1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
6 #include <linux/sched.h>
7 #include <linux/sched/signal.h>
8 #include <linux/pagemap.h>
9 #include <linux/writeback.h>
10 #include <linux/blkdev.h>
11 #include <linux/sort.h>
12 #include <linux/rcupdate.h>
13 #include <linux/kthread.h>
14 #include <linux/slab.h>
15 #include <linux/ratelimit.h>
16 #include <linux/percpu_counter.h>
17 #include <linux/lockdep.h>
18 #include <linux/crc32c.h>
21 #include "print-tree.h"
25 #include "free-space-cache.h"
26 #include "free-space-tree.h"
30 #include "ref-verify.h"
32 #undef SCRAMBLE_DELAYED_REFS
35 * control flags for do_chunk_alloc's force field
36 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
37 * if we really need one.
39 * CHUNK_ALLOC_LIMITED means to only try and allocate one
40 * if we have very few chunks already allocated. This is
41 * used as part of the clustering code to help make sure
42 * we have a good pool of storage to cluster in, without
43 * filling the FS with empty chunks
45 * CHUNK_ALLOC_FORCE means it must try to allocate one
49 CHUNK_ALLOC_NO_FORCE = 0,
50 CHUNK_ALLOC_LIMITED = 1,
51 CHUNK_ALLOC_FORCE = 2,
55 * Declare a helper function to detect underflow of various space info members
57 #define DECLARE_SPACE_INFO_UPDATE(name) \
58 static inline void update_##name(struct btrfs_space_info *sinfo, \
61 if (bytes < 0 && sinfo->name < -bytes) { \
66 sinfo->name += bytes; \
69 DECLARE_SPACE_INFO_UPDATE(bytes_may_use);
70 DECLARE_SPACE_INFO_UPDATE(bytes_pinned);
72 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
73 struct btrfs_delayed_ref_node *node, u64 parent,
74 u64 root_objectid, u64 owner_objectid,
75 u64 owner_offset, int refs_to_drop,
76 struct btrfs_delayed_extent_op *extra_op);
77 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
78 struct extent_buffer *leaf,
79 struct btrfs_extent_item *ei);
80 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
81 u64 parent, u64 root_objectid,
82 u64 flags, u64 owner, u64 offset,
83 struct btrfs_key *ins, int ref_mod);
84 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
85 struct btrfs_delayed_ref_node *node,
86 struct btrfs_delayed_extent_op *extent_op);
87 static int do_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags,
89 static int find_next_key(struct btrfs_path *path, int level,
90 struct btrfs_key *key);
91 static void dump_space_info(struct btrfs_fs_info *fs_info,
92 struct btrfs_space_info *info, u64 bytes,
93 int dump_block_groups);
94 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
96 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
97 struct btrfs_space_info *space_info,
99 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
100 struct btrfs_space_info *space_info,
104 block_group_cache_done(struct btrfs_block_group_cache *cache)
107 return cache->cached == BTRFS_CACHE_FINISHED ||
108 cache->cached == BTRFS_CACHE_ERROR;
111 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
113 return (cache->flags & bits) == bits;
116 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
118 atomic_inc(&cache->count);
121 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
123 if (atomic_dec_and_test(&cache->count)) {
124 WARN_ON(cache->pinned > 0);
125 WARN_ON(cache->reserved > 0);
128 * If not empty, someone is still holding mutex of
129 * full_stripe_lock, which can only be released by caller.
130 * And it will definitely cause use-after-free when caller
131 * tries to release full stripe lock.
133 * No better way to resolve, but only to warn.
135 WARN_ON(!RB_EMPTY_ROOT(&cache->full_stripe_locks_root.root));
136 kfree(cache->free_space_ctl);
142 * this adds the block group to the fs_info rb tree for the block group
145 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
146 struct btrfs_block_group_cache *block_group)
149 struct rb_node *parent = NULL;
150 struct btrfs_block_group_cache *cache;
152 spin_lock(&info->block_group_cache_lock);
153 p = &info->block_group_cache_tree.rb_node;
157 cache = rb_entry(parent, struct btrfs_block_group_cache,
159 if (block_group->key.objectid < cache->key.objectid) {
161 } else if (block_group->key.objectid > cache->key.objectid) {
164 spin_unlock(&info->block_group_cache_lock);
169 rb_link_node(&block_group->cache_node, parent, p);
170 rb_insert_color(&block_group->cache_node,
171 &info->block_group_cache_tree);
173 if (info->first_logical_byte > block_group->key.objectid)
174 info->first_logical_byte = block_group->key.objectid;
176 spin_unlock(&info->block_group_cache_lock);
182 * This will return the block group at or after bytenr if contains is 0, else
183 * it will return the block group that contains the bytenr
185 static struct btrfs_block_group_cache *
186 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
189 struct btrfs_block_group_cache *cache, *ret = NULL;
193 spin_lock(&info->block_group_cache_lock);
194 n = info->block_group_cache_tree.rb_node;
197 cache = rb_entry(n, struct btrfs_block_group_cache,
199 end = cache->key.objectid + cache->key.offset - 1;
200 start = cache->key.objectid;
202 if (bytenr < start) {
203 if (!contains && (!ret || start < ret->key.objectid))
206 } else if (bytenr > start) {
207 if (contains && bytenr <= end) {
218 btrfs_get_block_group(ret);
219 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
220 info->first_logical_byte = ret->key.objectid;
222 spin_unlock(&info->block_group_cache_lock);
227 static int add_excluded_extent(struct btrfs_fs_info *fs_info,
228 u64 start, u64 num_bytes)
230 u64 end = start + num_bytes - 1;
231 set_extent_bits(&fs_info->freed_extents[0],
232 start, end, EXTENT_UPTODATE);
233 set_extent_bits(&fs_info->freed_extents[1],
234 start, end, EXTENT_UPTODATE);
238 static void free_excluded_extents(struct btrfs_block_group_cache *cache)
240 struct btrfs_fs_info *fs_info = cache->fs_info;
243 start = cache->key.objectid;
244 end = start + cache->key.offset - 1;
246 clear_extent_bits(&fs_info->freed_extents[0],
247 start, end, EXTENT_UPTODATE);
248 clear_extent_bits(&fs_info->freed_extents[1],
249 start, end, EXTENT_UPTODATE);
252 static int exclude_super_stripes(struct btrfs_block_group_cache *cache)
254 struct btrfs_fs_info *fs_info = cache->fs_info;
260 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
261 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
262 cache->bytes_super += stripe_len;
263 ret = add_excluded_extent(fs_info, cache->key.objectid,
269 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
270 bytenr = btrfs_sb_offset(i);
271 ret = btrfs_rmap_block(fs_info, cache->key.objectid,
272 bytenr, &logical, &nr, &stripe_len);
279 if (logical[nr] > cache->key.objectid +
283 if (logical[nr] + stripe_len <= cache->key.objectid)
287 if (start < cache->key.objectid) {
288 start = cache->key.objectid;
289 len = (logical[nr] + stripe_len) - start;
291 len = min_t(u64, stripe_len,
292 cache->key.objectid +
293 cache->key.offset - start);
296 cache->bytes_super += len;
297 ret = add_excluded_extent(fs_info, start, len);
309 static struct btrfs_caching_control *
310 get_caching_control(struct btrfs_block_group_cache *cache)
312 struct btrfs_caching_control *ctl;
314 spin_lock(&cache->lock);
315 if (!cache->caching_ctl) {
316 spin_unlock(&cache->lock);
320 ctl = cache->caching_ctl;
321 refcount_inc(&ctl->count);
322 spin_unlock(&cache->lock);
326 static void put_caching_control(struct btrfs_caching_control *ctl)
328 if (refcount_dec_and_test(&ctl->count))
332 #ifdef CONFIG_BTRFS_DEBUG
333 static void fragment_free_space(struct btrfs_block_group_cache *block_group)
335 struct btrfs_fs_info *fs_info = block_group->fs_info;
336 u64 start = block_group->key.objectid;
337 u64 len = block_group->key.offset;
338 u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ?
339 fs_info->nodesize : fs_info->sectorsize;
340 u64 step = chunk << 1;
342 while (len > chunk) {
343 btrfs_remove_free_space(block_group, start, chunk);
354 * this is only called by cache_block_group, since we could have freed extents
355 * we need to check the pinned_extents for any extents that can't be used yet
356 * since their free space will be released as soon as the transaction commits.
358 u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
361 struct btrfs_fs_info *info = block_group->fs_info;
362 u64 extent_start, extent_end, size, total_added = 0;
365 while (start < end) {
366 ret = find_first_extent_bit(info->pinned_extents, start,
367 &extent_start, &extent_end,
368 EXTENT_DIRTY | EXTENT_UPTODATE,
373 if (extent_start <= start) {
374 start = extent_end + 1;
375 } else if (extent_start > start && extent_start < end) {
376 size = extent_start - start;
378 ret = btrfs_add_free_space(block_group, start,
380 BUG_ON(ret); /* -ENOMEM or logic error */
381 start = extent_end + 1;
390 ret = btrfs_add_free_space(block_group, start, size);
391 BUG_ON(ret); /* -ENOMEM or logic error */
397 static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl)
399 struct btrfs_block_group_cache *block_group = caching_ctl->block_group;
400 struct btrfs_fs_info *fs_info = block_group->fs_info;
401 struct btrfs_root *extent_root = fs_info->extent_root;
402 struct btrfs_path *path;
403 struct extent_buffer *leaf;
404 struct btrfs_key key;
411 path = btrfs_alloc_path();
415 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
417 #ifdef CONFIG_BTRFS_DEBUG
419 * If we're fragmenting we don't want to make anybody think we can
420 * allocate from this block group until we've had a chance to fragment
423 if (btrfs_should_fragment_free_space(block_group))
427 * We don't want to deadlock with somebody trying to allocate a new
428 * extent for the extent root while also trying to search the extent
429 * root to add free space. So we skip locking and search the commit
430 * root, since its read-only
432 path->skip_locking = 1;
433 path->search_commit_root = 1;
434 path->reada = READA_FORWARD;
438 key.type = BTRFS_EXTENT_ITEM_KEY;
441 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
445 leaf = path->nodes[0];
446 nritems = btrfs_header_nritems(leaf);
449 if (btrfs_fs_closing(fs_info) > 1) {
454 if (path->slots[0] < nritems) {
455 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
457 ret = find_next_key(path, 0, &key);
461 if (need_resched() ||
462 rwsem_is_contended(&fs_info->commit_root_sem)) {
464 caching_ctl->progress = last;
465 btrfs_release_path(path);
466 up_read(&fs_info->commit_root_sem);
467 mutex_unlock(&caching_ctl->mutex);
469 mutex_lock(&caching_ctl->mutex);
470 down_read(&fs_info->commit_root_sem);
474 ret = btrfs_next_leaf(extent_root, path);
479 leaf = path->nodes[0];
480 nritems = btrfs_header_nritems(leaf);
484 if (key.objectid < last) {
487 key.type = BTRFS_EXTENT_ITEM_KEY;
490 caching_ctl->progress = last;
491 btrfs_release_path(path);
495 if (key.objectid < block_group->key.objectid) {
500 if (key.objectid >= block_group->key.objectid +
501 block_group->key.offset)
504 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
505 key.type == BTRFS_METADATA_ITEM_KEY) {
506 total_found += add_new_free_space(block_group, last,
508 if (key.type == BTRFS_METADATA_ITEM_KEY)
509 last = key.objectid +
512 last = key.objectid + key.offset;
514 if (total_found > CACHING_CTL_WAKE_UP) {
517 wake_up(&caching_ctl->wait);
524 total_found += add_new_free_space(block_group, last,
525 block_group->key.objectid +
526 block_group->key.offset);
527 caching_ctl->progress = (u64)-1;
530 btrfs_free_path(path);
534 static noinline void caching_thread(struct btrfs_work *work)
536 struct btrfs_block_group_cache *block_group;
537 struct btrfs_fs_info *fs_info;
538 struct btrfs_caching_control *caching_ctl;
541 caching_ctl = container_of(work, struct btrfs_caching_control, work);
542 block_group = caching_ctl->block_group;
543 fs_info = block_group->fs_info;
545 mutex_lock(&caching_ctl->mutex);
546 down_read(&fs_info->commit_root_sem);
548 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
549 ret = load_free_space_tree(caching_ctl);
551 ret = load_extent_tree_free(caching_ctl);
553 spin_lock(&block_group->lock);
554 block_group->caching_ctl = NULL;
555 block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED;
556 spin_unlock(&block_group->lock);
558 #ifdef CONFIG_BTRFS_DEBUG
559 if (btrfs_should_fragment_free_space(block_group)) {
562 spin_lock(&block_group->space_info->lock);
563 spin_lock(&block_group->lock);
564 bytes_used = block_group->key.offset -
565 btrfs_block_group_used(&block_group->item);
566 block_group->space_info->bytes_used += bytes_used >> 1;
567 spin_unlock(&block_group->lock);
568 spin_unlock(&block_group->space_info->lock);
569 fragment_free_space(block_group);
573 caching_ctl->progress = (u64)-1;
575 up_read(&fs_info->commit_root_sem);
576 free_excluded_extents(block_group);
577 mutex_unlock(&caching_ctl->mutex);
579 wake_up(&caching_ctl->wait);
581 put_caching_control(caching_ctl);
582 btrfs_put_block_group(block_group);
585 static int cache_block_group(struct btrfs_block_group_cache *cache,
589 struct btrfs_fs_info *fs_info = cache->fs_info;
590 struct btrfs_caching_control *caching_ctl;
593 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
597 INIT_LIST_HEAD(&caching_ctl->list);
598 mutex_init(&caching_ctl->mutex);
599 init_waitqueue_head(&caching_ctl->wait);
600 caching_ctl->block_group = cache;
601 caching_ctl->progress = cache->key.objectid;
602 refcount_set(&caching_ctl->count, 1);
603 btrfs_init_work(&caching_ctl->work, btrfs_cache_helper,
604 caching_thread, NULL, NULL);
606 spin_lock(&cache->lock);
608 * This should be a rare occasion, but this could happen I think in the
609 * case where one thread starts to load the space cache info, and then
610 * some other thread starts a transaction commit which tries to do an
611 * allocation while the other thread is still loading the space cache
612 * info. The previous loop should have kept us from choosing this block
613 * group, but if we've moved to the state where we will wait on caching
614 * block groups we need to first check if we're doing a fast load here,
615 * so we can wait for it to finish, otherwise we could end up allocating
616 * from a block group who's cache gets evicted for one reason or
619 while (cache->cached == BTRFS_CACHE_FAST) {
620 struct btrfs_caching_control *ctl;
622 ctl = cache->caching_ctl;
623 refcount_inc(&ctl->count);
624 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
625 spin_unlock(&cache->lock);
629 finish_wait(&ctl->wait, &wait);
630 put_caching_control(ctl);
631 spin_lock(&cache->lock);
634 if (cache->cached != BTRFS_CACHE_NO) {
635 spin_unlock(&cache->lock);
639 WARN_ON(cache->caching_ctl);
640 cache->caching_ctl = caching_ctl;
641 cache->cached = BTRFS_CACHE_FAST;
642 spin_unlock(&cache->lock);
644 if (btrfs_test_opt(fs_info, SPACE_CACHE)) {
645 mutex_lock(&caching_ctl->mutex);
646 ret = load_free_space_cache(fs_info, cache);
648 spin_lock(&cache->lock);
650 cache->caching_ctl = NULL;
651 cache->cached = BTRFS_CACHE_FINISHED;
652 cache->last_byte_to_unpin = (u64)-1;
653 caching_ctl->progress = (u64)-1;
655 if (load_cache_only) {
656 cache->caching_ctl = NULL;
657 cache->cached = BTRFS_CACHE_NO;
659 cache->cached = BTRFS_CACHE_STARTED;
660 cache->has_caching_ctl = 1;
663 spin_unlock(&cache->lock);
664 #ifdef CONFIG_BTRFS_DEBUG
666 btrfs_should_fragment_free_space(cache)) {
669 spin_lock(&cache->space_info->lock);
670 spin_lock(&cache->lock);
671 bytes_used = cache->key.offset -
672 btrfs_block_group_used(&cache->item);
673 cache->space_info->bytes_used += bytes_used >> 1;
674 spin_unlock(&cache->lock);
675 spin_unlock(&cache->space_info->lock);
676 fragment_free_space(cache);
679 mutex_unlock(&caching_ctl->mutex);
681 wake_up(&caching_ctl->wait);
683 put_caching_control(caching_ctl);
684 free_excluded_extents(cache);
689 * We're either using the free space tree or no caching at all.
690 * Set cached to the appropriate value and wakeup any waiters.
692 spin_lock(&cache->lock);
693 if (load_cache_only) {
694 cache->caching_ctl = NULL;
695 cache->cached = BTRFS_CACHE_NO;
697 cache->cached = BTRFS_CACHE_STARTED;
698 cache->has_caching_ctl = 1;
700 spin_unlock(&cache->lock);
701 wake_up(&caching_ctl->wait);
704 if (load_cache_only) {
705 put_caching_control(caching_ctl);
709 down_write(&fs_info->commit_root_sem);
710 refcount_inc(&caching_ctl->count);
711 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
712 up_write(&fs_info->commit_root_sem);
714 btrfs_get_block_group(cache);
716 btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
722 * return the block group that starts at or after bytenr
724 static struct btrfs_block_group_cache *
725 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
727 return block_group_cache_tree_search(info, bytenr, 0);
731 * return the block group that contains the given bytenr
733 struct btrfs_block_group_cache *btrfs_lookup_block_group(
734 struct btrfs_fs_info *info,
737 return block_group_cache_tree_search(info, bytenr, 1);
740 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
743 struct list_head *head = &info->space_info;
744 struct btrfs_space_info *found;
746 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
749 list_for_each_entry_rcu(found, head, list) {
750 if (found->flags & flags) {
759 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, s64 num_bytes,
760 bool metadata, u64 root_objectid)
762 struct btrfs_space_info *space_info;
766 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
767 flags = BTRFS_BLOCK_GROUP_SYSTEM;
769 flags = BTRFS_BLOCK_GROUP_METADATA;
771 flags = BTRFS_BLOCK_GROUP_DATA;
774 space_info = __find_space_info(fs_info, flags);
776 percpu_counter_add_batch(&space_info->total_bytes_pinned, num_bytes,
777 BTRFS_TOTAL_BYTES_PINNED_BATCH);
781 * after adding space to the filesystem, we need to clear the full flags
782 * on all the space infos.
784 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
786 struct list_head *head = &info->space_info;
787 struct btrfs_space_info *found;
790 list_for_each_entry_rcu(found, head, list)
795 /* simple helper to search for an existing data extent at a given offset */
796 int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len)
799 struct btrfs_key key;
800 struct btrfs_path *path;
802 path = btrfs_alloc_path();
806 key.objectid = start;
808 key.type = BTRFS_EXTENT_ITEM_KEY;
809 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
810 btrfs_free_path(path);
815 * helper function to lookup reference count and flags of a tree block.
817 * the head node for delayed ref is used to store the sum of all the
818 * reference count modifications queued up in the rbtree. the head
819 * node may also store the extent flags to set. This way you can check
820 * to see what the reference count and extent flags would be if all of
821 * the delayed refs are not processed.
823 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
824 struct btrfs_fs_info *fs_info, u64 bytenr,
825 u64 offset, int metadata, u64 *refs, u64 *flags)
827 struct btrfs_delayed_ref_head *head;
828 struct btrfs_delayed_ref_root *delayed_refs;
829 struct btrfs_path *path;
830 struct btrfs_extent_item *ei;
831 struct extent_buffer *leaf;
832 struct btrfs_key key;
839 * If we don't have skinny metadata, don't bother doing anything
842 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) {
843 offset = fs_info->nodesize;
847 path = btrfs_alloc_path();
852 path->skip_locking = 1;
853 path->search_commit_root = 1;
857 key.objectid = bytenr;
860 key.type = BTRFS_METADATA_ITEM_KEY;
862 key.type = BTRFS_EXTENT_ITEM_KEY;
864 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
868 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
869 if (path->slots[0]) {
871 btrfs_item_key_to_cpu(path->nodes[0], &key,
873 if (key.objectid == bytenr &&
874 key.type == BTRFS_EXTENT_ITEM_KEY &&
875 key.offset == fs_info->nodesize)
881 leaf = path->nodes[0];
882 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
883 if (item_size >= sizeof(*ei)) {
884 ei = btrfs_item_ptr(leaf, path->slots[0],
885 struct btrfs_extent_item);
886 num_refs = btrfs_extent_refs(leaf, ei);
887 extent_flags = btrfs_extent_flags(leaf, ei);
890 btrfs_print_v0_err(fs_info);
892 btrfs_abort_transaction(trans, ret);
894 btrfs_handle_fs_error(fs_info, ret, NULL);
899 BUG_ON(num_refs == 0);
909 delayed_refs = &trans->transaction->delayed_refs;
910 spin_lock(&delayed_refs->lock);
911 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
913 if (!mutex_trylock(&head->mutex)) {
914 refcount_inc(&head->refs);
915 spin_unlock(&delayed_refs->lock);
917 btrfs_release_path(path);
920 * Mutex was contended, block until it's released and try
923 mutex_lock(&head->mutex);
924 mutex_unlock(&head->mutex);
925 btrfs_put_delayed_ref_head(head);
928 spin_lock(&head->lock);
929 if (head->extent_op && head->extent_op->update_flags)
930 extent_flags |= head->extent_op->flags_to_set;
932 BUG_ON(num_refs == 0);
934 num_refs += head->ref_mod;
935 spin_unlock(&head->lock);
936 mutex_unlock(&head->mutex);
938 spin_unlock(&delayed_refs->lock);
940 WARN_ON(num_refs == 0);
944 *flags = extent_flags;
946 btrfs_free_path(path);
951 * Back reference rules. Back refs have three main goals:
953 * 1) differentiate between all holders of references to an extent so that
954 * when a reference is dropped we can make sure it was a valid reference
955 * before freeing the extent.
957 * 2) Provide enough information to quickly find the holders of an extent
958 * if we notice a given block is corrupted or bad.
960 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
961 * maintenance. This is actually the same as #2, but with a slightly
962 * different use case.
964 * There are two kinds of back refs. The implicit back refs is optimized
965 * for pointers in non-shared tree blocks. For a given pointer in a block,
966 * back refs of this kind provide information about the block's owner tree
967 * and the pointer's key. These information allow us to find the block by
968 * b-tree searching. The full back refs is for pointers in tree blocks not
969 * referenced by their owner trees. The location of tree block is recorded
970 * in the back refs. Actually the full back refs is generic, and can be
971 * used in all cases the implicit back refs is used. The major shortcoming
972 * of the full back refs is its overhead. Every time a tree block gets
973 * COWed, we have to update back refs entry for all pointers in it.
975 * For a newly allocated tree block, we use implicit back refs for
976 * pointers in it. This means most tree related operations only involve
977 * implicit back refs. For a tree block created in old transaction, the
978 * only way to drop a reference to it is COW it. So we can detect the
979 * event that tree block loses its owner tree's reference and do the
980 * back refs conversion.
982 * When a tree block is COWed through a tree, there are four cases:
984 * The reference count of the block is one and the tree is the block's
985 * owner tree. Nothing to do in this case.
987 * The reference count of the block is one and the tree is not the
988 * block's owner tree. In this case, full back refs is used for pointers
989 * in the block. Remove these full back refs, add implicit back refs for
990 * every pointers in the new block.
992 * The reference count of the block is greater than one and the tree is
993 * the block's owner tree. In this case, implicit back refs is used for
994 * pointers in the block. Add full back refs for every pointers in the
995 * block, increase lower level extents' reference counts. The original
996 * implicit back refs are entailed to the new block.
998 * The reference count of the block is greater than one and the tree is
999 * not the block's owner tree. Add implicit back refs for every pointer in
1000 * the new block, increase lower level extents' reference count.
1002 * Back Reference Key composing:
1004 * The key objectid corresponds to the first byte in the extent,
1005 * The key type is used to differentiate between types of back refs.
1006 * There are different meanings of the key offset for different types
1009 * File extents can be referenced by:
1011 * - multiple snapshots, subvolumes, or different generations in one subvol
1012 * - different files inside a single subvolume
1013 * - different offsets inside a file (bookend extents in file.c)
1015 * The extent ref structure for the implicit back refs has fields for:
1017 * - Objectid of the subvolume root
1018 * - objectid of the file holding the reference
1019 * - original offset in the file
1020 * - how many bookend extents
1022 * The key offset for the implicit back refs is hash of the first
1025 * The extent ref structure for the full back refs has field for:
1027 * - number of pointers in the tree leaf
1029 * The key offset for the implicit back refs is the first byte of
1032 * When a file extent is allocated, The implicit back refs is used.
1033 * the fields are filled in:
1035 * (root_key.objectid, inode objectid, offset in file, 1)
1037 * When a file extent is removed file truncation, we find the
1038 * corresponding implicit back refs and check the following fields:
1040 * (btrfs_header_owner(leaf), inode objectid, offset in file)
1042 * Btree extents can be referenced by:
1044 * - Different subvolumes
1046 * Both the implicit back refs and the full back refs for tree blocks
1047 * only consist of key. The key offset for the implicit back refs is
1048 * objectid of block's owner tree. The key offset for the full back refs
1049 * is the first byte of parent block.
1051 * When implicit back refs is used, information about the lowest key and
1052 * level of the tree block are required. These information are stored in
1053 * tree block info structure.
1057 * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required,
1058 * is_data == BTRFS_REF_TYPE_DATA, data type is requiried,
1059 * is_data == BTRFS_REF_TYPE_ANY, either type is OK.
1061 int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb,
1062 struct btrfs_extent_inline_ref *iref,
1063 enum btrfs_inline_ref_type is_data)
1065 int type = btrfs_extent_inline_ref_type(eb, iref);
1066 u64 offset = btrfs_extent_inline_ref_offset(eb, iref);
1068 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1069 type == BTRFS_SHARED_BLOCK_REF_KEY ||
1070 type == BTRFS_SHARED_DATA_REF_KEY ||
1071 type == BTRFS_EXTENT_DATA_REF_KEY) {
1072 if (is_data == BTRFS_REF_TYPE_BLOCK) {
1073 if (type == BTRFS_TREE_BLOCK_REF_KEY)
1075 if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1076 ASSERT(eb->fs_info);
1078 * Every shared one has parent tree
1079 * block, which must be aligned to
1083 IS_ALIGNED(offset, eb->fs_info->nodesize))
1086 } else if (is_data == BTRFS_REF_TYPE_DATA) {
1087 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1089 if (type == BTRFS_SHARED_DATA_REF_KEY) {
1090 ASSERT(eb->fs_info);
1092 * Every shared one has parent tree
1093 * block, which must be aligned to
1097 IS_ALIGNED(offset, eb->fs_info->nodesize))
1101 ASSERT(is_data == BTRFS_REF_TYPE_ANY);
1106 btrfs_print_leaf((struct extent_buffer *)eb);
1107 btrfs_err(eb->fs_info, "eb %llu invalid extent inline ref type %d",
1111 return BTRFS_REF_TYPE_INVALID;
1114 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1116 u32 high_crc = ~(u32)0;
1117 u32 low_crc = ~(u32)0;
1120 lenum = cpu_to_le64(root_objectid);
1121 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1122 lenum = cpu_to_le64(owner);
1123 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1124 lenum = cpu_to_le64(offset);
1125 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1127 return ((u64)high_crc << 31) ^ (u64)low_crc;
1130 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1131 struct btrfs_extent_data_ref *ref)
1133 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1134 btrfs_extent_data_ref_objectid(leaf, ref),
1135 btrfs_extent_data_ref_offset(leaf, ref));
1138 static int match_extent_data_ref(struct extent_buffer *leaf,
1139 struct btrfs_extent_data_ref *ref,
1140 u64 root_objectid, u64 owner, u64 offset)
1142 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1143 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1144 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1149 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1150 struct btrfs_path *path,
1151 u64 bytenr, u64 parent,
1153 u64 owner, u64 offset)
1155 struct btrfs_root *root = trans->fs_info->extent_root;
1156 struct btrfs_key key;
1157 struct btrfs_extent_data_ref *ref;
1158 struct extent_buffer *leaf;
1164 key.objectid = bytenr;
1166 key.type = BTRFS_SHARED_DATA_REF_KEY;
1167 key.offset = parent;
1169 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1170 key.offset = hash_extent_data_ref(root_objectid,
1175 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1187 leaf = path->nodes[0];
1188 nritems = btrfs_header_nritems(leaf);
1190 if (path->slots[0] >= nritems) {
1191 ret = btrfs_next_leaf(root, path);
1197 leaf = path->nodes[0];
1198 nritems = btrfs_header_nritems(leaf);
1202 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1203 if (key.objectid != bytenr ||
1204 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1207 ref = btrfs_item_ptr(leaf, path->slots[0],
1208 struct btrfs_extent_data_ref);
1210 if (match_extent_data_ref(leaf, ref, root_objectid,
1213 btrfs_release_path(path);
1225 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1226 struct btrfs_path *path,
1227 u64 bytenr, u64 parent,
1228 u64 root_objectid, u64 owner,
1229 u64 offset, int refs_to_add)
1231 struct btrfs_root *root = trans->fs_info->extent_root;
1232 struct btrfs_key key;
1233 struct extent_buffer *leaf;
1238 key.objectid = bytenr;
1240 key.type = BTRFS_SHARED_DATA_REF_KEY;
1241 key.offset = parent;
1242 size = sizeof(struct btrfs_shared_data_ref);
1244 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1245 key.offset = hash_extent_data_ref(root_objectid,
1247 size = sizeof(struct btrfs_extent_data_ref);
1250 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1251 if (ret && ret != -EEXIST)
1254 leaf = path->nodes[0];
1256 struct btrfs_shared_data_ref *ref;
1257 ref = btrfs_item_ptr(leaf, path->slots[0],
1258 struct btrfs_shared_data_ref);
1260 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1262 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1263 num_refs += refs_to_add;
1264 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1267 struct btrfs_extent_data_ref *ref;
1268 while (ret == -EEXIST) {
1269 ref = btrfs_item_ptr(leaf, path->slots[0],
1270 struct btrfs_extent_data_ref);
1271 if (match_extent_data_ref(leaf, ref, root_objectid,
1274 btrfs_release_path(path);
1276 ret = btrfs_insert_empty_item(trans, root, path, &key,
1278 if (ret && ret != -EEXIST)
1281 leaf = path->nodes[0];
1283 ref = btrfs_item_ptr(leaf, path->slots[0],
1284 struct btrfs_extent_data_ref);
1286 btrfs_set_extent_data_ref_root(leaf, ref,
1288 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1289 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1290 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1292 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1293 num_refs += refs_to_add;
1294 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1297 btrfs_mark_buffer_dirty(leaf);
1300 btrfs_release_path(path);
1304 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1305 struct btrfs_path *path,
1306 int refs_to_drop, int *last_ref)
1308 struct btrfs_key key;
1309 struct btrfs_extent_data_ref *ref1 = NULL;
1310 struct btrfs_shared_data_ref *ref2 = NULL;
1311 struct extent_buffer *leaf;
1315 leaf = path->nodes[0];
1316 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1318 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1319 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1320 struct btrfs_extent_data_ref);
1321 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1322 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1323 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1324 struct btrfs_shared_data_ref);
1325 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1326 } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) {
1327 btrfs_print_v0_err(trans->fs_info);
1328 btrfs_abort_transaction(trans, -EINVAL);
1334 BUG_ON(num_refs < refs_to_drop);
1335 num_refs -= refs_to_drop;
1337 if (num_refs == 0) {
1338 ret = btrfs_del_item(trans, trans->fs_info->extent_root, path);
1341 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1342 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1343 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1344 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1345 btrfs_mark_buffer_dirty(leaf);
1350 static noinline u32 extent_data_ref_count(struct btrfs_path *path,
1351 struct btrfs_extent_inline_ref *iref)
1353 struct btrfs_key key;
1354 struct extent_buffer *leaf;
1355 struct btrfs_extent_data_ref *ref1;
1356 struct btrfs_shared_data_ref *ref2;
1360 leaf = path->nodes[0];
1361 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1363 BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY);
1366 * If type is invalid, we should have bailed out earlier than
1369 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
1370 ASSERT(type != BTRFS_REF_TYPE_INVALID);
1371 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1372 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1373 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1375 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1376 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1378 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1379 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1380 struct btrfs_extent_data_ref);
1381 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1382 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1383 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1384 struct btrfs_shared_data_ref);
1385 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1392 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1393 struct btrfs_path *path,
1394 u64 bytenr, u64 parent,
1397 struct btrfs_root *root = trans->fs_info->extent_root;
1398 struct btrfs_key key;
1401 key.objectid = bytenr;
1403 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1404 key.offset = parent;
1406 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1407 key.offset = root_objectid;
1410 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1416 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1417 struct btrfs_path *path,
1418 u64 bytenr, u64 parent,
1421 struct btrfs_key key;
1424 key.objectid = bytenr;
1426 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1427 key.offset = parent;
1429 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1430 key.offset = root_objectid;
1433 ret = btrfs_insert_empty_item(trans, trans->fs_info->extent_root,
1435 btrfs_release_path(path);
1439 static inline int extent_ref_type(u64 parent, u64 owner)
1442 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1444 type = BTRFS_SHARED_BLOCK_REF_KEY;
1446 type = BTRFS_TREE_BLOCK_REF_KEY;
1449 type = BTRFS_SHARED_DATA_REF_KEY;
1451 type = BTRFS_EXTENT_DATA_REF_KEY;
1456 static int find_next_key(struct btrfs_path *path, int level,
1457 struct btrfs_key *key)
1460 for (; level < BTRFS_MAX_LEVEL; level++) {
1461 if (!path->nodes[level])
1463 if (path->slots[level] + 1 >=
1464 btrfs_header_nritems(path->nodes[level]))
1467 btrfs_item_key_to_cpu(path->nodes[level], key,
1468 path->slots[level] + 1);
1470 btrfs_node_key_to_cpu(path->nodes[level], key,
1471 path->slots[level] + 1);
1478 * look for inline back ref. if back ref is found, *ref_ret is set
1479 * to the address of inline back ref, and 0 is returned.
1481 * if back ref isn't found, *ref_ret is set to the address where it
1482 * should be inserted, and -ENOENT is returned.
1484 * if insert is true and there are too many inline back refs, the path
1485 * points to the extent item, and -EAGAIN is returned.
1487 * NOTE: inline back refs are ordered in the same way that back ref
1488 * items in the tree are ordered.
1490 static noinline_for_stack
1491 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1492 struct btrfs_path *path,
1493 struct btrfs_extent_inline_ref **ref_ret,
1494 u64 bytenr, u64 num_bytes,
1495 u64 parent, u64 root_objectid,
1496 u64 owner, u64 offset, int insert)
1498 struct btrfs_fs_info *fs_info = trans->fs_info;
1499 struct btrfs_root *root = fs_info->extent_root;
1500 struct btrfs_key key;
1501 struct extent_buffer *leaf;
1502 struct btrfs_extent_item *ei;
1503 struct btrfs_extent_inline_ref *iref;
1513 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
1516 key.objectid = bytenr;
1517 key.type = BTRFS_EXTENT_ITEM_KEY;
1518 key.offset = num_bytes;
1520 want = extent_ref_type(parent, owner);
1522 extra_size = btrfs_extent_inline_ref_size(want);
1523 path->keep_locks = 1;
1528 * Owner is our level, so we can just add one to get the level for the
1529 * block we are interested in.
1531 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1532 key.type = BTRFS_METADATA_ITEM_KEY;
1537 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1544 * We may be a newly converted file system which still has the old fat
1545 * extent entries for metadata, so try and see if we have one of those.
