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_space_info *info, u64 bytes,
91 int dump_block_groups);
92 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache,
93 u64 ram_bytes, u64 num_bytes, int delalloc);
94 static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache,
95 u64 num_bytes, int delalloc);
96 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
98 int btrfs_pin_extent(struct btrfs_root *root,
99 u64 bytenr, u64 num_bytes, int reserved);
100 static int __reserve_metadata_bytes(struct btrfs_root *root,
101 struct btrfs_space_info *space_info,
103 enum btrfs_reserve_flush_enum flush);
104 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
105 struct btrfs_space_info *space_info,
107 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
108 struct btrfs_space_info *space_info,
112 block_group_cache_done(struct btrfs_block_group_cache *cache)
115 return cache->cached == BTRFS_CACHE_FINISHED ||
116 cache->cached == BTRFS_CACHE_ERROR;
119 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
121 return (cache->flags & bits) == bits;
124 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
126 atomic_inc(&cache->count);
129 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
131 if (atomic_dec_and_test(&cache->count)) {
132 WARN_ON(cache->pinned > 0);
133 WARN_ON(cache->reserved > 0);
134 kfree(cache->free_space_ctl);
140 * this adds the block group to the fs_info rb tree for the block group
143 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
144 struct btrfs_block_group_cache *block_group)
147 struct rb_node *parent = NULL;
148 struct btrfs_block_group_cache *cache;
150 spin_lock(&info->block_group_cache_lock);
151 p = &info->block_group_cache_tree.rb_node;
155 cache = rb_entry(parent, struct btrfs_block_group_cache,
157 if (block_group->key.objectid < cache->key.objectid) {
159 } else if (block_group->key.objectid > cache->key.objectid) {
162 spin_unlock(&info->block_group_cache_lock);
167 rb_link_node(&block_group->cache_node, parent, p);
168 rb_insert_color(&block_group->cache_node,
169 &info->block_group_cache_tree);
171 if (info->first_logical_byte > block_group->key.objectid)
172 info->first_logical_byte = block_group->key.objectid;
174 spin_unlock(&info->block_group_cache_lock);
180 * This will return the block group at or after bytenr if contains is 0, else
181 * it will return the block group that contains the bytenr
183 static struct btrfs_block_group_cache *
184 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
187 struct btrfs_block_group_cache *cache, *ret = NULL;
191 spin_lock(&info->block_group_cache_lock);
192 n = info->block_group_cache_tree.rb_node;
195 cache = rb_entry(n, struct btrfs_block_group_cache,
197 end = cache->key.objectid + cache->key.offset - 1;
198 start = cache->key.objectid;
200 if (bytenr < start) {
201 if (!contains && (!ret || start < ret->key.objectid))
204 } else if (bytenr > start) {
205 if (contains && bytenr <= end) {
216 btrfs_get_block_group(ret);
217 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
218 info->first_logical_byte = ret->key.objectid;
220 spin_unlock(&info->block_group_cache_lock);
225 static int add_excluded_extent(struct btrfs_root *root,
226 u64 start, u64 num_bytes)
228 u64 end = start + num_bytes - 1;
229 set_extent_bits(&root->fs_info->freed_extents[0],
230 start, end, EXTENT_UPTODATE);
231 set_extent_bits(&root->fs_info->freed_extents[1],
232 start, end, EXTENT_UPTODATE);
236 static void free_excluded_extents(struct btrfs_root *root,
237 struct btrfs_block_group_cache *cache)
241 start = cache->key.objectid;
242 end = start + cache->key.offset - 1;
244 clear_extent_bits(&root->fs_info->freed_extents[0],
245 start, end, EXTENT_UPTODATE);
246 clear_extent_bits(&root->fs_info->freed_extents[1],
247 start, end, EXTENT_UPTODATE);
250 static int exclude_super_stripes(struct btrfs_root *root,
251 struct btrfs_block_group_cache *cache)
258 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
259 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
260 cache->bytes_super += stripe_len;
261 ret = add_excluded_extent(root, cache->key.objectid,
267 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
268 bytenr = btrfs_sb_offset(i);
269 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
270 cache->key.objectid, bytenr,
271 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 BUG_ON(node->ref_mod != 1);
2356 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2357 BUG_ON(!extent_op || !extent_op->update_flags);
2358 ret = alloc_reserved_tree_block(trans, root,
2360 extent_op->flags_to_set,
2363 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2364 ret = __btrfs_inc_extent_ref(trans, root, node,
2368 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2369 ret = __btrfs_free_extent(trans, root, node,
2371 ref->level, 0, 1, extent_op);
2378 /* helper function to actually process a single delayed ref entry */
2379 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2380 struct btrfs_root *root,
2381 struct btrfs_delayed_ref_node *node,
2382 struct btrfs_delayed_extent_op *extent_op,
2383 int insert_reserved)
2387 if (trans->aborted) {
2388 if (insert_reserved)
2389 btrfs_pin_extent(root, node->bytenr,
2390 node->num_bytes, 1);
2394 if (btrfs_delayed_ref_is_head(node)) {
2395 struct btrfs_delayed_ref_head *head;
2397 * we've hit the end of the chain and we were supposed
2398 * to insert this extent into the tree. But, it got
2399 * deleted before we ever needed to insert it, so all
2400 * we have to do is clean up the accounting
2403 head = btrfs_delayed_node_to_head(node);
2404 trace_run_delayed_ref_head(root->fs_info, node, head,
2407 if (insert_reserved) {
2408 btrfs_pin_extent(root, node->bytenr,
2409 node->num_bytes, 1);
2410 if (head->is_data) {
2411 ret = btrfs_del_csums(trans, root,
2417 /* Also free its reserved qgroup space */
2418 btrfs_qgroup_free_delayed_ref(root->fs_info,
2419 head->qgroup_ref_root,
2420 head->qgroup_reserved);
2424 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2425 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2426 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2428 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2429 node->type == BTRFS_SHARED_DATA_REF_KEY)
2430 ret = run_delayed_data_ref(trans, root, node, extent_op,
2437 static inline struct btrfs_delayed_ref_node *
2438 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2440 struct btrfs_delayed_ref_node *ref;
2442 if (list_empty(&head->ref_list))
2446 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2447 * This is to prevent a ref count from going down to zero, which deletes
2448 * the extent item from the extent tree, when there still are references
2449 * to add, which would fail because they would not find the extent item.
2451 list_for_each_entry(ref, &head->ref_list, list) {
2452 if (ref->action == BTRFS_ADD_DELAYED_REF)
2456 return list_entry(head->ref_list.next, struct btrfs_delayed_ref_node,
2461 * Returns 0 on success or if called with an already aborted transaction.
2462 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2464 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2465 struct btrfs_root *root,
2468 struct btrfs_delayed_ref_root *delayed_refs;
2469 struct btrfs_delayed_ref_node *ref;
2470 struct btrfs_delayed_ref_head *locked_ref = NULL;
2471 struct btrfs_delayed_extent_op *extent_op;
2472 struct btrfs_fs_info *fs_info = root->fs_info;
2473 ktime_t start = ktime_get();
2475 unsigned long count = 0;
2476 unsigned long actual_count = 0;
2477 int must_insert_reserved = 0;
2479 delayed_refs = &trans->transaction->delayed_refs;
2485 spin_lock(&delayed_refs->lock);
2486 locked_ref = btrfs_select_ref_head(trans);
2488 spin_unlock(&delayed_refs->lock);
2492 /* grab the lock that says we are going to process
2493 * all the refs for this head */
2494 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2495 spin_unlock(&delayed_refs->lock);
2497 * we may have dropped the spin lock to get the head
2498 * mutex lock, and that might have given someone else
2499 * time to free the head. If that's true, it has been
2500 * removed from our list and we can move on.
2502 if (ret == -EAGAIN) {
2510 * We need to try and merge add/drops of the same ref since we
2511 * can run into issues with relocate dropping the implicit ref
2512 * and then it being added back again before the drop can
2513 * finish. If we merged anything we need to re-loop so we can
2515 * Or we can get node references of the same type that weren't
2516 * merged when created due to bumps in the tree mod seq, and
2517 * we need to merge them to prevent adding an inline extent
2518 * backref before dropping it (triggering a BUG_ON at
2519 * insert_inline_extent_backref()).
2521 spin_lock(&locked_ref->lock);
2522 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2526 * locked_ref is the head node, so we have to go one
2527 * node back for any delayed ref updates
2529 ref = select_delayed_ref(locked_ref);
2531 if (ref && ref->seq &&
2532 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2533 spin_unlock(&locked_ref->lock);
2534 btrfs_delayed_ref_unlock(locked_ref);
2535 spin_lock(&delayed_refs->lock);
2536 locked_ref->processing = 0;
2537 delayed_refs->num_heads_ready++;
2538 spin_unlock(&delayed_refs->lock);
2546 * record the must insert reserved flag before we
2547 * drop the spin lock.
2549 must_insert_reserved = locked_ref->must_insert_reserved;
2550 locked_ref->must_insert_reserved = 0;
2552 extent_op = locked_ref->extent_op;
2553 locked_ref->extent_op = NULL;
2558 /* All delayed refs have been processed, Go ahead
2559 * and send the head node to run_one_delayed_ref,
2560 * so that any accounting fixes can happen
2562 ref = &locked_ref->node;
2564 if (extent_op && must_insert_reserved) {
2565 btrfs_free_delayed_extent_op(extent_op);
2570 spin_unlock(&locked_ref->lock);
2571 ret = run_delayed_extent_op(trans, root,
2573 btrfs_free_delayed_extent_op(extent_op);
2577 * Need to reset must_insert_reserved if
2578 * there was an error so the abort stuff
2579 * can cleanup the reserved space
2582 if (must_insert_reserved)
2583 locked_ref->must_insert_reserved = 1;
2584 locked_ref->processing = 0;
2585 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2586 btrfs_delayed_ref_unlock(locked_ref);
2593 * Need to drop our head ref lock and re-acquire the
2594 * delayed ref lock and then re-check to make sure
2597 spin_unlock(&locked_ref->lock);
2598 spin_lock(&delayed_refs->lock);
2599 spin_lock(&locked_ref->lock);
2600 if (!list_empty(&locked_ref->ref_list) ||
2601 locked_ref->extent_op) {
2602 spin_unlock(&locked_ref->lock);
2603 spin_unlock(&delayed_refs->lock);
2607 delayed_refs->num_heads--;
2608 rb_erase(&locked_ref->href_node,
2609 &delayed_refs->href_root);
2610 spin_unlock(&delayed_refs->lock);
2614 list_del(&ref->list);
2616 atomic_dec(&delayed_refs->num_entries);
2618 if (!btrfs_delayed_ref_is_head(ref)) {
2620 * when we play the delayed ref, also correct the
2623 switch (ref->action) {
2624 case BTRFS_ADD_DELAYED_REF:
2625 case BTRFS_ADD_DELAYED_EXTENT:
2626 locked_ref->node.ref_mod -= ref->ref_mod;
2628 case BTRFS_DROP_DELAYED_REF:
2629 locked_ref->node.ref_mod += ref->ref_mod;
2635 spin_unlock(&locked_ref->lock);
2637 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2638 must_insert_reserved);
2640 btrfs_free_delayed_extent_op(extent_op);
2642 locked_ref->processing = 0;
2643 btrfs_delayed_ref_unlock(locked_ref);
2644 btrfs_put_delayed_ref(ref);
2645 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2650 * If this node is a head, that means all the refs in this head
2651 * have been dealt with, and we will pick the next head to deal
2652 * with, so we must unlock the head and drop it from the cluster
2653 * list before we release it.
2655 if (btrfs_delayed_ref_is_head(ref)) {
2656 if (locked_ref->is_data &&
2657 locked_ref->total_ref_mod < 0) {
2658 spin_lock(&delayed_refs->lock);
2659 delayed_refs->pending_csums -= ref->num_bytes;
2660 spin_unlock(&delayed_refs->lock);
2662 btrfs_delayed_ref_unlock(locked_ref);
2665 btrfs_put_delayed_ref(ref);
2671 * We don't want to include ref heads since we can have empty ref heads
2672 * and those will drastically skew our runtime down since we just do
2673 * accounting, no actual extent tree updates.
2675 if (actual_count > 0) {
2676 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2680 * We weigh the current average higher than our current runtime
2681 * to avoid large swings in the average.
2683 spin_lock(&delayed_refs->lock);
2684 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2685 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2686 spin_unlock(&delayed_refs->lock);
2691 #ifdef SCRAMBLE_DELAYED_REFS
2693 * Normally delayed refs get processed in ascending bytenr order. This
2694 * correlates in most cases to the order added. To expose dependencies on this
2695 * order, we start to process the tree in the middle instead of the beginning
2697 static u64 find_middle(struct rb_root *root)
2699 struct rb_node *n = root->rb_node;
2700 struct btrfs_delayed_ref_node *entry;
2703 u64 first = 0, last = 0;
2707 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2708 first = entry->bytenr;
2712 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2713 last = entry->bytenr;
2718 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2719 WARN_ON(!entry->in_tree);
2721 middle = entry->bytenr;
2734 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2738 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2739 sizeof(struct btrfs_extent_inline_ref));
2740 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2741 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2744 * We don't ever fill up leaves all the way so multiply by 2 just to be
2745 * closer to what we're really going to want to use.
2747 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2751 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2752 * would require to store the csums for that many bytes.
2754 u64 btrfs_csum_bytes_to_leaves(struct btrfs_root *root, u64 csum_bytes)
2757 u64 num_csums_per_leaf;
2760 csum_size = BTRFS_MAX_ITEM_SIZE(root);
2761 num_csums_per_leaf = div64_u64(csum_size,
2762 (u64)btrfs_super_csum_size(root->fs_info->super_copy));
2763 num_csums = div64_u64(csum_bytes, root->sectorsize);
2764 num_csums += num_csums_per_leaf - 1;
2765 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2769 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2770 struct btrfs_root *root)
2772 struct btrfs_block_rsv *global_rsv;
2773 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2774 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2775 u64 num_dirty_bgs = trans->transaction->num_dirty_bgs;
2776 u64 num_bytes, num_dirty_bgs_bytes;
2779 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2780 num_heads = heads_to_leaves(root, num_heads);
2782 num_bytes += (num_heads - 1) * root->nodesize;
2784 num_bytes += btrfs_csum_bytes_to_leaves(root, csum_bytes) * root->nodesize;
2785 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(root,
2787 global_rsv = &root->fs_info->global_block_rsv;
2790 * If we can't allocate any more chunks lets make sure we have _lots_ of
2791 * wiggle room since running delayed refs can create more delayed refs.
2793 if (global_rsv->space_info->full) {
2794 num_dirty_bgs_bytes <<= 1;
2798 spin_lock(&global_rsv->lock);
2799 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2801 spin_unlock(&global_rsv->lock);
2805 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2806 struct btrfs_root *root)
2808 struct btrfs_fs_info *fs_info = root->fs_info;
2810 atomic_read(&trans->transaction->delayed_refs.num_entries);
2815 avg_runtime = fs_info->avg_delayed_ref_runtime;
2816 val = num_entries * avg_runtime;
2817 if (num_entries * avg_runtime >= NSEC_PER_SEC)
2819 if (val >= NSEC_PER_SEC / 2)
2822 return btrfs_check_space_for_delayed_refs(trans, root);
2825 struct async_delayed_refs {
2826 struct btrfs_root *root;
2831 struct completion wait;
2832 struct btrfs_work work;
2835 static void delayed_ref_async_start(struct btrfs_work *work)
2837 struct async_delayed_refs *async;
2838 struct btrfs_trans_handle *trans;
2841 async = container_of(work, struct async_delayed_refs, work);
2843 /* if the commit is already started, we don't need to wait here */
2844 if (btrfs_transaction_blocked(async->root->fs_info))
2847 trans = btrfs_join_transaction(async->root);
2848 if (IS_ERR(trans)) {
2849 async->error = PTR_ERR(trans);
2854 * trans->sync means that when we call end_transaction, we won't
2855 * wait on delayed refs
2859 /* Don't bother flushing if we got into a different transaction */
2860 if (trans->transid > async->transid)
2863 ret = btrfs_run_delayed_refs(trans, async->root, async->count);
2867 ret = btrfs_end_transaction(trans, async->root);
2868 if (ret && !async->error)
2872 complete(&async->wait);
2877 int btrfs_async_run_delayed_refs(struct btrfs_root *root,
2878 unsigned long count, u64 transid, int wait)
2880 struct async_delayed_refs *async;
2883 async = kmalloc(sizeof(*async), GFP_NOFS);
2887 async->root = root->fs_info->tree_root;
2888 async->count = count;
2890 async->transid = transid;
2895 init_completion(&async->wait);
2897 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2898 delayed_ref_async_start, NULL, NULL);
2900 btrfs_queue_work(root->fs_info->extent_workers, &async->work);
2903 wait_for_completion(&async->wait);
2912 * this starts processing the delayed reference count updates and
2913 * extent insertions we have queued up so far. count can be
2914 * 0, which means to process everything in the tree at the start
2915 * of the run (but not newly added entries), or it can be some target
2916 * number you'd like to process.
2918 * Returns 0 on success or if called with an aborted transaction
2919 * Returns <0 on error and aborts the transaction
2921 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2922 struct btrfs_root *root, unsigned long count)
2924 struct rb_node *node;
2925 struct btrfs_delayed_ref_root *delayed_refs;
2926 struct btrfs_delayed_ref_head *head;
2928 int run_all = count == (unsigned long)-1;
2929 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
2931 /* We'll clean this up in btrfs_cleanup_transaction */
2935 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &root->fs_info->flags))
2938 if (root == root->fs_info->extent_root)
2939 root = root->fs_info->tree_root;
2941 delayed_refs = &trans->transaction->delayed_refs;
2943 count = atomic_read(&delayed_refs->num_entries) * 2;
2946 #ifdef SCRAMBLE_DELAYED_REFS
2947 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2949 trans->can_flush_pending_bgs = false;
2950 ret = __btrfs_run_delayed_refs(trans, root, count);
2952 btrfs_abort_transaction(trans, ret);
2957 if (!list_empty(&trans->new_bgs))
2958 btrfs_create_pending_block_groups(trans, root);
2960 spin_lock(&delayed_refs->lock);
2961 node = rb_first(&delayed_refs->href_root);
2963 spin_unlock(&delayed_refs->lock);
2966 count = (unsigned long)-1;
2969 head = rb_entry(node, struct btrfs_delayed_ref_head,
2971 if (btrfs_delayed_ref_is_head(&head->node)) {
2972 struct btrfs_delayed_ref_node *ref;
2975 atomic_inc(&ref->refs);
2977 spin_unlock(&delayed_refs->lock);
2979 * Mutex was contended, block until it's
2980 * released and try again
2982 mutex_lock(&head->mutex);
2983 mutex_unlock(&head->mutex);
2985 btrfs_put_delayed_ref(ref);
2991 node = rb_next(node);
2993 spin_unlock(&delayed_refs->lock);
2998 assert_qgroups_uptodate(trans);
2999 trans->can_flush_pending_bgs = can_flush_pending_bgs;
3003 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
3004 struct btrfs_root *root,
3005 u64 bytenr, u64 num_bytes, u64 flags,
3006 int level, int is_data)
3008 struct btrfs_delayed_extent_op *extent_op;
3011 extent_op = btrfs_alloc_delayed_extent_op();
3015 extent_op->flags_to_set = flags;
3016 extent_op->update_flags = true;
3017 extent_op->update_key = false;
3018 extent_op->is_data = is_data ? true : false;
3019 extent_op->level = level;
3021 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
3022 num_bytes, extent_op);
3024 btrfs_free_delayed_extent_op(extent_op);
3028 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
3029 struct btrfs_root *root,
3030 struct btrfs_path *path,
3031 u64 objectid, u64 offset, u64 bytenr)
3033 struct btrfs_delayed_ref_head *head;
3034 struct btrfs_delayed_ref_node *ref;
3035 struct btrfs_delayed_data_ref *data_ref;
3036 struct btrfs_delayed_ref_root *delayed_refs;
3039 delayed_refs = &trans->transaction->delayed_refs;
3040 spin_lock(&delayed_refs->lock);
3041 head = btrfs_find_delayed_ref_head(trans, bytenr);
3043 spin_unlock(&delayed_refs->lock);
3047 if (!mutex_trylock(&head->mutex)) {
3048 atomic_inc(&head->node.refs);
3049 spin_unlock(&delayed_refs->lock);
3051 btrfs_release_path(path);
3054 * Mutex was contended, block until it's released and let
3057 mutex_lock(&head->mutex);
3058 mutex_unlock(&head->mutex);
3059 btrfs_put_delayed_ref(&head->node);
3062 spin_unlock(&delayed_refs->lock);
3064 spin_lock(&head->lock);
3065 list_for_each_entry(ref, &head->ref_list, list) {
3066 /* If it's a shared ref we know a cross reference exists */
3067 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
3072 data_ref = btrfs_delayed_node_to_data_ref(ref);
3075 * If our ref doesn't match the one we're currently looking at
3076 * then we have a cross reference.
3078 if (data_ref->root != root->root_key.objectid ||
3079 data_ref->objectid != objectid ||
3080 data_ref->offset != offset) {
3085 spin_unlock(&head->lock);
3086 mutex_unlock(&head->mutex);
3090 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
3091 struct btrfs_root *root,
3092 struct btrfs_path *path,
3093 u64 objectid, u64 offset, u64 bytenr)
3095 struct btrfs_root *extent_root = root->fs_info->extent_root;
3096 struct extent_buffer *leaf;
3097 struct btrfs_extent_data_ref *ref;
3098 struct btrfs_extent_inline_ref *iref;
3099 struct btrfs_extent_item *ei;
3100 struct btrfs_key key;
3104 key.objectid = bytenr;
3105 key.offset = (u64)-1;
3106 key.type = BTRFS_EXTENT_ITEM_KEY;
3108 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3111 BUG_ON(ret == 0); /* Corruption */
3114 if (path->slots[0] == 0)
3118 leaf = path->nodes[0];
3119 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3121 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3125 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3126 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3127 if (item_size < sizeof(*ei)) {
3128 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3132 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3134 if (item_size != sizeof(*ei) +
3135 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3138 if (btrfs_extent_generation(leaf, ei) <=
3139 btrfs_root_last_snapshot(&root->root_item))
3142 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3143 if (btrfs_extent_inline_ref_type(leaf, iref) !=
3144 BTRFS_EXTENT_DATA_REF_KEY)
3147 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3148 if (btrfs_extent_refs(leaf, ei) !=
3149 btrfs_extent_data_ref_count(leaf, ref) ||
3150 btrfs_extent_data_ref_root(leaf, ref) !=
3151 root->root_key.objectid ||
3152 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3153 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3161 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
3162 struct btrfs_root *root,
3163 u64 objectid, u64 offset, u64 bytenr)
3165 struct btrfs_path *path;
3169 path = btrfs_alloc_path();
3174 ret = check_committed_ref(trans, root, path, objectid,
3176 if (ret && ret != -ENOENT)
3179 ret2 = check_delayed_ref(trans, root, path, objectid,
3181 } while (ret2 == -EAGAIN);
3183 if (ret2 && ret2 != -ENOENT) {
3188 if (ret != -ENOENT || ret2 != -ENOENT)
3191 btrfs_free_path(path);
3192 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3197 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3198 struct btrfs_root *root,
3199 struct extent_buffer *buf,
3200 int full_backref, int inc)
3207 struct btrfs_key key;
3208 struct btrfs_file_extent_item *fi;
3212 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3213 u64, u64, u64, u64, u64, u64);
3216 if (btrfs_is_testing(root->fs_info))
3219 ref_root = btrfs_header_owner(buf);
3220 nritems = btrfs_header_nritems(buf);
3221 level = btrfs_header_level(buf);
3223 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3227 process_func = btrfs_inc_extent_ref;
3229 process_func = btrfs_free_extent;
3232 parent = buf->start;
3236 for (i = 0; i < nritems; i++) {
3238 btrfs_item_key_to_cpu(buf, &key, i);
3239 if (key.type != BTRFS_EXTENT_DATA_KEY)
3241 fi = btrfs_item_ptr(buf, i,
3242 struct btrfs_file_extent_item);
3243 if (btrfs_file_extent_type(buf, fi) ==
3244 BTRFS_FILE_EXTENT_INLINE)
3246 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3250 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3251 key.offset -= btrfs_file_extent_offset(buf, fi);
3252 ret = process_func(trans, root, bytenr, num_bytes,
3253 parent, ref_root, key.objectid,
3258 bytenr = btrfs_node_blockptr(buf, i);
3259 num_bytes = root->nodesize;
3260 ret = process_func(trans, root, bytenr, num_bytes,
3261 parent, ref_root, level - 1, 0);
3271 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3272 struct extent_buffer *buf, int full_backref)
3274 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3277 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3278 struct extent_buffer *buf, int full_backref)
3280 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3283 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3284 struct btrfs_root *root,
3285 struct btrfs_path *path,
3286 struct btrfs_block_group_cache *cache)
3289 struct btrfs_root *extent_root = root->fs_info->extent_root;
3291 struct extent_buffer *leaf;
3293 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3300 leaf = path->nodes[0];
3301 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3302 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3303 btrfs_mark_buffer_dirty(leaf);
3305 btrfs_release_path(path);
3310 static struct btrfs_block_group_cache *
3311 next_block_group(struct btrfs_root *root,
3312 struct btrfs_block_group_cache *cache)
3314 struct rb_node *node;
3316 spin_lock(&root->fs_info->block_group_cache_lock);
3318 /* If our block group was removed, we need a full search. */
3319 if (RB_EMPTY_NODE(&cache->cache_node)) {
3320 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3322 spin_unlock(&root->fs_info->block_group_cache_lock);
3323 btrfs_put_block_group(cache);
3324 cache = btrfs_lookup_first_block_group(root->fs_info,
3328 node = rb_next(&cache->cache_node);
3329 btrfs_put_block_group(cache);
3331 cache = rb_entry(node, struct btrfs_block_group_cache,
3333 btrfs_get_block_group(cache);
3336 spin_unlock(&root->fs_info->block_group_cache_lock);
3340 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3341 struct btrfs_trans_handle *trans,
3342 struct btrfs_path *path)
3344 struct btrfs_root *root = block_group->fs_info->tree_root;
3345 struct inode *inode = NULL;
3347 int dcs = BTRFS_DC_ERROR;
3353 * If this block group is smaller than 100 megs don't bother caching the
3356 if (block_group->key.offset < (100 * SZ_1M)) {
3357 spin_lock(&block_group->lock);
3358 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3359 spin_unlock(&block_group->lock);
3366 inode = lookup_free_space_inode(root, block_group, path);
3367 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3368 ret = PTR_ERR(inode);
3369 btrfs_release_path(path);
3373 if (IS_ERR(inode)) {
3377 if (block_group->ro)
3380 ret = create_free_space_inode(root, trans, block_group, path);
3386 /* We've already setup this transaction, go ahead and exit */
3387 if (block_group->cache_generation == trans->transid &&
3388 i_size_read(inode)) {
3389 dcs = BTRFS_DC_SETUP;
3394 * We want to set the generation to 0, that way if anything goes wrong
3395 * from here on out we know not to trust this cache when we load up next
3398 BTRFS_I(inode)->generation = 0;
3399 ret = btrfs_update_inode(trans, root, inode);
3402 * So theoretically we could recover from this, simply set the
3403 * super cache generation to 0 so we know to invalidate the
3404 * cache, but then we'd have to keep track of the block groups
3405 * that fail this way so we know we _have_ to reset this cache
3406 * before the next commit or risk reading stale cache. So to
3407 * limit our exposure to horrible edge cases lets just abort the
3408 * transaction, this only happens in really bad situations
3411 btrfs_abort_transaction(trans, ret);
3416 if (i_size_read(inode) > 0) {
3417 ret = btrfs_check_trunc_cache_free_space(root,
3418 &root->fs_info->global_block_rsv);
3422 ret = btrfs_truncate_free_space_cache(root, trans, NULL, inode);
3427 spin_lock(&block_group->lock);
3428 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3429 !btrfs_test_opt(root->fs_info, SPACE_CACHE)) {
3431 * don't bother trying to write stuff out _if_
3432 * a) we're not cached,
3433 * b) we're with nospace_cache mount option.
