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(trans, block_group, path);
3421 * We want to set the generation to 0, that way if anything goes wrong
3422 * from here on out we know not to trust this cache when we load up next
3425 BTRFS_I(inode)->generation = 0;
3426 ret = btrfs_update_inode(trans, root, inode);
3429 * So theoretically we could recover from this, simply set the
3430 * super cache generation to 0 so we know to invalidate the
3431 * cache, but then we'd have to keep track of the block groups
3432 * that fail this way so we know we _have_ to reset this cache
3433 * before the next commit or risk reading stale cache. So to
3434 * limit our exposure to horrible edge cases lets just abort the
3435 * transaction, this only happens in really bad situations
3438 btrfs_abort_transaction(trans, ret);
3443 /* We've already setup this transaction, go ahead and exit */
3444 if (block_group->cache_generation == trans->transid &&
3445 i_size_read(inode)) {
3446 dcs = BTRFS_DC_SETUP;
3450 if (i_size_read(inode) > 0) {
3451 ret = btrfs_check_trunc_cache_free_space(fs_info,
3452 &fs_info->global_block_rsv);
3456 ret = btrfs_truncate_free_space_cache(trans, NULL, inode);
3461 spin_lock(&block_group->lock);
3462 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3463 !btrfs_test_opt(fs_info, SPACE_CACHE)) {
3465 * don't bother trying to write stuff out _if_
3466 * a) we're not cached,
3467 * b) we're with nospace_cache mount option,
3468 * c) we're with v2 space_cache (FREE_SPACE_TREE).
3470 dcs = BTRFS_DC_WRITTEN;
3471 spin_unlock(&block_group->lock);
3474 spin_unlock(&block_group->lock);
3477 * We hit an ENOSPC when setting up the cache in this transaction, just
3478 * skip doing the setup, we've already cleared the cache so we're safe.
3480 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
3486 * Try to preallocate enough space based on how big the block group is.
3487 * Keep in mind this has to include any pinned space which could end up
3488 * taking up quite a bit since it's not folded into the other space
3491 num_pages = div_u64(block_group->key.offset, SZ_256M);
3496 num_pages *= PAGE_SIZE;
3498 ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages);
3502 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3503 num_pages, num_pages,
3506 * Our cache requires contiguous chunks so that we don't modify a bunch
3507 * of metadata or split extents when writing the cache out, which means
3508 * we can enospc if we are heavily fragmented in addition to just normal
3509 * out of space conditions. So if we hit this just skip setting up any
3510 * other block groups for this transaction, maybe we'll unpin enough
3511 * space the next time around.
3514 dcs = BTRFS_DC_SETUP;
3515 else if (ret == -ENOSPC)
3516 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
3521 btrfs_release_path(path);
3523 spin_lock(&block_group->lock);
3524 if (!ret && dcs == BTRFS_DC_SETUP)
3525 block_group->cache_generation = trans->transid;
3526 block_group->disk_cache_state = dcs;
3527 spin_unlock(&block_group->lock);
3529 extent_changeset_free(data_reserved);
3533 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans)
3535 struct btrfs_fs_info *fs_info = trans->fs_info;
3536 struct btrfs_block_group_cache *cache, *tmp;
3537 struct btrfs_transaction *cur_trans = trans->transaction;
3538 struct btrfs_path *path;
3540 if (list_empty(&cur_trans->dirty_bgs) ||
3541 !btrfs_test_opt(fs_info, SPACE_CACHE))
3544 path = btrfs_alloc_path();
3548 /* Could add new block groups, use _safe just in case */
3549 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3551 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3552 cache_save_setup(cache, trans, path);
3555 btrfs_free_path(path);
3560 * transaction commit does final block group cache writeback during a
3561 * critical section where nothing is allowed to change the FS. This is
3562 * required in order for the cache to actually match the block group,
3563 * but can introduce a lot of latency into the commit.
3565 * So, btrfs_start_dirty_block_groups is here to kick off block group
3566 * cache IO. There's a chance we'll have to redo some of it if the
3567 * block group changes again during the commit, but it greatly reduces
3568 * the commit latency by getting rid of the easy block groups while
3569 * we're still allowing others to join the commit.
3571 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans)
3573 struct btrfs_fs_info *fs_info = trans->fs_info;
3574 struct btrfs_block_group_cache *cache;
3575 struct btrfs_transaction *cur_trans = trans->transaction;
3578 struct btrfs_path *path = NULL;
3580 struct list_head *io = &cur_trans->io_bgs;
3581 int num_started = 0;
3584 spin_lock(&cur_trans->dirty_bgs_lock);
3585 if (list_empty(&cur_trans->dirty_bgs)) {
3586 spin_unlock(&cur_trans->dirty_bgs_lock);
3589 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3590 spin_unlock(&cur_trans->dirty_bgs_lock);
3594 * make sure all the block groups on our dirty list actually
3597 btrfs_create_pending_block_groups(trans);
3600 path = btrfs_alloc_path();
3606 * cache_write_mutex is here only to save us from balance or automatic
3607 * removal of empty block groups deleting this block group while we are
3608 * writing out the cache
3610 mutex_lock(&trans->transaction->cache_write_mutex);
3611 while (!list_empty(&dirty)) {
3612 bool drop_reserve = true;
3614 cache = list_first_entry(&dirty,
3615 struct btrfs_block_group_cache,
3618 * this can happen if something re-dirties a block
3619 * group that is already under IO. Just wait for it to
3620 * finish and then do it all again
3622 if (!list_empty(&cache->io_list)) {
3623 list_del_init(&cache->io_list);
3624 btrfs_wait_cache_io(trans, cache, path);
3625 btrfs_put_block_group(cache);
3630 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3631 * if it should update the cache_state. Don't delete
3632 * until after we wait.
3634 * Since we're not running in the commit critical section
3635 * we need the dirty_bgs_lock to protect from update_block_group
3637 spin_lock(&cur_trans->dirty_bgs_lock);
3638 list_del_init(&cache->dirty_list);
3639 spin_unlock(&cur_trans->dirty_bgs_lock);
3643 cache_save_setup(cache, trans, path);
3645 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3646 cache->io_ctl.inode = NULL;
3647 ret = btrfs_write_out_cache(trans, cache, path);
3648 if (ret == 0 && cache->io_ctl.inode) {
3653 * The cache_write_mutex is protecting the
3654 * io_list, also refer to the definition of
3655 * btrfs_transaction::io_bgs for more details
3657 list_add_tail(&cache->io_list, io);
3660 * if we failed to write the cache, the
3661 * generation will be bad and life goes on
3667 ret = write_one_cache_group(trans, path, cache);
3669 * Our block group might still be attached to the list
3670 * of new block groups in the transaction handle of some
3671 * other task (struct btrfs_trans_handle->new_bgs). This
3672 * means its block group item isn't yet in the extent
3673 * tree. If this happens ignore the error, as we will
3674 * try again later in the critical section of the
3675 * transaction commit.
3677 if (ret == -ENOENT) {
3679 spin_lock(&cur_trans->dirty_bgs_lock);
3680 if (list_empty(&cache->dirty_list)) {
3681 list_add_tail(&cache->dirty_list,
3682 &cur_trans->dirty_bgs);
3683 btrfs_get_block_group(cache);
3684 drop_reserve = false;
3686 spin_unlock(&cur_trans->dirty_bgs_lock);
3688 btrfs_abort_transaction(trans, ret);
3692 /* if it's not on the io list, we need to put the block group */
3694 btrfs_put_block_group(cache);
3696 btrfs_delayed_refs_rsv_release(fs_info, 1);
3702 * Avoid blocking other tasks for too long. It might even save
3703 * us from writing caches for block groups that are going to be
3706 mutex_unlock(&trans->transaction->cache_write_mutex);
3707 mutex_lock(&trans->transaction->cache_write_mutex);
3709 mutex_unlock(&trans->transaction->cache_write_mutex);
3712 * go through delayed refs for all the stuff we've just kicked off
3713 * and then loop back (just once)
3715 ret = btrfs_run_delayed_refs(trans, 0);
3716 if (!ret && loops == 0) {
3718 spin_lock(&cur_trans->dirty_bgs_lock);
3719 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3721 * dirty_bgs_lock protects us from concurrent block group
3722 * deletes too (not just cache_write_mutex).
3724 if (!list_empty(&dirty)) {
3725 spin_unlock(&cur_trans->dirty_bgs_lock);
3728 spin_unlock(&cur_trans->dirty_bgs_lock);
3729 } else if (ret < 0) {
3730 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
3733 btrfs_free_path(path);
3737 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans)
3739 struct btrfs_fs_info *fs_info = trans->fs_info;
3740 struct btrfs_block_group_cache *cache;
3741 struct btrfs_transaction *cur_trans = trans->transaction;
3744 struct btrfs_path *path;
3745 struct list_head *io = &cur_trans->io_bgs;
3746 int num_started = 0;
3748 path = btrfs_alloc_path();
3753 * Even though we are in the critical section of the transaction commit,
3754 * we can still have concurrent tasks adding elements to this
3755 * transaction's list of dirty block groups. These tasks correspond to
3756 * endio free space workers started when writeback finishes for a
3757 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3758 * allocate new block groups as a result of COWing nodes of the root
3759 * tree when updating the free space inode. The writeback for the space
3760 * caches is triggered by an earlier call to
3761 * btrfs_start_dirty_block_groups() and iterations of the following
3763 * Also we want to do the cache_save_setup first and then run the
3764 * delayed refs to make sure we have the best chance at doing this all
3767 spin_lock(&cur_trans->dirty_bgs_lock);
3768 while (!list_empty(&cur_trans->dirty_bgs)) {
3769 cache = list_first_entry(&cur_trans->dirty_bgs,
3770 struct btrfs_block_group_cache,
3774 * this can happen if cache_save_setup re-dirties a block
3775 * group that is already under IO. Just wait for it to
3776 * finish and then do it all again
3778 if (!list_empty(&cache->io_list)) {
3779 spin_unlock(&cur_trans->dirty_bgs_lock);
3780 list_del_init(&cache->io_list);
3781 btrfs_wait_cache_io(trans, cache, path);
3782 btrfs_put_block_group(cache);
3783 spin_lock(&cur_trans->dirty_bgs_lock);
3787 * don't remove from the dirty list until after we've waited
3790 list_del_init(&cache->dirty_list);
3791 spin_unlock(&cur_trans->dirty_bgs_lock);
3794 cache_save_setup(cache, trans, path);
3797 ret = btrfs_run_delayed_refs(trans,
3798 (unsigned long) -1);
3800 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3801 cache->io_ctl.inode = NULL;
3802 ret = btrfs_write_out_cache(trans, cache, path);
3803 if (ret == 0 && cache->io_ctl.inode) {
3806 list_add_tail(&cache->io_list, io);
3809 * if we failed to write the cache, the
3810 * generation will be bad and life goes on
3816 ret = write_one_cache_group(trans, path, cache);
3818 * One of the free space endio workers might have
3819 * created a new block group while updating a free space
3820 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3821 * and hasn't released its transaction handle yet, in
3822 * which case the new block group is still attached to
3823 * its transaction handle and its creation has not
3824 * finished yet (no block group item in the extent tree
3825 * yet, etc). If this is the case, wait for all free
3826 * space endio workers to finish and retry. This is a
3827 * a very rare case so no need for a more efficient and
3830 if (ret == -ENOENT) {
3831 wait_event(cur_trans->writer_wait,
3832 atomic_read(&cur_trans->num_writers) == 1);
3833 ret = write_one_cache_group(trans, path, cache);
3836 btrfs_abort_transaction(trans, ret);
3839 /* if its not on the io list, we need to put the block group */
3841 btrfs_put_block_group(cache);
3842 btrfs_delayed_refs_rsv_release(fs_info, 1);
3843 spin_lock(&cur_trans->dirty_bgs_lock);
3845 spin_unlock(&cur_trans->dirty_bgs_lock);
3848 * Refer to the definition of io_bgs member for details why it's safe
3849 * to use it without any locking
3851 while (!list_empty(io)) {
3852 cache = list_first_entry(io, struct btrfs_block_group_cache,
3854 list_del_init(&cache->io_list);
3855 btrfs_wait_cache_io(trans, cache, path);
3856 btrfs_put_block_group(cache);
3859 btrfs_free_path(path);
3863 int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
3865 struct btrfs_block_group_cache *block_group;
3868 block_group = btrfs_lookup_block_group(fs_info, bytenr);
3869 if (!block_group || block_group->ro)
3872 btrfs_put_block_group(block_group);
3876 bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3878 struct btrfs_block_group_cache *bg;
3881 bg = btrfs_lookup_block_group(fs_info, bytenr);
3885 spin_lock(&bg->lock);
3889 atomic_inc(&bg->nocow_writers);
3890 spin_unlock(&bg->lock);
3892 /* no put on block group, done by btrfs_dec_nocow_writers */
3894 btrfs_put_block_group(bg);
3900 void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3902 struct btrfs_block_group_cache *bg;
3904 bg = btrfs_lookup_block_group(fs_info, bytenr);
3906 if (atomic_dec_and_test(&bg->nocow_writers))
3907 wake_up_var(&bg->nocow_writers);
3909 * Once for our lookup and once for the lookup done by a previous call
3910 * to btrfs_inc_nocow_writers()
3912 btrfs_put_block_group(bg);
3913 btrfs_put_block_group(bg);
3916 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg)
3918 wait_var_event(&bg->nocow_writers, !atomic_read(&bg->nocow_writers));
3921 static const char *alloc_name(u64 flags)
3924 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3926 case BTRFS_BLOCK_GROUP_METADATA:
3928 case BTRFS_BLOCK_GROUP_DATA:
3930 case BTRFS_BLOCK_GROUP_SYSTEM:
3934 return "invalid-combination";
3938 static int create_space_info(struct btrfs_fs_info *info, u64 flags)
3941 struct btrfs_space_info *space_info;
3945 space_info = kzalloc(sizeof(*space_info), GFP_NOFS);
3949 ret = percpu_counter_init(&space_info->total_bytes_pinned, 0,
3956 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3957 INIT_LIST_HEAD(&space_info->block_groups[i]);
3958 init_rwsem(&space_info->groups_sem);
3959 spin_lock_init(&space_info->lock);
3960 space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3961 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3962 init_waitqueue_head(&space_info->wait);
3963 INIT_LIST_HEAD(&space_info->ro_bgs);
3964 INIT_LIST_HEAD(&space_info->tickets);
3965 INIT_LIST_HEAD(&space_info->priority_tickets);
3967 ret = kobject_init_and_add(&space_info->kobj, &space_info_ktype,
3968 info->space_info_kobj, "%s",
3969 alloc_name(space_info->flags));
3971 percpu_counter_destroy(&space_info->total_bytes_pinned);
3976 list_add_rcu(&space_info->list, &info->space_info);
3977 if (flags & BTRFS_BLOCK_GROUP_DATA)
3978 info->data_sinfo = space_info;
3983 static void update_space_info(struct btrfs_fs_info *info, u64 flags,
3984 u64 total_bytes, u64 bytes_used,
3986 struct btrfs_space_info **space_info)
3988 struct btrfs_space_info *found;
3991 factor = btrfs_bg_type_to_factor(flags);
3993 found = __find_space_info(info, flags);
3995 spin_lock(&found->lock);
3996 found->total_bytes += total_bytes;
3997 found->disk_total += total_bytes * factor;
3998 found->bytes_used += bytes_used;
3999 found->disk_used += bytes_used * factor;
4000 found->bytes_readonly += bytes_readonly;
4001 if (total_bytes > 0)
4003 space_info_add_new_bytes(info, found, total_bytes -
4004 bytes_used - bytes_readonly);
4005 spin_unlock(&found->lock);
4006 *space_info = found;
4009 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
4011 u64 extra_flags = chunk_to_extended(flags) &
4012 BTRFS_EXTENDED_PROFILE_MASK;
4014 write_seqlock(&fs_info->profiles_lock);
4015 if (flags & BTRFS_BLOCK_GROUP_DATA)
4016 fs_info->avail_data_alloc_bits |= extra_flags;
4017 if (flags & BTRFS_BLOCK_GROUP_METADATA)
4018 fs_info->avail_metadata_alloc_bits |= extra_flags;
4019 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4020 fs_info->avail_system_alloc_bits |= extra_flags;
4021 write_sequnlock(&fs_info->profiles_lock);
4025 * returns target flags in extended format or 0 if restripe for this
4026 * chunk_type is not in progress
4028 * should be called with balance_lock held
4030 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
4032 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4038 if (flags & BTRFS_BLOCK_GROUP_DATA &&
4039 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4040 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
4041 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
4042 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4043 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
4044 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
4045 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4046 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
4053 * @flags: available profiles in extended format (see ctree.h)
4055 * Returns reduced profile in chunk format. If profile changing is in
4056 * progress (either running or paused) picks the target profile (if it's
4057 * already available), otherwise falls back to plain reducing.
4059 static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags)
4061 u64 num_devices = fs_info->fs_devices->rw_devices;
4067 * see if restripe for this chunk_type is in progress, if so
4068 * try to reduce to the target profile
4070 spin_lock(&fs_info->balance_lock);
4071 target = get_restripe_target(fs_info, flags);
4073 /* pick target profile only if it's already available */
4074 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
4075 spin_unlock(&fs_info->balance_lock);
4076 return extended_to_chunk(target);
4079 spin_unlock(&fs_info->balance_lock);
4081 /* First, mask out the RAID levels which aren't possible */
4082 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4083 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
4084 allowed |= btrfs_raid_array[raid_type].bg_flag;
4088 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
4089 allowed = BTRFS_BLOCK_GROUP_RAID6;
4090 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
4091 allowed = BTRFS_BLOCK_GROUP_RAID5;
4092 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
4093 allowed = BTRFS_BLOCK_GROUP_RAID10;
4094 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
4095 allowed = BTRFS_BLOCK_GROUP_RAID1;
4096 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
4097 allowed = BTRFS_BLOCK_GROUP_RAID0;
4099 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
4101 return extended_to_chunk(flags | allowed);
4104 static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags)
4111 seq = read_seqbegin(&fs_info->profiles_lock);
4113 if (flags & BTRFS_BLOCK_GROUP_DATA)
4114 flags |= fs_info->avail_data_alloc_bits;
4115 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4116 flags |= fs_info->avail_system_alloc_bits;
4117 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
4118 flags |= fs_info->avail_metadata_alloc_bits;
4119 } while (read_seqretry(&fs_info->profiles_lock, seq));
4121 return btrfs_reduce_alloc_profile(fs_info, flags);
4124 static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
4126 struct btrfs_fs_info *fs_info = root->fs_info;
4131 flags = BTRFS_BLOCK_GROUP_DATA;
4132 else if (root == fs_info->chunk_root)
4133 flags = BTRFS_BLOCK_GROUP_SYSTEM;
4135 flags = BTRFS_BLOCK_GROUP_METADATA;
4137 ret = get_alloc_profile(fs_info, flags);
4141 u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info)
4143 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA);
4146 u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info)
4148 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4151 u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info)
4153 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4156 static u64 btrfs_space_info_used(struct btrfs_space_info *s_info,
4157 bool may_use_included)
4160 return s_info->bytes_used + s_info->bytes_reserved +
4161 s_info->bytes_pinned + s_info->bytes_readonly +
4162 (may_use_included ? s_info->bytes_may_use : 0);
4165 int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes)
4167 struct btrfs_root *root = inode->root;
4168 struct btrfs_fs_info *fs_info = root->fs_info;
4169 struct btrfs_space_info *data_sinfo = fs_info->data_sinfo;
4172 int need_commit = 2;
4173 int have_pinned_space;
4175 /* make sure bytes are sectorsize aligned */
4176 bytes = ALIGN(bytes, fs_info->sectorsize);
4178 if (btrfs_is_free_space_inode(inode)) {
4180 ASSERT(current->journal_info);
4184 /* make sure we have enough space to handle the data first */
4185 spin_lock(&data_sinfo->lock);
4186 used = btrfs_space_info_used(data_sinfo, true);
4188 if (used + bytes > data_sinfo->total_bytes) {
4189 struct btrfs_trans_handle *trans;
4192 * if we don't have enough free bytes in this space then we need
4193 * to alloc a new chunk.
4195 if (!data_sinfo->full) {
4198 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4199 spin_unlock(&data_sinfo->lock);
4201 alloc_target = btrfs_data_alloc_profile(fs_info);
4203 * It is ugly that we don't call nolock join
4204 * transaction for the free space inode case here.
4205 * But it is safe because we only do the data space
4206 * reservation for the free space cache in the
4207 * transaction context, the common join transaction
4208 * just increase the counter of the current transaction
4209 * handler, doesn't try to acquire the trans_lock of
4212 trans = btrfs_join_transaction(root);
4214 return PTR_ERR(trans);
4216 ret = do_chunk_alloc(trans, alloc_target,
4217 CHUNK_ALLOC_NO_FORCE);
4218 btrfs_end_transaction(trans);
4223 have_pinned_space = 1;
4232 * If we don't have enough pinned space to deal with this
4233 * allocation, and no removed chunk in current transaction,
4234 * don't bother committing the transaction.
4236 have_pinned_space = __percpu_counter_compare(
4237 &data_sinfo->total_bytes_pinned,
4238 used + bytes - data_sinfo->total_bytes,
4239 BTRFS_TOTAL_BYTES_PINNED_BATCH);
4240 spin_unlock(&data_sinfo->lock);
4242 /* commit the current transaction and try again */
4247 if (need_commit > 0) {
4248 btrfs_start_delalloc_roots(fs_info, -1);
4249 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0,
4253 trans = btrfs_join_transaction(root);
4255 return PTR_ERR(trans);
4256 if (have_pinned_space >= 0 ||
4257 test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4258 &trans->transaction->flags) ||
4260 ret = btrfs_commit_transaction(trans);
4264 * The cleaner kthread might still be doing iput
4265 * operations. Wait for it to finish so that
4266 * more space is released. We don't need to
4267 * explicitly run the delayed iputs here because
4268 * the commit_transaction would have woken up
4271 ret = btrfs_wait_on_delayed_iputs(fs_info);
4276 btrfs_end_transaction(trans);
4280 trace_btrfs_space_reservation(fs_info,
4281 "space_info:enospc",
4282 data_sinfo->flags, bytes, 1);
4285 update_bytes_may_use(data_sinfo, bytes);
4286 trace_btrfs_space_reservation(fs_info, "space_info",
4287 data_sinfo->flags, bytes, 1);
4288 spin_unlock(&data_sinfo->lock);
4293 int btrfs_check_data_free_space(struct inode *inode,
4294 struct extent_changeset **reserved, u64 start, u64 len)
4296 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4299 /* align the range */
4300 len = round_up(start + len, fs_info->sectorsize) -
4301 round_down(start, fs_info->sectorsize);
4302 start = round_down(start, fs_info->sectorsize);
4304 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len);
4308 /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */
4309 ret = btrfs_qgroup_reserve_data(inode, reserved, start, len);
4311 btrfs_free_reserved_data_space_noquota(inode, start, len);
4318 * Called if we need to clear a data reservation for this inode
4319 * Normally in a error case.
4321 * This one will *NOT* use accurate qgroup reserved space API, just for case
4322 * which we can't sleep and is sure it won't affect qgroup reserved space.
4323 * Like clear_bit_hook().
4325 void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4328 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4329 struct btrfs_space_info *data_sinfo;
4331 /* Make sure the range is aligned to sectorsize */
4332 len = round_up(start + len, fs_info->sectorsize) -
4333 round_down(start, fs_info->sectorsize);
4334 start = round_down(start, fs_info->sectorsize);
4336 data_sinfo = fs_info->data_sinfo;
4337 spin_lock(&data_sinfo->lock);
4338 update_bytes_may_use(data_sinfo, -len);
4339 trace_btrfs_space_reservation(fs_info, "space_info",
4340 data_sinfo->flags, len, 0);
4341 spin_unlock(&data_sinfo->lock);
4345 * Called if we need to clear a data reservation for this inode
4346 * Normally in a error case.
4348 * This one will handle the per-inode data rsv map for accurate reserved
4351 void btrfs_free_reserved_data_space(struct inode *inode,
4352 struct extent_changeset *reserved, u64 start, u64 len)
4354 struct btrfs_root *root = BTRFS_I(inode)->root;
4356 /* Make sure the range is aligned to sectorsize */
4357 len = round_up(start + len, root->fs_info->sectorsize) -
4358 round_down(start, root->fs_info->sectorsize);
4359 start = round_down(start, root->fs_info->sectorsize);
4361 btrfs_free_reserved_data_space_noquota(inode, start, len);
4362 btrfs_qgroup_free_data(inode, reserved, start, len);
4365 static void force_metadata_allocation(struct btrfs_fs_info *info)
4367 struct list_head *head = &info->space_info;
4368 struct btrfs_space_info *found;
4371 list_for_each_entry_rcu(found, head, list) {
4372 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4373 found->force_alloc = CHUNK_ALLOC_FORCE;
4378 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4380 return (global->size << 1);
4383 static int should_alloc_chunk(struct btrfs_fs_info *fs_info,
4384 struct btrfs_space_info *sinfo, int force)
4386 u64 bytes_used = btrfs_space_info_used(sinfo, false);
4389 if (force == CHUNK_ALLOC_FORCE)
4393 * in limited mode, we want to have some free space up to
4394 * about 1% of the FS size.
4396 if (force == CHUNK_ALLOC_LIMITED) {
4397 thresh = btrfs_super_total_bytes(fs_info->super_copy);
4398 thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1));
4400 if (sinfo->total_bytes - bytes_used < thresh)
4404 if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8))
4409 static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type)
4413 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4414 BTRFS_BLOCK_GROUP_RAID0 |
4415 BTRFS_BLOCK_GROUP_RAID5 |
4416 BTRFS_BLOCK_GROUP_RAID6))
4417 num_dev = fs_info->fs_devices->rw_devices;
4418 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4421 num_dev = 1; /* DUP or single */
4427 * If @is_allocation is true, reserve space in the system space info necessary
4428 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4431 void check_system_chunk(struct btrfs_trans_handle *trans, u64 type)
4433 struct btrfs_fs_info *fs_info = trans->fs_info;
4434 struct btrfs_space_info *info;
4441 * Needed because we can end up allocating a system chunk and for an
4442 * atomic and race free space reservation in the chunk block reserve.
4444 lockdep_assert_held(&fs_info->chunk_mutex);
4446 info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4447 spin_lock(&info->lock);
4448 left = info->total_bytes - btrfs_space_info_used(info, true);
4449 spin_unlock(&info->lock);
4451 num_devs = get_profile_num_devs(fs_info, type);
4453 /* num_devs device items to update and 1 chunk item to add or remove */
4454 thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) +
4455 btrfs_calc_trans_metadata_size(fs_info, 1);
4457 if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4458 btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu",
4459 left, thresh, type);
4460 dump_space_info(fs_info, info, 0, 0);
4463 if (left < thresh) {
4464 u64 flags = btrfs_system_alloc_profile(fs_info);
4467 * Ignore failure to create system chunk. We might end up not
4468 * needing it, as we might not need to COW all nodes/leafs from
4469 * the paths we visit in the chunk tree (they were already COWed
4470 * or created in the current transaction for example).
