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/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include <linux/slab.h>
26 #include <linux/ratelimit.h>
27 #include <linux/percpu_counter.h>
31 #include "print-tree.h"
35 #include "free-space-cache.h"
36 #include "free-space-tree.h"
41 #undef SCRAMBLE_DELAYED_REFS
44 * control flags for do_chunk_alloc's force field
45 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
46 * if we really need one.
48 * CHUNK_ALLOC_LIMITED means to only try and allocate one
49 * if we have very few chunks already allocated. This is
50 * used as part of the clustering code to help make sure
51 * we have a good pool of storage to cluster in, without
52 * filling the FS with empty chunks
54 * CHUNK_ALLOC_FORCE means it must try to allocate one
58 CHUNK_ALLOC_NO_FORCE = 0,
59 CHUNK_ALLOC_LIMITED = 1,
60 CHUNK_ALLOC_FORCE = 2,
63 static int update_block_group(struct btrfs_trans_handle *trans,
64 struct btrfs_root *root, u64 bytenr,
65 u64 num_bytes, int alloc);
66 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
67 struct btrfs_root *root,
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_root *root,
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_root *root,
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_root *extent_root, 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 btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache,
94 u64 ram_bytes, u64 num_bytes, int delalloc);
95 static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache,
96 u64 num_bytes, int delalloc);
97 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
99 int btrfs_pin_extent(struct btrfs_root *root,
100 u64 bytenr, u64 num_bytes, int reserved);
101 static int __reserve_metadata_bytes(struct btrfs_root *root,
102 struct btrfs_space_info *space_info,
104 enum btrfs_reserve_flush_enum flush);
105 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
106 struct btrfs_space_info *space_info,
108 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
109 struct btrfs_space_info *space_info,
113 block_group_cache_done(struct btrfs_block_group_cache *cache)
116 return cache->cached == BTRFS_CACHE_FINISHED ||
117 cache->cached == BTRFS_CACHE_ERROR;
120 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
122 return (cache->flags & bits) == bits;
125 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
127 atomic_inc(&cache->count);
130 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
132 if (atomic_dec_and_test(&cache->count)) {
133 WARN_ON(cache->pinned > 0);
134 WARN_ON(cache->reserved > 0);
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_root *root,
227 u64 start, u64 num_bytes)
229 u64 end = start + num_bytes - 1;
230 set_extent_bits(&root->fs_info->freed_extents[0],
231 start, end, EXTENT_UPTODATE);
232 set_extent_bits(&root->fs_info->freed_extents[1],
233 start, end, EXTENT_UPTODATE);
237 static void free_excluded_extents(struct btrfs_root *root,
238 struct btrfs_block_group_cache *cache)
242 start = cache->key.objectid;
243 end = start + cache->key.offset - 1;
245 clear_extent_bits(&root->fs_info->freed_extents[0],
246 start, end, EXTENT_UPTODATE);
247 clear_extent_bits(&root->fs_info->freed_extents[1],
248 start, end, EXTENT_UPTODATE);
251 static int exclude_super_stripes(struct btrfs_root *root,
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(root, cache->key.objectid,
268 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
269 bytenr = btrfs_sb_offset(i);
270 ret = btrfs_rmap_block(root->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(root, 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 atomic_inc(&ctl->count);
321 spin_unlock(&cache->lock);
325 static void put_caching_control(struct btrfs_caching_control *ctl)
327 if (atomic_dec_and_test(&ctl->count))
331 #ifdef CONFIG_BTRFS_DEBUG
332 static void fragment_free_space(struct btrfs_root *root,
333 struct btrfs_block_group_cache *block_group)
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 root->nodesize : root->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;
398 struct btrfs_fs_info *fs_info;
399 struct btrfs_root *extent_root;
400 struct btrfs_path *path;
401 struct extent_buffer *leaf;
402 struct btrfs_key key;
409 block_group = caching_ctl->block_group;
410 fs_info = block_group->fs_info;
411 extent_root = fs_info->extent_root;
413 path = btrfs_alloc_path();
417 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
419 #ifdef CONFIG_BTRFS_DEBUG
421 * If we're fragmenting we don't want to make anybody think we can
422 * allocate from this block group until we've had a chance to fragment
425 if (btrfs_should_fragment_free_space(extent_root, block_group))
429 * We don't want to deadlock with somebody trying to allocate a new
430 * extent for the extent root while also trying to search the extent
431 * root to add free space. So we skip locking and search the commit
432 * root, since its read-only
434 path->skip_locking = 1;
435 path->search_commit_root = 1;
436 path->reada = READA_FORWARD;
440 key.type = BTRFS_EXTENT_ITEM_KEY;
443 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
447 leaf = path->nodes[0];
448 nritems = btrfs_header_nritems(leaf);
451 if (btrfs_fs_closing(fs_info) > 1) {
456 if (path->slots[0] < nritems) {
457 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
459 ret = find_next_key(path, 0, &key);
463 if (need_resched() ||
464 rwsem_is_contended(&fs_info->commit_root_sem)) {
466 caching_ctl->progress = last;
467 btrfs_release_path(path);
468 up_read(&fs_info->commit_root_sem);
469 mutex_unlock(&caching_ctl->mutex);
471 mutex_lock(&caching_ctl->mutex);
472 down_read(&fs_info->commit_root_sem);
476 ret = btrfs_next_leaf(extent_root, path);
481 leaf = path->nodes[0];
482 nritems = btrfs_header_nritems(leaf);
486 if (key.objectid < last) {
489 key.type = BTRFS_EXTENT_ITEM_KEY;
492 caching_ctl->progress = last;
493 btrfs_release_path(path);
497 if (key.objectid < block_group->key.objectid) {
502 if (key.objectid >= block_group->key.objectid +
503 block_group->key.offset)
506 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
507 key.type == BTRFS_METADATA_ITEM_KEY) {
508 total_found += add_new_free_space(block_group,
511 if (key.type == BTRFS_METADATA_ITEM_KEY)
512 last = key.objectid +
513 fs_info->tree_root->nodesize;
515 last = key.objectid + key.offset;
517 if (total_found > CACHING_CTL_WAKE_UP) {
520 wake_up(&caching_ctl->wait);
527 total_found += add_new_free_space(block_group, fs_info, last,
528 block_group->key.objectid +
529 block_group->key.offset);
530 caching_ctl->progress = (u64)-1;
533 btrfs_free_path(path);
537 static noinline void caching_thread(struct btrfs_work *work)
539 struct btrfs_block_group_cache *block_group;
540 struct btrfs_fs_info *fs_info;
541 struct btrfs_caching_control *caching_ctl;
542 struct btrfs_root *extent_root;
545 caching_ctl = container_of(work, struct btrfs_caching_control, work);
546 block_group = caching_ctl->block_group;
547 fs_info = block_group->fs_info;
548 extent_root = fs_info->extent_root;
550 mutex_lock(&caching_ctl->mutex);
551 down_read(&fs_info->commit_root_sem);
553 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
554 ret = load_free_space_tree(caching_ctl);
556 ret = load_extent_tree_free(caching_ctl);
558 spin_lock(&block_group->lock);
559 block_group->caching_ctl = NULL;
560 block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED;
561 spin_unlock(&block_group->lock);
563 #ifdef CONFIG_BTRFS_DEBUG
564 if (btrfs_should_fragment_free_space(extent_root, block_group)) {
567 spin_lock(&block_group->space_info->lock);
568 spin_lock(&block_group->lock);
569 bytes_used = block_group->key.offset -
570 btrfs_block_group_used(&block_group->item);
571 block_group->space_info->bytes_used += bytes_used >> 1;
572 spin_unlock(&block_group->lock);
573 spin_unlock(&block_group->space_info->lock);
574 fragment_free_space(extent_root, block_group);
578 caching_ctl->progress = (u64)-1;
580 up_read(&fs_info->commit_root_sem);
581 free_excluded_extents(fs_info->extent_root, block_group);
582 mutex_unlock(&caching_ctl->mutex);
584 wake_up(&caching_ctl->wait);
586 put_caching_control(caching_ctl);
587 btrfs_put_block_group(block_group);
590 static int cache_block_group(struct btrfs_block_group_cache *cache,
594 struct btrfs_fs_info *fs_info = cache->fs_info;
595 struct btrfs_caching_control *caching_ctl;
598 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
602 INIT_LIST_HEAD(&caching_ctl->list);
603 mutex_init(&caching_ctl->mutex);
604 init_waitqueue_head(&caching_ctl->wait);
605 caching_ctl->block_group = cache;
606 caching_ctl->progress = cache->key.objectid;
607 atomic_set(&caching_ctl->count, 1);
608 btrfs_init_work(&caching_ctl->work, btrfs_cache_helper,
609 caching_thread, NULL, NULL);
611 spin_lock(&cache->lock);
613 * This should be a rare occasion, but this could happen I think in the
614 * case where one thread starts to load the space cache info, and then
615 * some other thread starts a transaction commit which tries to do an
616 * allocation while the other thread is still loading the space cache
617 * info. The previous loop should have kept us from choosing this block
618 * group, but if we've moved to the state where we will wait on caching
619 * block groups we need to first check if we're doing a fast load here,
620 * so we can wait for it to finish, otherwise we could end up allocating
621 * from a block group who's cache gets evicted for one reason or
624 while (cache->cached == BTRFS_CACHE_FAST) {
625 struct btrfs_caching_control *ctl;
627 ctl = cache->caching_ctl;
628 atomic_inc(&ctl->count);
629 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
630 spin_unlock(&cache->lock);
634 finish_wait(&ctl->wait, &wait);
635 put_caching_control(ctl);
636 spin_lock(&cache->lock);
639 if (cache->cached != BTRFS_CACHE_NO) {
640 spin_unlock(&cache->lock);
644 WARN_ON(cache->caching_ctl);
645 cache->caching_ctl = caching_ctl;
646 cache->cached = BTRFS_CACHE_FAST;
647 spin_unlock(&cache->lock);
649 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
650 mutex_lock(&caching_ctl->mutex);
651 ret = load_free_space_cache(fs_info, cache);
653 spin_lock(&cache->lock);
655 cache->caching_ctl = NULL;
656 cache->cached = BTRFS_CACHE_FINISHED;
657 cache->last_byte_to_unpin = (u64)-1;
658 caching_ctl->progress = (u64)-1;
660 if (load_cache_only) {
661 cache->caching_ctl = NULL;
662 cache->cached = BTRFS_CACHE_NO;
664 cache->cached = BTRFS_CACHE_STARTED;
665 cache->has_caching_ctl = 1;
668 spin_unlock(&cache->lock);
669 #ifdef CONFIG_BTRFS_DEBUG
671 btrfs_should_fragment_free_space(fs_info->extent_root,
675 spin_lock(&cache->space_info->lock);
676 spin_lock(&cache->lock);
677 bytes_used = cache->key.offset -
678 btrfs_block_group_used(&cache->item);
679 cache->space_info->bytes_used += bytes_used >> 1;
680 spin_unlock(&cache->lock);
681 spin_unlock(&cache->space_info->lock);
682 fragment_free_space(fs_info->extent_root, cache);
685 mutex_unlock(&caching_ctl->mutex);
687 wake_up(&caching_ctl->wait);
689 put_caching_control(caching_ctl);
690 free_excluded_extents(fs_info->extent_root, cache);
695 * We're either using the free space tree or no caching at all.
696 * Set cached to the appropriate value and wakeup any waiters.
698 spin_lock(&cache->lock);
699 if (load_cache_only) {
700 cache->caching_ctl = NULL;
701 cache->cached = BTRFS_CACHE_NO;
703 cache->cached = BTRFS_CACHE_STARTED;
704 cache->has_caching_ctl = 1;
706 spin_unlock(&cache->lock);
707 wake_up(&caching_ctl->wait);
710 if (load_cache_only) {
711 put_caching_control(caching_ctl);
715 down_write(&fs_info->commit_root_sem);
716 atomic_inc(&caching_ctl->count);
717 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
718 up_write(&fs_info->commit_root_sem);
720 btrfs_get_block_group(cache);
722 btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
728 * return the block group that starts at or after bytenr
730 static struct btrfs_block_group_cache *
731 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
733 return block_group_cache_tree_search(info, bytenr, 0);
737 * return the block group that contains the given bytenr
739 struct btrfs_block_group_cache *btrfs_lookup_block_group(
740 struct btrfs_fs_info *info,
743 return block_group_cache_tree_search(info, bytenr, 1);
746 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
749 struct list_head *head = &info->space_info;
750 struct btrfs_space_info *found;
752 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
755 list_for_each_entry_rcu(found, head, list) {
756 if (found->flags & flags) {
766 * after adding space to the filesystem, we need to clear the full flags
767 * on all the space infos.
769 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
771 struct list_head *head = &info->space_info;
772 struct btrfs_space_info *found;
775 list_for_each_entry_rcu(found, head, list)
780 /* simple helper to search for an existing data extent at a given offset */
781 int btrfs_lookup_data_extent(struct btrfs_root *root, u64 start, u64 len)
784 struct btrfs_key key;
785 struct btrfs_path *path;
787 path = btrfs_alloc_path();
791 key.objectid = start;
793 key.type = BTRFS_EXTENT_ITEM_KEY;
794 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
796 btrfs_free_path(path);
801 * helper function to lookup reference count and flags of a tree block.
803 * the head node for delayed ref is used to store the sum of all the
804 * reference count modifications queued up in the rbtree. the head
805 * node may also store the extent flags to set. This way you can check
806 * to see what the reference count and extent flags would be if all of
807 * the delayed refs are not processed.
809 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
810 struct btrfs_root *root, u64 bytenr,
811 u64 offset, int metadata, u64 *refs, u64 *flags)
813 struct btrfs_delayed_ref_head *head;
814 struct btrfs_delayed_ref_root *delayed_refs;
815 struct btrfs_path *path;
816 struct btrfs_extent_item *ei;
817 struct extent_buffer *leaf;
818 struct btrfs_key key;
825 * If we don't have skinny metadata, don't bother doing anything
828 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
829 offset = root->nodesize;
833 path = btrfs_alloc_path();
838 path->skip_locking = 1;
839 path->search_commit_root = 1;
843 key.objectid = bytenr;
846 key.type = BTRFS_METADATA_ITEM_KEY;
848 key.type = BTRFS_EXTENT_ITEM_KEY;
850 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
855 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
856 if (path->slots[0]) {
858 btrfs_item_key_to_cpu(path->nodes[0], &key,
860 if (key.objectid == bytenr &&
861 key.type == BTRFS_EXTENT_ITEM_KEY &&
862 key.offset == root->nodesize)
868 leaf = path->nodes[0];
869 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
870 if (item_size >= sizeof(*ei)) {
871 ei = btrfs_item_ptr(leaf, path->slots[0],
872 struct btrfs_extent_item);
873 num_refs = btrfs_extent_refs(leaf, ei);
874 extent_flags = btrfs_extent_flags(leaf, ei);
876 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
877 struct btrfs_extent_item_v0 *ei0;
878 BUG_ON(item_size != sizeof(*ei0));
879 ei0 = btrfs_item_ptr(leaf, path->slots[0],
880 struct btrfs_extent_item_v0);
881 num_refs = btrfs_extent_refs_v0(leaf, ei0);
882 /* FIXME: this isn't correct for data */
883 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
888 BUG_ON(num_refs == 0);
898 delayed_refs = &trans->transaction->delayed_refs;
899 spin_lock(&delayed_refs->lock);
900 head = btrfs_find_delayed_ref_head(trans, bytenr);
902 if (!mutex_trylock(&head->mutex)) {
903 atomic_inc(&head->node.refs);
904 spin_unlock(&delayed_refs->lock);
906 btrfs_release_path(path);
909 * Mutex was contended, block until it's released and try
912 mutex_lock(&head->mutex);
913 mutex_unlock(&head->mutex);
914 btrfs_put_delayed_ref(&head->node);
917 spin_lock(&head->lock);
918 if (head->extent_op && head->extent_op->update_flags)
919 extent_flags |= head->extent_op->flags_to_set;
921 BUG_ON(num_refs == 0);
923 num_refs += head->node.ref_mod;
924 spin_unlock(&head->lock);
925 mutex_unlock(&head->mutex);
927 spin_unlock(&delayed_refs->lock);
929 WARN_ON(num_refs == 0);
933 *flags = extent_flags;
935 btrfs_free_path(path);
940 * Back reference rules. Back refs have three main goals:
942 * 1) differentiate between all holders of references to an extent so that
943 * when a reference is dropped we can make sure it was a valid reference
944 * before freeing the extent.
946 * 2) Provide enough information to quickly find the holders of an extent
947 * if we notice a given block is corrupted or bad.
949 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
950 * maintenance. This is actually the same as #2, but with a slightly
951 * different use case.
953 * There are two kinds of back refs. The implicit back refs is optimized
954 * for pointers in non-shared tree blocks. For a given pointer in a block,
955 * back refs of this kind provide information about the block's owner tree
956 * and the pointer's key. These information allow us to find the block by
957 * b-tree searching. The full back refs is for pointers in tree blocks not
958 * referenced by their owner trees. The location of tree block is recorded
959 * in the back refs. Actually the full back refs is generic, and can be
960 * used in all cases the implicit back refs is used. The major shortcoming
961 * of the full back refs is its overhead. Every time a tree block gets
962 * COWed, we have to update back refs entry for all pointers in it.
964 * For a newly allocated tree block, we use implicit back refs for
965 * pointers in it. This means most tree related operations only involve
966 * implicit back refs. For a tree block created in old transaction, the
967 * only way to drop a reference to it is COW it. So we can detect the
968 * event that tree block loses its owner tree's reference and do the
969 * back refs conversion.
971 * When a tree block is COWed through a tree, there are four cases:
973 * The reference count of the block is one and the tree is the block's
974 * owner tree. Nothing to do in this case.
976 * The reference count of the block is one and the tree is not the
977 * block's owner tree. In this case, full back refs is used for pointers
978 * in the block. Remove these full back refs, add implicit back refs for
979 * every pointers in the new block.
981 * The reference count of the block is greater than one and the tree is
982 * the block's owner tree. In this case, implicit back refs is used for
983 * pointers in the block. Add full back refs for every pointers in the
984 * block, increase lower level extents' reference counts. The original
985 * implicit back refs are entailed to the new block.
987 * The reference count of the block is greater than one and the tree is
988 * not the block's owner tree. Add implicit back refs for every pointer in
989 * the new block, increase lower level extents' reference count.
991 * Back Reference Key composing:
993 * The key objectid corresponds to the first byte in the extent,
994 * The key type is used to differentiate between types of back refs.
995 * There are different meanings of the key offset for different types
998 * File extents can be referenced by:
1000 * - multiple snapshots, subvolumes, or different generations in one subvol
1001 * - different files inside a single subvolume
1002 * - different offsets inside a file (bookend extents in file.c)
1004 * The extent ref structure for the implicit back refs has fields for:
1006 * - Objectid of the subvolume root
1007 * - objectid of the file holding the reference
1008 * - original offset in the file
1009 * - how many bookend extents
1011 * The key offset for the implicit back refs is hash of the first
1014 * The extent ref structure for the full back refs has field for:
1016 * - number of pointers in the tree leaf
1018 * The key offset for the implicit back refs is the first byte of
1021 * When a file extent is allocated, The implicit back refs is used.
1022 * the fields are filled in:
1024 * (root_key.objectid, inode objectid, offset in file, 1)
1026 * When a file extent is removed file truncation, we find the
1027 * corresponding implicit back refs and check the following fields:
1029 * (btrfs_header_owner(leaf), inode objectid, offset in file)
1031 * Btree extents can be referenced by:
1033 * - Different subvolumes
1035 * Both the implicit back refs and the full back refs for tree blocks
1036 * only consist of key. The key offset for the implicit back refs is
1037 * objectid of block's owner tree. The key offset for the full back refs
1038 * is the first byte of parent block.
1040 * When implicit back refs is used, information about the lowest key and
1041 * level of the tree block are required. These information are stored in
1042 * tree block info structure.
1045 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1046 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
1047 struct btrfs_root *root,
1048 struct btrfs_path *path,
1049 u64 owner, u32 extra_size)
1051 struct btrfs_extent_item *item;
1052 struct btrfs_extent_item_v0 *ei0;
1053 struct btrfs_extent_ref_v0 *ref0;
1054 struct btrfs_tree_block_info *bi;
1055 struct extent_buffer *leaf;
1056 struct btrfs_key key;
1057 struct btrfs_key found_key;
1058 u32 new_size = sizeof(*item);
1062 leaf = path->nodes[0];
1063 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
1065 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1066 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1067 struct btrfs_extent_item_v0);
1068 refs = btrfs_extent_refs_v0(leaf, ei0);
1070 if (owner == (u64)-1) {
1072 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1073 ret = btrfs_next_leaf(root, path);
1076 BUG_ON(ret > 0); /* Corruption */
1077 leaf = path->nodes[0];
1079 btrfs_item_key_to_cpu(leaf, &found_key,
1081 BUG_ON(key.objectid != found_key.objectid);
1082 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1086 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1087 struct btrfs_extent_ref_v0);
1088 owner = btrfs_ref_objectid_v0(leaf, ref0);
1092 btrfs_release_path(path);
1094 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1095 new_size += sizeof(*bi);
1097 new_size -= sizeof(*ei0);
1098 ret = btrfs_search_slot(trans, root, &key, path,
1099 new_size + extra_size, 1);
1102 BUG_ON(ret); /* Corruption */
1104 btrfs_extend_item(root, path, new_size);
1106 leaf = path->nodes[0];
1107 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1108 btrfs_set_extent_refs(leaf, item, refs);
1109 /* FIXME: get real generation */
1110 btrfs_set_extent_generation(leaf, item, 0);
1111 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1112 btrfs_set_extent_flags(leaf, item,
1113 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1114 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1115 bi = (struct btrfs_tree_block_info *)(item + 1);
1116 /* FIXME: get first key of the block */
1117 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1118 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1120 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1122 btrfs_mark_buffer_dirty(leaf);
1127 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1129 u32 high_crc = ~(u32)0;
1130 u32 low_crc = ~(u32)0;
1133 lenum = cpu_to_le64(root_objectid);
1134 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
1135 lenum = cpu_to_le64(owner);
1136 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1137 lenum = cpu_to_le64(offset);
1138 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1140 return ((u64)high_crc << 31) ^ (u64)low_crc;
1143 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1144 struct btrfs_extent_data_ref *ref)
1146 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1147 btrfs_extent_data_ref_objectid(leaf, ref),
1148 btrfs_extent_data_ref_offset(leaf, ref));
1151 static int match_extent_data_ref(struct extent_buffer *leaf,
1152 struct btrfs_extent_data_ref *ref,
1153 u64 root_objectid, u64 owner, u64 offset)
1155 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1156 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1157 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1162 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1163 struct btrfs_root *root,
1164 struct btrfs_path *path,
1165 u64 bytenr, u64 parent,
1167 u64 owner, u64 offset)
1169 struct btrfs_key key;
1170 struct btrfs_extent_data_ref *ref;
1171 struct extent_buffer *leaf;
1177 key.objectid = bytenr;
1179 key.type = BTRFS_SHARED_DATA_REF_KEY;
1180 key.offset = parent;
1182 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1183 key.offset = hash_extent_data_ref(root_objectid,
1188 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1197 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1198 key.type = BTRFS_EXTENT_REF_V0_KEY;
1199 btrfs_release_path(path);
1200 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1211 leaf = path->nodes[0];
1212 nritems = btrfs_header_nritems(leaf);
1214 if (path->slots[0] >= nritems) {
1215 ret = btrfs_next_leaf(root, path);
1221 leaf = path->nodes[0];
1222 nritems = btrfs_header_nritems(leaf);
1226 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1227 if (key.objectid != bytenr ||
1228 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1231 ref = btrfs_item_ptr(leaf, path->slots[0],
1232 struct btrfs_extent_data_ref);
1234 if (match_extent_data_ref(leaf, ref, root_objectid,
1237 btrfs_release_path(path);
1249 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1250 struct btrfs_root *root,
1251 struct btrfs_path *path,
1252 u64 bytenr, u64 parent,
1253 u64 root_objectid, u64 owner,
1254 u64 offset, int refs_to_add)
1256 struct btrfs_key key;
1257 struct extent_buffer *leaf;
1262 key.objectid = bytenr;
1264 key.type = BTRFS_SHARED_DATA_REF_KEY;
1265 key.offset = parent;
1266 size = sizeof(struct btrfs_shared_data_ref);
1268 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1269 key.offset = hash_extent_data_ref(root_objectid,
1271 size = sizeof(struct btrfs_extent_data_ref);
1274 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1275 if (ret && ret != -EEXIST)
1278 leaf = path->nodes[0];
1280 struct btrfs_shared_data_ref *ref;
1281 ref = btrfs_item_ptr(leaf, path->slots[0],
1282 struct btrfs_shared_data_ref);
1284 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1286 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1287 num_refs += refs_to_add;
1288 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1291 struct btrfs_extent_data_ref *ref;
1292 while (ret == -EEXIST) {
1293 ref = btrfs_item_ptr(leaf, path->slots[0],
1294 struct btrfs_extent_data_ref);
1295 if (match_extent_data_ref(leaf, ref, root_objectid,
1298 btrfs_release_path(path);
1300 ret = btrfs_insert_empty_item(trans, root, path, &key,
1302 if (ret && ret != -EEXIST)
1305 leaf = path->nodes[0];
1307 ref = btrfs_item_ptr(leaf, path->slots[0],
1308 struct btrfs_extent_data_ref);
1310 btrfs_set_extent_data_ref_root(leaf, ref,
1312 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1313 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1314 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1316 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1317 num_refs += refs_to_add;
1318 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1321 btrfs_mark_buffer_dirty(leaf);
1324 btrfs_release_path(path);
1328 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1329 struct btrfs_root *root,
1330 struct btrfs_path *path,
1331 int refs_to_drop, int *last_ref)
1333 struct btrfs_key key;
1334 struct btrfs_extent_data_ref *ref1 = NULL;
1335 struct btrfs_shared_data_ref *ref2 = NULL;
1336 struct extent_buffer *leaf;
1340 leaf = path->nodes[0];
1341 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1343 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1344 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1345 struct btrfs_extent_data_ref);
1346 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1347 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1348 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1349 struct btrfs_shared_data_ref);
1350 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1351 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1352 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1353 struct btrfs_extent_ref_v0 *ref0;
1354 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1355 struct btrfs_extent_ref_v0);
1356 num_refs = btrfs_ref_count_v0(leaf, ref0);
1362 BUG_ON(num_refs < refs_to_drop);
1363 num_refs -= refs_to_drop;
1365 if (num_refs == 0) {
1366 ret = btrfs_del_item(trans, root, path);
1369 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1370 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1371 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1372 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1373 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1375 struct btrfs_extent_ref_v0 *ref0;
1376 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1377 struct btrfs_extent_ref_v0);
1378 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1381 btrfs_mark_buffer_dirty(leaf);
1386 static noinline u32 extent_data_ref_count(struct btrfs_path *path,
1387 struct btrfs_extent_inline_ref *iref)
1389 struct btrfs_key key;
1390 struct extent_buffer *leaf;
1391 struct btrfs_extent_data_ref *ref1;
1392 struct btrfs_shared_data_ref *ref2;
1395 leaf = path->nodes[0];
1396 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1398 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1399 BTRFS_EXTENT_DATA_REF_KEY) {
1400 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1401 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1403 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1404 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1406 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1407 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1408 struct btrfs_extent_data_ref);
1409 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1410 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1411 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1412 struct btrfs_shared_data_ref);
1413 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1414 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1415 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1416 struct btrfs_extent_ref_v0 *ref0;
1417 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1418 struct btrfs_extent_ref_v0);
1419 num_refs = btrfs_ref_count_v0(leaf, ref0);
1427 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1428 struct btrfs_root *root,
1429 struct btrfs_path *path,
1430 u64 bytenr, u64 parent,
1433 struct btrfs_key key;
1436 key.objectid = bytenr;
1438 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1439 key.offset = parent;
1441 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1442 key.offset = root_objectid;
1445 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1448 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1449 if (ret == -ENOENT && parent) {
1450 btrfs_release_path(path);
1451 key.type = BTRFS_EXTENT_REF_V0_KEY;
1452 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1460 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1461 struct btrfs_root *root,
1462 struct btrfs_path *path,
1463 u64 bytenr, u64 parent,
1466 struct btrfs_key key;
1469 key.objectid = bytenr;
1471 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1472 key.offset = parent;
1474 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1475 key.offset = root_objectid;
1478 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1479 btrfs_release_path(path);
1483 static inline int extent_ref_type(u64 parent, u64 owner)
1486 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1488 type = BTRFS_SHARED_BLOCK_REF_KEY;
1490 type = BTRFS_TREE_BLOCK_REF_KEY;
1493 type = BTRFS_SHARED_DATA_REF_KEY;
1495 type = BTRFS_EXTENT_DATA_REF_KEY;
1500 static int find_next_key(struct btrfs_path *path, int level,
1501 struct btrfs_key *key)
1504 for (; level < BTRFS_MAX_LEVEL; level++) {
1505 if (!path->nodes[level])
1507 if (path->slots[level] + 1 >=
1508 btrfs_header_nritems(path->nodes[level]))
1511 btrfs_item_key_to_cpu(path->nodes[level], key,
1512 path->slots[level] + 1);
1514 btrfs_node_key_to_cpu(path->nodes[level], key,
1515 path->slots[level] + 1);
1522 * look for inline back ref. if back ref is found, *ref_ret is set
1523 * to the address of inline back ref, and 0 is returned.
1525 * if back ref isn't found, *ref_ret is set to the address where it
1526 * should be inserted, and -ENOENT is returned.
1528 * if insert is true and there are too many inline back refs, the path
1529 * points to the extent item, and -EAGAIN is returned.
1531 * NOTE: inline back refs are ordered in the same way that back ref
1532 * items in the tree are ordered.
1534 static noinline_for_stack
1535 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1536 struct btrfs_root *root,
1537 struct btrfs_path *path,
1538 struct btrfs_extent_inline_ref **ref_ret,
1539 u64 bytenr, u64 num_bytes,
1540 u64 parent, u64 root_objectid,
1541 u64 owner, u64 offset, int insert)
1543 struct btrfs_key key;
1544 struct extent_buffer *leaf;
1545 struct btrfs_extent_item *ei;
1546 struct btrfs_extent_inline_ref *iref;
1556 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1559 key.objectid = bytenr;
1560 key.type = BTRFS_EXTENT_ITEM_KEY;
1561 key.offset = num_bytes;
1563 want = extent_ref_type(parent, owner);
1565 extra_size = btrfs_extent_inline_ref_size(want);
1566 path->keep_locks = 1;
1571 * Owner is our parent level, so we can just add one to get the level
1572 * for the block we are interested in.
1574 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1575 key.type = BTRFS_METADATA_ITEM_KEY;
1580 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1587 * We may be a newly converted file system which still has the old fat
1588 * extent entries for metadata, so try and see if we have one of those.
1590 if (ret > 0 && skinny_metadata) {
1591 skinny_metadata = false;
1592 if (path->slots[0]) {
1594 btrfs_item_key_to_cpu(path->nodes[0], &key,
1596 if (key.objectid == bytenr &&
1597 key.type == BTRFS_EXTENT_ITEM_KEY &&
1598 key.offset == num_bytes)
1602 key.objectid = bytenr;
1603 key.type = BTRFS_EXTENT_ITEM_KEY;
1604 key.offset = num_bytes;
1605 btrfs_release_path(path);
1610 if (ret && !insert) {
1613 } else if (WARN_ON(ret)) {
1618 leaf = path->nodes[0];
1619 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1620 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1621 if (item_size < sizeof(*ei)) {
1626 ret = convert_extent_item_v0(trans, root, path, owner,
1632 leaf = path->nodes[0];
1633 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1636 BUG_ON(item_size < sizeof(*ei));
1638 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1639 flags = btrfs_extent_flags(leaf, ei);
1641 ptr = (unsigned long)(ei + 1);
1642 end = (unsigned long)ei + item_size;
1644 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1645 ptr += sizeof(struct btrfs_tree_block_info);
1655 iref = (struct btrfs_extent_inline_ref *)ptr;
1656 type = btrfs_extent_inline_ref_type(leaf, iref);
1660 ptr += btrfs_extent_inline_ref_size(type);
1664 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1665 struct btrfs_extent_data_ref *dref;
1666 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1667 if (match_extent_data_ref(leaf, dref, root_objectid,
1672 if (hash_extent_data_ref_item(leaf, dref) <
1673 hash_extent_data_ref(root_objectid, owner, offset))
1677 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1679 if (parent == ref_offset) {
1683 if (ref_offset < parent)
1686 if (root_objectid == ref_offset) {
1690 if (ref_offset < root_objectid)
1694 ptr += btrfs_extent_inline_ref_size(type);
1696 if (err == -ENOENT && insert) {
1697 if (item_size + extra_size >=
1698 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1703 * To add new inline back ref, we have to make sure
1704 * there is no corresponding back ref item.