3435 dcs = BTRFS_DC_WRITTEN;
3436 spin_unlock(&block_group->lock);
3439 spin_unlock(&block_group->lock);
3442 * We hit an ENOSPC when setting up the cache in this transaction, just
3443 * skip doing the setup, we've already cleared the cache so we're safe.
3445 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
3451 * Try to preallocate enough space based on how big the block group is.
3452 * Keep in mind this has to include any pinned space which could end up
3453 * taking up quite a bit since it's not folded into the other space
3456 num_pages = div_u64(block_group->key.offset, SZ_256M);
3461 num_pages *= PAGE_SIZE;
3463 ret = btrfs_check_data_free_space(inode, 0, num_pages);
3467 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3468 num_pages, num_pages,
3471 * Our cache requires contiguous chunks so that we don't modify a bunch
3472 * of metadata or split extents when writing the cache out, which means
3473 * we can enospc if we are heavily fragmented in addition to just normal
3474 * out of space conditions. So if we hit this just skip setting up any
3475 * other block groups for this transaction, maybe we'll unpin enough
3476 * space the next time around.
3479 dcs = BTRFS_DC_SETUP;
3480 else if (ret == -ENOSPC)
3481 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
3486 btrfs_release_path(path);
3488 spin_lock(&block_group->lock);
3489 if (!ret && dcs == BTRFS_DC_SETUP)
3490 block_group->cache_generation = trans->transid;
3491 block_group->disk_cache_state = dcs;
3492 spin_unlock(&block_group->lock);
3497 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3498 struct btrfs_root *root)
3500 struct btrfs_block_group_cache *cache, *tmp;
3501 struct btrfs_transaction *cur_trans = trans->transaction;
3502 struct btrfs_path *path;
3504 if (list_empty(&cur_trans->dirty_bgs) ||
3505 !btrfs_test_opt(root->fs_info, SPACE_CACHE))
3508 path = btrfs_alloc_path();
3512 /* Could add new block groups, use _safe just in case */
3513 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3515 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3516 cache_save_setup(cache, trans, path);
3519 btrfs_free_path(path);
3524 * transaction commit does final block group cache writeback during a
3525 * critical section where nothing is allowed to change the FS. This is
3526 * required in order for the cache to actually match the block group,
3527 * but can introduce a lot of latency into the commit.
3529 * So, btrfs_start_dirty_block_groups is here to kick off block group
3530 * cache IO. There's a chance we'll have to redo some of it if the
3531 * block group changes again during the commit, but it greatly reduces
3532 * the commit latency by getting rid of the easy block groups while
3533 * we're still allowing others to join the commit.
3535 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans,
3536 struct btrfs_root *root)
3538 struct btrfs_block_group_cache *cache;
3539 struct btrfs_transaction *cur_trans = trans->transaction;
3542 struct btrfs_path *path = NULL;
3544 struct list_head *io = &cur_trans->io_bgs;
3545 int num_started = 0;
3548 spin_lock(&cur_trans->dirty_bgs_lock);
3549 if (list_empty(&cur_trans->dirty_bgs)) {
3550 spin_unlock(&cur_trans->dirty_bgs_lock);
3553 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3554 spin_unlock(&cur_trans->dirty_bgs_lock);
3558 * make sure all the block groups on our dirty list actually
3561 btrfs_create_pending_block_groups(trans, root);
3564 path = btrfs_alloc_path();
3570 * cache_write_mutex is here only to save us from balance or automatic
3571 * removal of empty block groups deleting this block group while we are
3572 * writing out the cache
3574 mutex_lock(&trans->transaction->cache_write_mutex);
3575 while (!list_empty(&dirty)) {
3576 cache = list_first_entry(&dirty,
3577 struct btrfs_block_group_cache,
3580 * this can happen if something re-dirties a block
3581 * group that is already under IO. Just wait for it to
3582 * finish and then do it all again
3584 if (!list_empty(&cache->io_list)) {
3585 list_del_init(&cache->io_list);
3586 btrfs_wait_cache_io(root, trans, cache,
3587 &cache->io_ctl, path,
3588 cache->key.objectid);
3589 btrfs_put_block_group(cache);
3594 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3595 * if it should update the cache_state. Don't delete
3596 * until after we wait.
3598 * Since we're not running in the commit critical section
3599 * we need the dirty_bgs_lock to protect from update_block_group
3601 spin_lock(&cur_trans->dirty_bgs_lock);
3602 list_del_init(&cache->dirty_list);
3603 spin_unlock(&cur_trans->dirty_bgs_lock);
3607 cache_save_setup(cache, trans, path);
3609 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3610 cache->io_ctl.inode = NULL;
3611 ret = btrfs_write_out_cache(root, trans, cache, path);
3612 if (ret == 0 && cache->io_ctl.inode) {
3617 * the cache_write_mutex is protecting
3620 list_add_tail(&cache->io_list, io);
3623 * if we failed to write the cache, the
3624 * generation will be bad and life goes on
3630 ret = write_one_cache_group(trans, root, path, cache);
3632 * Our block group might still be attached to the list
3633 * of new block groups in the transaction handle of some
3634 * other task (struct btrfs_trans_handle->new_bgs). This
3635 * means its block group item isn't yet in the extent
3636 * tree. If this happens ignore the error, as we will
3637 * try again later in the critical section of the
3638 * transaction commit.
3640 if (ret == -ENOENT) {
3642 spin_lock(&cur_trans->dirty_bgs_lock);
3643 if (list_empty(&cache->dirty_list)) {
3644 list_add_tail(&cache->dirty_list,
3645 &cur_trans->dirty_bgs);
3646 btrfs_get_block_group(cache);
3648 spin_unlock(&cur_trans->dirty_bgs_lock);
3650 btrfs_abort_transaction(trans, ret);
3654 /* if its not on the io list, we need to put the block group */
3656 btrfs_put_block_group(cache);
3662 * Avoid blocking other tasks for too long. It might even save
3663 * us from writing caches for block groups that are going to be
3666 mutex_unlock(&trans->transaction->cache_write_mutex);
3667 mutex_lock(&trans->transaction->cache_write_mutex);
3669 mutex_unlock(&trans->transaction->cache_write_mutex);
3672 * go through delayed refs for all the stuff we've just kicked off
3673 * and then loop back (just once)
3675 ret = btrfs_run_delayed_refs(trans, root, 0);
3676 if (!ret && loops == 0) {
3678 spin_lock(&cur_trans->dirty_bgs_lock);
3679 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3681 * dirty_bgs_lock protects us from concurrent block group
3682 * deletes too (not just cache_write_mutex).
3684 if (!list_empty(&dirty)) {
3685 spin_unlock(&cur_trans->dirty_bgs_lock);
3688 spin_unlock(&cur_trans->dirty_bgs_lock);
3689 } else if (ret < 0) {
3690 btrfs_cleanup_dirty_bgs(cur_trans, root);
3693 btrfs_free_path(path);
3697 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3698 struct btrfs_root *root)
3700 struct btrfs_block_group_cache *cache;
3701 struct btrfs_transaction *cur_trans = trans->transaction;
3704 struct btrfs_path *path;
3705 struct list_head *io = &cur_trans->io_bgs;
3706 int num_started = 0;
3708 path = btrfs_alloc_path();
3713 * Even though we are in the critical section of the transaction commit,
3714 * we can still have concurrent tasks adding elements to this
3715 * transaction's list of dirty block groups. These tasks correspond to
3716 * endio free space workers started when writeback finishes for a
3717 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3718 * allocate new block groups as a result of COWing nodes of the root
3719 * tree when updating the free space inode. The writeback for the space
3720 * caches is triggered by an earlier call to
3721 * btrfs_start_dirty_block_groups() and iterations of the following
3723 * Also we want to do the cache_save_setup first and then run the
3724 * delayed refs to make sure we have the best chance at doing this all
3727 spin_lock(&cur_trans->dirty_bgs_lock);
3728 while (!list_empty(&cur_trans->dirty_bgs)) {
3729 cache = list_first_entry(&cur_trans->dirty_bgs,
3730 struct btrfs_block_group_cache,
3734 * this can happen if cache_save_setup re-dirties a block
3735 * group that is already under IO. Just wait for it to
3736 * finish and then do it all again
3738 if (!list_empty(&cache->io_list)) {
3739 spin_unlock(&cur_trans->dirty_bgs_lock);
3740 list_del_init(&cache->io_list);
3741 btrfs_wait_cache_io(root, trans, cache,
3742 &cache->io_ctl, path,
3743 cache->key.objectid);
3744 btrfs_put_block_group(cache);
3745 spin_lock(&cur_trans->dirty_bgs_lock);
3749 * don't remove from the dirty list until after we've waited
3752 list_del_init(&cache->dirty_list);
3753 spin_unlock(&cur_trans->dirty_bgs_lock);
3756 cache_save_setup(cache, trans, path);
3759 ret = btrfs_run_delayed_refs(trans, root, (unsigned long) -1);
3761 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3762 cache->io_ctl.inode = NULL;
3763 ret = btrfs_write_out_cache(root, trans, cache, path);
3764 if (ret == 0 && cache->io_ctl.inode) {
3767 list_add_tail(&cache->io_list, io);
3770 * if we failed to write the cache, the
3771 * generation will be bad and life goes on
3777 ret = write_one_cache_group(trans, root, path, cache);
3779 * One of the free space endio workers might have
3780 * created a new block group while updating a free space
3781 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3782 * and hasn't released its transaction handle yet, in
3783 * which case the new block group is still attached to
3784 * its transaction handle and its creation has not
3785 * finished yet (no block group item in the extent tree
3786 * yet, etc). If this is the case, wait for all free
3787 * space endio workers to finish and retry. This is a
3788 * a very rare case so no need for a more efficient and
3791 if (ret == -ENOENT) {
3792 wait_event(cur_trans->writer_wait,
3793 atomic_read(&cur_trans->num_writers) == 1);
3794 ret = write_one_cache_group(trans, root, path,
3798 btrfs_abort_transaction(trans, ret);
3801 /* if its not on the io list, we need to put the block group */
3803 btrfs_put_block_group(cache);
3804 spin_lock(&cur_trans->dirty_bgs_lock);
3806 spin_unlock(&cur_trans->dirty_bgs_lock);
3808 while (!list_empty(io)) {
3809 cache = list_first_entry(io, struct btrfs_block_group_cache,
3811 list_del_init(&cache->io_list);
3812 btrfs_wait_cache_io(root, trans, cache,
3813 &cache->io_ctl, path, cache->key.objectid);
3814 btrfs_put_block_group(cache);
3817 btrfs_free_path(path);
3821 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3823 struct btrfs_block_group_cache *block_group;
3826 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3827 if (!block_group || block_group->ro)
3830 btrfs_put_block_group(block_group);
3834 bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3836 struct btrfs_block_group_cache *bg;
3839 bg = btrfs_lookup_block_group(fs_info, bytenr);
3843 spin_lock(&bg->lock);
3847 atomic_inc(&bg->nocow_writers);
3848 spin_unlock(&bg->lock);
3850 /* no put on block group, done by btrfs_dec_nocow_writers */
3852 btrfs_put_block_group(bg);
3858 void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3860 struct btrfs_block_group_cache *bg;
3862 bg = btrfs_lookup_block_group(fs_info, bytenr);
3864 if (atomic_dec_and_test(&bg->nocow_writers))
3865 wake_up_atomic_t(&bg->nocow_writers);
3867 * Once for our lookup and once for the lookup done by a previous call
3868 * to btrfs_inc_nocow_writers()
3870 btrfs_put_block_group(bg);
3871 btrfs_put_block_group(bg);
3874 static int btrfs_wait_nocow_writers_atomic_t(atomic_t *a)
3880 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg)
3882 wait_on_atomic_t(&bg->nocow_writers,
3883 btrfs_wait_nocow_writers_atomic_t,
3884 TASK_UNINTERRUPTIBLE);
3887 static const char *alloc_name(u64 flags)
3890 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3892 case BTRFS_BLOCK_GROUP_METADATA:
3894 case BTRFS_BLOCK_GROUP_DATA:
3896 case BTRFS_BLOCK_GROUP_SYSTEM:
3900 return "invalid-combination";
3904 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3905 u64 total_bytes, u64 bytes_used,
3907 struct btrfs_space_info **space_info)
3909 struct btrfs_space_info *found;
3914 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3915 BTRFS_BLOCK_GROUP_RAID10))
3920 found = __find_space_info(info, flags);
3922 spin_lock(&found->lock);
3923 found->total_bytes += total_bytes;
3924 found->disk_total += total_bytes * factor;
3925 found->bytes_used += bytes_used;
3926 found->disk_used += bytes_used * factor;
3927 found->bytes_readonly += bytes_readonly;
3928 if (total_bytes > 0)
3930 space_info_add_new_bytes(info, found, total_bytes -
3931 bytes_used - bytes_readonly);
3932 spin_unlock(&found->lock);
3933 *space_info = found;
3936 found = kzalloc(sizeof(*found), GFP_NOFS);
3940 ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL);
3946 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3947 INIT_LIST_HEAD(&found->block_groups[i]);
3948 init_rwsem(&found->groups_sem);
3949 spin_lock_init(&found->lock);
3950 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3951 found->total_bytes = total_bytes;
3952 found->disk_total = total_bytes * factor;
3953 found->bytes_used = bytes_used;
3954 found->disk_used = bytes_used * factor;
3955 found->bytes_pinned = 0;
3956 found->bytes_reserved = 0;
3957 found->bytes_readonly = bytes_readonly;
3958 found->bytes_may_use = 0;
3960 found->max_extent_size = 0;
3961 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3962 found->chunk_alloc = 0;
3964 init_waitqueue_head(&found->wait);
3965 INIT_LIST_HEAD(&found->ro_bgs);
3966 INIT_LIST_HEAD(&found->tickets);
3967 INIT_LIST_HEAD(&found->priority_tickets);
3969 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3970 info->space_info_kobj, "%s",
3971 alloc_name(found->flags));
3977 *space_info = found;
3978 list_add_rcu(&found->list, &info->space_info);
3979 if (flags & BTRFS_BLOCK_GROUP_DATA)
3980 info->data_sinfo = found;
3985 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3987 u64 extra_flags = chunk_to_extended(flags) &
3988 BTRFS_EXTENDED_PROFILE_MASK;
3990 write_seqlock(&fs_info->profiles_lock);
3991 if (flags & BTRFS_BLOCK_GROUP_DATA)
3992 fs_info->avail_data_alloc_bits |= extra_flags;
3993 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3994 fs_info->avail_metadata_alloc_bits |= extra_flags;
3995 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3996 fs_info->avail_system_alloc_bits |= extra_flags;
3997 write_sequnlock(&fs_info->profiles_lock);
4001 * returns target flags in extended format or 0 if restripe for this
4002 * chunk_type is not in progress
4004 * should be called with either volume_mutex or balance_lock held
4006 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
4008 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4014 if (flags & BTRFS_BLOCK_GROUP_DATA &&
4015 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4016 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
4017 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
4018 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4019 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
4020 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
4021 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4022 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
4029 * @flags: available profiles in extended format (see ctree.h)
4031 * Returns reduced profile in chunk format. If profile changing is in
4032 * progress (either running or paused) picks the target profile (if it's
4033 * already available), otherwise falls back to plain reducing.
4035 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
4037 u64 num_devices = root->fs_info->fs_devices->rw_devices;
4043 * see if restripe for this chunk_type is in progress, if so
4044 * try to reduce to the target profile
4046 spin_lock(&root->fs_info->balance_lock);
4047 target = get_restripe_target(root->fs_info, flags);
4049 /* pick target profile only if it's already available */
4050 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
4051 spin_unlock(&root->fs_info->balance_lock);
4052 return extended_to_chunk(target);
4055 spin_unlock(&root->fs_info->balance_lock);
4057 /* First, mask out the RAID levels which aren't possible */
4058 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4059 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
4060 allowed |= btrfs_raid_group[raid_type];
4064 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
4065 allowed = BTRFS_BLOCK_GROUP_RAID6;
4066 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
4067 allowed = BTRFS_BLOCK_GROUP_RAID5;
4068 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
4069 allowed = BTRFS_BLOCK_GROUP_RAID10;
4070 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
4071 allowed = BTRFS_BLOCK_GROUP_RAID1;
4072 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
4073 allowed = BTRFS_BLOCK_GROUP_RAID0;
4075 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
4077 return extended_to_chunk(flags | allowed);
4080 static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags)
4087 seq = read_seqbegin(&root->fs_info->profiles_lock);
4089 if (flags & BTRFS_BLOCK_GROUP_DATA)
4090 flags |= root->fs_info->avail_data_alloc_bits;
4091 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4092 flags |= root->fs_info->avail_system_alloc_bits;
4093 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
4094 flags |= root->fs_info->avail_metadata_alloc_bits;
4095 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
4097 return btrfs_reduce_alloc_profile(root, flags);
4100 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
4106 flags = BTRFS_BLOCK_GROUP_DATA;
4107 else if (root == root->fs_info->chunk_root)
4108 flags = BTRFS_BLOCK_GROUP_SYSTEM;
4110 flags = BTRFS_BLOCK_GROUP_METADATA;
4112 ret = get_alloc_profile(root, flags);
4116 int btrfs_alloc_data_chunk_ondemand(struct inode *inode, u64 bytes)
4118 struct btrfs_space_info *data_sinfo;
4119 struct btrfs_root *root = BTRFS_I(inode)->root;
4120 struct btrfs_fs_info *fs_info = root->fs_info;
4123 int need_commit = 2;
4124 int have_pinned_space;
4126 /* make sure bytes are sectorsize aligned */
4127 bytes = ALIGN(bytes, root->sectorsize);
4129 if (btrfs_is_free_space_inode(inode)) {
4131 ASSERT(current->journal_info);
4134 data_sinfo = fs_info->data_sinfo;
4139 /* make sure we have enough space to handle the data first */
4140 spin_lock(&data_sinfo->lock);
4141 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
4142 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
4143 data_sinfo->bytes_may_use;
4145 if (used + bytes > data_sinfo->total_bytes) {
4146 struct btrfs_trans_handle *trans;
4149 * if we don't have enough free bytes in this space then we need
4150 * to alloc a new chunk.
4152 if (!data_sinfo->full) {
4155 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4156 spin_unlock(&data_sinfo->lock);
4158 alloc_target = btrfs_get_alloc_profile(root, 1);
4160 * It is ugly that we don't call nolock join
4161 * transaction for the free space inode case here.
4162 * But it is safe because we only do the data space
4163 * reservation for the free space cache in the
4164 * transaction context, the common join transaction
4165 * just increase the counter of the current transaction
4166 * handler, doesn't try to acquire the trans_lock of
4169 trans = btrfs_join_transaction(root);
4171 return PTR_ERR(trans);
4173 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4175 CHUNK_ALLOC_NO_FORCE);
4176 btrfs_end_transaction(trans, root);
4181 have_pinned_space = 1;
4187 data_sinfo = fs_info->data_sinfo;
4193 * If we don't have enough pinned space to deal with this
4194 * allocation, and no removed chunk in current transaction,
4195 * don't bother committing the transaction.
4197 have_pinned_space = percpu_counter_compare(
4198 &data_sinfo->total_bytes_pinned,
4199 used + bytes - data_sinfo->total_bytes);
4200 spin_unlock(&data_sinfo->lock);
4202 /* commit the current transaction and try again */
4205 !atomic_read(&root->fs_info->open_ioctl_trans)) {
4208 if (need_commit > 0) {
4209 btrfs_start_delalloc_roots(fs_info, 0, -1);
4210 btrfs_wait_ordered_roots(fs_info, -1, 0, (u64)-1);
4213 trans = btrfs_join_transaction(root);
4215 return PTR_ERR(trans);
4216 if (have_pinned_space >= 0 ||
4217 test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4218 &trans->transaction->flags) ||
4220 ret = btrfs_commit_transaction(trans, root);
4224 * The cleaner kthread might still be doing iput
4225 * operations. Wait for it to finish so that
4226 * more space is released.
4228 mutex_lock(&root->fs_info->cleaner_delayed_iput_mutex);
4229 mutex_unlock(&root->fs_info->cleaner_delayed_iput_mutex);
4232 btrfs_end_transaction(trans, root);
4236 trace_btrfs_space_reservation(root->fs_info,
4237 "space_info:enospc",
4238 data_sinfo->flags, bytes, 1);
4241 data_sinfo->bytes_may_use += bytes;
4242 trace_btrfs_space_reservation(root->fs_info, "space_info",
4243 data_sinfo->flags, bytes, 1);
4244 spin_unlock(&data_sinfo->lock);
4250 * New check_data_free_space() with ability for precious data reservation
4251 * Will replace old btrfs_check_data_free_space(), but for patch split,
4252 * add a new function first and then replace it.
4254 int btrfs_check_data_free_space(struct inode *inode, u64 start, u64 len)
4256 struct btrfs_root *root = BTRFS_I(inode)->root;
4259 /* align the range */
4260 len = round_up(start + len, root->sectorsize) -
4261 round_down(start, root->sectorsize);
4262 start = round_down(start, root->sectorsize);
4264 ret = btrfs_alloc_data_chunk_ondemand(inode, len);
4268 /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */
4269 ret = btrfs_qgroup_reserve_data(inode, start, len);
4271 btrfs_free_reserved_data_space_noquota(inode, start, len);
4276 * Called if we need to clear a data reservation for this inode
4277 * Normally in a error case.
4279 * This one will *NOT* use accurate qgroup reserved space API, just for case
4280 * which we can't sleep and is sure it won't affect qgroup reserved space.
4281 * Like clear_bit_hook().
4283 void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4286 struct btrfs_root *root = BTRFS_I(inode)->root;
4287 struct btrfs_space_info *data_sinfo;
4289 /* Make sure the range is aligned to sectorsize */
4290 len = round_up(start + len, root->sectorsize) -
4291 round_down(start, root->sectorsize);
4292 start = round_down(start, root->sectorsize);
4294 data_sinfo = root->fs_info->data_sinfo;
4295 spin_lock(&data_sinfo->lock);
4296 if (WARN_ON(data_sinfo->bytes_may_use < len))
4297 data_sinfo->bytes_may_use = 0;
4299 data_sinfo->bytes_may_use -= len;
4300 trace_btrfs_space_reservation(root->fs_info, "space_info",
4301 data_sinfo->flags, len, 0);
4302 spin_unlock(&data_sinfo->lock);
4306 * Called if we need to clear a data reservation for this inode
4307 * Normally in a error case.
4309 * This one will handle the per-inode data rsv map for accurate reserved
4312 void btrfs_free_reserved_data_space(struct inode *inode, u64 start, u64 len)
4314 btrfs_free_reserved_data_space_noquota(inode, start, len);
4315 btrfs_qgroup_free_data(inode, start, len);
4318 static void force_metadata_allocation(struct btrfs_fs_info *info)
4320 struct list_head *head = &info->space_info;
4321 struct btrfs_space_info *found;
4324 list_for_each_entry_rcu(found, head, list) {
4325 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4326 found->force_alloc = CHUNK_ALLOC_FORCE;
4331 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4333 return (global->size << 1);
4336 static int should_alloc_chunk(struct btrfs_root *root,
4337 struct btrfs_space_info *sinfo, int force)
4339 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4340 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
4341 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
4344 if (force == CHUNK_ALLOC_FORCE)
4348 * We need to take into account the global rsv because for all intents
4349 * and purposes it's used space. Don't worry about locking the
4350 * global_rsv, it doesn't change except when the transaction commits.
4352 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4353 num_allocated += calc_global_rsv_need_space(global_rsv);
4356 * in limited mode, we want to have some free space up to
4357 * about 1% of the FS size.
4359 if (force == CHUNK_ALLOC_LIMITED) {
4360 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
4361 thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1));
4363 if (num_bytes - num_allocated < thresh)
4367 if (num_allocated + SZ_2M < div_factor(num_bytes, 8))
4372 static u64 get_profile_num_devs(struct btrfs_root *root, u64 type)
4376 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4377 BTRFS_BLOCK_GROUP_RAID0 |
4378 BTRFS_BLOCK_GROUP_RAID5 |
4379 BTRFS_BLOCK_GROUP_RAID6))
4380 num_dev = root->fs_info->fs_devices->rw_devices;
4381 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4384 num_dev = 1; /* DUP or single */
4390 * If @is_allocation is true, reserve space in the system space info necessary
4391 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4394 void check_system_chunk(struct btrfs_trans_handle *trans,
4395 struct btrfs_root *root,
4398 struct btrfs_space_info *info;
4405 * Needed because we can end up allocating a system chunk and for an
4406 * atomic and race free space reservation in the chunk block reserve.
4408 ASSERT(mutex_is_locked(&root->fs_info->chunk_mutex));
4410 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4411 spin_lock(&info->lock);
4412 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
4413 info->bytes_reserved - info->bytes_readonly -
4414 info->bytes_may_use;
4415 spin_unlock(&info->lock);
4417 num_devs = get_profile_num_devs(root, type);
4419 /* num_devs device items to update and 1 chunk item to add or remove */
4420 thresh = btrfs_calc_trunc_metadata_size(root, num_devs) +
4421 btrfs_calc_trans_metadata_size(root, 1);
4423 if (left < thresh && btrfs_test_opt(root->fs_info, ENOSPC_DEBUG)) {
4424 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
4425 left, thresh, type);
4426 dump_space_info(info, 0, 0);
4429 if (left < thresh) {
4432 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
4434 * Ignore failure to create system chunk. We might end up not
4435 * needing it, as we might not need to COW all nodes/leafs from
4436 * the paths we visit in the chunk tree (they were already COWed
4437 * or created in the current transaction for example).
4439 ret = btrfs_alloc_chunk(trans, root, flags);
4443 ret = btrfs_block_rsv_add(root->fs_info->chunk_root,
4444 &root->fs_info->chunk_block_rsv,
4445 thresh, BTRFS_RESERVE_NO_FLUSH);
4447 trans->chunk_bytes_reserved += thresh;
4452 * If force is CHUNK_ALLOC_FORCE:
4453 * - return 1 if it successfully allocates a chunk,
4454 * - return errors including -ENOSPC otherwise.
4455 * If force is NOT CHUNK_ALLOC_FORCE:
4456 * - return 0 if it doesn't need to allocate a new chunk,
4457 * - return 1 if it successfully allocates a chunk,
4458 * - return errors including -ENOSPC otherwise.
4460 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4461 struct btrfs_root *extent_root, u64 flags, int force)
4463 struct btrfs_space_info *space_info;
4464 struct btrfs_fs_info *fs_info = extent_root->fs_info;
4465 int wait_for_alloc = 0;
4468 /* Don't re-enter if we're already allocating a chunk */
4469 if (trans->allocating_chunk)
4472 space_info = __find_space_info(extent_root->fs_info, flags);
4474 ret = update_space_info(extent_root->fs_info, flags,
4475 0, 0, 0, &space_info);
4476 BUG_ON(ret); /* -ENOMEM */
4478 BUG_ON(!space_info); /* Logic error */
4481 spin_lock(&space_info->lock);
4482 if (force < space_info->force_alloc)
4483 force = space_info->force_alloc;
4484 if (space_info->full) {
4485 if (should_alloc_chunk(extent_root, space_info, force))
4489 spin_unlock(&space_info->lock);
4493 if (!should_alloc_chunk(extent_root, space_info, force)) {
4494 spin_unlock(&space_info->lock);
4496 } else if (space_info->chunk_alloc) {
4499 space_info->chunk_alloc = 1;
4502 spin_unlock(&space_info->lock);
4504 mutex_lock(&fs_info->chunk_mutex);
4507 * The chunk_mutex is held throughout the entirety of a chunk
4508 * allocation, so once we've acquired the chunk_mutex we know that the
4509 * other guy is done and we need to recheck and see if we should
4512 if (wait_for_alloc) {
4513 mutex_unlock(&fs_info->chunk_mutex);
4518 trans->allocating_chunk = true;
4521 * If we have mixed data/metadata chunks we want to make sure we keep
4522 * allocating mixed chunks instead of individual chunks.