4472 ret = btrfs_alloc_chunk(trans, flags);
4476 ret = btrfs_block_rsv_add(fs_info->chunk_root,
4477 &fs_info->chunk_block_rsv,
4478 thresh, BTRFS_RESERVE_NO_FLUSH);
4480 trans->chunk_bytes_reserved += thresh;
4485 * If force is CHUNK_ALLOC_FORCE:
4486 * - return 1 if it successfully allocates a chunk,
4487 * - return errors including -ENOSPC otherwise.
4488 * If force is NOT CHUNK_ALLOC_FORCE:
4489 * - return 0 if it doesn't need to allocate a new chunk,
4490 * - return 1 if it successfully allocates a chunk,
4491 * - return errors including -ENOSPC otherwise.
4493 static int do_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags,
4496 struct btrfs_fs_info *fs_info = trans->fs_info;
4497 struct btrfs_space_info *space_info;
4498 bool wait_for_alloc = false;
4499 bool should_alloc = false;
4502 /* Don't re-enter if we're already allocating a chunk */
4503 if (trans->allocating_chunk)
4506 space_info = __find_space_info(fs_info, flags);
4510 spin_lock(&space_info->lock);
4511 if (force < space_info->force_alloc)
4512 force = space_info->force_alloc;
4513 should_alloc = should_alloc_chunk(fs_info, space_info, force);
4514 if (space_info->full) {
4515 /* No more free physical space */
4520 spin_unlock(&space_info->lock);
4522 } else if (!should_alloc) {
4523 spin_unlock(&space_info->lock);
4525 } else if (space_info->chunk_alloc) {
4527 * Someone is already allocating, so we need to block
4528 * until this someone is finished and then loop to
4529 * recheck if we should continue with our allocation
4532 wait_for_alloc = true;
4533 spin_unlock(&space_info->lock);
4534 mutex_lock(&fs_info->chunk_mutex);
4535 mutex_unlock(&fs_info->chunk_mutex);
4537 /* Proceed with allocation */
4538 space_info->chunk_alloc = 1;
4539 wait_for_alloc = false;
4540 spin_unlock(&space_info->lock);
4544 } while (wait_for_alloc);
4546 mutex_lock(&fs_info->chunk_mutex);
4547 trans->allocating_chunk = true;
4550 * If we have mixed data/metadata chunks we want to make sure we keep
4551 * allocating mixed chunks instead of individual chunks.
4553 if (btrfs_mixed_space_info(space_info))
4554 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4557 * if we're doing a data chunk, go ahead and make sure that
4558 * we keep a reasonable number of metadata chunks allocated in the
4561 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4562 fs_info->data_chunk_allocations++;
4563 if (!(fs_info->data_chunk_allocations %
4564 fs_info->metadata_ratio))
4565 force_metadata_allocation(fs_info);
4569 * Check if we have enough space in SYSTEM chunk because we may need
4570 * to update devices.
4572 check_system_chunk(trans, flags);
4574 ret = btrfs_alloc_chunk(trans, flags);
4575 trans->allocating_chunk = false;
4577 spin_lock(&space_info->lock);
4580 space_info->full = 1;
4585 space_info->max_extent_size = 0;
4588 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4590 space_info->chunk_alloc = 0;
4591 spin_unlock(&space_info->lock);
4592 mutex_unlock(&fs_info->chunk_mutex);
4594 * When we allocate a new chunk we reserve space in the chunk block
4595 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4596 * add new nodes/leafs to it if we end up needing to do it when
4597 * inserting the chunk item and updating device items as part of the
4598 * second phase of chunk allocation, performed by
4599 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4600 * large number of new block groups to create in our transaction
4601 * handle's new_bgs list to avoid exhausting the chunk block reserve
4602 * in extreme cases - like having a single transaction create many new
4603 * block groups when starting to write out the free space caches of all
4604 * the block groups that were made dirty during the lifetime of the
4607 if (trans->chunk_bytes_reserved >= (u64)SZ_2M)
4608 btrfs_create_pending_block_groups(trans);
4613 static int can_overcommit(struct btrfs_fs_info *fs_info,
4614 struct btrfs_space_info *space_info, u64 bytes,
4615 enum btrfs_reserve_flush_enum flush,
4618 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4625 /* Don't overcommit when in mixed mode. */
4626 if (space_info->flags & BTRFS_BLOCK_GROUP_DATA)
4630 profile = btrfs_system_alloc_profile(fs_info);
4632 profile = btrfs_metadata_alloc_profile(fs_info);
4634 used = btrfs_space_info_used(space_info, false);
4637 * We only want to allow over committing if we have lots of actual space
4638 * free, but if we don't have enough space to handle the global reserve
4639 * space then we could end up having a real enospc problem when trying
4640 * to allocate a chunk or some other such important allocation.
4642 spin_lock(&global_rsv->lock);
4643 space_size = calc_global_rsv_need_space(global_rsv);
4644 spin_unlock(&global_rsv->lock);
4645 if (used + space_size >= space_info->total_bytes)
4648 used += space_info->bytes_may_use;
4650 avail = atomic64_read(&fs_info->free_chunk_space);
4653 * If we have dup, raid1 or raid10 then only half of the free
4654 * space is actually usable. For raid56, the space info used
4655 * doesn't include the parity drive, so we don't have to
4658 factor = btrfs_bg_type_to_factor(profile);
4659 avail = div_u64(avail, factor);
4662 * If we aren't flushing all things, let us overcommit up to
4663 * 1/2th of the space. If we can flush, don't let us overcommit
4664 * too much, let it overcommit up to 1/8 of the space.
4666 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4671 if (used + bytes < space_info->total_bytes + avail)
4676 static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info,
4677 unsigned long nr_pages, int nr_items)
4679 struct super_block *sb = fs_info->sb;
4681 if (down_read_trylock(&sb->s_umount)) {
4682 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4683 up_read(&sb->s_umount);
4686 * We needn't worry the filesystem going from r/w to r/o though
4687 * we don't acquire ->s_umount mutex, because the filesystem
4688 * should guarantee the delalloc inodes list be empty after
4689 * the filesystem is readonly(all dirty pages are written to
4692 btrfs_start_delalloc_roots(fs_info, nr_items);
4693 if (!current->journal_info)
4694 btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1);
4698 static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info,
4704 bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
4705 nr = div64_u64(to_reclaim, bytes);
4711 #define EXTENT_SIZE_PER_ITEM SZ_256K
4714 * shrink metadata reservation for delalloc
4716 static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim,
4717 u64 orig, bool wait_ordered)
4719 struct btrfs_space_info *space_info;
4720 struct btrfs_trans_handle *trans;
4725 unsigned long nr_pages;
4728 /* Calc the number of the pages we need flush for space reservation */
4729 items = calc_reclaim_items_nr(fs_info, to_reclaim);
4730 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4732 trans = (struct btrfs_trans_handle *)current->journal_info;
4733 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4735 delalloc_bytes = percpu_counter_sum_positive(
4736 &fs_info->delalloc_bytes);
4737 if (delalloc_bytes == 0) {
4741 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4746 while (delalloc_bytes && loops < 3) {
4747 nr_pages = min(delalloc_bytes, to_reclaim) >> PAGE_SHIFT;
4750 * Triggers inode writeback for up to nr_pages. This will invoke
4751 * ->writepages callback and trigger delalloc filling
4752 * (btrfs_run_delalloc_range()).
4754 btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items);
4757 * We need to wait for the compressed pages to start before
4760 async_pages = atomic_read(&fs_info->async_delalloc_pages);
4765 * Calculate how many compressed pages we want to be written
4766 * before we continue. I.e if there are more async pages than we
4767 * require wait_event will wait until nr_pages are written.
4769 if (async_pages <= nr_pages)
4772 async_pages -= nr_pages;
4774 wait_event(fs_info->async_submit_wait,
4775 atomic_read(&fs_info->async_delalloc_pages) <=
4778 spin_lock(&space_info->lock);
4779 if (list_empty(&space_info->tickets) &&
4780 list_empty(&space_info->priority_tickets)) {
4781 spin_unlock(&space_info->lock);
4784 spin_unlock(&space_info->lock);
4787 if (wait_ordered && !trans) {
4788 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4790 time_left = schedule_timeout_killable(1);
4794 delalloc_bytes = percpu_counter_sum_positive(
4795 &fs_info->delalloc_bytes);
4799 struct reserve_ticket {
4803 struct list_head list;
4804 wait_queue_head_t wait;
4808 * maybe_commit_transaction - possibly commit the transaction if its ok to
4809 * @root - the root we're allocating for
4810 * @bytes - the number of bytes we want to reserve
4811 * @force - force the commit
4813 * This will check to make sure that committing the transaction will actually
4814 * get us somewhere and then commit the transaction if it does. Otherwise it
4815 * will return -ENOSPC.
4817 static int may_commit_transaction(struct btrfs_fs_info *fs_info,
4818 struct btrfs_space_info *space_info)
4820 struct reserve_ticket *ticket = NULL;
4821 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv;
4822 struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv;
4823 struct btrfs_trans_handle *trans;
4825 u64 reclaim_bytes = 0;
4827 trans = (struct btrfs_trans_handle *)current->journal_info;
4831 spin_lock(&space_info->lock);
4832 if (!list_empty(&space_info->priority_tickets))
4833 ticket = list_first_entry(&space_info->priority_tickets,
4834 struct reserve_ticket, list);
4835 else if (!list_empty(&space_info->tickets))
4836 ticket = list_first_entry(&space_info->tickets,
4837 struct reserve_ticket, list);
4838 bytes_needed = (ticket) ? ticket->bytes : 0;
4839 spin_unlock(&space_info->lock);
4844 trans = btrfs_join_transaction(fs_info->extent_root);
4846 return PTR_ERR(trans);
4849 * See if there is enough pinned space to make this reservation, or if
4850 * we have block groups that are going to be freed, allowing us to
4851 * possibly do a chunk allocation the next loop through.
4853 if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags) ||
4854 __percpu_counter_compare(&space_info->total_bytes_pinned,
4856 BTRFS_TOTAL_BYTES_PINNED_BATCH) >= 0)
4860 * See if there is some space in the delayed insertion reservation for
4863 if (space_info != delayed_rsv->space_info)
4866 spin_lock(&delayed_rsv->lock);
4867 reclaim_bytes += delayed_rsv->reserved;
4868 spin_unlock(&delayed_rsv->lock);
4870 spin_lock(&delayed_refs_rsv->lock);
4871 reclaim_bytes += delayed_refs_rsv->reserved;
4872 spin_unlock(&delayed_refs_rsv->lock);
4873 if (reclaim_bytes >= bytes_needed)
4875 bytes_needed -= reclaim_bytes;
4877 if (__percpu_counter_compare(&space_info->total_bytes_pinned,
4879 BTRFS_TOTAL_BYTES_PINNED_BATCH) < 0)
4883 return btrfs_commit_transaction(trans);
4885 btrfs_end_transaction(trans);
4890 * Try to flush some data based on policy set by @state. This is only advisory
4891 * and may fail for various reasons. The caller is supposed to examine the
4892 * state of @space_info to detect the outcome.
4894 static void flush_space(struct btrfs_fs_info *fs_info,
4895 struct btrfs_space_info *space_info, u64 num_bytes,
4898 struct btrfs_root *root = fs_info->extent_root;
4899 struct btrfs_trans_handle *trans;
4904 case FLUSH_DELAYED_ITEMS_NR:
4905 case FLUSH_DELAYED_ITEMS:
4906 if (state == FLUSH_DELAYED_ITEMS_NR)
4907 nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2;
4911 trans = btrfs_join_transaction(root);
4912 if (IS_ERR(trans)) {
4913 ret = PTR_ERR(trans);
4916 ret = btrfs_run_delayed_items_nr(trans, nr);
4917 btrfs_end_transaction(trans);
4919 case FLUSH_DELALLOC:
4920 case FLUSH_DELALLOC_WAIT:
4921 shrink_delalloc(fs_info, num_bytes * 2, num_bytes,
4922 state == FLUSH_DELALLOC_WAIT);
4924 case FLUSH_DELAYED_REFS_NR:
4925 case FLUSH_DELAYED_REFS:
4926 trans = btrfs_join_transaction(root);
4927 if (IS_ERR(trans)) {
4928 ret = PTR_ERR(trans);
4931 if (state == FLUSH_DELAYED_REFS_NR)
4932 nr = calc_reclaim_items_nr(fs_info, num_bytes);
4935 btrfs_run_delayed_refs(trans, nr);
4936 btrfs_end_transaction(trans);
4939 case ALLOC_CHUNK_FORCE:
4940 trans = btrfs_join_transaction(root);
4941 if (IS_ERR(trans)) {
4942 ret = PTR_ERR(trans);
4945 ret = do_chunk_alloc(trans,
4946 btrfs_metadata_alloc_profile(fs_info),
4947 (state == ALLOC_CHUNK) ?
4948 CHUNK_ALLOC_NO_FORCE : CHUNK_ALLOC_FORCE);
4949 btrfs_end_transaction(trans);
4950 if (ret > 0 || ret == -ENOSPC)
4955 * If we have pending delayed iputs then we could free up a
4956 * bunch of pinned space, so make sure we run the iputs before
4957 * we do our pinned bytes check below.
4959 btrfs_run_delayed_iputs(fs_info);
4960 btrfs_wait_on_delayed_iputs(fs_info);
4962 ret = may_commit_transaction(fs_info, space_info);
4969 trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, state,
4975 btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info,
4976 struct btrfs_space_info *space_info,
4979 struct reserve_ticket *ticket;
4984 list_for_each_entry(ticket, &space_info->tickets, list)
4985 to_reclaim += ticket->bytes;
4986 list_for_each_entry(ticket, &space_info->priority_tickets, list)
4987 to_reclaim += ticket->bytes;
4991 to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M);
4992 if (can_overcommit(fs_info, space_info, to_reclaim,
4993 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
4996 used = btrfs_space_info_used(space_info, true);
4998 if (can_overcommit(fs_info, space_info, SZ_1M,
4999 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
5000 expected = div_factor_fine(space_info->total_bytes, 95);
5002 expected = div_factor_fine(space_info->total_bytes, 90);
5004 if (used > expected)
5005 to_reclaim = used - expected;
5008 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
5009 space_info->bytes_reserved);
5013 static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info,
5014 struct btrfs_space_info *space_info,
5015 u64 used, bool system_chunk)
5017 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
5019 /* If we're just plain full then async reclaim just slows us down. */
5020 if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh)
5023 if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5027 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
5028 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
5031 static bool wake_all_tickets(struct list_head *head)
5033 struct reserve_ticket *ticket;
5035 while (!list_empty(head)) {
5036 ticket = list_first_entry(head, struct reserve_ticket, list);
5037 list_del_init(&ticket->list);
5038 ticket->error = -ENOSPC;
5039 wake_up(&ticket->wait);
5040 if (ticket->bytes != ticket->orig_bytes)
5047 * This is for normal flushers, we can wait all goddamned day if we want to. We
5048 * will loop and continuously try to flush as long as we are making progress.
5049 * We count progress as clearing off tickets each time we have to loop.
5051 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
5053 struct btrfs_fs_info *fs_info;
5054 struct btrfs_space_info *space_info;
5057 int commit_cycles = 0;
5058 u64 last_tickets_id;
5060 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
5061 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5063 spin_lock(&space_info->lock);
5064 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5067 space_info->flush = 0;
5068 spin_unlock(&space_info->lock);
5071 last_tickets_id = space_info->tickets_id;
5072 spin_unlock(&space_info->lock);
5074 flush_state = FLUSH_DELAYED_ITEMS_NR;
5076 flush_space(fs_info, space_info, to_reclaim, flush_state);
5077 spin_lock(&space_info->lock);
5078 if (list_empty(&space_info->tickets)) {
5079 space_info->flush = 0;
5080 spin_unlock(&space_info->lock);
5083 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info,
5086 if (last_tickets_id == space_info->tickets_id) {
5089 last_tickets_id = space_info->tickets_id;
5090 flush_state = FLUSH_DELAYED_ITEMS_NR;
5096 * We don't want to force a chunk allocation until we've tried
5097 * pretty hard to reclaim space. Think of the case where we
5098 * freed up a bunch of space and so have a lot of pinned space
5099 * to reclaim. We would rather use that than possibly create a
5100 * underutilized metadata chunk. So if this is our first run
5101 * through the flushing state machine skip ALLOC_CHUNK_FORCE and
5102 * commit the transaction. If nothing has changed the next go
5103 * around then we can force a chunk allocation.
5105 if (flush_state == ALLOC_CHUNK_FORCE && !commit_cycles)
5108 if (flush_state > COMMIT_TRANS) {
5110 if (commit_cycles > 2) {
5111 if (wake_all_tickets(&space_info->tickets)) {
5112 flush_state = FLUSH_DELAYED_ITEMS_NR;
5115 space_info->flush = 0;
5118 flush_state = FLUSH_DELAYED_ITEMS_NR;
5121 spin_unlock(&space_info->lock);
5122 } while (flush_state <= COMMIT_TRANS);
5125 void btrfs_init_async_reclaim_work(struct work_struct *work)
5127 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
5130 static const enum btrfs_flush_state priority_flush_states[] = {
5131 FLUSH_DELAYED_ITEMS_NR,
5132 FLUSH_DELAYED_ITEMS,
5136 static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
5137 struct btrfs_space_info *space_info,
5138 struct reserve_ticket *ticket)
5143 spin_lock(&space_info->lock);
5144 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5147 spin_unlock(&space_info->lock);
5150 spin_unlock(&space_info->lock);
5154 flush_space(fs_info, space_info, to_reclaim,
5155 priority_flush_states[flush_state]);
5157 spin_lock(&space_info->lock);
5158 if (ticket->bytes == 0) {
5159 spin_unlock(&space_info->lock);
5162 spin_unlock(&space_info->lock);
5163 } while (flush_state < ARRAY_SIZE(priority_flush_states));
5166 static int wait_reserve_ticket(struct btrfs_fs_info *fs_info,
5167 struct btrfs_space_info *space_info,
5168 struct reserve_ticket *ticket)
5172 u64 reclaim_bytes = 0;
5175 spin_lock(&space_info->lock);
5176 while (ticket->bytes > 0 && ticket->error == 0) {
5177 ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE);
5182 spin_unlock(&space_info->lock);
5186 finish_wait(&ticket->wait, &wait);
5187 spin_lock(&space_info->lock);
5190 ret = ticket->error;
5191 if (!list_empty(&ticket->list))
5192 list_del_init(&ticket->list);
5193 if (ticket->bytes && ticket->bytes < ticket->orig_bytes)
5194 reclaim_bytes = ticket->orig_bytes - ticket->bytes;
5195 spin_unlock(&space_info->lock);
5198 space_info_add_old_bytes(fs_info, space_info, reclaim_bytes);
5203 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5204 * @root - the root we're allocating for
5205 * @space_info - the space info we want to allocate from
5206 * @orig_bytes - the number of bytes we want
5207 * @flush - whether or not we can flush to make our reservation
5209 * This will reserve orig_bytes number of bytes from the space info associated
5210 * with the block_rsv. If there is not enough space it will make an attempt to
5211 * flush out space to make room. It will do this by flushing delalloc if
5212 * possible or committing the transaction. If flush is 0 then no attempts to
5213 * regain reservations will be made and this will fail if there is not enough
5216 static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info,
5217 struct btrfs_space_info *space_info,
5219 enum btrfs_reserve_flush_enum flush,
5222 struct reserve_ticket ticket;
5224 u64 reclaim_bytes = 0;
5228 ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL);
5230 spin_lock(&space_info->lock);
5232 used = btrfs_space_info_used(space_info, true);
5235 * If we have enough space then hooray, make our reservation and carry
5236 * on. If not see if we can overcommit, and if we can, hooray carry on.
5237 * If not things get more complicated.
5239 if (used + orig_bytes <= space_info->total_bytes) {
5240 update_bytes_may_use(space_info, orig_bytes);
5241 trace_btrfs_space_reservation(fs_info, "space_info",
5242 space_info->flags, orig_bytes, 1);
5244 } else if (can_overcommit(fs_info, space_info, orig_bytes, flush,
5246 update_bytes_may_use(space_info, orig_bytes);
5247 trace_btrfs_space_reservation(fs_info, "space_info",
5248 space_info->flags, orig_bytes, 1);
5253 * If we couldn't make a reservation then setup our reservation ticket
5254 * and kick the async worker if it's not already running.
5256 * If we are a priority flusher then we just need to add our ticket to
5257 * the list and we will do our own flushing further down.
5259 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
5260 ticket.orig_bytes = orig_bytes;
5261 ticket.bytes = orig_bytes;
5263 init_waitqueue_head(&ticket.wait);
5264 if (flush == BTRFS_RESERVE_FLUSH_ALL) {
5265 list_add_tail(&ticket.list, &space_info->tickets);
5266 if (!space_info->flush) {
5267 space_info->flush = 1;
5268 trace_btrfs_trigger_flush(fs_info,
5272 queue_work(system_unbound_wq,
5273 &fs_info->async_reclaim_work);
5276 list_add_tail(&ticket.list,
5277 &space_info->priority_tickets);
5279 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
5282 * We will do the space reservation dance during log replay,
5283 * which means we won't have fs_info->fs_root set, so don't do
5284 * the async reclaim as we will panic.
5286 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) &&
5287 need_do_async_reclaim(fs_info, space_info,
5288 used, system_chunk) &&
5289 !work_busy(&fs_info->async_reclaim_work)) {
5290 trace_btrfs_trigger_flush(fs_info, space_info->flags,
5291 orig_bytes, flush, "preempt");
5292 queue_work(system_unbound_wq,
5293 &fs_info->async_reclaim_work);
5296 spin_unlock(&space_info->lock);
5297 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
5300 if (flush == BTRFS_RESERVE_FLUSH_ALL)
5301 return wait_reserve_ticket(fs_info, space_info, &ticket);
5304 priority_reclaim_metadata_space(fs_info, space_info, &ticket);
5305 spin_lock(&space_info->lock);
5307 if (ticket.bytes < orig_bytes)
5308 reclaim_bytes = orig_bytes - ticket.bytes;
5309 list_del_init(&ticket.list);
5312 spin_unlock(&space_info->lock);
5315 space_info_add_old_bytes(fs_info, space_info, reclaim_bytes);
5316 ASSERT(list_empty(&ticket.list));
5321 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5322 * @root - the root we're allocating for
5323 * @block_rsv - the block_rsv we're allocating for
5324 * @orig_bytes - the number of bytes we want
5325 * @flush - whether or not we can flush to make our reservation
5327 * This will reserve orig_bytes number of bytes from the space info associated
5328 * with the block_rsv. If there is not enough space it will make an attempt to
5329 * flush out space to make room. It will do this by flushing delalloc if
5330 * possible or committing the transaction. If flush is 0 then no attempts to
5331 * regain reservations will be made and this will fail if there is not enough
5334 static int reserve_metadata_bytes(struct btrfs_root *root,
5335 struct btrfs_block_rsv *block_rsv,
5337 enum btrfs_reserve_flush_enum flush)
5339 struct btrfs_fs_info *fs_info = root->fs_info;
5340 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5342 bool system_chunk = (root == fs_info->chunk_root);
5344 ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info,
5345 orig_bytes, flush, system_chunk);
5346 if (ret == -ENOSPC &&
5347 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5348 if (block_rsv != global_rsv &&
5349 !block_rsv_use_bytes(global_rsv, orig_bytes))
5352 if (ret == -ENOSPC) {
5353 trace_btrfs_space_reservation(fs_info, "space_info:enospc",
5354 block_rsv->space_info->flags,
5357 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG))
5358 dump_space_info(fs_info, block_rsv->space_info,
5364 static struct btrfs_block_rsv *get_block_rsv(
5365 const struct btrfs_trans_handle *trans,
5366 const struct btrfs_root *root)
5368 struct btrfs_fs_info *fs_info = root->fs_info;
5369 struct btrfs_block_rsv *block_rsv = NULL;
5371 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5372 (root == fs_info->csum_root && trans->adding_csums) ||
5373 (root == fs_info->uuid_root))
5374 block_rsv = trans->block_rsv;
5377 block_rsv = root->block_rsv;
5380 block_rsv = &fs_info->empty_block_rsv;
5385 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5389 spin_lock(&block_rsv->lock);
5390 if (block_rsv->reserved >= num_bytes) {
5391 block_rsv->reserved -= num_bytes;
5392 if (block_rsv->reserved < block_rsv->size)
5393 block_rsv->full = 0;
5396 spin_unlock(&block_rsv->lock);
5400 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5401 u64 num_bytes, bool update_size)
5403 spin_lock(&block_rsv->lock);
5404 block_rsv->reserved += num_bytes;
5406 block_rsv->size += num_bytes;
5407 else if (block_rsv->reserved >= block_rsv->size)
5408 block_rsv->full = 1;
5409 spin_unlock(&block_rsv->lock);
5412 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5413 struct btrfs_block_rsv *dest, u64 num_bytes,
5416 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5419 if (global_rsv->space_info != dest->space_info)
5422 spin_lock(&global_rsv->lock);
5423 min_bytes = div_factor(global_rsv->size, min_factor);
5424 if (global_rsv->reserved < min_bytes + num_bytes) {
5425 spin_unlock(&global_rsv->lock);
5428 global_rsv->reserved -= num_bytes;
5429 if (global_rsv->reserved < global_rsv->size)
5430 global_rsv->full = 0;
5431 spin_unlock(&global_rsv->lock);
5433 block_rsv_add_bytes(dest, num_bytes, true);
5438 * btrfs_migrate_to_delayed_refs_rsv - transfer bytes to our delayed refs rsv.
5439 * @fs_info - the fs info for our fs.
5440 * @src - the source block rsv to transfer from.
5441 * @num_bytes - the number of bytes to transfer.
5443 * This transfers up to the num_bytes amount from the src rsv to the
5444 * delayed_refs_rsv. Any extra bytes are returned to the space info.
5446 void btrfs_migrate_to_delayed_refs_rsv(struct btrfs_fs_info *fs_info,
5447 struct btrfs_block_rsv *src,
5450 struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv;
5453 spin_lock(&src->lock);
5454 src->reserved -= num_bytes;
5455 src->size -= num_bytes;
5456 spin_unlock(&src->lock);
5458 spin_lock(&delayed_refs_rsv->lock);
5459 if (delayed_refs_rsv->size > delayed_refs_rsv->reserved) {
5460 u64 delta = delayed_refs_rsv->size -
5461 delayed_refs_rsv->reserved;
5462 if (num_bytes > delta) {
5463 to_free = num_bytes - delta;
5467 to_free = num_bytes;
5472 delayed_refs_rsv->reserved += num_bytes;
5473 if (delayed_refs_rsv->reserved >= delayed_refs_rsv->size)
5474 delayed_refs_rsv->full = 1;
5475 spin_unlock(&delayed_refs_rsv->lock);
5478 trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv",
5481 space_info_add_old_bytes(fs_info, delayed_refs_rsv->space_info,
5486 * btrfs_delayed_refs_rsv_refill - refill based on our delayed refs usage.
