *fs_info,
9402 struct btrfs_block_group_cache *cache)
9405 struct btrfs_trans_handle *trans;
9410 trans = btrfs_join_transaction(fs_info->extent_root);
9412 return PTR_ERR(trans);
9415 * we're not allowed to set block groups readonly after the dirty
9416 * block groups cache has started writing. If it already started,
9417 * back off and let this transaction commit
9419 mutex_lock(&fs_info->ro_block_group_mutex);
9420 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9421 u64 transid = trans->transid;
9423 mutex_unlock(&fs_info->ro_block_group_mutex);
9424 btrfs_end_transaction(trans);
9426 ret = btrfs_wait_for_commit(fs_info, transid);
9433 * if we are changing raid levels, try to allocate a corresponding
9434 * block group with the new raid level.
9436 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9437 if (alloc_flags != cache->flags) {
9438 ret = do_chunk_alloc(trans, fs_info, alloc_flags,
9441 * ENOSPC is allowed here, we may have enough space
9442 * already allocated at the new raid level to
9451 ret = inc_block_group_ro(cache, 0);
9454 alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags);
9455 ret = do_chunk_alloc(trans, fs_info, alloc_flags,
9459 ret = inc_block_group_ro(cache, 0);
9461 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9462 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9463 mutex_lock(&fs_info->chunk_mutex);
9464 check_system_chunk(trans, fs_info, alloc_flags);
9465 mutex_unlock(&fs_info->chunk_mutex);
9467 mutex_unlock(&fs_info->ro_block_group_mutex);
9469 btrfs_end_transaction(trans);
9473 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
9474 struct btrfs_fs_info *fs_info, u64 type)
9476 u64 alloc_flags = get_alloc_profile(fs_info, type);
9478 return do_chunk_alloc(trans, fs_info, alloc_flags, CHUNK_ALLOC_FORCE);
9482 * helper to account the unused space of all the readonly block group in the
9483 * space_info. takes mirrors into account.
9485 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9487 struct btrfs_block_group_cache *block_group;
9491 /* It's df, we don't care if it's racy */
9492 if (list_empty(&sinfo->ro_bgs))
9495 spin_lock(&sinfo->lock);
9496 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9497 spin_lock(&block_group->lock);
9499 if (!block_group->ro) {
9500 spin_unlock(&block_group->lock);
9504 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
9505 BTRFS_BLOCK_GROUP_RAID10 |
9506 BTRFS_BLOCK_GROUP_DUP))
9511 free_bytes += (block_group->key.offset -
9512 btrfs_block_group_used(&block_group->item)) *
9515 spin_unlock(&block_group->lock);
9517 spin_unlock(&sinfo->lock);
9522 void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache)
9524 struct btrfs_space_info *sinfo = cache->space_info;
9529 spin_lock(&sinfo->lock);
9530 spin_lock(&cache->lock);
9532 num_bytes = cache->key.offset - cache->reserved -
9533 cache->pinned - cache->bytes_super -
9534 btrfs_block_group_used(&cache->item);
9535 sinfo->bytes_readonly -= num_bytes;
9536 list_del_init(&cache->ro_list);
9538 spin_unlock(&cache->lock);
9539 spin_unlock(&sinfo->lock);
9543 * checks to see if its even possible to relocate this block group.
9545 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9546 * ok to go ahead and try.
9548 int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr)
9550 struct btrfs_root *root = fs_info->extent_root;
9551 struct btrfs_block_group_cache *block_group;
9552 struct btrfs_space_info *space_info;
9553 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
9554 struct btrfs_device *device;
9555 struct btrfs_trans_handle *trans;
9565 debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG);
9567 block_group = btrfs_lookup_block_group(fs_info, bytenr);
9569 /* odd, couldn't find the block group, leave it alone */
9573 "can't find block group for bytenr %llu",
9578 min_free = btrfs_block_group_used(&block_group->item);
9580 /* no bytes used, we're good */
9584 space_info = block_group->space_info;
9585 spin_lock(&space_info->lock);
9587 full = space_info->full;
9590 * if this is the last block group we have in this space, we can't
9591 * relocate it unless we're able to allocate a new chunk below.
9593 * Otherwise, we need to make sure we have room in the space to handle
9594 * all of the extents from this block group. If we can, we're good
9596 if ((space_info->total_bytes != block_group->key.offset) &&
9597 (btrfs_space_info_used(space_info, false) + min_free <
9598 space_info->total_bytes)) {
9599 spin_unlock(&space_info->lock);
9602 spin_unlock(&space_info->lock);
9605 * ok we don't have enough space, but maybe we have free space on our
9606 * devices to allocate new chunks for relocation, so loop through our
9607 * alloc devices and guess if we have enough space. if this block
9608 * group is going to be restriped, run checks against the target
9609 * profile instead of the current one.
9621 target = get_restripe_target(fs_info, block_group->flags);
9623 index = btrfs_bg_flags_to_raid_index(extended_to_chunk(target));
9626 * this is just a balance, so if we were marked as full
9627 * we know there is no space for a new chunk
9632 "no space to alloc new chunk for block group %llu",
9633 block_group->key.objectid);
9637 index = btrfs_bg_flags_to_raid_index(block_group->flags);
9640 if (index == BTRFS_RAID_RAID10) {
9644 } else if (index == BTRFS_RAID_RAID1) {
9646 } else if (index == BTRFS_RAID_DUP) {
9649 } else if (index == BTRFS_RAID_RAID0) {
9650 dev_min = fs_devices->rw_devices;
9651 min_free = div64_u64(min_free, dev_min);
9654 /* We need to do this so that we can look at pending chunks */
9655 trans = btrfs_join_transaction(root);
9656 if (IS_ERR(trans)) {
9657 ret = PTR_ERR(trans);
9661 mutex_lock(&fs_info->chunk_mutex);
9662 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9666 * check to make sure we can actually find a chunk with enough
9667 * space to fit our block group in.