4524 if (btrfs_mixed_space_info(space_info))
4525 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4528 * if we're doing a data chunk, go ahead and make sure that
4529 * we keep a reasonable number of metadata chunks allocated in the
4532 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4533 fs_info->data_chunk_allocations++;
4534 if (!(fs_info->data_chunk_allocations %
4535 fs_info->metadata_ratio))
4536 force_metadata_allocation(fs_info);
4540 * Check if we have enough space in SYSTEM chunk because we may need
4541 * to update devices.
4543 check_system_chunk(trans, extent_root, flags);
4545 ret = btrfs_alloc_chunk(trans, extent_root, flags);
4546 trans->allocating_chunk = false;
4548 spin_lock(&space_info->lock);
4549 if (ret < 0 && ret != -ENOSPC)
4552 space_info->full = 1;
4556 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4558 space_info->chunk_alloc = 0;
4559 spin_unlock(&space_info->lock);
4560 mutex_unlock(&fs_info->chunk_mutex);
4562 * When we allocate a new chunk we reserve space in the chunk block
4563 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4564 * add new nodes/leafs to it if we end up needing to do it when
4565 * inserting the chunk item and updating device items as part of the
4566 * second phase of chunk allocation, performed by
4567 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4568 * large number of new block groups to create in our transaction
4569 * handle's new_bgs list to avoid exhausting the chunk block reserve
4570 * in extreme cases - like having a single transaction create many new
4571 * block groups when starting to write out the free space caches of all
4572 * the block groups that were made dirty during the lifetime of the
4575 if (trans->can_flush_pending_bgs &&
4576 trans->chunk_bytes_reserved >= (u64)SZ_2M) {
4577 btrfs_create_pending_block_groups(trans, extent_root);
4578 btrfs_trans_release_chunk_metadata(trans);
4583 static int can_overcommit(struct btrfs_root *root,
4584 struct btrfs_space_info *space_info, u64 bytes,
4585 enum btrfs_reserve_flush_enum flush)
4587 struct btrfs_block_rsv *global_rsv;
4593 /* Don't overcommit when in mixed mode. */
4594 if (space_info->flags & BTRFS_BLOCK_GROUP_DATA)
4597 BUG_ON(root->fs_info == NULL);
4598 global_rsv = &root->fs_info->global_block_rsv;
4599 profile = btrfs_get_alloc_profile(root, 0);
4600 used = space_info->bytes_used + space_info->bytes_reserved +
4601 space_info->bytes_pinned + space_info->bytes_readonly;
4604 * We only want to allow over committing if we have lots of actual space
4605 * free, but if we don't have enough space to handle the global reserve
4606 * space then we could end up having a real enospc problem when trying
4607 * to allocate a chunk or some other such important allocation.
4609 spin_lock(&global_rsv->lock);
4610 space_size = calc_global_rsv_need_space(global_rsv);
4611 spin_unlock(&global_rsv->lock);
4612 if (used + space_size >= space_info->total_bytes)
4615 used += space_info->bytes_may_use;
4617 spin_lock(&root->fs_info->free_chunk_lock);
4618 avail = root->fs_info->free_chunk_space;
4619 spin_unlock(&root->fs_info->free_chunk_lock);
4622 * If we have dup, raid1 or raid10 then only half of the free
4623 * space is actually useable. For raid56, the space info used
4624 * doesn't include the parity drive, so we don't have to
4627 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4628 BTRFS_BLOCK_GROUP_RAID1 |
4629 BTRFS_BLOCK_GROUP_RAID10))
4633 * If we aren't flushing all things, let us overcommit up to
4634 * 1/2th of the space. If we can flush, don't let us overcommit
4635 * too much, let it overcommit up to 1/8 of the space.
4637 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4642 if (used + bytes < space_info->total_bytes + avail)
4647 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4648 unsigned long nr_pages, int nr_items)
4650 struct super_block *sb = root->fs_info->sb;
4652 if (down_read_trylock(&sb->s_umount)) {
4653 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4654 up_read(&sb->s_umount);
4657 * We needn't worry the filesystem going from r/w to r/o though
4658 * we don't acquire ->s_umount mutex, because the filesystem
4659 * should guarantee the delalloc inodes list be empty after
4660 * the filesystem is readonly(all dirty pages are written to
4663 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
4664 if (!current->journal_info)
4665 btrfs_wait_ordered_roots(root->fs_info, nr_items,
4670 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4675 bytes = btrfs_calc_trans_metadata_size(root, 1);
4676 nr = (int)div64_u64(to_reclaim, bytes);
4682 #define EXTENT_SIZE_PER_ITEM SZ_256K
4685 * shrink metadata reservation for delalloc
4687 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4690 struct btrfs_block_rsv *block_rsv;
4691 struct btrfs_space_info *space_info;
4692 struct btrfs_trans_handle *trans;
4696 unsigned long nr_pages;
4699 enum btrfs_reserve_flush_enum flush;
4701 /* Calc the number of the pages we need flush for space reservation */
4702 items = calc_reclaim_items_nr(root, to_reclaim);
4703 to_reclaim = (u64)items * EXTENT_SIZE_PER_ITEM;
4705 trans = (struct btrfs_trans_handle *)current->journal_info;
4706 block_rsv = &root->fs_info->delalloc_block_rsv;
4707 space_info = block_rsv->space_info;
4709 delalloc_bytes = percpu_counter_sum_positive(
4710 &root->fs_info->delalloc_bytes);
4711 if (delalloc_bytes == 0) {
4715 btrfs_wait_ordered_roots(root->fs_info, items,
4721 while (delalloc_bytes && loops < 3) {
4722 max_reclaim = min(delalloc_bytes, to_reclaim);
4723 nr_pages = max_reclaim >> PAGE_SHIFT;
4724 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4726 * We need to wait for the async pages to actually start before
4729 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4733 if (max_reclaim <= nr_pages)
4736 max_reclaim -= nr_pages;
4738 wait_event(root->fs_info->async_submit_wait,
4739 atomic_read(&root->fs_info->async_delalloc_pages) <=
4743 flush = BTRFS_RESERVE_FLUSH_ALL;
4745 flush = BTRFS_RESERVE_NO_FLUSH;
4746 spin_lock(&space_info->lock);
4747 if (can_overcommit(root, space_info, orig, flush)) {
4748 spin_unlock(&space_info->lock);
4751 if (list_empty(&space_info->tickets) &&
4752 list_empty(&space_info->priority_tickets)) {
4753 spin_unlock(&space_info->lock);
4756 spin_unlock(&space_info->lock);
4759 if (wait_ordered && !trans) {
4760 btrfs_wait_ordered_roots(root->fs_info, items,
4763 time_left = schedule_timeout_killable(1);
4767 delalloc_bytes = percpu_counter_sum_positive(
4768 &root->fs_info->delalloc_bytes);
4773 * maybe_commit_transaction - possibly commit the transaction if its ok to
4774 * @root - the root we're allocating for
4775 * @bytes - the number of bytes we want to reserve
4776 * @force - force the commit
4778 * This will check to make sure that committing the transaction will actually
4779 * get us somewhere and then commit the transaction if it does. Otherwise it
4780 * will return -ENOSPC.
4782 static int may_commit_transaction(struct btrfs_root *root,
4783 struct btrfs_space_info *space_info,
4784 u64 bytes, int force)
4786 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4787 struct btrfs_trans_handle *trans;
4789 trans = (struct btrfs_trans_handle *)current->journal_info;
4796 /* See if there is enough pinned space to make this reservation */
4797 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4802 * See if there is some space in the delayed insertion reservation for
4805 if (space_info != delayed_rsv->space_info)
4808 spin_lock(&delayed_rsv->lock);
4809 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4810 bytes - delayed_rsv->size) >= 0) {
4811 spin_unlock(&delayed_rsv->lock);
4814 spin_unlock(&delayed_rsv->lock);
4817 trans = btrfs_join_transaction(root);
4821 return btrfs_commit_transaction(trans, root);
4824 struct reserve_ticket {
4827 struct list_head list;
4828 wait_queue_head_t wait;
4831 static int flush_space(struct btrfs_root *root,
4832 struct btrfs_space_info *space_info, u64 num_bytes,
4833 u64 orig_bytes, int state)
4835 struct btrfs_trans_handle *trans;
4840 case FLUSH_DELAYED_ITEMS_NR:
4841 case FLUSH_DELAYED_ITEMS:
4842 if (state == FLUSH_DELAYED_ITEMS_NR)
4843 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4847 trans = btrfs_join_transaction(root);
4848 if (IS_ERR(trans)) {
4849 ret = PTR_ERR(trans);
4852 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4853 btrfs_end_transaction(trans, root);
4855 case FLUSH_DELALLOC:
4856 case FLUSH_DELALLOC_WAIT:
4857 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4858 state == FLUSH_DELALLOC_WAIT);
4861 trans = btrfs_join_transaction(root);
4862 if (IS_ERR(trans)) {
4863 ret = PTR_ERR(trans);
4866 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4867 btrfs_get_alloc_profile(root, 0),
4868 CHUNK_ALLOC_NO_FORCE);
4869 btrfs_end_transaction(trans, root);
4870 if (ret > 0 || ret == -ENOSPC)
4874 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4881 trace_btrfs_flush_space(root->fs_info, space_info->flags, num_bytes,
4882 orig_bytes, state, ret);
4887 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4888 struct btrfs_space_info *space_info)
4890 struct reserve_ticket *ticket;
4895 list_for_each_entry(ticket, &space_info->tickets, list)
4896 to_reclaim += ticket->bytes;
4897 list_for_each_entry(ticket, &space_info->priority_tickets, list)
4898 to_reclaim += ticket->bytes;
4902 to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M);
4903 if (can_overcommit(root, space_info, to_reclaim,
4904 BTRFS_RESERVE_FLUSH_ALL))
4907 used = space_info->bytes_used + space_info->bytes_reserved +
4908 space_info->bytes_pinned + space_info->bytes_readonly +
4909 space_info->bytes_may_use;
4910 if (can_overcommit(root, space_info, SZ_1M, BTRFS_RESERVE_FLUSH_ALL))
4911 expected = div_factor_fine(space_info->total_bytes, 95);
4913 expected = div_factor_fine(space_info->total_bytes, 90);
4915 if (used > expected)
4916 to_reclaim = used - expected;
4919 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4920 space_info->bytes_reserved);
4924 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4925 struct btrfs_root *root, u64 used)
4927 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
4929 /* If we're just plain full then async reclaim just slows us down. */
4930 if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh)
4933 if (!btrfs_calc_reclaim_metadata_size(root, space_info))
4936 return (used >= thresh && !btrfs_fs_closing(root->fs_info) &&
4937 !test_bit(BTRFS_FS_STATE_REMOUNTING,
4938 &root->fs_info->fs_state));
4941 static void wake_all_tickets(struct list_head *head)
4943 struct reserve_ticket *ticket;
4945 while (!list_empty(head)) {
4946 ticket = list_first_entry(head, struct reserve_ticket, list);
4947 list_del_init(&ticket->list);
4948 ticket->error = -ENOSPC;
4949 wake_up(&ticket->wait);
4954 * This is for normal flushers, we can wait all goddamned day if we want to. We
4955 * will loop and continuously try to flush as long as we are making progress.
4956 * We count progress as clearing off tickets each time we have to loop.
4958 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4960 struct btrfs_fs_info *fs_info;
4961 struct btrfs_space_info *space_info;
4964 int commit_cycles = 0;
4965 u64 last_tickets_id;
4967 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4968 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4970 spin_lock(&space_info->lock);
4971 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4974 space_info->flush = 0;
4975 spin_unlock(&space_info->lock);
4978 last_tickets_id = space_info->tickets_id;
4979 spin_unlock(&space_info->lock);
4981 flush_state = FLUSH_DELAYED_ITEMS_NR;
4983 struct reserve_ticket *ticket;
4986 ret = flush_space(fs_info->fs_root, space_info, to_reclaim,
4987 to_reclaim, flush_state);
4988 spin_lock(&space_info->lock);
4989 if (list_empty(&space_info->tickets)) {
4990 space_info->flush = 0;
4991 spin_unlock(&space_info->lock);
4994 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4996 ticket = list_first_entry(&space_info->tickets,
4997 struct reserve_ticket, list);
4998 if (last_tickets_id == space_info->tickets_id) {
5001 last_tickets_id = space_info->tickets_id;
5002 flush_state = FLUSH_DELAYED_ITEMS_NR;
5007 if (flush_state > COMMIT_TRANS) {
5009 if (commit_cycles > 2) {
5010 wake_all_tickets(&space_info->tickets);
5011 space_info->flush = 0;
5013 flush_state = FLUSH_DELAYED_ITEMS_NR;
5016 spin_unlock(&space_info->lock);
5017 } while (flush_state <= COMMIT_TRANS);
5020 void btrfs_init_async_reclaim_work(struct work_struct *work)
5022 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
5025 static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
5026 struct btrfs_space_info *space_info,
5027 struct reserve_ticket *ticket)
5030 int flush_state = FLUSH_DELAYED_ITEMS_NR;
5032 spin_lock(&space_info->lock);
5033 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
5036 spin_unlock(&space_info->lock);
5039 spin_unlock(&space_info->lock);
5042 flush_space(fs_info->fs_root, space_info, to_reclaim,
5043 to_reclaim, flush_state);
5045 spin_lock(&space_info->lock);
5046 if (ticket->bytes == 0) {
5047 spin_unlock(&space_info->lock);
5050 spin_unlock(&space_info->lock);
5053 * Priority flushers can't wait on delalloc without
5056 if (flush_state == FLUSH_DELALLOC ||
5057 flush_state == FLUSH_DELALLOC_WAIT)
5058 flush_state = ALLOC_CHUNK;
5059 } while (flush_state < COMMIT_TRANS);
5062 static int wait_reserve_ticket(struct btrfs_fs_info *fs_info,
5063 struct btrfs_space_info *space_info,
5064 struct reserve_ticket *ticket, u64 orig_bytes)
5070 spin_lock(&space_info->lock);
5071 while (ticket->bytes > 0 && ticket->error == 0) {
5072 ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE);
5077 spin_unlock(&space_info->lock);
5081 finish_wait(&ticket->wait, &wait);
5082 spin_lock(&space_info->lock);
5085 ret = ticket->error;
5086 if (!list_empty(&ticket->list))
5087 list_del_init(&ticket->list);
5088 if (ticket->bytes && ticket->bytes < orig_bytes) {
5089 u64 num_bytes = orig_bytes - ticket->bytes;
5090 space_info->bytes_may_use -= num_bytes;
5091 trace_btrfs_space_reservation(fs_info, "space_info",
5092 space_info->flags, num_bytes, 0);
5094 spin_unlock(&space_info->lock);
5100 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5101 * @root - the root we're allocating for
5102 * @space_info - the space info we want to allocate from
5103 * @orig_bytes - the number of bytes we want
5104 * @flush - whether or not we can flush to make our reservation
5106 * This will reserve orig_bytes number of bytes from the space info associated
5107 * with the block_rsv. If there is not enough space it will make an attempt to
5108 * flush out space to make room. It will do this by flushing delalloc if
5109 * possible or committing the transaction. If flush is 0 then no attempts to
5110 * regain reservations will be made and this will fail if there is not enough
5113 static int __reserve_metadata_bytes(struct btrfs_root *root,
5114 struct btrfs_space_info *space_info,
5116 enum btrfs_reserve_flush_enum flush)
5118 struct reserve_ticket ticket;
5123 ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL);
5125 spin_lock(&space_info->lock);
5127 used = space_info->bytes_used + space_info->bytes_reserved +
5128 space_info->bytes_pinned + space_info->bytes_readonly +
5129 space_info->bytes_may_use;
5132 * If we have enough space then hooray, make our reservation and carry
5133 * on. If not see if we can overcommit, and if we can, hooray carry on.
5134 * If not things get more complicated.
5136 if (used + orig_bytes <= space_info->total_bytes) {
5137 space_info->bytes_may_use += orig_bytes;
5138 trace_btrfs_space_reservation(root->fs_info, "space_info",
5139 space_info->flags, orig_bytes,
5142 } else if (can_overcommit(root, space_info, orig_bytes, flush)) {
5143 space_info->bytes_may_use += orig_bytes;
5144 trace_btrfs_space_reservation(root->fs_info, "space_info",
5145 space_info->flags, orig_bytes,
5151 * If we couldn't make a reservation then setup our reservation ticket
5152 * and kick the async worker if it's not already running.
5154 * If we are a priority flusher then we just need to add our ticket to
5155 * the list and we will do our own flushing further down.
5157 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
5158 ticket.bytes = orig_bytes;
5160 init_waitqueue_head(&ticket.wait);
5161 if (flush == BTRFS_RESERVE_FLUSH_ALL) {
5162 list_add_tail(&ticket.list, &space_info->tickets);
5163 if (!space_info->flush) {
5164 space_info->flush = 1;
5165 trace_btrfs_trigger_flush(root->fs_info,
5169 queue_work(system_unbound_wq,
5170 &root->fs_info->async_reclaim_work);
5173 list_add_tail(&ticket.list,
5174 &space_info->priority_tickets);
5176 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
5179 * We will do the space reservation dance during log replay,
5180 * which means we won't have fs_info->fs_root set, so don't do
5181 * the async reclaim as we will panic.
5183 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags) &&
5184 need_do_async_reclaim(space_info, root, used) &&
5185 !work_busy(&root->fs_info->async_reclaim_work)) {
5186 trace_btrfs_trigger_flush(root->fs_info,
5190 queue_work(system_unbound_wq,
5191 &root->fs_info->async_reclaim_work);
5194 spin_unlock(&space_info->lock);
5195 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
5198 if (flush == BTRFS_RESERVE_FLUSH_ALL)
5199 return wait_reserve_ticket(root->fs_info, space_info, &ticket,
5203 priority_reclaim_metadata_space(root->fs_info, space_info, &ticket);
5204 spin_lock(&space_info->lock);
5206 if (ticket.bytes < orig_bytes) {
5207 u64 num_bytes = orig_bytes - ticket.bytes;
5208 space_info->bytes_may_use -= num_bytes;
5209 trace_btrfs_space_reservation(root->fs_info,
5210 "space_info", space_info->flags,
5214 list_del_init(&ticket.list);
5217 spin_unlock(&space_info->lock);
5218 ASSERT(list_empty(&ticket.list));
5223 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5224 * @root - the root we're allocating for
5225 * @block_rsv - the block_rsv we're allocating for
5226 * @orig_bytes - the number of bytes we want
5227 * @flush - whether or not we can flush to make our reservation
5229 * This will reserve orgi_bytes number of bytes from the space info associated
5230 * with the block_rsv. If there is not enough space it will make an attempt to
5231 * flush out space to make room. It will do this by flushing delalloc if
5232 * possible or committing the transaction. If flush is 0 then no attempts to
5233 * regain reservations will be made and this will fail if there is not enough
5236 static int reserve_metadata_bytes(struct btrfs_root *root,
5237 struct btrfs_block_rsv *block_rsv,
5239 enum btrfs_reserve_flush_enum flush)
5243 ret = __reserve_metadata_bytes(root, block_rsv->space_info, orig_bytes,
5245 if (ret == -ENOSPC &&
5246 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5247 struct btrfs_block_rsv *global_rsv =
5248 &root->fs_info->global_block_rsv;
5250 if (block_rsv != global_rsv &&
5251 !block_rsv_use_bytes(global_rsv, orig_bytes))
5255 trace_btrfs_space_reservation(root->fs_info,
5256 "space_info:enospc",
5257 block_rsv->space_info->flags,
5262 static struct btrfs_block_rsv *get_block_rsv(
5263 const struct btrfs_trans_handle *trans,
5264 const struct btrfs_root *root)
5266 struct btrfs_block_rsv *block_rsv = NULL;
5268 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5269 (root == root->fs_info->csum_root && trans->adding_csums) ||
5270 (root == root->fs_info->uuid_root))
5271 block_rsv = trans->block_rsv;
5274 block_rsv = root->block_rsv;
5277 block_rsv = &root->fs_info->empty_block_rsv;
5282 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5286 spin_lock(&block_rsv->lock);
5287 if (block_rsv->reserved >= num_bytes) {
5288 block_rsv->reserved -= num_bytes;
5289 if (block_rsv->reserved < block_rsv->size)
5290 block_rsv->full = 0;
5293 spin_unlock(&block_rsv->lock);
5297 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5298 u64 num_bytes, int update_size)
5300 spin_lock(&block_rsv->lock);
5301 block_rsv->reserved += num_bytes;
5303 block_rsv->size += num_bytes;
5304 else if (block_rsv->reserved >= block_rsv->size)
5305 block_rsv->full = 1;
5306 spin_unlock(&block_rsv->lock);
5309 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5310 struct btrfs_block_rsv *dest, u64 num_bytes,
5313 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5316 if (global_rsv->space_info != dest->space_info)
5319 spin_lock(&global_rsv->lock);
5320 min_bytes = div_factor(global_rsv->size, min_factor);
5321 if (global_rsv->reserved < min_bytes + num_bytes) {
5322 spin_unlock(&global_rsv->lock);
5325 global_rsv->reserved -= num_bytes;
5326 if (global_rsv->reserved < global_rsv->size)
5327 global_rsv->full = 0;
5328 spin_unlock(&global_rsv->lock);
5330 block_rsv_add_bytes(dest, num_bytes, 1);
5335 * This is for space we already have accounted in space_info->bytes_may_use, so
5336 * basically when we're returning space from block_rsv's.
5338 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
5339 struct btrfs_space_info *space_info,
5342 struct reserve_ticket *ticket;
5343 struct list_head *head;
5345 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH;
5346 bool check_overcommit = false;
5348 spin_lock(&space_info->lock);
5349 head = &space_info->priority_tickets;
5352 * If we are over our limit then we need to check and see if we can
5353 * overcommit, and if we can't then we just need to free up our space
5354 * and not satisfy any requests.
5356 used = space_info->bytes_used + space_info->bytes_reserved +
5357 space_info->bytes_pinned + space_info->bytes_readonly +
5358 space_info->bytes_may_use;
5359 if (used - num_bytes >= space_info->total_bytes)
5360 check_overcommit = true;
5362 while (!list_empty(head) && num_bytes) {
5363 ticket = list_first_entry(head, struct reserve_ticket,
5366 * We use 0 bytes because this space is already reserved, so
5367 * adding the ticket space would be a double count.
5369 if (check_overcommit &&
5370 !can_overcommit(fs_info->extent_root, space_info, 0,
5373 if (num_bytes >= ticket->bytes) {
5374 list_del_init(&ticket->list);
5375 num_bytes -= ticket->bytes;
5377 space_info->tickets_id++;
5378 wake_up(&ticket->wait);
5380 ticket->bytes -= num_bytes;
5385 if (num_bytes && head == &space_info->priority_tickets) {
5386 head = &space_info->tickets;
5387 flush = BTRFS_RESERVE_FLUSH_ALL;
5390 space_info->bytes_may_use -= num_bytes;
5391 trace_btrfs_space_reservation(fs_info, "space_info",
5392 space_info->flags, num_bytes, 0);
5393 spin_unlock(&space_info->lock);
5397 * This is for newly allocated space that isn't accounted in
5398 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5399 * we use this helper.
5401 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
5402 struct btrfs_space_info *space_info,
5405 struct reserve_ticket *ticket;
5406 struct list_head *head = &space_info->priority_tickets;
5409 while (!list_empty(head) && num_bytes) {
5410 ticket = list_first_entry(head, struct reserve_ticket,
5412 if (num_bytes >= ticket->bytes) {
5413 trace_btrfs_space_reservation(fs_info, "space_info",
5416 list_del_init(&ticket->list);
5417 num_bytes -= ticket->bytes;
5418 space_info->bytes_may_use += ticket->bytes;
5420 space_info->tickets_id++;
5421 wake_up(&ticket->wait);
5423 trace_btrfs_space_reservation(fs_info, "space_info",
5426 space_info->bytes_may_use += num_bytes;
5427 ticket->bytes -= num_bytes;
5432 if (num_bytes && head == &space_info->priority_tickets) {
5433 head = &space_info->tickets;
5438 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5439 struct btrfs_block_rsv *block_rsv,
5440 struct btrfs_block_rsv *dest, u64 num_bytes)
5442 struct btrfs_space_info *space_info = block_rsv->space_info;
5444 spin_lock(&block_rsv->lock);
5445 if (num_bytes == (u64)-1)
5446 num_bytes = block_rsv->size;
5447 block_rsv->size -= num_bytes;
5448 if (block_rsv->reserved >= block_rsv->size) {
5449 num_bytes = block_rsv->reserved - block_rsv->size;
5450 block_rsv->reserved = block_rsv->size;
5451 block_rsv->full = 1;
5455 spin_unlock(&block_rsv->lock);
5457 if (num_bytes > 0) {
5459 spin_lock(&dest->lock);
5463 bytes_to_add = dest->size - dest->reserved;
5464 bytes_to_add = min(num_bytes, bytes_to_add);
5465 dest->reserved += bytes_to_add;
5466 if (dest->reserved >= dest->size)
5468 num_bytes -= bytes_to_add;
5470 spin_unlock(&dest->lock);
5473 space_info_add_old_bytes(fs_info, space_info,
5478 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
5479 struct btrfs_block_rsv *dst, u64 num_bytes,
5484 ret = block_rsv_use_bytes(src, num_bytes);
5488 block_rsv_add_bytes(dst, num_bytes, update_size);
5492 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5494 memset(rsv, 0, sizeof(*rsv));
5495 spin_lock_init(&rsv->lock);
5499 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
5500 unsigned short type)
5502 struct btrfs_block_rsv *block_rsv;
5503 struct btrfs_fs_info *fs_info = root->fs_info;
5505 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5509 btrfs_init_block_rsv(block_rsv, type);
5510 block_rsv->space_info = __find_space_info(fs_info,
5511 BTRFS_BLOCK_GROUP_METADATA);
5515 void btrfs_free_block_rsv(struct btrfs_root *root,
5516 struct btrfs_block_rsv *rsv)
5520 btrfs_block_rsv_release(root, rsv, (u64)-1);
5524 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
5529 int btrfs_block_rsv_add(struct btrfs_root *root,
5530 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5531 enum btrfs_reserve_flush_enum flush)
5538 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5540 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5547 int btrfs_block_rsv_check(struct btrfs_root *root,
5548 struct btrfs_block_rsv *block_rsv, int min_factor)
5556 spin_lock(&block_rsv->lock);
5557 num_bytes = div_factor(block_rsv->size, min_factor);
5558 if (block_rsv->reserved >= num_bytes)
5560 spin_unlock(&block_rsv->lock);
5565 int btrfs_block_rsv_refill(struct btrfs_root *root,
5566 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5567 enum btrfs_reserve_flush_enum flush)
5575 spin_lock(&block_rsv->lock);
5576 num_bytes = min_reserved;
5577 if (block_rsv->reserved >= num_bytes)
5580 num_bytes -= block_rsv->reserved;
5581 spin_unlock(&block_rsv->lock);
5586 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5588 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5595 void btrfs_block_rsv_release(struct btrfs_root *root,
5596 struct btrfs_block_rsv *block_rsv,
5599 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5600 if (global_rsv == block_rsv ||
5601 block_rsv->space_info != global_rsv->space_info)
5603 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
5607 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5609 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5610 struct btrfs_space_info *sinfo = block_rsv->space_info;
5614 * The global block rsv is based on the size of the extent tree, the
5615 * checksum tree and the root tree. If the fs is empty we want to set
5616 * it to a minimal amount for safety.