5487 * @fs_info - the fs_info for our fs.
5488 * @flush - control how we can flush for this reservation.
5490 * This will refill the delayed block_rsv up to 1 items size worth of space and
5491 * will return -ENOSPC if we can't make the reservation.
5493 int btrfs_delayed_refs_rsv_refill(struct btrfs_fs_info *fs_info,
5494 enum btrfs_reserve_flush_enum flush)
5496 struct btrfs_block_rsv *block_rsv = &fs_info->delayed_refs_rsv;
5497 u64 limit = btrfs_calc_trans_metadata_size(fs_info, 1);
5501 spin_lock(&block_rsv->lock);
5502 if (block_rsv->reserved < block_rsv->size) {
5503 num_bytes = block_rsv->size - block_rsv->reserved;
5504 num_bytes = min(num_bytes, limit);
5506 spin_unlock(&block_rsv->lock);
5511 ret = reserve_metadata_bytes(fs_info->extent_root, block_rsv,
5515 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5516 trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv",
5522 * This is for space we already have accounted in space_info->bytes_may_use, so
5523 * basically when we're returning space from block_rsv's.
5525 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
5526 struct btrfs_space_info *space_info,
5529 struct reserve_ticket *ticket;
5530 struct list_head *head;
5532 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH;
5533 bool check_overcommit = false;
5535 spin_lock(&space_info->lock);
5536 head = &space_info->priority_tickets;
5539 * If we are over our limit then we need to check and see if we can
5540 * overcommit, and if we can't then we just need to free up our space
5541 * and not satisfy any requests.
5543 used = btrfs_space_info_used(space_info, true);
5544 if (used - num_bytes >= space_info->total_bytes)
5545 check_overcommit = true;
5547 while (!list_empty(head) && num_bytes) {
5548 ticket = list_first_entry(head, struct reserve_ticket,
5551 * We use 0 bytes because this space is already reserved, so
5552 * adding the ticket space would be a double count.
5554 if (check_overcommit &&
5555 !can_overcommit(fs_info, space_info, 0, flush, false))
5557 if (num_bytes >= ticket->bytes) {
5558 list_del_init(&ticket->list);
5559 num_bytes -= ticket->bytes;
5561 space_info->tickets_id++;
5562 wake_up(&ticket->wait);
5564 ticket->bytes -= num_bytes;
5569 if (num_bytes && head == &space_info->priority_tickets) {
5570 head = &space_info->tickets;
5571 flush = BTRFS_RESERVE_FLUSH_ALL;
5574 update_bytes_may_use(space_info, -num_bytes);
5575 trace_btrfs_space_reservation(fs_info, "space_info",
5576 space_info->flags, num_bytes, 0);
5577 spin_unlock(&space_info->lock);
5581 * This is for newly allocated space that isn't accounted in
5582 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5583 * we use this helper.
5585 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
5586 struct btrfs_space_info *space_info,
5589 struct reserve_ticket *ticket;
5590 struct list_head *head = &space_info->priority_tickets;
5593 while (!list_empty(head) && num_bytes) {
5594 ticket = list_first_entry(head, struct reserve_ticket,
5596 if (num_bytes >= ticket->bytes) {
5597 trace_btrfs_space_reservation(fs_info, "space_info",
5600 list_del_init(&ticket->list);
5601 num_bytes -= ticket->bytes;
5602 update_bytes_may_use(space_info, ticket->bytes);
5604 space_info->tickets_id++;
5605 wake_up(&ticket->wait);
5607 trace_btrfs_space_reservation(fs_info, "space_info",
5610 update_bytes_may_use(space_info, num_bytes);
5611 ticket->bytes -= num_bytes;
5616 if (num_bytes && head == &space_info->priority_tickets) {
5617 head = &space_info->tickets;
5622 static u64 block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5623 struct btrfs_block_rsv *block_rsv,
5624 struct btrfs_block_rsv *dest, u64 num_bytes,
5625 u64 *qgroup_to_release_ret)
5627 struct btrfs_space_info *space_info = block_rsv->space_info;
5628 u64 qgroup_to_release = 0;
5631 spin_lock(&block_rsv->lock);
5632 if (num_bytes == (u64)-1) {
5633 num_bytes = block_rsv->size;
5634 qgroup_to_release = block_rsv->qgroup_rsv_size;
5636 block_rsv->size -= num_bytes;
5637 if (block_rsv->reserved >= block_rsv->size) {
5638 num_bytes = block_rsv->reserved - block_rsv->size;
5639 block_rsv->reserved = block_rsv->size;
5640 block_rsv->full = 1;
5644 if (block_rsv->qgroup_rsv_reserved >= block_rsv->qgroup_rsv_size) {
5645 qgroup_to_release = block_rsv->qgroup_rsv_reserved -
5646 block_rsv->qgroup_rsv_size;
5647 block_rsv->qgroup_rsv_reserved = block_rsv->qgroup_rsv_size;
5649 qgroup_to_release = 0;
5651 spin_unlock(&block_rsv->lock);
5654 if (num_bytes > 0) {
5656 spin_lock(&dest->lock);
5660 bytes_to_add = dest->size - dest->reserved;
5661 bytes_to_add = min(num_bytes, bytes_to_add);
5662 dest->reserved += bytes_to_add;
5663 if (dest->reserved >= dest->size)
5665 num_bytes -= bytes_to_add;
5667 spin_unlock(&dest->lock);
5670 space_info_add_old_bytes(fs_info, space_info,
5673 if (qgroup_to_release_ret)
5674 *qgroup_to_release_ret = qgroup_to_release;
5678 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
5679 struct btrfs_block_rsv *dst, u64 num_bytes,
5684 ret = block_rsv_use_bytes(src, num_bytes);
5688 block_rsv_add_bytes(dst, num_bytes, update_size);
5692 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5694 memset(rsv, 0, sizeof(*rsv));
5695 spin_lock_init(&rsv->lock);
5699 void btrfs_init_metadata_block_rsv(struct btrfs_fs_info *fs_info,
5700 struct btrfs_block_rsv *rsv,
5701 unsigned short type)
5703 btrfs_init_block_rsv(rsv, type);
5704 rsv->space_info = __find_space_info(fs_info,
5705 BTRFS_BLOCK_GROUP_METADATA);
5708 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info,
5709 unsigned short type)
5711 struct btrfs_block_rsv *block_rsv;
5713 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5717 btrfs_init_metadata_block_rsv(fs_info, block_rsv, type);
5721 void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info,
5722 struct btrfs_block_rsv *rsv)
5726 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
5730 int btrfs_block_rsv_add(struct btrfs_root *root,
5731 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5732 enum btrfs_reserve_flush_enum flush)
5739 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5741 block_rsv_add_bytes(block_rsv, num_bytes, true);
5746 int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor)
5754 spin_lock(&block_rsv->lock);
5755 num_bytes = div_factor(block_rsv->size, min_factor);
5756 if (block_rsv->reserved >= num_bytes)
5758 spin_unlock(&block_rsv->lock);
5763 int btrfs_block_rsv_refill(struct btrfs_root *root,
5764 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5765 enum btrfs_reserve_flush_enum flush)
5773 spin_lock(&block_rsv->lock);
5774 num_bytes = min_reserved;
5775 if (block_rsv->reserved >= num_bytes)
5778 num_bytes -= block_rsv->reserved;
5779 spin_unlock(&block_rsv->lock);
5784 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5786 block_rsv_add_bytes(block_rsv, num_bytes, false);
5793 static void calc_refill_bytes(struct btrfs_block_rsv *block_rsv,
5794 u64 *metadata_bytes, u64 *qgroup_bytes)
5796 *metadata_bytes = 0;
5799 spin_lock(&block_rsv->lock);
5800 if (block_rsv->reserved < block_rsv->size)
5801 *metadata_bytes = block_rsv->size - block_rsv->reserved;
5802 if (block_rsv->qgroup_rsv_reserved < block_rsv->qgroup_rsv_size)
5803 *qgroup_bytes = block_rsv->qgroup_rsv_size -
5804 block_rsv->qgroup_rsv_reserved;
5805 spin_unlock(&block_rsv->lock);
5809 * btrfs_inode_rsv_refill - refill the inode block rsv.
5810 * @inode - the inode we are refilling.
5811 * @flush - the flushing restriction.
5813 * Essentially the same as btrfs_block_rsv_refill, except it uses the
5814 * block_rsv->size as the minimum size. We'll either refill the missing amount
5815 * or return if we already have enough space. This will also handle the reserve
5816 * tracepoint for the reserved amount.
5818 static int btrfs_inode_rsv_refill(struct btrfs_inode *inode,
5819 enum btrfs_reserve_flush_enum flush)
5821 struct btrfs_root *root = inode->root;
5822 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5823 u64 num_bytes, last = 0;
5824 u64 qgroup_num_bytes;
5827 calc_refill_bytes(block_rsv, &num_bytes, &qgroup_num_bytes);
5832 ret = btrfs_qgroup_reserve_meta_prealloc(root, qgroup_num_bytes,
5836 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5838 btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes);
5841 * If we are fragmented we can end up with a lot of
5842 * outstanding extents which will make our size be much
5843 * larger than our reserved amount.
5845 * If the reservation happens here, it might be very
5846 * big though not needed in the end, if the delalloc
5849 * If this is the case try and do the reserve again.
5851 if (flush == BTRFS_RESERVE_FLUSH_ALL)
5852 calc_refill_bytes(block_rsv, &num_bytes,
5857 } while (ret && last != num_bytes);
5860 block_rsv_add_bytes(block_rsv, num_bytes, false);
5861 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5862 btrfs_ino(inode), num_bytes, 1);
5864 /* Don't forget to increase qgroup_rsv_reserved */
5865 spin_lock(&block_rsv->lock);
5866 block_rsv->qgroup_rsv_reserved += qgroup_num_bytes;
5867 spin_unlock(&block_rsv->lock);
5872 static u64 __btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
5873 struct btrfs_block_rsv *block_rsv,
5874 u64 num_bytes, u64 *qgroup_to_release)
5876 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5877 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_refs_rsv;
5878 struct btrfs_block_rsv *target = delayed_rsv;
5880 if (target->full || target == block_rsv)
5881 target = global_rsv;
5883 if (block_rsv->space_info != target->space_info)
5886 return block_rsv_release_bytes(fs_info, block_rsv, target, num_bytes,
5890 void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
5891 struct btrfs_block_rsv *block_rsv,
5894 __btrfs_block_rsv_release(fs_info, block_rsv, num_bytes, NULL);
5898 * btrfs_inode_rsv_release - release any excessive reservation.
5899 * @inode - the inode we need to release from.
5900 * @qgroup_free - free or convert qgroup meta.
5901 * Unlike normal operation, qgroup meta reservation needs to know if we are
5902 * freeing qgroup reservation or just converting it into per-trans. Normally
5903 * @qgroup_free is true for error handling, and false for normal release.
5905 * This is the same as btrfs_block_rsv_release, except that it handles the
5906 * tracepoint for the reservation.
5908 static void btrfs_inode_rsv_release(struct btrfs_inode *inode, bool qgroup_free)
5910 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5911 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5913 u64 qgroup_to_release = 0;
5916 * Since we statically set the block_rsv->size we just want to say we
5917 * are releasing 0 bytes, and then we'll just get the reservation over
5920 released = __btrfs_block_rsv_release(fs_info, block_rsv, 0,
5921 &qgroup_to_release);
5923 trace_btrfs_space_reservation(fs_info, "delalloc",
5924 btrfs_ino(inode), released, 0);
5926 btrfs_qgroup_free_meta_prealloc(inode->root, qgroup_to_release);
5928 btrfs_qgroup_convert_reserved_meta(inode->root,
5933 * btrfs_delayed_refs_rsv_release - release a ref head's reservation.
5934 * @fs_info - the fs_info for our fs.
5935 * @nr - the number of items to drop.
5937 * This drops the delayed ref head's count from the delayed refs rsv and frees
5938 * any excess reservation we had.
5940 void btrfs_delayed_refs_rsv_release(struct btrfs_fs_info *fs_info, int nr)
5942 struct btrfs_block_rsv *block_rsv = &fs_info->delayed_refs_rsv;
5943 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5944 u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, nr);
5947 released = block_rsv_release_bytes(fs_info, block_rsv, global_rsv,
5950 trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv",
5954 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5956 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5957 struct btrfs_space_info *sinfo = block_rsv->space_info;
5961 * The global block rsv is based on the size of the extent tree, the
5962 * checksum tree and the root tree. If the fs is empty we want to set
5963 * it to a minimal amount for safety.
5965 num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) +
5966 btrfs_root_used(&fs_info->csum_root->root_item) +
5967 btrfs_root_used(&fs_info->tree_root->root_item);
5968 num_bytes = max_t(u64, num_bytes, SZ_16M);
5970 spin_lock(&sinfo->lock);
5971 spin_lock(&block_rsv->lock);
5973 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
5975 if (block_rsv->reserved < block_rsv->size) {
5976 num_bytes = btrfs_space_info_used(sinfo, true);
5977 if (sinfo->total_bytes > num_bytes) {
5978 num_bytes = sinfo->total_bytes - num_bytes;
5979 num_bytes = min(num_bytes,
5980 block_rsv->size - block_rsv->reserved);
5981 block_rsv->reserved += num_bytes;
5982 update_bytes_may_use(sinfo, num_bytes);
5983 trace_btrfs_space_reservation(fs_info, "space_info",
5984 sinfo->flags, num_bytes,
5987 } else if (block_rsv->reserved > block_rsv->size) {
5988 num_bytes = block_rsv->reserved - block_rsv->size;
5989 update_bytes_may_use(sinfo, -num_bytes);
5990 trace_btrfs_space_reservation(fs_info, "space_info",
5991 sinfo->flags, num_bytes, 0);
5992 block_rsv->reserved = block_rsv->size;
5995 if (block_rsv->reserved == block_rsv->size)
5996 block_rsv->full = 1;
5998 block_rsv->full = 0;
6000 spin_unlock(&block_rsv->lock);
6001 spin_unlock(&sinfo->lock);
6004 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
6006 struct btrfs_space_info *space_info;
6008 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
6009 fs_info->chunk_block_rsv.space_info = space_info;
6011 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
6012 fs_info->global_block_rsv.space_info = space_info;
6013 fs_info->trans_block_rsv.space_info = space_info;
6014 fs_info->empty_block_rsv.space_info = space_info;
6015 fs_info->delayed_block_rsv.space_info = space_info;
6016 fs_info->delayed_refs_rsv.space_info = space_info;
6018 fs_info->extent_root->block_rsv = &fs_info->delayed_refs_rsv;
6019 fs_info->csum_root->block_rsv = &fs_info->delayed_refs_rsv;
6020 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
6021 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
6022 if (fs_info->quota_root)
6023 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
6024 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
6026 update_global_block_rsv(fs_info);
6029 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
6031 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
6033 WARN_ON(fs_info->trans_block_rsv.size > 0);
6034 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
6035 WARN_ON(fs_info->chunk_block_rsv.size > 0);
6036 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
6037 WARN_ON(fs_info->delayed_block_rsv.size > 0);
6038 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
6039 WARN_ON(fs_info->delayed_refs_rsv.reserved > 0);
6040 WARN_ON(fs_info->delayed_refs_rsv.size > 0);
6044 * btrfs_update_delayed_refs_rsv - adjust the size of the delayed refs rsv
6045 * @trans - the trans that may have generated delayed refs
6047 * This is to be called anytime we may have adjusted trans->delayed_ref_updates,
6048 * it'll calculate the additional size and add it to the delayed_refs_rsv.
6050 void btrfs_update_delayed_refs_rsv(struct btrfs_trans_handle *trans)
6052 struct btrfs_fs_info *fs_info = trans->fs_info;
6053 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_refs_rsv;
6056 if (!trans->delayed_ref_updates)
6059 num_bytes = btrfs_calc_trans_metadata_size(fs_info,
6060 trans->delayed_ref_updates);
6061 spin_lock(&delayed_rsv->lock);
6062 delayed_rsv->size += num_bytes;
6063 delayed_rsv->full = 0;
6064 spin_unlock(&delayed_rsv->lock);
6065 trans->delayed_ref_updates = 0;
6069 * To be called after all the new block groups attached to the transaction
6070 * handle have been created (btrfs_create_pending_block_groups()).
6072 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
6074 struct btrfs_fs_info *fs_info = trans->fs_info;
6076 if (!trans->chunk_bytes_reserved)
6079 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
6081 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
6082 trans->chunk_bytes_reserved, NULL);
6083 trans->chunk_bytes_reserved = 0;
6087 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
6088 * root: the root of the parent directory
6089 * rsv: block reservation
6090 * items: the number of items that we need do reservation
6091 * use_global_rsv: allow fallback to the global block reservation
6093 * This function is used to reserve the space for snapshot/subvolume
6094 * creation and deletion. Those operations are different with the
6095 * common file/directory operations, they change two fs/file trees
6096 * and root tree, the number of items that the qgroup reserves is
6097 * different with the free space reservation. So we can not use
6098 * the space reservation mechanism in start_transaction().
6100 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
6101 struct btrfs_block_rsv *rsv, int items,
6102 bool use_global_rsv)
6104 u64 qgroup_num_bytes = 0;
6107 struct btrfs_fs_info *fs_info = root->fs_info;
6108 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6110 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
6111 /* One for parent inode, two for dir entries */
6112 qgroup_num_bytes = 3 * fs_info->nodesize;
6113 ret = btrfs_qgroup_reserve_meta_prealloc(root,
6114 qgroup_num_bytes, true);
6119 num_bytes = btrfs_calc_trans_metadata_size(fs_info, items);
6120 rsv->space_info = __find_space_info(fs_info,
6121 BTRFS_BLOCK_GROUP_METADATA);
6122 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
6123 BTRFS_RESERVE_FLUSH_ALL);
6125 if (ret == -ENOSPC && use_global_rsv)
6126 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, true);
6128 if (ret && qgroup_num_bytes)
6129 btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes);
6134 void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info,
6135 struct btrfs_block_rsv *rsv)
6137 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
6140 static void btrfs_calculate_inode_block_rsv_size(struct btrfs_fs_info *fs_info,
6141 struct btrfs_inode *inode)
6143 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
6144 u64 reserve_size = 0;
6145 u64 qgroup_rsv_size = 0;
6147 unsigned outstanding_extents;
6149 lockdep_assert_held(&inode->lock);
6150 outstanding_extents = inode->outstanding_extents;
6151 if (outstanding_extents)
6152 reserve_size = btrfs_calc_trans_metadata_size(fs_info,
6153 outstanding_extents + 1);
6154 csum_leaves = btrfs_csum_bytes_to_leaves(fs_info,
6156 reserve_size += btrfs_calc_trans_metadata_size(fs_info,
6159 * For qgroup rsv, the calculation is very simple:
6160 * account one nodesize for each outstanding extent
6162 * This is overestimating in most cases.
6164 qgroup_rsv_size = (u64)outstanding_extents * fs_info->nodesize;
6166 spin_lock(&block_rsv->lock);
6167 block_rsv->size = reserve_size;
6168 block_rsv->qgroup_rsv_size = qgroup_rsv_size;
6169 spin_unlock(&block_rsv->lock);
6172 int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes)
6174 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6175 unsigned nr_extents;
6176 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
6178 bool delalloc_lock = true;
6180 /* If we are a free space inode we need to not flush since we will be in
6181 * the middle of a transaction commit. We also don't need the delalloc
6182 * mutex since we won't race with anybody. We need this mostly to make
6183 * lockdep shut its filthy mouth.
6185 * If we have a transaction open (can happen if we call truncate_block
6186 * from truncate), then we need FLUSH_LIMIT so we don't deadlock.
6188 if (btrfs_is_free_space_inode(inode)) {
6189 flush = BTRFS_RESERVE_NO_FLUSH;
6190 delalloc_lock = false;
6192 if (current->journal_info)
6193 flush = BTRFS_RESERVE_FLUSH_LIMIT;
6195 if (btrfs_transaction_in_commit(fs_info))
6196 schedule_timeout(1);
6200 mutex_lock(&inode->delalloc_mutex);
6202 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6204 /* Add our new extents and calculate the new rsv size. */
6205 spin_lock(&inode->lock);
6206 nr_extents = count_max_extents(num_bytes);
6207 btrfs_mod_outstanding_extents(inode, nr_extents);
6208 inode->csum_bytes += num_bytes;
6209 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6210 spin_unlock(&inode->lock);
6212 ret = btrfs_inode_rsv_refill(inode, flush);
6217 mutex_unlock(&inode->delalloc_mutex);
6221 spin_lock(&inode->lock);
6222 nr_extents = count_max_extents(num_bytes);
6223 btrfs_mod_outstanding_extents(inode, -nr_extents);
6224 inode->csum_bytes -= num_bytes;
6225 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6226 spin_unlock(&inode->lock);
6228 btrfs_inode_rsv_release(inode, true);
6230 mutex_unlock(&inode->delalloc_mutex);
6235 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
6236 * @inode: the inode to release the reservation for.
6237 * @num_bytes: the number of bytes we are releasing.
6238 * @qgroup_free: free qgroup reservation or convert it to per-trans reservation
6240 * This will release the metadata reservation for an inode. This can be called
6241 * once we complete IO for a given set of bytes to release their metadata
6242 * reservations, or on error for the same reason.
6244 void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes,
6247 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6249 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6250 spin_lock(&inode->lock);
6251 inode->csum_bytes -= num_bytes;
6252 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6253 spin_unlock(&inode->lock);
6255 if (btrfs_is_testing(fs_info))
6258 btrfs_inode_rsv_release(inode, qgroup_free);
6262 * btrfs_delalloc_release_extents - release our outstanding_extents
6263 * @inode: the inode to balance the reservation for.
6264 * @num_bytes: the number of bytes we originally reserved with
6265 * @qgroup_free: do we need to free qgroup meta reservation or convert them.
6267 * When we reserve space we increase outstanding_extents for the extents we may
6268 * add. Once we've set the range as delalloc or created our ordered extents we
6269 * have outstanding_extents to track the real usage, so we use this to free our
6270 * temporarily tracked outstanding_extents. This _must_ be used in conjunction
6271 * with btrfs_delalloc_reserve_metadata.
6273 void btrfs_delalloc_release_extents(struct btrfs_inode *inode, u64 num_bytes,
6276 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6277 unsigned num_extents;
6279 spin_lock(&inode->lock);
6280 num_extents = count_max_extents(num_bytes);
6281 btrfs_mod_outstanding_extents(inode, -num_extents);
6282 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6283 spin_unlock(&inode->lock);
6285 if (btrfs_is_testing(fs_info))
6288 btrfs_inode_rsv_release(inode, qgroup_free);
6292 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6294 * @inode: inode we're writing to
6295 * @start: start range we are writing to
6296 * @len: how long the range we are writing to
6297 * @reserved: mandatory parameter, record actually reserved qgroup ranges of
6298 * current reservation.
6300 * This will do the following things
6302 * o reserve space in data space info for num bytes
6303 * and reserve precious corresponding qgroup space
6304 * (Done in check_data_free_space)
6306 * o reserve space for metadata space, based on the number of outstanding
6307 * extents and how much csums will be needed
6308 * also reserve metadata space in a per root over-reserve method.
6309 * o add to the inodes->delalloc_bytes
6310 * o add it to the fs_info's delalloc inodes list.
6311 * (Above 3 all done in delalloc_reserve_metadata)
6313 * Return 0 for success
6314 * Return <0 for error(-ENOSPC or -EQUOT)
6316 int btrfs_delalloc_reserve_space(struct inode *inode,
6317 struct extent_changeset **reserved, u64 start, u64 len)
6321 ret = btrfs_check_data_free_space(inode, reserved, start, len);
6324 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len);
6326 btrfs_free_reserved_data_space(inode, *reserved, start, len);
6331 * btrfs_delalloc_release_space - release data and metadata space for delalloc
6332 * @inode: inode we're releasing space for
6333 * @start: start position of the space already reserved
6334 * @len: the len of the space already reserved
6335 * @release_bytes: the len of the space we consumed or didn't use
6337 * This function will release the metadata space that was not used and will
6338 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
6339 * list if there are no delalloc bytes left.
6340 * Also it will handle the qgroup reserved space.
6342 void btrfs_delalloc_release_space(struct inode *inode,
6343 struct extent_changeset *reserved,
6344 u64 start, u64 len, bool qgroup_free)
6346 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, qgroup_free);
6347 btrfs_free_reserved_data_space(inode, reserved, start, len);
6350 static int update_block_group(struct btrfs_trans_handle *trans,
6351 u64 bytenr, u64 num_bytes, int alloc)
6353 struct btrfs_fs_info *info = trans->fs_info;
6354 struct btrfs_block_group_cache *cache = NULL;
6355 u64 total = num_bytes;
6361 /* block accounting for super block */
6362 spin_lock(&info->delalloc_root_lock);
6363 old_val = btrfs_super_bytes_used(info->super_copy);
6365 old_val += num_bytes;
6367 old_val -= num_bytes;
6368 btrfs_set_super_bytes_used(info->super_copy, old_val);
6369 spin_unlock(&info->delalloc_root_lock);
6372 cache = btrfs_lookup_block_group(info, bytenr);
6377 factor = btrfs_bg_type_to_factor(cache->flags);
6380 * If this block group has free space cache written out, we
6381 * need to make sure to load it if we are removing space. This
6382 * is because we need the unpinning stage to actually add the
6383 * space back to the block group, otherwise we will leak space.
