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>
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 = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
1207 lenum = cpu_to_le64(owner);
1208 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1209 lenum = cpu_to_le64(offset);
1210 low_crc = btrfs_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 if (!stripe->dev->can_discard)
2151 ret = btrfs_issue_discard(stripe->dev->bdev,
2156 discarded_bytes += bytes;
2157 else if (ret != -EOPNOTSUPP)
2158 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2161 * Just in case we get back EOPNOTSUPP for some reason,
2162 * just ignore the return value so we don't screw up
2163 * people calling discard_extent.
2167 btrfs_put_bbio(bbio);
2169 btrfs_bio_counter_dec(fs_info);
2172 *actual_bytes = discarded_bytes;
2175 if (ret == -EOPNOTSUPP)
2180 /* Can return -ENOMEM */
2181 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2182 struct btrfs_root *root,
2183 u64 bytenr, u64 num_bytes, u64 parent,
2184 u64 root_objectid, u64 owner, u64 offset)
2186 struct btrfs_fs_info *fs_info = root->fs_info;
2187 int old_ref_mod, new_ref_mod;
2190 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
2191 root_objectid == BTRFS_TREE_LOG_OBJECTID);
2193 btrfs_ref_tree_mod(root, bytenr, num_bytes, parent, root_objectid,
2194 owner, offset, BTRFS_ADD_DELAYED_REF);
2196 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2197 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
2199 root_objectid, (int)owner,
2200 BTRFS_ADD_DELAYED_REF, NULL,
2201 &old_ref_mod, &new_ref_mod);
2203 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
2205 root_objectid, owner, offset,
2206 0, BTRFS_ADD_DELAYED_REF,
2207 &old_ref_mod, &new_ref_mod);
2210 if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0)
2211 add_pinned_bytes(fs_info, -num_bytes, owner, root_objectid);
2216 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2217 struct btrfs_fs_info *fs_info,
2218 struct btrfs_delayed_ref_node *node,
2219 u64 parent, u64 root_objectid,
2220 u64 owner, u64 offset, int refs_to_add,
2221 struct btrfs_delayed_extent_op *extent_op)
2223 struct btrfs_path *path;
2224 struct extent_buffer *leaf;
2225 struct btrfs_extent_item *item;
2226 struct btrfs_key key;
2227 u64 bytenr = node->bytenr;
2228 u64 num_bytes = node->num_bytes;
2232 path = btrfs_alloc_path();
2236 path->reada = READA_FORWARD;
2237 path->leave_spinning = 1;
2238 /* this will setup the path even if it fails to insert the back ref */
2239 ret = insert_inline_extent_backref(trans, fs_info, path, bytenr,
2240 num_bytes, parent, root_objectid,
2242 refs_to_add, extent_op);
2243 if ((ret < 0 && ret != -EAGAIN) || !ret)
2247 * Ok we had -EAGAIN which means we didn't have space to insert and
2248 * inline extent ref, so just update the reference count and add a
2251 leaf = path->nodes[0];
2252 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2253 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2254 refs = btrfs_extent_refs(leaf, item);
2255 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2257 __run_delayed_extent_op(extent_op, leaf, item);
2259 btrfs_mark_buffer_dirty(leaf);
2260 btrfs_release_path(path);
2262 path->reada = READA_FORWARD;
2263 path->leave_spinning = 1;
2264 /* now insert the actual backref */
2265 ret = insert_extent_backref(trans, fs_info, path, bytenr, parent,
2266 root_objectid, owner, offset, refs_to_add);
2268 btrfs_abort_transaction(trans, ret);
2270 btrfs_free_path(path);
2274 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2275 struct btrfs_fs_info *fs_info,
2276 struct btrfs_delayed_ref_node *node,
2277 struct btrfs_delayed_extent_op *extent_op,
2278 int insert_reserved)
2281 struct btrfs_delayed_data_ref *ref;
2282 struct btrfs_key ins;
2287 ins.objectid = node->bytenr;
2288 ins.offset = node->num_bytes;
2289 ins.type = BTRFS_EXTENT_ITEM_KEY;
2291 ref = btrfs_delayed_node_to_data_ref(node);
2292 trace_run_delayed_data_ref(fs_info, node, ref, node->action);
2294 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2295 parent = ref->parent;
2296 ref_root = ref->root;
2298 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2300 flags |= extent_op->flags_to_set;
2301 ret = alloc_reserved_file_extent(trans, fs_info,
2302 parent, ref_root, flags,
2303 ref->objectid, ref->offset,
2304 &ins, node->ref_mod);
2305 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2306 ret = __btrfs_inc_extent_ref(trans, fs_info, node, parent,
2307 ref_root, ref->objectid,
2308 ref->offset, node->ref_mod,
2310 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2311 ret = __btrfs_free_extent(trans, fs_info, node, parent,
2312 ref_root, ref->objectid,
2313 ref->offset, node->ref_mod,
2321 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2322 struct extent_buffer *leaf,
2323 struct btrfs_extent_item *ei)
2325 u64 flags = btrfs_extent_flags(leaf, ei);
2326 if (extent_op->update_flags) {
2327 flags |= extent_op->flags_to_set;
2328 btrfs_set_extent_flags(leaf, ei, flags);
2331 if (extent_op->update_key) {
2332 struct btrfs_tree_block_info *bi;
2333 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2334 bi = (struct btrfs_tree_block_info *)(ei + 1);
2335 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2339 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2340 struct btrfs_fs_info *fs_info,
2341 struct btrfs_delayed_ref_head *head,
2342 struct btrfs_delayed_extent_op *extent_op)
2344 struct btrfs_key key;
2345 struct btrfs_path *path;
2346 struct btrfs_extent_item *ei;
2347 struct extent_buffer *leaf;
2351 int metadata = !extent_op->is_data;
2356 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2359 path = btrfs_alloc_path();
2363 key.objectid = head->bytenr;
2366 key.type = BTRFS_METADATA_ITEM_KEY;
2367 key.offset = extent_op->level;
2369 key.type = BTRFS_EXTENT_ITEM_KEY;
2370 key.offset = head->num_bytes;
2374 path->reada = READA_FORWARD;
2375 path->leave_spinning = 1;
2376 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1);
2383 if (path->slots[0] > 0) {
2385 btrfs_item_key_to_cpu(path->nodes[0], &key,
2387 if (key.objectid == head->bytenr &&
2388 key.type == BTRFS_EXTENT_ITEM_KEY &&
2389 key.offset == head->num_bytes)
2393 btrfs_release_path(path);
2396 key.objectid = head->bytenr;
2397 key.offset = head->num_bytes;
2398 key.type = BTRFS_EXTENT_ITEM_KEY;
2407 leaf = path->nodes[0];
2408 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2409 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2410 if (item_size < sizeof(*ei)) {
2411 ret = convert_extent_item_v0(trans, fs_info, path, (u64)-1, 0);
2416 leaf = path->nodes[0];
2417 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2420 BUG_ON(item_size < sizeof(*ei));
2421 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2422 __run_delayed_extent_op(extent_op, leaf, ei);
2424 btrfs_mark_buffer_dirty(leaf);
2426 btrfs_free_path(path);
2430 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2431 struct btrfs_fs_info *fs_info,
2432 struct btrfs_delayed_ref_node *node,
2433 struct btrfs_delayed_extent_op *extent_op,
2434 int insert_reserved)
2437 struct btrfs_delayed_tree_ref *ref;
2438 struct btrfs_key ins;
2441 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
2443 ref = btrfs_delayed_node_to_tree_ref(node);
2444 trace_run_delayed_tree_ref(fs_info, node, ref, node->action);
2446 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2447 parent = ref->parent;
2448 ref_root = ref->root;
2450 ins.objectid = node->bytenr;
2451 if (skinny_metadata) {
2452 ins.offset = ref->level;
2453 ins.type = BTRFS_METADATA_ITEM_KEY;
2455 ins.offset = node->num_bytes;
2456 ins.type = BTRFS_EXTENT_ITEM_KEY;
2459 if (node->ref_mod != 1) {
2461 "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu",
2462 node->bytenr, node->ref_mod, node->action, ref_root,
2466 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2467 BUG_ON(!extent_op || !extent_op->update_flags);
2468 ret = alloc_reserved_tree_block(trans, fs_info,
2470 extent_op->flags_to_set,
2473 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2474 ret = __btrfs_inc_extent_ref(trans, fs_info, node,
2478 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2479 ret = __btrfs_free_extent(trans, fs_info, node,
2481 ref->level, 0, 1, extent_op);
2488 /* helper function to actually process a single delayed ref entry */
2489 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2490 struct btrfs_fs_info *fs_info,
2491 struct btrfs_delayed_ref_node *node,
2492 struct btrfs_delayed_extent_op *extent_op,
2493 int insert_reserved)
2497 if (trans->aborted) {
2498 if (insert_reserved)
2499 btrfs_pin_extent(fs_info, node->bytenr,
2500 node->num_bytes, 1);
2504 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2505 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2506 ret = run_delayed_tree_ref(trans, fs_info, node, extent_op,
2508 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2509 node->type == BTRFS_SHARED_DATA_REF_KEY)
2510 ret = run_delayed_data_ref(trans, fs_info, node, extent_op,
2517 static inline struct btrfs_delayed_ref_node *
2518 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2520 struct btrfs_delayed_ref_node *ref;
2522 if (RB_EMPTY_ROOT(&head->ref_tree))
2526 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2527 * This is to prevent a ref count from going down to zero, which deletes
2528 * the extent item from the extent tree, when there still are references
2529 * to add, which would fail because they would not find the extent item.
2531 if (!list_empty(&head->ref_add_list))
2532 return list_first_entry(&head->ref_add_list,
2533 struct btrfs_delayed_ref_node, add_list);
2535 ref = rb_entry(rb_first(&head->ref_tree),
2536 struct btrfs_delayed_ref_node, ref_node);
2537 ASSERT(list_empty(&ref->add_list));
2541 static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
2542 struct btrfs_delayed_ref_head *head)
2544 spin_lock(&delayed_refs->lock);
2545 head->processing = 0;
2546 delayed_refs->num_heads_ready++;
2547 spin_unlock(&delayed_refs->lock);
2548 btrfs_delayed_ref_unlock(head);
2551 static int cleanup_extent_op(struct btrfs_trans_handle *trans,
2552 struct btrfs_fs_info *fs_info,
2553 struct btrfs_delayed_ref_head *head)
2555 struct btrfs_delayed_extent_op *extent_op = head->extent_op;
2560 head->extent_op = NULL;
2561 if (head->must_insert_reserved) {
2562 btrfs_free_delayed_extent_op(extent_op);
2565 spin_unlock(&head->lock);
2566 ret = run_delayed_extent_op(trans, fs_info, head, extent_op);
2567 btrfs_free_delayed_extent_op(extent_op);
2568 return ret ? ret : 1;
2571 static int cleanup_ref_head(struct btrfs_trans_handle *trans,
2572 struct btrfs_fs_info *fs_info,
2573 struct btrfs_delayed_ref_head *head)
2575 struct btrfs_delayed_ref_root *delayed_refs;
2578 delayed_refs = &trans->transaction->delayed_refs;
2580 ret = cleanup_extent_op(trans, fs_info, head);
2582 unselect_delayed_ref_head(delayed_refs, head);
2583 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2590 * Need to drop our head ref lock and re-acquire the delayed ref lock
2591 * and then re-check to make sure nobody got added.
2593 spin_unlock(&head->lock);
2594 spin_lock(&delayed_refs->lock);
2595 spin_lock(&head->lock);
2596 if (!RB_EMPTY_ROOT(&head->ref_tree) || head->extent_op) {
2597 spin_unlock(&head->lock);
2598 spin_unlock(&delayed_refs->lock);
2601 delayed_refs->num_heads--;
2602 rb_erase(&head->href_node, &delayed_refs->href_root);
2603 RB_CLEAR_NODE(&head->href_node);
2604 spin_unlock(&delayed_refs->lock);
2605 spin_unlock(&head->lock);
2606 atomic_dec(&delayed_refs->num_entries);
2608 trace_run_delayed_ref_head(fs_info, head, 0);
2610 if (head->total_ref_mod < 0) {
2611 struct btrfs_block_group_cache *cache;
2613 cache = btrfs_lookup_block_group(fs_info, head->bytenr);
2615 percpu_counter_add(&cache->space_info->total_bytes_pinned,
2617 btrfs_put_block_group(cache);
2619 if (head->is_data) {
2620 spin_lock(&delayed_refs->lock);
2621 delayed_refs->pending_csums -= head->num_bytes;
2622 spin_unlock(&delayed_refs->lock);
2626 if (head->must_insert_reserved) {
2627 btrfs_pin_extent(fs_info, head->bytenr,
2628 head->num_bytes, 1);
2629 if (head->is_data) {
2630 ret = btrfs_del_csums(trans, fs_info, head->bytenr,
2635 /* Also free its reserved qgroup space */
2636 btrfs_qgroup_free_delayed_ref(fs_info, head->qgroup_ref_root,
2637 head->qgroup_reserved);
2638 btrfs_delayed_ref_unlock(head);
2639 btrfs_put_delayed_ref_head(head);
2644 * Returns 0 on success or if called with an already aborted transaction.
2645 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2647 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2648 struct btrfs_fs_info *fs_info,
2651 struct btrfs_delayed_ref_root *delayed_refs;
2652 struct btrfs_delayed_ref_node *ref;
2653 struct btrfs_delayed_ref_head *locked_ref = NULL;
2654 struct btrfs_delayed_extent_op *extent_op;
2655 ktime_t start = ktime_get();
2657 unsigned long count = 0;
2658 unsigned long actual_count = 0;
2659 int must_insert_reserved = 0;
2661 delayed_refs = &trans->transaction->delayed_refs;
2667 spin_lock(&delayed_refs->lock);
2668 locked_ref = btrfs_select_ref_head(trans);
2670 spin_unlock(&delayed_refs->lock);
2674 /* grab the lock that says we are going to process
2675 * all the refs for this head */
2676 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2677 spin_unlock(&delayed_refs->lock);
2679 * we may have dropped the spin lock to get the head
2680 * mutex lock, and that might have given someone else
2681 * time to free the head. If that's true, it has been
2682 * removed from our list and we can move on.
2684 if (ret == -EAGAIN) {
2692 * We need to try and merge add/drops of the same ref since we
2693 * can run into issues with relocate dropping the implicit ref
2694 * and then it being added back again before the drop can
2695 * finish. If we merged anything we need to re-loop so we can
2697 * Or we can get node references of the same type that weren't
2698 * merged when created due to bumps in the tree mod seq, and
2699 * we need to merge them to prevent adding an inline extent
2700 * backref before dropping it (triggering a BUG_ON at
2701 * insert_inline_extent_backref()).
2703 spin_lock(&locked_ref->lock);
2704 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2708 * locked_ref is the head node, so we have to go one
2709 * node back for any delayed ref updates
2711 ref = select_delayed_ref(locked_ref);
2713 if (ref && ref->seq &&
2714 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2715 spin_unlock(&locked_ref->lock);
2716 unselect_delayed_ref_head(delayed_refs, locked_ref);
2724 * We're done processing refs in this ref_head, clean everything
2725 * up and move on to the next ref_head.
2728 ret = cleanup_ref_head(trans, fs_info, locked_ref);
2730 /* We dropped our lock, we need to loop. */
2743 rb_erase(&ref->ref_node, &locked_ref->ref_tree);
2744 RB_CLEAR_NODE(&ref->ref_node);
2745 if (!list_empty(&ref->add_list))
2746 list_del(&ref->add_list);
2748 * When we play the delayed ref, also correct the ref_mod on
2751 switch (ref->action) {
2752 case BTRFS_ADD_DELAYED_REF:
2753 case BTRFS_ADD_DELAYED_EXTENT:
2754 locked_ref->ref_mod -= ref->ref_mod;
2756 case BTRFS_DROP_DELAYED_REF:
2757 locked_ref->ref_mod += ref->ref_mod;
2762 atomic_dec(&delayed_refs->num_entries);
2765 * Record the must-insert_reserved flag before we drop the spin
2768 must_insert_reserved = locked_ref->must_insert_reserved;
2769 locked_ref->must_insert_reserved = 0;
2771 extent_op = locked_ref->extent_op;
2772 locked_ref->extent_op = NULL;
2773 spin_unlock(&locked_ref->lock);
2775 ret = run_one_delayed_ref(trans, fs_info, ref, extent_op,
2776 must_insert_reserved);
2778 btrfs_free_delayed_extent_op(extent_op);
2780 unselect_delayed_ref_head(delayed_refs, locked_ref);
2781 btrfs_put_delayed_ref(ref);
2782 btrfs_debug(fs_info, "run_one_delayed_ref returned %d",
2787 btrfs_put_delayed_ref(ref);
2793 * We don't want to include ref heads since we can have empty ref heads
2794 * and those will drastically skew our runtime down since we just do
2795 * accounting, no actual extent tree updates.
2797 if (actual_count > 0) {
2798 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2802 * We weigh the current average higher than our current runtime
2803 * to avoid large swings in the average.
2805 spin_lock(&delayed_refs->lock);
2806 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2807 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2808 spin_unlock(&delayed_refs->lock);
2813 #ifdef SCRAMBLE_DELAYED_REFS
2815 * Normally delayed refs get processed in ascending bytenr order. This
2816 * correlates in most cases to the order added. To expose dependencies on this
2817 * order, we start to process the tree in the middle instead of the beginning
2819 static u64 find_middle(struct rb_root *root)
2821 struct rb_node *n = root->rb_node;
2822 struct btrfs_delayed_ref_node *entry;
2825 u64 first = 0, last = 0;
2829 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2830 first = entry->bytenr;
2834 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2835 last = entry->bytenr;
2840 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2841 WARN_ON(!entry->in_tree);
2843 middle = entry->bytenr;
2856 static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads)
2860 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2861 sizeof(struct btrfs_extent_inline_ref));
2862 if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2863 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2866 * We don't ever fill up leaves all the way so multiply by 2 just to be
2867 * closer to what we're really going to want to use.
2869 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info));
2873 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2874 * would require to store the csums for that many bytes.
2876 u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes)
2879 u64 num_csums_per_leaf;
2882 csum_size = BTRFS_MAX_ITEM_SIZE(fs_info);
2883 num_csums_per_leaf = div64_u64(csum_size,
2884 (u64)btrfs_super_csum_size(fs_info->super_copy));
2885 num_csums = div64_u64(csum_bytes, fs_info->sectorsize);
2886 num_csums += num_csums_per_leaf - 1;
2887 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2891 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2892 struct btrfs_fs_info *fs_info)
2894 struct btrfs_block_rsv *global_rsv;
2895 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2896 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2897 u64 num_dirty_bgs = trans->transaction->num_dirty_bgs;
2898 u64 num_bytes, num_dirty_bgs_bytes;
2901 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
2902 num_heads = heads_to_leaves(fs_info, num_heads);
2904 num_bytes += (num_heads - 1) * fs_info->nodesize;
2906 num_bytes += btrfs_csum_bytes_to_leaves(fs_info, csum_bytes) *
2908 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(fs_info,
2910 global_rsv = &fs_info->global_block_rsv;
2913 * If we can't allocate any more chunks lets make sure we have _lots_ of
2914 * wiggle room since running delayed refs can create more delayed refs.
2916 if (global_rsv->space_info->full) {
2917 num_dirty_bgs_bytes <<= 1;
2921 spin_lock(&global_rsv->lock);
2922 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2924 spin_unlock(&global_rsv->lock);
2928 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2929 struct btrfs_fs_info *fs_info)
2932 atomic_read(&trans->transaction->delayed_refs.num_entries);
2937 avg_runtime = fs_info->avg_delayed_ref_runtime;
2938 val = num_entries * avg_runtime;
2939 if (val >= NSEC_PER_SEC)
2941 if (val >= NSEC_PER_SEC / 2)
2944 return btrfs_check_space_for_delayed_refs(trans, fs_info);
2947 struct async_delayed_refs {
2948 struct btrfs_root *root;
2953 struct completion wait;
2954 struct btrfs_work work;
2957 static inline struct async_delayed_refs *
2958 to_async_delayed_refs(struct btrfs_work *work)
2960 return container_of(work, struct async_delayed_refs, work);
2963 static void delayed_ref_async_start(struct btrfs_work *work)
2965 struct async_delayed_refs *async = to_async_delayed_refs(work);
2966 struct btrfs_trans_handle *trans;
2967 struct btrfs_fs_info *fs_info = async->root->fs_info;
2970 /* if the commit is already started, we don't need to wait here */
2971 if (btrfs_transaction_blocked(fs_info))
2974 trans = btrfs_join_transaction(async->root);
2975 if (IS_ERR(trans)) {
2976 async->error = PTR_ERR(trans);
2981 * trans->sync means that when we call end_transaction, we won't
2982 * wait on delayed refs
2986 /* Don't bother flushing if we got into a different transaction */
2987 if (trans->transid > async->transid)
2990 ret = btrfs_run_delayed_refs(trans, fs_info, async->count);
2994 ret = btrfs_end_transaction(trans);
2995 if (ret && !async->error)
2999 complete(&async->wait);
3004 int btrfs_async_run_delayed_refs(struct btrfs_fs_info *fs_info,
3005 unsigned long count, u64 transid, int wait)
3007 struct async_delayed_refs *async;
3010 async = kmalloc(sizeof(*async), GFP_NOFS);
3014 async->root = fs_info->tree_root;
3015 async->count = count;
3017 async->transid = transid;
3022 init_completion(&async->wait);
3024 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
3025 delayed_ref_async_start, NULL, NULL);
3027 btrfs_queue_work(fs_info->extent_workers, &async->work);
3030 wait_for_completion(&async->wait);
3039 * this starts processing the delayed reference count updates and
3040 * extent insertions we have queued up so far. count can be
3041 * 0, which means to process everything in the tree at the start
3042 * of the run (but not newly added entries), or it can be some target
3043 * number you'd like to process.
3045 * Returns 0 on success or if called with an aborted transaction
3046 * Returns <0 on error and aborts the transaction
3048 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
3049 struct btrfs_fs_info *fs_info, unsigned long count)
3051 struct rb_node *node;
3052 struct btrfs_delayed_ref_root *delayed_refs;
3053 struct btrfs_delayed_ref_head *head;
3055 int run_all = count == (unsigned long)-1;
3056 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
3058 /* We'll clean this up in btrfs_cleanup_transaction */
3062 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
3065 delayed_refs = &trans->transaction->delayed_refs;
3067 count = atomic_read(&delayed_refs->num_entries) * 2;
3070 #ifdef SCRAMBLE_DELAYED_REFS
3071 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
3073 trans->can_flush_pending_bgs = false;
3074 ret = __btrfs_run_delayed_refs(trans, fs_info, count);
3076 btrfs_abort_transaction(trans, ret);
3081 if (!list_empty(&trans->new_bgs))
3082 btrfs_create_pending_block_groups(trans, fs_info);
3084 spin_lock(&delayed_refs->lock);
3085 node = rb_first(&delayed_refs->href_root);
3087 spin_unlock(&delayed_refs->lock);
3090 head = rb_entry(node, struct btrfs_delayed_ref_head,
3092 refcount_inc(&head->refs);
3093 spin_unlock(&delayed_refs->lock);
3095 /* Mutex was contended, block until it's released and retry. */
3096 mutex_lock(&head->mutex);
3097 mutex_unlock(&head->mutex);
3099 btrfs_put_delayed_ref_head(head);
3104 trans->can_flush_pending_bgs = can_flush_pending_bgs;
3108 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
3109 struct btrfs_fs_info *fs_info,
3110 u64 bytenr, u64 num_bytes, u64 flags,
3111 int level, int is_data)
3113 struct btrfs_delayed_extent_op *extent_op;
3116 extent_op = btrfs_alloc_delayed_extent_op();
3120 extent_op->flags_to_set = flags;
3121 extent_op->update_flags = true;
3122 extent_op->update_key = false;
3123 extent_op->is_data = is_data ? true : false;
3124 extent_op->level = level;
3126 ret = btrfs_add_delayed_extent_op(fs_info, trans, bytenr,
3127 num_bytes, extent_op);
3129 btrfs_free_delayed_extent_op(extent_op);
3133 static noinline int check_delayed_ref(struct btrfs_root *root,
3134 struct btrfs_path *path,
3135 u64 objectid, u64 offset, u64 bytenr)
3137 struct btrfs_delayed_ref_head *head;
3138 struct btrfs_delayed_ref_node *ref;
3139 struct btrfs_delayed_data_ref *data_ref;
3140 struct btrfs_delayed_ref_root *delayed_refs;
3141 struct btrfs_transaction *cur_trans;
3142 struct rb_node *node;
3145 cur_trans = root->fs_info->running_transaction;
3149 delayed_refs = &cur_trans->delayed_refs;
3150 spin_lock(&delayed_refs->lock);
3151 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
3153 spin_unlock(&delayed_refs->lock);
3157 if (!mutex_trylock(&head->mutex)) {
3158 refcount_inc(&head->refs);
3159 spin_unlock(&delayed_refs->lock);
3161 btrfs_release_path(path);
3164 * Mutex was contended, block until it's released and let
3167 mutex_lock(&head->mutex);
3168 mutex_unlock(&head->mutex);
3169 btrfs_put_delayed_ref_head(head);
3172 spin_unlock(&delayed_refs->lock);
3174 spin_lock(&head->lock);
3176 * XXX: We should replace this with a proper search function in the
3179 for (node = rb_first(&head->ref_tree); node; node = rb_next(node)) {
3180 ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node);
3181 /* If it's a shared ref we know a cross reference exists */
3182 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
3187 data_ref = btrfs_delayed_node_to_data_ref(ref);
3190 * If our ref doesn't match the one we're currently looking at
3191 * then we have a cross reference.