5618 num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) +
5619 btrfs_root_used(&fs_info->csum_root->root_item) +
5620 btrfs_root_used(&fs_info->tree_root->root_item);
5621 num_bytes = max_t(u64, num_bytes, SZ_16M);
5623 spin_lock(&sinfo->lock);
5624 spin_lock(&block_rsv->lock);
5626 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
5628 if (block_rsv->reserved < block_rsv->size) {
5629 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
5630 sinfo->bytes_reserved + sinfo->bytes_readonly +
5631 sinfo->bytes_may_use;
5632 if (sinfo->total_bytes > num_bytes) {
5633 num_bytes = sinfo->total_bytes - num_bytes;
5634 num_bytes = min(num_bytes,
5635 block_rsv->size - block_rsv->reserved);
5636 block_rsv->reserved += num_bytes;
5637 sinfo->bytes_may_use += num_bytes;
5638 trace_btrfs_space_reservation(fs_info, "space_info",
5639 sinfo->flags, num_bytes,
5642 } else if (block_rsv->reserved > block_rsv->size) {
5643 num_bytes = block_rsv->reserved - block_rsv->size;
5644 sinfo->bytes_may_use -= num_bytes;
5645 trace_btrfs_space_reservation(fs_info, "space_info",
5646 sinfo->flags, num_bytes, 0);
5647 block_rsv->reserved = block_rsv->size;
5650 if (block_rsv->reserved == block_rsv->size)
5651 block_rsv->full = 1;
5653 block_rsv->full = 0;
5655 spin_unlock(&block_rsv->lock);
5656 spin_unlock(&sinfo->lock);
5659 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5661 struct btrfs_space_info *space_info;
5663 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5664 fs_info->chunk_block_rsv.space_info = space_info;
5666 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5667 fs_info->global_block_rsv.space_info = space_info;
5668 fs_info->delalloc_block_rsv.space_info = space_info;
5669 fs_info->trans_block_rsv.space_info = space_info;
5670 fs_info->empty_block_rsv.space_info = space_info;
5671 fs_info->delayed_block_rsv.space_info = space_info;
5673 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5674 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5675 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5676 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5677 if (fs_info->quota_root)
5678 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5679 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5681 update_global_block_rsv(fs_info);
5684 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5686 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5688 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
5689 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
5690 WARN_ON(fs_info->trans_block_rsv.size > 0);
5691 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5692 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5693 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5694 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5695 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5698 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
5699 struct btrfs_root *root)
5701 if (!trans->block_rsv)
5704 if (!trans->bytes_reserved)
5707 trace_btrfs_space_reservation(root->fs_info, "transaction",
5708 trans->transid, trans->bytes_reserved, 0);
5709 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
5710 trans->bytes_reserved = 0;
5714 * To be called after all the new block groups attached to the transaction
5715 * handle have been created (btrfs_create_pending_block_groups()).
5717 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5719 struct btrfs_fs_info *fs_info = trans->fs_info;
5721 if (!trans->chunk_bytes_reserved)
5724 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5726 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5727 trans->chunk_bytes_reserved);
5728 trans->chunk_bytes_reserved = 0;
5731 /* Can only return 0 or -ENOSPC */
5732 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5733 struct inode *inode)
5735 struct btrfs_root *root = BTRFS_I(inode)->root;
5737 * We always use trans->block_rsv here as we will have reserved space
5738 * for our orphan when starting the transaction, using get_block_rsv()
5739 * here will sometimes make us choose the wrong block rsv as we could be
5740 * doing a reloc inode for a non refcounted root.
5742 struct btrfs_block_rsv *src_rsv = trans->block_rsv;
5743 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5746 * We need to hold space in order to delete our orphan item once we've
5747 * added it, so this takes the reservation so we can release it later
5748 * when we are truly done with the orphan item.
5750 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5751 trace_btrfs_space_reservation(root->fs_info, "orphan",
5752 btrfs_ino(inode), num_bytes, 1);
5753 return btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1);
5756 void btrfs_orphan_release_metadata(struct inode *inode)
5758 struct btrfs_root *root = BTRFS_I(inode)->root;
5759 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5760 trace_btrfs_space_reservation(root->fs_info, "orphan",
5761 btrfs_ino(inode), num_bytes, 0);
5762 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
5766 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5767 * root: the root of the parent directory
5768 * rsv: block reservation
5769 * items: the number of items that we need do reservation
5770 * qgroup_reserved: used to return the reserved size in qgroup
5772 * This function is used to reserve the space for snapshot/subvolume
5773 * creation and deletion. Those operations are different with the
5774 * common file/directory operations, they change two fs/file trees
5775 * and root tree, the number of items that the qgroup reserves is
5776 * different with the free space reservation. So we can not use
5777 * the space reservation mechanism in start_transaction().
5779 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5780 struct btrfs_block_rsv *rsv,
5782 u64 *qgroup_reserved,
5783 bool use_global_rsv)
5787 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5789 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags)) {
5790 /* One for parent inode, two for dir entries */
5791 num_bytes = 3 * root->nodesize;
5792 ret = btrfs_qgroup_reserve_meta(root, num_bytes);
5799 *qgroup_reserved = num_bytes;
5801 num_bytes = btrfs_calc_trans_metadata_size(root, items);
5802 rsv->space_info = __find_space_info(root->fs_info,
5803 BTRFS_BLOCK_GROUP_METADATA);
5804 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5805 BTRFS_RESERVE_FLUSH_ALL);
5807 if (ret == -ENOSPC && use_global_rsv)
5808 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, 1);
5810 if (ret && *qgroup_reserved)
5811 btrfs_qgroup_free_meta(root, *qgroup_reserved);
5816 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
5817 struct btrfs_block_rsv *rsv,
5818 u64 qgroup_reserved)
5820 btrfs_block_rsv_release(root, rsv, (u64)-1);
5824 * drop_outstanding_extent - drop an outstanding extent
5825 * @inode: the inode we're dropping the extent for
5826 * @num_bytes: the number of bytes we're releasing.
5828 * This is called when we are freeing up an outstanding extent, either called
5829 * after an error or after an extent is written. This will return the number of
5830 * reserved extents that need to be freed. This must be called with
5831 * BTRFS_I(inode)->lock held.
5833 static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes)
5835 unsigned drop_inode_space = 0;
5836 unsigned dropped_extents = 0;
5837 unsigned num_extents = 0;
5839 num_extents = (unsigned)div64_u64(num_bytes +
5840 BTRFS_MAX_EXTENT_SIZE - 1,
5841 BTRFS_MAX_EXTENT_SIZE);
5842 ASSERT(num_extents);
5843 ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents);
5844 BTRFS_I(inode)->outstanding_extents -= num_extents;
5846 if (BTRFS_I(inode)->outstanding_extents == 0 &&
5847 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5848 &BTRFS_I(inode)->runtime_flags))
5849 drop_inode_space = 1;
5852 * If we have more or the same amount of outstanding extents than we have
5853 * reserved then we need to leave the reserved extents count alone.
5855 if (BTRFS_I(inode)->outstanding_extents >=
5856 BTRFS_I(inode)->reserved_extents)
5857 return drop_inode_space;
5859 dropped_extents = BTRFS_I(inode)->reserved_extents -
5860 BTRFS_I(inode)->outstanding_extents;
5861 BTRFS_I(inode)->reserved_extents -= dropped_extents;
5862 return dropped_extents + drop_inode_space;
5866 * calc_csum_metadata_size - return the amount of metadata space that must be
5867 * reserved/freed for the given bytes.
5868 * @inode: the inode we're manipulating
5869 * @num_bytes: the number of bytes in question
5870 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5872 * This adjusts the number of csum_bytes in the inode and then returns the
5873 * correct amount of metadata that must either be reserved or freed. We
5874 * calculate how many checksums we can fit into one leaf and then divide the
5875 * number of bytes that will need to be checksumed by this value to figure out
5876 * how many checksums will be required. If we are adding bytes then the number
5877 * may go up and we will return the number of additional bytes that must be
5878 * reserved. If it is going down we will return the number of bytes that must
5881 * This must be called with BTRFS_I(inode)->lock held.
5883 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
5886 struct btrfs_root *root = BTRFS_I(inode)->root;
5887 u64 old_csums, num_csums;
5889 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5890 BTRFS_I(inode)->csum_bytes == 0)
5893 old_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5895 BTRFS_I(inode)->csum_bytes += num_bytes;
5897 BTRFS_I(inode)->csum_bytes -= num_bytes;
5898 num_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5900 /* No change, no need to reserve more */
5901 if (old_csums == num_csums)
5905 return btrfs_calc_trans_metadata_size(root,
5906 num_csums - old_csums);
5908 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5911 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5913 struct btrfs_root *root = BTRFS_I(inode)->root;
5914 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5917 unsigned nr_extents = 0;
5918 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5920 bool delalloc_lock = true;
5923 bool release_extra = false;
5925 /* If we are a free space inode we need to not flush since we will be in
5926 * the middle of a transaction commit. We also don't need the delalloc
5927 * mutex since we won't race with anybody. We need this mostly to make
5928 * lockdep shut its filthy mouth.
5930 * If we have a transaction open (can happen if we call truncate_block
5931 * from truncate), then we need FLUSH_LIMIT so we don't deadlock.
5933 if (btrfs_is_free_space_inode(inode)) {
5934 flush = BTRFS_RESERVE_NO_FLUSH;
5935 delalloc_lock = false;
5936 } else if (current->journal_info) {
5937 flush = BTRFS_RESERVE_FLUSH_LIMIT;
5940 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5941 btrfs_transaction_in_commit(root->fs_info))
5942 schedule_timeout(1);
5945 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5947 num_bytes = ALIGN(num_bytes, root->sectorsize);
5949 spin_lock(&BTRFS_I(inode)->lock);
5950 nr_extents = (unsigned)div64_u64(num_bytes +
5951 BTRFS_MAX_EXTENT_SIZE - 1,
5952 BTRFS_MAX_EXTENT_SIZE);
5953 BTRFS_I(inode)->outstanding_extents += nr_extents;
5956 if (BTRFS_I(inode)->outstanding_extents >
5957 BTRFS_I(inode)->reserved_extents)
5958 nr_extents += BTRFS_I(inode)->outstanding_extents -
5959 BTRFS_I(inode)->reserved_extents;
5961 /* We always want to reserve a slot for updating the inode. */
5962 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents + 1);
5963 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5964 csum_bytes = BTRFS_I(inode)->csum_bytes;
5965 spin_unlock(&BTRFS_I(inode)->lock);
5967 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags)) {
5968 ret = btrfs_qgroup_reserve_meta(root,
5969 nr_extents * root->nodesize);
5974 ret = btrfs_block_rsv_add(root, block_rsv, to_reserve, flush);
5975 if (unlikely(ret)) {
5976 btrfs_qgroup_free_meta(root, nr_extents * root->nodesize);
5980 spin_lock(&BTRFS_I(inode)->lock);
5981 if (test_and_set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5982 &BTRFS_I(inode)->runtime_flags)) {
5983 to_reserve -= btrfs_calc_trans_metadata_size(root, 1);
5984 release_extra = true;
5986 BTRFS_I(inode)->reserved_extents += nr_extents;
5987 spin_unlock(&BTRFS_I(inode)->lock);
5990 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5993 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5994 btrfs_ino(inode), to_reserve, 1);
5996 btrfs_block_rsv_release(root, block_rsv,
5997 btrfs_calc_trans_metadata_size(root,
6002 spin_lock(&BTRFS_I(inode)->lock);
6003 dropped = drop_outstanding_extent(inode, num_bytes);
6005 * If the inodes csum_bytes is the same as the original
6006 * csum_bytes then we know we haven't raced with any free()ers
6007 * so we can just reduce our inodes csum bytes and carry on.
6009 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
6010 calc_csum_metadata_size(inode, num_bytes, 0);
6012 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
6016 * This is tricky, but first we need to figure out how much we
6017 * freed from any free-ers that occurred during this
6018 * reservation, so we reset ->csum_bytes to the csum_bytes
6019 * before we dropped our lock, and then call the free for the
6020 * number of bytes that were freed while we were trying our
6023 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
6024 BTRFS_I(inode)->csum_bytes = csum_bytes;
6025 to_free = calc_csum_metadata_size(inode, bytes, 0);
6029 * Now we need to see how much we would have freed had we not
6030 * been making this reservation and our ->csum_bytes were not
6031 * artificially inflated.
6033 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
6034 bytes = csum_bytes - orig_csum_bytes;
6035 bytes = calc_csum_metadata_size(inode, bytes, 0);
6038 * Now reset ->csum_bytes to what it should be. If bytes is
6039 * more than to_free then we would have freed more space had we
6040 * not had an artificially high ->csum_bytes, so we need to free
6041 * the remainder. If bytes is the same or less then we don't
6042 * need to do anything, the other free-ers did the correct
6045 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
6046 if (bytes > to_free)
6047 to_free = bytes - to_free;
6051 spin_unlock(&BTRFS_I(inode)->lock);
6053 to_free += btrfs_calc_trans_metadata_size(root, dropped);
6056 btrfs_block_rsv_release(root, block_rsv, to_free);
6057 trace_btrfs_space_reservation(root->fs_info, "delalloc",
6058 btrfs_ino(inode), to_free, 0);
6061 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
6066 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
6067 * @inode: the inode to release the reservation for
6068 * @num_bytes: the number of bytes we're releasing
6070 * This will release the metadata reservation for an inode. This can be called
6071 * once we complete IO for a given set of bytes to release their metadata
6074 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
6076 struct btrfs_root *root = BTRFS_I(inode)->root;
6080 num_bytes = ALIGN(num_bytes, root->sectorsize);
6081 spin_lock(&BTRFS_I(inode)->lock);
6082 dropped = drop_outstanding_extent(inode, num_bytes);
6085 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
6086 spin_unlock(&BTRFS_I(inode)->lock);
6088 to_free += btrfs_calc_trans_metadata_size(root, dropped);
6090 if (btrfs_is_testing(root->fs_info))
6093 trace_btrfs_space_reservation(root->fs_info, "delalloc",
6094 btrfs_ino(inode), to_free, 0);
6096 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
6101 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6103 * @inode: inode we're writing to
6104 * @start: start range we are writing to
6105 * @len: how long the range we are writing to
6107 * TODO: This function will finally replace old btrfs_delalloc_reserve_space()
6109 * This will do the following things
6111 * o reserve space in data space info for num bytes
6112 * and reserve precious corresponding qgroup space
6113 * (Done in check_data_free_space)
6115 * o reserve space for metadata space, based on the number of outstanding
6116 * extents and how much csums will be needed
6117 * also reserve metadata space in a per root over-reserve method.
6118 * o add to the inodes->delalloc_bytes
6119 * o add it to the fs_info's delalloc inodes list.
6120 * (Above 3 all done in delalloc_reserve_metadata)
6122 * Return 0 for success
6123 * Return <0 for error(-ENOSPC or -EQUOT)
6125 int btrfs_delalloc_reserve_space(struct inode *inode, u64 start, u64 len)
6129 ret = btrfs_check_data_free_space(inode, start, len);
6132 ret = btrfs_delalloc_reserve_metadata(inode, len);
6134 btrfs_free_reserved_data_space(inode, start, len);
6139 * btrfs_delalloc_release_space - release data and metadata space for delalloc
6140 * @inode: inode we're releasing space for
6141 * @start: start position of the space already reserved
6142 * @len: the len of the space already reserved
6144 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
6145 * called in the case that we don't need the metadata AND data reservations
6146 * anymore. So if there is an error or we insert an inline extent.
6148 * This function will release the metadata space that was not used and will
6149 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
6150 * list if there are no delalloc bytes left.
6151 * Also it will handle the qgroup reserved space.
6153 void btrfs_delalloc_release_space(struct inode *inode, u64 start, u64 len)
6155 btrfs_delalloc_release_metadata(inode, len);
6156 btrfs_free_reserved_data_space(inode, start, len);
6159 static int update_block_group(struct btrfs_trans_handle *trans,
6160 struct btrfs_root *root, u64 bytenr,
6161 u64 num_bytes, int alloc)
6163 struct btrfs_block_group_cache *cache = NULL;
6164 struct btrfs_fs_info *info = root->fs_info;
6165 u64 total = num_bytes;
6170 /* block accounting for super block */
6171 spin_lock(&info->delalloc_root_lock);
6172 old_val = btrfs_super_bytes_used(info->super_copy);
6174 old_val += num_bytes;
6176 old_val -= num_bytes;
6177 btrfs_set_super_bytes_used(info->super_copy, old_val);
6178 spin_unlock(&info->delalloc_root_lock);
6181 cache = btrfs_lookup_block_group(info, bytenr);
6184 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
6185 BTRFS_BLOCK_GROUP_RAID1 |
6186 BTRFS_BLOCK_GROUP_RAID10))
6191 * If this block group has free space cache written out, we
6192 * need to make sure to load it if we are removing space. This
6193 * is because we need the unpinning stage to actually add the
6194 * space back to the block group, otherwise we will leak space.
6196 if (!alloc && cache->cached == BTRFS_CACHE_NO)
6197 cache_block_group(cache, 1);
6199 byte_in_group = bytenr - cache->key.objectid;
6200 WARN_ON(byte_in_group > cache->key.offset);
6202 spin_lock(&cache->space_info->lock);
6203 spin_lock(&cache->lock);
6205 if (btrfs_test_opt(root->fs_info, SPACE_CACHE) &&
6206 cache->disk_cache_state < BTRFS_DC_CLEAR)
6207 cache->disk_cache_state = BTRFS_DC_CLEAR;
6209 old_val = btrfs_block_group_used(&cache->item);
6210 num_bytes = min(total, cache->key.offset - byte_in_group);
6212 old_val += num_bytes;
6213 btrfs_set_block_group_used(&cache->item, old_val);
6214 cache->reserved -= num_bytes;
6215 cache->space_info->bytes_reserved -= num_bytes;
6216 cache->space_info->bytes_used += num_bytes;
6217 cache->space_info->disk_used += num_bytes * factor;
6218 spin_unlock(&cache->lock);
6219 spin_unlock(&cache->space_info->lock);
6221 old_val -= num_bytes;
6222 btrfs_set_block_group_used(&cache->item, old_val);
6223 cache->pinned += num_bytes;
6224 cache->space_info->bytes_pinned += num_bytes;
6225 cache->space_info->bytes_used -= num_bytes;
6226 cache->space_info->disk_used -= num_bytes * factor;
6227 spin_unlock(&cache->lock);
6228 spin_unlock(&cache->space_info->lock);
6230 trace_btrfs_space_reservation(root->fs_info, "pinned",
6231 cache->space_info->flags,
6233 set_extent_dirty(info->pinned_extents,
6234 bytenr, bytenr + num_bytes - 1,
6235 GFP_NOFS | __GFP_NOFAIL);
6238 spin_lock(&trans->transaction->dirty_bgs_lock);
6239 if (list_empty(&cache->dirty_list)) {
6240 list_add_tail(&cache->dirty_list,
6241 &trans->transaction->dirty_bgs);
6242 trans->transaction->num_dirty_bgs++;
6243 btrfs_get_block_group(cache);
6245 spin_unlock(&trans->transaction->dirty_bgs_lock);
6248 * No longer have used bytes in this block group, queue it for
6249 * deletion. We do this after adding the block group to the
6250 * dirty list to avoid races between cleaner kthread and space
6253 if (!alloc && old_val == 0) {
6254 spin_lock(&info->unused_bgs_lock);
6255 if (list_empty(&cache->bg_list)) {
6256 btrfs_get_block_group(cache);
6257 list_add_tail(&cache->bg_list,
6260 spin_unlock(&info->unused_bgs_lock);
6263 btrfs_put_block_group(cache);
6265 bytenr += num_bytes;
6270 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
6272 struct btrfs_block_group_cache *cache;
6275 spin_lock(&root->fs_info->block_group_cache_lock);
6276 bytenr = root->fs_info->first_logical_byte;
6277 spin_unlock(&root->fs_info->block_group_cache_lock);
6279 if (bytenr < (u64)-1)
6282 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
6286 bytenr = cache->key.objectid;
6287 btrfs_put_block_group(cache);
6292 static int pin_down_extent(struct btrfs_root *root,
6293 struct btrfs_block_group_cache *cache,
6294 u64 bytenr, u64 num_bytes, int reserved)
6296 spin_lock(&cache->space_info->lock);
6297 spin_lock(&cache->lock);
6298 cache->pinned += num_bytes;
6299 cache->space_info->bytes_pinned += num_bytes;
6301 cache->reserved -= num_bytes;
6302 cache->space_info->bytes_reserved -= num_bytes;
6304 spin_unlock(&cache->lock);
6305 spin_unlock(&cache->space_info->lock);
6307 trace_btrfs_space_reservation(root->fs_info, "pinned",
6308 cache->space_info->flags, num_bytes, 1);
6309 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
6310 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
6315 * this function must be called within transaction
6317 int btrfs_pin_extent(struct btrfs_root *root,
6318 u64 bytenr, u64 num_bytes, int reserved)
6320 struct btrfs_block_group_cache *cache;
6322 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
6323 BUG_ON(!cache); /* Logic error */
6325 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
6327 btrfs_put_block_group(cache);
6332 * this function must be called within transaction
6334 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
6335 u64 bytenr, u64 num_bytes)
6337 struct btrfs_block_group_cache *cache;
6340 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
6345 * pull in the free space cache (if any) so that our pin
6346 * removes the free space from the cache. We have load_only set
6347 * to one because the slow code to read in the free extents does check
6348 * the pinned extents.
6350 cache_block_group(cache, 1);
6352 pin_down_extent(root, cache, bytenr, num_bytes, 0);
6354 /* remove us from the free space cache (if we're there at all) */
6355 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6356 btrfs_put_block_group(cache);
6360 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
6363 struct btrfs_block_group_cache *block_group;
6364 struct btrfs_caching_control *caching_ctl;
6366 block_group = btrfs_lookup_block_group(root->fs_info, start);
6370 cache_block_group(block_group, 0);
6371 caching_ctl = get_caching_control(block_group);
6375 BUG_ON(!block_group_cache_done(block_group));
6376 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6378 mutex_lock(&caching_ctl->mutex);
6380 if (start >= caching_ctl->progress) {
6381 ret = add_excluded_extent(root, start, num_bytes);
6382 } else if (start + num_bytes <= caching_ctl->progress) {
6383 ret = btrfs_remove_free_space(block_group,
6386 num_bytes = caching_ctl->progress - start;
6387 ret = btrfs_remove_free_space(block_group,
6392 num_bytes = (start + num_bytes) -
6393 caching_ctl->progress;
6394 start = caching_ctl->progress;
6395 ret = add_excluded_extent(root, start, num_bytes);
6398 mutex_unlock(&caching_ctl->mutex);
6399 put_caching_control(caching_ctl);
6401 btrfs_put_block_group(block_group);
6405 int btrfs_exclude_logged_extents(struct btrfs_root *log,
6406 struct extent_buffer *eb)
6408 struct btrfs_file_extent_item *item;
6409 struct btrfs_key key;
6413 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
6416 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6417 btrfs_item_key_to_cpu(eb, &key, i);
6418 if (key.type != BTRFS_EXTENT_DATA_KEY)
6420 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6421 found_type = btrfs_file_extent_type(eb, item);
6422 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6424 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6426 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6427 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6428 __exclude_logged_extent(log, key.objectid, key.offset);
6435 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg)
6437 atomic_inc(&bg->reservations);
6440 void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info,
6443 struct btrfs_block_group_cache *bg;
6445 bg = btrfs_lookup_block_group(fs_info, start);
6447 if (atomic_dec_and_test(&bg->reservations))
6448 wake_up_atomic_t(&bg->reservations);
6449 btrfs_put_block_group(bg);
6452 static int btrfs_wait_bg_reservations_atomic_t(atomic_t *a)
6458 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg)
6460 struct btrfs_space_info *space_info = bg->space_info;
6464 if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA))
6468 * Our block group is read only but before we set it to read only,
6469 * some task might have had allocated an extent from it already, but it
6470 * has not yet created a respective ordered extent (and added it to a
6471 * root's list of ordered extents).
6472 * Therefore wait for any task currently allocating extents, since the
6473 * block group's reservations counter is incremented while a read lock
6474 * on the groups' semaphore is held and decremented after releasing
6475 * the read access on that semaphore and creating the ordered extent.
6477 down_write(&space_info->groups_sem);
6478 up_write(&space_info->groups_sem);
6480 wait_on_atomic_t(&bg->reservations,
6481 btrfs_wait_bg_reservations_atomic_t,
6482 TASK_UNINTERRUPTIBLE);
6486 * btrfs_add_reserved_bytes - update the block_group and space info counters
6487 * @cache: The cache we are manipulating
6488 * @ram_bytes: The number of bytes of file content, and will be same to
6489 * @num_bytes except for the compress path.
6490 * @num_bytes: The number of bytes in question
6491 * @delalloc: The blocks are allocated for the delalloc write
6493 * This is called by the allocator when it reserves space. Metadata
6494 * reservations should be called with RESERVE_ALLOC so we do the proper
6495 * ENOSPC accounting. For data we handle the reservation through clearing the
6496 * delalloc bits in the io_tree. We have to do this since we could end up
6497 * allocating less disk space for the amount of data we have reserved in the
6498 * case of compression.
6500 * If this is a reservation and the block group has become read only we cannot
6501 * make the reservation and return -EAGAIN, otherwise this function always
6504 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache,
6505 u64 ram_bytes, u64 num_bytes, int delalloc)
6507 struct btrfs_space_info *space_info = cache->space_info;
6510 spin_lock(&space_info->lock);
6511 spin_lock(&cache->lock);
6515 cache->reserved += num_bytes;
6516 space_info->bytes_reserved += num_bytes;
6518 trace_btrfs_space_reservation(cache->fs_info,
6519 "space_info", space_info->flags,
6521 space_info->bytes_may_use -= ram_bytes;
6523 cache->delalloc_bytes += num_bytes;
6525 spin_unlock(&cache->lock);
6526 spin_unlock(&space_info->lock);
6531 * btrfs_free_reserved_bytes - update the block_group and space info counters
6532 * @cache: The cache we are manipulating
6533 * @num_bytes: The number of bytes in question
6534 * @delalloc: The blocks are allocated for the delalloc write
6536 * This is called by somebody who is freeing space that was never actually used
6537 * on disk. For example if you reserve some space for a new leaf in transaction
6538 * A and before transaction A commits you free that leaf, you call this with
6539 * reserve set to 0 in order to clear the reservation.
6542 static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache,
6543 u64 num_bytes, int delalloc)
6545 struct btrfs_space_info *space_info = cache->space_info;
6548 spin_lock(&space_info->lock);
6549 spin_lock(&cache->lock);
6551 space_info->bytes_readonly += num_bytes;
6552 cache->reserved -= num_bytes;
6553 space_info->bytes_reserved -= num_bytes;
6556 cache->delalloc_bytes -= num_bytes;
6557 spin_unlock(&cache->lock);
6558 spin_unlock(&space_info->lock);
6561 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
6562 struct btrfs_root *root)
6564 struct btrfs_fs_info *fs_info = root->fs_info;
6565 struct btrfs_caching_control *next;
6566 struct btrfs_caching_control *caching_ctl;
6567 struct btrfs_block_group_cache *cache;
6569 down_write(&fs_info->commit_root_sem);
6571 list_for_each_entry_safe(caching_ctl, next,
6572 &fs_info->caching_block_groups, list) {
6573 cache = caching_ctl->block_group;
6574 if (block_group_cache_done(cache)) {
6575 cache->last_byte_to_unpin = (u64)-1;
6576 list_del_init(&caching_ctl->list);
6577 put_caching_control(caching_ctl);
6579 cache->last_byte_to_unpin = caching_ctl->progress;
6583 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6584 fs_info->pinned_extents = &fs_info->freed_extents[1];
6586 fs_info->pinned_extents = &fs_info->freed_extents[0];
6588 up_write(&fs_info->commit_root_sem);
6590 update_global_block_rsv(fs_info);
6594 * Returns the free cluster for the given space info and sets empty_cluster to
6595 * what it should be based on the mount options.