1705 * For simplicity, we just do not add new inline back
1706 * ref if there is any kind of item for this block
1708 if (find_next_key(path, 0, &key) == 0 &&
1709 key.objectid == bytenr &&
1710 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1715 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1718 path->keep_locks = 0;
1719 btrfs_unlock_up_safe(path, 1);
1725 * helper to add new inline back ref
1727 static noinline_for_stack
1728 void setup_inline_extent_backref(struct btrfs_root *root,
1729 struct btrfs_path *path,
1730 struct btrfs_extent_inline_ref *iref,
1731 u64 parent, u64 root_objectid,
1732 u64 owner, u64 offset, int refs_to_add,
1733 struct btrfs_delayed_extent_op *extent_op)
1735 struct extent_buffer *leaf;
1736 struct btrfs_extent_item *ei;
1739 unsigned long item_offset;
1744 leaf = path->nodes[0];
1745 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1746 item_offset = (unsigned long)iref - (unsigned long)ei;
1748 type = extent_ref_type(parent, owner);
1749 size = btrfs_extent_inline_ref_size(type);
1751 btrfs_extend_item(root, path, size);
1753 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1754 refs = btrfs_extent_refs(leaf, ei);
1755 refs += refs_to_add;
1756 btrfs_set_extent_refs(leaf, ei, refs);
1758 __run_delayed_extent_op(extent_op, leaf, ei);
1760 ptr = (unsigned long)ei + item_offset;
1761 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1762 if (ptr < end - size)
1763 memmove_extent_buffer(leaf, ptr + size, ptr,
1766 iref = (struct btrfs_extent_inline_ref *)ptr;
1767 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1768 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1769 struct btrfs_extent_data_ref *dref;
1770 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1771 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1772 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1773 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1774 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1775 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1776 struct btrfs_shared_data_ref *sref;
1777 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1778 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1779 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1780 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1781 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1783 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1785 btrfs_mark_buffer_dirty(leaf);
1788 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1789 struct btrfs_root *root,
1790 struct btrfs_path *path,
1791 struct btrfs_extent_inline_ref **ref_ret,
1792 u64 bytenr, u64 num_bytes, u64 parent,
1793 u64 root_objectid, u64 owner, u64 offset)
1797 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1798 bytenr, num_bytes, parent,
1799 root_objectid, owner, offset, 0);
1803 btrfs_release_path(path);
1806 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1807 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1810 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1811 root_objectid, owner, offset);
1817 * helper to update/remove inline back ref
1819 static noinline_for_stack
1820 void update_inline_extent_backref(struct btrfs_root *root,
1821 struct btrfs_path *path,
1822 struct btrfs_extent_inline_ref *iref,
1824 struct btrfs_delayed_extent_op *extent_op,
1827 struct extent_buffer *leaf;
1828 struct btrfs_extent_item *ei;
1829 struct btrfs_extent_data_ref *dref = NULL;
1830 struct btrfs_shared_data_ref *sref = NULL;
1838 leaf = path->nodes[0];
1839 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1840 refs = btrfs_extent_refs(leaf, ei);
1841 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1842 refs += refs_to_mod;
1843 btrfs_set_extent_refs(leaf, ei, refs);
1845 __run_delayed_extent_op(extent_op, leaf, ei);
1847 type = btrfs_extent_inline_ref_type(leaf, iref);
1849 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1850 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1851 refs = btrfs_extent_data_ref_count(leaf, dref);
1852 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1853 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1854 refs = btrfs_shared_data_ref_count(leaf, sref);
1857 BUG_ON(refs_to_mod != -1);
1860 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1861 refs += refs_to_mod;
1864 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1865 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1867 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1870 size = btrfs_extent_inline_ref_size(type);
1871 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1872 ptr = (unsigned long)iref;
1873 end = (unsigned long)ei + item_size;
1874 if (ptr + size < end)
1875 memmove_extent_buffer(leaf, ptr, ptr + size,
1878 btrfs_truncate_item(root, path, item_size, 1);
1880 btrfs_mark_buffer_dirty(leaf);
1883 static noinline_for_stack
1884 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1885 struct btrfs_root *root,
1886 struct btrfs_path *path,
1887 u64 bytenr, u64 num_bytes, u64 parent,
1888 u64 root_objectid, u64 owner,
1889 u64 offset, int refs_to_add,
1890 struct btrfs_delayed_extent_op *extent_op)
1892 struct btrfs_extent_inline_ref *iref;
1895 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1896 bytenr, num_bytes, parent,
1897 root_objectid, owner, offset, 1);
1899 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1900 update_inline_extent_backref(root, path, iref,
1901 refs_to_add, extent_op, NULL);
1902 } else if (ret == -ENOENT) {
1903 setup_inline_extent_backref(root, path, iref, parent,
1904 root_objectid, owner, offset,
1905 refs_to_add, extent_op);
1911 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1912 struct btrfs_root *root,
1913 struct btrfs_path *path,
1914 u64 bytenr, u64 parent, u64 root_objectid,
1915 u64 owner, u64 offset, int refs_to_add)
1918 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1919 BUG_ON(refs_to_add != 1);
1920 ret = insert_tree_block_ref(trans, root, path, bytenr,
1921 parent, root_objectid);
1923 ret = insert_extent_data_ref(trans, root, path, bytenr,
1924 parent, root_objectid,
1925 owner, offset, refs_to_add);
1930 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1931 struct btrfs_root *root,
1932 struct btrfs_path *path,
1933 struct btrfs_extent_inline_ref *iref,
1934 int refs_to_drop, int is_data, int *last_ref)
1938 BUG_ON(!is_data && refs_to_drop != 1);
1940 update_inline_extent_backref(root, path, iref,
1941 -refs_to_drop, NULL, last_ref);
1942 } else if (is_data) {
1943 ret = remove_extent_data_ref(trans, root, path, refs_to_drop,
1947 ret = btrfs_del_item(trans, root, path);
1952 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
1953 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
1954 u64 *discarded_bytes)
1957 u64 bytes_left, end;
1958 u64 aligned_start = ALIGN(start, 1 << 9);
1960 if (WARN_ON(start != aligned_start)) {
1961 len -= aligned_start - start;
1962 len = round_down(len, 1 << 9);
1963 start = aligned_start;
1966 *discarded_bytes = 0;
1974 /* Skip any superblocks on this device. */
1975 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
1976 u64 sb_start = btrfs_sb_offset(j);
1977 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
1978 u64 size = sb_start - start;
1980 if (!in_range(sb_start, start, bytes_left) &&
1981 !in_range(sb_end, start, bytes_left) &&
1982 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
1986 * Superblock spans beginning of range. Adjust start and
1989 if (sb_start <= start) {
1990 start += sb_end - start;
1995 bytes_left = end - start;
2000 ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
2003 *discarded_bytes += size;
2004 else if (ret != -EOPNOTSUPP)
2013 bytes_left = end - start;
2017 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
2020 *discarded_bytes += bytes_left;
2025 int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
2026 u64 num_bytes, u64 *actual_bytes)
2029 u64 discarded_bytes = 0;
2030 struct btrfs_bio *bbio = NULL;
2034 * Avoid races with device replace and make sure our bbio has devices
2035 * associated to its stripes that don't go away while we are discarding.
2037 btrfs_bio_counter_inc_blocked(root->fs_info);
2038 /* Tell the block device(s) that the sectors can be discarded */
2039 ret = btrfs_map_block(root->fs_info, REQ_OP_DISCARD,
2040 bytenr, &num_bytes, &bbio, 0);
2041 /* Error condition is -ENOMEM */
2043 struct btrfs_bio_stripe *stripe = bbio->stripes;
2047 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
2049 if (!stripe->dev->can_discard)
2052 ret = btrfs_issue_discard(stripe->dev->bdev,
2057 discarded_bytes += bytes;
2058 else if (ret != -EOPNOTSUPP)
2059 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2062 * Just in case we get back EOPNOTSUPP for some reason,
2063 * just ignore the return value so we don't screw up
2064 * people calling discard_extent.
2068 btrfs_put_bbio(bbio);
2070 btrfs_bio_counter_dec(root->fs_info);
2073 *actual_bytes = discarded_bytes;
2076 if (ret == -EOPNOTSUPP)
2081 /* Can return -ENOMEM */
2082 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2083 struct btrfs_root *root,
2084 u64 bytenr, u64 num_bytes, u64 parent,
2085 u64 root_objectid, u64 owner, u64 offset)
2088 struct btrfs_fs_info *fs_info = root->fs_info;
2090 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
2091 root_objectid == BTRFS_TREE_LOG_OBJECTID);
2093 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2094 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
2096 parent, root_objectid, (int)owner,
2097 BTRFS_ADD_DELAYED_REF, NULL);
2099 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
2100 num_bytes, parent, root_objectid,
2102 BTRFS_ADD_DELAYED_REF, NULL);
2107 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2108 struct btrfs_root *root,
2109 struct btrfs_delayed_ref_node *node,
2110 u64 parent, u64 root_objectid,
2111 u64 owner, u64 offset, int refs_to_add,
2112 struct btrfs_delayed_extent_op *extent_op)
2114 struct btrfs_fs_info *fs_info = root->fs_info;
2115 struct btrfs_path *path;
2116 struct extent_buffer *leaf;
2117 struct btrfs_extent_item *item;
2118 struct btrfs_key key;
2119 u64 bytenr = node->bytenr;
2120 u64 num_bytes = node->num_bytes;
2124 path = btrfs_alloc_path();
2128 path->reada = READA_FORWARD;
2129 path->leave_spinning = 1;
2130 /* this will setup the path even if it fails to insert the back ref */
2131 ret = insert_inline_extent_backref(trans, fs_info->extent_root, path,
2132 bytenr, num_bytes, parent,
2133 root_objectid, owner, offset,
2134 refs_to_add, extent_op);
2135 if ((ret < 0 && ret != -EAGAIN) || !ret)
2139 * Ok we had -EAGAIN which means we didn't have space to insert and
2140 * inline extent ref, so just update the reference count and add a
2143 leaf = path->nodes[0];
2144 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2145 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2146 refs = btrfs_extent_refs(leaf, item);
2147 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2149 __run_delayed_extent_op(extent_op, leaf, item);
2151 btrfs_mark_buffer_dirty(leaf);
2152 btrfs_release_path(path);
2154 path->reada = READA_FORWARD;
2155 path->leave_spinning = 1;
2156 /* now insert the actual backref */
2157 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2158 path, bytenr, parent, root_objectid,
2159 owner, offset, refs_to_add);
2161 btrfs_abort_transaction(trans, ret);
2163 btrfs_free_path(path);
2167 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2168 struct btrfs_root *root,
2169 struct btrfs_delayed_ref_node *node,
2170 struct btrfs_delayed_extent_op *extent_op,
2171 int insert_reserved)
2174 struct btrfs_delayed_data_ref *ref;
2175 struct btrfs_key ins;
2180 ins.objectid = node->bytenr;
2181 ins.offset = node->num_bytes;
2182 ins.type = BTRFS_EXTENT_ITEM_KEY;
2184 ref = btrfs_delayed_node_to_data_ref(node);
2185 trace_run_delayed_data_ref(root->fs_info, node, ref, node->action);
2187 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2188 parent = ref->parent;
2189 ref_root = ref->root;
2191 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2193 flags |= extent_op->flags_to_set;
2194 ret = alloc_reserved_file_extent(trans, root,
2195 parent, ref_root, flags,
2196 ref->objectid, ref->offset,
2197 &ins, node->ref_mod);
2198 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2199 ret = __btrfs_inc_extent_ref(trans, root, node, parent,
2200 ref_root, ref->objectid,
2201 ref->offset, node->ref_mod,
2203 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2204 ret = __btrfs_free_extent(trans, root, node, parent,
2205 ref_root, ref->objectid,
2206 ref->offset, node->ref_mod,
2214 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2215 struct extent_buffer *leaf,
2216 struct btrfs_extent_item *ei)
2218 u64 flags = btrfs_extent_flags(leaf, ei);
2219 if (extent_op->update_flags) {
2220 flags |= extent_op->flags_to_set;
2221 btrfs_set_extent_flags(leaf, ei, flags);
2224 if (extent_op->update_key) {
2225 struct btrfs_tree_block_info *bi;
2226 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2227 bi = (struct btrfs_tree_block_info *)(ei + 1);
2228 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2232 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2233 struct btrfs_root *root,
2234 struct btrfs_delayed_ref_node *node,
2235 struct btrfs_delayed_extent_op *extent_op)
2237 struct btrfs_key key;
2238 struct btrfs_path *path;
2239 struct btrfs_extent_item *ei;
2240 struct extent_buffer *leaf;
2244 int metadata = !extent_op->is_data;
2249 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2252 path = btrfs_alloc_path();
2256 key.objectid = node->bytenr;
2259 key.type = BTRFS_METADATA_ITEM_KEY;
2260 key.offset = extent_op->level;
2262 key.type = BTRFS_EXTENT_ITEM_KEY;
2263 key.offset = node->num_bytes;
2267 path->reada = READA_FORWARD;
2268 path->leave_spinning = 1;
2269 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2277 if (path->slots[0] > 0) {
2279 btrfs_item_key_to_cpu(path->nodes[0], &key,
2281 if (key.objectid == node->bytenr &&
2282 key.type == BTRFS_EXTENT_ITEM_KEY &&
2283 key.offset == node->num_bytes)
2287 btrfs_release_path(path);
2290 key.objectid = node->bytenr;
2291 key.offset = node->num_bytes;
2292 key.type = BTRFS_EXTENT_ITEM_KEY;
2301 leaf = path->nodes[0];
2302 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2303 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2304 if (item_size < sizeof(*ei)) {
2305 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2311 leaf = path->nodes[0];
2312 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2315 BUG_ON(item_size < sizeof(*ei));
2316 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2317 __run_delayed_extent_op(extent_op, leaf, ei);
2319 btrfs_mark_buffer_dirty(leaf);
2321 btrfs_free_path(path);
2325 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2326 struct btrfs_root *root,
2327 struct btrfs_delayed_ref_node *node,
2328 struct btrfs_delayed_extent_op *extent_op,
2329 int insert_reserved)
2332 struct btrfs_delayed_tree_ref *ref;
2333 struct btrfs_key ins;
2336 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2339 ref = btrfs_delayed_node_to_tree_ref(node);
2340 trace_run_delayed_tree_ref(root->fs_info, node, ref, node->action);
2342 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2343 parent = ref->parent;
2344 ref_root = ref->root;
2346 ins.objectid = node->bytenr;
2347 if (skinny_metadata) {
2348 ins.offset = ref->level;
2349 ins.type = BTRFS_METADATA_ITEM_KEY;
2351 ins.offset = node->num_bytes;
2352 ins.type = BTRFS_EXTENT_ITEM_KEY;
2355 if (node->ref_mod != 1) {
2356 btrfs_err(root->fs_info,
2357 "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu",
2358 node->bytenr, node->ref_mod, node->action, ref_root,
2362 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2363 BUG_ON(!extent_op || !extent_op->update_flags);
2364 ret = alloc_reserved_tree_block(trans, root,
2366 extent_op->flags_to_set,
2369 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2370 ret = __btrfs_inc_extent_ref(trans, root, node,
2374 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2375 ret = __btrfs_free_extent(trans, root, node,
2377 ref->level, 0, 1, extent_op);
2384 /* helper function to actually process a single delayed ref entry */
2385 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2386 struct btrfs_root *root,
2387 struct btrfs_delayed_ref_node *node,
2388 struct btrfs_delayed_extent_op *extent_op,
2389 int insert_reserved)
2393 if (trans->aborted) {
2394 if (insert_reserved)
2395 btrfs_pin_extent(root, node->bytenr,
2396 node->num_bytes, 1);
2400 if (btrfs_delayed_ref_is_head(node)) {
2401 struct btrfs_delayed_ref_head *head;
2403 * we've hit the end of the chain and we were supposed
2404 * to insert this extent into the tree. But, it got
2405 * deleted before we ever needed to insert it, so all
2406 * we have to do is clean up the accounting
2409 head = btrfs_delayed_node_to_head(node);
2410 trace_run_delayed_ref_head(root->fs_info, node, head,
2413 if (insert_reserved) {
2414 btrfs_pin_extent(root, node->bytenr,
2415 node->num_bytes, 1);
2416 if (head->is_data) {
2417 ret = btrfs_del_csums(trans, root,
2423 /* Also free its reserved qgroup space */
2424 btrfs_qgroup_free_delayed_ref(root->fs_info,
2425 head->qgroup_ref_root,
2426 head->qgroup_reserved);
2430 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2431 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2432 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2434 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2435 node->type == BTRFS_SHARED_DATA_REF_KEY)
2436 ret = run_delayed_data_ref(trans, root, node, extent_op,
2443 static inline struct btrfs_delayed_ref_node *
2444 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2446 struct btrfs_delayed_ref_node *ref;
2448 if (list_empty(&head->ref_list))
2452 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2453 * This is to prevent a ref count from going down to zero, which deletes
2454 * the extent item from the extent tree, when there still are references
2455 * to add, which would fail because they would not find the extent item.
2457 list_for_each_entry(ref, &head->ref_list, list) {
2458 if (ref->action == BTRFS_ADD_DELAYED_REF)
2462 return list_entry(head->ref_list.next, struct btrfs_delayed_ref_node,
2467 * Returns 0 on success or if called with an already aborted transaction.
2468 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2470 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2471 struct btrfs_root *root,
2474 struct btrfs_delayed_ref_root *delayed_refs;
2475 struct btrfs_delayed_ref_node *ref;
2476 struct btrfs_delayed_ref_head *locked_ref = NULL;
2477 struct btrfs_delayed_extent_op *extent_op;
2478 struct btrfs_fs_info *fs_info = root->fs_info;
2479 ktime_t start = ktime_get();
2481 unsigned long count = 0;
2482 unsigned long actual_count = 0;
2483 int must_insert_reserved = 0;
2485 delayed_refs = &trans->transaction->delayed_refs;
2491 spin_lock(&delayed_refs->lock);
2492 locked_ref = btrfs_select_ref_head(trans);
2494 spin_unlock(&delayed_refs->lock);
2498 /* grab the lock that says we are going to process
2499 * all the refs for this head */
2500 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2501 spin_unlock(&delayed_refs->lock);
2503 * we may have dropped the spin lock to get the head
2504 * mutex lock, and that might have given someone else
2505 * time to free the head. If that's true, it has been
2506 * removed from our list and we can move on.
2508 if (ret == -EAGAIN) {
2516 * We need to try and merge add/drops of the same ref since we
2517 * can run into issues with relocate dropping the implicit ref
2518 * and then it being added back again before the drop can
2519 * finish. If we merged anything we need to re-loop so we can
2521 * Or we can get node references of the same type that weren't
2522 * merged when created due to bumps in the tree mod seq, and
2523 * we need to merge them to prevent adding an inline extent
2524 * backref before dropping it (triggering a BUG_ON at
2525 * insert_inline_extent_backref()).
2527 spin_lock(&locked_ref->lock);
2528 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2532 * locked_ref is the head node, so we have to go one
2533 * node back for any delayed ref updates
2535 ref = select_delayed_ref(locked_ref);
2537 if (ref && ref->seq &&
2538 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2539 spin_unlock(&locked_ref->lock);
2540 spin_lock(&delayed_refs->lock);
2541 locked_ref->processing = 0;
2542 delayed_refs->num_heads_ready++;
2543 spin_unlock(&delayed_refs->lock);
2544 btrfs_delayed_ref_unlock(locked_ref);
2552 * record the must insert reserved flag before we
2553 * drop the spin lock.
2555 must_insert_reserved = locked_ref->must_insert_reserved;
2556 locked_ref->must_insert_reserved = 0;
2558 extent_op = locked_ref->extent_op;
2559 locked_ref->extent_op = NULL;
2564 /* All delayed refs have been processed, Go ahead
2565 * and send the head node to run_one_delayed_ref,
2566 * so that any accounting fixes can happen
2568 ref = &locked_ref->node;
2570 if (extent_op && must_insert_reserved) {
2571 btrfs_free_delayed_extent_op(extent_op);
2576 spin_unlock(&locked_ref->lock);
2577 ret = run_delayed_extent_op(trans, root,
2579 btrfs_free_delayed_extent_op(extent_op);
2583 * Need to reset must_insert_reserved if
2584 * there was an error so the abort stuff
2585 * can cleanup the reserved space
2588 if (must_insert_reserved)
2589 locked_ref->must_insert_reserved = 1;
2590 spin_lock(&delayed_refs->lock);
2591 locked_ref->processing = 0;
2592 delayed_refs->num_heads_ready++;
2593 spin_unlock(&delayed_refs->lock);
2594 btrfs_debug(fs_info,
2595 "run_delayed_extent_op returned %d",
2597 btrfs_delayed_ref_unlock(locked_ref);
2604 * Need to drop our head ref lock and re-acquire the
2605 * delayed ref lock and then re-check to make sure
2608 spin_unlock(&locked_ref->lock);
2609 spin_lock(&delayed_refs->lock);
2610 spin_lock(&locked_ref->lock);
2611 if (!list_empty(&locked_ref->ref_list) ||
2612 locked_ref->extent_op) {
2613 spin_unlock(&locked_ref->lock);
2614 spin_unlock(&delayed_refs->lock);
2618 delayed_refs->num_heads--;
2619 rb_erase(&locked_ref->href_node,
2620 &delayed_refs->href_root);
2621 spin_unlock(&delayed_refs->lock);
2625 list_del(&ref->list);
2627 atomic_dec(&delayed_refs->num_entries);
2629 if (!btrfs_delayed_ref_is_head(ref)) {
2631 * when we play the delayed ref, also correct the
2634 switch (ref->action) {
2635 case BTRFS_ADD_DELAYED_REF:
2636 case BTRFS_ADD_DELAYED_EXTENT:
2637 locked_ref->node.ref_mod -= ref->ref_mod;
2639 case BTRFS_DROP_DELAYED_REF:
2640 locked_ref->node.ref_mod += ref->ref_mod;
2646 spin_unlock(&locked_ref->lock);
2648 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2649 must_insert_reserved);
2651 btrfs_free_delayed_extent_op(extent_op);
2653 spin_lock(&delayed_refs->lock);
2654 locked_ref->processing = 0;
2655 delayed_refs->num_heads_ready++;
2656 spin_unlock(&delayed_refs->lock);
2657 btrfs_delayed_ref_unlock(locked_ref);
2658 btrfs_put_delayed_ref(ref);
2659 btrfs_debug(fs_info, "run_one_delayed_ref returned %d",
2665 * If this node is a head, that means all the refs in this head
2666 * have been dealt with, and we will pick the next head to deal
2667 * with, so we must unlock the head and drop it from the cluster
2668 * list before we release it.
2670 if (btrfs_delayed_ref_is_head(ref)) {
2671 if (locked_ref->is_data &&
2672 locked_ref->total_ref_mod < 0) {
2673 spin_lock(&delayed_refs->lock);
2674 delayed_refs->pending_csums -= ref->num_bytes;
2675 spin_unlock(&delayed_refs->lock);
2677 btrfs_delayed_ref_unlock(locked_ref);
2680 btrfs_put_delayed_ref(ref);
2686 * We don't want to include ref heads since we can have empty ref heads
2687 * and those will drastically skew our runtime down since we just do
2688 * accounting, no actual extent tree updates.
2690 if (actual_count > 0) {
2691 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2695 * We weigh the current average higher than our current runtime
2696 * to avoid large swings in the average.
2698 spin_lock(&delayed_refs->lock);
2699 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2700 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2701 spin_unlock(&delayed_refs->lock);
2706 #ifdef SCRAMBLE_DELAYED_REFS
2708 * Normally delayed refs get processed in ascending bytenr order. This
2709 * correlates in most cases to the order added. To expose dependencies on this
2710 * order, we start to process the tree in the middle instead of the beginning
2712 static u64 find_middle(struct rb_root *root)
2714 struct rb_node *n = root->rb_node;
2715 struct btrfs_delayed_ref_node *entry;
2718 u64 first = 0, last = 0;
2722 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2723 first = entry->bytenr;
2727 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2728 last = entry->bytenr;
2733 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2734 WARN_ON(!entry->in_tree);
2736 middle = entry->bytenr;
2749 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2753 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2754 sizeof(struct btrfs_extent_inline_ref));
2755 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2756 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2759 * We don't ever fill up leaves all the way so multiply by 2 just to be
2760 * closer to what we're really going to want to use.
2762 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2766 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2767 * would require to store the csums for that many bytes.
2769 u64 btrfs_csum_bytes_to_leaves(struct btrfs_root *root, u64 csum_bytes)
2772 u64 num_csums_per_leaf;
2775 csum_size = BTRFS_MAX_ITEM_SIZE(root);
2776 num_csums_per_leaf = div64_u64(csum_size,
2777 (u64)btrfs_super_csum_size(root->fs_info->super_copy));
2778 num_csums = div64_u64(csum_bytes, root->sectorsize);
2779 num_csums += num_csums_per_leaf - 1;
2780 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2784 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2785 struct btrfs_root *root)
2787 struct btrfs_block_rsv *global_rsv;
2788 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2789 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2790 u64 num_dirty_bgs = trans->transaction->num_dirty_bgs;
2791 u64 num_bytes, num_dirty_bgs_bytes;
2794 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2795 num_heads = heads_to_leaves(root, num_heads);
2797 num_bytes += (num_heads - 1) * root->nodesize;
2799 num_bytes += btrfs_csum_bytes_to_leaves(root, csum_bytes) * root->nodesize;
2800 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(root,
2802 global_rsv = &root->fs_info->global_block_rsv;
2805 * If we can't allocate any more chunks lets make sure we have _lots_ of
2806 * wiggle room since running delayed refs can create more delayed refs.
2808 if (global_rsv->space_info->full) {
2809 num_dirty_bgs_bytes <<= 1;
2813 spin_lock(&global_rsv->lock);
2814 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2816 spin_unlock(&global_rsv->lock);
2820 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2821 struct btrfs_root *root)
2823 struct btrfs_fs_info *fs_info = root->fs_info;
2825 atomic_read(&trans->transaction->delayed_refs.num_entries);
2830 avg_runtime = fs_info->avg_delayed_ref_runtime;
2831 val = num_entries * avg_runtime;
2832 if (num_entries * avg_runtime >= NSEC_PER_SEC)
2834 if (val >= NSEC_PER_SEC / 2)
2837 return btrfs_check_space_for_delayed_refs(trans, root);
2840 struct async_delayed_refs {
2841 struct btrfs_root *root;
2846 struct completion wait;
2847 struct btrfs_work work;
2850 static void delayed_ref_async_start(struct btrfs_work *work)
2852 struct async_delayed_refs *async;
2853 struct btrfs_trans_handle *trans;
2856 async = container_of(work, struct async_delayed_refs, work);
2858 /* if the commit is already started, we don't need to wait here */
2859 if (btrfs_transaction_blocked(async->root->fs_info))
2862 trans = btrfs_join_transaction(async->root);
2863 if (IS_ERR(trans)) {
2864 async->error = PTR_ERR(trans);
2869 * trans->sync means that when we call end_transaction, we won't
2870 * wait on delayed refs
2874 /* Don't bother flushing if we got into a different transaction */
2875 if (trans->transid > async->transid)
2878 ret = btrfs_run_delayed_refs(trans, async->root, async->count);
2882 ret = btrfs_end_transaction(trans, async->root);
2883 if (ret && !async->error)
2887 complete(&async->wait);
2892 int btrfs_async_run_delayed_refs(struct btrfs_root *root,
2893 unsigned long count, u64 transid, int wait)
2895 struct async_delayed_refs *async;
2898 async = kmalloc(sizeof(*async), GFP_NOFS);
2902 async->root = root->fs_info->tree_root;
2903 async->count = count;
2905 async->transid = transid;
2910 init_completion(&async->wait);
2912 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2913 delayed_ref_async_start, NULL, NULL);
2915 btrfs_queue_work(root->fs_info->extent_workers, &async->work);
2918 wait_for_completion(&async->wait);
2927 * this starts processing the delayed reference count updates and
2928 * extent insertions we have queued up so far. count can be
2929 * 0, which means to process everything in the tree at the start
2930 * of the run (but not newly added entries), or it can be some target
2931 * number you'd like to process.
2933 * Returns 0 on success or if called with an aborted transaction
2934 * Returns <0 on error and aborts the transaction
2936 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2937 struct btrfs_root *root, unsigned long count)
2939 struct rb_node *node;
2940 struct btrfs_delayed_ref_root *delayed_refs;
2941 struct btrfs_delayed_ref_head *head;
2943 int run_all = count == (unsigned long)-1;
2944 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
2946 /* We'll clean this up in btrfs_cleanup_transaction */
2950 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &root->fs_info->flags))
2953 if (root == root->fs_info->extent_root)
2954 root = root->fs_info->tree_root;
2956 delayed_refs = &trans->transaction->delayed_refs;
2958 count = atomic_read(&delayed_refs->num_entries) * 2;
2961 #ifdef SCRAMBLE_DELAYED_REFS
2962 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2964 trans->can_flush_pending_bgs = false;
2965 ret = __btrfs_run_delayed_refs(trans, root, count);
2967 btrfs_abort_transaction(trans, ret);
2972 if (!list_empty(&trans->new_bgs))
2973 btrfs_create_pending_block_groups(trans, root);
2975 spin_lock(&delayed_refs->lock);
2976 node = rb_first(&delayed_refs->href_root);
2978 spin_unlock(&delayed_refs->lock);
2983 head = rb_entry(node, struct btrfs_delayed_ref_head,
2985 if (btrfs_delayed_ref_is_head(&head->node)) {
2986 struct btrfs_delayed_ref_node *ref;
2989 atomic_inc(&ref->refs);
2991 spin_unlock(&delayed_refs->lock);
2993 * Mutex was contended, block until it's
2994 * released and try again
2996 mutex_lock(&head->mutex);
2997 mutex_unlock(&head->mutex);
2999 btrfs_put_delayed_ref(ref);
3005 node = rb_next(node);
3007 spin_unlock(&delayed_refs->lock);
3012 assert_qgroups_uptodate(trans);
3013 trans->can_flush_pending_bgs = can_flush_pending_bgs;
3017 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
3018 struct btrfs_root *root,
3019 u64 bytenr, u64 num_bytes, u64 flags,
3020 int level, int is_data)
3022 struct btrfs_delayed_extent_op *extent_op;
3025 extent_op = btrfs_alloc_delayed_extent_op();
3029 extent_op->flags_to_set = flags;
3030 extent_op->update_flags = true;
3031 extent_op->update_key = false;
3032 extent_op->is_data = is_data ? true : false;
3033 extent_op->level = level;
3035 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
3036 num_bytes, extent_op);
3038 btrfs_free_delayed_extent_op(extent_op);
3042 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
3043 struct btrfs_root *root,
3044 struct btrfs_path *path,
3045 u64 objectid, u64 offset, u64 bytenr)
3047 struct btrfs_delayed_ref_head *head;
3048 struct btrfs_delayed_ref_node *ref;
3049 struct btrfs_delayed_data_ref *data_ref;
3050 struct btrfs_delayed_ref_root *delayed_refs;
3053 delayed_refs = &trans->transaction->delayed_refs;
3054 spin_lock(&delayed_refs->lock);
3055 head = btrfs_find_delayed_ref_head(trans, bytenr);
3057 spin_unlock(&delayed_refs->lock);
3061 if (!mutex_trylock(&head->mutex)) {
3062 atomic_inc(&head->node.refs);
3063 spin_unlock(&delayed_refs->lock);
3065 btrfs_release_path(path);
3068 * Mutex was contended, block until it's released and let
3071 mutex_lock(&head->mutex);
3072 mutex_unlock(&head->mutex);
3073 btrfs_put_delayed_ref(&head->node);
3076 spin_unlock(&delayed_refs->lock);
3078 spin_lock(&head->lock);
3079 list_for_each_entry(ref, &head->ref_list, list) {
3080 /* If it's a shared ref we know a cross reference exists */
3081 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
3086 data_ref = btrfs_delayed_node_to_data_ref(ref);
3089 * If our ref doesn't match the one we're currently looking at
3090 * then we have a cross reference.
