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(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 u64 generic_ref_to_space_flags(struct btrfs_ref *ref)
761 if (ref->type == BTRFS_REF_METADATA) {
762 if (ref->tree_ref.root == BTRFS_CHUNK_TREE_OBJECTID)
763 return BTRFS_BLOCK_GROUP_SYSTEM;
765 return BTRFS_BLOCK_GROUP_METADATA;
767 return BTRFS_BLOCK_GROUP_DATA;
770 static void add_pinned_bytes(struct btrfs_fs_info *fs_info,
771 struct btrfs_ref *ref)
773 struct btrfs_space_info *space_info;
774 u64 flags = generic_ref_to_space_flags(ref);
776 space_info = __find_space_info(fs_info, flags);
778 percpu_counter_add_batch(&space_info->total_bytes_pinned, ref->len,
779 BTRFS_TOTAL_BYTES_PINNED_BATCH);
782 static void sub_pinned_bytes(struct btrfs_fs_info *fs_info,
783 struct btrfs_ref *ref)
785 struct btrfs_space_info *space_info;
786 u64 flags = generic_ref_to_space_flags(ref);
788 space_info = __find_space_info(fs_info, flags);
790 percpu_counter_add_batch(&space_info->total_bytes_pinned, -ref->len,
791 BTRFS_TOTAL_BYTES_PINNED_BATCH);
795 * after adding space to the filesystem, we need to clear the full flags
796 * on all the space infos.
798 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
800 struct list_head *head = &info->space_info;
801 struct btrfs_space_info *found;
804 list_for_each_entry_rcu(found, head, list)
809 /* simple helper to search for an existing data extent at a given offset */
810 int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len)
813 struct btrfs_key key;
814 struct btrfs_path *path;
816 path = btrfs_alloc_path();
820 key.objectid = start;
822 key.type = BTRFS_EXTENT_ITEM_KEY;
823 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
824 btrfs_free_path(path);
829 * helper function to lookup reference count and flags of a tree block.
831 * the head node for delayed ref is used to store the sum of all the
832 * reference count modifications queued up in the rbtree. the head
833 * node may also store the extent flags to set. This way you can check
834 * to see what the reference count and extent flags would be if all of
835 * the delayed refs are not processed.
837 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
838 struct btrfs_fs_info *fs_info, u64 bytenr,
839 u64 offset, int metadata, u64 *refs, u64 *flags)
841 struct btrfs_delayed_ref_head *head;
842 struct btrfs_delayed_ref_root *delayed_refs;
843 struct btrfs_path *path;
844 struct btrfs_extent_item *ei;
845 struct extent_buffer *leaf;
846 struct btrfs_key key;
853 * If we don't have skinny metadata, don't bother doing anything
856 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) {
857 offset = fs_info->nodesize;
861 path = btrfs_alloc_path();
866 path->skip_locking = 1;
867 path->search_commit_root = 1;
871 key.objectid = bytenr;
874 key.type = BTRFS_METADATA_ITEM_KEY;
876 key.type = BTRFS_EXTENT_ITEM_KEY;
878 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
882 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
883 if (path->slots[0]) {
885 btrfs_item_key_to_cpu(path->nodes[0], &key,
887 if (key.objectid == bytenr &&
888 key.type == BTRFS_EXTENT_ITEM_KEY &&
889 key.offset == fs_info->nodesize)
895 leaf = path->nodes[0];
896 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
897 if (item_size >= sizeof(*ei)) {
898 ei = btrfs_item_ptr(leaf, path->slots[0],
899 struct btrfs_extent_item);
900 num_refs = btrfs_extent_refs(leaf, ei);
901 extent_flags = btrfs_extent_flags(leaf, ei);
904 btrfs_print_v0_err(fs_info);
906 btrfs_abort_transaction(trans, ret);
908 btrfs_handle_fs_error(fs_info, ret, NULL);
913 BUG_ON(num_refs == 0);
923 delayed_refs = &trans->transaction->delayed_refs;
924 spin_lock(&delayed_refs->lock);
925 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
927 if (!mutex_trylock(&head->mutex)) {
928 refcount_inc(&head->refs);
929 spin_unlock(&delayed_refs->lock);
931 btrfs_release_path(path);
934 * Mutex was contended, block until it's released and try
937 mutex_lock(&head->mutex);
938 mutex_unlock(&head->mutex);
939 btrfs_put_delayed_ref_head(head);
942 spin_lock(&head->lock);
943 if (head->extent_op && head->extent_op->update_flags)
944 extent_flags |= head->extent_op->flags_to_set;
946 BUG_ON(num_refs == 0);
948 num_refs += head->ref_mod;
949 spin_unlock(&head->lock);
950 mutex_unlock(&head->mutex);
952 spin_unlock(&delayed_refs->lock);
954 WARN_ON(num_refs == 0);
958 *flags = extent_flags;
960 btrfs_free_path(path);
965 * Back reference rules. Back refs have three main goals:
967 * 1) differentiate between all holders of references to an extent so that
968 * when a reference is dropped we can make sure it was a valid reference
969 * before freeing the extent.
971 * 2) Provide enough information to quickly find the holders of an extent
972 * if we notice a given block is corrupted or bad.
974 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
975 * maintenance. This is actually the same as #2, but with a slightly
976 * different use case.
978 * There are two kinds of back refs. The implicit back refs is optimized
979 * for pointers in non-shared tree blocks. For a given pointer in a block,
980 * back refs of this kind provide information about the block's owner tree
981 * and the pointer's key. These information allow us to find the block by
982 * b-tree searching. The full back refs is for pointers in tree blocks not
983 * referenced by their owner trees. The location of tree block is recorded
984 * in the back refs. Actually the full back refs is generic, and can be
985 * used in all cases the implicit back refs is used. The major shortcoming
986 * of the full back refs is its overhead. Every time a tree block gets
987 * COWed, we have to update back refs entry for all pointers in it.
989 * For a newly allocated tree block, we use implicit back refs for
990 * pointers in it. This means most tree related operations only involve
991 * implicit back refs. For a tree block created in old transaction, the
992 * only way to drop a reference to it is COW it. So we can detect the
993 * event that tree block loses its owner tree's reference and do the
994 * back refs conversion.
996 * When a tree block is COWed through a tree, there are four cases:
998 * The reference count of the block is one and the tree is the block's
999 * owner tree. Nothing to do in this case.
1001 * The reference count of the block is one and the tree is not the
1002 * block's owner tree. In this case, full back refs is used for pointers
1003 * in the block. Remove these full back refs, add implicit back refs for
1004 * every pointers in the new block.
1006 * The reference count of the block is greater than one and the tree is
1007 * the block's owner tree. In this case, implicit back refs is used for
1008 * pointers in the block. Add full back refs for every pointers in the
1009 * block, increase lower level extents' reference counts. The original
1010 * implicit back refs are entailed to the new block.
1012 * The reference count of the block is greater than one and the tree is
1013 * not the block's owner tree. Add implicit back refs for every pointer in
1014 * the new block, increase lower level extents' reference count.
1016 * Back Reference Key composing:
1018 * The key objectid corresponds to the first byte in the extent,
1019 * The key type is used to differentiate between types of back refs.
1020 * There are different meanings of the key offset for different types
1023 * File extents can be referenced by:
1025 * - multiple snapshots, subvolumes, or different generations in one subvol
1026 * - different files inside a single subvolume
1027 * - different offsets inside a file (bookend extents in file.c)
1029 * The extent ref structure for the implicit back refs has fields for:
1031 * - Objectid of the subvolume root
1032 * - objectid of the file holding the reference
1033 * - original offset in the file
1034 * - how many bookend extents
1036 * The key offset for the implicit back refs is hash of the first
1039 * The extent ref structure for the full back refs has field for:
1041 * - number of pointers in the tree leaf
1043 * The key offset for the implicit back refs is the first byte of
1046 * When a file extent is allocated, The implicit back refs is used.
1047 * the fields are filled in:
1049 * (root_key.objectid, inode objectid, offset in file, 1)
1051 * When a file extent is removed file truncation, we find the
1052 * corresponding implicit back refs and check the following fields:
1054 * (btrfs_header_owner(leaf), inode objectid, offset in file)
1056 * Btree extents can be referenced by:
1058 * - Different subvolumes
1060 * Both the implicit back refs and the full back refs for tree blocks
1061 * only consist of key. The key offset for the implicit back refs is
1062 * objectid of block's owner tree. The key offset for the full back refs
1063 * is the first byte of parent block.
1065 * When implicit back refs is used, information about the lowest key and
1066 * level of the tree block are required. These information are stored in
1067 * tree block info structure.
1071 * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required,
1072 * is_data == BTRFS_REF_TYPE_DATA, data type is requiried,
1073 * is_data == BTRFS_REF_TYPE_ANY, either type is OK.
1075 int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb,
1076 struct btrfs_extent_inline_ref *iref,
1077 enum btrfs_inline_ref_type is_data)
1079 int type = btrfs_extent_inline_ref_type(eb, iref);
1080 u64 offset = btrfs_extent_inline_ref_offset(eb, iref);
1082 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1083 type == BTRFS_SHARED_BLOCK_REF_KEY ||
1084 type == BTRFS_SHARED_DATA_REF_KEY ||
1085 type == BTRFS_EXTENT_DATA_REF_KEY) {
1086 if (is_data == BTRFS_REF_TYPE_BLOCK) {
1087 if (type == BTRFS_TREE_BLOCK_REF_KEY)
1089 if (type == BTRFS_SHARED_BLOCK_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))
1100 } else if (is_data == BTRFS_REF_TYPE_DATA) {
1101 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1103 if (type == BTRFS_SHARED_DATA_REF_KEY) {
1104 ASSERT(eb->fs_info);
1106 * Every shared one has parent tree
1107 * block, which must be aligned to
1111 IS_ALIGNED(offset, eb->fs_info->nodesize))
1115 ASSERT(is_data == BTRFS_REF_TYPE_ANY);
1120 btrfs_print_leaf((struct extent_buffer *)eb);
1121 btrfs_err(eb->fs_info, "eb %llu invalid extent inline ref type %d",
1125 return BTRFS_REF_TYPE_INVALID;
1128 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1130 u32 high_crc = ~(u32)0;
1131 u32 low_crc = ~(u32)0;
1134 lenum = cpu_to_le64(root_objectid);
1135 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1136 lenum = cpu_to_le64(owner);
1137 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1138 lenum = cpu_to_le64(offset);
1139 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1141 return ((u64)high_crc << 31) ^ (u64)low_crc;
1144 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1145 struct btrfs_extent_data_ref *ref)
1147 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1148 btrfs_extent_data_ref_objectid(leaf, ref),
1149 btrfs_extent_data_ref_offset(leaf, ref));
1152 static int match_extent_data_ref(struct extent_buffer *leaf,
1153 struct btrfs_extent_data_ref *ref,
1154 u64 root_objectid, u64 owner, u64 offset)
1156 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1157 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1158 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1163 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1164 struct btrfs_path *path,
1165 u64 bytenr, u64 parent,
1167 u64 owner, u64 offset)
1169 struct btrfs_root *root = trans->fs_info->extent_root;
1170 struct btrfs_key key;
1171 struct btrfs_extent_data_ref *ref;
1172 struct extent_buffer *leaf;
1178 key.objectid = bytenr;
1180 key.type = BTRFS_SHARED_DATA_REF_KEY;
1181 key.offset = parent;
1183 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1184 key.offset = hash_extent_data_ref(root_objectid,
1189 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1201 leaf = path->nodes[0];
1202 nritems = btrfs_header_nritems(leaf);
1204 if (path->slots[0] >= nritems) {
1205 ret = btrfs_next_leaf(root, path);
1211 leaf = path->nodes[0];
1212 nritems = btrfs_header_nritems(leaf);
1216 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1217 if (key.objectid != bytenr ||
1218 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1221 ref = btrfs_item_ptr(leaf, path->slots[0],
1222 struct btrfs_extent_data_ref);
1224 if (match_extent_data_ref(leaf, ref, root_objectid,
1227 btrfs_release_path(path);
1239 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1240 struct btrfs_path *path,
1241 u64 bytenr, u64 parent,
1242 u64 root_objectid, u64 owner,
1243 u64 offset, int refs_to_add)
1245 struct btrfs_root *root = trans->fs_info->extent_root;
1246 struct btrfs_key key;
1247 struct extent_buffer *leaf;
1252 key.objectid = bytenr;
1254 key.type = BTRFS_SHARED_DATA_REF_KEY;
1255 key.offset = parent;
1256 size = sizeof(struct btrfs_shared_data_ref);
1258 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1259 key.offset = hash_extent_data_ref(root_objectid,
1261 size = sizeof(struct btrfs_extent_data_ref);
1264 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1265 if (ret && ret != -EEXIST)
1268 leaf = path->nodes[0];
1270 struct btrfs_shared_data_ref *ref;
1271 ref = btrfs_item_ptr(leaf, path->slots[0],
1272 struct btrfs_shared_data_ref);
1274 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1276 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1277 num_refs += refs_to_add;
1278 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1281 struct btrfs_extent_data_ref *ref;
1282 while (ret == -EEXIST) {
1283 ref = btrfs_item_ptr(leaf, path->slots[0],
1284 struct btrfs_extent_data_ref);
1285 if (match_extent_data_ref(leaf, ref, root_objectid,
1288 btrfs_release_path(path);
1290 ret = btrfs_insert_empty_item(trans, root, path, &key,
1292 if (ret && ret != -EEXIST)
1295 leaf = path->nodes[0];
1297 ref = btrfs_item_ptr(leaf, path->slots[0],
1298 struct btrfs_extent_data_ref);
1300 btrfs_set_extent_data_ref_root(leaf, ref,
1302 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1303 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1304 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1306 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1307 num_refs += refs_to_add;
1308 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1311 btrfs_mark_buffer_dirty(leaf);
1314 btrfs_release_path(path);
1318 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1319 struct btrfs_path *path,
1320 int refs_to_drop, int *last_ref)
1322 struct btrfs_key key;
1323 struct btrfs_extent_data_ref *ref1 = NULL;
1324 struct btrfs_shared_data_ref *ref2 = NULL;
1325 struct extent_buffer *leaf;
1329 leaf = path->nodes[0];
1330 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1332 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1333 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1334 struct btrfs_extent_data_ref);
1335 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1336 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1337 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1338 struct btrfs_shared_data_ref);
1339 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1340 } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) {
1341 btrfs_print_v0_err(trans->fs_info);
1342 btrfs_abort_transaction(trans, -EINVAL);
1348 BUG_ON(num_refs < refs_to_drop);
1349 num_refs -= refs_to_drop;
1351 if (num_refs == 0) {
1352 ret = btrfs_del_item(trans, trans->fs_info->extent_root, path);
1355 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1356 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1357 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1358 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1359 btrfs_mark_buffer_dirty(leaf);
1364 static noinline u32 extent_data_ref_count(struct btrfs_path *path,
1365 struct btrfs_extent_inline_ref *iref)
1367 struct btrfs_key key;
1368 struct extent_buffer *leaf;
1369 struct btrfs_extent_data_ref *ref1;
1370 struct btrfs_shared_data_ref *ref2;
1374 leaf = path->nodes[0];
1375 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1377 BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY);
1380 * If type is invalid, we should have bailed out earlier than
1383 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
1384 ASSERT(type != BTRFS_REF_TYPE_INVALID);
1385 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1386 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1387 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1389 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1390 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1392 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1393 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1394 struct btrfs_extent_data_ref);
1395 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1396 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1397 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1398 struct btrfs_shared_data_ref);
1399 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1406 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1407 struct btrfs_path *path,
1408 u64 bytenr, u64 parent,
1411 struct btrfs_root *root = trans->fs_info->extent_root;
1412 struct btrfs_key key;
1415 key.objectid = bytenr;
1417 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1418 key.offset = parent;
1420 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1421 key.offset = root_objectid;
1424 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1430 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1431 struct btrfs_path *path,
1432 u64 bytenr, u64 parent,
1435 struct btrfs_key key;
1438 key.objectid = bytenr;
1440 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1441 key.offset = parent;
1443 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1444 key.offset = root_objectid;
1447 ret = btrfs_insert_empty_item(trans, trans->fs_info->extent_root,
1449 btrfs_release_path(path);
1453 static inline int extent_ref_type(u64 parent, u64 owner)
1456 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1458 type = BTRFS_SHARED_BLOCK_REF_KEY;
1460 type = BTRFS_TREE_BLOCK_REF_KEY;
1463 type = BTRFS_SHARED_DATA_REF_KEY;
1465 type = BTRFS_EXTENT_DATA_REF_KEY;
1470 static int find_next_key(struct btrfs_path *path, int level,
1471 struct btrfs_key *key)
1474 for (; level < BTRFS_MAX_LEVEL; level++) {
1475 if (!path->nodes[level])
1477 if (path->slots[level] + 1 >=
1478 btrfs_header_nritems(path->nodes[level]))
1481 btrfs_item_key_to_cpu(path->nodes[level], key,
1482 path->slots[level] + 1);
1484 btrfs_node_key_to_cpu(path->nodes[level], key,
1485 path->slots[level] + 1);
1492 * look for inline back ref. if back ref is found, *ref_ret is set
1493 * to the address of inline back ref, and 0 is returned.
1495 * if back ref isn't found, *ref_ret is set to the address where it
1496 * should be inserted, and -ENOENT is returned.
1498 * if insert is true and there are too many inline back refs, the path
1499 * points to the extent item, and -EAGAIN is returned.
1501 * NOTE: inline back refs are ordered in the same way that back ref
1502 * items in the tree are ordered.
1504 static noinline_for_stack
1505 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1506 struct btrfs_path *path,
1507 struct btrfs_extent_inline_ref **ref_ret,
1508 u64 bytenr, u64 num_bytes,
1509 u64 parent, u64 root_objectid,
1510 u64 owner, u64 offset, int insert)
1512 struct btrfs_fs_info *fs_info = trans->fs_info;
1513 struct btrfs_root *root = fs_info->extent_root;
1514 struct btrfs_key key;
1515 struct extent_buffer *leaf;
1516 struct btrfs_extent_item *ei;
1517 struct btrfs_extent_inline_ref *iref;
1527 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
1530 key.objectid = bytenr;
1531 key.type = BTRFS_EXTENT_ITEM_KEY;
1532 key.offset = num_bytes;
1534 want = extent_ref_type(parent, owner);
1536 extra_size = btrfs_extent_inline_ref_size(want);
1537 path->keep_locks = 1;
1542 * Owner is our level, so we can just add one to get the level for the
1543 * block we are interested in.
1545 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1546 key.type = BTRFS_METADATA_ITEM_KEY;
1551 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1558 * We may be a newly converted file system which still has the old fat
1559 * extent entries for metadata, so try and see if we have one of those.
1561 if (ret > 0 && skinny_metadata) {
1562 skinny_metadata = false;
1563 if (path->slots[0]) {
1565 btrfs_item_key_to_cpu(path->nodes[0], &key,
1567 if (key.objectid == bytenr &&
1568 key.type == BTRFS_EXTENT_ITEM_KEY &&
1569 key.offset == num_bytes)
1573 key.objectid = bytenr;
1574 key.type = BTRFS_EXTENT_ITEM_KEY;
1575 key.offset = num_bytes;
1576 btrfs_release_path(path);
1581 if (ret && !insert) {
1584 } else if (WARN_ON(ret)) {
1589 leaf = path->nodes[0];
1590 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1591 if (unlikely(item_size < sizeof(*ei))) {
1593 btrfs_print_v0_err(fs_info);
1594 btrfs_abort_transaction(trans, err);
1598 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1599 flags = btrfs_extent_flags(leaf, ei);
1601 ptr = (unsigned long)(ei + 1);
1602 end = (unsigned long)ei + item_size;
1604 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1605 ptr += sizeof(struct btrfs_tree_block_info);
1609 if (owner >= BTRFS_FIRST_FREE_OBJECTID)
1610 needed = BTRFS_REF_TYPE_DATA;
1612 needed = BTRFS_REF_TYPE_BLOCK;
1620 iref = (struct btrfs_extent_inline_ref *)ptr;
1621 type = btrfs_get_extent_inline_ref_type(leaf, iref, needed);
1622 if (type == BTRFS_REF_TYPE_INVALID) {
1630 ptr += btrfs_extent_inline_ref_size(type);
1634 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1635 struct btrfs_extent_data_ref *dref;
1636 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1637 if (match_extent_data_ref(leaf, dref, root_objectid,
1642 if (hash_extent_data_ref_item(leaf, dref) <
1643 hash_extent_data_ref(root_objectid, owner, offset))
1647 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1649 if (parent == ref_offset) {
1653 if (ref_offset < parent)
1656 if (root_objectid == ref_offset) {
1660 if (ref_offset < root_objectid)
1664 ptr += btrfs_extent_inline_ref_size(type);
1666 if (err == -ENOENT && insert) {
1667 if (item_size + extra_size >=
1668 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1673 * To add new inline back ref, we have to make sure
1674 * there is no corresponding back ref item.