1547 if (ret > 0 && skinny_metadata) {
1548 skinny_metadata = false;
1549 if (path->slots[0]) {
1551 btrfs_item_key_to_cpu(path->nodes[0], &key,
1553 if (key.objectid == bytenr &&
1554 key.type == BTRFS_EXTENT_ITEM_KEY &&
1555 key.offset == num_bytes)
1559 key.objectid = bytenr;
1560 key.type = BTRFS_EXTENT_ITEM_KEY;
1561 key.offset = num_bytes;
1562 btrfs_release_path(path);
1567 if (ret && !insert) {
1570 } else if (WARN_ON(ret)) {
1575 leaf = path->nodes[0];
1576 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1577 if (unlikely(item_size < sizeof(*ei))) {
1579 btrfs_print_v0_err(fs_info);
1580 btrfs_abort_transaction(trans, err);
1584 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1585 flags = btrfs_extent_flags(leaf, ei);
1587 ptr = (unsigned long)(ei + 1);
1588 end = (unsigned long)ei + item_size;
1590 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1591 ptr += sizeof(struct btrfs_tree_block_info);
1595 if (owner >= BTRFS_FIRST_FREE_OBJECTID)
1596 needed = BTRFS_REF_TYPE_DATA;
1598 needed = BTRFS_REF_TYPE_BLOCK;
1606 iref = (struct btrfs_extent_inline_ref *)ptr;
1607 type = btrfs_get_extent_inline_ref_type(leaf, iref, needed);
1608 if (type == BTRFS_REF_TYPE_INVALID) {
1616 ptr += btrfs_extent_inline_ref_size(type);
1620 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1621 struct btrfs_extent_data_ref *dref;
1622 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1623 if (match_extent_data_ref(leaf, dref, root_objectid,
1628 if (hash_extent_data_ref_item(leaf, dref) <
1629 hash_extent_data_ref(root_objectid, owner, offset))
1633 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1635 if (parent == ref_offset) {
1639 if (ref_offset < parent)
1642 if (root_objectid == ref_offset) {
1646 if (ref_offset < root_objectid)
1650 ptr += btrfs_extent_inline_ref_size(type);
1652 if (err == -ENOENT && insert) {
1653 if (item_size + extra_size >=
1654 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1659 * To add new inline back ref, we have to make sure
1660 * there is no corresponding back ref item.
1661 * For simplicity, we just do not add new inline back
1662 * ref if there is any kind of item for this block
1664 if (find_next_key(path, 0, &key) == 0 &&
1665 key.objectid == bytenr &&
1666 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1671 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1674 path->keep_locks = 0;
1675 btrfs_unlock_up_safe(path, 1);
1681 * helper to add new inline back ref
1683 static noinline_for_stack
1684 void setup_inline_extent_backref(struct btrfs_fs_info *fs_info,
1685 struct btrfs_path *path,
1686 struct btrfs_extent_inline_ref *iref,
1687 u64 parent, u64 root_objectid,
1688 u64 owner, u64 offset, int refs_to_add,
1689 struct btrfs_delayed_extent_op *extent_op)
1691 struct extent_buffer *leaf;
1692 struct btrfs_extent_item *ei;
1695 unsigned long item_offset;
1700 leaf = path->nodes[0];
1701 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1702 item_offset = (unsigned long)iref - (unsigned long)ei;
1704 type = extent_ref_type(parent, owner);
1705 size = btrfs_extent_inline_ref_size(type);
1707 btrfs_extend_item(fs_info, path, size);
1709 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1710 refs = btrfs_extent_refs(leaf, ei);
1711 refs += refs_to_add;
1712 btrfs_set_extent_refs(leaf, ei, refs);
1714 __run_delayed_extent_op(extent_op, leaf, ei);
1716 ptr = (unsigned long)ei + item_offset;
1717 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1718 if (ptr < end - size)
1719 memmove_extent_buffer(leaf, ptr + size, ptr,
1722 iref = (struct btrfs_extent_inline_ref *)ptr;
1723 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1724 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1725 struct btrfs_extent_data_ref *dref;
1726 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1727 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1728 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1729 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1730 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1731 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1732 struct btrfs_shared_data_ref *sref;
1733 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1734 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1735 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1736 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1737 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1739 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1741 btrfs_mark_buffer_dirty(leaf);
1744 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1745 struct btrfs_path *path,
1746 struct btrfs_extent_inline_ref **ref_ret,
1747 u64 bytenr, u64 num_bytes, u64 parent,
1748 u64 root_objectid, u64 owner, u64 offset)
1752 ret = lookup_inline_extent_backref(trans, path, ref_ret, bytenr,
1753 num_bytes, parent, root_objectid,
1758 btrfs_release_path(path);
1761 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1762 ret = lookup_tree_block_ref(trans, path, bytenr, parent,
1765 ret = lookup_extent_data_ref(trans, path, bytenr, parent,
1766 root_objectid, owner, offset);
1772 * helper to update/remove inline back ref
1774 static noinline_for_stack
1775 void update_inline_extent_backref(struct btrfs_path *path,
1776 struct btrfs_extent_inline_ref *iref,
1778 struct btrfs_delayed_extent_op *extent_op,
1781 struct extent_buffer *leaf = path->nodes[0];
1782 struct btrfs_fs_info *fs_info = leaf->fs_info;
1783 struct btrfs_extent_item *ei;
1784 struct btrfs_extent_data_ref *dref = NULL;
1785 struct btrfs_shared_data_ref *sref = NULL;
1793 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1794 refs = btrfs_extent_refs(leaf, ei);
1795 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1796 refs += refs_to_mod;
1797 btrfs_set_extent_refs(leaf, ei, refs);
1799 __run_delayed_extent_op(extent_op, leaf, ei);
1802 * If type is invalid, we should have bailed out after
1803 * lookup_inline_extent_backref().
1805 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY);
1806 ASSERT(type != BTRFS_REF_TYPE_INVALID);
1808 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1809 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1810 refs = btrfs_extent_data_ref_count(leaf, dref);
1811 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1812 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1813 refs = btrfs_shared_data_ref_count(leaf, sref);
1816 BUG_ON(refs_to_mod != -1);
1819 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1820 refs += refs_to_mod;
1823 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1824 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1826 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1829 size = btrfs_extent_inline_ref_size(type);
1830 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1831 ptr = (unsigned long)iref;
1832 end = (unsigned long)ei + item_size;
1833 if (ptr + size < end)
1834 memmove_extent_buffer(leaf, ptr, ptr + size,
1837 btrfs_truncate_item(fs_info, path, item_size, 1);
1839 btrfs_mark_buffer_dirty(leaf);
1842 static noinline_for_stack
1843 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1844 struct btrfs_path *path,
1845 u64 bytenr, u64 num_bytes, u64 parent,
1846 u64 root_objectid, u64 owner,
1847 u64 offset, int refs_to_add,
1848 struct btrfs_delayed_extent_op *extent_op)
1850 struct btrfs_extent_inline_ref *iref;
1853 ret = lookup_inline_extent_backref(trans, path, &iref, bytenr,
1854 num_bytes, parent, root_objectid,
1857 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1858 update_inline_extent_backref(path, iref, refs_to_add,
1860 } else if (ret == -ENOENT) {
1861 setup_inline_extent_backref(trans->fs_info, path, iref, parent,
1862 root_objectid, owner, offset,
1863 refs_to_add, extent_op);
1869 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1870 struct btrfs_path *path,
1871 u64 bytenr, u64 parent, u64 root_objectid,
1872 u64 owner, u64 offset, int refs_to_add)
1875 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1876 BUG_ON(refs_to_add != 1);
1877 ret = insert_tree_block_ref(trans, path, bytenr, parent,
1880 ret = insert_extent_data_ref(trans, path, bytenr, parent,
1881 root_objectid, owner, offset,
1887 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1888 struct btrfs_path *path,
1889 struct btrfs_extent_inline_ref *iref,
1890 int refs_to_drop, int is_data, int *last_ref)
1894 BUG_ON(!is_data && refs_to_drop != 1);
1896 update_inline_extent_backref(path, iref, -refs_to_drop, NULL,
1898 } else if (is_data) {
1899 ret = remove_extent_data_ref(trans, path, refs_to_drop,
1903 ret = btrfs_del_item(trans, trans->fs_info->extent_root, path);
1908 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
1909 u64 *discarded_bytes)
1912 u64 bytes_left, end;
1913 u64 aligned_start = ALIGN(start, 1 << 9);
1915 if (WARN_ON(start != aligned_start)) {
1916 len -= aligned_start - start;
1917 len = round_down(len, 1 << 9);
1918 start = aligned_start;
1921 *discarded_bytes = 0;
1929 /* Skip any superblocks on this device. */
1930 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
1931 u64 sb_start = btrfs_sb_offset(j);
1932 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
1933 u64 size = sb_start - start;
1935 if (!in_range(sb_start, start, bytes_left) &&
1936 !in_range(sb_end, start, bytes_left) &&
1937 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
1941 * Superblock spans beginning of range. Adjust start and
1944 if (sb_start <= start) {
1945 start += sb_end - start;
1950 bytes_left = end - start;
1955 ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
1958 *discarded_bytes += size;
1959 else if (ret != -EOPNOTSUPP)
1968 bytes_left = end - start;
1972 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
1975 *discarded_bytes += bytes_left;
1980 int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr,
1981 u64 num_bytes, u64 *actual_bytes)
1984 u64 discarded_bytes = 0;
1985 struct btrfs_bio *bbio = NULL;
1989 * Avoid races with device replace and make sure our bbio has devices
1990 * associated to its stripes that don't go away while we are discarding.
1992 btrfs_bio_counter_inc_blocked(fs_info);
1993 /* Tell the block device(s) that the sectors can be discarded */
1994 ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, bytenr, &num_bytes,
1996 /* Error condition is -ENOMEM */
1998 struct btrfs_bio_stripe *stripe = bbio->stripes;
2002 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
2004 struct request_queue *req_q;
2006 if (!stripe->dev->bdev) {
2007 ASSERT(btrfs_test_opt(fs_info, DEGRADED));
2010 req_q = bdev_get_queue(stripe->dev->bdev);
2011 if (!blk_queue_discard(req_q))
2014 ret = btrfs_issue_discard(stripe->dev->bdev,
2019 discarded_bytes += bytes;
2020 else if (ret != -EOPNOTSUPP)
2021 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2024 * Just in case we get back EOPNOTSUPP for some reason,
2025 * just ignore the return value so we don't screw up
2026 * people calling discard_extent.
2030 btrfs_put_bbio(bbio);
2032 btrfs_bio_counter_dec(fs_info);
2035 *actual_bytes = discarded_bytes;
2038 if (ret == -EOPNOTSUPP)
2043 /* Can return -ENOMEM */
2044 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2045 struct btrfs_root *root,
2046 u64 bytenr, u64 num_bytes, u64 parent,
2047 u64 root_objectid, u64 owner, u64 offset)
2049 struct btrfs_fs_info *fs_info = root->fs_info;
2050 int old_ref_mod, new_ref_mod;
2053 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
2054 root_objectid == BTRFS_TREE_LOG_OBJECTID);
2056 btrfs_ref_tree_mod(root, bytenr, num_bytes, parent, root_objectid,
2057 owner, offset, BTRFS_ADD_DELAYED_REF);
2059 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2060 ret = btrfs_add_delayed_tree_ref(trans, bytenr,
2062 root_objectid, (int)owner,
2063 BTRFS_ADD_DELAYED_REF, NULL,
2064 &old_ref_mod, &new_ref_mod);
2066 ret = btrfs_add_delayed_data_ref(trans, bytenr,
2068 root_objectid, owner, offset,
2069 0, BTRFS_ADD_DELAYED_REF,
2070 &old_ref_mod, &new_ref_mod);
2073 if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0) {
2074 bool metadata = owner < BTRFS_FIRST_FREE_OBJECTID;
2076 add_pinned_bytes(fs_info, -num_bytes, metadata, root_objectid);
2083 * __btrfs_inc_extent_ref - insert backreference for a given extent
2085 * @trans: Handle of transaction
2087 * @node: The delayed ref node used to get the bytenr/length for
2088 * extent whose references are incremented.
2090 * @parent: If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/
2091 * BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical
2092 * bytenr of the parent block. Since new extents are always
2093 * created with indirect references, this will only be the case
2094 * when relocating a shared extent. In that case, root_objectid
2095 * will be BTRFS_TREE_RELOC_OBJECTID. Otheriwse, parent must
2098 * @root_objectid: The id of the root where this modification has originated,
2099 * this can be either one of the well-known metadata trees or
2100 * the subvolume id which references this extent.
2102 * @owner: For data extents it is the inode number of the owning file.
2103 * For metadata extents this parameter holds the level in the
2104 * tree of the extent.
2106 * @offset: For metadata extents the offset is ignored and is currently
2107 * always passed as 0. For data extents it is the fileoffset
2108 * this extent belongs to.
2110 * @refs_to_add Number of references to add
2112 * @extent_op Pointer to a structure, holding information necessary when
2113 * updating a tree block's flags
2116 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2117 struct btrfs_delayed_ref_node *node,
2118 u64 parent, u64 root_objectid,
2119 u64 owner, u64 offset, int refs_to_add,
2120 struct btrfs_delayed_extent_op *extent_op)
2122 struct btrfs_path *path;
2123 struct extent_buffer *leaf;
2124 struct btrfs_extent_item *item;
2125 struct btrfs_key key;
2126 u64 bytenr = node->bytenr;
2127 u64 num_bytes = node->num_bytes;
2131 path = btrfs_alloc_path();
2135 path->reada = READA_FORWARD;
2136 path->leave_spinning = 1;
2137 /* this will setup the path even if it fails to insert the back ref */
2138 ret = insert_inline_extent_backref(trans, path, bytenr, num_bytes,
2139 parent, root_objectid, owner,
2140 offset, refs_to_add, extent_op);
2141 if ((ret < 0 && ret != -EAGAIN) || !ret)
2145 * Ok we had -EAGAIN which means we didn't have space to insert and
2146 * inline extent ref, so just update the reference count and add a
2149 leaf = path->nodes[0];
2150 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2151 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2152 refs = btrfs_extent_refs(leaf, item);
2153 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2155 __run_delayed_extent_op(extent_op, leaf, item);
2157 btrfs_mark_buffer_dirty(leaf);
2158 btrfs_release_path(path);
2160 path->reada = READA_FORWARD;
2161 path->leave_spinning = 1;
2162 /* now insert the actual backref */
2163 ret = insert_extent_backref(trans, path, bytenr, parent, root_objectid,
2164 owner, offset, refs_to_add);
2166 btrfs_abort_transaction(trans, ret);
2168 btrfs_free_path(path);
2172 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2173 struct btrfs_delayed_ref_node *node,
2174 struct btrfs_delayed_extent_op *extent_op,
2175 int insert_reserved)
2178 struct btrfs_delayed_data_ref *ref;
2179 struct btrfs_key ins;
2184 ins.objectid = node->bytenr;
2185 ins.offset = node->num_bytes;
2186 ins.type = BTRFS_EXTENT_ITEM_KEY;
2188 ref = btrfs_delayed_node_to_data_ref(node);
2189 trace_run_delayed_data_ref(trans->fs_info, node, ref, node->action);
2191 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2192 parent = ref->parent;
2193 ref_root = ref->root;
2195 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2197 flags |= extent_op->flags_to_set;
2198 ret = alloc_reserved_file_extent(trans, parent, ref_root,
2199 flags, ref->objectid,
2202 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2203 ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root,
2204 ref->objectid, ref->offset,
2205 node->ref_mod, extent_op);
2206 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2207 ret = __btrfs_free_extent(trans, node, parent,
2208 ref_root, ref->objectid,
2209 ref->offset, node->ref_mod,
2217 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2218 struct extent_buffer *leaf,
2219 struct btrfs_extent_item *ei)
2221 u64 flags = btrfs_extent_flags(leaf, ei);
2222 if (extent_op->update_flags) {
2223 flags |= extent_op->flags_to_set;
2224 btrfs_set_extent_flags(leaf, ei, flags);
2227 if (extent_op->update_key) {
2228 struct btrfs_tree_block_info *bi;
2229 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2230 bi = (struct btrfs_tree_block_info *)(ei + 1);
2231 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2235 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2236 struct btrfs_delayed_ref_head *head,
2237 struct btrfs_delayed_extent_op *extent_op)
2239 struct btrfs_fs_info *fs_info = trans->fs_info;
2240 struct btrfs_key key;
2241 struct btrfs_path *path;
2242 struct btrfs_extent_item *ei;
2243 struct extent_buffer *leaf;
2247 int metadata = !extent_op->is_data;
2252 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2255 path = btrfs_alloc_path();
2259 key.objectid = head->bytenr;
2262 key.type = BTRFS_METADATA_ITEM_KEY;
2263 key.offset = extent_op->level;
2265 key.type = BTRFS_EXTENT_ITEM_KEY;
2266 key.offset = head->num_bytes;
2270 path->reada = READA_FORWARD;
2271 path->leave_spinning = 1;
2272 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1);
2279 if (path->slots[0] > 0) {
2281 btrfs_item_key_to_cpu(path->nodes[0], &key,
2283 if (key.objectid == head->bytenr &&
2284 key.type == BTRFS_EXTENT_ITEM_KEY &&
2285 key.offset == head->num_bytes)
2289 btrfs_release_path(path);
2292 key.objectid = head->bytenr;
2293 key.offset = head->num_bytes;
2294 key.type = BTRFS_EXTENT_ITEM_KEY;
2303 leaf = path->nodes[0];
2304 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2306 if (unlikely(item_size < sizeof(*ei))) {
2308 btrfs_print_v0_err(fs_info);
2309 btrfs_abort_transaction(trans, err);
2313 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2314 __run_delayed_extent_op(extent_op, leaf, ei);
2316 btrfs_mark_buffer_dirty(leaf);
2318 btrfs_free_path(path);
2322 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2323 struct btrfs_delayed_ref_node *node,
2324 struct btrfs_delayed_extent_op *extent_op,
2325 int insert_reserved)
2328 struct btrfs_delayed_tree_ref *ref;
2332 ref = btrfs_delayed_node_to_tree_ref(node);
2333 trace_run_delayed_tree_ref(trans->fs_info, node, ref, node->action);
2335 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2336 parent = ref->parent;
2337 ref_root = ref->root;
2339 if (node->ref_mod != 1) {
2340 btrfs_err(trans->fs_info,
2341 "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu",
2342 node->bytenr, node->ref_mod, node->action, ref_root,
2346 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2347 BUG_ON(!extent_op || !extent_op->update_flags);
2348 ret = alloc_reserved_tree_block(trans, node, extent_op);
2349 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2350 ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root,
2351 ref->level, 0, 1, extent_op);
2352 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2353 ret = __btrfs_free_extent(trans, node, parent, ref_root,
2354 ref->level, 0, 1, extent_op);
2361 /* helper function to actually process a single delayed ref entry */
2362 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2363 struct btrfs_delayed_ref_node *node,
2364 struct btrfs_delayed_extent_op *extent_op,
2365 int insert_reserved)
2369 if (trans->aborted) {
2370 if (insert_reserved)
2371 btrfs_pin_extent(trans->fs_info, node->bytenr,
2372 node->num_bytes, 1);
2376 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2377 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2378 ret = run_delayed_tree_ref(trans, node, extent_op,
2380 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2381 node->type == BTRFS_SHARED_DATA_REF_KEY)
2382 ret = run_delayed_data_ref(trans, node, extent_op,
2386 if (ret && insert_reserved)
2387 btrfs_pin_extent(trans->fs_info, node->bytenr,
2388 node->num_bytes, 1);
2392 static inline struct btrfs_delayed_ref_node *
2393 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2395 struct btrfs_delayed_ref_node *ref;
2397 if (RB_EMPTY_ROOT(&head->ref_tree.rb_root))
2401 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2402 * This is to prevent a ref count from going down to zero, which deletes
2403 * the extent item from the extent tree, when there still are references
2404 * to add, which would fail because they would not find the extent item.
2406 if (!list_empty(&head->ref_add_list))
2407 return list_first_entry(&head->ref_add_list,
2408 struct btrfs_delayed_ref_node, add_list);
2410 ref = rb_entry(rb_first_cached(&head->ref_tree),
2411 struct btrfs_delayed_ref_node, ref_node);
2412 ASSERT(list_empty(&ref->add_list));
2416 static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
2417 struct btrfs_delayed_ref_head *head)
2419 spin_lock(&delayed_refs->lock);
2420 head->processing = 0;
2421 delayed_refs->num_heads_ready++;
2422 spin_unlock(&delayed_refs->lock);
2423 btrfs_delayed_ref_unlock(head);
2426 static struct btrfs_delayed_extent_op *cleanup_extent_op(
2427 struct btrfs_delayed_ref_head *head)
2429 struct btrfs_delayed_extent_op *extent_op = head->extent_op;
2434 if (head->must_insert_reserved) {
2435 head->extent_op = NULL;
2436 btrfs_free_delayed_extent_op(extent_op);
2442 static int run_and_cleanup_extent_op(struct btrfs_trans_handle *trans,
2443 struct btrfs_delayed_ref_head *head)
2445 struct btrfs_delayed_extent_op *extent_op;
2448 extent_op = cleanup_extent_op(head);
2451 head->extent_op = NULL;
2452 spin_unlock(&head->lock);
2453 ret = run_delayed_extent_op(trans, head, extent_op);
2454 btrfs_free_delayed_extent_op(extent_op);
2455 return ret ? ret : 1;
2458 void btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info *fs_info,
2459 struct btrfs_delayed_ref_root *delayed_refs,
2460 struct btrfs_delayed_ref_head *head)
2462 int nr_items = 1; /* Dropping this ref head update. */
2464 if (head->total_ref_mod < 0) {
2465 struct btrfs_space_info *space_info;
2469 flags = BTRFS_BLOCK_GROUP_DATA;
2470 else if (head->is_system)
2471 flags = BTRFS_BLOCK_GROUP_SYSTEM;
2473 flags = BTRFS_BLOCK_GROUP_METADATA;
2474 space_info = __find_space_info(fs_info, flags);
2476 percpu_counter_add_batch(&space_info->total_bytes_pinned,
2478 BTRFS_TOTAL_BYTES_PINNED_BATCH);
2481 * We had csum deletions accounted for in our delayed refs rsv,
2482 * we need to drop the csum leaves for this update from our
2485 if (head->is_data) {
2486 spin_lock(&delayed_refs->lock);
2487 delayed_refs->pending_csums -= head->num_bytes;
2488 spin_unlock(&delayed_refs->lock);
2489 nr_items += btrfs_csum_bytes_to_leaves(fs_info,
2494 btrfs_delayed_refs_rsv_release(fs_info, nr_items);
2497 static int cleanup_ref_head(struct btrfs_trans_handle *trans,
2498 struct btrfs_delayed_ref_head *head)
2501 struct btrfs_fs_info *fs_info = trans->fs_info;
2502 struct btrfs_delayed_ref_root *delayed_refs;
2505 delayed_refs = &trans->transaction->delayed_refs;
2507 ret = run_and_cleanup_extent_op(trans, head);
2509 unselect_delayed_ref_head(delayed_refs, head);
2510 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2517 * Need to drop our head ref lock and re-acquire the delayed ref lock
2518 * and then re-check to make sure nobody got added.
2520 spin_unlock(&head->lock);
2521 spin_lock(&delayed_refs->lock);
2522 spin_lock(&head->lock);
2523 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root) || head->extent_op) {
2524 spin_unlock(&head->lock);
2525 spin_unlock(&delayed_refs->lock);
2528 btrfs_delete_ref_head(delayed_refs, head);
2529 spin_unlock(&head->lock);
2530 spin_unlock(&delayed_refs->lock);
2532 if (head->must_insert_reserved) {
2533 btrfs_pin_extent(fs_info, head->bytenr,
2534 head->num_bytes, 1);
2535 if (head->is_data) {
2536 ret = btrfs_del_csums(trans, fs_info, head->bytenr,
2541 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
2543 trace_run_delayed_ref_head(fs_info, head, 0);
2544 btrfs_delayed_ref_unlock(head);
2545 btrfs_put_delayed_ref_head(head);
2549 static struct btrfs_delayed_ref_head *btrfs_obtain_ref_head(
2550 struct btrfs_trans_handle *trans)
2552 struct btrfs_delayed_ref_root *delayed_refs =
2553 &trans->transaction->delayed_refs;
2554 struct btrfs_delayed_ref_head *head = NULL;
2557 spin_lock(&delayed_refs->lock);
2558 head = btrfs_select_ref_head(delayed_refs);
2560 spin_unlock(&delayed_refs->lock);
2565 * Grab the lock that says we are going to process all the refs for
2568 ret = btrfs_delayed_ref_lock(delayed_refs, head);
2569 spin_unlock(&delayed_refs->lock);
2572 * We may have dropped the spin lock to get the head mutex lock, and
2573 * that might have given someone else time to free the head. If that's
2574 * true, it has been removed from our list and we can move on.
2577 head = ERR_PTR(-EAGAIN);
2582 static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans,
2583 struct btrfs_delayed_ref_head *locked_ref,
2584 unsigned long *run_refs)
2586 struct btrfs_fs_info *fs_info = trans->fs_info;
2587 struct btrfs_delayed_ref_root *delayed_refs;
2588 struct btrfs_delayed_extent_op *extent_op;
2589 struct btrfs_delayed_ref_node *ref;
2590 int must_insert_reserved = 0;
2593 delayed_refs = &trans->transaction->delayed_refs;
2595 lockdep_assert_held(&locked_ref->mutex);
2596 lockdep_assert_held(&locked_ref->lock);
2598 while ((ref = select_delayed_ref(locked_ref))) {
2600 btrfs_check_delayed_seq(fs_info, ref->seq)) {
2601 spin_unlock(&locked_ref->lock);
2602 unselect_delayed_ref_head(delayed_refs, locked_ref);
2608 rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree);
2609 RB_CLEAR_NODE(&ref->ref_node);
2610 if (!list_empty(&ref->add_list))
2611 list_del(&ref->add_list);
2613 * When we play the delayed ref, also correct the ref_mod on
2616 switch (ref->action) {
2617 case BTRFS_ADD_DELAYED_REF:
2618 case BTRFS_ADD_DELAYED_EXTENT:
2619 locked_ref->ref_mod -= ref->ref_mod;
2621 case BTRFS_DROP_DELAYED_REF:
2622 locked_ref->ref_mod += ref->ref_mod;
2627 atomic_dec(&delayed_refs->num_entries);
2630 * Record the must_insert_reserved flag before we drop the
2633 must_insert_reserved = locked_ref->must_insert_reserved;
2634 locked_ref->must_insert_reserved = 0;
2636 extent_op = locked_ref->extent_op;
2637 locked_ref->extent_op = NULL;
2638 spin_unlock(&locked_ref->lock);
2640 ret = run_one_delayed_ref(trans, ref, extent_op,
2641 must_insert_reserved);
2643 btrfs_free_delayed_extent_op(extent_op);
2645 unselect_delayed_ref_head(delayed_refs, locked_ref);
2646 btrfs_put_delayed_ref(ref);
2647 btrfs_debug(fs_info, "run_one_delayed_ref returned %d",
2652 btrfs_put_delayed_ref(ref);
2655 spin_lock(&locked_ref->lock);
2656 btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref);
2663 * Returns 0 on success or if called with an already aborted transaction.
2664 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2666 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2669 struct btrfs_fs_info *fs_info = trans->fs_info;
2670 struct btrfs_delayed_ref_root *delayed_refs;
2671 struct btrfs_delayed_ref_head *locked_ref = NULL;
2672 ktime_t start = ktime_get();
2674 unsigned long count = 0;
2675 unsigned long actual_count = 0;
2677 delayed_refs = &trans->transaction->delayed_refs;
2680 locked_ref = btrfs_obtain_ref_head(trans);
2681 if (IS_ERR_OR_NULL(locked_ref)) {
2682 if (PTR_ERR(locked_ref) == -EAGAIN) {
2691 * We need to try and merge add/drops of the same ref since we
2692 * can run into issues with relocate dropping the implicit ref
2693 * and then it being added back again before the drop can
2694 * finish. If we merged anything we need to re-loop so we can
2696 * Or we can get node references of the same type that weren't
2697 * merged when created due to bumps in the tree mod seq, and
2698 * we need to merge them to prevent adding an inline extent
2699 * backref before dropping it (triggering a BUG_ON at
2700 * insert_inline_extent_backref()).
2702 spin_lock(&locked_ref->lock);
2703 btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref);
2705 ret = btrfs_run_delayed_refs_for_head(trans, locked_ref,
2707 if (ret < 0 && ret != -EAGAIN) {
2709 * Error, btrfs_run_delayed_refs_for_head already
2710 * unlocked everything so just bail out
2715 * Success, perform the usual cleanup of a processed
2718 ret = cleanup_ref_head(trans, locked_ref);
2720 /* We dropped our lock, we need to loop. */
2729 * Either success case or btrfs_run_delayed_refs_for_head
2730 * returned -EAGAIN, meaning we need to select another head
2735 } while ((nr != -1 && count < nr) || locked_ref);
2738 * We don't want to include ref heads since we can have empty ref heads
2739 * and those will drastically skew our runtime down since we just do
2740 * accounting, no actual extent tree updates.
2742 if (actual_count > 0) {
2743 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2747 * We weigh the current average higher than our current runtime
2748 * to avoid large swings in the average.
2750 spin_lock(&delayed_refs->lock);
2751 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2752 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2753 spin_unlock(&delayed_refs->lock);
2758 #ifdef SCRAMBLE_DELAYED_REFS
2760 * Normally delayed refs get processed in ascending bytenr order. This
2761 * correlates in most cases to the order added. To expose dependencies on this
2762 * order, we start to process the tree in the middle instead of the beginning
2764 static u64 find_middle(struct rb_root *root)
2766 struct rb_node *n = root->rb_node;
2767 struct btrfs_delayed_ref_node *entry;
2770 u64 first = 0, last = 0;
2774 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2775 first = entry->bytenr;
2779 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2780 last = entry->bytenr;
2785 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2786 WARN_ON(!entry->in_tree);
2788 middle = entry->bytenr;
2801 static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads)
2805 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2806 sizeof(struct btrfs_extent_inline_ref));
2807 if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2808 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2811 * We don't ever fill up leaves all the way so multiply by 2 just to be
2812 * closer to what we're really going to want to use.
2814 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info));
2818 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2819 * would require to store the csums for that many bytes.
2821 u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes)
2824 u64 num_csums_per_leaf;
2827 csum_size = BTRFS_MAX_ITEM_SIZE(fs_info);
2828 num_csums_per_leaf = div64_u64(csum_size,
2829 (u64)btrfs_super_csum_size(fs_info->super_copy));
2830 num_csums = div64_u64(csum_bytes, fs_info->sectorsize);
2831 num_csums += num_csums_per_leaf - 1;
2832 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2836 bool btrfs_check_space_for_delayed_refs(struct btrfs_fs_info *fs_info)
2838 struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv;
2839 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
2843 spin_lock(&global_rsv->lock);
2844 reserved = global_rsv->reserved;
2845 spin_unlock(&global_rsv->lock);
2848 * Since the global reserve is just kind of magic we don't really want
2849 * to rely on it to save our bacon, so if our size is more than the
2850 * delayed_refs_rsv and the global rsv then it's time to think about
2853 spin_lock(&delayed_refs_rsv->lock);
2854 reserved += delayed_refs_rsv->reserved;
2855 if (delayed_refs_rsv->size >= reserved)
2857 spin_unlock(&delayed_refs_rsv->lock);
2861 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans)
2864 atomic_read(&trans->transaction->delayed_refs.num_entries);
2869 avg_runtime = trans->fs_info->avg_delayed_ref_runtime;
2870 val = num_entries * avg_runtime;
2871 if (val >= NSEC_PER_SEC)
2873 if (val >= NSEC_PER_SEC / 2)
2876 return btrfs_check_space_for_delayed_refs(trans->fs_info);
2879 struct async_delayed_refs {
2880 struct btrfs_root *root;
2885 struct completion wait;
2886 struct btrfs_work work;
2889 static inline struct async_delayed_refs *
2890 to_async_delayed_refs(struct btrfs_work *work)
2892 return container_of(work, struct async_delayed_refs, work);
2895 static void delayed_ref_async_start(struct btrfs_work *work)
2897 struct async_delayed_refs *async = to_async_delayed_refs(work);
2898 struct btrfs_trans_handle *trans;
2899 struct btrfs_fs_info *fs_info = async->root->fs_info;
2902 /* if the commit is already started, we don't need to wait here */
2903 if (btrfs_transaction_blocked(fs_info))
2906 trans = btrfs_join_transaction(async->root);
2907 if (IS_ERR(trans)) {
2908 async->error = PTR_ERR(trans);
2912 /* Don't bother flushing if we got into a different transaction */
2913 if (trans->transid > async->transid)
2916 ret = btrfs_run_delayed_refs(trans, async->count);
2920 ret = btrfs_end_transaction(trans);
2921 if (ret && !async->error)
2925 complete(&async->wait);
2930 int btrfs_async_run_delayed_refs(struct btrfs_fs_info *fs_info,
2931 unsigned long count, u64 transid, int wait)
2933 struct async_delayed_refs *async;
2936 async = kmalloc(sizeof(*async), GFP_NOFS);
2940 async->root = fs_info->tree_root;
2941 async->count = count;
2943 async->transid = transid;
2948 init_completion(&async->wait);
2950 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2951 delayed_ref_async_start, NULL, NULL);
2953 btrfs_queue_work(fs_info->extent_workers, &async->work);
2956 wait_for_completion(&async->wait);
2965 * this starts processing the delayed reference count updates and
2966 * extent insertions we have queued up so far. count can be
2967 * 0, which means to process everything in the tree at the start
2968 * of the run (but not newly added entries), or it can be some target
2969 * number you'd like to process.
2971 * Returns 0 on success or if called with an aborted transaction
2972 * Returns <0 on error and aborts the transaction
2974 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2975 unsigned long count)
2977 struct btrfs_fs_info *fs_info = trans->fs_info;
2978 struct rb_node *node;
2979 struct btrfs_delayed_ref_root *delayed_refs;
2980 struct btrfs_delayed_ref_head *head;
2982 int run_all = count == (unsigned long)-1;
2984 /* We'll clean this up in btrfs_cleanup_transaction */
2988 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
2991 delayed_refs = &trans->transaction->delayed_refs;
2993 count = atomic_read(&delayed_refs->num_entries) * 2;
2996 #ifdef SCRAMBLE_DELAYED_REFS
2997 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2999 ret = __btrfs_run_delayed_refs(trans, count);
3001 btrfs_abort_transaction(trans, ret);
3006 btrfs_create_pending_block_groups(trans);
3008 spin_lock(&delayed_refs->lock);
3009 node = rb_first_cached(&delayed_refs->href_root);
3011 spin_unlock(&delayed_refs->lock);
3014 head = rb_entry(node, struct btrfs_delayed_ref_head,
3016 refcount_inc(&head->refs);
3017 spin_unlock(&delayed_refs->lock);
3019 /* Mutex was contended, block until it's released and retry. */
3020 mutex_lock(&head->mutex);
3021 mutex_unlock(&head->mutex);
3023 btrfs_put_delayed_ref_head(head);
3031 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
3032 struct btrfs_fs_info *fs_info,
3033 u64 bytenr, u64 num_bytes, u64 flags,
3034 int level, int is_data)
3036 struct btrfs_delayed_extent_op *extent_op;
3039 extent_op = btrfs_alloc_delayed_extent_op();
3043 extent_op->flags_to_set = flags;
3044 extent_op->update_flags = true;
3045 extent_op->update_key = false;
3046 extent_op->is_data = is_data ? true : false;
3047 extent_op->level = level;
3049 ret = btrfs_add_delayed_extent_op(fs_info, trans, bytenr,
3050 num_bytes, extent_op);
3052 btrfs_free_delayed_extent_op(extent_op);
3056 static noinline int check_delayed_ref(struct btrfs_root *root,
3057 struct btrfs_path *path,
3058 u64 objectid, u64 offset, u64 bytenr)
3060 struct btrfs_delayed_ref_head *head;
3061 struct btrfs_delayed_ref_node *ref;
3062 struct btrfs_delayed_data_ref *data_ref;
3063 struct btrfs_delayed_ref_root *delayed_refs;
3064 struct btrfs_transaction *cur_trans;
3065 struct rb_node *node;
3068 spin_lock(&root->fs_info->trans_lock);
3069 cur_trans = root->fs_info->running_transaction;
3071 refcount_inc(&cur_trans->use_count);
3072 spin_unlock(&root->fs_info->trans_lock);
3076 delayed_refs = &cur_trans->delayed_refs;
3077 spin_lock(&delayed_refs->lock);
3078 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
3080 spin_unlock(&delayed_refs->lock);
3081 btrfs_put_transaction(cur_trans);
3085 if (!mutex_trylock(&head->mutex)) {
3086 refcount_inc(&head->refs);
3087 spin_unlock(&delayed_refs->lock);
3089 btrfs_release_path(path);
3092 * Mutex was contended, block until it's released and let
3095 mutex_lock(&head->mutex);
3096 mutex_unlock(&head->mutex);
3097 btrfs_put_delayed_ref_head(head);
3098 btrfs_put_transaction(cur_trans);
3101 spin_unlock(&delayed_refs->lock);
3103 spin_lock(&head->lock);
3105 * XXX: We should replace this with a proper search function in the
3108 for (node = rb_first_cached(&head->ref_tree); node;
3109 node = rb_next(node)) {
3110 ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node);
3111 /* If it's a shared ref we know a cross reference exists */
3112 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
3117 data_ref = btrfs_delayed_node_to_data_ref(ref);
3120 * If our ref doesn't match the one we're currently looking at
3121 * then we have a cross reference.