9669 if (device->total_bytes > device->bytes_used + min_free &&
9670 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
9671 ret = find_free_dev_extent(trans, device, min_free,
9676 if (dev_nr >= dev_min)
9682 if (debug && ret == -1)
9684 "no space to allocate a new chunk for block group %llu",
9685 block_group->key.objectid);
9686 mutex_unlock(&fs_info->chunk_mutex);
9687 btrfs_end_transaction(trans);
9689 btrfs_put_block_group(block_group);
9693 static int find_first_block_group(struct btrfs_fs_info *fs_info,
9694 struct btrfs_path *path,
9695 struct btrfs_key *key)
9697 struct btrfs_root *root = fs_info->extent_root;
9699 struct btrfs_key found_key;
9700 struct extent_buffer *leaf;
9703 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9708 slot = path->slots[0];
9709 leaf = path->nodes[0];
9710 if (slot >= btrfs_header_nritems(leaf)) {
9711 ret = btrfs_next_leaf(root, path);
9718 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9720 if (found_key.objectid >= key->objectid &&
9721 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9722 struct extent_map_tree *em_tree;
9723 struct extent_map *em;
9725 em_tree = &root->fs_info->mapping_tree.map_tree;
9726 read_lock(&em_tree->lock);
9727 em = lookup_extent_mapping(em_tree, found_key.objectid,
9729 read_unlock(&em_tree->lock);
9732 "logical %llu len %llu found bg but no related chunk",
9733 found_key.objectid, found_key.offset);
9738 free_extent_map(em);
9747 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9749 struct btrfs_block_group_cache *block_group;
9753 struct inode *inode;
9755 block_group = btrfs_lookup_first_block_group(info, last);
9756 while (block_group) {
9757 spin_lock(&block_group->lock);
9758 if (block_group->iref)
9760 spin_unlock(&block_group->lock);
9761 block_group = next_block_group(info, block_group);
9770 inode = block_group->inode;
9771 block_group->iref = 0;
9772 block_group->inode = NULL;
9773 spin_unlock(&block_group->lock);
9774 ASSERT(block_group->io_ctl.inode == NULL);
9776 last = block_group->key.objectid + block_group->key.offset;
9777 btrfs_put_block_group(block_group);
9782 * Must be called only after stopping all workers, since we could have block
9783 * group caching kthreads running, and therefore they could race with us if we
9784 * freed the block groups before stopping them.
9786 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9788 struct btrfs_block_group_cache *block_group;
9789 struct btrfs_space_info *space_info;
9790 struct btrfs_caching_control *caching_ctl;
9793 down_write(&info->commit_root_sem);
9794 while (!list_empty(&info->caching_block_groups)) {
9795 caching_ctl = list_entry(info->caching_block_groups.next,
9796 struct btrfs_caching_control, list);
9797 list_del(&caching_ctl->list);
9798 put_caching_control(caching_ctl);
9800 up_write(&info->commit_root_sem);
9802 spin_lock(&info->unused_bgs_lock);
9803 while (!list_empty(&info->unused_bgs)) {
9804 block_group = list_first_entry(&info->unused_bgs,
9805 struct btrfs_block_group_cache,
9807 list_del_init(&block_group->bg_list);
9808 btrfs_put_block_group(block_group);
9810 spin_unlock(&info->unused_bgs_lock);
9812 spin_lock(&info->block_group_cache_lock);
9813 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9814 block_group = rb_entry(n, struct btrfs_block_group_cache,
9816 rb_erase(&block_group->cache_node,
9817 &info->block_group_cache_tree);
9818 RB_CLEAR_NODE(&block_group->cache_node);
9819 spin_unlock(&info->block_group_cache_lock);
9821 down_write(&block_group->space_info->groups_sem);
9822 list_del(&block_group->list);
9823 up_write(&block_group->space_info->groups_sem);
9826 * We haven't cached this block group, which means we could
9827 * possibly have excluded extents on this block group.
9829 if (block_group->cached == BTRFS_CACHE_NO ||
9830 block_group->cached == BTRFS_CACHE_ERROR)
9831 free_excluded_extents(info, block_group);
9833 btrfs_remove_free_space_cache(block_group);
9834 ASSERT(block_group->cached != BTRFS_CACHE_STARTED);
9835 ASSERT(list_empty(&block_group->dirty_list));
9836 ASSERT(list_empty(&block_group->io_list));
9837 ASSERT(list_empty(&block_group->bg_list));
9838 ASSERT(atomic_read(&block_group->count) == 1);
9839 btrfs_put_block_group(block_group);
9841 spin_lock(&info->block_group_cache_lock);
9843 spin_unlock(&info->block_group_cache_lock);
9845 /* now that all the block groups are freed, go through and
9846 * free all the space_info structs. This is only called during
9847 * the final stages of unmount, and so we know nobody is
9848 * using them. We call synchronize_rcu() once before we start,
9849 * just to be on the safe side.