6597 static struct btrfs_free_cluster *
6598 fetch_cluster_info(struct btrfs_root *root, struct btrfs_space_info *space_info,
6601 struct btrfs_free_cluster *ret = NULL;
6602 bool ssd = btrfs_test_opt(root->fs_info, SSD);
6605 if (btrfs_mixed_space_info(space_info))
6609 *empty_cluster = SZ_2M;
6610 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
6611 ret = &root->fs_info->meta_alloc_cluster;
6613 *empty_cluster = SZ_64K;
6614 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && ssd) {
6615 ret = &root->fs_info->data_alloc_cluster;
6621 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end,
6622 const bool return_free_space)
6624 struct btrfs_fs_info *fs_info = root->fs_info;
6625 struct btrfs_block_group_cache *cache = NULL;
6626 struct btrfs_space_info *space_info;
6627 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6628 struct btrfs_free_cluster *cluster = NULL;
6630 u64 total_unpinned = 0;
6631 u64 empty_cluster = 0;
6634 while (start <= end) {
6637 start >= cache->key.objectid + cache->key.offset) {
6639 btrfs_put_block_group(cache);
6641 cache = btrfs_lookup_block_group(fs_info, start);
6642 BUG_ON(!cache); /* Logic error */
6644 cluster = fetch_cluster_info(root,
6647 empty_cluster <<= 1;
6650 len = cache->key.objectid + cache->key.offset - start;
6651 len = min(len, end + 1 - start);
6653 if (start < cache->last_byte_to_unpin) {
6654 len = min(len, cache->last_byte_to_unpin - start);
6655 if (return_free_space)
6656 btrfs_add_free_space(cache, start, len);
6660 total_unpinned += len;
6661 space_info = cache->space_info;
6664 * If this space cluster has been marked as fragmented and we've
6665 * unpinned enough in this block group to potentially allow a
6666 * cluster to be created inside of it go ahead and clear the
6669 if (cluster && cluster->fragmented &&
6670 total_unpinned > empty_cluster) {
6671 spin_lock(&cluster->lock);
6672 cluster->fragmented = 0;
6673 spin_unlock(&cluster->lock);
6676 spin_lock(&space_info->lock);
6677 spin_lock(&cache->lock);
6678 cache->pinned -= len;
6679 space_info->bytes_pinned -= len;
6681 trace_btrfs_space_reservation(fs_info, "pinned",
6682 space_info->flags, len, 0);
6683 space_info->max_extent_size = 0;
6684 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6686 space_info->bytes_readonly += len;
6689 spin_unlock(&cache->lock);
6690 if (!readonly && return_free_space &&
6691 global_rsv->space_info == space_info) {
6693 WARN_ON(!return_free_space);
6694 spin_lock(&global_rsv->lock);
6695 if (!global_rsv->full) {
6696 to_add = min(len, global_rsv->size -
6697 global_rsv->reserved);
6698 global_rsv->reserved += to_add;
6699 space_info->bytes_may_use += to_add;
6700 if (global_rsv->reserved >= global_rsv->size)
6701 global_rsv->full = 1;
6702 trace_btrfs_space_reservation(fs_info,
6708 spin_unlock(&global_rsv->lock);
6709 /* Add to any tickets we may have */
6711 space_info_add_new_bytes(fs_info, space_info,
6714 spin_unlock(&space_info->lock);
6718 btrfs_put_block_group(cache);
6722 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
6723 struct btrfs_root *root)
6725 struct btrfs_fs_info *fs_info = root->fs_info;
6726 struct btrfs_block_group_cache *block_group, *tmp;
6727 struct list_head *deleted_bgs;
6728 struct extent_io_tree *unpin;
6733 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6734 unpin = &fs_info->freed_extents[1];
6736 unpin = &fs_info->freed_extents[0];
6738 while (!trans->aborted) {
6739 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6740 ret = find_first_extent_bit(unpin, 0, &start, &end,
6741 EXTENT_DIRTY, NULL);
6743 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6747 if (btrfs_test_opt(root->fs_info, DISCARD))
6748 ret = btrfs_discard_extent(root, start,
6749 end + 1 - start, NULL);
6751 clear_extent_dirty(unpin, start, end);
6752 unpin_extent_range(root, start, end, true);
6753 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6758 * Transaction is finished. We don't need the lock anymore. We
6759 * do need to clean up the block groups in case of a transaction
6762 deleted_bgs = &trans->transaction->deleted_bgs;
6763 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6767 if (!trans->aborted)
6768 ret = btrfs_discard_extent(root,
6769 block_group->key.objectid,
6770 block_group->key.offset,
6773 list_del_init(&block_group->bg_list);
6774 btrfs_put_block_group_trimming(block_group);
6775 btrfs_put_block_group(block_group);
6778 const char *errstr = btrfs_decode_error(ret);
6780 "Discard failed while removing blockgroup: errno=%d %s\n",
6788 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
6789 u64 owner, u64 root_objectid)
6791 struct btrfs_space_info *space_info;
6794 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6795 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
6796 flags = BTRFS_BLOCK_GROUP_SYSTEM;
6798 flags = BTRFS_BLOCK_GROUP_METADATA;
6800 flags = BTRFS_BLOCK_GROUP_DATA;
6803 space_info = __find_space_info(fs_info, flags);
6804 BUG_ON(!space_info); /* Logic bug */
6805 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
6809 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6810 struct btrfs_root *root,
6811 struct btrfs_delayed_ref_node *node, u64 parent,
6812 u64 root_objectid, u64 owner_objectid,
6813 u64 owner_offset, int refs_to_drop,
6814 struct btrfs_delayed_extent_op *extent_op)
6816 struct btrfs_key key;
6817 struct btrfs_path *path;
6818 struct btrfs_fs_info *info = root->fs_info;
6819 struct btrfs_root *extent_root = info->extent_root;
6820 struct extent_buffer *leaf;
6821 struct btrfs_extent_item *ei;
6822 struct btrfs_extent_inline_ref *iref;
6825 int extent_slot = 0;
6826 int found_extent = 0;
6830 u64 bytenr = node->bytenr;
6831 u64 num_bytes = node->num_bytes;
6833 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6836 path = btrfs_alloc_path();
6840 path->reada = READA_FORWARD;
6841 path->leave_spinning = 1;
6843 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6844 BUG_ON(!is_data && refs_to_drop != 1);
6847 skinny_metadata = 0;
6849 ret = lookup_extent_backref(trans, extent_root, path, &iref,
6850 bytenr, num_bytes, parent,
6851 root_objectid, owner_objectid,
6854 extent_slot = path->slots[0];
6855 while (extent_slot >= 0) {
6856 btrfs_item_key_to_cpu(path->nodes[0], &key,
6858 if (key.objectid != bytenr)
6860 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6861 key.offset == num_bytes) {
6865 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6866 key.offset == owner_objectid) {
6870 if (path->slots[0] - extent_slot > 5)
6874 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6875 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6876 if (found_extent && item_size < sizeof(*ei))
6879 if (!found_extent) {
6881 ret = remove_extent_backref(trans, extent_root, path,
6883 is_data, &last_ref);
6885 btrfs_abort_transaction(trans, ret);
6888 btrfs_release_path(path);
6889 path->leave_spinning = 1;
6891 key.objectid = bytenr;
6892 key.type = BTRFS_EXTENT_ITEM_KEY;
6893 key.offset = num_bytes;
6895 if (!is_data && skinny_metadata) {
6896 key.type = BTRFS_METADATA_ITEM_KEY;
6897 key.offset = owner_objectid;
6900 ret = btrfs_search_slot(trans, extent_root,
6902 if (ret > 0 && skinny_metadata && path->slots[0]) {
6904 * Couldn't find our skinny metadata item,
6905 * see if we have ye olde extent item.
6908 btrfs_item_key_to_cpu(path->nodes[0], &key,
6910 if (key.objectid == bytenr &&
6911 key.type == BTRFS_EXTENT_ITEM_KEY &&
6912 key.offset == num_bytes)
6916 if (ret > 0 && skinny_metadata) {
6917 skinny_metadata = false;
6918 key.objectid = bytenr;
6919 key.type = BTRFS_EXTENT_ITEM_KEY;
6920 key.offset = num_bytes;
6921 btrfs_release_path(path);
6922 ret = btrfs_search_slot(trans, extent_root,
6927 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6930 btrfs_print_leaf(extent_root,
6934 btrfs_abort_transaction(trans, ret);
6937 extent_slot = path->slots[0];
6939 } else if (WARN_ON(ret == -ENOENT)) {
6940 btrfs_print_leaf(extent_root, path->nodes[0]);
6942 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6943 bytenr, parent, root_objectid, owner_objectid,
6945 btrfs_abort_transaction(trans, ret);
6948 btrfs_abort_transaction(trans, ret);
6952 leaf = path->nodes[0];
6953 item_size = btrfs_item_size_nr(leaf, extent_slot);
6954 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6955 if (item_size < sizeof(*ei)) {
6956 BUG_ON(found_extent || extent_slot != path->slots[0]);
6957 ret = convert_extent_item_v0(trans, extent_root, path,
6960 btrfs_abort_transaction(trans, ret);
6964 btrfs_release_path(path);
6965 path->leave_spinning = 1;
6967 key.objectid = bytenr;
6968 key.type = BTRFS_EXTENT_ITEM_KEY;
6969 key.offset = num_bytes;
6971 ret = btrfs_search_slot(trans, extent_root, &key, path,
6974 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6976 btrfs_print_leaf(extent_root, path->nodes[0]);
6979 btrfs_abort_transaction(trans, ret);
6983 extent_slot = path->slots[0];
6984 leaf = path->nodes[0];
6985 item_size = btrfs_item_size_nr(leaf, extent_slot);
6988 BUG_ON(item_size < sizeof(*ei));
6989 ei = btrfs_item_ptr(leaf, extent_slot,
6990 struct btrfs_extent_item);
6991 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
6992 key.type == BTRFS_EXTENT_ITEM_KEY) {
6993 struct btrfs_tree_block_info *bi;
6994 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
6995 bi = (struct btrfs_tree_block_info *)(ei + 1);
6996 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
6999 refs = btrfs_extent_refs(leaf, ei);
7000 if (refs < refs_to_drop) {
7001 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
7002 "for bytenr %Lu", refs_to_drop, refs, bytenr);
7004 btrfs_abort_transaction(trans, ret);
7007 refs -= refs_to_drop;
7011 __run_delayed_extent_op(extent_op, leaf, ei);
7013 * In the case of inline back ref, reference count will
7014 * be updated by remove_extent_backref
7017 BUG_ON(!found_extent);
7019 btrfs_set_extent_refs(leaf, ei, refs);
7020 btrfs_mark_buffer_dirty(leaf);
7023 ret = remove_extent_backref(trans, extent_root, path,
7025 is_data, &last_ref);
7027 btrfs_abort_transaction(trans, ret);
7031 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
7035 BUG_ON(is_data && refs_to_drop !=
7036 extent_data_ref_count(path, iref));
7038 BUG_ON(path->slots[0] != extent_slot);
7040 BUG_ON(path->slots[0] != extent_slot + 1);
7041 path->slots[0] = extent_slot;
7047 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
7050 btrfs_abort_transaction(trans, ret);
7053 btrfs_release_path(path);
7056 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
7058 btrfs_abort_transaction(trans, ret);
7063 ret = add_to_free_space_tree(trans, root->fs_info, bytenr,
7066 btrfs_abort_transaction(trans, ret);
7070 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
7072 btrfs_abort_transaction(trans, ret);
7076 btrfs_release_path(path);
7079 btrfs_free_path(path);
7084 * when we free an block, it is possible (and likely) that we free the last
7085 * delayed ref for that extent as well. This searches the delayed ref tree for
7086 * a given extent, and if there are no other delayed refs to be processed, it
7087 * removes it from the tree.
7089 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
7090 struct btrfs_root *root, u64 bytenr)
7092 struct btrfs_delayed_ref_head *head;
7093 struct btrfs_delayed_ref_root *delayed_refs;
7096 delayed_refs = &trans->transaction->delayed_refs;
7097 spin_lock(&delayed_refs->lock);
7098 head = btrfs_find_delayed_ref_head(trans, bytenr);
7100 goto out_delayed_unlock;
7102 spin_lock(&head->lock);
7103 if (!list_empty(&head->ref_list))
7106 if (head->extent_op) {
7107 if (!head->must_insert_reserved)
7109 btrfs_free_delayed_extent_op(head->extent_op);
7110 head->extent_op = NULL;
7114 * waiting for the lock here would deadlock. If someone else has it
7115 * locked they are already in the process of dropping it anyway
7117 if (!mutex_trylock(&head->mutex))
7121 * at this point we have a head with no other entries. Go
7122 * ahead and process it.
7124 head->node.in_tree = 0;
7125 rb_erase(&head->href_node, &delayed_refs->href_root);
7127 atomic_dec(&delayed_refs->num_entries);
7130 * we don't take a ref on the node because we're removing it from the
7131 * tree, so we just steal the ref the tree was holding.
7133 delayed_refs->num_heads--;
7134 if (head->processing == 0)
7135 delayed_refs->num_heads_ready--;
7136 head->processing = 0;
7137 spin_unlock(&head->lock);
7138 spin_unlock(&delayed_refs->lock);
7140 BUG_ON(head->extent_op);
7141 if (head->must_insert_reserved)
7144 mutex_unlock(&head->mutex);
7145 btrfs_put_delayed_ref(&head->node);
7148 spin_unlock(&head->lock);
7151 spin_unlock(&delayed_refs->lock);
7155 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
7156 struct btrfs_root *root,
7157 struct extent_buffer *buf,
7158 u64 parent, int last_ref)
7163 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7164 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7165 buf->start, buf->len,
7166 parent, root->root_key.objectid,
7167 btrfs_header_level(buf),
7168 BTRFS_DROP_DELAYED_REF, NULL);
7169 BUG_ON(ret); /* -ENOMEM */
7175 if (btrfs_header_generation(buf) == trans->transid) {
7176 struct btrfs_block_group_cache *cache;
7178 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7179 ret = check_ref_cleanup(trans, root, buf->start);
7184 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
7186 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
7187 pin_down_extent(root, cache, buf->start, buf->len, 1);
7188 btrfs_put_block_group(cache);
7192 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
7194 btrfs_add_free_space(cache, buf->start, buf->len);
7195 btrfs_free_reserved_bytes(cache, buf->len, 0);
7196 btrfs_put_block_group(cache);
7197 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
7202 add_pinned_bytes(root->fs_info, buf->len,
7203 btrfs_header_level(buf),
7204 root->root_key.objectid);
7207 * Deleting the buffer, clear the corrupt flag since it doesn't matter
7210 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
7213 /* Can return -ENOMEM */
7214 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
7215 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
7216 u64 owner, u64 offset)
7219 struct btrfs_fs_info *fs_info = root->fs_info;
7221 if (btrfs_is_testing(fs_info))
7224 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
7227 * tree log blocks never actually go into the extent allocation
7228 * tree, just update pinning info and exit early.
7230 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
7231 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
7232 /* unlocks the pinned mutex */
7233 btrfs_pin_extent(root, bytenr, num_bytes, 1);
7235 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
7236 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
7238 parent, root_objectid, (int)owner,
7239 BTRFS_DROP_DELAYED_REF, NULL);
7241 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
7243 parent, root_objectid, owner,
7245 BTRFS_DROP_DELAYED_REF, NULL);
7251 * when we wait for progress in the block group caching, its because
7252 * our allocation attempt failed at least once. So, we must sleep
7253 * and let some progress happen before we try again.
7255 * This function will sleep at least once waiting for new free space to
7256 * show up, and then it will check the block group free space numbers
7257 * for our min num_bytes. Another option is to have it go ahead
7258 * and look in the rbtree for a free extent of a given size, but this
7261 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7262 * any of the information in this block group.
7264 static noinline void
7265 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
7268 struct btrfs_caching_control *caching_ctl;
7270 caching_ctl = get_caching_control(cache);
7274 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
7275 (cache->free_space_ctl->free_space >= num_bytes));
7277 put_caching_control(caching_ctl);
7281 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
7283 struct btrfs_caching_control *caching_ctl;
7286 caching_ctl = get_caching_control(cache);
7288 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
7290 wait_event(caching_ctl->wait, block_group_cache_done(cache));
7291 if (cache->cached == BTRFS_CACHE_ERROR)
7293 put_caching_control(caching_ctl);
7297 int __get_raid_index(u64 flags)
7299 if (flags & BTRFS_BLOCK_GROUP_RAID10)
7300 return BTRFS_RAID_RAID10;
7301 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
7302 return BTRFS_RAID_RAID1;
7303 else if (flags & BTRFS_BLOCK_GROUP_DUP)
7304 return BTRFS_RAID_DUP;
7305 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
7306 return BTRFS_RAID_RAID0;
7307 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
7308 return BTRFS_RAID_RAID5;
7309 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
7310 return BTRFS_RAID_RAID6;
7312 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
7315 int get_block_group_index(struct btrfs_block_group_cache *cache)
7317 return __get_raid_index(cache->flags);
7320 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
7321 [BTRFS_RAID_RAID10] = "raid10",
7322 [BTRFS_RAID_RAID1] = "raid1",
7323 [BTRFS_RAID_DUP] = "dup",
7324 [BTRFS_RAID_RAID0] = "raid0",
7325 [BTRFS_RAID_SINGLE] = "single",
7326 [BTRFS_RAID_RAID5] = "raid5",
7327 [BTRFS_RAID_RAID6] = "raid6",
7330 static const char *get_raid_name(enum btrfs_raid_types type)
7332 if (type >= BTRFS_NR_RAID_TYPES)
7335 return btrfs_raid_type_names[type];
7338 enum btrfs_loop_type {
7339 LOOP_CACHING_NOWAIT = 0,
7340 LOOP_CACHING_WAIT = 1,
7341 LOOP_ALLOC_CHUNK = 2,
7342 LOOP_NO_EMPTY_SIZE = 3,
7346 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
7350 down_read(&cache->data_rwsem);
7354 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
7357 btrfs_get_block_group(cache);
7359 down_read(&cache->data_rwsem);
7362 static struct btrfs_block_group_cache *
7363 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
7364 struct btrfs_free_cluster *cluster,
7367 struct btrfs_block_group_cache *used_bg = NULL;
7369 spin_lock(&cluster->refill_lock);
7371 used_bg = cluster->block_group;
7375 if (used_bg == block_group)
7378 btrfs_get_block_group(used_bg);
7383 if (down_read_trylock(&used_bg->data_rwsem))
7386 spin_unlock(&cluster->refill_lock);
7388 down_read(&used_bg->data_rwsem);
7390 spin_lock(&cluster->refill_lock);
7391 if (used_bg == cluster->block_group)
7394 up_read(&used_bg->data_rwsem);
7395 btrfs_put_block_group(used_bg);
7400 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
7404 up_read(&cache->data_rwsem);
7405 btrfs_put_block_group(cache);
7409 * walks the btree of allocated extents and find a hole of a given size.
7410 * The key ins is changed to record the hole:
7411 * ins->objectid == start position
7412 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7413 * ins->offset == the size of the hole.
7414 * Any available blocks before search_start are skipped.
7416 * If there is no suitable free space, we will record the max size of
7417 * the free space extent currently.
7419 static noinline int find_free_extent(struct btrfs_root *orig_root,
7420 u64 ram_bytes, u64 num_bytes, u64 empty_size,
7421 u64 hint_byte, struct btrfs_key *ins,
7422 u64 flags, int delalloc)
7425 struct btrfs_root *root = orig_root->fs_info->extent_root;
7426 struct btrfs_free_cluster *last_ptr = NULL;
7427 struct btrfs_block_group_cache *block_group = NULL;
7428 u64 search_start = 0;
7429 u64 max_extent_size = 0;
7430 u64 empty_cluster = 0;
7431 struct btrfs_space_info *space_info;
7433 int index = __get_raid_index(flags);
7434 bool failed_cluster_refill = false;
7435 bool failed_alloc = false;
7436 bool use_cluster = true;
7437 bool have_caching_bg = false;
7438 bool orig_have_caching_bg = false;
7439 bool full_search = false;
7441 WARN_ON(num_bytes < root->sectorsize);
7442 ins->type = BTRFS_EXTENT_ITEM_KEY;
7446 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
7448 space_info = __find_space_info(root->fs_info, flags);
7450 btrfs_err(root->fs_info, "No space info for %llu", flags);
7455 * If our free space is heavily fragmented we may not be able to make
7456 * big contiguous allocations, so instead of doing the expensive search
7457 * for free space, simply return ENOSPC with our max_extent_size so we
7458 * can go ahead and search for a more manageable chunk.
7460 * If our max_extent_size is large enough for our allocation simply
7461 * disable clustering since we will likely not be able to find enough
7462 * space to create a cluster and induce latency trying.
7464 if (unlikely(space_info->max_extent_size)) {
7465 spin_lock(&space_info->lock);
7466 if (space_info->max_extent_size &&
7467 num_bytes > space_info->max_extent_size) {
7468 ins->offset = space_info->max_extent_size;
7469 spin_unlock(&space_info->lock);
7471 } else if (space_info->max_extent_size) {
7472 use_cluster = false;
7474 spin_unlock(&space_info->lock);
7477 last_ptr = fetch_cluster_info(orig_root, space_info, &empty_cluster);
7479 spin_lock(&last_ptr->lock);
7480 if (last_ptr->block_group)
7481 hint_byte = last_ptr->window_start;
7482 if (last_ptr->fragmented) {
7484 * We still set window_start so we can keep track of the
7485 * last place we found an allocation to try and save
7488 hint_byte = last_ptr->window_start;
7489 use_cluster = false;
7491 spin_unlock(&last_ptr->lock);
7494 search_start = max(search_start, first_logical_byte(root, 0));
7495 search_start = max(search_start, hint_byte);
7496 if (search_start == hint_byte) {
7497 block_group = btrfs_lookup_block_group(root->fs_info,
7500 * we don't want to use the block group if it doesn't match our
7501 * allocation bits, or if its not cached.
7503 * However if we are re-searching with an ideal block group
7504 * picked out then we don't care that the block group is cached.
7506 if (block_group && block_group_bits(block_group, flags) &&
7507 block_group->cached != BTRFS_CACHE_NO) {
7508 down_read(&space_info->groups_sem);
7509 if (list_empty(&block_group->list) ||
7512 * someone is removing this block group,
7513 * we can't jump into the have_block_group
7514 * target because our list pointers are not
7517 btrfs_put_block_group(block_group);
7518 up_read(&space_info->groups_sem);
7520 index = get_block_group_index(block_group);
7521 btrfs_lock_block_group(block_group, delalloc);
7522 goto have_block_group;
7524 } else if (block_group) {
7525 btrfs_put_block_group(block_group);
7529 have_caching_bg = false;
7530 if (index == 0 || index == __get_raid_index(flags))
7532 down_read(&space_info->groups_sem);
7533 list_for_each_entry(block_group, &space_info->block_groups[index],
7538 btrfs_grab_block_group(block_group, delalloc);
7539 search_start = block_group->key.objectid;
7542 * this can happen if we end up cycling through all the
7543 * raid types, but we want to make sure we only allocate
7544 * for the proper type.
7546 if (!block_group_bits(block_group, flags)) {
7547 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7548 BTRFS_BLOCK_GROUP_RAID1 |
7549 BTRFS_BLOCK_GROUP_RAID5 |
7550 BTRFS_BLOCK_GROUP_RAID6 |
7551 BTRFS_BLOCK_GROUP_RAID10;
7554 * if they asked for extra copies and this block group
7555 * doesn't provide them, bail. This does allow us to
7556 * fill raid0 from raid1.
7558 if ((flags & extra) && !(block_group->flags & extra))
7563 cached = block_group_cache_done(block_group);
7564 if (unlikely(!cached)) {
7565 have_caching_bg = true;
7566 ret = cache_block_group(block_group, 0);
7571 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7573 if (unlikely(block_group->ro))
7577 * Ok we want to try and use the cluster allocator, so
7580 if (last_ptr && use_cluster) {
7581 struct btrfs_block_group_cache *used_block_group;
7582 unsigned long aligned_cluster;
7584 * the refill lock keeps out other
7585 * people trying to start a new cluster
7587 used_block_group = btrfs_lock_cluster(block_group,
7590 if (!used_block_group)
7591 goto refill_cluster;
7593 if (used_block_group != block_group &&
7594 (used_block_group->ro ||
7595 !block_group_bits(used_block_group, flags)))
7596 goto release_cluster;
7598 offset = btrfs_alloc_from_cluster(used_block_group,
7601 used_block_group->key.objectid,
7604 /* we have a block, we're done */
7605 spin_unlock(&last_ptr->refill_lock);
7606 trace_btrfs_reserve_extent_cluster(root,
7608 search_start, num_bytes);
7609 if (used_block_group != block_group) {
7610 btrfs_release_block_group(block_group,
7612 block_group = used_block_group;
7617 WARN_ON(last_ptr->block_group != used_block_group);
7619 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7620 * set up a new clusters, so lets just skip it
7621 * and let the allocator find whatever block
7622 * it can find. If we reach this point, we
7623 * will have tried the cluster allocator
7624 * plenty of times and not have found
7625 * anything, so we are likely way too
7626 * fragmented for the clustering stuff to find
7629 * However, if the cluster is taken from the
7630 * current block group, release the cluster
7631 * first, so that we stand a better chance of
7632 * succeeding in the unclustered
7634 if (loop >= LOOP_NO_EMPTY_SIZE &&
7635 used_block_group != block_group) {
7636 spin_unlock(&last_ptr->refill_lock);
7637 btrfs_release_block_group(used_block_group,
7639 goto unclustered_alloc;
7643 * this cluster didn't work out, free it and
7646 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7648 if (used_block_group != block_group)
7649 btrfs_release_block_group(used_block_group,
7652 if (loop >= LOOP_NO_EMPTY_SIZE) {
7653 spin_unlock(&last_ptr->refill_lock);
7654 goto unclustered_alloc;
7657 aligned_cluster = max_t(unsigned long,
7658 empty_cluster + empty_size,
7659 block_group->full_stripe_len);
7661 /* allocate a cluster in this block group */
7662 ret = btrfs_find_space_cluster(root, block_group,
7663 last_ptr, search_start,
7668 * now pull our allocation out of this
7671 offset = btrfs_alloc_from_cluster(block_group,
7677 /* we found one, proceed */
7678 spin_unlock(&last_ptr->refill_lock);
7679 trace_btrfs_reserve_extent_cluster(root,
7680 block_group, search_start,
7684 } else if (!cached && loop > LOOP_CACHING_NOWAIT
7685 && !failed_cluster_refill) {
7686 spin_unlock(&last_ptr->refill_lock);
7688 failed_cluster_refill = true;
7689 wait_block_group_cache_progress(block_group,
7690 num_bytes + empty_cluster + empty_size);
7691 goto have_block_group;
7695 * at this point we either didn't find a cluster
7696 * or we weren't able to allocate a block from our
7697 * cluster. Free the cluster we've been trying
7698 * to use, and go to the next block group
7700 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7701 spin_unlock(&last_ptr->refill_lock);
7707 * We are doing an unclustered alloc, set the fragmented flag so
7708 * we don't bother trying to setup a cluster again until we get
7711 if (unlikely(last_ptr)) {
7712 spin_lock(&last_ptr->lock);
7713 last_ptr->fragmented = 1;
7714 spin_unlock(&last_ptr->lock);
7716 spin_lock(&block_group->free_space_ctl->tree_lock);
7718 block_group->free_space_ctl->free_space <
7719 num_bytes + empty_cluster + empty_size) {
7720 if (block_group->free_space_ctl->free_space >
7723 block_group->free_space_ctl->free_space;
7724 spin_unlock(&block_group->free_space_ctl->tree_lock);
7727 spin_unlock(&block_group->free_space_ctl->tree_lock);
7729 offset = btrfs_find_space_for_alloc(block_group, search_start,
7730 num_bytes, empty_size,
7733 * If we didn't find a chunk, and we haven't failed on this
7734 * block group before, and this block group is in the middle of
7735 * caching and we are ok with waiting, then go ahead and wait
7736 * for progress to be made, and set failed_alloc to true.
7738 * If failed_alloc is true then we've already waited on this
7739 * block group once and should move on to the next block group.