3123 if (data_ref->root != root->root_key.objectid ||
3124 data_ref->objectid != objectid ||
3125 data_ref->offset != offset) {
3130 spin_unlock(&head->lock);
3131 mutex_unlock(&head->mutex);
3132 btrfs_put_transaction(cur_trans);
3136 static noinline int check_committed_ref(struct btrfs_root *root,
3137 struct btrfs_path *path,
3138 u64 objectid, u64 offset, u64 bytenr)
3140 struct btrfs_fs_info *fs_info = root->fs_info;
3141 struct btrfs_root *extent_root = fs_info->extent_root;
3142 struct extent_buffer *leaf;
3143 struct btrfs_extent_data_ref *ref;
3144 struct btrfs_extent_inline_ref *iref;
3145 struct btrfs_extent_item *ei;
3146 struct btrfs_key key;
3151 key.objectid = bytenr;
3152 key.offset = (u64)-1;
3153 key.type = BTRFS_EXTENT_ITEM_KEY;
3155 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3158 BUG_ON(ret == 0); /* Corruption */
3161 if (path->slots[0] == 0)
3165 leaf = path->nodes[0];
3166 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3168 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3172 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3173 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3175 if (item_size != sizeof(*ei) +
3176 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3179 if (btrfs_extent_generation(leaf, ei) <=
3180 btrfs_root_last_snapshot(&root->root_item))
3183 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3185 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
3186 if (type != BTRFS_EXTENT_DATA_REF_KEY)
3189 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3190 if (btrfs_extent_refs(leaf, ei) !=
3191 btrfs_extent_data_ref_count(leaf, ref) ||
3192 btrfs_extent_data_ref_root(leaf, ref) !=
3193 root->root_key.objectid ||
3194 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3195 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3203 int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
3206 struct btrfs_path *path;
3209 path = btrfs_alloc_path();
3214 ret = check_committed_ref(root, path, objectid,
3216 if (ret && ret != -ENOENT)
3219 ret = check_delayed_ref(root, path, objectid, offset, bytenr);
3220 } while (ret == -EAGAIN);
3223 btrfs_free_path(path);
3224 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3229 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3230 struct btrfs_root *root,
3231 struct extent_buffer *buf,
3232 int full_backref, int inc)
3234 struct btrfs_fs_info *fs_info = root->fs_info;
3240 struct btrfs_key key;
3241 struct btrfs_file_extent_item *fi;
3245 int (*process_func)(struct btrfs_trans_handle *,
3246 struct btrfs_root *,
3247 u64, u64, u64, u64, u64, u64);
3250 if (btrfs_is_testing(fs_info))
3253 ref_root = btrfs_header_owner(buf);
3254 nritems = btrfs_header_nritems(buf);
3255 level = btrfs_header_level(buf);
3257 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3261 process_func = btrfs_inc_extent_ref;
3263 process_func = btrfs_free_extent;
3266 parent = buf->start;
3270 for (i = 0; i < nritems; i++) {
3272 btrfs_item_key_to_cpu(buf, &key, i);
3273 if (key.type != BTRFS_EXTENT_DATA_KEY)
3275 fi = btrfs_item_ptr(buf, i,
3276 struct btrfs_file_extent_item);
3277 if (btrfs_file_extent_type(buf, fi) ==
3278 BTRFS_FILE_EXTENT_INLINE)
3280 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3284 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3285 key.offset -= btrfs_file_extent_offset(buf, fi);
3286 ret = process_func(trans, root, bytenr, num_bytes,
3287 parent, ref_root, key.objectid,
3292 bytenr = btrfs_node_blockptr(buf, i);
3293 num_bytes = fs_info->nodesize;
3294 ret = process_func(trans, root, bytenr, num_bytes,
3295 parent, ref_root, level - 1, 0);
3305 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3306 struct extent_buffer *buf, int full_backref)
3308 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3311 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3312 struct extent_buffer *buf, int full_backref)
3314 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3317 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3318 struct btrfs_path *path,
3319 struct btrfs_block_group_cache *cache)
3321 struct btrfs_fs_info *fs_info = trans->fs_info;
3323 struct btrfs_root *extent_root = fs_info->extent_root;
3325 struct extent_buffer *leaf;
3327 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3334 leaf = path->nodes[0];
3335 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3336 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3337 btrfs_mark_buffer_dirty(leaf);
3339 btrfs_release_path(path);
3344 static struct btrfs_block_group_cache *
3345 next_block_group(struct btrfs_fs_info *fs_info,
3346 struct btrfs_block_group_cache *cache)
3348 struct rb_node *node;
3350 spin_lock(&fs_info->block_group_cache_lock);
3352 /* If our block group was removed, we need a full search. */
3353 if (RB_EMPTY_NODE(&cache->cache_node)) {
3354 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3356 spin_unlock(&fs_info->block_group_cache_lock);
3357 btrfs_put_block_group(cache);
3358 cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache;
3360 node = rb_next(&cache->cache_node);
3361 btrfs_put_block_group(cache);
3363 cache = rb_entry(node, struct btrfs_block_group_cache,
3365 btrfs_get_block_group(cache);
3368 spin_unlock(&fs_info->block_group_cache_lock);
3372 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3373 struct btrfs_trans_handle *trans,
3374 struct btrfs_path *path)
3376 struct btrfs_fs_info *fs_info = block_group->fs_info;
3377 struct btrfs_root *root = fs_info->tree_root;
3378 struct inode *inode = NULL;
3379 struct extent_changeset *data_reserved = NULL;
3381 int dcs = BTRFS_DC_ERROR;
3387 * If this block group is smaller than 100 megs don't bother caching the
3390 if (block_group->key.offset < (100 * SZ_1M)) {
3391 spin_lock(&block_group->lock);
3392 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3393 spin_unlock(&block_group->lock);
3400 inode = lookup_free_space_inode(fs_info, block_group, path);
3401 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3402 ret = PTR_ERR(inode);
3403 btrfs_release_path(path);
3407 if (IS_ERR(inode)) {
3411 if (block_group->ro)
3414 ret = create_free_space_inode(fs_info, trans, block_group,
3422 * We want to set the generation to 0, that way if anything goes wrong
3423 * from here on out we know not to trust this cache when we load up next
3426 BTRFS_I(inode)->generation = 0;
3427 ret = btrfs_update_inode(trans, root, inode);
3430 * So theoretically we could recover from this, simply set the
3431 * super cache generation to 0 so we know to invalidate the
3432 * cache, but then we'd have to keep track of the block groups
3433 * that fail this way so we know we _have_ to reset this cache
3434 * before the next commit or risk reading stale cache. So to
3435 * limit our exposure to horrible edge cases lets just abort the
3436 * transaction, this only happens in really bad situations
3439 btrfs_abort_transaction(trans, ret);
3444 /* We've already setup this transaction, go ahead and exit */
3445 if (block_group->cache_generation == trans->transid &&
3446 i_size_read(inode)) {
3447 dcs = BTRFS_DC_SETUP;
3451 if (i_size_read(inode) > 0) {
3452 ret = btrfs_check_trunc_cache_free_space(fs_info,
3453 &fs_info->global_block_rsv);
3457 ret = btrfs_truncate_free_space_cache(trans, NULL, inode);
3462 spin_lock(&block_group->lock);
3463 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3464 !btrfs_test_opt(fs_info, SPACE_CACHE)) {
3466 * don't bother trying to write stuff out _if_
3467 * a) we're not cached,
3468 * b) we're with nospace_cache mount option,
3469 * c) we're with v2 space_cache (FREE_SPACE_TREE).
3471 dcs = BTRFS_DC_WRITTEN;
3472 spin_unlock(&block_group->lock);
3475 spin_unlock(&block_group->lock);
3478 * We hit an ENOSPC when setting up the cache in this transaction, just
3479 * skip doing the setup, we've already cleared the cache so we're safe.
3481 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
3487 * Try to preallocate enough space based on how big the block group is.
3488 * Keep in mind this has to include any pinned space which could end up
3489 * taking up quite a bit since it's not folded into the other space
3492 num_pages = div_u64(block_group->key.offset, SZ_256M);
3497 num_pages *= PAGE_SIZE;
3499 ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages);
3503 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3504 num_pages, num_pages,
3507 * Our cache requires contiguous chunks so that we don't modify a bunch
3508 * of metadata or split extents when writing the cache out, which means
3509 * we can enospc if we are heavily fragmented in addition to just normal
3510 * out of space conditions. So if we hit this just skip setting up any
3511 * other block groups for this transaction, maybe we'll unpin enough
3512 * space the next time around.
3515 dcs = BTRFS_DC_SETUP;
3516 else if (ret == -ENOSPC)
3517 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
3522 btrfs_release_path(path);
3524 spin_lock(&block_group->lock);
3525 if (!ret && dcs == BTRFS_DC_SETUP)
3526 block_group->cache_generation = trans->transid;
3527 block_group->disk_cache_state = dcs;
3528 spin_unlock(&block_group->lock);
3530 extent_changeset_free(data_reserved);
3534 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3535 struct btrfs_fs_info *fs_info)
3537 struct btrfs_block_group_cache *cache, *tmp;
3538 struct btrfs_transaction *cur_trans = trans->transaction;
3539 struct btrfs_path *path;
3541 if (list_empty(&cur_trans->dirty_bgs) ||
3542 !btrfs_test_opt(fs_info, SPACE_CACHE))
3545 path = btrfs_alloc_path();
3549 /* Could add new block groups, use _safe just in case */
3550 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3552 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3553 cache_save_setup(cache, trans, path);
3556 btrfs_free_path(path);
3561 * transaction commit does final block group cache writeback during a
3562 * critical section where nothing is allowed to change the FS. This is
3563 * required in order for the cache to actually match the block group,
3564 * but can introduce a lot of latency into the commit.
3566 * So, btrfs_start_dirty_block_groups is here to kick off block group
3567 * cache IO. There's a chance we'll have to redo some of it if the
3568 * block group changes again during the commit, but it greatly reduces
3569 * the commit latency by getting rid of the easy block groups while
3570 * we're still allowing others to join the commit.
3572 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans)
3574 struct btrfs_fs_info *fs_info = trans->fs_info;
3575 struct btrfs_block_group_cache *cache;
3576 struct btrfs_transaction *cur_trans = trans->transaction;
3579 struct btrfs_path *path = NULL;
3581 struct list_head *io = &cur_trans->io_bgs;
3582 int num_started = 0;
3585 spin_lock(&cur_trans->dirty_bgs_lock);
3586 if (list_empty(&cur_trans->dirty_bgs)) {
3587 spin_unlock(&cur_trans->dirty_bgs_lock);
3590 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3591 spin_unlock(&cur_trans->dirty_bgs_lock);
3595 * make sure all the block groups on our dirty list actually
3598 btrfs_create_pending_block_groups(trans);
3601 path = btrfs_alloc_path();
3607 * cache_write_mutex is here only to save us from balance or automatic
3608 * removal of empty block groups deleting this block group while we are
3609 * writing out the cache
3611 mutex_lock(&trans->transaction->cache_write_mutex);
3612 while (!list_empty(&dirty)) {
3613 bool drop_reserve = true;
3615 cache = list_first_entry(&dirty,
3616 struct btrfs_block_group_cache,
3619 * this can happen if something re-dirties a block
3620 * group that is already under IO. Just wait for it to
3621 * finish and then do it all again
3623 if (!list_empty(&cache->io_list)) {
3624 list_del_init(&cache->io_list);
3625 btrfs_wait_cache_io(trans, cache, path);
3626 btrfs_put_block_group(cache);
3631 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3632 * if it should update the cache_state. Don't delete
3633 * until after we wait.
3635 * Since we're not running in the commit critical section
3636 * we need the dirty_bgs_lock to protect from update_block_group
3638 spin_lock(&cur_trans->dirty_bgs_lock);
3639 list_del_init(&cache->dirty_list);
3640 spin_unlock(&cur_trans->dirty_bgs_lock);
3644 cache_save_setup(cache, trans, path);
3646 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3647 cache->io_ctl.inode = NULL;
3648 ret = btrfs_write_out_cache(fs_info, trans,
3650 if (ret == 0 && cache->io_ctl.inode) {
3655 * The cache_write_mutex is protecting the
3656 * io_list, also refer to the definition of
3657 * btrfs_transaction::io_bgs for more details
3659 list_add_tail(&cache->io_list, io);
3662 * if we failed to write the cache, the
3663 * generation will be bad and life goes on
3669 ret = write_one_cache_group(trans, path, cache);
3671 * Our block group might still be attached to the list
3672 * of new block groups in the transaction handle of some
3673 * other task (struct btrfs_trans_handle->new_bgs). This
3674 * means its block group item isn't yet in the extent
3675 * tree. If this happens ignore the error, as we will
3676 * try again later in the critical section of the
3677 * transaction commit.
3679 if (ret == -ENOENT) {
3681 spin_lock(&cur_trans->dirty_bgs_lock);
3682 if (list_empty(&cache->dirty_list)) {
3683 list_add_tail(&cache->dirty_list,
3684 &cur_trans->dirty_bgs);
3685 btrfs_get_block_group(cache);
3686 drop_reserve = false;
3688 spin_unlock(&cur_trans->dirty_bgs_lock);
3690 btrfs_abort_transaction(trans, ret);
3694 /* if it's not on the io list, we need to put the block group */
3696 btrfs_put_block_group(cache);
3698 btrfs_delayed_refs_rsv_release(fs_info, 1);
3704 * Avoid blocking other tasks for too long. It might even save
3705 * us from writing caches for block groups that are going to be
3708 mutex_unlock(&trans->transaction->cache_write_mutex);
3709 mutex_lock(&trans->transaction->cache_write_mutex);
3711 mutex_unlock(&trans->transaction->cache_write_mutex);
3714 * go through delayed refs for all the stuff we've just kicked off
3715 * and then loop back (just once)
3717 ret = btrfs_run_delayed_refs(trans, 0);
3718 if (!ret && loops == 0) {
3720 spin_lock(&cur_trans->dirty_bgs_lock);
3721 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3723 * dirty_bgs_lock protects us from concurrent block group
3724 * deletes too (not just cache_write_mutex).
3726 if (!list_empty(&dirty)) {
3727 spin_unlock(&cur_trans->dirty_bgs_lock);
3730 spin_unlock(&cur_trans->dirty_bgs_lock);
3731 } else if (ret < 0) {
3732 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
3735 btrfs_free_path(path);
3739 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3740 struct btrfs_fs_info *fs_info)
3742 struct btrfs_block_group_cache *cache;
3743 struct btrfs_transaction *cur_trans = trans->transaction;
3746 struct btrfs_path *path;
3747 struct list_head *io = &cur_trans->io_bgs;
3748 int num_started = 0;
3750 path = btrfs_alloc_path();
3755 * Even though we are in the critical section of the transaction commit,
3756 * we can still have concurrent tasks adding elements to this
3757 * transaction's list of dirty block groups. These tasks correspond to
3758 * endio free space workers started when writeback finishes for a
3759 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3760 * allocate new block groups as a result of COWing nodes of the root
3761 * tree when updating the free space inode. The writeback for the space
3762 * caches is triggered by an earlier call to
3763 * btrfs_start_dirty_block_groups() and iterations of the following
3765 * Also we want to do the cache_save_setup first and then run the
3766 * delayed refs to make sure we have the best chance at doing this all
3769 spin_lock(&cur_trans->dirty_bgs_lock);
3770 while (!list_empty(&cur_trans->dirty_bgs)) {
3771 cache = list_first_entry(&cur_trans->dirty_bgs,
3772 struct btrfs_block_group_cache,
3776 * this can happen if cache_save_setup re-dirties a block
3777 * group that is already under IO. Just wait for it to
3778 * finish and then do it all again
3780 if (!list_empty(&cache->io_list)) {
3781 spin_unlock(&cur_trans->dirty_bgs_lock);
3782 list_del_init(&cache->io_list);
3783 btrfs_wait_cache_io(trans, cache, path);
3784 btrfs_put_block_group(cache);
3785 spin_lock(&cur_trans->dirty_bgs_lock);
3789 * don't remove from the dirty list until after we've waited
3792 list_del_init(&cache->dirty_list);
3793 spin_unlock(&cur_trans->dirty_bgs_lock);
3796 cache_save_setup(cache, trans, path);
3799 ret = btrfs_run_delayed_refs(trans,
3800 (unsigned long) -1);
3802 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3803 cache->io_ctl.inode = NULL;
3804 ret = btrfs_write_out_cache(fs_info, trans,
3806 if (ret == 0 && cache->io_ctl.inode) {
3809 list_add_tail(&cache->io_list, io);
3812 * if we failed to write the cache, the
3813 * generation will be bad and life goes on
3819 ret = write_one_cache_group(trans, path, cache);
3821 * One of the free space endio workers might have
3822 * created a new block group while updating a free space
3823 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3824 * and hasn't released its transaction handle yet, in
3825 * which case the new block group is still attached to
3826 * its transaction handle and its creation has not
3827 * finished yet (no block group item in the extent tree
3828 * yet, etc). If this is the case, wait for all free
3829 * space endio workers to finish and retry. This is a
3830 * a very rare case so no need for a more efficient and
3833 if (ret == -ENOENT) {
3834 wait_event(cur_trans->writer_wait,
3835 atomic_read(&cur_trans->num_writers) == 1);
3836 ret = write_one_cache_group(trans, path, cache);
3839 btrfs_abort_transaction(trans, ret);
3842 /* if its not on the io list, we need to put the block group */
3844 btrfs_put_block_group(cache);
3845 btrfs_delayed_refs_rsv_release(fs_info, 1);
3846 spin_lock(&cur_trans->dirty_bgs_lock);
3848 spin_unlock(&cur_trans->dirty_bgs_lock);
3851 * Refer to the definition of io_bgs member for details why it's safe
3852 * to use it without any locking
3854 while (!list_empty(io)) {
3855 cache = list_first_entry(io, struct btrfs_block_group_cache,
3857 list_del_init(&cache->io_list);
3858 btrfs_wait_cache_io(trans, cache, path);
3859 btrfs_put_block_group(cache);
3862 btrfs_free_path(path);
3866 int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
3868 struct btrfs_block_group_cache *block_group;
3871 block_group = btrfs_lookup_block_group(fs_info, bytenr);
3872 if (!block_group || block_group->ro)
3875 btrfs_put_block_group(block_group);
3879 bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3881 struct btrfs_block_group_cache *bg;
3884 bg = btrfs_lookup_block_group(fs_info, bytenr);
3888 spin_lock(&bg->lock);
3892 atomic_inc(&bg->nocow_writers);
3893 spin_unlock(&bg->lock);
3895 /* no put on block group, done by btrfs_dec_nocow_writers */
3897 btrfs_put_block_group(bg);
3903 void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3905 struct btrfs_block_group_cache *bg;
3907 bg = btrfs_lookup_block_group(fs_info, bytenr);
3909 if (atomic_dec_and_test(&bg->nocow_writers))
3910 wake_up_var(&bg->nocow_writers);
3912 * Once for our lookup and once for the lookup done by a previous call
3913 * to btrfs_inc_nocow_writers()
3915 btrfs_put_block_group(bg);
3916 btrfs_put_block_group(bg);
3919 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg)
3921 wait_var_event(&bg->nocow_writers, !atomic_read(&bg->nocow_writers));
3924 static const char *alloc_name(u64 flags)
3927 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3929 case BTRFS_BLOCK_GROUP_METADATA:
3931 case BTRFS_BLOCK_GROUP_DATA:
3933 case BTRFS_BLOCK_GROUP_SYSTEM:
3937 return "invalid-combination";
3941 static int create_space_info(struct btrfs_fs_info *info, u64 flags)
3944 struct btrfs_space_info *space_info;
3948 space_info = kzalloc(sizeof(*space_info), GFP_NOFS);
3952 ret = percpu_counter_init(&space_info->total_bytes_pinned, 0,
3959 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3960 INIT_LIST_HEAD(&space_info->block_groups[i]);
3961 init_rwsem(&space_info->groups_sem);
3962 spin_lock_init(&space_info->lock);
3963 space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3964 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3965 init_waitqueue_head(&space_info->wait);
3966 INIT_LIST_HEAD(&space_info->ro_bgs);
3967 INIT_LIST_HEAD(&space_info->tickets);
3968 INIT_LIST_HEAD(&space_info->priority_tickets);
3970 ret = kobject_init_and_add(&space_info->kobj, &space_info_ktype,
3971 info->space_info_kobj, "%s",
3972 alloc_name(space_info->flags));
3974 percpu_counter_destroy(&space_info->total_bytes_pinned);
3979 list_add_rcu(&space_info->list, &info->space_info);
3980 if (flags & BTRFS_BLOCK_GROUP_DATA)
3981 info->data_sinfo = space_info;
3986 static void update_space_info(struct btrfs_fs_info *info, u64 flags,
3987 u64 total_bytes, u64 bytes_used,
3989 struct btrfs_space_info **space_info)
3991 struct btrfs_space_info *found;
3994 factor = btrfs_bg_type_to_factor(flags);
3996 found = __find_space_info(info, flags);
3998 spin_lock(&found->lock);
3999 found->total_bytes += total_bytes;
4000 found->disk_total += total_bytes * factor;
4001 found->bytes_used += bytes_used;
4002 found->disk_used += bytes_used * factor;
4003 found->bytes_readonly += bytes_readonly;
4004 if (total_bytes > 0)
4006 space_info_add_new_bytes(info, found, total_bytes -
4007 bytes_used - bytes_readonly);
4008 spin_unlock(&found->lock);
4009 *space_info = found;
4012 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
4014 u64 extra_flags = chunk_to_extended(flags) &
4015 BTRFS_EXTENDED_PROFILE_MASK;
4017 write_seqlock(&fs_info->profiles_lock);
4018 if (flags & BTRFS_BLOCK_GROUP_DATA)
4019 fs_info->avail_data_alloc_bits |= extra_flags;
4020 if (flags & BTRFS_BLOCK_GROUP_METADATA)
4021 fs_info->avail_metadata_alloc_bits |= extra_flags;
4022 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4023 fs_info->avail_system_alloc_bits |= extra_flags;
4024 write_sequnlock(&fs_info->profiles_lock);
4028 * returns target flags in extended format or 0 if restripe for this
4029 * chunk_type is not in progress
4031 * should be called with balance_lock held
4033 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
4035 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4041 if (flags & BTRFS_BLOCK_GROUP_DATA &&
4042 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4043 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
4044 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
4045 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4046 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
4047 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
4048 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4049 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
4056 * @flags: available profiles in extended format (see ctree.h)
4058 * Returns reduced profile in chunk format. If profile changing is in
4059 * progress (either running or paused) picks the target profile (if it's
4060 * already available), otherwise falls back to plain reducing.
4062 static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags)
4064 u64 num_devices = fs_info->fs_devices->rw_devices;
4070 * see if restripe for this chunk_type is in progress, if so
4071 * try to reduce to the target profile
4073 spin_lock(&fs_info->balance_lock);
4074 target = get_restripe_target(fs_info, flags);
4076 /* pick target profile only if it's already available */
4077 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
4078 spin_unlock(&fs_info->balance_lock);
4079 return extended_to_chunk(target);
4082 spin_unlock(&fs_info->balance_lock);
4084 /* First, mask out the RAID levels which aren't possible */
4085 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4086 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
4087 allowed |= btrfs_raid_array[raid_type].bg_flag;
4091 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
4092 allowed = BTRFS_BLOCK_GROUP_RAID6;
4093 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
4094 allowed = BTRFS_BLOCK_GROUP_RAID5;
4095 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
4096 allowed = BTRFS_BLOCK_GROUP_RAID10;
4097 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
4098 allowed = BTRFS_BLOCK_GROUP_RAID1;
4099 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
4100 allowed = BTRFS_BLOCK_GROUP_RAID0;
4102 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
4104 return extended_to_chunk(flags | allowed);
4107 static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags)
4114 seq = read_seqbegin(&fs_info->profiles_lock);
4116 if (flags & BTRFS_BLOCK_GROUP_DATA)
4117 flags |= fs_info->avail_data_alloc_bits;
4118 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4119 flags |= fs_info->avail_system_alloc_bits;
4120 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
4121 flags |= fs_info->avail_metadata_alloc_bits;
4122 } while (read_seqretry(&fs_info->profiles_lock, seq));
4124 return btrfs_reduce_alloc_profile(fs_info, flags);
4127 static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
4129 struct btrfs_fs_info *fs_info = root->fs_info;
4134 flags = BTRFS_BLOCK_GROUP_DATA;
4135 else if (root == fs_info->chunk_root)
4136 flags = BTRFS_BLOCK_GROUP_SYSTEM;
4138 flags = BTRFS_BLOCK_GROUP_METADATA;
4140 ret = get_alloc_profile(fs_info, flags);
4144 u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info)
4146 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA);
4149 u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info)
4151 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4154 u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info)
4156 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4159 static u64 btrfs_space_info_used(struct btrfs_space_info *s_info,
4160 bool may_use_included)
4163 return s_info->bytes_used + s_info->bytes_reserved +
4164 s_info->bytes_pinned + s_info->bytes_readonly +
4165 (may_use_included ? s_info->bytes_may_use : 0);
4168 int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes)
4170 struct btrfs_root *root = inode->root;
4171 struct btrfs_fs_info *fs_info = root->fs_info;
4172 struct btrfs_space_info *data_sinfo = fs_info->data_sinfo;
4175 int need_commit = 2;
4176 int have_pinned_space;
4178 /* make sure bytes are sectorsize aligned */
4179 bytes = ALIGN(bytes, fs_info->sectorsize);
4181 if (btrfs_is_free_space_inode(inode)) {
4183 ASSERT(current->journal_info);
4187 /* make sure we have enough space to handle the data first */
4188 spin_lock(&data_sinfo->lock);
4189 used = btrfs_space_info_used(data_sinfo, true);
4191 if (used + bytes > data_sinfo->total_bytes) {
4192 struct btrfs_trans_handle *trans;
4195 * if we don't have enough free bytes in this space then we need
4196 * to alloc a new chunk.
4198 if (!data_sinfo->full) {
4201 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4202 spin_unlock(&data_sinfo->lock);
4204 alloc_target = btrfs_data_alloc_profile(fs_info);
4206 * It is ugly that we don't call nolock join
4207 * transaction for the free space inode case here.
4208 * But it is safe because we only do the data space
4209 * reservation for the free space cache in the
4210 * transaction context, the common join transaction
4211 * just increase the counter of the current transaction
4212 * handler, doesn't try to acquire the trans_lock of
4215 trans = btrfs_join_transaction(root);
4217 return PTR_ERR(trans);
4219 ret = do_chunk_alloc(trans, alloc_target,
4220 CHUNK_ALLOC_NO_FORCE);
4221 btrfs_end_transaction(trans);
4226 have_pinned_space = 1;
4235 * If we don't have enough pinned space to deal with this
4236 * allocation, and no removed chunk in current transaction,
4237 * don't bother committing the transaction.
4239 have_pinned_space = __percpu_counter_compare(
4240 &data_sinfo->total_bytes_pinned,
4241 used + bytes - data_sinfo->total_bytes,
4242 BTRFS_TOTAL_BYTES_PINNED_BATCH);
4243 spin_unlock(&data_sinfo->lock);
4245 /* commit the current transaction and try again */
4250 if (need_commit > 0) {
4251 btrfs_start_delalloc_roots(fs_info, -1);
4252 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0,
4256 trans = btrfs_join_transaction(root);
4258 return PTR_ERR(trans);
4259 if (have_pinned_space >= 0 ||
4260 test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4261 &trans->transaction->flags) ||
4263 ret = btrfs_commit_transaction(trans);
4267 * The cleaner kthread might still be doing iput
4268 * operations. Wait for it to finish so that
4269 * more space is released. We don't need to
4270 * explicitly run the delayed iputs here because
4271 * the commit_transaction would have woken up
4274 ret = btrfs_wait_on_delayed_iputs(fs_info);
4279 btrfs_end_transaction(trans);
4283 trace_btrfs_space_reservation(fs_info,
4284 "space_info:enospc",
4285 data_sinfo->flags, bytes, 1);
4288 update_bytes_may_use(data_sinfo, bytes);
4289 trace_btrfs_space_reservation(fs_info, "space_info",
4290 data_sinfo->flags, bytes, 1);
4291 spin_unlock(&data_sinfo->lock);
4296 int btrfs_check_data_free_space(struct inode *inode,
4297 struct extent_changeset **reserved, u64 start, u64 len)
4299 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4302 /* align the range */
4303 len = round_up(start + len, fs_info->sectorsize) -
4304 round_down(start, fs_info->sectorsize);
4305 start = round_down(start, fs_info->sectorsize);
4307 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len);
4311 /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */
4312 ret = btrfs_qgroup_reserve_data(inode, reserved, start, len);
4314 btrfs_free_reserved_data_space_noquota(inode, start, len);
4321 * Called if we need to clear a data reservation for this inode
4322 * Normally in a error case.
4324 * This one will *NOT* use accurate qgroup reserved space API, just for case
4325 * which we can't sleep and is sure it won't affect qgroup reserved space.
4326 * Like clear_bit_hook().
4328 void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4331 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4332 struct btrfs_space_info *data_sinfo;
4334 /* Make sure the range is aligned to sectorsize */
4335 len = round_up(start + len, fs_info->sectorsize) -
4336 round_down(start, fs_info->sectorsize);
4337 start = round_down(start, fs_info->sectorsize);
4339 data_sinfo = fs_info->data_sinfo;
4340 spin_lock(&data_sinfo->lock);
4341 update_bytes_may_use(data_sinfo, -len);
4342 trace_btrfs_space_reservation(fs_info, "space_info",
4343 data_sinfo->flags, len, 0);
4344 spin_unlock(&data_sinfo->lock);
4348 * Called if we need to clear a data reservation for this inode
4349 * Normally in a error case.
4351 * This one will handle the per-inode data rsv map for accurate reserved
4354 void btrfs_free_reserved_data_space(struct inode *inode,
4355 struct extent_changeset *reserved, u64 start, u64 len)
4357 struct btrfs_root *root = BTRFS_I(inode)->root;
4359 /* Make sure the range is aligned to sectorsize */
4360 len = round_up(start + len, root->fs_info->sectorsize) -
4361 round_down(start, root->fs_info->sectorsize);
4362 start = round_down(start, root->fs_info->sectorsize);
4364 btrfs_free_reserved_data_space_noquota(inode, start, len);
4365 btrfs_qgroup_free_data(inode, reserved, start, len);
4368 static void force_metadata_allocation(struct btrfs_fs_info *info)
4370 struct list_head *head = &info->space_info;
4371 struct btrfs_space_info *found;
4374 list_for_each_entry_rcu(found, head, list) {
4375 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4376 found->force_alloc = CHUNK_ALLOC_FORCE;
4381 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4383 return (global->size << 1);
4386 static int should_alloc_chunk(struct btrfs_fs_info *fs_info,
4387 struct btrfs_space_info *sinfo, int force)
4389 u64 bytes_used = btrfs_space_info_used(sinfo, false);
4392 if (force == CHUNK_ALLOC_FORCE)
4396 * in limited mode, we want to have some free space up to
4397 * about 1% of the FS size.
4399 if (force == CHUNK_ALLOC_LIMITED) {
4400 thresh = btrfs_super_total_bytes(fs_info->super_copy);
4401 thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1));
4403 if (sinfo->total_bytes - bytes_used < thresh)
4407 if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8))
4412 static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type)
4416 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4417 BTRFS_BLOCK_GROUP_RAID0 |
4418 BTRFS_BLOCK_GROUP_RAID5 |
4419 BTRFS_BLOCK_GROUP_RAID6))
4420 num_dev = fs_info->fs_devices->rw_devices;
4421 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4424 num_dev = 1; /* DUP or single */
4430 * If @is_allocation is true, reserve space in the system space info necessary
4431 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4434 void check_system_chunk(struct btrfs_trans_handle *trans, u64 type)
4436 struct btrfs_fs_info *fs_info = trans->fs_info;
4437 struct btrfs_space_info *info;
4444 * Needed because we can end up allocating a system chunk and for an
4445 * atomic and race free space reservation in the chunk block reserve.
4447 lockdep_assert_held(&fs_info->chunk_mutex);
4449 info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4450 spin_lock(&info->lock);
4451 left = info->total_bytes - btrfs_space_info_used(info, true);
4452 spin_unlock(&info->lock);
4454 num_devs = get_profile_num_devs(fs_info, type);
4456 /* num_devs device items to update and 1 chunk item to add or remove */
4457 thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) +
4458 btrfs_calc_trans_metadata_size(fs_info, 1);
4460 if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4461 btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu",
4462 left, thresh, type);
4463 dump_space_info(fs_info, info, 0, 0);
4466 if (left < thresh) {
4467 u64 flags = btrfs_system_alloc_profile(fs_info);
4470 * Ignore failure to create system chunk. We might end up not
4471 * needing it, as we might not need to COW all nodes/leafs from
4472 * the paths we visit in the chunk tree (they were already COWed
4473 * or created in the current transaction for example).
4475 ret = btrfs_alloc_chunk(trans, flags);
4479 ret = btrfs_block_rsv_add(fs_info->chunk_root,
4480 &fs_info->chunk_block_rsv,
4481 thresh, BTRFS_RESERVE_NO_FLUSH);
4483 trans->chunk_bytes_reserved += thresh;
4488 * If force is CHUNK_ALLOC_FORCE:
4489 * - return 1 if it successfully allocates a chunk,
4490 * - return errors including -ENOSPC otherwise.
4491 * If force is NOT CHUNK_ALLOC_FORCE:
4492 * - return 0 if it doesn't need to allocate a new chunk,
4493 * - return 1 if it successfully allocates a chunk,
4494 * - return errors including -ENOSPC otherwise.
4496 static int do_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags,
4499 struct btrfs_fs_info *fs_info = trans->fs_info;
4500 struct btrfs_space_info *space_info;
4501 bool wait_for_alloc = false;
4502 bool should_alloc = false;
4505 /* Don't re-enter if we're already allocating a chunk */
4506 if (trans->allocating_chunk)
4509 space_info = __find_space_info(fs_info, flags);
4513 spin_lock(&space_info->lock);
4514 if (force < space_info->force_alloc)
4515 force = space_info->force_alloc;
4516 should_alloc = should_alloc_chunk(fs_info, space_info, force);
4517 if (space_info->full) {
4518 /* No more free physical space */
4523 spin_unlock(&space_info->lock);
4525 } else if (!should_alloc) {
4526 spin_unlock(&space_info->lock);
4528 } else if (space_info->chunk_alloc) {
4530 * Someone is already allocating, so we need to block
4531 * until this someone is finished and then loop to
4532 * recheck if we should continue with our allocation
4535 wait_for_alloc = true;
4536 spin_unlock(&space_info->lock);
4537 mutex_lock(&fs_info->chunk_mutex);
4538 mutex_unlock(&fs_info->chunk_mutex);
4540 /* Proceed with allocation */
4541 space_info->chunk_alloc = 1;
4542 wait_for_alloc = false;
4543 spin_unlock(&space_info->lock);
4547 } while (wait_for_alloc);
4549 mutex_lock(&fs_info->chunk_mutex);
4550 trans->allocating_chunk = true;
4553 * If we have mixed data/metadata chunks we want to make sure we keep
4554 * allocating mixed chunks instead of individual chunks.
4556 if (btrfs_mixed_space_info(space_info))
4557 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4560 * if we're doing a data chunk, go ahead and make sure that
4561 * we keep a reasonable number of metadata chunks allocated in the
4564 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4565 fs_info->data_chunk_allocations++;
4566 if (!(fs_info->data_chunk_allocations %
4567 fs_info->metadata_ratio))
4568 force_metadata_allocation(fs_info);
4572 * Check if we have enough space in SYSTEM chunk because we may need
4573 * to update devices.
4575 check_system_chunk(trans, flags);
4577 ret = btrfs_alloc_chunk(trans, flags);
4578 trans->allocating_chunk = false;
4580 spin_lock(&space_info->lock);
4583 space_info->full = 1;
4588 space_info->max_extent_size = 0;
4591 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4593 space_info->chunk_alloc = 0;
4594 spin_unlock(&space_info->lock);
4595 mutex_unlock(&fs_info->chunk_mutex);
4597 * When we allocate a new chunk we reserve space in the chunk block
4598 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4599 * add new nodes/leafs to it if we end up needing to do it when
4600 * inserting the chunk item and updating device items as part of the
4601 * second phase of chunk allocation, performed by
4602 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4603 * large number of new block groups to create in our transaction
4604 * handle's new_bgs list to avoid exhausting the chunk block reserve
4605 * in extreme cases - like having a single transaction create many new
4606 * block groups when starting to write out the free space caches of all
4607 * the block groups that were made dirty during the lifetime of the
4610 if (trans->chunk_bytes_reserved >= (u64)SZ_2M)
4611 btrfs_create_pending_block_groups(trans);
4616 static int can_overcommit(struct btrfs_fs_info *fs_info,
4617 struct btrfs_space_info *space_info, u64 bytes,
4618 enum btrfs_reserve_flush_enum flush,
4621 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4628 /* Don't overcommit when in mixed mode. */
4629 if (space_info->flags & BTRFS_BLOCK_GROUP_DATA)
4633 profile = btrfs_system_alloc_profile(fs_info);
4635 profile = btrfs_metadata_alloc_profile(fs_info);
4637 used = btrfs_space_info_used(space_info, false);
4640 * We only want to allow over committing if we have lots of actual space
4641 * free, but if we don't have enough space to handle the global reserve
4642 * space then we could end up having a real enospc problem when trying
4643 * to allocate a chunk or some other such important allocation.