3092 if (data_ref->root != root->root_key.objectid ||
3093 data_ref->objectid != objectid ||
3094 data_ref->offset != offset) {
3099 spin_unlock(&head->lock);
3100 mutex_unlock(&head->mutex);
3104 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
3105 struct btrfs_root *root,
3106 struct btrfs_path *path,
3107 u64 objectid, u64 offset, u64 bytenr)
3109 struct btrfs_root *extent_root = root->fs_info->extent_root;
3110 struct extent_buffer *leaf;
3111 struct btrfs_extent_data_ref *ref;
3112 struct btrfs_extent_inline_ref *iref;
3113 struct btrfs_extent_item *ei;
3114 struct btrfs_key key;
3118 key.objectid = bytenr;
3119 key.offset = (u64)-1;
3120 key.type = BTRFS_EXTENT_ITEM_KEY;
3122 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3125 BUG_ON(ret == 0); /* Corruption */
3128 if (path->slots[0] == 0)
3132 leaf = path->nodes[0];
3133 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3135 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3139 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3140 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3141 if (item_size < sizeof(*ei)) {
3142 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3146 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3148 if (item_size != sizeof(*ei) +
3149 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3152 if (btrfs_extent_generation(leaf, ei) <=
3153 btrfs_root_last_snapshot(&root->root_item))
3156 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3157 if (btrfs_extent_inline_ref_type(leaf, iref) !=
3158 BTRFS_EXTENT_DATA_REF_KEY)
3161 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3162 if (btrfs_extent_refs(leaf, ei) !=
3163 btrfs_extent_data_ref_count(leaf, ref) ||
3164 btrfs_extent_data_ref_root(leaf, ref) !=
3165 root->root_key.objectid ||
3166 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3167 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3175 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
3176 struct btrfs_root *root,
3177 u64 objectid, u64 offset, u64 bytenr)
3179 struct btrfs_path *path;
3183 path = btrfs_alloc_path();
3188 ret = check_committed_ref(trans, root, path, objectid,
3190 if (ret && ret != -ENOENT)
3193 ret2 = check_delayed_ref(trans, root, path, objectid,
3195 } while (ret2 == -EAGAIN);
3197 if (ret2 && ret2 != -ENOENT) {
3202 if (ret != -ENOENT || ret2 != -ENOENT)
3205 btrfs_free_path(path);
3206 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3211 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3212 struct btrfs_root *root,
3213 struct extent_buffer *buf,
3214 int full_backref, int inc)
3221 struct btrfs_key key;
3222 struct btrfs_file_extent_item *fi;
3226 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3227 u64, u64, u64, u64, u64, u64);
3230 if (btrfs_is_testing(root->fs_info))
3233 ref_root = btrfs_header_owner(buf);
3234 nritems = btrfs_header_nritems(buf);
3235 level = btrfs_header_level(buf);
3237 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3241 process_func = btrfs_inc_extent_ref;
3243 process_func = btrfs_free_extent;
3246 parent = buf->start;
3250 for (i = 0; i < nritems; i++) {
3252 btrfs_item_key_to_cpu(buf, &key, i);
3253 if (key.type != BTRFS_EXTENT_DATA_KEY)
3255 fi = btrfs_item_ptr(buf, i,
3256 struct btrfs_file_extent_item);
3257 if (btrfs_file_extent_type(buf, fi) ==
3258 BTRFS_FILE_EXTENT_INLINE)
3260 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3264 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3265 key.offset -= btrfs_file_extent_offset(buf, fi);
3266 ret = process_func(trans, root, bytenr, num_bytes,
3267 parent, ref_root, key.objectid,
3272 bytenr = btrfs_node_blockptr(buf, i);
3273 num_bytes = root->nodesize;
3274 ret = process_func(trans, root, bytenr, num_bytes,
3275 parent, ref_root, level - 1, 0);
3285 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3286 struct extent_buffer *buf, int full_backref)
3288 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3291 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3292 struct extent_buffer *buf, int full_backref)
3294 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3297 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3298 struct btrfs_root *root,
3299 struct btrfs_path *path,
3300 struct btrfs_block_group_cache *cache)
3303 struct btrfs_root *extent_root = root->fs_info->extent_root;
3305 struct extent_buffer *leaf;
3307 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3314 leaf = path->nodes[0];
3315 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3316 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3317 btrfs_mark_buffer_dirty(leaf);
3319 btrfs_release_path(path);
3324 static struct btrfs_block_group_cache *
3325 next_block_group(struct btrfs_root *root,
3326 struct btrfs_block_group_cache *cache)
3328 struct rb_node *node;
3330 spin_lock(&root->fs_info->block_group_cache_lock);
3332 /* If our block group was removed, we need a full search. */
3333 if (RB_EMPTY_NODE(&cache->cache_node)) {
3334 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3336 spin_unlock(&root->fs_info->block_group_cache_lock);
3337 btrfs_put_block_group(cache);
3338 cache = btrfs_lookup_first_block_group(root->fs_info,
3342 node = rb_next(&cache->cache_node);
3343 btrfs_put_block_group(cache);
3345 cache = rb_entry(node, struct btrfs_block_group_cache,
3347 btrfs_get_block_group(cache);
3350 spin_unlock(&root->fs_info->block_group_cache_lock);
3354 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3355 struct btrfs_trans_handle *trans,
3356 struct btrfs_path *path)
3358 struct btrfs_root *root = block_group->fs_info->tree_root;
3359 struct inode *inode = NULL;
3361 int dcs = BTRFS_DC_ERROR;
3367 * If this block group is smaller than 100 megs don't bother caching the
3370 if (block_group->key.offset < (100 * SZ_1M)) {
3371 spin_lock(&block_group->lock);
3372 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3373 spin_unlock(&block_group->lock);
3380 inode = lookup_free_space_inode(root, block_group, path);
3381 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3382 ret = PTR_ERR(inode);
3383 btrfs_release_path(path);
3387 if (IS_ERR(inode)) {
3391 if (block_group->ro)
3394 ret = create_free_space_inode(root, trans, block_group, path);
3401 * We want to set the generation to 0, that way if anything goes wrong
3402 * from here on out we know not to trust this cache when we load up next
3405 BTRFS_I(inode)->generation = 0;
3406 ret = btrfs_update_inode(trans, root, inode);
3409 * So theoretically we could recover from this, simply set the
3410 * super cache generation to 0 so we know to invalidate the
3411 * cache, but then we'd have to keep track of the block groups
3412 * that fail this way so we know we _have_ to reset this cache
3413 * before the next commit or risk reading stale cache. So to
3414 * limit our exposure to horrible edge cases lets just abort the
3415 * transaction, this only happens in really bad situations
3418 btrfs_abort_transaction(trans, ret);
3423 /* We've already setup this transaction, go ahead and exit */
3424 if (block_group->cache_generation == trans->transid &&
3425 i_size_read(inode)) {
3426 dcs = BTRFS_DC_SETUP;
3430 if (i_size_read(inode) > 0) {
3431 ret = btrfs_check_trunc_cache_free_space(root,
3432 &root->fs_info->global_block_rsv);
3436 ret = btrfs_truncate_free_space_cache(root, trans, NULL, inode);
3441 spin_lock(&block_group->lock);
3442 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3443 !btrfs_test_opt(root->fs_info, SPACE_CACHE)) {
3445 * don't bother trying to write stuff out _if_
3446 * a) we're not cached,
3447 * b) we're with nospace_cache mount option.
3449 dcs = BTRFS_DC_WRITTEN;
3450 spin_unlock(&block_group->lock);
3453 spin_unlock(&block_group->lock);
3456 * We hit an ENOSPC when setting up the cache in this transaction, just
3457 * skip doing the setup, we've already cleared the cache so we're safe.
3459 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
3465 * Try to preallocate enough space based on how big the block group is.
3466 * Keep in mind this has to include any pinned space which could end up
3467 * taking up quite a bit since it's not folded into the other space
3470 num_pages = div_u64(block_group->key.offset, SZ_256M);
3475 num_pages *= PAGE_SIZE;
3477 ret = btrfs_check_data_free_space(inode, 0, num_pages);
3481 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3482 num_pages, num_pages,
3485 * Our cache requires contiguous chunks so that we don't modify a bunch
3486 * of metadata or split extents when writing the cache out, which means
3487 * we can enospc if we are heavily fragmented in addition to just normal
3488 * out of space conditions. So if we hit this just skip setting up any
3489 * other block groups for this transaction, maybe we'll unpin enough
3490 * space the next time around.
3493 dcs = BTRFS_DC_SETUP;
3494 else if (ret == -ENOSPC)
3495 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
3500 btrfs_release_path(path);
3502 spin_lock(&block_group->lock);
3503 if (!ret && dcs == BTRFS_DC_SETUP)
3504 block_group->cache_generation = trans->transid;
3505 block_group->disk_cache_state = dcs;
3506 spin_unlock(&block_group->lock);
3511 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3512 struct btrfs_root *root)
3514 struct btrfs_block_group_cache *cache, *tmp;
3515 struct btrfs_transaction *cur_trans = trans->transaction;
3516 struct btrfs_path *path;
3518 if (list_empty(&cur_trans->dirty_bgs) ||
3519 !btrfs_test_opt(root->fs_info, SPACE_CACHE))
3522 path = btrfs_alloc_path();
3526 /* Could add new block groups, use _safe just in case */
3527 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3529 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3530 cache_save_setup(cache, trans, path);
3533 btrfs_free_path(path);
3538 * transaction commit does final block group cache writeback during a
3539 * critical section where nothing is allowed to change the FS. This is
3540 * required in order for the cache to actually match the block group,
3541 * but can introduce a lot of latency into the commit.
3543 * So, btrfs_start_dirty_block_groups is here to kick off block group
3544 * cache IO. There's a chance we'll have to redo some of it if the
3545 * block group changes again during the commit, but it greatly reduces
3546 * the commit latency by getting rid of the easy block groups while
3547 * we're still allowing others to join the commit.
3549 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans,
3550 struct btrfs_root *root)
3552 struct btrfs_block_group_cache *cache;
3553 struct btrfs_transaction *cur_trans = trans->transaction;
3556 struct btrfs_path *path = NULL;
3558 struct list_head *io = &cur_trans->io_bgs;
3559 int num_started = 0;
3562 spin_lock(&cur_trans->dirty_bgs_lock);
3563 if (list_empty(&cur_trans->dirty_bgs)) {
3564 spin_unlock(&cur_trans->dirty_bgs_lock);
3567 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3568 spin_unlock(&cur_trans->dirty_bgs_lock);
3572 * make sure all the block groups on our dirty list actually
3575 btrfs_create_pending_block_groups(trans, root);
3578 path = btrfs_alloc_path();
3584 * cache_write_mutex is here only to save us from balance or automatic
3585 * removal of empty block groups deleting this block group while we are
3586 * writing out the cache
3588 mutex_lock(&trans->transaction->cache_write_mutex);
3589 while (!list_empty(&dirty)) {
3590 cache = list_first_entry(&dirty,
3591 struct btrfs_block_group_cache,
3594 * this can happen if something re-dirties a block
3595 * group that is already under IO. Just wait for it to
3596 * finish and then do it all again
3598 if (!list_empty(&cache->io_list)) {
3599 list_del_init(&cache->io_list);
3600 btrfs_wait_cache_io(root, trans, cache,
3601 &cache->io_ctl, path,
3602 cache->key.objectid);
3603 btrfs_put_block_group(cache);
3608 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3609 * if it should update the cache_state. Don't delete
3610 * until after we wait.
3612 * Since we're not running in the commit critical section
3613 * we need the dirty_bgs_lock to protect from update_block_group
3615 spin_lock(&cur_trans->dirty_bgs_lock);
3616 list_del_init(&cache->dirty_list);
3617 spin_unlock(&cur_trans->dirty_bgs_lock);
3621 cache_save_setup(cache, trans, path);
3623 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3624 cache->io_ctl.inode = NULL;
3625 ret = btrfs_write_out_cache(root, trans, cache, path);
3626 if (ret == 0 && cache->io_ctl.inode) {
3631 * the cache_write_mutex is protecting
3634 list_add_tail(&cache->io_list, io);
3637 * if we failed to write the cache, the
3638 * generation will be bad and life goes on
3644 ret = write_one_cache_group(trans, root, path, cache);
3646 * Our block group might still be attached to the list
3647 * of new block groups in the transaction handle of some
3648 * other task (struct btrfs_trans_handle->new_bgs). This
3649 * means its block group item isn't yet in the extent
3650 * tree. If this happens ignore the error, as we will
3651 * try again later in the critical section of the
3652 * transaction commit.
3654 if (ret == -ENOENT) {
3656 spin_lock(&cur_trans->dirty_bgs_lock);
3657 if (list_empty(&cache->dirty_list)) {
3658 list_add_tail(&cache->dirty_list,
3659 &cur_trans->dirty_bgs);
3660 btrfs_get_block_group(cache);
3662 spin_unlock(&cur_trans->dirty_bgs_lock);
3664 btrfs_abort_transaction(trans, ret);
3668 /* if its not on the io list, we need to put the block group */
3670 btrfs_put_block_group(cache);
3676 * Avoid blocking other tasks for too long. It might even save
3677 * us from writing caches for block groups that are going to be
3680 mutex_unlock(&trans->transaction->cache_write_mutex);
3681 mutex_lock(&trans->transaction->cache_write_mutex);
3683 mutex_unlock(&trans->transaction->cache_write_mutex);
3686 * go through delayed refs for all the stuff we've just kicked off
3687 * and then loop back (just once)
3689 ret = btrfs_run_delayed_refs(trans, root, 0);
3690 if (!ret && loops == 0) {
3692 spin_lock(&cur_trans->dirty_bgs_lock);
3693 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3695 * dirty_bgs_lock protects us from concurrent block group
3696 * deletes too (not just cache_write_mutex).
3698 if (!list_empty(&dirty)) {
3699 spin_unlock(&cur_trans->dirty_bgs_lock);
3702 spin_unlock(&cur_trans->dirty_bgs_lock);
3703 } else if (ret < 0) {
3704 btrfs_cleanup_dirty_bgs(cur_trans, root);
3707 btrfs_free_path(path);
3711 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3712 struct btrfs_root *root)
3714 struct btrfs_block_group_cache *cache;
3715 struct btrfs_transaction *cur_trans = trans->transaction;
3718 struct btrfs_path *path;
3719 struct list_head *io = &cur_trans->io_bgs;
3720 int num_started = 0;
3722 path = btrfs_alloc_path();
3727 * Even though we are in the critical section of the transaction commit,
3728 * we can still have concurrent tasks adding elements to this
3729 * transaction's list of dirty block groups. These tasks correspond to
3730 * endio free space workers started when writeback finishes for a
3731 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3732 * allocate new block groups as a result of COWing nodes of the root
3733 * tree when updating the free space inode. The writeback for the space
3734 * caches is triggered by an earlier call to
3735 * btrfs_start_dirty_block_groups() and iterations of the following
3737 * Also we want to do the cache_save_setup first and then run the
3738 * delayed refs to make sure we have the best chance at doing this all
3741 spin_lock(&cur_trans->dirty_bgs_lock);
3742 while (!list_empty(&cur_trans->dirty_bgs)) {
3743 cache = list_first_entry(&cur_trans->dirty_bgs,
3744 struct btrfs_block_group_cache,
3748 * this can happen if cache_save_setup re-dirties a block
3749 * group that is already under IO. Just wait for it to
3750 * finish and then do it all again
3752 if (!list_empty(&cache->io_list)) {
3753 spin_unlock(&cur_trans->dirty_bgs_lock);
3754 list_del_init(&cache->io_list);
3755 btrfs_wait_cache_io(root, trans, cache,
3756 &cache->io_ctl, path,
3757 cache->key.objectid);
3758 btrfs_put_block_group(cache);
3759 spin_lock(&cur_trans->dirty_bgs_lock);
3763 * don't remove from the dirty list until after we've waited
3766 list_del_init(&cache->dirty_list);
3767 spin_unlock(&cur_trans->dirty_bgs_lock);
3770 cache_save_setup(cache, trans, path);
3773 ret = btrfs_run_delayed_refs(trans, root, (unsigned long) -1);
3775 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3776 cache->io_ctl.inode = NULL;
3777 ret = btrfs_write_out_cache(root, trans, cache, path);
3778 if (ret == 0 && cache->io_ctl.inode) {
3781 list_add_tail(&cache->io_list, io);
3784 * if we failed to write the cache, the
3785 * generation will be bad and life goes on
3791 ret = write_one_cache_group(trans, root, path, cache);
3793 * One of the free space endio workers might have
3794 * created a new block group while updating a free space
3795 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3796 * and hasn't released its transaction handle yet, in
3797 * which case the new block group is still attached to
3798 * its transaction handle and its creation has not
3799 * finished yet (no block group item in the extent tree
3800 * yet, etc). If this is the case, wait for all free
3801 * space endio workers to finish and retry. This is a
3802 * a very rare case so no need for a more efficient and
3805 if (ret == -ENOENT) {
3806 wait_event(cur_trans->writer_wait,
3807 atomic_read(&cur_trans->num_writers) == 1);
3808 ret = write_one_cache_group(trans, root, path,
3812 btrfs_abort_transaction(trans, ret);
3815 /* if its not on the io list, we need to put the block group */
3817 btrfs_put_block_group(cache);
3818 spin_lock(&cur_trans->dirty_bgs_lock);
3820 spin_unlock(&cur_trans->dirty_bgs_lock);
3822 while (!list_empty(io)) {
3823 cache = list_first_entry(io, struct btrfs_block_group_cache,
3825 list_del_init(&cache->io_list);
3826 btrfs_wait_cache_io(root, trans, cache,
3827 &cache->io_ctl, path, cache->key.objectid);
3828 btrfs_put_block_group(cache);
3831 btrfs_free_path(path);
3835 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3837 struct btrfs_block_group_cache *block_group;
3840 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3841 if (!block_group || block_group->ro)
3844 btrfs_put_block_group(block_group);
3848 bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3850 struct btrfs_block_group_cache *bg;
3853 bg = btrfs_lookup_block_group(fs_info, bytenr);
3857 spin_lock(&bg->lock);
3861 atomic_inc(&bg->nocow_writers);
3862 spin_unlock(&bg->lock);
3864 /* no put on block group, done by btrfs_dec_nocow_writers */
3866 btrfs_put_block_group(bg);
3872 void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3874 struct btrfs_block_group_cache *bg;
3876 bg = btrfs_lookup_block_group(fs_info, bytenr);
3878 if (atomic_dec_and_test(&bg->nocow_writers))
3879 wake_up_atomic_t(&bg->nocow_writers);
3881 * Once for our lookup and once for the lookup done by a previous call
3882 * to btrfs_inc_nocow_writers()
3884 btrfs_put_block_group(bg);
3885 btrfs_put_block_group(bg);
3888 static int btrfs_wait_nocow_writers_atomic_t(atomic_t *a)
3894 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg)
3896 wait_on_atomic_t(&bg->nocow_writers,
3897 btrfs_wait_nocow_writers_atomic_t,
3898 TASK_UNINTERRUPTIBLE);
3901 static const char *alloc_name(u64 flags)
3904 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3906 case BTRFS_BLOCK_GROUP_METADATA:
3908 case BTRFS_BLOCK_GROUP_DATA:
3910 case BTRFS_BLOCK_GROUP_SYSTEM:
3914 return "invalid-combination";
3918 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3919 u64 total_bytes, u64 bytes_used,
3921 struct btrfs_space_info **space_info)
3923 struct btrfs_space_info *found;
3928 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3929 BTRFS_BLOCK_GROUP_RAID10))
3934 found = __find_space_info(info, flags);
3936 spin_lock(&found->lock);
3937 found->total_bytes += total_bytes;
3938 found->disk_total += total_bytes * factor;
3939 found->bytes_used += bytes_used;
3940 found->disk_used += bytes_used * factor;
3941 found->bytes_readonly += bytes_readonly;
3942 if (total_bytes > 0)
3944 space_info_add_new_bytes(info, found, total_bytes -
3945 bytes_used - bytes_readonly);
3946 spin_unlock(&found->lock);
3947 *space_info = found;
3950 found = kzalloc(sizeof(*found), GFP_NOFS);
3954 ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL);
3960 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3961 INIT_LIST_HEAD(&found->block_groups[i]);
3962 init_rwsem(&found->groups_sem);
3963 spin_lock_init(&found->lock);
3964 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3965 found->total_bytes = total_bytes;
3966 found->disk_total = total_bytes * factor;
3967 found->bytes_used = bytes_used;
3968 found->disk_used = bytes_used * factor;
3969 found->bytes_pinned = 0;
3970 found->bytes_reserved = 0;
3971 found->bytes_readonly = bytes_readonly;
3972 found->bytes_may_use = 0;
3974 found->max_extent_size = 0;
3975 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3976 found->chunk_alloc = 0;
3978 init_waitqueue_head(&found->wait);
3979 INIT_LIST_HEAD(&found->ro_bgs);
3980 INIT_LIST_HEAD(&found->tickets);
3981 INIT_LIST_HEAD(&found->priority_tickets);
3983 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3984 info->space_info_kobj, "%s",
3985 alloc_name(found->flags));
3987 percpu_counter_destroy(&found->total_bytes_pinned);
3992 *space_info = found;
3993 list_add_rcu(&found->list, &info->space_info);
3994 if (flags & BTRFS_BLOCK_GROUP_DATA)
3995 info->data_sinfo = found;
4000 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
4002 u64 extra_flags = chunk_to_extended(flags) &
4003 BTRFS_EXTENDED_PROFILE_MASK;
4005 write_seqlock(&fs_info->profiles_lock);
4006 if (flags & BTRFS_BLOCK_GROUP_DATA)
4007 fs_info->avail_data_alloc_bits |= extra_flags;
4008 if (flags & BTRFS_BLOCK_GROUP_METADATA)
4009 fs_info->avail_metadata_alloc_bits |= extra_flags;
4010 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4011 fs_info->avail_system_alloc_bits |= extra_flags;
4012 write_sequnlock(&fs_info->profiles_lock);
4016 * returns target flags in extended format or 0 if restripe for this
4017 * chunk_type is not in progress
4019 * should be called with either volume_mutex or balance_lock held
4021 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
4023 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4029 if (flags & BTRFS_BLOCK_GROUP_DATA &&
4030 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4031 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
4032 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
4033 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4034 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
4035 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
4036 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4037 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
4044 * @flags: available profiles in extended format (see ctree.h)
4046 * Returns reduced profile in chunk format. If profile changing is in
4047 * progress (either running or paused) picks the target profile (if it's
4048 * already available), otherwise falls back to plain reducing.
4050 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
4052 u64 num_devices = root->fs_info->fs_devices->rw_devices;
4058 * see if restripe for this chunk_type is in progress, if so
4059 * try to reduce to the target profile
4061 spin_lock(&root->fs_info->balance_lock);
4062 target = get_restripe_target(root->fs_info, flags);
4064 /* pick target profile only if it's already available */
4065 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
4066 spin_unlock(&root->fs_info->balance_lock);
4067 return extended_to_chunk(target);
4070 spin_unlock(&root->fs_info->balance_lock);
4072 /* First, mask out the RAID levels which aren't possible */
4073 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4074 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
4075 allowed |= btrfs_raid_group[raid_type];
4079 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
4080 allowed = BTRFS_BLOCK_GROUP_RAID6;
4081 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
4082 allowed = BTRFS_BLOCK_GROUP_RAID5;
4083 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
4084 allowed = BTRFS_BLOCK_GROUP_RAID10;
4085 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
4086 allowed = BTRFS_BLOCK_GROUP_RAID1;
4087 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
4088 allowed = BTRFS_BLOCK_GROUP_RAID0;
4090 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
4092 return extended_to_chunk(flags | allowed);
4095 static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags)
4102 seq = read_seqbegin(&root->fs_info->profiles_lock);
4104 if (flags & BTRFS_BLOCK_GROUP_DATA)
4105 flags |= root->fs_info->avail_data_alloc_bits;
4106 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4107 flags |= root->fs_info->avail_system_alloc_bits;
4108 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
4109 flags |= root->fs_info->avail_metadata_alloc_bits;
4110 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
4112 return btrfs_reduce_alloc_profile(root, flags);
4115 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
4121 flags = BTRFS_BLOCK_GROUP_DATA;
4122 else if (root == root->fs_info->chunk_root)
4123 flags = BTRFS_BLOCK_GROUP_SYSTEM;
4125 flags = BTRFS_BLOCK_GROUP_METADATA;
4127 ret = get_alloc_profile(root, flags);
4131 int btrfs_alloc_data_chunk_ondemand(struct inode *inode, u64 bytes)
4133 struct btrfs_space_info *data_sinfo;
4134 struct btrfs_root *root = BTRFS_I(inode)->root;
4135 struct btrfs_fs_info *fs_info = root->fs_info;
4138 int need_commit = 2;
4139 int have_pinned_space;
4141 /* make sure bytes are sectorsize aligned */
4142 bytes = ALIGN(bytes, root->sectorsize);
4144 if (btrfs_is_free_space_inode(inode)) {
4146 ASSERT(current->journal_info);
4149 data_sinfo = fs_info->data_sinfo;
4154 /* make sure we have enough space to handle the data first */
4155 spin_lock(&data_sinfo->lock);
4156 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
4157 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
4158 data_sinfo->bytes_may_use;
4160 if (used + bytes > data_sinfo->total_bytes) {
4161 struct btrfs_trans_handle *trans;
4164 * if we don't have enough free bytes in this space then we need
4165 * to alloc a new chunk.
4167 if (!data_sinfo->full) {
4170 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4171 spin_unlock(&data_sinfo->lock);
4173 alloc_target = btrfs_get_alloc_profile(root, 1);
4175 * It is ugly that we don't call nolock join
4176 * transaction for the free space inode case here.
4177 * But it is safe because we only do the data space
4178 * reservation for the free space cache in the
4179 * transaction context, the common join transaction
4180 * just increase the counter of the current transaction
4181 * handler, doesn't try to acquire the trans_lock of
4184 trans = btrfs_join_transaction(root);
4186 return PTR_ERR(trans);
4188 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4190 CHUNK_ALLOC_NO_FORCE);
4191 btrfs_end_transaction(trans, root);
4196 have_pinned_space = 1;
4202 data_sinfo = fs_info->data_sinfo;
4208 * If we don't have enough pinned space to deal with this
4209 * allocation, and no removed chunk in current transaction,
4210 * don't bother committing the transaction.
4212 have_pinned_space = percpu_counter_compare(
4213 &data_sinfo->total_bytes_pinned,
4214 used + bytes - data_sinfo->total_bytes);
4215 spin_unlock(&data_sinfo->lock);
4217 /* commit the current transaction and try again */
4220 !atomic_read(&root->fs_info->open_ioctl_trans)) {
4223 if (need_commit > 0) {
4224 btrfs_start_delalloc_roots(fs_info, 0, -1);
4225 btrfs_wait_ordered_roots(fs_info, -1, 0, (u64)-1);
4228 trans = btrfs_join_transaction(root);
4230 return PTR_ERR(trans);
4231 if (have_pinned_space >= 0 ||
4232 test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4233 &trans->transaction->flags) ||
4235 ret = btrfs_commit_transaction(trans, root);
4239 * The cleaner kthread might still be doing iput
4240 * operations. Wait for it to finish so that
4241 * more space is released.
4243 mutex_lock(&root->fs_info->cleaner_delayed_iput_mutex);
4244 mutex_unlock(&root->fs_info->cleaner_delayed_iput_mutex);
4247 btrfs_end_transaction(trans, root);
4251 trace_btrfs_space_reservation(root->fs_info,
4252 "space_info:enospc",
4253 data_sinfo->flags, bytes, 1);
4256 data_sinfo->bytes_may_use += bytes;
4257 trace_btrfs_space_reservation(root->fs_info, "space_info",
4258 data_sinfo->flags, bytes, 1);
4259 spin_unlock(&data_sinfo->lock);
4265 * New check_data_free_space() with ability for precious data reservation
4266 * Will replace old btrfs_check_data_free_space(), but for patch split,
4267 * add a new function first and then replace it.
4269 int btrfs_check_data_free_space(struct inode *inode, u64 start, u64 len)
4271 struct btrfs_root *root = BTRFS_I(inode)->root;
4274 /* align the range */
4275 len = round_up(start + len, root->sectorsize) -
4276 round_down(start, root->sectorsize);
4277 start = round_down(start, root->sectorsize);
4279 ret = btrfs_alloc_data_chunk_ondemand(inode, len);
4283 /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */
4284 ret = btrfs_qgroup_reserve_data(inode, start, len);
4286 btrfs_free_reserved_data_space_noquota(inode, start, len);
4291 * Called if we need to clear a data reservation for this inode
4292 * Normally in a error case.
4294 * This one will *NOT* use accurate qgroup reserved space API, just for case
4295 * which we can't sleep and is sure it won't affect qgroup reserved space.
4296 * Like clear_bit_hook().
4298 void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4301 struct btrfs_root *root = BTRFS_I(inode)->root;
4302 struct btrfs_space_info *data_sinfo;
4304 /* Make sure the range is aligned to sectorsize */
4305 len = round_up(start + len, root->sectorsize) -
4306 round_down(start, root->sectorsize);
4307 start = round_down(start, root->sectorsize);
4309 data_sinfo = root->fs_info->data_sinfo;
4310 spin_lock(&data_sinfo->lock);
4311 if (WARN_ON(data_sinfo->bytes_may_use < len))
4312 data_sinfo->bytes_may_use = 0;
4314 data_sinfo->bytes_may_use -= len;
4315 trace_btrfs_space_reservation(root->fs_info, "space_info",
4316 data_sinfo->flags, len, 0);
4317 spin_unlock(&data_sinfo->lock);
4321 * Called if we need to clear a data reservation for this inode
4322 * Normally in a error case.
4324 * This one will handle the per-inode data rsv map for accurate reserved
4327 void btrfs_free_reserved_data_space(struct inode *inode, u64 start, u64 len)
4329 btrfs_free_reserved_data_space_noquota(inode, start, len);
4330 btrfs_qgroup_free_data(inode, start, len);
4333 static void force_metadata_allocation(struct btrfs_fs_info *info)
4335 struct list_head *head = &info->space_info;
4336 struct btrfs_space_info *found;
4339 list_for_each_entry_rcu(found, head, list) {
4340 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4341 found->force_alloc = CHUNK_ALLOC_FORCE;
4346 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4348 return (global->size << 1);
4351 static int should_alloc_chunk(struct btrfs_root *root,
4352 struct btrfs_space_info *sinfo, int force)
4354 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4355 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
4356 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
4359 if (force == CHUNK_ALLOC_FORCE)
4363 * We need to take into account the global rsv because for all intents
4364 * and purposes it's used space. Don't worry about locking the
4365 * global_rsv, it doesn't change except when the transaction commits.
4367 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4368 num_allocated += calc_global_rsv_need_space(global_rsv);
4371 * in limited mode, we want to have some free space up to
4372 * about 1% of the FS size.
4374 if (force == CHUNK_ALLOC_LIMITED) {
4375 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
4376 thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1));
4378 if (num_bytes - num_allocated < thresh)
4382 if (num_allocated + SZ_2M < div_factor(num_bytes, 8))
4387 static u64 get_profile_num_devs(struct btrfs_root *root, u64 type)
4391 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4392 BTRFS_BLOCK_GROUP_RAID0 |
4393 BTRFS_BLOCK_GROUP_RAID5 |
4394 BTRFS_BLOCK_GROUP_RAID6))
4395 num_dev = root->fs_info->fs_devices->rw_devices;
4396 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4399 num_dev = 1; /* DUP or single */
4405 * If @is_allocation is true, reserve space in the system space info necessary
4406 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4409 void check_system_chunk(struct btrfs_trans_handle *trans,
4410 struct btrfs_root *root,
4413 struct btrfs_space_info *info;
4420 * Needed because we can end up allocating a system chunk and for an
4421 * atomic and race free space reservation in the chunk block reserve.
4423 ASSERT(mutex_is_locked(&root->fs_info->chunk_mutex));
4425 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4426 spin_lock(&info->lock);
4427 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
4428 info->bytes_reserved - info->bytes_readonly -
4429 info->bytes_may_use;
4430 spin_unlock(&info->lock);
4432 num_devs = get_profile_num_devs(root, type);
4434 /* num_devs device items to update and 1 chunk item to add or remove */
4435 thresh = btrfs_calc_trunc_metadata_size(root, num_devs) +
4436 btrfs_calc_trans_metadata_size(root, 1);
4438 if (left < thresh && btrfs_test_opt(root->fs_info, ENOSPC_DEBUG)) {
4439 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
4440 left, thresh, type);
4441 dump_space_info(root->fs_info, info, 0, 0);
4444 if (left < thresh) {
4447 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
4449 * Ignore failure to create system chunk. We might end up not
4450 * needing it, as we might not need to COW all nodes/leafs from
4451 * the paths we visit in the chunk tree (they were already COWed
4452 * or created in the current transaction for example).
4454 ret = btrfs_alloc_chunk(trans, root, flags);
4458 ret = btrfs_block_rsv_add(root->fs_info->chunk_root,
4459 &root->fs_info->chunk_block_rsv,
4460 thresh, BTRFS_RESERVE_NO_FLUSH);
4462 trans->chunk_bytes_reserved += thresh;
4467 * If force is CHUNK_ALLOC_FORCE:
4468 * - return 1 if it successfully allocates a chunk,
4469 * - return errors including -ENOSPC otherwise.
4470 * If force is NOT CHUNK_ALLOC_FORCE:
4471 * - return 0 if it doesn't need to allocate a new chunk,
4472 * - return 1 if it successfully allocates a chunk,
4473 * - return errors including -ENOSPC otherwise.
4475 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4476 struct btrfs_root *extent_root, u64 flags, int force)
4478 struct btrfs_space_info *space_info;
4479 struct btrfs_fs_info *fs_info = extent_root->fs_info;
4480 int wait_for_alloc = 0;
4483 /* Don't re-enter if we're already allocating a chunk */
4484 if (trans->allocating_chunk)
4487 space_info = __find_space_info(extent_root->fs_info, flags);
4489 ret = update_space_info(extent_root->fs_info, flags,
4490 0, 0, 0, &space_info);
4491 BUG_ON(ret); /* -ENOMEM */
4493 BUG_ON(!space_info); /* Logic error */
4496 spin_lock(&space_info->lock);
4497 if (force < space_info->force_alloc)
4498 force = space_info->force_alloc;
4499 if (space_info->full) {
4500 if (should_alloc_chunk(extent_root, space_info, force))
4504 spin_unlock(&space_info->lock);
4508 if (!should_alloc_chunk(extent_root, space_info, force)) {
4509 spin_unlock(&space_info->lock);
4511 } else if (space_info->chunk_alloc) {
4514 space_info->chunk_alloc = 1;
4517 spin_unlock(&space_info->lock);
4519 mutex_lock(&fs_info->chunk_mutex);
4522 * The chunk_mutex is held throughout the entirety of a chunk
4523 * allocation, so once we've acquired the chunk_mutex we know that the
4524 * other guy is done and we need to recheck and see if we should
4527 if (wait_for_alloc) {
4528 mutex_unlock(&fs_info->chunk_mutex);
4534 trans->allocating_chunk = true;
4537 * If we have mixed data/metadata chunks we want to make sure we keep
4538 * allocating mixed chunks instead of individual chunks.
4540 if (btrfs_mixed_space_info(space_info))
4541 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4544 * if we're doing a data chunk, go ahead and make sure that
4545 * we keep a reasonable number of metadata chunks allocated in the
4548 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4549 fs_info->data_chunk_allocations++;
4550 if (!(fs_info->data_chunk_allocations %
4551 fs_info->metadata_ratio))
4552 force_metadata_allocation(fs_info);
4556 * Check if we have enough space in SYSTEM chunk because we may need
4557 * to update devices.