3193 if (data_ref->root != root->root_key.objectid ||
3194 data_ref->objectid != objectid ||
3195 data_ref->offset != offset) {
3200 spin_unlock(&head->lock);
3201 mutex_unlock(&head->mutex);
3205 static noinline int check_committed_ref(struct btrfs_root *root,
3206 struct btrfs_path *path,
3207 u64 objectid, u64 offset, u64 bytenr)
3209 struct btrfs_fs_info *fs_info = root->fs_info;
3210 struct btrfs_root *extent_root = fs_info->extent_root;
3211 struct extent_buffer *leaf;
3212 struct btrfs_extent_data_ref *ref;
3213 struct btrfs_extent_inline_ref *iref;
3214 struct btrfs_extent_item *ei;
3215 struct btrfs_key key;
3220 key.objectid = bytenr;
3221 key.offset = (u64)-1;
3222 key.type = BTRFS_EXTENT_ITEM_KEY;
3224 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3227 BUG_ON(ret == 0); /* Corruption */
3230 if (path->slots[0] == 0)
3234 leaf = path->nodes[0];
3235 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3237 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3241 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3242 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3243 if (item_size < sizeof(*ei)) {
3244 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3248 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3250 if (item_size != sizeof(*ei) +
3251 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3254 if (btrfs_extent_generation(leaf, ei) <=
3255 btrfs_root_last_snapshot(&root->root_item))
3258 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3260 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
3261 if (type != BTRFS_EXTENT_DATA_REF_KEY)
3264 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3265 if (btrfs_extent_refs(leaf, ei) !=
3266 btrfs_extent_data_ref_count(leaf, ref) ||
3267 btrfs_extent_data_ref_root(leaf, ref) !=
3268 root->root_key.objectid ||
3269 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3270 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3278 int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
3281 struct btrfs_path *path;
3285 path = btrfs_alloc_path();
3290 ret = check_committed_ref(root, path, objectid,
3292 if (ret && ret != -ENOENT)
3295 ret2 = check_delayed_ref(root, path, objectid,
3297 } while (ret2 == -EAGAIN);
3299 if (ret2 && ret2 != -ENOENT) {
3304 if (ret != -ENOENT || ret2 != -ENOENT)
3307 btrfs_free_path(path);
3308 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3313 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3314 struct btrfs_root *root,
3315 struct extent_buffer *buf,
3316 int full_backref, int inc)
3318 struct btrfs_fs_info *fs_info = root->fs_info;
3324 struct btrfs_key key;
3325 struct btrfs_file_extent_item *fi;
3329 int (*process_func)(struct btrfs_trans_handle *,
3330 struct btrfs_root *,
3331 u64, u64, u64, u64, u64, u64);
3334 if (btrfs_is_testing(fs_info))
3337 ref_root = btrfs_header_owner(buf);
3338 nritems = btrfs_header_nritems(buf);
3339 level = btrfs_header_level(buf);
3341 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3345 process_func = btrfs_inc_extent_ref;
3347 process_func = btrfs_free_extent;
3350 parent = buf->start;
3354 for (i = 0; i < nritems; i++) {
3356 btrfs_item_key_to_cpu(buf, &key, i);
3357 if (key.type != BTRFS_EXTENT_DATA_KEY)
3359 fi = btrfs_item_ptr(buf, i,
3360 struct btrfs_file_extent_item);
3361 if (btrfs_file_extent_type(buf, fi) ==
3362 BTRFS_FILE_EXTENT_INLINE)
3364 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3368 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3369 key.offset -= btrfs_file_extent_offset(buf, fi);
3370 ret = process_func(trans, root, bytenr, num_bytes,
3371 parent, ref_root, key.objectid,
3376 bytenr = btrfs_node_blockptr(buf, i);
3377 num_bytes = fs_info->nodesize;
3378 ret = process_func(trans, root, bytenr, num_bytes,
3379 parent, ref_root, level - 1, 0);
3389 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3390 struct extent_buffer *buf, int full_backref)
3392 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3395 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3396 struct extent_buffer *buf, int full_backref)
3398 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3401 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3402 struct btrfs_fs_info *fs_info,
3403 struct btrfs_path *path,
3404 struct btrfs_block_group_cache *cache)
3407 struct btrfs_root *extent_root = fs_info->extent_root;
3409 struct extent_buffer *leaf;
3411 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3418 leaf = path->nodes[0];
3419 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3420 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3421 btrfs_mark_buffer_dirty(leaf);
3423 btrfs_release_path(path);
3428 static struct btrfs_block_group_cache *
3429 next_block_group(struct btrfs_fs_info *fs_info,
3430 struct btrfs_block_group_cache *cache)
3432 struct rb_node *node;
3434 spin_lock(&fs_info->block_group_cache_lock);
3436 /* If our block group was removed, we need a full search. */
3437 if (RB_EMPTY_NODE(&cache->cache_node)) {
3438 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3440 spin_unlock(&fs_info->block_group_cache_lock);
3441 btrfs_put_block_group(cache);
3442 cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache;
3444 node = rb_next(&cache->cache_node);
3445 btrfs_put_block_group(cache);
3447 cache = rb_entry(node, struct btrfs_block_group_cache,
3449 btrfs_get_block_group(cache);
3452 spin_unlock(&fs_info->block_group_cache_lock);
3456 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3457 struct btrfs_trans_handle *trans,
3458 struct btrfs_path *path)
3460 struct btrfs_fs_info *fs_info = block_group->fs_info;
3461 struct btrfs_root *root = fs_info->tree_root;
3462 struct inode *inode = NULL;
3463 struct extent_changeset *data_reserved = NULL;
3465 int dcs = BTRFS_DC_ERROR;
3471 * If this block group is smaller than 100 megs don't bother caching the
3474 if (block_group->key.offset < (100 * SZ_1M)) {
3475 spin_lock(&block_group->lock);
3476 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3477 spin_unlock(&block_group->lock);
3484 inode = lookup_free_space_inode(fs_info, block_group, path);
3485 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3486 ret = PTR_ERR(inode);
3487 btrfs_release_path(path);
3491 if (IS_ERR(inode)) {
3495 if (block_group->ro)
3498 ret = create_free_space_inode(fs_info, trans, block_group,
3505 /* We've already setup this transaction, go ahead and exit */
3506 if (block_group->cache_generation == trans->transid &&
3507 i_size_read(inode)) {
3508 dcs = BTRFS_DC_SETUP;
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 if (i_size_read(inode) > 0) {
3536 ret = btrfs_check_trunc_cache_free_space(fs_info,
3537 &fs_info->global_block_rsv);
3541 ret = btrfs_truncate_free_space_cache(trans, NULL, inode);
3546 spin_lock(&block_group->lock);
3547 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3548 !btrfs_test_opt(fs_info, SPACE_CACHE)) {
3550 * don't bother trying to write stuff out _if_
3551 * a) we're not cached,
3552 * b) we're with nospace_cache mount option,
3553 * c) we're with v2 space_cache (FREE_SPACE_TREE).
3555 dcs = BTRFS_DC_WRITTEN;
3556 spin_unlock(&block_group->lock);
3559 spin_unlock(&block_group->lock);
3562 * We hit an ENOSPC when setting up the cache in this transaction, just
3563 * skip doing the setup, we've already cleared the cache so we're safe.
3565 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
3571 * Try to preallocate enough space based on how big the block group is.
3572 * Keep in mind this has to include any pinned space which could end up
3573 * taking up quite a bit since it's not folded into the other space
3576 num_pages = div_u64(block_group->key.offset, SZ_256M);
3581 num_pages *= PAGE_SIZE;
3583 ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages);
3587 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3588 num_pages, num_pages,
3591 * Our cache requires contiguous chunks so that we don't modify a bunch
3592 * of metadata or split extents when writing the cache out, which means
3593 * we can enospc if we are heavily fragmented in addition to just normal
3594 * out of space conditions. So if we hit this just skip setting up any
3595 * other block groups for this transaction, maybe we'll unpin enough
3596 * space the next time around.
3599 dcs = BTRFS_DC_SETUP;
3600 else if (ret == -ENOSPC)
3601 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
3606 btrfs_release_path(path);
3608 spin_lock(&block_group->lock);
3609 if (!ret && dcs == BTRFS_DC_SETUP)
3610 block_group->cache_generation = trans->transid;
3611 block_group->disk_cache_state = dcs;
3612 spin_unlock(&block_group->lock);
3614 extent_changeset_free(data_reserved);
3618 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3619 struct btrfs_fs_info *fs_info)
3621 struct btrfs_block_group_cache *cache, *tmp;
3622 struct btrfs_transaction *cur_trans = trans->transaction;
3623 struct btrfs_path *path;
3625 if (list_empty(&cur_trans->dirty_bgs) ||
3626 !btrfs_test_opt(fs_info, SPACE_CACHE))
3629 path = btrfs_alloc_path();
3633 /* Could add new block groups, use _safe just in case */
3634 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3636 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3637 cache_save_setup(cache, trans, path);
3640 btrfs_free_path(path);
3645 * transaction commit does final block group cache writeback during a
3646 * critical section where nothing is allowed to change the FS. This is
3647 * required in order for the cache to actually match the block group,
3648 * but can introduce a lot of latency into the commit.
3650 * So, btrfs_start_dirty_block_groups is here to kick off block group
3651 * cache IO. There's a chance we'll have to redo some of it if the
3652 * block group changes again during the commit, but it greatly reduces
3653 * the commit latency by getting rid of the easy block groups while
3654 * we're still allowing others to join the commit.
3656 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans,
3657 struct btrfs_fs_info *fs_info)
3659 struct btrfs_block_group_cache *cache;
3660 struct btrfs_transaction *cur_trans = trans->transaction;
3663 struct btrfs_path *path = NULL;
3665 struct list_head *io = &cur_trans->io_bgs;
3666 int num_started = 0;
3669 spin_lock(&cur_trans->dirty_bgs_lock);
3670 if (list_empty(&cur_trans->dirty_bgs)) {
3671 spin_unlock(&cur_trans->dirty_bgs_lock);
3674 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3675 spin_unlock(&cur_trans->dirty_bgs_lock);
3679 * make sure all the block groups on our dirty list actually
3682 btrfs_create_pending_block_groups(trans, fs_info);
3685 path = btrfs_alloc_path();
3691 * cache_write_mutex is here only to save us from balance or automatic
3692 * removal of empty block groups deleting this block group while we are
3693 * writing out the cache
3695 mutex_lock(&trans->transaction->cache_write_mutex);
3696 while (!list_empty(&dirty)) {
3697 cache = list_first_entry(&dirty,
3698 struct btrfs_block_group_cache,
3701 * this can happen if something re-dirties a block
3702 * group that is already under IO. Just wait for it to
3703 * finish and then do it all again
3705 if (!list_empty(&cache->io_list)) {
3706 list_del_init(&cache->io_list);
3707 btrfs_wait_cache_io(trans, cache, path);
3708 btrfs_put_block_group(cache);
3713 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3714 * if it should update the cache_state. Don't delete
3715 * until after we wait.
3717 * Since we're not running in the commit critical section
3718 * we need the dirty_bgs_lock to protect from update_block_group
3720 spin_lock(&cur_trans->dirty_bgs_lock);
3721 list_del_init(&cache->dirty_list);
3722 spin_unlock(&cur_trans->dirty_bgs_lock);
3726 cache_save_setup(cache, trans, path);
3728 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3729 cache->io_ctl.inode = NULL;
3730 ret = btrfs_write_out_cache(fs_info, trans,
3732 if (ret == 0 && cache->io_ctl.inode) {
3737 * the cache_write_mutex is protecting
3740 list_add_tail(&cache->io_list, io);
3743 * if we failed to write the cache, the
3744 * generation will be bad and life goes on
3750 ret = write_one_cache_group(trans, fs_info,
3753 * Our block group might still be attached to the list
3754 * of new block groups in the transaction handle of some
3755 * other task (struct btrfs_trans_handle->new_bgs). This
3756 * means its block group item isn't yet in the extent
3757 * tree. If this happens ignore the error, as we will
3758 * try again later in the critical section of the
3759 * transaction commit.
3761 if (ret == -ENOENT) {
3763 spin_lock(&cur_trans->dirty_bgs_lock);
3764 if (list_empty(&cache->dirty_list)) {
3765 list_add_tail(&cache->dirty_list,
3766 &cur_trans->dirty_bgs);
3767 btrfs_get_block_group(cache);
3769 spin_unlock(&cur_trans->dirty_bgs_lock);
3771 btrfs_abort_transaction(trans, ret);
3775 /* if its not on the io list, we need to put the block group */
3777 btrfs_put_block_group(cache);
3783 * Avoid blocking other tasks for too long. It might even save
3784 * us from writing caches for block groups that are going to be
3787 mutex_unlock(&trans->transaction->cache_write_mutex);
3788 mutex_lock(&trans->transaction->cache_write_mutex);
3790 mutex_unlock(&trans->transaction->cache_write_mutex);
3793 * go through delayed refs for all the stuff we've just kicked off
3794 * and then loop back (just once)
3796 ret = btrfs_run_delayed_refs(trans, fs_info, 0);
3797 if (!ret && loops == 0) {
3799 spin_lock(&cur_trans->dirty_bgs_lock);
3800 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3802 * dirty_bgs_lock protects us from concurrent block group
3803 * deletes too (not just cache_write_mutex).
3805 if (!list_empty(&dirty)) {
3806 spin_unlock(&cur_trans->dirty_bgs_lock);
3809 spin_unlock(&cur_trans->dirty_bgs_lock);
3810 } else if (ret < 0) {
3811 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
3814 btrfs_free_path(path);
3818 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3819 struct btrfs_fs_info *fs_info)
3821 struct btrfs_block_group_cache *cache;
3822 struct btrfs_transaction *cur_trans = trans->transaction;
3825 struct btrfs_path *path;
3826 struct list_head *io = &cur_trans->io_bgs;
3827 int num_started = 0;
3829 path = btrfs_alloc_path();
3834 * Even though we are in the critical section of the transaction commit,
3835 * we can still have concurrent tasks adding elements to this
3836 * transaction's list of dirty block groups. These tasks correspond to
3837 * endio free space workers started when writeback finishes for a
3838 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3839 * allocate new block groups as a result of COWing nodes of the root
3840 * tree when updating the free space inode. The writeback for the space
3841 * caches is triggered by an earlier call to
3842 * btrfs_start_dirty_block_groups() and iterations of the following
3844 * Also we want to do the cache_save_setup first and then run the
3845 * delayed refs to make sure we have the best chance at doing this all
3848 spin_lock(&cur_trans->dirty_bgs_lock);
3849 while (!list_empty(&cur_trans->dirty_bgs)) {
3850 cache = list_first_entry(&cur_trans->dirty_bgs,
3851 struct btrfs_block_group_cache,
3855 * this can happen if cache_save_setup re-dirties a block
3856 * group that is already under IO. Just wait for it to
3857 * finish and then do it all again
3859 if (!list_empty(&cache->io_list)) {
3860 spin_unlock(&cur_trans->dirty_bgs_lock);
3861 list_del_init(&cache->io_list);
3862 btrfs_wait_cache_io(trans, cache, path);
3863 btrfs_put_block_group(cache);
3864 spin_lock(&cur_trans->dirty_bgs_lock);
3868 * don't remove from the dirty list until after we've waited
3871 list_del_init(&cache->dirty_list);
3872 spin_unlock(&cur_trans->dirty_bgs_lock);
3875 cache_save_setup(cache, trans, path);
3878 ret = btrfs_run_delayed_refs(trans, fs_info,
3879 (unsigned long) -1);
3881 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3882 cache->io_ctl.inode = NULL;
3883 ret = btrfs_write_out_cache(fs_info, trans,
3885 if (ret == 0 && cache->io_ctl.inode) {
3888 list_add_tail(&cache->io_list, io);
3891 * if we failed to write the cache, the
3892 * generation will be bad and life goes on
3898 ret = write_one_cache_group(trans, fs_info,
3901 * One of the free space endio workers might have
3902 * created a new block group while updating a free space
3903 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3904 * and hasn't released its transaction handle yet, in
3905 * which case the new block group is still attached to
3906 * its transaction handle and its creation has not
3907 * finished yet (no block group item in the extent tree
3908 * yet, etc). If this is the case, wait for all free
3909 * space endio workers to finish and retry. This is a
3910 * a very rare case so no need for a more efficient and
3913 if (ret == -ENOENT) {
3914 wait_event(cur_trans->writer_wait,
3915 atomic_read(&cur_trans->num_writers) == 1);
3916 ret = write_one_cache_group(trans, fs_info,
3920 btrfs_abort_transaction(trans, ret);
3923 /* if its not on the io list, we need to put the block group */
3925 btrfs_put_block_group(cache);
3926 spin_lock(&cur_trans->dirty_bgs_lock);
3928 spin_unlock(&cur_trans->dirty_bgs_lock);
3930 while (!list_empty(io)) {
3931 cache = list_first_entry(io, struct btrfs_block_group_cache,
3933 list_del_init(&cache->io_list);
3934 btrfs_wait_cache_io(trans, cache, path);
3935 btrfs_put_block_group(cache);
3938 btrfs_free_path(path);
3942 int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
3944 struct btrfs_block_group_cache *block_group;
3947 block_group = btrfs_lookup_block_group(fs_info, bytenr);
3948 if (!block_group || block_group->ro)
3951 btrfs_put_block_group(block_group);
3955 bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3957 struct btrfs_block_group_cache *bg;
3960 bg = btrfs_lookup_block_group(fs_info, bytenr);
3964 spin_lock(&bg->lock);
3968 atomic_inc(&bg->nocow_writers);
3969 spin_unlock(&bg->lock);
3971 /* no put on block group, done by btrfs_dec_nocow_writers */
3973 btrfs_put_block_group(bg);
3979 void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3981 struct btrfs_block_group_cache *bg;
3983 bg = btrfs_lookup_block_group(fs_info, bytenr);
3985 if (atomic_dec_and_test(&bg->nocow_writers))
3986 wake_up_atomic_t(&bg->nocow_writers);
3988 * Once for our lookup and once for the lookup done by a previous call
3989 * to btrfs_inc_nocow_writers()
3991 btrfs_put_block_group(bg);
3992 btrfs_put_block_group(bg);
3995 static int btrfs_wait_nocow_writers_atomic_t(atomic_t *a)
4001 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg)
4003 wait_on_atomic_t(&bg->nocow_writers,
4004 btrfs_wait_nocow_writers_atomic_t,
4005 TASK_UNINTERRUPTIBLE);
4008 static const char *alloc_name(u64 flags)
4011 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
4013 case BTRFS_BLOCK_GROUP_METADATA:
4015 case BTRFS_BLOCK_GROUP_DATA:
4017 case BTRFS_BLOCK_GROUP_SYSTEM:
4021 return "invalid-combination";
4025 static int create_space_info(struct btrfs_fs_info *info, u64 flags,
4026 struct btrfs_space_info **new)
4029 struct btrfs_space_info *space_info;
4033 space_info = kzalloc(sizeof(*space_info), GFP_NOFS);
4037 ret = percpu_counter_init(&space_info->total_bytes_pinned, 0,
4044 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
4045 INIT_LIST_HEAD(&space_info->block_groups[i]);
4046 init_rwsem(&space_info->groups_sem);
4047 spin_lock_init(&space_info->lock);
4048 space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
4049 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4050 init_waitqueue_head(&space_info->wait);
4051 INIT_LIST_HEAD(&space_info->ro_bgs);
4052 INIT_LIST_HEAD(&space_info->tickets);
4053 INIT_LIST_HEAD(&space_info->priority_tickets);
4055 ret = kobject_init_and_add(&space_info->kobj, &space_info_ktype,
4056 info->space_info_kobj, "%s",
4057 alloc_name(space_info->flags));
4059 percpu_counter_destroy(&space_info->total_bytes_pinned);
4065 list_add_rcu(&space_info->list, &info->space_info);
4066 if (flags & BTRFS_BLOCK_GROUP_DATA)
4067 info->data_sinfo = space_info;
4072 static void update_space_info(struct btrfs_fs_info *info, u64 flags,
4073 u64 total_bytes, u64 bytes_used,
4075 struct btrfs_space_info **space_info)
4077 struct btrfs_space_info *found;
4080 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
4081 BTRFS_BLOCK_GROUP_RAID10))
4086 found = __find_space_info(info, flags);
4088 spin_lock(&found->lock);
4089 found->total_bytes += total_bytes;
4090 found->disk_total += total_bytes * factor;
4091 found->bytes_used += bytes_used;
4092 found->disk_used += bytes_used * factor;
4093 found->bytes_readonly += bytes_readonly;
4094 if (total_bytes > 0)
4096 space_info_add_new_bytes(info, found, total_bytes -
4097 bytes_used - bytes_readonly);
4098 spin_unlock(&found->lock);
4099 *space_info = found;
4102 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
4104 u64 extra_flags = chunk_to_extended(flags) &
4105 BTRFS_EXTENDED_PROFILE_MASK;
4107 write_seqlock(&fs_info->profiles_lock);
4108 if (flags & BTRFS_BLOCK_GROUP_DATA)
4109 fs_info->avail_data_alloc_bits |= extra_flags;
4110 if (flags & BTRFS_BLOCK_GROUP_METADATA)
4111 fs_info->avail_metadata_alloc_bits |= extra_flags;
4112 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4113 fs_info->avail_system_alloc_bits |= extra_flags;
4114 write_sequnlock(&fs_info->profiles_lock);
4118 * returns target flags in extended format or 0 if restripe for this
4119 * chunk_type is not in progress
4121 * should be called with either volume_mutex or balance_lock held
4123 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
4125 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4131 if (flags & BTRFS_BLOCK_GROUP_DATA &&
4132 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4133 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
4134 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
4135 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4136 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
4137 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
4138 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4139 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
4146 * @flags: available profiles in extended format (see ctree.h)
4148 * Returns reduced profile in chunk format. If profile changing is in
4149 * progress (either running or paused) picks the target profile (if it's
4150 * already available), otherwise falls back to plain reducing.
4152 static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags)
4154 u64 num_devices = fs_info->fs_devices->rw_devices;
4160 * see if restripe for this chunk_type is in progress, if so
4161 * try to reduce to the target profile
4163 spin_lock(&fs_info->balance_lock);
4164 target = get_restripe_target(fs_info, flags);
4166 /* pick target profile only if it's already available */
4167 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
4168 spin_unlock(&fs_info->balance_lock);
4169 return extended_to_chunk(target);
4172 spin_unlock(&fs_info->balance_lock);
4174 /* First, mask out the RAID levels which aren't possible */
4175 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4176 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
4177 allowed |= btrfs_raid_group[raid_type];
4181 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
4182 allowed = BTRFS_BLOCK_GROUP_RAID6;
4183 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
4184 allowed = BTRFS_BLOCK_GROUP_RAID5;
4185 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
4186 allowed = BTRFS_BLOCK_GROUP_RAID10;
4187 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
4188 allowed = BTRFS_BLOCK_GROUP_RAID1;
4189 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
4190 allowed = BTRFS_BLOCK_GROUP_RAID0;
4192 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
4194 return extended_to_chunk(flags | allowed);
4197 static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags)
4204 seq = read_seqbegin(&fs_info->profiles_lock);
4206 if (flags & BTRFS_BLOCK_GROUP_DATA)
4207 flags |= fs_info->avail_data_alloc_bits;
4208 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4209 flags |= fs_info->avail_system_alloc_bits;
4210 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
4211 flags |= fs_info->avail_metadata_alloc_bits;
4212 } while (read_seqretry(&fs_info->profiles_lock, seq));
4214 return btrfs_reduce_alloc_profile(fs_info, flags);
4217 static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
4219 struct btrfs_fs_info *fs_info = root->fs_info;
4224 flags = BTRFS_BLOCK_GROUP_DATA;
4225 else if (root == fs_info->chunk_root)
4226 flags = BTRFS_BLOCK_GROUP_SYSTEM;
4228 flags = BTRFS_BLOCK_GROUP_METADATA;
4230 ret = get_alloc_profile(fs_info, flags);
4234 u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info)
4236 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA);
4239 u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info)
4241 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4244 u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info)
4246 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4249 static u64 btrfs_space_info_used(struct btrfs_space_info *s_info,
4250 bool may_use_included)
4253 return s_info->bytes_used + s_info->bytes_reserved +
4254 s_info->bytes_pinned + s_info->bytes_readonly +
4255 (may_use_included ? s_info->bytes_may_use : 0);
4258 int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes)
4260 struct btrfs_root *root = inode->root;
4261 struct btrfs_fs_info *fs_info = root->fs_info;
4262 struct btrfs_space_info *data_sinfo = fs_info->data_sinfo;
4265 int need_commit = 2;
4266 int have_pinned_space;
4268 /* make sure bytes are sectorsize aligned */
4269 bytes = ALIGN(bytes, fs_info->sectorsize);
4271 if (btrfs_is_free_space_inode(inode)) {
4273 ASSERT(current->journal_info);
4277 /* make sure we have enough space to handle the data first */
4278 spin_lock(&data_sinfo->lock);
4279 used = btrfs_space_info_used(data_sinfo, true);
4281 if (used + bytes > data_sinfo->total_bytes) {
4282 struct btrfs_trans_handle *trans;
4285 * if we don't have enough free bytes in this space then we need
4286 * to alloc a new chunk.
4288 if (!data_sinfo->full) {
4291 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4292 spin_unlock(&data_sinfo->lock);
4294 alloc_target = btrfs_data_alloc_profile(fs_info);
4296 * It is ugly that we don't call nolock join
4297 * transaction for the free space inode case here.
4298 * But it is safe because we only do the data space
4299 * reservation for the free space cache in the
4300 * transaction context, the common join transaction
4301 * just increase the counter of the current transaction
4302 * handler, doesn't try to acquire the trans_lock of
4305 trans = btrfs_join_transaction(root);
4307 return PTR_ERR(trans);
4309 ret = do_chunk_alloc(trans, fs_info, alloc_target,
4310 CHUNK_ALLOC_NO_FORCE);
4311 btrfs_end_transaction(trans);
4316 have_pinned_space = 1;
4325 * If we don't have enough pinned space to deal with this
4326 * allocation, and no removed chunk in current transaction,
4327 * don't bother committing the transaction.
4329 have_pinned_space = percpu_counter_compare(
4330 &data_sinfo->total_bytes_pinned,
4331 used + bytes - data_sinfo->total_bytes);
4332 spin_unlock(&data_sinfo->lock);
4334 /* commit the current transaction and try again */
4337 !atomic_read(&fs_info->open_ioctl_trans)) {
4340 if (need_commit > 0) {
4341 btrfs_start_delalloc_roots(fs_info, 0, -1);
4342 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0,
4346 trans = btrfs_join_transaction(root);
4348 return PTR_ERR(trans);
4349 if (have_pinned_space >= 0 ||
4350 test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4351 &trans->transaction->flags) ||
4353 ret = btrfs_commit_transaction(trans);
4357 * The cleaner kthread might still be doing iput
4358 * operations. Wait for it to finish so that
4359 * more space is released.
4361 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
4362 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
4365 btrfs_end_transaction(trans);
4369 trace_btrfs_space_reservation(fs_info,
4370 "space_info:enospc",
4371 data_sinfo->flags, bytes, 1);
4374 data_sinfo->bytes_may_use += bytes;
4375 trace_btrfs_space_reservation(fs_info, "space_info",
4376 data_sinfo->flags, bytes, 1);
4377 spin_unlock(&data_sinfo->lock);
4382 int btrfs_check_data_free_space(struct inode *inode,
4383 struct extent_changeset **reserved, u64 start, u64 len)
4385 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4388 /* align the range */
4389 len = round_up(start + len, fs_info->sectorsize) -
4390 round_down(start, fs_info->sectorsize);
4391 start = round_down(start, fs_info->sectorsize);
4393 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len);
4397 /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */
4398 ret = btrfs_qgroup_reserve_data(inode, reserved, start, len);
4400 btrfs_free_reserved_data_space_noquota(inode, start, len);
4407 * Called if we need to clear a data reservation for this inode
4408 * Normally in a error case.
4410 * This one will *NOT* use accurate qgroup reserved space API, just for case
4411 * which we can't sleep and is sure it won't affect qgroup reserved space.
4412 * Like clear_bit_hook().
4414 void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4417 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4418 struct btrfs_space_info *data_sinfo;
4420 /* Make sure the range is aligned to sectorsize */
4421 len = round_up(start + len, fs_info->sectorsize) -
4422 round_down(start, fs_info->sectorsize);
4423 start = round_down(start, fs_info->sectorsize);
4425 data_sinfo = fs_info->data_sinfo;
4426 spin_lock(&data_sinfo->lock);
4427 if (WARN_ON(data_sinfo->bytes_may_use < len))
4428 data_sinfo->bytes_may_use = 0;
4430 data_sinfo->bytes_may_use -= len;
4431 trace_btrfs_space_reservation(fs_info, "space_info",
4432 data_sinfo->flags, len, 0);
4433 spin_unlock(&data_sinfo->lock);
4437 * Called if we need to clear a data reservation for this inode
4438 * Normally in a error case.
4440 * This one will handle the per-inode data rsv map for accurate reserved
4443 void btrfs_free_reserved_data_space(struct inode *inode,
4444 struct extent_changeset *reserved, u64 start, u64 len)
4446 struct btrfs_root *root = BTRFS_I(inode)->root;
4448 /* Make sure the range is aligned to sectorsize */
4449 len = round_up(start + len, root->fs_info->sectorsize) -
4450 round_down(start, root->fs_info->sectorsize);
4451 start = round_down(start, root->fs_info->sectorsize);
4453 btrfs_free_reserved_data_space_noquota(inode, start, len);
4454 btrfs_qgroup_free_data(inode, reserved, start, len);
4457 static void force_metadata_allocation(struct btrfs_fs_info *info)
4459 struct list_head *head = &info->space_info;
4460 struct btrfs_space_info *found;
4463 list_for_each_entry_rcu(found, head, list) {
4464 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4465 found->force_alloc = CHUNK_ALLOC_FORCE;
4470 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4472 return (global->size << 1);
4475 static int should_alloc_chunk(struct btrfs_fs_info *fs_info,
4476 struct btrfs_space_info *sinfo, int force)
4478 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4479 u64 bytes_used = btrfs_space_info_used(sinfo, false);
4482 if (force == CHUNK_ALLOC_FORCE)
4486 * We need to take into account the global rsv because for all intents
4487 * and purposes it's used space. Don't worry about locking the
4488 * global_rsv, it doesn't change except when the transaction commits.