4645 spin_lock(&global_rsv->lock);
4646 space_size = calc_global_rsv_need_space(global_rsv);
4647 spin_unlock(&global_rsv->lock);
4648 if (used + space_size >= space_info->total_bytes)
4651 used += space_info->bytes_may_use;
4653 avail = atomic64_read(&fs_info->free_chunk_space);
4656 * If we have dup, raid1 or raid10 then only half of the free
4657 * space is actually usable. For raid56, the space info used
4658 * doesn't include the parity drive, so we don't have to
4661 factor = btrfs_bg_type_to_factor(profile);
4662 avail = div_u64(avail, factor);
4665 * If we aren't flushing all things, let us overcommit up to
4666 * 1/2th of the space. If we can flush, don't let us overcommit
4667 * too much, let it overcommit up to 1/8 of the space.
4669 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4674 if (used + bytes < space_info->total_bytes + avail)
4679 static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info,
4680 unsigned long nr_pages, int nr_items)
4682 struct super_block *sb = fs_info->sb;
4684 if (down_read_trylock(&sb->s_umount)) {
4685 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4686 up_read(&sb->s_umount);
4689 * We needn't worry the filesystem going from r/w to r/o though
4690 * we don't acquire ->s_umount mutex, because the filesystem
4691 * should guarantee the delalloc inodes list be empty after
4692 * the filesystem is readonly(all dirty pages are written to
4695 btrfs_start_delalloc_roots(fs_info, nr_items);
4696 if (!current->journal_info)
4697 btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1);
4701 static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info,
4707 bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
4708 nr = div64_u64(to_reclaim, bytes);
4714 #define EXTENT_SIZE_PER_ITEM SZ_256K
4717 * shrink metadata reservation for delalloc
4719 static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim,
4720 u64 orig, bool wait_ordered)
4722 struct btrfs_space_info *space_info;
4723 struct btrfs_trans_handle *trans;
4728 unsigned long nr_pages;
4731 /* Calc the number of the pages we need flush for space reservation */
4732 items = calc_reclaim_items_nr(fs_info, to_reclaim);
4733 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4735 trans = (struct btrfs_trans_handle *)current->journal_info;
4736 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4738 delalloc_bytes = percpu_counter_sum_positive(
4739 &fs_info->delalloc_bytes);
4740 if (delalloc_bytes == 0) {
4744 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4749 while (delalloc_bytes && loops < 3) {
4750 nr_pages = min(delalloc_bytes, to_reclaim) >> PAGE_SHIFT;
4753 * Triggers inode writeback for up to nr_pages. This will invoke
4754 * ->writepages callback and trigger delalloc filling
4755 * (btrfs_run_delalloc_range()).
4757 btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items);
4760 * We need to wait for the compressed pages to start before
4763 async_pages = atomic_read(&fs_info->async_delalloc_pages);
4768 * Calculate how many compressed pages we want to be written
4769 * before we continue. I.e if there are more async pages than we
4770 * require wait_event will wait until nr_pages are written.
4772 if (async_pages <= nr_pages)
4775 async_pages -= nr_pages;
4777 wait_event(fs_info->async_submit_wait,
4778 atomic_read(&fs_info->async_delalloc_pages) <=
4781 spin_lock(&space_info->lock);
4782 if (list_empty(&space_info->tickets) &&
4783 list_empty(&space_info->priority_tickets)) {
4784 spin_unlock(&space_info->lock);
4787 spin_unlock(&space_info->lock);
4790 if (wait_ordered && !trans) {
4791 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4793 time_left = schedule_timeout_killable(1);
4797 delalloc_bytes = percpu_counter_sum_positive(
4798 &fs_info->delalloc_bytes);
4802 struct reserve_ticket {
4806 struct list_head list;
4807 wait_queue_head_t wait;
4811 * maybe_commit_transaction - possibly commit the transaction if its ok to
4812 * @root - the root we're allocating for
4813 * @bytes - the number of bytes we want to reserve
4814 * @force - force the commit
4816 * This will check to make sure that committing the transaction will actually
4817 * get us somewhere and then commit the transaction if it does. Otherwise it
4818 * will return -ENOSPC.
4820 static int may_commit_transaction(struct btrfs_fs_info *fs_info,
4821 struct btrfs_space_info *space_info)
4823 struct reserve_ticket *ticket = NULL;
4824 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv;
4825 struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv;
4826 struct btrfs_trans_handle *trans;
4828 u64 reclaim_bytes = 0;
4830 trans = (struct btrfs_trans_handle *)current->journal_info;
4834 spin_lock(&space_info->lock);
4835 if (!list_empty(&space_info->priority_tickets))
4836 ticket = list_first_entry(&space_info->priority_tickets,
4837 struct reserve_ticket, list);
4838 else if (!list_empty(&space_info->tickets))
4839 ticket = list_first_entry(&space_info->tickets,
4840 struct reserve_ticket, list);
4841 bytes_needed = (ticket) ? ticket->bytes : 0;
4842 spin_unlock(&space_info->lock);
4847 trans = btrfs_join_transaction(fs_info->extent_root);
4849 return PTR_ERR(trans);
4852 * See if there is enough pinned space to make this reservation, or if
4853 * we have block groups that are going to be freed, allowing us to
4854 * possibly do a chunk allocation the next loop through.
4856 if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags) ||
4857 __percpu_counter_compare(&space_info->total_bytes_pinned,
4859 BTRFS_TOTAL_BYTES_PINNED_BATCH) >= 0)
4863 * See if there is some space in the delayed insertion reservation for
4866 if (space_info != delayed_rsv->space_info)
4869 spin_lock(&delayed_rsv->lock);
4870 reclaim_bytes += delayed_rsv->reserved;
4871 spin_unlock(&delayed_rsv->lock);
4873 spin_lock(&delayed_refs_rsv->lock);
4874 reclaim_bytes += delayed_refs_rsv->reserved;
4875 spin_unlock(&delayed_refs_rsv->lock);
4876 if (reclaim_bytes >= bytes_needed)
4878 bytes_needed -= reclaim_bytes;
4880 if (__percpu_counter_compare(&space_info->total_bytes_pinned,
4882 BTRFS_TOTAL_BYTES_PINNED_BATCH) < 0)
4886 return btrfs_commit_transaction(trans);
4888 btrfs_end_transaction(trans);
4893 * Try to flush some data based on policy set by @state. This is only advisory
4894 * and may fail for various reasons. The caller is supposed to examine the
4895 * state of @space_info to detect the outcome.
4897 static void flush_space(struct btrfs_fs_info *fs_info,
4898 struct btrfs_space_info *space_info, u64 num_bytes,
4901 struct btrfs_root *root = fs_info->extent_root;
4902 struct btrfs_trans_handle *trans;
4907 case FLUSH_DELAYED_ITEMS_NR:
4908 case FLUSH_DELAYED_ITEMS:
4909 if (state == FLUSH_DELAYED_ITEMS_NR)
4910 nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2;
4914 trans = btrfs_join_transaction(root);
4915 if (IS_ERR(trans)) {
4916 ret = PTR_ERR(trans);
4919 ret = btrfs_run_delayed_items_nr(trans, nr);
4920 btrfs_end_transaction(trans);
4922 case FLUSH_DELALLOC:
4923 case FLUSH_DELALLOC_WAIT:
4924 shrink_delalloc(fs_info, num_bytes * 2, num_bytes,
4925 state == FLUSH_DELALLOC_WAIT);
4927 case FLUSH_DELAYED_REFS_NR:
4928 case FLUSH_DELAYED_REFS:
4929 trans = btrfs_join_transaction(root);
4930 if (IS_ERR(trans)) {
4931 ret = PTR_ERR(trans);
4934 if (state == FLUSH_DELAYED_REFS_NR)
4935 nr = calc_reclaim_items_nr(fs_info, num_bytes);
4938 btrfs_run_delayed_refs(trans, nr);
4939 btrfs_end_transaction(trans);
4942 case ALLOC_CHUNK_FORCE:
4943 trans = btrfs_join_transaction(root);
4944 if (IS_ERR(trans)) {
4945 ret = PTR_ERR(trans);
4948 ret = do_chunk_alloc(trans,
4949 btrfs_metadata_alloc_profile(fs_info),
4950 (state == ALLOC_CHUNK) ?
4951 CHUNK_ALLOC_NO_FORCE : CHUNK_ALLOC_FORCE);
4952 btrfs_end_transaction(trans);
4953 if (ret > 0 || ret == -ENOSPC)
4958 * If we have pending delayed iputs then we could free up a
4959 * bunch of pinned space, so make sure we run the iputs before
4960 * we do our pinned bytes check below.
4962 btrfs_run_delayed_iputs(fs_info);
4963 btrfs_wait_on_delayed_iputs(fs_info);
4965 ret = may_commit_transaction(fs_info, space_info);
4972 trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, state,
4978 btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info,
4979 struct btrfs_space_info *space_info,
4982 struct reserve_ticket *ticket;
4987 list_for_each_entry(ticket, &space_info->tickets, list)
4988 to_reclaim += ticket->bytes;
4989 list_for_each_entry(ticket, &space_info->priority_tickets, list)
4990 to_reclaim += ticket->bytes;
4994 to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M);
4995 if (can_overcommit(fs_info, space_info, to_reclaim,
4996 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
4999 used = btrfs_space_info_used(space_info, true);
5001 if (can_overcommit(fs_info, space_info, SZ_1M,
5002 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
5003 expected = div_factor_fine(space_info->total_bytes, 95);
5005 expected = div_factor_fine(space_info->total_bytes, 90);
5007 if (used > expected)
5008 to_reclaim = used - expected;
5011 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
5012 space_info->bytes_reserved);
5016 static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info,
5017 struct btrfs_space_info *space_info,
5018 u64 used, bool system_chunk)
5020 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
5022 /* If we're just plain full then async reclaim just slows us down. */
5023 if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh)
5026 if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5030 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
5031 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
5034 static bool wake_all_tickets(struct list_head *head)
5036 struct reserve_ticket *ticket;
5038 while (!list_empty(head)) {
5039 ticket = list_first_entry(head, struct reserve_ticket, list);
5040 list_del_init(&ticket->list);
5041 ticket->error = -ENOSPC;
5042 wake_up(&ticket->wait);
5043 if (ticket->bytes != ticket->orig_bytes)
5050 * This is for normal flushers, we can wait all goddamned day if we want to. We
5051 * will loop and continuously try to flush as long as we are making progress.
5052 * We count progress as clearing off tickets each time we have to loop.
5054 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
5056 struct btrfs_fs_info *fs_info;
5057 struct btrfs_space_info *space_info;
5060 int commit_cycles = 0;
5061 u64 last_tickets_id;
5063 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
5064 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5066 spin_lock(&space_info->lock);
5067 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5070 space_info->flush = 0;
5071 spin_unlock(&space_info->lock);
5074 last_tickets_id = space_info->tickets_id;
5075 spin_unlock(&space_info->lock);
5077 flush_state = FLUSH_DELAYED_ITEMS_NR;
5079 flush_space(fs_info, space_info, to_reclaim, flush_state);
5080 spin_lock(&space_info->lock);
5081 if (list_empty(&space_info->tickets)) {
5082 space_info->flush = 0;
5083 spin_unlock(&space_info->lock);
5086 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info,
5089 if (last_tickets_id == space_info->tickets_id) {
5092 last_tickets_id = space_info->tickets_id;
5093 flush_state = FLUSH_DELAYED_ITEMS_NR;
5099 * We don't want to force a chunk allocation until we've tried
5100 * pretty hard to reclaim space. Think of the case where we
5101 * freed up a bunch of space and so have a lot of pinned space
5102 * to reclaim. We would rather use that than possibly create a
5103 * underutilized metadata chunk. So if this is our first run
5104 * through the flushing state machine skip ALLOC_CHUNK_FORCE and
5105 * commit the transaction. If nothing has changed the next go
5106 * around then we can force a chunk allocation.
5108 if (flush_state == ALLOC_CHUNK_FORCE && !commit_cycles)
5111 if (flush_state > COMMIT_TRANS) {
5113 if (commit_cycles > 2) {
5114 if (wake_all_tickets(&space_info->tickets)) {
5115 flush_state = FLUSH_DELAYED_ITEMS_NR;
5118 space_info->flush = 0;
5121 flush_state = FLUSH_DELAYED_ITEMS_NR;
5124 spin_unlock(&space_info->lock);
5125 } while (flush_state <= COMMIT_TRANS);
5128 void btrfs_init_async_reclaim_work(struct work_struct *work)
5130 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
5133 static const enum btrfs_flush_state priority_flush_states[] = {
5134 FLUSH_DELAYED_ITEMS_NR,
5135 FLUSH_DELAYED_ITEMS,
5139 static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
5140 struct btrfs_space_info *space_info,
5141 struct reserve_ticket *ticket)
5146 spin_lock(&space_info->lock);
5147 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5150 spin_unlock(&space_info->lock);
5153 spin_unlock(&space_info->lock);
5157 flush_space(fs_info, space_info, to_reclaim,
5158 priority_flush_states[flush_state]);
5160 spin_lock(&space_info->lock);
5161 if (ticket->bytes == 0) {
5162 spin_unlock(&space_info->lock);
5165 spin_unlock(&space_info->lock);
5166 } while (flush_state < ARRAY_SIZE(priority_flush_states));
5169 static int wait_reserve_ticket(struct btrfs_fs_info *fs_info,
5170 struct btrfs_space_info *space_info,
5171 struct reserve_ticket *ticket)
5175 u64 reclaim_bytes = 0;
5178 spin_lock(&space_info->lock);
5179 while (ticket->bytes > 0 && ticket->error == 0) {
5180 ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE);
5185 spin_unlock(&space_info->lock);
5189 finish_wait(&ticket->wait, &wait);
5190 spin_lock(&space_info->lock);
5193 ret = ticket->error;
5194 if (!list_empty(&ticket->list))
5195 list_del_init(&ticket->list);
5196 if (ticket->bytes && ticket->bytes < ticket->orig_bytes)
5197 reclaim_bytes = ticket->orig_bytes - ticket->bytes;
5198 spin_unlock(&space_info->lock);
5201 space_info_add_old_bytes(fs_info, space_info, reclaim_bytes);
5206 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5207 * @root - the root we're allocating for
5208 * @space_info - the space info we want to allocate from
5209 * @orig_bytes - the number of bytes we want
5210 * @flush - whether or not we can flush to make our reservation
5212 * This will reserve orig_bytes number of bytes from the space info associated
5213 * with the block_rsv. If there is not enough space it will make an attempt to
5214 * flush out space to make room. It will do this by flushing delalloc if
5215 * possible or committing the transaction. If flush is 0 then no attempts to
5216 * regain reservations will be made and this will fail if there is not enough
5219 static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info,
5220 struct btrfs_space_info *space_info,
5222 enum btrfs_reserve_flush_enum flush,
5225 struct reserve_ticket ticket;
5227 u64 reclaim_bytes = 0;
5231 ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL);
5233 spin_lock(&space_info->lock);
5235 used = btrfs_space_info_used(space_info, true);
5238 * If we have enough space then hooray, make our reservation and carry
5239 * on. If not see if we can overcommit, and if we can, hooray carry on.
5240 * If not things get more complicated.
5242 if (used + orig_bytes <= space_info->total_bytes) {
5243 update_bytes_may_use(space_info, orig_bytes);
5244 trace_btrfs_space_reservation(fs_info, "space_info",
5245 space_info->flags, orig_bytes, 1);
5247 } else if (can_overcommit(fs_info, space_info, orig_bytes, flush,
5249 update_bytes_may_use(space_info, orig_bytes);
5250 trace_btrfs_space_reservation(fs_info, "space_info",
5251 space_info->flags, orig_bytes, 1);
5256 * If we couldn't make a reservation then setup our reservation ticket
5257 * and kick the async worker if it's not already running.
5259 * If we are a priority flusher then we just need to add our ticket to
5260 * the list and we will do our own flushing further down.
5262 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
5263 ticket.orig_bytes = orig_bytes;
5264 ticket.bytes = orig_bytes;
5266 init_waitqueue_head(&ticket.wait);
5267 if (flush == BTRFS_RESERVE_FLUSH_ALL) {
5268 list_add_tail(&ticket.list, &space_info->tickets);
5269 if (!space_info->flush) {
5270 space_info->flush = 1;
5271 trace_btrfs_trigger_flush(fs_info,
5275 queue_work(system_unbound_wq,
5276 &fs_info->async_reclaim_work);
5279 list_add_tail(&ticket.list,
5280 &space_info->priority_tickets);
5282 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
5285 * We will do the space reservation dance during log replay,
5286 * which means we won't have fs_info->fs_root set, so don't do
5287 * the async reclaim as we will panic.
5289 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) &&
5290 need_do_async_reclaim(fs_info, space_info,
5291 used, system_chunk) &&
5292 !work_busy(&fs_info->async_reclaim_work)) {
5293 trace_btrfs_trigger_flush(fs_info, space_info->flags,
5294 orig_bytes, flush, "preempt");
5295 queue_work(system_unbound_wq,
5296 &fs_info->async_reclaim_work);
5299 spin_unlock(&space_info->lock);
5300 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
5303 if (flush == BTRFS_RESERVE_FLUSH_ALL)
5304 return wait_reserve_ticket(fs_info, space_info, &ticket);
5307 priority_reclaim_metadata_space(fs_info, space_info, &ticket);
5308 spin_lock(&space_info->lock);
5310 if (ticket.bytes < orig_bytes)
5311 reclaim_bytes = orig_bytes - ticket.bytes;
5312 list_del_init(&ticket.list);
5315 spin_unlock(&space_info->lock);
5318 space_info_add_old_bytes(fs_info, space_info, reclaim_bytes);
5319 ASSERT(list_empty(&ticket.list));
5324 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5325 * @root - the root we're allocating for
5326 * @block_rsv - the block_rsv we're allocating for
5327 * @orig_bytes - the number of bytes we want
5328 * @flush - whether or not we can flush to make our reservation
5330 * This will reserve orig_bytes number of bytes from the space info associated
5331 * with the block_rsv. If there is not enough space it will make an attempt to
5332 * flush out space to make room. It will do this by flushing delalloc if
5333 * possible or committing the transaction. If flush is 0 then no attempts to
5334 * regain reservations will be made and this will fail if there is not enough
5337 static int reserve_metadata_bytes(struct btrfs_root *root,
5338 struct btrfs_block_rsv *block_rsv,
5340 enum btrfs_reserve_flush_enum flush)
5342 struct btrfs_fs_info *fs_info = root->fs_info;
5343 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5345 bool system_chunk = (root == fs_info->chunk_root);
5347 ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info,
5348 orig_bytes, flush, system_chunk);
5349 if (ret == -ENOSPC &&
5350 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5351 if (block_rsv != global_rsv &&
5352 !block_rsv_use_bytes(global_rsv, orig_bytes))
5355 if (ret == -ENOSPC) {
5356 trace_btrfs_space_reservation(fs_info, "space_info:enospc",
5357 block_rsv->space_info->flags,
5360 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG))
5361 dump_space_info(fs_info, block_rsv->space_info,
5367 static struct btrfs_block_rsv *get_block_rsv(
5368 const struct btrfs_trans_handle *trans,
5369 const struct btrfs_root *root)
5371 struct btrfs_fs_info *fs_info = root->fs_info;
5372 struct btrfs_block_rsv *block_rsv = NULL;
5374 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5375 (root == fs_info->csum_root && trans->adding_csums) ||
5376 (root == fs_info->uuid_root))
5377 block_rsv = trans->block_rsv;
5380 block_rsv = root->block_rsv;
5383 block_rsv = &fs_info->empty_block_rsv;
5388 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5392 spin_lock(&block_rsv->lock);
5393 if (block_rsv->reserved >= num_bytes) {
5394 block_rsv->reserved -= num_bytes;
5395 if (block_rsv->reserved < block_rsv->size)
5396 block_rsv->full = 0;
5399 spin_unlock(&block_rsv->lock);
5403 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5404 u64 num_bytes, bool update_size)
5406 spin_lock(&block_rsv->lock);
5407 block_rsv->reserved += num_bytes;
5409 block_rsv->size += num_bytes;
5410 else if (block_rsv->reserved >= block_rsv->size)
5411 block_rsv->full = 1;
5412 spin_unlock(&block_rsv->lock);
5415 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5416 struct btrfs_block_rsv *dest, u64 num_bytes,
5419 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5422 if (global_rsv->space_info != dest->space_info)
5425 spin_lock(&global_rsv->lock);
5426 min_bytes = div_factor(global_rsv->size, min_factor);
5427 if (global_rsv->reserved < min_bytes + num_bytes) {
5428 spin_unlock(&global_rsv->lock);
5431 global_rsv->reserved -= num_bytes;
5432 if (global_rsv->reserved < global_rsv->size)
5433 global_rsv->full = 0;
5434 spin_unlock(&global_rsv->lock);
5436 block_rsv_add_bytes(dest, num_bytes, true);
5441 * btrfs_migrate_to_delayed_refs_rsv - transfer bytes to our delayed refs rsv.
5442 * @fs_info - the fs info for our fs.
5443 * @src - the source block rsv to transfer from.
5444 * @num_bytes - the number of bytes to transfer.
5446 * This transfers up to the num_bytes amount from the src rsv to the
5447 * delayed_refs_rsv. Any extra bytes are returned to the space info.
5449 void btrfs_migrate_to_delayed_refs_rsv(struct btrfs_fs_info *fs_info,
5450 struct btrfs_block_rsv *src,
5453 struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv;
5456 spin_lock(&src->lock);
5457 src->reserved -= num_bytes;
5458 src->size -= num_bytes;
5459 spin_unlock(&src->lock);
5461 spin_lock(&delayed_refs_rsv->lock);
5462 if (delayed_refs_rsv->size > delayed_refs_rsv->reserved) {
5463 u64 delta = delayed_refs_rsv->size -
5464 delayed_refs_rsv->reserved;
5465 if (num_bytes > delta) {
5466 to_free = num_bytes - delta;
5470 to_free = num_bytes;
5475 delayed_refs_rsv->reserved += num_bytes;
5476 if (delayed_refs_rsv->reserved >= delayed_refs_rsv->size)
5477 delayed_refs_rsv->full = 1;
5478 spin_unlock(&delayed_refs_rsv->lock);
5481 trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv",
5484 space_info_add_old_bytes(fs_info, delayed_refs_rsv->space_info,
5489 * btrfs_delayed_refs_rsv_refill - refill based on our delayed refs usage.
5490 * @fs_info - the fs_info for our fs.
5491 * @flush - control how we can flush for this reservation.
5493 * This will refill the delayed block_rsv up to 1 items size worth of space and
5494 * will return -ENOSPC if we can't make the reservation.
5496 int btrfs_delayed_refs_rsv_refill(struct btrfs_fs_info *fs_info,
5497 enum btrfs_reserve_flush_enum flush)
5499 struct btrfs_block_rsv *block_rsv = &fs_info->delayed_refs_rsv;
5500 u64 limit = btrfs_calc_trans_metadata_size(fs_info, 1);
5504 spin_lock(&block_rsv->lock);
5505 if (block_rsv->reserved < block_rsv->size) {
5506 num_bytes = block_rsv->size - block_rsv->reserved;
5507 num_bytes = min(num_bytes, limit);
5509 spin_unlock(&block_rsv->lock);
5514 ret = reserve_metadata_bytes(fs_info->extent_root, block_rsv,
5518 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5519 trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv",
5525 * This is for space we already have accounted in space_info->bytes_may_use, so
5526 * basically when we're returning space from block_rsv's.
5528 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
5529 struct btrfs_space_info *space_info,
5532 struct reserve_ticket *ticket;
5533 struct list_head *head;
5535 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH;
5536 bool check_overcommit = false;
5538 spin_lock(&space_info->lock);
5539 head = &space_info->priority_tickets;
5542 * If we are over our limit then we need to check and see if we can
5543 * overcommit, and if we can't then we just need to free up our space
5544 * and not satisfy any requests.
5546 used = btrfs_space_info_used(space_info, true);
5547 if (used - num_bytes >= space_info->total_bytes)
5548 check_overcommit = true;
5550 while (!list_empty(head) && num_bytes) {
5551 ticket = list_first_entry(head, struct reserve_ticket,
5554 * We use 0 bytes because this space is already reserved, so
5555 * adding the ticket space would be a double count.
5557 if (check_overcommit &&
5558 !can_overcommit(fs_info, space_info, 0, flush, false))
5560 if (num_bytes >= ticket->bytes) {
5561 list_del_init(&ticket->list);
5562 num_bytes -= ticket->bytes;
5564 space_info->tickets_id++;
5565 wake_up(&ticket->wait);
5567 ticket->bytes -= num_bytes;
5572 if (num_bytes && head == &space_info->priority_tickets) {
5573 head = &space_info->tickets;
5574 flush = BTRFS_RESERVE_FLUSH_ALL;
5577 update_bytes_may_use(space_info, -num_bytes);
5578 trace_btrfs_space_reservation(fs_info, "space_info",
5579 space_info->flags, num_bytes, 0);
5580 spin_unlock(&space_info->lock);
5584 * This is for newly allocated space that isn't accounted in
5585 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5586 * we use this helper.
5588 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
5589 struct btrfs_space_info *space_info,
5592 struct reserve_ticket *ticket;
5593 struct list_head *head = &space_info->priority_tickets;
5596 while (!list_empty(head) && num_bytes) {
5597 ticket = list_first_entry(head, struct reserve_ticket,
5599 if (num_bytes >= ticket->bytes) {
5600 trace_btrfs_space_reservation(fs_info, "space_info",
5603 list_del_init(&ticket->list);
5604 num_bytes -= ticket->bytes;
5605 update_bytes_may_use(space_info, ticket->bytes);
5607 space_info->tickets_id++;
5608 wake_up(&ticket->wait);
5610 trace_btrfs_space_reservation(fs_info, "space_info",
5613 update_bytes_may_use(space_info, num_bytes);
5614 ticket->bytes -= num_bytes;
5619 if (num_bytes && head == &space_info->priority_tickets) {
5620 head = &space_info->tickets;
5625 static u64 block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5626 struct btrfs_block_rsv *block_rsv,
5627 struct btrfs_block_rsv *dest, u64 num_bytes,
5628 u64 *qgroup_to_release_ret)
5630 struct btrfs_space_info *space_info = block_rsv->space_info;
5631 u64 qgroup_to_release = 0;
5634 spin_lock(&block_rsv->lock);
5635 if (num_bytes == (u64)-1) {
5636 num_bytes = block_rsv->size;
5637 qgroup_to_release = block_rsv->qgroup_rsv_size;
5639 block_rsv->size -= num_bytes;
5640 if (block_rsv->reserved >= block_rsv->size) {
5641 num_bytes = block_rsv->reserved - block_rsv->size;
5642 block_rsv->reserved = block_rsv->size;
5643 block_rsv->full = 1;
5647 if (block_rsv->qgroup_rsv_reserved >= block_rsv->qgroup_rsv_size) {
5648 qgroup_to_release = block_rsv->qgroup_rsv_reserved -
5649 block_rsv->qgroup_rsv_size;
5650 block_rsv->qgroup_rsv_reserved = block_rsv->qgroup_rsv_size;
5652 qgroup_to_release = 0;
5654 spin_unlock(&block_rsv->lock);
5657 if (num_bytes > 0) {
5659 spin_lock(&dest->lock);
5663 bytes_to_add = dest->size - dest->reserved;
5664 bytes_to_add = min(num_bytes, bytes_to_add);
5665 dest->reserved += bytes_to_add;
5666 if (dest->reserved >= dest->size)
5668 num_bytes -= bytes_to_add;
5670 spin_unlock(&dest->lock);
5673 space_info_add_old_bytes(fs_info, space_info,
5676 if (qgroup_to_release_ret)
5677 *qgroup_to_release_ret = qgroup_to_release;
5681 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
5682 struct btrfs_block_rsv *dst, u64 num_bytes,
5687 ret = block_rsv_use_bytes(src, num_bytes);
5691 block_rsv_add_bytes(dst, num_bytes, update_size);
5695 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5697 memset(rsv, 0, sizeof(*rsv));
5698 spin_lock_init(&rsv->lock);
5702 void btrfs_init_metadata_block_rsv(struct btrfs_fs_info *fs_info,
5703 struct btrfs_block_rsv *rsv,
5704 unsigned short type)
5706 btrfs_init_block_rsv(rsv, type);
5707 rsv->space_info = __find_space_info(fs_info,
5708 BTRFS_BLOCK_GROUP_METADATA);
5711 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info,
5712 unsigned short type)
5714 struct btrfs_block_rsv *block_rsv;
5716 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5720 btrfs_init_metadata_block_rsv(fs_info, block_rsv, type);
5724 void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info,
5725 struct btrfs_block_rsv *rsv)
5729 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
5733 int btrfs_block_rsv_add(struct btrfs_root *root,
5734 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5735 enum btrfs_reserve_flush_enum flush)
5742 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5744 block_rsv_add_bytes(block_rsv, num_bytes, true);
5749 int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor)
5757 spin_lock(&block_rsv->lock);
5758 num_bytes = div_factor(block_rsv->size, min_factor);
5759 if (block_rsv->reserved >= num_bytes)
5761 spin_unlock(&block_rsv->lock);
5766 int btrfs_block_rsv_refill(struct btrfs_root *root,
5767 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5768 enum btrfs_reserve_flush_enum flush)
5776 spin_lock(&block_rsv->lock);
5777 num_bytes = min_reserved;
5778 if (block_rsv->reserved >= num_bytes)
5781 num_bytes -= block_rsv->reserved;
5782 spin_unlock(&block_rsv->lock);
5787 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5789 block_rsv_add_bytes(block_rsv, num_bytes, false);
5796 static void calc_refill_bytes(struct btrfs_block_rsv *block_rsv,
5797 u64 *metadata_bytes, u64 *qgroup_bytes)
5799 *metadata_bytes = 0;
5802 spin_lock(&block_rsv->lock);
5803 if (block_rsv->reserved < block_rsv->size)
5804 *metadata_bytes = block_rsv->size - block_rsv->reserved;
5805 if (block_rsv->qgroup_rsv_reserved < block_rsv->qgroup_rsv_size)
5806 *qgroup_bytes = block_rsv->qgroup_rsv_size -
5807 block_rsv->qgroup_rsv_reserved;
5808 spin_unlock(&block_rsv->lock);
5812 * btrfs_inode_rsv_refill - refill the inode block rsv.
5813 * @inode - the inode we are refilling.
5814 * @flush - the flushing restriction.
5816 * Essentially the same as btrfs_block_rsv_refill, except it uses the
5817 * block_rsv->size as the minimum size. We'll either refill the missing amount
5818 * or return if we already have enough space. This will also handle the reserve
5819 * tracepoint for the reserved amount.
5821 static int btrfs_inode_rsv_refill(struct btrfs_inode *inode,
5822 enum btrfs_reserve_flush_enum flush)
5824 struct btrfs_root *root = inode->root;
5825 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5826 u64 num_bytes, last = 0;
5827 u64 qgroup_num_bytes;
5830 calc_refill_bytes(block_rsv, &num_bytes, &qgroup_num_bytes);
5835 ret = btrfs_qgroup_reserve_meta_prealloc(root, qgroup_num_bytes,
5839 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5841 btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes);
5844 * If we are fragmented we can end up with a lot of
5845 * outstanding extents which will make our size be much
5846 * larger than our reserved amount.
5848 * If the reservation happens here, it might be very
5849 * big though not needed in the end, if the delalloc
5852 * If this is the case try and do the reserve again.
5854 if (flush == BTRFS_RESERVE_FLUSH_ALL)
5855 calc_refill_bytes(block_rsv, &num_bytes,
5860 } while (ret && last != num_bytes);
5863 block_rsv_add_bytes(block_rsv, num_bytes, false);
5864 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5865 btrfs_ino(inode), num_bytes, 1);
5867 /* Don't forget to increase qgroup_rsv_reserved */
5868 spin_lock(&block_rsv->lock);
5869 block_rsv->qgroup_rsv_reserved += qgroup_num_bytes;
5870 spin_unlock(&block_rsv->lock);
5875 static u64 __btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
5876 struct btrfs_block_rsv *block_rsv,
5877 u64 num_bytes, u64 *qgroup_to_release)
5879 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5880 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_refs_rsv;
5881 struct btrfs_block_rsv *target = delayed_rsv;
5883 if (target->full || target == block_rsv)
5884 target = global_rsv;
5886 if (block_rsv->space_info != target->space_info)
5889 return block_rsv_release_bytes(fs_info, block_rsv, target, num_bytes,
5893 void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
5894 struct btrfs_block_rsv *block_rsv,
5897 __btrfs_block_rsv_release(fs_info, block_rsv, num_bytes, NULL);
5901 * btrfs_inode_rsv_release - release any excessive reservation.
5902 * @inode - the inode we need to release from.
5903 * @qgroup_free - free or convert qgroup meta.
5904 * Unlike normal operation, qgroup meta reservation needs to know if we are
5905 * freeing qgroup reservation or just converting it into per-trans. Normally
5906 * @qgroup_free is true for error handling, and false for normal release.
5908 * This is the same as btrfs_block_rsv_release, except that it handles the
5909 * tracepoint for the reservation.
5911 static void btrfs_inode_rsv_release(struct btrfs_inode *inode, bool qgroup_free)
5913 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5914 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5916 u64 qgroup_to_release = 0;
5919 * Since we statically set the block_rsv->size we just want to say we
5920 * are releasing 0 bytes, and then we'll just get the reservation over
5923 released = __btrfs_block_rsv_release(fs_info, block_rsv, 0,
5924 &qgroup_to_release);
5926 trace_btrfs_space_reservation(fs_info, "delalloc",
5927 btrfs_ino(inode), released, 0);
5929 btrfs_qgroup_free_meta_prealloc(inode->root, qgroup_to_release);
5931 btrfs_qgroup_convert_reserved_meta(inode->root,
5936 * btrfs_delayed_refs_rsv_release - release a ref head's reservation.
5937 * @fs_info - the fs_info for our fs.
5938 * @nr - the number of items to drop.
5940 * This drops the delayed ref head's count from the delayed refs rsv and frees
5941 * any excess reservation we had.
5943 void btrfs_delayed_refs_rsv_release(struct btrfs_fs_info *fs_info, int nr)
5945 struct btrfs_block_rsv *block_rsv = &fs_info->delayed_refs_rsv;
5946 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5947 u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, nr);
5950 released = block_rsv_release_bytes(fs_info, block_rsv, global_rsv,
5953 trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv",
5957 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5959 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5960 struct btrfs_space_info *sinfo = block_rsv->space_info;
5964 * The global block rsv is based on the size of the extent tree, the
5965 * checksum tree and the root tree. If the fs is empty we want to set
5966 * it to a minimal amount for safety.