4559 check_system_chunk(trans, extent_root, flags);
4561 ret = btrfs_alloc_chunk(trans, extent_root, flags);
4562 trans->allocating_chunk = false;
4564 spin_lock(&space_info->lock);
4565 if (ret < 0 && ret != -ENOSPC)
4568 space_info->full = 1;
4572 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4574 space_info->chunk_alloc = 0;
4575 spin_unlock(&space_info->lock);
4576 mutex_unlock(&fs_info->chunk_mutex);
4578 * When we allocate a new chunk we reserve space in the chunk block
4579 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4580 * add new nodes/leafs to it if we end up needing to do it when
4581 * inserting the chunk item and updating device items as part of the
4582 * second phase of chunk allocation, performed by
4583 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4584 * large number of new block groups to create in our transaction
4585 * handle's new_bgs list to avoid exhausting the chunk block reserve
4586 * in extreme cases - like having a single transaction create many new
4587 * block groups when starting to write out the free space caches of all
4588 * the block groups that were made dirty during the lifetime of the
4591 if (trans->can_flush_pending_bgs &&
4592 trans->chunk_bytes_reserved >= (u64)SZ_2M) {
4593 btrfs_create_pending_block_groups(trans, extent_root);
4594 btrfs_trans_release_chunk_metadata(trans);
4599 static int can_overcommit(struct btrfs_root *root,
4600 struct btrfs_space_info *space_info, u64 bytes,
4601 enum btrfs_reserve_flush_enum flush)
4603 struct btrfs_block_rsv *global_rsv;
4609 /* Don't overcommit when in mixed mode. */
4610 if (space_info->flags & BTRFS_BLOCK_GROUP_DATA)
4613 BUG_ON(root->fs_info == NULL);
4614 global_rsv = &root->fs_info->global_block_rsv;
4615 profile = btrfs_get_alloc_profile(root, 0);
4616 used = space_info->bytes_used + space_info->bytes_reserved +
4617 space_info->bytes_pinned + space_info->bytes_readonly;
4620 * We only want to allow over committing if we have lots of actual space
4621 * free, but if we don't have enough space to handle the global reserve
4622 * space then we could end up having a real enospc problem when trying
4623 * to allocate a chunk or some other such important allocation.
4625 spin_lock(&global_rsv->lock);
4626 space_size = calc_global_rsv_need_space(global_rsv);
4627 spin_unlock(&global_rsv->lock);
4628 if (used + space_size >= space_info->total_bytes)
4631 used += space_info->bytes_may_use;
4633 spin_lock(&root->fs_info->free_chunk_lock);
4634 avail = root->fs_info->free_chunk_space;
4635 spin_unlock(&root->fs_info->free_chunk_lock);
4638 * If we have dup, raid1 or raid10 then only half of the free
4639 * space is actually useable. For raid56, the space info used
4640 * doesn't include the parity drive, so we don't have to
4643 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4644 BTRFS_BLOCK_GROUP_RAID1 |
4645 BTRFS_BLOCK_GROUP_RAID10))
4649 * If we aren't flushing all things, let us overcommit up to
4650 * 1/2th of the space. If we can flush, don't let us overcommit
4651 * too much, let it overcommit up to 1/8 of the space.
4653 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4658 if (used + bytes < space_info->total_bytes + avail)
4663 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4664 unsigned long nr_pages, int nr_items)
4666 struct super_block *sb = root->fs_info->sb;
4668 if (down_read_trylock(&sb->s_umount)) {
4669 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4670 up_read(&sb->s_umount);
4673 * We needn't worry the filesystem going from r/w to r/o though
4674 * we don't acquire ->s_umount mutex, because the filesystem
4675 * should guarantee the delalloc inodes list be empty after
4676 * the filesystem is readonly(all dirty pages are written to
4679 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
4680 if (!current->journal_info)
4681 btrfs_wait_ordered_roots(root->fs_info, nr_items,
4686 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4691 bytes = btrfs_calc_trans_metadata_size(root, 1);
4692 nr = (int)div64_u64(to_reclaim, bytes);
4698 #define EXTENT_SIZE_PER_ITEM SZ_256K
4701 * shrink metadata reservation for delalloc
4703 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4706 struct btrfs_block_rsv *block_rsv;
4707 struct btrfs_space_info *space_info;
4708 struct btrfs_trans_handle *trans;
4712 unsigned long nr_pages;
4715 enum btrfs_reserve_flush_enum flush;
4717 /* Calc the number of the pages we need flush for space reservation */
4718 items = calc_reclaim_items_nr(root, to_reclaim);
4719 to_reclaim = (u64)items * EXTENT_SIZE_PER_ITEM;
4721 trans = (struct btrfs_trans_handle *)current->journal_info;
4722 block_rsv = &root->fs_info->delalloc_block_rsv;
4723 space_info = block_rsv->space_info;
4725 delalloc_bytes = percpu_counter_sum_positive(
4726 &root->fs_info->delalloc_bytes);
4727 if (delalloc_bytes == 0) {
4731 btrfs_wait_ordered_roots(root->fs_info, items,
4737 while (delalloc_bytes && loops < 3) {
4738 max_reclaim = min(delalloc_bytes, to_reclaim);
4739 nr_pages = max_reclaim >> PAGE_SHIFT;
4740 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4742 * We need to wait for the async pages to actually start before
4745 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4749 if (max_reclaim <= nr_pages)
4752 max_reclaim -= nr_pages;
4754 wait_event(root->fs_info->async_submit_wait,
4755 atomic_read(&root->fs_info->async_delalloc_pages) <=
4759 flush = BTRFS_RESERVE_FLUSH_ALL;
4761 flush = BTRFS_RESERVE_NO_FLUSH;
4762 spin_lock(&space_info->lock);
4763 if (list_empty(&space_info->tickets) &&
4764 list_empty(&space_info->priority_tickets)) {
4765 spin_unlock(&space_info->lock);
4768 spin_unlock(&space_info->lock);
4771 if (wait_ordered && !trans) {
4772 btrfs_wait_ordered_roots(root->fs_info, items,
4775 time_left = schedule_timeout_killable(1);
4779 delalloc_bytes = percpu_counter_sum_positive(
4780 &root->fs_info->delalloc_bytes);
4785 * maybe_commit_transaction - possibly commit the transaction if its ok to
4786 * @root - the root we're allocating for
4787 * @bytes - the number of bytes we want to reserve
4788 * @force - force the commit
4790 * This will check to make sure that committing the transaction will actually
4791 * get us somewhere and then commit the transaction if it does. Otherwise it
4792 * will return -ENOSPC.
4794 static int may_commit_transaction(struct btrfs_root *root,
4795 struct btrfs_space_info *space_info,
4796 u64 bytes, int force)
4798 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4799 struct btrfs_trans_handle *trans;
4801 trans = (struct btrfs_trans_handle *)current->journal_info;
4808 /* See if there is enough pinned space to make this reservation */
4809 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4814 * See if there is some space in the delayed insertion reservation for
4817 if (space_info != delayed_rsv->space_info)
4820 spin_lock(&delayed_rsv->lock);
4821 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4822 bytes - delayed_rsv->size) >= 0) {
4823 spin_unlock(&delayed_rsv->lock);
4826 spin_unlock(&delayed_rsv->lock);
4829 trans = btrfs_join_transaction(root);
4833 return btrfs_commit_transaction(trans, root);
4836 struct reserve_ticket {
4839 struct list_head list;
4840 wait_queue_head_t wait;
4843 static int flush_space(struct btrfs_root *root,
4844 struct btrfs_space_info *space_info, u64 num_bytes,
4845 u64 orig_bytes, int state)
4847 struct btrfs_trans_handle *trans;
4852 case FLUSH_DELAYED_ITEMS_NR:
4853 case FLUSH_DELAYED_ITEMS:
4854 if (state == FLUSH_DELAYED_ITEMS_NR)
4855 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4859 trans = btrfs_join_transaction(root);
4860 if (IS_ERR(trans)) {
4861 ret = PTR_ERR(trans);
4864 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4865 btrfs_end_transaction(trans, root);
4867 case FLUSH_DELALLOC:
4868 case FLUSH_DELALLOC_WAIT:
4869 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4870 state == FLUSH_DELALLOC_WAIT);
4873 trans = btrfs_join_transaction(root);
4874 if (IS_ERR(trans)) {
4875 ret = PTR_ERR(trans);
4878 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4879 btrfs_get_alloc_profile(root, 0),
4880 CHUNK_ALLOC_NO_FORCE);
4881 btrfs_end_transaction(trans, root);
4882 if (ret > 0 || ret == -ENOSPC)
4886 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4893 trace_btrfs_flush_space(root->fs_info, space_info->flags, num_bytes,
4894 orig_bytes, state, ret);
4899 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4900 struct btrfs_space_info *space_info)
4902 struct reserve_ticket *ticket;
4907 list_for_each_entry(ticket, &space_info->tickets, list)
4908 to_reclaim += ticket->bytes;
4909 list_for_each_entry(ticket, &space_info->priority_tickets, list)
4910 to_reclaim += ticket->bytes;
4914 to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M);
4915 if (can_overcommit(root, space_info, to_reclaim,
4916 BTRFS_RESERVE_FLUSH_ALL))
4919 used = space_info->bytes_used + space_info->bytes_reserved +
4920 space_info->bytes_pinned + space_info->bytes_readonly +
4921 space_info->bytes_may_use;
4922 if (can_overcommit(root, space_info, SZ_1M, BTRFS_RESERVE_FLUSH_ALL))
4923 expected = div_factor_fine(space_info->total_bytes, 95);
4925 expected = div_factor_fine(space_info->total_bytes, 90);
4927 if (used > expected)
4928 to_reclaim = used - expected;
4931 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4932 space_info->bytes_reserved);
4936 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4937 struct btrfs_root *root, u64 used)
4939 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
4941 /* If we're just plain full then async reclaim just slows us down. */
4942 if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh)
4945 if (!btrfs_calc_reclaim_metadata_size(root, space_info))
4948 return (used >= thresh && !btrfs_fs_closing(root->fs_info) &&
4949 !test_bit(BTRFS_FS_STATE_REMOUNTING,
4950 &root->fs_info->fs_state));
4953 static void wake_all_tickets(struct list_head *head)
4955 struct reserve_ticket *ticket;
4957 while (!list_empty(head)) {
4958 ticket = list_first_entry(head, struct reserve_ticket, list);
4959 list_del_init(&ticket->list);
4960 ticket->error = -ENOSPC;
4961 wake_up(&ticket->wait);
4966 * This is for normal flushers, we can wait all goddamned day if we want to. We
4967 * will loop and continuously try to flush as long as we are making progress.
4968 * We count progress as clearing off tickets each time we have to loop.
4970 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4972 struct btrfs_fs_info *fs_info;
4973 struct btrfs_space_info *space_info;
4976 int commit_cycles = 0;
4977 u64 last_tickets_id;
4979 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4980 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4982 spin_lock(&space_info->lock);
4983 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4986 space_info->flush = 0;
4987 spin_unlock(&space_info->lock);
4990 last_tickets_id = space_info->tickets_id;
4991 spin_unlock(&space_info->lock);
4993 flush_state = FLUSH_DELAYED_ITEMS_NR;
4995 struct reserve_ticket *ticket;
4998 ret = flush_space(fs_info->fs_root, space_info, to_reclaim,
4999 to_reclaim, flush_state);
5000 spin_lock(&space_info->lock);
5001 if (list_empty(&space_info->tickets)) {
5002 space_info->flush = 0;
5003 spin_unlock(&space_info->lock);
5006 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
5008 ticket = list_first_entry(&space_info->tickets,
5009 struct reserve_ticket, list);
5010 if (last_tickets_id == space_info->tickets_id) {
5013 last_tickets_id = space_info->tickets_id;
5014 flush_state = FLUSH_DELAYED_ITEMS_NR;
5019 if (flush_state > COMMIT_TRANS) {
5021 if (commit_cycles > 2) {
5022 wake_all_tickets(&space_info->tickets);
5023 space_info->flush = 0;
5025 flush_state = FLUSH_DELAYED_ITEMS_NR;
5028 spin_unlock(&space_info->lock);
5029 } while (flush_state <= COMMIT_TRANS);
5032 void btrfs_init_async_reclaim_work(struct work_struct *work)
5034 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
5037 static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
5038 struct btrfs_space_info *space_info,
5039 struct reserve_ticket *ticket)
5042 int flush_state = FLUSH_DELAYED_ITEMS_NR;
5044 spin_lock(&space_info->lock);
5045 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
5048 spin_unlock(&space_info->lock);
5051 spin_unlock(&space_info->lock);
5054 flush_space(fs_info->fs_root, space_info, to_reclaim,
5055 to_reclaim, flush_state);
5057 spin_lock(&space_info->lock);
5058 if (ticket->bytes == 0) {
5059 spin_unlock(&space_info->lock);
5062 spin_unlock(&space_info->lock);
5065 * Priority flushers can't wait on delalloc without
5068 if (flush_state == FLUSH_DELALLOC ||
5069 flush_state == FLUSH_DELALLOC_WAIT)
5070 flush_state = ALLOC_CHUNK;
5071 } while (flush_state < COMMIT_TRANS);
5074 static int wait_reserve_ticket(struct btrfs_fs_info *fs_info,
5075 struct btrfs_space_info *space_info,
5076 struct reserve_ticket *ticket, u64 orig_bytes)
5082 spin_lock(&space_info->lock);
5083 while (ticket->bytes > 0 && ticket->error == 0) {
5084 ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE);
5089 spin_unlock(&space_info->lock);
5093 finish_wait(&ticket->wait, &wait);
5094 spin_lock(&space_info->lock);
5097 ret = ticket->error;
5098 if (!list_empty(&ticket->list))
5099 list_del_init(&ticket->list);
5100 if (ticket->bytes && ticket->bytes < orig_bytes) {
5101 u64 num_bytes = orig_bytes - ticket->bytes;
5102 space_info->bytes_may_use -= num_bytes;
5103 trace_btrfs_space_reservation(fs_info, "space_info",
5104 space_info->flags, num_bytes, 0);
5106 spin_unlock(&space_info->lock);
5112 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5113 * @root - the root we're allocating for
5114 * @space_info - the space info we want to allocate from
5115 * @orig_bytes - the number of bytes we want
5116 * @flush - whether or not we can flush to make our reservation
5118 * This will reserve orig_bytes number of bytes from the space info associated
5119 * with the block_rsv. If there is not enough space it will make an attempt to
5120 * flush out space to make room. It will do this by flushing delalloc if
5121 * possible or committing the transaction. If flush is 0 then no attempts to
5122 * regain reservations will be made and this will fail if there is not enough
5125 static int __reserve_metadata_bytes(struct btrfs_root *root,
5126 struct btrfs_space_info *space_info,
5128 enum btrfs_reserve_flush_enum flush)
5130 struct reserve_ticket ticket;
5135 ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL);
5137 spin_lock(&space_info->lock);
5139 used = space_info->bytes_used + space_info->bytes_reserved +
5140 space_info->bytes_pinned + space_info->bytes_readonly +
5141 space_info->bytes_may_use;
5144 * If we have enough space then hooray, make our reservation and carry
5145 * on. If not see if we can overcommit, and if we can, hooray carry on.
5146 * If not things get more complicated.
5148 if (used + orig_bytes <= space_info->total_bytes) {
5149 space_info->bytes_may_use += orig_bytes;
5150 trace_btrfs_space_reservation(root->fs_info, "space_info",
5151 space_info->flags, orig_bytes,
5154 } else if (can_overcommit(root, space_info, orig_bytes, flush)) {
5155 space_info->bytes_may_use += orig_bytes;
5156 trace_btrfs_space_reservation(root->fs_info, "space_info",
5157 space_info->flags, orig_bytes,
5163 * If we couldn't make a reservation then setup our reservation ticket
5164 * and kick the async worker if it's not already running.
5166 * If we are a priority flusher then we just need to add our ticket to
5167 * the list and we will do our own flushing further down.
5169 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
5170 ticket.bytes = orig_bytes;
5172 init_waitqueue_head(&ticket.wait);
5173 if (flush == BTRFS_RESERVE_FLUSH_ALL) {
5174 list_add_tail(&ticket.list, &space_info->tickets);
5175 if (!space_info->flush) {
5176 space_info->flush = 1;
5177 trace_btrfs_trigger_flush(root->fs_info,
5181 queue_work(system_unbound_wq,
5182 &root->fs_info->async_reclaim_work);
5185 list_add_tail(&ticket.list,
5186 &space_info->priority_tickets);
5188 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
5191 * We will do the space reservation dance during log replay,
5192 * which means we won't have fs_info->fs_root set, so don't do
5193 * the async reclaim as we will panic.
5195 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags) &&
5196 need_do_async_reclaim(space_info, root, used) &&
5197 !work_busy(&root->fs_info->async_reclaim_work)) {
5198 trace_btrfs_trigger_flush(root->fs_info,
5202 queue_work(system_unbound_wq,
5203 &root->fs_info->async_reclaim_work);
5206 spin_unlock(&space_info->lock);
5207 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
5210 if (flush == BTRFS_RESERVE_FLUSH_ALL)
5211 return wait_reserve_ticket(root->fs_info, space_info, &ticket,
5215 priority_reclaim_metadata_space(root->fs_info, space_info, &ticket);
5216 spin_lock(&space_info->lock);
5218 if (ticket.bytes < orig_bytes) {
5219 u64 num_bytes = orig_bytes - ticket.bytes;
5220 space_info->bytes_may_use -= num_bytes;
5221 trace_btrfs_space_reservation(root->fs_info,
5222 "space_info", space_info->flags,
5226 list_del_init(&ticket.list);
5229 spin_unlock(&space_info->lock);
5230 ASSERT(list_empty(&ticket.list));
5235 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5236 * @root - the root we're allocating for
5237 * @block_rsv - the block_rsv we're allocating for
5238 * @orig_bytes - the number of bytes we want
5239 * @flush - whether or not we can flush to make our reservation
5241 * This will reserve orgi_bytes number of bytes from the space info associated
5242 * with the block_rsv. If there is not enough space it will make an attempt to
5243 * flush out space to make room. It will do this by flushing delalloc if
5244 * possible or committing the transaction. If flush is 0 then no attempts to
5245 * regain reservations will be made and this will fail if there is not enough
5248 static int reserve_metadata_bytes(struct btrfs_root *root,
5249 struct btrfs_block_rsv *block_rsv,
5251 enum btrfs_reserve_flush_enum flush)
5255 ret = __reserve_metadata_bytes(root, block_rsv->space_info, orig_bytes,
5257 if (ret == -ENOSPC &&
5258 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5259 struct btrfs_block_rsv *global_rsv =
5260 &root->fs_info->global_block_rsv;
5262 if (block_rsv != global_rsv &&
5263 !block_rsv_use_bytes(global_rsv, orig_bytes))
5267 trace_btrfs_space_reservation(root->fs_info,
5268 "space_info:enospc",
5269 block_rsv->space_info->flags,
5274 static struct btrfs_block_rsv *get_block_rsv(
5275 const struct btrfs_trans_handle *trans,
5276 const struct btrfs_root *root)
5278 struct btrfs_block_rsv *block_rsv = NULL;
5280 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5281 (root == root->fs_info->csum_root && trans->adding_csums) ||
5282 (root == root->fs_info->uuid_root))
5283 block_rsv = trans->block_rsv;
5286 block_rsv = root->block_rsv;
5289 block_rsv = &root->fs_info->empty_block_rsv;
5294 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5298 spin_lock(&block_rsv->lock);
5299 if (block_rsv->reserved >= num_bytes) {
5300 block_rsv->reserved -= num_bytes;
5301 if (block_rsv->reserved < block_rsv->size)
5302 block_rsv->full = 0;
5305 spin_unlock(&block_rsv->lock);
5309 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5310 u64 num_bytes, int update_size)
5312 spin_lock(&block_rsv->lock);
5313 block_rsv->reserved += num_bytes;
5315 block_rsv->size += num_bytes;
5316 else if (block_rsv->reserved >= block_rsv->size)
5317 block_rsv->full = 1;
5318 spin_unlock(&block_rsv->lock);
5321 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5322 struct btrfs_block_rsv *dest, u64 num_bytes,
5325 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5328 if (global_rsv->space_info != dest->space_info)
5331 spin_lock(&global_rsv->lock);
5332 min_bytes = div_factor(global_rsv->size, min_factor);
5333 if (global_rsv->reserved < min_bytes + num_bytes) {
5334 spin_unlock(&global_rsv->lock);
5337 global_rsv->reserved -= num_bytes;
5338 if (global_rsv->reserved < global_rsv->size)
5339 global_rsv->full = 0;
5340 spin_unlock(&global_rsv->lock);
5342 block_rsv_add_bytes(dest, num_bytes, 1);
5347 * This is for space we already have accounted in space_info->bytes_may_use, so
5348 * basically when we're returning space from block_rsv's.
5350 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
5351 struct btrfs_space_info *space_info,
5354 struct reserve_ticket *ticket;
5355 struct list_head *head;
5357 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH;
5358 bool check_overcommit = false;
5360 spin_lock(&space_info->lock);
5361 head = &space_info->priority_tickets;
5364 * If we are over our limit then we need to check and see if we can
5365 * overcommit, and if we can't then we just need to free up our space
5366 * and not satisfy any requests.
5368 used = space_info->bytes_used + space_info->bytes_reserved +
5369 space_info->bytes_pinned + space_info->bytes_readonly +
5370 space_info->bytes_may_use;
5371 if (used - num_bytes >= space_info->total_bytes)
5372 check_overcommit = true;
5374 while (!list_empty(head) && num_bytes) {
5375 ticket = list_first_entry(head, struct reserve_ticket,
5378 * We use 0 bytes because this space is already reserved, so
5379 * adding the ticket space would be a double count.
5381 if (check_overcommit &&
5382 !can_overcommit(fs_info->extent_root, space_info, 0,
5385 if (num_bytes >= ticket->bytes) {
5386 list_del_init(&ticket->list);
5387 num_bytes -= ticket->bytes;
5389 space_info->tickets_id++;
5390 wake_up(&ticket->wait);
5392 ticket->bytes -= num_bytes;
5397 if (num_bytes && head == &space_info->priority_tickets) {
5398 head = &space_info->tickets;
5399 flush = BTRFS_RESERVE_FLUSH_ALL;
5402 space_info->bytes_may_use -= num_bytes;
5403 trace_btrfs_space_reservation(fs_info, "space_info",
5404 space_info->flags, num_bytes, 0);
5405 spin_unlock(&space_info->lock);
5409 * This is for newly allocated space that isn't accounted in
5410 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5411 * we use this helper.
5413 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
5414 struct btrfs_space_info *space_info,
5417 struct reserve_ticket *ticket;
5418 struct list_head *head = &space_info->priority_tickets;
5421 while (!list_empty(head) && num_bytes) {
5422 ticket = list_first_entry(head, struct reserve_ticket,
5424 if (num_bytes >= ticket->bytes) {
5425 trace_btrfs_space_reservation(fs_info, "space_info",
5428 list_del_init(&ticket->list);
5429 num_bytes -= ticket->bytes;
5430 space_info->bytes_may_use += ticket->bytes;
5432 space_info->tickets_id++;
5433 wake_up(&ticket->wait);
5435 trace_btrfs_space_reservation(fs_info, "space_info",
5438 space_info->bytes_may_use += num_bytes;
5439 ticket->bytes -= num_bytes;
5444 if (num_bytes && head == &space_info->priority_tickets) {
5445 head = &space_info->tickets;
5450 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5451 struct btrfs_block_rsv *block_rsv,
5452 struct btrfs_block_rsv *dest, u64 num_bytes)
5454 struct btrfs_space_info *space_info = block_rsv->space_info;
5456 spin_lock(&block_rsv->lock);
5457 if (num_bytes == (u64)-1)
5458 num_bytes = block_rsv->size;
5459 block_rsv->size -= num_bytes;
5460 if (block_rsv->reserved >= block_rsv->size) {
5461 num_bytes = block_rsv->reserved - block_rsv->size;
5462 block_rsv->reserved = block_rsv->size;
5463 block_rsv->full = 1;
5467 spin_unlock(&block_rsv->lock);
5469 if (num_bytes > 0) {
5471 spin_lock(&dest->lock);
5475 bytes_to_add = dest->size - dest->reserved;
5476 bytes_to_add = min(num_bytes, bytes_to_add);
5477 dest->reserved += bytes_to_add;
5478 if (dest->reserved >= dest->size)
5480 num_bytes -= bytes_to_add;
5482 spin_unlock(&dest->lock);
5485 space_info_add_old_bytes(fs_info, space_info,
5490 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
5491 struct btrfs_block_rsv *dst, u64 num_bytes,
5496 ret = block_rsv_use_bytes(src, num_bytes);
5500 block_rsv_add_bytes(dst, num_bytes, update_size);
5504 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5506 memset(rsv, 0, sizeof(*rsv));
5507 spin_lock_init(&rsv->lock);
5511 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
5512 unsigned short type)
5514 struct btrfs_block_rsv *block_rsv;
5515 struct btrfs_fs_info *fs_info = root->fs_info;
5517 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5521 btrfs_init_block_rsv(block_rsv, type);
5522 block_rsv->space_info = __find_space_info(fs_info,
5523 BTRFS_BLOCK_GROUP_METADATA);
5527 void btrfs_free_block_rsv(struct btrfs_root *root,
5528 struct btrfs_block_rsv *rsv)
5532 btrfs_block_rsv_release(root, rsv, (u64)-1);
5536 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
5541 int btrfs_block_rsv_add(struct btrfs_root *root,
5542 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5543 enum btrfs_reserve_flush_enum flush)
5550 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5552 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5559 int btrfs_block_rsv_check(struct btrfs_root *root,
5560 struct btrfs_block_rsv *block_rsv, int min_factor)
5568 spin_lock(&block_rsv->lock);
5569 num_bytes = div_factor(block_rsv->size, min_factor);
5570 if (block_rsv->reserved >= num_bytes)
5572 spin_unlock(&block_rsv->lock);
5577 int btrfs_block_rsv_refill(struct btrfs_root *root,
5578 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5579 enum btrfs_reserve_flush_enum flush)
5587 spin_lock(&block_rsv->lock);
5588 num_bytes = min_reserved;
5589 if (block_rsv->reserved >= num_bytes)
5592 num_bytes -= block_rsv->reserved;
5593 spin_unlock(&block_rsv->lock);
5598 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5600 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5607 void btrfs_block_rsv_release(struct btrfs_root *root,
5608 struct btrfs_block_rsv *block_rsv,
5611 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5612 if (global_rsv == block_rsv ||
5613 block_rsv->space_info != global_rsv->space_info)
5615 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
5619 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5621 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5622 struct btrfs_space_info *sinfo = block_rsv->space_info;
5626 * The global block rsv is based on the size of the extent tree, the
5627 * checksum tree and the root tree. If the fs is empty we want to set
5628 * it to a minimal amount for safety.
5630 num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) +
5631 btrfs_root_used(&fs_info->csum_root->root_item) +
5632 btrfs_root_used(&fs_info->tree_root->root_item);
5633 num_bytes = max_t(u64, num_bytes, SZ_16M);
5635 spin_lock(&sinfo->lock);
5636 spin_lock(&block_rsv->lock);
5638 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
5640 if (block_rsv->reserved < block_rsv->size) {
5641 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
5642 sinfo->bytes_reserved + sinfo->bytes_readonly +
5643 sinfo->bytes_may_use;
5644 if (sinfo->total_bytes > num_bytes) {
5645 num_bytes = sinfo->total_bytes - num_bytes;
5646 num_bytes = min(num_bytes,
5647 block_rsv->size - block_rsv->reserved);
5648 block_rsv->reserved += num_bytes;
5649 sinfo->bytes_may_use += num_bytes;
5650 trace_btrfs_space_reservation(fs_info, "space_info",
5651 sinfo->flags, num_bytes,
5654 } else if (block_rsv->reserved > block_rsv->size) {
5655 num_bytes = block_rsv->reserved - block_rsv->size;
5656 sinfo->bytes_may_use -= num_bytes;
5657 trace_btrfs_space_reservation(fs_info, "space_info",
5658 sinfo->flags, num_bytes, 0);
5659 block_rsv->reserved = block_rsv->size;
5662 if (block_rsv->reserved == block_rsv->size)
5663 block_rsv->full = 1;
5665 block_rsv->full = 0;
5667 spin_unlock(&block_rsv->lock);
5668 spin_unlock(&sinfo->lock);
5671 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5673 struct btrfs_space_info *space_info;
5675 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5676 fs_info->chunk_block_rsv.space_info = space_info;
5678 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5679 fs_info->global_block_rsv.space_info = space_info;
5680 fs_info->delalloc_block_rsv.space_info = space_info;
5681 fs_info->trans_block_rsv.space_info = space_info;
5682 fs_info->empty_block_rsv.space_info = space_info;
5683 fs_info->delayed_block_rsv.space_info = space_info;
5685 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5686 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5687 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5688 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5689 if (fs_info->quota_root)
5690 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5691 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5693 update_global_block_rsv(fs_info);
5696 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5698 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5700 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
5701 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
5702 WARN_ON(fs_info->trans_block_rsv.size > 0);
5703 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5704 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5705 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5706 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5707 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5710 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
5711 struct btrfs_root *root)
5713 if (!trans->block_rsv)
5716 if (!trans->bytes_reserved)
5719 trace_btrfs_space_reservation(root->fs_info, "transaction",
5720 trans->transid, trans->bytes_reserved, 0);
5721 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
5722 trans->bytes_reserved = 0;
5726 * To be called after all the new block groups attached to the transaction
5727 * handle have been created (btrfs_create_pending_block_groups()).
5729 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5731 struct btrfs_fs_info *fs_info = trans->fs_info;
5733 if (!trans->chunk_bytes_reserved)
5736 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5738 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5739 trans->chunk_bytes_reserved);
5740 trans->chunk_bytes_reserved = 0;
5743 /* Can only return 0 or -ENOSPC */
5744 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5745 struct inode *inode)
5747 struct btrfs_root *root = BTRFS_I(inode)->root;
5749 * We always use trans->block_rsv here as we will have reserved space
5750 * for our orphan when starting the transaction, using get_block_rsv()
5751 * here will sometimes make us choose the wrong block rsv as we could be
5752 * doing a reloc inode for a non refcounted root.
5754 struct btrfs_block_rsv *src_rsv = trans->block_rsv;
5755 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5758 * We need to hold space in order to delete our orphan item once we've
5759 * added it, so this takes the reservation so we can release it later
5760 * when we are truly done with the orphan item.
5762 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5763 trace_btrfs_space_reservation(root->fs_info, "orphan",
5764 btrfs_ino(inode), num_bytes, 1);
5765 return btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1);
5768 void btrfs_orphan_release_metadata(struct inode *inode)
5770 struct btrfs_root *root = BTRFS_I(inode)->root;
5771 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5772 trace_btrfs_space_reservation(root->fs_info, "orphan",
5773 btrfs_ino(inode), num_bytes, 0);
5774 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
5778 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5779 * root: the root of the parent directory
5780 * rsv: block reservation
5781 * items: the number of items that we need do reservation
5782 * qgroup_reserved: used to return the reserved size in qgroup
5784 * This function is used to reserve the space for snapshot/subvolume
5785 * creation and deletion. Those operations are different with the
5786 * common file/directory operations, they change two fs/file trees
5787 * and root tree, the number of items that the qgroup reserves is
5788 * different with the free space reservation. So we can not use
5789 * the space reservation mechanism in start_transaction().
5791 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5792 struct btrfs_block_rsv *rsv,
5794 u64 *qgroup_reserved,
5795 bool use_global_rsv)
5799 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5801 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags)) {
5802 /* One for parent inode, two for dir entries */
5803 num_bytes = 3 * root->nodesize;
5804 ret = btrfs_qgroup_reserve_meta(root, num_bytes);
5811 *qgroup_reserved = num_bytes;
5813 num_bytes = btrfs_calc_trans_metadata_size(root, items);
5814 rsv->space_info = __find_space_info(root->fs_info,
5815 BTRFS_BLOCK_GROUP_METADATA);
5816 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5817 BTRFS_RESERVE_FLUSH_ALL);
5819 if (ret == -ENOSPC && use_global_rsv)
5820 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, 1);
5822 if (ret && *qgroup_reserved)
5823 btrfs_qgroup_free_meta(root, *qgroup_reserved);
5828 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
5829 struct btrfs_block_rsv *rsv,
5830 u64 qgroup_reserved)
5832 btrfs_block_rsv_release(root, rsv, (u64)-1);
5836 * drop_outstanding_extent - drop an outstanding extent
5837 * @inode: the inode we're dropping the extent for
5838 * @num_bytes: the number of bytes we're releasing.
5840 * This is called when we are freeing up an outstanding extent, either called
5841 * after an error or after an extent is written. This will return the number of
5842 * reserved extents that need to be freed. This must be called with
5843 * BTRFS_I(inode)->lock held.
5845 static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes)
5847 unsigned drop_inode_space = 0;
5848 unsigned dropped_extents = 0;
5849 unsigned num_extents = 0;
5851 num_extents = (unsigned)div64_u64(num_bytes +
5852 BTRFS_MAX_EXTENT_SIZE - 1,
5853 BTRFS_MAX_EXTENT_SIZE);
5854 ASSERT(num_extents);
5855 ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents);
5856 BTRFS_I(inode)->outstanding_extents -= num_extents;
5858 if (BTRFS_I(inode)->outstanding_extents == 0 &&
5859 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5860 &BTRFS_I(inode)->runtime_flags))
5861 drop_inode_space = 1;
5864 * If we have more or the same amount of outstanding extents than we have
5865 * reserved then we need to leave the reserved extents count alone.