4490 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4491 bytes_used += calc_global_rsv_need_space(global_rsv);
4494 * in limited mode, we want to have some free space up to
4495 * about 1% of the FS size.
4497 if (force == CHUNK_ALLOC_LIMITED) {
4498 thresh = btrfs_super_total_bytes(fs_info->super_copy);
4499 thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1));
4501 if (sinfo->total_bytes - bytes_used < thresh)
4505 if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8))
4510 static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type)
4514 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4515 BTRFS_BLOCK_GROUP_RAID0 |
4516 BTRFS_BLOCK_GROUP_RAID5 |
4517 BTRFS_BLOCK_GROUP_RAID6))
4518 num_dev = fs_info->fs_devices->rw_devices;
4519 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4522 num_dev = 1; /* DUP or single */
4528 * If @is_allocation is true, reserve space in the system space info necessary
4529 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4532 void check_system_chunk(struct btrfs_trans_handle *trans,
4533 struct btrfs_fs_info *fs_info, u64 type)
4535 struct btrfs_space_info *info;
4542 * Needed because we can end up allocating a system chunk and for an
4543 * atomic and race free space reservation in the chunk block reserve.
4545 ASSERT(mutex_is_locked(&fs_info->chunk_mutex));
4547 info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4548 spin_lock(&info->lock);
4549 left = info->total_bytes - btrfs_space_info_used(info, true);
4550 spin_unlock(&info->lock);
4552 num_devs = get_profile_num_devs(fs_info, type);
4554 /* num_devs device items to update and 1 chunk item to add or remove */
4555 thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) +
4556 btrfs_calc_trans_metadata_size(fs_info, 1);
4558 if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4559 btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu",
4560 left, thresh, type);
4561 dump_space_info(fs_info, info, 0, 0);
4564 if (left < thresh) {
4565 u64 flags = btrfs_system_alloc_profile(fs_info);
4568 * Ignore failure to create system chunk. We might end up not
4569 * needing it, as we might not need to COW all nodes/leafs from
4570 * the paths we visit in the chunk tree (they were already COWed
4571 * or created in the current transaction for example).
4573 ret = btrfs_alloc_chunk(trans, fs_info, flags);
4577 ret = btrfs_block_rsv_add(fs_info->chunk_root,
4578 &fs_info->chunk_block_rsv,
4579 thresh, BTRFS_RESERVE_NO_FLUSH);
4581 trans->chunk_bytes_reserved += thresh;
4586 * If force is CHUNK_ALLOC_FORCE:
4587 * - return 1 if it successfully allocates a chunk,
4588 * - return errors including -ENOSPC otherwise.
4589 * If force is NOT CHUNK_ALLOC_FORCE:
4590 * - return 0 if it doesn't need to allocate a new chunk,
4591 * - return 1 if it successfully allocates a chunk,
4592 * - return errors including -ENOSPC otherwise.
4594 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4595 struct btrfs_fs_info *fs_info, u64 flags, int force)
4597 struct btrfs_space_info *space_info;
4598 int wait_for_alloc = 0;
4601 /* Don't re-enter if we're already allocating a chunk */
4602 if (trans->allocating_chunk)
4605 space_info = __find_space_info(fs_info, flags);
4607 ret = create_space_info(fs_info, flags, &space_info);
4613 spin_lock(&space_info->lock);
4614 if (force < space_info->force_alloc)
4615 force = space_info->force_alloc;
4616 if (space_info->full) {
4617 if (should_alloc_chunk(fs_info, space_info, force))
4621 spin_unlock(&space_info->lock);
4625 if (!should_alloc_chunk(fs_info, space_info, force)) {
4626 spin_unlock(&space_info->lock);
4628 } else if (space_info->chunk_alloc) {
4631 space_info->chunk_alloc = 1;
4634 spin_unlock(&space_info->lock);
4636 mutex_lock(&fs_info->chunk_mutex);
4639 * The chunk_mutex is held throughout the entirety of a chunk
4640 * allocation, so once we've acquired the chunk_mutex we know that the
4641 * other guy is done and we need to recheck and see if we should
4644 if (wait_for_alloc) {
4645 mutex_unlock(&fs_info->chunk_mutex);
4650 trans->allocating_chunk = true;
4653 * If we have mixed data/metadata chunks we want to make sure we keep
4654 * allocating mixed chunks instead of individual chunks.
4656 if (btrfs_mixed_space_info(space_info))
4657 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4660 * if we're doing a data chunk, go ahead and make sure that
4661 * we keep a reasonable number of metadata chunks allocated in the
4664 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4665 fs_info->data_chunk_allocations++;
4666 if (!(fs_info->data_chunk_allocations %
4667 fs_info->metadata_ratio))
4668 force_metadata_allocation(fs_info);
4672 * Check if we have enough space in SYSTEM chunk because we may need
4673 * to update devices.
4675 check_system_chunk(trans, fs_info, flags);
4677 ret = btrfs_alloc_chunk(trans, fs_info, flags);
4678 trans->allocating_chunk = false;
4680 spin_lock(&space_info->lock);
4681 if (ret < 0 && ret != -ENOSPC)
4684 space_info->full = 1;
4688 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4690 space_info->chunk_alloc = 0;
4691 spin_unlock(&space_info->lock);
4692 mutex_unlock(&fs_info->chunk_mutex);
4694 * When we allocate a new chunk we reserve space in the chunk block
4695 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4696 * add new nodes/leafs to it if we end up needing to do it when
4697 * inserting the chunk item and updating device items as part of the
4698 * second phase of chunk allocation, performed by
4699 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4700 * large number of new block groups to create in our transaction
4701 * handle's new_bgs list to avoid exhausting the chunk block reserve
4702 * in extreme cases - like having a single transaction create many new
4703 * block groups when starting to write out the free space caches of all
4704 * the block groups that were made dirty during the lifetime of the
4707 if (trans->can_flush_pending_bgs &&
4708 trans->chunk_bytes_reserved >= (u64)SZ_2M) {
4709 btrfs_create_pending_block_groups(trans, fs_info);
4710 btrfs_trans_release_chunk_metadata(trans);
4715 static int can_overcommit(struct btrfs_fs_info *fs_info,
4716 struct btrfs_space_info *space_info, u64 bytes,
4717 enum btrfs_reserve_flush_enum flush,
4720 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4726 /* Don't overcommit when in mixed mode. */
4727 if (space_info->flags & BTRFS_BLOCK_GROUP_DATA)
4731 profile = btrfs_system_alloc_profile(fs_info);
4733 profile = btrfs_metadata_alloc_profile(fs_info);
4735 used = btrfs_space_info_used(space_info, false);
4738 * We only want to allow over committing if we have lots of actual space
4739 * free, but if we don't have enough space to handle the global reserve
4740 * space then we could end up having a real enospc problem when trying
4741 * to allocate a chunk or some other such important allocation.
4743 spin_lock(&global_rsv->lock);
4744 space_size = calc_global_rsv_need_space(global_rsv);
4745 spin_unlock(&global_rsv->lock);
4746 if (used + space_size >= space_info->total_bytes)
4749 used += space_info->bytes_may_use;
4751 avail = atomic64_read(&fs_info->free_chunk_space);
4754 * If we have dup, raid1 or raid10 then only half of the free
4755 * space is actually useable. For raid56, the space info used
4756 * doesn't include the parity drive, so we don't have to
4759 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4760 BTRFS_BLOCK_GROUP_RAID1 |
4761 BTRFS_BLOCK_GROUP_RAID10))
4765 * If we aren't flushing all things, let us overcommit up to
4766 * 1/2th of the space. If we can flush, don't let us overcommit
4767 * too much, let it overcommit up to 1/8 of the space.
4769 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4774 if (used + bytes < space_info->total_bytes + avail)
4779 static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info,
4780 unsigned long nr_pages, int nr_items)
4782 struct super_block *sb = fs_info->sb;
4784 if (down_read_trylock(&sb->s_umount)) {
4785 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4786 up_read(&sb->s_umount);
4789 * We needn't worry the filesystem going from r/w to r/o though
4790 * we don't acquire ->s_umount mutex, because the filesystem
4791 * should guarantee the delalloc inodes list be empty after
4792 * the filesystem is readonly(all dirty pages are written to
4795 btrfs_start_delalloc_roots(fs_info, 0, nr_items);
4796 if (!current->journal_info)
4797 btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1);
4801 static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info,
4807 bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
4808 nr = div64_u64(to_reclaim, bytes);
4814 #define EXTENT_SIZE_PER_ITEM SZ_256K
4817 * shrink metadata reservation for delalloc
4819 static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim,
4820 u64 orig, bool wait_ordered)
4822 struct btrfs_space_info *space_info;
4823 struct btrfs_trans_handle *trans;
4828 unsigned long nr_pages;
4830 enum btrfs_reserve_flush_enum flush;
4832 /* Calc the number of the pages we need flush for space reservation */
4833 items = calc_reclaim_items_nr(fs_info, to_reclaim);
4834 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4836 trans = (struct btrfs_trans_handle *)current->journal_info;
4837 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4839 delalloc_bytes = percpu_counter_sum_positive(
4840 &fs_info->delalloc_bytes);
4841 if (delalloc_bytes == 0) {
4845 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4850 while (delalloc_bytes && loops < 3) {
4851 max_reclaim = min(delalloc_bytes, to_reclaim);
4852 nr_pages = max_reclaim >> PAGE_SHIFT;
4853 btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items);
4855 * We need to wait for the async pages to actually start before
4858 max_reclaim = atomic_read(&fs_info->async_delalloc_pages);
4862 if (max_reclaim <= nr_pages)
4865 max_reclaim -= nr_pages;
4867 wait_event(fs_info->async_submit_wait,
4868 atomic_read(&fs_info->async_delalloc_pages) <=
4872 flush = BTRFS_RESERVE_FLUSH_ALL;
4874 flush = BTRFS_RESERVE_NO_FLUSH;
4875 spin_lock(&space_info->lock);
4876 if (list_empty(&space_info->tickets) &&
4877 list_empty(&space_info->priority_tickets)) {
4878 spin_unlock(&space_info->lock);
4881 spin_unlock(&space_info->lock);
4884 if (wait_ordered && !trans) {
4885 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4887 time_left = schedule_timeout_killable(1);
4891 delalloc_bytes = percpu_counter_sum_positive(
4892 &fs_info->delalloc_bytes);
4896 struct reserve_ticket {
4899 struct list_head list;
4900 wait_queue_head_t wait;
4904 * maybe_commit_transaction - possibly commit the transaction if its ok to
4905 * @root - the root we're allocating for
4906 * @bytes - the number of bytes we want to reserve
4907 * @force - force the commit
4909 * This will check to make sure that committing the transaction will actually
4910 * get us somewhere and then commit the transaction if it does. Otherwise it
4911 * will return -ENOSPC.
4913 static int may_commit_transaction(struct btrfs_fs_info *fs_info,
4914 struct btrfs_space_info *space_info)
4916 struct reserve_ticket *ticket = NULL;
4917 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv;
4918 struct btrfs_trans_handle *trans;
4921 trans = (struct btrfs_trans_handle *)current->journal_info;
4925 spin_lock(&space_info->lock);
4926 if (!list_empty(&space_info->priority_tickets))
4927 ticket = list_first_entry(&space_info->priority_tickets,
4928 struct reserve_ticket, list);
4929 else if (!list_empty(&space_info->tickets))
4930 ticket = list_first_entry(&space_info->tickets,
4931 struct reserve_ticket, list);
4932 bytes = (ticket) ? ticket->bytes : 0;
4933 spin_unlock(&space_info->lock);
4938 /* See if there is enough pinned space to make this reservation */
4939 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4944 * See if there is some space in the delayed insertion reservation for
4947 if (space_info != delayed_rsv->space_info)
4950 spin_lock(&delayed_rsv->lock);
4951 if (delayed_rsv->size > bytes)
4954 bytes -= delayed_rsv->size;
4955 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4957 spin_unlock(&delayed_rsv->lock);
4960 spin_unlock(&delayed_rsv->lock);
4963 trans = btrfs_join_transaction(fs_info->extent_root);
4967 return btrfs_commit_transaction(trans);
4971 * Try to flush some data based on policy set by @state. This is only advisory
4972 * and may fail for various reasons. The caller is supposed to examine the
4973 * state of @space_info to detect the outcome.
4975 static void flush_space(struct btrfs_fs_info *fs_info,
4976 struct btrfs_space_info *space_info, u64 num_bytes,
4979 struct btrfs_root *root = fs_info->extent_root;
4980 struct btrfs_trans_handle *trans;
4985 case FLUSH_DELAYED_ITEMS_NR:
4986 case FLUSH_DELAYED_ITEMS:
4987 if (state == FLUSH_DELAYED_ITEMS_NR)
4988 nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2;
4992 trans = btrfs_join_transaction(root);
4993 if (IS_ERR(trans)) {
4994 ret = PTR_ERR(trans);
4997 ret = btrfs_run_delayed_items_nr(trans, fs_info, nr);
4998 btrfs_end_transaction(trans);
5000 case FLUSH_DELALLOC:
5001 case FLUSH_DELALLOC_WAIT:
5002 shrink_delalloc(fs_info, num_bytes * 2, num_bytes,
5003 state == FLUSH_DELALLOC_WAIT);
5006 trans = btrfs_join_transaction(root);
5007 if (IS_ERR(trans)) {
5008 ret = PTR_ERR(trans);
5011 ret = do_chunk_alloc(trans, fs_info,
5012 btrfs_metadata_alloc_profile(fs_info),
5013 CHUNK_ALLOC_NO_FORCE);
5014 btrfs_end_transaction(trans);
5015 if (ret > 0 || ret == -ENOSPC)
5019 ret = may_commit_transaction(fs_info, space_info);
5026 trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, state,
5032 btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info,
5033 struct btrfs_space_info *space_info,
5036 struct reserve_ticket *ticket;
5041 list_for_each_entry(ticket, &space_info->tickets, list)
5042 to_reclaim += ticket->bytes;
5043 list_for_each_entry(ticket, &space_info->priority_tickets, list)
5044 to_reclaim += ticket->bytes;
5048 to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M);
5049 if (can_overcommit(fs_info, space_info, to_reclaim,
5050 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
5053 used = btrfs_space_info_used(space_info, true);
5055 if (can_overcommit(fs_info, space_info, SZ_1M,
5056 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
5057 expected = div_factor_fine(space_info->total_bytes, 95);
5059 expected = div_factor_fine(space_info->total_bytes, 90);
5061 if (used > expected)
5062 to_reclaim = used - expected;
5065 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
5066 space_info->bytes_reserved);
5070 static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info,
5071 struct btrfs_space_info *space_info,
5072 u64 used, bool system_chunk)
5074 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
5076 /* If we're just plain full then async reclaim just slows us down. */
5077 if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh)
5080 if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5084 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
5085 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
5088 static void wake_all_tickets(struct list_head *head)
5090 struct reserve_ticket *ticket;
5092 while (!list_empty(head)) {
5093 ticket = list_first_entry(head, struct reserve_ticket, list);
5094 list_del_init(&ticket->list);
5095 ticket->error = -ENOSPC;
5096 wake_up(&ticket->wait);
5101 * This is for normal flushers, we can wait all goddamned day if we want to. We
5102 * will loop and continuously try to flush as long as we are making progress.
5103 * We count progress as clearing off tickets each time we have to loop.
5105 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
5107 struct btrfs_fs_info *fs_info;
5108 struct btrfs_space_info *space_info;
5111 int commit_cycles = 0;
5112 u64 last_tickets_id;
5114 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
5115 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5117 spin_lock(&space_info->lock);
5118 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5121 space_info->flush = 0;
5122 spin_unlock(&space_info->lock);
5125 last_tickets_id = space_info->tickets_id;
5126 spin_unlock(&space_info->lock);
5128 flush_state = FLUSH_DELAYED_ITEMS_NR;
5130 flush_space(fs_info, space_info, to_reclaim, flush_state);
5131 spin_lock(&space_info->lock);
5132 if (list_empty(&space_info->tickets)) {
5133 space_info->flush = 0;
5134 spin_unlock(&space_info->lock);
5137 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info,
5140 if (last_tickets_id == space_info->tickets_id) {
5143 last_tickets_id = space_info->tickets_id;
5144 flush_state = FLUSH_DELAYED_ITEMS_NR;
5149 if (flush_state > COMMIT_TRANS) {
5151 if (commit_cycles > 2) {
5152 wake_all_tickets(&space_info->tickets);
5153 space_info->flush = 0;
5155 flush_state = FLUSH_DELAYED_ITEMS_NR;
5158 spin_unlock(&space_info->lock);
5159 } while (flush_state <= COMMIT_TRANS);
5162 void btrfs_init_async_reclaim_work(struct work_struct *work)
5164 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
5167 static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
5168 struct btrfs_space_info *space_info,
5169 struct reserve_ticket *ticket)
5172 int flush_state = FLUSH_DELAYED_ITEMS_NR;
5174 spin_lock(&space_info->lock);
5175 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5178 spin_unlock(&space_info->lock);
5181 spin_unlock(&space_info->lock);
5184 flush_space(fs_info, space_info, to_reclaim, flush_state);
5186 spin_lock(&space_info->lock);
5187 if (ticket->bytes == 0) {
5188 spin_unlock(&space_info->lock);
5191 spin_unlock(&space_info->lock);
5194 * Priority flushers can't wait on delalloc without
5197 if (flush_state == FLUSH_DELALLOC ||
5198 flush_state == FLUSH_DELALLOC_WAIT)
5199 flush_state = ALLOC_CHUNK;
5200 } while (flush_state < COMMIT_TRANS);
5203 static int wait_reserve_ticket(struct btrfs_fs_info *fs_info,
5204 struct btrfs_space_info *space_info,
5205 struct reserve_ticket *ticket, u64 orig_bytes)
5211 spin_lock(&space_info->lock);
5212 while (ticket->bytes > 0 && ticket->error == 0) {
5213 ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE);
5218 spin_unlock(&space_info->lock);
5222 finish_wait(&ticket->wait, &wait);
5223 spin_lock(&space_info->lock);
5226 ret = ticket->error;
5227 if (!list_empty(&ticket->list))
5228 list_del_init(&ticket->list);
5229 if (ticket->bytes && ticket->bytes < orig_bytes) {
5230 u64 num_bytes = orig_bytes - ticket->bytes;
5231 space_info->bytes_may_use -= num_bytes;
5232 trace_btrfs_space_reservation(fs_info, "space_info",
5233 space_info->flags, num_bytes, 0);
5235 spin_unlock(&space_info->lock);
5241 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5242 * @root - the root we're allocating for
5243 * @space_info - the space info we want to allocate from
5244 * @orig_bytes - the number of bytes we want
5245 * @flush - whether or not we can flush to make our reservation
5247 * This will reserve orig_bytes number of bytes from the space info associated
5248 * with the block_rsv. If there is not enough space it will make an attempt to
5249 * flush out space to make room. It will do this by flushing delalloc if
5250 * possible or committing the transaction. If flush is 0 then no attempts to
5251 * regain reservations will be made and this will fail if there is not enough
5254 static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info,
5255 struct btrfs_space_info *space_info,
5257 enum btrfs_reserve_flush_enum flush,
5260 struct reserve_ticket ticket;
5265 ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL);
5267 spin_lock(&space_info->lock);
5269 used = btrfs_space_info_used(space_info, true);
5272 * If we have enough space then hooray, make our reservation and carry
5273 * on. If not see if we can overcommit, and if we can, hooray carry on.
5274 * If not things get more complicated.
5276 if (used + orig_bytes <= space_info->total_bytes) {
5277 space_info->bytes_may_use += orig_bytes;
5278 trace_btrfs_space_reservation(fs_info, "space_info",
5279 space_info->flags, orig_bytes, 1);
5281 } else if (can_overcommit(fs_info, space_info, orig_bytes, flush,
5283 space_info->bytes_may_use += orig_bytes;
5284 trace_btrfs_space_reservation(fs_info, "space_info",
5285 space_info->flags, orig_bytes, 1);
5290 * If we couldn't make a reservation then setup our reservation ticket
5291 * and kick the async worker if it's not already running.
5293 * If we are a priority flusher then we just need to add our ticket to
5294 * the list and we will do our own flushing further down.
5296 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
5297 ticket.bytes = orig_bytes;
5299 init_waitqueue_head(&ticket.wait);
5300 if (flush == BTRFS_RESERVE_FLUSH_ALL) {
5301 list_add_tail(&ticket.list, &space_info->tickets);
5302 if (!space_info->flush) {
5303 space_info->flush = 1;
5304 trace_btrfs_trigger_flush(fs_info,
5308 queue_work(system_unbound_wq,
5309 &fs_info->async_reclaim_work);
5312 list_add_tail(&ticket.list,
5313 &space_info->priority_tickets);
5315 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
5318 * We will do the space reservation dance during log replay,
5319 * which means we won't have fs_info->fs_root set, so don't do
5320 * the async reclaim as we will panic.
5322 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) &&
5323 need_do_async_reclaim(fs_info, space_info,
5324 used, system_chunk) &&
5325 !work_busy(&fs_info->async_reclaim_work)) {
5326 trace_btrfs_trigger_flush(fs_info, space_info->flags,
5327 orig_bytes, flush, "preempt");
5328 queue_work(system_unbound_wq,
5329 &fs_info->async_reclaim_work);
5332 spin_unlock(&space_info->lock);
5333 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
5336 if (flush == BTRFS_RESERVE_FLUSH_ALL)
5337 return wait_reserve_ticket(fs_info, space_info, &ticket,
5341 priority_reclaim_metadata_space(fs_info, space_info, &ticket);
5342 spin_lock(&space_info->lock);
5344 if (ticket.bytes < orig_bytes) {
5345 u64 num_bytes = orig_bytes - ticket.bytes;
5346 space_info->bytes_may_use -= num_bytes;
5347 trace_btrfs_space_reservation(fs_info, "space_info",
5352 list_del_init(&ticket.list);
5355 spin_unlock(&space_info->lock);
5356 ASSERT(list_empty(&ticket.list));
5361 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5362 * @root - the root we're allocating for
5363 * @block_rsv - the block_rsv we're allocating for
5364 * @orig_bytes - the number of bytes we want
5365 * @flush - whether or not we can flush to make our reservation
5367 * This will reserve orgi_bytes number of bytes from the space info associated
5368 * with the block_rsv. If there is not enough space it will make an attempt to
5369 * flush out space to make room. It will do this by flushing delalloc if
5370 * possible or committing the transaction. If flush is 0 then no attempts to
5371 * regain reservations will be made and this will fail if there is not enough
5374 static int reserve_metadata_bytes(struct btrfs_root *root,
5375 struct btrfs_block_rsv *block_rsv,
5377 enum btrfs_reserve_flush_enum flush)
5379 struct btrfs_fs_info *fs_info = root->fs_info;
5380 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5382 bool system_chunk = (root == fs_info->chunk_root);
5384 ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info,
5385 orig_bytes, flush, system_chunk);
5386 if (ret == -ENOSPC &&
5387 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5388 if (block_rsv != global_rsv &&
5389 !block_rsv_use_bytes(global_rsv, orig_bytes))
5393 trace_btrfs_space_reservation(fs_info, "space_info:enospc",
5394 block_rsv->space_info->flags,
5399 static struct btrfs_block_rsv *get_block_rsv(
5400 const struct btrfs_trans_handle *trans,
5401 const struct btrfs_root *root)
5403 struct btrfs_fs_info *fs_info = root->fs_info;
5404 struct btrfs_block_rsv *block_rsv = NULL;
5406 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5407 (root == fs_info->csum_root && trans->adding_csums) ||
5408 (root == fs_info->uuid_root))
5409 block_rsv = trans->block_rsv;
5412 block_rsv = root->block_rsv;
5415 block_rsv = &fs_info->empty_block_rsv;
5420 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5424 spin_lock(&block_rsv->lock);
5425 if (block_rsv->reserved >= num_bytes) {
5426 block_rsv->reserved -= num_bytes;
5427 if (block_rsv->reserved < block_rsv->size)
5428 block_rsv->full = 0;
5431 spin_unlock(&block_rsv->lock);
5435 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5436 u64 num_bytes, int update_size)
5438 spin_lock(&block_rsv->lock);
5439 block_rsv->reserved += num_bytes;
5441 block_rsv->size += num_bytes;
5442 else if (block_rsv->reserved >= block_rsv->size)
5443 block_rsv->full = 1;
5444 spin_unlock(&block_rsv->lock);
5447 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5448 struct btrfs_block_rsv *dest, u64 num_bytes,
5451 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5454 if (global_rsv->space_info != dest->space_info)
5457 spin_lock(&global_rsv->lock);
5458 min_bytes = div_factor(global_rsv->size, min_factor);
5459 if (global_rsv->reserved < min_bytes + num_bytes) {
5460 spin_unlock(&global_rsv->lock);
5463 global_rsv->reserved -= num_bytes;
5464 if (global_rsv->reserved < global_rsv->size)
5465 global_rsv->full = 0;
5466 spin_unlock(&global_rsv->lock);
5468 block_rsv_add_bytes(dest, num_bytes, 1);
5473 * This is for space we already have accounted in space_info->bytes_may_use, so
5474 * basically when we're returning space from block_rsv's.
5476 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
5477 struct btrfs_space_info *space_info,
5480 struct reserve_ticket *ticket;
5481 struct list_head *head;
5483 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH;
5484 bool check_overcommit = false;
5486 spin_lock(&space_info->lock);
5487 head = &space_info->priority_tickets;
5490 * If we are over our limit then we need to check and see if we can
5491 * overcommit, and if we can't then we just need to free up our space
5492 * and not satisfy any requests.
5494 used = btrfs_space_info_used(space_info, true);
5495 if (used - num_bytes >= space_info->total_bytes)
5496 check_overcommit = true;
5498 while (!list_empty(head) && num_bytes) {
5499 ticket = list_first_entry(head, struct reserve_ticket,
5502 * We use 0 bytes because this space is already reserved, so
5503 * adding the ticket space would be a double count.
5505 if (check_overcommit &&
5506 !can_overcommit(fs_info, space_info, 0, flush, false))
5508 if (num_bytes >= ticket->bytes) {
5509 list_del_init(&ticket->list);
5510 num_bytes -= ticket->bytes;
5512 space_info->tickets_id++;
5513 wake_up(&ticket->wait);
5515 ticket->bytes -= num_bytes;
5520 if (num_bytes && head == &space_info->priority_tickets) {
5521 head = &space_info->tickets;
5522 flush = BTRFS_RESERVE_FLUSH_ALL;
5525 space_info->bytes_may_use -= num_bytes;
5526 trace_btrfs_space_reservation(fs_info, "space_info",
5527 space_info->flags, num_bytes, 0);
5528 spin_unlock(&space_info->lock);
5532 * This is for newly allocated space that isn't accounted in
5533 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5534 * we use this helper.