5968 num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) +
5969 btrfs_root_used(&fs_info->csum_root->root_item) +
5970 btrfs_root_used(&fs_info->tree_root->root_item);
5971 num_bytes = max_t(u64, num_bytes, SZ_16M);
5973 spin_lock(&sinfo->lock);
5974 spin_lock(&block_rsv->lock);
5976 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
5978 if (block_rsv->reserved < block_rsv->size) {
5979 num_bytes = btrfs_space_info_used(sinfo, true);
5980 if (sinfo->total_bytes > num_bytes) {
5981 num_bytes = sinfo->total_bytes - num_bytes;
5982 num_bytes = min(num_bytes,
5983 block_rsv->size - block_rsv->reserved);
5984 block_rsv->reserved += num_bytes;
5985 update_bytes_may_use(sinfo, num_bytes);
5986 trace_btrfs_space_reservation(fs_info, "space_info",
5987 sinfo->flags, num_bytes,
5990 } else if (block_rsv->reserved > block_rsv->size) {
5991 num_bytes = block_rsv->reserved - block_rsv->size;
5992 update_bytes_may_use(sinfo, -num_bytes);
5993 trace_btrfs_space_reservation(fs_info, "space_info",
5994 sinfo->flags, num_bytes, 0);
5995 block_rsv->reserved = block_rsv->size;
5998 if (block_rsv->reserved == block_rsv->size)
5999 block_rsv->full = 1;
6001 block_rsv->full = 0;
6003 spin_unlock(&block_rsv->lock);
6004 spin_unlock(&sinfo->lock);
6007 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
6009 struct btrfs_space_info *space_info;
6011 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
6012 fs_info->chunk_block_rsv.space_info = space_info;
6014 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
6015 fs_info->global_block_rsv.space_info = space_info;
6016 fs_info->trans_block_rsv.space_info = space_info;
6017 fs_info->empty_block_rsv.space_info = space_info;
6018 fs_info->delayed_block_rsv.space_info = space_info;
6019 fs_info->delayed_refs_rsv.space_info = space_info;
6021 fs_info->extent_root->block_rsv = &fs_info->delayed_refs_rsv;
6022 fs_info->csum_root->block_rsv = &fs_info->delayed_refs_rsv;
6023 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
6024 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
6025 if (fs_info->quota_root)
6026 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
6027 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
6029 update_global_block_rsv(fs_info);
6032 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
6034 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
6036 WARN_ON(fs_info->trans_block_rsv.size > 0);
6037 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
6038 WARN_ON(fs_info->chunk_block_rsv.size > 0);
6039 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
6040 WARN_ON(fs_info->delayed_block_rsv.size > 0);
6041 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
6042 WARN_ON(fs_info->delayed_refs_rsv.reserved > 0);
6043 WARN_ON(fs_info->delayed_refs_rsv.size > 0);
6047 * btrfs_update_delayed_refs_rsv - adjust the size of the delayed refs rsv
6048 * @trans - the trans that may have generated delayed refs
6050 * This is to be called anytime we may have adjusted trans->delayed_ref_updates,
6051 * it'll calculate the additional size and add it to the delayed_refs_rsv.
6053 void btrfs_update_delayed_refs_rsv(struct btrfs_trans_handle *trans)
6055 struct btrfs_fs_info *fs_info = trans->fs_info;
6056 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_refs_rsv;
6059 if (!trans->delayed_ref_updates)
6062 num_bytes = btrfs_calc_trans_metadata_size(fs_info,
6063 trans->delayed_ref_updates);
6064 spin_lock(&delayed_rsv->lock);
6065 delayed_rsv->size += num_bytes;
6066 delayed_rsv->full = 0;
6067 spin_unlock(&delayed_rsv->lock);
6068 trans->delayed_ref_updates = 0;
6072 * To be called after all the new block groups attached to the transaction
6073 * handle have been created (btrfs_create_pending_block_groups()).
6075 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
6077 struct btrfs_fs_info *fs_info = trans->fs_info;
6079 if (!trans->chunk_bytes_reserved)
6082 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
6084 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
6085 trans->chunk_bytes_reserved, NULL);
6086 trans->chunk_bytes_reserved = 0;
6090 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
6091 * root: the root of the parent directory
6092 * rsv: block reservation
6093 * items: the number of items that we need do reservation
6094 * use_global_rsv: allow fallback to the global block reservation
6096 * This function is used to reserve the space for snapshot/subvolume
6097 * creation and deletion. Those operations are different with the
6098 * common file/directory operations, they change two fs/file trees
6099 * and root tree, the number of items that the qgroup reserves is
6100 * different with the free space reservation. So we can not use
6101 * the space reservation mechanism in start_transaction().
6103 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
6104 struct btrfs_block_rsv *rsv, int items,
6105 bool use_global_rsv)
6107 u64 qgroup_num_bytes = 0;
6110 struct btrfs_fs_info *fs_info = root->fs_info;
6111 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6113 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
6114 /* One for parent inode, two for dir entries */
6115 qgroup_num_bytes = 3 * fs_info->nodesize;
6116 ret = btrfs_qgroup_reserve_meta_prealloc(root,
6117 qgroup_num_bytes, true);
6122 num_bytes = btrfs_calc_trans_metadata_size(fs_info, items);
6123 rsv->space_info = __find_space_info(fs_info,
6124 BTRFS_BLOCK_GROUP_METADATA);
6125 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
6126 BTRFS_RESERVE_FLUSH_ALL);
6128 if (ret == -ENOSPC && use_global_rsv)
6129 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, true);
6131 if (ret && qgroup_num_bytes)
6132 btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes);
6137 void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info,
6138 struct btrfs_block_rsv *rsv)
6140 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
6143 static void btrfs_calculate_inode_block_rsv_size(struct btrfs_fs_info *fs_info,
6144 struct btrfs_inode *inode)
6146 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
6147 u64 reserve_size = 0;
6148 u64 qgroup_rsv_size = 0;
6150 unsigned outstanding_extents;
6152 lockdep_assert_held(&inode->lock);
6153 outstanding_extents = inode->outstanding_extents;
6154 if (outstanding_extents)
6155 reserve_size = btrfs_calc_trans_metadata_size(fs_info,
6156 outstanding_extents + 1);
6157 csum_leaves = btrfs_csum_bytes_to_leaves(fs_info,
6159 reserve_size += btrfs_calc_trans_metadata_size(fs_info,
6162 * For qgroup rsv, the calculation is very simple:
6163 * account one nodesize for each outstanding extent
6165 * This is overestimating in most cases.
6167 qgroup_rsv_size = (u64)outstanding_extents * fs_info->nodesize;
6169 spin_lock(&block_rsv->lock);
6170 block_rsv->size = reserve_size;
6171 block_rsv->qgroup_rsv_size = qgroup_rsv_size;
6172 spin_unlock(&block_rsv->lock);
6175 int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes)
6177 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6178 unsigned nr_extents;
6179 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
6181 bool delalloc_lock = true;
6183 /* If we are a free space inode we need to not flush since we will be in
6184 * the middle of a transaction commit. We also don't need the delalloc
6185 * mutex since we won't race with anybody. We need this mostly to make
6186 * lockdep shut its filthy mouth.
6188 * If we have a transaction open (can happen if we call truncate_block
6189 * from truncate), then we need FLUSH_LIMIT so we don't deadlock.
6191 if (btrfs_is_free_space_inode(inode)) {
6192 flush = BTRFS_RESERVE_NO_FLUSH;
6193 delalloc_lock = false;
6195 if (current->journal_info)
6196 flush = BTRFS_RESERVE_FLUSH_LIMIT;
6198 if (btrfs_transaction_in_commit(fs_info))
6199 schedule_timeout(1);
6203 mutex_lock(&inode->delalloc_mutex);
6205 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6207 /* Add our new extents and calculate the new rsv size. */
6208 spin_lock(&inode->lock);
6209 nr_extents = count_max_extents(num_bytes);
6210 btrfs_mod_outstanding_extents(inode, nr_extents);
6211 inode->csum_bytes += num_bytes;
6212 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6213 spin_unlock(&inode->lock);
6215 ret = btrfs_inode_rsv_refill(inode, flush);
6220 mutex_unlock(&inode->delalloc_mutex);
6224 spin_lock(&inode->lock);
6225 nr_extents = count_max_extents(num_bytes);
6226 btrfs_mod_outstanding_extents(inode, -nr_extents);
6227 inode->csum_bytes -= num_bytes;
6228 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6229 spin_unlock(&inode->lock);
6231 btrfs_inode_rsv_release(inode, true);
6233 mutex_unlock(&inode->delalloc_mutex);
6238 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
6239 * @inode: the inode to release the reservation for.
6240 * @num_bytes: the number of bytes we are releasing.
6241 * @qgroup_free: free qgroup reservation or convert it to per-trans reservation
6243 * This will release the metadata reservation for an inode. This can be called
6244 * once we complete IO for a given set of bytes to release their metadata
6245 * reservations, or on error for the same reason.
6247 void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes,
6250 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6252 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6253 spin_lock(&inode->lock);
6254 inode->csum_bytes -= num_bytes;
6255 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6256 spin_unlock(&inode->lock);
6258 if (btrfs_is_testing(fs_info))
6261 btrfs_inode_rsv_release(inode, qgroup_free);
6265 * btrfs_delalloc_release_extents - release our outstanding_extents
6266 * @inode: the inode to balance the reservation for.
6267 * @num_bytes: the number of bytes we originally reserved with
6268 * @qgroup_free: do we need to free qgroup meta reservation or convert them.
6270 * When we reserve space we increase outstanding_extents for the extents we may
6271 * add. Once we've set the range as delalloc or created our ordered extents we
6272 * have outstanding_extents to track the real usage, so we use this to free our
6273 * temporarily tracked outstanding_extents. This _must_ be used in conjunction
6274 * with btrfs_delalloc_reserve_metadata.
6276 void btrfs_delalloc_release_extents(struct btrfs_inode *inode, u64 num_bytes,
6279 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6280 unsigned num_extents;
6282 spin_lock(&inode->lock);
6283 num_extents = count_max_extents(num_bytes);
6284 btrfs_mod_outstanding_extents(inode, -num_extents);
6285 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6286 spin_unlock(&inode->lock);
6288 if (btrfs_is_testing(fs_info))
6291 btrfs_inode_rsv_release(inode, qgroup_free);
6295 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6297 * @inode: inode we're writing to
6298 * @start: start range we are writing to
6299 * @len: how long the range we are writing to
6300 * @reserved: mandatory parameter, record actually reserved qgroup ranges of
6301 * current reservation.
6303 * This will do the following things
6305 * o reserve space in data space info for num bytes
6306 * and reserve precious corresponding qgroup space
6307 * (Done in check_data_free_space)
6309 * o reserve space for metadata space, based on the number of outstanding
6310 * extents and how much csums will be needed
6311 * also reserve metadata space in a per root over-reserve method.
6312 * o add to the inodes->delalloc_bytes
6313 * o add it to the fs_info's delalloc inodes list.
6314 * (Above 3 all done in delalloc_reserve_metadata)
6316 * Return 0 for success
6317 * Return <0 for error(-ENOSPC or -EQUOT)
6319 int btrfs_delalloc_reserve_space(struct inode *inode,
6320 struct extent_changeset **reserved, u64 start, u64 len)
6324 ret = btrfs_check_data_free_space(inode, reserved, start, len);
6327 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len);
6329 btrfs_free_reserved_data_space(inode, *reserved, start, len);
6334 * btrfs_delalloc_release_space - release data and metadata space for delalloc
6335 * @inode: inode we're releasing space for
6336 * @start: start position of the space already reserved
6337 * @len: the len of the space already reserved
6338 * @release_bytes: the len of the space we consumed or didn't use
6340 * This function will release the metadata space that was not used and will
6341 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
6342 * list if there are no delalloc bytes left.
6343 * Also it will handle the qgroup reserved space.
6345 void btrfs_delalloc_release_space(struct inode *inode,
6346 struct extent_changeset *reserved,
6347 u64 start, u64 len, bool qgroup_free)
6349 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, qgroup_free);
6350 btrfs_free_reserved_data_space(inode, reserved, start, len);
6353 static int update_block_group(struct btrfs_trans_handle *trans,
6354 struct btrfs_fs_info *info, u64 bytenr,
6355 u64 num_bytes, int alloc)
6357 struct btrfs_block_group_cache *cache = NULL;
6358 u64 total = num_bytes;
6364 /* block accounting for super block */
6365 spin_lock(&info->delalloc_root_lock);
6366 old_val = btrfs_super_bytes_used(info->super_copy);
6368 old_val += num_bytes;
6370 old_val -= num_bytes;
6371 btrfs_set_super_bytes_used(info->super_copy, old_val);
6372 spin_unlock(&info->delalloc_root_lock);
6375 cache = btrfs_lookup_block_group(info, bytenr);
6380 factor = btrfs_bg_type_to_factor(cache->flags);
6383 * If this block group has free space cache written out, we
6384 * need to make sure to load it if we are removing space. This
6385 * is because we need the unpinning stage to actually add the
6386 * space back to the block group, otherwise we will leak space.
6388 if (!alloc && cache->cached == BTRFS_CACHE_NO)
6389 cache_block_group(cache, 1);
6391 byte_in_group = bytenr - cache->key.objectid;
6392 WARN_ON(byte_in_group > cache->key.offset);
6394 spin_lock(&cache->space_info->lock);
6395 spin_lock(&cache->lock);
6397 if (btrfs_test_opt(info, SPACE_CACHE) &&
6398 cache->disk_cache_state < BTRFS_DC_CLEAR)
6399 cache->disk_cache_state = BTRFS_DC_CLEAR;
6401 old_val = btrfs_block_group_used(&cache->item);
6402 num_bytes = min(total, cache->key.offset - byte_in_group);
6404 old_val += num_bytes;
6405 btrfs_set_block_group_used(&cache->item, old_val);
6406 cache->reserved -= num_bytes;
6407 cache->space_info->bytes_reserved -= num_bytes;
6408 cache->space_info->bytes_used += num_bytes;
6409 cache->space_info->disk_used += num_bytes * factor;
6410 spin_unlock(&cache->lock);
6411 spin_unlock(&cache->space_info->lock);
6413 old_val -= num_bytes;
6414 btrfs_set_block_group_used(&cache->item, old_val);
6415 cache->pinned += num_bytes;
6416 update_bytes_pinned(cache->space_info, num_bytes);
6417 cache->space_info->bytes_used -= num_bytes;
6418 cache->space_info->disk_used -= num_bytes * factor;
6419 spin_unlock(&cache->lock);
6420 spin_unlock(&cache->space_info->lock);
6422 trace_btrfs_space_reservation(info, "pinned",
6423 cache->space_info->flags,
6425 percpu_counter_add_batch(&cache->space_info->total_bytes_pinned,
6427 BTRFS_TOTAL_BYTES_PINNED_BATCH);
6428 set_extent_dirty(info->pinned_extents,
6429 bytenr, bytenr + num_bytes - 1,
6430 GFP_NOFS | __GFP_NOFAIL);
6433 spin_lock(&trans->transaction->dirty_bgs_lock);
6434 if (list_empty(&cache->dirty_list)) {
6435 list_add_tail(&cache->dirty_list,
6436 &trans->transaction->dirty_bgs);
6437 trans->transaction->num_dirty_bgs++;
6438 trans->delayed_ref_updates++;
6439 btrfs_get_block_group(cache);
6441 spin_unlock(&trans->transaction->dirty_bgs_lock);
6444 * No longer have used bytes in this block group, queue it for
6445 * deletion. We do this after adding the block group to the
6446 * dirty list to avoid races between cleaner kthread and space
6449 if (!alloc && old_val == 0)
6450 btrfs_mark_bg_unused(cache);
6452 btrfs_put_block_group(cache);
6454 bytenr += num_bytes;
6457 /* Modified block groups are accounted for in the delayed_refs_rsv. */
6458 btrfs_update_delayed_refs_rsv(trans);
6462 static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start)
6464 struct btrfs_block_group_cache *cache;
6467 spin_lock(&fs_info->block_group_cache_lock);
6468 bytenr = fs_info->first_logical_byte;
6469 spin_unlock(&fs_info->block_group_cache_lock);
6471 if (bytenr < (u64)-1)
6474 cache = btrfs_lookup_first_block_group(fs_info, search_start);
6478 bytenr = cache->key.objectid;
6479 btrfs_put_block_group(cache);
6484 static int pin_down_extent(struct btrfs_fs_info *fs_info,
6485 struct btrfs_block_group_cache *cache,
6486 u64 bytenr, u64 num_bytes, int reserved)
6488 spin_lock(&cache->space_info->lock);
6489 spin_lock(&cache->lock);
6490 cache->pinned += num_bytes;
6491 update_bytes_pinned(cache->space_info, num_bytes);
6493 cache->reserved -= num_bytes;
6494 cache->space_info->bytes_reserved -= num_bytes;
6496 spin_unlock(&cache->lock);
6497 spin_unlock(&cache->space_info->lock);
6499 trace_btrfs_space_reservation(fs_info, "pinned",
6500 cache->space_info->flags, num_bytes, 1);
6501 percpu_counter_add_batch(&cache->space_info->total_bytes_pinned,
6502 num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH);
6503 set_extent_dirty(fs_info->pinned_extents, bytenr,
6504 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
6509 * this function must be called within transaction
6511 int btrfs_pin_extent(struct btrfs_fs_info *fs_info,
6512 u64 bytenr, u64 num_bytes, int reserved)
6514 struct btrfs_block_group_cache *cache;
6516 cache = btrfs_lookup_block_group(fs_info, bytenr);
6517 BUG_ON(!cache); /* Logic error */
6519 pin_down_extent(fs_info, cache, bytenr, num_bytes, reserved);
6521 btrfs_put_block_group(cache);
6526 * this function must be called within transaction
6528 int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info,
6529 u64 bytenr, u64 num_bytes)
6531 struct btrfs_block_group_cache *cache;
6534 cache = btrfs_lookup_block_group(fs_info, bytenr);
6539 * pull in the free space cache (if any) so that our pin
6540 * removes the free space from the cache. We have load_only set
6541 * to one because the slow code to read in the free extents does check
6542 * the pinned extents.
6544 cache_block_group(cache, 1);
6546 pin_down_extent(fs_info, cache, bytenr, num_bytes, 0);
6548 /* remove us from the free space cache (if we're there at all) */
6549 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6550 btrfs_put_block_group(cache);
6554 static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
6555 u64 start, u64 num_bytes)
6558 struct btrfs_block_group_cache *block_group;
6559 struct btrfs_caching_control *caching_ctl;
6561 block_group = btrfs_lookup_block_group(fs_info, start);
6565 cache_block_group(block_group, 0);
6566 caching_ctl = get_caching_control(block_group);
6570 BUG_ON(!block_group_cache_done(block_group));
6571 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6573 mutex_lock(&caching_ctl->mutex);
6575 if (start >= caching_ctl->progress) {
6576 ret = add_excluded_extent(fs_info, start, num_bytes);
6577 } else if (start + num_bytes <= caching_ctl->progress) {
6578 ret = btrfs_remove_free_space(block_group,
6581 num_bytes = caching_ctl->progress - start;
6582 ret = btrfs_remove_free_space(block_group,
6587 num_bytes = (start + num_bytes) -
6588 caching_ctl->progress;
6589 start = caching_ctl->progress;
6590 ret = add_excluded_extent(fs_info, start, num_bytes);
6593 mutex_unlock(&caching_ctl->mutex);
6594 put_caching_control(caching_ctl);
6596 btrfs_put_block_group(block_group);
6600 int btrfs_exclude_logged_extents(struct extent_buffer *eb)
6602 struct btrfs_fs_info *fs_info = eb->fs_info;
6603 struct btrfs_file_extent_item *item;
6604 struct btrfs_key key;
6609 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
6612 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6613 btrfs_item_key_to_cpu(eb, &key, i);
6614 if (key.type != BTRFS_EXTENT_DATA_KEY)
6616 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6617 found_type = btrfs_file_extent_type(eb, item);
6618 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6620 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6622 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6623 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6624 ret = __exclude_logged_extent(fs_info, key.objectid, key.offset);
6633 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg)
6635 atomic_inc(&bg->reservations);
6638 void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info,
6641 struct btrfs_block_group_cache *bg;
6643 bg = btrfs_lookup_block_group(fs_info, start);
6645 if (atomic_dec_and_test(&bg->reservations))
6646 wake_up_var(&bg->reservations);
6647 btrfs_put_block_group(bg);
6650 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg)
6652 struct btrfs_space_info *space_info = bg->space_info;
6656 if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA))
6660 * Our block group is read only but before we set it to read only,
6661 * some task might have had allocated an extent from it already, but it
6662 * has not yet created a respective ordered extent (and added it to a
6663 * root's list of ordered extents).
6664 * Therefore wait for any task currently allocating extents, since the
6665 * block group's reservations counter is incremented while a read lock
6666 * on the groups' semaphore is held and decremented after releasing
6667 * the read access on that semaphore and creating the ordered extent.
6669 down_write(&space_info->groups_sem);
6670 up_write(&space_info->groups_sem);
6672 wait_var_event(&bg->reservations, !atomic_read(&bg->reservations));
6676 * btrfs_add_reserved_bytes - update the block_group and space info counters
6677 * @cache: The cache we are manipulating
6678 * @ram_bytes: The number of bytes of file content, and will be same to
6679 * @num_bytes except for the compress path.
6680 * @num_bytes: The number of bytes in question
6681 * @delalloc: The blocks are allocated for the delalloc write
6683 * This is called by the allocator when it reserves space. If this is a
6684 * reservation and the block group has become read only we cannot make the
6685 * reservation and return -EAGAIN, otherwise this function always succeeds.
6687 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache,
6688 u64 ram_bytes, u64 num_bytes, int delalloc)
6690 struct btrfs_space_info *space_info = cache->space_info;
6693 spin_lock(&space_info->lock);
6694 spin_lock(&cache->lock);
6698 cache->reserved += num_bytes;
6699 space_info->bytes_reserved += num_bytes;
6700 update_bytes_may_use(space_info, -ram_bytes);
6702 cache->delalloc_bytes += num_bytes;
6704 spin_unlock(&cache->lock);
6705 spin_unlock(&space_info->lock);
6710 * btrfs_free_reserved_bytes - update the block_group and space info counters
6711 * @cache: The cache we are manipulating
6712 * @num_bytes: The number of bytes in question
6713 * @delalloc: The blocks are allocated for the delalloc write
6715 * This is called by somebody who is freeing space that was never actually used
6716 * on disk. For example if you reserve some space for a new leaf in transaction
6717 * A and before transaction A commits you free that leaf, you call this with
6718 * reserve set to 0 in order to clear the reservation.
6721 static void btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache,
6722 u64 num_bytes, int delalloc)
6724 struct btrfs_space_info *space_info = cache->space_info;
6726 spin_lock(&space_info->lock);
6727 spin_lock(&cache->lock);
6729 space_info->bytes_readonly += num_bytes;
6730 cache->reserved -= num_bytes;
6731 space_info->bytes_reserved -= num_bytes;
6732 space_info->max_extent_size = 0;
6735 cache->delalloc_bytes -= num_bytes;
6736 spin_unlock(&cache->lock);
6737 spin_unlock(&space_info->lock);
6739 void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info)
6741 struct btrfs_caching_control *next;
6742 struct btrfs_caching_control *caching_ctl;
6743 struct btrfs_block_group_cache *cache;
6745 down_write(&fs_info->commit_root_sem);
6747 list_for_each_entry_safe(caching_ctl, next,
6748 &fs_info->caching_block_groups, list) {
6749 cache = caching_ctl->block_group;
6750 if (block_group_cache_done(cache)) {
6751 cache->last_byte_to_unpin = (u64)-1;
6752 list_del_init(&caching_ctl->list);
6753 put_caching_control(caching_ctl);
6755 cache->last_byte_to_unpin = caching_ctl->progress;
6759 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6760 fs_info->pinned_extents = &fs_info->freed_extents[1];
6762 fs_info->pinned_extents = &fs_info->freed_extents[0];
6764 up_write(&fs_info->commit_root_sem);
6766 update_global_block_rsv(fs_info);
6770 * Returns the free cluster for the given space info and sets empty_cluster to
6771 * what it should be based on the mount options.
6773 static struct btrfs_free_cluster *
6774 fetch_cluster_info(struct btrfs_fs_info *fs_info,
6775 struct btrfs_space_info *space_info, u64 *empty_cluster)
6777 struct btrfs_free_cluster *ret = NULL;
6780 if (btrfs_mixed_space_info(space_info))
6783 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
6784 ret = &fs_info->meta_alloc_cluster;
6785 if (btrfs_test_opt(fs_info, SSD))
6786 *empty_cluster = SZ_2M;
6788 *empty_cluster = SZ_64K;
6789 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
6790 btrfs_test_opt(fs_info, SSD_SPREAD)) {
6791 *empty_cluster = SZ_2M;
6792 ret = &fs_info->data_alloc_cluster;
6798 static int unpin_extent_range(struct btrfs_fs_info *fs_info,
6800 const bool return_free_space)
6802 struct btrfs_block_group_cache *cache = NULL;
6803 struct btrfs_space_info *space_info;
6804 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6805 struct btrfs_free_cluster *cluster = NULL;
6807 u64 total_unpinned = 0;
6808 u64 empty_cluster = 0;
6811 while (start <= end) {
6814 start >= cache->key.objectid + cache->key.offset) {
6816 btrfs_put_block_group(cache);
6818 cache = btrfs_lookup_block_group(fs_info, start);
6819 BUG_ON(!cache); /* Logic error */
6821 cluster = fetch_cluster_info(fs_info,
6824 empty_cluster <<= 1;
6827 len = cache->key.objectid + cache->key.offset - start;
6828 len = min(len, end + 1 - start);
6830 if (start < cache->last_byte_to_unpin) {
6831 len = min(len, cache->last_byte_to_unpin - start);
6832 if (return_free_space)
6833 btrfs_add_free_space(cache, start, len);
6837 total_unpinned += len;
6838 space_info = cache->space_info;
6841 * If this space cluster has been marked as fragmented and we've
6842 * unpinned enough in this block group to potentially allow a
6843 * cluster to be created inside of it go ahead and clear the
6846 if (cluster && cluster->fragmented &&
6847 total_unpinned > empty_cluster) {
6848 spin_lock(&cluster->lock);
6849 cluster->fragmented = 0;
6850 spin_unlock(&cluster->lock);
6853 spin_lock(&space_info->lock);
6854 spin_lock(&cache->lock);
6855 cache->pinned -= len;
6856 update_bytes_pinned(space_info, -len);
6858 trace_btrfs_space_reservation(fs_info, "pinned",
6859 space_info->flags, len, 0);
6860 space_info->max_extent_size = 0;
6861 percpu_counter_add_batch(&space_info->total_bytes_pinned,
6862 -len, BTRFS_TOTAL_BYTES_PINNED_BATCH);
6864 space_info->bytes_readonly += len;
6867 spin_unlock(&cache->lock);
6868 if (!readonly && return_free_space &&
6869 global_rsv->space_info == space_info) {
6872 spin_lock(&global_rsv->lock);
6873 if (!global_rsv->full) {
6874 to_add = min(len, global_rsv->size -
6875 global_rsv->reserved);
6876 global_rsv->reserved += to_add;
6877 update_bytes_may_use(space_info, to_add);
6878 if (global_rsv->reserved >= global_rsv->size)
6879 global_rsv->full = 1;
6880 trace_btrfs_space_reservation(fs_info,
6886 spin_unlock(&global_rsv->lock);
6887 /* Add to any tickets we may have */
6889 space_info_add_new_bytes(fs_info, space_info,
6892 spin_unlock(&space_info->lock);
6896 btrfs_put_block_group(cache);
6900 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans)
6902 struct btrfs_fs_info *fs_info = trans->fs_info;
6903 struct btrfs_block_group_cache *block_group, *tmp;
6904 struct list_head *deleted_bgs;
6905 struct extent_io_tree *unpin;
6910 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6911 unpin = &fs_info->freed_extents[1];
6913 unpin = &fs_info->freed_extents[0];
6915 while (!trans->aborted) {
6916 struct extent_state *cached_state = NULL;
6918 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6919 ret = find_first_extent_bit(unpin, 0, &start, &end,
6920 EXTENT_DIRTY, &cached_state);
6922 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6926 if (btrfs_test_opt(fs_info, DISCARD))
6927 ret = btrfs_discard_extent(fs_info, start,
6928 end + 1 - start, NULL);
6930 clear_extent_dirty(unpin, start, end, &cached_state);
6931 unpin_extent_range(fs_info, start, end, true);
6932 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6933 free_extent_state(cached_state);
6938 * Transaction is finished. We don't need the lock anymore. We
6939 * do need to clean up the block groups in case of a transaction
6942 deleted_bgs = &trans->transaction->deleted_bgs;
6943 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6947 if (!trans->aborted)
6948 ret = btrfs_discard_extent(fs_info,
6949 block_group->key.objectid,
6950 block_group->key.offset,
6953 list_del_init(&block_group->bg_list);
6954 btrfs_put_block_group_trimming(block_group);
6955 btrfs_put_block_group(block_group);
6958 const char *errstr = btrfs_decode_error(ret);
6960 "discard failed while removing blockgroup: errno=%d %s",
6968 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6969 struct btrfs_delayed_ref_node *node, u64 parent,
6970 u64 root_objectid, u64 owner_objectid,
6971 u64 owner_offset, int refs_to_drop,
6972 struct btrfs_delayed_extent_op *extent_op)
6974 struct btrfs_fs_info *info = trans->fs_info;
6975 struct btrfs_key key;
6976 struct btrfs_path *path;
6977 struct btrfs_root *extent_root = info->extent_root;
6978 struct extent_buffer *leaf;
6979 struct btrfs_extent_item *ei;
6980 struct btrfs_extent_inline_ref *iref;
6983 int extent_slot = 0;
6984 int found_extent = 0;
6988 u64 bytenr = node->bytenr;
6989 u64 num_bytes = node->num_bytes;
6991 bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
6993 path = btrfs_alloc_path();
6997 path->reada = READA_FORWARD;
6998 path->leave_spinning = 1;
7000 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
7001 BUG_ON(!is_data && refs_to_drop != 1);
7004 skinny_metadata = false;
7006 ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes,
7007 parent, root_objectid, owner_objectid,
7010 extent_slot = path->slots[0];
7011 while (extent_slot >= 0) {
7012 btrfs_item_key_to_cpu(path->nodes[0], &key,
7014 if (key.objectid != bytenr)
7016 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
7017 key.offset == num_bytes) {
7021 if (key.type == BTRFS_METADATA_ITEM_KEY &&
7022 key.offset == owner_objectid) {
7026 if (path->slots[0] - extent_slot > 5)
7031 if (!found_extent) {
7033 ret = remove_extent_backref(trans, path, NULL,
7035 is_data, &last_ref);
7037 btrfs_abort_transaction(trans, ret);
7040 btrfs_release_path(path);
7041 path->leave_spinning = 1;
7043 key.objectid = bytenr;
7044 key.type = BTRFS_EXTENT_ITEM_KEY;
7045 key.offset = num_bytes;
7047 if (!is_data && skinny_metadata) {
7048 key.type = BTRFS_METADATA_ITEM_KEY;
7049 key.offset = owner_objectid;
7052 ret = btrfs_search_slot(trans, extent_root,
7054 if (ret > 0 && skinny_metadata && path->slots[0]) {
7056 * Couldn't find our skinny metadata item,
7057 * see if we have ye olde extent item.
7060 btrfs_item_key_to_cpu(path->nodes[0], &key,
7062 if (key.objectid == bytenr &&
7063 key.type == BTRFS_EXTENT_ITEM_KEY &&
7064 key.offset == num_bytes)
7068 if (ret > 0 && skinny_metadata) {
7069 skinny_metadata = false;
7070 key.objectid = bytenr;
7071 key.type = BTRFS_EXTENT_ITEM_KEY;
7072 key.offset = num_bytes;
7073 btrfs_release_path(path);
7074 ret = btrfs_search_slot(trans, extent_root,
7080 "umm, got %d back from search, was looking for %llu",
7083 btrfs_print_leaf(path->nodes[0]);
7086 btrfs_abort_transaction(trans, ret);
7089 extent_slot = path->slots[0];
7091 } else if (WARN_ON(ret == -ENOENT)) {
7092 btrfs_print_leaf(path->nodes[0]);
7094 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
7095 bytenr, parent, root_objectid, owner_objectid,
7097 btrfs_abort_transaction(trans, ret);
7100 btrfs_abort_transaction(trans, ret);
7104 leaf = path->nodes[0];
7105 item_size = btrfs_item_size_nr(leaf, extent_slot);
7106 if (unlikely(item_size < sizeof(*ei))) {
7108 btrfs_print_v0_err(info);
7109 btrfs_abort_transaction(trans, ret);
7112 ei = btrfs_item_ptr(leaf, extent_slot,
7113 struct btrfs_extent_item);
7114 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
7115 key.type == BTRFS_EXTENT_ITEM_KEY) {
7116 struct btrfs_tree_block_info *bi;
7117 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
7118 bi = (struct btrfs_tree_block_info *)(ei + 1);
7119 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
7122 refs = btrfs_extent_refs(leaf, ei);
7123 if (refs < refs_to_drop) {
7125 "trying to drop %d refs but we only have %Lu for bytenr %Lu",
7126 refs_to_drop, refs, bytenr);
7128 btrfs_abort_transaction(trans, ret);
7131 refs -= refs_to_drop;
7135 __run_delayed_extent_op(extent_op, leaf, ei);
7137 * In the case of inline back ref, reference count will
7138 * be updated by remove_extent_backref
7141 BUG_ON(!found_extent);
7143 btrfs_set_extent_refs(leaf, ei, refs);
7144 btrfs_mark_buffer_dirty(leaf);
7147 ret = remove_extent_backref(trans, path, iref,
7148 refs_to_drop, is_data,
7151 btrfs_abort_transaction(trans, ret);
7157 BUG_ON(is_data && refs_to_drop !=
7158 extent_data_ref_count(path, iref));
7160 BUG_ON(path->slots[0] != extent_slot);
7162 BUG_ON(path->slots[0] != extent_slot + 1);
7163 path->slots[0] = extent_slot;
7169 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
7172 btrfs_abort_transaction(trans, ret);
7175 btrfs_release_path(path);
7178 ret = btrfs_del_csums(trans, info, bytenr, num_bytes);
7180 btrfs_abort_transaction(trans, ret);
7185 ret = add_to_free_space_tree(trans, bytenr, num_bytes);
7187 btrfs_abort_transaction(trans, ret);
7191 ret = update_block_group(trans, info, bytenr, num_bytes, 0);
7193 btrfs_abort_transaction(trans, ret);
7197 btrfs_release_path(path);
7200 btrfs_free_path(path);
7205 * when we free an block, it is possible (and likely) that we free the last
7206 * delayed ref for that extent as well. This searches the delayed ref tree for
7207 * a given extent, and if there are no other delayed refs to be processed, it
7208 * removes it from the tree.
7210 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
7213 struct btrfs_delayed_ref_head *head;
7214 struct btrfs_delayed_ref_root *delayed_refs;
7217 delayed_refs = &trans->transaction->delayed_refs;
7218 spin_lock(&delayed_refs->lock);
7219 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
7221 goto out_delayed_unlock;
7223 spin_lock(&head->lock);
7224 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root))
7227 if (cleanup_extent_op(head) != NULL)
7231 * waiting for the lock here would deadlock. If someone else has it
7232 * locked they are already in the process of dropping it anyway
7234 if (!mutex_trylock(&head->mutex))
7237 btrfs_delete_ref_head(delayed_refs, head);
7238 head->processing = 0;
7240 spin_unlock(&head->lock);
7241 spin_unlock(&delayed_refs->lock);
7243 BUG_ON(head->extent_op);
7244 if (head->must_insert_reserved)
7247 btrfs_cleanup_ref_head_accounting(trans->fs_info, delayed_refs, head);
7248 mutex_unlock(&head->mutex);
7249 btrfs_put_delayed_ref_head(head);
7252 spin_unlock(&head->lock);
7255 spin_unlock(&delayed_refs->lock);
7259 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
7260 struct btrfs_root *root,
7261 struct extent_buffer *buf,
7262 u64 parent, int last_ref)
7264 struct btrfs_fs_info *fs_info = root->fs_info;
7268 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7269 int old_ref_mod, new_ref_mod;
7271 btrfs_ref_tree_mod(root, buf->start, buf->len, parent,
7272 root->root_key.objectid,
7273 btrfs_header_level(buf), 0,
7274 BTRFS_DROP_DELAYED_REF);
7275 ret = btrfs_add_delayed_tree_ref(trans, buf->start,
7277 root->root_key.objectid,
7278 btrfs_header_level(buf),
7279 BTRFS_DROP_DELAYED_REF, NULL,
7280 &old_ref_mod, &new_ref_mod);
7281 BUG_ON(ret); /* -ENOMEM */
7282 pin = old_ref_mod >= 0 && new_ref_mod < 0;
7285 if (last_ref && btrfs_header_generation(buf) == trans->transid) {
7286 struct btrfs_block_group_cache *cache;
7288 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7289 ret = check_ref_cleanup(trans, buf->start);
7295 cache = btrfs_lookup_block_group(fs_info, buf->start);
7297 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
7298 pin_down_extent(fs_info, cache, buf->start,
7300 btrfs_put_block_group(cache);
7304 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
7306 btrfs_add_free_space(cache, buf->start, buf->len);
7307 btrfs_free_reserved_bytes(cache, buf->len, 0);
7308 btrfs_put_block_group(cache);
7309 trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
7313 add_pinned_bytes(fs_info, buf->len, true,
7314 root->root_key.objectid);
7318 * Deleting the buffer, clear the corrupt flag since it doesn't
7321 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
7325 /* Can return -ENOMEM */
7326 int btrfs_free_extent(struct btrfs_trans_handle *trans,
7327 struct btrfs_root *root,
7328 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
7329 u64 owner, u64 offset)
7331 struct btrfs_fs_info *fs_info = root->fs_info;
7332 int old_ref_mod, new_ref_mod;
7335 if (btrfs_is_testing(fs_info))
7338 if (root_objectid != BTRFS_TREE_LOG_OBJECTID)
7339 btrfs_ref_tree_mod(root, bytenr, num_bytes, parent,
7340 root_objectid, owner, offset,
7341 BTRFS_DROP_DELAYED_REF);
7344 * tree log blocks never actually go into the extent allocation
7345 * tree, just update pinning info and exit early.