5867 if (BTRFS_I(inode)->outstanding_extents >=
5868 BTRFS_I(inode)->reserved_extents)
5869 return drop_inode_space;
5871 dropped_extents = BTRFS_I(inode)->reserved_extents -
5872 BTRFS_I(inode)->outstanding_extents;
5873 BTRFS_I(inode)->reserved_extents -= dropped_extents;
5874 return dropped_extents + drop_inode_space;
5878 * calc_csum_metadata_size - return the amount of metadata space that must be
5879 * reserved/freed for the given bytes.
5880 * @inode: the inode we're manipulating
5881 * @num_bytes: the number of bytes in question
5882 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5884 * This adjusts the number of csum_bytes in the inode and then returns the
5885 * correct amount of metadata that must either be reserved or freed. We
5886 * calculate how many checksums we can fit into one leaf and then divide the
5887 * number of bytes that will need to be checksumed by this value to figure out
5888 * how many checksums will be required. If we are adding bytes then the number
5889 * may go up and we will return the number of additional bytes that must be
5890 * reserved. If it is going down we will return the number of bytes that must
5893 * This must be called with BTRFS_I(inode)->lock held.
5895 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
5898 struct btrfs_root *root = BTRFS_I(inode)->root;
5899 u64 old_csums, num_csums;
5901 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5902 BTRFS_I(inode)->csum_bytes == 0)
5905 old_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5907 BTRFS_I(inode)->csum_bytes += num_bytes;
5909 BTRFS_I(inode)->csum_bytes -= num_bytes;
5910 num_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5912 /* No change, no need to reserve more */
5913 if (old_csums == num_csums)
5917 return btrfs_calc_trans_metadata_size(root,
5918 num_csums - old_csums);
5920 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5923 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5925 struct btrfs_root *root = BTRFS_I(inode)->root;
5926 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5929 unsigned nr_extents = 0;
5930 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5932 bool delalloc_lock = true;
5935 bool release_extra = false;
5937 /* If we are a free space inode we need to not flush since we will be in
5938 * the middle of a transaction commit. We also don't need the delalloc
5939 * mutex since we won't race with anybody. We need this mostly to make
5940 * lockdep shut its filthy mouth.
5942 * If we have a transaction open (can happen if we call truncate_block
5943 * from truncate), then we need FLUSH_LIMIT so we don't deadlock.
5945 if (btrfs_is_free_space_inode(inode)) {
5946 flush = BTRFS_RESERVE_NO_FLUSH;
5947 delalloc_lock = false;
5948 } else if (current->journal_info) {
5949 flush = BTRFS_RESERVE_FLUSH_LIMIT;
5952 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5953 btrfs_transaction_in_commit(root->fs_info))
5954 schedule_timeout(1);
5957 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5959 num_bytes = ALIGN(num_bytes, root->sectorsize);
5961 spin_lock(&BTRFS_I(inode)->lock);
5962 nr_extents = (unsigned)div64_u64(num_bytes +
5963 BTRFS_MAX_EXTENT_SIZE - 1,
5964 BTRFS_MAX_EXTENT_SIZE);
5965 BTRFS_I(inode)->outstanding_extents += nr_extents;
5968 if (BTRFS_I(inode)->outstanding_extents >
5969 BTRFS_I(inode)->reserved_extents)
5970 nr_extents += BTRFS_I(inode)->outstanding_extents -
5971 BTRFS_I(inode)->reserved_extents;
5973 /* We always want to reserve a slot for updating the inode. */
5974 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents + 1);
5975 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5976 csum_bytes = BTRFS_I(inode)->csum_bytes;
5977 spin_unlock(&BTRFS_I(inode)->lock);
5979 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags)) {
5980 ret = btrfs_qgroup_reserve_meta(root,
5981 nr_extents * root->nodesize);
5986 ret = btrfs_block_rsv_add(root, block_rsv, to_reserve, flush);
5987 if (unlikely(ret)) {
5988 btrfs_qgroup_free_meta(root, nr_extents * root->nodesize);
5992 spin_lock(&BTRFS_I(inode)->lock);
5993 if (test_and_set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5994 &BTRFS_I(inode)->runtime_flags)) {
5995 to_reserve -= btrfs_calc_trans_metadata_size(root, 1);
5996 release_extra = true;
5998 BTRFS_I(inode)->reserved_extents += nr_extents;
5999 spin_unlock(&BTRFS_I(inode)->lock);
6002 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
6005 trace_btrfs_space_reservation(root->fs_info, "delalloc",
6006 btrfs_ino(inode), to_reserve, 1);
6008 btrfs_block_rsv_release(root, block_rsv,
6009 btrfs_calc_trans_metadata_size(root,
6014 spin_lock(&BTRFS_I(inode)->lock);
6015 dropped = drop_outstanding_extent(inode, num_bytes);
6017 * If the inodes csum_bytes is the same as the original
6018 * csum_bytes then we know we haven't raced with any free()ers
6019 * so we can just reduce our inodes csum bytes and carry on.
6021 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
6022 calc_csum_metadata_size(inode, num_bytes, 0);
6024 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
6028 * This is tricky, but first we need to figure out how much we
6029 * freed from any free-ers that occurred during this
6030 * reservation, so we reset ->csum_bytes to the csum_bytes
6031 * before we dropped our lock, and then call the free for the
6032 * number of bytes that were freed while we were trying our
6035 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
6036 BTRFS_I(inode)->csum_bytes = csum_bytes;
6037 to_free = calc_csum_metadata_size(inode, bytes, 0);
6041 * Now we need to see how much we would have freed had we not
6042 * been making this reservation and our ->csum_bytes were not
6043 * artificially inflated.
6045 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
6046 bytes = csum_bytes - orig_csum_bytes;
6047 bytes = calc_csum_metadata_size(inode, bytes, 0);
6050 * Now reset ->csum_bytes to what it should be. If bytes is
6051 * more than to_free then we would have freed more space had we
6052 * not had an artificially high ->csum_bytes, so we need to free
6053 * the remainder. If bytes is the same or less then we don't
6054 * need to do anything, the other free-ers did the correct
6057 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
6058 if (bytes > to_free)
6059 to_free = bytes - to_free;
6063 spin_unlock(&BTRFS_I(inode)->lock);
6065 to_free += btrfs_calc_trans_metadata_size(root, dropped);
6068 btrfs_block_rsv_release(root, block_rsv, to_free);
6069 trace_btrfs_space_reservation(root->fs_info, "delalloc",
6070 btrfs_ino(inode), to_free, 0);
6073 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
6078 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
6079 * @inode: the inode to release the reservation for
6080 * @num_bytes: the number of bytes we're releasing
6082 * This will release the metadata reservation for an inode. This can be called
6083 * once we complete IO for a given set of bytes to release their metadata
6086 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
6088 struct btrfs_root *root = BTRFS_I(inode)->root;
6092 num_bytes = ALIGN(num_bytes, root->sectorsize);
6093 spin_lock(&BTRFS_I(inode)->lock);
6094 dropped = drop_outstanding_extent(inode, num_bytes);
6097 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
6098 spin_unlock(&BTRFS_I(inode)->lock);
6100 to_free += btrfs_calc_trans_metadata_size(root, dropped);
6102 if (btrfs_is_testing(root->fs_info))
6105 trace_btrfs_space_reservation(root->fs_info, "delalloc",
6106 btrfs_ino(inode), to_free, 0);
6108 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
6113 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6115 * @inode: inode we're writing to
6116 * @start: start range we are writing to
6117 * @len: how long the range we are writing to
6119 * This will do the following things
6121 * o reserve space in data space info for num bytes
6122 * and reserve precious corresponding qgroup space
6123 * (Done in check_data_free_space)
6125 * o reserve space for metadata space, based on the number of outstanding
6126 * extents and how much csums will be needed
6127 * also reserve metadata space in a per root over-reserve method.
6128 * o add to the inodes->delalloc_bytes
6129 * o add it to the fs_info's delalloc inodes list.
6130 * (Above 3 all done in delalloc_reserve_metadata)
6132 * Return 0 for success
6133 * Return <0 for error(-ENOSPC or -EQUOT)
6135 int btrfs_delalloc_reserve_space(struct inode *inode, u64 start, u64 len)
6139 ret = btrfs_check_data_free_space(inode, start, len);
6142 ret = btrfs_delalloc_reserve_metadata(inode, len);
6144 btrfs_free_reserved_data_space(inode, start, len);
6149 * btrfs_delalloc_release_space - release data and metadata space for delalloc
6150 * @inode: inode we're releasing space for
6151 * @start: start position of the space already reserved
6152 * @len: the len of the space already reserved
6154 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
6155 * called in the case that we don't need the metadata AND data reservations
6156 * anymore. So if there is an error or we insert an inline extent.
6158 * This function will release the metadata space that was not used and will
6159 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
6160 * list if there are no delalloc bytes left.
6161 * Also it will handle the qgroup reserved space.
6163 void btrfs_delalloc_release_space(struct inode *inode, u64 start, u64 len)
6165 btrfs_delalloc_release_metadata(inode, len);
6166 btrfs_free_reserved_data_space(inode, start, len);
6169 static int update_block_group(struct btrfs_trans_handle *trans,
6170 struct btrfs_root *root, u64 bytenr,
6171 u64 num_bytes, int alloc)
6173 struct btrfs_block_group_cache *cache = NULL;
6174 struct btrfs_fs_info *info = root->fs_info;
6175 u64 total = num_bytes;
6180 /* block accounting for super block */
6181 spin_lock(&info->delalloc_root_lock);
6182 old_val = btrfs_super_bytes_used(info->super_copy);
6184 old_val += num_bytes;
6186 old_val -= num_bytes;
6187 btrfs_set_super_bytes_used(info->super_copy, old_val);
6188 spin_unlock(&info->delalloc_root_lock);
6191 cache = btrfs_lookup_block_group(info, bytenr);
6194 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
6195 BTRFS_BLOCK_GROUP_RAID1 |
6196 BTRFS_BLOCK_GROUP_RAID10))
6201 * If this block group has free space cache written out, we
6202 * need to make sure to load it if we are removing space. This
6203 * is because we need the unpinning stage to actually add the
6204 * space back to the block group, otherwise we will leak space.
6206 if (!alloc && cache->cached == BTRFS_CACHE_NO)
6207 cache_block_group(cache, 1);
6209 byte_in_group = bytenr - cache->key.objectid;
6210 WARN_ON(byte_in_group > cache->key.offset);
6212 spin_lock(&cache->space_info->lock);
6213 spin_lock(&cache->lock);
6215 if (btrfs_test_opt(root->fs_info, SPACE_CACHE) &&
6216 cache->disk_cache_state < BTRFS_DC_CLEAR)
6217 cache->disk_cache_state = BTRFS_DC_CLEAR;
6219 old_val = btrfs_block_group_used(&cache->item);
6220 num_bytes = min(total, cache->key.offset - byte_in_group);
6222 old_val += num_bytes;
6223 btrfs_set_block_group_used(&cache->item, old_val);
6224 cache->reserved -= num_bytes;
6225 cache->space_info->bytes_reserved -= num_bytes;
6226 cache->space_info->bytes_used += num_bytes;
6227 cache->space_info->disk_used += num_bytes * factor;
6228 spin_unlock(&cache->lock);
6229 spin_unlock(&cache->space_info->lock);
6231 old_val -= num_bytes;
6232 btrfs_set_block_group_used(&cache->item, old_val);
6233 cache->pinned += num_bytes;
6234 cache->space_info->bytes_pinned += num_bytes;
6235 cache->space_info->bytes_used -= num_bytes;
6236 cache->space_info->disk_used -= num_bytes * factor;
6237 spin_unlock(&cache->lock);
6238 spin_unlock(&cache->space_info->lock);
6240 trace_btrfs_space_reservation(root->fs_info, "pinned",
6241 cache->space_info->flags,
6243 set_extent_dirty(info->pinned_extents,
6244 bytenr, bytenr + num_bytes - 1,
6245 GFP_NOFS | __GFP_NOFAIL);
6248 spin_lock(&trans->transaction->dirty_bgs_lock);
6249 if (list_empty(&cache->dirty_list)) {
6250 list_add_tail(&cache->dirty_list,
6251 &trans->transaction->dirty_bgs);
6252 trans->transaction->num_dirty_bgs++;
6253 btrfs_get_block_group(cache);
6255 spin_unlock(&trans->transaction->dirty_bgs_lock);
6258 * No longer have used bytes in this block group, queue it for
6259 * deletion. We do this after adding the block group to the
6260 * dirty list to avoid races between cleaner kthread and space
6263 if (!alloc && old_val == 0) {
6264 spin_lock(&info->unused_bgs_lock);
6265 if (list_empty(&cache->bg_list)) {
6266 btrfs_get_block_group(cache);
6267 list_add_tail(&cache->bg_list,
6270 spin_unlock(&info->unused_bgs_lock);
6273 btrfs_put_block_group(cache);
6275 bytenr += num_bytes;
6280 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
6282 struct btrfs_block_group_cache *cache;
6285 spin_lock(&root->fs_info->block_group_cache_lock);
6286 bytenr = root->fs_info->first_logical_byte;
6287 spin_unlock(&root->fs_info->block_group_cache_lock);
6289 if (bytenr < (u64)-1)
6292 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
6296 bytenr = cache->key.objectid;
6297 btrfs_put_block_group(cache);
6302 static int pin_down_extent(struct btrfs_root *root,
6303 struct btrfs_block_group_cache *cache,
6304 u64 bytenr, u64 num_bytes, int reserved)
6306 spin_lock(&cache->space_info->lock);
6307 spin_lock(&cache->lock);
6308 cache->pinned += num_bytes;
6309 cache->space_info->bytes_pinned += num_bytes;
6311 cache->reserved -= num_bytes;
6312 cache->space_info->bytes_reserved -= num_bytes;
6314 spin_unlock(&cache->lock);
6315 spin_unlock(&cache->space_info->lock);
6317 trace_btrfs_space_reservation(root->fs_info, "pinned",
6318 cache->space_info->flags, num_bytes, 1);
6319 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
6320 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
6325 * this function must be called within transaction
6327 int btrfs_pin_extent(struct btrfs_root *root,
6328 u64 bytenr, u64 num_bytes, int reserved)
6330 struct btrfs_block_group_cache *cache;
6332 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
6333 BUG_ON(!cache); /* Logic error */
6335 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
6337 btrfs_put_block_group(cache);
6342 * this function must be called within transaction
6344 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
6345 u64 bytenr, u64 num_bytes)
6347 struct btrfs_block_group_cache *cache;
6350 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
6355 * pull in the free space cache (if any) so that our pin
6356 * removes the free space from the cache. We have load_only set
6357 * to one because the slow code to read in the free extents does check
6358 * the pinned extents.
6360 cache_block_group(cache, 1);
6362 pin_down_extent(root, cache, bytenr, num_bytes, 0);
6364 /* remove us from the free space cache (if we're there at all) */
6365 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6366 btrfs_put_block_group(cache);
6370 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
6373 struct btrfs_block_group_cache *block_group;
6374 struct btrfs_caching_control *caching_ctl;
6376 block_group = btrfs_lookup_block_group(root->fs_info, start);
6380 cache_block_group(block_group, 0);
6381 caching_ctl = get_caching_control(block_group);
6385 BUG_ON(!block_group_cache_done(block_group));
6386 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6388 mutex_lock(&caching_ctl->mutex);
6390 if (start >= caching_ctl->progress) {
6391 ret = add_excluded_extent(root, start, num_bytes);
6392 } else if (start + num_bytes <= caching_ctl->progress) {
6393 ret = btrfs_remove_free_space(block_group,
6396 num_bytes = caching_ctl->progress - start;
6397 ret = btrfs_remove_free_space(block_group,
6402 num_bytes = (start + num_bytes) -
6403 caching_ctl->progress;
6404 start = caching_ctl->progress;
6405 ret = add_excluded_extent(root, start, num_bytes);
6408 mutex_unlock(&caching_ctl->mutex);
6409 put_caching_control(caching_ctl);
6411 btrfs_put_block_group(block_group);
6415 int btrfs_exclude_logged_extents(struct btrfs_root *log,
6416 struct extent_buffer *eb)
6418 struct btrfs_file_extent_item *item;
6419 struct btrfs_key key;
6423 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
6426 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6427 btrfs_item_key_to_cpu(eb, &key, i);
6428 if (key.type != BTRFS_EXTENT_DATA_KEY)
6430 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6431 found_type = btrfs_file_extent_type(eb, item);
6432 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6434 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6436 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6437 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6438 __exclude_logged_extent(log, key.objectid, key.offset);
6445 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg)
6447 atomic_inc(&bg->reservations);
6450 void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info,
6453 struct btrfs_block_group_cache *bg;
6455 bg = btrfs_lookup_block_group(fs_info, start);
6457 if (atomic_dec_and_test(&bg->reservations))
6458 wake_up_atomic_t(&bg->reservations);
6459 btrfs_put_block_group(bg);
6462 static int btrfs_wait_bg_reservations_atomic_t(atomic_t *a)
6468 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg)
6470 struct btrfs_space_info *space_info = bg->space_info;
6474 if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA))
6478 * Our block group is read only but before we set it to read only,
6479 * some task might have had allocated an extent from it already, but it
6480 * has not yet created a respective ordered extent (and added it to a
6481 * root's list of ordered extents).
6482 * Therefore wait for any task currently allocating extents, since the
6483 * block group's reservations counter is incremented while a read lock
6484 * on the groups' semaphore is held and decremented after releasing
6485 * the read access on that semaphore and creating the ordered extent.
6487 down_write(&space_info->groups_sem);
6488 up_write(&space_info->groups_sem);
6490 wait_on_atomic_t(&bg->reservations,
6491 btrfs_wait_bg_reservations_atomic_t,
6492 TASK_UNINTERRUPTIBLE);
6496 * btrfs_add_reserved_bytes - update the block_group and space info counters
6497 * @cache: The cache we are manipulating
6498 * @ram_bytes: The number of bytes of file content, and will be same to
6499 * @num_bytes except for the compress path.
6500 * @num_bytes: The number of bytes in question
6501 * @delalloc: The blocks are allocated for the delalloc write
6503 * This is called by the allocator when it reserves space. Metadata
6504 * reservations should be called with RESERVE_ALLOC so we do the proper
6505 * ENOSPC accounting. For data we handle the reservation through clearing the
6506 * delalloc bits in the io_tree. We have to do this since we could end up
6507 * allocating less disk space for the amount of data we have reserved in the
6508 * case of compression.
6510 * If this is a reservation and the block group has become read only we cannot
6511 * make the reservation and return -EAGAIN, otherwise this function always
6514 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache,
6515 u64 ram_bytes, u64 num_bytes, int delalloc)
6517 struct btrfs_space_info *space_info = cache->space_info;
6520 spin_lock(&space_info->lock);
6521 spin_lock(&cache->lock);
6525 cache->reserved += num_bytes;
6526 space_info->bytes_reserved += num_bytes;
6528 trace_btrfs_space_reservation(cache->fs_info,
6529 "space_info", space_info->flags,
6531 space_info->bytes_may_use -= ram_bytes;
6533 cache->delalloc_bytes += num_bytes;
6535 spin_unlock(&cache->lock);
6536 spin_unlock(&space_info->lock);
6541 * btrfs_free_reserved_bytes - update the block_group and space info counters
6542 * @cache: The cache we are manipulating
6543 * @num_bytes: The number of bytes in question
6544 * @delalloc: The blocks are allocated for the delalloc write
6546 * This is called by somebody who is freeing space that was never actually used
6547 * on disk. For example if you reserve some space for a new leaf in transaction
6548 * A and before transaction A commits you free that leaf, you call this with
6549 * reserve set to 0 in order to clear the reservation.
6552 static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache,
6553 u64 num_bytes, int delalloc)
6555 struct btrfs_space_info *space_info = cache->space_info;
6558 spin_lock(&space_info->lock);
6559 spin_lock(&cache->lock);
6561 space_info->bytes_readonly += num_bytes;
6562 cache->reserved -= num_bytes;
6563 space_info->bytes_reserved -= num_bytes;
6566 cache->delalloc_bytes -= num_bytes;
6567 spin_unlock(&cache->lock);
6568 spin_unlock(&space_info->lock);
6571 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
6572 struct btrfs_root *root)
6574 struct btrfs_fs_info *fs_info = root->fs_info;
6575 struct btrfs_caching_control *next;
6576 struct btrfs_caching_control *caching_ctl;
6577 struct btrfs_block_group_cache *cache;
6579 down_write(&fs_info->commit_root_sem);
6581 list_for_each_entry_safe(caching_ctl, next,
6582 &fs_info->caching_block_groups, list) {
6583 cache = caching_ctl->block_group;
6584 if (block_group_cache_done(cache)) {
6585 cache->last_byte_to_unpin = (u64)-1;
6586 list_del_init(&caching_ctl->list);
6587 put_caching_control(caching_ctl);
6589 cache->last_byte_to_unpin = caching_ctl->progress;
6593 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6594 fs_info->pinned_extents = &fs_info->freed_extents[1];
6596 fs_info->pinned_extents = &fs_info->freed_extents[0];
6598 up_write(&fs_info->commit_root_sem);
6600 update_global_block_rsv(fs_info);
6604 * Returns the free cluster for the given space info and sets empty_cluster to
6605 * what it should be based on the mount options.
6607 static struct btrfs_free_cluster *
6608 fetch_cluster_info(struct btrfs_root *root, struct btrfs_space_info *space_info,
6611 struct btrfs_free_cluster *ret = NULL;
6612 bool ssd = btrfs_test_opt(root->fs_info, SSD);
6615 if (btrfs_mixed_space_info(space_info))
6619 *empty_cluster = SZ_2M;
6620 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
6621 ret = &root->fs_info->meta_alloc_cluster;
6623 *empty_cluster = SZ_64K;
6624 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && ssd) {
6625 ret = &root->fs_info->data_alloc_cluster;
6631 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end,
6632 const bool return_free_space)
6634 struct btrfs_fs_info *fs_info = root->fs_info;
6635 struct btrfs_block_group_cache *cache = NULL;
6636 struct btrfs_space_info *space_info;
6637 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6638 struct btrfs_free_cluster *cluster = NULL;
6640 u64 total_unpinned = 0;
6641 u64 empty_cluster = 0;
6644 while (start <= end) {
6647 start >= cache->key.objectid + cache->key.offset) {
6649 btrfs_put_block_group(cache);
6651 cache = btrfs_lookup_block_group(fs_info, start);
6652 BUG_ON(!cache); /* Logic error */
6654 cluster = fetch_cluster_info(root,
6657 empty_cluster <<= 1;
6660 len = cache->key.objectid + cache->key.offset - start;
6661 len = min(len, end + 1 - start);
6663 if (start < cache->last_byte_to_unpin) {
6664 len = min(len, cache->last_byte_to_unpin - start);
6665 if (return_free_space)
6666 btrfs_add_free_space(cache, start, len);
6670 total_unpinned += len;
6671 space_info = cache->space_info;
6674 * If this space cluster has been marked as fragmented and we've
6675 * unpinned enough in this block group to potentially allow a
6676 * cluster to be created inside of it go ahead and clear the
6679 if (cluster && cluster->fragmented &&
6680 total_unpinned > empty_cluster) {
6681 spin_lock(&cluster->lock);
6682 cluster->fragmented = 0;
6683 spin_unlock(&cluster->lock);
6686 spin_lock(&space_info->lock);
6687 spin_lock(&cache->lock);
6688 cache->pinned -= len;
6689 space_info->bytes_pinned -= len;
6691 trace_btrfs_space_reservation(fs_info, "pinned",
6692 space_info->flags, len, 0);
6693 space_info->max_extent_size = 0;
6694 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6696 space_info->bytes_readonly += len;
6699 spin_unlock(&cache->lock);
6700 if (!readonly && return_free_space &&
6701 global_rsv->space_info == space_info) {
6703 WARN_ON(!return_free_space);
6704 spin_lock(&global_rsv->lock);
6705 if (!global_rsv->full) {
6706 to_add = min(len, global_rsv->size -
6707 global_rsv->reserved);
6708 global_rsv->reserved += to_add;
6709 space_info->bytes_may_use += to_add;
6710 if (global_rsv->reserved >= global_rsv->size)
6711 global_rsv->full = 1;
6712 trace_btrfs_space_reservation(fs_info,
6718 spin_unlock(&global_rsv->lock);
6719 /* Add to any tickets we may have */
6721 space_info_add_new_bytes(fs_info, space_info,
6724 spin_unlock(&space_info->lock);
6728 btrfs_put_block_group(cache);
6732 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
6733 struct btrfs_root *root)
6735 struct btrfs_fs_info *fs_info = root->fs_info;
6736 struct btrfs_block_group_cache *block_group, *tmp;
6737 struct list_head *deleted_bgs;
6738 struct extent_io_tree *unpin;
6743 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6744 unpin = &fs_info->freed_extents[1];
6746 unpin = &fs_info->freed_extents[0];
6748 while (!trans->aborted) {
6749 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6750 ret = find_first_extent_bit(unpin, 0, &start, &end,
6751 EXTENT_DIRTY, NULL);
6753 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6757 if (btrfs_test_opt(root->fs_info, DISCARD))
6758 ret = btrfs_discard_extent(root, start,
6759 end + 1 - start, NULL);
6761 clear_extent_dirty(unpin, start, end);
6762 unpin_extent_range(root, start, end, true);
6763 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6768 * Transaction is finished. We don't need the lock anymore. We
6769 * do need to clean up the block groups in case of a transaction
6772 deleted_bgs = &trans->transaction->deleted_bgs;
6773 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6777 if (!trans->aborted)
6778 ret = btrfs_discard_extent(root,
6779 block_group->key.objectid,
6780 block_group->key.offset,
6783 list_del_init(&block_group->bg_list);
6784 btrfs_put_block_group_trimming(block_group);
6785 btrfs_put_block_group(block_group);
6788 const char *errstr = btrfs_decode_error(ret);
6790 "Discard failed while removing blockgroup: errno=%d %s\n",
6798 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
6799 u64 owner, u64 root_objectid)
6801 struct btrfs_space_info *space_info;
6804 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6805 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
6806 flags = BTRFS_BLOCK_GROUP_SYSTEM;
6808 flags = BTRFS_BLOCK_GROUP_METADATA;
6810 flags = BTRFS_BLOCK_GROUP_DATA;
6813 space_info = __find_space_info(fs_info, flags);
6814 BUG_ON(!space_info); /* Logic bug */
6815 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
6819 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6820 struct btrfs_root *root,
6821 struct btrfs_delayed_ref_node *node, u64 parent,
6822 u64 root_objectid, u64 owner_objectid,
6823 u64 owner_offset, int refs_to_drop,
6824 struct btrfs_delayed_extent_op *extent_op)
6826 struct btrfs_key key;
6827 struct btrfs_path *path;
6828 struct btrfs_fs_info *info = root->fs_info;
6829 struct btrfs_root *extent_root = info->extent_root;
6830 struct extent_buffer *leaf;
6831 struct btrfs_extent_item *ei;
6832 struct btrfs_extent_inline_ref *iref;
6835 int extent_slot = 0;
6836 int found_extent = 0;
6840 u64 bytenr = node->bytenr;
6841 u64 num_bytes = node->num_bytes;
6843 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6846 path = btrfs_alloc_path();
6850 path->reada = READA_FORWARD;
6851 path->leave_spinning = 1;
6853 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6854 BUG_ON(!is_data && refs_to_drop != 1);
6857 skinny_metadata = 0;
6859 ret = lookup_extent_backref(trans, extent_root, path, &iref,
6860 bytenr, num_bytes, parent,
6861 root_objectid, owner_objectid,
6864 extent_slot = path->slots[0];
6865 while (extent_slot >= 0) {
6866 btrfs_item_key_to_cpu(path->nodes[0], &key,
6868 if (key.objectid != bytenr)
6870 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6871 key.offset == num_bytes) {
6875 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6876 key.offset == owner_objectid) {
6880 if (path->slots[0] - extent_slot > 5)
6884 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6885 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6886 if (found_extent && item_size < sizeof(*ei))
6889 if (!found_extent) {
6891 ret = remove_extent_backref(trans, extent_root, path,
6893 is_data, &last_ref);
6895 btrfs_abort_transaction(trans, ret);
6898 btrfs_release_path(path);
6899 path->leave_spinning = 1;
6901 key.objectid = bytenr;
6902 key.type = BTRFS_EXTENT_ITEM_KEY;
6903 key.offset = num_bytes;
6905 if (!is_data && skinny_metadata) {
6906 key.type = BTRFS_METADATA_ITEM_KEY;
6907 key.offset = owner_objectid;
6910 ret = btrfs_search_slot(trans, extent_root,
6912 if (ret > 0 && skinny_metadata && path->slots[0]) {
6914 * Couldn't find our skinny metadata item,
6915 * see if we have ye olde extent item.
6918 btrfs_item_key_to_cpu(path->nodes[0], &key,
6920 if (key.objectid == bytenr &&
6921 key.type == BTRFS_EXTENT_ITEM_KEY &&
6922 key.offset == num_bytes)
6926 if (ret > 0 && skinny_metadata) {
6927 skinny_metadata = false;
6928 key.objectid = bytenr;
6929 key.type = BTRFS_EXTENT_ITEM_KEY;
6930 key.offset = num_bytes;
6931 btrfs_release_path(path);
6932 ret = btrfs_search_slot(trans, extent_root,
6938 "umm, got %d back from search, was looking for %llu",
6941 btrfs_print_leaf(extent_root,
6945 btrfs_abort_transaction(trans, ret);
6948 extent_slot = path->slots[0];
6950 } else if (WARN_ON(ret == -ENOENT)) {
6951 btrfs_print_leaf(extent_root, path->nodes[0]);
6953 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6954 bytenr, parent, root_objectid, owner_objectid,
6956 btrfs_abort_transaction(trans, ret);
6959 btrfs_abort_transaction(trans, ret);
6963 leaf = path->nodes[0];
6964 item_size = btrfs_item_size_nr(leaf, extent_slot);
6965 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6966 if (item_size < sizeof(*ei)) {
6967 BUG_ON(found_extent || extent_slot != path->slots[0]);
6968 ret = convert_extent_item_v0(trans, extent_root, path,
6971 btrfs_abort_transaction(trans, ret);
6975 btrfs_release_path(path);
6976 path->leave_spinning = 1;
6978 key.objectid = bytenr;
6979 key.type = BTRFS_EXTENT_ITEM_KEY;
6980 key.offset = num_bytes;
6982 ret = btrfs_search_slot(trans, extent_root, &key, path,
6986 "umm, got %d back from search, was looking for %llu",
6988 btrfs_print_leaf(extent_root, path->nodes[0]);
6991 btrfs_abort_transaction(trans, ret);
6995 extent_slot = path->slots[0];
6996 leaf = path->nodes[0];
6997 item_size = btrfs_item_size_nr(leaf, extent_slot);
7000 BUG_ON(item_size < sizeof(*ei));
7001 ei = btrfs_item_ptr(leaf, extent_slot,
7002 struct btrfs_extent_item);
7003 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
7004 key.type == BTRFS_EXTENT_ITEM_KEY) {
7005 struct btrfs_tree_block_info *bi;
7006 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
7007 bi = (struct btrfs_tree_block_info *)(ei + 1);
7008 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
7011 refs = btrfs_extent_refs(leaf, ei);
7012 if (refs < refs_to_drop) {
7014 "trying to drop %d refs but we only have %Lu for bytenr %Lu",
7015 refs_to_drop, refs, bytenr);
7017 btrfs_abort_transaction(trans, ret);
7020 refs -= refs_to_drop;
7024 __run_delayed_extent_op(extent_op, leaf, ei);
7026 * In the case of inline back ref, reference count will
7027 * be updated by remove_extent_backref
7030 BUG_ON(!found_extent);
7032 btrfs_set_extent_refs(leaf, ei, refs);
7033 btrfs_mark_buffer_dirty(leaf);
7036 ret = remove_extent_backref(trans, extent_root, path,
7038 is_data, &last_ref);
7040 btrfs_abort_transaction(trans, ret);
7044 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
7048 BUG_ON(is_data && refs_to_drop !=
7049 extent_data_ref_count(path, iref));
7051 BUG_ON(path->slots[0] != extent_slot);
7053 BUG_ON(path->slots[0] != extent_slot + 1);
7054 path->slots[0] = extent_slot;
7060 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
7063 btrfs_abort_transaction(trans, ret);
7066 btrfs_release_path(path);
7069 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
7071 btrfs_abort_transaction(trans, ret);
7076 ret = add_to_free_space_tree(trans, root->fs_info, bytenr,
7079 btrfs_abort_transaction(trans, ret);
7083 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
7085 btrfs_abort_transaction(trans, ret);
7089 btrfs_release_path(path);
7092 btrfs_free_path(path);
7097 * when we free an block, it is possible (and likely) that we free the last
7098 * delayed ref for that extent as well. This searches the delayed ref tree for
7099 * a given extent, and if there are no other delayed refs to be processed, it
7100 * removes it from the tree.