6385 if (!alloc && cache->cached == BTRFS_CACHE_NO)
6386 cache_block_group(cache, 1);
6388 byte_in_group = bytenr - cache->key.objectid;
6389 WARN_ON(byte_in_group > cache->key.offset);
6391 spin_lock(&cache->space_info->lock);
6392 spin_lock(&cache->lock);
6394 if (btrfs_test_opt(info, SPACE_CACHE) &&
6395 cache->disk_cache_state < BTRFS_DC_CLEAR)
6396 cache->disk_cache_state = BTRFS_DC_CLEAR;
6398 old_val = btrfs_block_group_used(&cache->item);
6399 num_bytes = min(total, cache->key.offset - byte_in_group);
6401 old_val += num_bytes;
6402 btrfs_set_block_group_used(&cache->item, old_val);
6403 cache->reserved -= num_bytes;
6404 cache->space_info->bytes_reserved -= num_bytes;
6405 cache->space_info->bytes_used += num_bytes;
6406 cache->space_info->disk_used += num_bytes * factor;
6407 spin_unlock(&cache->lock);
6408 spin_unlock(&cache->space_info->lock);
6410 old_val -= num_bytes;
6411 btrfs_set_block_group_used(&cache->item, old_val);
6412 cache->pinned += num_bytes;
6413 update_bytes_pinned(cache->space_info, num_bytes);
6414 cache->space_info->bytes_used -= num_bytes;
6415 cache->space_info->disk_used -= num_bytes * factor;
6416 spin_unlock(&cache->lock);
6417 spin_unlock(&cache->space_info->lock);
6419 trace_btrfs_space_reservation(info, "pinned",
6420 cache->space_info->flags,
6422 percpu_counter_add_batch(&cache->space_info->total_bytes_pinned,
6424 BTRFS_TOTAL_BYTES_PINNED_BATCH);
6425 set_extent_dirty(info->pinned_extents,
6426 bytenr, bytenr + num_bytes - 1,
6427 GFP_NOFS | __GFP_NOFAIL);
6430 spin_lock(&trans->transaction->dirty_bgs_lock);
6431 if (list_empty(&cache->dirty_list)) {
6432 list_add_tail(&cache->dirty_list,
6433 &trans->transaction->dirty_bgs);
6434 trans->transaction->num_dirty_bgs++;
6435 trans->delayed_ref_updates++;
6436 btrfs_get_block_group(cache);
6438 spin_unlock(&trans->transaction->dirty_bgs_lock);
6441 * No longer have used bytes in this block group, queue it for
6442 * deletion. We do this after adding the block group to the
6443 * dirty list to avoid races between cleaner kthread and space
6446 if (!alloc && old_val == 0)
6447 btrfs_mark_bg_unused(cache);
6449 btrfs_put_block_group(cache);
6451 bytenr += num_bytes;
6454 /* Modified block groups are accounted for in the delayed_refs_rsv. */
6455 btrfs_update_delayed_refs_rsv(trans);
6459 static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start)
6461 struct btrfs_block_group_cache *cache;
6464 spin_lock(&fs_info->block_group_cache_lock);
6465 bytenr = fs_info->first_logical_byte;
6466 spin_unlock(&fs_info->block_group_cache_lock);
6468 if (bytenr < (u64)-1)
6471 cache = btrfs_lookup_first_block_group(fs_info, search_start);
6475 bytenr = cache->key.objectid;
6476 btrfs_put_block_group(cache);
6481 static int pin_down_extent(struct btrfs_fs_info *fs_info,
6482 struct btrfs_block_group_cache *cache,
6483 u64 bytenr, u64 num_bytes, int reserved)
6485 spin_lock(&cache->space_info->lock);
6486 spin_lock(&cache->lock);
6487 cache->pinned += num_bytes;
6488 update_bytes_pinned(cache->space_info, num_bytes);
6490 cache->reserved -= num_bytes;
6491 cache->space_info->bytes_reserved -= num_bytes;
6493 spin_unlock(&cache->lock);
6494 spin_unlock(&cache->space_info->lock);
6496 trace_btrfs_space_reservation(fs_info, "pinned",
6497 cache->space_info->flags, num_bytes, 1);
6498 percpu_counter_add_batch(&cache->space_info->total_bytes_pinned,
6499 num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH);
6500 set_extent_dirty(fs_info->pinned_extents, bytenr,
6501 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
6506 * this function must be called within transaction
6508 int btrfs_pin_extent(struct btrfs_fs_info *fs_info,
6509 u64 bytenr, u64 num_bytes, int reserved)
6511 struct btrfs_block_group_cache *cache;
6513 cache = btrfs_lookup_block_group(fs_info, bytenr);
6514 BUG_ON(!cache); /* Logic error */
6516 pin_down_extent(fs_info, cache, bytenr, num_bytes, reserved);
6518 btrfs_put_block_group(cache);
6523 * this function must be called within transaction
6525 int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info,
6526 u64 bytenr, u64 num_bytes)
6528 struct btrfs_block_group_cache *cache;
6531 cache = btrfs_lookup_block_group(fs_info, bytenr);
6536 * pull in the free space cache (if any) so that our pin
6537 * removes the free space from the cache. We have load_only set
6538 * to one because the slow code to read in the free extents does check
6539 * the pinned extents.
6541 cache_block_group(cache, 1);
6543 pin_down_extent(fs_info, cache, bytenr, num_bytes, 0);
6545 /* remove us from the free space cache (if we're there at all) */
6546 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6547 btrfs_put_block_group(cache);
6551 static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
6552 u64 start, u64 num_bytes)
6555 struct btrfs_block_group_cache *block_group;
6556 struct btrfs_caching_control *caching_ctl;
6558 block_group = btrfs_lookup_block_group(fs_info, start);
6562 cache_block_group(block_group, 0);
6563 caching_ctl = get_caching_control(block_group);
6567 BUG_ON(!block_group_cache_done(block_group));
6568 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6570 mutex_lock(&caching_ctl->mutex);
6572 if (start >= caching_ctl->progress) {
6573 ret = add_excluded_extent(fs_info, start, num_bytes);
6574 } else if (start + num_bytes <= caching_ctl->progress) {
6575 ret = btrfs_remove_free_space(block_group,
6578 num_bytes = caching_ctl->progress - start;
6579 ret = btrfs_remove_free_space(block_group,
6584 num_bytes = (start + num_bytes) -
6585 caching_ctl->progress;
6586 start = caching_ctl->progress;
6587 ret = add_excluded_extent(fs_info, start, num_bytes);
6590 mutex_unlock(&caching_ctl->mutex);
6591 put_caching_control(caching_ctl);
6593 btrfs_put_block_group(block_group);
6597 int btrfs_exclude_logged_extents(struct extent_buffer *eb)
6599 struct btrfs_fs_info *fs_info = eb->fs_info;
6600 struct btrfs_file_extent_item *item;
6601 struct btrfs_key key;
6606 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
6609 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6610 btrfs_item_key_to_cpu(eb, &key, i);
6611 if (key.type != BTRFS_EXTENT_DATA_KEY)
6613 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6614 found_type = btrfs_file_extent_type(eb, item);
6615 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6617 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6619 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6620 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6621 ret = __exclude_logged_extent(fs_info, key.objectid, key.offset);
6630 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg)
6632 atomic_inc(&bg->reservations);
6635 void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info,
6638 struct btrfs_block_group_cache *bg;
6640 bg = btrfs_lookup_block_group(fs_info, start);
6642 if (atomic_dec_and_test(&bg->reservations))
6643 wake_up_var(&bg->reservations);
6644 btrfs_put_block_group(bg);
6647 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg)
6649 struct btrfs_space_info *space_info = bg->space_info;
6653 if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA))
6657 * Our block group is read only but before we set it to read only,
6658 * some task might have had allocated an extent from it already, but it
6659 * has not yet created a respective ordered extent (and added it to a
6660 * root's list of ordered extents).
6661 * Therefore wait for any task currently allocating extents, since the
6662 * block group's reservations counter is incremented while a read lock
6663 * on the groups' semaphore is held and decremented after releasing
6664 * the read access on that semaphore and creating the ordered extent.
6666 down_write(&space_info->groups_sem);
6667 up_write(&space_info->groups_sem);
6669 wait_var_event(&bg->reservations, !atomic_read(&bg->reservations));
6673 * btrfs_add_reserved_bytes - update the block_group and space info counters
6674 * @cache: The cache we are manipulating
6675 * @ram_bytes: The number of bytes of file content, and will be same to
6676 * @num_bytes except for the compress path.
6677 * @num_bytes: The number of bytes in question
6678 * @delalloc: The blocks are allocated for the delalloc write
6680 * This is called by the allocator when it reserves space. If this is a
6681 * reservation and the block group has become read only we cannot make the
6682 * reservation and return -EAGAIN, otherwise this function always succeeds.
6684 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache,
6685 u64 ram_bytes, u64 num_bytes, int delalloc)
6687 struct btrfs_space_info *space_info = cache->space_info;
6690 spin_lock(&space_info->lock);
6691 spin_lock(&cache->lock);
6695 cache->reserved += num_bytes;
6696 space_info->bytes_reserved += num_bytes;
6697 update_bytes_may_use(space_info, -ram_bytes);
6699 cache->delalloc_bytes += num_bytes;
6701 spin_unlock(&cache->lock);
6702 spin_unlock(&space_info->lock);
6707 * btrfs_free_reserved_bytes - update the block_group and space info counters
6708 * @cache: The cache we are manipulating
6709 * @num_bytes: The number of bytes in question
6710 * @delalloc: The blocks are allocated for the delalloc write
6712 * This is called by somebody who is freeing space that was never actually used
6713 * on disk. For example if you reserve some space for a new leaf in transaction
6714 * A and before transaction A commits you free that leaf, you call this with
6715 * reserve set to 0 in order to clear the reservation.
6718 static void btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache,
6719 u64 num_bytes, int delalloc)
6721 struct btrfs_space_info *space_info = cache->space_info;
6723 spin_lock(&space_info->lock);
6724 spin_lock(&cache->lock);
6726 space_info->bytes_readonly += num_bytes;
6727 cache->reserved -= num_bytes;
6728 space_info->bytes_reserved -= num_bytes;
6729 space_info->max_extent_size = 0;
6732 cache->delalloc_bytes -= num_bytes;
6733 spin_unlock(&cache->lock);
6734 spin_unlock(&space_info->lock);
6736 void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info)
6738 struct btrfs_caching_control *next;
6739 struct btrfs_caching_control *caching_ctl;
6740 struct btrfs_block_group_cache *cache;
6742 down_write(&fs_info->commit_root_sem);
6744 list_for_each_entry_safe(caching_ctl, next,
6745 &fs_info->caching_block_groups, list) {
6746 cache = caching_ctl->block_group;
6747 if (block_group_cache_done(cache)) {
6748 cache->last_byte_to_unpin = (u64)-1;
6749 list_del_init(&caching_ctl->list);
6750 put_caching_control(caching_ctl);
6752 cache->last_byte_to_unpin = caching_ctl->progress;
6756 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6757 fs_info->pinned_extents = &fs_info->freed_extents[1];
6759 fs_info->pinned_extents = &fs_info->freed_extents[0];
6761 up_write(&fs_info->commit_root_sem);
6763 update_global_block_rsv(fs_info);
6767 * Returns the free cluster for the given space info and sets empty_cluster to
6768 * what it should be based on the mount options.
6770 static struct btrfs_free_cluster *
6771 fetch_cluster_info(struct btrfs_fs_info *fs_info,
6772 struct btrfs_space_info *space_info, u64 *empty_cluster)
6774 struct btrfs_free_cluster *ret = NULL;
6777 if (btrfs_mixed_space_info(space_info))
6780 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
6781 ret = &fs_info->meta_alloc_cluster;
6782 if (btrfs_test_opt(fs_info, SSD))
6783 *empty_cluster = SZ_2M;
6785 *empty_cluster = SZ_64K;
6786 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
6787 btrfs_test_opt(fs_info, SSD_SPREAD)) {
6788 *empty_cluster = SZ_2M;
6789 ret = &fs_info->data_alloc_cluster;
6795 static int unpin_extent_range(struct btrfs_fs_info *fs_info,
6797 const bool return_free_space)
6799 struct btrfs_block_group_cache *cache = NULL;
6800 struct btrfs_space_info *space_info;
6801 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6802 struct btrfs_free_cluster *cluster = NULL;
6804 u64 total_unpinned = 0;
6805 u64 empty_cluster = 0;
6808 while (start <= end) {
6811 start >= cache->key.objectid + cache->key.offset) {
6813 btrfs_put_block_group(cache);
6815 cache = btrfs_lookup_block_group(fs_info, start);
6816 BUG_ON(!cache); /* Logic error */
6818 cluster = fetch_cluster_info(fs_info,
6821 empty_cluster <<= 1;
6824 len = cache->key.objectid + cache->key.offset - start;
6825 len = min(len, end + 1 - start);
6827 if (start < cache->last_byte_to_unpin) {
6828 len = min(len, cache->last_byte_to_unpin - start);
6829 if (return_free_space)
6830 btrfs_add_free_space(cache, start, len);
6834 total_unpinned += len;
6835 space_info = cache->space_info;
6838 * If this space cluster has been marked as fragmented and we've
6839 * unpinned enough in this block group to potentially allow a
6840 * cluster to be created inside of it go ahead and clear the
6843 if (cluster && cluster->fragmented &&
6844 total_unpinned > empty_cluster) {
6845 spin_lock(&cluster->lock);
6846 cluster->fragmented = 0;
6847 spin_unlock(&cluster->lock);
6850 spin_lock(&space_info->lock);
6851 spin_lock(&cache->lock);
6852 cache->pinned -= len;
6853 update_bytes_pinned(space_info, -len);
6855 trace_btrfs_space_reservation(fs_info, "pinned",
6856 space_info->flags, len, 0);
6857 space_info->max_extent_size = 0;
6858 percpu_counter_add_batch(&space_info->total_bytes_pinned,
6859 -len, BTRFS_TOTAL_BYTES_PINNED_BATCH);
6861 space_info->bytes_readonly += len;
6864 spin_unlock(&cache->lock);
6865 if (!readonly && return_free_space &&
6866 global_rsv->space_info == space_info) {
6869 spin_lock(&global_rsv->lock);
6870 if (!global_rsv->full) {
6871 to_add = min(len, global_rsv->size -
6872 global_rsv->reserved);
6873 global_rsv->reserved += to_add;
6874 update_bytes_may_use(space_info, to_add);
6875 if (global_rsv->reserved >= global_rsv->size)
6876 global_rsv->full = 1;
6877 trace_btrfs_space_reservation(fs_info,
6883 spin_unlock(&global_rsv->lock);
6884 /* Add to any tickets we may have */
6886 space_info_add_new_bytes(fs_info, space_info,
6889 spin_unlock(&space_info->lock);
6893 btrfs_put_block_group(cache);
6897 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans)
6899 struct btrfs_fs_info *fs_info = trans->fs_info;
6900 struct btrfs_block_group_cache *block_group, *tmp;
6901 struct list_head *deleted_bgs;
6902 struct extent_io_tree *unpin;
6907 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6908 unpin = &fs_info->freed_extents[1];
6910 unpin = &fs_info->freed_extents[0];
6912 while (!trans->aborted) {
6913 struct extent_state *cached_state = NULL;
6915 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6916 ret = find_first_extent_bit(unpin, 0, &start, &end,
6917 EXTENT_DIRTY, &cached_state);
6919 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6923 if (btrfs_test_opt(fs_info, DISCARD))
6924 ret = btrfs_discard_extent(fs_info, start,
6925 end + 1 - start, NULL);
6927 clear_extent_dirty(unpin, start, end, &cached_state);
6928 unpin_extent_range(fs_info, start, end, true);
6929 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6930 free_extent_state(cached_state);
6935 * Transaction is finished. We don't need the lock anymore. We
6936 * do need to clean up the block groups in case of a transaction
6939 deleted_bgs = &trans->transaction->deleted_bgs;
6940 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6944 if (!trans->aborted)
6945 ret = btrfs_discard_extent(fs_info,
6946 block_group->key.objectid,
6947 block_group->key.offset,
6950 list_del_init(&block_group->bg_list);
6951 btrfs_put_block_group_trimming(block_group);
6952 btrfs_put_block_group(block_group);
6955 const char *errstr = btrfs_decode_error(ret);
6957 "discard failed while removing blockgroup: errno=%d %s",
6965 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6966 struct btrfs_delayed_ref_node *node, u64 parent,
6967 u64 root_objectid, u64 owner_objectid,
6968 u64 owner_offset, int refs_to_drop,
6969 struct btrfs_delayed_extent_op *extent_op)
6971 struct btrfs_fs_info *info = trans->fs_info;
6972 struct btrfs_key key;
6973 struct btrfs_path *path;
6974 struct btrfs_root *extent_root = info->extent_root;
6975 struct extent_buffer *leaf;
6976 struct btrfs_extent_item *ei;
6977 struct btrfs_extent_inline_ref *iref;
6980 int extent_slot = 0;
6981 int found_extent = 0;
6985 u64 bytenr = node->bytenr;
6986 u64 num_bytes = node->num_bytes;
6988 bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
6990 path = btrfs_alloc_path();
6994 path->reada = READA_FORWARD;
6995 path->leave_spinning = 1;
6997 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6998 BUG_ON(!is_data && refs_to_drop != 1);
7001 skinny_metadata = false;
7003 ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes,
7004 parent, root_objectid, owner_objectid,
7007 extent_slot = path->slots[0];
7008 while (extent_slot >= 0) {
7009 btrfs_item_key_to_cpu(path->nodes[0], &key,
7011 if (key.objectid != bytenr)
7013 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
7014 key.offset == num_bytes) {
7018 if (key.type == BTRFS_METADATA_ITEM_KEY &&
7019 key.offset == owner_objectid) {
7023 if (path->slots[0] - extent_slot > 5)
7028 if (!found_extent) {
7030 ret = remove_extent_backref(trans, path, NULL,
7032 is_data, &last_ref);
7034 btrfs_abort_transaction(trans, ret);
7037 btrfs_release_path(path);
7038 path->leave_spinning = 1;
7040 key.objectid = bytenr;
7041 key.type = BTRFS_EXTENT_ITEM_KEY;
7042 key.offset = num_bytes;
7044 if (!is_data && skinny_metadata) {
7045 key.type = BTRFS_METADATA_ITEM_KEY;
7046 key.offset = owner_objectid;
7049 ret = btrfs_search_slot(trans, extent_root,
7051 if (ret > 0 && skinny_metadata && path->slots[0]) {
7053 * Couldn't find our skinny metadata item,
7054 * see if we have ye olde extent item.
7057 btrfs_item_key_to_cpu(path->nodes[0], &key,
7059 if (key.objectid == bytenr &&
7060 key.type == BTRFS_EXTENT_ITEM_KEY &&
7061 key.offset == num_bytes)
7065 if (ret > 0 && skinny_metadata) {
7066 skinny_metadata = false;
7067 key.objectid = bytenr;
7068 key.type = BTRFS_EXTENT_ITEM_KEY;
7069 key.offset = num_bytes;
7070 btrfs_release_path(path);
7071 ret = btrfs_search_slot(trans, extent_root,
7077 "umm, got %d back from search, was looking for %llu",
7080 btrfs_print_leaf(path->nodes[0]);
7083 btrfs_abort_transaction(trans, ret);
7086 extent_slot = path->slots[0];
7088 } else if (WARN_ON(ret == -ENOENT)) {
7089 btrfs_print_leaf(path->nodes[0]);
7091 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
7092 bytenr, parent, root_objectid, owner_objectid,
7094 btrfs_abort_transaction(trans, ret);
7097 btrfs_abort_transaction(trans, ret);
7101 leaf = path->nodes[0];
7102 item_size = btrfs_item_size_nr(leaf, extent_slot);
7103 if (unlikely(item_size < sizeof(*ei))) {
7105 btrfs_print_v0_err(info);
7106 btrfs_abort_transaction(trans, ret);
7109 ei = btrfs_item_ptr(leaf, extent_slot,
7110 struct btrfs_extent_item);
7111 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
7112 key.type == BTRFS_EXTENT_ITEM_KEY) {
7113 struct btrfs_tree_block_info *bi;
7114 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
7115 bi = (struct btrfs_tree_block_info *)(ei + 1);
7116 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
7119 refs = btrfs_extent_refs(leaf, ei);
7120 if (refs < refs_to_drop) {
7122 "trying to drop %d refs but we only have %Lu for bytenr %Lu",
7123 refs_to_drop, refs, bytenr);
7125 btrfs_abort_transaction(trans, ret);
7128 refs -= refs_to_drop;
7132 __run_delayed_extent_op(extent_op, leaf, ei);
7134 * In the case of inline back ref, reference count will
7135 * be updated by remove_extent_backref
7138 BUG_ON(!found_extent);
7140 btrfs_set_extent_refs(leaf, ei, refs);
7141 btrfs_mark_buffer_dirty(leaf);
7144 ret = remove_extent_backref(trans, path, iref,
7145 refs_to_drop, is_data,
7148 btrfs_abort_transaction(trans, ret);
7154 BUG_ON(is_data && refs_to_drop !=
7155 extent_data_ref_count(path, iref));
7157 BUG_ON(path->slots[0] != extent_slot);
7159 BUG_ON(path->slots[0] != extent_slot + 1);
7160 path->slots[0] = extent_slot;
7166 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
7169 btrfs_abort_transaction(trans, ret);
7172 btrfs_release_path(path);
7175 ret = btrfs_del_csums(trans, info, bytenr, num_bytes);
7177 btrfs_abort_transaction(trans, ret);
7182 ret = add_to_free_space_tree(trans, bytenr, num_bytes);
7184 btrfs_abort_transaction(trans, ret);
7188 ret = update_block_group(trans, bytenr, num_bytes, 0);
7190 btrfs_abort_transaction(trans, ret);
7194 btrfs_release_path(path);
7197 btrfs_free_path(path);
7202 * when we free an block, it is possible (and likely) that we free the last
7203 * delayed ref for that extent as well. This searches the delayed ref tree for
7204 * a given extent, and if there are no other delayed refs to be processed, it
7205 * removes it from the tree.
7207 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
7210 struct btrfs_delayed_ref_head *head;
7211 struct btrfs_delayed_ref_root *delayed_refs;
7214 delayed_refs = &trans->transaction->delayed_refs;
7215 spin_lock(&delayed_refs->lock);
7216 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
7218 goto out_delayed_unlock;
7220 spin_lock(&head->lock);
7221 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root))
7224 if (cleanup_extent_op(head) != NULL)
7228 * waiting for the lock here would deadlock. If someone else has it
7229 * locked they are already in the process of dropping it anyway
7231 if (!mutex_trylock(&head->mutex))
7234 btrfs_delete_ref_head(delayed_refs, head);
7235 head->processing = 0;
7237 spin_unlock(&head->lock);
7238 spin_unlock(&delayed_refs->lock);
7240 BUG_ON(head->extent_op);
7241 if (head->must_insert_reserved)
7244 btrfs_cleanup_ref_head_accounting(trans->fs_info, delayed_refs, head);
7245 mutex_unlock(&head->mutex);
7246 btrfs_put_delayed_ref_head(head);
7249 spin_unlock(&head->lock);
7252 spin_unlock(&delayed_refs->lock);
7256 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
7257 struct btrfs_root *root,
7258 struct extent_buffer *buf,
7259 u64 parent, int last_ref)
7261 struct btrfs_fs_info *fs_info = root->fs_info;
7265 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7266 int old_ref_mod, new_ref_mod;
7268 btrfs_ref_tree_mod(root, buf->start, buf->len, parent,
7269 root->root_key.objectid,
7270 btrfs_header_level(buf), 0,
7271 BTRFS_DROP_DELAYED_REF);
7272 ret = btrfs_add_delayed_tree_ref(trans, buf->start,
7274 root->root_key.objectid,
7275 btrfs_header_level(buf),
7276 BTRFS_DROP_DELAYED_REF, NULL,
7277 &old_ref_mod, &new_ref_mod);
7278 BUG_ON(ret); /* -ENOMEM */
7279 pin = old_ref_mod >= 0 && new_ref_mod < 0;
7282 if (last_ref && btrfs_header_generation(buf) == trans->transid) {
7283 struct btrfs_block_group_cache *cache;
7285 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7286 ret = check_ref_cleanup(trans, buf->start);
7292 cache = btrfs_lookup_block_group(fs_info, buf->start);
7294 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
7295 pin_down_extent(fs_info, cache, buf->start,
7297 btrfs_put_block_group(cache);
7301 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
7303 btrfs_add_free_space(cache, buf->start, buf->len);
7304 btrfs_free_reserved_bytes(cache, buf->len, 0);
7305 btrfs_put_block_group(cache);
7306 trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
7310 add_pinned_bytes(fs_info, buf->len, true,
7311 root->root_key.objectid);
7315 * Deleting the buffer, clear the corrupt flag since it doesn't
7318 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
7322 /* Can return -ENOMEM */
7323 int btrfs_free_extent(struct btrfs_trans_handle *trans,
7324 struct btrfs_root *root,
7325 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
7326 u64 owner, u64 offset)
7328 struct btrfs_fs_info *fs_info = root->fs_info;
7329 int old_ref_mod, new_ref_mod;
7332 if (btrfs_is_testing(fs_info))
7335 if (root_objectid != BTRFS_TREE_LOG_OBJECTID)
7336 btrfs_ref_tree_mod(root, bytenr, num_bytes, parent,
7337 root_objectid, owner, offset,
7338 BTRFS_DROP_DELAYED_REF);
7341 * tree log blocks never actually go into the extent allocation
7342 * tree, just update pinning info and exit early.
7344 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
7345 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
7346 /* unlocks the pinned mutex */
7347 btrfs_pin_extent(fs_info, bytenr, num_bytes, 1);
7348 old_ref_mod = new_ref_mod = 0;
7350 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
7351 ret = btrfs_add_delayed_tree_ref(trans, bytenr,
7353 root_objectid, (int)owner,
7354 BTRFS_DROP_DELAYED_REF, NULL,
7355 &old_ref_mod, &new_ref_mod);
7357 ret = btrfs_add_delayed_data_ref(trans, bytenr,
7359 root_objectid, owner, offset,
7360 0, BTRFS_DROP_DELAYED_REF,
7361 &old_ref_mod, &new_ref_mod);
7364 if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0) {
7365 bool metadata = owner < BTRFS_FIRST_FREE_OBJECTID;
7367 add_pinned_bytes(fs_info, num_bytes, metadata, root_objectid);
7374 * when we wait for progress in the block group caching, its because
7375 * our allocation attempt failed at least once. So, we must sleep
7376 * and let some progress happen before we try again.
7378 * This function will sleep at least once waiting for new free space to
7379 * show up, and then it will check the block group free space numbers
7380 * for our min num_bytes. Another option is to have it go ahead
7381 * and look in the rbtree for a free extent of a given size, but this
7384 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7385 * any of the information in this block group.