5536 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
5537 struct btrfs_space_info *space_info,
5540 struct reserve_ticket *ticket;
5541 struct list_head *head = &space_info->priority_tickets;
5544 while (!list_empty(head) && num_bytes) {
5545 ticket = list_first_entry(head, struct reserve_ticket,
5547 if (num_bytes >= ticket->bytes) {
5548 trace_btrfs_space_reservation(fs_info, "space_info",
5551 list_del_init(&ticket->list);
5552 num_bytes -= ticket->bytes;
5553 space_info->bytes_may_use += ticket->bytes;
5555 space_info->tickets_id++;
5556 wake_up(&ticket->wait);
5558 trace_btrfs_space_reservation(fs_info, "space_info",
5561 space_info->bytes_may_use += num_bytes;
5562 ticket->bytes -= num_bytes;
5567 if (num_bytes && head == &space_info->priority_tickets) {
5568 head = &space_info->tickets;
5573 static u64 block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5574 struct btrfs_block_rsv *block_rsv,
5575 struct btrfs_block_rsv *dest, u64 num_bytes)
5577 struct btrfs_space_info *space_info = block_rsv->space_info;
5580 spin_lock(&block_rsv->lock);
5581 if (num_bytes == (u64)-1)
5582 num_bytes = block_rsv->size;
5583 block_rsv->size -= num_bytes;
5584 if (block_rsv->reserved >= block_rsv->size) {
5585 num_bytes = block_rsv->reserved - block_rsv->size;
5586 block_rsv->reserved = block_rsv->size;
5587 block_rsv->full = 1;
5591 spin_unlock(&block_rsv->lock);
5594 if (num_bytes > 0) {
5596 spin_lock(&dest->lock);
5600 bytes_to_add = dest->size - dest->reserved;
5601 bytes_to_add = min(num_bytes, bytes_to_add);
5602 dest->reserved += bytes_to_add;
5603 if (dest->reserved >= dest->size)
5605 num_bytes -= bytes_to_add;
5607 spin_unlock(&dest->lock);
5610 space_info_add_old_bytes(fs_info, space_info,
5616 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
5617 struct btrfs_block_rsv *dst, u64 num_bytes,
5622 ret = block_rsv_use_bytes(src, num_bytes);
5626 block_rsv_add_bytes(dst, num_bytes, update_size);
5630 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5632 memset(rsv, 0, sizeof(*rsv));
5633 spin_lock_init(&rsv->lock);
5637 void btrfs_init_metadata_block_rsv(struct btrfs_fs_info *fs_info,
5638 struct btrfs_block_rsv *rsv,
5639 unsigned short type)
5641 btrfs_init_block_rsv(rsv, type);
5642 rsv->space_info = __find_space_info(fs_info,
5643 BTRFS_BLOCK_GROUP_METADATA);
5646 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info,
5647 unsigned short type)
5649 struct btrfs_block_rsv *block_rsv;
5651 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5655 btrfs_init_metadata_block_rsv(fs_info, block_rsv, type);
5659 void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info,
5660 struct btrfs_block_rsv *rsv)
5664 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
5668 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
5673 int btrfs_block_rsv_add(struct btrfs_root *root,
5674 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5675 enum btrfs_reserve_flush_enum flush)
5682 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5684 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5691 int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor)
5699 spin_lock(&block_rsv->lock);
5700 num_bytes = div_factor(block_rsv->size, min_factor);
5701 if (block_rsv->reserved >= num_bytes)
5703 spin_unlock(&block_rsv->lock);
5708 int btrfs_block_rsv_refill(struct btrfs_root *root,
5709 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5710 enum btrfs_reserve_flush_enum flush)
5718 spin_lock(&block_rsv->lock);
5719 num_bytes = min_reserved;
5720 if (block_rsv->reserved >= num_bytes)
5723 num_bytes -= block_rsv->reserved;
5724 spin_unlock(&block_rsv->lock);
5729 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5731 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5739 * btrfs_inode_rsv_refill - refill the inode block rsv.
5740 * @inode - the inode we are refilling.
5741 * @flush - the flusing restriction.
5743 * Essentially the same as btrfs_block_rsv_refill, except it uses the
5744 * block_rsv->size as the minimum size. We'll either refill the missing amount
5745 * or return if we already have enough space. This will also handle the resreve
5746 * tracepoint for the reserved amount.
5748 int btrfs_inode_rsv_refill(struct btrfs_inode *inode,
5749 enum btrfs_reserve_flush_enum flush)
5751 struct btrfs_root *root = inode->root;
5752 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5756 spin_lock(&block_rsv->lock);
5757 if (block_rsv->reserved < block_rsv->size)
5758 num_bytes = block_rsv->size - block_rsv->reserved;
5759 spin_unlock(&block_rsv->lock);
5764 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5766 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5767 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5768 btrfs_ino(inode), num_bytes, 1);
5774 * btrfs_inode_rsv_release - release any excessive reservation.
5775 * @inode - the inode we need to release from.
5777 * This is the same as btrfs_block_rsv_release, except that it handles the
5778 * tracepoint for the reservation.
5780 void btrfs_inode_rsv_release(struct btrfs_inode *inode)
5782 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5783 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5784 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5788 * Since we statically set the block_rsv->size we just want to say we
5789 * are releasing 0 bytes, and then we'll just get the reservation over
5792 released = block_rsv_release_bytes(fs_info, block_rsv, global_rsv, 0);
5794 trace_btrfs_space_reservation(fs_info, "delalloc",
5795 btrfs_ino(inode), released, 0);
5798 void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
5799 struct btrfs_block_rsv *block_rsv,
5802 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5804 if (global_rsv == block_rsv ||
5805 block_rsv->space_info != global_rsv->space_info)
5807 block_rsv_release_bytes(fs_info, block_rsv, global_rsv, num_bytes);
5810 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5812 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5813 struct btrfs_space_info *sinfo = block_rsv->space_info;
5817 * The global block rsv is based on the size of the extent tree, the
5818 * checksum tree and the root tree. If the fs is empty we want to set
5819 * it to a minimal amount for safety.
5821 num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) +
5822 btrfs_root_used(&fs_info->csum_root->root_item) +
5823 btrfs_root_used(&fs_info->tree_root->root_item);
5824 num_bytes = max_t(u64, num_bytes, SZ_16M);
5826 spin_lock(&sinfo->lock);
5827 spin_lock(&block_rsv->lock);
5829 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
5831 if (block_rsv->reserved < block_rsv->size) {
5832 num_bytes = btrfs_space_info_used(sinfo, true);
5833 if (sinfo->total_bytes > num_bytes) {
5834 num_bytes = sinfo->total_bytes - num_bytes;
5835 num_bytes = min(num_bytes,
5836 block_rsv->size - block_rsv->reserved);
5837 block_rsv->reserved += num_bytes;
5838 sinfo->bytes_may_use += num_bytes;
5839 trace_btrfs_space_reservation(fs_info, "space_info",
5840 sinfo->flags, num_bytes,
5843 } else if (block_rsv->reserved > block_rsv->size) {
5844 num_bytes = block_rsv->reserved - block_rsv->size;
5845 sinfo->bytes_may_use -= num_bytes;
5846 trace_btrfs_space_reservation(fs_info, "space_info",
5847 sinfo->flags, num_bytes, 0);
5848 block_rsv->reserved = block_rsv->size;
5851 if (block_rsv->reserved == block_rsv->size)
5852 block_rsv->full = 1;
5854 block_rsv->full = 0;
5856 spin_unlock(&block_rsv->lock);
5857 spin_unlock(&sinfo->lock);
5860 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5862 struct btrfs_space_info *space_info;
5864 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5865 fs_info->chunk_block_rsv.space_info = space_info;
5867 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5868 fs_info->global_block_rsv.space_info = space_info;
5869 fs_info->trans_block_rsv.space_info = space_info;
5870 fs_info->empty_block_rsv.space_info = space_info;
5871 fs_info->delayed_block_rsv.space_info = space_info;
5873 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5874 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5875 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5876 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5877 if (fs_info->quota_root)
5878 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5879 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5881 update_global_block_rsv(fs_info);
5884 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5886 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5888 WARN_ON(fs_info->trans_block_rsv.size > 0);
5889 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5890 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5891 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5892 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5893 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5896 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
5897 struct btrfs_fs_info *fs_info)
5899 if (!trans->block_rsv) {
5900 ASSERT(!trans->bytes_reserved);
5904 if (!trans->bytes_reserved)
5907 ASSERT(trans->block_rsv == &fs_info->trans_block_rsv);
5908 trace_btrfs_space_reservation(fs_info, "transaction",
5909 trans->transid, trans->bytes_reserved, 0);
5910 btrfs_block_rsv_release(fs_info, trans->block_rsv,
5911 trans->bytes_reserved);
5912 trans->bytes_reserved = 0;
5916 * To be called after all the new block groups attached to the transaction
5917 * handle have been created (btrfs_create_pending_block_groups()).
5919 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5921 struct btrfs_fs_info *fs_info = trans->fs_info;
5923 if (!trans->chunk_bytes_reserved)
5926 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5928 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5929 trans->chunk_bytes_reserved);
5930 trans->chunk_bytes_reserved = 0;
5933 /* Can only return 0 or -ENOSPC */
5934 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5935 struct btrfs_inode *inode)
5937 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5938 struct btrfs_root *root = inode->root;
5940 * We always use trans->block_rsv here as we will have reserved space
5941 * for our orphan when starting the transaction, using get_block_rsv()
5942 * here will sometimes make us choose the wrong block rsv as we could be
5943 * doing a reloc inode for a non refcounted root.
5945 struct btrfs_block_rsv *src_rsv = trans->block_rsv;
5946 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5949 * We need to hold space in order to delete our orphan item once we've
5950 * added it, so this takes the reservation so we can release it later
5951 * when we are truly done with the orphan item.
5953 u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
5955 trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode),
5957 return btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1);
5960 void btrfs_orphan_release_metadata(struct btrfs_inode *inode)
5962 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5963 struct btrfs_root *root = inode->root;
5964 u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
5966 trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode),
5968 btrfs_block_rsv_release(fs_info, root->orphan_block_rsv, num_bytes);
5972 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5973 * root: the root of the parent directory
5974 * rsv: block reservation
5975 * items: the number of items that we need do reservation
5976 * qgroup_reserved: used to return the reserved size in qgroup
5978 * This function is used to reserve the space for snapshot/subvolume
5979 * creation and deletion. Those operations are different with the
5980 * common file/directory operations, they change two fs/file trees
5981 * and root tree, the number of items that the qgroup reserves is
5982 * different with the free space reservation. So we can not use
5983 * the space reservation mechanism in start_transaction().
5985 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5986 struct btrfs_block_rsv *rsv,
5988 u64 *qgroup_reserved,
5989 bool use_global_rsv)
5993 struct btrfs_fs_info *fs_info = root->fs_info;
5994 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5996 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
5997 /* One for parent inode, two for dir entries */
5998 num_bytes = 3 * fs_info->nodesize;
5999 ret = btrfs_qgroup_reserve_meta(root, num_bytes, true);
6006 *qgroup_reserved = num_bytes;
6008 num_bytes = btrfs_calc_trans_metadata_size(fs_info, items);
6009 rsv->space_info = __find_space_info(fs_info,
6010 BTRFS_BLOCK_GROUP_METADATA);
6011 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
6012 BTRFS_RESERVE_FLUSH_ALL);
6014 if (ret == -ENOSPC && use_global_rsv)
6015 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, 1);
6017 if (ret && *qgroup_reserved)
6018 btrfs_qgroup_free_meta(root, *qgroup_reserved);
6023 void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info,
6024 struct btrfs_block_rsv *rsv)
6026 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
6029 static void btrfs_calculate_inode_block_rsv_size(struct btrfs_fs_info *fs_info,
6030 struct btrfs_inode *inode)
6032 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
6033 u64 reserve_size = 0;
6035 unsigned outstanding_extents;
6037 lockdep_assert_held(&inode->lock);
6038 outstanding_extents = inode->outstanding_extents;
6039 if (outstanding_extents)
6040 reserve_size = btrfs_calc_trans_metadata_size(fs_info,
6041 outstanding_extents + 1);
6042 csum_leaves = btrfs_csum_bytes_to_leaves(fs_info,
6044 reserve_size += btrfs_calc_trans_metadata_size(fs_info,
6047 spin_lock(&block_rsv->lock);
6048 block_rsv->size = reserve_size;
6049 spin_unlock(&block_rsv->lock);
6052 int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes)
6054 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6055 struct btrfs_root *root = inode->root;
6056 unsigned nr_extents;
6057 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
6059 bool delalloc_lock = true;
6061 /* If we are a free space inode we need to not flush since we will be in
6062 * the middle of a transaction commit. We also don't need the delalloc
6063 * mutex since we won't race with anybody. We need this mostly to make
6064 * lockdep shut its filthy mouth.
6066 * If we have a transaction open (can happen if we call truncate_block
6067 * from truncate), then we need FLUSH_LIMIT so we don't deadlock.
6069 if (btrfs_is_free_space_inode(inode)) {
6070 flush = BTRFS_RESERVE_NO_FLUSH;
6071 delalloc_lock = false;
6072 } else if (current->journal_info) {
6073 flush = BTRFS_RESERVE_FLUSH_LIMIT;
6076 if (flush != BTRFS_RESERVE_NO_FLUSH &&
6077 btrfs_transaction_in_commit(fs_info))
6078 schedule_timeout(1);
6081 mutex_lock(&inode->delalloc_mutex);
6083 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6085 /* Add our new extents and calculate the new rsv size. */
6086 spin_lock(&inode->lock);
6087 nr_extents = count_max_extents(num_bytes);
6088 btrfs_mod_outstanding_extents(inode, nr_extents);
6089 inode->csum_bytes += num_bytes;
6090 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6091 spin_unlock(&inode->lock);
6093 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
6094 ret = btrfs_qgroup_reserve_meta(root,
6095 nr_extents * fs_info->nodesize, true);
6100 ret = btrfs_inode_rsv_refill(inode, flush);
6101 if (unlikely(ret)) {
6102 btrfs_qgroup_free_meta(root,
6103 nr_extents * fs_info->nodesize);
6108 mutex_unlock(&inode->delalloc_mutex);
6112 spin_lock(&inode->lock);
6113 nr_extents = count_max_extents(num_bytes);
6114 btrfs_mod_outstanding_extents(inode, -nr_extents);
6115 inode->csum_bytes -= num_bytes;
6116 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6117 spin_unlock(&inode->lock);
6119 btrfs_inode_rsv_release(inode);
6121 mutex_unlock(&inode->delalloc_mutex);
6126 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
6127 * @inode: the inode to release the reservation for.
6128 * @num_bytes: the number of bytes we are releasing.
6130 * This will release the metadata reservation for an inode. This can be called
6131 * once we complete IO for a given set of bytes to release their metadata
6132 * reservations, or on error for the same reason.
6134 void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes)
6136 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6138 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6139 spin_lock(&inode->lock);
6140 inode->csum_bytes -= num_bytes;
6141 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6142 spin_unlock(&inode->lock);
6144 if (btrfs_is_testing(fs_info))
6147 btrfs_inode_rsv_release(inode);
6151 * btrfs_delalloc_release_extents - release our outstanding_extents
6152 * @inode: the inode to balance the reservation for.
6153 * @num_bytes: the number of bytes we originally reserved with
6155 * When we reserve space we increase outstanding_extents for the extents we may
6156 * add. Once we've set the range as delalloc or created our ordered extents we
6157 * have outstanding_extents to track the real usage, so we use this to free our
6158 * temporarily tracked outstanding_extents. This _must_ be used in conjunction
6159 * with btrfs_delalloc_reserve_metadata.
6161 void btrfs_delalloc_release_extents(struct btrfs_inode *inode, u64 num_bytes)
6163 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6164 unsigned num_extents;
6166 spin_lock(&inode->lock);
6167 num_extents = count_max_extents(num_bytes);
6168 btrfs_mod_outstanding_extents(inode, -num_extents);
6169 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6170 spin_unlock(&inode->lock);
6172 if (btrfs_is_testing(fs_info))
6175 btrfs_inode_rsv_release(inode);
6179 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6181 * @inode: inode we're writing to
6182 * @start: start range we are writing to
6183 * @len: how long the range we are writing to
6184 * @reserved: mandatory parameter, record actually reserved qgroup ranges of
6185 * current reservation.
6187 * This will do the following things
6189 * o reserve space in data space info for num bytes
6190 * and reserve precious corresponding qgroup space
6191 * (Done in check_data_free_space)
6193 * o reserve space for metadata space, based on the number of outstanding
6194 * extents and how much csums will be needed
6195 * also reserve metadata space in a per root over-reserve method.
6196 * o add to the inodes->delalloc_bytes
6197 * o add it to the fs_info's delalloc inodes list.
6198 * (Above 3 all done in delalloc_reserve_metadata)
6200 * Return 0 for success
6201 * Return <0 for error(-ENOSPC or -EQUOT)
6203 int btrfs_delalloc_reserve_space(struct inode *inode,
6204 struct extent_changeset **reserved, u64 start, u64 len)
6208 ret = btrfs_check_data_free_space(inode, reserved, start, len);
6211 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len);
6213 btrfs_free_reserved_data_space(inode, *reserved, start, len);
6218 * btrfs_delalloc_release_space - release data and metadata space for delalloc
6219 * @inode: inode we're releasing space for
6220 * @start: start position of the space already reserved
6221 * @len: the len of the space already reserved
6222 * @release_bytes: the len of the space we consumed or didn't use
6224 * This function will release the metadata space that was not used and will
6225 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
6226 * list if there are no delalloc bytes left.
6227 * Also it will handle the qgroup reserved space.
6229 void btrfs_delalloc_release_space(struct inode *inode,
6230 struct extent_changeset *reserved,
6233 btrfs_delalloc_release_metadata(BTRFS_I(inode), len);
6234 btrfs_free_reserved_data_space(inode, reserved, start, len);
6237 static int update_block_group(struct btrfs_trans_handle *trans,
6238 struct btrfs_fs_info *info, u64 bytenr,
6239 u64 num_bytes, int alloc)
6241 struct btrfs_block_group_cache *cache = NULL;
6242 u64 total = num_bytes;
6247 /* block accounting for super block */
6248 spin_lock(&info->delalloc_root_lock);
6249 old_val = btrfs_super_bytes_used(info->super_copy);
6251 old_val += num_bytes;
6253 old_val -= num_bytes;
6254 btrfs_set_super_bytes_used(info->super_copy, old_val);
6255 spin_unlock(&info->delalloc_root_lock);
6258 cache = btrfs_lookup_block_group(info, bytenr);
6261 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
6262 BTRFS_BLOCK_GROUP_RAID1 |
6263 BTRFS_BLOCK_GROUP_RAID10))
6268 * If this block group has free space cache written out, we
6269 * need to make sure to load it if we are removing space. This
6270 * is because we need the unpinning stage to actually add the
6271 * space back to the block group, otherwise we will leak space.
6273 if (!alloc && cache->cached == BTRFS_CACHE_NO)
6274 cache_block_group(cache, 1);
6276 byte_in_group = bytenr - cache->key.objectid;
6277 WARN_ON(byte_in_group > cache->key.offset);
6279 spin_lock(&cache->space_info->lock);
6280 spin_lock(&cache->lock);
6282 if (btrfs_test_opt(info, SPACE_CACHE) &&
6283 cache->disk_cache_state < BTRFS_DC_CLEAR)
6284 cache->disk_cache_state = BTRFS_DC_CLEAR;
6286 old_val = btrfs_block_group_used(&cache->item);
6287 num_bytes = min(total, cache->key.offset - byte_in_group);
6289 old_val += num_bytes;
6290 btrfs_set_block_group_used(&cache->item, old_val);
6291 cache->reserved -= num_bytes;
6292 cache->space_info->bytes_reserved -= 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 old_val -= num_bytes;
6299 btrfs_set_block_group_used(&cache->item, old_val);
6300 cache->pinned += num_bytes;
6301 cache->space_info->bytes_pinned += num_bytes;
6302 cache->space_info->bytes_used -= num_bytes;
6303 cache->space_info->disk_used -= num_bytes * factor;
6304 spin_unlock(&cache->lock);
6305 spin_unlock(&cache->space_info->lock);
6307 trace_btrfs_space_reservation(info, "pinned",
6308 cache->space_info->flags,
6310 percpu_counter_add(&cache->space_info->total_bytes_pinned,
6312 set_extent_dirty(info->pinned_extents,
6313 bytenr, bytenr + num_bytes - 1,
6314 GFP_NOFS | __GFP_NOFAIL);
6317 spin_lock(&trans->transaction->dirty_bgs_lock);
6318 if (list_empty(&cache->dirty_list)) {
6319 list_add_tail(&cache->dirty_list,
6320 &trans->transaction->dirty_bgs);
6321 trans->transaction->num_dirty_bgs++;
6322 btrfs_get_block_group(cache);
6324 spin_unlock(&trans->transaction->dirty_bgs_lock);
6327 * No longer have used bytes in this block group, queue it for
6328 * deletion. We do this after adding the block group to the
6329 * dirty list to avoid races between cleaner kthread and space
6332 if (!alloc && old_val == 0) {
6333 spin_lock(&info->unused_bgs_lock);
6334 if (list_empty(&cache->bg_list)) {
6335 btrfs_get_block_group(cache);
6336 list_add_tail(&cache->bg_list,
6339 spin_unlock(&info->unused_bgs_lock);
6342 btrfs_put_block_group(cache);
6344 bytenr += num_bytes;
6349 static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start)
6351 struct btrfs_block_group_cache *cache;
6354 spin_lock(&fs_info->block_group_cache_lock);
6355 bytenr = fs_info->first_logical_byte;
6356 spin_unlock(&fs_info->block_group_cache_lock);
6358 if (bytenr < (u64)-1)
6361 cache = btrfs_lookup_first_block_group(fs_info, search_start);
6365 bytenr = cache->key.objectid;
6366 btrfs_put_block_group(cache);
6371 static int pin_down_extent(struct btrfs_fs_info *fs_info,
6372 struct btrfs_block_group_cache *cache,
6373 u64 bytenr, u64 num_bytes, int reserved)
6375 spin_lock(&cache->space_info->lock);
6376 spin_lock(&cache->lock);
6377 cache->pinned += num_bytes;
6378 cache->space_info->bytes_pinned += num_bytes;
6380 cache->reserved -= num_bytes;
6381 cache->space_info->bytes_reserved -= num_bytes;
6383 spin_unlock(&cache->lock);
6384 spin_unlock(&cache->space_info->lock);
6386 trace_btrfs_space_reservation(fs_info, "pinned",
6387 cache->space_info->flags, num_bytes, 1);
6388 percpu_counter_add(&cache->space_info->total_bytes_pinned, num_bytes);
6389 set_extent_dirty(fs_info->pinned_extents, bytenr,
6390 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
6395 * this function must be called within transaction
6397 int btrfs_pin_extent(struct btrfs_fs_info *fs_info,
6398 u64 bytenr, u64 num_bytes, int reserved)
6400 struct btrfs_block_group_cache *cache;
6402 cache = btrfs_lookup_block_group(fs_info, bytenr);
6403 BUG_ON(!cache); /* Logic error */
6405 pin_down_extent(fs_info, cache, bytenr, num_bytes, reserved);
6407 btrfs_put_block_group(cache);
6412 * this function must be called within transaction
6414 int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info,
6415 u64 bytenr, u64 num_bytes)
6417 struct btrfs_block_group_cache *cache;
6420 cache = btrfs_lookup_block_group(fs_info, bytenr);
6425 * pull in the free space cache (if any) so that our pin
6426 * removes the free space from the cache. We have load_only set
6427 * to one because the slow code to read in the free extents does check
6428 * the pinned extents.
6430 cache_block_group(cache, 1);
6432 pin_down_extent(fs_info, cache, bytenr, num_bytes, 0);
6434 /* remove us from the free space cache (if we're there at all) */
6435 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6436 btrfs_put_block_group(cache);
6440 static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
6441 u64 start, u64 num_bytes)
6444 struct btrfs_block_group_cache *block_group;
6445 struct btrfs_caching_control *caching_ctl;
6447 block_group = btrfs_lookup_block_group(fs_info, start);
6451 cache_block_group(block_group, 0);
6452 caching_ctl = get_caching_control(block_group);
6456 BUG_ON(!block_group_cache_done(block_group));
6457 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6459 mutex_lock(&caching_ctl->mutex);
6461 if (start >= caching_ctl->progress) {
6462 ret = add_excluded_extent(fs_info, start, num_bytes);
6463 } else if (start + num_bytes <= caching_ctl->progress) {
6464 ret = btrfs_remove_free_space(block_group,
6467 num_bytes = caching_ctl->progress - start;
6468 ret = btrfs_remove_free_space(block_group,
6473 num_bytes = (start + num_bytes) -
6474 caching_ctl->progress;
6475 start = caching_ctl->progress;
6476 ret = add_excluded_extent(fs_info, start, num_bytes);
6479 mutex_unlock(&caching_ctl->mutex);
6480 put_caching_control(caching_ctl);
6482 btrfs_put_block_group(block_group);
6486 int btrfs_exclude_logged_extents(struct btrfs_fs_info *fs_info,
6487 struct extent_buffer *eb)
6489 struct btrfs_file_extent_item *item;
6490 struct btrfs_key key;
6494 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
6497 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6498 btrfs_item_key_to_cpu(eb, &key, i);
6499 if (key.type != BTRFS_EXTENT_DATA_KEY)
6501 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6502 found_type = btrfs_file_extent_type(eb, item);
6503 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6505 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6507 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6508 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6509 __exclude_logged_extent(fs_info, key.objectid, key.offset);
6516 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg)
6518 atomic_inc(&bg->reservations);
6521 void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info,
6524 struct btrfs_block_group_cache *bg;
6526 bg = btrfs_lookup_block_group(fs_info, start);
6528 if (atomic_dec_and_test(&bg->reservations))
6529 wake_up_atomic_t(&bg->reservations);
6530 btrfs_put_block_group(bg);
6533 static int btrfs_wait_bg_reservations_atomic_t(atomic_t *a)
6539 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg)
6541 struct btrfs_space_info *space_info = bg->space_info;
6545 if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA))
6549 * Our block group is read only but before we set it to read only,
6550 * some task might have had allocated an extent from it already, but it
6551 * has not yet created a respective ordered extent (and added it to a
6552 * root's list of ordered extents).
6553 * Therefore wait for any task currently allocating extents, since the
6554 * block group's reservations counter is incremented while a read lock
6555 * on the groups' semaphore is held and decremented after releasing
6556 * the read access on that semaphore and creating the ordered extent.
6558 down_write(&space_info->groups_sem);
6559 up_write(&space_info->groups_sem);
6561 wait_on_atomic_t(&bg->reservations,
6562 btrfs_wait_bg_reservations_atomic_t,
6563 TASK_UNINTERRUPTIBLE);
6567 * btrfs_add_reserved_bytes - update the block_group and space info counters
6568 * @cache: The cache we are manipulating
6569 * @ram_bytes: The number of bytes of file content, and will be same to
6570 * @num_bytes except for the compress path.
6571 * @num_bytes: The number of bytes in question
6572 * @delalloc: The blocks are allocated for the delalloc write
6574 * This is called by the allocator when it reserves space. If this is a
6575 * reservation and the block group has become read only we cannot make the
6576 * reservation and return -EAGAIN, otherwise this function always succeeds.
6578 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache,
6579 u64 ram_bytes, u64 num_bytes, int delalloc)
6581 struct btrfs_space_info *space_info = cache->space_info;
6584 spin_lock(&space_info->lock);
6585 spin_lock(&cache->lock);
6589 cache->reserved += num_bytes;
6590 space_info->bytes_reserved += num_bytes;
6592 trace_btrfs_space_reservation(cache->fs_info,
6593 "space_info", space_info->flags,
6595 space_info->bytes_may_use -= ram_bytes;
6597 cache->delalloc_bytes += num_bytes;
6599 spin_unlock(&cache->lock);
6600 spin_unlock(&space_info->lock);
6605 * btrfs_free_reserved_bytes - update the block_group and space info counters
6606 * @cache: The cache we are manipulating
6607 * @num_bytes: The number of bytes in question
6608 * @delalloc: The blocks are allocated for the delalloc write
6610 * This is called by somebody who is freeing space that was never actually used
6611 * on disk. For example if you reserve some space for a new leaf in transaction
6612 * A and before transaction A commits you free that leaf, you call this with
6613 * reserve set to 0 in order to clear the reservation.