7102 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
7103 struct btrfs_root *root, u64 bytenr)
7105 struct btrfs_delayed_ref_head *head;
7106 struct btrfs_delayed_ref_root *delayed_refs;
7109 delayed_refs = &trans->transaction->delayed_refs;
7110 spin_lock(&delayed_refs->lock);
7111 head = btrfs_find_delayed_ref_head(trans, bytenr);
7113 goto out_delayed_unlock;
7115 spin_lock(&head->lock);
7116 if (!list_empty(&head->ref_list))
7119 if (head->extent_op) {
7120 if (!head->must_insert_reserved)
7122 btrfs_free_delayed_extent_op(head->extent_op);
7123 head->extent_op = NULL;
7127 * waiting for the lock here would deadlock. If someone else has it
7128 * locked they are already in the process of dropping it anyway
7130 if (!mutex_trylock(&head->mutex))
7134 * at this point we have a head with no other entries. Go
7135 * ahead and process it.
7137 head->node.in_tree = 0;
7138 rb_erase(&head->href_node, &delayed_refs->href_root);
7140 atomic_dec(&delayed_refs->num_entries);
7143 * we don't take a ref on the node because we're removing it from the
7144 * tree, so we just steal the ref the tree was holding.
7146 delayed_refs->num_heads--;
7147 if (head->processing == 0)
7148 delayed_refs->num_heads_ready--;
7149 head->processing = 0;
7150 spin_unlock(&head->lock);
7151 spin_unlock(&delayed_refs->lock);
7153 BUG_ON(head->extent_op);
7154 if (head->must_insert_reserved)
7157 mutex_unlock(&head->mutex);
7158 btrfs_put_delayed_ref(&head->node);
7161 spin_unlock(&head->lock);
7164 spin_unlock(&delayed_refs->lock);
7168 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
7169 struct btrfs_root *root,
7170 struct extent_buffer *buf,
7171 u64 parent, int last_ref)
7176 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7177 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7178 buf->start, buf->len,
7179 parent, root->root_key.objectid,
7180 btrfs_header_level(buf),
7181 BTRFS_DROP_DELAYED_REF, NULL);
7182 BUG_ON(ret); /* -ENOMEM */
7188 if (btrfs_header_generation(buf) == trans->transid) {
7189 struct btrfs_block_group_cache *cache;
7191 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7192 ret = check_ref_cleanup(trans, root, buf->start);
7197 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
7199 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
7200 pin_down_extent(root, cache, buf->start, buf->len, 1);
7201 btrfs_put_block_group(cache);
7205 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
7207 btrfs_add_free_space(cache, buf->start, buf->len);
7208 btrfs_free_reserved_bytes(cache, buf->len, 0);
7209 btrfs_put_block_group(cache);
7210 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
7215 add_pinned_bytes(root->fs_info, buf->len,
7216 btrfs_header_level(buf),
7217 root->root_key.objectid);
7220 * Deleting the buffer, clear the corrupt flag since it doesn't matter
7223 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
7226 /* Can return -ENOMEM */
7227 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
7228 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
7229 u64 owner, u64 offset)
7232 struct btrfs_fs_info *fs_info = root->fs_info;
7234 if (btrfs_is_testing(fs_info))
7237 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
7240 * tree log blocks never actually go into the extent allocation
7241 * tree, just update pinning info and exit early.
7243 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
7244 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
7245 /* unlocks the pinned mutex */
7246 btrfs_pin_extent(root, bytenr, num_bytes, 1);
7248 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
7249 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
7251 parent, root_objectid, (int)owner,
7252 BTRFS_DROP_DELAYED_REF, NULL);
7254 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
7256 parent, root_objectid, owner,
7258 BTRFS_DROP_DELAYED_REF, NULL);
7264 * when we wait for progress in the block group caching, its because
7265 * our allocation attempt failed at least once. So, we must sleep
7266 * and let some progress happen before we try again.
7268 * This function will sleep at least once waiting for new free space to
7269 * show up, and then it will check the block group free space numbers
7270 * for our min num_bytes. Another option is to have it go ahead
7271 * and look in the rbtree for a free extent of a given size, but this
7274 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7275 * any of the information in this block group.
7277 static noinline void
7278 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
7281 struct btrfs_caching_control *caching_ctl;
7283 caching_ctl = get_caching_control(cache);
7287 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
7288 (cache->free_space_ctl->free_space >= num_bytes));
7290 put_caching_control(caching_ctl);
7294 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
7296 struct btrfs_caching_control *caching_ctl;
7299 caching_ctl = get_caching_control(cache);
7301 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
7303 wait_event(caching_ctl->wait, block_group_cache_done(cache));
7304 if (cache->cached == BTRFS_CACHE_ERROR)
7306 put_caching_control(caching_ctl);
7310 int __get_raid_index(u64 flags)
7312 if (flags & BTRFS_BLOCK_GROUP_RAID10)
7313 return BTRFS_RAID_RAID10;
7314 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
7315 return BTRFS_RAID_RAID1;
7316 else if (flags & BTRFS_BLOCK_GROUP_DUP)
7317 return BTRFS_RAID_DUP;
7318 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
7319 return BTRFS_RAID_RAID0;
7320 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
7321 return BTRFS_RAID_RAID5;
7322 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
7323 return BTRFS_RAID_RAID6;
7325 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
7328 int get_block_group_index(struct btrfs_block_group_cache *cache)
7330 return __get_raid_index(cache->flags);
7333 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
7334 [BTRFS_RAID_RAID10] = "raid10",
7335 [BTRFS_RAID_RAID1] = "raid1",
7336 [BTRFS_RAID_DUP] = "dup",
7337 [BTRFS_RAID_RAID0] = "raid0",
7338 [BTRFS_RAID_SINGLE] = "single",
7339 [BTRFS_RAID_RAID5] = "raid5",
7340 [BTRFS_RAID_RAID6] = "raid6",
7343 static const char *get_raid_name(enum btrfs_raid_types type)
7345 if (type >= BTRFS_NR_RAID_TYPES)
7348 return btrfs_raid_type_names[type];
7351 enum btrfs_loop_type {
7352 LOOP_CACHING_NOWAIT = 0,
7353 LOOP_CACHING_WAIT = 1,
7354 LOOP_ALLOC_CHUNK = 2,
7355 LOOP_NO_EMPTY_SIZE = 3,
7359 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
7363 down_read(&cache->data_rwsem);
7367 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
7370 btrfs_get_block_group(cache);
7372 down_read(&cache->data_rwsem);
7375 static struct btrfs_block_group_cache *
7376 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
7377 struct btrfs_free_cluster *cluster,
7380 struct btrfs_block_group_cache *used_bg = NULL;
7382 spin_lock(&cluster->refill_lock);
7384 used_bg = cluster->block_group;
7388 if (used_bg == block_group)
7391 btrfs_get_block_group(used_bg);
7396 if (down_read_trylock(&used_bg->data_rwsem))
7399 spin_unlock(&cluster->refill_lock);
7401 /* We should only have one-level nested. */
7402 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
7404 spin_lock(&cluster->refill_lock);
7405 if (used_bg == cluster->block_group)
7408 up_read(&used_bg->data_rwsem);
7409 btrfs_put_block_group(used_bg);
7414 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
7418 up_read(&cache->data_rwsem);
7419 btrfs_put_block_group(cache);
7423 * walks the btree of allocated extents and find a hole of a given size.
7424 * The key ins is changed to record the hole:
7425 * ins->objectid == start position
7426 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7427 * ins->offset == the size of the hole.
7428 * Any available blocks before search_start are skipped.
7430 * If there is no suitable free space, we will record the max size of
7431 * the free space extent currently.
7433 static noinline int find_free_extent(struct btrfs_root *orig_root,
7434 u64 ram_bytes, u64 num_bytes, u64 empty_size,
7435 u64 hint_byte, struct btrfs_key *ins,
7436 u64 flags, int delalloc)
7439 struct btrfs_root *root = orig_root->fs_info->extent_root;
7440 struct btrfs_free_cluster *last_ptr = NULL;
7441 struct btrfs_block_group_cache *block_group = NULL;
7442 u64 search_start = 0;
7443 u64 max_extent_size = 0;
7444 u64 empty_cluster = 0;
7445 struct btrfs_space_info *space_info;
7447 int index = __get_raid_index(flags);
7448 bool failed_cluster_refill = false;
7449 bool failed_alloc = false;
7450 bool use_cluster = true;
7451 bool have_caching_bg = false;
7452 bool orig_have_caching_bg = false;
7453 bool full_search = false;
7455 WARN_ON(num_bytes < root->sectorsize);
7456 ins->type = BTRFS_EXTENT_ITEM_KEY;
7460 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
7462 space_info = __find_space_info(root->fs_info, flags);
7464 btrfs_err(root->fs_info, "No space info for %llu", flags);
7469 * If our free space is heavily fragmented we may not be able to make
7470 * big contiguous allocations, so instead of doing the expensive search
7471 * for free space, simply return ENOSPC with our max_extent_size so we
7472 * can go ahead and search for a more manageable chunk.
7474 * If our max_extent_size is large enough for our allocation simply
7475 * disable clustering since we will likely not be able to find enough
7476 * space to create a cluster and induce latency trying.
7478 if (unlikely(space_info->max_extent_size)) {
7479 spin_lock(&space_info->lock);
7480 if (space_info->max_extent_size &&
7481 num_bytes > space_info->max_extent_size) {
7482 ins->offset = space_info->max_extent_size;
7483 spin_unlock(&space_info->lock);
7485 } else if (space_info->max_extent_size) {
7486 use_cluster = false;
7488 spin_unlock(&space_info->lock);
7491 last_ptr = fetch_cluster_info(orig_root, space_info, &empty_cluster);
7493 spin_lock(&last_ptr->lock);
7494 if (last_ptr->block_group)
7495 hint_byte = last_ptr->window_start;
7496 if (last_ptr->fragmented) {
7498 * We still set window_start so we can keep track of the
7499 * last place we found an allocation to try and save
7502 hint_byte = last_ptr->window_start;
7503 use_cluster = false;
7505 spin_unlock(&last_ptr->lock);
7508 search_start = max(search_start, first_logical_byte(root, 0));
7509 search_start = max(search_start, hint_byte);
7510 if (search_start == hint_byte) {
7511 block_group = btrfs_lookup_block_group(root->fs_info,
7514 * we don't want to use the block group if it doesn't match our
7515 * allocation bits, or if its not cached.
7517 * However if we are re-searching with an ideal block group
7518 * picked out then we don't care that the block group is cached.
7520 if (block_group && block_group_bits(block_group, flags) &&
7521 block_group->cached != BTRFS_CACHE_NO) {
7522 down_read(&space_info->groups_sem);
7523 if (list_empty(&block_group->list) ||
7526 * someone is removing this block group,
7527 * we can't jump into the have_block_group
7528 * target because our list pointers are not
7531 btrfs_put_block_group(block_group);
7532 up_read(&space_info->groups_sem);
7534 index = get_block_group_index(block_group);
7535 btrfs_lock_block_group(block_group, delalloc);
7536 goto have_block_group;
7538 } else if (block_group) {
7539 btrfs_put_block_group(block_group);
7543 have_caching_bg = false;
7544 if (index == 0 || index == __get_raid_index(flags))
7546 down_read(&space_info->groups_sem);
7547 list_for_each_entry(block_group, &space_info->block_groups[index],
7552 btrfs_grab_block_group(block_group, delalloc);
7553 search_start = block_group->key.objectid;
7556 * this can happen if we end up cycling through all the
7557 * raid types, but we want to make sure we only allocate
7558 * for the proper type.
7560 if (!block_group_bits(block_group, flags)) {
7561 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7562 BTRFS_BLOCK_GROUP_RAID1 |
7563 BTRFS_BLOCK_GROUP_RAID5 |
7564 BTRFS_BLOCK_GROUP_RAID6 |
7565 BTRFS_BLOCK_GROUP_RAID10;
7568 * if they asked for extra copies and this block group
7569 * doesn't provide them, bail. This does allow us to
7570 * fill raid0 from raid1.
7572 if ((flags & extra) && !(block_group->flags & extra))
7577 cached = block_group_cache_done(block_group);
7578 if (unlikely(!cached)) {
7579 have_caching_bg = true;
7580 ret = cache_block_group(block_group, 0);
7585 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7587 if (unlikely(block_group->ro))
7591 * Ok we want to try and use the cluster allocator, so
7594 if (last_ptr && use_cluster) {
7595 struct btrfs_block_group_cache *used_block_group;
7596 unsigned long aligned_cluster;
7598 * the refill lock keeps out other
7599 * people trying to start a new cluster
7601 used_block_group = btrfs_lock_cluster(block_group,
7604 if (!used_block_group)
7605 goto refill_cluster;
7607 if (used_block_group != block_group &&
7608 (used_block_group->ro ||
7609 !block_group_bits(used_block_group, flags)))
7610 goto release_cluster;
7612 offset = btrfs_alloc_from_cluster(used_block_group,
7615 used_block_group->key.objectid,
7618 /* we have a block, we're done */
7619 spin_unlock(&last_ptr->refill_lock);
7620 trace_btrfs_reserve_extent_cluster(root,
7622 search_start, num_bytes);
7623 if (used_block_group != block_group) {
7624 btrfs_release_block_group(block_group,
7626 block_group = used_block_group;
7631 WARN_ON(last_ptr->block_group != used_block_group);
7633 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7634 * set up a new clusters, so lets just skip it
7635 * and let the allocator find whatever block
7636 * it can find. If we reach this point, we
7637 * will have tried the cluster allocator
7638 * plenty of times and not have found
7639 * anything, so we are likely way too
7640 * fragmented for the clustering stuff to find
7643 * However, if the cluster is taken from the
7644 * current block group, release the cluster
7645 * first, so that we stand a better chance of
7646 * succeeding in the unclustered
7648 if (loop >= LOOP_NO_EMPTY_SIZE &&
7649 used_block_group != block_group) {
7650 spin_unlock(&last_ptr->refill_lock);
7651 btrfs_release_block_group(used_block_group,
7653 goto unclustered_alloc;
7657 * this cluster didn't work out, free it and
7660 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7662 if (used_block_group != block_group)
7663 btrfs_release_block_group(used_block_group,
7666 if (loop >= LOOP_NO_EMPTY_SIZE) {
7667 spin_unlock(&last_ptr->refill_lock);
7668 goto unclustered_alloc;
7671 aligned_cluster = max_t(unsigned long,
7672 empty_cluster + empty_size,
7673 block_group->full_stripe_len);
7675 /* allocate a cluster in this block group */
7676 ret = btrfs_find_space_cluster(root, block_group,
7677 last_ptr, search_start,
7682 * now pull our allocation out of this
7685 offset = btrfs_alloc_from_cluster(block_group,
7691 /* we found one, proceed */
7692 spin_unlock(&last_ptr->refill_lock);
7693 trace_btrfs_reserve_extent_cluster(root,
7694 block_group, search_start,
7698 } else if (!cached && loop > LOOP_CACHING_NOWAIT
7699 && !failed_cluster_refill) {
7700 spin_unlock(&last_ptr->refill_lock);
7702 failed_cluster_refill = true;
7703 wait_block_group_cache_progress(block_group,
7704 num_bytes + empty_cluster + empty_size);
7705 goto have_block_group;
7709 * at this point we either didn't find a cluster
7710 * or we weren't able to allocate a block from our
7711 * cluster. Free the cluster we've been trying
7712 * to use, and go to the next block group
7714 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7715 spin_unlock(&last_ptr->refill_lock);
7721 * We are doing an unclustered alloc, set the fragmented flag so
7722 * we don't bother trying to setup a cluster again until we get
7725 if (unlikely(last_ptr)) {
7726 spin_lock(&last_ptr->lock);
7727 last_ptr->fragmented = 1;
7728 spin_unlock(&last_ptr->lock);
7730 spin_lock(&block_group->free_space_ctl->tree_lock);
7732 block_group->free_space_ctl->free_space <
7733 num_bytes + empty_cluster + empty_size) {
7734 if (block_group->free_space_ctl->free_space >
7737 block_group->free_space_ctl->free_space;
7738 spin_unlock(&block_group->free_space_ctl->tree_lock);
7741 spin_unlock(&block_group->free_space_ctl->tree_lock);
7743 offset = btrfs_find_space_for_alloc(block_group, search_start,
7744 num_bytes, empty_size,
7747 * If we didn't find a chunk, and we haven't failed on this
7748 * block group before, and this block group is in the middle of
7749 * caching and we are ok with waiting, then go ahead and wait
7750 * for progress to be made, and set failed_alloc to true.
7752 * If failed_alloc is true then we've already waited on this
7753 * block group once and should move on to the next block group.
7755 if (!offset && !failed_alloc && !cached &&
7756 loop > LOOP_CACHING_NOWAIT) {
7757 wait_block_group_cache_progress(block_group,
7758 num_bytes + empty_size);
7759 failed_alloc = true;
7760 goto have_block_group;
7761 } else if (!offset) {
7765 search_start = ALIGN(offset, root->stripesize);
7767 /* move on to the next group */
7768 if (search_start + num_bytes >
7769 block_group->key.objectid + block_group->key.offset) {
7770 btrfs_add_free_space(block_group, offset, num_bytes);
7774 if (offset < search_start)
7775 btrfs_add_free_space(block_group, offset,
7776 search_start - offset);
7777 BUG_ON(offset > search_start);
7779 ret = btrfs_add_reserved_bytes(block_group, ram_bytes,
7780 num_bytes, delalloc);
7781 if (ret == -EAGAIN) {
7782 btrfs_add_free_space(block_group, offset, num_bytes);
7785 btrfs_inc_block_group_reservations(block_group);
7787 /* we are all good, lets return */
7788 ins->objectid = search_start;
7789 ins->offset = num_bytes;
7791 trace_btrfs_reserve_extent(orig_root, block_group,
7792 search_start, num_bytes);
7793 btrfs_release_block_group(block_group, delalloc);
7796 failed_cluster_refill = false;
7797 failed_alloc = false;
7798 BUG_ON(index != get_block_group_index(block_group));
7799 btrfs_release_block_group(block_group, delalloc);
7801 up_read(&space_info->groups_sem);
7803 if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg
7804 && !orig_have_caching_bg)
7805 orig_have_caching_bg = true;
7807 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7810 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7814 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7815 * caching kthreads as we move along
7816 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7817 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7818 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7821 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7823 if (loop == LOOP_CACHING_NOWAIT) {
7825 * We want to skip the LOOP_CACHING_WAIT step if we
7826 * don't have any uncached bgs and we've already done a
7827 * full search through.
7829 if (orig_have_caching_bg || !full_search)
7830 loop = LOOP_CACHING_WAIT;
7832 loop = LOOP_ALLOC_CHUNK;
7837 if (loop == LOOP_ALLOC_CHUNK) {
7838 struct btrfs_trans_handle *trans;
7841 trans = current->journal_info;
7845 trans = btrfs_join_transaction(root);
7847 if (IS_ERR(trans)) {
7848 ret = PTR_ERR(trans);
7852 ret = do_chunk_alloc(trans, root, flags,
7856 * If we can't allocate a new chunk we've already looped
7857 * through at least once, move on to the NO_EMPTY_SIZE
7861 loop = LOOP_NO_EMPTY_SIZE;
7864 * Do not bail out on ENOSPC since we
7865 * can do more things.
7867 if (ret < 0 && ret != -ENOSPC)
7868 btrfs_abort_transaction(trans, ret);
7872 btrfs_end_transaction(trans, root);
7877 if (loop == LOOP_NO_EMPTY_SIZE) {
7879 * Don't loop again if we already have no empty_size and
7882 if (empty_size == 0 &&
7883 empty_cluster == 0) {
7892 } else if (!ins->objectid) {
7894 } else if (ins->objectid) {
7895 if (!use_cluster && last_ptr) {
7896 spin_lock(&last_ptr->lock);
7897 last_ptr->window_start = ins->objectid;
7898 spin_unlock(&last_ptr->lock);
7903 if (ret == -ENOSPC) {
7904 spin_lock(&space_info->lock);
7905 space_info->max_extent_size = max_extent_size;
7906 spin_unlock(&space_info->lock);
7907 ins->offset = max_extent_size;
7912 static void dump_space_info(struct btrfs_fs_info *fs_info,
7913 struct btrfs_space_info *info, u64 bytes,
7914 int dump_block_groups)
7916 struct btrfs_block_group_cache *cache;
7919 spin_lock(&info->lock);
7920 btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull",
7922 info->total_bytes - info->bytes_used - info->bytes_pinned -
7923 info->bytes_reserved - info->bytes_readonly -
7924 info->bytes_may_use, (info->full) ? "" : "not ");
7926 "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu",
7927 info->total_bytes, info->bytes_used, info->bytes_pinned,
7928 info->bytes_reserved, info->bytes_may_use,
7929 info->bytes_readonly);
7930 spin_unlock(&info->lock);
7932 if (!dump_block_groups)
7935 down_read(&info->groups_sem);
7937 list_for_each_entry(cache, &info->block_groups[index], list) {
7938 spin_lock(&cache->lock);
7940 "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s",
7941 cache->key.objectid, cache->key.offset,
7942 btrfs_block_group_used(&cache->item), cache->pinned,
7943 cache->reserved, cache->ro ? "[readonly]" : "");
7944 btrfs_dump_free_space(cache, bytes);
7945 spin_unlock(&cache->lock);
7947 if (++index < BTRFS_NR_RAID_TYPES)
7949 up_read(&info->groups_sem);
7952 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
7953 u64 num_bytes, u64 min_alloc_size,
7954 u64 empty_size, u64 hint_byte,
7955 struct btrfs_key *ins, int is_data, int delalloc)
7957 struct btrfs_fs_info *fs_info = root->fs_info;
7958 bool final_tried = num_bytes == min_alloc_size;
7962 flags = btrfs_get_alloc_profile(root, is_data);
7964 WARN_ON(num_bytes < root->sectorsize);
7965 ret = find_free_extent(root, ram_bytes, num_bytes, empty_size,
7966 hint_byte, ins, flags, delalloc);
7967 if (!ret && !is_data) {
7968 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
7969 } else if (ret == -ENOSPC) {
7970 if (!final_tried && ins->offset) {
7971 num_bytes = min(num_bytes >> 1, ins->offset);
7972 num_bytes = round_down(num_bytes, root->sectorsize);
7973 num_bytes = max(num_bytes, min_alloc_size);
7974 ram_bytes = num_bytes;
7975 if (num_bytes == min_alloc_size)
7978 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
7979 struct btrfs_space_info *sinfo;
7981 sinfo = __find_space_info(fs_info, flags);
7982 btrfs_err(root->fs_info,
7983 "allocation failed flags %llu, wanted %llu",
7986 dump_space_info(fs_info, sinfo, num_bytes, 1);
7993 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
7995 int pin, int delalloc)
7997 struct btrfs_block_group_cache *cache;
8000 cache = btrfs_lookup_block_group(root->fs_info, start);
8002 btrfs_err(root->fs_info, "Unable to find block group for %llu",
8008 pin_down_extent(root, cache, start, len, 1);
8010 if (btrfs_test_opt(root->fs_info, DISCARD))
8011 ret = btrfs_discard_extent(root, start, len, NULL);
8012 btrfs_add_free_space(cache, start, len);
8013 btrfs_free_reserved_bytes(cache, len, delalloc);
8014 trace_btrfs_reserved_extent_free(root, start, len);
8017 btrfs_put_block_group(cache);
8021 int btrfs_free_reserved_extent(struct btrfs_root *root,
8022 u64 start, u64 len, int delalloc)
8024 return __btrfs_free_reserved_extent(root, start, len, 0, delalloc);
8027 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
8030 return __btrfs_free_reserved_extent(root, start, len, 1, 0);
8033 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8034 struct btrfs_root *root,
8035 u64 parent, u64 root_objectid,
8036 u64 flags, u64 owner, u64 offset,
8037 struct btrfs_key *ins, int ref_mod)
8040 struct btrfs_fs_info *fs_info = root->fs_info;
8041 struct btrfs_extent_item *extent_item;
8042 struct btrfs_extent_inline_ref *iref;
8043 struct btrfs_path *path;
8044 struct extent_buffer *leaf;
8049 type = BTRFS_SHARED_DATA_REF_KEY;
8051 type = BTRFS_EXTENT_DATA_REF_KEY;
8053 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
8055 path = btrfs_alloc_path();
8059 path->leave_spinning = 1;
8060 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8063 btrfs_free_path(path);
8067 leaf = path->nodes[0];
8068 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8069 struct btrfs_extent_item);
8070 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
8071 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8072 btrfs_set_extent_flags(leaf, extent_item,
8073 flags | BTRFS_EXTENT_FLAG_DATA);
8075 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8076 btrfs_set_extent_inline_ref_type(leaf, iref, type);
8078 struct btrfs_shared_data_ref *ref;
8079 ref = (struct btrfs_shared_data_ref *)(iref + 1);
8080 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8081 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
8083 struct btrfs_extent_data_ref *ref;
8084 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
8085 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
8086 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
8087 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
8088 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
8091 btrfs_mark_buffer_dirty(path->nodes[0]);
8092 btrfs_free_path(path);
8094 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
8099 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
8100 if (ret) { /* -ENOENT, logic error */
8101 btrfs_err(fs_info, "update block group failed for %llu %llu",
8102 ins->objectid, ins->offset);
8105 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
8109 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
8110 struct btrfs_root *root,
8111 u64 parent, u64 root_objectid,
8112 u64 flags, struct btrfs_disk_key *key,
8113 int level, struct btrfs_key *ins)
8116 struct btrfs_fs_info *fs_info = root->fs_info;
8117 struct btrfs_extent_item *extent_item;
8118 struct btrfs_tree_block_info *block_info;
8119 struct btrfs_extent_inline_ref *iref;
8120 struct btrfs_path *path;
8121 struct extent_buffer *leaf;
8122 u32 size = sizeof(*extent_item) + sizeof(*iref);
8123 u64 num_bytes = ins->offset;
8124 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
8127 if (!skinny_metadata)
8128 size += sizeof(*block_info);
8130 path = btrfs_alloc_path();
8132 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
8137 path->leave_spinning = 1;
8138 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8141 btrfs_free_path(path);
8142 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
8147 leaf = path->nodes[0];
8148 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8149 struct btrfs_extent_item);
8150 btrfs_set_extent_refs(leaf, extent_item, 1);
8151 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8152 btrfs_set_extent_flags(leaf, extent_item,
8153 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
8155 if (skinny_metadata) {
8156 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8157 num_bytes = root->nodesize;
8159 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
8160 btrfs_set_tree_block_key(leaf, block_info, key);
8161 btrfs_set_tree_block_level(leaf, block_info, level);
8162 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
8166 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
8167 btrfs_set_extent_inline_ref_type(leaf, iref,
8168 BTRFS_SHARED_BLOCK_REF_KEY);
8169 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8171 btrfs_set_extent_inline_ref_type(leaf, iref,
8172 BTRFS_TREE_BLOCK_REF_KEY);
8173 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
8176 btrfs_mark_buffer_dirty(leaf);
8177 btrfs_free_path(path);
8179 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
8184 ret = update_block_group(trans, root, ins->objectid, root->nodesize,
8186 if (ret) { /* -ENOENT, logic error */
8187 btrfs_err(fs_info, "update block group failed for %llu %llu",
8188 ins->objectid, ins->offset);
8192 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize);
8196 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8197 struct btrfs_root *root,
8198 u64 root_objectid, u64 owner,
8199 u64 offset, u64 ram_bytes,
8200 struct btrfs_key *ins)
8204 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
8206 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
8208 root_objectid, owner, offset,
8209 ram_bytes, BTRFS_ADD_DELAYED_EXTENT,
8215 * this is used by the tree logging recovery code. It records that
8216 * an extent has been allocated and makes sure to clear the free
8217 * space cache bits as well
8219 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
8220 struct btrfs_root *root,
8221 u64 root_objectid, u64 owner, u64 offset,
8222 struct btrfs_key *ins)
8225 struct btrfs_block_group_cache *block_group;
8226 struct btrfs_space_info *space_info;
8229 * Mixed block groups will exclude before processing the log so we only
8230 * need to do the exclude dance if this fs isn't mixed.
8232 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
8233 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
8238 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
8242 space_info = block_group->space_info;
8243 spin_lock(&space_info->lock);
8244 spin_lock(&block_group->lock);
8245 space_info->bytes_reserved += ins->offset;
8246 block_group->reserved += ins->offset;
8247 spin_unlock(&block_group->lock);
8248 spin_unlock(&space_info->lock);
8250 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
8251 0, owner, offset, ins, 1);
8252 btrfs_put_block_group(block_group);
8256 static struct extent_buffer *
8257 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
8258 u64 bytenr, int level)
8260 struct extent_buffer *buf;
8262 buf = btrfs_find_create_tree_block(root, bytenr);
8266 btrfs_set_header_generation(buf, trans->transid);
8267 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
8268 btrfs_tree_lock(buf);
8269 clean_tree_block(trans, root->fs_info, buf);
8270 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
8272 btrfs_set_lock_blocking(buf);
8273 set_extent_buffer_uptodate(buf);
8275 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
8276 buf->log_index = root->log_transid % 2;
8278 * we allow two log transactions at a time, use different
8279 * EXENT bit to differentiate dirty pages.
8281 if (buf->log_index == 0)
8282 set_extent_dirty(&root->dirty_log_pages, buf->start,
8283 buf->start + buf->len - 1, GFP_NOFS);
8285 set_extent_new(&root->dirty_log_pages, buf->start,
8286 buf->start + buf->len - 1);
8288 buf->log_index = -1;
8289 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
8290 buf->start + buf->len - 1, GFP_NOFS);
8292 trans->dirty = true;
8293 /* this returns a buffer locked for blocking */
8297 static struct btrfs_block_rsv *
8298 use_block_rsv(struct btrfs_trans_handle *trans,
8299 struct btrfs_root *root, u32 blocksize)
8301 struct btrfs_block_rsv *block_rsv;
8302 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
8304 bool global_updated = false;
8306 block_rsv = get_block_rsv(trans, root);
8308 if (unlikely(block_rsv->size == 0))
8311 ret = block_rsv_use_bytes(block_rsv, blocksize);
8315 if (block_rsv->failfast)
8316 return ERR_PTR(ret);
8318 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
8319 global_updated = true;
8320 update_global_block_rsv(root->fs_info);
8324 if (btrfs_test_opt(root->fs_info, ENOSPC_DEBUG)) {
8325 static DEFINE_RATELIMIT_STATE(_rs,
8326 DEFAULT_RATELIMIT_INTERVAL * 10,
8327 /*DEFAULT_RATELIMIT_BURST*/ 1);
8328 if (__ratelimit(&_rs))
8330 "BTRFS: block rsv returned %d\n", ret);
8333 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
8334 BTRFS_RESERVE_NO_FLUSH);
8338 * If we couldn't reserve metadata bytes try and use some from
8339 * the global reserve if its space type is the same as the global
8342 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
8343 block_rsv->space_info == global_rsv->space_info) {
8344 ret = block_rsv_use_bytes(global_rsv, blocksize);
8348 return ERR_PTR(ret);
8351 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
8352 struct btrfs_block_rsv *block_rsv, u32 blocksize)
8354 block_rsv_add_bytes(block_rsv, blocksize, 0);
8355 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
8359 * finds a free extent and does all the dirty work required for allocation
8360 * returns the tree buffer or an ERR_PTR on error.
8362 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
8363 struct btrfs_root *root,
8364 u64 parent, u64 root_objectid,
8365 struct btrfs_disk_key *key, int level,
8366 u64 hint, u64 empty_size)
8368 struct btrfs_key ins;
8369 struct btrfs_block_rsv *block_rsv;
8370 struct extent_buffer *buf;
8371 struct btrfs_delayed_extent_op *extent_op;
8374 u32 blocksize = root->nodesize;
8375 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
8378 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8379 if (btrfs_is_testing(root->fs_info)) {
8380 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
8383 root->alloc_bytenr += blocksize;
8388 block_rsv = use_block_rsv(trans, root, blocksize);
8389 if (IS_ERR(block_rsv))
8390 return ERR_CAST(block_rsv);
8392 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
8393 empty_size, hint, &ins, 0, 0);
8397 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
8400 goto out_free_reserved;
8403 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
8405 parent = ins.objectid;
8406 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
8410 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
8411 extent_op = btrfs_alloc_delayed_extent_op();
8417 memcpy(&extent_op->key, key, sizeof(extent_op->key));
8419 memset(&extent_op->key, 0, sizeof(extent_op->key));
8420 extent_op->flags_to_set = flags;
8421 extent_op->update_key = skinny_metadata ? false : true;
8422 extent_op->update_flags = true;
8423 extent_op->is_data = false;
8424 extent_op->level = level;
8426 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
8427 ins.objectid, ins.offset,
8428 parent, root_objectid, level,
8429 BTRFS_ADD_DELAYED_EXTENT,
8432 goto out_free_delayed;
8437 btrfs_free_delayed_extent_op(extent_op);
8439 free_extent_buffer(buf);
8441 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 0);
8443 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
8444 return ERR_PTR(ret);
8447 struct walk_control {
8448 u64 refs[BTRFS_MAX_LEVEL];
8449 u64 flags[BTRFS_MAX_LEVEL];
8450 struct btrfs_key update_progress;
8461 #define DROP_REFERENCE 1
8462 #define UPDATE_BACKREF 2
8464 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
8465 struct btrfs_root *root,
8466 struct walk_control *wc,
8467 struct btrfs_path *path)
8474 struct btrfs_key key;
8475 struct extent_buffer *eb;
8480 if (path->slots[wc->level] < wc->reada_slot) {
8481 wc->reada_count = wc->reada_count * 2 / 3;
8482 wc->reada_count = max(wc->reada_count, 2);
8484 wc->reada_count = wc->reada_count * 3 / 2;
8485 wc->reada_count = min_t(int, wc->reada_count,
8486 BTRFS_NODEPTRS_PER_BLOCK(root));
8489 eb = path->nodes[wc->level];
8490 nritems = btrfs_header_nritems(eb);
8492 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
8493 if (nread >= wc->reada_count)
8497 bytenr = btrfs_node_blockptr(eb, slot);
8498 generation = btrfs_node_ptr_generation(eb, slot);
8500 if (slot == path->slots[wc->level])
8503 if (wc->stage == UPDATE_BACKREF &&
8504 generation <= root->root_key.offset)
8507 /* We don't lock the tree block, it's OK to be racy here */
8508 ret = btrfs_lookup_extent_info(trans, root, bytenr,
8509 wc->level - 1, 1, &refs,
8511 /* We don't care about errors in readahead. */
8516 if (wc->stage == DROP_REFERENCE) {
8520 if (wc->level == 1 &&
8521 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8523 if (!wc->update_ref ||
8524 generation <= root->root_key.offset)
8526 btrfs_node_key_to_cpu(eb, &key, slot);
8527 ret = btrfs_comp_cpu_keys(&key,
8528 &wc->update_progress);
8532 if (wc->level == 1 &&
8533 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8537 readahead_tree_block(root, bytenr);
8540 wc->reada_slot = slot;
8543 static int account_leaf_items(struct btrfs_trans_handle *trans,
8544 struct btrfs_root *root,
8545 struct extent_buffer *eb)
8547 int nr = btrfs_header_nritems(eb);
8548 int i, extent_type, ret;
8549 struct btrfs_key key;
8550 struct btrfs_file_extent_item *fi;
8551 u64 bytenr, num_bytes;
8553 /* We can be called directly from walk_up_proc() */
8554 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags))
8557 for (i = 0; i < nr; i++) {
8558 btrfs_item_key_to_cpu(eb, &key, i);
8560 if (key.type != BTRFS_EXTENT_DATA_KEY)
8563 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
8564 /* filter out non qgroup-accountable extents */
8565 extent_type = btrfs_file_extent_type(eb, fi);
8567 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
8570 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
8574 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
8576 ret = btrfs_qgroup_insert_dirty_extent(trans, root->fs_info,
8577 bytenr, num_bytes, GFP_NOFS);
8585 * Walk up the tree from the bottom, freeing leaves and any interior
8586 * nodes which have had all slots visited. If a node (leaf or
8587 * interior) is freed, the node above it will have it's slot
8588 * incremented. The root node will never be freed.