9853 release_global_block_rsv(info);
9855 while (!list_empty(&info->space_info)) {
9858 space_info = list_entry(info->space_info.next,
9859 struct btrfs_space_info,
9863 * Do not hide this behind enospc_debug, this is actually
9864 * important and indicates a real bug if this happens.
9866 if (WARN_ON(space_info->bytes_pinned > 0 ||
9867 space_info->bytes_reserved > 0 ||
9868 space_info->bytes_may_use > 0))
9869 dump_space_info(info, space_info, 0, 0);
9870 list_del(&space_info->list);
9871 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9872 struct kobject *kobj;
9873 kobj = space_info->block_group_kobjs[i];
9874 space_info->block_group_kobjs[i] = NULL;
9880 kobject_del(&space_info->kobj);
9881 kobject_put(&space_info->kobj);
9886 static void link_block_group(struct btrfs_block_group_cache *cache)
9888 struct btrfs_space_info *space_info = cache->space_info;
9889 int index = btrfs_bg_flags_to_raid_index(cache->flags);
9892 down_write(&space_info->groups_sem);
9893 if (list_empty(&space_info->block_groups[index]))
9895 list_add_tail(&cache->list, &space_info->block_groups[index]);
9896 up_write(&space_info->groups_sem);
9899 struct raid_kobject *rkobj;
9902 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9905 rkobj->raid_type = index;
9906 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9907 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9908 "%s", get_raid_name(index));
9910 kobject_put(&rkobj->kobj);
9913 space_info->block_group_kobjs[index] = &rkobj->kobj;
9918 btrfs_warn(cache->fs_info,
9919 "failed to add kobject for block cache, ignoring");
9922 static struct btrfs_block_group_cache *
9923 btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info,
9924 u64 start, u64 size)
9926 struct btrfs_block_group_cache *cache;
9928 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9932 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
9934 if (!cache->free_space_ctl) {
9939 cache->key.objectid = start;
9940 cache->key.offset = size;
9941 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9943 cache->fs_info = fs_info;
9944 cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start);
9945 set_free_space_tree_thresholds(cache);
9947 atomic_set(&cache->count, 1);
9948 spin_lock_init(&cache->lock);
9949 init_rwsem(&cache->data_rwsem);
9950 INIT_LIST_HEAD(&cache->list);
9951 INIT_LIST_HEAD(&cache->cluster_list);
9952 INIT_LIST_HEAD(&cache->bg_list);
9953 INIT_LIST_HEAD(&cache->ro_list);
9954 INIT_LIST_HEAD(&cache->dirty_list);
9955 INIT_LIST_HEAD(&cache->io_list);
9956 btrfs_init_free_space_ctl(cache);
9957 atomic_set(&cache->trimming, 0);
9958 mutex_init(&cache->free_space_lock);
9959 btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root);
9964 int btrfs_read_block_groups(struct btrfs_fs_info *info)
9966 struct btrfs_path *path;
9968 struct btrfs_block_group_cache *cache;
9969 struct btrfs_space_info *space_info;
9970 struct btrfs_key key;
9971 struct btrfs_key found_key;
9972 struct extent_buffer *leaf;
9978 feature = btrfs_super_incompat_flags(info->super_copy);
9979 mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS);
9983 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9984 path = btrfs_alloc_path();
9987 path->reada = READA_FORWARD;
9989 cache_gen = btrfs_super_cache_generation(info->super_copy);
9990 if (btrfs_test_opt(info, SPACE_CACHE) &&
9991 btrfs_super_generation(info->super_copy) != cache_gen)
9993 if (btrfs_test_opt(info, CLEAR_CACHE))
9997 ret = find_first_block_group(info, path, &key);
10003 leaf = path->nodes[0];
10004 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
10006 cache = btrfs_create_block_group_cache(info, found_key.objectid,
10015 * When we mount with old space cache, we need to
10016 * set BTRFS_DC_CLEAR and set dirty flag.
10018 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
10019 * truncate the old free space cache inode and
10021 * b) Setting 'dirty flag' makes sure that we flush
10022 * the new space cache info onto disk.
10024 if (btrfs_test_opt(info, SPACE_CACHE))
10025 cache->disk_cache_state = BTRFS_DC_CLEAR;
10028 read_extent_buffer(leaf, &cache->item,
10029 btrfs_item_ptr_offset(leaf, path->slots[0]),
10030 sizeof(cache->item));
10031 cache->flags = btrfs_block_group_flags(&cache->item);
10033 ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) &&
10034 (cache->flags & BTRFS_BLOCK_GROUP_DATA))) {
10036 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
10037 cache->key.objectid);
10042 key.objectid = found_key.objectid + found_key.offset;
10043 btrfs_release_path(path);
10046 * We need to exclude the super stripes now so that the space
10047 * info has super bytes accounted for, otherwise we'll think
10048 * we have more space than we actually do.
10050 ret = exclude_super_stripes(info, cache);
10053 * We may have excluded something, so call this just in
10056 free_excluded_extents(info, cache);
10057 btrfs_put_block_group(cache);
10062 * check for two cases, either we are full, and therefore
10063 * don't need to bother with the caching work since we won't
10064 * find any space, or we are empty, and we can just add all
10065 * the space in and be done with it. This saves us _alot_ of
10066 * time, particularly in the full case.