7387 static noinline void
7388 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
7391 struct btrfs_caching_control *caching_ctl;
7393 caching_ctl = get_caching_control(cache);
7397 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
7398 (cache->free_space_ctl->free_space >= num_bytes));
7400 put_caching_control(caching_ctl);
7404 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
7406 struct btrfs_caching_control *caching_ctl;
7409 caching_ctl = get_caching_control(cache);
7411 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
7413 wait_event(caching_ctl->wait, block_group_cache_done(cache));
7414 if (cache->cached == BTRFS_CACHE_ERROR)
7416 put_caching_control(caching_ctl);
7420 enum btrfs_loop_type {
7421 LOOP_CACHING_NOWAIT = 0,
7422 LOOP_CACHING_WAIT = 1,
7423 LOOP_ALLOC_CHUNK = 2,
7424 LOOP_NO_EMPTY_SIZE = 3,
7428 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
7432 down_read(&cache->data_rwsem);
7436 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
7439 btrfs_get_block_group(cache);
7441 down_read(&cache->data_rwsem);
7444 static struct btrfs_block_group_cache *
7445 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
7446 struct btrfs_free_cluster *cluster,
7449 struct btrfs_block_group_cache *used_bg = NULL;
7451 spin_lock(&cluster->refill_lock);
7453 used_bg = cluster->block_group;
7457 if (used_bg == block_group)
7460 btrfs_get_block_group(used_bg);
7465 if (down_read_trylock(&used_bg->data_rwsem))
7468 spin_unlock(&cluster->refill_lock);
7470 /* We should only have one-level nested. */
7471 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
7473 spin_lock(&cluster->refill_lock);
7474 if (used_bg == cluster->block_group)
7477 up_read(&used_bg->data_rwsem);
7478 btrfs_put_block_group(used_bg);
7483 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
7487 up_read(&cache->data_rwsem);
7488 btrfs_put_block_group(cache);
7492 * Structure used internally for find_free_extent() function. Wraps needed
7495 struct find_free_extent_ctl {
7496 /* Basic allocation info */
7503 /* Where to start the search inside the bg */
7506 /* For clustered allocation */
7509 bool have_caching_bg;
7510 bool orig_have_caching_bg;
7512 /* RAID index, converted from flags */
7516 * Current loop number, check find_free_extent_update_loop() for details
7521 * Whether we're refilling a cluster, if true we need to re-search
7522 * current block group but don't try to refill the cluster again.
7524 bool retry_clustered;
7527 * Whether we're updating free space cache, if true we need to re-search
7528 * current block group but don't try updating free space cache again.
7530 bool retry_unclustered;
7532 /* If current block group is cached */
7535 /* Max contiguous hole found */
7536 u64 max_extent_size;
7538 /* Total free space from free space cache, not always contiguous */
7539 u64 total_free_space;
7547 * Helper function for find_free_extent().
7549 * Return -ENOENT to inform caller that we need fallback to unclustered mode.
7550 * Return -EAGAIN to inform caller that we need to re-search this block group
7551 * Return >0 to inform caller that we find nothing
7552 * Return 0 means we have found a location and set ffe_ctl->found_offset.
7554 static int find_free_extent_clustered(struct btrfs_block_group_cache *bg,
7555 struct btrfs_free_cluster *last_ptr,
7556 struct find_free_extent_ctl *ffe_ctl,
7557 struct btrfs_block_group_cache **cluster_bg_ret)
7559 struct btrfs_fs_info *fs_info = bg->fs_info;
7560 struct btrfs_block_group_cache *cluster_bg;
7561 u64 aligned_cluster;
7565 cluster_bg = btrfs_lock_cluster(bg, last_ptr, ffe_ctl->delalloc);
7567 goto refill_cluster;
7568 if (cluster_bg != bg && (cluster_bg->ro ||
7569 !block_group_bits(cluster_bg, ffe_ctl->flags)))
7570 goto release_cluster;
7572 offset = btrfs_alloc_from_cluster(cluster_bg, last_ptr,
7573 ffe_ctl->num_bytes, cluster_bg->key.objectid,
7574 &ffe_ctl->max_extent_size);
7576 /* We have a block, we're done */
7577 spin_unlock(&last_ptr->refill_lock);
7578 trace_btrfs_reserve_extent_cluster(cluster_bg,
7579 ffe_ctl->search_start, ffe_ctl->num_bytes);
7580 *cluster_bg_ret = cluster_bg;
7581 ffe_ctl->found_offset = offset;
7584 WARN_ON(last_ptr->block_group != cluster_bg);
7588 * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so
7589 * lets just skip it and let the allocator find whatever block it can
7590 * find. If we reach this point, we will have tried the cluster
7591 * allocator plenty of times and not have found anything, so we are
7592 * likely way too fragmented for the clustering stuff to find anything.
7594 * However, if the cluster is taken from the current block group,
7595 * release the cluster first, so that we stand a better chance of
7596 * succeeding in the unclustered allocation.
7598 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE && cluster_bg != bg) {
7599 spin_unlock(&last_ptr->refill_lock);
7600 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
7604 /* This cluster didn't work out, free it and start over */
7605 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7607 if (cluster_bg != bg)
7608 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
7611 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE) {
7612 spin_unlock(&last_ptr->refill_lock);
7616 aligned_cluster = max_t(u64,
7617 ffe_ctl->empty_cluster + ffe_ctl->empty_size,
7618 bg->full_stripe_len);
7619 ret = btrfs_find_space_cluster(fs_info, bg, last_ptr,
7620 ffe_ctl->search_start, ffe_ctl->num_bytes,
7623 /* Now pull our allocation out of this cluster */
7624 offset = btrfs_alloc_from_cluster(bg, last_ptr,
7625 ffe_ctl->num_bytes, ffe_ctl->search_start,
7626 &ffe_ctl->max_extent_size);
7628 /* We found one, proceed */
7629 spin_unlock(&last_ptr->refill_lock);
7630 trace_btrfs_reserve_extent_cluster(bg,
7631 ffe_ctl->search_start,
7632 ffe_ctl->num_bytes);
7633 ffe_ctl->found_offset = offset;
7636 } else if (!ffe_ctl->cached && ffe_ctl->loop > LOOP_CACHING_NOWAIT &&
7637 !ffe_ctl->retry_clustered) {
7638 spin_unlock(&last_ptr->refill_lock);
7640 ffe_ctl->retry_clustered = true;
7641 wait_block_group_cache_progress(bg, ffe_ctl->num_bytes +
7642 ffe_ctl->empty_cluster + ffe_ctl->empty_size);
7646 * At this point we either didn't find a cluster or we weren't able to
7647 * allocate a block from our cluster. Free the cluster we've been
7648 * trying to use, and go to the next block group.
7650 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7651 spin_unlock(&last_ptr->refill_lock);
7656 * Return >0 to inform caller that we find nothing
7657 * Return 0 when we found an free extent and set ffe_ctrl->found_offset
7658 * Return -EAGAIN to inform caller that we need to re-search this block group
7660 static int find_free_extent_unclustered(struct btrfs_block_group_cache *bg,
7661 struct btrfs_free_cluster *last_ptr,
7662 struct find_free_extent_ctl *ffe_ctl)
7667 * We are doing an unclustered allocation, set the fragmented flag so
7668 * we don't bother trying to setup a cluster again until we get more
7671 if (unlikely(last_ptr)) {
7672 spin_lock(&last_ptr->lock);
7673 last_ptr->fragmented = 1;
7674 spin_unlock(&last_ptr->lock);
7676 if (ffe_ctl->cached) {
7677 struct btrfs_free_space_ctl *free_space_ctl;
7679 free_space_ctl = bg->free_space_ctl;
7680 spin_lock(&free_space_ctl->tree_lock);
7681 if (free_space_ctl->free_space <
7682 ffe_ctl->num_bytes + ffe_ctl->empty_cluster +
7683 ffe_ctl->empty_size) {
7684 ffe_ctl->total_free_space = max_t(u64,
7685 ffe_ctl->total_free_space,
7686 free_space_ctl->free_space);
7687 spin_unlock(&free_space_ctl->tree_lock);
7690 spin_unlock(&free_space_ctl->tree_lock);
7693 offset = btrfs_find_space_for_alloc(bg, ffe_ctl->search_start,
7694 ffe_ctl->num_bytes, ffe_ctl->empty_size,
7695 &ffe_ctl->max_extent_size);
7698 * If we didn't find a chunk, and we haven't failed on this block group
7699 * before, and this block group is in the middle of caching and we are
7700 * ok with waiting, then go ahead and wait for progress to be made, and
7701 * set @retry_unclustered to true.
7703 * If @retry_unclustered is true then we've already waited on this
7704 * block group once and should move on to the next block group.
7706 if (!offset && !ffe_ctl->retry_unclustered && !ffe_ctl->cached &&
7707 ffe_ctl->loop > LOOP_CACHING_NOWAIT) {
7708 wait_block_group_cache_progress(bg, ffe_ctl->num_bytes +
7709 ffe_ctl->empty_size);
7710 ffe_ctl->retry_unclustered = true;
7712 } else if (!offset) {
7715 ffe_ctl->found_offset = offset;
7720 * Return >0 means caller needs to re-search for free extent
7721 * Return 0 means we have the needed free extent.
7722 * Return <0 means we failed to locate any free extent.
7724 static int find_free_extent_update_loop(struct btrfs_fs_info *fs_info,
7725 struct btrfs_free_cluster *last_ptr,
7726 struct btrfs_key *ins,
7727 struct find_free_extent_ctl *ffe_ctl,
7728 int full_search, bool use_cluster)
7730 struct btrfs_root *root = fs_info->extent_root;
7733 if ((ffe_ctl->loop == LOOP_CACHING_NOWAIT) &&
7734 ffe_ctl->have_caching_bg && !ffe_ctl->orig_have_caching_bg)
7735 ffe_ctl->orig_have_caching_bg = true;
7737 if (!ins->objectid && ffe_ctl->loop >= LOOP_CACHING_WAIT &&
7738 ffe_ctl->have_caching_bg)
7741 if (!ins->objectid && ++(ffe_ctl->index) < BTRFS_NR_RAID_TYPES)
7744 if (ins->objectid) {
7745 if (!use_cluster && last_ptr) {
7746 spin_lock(&last_ptr->lock);
7747 last_ptr->window_start = ins->objectid;
7748 spin_unlock(&last_ptr->lock);
7754 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7755 * caching kthreads as we move along
7756 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7757 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7758 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7761 if (ffe_ctl->loop < LOOP_NO_EMPTY_SIZE) {
7763 if (ffe_ctl->loop == LOOP_CACHING_NOWAIT) {
7765 * We want to skip the LOOP_CACHING_WAIT step if we
7766 * don't have any uncached bgs and we've already done a
7767 * full search through.
7769 if (ffe_ctl->orig_have_caching_bg || !full_search)
7770 ffe_ctl->loop = LOOP_CACHING_WAIT;
7772 ffe_ctl->loop = LOOP_ALLOC_CHUNK;
7777 if (ffe_ctl->loop == LOOP_ALLOC_CHUNK) {
7778 struct btrfs_trans_handle *trans;
7781 trans = current->journal_info;
7785 trans = btrfs_join_transaction(root);
7787 if (IS_ERR(trans)) {
7788 ret = PTR_ERR(trans);
7792 ret = do_chunk_alloc(trans, ffe_ctl->flags,
7796 * If we can't allocate a new chunk we've already looped
7797 * through at least once, move on to the NO_EMPTY_SIZE
7801 ffe_ctl->loop = LOOP_NO_EMPTY_SIZE;
7803 /* Do not bail out on ENOSPC since we can do more. */
7804 if (ret < 0 && ret != -ENOSPC)
7805 btrfs_abort_transaction(trans, ret);
7809 btrfs_end_transaction(trans);
7814 if (ffe_ctl->loop == LOOP_NO_EMPTY_SIZE) {
7816 * Don't loop again if we already have no empty_size and
7819 if (ffe_ctl->empty_size == 0 &&
7820 ffe_ctl->empty_cluster == 0)
7822 ffe_ctl->empty_size = 0;
7823 ffe_ctl->empty_cluster = 0;
7831 * walks the btree of allocated extents and find a hole of a given size.
7832 * The key ins is changed to record the hole:
7833 * ins->objectid == start position
7834 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7835 * ins->offset == the size of the hole.
7836 * Any available blocks before search_start are skipped.
7838 * If there is no suitable free space, we will record the max size of
7839 * the free space extent currently.
7841 * The overall logic and call chain:
7843 * find_free_extent()
7844 * |- Iterate through all block groups
7845 * | |- Get a valid block group
7846 * | |- Try to do clustered allocation in that block group
7847 * | |- Try to do unclustered allocation in that block group
7848 * | |- Check if the result is valid
7849 * | | |- If valid, then exit
7850 * | |- Jump to next block group
7852 * |- Push harder to find free extents
7853 * |- If not found, re-iterate all block groups
7855 static noinline int find_free_extent(struct btrfs_fs_info *fs_info,
7856 u64 ram_bytes, u64 num_bytes, u64 empty_size,
7857 u64 hint_byte, struct btrfs_key *ins,
7858 u64 flags, int delalloc)
7861 struct btrfs_free_cluster *last_ptr = NULL;
7862 struct btrfs_block_group_cache *block_group = NULL;
7863 struct find_free_extent_ctl ffe_ctl = {0};
7864 struct btrfs_space_info *space_info;
7865 bool use_cluster = true;
7866 bool full_search = false;
7868 WARN_ON(num_bytes < fs_info->sectorsize);
7870 ffe_ctl.ram_bytes = ram_bytes;
7871 ffe_ctl.num_bytes = num_bytes;
7872 ffe_ctl.empty_size = empty_size;
7873 ffe_ctl.flags = flags;
7874 ffe_ctl.search_start = 0;
7875 ffe_ctl.retry_clustered = false;
7876 ffe_ctl.retry_unclustered = false;
7877 ffe_ctl.delalloc = delalloc;
7878 ffe_ctl.index = btrfs_bg_flags_to_raid_index(flags);
7879 ffe_ctl.have_caching_bg = false;
7880 ffe_ctl.orig_have_caching_bg = false;
7881 ffe_ctl.found_offset = 0;
7883 ins->type = BTRFS_EXTENT_ITEM_KEY;
7887 trace_find_free_extent(fs_info, num_bytes, empty_size, flags);
7889 space_info = __find_space_info(fs_info, flags);
7891 btrfs_err(fs_info, "No space info for %llu", flags);
7896 * If our free space is heavily fragmented we may not be able to make
7897 * big contiguous allocations, so instead of doing the expensive search
7898 * for free space, simply return ENOSPC with our max_extent_size so we
7899 * can go ahead and search for a more manageable chunk.
7901 * If our max_extent_size is large enough for our allocation simply
7902 * disable clustering since we will likely not be able to find enough
7903 * space to create a cluster and induce latency trying.
7905 if (unlikely(space_info->max_extent_size)) {
7906 spin_lock(&space_info->lock);
7907 if (space_info->max_extent_size &&
7908 num_bytes > space_info->max_extent_size) {
7909 ins->offset = space_info->max_extent_size;
7910 spin_unlock(&space_info->lock);
7912 } else if (space_info->max_extent_size) {
7913 use_cluster = false;
7915 spin_unlock(&space_info->lock);
7918 last_ptr = fetch_cluster_info(fs_info, space_info,
7919 &ffe_ctl.empty_cluster);
7921 spin_lock(&last_ptr->lock);
7922 if (last_ptr->block_group)
7923 hint_byte = last_ptr->window_start;
7924 if (last_ptr->fragmented) {
7926 * We still set window_start so we can keep track of the
7927 * last place we found an allocation to try and save
7930 hint_byte = last_ptr->window_start;
7931 use_cluster = false;
7933 spin_unlock(&last_ptr->lock);
7936 ffe_ctl.search_start = max(ffe_ctl.search_start,
7937 first_logical_byte(fs_info, 0));
7938 ffe_ctl.search_start = max(ffe_ctl.search_start, hint_byte);
7939 if (ffe_ctl.search_start == hint_byte) {
7940 block_group = btrfs_lookup_block_group(fs_info,
7941 ffe_ctl.search_start);
7943 * we don't want to use the block group if it doesn't match our
7944 * allocation bits, or if its not cached.
7946 * However if we are re-searching with an ideal block group
7947 * picked out then we don't care that the block group is cached.
7949 if (block_group && block_group_bits(block_group, flags) &&
7950 block_group->cached != BTRFS_CACHE_NO) {
7951 down_read(&space_info->groups_sem);
7952 if (list_empty(&block_group->list) ||
7955 * someone is removing this block group,
7956 * we can't jump into the have_block_group
7957 * target because our list pointers are not
7960 btrfs_put_block_group(block_group);
7961 up_read(&space_info->groups_sem);
7963 ffe_ctl.index = btrfs_bg_flags_to_raid_index(
7964 block_group->flags);
7965 btrfs_lock_block_group(block_group, delalloc);
7966 goto have_block_group;
7968 } else if (block_group) {
7969 btrfs_put_block_group(block_group);
7973 ffe_ctl.have_caching_bg = false;
7974 if (ffe_ctl.index == btrfs_bg_flags_to_raid_index(flags) ||
7977 down_read(&space_info->groups_sem);
7978 list_for_each_entry(block_group,
7979 &space_info->block_groups[ffe_ctl.index], list) {
7980 /* If the block group is read-only, we can skip it entirely. */
7981 if (unlikely(block_group->ro))
7984 btrfs_grab_block_group(block_group, delalloc);
7985 ffe_ctl.search_start = block_group->key.objectid;
7988 * this can happen if we end up cycling through all the
7989 * raid types, but we want to make sure we only allocate
7990 * for the proper type.
7992 if (!block_group_bits(block_group, flags)) {
7993 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7994 BTRFS_BLOCK_GROUP_RAID1 |
7995 BTRFS_BLOCK_GROUP_RAID5 |
7996 BTRFS_BLOCK_GROUP_RAID6 |
7997 BTRFS_BLOCK_GROUP_RAID10;
8000 * if they asked for extra copies and this block group
8001 * doesn't provide them, bail. This does allow us to
8002 * fill raid0 from raid1.
8004 if ((flags & extra) && !(block_group->flags & extra))
8009 ffe_ctl.cached = block_group_cache_done(block_group);
8010 if (unlikely(!ffe_ctl.cached)) {
8011 ffe_ctl.have_caching_bg = true;
8012 ret = cache_block_group(block_group, 0);
8017 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
8021 * Ok we want to try and use the cluster allocator, so
8024 if (last_ptr && use_cluster) {
8025 struct btrfs_block_group_cache *cluster_bg = NULL;
8027 ret = find_free_extent_clustered(block_group, last_ptr,
8028 &ffe_ctl, &cluster_bg);
8031 if (cluster_bg && cluster_bg != block_group) {
8032 btrfs_release_block_group(block_group,
8034 block_group = cluster_bg;
8037 } else if (ret == -EAGAIN) {
8038 goto have_block_group;
8039 } else if (ret > 0) {
8042 /* ret == -ENOENT case falls through */
8045 ret = find_free_extent_unclustered(block_group, last_ptr,
8048 goto have_block_group;
8051 /* ret == 0 case falls through */
8053 ffe_ctl.search_start = round_up(ffe_ctl.found_offset,
8054 fs_info->stripesize);
8056 /* move on to the next group */
8057 if (ffe_ctl.search_start + num_bytes >
8058 block_group->key.objectid + block_group->key.offset) {
8059 btrfs_add_free_space(block_group, ffe_ctl.found_offset,
8064 if (ffe_ctl.found_offset < ffe_ctl.search_start)
8065 btrfs_add_free_space(block_group, ffe_ctl.found_offset,
8066 ffe_ctl.search_start - ffe_ctl.found_offset);
8068 ret = btrfs_add_reserved_bytes(block_group, ram_bytes,
8069 num_bytes, delalloc);
8070 if (ret == -EAGAIN) {
8071 btrfs_add_free_space(block_group, ffe_ctl.found_offset,
8075 btrfs_inc_block_group_reservations(block_group);
8077 /* we are all good, lets return */
8078 ins->objectid = ffe_ctl.search_start;
8079 ins->offset = num_bytes;
8081 trace_btrfs_reserve_extent(block_group, ffe_ctl.search_start,
8083 btrfs_release_block_group(block_group, delalloc);
8086 ffe_ctl.retry_clustered = false;
8087 ffe_ctl.retry_unclustered = false;
8088 BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) !=
8090 btrfs_release_block_group(block_group, delalloc);
8093 up_read(&space_info->groups_sem);
8095 ret = find_free_extent_update_loop(fs_info, last_ptr, ins, &ffe_ctl,
8096 full_search, use_cluster);
8100 if (ret == -ENOSPC) {
8102 * Use ffe_ctl->total_free_space as fallback if we can't find
8103 * any contiguous hole.
8105 if (!ffe_ctl.max_extent_size)
8106 ffe_ctl.max_extent_size = ffe_ctl.total_free_space;
8107 spin_lock(&space_info->lock);
8108 space_info->max_extent_size = ffe_ctl.max_extent_size;
8109 spin_unlock(&space_info->lock);
8110 ins->offset = ffe_ctl.max_extent_size;
8115 #define DUMP_BLOCK_RSV(fs_info, rsv_name) \
8117 struct btrfs_block_rsv *__rsv = &(fs_info)->rsv_name; \
8118 spin_lock(&__rsv->lock); \
8119 btrfs_info(fs_info, #rsv_name ": size %llu reserved %llu", \
8120 __rsv->size, __rsv->reserved); \
8121 spin_unlock(&__rsv->lock); \
8124 static void dump_space_info(struct btrfs_fs_info *fs_info,
8125 struct btrfs_space_info *info, u64 bytes,
8126 int dump_block_groups)
8128 struct btrfs_block_group_cache *cache;
8131 spin_lock(&info->lock);
8132 btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull",
8134 info->total_bytes - btrfs_space_info_used(info, true),
8135 info->full ? "" : "not ");
8137 "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu",
8138 info->total_bytes, info->bytes_used, info->bytes_pinned,
8139 info->bytes_reserved, info->bytes_may_use,
8140 info->bytes_readonly);
8141 spin_unlock(&info->lock);
8143 DUMP_BLOCK_RSV(fs_info, global_block_rsv);
8144 DUMP_BLOCK_RSV(fs_info, trans_block_rsv);
8145 DUMP_BLOCK_RSV(fs_info, chunk_block_rsv);
8146 DUMP_BLOCK_RSV(fs_info, delayed_block_rsv);
8147 DUMP_BLOCK_RSV(fs_info, delayed_refs_rsv);
8149 if (!dump_block_groups)
8152 down_read(&info->groups_sem);
8154 list_for_each_entry(cache, &info->block_groups[index], list) {
8155 spin_lock(&cache->lock);
8157 "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s",
8158 cache->key.objectid, cache->key.offset,
8159 btrfs_block_group_used(&cache->item), cache->pinned,
8160 cache->reserved, cache->ro ? "[readonly]" : "");
8161 btrfs_dump_free_space(cache, bytes);
8162 spin_unlock(&cache->lock);
8164 if (++index < BTRFS_NR_RAID_TYPES)
8166 up_read(&info->groups_sem);
8170 * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a
8171 * hole that is at least as big as @num_bytes.
8173 * @root - The root that will contain this extent
8175 * @ram_bytes - The amount of space in ram that @num_bytes take. This
8176 * is used for accounting purposes. This value differs
8177 * from @num_bytes only in the case of compressed extents.
8179 * @num_bytes - Number of bytes to allocate on-disk.
8181 * @min_alloc_size - Indicates the minimum amount of space that the
8182 * allocator should try to satisfy. In some cases
8183 * @num_bytes may be larger than what is required and if
8184 * the filesystem is fragmented then allocation fails.
8185 * However, the presence of @min_alloc_size gives a
8186 * chance to try and satisfy the smaller allocation.
8188 * @empty_size - A hint that you plan on doing more COW. This is the
8189 * size in bytes the allocator should try to find free
8190 * next to the block it returns. This is just a hint and
8191 * may be ignored by the allocator.
8193 * @hint_byte - Hint to the allocator to start searching above the byte
8194 * address passed. It might be ignored.
8196 * @ins - This key is modified to record the found hole. It will
8197 * have the following values:
8198 * ins->objectid == start position
8199 * ins->flags = BTRFS_EXTENT_ITEM_KEY
8200 * ins->offset == the size of the hole.
8202 * @is_data - Boolean flag indicating whether an extent is
8203 * allocated for data (true) or metadata (false)
8205 * @delalloc - Boolean flag indicating whether this allocation is for
8206 * delalloc or not. If 'true' data_rwsem of block groups
8207 * is going to be acquired.
8210 * Returns 0 when an allocation succeeded or < 0 when an error occurred. In
8211 * case -ENOSPC is returned then @ins->offset will contain the size of the
8212 * largest available hole the allocator managed to find.