8590 * At the end of this function, we should have a path which has all
8591 * slots incremented to the next position for a search. If we need to
8592 * read a new node it will be NULL and the node above it will have the
8593 * correct slot selected for a later read.
8595 * If we increment the root nodes slot counter past the number of
8596 * elements, 1 is returned to signal completion of the search.
8598 static int adjust_slots_upwards(struct btrfs_root *root,
8599 struct btrfs_path *path, int root_level)
8603 struct extent_buffer *eb;
8605 if (root_level == 0)
8608 while (level <= root_level) {
8609 eb = path->nodes[level];
8610 nr = btrfs_header_nritems(eb);
8611 path->slots[level]++;
8612 slot = path->slots[level];
8613 if (slot >= nr || level == 0) {
8615 * Don't free the root - we will detect this
8616 * condition after our loop and return a
8617 * positive value for caller to stop walking the tree.
8619 if (level != root_level) {
8620 btrfs_tree_unlock_rw(eb, path->locks[level]);
8621 path->locks[level] = 0;
8623 free_extent_buffer(eb);
8624 path->nodes[level] = NULL;
8625 path->slots[level] = 0;
8629 * We have a valid slot to walk back down
8630 * from. Stop here so caller can process these
8639 eb = path->nodes[root_level];
8640 if (path->slots[root_level] >= btrfs_header_nritems(eb))
8647 * root_eb is the subtree root and is locked before this function is called.
8649 static int account_shared_subtree(struct btrfs_trans_handle *trans,
8650 struct btrfs_root *root,
8651 struct extent_buffer *root_eb,
8657 struct extent_buffer *eb = root_eb;
8658 struct btrfs_path *path = NULL;
8660 BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
8661 BUG_ON(root_eb == NULL);
8663 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags))
8666 if (!extent_buffer_uptodate(root_eb)) {
8667 ret = btrfs_read_buffer(root_eb, root_gen);
8672 if (root_level == 0) {
8673 ret = account_leaf_items(trans, root, root_eb);
8677 path = btrfs_alloc_path();
8682 * Walk down the tree. Missing extent blocks are filled in as
8683 * we go. Metadata is accounted every time we read a new
8686 * When we reach a leaf, we account for file extent items in it,
8687 * walk back up the tree (adjusting slot pointers as we go)
8688 * and restart the search process.
8690 extent_buffer_get(root_eb); /* For path */
8691 path->nodes[root_level] = root_eb;
8692 path->slots[root_level] = 0;
8693 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
8696 while (level >= 0) {
8697 if (path->nodes[level] == NULL) {
8702 /* We need to get child blockptr/gen from
8703 * parent before we can read it. */
8704 eb = path->nodes[level + 1];
8705 parent_slot = path->slots[level + 1];
8706 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
8707 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
8709 eb = read_tree_block(root, child_bytenr, child_gen);
8713 } else if (!extent_buffer_uptodate(eb)) {
8714 free_extent_buffer(eb);
8719 path->nodes[level] = eb;
8720 path->slots[level] = 0;
8722 btrfs_tree_read_lock(eb);
8723 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
8724 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
8726 ret = btrfs_qgroup_insert_dirty_extent(trans,
8727 root->fs_info, child_bytenr,
8728 root->nodesize, GFP_NOFS);
8734 ret = account_leaf_items(trans, root, path->nodes[level]);
8738 /* Nonzero return here means we completed our search */
8739 ret = adjust_slots_upwards(root, path, root_level);
8743 /* Restart search with new slots */
8752 btrfs_free_path(path);
8758 * helper to process tree block while walking down the tree.
8760 * when wc->stage == UPDATE_BACKREF, this function updates
8761 * back refs for pointers in the block.
8763 * NOTE: return value 1 means we should stop walking down.
8765 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8766 struct btrfs_root *root,
8767 struct btrfs_path *path,
8768 struct walk_control *wc, int lookup_info)
8770 int level = wc->level;
8771 struct extent_buffer *eb = path->nodes[level];
8772 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8775 if (wc->stage == UPDATE_BACKREF &&
8776 btrfs_header_owner(eb) != root->root_key.objectid)
8780 * when reference count of tree block is 1, it won't increase
8781 * again. once full backref flag is set, we never clear it.
8784 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8785 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8786 BUG_ON(!path->locks[level]);
8787 ret = btrfs_lookup_extent_info(trans, root,
8788 eb->start, level, 1,
8791 BUG_ON(ret == -ENOMEM);
8794 BUG_ON(wc->refs[level] == 0);
8797 if (wc->stage == DROP_REFERENCE) {
8798 if (wc->refs[level] > 1)
8801 if (path->locks[level] && !wc->keep_locks) {
8802 btrfs_tree_unlock_rw(eb, path->locks[level]);
8803 path->locks[level] = 0;
8808 /* wc->stage == UPDATE_BACKREF */
8809 if (!(wc->flags[level] & flag)) {
8810 BUG_ON(!path->locks[level]);
8811 ret = btrfs_inc_ref(trans, root, eb, 1);
8812 BUG_ON(ret); /* -ENOMEM */
8813 ret = btrfs_dec_ref(trans, root, eb, 0);
8814 BUG_ON(ret); /* -ENOMEM */
8815 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
8817 btrfs_header_level(eb), 0);
8818 BUG_ON(ret); /* -ENOMEM */
8819 wc->flags[level] |= flag;
8823 * the block is shared by multiple trees, so it's not good to
8824 * keep the tree lock
8826 if (path->locks[level] && level > 0) {
8827 btrfs_tree_unlock_rw(eb, path->locks[level]);
8828 path->locks[level] = 0;
8834 * helper to process tree block pointer.
8836 * when wc->stage == DROP_REFERENCE, this function checks
8837 * reference count of the block pointed to. if the block
8838 * is shared and we need update back refs for the subtree
8839 * rooted at the block, this function changes wc->stage to
8840 * UPDATE_BACKREF. if the block is shared and there is no
8841 * need to update back, this function drops the reference
8844 * NOTE: return value 1 means we should stop walking down.
8846 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8847 struct btrfs_root *root,
8848 struct btrfs_path *path,
8849 struct walk_control *wc, int *lookup_info)
8855 struct btrfs_key key;
8856 struct extent_buffer *next;
8857 int level = wc->level;
8860 bool need_account = false;
8862 generation = btrfs_node_ptr_generation(path->nodes[level],
8863 path->slots[level]);
8865 * if the lower level block was created before the snapshot
8866 * was created, we know there is no need to update back refs
8869 if (wc->stage == UPDATE_BACKREF &&
8870 generation <= root->root_key.offset) {
8875 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8876 blocksize = root->nodesize;
8878 next = btrfs_find_tree_block(root->fs_info, bytenr);
8880 next = btrfs_find_create_tree_block(root, bytenr);
8882 return PTR_ERR(next);
8884 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8888 btrfs_tree_lock(next);
8889 btrfs_set_lock_blocking(next);
8891 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
8892 &wc->refs[level - 1],
8893 &wc->flags[level - 1]);
8897 if (unlikely(wc->refs[level - 1] == 0)) {
8898 btrfs_err(root->fs_info, "Missing references.");
8904 if (wc->stage == DROP_REFERENCE) {
8905 if (wc->refs[level - 1] > 1) {
8906 need_account = true;
8908 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8911 if (!wc->update_ref ||
8912 generation <= root->root_key.offset)
8915 btrfs_node_key_to_cpu(path->nodes[level], &key,
8916 path->slots[level]);
8917 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8921 wc->stage = UPDATE_BACKREF;
8922 wc->shared_level = level - 1;
8926 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8930 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8931 btrfs_tree_unlock(next);
8932 free_extent_buffer(next);
8938 if (reada && level == 1)
8939 reada_walk_down(trans, root, wc, path);
8940 next = read_tree_block(root, bytenr, generation);
8942 return PTR_ERR(next);
8943 } else if (!extent_buffer_uptodate(next)) {
8944 free_extent_buffer(next);
8947 btrfs_tree_lock(next);
8948 btrfs_set_lock_blocking(next);
8952 ASSERT(level == btrfs_header_level(next));
8953 if (level != btrfs_header_level(next)) {
8954 btrfs_err(root->fs_info, "mismatched level");
8958 path->nodes[level] = next;
8959 path->slots[level] = 0;
8960 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8966 wc->refs[level - 1] = 0;
8967 wc->flags[level - 1] = 0;
8968 if (wc->stage == DROP_REFERENCE) {
8969 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8970 parent = path->nodes[level]->start;
8972 ASSERT(root->root_key.objectid ==
8973 btrfs_header_owner(path->nodes[level]));
8974 if (root->root_key.objectid !=
8975 btrfs_header_owner(path->nodes[level])) {
8976 btrfs_err(root->fs_info,
8977 "mismatched block owner");
8985 ret = account_shared_subtree(trans, root, next,
8986 generation, level - 1);
8988 btrfs_err_rl(root->fs_info,
8989 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
8993 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
8994 root->root_key.objectid, level - 1, 0);
9003 btrfs_tree_unlock(next);
9004 free_extent_buffer(next);
9010 * helper to process tree block while walking up the tree.
9012 * when wc->stage == DROP_REFERENCE, this function drops
9013 * reference count on the block.
9015 * when wc->stage == UPDATE_BACKREF, this function changes
9016 * wc->stage back to DROP_REFERENCE if we changed wc->stage
9017 * to UPDATE_BACKREF previously while processing the block.
9019 * NOTE: return value 1 means we should stop walking up.
9021 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
9022 struct btrfs_root *root,
9023 struct btrfs_path *path,
9024 struct walk_control *wc)
9027 int level = wc->level;
9028 struct extent_buffer *eb = path->nodes[level];
9031 if (wc->stage == UPDATE_BACKREF) {
9032 BUG_ON(wc->shared_level < level);
9033 if (level < wc->shared_level)
9036 ret = find_next_key(path, level + 1, &wc->update_progress);
9040 wc->stage = DROP_REFERENCE;
9041 wc->shared_level = -1;
9042 path->slots[level] = 0;
9045 * check reference count again if the block isn't locked.
9046 * we should start walking down the tree again if reference
9049 if (!path->locks[level]) {
9051 btrfs_tree_lock(eb);
9052 btrfs_set_lock_blocking(eb);
9053 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9055 ret = btrfs_lookup_extent_info(trans, root,
9056 eb->start, level, 1,
9060 btrfs_tree_unlock_rw(eb, path->locks[level]);
9061 path->locks[level] = 0;
9064 BUG_ON(wc->refs[level] == 0);
9065 if (wc->refs[level] == 1) {
9066 btrfs_tree_unlock_rw(eb, path->locks[level]);
9067 path->locks[level] = 0;
9073 /* wc->stage == DROP_REFERENCE */
9074 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
9076 if (wc->refs[level] == 1) {
9078 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9079 ret = btrfs_dec_ref(trans, root, eb, 1);
9081 ret = btrfs_dec_ref(trans, root, eb, 0);
9082 BUG_ON(ret); /* -ENOMEM */
9083 ret = account_leaf_items(trans, root, eb);
9085 btrfs_err_rl(root->fs_info,
9086 "error %d accounting leaf items. Quota is out of sync, rescan required.",
9090 /* make block locked assertion in clean_tree_block happy */
9091 if (!path->locks[level] &&
9092 btrfs_header_generation(eb) == trans->transid) {
9093 btrfs_tree_lock(eb);
9094 btrfs_set_lock_blocking(eb);
9095 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9097 clean_tree_block(trans, root->fs_info, eb);
9100 if (eb == root->node) {
9101 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9104 BUG_ON(root->root_key.objectid !=
9105 btrfs_header_owner(eb));
9107 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9108 parent = path->nodes[level + 1]->start;
9110 BUG_ON(root->root_key.objectid !=
9111 btrfs_header_owner(path->nodes[level + 1]));
9114 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
9116 wc->refs[level] = 0;
9117 wc->flags[level] = 0;
9121 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
9122 struct btrfs_root *root,
9123 struct btrfs_path *path,
9124 struct walk_control *wc)
9126 int level = wc->level;
9127 int lookup_info = 1;
9130 while (level >= 0) {
9131 ret = walk_down_proc(trans, root, path, wc, lookup_info);
9138 if (path->slots[level] >=
9139 btrfs_header_nritems(path->nodes[level]))
9142 ret = do_walk_down(trans, root, path, wc, &lookup_info);
9144 path->slots[level]++;
9153 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
9154 struct btrfs_root *root,
9155 struct btrfs_path *path,
9156 struct walk_control *wc, int max_level)
9158 int level = wc->level;
9161 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
9162 while (level < max_level && path->nodes[level]) {
9164 if (path->slots[level] + 1 <
9165 btrfs_header_nritems(path->nodes[level])) {
9166 path->slots[level]++;
9169 ret = walk_up_proc(trans, root, path, wc);
9173 if (path->locks[level]) {
9174 btrfs_tree_unlock_rw(path->nodes[level],
9175 path->locks[level]);
9176 path->locks[level] = 0;
9178 free_extent_buffer(path->nodes[level]);
9179 path->nodes[level] = NULL;
9187 * drop a subvolume tree.
9189 * this function traverses the tree freeing any blocks that only
9190 * referenced by the tree.
9192 * when a shared tree block is found. this function decreases its
9193 * reference count by one. if update_ref is true, this function
9194 * also make sure backrefs for the shared block and all lower level
9195 * blocks are properly updated.
9197 * If called with for_reloc == 0, may exit early with -EAGAIN
9199 int btrfs_drop_snapshot(struct btrfs_root *root,
9200 struct btrfs_block_rsv *block_rsv, int update_ref,
9203 struct btrfs_fs_info *fs_info = root->fs_info;
9204 struct btrfs_path *path;
9205 struct btrfs_trans_handle *trans;
9206 struct btrfs_root *tree_root = fs_info->tree_root;
9207 struct btrfs_root_item *root_item = &root->root_item;
9208 struct walk_control *wc;
9209 struct btrfs_key key;
9213 bool root_dropped = false;
9215 btrfs_debug(fs_info, "Drop subvolume %llu", root->objectid);
9217 path = btrfs_alloc_path();
9223 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9225 btrfs_free_path(path);
9230 trans = btrfs_start_transaction(tree_root, 0);
9231 if (IS_ERR(trans)) {
9232 err = PTR_ERR(trans);
9237 trans->block_rsv = block_rsv;
9239 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
9240 level = btrfs_header_level(root->node);
9241 path->nodes[level] = btrfs_lock_root_node(root);
9242 btrfs_set_lock_blocking(path->nodes[level]);
9243 path->slots[level] = 0;
9244 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9245 memset(&wc->update_progress, 0,
9246 sizeof(wc->update_progress));
9248 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
9249 memcpy(&wc->update_progress, &key,
9250 sizeof(wc->update_progress));
9252 level = root_item->drop_level;
9254 path->lowest_level = level;
9255 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
9256 path->lowest_level = 0;
9264 * unlock our path, this is safe because only this
9265 * function is allowed to delete this snapshot
9267 btrfs_unlock_up_safe(path, 0);
9269 level = btrfs_header_level(root->node);
9271 btrfs_tree_lock(path->nodes[level]);
9272 btrfs_set_lock_blocking(path->nodes[level]);
9273 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9275 ret = btrfs_lookup_extent_info(trans, root,
9276 path->nodes[level]->start,
9277 level, 1, &wc->refs[level],
9283 BUG_ON(wc->refs[level] == 0);
9285 if (level == root_item->drop_level)
9288 btrfs_tree_unlock(path->nodes[level]);
9289 path->locks[level] = 0;
9290 WARN_ON(wc->refs[level] != 1);
9296 wc->shared_level = -1;
9297 wc->stage = DROP_REFERENCE;
9298 wc->update_ref = update_ref;
9300 wc->for_reloc = for_reloc;
9301 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
9305 ret = walk_down_tree(trans, root, path, wc);
9311 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
9318 BUG_ON(wc->stage != DROP_REFERENCE);
9322 if (wc->stage == DROP_REFERENCE) {
9324 btrfs_node_key(path->nodes[level],
9325 &root_item->drop_progress,
9326 path->slots[level]);
9327 root_item->drop_level = level;
9330 BUG_ON(wc->level == 0);
9331 if (btrfs_should_end_transaction(trans, tree_root) ||
9332 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
9333 ret = btrfs_update_root(trans, tree_root,
9337 btrfs_abort_transaction(trans, ret);
9342 btrfs_end_transaction_throttle(trans, tree_root);
9343 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
9344 btrfs_debug(fs_info,
9345 "drop snapshot early exit");
9350 trans = btrfs_start_transaction(tree_root, 0);
9351 if (IS_ERR(trans)) {
9352 err = PTR_ERR(trans);
9356 trans->block_rsv = block_rsv;
9359 btrfs_release_path(path);
9363 ret = btrfs_del_root(trans, tree_root, &root->root_key);
9365 btrfs_abort_transaction(trans, ret);
9370 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
9371 ret = btrfs_find_root(tree_root, &root->root_key, path,
9374 btrfs_abort_transaction(trans, ret);
9377 } else if (ret > 0) {
9378 /* if we fail to delete the orphan item this time
9379 * around, it'll get picked up the next time.
9381 * The most common failure here is just -ENOENT.
9383 btrfs_del_orphan_item(trans, tree_root,
9384 root->root_key.objectid);
9388 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
9389 btrfs_add_dropped_root(trans, root);
9391 free_extent_buffer(root->node);
9392 free_extent_buffer(root->commit_root);
9393 btrfs_put_fs_root(root);
9395 root_dropped = true;
9397 btrfs_end_transaction_throttle(trans, tree_root);
9400 btrfs_free_path(path);
9403 * So if we need to stop dropping the snapshot for whatever reason we
9404 * need to make sure to add it back to the dead root list so that we
9405 * keep trying to do the work later. This also cleans up roots if we
9406 * don't have it in the radix (like when we recover after a power fail
9407 * or unmount) so we don't leak memory.
9409 if (!for_reloc && root_dropped == false)
9410 btrfs_add_dead_root(root);
9411 if (err && err != -EAGAIN)
9412 btrfs_handle_fs_error(fs_info, err, NULL);
9417 * drop subtree rooted at tree block 'node'.
9419 * NOTE: this function will unlock and release tree block 'node'
9420 * only used by relocation code
9422 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
9423 struct btrfs_root *root,
9424 struct extent_buffer *node,
9425 struct extent_buffer *parent)
9427 struct btrfs_path *path;
9428 struct walk_control *wc;
9434 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
9436 path = btrfs_alloc_path();
9440 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9442 btrfs_free_path(path);
9446 btrfs_assert_tree_locked(parent);
9447 parent_level = btrfs_header_level(parent);
9448 extent_buffer_get(parent);
9449 path->nodes[parent_level] = parent;
9450 path->slots[parent_level] = btrfs_header_nritems(parent);
9452 btrfs_assert_tree_locked(node);
9453 level = btrfs_header_level(node);
9454 path->nodes[level] = node;
9455 path->slots[level] = 0;
9456 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9458 wc->refs[parent_level] = 1;
9459 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
9461 wc->shared_level = -1;
9462 wc->stage = DROP_REFERENCE;
9466 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
9469 wret = walk_down_tree(trans, root, path, wc);
9475 wret = walk_up_tree(trans, root, path, wc, parent_level);
9483 btrfs_free_path(path);
9487 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
9493 * if restripe for this chunk_type is on pick target profile and
9494 * return, otherwise do the usual balance
9496 stripped = get_restripe_target(root->fs_info, flags);
9498 return extended_to_chunk(stripped);
9500 num_devices = root->fs_info->fs_devices->rw_devices;
9502 stripped = BTRFS_BLOCK_GROUP_RAID0 |
9503 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
9504 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
9506 if (num_devices == 1) {
9507 stripped |= BTRFS_BLOCK_GROUP_DUP;
9508 stripped = flags & ~stripped;
9510 /* turn raid0 into single device chunks */
9511 if (flags & BTRFS_BLOCK_GROUP_RAID0)
9514 /* turn mirroring into duplication */
9515 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
9516 BTRFS_BLOCK_GROUP_RAID10))
9517 return stripped | BTRFS_BLOCK_GROUP_DUP;
9519 /* they already had raid on here, just return */
9520 if (flags & stripped)
9523 stripped |= BTRFS_BLOCK_GROUP_DUP;
9524 stripped = flags & ~stripped;
9526 /* switch duplicated blocks with raid1 */
9527 if (flags & BTRFS_BLOCK_GROUP_DUP)
9528 return stripped | BTRFS_BLOCK_GROUP_RAID1;
9530 /* this is drive concat, leave it alone */
9536 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
9538 struct btrfs_space_info *sinfo = cache->space_info;
9540 u64 min_allocable_bytes;
9544 * We need some metadata space and system metadata space for
9545 * allocating chunks in some corner cases until we force to set
9546 * it to be readonly.
9549 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
9551 min_allocable_bytes = SZ_1M;
9553 min_allocable_bytes = 0;
9555 spin_lock(&sinfo->lock);
9556 spin_lock(&cache->lock);
9564 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
9565 cache->bytes_super - btrfs_block_group_used(&cache->item);
9567 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
9568 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
9569 min_allocable_bytes <= sinfo->total_bytes) {
9570 sinfo->bytes_readonly += num_bytes;
9572 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
9576 spin_unlock(&cache->lock);
9577 spin_unlock(&sinfo->lock);
9581 int btrfs_inc_block_group_ro(struct btrfs_root *root,
9582 struct btrfs_block_group_cache *cache)
9585 struct btrfs_trans_handle *trans;
9590 trans = btrfs_join_transaction(root);
9592 return PTR_ERR(trans);
9595 * we're not allowed to set block groups readonly after the dirty
9596 * block groups cache has started writing. If it already started,
9597 * back off and let this transaction commit
9599 mutex_lock(&root->fs_info->ro_block_group_mutex);
9600 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9601 u64 transid = trans->transid;
9603 mutex_unlock(&root->fs_info->ro_block_group_mutex);
9604 btrfs_end_transaction(trans, root);
9606 ret = btrfs_wait_for_commit(root, transid);
9613 * if we are changing raid levels, try to allocate a corresponding
9614 * block group with the new raid level.
9616 alloc_flags = update_block_group_flags(root, cache->flags);
9617 if (alloc_flags != cache->flags) {
9618 ret = do_chunk_alloc(trans, root, alloc_flags,
9621 * ENOSPC is allowed here, we may have enough space
9622 * already allocated at the new raid level to
9631 ret = inc_block_group_ro(cache, 0);
9634 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
9635 ret = do_chunk_alloc(trans, root, alloc_flags,
9639 ret = inc_block_group_ro(cache, 0);
9641 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9642 alloc_flags = update_block_group_flags(root, cache->flags);
9643 lock_chunks(root->fs_info->chunk_root);
9644 check_system_chunk(trans, root, alloc_flags);
9645 unlock_chunks(root->fs_info->chunk_root);
9647 mutex_unlock(&root->fs_info->ro_block_group_mutex);
9649 btrfs_end_transaction(trans, root);
9653 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
9654 struct btrfs_root *root, u64 type)
9656 u64 alloc_flags = get_alloc_profile(root, type);
9657 return do_chunk_alloc(trans, root, alloc_flags,
9662 * helper to account the unused space of all the readonly block group in the
9663 * space_info. takes mirrors into account.
9665 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9667 struct btrfs_block_group_cache *block_group;
9671 /* It's df, we don't care if it's racy */
9672 if (list_empty(&sinfo->ro_bgs))
9675 spin_lock(&sinfo->lock);
9676 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9677 spin_lock(&block_group->lock);
9679 if (!block_group->ro) {
9680 spin_unlock(&block_group->lock);
9684 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
9685 BTRFS_BLOCK_GROUP_RAID10 |
9686 BTRFS_BLOCK_GROUP_DUP))
9691 free_bytes += (block_group->key.offset -
9692 btrfs_block_group_used(&block_group->item)) *
9695 spin_unlock(&block_group->lock);
9697 spin_unlock(&sinfo->lock);
9702 void btrfs_dec_block_group_ro(struct btrfs_root *root,
9703 struct btrfs_block_group_cache *cache)
9705 struct btrfs_space_info *sinfo = cache->space_info;
9710 spin_lock(&sinfo->lock);
9711 spin_lock(&cache->lock);
9713 num_bytes = cache->key.offset - cache->reserved -
9714 cache->pinned - cache->bytes_super -
9715 btrfs_block_group_used(&cache->item);
9716 sinfo->bytes_readonly -= num_bytes;
9717 list_del_init(&cache->ro_list);
9719 spin_unlock(&cache->lock);
9720 spin_unlock(&sinfo->lock);
9724 * checks to see if its even possible to relocate this block group.
9726 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9727 * ok to go ahead and try.
9729 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
9731 struct btrfs_block_group_cache *block_group;
9732 struct btrfs_space_info *space_info;
9733 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
9734 struct btrfs_device *device;
9735 struct btrfs_trans_handle *trans;
9745 debug = btrfs_test_opt(root->fs_info, ENOSPC_DEBUG);
9747 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
9749 /* odd, couldn't find the block group, leave it alone */
9752 btrfs_warn(root->fs_info,
9753 "can't find block group for bytenr %llu",
9758 min_free = btrfs_block_group_used(&block_group->item);
9760 /* no bytes used, we're good */
9764 space_info = block_group->space_info;
9765 spin_lock(&space_info->lock);
9767 full = space_info->full;
9770 * if this is the last block group we have in this space, we can't
9771 * relocate it unless we're able to allocate a new chunk below.
9773 * Otherwise, we need to make sure we have room in the space to handle
9774 * all of the extents from this block group. If we can, we're good
9776 if ((space_info->total_bytes != block_group->key.offset) &&
9777 (space_info->bytes_used + space_info->bytes_reserved +
9778 space_info->bytes_pinned + space_info->bytes_readonly +
9779 min_free < space_info->total_bytes)) {
9780 spin_unlock(&space_info->lock);
9783 spin_unlock(&space_info->lock);
9786 * ok we don't have enough space, but maybe we have free space on our
9787 * devices to allocate new chunks for relocation, so loop through our
9788 * alloc devices and guess if we have enough space. if this block
9789 * group is going to be restriped, run checks against the target
9790 * profile instead of the current one.
9802 target = get_restripe_target(root->fs_info, block_group->flags);
9804 index = __get_raid_index(extended_to_chunk(target));
9807 * this is just a balance, so if we were marked as full
9808 * we know there is no space for a new chunk
9812 btrfs_warn(root->fs_info,
9813 "no space to alloc new chunk for block group %llu",
9814 block_group->key.objectid);
9818 index = get_block_group_index(block_group);
9821 if (index == BTRFS_RAID_RAID10) {
9825 } else if (index == BTRFS_RAID_RAID1) {
9827 } else if (index == BTRFS_RAID_DUP) {
9830 } else if (index == BTRFS_RAID_RAID0) {
9831 dev_min = fs_devices->rw_devices;
9832 min_free = div64_u64(min_free, dev_min);
9835 /* We need to do this so that we can look at pending chunks */
9836 trans = btrfs_join_transaction(root);
9837 if (IS_ERR(trans)) {
9838 ret = PTR_ERR(trans);
9842 mutex_lock(&root->fs_info->chunk_mutex);
9843 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9847 * check to make sure we can actually find a chunk with enough
9848 * space to fit our block group in.
9850 if (device->total_bytes > device->bytes_used + min_free &&
9851 !device->is_tgtdev_for_dev_replace) {
9852 ret = find_free_dev_extent(trans, device, min_free,
9857 if (dev_nr >= dev_min)
9863 if (debug && ret == -1)
9864 btrfs_warn(root->fs_info,
9865 "no space to allocate a new chunk for block group %llu",
9866 block_group->key.objectid);
9867 mutex_unlock(&root->fs_info->chunk_mutex);
9868 btrfs_end_transaction(trans, root);
9870 btrfs_put_block_group(block_group);
9874 static int find_first_block_group(struct btrfs_root *root,
9875 struct btrfs_path *path, struct btrfs_key *key)
9878 struct btrfs_key found_key;
9879 struct extent_buffer *leaf;
9882 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9887 slot = path->slots[0];
9888 leaf = path->nodes[0];
9889 if (slot >= btrfs_header_nritems(leaf)) {
9890 ret = btrfs_next_leaf(root, path);
9897 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9899 if (found_key.objectid >= key->objectid &&
9900 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9901 struct extent_map_tree *em_tree;
9902 struct extent_map *em;
9904 em_tree = &root->fs_info->mapping_tree.map_tree;
9905 read_lock(&em_tree->lock);
9906 em = lookup_extent_mapping(em_tree, found_key.objectid,
9908 read_unlock(&em_tree->lock);
9910 btrfs_err(root->fs_info,
9911 "logical %llu len %llu found bg but no related chunk",
9912 found_key.objectid, found_key.offset);
9917 free_extent_map(em);
9926 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9928 struct btrfs_block_group_cache *block_group;
9932 struct inode *inode;
9934 block_group = btrfs_lookup_first_block_group(info, last);
9935 while (block_group) {
9936 spin_lock(&block_group->lock);
9937 if (block_group->iref)
9939 spin_unlock(&block_group->lock);
9940 block_group = next_block_group(info->tree_root,
9950 inode = block_group->inode;
9951 block_group->iref = 0;
9952 block_group->inode = NULL;
9953 spin_unlock(&block_group->lock);
9954 ASSERT(block_group->io_ctl.inode == NULL);
9956 last = block_group->key.objectid + block_group->key.offset;
9957 btrfs_put_block_group(block_group);
9961 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9963 struct btrfs_block_group_cache *block_group;
9964 struct btrfs_space_info *space_info;
9965 struct btrfs_caching_control *caching_ctl;
9968 down_write(&info->commit_root_sem);
9969 while (!list_empty(&info->caching_block_groups)) {
9970 caching_ctl = list_entry(info->caching_block_groups.next,
9971 struct btrfs_caching_control, list);
9972 list_del(&caching_ctl->list);
9973 put_caching_control(caching_ctl);
9975 up_write(&info->commit_root_sem);
9977 spin_lock(&info->unused_bgs_lock);
9978 while (!list_empty(&info->unused_bgs)) {
9979 block_group = list_first_entry(&info->unused_bgs,
9980 struct btrfs_block_group_cache,
9982 list_del_init(&block_group->bg_list);
9983 btrfs_put_block_group(block_group);
9985 spin_unlock(&info->unused_bgs_lock);
9987 spin_lock(&info->block_group_cache_lock);
9988 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9989 block_group = rb_entry(n, struct btrfs_block_group_cache,
9991 rb_erase(&block_group->cache_node,
9992 &info->block_group_cache_tree);
9993 RB_CLEAR_NODE(&block_group->cache_node);
9994 spin_unlock(&info->block_group_cache_lock);
9996 down_write(&block_group->space_info->groups_sem);
9997 list_del(&block_group->list);
9998 up_write(&block_group->space_info->groups_sem);
10000 if (block_group->cached == BTRFS_CACHE_STARTED)
10001 wait_block_group_cache_done(block_group);
10004 * We haven't cached this block group, which means we could
10005 * possibly have excluded extents on this block group.