6616 static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache,
6617 u64 num_bytes, int delalloc)
6619 struct btrfs_space_info *space_info = cache->space_info;
6622 spin_lock(&space_info->lock);
6623 spin_lock(&cache->lock);
6625 space_info->bytes_readonly += num_bytes;
6626 cache->reserved -= num_bytes;
6627 space_info->bytes_reserved -= num_bytes;
6630 cache->delalloc_bytes -= num_bytes;
6631 spin_unlock(&cache->lock);
6632 spin_unlock(&space_info->lock);
6635 void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info)
6637 struct btrfs_caching_control *next;
6638 struct btrfs_caching_control *caching_ctl;
6639 struct btrfs_block_group_cache *cache;
6641 down_write(&fs_info->commit_root_sem);
6643 list_for_each_entry_safe(caching_ctl, next,
6644 &fs_info->caching_block_groups, list) {
6645 cache = caching_ctl->block_group;
6646 if (block_group_cache_done(cache)) {
6647 cache->last_byte_to_unpin = (u64)-1;
6648 list_del_init(&caching_ctl->list);
6649 put_caching_control(caching_ctl);
6651 cache->last_byte_to_unpin = caching_ctl->progress;
6655 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6656 fs_info->pinned_extents = &fs_info->freed_extents[1];
6658 fs_info->pinned_extents = &fs_info->freed_extents[0];
6660 up_write(&fs_info->commit_root_sem);
6662 update_global_block_rsv(fs_info);
6666 * Returns the free cluster for the given space info and sets empty_cluster to
6667 * what it should be based on the mount options.
6669 static struct btrfs_free_cluster *
6670 fetch_cluster_info(struct btrfs_fs_info *fs_info,
6671 struct btrfs_space_info *space_info, u64 *empty_cluster)
6673 struct btrfs_free_cluster *ret = NULL;
6676 if (btrfs_mixed_space_info(space_info))
6679 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
6680 ret = &fs_info->meta_alloc_cluster;
6681 if (btrfs_test_opt(fs_info, SSD))
6682 *empty_cluster = SZ_2M;
6684 *empty_cluster = SZ_64K;
6685 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
6686 btrfs_test_opt(fs_info, SSD_SPREAD)) {
6687 *empty_cluster = SZ_2M;
6688 ret = &fs_info->data_alloc_cluster;
6694 static int unpin_extent_range(struct btrfs_fs_info *fs_info,
6696 const bool return_free_space)
6698 struct btrfs_block_group_cache *cache = NULL;
6699 struct btrfs_space_info *space_info;
6700 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6701 struct btrfs_free_cluster *cluster = NULL;
6703 u64 total_unpinned = 0;
6704 u64 empty_cluster = 0;
6707 while (start <= end) {
6710 start >= cache->key.objectid + cache->key.offset) {
6712 btrfs_put_block_group(cache);
6714 cache = btrfs_lookup_block_group(fs_info, start);
6715 BUG_ON(!cache); /* Logic error */
6717 cluster = fetch_cluster_info(fs_info,
6720 empty_cluster <<= 1;
6723 len = cache->key.objectid + cache->key.offset - start;
6724 len = min(len, end + 1 - start);
6726 if (start < cache->last_byte_to_unpin) {
6727 len = min(len, cache->last_byte_to_unpin - start);
6728 if (return_free_space)
6729 btrfs_add_free_space(cache, start, len);
6733 total_unpinned += len;
6734 space_info = cache->space_info;
6737 * If this space cluster has been marked as fragmented and we've
6738 * unpinned enough in this block group to potentially allow a
6739 * cluster to be created inside of it go ahead and clear the
6742 if (cluster && cluster->fragmented &&
6743 total_unpinned > empty_cluster) {
6744 spin_lock(&cluster->lock);
6745 cluster->fragmented = 0;
6746 spin_unlock(&cluster->lock);
6749 spin_lock(&space_info->lock);
6750 spin_lock(&cache->lock);
6751 cache->pinned -= len;
6752 space_info->bytes_pinned -= len;
6754 trace_btrfs_space_reservation(fs_info, "pinned",
6755 space_info->flags, len, 0);
6756 space_info->max_extent_size = 0;
6757 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6759 space_info->bytes_readonly += len;
6762 spin_unlock(&cache->lock);
6763 if (!readonly && return_free_space &&
6764 global_rsv->space_info == space_info) {
6767 spin_lock(&global_rsv->lock);
6768 if (!global_rsv->full) {
6769 to_add = min(len, global_rsv->size -
6770 global_rsv->reserved);
6771 global_rsv->reserved += to_add;
6772 space_info->bytes_may_use += to_add;
6773 if (global_rsv->reserved >= global_rsv->size)
6774 global_rsv->full = 1;
6775 trace_btrfs_space_reservation(fs_info,
6781 spin_unlock(&global_rsv->lock);
6782 /* Add to any tickets we may have */
6784 space_info_add_new_bytes(fs_info, space_info,
6787 spin_unlock(&space_info->lock);
6791 btrfs_put_block_group(cache);
6795 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
6796 struct btrfs_fs_info *fs_info)
6798 struct btrfs_block_group_cache *block_group, *tmp;
6799 struct list_head *deleted_bgs;
6800 struct extent_io_tree *unpin;
6805 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6806 unpin = &fs_info->freed_extents[1];
6808 unpin = &fs_info->freed_extents[0];
6810 while (!trans->aborted) {
6811 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6812 ret = find_first_extent_bit(unpin, 0, &start, &end,
6813 EXTENT_DIRTY, NULL);
6815 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6819 if (btrfs_test_opt(fs_info, DISCARD))
6820 ret = btrfs_discard_extent(fs_info, start,
6821 end + 1 - start, NULL);
6823 clear_extent_dirty(unpin, start, end);
6824 unpin_extent_range(fs_info, start, end, true);
6825 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6830 * Transaction is finished. We don't need the lock anymore. We
6831 * do need to clean up the block groups in case of a transaction
6834 deleted_bgs = &trans->transaction->deleted_bgs;
6835 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6839 if (!trans->aborted)
6840 ret = btrfs_discard_extent(fs_info,
6841 block_group->key.objectid,
6842 block_group->key.offset,
6845 list_del_init(&block_group->bg_list);
6846 btrfs_put_block_group_trimming(block_group);
6847 btrfs_put_block_group(block_group);
6850 const char *errstr = btrfs_decode_error(ret);
6852 "discard failed while removing blockgroup: errno=%d %s",
6860 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6861 struct btrfs_fs_info *info,
6862 struct btrfs_delayed_ref_node *node, u64 parent,
6863 u64 root_objectid, u64 owner_objectid,
6864 u64 owner_offset, int refs_to_drop,
6865 struct btrfs_delayed_extent_op *extent_op)
6867 struct btrfs_key key;
6868 struct btrfs_path *path;
6869 struct btrfs_root *extent_root = info->extent_root;
6870 struct extent_buffer *leaf;
6871 struct btrfs_extent_item *ei;
6872 struct btrfs_extent_inline_ref *iref;
6875 int extent_slot = 0;
6876 int found_extent = 0;
6880 u64 bytenr = node->bytenr;
6881 u64 num_bytes = node->num_bytes;
6883 bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
6885 path = btrfs_alloc_path();
6889 path->reada = READA_FORWARD;
6890 path->leave_spinning = 1;
6892 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6893 BUG_ON(!is_data && refs_to_drop != 1);
6896 skinny_metadata = false;
6898 ret = lookup_extent_backref(trans, info, path, &iref,
6899 bytenr, num_bytes, parent,
6900 root_objectid, owner_objectid,
6903 extent_slot = path->slots[0];
6904 while (extent_slot >= 0) {
6905 btrfs_item_key_to_cpu(path->nodes[0], &key,
6907 if (key.objectid != bytenr)
6909 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6910 key.offset == num_bytes) {
6914 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6915 key.offset == owner_objectid) {
6919 if (path->slots[0] - extent_slot > 5)
6923 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6924 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6925 if (found_extent && item_size < sizeof(*ei))
6928 if (!found_extent) {
6930 ret = remove_extent_backref(trans, info, path, NULL,
6932 is_data, &last_ref);
6934 btrfs_abort_transaction(trans, ret);
6937 btrfs_release_path(path);
6938 path->leave_spinning = 1;
6940 key.objectid = bytenr;
6941 key.type = BTRFS_EXTENT_ITEM_KEY;
6942 key.offset = num_bytes;
6944 if (!is_data && skinny_metadata) {
6945 key.type = BTRFS_METADATA_ITEM_KEY;
6946 key.offset = owner_objectid;
6949 ret = btrfs_search_slot(trans, extent_root,
6951 if (ret > 0 && skinny_metadata && path->slots[0]) {
6953 * Couldn't find our skinny metadata item,
6954 * see if we have ye olde extent item.
6957 btrfs_item_key_to_cpu(path->nodes[0], &key,
6959 if (key.objectid == bytenr &&
6960 key.type == BTRFS_EXTENT_ITEM_KEY &&
6961 key.offset == num_bytes)
6965 if (ret > 0 && skinny_metadata) {
6966 skinny_metadata = false;
6967 key.objectid = bytenr;
6968 key.type = BTRFS_EXTENT_ITEM_KEY;
6969 key.offset = num_bytes;
6970 btrfs_release_path(path);
6971 ret = btrfs_search_slot(trans, extent_root,
6977 "umm, got %d back from search, was looking for %llu",
6980 btrfs_print_leaf(path->nodes[0]);
6983 btrfs_abort_transaction(trans, ret);
6986 extent_slot = path->slots[0];
6988 } else if (WARN_ON(ret == -ENOENT)) {
6989 btrfs_print_leaf(path->nodes[0]);
6991 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6992 bytenr, parent, root_objectid, owner_objectid,
6994 btrfs_abort_transaction(trans, ret);
6997 btrfs_abort_transaction(trans, ret);
7001 leaf = path->nodes[0];
7002 item_size = btrfs_item_size_nr(leaf, extent_slot);
7003 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
7004 if (item_size < sizeof(*ei)) {
7005 BUG_ON(found_extent || extent_slot != path->slots[0]);
7006 ret = convert_extent_item_v0(trans, info, path, owner_objectid,
7009 btrfs_abort_transaction(trans, ret);
7013 btrfs_release_path(path);
7014 path->leave_spinning = 1;
7016 key.objectid = bytenr;
7017 key.type = BTRFS_EXTENT_ITEM_KEY;
7018 key.offset = num_bytes;
7020 ret = btrfs_search_slot(trans, extent_root, &key, path,
7024 "umm, got %d back from search, was looking for %llu",
7026 btrfs_print_leaf(path->nodes[0]);
7029 btrfs_abort_transaction(trans, ret);
7033 extent_slot = path->slots[0];
7034 leaf = path->nodes[0];
7035 item_size = btrfs_item_size_nr(leaf, extent_slot);
7038 BUG_ON(item_size < sizeof(*ei));
7039 ei = btrfs_item_ptr(leaf, extent_slot,
7040 struct btrfs_extent_item);
7041 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
7042 key.type == BTRFS_EXTENT_ITEM_KEY) {
7043 struct btrfs_tree_block_info *bi;
7044 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
7045 bi = (struct btrfs_tree_block_info *)(ei + 1);
7046 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
7049 refs = btrfs_extent_refs(leaf, ei);
7050 if (refs < refs_to_drop) {
7052 "trying to drop %d refs but we only have %Lu for bytenr %Lu",
7053 refs_to_drop, refs, bytenr);
7055 btrfs_abort_transaction(trans, ret);
7058 refs -= refs_to_drop;
7062 __run_delayed_extent_op(extent_op, leaf, ei);
7064 * In the case of inline back ref, reference count will
7065 * be updated by remove_extent_backref
7068 BUG_ON(!found_extent);
7070 btrfs_set_extent_refs(leaf, ei, refs);
7071 btrfs_mark_buffer_dirty(leaf);
7074 ret = remove_extent_backref(trans, info, path,
7076 is_data, &last_ref);
7078 btrfs_abort_transaction(trans, ret);
7084 BUG_ON(is_data && refs_to_drop !=
7085 extent_data_ref_count(path, iref));
7087 BUG_ON(path->slots[0] != extent_slot);
7089 BUG_ON(path->slots[0] != extent_slot + 1);
7090 path->slots[0] = extent_slot;
7096 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
7099 btrfs_abort_transaction(trans, ret);
7102 btrfs_release_path(path);
7105 ret = btrfs_del_csums(trans, info, bytenr, num_bytes);
7107 btrfs_abort_transaction(trans, ret);
7112 ret = add_to_free_space_tree(trans, info, bytenr, num_bytes);
7114 btrfs_abort_transaction(trans, ret);
7118 ret = update_block_group(trans, info, bytenr, num_bytes, 0);
7120 btrfs_abort_transaction(trans, ret);
7124 btrfs_release_path(path);
7127 btrfs_free_path(path);
7132 * when we free an block, it is possible (and likely) that we free the last
7133 * delayed ref for that extent as well. This searches the delayed ref tree for
7134 * a given extent, and if there are no other delayed refs to be processed, it
7135 * removes it from the tree.
7137 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
7140 struct btrfs_delayed_ref_head *head;
7141 struct btrfs_delayed_ref_root *delayed_refs;
7144 delayed_refs = &trans->transaction->delayed_refs;
7145 spin_lock(&delayed_refs->lock);
7146 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
7148 goto out_delayed_unlock;
7150 spin_lock(&head->lock);
7151 if (!RB_EMPTY_ROOT(&head->ref_tree))
7154 if (head->extent_op) {
7155 if (!head->must_insert_reserved)
7157 btrfs_free_delayed_extent_op(head->extent_op);
7158 head->extent_op = NULL;
7162 * waiting for the lock here would deadlock. If someone else has it
7163 * locked they are already in the process of dropping it anyway
7165 if (!mutex_trylock(&head->mutex))
7169 * at this point we have a head with no other entries. Go
7170 * ahead and process it.
7172 rb_erase(&head->href_node, &delayed_refs->href_root);
7173 RB_CLEAR_NODE(&head->href_node);
7174 atomic_dec(&delayed_refs->num_entries);
7177 * we don't take a ref on the node because we're removing it from the
7178 * tree, so we just steal the ref the tree was holding.
7180 delayed_refs->num_heads--;
7181 if (head->processing == 0)
7182 delayed_refs->num_heads_ready--;
7183 head->processing = 0;
7184 spin_unlock(&head->lock);
7185 spin_unlock(&delayed_refs->lock);
7187 BUG_ON(head->extent_op);
7188 if (head->must_insert_reserved)
7191 mutex_unlock(&head->mutex);
7192 btrfs_put_delayed_ref_head(head);
7195 spin_unlock(&head->lock);
7198 spin_unlock(&delayed_refs->lock);
7202 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
7203 struct btrfs_root *root,
7204 struct extent_buffer *buf,
7205 u64 parent, int last_ref)
7207 struct btrfs_fs_info *fs_info = root->fs_info;
7211 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7212 int old_ref_mod, new_ref_mod;
7214 btrfs_ref_tree_mod(root, buf->start, buf->len, parent,
7215 root->root_key.objectid,
7216 btrfs_header_level(buf), 0,
7217 BTRFS_DROP_DELAYED_REF);
7218 ret = btrfs_add_delayed_tree_ref(fs_info, trans, buf->start,
7220 root->root_key.objectid,
7221 btrfs_header_level(buf),
7222 BTRFS_DROP_DELAYED_REF, NULL,
7223 &old_ref_mod, &new_ref_mod);
7224 BUG_ON(ret); /* -ENOMEM */
7225 pin = old_ref_mod >= 0 && new_ref_mod < 0;
7228 if (last_ref && btrfs_header_generation(buf) == trans->transid) {
7229 struct btrfs_block_group_cache *cache;
7231 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7232 ret = check_ref_cleanup(trans, buf->start);
7238 cache = btrfs_lookup_block_group(fs_info, buf->start);
7240 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
7241 pin_down_extent(fs_info, cache, buf->start,
7243 btrfs_put_block_group(cache);
7247 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
7249 btrfs_add_free_space(cache, buf->start, buf->len);
7250 btrfs_free_reserved_bytes(cache, buf->len, 0);
7251 btrfs_put_block_group(cache);
7252 trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
7256 add_pinned_bytes(fs_info, buf->len, btrfs_header_level(buf),
7257 root->root_key.objectid);
7261 * Deleting the buffer, clear the corrupt flag since it doesn't
7264 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
7268 /* Can return -ENOMEM */
7269 int btrfs_free_extent(struct btrfs_trans_handle *trans,
7270 struct btrfs_root *root,
7271 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
7272 u64 owner, u64 offset)
7274 struct btrfs_fs_info *fs_info = root->fs_info;
7275 int old_ref_mod, new_ref_mod;
7278 if (btrfs_is_testing(fs_info))
7281 if (root_objectid != BTRFS_TREE_LOG_OBJECTID)
7282 btrfs_ref_tree_mod(root, bytenr, num_bytes, parent,
7283 root_objectid, owner, offset,
7284 BTRFS_DROP_DELAYED_REF);
7287 * tree log blocks never actually go into the extent allocation
7288 * tree, just update pinning info and exit early.
7290 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
7291 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
7292 /* unlocks the pinned mutex */
7293 btrfs_pin_extent(fs_info, bytenr, num_bytes, 1);
7294 old_ref_mod = new_ref_mod = 0;
7296 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
7297 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
7299 root_objectid, (int)owner,
7300 BTRFS_DROP_DELAYED_REF, NULL,
7301 &old_ref_mod, &new_ref_mod);
7303 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
7305 root_objectid, owner, offset,
7306 0, BTRFS_DROP_DELAYED_REF,
7307 &old_ref_mod, &new_ref_mod);
7310 if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0)
7311 add_pinned_bytes(fs_info, num_bytes, owner, root_objectid);
7317 * when we wait for progress in the block group caching, its because
7318 * our allocation attempt failed at least once. So, we must sleep
7319 * and let some progress happen before we try again.
7321 * This function will sleep at least once waiting for new free space to
7322 * show up, and then it will check the block group free space numbers
7323 * for our min num_bytes. Another option is to have it go ahead
7324 * and look in the rbtree for a free extent of a given size, but this
7327 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7328 * any of the information in this block group.
7330 static noinline void
7331 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
7334 struct btrfs_caching_control *caching_ctl;
7336 caching_ctl = get_caching_control(cache);
7340 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
7341 (cache->free_space_ctl->free_space >= num_bytes));
7343 put_caching_control(caching_ctl);
7347 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
7349 struct btrfs_caching_control *caching_ctl;
7352 caching_ctl = get_caching_control(cache);
7354 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
7356 wait_event(caching_ctl->wait, block_group_cache_done(cache));
7357 if (cache->cached == BTRFS_CACHE_ERROR)
7359 put_caching_control(caching_ctl);
7363 int __get_raid_index(u64 flags)
7365 if (flags & BTRFS_BLOCK_GROUP_RAID10)
7366 return BTRFS_RAID_RAID10;
7367 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
7368 return BTRFS_RAID_RAID1;
7369 else if (flags & BTRFS_BLOCK_GROUP_DUP)
7370 return BTRFS_RAID_DUP;
7371 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
7372 return BTRFS_RAID_RAID0;
7373 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
7374 return BTRFS_RAID_RAID5;
7375 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
7376 return BTRFS_RAID_RAID6;
7378 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
7381 int get_block_group_index(struct btrfs_block_group_cache *cache)
7383 return __get_raid_index(cache->flags);
7386 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
7387 [BTRFS_RAID_RAID10] = "raid10",
7388 [BTRFS_RAID_RAID1] = "raid1",
7389 [BTRFS_RAID_DUP] = "dup",
7390 [BTRFS_RAID_RAID0] = "raid0",
7391 [BTRFS_RAID_SINGLE] = "single",
7392 [BTRFS_RAID_RAID5] = "raid5",
7393 [BTRFS_RAID_RAID6] = "raid6",
7396 static const char *get_raid_name(enum btrfs_raid_types type)
7398 if (type >= BTRFS_NR_RAID_TYPES)
7401 return btrfs_raid_type_names[type];
7404 enum btrfs_loop_type {
7405 LOOP_CACHING_NOWAIT = 0,
7406 LOOP_CACHING_WAIT = 1,
7407 LOOP_ALLOC_CHUNK = 2,
7408 LOOP_NO_EMPTY_SIZE = 3,
7412 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
7416 down_read(&cache->data_rwsem);
7420 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
7423 btrfs_get_block_group(cache);
7425 down_read(&cache->data_rwsem);
7428 static struct btrfs_block_group_cache *
7429 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
7430 struct btrfs_free_cluster *cluster,
7433 struct btrfs_block_group_cache *used_bg = NULL;
7435 spin_lock(&cluster->refill_lock);
7437 used_bg = cluster->block_group;
7441 if (used_bg == block_group)
7444 btrfs_get_block_group(used_bg);
7449 if (down_read_trylock(&used_bg->data_rwsem))
7452 spin_unlock(&cluster->refill_lock);
7454 /* We should only have one-level nested. */
7455 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
7457 spin_lock(&cluster->refill_lock);
7458 if (used_bg == cluster->block_group)
7461 up_read(&used_bg->data_rwsem);
7462 btrfs_put_block_group(used_bg);
7467 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
7471 up_read(&cache->data_rwsem);
7472 btrfs_put_block_group(cache);
7476 * walks the btree of allocated extents and find a hole of a given size.
7477 * The key ins is changed to record the hole:
7478 * ins->objectid == start position
7479 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7480 * ins->offset == the size of the hole.
7481 * Any available blocks before search_start are skipped.
7483 * If there is no suitable free space, we will record the max size of
7484 * the free space extent currently.
7486 static noinline int find_free_extent(struct btrfs_fs_info *fs_info,
7487 u64 ram_bytes, u64 num_bytes, u64 empty_size,
7488 u64 hint_byte, struct btrfs_key *ins,
7489 u64 flags, int delalloc)
7492 struct btrfs_root *root = fs_info->extent_root;
7493 struct btrfs_free_cluster *last_ptr = NULL;
7494 struct btrfs_block_group_cache *block_group = NULL;
7495 u64 search_start = 0;
7496 u64 max_extent_size = 0;
7497 u64 empty_cluster = 0;
7498 struct btrfs_space_info *space_info;
7500 int index = __get_raid_index(flags);
7501 bool failed_cluster_refill = false;
7502 bool failed_alloc = false;
7503 bool use_cluster = true;
7504 bool have_caching_bg = false;
7505 bool orig_have_caching_bg = false;
7506 bool full_search = false;
7508 WARN_ON(num_bytes < fs_info->sectorsize);
7509 ins->type = BTRFS_EXTENT_ITEM_KEY;
7513 trace_find_free_extent(fs_info, num_bytes, empty_size, flags);
7515 space_info = __find_space_info(fs_info, flags);
7517 btrfs_err(fs_info, "No space info for %llu", flags);
7522 * If our free space is heavily fragmented we may not be able to make
7523 * big contiguous allocations, so instead of doing the expensive search
7524 * for free space, simply return ENOSPC with our max_extent_size so we
7525 * can go ahead and search for a more manageable chunk.
7527 * If our max_extent_size is large enough for our allocation simply
7528 * disable clustering since we will likely not be able to find enough
7529 * space to create a cluster and induce latency trying.
7531 if (unlikely(space_info->max_extent_size)) {
7532 spin_lock(&space_info->lock);
7533 if (space_info->max_extent_size &&
7534 num_bytes > space_info->max_extent_size) {
7535 ins->offset = space_info->max_extent_size;
7536 spin_unlock(&space_info->lock);
7538 } else if (space_info->max_extent_size) {
7539 use_cluster = false;
7541 spin_unlock(&space_info->lock);
7544 last_ptr = fetch_cluster_info(fs_info, space_info, &empty_cluster);
7546 spin_lock(&last_ptr->lock);
7547 if (last_ptr->block_group)
7548 hint_byte = last_ptr->window_start;
7549 if (last_ptr->fragmented) {
7551 * We still set window_start so we can keep track of the
7552 * last place we found an allocation to try and save
7555 hint_byte = last_ptr->window_start;
7556 use_cluster = false;
7558 spin_unlock(&last_ptr->lock);
7561 search_start = max(search_start, first_logical_byte(fs_info, 0));
7562 search_start = max(search_start, hint_byte);
7563 if (search_start == hint_byte) {
7564 block_group = btrfs_lookup_block_group(fs_info, search_start);
7566 * we don't want to use the block group if it doesn't match our
7567 * allocation bits, or if its not cached.
7569 * However if we are re-searching with an ideal block group
7570 * picked out then we don't care that the block group is cached.
7572 if (block_group && block_group_bits(block_group, flags) &&
7573 block_group->cached != BTRFS_CACHE_NO) {
7574 down_read(&space_info->groups_sem);
7575 if (list_empty(&block_group->list) ||
7578 * someone is removing this block group,
7579 * we can't jump into the have_block_group
7580 * target because our list pointers are not
7583 btrfs_put_block_group(block_group);
7584 up_read(&space_info->groups_sem);
7586 index = get_block_group_index(block_group);
7587 btrfs_lock_block_group(block_group, delalloc);
7588 goto have_block_group;
7590 } else if (block_group) {
7591 btrfs_put_block_group(block_group);
7595 have_caching_bg = false;
7596 if (index == 0 || index == __get_raid_index(flags))
7598 down_read(&space_info->groups_sem);
7599 list_for_each_entry(block_group, &space_info->block_groups[index],
7604 /* If the block group is read-only, we can skip it entirely. */
7605 if (unlikely(block_group->ro))
7608 btrfs_grab_block_group(block_group, delalloc);
7609 search_start = block_group->key.objectid;
7612 * this can happen if we end up cycling through all the
7613 * raid types, but we want to make sure we only allocate
7614 * for the proper type.
7616 if (!block_group_bits(block_group, flags)) {
7617 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7618 BTRFS_BLOCK_GROUP_RAID1 |
7619 BTRFS_BLOCK_GROUP_RAID5 |
7620 BTRFS_BLOCK_GROUP_RAID6 |
7621 BTRFS_BLOCK_GROUP_RAID10;
7624 * if they asked for extra copies and this block group
7625 * doesn't provide them, bail. This does allow us to
7626 * fill raid0 from raid1.