7347 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
7348 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
7349 /* unlocks the pinned mutex */
7350 btrfs_pin_extent(fs_info, bytenr, num_bytes, 1);
7351 old_ref_mod = new_ref_mod = 0;
7353 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
7354 ret = btrfs_add_delayed_tree_ref(trans, bytenr,
7356 root_objectid, (int)owner,
7357 BTRFS_DROP_DELAYED_REF, NULL,
7358 &old_ref_mod, &new_ref_mod);
7360 ret = btrfs_add_delayed_data_ref(trans, bytenr,
7362 root_objectid, owner, offset,
7363 0, BTRFS_DROP_DELAYED_REF,
7364 &old_ref_mod, &new_ref_mod);
7367 if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0) {
7368 bool metadata = owner < BTRFS_FIRST_FREE_OBJECTID;
7370 add_pinned_bytes(fs_info, num_bytes, metadata, root_objectid);
7377 * when we wait for progress in the block group caching, its because
7378 * our allocation attempt failed at least once. So, we must sleep
7379 * and let some progress happen before we try again.
7381 * This function will sleep at least once waiting for new free space to
7382 * show up, and then it will check the block group free space numbers
7383 * for our min num_bytes. Another option is to have it go ahead
7384 * and look in the rbtree for a free extent of a given size, but this
7387 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7388 * any of the information in this block group.
7390 static noinline void
7391 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
7394 struct btrfs_caching_control *caching_ctl;
7396 caching_ctl = get_caching_control(cache);
7400 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
7401 (cache->free_space_ctl->free_space >= num_bytes));
7403 put_caching_control(caching_ctl);
7407 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
7409 struct btrfs_caching_control *caching_ctl;
7412 caching_ctl = get_caching_control(cache);
7414 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
7416 wait_event(caching_ctl->wait, block_group_cache_done(cache));
7417 if (cache->cached == BTRFS_CACHE_ERROR)
7419 put_caching_control(caching_ctl);
7423 enum btrfs_loop_type {
7424 LOOP_CACHING_NOWAIT = 0,
7425 LOOP_CACHING_WAIT = 1,
7426 LOOP_ALLOC_CHUNK = 2,
7427 LOOP_NO_EMPTY_SIZE = 3,
7431 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
7435 down_read(&cache->data_rwsem);
7439 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
7442 btrfs_get_block_group(cache);
7444 down_read(&cache->data_rwsem);
7447 static struct btrfs_block_group_cache *
7448 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
7449 struct btrfs_free_cluster *cluster,
7452 struct btrfs_block_group_cache *used_bg = NULL;
7454 spin_lock(&cluster->refill_lock);
7456 used_bg = cluster->block_group;
7460 if (used_bg == block_group)
7463 btrfs_get_block_group(used_bg);
7468 if (down_read_trylock(&used_bg->data_rwsem))
7471 spin_unlock(&cluster->refill_lock);
7473 /* We should only have one-level nested. */
7474 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
7476 spin_lock(&cluster->refill_lock);
7477 if (used_bg == cluster->block_group)
7480 up_read(&used_bg->data_rwsem);
7481 btrfs_put_block_group(used_bg);
7486 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
7490 up_read(&cache->data_rwsem);
7491 btrfs_put_block_group(cache);
7495 * Structure used internally for find_free_extent() function. Wraps needed
7498 struct find_free_extent_ctl {
7499 /* Basic allocation info */
7506 /* Where to start the search inside the bg */
7509 /* For clustered allocation */
7512 bool have_caching_bg;
7513 bool orig_have_caching_bg;
7515 /* RAID index, converted from flags */
7519 * Current loop number, check find_free_extent_update_loop() for details
7524 * Whether we're refilling a cluster, if true we need to re-search
7525 * current block group but don't try to refill the cluster again.
7527 bool retry_clustered;
7530 * Whether we're updating free space cache, if true we need to re-search
7531 * current block group but don't try updating free space cache again.
7533 bool retry_unclustered;
7535 /* If current block group is cached */
7538 /* Max contiguous hole found */
7539 u64 max_extent_size;
7541 /* Total free space from free space cache, not always contiguous */
7542 u64 total_free_space;
7550 * Helper function for find_free_extent().
7552 * Return -ENOENT to inform caller that we need fallback to unclustered mode.
7553 * Return -EAGAIN to inform caller that we need to re-search this block group
7554 * Return >0 to inform caller that we find nothing
7555 * Return 0 means we have found a location and set ffe_ctl->found_offset.
7557 static int find_free_extent_clustered(struct btrfs_block_group_cache *bg,
7558 struct btrfs_free_cluster *last_ptr,
7559 struct find_free_extent_ctl *ffe_ctl,
7560 struct btrfs_block_group_cache **cluster_bg_ret)
7562 struct btrfs_fs_info *fs_info = bg->fs_info;
7563 struct btrfs_block_group_cache *cluster_bg;
7564 u64 aligned_cluster;
7568 cluster_bg = btrfs_lock_cluster(bg, last_ptr, ffe_ctl->delalloc);
7570 goto refill_cluster;
7571 if (cluster_bg != bg && (cluster_bg->ro ||
7572 !block_group_bits(cluster_bg, ffe_ctl->flags)))
7573 goto release_cluster;
7575 offset = btrfs_alloc_from_cluster(cluster_bg, last_ptr,
7576 ffe_ctl->num_bytes, cluster_bg->key.objectid,
7577 &ffe_ctl->max_extent_size);
7579 /* We have a block, we're done */
7580 spin_unlock(&last_ptr->refill_lock);
7581 trace_btrfs_reserve_extent_cluster(cluster_bg,
7582 ffe_ctl->search_start, ffe_ctl->num_bytes);
7583 *cluster_bg_ret = cluster_bg;
7584 ffe_ctl->found_offset = offset;
7587 WARN_ON(last_ptr->block_group != cluster_bg);
7591 * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so
7592 * lets just skip it and let the allocator find whatever block it can
7593 * find. If we reach this point, we will have tried the cluster
7594 * allocator plenty of times and not have found anything, so we are
7595 * likely way too fragmented for the clustering stuff to find anything.
7597 * However, if the cluster is taken from the current block group,
7598 * release the cluster first, so that we stand a better chance of
7599 * succeeding in the unclustered allocation.
7601 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE && cluster_bg != bg) {
7602 spin_unlock(&last_ptr->refill_lock);
7603 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
7607 /* This cluster didn't work out, free it and start over */
7608 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7610 if (cluster_bg != bg)
7611 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
7614 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE) {
7615 spin_unlock(&last_ptr->refill_lock);
7619 aligned_cluster = max_t(u64,
7620 ffe_ctl->empty_cluster + ffe_ctl->empty_size,
7621 bg->full_stripe_len);
7622 ret = btrfs_find_space_cluster(fs_info, bg, last_ptr,
7623 ffe_ctl->search_start, ffe_ctl->num_bytes,
7626 /* Now pull our allocation out of this cluster */
7627 offset = btrfs_alloc_from_cluster(bg, last_ptr,
7628 ffe_ctl->num_bytes, ffe_ctl->search_start,
7629 &ffe_ctl->max_extent_size);
7631 /* We found one, proceed */
7632 spin_unlock(&last_ptr->refill_lock);
7633 trace_btrfs_reserve_extent_cluster(bg,
7634 ffe_ctl->search_start,
7635 ffe_ctl->num_bytes);
7636 ffe_ctl->found_offset = offset;
7639 } else if (!ffe_ctl->cached && ffe_ctl->loop > LOOP_CACHING_NOWAIT &&
7640 !ffe_ctl->retry_clustered) {
7641 spin_unlock(&last_ptr->refill_lock);
7643 ffe_ctl->retry_clustered = true;
7644 wait_block_group_cache_progress(bg, ffe_ctl->num_bytes +
7645 ffe_ctl->empty_cluster + ffe_ctl->empty_size);
7649 * At this point we either didn't find a cluster or we weren't able to
7650 * allocate a block from our cluster. Free the cluster we've been
7651 * trying to use, and go to the next block group.
7653 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7654 spin_unlock(&last_ptr->refill_lock);
7659 * Return >0 to inform caller that we find nothing
7660 * Return 0 when we found an free extent and set ffe_ctrl->found_offset
7661 * Return -EAGAIN to inform caller that we need to re-search this block group
7663 static int find_free_extent_unclustered(struct btrfs_block_group_cache *bg,
7664 struct btrfs_free_cluster *last_ptr,
7665 struct find_free_extent_ctl *ffe_ctl)
7670 * We are doing an unclustered allocation, set the fragmented flag so
7671 * we don't bother trying to setup a cluster again until we get more
7674 if (unlikely(last_ptr)) {
7675 spin_lock(&last_ptr->lock);
7676 last_ptr->fragmented = 1;
7677 spin_unlock(&last_ptr->lock);
7679 if (ffe_ctl->cached) {
7680 struct btrfs_free_space_ctl *free_space_ctl;
7682 free_space_ctl = bg->free_space_ctl;
7683 spin_lock(&free_space_ctl->tree_lock);
7684 if (free_space_ctl->free_space <
7685 ffe_ctl->num_bytes + ffe_ctl->empty_cluster +
7686 ffe_ctl->empty_size) {
7687 ffe_ctl->total_free_space = max_t(u64,
7688 ffe_ctl->total_free_space,
7689 free_space_ctl->free_space);
7690 spin_unlock(&free_space_ctl->tree_lock);
7693 spin_unlock(&free_space_ctl->tree_lock);
7696 offset = btrfs_find_space_for_alloc(bg, ffe_ctl->search_start,
7697 ffe_ctl->num_bytes, ffe_ctl->empty_size,
7698 &ffe_ctl->max_extent_size);
7701 * If we didn't find a chunk, and we haven't failed on this block group
7702 * before, and this block group is in the middle of caching and we are
7703 * ok with waiting, then go ahead and wait for progress to be made, and
7704 * set @retry_unclustered to true.
7706 * If @retry_unclustered is true then we've already waited on this
7707 * block group once and should move on to the next block group.
7709 if (!offset && !ffe_ctl->retry_unclustered && !ffe_ctl->cached &&
7710 ffe_ctl->loop > LOOP_CACHING_NOWAIT) {
7711 wait_block_group_cache_progress(bg, ffe_ctl->num_bytes +
7712 ffe_ctl->empty_size);
7713 ffe_ctl->retry_unclustered = true;
7715 } else if (!offset) {
7718 ffe_ctl->found_offset = offset;
7723 * Return >0 means caller needs to re-search for free extent
7724 * Return 0 means we have the needed free extent.
7725 * Return <0 means we failed to locate any free extent.
7727 static int find_free_extent_update_loop(struct btrfs_fs_info *fs_info,
7728 struct btrfs_free_cluster *last_ptr,
7729 struct btrfs_key *ins,
7730 struct find_free_extent_ctl *ffe_ctl,
7731 int full_search, bool use_cluster)
7733 struct btrfs_root *root = fs_info->extent_root;
7736 if ((ffe_ctl->loop == LOOP_CACHING_NOWAIT) &&
7737 ffe_ctl->have_caching_bg && !ffe_ctl->orig_have_caching_bg)
7738 ffe_ctl->orig_have_caching_bg = true;
7740 if (!ins->objectid && ffe_ctl->loop >= LOOP_CACHING_WAIT &&
7741 ffe_ctl->have_caching_bg)
7744 if (!ins->objectid && ++(ffe_ctl->index) < BTRFS_NR_RAID_TYPES)
7747 if (ins->objectid) {
7748 if (!use_cluster && last_ptr) {
7749 spin_lock(&last_ptr->lock);
7750 last_ptr->window_start = ins->objectid;
7751 spin_unlock(&last_ptr->lock);
7757 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7758 * caching kthreads as we move along
7759 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7760 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7761 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7764 if (ffe_ctl->loop < LOOP_NO_EMPTY_SIZE) {
7766 if (ffe_ctl->loop == LOOP_CACHING_NOWAIT) {
7768 * We want to skip the LOOP_CACHING_WAIT step if we
7769 * don't have any uncached bgs and we've already done a
7770 * full search through.
7772 if (ffe_ctl->orig_have_caching_bg || !full_search)
7773 ffe_ctl->loop = LOOP_CACHING_WAIT;
7775 ffe_ctl->loop = LOOP_ALLOC_CHUNK;
7780 if (ffe_ctl->loop == LOOP_ALLOC_CHUNK) {
7781 struct btrfs_trans_handle *trans;
7784 trans = current->journal_info;
7788 trans = btrfs_join_transaction(root);
7790 if (IS_ERR(trans)) {
7791 ret = PTR_ERR(trans);
7795 ret = do_chunk_alloc(trans, ffe_ctl->flags,
7799 * If we can't allocate a new chunk we've already looped
7800 * through at least once, move on to the NO_EMPTY_SIZE
7804 ffe_ctl->loop = LOOP_NO_EMPTY_SIZE;
7806 /* Do not bail out on ENOSPC since we can do more. */
7807 if (ret < 0 && ret != -ENOSPC)
7808 btrfs_abort_transaction(trans, ret);
7812 btrfs_end_transaction(trans);
7817 if (ffe_ctl->loop == LOOP_NO_EMPTY_SIZE) {
7819 * Don't loop again if we already have no empty_size and
7822 if (ffe_ctl->empty_size == 0 &&
7823 ffe_ctl->empty_cluster == 0)
7825 ffe_ctl->empty_size = 0;
7826 ffe_ctl->empty_cluster = 0;
7834 * walks the btree of allocated extents and find a hole of a given size.
7835 * The key ins is changed to record the hole:
7836 * ins->objectid == start position
7837 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7838 * ins->offset == the size of the hole.
7839 * Any available blocks before search_start are skipped.
7841 * If there is no suitable free space, we will record the max size of
7842 * the free space extent currently.
7844 * The overall logic and call chain:
7846 * find_free_extent()
7847 * |- Iterate through all block groups
7848 * | |- Get a valid block group
7849 * | |- Try to do clustered allocation in that block group
7850 * | |- Try to do unclustered allocation in that block group
7851 * | |- Check if the result is valid
7852 * | | |- If valid, then exit
7853 * | |- Jump to next block group
7855 * |- Push harder to find free extents
7856 * |- If not found, re-iterate all block groups
7858 static noinline int find_free_extent(struct btrfs_fs_info *fs_info,
7859 u64 ram_bytes, u64 num_bytes, u64 empty_size,
7860 u64 hint_byte, struct btrfs_key *ins,
7861 u64 flags, int delalloc)
7864 struct btrfs_free_cluster *last_ptr = NULL;
7865 struct btrfs_block_group_cache *block_group = NULL;
7866 struct find_free_extent_ctl ffe_ctl = {0};
7867 struct btrfs_space_info *space_info;
7868 bool use_cluster = true;
7869 bool full_search = false;
7871 WARN_ON(num_bytes < fs_info->sectorsize);
7873 ffe_ctl.ram_bytes = ram_bytes;
7874 ffe_ctl.num_bytes = num_bytes;
7875 ffe_ctl.empty_size = empty_size;
7876 ffe_ctl.flags = flags;
7877 ffe_ctl.search_start = 0;
7878 ffe_ctl.retry_clustered = false;
7879 ffe_ctl.retry_unclustered = false;
7880 ffe_ctl.delalloc = delalloc;
7881 ffe_ctl.index = btrfs_bg_flags_to_raid_index(flags);
7882 ffe_ctl.have_caching_bg = false;
7883 ffe_ctl.orig_have_caching_bg = false;
7884 ffe_ctl.found_offset = 0;
7886 ins->type = BTRFS_EXTENT_ITEM_KEY;
7890 trace_find_free_extent(fs_info, num_bytes, empty_size, flags);
7892 space_info = __find_space_info(fs_info, flags);
7894 btrfs_err(fs_info, "No space info for %llu", flags);
7899 * If our free space is heavily fragmented we may not be able to make
7900 * big contiguous allocations, so instead of doing the expensive search
7901 * for free space, simply return ENOSPC with our max_extent_size so we
7902 * can go ahead and search for a more manageable chunk.
7904 * If our max_extent_size is large enough for our allocation simply
7905 * disable clustering since we will likely not be able to find enough
7906 * space to create a cluster and induce latency trying.
7908 if (unlikely(space_info->max_extent_size)) {
7909 spin_lock(&space_info->lock);
7910 if (space_info->max_extent_size &&
7911 num_bytes > space_info->max_extent_size) {
7912 ins->offset = space_info->max_extent_size;
7913 spin_unlock(&space_info->lock);
7915 } else if (space_info->max_extent_size) {
7916 use_cluster = false;
7918 spin_unlock(&space_info->lock);
7921 last_ptr = fetch_cluster_info(fs_info, space_info,
7922 &ffe_ctl.empty_cluster);
7924 spin_lock(&last_ptr->lock);
7925 if (last_ptr->block_group)
7926 hint_byte = last_ptr->window_start;
7927 if (last_ptr->fragmented) {
7929 * We still set window_start so we can keep track of the
7930 * last place we found an allocation to try and save
7933 hint_byte = last_ptr->window_start;
7934 use_cluster = false;
7936 spin_unlock(&last_ptr->lock);
7939 ffe_ctl.search_start = max(ffe_ctl.search_start,
7940 first_logical_byte(fs_info, 0));
7941 ffe_ctl.search_start = max(ffe_ctl.search_start, hint_byte);
7942 if (ffe_ctl.search_start == hint_byte) {
7943 block_group = btrfs_lookup_block_group(fs_info,
7944 ffe_ctl.search_start);
7946 * we don't want to use the block group if it doesn't match our
7947 * allocation bits, or if its not cached.
7949 * However if we are re-searching with an ideal block group
7950 * picked out then we don't care that the block group is cached.
7952 if (block_group && block_group_bits(block_group, flags) &&
7953 block_group->cached != BTRFS_CACHE_NO) {
7954 down_read(&space_info->groups_sem);
7955 if (list_empty(&block_group->list) ||
7958 * someone is removing this block group,
7959 * we can't jump into the have_block_group
7960 * target because our list pointers are not
7963 btrfs_put_block_group(block_group);
7964 up_read(&space_info->groups_sem);
7966 ffe_ctl.index = btrfs_bg_flags_to_raid_index(
7967 block_group->flags);
7968 btrfs_lock_block_group(block_group, delalloc);
7969 goto have_block_group;
7971 } else if (block_group) {
7972 btrfs_put_block_group(block_group);
7976 ffe_ctl.have_caching_bg = false;
7977 if (ffe_ctl.index == btrfs_bg_flags_to_raid_index(flags) ||
7980 down_read(&space_info->groups_sem);
7981 list_for_each_entry(block_group,
7982 &space_info->block_groups[ffe_ctl.index], list) {
7983 /* If the block group is read-only, we can skip it entirely. */
7984 if (unlikely(block_group->ro))
7987 btrfs_grab_block_group(block_group, delalloc);
7988 ffe_ctl.search_start = block_group->key.objectid;
7991 * this can happen if we end up cycling through all the
7992 * raid types, but we want to make sure we only allocate
7993 * for the proper type.
7995 if (!block_group_bits(block_group, flags)) {
7996 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7997 BTRFS_BLOCK_GROUP_RAID1 |
7998 BTRFS_BLOCK_GROUP_RAID5 |
7999 BTRFS_BLOCK_GROUP_RAID6 |
8000 BTRFS_BLOCK_GROUP_RAID10;
8003 * if they asked for extra copies and this block group
8004 * doesn't provide them, bail. This does allow us to
8005 * fill raid0 from raid1.
8007 if ((flags & extra) && !(block_group->flags & extra))
8012 ffe_ctl.cached = block_group_cache_done(block_group);
8013 if (unlikely(!ffe_ctl.cached)) {
8014 ffe_ctl.have_caching_bg = true;
8015 ret = cache_block_group(block_group, 0);
8020 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
8024 * Ok we want to try and use the cluster allocator, so
8027 if (last_ptr && use_cluster) {
8028 struct btrfs_block_group_cache *cluster_bg = NULL;
8030 ret = find_free_extent_clustered(block_group, last_ptr,
8031 &ffe_ctl, &cluster_bg);
8034 if (cluster_bg && cluster_bg != block_group) {
8035 btrfs_release_block_group(block_group,
8037 block_group = cluster_bg;
8040 } else if (ret == -EAGAIN) {
8041 goto have_block_group;
8042 } else if (ret > 0) {
8045 /* ret == -ENOENT case falls through */
8048 ret = find_free_extent_unclustered(block_group, last_ptr,
8051 goto have_block_group;
8054 /* ret == 0 case falls through */
8056 ffe_ctl.search_start = round_up(ffe_ctl.found_offset,
8057 fs_info->stripesize);
8059 /* move on to the next group */
8060 if (ffe_ctl.search_start + num_bytes >
8061 block_group->key.objectid + block_group->key.offset) {
8062 btrfs_add_free_space(block_group, ffe_ctl.found_offset,
8067 if (ffe_ctl.found_offset < ffe_ctl.search_start)
8068 btrfs_add_free_space(block_group, ffe_ctl.found_offset,
8069 ffe_ctl.search_start - ffe_ctl.found_offset);
8071 ret = btrfs_add_reserved_bytes(block_group, ram_bytes,
8072 num_bytes, delalloc);
8073 if (ret == -EAGAIN) {
8074 btrfs_add_free_space(block_group, ffe_ctl.found_offset,
8078 btrfs_inc_block_group_reservations(block_group);
8080 /* we are all good, lets return */
8081 ins->objectid = ffe_ctl.search_start;
8082 ins->offset = num_bytes;
8084 trace_btrfs_reserve_extent(block_group, ffe_ctl.search_start,
8086 btrfs_release_block_group(block_group, delalloc);
8089 ffe_ctl.retry_clustered = false;
8090 ffe_ctl.retry_unclustered = false;
8091 BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) !=
8093 btrfs_release_block_group(block_group, delalloc);
8096 up_read(&space_info->groups_sem);
8098 ret = find_free_extent_update_loop(fs_info, last_ptr, ins, &ffe_ctl,
8099 full_search, use_cluster);
8103 if (ret == -ENOSPC) {
8105 * Use ffe_ctl->total_free_space as fallback if we can't find
8106 * any contiguous hole.
8108 if (!ffe_ctl.max_extent_size)
8109 ffe_ctl.max_extent_size = ffe_ctl.total_free_space;
8110 spin_lock(&space_info->lock);
8111 space_info->max_extent_size = ffe_ctl.max_extent_size;
8112 spin_unlock(&space_info->lock);
8113 ins->offset = ffe_ctl.max_extent_size;
8118 #define DUMP_BLOCK_RSV(fs_info, rsv_name) \
8120 struct btrfs_block_rsv *__rsv = &(fs_info)->rsv_name; \
8121 spin_lock(&__rsv->lock); \
8122 btrfs_info(fs_info, #rsv_name ": size %llu reserved %llu", \
8123 __rsv->size, __rsv->reserved); \
8124 spin_unlock(&__rsv->lock); \
8127 static void dump_space_info(struct btrfs_fs_info *fs_info,
8128 struct btrfs_space_info *info, u64 bytes,
8129 int dump_block_groups)
8131 struct btrfs_block_group_cache *cache;
8134 spin_lock(&info->lock);
8135 btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull",
8137 info->total_bytes - btrfs_space_info_used(info, true),
8138 info->full ? "" : "not ");
8140 "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu",
8141 info->total_bytes, info->bytes_used, info->bytes_pinned,
8142 info->bytes_reserved, info->bytes_may_use,
8143 info->bytes_readonly);
8144 spin_unlock(&info->lock);
8146 DUMP_BLOCK_RSV(fs_info, global_block_rsv);
8147 DUMP_BLOCK_RSV(fs_info, trans_block_rsv);
8148 DUMP_BLOCK_RSV(fs_info, chunk_block_rsv);
8149 DUMP_BLOCK_RSV(fs_info, delayed_block_rsv);
8150 DUMP_BLOCK_RSV(fs_info, delayed_refs_rsv);
8152 if (!dump_block_groups)
8155 down_read(&info->groups_sem);
8157 list_for_each_entry(cache, &info->block_groups[index], list) {
8158 spin_lock(&cache->lock);
8160 "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s",
8161 cache->key.objectid, cache->key.offset,
8162 btrfs_block_group_used(&cache->item), cache->pinned,
8163 cache->reserved, cache->ro ? "[readonly]" : "");
8164 btrfs_dump_free_space(cache, bytes);
8165 spin_unlock(&cache->lock);
8167 if (++index < BTRFS_NR_RAID_TYPES)
8169 up_read(&info->groups_sem);
8173 * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a
8174 * hole that is at least as big as @num_bytes.
8176 * @root - The root that will contain this extent
8178 * @ram_bytes - The amount of space in ram that @num_bytes take. This
8179 * is used for accounting purposes. This value differs
8180 * from @num_bytes only in the case of compressed extents.
8182 * @num_bytes - Number of bytes to allocate on-disk.
8184 * @min_alloc_size - Indicates the minimum amount of space that the
8185 * allocator should try to satisfy. In some cases
8186 * @num_bytes may be larger than what is required and if
8187 * the filesystem is fragmented then allocation fails.
8188 * However, the presence of @min_alloc_size gives a
8189 * chance to try and satisfy the smaller allocation.
8191 * @empty_size - A hint that you plan on doing more COW. This is the
8192 * size in bytes the allocator should try to find free
8193 * next to the block it returns. This is just a hint and
8194 * may be ignored by the allocator.
8196 * @hint_byte - Hint to the allocator to start searching above the byte
8197 * address passed. It might be ignored.
8199 * @ins - This key is modified to record the found hole. It will
8200 * have the following values:
8201 * ins->objectid == start position
8202 * ins->flags = BTRFS_EXTENT_ITEM_KEY
8203 * ins->offset == the size of the hole.
8205 * @is_data - Boolean flag indicating whether an extent is
8206 * allocated for data (true) or metadata (false)
8208 * @delalloc - Boolean flag indicating whether this allocation is for
8209 * delalloc or not. If 'true' data_rwsem of block groups
8210 * is going to be acquired.
8213 * Returns 0 when an allocation succeeded or < 0 when an error occurred. In
8214 * case -ENOSPC is returned then @ins->offset will contain the size of the
8215 * largest available hole the allocator managed to find.
8217 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
8218 u64 num_bytes, u64 min_alloc_size,
8219 u64 empty_size, u64 hint_byte,
8220 struct btrfs_key *ins, int is_data, int delalloc)
8222 struct btrfs_fs_info *fs_info = root->fs_info;
8223 bool final_tried = num_bytes == min_alloc_size;
8227 flags = get_alloc_profile_by_root(root, is_data);
8229 WARN_ON(num_bytes < fs_info->sectorsize);
8230 ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size,
8231 hint_byte, ins, flags, delalloc);
8232 if (!ret && !is_data) {
8233 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
8234 } else if (ret == -ENOSPC) {
8235 if (!final_tried && ins->offset) {
8236 num_bytes = min(num_bytes >> 1, ins->offset);
8237 num_bytes = round_down(num_bytes,
8238 fs_info->sectorsize);
8239 num_bytes = max(num_bytes, min_alloc_size);
8240 ram_bytes = num_bytes;
8241 if (num_bytes == min_alloc_size)
8244 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8245 struct btrfs_space_info *sinfo;
8247 sinfo = __find_space_info(fs_info, flags);
8249 "allocation failed flags %llu, wanted %llu",
8252 dump_space_info(fs_info, sinfo, num_bytes, 1);
8259 static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8261 int pin, int delalloc)
8263 struct btrfs_block_group_cache *cache;
8266 cache = btrfs_lookup_block_group(fs_info, start);
8268 btrfs_err(fs_info, "Unable to find block group for %llu",
8274 pin_down_extent(fs_info, cache, start, len, 1);
8276 if (btrfs_test_opt(fs_info, DISCARD))
8277 ret = btrfs_discard_extent(fs_info, start, len, NULL);
8278 btrfs_add_free_space(cache, start, len);
8279 btrfs_free_reserved_bytes(cache, len, delalloc);
8280 trace_btrfs_reserved_extent_free(fs_info, start, len);
8283 btrfs_put_block_group(cache);
8287 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8288 u64 start, u64 len, int delalloc)
8290 return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc);
8293 int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info,
8296 return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0);
8299 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8300 u64 parent, u64 root_objectid,
8301 u64 flags, u64 owner, u64 offset,
8302 struct btrfs_key *ins, int ref_mod)
8304 struct btrfs_fs_info *fs_info = trans->fs_info;
8306 struct btrfs_extent_item *extent_item;
8307 struct btrfs_extent_inline_ref *iref;
8308 struct btrfs_path *path;
8309 struct extent_buffer *leaf;
8314 type = BTRFS_SHARED_DATA_REF_KEY;
8316 type = BTRFS_EXTENT_DATA_REF_KEY;
8318 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
8320 path = btrfs_alloc_path();
8324 path->leave_spinning = 1;
8325 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8328 btrfs_free_path(path);
8332 leaf = path->nodes[0];
8333 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8334 struct btrfs_extent_item);
8335 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
8336 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8337 btrfs_set_extent_flags(leaf, extent_item,
8338 flags | BTRFS_EXTENT_FLAG_DATA);
8340 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8341 btrfs_set_extent_inline_ref_type(leaf, iref, type);
8343 struct btrfs_shared_data_ref *ref;
8344 ref = (struct btrfs_shared_data_ref *)(iref + 1);
8345 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8346 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
8348 struct btrfs_extent_data_ref *ref;
8349 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
8350 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
8351 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
8352 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
8353 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
8356 btrfs_mark_buffer_dirty(path->nodes[0]);
8357 btrfs_free_path(path);
8359 ret = remove_from_free_space_tree(trans, ins->objectid, ins->offset);
8363 ret = update_block_group(trans, fs_info, ins->objectid, ins->offset, 1);
8364 if (ret) { /* -ENOENT, logic error */
8365 btrfs_err(fs_info, "update block group failed for %llu %llu",
8366 ins->objectid, ins->offset);
8369 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset);
8373 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
8374 struct btrfs_delayed_ref_node *node,
8375 struct btrfs_delayed_extent_op *extent_op)
8377 struct btrfs_fs_info *fs_info = trans->fs_info;
8379 struct btrfs_extent_item *extent_item;
8380 struct btrfs_key extent_key;
8381 struct btrfs_tree_block_info *block_info;
8382 struct btrfs_extent_inline_ref *iref;
8383 struct btrfs_path *path;
8384 struct extent_buffer *leaf;
8385 struct btrfs_delayed_tree_ref *ref;
8386 u32 size = sizeof(*extent_item) + sizeof(*iref);
8388 u64 flags = extent_op->flags_to_set;
8389 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8391 ref = btrfs_delayed_node_to_tree_ref(node);
8393 extent_key.objectid = node->bytenr;
8394 if (skinny_metadata) {
8395 extent_key.offset = ref->level;
8396 extent_key.type = BTRFS_METADATA_ITEM_KEY;
8397 num_bytes = fs_info->nodesize;
8399 extent_key.offset = node->num_bytes;
8400 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
8401 size += sizeof(*block_info);
8402 num_bytes = node->num_bytes;
8405 path = btrfs_alloc_path();
8409 path->leave_spinning = 1;
8410 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8413 btrfs_free_path(path);
8417 leaf = path->nodes[0];
8418 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8419 struct btrfs_extent_item);
8420 btrfs_set_extent_refs(leaf, extent_item, 1);
8421 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8422 btrfs_set_extent_flags(leaf, extent_item,
8423 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
8425 if (skinny_metadata) {
8426 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8428 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
8429 btrfs_set_tree_block_key(leaf, block_info, &extent_op->key);
8430 btrfs_set_tree_block_level(leaf, block_info, ref->level);
8431 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
8434 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) {
8435 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
8436 btrfs_set_extent_inline_ref_type(leaf, iref,
8437 BTRFS_SHARED_BLOCK_REF_KEY);
8438 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent);
8440 btrfs_set_extent_inline_ref_type(leaf, iref,
8441 BTRFS_TREE_BLOCK_REF_KEY);
8442 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root);
8445 btrfs_mark_buffer_dirty(leaf);
8446 btrfs_free_path(path);
8448 ret = remove_from_free_space_tree(trans, extent_key.objectid,
8453 ret = update_block_group(trans, fs_info, extent_key.objectid,
8454 fs_info->nodesize, 1);
8455 if (ret) { /* -ENOENT, logic error */
8456 btrfs_err(fs_info, "update block group failed for %llu %llu",
8457 extent_key.objectid, extent_key.offset);
8461 trace_btrfs_reserved_extent_alloc(fs_info, extent_key.objectid,
8466 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8467 struct btrfs_root *root, u64 owner,
8468 u64 offset, u64 ram_bytes,
8469 struct btrfs_key *ins)
8473 BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
8475 btrfs_ref_tree_mod(root, ins->objectid, ins->offset, 0,
8476 root->root_key.objectid, owner, offset,
8477 BTRFS_ADD_DELAYED_EXTENT);
8479 ret = btrfs_add_delayed_data_ref(trans, ins->objectid,
8481 root->root_key.objectid, owner,
8483 BTRFS_ADD_DELAYED_EXTENT, NULL, NULL);
8488 * this is used by the tree logging recovery code. It records that
8489 * an extent has been allocated and makes sure to clear the free
8490 * space cache bits as well
8492 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
8493 u64 root_objectid, u64 owner, u64 offset,
8494 struct btrfs_key *ins)
8496 struct btrfs_fs_info *fs_info = trans->fs_info;
8498 struct btrfs_block_group_cache *block_group;
8499 struct btrfs_space_info *space_info;
8502 * Mixed block groups will exclude before processing the log so we only
8503 * need to do the exclude dance if this fs isn't mixed.
8505 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
8506 ret = __exclude_logged_extent(fs_info, ins->objectid,
8512 block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
8516 space_info = block_group->space_info;
8517 spin_lock(&space_info->lock);
8518 spin_lock(&block_group->lock);
8519 space_info->bytes_reserved += ins->offset;
8520 block_group->reserved += ins->offset;
8521 spin_unlock(&block_group->lock);
8522 spin_unlock(&space_info->lock);
8524 ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner,
8526 btrfs_put_block_group(block_group);
8530 static struct extent_buffer *
8531 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
8532 u64 bytenr, int level, u64 owner)
8534 struct btrfs_fs_info *fs_info = root->fs_info;
8535 struct extent_buffer *buf;
8537 buf = btrfs_find_create_tree_block(fs_info, bytenr);
8542 * Extra safety check in case the extent tree is corrupted and extent
8543 * allocator chooses to use a tree block which is already used and
8546 if (buf->lock_owner == current->pid) {
8547 btrfs_err_rl(fs_info,
8548 "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected",
8549 buf->start, btrfs_header_owner(buf), current->pid);
8550 free_extent_buffer(buf);
8551 return ERR_PTR(-EUCLEAN);
8554 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
8555 btrfs_tree_lock(buf);
8556 btrfs_clean_tree_block(buf);
8557 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
8559 btrfs_set_lock_blocking_write(buf);
8560 set_extent_buffer_uptodate(buf);
8562 memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header));
8563 btrfs_set_header_level(buf, level);
8564 btrfs_set_header_bytenr(buf, buf->start);
8565 btrfs_set_header_generation(buf, trans->transid);
8566 btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV);
8567 btrfs_set_header_owner(buf, owner);
8568 write_extent_buffer_fsid(buf, fs_info->fs_devices->metadata_uuid);
8569 write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid);
8570 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
8571 buf->log_index = root->log_transid % 2;
8573 * we allow two log transactions at a time, use different
8574 * EXTENT bit to differentiate dirty pages.
8576 if (buf->log_index == 0)
8577 set_extent_dirty(&root->dirty_log_pages, buf->start,
8578 buf->start + buf->len - 1, GFP_NOFS);
8580 set_extent_new(&root->dirty_log_pages, buf->start,
8581 buf->start + buf->len - 1);
8583 buf->log_index = -1;
8584 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
8585 buf->start + buf->len - 1, GFP_NOFS);
8587 trans->dirty = true;
8588 /* this returns a buffer locked for blocking */
8592 static struct btrfs_block_rsv *
8593 use_block_rsv(struct btrfs_trans_handle *trans,
8594 struct btrfs_root *root, u32 blocksize)
8596 struct btrfs_fs_info *fs_info = root->fs_info;
8597 struct btrfs_block_rsv *block_rsv;
8598 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
8600 bool global_updated = false;
8602 block_rsv = get_block_rsv(trans, root);
8604 if (unlikely(block_rsv->size == 0))
8607 ret = block_rsv_use_bytes(block_rsv, blocksize);
8611 if (block_rsv->failfast)
8612 return ERR_PTR(ret);
8614 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
8615 global_updated = true;
8616 update_global_block_rsv(fs_info);
8621 * The global reserve still exists to save us from ourselves, so don't
8622 * warn_on if we are short on our delayed refs reserve.