7741 if (!offset && !failed_alloc && !cached &&
7742 loop > LOOP_CACHING_NOWAIT) {
7743 wait_block_group_cache_progress(block_group,
7744 num_bytes + empty_size);
7745 failed_alloc = true;
7746 goto have_block_group;
7747 } else if (!offset) {
7751 search_start = ALIGN(offset, root->stripesize);
7753 /* move on to the next group */
7754 if (search_start + num_bytes >
7755 block_group->key.objectid + block_group->key.offset) {
7756 btrfs_add_free_space(block_group, offset, num_bytes);
7760 if (offset < search_start)
7761 btrfs_add_free_space(block_group, offset,
7762 search_start - offset);
7763 BUG_ON(offset > search_start);
7765 ret = btrfs_add_reserved_bytes(block_group, ram_bytes,
7766 num_bytes, delalloc);
7767 if (ret == -EAGAIN) {
7768 btrfs_add_free_space(block_group, offset, num_bytes);
7771 btrfs_inc_block_group_reservations(block_group);
7773 /* we are all good, lets return */
7774 ins->objectid = search_start;
7775 ins->offset = num_bytes;
7777 trace_btrfs_reserve_extent(orig_root, block_group,
7778 search_start, num_bytes);
7779 btrfs_release_block_group(block_group, delalloc);
7782 failed_cluster_refill = false;
7783 failed_alloc = false;
7784 BUG_ON(index != get_block_group_index(block_group));
7785 btrfs_release_block_group(block_group, delalloc);
7787 up_read(&space_info->groups_sem);
7789 if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg
7790 && !orig_have_caching_bg)
7791 orig_have_caching_bg = true;
7793 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7796 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7800 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7801 * caching kthreads as we move along
7802 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7803 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7804 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7807 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7809 if (loop == LOOP_CACHING_NOWAIT) {
7811 * We want to skip the LOOP_CACHING_WAIT step if we
7812 * don't have any uncached bgs and we've already done a
7813 * full search through.
7815 if (orig_have_caching_bg || !full_search)
7816 loop = LOOP_CACHING_WAIT;
7818 loop = LOOP_ALLOC_CHUNK;
7823 if (loop == LOOP_ALLOC_CHUNK) {
7824 struct btrfs_trans_handle *trans;
7827 trans = current->journal_info;
7831 trans = btrfs_join_transaction(root);
7833 if (IS_ERR(trans)) {
7834 ret = PTR_ERR(trans);
7838 ret = do_chunk_alloc(trans, root, flags,
7842 * If we can't allocate a new chunk we've already looped
7843 * through at least once, move on to the NO_EMPTY_SIZE
7847 loop = LOOP_NO_EMPTY_SIZE;
7850 * Do not bail out on ENOSPC since we
7851 * can do more things.
7853 if (ret < 0 && ret != -ENOSPC)
7854 btrfs_abort_transaction(trans, ret);
7858 btrfs_end_transaction(trans, root);
7863 if (loop == LOOP_NO_EMPTY_SIZE) {
7865 * Don't loop again if we already have no empty_size and
7868 if (empty_size == 0 &&
7869 empty_cluster == 0) {
7878 } else if (!ins->objectid) {
7880 } else if (ins->objectid) {
7881 if (!use_cluster && last_ptr) {
7882 spin_lock(&last_ptr->lock);
7883 last_ptr->window_start = ins->objectid;
7884 spin_unlock(&last_ptr->lock);
7889 if (ret == -ENOSPC) {
7890 spin_lock(&space_info->lock);
7891 space_info->max_extent_size = max_extent_size;
7892 spin_unlock(&space_info->lock);
7893 ins->offset = max_extent_size;
7898 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
7899 int dump_block_groups)
7901 struct btrfs_block_group_cache *cache;
7904 spin_lock(&info->lock);
7905 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
7907 info->total_bytes - info->bytes_used - info->bytes_pinned -
7908 info->bytes_reserved - info->bytes_readonly -
7909 info->bytes_may_use, (info->full) ? "" : "not ");
7910 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7911 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7912 info->total_bytes, info->bytes_used, info->bytes_pinned,
7913 info->bytes_reserved, info->bytes_may_use,
7914 info->bytes_readonly);
7915 spin_unlock(&info->lock);
7917 if (!dump_block_groups)
7920 down_read(&info->groups_sem);
7922 list_for_each_entry(cache, &info->block_groups[index], list) {
7923 spin_lock(&cache->lock);
7924 printk(KERN_INFO "BTRFS: "
7925 "block group %llu has %llu bytes, "
7926 "%llu used %llu pinned %llu reserved %s\n",
7927 cache->key.objectid, cache->key.offset,
7928 btrfs_block_group_used(&cache->item), cache->pinned,
7929 cache->reserved, cache->ro ? "[readonly]" : "");
7930 btrfs_dump_free_space(cache, bytes);
7931 spin_unlock(&cache->lock);
7933 if (++index < BTRFS_NR_RAID_TYPES)
7935 up_read(&info->groups_sem);
7938 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
7939 u64 num_bytes, u64 min_alloc_size,
7940 u64 empty_size, u64 hint_byte,
7941 struct btrfs_key *ins, int is_data, int delalloc)
7943 bool final_tried = num_bytes == min_alloc_size;
7947 flags = btrfs_get_alloc_profile(root, is_data);
7949 WARN_ON(num_bytes < root->sectorsize);
7950 ret = find_free_extent(root, ram_bytes, num_bytes, empty_size,
7951 hint_byte, ins, flags, delalloc);
7952 if (!ret && !is_data) {
7953 btrfs_dec_block_group_reservations(root->fs_info,
7955 } else if (ret == -ENOSPC) {
7956 if (!final_tried && ins->offset) {
7957 num_bytes = min(num_bytes >> 1, ins->offset);
7958 num_bytes = round_down(num_bytes, root->sectorsize);
7959 num_bytes = max(num_bytes, min_alloc_size);
7960 ram_bytes = num_bytes;
7961 if (num_bytes == min_alloc_size)
7964 } else if (btrfs_test_opt(root->fs_info, ENOSPC_DEBUG)) {
7965 struct btrfs_space_info *sinfo;
7967 sinfo = __find_space_info(root->fs_info, flags);
7968 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
7971 dump_space_info(sinfo, num_bytes, 1);
7978 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
7980 int pin, int delalloc)
7982 struct btrfs_block_group_cache *cache;
7985 cache = btrfs_lookup_block_group(root->fs_info, start);
7987 btrfs_err(root->fs_info, "Unable to find block group for %llu",
7993 pin_down_extent(root, cache, start, len, 1);
7995 if (btrfs_test_opt(root->fs_info, DISCARD))
7996 ret = btrfs_discard_extent(root, start, len, NULL);
7997 btrfs_add_free_space(cache, start, len);
7998 btrfs_free_reserved_bytes(cache, len, delalloc);
7999 trace_btrfs_reserved_extent_free(root, start, len);
8002 btrfs_put_block_group(cache);
8006 int btrfs_free_reserved_extent(struct btrfs_root *root,
8007 u64 start, u64 len, int delalloc)
8009 return __btrfs_free_reserved_extent(root, start, len, 0, delalloc);
8012 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
8015 return __btrfs_free_reserved_extent(root, start, len, 1, 0);
8018 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8019 struct btrfs_root *root,
8020 u64 parent, u64 root_objectid,
8021 u64 flags, u64 owner, u64 offset,
8022 struct btrfs_key *ins, int ref_mod)
8025 struct btrfs_fs_info *fs_info = root->fs_info;
8026 struct btrfs_extent_item *extent_item;
8027 struct btrfs_extent_inline_ref *iref;
8028 struct btrfs_path *path;
8029 struct extent_buffer *leaf;
8034 type = BTRFS_SHARED_DATA_REF_KEY;
8036 type = BTRFS_EXTENT_DATA_REF_KEY;
8038 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
8040 path = btrfs_alloc_path();
8044 path->leave_spinning = 1;
8045 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8048 btrfs_free_path(path);
8052 leaf = path->nodes[0];
8053 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8054 struct btrfs_extent_item);
8055 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
8056 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8057 btrfs_set_extent_flags(leaf, extent_item,
8058 flags | BTRFS_EXTENT_FLAG_DATA);
8060 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8061 btrfs_set_extent_inline_ref_type(leaf, iref, type);
8063 struct btrfs_shared_data_ref *ref;
8064 ref = (struct btrfs_shared_data_ref *)(iref + 1);
8065 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8066 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
8068 struct btrfs_extent_data_ref *ref;
8069 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
8070 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
8071 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
8072 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
8073 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
8076 btrfs_mark_buffer_dirty(path->nodes[0]);
8077 btrfs_free_path(path);
8079 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
8084 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
8085 if (ret) { /* -ENOENT, logic error */
8086 btrfs_err(fs_info, "update block group failed for %llu %llu",
8087 ins->objectid, ins->offset);
8090 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
8094 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
8095 struct btrfs_root *root,
8096 u64 parent, u64 root_objectid,
8097 u64 flags, struct btrfs_disk_key *key,
8098 int level, struct btrfs_key *ins)
8101 struct btrfs_fs_info *fs_info = root->fs_info;
8102 struct btrfs_extent_item *extent_item;
8103 struct btrfs_tree_block_info *block_info;
8104 struct btrfs_extent_inline_ref *iref;
8105 struct btrfs_path *path;
8106 struct extent_buffer *leaf;
8107 u32 size = sizeof(*extent_item) + sizeof(*iref);
8108 u64 num_bytes = ins->offset;
8109 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
8112 if (!skinny_metadata)
8113 size += sizeof(*block_info);
8115 path = btrfs_alloc_path();
8117 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
8122 path->leave_spinning = 1;
8123 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8126 btrfs_free_path(path);
8127 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
8132 leaf = path->nodes[0];
8133 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8134 struct btrfs_extent_item);
8135 btrfs_set_extent_refs(leaf, extent_item, 1);
8136 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8137 btrfs_set_extent_flags(leaf, extent_item,
8138 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
8140 if (skinny_metadata) {
8141 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8142 num_bytes = root->nodesize;
8144 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
8145 btrfs_set_tree_block_key(leaf, block_info, key);
8146 btrfs_set_tree_block_level(leaf, block_info, level);
8147 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
8151 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
8152 btrfs_set_extent_inline_ref_type(leaf, iref,
8153 BTRFS_SHARED_BLOCK_REF_KEY);
8154 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8156 btrfs_set_extent_inline_ref_type(leaf, iref,
8157 BTRFS_TREE_BLOCK_REF_KEY);
8158 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
8161 btrfs_mark_buffer_dirty(leaf);
8162 btrfs_free_path(path);
8164 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
8169 ret = update_block_group(trans, root, ins->objectid, root->nodesize,
8171 if (ret) { /* -ENOENT, logic error */
8172 btrfs_err(fs_info, "update block group failed for %llu %llu",
8173 ins->objectid, ins->offset);
8177 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize);
8181 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8182 struct btrfs_root *root,
8183 u64 root_objectid, u64 owner,
8184 u64 offset, u64 ram_bytes,
8185 struct btrfs_key *ins)
8189 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
8191 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
8193 root_objectid, owner, offset,
8194 ram_bytes, BTRFS_ADD_DELAYED_EXTENT,
8200 * this is used by the tree logging recovery code. It records that
8201 * an extent has been allocated and makes sure to clear the free
8202 * space cache bits as well
8204 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
8205 struct btrfs_root *root,
8206 u64 root_objectid, u64 owner, u64 offset,
8207 struct btrfs_key *ins)
8210 struct btrfs_block_group_cache *block_group;
8211 struct btrfs_space_info *space_info;
8214 * Mixed block groups will exclude before processing the log so we only
8215 * need to do the exclude dance if this fs isn't mixed.
8217 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
8218 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
8223 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
8227 space_info = block_group->space_info;
8228 spin_lock(&space_info->lock);
8229 spin_lock(&block_group->lock);
8230 space_info->bytes_reserved += ins->offset;
8231 block_group->reserved += ins->offset;
8232 spin_unlock(&block_group->lock);
8233 spin_unlock(&space_info->lock);
8235 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
8236 0, owner, offset, ins, 1);
8237 btrfs_put_block_group(block_group);
8241 static struct extent_buffer *
8242 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
8243 u64 bytenr, int level)
8245 struct extent_buffer *buf;
8247 buf = btrfs_find_create_tree_block(root, bytenr);
8251 btrfs_set_header_generation(buf, trans->transid);
8252 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
8253 btrfs_tree_lock(buf);
8254 clean_tree_block(trans, root->fs_info, buf);
8255 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
8257 btrfs_set_lock_blocking(buf);
8258 set_extent_buffer_uptodate(buf);
8260 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
8261 buf->log_index = root->log_transid % 2;
8263 * we allow two log transactions at a time, use different
8264 * EXENT bit to differentiate dirty pages.
8266 if (buf->log_index == 0)
8267 set_extent_dirty(&root->dirty_log_pages, buf->start,
8268 buf->start + buf->len - 1, GFP_NOFS);
8270 set_extent_new(&root->dirty_log_pages, buf->start,
8271 buf->start + buf->len - 1);
8273 buf->log_index = -1;
8274 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
8275 buf->start + buf->len - 1, GFP_NOFS);
8277 trans->dirty = true;
8278 /* this returns a buffer locked for blocking */
8282 static struct btrfs_block_rsv *
8283 use_block_rsv(struct btrfs_trans_handle *trans,
8284 struct btrfs_root *root, u32 blocksize)
8286 struct btrfs_block_rsv *block_rsv;
8287 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
8289 bool global_updated = false;
8291 block_rsv = get_block_rsv(trans, root);
8293 if (unlikely(block_rsv->size == 0))
8296 ret = block_rsv_use_bytes(block_rsv, blocksize);
8300 if (block_rsv->failfast)
8301 return ERR_PTR(ret);
8303 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
8304 global_updated = true;
8305 update_global_block_rsv(root->fs_info);
8309 if (btrfs_test_opt(root->fs_info, ENOSPC_DEBUG)) {
8310 static DEFINE_RATELIMIT_STATE(_rs,
8311 DEFAULT_RATELIMIT_INTERVAL * 10,
8312 /*DEFAULT_RATELIMIT_BURST*/ 1);
8313 if (__ratelimit(&_rs))
8315 "BTRFS: block rsv returned %d\n", ret);
8318 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
8319 BTRFS_RESERVE_NO_FLUSH);
8323 * If we couldn't reserve metadata bytes try and use some from
8324 * the global reserve if its space type is the same as the global
8327 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
8328 block_rsv->space_info == global_rsv->space_info) {
8329 ret = block_rsv_use_bytes(global_rsv, blocksize);
8333 return ERR_PTR(ret);
8336 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
8337 struct btrfs_block_rsv *block_rsv, u32 blocksize)
8339 block_rsv_add_bytes(block_rsv, blocksize, 0);
8340 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
8344 * finds a free extent and does all the dirty work required for allocation
8345 * returns the tree buffer or an ERR_PTR on error.
8347 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
8348 struct btrfs_root *root,
8349 u64 parent, u64 root_objectid,
8350 struct btrfs_disk_key *key, int level,
8351 u64 hint, u64 empty_size)
8353 struct btrfs_key ins;
8354 struct btrfs_block_rsv *block_rsv;
8355 struct extent_buffer *buf;
8356 struct btrfs_delayed_extent_op *extent_op;
8359 u32 blocksize = root->nodesize;
8360 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
8363 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8364 if (btrfs_is_testing(root->fs_info)) {
8365 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
8368 root->alloc_bytenr += blocksize;
8373 block_rsv = use_block_rsv(trans, root, blocksize);
8374 if (IS_ERR(block_rsv))
8375 return ERR_CAST(block_rsv);
8377 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
8378 empty_size, hint, &ins, 0, 0);
8382 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
8385 goto out_free_reserved;
8388 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
8390 parent = ins.objectid;
8391 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
8395 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
8396 extent_op = btrfs_alloc_delayed_extent_op();
8402 memcpy(&extent_op->key, key, sizeof(extent_op->key));
8404 memset(&extent_op->key, 0, sizeof(extent_op->key));
8405 extent_op->flags_to_set = flags;
8406 extent_op->update_key = skinny_metadata ? false : true;
8407 extent_op->update_flags = true;
8408 extent_op->is_data = false;
8409 extent_op->level = level;
8411 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
8412 ins.objectid, ins.offset,
8413 parent, root_objectid, level,
8414 BTRFS_ADD_DELAYED_EXTENT,
8417 goto out_free_delayed;
8422 btrfs_free_delayed_extent_op(extent_op);
8424 free_extent_buffer(buf);
8426 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 0);
8428 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
8429 return ERR_PTR(ret);
8432 struct walk_control {
8433 u64 refs[BTRFS_MAX_LEVEL];
8434 u64 flags[BTRFS_MAX_LEVEL];
8435 struct btrfs_key update_progress;
8446 #define DROP_REFERENCE 1
8447 #define UPDATE_BACKREF 2
8449 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
8450 struct btrfs_root *root,
8451 struct walk_control *wc,
8452 struct btrfs_path *path)
8459 struct btrfs_key key;
8460 struct extent_buffer *eb;
8465 if (path->slots[wc->level] < wc->reada_slot) {
8466 wc->reada_count = wc->reada_count * 2 / 3;
8467 wc->reada_count = max(wc->reada_count, 2);
8469 wc->reada_count = wc->reada_count * 3 / 2;
8470 wc->reada_count = min_t(int, wc->reada_count,
8471 BTRFS_NODEPTRS_PER_BLOCK(root));
8474 eb = path->nodes[wc->level];
8475 nritems = btrfs_header_nritems(eb);
8477 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
8478 if (nread >= wc->reada_count)
8482 bytenr = btrfs_node_blockptr(eb, slot);
8483 generation = btrfs_node_ptr_generation(eb, slot);
8485 if (slot == path->slots[wc->level])
8488 if (wc->stage == UPDATE_BACKREF &&
8489 generation <= root->root_key.offset)
8492 /* We don't lock the tree block, it's OK to be racy here */
8493 ret = btrfs_lookup_extent_info(trans, root, bytenr,
8494 wc->level - 1, 1, &refs,
8496 /* We don't care about errors in readahead. */
8501 if (wc->stage == DROP_REFERENCE) {
8505 if (wc->level == 1 &&
8506 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8508 if (!wc->update_ref ||
8509 generation <= root->root_key.offset)
8511 btrfs_node_key_to_cpu(eb, &key, slot);
8512 ret = btrfs_comp_cpu_keys(&key,
8513 &wc->update_progress);
8517 if (wc->level == 1 &&
8518 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8522 readahead_tree_block(root, bytenr);
8525 wc->reada_slot = slot;
8528 static int account_leaf_items(struct btrfs_trans_handle *trans,
8529 struct btrfs_root *root,
8530 struct extent_buffer *eb)
8532 int nr = btrfs_header_nritems(eb);
8533 int i, extent_type, ret;
8534 struct btrfs_key key;
8535 struct btrfs_file_extent_item *fi;
8536 u64 bytenr, num_bytes;
8538 /* We can be called directly from walk_up_proc() */
8539 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags))
8542 for (i = 0; i < nr; i++) {
8543 btrfs_item_key_to_cpu(eb, &key, i);
8545 if (key.type != BTRFS_EXTENT_DATA_KEY)
8548 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
8549 /* filter out non qgroup-accountable extents */
8550 extent_type = btrfs_file_extent_type(eb, fi);
8552 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
8555 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
8559 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
8561 ret = btrfs_qgroup_insert_dirty_extent(trans, root->fs_info,
8562 bytenr, num_bytes, GFP_NOFS);
8570 * Walk up the tree from the bottom, freeing leaves and any interior
8571 * nodes which have had all slots visited. If a node (leaf or
8572 * interior) is freed, the node above it will have it's slot
8573 * incremented. The root node will never be freed.
8575 * At the end of this function, we should have a path which has all
8576 * slots incremented to the next position for a search. If we need to
8577 * read a new node it will be NULL and the node above it will have the
8578 * correct slot selected for a later read.
8580 * If we increment the root nodes slot counter past the number of
8581 * elements, 1 is returned to signal completion of the search.
8583 static int adjust_slots_upwards(struct btrfs_root *root,
8584 struct btrfs_path *path, int root_level)
8588 struct extent_buffer *eb;
8590 if (root_level == 0)
8593 while (level <= root_level) {
8594 eb = path->nodes[level];
8595 nr = btrfs_header_nritems(eb);
8596 path->slots[level]++;
8597 slot = path->slots[level];
8598 if (slot >= nr || level == 0) {
8600 * Don't free the root - we will detect this
8601 * condition after our loop and return a
8602 * positive value for caller to stop walking the tree.
8604 if (level != root_level) {
8605 btrfs_tree_unlock_rw(eb, path->locks[level]);
8606 path->locks[level] = 0;
8608 free_extent_buffer(eb);
8609 path->nodes[level] = NULL;
8610 path->slots[level] = 0;
8614 * We have a valid slot to walk back down
8615 * from. Stop here so caller can process these
8624 eb = path->nodes[root_level];
8625 if (path->slots[root_level] >= btrfs_header_nritems(eb))
8632 * root_eb is the subtree root and is locked before this function is called.
8634 static int account_shared_subtree(struct btrfs_trans_handle *trans,
8635 struct btrfs_root *root,
8636 struct extent_buffer *root_eb,
8642 struct extent_buffer *eb = root_eb;
8643 struct btrfs_path *path = NULL;
8645 BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
8646 BUG_ON(root_eb == NULL);
8648 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags))
8651 if (!extent_buffer_uptodate(root_eb)) {
8652 ret = btrfs_read_buffer(root_eb, root_gen);
8657 if (root_level == 0) {
8658 ret = account_leaf_items(trans, root, root_eb);
8662 path = btrfs_alloc_path();
8667 * Walk down the tree. Missing extent blocks are filled in as
8668 * we go. Metadata is accounted every time we read a new
8671 * When we reach a leaf, we account for file extent items in it,
8672 * walk back up the tree (adjusting slot pointers as we go)
8673 * and restart the search process.
8675 extent_buffer_get(root_eb); /* For path */
8676 path->nodes[root_level] = root_eb;
8677 path->slots[root_level] = 0;
8678 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
8681 while (level >= 0) {
8682 if (path->nodes[level] == NULL) {
8687 /* We need to get child blockptr/gen from
8688 * parent before we can read it. */
8689 eb = path->nodes[level + 1];
8690 parent_slot = path->slots[level + 1];
8691 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
8692 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
8694 eb = read_tree_block(root, child_bytenr, child_gen);
8698 } else if (!extent_buffer_uptodate(eb)) {
8699 free_extent_buffer(eb);
8704 path->nodes[level] = eb;
8705 path->slots[level] = 0;
8707 btrfs_tree_read_lock(eb);
8708 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
8709 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
8711 ret = btrfs_qgroup_insert_dirty_extent(trans,
8712 root->fs_info, child_bytenr,
8713 root->nodesize, GFP_NOFS);
8719 ret = account_leaf_items(trans, root, path->nodes[level]);
8723 /* Nonzero return here means we completed our search */
8724 ret = adjust_slots_upwards(root, path, root_level);
8728 /* Restart search with new slots */
8737 btrfs_free_path(path);
8743 * helper to process tree block while walking down the tree.
8745 * when wc->stage == UPDATE_BACKREF, this function updates
8746 * back refs for pointers in the block.
8748 * NOTE: return value 1 means we should stop walking down.
8750 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8751 struct btrfs_root *root,
8752 struct btrfs_path *path,
8753 struct walk_control *wc, int lookup_info)
8755 int level = wc->level;
8756 struct extent_buffer *eb = path->nodes[level];
8757 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8760 if (wc->stage == UPDATE_BACKREF &&
8761 btrfs_header_owner(eb) != root->root_key.objectid)
8765 * when reference count of tree block is 1, it won't increase
8766 * again. once full backref flag is set, we never clear it.
8769 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8770 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8771 BUG_ON(!path->locks[level]);
8772 ret = btrfs_lookup_extent_info(trans, root,
8773 eb->start, level, 1,
8776 BUG_ON(ret == -ENOMEM);
8779 BUG_ON(wc->refs[level] == 0);
8782 if (wc->stage == DROP_REFERENCE) {
8783 if (wc->refs[level] > 1)
8786 if (path->locks[level] && !wc->keep_locks) {
8787 btrfs_tree_unlock_rw(eb, path->locks[level]);
8788 path->locks[level] = 0;
8793 /* wc->stage == UPDATE_BACKREF */
8794 if (!(wc->flags[level] & flag)) {
8795 BUG_ON(!path->locks[level]);
8796 ret = btrfs_inc_ref(trans, root, eb, 1);
8797 BUG_ON(ret); /* -ENOMEM */
8798 ret = btrfs_dec_ref(trans, root, eb, 0);
8799 BUG_ON(ret); /* -ENOMEM */
8800 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
8802 btrfs_header_level(eb), 0);
8803 BUG_ON(ret); /* -ENOMEM */
8804 wc->flags[level] |= flag;
8808 * the block is shared by multiple trees, so it's not good to
8809 * keep the tree lock
8811 if (path->locks[level] && level > 0) {
8812 btrfs_tree_unlock_rw(eb, path->locks[level]);
8813 path->locks[level] = 0;
8819 * helper to process tree block pointer.
8821 * when wc->stage == DROP_REFERENCE, this function checks
8822 * reference count of the block pointed to. if the block
8823 * is shared and we need update back refs for the subtree
8824 * rooted at the block, this function changes wc->stage to
8825 * UPDATE_BACKREF. if the block is shared and there is no
8826 * need to update back, this function drops the reference
8829 * NOTE: return value 1 means we should stop walking down.
8831 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8832 struct btrfs_root *root,
8833 struct btrfs_path *path,
8834 struct walk_control *wc, int *lookup_info)
8840 struct btrfs_key key;
8841 struct extent_buffer *next;
8842 int level = wc->level;
8845 bool need_account = false;
8847 generation = btrfs_node_ptr_generation(path->nodes[level],
8848 path->slots[level]);
8850 * if the lower level block was created before the snapshot
8851 * was created, we know there is no need to update back refs
8854 if (wc->stage == UPDATE_BACKREF &&
8855 generation <= root->root_key.offset) {
8860 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8861 blocksize = root->nodesize;
8863 next = btrfs_find_tree_block(root->fs_info, bytenr);
8865 next = btrfs_find_create_tree_block(root, bytenr);
8867 return PTR_ERR(next);
8869 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8873 btrfs_tree_lock(next);
8874 btrfs_set_lock_blocking(next);
8876 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
8877 &wc->refs[level - 1],
8878 &wc->flags[level - 1]);
8880 btrfs_tree_unlock(next);
8884 if (unlikely(wc->refs[level - 1] == 0)) {
8885 btrfs_err(root->fs_info, "Missing references.");
8890 if (wc->stage == DROP_REFERENCE) {
8891 if (wc->refs[level - 1] > 1) {
8892 need_account = true;
8894 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8897 if (!wc->update_ref ||
8898 generation <= root->root_key.offset)
8901 btrfs_node_key_to_cpu(path->nodes[level], &key,
8902 path->slots[level]);
8903 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8907 wc->stage = UPDATE_BACKREF;
8908 wc->shared_level = level - 1;
8912 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8916 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8917 btrfs_tree_unlock(next);
8918 free_extent_buffer(next);
8924 if (reada && level == 1)
8925 reada_walk_down(trans, root, wc, path);
8926 next = read_tree_block(root, bytenr, generation);
8928 return PTR_ERR(next);
8929 } else if (!extent_buffer_uptodate(next)) {
8930 free_extent_buffer(next);
8933 btrfs_tree_lock(next);
8934 btrfs_set_lock_blocking(next);
8938 BUG_ON(level != btrfs_header_level(next));
8939 path->nodes[level] = next;
8940 path->slots[level] = 0;
8941 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8947 wc->refs[level - 1] = 0;
8948 wc->flags[level - 1] = 0;
8949 if (wc->stage == DROP_REFERENCE) {
8950 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8951 parent = path->nodes[level]->start;
8953 BUG_ON(root->root_key.objectid !=
8954 btrfs_header_owner(path->nodes[level]));
8959 ret = account_shared_subtree(trans, root, next,
8960 generation, level - 1);
8962 btrfs_err_rl(root->fs_info,
8964 "%d accounting shared subtree. Quota "
8965 "is out of sync, rescan required.",
8969 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
8970 root->root_key.objectid, level - 1, 0);
8971 BUG_ON(ret); /* -ENOMEM */
8973 btrfs_tree_unlock(next);
8974 free_extent_buffer(next);
8980 * helper to process tree block while walking up the tree.