10068 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
10069 cache->last_byte_to_unpin = (u64)-1;
10070 cache->cached = BTRFS_CACHE_FINISHED;
10071 free_excluded_extents(info, cache);
10072 } else if (btrfs_block_group_used(&cache->item) == 0) {
10073 cache->last_byte_to_unpin = (u64)-1;
10074 cache->cached = BTRFS_CACHE_FINISHED;
10075 add_new_free_space(cache, info,
10076 found_key.objectid,
10077 found_key.objectid +
10079 free_excluded_extents(info, cache);
10082 ret = btrfs_add_block_group_cache(info, cache);
10084 btrfs_remove_free_space_cache(cache);
10085 btrfs_put_block_group(cache);
10089 trace_btrfs_add_block_group(info, cache, 0);
10090 update_space_info(info, cache->flags, found_key.offset,
10091 btrfs_block_group_used(&cache->item),
10092 cache->bytes_super, &space_info);
10094 cache->space_info = space_info;
10096 link_block_group(cache);
10098 set_avail_alloc_bits(info, cache->flags);
10099 if (btrfs_chunk_readonly(info, cache->key.objectid)) {
10100 inc_block_group_ro(cache, 1);
10101 } else if (btrfs_block_group_used(&cache->item) == 0) {
10102 spin_lock(&info->unused_bgs_lock);
10103 /* Should always be true but just in case. */
10104 if (list_empty(&cache->bg_list)) {
10105 btrfs_get_block_group(cache);
10106 list_add_tail(&cache->bg_list,
10107 &info->unused_bgs);
10109 spin_unlock(&info->unused_bgs_lock);
10113 list_for_each_entry_rcu(space_info, &info->space_info, list) {
10114 if (!(get_alloc_profile(info, space_info->flags) &
10115 (BTRFS_BLOCK_GROUP_RAID10 |
10116 BTRFS_BLOCK_GROUP_RAID1 |
10117 BTRFS_BLOCK_GROUP_RAID5 |
10118 BTRFS_BLOCK_GROUP_RAID6 |
10119 BTRFS_BLOCK_GROUP_DUP)))
10122 * avoid allocating from un-mirrored block group if there are
10123 * mirrored block groups.
10125 list_for_each_entry(cache,
10126 &space_info->block_groups[BTRFS_RAID_RAID0],
10128 inc_block_group_ro(cache, 1);
10129 list_for_each_entry(cache,
10130 &space_info->block_groups[BTRFS_RAID_SINGLE],
10132 inc_block_group_ro(cache, 1);
10135 init_global_block_rsv(info);
10138 btrfs_free_path(path);
10142 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans)
10144 struct btrfs_fs_info *fs_info = trans->fs_info;
10145 struct btrfs_block_group_cache *block_group, *tmp;
10146 struct btrfs_root *extent_root = fs_info->extent_root;
10147 struct btrfs_block_group_item item;
10148 struct btrfs_key key;
10150 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
10152 trans->can_flush_pending_bgs = false;
10153 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
10157 spin_lock(&block_group->lock);
10158 memcpy(&item, &block_group->item, sizeof(item));
10159 memcpy(&key, &block_group->key, sizeof(key));
10160 spin_unlock(&block_group->lock);
10162 ret = btrfs_insert_item(trans, extent_root, &key, &item,
10165 btrfs_abort_transaction(trans, ret);
10166 ret = btrfs_finish_chunk_alloc(trans, fs_info, key.objectid,
10169 btrfs_abort_transaction(trans, ret);
10170 add_block_group_free_space(trans, fs_info, block_group);
10171 /* already aborted the transaction if it failed. */
10173 list_del_init(&block_group->bg_list);
10175 trans->can_flush_pending_bgs = can_flush_pending_bgs;
10178 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
10179 struct btrfs_fs_info *fs_info, u64 bytes_used,
10180 u64 type, u64 chunk_offset, u64 size)
10182 struct btrfs_block_group_cache *cache;
10185 btrfs_set_log_full_commit(fs_info, trans);
10187 cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size);
10191 btrfs_set_block_group_used(&cache->item, bytes_used);
10192 btrfs_set_block_group_chunk_objectid(&cache->item,
10193 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
10194 btrfs_set_block_group_flags(&cache->item, type);
10196 cache->flags = type;
10197 cache->last_byte_to_unpin = (u64)-1;
10198 cache->cached = BTRFS_CACHE_FINISHED;
10199 cache->needs_free_space = 1;
10200 ret = exclude_super_stripes(fs_info, cache);
10203 * We may have excluded something, so call this just in
10206 free_excluded_extents(fs_info, cache);
10207 btrfs_put_block_group(cache);
10211 add_new_free_space(cache, fs_info, chunk_offset, chunk_offset + size);
10213 free_excluded_extents(fs_info, cache);
10215 #ifdef CONFIG_BTRFS_DEBUG
10216 if (btrfs_should_fragment_free_space(cache)) {
10217 u64 new_bytes_used = size - bytes_used;
10219 bytes_used += new_bytes_used >> 1;
10220 fragment_free_space(cache);
10224 * Ensure the corresponding space_info object is created and
10225 * assigned to our block group. We want our bg to be added to the rbtree
10226 * with its ->space_info set.
10228 cache->space_info = __find_space_info(fs_info, cache->flags);
10229 if (!cache->space_info) {
10230 ret = create_space_info(fs_info, cache->flags,
10231 &cache->space_info);
10233 btrfs_remove_free_space_cache(cache);
10234 btrfs_put_block_group(cache);
10239 ret = btrfs_add_block_group_cache(fs_info, cache);
10241 btrfs_remove_free_space_cache(cache);
10242 btrfs_put_block_group(cache);
10247 * Now that our block group has its ->space_info set and is inserted in
10248 * the rbtree, update the space info's counters.