8624 if (block_rsv->type != BTRFS_BLOCK_RSV_DELREFS &&
8625 btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8626 static DEFINE_RATELIMIT_STATE(_rs,
8627 DEFAULT_RATELIMIT_INTERVAL * 10,
8628 /*DEFAULT_RATELIMIT_BURST*/ 1);
8629 if (__ratelimit(&_rs))
8631 "BTRFS: block rsv returned %d\n", ret);
8634 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
8635 BTRFS_RESERVE_NO_FLUSH);
8639 * If we couldn't reserve metadata bytes try and use some from
8640 * the global reserve if its space type is the same as the global
8643 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
8644 block_rsv->space_info == global_rsv->space_info) {
8645 ret = block_rsv_use_bytes(global_rsv, blocksize);
8649 return ERR_PTR(ret);
8652 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
8653 struct btrfs_block_rsv *block_rsv, u32 blocksize)
8655 block_rsv_add_bytes(block_rsv, blocksize, false);
8656 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0, NULL);
8660 * finds a free extent and does all the dirty work required for allocation
8661 * returns the tree buffer or an ERR_PTR on error.
8663 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
8664 struct btrfs_root *root,
8665 u64 parent, u64 root_objectid,
8666 const struct btrfs_disk_key *key,
8667 int level, u64 hint,
8670 struct btrfs_fs_info *fs_info = root->fs_info;
8671 struct btrfs_key ins;
8672 struct btrfs_block_rsv *block_rsv;
8673 struct extent_buffer *buf;
8674 struct btrfs_delayed_extent_op *extent_op;
8677 u32 blocksize = fs_info->nodesize;
8678 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8680 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8681 if (btrfs_is_testing(fs_info)) {
8682 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
8683 level, root_objectid);
8685 root->alloc_bytenr += blocksize;
8690 block_rsv = use_block_rsv(trans, root, blocksize);
8691 if (IS_ERR(block_rsv))
8692 return ERR_CAST(block_rsv);
8694 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
8695 empty_size, hint, &ins, 0, 0);
8699 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level,
8703 goto out_free_reserved;
8706 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
8708 parent = ins.objectid;
8709 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
8713 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
8714 extent_op = btrfs_alloc_delayed_extent_op();
8720 memcpy(&extent_op->key, key, sizeof(extent_op->key));
8722 memset(&extent_op->key, 0, sizeof(extent_op->key));
8723 extent_op->flags_to_set = flags;
8724 extent_op->update_key = skinny_metadata ? false : true;
8725 extent_op->update_flags = true;
8726 extent_op->is_data = false;
8727 extent_op->level = level;
8729 btrfs_ref_tree_mod(root, ins.objectid, ins.offset, parent,
8730 root_objectid, level, 0,
8731 BTRFS_ADD_DELAYED_EXTENT);
8732 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
8734 root_objectid, level,
8735 BTRFS_ADD_DELAYED_EXTENT,
8736 extent_op, NULL, NULL);
8738 goto out_free_delayed;
8743 btrfs_free_delayed_extent_op(extent_op);
8745 free_extent_buffer(buf);
8747 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
8749 unuse_block_rsv(fs_info, block_rsv, blocksize);
8750 return ERR_PTR(ret);
8753 struct walk_control {
8754 u64 refs[BTRFS_MAX_LEVEL];
8755 u64 flags[BTRFS_MAX_LEVEL];
8756 struct btrfs_key update_progress;
8757 struct btrfs_key drop_progress;
8769 #define DROP_REFERENCE 1
8770 #define UPDATE_BACKREF 2
8772 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
8773 struct btrfs_root *root,
8774 struct walk_control *wc,
8775 struct btrfs_path *path)
8777 struct btrfs_fs_info *fs_info = root->fs_info;
8783 struct btrfs_key key;
8784 struct extent_buffer *eb;
8789 if (path->slots[wc->level] < wc->reada_slot) {
8790 wc->reada_count = wc->reada_count * 2 / 3;
8791 wc->reada_count = max(wc->reada_count, 2);
8793 wc->reada_count = wc->reada_count * 3 / 2;
8794 wc->reada_count = min_t(int, wc->reada_count,
8795 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
8798 eb = path->nodes[wc->level];
8799 nritems = btrfs_header_nritems(eb);
8801 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
8802 if (nread >= wc->reada_count)
8806 bytenr = btrfs_node_blockptr(eb, slot);
8807 generation = btrfs_node_ptr_generation(eb, slot);
8809 if (slot == path->slots[wc->level])
8812 if (wc->stage == UPDATE_BACKREF &&
8813 generation <= root->root_key.offset)
8816 /* We don't lock the tree block, it's OK to be racy here */
8817 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
8818 wc->level - 1, 1, &refs,
8820 /* We don't care about errors in readahead. */
8825 if (wc->stage == DROP_REFERENCE) {
8829 if (wc->level == 1 &&
8830 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8832 if (!wc->update_ref ||
8833 generation <= root->root_key.offset)
8835 btrfs_node_key_to_cpu(eb, &key, slot);
8836 ret = btrfs_comp_cpu_keys(&key,
8837 &wc->update_progress);
8841 if (wc->level == 1 &&
8842 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8846 readahead_tree_block(fs_info, bytenr);
8849 wc->reada_slot = slot;
8853 * helper to process tree block while walking down the tree.
8855 * when wc->stage == UPDATE_BACKREF, this function updates
8856 * back refs for pointers in the block.
8858 * NOTE: return value 1 means we should stop walking down.
8860 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8861 struct btrfs_root *root,
8862 struct btrfs_path *path,
8863 struct walk_control *wc, int lookup_info)
8865 struct btrfs_fs_info *fs_info = root->fs_info;
8866 int level = wc->level;
8867 struct extent_buffer *eb = path->nodes[level];
8868 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8871 if (wc->stage == UPDATE_BACKREF &&
8872 btrfs_header_owner(eb) != root->root_key.objectid)
8876 * when reference count of tree block is 1, it won't increase
8877 * again. once full backref flag is set, we never clear it.
8880 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8881 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8882 BUG_ON(!path->locks[level]);
8883 ret = btrfs_lookup_extent_info(trans, fs_info,
8884 eb->start, level, 1,
8887 BUG_ON(ret == -ENOMEM);
8890 BUG_ON(wc->refs[level] == 0);
8893 if (wc->stage == DROP_REFERENCE) {
8894 if (wc->refs[level] > 1)
8897 if (path->locks[level] && !wc->keep_locks) {
8898 btrfs_tree_unlock_rw(eb, path->locks[level]);
8899 path->locks[level] = 0;
8904 /* wc->stage == UPDATE_BACKREF */
8905 if (!(wc->flags[level] & flag)) {
8906 BUG_ON(!path->locks[level]);
8907 ret = btrfs_inc_ref(trans, root, eb, 1);
8908 BUG_ON(ret); /* -ENOMEM */
8909 ret = btrfs_dec_ref(trans, root, eb, 0);
8910 BUG_ON(ret); /* -ENOMEM */
8911 ret = btrfs_set_disk_extent_flags(trans, fs_info, eb->start,
8913 btrfs_header_level(eb), 0);
8914 BUG_ON(ret); /* -ENOMEM */
8915 wc->flags[level] |= flag;
8919 * the block is shared by multiple trees, so it's not good to
8920 * keep the tree lock
8922 if (path->locks[level] && level > 0) {
8923 btrfs_tree_unlock_rw(eb, path->locks[level]);
8924 path->locks[level] = 0;
8930 * This is used to verify a ref exists for this root to deal with a bug where we
8931 * would have a drop_progress key that hadn't been updated properly.
8933 static int check_ref_exists(struct btrfs_trans_handle *trans,
8934 struct btrfs_root *root, u64 bytenr, u64 parent,
8937 struct btrfs_path *path;
8938 struct btrfs_extent_inline_ref *iref;
8941 path = btrfs_alloc_path();
8945 ret = lookup_extent_backref(trans, path, &iref, bytenr,
8946 root->fs_info->nodesize, parent,
8947 root->root_key.objectid, level, 0);
8948 btrfs_free_path(path);
8957 * helper to process tree block pointer.
8959 * when wc->stage == DROP_REFERENCE, this function checks
8960 * reference count of the block pointed to. if the block
8961 * is shared and we need update back refs for the subtree
8962 * rooted at the block, this function changes wc->stage to
8963 * UPDATE_BACKREF. if the block is shared and there is no
8964 * need to update back, this function drops the reference
8967 * NOTE: return value 1 means we should stop walking down.
8969 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8970 struct btrfs_root *root,
8971 struct btrfs_path *path,
8972 struct walk_control *wc, int *lookup_info)
8974 struct btrfs_fs_info *fs_info = root->fs_info;
8978 struct btrfs_key key;
8979 struct btrfs_key first_key;
8980 struct extent_buffer *next;
8981 int level = wc->level;
8984 bool need_account = false;
8986 generation = btrfs_node_ptr_generation(path->nodes[level],
8987 path->slots[level]);
8989 * if the lower level block was created before the snapshot
8990 * was created, we know there is no need to update back refs
8993 if (wc->stage == UPDATE_BACKREF &&
8994 generation <= root->root_key.offset) {
8999 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
9000 btrfs_node_key_to_cpu(path->nodes[level], &first_key,
9001 path->slots[level]);
9003 next = find_extent_buffer(fs_info, bytenr);
9005 next = btrfs_find_create_tree_block(fs_info, bytenr);
9007 return PTR_ERR(next);
9009 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
9013 btrfs_tree_lock(next);
9014 btrfs_set_lock_blocking_write(next);
9016 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
9017 &wc->refs[level - 1],
9018 &wc->flags[level - 1]);
9022 if (unlikely(wc->refs[level - 1] == 0)) {
9023 btrfs_err(fs_info, "Missing references.");
9029 if (wc->stage == DROP_REFERENCE) {
9030 if (wc->refs[level - 1] > 1) {
9031 need_account = true;
9033 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
9036 if (!wc->update_ref ||
9037 generation <= root->root_key.offset)
9040 btrfs_node_key_to_cpu(path->nodes[level], &key,
9041 path->slots[level]);
9042 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
9046 wc->stage = UPDATE_BACKREF;
9047 wc->shared_level = level - 1;
9051 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
9055 if (!btrfs_buffer_uptodate(next, generation, 0)) {
9056 btrfs_tree_unlock(next);
9057 free_extent_buffer(next);
9063 if (reada && level == 1)
9064 reada_walk_down(trans, root, wc, path);
9065 next = read_tree_block(fs_info, bytenr, generation, level - 1,
9068 return PTR_ERR(next);
9069 } else if (!extent_buffer_uptodate(next)) {
9070 free_extent_buffer(next);
9073 btrfs_tree_lock(next);
9074 btrfs_set_lock_blocking_write(next);
9078 ASSERT(level == btrfs_header_level(next));
9079 if (level != btrfs_header_level(next)) {
9080 btrfs_err(root->fs_info, "mismatched level");
9084 path->nodes[level] = next;
9085 path->slots[level] = 0;
9086 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9092 wc->refs[level - 1] = 0;
9093 wc->flags[level - 1] = 0;
9094 if (wc->stage == DROP_REFERENCE) {
9095 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
9096 parent = path->nodes[level]->start;
9098 ASSERT(root->root_key.objectid ==
9099 btrfs_header_owner(path->nodes[level]));
9100 if (root->root_key.objectid !=
9101 btrfs_header_owner(path->nodes[level])) {
9102 btrfs_err(root->fs_info,
9103 "mismatched block owner");
9111 * If we had a drop_progress we need to verify the refs are set
9112 * as expected. If we find our ref then we know that from here
9113 * on out everything should be correct, and we can clear the
9116 if (wc->restarted) {
9117 ret = check_ref_exists(trans, root, bytenr, parent,
9128 * Reloc tree doesn't contribute to qgroup numbers, and we have
9129 * already accounted them at merge time (replace_path),
9130 * thus we could skip expensive subtree trace here.
9132 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
9134 ret = btrfs_qgroup_trace_subtree(trans, next,
9135 generation, level - 1);
9137 btrfs_err_rl(fs_info,
9138 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
9144 * We need to update the next key in our walk control so we can
9145 * update the drop_progress key accordingly. We don't care if
9146 * find_next_key doesn't find a key because that means we're at
9147 * the end and are going to clean up now.
9149 wc->drop_level = level;
9150 find_next_key(path, level, &wc->drop_progress);
9152 ret = btrfs_free_extent(trans, root, bytenr, fs_info->nodesize,
9153 parent, root->root_key.objectid,
9163 btrfs_tree_unlock(next);
9164 free_extent_buffer(next);
9170 * helper to process tree block while walking up the tree.
9172 * when wc->stage == DROP_REFERENCE, this function drops
9173 * reference count on the block.
9175 * when wc->stage == UPDATE_BACKREF, this function changes
9176 * wc->stage back to DROP_REFERENCE if we changed wc->stage
9177 * to UPDATE_BACKREF previously while processing the block.
9179 * NOTE: return value 1 means we should stop walking up.
9181 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
9182 struct btrfs_root *root,
9183 struct btrfs_path *path,
9184 struct walk_control *wc)
9186 struct btrfs_fs_info *fs_info = root->fs_info;
9188 int level = wc->level;
9189 struct extent_buffer *eb = path->nodes[level];
9192 if (wc->stage == UPDATE_BACKREF) {
9193 BUG_ON(wc->shared_level < level);
9194 if (level < wc->shared_level)
9197 ret = find_next_key(path, level + 1, &wc->update_progress);
9201 wc->stage = DROP_REFERENCE;
9202 wc->shared_level = -1;
9203 path->slots[level] = 0;
9206 * check reference count again if the block isn't locked.
9207 * we should start walking down the tree again if reference
9210 if (!path->locks[level]) {
9212 btrfs_tree_lock(eb);
9213 btrfs_set_lock_blocking_write(eb);
9214 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9216 ret = btrfs_lookup_extent_info(trans, fs_info,
9217 eb->start, level, 1,
9221 btrfs_tree_unlock_rw(eb, path->locks[level]);
9222 path->locks[level] = 0;
9225 BUG_ON(wc->refs[level] == 0);
9226 if (wc->refs[level] == 1) {
9227 btrfs_tree_unlock_rw(eb, path->locks[level]);
9228 path->locks[level] = 0;
9234 /* wc->stage == DROP_REFERENCE */
9235 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
9237 if (wc->refs[level] == 1) {
9239 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9240 ret = btrfs_dec_ref(trans, root, eb, 1);
9242 ret = btrfs_dec_ref(trans, root, eb, 0);
9243 BUG_ON(ret); /* -ENOMEM */
9244 ret = btrfs_qgroup_trace_leaf_items(trans, eb);
9246 btrfs_err_rl(fs_info,
9247 "error %d accounting leaf items. Quota is out of sync, rescan required.",
9251 /* make block locked assertion in btrfs_clean_tree_block happy */
9252 if (!path->locks[level] &&
9253 btrfs_header_generation(eb) == trans->transid) {
9254 btrfs_tree_lock(eb);
9255 btrfs_set_lock_blocking_write(eb);
9256 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9258 btrfs_clean_tree_block(eb);
9261 if (eb == root->node) {
9262 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9264 else if (root->root_key.objectid != btrfs_header_owner(eb))
9265 goto owner_mismatch;
9267 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9268 parent = path->nodes[level + 1]->start;
9269 else if (root->root_key.objectid !=
9270 btrfs_header_owner(path->nodes[level + 1]))
9271 goto owner_mismatch;
9274 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
9276 wc->refs[level] = 0;
9277 wc->flags[level] = 0;
9281 btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu",
9282 btrfs_header_owner(eb), root->root_key.objectid);
9286 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
9287 struct btrfs_root *root,
9288 struct btrfs_path *path,
9289 struct walk_control *wc)
9291 int level = wc->level;
9292 int lookup_info = 1;
9295 while (level >= 0) {
9296 ret = walk_down_proc(trans, root, path, wc, lookup_info);
9303 if (path->slots[level] >=
9304 btrfs_header_nritems(path->nodes[level]))
9307 ret = do_walk_down(trans, root, path, wc, &lookup_info);
9309 path->slots[level]++;
9318 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
9319 struct btrfs_root *root,
9320 struct btrfs_path *path,
9321 struct walk_control *wc, int max_level)
9323 int level = wc->level;
9326 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
9327 while (level < max_level && path->nodes[level]) {
9329 if (path->slots[level] + 1 <
9330 btrfs_header_nritems(path->nodes[level])) {
9331 path->slots[level]++;
9334 ret = walk_up_proc(trans, root, path, wc);
9340 if (path->locks[level]) {
9341 btrfs_tree_unlock_rw(path->nodes[level],
9342 path->locks[level]);
9343 path->locks[level] = 0;
9345 free_extent_buffer(path->nodes[level]);
9346 path->nodes[level] = NULL;
9354 * drop a subvolume tree.
9356 * this function traverses the tree freeing any blocks that only
9357 * referenced by the tree.
9359 * when a shared tree block is found. this function decreases its
9360 * reference count by one. if update_ref is true, this function
9361 * also make sure backrefs for the shared block and all lower level
9362 * blocks are properly updated.
9364 * If called with for_reloc == 0, may exit early with -EAGAIN
9366 int btrfs_drop_snapshot(struct btrfs_root *root,
9367 struct btrfs_block_rsv *block_rsv, int update_ref,
9370 struct btrfs_fs_info *fs_info = root->fs_info;
9371 struct btrfs_path *path;
9372 struct btrfs_trans_handle *trans;
9373 struct btrfs_root *tree_root = fs_info->tree_root;
9374 struct btrfs_root_item *root_item = &root->root_item;
9375 struct walk_control *wc;
9376 struct btrfs_key key;
9380 bool root_dropped = false;
9382 btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid);
9384 path = btrfs_alloc_path();
9390 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9392 btrfs_free_path(path);
9397 trans = btrfs_start_transaction(tree_root, 0);
9398 if (IS_ERR(trans)) {
9399 err = PTR_ERR(trans);
9403 err = btrfs_run_delayed_items(trans);
9408 trans->block_rsv = block_rsv;
9411 * This will help us catch people modifying the fs tree while we're
9412 * dropping it. It is unsafe to mess with the fs tree while it's being
9413 * dropped as we unlock the root node and parent nodes as we walk down
9414 * the tree, assuming nothing will change. If something does change
9415 * then we'll have stale information and drop references to blocks we've
9418 set_bit(BTRFS_ROOT_DELETING, &root->state);
9419 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
9420 level = btrfs_header_level(root->node);
9421 path->nodes[level] = btrfs_lock_root_node(root);
9422 btrfs_set_lock_blocking_write(path->nodes[level]);
9423 path->slots[level] = 0;
9424 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9425 memset(&wc->update_progress, 0,
9426 sizeof(wc->update_progress));
9428 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
9429 memcpy(&wc->update_progress, &key,
9430 sizeof(wc->update_progress));
9432 level = root_item->drop_level;
9434 path->lowest_level = level;
9435 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
9436 path->lowest_level = 0;
9444 * unlock our path, this is safe because only this
9445 * function is allowed to delete this snapshot
9447 btrfs_unlock_up_safe(path, 0);
9449 level = btrfs_header_level(root->node);
9451 btrfs_tree_lock(path->nodes[level]);
9452 btrfs_set_lock_blocking_write(path->nodes[level]);
9453 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9455 ret = btrfs_lookup_extent_info(trans, fs_info,
9456 path->nodes[level]->start,
9457 level, 1, &wc->refs[level],
9463 BUG_ON(wc->refs[level] == 0);
9465 if (level == root_item->drop_level)
9468 btrfs_tree_unlock(path->nodes[level]);
9469 path->locks[level] = 0;
9470 WARN_ON(wc->refs[level] != 1);
9475 wc->restarted = test_bit(BTRFS_ROOT_DEAD_TREE, &root->state);
9477 wc->shared_level = -1;
9478 wc->stage = DROP_REFERENCE;
9479 wc->update_ref = update_ref;
9481 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9485 ret = walk_down_tree(trans, root, path, wc);
9491 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
9498 BUG_ON(wc->stage != DROP_REFERENCE);
9502 if (wc->stage == DROP_REFERENCE) {
9503 wc->drop_level = wc->level;
9504 btrfs_node_key_to_cpu(path->nodes[wc->drop_level],
9506 path->slots[wc->drop_level]);
9508 btrfs_cpu_key_to_disk(&root_item->drop_progress,
9509 &wc->drop_progress);
9510 root_item->drop_level = wc->drop_level;
9512 BUG_ON(wc->level == 0);
9513 if (btrfs_should_end_transaction(trans) ||
9514 (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
9515 ret = btrfs_update_root(trans, tree_root,
9519 btrfs_abort_transaction(trans, ret);
9524 btrfs_end_transaction_throttle(trans);
9525 if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
9526 btrfs_debug(fs_info,
9527 "drop snapshot early exit");
9532 trans = btrfs_start_transaction(tree_root, 0);
9533 if (IS_ERR(trans)) {
9534 err = PTR_ERR(trans);
9538 trans->block_rsv = block_rsv;
9541 btrfs_release_path(path);
9545 ret = btrfs_del_root(trans, &root->root_key);
9547 btrfs_abort_transaction(trans, ret);
9552 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
9553 ret = btrfs_find_root(tree_root, &root->root_key, path,
9556 btrfs_abort_transaction(trans, ret);
9559 } else if (ret > 0) {
9560 /* if we fail to delete the orphan item this time
9561 * around, it'll get picked up the next time.
9563 * The most common failure here is just -ENOENT.
9565 btrfs_del_orphan_item(trans, tree_root,
9566 root->root_key.objectid);
9570 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
9571 btrfs_add_dropped_root(trans, root);
9573 free_extent_buffer(root->node);
9574 free_extent_buffer(root->commit_root);
9575 btrfs_put_fs_root(root);
9577 root_dropped = true;
9579 btrfs_end_transaction_throttle(trans);
9582 btrfs_free_path(path);
9585 * So if we need to stop dropping the snapshot for whatever reason we
9586 * need to make sure to add it back to the dead root list so that we
9587 * keep trying to do the work later. This also cleans up roots if we
9588 * don't have it in the radix (like when we recover after a power fail
9589 * or unmount) so we don't leak memory.
9591 if (!for_reloc && !root_dropped)
9592 btrfs_add_dead_root(root);
9593 if (err && err != -EAGAIN)
9594 btrfs_handle_fs_error(fs_info, err, NULL);
9599 * drop subtree rooted at tree block 'node'.
9601 * NOTE: this function will unlock and release tree block 'node'
9602 * only used by relocation code
9604 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
9605 struct btrfs_root *root,
9606 struct extent_buffer *node,
9607 struct extent_buffer *parent)
9609 struct btrfs_fs_info *fs_info = root->fs_info;
9610 struct btrfs_path *path;
9611 struct walk_control *wc;
9617 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
9619 path = btrfs_alloc_path();
9623 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9625 btrfs_free_path(path);
9629 btrfs_assert_tree_locked(parent);
9630 parent_level = btrfs_header_level(parent);
9631 extent_buffer_get(parent);
9632 path->nodes[parent_level] = parent;
9633 path->slots[parent_level] = btrfs_header_nritems(parent);
9635 btrfs_assert_tree_locked(node);
9636 level = btrfs_header_level(node);
9637 path->nodes[level] = node;
9638 path->slots[level] = 0;
9639 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9641 wc->refs[parent_level] = 1;
9642 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
9644 wc->shared_level = -1;
9645 wc->stage = DROP_REFERENCE;
9648 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9651 wret = walk_down_tree(trans, root, path, wc);
9657 wret = walk_up_tree(trans, root, path, wc, parent_level);
9665 btrfs_free_path(path);
9669 static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags)
9675 * if restripe for this chunk_type is on pick target profile and
9676 * return, otherwise do the usual balance
9678 stripped = get_restripe_target(fs_info, flags);
9680 return extended_to_chunk(stripped);
9682 num_devices = fs_info->fs_devices->rw_devices;
9684 stripped = BTRFS_BLOCK_GROUP_RAID0 |
9685 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
9686 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
9688 if (num_devices == 1) {
9689 stripped |= BTRFS_BLOCK_GROUP_DUP;
9690 stripped = flags & ~stripped;
9692 /* turn raid0 into single device chunks */
9693 if (flags & BTRFS_BLOCK_GROUP_RAID0)
9696 /* turn mirroring into duplication */
9697 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
9698 BTRFS_BLOCK_GROUP_RAID10))
9699 return stripped | BTRFS_BLOCK_GROUP_DUP;
9701 /* they already had raid on here, just return */
9702 if (flags & stripped)
9705 stripped |= BTRFS_BLOCK_GROUP_DUP;
9706 stripped = flags & ~stripped;
9708 /* switch duplicated blocks with raid1 */
9709 if (flags & BTRFS_BLOCK_GROUP_DUP)
9710 return stripped | BTRFS_BLOCK_GROUP_RAID1;
9712 /* this is drive concat, leave it alone */
9718 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
9720 struct btrfs_space_info *sinfo = cache->space_info;
9723 u64 min_allocable_bytes;
9727 * We need some metadata space and system metadata space for
9728 * allocating chunks in some corner cases until we force to set
9729 * it to be readonly.
9732 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
9734 min_allocable_bytes = SZ_1M;
9736 min_allocable_bytes = 0;
9738 spin_lock(&sinfo->lock);
9739 spin_lock(&cache->lock);
9747 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
9748 cache->bytes_super - btrfs_block_group_used(&cache->item);
9749 sinfo_used = btrfs_space_info_used(sinfo, true);
9751 if (sinfo_used + num_bytes + min_allocable_bytes <=
9752 sinfo->total_bytes) {
9753 sinfo->bytes_readonly += num_bytes;
9755 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
9759 spin_unlock(&cache->lock);
9760 spin_unlock(&sinfo->lock);
9761 if (ret == -ENOSPC && btrfs_test_opt(cache->fs_info, ENOSPC_DEBUG)) {
9762 btrfs_info(cache->fs_info,
9763 "unable to make block group %llu ro",
9764 cache->key.objectid);
9765 btrfs_info(cache->fs_info,
9766 "sinfo_used=%llu bg_num_bytes=%llu min_allocable=%llu",
9767 sinfo_used, num_bytes, min_allocable_bytes);
9768 dump_space_info(cache->fs_info, cache->space_info, 0, 0);
9773 int btrfs_inc_block_group_ro(struct btrfs_block_group_cache *cache)
9776 struct btrfs_fs_info *fs_info = cache->fs_info;
9777 struct btrfs_trans_handle *trans;
9782 trans = btrfs_join_transaction(fs_info->extent_root);
9784 return PTR_ERR(trans);
9787 * we're not allowed to set block groups readonly after the dirty
9788 * block groups cache has started writing. If it already started,
9789 * back off and let this transaction commit
9791 mutex_lock(&fs_info->ro_block_group_mutex);
9792 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9793 u64 transid = trans->transid;
9795 mutex_unlock(&fs_info->ro_block_group_mutex);
9796 btrfs_end_transaction(trans);
9798 ret = btrfs_wait_for_commit(fs_info, transid);
9805 * if we are changing raid levels, try to allocate a corresponding
9806 * block group with the new raid level.
9808 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9809 if (alloc_flags != cache->flags) {
9810 ret = do_chunk_alloc(trans, alloc_flags,
9813 * ENOSPC is allowed here, we may have enough space
9814 * already allocated at the new raid level to
9823 ret = inc_block_group_ro(cache, 0);
9826 alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags);
9827 ret = do_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE);
9830 ret = inc_block_group_ro(cache, 0);
9832 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9833 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9834 mutex_lock(&fs_info->chunk_mutex);
9835 check_system_chunk(trans, alloc_flags);
9836 mutex_unlock(&fs_info->chunk_mutex);
9838 mutex_unlock(&fs_info->ro_block_group_mutex);
9840 btrfs_end_transaction(trans);
9844 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, u64 type)
9846 u64 alloc_flags = get_alloc_profile(trans->fs_info, type);
9848 return do_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE);
9852 * helper to account the unused space of all the readonly block group in the
9853 * space_info. takes mirrors into account.
9855 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9857 struct btrfs_block_group_cache *block_group;
9861 /* It's df, we don't care if it's racy */
9862 if (list_empty(&sinfo->ro_bgs))
9865 spin_lock(&sinfo->lock);
9866 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9867 spin_lock(&block_group->lock);
9869 if (!block_group->ro) {
9870 spin_unlock(&block_group->lock);
9874 factor = btrfs_bg_type_to_factor(block_group->flags);
9875 free_bytes += (block_group->key.offset -
9876 btrfs_block_group_used(&block_group->item)) *
9879 spin_unlock(&block_group->lock);
9881 spin_unlock(&sinfo->lock);
9886 void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache)
9888 struct btrfs_space_info *sinfo = cache->space_info;
9893 spin_lock(&sinfo->lock);
9894 spin_lock(&cache->lock);
9896 num_bytes = cache->key.offset - cache->reserved -
9897 cache->pinned - cache->bytes_super -
9898 btrfs_block_group_used(&cache->item);
9899 sinfo->bytes_readonly -= num_bytes;
9900 list_del_init(&cache->ro_list);
9902 spin_unlock(&cache->lock);
9903 spin_unlock(&sinfo->lock);
9907 * Checks to see if it's even possible to relocate this block group.
9909 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9910 * ok to go ahead and try.
9912 int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr)
9914 struct btrfs_block_group_cache *block_group;
9915 struct btrfs_space_info *space_info;
9916 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
9917 struct btrfs_device *device;
9927 debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG);
9929 block_group = btrfs_lookup_block_group(fs_info, bytenr);
9931 /* odd, couldn't find the block group, leave it alone */
9935 "can't find block group for bytenr %llu",
9940 min_free = btrfs_block_group_used(&block_group->item);
9942 /* no bytes used, we're good */
9946 space_info = block_group->space_info;
9947 spin_lock(&space_info->lock);
9949 full = space_info->full;
9952 * if this is the last block group we have in this space, we can't
9953 * relocate it unless we're able to allocate a new chunk below.
9955 * Otherwise, we need to make sure we have room in the space to handle
9956 * all of the extents from this block group. If we can, we're good
9958 if ((space_info->total_bytes != block_group->key.offset) &&
9959 (btrfs_space_info_used(space_info, false) + min_free <
9960 space_info->total_bytes)) {
9961 spin_unlock(&space_info->lock);
9964 spin_unlock(&space_info->lock);
9967 * ok we don't have enough space, but maybe we have free space on our
9968 * devices to allocate new chunks for relocation, so loop through our
9969 * alloc devices and guess if we have enough space. if this block
9970 * group is going to be restriped, run checks against the target
9971 * profile instead of the current one.
9983 target = get_restripe_target(fs_info, block_group->flags);
9985 index = btrfs_bg_flags_to_raid_index(extended_to_chunk(target));
9988 * this is just a balance, so if we were marked as full
9989 * we know there is no space for a new chunk
9994 "no space to alloc new chunk for block group %llu",
9995 block_group->key.objectid);
9999 index = btrfs_bg_flags_to_raid_index(block_group->flags);
10002 if (index == BTRFS_RAID_RAID10) {
10006 } else if (index == BTRFS_RAID_RAID1) {
10008 } else if (index == BTRFS_RAID_DUP) {
10009 /* Multiply by 2 */
10011 } else if (index == BTRFS_RAID_RAID0) {
10012 dev_min = fs_devices->rw_devices;
10013 min_free = div64_u64(min_free, dev_min);
10016 mutex_lock(&fs_info->chunk_mutex);
10017 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
10021 * check to make sure we can actually find a chunk with enough
10022 * space to fit our block group in.
10024 if (device->total_bytes > device->bytes_used + min_free &&
10025 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
10026 ret = find_free_dev_extent(device, min_free,
10027 &dev_offset, NULL);
10031 if (dev_nr >= dev_min)
10037 if (debug && ret == -1)
10038 btrfs_warn(fs_info,
10039 "no space to allocate a new chunk for block group %llu",
10040 block_group->key.objectid);
10041 mutex_unlock(&fs_info->chunk_mutex);
10043 btrfs_put_block_group(block_group);
10047 static int find_first_block_group(struct btrfs_fs_info *fs_info,
10048 struct btrfs_path *path,
10049 struct btrfs_key *key)
10051 struct btrfs_root *root = fs_info->extent_root;
10053 struct btrfs_key found_key;
10054 struct extent_buffer *leaf;
10055 struct btrfs_block_group_item bg;
10059 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
10064 slot = path->slots[0];
10065 leaf = path->nodes[0];
10066 if (slot >= btrfs_header_nritems(leaf)) {
10067 ret = btrfs_next_leaf(root, path);
10074 btrfs_item_key_to_cpu(leaf, &found_key, slot);
10076 if (found_key.objectid >= key->objectid &&
10077 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
10078 struct extent_map_tree *em_tree;
10079 struct extent_map *em;
10081 em_tree = &root->fs_info->mapping_tree.map_tree;
10082 read_lock(&em_tree->lock);
10083 em = lookup_extent_mapping(em_tree, found_key.objectid,
10085 read_unlock(&em_tree->lock);
10088 "logical %llu len %llu found bg but no related chunk",
10089 found_key.objectid, found_key.offset);
10091 } else if (em->start != found_key.objectid ||
10092 em->len != found_key.offset) {
10094 "block group %llu len %llu mismatch with chunk %llu len %llu",
10095 found_key.objectid, found_key.offset,
10096 em->start, em->len);
10099 read_extent_buffer(leaf, &bg,
10100 btrfs_item_ptr_offset(leaf, slot),
10102 flags = btrfs_block_group_flags(&bg) &
10103 BTRFS_BLOCK_GROUP_TYPE_MASK;
10105 if (flags != (em->map_lookup->type &
10106 BTRFS_BLOCK_GROUP_TYPE_MASK)) {
10108 "block group %llu len %llu type flags 0x%llx mismatch with chunk type flags 0x%llx",
10109 found_key.objectid,
10110 found_key.offset, flags,
10111 (BTRFS_BLOCK_GROUP_TYPE_MASK &
10112 em->map_lookup->type));
10118 free_extent_map(em);
10127 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
10129 struct btrfs_block_group_cache *block_group;
10133 struct inode *inode;
10135 block_group = btrfs_lookup_first_block_group(info, last);
10136 while (block_group) {
10137 wait_block_group_cache_done(block_group);
10138 spin_lock(&block_group->lock);
10139 if (block_group->iref)
10141 spin_unlock(&block_group->lock);
10142 block_group = next_block_group(info, block_group);
10144 if (!block_group) {
10151 inode = block_group->inode;
10152 block_group->iref = 0;
10153 block_group->inode = NULL;
10154 spin_unlock(&block_group->lock);
10155 ASSERT(block_group->io_ctl.inode == NULL);
10157 last = block_group->key.objectid + block_group->key.offset;
10158 btrfs_put_block_group(block_group);
10163 * Must be called only after stopping all workers, since we could have block
10164 * group caching kthreads running, and therefore they could race with us if we
10165 * freed the block groups before stopping them.
10167 int btrfs_free_block_groups(struct btrfs_fs_info *info)
10169 struct btrfs_block_group_cache *block_group;
10170 struct btrfs_space_info *space_info;
10171 struct btrfs_caching_control *caching_ctl;
10174 down_write(&info->commit_root_sem);
10175 while (!list_empty(&info->caching_block_groups)) {
10176 caching_ctl = list_entry(info->caching_block_groups.next,
10177 struct btrfs_caching_control, list);
10178 list_del(&caching_ctl->list);
10179 put_caching_control(caching_ctl);
10181 up_write(&info->commit_root_sem);
10183 spin_lock(&info->unused_bgs_lock);
10184 while (!list_empty(&info->unused_bgs)) {
10185 block_group = list_first_entry(&info->unused_bgs,
10186 struct btrfs_block_group_cache,
10188 list_del_init(&block_group->bg_list);
10189 btrfs_put_block_group(block_group);
10191 spin_unlock(&info->unused_bgs_lock);
10193 spin_lock(&info->block_group_cache_lock);
10194 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
10195 block_group = rb_entry(n, struct btrfs_block_group_cache,
10197 rb_erase(&block_group->cache_node,
10198 &info->block_group_cache_tree);
10199 RB_CLEAR_NODE(&block_group->cache_node);
10200 spin_unlock(&info->block_group_cache_lock);
10202 down_write(&block_group->space_info->groups_sem);
10203 list_del(&block_group->list);
10204 up_write(&block_group->space_info->groups_sem);
10207 * We haven't cached this block group, which means we could
10208 * possibly have excluded extents on this block group.