8214 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
8215 u64 num_bytes, u64 min_alloc_size,
8216 u64 empty_size, u64 hint_byte,
8217 struct btrfs_key *ins, int is_data, int delalloc)
8219 struct btrfs_fs_info *fs_info = root->fs_info;
8220 bool final_tried = num_bytes == min_alloc_size;
8224 flags = get_alloc_profile_by_root(root, is_data);
8226 WARN_ON(num_bytes < fs_info->sectorsize);
8227 ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size,
8228 hint_byte, ins, flags, delalloc);
8229 if (!ret && !is_data) {
8230 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
8231 } else if (ret == -ENOSPC) {
8232 if (!final_tried && ins->offset) {
8233 num_bytes = min(num_bytes >> 1, ins->offset);
8234 num_bytes = round_down(num_bytes,
8235 fs_info->sectorsize);
8236 num_bytes = max(num_bytes, min_alloc_size);
8237 ram_bytes = num_bytes;
8238 if (num_bytes == min_alloc_size)
8241 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8242 struct btrfs_space_info *sinfo;
8244 sinfo = __find_space_info(fs_info, flags);
8246 "allocation failed flags %llu, wanted %llu",
8249 dump_space_info(fs_info, sinfo, num_bytes, 1);
8256 static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8258 int pin, int delalloc)
8260 struct btrfs_block_group_cache *cache;
8263 cache = btrfs_lookup_block_group(fs_info, start);
8265 btrfs_err(fs_info, "Unable to find block group for %llu",
8271 pin_down_extent(fs_info, cache, start, len, 1);
8273 if (btrfs_test_opt(fs_info, DISCARD))
8274 ret = btrfs_discard_extent(fs_info, start, len, NULL);
8275 btrfs_add_free_space(cache, start, len);
8276 btrfs_free_reserved_bytes(cache, len, delalloc);
8277 trace_btrfs_reserved_extent_free(fs_info, start, len);
8280 btrfs_put_block_group(cache);
8284 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8285 u64 start, u64 len, int delalloc)
8287 return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc);
8290 int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info,
8293 return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0);
8296 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8297 u64 parent, u64 root_objectid,
8298 u64 flags, u64 owner, u64 offset,
8299 struct btrfs_key *ins, int ref_mod)
8301 struct btrfs_fs_info *fs_info = trans->fs_info;
8303 struct btrfs_extent_item *extent_item;
8304 struct btrfs_extent_inline_ref *iref;
8305 struct btrfs_path *path;
8306 struct extent_buffer *leaf;
8311 type = BTRFS_SHARED_DATA_REF_KEY;
8313 type = BTRFS_EXTENT_DATA_REF_KEY;
8315 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
8317 path = btrfs_alloc_path();
8321 path->leave_spinning = 1;
8322 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8325 btrfs_free_path(path);
8329 leaf = path->nodes[0];
8330 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8331 struct btrfs_extent_item);
8332 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
8333 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8334 btrfs_set_extent_flags(leaf, extent_item,
8335 flags | BTRFS_EXTENT_FLAG_DATA);
8337 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8338 btrfs_set_extent_inline_ref_type(leaf, iref, type);
8340 struct btrfs_shared_data_ref *ref;
8341 ref = (struct btrfs_shared_data_ref *)(iref + 1);
8342 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8343 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
8345 struct btrfs_extent_data_ref *ref;
8346 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
8347 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
8348 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
8349 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
8350 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
8353 btrfs_mark_buffer_dirty(path->nodes[0]);
8354 btrfs_free_path(path);
8356 ret = remove_from_free_space_tree(trans, ins->objectid, ins->offset);
8360 ret = update_block_group(trans, ins->objectid, ins->offset, 1);
8361 if (ret) { /* -ENOENT, logic error */
8362 btrfs_err(fs_info, "update block group failed for %llu %llu",
8363 ins->objectid, ins->offset);
8366 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset);
8370 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
8371 struct btrfs_delayed_ref_node *node,
8372 struct btrfs_delayed_extent_op *extent_op)
8374 struct btrfs_fs_info *fs_info = trans->fs_info;
8376 struct btrfs_extent_item *extent_item;
8377 struct btrfs_key extent_key;
8378 struct btrfs_tree_block_info *block_info;
8379 struct btrfs_extent_inline_ref *iref;
8380 struct btrfs_path *path;
8381 struct extent_buffer *leaf;
8382 struct btrfs_delayed_tree_ref *ref;
8383 u32 size = sizeof(*extent_item) + sizeof(*iref);
8385 u64 flags = extent_op->flags_to_set;
8386 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8388 ref = btrfs_delayed_node_to_tree_ref(node);
8390 extent_key.objectid = node->bytenr;
8391 if (skinny_metadata) {
8392 extent_key.offset = ref->level;
8393 extent_key.type = BTRFS_METADATA_ITEM_KEY;
8394 num_bytes = fs_info->nodesize;
8396 extent_key.offset = node->num_bytes;
8397 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
8398 size += sizeof(*block_info);
8399 num_bytes = node->num_bytes;
8402 path = btrfs_alloc_path();
8406 path->leave_spinning = 1;
8407 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8410 btrfs_free_path(path);
8414 leaf = path->nodes[0];
8415 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8416 struct btrfs_extent_item);
8417 btrfs_set_extent_refs(leaf, extent_item, 1);
8418 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8419 btrfs_set_extent_flags(leaf, extent_item,
8420 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
8422 if (skinny_metadata) {
8423 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8425 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
8426 btrfs_set_tree_block_key(leaf, block_info, &extent_op->key);
8427 btrfs_set_tree_block_level(leaf, block_info, ref->level);
8428 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
8431 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) {
8432 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
8433 btrfs_set_extent_inline_ref_type(leaf, iref,
8434 BTRFS_SHARED_BLOCK_REF_KEY);
8435 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent);
8437 btrfs_set_extent_inline_ref_type(leaf, iref,
8438 BTRFS_TREE_BLOCK_REF_KEY);
8439 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root);
8442 btrfs_mark_buffer_dirty(leaf);
8443 btrfs_free_path(path);
8445 ret = remove_from_free_space_tree(trans, extent_key.objectid,
8450 ret = update_block_group(trans, extent_key.objectid,
8451 fs_info->nodesize, 1);
8452 if (ret) { /* -ENOENT, logic error */
8453 btrfs_err(fs_info, "update block group failed for %llu %llu",
8454 extent_key.objectid, extent_key.offset);
8458 trace_btrfs_reserved_extent_alloc(fs_info, extent_key.objectid,
8463 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8464 struct btrfs_root *root, u64 owner,
8465 u64 offset, u64 ram_bytes,
8466 struct btrfs_key *ins)
8470 BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
8472 btrfs_ref_tree_mod(root, ins->objectid, ins->offset, 0,
8473 root->root_key.objectid, owner, offset,
8474 BTRFS_ADD_DELAYED_EXTENT);
8476 ret = btrfs_add_delayed_data_ref(trans, ins->objectid,
8478 root->root_key.objectid, owner,
8480 BTRFS_ADD_DELAYED_EXTENT, NULL, NULL);
8485 * this is used by the tree logging recovery code. It records that
8486 * an extent has been allocated and makes sure to clear the free
8487 * space cache bits as well
8489 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
8490 u64 root_objectid, u64 owner, u64 offset,
8491 struct btrfs_key *ins)
8493 struct btrfs_fs_info *fs_info = trans->fs_info;
8495 struct btrfs_block_group_cache *block_group;
8496 struct btrfs_space_info *space_info;
8499 * Mixed block groups will exclude before processing the log so we only
8500 * need to do the exclude dance if this fs isn't mixed.
8502 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
8503 ret = __exclude_logged_extent(fs_info, ins->objectid,
8509 block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
8513 space_info = block_group->space_info;
8514 spin_lock(&space_info->lock);
8515 spin_lock(&block_group->lock);
8516 space_info->bytes_reserved += ins->offset;
8517 block_group->reserved += ins->offset;
8518 spin_unlock(&block_group->lock);
8519 spin_unlock(&space_info->lock);
8521 ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner,
8523 btrfs_put_block_group(block_group);
8527 static struct extent_buffer *
8528 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
8529 u64 bytenr, int level, u64 owner)
8531 struct btrfs_fs_info *fs_info = root->fs_info;
8532 struct extent_buffer *buf;
8534 buf = btrfs_find_create_tree_block(fs_info, bytenr);
8539 * Extra safety check in case the extent tree is corrupted and extent
8540 * allocator chooses to use a tree block which is already used and
8543 if (buf->lock_owner == current->pid) {
8544 btrfs_err_rl(fs_info,
8545 "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected",
8546 buf->start, btrfs_header_owner(buf), current->pid);
8547 free_extent_buffer(buf);
8548 return ERR_PTR(-EUCLEAN);
8551 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
8552 btrfs_tree_lock(buf);
8553 btrfs_clean_tree_block(buf);
8554 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
8556 btrfs_set_lock_blocking_write(buf);
8557 set_extent_buffer_uptodate(buf);
8559 memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header));
8560 btrfs_set_header_level(buf, level);
8561 btrfs_set_header_bytenr(buf, buf->start);
8562 btrfs_set_header_generation(buf, trans->transid);
8563 btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV);
8564 btrfs_set_header_owner(buf, owner);
8565 write_extent_buffer_fsid(buf, fs_info->fs_devices->metadata_uuid);
8566 write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid);
8567 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
8568 buf->log_index = root->log_transid % 2;
8570 * we allow two log transactions at a time, use different
8571 * EXTENT bit to differentiate dirty pages.
8573 if (buf->log_index == 0)
8574 set_extent_dirty(&root->dirty_log_pages, buf->start,
8575 buf->start + buf->len - 1, GFP_NOFS);
8577 set_extent_new(&root->dirty_log_pages, buf->start,
8578 buf->start + buf->len - 1);
8580 buf->log_index = -1;
8581 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
8582 buf->start + buf->len - 1, GFP_NOFS);
8584 trans->dirty = true;
8585 /* this returns a buffer locked for blocking */
8589 static struct btrfs_block_rsv *
8590 use_block_rsv(struct btrfs_trans_handle *trans,
8591 struct btrfs_root *root, u32 blocksize)
8593 struct btrfs_fs_info *fs_info = root->fs_info;
8594 struct btrfs_block_rsv *block_rsv;
8595 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
8597 bool global_updated = false;
8599 block_rsv = get_block_rsv(trans, root);
8601 if (unlikely(block_rsv->size == 0))
8604 ret = block_rsv_use_bytes(block_rsv, blocksize);
8608 if (block_rsv->failfast)
8609 return ERR_PTR(ret);
8611 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
8612 global_updated = true;
8613 update_global_block_rsv(fs_info);
8618 * The global reserve still exists to save us from ourselves, so don't
8619 * warn_on if we are short on our delayed refs reserve.
8621 if (block_rsv->type != BTRFS_BLOCK_RSV_DELREFS &&
8622 btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8623 static DEFINE_RATELIMIT_STATE(_rs,
8624 DEFAULT_RATELIMIT_INTERVAL * 10,
8625 /*DEFAULT_RATELIMIT_BURST*/ 1);
8626 if (__ratelimit(&_rs))
8628 "BTRFS: block rsv returned %d\n", ret);
8631 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
8632 BTRFS_RESERVE_NO_FLUSH);
8636 * If we couldn't reserve metadata bytes try and use some from
8637 * the global reserve if its space type is the same as the global
8640 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
8641 block_rsv->space_info == global_rsv->space_info) {
8642 ret = block_rsv_use_bytes(global_rsv, blocksize);
8646 return ERR_PTR(ret);
8649 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
8650 struct btrfs_block_rsv *block_rsv, u32 blocksize)
8652 block_rsv_add_bytes(block_rsv, blocksize, false);
8653 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0, NULL);
8657 * finds a free extent and does all the dirty work required for allocation
8658 * returns the tree buffer or an ERR_PTR on error.
8660 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
8661 struct btrfs_root *root,
8662 u64 parent, u64 root_objectid,
8663 const struct btrfs_disk_key *key,
8664 int level, u64 hint,
8667 struct btrfs_fs_info *fs_info = root->fs_info;
8668 struct btrfs_key ins;
8669 struct btrfs_block_rsv *block_rsv;
8670 struct extent_buffer *buf;
8671 struct btrfs_delayed_extent_op *extent_op;
8674 u32 blocksize = fs_info->nodesize;
8675 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8677 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8678 if (btrfs_is_testing(fs_info)) {
8679 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
8680 level, root_objectid);
8682 root->alloc_bytenr += blocksize;
8687 block_rsv = use_block_rsv(trans, root, blocksize);
8688 if (IS_ERR(block_rsv))
8689 return ERR_CAST(block_rsv);
8691 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
8692 empty_size, hint, &ins, 0, 0);
8696 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level,
8700 goto out_free_reserved;
8703 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
8705 parent = ins.objectid;
8706 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
8710 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
8711 extent_op = btrfs_alloc_delayed_extent_op();
8717 memcpy(&extent_op->key, key, sizeof(extent_op->key));
8719 memset(&extent_op->key, 0, sizeof(extent_op->key));
8720 extent_op->flags_to_set = flags;
8721 extent_op->update_key = skinny_metadata ? false : true;
8722 extent_op->update_flags = true;
8723 extent_op->is_data = false;
8724 extent_op->level = level;
8726 btrfs_ref_tree_mod(root, ins.objectid, ins.offset, parent,
8727 root_objectid, level, 0,
8728 BTRFS_ADD_DELAYED_EXTENT);
8729 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
8731 root_objectid, level,
8732 BTRFS_ADD_DELAYED_EXTENT,
8733 extent_op, NULL, NULL);
8735 goto out_free_delayed;
8740 btrfs_free_delayed_extent_op(extent_op);
8742 free_extent_buffer(buf);
8744 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
8746 unuse_block_rsv(fs_info, block_rsv, blocksize);
8747 return ERR_PTR(ret);
8750 struct walk_control {
8751 u64 refs[BTRFS_MAX_LEVEL];
8752 u64 flags[BTRFS_MAX_LEVEL];
8753 struct btrfs_key update_progress;
8754 struct btrfs_key drop_progress;
8766 #define DROP_REFERENCE 1
8767 #define UPDATE_BACKREF 2
8769 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
8770 struct btrfs_root *root,
8771 struct walk_control *wc,
8772 struct btrfs_path *path)
8774 struct btrfs_fs_info *fs_info = root->fs_info;
8780 struct btrfs_key key;
8781 struct extent_buffer *eb;
8786 if (path->slots[wc->level] < wc->reada_slot) {
8787 wc->reada_count = wc->reada_count * 2 / 3;
8788 wc->reada_count = max(wc->reada_count, 2);
8790 wc->reada_count = wc->reada_count * 3 / 2;
8791 wc->reada_count = min_t(int, wc->reada_count,
8792 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
8795 eb = path->nodes[wc->level];
8796 nritems = btrfs_header_nritems(eb);
8798 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
8799 if (nread >= wc->reada_count)
8803 bytenr = btrfs_node_blockptr(eb, slot);
8804 generation = btrfs_node_ptr_generation(eb, slot);
8806 if (slot == path->slots[wc->level])
8809 if (wc->stage == UPDATE_BACKREF &&
8810 generation <= root->root_key.offset)
8813 /* We don't lock the tree block, it's OK to be racy here */
8814 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
8815 wc->level - 1, 1, &refs,
8817 /* We don't care about errors in readahead. */
8822 if (wc->stage == DROP_REFERENCE) {
8826 if (wc->level == 1 &&
8827 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8829 if (!wc->update_ref ||
8830 generation <= root->root_key.offset)
8832 btrfs_node_key_to_cpu(eb, &key, slot);
8833 ret = btrfs_comp_cpu_keys(&key,
8834 &wc->update_progress);
8838 if (wc->level == 1 &&
8839 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8843 readahead_tree_block(fs_info, bytenr);
8846 wc->reada_slot = slot;
8850 * helper to process tree block while walking down the tree.
8852 * when wc->stage == UPDATE_BACKREF, this function updates
8853 * back refs for pointers in the block.
8855 * NOTE: return value 1 means we should stop walking down.
8857 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8858 struct btrfs_root *root,
8859 struct btrfs_path *path,
8860 struct walk_control *wc, int lookup_info)
8862 struct btrfs_fs_info *fs_info = root->fs_info;
8863 int level = wc->level;
8864 struct extent_buffer *eb = path->nodes[level];
8865 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8868 if (wc->stage == UPDATE_BACKREF &&
8869 btrfs_header_owner(eb) != root->root_key.objectid)
8873 * when reference count of tree block is 1, it won't increase
8874 * again. once full backref flag is set, we never clear it.
8877 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8878 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8879 BUG_ON(!path->locks[level]);
8880 ret = btrfs_lookup_extent_info(trans, fs_info,
8881 eb->start, level, 1,
8884 BUG_ON(ret == -ENOMEM);
8887 BUG_ON(wc->refs[level] == 0);
8890 if (wc->stage == DROP_REFERENCE) {
8891 if (wc->refs[level] > 1)
8894 if (path->locks[level] && !wc->keep_locks) {
8895 btrfs_tree_unlock_rw(eb, path->locks[level]);
8896 path->locks[level] = 0;
8901 /* wc->stage == UPDATE_BACKREF */
8902 if (!(wc->flags[level] & flag)) {
8903 BUG_ON(!path->locks[level]);
8904 ret = btrfs_inc_ref(trans, root, eb, 1);
8905 BUG_ON(ret); /* -ENOMEM */
8906 ret = btrfs_dec_ref(trans, root, eb, 0);
8907 BUG_ON(ret); /* -ENOMEM */
8908 ret = btrfs_set_disk_extent_flags(trans, fs_info, eb->start,
8910 btrfs_header_level(eb), 0);
8911 BUG_ON(ret); /* -ENOMEM */
8912 wc->flags[level] |= flag;
8916 * the block is shared by multiple trees, so it's not good to
8917 * keep the tree lock
8919 if (path->locks[level] && level > 0) {
8920 btrfs_tree_unlock_rw(eb, path->locks[level]);
8921 path->locks[level] = 0;
8927 * This is used to verify a ref exists for this root to deal with a bug where we
8928 * would have a drop_progress key that hadn't been updated properly.
8930 static int check_ref_exists(struct btrfs_trans_handle *trans,
8931 struct btrfs_root *root, u64 bytenr, u64 parent,
8934 struct btrfs_path *path;
8935 struct btrfs_extent_inline_ref *iref;
8938 path = btrfs_alloc_path();
8942 ret = lookup_extent_backref(trans, path, &iref, bytenr,
8943 root->fs_info->nodesize, parent,
8944 root->root_key.objectid, level, 0);
8945 btrfs_free_path(path);
8954 * helper to process tree block pointer.
8956 * when wc->stage == DROP_REFERENCE, this function checks
8957 * reference count of the block pointed to. if the block
8958 * is shared and we need update back refs for the subtree
8959 * rooted at the block, this function changes wc->stage to
8960 * UPDATE_BACKREF. if the block is shared and there is no
8961 * need to update back, this function drops the reference
8964 * NOTE: return value 1 means we should stop walking down.
8966 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8967 struct btrfs_root *root,
8968 struct btrfs_path *path,
8969 struct walk_control *wc, int *lookup_info)
8971 struct btrfs_fs_info *fs_info = root->fs_info;
8975 struct btrfs_key key;
8976 struct btrfs_key first_key;
8977 struct extent_buffer *next;
8978 int level = wc->level;
8981 bool need_account = false;
8983 generation = btrfs_node_ptr_generation(path->nodes[level],
8984 path->slots[level]);
8986 * if the lower level block was created before the snapshot
8987 * was created, we know there is no need to update back refs
8990 if (wc->stage == UPDATE_BACKREF &&
8991 generation <= root->root_key.offset) {
8996 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8997 btrfs_node_key_to_cpu(path->nodes[level], &first_key,
8998 path->slots[level]);
9000 next = find_extent_buffer(fs_info, bytenr);
9002 next = btrfs_find_create_tree_block(fs_info, bytenr);
9004 return PTR_ERR(next);
9006 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
9010 btrfs_tree_lock(next);
9011 btrfs_set_lock_blocking_write(next);
9013 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
9014 &wc->refs[level - 1],
9015 &wc->flags[level - 1]);
9019 if (unlikely(wc->refs[level - 1] == 0)) {
9020 btrfs_err(fs_info, "Missing references.");
9026 if (wc->stage == DROP_REFERENCE) {
9027 if (wc->refs[level - 1] > 1) {
9028 need_account = true;
9030 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
9033 if (!wc->update_ref ||
9034 generation <= root->root_key.offset)
9037 btrfs_node_key_to_cpu(path->nodes[level], &key,
9038 path->slots[level]);
9039 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
9043 wc->stage = UPDATE_BACKREF;
9044 wc->shared_level = level - 1;
9048 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
9052 if (!btrfs_buffer_uptodate(next, generation, 0)) {
9053 btrfs_tree_unlock(next);
9054 free_extent_buffer(next);
9060 if (reada && level == 1)
9061 reada_walk_down(trans, root, wc, path);
9062 next = read_tree_block(fs_info, bytenr, generation, level - 1,
9065 return PTR_ERR(next);
9066 } else if (!extent_buffer_uptodate(next)) {
9067 free_extent_buffer(next);
9070 btrfs_tree_lock(next);
9071 btrfs_set_lock_blocking_write(next);
9075 ASSERT(level == btrfs_header_level(next));
9076 if (level != btrfs_header_level(next)) {
9077 btrfs_err(root->fs_info, "mismatched level");
9081 path->nodes[level] = next;
9082 path->slots[level] = 0;
9083 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9089 wc->refs[level - 1] = 0;
9090 wc->flags[level - 1] = 0;
9091 if (wc->stage == DROP_REFERENCE) {
9092 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
9093 parent = path->nodes[level]->start;
9095 ASSERT(root->root_key.objectid ==
9096 btrfs_header_owner(path->nodes[level]));
9097 if (root->root_key.objectid !=
9098 btrfs_header_owner(path->nodes[level])) {
9099 btrfs_err(root->fs_info,
9100 "mismatched block owner");
9108 * If we had a drop_progress we need to verify the refs are set
9109 * as expected. If we find our ref then we know that from here
9110 * on out everything should be correct, and we can clear the
9113 if (wc->restarted) {
9114 ret = check_ref_exists(trans, root, bytenr, parent,
9125 * Reloc tree doesn't contribute to qgroup numbers, and we have
9126 * already accounted them at merge time (replace_path),
9127 * thus we could skip expensive subtree trace here.
9129 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
9131 ret = btrfs_qgroup_trace_subtree(trans, next,
9132 generation, level - 1);
9134 btrfs_err_rl(fs_info,
9135 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
9141 * We need to update the next key in our walk control so we can
9142 * update the drop_progress key accordingly. We don't care if
9143 * find_next_key doesn't find a key because that means we're at
9144 * the end and are going to clean up now.
9146 wc->drop_level = level;
9147 find_next_key(path, level, &wc->drop_progress);
9149 ret = btrfs_free_extent(trans, root, bytenr, fs_info->nodesize,
9150 parent, root->root_key.objectid,
9160 btrfs_tree_unlock(next);
9161 free_extent_buffer(next);
9167 * helper to process tree block while walking up the tree.
9169 * when wc->stage == DROP_REFERENCE, this function drops
9170 * reference count on the block.
9172 * when wc->stage == UPDATE_BACKREF, this function changes
9173 * wc->stage back to DROP_REFERENCE if we changed wc->stage
9174 * to UPDATE_BACKREF previously while processing the block.
9176 * NOTE: return value 1 means we should stop walking up.
9178 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
9179 struct btrfs_root *root,
9180 struct btrfs_path *path,
9181 struct walk_control *wc)
9183 struct btrfs_fs_info *fs_info = root->fs_info;
9185 int level = wc->level;
9186 struct extent_buffer *eb = path->nodes[level];
9189 if (wc->stage == UPDATE_BACKREF) {
9190 BUG_ON(wc->shared_level < level);
9191 if (level < wc->shared_level)
9194 ret = find_next_key(path, level + 1, &wc->update_progress);
9198 wc->stage = DROP_REFERENCE;
9199 wc->shared_level = -1;
9200 path->slots[level] = 0;
9203 * check reference count again if the block isn't locked.
9204 * we should start walking down the tree again if reference
9207 if (!path->locks[level]) {
9209 btrfs_tree_lock(eb);
9210 btrfs_set_lock_blocking_write(eb);
9211 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9213 ret = btrfs_lookup_extent_info(trans, fs_info,
9214 eb->start, level, 1,
9218 btrfs_tree_unlock_rw(eb, path->locks[level]);
9219 path->locks[level] = 0;
9222 BUG_ON(wc->refs[level] == 0);
9223 if (wc->refs[level] == 1) {
9224 btrfs_tree_unlock_rw(eb, path->locks[level]);
9225 path->locks[level] = 0;
9231 /* wc->stage == DROP_REFERENCE */
9232 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
9234 if (wc->refs[level] == 1) {
9236 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9237 ret = btrfs_dec_ref(trans, root, eb, 1);
9239 ret = btrfs_dec_ref(trans, root, eb, 0);
9240 BUG_ON(ret); /* -ENOMEM */
9241 ret = btrfs_qgroup_trace_leaf_items(trans, eb);
9243 btrfs_err_rl(fs_info,
9244 "error %d accounting leaf items. Quota is out of sync, rescan required.",
9248 /* make block locked assertion in btrfs_clean_tree_block happy */
9249 if (!path->locks[level] &&
9250 btrfs_header_generation(eb) == trans->transid) {
9251 btrfs_tree_lock(eb);
9252 btrfs_set_lock_blocking_write(eb);
9253 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9255 btrfs_clean_tree_block(eb);
9258 if (eb == root->node) {
9259 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9261 else if (root->root_key.objectid != btrfs_header_owner(eb))
9262 goto owner_mismatch;
9264 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9265 parent = path->nodes[level + 1]->start;
9266 else if (root->root_key.objectid !=
9267 btrfs_header_owner(path->nodes[level + 1]))
9268 goto owner_mismatch;
9271 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
9273 wc->refs[level] = 0;
9274 wc->flags[level] = 0;
9278 btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu",
9279 btrfs_header_owner(eb), root->root_key.objectid);
9283 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
9284 struct btrfs_root *root,
9285 struct btrfs_path *path,
9286 struct walk_control *wc)
9288 int level = wc->level;
9289 int lookup_info = 1;
9292 while (level >= 0) {
9293 ret = walk_down_proc(trans, root, path, wc, lookup_info);
9300 if (path->slots[level] >=
9301 btrfs_header_nritems(path->nodes[level]))
9304 ret = do_walk_down(trans, root, path, wc, &lookup_info);
9306 path->slots[level]++;
9315 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
9316 struct btrfs_root *root,
9317 struct btrfs_path *path,
9318 struct walk_control *wc, int max_level)
9320 int level = wc->level;
9323 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
9324 while (level < max_level && path->nodes[level]) {
9326 if (path->slots[level] + 1 <
9327 btrfs_header_nritems(path->nodes[level])) {
9328 path->slots[level]++;
9331 ret = walk_up_proc(trans, root, path, wc);
9337 if (path->locks[level]) {
9338 btrfs_tree_unlock_rw(path->nodes[level],
9339 path->locks[level]);
9340 path->locks[level] = 0;
9342 free_extent_buffer(path->nodes[level]);
9343 path->nodes[level] = NULL;
9351 * drop a subvolume tree.
9353 * this function traverses the tree freeing any blocks that only
9354 * referenced by the tree.
9356 * when a shared tree block is found. this function decreases its
9357 * reference count by one. if update_ref is true, this function
9358 * also make sure backrefs for the shared block and all lower level
9359 * blocks are properly updated.
9361 * If called with for_reloc == 0, may exit early with -EAGAIN
9363 int btrfs_drop_snapshot(struct btrfs_root *root,
9364 struct btrfs_block_rsv *block_rsv, int update_ref,
9367 struct btrfs_fs_info *fs_info = root->fs_info;
9368 struct btrfs_path *path;
9369 struct btrfs_trans_handle *trans;
9370 struct btrfs_root *tree_root = fs_info->tree_root;
9371 struct btrfs_root_item *root_item = &root->root_item;
9372 struct walk_control *wc;
9373 struct btrfs_key key;
9377 bool root_dropped = false;
9379 btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid);
9381 path = btrfs_alloc_path();
9387 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9389 btrfs_free_path(path);
9394 trans = btrfs_start_transaction(tree_root, 0);
9395 if (IS_ERR(trans)) {
9396 err = PTR_ERR(trans);
9400 err = btrfs_run_delayed_items(trans);
9405 trans->block_rsv = block_rsv;
9408 * This will help us catch people modifying the fs tree while we're
9409 * dropping it. It is unsafe to mess with the fs tree while it's being
9410 * dropped as we unlock the root node and parent nodes as we walk down
9411 * the tree, assuming nothing will change. If something does change
9412 * then we'll have stale information and drop references to blocks we've
9415 set_bit(BTRFS_ROOT_DELETING, &root->state);
9416 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
9417 level = btrfs_header_level(root->node);
9418 path->nodes[level] = btrfs_lock_root_node(root);
9419 btrfs_set_lock_blocking_write(path->nodes[level]);
9420 path->slots[level] = 0;
9421 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9422 memset(&wc->update_progress, 0,
9423 sizeof(wc->update_progress));
9425 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
9426 memcpy(&wc->update_progress, &key,
9427 sizeof(wc->update_progress));
9429 level = root_item->drop_level;
9431 path->lowest_level = level;
9432 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
9433 path->lowest_level = 0;
9441 * unlock our path, this is safe because only this
9442 * function is allowed to delete this snapshot
9444 btrfs_unlock_up_safe(path, 0);
9446 level = btrfs_header_level(root->node);
9448 btrfs_tree_lock(path->nodes[level]);
9449 btrfs_set_lock_blocking_write(path->nodes[level]);
9450 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9452 ret = btrfs_lookup_extent_info(trans, fs_info,
9453 path->nodes[level]->start,
9454 level, 1, &wc->refs[level],
9460 BUG_ON(wc->refs[level] == 0);
9462 if (level == root_item->drop_level)
9465 btrfs_tree_unlock(path->nodes[level]);
9466 path->locks[level] = 0;
9467 WARN_ON(wc->refs[level] != 1);
9472 wc->restarted = test_bit(BTRFS_ROOT_DEAD_TREE, &root->state);
9474 wc->shared_level = -1;
9475 wc->stage = DROP_REFERENCE;
9476 wc->update_ref = update_ref;
9478 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9482 ret = walk_down_tree(trans, root, path, wc);
9488 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
9495 BUG_ON(wc->stage != DROP_REFERENCE);
9499 if (wc->stage == DROP_REFERENCE) {
9500 wc->drop_level = wc->level;
9501 btrfs_node_key_to_cpu(path->nodes[wc->drop_level],
9503 path->slots[wc->drop_level]);
9505 btrfs_cpu_key_to_disk(&root_item->drop_progress,
9506 &wc->drop_progress);
9507 root_item->drop_level = wc->drop_level;
9509 BUG_ON(wc->level == 0);
9510 if (btrfs_should_end_transaction(trans) ||
9511 (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
9512 ret = btrfs_update_root(trans, tree_root,
9516 btrfs_abort_transaction(trans, ret);
9521 btrfs_end_transaction_throttle(trans);
9522 if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
9523 btrfs_debug(fs_info,
9524 "drop snapshot early exit");
9529 trans = btrfs_start_transaction(tree_root, 0);
9530 if (IS_ERR(trans)) {
9531 err = PTR_ERR(trans);
9535 trans->block_rsv = block_rsv;
9538 btrfs_release_path(path);
9542 ret = btrfs_del_root(trans, &root->root_key);
9544 btrfs_abort_transaction(trans, ret);
9549 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
9550 ret = btrfs_find_root(tree_root, &root->root_key, path,
9553 btrfs_abort_transaction(trans, ret);
9556 } else if (ret > 0) {
9557 /* if we fail to delete the orphan item this time
9558 * around, it'll get picked up the next time.