10250 trace_btrfs_add_block_group(fs_info, cache, 1);
10251 update_space_info(fs_info, cache->flags, size, bytes_used,
10252 cache->bytes_super, &cache->space_info);
10253 update_global_block_rsv(fs_info);
10255 link_block_group(cache);
10257 list_add_tail(&cache->bg_list, &trans->new_bgs);
10259 set_avail_alloc_bits(fs_info, type);
10263 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
10265 u64 extra_flags = chunk_to_extended(flags) &
10266 BTRFS_EXTENDED_PROFILE_MASK;
10268 write_seqlock(&fs_info->profiles_lock);
10269 if (flags & BTRFS_BLOCK_GROUP_DATA)
10270 fs_info->avail_data_alloc_bits &= ~extra_flags;
10271 if (flags & BTRFS_BLOCK_GROUP_METADATA)
10272 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
10273 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
10274 fs_info->avail_system_alloc_bits &= ~extra_flags;
10275 write_sequnlock(&fs_info->profiles_lock);
10278 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
10279 struct btrfs_fs_info *fs_info, u64 group_start,
10280 struct extent_map *em)
10282 struct btrfs_root *root = fs_info->extent_root;
10283 struct btrfs_path *path;
10284 struct btrfs_block_group_cache *block_group;
10285 struct btrfs_free_cluster *cluster;
10286 struct btrfs_root *tree_root = fs_info->tree_root;
10287 struct btrfs_key key;
10288 struct inode *inode;
10289 struct kobject *kobj = NULL;
10293 struct btrfs_caching_control *caching_ctl = NULL;
10296 block_group = btrfs_lookup_block_group(fs_info, group_start);
10297 BUG_ON(!block_group);
10298 BUG_ON(!block_group->ro);
10301 * Free the reserved super bytes from this block group before
10304 free_excluded_extents(fs_info, block_group);
10305 btrfs_free_ref_tree_range(fs_info, block_group->key.objectid,
10306 block_group->key.offset);
10308 memcpy(&key, &block_group->key, sizeof(key));
10309 index = btrfs_bg_flags_to_raid_index(block_group->flags);
10310 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
10311 BTRFS_BLOCK_GROUP_RAID1 |
10312 BTRFS_BLOCK_GROUP_RAID10))
10317 /* make sure this block group isn't part of an allocation cluster */
10318 cluster = &fs_info->data_alloc_cluster;
10319 spin_lock(&cluster->refill_lock);
10320 btrfs_return_cluster_to_free_space(block_group, cluster);
10321 spin_unlock(&cluster->refill_lock);
10324 * make sure this block group isn't part of a metadata
10325 * allocation cluster
10327 cluster = &fs_info->meta_alloc_cluster;
10328 spin_lock(&cluster->refill_lock);
10329 btrfs_return_cluster_to_free_space(block_group, cluster);
10330 spin_unlock(&cluster->refill_lock);
10332 path = btrfs_alloc_path();
10339 * get the inode first so any iput calls done for the io_list
10340 * aren't the final iput (no unlinks allowed now)
10342 inode = lookup_free_space_inode(fs_info, block_group, path);
10344 mutex_lock(&trans->transaction->cache_write_mutex);
10346 * make sure our free spache cache IO is done before remove the
10349 spin_lock(&trans->transaction->dirty_bgs_lock);
10350 if (!list_empty(&block_group->io_list)) {
10351 list_del_init(&block_group->io_list);
10353 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
10355 spin_unlock(&trans->transaction->dirty_bgs_lock);
10356 btrfs_wait_cache_io(trans, block_group, path);
10357 btrfs_put_block_group(block_group);
10358 spin_lock(&trans->transaction->dirty_bgs_lock);
10361 if (!list_empty(&block_group->dirty_list)) {
10362 list_del_init(&block_group->dirty_list);
10363 btrfs_put_block_group(block_group);
10365 spin_unlock(&trans->transaction->dirty_bgs_lock);
10366 mutex_unlock(&trans->transaction->cache_write_mutex);
10368 if (!IS_ERR(inode)) {
10369 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10371 btrfs_add_delayed_iput(inode);
10374 clear_nlink(inode);
10375 /* One for the block groups ref */
10376 spin_lock(&block_group->lock);
10377 if (block_group->iref) {
10378 block_group->iref = 0;
10379 block_group->inode = NULL;
10380 spin_unlock(&block_group->lock);
10383 spin_unlock(&block_group->lock);
10385 /* One for our lookup ref */
10386 btrfs_add_delayed_iput(inode);
10389 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
10390 key.offset = block_group->key.objectid;
10393 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
10397 btrfs_release_path(path);
10399 ret = btrfs_del_item(trans, tree_root, path);
10402 btrfs_release_path(path);
10405 spin_lock(&fs_info->block_group_cache_lock);
10406 rb_erase(&block_group->cache_node,
10407 &fs_info->block_group_cache_tree);
10408 RB_CLEAR_NODE(&block_group->cache_node);
10410 if (fs_info->first_logical_byte == block_group->key.objectid)
10411 fs_info->first_logical_byte = (u64)-1;
10412 spin_unlock(&fs_info->block_group_cache_lock);
10414 down_write(&block_group->space_info->groups_sem);
10416 * we must use list_del_init so people can check to see if they
10417 * are still on the list after taking the semaphore
10419 list_del_init(&block_group->list);
10420 if (list_empty(&block_group->space_info->block_groups[index])) {
10421 kobj = block_group->space_info->block_group_kobjs[index];
10422 block_group->space_info->block_group_kobjs[index] = NULL;
10423 clear_avail_alloc_bits(fs_info, block_group->flags);
10425 up_write(&block_group->space_info->groups_sem);
10431 if (block_group->has_caching_ctl)
10432 caching_ctl = get_caching_control(block_group);
10433 if (block_group->cached == BTRFS_CACHE_STARTED)
10434 wait_block_group_cache_done(block_group);
10435 if (block_group->has_caching_ctl) {
10436 down_write(&fs_info->commit_root_sem);
10437 if (!caching_ctl) {
10438 struct btrfs_caching_control *ctl;