1675 * For simplicity, we just do not add new inline back
1676 * ref if there is any kind of item for this block
1678 if (find_next_key(path, 0, &key) == 0 &&
1679 key.objectid == bytenr &&
1680 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1685 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1688 path->keep_locks = 0;
1689 btrfs_unlock_up_safe(path, 1);
1695 * helper to add new inline back ref
1697 static noinline_for_stack
1698 void setup_inline_extent_backref(struct btrfs_fs_info *fs_info,
1699 struct btrfs_path *path,
1700 struct btrfs_extent_inline_ref *iref,
1701 u64 parent, u64 root_objectid,
1702 u64 owner, u64 offset, int refs_to_add,
1703 struct btrfs_delayed_extent_op *extent_op)
1705 struct extent_buffer *leaf;
1706 struct btrfs_extent_item *ei;
1709 unsigned long item_offset;
1714 leaf = path->nodes[0];
1715 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1716 item_offset = (unsigned long)iref - (unsigned long)ei;
1718 type = extent_ref_type(parent, owner);
1719 size = btrfs_extent_inline_ref_size(type);
1721 btrfs_extend_item(path, size);
1723 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1724 refs = btrfs_extent_refs(leaf, ei);
1725 refs += refs_to_add;
1726 btrfs_set_extent_refs(leaf, ei, refs);
1728 __run_delayed_extent_op(extent_op, leaf, ei);
1730 ptr = (unsigned long)ei + item_offset;
1731 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1732 if (ptr < end - size)
1733 memmove_extent_buffer(leaf, ptr + size, ptr,
1736 iref = (struct btrfs_extent_inline_ref *)ptr;
1737 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1738 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1739 struct btrfs_extent_data_ref *dref;
1740 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1741 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1742 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1743 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1744 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1745 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1746 struct btrfs_shared_data_ref *sref;
1747 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1748 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1749 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1750 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1751 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1753 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1755 btrfs_mark_buffer_dirty(leaf);
1758 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1759 struct btrfs_path *path,
1760 struct btrfs_extent_inline_ref **ref_ret,
1761 u64 bytenr, u64 num_bytes, u64 parent,
1762 u64 root_objectid, u64 owner, u64 offset)
1766 ret = lookup_inline_extent_backref(trans, path, ref_ret, bytenr,
1767 num_bytes, parent, root_objectid,
1772 btrfs_release_path(path);
1775 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1776 ret = lookup_tree_block_ref(trans, path, bytenr, parent,
1779 ret = lookup_extent_data_ref(trans, path, bytenr, parent,
1780 root_objectid, owner, offset);
1786 * helper to update/remove inline back ref
1788 static noinline_for_stack
1789 void update_inline_extent_backref(struct btrfs_path *path,
1790 struct btrfs_extent_inline_ref *iref,
1792 struct btrfs_delayed_extent_op *extent_op,
1795 struct extent_buffer *leaf = path->nodes[0];
1796 struct btrfs_extent_item *ei;
1797 struct btrfs_extent_data_ref *dref = NULL;
1798 struct btrfs_shared_data_ref *sref = NULL;
1806 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1807 refs = btrfs_extent_refs(leaf, ei);
1808 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1809 refs += refs_to_mod;
1810 btrfs_set_extent_refs(leaf, ei, refs);
1812 __run_delayed_extent_op(extent_op, leaf, ei);
1815 * If type is invalid, we should have bailed out after
1816 * lookup_inline_extent_backref().
1818 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY);
1819 ASSERT(type != BTRFS_REF_TYPE_INVALID);
1821 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1822 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1823 refs = btrfs_extent_data_ref_count(leaf, dref);
1824 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1825 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1826 refs = btrfs_shared_data_ref_count(leaf, sref);
1829 BUG_ON(refs_to_mod != -1);
1832 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1833 refs += refs_to_mod;
1836 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1837 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1839 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1842 size = btrfs_extent_inline_ref_size(type);
1843 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1844 ptr = (unsigned long)iref;
1845 end = (unsigned long)ei + item_size;
1846 if (ptr + size < end)
1847 memmove_extent_buffer(leaf, ptr, ptr + size,
1850 btrfs_truncate_item(path, item_size, 1);
1852 btrfs_mark_buffer_dirty(leaf);
1855 static noinline_for_stack
1856 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1857 struct btrfs_path *path,
1858 u64 bytenr, u64 num_bytes, u64 parent,
1859 u64 root_objectid, u64 owner,
1860 u64 offset, int refs_to_add,
1861 struct btrfs_delayed_extent_op *extent_op)
1863 struct btrfs_extent_inline_ref *iref;
1866 ret = lookup_inline_extent_backref(trans, path, &iref, bytenr,
1867 num_bytes, parent, root_objectid,
1870 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1871 update_inline_extent_backref(path, iref, refs_to_add,
1873 } else if (ret == -ENOENT) {
1874 setup_inline_extent_backref(trans->fs_info, path, iref, parent,
1875 root_objectid, owner, offset,
1876 refs_to_add, extent_op);
1882 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1883 struct btrfs_path *path,
1884 u64 bytenr, u64 parent, u64 root_objectid,
1885 u64 owner, u64 offset, int refs_to_add)
1888 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1889 BUG_ON(refs_to_add != 1);
1890 ret = insert_tree_block_ref(trans, path, bytenr, parent,
1893 ret = insert_extent_data_ref(trans, path, bytenr, parent,
1894 root_objectid, owner, offset,
1900 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1901 struct btrfs_path *path,
1902 struct btrfs_extent_inline_ref *iref,
1903 int refs_to_drop, int is_data, int *last_ref)
1907 BUG_ON(!is_data && refs_to_drop != 1);
1909 update_inline_extent_backref(path, iref, -refs_to_drop, NULL,
1911 } else if (is_data) {
1912 ret = remove_extent_data_ref(trans, path, refs_to_drop,
1916 ret = btrfs_del_item(trans, trans->fs_info->extent_root, path);
1921 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
1922 u64 *discarded_bytes)
1925 u64 bytes_left, end;
1926 u64 aligned_start = ALIGN(start, 1 << 9);
1928 if (WARN_ON(start != aligned_start)) {
1929 len -= aligned_start - start;
1930 len = round_down(len, 1 << 9);
1931 start = aligned_start;
1934 *discarded_bytes = 0;
1942 /* Skip any superblocks on this device. */
1943 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
1944 u64 sb_start = btrfs_sb_offset(j);
1945 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
1946 u64 size = sb_start - start;
1948 if (!in_range(sb_start, start, bytes_left) &&
1949 !in_range(sb_end, start, bytes_left) &&
1950 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
1954 * Superblock spans beginning of range. Adjust start and
1957 if (sb_start <= start) {
1958 start += sb_end - start;
1963 bytes_left = end - start;
1968 ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
1971 *discarded_bytes += size;
1972 else if (ret != -EOPNOTSUPP)
1981 bytes_left = end - start;
1985 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
1988 *discarded_bytes += bytes_left;
1993 int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr,
1994 u64 num_bytes, u64 *actual_bytes)
1997 u64 discarded_bytes = 0;
1998 struct btrfs_bio *bbio = NULL;
2002 * Avoid races with device replace and make sure our bbio has devices
2003 * associated to its stripes that don't go away while we are discarding.
2005 btrfs_bio_counter_inc_blocked(fs_info);
2006 /* Tell the block device(s) that the sectors can be discarded */
2007 ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, bytenr, &num_bytes,
2009 /* Error condition is -ENOMEM */
2011 struct btrfs_bio_stripe *stripe = bbio->stripes;
2015 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
2017 struct request_queue *req_q;
2019 if (!stripe->dev->bdev) {
2020 ASSERT(btrfs_test_opt(fs_info, DEGRADED));
2023 req_q = bdev_get_queue(stripe->dev->bdev);
2024 if (!blk_queue_discard(req_q))
2027 ret = btrfs_issue_discard(stripe->dev->bdev,
2032 discarded_bytes += bytes;
2033 else if (ret != -EOPNOTSUPP)
2034 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2037 * Just in case we get back EOPNOTSUPP for some reason,
2038 * just ignore the return value so we don't screw up
2039 * people calling discard_extent.
2043 btrfs_put_bbio(bbio);
2045 btrfs_bio_counter_dec(fs_info);
2048 *actual_bytes = discarded_bytes;
2051 if (ret == -EOPNOTSUPP)
2056 /* Can return -ENOMEM */
2057 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2058 struct btrfs_ref *generic_ref)
2060 struct btrfs_fs_info *fs_info = trans->fs_info;
2061 int old_ref_mod, new_ref_mod;
2064 ASSERT(generic_ref->type != BTRFS_REF_NOT_SET &&
2065 generic_ref->action);
2066 BUG_ON(generic_ref->type == BTRFS_REF_METADATA &&
2067 generic_ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID);
2069 if (generic_ref->type == BTRFS_REF_METADATA)
2070 ret = btrfs_add_delayed_tree_ref(trans, generic_ref,
2071 NULL, &old_ref_mod, &new_ref_mod);
2073 ret = btrfs_add_delayed_data_ref(trans, generic_ref, 0,
2074 &old_ref_mod, &new_ref_mod);
2076 btrfs_ref_tree_mod(fs_info, generic_ref);
2078 if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0)
2079 sub_pinned_bytes(fs_info, generic_ref);
2085 * __btrfs_inc_extent_ref - insert backreference for a given extent
2087 * @trans: Handle of transaction
2089 * @node: The delayed ref node used to get the bytenr/length for
2090 * extent whose references are incremented.
2092 * @parent: If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/
2093 * BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical
2094 * bytenr of the parent block. Since new extents are always
2095 * created with indirect references, this will only be the case
2096 * when relocating a shared extent. In that case, root_objectid
2097 * will be BTRFS_TREE_RELOC_OBJECTID. Otheriwse, parent must
2100 * @root_objectid: The id of the root where this modification has originated,
2101 * this can be either one of the well-known metadata trees or
2102 * the subvolume id which references this extent.
2104 * @owner: For data extents it is the inode number of the owning file.
2105 * For metadata extents this parameter holds the level in the
2106 * tree of the extent.
2108 * @offset: For metadata extents the offset is ignored and is currently
2109 * always passed as 0. For data extents it is the fileoffset
2110 * this extent belongs to.
2112 * @refs_to_add Number of references to add
2114 * @extent_op Pointer to a structure, holding information necessary when
2115 * updating a tree block's flags
2118 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2119 struct btrfs_delayed_ref_node *node,
2120 u64 parent, u64 root_objectid,
2121 u64 owner, u64 offset, int refs_to_add,
2122 struct btrfs_delayed_extent_op *extent_op)
2124 struct btrfs_path *path;
2125 struct extent_buffer *leaf;
2126 struct btrfs_extent_item *item;
2127 struct btrfs_key key;
2128 u64 bytenr = node->bytenr;
2129 u64 num_bytes = node->num_bytes;
2133 path = btrfs_alloc_path();
2137 path->reada = READA_FORWARD;
2138 path->leave_spinning = 1;
2139 /* this will setup the path even if it fails to insert the back ref */
2140 ret = insert_inline_extent_backref(trans, path, bytenr, num_bytes,
2141 parent, root_objectid, owner,
2142 offset, refs_to_add, extent_op);
2143 if ((ret < 0 && ret != -EAGAIN) || !ret)
2147 * Ok we had -EAGAIN which means we didn't have space to insert and
2148 * inline extent ref, so just update the reference count and add a
2151 leaf = path->nodes[0];
2152 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2153 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2154 refs = btrfs_extent_refs(leaf, item);
2155 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2157 __run_delayed_extent_op(extent_op, leaf, item);
2159 btrfs_mark_buffer_dirty(leaf);
2160 btrfs_release_path(path);
2162 path->reada = READA_FORWARD;
2163 path->leave_spinning = 1;
2164 /* now insert the actual backref */
2165 ret = insert_extent_backref(trans, path, bytenr, parent, root_objectid,
2166 owner, offset, refs_to_add);
2168 btrfs_abort_transaction(trans, ret);
2170 btrfs_free_path(path);
2174 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2175 struct btrfs_delayed_ref_node *node,
2176 struct btrfs_delayed_extent_op *extent_op,
2177 int insert_reserved)
2180 struct btrfs_delayed_data_ref *ref;
2181 struct btrfs_key ins;
2186 ins.objectid = node->bytenr;
2187 ins.offset = node->num_bytes;
2188 ins.type = BTRFS_EXTENT_ITEM_KEY;
2190 ref = btrfs_delayed_node_to_data_ref(node);
2191 trace_run_delayed_data_ref(trans->fs_info, node, ref, node->action);
2193 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2194 parent = ref->parent;
2195 ref_root = ref->root;
2197 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2199 flags |= extent_op->flags_to_set;
2200 ret = alloc_reserved_file_extent(trans, parent, ref_root,
2201 flags, ref->objectid,
2204 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2205 ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root,
2206 ref->objectid, ref->offset,
2207 node->ref_mod, extent_op);
2208 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2209 ret = __btrfs_free_extent(trans, node, parent,
2210 ref_root, ref->objectid,
2211 ref->offset, node->ref_mod,
2219 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2220 struct extent_buffer *leaf,
2221 struct btrfs_extent_item *ei)
2223 u64 flags = btrfs_extent_flags(leaf, ei);
2224 if (extent_op->update_flags) {
2225 flags |= extent_op->flags_to_set;
2226 btrfs_set_extent_flags(leaf, ei, flags);
2229 if (extent_op->update_key) {
2230 struct btrfs_tree_block_info *bi;
2231 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2232 bi = (struct btrfs_tree_block_info *)(ei + 1);
2233 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2237 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2238 struct btrfs_delayed_ref_head *head,
2239 struct btrfs_delayed_extent_op *extent_op)
2241 struct btrfs_fs_info *fs_info = trans->fs_info;
2242 struct btrfs_key key;
2243 struct btrfs_path *path;
2244 struct btrfs_extent_item *ei;
2245 struct extent_buffer *leaf;
2249 int metadata = !extent_op->is_data;
2254 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2257 path = btrfs_alloc_path();
2261 key.objectid = head->bytenr;
2264 key.type = BTRFS_METADATA_ITEM_KEY;
2265 key.offset = extent_op->level;
2267 key.type = BTRFS_EXTENT_ITEM_KEY;
2268 key.offset = head->num_bytes;
2272 path->reada = READA_FORWARD;
2273 path->leave_spinning = 1;
2274 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1);
2281 if (path->slots[0] > 0) {
2283 btrfs_item_key_to_cpu(path->nodes[0], &key,
2285 if (key.objectid == head->bytenr &&
2286 key.type == BTRFS_EXTENT_ITEM_KEY &&
2287 key.offset == head->num_bytes)
2291 btrfs_release_path(path);
2294 key.objectid = head->bytenr;
2295 key.offset = head->num_bytes;
2296 key.type = BTRFS_EXTENT_ITEM_KEY;
2305 leaf = path->nodes[0];
2306 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2308 if (unlikely(item_size < sizeof(*ei))) {
2310 btrfs_print_v0_err(fs_info);
2311 btrfs_abort_transaction(trans, err);
2315 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2316 __run_delayed_extent_op(extent_op, leaf, ei);
2318 btrfs_mark_buffer_dirty(leaf);
2320 btrfs_free_path(path);
2324 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2325 struct btrfs_delayed_ref_node *node,
2326 struct btrfs_delayed_extent_op *extent_op,
2327 int insert_reserved)
2330 struct btrfs_delayed_tree_ref *ref;
2334 ref = btrfs_delayed_node_to_tree_ref(node);
2335 trace_run_delayed_tree_ref(trans->fs_info, node, ref, node->action);
2337 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2338 parent = ref->parent;
2339 ref_root = ref->root;
2341 if (node->ref_mod != 1) {
2342 btrfs_err(trans->fs_info,
2343 "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu",
2344 node->bytenr, node->ref_mod, node->action, ref_root,
2348 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2349 BUG_ON(!extent_op || !extent_op->update_flags);
2350 ret = alloc_reserved_tree_block(trans, node, extent_op);
2351 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2352 ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root,
2353 ref->level, 0, 1, extent_op);
2354 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2355 ret = __btrfs_free_extent(trans, node, parent, ref_root,
2356 ref->level, 0, 1, extent_op);
2363 /* helper function to actually process a single delayed ref entry */
2364 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2365 struct btrfs_delayed_ref_node *node,
2366 struct btrfs_delayed_extent_op *extent_op,
2367 int insert_reserved)
2371 if (trans->aborted) {
2372 if (insert_reserved)
2373 btrfs_pin_extent(trans->fs_info, node->bytenr,
2374 node->num_bytes, 1);
2378 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2379 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2380 ret = run_delayed_tree_ref(trans, node, extent_op,
2382 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2383 node->type == BTRFS_SHARED_DATA_REF_KEY)
2384 ret = run_delayed_data_ref(trans, node, extent_op,
2388 if (ret && insert_reserved)
2389 btrfs_pin_extent(trans->fs_info, node->bytenr,
2390 node->num_bytes, 1);
2394 static inline struct btrfs_delayed_ref_node *
2395 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2397 struct btrfs_delayed_ref_node *ref;
2399 if (RB_EMPTY_ROOT(&head->ref_tree.rb_root))
2403 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2404 * This is to prevent a ref count from going down to zero, which deletes
2405 * the extent item from the extent tree, when there still are references
2406 * to add, which would fail because they would not find the extent item.
2408 if (!list_empty(&head->ref_add_list))
2409 return list_first_entry(&head->ref_add_list,
2410 struct btrfs_delayed_ref_node, add_list);
2412 ref = rb_entry(rb_first_cached(&head->ref_tree),
2413 struct btrfs_delayed_ref_node, ref_node);
2414 ASSERT(list_empty(&ref->add_list));
2418 static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
2419 struct btrfs_delayed_ref_head *head)
2421 spin_lock(&delayed_refs->lock);
2422 head->processing = 0;
2423 delayed_refs->num_heads_ready++;
2424 spin_unlock(&delayed_refs->lock);
2425 btrfs_delayed_ref_unlock(head);
2428 static struct btrfs_delayed_extent_op *cleanup_extent_op(
2429 struct btrfs_delayed_ref_head *head)
2431 struct btrfs_delayed_extent_op *extent_op = head->extent_op;
2436 if (head->must_insert_reserved) {
2437 head->extent_op = NULL;
2438 btrfs_free_delayed_extent_op(extent_op);
2444 static int run_and_cleanup_extent_op(struct btrfs_trans_handle *trans,
2445 struct btrfs_delayed_ref_head *head)
2447 struct btrfs_delayed_extent_op *extent_op;
2450 extent_op = cleanup_extent_op(head);
2453 head->extent_op = NULL;
2454 spin_unlock(&head->lock);
2455 ret = run_delayed_extent_op(trans, head, extent_op);
2456 btrfs_free_delayed_extent_op(extent_op);
2457 return ret ? ret : 1;
2460 void btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info *fs_info,
2461 struct btrfs_delayed_ref_root *delayed_refs,
2462 struct btrfs_delayed_ref_head *head)
2464 int nr_items = 1; /* Dropping this ref head update. */
2466 if (head->total_ref_mod < 0) {
2467 struct btrfs_space_info *space_info;
2471 flags = BTRFS_BLOCK_GROUP_DATA;
2472 else if (head->is_system)
2473 flags = BTRFS_BLOCK_GROUP_SYSTEM;
2475 flags = BTRFS_BLOCK_GROUP_METADATA;
2476 space_info = __find_space_info(fs_info, flags);
2478 percpu_counter_add_batch(&space_info->total_bytes_pinned,
2480 BTRFS_TOTAL_BYTES_PINNED_BATCH);
2483 * We had csum deletions accounted for in our delayed refs rsv,
2484 * we need to drop the csum leaves for this update from our
2487 if (head->is_data) {
2488 spin_lock(&delayed_refs->lock);
2489 delayed_refs->pending_csums -= head->num_bytes;
2490 spin_unlock(&delayed_refs->lock);
2491 nr_items += btrfs_csum_bytes_to_leaves(fs_info,
2496 btrfs_delayed_refs_rsv_release(fs_info, nr_items);
2499 static int cleanup_ref_head(struct btrfs_trans_handle *trans,
2500 struct btrfs_delayed_ref_head *head)
2503 struct btrfs_fs_info *fs_info = trans->fs_info;
2504 struct btrfs_delayed_ref_root *delayed_refs;
2507 delayed_refs = &trans->transaction->delayed_refs;
2509 ret = run_and_cleanup_extent_op(trans, head);
2511 unselect_delayed_ref_head(delayed_refs, head);
2512 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2519 * Need to drop our head ref lock and re-acquire the delayed ref lock
2520 * and then re-check to make sure nobody got added.
2522 spin_unlock(&head->lock);
2523 spin_lock(&delayed_refs->lock);
2524 spin_lock(&head->lock);
2525 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root) || head->extent_op) {
2526 spin_unlock(&head->lock);
2527 spin_unlock(&delayed_refs->lock);
2530 btrfs_delete_ref_head(delayed_refs, head);
2531 spin_unlock(&head->lock);
2532 spin_unlock(&delayed_refs->lock);
2534 if (head->must_insert_reserved) {
2535 btrfs_pin_extent(fs_info, head->bytenr,
2536 head->num_bytes, 1);
2537 if (head->is_data) {
2538 ret = btrfs_del_csums(trans, fs_info, head->bytenr,
2543 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
2545 trace_run_delayed_ref_head(fs_info, head, 0);
2546 btrfs_delayed_ref_unlock(head);
2547 btrfs_put_delayed_ref_head(head);
2551 static struct btrfs_delayed_ref_head *btrfs_obtain_ref_head(
2552 struct btrfs_trans_handle *trans)
2554 struct btrfs_delayed_ref_root *delayed_refs =
2555 &trans->transaction->delayed_refs;
2556 struct btrfs_delayed_ref_head *head = NULL;
2559 spin_lock(&delayed_refs->lock);
2560 head = btrfs_select_ref_head(delayed_refs);
2562 spin_unlock(&delayed_refs->lock);
2567 * Grab the lock that says we are going to process all the refs for
2570 ret = btrfs_delayed_ref_lock(delayed_refs, head);
2571 spin_unlock(&delayed_refs->lock);
2574 * We may have dropped the spin lock to get the head mutex lock, and
2575 * that might have given someone else time to free the head. If that's
2576 * true, it has been removed from our list and we can move on.