8982 * when wc->stage == DROP_REFERENCE, this function drops
8983 * reference count on the block.
8985 * when wc->stage == UPDATE_BACKREF, this function changes
8986 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8987 * to UPDATE_BACKREF previously while processing the block.
8989 * NOTE: return value 1 means we should stop walking up.
8991 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8992 struct btrfs_root *root,
8993 struct btrfs_path *path,
8994 struct walk_control *wc)
8997 int level = wc->level;
8998 struct extent_buffer *eb = path->nodes[level];
9001 if (wc->stage == UPDATE_BACKREF) {
9002 BUG_ON(wc->shared_level < level);
9003 if (level < wc->shared_level)
9006 ret = find_next_key(path, level + 1, &wc->update_progress);
9010 wc->stage = DROP_REFERENCE;
9011 wc->shared_level = -1;
9012 path->slots[level] = 0;
9015 * check reference count again if the block isn't locked.
9016 * we should start walking down the tree again if reference
9019 if (!path->locks[level]) {
9021 btrfs_tree_lock(eb);
9022 btrfs_set_lock_blocking(eb);
9023 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9025 ret = btrfs_lookup_extent_info(trans, root,
9026 eb->start, level, 1,
9030 btrfs_tree_unlock_rw(eb, path->locks[level]);
9031 path->locks[level] = 0;
9034 BUG_ON(wc->refs[level] == 0);
9035 if (wc->refs[level] == 1) {
9036 btrfs_tree_unlock_rw(eb, path->locks[level]);
9037 path->locks[level] = 0;
9043 /* wc->stage == DROP_REFERENCE */
9044 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
9046 if (wc->refs[level] == 1) {
9048 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9049 ret = btrfs_dec_ref(trans, root, eb, 1);
9051 ret = btrfs_dec_ref(trans, root, eb, 0);
9052 BUG_ON(ret); /* -ENOMEM */
9053 ret = account_leaf_items(trans, root, eb);
9055 btrfs_err_rl(root->fs_info,
9057 "%d accounting leaf items. Quota "
9058 "is out of sync, rescan required.",
9062 /* make block locked assertion in clean_tree_block happy */
9063 if (!path->locks[level] &&
9064 btrfs_header_generation(eb) == trans->transid) {
9065 btrfs_tree_lock(eb);
9066 btrfs_set_lock_blocking(eb);
9067 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9069 clean_tree_block(trans, root->fs_info, eb);
9072 if (eb == root->node) {
9073 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9076 BUG_ON(root->root_key.objectid !=
9077 btrfs_header_owner(eb));
9079 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9080 parent = path->nodes[level + 1]->start;
9082 BUG_ON(root->root_key.objectid !=
9083 btrfs_header_owner(path->nodes[level + 1]));
9086 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
9088 wc->refs[level] = 0;
9089 wc->flags[level] = 0;
9093 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
9094 struct btrfs_root *root,
9095 struct btrfs_path *path,
9096 struct walk_control *wc)
9098 int level = wc->level;
9099 int lookup_info = 1;
9102 while (level >= 0) {
9103 ret = walk_down_proc(trans, root, path, wc, lookup_info);
9110 if (path->slots[level] >=
9111 btrfs_header_nritems(path->nodes[level]))
9114 ret = do_walk_down(trans, root, path, wc, &lookup_info);
9116 path->slots[level]++;
9125 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
9126 struct btrfs_root *root,
9127 struct btrfs_path *path,
9128 struct walk_control *wc, int max_level)
9130 int level = wc->level;
9133 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
9134 while (level < max_level && path->nodes[level]) {
9136 if (path->slots[level] + 1 <
9137 btrfs_header_nritems(path->nodes[level])) {
9138 path->slots[level]++;
9141 ret = walk_up_proc(trans, root, path, wc);
9145 if (path->locks[level]) {
9146 btrfs_tree_unlock_rw(path->nodes[level],
9147 path->locks[level]);
9148 path->locks[level] = 0;
9150 free_extent_buffer(path->nodes[level]);
9151 path->nodes[level] = NULL;
9159 * drop a subvolume tree.
9161 * this function traverses the tree freeing any blocks that only
9162 * referenced by the tree.
9164 * when a shared tree block is found. this function decreases its
9165 * reference count by one. if update_ref is true, this function
9166 * also make sure backrefs for the shared block and all lower level
9167 * blocks are properly updated.
9169 * If called with for_reloc == 0, may exit early with -EAGAIN
9171 int btrfs_drop_snapshot(struct btrfs_root *root,
9172 struct btrfs_block_rsv *block_rsv, int update_ref,
9175 struct btrfs_path *path;
9176 struct btrfs_trans_handle *trans;
9177 struct btrfs_root *tree_root = root->fs_info->tree_root;
9178 struct btrfs_root_item *root_item = &root->root_item;
9179 struct walk_control *wc;
9180 struct btrfs_key key;
9184 bool root_dropped = false;
9186 btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid);
9188 path = btrfs_alloc_path();
9194 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9196 btrfs_free_path(path);
9201 trans = btrfs_start_transaction(tree_root, 0);
9202 if (IS_ERR(trans)) {
9203 err = PTR_ERR(trans);
9208 trans->block_rsv = block_rsv;
9210 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
9211 level = btrfs_header_level(root->node);
9212 path->nodes[level] = btrfs_lock_root_node(root);
9213 btrfs_set_lock_blocking(path->nodes[level]);
9214 path->slots[level] = 0;
9215 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9216 memset(&wc->update_progress, 0,
9217 sizeof(wc->update_progress));
9219 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
9220 memcpy(&wc->update_progress, &key,
9221 sizeof(wc->update_progress));
9223 level = root_item->drop_level;
9225 path->lowest_level = level;
9226 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
9227 path->lowest_level = 0;
9235 * unlock our path, this is safe because only this
9236 * function is allowed to delete this snapshot
9238 btrfs_unlock_up_safe(path, 0);
9240 level = btrfs_header_level(root->node);
9242 btrfs_tree_lock(path->nodes[level]);
9243 btrfs_set_lock_blocking(path->nodes[level]);
9244 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9246 ret = btrfs_lookup_extent_info(trans, root,
9247 path->nodes[level]->start,
9248 level, 1, &wc->refs[level],
9254 BUG_ON(wc->refs[level] == 0);
9256 if (level == root_item->drop_level)
9259 btrfs_tree_unlock(path->nodes[level]);
9260 path->locks[level] = 0;
9261 WARN_ON(wc->refs[level] != 1);
9267 wc->shared_level = -1;
9268 wc->stage = DROP_REFERENCE;
9269 wc->update_ref = update_ref;
9271 wc->for_reloc = for_reloc;
9272 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
9276 ret = walk_down_tree(trans, root, path, wc);
9282 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
9289 BUG_ON(wc->stage != DROP_REFERENCE);
9293 if (wc->stage == DROP_REFERENCE) {
9295 btrfs_node_key(path->nodes[level],
9296 &root_item->drop_progress,
9297 path->slots[level]);
9298 root_item->drop_level = level;
9301 BUG_ON(wc->level == 0);
9302 if (btrfs_should_end_transaction(trans, tree_root) ||
9303 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
9304 ret = btrfs_update_root(trans, tree_root,
9308 btrfs_abort_transaction(trans, ret);
9313 btrfs_end_transaction_throttle(trans, tree_root);
9314 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
9315 pr_debug("BTRFS: drop snapshot early exit\n");
9320 trans = btrfs_start_transaction(tree_root, 0);
9321 if (IS_ERR(trans)) {
9322 err = PTR_ERR(trans);
9326 trans->block_rsv = block_rsv;
9329 btrfs_release_path(path);
9333 ret = btrfs_del_root(trans, tree_root, &root->root_key);
9335 btrfs_abort_transaction(trans, ret);
9339 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
9340 ret = btrfs_find_root(tree_root, &root->root_key, path,
9343 btrfs_abort_transaction(trans, ret);
9346 } else if (ret > 0) {
9347 /* if we fail to delete the orphan item this time
9348 * around, it'll get picked up the next time.
9350 * The most common failure here is just -ENOENT.
9352 btrfs_del_orphan_item(trans, tree_root,
9353 root->root_key.objectid);
9357 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
9358 btrfs_add_dropped_root(trans, root);
9360 free_extent_buffer(root->node);
9361 free_extent_buffer(root->commit_root);
9362 btrfs_put_fs_root(root);
9364 root_dropped = true;
9366 btrfs_end_transaction_throttle(trans, tree_root);
9369 btrfs_free_path(path);
9372 * So if we need to stop dropping the snapshot for whatever reason we
9373 * need to make sure to add it back to the dead root list so that we
9374 * keep trying to do the work later. This also cleans up roots if we
9375 * don't have it in the radix (like when we recover after a power fail
9376 * or unmount) so we don't leak memory.
9378 if (!for_reloc && root_dropped == false)
9379 btrfs_add_dead_root(root);
9380 if (err && err != -EAGAIN)
9381 btrfs_handle_fs_error(root->fs_info, err, NULL);
9386 * drop subtree rooted at tree block 'node'.
9388 * NOTE: this function will unlock and release tree block 'node'
9389 * only used by relocation code
9391 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
9392 struct btrfs_root *root,
9393 struct extent_buffer *node,
9394 struct extent_buffer *parent)
9396 struct btrfs_path *path;
9397 struct walk_control *wc;
9403 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
9405 path = btrfs_alloc_path();
9409 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9411 btrfs_free_path(path);
9415 btrfs_assert_tree_locked(parent);
9416 parent_level = btrfs_header_level(parent);
9417 extent_buffer_get(parent);
9418 path->nodes[parent_level] = parent;
9419 path->slots[parent_level] = btrfs_header_nritems(parent);
9421 btrfs_assert_tree_locked(node);
9422 level = btrfs_header_level(node);
9423 path->nodes[level] = node;
9424 path->slots[level] = 0;
9425 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9427 wc->refs[parent_level] = 1;
9428 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
9430 wc->shared_level = -1;
9431 wc->stage = DROP_REFERENCE;
9435 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
9438 wret = walk_down_tree(trans, root, path, wc);
9444 wret = walk_up_tree(trans, root, path, wc, parent_level);
9452 btrfs_free_path(path);
9456 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
9462 * if restripe for this chunk_type is on pick target profile and
9463 * return, otherwise do the usual balance
9465 stripped = get_restripe_target(root->fs_info, flags);
9467 return extended_to_chunk(stripped);
9469 num_devices = root->fs_info->fs_devices->rw_devices;
9471 stripped = BTRFS_BLOCK_GROUP_RAID0 |
9472 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
9473 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
9475 if (num_devices == 1) {
9476 stripped |= BTRFS_BLOCK_GROUP_DUP;
9477 stripped = flags & ~stripped;
9479 /* turn raid0 into single device chunks */
9480 if (flags & BTRFS_BLOCK_GROUP_RAID0)
9483 /* turn mirroring into duplication */
9484 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
9485 BTRFS_BLOCK_GROUP_RAID10))
9486 return stripped | BTRFS_BLOCK_GROUP_DUP;
9488 /* they already had raid on here, just return */
9489 if (flags & stripped)
9492 stripped |= BTRFS_BLOCK_GROUP_DUP;
9493 stripped = flags & ~stripped;
9495 /* switch duplicated blocks with raid1 */
9496 if (flags & BTRFS_BLOCK_GROUP_DUP)
9497 return stripped | BTRFS_BLOCK_GROUP_RAID1;
9499 /* this is drive concat, leave it alone */
9505 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
9507 struct btrfs_space_info *sinfo = cache->space_info;
9509 u64 min_allocable_bytes;
9513 * We need some metadata space and system metadata space for
9514 * allocating chunks in some corner cases until we force to set
9515 * it to be readonly.
9518 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
9520 min_allocable_bytes = SZ_1M;
9522 min_allocable_bytes = 0;
9524 spin_lock(&sinfo->lock);
9525 spin_lock(&cache->lock);
9533 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
9534 cache->bytes_super - btrfs_block_group_used(&cache->item);
9536 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
9537 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
9538 min_allocable_bytes <= sinfo->total_bytes) {
9539 sinfo->bytes_readonly += num_bytes;
9541 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
9545 spin_unlock(&cache->lock);
9546 spin_unlock(&sinfo->lock);
9550 int btrfs_inc_block_group_ro(struct btrfs_root *root,
9551 struct btrfs_block_group_cache *cache)
9554 struct btrfs_trans_handle *trans;
9559 trans = btrfs_join_transaction(root);
9561 return PTR_ERR(trans);
9564 * we're not allowed to set block groups readonly after the dirty
9565 * block groups cache has started writing. If it already started,
9566 * back off and let this transaction commit
9568 mutex_lock(&root->fs_info->ro_block_group_mutex);
9569 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9570 u64 transid = trans->transid;
9572 mutex_unlock(&root->fs_info->ro_block_group_mutex);
9573 btrfs_end_transaction(trans, root);
9575 ret = btrfs_wait_for_commit(root, transid);
9582 * if we are changing raid levels, try to allocate a corresponding
9583 * block group with the new raid level.
9585 alloc_flags = update_block_group_flags(root, cache->flags);
9586 if (alloc_flags != cache->flags) {
9587 ret = do_chunk_alloc(trans, root, alloc_flags,
9590 * ENOSPC is allowed here, we may have enough space
9591 * already allocated at the new raid level to
9600 ret = inc_block_group_ro(cache, 0);
9603 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
9604 ret = do_chunk_alloc(trans, root, alloc_flags,
9608 ret = inc_block_group_ro(cache, 0);
9610 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9611 alloc_flags = update_block_group_flags(root, cache->flags);
9612 lock_chunks(root->fs_info->chunk_root);
9613 check_system_chunk(trans, root, alloc_flags);
9614 unlock_chunks(root->fs_info->chunk_root);
9616 mutex_unlock(&root->fs_info->ro_block_group_mutex);
9618 btrfs_end_transaction(trans, root);
9622 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
9623 struct btrfs_root *root, u64 type)
9625 u64 alloc_flags = get_alloc_profile(root, type);
9626 return do_chunk_alloc(trans, root, alloc_flags,
9631 * helper to account the unused space of all the readonly block group in the
9632 * space_info. takes mirrors into account.
9634 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9636 struct btrfs_block_group_cache *block_group;
9640 /* It's df, we don't care if it's racy */
9641 if (list_empty(&sinfo->ro_bgs))
9644 spin_lock(&sinfo->lock);
9645 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9646 spin_lock(&block_group->lock);
9648 if (!block_group->ro) {
9649 spin_unlock(&block_group->lock);
9653 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
9654 BTRFS_BLOCK_GROUP_RAID10 |
9655 BTRFS_BLOCK_GROUP_DUP))
9660 free_bytes += (block_group->key.offset -
9661 btrfs_block_group_used(&block_group->item)) *
9664 spin_unlock(&block_group->lock);
9666 spin_unlock(&sinfo->lock);
9671 void btrfs_dec_block_group_ro(struct btrfs_root *root,
9672 struct btrfs_block_group_cache *cache)
9674 struct btrfs_space_info *sinfo = cache->space_info;
9679 spin_lock(&sinfo->lock);
9680 spin_lock(&cache->lock);
9682 num_bytes = cache->key.offset - cache->reserved -
9683 cache->pinned - cache->bytes_super -
9684 btrfs_block_group_used(&cache->item);
9685 sinfo->bytes_readonly -= num_bytes;
9686 list_del_init(&cache->ro_list);
9688 spin_unlock(&cache->lock);
9689 spin_unlock(&sinfo->lock);
9693 * checks to see if its even possible to relocate this block group.
9695 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9696 * ok to go ahead and try.
9698 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
9700 struct btrfs_block_group_cache *block_group;
9701 struct btrfs_space_info *space_info;
9702 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
9703 struct btrfs_device *device;
9704 struct btrfs_trans_handle *trans;
9714 debug = btrfs_test_opt(root->fs_info, ENOSPC_DEBUG);
9716 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
9718 /* odd, couldn't find the block group, leave it alone */
9721 btrfs_warn(root->fs_info,
9722 "can't find block group for bytenr %llu",
9727 min_free = btrfs_block_group_used(&block_group->item);
9729 /* no bytes used, we're good */
9733 space_info = block_group->space_info;
9734 spin_lock(&space_info->lock);
9736 full = space_info->full;
9739 * if this is the last block group we have in this space, we can't
9740 * relocate it unless we're able to allocate a new chunk below.
9742 * Otherwise, we need to make sure we have room in the space to handle
9743 * all of the extents from this block group. If we can, we're good
9745 if ((space_info->total_bytes != block_group->key.offset) &&
9746 (space_info->bytes_used + space_info->bytes_reserved +
9747 space_info->bytes_pinned + space_info->bytes_readonly +
9748 min_free < space_info->total_bytes)) {
9749 spin_unlock(&space_info->lock);
9752 spin_unlock(&space_info->lock);
9755 * ok we don't have enough space, but maybe we have free space on our
9756 * devices to allocate new chunks for relocation, so loop through our
9757 * alloc devices and guess if we have enough space. if this block
9758 * group is going to be restriped, run checks against the target
9759 * profile instead of the current one.
9771 target = get_restripe_target(root->fs_info, block_group->flags);
9773 index = __get_raid_index(extended_to_chunk(target));
9776 * this is just a balance, so if we were marked as full
9777 * we know there is no space for a new chunk
9781 btrfs_warn(root->fs_info,
9782 "no space to alloc new chunk for block group %llu",
9783 block_group->key.objectid);
9787 index = get_block_group_index(block_group);
9790 if (index == BTRFS_RAID_RAID10) {
9794 } else if (index == BTRFS_RAID_RAID1) {
9796 } else if (index == BTRFS_RAID_DUP) {
9799 } else if (index == BTRFS_RAID_RAID0) {
9800 dev_min = fs_devices->rw_devices;
9801 min_free = div64_u64(min_free, dev_min);
9804 /* We need to do this so that we can look at pending chunks */
9805 trans = btrfs_join_transaction(root);
9806 if (IS_ERR(trans)) {
9807 ret = PTR_ERR(trans);
9811 mutex_lock(&root->fs_info->chunk_mutex);
9812 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9816 * check to make sure we can actually find a chunk with enough
9817 * space to fit our block group in.
9819 if (device->total_bytes > device->bytes_used + min_free &&
9820 !device->is_tgtdev_for_dev_replace) {
9821 ret = find_free_dev_extent(trans, device, min_free,
9826 if (dev_nr >= dev_min)
9832 if (debug && ret == -1)
9833 btrfs_warn(root->fs_info,
9834 "no space to allocate a new chunk for block group %llu",
9835 block_group->key.objectid);
9836 mutex_unlock(&root->fs_info->chunk_mutex);
9837 btrfs_end_transaction(trans, root);
9839 btrfs_put_block_group(block_group);
9843 static int find_first_block_group(struct btrfs_root *root,
9844 struct btrfs_path *path, struct btrfs_key *key)
9847 struct btrfs_key found_key;
9848 struct extent_buffer *leaf;
9851 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9856 slot = path->slots[0];
9857 leaf = path->nodes[0];
9858 if (slot >= btrfs_header_nritems(leaf)) {
9859 ret = btrfs_next_leaf(root, path);
9866 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9868 if (found_key.objectid >= key->objectid &&
9869 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9870 struct extent_map_tree *em_tree;
9871 struct extent_map *em;
9873 em_tree = &root->fs_info->mapping_tree.map_tree;
9874 read_lock(&em_tree->lock);
9875 em = lookup_extent_mapping(em_tree, found_key.objectid,
9877 read_unlock(&em_tree->lock);
9879 btrfs_err(root->fs_info,
9880 "logical %llu len %llu found bg but no related chunk",
9881 found_key.objectid, found_key.offset);
9886 free_extent_map(em);
9895 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9897 struct btrfs_block_group_cache *block_group;
9901 struct inode *inode;
9903 block_group = btrfs_lookup_first_block_group(info, last);
9904 while (block_group) {
9905 spin_lock(&block_group->lock);
9906 if (block_group->iref)
9908 spin_unlock(&block_group->lock);
9909 block_group = next_block_group(info->tree_root,
9919 inode = block_group->inode;
9920 block_group->iref = 0;
9921 block_group->inode = NULL;
9922 spin_unlock(&block_group->lock);
9923 ASSERT(block_group->io_ctl.inode == NULL);
9925 last = block_group->key.objectid + block_group->key.offset;
9926 btrfs_put_block_group(block_group);
9930 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9932 struct btrfs_block_group_cache *block_group;
9933 struct btrfs_space_info *space_info;
9934 struct btrfs_caching_control *caching_ctl;
9937 down_write(&info->commit_root_sem);
9938 while (!list_empty(&info->caching_block_groups)) {
9939 caching_ctl = list_entry(info->caching_block_groups.next,
9940 struct btrfs_caching_control, list);
9941 list_del(&caching_ctl->list);
9942 put_caching_control(caching_ctl);
9944 up_write(&info->commit_root_sem);
9946 spin_lock(&info->unused_bgs_lock);
9947 while (!list_empty(&info->unused_bgs)) {
9948 block_group = list_first_entry(&info->unused_bgs,
9949 struct btrfs_block_group_cache,
9951 list_del_init(&block_group->bg_list);
9952 btrfs_put_block_group(block_group);
9954 spin_unlock(&info->unused_bgs_lock);
9956 spin_lock(&info->block_group_cache_lock);
9957 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9958 block_group = rb_entry(n, struct btrfs_block_group_cache,
9960 rb_erase(&block_group->cache_node,
9961 &info->block_group_cache_tree);
9962 RB_CLEAR_NODE(&block_group->cache_node);
9963 spin_unlock(&info->block_group_cache_lock);
9965 down_write(&block_group->space_info->groups_sem);
9966 list_del(&block_group->list);
9967 up_write(&block_group->space_info->groups_sem);
9969 if (block_group->cached == BTRFS_CACHE_STARTED)
9970 wait_block_group_cache_done(block_group);
9973 * We haven't cached this block group, which means we could
9974 * possibly have excluded extents on this block group.
9976 if (block_group->cached == BTRFS_CACHE_NO ||
9977 block_group->cached == BTRFS_CACHE_ERROR)
9978 free_excluded_extents(info->extent_root, block_group);
9980 btrfs_remove_free_space_cache(block_group);
9981 ASSERT(list_empty(&block_group->dirty_list));
9982 ASSERT(list_empty(&block_group->io_list));
9983 ASSERT(list_empty(&block_group->bg_list));
9984 ASSERT(atomic_read(&block_group->count) == 1);
9985 btrfs_put_block_group(block_group);
9987 spin_lock(&info->block_group_cache_lock);
9989 spin_unlock(&info->block_group_cache_lock);
9991 /* now that all the block groups are freed, go through and
9992 * free all the space_info structs. This is only called during
9993 * the final stages of unmount, and so we know nobody is
9994 * using them. We call synchronize_rcu() once before we start,
9995 * just to be on the safe side.
9999 release_global_block_rsv(info);
10001 while (!list_empty(&info->space_info)) {
10004 space_info = list_entry(info->space_info.next,
10005 struct btrfs_space_info,
10009 * Do not hide this behind enospc_debug, this is actually
10010 * important and indicates a real bug if this happens.
10012 if (WARN_ON(space_info->bytes_pinned > 0 ||
10013 space_info->bytes_reserved > 0 ||
10014 space_info->bytes_may_use > 0))
10015 dump_space_info(space_info, 0, 0);
10016 list_del(&space_info->list);
10017 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
10018 struct kobject *kobj;
10019 kobj = space_info->block_group_kobjs[i];
10020 space_info->block_group_kobjs[i] = NULL;
10026 kobject_del(&space_info->kobj);
10027 kobject_put(&space_info->kobj);
10032 static void __link_block_group(struct btrfs_space_info *space_info,
10033 struct btrfs_block_group_cache *cache)
10035 int index = get_block_group_index(cache);
10036 bool first = false;
10038 down_write(&space_info->groups_sem);
10039 if (list_empty(&space_info->block_groups[index]))
10041 list_add_tail(&cache->list, &space_info->block_groups[index]);
10042 up_write(&space_info->groups_sem);
10045 struct raid_kobject *rkobj;
10048 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
10051 rkobj->raid_type = index;
10052 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
10053 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
10054 "%s", get_raid_name(index));
10056 kobject_put(&rkobj->kobj);
10059 space_info->block_group_kobjs[index] = &rkobj->kobj;
10064 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
10067 static struct btrfs_block_group_cache *
10068 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
10070 struct btrfs_block_group_cache *cache;
10072 cache = kzalloc(sizeof(*cache), GFP_NOFS);
10076 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
10078 if (!cache->free_space_ctl) {
10083 cache->key.objectid = start;
10084 cache->key.offset = size;
10085 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10087 cache->sectorsize = root->sectorsize;
10088 cache->fs_info = root->fs_info;
10089 cache->full_stripe_len = btrfs_full_stripe_len(root,
10090 &root->fs_info->mapping_tree,
10092 set_free_space_tree_thresholds(cache);
10094 atomic_set(&cache->count, 1);
10095 spin_lock_init(&cache->lock);
10096 init_rwsem(&cache->data_rwsem);
10097 INIT_LIST_HEAD(&cache->list);
10098 INIT_LIST_HEAD(&cache->cluster_list);
10099 INIT_LIST_HEAD(&cache->bg_list);
10100 INIT_LIST_HEAD(&cache->ro_list);
10101 INIT_LIST_HEAD(&cache->dirty_list);
10102 INIT_LIST_HEAD(&cache->io_list);
10103 btrfs_init_free_space_ctl(cache);
10104 atomic_set(&cache->trimming, 0);
10105 mutex_init(&cache->free_space_lock);
10110 int btrfs_read_block_groups(struct btrfs_root *root)
10112 struct btrfs_path *path;
10114 struct btrfs_block_group_cache *cache;
10115 struct btrfs_fs_info *info = root->fs_info;
10116 struct btrfs_space_info *space_info;
10117 struct btrfs_key key;
10118 struct btrfs_key found_key;
10119 struct extent_buffer *leaf;
10120 int need_clear = 0;
10125 feature = btrfs_super_incompat_flags(info->super_copy);
10126 mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS);
10128 root = info->extent_root;
10131 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10132 path = btrfs_alloc_path();
10135 path->reada = READA_FORWARD;
10137 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
10138 if (btrfs_test_opt(root->fs_info, SPACE_CACHE) &&
10139 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
10141 if (btrfs_test_opt(root->fs_info, CLEAR_CACHE))
10145 ret = find_first_block_group(root, path, &key);
10151 leaf = path->nodes[0];
10152 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
10154 cache = btrfs_create_block_group_cache(root, found_key.objectid,
10163 * When we mount with old space cache, we need to
10164 * set BTRFS_DC_CLEAR and set dirty flag.
10166 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
10167 * truncate the old free space cache inode and
10169 * b) Setting 'dirty flag' makes sure that we flush
10170 * the new space cache info onto disk.
10172 if (btrfs_test_opt(root->fs_info, SPACE_CACHE))
10173 cache->disk_cache_state = BTRFS_DC_CLEAR;
10176 read_extent_buffer(leaf, &cache->item,
10177 btrfs_item_ptr_offset(leaf, path->slots[0]),
10178 sizeof(cache->item));
10179 cache->flags = btrfs_block_group_flags(&cache->item);
10181 ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) &&
10182 (cache->flags & BTRFS_BLOCK_GROUP_DATA))) {
10184 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
10185 cache->key.objectid);
10190 key.objectid = found_key.objectid + found_key.offset;
10191 btrfs_release_path(path);
10194 * We need to exclude the super stripes now so that the space
10195 * info has super bytes accounted for, otherwise we'll think
10196 * we have more space than we actually do.