10440 list_for_each_entry(ctl,
10441 &fs_info->caching_block_groups, list)
10442 if (ctl->block_group == block_group) {
10444 refcount_inc(&caching_ctl->count);
10449 list_del_init(&caching_ctl->list);
10450 up_write(&fs_info->commit_root_sem);
10452 /* Once for the caching bgs list and once for us. */
10453 put_caching_control(caching_ctl);
10454 put_caching_control(caching_ctl);
10458 spin_lock(&trans->transaction->dirty_bgs_lock);
10459 if (!list_empty(&block_group->dirty_list)) {
10462 if (!list_empty(&block_group->io_list)) {
10465 spin_unlock(&trans->transaction->dirty_bgs_lock);
10466 btrfs_remove_free_space_cache(block_group);
10468 spin_lock(&block_group->space_info->lock);
10469 list_del_init(&block_group->ro_list);
10471 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
10472 WARN_ON(block_group->space_info->total_bytes
10473 < block_group->key.offset);
10474 WARN_ON(block_group->space_info->bytes_readonly
10475 < block_group->key.offset);
10476 WARN_ON(block_group->space_info->disk_total
10477 < block_group->key.offset * factor);
10479 block_group->space_info->total_bytes -= block_group->key.offset;
10480 block_group->space_info->bytes_readonly -= block_group->key.offset;
10481 block_group->space_info->disk_total -= block_group->key.offset * factor;
10483 spin_unlock(&block_group->space_info->lock);
10485 memcpy(&key, &block_group->key, sizeof(key));
10487 mutex_lock(&fs_info->chunk_mutex);
10488 if (!list_empty(&em->list)) {
10489 /* We're in the transaction->pending_chunks list. */
10490 free_extent_map(em);
10492 spin_lock(&block_group->lock);
10493 block_group->removed = 1;
10495 * At this point trimming can't start on this block group, because we
10496 * removed the block group from the tree fs_info->block_group_cache_tree
10497 * so no one can't find it anymore and even if someone already got this
10498 * block group before we removed it from the rbtree, they have already
10499 * incremented block_group->trimming - if they didn't, they won't find
10500 * any free space entries because we already removed them all when we
10501 * called btrfs_remove_free_space_cache().
10503 * And we must not remove the extent map from the fs_info->mapping_tree
10504 * to prevent the same logical address range and physical device space
10505 * ranges from being reused for a new block group. This is because our
10506 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10507 * completely transactionless, so while it is trimming a range the
10508 * currently running transaction might finish and a new one start,
10509 * allowing for new block groups to be created that can reuse the same
10510 * physical device locations unless we take this special care.
10512 * There may also be an implicit trim operation if the file system
10513 * is mounted with -odiscard. The same protections must remain
10514 * in place until the extents have been discarded completely when
10515 * the transaction commit has completed.
10517 remove_em = (atomic_read(&block_group->trimming) == 0);
10519 * Make sure a trimmer task always sees the em in the pinned_chunks list
10520 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10521 * before checking block_group->removed).
10525 * Our em might be in trans->transaction->pending_chunks which
10526 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10527 * and so is the fs_info->pinned_chunks list.
10529 * So at this point we must be holding the chunk_mutex to avoid
10530 * any races with chunk allocation (more specifically at
10531 * volumes.c:contains_pending_extent()), to ensure it always
10532 * sees the em, either in the pending_chunks list or in the
10533 * pinned_chunks list.
10535 list_move_tail(&em->list, &fs_info->pinned_chunks);
10537 spin_unlock(&block_group->lock);
10540 struct extent_map_tree *em_tree;
10542 em_tree = &fs_info->mapping_tree.map_tree;
10543 write_lock(&em_tree->lock);
10545 * The em might be in the pending_chunks list, so make sure the
10546 * chunk mutex is locked, since remove_extent_mapping() will
10547 * delete us from that list.
10549 remove_extent_mapping(em_tree, em);
10550 write_unlock(&em_tree->lock);
10551 /* once for the tree */
10552 free_extent_map(em);
10555 mutex_unlock(&fs_info->chunk_mutex);
10557 ret = remove_block_group_free_space(trans, fs_info, block_group);
10561 btrfs_put_block_group(block_group);
10562 btrfs_put_block_group(block_group);
10564 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10570 ret = btrfs_del_item(trans, root, path);
10572 btrfs_free_path(path);
10576 struct btrfs_trans_handle *
10577 btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info,
10578 const u64 chunk_offset)
10580 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
10581 struct extent_map *em;
10582 struct map_lookup *map;
10583 unsigned int num_items;
10585 read_lock(&em_tree->lock);
10586 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
10587 read_unlock(&em_tree->lock);
10588 ASSERT(em && em->start == chunk_offset);
10591 * We need to reserve 3 + N units from the metadata space info in order
10592 * to remove a block group (done at btrfs_remove_chunk() and at
10593 * btrfs_remove_block_group()), which are used for:
10595 * 1 unit for adding the free space inode's orphan (located in the tree
10597 * 1 unit for deleting the block group item (located in the extent
10599 * 1 unit for deleting the free space item (located in tree of tree
10601 * N units for deleting N device extent items corresponding to each
10602 * stripe (located in the device tree).