2579 head = ERR_PTR(-EAGAIN);
2584 static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans,
2585 struct btrfs_delayed_ref_head *locked_ref,
2586 unsigned long *run_refs)
2588 struct btrfs_fs_info *fs_info = trans->fs_info;
2589 struct btrfs_delayed_ref_root *delayed_refs;
2590 struct btrfs_delayed_extent_op *extent_op;
2591 struct btrfs_delayed_ref_node *ref;
2592 int must_insert_reserved = 0;
2595 delayed_refs = &trans->transaction->delayed_refs;
2597 lockdep_assert_held(&locked_ref->mutex);
2598 lockdep_assert_held(&locked_ref->lock);
2600 while ((ref = select_delayed_ref(locked_ref))) {
2602 btrfs_check_delayed_seq(fs_info, ref->seq)) {
2603 spin_unlock(&locked_ref->lock);
2604 unselect_delayed_ref_head(delayed_refs, locked_ref);
2610 rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree);
2611 RB_CLEAR_NODE(&ref->ref_node);
2612 if (!list_empty(&ref->add_list))
2613 list_del(&ref->add_list);
2615 * When we play the delayed ref, also correct the ref_mod on
2618 switch (ref->action) {
2619 case BTRFS_ADD_DELAYED_REF:
2620 case BTRFS_ADD_DELAYED_EXTENT:
2621 locked_ref->ref_mod -= ref->ref_mod;
2623 case BTRFS_DROP_DELAYED_REF:
2624 locked_ref->ref_mod += ref->ref_mod;
2629 atomic_dec(&delayed_refs->num_entries);
2632 * Record the must_insert_reserved flag before we drop the
2635 must_insert_reserved = locked_ref->must_insert_reserved;
2636 locked_ref->must_insert_reserved = 0;
2638 extent_op = locked_ref->extent_op;
2639 locked_ref->extent_op = NULL;
2640 spin_unlock(&locked_ref->lock);
2642 ret = run_one_delayed_ref(trans, ref, extent_op,
2643 must_insert_reserved);
2645 btrfs_free_delayed_extent_op(extent_op);
2647 unselect_delayed_ref_head(delayed_refs, locked_ref);
2648 btrfs_put_delayed_ref(ref);
2649 btrfs_debug(fs_info, "run_one_delayed_ref returned %d",
2654 btrfs_put_delayed_ref(ref);
2657 spin_lock(&locked_ref->lock);
2658 btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref);
2665 * Returns 0 on success or if called with an already aborted transaction.
2666 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2668 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2671 struct btrfs_fs_info *fs_info = trans->fs_info;
2672 struct btrfs_delayed_ref_root *delayed_refs;
2673 struct btrfs_delayed_ref_head *locked_ref = NULL;
2674 ktime_t start = ktime_get();
2676 unsigned long count = 0;
2677 unsigned long actual_count = 0;
2679 delayed_refs = &trans->transaction->delayed_refs;
2682 locked_ref = btrfs_obtain_ref_head(trans);
2683 if (IS_ERR_OR_NULL(locked_ref)) {
2684 if (PTR_ERR(locked_ref) == -EAGAIN) {
2693 * We need to try and merge add/drops of the same ref since we
2694 * can run into issues with relocate dropping the implicit ref
2695 * and then it being added back again before the drop can
2696 * finish. If we merged anything we need to re-loop so we can
2698 * Or we can get node references of the same type that weren't
2699 * merged when created due to bumps in the tree mod seq, and
2700 * we need to merge them to prevent adding an inline extent
2701 * backref before dropping it (triggering a BUG_ON at
2702 * insert_inline_extent_backref()).
2704 spin_lock(&locked_ref->lock);
2705 btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref);
2707 ret = btrfs_run_delayed_refs_for_head(trans, locked_ref,
2709 if (ret < 0 && ret != -EAGAIN) {
2711 * Error, btrfs_run_delayed_refs_for_head already
2712 * unlocked everything so just bail out
2717 * Success, perform the usual cleanup of a processed
2720 ret = cleanup_ref_head(trans, locked_ref);
2722 /* We dropped our lock, we need to loop. */
2731 * Either success case or btrfs_run_delayed_refs_for_head
2732 * returned -EAGAIN, meaning we need to select another head
2737 } while ((nr != -1 && count < nr) || locked_ref);
2740 * We don't want to include ref heads since we can have empty ref heads
2741 * and those will drastically skew our runtime down since we just do
2742 * accounting, no actual extent tree updates.
2744 if (actual_count > 0) {
2745 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2749 * We weigh the current average higher than our current runtime
2750 * to avoid large swings in the average.
2752 spin_lock(&delayed_refs->lock);
2753 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2754 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2755 spin_unlock(&delayed_refs->lock);
2760 #ifdef SCRAMBLE_DELAYED_REFS
2762 * Normally delayed refs get processed in ascending bytenr order. This
2763 * correlates in most cases to the order added. To expose dependencies on this
2764 * order, we start to process the tree in the middle instead of the beginning
2766 static u64 find_middle(struct rb_root *root)
2768 struct rb_node *n = root->rb_node;
2769 struct btrfs_delayed_ref_node *entry;
2772 u64 first = 0, last = 0;
2776 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2777 first = entry->bytenr;
2781 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2782 last = entry->bytenr;
2787 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2788 WARN_ON(!entry->in_tree);
2790 middle = entry->bytenr;
2803 static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads)
2807 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2808 sizeof(struct btrfs_extent_inline_ref));
2809 if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2810 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2813 * We don't ever fill up leaves all the way so multiply by 2 just to be
2814 * closer to what we're really going to want to use.
2816 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info));
2820 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2821 * would require to store the csums for that many bytes.
2823 u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes)
2826 u64 num_csums_per_leaf;
2829 csum_size = BTRFS_MAX_ITEM_SIZE(fs_info);
2830 num_csums_per_leaf = div64_u64(csum_size,
2831 (u64)btrfs_super_csum_size(fs_info->super_copy));
2832 num_csums = div64_u64(csum_bytes, fs_info->sectorsize);
2833 num_csums += num_csums_per_leaf - 1;
2834 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2838 bool btrfs_check_space_for_delayed_refs(struct btrfs_fs_info *fs_info)
2840 struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv;
2841 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
2845 spin_lock(&global_rsv->lock);
2846 reserved = global_rsv->reserved;
2847 spin_unlock(&global_rsv->lock);
2850 * Since the global reserve is just kind of magic we don't really want
2851 * to rely on it to save our bacon, so if our size is more than the
2852 * delayed_refs_rsv and the global rsv then it's time to think about
2855 spin_lock(&delayed_refs_rsv->lock);
2856 reserved += delayed_refs_rsv->reserved;
2857 if (delayed_refs_rsv->size >= reserved)
2859 spin_unlock(&delayed_refs_rsv->lock);
2863 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans)
2866 atomic_read(&trans->transaction->delayed_refs.num_entries);
2871 avg_runtime = trans->fs_info->avg_delayed_ref_runtime;
2872 val = num_entries * avg_runtime;
2873 if (val >= NSEC_PER_SEC)
2875 if (val >= NSEC_PER_SEC / 2)
2878 return btrfs_check_space_for_delayed_refs(trans->fs_info);
2882 * this starts processing the delayed reference count updates and
2883 * extent insertions we have queued up so far. count can be
2884 * 0, which means to process everything in the tree at the start
2885 * of the run (but not newly added entries), or it can be some target
2886 * number you'd like to process.
2888 * Returns 0 on success or if called with an aborted transaction
2889 * Returns <0 on error and aborts the transaction
2891 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2892 unsigned long count)
2894 struct btrfs_fs_info *fs_info = trans->fs_info;
2895 struct rb_node *node;
2896 struct btrfs_delayed_ref_root *delayed_refs;
2897 struct btrfs_delayed_ref_head *head;
2899 int run_all = count == (unsigned long)-1;
2901 /* We'll clean this up in btrfs_cleanup_transaction */
2905 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
2908 delayed_refs = &trans->transaction->delayed_refs;
2910 count = atomic_read(&delayed_refs->num_entries) * 2;
2913 #ifdef SCRAMBLE_DELAYED_REFS
2914 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2916 ret = __btrfs_run_delayed_refs(trans, count);
2918 btrfs_abort_transaction(trans, ret);
2923 btrfs_create_pending_block_groups(trans);
2925 spin_lock(&delayed_refs->lock);
2926 node = rb_first_cached(&delayed_refs->href_root);
2928 spin_unlock(&delayed_refs->lock);
2931 head = rb_entry(node, struct btrfs_delayed_ref_head,
2933 refcount_inc(&head->refs);
2934 spin_unlock(&delayed_refs->lock);
2936 /* Mutex was contended, block until it's released and retry. */
2937 mutex_lock(&head->mutex);
2938 mutex_unlock(&head->mutex);
2940 btrfs_put_delayed_ref_head(head);
2948 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2949 u64 bytenr, u64 num_bytes, u64 flags,
2950 int level, int is_data)
2952 struct btrfs_delayed_extent_op *extent_op;
2955 extent_op = btrfs_alloc_delayed_extent_op();
2959 extent_op->flags_to_set = flags;
2960 extent_op->update_flags = true;
2961 extent_op->update_key = false;
2962 extent_op->is_data = is_data ? true : false;
2963 extent_op->level = level;
2965 ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2967 btrfs_free_delayed_extent_op(extent_op);
2971 static noinline int check_delayed_ref(struct btrfs_root *root,
2972 struct btrfs_path *path,
2973 u64 objectid, u64 offset, u64 bytenr)
2975 struct btrfs_delayed_ref_head *head;
2976 struct btrfs_delayed_ref_node *ref;
2977 struct btrfs_delayed_data_ref *data_ref;
2978 struct btrfs_delayed_ref_root *delayed_refs;
2979 struct btrfs_transaction *cur_trans;
2980 struct rb_node *node;
2983 spin_lock(&root->fs_info->trans_lock);
2984 cur_trans = root->fs_info->running_transaction;
2986 refcount_inc(&cur_trans->use_count);
2987 spin_unlock(&root->fs_info->trans_lock);
2991 delayed_refs = &cur_trans->delayed_refs;
2992 spin_lock(&delayed_refs->lock);
2993 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
2995 spin_unlock(&delayed_refs->lock);
2996 btrfs_put_transaction(cur_trans);
3000 if (!mutex_trylock(&head->mutex)) {
3001 refcount_inc(&head->refs);
3002 spin_unlock(&delayed_refs->lock);
3004 btrfs_release_path(path);
3007 * Mutex was contended, block until it's released and let
3010 mutex_lock(&head->mutex);
3011 mutex_unlock(&head->mutex);
3012 btrfs_put_delayed_ref_head(head);
3013 btrfs_put_transaction(cur_trans);
3016 spin_unlock(&delayed_refs->lock);
3018 spin_lock(&head->lock);
3020 * XXX: We should replace this with a proper search function in the
3023 for (node = rb_first_cached(&head->ref_tree); node;
3024 node = rb_next(node)) {
3025 ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node);
3026 /* If it's a shared ref we know a cross reference exists */
3027 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
3032 data_ref = btrfs_delayed_node_to_data_ref(ref);
3035 * If our ref doesn't match the one we're currently looking at
3036 * then we have a cross reference.
3038 if (data_ref->root != root->root_key.objectid ||
3039 data_ref->objectid != objectid ||
3040 data_ref->offset != offset) {
3045 spin_unlock(&head->lock);
3046 mutex_unlock(&head->mutex);
3047 btrfs_put_transaction(cur_trans);
3051 static noinline int check_committed_ref(struct btrfs_root *root,
3052 struct btrfs_path *path,
3053 u64 objectid, u64 offset, u64 bytenr)
3055 struct btrfs_fs_info *fs_info = root->fs_info;
3056 struct btrfs_root *extent_root = fs_info->extent_root;
3057 struct extent_buffer *leaf;
3058 struct btrfs_extent_data_ref *ref;
3059 struct btrfs_extent_inline_ref *iref;
3060 struct btrfs_extent_item *ei;
3061 struct btrfs_key key;
3066 key.objectid = bytenr;
3067 key.offset = (u64)-1;
3068 key.type = BTRFS_EXTENT_ITEM_KEY;
3070 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3073 BUG_ON(ret == 0); /* Corruption */
3076 if (path->slots[0] == 0)
3080 leaf = path->nodes[0];
3081 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3083 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3087 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3088 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3090 if (item_size != sizeof(*ei) +
3091 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3094 if (btrfs_extent_generation(leaf, ei) <=
3095 btrfs_root_last_snapshot(&root->root_item))
3098 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3100 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
3101 if (type != BTRFS_EXTENT_DATA_REF_KEY)
3104 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3105 if (btrfs_extent_refs(leaf, ei) !=
3106 btrfs_extent_data_ref_count(leaf, ref) ||
3107 btrfs_extent_data_ref_root(leaf, ref) !=
3108 root->root_key.objectid ||
3109 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3110 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3118 int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
3121 struct btrfs_path *path;
3124 path = btrfs_alloc_path();
3129 ret = check_committed_ref(root, path, objectid,
3131 if (ret && ret != -ENOENT)
3134 ret = check_delayed_ref(root, path, objectid, offset, bytenr);
3135 } while (ret == -EAGAIN);
3138 btrfs_free_path(path);
3139 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3144 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3145 struct btrfs_root *root,
3146 struct extent_buffer *buf,
3147 int full_backref, int inc)
3149 struct btrfs_fs_info *fs_info = root->fs_info;
3155 struct btrfs_key key;
3156 struct btrfs_file_extent_item *fi;
3157 struct btrfs_ref generic_ref = { 0 };
3158 bool for_reloc = btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC);
3164 if (btrfs_is_testing(fs_info))
3167 ref_root = btrfs_header_owner(buf);
3168 nritems = btrfs_header_nritems(buf);
3169 level = btrfs_header_level(buf);
3171 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3175 parent = buf->start;
3179 action = BTRFS_ADD_DELAYED_REF;
3181 action = BTRFS_DROP_DELAYED_REF;
3183 for (i = 0; i < nritems; i++) {
3185 btrfs_item_key_to_cpu(buf, &key, i);
3186 if (key.type != BTRFS_EXTENT_DATA_KEY)
3188 fi = btrfs_item_ptr(buf, i,
3189 struct btrfs_file_extent_item);
3190 if (btrfs_file_extent_type(buf, fi) ==
3191 BTRFS_FILE_EXTENT_INLINE)
3193 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3197 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3198 key.offset -= btrfs_file_extent_offset(buf, fi);
3199 btrfs_init_generic_ref(&generic_ref, action, bytenr,
3201 generic_ref.real_root = root->root_key.objectid;
3202 btrfs_init_data_ref(&generic_ref, ref_root, key.objectid,
3204 generic_ref.skip_qgroup = for_reloc;
3206 ret = btrfs_inc_extent_ref(trans, &generic_ref);
3208 ret = btrfs_free_extent(trans, &generic_ref);
3212 bytenr = btrfs_node_blockptr(buf, i);
3213 num_bytes = fs_info->nodesize;
3214 btrfs_init_generic_ref(&generic_ref, action, bytenr,
3216 generic_ref.real_root = root->root_key.objectid;
3217 btrfs_init_tree_ref(&generic_ref, level - 1, ref_root);
3218 generic_ref.skip_qgroup = for_reloc;
3220 ret = btrfs_inc_extent_ref(trans, &generic_ref);
3222 ret = btrfs_free_extent(trans, &generic_ref);
3232 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3233 struct extent_buffer *buf, int full_backref)
3235 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3238 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3239 struct extent_buffer *buf, int full_backref)
3241 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3244 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3245 struct btrfs_path *path,
3246 struct btrfs_block_group_cache *cache)
3248 struct btrfs_fs_info *fs_info = trans->fs_info;
3250 struct btrfs_root *extent_root = fs_info->extent_root;
3252 struct extent_buffer *leaf;
3254 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3261 leaf = path->nodes[0];
3262 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3263 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3264 btrfs_mark_buffer_dirty(leaf);
3266 btrfs_release_path(path);
3271 static struct btrfs_block_group_cache *next_block_group(
3272 struct btrfs_block_group_cache *cache)
3274 struct btrfs_fs_info *fs_info = cache->fs_info;
3275 struct rb_node *node;
3277 spin_lock(&fs_info->block_group_cache_lock);
3279 /* If our block group was removed, we need a full search. */
3280 if (RB_EMPTY_NODE(&cache->cache_node)) {
3281 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3283 spin_unlock(&fs_info->block_group_cache_lock);
3284 btrfs_put_block_group(cache);
3285 cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache;
3287 node = rb_next(&cache->cache_node);
3288 btrfs_put_block_group(cache);
3290 cache = rb_entry(node, struct btrfs_block_group_cache,
3292 btrfs_get_block_group(cache);
3295 spin_unlock(&fs_info->block_group_cache_lock);
3299 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3300 struct btrfs_trans_handle *trans,
3301 struct btrfs_path *path)
3303 struct btrfs_fs_info *fs_info = block_group->fs_info;
3304 struct btrfs_root *root = fs_info->tree_root;
3305 struct inode *inode = NULL;
3306 struct extent_changeset *data_reserved = NULL;
3308 int dcs = BTRFS_DC_ERROR;
3314 * If this block group is smaller than 100 megs don't bother caching the
3317 if (block_group->key.offset < (100 * SZ_1M)) {
3318 spin_lock(&block_group->lock);
3319 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3320 spin_unlock(&block_group->lock);
3327 inode = lookup_free_space_inode(block_group, path);
3328 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3329 ret = PTR_ERR(inode);
3330 btrfs_release_path(path);
3334 if (IS_ERR(inode)) {
3338 if (block_group->ro)
3341 ret = create_free_space_inode(trans, block_group, path);
3348 * We want to set the generation to 0, that way if anything goes wrong
3349 * from here on out we know not to trust this cache when we load up next
3352 BTRFS_I(inode)->generation = 0;
3353 ret = btrfs_update_inode(trans, root, inode);
3356 * So theoretically we could recover from this, simply set the
3357 * super cache generation to 0 so we know to invalidate the
3358 * cache, but then we'd have to keep track of the block groups
3359 * that fail this way so we know we _have_ to reset this cache
3360 * before the next commit or risk reading stale cache. So to
3361 * limit our exposure to horrible edge cases lets just abort the
3362 * transaction, this only happens in really bad situations
3365 btrfs_abort_transaction(trans, ret);
3370 /* We've already setup this transaction, go ahead and exit */
3371 if (block_group->cache_generation == trans->transid &&
3372 i_size_read(inode)) {
3373 dcs = BTRFS_DC_SETUP;
3377 if (i_size_read(inode) > 0) {
3378 ret = btrfs_check_trunc_cache_free_space(fs_info,
3379 &fs_info->global_block_rsv);
3383 ret = btrfs_truncate_free_space_cache(trans, NULL, inode);
3388 spin_lock(&block_group->lock);
3389 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3390 !btrfs_test_opt(fs_info, SPACE_CACHE)) {
3392 * don't bother trying to write stuff out _if_
3393 * a) we're not cached,
3394 * b) we're with nospace_cache mount option,
3395 * c) we're with v2 space_cache (FREE_SPACE_TREE).
3397 dcs = BTRFS_DC_WRITTEN;
3398 spin_unlock(&block_group->lock);
3401 spin_unlock(&block_group->lock);
3404 * We hit an ENOSPC when setting up the cache in this transaction, just
3405 * skip doing the setup, we've already cleared the cache so we're safe.
3407 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
3413 * Try to preallocate enough space based on how big the block group is.
3414 * Keep in mind this has to include any pinned space which could end up
3415 * taking up quite a bit since it's not folded into the other space
3418 num_pages = div_u64(block_group->key.offset, SZ_256M);
3423 num_pages *= PAGE_SIZE;
3425 ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages);
3429 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3430 num_pages, num_pages,
3433 * Our cache requires contiguous chunks so that we don't modify a bunch
3434 * of metadata or split extents when writing the cache out, which means
3435 * we can enospc if we are heavily fragmented in addition to just normal
3436 * out of space conditions. So if we hit this just skip setting up any
3437 * other block groups for this transaction, maybe we'll unpin enough
3438 * space the next time around.
3441 dcs = BTRFS_DC_SETUP;
3442 else if (ret == -ENOSPC)
3443 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
3448 btrfs_release_path(path);
3450 spin_lock(&block_group->lock);
3451 if (!ret && dcs == BTRFS_DC_SETUP)
3452 block_group->cache_generation = trans->transid;
3453 block_group->disk_cache_state = dcs;
3454 spin_unlock(&block_group->lock);
3456 extent_changeset_free(data_reserved);
3460 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans)
3462 struct btrfs_fs_info *fs_info = trans->fs_info;
3463 struct btrfs_block_group_cache *cache, *tmp;
3464 struct btrfs_transaction *cur_trans = trans->transaction;
3465 struct btrfs_path *path;
3467 if (list_empty(&cur_trans->dirty_bgs) ||
3468 !btrfs_test_opt(fs_info, SPACE_CACHE))
3471 path = btrfs_alloc_path();
3475 /* Could add new block groups, use _safe just in case */
3476 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3478 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3479 cache_save_setup(cache, trans, path);
3482 btrfs_free_path(path);
3487 * transaction commit does final block group cache writeback during a
3488 * critical section where nothing is allowed to change the FS. This is
3489 * required in order for the cache to actually match the block group,
3490 * but can introduce a lot of latency into the commit.
3492 * So, btrfs_start_dirty_block_groups is here to kick off block group
3493 * cache IO. There's a chance we'll have to redo some of it if the
3494 * block group changes again during the commit, but it greatly reduces
3495 * the commit latency by getting rid of the easy block groups while
3496 * we're still allowing others to join the commit.