10210 if (block_group->cached == BTRFS_CACHE_NO ||
10211 block_group->cached == BTRFS_CACHE_ERROR)
10212 free_excluded_extents(block_group);
10214 btrfs_remove_free_space_cache(block_group);
10215 ASSERT(block_group->cached != BTRFS_CACHE_STARTED);
10216 ASSERT(list_empty(&block_group->dirty_list));
10217 ASSERT(list_empty(&block_group->io_list));
10218 ASSERT(list_empty(&block_group->bg_list));
10219 ASSERT(atomic_read(&block_group->count) == 1);
10220 btrfs_put_block_group(block_group);
10222 spin_lock(&info->block_group_cache_lock);
10224 spin_unlock(&info->block_group_cache_lock);
10226 /* now that all the block groups are freed, go through and
10227 * free all the space_info structs. This is only called during
10228 * the final stages of unmount, and so we know nobody is
10229 * using them. We call synchronize_rcu() once before we start,
10230 * just to be on the safe side.
10234 release_global_block_rsv(info);
10236 while (!list_empty(&info->space_info)) {
10239 space_info = list_entry(info->space_info.next,
10240 struct btrfs_space_info,
10244 * Do not hide this behind enospc_debug, this is actually
10245 * important and indicates a real bug if this happens.
10247 if (WARN_ON(space_info->bytes_pinned > 0 ||
10248 space_info->bytes_reserved > 0 ||
10249 space_info->bytes_may_use > 0))
10250 dump_space_info(info, space_info, 0, 0);
10251 list_del(&space_info->list);
10252 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
10253 struct kobject *kobj;
10254 kobj = space_info->block_group_kobjs[i];
10255 space_info->block_group_kobjs[i] = NULL;
10261 kobject_del(&space_info->kobj);
10262 kobject_put(&space_info->kobj);
10267 /* link_block_group will queue up kobjects to add when we're reclaim-safe */
10268 void btrfs_add_raid_kobjects(struct btrfs_fs_info *fs_info)
10270 struct btrfs_space_info *space_info;
10271 struct raid_kobject *rkobj;
10276 spin_lock(&fs_info->pending_raid_kobjs_lock);
10277 list_splice_init(&fs_info->pending_raid_kobjs, &list);
10278 spin_unlock(&fs_info->pending_raid_kobjs_lock);
10280 list_for_each_entry(rkobj, &list, list) {
10281 space_info = __find_space_info(fs_info, rkobj->flags);
10282 index = btrfs_bg_flags_to_raid_index(rkobj->flags);
10284 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
10285 "%s", get_raid_name(index));
10287 kobject_put(&rkobj->kobj);
10292 btrfs_warn(fs_info,
10293 "failed to add kobject for block cache, ignoring");
10296 static void link_block_group(struct btrfs_block_group_cache *cache)
10298 struct btrfs_space_info *space_info = cache->space_info;
10299 struct btrfs_fs_info *fs_info = cache->fs_info;
10300 int index = btrfs_bg_flags_to_raid_index(cache->flags);
10301 bool first = false;
10303 down_write(&space_info->groups_sem);
10304 if (list_empty(&space_info->block_groups[index]))
10306 list_add_tail(&cache->list, &space_info->block_groups[index]);
10307 up_write(&space_info->groups_sem);
10310 struct raid_kobject *rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
10312 btrfs_warn(cache->fs_info,
10313 "couldn't alloc memory for raid level kobject");
10316 rkobj->flags = cache->flags;
10317 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
10319 spin_lock(&fs_info->pending_raid_kobjs_lock);
10320 list_add_tail(&rkobj->list, &fs_info->pending_raid_kobjs);
10321 spin_unlock(&fs_info->pending_raid_kobjs_lock);
10322 space_info->block_group_kobjs[index] = &rkobj->kobj;
10326 static struct btrfs_block_group_cache *
10327 btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info,
10328 u64 start, u64 size)
10330 struct btrfs_block_group_cache *cache;
10332 cache = kzalloc(sizeof(*cache), GFP_NOFS);
10336 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
10338 if (!cache->free_space_ctl) {
10343 cache->key.objectid = start;
10344 cache->key.offset = size;
10345 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10347 cache->fs_info = fs_info;
10348 cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start);
10349 set_free_space_tree_thresholds(cache);
10351 atomic_set(&cache->count, 1);
10352 spin_lock_init(&cache->lock);
10353 init_rwsem(&cache->data_rwsem);
10354 INIT_LIST_HEAD(&cache->list);
10355 INIT_LIST_HEAD(&cache->cluster_list);
10356 INIT_LIST_HEAD(&cache->bg_list);
10357 INIT_LIST_HEAD(&cache->ro_list);
10358 INIT_LIST_HEAD(&cache->dirty_list);
10359 INIT_LIST_HEAD(&cache->io_list);
10360 btrfs_init_free_space_ctl(cache);
10361 atomic_set(&cache->trimming, 0);
10362 mutex_init(&cache->free_space_lock);
10363 btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root);
10370 * Iterate all chunks and verify that each of them has the corresponding block
10373 static int check_chunk_block_group_mappings(struct btrfs_fs_info *fs_info)
10375 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
10376 struct extent_map *em;
10377 struct btrfs_block_group_cache *bg;
10382 read_lock(&map_tree->map_tree.lock);
10384 * lookup_extent_mapping will return the first extent map
10385 * intersecting the range, so setting @len to 1 is enough to
10386 * get the first chunk.
10388 em = lookup_extent_mapping(&map_tree->map_tree, start, 1);
10389 read_unlock(&map_tree->map_tree.lock);
10393 bg = btrfs_lookup_block_group(fs_info, em->start);
10396 "chunk start=%llu len=%llu doesn't have corresponding block group",
10397 em->start, em->len);
10399 free_extent_map(em);
10402 if (bg->key.objectid != em->start ||
10403 bg->key.offset != em->len ||
10404 (bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK) !=
10405 (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
10407 "chunk start=%llu len=%llu flags=0x%llx doesn't match block group start=%llu len=%llu flags=0x%llx",
10408 em->start, em->len,
10409 em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK,
10410 bg->key.objectid, bg->key.offset,
10411 bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK);
10413 free_extent_map(em);
10414 btrfs_put_block_group(bg);
10417 start = em->start + em->len;
10418 free_extent_map(em);
10419 btrfs_put_block_group(bg);
10424 int btrfs_read_block_groups(struct btrfs_fs_info *info)
10426 struct btrfs_path *path;
10428 struct btrfs_block_group_cache *cache;
10429 struct btrfs_space_info *space_info;
10430 struct btrfs_key key;
10431 struct btrfs_key found_key;
10432 struct extent_buffer *leaf;
10433 int need_clear = 0;
10438 feature = btrfs_super_incompat_flags(info->super_copy);
10439 mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS);
10443 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10444 path = btrfs_alloc_path();
10447 path->reada = READA_FORWARD;
10449 cache_gen = btrfs_super_cache_generation(info->super_copy);
10450 if (btrfs_test_opt(info, SPACE_CACHE) &&
10451 btrfs_super_generation(info->super_copy) != cache_gen)
10453 if (btrfs_test_opt(info, CLEAR_CACHE))
10457 ret = find_first_block_group(info, path, &key);
10463 leaf = path->nodes[0];
10464 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
10466 cache = btrfs_create_block_group_cache(info, found_key.objectid,
10475 * When we mount with old space cache, we need to
10476 * set BTRFS_DC_CLEAR and set dirty flag.
10478 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
10479 * truncate the old free space cache inode and
10481 * b) Setting 'dirty flag' makes sure that we flush
10482 * the new space cache info onto disk.
10484 if (btrfs_test_opt(info, SPACE_CACHE))
10485 cache->disk_cache_state = BTRFS_DC_CLEAR;
10488 read_extent_buffer(leaf, &cache->item,
10489 btrfs_item_ptr_offset(leaf, path->slots[0]),
10490 sizeof(cache->item));
10491 cache->flags = btrfs_block_group_flags(&cache->item);
10493 ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) &&
10494 (cache->flags & BTRFS_BLOCK_GROUP_DATA))) {
10496 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
10497 cache->key.objectid);
10502 key.objectid = found_key.objectid + found_key.offset;
10503 btrfs_release_path(path);
10506 * We need to exclude the super stripes now so that the space
10507 * info has super bytes accounted for, otherwise we'll think
10508 * we have more space than we actually do.
10510 ret = exclude_super_stripes(cache);
10513 * We may have excluded something, so call this just in
10516 free_excluded_extents(cache);
10517 btrfs_put_block_group(cache);
10522 * check for two cases, either we are full, and therefore
10523 * don't need to bother with the caching work since we won't
10524 * find any space, or we are empty, and we can just add all
10525 * the space in and be done with it. This saves us _a_lot_ of
10526 * time, particularly in the full case.
10528 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
10529 cache->last_byte_to_unpin = (u64)-1;
10530 cache->cached = BTRFS_CACHE_FINISHED;
10531 free_excluded_extents(cache);
10532 } else if (btrfs_block_group_used(&cache->item) == 0) {
10533 cache->last_byte_to_unpin = (u64)-1;
10534 cache->cached = BTRFS_CACHE_FINISHED;
10535 add_new_free_space(cache, found_key.objectid,
10536 found_key.objectid +
10538 free_excluded_extents(cache);
10541 ret = btrfs_add_block_group_cache(info, cache);
10543 btrfs_remove_free_space_cache(cache);
10544 btrfs_put_block_group(cache);
10548 trace_btrfs_add_block_group(info, cache, 0);
10549 update_space_info(info, cache->flags, found_key.offset,
10550 btrfs_block_group_used(&cache->item),
10551 cache->bytes_super, &space_info);
10553 cache->space_info = space_info;
10555 link_block_group(cache);
10557 set_avail_alloc_bits(info, cache->flags);
10558 if (btrfs_chunk_readonly(info, cache->key.objectid)) {
10559 inc_block_group_ro(cache, 1);
10560 } else if (btrfs_block_group_used(&cache->item) == 0) {
10561 ASSERT(list_empty(&cache->bg_list));
10562 btrfs_mark_bg_unused(cache);
10566 list_for_each_entry_rcu(space_info, &info->space_info, list) {
10567 if (!(get_alloc_profile(info, space_info->flags) &
10568 (BTRFS_BLOCK_GROUP_RAID10 |
10569 BTRFS_BLOCK_GROUP_RAID1 |
10570 BTRFS_BLOCK_GROUP_RAID5 |
10571 BTRFS_BLOCK_GROUP_RAID6 |
10572 BTRFS_BLOCK_GROUP_DUP)))
10575 * avoid allocating from un-mirrored block group if there are
10576 * mirrored block groups.
10578 list_for_each_entry(cache,
10579 &space_info->block_groups[BTRFS_RAID_RAID0],
10581 inc_block_group_ro(cache, 1);
10582 list_for_each_entry(cache,
10583 &space_info->block_groups[BTRFS_RAID_SINGLE],
10585 inc_block_group_ro(cache, 1);
10588 btrfs_add_raid_kobjects(info);
10589 init_global_block_rsv(info);
10590 ret = check_chunk_block_group_mappings(info);
10592 btrfs_free_path(path);
10596 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans)
10598 struct btrfs_fs_info *fs_info = trans->fs_info;
10599 struct btrfs_block_group_cache *block_group;
10600 struct btrfs_root *extent_root = fs_info->extent_root;
10601 struct btrfs_block_group_item item;
10602 struct btrfs_key key;
10605 if (!trans->can_flush_pending_bgs)
10608 while (!list_empty(&trans->new_bgs)) {
10609 block_group = list_first_entry(&trans->new_bgs,
10610 struct btrfs_block_group_cache,
10615 spin_lock(&block_group->lock);
10616 memcpy(&item, &block_group->item, sizeof(item));
10617 memcpy(&key, &block_group->key, sizeof(key));
10618 spin_unlock(&block_group->lock);
10620 ret = btrfs_insert_item(trans, extent_root, &key, &item,
10623 btrfs_abort_transaction(trans, ret);
10624 ret = btrfs_finish_chunk_alloc(trans, key.objectid, key.offset);
10626 btrfs_abort_transaction(trans, ret);
10627 add_block_group_free_space(trans, block_group);
10628 /* already aborted the transaction if it failed. */
10630 btrfs_delayed_refs_rsv_release(fs_info, 1);
10631 list_del_init(&block_group->bg_list);
10633 btrfs_trans_release_chunk_metadata(trans);
10636 int btrfs_make_block_group(struct btrfs_trans_handle *trans, u64 bytes_used,
10637 u64 type, u64 chunk_offset, u64 size)
10639 struct btrfs_fs_info *fs_info = trans->fs_info;
10640 struct btrfs_block_group_cache *cache;
10643 btrfs_set_log_full_commit(fs_info, trans);
10645 cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size);
10649 btrfs_set_block_group_used(&cache->item, bytes_used);
10650 btrfs_set_block_group_chunk_objectid(&cache->item,
10651 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
10652 btrfs_set_block_group_flags(&cache->item, type);
10654 cache->flags = type;
10655 cache->last_byte_to_unpin = (u64)-1;
10656 cache->cached = BTRFS_CACHE_FINISHED;
10657 cache->needs_free_space = 1;
10658 ret = exclude_super_stripes(cache);
10661 * We may have excluded something, so call this just in
10664 free_excluded_extents(cache);
10665 btrfs_put_block_group(cache);
10669 add_new_free_space(cache, chunk_offset, chunk_offset + size);
10671 free_excluded_extents(cache);
10673 #ifdef CONFIG_BTRFS_DEBUG
10674 if (btrfs_should_fragment_free_space(cache)) {
10675 u64 new_bytes_used = size - bytes_used;
10677 bytes_used += new_bytes_used >> 1;
10678 fragment_free_space(cache);
10682 * Ensure the corresponding space_info object is created and
10683 * assigned to our block group. We want our bg to be added to the rbtree
10684 * with its ->space_info set.
10686 cache->space_info = __find_space_info(fs_info, cache->flags);
10687 ASSERT(cache->space_info);
10689 ret = btrfs_add_block_group_cache(fs_info, cache);
10691 btrfs_remove_free_space_cache(cache);
10692 btrfs_put_block_group(cache);
10697 * Now that our block group has its ->space_info set and is inserted in
10698 * the rbtree, update the space info's counters.
10700 trace_btrfs_add_block_group(fs_info, cache, 1);
10701 update_space_info(fs_info, cache->flags, size, bytes_used,
10702 cache->bytes_super, &cache->space_info);
10703 update_global_block_rsv(fs_info);
10705 link_block_group(cache);
10707 list_add_tail(&cache->bg_list, &trans->new_bgs);
10708 trans->delayed_ref_updates++;
10709 btrfs_update_delayed_refs_rsv(trans);
10711 set_avail_alloc_bits(fs_info, type);
10715 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
10717 u64 extra_flags = chunk_to_extended(flags) &
10718 BTRFS_EXTENDED_PROFILE_MASK;
10720 write_seqlock(&fs_info->profiles_lock);
10721 if (flags & BTRFS_BLOCK_GROUP_DATA)
10722 fs_info->avail_data_alloc_bits &= ~extra_flags;
10723 if (flags & BTRFS_BLOCK_GROUP_METADATA)
10724 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
10725 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
10726 fs_info->avail_system_alloc_bits &= ~extra_flags;
10727 write_sequnlock(&fs_info->profiles_lock);
10730 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
10731 u64 group_start, struct extent_map *em)
10733 struct btrfs_fs_info *fs_info = trans->fs_info;
10734 struct btrfs_root *root = fs_info->extent_root;
10735 struct btrfs_path *path;
10736 struct btrfs_block_group_cache *block_group;
10737 struct btrfs_free_cluster *cluster;
10738 struct btrfs_root *tree_root = fs_info->tree_root;
10739 struct btrfs_key key;
10740 struct inode *inode;
10741 struct kobject *kobj = NULL;
10745 struct btrfs_caching_control *caching_ctl = NULL;
10747 bool remove_rsv = false;
10749 block_group = btrfs_lookup_block_group(fs_info, group_start);
10750 BUG_ON(!block_group);
10751 BUG_ON(!block_group->ro);
10753 trace_btrfs_remove_block_group(block_group);
10755 * Free the reserved super bytes from this block group before
10758 free_excluded_extents(block_group);
10759 btrfs_free_ref_tree_range(fs_info, block_group->key.objectid,
10760 block_group->key.offset);
10762 memcpy(&key, &block_group->key, sizeof(key));
10763 index = btrfs_bg_flags_to_raid_index(block_group->flags);
10764 factor = btrfs_bg_type_to_factor(block_group->flags);
10766 /* make sure this block group isn't part of an allocation cluster */
10767 cluster = &fs_info->data_alloc_cluster;
10768 spin_lock(&cluster->refill_lock);
10769 btrfs_return_cluster_to_free_space(block_group, cluster);
10770 spin_unlock(&cluster->refill_lock);
10773 * make sure this block group isn't part of a metadata
10774 * allocation cluster
10776 cluster = &fs_info->meta_alloc_cluster;
10777 spin_lock(&cluster->refill_lock);
10778 btrfs_return_cluster_to_free_space(block_group, cluster);
10779 spin_unlock(&cluster->refill_lock);
10781 path = btrfs_alloc_path();
10788 * get the inode first so any iput calls done for the io_list
10789 * aren't the final iput (no unlinks allowed now)
10791 inode = lookup_free_space_inode(fs_info, block_group, path);
10793 mutex_lock(&trans->transaction->cache_write_mutex);
10795 * Make sure our free space cache IO is done before removing the
10798 spin_lock(&trans->transaction->dirty_bgs_lock);
10799 if (!list_empty(&block_group->io_list)) {
10800 list_del_init(&block_group->io_list);
10802 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
10804 spin_unlock(&trans->transaction->dirty_bgs_lock);
10805 btrfs_wait_cache_io(trans, block_group, path);
10806 btrfs_put_block_group(block_group);
10807 spin_lock(&trans->transaction->dirty_bgs_lock);
10810 if (!list_empty(&block_group->dirty_list)) {
10811 list_del_init(&block_group->dirty_list);
10813 btrfs_put_block_group(block_group);
10815 spin_unlock(&trans->transaction->dirty_bgs_lock);
10816 mutex_unlock(&trans->transaction->cache_write_mutex);
10818 if (!IS_ERR(inode)) {
10819 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10821 btrfs_add_delayed_iput(inode);
10824 clear_nlink(inode);
10825 /* One for the block groups ref */
10826 spin_lock(&block_group->lock);
10827 if (block_group->iref) {
10828 block_group->iref = 0;
10829 block_group->inode = NULL;
10830 spin_unlock(&block_group->lock);
10833 spin_unlock(&block_group->lock);
10835 /* One for our lookup ref */
10836 btrfs_add_delayed_iput(inode);
10839 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
10840 key.offset = block_group->key.objectid;
10843 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
10847 btrfs_release_path(path);
10849 ret = btrfs_del_item(trans, tree_root, path);
10852 btrfs_release_path(path);
10855 spin_lock(&fs_info->block_group_cache_lock);
10856 rb_erase(&block_group->cache_node,
10857 &fs_info->block_group_cache_tree);
10858 RB_CLEAR_NODE(&block_group->cache_node);
10860 if (fs_info->first_logical_byte == block_group->key.objectid)
10861 fs_info->first_logical_byte = (u64)-1;
10862 spin_unlock(&fs_info->block_group_cache_lock);
10864 down_write(&block_group->space_info->groups_sem);
10866 * we must use list_del_init so people can check to see if they
10867 * are still on the list after taking the semaphore
10869 list_del_init(&block_group->list);
10870 if (list_empty(&block_group->space_info->block_groups[index])) {
10871 kobj = block_group->space_info->block_group_kobjs[index];
10872 block_group->space_info->block_group_kobjs[index] = NULL;
10873 clear_avail_alloc_bits(fs_info, block_group->flags);
10875 up_write(&block_group->space_info->groups_sem);
10881 if (block_group->has_caching_ctl)
10882 caching_ctl = get_caching_control(block_group);
10883 if (block_group->cached == BTRFS_CACHE_STARTED)
10884 wait_block_group_cache_done(block_group);
10885 if (block_group->has_caching_ctl) {
10886 down_write(&fs_info->commit_root_sem);
10887 if (!caching_ctl) {
10888 struct btrfs_caching_control *ctl;
10890 list_for_each_entry(ctl,
10891 &fs_info->caching_block_groups, list)
10892 if (ctl->block_group == block_group) {
10894 refcount_inc(&caching_ctl->count);
10899 list_del_init(&caching_ctl->list);
10900 up_write(&fs_info->commit_root_sem);
10902 /* Once for the caching bgs list and once for us. */
10903 put_caching_control(caching_ctl);
10904 put_caching_control(caching_ctl);
10908 spin_lock(&trans->transaction->dirty_bgs_lock);
10909 WARN_ON(!list_empty(&block_group->dirty_list));
10910 WARN_ON(!list_empty(&block_group->io_list));
10911 spin_unlock(&trans->transaction->dirty_bgs_lock);
10913 btrfs_remove_free_space_cache(block_group);
10915 spin_lock(&block_group->space_info->lock);
10916 list_del_init(&block_group->ro_list);
10918 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
10919 WARN_ON(block_group->space_info->total_bytes
10920 < block_group->key.offset);
10921 WARN_ON(block_group->space_info->bytes_readonly
10922 < block_group->key.offset);
10923 WARN_ON(block_group->space_info->disk_total
10924 < block_group->key.offset * factor);
10926 block_group->space_info->total_bytes -= block_group->key.offset;
10927 block_group->space_info->bytes_readonly -= block_group->key.offset;
10928 block_group->space_info->disk_total -= block_group->key.offset * factor;
10930 spin_unlock(&block_group->space_info->lock);
10932 memcpy(&key, &block_group->key, sizeof(key));
10934 mutex_lock(&fs_info->chunk_mutex);
10935 spin_lock(&block_group->lock);
10936 block_group->removed = 1;
10938 * At this point trimming can't start on this block group, because we
10939 * removed the block group from the tree fs_info->block_group_cache_tree
10940 * so no one can't find it anymore and even if someone already got this
10941 * block group before we removed it from the rbtree, they have already
10942 * incremented block_group->trimming - if they didn't, they won't find
10943 * any free space entries because we already removed them all when we
10944 * called btrfs_remove_free_space_cache().
10946 * And we must not remove the extent map from the fs_info->mapping_tree
10947 * to prevent the same logical address range and physical device space
10948 * ranges from being reused for a new block group. This is because our
10949 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10950 * completely transactionless, so while it is trimming a range the
10951 * currently running transaction might finish and a new one start,
10952 * allowing for new block groups to be created that can reuse the same
10953 * physical device locations unless we take this special care.
10955 * There may also be an implicit trim operation if the file system
10956 * is mounted with -odiscard. The same protections must remain
10957 * in place until the extents have been discarded completely when
10958 * the transaction commit has completed.
10960 remove_em = (atomic_read(&block_group->trimming) == 0);
10961 spin_unlock(&block_group->lock);
10964 struct extent_map_tree *em_tree;
10966 em_tree = &fs_info->mapping_tree.map_tree;
10967 write_lock(&em_tree->lock);
10968 remove_extent_mapping(em_tree, em);
10969 write_unlock(&em_tree->lock);
10970 /* once for the tree */
10971 free_extent_map(em);
10974 mutex_unlock(&fs_info->chunk_mutex);
10976 ret = remove_block_group_free_space(trans, block_group);
10980 btrfs_put_block_group(block_group);
10981 btrfs_put_block_group(block_group);
10983 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10989 ret = btrfs_del_item(trans, root, path);
10992 btrfs_delayed_refs_rsv_release(fs_info, 1);
10993 btrfs_free_path(path);
10997 struct btrfs_trans_handle *
10998 btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info,
10999 const u64 chunk_offset)
11001 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
11002 struct extent_map *em;
11003 struct map_lookup *map;
11004 unsigned int num_items;
11006 read_lock(&em_tree->lock);
11007 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
11008 read_unlock(&em_tree->lock);
11009 ASSERT(em && em->start == chunk_offset);
11012 * We need to reserve 3 + N units from the metadata space info in order
11013 * to remove a block group (done at btrfs_remove_chunk() and at
11014 * btrfs_remove_block_group()), which are used for:
11016 * 1 unit for adding the free space inode's orphan (located in the tree
11018 * 1 unit for deleting the block group item (located in the extent
11020 * 1 unit for deleting the free space item (located in tree of tree
11022 * N units for deleting N device extent items corresponding to each
11023 * stripe (located in the device tree).
11025 * In order to remove a block group we also need to reserve units in the
11026 * system space info in order to update the chunk tree (update one or
11027 * more device items and remove one chunk item), but this is done at
11028 * btrfs_remove_chunk() through a call to check_system_chunk().
11030 map = em->map_lookup;
11031 num_items = 3 + map->num_stripes;
11032 free_extent_map(em);
11034 return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root,
11039 * Process the unused_bgs list and remove any that don't have any allocated
11040 * space inside of them.
11042 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
11044 struct btrfs_block_group_cache *block_group;
11045 struct btrfs_space_info *space_info;
11046 struct btrfs_trans_handle *trans;
11049 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
11052 spin_lock(&fs_info->unused_bgs_lock);
11053 while (!list_empty(&fs_info->unused_bgs)) {
11057 block_group = list_first_entry(&fs_info->unused_bgs,
11058 struct btrfs_block_group_cache,
11060 list_del_init(&block_group->bg_list);
11062 space_info = block_group->space_info;
11064 if (ret || btrfs_mixed_space_info(space_info)) {
11065 btrfs_put_block_group(block_group);
11068 spin_unlock(&fs_info->unused_bgs_lock);
11070 mutex_lock(&fs_info->delete_unused_bgs_mutex);
11072 /* Don't want to race with allocators so take the groups_sem */
11073 down_write(&space_info->groups_sem);
11074 spin_lock(&block_group->lock);
11075 if (block_group->reserved || block_group->pinned ||
11076 btrfs_block_group_used(&block_group->item) ||
11078 list_is_singular(&block_group->list)) {
11080 * We want to bail if we made new allocations or have
11081 * outstanding allocations in this block group. We do
11082 * the ro check in case balance is currently acting on
11083 * this block group.
11085 trace_btrfs_skip_unused_block_group(block_group);
11086 spin_unlock(&block_group->lock);
11087 up_write(&space_info->groups_sem);
11090 spin_unlock(&block_group->lock);
11092 /* We don't want to force the issue, only flip if it's ok. */
11093 ret = inc_block_group_ro(block_group, 0);
11094 up_write(&space_info->groups_sem);
11101 * Want to do this before we do anything else so we can recover
11102 * properly if we fail to join the transaction.
11104 trans = btrfs_start_trans_remove_block_group(fs_info,
11105 block_group->key.objectid);
11106 if (IS_ERR(trans)) {
11107 btrfs_dec_block_group_ro(block_group);
11108 ret = PTR_ERR(trans);
11113 * We could have pending pinned extents for this block group,
11114 * just delete them, we don't care about them anymore.
11116 start = block_group->key.objectid;
11117 end = start + block_group->key.offset - 1;
11119 * Hold the unused_bg_unpin_mutex lock to avoid racing with
11120 * btrfs_finish_extent_commit(). If we are at transaction N,
11121 * another task might be running finish_extent_commit() for the
11122 * previous transaction N - 1, and have seen a range belonging
11123 * to the block group in freed_extents[] before we were able to
11124 * clear the whole block group range from freed_extents[]. This
11125 * means that task can lookup for the block group after we
11126 * unpinned it from freed_extents[] and removed it, leading to
11127 * a BUG_ON() at btrfs_unpin_extent_range().
11129 mutex_lock(&fs_info->unused_bg_unpin_mutex);
11130 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
11133 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
11134 btrfs_dec_block_group_ro(block_group);
11137 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
11140 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
11141 btrfs_dec_block_group_ro(block_group);
11144 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
11146 /* Reset pinned so btrfs_put_block_group doesn't complain */
11147 spin_lock(&space_info->lock);
11148 spin_lock(&block_group->lock);
11150 update_bytes_pinned(space_info, -block_group->pinned);
11151 space_info->bytes_readonly += block_group->pinned;
11152 percpu_counter_add_batch(&space_info->total_bytes_pinned,
11153 -block_group->pinned,
11154 BTRFS_TOTAL_BYTES_PINNED_BATCH);
11155 block_group->pinned = 0;
11157 spin_unlock(&block_group->lock);
11158 spin_unlock(&space_info->lock);
11160 /* DISCARD can flip during remount */
11161 trimming = btrfs_test_opt(fs_info, DISCARD);
11163 /* Implicit trim during transaction commit. */
11165 btrfs_get_block_group_trimming(block_group);
11168 * Btrfs_remove_chunk will abort the transaction if things go
11171 ret = btrfs_remove_chunk(trans, block_group->key.objectid);
11175 btrfs_put_block_group_trimming(block_group);
11180 * If we're not mounted with -odiscard, we can just forget
11181 * about this block group. Otherwise we'll need to wait
11182 * until transaction commit to do the actual discard.
11185 spin_lock(&fs_info->unused_bgs_lock);
11187 * A concurrent scrub might have added us to the list
11188 * fs_info->unused_bgs, so use a list_move operation
11189 * to add the block group to the deleted_bgs list.
11191 list_move(&block_group->bg_list,
11192 &trans->transaction->deleted_bgs);
11193 spin_unlock(&fs_info->unused_bgs_lock);
11194 btrfs_get_block_group(block_group);
11197 btrfs_end_transaction(trans);
11199 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
11200 btrfs_put_block_group(block_group);
11201 spin_lock(&fs_info->unused_bgs_lock);
11203 spin_unlock(&fs_info->unused_bgs_lock);
11206 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
11208 struct btrfs_super_block *disk_super;
11214 disk_super = fs_info->super_copy;
11215 if (!btrfs_super_root(disk_super))
11218 features = btrfs_super_incompat_flags(disk_super);
11219 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
11222 flags = BTRFS_BLOCK_GROUP_SYSTEM;
11223 ret = create_space_info(fs_info, flags);
11228 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
11229 ret = create_space_info(fs_info, flags);
11231 flags = BTRFS_BLOCK_GROUP_METADATA;
11232 ret = create_space_info(fs_info, flags);
11236 flags = BTRFS_BLOCK_GROUP_DATA;
11237 ret = create_space_info(fs_info, flags);
11243 int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
11244 u64 start, u64 end)
11246 return unpin_extent_range(fs_info, start, end, false);
11250 * It used to be that old block groups would be left around forever.
11251 * Iterating over them would be enough to trim unused space. Since we
11252 * now automatically remove them, we also need to iterate over unallocated
11255 * We don't want a transaction for this since the discard may take a
11256 * substantial amount of time. We don't require that a transaction be
11257 * running, but we do need to take a running transaction into account
11258 * to ensure that we're not discarding chunks that were released or
11259 * allocated in the current transaction.
11261 * Holding the chunks lock will prevent other threads from allocating
11262 * or releasing chunks, but it won't prevent a running transaction
11263 * from committing and releasing the memory that the pending chunks
11264 * list head uses. For that, we need to take a reference to the
11265 * transaction and hold the commit root sem. We only need to hold
11266 * it while performing the free space search since we have already
11267 * held back allocations.
11269 static int btrfs_trim_free_extents(struct btrfs_device *device,
11270 struct fstrim_range *range, u64 *trimmed)
11272 u64 start, len = 0, end = 0;
11275 start = max_t(u64, range->start, SZ_1M);
11278 /* Discard not supported = nothing to do. */
11279 if (!blk_queue_discard(bdev_get_queue(device->bdev)))
11282 /* Not writable = nothing to do. */
11283 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
11286 /* No free space = nothing to do. */
11287 if (device->total_bytes <= device->bytes_used)
11293 struct btrfs_fs_info *fs_info = device->fs_info;
11296 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
11300 find_first_clear_extent_bit(&device->alloc_state, start,
11302 CHUNK_TRIMMED | CHUNK_ALLOCATED);
11304 * If find_first_clear_extent_bit find a range that spans the
11305 * end of the device it will set end to -1, in this case it's up
11306 * to the caller to trim the value to the size of the device.
11308 end = min(end, device->total_bytes - 1);
11309 len = end - start + 1;
11311 /* We didn't find any extents */
11313 mutex_unlock(&fs_info->chunk_mutex);
11318 /* Keep going until we satisfy minlen or reach end of space */
11319 if (len < range->minlen) {
11320 mutex_unlock(&fs_info->chunk_mutex);
11325 /* If we are out of the passed range break */
11326 if (start > range->start + range->len - 1) {
11327 mutex_unlock(&fs_info->chunk_mutex);
11331 start = max(range->start, start);
11332 len = min(range->len, len);
11334 ret = btrfs_issue_discard(device->bdev, start, len,
11337 set_extent_bits(&device->alloc_state, start,
11340 mutex_unlock(&fs_info->chunk_mutex);
11348 /* We've trimmed enough */
11349 if (*trimmed >= range->len)
11352 if (fatal_signal_pending(current)) {
11353 ret = -ERESTARTSYS;
11364 * Trim the whole filesystem by:
11365 * 1) trimming the free space in each block group
11366 * 2) trimming the unallocated space on each device
11368 * This will also continue trimming even if a block group or device encounters
11369 * an error. The return value will be the last error, or 0 if nothing bad
11372 int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
11374 struct btrfs_block_group_cache *cache = NULL;
11375 struct btrfs_device *device;
11376 struct list_head *devices;
11382 u64 dev_failed = 0;
11387 cache = btrfs_lookup_first_block_group(fs_info, range->start);
11388 for (; cache; cache = next_block_group(fs_info, cache)) {
11389 if (cache->key.objectid >= (range->start + range->len)) {
11390 btrfs_put_block_group(cache);
11394 start = max(range->start, cache->key.objectid);
11395 end = min(range->start + range->len,
11396 cache->key.objectid + cache->key.offset);
11398 if (end - start >= range->minlen) {
11399 if (!block_group_cache_done(cache)) {
11400 ret = cache_block_group(cache, 0);
11406 ret = wait_block_group_cache_done(cache);
11413 ret = btrfs_trim_block_group(cache,
11419 trimmed += group_trimmed;
11429 btrfs_warn(fs_info,
11430 "failed to trim %llu block group(s), last error %d",
11431 bg_failed, bg_ret);
11432 mutex_lock(&fs_info->fs_devices->device_list_mutex);
11433 devices = &fs_info->fs_devices->devices;
11434 list_for_each_entry(device, devices, dev_list) {
11435 ret = btrfs_trim_free_extents(device, range, &group_trimmed);
11442 trimmed += group_trimmed;
11444 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
11447 btrfs_warn(fs_info,
11448 "failed to trim %llu device(s), last error %d",
11449 dev_failed, dev_ret);
11450 range->len = trimmed;
11457 * btrfs_{start,end}_write_no_snapshotting() are similar to
11458 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
11459 * data into the page cache through nocow before the subvolume is snapshoted,
11460 * but flush the data into disk after the snapshot creation, or to prevent
11461 * operations while snapshotting is ongoing and that cause the snapshot to be
11462 * inconsistent (writes followed by expanding truncates for example).
11464 void btrfs_end_write_no_snapshotting(struct btrfs_root *root)
11466 percpu_counter_dec(&root->subv_writers->counter);
11467 cond_wake_up(&root->subv_writers->wait);
11470 int btrfs_start_write_no_snapshotting(struct btrfs_root *root)
11472 if (atomic_read(&root->will_be_snapshotted))
11475 percpu_counter_inc(&root->subv_writers->counter);
11477 * Make sure counter is updated before we check for snapshot creation.
11480 if (atomic_read(&root->will_be_snapshotted)) {
11481 btrfs_end_write_no_snapshotting(root);
11487 void btrfs_wait_for_snapshot_creation(struct btrfs_root *root)
11492 ret = btrfs_start_write_no_snapshotting(root);
11495 wait_var_event(&root->will_be_snapshotted,
11496 !atomic_read(&root->will_be_snapshotted));
11500 void btrfs_mark_bg_unused(struct btrfs_block_group_cache *bg)
11502 struct btrfs_fs_info *fs_info = bg->fs_info;
11504 spin_lock(&fs_info->unused_bgs_lock);
11505 if (list_empty(&bg->bg_list)) {
11506 btrfs_get_block_group(bg);
11507 trace_btrfs_add_unused_block_group(bg);
11508 list_add_tail(&bg->bg_list, &fs_info->unused_bgs);
11510 spin_unlock(&fs_info->unused_bgs_lock);