10604 * In order to remove a block group we also need to reserve units in the
10605 * system space info in order to update the chunk tree (update one or
10606 * more device items and remove one chunk item), but this is done at
10607 * btrfs_remove_chunk() through a call to check_system_chunk().
10609 map = em->map_lookup;
10610 num_items = 3 + map->num_stripes;
10611 free_extent_map(em);
10613 return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root,
10618 * Process the unused_bgs list and remove any that don't have any allocated
10619 * space inside of them.
10621 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10623 struct btrfs_block_group_cache *block_group;
10624 struct btrfs_space_info *space_info;
10625 struct btrfs_trans_handle *trans;
10628 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
10631 spin_lock(&fs_info->unused_bgs_lock);
10632 while (!list_empty(&fs_info->unused_bgs)) {
10636 block_group = list_first_entry(&fs_info->unused_bgs,
10637 struct btrfs_block_group_cache,
10639 list_del_init(&block_group->bg_list);
10641 space_info = block_group->space_info;
10643 if (ret || btrfs_mixed_space_info(space_info)) {
10644 btrfs_put_block_group(block_group);
10647 spin_unlock(&fs_info->unused_bgs_lock);
10649 mutex_lock(&fs_info->delete_unused_bgs_mutex);
10651 /* Don't want to race with allocators so take the groups_sem */
10652 down_write(&space_info->groups_sem);
10653 spin_lock(&block_group->lock);
10654 if (block_group->reserved ||
10655 btrfs_block_group_used(&block_group->item) ||
10657 list_is_singular(&block_group->list)) {
10659 * We want to bail if we made new allocations or have
10660 * outstanding allocations in this block group. We do
10661 * the ro check in case balance is currently acting on
10662 * this block group.
10664 spin_unlock(&block_group->lock);
10665 up_write(&space_info->groups_sem);
10668 spin_unlock(&block_group->lock);
10670 /* We don't want to force the issue, only flip if it's ok. */
10671 ret = inc_block_group_ro(block_group, 0);
10672 up_write(&space_info->groups_sem);
10679 * Want to do this before we do anything else so we can recover
10680 * properly if we fail to join the transaction.
10682 trans = btrfs_start_trans_remove_block_group(fs_info,
10683 block_group->key.objectid);
10684 if (IS_ERR(trans)) {
10685 btrfs_dec_block_group_ro(block_group);
10686 ret = PTR_ERR(trans);
10691 * We could have pending pinned extents for this block group,
10692 * just delete them, we don't care about them anymore.
10694 start = block_group->key.objectid;
10695 end = start + block_group->key.offset - 1;
10697 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10698 * btrfs_finish_extent_commit(). If we are at transaction N,
10699 * another task might be running finish_extent_commit() for the
10700 * previous transaction N - 1, and have seen a range belonging
10701 * to the block group in freed_extents[] before we were able to
10702 * clear the whole block group range from freed_extents[]. This
10703 * means that task can lookup for the block group after we
10704 * unpinned it from freed_extents[] and removed it, leading to
10705 * a BUG_ON() at btrfs_unpin_extent_range().
10707 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10708 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10711 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10712 btrfs_dec_block_group_ro(block_group);
10715 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10718 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10719 btrfs_dec_block_group_ro(block_group);
10722 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10724 /* Reset pinned so btrfs_put_block_group doesn't complain */
10725 spin_lock(&space_info->lock);
10726 spin_lock(&block_group->lock);
10728 space_info->bytes_pinned -= block_group->pinned;
10729 space_info->bytes_readonly += block_group->pinned;
10730 percpu_counter_add(&space_info->total_bytes_pinned,
10731 -block_group->pinned);
10732 block_group->pinned = 0;
10734 spin_unlock(&block_group->lock);
10735 spin_unlock(&space_info->lock);
10737 /* DISCARD can flip during remount */
10738 trimming = btrfs_test_opt(fs_info, DISCARD);
10740 /* Implicit trim during transaction commit. */
10742 btrfs_get_block_group_trimming(block_group);
10745 * Btrfs_remove_chunk will abort the transaction if things go
10748 ret = btrfs_remove_chunk(trans, fs_info,
10749 block_group->key.objectid);
10753 btrfs_put_block_group_trimming(block_group);
10758 * If we're not mounted with -odiscard, we can just forget
10759 * about this block group. Otherwise we'll need to wait
10760 * until transaction commit to do the actual discard.
10763 spin_lock(&fs_info->unused_bgs_lock);
10765 * A concurrent scrub might have added us to the list
10766 * fs_info->unused_bgs, so use a list_move operation
10767 * to add the block group to the deleted_bgs list.