3498 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans)
3500 struct btrfs_fs_info *fs_info = trans->fs_info;
3501 struct btrfs_block_group_cache *cache;
3502 struct btrfs_transaction *cur_trans = trans->transaction;
3505 struct btrfs_path *path = NULL;
3507 struct list_head *io = &cur_trans->io_bgs;
3508 int num_started = 0;
3511 spin_lock(&cur_trans->dirty_bgs_lock);
3512 if (list_empty(&cur_trans->dirty_bgs)) {
3513 spin_unlock(&cur_trans->dirty_bgs_lock);
3516 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3517 spin_unlock(&cur_trans->dirty_bgs_lock);
3521 * make sure all the block groups on our dirty list actually
3524 btrfs_create_pending_block_groups(trans);
3527 path = btrfs_alloc_path();
3533 * cache_write_mutex is here only to save us from balance or automatic
3534 * removal of empty block groups deleting this block group while we are
3535 * writing out the cache
3537 mutex_lock(&trans->transaction->cache_write_mutex);
3538 while (!list_empty(&dirty)) {
3539 bool drop_reserve = true;
3541 cache = list_first_entry(&dirty,
3542 struct btrfs_block_group_cache,
3545 * this can happen if something re-dirties a block
3546 * group that is already under IO. Just wait for it to
3547 * finish and then do it all again
3549 if (!list_empty(&cache->io_list)) {
3550 list_del_init(&cache->io_list);
3551 btrfs_wait_cache_io(trans, cache, path);
3552 btrfs_put_block_group(cache);
3557 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3558 * if it should update the cache_state. Don't delete
3559 * until after we wait.
3561 * Since we're not running in the commit critical section
3562 * we need the dirty_bgs_lock to protect from update_block_group
3564 spin_lock(&cur_trans->dirty_bgs_lock);
3565 list_del_init(&cache->dirty_list);
3566 spin_unlock(&cur_trans->dirty_bgs_lock);
3570 cache_save_setup(cache, trans, path);
3572 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3573 cache->io_ctl.inode = NULL;
3574 ret = btrfs_write_out_cache(trans, cache, path);
3575 if (ret == 0 && cache->io_ctl.inode) {
3580 * The cache_write_mutex is protecting the
3581 * io_list, also refer to the definition of
3582 * btrfs_transaction::io_bgs for more details
3584 list_add_tail(&cache->io_list, io);
3587 * if we failed to write the cache, the
3588 * generation will be bad and life goes on
3594 ret = write_one_cache_group(trans, path, cache);
3596 * Our block group might still be attached to the list
3597 * of new block groups in the transaction handle of some
3598 * other task (struct btrfs_trans_handle->new_bgs). This
3599 * means its block group item isn't yet in the extent
3600 * tree. If this happens ignore the error, as we will
3601 * try again later in the critical section of the
3602 * transaction commit.
3604 if (ret == -ENOENT) {
3606 spin_lock(&cur_trans->dirty_bgs_lock);
3607 if (list_empty(&cache->dirty_list)) {
3608 list_add_tail(&cache->dirty_list,
3609 &cur_trans->dirty_bgs);
3610 btrfs_get_block_group(cache);
3611 drop_reserve = false;
3613 spin_unlock(&cur_trans->dirty_bgs_lock);
3615 btrfs_abort_transaction(trans, ret);
3619 /* if it's not on the io list, we need to put the block group */
3621 btrfs_put_block_group(cache);
3623 btrfs_delayed_refs_rsv_release(fs_info, 1);
3629 * Avoid blocking other tasks for too long. It might even save
3630 * us from writing caches for block groups that are going to be
3633 mutex_unlock(&trans->transaction->cache_write_mutex);
3634 mutex_lock(&trans->transaction->cache_write_mutex);
3636 mutex_unlock(&trans->transaction->cache_write_mutex);
3639 * go through delayed refs for all the stuff we've just kicked off
3640 * and then loop back (just once)
3642 ret = btrfs_run_delayed_refs(trans, 0);
3643 if (!ret && loops == 0) {
3645 spin_lock(&cur_trans->dirty_bgs_lock);
3646 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3648 * dirty_bgs_lock protects us from concurrent block group
3649 * deletes too (not just cache_write_mutex).
3651 if (!list_empty(&dirty)) {
3652 spin_unlock(&cur_trans->dirty_bgs_lock);
3655 spin_unlock(&cur_trans->dirty_bgs_lock);
3656 } else if (ret < 0) {
3657 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
3660 btrfs_free_path(path);
3664 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans)
3666 struct btrfs_fs_info *fs_info = trans->fs_info;
3667 struct btrfs_block_group_cache *cache;
3668 struct btrfs_transaction *cur_trans = trans->transaction;
3671 struct btrfs_path *path;
3672 struct list_head *io = &cur_trans->io_bgs;
3673 int num_started = 0;
3675 path = btrfs_alloc_path();
3680 * Even though we are in the critical section of the transaction commit,
3681 * we can still have concurrent tasks adding elements to this
3682 * transaction's list of dirty block groups. These tasks correspond to
3683 * endio free space workers started when writeback finishes for a
3684 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3685 * allocate new block groups as a result of COWing nodes of the root
3686 * tree when updating the free space inode. The writeback for the space
3687 * caches is triggered by an earlier call to
3688 * btrfs_start_dirty_block_groups() and iterations of the following
3690 * Also we want to do the cache_save_setup first and then run the
3691 * delayed refs to make sure we have the best chance at doing this all
3694 spin_lock(&cur_trans->dirty_bgs_lock);
3695 while (!list_empty(&cur_trans->dirty_bgs)) {
3696 cache = list_first_entry(&cur_trans->dirty_bgs,
3697 struct btrfs_block_group_cache,
3701 * this can happen if cache_save_setup re-dirties a block
3702 * group that is already under IO. Just wait for it to
3703 * finish and then do it all again
3705 if (!list_empty(&cache->io_list)) {
3706 spin_unlock(&cur_trans->dirty_bgs_lock);
3707 list_del_init(&cache->io_list);
3708 btrfs_wait_cache_io(trans, cache, path);
3709 btrfs_put_block_group(cache);
3710 spin_lock(&cur_trans->dirty_bgs_lock);
3714 * don't remove from the dirty list until after we've waited
3717 list_del_init(&cache->dirty_list);
3718 spin_unlock(&cur_trans->dirty_bgs_lock);
3721 cache_save_setup(cache, trans, path);
3724 ret = btrfs_run_delayed_refs(trans,
3725 (unsigned long) -1);
3727 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3728 cache->io_ctl.inode = NULL;
3729 ret = btrfs_write_out_cache(trans, cache, path);
3730 if (ret == 0 && cache->io_ctl.inode) {
3733 list_add_tail(&cache->io_list, io);
3736 * if we failed to write the cache, the
3737 * generation will be bad and life goes on
3743 ret = write_one_cache_group(trans, path, cache);
3745 * One of the free space endio workers might have
3746 * created a new block group while updating a free space
3747 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3748 * and hasn't released its transaction handle yet, in
3749 * which case the new block group is still attached to
3750 * its transaction handle and its creation has not
3751 * finished yet (no block group item in the extent tree
3752 * yet, etc). If this is the case, wait for all free
3753 * space endio workers to finish and retry. This is a
3754 * a very rare case so no need for a more efficient and
3757 if (ret == -ENOENT) {
3758 wait_event(cur_trans->writer_wait,
3759 atomic_read(&cur_trans->num_writers) == 1);
3760 ret = write_one_cache_group(trans, path, cache);
3763 btrfs_abort_transaction(trans, ret);
3766 /* if its not on the io list, we need to put the block group */
3768 btrfs_put_block_group(cache);
3769 btrfs_delayed_refs_rsv_release(fs_info, 1);
3770 spin_lock(&cur_trans->dirty_bgs_lock);
3772 spin_unlock(&cur_trans->dirty_bgs_lock);
3775 * Refer to the definition of io_bgs member for details why it's safe
3776 * to use it without any locking
3778 while (!list_empty(io)) {
3779 cache = list_first_entry(io, struct btrfs_block_group_cache,
3781 list_del_init(&cache->io_list);
3782 btrfs_wait_cache_io(trans, cache, path);
3783 btrfs_put_block_group(cache);
3786 btrfs_free_path(path);
3790 int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
3792 struct btrfs_block_group_cache *block_group;
3795 block_group = btrfs_lookup_block_group(fs_info, bytenr);
3796 if (!block_group || block_group->ro)
3799 btrfs_put_block_group(block_group);
3803 bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3805 struct btrfs_block_group_cache *bg;
3808 bg = btrfs_lookup_block_group(fs_info, bytenr);
3812 spin_lock(&bg->lock);
3816 atomic_inc(&bg->nocow_writers);
3817 spin_unlock(&bg->lock);
3819 /* no put on block group, done by btrfs_dec_nocow_writers */
3821 btrfs_put_block_group(bg);
3827 void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3829 struct btrfs_block_group_cache *bg;
3831 bg = btrfs_lookup_block_group(fs_info, bytenr);
3833 if (atomic_dec_and_test(&bg->nocow_writers))
3834 wake_up_var(&bg->nocow_writers);
3836 * Once for our lookup and once for the lookup done by a previous call
3837 * to btrfs_inc_nocow_writers()
3839 btrfs_put_block_group(bg);
3840 btrfs_put_block_group(bg);
3843 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg)
3845 wait_var_event(&bg->nocow_writers, !atomic_read(&bg->nocow_writers));
3848 static const char *alloc_name(u64 flags)
3851 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3853 case BTRFS_BLOCK_GROUP_METADATA:
3855 case BTRFS_BLOCK_GROUP_DATA:
3857 case BTRFS_BLOCK_GROUP_SYSTEM:
3861 return "invalid-combination";
3865 static int create_space_info(struct btrfs_fs_info *info, u64 flags)
3868 struct btrfs_space_info *space_info;
3872 space_info = kzalloc(sizeof(*space_info), GFP_NOFS);
3876 ret = percpu_counter_init(&space_info->total_bytes_pinned, 0,
3883 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3884 INIT_LIST_HEAD(&space_info->block_groups[i]);
3885 init_rwsem(&space_info->groups_sem);
3886 spin_lock_init(&space_info->lock);
3887 space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3888 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3889 init_waitqueue_head(&space_info->wait);
3890 INIT_LIST_HEAD(&space_info->ro_bgs);
3891 INIT_LIST_HEAD(&space_info->tickets);
3892 INIT_LIST_HEAD(&space_info->priority_tickets);
3894 ret = kobject_init_and_add(&space_info->kobj, &space_info_ktype,
3895 info->space_info_kobj, "%s",
3896 alloc_name(space_info->flags));
3898 kobject_put(&space_info->kobj);
3902 list_add_rcu(&space_info->list, &info->space_info);
3903 if (flags & BTRFS_BLOCK_GROUP_DATA)
3904 info->data_sinfo = space_info;
3909 static void update_space_info(struct btrfs_fs_info *info, u64 flags,
3910 u64 total_bytes, u64 bytes_used,
3912 struct btrfs_space_info **space_info)
3914 struct btrfs_space_info *found;
3917 factor = btrfs_bg_type_to_factor(flags);
3919 found = __find_space_info(info, flags);
3921 spin_lock(&found->lock);
3922 found->total_bytes += total_bytes;
3923 found->disk_total += total_bytes * factor;
3924 found->bytes_used += bytes_used;
3925 found->disk_used += bytes_used * factor;
3926 found->bytes_readonly += bytes_readonly;
3927 if (total_bytes > 0)
3929 space_info_add_new_bytes(info, found, total_bytes -
3930 bytes_used - bytes_readonly);
3931 spin_unlock(&found->lock);
3932 *space_info = found;
3935 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3937 u64 extra_flags = chunk_to_extended(flags) &
3938 BTRFS_EXTENDED_PROFILE_MASK;
3940 write_seqlock(&fs_info->profiles_lock);
3941 if (flags & BTRFS_BLOCK_GROUP_DATA)
3942 fs_info->avail_data_alloc_bits |= extra_flags;
3943 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3944 fs_info->avail_metadata_alloc_bits |= extra_flags;
3945 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3946 fs_info->avail_system_alloc_bits |= extra_flags;
3947 write_sequnlock(&fs_info->profiles_lock);
3951 * returns target flags in extended format or 0 if restripe for this
3952 * chunk_type is not in progress
3954 * should be called with balance_lock held
3956 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3958 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3964 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3965 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3966 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3967 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3968 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3969 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3970 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3971 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3972 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3979 * @flags: available profiles in extended format (see ctree.h)
3981 * Returns reduced profile in chunk format. If profile changing is in
3982 * progress (either running or paused) picks the target profile (if it's
3983 * already available), otherwise falls back to plain reducing.
3985 static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags)
3987 u64 num_devices = fs_info->fs_devices->rw_devices;
3993 * see if restripe for this chunk_type is in progress, if so
3994 * try to reduce to the target profile
3996 spin_lock(&fs_info->balance_lock);
3997 target = get_restripe_target(fs_info, flags);
3999 /* pick target profile only if it's already available */
4000 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
4001 spin_unlock(&fs_info->balance_lock);
4002 return extended_to_chunk(target);
4005 spin_unlock(&fs_info->balance_lock);
4007 /* First, mask out the RAID levels which aren't possible */
4008 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4009 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
4010 allowed |= btrfs_raid_array[raid_type].bg_flag;
4014 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
4015 allowed = BTRFS_BLOCK_GROUP_RAID6;
4016 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
4017 allowed = BTRFS_BLOCK_GROUP_RAID5;
4018 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
4019 allowed = BTRFS_BLOCK_GROUP_RAID10;
4020 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
4021 allowed = BTRFS_BLOCK_GROUP_RAID1;
4022 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
4023 allowed = BTRFS_BLOCK_GROUP_RAID0;
4025 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
4027 return extended_to_chunk(flags | allowed);
4030 static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags)
4037 seq = read_seqbegin(&fs_info->profiles_lock);
4039 if (flags & BTRFS_BLOCK_GROUP_DATA)
4040 flags |= fs_info->avail_data_alloc_bits;
4041 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4042 flags |= fs_info->avail_system_alloc_bits;
4043 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
4044 flags |= fs_info->avail_metadata_alloc_bits;
4045 } while (read_seqretry(&fs_info->profiles_lock, seq));
4047 return btrfs_reduce_alloc_profile(fs_info, flags);
4050 static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
4052 struct btrfs_fs_info *fs_info = root->fs_info;
4057 flags = BTRFS_BLOCK_GROUP_DATA;
4058 else if (root == fs_info->chunk_root)
4059 flags = BTRFS_BLOCK_GROUP_SYSTEM;
4061 flags = BTRFS_BLOCK_GROUP_METADATA;
4063 ret = get_alloc_profile(fs_info, flags);
4067 u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info)
4069 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA);
4072 u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info)
4074 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4077 u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info)
4079 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4082 static u64 btrfs_space_info_used(struct btrfs_space_info *s_info,
4083 bool may_use_included)
4086 return s_info->bytes_used + s_info->bytes_reserved +
4087 s_info->bytes_pinned + s_info->bytes_readonly +
4088 (may_use_included ? s_info->bytes_may_use : 0);
4091 int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes)
4093 struct btrfs_root *root = inode->root;
4094 struct btrfs_fs_info *fs_info = root->fs_info;
4095 struct btrfs_space_info *data_sinfo = fs_info->data_sinfo;
4098 int need_commit = 2;
4099 int have_pinned_space;
4101 /* make sure bytes are sectorsize aligned */
4102 bytes = ALIGN(bytes, fs_info->sectorsize);
4104 if (btrfs_is_free_space_inode(inode)) {
4106 ASSERT(current->journal_info);
4110 /* make sure we have enough space to handle the data first */
4111 spin_lock(&data_sinfo->lock);
4112 used = btrfs_space_info_used(data_sinfo, true);
4114 if (used + bytes > data_sinfo->total_bytes) {
4115 struct btrfs_trans_handle *trans;
4118 * if we don't have enough free bytes in this space then we need
4119 * to alloc a new chunk.
4121 if (!data_sinfo->full) {
4124 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4125 spin_unlock(&data_sinfo->lock);
4127 alloc_target = btrfs_data_alloc_profile(fs_info);
4129 * It is ugly that we don't call nolock join
4130 * transaction for the free space inode case here.
4131 * But it is safe because we only do the data space
4132 * reservation for the free space cache in the
4133 * transaction context, the common join transaction
4134 * just increase the counter of the current transaction
4135 * handler, doesn't try to acquire the trans_lock of
4138 trans = btrfs_join_transaction(root);
4140 return PTR_ERR(trans);
4142 ret = do_chunk_alloc(trans, alloc_target,
4143 CHUNK_ALLOC_NO_FORCE);
4144 btrfs_end_transaction(trans);
4149 have_pinned_space = 1;
4158 * If we don't have enough pinned space to deal with this
4159 * allocation, and no removed chunk in current transaction,
4160 * don't bother committing the transaction.
4162 have_pinned_space = __percpu_counter_compare(
4163 &data_sinfo->total_bytes_pinned,
4164 used + bytes - data_sinfo->total_bytes,
4165 BTRFS_TOTAL_BYTES_PINNED_BATCH);
4166 spin_unlock(&data_sinfo->lock);
4168 /* commit the current transaction and try again */
4173 if (need_commit > 0) {
4174 btrfs_start_delalloc_roots(fs_info, -1);
4175 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0,
4179 trans = btrfs_join_transaction(root);
4181 return PTR_ERR(trans);
4182 if (have_pinned_space >= 0 ||
4183 test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4184 &trans->transaction->flags) ||
4186 ret = btrfs_commit_transaction(trans);
4190 * The cleaner kthread might still be doing iput
4191 * operations. Wait for it to finish so that
4192 * more space is released. We don't need to
4193 * explicitly run the delayed iputs here because
4194 * the commit_transaction would have woken up
4197 ret = btrfs_wait_on_delayed_iputs(fs_info);
4202 btrfs_end_transaction(trans);
4206 trace_btrfs_space_reservation(fs_info,
4207 "space_info:enospc",
4208 data_sinfo->flags, bytes, 1);
4211 update_bytes_may_use(data_sinfo, bytes);
4212 trace_btrfs_space_reservation(fs_info, "space_info",
4213 data_sinfo->flags, bytes, 1);
4214 spin_unlock(&data_sinfo->lock);
4219 int btrfs_check_data_free_space(struct inode *inode,
4220 struct extent_changeset **reserved, u64 start, u64 len)
4222 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4225 /* align the range */
4226 len = round_up(start + len, fs_info->sectorsize) -
4227 round_down(start, fs_info->sectorsize);
4228 start = round_down(start, fs_info->sectorsize);
4230 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len);
4234 /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */
4235 ret = btrfs_qgroup_reserve_data(inode, reserved, start, len);
4237 btrfs_free_reserved_data_space_noquota(inode, start, len);
4244 * Called if we need to clear a data reservation for this inode
4245 * Normally in a error case.
4247 * This one will *NOT* use accurate qgroup reserved space API, just for case
4248 * which we can't sleep and is sure it won't affect qgroup reserved space.
4249 * Like clear_bit_hook().
4251 void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4254 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4255 struct btrfs_space_info *data_sinfo;
4257 /* Make sure the range is aligned to sectorsize */
4258 len = round_up(start + len, fs_info->sectorsize) -
4259 round_down(start, fs_info->sectorsize);
4260 start = round_down(start, fs_info->sectorsize);
4262 data_sinfo = fs_info->data_sinfo;
4263 spin_lock(&data_sinfo->lock);
4264 update_bytes_may_use(data_sinfo, -len);
4265 trace_btrfs_space_reservation(fs_info, "space_info",
4266 data_sinfo->flags, len, 0);
4267 spin_unlock(&data_sinfo->lock);
4271 * Called if we need to clear a data reservation for this inode
4272 * Normally in a error case.
4274 * This one will handle the per-inode data rsv map for accurate reserved
4277 void btrfs_free_reserved_data_space(struct inode *inode,
4278 struct extent_changeset *reserved, u64 start, u64 len)
4280 struct btrfs_root *root = BTRFS_I(inode)->root;
4282 /* Make sure the range is aligned to sectorsize */
4283 len = round_up(start + len, root->fs_info->sectorsize) -
4284 round_down(start, root->fs_info->sectorsize);
4285 start = round_down(start, root->fs_info->sectorsize);
4287 btrfs_free_reserved_data_space_noquota(inode, start, len);
4288 btrfs_qgroup_free_data(inode, reserved, start, len);
4291 static void force_metadata_allocation(struct btrfs_fs_info *info)
4293 struct list_head *head = &info->space_info;
4294 struct btrfs_space_info *found;
4297 list_for_each_entry_rcu(found, head, list) {
4298 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4299 found->force_alloc = CHUNK_ALLOC_FORCE;
4304 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4306 return (global->size << 1);
4309 static int should_alloc_chunk(struct btrfs_fs_info *fs_info,
4310 struct btrfs_space_info *sinfo, int force)
4312 u64 bytes_used = btrfs_space_info_used(sinfo, false);
4315 if (force == CHUNK_ALLOC_FORCE)
4319 * in limited mode, we want to have some free space up to
4320 * about 1% of the FS size.
4322 if (force == CHUNK_ALLOC_LIMITED) {
4323 thresh = btrfs_super_total_bytes(fs_info->super_copy);
4324 thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1));
4326 if (sinfo->total_bytes - bytes_used < thresh)
4330 if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8))
4335 static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type)
4339 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4340 BTRFS_BLOCK_GROUP_RAID0 |
4341 BTRFS_BLOCK_GROUP_RAID5 |
4342 BTRFS_BLOCK_GROUP_RAID6))
4343 num_dev = fs_info->fs_devices->rw_devices;
4344 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4347 num_dev = 1; /* DUP or single */
4353 * If @is_allocation is true, reserve space in the system space info necessary
4354 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4357 void check_system_chunk(struct btrfs_trans_handle *trans, u64 type)
4359 struct btrfs_fs_info *fs_info = trans->fs_info;
4360 struct btrfs_space_info *info;
4367 * Needed because we can end up allocating a system chunk and for an
4368 * atomic and race free space reservation in the chunk block reserve.