9560 * The most common failure here is just -ENOENT.
9562 btrfs_del_orphan_item(trans, tree_root,
9563 root->root_key.objectid);
9567 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
9568 btrfs_add_dropped_root(trans, root);
9570 free_extent_buffer(root->node);
9571 free_extent_buffer(root->commit_root);
9572 btrfs_put_fs_root(root);
9574 root_dropped = true;
9576 btrfs_end_transaction_throttle(trans);
9579 btrfs_free_path(path);
9582 * So if we need to stop dropping the snapshot for whatever reason we
9583 * need to make sure to add it back to the dead root list so that we
9584 * keep trying to do the work later. This also cleans up roots if we
9585 * don't have it in the radix (like when we recover after a power fail
9586 * or unmount) so we don't leak memory.
9588 if (!for_reloc && !root_dropped)
9589 btrfs_add_dead_root(root);
9590 if (err && err != -EAGAIN)
9591 btrfs_handle_fs_error(fs_info, err, NULL);
9596 * drop subtree rooted at tree block 'node'.
9598 * NOTE: this function will unlock and release tree block 'node'
9599 * only used by relocation code
9601 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
9602 struct btrfs_root *root,
9603 struct extent_buffer *node,
9604 struct extent_buffer *parent)
9606 struct btrfs_fs_info *fs_info = root->fs_info;
9607 struct btrfs_path *path;
9608 struct walk_control *wc;
9614 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
9616 path = btrfs_alloc_path();
9620 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9622 btrfs_free_path(path);
9626 btrfs_assert_tree_locked(parent);
9627 parent_level = btrfs_header_level(parent);
9628 extent_buffer_get(parent);
9629 path->nodes[parent_level] = parent;
9630 path->slots[parent_level] = btrfs_header_nritems(parent);
9632 btrfs_assert_tree_locked(node);
9633 level = btrfs_header_level(node);
9634 path->nodes[level] = node;
9635 path->slots[level] = 0;
9636 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9638 wc->refs[parent_level] = 1;
9639 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
9641 wc->shared_level = -1;
9642 wc->stage = DROP_REFERENCE;
9645 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9648 wret = walk_down_tree(trans, root, path, wc);
9654 wret = walk_up_tree(trans, root, path, wc, parent_level);
9662 btrfs_free_path(path);
9666 static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags)
9672 * if restripe for this chunk_type is on pick target profile and
9673 * return, otherwise do the usual balance
9675 stripped = get_restripe_target(fs_info, flags);
9677 return extended_to_chunk(stripped);
9679 num_devices = fs_info->fs_devices->rw_devices;
9681 stripped = BTRFS_BLOCK_GROUP_RAID0 |
9682 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
9683 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
9685 if (num_devices == 1) {
9686 stripped |= BTRFS_BLOCK_GROUP_DUP;
9687 stripped = flags & ~stripped;
9689 /* turn raid0 into single device chunks */
9690 if (flags & BTRFS_BLOCK_GROUP_RAID0)
9693 /* turn mirroring into duplication */
9694 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
9695 BTRFS_BLOCK_GROUP_RAID10))
9696 return stripped | BTRFS_BLOCK_GROUP_DUP;
9698 /* they already had raid on here, just return */
9699 if (flags & stripped)
9702 stripped |= BTRFS_BLOCK_GROUP_DUP;
9703 stripped = flags & ~stripped;
9705 /* switch duplicated blocks with raid1 */
9706 if (flags & BTRFS_BLOCK_GROUP_DUP)
9707 return stripped | BTRFS_BLOCK_GROUP_RAID1;
9709 /* this is drive concat, leave it alone */
9715 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
9717 struct btrfs_space_info *sinfo = cache->space_info;
9720 u64 min_allocable_bytes;
9724 * We need some metadata space and system metadata space for
9725 * allocating chunks in some corner cases until we force to set
9726 * it to be readonly.
9729 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
9731 min_allocable_bytes = SZ_1M;
9733 min_allocable_bytes = 0;
9735 spin_lock(&sinfo->lock);
9736 spin_lock(&cache->lock);
9744 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
9745 cache->bytes_super - btrfs_block_group_used(&cache->item);
9746 sinfo_used = btrfs_space_info_used(sinfo, true);
9748 if (sinfo_used + num_bytes + min_allocable_bytes <=
9749 sinfo->total_bytes) {
9750 sinfo->bytes_readonly += num_bytes;
9752 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
9756 spin_unlock(&cache->lock);
9757 spin_unlock(&sinfo->lock);
9758 if (ret == -ENOSPC && btrfs_test_opt(cache->fs_info, ENOSPC_DEBUG)) {
9759 btrfs_info(cache->fs_info,
9760 "unable to make block group %llu ro",
9761 cache->key.objectid);
9762 btrfs_info(cache->fs_info,
9763 "sinfo_used=%llu bg_num_bytes=%llu min_allocable=%llu",
9764 sinfo_used, num_bytes, min_allocable_bytes);
9765 dump_space_info(cache->fs_info, cache->space_info, 0, 0);
9770 int btrfs_inc_block_group_ro(struct btrfs_block_group_cache *cache)
9773 struct btrfs_fs_info *fs_info = cache->fs_info;
9774 struct btrfs_trans_handle *trans;
9779 trans = btrfs_join_transaction(fs_info->extent_root);
9781 return PTR_ERR(trans);
9784 * we're not allowed to set block groups readonly after the dirty
9785 * block groups cache has started writing. If it already started,
9786 * back off and let this transaction commit
9788 mutex_lock(&fs_info->ro_block_group_mutex);
9789 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9790 u64 transid = trans->transid;
9792 mutex_unlock(&fs_info->ro_block_group_mutex);
9793 btrfs_end_transaction(trans);
9795 ret = btrfs_wait_for_commit(fs_info, transid);
9802 * if we are changing raid levels, try to allocate a corresponding
9803 * block group with the new raid level.
9805 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9806 if (alloc_flags != cache->flags) {
9807 ret = do_chunk_alloc(trans, alloc_flags,
9810 * ENOSPC is allowed here, we may have enough space
9811 * already allocated at the new raid level to
9820 ret = inc_block_group_ro(cache, 0);
9823 alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags);
9824 ret = do_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE);
9827 ret = inc_block_group_ro(cache, 0);
9829 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9830 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9831 mutex_lock(&fs_info->chunk_mutex);
9832 check_system_chunk(trans, alloc_flags);
9833 mutex_unlock(&fs_info->chunk_mutex);
9835 mutex_unlock(&fs_info->ro_block_group_mutex);
9837 btrfs_end_transaction(trans);
9841 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, u64 type)
9843 u64 alloc_flags = get_alloc_profile(trans->fs_info, type);
9845 return do_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE);
9849 * helper to account the unused space of all the readonly block group in the
9850 * space_info. takes mirrors into account.
9852 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9854 struct btrfs_block_group_cache *block_group;
9858 /* It's df, we don't care if it's racy */
9859 if (list_empty(&sinfo->ro_bgs))
9862 spin_lock(&sinfo->lock);
9863 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9864 spin_lock(&block_group->lock);
9866 if (!block_group->ro) {
9867 spin_unlock(&block_group->lock);
9871 factor = btrfs_bg_type_to_factor(block_group->flags);
9872 free_bytes += (block_group->key.offset -
9873 btrfs_block_group_used(&block_group->item)) *
9876 spin_unlock(&block_group->lock);
9878 spin_unlock(&sinfo->lock);
9883 void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache)
9885 struct btrfs_space_info *sinfo = cache->space_info;
9890 spin_lock(&sinfo->lock);
9891 spin_lock(&cache->lock);
9893 num_bytes = cache->key.offset - cache->reserved -
9894 cache->pinned - cache->bytes_super -
9895 btrfs_block_group_used(&cache->item);
9896 sinfo->bytes_readonly -= num_bytes;
9897 list_del_init(&cache->ro_list);
9899 spin_unlock(&cache->lock);
9900 spin_unlock(&sinfo->lock);
9904 * Checks to see if it's even possible to relocate this block group.
9906 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9907 * ok to go ahead and try.
9909 int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr)
9911 struct btrfs_block_group_cache *block_group;
9912 struct btrfs_space_info *space_info;
9913 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
9914 struct btrfs_device *device;
9924 debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG);
9926 block_group = btrfs_lookup_block_group(fs_info, bytenr);
9928 /* odd, couldn't find the block group, leave it alone */
9932 "can't find block group for bytenr %llu",
9937 min_free = btrfs_block_group_used(&block_group->item);
9939 /* no bytes used, we're good */
9943 space_info = block_group->space_info;
9944 spin_lock(&space_info->lock);
9946 full = space_info->full;
9949 * if this is the last block group we have in this space, we can't
9950 * relocate it unless we're able to allocate a new chunk below.
9952 * Otherwise, we need to make sure we have room in the space to handle
9953 * all of the extents from this block group. If we can, we're good
9955 if ((space_info->total_bytes != block_group->key.offset) &&
9956 (btrfs_space_info_used(space_info, false) + min_free <
9957 space_info->total_bytes)) {
9958 spin_unlock(&space_info->lock);
9961 spin_unlock(&space_info->lock);
9964 * ok we don't have enough space, but maybe we have free space on our
9965 * devices to allocate new chunks for relocation, so loop through our
9966 * alloc devices and guess if we have enough space. if this block
9967 * group is going to be restriped, run checks against the target
9968 * profile instead of the current one.
9980 target = get_restripe_target(fs_info, block_group->flags);
9982 index = btrfs_bg_flags_to_raid_index(extended_to_chunk(target));
9985 * this is just a balance, so if we were marked as full
9986 * we know there is no space for a new chunk
9991 "no space to alloc new chunk for block group %llu",
9992 block_group->key.objectid);
9996 index = btrfs_bg_flags_to_raid_index(block_group->flags);
9999 if (index == BTRFS_RAID_RAID10) {
10003 } else if (index == BTRFS_RAID_RAID1) {
10005 } else if (index == BTRFS_RAID_DUP) {
10006 /* Multiply by 2 */
10008 } else if (index == BTRFS_RAID_RAID0) {
10009 dev_min = fs_devices->rw_devices;
10010 min_free = div64_u64(min_free, dev_min);
10013 mutex_lock(&fs_info->chunk_mutex);
10014 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
10018 * check to make sure we can actually find a chunk with enough
10019 * space to fit our block group in.
10021 if (device->total_bytes > device->bytes_used + min_free &&
10022 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
10023 ret = find_free_dev_extent(device, min_free,
10024 &dev_offset, NULL);
10028 if (dev_nr >= dev_min)
10034 if (debug && ret == -1)
10035 btrfs_warn(fs_info,
10036 "no space to allocate a new chunk for block group %llu",
10037 block_group->key.objectid);
10038 mutex_unlock(&fs_info->chunk_mutex);
10040 btrfs_put_block_group(block_group);
10044 static int find_first_block_group(struct btrfs_fs_info *fs_info,
10045 struct btrfs_path *path,
10046 struct btrfs_key *key)
10048 struct btrfs_root *root = fs_info->extent_root;
10050 struct btrfs_key found_key;
10051 struct extent_buffer *leaf;
10052 struct btrfs_block_group_item bg;
10056 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
10061 slot = path->slots[0];
10062 leaf = path->nodes[0];
10063 if (slot >= btrfs_header_nritems(leaf)) {
10064 ret = btrfs_next_leaf(root, path);
10071 btrfs_item_key_to_cpu(leaf, &found_key, slot);
10073 if (found_key.objectid >= key->objectid &&
10074 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
10075 struct extent_map_tree *em_tree;
10076 struct extent_map *em;
10078 em_tree = &root->fs_info->mapping_tree.map_tree;
10079 read_lock(&em_tree->lock);
10080 em = lookup_extent_mapping(em_tree, found_key.objectid,
10082 read_unlock(&em_tree->lock);
10085 "logical %llu len %llu found bg but no related chunk",
10086 found_key.objectid, found_key.offset);
10088 } else if (em->start != found_key.objectid ||
10089 em->len != found_key.offset) {
10091 "block group %llu len %llu mismatch with chunk %llu len %llu",
10092 found_key.objectid, found_key.offset,
10093 em->start, em->len);
10096 read_extent_buffer(leaf, &bg,
10097 btrfs_item_ptr_offset(leaf, slot),
10099 flags = btrfs_block_group_flags(&bg) &
10100 BTRFS_BLOCK_GROUP_TYPE_MASK;
10102 if (flags != (em->map_lookup->type &
10103 BTRFS_BLOCK_GROUP_TYPE_MASK)) {
10105 "block group %llu len %llu type flags 0x%llx mismatch with chunk type flags 0x%llx",
10106 found_key.objectid,
10107 found_key.offset, flags,
10108 (BTRFS_BLOCK_GROUP_TYPE_MASK &
10109 em->map_lookup->type));
10115 free_extent_map(em);
10124 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
10126 struct btrfs_block_group_cache *block_group;
10130 struct inode *inode;
10132 block_group = btrfs_lookup_first_block_group(info, last);
10133 while (block_group) {
10134 wait_block_group_cache_done(block_group);
10135 spin_lock(&block_group->lock);
10136 if (block_group->iref)
10138 spin_unlock(&block_group->lock);
10139 block_group = next_block_group(info, block_group);
10141 if (!block_group) {
10148 inode = block_group->inode;
10149 block_group->iref = 0;
10150 block_group->inode = NULL;
10151 spin_unlock(&block_group->lock);
10152 ASSERT(block_group->io_ctl.inode == NULL);
10154 last = block_group->key.objectid + block_group->key.offset;
10155 btrfs_put_block_group(block_group);
10160 * Must be called only after stopping all workers, since we could have block
10161 * group caching kthreads running, and therefore they could race with us if we
10162 * freed the block groups before stopping them.
10164 int btrfs_free_block_groups(struct btrfs_fs_info *info)
10166 struct btrfs_block_group_cache *block_group;
10167 struct btrfs_space_info *space_info;
10168 struct btrfs_caching_control *caching_ctl;
10171 down_write(&info->commit_root_sem);
10172 while (!list_empty(&info->caching_block_groups)) {
10173 caching_ctl = list_entry(info->caching_block_groups.next,
10174 struct btrfs_caching_control, list);
10175 list_del(&caching_ctl->list);
10176 put_caching_control(caching_ctl);
10178 up_write(&info->commit_root_sem);
10180 spin_lock(&info->unused_bgs_lock);
10181 while (!list_empty(&info->unused_bgs)) {
10182 block_group = list_first_entry(&info->unused_bgs,
10183 struct btrfs_block_group_cache,
10185 list_del_init(&block_group->bg_list);
10186 btrfs_put_block_group(block_group);
10188 spin_unlock(&info->unused_bgs_lock);
10190 spin_lock(&info->block_group_cache_lock);
10191 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
10192 block_group = rb_entry(n, struct btrfs_block_group_cache,
10194 rb_erase(&block_group->cache_node,
10195 &info->block_group_cache_tree);
10196 RB_CLEAR_NODE(&block_group->cache_node);
10197 spin_unlock(&info->block_group_cache_lock);
10199 down_write(&block_group->space_info->groups_sem);
10200 list_del(&block_group->list);
10201 up_write(&block_group->space_info->groups_sem);
10204 * We haven't cached this block group, which means we could
10205 * possibly have excluded extents on this block group.
10207 if (block_group->cached == BTRFS_CACHE_NO ||
10208 block_group->cached == BTRFS_CACHE_ERROR)
10209 free_excluded_extents(block_group);
10211 btrfs_remove_free_space_cache(block_group);
10212 ASSERT(block_group->cached != BTRFS_CACHE_STARTED);
10213 ASSERT(list_empty(&block_group->dirty_list));
10214 ASSERT(list_empty(&block_group->io_list));
10215 ASSERT(list_empty(&block_group->bg_list));
10216 ASSERT(atomic_read(&block_group->count) == 1);
10217 btrfs_put_block_group(block_group);
10219 spin_lock(&info->block_group_cache_lock);
10221 spin_unlock(&info->block_group_cache_lock);
10223 /* now that all the block groups are freed, go through and
10224 * free all the space_info structs. This is only called during
10225 * the final stages of unmount, and so we know nobody is
10226 * using them. We call synchronize_rcu() once before we start,
10227 * just to be on the safe side.
10231 release_global_block_rsv(info);
10233 while (!list_empty(&info->space_info)) {
10236 space_info = list_entry(info->space_info.next,
10237 struct btrfs_space_info,
10241 * Do not hide this behind enospc_debug, this is actually
10242 * important and indicates a real bug if this happens.
10244 if (WARN_ON(space_info->bytes_pinned > 0 ||
10245 space_info->bytes_reserved > 0 ||
10246 space_info->bytes_may_use > 0))
10247 dump_space_info(info, space_info, 0, 0);
10248 list_del(&space_info->list);
10249 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
10250 struct kobject *kobj;
10251 kobj = space_info->block_group_kobjs[i];
10252 space_info->block_group_kobjs[i] = NULL;
10258 kobject_del(&space_info->kobj);
10259 kobject_put(&space_info->kobj);
10264 /* link_block_group will queue up kobjects to add when we're reclaim-safe */
10265 void btrfs_add_raid_kobjects(struct btrfs_fs_info *fs_info)
10267 struct btrfs_space_info *space_info;
10268 struct raid_kobject *rkobj;
10273 spin_lock(&fs_info->pending_raid_kobjs_lock);
10274 list_splice_init(&fs_info->pending_raid_kobjs, &list);
10275 spin_unlock(&fs_info->pending_raid_kobjs_lock);
10277 list_for_each_entry(rkobj, &list, list) {
10278 space_info = __find_space_info(fs_info, rkobj->flags);
10279 index = btrfs_bg_flags_to_raid_index(rkobj->flags);
10281 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
10282 "%s", get_raid_name(index));
10284 kobject_put(&rkobj->kobj);
10289 btrfs_warn(fs_info,
10290 "failed to add kobject for block cache, ignoring");
10293 static void link_block_group(struct btrfs_block_group_cache *cache)
10295 struct btrfs_space_info *space_info = cache->space_info;
10296 struct btrfs_fs_info *fs_info = cache->fs_info;
10297 int index = btrfs_bg_flags_to_raid_index(cache->flags);
10298 bool first = false;
10300 down_write(&space_info->groups_sem);
10301 if (list_empty(&space_info->block_groups[index]))
10303 list_add_tail(&cache->list, &space_info->block_groups[index]);
10304 up_write(&space_info->groups_sem);
10307 struct raid_kobject *rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
10309 btrfs_warn(cache->fs_info,
10310 "couldn't alloc memory for raid level kobject");
10313 rkobj->flags = cache->flags;
10314 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
10316 spin_lock(&fs_info->pending_raid_kobjs_lock);
10317 list_add_tail(&rkobj->list, &fs_info->pending_raid_kobjs);
10318 spin_unlock(&fs_info->pending_raid_kobjs_lock);
10319 space_info->block_group_kobjs[index] = &rkobj->kobj;
10323 static struct btrfs_block_group_cache *
10324 btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info,
10325 u64 start, u64 size)
10327 struct btrfs_block_group_cache *cache;
10329 cache = kzalloc(sizeof(*cache), GFP_NOFS);
10333 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
10335 if (!cache->free_space_ctl) {
10340 cache->key.objectid = start;
10341 cache->key.offset = size;
10342 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10344 cache->fs_info = fs_info;
10345 cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start);
10346 set_free_space_tree_thresholds(cache);
10348 atomic_set(&cache->count, 1);
10349 spin_lock_init(&cache->lock);
10350 init_rwsem(&cache->data_rwsem);
10351 INIT_LIST_HEAD(&cache->list);
10352 INIT_LIST_HEAD(&cache->cluster_list);
10353 INIT_LIST_HEAD(&cache->bg_list);
10354 INIT_LIST_HEAD(&cache->ro_list);
10355 INIT_LIST_HEAD(&cache->dirty_list);
10356 INIT_LIST_HEAD(&cache->io_list);
10357 btrfs_init_free_space_ctl(cache);
10358 atomic_set(&cache->trimming, 0);
10359 mutex_init(&cache->free_space_lock);
10360 btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root);
10367 * Iterate all chunks and verify that each of them has the corresponding block
10370 static int check_chunk_block_group_mappings(struct btrfs_fs_info *fs_info)
10372 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
10373 struct extent_map *em;
10374 struct btrfs_block_group_cache *bg;
10379 read_lock(&map_tree->map_tree.lock);
10381 * lookup_extent_mapping will return the first extent map
10382 * intersecting the range, so setting @len to 1 is enough to
10383 * get the first chunk.
10385 em = lookup_extent_mapping(&map_tree->map_tree, start, 1);
10386 read_unlock(&map_tree->map_tree.lock);
10390 bg = btrfs_lookup_block_group(fs_info, em->start);
10393 "chunk start=%llu len=%llu doesn't have corresponding block group",
10394 em->start, em->len);
10396 free_extent_map(em);
10399 if (bg->key.objectid != em->start ||
10400 bg->key.offset != em->len ||
10401 (bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK) !=
10402 (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
10404 "chunk start=%llu len=%llu flags=0x%llx doesn't match block group start=%llu len=%llu flags=0x%llx",
10405 em->start, em->len,
10406 em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK,
10407 bg->key.objectid, bg->key.offset,
10408 bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK);
10410 free_extent_map(em);
10411 btrfs_put_block_group(bg);
10414 start = em->start + em->len;
10415 free_extent_map(em);
10416 btrfs_put_block_group(bg);
10421 int btrfs_read_block_groups(struct btrfs_fs_info *info)
10423 struct btrfs_path *path;
10425 struct btrfs_block_group_cache *cache;
10426 struct btrfs_space_info *space_info;
10427 struct btrfs_key key;
10428 struct btrfs_key found_key;
10429 struct extent_buffer *leaf;
10430 int need_clear = 0;
10435 feature = btrfs_super_incompat_flags(info->super_copy);
10436 mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS);
10440 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10441 path = btrfs_alloc_path();
10444 path->reada = READA_FORWARD;
10446 cache_gen = btrfs_super_cache_generation(info->super_copy);
10447 if (btrfs_test_opt(info, SPACE_CACHE) &&
10448 btrfs_super_generation(info->super_copy) != cache_gen)
10450 if (btrfs_test_opt(info, CLEAR_CACHE))
10454 ret = find_first_block_group(info, path, &key);
10460 leaf = path->nodes[0];
10461 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
10463 cache = btrfs_create_block_group_cache(info, found_key.objectid,
10472 * When we mount with old space cache, we need to
10473 * set BTRFS_DC_CLEAR and set dirty flag.
10475 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
10476 * truncate the old free space cache inode and
10478 * b) Setting 'dirty flag' makes sure that we flush
10479 * the new space cache info onto disk.
10481 if (btrfs_test_opt(info, SPACE_CACHE))
10482 cache->disk_cache_state = BTRFS_DC_CLEAR;
10485 read_extent_buffer(leaf, &cache->item,
10486 btrfs_item_ptr_offset(leaf, path->slots[0]),
10487 sizeof(cache->item));
10488 cache->flags = btrfs_block_group_flags(&cache->item);
10490 ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) &&
10491 (cache->flags & BTRFS_BLOCK_GROUP_DATA))) {
10493 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
10494 cache->key.objectid);
10499 key.objectid = found_key.objectid + found_key.offset;
10500 btrfs_release_path(path);
10503 * We need to exclude the super stripes now so that the space
10504 * info has super bytes accounted for, otherwise we'll think
10505 * we have more space than we actually do.
10507 ret = exclude_super_stripes(cache);
10510 * We may have excluded something, so call this just in
10513 free_excluded_extents(cache);
10514 btrfs_put_block_group(cache);
10519 * check for two cases, either we are full, and therefore
10520 * don't need to bother with the caching work since we won't
10521 * find any space, or we are empty, and we can just add all
10522 * the space in and be done with it. This saves us _a_lot_ of
10523 * time, particularly in the full case.
10525 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
10526 cache->last_byte_to_unpin = (u64)-1;
10527 cache->cached = BTRFS_CACHE_FINISHED;
10528 free_excluded_extents(cache);
10529 } else if (btrfs_block_group_used(&cache->item) == 0) {
10530 cache->last_byte_to_unpin = (u64)-1;
10531 cache->cached = BTRFS_CACHE_FINISHED;
10532 add_new_free_space(cache, found_key.objectid,
10533 found_key.objectid +
10535 free_excluded_extents(cache);
10538 ret = btrfs_add_block_group_cache(info, cache);
10540 btrfs_remove_free_space_cache(cache);
10541 btrfs_put_block_group(cache);
10545 trace_btrfs_add_block_group(info, cache, 0);
10546 update_space_info(info, cache->flags, found_key.offset,
10547 btrfs_block_group_used(&cache->item),
10548 cache->bytes_super, &space_info);
10550 cache->space_info = space_info;
10552 link_block_group(cache);
10554 set_avail_alloc_bits(info, cache->flags);
10555 if (btrfs_chunk_readonly(info, cache->key.objectid)) {
10556 inc_block_group_ro(cache, 1);
10557 } else if (btrfs_block_group_used(&cache->item) == 0) {
10558 ASSERT(list_empty(&cache->bg_list));
10559 btrfs_mark_bg_unused(cache);
10563 list_for_each_entry_rcu(space_info, &info->space_info, list) {
10564 if (!(get_alloc_profile(info, space_info->flags) &
10565 (BTRFS_BLOCK_GROUP_RAID10 |
10566 BTRFS_BLOCK_GROUP_RAID1 |
10567 BTRFS_BLOCK_GROUP_RAID5 |
10568 BTRFS_BLOCK_GROUP_RAID6 |
10569 BTRFS_BLOCK_GROUP_DUP)))
10572 * avoid allocating from un-mirrored block group if there are
10573 * mirrored block groups.