10769 list_move(&block_group->bg_list,
10770 &trans->transaction->deleted_bgs);
10771 spin_unlock(&fs_info->unused_bgs_lock);
10772 btrfs_get_block_group(block_group);
10775 btrfs_end_transaction(trans);
10777 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
10778 btrfs_put_block_group(block_group);
10779 spin_lock(&fs_info->unused_bgs_lock);
10781 spin_unlock(&fs_info->unused_bgs_lock);
10784 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10786 struct btrfs_space_info *space_info;
10787 struct btrfs_super_block *disk_super;
10793 disk_super = fs_info->super_copy;
10794 if (!btrfs_super_root(disk_super))
10797 features = btrfs_super_incompat_flags(disk_super);
10798 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10801 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10802 ret = create_space_info(fs_info, flags, &space_info);
10807 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10808 ret = create_space_info(fs_info, flags, &space_info);
10810 flags = BTRFS_BLOCK_GROUP_METADATA;
10811 ret = create_space_info(fs_info, flags, &space_info);
10815 flags = BTRFS_BLOCK_GROUP_DATA;
10816 ret = create_space_info(fs_info, flags, &space_info);
10822 int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
10823 u64 start, u64 end)
10825 return unpin_extent_range(fs_info, start, end, false);
10829 * It used to be that old block groups would be left around forever.
10830 * Iterating over them would be enough to trim unused space. Since we
10831 * now automatically remove them, we also need to iterate over unallocated
10834 * We don't want a transaction for this since the discard may take a
10835 * substantial amount of time. We don't require that a transaction be
10836 * running, but we do need to take a running transaction into account
10837 * to ensure that we're not discarding chunks that were released in
10838 * the current transaction.
10840 * Holding the chunks lock will prevent other threads from allocating
10841 * or releasing chunks, but it won't prevent a running transaction
10842 * from committing and releasing the memory that the pending chunks
10843 * list head uses. For that, we need to take a reference to the
10846 static int btrfs_trim_free_extents(struct btrfs_device *device,
10847 u64 minlen, u64 *trimmed)
10849 u64 start = 0, len = 0;
10854 /* Not writeable = nothing to do. */
10855 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
10858 /* No free space = nothing to do. */
10859 if (device->total_bytes <= device->bytes_used)
10865 struct btrfs_fs_info *fs_info = device->fs_info;
10866 struct btrfs_transaction *trans;
10869 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
10873 down_read(&fs_info->commit_root_sem);
10875 spin_lock(&fs_info->trans_lock);
10876 trans = fs_info->running_transaction;
10878 refcount_inc(&trans->use_count);
10879 spin_unlock(&fs_info->trans_lock);
10881 ret = find_free_dev_extent_start(trans, device, minlen, start,
10884 btrfs_put_transaction(trans);
10887 up_read(&fs_info->commit_root_sem);
10888 mutex_unlock(&fs_info->chunk_mutex);
10889 if (ret == -ENOSPC)
10894 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
10895 up_read(&fs_info->commit_root_sem);
10896 mutex_unlock(&fs_info->chunk_mutex);
10904 if (fatal_signal_pending(current)) {
10905 ret = -ERESTARTSYS;
10915 int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
10917 struct btrfs_block_group_cache *cache = NULL;
10918 struct btrfs_device *device;
10919 struct list_head *devices;
10924 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
10928 * try to trim all FS space, our block group may start from non-zero.
10930 if (range->len == total_bytes)
10931 cache = btrfs_lookup_first_block_group(fs_info, range->start);
10933 cache = btrfs_lookup_block_group(fs_info, range->start);
10936 if (cache->key.objectid >= (range->start + range->len)) {
10937 btrfs_put_block_group(cache);
10941 start = max(range->start, cache->key.objectid);
10942 end = min(range->start + range->len,
10943 cache->key.objectid + cache->key.offset);
10945 if (end - start >= range->minlen) {
10946 if (!block_group_cache_done(cache)) {
10947 ret = cache_block_group(cache, 0);
10949 btrfs_put_block_group(cache);
10952 ret = wait_block_group_cache_done(cache);
10954 btrfs_put_block_group(cache);
10958 ret = btrfs_trim_block_group(cache,
10964 trimmed += group_trimmed;
10966 btrfs_put_block_group(cache);
10971 cache = next_block_group(fs_info, cache);
10974 mutex_lock(&fs_info->fs_devices->device_list_mutex);
10975 devices = &fs_info->fs_devices->alloc_list;
10976 list_for_each_entry(device, devices, dev_alloc_list) {
10977 ret = btrfs_trim_free_extents(device, range->minlen,
10982 trimmed += group_trimmed;
10984 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
10986 range->len = trimmed;
10991 * btrfs_{start,end}_write_no_snapshotting() are similar to
10992 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10993 * data into the page cache through nocow before the subvolume is snapshoted,
10994 * but flush the data into disk after the snapshot creation, or to prevent
10995 * operations while snapshotting is ongoing and that cause the snapshot to be
10996 * inconsistent (writes followed by expanding truncates for example).
10998 void btrfs_end_write_no_snapshotting(struct btrfs_root *root)
11000 percpu_counter_dec(&root->subv_writers->counter);
11002 * Make sure counter is updated before we wake up waiters.
11005 if (waitqueue_active(&root->subv_writers->wait))
11006 wake_up(&root->subv_writers->wait);
11009 int btrfs_start_write_no_snapshotting(struct btrfs_root *root)
11011 if (atomic_read(&root->will_be_snapshotted))
11014 percpu_counter_inc(&root->subv_writers->counter);
11016 * Make sure counter is updated before we check for snapshot creation.
11019 if (atomic_read(&root->will_be_snapshotted)) {
11020 btrfs_end_write_no_snapshotting(root);
11026 void btrfs_wait_for_snapshot_creation(struct btrfs_root *root)
11031 ret = btrfs_start_write_no_snapshotting(root);
11034 wait_on_atomic_t(&root->will_be_snapshotted, atomic_t_wait,
11035 TASK_UNINTERRUPTIBLE);