4370 lockdep_assert_held(&fs_info->chunk_mutex);
4372 info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4373 spin_lock(&info->lock);
4374 left = info->total_bytes - btrfs_space_info_used(info, true);
4375 spin_unlock(&info->lock);
4377 num_devs = get_profile_num_devs(fs_info, type);
4379 /* num_devs device items to update and 1 chunk item to add or remove */
4380 thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) +
4381 btrfs_calc_trans_metadata_size(fs_info, 1);
4383 if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4384 btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu",
4385 left, thresh, type);
4386 dump_space_info(fs_info, info, 0, 0);
4389 if (left < thresh) {
4390 u64 flags = btrfs_system_alloc_profile(fs_info);
4393 * Ignore failure to create system chunk. We might end up not
4394 * needing it, as we might not need to COW all nodes/leafs from
4395 * the paths we visit in the chunk tree (they were already COWed
4396 * or created in the current transaction for example).
4398 ret = btrfs_alloc_chunk(trans, flags);
4402 ret = btrfs_block_rsv_add(fs_info->chunk_root,
4403 &fs_info->chunk_block_rsv,
4404 thresh, BTRFS_RESERVE_NO_FLUSH);
4406 trans->chunk_bytes_reserved += thresh;
4411 * If force is CHUNK_ALLOC_FORCE:
4412 * - return 1 if it successfully allocates a chunk,
4413 * - return errors including -ENOSPC otherwise.
4414 * If force is NOT CHUNK_ALLOC_FORCE:
4415 * - return 0 if it doesn't need to allocate a new chunk,
4416 * - return 1 if it successfully allocates a chunk,
4417 * - return errors including -ENOSPC otherwise.
4419 static int do_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags,
4422 struct btrfs_fs_info *fs_info = trans->fs_info;
4423 struct btrfs_space_info *space_info;
4424 bool wait_for_alloc = false;
4425 bool should_alloc = false;
4428 /* Don't re-enter if we're already allocating a chunk */
4429 if (trans->allocating_chunk)
4432 space_info = __find_space_info(fs_info, flags);
4436 spin_lock(&space_info->lock);
4437 if (force < space_info->force_alloc)
4438 force = space_info->force_alloc;
4439 should_alloc = should_alloc_chunk(fs_info, space_info, force);
4440 if (space_info->full) {
4441 /* No more free physical space */
4446 spin_unlock(&space_info->lock);
4448 } else if (!should_alloc) {
4449 spin_unlock(&space_info->lock);
4451 } else if (space_info->chunk_alloc) {
4453 * Someone is already allocating, so we need to block
4454 * until this someone is finished and then loop to
4455 * recheck if we should continue with our allocation
4458 wait_for_alloc = true;
4459 spin_unlock(&space_info->lock);
4460 mutex_lock(&fs_info->chunk_mutex);
4461 mutex_unlock(&fs_info->chunk_mutex);
4463 /* Proceed with allocation */
4464 space_info->chunk_alloc = 1;
4465 wait_for_alloc = false;
4466 spin_unlock(&space_info->lock);
4470 } while (wait_for_alloc);
4472 mutex_lock(&fs_info->chunk_mutex);
4473 trans->allocating_chunk = true;
4476 * If we have mixed data/metadata chunks we want to make sure we keep
4477 * allocating mixed chunks instead of individual chunks.
4479 if (btrfs_mixed_space_info(space_info))
4480 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4483 * if we're doing a data chunk, go ahead and make sure that
4484 * we keep a reasonable number of metadata chunks allocated in the
4487 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4488 fs_info->data_chunk_allocations++;
4489 if (!(fs_info->data_chunk_allocations %
4490 fs_info->metadata_ratio))
4491 force_metadata_allocation(fs_info);
4495 * Check if we have enough space in SYSTEM chunk because we may need
4496 * to update devices.
4498 check_system_chunk(trans, flags);
4500 ret = btrfs_alloc_chunk(trans, flags);
4501 trans->allocating_chunk = false;
4503 spin_lock(&space_info->lock);
4506 space_info->full = 1;
4511 space_info->max_extent_size = 0;
4514 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4516 space_info->chunk_alloc = 0;
4517 spin_unlock(&space_info->lock);
4518 mutex_unlock(&fs_info->chunk_mutex);
4520 * When we allocate a new chunk we reserve space in the chunk block
4521 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4522 * add new nodes/leafs to it if we end up needing to do it when
4523 * inserting the chunk item and updating device items as part of the
4524 * second phase of chunk allocation, performed by
4525 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4526 * large number of new block groups to create in our transaction
4527 * handle's new_bgs list to avoid exhausting the chunk block reserve
4528 * in extreme cases - like having a single transaction create many new
4529 * block groups when starting to write out the free space caches of all
4530 * the block groups that were made dirty during the lifetime of the
4533 if (trans->chunk_bytes_reserved >= (u64)SZ_2M)
4534 btrfs_create_pending_block_groups(trans);
4539 static int can_overcommit(struct btrfs_fs_info *fs_info,
4540 struct btrfs_space_info *space_info, u64 bytes,
4541 enum btrfs_reserve_flush_enum flush,
4544 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4551 /* Don't overcommit when in mixed mode. */
4552 if (space_info->flags & BTRFS_BLOCK_GROUP_DATA)
4556 profile = btrfs_system_alloc_profile(fs_info);
4558 profile = btrfs_metadata_alloc_profile(fs_info);
4560 used = btrfs_space_info_used(space_info, false);
4563 * We only want to allow over committing if we have lots of actual space
4564 * free, but if we don't have enough space to handle the global reserve
4565 * space then we could end up having a real enospc problem when trying
4566 * to allocate a chunk or some other such important allocation.
4568 spin_lock(&global_rsv->lock);
4569 space_size = calc_global_rsv_need_space(global_rsv);
4570 spin_unlock(&global_rsv->lock);
4571 if (used + space_size >= space_info->total_bytes)
4574 used += space_info->bytes_may_use;
4576 avail = atomic64_read(&fs_info->free_chunk_space);
4579 * If we have dup, raid1 or raid10 then only half of the free
4580 * space is actually usable. For raid56, the space info used
4581 * doesn't include the parity drive, so we don't have to
4584 factor = btrfs_bg_type_to_factor(profile);
4585 avail = div_u64(avail, factor);
4588 * If we aren't flushing all things, let us overcommit up to
4589 * 1/2th of the space. If we can flush, don't let us overcommit
4590 * too much, let it overcommit up to 1/8 of the space.
4592 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4597 if (used + bytes < space_info->total_bytes + avail)
4602 static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info,
4603 unsigned long nr_pages, int nr_items)
4605 struct super_block *sb = fs_info->sb;
4607 if (down_read_trylock(&sb->s_umount)) {
4608 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4609 up_read(&sb->s_umount);
4612 * We needn't worry the filesystem going from r/w to r/o though
4613 * we don't acquire ->s_umount mutex, because the filesystem
4614 * should guarantee the delalloc inodes list be empty after
4615 * the filesystem is readonly(all dirty pages are written to
4618 btrfs_start_delalloc_roots(fs_info, nr_items);
4619 if (!current->journal_info)
4620 btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1);
4624 static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info,
4630 bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
4631 nr = div64_u64(to_reclaim, bytes);
4637 #define EXTENT_SIZE_PER_ITEM SZ_256K
4640 * shrink metadata reservation for delalloc
4642 static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim,
4643 u64 orig, bool wait_ordered)
4645 struct btrfs_space_info *space_info;
4646 struct btrfs_trans_handle *trans;
4652 unsigned long nr_pages;
4655 /* Calc the number of the pages we need flush for space reservation */
4656 items = calc_reclaim_items_nr(fs_info, to_reclaim);
4657 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4659 trans = (struct btrfs_trans_handle *)current->journal_info;
4660 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4662 delalloc_bytes = percpu_counter_sum_positive(
4663 &fs_info->delalloc_bytes);
4664 dio_bytes = percpu_counter_sum_positive(&fs_info->dio_bytes);
4665 if (delalloc_bytes == 0 && dio_bytes == 0) {
4669 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4674 * If we are doing more ordered than delalloc we need to just wait on
4675 * ordered extents, otherwise we'll waste time trying to flush delalloc
4676 * that likely won't give us the space back we need.
4678 if (dio_bytes > delalloc_bytes)
4679 wait_ordered = true;
4682 while ((delalloc_bytes || dio_bytes) && loops < 3) {
4683 nr_pages = min(delalloc_bytes, to_reclaim) >> PAGE_SHIFT;
4686 * Triggers inode writeback for up to nr_pages. This will invoke
4687 * ->writepages callback and trigger delalloc filling
4688 * (btrfs_run_delalloc_range()).
4690 btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items);
4693 * We need to wait for the compressed pages to start before
4696 async_pages = atomic_read(&fs_info->async_delalloc_pages);
4701 * Calculate how many compressed pages we want to be written
4702 * before we continue. I.e if there are more async pages than we
4703 * require wait_event will wait until nr_pages are written.
4705 if (async_pages <= nr_pages)
4708 async_pages -= nr_pages;
4710 wait_event(fs_info->async_submit_wait,
4711 atomic_read(&fs_info->async_delalloc_pages) <=
4714 spin_lock(&space_info->lock);
4715 if (list_empty(&space_info->tickets) &&
4716 list_empty(&space_info->priority_tickets)) {
4717 spin_unlock(&space_info->lock);
4720 spin_unlock(&space_info->lock);
4723 if (wait_ordered && !trans) {
4724 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4726 time_left = schedule_timeout_killable(1);
4730 delalloc_bytes = percpu_counter_sum_positive(
4731 &fs_info->delalloc_bytes);
4732 dio_bytes = percpu_counter_sum_positive(&fs_info->dio_bytes);
4736 struct reserve_ticket {
4740 struct list_head list;
4741 wait_queue_head_t wait;
4745 * maybe_commit_transaction - possibly commit the transaction if its ok to
4746 * @root - the root we're allocating for
4747 * @bytes - the number of bytes we want to reserve
4748 * @force - force the commit
4750 * This will check to make sure that committing the transaction will actually
4751 * get us somewhere and then commit the transaction if it does. Otherwise it
4752 * will return -ENOSPC.
4754 static int may_commit_transaction(struct btrfs_fs_info *fs_info,
4755 struct btrfs_space_info *space_info)
4757 struct reserve_ticket *ticket = NULL;
4758 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv;
4759 struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv;
4760 struct btrfs_trans_handle *trans;
4762 u64 reclaim_bytes = 0;
4764 trans = (struct btrfs_trans_handle *)current->journal_info;
4768 spin_lock(&space_info->lock);
4769 if (!list_empty(&space_info->priority_tickets))
4770 ticket = list_first_entry(&space_info->priority_tickets,
4771 struct reserve_ticket, list);
4772 else if (!list_empty(&space_info->tickets))
4773 ticket = list_first_entry(&space_info->tickets,
4774 struct reserve_ticket, list);
4775 bytes_needed = (ticket) ? ticket->bytes : 0;
4776 spin_unlock(&space_info->lock);
4781 trans = btrfs_join_transaction(fs_info->extent_root);
4783 return PTR_ERR(trans);
4786 * See if there is enough pinned space to make this reservation, or if
4787 * we have block groups that are going to be freed, allowing us to
4788 * possibly do a chunk allocation the next loop through.
4790 if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags) ||
4791 __percpu_counter_compare(&space_info->total_bytes_pinned,
4793 BTRFS_TOTAL_BYTES_PINNED_BATCH) >= 0)
4797 * See if there is some space in the delayed insertion reservation for
4800 if (space_info != delayed_rsv->space_info)
4803 spin_lock(&delayed_rsv->lock);
4804 reclaim_bytes += delayed_rsv->reserved;
4805 spin_unlock(&delayed_rsv->lock);
4807 spin_lock(&delayed_refs_rsv->lock);
4808 reclaim_bytes += delayed_refs_rsv->reserved;
4809 spin_unlock(&delayed_refs_rsv->lock);
4810 if (reclaim_bytes >= bytes_needed)
4812 bytes_needed -= reclaim_bytes;
4814 if (__percpu_counter_compare(&space_info->total_bytes_pinned,
4816 BTRFS_TOTAL_BYTES_PINNED_BATCH) < 0)
4820 return btrfs_commit_transaction(trans);
4822 btrfs_end_transaction(trans);
4827 * Try to flush some data based on policy set by @state. This is only advisory
4828 * and may fail for various reasons. The caller is supposed to examine the
4829 * state of @space_info to detect the outcome.
4831 static void flush_space(struct btrfs_fs_info *fs_info,
4832 struct btrfs_space_info *space_info, u64 num_bytes,
4835 struct btrfs_root *root = fs_info->extent_root;
4836 struct btrfs_trans_handle *trans;
4841 case FLUSH_DELAYED_ITEMS_NR:
4842 case FLUSH_DELAYED_ITEMS:
4843 if (state == FLUSH_DELAYED_ITEMS_NR)
4844 nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2;
4848 trans = btrfs_join_transaction(root);
4849 if (IS_ERR(trans)) {
4850 ret = PTR_ERR(trans);
4853 ret = btrfs_run_delayed_items_nr(trans, nr);
4854 btrfs_end_transaction(trans);
4856 case FLUSH_DELALLOC:
4857 case FLUSH_DELALLOC_WAIT:
4858 shrink_delalloc(fs_info, num_bytes * 2, num_bytes,
4859 state == FLUSH_DELALLOC_WAIT);
4861 case FLUSH_DELAYED_REFS_NR:
4862 case FLUSH_DELAYED_REFS:
4863 trans = btrfs_join_transaction(root);
4864 if (IS_ERR(trans)) {
4865 ret = PTR_ERR(trans);
4868 if (state == FLUSH_DELAYED_REFS_NR)
4869 nr = calc_reclaim_items_nr(fs_info, num_bytes);
4872 btrfs_run_delayed_refs(trans, nr);
4873 btrfs_end_transaction(trans);
4876 case ALLOC_CHUNK_FORCE:
4877 trans = btrfs_join_transaction(root);
4878 if (IS_ERR(trans)) {
4879 ret = PTR_ERR(trans);
4882 ret = do_chunk_alloc(trans,
4883 btrfs_metadata_alloc_profile(fs_info),
4884 (state == ALLOC_CHUNK) ?
4885 CHUNK_ALLOC_NO_FORCE : CHUNK_ALLOC_FORCE);
4886 btrfs_end_transaction(trans);
4887 if (ret > 0 || ret == -ENOSPC)
4892 * If we have pending delayed iputs then we could free up a
4893 * bunch of pinned space, so make sure we run the iputs before
4894 * we do our pinned bytes check below.
4896 btrfs_run_delayed_iputs(fs_info);
4897 btrfs_wait_on_delayed_iputs(fs_info);
4899 ret = may_commit_transaction(fs_info, space_info);
4906 trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, state,
4912 btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info,
4913 struct btrfs_space_info *space_info,
4916 struct reserve_ticket *ticket;
4921 list_for_each_entry(ticket, &space_info->tickets, list)
4922 to_reclaim += ticket->bytes;
4923 list_for_each_entry(ticket, &space_info->priority_tickets, list)
4924 to_reclaim += ticket->bytes;
4928 to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M);
4929 if (can_overcommit(fs_info, space_info, to_reclaim,
4930 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
4933 used = btrfs_space_info_used(space_info, true);
4935 if (can_overcommit(fs_info, space_info, SZ_1M,
4936 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
4937 expected = div_factor_fine(space_info->total_bytes, 95);
4939 expected = div_factor_fine(space_info->total_bytes, 90);
4941 if (used > expected)
4942 to_reclaim = used - expected;
4945 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4946 space_info->bytes_reserved);
4950 static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info,
4951 struct btrfs_space_info *space_info,
4952 u64 used, bool system_chunk)
4954 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
4956 /* If we're just plain full then async reclaim just slows us down. */
4957 if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh)
4960 if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info,
4964 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
4965 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4968 static bool wake_all_tickets(struct list_head *head)
4970 struct reserve_ticket *ticket;
4972 while (!list_empty(head)) {
4973 ticket = list_first_entry(head, struct reserve_ticket, list);
4974 list_del_init(&ticket->list);
4975 ticket->error = -ENOSPC;
4976 wake_up(&ticket->wait);
4977 if (ticket->bytes != ticket->orig_bytes)
4984 * This is for normal flushers, we can wait all goddamned day if we want to. We
4985 * will loop and continuously try to flush as long as we are making progress.
4986 * We count progress as clearing off tickets each time we have to loop.
4988 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4990 struct btrfs_fs_info *fs_info;
4991 struct btrfs_space_info *space_info;
4994 int commit_cycles = 0;
4995 u64 last_tickets_id;
4997 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4998 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5000 spin_lock(&space_info->lock);
5001 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5004 space_info->flush = 0;
5005 spin_unlock(&space_info->lock);
5008 last_tickets_id = space_info->tickets_id;
5009 spin_unlock(&space_info->lock);
5011 flush_state = FLUSH_DELAYED_ITEMS_NR;
5013 flush_space(fs_info, space_info, to_reclaim, flush_state);
5014 spin_lock(&space_info->lock);
5015 if (list_empty(&space_info->tickets)) {
5016 space_info->flush = 0;
5017 spin_unlock(&space_info->lock);
5020 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info,
5023 if (last_tickets_id == space_info->tickets_id) {
5026 last_tickets_id = space_info->tickets_id;
5027 flush_state = FLUSH_DELAYED_ITEMS_NR;
5033 * We don't want to force a chunk allocation until we've tried
5034 * pretty hard to reclaim space. Think of the case where we
5035 * freed up a bunch of space and so have a lot of pinned space
5036 * to reclaim. We would rather use that than possibly create a
5037 * underutilized metadata chunk. So if this is our first run
5038 * through the flushing state machine skip ALLOC_CHUNK_FORCE and
5039 * commit the transaction. If nothing has changed the next go
5040 * around then we can force a chunk allocation.
5042 if (flush_state == ALLOC_CHUNK_FORCE && !commit_cycles)
5045 if (flush_state > COMMIT_TRANS) {
5047 if (commit_cycles > 2) {
5048 if (wake_all_tickets(&space_info->tickets)) {
5049 flush_state = FLUSH_DELAYED_ITEMS_NR;
5052 space_info->flush = 0;
5055 flush_state = FLUSH_DELAYED_ITEMS_NR;
5058 spin_unlock(&space_info->lock);
5059 } while (flush_state <= COMMIT_TRANS);
5062 void btrfs_init_async_reclaim_work(struct work_struct *work)
5064 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
5067 static const enum btrfs_flush_state priority_flush_states[] = {
5068 FLUSH_DELAYED_ITEMS_NR,
5069 FLUSH_DELAYED_ITEMS,
5073 static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
5074 struct btrfs_space_info *space_info,
5075 struct reserve_ticket *ticket)
5080 spin_lock(&space_info->lock);
5081 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5084 spin_unlock(&space_info->lock);
5087 spin_unlock(&space_info->lock);
5091 flush_space(fs_info, space_info, to_reclaim,
5092 priority_flush_states[flush_state]);
5094 spin_lock(&space_info->lock);
5095 if (ticket->bytes == 0) {
5096 spin_unlock(&space_info->lock);
5099 spin_unlock(&space_info->lock);
5100 } while (flush_state < ARRAY_SIZE(priority_flush_states));
5103 static int wait_reserve_ticket(struct btrfs_fs_info *fs_info,
5104 struct btrfs_space_info *space_info,
5105 struct reserve_ticket *ticket)
5109 u64 reclaim_bytes = 0;
5112 spin_lock(&space_info->lock);
5113 while (ticket->bytes > 0 && ticket->error == 0) {
5114 ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE);
5119 spin_unlock(&space_info->lock);
5123 finish_wait(&ticket->wait, &wait);
5124 spin_lock(&space_info->lock);
5127 ret = ticket->error;
5128 if (!list_empty(&ticket->list))
5129 list_del_init(&ticket->list);
5130 if (ticket->bytes && ticket->bytes < ticket->orig_bytes)
5131 reclaim_bytes = ticket->orig_bytes - ticket->bytes;
5132 spin_unlock(&space_info->lock);
5135 space_info_add_old_bytes(fs_info, space_info, reclaim_bytes);
5140 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5141 * @root - the root we're allocating for
5142 * @space_info - the space info we want to allocate from
5143 * @orig_bytes - the number of bytes we want
5144 * @flush - whether or not we can flush to make our reservation
5146 * This will reserve orig_bytes number of bytes from the space info associated
5147 * with the block_rsv. If there is not enough space it will make an attempt to
5148 * flush out space to make room. It will do this by flushing delalloc if
5149 * possible or committing the transaction. If flush is 0 then no attempts to
5150 * regain reservations will be made and this will fail if there is not enough
5153 static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info,
5154 struct btrfs_space_info *space_info,
5156 enum btrfs_reserve_flush_enum flush,
5159 struct reserve_ticket ticket;
5161 u64 reclaim_bytes = 0;
5165 ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL);
5167 spin_lock(&space_info->lock);
5169 used = btrfs_space_info_used(space_info, true);
5172 * If we have enough space then hooray, make our reservation and carry
5173 * on. If not see if we can overcommit, and if we can, hooray carry on.
5174 * If not things get more complicated.