7628 if ((flags & extra) && !(block_group->flags & extra))
7633 cached = block_group_cache_done(block_group);
7634 if (unlikely(!cached)) {
7635 have_caching_bg = true;
7636 ret = cache_block_group(block_group, 0);
7641 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7645 * Ok we want to try and use the cluster allocator, so
7648 if (last_ptr && use_cluster) {
7649 struct btrfs_block_group_cache *used_block_group;
7650 unsigned long aligned_cluster;
7652 * the refill lock keeps out other
7653 * people trying to start a new cluster
7655 used_block_group = btrfs_lock_cluster(block_group,
7658 if (!used_block_group)
7659 goto refill_cluster;
7661 if (used_block_group != block_group &&
7662 (used_block_group->ro ||
7663 !block_group_bits(used_block_group, flags)))
7664 goto release_cluster;
7666 offset = btrfs_alloc_from_cluster(used_block_group,
7669 used_block_group->key.objectid,
7672 /* we have a block, we're done */
7673 spin_unlock(&last_ptr->refill_lock);
7674 trace_btrfs_reserve_extent_cluster(fs_info,
7676 search_start, num_bytes);
7677 if (used_block_group != block_group) {
7678 btrfs_release_block_group(block_group,
7680 block_group = used_block_group;
7685 WARN_ON(last_ptr->block_group != used_block_group);
7687 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7688 * set up a new clusters, so lets just skip it
7689 * and let the allocator find whatever block
7690 * it can find. If we reach this point, we
7691 * will have tried the cluster allocator
7692 * plenty of times and not have found
7693 * anything, so we are likely way too
7694 * fragmented for the clustering stuff to find
7697 * However, if the cluster is taken from the
7698 * current block group, release the cluster
7699 * first, so that we stand a better chance of
7700 * succeeding in the unclustered
7702 if (loop >= LOOP_NO_EMPTY_SIZE &&
7703 used_block_group != block_group) {
7704 spin_unlock(&last_ptr->refill_lock);
7705 btrfs_release_block_group(used_block_group,
7707 goto unclustered_alloc;
7711 * this cluster didn't work out, free it and
7714 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7716 if (used_block_group != block_group)
7717 btrfs_release_block_group(used_block_group,
7720 if (loop >= LOOP_NO_EMPTY_SIZE) {
7721 spin_unlock(&last_ptr->refill_lock);
7722 goto unclustered_alloc;
7725 aligned_cluster = max_t(unsigned long,
7726 empty_cluster + empty_size,
7727 block_group->full_stripe_len);
7729 /* allocate a cluster in this block group */
7730 ret = btrfs_find_space_cluster(fs_info, block_group,
7731 last_ptr, search_start,
7736 * now pull our allocation out of this
7739 offset = btrfs_alloc_from_cluster(block_group,
7745 /* we found one, proceed */
7746 spin_unlock(&last_ptr->refill_lock);
7747 trace_btrfs_reserve_extent_cluster(fs_info,
7748 block_group, search_start,
7752 } else if (!cached && loop > LOOP_CACHING_NOWAIT
7753 && !failed_cluster_refill) {
7754 spin_unlock(&last_ptr->refill_lock);
7756 failed_cluster_refill = true;
7757 wait_block_group_cache_progress(block_group,
7758 num_bytes + empty_cluster + empty_size);
7759 goto have_block_group;
7763 * at this point we either didn't find a cluster
7764 * or we weren't able to allocate a block from our
7765 * cluster. Free the cluster we've been trying
7766 * to use, and go to the next block group
7768 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7769 spin_unlock(&last_ptr->refill_lock);
7775 * We are doing an unclustered alloc, set the fragmented flag so
7776 * we don't bother trying to setup a cluster again until we get
7779 if (unlikely(last_ptr)) {
7780 spin_lock(&last_ptr->lock);
7781 last_ptr->fragmented = 1;
7782 spin_unlock(&last_ptr->lock);
7785 struct btrfs_free_space_ctl *ctl =
7786 block_group->free_space_ctl;
7788 spin_lock(&ctl->tree_lock);
7789 if (ctl->free_space <
7790 num_bytes + empty_cluster + empty_size) {
7791 if (ctl->free_space > max_extent_size)
7792 max_extent_size = ctl->free_space;
7793 spin_unlock(&ctl->tree_lock);
7796 spin_unlock(&ctl->tree_lock);
7799 offset = btrfs_find_space_for_alloc(block_group, search_start,
7800 num_bytes, empty_size,
7803 * If we didn't find a chunk, and we haven't failed on this
7804 * block group before, and this block group is in the middle of
7805 * caching and we are ok with waiting, then go ahead and wait
7806 * for progress to be made, and set failed_alloc to true.
7808 * If failed_alloc is true then we've already waited on this
7809 * block group once and should move on to the next block group.
7811 if (!offset && !failed_alloc && !cached &&
7812 loop > LOOP_CACHING_NOWAIT) {
7813 wait_block_group_cache_progress(block_group,
7814 num_bytes + empty_size);
7815 failed_alloc = true;
7816 goto have_block_group;
7817 } else if (!offset) {
7821 search_start = ALIGN(offset, fs_info->stripesize);
7823 /* move on to the next group */
7824 if (search_start + num_bytes >
7825 block_group->key.objectid + block_group->key.offset) {
7826 btrfs_add_free_space(block_group, offset, num_bytes);
7830 if (offset < search_start)
7831 btrfs_add_free_space(block_group, offset,
7832 search_start - offset);
7833 BUG_ON(offset > search_start);
7835 ret = btrfs_add_reserved_bytes(block_group, ram_bytes,
7836 num_bytes, delalloc);
7837 if (ret == -EAGAIN) {
7838 btrfs_add_free_space(block_group, offset, num_bytes);
7841 btrfs_inc_block_group_reservations(block_group);
7843 /* we are all good, lets return */
7844 ins->objectid = search_start;
7845 ins->offset = num_bytes;
7847 trace_btrfs_reserve_extent(fs_info, block_group,
7848 search_start, num_bytes);
7849 btrfs_release_block_group(block_group, delalloc);
7852 failed_cluster_refill = false;
7853 failed_alloc = false;
7854 BUG_ON(index != get_block_group_index(block_group));
7855 btrfs_release_block_group(block_group, delalloc);
7858 up_read(&space_info->groups_sem);
7860 if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg
7861 && !orig_have_caching_bg)
7862 orig_have_caching_bg = true;
7864 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7867 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7871 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7872 * caching kthreads as we move along
7873 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7874 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7875 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7878 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7880 if (loop == LOOP_CACHING_NOWAIT) {
7882 * We want to skip the LOOP_CACHING_WAIT step if we
7883 * don't have any uncached bgs and we've already done a
7884 * full search through.
7886 if (orig_have_caching_bg || !full_search)
7887 loop = LOOP_CACHING_WAIT;
7889 loop = LOOP_ALLOC_CHUNK;
7894 if (loop == LOOP_ALLOC_CHUNK) {
7895 struct btrfs_trans_handle *trans;
7898 trans = current->journal_info;
7902 trans = btrfs_join_transaction(root);
7904 if (IS_ERR(trans)) {
7905 ret = PTR_ERR(trans);
7909 ret = do_chunk_alloc(trans, fs_info, flags,
7913 * If we can't allocate a new chunk we've already looped
7914 * through at least once, move on to the NO_EMPTY_SIZE
7918 loop = LOOP_NO_EMPTY_SIZE;
7921 * Do not bail out on ENOSPC since we
7922 * can do more things.
7924 if (ret < 0 && ret != -ENOSPC)
7925 btrfs_abort_transaction(trans, ret);
7929 btrfs_end_transaction(trans);
7934 if (loop == LOOP_NO_EMPTY_SIZE) {
7936 * Don't loop again if we already have no empty_size and
7939 if (empty_size == 0 &&
7940 empty_cluster == 0) {
7949 } else if (!ins->objectid) {
7951 } else if (ins->objectid) {
7952 if (!use_cluster && last_ptr) {
7953 spin_lock(&last_ptr->lock);
7954 last_ptr->window_start = ins->objectid;
7955 spin_unlock(&last_ptr->lock);
7960 if (ret == -ENOSPC) {
7961 spin_lock(&space_info->lock);
7962 space_info->max_extent_size = max_extent_size;
7963 spin_unlock(&space_info->lock);
7964 ins->offset = max_extent_size;
7969 static void dump_space_info(struct btrfs_fs_info *fs_info,
7970 struct btrfs_space_info *info, u64 bytes,
7971 int dump_block_groups)
7973 struct btrfs_block_group_cache *cache;
7976 spin_lock(&info->lock);
7977 btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull",
7979 info->total_bytes - btrfs_space_info_used(info, true),
7980 info->full ? "" : "not ");
7982 "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu",
7983 info->total_bytes, info->bytes_used, info->bytes_pinned,
7984 info->bytes_reserved, info->bytes_may_use,
7985 info->bytes_readonly);
7986 spin_unlock(&info->lock);
7988 if (!dump_block_groups)
7991 down_read(&info->groups_sem);
7993 list_for_each_entry(cache, &info->block_groups[index], list) {
7994 spin_lock(&cache->lock);
7996 "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s",
7997 cache->key.objectid, cache->key.offset,
7998 btrfs_block_group_used(&cache->item), cache->pinned,
7999 cache->reserved, cache->ro ? "[readonly]" : "");
8000 btrfs_dump_free_space(cache, bytes);
8001 spin_unlock(&cache->lock);
8003 if (++index < BTRFS_NR_RAID_TYPES)
8005 up_read(&info->groups_sem);
8008 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
8009 u64 num_bytes, u64 min_alloc_size,
8010 u64 empty_size, u64 hint_byte,
8011 struct btrfs_key *ins, int is_data, int delalloc)
8013 struct btrfs_fs_info *fs_info = root->fs_info;
8014 bool final_tried = num_bytes == min_alloc_size;
8018 flags = get_alloc_profile_by_root(root, is_data);
8020 WARN_ON(num_bytes < fs_info->sectorsize);
8021 ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size,
8022 hint_byte, ins, flags, delalloc);
8023 if (!ret && !is_data) {
8024 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
8025 } else if (ret == -ENOSPC) {
8026 if (!final_tried && ins->offset) {
8027 num_bytes = min(num_bytes >> 1, ins->offset);
8028 num_bytes = round_down(num_bytes,
8029 fs_info->sectorsize);
8030 num_bytes = max(num_bytes, min_alloc_size);
8031 ram_bytes = num_bytes;
8032 if (num_bytes == min_alloc_size)
8035 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8036 struct btrfs_space_info *sinfo;
8038 sinfo = __find_space_info(fs_info, flags);
8040 "allocation failed flags %llu, wanted %llu",
8043 dump_space_info(fs_info, sinfo, num_bytes, 1);
8050 static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8052 int pin, int delalloc)
8054 struct btrfs_block_group_cache *cache;
8057 cache = btrfs_lookup_block_group(fs_info, start);
8059 btrfs_err(fs_info, "Unable to find block group for %llu",
8065 pin_down_extent(fs_info, cache, start, len, 1);
8067 if (btrfs_test_opt(fs_info, DISCARD))
8068 ret = btrfs_discard_extent(fs_info, start, len, NULL);
8069 btrfs_add_free_space(cache, start, len);
8070 btrfs_free_reserved_bytes(cache, len, delalloc);
8071 trace_btrfs_reserved_extent_free(fs_info, start, len);
8074 btrfs_put_block_group(cache);
8078 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8079 u64 start, u64 len, int delalloc)
8081 return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc);
8084 int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info,
8087 return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0);
8090 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8091 struct btrfs_fs_info *fs_info,
8092 u64 parent, u64 root_objectid,
8093 u64 flags, u64 owner, u64 offset,
8094 struct btrfs_key *ins, int ref_mod)
8097 struct btrfs_extent_item *extent_item;
8098 struct btrfs_extent_inline_ref *iref;
8099 struct btrfs_path *path;
8100 struct extent_buffer *leaf;
8105 type = BTRFS_SHARED_DATA_REF_KEY;
8107 type = BTRFS_EXTENT_DATA_REF_KEY;
8109 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
8111 path = btrfs_alloc_path();
8115 path->leave_spinning = 1;
8116 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8119 btrfs_free_path(path);
8123 leaf = path->nodes[0];
8124 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8125 struct btrfs_extent_item);
8126 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
8127 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8128 btrfs_set_extent_flags(leaf, extent_item,
8129 flags | BTRFS_EXTENT_FLAG_DATA);
8131 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8132 btrfs_set_extent_inline_ref_type(leaf, iref, type);
8134 struct btrfs_shared_data_ref *ref;
8135 ref = (struct btrfs_shared_data_ref *)(iref + 1);
8136 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8137 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
8139 struct btrfs_extent_data_ref *ref;
8140 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
8141 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
8142 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
8143 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
8144 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
8147 btrfs_mark_buffer_dirty(path->nodes[0]);
8148 btrfs_free_path(path);
8150 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
8155 ret = update_block_group(trans, fs_info, ins->objectid, ins->offset, 1);
8156 if (ret) { /* -ENOENT, logic error */
8157 btrfs_err(fs_info, "update block group failed for %llu %llu",
8158 ins->objectid, ins->offset);
8161 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset);
8165 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
8166 struct btrfs_fs_info *fs_info,
8167 u64 parent, u64 root_objectid,
8168 u64 flags, struct btrfs_disk_key *key,
8169 int level, struct btrfs_key *ins)
8172 struct btrfs_extent_item *extent_item;
8173 struct btrfs_tree_block_info *block_info;
8174 struct btrfs_extent_inline_ref *iref;
8175 struct btrfs_path *path;
8176 struct extent_buffer *leaf;
8177 u32 size = sizeof(*extent_item) + sizeof(*iref);
8178 u64 num_bytes = ins->offset;
8179 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8181 if (!skinny_metadata)
8182 size += sizeof(*block_info);
8184 path = btrfs_alloc_path();
8186 btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid,
8191 path->leave_spinning = 1;
8192 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8195 btrfs_free_path(path);
8196 btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid,
8201 leaf = path->nodes[0];
8202 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8203 struct btrfs_extent_item);
8204 btrfs_set_extent_refs(leaf, extent_item, 1);
8205 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8206 btrfs_set_extent_flags(leaf, extent_item,
8207 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
8209 if (skinny_metadata) {
8210 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8211 num_bytes = fs_info->nodesize;
8213 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
8214 btrfs_set_tree_block_key(leaf, block_info, key);
8215 btrfs_set_tree_block_level(leaf, block_info, level);
8216 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
8220 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
8221 btrfs_set_extent_inline_ref_type(leaf, iref,
8222 BTRFS_SHARED_BLOCK_REF_KEY);
8223 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8225 btrfs_set_extent_inline_ref_type(leaf, iref,
8226 BTRFS_TREE_BLOCK_REF_KEY);
8227 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
8230 btrfs_mark_buffer_dirty(leaf);
8231 btrfs_free_path(path);
8233 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
8238 ret = update_block_group(trans, fs_info, ins->objectid,
8239 fs_info->nodesize, 1);
8240 if (ret) { /* -ENOENT, logic error */
8241 btrfs_err(fs_info, "update block group failed for %llu %llu",
8242 ins->objectid, ins->offset);
8246 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid,
8251 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8252 struct btrfs_root *root, u64 owner,
8253 u64 offset, u64 ram_bytes,
8254 struct btrfs_key *ins)
8256 struct btrfs_fs_info *fs_info = root->fs_info;
8259 BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
8261 btrfs_ref_tree_mod(root, ins->objectid, ins->offset, 0,
8262 root->root_key.objectid, owner, offset,
8263 BTRFS_ADD_DELAYED_EXTENT);
8265 ret = btrfs_add_delayed_data_ref(fs_info, trans, ins->objectid,
8267 root->root_key.objectid, owner,
8269 BTRFS_ADD_DELAYED_EXTENT, NULL, NULL);
8274 * this is used by the tree logging recovery code. It records that
8275 * an extent has been allocated and makes sure to clear the free
8276 * space cache bits as well
8278 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
8279 struct btrfs_fs_info *fs_info,
8280 u64 root_objectid, u64 owner, u64 offset,
8281 struct btrfs_key *ins)
8284 struct btrfs_block_group_cache *block_group;
8285 struct btrfs_space_info *space_info;
8288 * Mixed block groups will exclude before processing the log so we only
8289 * need to do the exclude dance if this fs isn't mixed.
8291 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
8292 ret = __exclude_logged_extent(fs_info, ins->objectid,
8298 block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
8302 space_info = block_group->space_info;
8303 spin_lock(&space_info->lock);
8304 spin_lock(&block_group->lock);
8305 space_info->bytes_reserved += ins->offset;
8306 block_group->reserved += ins->offset;
8307 spin_unlock(&block_group->lock);
8308 spin_unlock(&space_info->lock);
8310 ret = alloc_reserved_file_extent(trans, fs_info, 0, root_objectid,
8311 0, owner, offset, ins, 1);
8312 btrfs_put_block_group(block_group);
8316 static struct extent_buffer *
8317 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
8318 u64 bytenr, int level)
8320 struct btrfs_fs_info *fs_info = root->fs_info;
8321 struct extent_buffer *buf;
8323 buf = btrfs_find_create_tree_block(fs_info, bytenr);
8327 btrfs_set_header_generation(buf, trans->transid);
8328 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
8329 btrfs_tree_lock(buf);
8330 clean_tree_block(fs_info, buf);
8331 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
8333 btrfs_set_lock_blocking(buf);
8334 set_extent_buffer_uptodate(buf);
8336 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
8337 buf->log_index = root->log_transid % 2;
8339 * we allow two log transactions at a time, use different
8340 * EXENT bit to differentiate dirty pages.
8342 if (buf->log_index == 0)
8343 set_extent_dirty(&root->dirty_log_pages, buf->start,
8344 buf->start + buf->len - 1, GFP_NOFS);
8346 set_extent_new(&root->dirty_log_pages, buf->start,
8347 buf->start + buf->len - 1);
8349 buf->log_index = -1;
8350 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
8351 buf->start + buf->len - 1, GFP_NOFS);
8353 trans->dirty = true;
8354 /* this returns a buffer locked for blocking */
8358 static struct btrfs_block_rsv *
8359 use_block_rsv(struct btrfs_trans_handle *trans,
8360 struct btrfs_root *root, u32 blocksize)
8362 struct btrfs_fs_info *fs_info = root->fs_info;
8363 struct btrfs_block_rsv *block_rsv;
8364 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
8366 bool global_updated = false;
8368 block_rsv = get_block_rsv(trans, root);
8370 if (unlikely(block_rsv->size == 0))
8373 ret = block_rsv_use_bytes(block_rsv, blocksize);
8377 if (block_rsv->failfast)
8378 return ERR_PTR(ret);
8380 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
8381 global_updated = true;
8382 update_global_block_rsv(fs_info);
8386 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8387 static DEFINE_RATELIMIT_STATE(_rs,
8388 DEFAULT_RATELIMIT_INTERVAL * 10,
8389 /*DEFAULT_RATELIMIT_BURST*/ 1);
8390 if (__ratelimit(&_rs))
8392 "BTRFS: block rsv returned %d\n", ret);
8395 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
8396 BTRFS_RESERVE_NO_FLUSH);
8400 * If we couldn't reserve metadata bytes try and use some from
8401 * the global reserve if its space type is the same as the global
8404 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
8405 block_rsv->space_info == global_rsv->space_info) {
8406 ret = block_rsv_use_bytes(global_rsv, blocksize);
8410 return ERR_PTR(ret);
8413 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
8414 struct btrfs_block_rsv *block_rsv, u32 blocksize)
8416 block_rsv_add_bytes(block_rsv, blocksize, 0);
8417 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
8421 * finds a free extent and does all the dirty work required for allocation
8422 * returns the tree buffer or an ERR_PTR on error.
8424 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
8425 struct btrfs_root *root,
8426 u64 parent, u64 root_objectid,
8427 const struct btrfs_disk_key *key,
8428 int level, u64 hint,
8431 struct btrfs_fs_info *fs_info = root->fs_info;
8432 struct btrfs_key ins;
8433 struct btrfs_block_rsv *block_rsv;
8434 struct extent_buffer *buf;
8435 struct btrfs_delayed_extent_op *extent_op;
8438 u32 blocksize = fs_info->nodesize;
8439 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8441 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8442 if (btrfs_is_testing(fs_info)) {
8443 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
8446 root->alloc_bytenr += blocksize;
8451 block_rsv = use_block_rsv(trans, root, blocksize);
8452 if (IS_ERR(block_rsv))
8453 return ERR_CAST(block_rsv);
8455 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
8456 empty_size, hint, &ins, 0, 0);
8460 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
8463 goto out_free_reserved;
8466 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
8468 parent = ins.objectid;
8469 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
8473 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
8474 extent_op = btrfs_alloc_delayed_extent_op();
8480 memcpy(&extent_op->key, key, sizeof(extent_op->key));
8482 memset(&extent_op->key, 0, sizeof(extent_op->key));
8483 extent_op->flags_to_set = flags;
8484 extent_op->update_key = skinny_metadata ? false : true;
8485 extent_op->update_flags = true;
8486 extent_op->is_data = false;
8487 extent_op->level = level;
8489 btrfs_ref_tree_mod(root, ins.objectid, ins.offset, parent,
8490 root_objectid, level, 0,
8491 BTRFS_ADD_DELAYED_EXTENT);
8492 ret = btrfs_add_delayed_tree_ref(fs_info, trans, ins.objectid,
8494 root_objectid, level,
8495 BTRFS_ADD_DELAYED_EXTENT,
8496 extent_op, NULL, NULL);
8498 goto out_free_delayed;
8503 btrfs_free_delayed_extent_op(extent_op);
8505 free_extent_buffer(buf);
8507 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
8509 unuse_block_rsv(fs_info, block_rsv, blocksize);
8510 return ERR_PTR(ret);
8513 struct walk_control {
8514 u64 refs[BTRFS_MAX_LEVEL];
8515 u64 flags[BTRFS_MAX_LEVEL];
8516 struct btrfs_key update_progress;
8527 #define DROP_REFERENCE 1
8528 #define UPDATE_BACKREF 2
8530 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
8531 struct btrfs_root *root,
8532 struct walk_control *wc,
8533 struct btrfs_path *path)
8535 struct btrfs_fs_info *fs_info = root->fs_info;
8541 struct btrfs_key key;
8542 struct extent_buffer *eb;
8547 if (path->slots[wc->level] < wc->reada_slot) {
8548 wc->reada_count = wc->reada_count * 2 / 3;
8549 wc->reada_count = max(wc->reada_count, 2);
8551 wc->reada_count = wc->reada_count * 3 / 2;
8552 wc->reada_count = min_t(int, wc->reada_count,
8553 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
8556 eb = path->nodes[wc->level];
8557 nritems = btrfs_header_nritems(eb);
8559 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
8560 if (nread >= wc->reada_count)
8564 bytenr = btrfs_node_blockptr(eb, slot);
8565 generation = btrfs_node_ptr_generation(eb, slot);
8567 if (slot == path->slots[wc->level])
8570 if (wc->stage == UPDATE_BACKREF &&
8571 generation <= root->root_key.offset)
8574 /* We don't lock the tree block, it's OK to be racy here */
8575 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
8576 wc->level - 1, 1, &refs,
8578 /* We don't care about errors in readahead. */
8583 if (wc->stage == DROP_REFERENCE) {
8587 if (wc->level == 1 &&
8588 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8590 if (!wc->update_ref ||
8591 generation <= root->root_key.offset)
8593 btrfs_node_key_to_cpu(eb, &key, slot);
8594 ret = btrfs_comp_cpu_keys(&key,
8595 &wc->update_progress);
8599 if (wc->level == 1 &&
8600 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8604 readahead_tree_block(fs_info, bytenr);
8607 wc->reada_slot = slot;
8611 * helper to process tree block while walking down the tree.
8613 * when wc->stage == UPDATE_BACKREF, this function updates
8614 * back refs for pointers in the block.
8616 * NOTE: return value 1 means we should stop walking down.
8618 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8619 struct btrfs_root *root,
8620 struct btrfs_path *path,
8621 struct walk_control *wc, int lookup_info)
8623 struct btrfs_fs_info *fs_info = root->fs_info;
8624 int level = wc->level;
8625 struct extent_buffer *eb = path->nodes[level];
8626 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8629 if (wc->stage == UPDATE_BACKREF &&
8630 btrfs_header_owner(eb) != root->root_key.objectid)
8634 * when reference count of tree block is 1, it won't increase
8635 * again. once full backref flag is set, we never clear it.
8638 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8639 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8640 BUG_ON(!path->locks[level]);
8641 ret = btrfs_lookup_extent_info(trans, fs_info,
8642 eb->start, level, 1,
8645 BUG_ON(ret == -ENOMEM);
8648 BUG_ON(wc->refs[level] == 0);
8651 if (wc->stage == DROP_REFERENCE) {
8652 if (wc->refs[level] > 1)
8655 if (path->locks[level] && !wc->keep_locks) {
8656 btrfs_tree_unlock_rw(eb, path->locks[level]);
8657 path->locks[level] = 0;
8662 /* wc->stage == UPDATE_BACKREF */
8663 if (!(wc->flags[level] & flag)) {
8664 BUG_ON(!path->locks[level]);
8665 ret = btrfs_inc_ref(trans, root, eb, 1);
8666 BUG_ON(ret); /* -ENOMEM */
8667 ret = btrfs_dec_ref(trans, root, eb, 0);
8668 BUG_ON(ret); /* -ENOMEM */
8669 ret = btrfs_set_disk_extent_flags(trans, fs_info, eb->start,
8671 btrfs_header_level(eb), 0);
8672 BUG_ON(ret); /* -ENOMEM */
8673 wc->flags[level] |= flag;
8677 * the block is shared by multiple trees, so it's not good to
8678 * keep the tree lock
8680 if (path->locks[level] && level > 0) {
8681 btrfs_tree_unlock_rw(eb, path->locks[level]);
8682 path->locks[level] = 0;
8688 * helper to process tree block pointer.
8690 * when wc->stage == DROP_REFERENCE, this function checks
8691 * reference count of the block pointed to. if the block
8692 * is shared and we need update back refs for the subtree
8693 * rooted at the block, this function changes wc->stage to
8694 * UPDATE_BACKREF. if the block is shared and there is no
8695 * need to update back, this function drops the reference
8698 * NOTE: return value 1 means we should stop walking down.
8700 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8701 struct btrfs_root *root,
8702 struct btrfs_path *path,
8703 struct walk_control *wc, int *lookup_info)
8705 struct btrfs_fs_info *fs_info = root->fs_info;
8710 struct btrfs_key key;
8711 struct extent_buffer *next;
8712 int level = wc->level;
8715 bool need_account = false;
8717 generation = btrfs_node_ptr_generation(path->nodes[level],
8718 path->slots[level]);
8720 * if the lower level block was created before the snapshot
8721 * was created, we know there is no need to update back refs
8724 if (wc->stage == UPDATE_BACKREF &&
8725 generation <= root->root_key.offset) {
8730 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8731 blocksize = fs_info->nodesize;
8733 next = find_extent_buffer(fs_info, bytenr);
8735 next = btrfs_find_create_tree_block(fs_info, bytenr);
8737 return PTR_ERR(next);
8739 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8743 btrfs_tree_lock(next);
8744 btrfs_set_lock_blocking(next);
8746 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
8747 &wc->refs[level - 1],
8748 &wc->flags[level - 1]);
8752 if (unlikely(wc->refs[level - 1] == 0)) {
8753 btrfs_err(fs_info, "Missing references.");
8759 if (wc->stage == DROP_REFERENCE) {
8760 if (wc->refs[level - 1] > 1) {
8761 need_account = true;
8763 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8766 if (!wc->update_ref ||
8767 generation <= root->root_key.offset)
8770 btrfs_node_key_to_cpu(path->nodes[level], &key,
8771 path->slots[level]);
8772 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8776 wc->stage = UPDATE_BACKREF;
8777 wc->shared_level = level - 1;
8781 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8785 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8786 btrfs_tree_unlock(next);
8787 free_extent_buffer(next);
8793 if (reada && level == 1)
8794 reada_walk_down(trans, root, wc, path);
8795 next = read_tree_block(fs_info, bytenr, generation);
8797 return PTR_ERR(next);
8798 } else if (!extent_buffer_uptodate(next)) {
8799 free_extent_buffer(next);
8802 btrfs_tree_lock(next);
8803 btrfs_set_lock_blocking(next);
8807 ASSERT(level == btrfs_header_level(next));
8808 if (level != btrfs_header_level(next)) {
8809 btrfs_err(root->fs_info, "mismatched level");
8813 path->nodes[level] = next;
8814 path->slots[level] = 0;
8815 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8821 wc->refs[level - 1] = 0;
8822 wc->flags[level - 1] = 0;
8823 if (wc->stage == DROP_REFERENCE) {
8824 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8825 parent = path->nodes[level]->start;
8827 ASSERT(root->root_key.objectid ==
8828 btrfs_header_owner(path->nodes[level]));
8829 if (root->root_key.objectid !=
8830 btrfs_header_owner(path->nodes[level])) {
8831 btrfs_err(root->fs_info,
8832 "mismatched block owner");
8840 ret = btrfs_qgroup_trace_subtree(trans, root, next,
8841 generation, level - 1);
8843 btrfs_err_rl(fs_info,
8844 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
8848 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
8849 parent, root->root_key.objectid,
8859 btrfs_tree_unlock(next);
8860 free_extent_buffer(next);
8866 * helper to process tree block while walking up the tree.