10575 list_for_each_entry(cache,
10576 &space_info->block_groups[BTRFS_RAID_RAID0],
10578 inc_block_group_ro(cache, 1);
10579 list_for_each_entry(cache,
10580 &space_info->block_groups[BTRFS_RAID_SINGLE],
10582 inc_block_group_ro(cache, 1);
10585 btrfs_add_raid_kobjects(info);
10586 init_global_block_rsv(info);
10587 ret = check_chunk_block_group_mappings(info);
10589 btrfs_free_path(path);
10593 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans)
10595 struct btrfs_fs_info *fs_info = trans->fs_info;
10596 struct btrfs_block_group_cache *block_group;
10597 struct btrfs_root *extent_root = fs_info->extent_root;
10598 struct btrfs_block_group_item item;
10599 struct btrfs_key key;
10602 if (!trans->can_flush_pending_bgs)
10605 while (!list_empty(&trans->new_bgs)) {
10606 block_group = list_first_entry(&trans->new_bgs,
10607 struct btrfs_block_group_cache,
10612 spin_lock(&block_group->lock);
10613 memcpy(&item, &block_group->item, sizeof(item));
10614 memcpy(&key, &block_group->key, sizeof(key));
10615 spin_unlock(&block_group->lock);
10617 ret = btrfs_insert_item(trans, extent_root, &key, &item,
10620 btrfs_abort_transaction(trans, ret);
10621 ret = btrfs_finish_chunk_alloc(trans, key.objectid, key.offset);
10623 btrfs_abort_transaction(trans, ret);
10624 add_block_group_free_space(trans, block_group);
10625 /* already aborted the transaction if it failed. */
10627 btrfs_delayed_refs_rsv_release(fs_info, 1);
10628 list_del_init(&block_group->bg_list);
10630 btrfs_trans_release_chunk_metadata(trans);
10633 int btrfs_make_block_group(struct btrfs_trans_handle *trans, u64 bytes_used,
10634 u64 type, u64 chunk_offset, u64 size)
10636 struct btrfs_fs_info *fs_info = trans->fs_info;
10637 struct btrfs_block_group_cache *cache;
10640 btrfs_set_log_full_commit(trans);
10642 cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size);
10646 btrfs_set_block_group_used(&cache->item, bytes_used);
10647 btrfs_set_block_group_chunk_objectid(&cache->item,
10648 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
10649 btrfs_set_block_group_flags(&cache->item, type);
10651 cache->flags = type;
10652 cache->last_byte_to_unpin = (u64)-1;
10653 cache->cached = BTRFS_CACHE_FINISHED;
10654 cache->needs_free_space = 1;
10655 ret = exclude_super_stripes(cache);
10658 * We may have excluded something, so call this just in
10661 free_excluded_extents(cache);
10662 btrfs_put_block_group(cache);
10666 add_new_free_space(cache, chunk_offset, chunk_offset + size);
10668 free_excluded_extents(cache);
10670 #ifdef CONFIG_BTRFS_DEBUG
10671 if (btrfs_should_fragment_free_space(cache)) {
10672 u64 new_bytes_used = size - bytes_used;
10674 bytes_used += new_bytes_used >> 1;
10675 fragment_free_space(cache);
10679 * Ensure the corresponding space_info object is created and
10680 * assigned to our block group. We want our bg to be added to the rbtree
10681 * with its ->space_info set.
10683 cache->space_info = __find_space_info(fs_info, cache->flags);
10684 ASSERT(cache->space_info);
10686 ret = btrfs_add_block_group_cache(fs_info, cache);
10688 btrfs_remove_free_space_cache(cache);
10689 btrfs_put_block_group(cache);
10694 * Now that our block group has its ->space_info set and is inserted in
10695 * the rbtree, update the space info's counters.
10697 trace_btrfs_add_block_group(fs_info, cache, 1);
10698 update_space_info(fs_info, cache->flags, size, bytes_used,
10699 cache->bytes_super, &cache->space_info);
10700 update_global_block_rsv(fs_info);
10702 link_block_group(cache);
10704 list_add_tail(&cache->bg_list, &trans->new_bgs);
10705 trans->delayed_ref_updates++;
10706 btrfs_update_delayed_refs_rsv(trans);
10708 set_avail_alloc_bits(fs_info, type);
10712 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
10714 u64 extra_flags = chunk_to_extended(flags) &
10715 BTRFS_EXTENDED_PROFILE_MASK;
10717 write_seqlock(&fs_info->profiles_lock);
10718 if (flags & BTRFS_BLOCK_GROUP_DATA)
10719 fs_info->avail_data_alloc_bits &= ~extra_flags;
10720 if (flags & BTRFS_BLOCK_GROUP_METADATA)
10721 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
10722 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
10723 fs_info->avail_system_alloc_bits &= ~extra_flags;
10724 write_sequnlock(&fs_info->profiles_lock);
10727 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
10728 u64 group_start, struct extent_map *em)
10730 struct btrfs_fs_info *fs_info = trans->fs_info;
10731 struct btrfs_root *root = fs_info->extent_root;
10732 struct btrfs_path *path;
10733 struct btrfs_block_group_cache *block_group;
10734 struct btrfs_free_cluster *cluster;
10735 struct btrfs_root *tree_root = fs_info->tree_root;
10736 struct btrfs_key key;
10737 struct inode *inode;
10738 struct kobject *kobj = NULL;
10742 struct btrfs_caching_control *caching_ctl = NULL;
10744 bool remove_rsv = false;
10746 block_group = btrfs_lookup_block_group(fs_info, group_start);
10747 BUG_ON(!block_group);
10748 BUG_ON(!block_group->ro);
10750 trace_btrfs_remove_block_group(block_group);
10752 * Free the reserved super bytes from this block group before
10755 free_excluded_extents(block_group);
10756 btrfs_free_ref_tree_range(fs_info, block_group->key.objectid,
10757 block_group->key.offset);
10759 memcpy(&key, &block_group->key, sizeof(key));
10760 index = btrfs_bg_flags_to_raid_index(block_group->flags);
10761 factor = btrfs_bg_type_to_factor(block_group->flags);
10763 /* make sure this block group isn't part of an allocation cluster */
10764 cluster = &fs_info->data_alloc_cluster;
10765 spin_lock(&cluster->refill_lock);
10766 btrfs_return_cluster_to_free_space(block_group, cluster);
10767 spin_unlock(&cluster->refill_lock);
10770 * make sure this block group isn't part of a metadata
10771 * allocation cluster
10773 cluster = &fs_info->meta_alloc_cluster;
10774 spin_lock(&cluster->refill_lock);
10775 btrfs_return_cluster_to_free_space(block_group, cluster);
10776 spin_unlock(&cluster->refill_lock);
10778 path = btrfs_alloc_path();
10785 * get the inode first so any iput calls done for the io_list
10786 * aren't the final iput (no unlinks allowed now)
10788 inode = lookup_free_space_inode(fs_info, block_group, path);
10790 mutex_lock(&trans->transaction->cache_write_mutex);
10792 * Make sure our free space cache IO is done before removing the
10795 spin_lock(&trans->transaction->dirty_bgs_lock);
10796 if (!list_empty(&block_group->io_list)) {
10797 list_del_init(&block_group->io_list);
10799 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
10801 spin_unlock(&trans->transaction->dirty_bgs_lock);
10802 btrfs_wait_cache_io(trans, block_group, path);
10803 btrfs_put_block_group(block_group);
10804 spin_lock(&trans->transaction->dirty_bgs_lock);
10807 if (!list_empty(&block_group->dirty_list)) {
10808 list_del_init(&block_group->dirty_list);
10810 btrfs_put_block_group(block_group);
10812 spin_unlock(&trans->transaction->dirty_bgs_lock);
10813 mutex_unlock(&trans->transaction->cache_write_mutex);
10815 if (!IS_ERR(inode)) {
10816 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10818 btrfs_add_delayed_iput(inode);
10821 clear_nlink(inode);
10822 /* One for the block groups ref */
10823 spin_lock(&block_group->lock);
10824 if (block_group->iref) {
10825 block_group->iref = 0;
10826 block_group->inode = NULL;
10827 spin_unlock(&block_group->lock);
10830 spin_unlock(&block_group->lock);
10832 /* One for our lookup ref */
10833 btrfs_add_delayed_iput(inode);
10836 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
10837 key.offset = block_group->key.objectid;
10840 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
10844 btrfs_release_path(path);
10846 ret = btrfs_del_item(trans, tree_root, path);
10849 btrfs_release_path(path);
10852 spin_lock(&fs_info->block_group_cache_lock);
10853 rb_erase(&block_group->cache_node,
10854 &fs_info->block_group_cache_tree);
10855 RB_CLEAR_NODE(&block_group->cache_node);
10857 if (fs_info->first_logical_byte == block_group->key.objectid)
10858 fs_info->first_logical_byte = (u64)-1;
10859 spin_unlock(&fs_info->block_group_cache_lock);
10861 down_write(&block_group->space_info->groups_sem);
10863 * we must use list_del_init so people can check to see if they
10864 * are still on the list after taking the semaphore
10866 list_del_init(&block_group->list);
10867 if (list_empty(&block_group->space_info->block_groups[index])) {
10868 kobj = block_group->space_info->block_group_kobjs[index];
10869 block_group->space_info->block_group_kobjs[index] = NULL;
10870 clear_avail_alloc_bits(fs_info, block_group->flags);
10872 up_write(&block_group->space_info->groups_sem);
10878 if (block_group->has_caching_ctl)
10879 caching_ctl = get_caching_control(block_group);
10880 if (block_group->cached == BTRFS_CACHE_STARTED)
10881 wait_block_group_cache_done(block_group);
10882 if (block_group->has_caching_ctl) {
10883 down_write(&fs_info->commit_root_sem);
10884 if (!caching_ctl) {
10885 struct btrfs_caching_control *ctl;
10887 list_for_each_entry(ctl,
10888 &fs_info->caching_block_groups, list)
10889 if (ctl->block_group == block_group) {
10891 refcount_inc(&caching_ctl->count);
10896 list_del_init(&caching_ctl->list);
10897 up_write(&fs_info->commit_root_sem);
10899 /* Once for the caching bgs list and once for us. */
10900 put_caching_control(caching_ctl);
10901 put_caching_control(caching_ctl);
10905 spin_lock(&trans->transaction->dirty_bgs_lock);
10906 WARN_ON(!list_empty(&block_group->dirty_list));
10907 WARN_ON(!list_empty(&block_group->io_list));
10908 spin_unlock(&trans->transaction->dirty_bgs_lock);
10910 btrfs_remove_free_space_cache(block_group);
10912 spin_lock(&block_group->space_info->lock);
10913 list_del_init(&block_group->ro_list);
10915 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
10916 WARN_ON(block_group->space_info->total_bytes
10917 < block_group->key.offset);
10918 WARN_ON(block_group->space_info->bytes_readonly
10919 < block_group->key.offset);
10920 WARN_ON(block_group->space_info->disk_total
10921 < block_group->key.offset * factor);
10923 block_group->space_info->total_bytes -= block_group->key.offset;
10924 block_group->space_info->bytes_readonly -= block_group->key.offset;
10925 block_group->space_info->disk_total -= block_group->key.offset * factor;
10927 spin_unlock(&block_group->space_info->lock);
10929 memcpy(&key, &block_group->key, sizeof(key));
10931 mutex_lock(&fs_info->chunk_mutex);
10932 spin_lock(&block_group->lock);
10933 block_group->removed = 1;
10935 * At this point trimming can't start on this block group, because we
10936 * removed the block group from the tree fs_info->block_group_cache_tree
10937 * so no one can't find it anymore and even if someone already got this
10938 * block group before we removed it from the rbtree, they have already
10939 * incremented block_group->trimming - if they didn't, they won't find
10940 * any free space entries because we already removed them all when we
10941 * called btrfs_remove_free_space_cache().
10943 * And we must not remove the extent map from the fs_info->mapping_tree
10944 * to prevent the same logical address range and physical device space
10945 * ranges from being reused for a new block group. This is because our
10946 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10947 * completely transactionless, so while it is trimming a range the
10948 * currently running transaction might finish and a new one start,
10949 * allowing for new block groups to be created that can reuse the same
10950 * physical device locations unless we take this special care.
10952 * There may also be an implicit trim operation if the file system
10953 * is mounted with -odiscard. The same protections must remain
10954 * in place until the extents have been discarded completely when
10955 * the transaction commit has completed.
10957 remove_em = (atomic_read(&block_group->trimming) == 0);
10958 spin_unlock(&block_group->lock);
10961 struct extent_map_tree *em_tree;
10963 em_tree = &fs_info->mapping_tree.map_tree;
10964 write_lock(&em_tree->lock);
10965 remove_extent_mapping(em_tree, em);
10966 write_unlock(&em_tree->lock);
10967 /* once for the tree */
10968 free_extent_map(em);
10971 mutex_unlock(&fs_info->chunk_mutex);
10973 ret = remove_block_group_free_space(trans, block_group);
10977 btrfs_put_block_group(block_group);
10978 btrfs_put_block_group(block_group);
10980 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10986 ret = btrfs_del_item(trans, root, path);
10989 btrfs_delayed_refs_rsv_release(fs_info, 1);
10990 btrfs_free_path(path);
10994 struct btrfs_trans_handle *
10995 btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info,
10996 const u64 chunk_offset)
10998 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
10999 struct extent_map *em;
11000 struct map_lookup *map;
11001 unsigned int num_items;
11003 read_lock(&em_tree->lock);
11004 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
11005 read_unlock(&em_tree->lock);
11006 ASSERT(em && em->start == chunk_offset);
11009 * We need to reserve 3 + N units from the metadata space info in order
11010 * to remove a block group (done at btrfs_remove_chunk() and at
11011 * btrfs_remove_block_group()), which are used for:
11013 * 1 unit for adding the free space inode's orphan (located in the tree
11015 * 1 unit for deleting the block group item (located in the extent
11017 * 1 unit for deleting the free space item (located in tree of tree
11019 * N units for deleting N device extent items corresponding to each
11020 * stripe (located in the device tree).
11022 * In order to remove a block group we also need to reserve units in the
11023 * system space info in order to update the chunk tree (update one or
11024 * more device items and remove one chunk item), but this is done at
11025 * btrfs_remove_chunk() through a call to check_system_chunk().
11027 map = em->map_lookup;
11028 num_items = 3 + map->num_stripes;
11029 free_extent_map(em);
11031 return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root,
11036 * Process the unused_bgs list and remove any that don't have any allocated
11037 * space inside of them.
11039 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
11041 struct btrfs_block_group_cache *block_group;
11042 struct btrfs_space_info *space_info;
11043 struct btrfs_trans_handle *trans;
11046 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
11049 spin_lock(&fs_info->unused_bgs_lock);
11050 while (!list_empty(&fs_info->unused_bgs)) {
11054 block_group = list_first_entry(&fs_info->unused_bgs,
11055 struct btrfs_block_group_cache,
11057 list_del_init(&block_group->bg_list);
11059 space_info = block_group->space_info;
11061 if (ret || btrfs_mixed_space_info(space_info)) {
11062 btrfs_put_block_group(block_group);
11065 spin_unlock(&fs_info->unused_bgs_lock);
11067 mutex_lock(&fs_info->delete_unused_bgs_mutex);
11069 /* Don't want to race with allocators so take the groups_sem */
11070 down_write(&space_info->groups_sem);
11071 spin_lock(&block_group->lock);
11072 if (block_group->reserved || block_group->pinned ||
11073 btrfs_block_group_used(&block_group->item) ||
11075 list_is_singular(&block_group->list)) {
11077 * We want to bail if we made new allocations or have
11078 * outstanding allocations in this block group. We do
11079 * the ro check in case balance is currently acting on
11080 * this block group.
11082 trace_btrfs_skip_unused_block_group(block_group);
11083 spin_unlock(&block_group->lock);
11084 up_write(&space_info->groups_sem);
11087 spin_unlock(&block_group->lock);
11089 /* We don't want to force the issue, only flip if it's ok. */
11090 ret = inc_block_group_ro(block_group, 0);
11091 up_write(&space_info->groups_sem);
11098 * Want to do this before we do anything else so we can recover
11099 * properly if we fail to join the transaction.
11101 trans = btrfs_start_trans_remove_block_group(fs_info,
11102 block_group->key.objectid);
11103 if (IS_ERR(trans)) {
11104 btrfs_dec_block_group_ro(block_group);
11105 ret = PTR_ERR(trans);
11110 * We could have pending pinned extents for this block group,
11111 * just delete them, we don't care about them anymore.
11113 start = block_group->key.objectid;
11114 end = start + block_group->key.offset - 1;
11116 * Hold the unused_bg_unpin_mutex lock to avoid racing with
11117 * btrfs_finish_extent_commit(). If we are at transaction N,
11118 * another task might be running finish_extent_commit() for the
11119 * previous transaction N - 1, and have seen a range belonging
11120 * to the block group in freed_extents[] before we were able to
11121 * clear the whole block group range from freed_extents[]. This
11122 * means that task can lookup for the block group after we
11123 * unpinned it from freed_extents[] and removed it, leading to
11124 * a BUG_ON() at btrfs_unpin_extent_range().
11126 mutex_lock(&fs_info->unused_bg_unpin_mutex);
11127 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
11130 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
11131 btrfs_dec_block_group_ro(block_group);
11134 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
11137 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
11138 btrfs_dec_block_group_ro(block_group);
11141 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
11143 /* Reset pinned so btrfs_put_block_group doesn't complain */
11144 spin_lock(&space_info->lock);
11145 spin_lock(&block_group->lock);
11147 update_bytes_pinned(space_info, -block_group->pinned);
11148 space_info->bytes_readonly += block_group->pinned;
11149 percpu_counter_add_batch(&space_info->total_bytes_pinned,
11150 -block_group->pinned,
11151 BTRFS_TOTAL_BYTES_PINNED_BATCH);
11152 block_group->pinned = 0;
11154 spin_unlock(&block_group->lock);
11155 spin_unlock(&space_info->lock);
11157 /* DISCARD can flip during remount */
11158 trimming = btrfs_test_opt(fs_info, DISCARD);
11160 /* Implicit trim during transaction commit. */
11162 btrfs_get_block_group_trimming(block_group);
11165 * Btrfs_remove_chunk will abort the transaction if things go
11168 ret = btrfs_remove_chunk(trans, block_group->key.objectid);
11172 btrfs_put_block_group_trimming(block_group);
11177 * If we're not mounted with -odiscard, we can just forget
11178 * about this block group. Otherwise we'll need to wait
11179 * until transaction commit to do the actual discard.
11182 spin_lock(&fs_info->unused_bgs_lock);
11184 * A concurrent scrub might have added us to the list
11185 * fs_info->unused_bgs, so use a list_move operation
11186 * to add the block group to the deleted_bgs list.
11188 list_move(&block_group->bg_list,
11189 &trans->transaction->deleted_bgs);
11190 spin_unlock(&fs_info->unused_bgs_lock);
11191 btrfs_get_block_group(block_group);
11194 btrfs_end_transaction(trans);
11196 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
11197 btrfs_put_block_group(block_group);
11198 spin_lock(&fs_info->unused_bgs_lock);
11200 spin_unlock(&fs_info->unused_bgs_lock);
11203 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
11205 struct btrfs_super_block *disk_super;
11211 disk_super = fs_info->super_copy;
11212 if (!btrfs_super_root(disk_super))
11215 features = btrfs_super_incompat_flags(disk_super);
11216 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
11219 flags = BTRFS_BLOCK_GROUP_SYSTEM;
11220 ret = create_space_info(fs_info, flags);
11225 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
11226 ret = create_space_info(fs_info, flags);
11228 flags = BTRFS_BLOCK_GROUP_METADATA;
11229 ret = create_space_info(fs_info, flags);
11233 flags = BTRFS_BLOCK_GROUP_DATA;
11234 ret = create_space_info(fs_info, flags);
11240 int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
11241 u64 start, u64 end)
11243 return unpin_extent_range(fs_info, start, end, false);
11247 * It used to be that old block groups would be left around forever.
11248 * Iterating over them would be enough to trim unused space. Since we
11249 * now automatically remove them, we also need to iterate over unallocated
11252 * We don't want a transaction for this since the discard may take a
11253 * substantial amount of time. We don't require that a transaction be
11254 * running, but we do need to take a running transaction into account
11255 * to ensure that we're not discarding chunks that were released or
11256 * allocated in the current transaction.
11258 * Holding the chunks lock will prevent other threads from allocating
11259 * or releasing chunks, but it won't prevent a running transaction
11260 * from committing and releasing the memory that the pending chunks
11261 * list head uses. For that, we need to take a reference to the
11262 * transaction and hold the commit root sem. We only need to hold
11263 * it while performing the free space search since we have already
11264 * held back allocations.
11266 static int btrfs_trim_free_extents(struct btrfs_device *device,
11267 struct fstrim_range *range, u64 *trimmed)
11269 u64 start, len = 0, end = 0;
11272 start = max_t(u64, range->start, SZ_1M);
11275 /* Discard not supported = nothing to do. */
11276 if (!blk_queue_discard(bdev_get_queue(device->bdev)))
11279 /* Not writable = nothing to do. */
11280 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
11283 /* No free space = nothing to do. */
11284 if (device->total_bytes <= device->bytes_used)
11290 struct btrfs_fs_info *fs_info = device->fs_info;
11293 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
11297 find_first_clear_extent_bit(&device->alloc_state, start,
11299 CHUNK_TRIMMED | CHUNK_ALLOCATED);
11301 * If find_first_clear_extent_bit find a range that spans the
11302 * end of the device it will set end to -1, in this case it's up
11303 * to the caller to trim the value to the size of the device.
11305 end = min(end, device->total_bytes - 1);
11306 len = end - start + 1;
11308 /* We didn't find any extents */
11310 mutex_unlock(&fs_info->chunk_mutex);
11315 /* Keep going until we satisfy minlen or reach end of space */
11316 if (len < range->minlen) {
11317 mutex_unlock(&fs_info->chunk_mutex);
11322 /* If we are out of the passed range break */
11323 if (start > range->start + range->len - 1) {
11324 mutex_unlock(&fs_info->chunk_mutex);
11328 start = max(range->start, start);
11329 len = min(range->len, len);
11331 ret = btrfs_issue_discard(device->bdev, start, len,
11334 set_extent_bits(&device->alloc_state, start,
11337 mutex_unlock(&fs_info->chunk_mutex);
11345 /* We've trimmed enough */
11346 if (*trimmed >= range->len)
11349 if (fatal_signal_pending(current)) {
11350 ret = -ERESTARTSYS;
11361 * Trim the whole filesystem by:
11362 * 1) trimming the free space in each block group
11363 * 2) trimming the unallocated space on each device
11365 * This will also continue trimming even if a block group or device encounters
11366 * an error. The return value will be the last error, or 0 if nothing bad
11369 int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
11371 struct btrfs_block_group_cache *cache = NULL;
11372 struct btrfs_device *device;
11373 struct list_head *devices;
11379 u64 dev_failed = 0;
11384 cache = btrfs_lookup_first_block_group(fs_info, range->start);
11385 for (; cache; cache = next_block_group(fs_info, cache)) {
11386 if (cache->key.objectid >= (range->start + range->len)) {
11387 btrfs_put_block_group(cache);
11391 start = max(range->start, cache->key.objectid);
11392 end = min(range->start + range->len,
11393 cache->key.objectid + cache->key.offset);
11395 if (end - start >= range->minlen) {
11396 if (!block_group_cache_done(cache)) {
11397 ret = cache_block_group(cache, 0);
11403 ret = wait_block_group_cache_done(cache);
11410 ret = btrfs_trim_block_group(cache,
11416 trimmed += group_trimmed;
11426 btrfs_warn(fs_info,
11427 "failed to trim %llu block group(s), last error %d",
11428 bg_failed, bg_ret);
11429 mutex_lock(&fs_info->fs_devices->device_list_mutex);
11430 devices = &fs_info->fs_devices->devices;
11431 list_for_each_entry(device, devices, dev_list) {
11432 ret = btrfs_trim_free_extents(device, range, &group_trimmed);
11439 trimmed += group_trimmed;
11441 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
11444 btrfs_warn(fs_info,
11445 "failed to trim %llu device(s), last error %d",
11446 dev_failed, dev_ret);
11447 range->len = trimmed;
11454 * btrfs_{start,end}_write_no_snapshotting() are similar to
11455 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
11456 * data into the page cache through nocow before the subvolume is snapshoted,
11457 * but flush the data into disk after the snapshot creation, or to prevent
11458 * operations while snapshotting is ongoing and that cause the snapshot to be
11459 * inconsistent (writes followed by expanding truncates for example).
11461 void btrfs_end_write_no_snapshotting(struct btrfs_root *root)
11463 percpu_counter_dec(&root->subv_writers->counter);
11464 cond_wake_up(&root->subv_writers->wait);
11467 int btrfs_start_write_no_snapshotting(struct btrfs_root *root)
11469 if (atomic_read(&root->will_be_snapshotted))
11472 percpu_counter_inc(&root->subv_writers->counter);
11474 * Make sure counter is updated before we check for snapshot creation.
11477 if (atomic_read(&root->will_be_snapshotted)) {
11478 btrfs_end_write_no_snapshotting(root);
11484 void btrfs_wait_for_snapshot_creation(struct btrfs_root *root)
11489 ret = btrfs_start_write_no_snapshotting(root);
11492 wait_var_event(&root->will_be_snapshotted,
11493 !atomic_read(&root->will_be_snapshotted));
11497 void btrfs_mark_bg_unused(struct btrfs_block_group_cache *bg)
11499 struct btrfs_fs_info *fs_info = bg->fs_info;
11501 spin_lock(&fs_info->unused_bgs_lock);
11502 if (list_empty(&bg->bg_list)) {
11503 btrfs_get_block_group(bg);
11504 trace_btrfs_add_unused_block_group(bg);
11505 list_add_tail(&bg->bg_list, &fs_info->unused_bgs);
11507 spin_unlock(&fs_info->unused_bgs_lock);