5176 if (used + orig_bytes <= space_info->total_bytes) {
5177 update_bytes_may_use(space_info, orig_bytes);
5178 trace_btrfs_space_reservation(fs_info, "space_info",
5179 space_info->flags, orig_bytes, 1);
5181 } else if (can_overcommit(fs_info, space_info, orig_bytes, flush,
5183 update_bytes_may_use(space_info, orig_bytes);
5184 trace_btrfs_space_reservation(fs_info, "space_info",
5185 space_info->flags, orig_bytes, 1);
5190 * If we couldn't make a reservation then setup our reservation ticket
5191 * and kick the async worker if it's not already running.
5193 * If we are a priority flusher then we just need to add our ticket to
5194 * the list and we will do our own flushing further down.
5196 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
5197 ticket.orig_bytes = orig_bytes;
5198 ticket.bytes = orig_bytes;
5200 init_waitqueue_head(&ticket.wait);
5201 if (flush == BTRFS_RESERVE_FLUSH_ALL) {
5202 list_add_tail(&ticket.list, &space_info->tickets);
5203 if (!space_info->flush) {
5204 space_info->flush = 1;
5205 trace_btrfs_trigger_flush(fs_info,
5209 queue_work(system_unbound_wq,
5210 &fs_info->async_reclaim_work);
5213 list_add_tail(&ticket.list,
5214 &space_info->priority_tickets);
5216 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
5219 * We will do the space reservation dance during log replay,
5220 * which means we won't have fs_info->fs_root set, so don't do
5221 * the async reclaim as we will panic.
5223 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) &&
5224 need_do_async_reclaim(fs_info, space_info,
5225 used, system_chunk) &&
5226 !work_busy(&fs_info->async_reclaim_work)) {
5227 trace_btrfs_trigger_flush(fs_info, space_info->flags,
5228 orig_bytes, flush, "preempt");
5229 queue_work(system_unbound_wq,
5230 &fs_info->async_reclaim_work);
5233 spin_unlock(&space_info->lock);
5234 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
5237 if (flush == BTRFS_RESERVE_FLUSH_ALL)
5238 return wait_reserve_ticket(fs_info, space_info, &ticket);
5241 priority_reclaim_metadata_space(fs_info, space_info, &ticket);
5242 spin_lock(&space_info->lock);
5244 if (ticket.bytes < orig_bytes)
5245 reclaim_bytes = orig_bytes - ticket.bytes;
5246 list_del_init(&ticket.list);
5249 spin_unlock(&space_info->lock);
5252 space_info_add_old_bytes(fs_info, space_info, reclaim_bytes);
5253 ASSERT(list_empty(&ticket.list));
5258 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5259 * @root - the root we're allocating for
5260 * @block_rsv - the block_rsv we're allocating for
5261 * @orig_bytes - the number of bytes we want
5262 * @flush - whether or not we can flush to make our reservation
5264 * This will reserve orig_bytes number of bytes from the space info associated
5265 * with the block_rsv. If there is not enough space it will make an attempt to
5266 * flush out space to make room. It will do this by flushing delalloc if
5267 * possible or committing the transaction. If flush is 0 then no attempts to
5268 * regain reservations will be made and this will fail if there is not enough
5271 static int reserve_metadata_bytes(struct btrfs_root *root,
5272 struct btrfs_block_rsv *block_rsv,
5274 enum btrfs_reserve_flush_enum flush)
5276 struct btrfs_fs_info *fs_info = root->fs_info;
5277 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5279 bool system_chunk = (root == fs_info->chunk_root);
5281 ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info,
5282 orig_bytes, flush, system_chunk);
5283 if (ret == -ENOSPC &&
5284 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5285 if (block_rsv != global_rsv &&
5286 !block_rsv_use_bytes(global_rsv, orig_bytes))
5289 if (ret == -ENOSPC) {
5290 trace_btrfs_space_reservation(fs_info, "space_info:enospc",
5291 block_rsv->space_info->flags,
5294 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG))
5295 dump_space_info(fs_info, block_rsv->space_info,
5301 static struct btrfs_block_rsv *get_block_rsv(
5302 const struct btrfs_trans_handle *trans,
5303 const struct btrfs_root *root)
5305 struct btrfs_fs_info *fs_info = root->fs_info;
5306 struct btrfs_block_rsv *block_rsv = NULL;
5308 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5309 (root == fs_info->csum_root && trans->adding_csums) ||
5310 (root == fs_info->uuid_root))
5311 block_rsv = trans->block_rsv;
5314 block_rsv = root->block_rsv;
5317 block_rsv = &fs_info->empty_block_rsv;
5322 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5326 spin_lock(&block_rsv->lock);
5327 if (block_rsv->reserved >= num_bytes) {
5328 block_rsv->reserved -= num_bytes;
5329 if (block_rsv->reserved < block_rsv->size)
5330 block_rsv->full = 0;
5333 spin_unlock(&block_rsv->lock);
5337 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5338 u64 num_bytes, bool update_size)
5340 spin_lock(&block_rsv->lock);
5341 block_rsv->reserved += num_bytes;
5343 block_rsv->size += num_bytes;
5344 else if (block_rsv->reserved >= block_rsv->size)
5345 block_rsv->full = 1;
5346 spin_unlock(&block_rsv->lock);
5349 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5350 struct btrfs_block_rsv *dest, u64 num_bytes,
5353 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5356 if (global_rsv->space_info != dest->space_info)
5359 spin_lock(&global_rsv->lock);
5360 min_bytes = div_factor(global_rsv->size, min_factor);
5361 if (global_rsv->reserved < min_bytes + num_bytes) {
5362 spin_unlock(&global_rsv->lock);
5365 global_rsv->reserved -= num_bytes;
5366 if (global_rsv->reserved < global_rsv->size)
5367 global_rsv->full = 0;
5368 spin_unlock(&global_rsv->lock);
5370 block_rsv_add_bytes(dest, num_bytes, true);
5375 * btrfs_migrate_to_delayed_refs_rsv - transfer bytes to our delayed refs rsv.
5376 * @fs_info - the fs info for our fs.
5377 * @src - the source block rsv to transfer from.
5378 * @num_bytes - the number of bytes to transfer.
5380 * This transfers up to the num_bytes amount from the src rsv to the
5381 * delayed_refs_rsv. Any extra bytes are returned to the space info.
5383 void btrfs_migrate_to_delayed_refs_rsv(struct btrfs_fs_info *fs_info,
5384 struct btrfs_block_rsv *src,
5387 struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv;
5390 spin_lock(&src->lock);
5391 src->reserved -= num_bytes;
5392 src->size -= num_bytes;
5393 spin_unlock(&src->lock);
5395 spin_lock(&delayed_refs_rsv->lock);
5396 if (delayed_refs_rsv->size > delayed_refs_rsv->reserved) {
5397 u64 delta = delayed_refs_rsv->size -
5398 delayed_refs_rsv->reserved;
5399 if (num_bytes > delta) {
5400 to_free = num_bytes - delta;
5404 to_free = num_bytes;
5409 delayed_refs_rsv->reserved += num_bytes;
5410 if (delayed_refs_rsv->reserved >= delayed_refs_rsv->size)
5411 delayed_refs_rsv->full = 1;
5412 spin_unlock(&delayed_refs_rsv->lock);
5415 trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv",
5418 space_info_add_old_bytes(fs_info, delayed_refs_rsv->space_info,
5423 * btrfs_delayed_refs_rsv_refill - refill based on our delayed refs usage.
5424 * @fs_info - the fs_info for our fs.
5425 * @flush - control how we can flush for this reservation.
5427 * This will refill the delayed block_rsv up to 1 items size worth of space and
5428 * will return -ENOSPC if we can't make the reservation.
5430 int btrfs_delayed_refs_rsv_refill(struct btrfs_fs_info *fs_info,
5431 enum btrfs_reserve_flush_enum flush)
5433 struct btrfs_block_rsv *block_rsv = &fs_info->delayed_refs_rsv;
5434 u64 limit = btrfs_calc_trans_metadata_size(fs_info, 1);
5438 spin_lock(&block_rsv->lock);
5439 if (block_rsv->reserved < block_rsv->size) {
5440 num_bytes = block_rsv->size - block_rsv->reserved;
5441 num_bytes = min(num_bytes, limit);
5443 spin_unlock(&block_rsv->lock);
5448 ret = reserve_metadata_bytes(fs_info->extent_root, block_rsv,
5452 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5453 trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv",
5459 * This is for space we already have accounted in space_info->bytes_may_use, so
5460 * basically when we're returning space from block_rsv's.
5462 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
5463 struct btrfs_space_info *space_info,
5466 struct reserve_ticket *ticket;
5467 struct list_head *head;
5469 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH;
5470 bool check_overcommit = false;
5472 spin_lock(&space_info->lock);
5473 head = &space_info->priority_tickets;
5476 * If we are over our limit then we need to check and see if we can
5477 * overcommit, and if we can't then we just need to free up our space
5478 * and not satisfy any requests.
5480 used = btrfs_space_info_used(space_info, true);
5481 if (used - num_bytes >= space_info->total_bytes)
5482 check_overcommit = true;
5484 while (!list_empty(head) && num_bytes) {
5485 ticket = list_first_entry(head, struct reserve_ticket,
5488 * We use 0 bytes because this space is already reserved, so
5489 * adding the ticket space would be a double count.
5491 if (check_overcommit &&
5492 !can_overcommit(fs_info, space_info, 0, flush, false))
5494 if (num_bytes >= ticket->bytes) {
5495 list_del_init(&ticket->list);
5496 num_bytes -= ticket->bytes;
5498 space_info->tickets_id++;
5499 wake_up(&ticket->wait);
5501 ticket->bytes -= num_bytes;
5506 if (num_bytes && head == &space_info->priority_tickets) {
5507 head = &space_info->tickets;
5508 flush = BTRFS_RESERVE_FLUSH_ALL;
5511 update_bytes_may_use(space_info, -num_bytes);
5512 trace_btrfs_space_reservation(fs_info, "space_info",
5513 space_info->flags, num_bytes, 0);
5514 spin_unlock(&space_info->lock);
5518 * This is for newly allocated space that isn't accounted in
5519 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5520 * we use this helper.
5522 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
5523 struct btrfs_space_info *space_info,
5526 struct reserve_ticket *ticket;
5527 struct list_head *head = &space_info->priority_tickets;
5530 while (!list_empty(head) && num_bytes) {
5531 ticket = list_first_entry(head, struct reserve_ticket,
5533 if (num_bytes >= ticket->bytes) {
5534 trace_btrfs_space_reservation(fs_info, "space_info",
5537 list_del_init(&ticket->list);
5538 num_bytes -= ticket->bytes;
5539 update_bytes_may_use(space_info, ticket->bytes);
5541 space_info->tickets_id++;
5542 wake_up(&ticket->wait);
5544 trace_btrfs_space_reservation(fs_info, "space_info",
5547 update_bytes_may_use(space_info, num_bytes);
5548 ticket->bytes -= num_bytes;
5553 if (num_bytes && head == &space_info->priority_tickets) {
5554 head = &space_info->tickets;
5559 static u64 block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5560 struct btrfs_block_rsv *block_rsv,
5561 struct btrfs_block_rsv *dest, u64 num_bytes,
5562 u64 *qgroup_to_release_ret)
5564 struct btrfs_space_info *space_info = block_rsv->space_info;
5565 u64 qgroup_to_release = 0;
5568 spin_lock(&block_rsv->lock);
5569 if (num_bytes == (u64)-1) {
5570 num_bytes = block_rsv->size;
5571 qgroup_to_release = block_rsv->qgroup_rsv_size;
5573 block_rsv->size -= num_bytes;
5574 if (block_rsv->reserved >= block_rsv->size) {
5575 num_bytes = block_rsv->reserved - block_rsv->size;
5576 block_rsv->reserved = block_rsv->size;
5577 block_rsv->full = 1;
5581 if (block_rsv->qgroup_rsv_reserved >= block_rsv->qgroup_rsv_size) {
5582 qgroup_to_release = block_rsv->qgroup_rsv_reserved -
5583 block_rsv->qgroup_rsv_size;
5584 block_rsv->qgroup_rsv_reserved = block_rsv->qgroup_rsv_size;
5586 qgroup_to_release = 0;
5588 spin_unlock(&block_rsv->lock);
5591 if (num_bytes > 0) {
5593 spin_lock(&dest->lock);
5597 bytes_to_add = dest->size - dest->reserved;
5598 bytes_to_add = min(num_bytes, bytes_to_add);
5599 dest->reserved += bytes_to_add;
5600 if (dest->reserved >= dest->size)
5602 num_bytes -= bytes_to_add;
5604 spin_unlock(&dest->lock);
5607 space_info_add_old_bytes(fs_info, space_info,
5610 if (qgroup_to_release_ret)
5611 *qgroup_to_release_ret = qgroup_to_release;
5615 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
5616 struct btrfs_block_rsv *dst, u64 num_bytes,
5621 ret = block_rsv_use_bytes(src, num_bytes);
5625 block_rsv_add_bytes(dst, num_bytes, update_size);
5629 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5631 memset(rsv, 0, sizeof(*rsv));
5632 spin_lock_init(&rsv->lock);
5636 void btrfs_init_metadata_block_rsv(struct btrfs_fs_info *fs_info,
5637 struct btrfs_block_rsv *rsv,
5638 unsigned short type)
5640 btrfs_init_block_rsv(rsv, type);
5641 rsv->space_info = __find_space_info(fs_info,
5642 BTRFS_BLOCK_GROUP_METADATA);
5645 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info,
5646 unsigned short type)
5648 struct btrfs_block_rsv *block_rsv;
5650 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5654 btrfs_init_metadata_block_rsv(fs_info, block_rsv, type);
5658 void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info,
5659 struct btrfs_block_rsv *rsv)
5663 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
5667 int btrfs_block_rsv_add(struct btrfs_root *root,
5668 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5669 enum btrfs_reserve_flush_enum flush)
5676 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5678 block_rsv_add_bytes(block_rsv, num_bytes, true);
5683 int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor)
5691 spin_lock(&block_rsv->lock);
5692 num_bytes = div_factor(block_rsv->size, min_factor);
5693 if (block_rsv->reserved >= num_bytes)
5695 spin_unlock(&block_rsv->lock);
5700 int btrfs_block_rsv_refill(struct btrfs_root *root,
5701 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5702 enum btrfs_reserve_flush_enum flush)
5710 spin_lock(&block_rsv->lock);
5711 num_bytes = min_reserved;
5712 if (block_rsv->reserved >= num_bytes)
5715 num_bytes -= block_rsv->reserved;
5716 spin_unlock(&block_rsv->lock);
5721 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5723 block_rsv_add_bytes(block_rsv, num_bytes, false);
5730 static u64 __btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
5731 struct btrfs_block_rsv *block_rsv,
5732 u64 num_bytes, u64 *qgroup_to_release)
5734 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5735 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_refs_rsv;
5736 struct btrfs_block_rsv *target = delayed_rsv;
5738 if (target->full || target == block_rsv)
5739 target = global_rsv;
5741 if (block_rsv->space_info != target->space_info)
5744 return block_rsv_release_bytes(fs_info, block_rsv, target, num_bytes,
5748 void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
5749 struct btrfs_block_rsv *block_rsv,
5752 __btrfs_block_rsv_release(fs_info, block_rsv, num_bytes, NULL);
5756 * btrfs_inode_rsv_release - release any excessive reservation.
5757 * @inode - the inode we need to release from.
5758 * @qgroup_free - free or convert qgroup meta.
5759 * Unlike normal operation, qgroup meta reservation needs to know if we are
5760 * freeing qgroup reservation or just converting it into per-trans. Normally
5761 * @qgroup_free is true for error handling, and false for normal release.
5763 * This is the same as btrfs_block_rsv_release, except that it handles the
5764 * tracepoint for the reservation.
5766 static void btrfs_inode_rsv_release(struct btrfs_inode *inode, bool qgroup_free)
5768 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5769 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5771 u64 qgroup_to_release = 0;
5774 * Since we statically set the block_rsv->size we just want to say we
5775 * are releasing 0 bytes, and then we'll just get the reservation over
5778 released = __btrfs_block_rsv_release(fs_info, block_rsv, 0,
5779 &qgroup_to_release);
5781 trace_btrfs_space_reservation(fs_info, "delalloc",
5782 btrfs_ino(inode), released, 0);
5784 btrfs_qgroup_free_meta_prealloc(inode->root, qgroup_to_release);
5786 btrfs_qgroup_convert_reserved_meta(inode->root,
5791 * btrfs_delayed_refs_rsv_release - release a ref head's reservation.
5792 * @fs_info - the fs_info for our fs.
5793 * @nr - the number of items to drop.
5795 * This drops the delayed ref head's count from the delayed refs rsv and frees
5796 * any excess reservation we had.
5798 void btrfs_delayed_refs_rsv_release(struct btrfs_fs_info *fs_info, int nr)
5800 struct btrfs_block_rsv *block_rsv = &fs_info->delayed_refs_rsv;
5801 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5802 u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, nr);
5805 released = block_rsv_release_bytes(fs_info, block_rsv, global_rsv,
5808 trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv",
5812 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5814 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5815 struct btrfs_space_info *sinfo = block_rsv->space_info;
5819 * The global block rsv is based on the size of the extent tree, the
5820 * checksum tree and the root tree. If the fs is empty we want to set
5821 * it to a minimal amount for safety.
5823 num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) +
5824 btrfs_root_used(&fs_info->csum_root->root_item) +
5825 btrfs_root_used(&fs_info->tree_root->root_item);
5826 num_bytes = max_t(u64, num_bytes, SZ_16M);
5828 spin_lock(&sinfo->lock);
5829 spin_lock(&block_rsv->lock);
5831 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
5833 if (block_rsv->reserved < block_rsv->size) {
5834 num_bytes = btrfs_space_info_used(sinfo, true);
5835 if (sinfo->total_bytes > num_bytes) {
5836 num_bytes = sinfo->total_bytes - num_bytes;
5837 num_bytes = min(num_bytes,
5838 block_rsv->size - block_rsv->reserved);
5839 block_rsv->reserved += num_bytes;
5840 update_bytes_may_use(sinfo, num_bytes);
5841 trace_btrfs_space_reservation(fs_info, "space_info",
5842 sinfo->flags, num_bytes,
5845 } else if (block_rsv->reserved > block_rsv->size) {
5846 num_bytes = block_rsv->reserved - block_rsv->size;
5847 update_bytes_may_use(sinfo, -num_bytes);
5848 trace_btrfs_space_reservation(fs_info, "space_info",
5849 sinfo->flags, num_bytes, 0);
5850 block_rsv->reserved = block_rsv->size;
5853 if (block_rsv->reserved == block_rsv->size)
5854 block_rsv->full = 1;
5856 block_rsv->full = 0;
5858 spin_unlock(&block_rsv->lock);
5859 spin_unlock(&sinfo->lock);
5862 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5864 struct btrfs_space_info *space_info;
5866 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5867 fs_info->chunk_block_rsv.space_info = space_info;
5869 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5870 fs_info->global_block_rsv.space_info = space_info;
5871 fs_info->trans_block_rsv.space_info = space_info;
5872 fs_info->empty_block_rsv.space_info = space_info;
5873 fs_info->delayed_block_rsv.space_info = space_info;
5874 fs_info->delayed_refs_rsv.space_info = space_info;
5876 fs_info->extent_root->block_rsv = &fs_info->delayed_refs_rsv;
5877 fs_info->csum_root->block_rsv = &fs_info->delayed_refs_rsv;
5878 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5879 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5880 if (fs_info->quota_root)
5881 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5882 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5884 update_global_block_rsv(fs_info);
5887 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5889 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5891 WARN_ON(fs_info->trans_block_rsv.size > 0);
5892 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5893 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5894 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5895 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5896 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5897 WARN_ON(fs_info->delayed_refs_rsv.reserved > 0);
5898 WARN_ON(fs_info->delayed_refs_rsv.size > 0);
5902 * btrfs_update_delayed_refs_rsv - adjust the size of the delayed refs rsv
5903 * @trans - the trans that may have generated delayed refs
5905 * This is to be called anytime we may have adjusted trans->delayed_ref_updates,
5906 * it'll calculate the additional size and add it to the delayed_refs_rsv.
5908 void btrfs_update_delayed_refs_rsv(struct btrfs_trans_handle *trans)
5910 struct btrfs_fs_info *fs_info = trans->fs_info;
5911 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_refs_rsv;
5914 if (!trans->delayed_ref_updates)
5917 num_bytes = btrfs_calc_trans_metadata_size(fs_info,
5918 trans->delayed_ref_updates);
5919 spin_lock(&delayed_rsv->lock);
5920 delayed_rsv->size += num_bytes;
5921 delayed_rsv->full = 0;
5922 spin_unlock(&delayed_rsv->lock);
5923 trans->delayed_ref_updates = 0;
5927 * To be called after all the new block groups attached to the transaction
5928 * handle have been created (btrfs_create_pending_block_groups()).
5930 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5932 struct btrfs_fs_info *fs_info = trans->fs_info;
5934 if (!trans->chunk_bytes_reserved)
5937 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5939 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5940 trans->chunk_bytes_reserved, NULL);
5941 trans->chunk_bytes_reserved = 0;
5945 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5946 * root: the root of the parent directory
5947 * rsv: block reservation
5948 * items: the number of items that we need do reservation
5949 * use_global_rsv: allow fallback to the global block reservation
5951 * This function is used to reserve the space for snapshot/subvolume
5952 * creation and deletion. Those operations are different with the
5953 * common file/directory operations, they change two fs/file trees
5954 * and root tree, the number of items that the qgroup reserves is
5955 * different with the free space reservation. So we can not use
5956 * the space reservation mechanism in start_transaction().