8868 * when wc->stage == DROP_REFERENCE, this function drops
8869 * reference count on the block.
8871 * when wc->stage == UPDATE_BACKREF, this function changes
8872 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8873 * to UPDATE_BACKREF previously while processing the block.
8875 * NOTE: return value 1 means we should stop walking up.
8877 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8878 struct btrfs_root *root,
8879 struct btrfs_path *path,
8880 struct walk_control *wc)
8882 struct btrfs_fs_info *fs_info = root->fs_info;
8884 int level = wc->level;
8885 struct extent_buffer *eb = path->nodes[level];
8888 if (wc->stage == UPDATE_BACKREF) {
8889 BUG_ON(wc->shared_level < level);
8890 if (level < wc->shared_level)
8893 ret = find_next_key(path, level + 1, &wc->update_progress);
8897 wc->stage = DROP_REFERENCE;
8898 wc->shared_level = -1;
8899 path->slots[level] = 0;
8902 * check reference count again if the block isn't locked.
8903 * we should start walking down the tree again if reference
8906 if (!path->locks[level]) {
8908 btrfs_tree_lock(eb);
8909 btrfs_set_lock_blocking(eb);
8910 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8912 ret = btrfs_lookup_extent_info(trans, fs_info,
8913 eb->start, level, 1,
8917 btrfs_tree_unlock_rw(eb, path->locks[level]);
8918 path->locks[level] = 0;
8921 BUG_ON(wc->refs[level] == 0);
8922 if (wc->refs[level] == 1) {
8923 btrfs_tree_unlock_rw(eb, path->locks[level]);
8924 path->locks[level] = 0;
8930 /* wc->stage == DROP_REFERENCE */
8931 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8933 if (wc->refs[level] == 1) {
8935 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8936 ret = btrfs_dec_ref(trans, root, eb, 1);
8938 ret = btrfs_dec_ref(trans, root, eb, 0);
8939 BUG_ON(ret); /* -ENOMEM */
8940 ret = btrfs_qgroup_trace_leaf_items(trans, fs_info, eb);
8942 btrfs_err_rl(fs_info,
8943 "error %d accounting leaf items. Quota is out of sync, rescan required.",
8947 /* make block locked assertion in clean_tree_block happy */
8948 if (!path->locks[level] &&
8949 btrfs_header_generation(eb) == trans->transid) {
8950 btrfs_tree_lock(eb);
8951 btrfs_set_lock_blocking(eb);
8952 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8954 clean_tree_block(fs_info, eb);
8957 if (eb == root->node) {
8958 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8961 BUG_ON(root->root_key.objectid !=
8962 btrfs_header_owner(eb));
8964 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8965 parent = path->nodes[level + 1]->start;
8967 BUG_ON(root->root_key.objectid !=
8968 btrfs_header_owner(path->nodes[level + 1]));
8971 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8973 wc->refs[level] = 0;
8974 wc->flags[level] = 0;
8978 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8979 struct btrfs_root *root,
8980 struct btrfs_path *path,
8981 struct walk_control *wc)
8983 int level = wc->level;
8984 int lookup_info = 1;
8987 while (level >= 0) {
8988 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8995 if (path->slots[level] >=
8996 btrfs_header_nritems(path->nodes[level]))
8999 ret = do_walk_down(trans, root, path, wc, &lookup_info);
9001 path->slots[level]++;
9010 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
9011 struct btrfs_root *root,
9012 struct btrfs_path *path,
9013 struct walk_control *wc, int max_level)
9015 int level = wc->level;
9018 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
9019 while (level < max_level && path->nodes[level]) {
9021 if (path->slots[level] + 1 <
9022 btrfs_header_nritems(path->nodes[level])) {
9023 path->slots[level]++;
9026 ret = walk_up_proc(trans, root, path, wc);
9030 if (path->locks[level]) {
9031 btrfs_tree_unlock_rw(path->nodes[level],
9032 path->locks[level]);
9033 path->locks[level] = 0;
9035 free_extent_buffer(path->nodes[level]);
9036 path->nodes[level] = NULL;
9044 * drop a subvolume tree.
9046 * this function traverses the tree freeing any blocks that only
9047 * referenced by the tree.
9049 * when a shared tree block is found. this function decreases its
9050 * reference count by one. if update_ref is true, this function
9051 * also make sure backrefs for the shared block and all lower level
9052 * blocks are properly updated.
9054 * If called with for_reloc == 0, may exit early with -EAGAIN
9056 int btrfs_drop_snapshot(struct btrfs_root *root,
9057 struct btrfs_block_rsv *block_rsv, int update_ref,
9060 struct btrfs_fs_info *fs_info = root->fs_info;
9061 struct btrfs_path *path;
9062 struct btrfs_trans_handle *trans;
9063 struct btrfs_root *tree_root = fs_info->tree_root;
9064 struct btrfs_root_item *root_item = &root->root_item;
9065 struct walk_control *wc;
9066 struct btrfs_key key;
9070 bool root_dropped = false;
9072 btrfs_debug(fs_info, "Drop subvolume %llu", root->objectid);
9074 path = btrfs_alloc_path();
9080 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9082 btrfs_free_path(path);
9087 trans = btrfs_start_transaction(tree_root, 0);
9088 if (IS_ERR(trans)) {
9089 err = PTR_ERR(trans);
9094 trans->block_rsv = block_rsv;
9096 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
9097 level = btrfs_header_level(root->node);
9098 path->nodes[level] = btrfs_lock_root_node(root);
9099 btrfs_set_lock_blocking(path->nodes[level]);
9100 path->slots[level] = 0;
9101 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9102 memset(&wc->update_progress, 0,
9103 sizeof(wc->update_progress));
9105 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
9106 memcpy(&wc->update_progress, &key,
9107 sizeof(wc->update_progress));
9109 level = root_item->drop_level;
9111 path->lowest_level = level;
9112 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
9113 path->lowest_level = 0;
9121 * unlock our path, this is safe because only this
9122 * function is allowed to delete this snapshot
9124 btrfs_unlock_up_safe(path, 0);
9126 level = btrfs_header_level(root->node);
9128 btrfs_tree_lock(path->nodes[level]);
9129 btrfs_set_lock_blocking(path->nodes[level]);
9130 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9132 ret = btrfs_lookup_extent_info(trans, fs_info,
9133 path->nodes[level]->start,
9134 level, 1, &wc->refs[level],
9140 BUG_ON(wc->refs[level] == 0);
9142 if (level == root_item->drop_level)
9145 btrfs_tree_unlock(path->nodes[level]);
9146 path->locks[level] = 0;
9147 WARN_ON(wc->refs[level] != 1);
9153 wc->shared_level = -1;
9154 wc->stage = DROP_REFERENCE;
9155 wc->update_ref = update_ref;
9157 wc->for_reloc = for_reloc;
9158 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9162 ret = walk_down_tree(trans, root, path, wc);
9168 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
9175 BUG_ON(wc->stage != DROP_REFERENCE);
9179 if (wc->stage == DROP_REFERENCE) {
9181 btrfs_node_key(path->nodes[level],
9182 &root_item->drop_progress,
9183 path->slots[level]);
9184 root_item->drop_level = level;
9187 BUG_ON(wc->level == 0);
9188 if (btrfs_should_end_transaction(trans) ||
9189 (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
9190 ret = btrfs_update_root(trans, tree_root,
9194 btrfs_abort_transaction(trans, ret);
9199 btrfs_end_transaction_throttle(trans);
9200 if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
9201 btrfs_debug(fs_info,
9202 "drop snapshot early exit");
9207 trans = btrfs_start_transaction(tree_root, 0);
9208 if (IS_ERR(trans)) {
9209 err = PTR_ERR(trans);
9213 trans->block_rsv = block_rsv;
9216 btrfs_release_path(path);
9220 ret = btrfs_del_root(trans, fs_info, &root->root_key);
9222 btrfs_abort_transaction(trans, ret);
9226 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
9227 ret = btrfs_find_root(tree_root, &root->root_key, path,
9230 btrfs_abort_transaction(trans, ret);
9233 } else if (ret > 0) {
9234 /* if we fail to delete the orphan item this time
9235 * around, it'll get picked up the next time.
9237 * The most common failure here is just -ENOENT.
9239 btrfs_del_orphan_item(trans, tree_root,
9240 root->root_key.objectid);
9244 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
9245 btrfs_add_dropped_root(trans, root);
9247 free_extent_buffer(root->node);
9248 free_extent_buffer(root->commit_root);
9249 btrfs_put_fs_root(root);
9251 root_dropped = true;
9253 btrfs_end_transaction_throttle(trans);
9256 btrfs_free_path(path);
9259 * So if we need to stop dropping the snapshot for whatever reason we
9260 * need to make sure to add it back to the dead root list so that we
9261 * keep trying to do the work later. This also cleans up roots if we
9262 * don't have it in the radix (like when we recover after a power fail
9263 * or unmount) so we don't leak memory.
9265 if (!for_reloc && !root_dropped)
9266 btrfs_add_dead_root(root);
9267 if (err && err != -EAGAIN)
9268 btrfs_handle_fs_error(fs_info, err, NULL);
9273 * drop subtree rooted at tree block 'node'.
9275 * NOTE: this function will unlock and release tree block 'node'
9276 * only used by relocation code
9278 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
9279 struct btrfs_root *root,
9280 struct extent_buffer *node,
9281 struct extent_buffer *parent)
9283 struct btrfs_fs_info *fs_info = root->fs_info;
9284 struct btrfs_path *path;
9285 struct walk_control *wc;
9291 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
9293 path = btrfs_alloc_path();
9297 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9299 btrfs_free_path(path);
9303 btrfs_assert_tree_locked(parent);
9304 parent_level = btrfs_header_level(parent);
9305 extent_buffer_get(parent);
9306 path->nodes[parent_level] = parent;
9307 path->slots[parent_level] = btrfs_header_nritems(parent);
9309 btrfs_assert_tree_locked(node);
9310 level = btrfs_header_level(node);
9311 path->nodes[level] = node;
9312 path->slots[level] = 0;
9313 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9315 wc->refs[parent_level] = 1;
9316 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
9318 wc->shared_level = -1;
9319 wc->stage = DROP_REFERENCE;
9323 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9326 wret = walk_down_tree(trans, root, path, wc);
9332 wret = walk_up_tree(trans, root, path, wc, parent_level);
9340 btrfs_free_path(path);
9344 static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags)
9350 * if restripe for this chunk_type is on pick target profile and
9351 * return, otherwise do the usual balance
9353 stripped = get_restripe_target(fs_info, flags);
9355 return extended_to_chunk(stripped);
9357 num_devices = fs_info->fs_devices->rw_devices;
9359 stripped = BTRFS_BLOCK_GROUP_RAID0 |
9360 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
9361 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
9363 if (num_devices == 1) {
9364 stripped |= BTRFS_BLOCK_GROUP_DUP;
9365 stripped = flags & ~stripped;
9367 /* turn raid0 into single device chunks */
9368 if (flags & BTRFS_BLOCK_GROUP_RAID0)
9371 /* turn mirroring into duplication */
9372 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
9373 BTRFS_BLOCK_GROUP_RAID10))
9374 return stripped | BTRFS_BLOCK_GROUP_DUP;
9376 /* they already had raid on here, just return */
9377 if (flags & stripped)
9380 stripped |= BTRFS_BLOCK_GROUP_DUP;
9381 stripped = flags & ~stripped;
9383 /* switch duplicated blocks with raid1 */
9384 if (flags & BTRFS_BLOCK_GROUP_DUP)
9385 return stripped | BTRFS_BLOCK_GROUP_RAID1;
9387 /* this is drive concat, leave it alone */
9393 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
9395 struct btrfs_space_info *sinfo = cache->space_info;
9397 u64 min_allocable_bytes;
9401 * We need some metadata space and system metadata space for
9402 * allocating chunks in some corner cases until we force to set
9403 * it to be readonly.
9406 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
9408 min_allocable_bytes = SZ_1M;
9410 min_allocable_bytes = 0;
9412 spin_lock(&sinfo->lock);
9413 spin_lock(&cache->lock);
9421 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
9422 cache->bytes_super - btrfs_block_group_used(&cache->item);
9424 if (btrfs_space_info_used(sinfo, true) + num_bytes +
9425 min_allocable_bytes <= sinfo->total_bytes) {
9426 sinfo->bytes_readonly += num_bytes;
9428 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
9432 spin_unlock(&cache->lock);
9433 spin_unlock(&sinfo->lock);
9437 int btrfs_inc_block_group_ro(struct btrfs_fs_info *fs_info,
9438 struct btrfs_block_group_cache *cache)
9441 struct btrfs_trans_handle *trans;
9446 trans = btrfs_join_transaction(fs_info->extent_root);
9448 return PTR_ERR(trans);
9451 * we're not allowed to set block groups readonly after the dirty
9452 * block groups cache has started writing. If it already started,
9453 * back off and let this transaction commit
9455 mutex_lock(&fs_info->ro_block_group_mutex);
9456 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9457 u64 transid = trans->transid;
9459 mutex_unlock(&fs_info->ro_block_group_mutex);
9460 btrfs_end_transaction(trans);
9462 ret = btrfs_wait_for_commit(fs_info, transid);
9469 * if we are changing raid levels, try to allocate a corresponding
9470 * block group with the new raid level.
9472 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9473 if (alloc_flags != cache->flags) {
9474 ret = do_chunk_alloc(trans, fs_info, alloc_flags,
9477 * ENOSPC is allowed here, we may have enough space
9478 * already allocated at the new raid level to
9487 ret = inc_block_group_ro(cache, 0);
9490 alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags);
9491 ret = do_chunk_alloc(trans, fs_info, alloc_flags,
9495 ret = inc_block_group_ro(cache, 0);
9497 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9498 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9499 mutex_lock(&fs_info->chunk_mutex);
9500 check_system_chunk(trans, fs_info, alloc_flags);
9501 mutex_unlock(&fs_info->chunk_mutex);
9503 mutex_unlock(&fs_info->ro_block_group_mutex);
9505 btrfs_end_transaction(trans);
9509 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
9510 struct btrfs_fs_info *fs_info, u64 type)
9512 u64 alloc_flags = get_alloc_profile(fs_info, type);
9514 return do_chunk_alloc(trans, fs_info, alloc_flags, CHUNK_ALLOC_FORCE);
9518 * helper to account the unused space of all the readonly block group in the
9519 * space_info. takes mirrors into account.
9521 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9523 struct btrfs_block_group_cache *block_group;
9527 /* It's df, we don't care if it's racy */
9528 if (list_empty(&sinfo->ro_bgs))
9531 spin_lock(&sinfo->lock);
9532 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9533 spin_lock(&block_group->lock);
9535 if (!block_group->ro) {
9536 spin_unlock(&block_group->lock);
9540 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
9541 BTRFS_BLOCK_GROUP_RAID10 |
9542 BTRFS_BLOCK_GROUP_DUP))
9547 free_bytes += (block_group->key.offset -
9548 btrfs_block_group_used(&block_group->item)) *
9551 spin_unlock(&block_group->lock);
9553 spin_unlock(&sinfo->lock);
9558 void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache)
9560 struct btrfs_space_info *sinfo = cache->space_info;
9565 spin_lock(&sinfo->lock);
9566 spin_lock(&cache->lock);
9568 num_bytes = cache->key.offset - cache->reserved -
9569 cache->pinned - cache->bytes_super -
9570 btrfs_block_group_used(&cache->item);
9571 sinfo->bytes_readonly -= num_bytes;
9572 list_del_init(&cache->ro_list);
9574 spin_unlock(&cache->lock);
9575 spin_unlock(&sinfo->lock);
9579 * checks to see if its even possible to relocate this block group.
9581 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9582 * ok to go ahead and try.
9584 int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr)
9586 struct btrfs_root *root = fs_info->extent_root;
9587 struct btrfs_block_group_cache *block_group;
9588 struct btrfs_space_info *space_info;
9589 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
9590 struct btrfs_device *device;
9591 struct btrfs_trans_handle *trans;
9601 debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG);
9603 block_group = btrfs_lookup_block_group(fs_info, bytenr);
9605 /* odd, couldn't find the block group, leave it alone */
9609 "can't find block group for bytenr %llu",
9614 min_free = btrfs_block_group_used(&block_group->item);
9616 /* no bytes used, we're good */
9620 space_info = block_group->space_info;
9621 spin_lock(&space_info->lock);
9623 full = space_info->full;
9626 * if this is the last block group we have in this space, we can't
9627 * relocate it unless we're able to allocate a new chunk below.
9629 * Otherwise, we need to make sure we have room in the space to handle
9630 * all of the extents from this block group. If we can, we're good
9632 if ((space_info->total_bytes != block_group->key.offset) &&
9633 (btrfs_space_info_used(space_info, false) + min_free <
9634 space_info->total_bytes)) {
9635 spin_unlock(&space_info->lock);
9638 spin_unlock(&space_info->lock);
9641 * ok we don't have enough space, but maybe we have free space on our
9642 * devices to allocate new chunks for relocation, so loop through our
9643 * alloc devices and guess if we have enough space. if this block
9644 * group is going to be restriped, run checks against the target
9645 * profile instead of the current one.
9657 target = get_restripe_target(fs_info, block_group->flags);
9659 index = __get_raid_index(extended_to_chunk(target));
9662 * this is just a balance, so if we were marked as full
9663 * we know there is no space for a new chunk
9668 "no space to alloc new chunk for block group %llu",
9669 block_group->key.objectid);
9673 index = get_block_group_index(block_group);
9676 if (index == BTRFS_RAID_RAID10) {
9680 } else if (index == BTRFS_RAID_RAID1) {
9682 } else if (index == BTRFS_RAID_DUP) {
9685 } else if (index == BTRFS_RAID_RAID0) {
9686 dev_min = fs_devices->rw_devices;
9687 min_free = div64_u64(min_free, dev_min);
9690 /* We need to do this so that we can look at pending chunks */
9691 trans = btrfs_join_transaction(root);
9692 if (IS_ERR(trans)) {
9693 ret = PTR_ERR(trans);
9697 mutex_lock(&fs_info->chunk_mutex);
9698 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9702 * check to make sure we can actually find a chunk with enough
9703 * space to fit our block group in.
9705 if (device->total_bytes > device->bytes_used + min_free &&
9706 !device->is_tgtdev_for_dev_replace) {
9707 ret = find_free_dev_extent(trans, device, min_free,
9712 if (dev_nr >= dev_min)
9718 if (debug && ret == -1)
9720 "no space to allocate a new chunk for block group %llu",
9721 block_group->key.objectid);
9722 mutex_unlock(&fs_info->chunk_mutex);
9723 btrfs_end_transaction(trans);
9725 btrfs_put_block_group(block_group);
9729 static int find_first_block_group(struct btrfs_fs_info *fs_info,
9730 struct btrfs_path *path,
9731 struct btrfs_key *key)
9733 struct btrfs_root *root = fs_info->extent_root;
9735 struct btrfs_key found_key;
9736 struct extent_buffer *leaf;
9739 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9744 slot = path->slots[0];
9745 leaf = path->nodes[0];
9746 if (slot >= btrfs_header_nritems(leaf)) {
9747 ret = btrfs_next_leaf(root, path);
9754 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9756 if (found_key.objectid >= key->objectid &&
9757 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9758 struct extent_map_tree *em_tree;
9759 struct extent_map *em;
9761 em_tree = &root->fs_info->mapping_tree.map_tree;
9762 read_lock(&em_tree->lock);
9763 em = lookup_extent_mapping(em_tree, found_key.objectid,
9765 read_unlock(&em_tree->lock);
9768 "logical %llu len %llu found bg but no related chunk",
9769 found_key.objectid, found_key.offset);
9774 free_extent_map(em);
9783 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9785 struct btrfs_block_group_cache *block_group;
9789 struct inode *inode;
9791 block_group = btrfs_lookup_first_block_group(info, last);
9792 while (block_group) {
9793 spin_lock(&block_group->lock);
9794 if (block_group->iref)
9796 spin_unlock(&block_group->lock);
9797 block_group = next_block_group(info, block_group);
9806 inode = block_group->inode;
9807 block_group->iref = 0;
9808 block_group->inode = NULL;
9809 spin_unlock(&block_group->lock);
9810 ASSERT(block_group->io_ctl.inode == NULL);
9812 last = block_group->key.objectid + block_group->key.offset;
9813 btrfs_put_block_group(block_group);
9818 * Must be called only after stopping all workers, since we could have block
9819 * group caching kthreads running, and therefore they could race with us if we
9820 * freed the block groups before stopping them.
9822 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9824 struct btrfs_block_group_cache *block_group;
9825 struct btrfs_space_info *space_info;
9826 struct btrfs_caching_control *caching_ctl;
9829 down_write(&info->commit_root_sem);
9830 while (!list_empty(&info->caching_block_groups)) {
9831 caching_ctl = list_entry(info->caching_block_groups.next,
9832 struct btrfs_caching_control, list);
9833 list_del(&caching_ctl->list);
9834 put_caching_control(caching_ctl);
9836 up_write(&info->commit_root_sem);
9838 spin_lock(&info->unused_bgs_lock);
9839 while (!list_empty(&info->unused_bgs)) {
9840 block_group = list_first_entry(&info->unused_bgs,
9841 struct btrfs_block_group_cache,
9843 list_del_init(&block_group->bg_list);
9844 btrfs_put_block_group(block_group);
9846 spin_unlock(&info->unused_bgs_lock);
9848 spin_lock(&info->block_group_cache_lock);
9849 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9850 block_group = rb_entry(n, struct btrfs_block_group_cache,
9852 rb_erase(&block_group->cache_node,
9853 &info->block_group_cache_tree);
9854 RB_CLEAR_NODE(&block_group->cache_node);
9855 spin_unlock(&info->block_group_cache_lock);
9857 down_write(&block_group->space_info->groups_sem);
9858 list_del(&block_group->list);
9859 up_write(&block_group->space_info->groups_sem);
9862 * We haven't cached this block group, which means we could
9863 * possibly have excluded extents on this block group.
9865 if (block_group->cached == BTRFS_CACHE_NO ||
9866 block_group->cached == BTRFS_CACHE_ERROR)
9867 free_excluded_extents(info, block_group);
9869 btrfs_remove_free_space_cache(block_group);
9870 ASSERT(block_group->cached != BTRFS_CACHE_STARTED);
9871 ASSERT(list_empty(&block_group->dirty_list));
9872 ASSERT(list_empty(&block_group->io_list));
9873 ASSERT(list_empty(&block_group->bg_list));
9874 ASSERT(atomic_read(&block_group->count) == 1);
9875 btrfs_put_block_group(block_group);
9877 spin_lock(&info->block_group_cache_lock);
9879 spin_unlock(&info->block_group_cache_lock);
9881 /* now that all the block groups are freed, go through and
9882 * free all the space_info structs. This is only called during
9883 * the final stages of unmount, and so we know nobody is
9884 * using them. We call synchronize_rcu() once before we start,
9885 * just to be on the safe side.
9889 release_global_block_rsv(info);
9891 while (!list_empty(&info->space_info)) {
9894 space_info = list_entry(info->space_info.next,
9895 struct btrfs_space_info,
9899 * Do not hide this behind enospc_debug, this is actually
9900 * important and indicates a real bug if this happens.
9902 if (WARN_ON(space_info->bytes_pinned > 0 ||
9903 space_info->bytes_reserved > 0 ||
9904 space_info->bytes_may_use > 0))
9905 dump_space_info(info, space_info, 0, 0);
9906 list_del(&space_info->list);
9907 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9908 struct kobject *kobj;
9909 kobj = space_info->block_group_kobjs[i];
9910 space_info->block_group_kobjs[i] = NULL;
9916 kobject_del(&space_info->kobj);
9917 kobject_put(&space_info->kobj);
9922 static void link_block_group(struct btrfs_block_group_cache *cache)
9924 struct btrfs_space_info *space_info = cache->space_info;
9925 int index = get_block_group_index(cache);
9928 down_write(&space_info->groups_sem);
9929 if (list_empty(&space_info->block_groups[index]))
9931 list_add_tail(&cache->list, &space_info->block_groups[index]);
9932 up_write(&space_info->groups_sem);
9935 struct raid_kobject *rkobj;
9938 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9941 rkobj->raid_type = index;
9942 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9943 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9944 "%s", get_raid_name(index));
9946 kobject_put(&rkobj->kobj);
9949 space_info->block_group_kobjs[index] = &rkobj->kobj;
9954 btrfs_warn(cache->fs_info,
9955 "failed to add kobject for block cache, ignoring");
9958 static struct btrfs_block_group_cache *
9959 btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info,
9960 u64 start, u64 size)
9962 struct btrfs_block_group_cache *cache;
9964 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9968 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
9970 if (!cache->free_space_ctl) {
9975 cache->key.objectid = start;
9976 cache->key.offset = size;
9977 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9979 cache->fs_info = fs_info;
9980 cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start);
9981 set_free_space_tree_thresholds(cache);
9983 atomic_set(&cache->count, 1);
9984 spin_lock_init(&cache->lock);
9985 init_rwsem(&cache->data_rwsem);
9986 INIT_LIST_HEAD(&cache->list);
9987 INIT_LIST_HEAD(&cache->cluster_list);
9988 INIT_LIST_HEAD(&cache->bg_list);
9989 INIT_LIST_HEAD(&cache->ro_list);
9990 INIT_LIST_HEAD(&cache->dirty_list);
9991 INIT_LIST_HEAD(&cache->io_list);
9992 btrfs_init_free_space_ctl(cache);
9993 atomic_set(&cache->trimming, 0);
9994 mutex_init(&cache->free_space_lock);
9995 btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root);
10000 int btrfs_read_block_groups(struct btrfs_fs_info *info)
10002 struct btrfs_path *path;
10004 struct btrfs_block_group_cache *cache;
10005 struct btrfs_space_info *space_info;
10006 struct btrfs_key key;
10007 struct btrfs_key found_key;
10008 struct extent_buffer *leaf;
10009 int need_clear = 0;
10014 feature = btrfs_super_incompat_flags(info->super_copy);
10015 mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS);
10019 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10020 path = btrfs_alloc_path();
10023 path->reada = READA_FORWARD;
10025 cache_gen = btrfs_super_cache_generation(info->super_copy);
10026 if (btrfs_test_opt(info, SPACE_CACHE) &&
10027 btrfs_super_generation(info->super_copy) != cache_gen)
10029 if (btrfs_test_opt(info, CLEAR_CACHE))
10033 ret = find_first_block_group(info, path, &key);
10039 leaf = path->nodes[0];
10040 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
10042 cache = btrfs_create_block_group_cache(info, found_key.objectid,
10051 * When we mount with old space cache, we need to
10052 * set BTRFS_DC_CLEAR and set dirty flag.
10054 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
10055 * truncate the old free space cache inode and
10057 * b) Setting 'dirty flag' makes sure that we flush
10058 * the new space cache info onto disk.
10060 if (btrfs_test_opt(info, SPACE_CACHE))
10061 cache->disk_cache_state = BTRFS_DC_CLEAR;
10064 read_extent_buffer(leaf, &cache->item,
10065 btrfs_item_ptr_offset(leaf, path->slots[0]),
10066 sizeof(cache->item));
10067 cache->flags = btrfs_block_group_flags(&cache->item);
10069 ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) &&
10070 (cache->flags & BTRFS_BLOCK_GROUP_DATA))) {
10072 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
10073 cache->key.objectid);
10078 key.objectid = found_key.objectid + found_key.offset;
10079 btrfs_release_path(path);
10082 * We need to exclude the super stripes now so that the space
10083 * info has super bytes accounted for, otherwise we'll think
10084 * we have more space than we actually do.