10198 ret = exclude_super_stripes(root, cache);
10201 * We may have excluded something, so call this just in
10204 free_excluded_extents(root, cache);
10205 btrfs_put_block_group(cache);
10210 * check for two cases, either we are full, and therefore
10211 * don't need to bother with the caching work since we won't
10212 * find any space, or we are empty, and we can just add all
10213 * the space in and be done with it. This saves us _alot_ of
10214 * time, particularly in the full case.
10216 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
10217 cache->last_byte_to_unpin = (u64)-1;
10218 cache->cached = BTRFS_CACHE_FINISHED;
10219 free_excluded_extents(root, cache);
10220 } else if (btrfs_block_group_used(&cache->item) == 0) {
10221 cache->last_byte_to_unpin = (u64)-1;
10222 cache->cached = BTRFS_CACHE_FINISHED;
10223 add_new_free_space(cache, root->fs_info,
10224 found_key.objectid,
10225 found_key.objectid +
10227 free_excluded_extents(root, cache);
10230 ret = btrfs_add_block_group_cache(root->fs_info, cache);
10232 btrfs_remove_free_space_cache(cache);
10233 btrfs_put_block_group(cache);
10237 trace_btrfs_add_block_group(root->fs_info, cache, 0);
10238 ret = update_space_info(info, cache->flags, found_key.offset,
10239 btrfs_block_group_used(&cache->item),
10240 cache->bytes_super, &space_info);
10242 btrfs_remove_free_space_cache(cache);
10243 spin_lock(&info->block_group_cache_lock);
10244 rb_erase(&cache->cache_node,
10245 &info->block_group_cache_tree);
10246 RB_CLEAR_NODE(&cache->cache_node);
10247 spin_unlock(&info->block_group_cache_lock);
10248 btrfs_put_block_group(cache);
10252 cache->space_info = space_info;
10254 __link_block_group(space_info, cache);
10256 set_avail_alloc_bits(root->fs_info, cache->flags);
10257 if (btrfs_chunk_readonly(root, cache->key.objectid)) {
10258 inc_block_group_ro(cache, 1);
10259 } else if (btrfs_block_group_used(&cache->item) == 0) {
10260 spin_lock(&info->unused_bgs_lock);
10261 /* Should always be true but just in case. */
10262 if (list_empty(&cache->bg_list)) {
10263 btrfs_get_block_group(cache);
10264 list_add_tail(&cache->bg_list,
10265 &info->unused_bgs);
10267 spin_unlock(&info->unused_bgs_lock);
10271 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
10272 if (!(get_alloc_profile(root, space_info->flags) &
10273 (BTRFS_BLOCK_GROUP_RAID10 |
10274 BTRFS_BLOCK_GROUP_RAID1 |
10275 BTRFS_BLOCK_GROUP_RAID5 |
10276 BTRFS_BLOCK_GROUP_RAID6 |
10277 BTRFS_BLOCK_GROUP_DUP)))
10280 * avoid allocating from un-mirrored block group if there are
10281 * mirrored block groups.
10283 list_for_each_entry(cache,
10284 &space_info->block_groups[BTRFS_RAID_RAID0],
10286 inc_block_group_ro(cache, 1);
10287 list_for_each_entry(cache,
10288 &space_info->block_groups[BTRFS_RAID_SINGLE],
10290 inc_block_group_ro(cache, 1);
10293 init_global_block_rsv(info);
10296 btrfs_free_path(path);
10300 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
10301 struct btrfs_root *root)
10303 struct btrfs_block_group_cache *block_group, *tmp;
10304 struct btrfs_root *extent_root = root->fs_info->extent_root;
10305 struct btrfs_block_group_item item;
10306 struct btrfs_key key;
10308 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
10310 trans->can_flush_pending_bgs = false;
10311 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
10315 spin_lock(&block_group->lock);
10316 memcpy(&item, &block_group->item, sizeof(item));
10317 memcpy(&key, &block_group->key, sizeof(key));
10318 spin_unlock(&block_group->lock);
10320 ret = btrfs_insert_item(trans, extent_root, &key, &item,
10323 btrfs_abort_transaction(trans, ret);
10324 ret = btrfs_finish_chunk_alloc(trans, extent_root,
10325 key.objectid, key.offset);
10327 btrfs_abort_transaction(trans, ret);
10328 add_block_group_free_space(trans, root->fs_info, block_group);
10329 /* already aborted the transaction if it failed. */
10331 list_del_init(&block_group->bg_list);
10333 trans->can_flush_pending_bgs = can_flush_pending_bgs;
10336 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
10337 struct btrfs_root *root, u64 bytes_used,
10338 u64 type, u64 chunk_objectid, u64 chunk_offset,
10342 struct btrfs_root *extent_root;
10343 struct btrfs_block_group_cache *cache;
10344 extent_root = root->fs_info->extent_root;
10346 btrfs_set_log_full_commit(root->fs_info, trans);
10348 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
10352 btrfs_set_block_group_used(&cache->item, bytes_used);
10353 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
10354 btrfs_set_block_group_flags(&cache->item, type);
10356 cache->flags = type;
10357 cache->last_byte_to_unpin = (u64)-1;
10358 cache->cached = BTRFS_CACHE_FINISHED;
10359 cache->needs_free_space = 1;
10360 ret = exclude_super_stripes(root, cache);
10363 * We may have excluded something, so call this just in
10366 free_excluded_extents(root, cache);
10367 btrfs_put_block_group(cache);
10371 add_new_free_space(cache, root->fs_info, chunk_offset,
10372 chunk_offset + size);
10374 free_excluded_extents(root, cache);
10376 #ifdef CONFIG_BTRFS_DEBUG
10377 if (btrfs_should_fragment_free_space(root, cache)) {
10378 u64 new_bytes_used = size - bytes_used;
10380 bytes_used += new_bytes_used >> 1;
10381 fragment_free_space(root, cache);
10385 * Call to ensure the corresponding space_info object is created and
10386 * assigned to our block group, but don't update its counters just yet.
10387 * We want our bg to be added to the rbtree with its ->space_info set.
10389 ret = update_space_info(root->fs_info, cache->flags, 0, 0, 0,
10390 &cache->space_info);
10392 btrfs_remove_free_space_cache(cache);
10393 btrfs_put_block_group(cache);
10397 ret = btrfs_add_block_group_cache(root->fs_info, cache);
10399 btrfs_remove_free_space_cache(cache);
10400 btrfs_put_block_group(cache);
10405 * Now that our block group has its ->space_info set and is inserted in
10406 * the rbtree, update the space info's counters.
10408 trace_btrfs_add_block_group(root->fs_info, cache, 1);
10409 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
10410 cache->bytes_super, &cache->space_info);
10412 btrfs_remove_free_space_cache(cache);
10413 spin_lock(&root->fs_info->block_group_cache_lock);
10414 rb_erase(&cache->cache_node,
10415 &root->fs_info->block_group_cache_tree);
10416 RB_CLEAR_NODE(&cache->cache_node);
10417 spin_unlock(&root->fs_info->block_group_cache_lock);
10418 btrfs_put_block_group(cache);
10421 update_global_block_rsv(root->fs_info);
10423 __link_block_group(cache->space_info, cache);
10425 list_add_tail(&cache->bg_list, &trans->new_bgs);
10427 set_avail_alloc_bits(extent_root->fs_info, type);
10431 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
10433 u64 extra_flags = chunk_to_extended(flags) &
10434 BTRFS_EXTENDED_PROFILE_MASK;
10436 write_seqlock(&fs_info->profiles_lock);
10437 if (flags & BTRFS_BLOCK_GROUP_DATA)
10438 fs_info->avail_data_alloc_bits &= ~extra_flags;
10439 if (flags & BTRFS_BLOCK_GROUP_METADATA)
10440 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
10441 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
10442 fs_info->avail_system_alloc_bits &= ~extra_flags;
10443 write_sequnlock(&fs_info->profiles_lock);
10446 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
10447 struct btrfs_root *root, u64 group_start,
10448 struct extent_map *em)
10450 struct btrfs_path *path;
10451 struct btrfs_block_group_cache *block_group;
10452 struct btrfs_free_cluster *cluster;
10453 struct btrfs_root *tree_root = root->fs_info->tree_root;
10454 struct btrfs_key key;
10455 struct inode *inode;
10456 struct kobject *kobj = NULL;
10460 struct btrfs_caching_control *caching_ctl = NULL;
10463 root = root->fs_info->extent_root;
10465 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
10466 BUG_ON(!block_group);
10467 BUG_ON(!block_group->ro);
10470 * Free the reserved super bytes from this block group before
10473 free_excluded_extents(root, block_group);
10475 memcpy(&key, &block_group->key, sizeof(key));
10476 index = get_block_group_index(block_group);
10477 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
10478 BTRFS_BLOCK_GROUP_RAID1 |
10479 BTRFS_BLOCK_GROUP_RAID10))
10484 /* make sure this block group isn't part of an allocation cluster */
10485 cluster = &root->fs_info->data_alloc_cluster;
10486 spin_lock(&cluster->refill_lock);
10487 btrfs_return_cluster_to_free_space(block_group, cluster);
10488 spin_unlock(&cluster->refill_lock);
10491 * make sure this block group isn't part of a metadata
10492 * allocation cluster
10494 cluster = &root->fs_info->meta_alloc_cluster;
10495 spin_lock(&cluster->refill_lock);
10496 btrfs_return_cluster_to_free_space(block_group, cluster);
10497 spin_unlock(&cluster->refill_lock);
10499 path = btrfs_alloc_path();
10506 * get the inode first so any iput calls done for the io_list
10507 * aren't the final iput (no unlinks allowed now)
10509 inode = lookup_free_space_inode(tree_root, block_group, path);
10511 mutex_lock(&trans->transaction->cache_write_mutex);
10513 * make sure our free spache cache IO is done before remove the
10516 spin_lock(&trans->transaction->dirty_bgs_lock);
10517 if (!list_empty(&block_group->io_list)) {
10518 list_del_init(&block_group->io_list);
10520 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
10522 spin_unlock(&trans->transaction->dirty_bgs_lock);
10523 btrfs_wait_cache_io(root, trans, block_group,
10524 &block_group->io_ctl, path,
10525 block_group->key.objectid);
10526 btrfs_put_block_group(block_group);
10527 spin_lock(&trans->transaction->dirty_bgs_lock);
10530 if (!list_empty(&block_group->dirty_list)) {
10531 list_del_init(&block_group->dirty_list);
10532 btrfs_put_block_group(block_group);
10534 spin_unlock(&trans->transaction->dirty_bgs_lock);
10535 mutex_unlock(&trans->transaction->cache_write_mutex);
10537 if (!IS_ERR(inode)) {
10538 ret = btrfs_orphan_add(trans, inode);
10540 btrfs_add_delayed_iput(inode);
10543 clear_nlink(inode);
10544 /* One for the block groups ref */
10545 spin_lock(&block_group->lock);
10546 if (block_group->iref) {
10547 block_group->iref = 0;
10548 block_group->inode = NULL;
10549 spin_unlock(&block_group->lock);
10552 spin_unlock(&block_group->lock);
10554 /* One for our lookup ref */
10555 btrfs_add_delayed_iput(inode);
10558 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
10559 key.offset = block_group->key.objectid;
10562 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
10566 btrfs_release_path(path);
10568 ret = btrfs_del_item(trans, tree_root, path);
10571 btrfs_release_path(path);
10574 spin_lock(&root->fs_info->block_group_cache_lock);
10575 rb_erase(&block_group->cache_node,
10576 &root->fs_info->block_group_cache_tree);
10577 RB_CLEAR_NODE(&block_group->cache_node);
10579 if (root->fs_info->first_logical_byte == block_group->key.objectid)
10580 root->fs_info->first_logical_byte = (u64)-1;
10581 spin_unlock(&root->fs_info->block_group_cache_lock);
10583 down_write(&block_group->space_info->groups_sem);
10585 * we must use list_del_init so people can check to see if they
10586 * are still on the list after taking the semaphore
10588 list_del_init(&block_group->list);
10589 if (list_empty(&block_group->space_info->block_groups[index])) {
10590 kobj = block_group->space_info->block_group_kobjs[index];
10591 block_group->space_info->block_group_kobjs[index] = NULL;
10592 clear_avail_alloc_bits(root->fs_info, block_group->flags);
10594 up_write(&block_group->space_info->groups_sem);
10600 if (block_group->has_caching_ctl)
10601 caching_ctl = get_caching_control(block_group);
10602 if (block_group->cached == BTRFS_CACHE_STARTED)
10603 wait_block_group_cache_done(block_group);
10604 if (block_group->has_caching_ctl) {
10605 down_write(&root->fs_info->commit_root_sem);
10606 if (!caching_ctl) {
10607 struct btrfs_caching_control *ctl;
10609 list_for_each_entry(ctl,
10610 &root->fs_info->caching_block_groups, list)
10611 if (ctl->block_group == block_group) {
10613 atomic_inc(&caching_ctl->count);
10618 list_del_init(&caching_ctl->list);
10619 up_write(&root->fs_info->commit_root_sem);
10621 /* Once for the caching bgs list and once for us. */
10622 put_caching_control(caching_ctl);
10623 put_caching_control(caching_ctl);
10627 spin_lock(&trans->transaction->dirty_bgs_lock);
10628 if (!list_empty(&block_group->dirty_list)) {
10631 if (!list_empty(&block_group->io_list)) {
10634 spin_unlock(&trans->transaction->dirty_bgs_lock);
10635 btrfs_remove_free_space_cache(block_group);
10637 spin_lock(&block_group->space_info->lock);
10638 list_del_init(&block_group->ro_list);
10640 if (btrfs_test_opt(root->fs_info, ENOSPC_DEBUG)) {
10641 WARN_ON(block_group->space_info->total_bytes
10642 < block_group->key.offset);
10643 WARN_ON(block_group->space_info->bytes_readonly
10644 < block_group->key.offset);
10645 WARN_ON(block_group->space_info->disk_total
10646 < block_group->key.offset * factor);
10648 block_group->space_info->total_bytes -= block_group->key.offset;
10649 block_group->space_info->bytes_readonly -= block_group->key.offset;
10650 block_group->space_info->disk_total -= block_group->key.offset * factor;
10652 spin_unlock(&block_group->space_info->lock);
10654 memcpy(&key, &block_group->key, sizeof(key));
10657 if (!list_empty(&em->list)) {
10658 /* We're in the transaction->pending_chunks list. */
10659 free_extent_map(em);
10661 spin_lock(&block_group->lock);
10662 block_group->removed = 1;
10664 * At this point trimming can't start on this block group, because we
10665 * removed the block group from the tree fs_info->block_group_cache_tree
10666 * so no one can't find it anymore and even if someone already got this
10667 * block group before we removed it from the rbtree, they have already
10668 * incremented block_group->trimming - if they didn't, they won't find
10669 * any free space entries because we already removed them all when we
10670 * called btrfs_remove_free_space_cache().
10672 * And we must not remove the extent map from the fs_info->mapping_tree
10673 * to prevent the same logical address range and physical device space
10674 * ranges from being reused for a new block group. This is because our
10675 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10676 * completely transactionless, so while it is trimming a range the
10677 * currently running transaction might finish and a new one start,
10678 * allowing for new block groups to be created that can reuse the same
10679 * physical device locations unless we take this special care.
10681 * There may also be an implicit trim operation if the file system
10682 * is mounted with -odiscard. The same protections must remain
10683 * in place until the extents have been discarded completely when
10684 * the transaction commit has completed.
10686 remove_em = (atomic_read(&block_group->trimming) == 0);
10688 * Make sure a trimmer task always sees the em in the pinned_chunks list
10689 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10690 * before checking block_group->removed).
10694 * Our em might be in trans->transaction->pending_chunks which
10695 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10696 * and so is the fs_info->pinned_chunks list.
10698 * So at this point we must be holding the chunk_mutex to avoid
10699 * any races with chunk allocation (more specifically at
10700 * volumes.c:contains_pending_extent()), to ensure it always
10701 * sees the em, either in the pending_chunks list or in the
10702 * pinned_chunks list.
10704 list_move_tail(&em->list, &root->fs_info->pinned_chunks);
10706 spin_unlock(&block_group->lock);
10709 struct extent_map_tree *em_tree;
10711 em_tree = &root->fs_info->mapping_tree.map_tree;
10712 write_lock(&em_tree->lock);
10714 * The em might be in the pending_chunks list, so make sure the
10715 * chunk mutex is locked, since remove_extent_mapping() will
10716 * delete us from that list.
10718 remove_extent_mapping(em_tree, em);
10719 write_unlock(&em_tree->lock);
10720 /* once for the tree */
10721 free_extent_map(em);
10724 unlock_chunks(root);
10726 ret = remove_block_group_free_space(trans, root->fs_info, block_group);
10730 btrfs_put_block_group(block_group);
10731 btrfs_put_block_group(block_group);
10733 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10739 ret = btrfs_del_item(trans, root, path);
10741 btrfs_free_path(path);
10745 struct btrfs_trans_handle *
10746 btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info,
10747 const u64 chunk_offset)
10749 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
10750 struct extent_map *em;
10751 struct map_lookup *map;
10752 unsigned int num_items;
10754 read_lock(&em_tree->lock);
10755 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
10756 read_unlock(&em_tree->lock);
10757 ASSERT(em && em->start == chunk_offset);
10760 * We need to reserve 3 + N units from the metadata space info in order
10761 * to remove a block group (done at btrfs_remove_chunk() and at
10762 * btrfs_remove_block_group()), which are used for:
10764 * 1 unit for adding the free space inode's orphan (located in the tree
10766 * 1 unit for deleting the block group item (located in the extent
10768 * 1 unit for deleting the free space item (located in tree of tree
10770 * N units for deleting N device extent items corresponding to each
10771 * stripe (located in the device tree).
10773 * In order to remove a block group we also need to reserve units in the
10774 * system space info in order to update the chunk tree (update one or
10775 * more device items and remove one chunk item), but this is done at
10776 * btrfs_remove_chunk() through a call to check_system_chunk().
10778 map = em->map_lookup;
10779 num_items = 3 + map->num_stripes;
10780 free_extent_map(em);
10782 return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root,
10787 * Process the unused_bgs list and remove any that don't have any allocated
10788 * space inside of them.
10790 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10792 struct btrfs_block_group_cache *block_group;
10793 struct btrfs_space_info *space_info;
10794 struct btrfs_root *root = fs_info->extent_root;
10795 struct btrfs_trans_handle *trans;
10798 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
10801 spin_lock(&fs_info->unused_bgs_lock);
10802 while (!list_empty(&fs_info->unused_bgs)) {
10806 block_group = list_first_entry(&fs_info->unused_bgs,
10807 struct btrfs_block_group_cache,
10809 list_del_init(&block_group->bg_list);
10811 space_info = block_group->space_info;
10813 if (ret || btrfs_mixed_space_info(space_info)) {
10814 btrfs_put_block_group(block_group);
10817 spin_unlock(&fs_info->unused_bgs_lock);
10819 mutex_lock(&fs_info->delete_unused_bgs_mutex);
10821 /* Don't want to race with allocators so take the groups_sem */
10822 down_write(&space_info->groups_sem);
10823 spin_lock(&block_group->lock);
10824 if (block_group->reserved ||
10825 btrfs_block_group_used(&block_group->item) ||
10826 (block_group->ro && !block_group->removed) ||
10827 list_is_singular(&block_group->list)) {
10829 * We want to bail if we made new allocations or have
10830 * outstanding allocations in this block group. We do
10831 * the ro check in case balance is currently acting on
10832 * this block group.
10834 spin_unlock(&block_group->lock);
10835 up_write(&space_info->groups_sem);
10838 spin_unlock(&block_group->lock);
10840 /* We don't want to force the issue, only flip if it's ok. */
10841 ret = inc_block_group_ro(block_group, 0);
10842 up_write(&space_info->groups_sem);
10849 * Want to do this before we do anything else so we can recover
10850 * properly if we fail to join the transaction.
10852 trans = btrfs_start_trans_remove_block_group(fs_info,
10853 block_group->key.objectid);
10854 if (IS_ERR(trans)) {
10855 btrfs_dec_block_group_ro(root, block_group);
10856 ret = PTR_ERR(trans);
10861 * We could have pending pinned extents for this block group,
10862 * just delete them, we don't care about them anymore.
10864 start = block_group->key.objectid;
10865 end = start + block_group->key.offset - 1;
10867 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10868 * btrfs_finish_extent_commit(). If we are at transaction N,
10869 * another task might be running finish_extent_commit() for the
10870 * previous transaction N - 1, and have seen a range belonging
10871 * to the block group in freed_extents[] before we were able to
10872 * clear the whole block group range from freed_extents[]. This
10873 * means that task can lookup for the block group after we
10874 * unpinned it from freed_extents[] and removed it, leading to
10875 * a BUG_ON() at btrfs_unpin_extent_range().
10877 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10878 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10881 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10882 btrfs_dec_block_group_ro(root, block_group);
10885 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10888 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10889 btrfs_dec_block_group_ro(root, block_group);
10892 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10894 /* Reset pinned so btrfs_put_block_group doesn't complain */
10895 spin_lock(&space_info->lock);
10896 spin_lock(&block_group->lock);
10898 space_info->bytes_pinned -= block_group->pinned;
10899 space_info->bytes_readonly += block_group->pinned;
10900 percpu_counter_add(&space_info->total_bytes_pinned,
10901 -block_group->pinned);
10902 block_group->pinned = 0;
10904 spin_unlock(&block_group->lock);
10905 spin_unlock(&space_info->lock);
10907 /* DISCARD can flip during remount */
10908 trimming = btrfs_test_opt(root->fs_info, DISCARD);
10910 /* Implicit trim during transaction commit. */
10912 btrfs_get_block_group_trimming(block_group);
10915 * Btrfs_remove_chunk will abort the transaction if things go
10918 ret = btrfs_remove_chunk(trans, root,
10919 block_group->key.objectid);
10923 btrfs_put_block_group_trimming(block_group);
10928 * If we're not mounted with -odiscard, we can just forget
10929 * about this block group. Otherwise we'll need to wait
10930 * until transaction commit to do the actual discard.
10933 spin_lock(&fs_info->unused_bgs_lock);
10935 * A concurrent scrub might have added us to the list
10936 * fs_info->unused_bgs, so use a list_move operation
10937 * to add the block group to the deleted_bgs list.
10939 list_move(&block_group->bg_list,
10940 &trans->transaction->deleted_bgs);
10941 spin_unlock(&fs_info->unused_bgs_lock);
10942 btrfs_get_block_group(block_group);
10945 btrfs_end_transaction(trans, root);
10947 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
10948 btrfs_put_block_group(block_group);
10949 spin_lock(&fs_info->unused_bgs_lock);
10951 spin_unlock(&fs_info->unused_bgs_lock);
10954 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10956 struct btrfs_space_info *space_info;
10957 struct btrfs_super_block *disk_super;
10963 disk_super = fs_info->super_copy;
10964 if (!btrfs_super_root(disk_super))
10967 features = btrfs_super_incompat_flags(disk_super);
10968 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10971 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10972 ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info);
10977 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10978 ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info);
10980 flags = BTRFS_BLOCK_GROUP_METADATA;
10981 ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info);
10985 flags = BTRFS_BLOCK_GROUP_DATA;
10986 ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info);
10992 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
10994 return unpin_extent_range(root, start, end, false);
10998 * It used to be that old block groups would be left around forever.
10999 * Iterating over them would be enough to trim unused space. Since we
11000 * now automatically remove them, we also need to iterate over unallocated
11003 * We don't want a transaction for this since the discard may take a
11004 * substantial amount of time. We don't require that a transaction be
11005 * running, but we do need to take a running transaction into account
11006 * to ensure that we're not discarding chunks that were released in
11007 * the current transaction.
11009 * Holding the chunks lock will prevent other threads from allocating
11010 * or releasing chunks, but it won't prevent a running transaction
11011 * from committing and releasing the memory that the pending chunks
11012 * list head uses. For that, we need to take a reference to the
11015 static int btrfs_trim_free_extents(struct btrfs_device *device,
11016 u64 minlen, u64 *trimmed)
11018 u64 start = 0, len = 0;
11023 /* Not writeable = nothing to do. */
11024 if (!device->writeable)
11027 /* No free space = nothing to do. */
11028 if (device->total_bytes <= device->bytes_used)
11034 struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
11035 struct btrfs_transaction *trans;
11038 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
11042 down_read(&fs_info->commit_root_sem);
11044 spin_lock(&fs_info->trans_lock);
11045 trans = fs_info->running_transaction;
11047 atomic_inc(&trans->use_count);
11048 spin_unlock(&fs_info->trans_lock);
11050 ret = find_free_dev_extent_start(trans, device, minlen, start,
11053 btrfs_put_transaction(trans);
11056 up_read(&fs_info->commit_root_sem);
11057 mutex_unlock(&fs_info->chunk_mutex);
11058 if (ret == -ENOSPC)
11063 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
11064 up_read(&fs_info->commit_root_sem);
11065 mutex_unlock(&fs_info->chunk_mutex);
11073 if (fatal_signal_pending(current)) {
11074 ret = -ERESTARTSYS;
11084 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
11086 struct btrfs_fs_info *fs_info = root->fs_info;
11087 struct btrfs_block_group_cache *cache = NULL;
11088 struct btrfs_device *device;
11089 struct list_head *devices;
11094 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
11098 * try to trim all FS space, our block group may start from non-zero.
11100 if (range->len == total_bytes)
11101 cache = btrfs_lookup_first_block_group(fs_info, range->start);
11103 cache = btrfs_lookup_block_group(fs_info, range->start);
11106 if (cache->key.objectid >= (range->start + range->len)) {
11107 btrfs_put_block_group(cache);
11111 start = max(range->start, cache->key.objectid);
11112 end = min(range->start + range->len,
11113 cache->key.objectid + cache->key.offset);
11115 if (end - start >= range->minlen) {
11116 if (!block_group_cache_done(cache)) {
11117 ret = cache_block_group(cache, 0);
11119 btrfs_put_block_group(cache);
11122 ret = wait_block_group_cache_done(cache);
11124 btrfs_put_block_group(cache);
11128 ret = btrfs_trim_block_group(cache,
11134 trimmed += group_trimmed;
11136 btrfs_put_block_group(cache);
11141 cache = next_block_group(fs_info->tree_root, cache);
11144 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
11145 devices = &root->fs_info->fs_devices->alloc_list;
11146 list_for_each_entry(device, devices, dev_alloc_list) {
11147 ret = btrfs_trim_free_extents(device, range->minlen,
11152 trimmed += group_trimmed;
11154 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
11156 range->len = trimmed;
11161 * btrfs_{start,end}_write_no_snapshoting() are similar to
11162 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
11163 * data into the page cache through nocow before the subvolume is snapshoted,
11164 * but flush the data into disk after the snapshot creation, or to prevent
11165 * operations while snapshoting is ongoing and that cause the snapshot to be
11166 * inconsistent (writes followed by expanding truncates for example).
11168 void btrfs_end_write_no_snapshoting(struct btrfs_root *root)
11170 percpu_counter_dec(&root->subv_writers->counter);
11172 * Make sure counter is updated before we wake up waiters.
11175 if (waitqueue_active(&root->subv_writers->wait))
11176 wake_up(&root->subv_writers->wait);
11179 int btrfs_start_write_no_snapshoting(struct btrfs_root *root)
11181 if (atomic_read(&root->will_be_snapshoted))
11184 percpu_counter_inc(&root->subv_writers->counter);
11186 * Make sure counter is updated before we check for snapshot creation.
11189 if (atomic_read(&root->will_be_snapshoted)) {
11190 btrfs_end_write_no_snapshoting(root);
11196 static int wait_snapshoting_atomic_t(atomic_t *a)
11202 void btrfs_wait_for_snapshot_creation(struct btrfs_root *root)
11207 ret = btrfs_start_write_no_snapshoting(root);
11210 wait_on_atomic_t(&root->will_be_snapshoted,
11211 wait_snapshoting_atomic_t,
11212 TASK_UNINTERRUPTIBLE);