5958 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5959 struct btrfs_block_rsv *rsv, int items,
5960 bool use_global_rsv)
5962 u64 qgroup_num_bytes = 0;
5965 struct btrfs_fs_info *fs_info = root->fs_info;
5966 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5968 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
5969 /* One for parent inode, two for dir entries */
5970 qgroup_num_bytes = 3 * fs_info->nodesize;
5971 ret = btrfs_qgroup_reserve_meta_prealloc(root,
5972 qgroup_num_bytes, true);
5977 num_bytes = btrfs_calc_trans_metadata_size(fs_info, items);
5978 rsv->space_info = __find_space_info(fs_info,
5979 BTRFS_BLOCK_GROUP_METADATA);
5980 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5981 BTRFS_RESERVE_FLUSH_ALL);
5983 if (ret == -ENOSPC && use_global_rsv)
5984 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, true);
5986 if (ret && qgroup_num_bytes)
5987 btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes);
5992 void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info,
5993 struct btrfs_block_rsv *rsv)
5995 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
5998 static void btrfs_calculate_inode_block_rsv_size(struct btrfs_fs_info *fs_info,
5999 struct btrfs_inode *inode)
6001 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
6002 u64 reserve_size = 0;
6003 u64 qgroup_rsv_size = 0;
6005 unsigned outstanding_extents;
6007 lockdep_assert_held(&inode->lock);
6008 outstanding_extents = inode->outstanding_extents;
6009 if (outstanding_extents)
6010 reserve_size = btrfs_calc_trans_metadata_size(fs_info,
6011 outstanding_extents + 1);
6012 csum_leaves = btrfs_csum_bytes_to_leaves(fs_info,
6014 reserve_size += btrfs_calc_trans_metadata_size(fs_info,
6017 * For qgroup rsv, the calculation is very simple:
6018 * account one nodesize for each outstanding extent
6020 * This is overestimating in most cases.
6022 qgroup_rsv_size = (u64)outstanding_extents * fs_info->nodesize;
6024 spin_lock(&block_rsv->lock);
6025 block_rsv->size = reserve_size;
6026 block_rsv->qgroup_rsv_size = qgroup_rsv_size;
6027 spin_unlock(&block_rsv->lock);
6030 static void calc_inode_reservations(struct btrfs_fs_info *fs_info,
6031 u64 num_bytes, u64 *meta_reserve,
6032 u64 *qgroup_reserve)
6034 u64 nr_extents = count_max_extents(num_bytes);
6035 u64 csum_leaves = btrfs_csum_bytes_to_leaves(fs_info, num_bytes);
6037 /* We add one for the inode update at finish ordered time */
6038 *meta_reserve = btrfs_calc_trans_metadata_size(fs_info,
6039 nr_extents + csum_leaves + 1);
6040 *qgroup_reserve = nr_extents * fs_info->nodesize;
6043 int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes)
6045 struct btrfs_root *root = inode->root;
6046 struct btrfs_fs_info *fs_info = root->fs_info;
6047 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
6048 u64 meta_reserve, qgroup_reserve;
6049 unsigned nr_extents;
6050 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
6052 bool delalloc_lock = true;
6054 /* If we are a free space inode we need to not flush since we will be in
6055 * the middle of a transaction commit. We also don't need the delalloc
6056 * mutex since we won't race with anybody. We need this mostly to make
6057 * lockdep shut its filthy mouth.
6059 * If we have a transaction open (can happen if we call truncate_block
6060 * from truncate), then we need FLUSH_LIMIT so we don't deadlock.
6062 if (btrfs_is_free_space_inode(inode)) {
6063 flush = BTRFS_RESERVE_NO_FLUSH;
6064 delalloc_lock = false;
6066 if (current->journal_info)
6067 flush = BTRFS_RESERVE_FLUSH_LIMIT;
6069 if (btrfs_transaction_in_commit(fs_info))
6070 schedule_timeout(1);
6074 mutex_lock(&inode->delalloc_mutex);
6076 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6079 * We always want to do it this way, every other way is wrong and ends
6080 * in tears. Pre-reserving the amount we are going to add will always
6081 * be the right way, because otherwise if we have enough parallelism we
6082 * could end up with thousands of inodes all holding little bits of
6083 * reservations they were able to make previously and the only way to
6084 * reclaim that space is to ENOSPC out the operations and clear
6085 * everything out and try again, which is bad. This way we just
6086 * over-reserve slightly, and clean up the mess when we are done.
6088 calc_inode_reservations(fs_info, num_bytes, &meta_reserve,
6090 ret = btrfs_qgroup_reserve_meta_prealloc(root, qgroup_reserve, true);
6093 ret = reserve_metadata_bytes(root, block_rsv, meta_reserve, flush);
6098 * Now we need to update our outstanding extents and csum bytes _first_
6099 * and then add the reservation to the block_rsv. This keeps us from
6100 * racing with an ordered completion or some such that would think it
6101 * needs to free the reservation we just made.
6103 spin_lock(&inode->lock);
6104 nr_extents = count_max_extents(num_bytes);
6105 btrfs_mod_outstanding_extents(inode, nr_extents);
6106 inode->csum_bytes += num_bytes;
6107 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6108 spin_unlock(&inode->lock);
6110 /* Now we can safely add our space to our block rsv */
6111 block_rsv_add_bytes(block_rsv, meta_reserve, false);
6112 trace_btrfs_space_reservation(root->fs_info, "delalloc",
6113 btrfs_ino(inode), meta_reserve, 1);
6115 spin_lock(&block_rsv->lock);
6116 block_rsv->qgroup_rsv_reserved += qgroup_reserve;
6117 spin_unlock(&block_rsv->lock);
6120 mutex_unlock(&inode->delalloc_mutex);
6123 btrfs_qgroup_free_meta_prealloc(root, qgroup_reserve);
6125 btrfs_inode_rsv_release(inode, true);
6127 mutex_unlock(&inode->delalloc_mutex);
6132 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
6133 * @inode: the inode to release the reservation for.
6134 * @num_bytes: the number of bytes we are releasing.
6135 * @qgroup_free: free qgroup reservation or convert it to per-trans reservation
6137 * This will release the metadata reservation for an inode. This can be called
6138 * once we complete IO for a given set of bytes to release their metadata
6139 * reservations, or on error for the same reason.
6141 void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes,
6144 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6146 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6147 spin_lock(&inode->lock);
6148 inode->csum_bytes -= num_bytes;
6149 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6150 spin_unlock(&inode->lock);
6152 if (btrfs_is_testing(fs_info))
6155 btrfs_inode_rsv_release(inode, qgroup_free);
6159 * btrfs_delalloc_release_extents - release our outstanding_extents
6160 * @inode: the inode to balance the reservation for.
6161 * @num_bytes: the number of bytes we originally reserved with
6162 * @qgroup_free: do we need to free qgroup meta reservation or convert them.
6164 * When we reserve space we increase outstanding_extents for the extents we may
6165 * add. Once we've set the range as delalloc or created our ordered extents we
6166 * have outstanding_extents to track the real usage, so we use this to free our
6167 * temporarily tracked outstanding_extents. This _must_ be used in conjunction
6168 * with btrfs_delalloc_reserve_metadata.
6170 void btrfs_delalloc_release_extents(struct btrfs_inode *inode, u64 num_bytes,
6173 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6174 unsigned num_extents;
6176 spin_lock(&inode->lock);
6177 num_extents = count_max_extents(num_bytes);
6178 btrfs_mod_outstanding_extents(inode, -num_extents);
6179 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6180 spin_unlock(&inode->lock);
6182 if (btrfs_is_testing(fs_info))
6185 btrfs_inode_rsv_release(inode, qgroup_free);
6189 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6191 * @inode: inode we're writing to
6192 * @start: start range we are writing to
6193 * @len: how long the range we are writing to
6194 * @reserved: mandatory parameter, record actually reserved qgroup ranges of
6195 * current reservation.
6197 * This will do the following things
6199 * o reserve space in data space info for num bytes
6200 * and reserve precious corresponding qgroup space
6201 * (Done in check_data_free_space)
6203 * o reserve space for metadata space, based on the number of outstanding
6204 * extents and how much csums will be needed
6205 * also reserve metadata space in a per root over-reserve method.
6206 * o add to the inodes->delalloc_bytes
6207 * o add it to the fs_info's delalloc inodes list.
6208 * (Above 3 all done in delalloc_reserve_metadata)
6210 * Return 0 for success
6211 * Return <0 for error(-ENOSPC or -EQUOT)
6213 int btrfs_delalloc_reserve_space(struct inode *inode,
6214 struct extent_changeset **reserved, u64 start, u64 len)
6218 ret = btrfs_check_data_free_space(inode, reserved, start, len);
6221 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len);
6223 btrfs_free_reserved_data_space(inode, *reserved, start, len);
6228 * btrfs_delalloc_release_space - release data and metadata space for delalloc
6229 * @inode: inode we're releasing space for
6230 * @start: start position of the space already reserved
6231 * @len: the len of the space already reserved
6232 * @release_bytes: the len of the space we consumed or didn't use
6234 * This function will release the metadata space that was not used and will
6235 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
6236 * list if there are no delalloc bytes left.
6237 * Also it will handle the qgroup reserved space.
6239 void btrfs_delalloc_release_space(struct inode *inode,
6240 struct extent_changeset *reserved,
6241 u64 start, u64 len, bool qgroup_free)
6243 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, qgroup_free);
6244 btrfs_free_reserved_data_space(inode, reserved, start, len);
6247 static int update_block_group(struct btrfs_trans_handle *trans,
6248 u64 bytenr, u64 num_bytes, int alloc)
6250 struct btrfs_fs_info *info = trans->fs_info;
6251 struct btrfs_block_group_cache *cache = NULL;
6252 u64 total = num_bytes;
6258 /* block accounting for super block */
6259 spin_lock(&info->delalloc_root_lock);
6260 old_val = btrfs_super_bytes_used(info->super_copy);
6262 old_val += num_bytes;
6264 old_val -= num_bytes;
6265 btrfs_set_super_bytes_used(info->super_copy, old_val);
6266 spin_unlock(&info->delalloc_root_lock);
6269 cache = btrfs_lookup_block_group(info, bytenr);
6274 factor = btrfs_bg_type_to_factor(cache->flags);
6277 * If this block group has free space cache written out, we
6278 * need to make sure to load it if we are removing space. This
6279 * is because we need the unpinning stage to actually add the
6280 * space back to the block group, otherwise we will leak space.
6282 if (!alloc && cache->cached == BTRFS_CACHE_NO)
6283 cache_block_group(cache, 1);
6285 byte_in_group = bytenr - cache->key.objectid;
6286 WARN_ON(byte_in_group > cache->key.offset);
6288 spin_lock(&cache->space_info->lock);
6289 spin_lock(&cache->lock);
6291 if (btrfs_test_opt(info, SPACE_CACHE) &&
6292 cache->disk_cache_state < BTRFS_DC_CLEAR)
6293 cache->disk_cache_state = BTRFS_DC_CLEAR;
6295 old_val = btrfs_block_group_used(&cache->item);
6296 num_bytes = min(total, cache->key.offset - byte_in_group);
6298 old_val += num_bytes;
6299 btrfs_set_block_group_used(&cache->item, old_val);
6300 cache->reserved -= num_bytes;
6301 cache->space_info->bytes_reserved -= num_bytes;
6302 cache->space_info->bytes_used += num_bytes;
6303 cache->space_info->disk_used += num_bytes * factor;
6304 spin_unlock(&cache->lock);
6305 spin_unlock(&cache->space_info->lock);
6307 old_val -= num_bytes;
6308 btrfs_set_block_group_used(&cache->item, old_val);
6309 cache->pinned += num_bytes;
6310 update_bytes_pinned(cache->space_info, num_bytes);
6311 cache->space_info->bytes_used -= num_bytes;
6312 cache->space_info->disk_used -= num_bytes * factor;
6313 spin_unlock(&cache->lock);
6314 spin_unlock(&cache->space_info->lock);
6316 trace_btrfs_space_reservation(info, "pinned",
6317 cache->space_info->flags,
6319 percpu_counter_add_batch(&cache->space_info->total_bytes_pinned,
6321 BTRFS_TOTAL_BYTES_PINNED_BATCH);
6322 set_extent_dirty(info->pinned_extents,
6323 bytenr, bytenr + num_bytes - 1,
6324 GFP_NOFS | __GFP_NOFAIL);
6327 spin_lock(&trans->transaction->dirty_bgs_lock);
6328 if (list_empty(&cache->dirty_list)) {
6329 list_add_tail(&cache->dirty_list,
6330 &trans->transaction->dirty_bgs);
6331 trans->delayed_ref_updates++;
6332 btrfs_get_block_group(cache);
6334 spin_unlock(&trans->transaction->dirty_bgs_lock);
6337 * No longer have used bytes in this block group, queue it for
6338 * deletion. We do this after adding the block group to the
6339 * dirty list to avoid races between cleaner kthread and space
6342 if (!alloc && old_val == 0)
6343 btrfs_mark_bg_unused(cache);
6345 btrfs_put_block_group(cache);
6347 bytenr += num_bytes;
6350 /* Modified block groups are accounted for in the delayed_refs_rsv. */
6351 btrfs_update_delayed_refs_rsv(trans);
6355 static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start)
6357 struct btrfs_block_group_cache *cache;
6360 spin_lock(&fs_info->block_group_cache_lock);
6361 bytenr = fs_info->first_logical_byte;
6362 spin_unlock(&fs_info->block_group_cache_lock);
6364 if (bytenr < (u64)-1)
6367 cache = btrfs_lookup_first_block_group(fs_info, search_start);
6371 bytenr = cache->key.objectid;
6372 btrfs_put_block_group(cache);
6377 static int pin_down_extent(struct btrfs_block_group_cache *cache,
6378 u64 bytenr, u64 num_bytes, int reserved)
6380 struct btrfs_fs_info *fs_info = cache->fs_info;
6382 spin_lock(&cache->space_info->lock);
6383 spin_lock(&cache->lock);
6384 cache->pinned += num_bytes;
6385 update_bytes_pinned(cache->space_info, num_bytes);
6387 cache->reserved -= num_bytes;
6388 cache->space_info->bytes_reserved -= num_bytes;
6390 spin_unlock(&cache->lock);
6391 spin_unlock(&cache->space_info->lock);
6393 trace_btrfs_space_reservation(fs_info, "pinned",
6394 cache->space_info->flags, num_bytes, 1);
6395 percpu_counter_add_batch(&cache->space_info->total_bytes_pinned,
6396 num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH);
6397 set_extent_dirty(fs_info->pinned_extents, bytenr,
6398 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
6403 * this function must be called within transaction
6405 int btrfs_pin_extent(struct btrfs_fs_info *fs_info,
6406 u64 bytenr, u64 num_bytes, int reserved)
6408 struct btrfs_block_group_cache *cache;
6410 cache = btrfs_lookup_block_group(fs_info, bytenr);
6411 BUG_ON(!cache); /* Logic error */
6413 pin_down_extent(cache, bytenr, num_bytes, reserved);
6415 btrfs_put_block_group(cache);
6420 * this function must be called within transaction
6422 int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info,
6423 u64 bytenr, u64 num_bytes)
6425 struct btrfs_block_group_cache *cache;
6428 cache = btrfs_lookup_block_group(fs_info, bytenr);
6433 * pull in the free space cache (if any) so that our pin
6434 * removes the free space from the cache. We have load_only set
6435 * to one because the slow code to read in the free extents does check
6436 * the pinned extents.
6438 cache_block_group(cache, 1);
6440 pin_down_extent(cache, bytenr, num_bytes, 0);
6442 /* remove us from the free space cache (if we're there at all) */
6443 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6444 btrfs_put_block_group(cache);
6448 static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
6449 u64 start, u64 num_bytes)
6452 struct btrfs_block_group_cache *block_group;
6453 struct btrfs_caching_control *caching_ctl;
6455 block_group = btrfs_lookup_block_group(fs_info, start);
6459 cache_block_group(block_group, 0);
6460 caching_ctl = get_caching_control(block_group);
6464 BUG_ON(!block_group_cache_done(block_group));
6465 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6467 mutex_lock(&caching_ctl->mutex);
6469 if (start >= caching_ctl->progress) {
6470 ret = add_excluded_extent(fs_info, start, num_bytes);
6471 } else if (start + num_bytes <= caching_ctl->progress) {
6472 ret = btrfs_remove_free_space(block_group,
6475 num_bytes = caching_ctl->progress - start;
6476 ret = btrfs_remove_free_space(block_group,
6481 num_bytes = (start + num_bytes) -
6482 caching_ctl->progress;
6483 start = caching_ctl->progress;
6484 ret = add_excluded_extent(fs_info, start, num_bytes);
6487 mutex_unlock(&caching_ctl->mutex);
6488 put_caching_control(caching_ctl);
6490 btrfs_put_block_group(block_group);
6494 int btrfs_exclude_logged_extents(struct extent_buffer *eb)
6496 struct btrfs_fs_info *fs_info = eb->fs_info;
6497 struct btrfs_file_extent_item *item;
6498 struct btrfs_key key;
6503 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
6506 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6507 btrfs_item_key_to_cpu(eb, &key, i);
6508 if (key.type != BTRFS_EXTENT_DATA_KEY)
6510 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6511 found_type = btrfs_file_extent_type(eb, item);
6512 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6514 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6516 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6517 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6518 ret = __exclude_logged_extent(fs_info, key.objectid, key.offset);
6527 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg)
6529 atomic_inc(&bg->reservations);
6532 void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info,
6535 struct btrfs_block_group_cache *bg;
6537 bg = btrfs_lookup_block_group(fs_info, start);
6539 if (atomic_dec_and_test(&bg->reservations))
6540 wake_up_var(&bg->reservations);
6541 btrfs_put_block_group(bg);
6544 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg)
6546 struct btrfs_space_info *space_info = bg->space_info;
6550 if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA))
6554 * Our block group is read only but before we set it to read only,
6555 * some task might have had allocated an extent from it already, but it
6556 * has not yet created a respective ordered extent (and added it to a
6557 * root's list of ordered extents).
6558 * Therefore wait for any task currently allocating extents, since the
6559 * block group's reservations counter is incremented while a read lock
6560 * on the groups' semaphore is held and decremented after releasing
6561 * the read access on that semaphore and creating the ordered extent.
6563 down_write(&space_info->groups_sem);
6564 up_write(&space_info->groups_sem);
6566 wait_var_event(&bg->reservations, !atomic_read(&bg->reservations));
6570 * btrfs_add_reserved_bytes - update the block_group and space info counters
6571 * @cache: The cache we are manipulating
6572 * @ram_bytes: The number of bytes of file content, and will be same to
6573 * @num_bytes except for the compress path.
6574 * @num_bytes: The number of bytes in question
6575 * @delalloc: The blocks are allocated for the delalloc write
6577 * This is called by the allocator when it reserves space. If this is a
6578 * reservation and the block group has become read only we cannot make the
6579 * reservation and return -EAGAIN, otherwise this function always succeeds.
6581 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache,
6582 u64 ram_bytes, u64 num_bytes, int delalloc)
6584 struct btrfs_space_info *space_info = cache->space_info;
6587 spin_lock(&space_info->lock);
6588 spin_lock(&cache->lock);
6592 cache->reserved += num_bytes;
6593 space_info->bytes_reserved += num_bytes;
6594 update_bytes_may_use(space_info, -ram_bytes);
6596 cache->delalloc_bytes += num_bytes;
6598 spin_unlock(&cache->lock);
6599 spin_unlock(&space_info->lock);
6604 * btrfs_free_reserved_bytes - update the block_group and space info counters
6605 * @cache: The cache we are manipulating
6606 * @num_bytes: The number of bytes in question
6607 * @delalloc: The blocks are allocated for the delalloc write
6609 * This is called by somebody who is freeing space that was never actually used
6610 * on disk. For example if you reserve some space for a new leaf in transaction
6611 * A and before transaction A commits you free that leaf, you call this with
6612 * reserve set to 0 in order to clear the reservation.
6615 static void btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache,
6616 u64 num_bytes, int delalloc)
6618 struct btrfs_space_info *space_info = cache->space_info;
6620 spin_lock(&space_info->lock);
6621 spin_lock(&cache->lock);
6623 space_info->bytes_readonly += num_bytes;
6624 cache->reserved -= num_bytes;
6625 space_info->bytes_reserved -= num_bytes;
6626 space_info->max_extent_size = 0;
6629 cache->delalloc_bytes -= num_bytes;
6630 spin_unlock(&cache->lock);
6631 spin_unlock(&space_info->lock);
6633 void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info)
6635 struct btrfs_caching_control *next;
6636 struct btrfs_caching_control *caching_ctl;
6637 struct btrfs_block_group_cache *cache;
6639 down_write(&fs_info->commit_root_sem);
6641 list_for_each_entry_safe(caching_ctl, next,
6642 &fs_info->caching_block_groups, list) {
6643 cache = caching_ctl->block_group;
6644 if (block_group_cache_done(cache)) {
6645 cache->last_byte_to_unpin = (u64)-1;
6646 list_del_init(&caching_ctl->list);
6647 put_caching_control(caching_ctl);
6649 cache->last_byte_to_unpin = caching_ctl->progress;
6653 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6654 fs_info->pinned_extents = &fs_info->freed_extents[1];
6656 fs_info->pinned_extents = &fs_info->freed_extents[0];
6658 up_write(&fs_info->commit_root_sem);
6660 update_global_block_rsv(fs_info);
6664 * Returns the free cluster for the given space info and sets empty_cluster to
6665 * what it should be based on the mount options.