10086 ret = exclude_super_stripes(info, cache);
10089 * We may have excluded something, so call this just in
10092 free_excluded_extents(info, cache);
10093 btrfs_put_block_group(cache);
10098 * check for two cases, either we are full, and therefore
10099 * don't need to bother with the caching work since we won't
10100 * find any space, or we are empty, and we can just add all
10101 * the space in and be done with it. This saves us _alot_ of
10102 * time, particularly in the full case.
10104 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
10105 cache->last_byte_to_unpin = (u64)-1;
10106 cache->cached = BTRFS_CACHE_FINISHED;
10107 free_excluded_extents(info, cache);
10108 } else if (btrfs_block_group_used(&cache->item) == 0) {
10109 cache->last_byte_to_unpin = (u64)-1;
10110 cache->cached = BTRFS_CACHE_FINISHED;
10111 add_new_free_space(cache, info,
10112 found_key.objectid,
10113 found_key.objectid +
10115 free_excluded_extents(info, cache);
10118 ret = btrfs_add_block_group_cache(info, cache);
10120 btrfs_remove_free_space_cache(cache);
10121 btrfs_put_block_group(cache);
10125 trace_btrfs_add_block_group(info, cache, 0);
10126 update_space_info(info, cache->flags, found_key.offset,
10127 btrfs_block_group_used(&cache->item),
10128 cache->bytes_super, &space_info);
10130 cache->space_info = space_info;
10132 link_block_group(cache);
10134 set_avail_alloc_bits(info, cache->flags);
10135 if (btrfs_chunk_readonly(info, cache->key.objectid)) {
10136 inc_block_group_ro(cache, 1);
10137 } else if (btrfs_block_group_used(&cache->item) == 0) {
10138 spin_lock(&info->unused_bgs_lock);
10139 /* Should always be true but just in case. */
10140 if (list_empty(&cache->bg_list)) {
10141 btrfs_get_block_group(cache);
10142 list_add_tail(&cache->bg_list,
10143 &info->unused_bgs);
10145 spin_unlock(&info->unused_bgs_lock);
10149 list_for_each_entry_rcu(space_info, &info->space_info, list) {
10150 if (!(get_alloc_profile(info, space_info->flags) &
10151 (BTRFS_BLOCK_GROUP_RAID10 |
10152 BTRFS_BLOCK_GROUP_RAID1 |
10153 BTRFS_BLOCK_GROUP_RAID5 |
10154 BTRFS_BLOCK_GROUP_RAID6 |
10155 BTRFS_BLOCK_GROUP_DUP)))
10158 * avoid allocating from un-mirrored block group if there are
10159 * mirrored block groups.
10161 list_for_each_entry(cache,
10162 &space_info->block_groups[BTRFS_RAID_RAID0],
10164 inc_block_group_ro(cache, 1);
10165 list_for_each_entry(cache,
10166 &space_info->block_groups[BTRFS_RAID_SINGLE],
10168 inc_block_group_ro(cache, 1);
10171 init_global_block_rsv(info);
10174 btrfs_free_path(path);
10178 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
10179 struct btrfs_fs_info *fs_info)
10181 struct btrfs_block_group_cache *block_group, *tmp;
10182 struct btrfs_root *extent_root = fs_info->extent_root;
10183 struct btrfs_block_group_item item;
10184 struct btrfs_key key;
10186 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
10188 trans->can_flush_pending_bgs = false;
10189 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
10193 spin_lock(&block_group->lock);
10194 memcpy(&item, &block_group->item, sizeof(item));
10195 memcpy(&key, &block_group->key, sizeof(key));
10196 spin_unlock(&block_group->lock);
10198 ret = btrfs_insert_item(trans, extent_root, &key, &item,
10201 btrfs_abort_transaction(trans, ret);
10202 ret = btrfs_finish_chunk_alloc(trans, fs_info, key.objectid,
10205 btrfs_abort_transaction(trans, ret);
10206 add_block_group_free_space(trans, fs_info, block_group);
10207 /* already aborted the transaction if it failed. */
10209 list_del_init(&block_group->bg_list);
10211 trans->can_flush_pending_bgs = can_flush_pending_bgs;
10214 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
10215 struct btrfs_fs_info *fs_info, u64 bytes_used,
10216 u64 type, u64 chunk_offset, u64 size)
10218 struct btrfs_block_group_cache *cache;
10221 btrfs_set_log_full_commit(fs_info, trans);
10223 cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size);
10227 btrfs_set_block_group_used(&cache->item, bytes_used);
10228 btrfs_set_block_group_chunk_objectid(&cache->item,
10229 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
10230 btrfs_set_block_group_flags(&cache->item, type);
10232 cache->flags = type;
10233 cache->last_byte_to_unpin = (u64)-1;
10234 cache->cached = BTRFS_CACHE_FINISHED;
10235 cache->needs_free_space = 1;
10236 ret = exclude_super_stripes(fs_info, cache);
10239 * We may have excluded something, so call this just in
10242 free_excluded_extents(fs_info, cache);
10243 btrfs_put_block_group(cache);
10247 add_new_free_space(cache, fs_info, chunk_offset, chunk_offset + size);
10249 free_excluded_extents(fs_info, cache);
10251 #ifdef CONFIG_BTRFS_DEBUG
10252 if (btrfs_should_fragment_free_space(cache)) {
10253 u64 new_bytes_used = size - bytes_used;
10255 bytes_used += new_bytes_used >> 1;
10256 fragment_free_space(cache);
10260 * Ensure the corresponding space_info object is created and
10261 * assigned to our block group. We want our bg to be added to the rbtree
10262 * with its ->space_info set.
10264 cache->space_info = __find_space_info(fs_info, cache->flags);
10265 if (!cache->space_info) {
10266 ret = create_space_info(fs_info, cache->flags,
10267 &cache->space_info);
10269 btrfs_remove_free_space_cache(cache);
10270 btrfs_put_block_group(cache);
10275 ret = btrfs_add_block_group_cache(fs_info, cache);
10277 btrfs_remove_free_space_cache(cache);
10278 btrfs_put_block_group(cache);
10283 * Now that our block group has its ->space_info set and is inserted in
10284 * the rbtree, update the space info's counters.
10286 trace_btrfs_add_block_group(fs_info, cache, 1);
10287 update_space_info(fs_info, cache->flags, size, bytes_used,
10288 cache->bytes_super, &cache->space_info);
10289 update_global_block_rsv(fs_info);
10291 link_block_group(cache);
10293 list_add_tail(&cache->bg_list, &trans->new_bgs);
10295 set_avail_alloc_bits(fs_info, type);
10299 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
10301 u64 extra_flags = chunk_to_extended(flags) &
10302 BTRFS_EXTENDED_PROFILE_MASK;
10304 write_seqlock(&fs_info->profiles_lock);
10305 if (flags & BTRFS_BLOCK_GROUP_DATA)
10306 fs_info->avail_data_alloc_bits &= ~extra_flags;
10307 if (flags & BTRFS_BLOCK_GROUP_METADATA)
10308 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
10309 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
10310 fs_info->avail_system_alloc_bits &= ~extra_flags;
10311 write_sequnlock(&fs_info->profiles_lock);
10314 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
10315 struct btrfs_fs_info *fs_info, u64 group_start,
10316 struct extent_map *em)
10318 struct btrfs_root *root = fs_info->extent_root;
10319 struct btrfs_path *path;
10320 struct btrfs_block_group_cache *block_group;
10321 struct btrfs_free_cluster *cluster;
10322 struct btrfs_root *tree_root = fs_info->tree_root;
10323 struct btrfs_key key;
10324 struct inode *inode;
10325 struct kobject *kobj = NULL;
10329 struct btrfs_caching_control *caching_ctl = NULL;
10332 block_group = btrfs_lookup_block_group(fs_info, group_start);
10333 BUG_ON(!block_group);
10334 BUG_ON(!block_group->ro);
10337 * Free the reserved super bytes from this block group before
10340 free_excluded_extents(fs_info, block_group);
10341 btrfs_free_ref_tree_range(fs_info, block_group->key.objectid,
10342 block_group->key.offset);
10344 memcpy(&key, &block_group->key, sizeof(key));
10345 index = get_block_group_index(block_group);
10346 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
10347 BTRFS_BLOCK_GROUP_RAID1 |
10348 BTRFS_BLOCK_GROUP_RAID10))
10353 /* make sure this block group isn't part of an allocation cluster */
10354 cluster = &fs_info->data_alloc_cluster;
10355 spin_lock(&cluster->refill_lock);
10356 btrfs_return_cluster_to_free_space(block_group, cluster);
10357 spin_unlock(&cluster->refill_lock);
10360 * make sure this block group isn't part of a metadata
10361 * allocation cluster
10363 cluster = &fs_info->meta_alloc_cluster;
10364 spin_lock(&cluster->refill_lock);
10365 btrfs_return_cluster_to_free_space(block_group, cluster);
10366 spin_unlock(&cluster->refill_lock);
10368 path = btrfs_alloc_path();
10375 * get the inode first so any iput calls done for the io_list
10376 * aren't the final iput (no unlinks allowed now)
10378 inode = lookup_free_space_inode(fs_info, block_group, path);
10380 mutex_lock(&trans->transaction->cache_write_mutex);
10382 * make sure our free spache cache IO is done before remove the
10385 spin_lock(&trans->transaction->dirty_bgs_lock);
10386 if (!list_empty(&block_group->io_list)) {
10387 list_del_init(&block_group->io_list);
10389 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
10391 spin_unlock(&trans->transaction->dirty_bgs_lock);
10392 btrfs_wait_cache_io(trans, block_group, path);
10393 btrfs_put_block_group(block_group);
10394 spin_lock(&trans->transaction->dirty_bgs_lock);
10397 if (!list_empty(&block_group->dirty_list)) {
10398 list_del_init(&block_group->dirty_list);
10399 btrfs_put_block_group(block_group);
10401 spin_unlock(&trans->transaction->dirty_bgs_lock);
10402 mutex_unlock(&trans->transaction->cache_write_mutex);
10404 if (!IS_ERR(inode)) {
10405 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10407 btrfs_add_delayed_iput(inode);
10410 clear_nlink(inode);
10411 /* One for the block groups ref */
10412 spin_lock(&block_group->lock);
10413 if (block_group->iref) {
10414 block_group->iref = 0;
10415 block_group->inode = NULL;
10416 spin_unlock(&block_group->lock);
10419 spin_unlock(&block_group->lock);
10421 /* One for our lookup ref */
10422 btrfs_add_delayed_iput(inode);
10425 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
10426 key.offset = block_group->key.objectid;
10429 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
10433 btrfs_release_path(path);
10435 ret = btrfs_del_item(trans, tree_root, path);
10438 btrfs_release_path(path);
10441 spin_lock(&fs_info->block_group_cache_lock);
10442 rb_erase(&block_group->cache_node,
10443 &fs_info->block_group_cache_tree);
10444 RB_CLEAR_NODE(&block_group->cache_node);
10446 if (fs_info->first_logical_byte == block_group->key.objectid)
10447 fs_info->first_logical_byte = (u64)-1;
10448 spin_unlock(&fs_info->block_group_cache_lock);
10450 down_write(&block_group->space_info->groups_sem);
10452 * we must use list_del_init so people can check to see if they
10453 * are still on the list after taking the semaphore
10455 list_del_init(&block_group->list);
10456 if (list_empty(&block_group->space_info->block_groups[index])) {
10457 kobj = block_group->space_info->block_group_kobjs[index];
10458 block_group->space_info->block_group_kobjs[index] = NULL;
10459 clear_avail_alloc_bits(fs_info, block_group->flags);
10461 up_write(&block_group->space_info->groups_sem);
10467 if (block_group->has_caching_ctl)
10468 caching_ctl = get_caching_control(block_group);
10469 if (block_group->cached == BTRFS_CACHE_STARTED)
10470 wait_block_group_cache_done(block_group);
10471 if (block_group->has_caching_ctl) {
10472 down_write(&fs_info->commit_root_sem);
10473 if (!caching_ctl) {
10474 struct btrfs_caching_control *ctl;
10476 list_for_each_entry(ctl,
10477 &fs_info->caching_block_groups, list)
10478 if (ctl->block_group == block_group) {
10480 refcount_inc(&caching_ctl->count);
10485 list_del_init(&caching_ctl->list);
10486 up_write(&fs_info->commit_root_sem);
10488 /* Once for the caching bgs list and once for us. */
10489 put_caching_control(caching_ctl);
10490 put_caching_control(caching_ctl);
10494 spin_lock(&trans->transaction->dirty_bgs_lock);
10495 if (!list_empty(&block_group->dirty_list)) {
10498 if (!list_empty(&block_group->io_list)) {
10501 spin_unlock(&trans->transaction->dirty_bgs_lock);
10502 btrfs_remove_free_space_cache(block_group);
10504 spin_lock(&block_group->space_info->lock);
10505 list_del_init(&block_group->ro_list);
10507 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
10508 WARN_ON(block_group->space_info->total_bytes
10509 < block_group->key.offset);
10510 WARN_ON(block_group->space_info->bytes_readonly
10511 < block_group->key.offset);
10512 WARN_ON(block_group->space_info->disk_total
10513 < block_group->key.offset * factor);
10515 block_group->space_info->total_bytes -= block_group->key.offset;
10516 block_group->space_info->bytes_readonly -= block_group->key.offset;
10517 block_group->space_info->disk_total -= block_group->key.offset * factor;
10519 spin_unlock(&block_group->space_info->lock);
10521 memcpy(&key, &block_group->key, sizeof(key));
10523 mutex_lock(&fs_info->chunk_mutex);
10524 if (!list_empty(&em->list)) {
10525 /* We're in the transaction->pending_chunks list. */
10526 free_extent_map(em);
10528 spin_lock(&block_group->lock);
10529 block_group->removed = 1;
10531 * At this point trimming can't start on this block group, because we
10532 * removed the block group from the tree fs_info->block_group_cache_tree
10533 * so no one can't find it anymore and even if someone already got this
10534 * block group before we removed it from the rbtree, they have already
10535 * incremented block_group->trimming - if they didn't, they won't find
10536 * any free space entries because we already removed them all when we
10537 * called btrfs_remove_free_space_cache().
10539 * And we must not remove the extent map from the fs_info->mapping_tree
10540 * to prevent the same logical address range and physical device space
10541 * ranges from being reused for a new block group. This is because our
10542 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10543 * completely transactionless, so while it is trimming a range the
10544 * currently running transaction might finish and a new one start,
10545 * allowing for new block groups to be created that can reuse the same
10546 * physical device locations unless we take this special care.
10548 * There may also be an implicit trim operation if the file system
10549 * is mounted with -odiscard. The same protections must remain
10550 * in place until the extents have been discarded completely when
10551 * the transaction commit has completed.
10553 remove_em = (atomic_read(&block_group->trimming) == 0);
10555 * Make sure a trimmer task always sees the em in the pinned_chunks list
10556 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10557 * before checking block_group->removed).
10561 * Our em might be in trans->transaction->pending_chunks which
10562 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10563 * and so is the fs_info->pinned_chunks list.
10565 * So at this point we must be holding the chunk_mutex to avoid
10566 * any races with chunk allocation (more specifically at
10567 * volumes.c:contains_pending_extent()), to ensure it always
10568 * sees the em, either in the pending_chunks list or in the
10569 * pinned_chunks list.
10571 list_move_tail(&em->list, &fs_info->pinned_chunks);
10573 spin_unlock(&block_group->lock);
10576 struct extent_map_tree *em_tree;
10578 em_tree = &fs_info->mapping_tree.map_tree;
10579 write_lock(&em_tree->lock);
10581 * The em might be in the pending_chunks list, so make sure the
10582 * chunk mutex is locked, since remove_extent_mapping() will
10583 * delete us from that list.
10585 remove_extent_mapping(em_tree, em);
10586 write_unlock(&em_tree->lock);
10587 /* once for the tree */
10588 free_extent_map(em);
10591 mutex_unlock(&fs_info->chunk_mutex);
10593 ret = remove_block_group_free_space(trans, fs_info, block_group);
10597 btrfs_put_block_group(block_group);
10598 btrfs_put_block_group(block_group);
10600 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10606 ret = btrfs_del_item(trans, root, path);
10608 btrfs_free_path(path);
10612 struct btrfs_trans_handle *
10613 btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info,
10614 const u64 chunk_offset)
10616 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
10617 struct extent_map *em;
10618 struct map_lookup *map;
10619 unsigned int num_items;
10621 read_lock(&em_tree->lock);
10622 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
10623 read_unlock(&em_tree->lock);
10624 ASSERT(em && em->start == chunk_offset);
10627 * We need to reserve 3 + N units from the metadata space info in order
10628 * to remove a block group (done at btrfs_remove_chunk() and at
10629 * btrfs_remove_block_group()), which are used for:
10631 * 1 unit for adding the free space inode's orphan (located in the tree
10633 * 1 unit for deleting the block group item (located in the extent
10635 * 1 unit for deleting the free space item (located in tree of tree
10637 * N units for deleting N device extent items corresponding to each
10638 * stripe (located in the device tree).
10640 * In order to remove a block group we also need to reserve units in the
10641 * system space info in order to update the chunk tree (update one or
10642 * more device items and remove one chunk item), but this is done at
10643 * btrfs_remove_chunk() through a call to check_system_chunk().
10645 map = em->map_lookup;
10646 num_items = 3 + map->num_stripes;
10647 free_extent_map(em);
10649 return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root,
10654 * Process the unused_bgs list and remove any that don't have any allocated
10655 * space inside of them.
10657 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10659 struct btrfs_block_group_cache *block_group;
10660 struct btrfs_space_info *space_info;
10661 struct btrfs_trans_handle *trans;
10664 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
10667 spin_lock(&fs_info->unused_bgs_lock);
10668 while (!list_empty(&fs_info->unused_bgs)) {
10672 block_group = list_first_entry(&fs_info->unused_bgs,
10673 struct btrfs_block_group_cache,
10675 list_del_init(&block_group->bg_list);
10677 space_info = block_group->space_info;
10679 if (ret || btrfs_mixed_space_info(space_info)) {
10680 btrfs_put_block_group(block_group);
10683 spin_unlock(&fs_info->unused_bgs_lock);
10685 mutex_lock(&fs_info->delete_unused_bgs_mutex);
10687 /* Don't want to race with allocators so take the groups_sem */
10688 down_write(&space_info->groups_sem);
10689 spin_lock(&block_group->lock);
10690 if (block_group->reserved ||
10691 btrfs_block_group_used(&block_group->item) ||
10693 list_is_singular(&block_group->list)) {
10695 * We want to bail if we made new allocations or have
10696 * outstanding allocations in this block group. We do
10697 * the ro check in case balance is currently acting on
10698 * this block group.
10700 spin_unlock(&block_group->lock);
10701 up_write(&space_info->groups_sem);
10704 spin_unlock(&block_group->lock);
10706 /* We don't want to force the issue, only flip if it's ok. */
10707 ret = inc_block_group_ro(block_group, 0);
10708 up_write(&space_info->groups_sem);
10715 * Want to do this before we do anything else so we can recover
10716 * properly if we fail to join the transaction.
10718 trans = btrfs_start_trans_remove_block_group(fs_info,
10719 block_group->key.objectid);
10720 if (IS_ERR(trans)) {
10721 btrfs_dec_block_group_ro(block_group);
10722 ret = PTR_ERR(trans);
10727 * We could have pending pinned extents for this block group,
10728 * just delete them, we don't care about them anymore.
10730 start = block_group->key.objectid;
10731 end = start + block_group->key.offset - 1;
10733 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10734 * btrfs_finish_extent_commit(). If we are at transaction N,
10735 * another task might be running finish_extent_commit() for the
10736 * previous transaction N - 1, and have seen a range belonging
10737 * to the block group in freed_extents[] before we were able to
10738 * clear the whole block group range from freed_extents[]. This
10739 * means that task can lookup for the block group after we
10740 * unpinned it from freed_extents[] and removed it, leading to
10741 * a BUG_ON() at btrfs_unpin_extent_range().
10743 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10744 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10747 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10748 btrfs_dec_block_group_ro(block_group);
10751 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10754 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10755 btrfs_dec_block_group_ro(block_group);
10758 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10760 /* Reset pinned so btrfs_put_block_group doesn't complain */
10761 spin_lock(&space_info->lock);
10762 spin_lock(&block_group->lock);
10764 space_info->bytes_pinned -= block_group->pinned;
10765 space_info->bytes_readonly += block_group->pinned;
10766 percpu_counter_add(&space_info->total_bytes_pinned,
10767 -block_group->pinned);
10768 block_group->pinned = 0;
10770 spin_unlock(&block_group->lock);
10771 spin_unlock(&space_info->lock);
10773 /* DISCARD can flip during remount */
10774 trimming = btrfs_test_opt(fs_info, DISCARD);
10776 /* Implicit trim during transaction commit. */
10778 btrfs_get_block_group_trimming(block_group);
10781 * Btrfs_remove_chunk will abort the transaction if things go
10784 ret = btrfs_remove_chunk(trans, fs_info,
10785 block_group->key.objectid);
10789 btrfs_put_block_group_trimming(block_group);
10794 * If we're not mounted with -odiscard, we can just forget
10795 * about this block group. Otherwise we'll need to wait
10796 * until transaction commit to do the actual discard.
10799 spin_lock(&fs_info->unused_bgs_lock);
10801 * A concurrent scrub might have added us to the list
10802 * fs_info->unused_bgs, so use a list_move operation
10803 * to add the block group to the deleted_bgs list.
10805 list_move(&block_group->bg_list,
10806 &trans->transaction->deleted_bgs);
10807 spin_unlock(&fs_info->unused_bgs_lock);
10808 btrfs_get_block_group(block_group);
10811 btrfs_end_transaction(trans);
10813 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
10814 btrfs_put_block_group(block_group);
10815 spin_lock(&fs_info->unused_bgs_lock);
10817 spin_unlock(&fs_info->unused_bgs_lock);
10820 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10822 struct btrfs_space_info *space_info;
10823 struct btrfs_super_block *disk_super;
10829 disk_super = fs_info->super_copy;
10830 if (!btrfs_super_root(disk_super))
10833 features = btrfs_super_incompat_flags(disk_super);
10834 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10837 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10838 ret = create_space_info(fs_info, flags, &space_info);
10843 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10844 ret = create_space_info(fs_info, flags, &space_info);
10846 flags = BTRFS_BLOCK_GROUP_METADATA;
10847 ret = create_space_info(fs_info, flags, &space_info);
10851 flags = BTRFS_BLOCK_GROUP_DATA;
10852 ret = create_space_info(fs_info, flags, &space_info);
10858 int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
10859 u64 start, u64 end)
10861 return unpin_extent_range(fs_info, start, end, false);
10865 * It used to be that old block groups would be left around forever.
10866 * Iterating over them would be enough to trim unused space. Since we
10867 * now automatically remove them, we also need to iterate over unallocated
10870 * We don't want a transaction for this since the discard may take a
10871 * substantial amount of time. We don't require that a transaction be
10872 * running, but we do need to take a running transaction into account
10873 * to ensure that we're not discarding chunks that were released in
10874 * the current transaction.
10876 * Holding the chunks lock will prevent other threads from allocating
10877 * or releasing chunks, but it won't prevent a running transaction
10878 * from committing and releasing the memory that the pending chunks
10879 * list head uses. For that, we need to take a reference to the
10882 static int btrfs_trim_free_extents(struct btrfs_device *device,
10883 u64 minlen, u64 *trimmed)
10885 u64 start = 0, len = 0;
10890 /* Not writeable = nothing to do. */
10891 if (!device->writeable)
10894 /* No free space = nothing to do. */
10895 if (device->total_bytes <= device->bytes_used)
10901 struct btrfs_fs_info *fs_info = device->fs_info;
10902 struct btrfs_transaction *trans;
10905 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
10909 down_read(&fs_info->commit_root_sem);
10911 spin_lock(&fs_info->trans_lock);
10912 trans = fs_info->running_transaction;
10914 refcount_inc(&trans->use_count);
10915 spin_unlock(&fs_info->trans_lock);
10917 ret = find_free_dev_extent_start(trans, device, minlen, start,
10920 btrfs_put_transaction(trans);
10923 up_read(&fs_info->commit_root_sem);
10924 mutex_unlock(&fs_info->chunk_mutex);
10925 if (ret == -ENOSPC)
10930 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
10931 up_read(&fs_info->commit_root_sem);
10932 mutex_unlock(&fs_info->chunk_mutex);
10940 if (fatal_signal_pending(current)) {
10941 ret = -ERESTARTSYS;
10951 int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
10953 struct btrfs_block_group_cache *cache = NULL;
10954 struct btrfs_device *device;
10955 struct list_head *devices;
10960 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
10964 * try to trim all FS space, our block group may start from non-zero.
10966 if (range->len == total_bytes)
10967 cache = btrfs_lookup_first_block_group(fs_info, range->start);
10969 cache = btrfs_lookup_block_group(fs_info, range->start);
10972 if (cache->key.objectid >= (range->start + range->len)) {
10973 btrfs_put_block_group(cache);
10977 start = max(range->start, cache->key.objectid);
10978 end = min(range->start + range->len,
10979 cache->key.objectid + cache->key.offset);
10981 if (end - start >= range->minlen) {
10982 if (!block_group_cache_done(cache)) {
10983 ret = cache_block_group(cache, 0);
10985 btrfs_put_block_group(cache);
10988 ret = wait_block_group_cache_done(cache);
10990 btrfs_put_block_group(cache);
10994 ret = btrfs_trim_block_group(cache,
11000 trimmed += group_trimmed;
11002 btrfs_put_block_group(cache);
11007 cache = next_block_group(fs_info, cache);
11010 mutex_lock(&fs_info->fs_devices->device_list_mutex);
11011 devices = &fs_info->fs_devices->alloc_list;
11012 list_for_each_entry(device, devices, dev_alloc_list) {
11013 ret = btrfs_trim_free_extents(device, range->minlen,
11018 trimmed += group_trimmed;
11020 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
11022 range->len = trimmed;
11027 * btrfs_{start,end}_write_no_snapshotting() are similar to
11028 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
11029 * data into the page cache through nocow before the subvolume is snapshoted,
11030 * but flush the data into disk after the snapshot creation, or to prevent
11031 * operations while snapshotting is ongoing and that cause the snapshot to be
11032 * inconsistent (writes followed by expanding truncates for example).
11034 void btrfs_end_write_no_snapshotting(struct btrfs_root *root)
11036 percpu_counter_dec(&root->subv_writers->counter);
11038 * Make sure counter is updated before we wake up waiters.
11041 if (waitqueue_active(&root->subv_writers->wait))
11042 wake_up(&root->subv_writers->wait);
11045 int btrfs_start_write_no_snapshotting(struct btrfs_root *root)
11047 if (atomic_read(&root->will_be_snapshotted))
11050 percpu_counter_inc(&root->subv_writers->counter);
11052 * Make sure counter is updated before we check for snapshot creation.
11055 if (atomic_read(&root->will_be_snapshotted)) {
11056 btrfs_end_write_no_snapshotting(root);
11062 static int wait_snapshotting_atomic_t(atomic_t *a)
11068 void btrfs_wait_for_snapshot_creation(struct btrfs_root *root)
11073 ret = btrfs_start_write_no_snapshotting(root);
11076 wait_on_atomic_t(&root->will_be_snapshotted,
11077 wait_snapshotting_atomic_t,
11078 TASK_UNINTERRUPTIBLE);