10007 if (block_group->cached == BTRFS_CACHE_NO ||
10008 block_group->cached == BTRFS_CACHE_ERROR)
10009 free_excluded_extents(info->extent_root, block_group);
10011 btrfs_remove_free_space_cache(block_group);
10012 ASSERT(list_empty(&block_group->dirty_list));
10013 ASSERT(list_empty(&block_group->io_list));
10014 ASSERT(list_empty(&block_group->bg_list));
10015 ASSERT(atomic_read(&block_group->count) == 1);
10016 btrfs_put_block_group(block_group);
10018 spin_lock(&info->block_group_cache_lock);
10020 spin_unlock(&info->block_group_cache_lock);
10022 /* now that all the block groups are freed, go through and
10023 * free all the space_info structs. This is only called during
10024 * the final stages of unmount, and so we know nobody is
10025 * using them. We call synchronize_rcu() once before we start,
10026 * just to be on the safe side.
10030 release_global_block_rsv(info);
10032 while (!list_empty(&info->space_info)) {
10035 space_info = list_entry(info->space_info.next,
10036 struct btrfs_space_info,
10040 * Do not hide this behind enospc_debug, this is actually
10041 * important and indicates a real bug if this happens.
10043 if (WARN_ON(space_info->bytes_pinned > 0 ||
10044 space_info->bytes_reserved > 0 ||
10045 space_info->bytes_may_use > 0))
10046 dump_space_info(info, space_info, 0, 0);
10047 list_del(&space_info->list);
10048 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
10049 struct kobject *kobj;
10050 kobj = space_info->block_group_kobjs[i];
10051 space_info->block_group_kobjs[i] = NULL;
10057 kobject_del(&space_info->kobj);
10058 kobject_put(&space_info->kobj);
10063 static void __link_block_group(struct btrfs_space_info *space_info,
10064 struct btrfs_block_group_cache *cache)
10066 int index = get_block_group_index(cache);
10067 bool first = false;
10069 down_write(&space_info->groups_sem);
10070 if (list_empty(&space_info->block_groups[index]))
10072 list_add_tail(&cache->list, &space_info->block_groups[index]);
10073 up_write(&space_info->groups_sem);
10076 struct raid_kobject *rkobj;
10079 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
10082 rkobj->raid_type = index;
10083 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
10084 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
10085 "%s", get_raid_name(index));
10087 kobject_put(&rkobj->kobj);
10090 space_info->block_group_kobjs[index] = &rkobj->kobj;
10095 btrfs_warn(cache->fs_info,
10096 "failed to add kobject for block cache, ignoring");
10099 static struct btrfs_block_group_cache *
10100 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
10102 struct btrfs_block_group_cache *cache;
10104 cache = kzalloc(sizeof(*cache), GFP_NOFS);
10108 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
10110 if (!cache->free_space_ctl) {
10115 cache->key.objectid = start;
10116 cache->key.offset = size;
10117 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10119 cache->sectorsize = root->sectorsize;
10120 cache->fs_info = root->fs_info;
10121 cache->full_stripe_len = btrfs_full_stripe_len(root,
10122 &root->fs_info->mapping_tree,
10124 set_free_space_tree_thresholds(cache);
10126 atomic_set(&cache->count, 1);
10127 spin_lock_init(&cache->lock);
10128 init_rwsem(&cache->data_rwsem);
10129 INIT_LIST_HEAD(&cache->list);
10130 INIT_LIST_HEAD(&cache->cluster_list);
10131 INIT_LIST_HEAD(&cache->bg_list);
10132 INIT_LIST_HEAD(&cache->ro_list);
10133 INIT_LIST_HEAD(&cache->dirty_list);
10134 INIT_LIST_HEAD(&cache->io_list);
10135 btrfs_init_free_space_ctl(cache);
10136 atomic_set(&cache->trimming, 0);
10137 mutex_init(&cache->free_space_lock);
10142 int btrfs_read_block_groups(struct btrfs_root *root)
10144 struct btrfs_path *path;
10146 struct btrfs_block_group_cache *cache;
10147 struct btrfs_fs_info *info = root->fs_info;
10148 struct btrfs_space_info *space_info;
10149 struct btrfs_key key;
10150 struct btrfs_key found_key;
10151 struct extent_buffer *leaf;
10152 int need_clear = 0;
10157 feature = btrfs_super_incompat_flags(info->super_copy);
10158 mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS);
10160 root = info->extent_root;
10163 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10164 path = btrfs_alloc_path();
10167 path->reada = READA_FORWARD;
10169 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
10170 if (btrfs_test_opt(root->fs_info, SPACE_CACHE) &&
10171 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
10173 if (btrfs_test_opt(root->fs_info, CLEAR_CACHE))
10177 ret = find_first_block_group(root, path, &key);
10183 leaf = path->nodes[0];
10184 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
10186 cache = btrfs_create_block_group_cache(root, found_key.objectid,
10195 * When we mount with old space cache, we need to
10196 * set BTRFS_DC_CLEAR and set dirty flag.
10198 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
10199 * truncate the old free space cache inode and
10201 * b) Setting 'dirty flag' makes sure that we flush
10202 * the new space cache info onto disk.
10204 if (btrfs_test_opt(root->fs_info, SPACE_CACHE))
10205 cache->disk_cache_state = BTRFS_DC_CLEAR;
10208 read_extent_buffer(leaf, &cache->item,
10209 btrfs_item_ptr_offset(leaf, path->slots[0]),
10210 sizeof(cache->item));
10211 cache->flags = btrfs_block_group_flags(&cache->item);
10213 ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) &&
10214 (cache->flags & BTRFS_BLOCK_GROUP_DATA))) {
10216 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
10217 cache->key.objectid);
10222 key.objectid = found_key.objectid + found_key.offset;
10223 btrfs_release_path(path);
10226 * We need to exclude the super stripes now so that the space
10227 * info has super bytes accounted for, otherwise we'll think
10228 * we have more space than we actually do.
10230 ret = exclude_super_stripes(root, cache);
10233 * We may have excluded something, so call this just in
10236 free_excluded_extents(root, cache);
10237 btrfs_put_block_group(cache);
10242 * check for two cases, either we are full, and therefore
10243 * don't need to bother with the caching work since we won't
10244 * find any space, or we are empty, and we can just add all
10245 * the space in and be done with it. This saves us _alot_ of
10246 * time, particularly in the full case.
10248 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
10249 cache->last_byte_to_unpin = (u64)-1;
10250 cache->cached = BTRFS_CACHE_FINISHED;
10251 free_excluded_extents(root, cache);
10252 } else if (btrfs_block_group_used(&cache->item) == 0) {
10253 cache->last_byte_to_unpin = (u64)-1;
10254 cache->cached = BTRFS_CACHE_FINISHED;
10255 add_new_free_space(cache, root->fs_info,
10256 found_key.objectid,
10257 found_key.objectid +
10259 free_excluded_extents(root, cache);
10262 ret = btrfs_add_block_group_cache(root->fs_info, cache);
10264 btrfs_remove_free_space_cache(cache);
10265 btrfs_put_block_group(cache);
10269 trace_btrfs_add_block_group(root->fs_info, cache, 0);
10270 ret = update_space_info(info, cache->flags, found_key.offset,
10271 btrfs_block_group_used(&cache->item),
10272 cache->bytes_super, &space_info);
10274 btrfs_remove_free_space_cache(cache);
10275 spin_lock(&info->block_group_cache_lock);
10276 rb_erase(&cache->cache_node,
10277 &info->block_group_cache_tree);
10278 RB_CLEAR_NODE(&cache->cache_node);
10279 spin_unlock(&info->block_group_cache_lock);
10280 btrfs_put_block_group(cache);
10284 cache->space_info = space_info;
10286 __link_block_group(space_info, cache);
10288 set_avail_alloc_bits(root->fs_info, cache->flags);
10289 if (btrfs_chunk_readonly(root, cache->key.objectid)) {
10290 inc_block_group_ro(cache, 1);
10291 } else if (btrfs_block_group_used(&cache->item) == 0) {
10292 spin_lock(&info->unused_bgs_lock);
10293 /* Should always be true but just in case. */
10294 if (list_empty(&cache->bg_list)) {
10295 btrfs_get_block_group(cache);
10296 list_add_tail(&cache->bg_list,
10297 &info->unused_bgs);
10299 spin_unlock(&info->unused_bgs_lock);
10303 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
10304 if (!(get_alloc_profile(root, space_info->flags) &
10305 (BTRFS_BLOCK_GROUP_RAID10 |
10306 BTRFS_BLOCK_GROUP_RAID1 |
10307 BTRFS_BLOCK_GROUP_RAID5 |
10308 BTRFS_BLOCK_GROUP_RAID6 |
10309 BTRFS_BLOCK_GROUP_DUP)))
10312 * avoid allocating from un-mirrored block group if there are
10313 * mirrored block groups.
10315 list_for_each_entry(cache,
10316 &space_info->block_groups[BTRFS_RAID_RAID0],
10318 inc_block_group_ro(cache, 1);
10319 list_for_each_entry(cache,
10320 &space_info->block_groups[BTRFS_RAID_SINGLE],
10322 inc_block_group_ro(cache, 1);
10325 init_global_block_rsv(info);
10328 btrfs_free_path(path);
10332 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
10333 struct btrfs_root *root)
10335 struct btrfs_block_group_cache *block_group, *tmp;
10336 struct btrfs_root *extent_root = root->fs_info->extent_root;
10337 struct btrfs_block_group_item item;
10338 struct btrfs_key key;
10340 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
10342 trans->can_flush_pending_bgs = false;
10343 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
10347 spin_lock(&block_group->lock);
10348 memcpy(&item, &block_group->item, sizeof(item));
10349 memcpy(&key, &block_group->key, sizeof(key));
10350 spin_unlock(&block_group->lock);
10352 ret = btrfs_insert_item(trans, extent_root, &key, &item,
10355 btrfs_abort_transaction(trans, ret);
10356 ret = btrfs_finish_chunk_alloc(trans, extent_root,
10357 key.objectid, key.offset);
10359 btrfs_abort_transaction(trans, ret);
10360 add_block_group_free_space(trans, root->fs_info, block_group);
10361 /* already aborted the transaction if it failed. */
10363 list_del_init(&block_group->bg_list);
10365 trans->can_flush_pending_bgs = can_flush_pending_bgs;
10368 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
10369 struct btrfs_root *root, u64 bytes_used,
10370 u64 type, u64 chunk_objectid, u64 chunk_offset,
10374 struct btrfs_root *extent_root;
10375 struct btrfs_block_group_cache *cache;
10376 extent_root = root->fs_info->extent_root;
10378 btrfs_set_log_full_commit(root->fs_info, trans);
10380 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
10384 btrfs_set_block_group_used(&cache->item, bytes_used);
10385 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
10386 btrfs_set_block_group_flags(&cache->item, type);
10388 cache->flags = type;
10389 cache->last_byte_to_unpin = (u64)-1;
10390 cache->cached = BTRFS_CACHE_FINISHED;
10391 cache->needs_free_space = 1;
10392 ret = exclude_super_stripes(root, cache);
10395 * We may have excluded something, so call this just in
10398 free_excluded_extents(root, cache);
10399 btrfs_put_block_group(cache);
10403 add_new_free_space(cache, root->fs_info, chunk_offset,
10404 chunk_offset + size);
10406 free_excluded_extents(root, cache);
10408 #ifdef CONFIG_BTRFS_DEBUG
10409 if (btrfs_should_fragment_free_space(root, cache)) {
10410 u64 new_bytes_used = size - bytes_used;
10412 bytes_used += new_bytes_used >> 1;
10413 fragment_free_space(root, cache);
10417 * Call to ensure the corresponding space_info object is created and
10418 * assigned to our block group, but don't update its counters just yet.
10419 * We want our bg to be added to the rbtree with its ->space_info set.
10421 ret = update_space_info(root->fs_info, cache->flags, 0, 0, 0,
10422 &cache->space_info);
10424 btrfs_remove_free_space_cache(cache);
10425 btrfs_put_block_group(cache);
10429 ret = btrfs_add_block_group_cache(root->fs_info, cache);
10431 btrfs_remove_free_space_cache(cache);
10432 btrfs_put_block_group(cache);
10437 * Now that our block group has its ->space_info set and is inserted in
10438 * the rbtree, update the space info's counters.
10440 trace_btrfs_add_block_group(root->fs_info, cache, 1);
10441 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
10442 cache->bytes_super, &cache->space_info);
10444 btrfs_remove_free_space_cache(cache);
10445 spin_lock(&root->fs_info->block_group_cache_lock);
10446 rb_erase(&cache->cache_node,
10447 &root->fs_info->block_group_cache_tree);
10448 RB_CLEAR_NODE(&cache->cache_node);
10449 spin_unlock(&root->fs_info->block_group_cache_lock);
10450 btrfs_put_block_group(cache);
10453 update_global_block_rsv(root->fs_info);
10455 __link_block_group(cache->space_info, cache);
10457 list_add_tail(&cache->bg_list, &trans->new_bgs);
10459 set_avail_alloc_bits(extent_root->fs_info, type);
10463 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
10465 u64 extra_flags = chunk_to_extended(flags) &
10466 BTRFS_EXTENDED_PROFILE_MASK;
10468 write_seqlock(&fs_info->profiles_lock);
10469 if (flags & BTRFS_BLOCK_GROUP_DATA)
10470 fs_info->avail_data_alloc_bits &= ~extra_flags;
10471 if (flags & BTRFS_BLOCK_GROUP_METADATA)
10472 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
10473 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
10474 fs_info->avail_system_alloc_bits &= ~extra_flags;
10475 write_sequnlock(&fs_info->profiles_lock);
10478 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
10479 struct btrfs_root *root, u64 group_start,
10480 struct extent_map *em)
10482 struct btrfs_path *path;
10483 struct btrfs_block_group_cache *block_group;
10484 struct btrfs_free_cluster *cluster;
10485 struct btrfs_root *tree_root = root->fs_info->tree_root;
10486 struct btrfs_key key;
10487 struct inode *inode;
10488 struct kobject *kobj = NULL;
10492 struct btrfs_caching_control *caching_ctl = NULL;
10495 root = root->fs_info->extent_root;
10497 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
10498 BUG_ON(!block_group);
10499 BUG_ON(!block_group->ro);
10502 * Free the reserved super bytes from this block group before
10505 free_excluded_extents(root, block_group);
10507 memcpy(&key, &block_group->key, sizeof(key));
10508 index = get_block_group_index(block_group);
10509 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
10510 BTRFS_BLOCK_GROUP_RAID1 |
10511 BTRFS_BLOCK_GROUP_RAID10))
10516 /* make sure this block group isn't part of an allocation cluster */
10517 cluster = &root->fs_info->data_alloc_cluster;
10518 spin_lock(&cluster->refill_lock);
10519 btrfs_return_cluster_to_free_space(block_group, cluster);
10520 spin_unlock(&cluster->refill_lock);
10523 * make sure this block group isn't part of a metadata
10524 * allocation cluster
10526 cluster = &root->fs_info->meta_alloc_cluster;
10527 spin_lock(&cluster->refill_lock);
10528 btrfs_return_cluster_to_free_space(block_group, cluster);
10529 spin_unlock(&cluster->refill_lock);
10531 path = btrfs_alloc_path();
10538 * get the inode first so any iput calls done for the io_list
10539 * aren't the final iput (no unlinks allowed now)
10541 inode = lookup_free_space_inode(tree_root, block_group, path);
10543 mutex_lock(&trans->transaction->cache_write_mutex);
10545 * make sure our free spache cache IO is done before remove the
10548 spin_lock(&trans->transaction->dirty_bgs_lock);
10549 if (!list_empty(&block_group->io_list)) {
10550 list_del_init(&block_group->io_list);
10552 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
10554 spin_unlock(&trans->transaction->dirty_bgs_lock);
10555 btrfs_wait_cache_io(root, trans, block_group,
10556 &block_group->io_ctl, path,
10557 block_group->key.objectid);
10558 btrfs_put_block_group(block_group);
10559 spin_lock(&trans->transaction->dirty_bgs_lock);
10562 if (!list_empty(&block_group->dirty_list)) {
10563 list_del_init(&block_group->dirty_list);
10564 btrfs_put_block_group(block_group);
10566 spin_unlock(&trans->transaction->dirty_bgs_lock);
10567 mutex_unlock(&trans->transaction->cache_write_mutex);
10569 if (!IS_ERR(inode)) {
10570 ret = btrfs_orphan_add(trans, inode);
10572 btrfs_add_delayed_iput(inode);
10575 clear_nlink(inode);
10576 /* One for the block groups ref */
10577 spin_lock(&block_group->lock);
10578 if (block_group->iref) {
10579 block_group->iref = 0;
10580 block_group->inode = NULL;
10581 spin_unlock(&block_group->lock);
10584 spin_unlock(&block_group->lock);
10586 /* One for our lookup ref */
10587 btrfs_add_delayed_iput(inode);
10590 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
10591 key.offset = block_group->key.objectid;
10594 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
10598 btrfs_release_path(path);
10600 ret = btrfs_del_item(trans, tree_root, path);
10603 btrfs_release_path(path);
10606 spin_lock(&root->fs_info->block_group_cache_lock);
10607 rb_erase(&block_group->cache_node,
10608 &root->fs_info->block_group_cache_tree);
10609 RB_CLEAR_NODE(&block_group->cache_node);
10611 if (root->fs_info->first_logical_byte == block_group->key.objectid)
10612 root->fs_info->first_logical_byte = (u64)-1;
10613 spin_unlock(&root->fs_info->block_group_cache_lock);
10615 down_write(&block_group->space_info->groups_sem);
10617 * we must use list_del_init so people can check to see if they
10618 * are still on the list after taking the semaphore
10620 list_del_init(&block_group->list);
10621 if (list_empty(&block_group->space_info->block_groups[index])) {
10622 kobj = block_group->space_info->block_group_kobjs[index];
10623 block_group->space_info->block_group_kobjs[index] = NULL;
10624 clear_avail_alloc_bits(root->fs_info, block_group->flags);
10626 up_write(&block_group->space_info->groups_sem);
10632 if (block_group->has_caching_ctl)
10633 caching_ctl = get_caching_control(block_group);
10634 if (block_group->cached == BTRFS_CACHE_STARTED)
10635 wait_block_group_cache_done(block_group);
10636 if (block_group->has_caching_ctl) {
10637 down_write(&root->fs_info->commit_root_sem);
10638 if (!caching_ctl) {
10639 struct btrfs_caching_control *ctl;
10641 list_for_each_entry(ctl,
10642 &root->fs_info->caching_block_groups, list)
10643 if (ctl->block_group == block_group) {
10645 atomic_inc(&caching_ctl->count);
10650 list_del_init(&caching_ctl->list);
10651 up_write(&root->fs_info->commit_root_sem);
10653 /* Once for the caching bgs list and once for us. */
10654 put_caching_control(caching_ctl);
10655 put_caching_control(caching_ctl);
10659 spin_lock(&trans->transaction->dirty_bgs_lock);
10660 if (!list_empty(&block_group->dirty_list)) {
10663 if (!list_empty(&block_group->io_list)) {
10666 spin_unlock(&trans->transaction->dirty_bgs_lock);
10667 btrfs_remove_free_space_cache(block_group);
10669 spin_lock(&block_group->space_info->lock);
10670 list_del_init(&block_group->ro_list);
10672 if (btrfs_test_opt(root->fs_info, ENOSPC_DEBUG)) {
10673 WARN_ON(block_group->space_info->total_bytes
10674 < block_group->key.offset);
10675 WARN_ON(block_group->space_info->bytes_readonly
10676 < block_group->key.offset);
10677 WARN_ON(block_group->space_info->disk_total
10678 < block_group->key.offset * factor);
10680 block_group->space_info->total_bytes -= block_group->key.offset;
10681 block_group->space_info->bytes_readonly -= block_group->key.offset;
10682 block_group->space_info->disk_total -= block_group->key.offset * factor;
10684 spin_unlock(&block_group->space_info->lock);
10686 memcpy(&key, &block_group->key, sizeof(key));
10689 if (!list_empty(&em->list)) {
10690 /* We're in the transaction->pending_chunks list. */
10691 free_extent_map(em);
10693 spin_lock(&block_group->lock);
10694 block_group->removed = 1;
10696 * At this point trimming can't start on this block group, because we
10697 * removed the block group from the tree fs_info->block_group_cache_tree
10698 * so no one can't find it anymore and even if someone already got this
10699 * block group before we removed it from the rbtree, they have already
10700 * incremented block_group->trimming - if they didn't, they won't find
10701 * any free space entries because we already removed them all when we
10702 * called btrfs_remove_free_space_cache().
10704 * And we must not remove the extent map from the fs_info->mapping_tree
10705 * to prevent the same logical address range and physical device space
10706 * ranges from being reused for a new block group. This is because our
10707 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10708 * completely transactionless, so while it is trimming a range the
10709 * currently running transaction might finish and a new one start,
10710 * allowing for new block groups to be created that can reuse the same
10711 * physical device locations unless we take this special care.
10713 * There may also be an implicit trim operation if the file system
10714 * is mounted with -odiscard. The same protections must remain
10715 * in place until the extents have been discarded completely when
10716 * the transaction commit has completed.
10718 remove_em = (atomic_read(&block_group->trimming) == 0);
10720 * Make sure a trimmer task always sees the em in the pinned_chunks list
10721 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10722 * before checking block_group->removed).
10726 * Our em might be in trans->transaction->pending_chunks which
10727 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10728 * and so is the fs_info->pinned_chunks list.
10730 * So at this point we must be holding the chunk_mutex to avoid
10731 * any races with chunk allocation (more specifically at
10732 * volumes.c:contains_pending_extent()), to ensure it always
10733 * sees the em, either in the pending_chunks list or in the
10734 * pinned_chunks list.
10736 list_move_tail(&em->list, &root->fs_info->pinned_chunks);
10738 spin_unlock(&block_group->lock);
10741 struct extent_map_tree *em_tree;
10743 em_tree = &root->fs_info->mapping_tree.map_tree;
10744 write_lock(&em_tree->lock);
10746 * The em might be in the pending_chunks list, so make sure the
10747 * chunk mutex is locked, since remove_extent_mapping() will
10748 * delete us from that list.
10750 remove_extent_mapping(em_tree, em);
10751 write_unlock(&em_tree->lock);
10752 /* once for the tree */
10753 free_extent_map(em);
10756 unlock_chunks(root);
10758 ret = remove_block_group_free_space(trans, root->fs_info, block_group);
10762 btrfs_put_block_group(block_group);
10763 btrfs_put_block_group(block_group);
10765 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10771 ret = btrfs_del_item(trans, root, path);
10773 btrfs_free_path(path);
10777 struct btrfs_trans_handle *
10778 btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info,
10779 const u64 chunk_offset)
10781 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
10782 struct extent_map *em;
10783 struct map_lookup *map;
10784 unsigned int num_items;
10786 read_lock(&em_tree->lock);
10787 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
10788 read_unlock(&em_tree->lock);
10789 ASSERT(em && em->start == chunk_offset);
10792 * We need to reserve 3 + N units from the metadata space info in order
10793 * to remove a block group (done at btrfs_remove_chunk() and at
10794 * btrfs_remove_block_group()), which are used for:
10796 * 1 unit for adding the free space inode's orphan (located in the tree
10798 * 1 unit for deleting the block group item (located in the extent
10800 * 1 unit for deleting the free space item (located in tree of tree
10802 * N units for deleting N device extent items corresponding to each
10803 * stripe (located in the device tree).
10805 * In order to remove a block group we also need to reserve units in the
10806 * system space info in order to update the chunk tree (update one or
10807 * more device items and remove one chunk item), but this is done at
10808 * btrfs_remove_chunk() through a call to check_system_chunk().
10810 map = em->map_lookup;
10811 num_items = 3 + map->num_stripes;
10812 free_extent_map(em);
10814 return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root,
10819 * Process the unused_bgs list and remove any that don't have any allocated
10820 * space inside of them.
10822 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10824 struct btrfs_block_group_cache *block_group;
10825 struct btrfs_space_info *space_info;
10826 struct btrfs_root *root = fs_info->extent_root;
10827 struct btrfs_trans_handle *trans;
10830 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
10833 spin_lock(&fs_info->unused_bgs_lock);
10834 while (!list_empty(&fs_info->unused_bgs)) {
10838 block_group = list_first_entry(&fs_info->unused_bgs,
10839 struct btrfs_block_group_cache,
10841 list_del_init(&block_group->bg_list);
10843 space_info = block_group->space_info;
10845 if (ret || btrfs_mixed_space_info(space_info)) {
10846 btrfs_put_block_group(block_group);
10849 spin_unlock(&fs_info->unused_bgs_lock);
10851 mutex_lock(&fs_info->delete_unused_bgs_mutex);
10853 /* Don't want to race with allocators so take the groups_sem */
10854 down_write(&space_info->groups_sem);
10855 spin_lock(&block_group->lock);
10856 if (block_group->reserved || block_group->pinned ||
10857 btrfs_block_group_used(&block_group->item) ||
10859 list_is_singular(&block_group->list)) {
10861 * We want to bail if we made new allocations or have
10862 * outstanding allocations in this block group. We do
10863 * the ro check in case balance is currently acting on
10864 * this block group.
10866 spin_unlock(&block_group->lock);
10867 up_write(&space_info->groups_sem);
10870 spin_unlock(&block_group->lock);
10872 /* We don't want to force the issue, only flip if it's ok. */
10873 ret = inc_block_group_ro(block_group, 0);
10874 up_write(&space_info->groups_sem);
10881 * Want to do this before we do anything else so we can recover
10882 * properly if we fail to join the transaction.
10884 trans = btrfs_start_trans_remove_block_group(fs_info,
10885 block_group->key.objectid);
10886 if (IS_ERR(trans)) {
10887 btrfs_dec_block_group_ro(root, block_group);
10888 ret = PTR_ERR(trans);
10893 * We could have pending pinned extents for this block group,
10894 * just delete them, we don't care about them anymore.
10896 start = block_group->key.objectid;
10897 end = start + block_group->key.offset - 1;
10899 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10900 * btrfs_finish_extent_commit(). If we are at transaction N,
10901 * another task might be running finish_extent_commit() for the
10902 * previous transaction N - 1, and have seen a range belonging
10903 * to the block group in freed_extents[] before we were able to
10904 * clear the whole block group range from freed_extents[]. This
10905 * means that task can lookup for the block group after we
10906 * unpinned it from freed_extents[] and removed it, leading to
10907 * a BUG_ON() at btrfs_unpin_extent_range().
10909 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10910 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10913 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10914 btrfs_dec_block_group_ro(root, block_group);
10917 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10920 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10921 btrfs_dec_block_group_ro(root, block_group);
10924 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10926 /* Reset pinned so btrfs_put_block_group doesn't complain */
10927 spin_lock(&space_info->lock);
10928 spin_lock(&block_group->lock);
10930 space_info->bytes_pinned -= block_group->pinned;
10931 space_info->bytes_readonly += block_group->pinned;
10932 percpu_counter_add(&space_info->total_bytes_pinned,
10933 -block_group->pinned);
10934 block_group->pinned = 0;
10936 spin_unlock(&block_group->lock);
10937 spin_unlock(&space_info->lock);
10939 /* DISCARD can flip during remount */
10940 trimming = btrfs_test_opt(root->fs_info, DISCARD);
10942 /* Implicit trim during transaction commit. */
10944 btrfs_get_block_group_trimming(block_group);
10947 * Btrfs_remove_chunk will abort the transaction if things go
10950 ret = btrfs_remove_chunk(trans, root,
10951 block_group->key.objectid);
10955 btrfs_put_block_group_trimming(block_group);
10960 * If we're not mounted with -odiscard, we can just forget
10961 * about this block group. Otherwise we'll need to wait
10962 * until transaction commit to do the actual discard.
10965 spin_lock(&fs_info->unused_bgs_lock);
10967 * A concurrent scrub might have added us to the list
10968 * fs_info->unused_bgs, so use a list_move operation
10969 * to add the block group to the deleted_bgs list.
10971 list_move(&block_group->bg_list,
10972 &trans->transaction->deleted_bgs);
10973 spin_unlock(&fs_info->unused_bgs_lock);
10974 btrfs_get_block_group(block_group);
10977 btrfs_end_transaction(trans, root);
10979 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
10980 btrfs_put_block_group(block_group);
10981 spin_lock(&fs_info->unused_bgs_lock);
10983 spin_unlock(&fs_info->unused_bgs_lock);
10986 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10988 struct btrfs_space_info *space_info;
10989 struct btrfs_super_block *disk_super;
10995 disk_super = fs_info->super_copy;
10996 if (!btrfs_super_root(disk_super))
10999 features = btrfs_super_incompat_flags(disk_super);
11000 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
11003 flags = BTRFS_BLOCK_GROUP_SYSTEM;
11004 ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info);
11009 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
11010 ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info);
11012 flags = BTRFS_BLOCK_GROUP_METADATA;
11013 ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info);
11017 flags = BTRFS_BLOCK_GROUP_DATA;
11018 ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info);
11024 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
11026 return unpin_extent_range(root, start, end, false);
11030 * It used to be that old block groups would be left around forever.
11031 * Iterating over them would be enough to trim unused space. Since we
11032 * now automatically remove them, we also need to iterate over unallocated
11035 * We don't want a transaction for this since the discard may take a
11036 * substantial amount of time. We don't require that a transaction be
11037 * running, but we do need to take a running transaction into account
11038 * to ensure that we're not discarding chunks that were released in
11039 * the current transaction.
11041 * Holding the chunks lock will prevent other threads from allocating
11042 * or releasing chunks, but it won't prevent a running transaction
11043 * from committing and releasing the memory that the pending chunks
11044 * list head uses. For that, we need to take a reference to the
11047 static int btrfs_trim_free_extents(struct btrfs_device *device,
11048 u64 minlen, u64 *trimmed)
11050 u64 start = 0, len = 0;
11055 /* Not writeable = nothing to do. */
11056 if (!device->writeable)
11059 /* No free space = nothing to do. */
11060 if (device->total_bytes <= device->bytes_used)
11066 struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
11067 struct btrfs_transaction *trans;
11070 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
11074 down_read(&fs_info->commit_root_sem);
11076 spin_lock(&fs_info->trans_lock);
11077 trans = fs_info->running_transaction;
11079 atomic_inc(&trans->use_count);
11080 spin_unlock(&fs_info->trans_lock);
11082 ret = find_free_dev_extent_start(trans, device, minlen, start,
11085 btrfs_put_transaction(trans);
11088 up_read(&fs_info->commit_root_sem);
11089 mutex_unlock(&fs_info->chunk_mutex);
11090 if (ret == -ENOSPC)
11095 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
11096 up_read(&fs_info->commit_root_sem);
11097 mutex_unlock(&fs_info->chunk_mutex);
11105 if (fatal_signal_pending(current)) {
11106 ret = -ERESTARTSYS;
11116 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
11118 struct btrfs_fs_info *fs_info = root->fs_info;
11119 struct btrfs_block_group_cache *cache = NULL;
11120 struct btrfs_device *device;
11121 struct list_head *devices;
11126 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
11130 * try to trim all FS space, our block group may start from non-zero.
11132 if (range->len == total_bytes)
11133 cache = btrfs_lookup_first_block_group(fs_info, range->start);
11135 cache = btrfs_lookup_block_group(fs_info, range->start);
11138 if (cache->key.objectid >= (range->start + range->len)) {
11139 btrfs_put_block_group(cache);
11143 start = max(range->start, cache->key.objectid);
11144 end = min(range->start + range->len,
11145 cache->key.objectid + cache->key.offset);
11147 if (end - start >= range->minlen) {
11148 if (!block_group_cache_done(cache)) {
11149 ret = cache_block_group(cache, 0);
11151 btrfs_put_block_group(cache);
11154 ret = wait_block_group_cache_done(cache);
11156 btrfs_put_block_group(cache);
11160 ret = btrfs_trim_block_group(cache,
11166 trimmed += group_trimmed;
11168 btrfs_put_block_group(cache);
11173 cache = next_block_group(fs_info->tree_root, cache);
11176 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
11177 devices = &root->fs_info->fs_devices->alloc_list;
11178 list_for_each_entry(device, devices, dev_alloc_list) {
11179 ret = btrfs_trim_free_extents(device, range->minlen,
11184 trimmed += group_trimmed;
11186 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
11188 range->len = trimmed;
11193 * btrfs_{start,end}_write_no_snapshoting() are similar to
11194 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
11195 * data into the page cache through nocow before the subvolume is snapshoted,
11196 * but flush the data into disk after the snapshot creation, or to prevent
11197 * operations while snapshoting is ongoing and that cause the snapshot to be
11198 * inconsistent (writes followed by expanding truncates for example).
11200 void btrfs_end_write_no_snapshoting(struct btrfs_root *root)
11202 percpu_counter_dec(&root->subv_writers->counter);
11204 * Make sure counter is updated before we wake up waiters.
11207 if (waitqueue_active(&root->subv_writers->wait))
11208 wake_up(&root->subv_writers->wait);
11211 int btrfs_start_write_no_snapshoting(struct btrfs_root *root)
11213 if (atomic_read(&root->will_be_snapshoted))
11216 percpu_counter_inc(&root->subv_writers->counter);
11218 * Make sure counter is updated before we check for snapshot creation.
11221 if (atomic_read(&root->will_be_snapshoted)) {
11222 btrfs_end_write_no_snapshoting(root);
11228 static int wait_snapshoting_atomic_t(atomic_t *a)
11234 void btrfs_wait_for_snapshot_creation(struct btrfs_root *root)
11239 ret = btrfs_start_write_no_snapshoting(root);
11242 wait_on_atomic_t(&root->will_be_snapshoted,
11243 wait_snapshoting_atomic_t,
11244 TASK_UNINTERRUPTIBLE);