6667 static struct btrfs_free_cluster *
6668 fetch_cluster_info(struct btrfs_fs_info *fs_info,
6669 struct btrfs_space_info *space_info, u64 *empty_cluster)
6671 struct btrfs_free_cluster *ret = NULL;
6674 if (btrfs_mixed_space_info(space_info))
6677 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
6678 ret = &fs_info->meta_alloc_cluster;
6679 if (btrfs_test_opt(fs_info, SSD))
6680 *empty_cluster = SZ_2M;
6682 *empty_cluster = SZ_64K;
6683 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
6684 btrfs_test_opt(fs_info, SSD_SPREAD)) {
6685 *empty_cluster = SZ_2M;
6686 ret = &fs_info->data_alloc_cluster;
6692 static int unpin_extent_range(struct btrfs_fs_info *fs_info,
6694 const bool return_free_space)
6696 struct btrfs_block_group_cache *cache = NULL;
6697 struct btrfs_space_info *space_info;
6698 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6699 struct btrfs_free_cluster *cluster = NULL;
6701 u64 total_unpinned = 0;
6702 u64 empty_cluster = 0;
6705 while (start <= end) {
6708 start >= cache->key.objectid + cache->key.offset) {
6710 btrfs_put_block_group(cache);
6712 cache = btrfs_lookup_block_group(fs_info, start);
6713 BUG_ON(!cache); /* Logic error */
6715 cluster = fetch_cluster_info(fs_info,
6718 empty_cluster <<= 1;
6721 len = cache->key.objectid + cache->key.offset - start;
6722 len = min(len, end + 1 - start);
6724 if (start < cache->last_byte_to_unpin) {
6725 len = min(len, cache->last_byte_to_unpin - start);
6726 if (return_free_space)
6727 btrfs_add_free_space(cache, start, len);
6731 total_unpinned += len;
6732 space_info = cache->space_info;
6735 * If this space cluster has been marked as fragmented and we've
6736 * unpinned enough in this block group to potentially allow a
6737 * cluster to be created inside of it go ahead and clear the
6740 if (cluster && cluster->fragmented &&
6741 total_unpinned > empty_cluster) {
6742 spin_lock(&cluster->lock);
6743 cluster->fragmented = 0;
6744 spin_unlock(&cluster->lock);
6747 spin_lock(&space_info->lock);
6748 spin_lock(&cache->lock);
6749 cache->pinned -= len;
6750 update_bytes_pinned(space_info, -len);
6752 trace_btrfs_space_reservation(fs_info, "pinned",
6753 space_info->flags, len, 0);
6754 space_info->max_extent_size = 0;
6755 percpu_counter_add_batch(&space_info->total_bytes_pinned,
6756 -len, BTRFS_TOTAL_BYTES_PINNED_BATCH);
6758 space_info->bytes_readonly += len;
6761 spin_unlock(&cache->lock);
6762 if (!readonly && return_free_space &&
6763 global_rsv->space_info == space_info) {
6766 spin_lock(&global_rsv->lock);
6767 if (!global_rsv->full) {
6768 to_add = min(len, global_rsv->size -
6769 global_rsv->reserved);
6770 global_rsv->reserved += to_add;
6771 update_bytes_may_use(space_info, to_add);
6772 if (global_rsv->reserved >= global_rsv->size)
6773 global_rsv->full = 1;
6774 trace_btrfs_space_reservation(fs_info,
6780 spin_unlock(&global_rsv->lock);
6781 /* Add to any tickets we may have */
6783 space_info_add_new_bytes(fs_info, space_info,
6786 spin_unlock(&space_info->lock);
6790 btrfs_put_block_group(cache);
6794 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans)
6796 struct btrfs_fs_info *fs_info = trans->fs_info;
6797 struct btrfs_block_group_cache *block_group, *tmp;
6798 struct list_head *deleted_bgs;
6799 struct extent_io_tree *unpin;
6804 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6805 unpin = &fs_info->freed_extents[1];
6807 unpin = &fs_info->freed_extents[0];
6809 while (!trans->aborted) {
6810 struct extent_state *cached_state = NULL;
6812 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6813 ret = find_first_extent_bit(unpin, 0, &start, &end,
6814 EXTENT_DIRTY, &cached_state);
6816 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6820 if (btrfs_test_opt(fs_info, DISCARD))
6821 ret = btrfs_discard_extent(fs_info, start,
6822 end + 1 - start, NULL);
6824 clear_extent_dirty(unpin, start, end, &cached_state);
6825 unpin_extent_range(fs_info, start, end, true);
6826 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6827 free_extent_state(cached_state);
6832 * Transaction is finished. We don't need the lock anymore. We
6833 * do need to clean up the block groups in case of a transaction
6836 deleted_bgs = &trans->transaction->deleted_bgs;
6837 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6841 if (!trans->aborted)
6842 ret = btrfs_discard_extent(fs_info,
6843 block_group->key.objectid,
6844 block_group->key.offset,
6847 list_del_init(&block_group->bg_list);
6848 btrfs_put_block_group_trimming(block_group);
6849 btrfs_put_block_group(block_group);
6852 const char *errstr = btrfs_decode_error(ret);
6854 "discard failed while removing blockgroup: errno=%d %s",
6862 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6863 struct btrfs_delayed_ref_node *node, u64 parent,
6864 u64 root_objectid, u64 owner_objectid,
6865 u64 owner_offset, int refs_to_drop,
6866 struct btrfs_delayed_extent_op *extent_op)
6868 struct btrfs_fs_info *info = trans->fs_info;
6869 struct btrfs_key key;
6870 struct btrfs_path *path;
6871 struct btrfs_root *extent_root = info->extent_root;
6872 struct extent_buffer *leaf;
6873 struct btrfs_extent_item *ei;
6874 struct btrfs_extent_inline_ref *iref;
6877 int extent_slot = 0;
6878 int found_extent = 0;
6882 u64 bytenr = node->bytenr;
6883 u64 num_bytes = node->num_bytes;
6885 bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
6887 path = btrfs_alloc_path();
6891 path->reada = READA_FORWARD;
6892 path->leave_spinning = 1;
6894 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6895 BUG_ON(!is_data && refs_to_drop != 1);
6898 skinny_metadata = false;
6900 ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes,
6901 parent, root_objectid, owner_objectid,
6904 extent_slot = path->slots[0];
6905 while (extent_slot >= 0) {
6906 btrfs_item_key_to_cpu(path->nodes[0], &key,
6908 if (key.objectid != bytenr)
6910 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6911 key.offset == num_bytes) {
6915 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6916 key.offset == owner_objectid) {
6920 if (path->slots[0] - extent_slot > 5)
6925 if (!found_extent) {
6927 ret = remove_extent_backref(trans, path, NULL,
6929 is_data, &last_ref);
6931 btrfs_abort_transaction(trans, ret);
6934 btrfs_release_path(path);
6935 path->leave_spinning = 1;
6937 key.objectid = bytenr;
6938 key.type = BTRFS_EXTENT_ITEM_KEY;
6939 key.offset = num_bytes;
6941 if (!is_data && skinny_metadata) {
6942 key.type = BTRFS_METADATA_ITEM_KEY;
6943 key.offset = owner_objectid;
6946 ret = btrfs_search_slot(trans, extent_root,
6948 if (ret > 0 && skinny_metadata && path->slots[0]) {
6950 * Couldn't find our skinny metadata item,
6951 * see if we have ye olde extent item.
6954 btrfs_item_key_to_cpu(path->nodes[0], &key,
6956 if (key.objectid == bytenr &&
6957 key.type == BTRFS_EXTENT_ITEM_KEY &&
6958 key.offset == num_bytes)
6962 if (ret > 0 && skinny_metadata) {
6963 skinny_metadata = false;
6964 key.objectid = bytenr;
6965 key.type = BTRFS_EXTENT_ITEM_KEY;
6966 key.offset = num_bytes;
6967 btrfs_release_path(path);
6968 ret = btrfs_search_slot(trans, extent_root,
6974 "umm, got %d back from search, was looking for %llu",
6977 btrfs_print_leaf(path->nodes[0]);
6980 btrfs_abort_transaction(trans, ret);
6983 extent_slot = path->slots[0];
6985 } else if (WARN_ON(ret == -ENOENT)) {
6986 btrfs_print_leaf(path->nodes[0]);
6988 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6989 bytenr, parent, root_objectid, owner_objectid,
6991 btrfs_abort_transaction(trans, ret);
6994 btrfs_abort_transaction(trans, ret);
6998 leaf = path->nodes[0];
6999 item_size = btrfs_item_size_nr(leaf, extent_slot);
7000 if (unlikely(item_size < sizeof(*ei))) {
7002 btrfs_print_v0_err(info);
7003 btrfs_abort_transaction(trans, ret);
7006 ei = btrfs_item_ptr(leaf, extent_slot,
7007 struct btrfs_extent_item);
7008 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
7009 key.type == BTRFS_EXTENT_ITEM_KEY) {
7010 struct btrfs_tree_block_info *bi;
7011 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
7012 bi = (struct btrfs_tree_block_info *)(ei + 1);
7013 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
7016 refs = btrfs_extent_refs(leaf, ei);
7017 if (refs < refs_to_drop) {
7019 "trying to drop %d refs but we only have %Lu for bytenr %Lu",
7020 refs_to_drop, refs, bytenr);
7022 btrfs_abort_transaction(trans, ret);
7025 refs -= refs_to_drop;
7029 __run_delayed_extent_op(extent_op, leaf, ei);
7031 * In the case of inline back ref, reference count will
7032 * be updated by remove_extent_backref
7035 BUG_ON(!found_extent);
7037 btrfs_set_extent_refs(leaf, ei, refs);
7038 btrfs_mark_buffer_dirty(leaf);
7041 ret = remove_extent_backref(trans, path, iref,
7042 refs_to_drop, is_data,
7045 btrfs_abort_transaction(trans, ret);
7051 BUG_ON(is_data && refs_to_drop !=
7052 extent_data_ref_count(path, iref));
7054 BUG_ON(path->slots[0] != extent_slot);
7056 BUG_ON(path->slots[0] != extent_slot + 1);
7057 path->slots[0] = extent_slot;
7063 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
7066 btrfs_abort_transaction(trans, ret);
7069 btrfs_release_path(path);
7072 ret = btrfs_del_csums(trans, info, bytenr, num_bytes);
7074 btrfs_abort_transaction(trans, ret);
7079 ret = add_to_free_space_tree(trans, bytenr, num_bytes);
7081 btrfs_abort_transaction(trans, ret);
7085 ret = update_block_group(trans, bytenr, num_bytes, 0);
7087 btrfs_abort_transaction(trans, ret);
7091 btrfs_release_path(path);
7094 btrfs_free_path(path);
7099 * when we free an block, it is possible (and likely) that we free the last
7100 * delayed ref for that extent as well. This searches the delayed ref tree for
7101 * a given extent, and if there are no other delayed refs to be processed, it
7102 * removes it from the tree.
7104 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
7107 struct btrfs_delayed_ref_head *head;
7108 struct btrfs_delayed_ref_root *delayed_refs;
7111 delayed_refs = &trans->transaction->delayed_refs;
7112 spin_lock(&delayed_refs->lock);
7113 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
7115 goto out_delayed_unlock;
7117 spin_lock(&head->lock);
7118 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root))
7121 if (cleanup_extent_op(head) != NULL)
7125 * waiting for the lock here would deadlock. If someone else has it
7126 * locked they are already in the process of dropping it anyway
7128 if (!mutex_trylock(&head->mutex))
7131 btrfs_delete_ref_head(delayed_refs, head);
7132 head->processing = 0;
7134 spin_unlock(&head->lock);
7135 spin_unlock(&delayed_refs->lock);
7137 BUG_ON(head->extent_op);
7138 if (head->must_insert_reserved)
7141 btrfs_cleanup_ref_head_accounting(trans->fs_info, delayed_refs, head);
7142 mutex_unlock(&head->mutex);
7143 btrfs_put_delayed_ref_head(head);
7146 spin_unlock(&head->lock);
7149 spin_unlock(&delayed_refs->lock);
7153 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
7154 struct btrfs_root *root,
7155 struct extent_buffer *buf,
7156 u64 parent, int last_ref)
7158 struct btrfs_fs_info *fs_info = root->fs_info;
7159 struct btrfs_ref generic_ref = { 0 };
7163 btrfs_init_generic_ref(&generic_ref, BTRFS_DROP_DELAYED_REF,
7164 buf->start, buf->len, parent);
7165 btrfs_init_tree_ref(&generic_ref, btrfs_header_level(buf),
7166 root->root_key.objectid);
7168 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7169 int old_ref_mod, new_ref_mod;
7171 btrfs_ref_tree_mod(fs_info, &generic_ref);
7172 ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, NULL,
7173 &old_ref_mod, &new_ref_mod);
7174 BUG_ON(ret); /* -ENOMEM */
7175 pin = old_ref_mod >= 0 && new_ref_mod < 0;
7178 if (last_ref && btrfs_header_generation(buf) == trans->transid) {
7179 struct btrfs_block_group_cache *cache;
7181 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7182 ret = check_ref_cleanup(trans, buf->start);
7188 cache = btrfs_lookup_block_group(fs_info, buf->start);
7190 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
7191 pin_down_extent(cache, buf->start, buf->len, 1);
7192 btrfs_put_block_group(cache);
7196 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
7198 btrfs_add_free_space(cache, buf->start, buf->len);
7199 btrfs_free_reserved_bytes(cache, buf->len, 0);
7200 btrfs_put_block_group(cache);
7201 trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
7205 add_pinned_bytes(fs_info, &generic_ref);
7209 * Deleting the buffer, clear the corrupt flag since it doesn't
7212 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
7216 /* Can return -ENOMEM */
7217 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_ref *ref)
7219 struct btrfs_fs_info *fs_info = trans->fs_info;
7220 int old_ref_mod, new_ref_mod;
7223 if (btrfs_is_testing(fs_info))
7227 * tree log blocks never actually go into the extent allocation
7228 * tree, just update pinning info and exit early.
7230 if ((ref->type == BTRFS_REF_METADATA &&
7231 ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID) ||
7232 (ref->type == BTRFS_REF_DATA &&
7233 ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID)) {
7234 /* unlocks the pinned mutex */
7235 btrfs_pin_extent(fs_info, ref->bytenr, ref->len, 1);
7236 old_ref_mod = new_ref_mod = 0;
7238 } else if (ref->type == BTRFS_REF_METADATA) {
7239 ret = btrfs_add_delayed_tree_ref(trans, ref, NULL,
7240 &old_ref_mod, &new_ref_mod);
7242 ret = btrfs_add_delayed_data_ref(trans, ref, 0,
7243 &old_ref_mod, &new_ref_mod);
7246 if (!((ref->type == BTRFS_REF_METADATA &&
7247 ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID) ||
7248 (ref->type == BTRFS_REF_DATA &&
7249 ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID)))
7250 btrfs_ref_tree_mod(fs_info, ref);
7252 if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0)
7253 add_pinned_bytes(fs_info, ref);
7259 * when we wait for progress in the block group caching, its because
7260 * our allocation attempt failed at least once. So, we must sleep
7261 * and let some progress happen before we try again.
7263 * This function will sleep at least once waiting for new free space to
7264 * show up, and then it will check the block group free space numbers
7265 * for our min num_bytes. Another option is to have it go ahead
7266 * and look in the rbtree for a free extent of a given size, but this
7269 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7270 * any of the information in this block group.
7272 static noinline void
7273 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
7276 struct btrfs_caching_control *caching_ctl;
7278 caching_ctl = get_caching_control(cache);
7282 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
7283 (cache->free_space_ctl->free_space >= num_bytes));
7285 put_caching_control(caching_ctl);
7289 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
7291 struct btrfs_caching_control *caching_ctl;
7294 caching_ctl = get_caching_control(cache);
7296 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
7298 wait_event(caching_ctl->wait, block_group_cache_done(cache));
7299 if (cache->cached == BTRFS_CACHE_ERROR)
7301 put_caching_control(caching_ctl);
7305 enum btrfs_loop_type {
7306 LOOP_CACHING_NOWAIT = 0,
7307 LOOP_CACHING_WAIT = 1,
7308 LOOP_ALLOC_CHUNK = 2,
7309 LOOP_NO_EMPTY_SIZE = 3,
7313 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
7317 down_read(&cache->data_rwsem);
7321 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
7324 btrfs_get_block_group(cache);
7326 down_read(&cache->data_rwsem);
7329 static struct btrfs_block_group_cache *
7330 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
7331 struct btrfs_free_cluster *cluster,
7334 struct btrfs_block_group_cache *used_bg = NULL;
7336 spin_lock(&cluster->refill_lock);
7338 used_bg = cluster->block_group;
7342 if (used_bg == block_group)
7345 btrfs_get_block_group(used_bg);
7350 if (down_read_trylock(&used_bg->data_rwsem))
7353 spin_unlock(&cluster->refill_lock);
7355 /* We should only have one-level nested. */
7356 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
7358 spin_lock(&cluster->refill_lock);
7359 if (used_bg == cluster->block_group)
7362 up_read(&used_bg->data_rwsem);
7363 btrfs_put_block_group(used_bg);
7368 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
7372 up_read(&cache->data_rwsem);
7373 btrfs_put_block_group(cache);
7377 * Structure used internally for find_free_extent() function. Wraps needed
7380 struct find_free_extent_ctl {
7381 /* Basic allocation info */
7388 /* Where to start the search inside the bg */
7391 /* For clustered allocation */
7394 bool have_caching_bg;
7395 bool orig_have_caching_bg;
7397 /* RAID index, converted from flags */
7401 * Current loop number, check find_free_extent_update_loop() for details
7406 * Whether we're refilling a cluster, if true we need to re-search
7407 * current block group but don't try to refill the cluster again.
7409 bool retry_clustered;
7412 * Whether we're updating free space cache, if true we need to re-search
7413 * current block group but don't try updating free space cache again.
7415 bool retry_unclustered;
7417 /* If current block group is cached */
7420 /* Max contiguous hole found */
7421 u64 max_extent_size;
7423 /* Total free space from free space cache, not always contiguous */
7424 u64 total_free_space;
7432 * Helper function for find_free_extent().
7434 * Return -ENOENT to inform caller that we need fallback to unclustered mode.
7435 * Return -EAGAIN to inform caller that we need to re-search this block group
7436 * Return >0 to inform caller that we find nothing
7437 * Return 0 means we have found a location and set ffe_ctl->found_offset.
7439 static int find_free_extent_clustered(struct btrfs_block_group_cache *bg,
7440 struct btrfs_free_cluster *last_ptr,
7441 struct find_free_extent_ctl *ffe_ctl,
7442 struct btrfs_block_group_cache **cluster_bg_ret)
7444 struct btrfs_block_group_cache *cluster_bg;
7445 u64 aligned_cluster;
7449 cluster_bg = btrfs_lock_cluster(bg, last_ptr, ffe_ctl->delalloc);
7451 goto refill_cluster;
7452 if (cluster_bg != bg && (cluster_bg->ro ||
7453 !block_group_bits(cluster_bg, ffe_ctl->flags)))
7454 goto release_cluster;
7456 offset = btrfs_alloc_from_cluster(cluster_bg, last_ptr,
7457 ffe_ctl->num_bytes, cluster_bg->key.objectid,
7458 &ffe_ctl->max_extent_size);
7460 /* We have a block, we're done */
7461 spin_unlock(&last_ptr->refill_lock);
7462 trace_btrfs_reserve_extent_cluster(cluster_bg,
7463 ffe_ctl->search_start, ffe_ctl->num_bytes);
7464 *cluster_bg_ret = cluster_bg;
7465 ffe_ctl->found_offset = offset;
7468 WARN_ON(last_ptr->block_group != cluster_bg);
7472 * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so
7473 * lets just skip it and let the allocator find whatever block it can
7474 * find. If we reach this point, we will have tried the cluster
7475 * allocator plenty of times and not have found anything, so we are
7476 * likely way too fragmented for the clustering stuff to find anything.
7478 * However, if the cluster is taken from the current block group,
7479 * release the cluster first, so that we stand a better chance of
7480 * succeeding in the unclustered allocation.
7482 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE && cluster_bg != bg) {
7483 spin_unlock(&last_ptr->refill_lock);
7484 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
7488 /* This cluster didn't work out, free it and start over */
7489 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7491 if (cluster_bg != bg)
7492 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
7495 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE) {
7496 spin_unlock(&last_ptr->refill_lock);
7500 aligned_cluster = max_t(u64,
7501 ffe_ctl->empty_cluster + ffe_ctl->empty_size,
7502 bg->full_stripe_len);
7503 ret = btrfs_find_space_cluster(bg, last_ptr, ffe_ctl->search_start,
7504 ffe_ctl->num_bytes, aligned_cluster);
7506 /* Now pull our allocation out of this cluster */
7507 offset = btrfs_alloc_from_cluster(bg, last_ptr,
7508 ffe_ctl->num_bytes, ffe_ctl->search_start,
7509 &ffe_ctl->max_extent_size);
7511 /* We found one, proceed */
7512 spin_unlock(&last_ptr->refill_lock);
7513 trace_btrfs_reserve_extent_cluster(bg,
7514 ffe_ctl->search_start,
7515 ffe_ctl->num_bytes);
7516 ffe_ctl->found_offset = offset;
7519 } else if (!ffe_ctl->cached && ffe_ctl->loop > LOOP_CACHING_NOWAIT &&
7520 !ffe_ctl->retry_clustered) {
7521 spin_unlock(&last_ptr->refill_lock);
7523 ffe_ctl->retry_clustered = true;
7524 wait_block_group_cache_progress(bg, ffe_ctl->num_bytes +
7525 ffe_ctl->empty_cluster + ffe_ctl->empty_size);
7529 * At this point we either didn't find a cluster or we weren't able to
7530 * allocate a block from our cluster. Free the cluster we've been
7531 * trying to use, and go to the next block group.
7533 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7534 spin_unlock(&last_ptr->refill_lock);
7539 * Return >0 to inform caller that we find nothing
7540 * Return 0 when we found an free extent and set ffe_ctrl->found_offset
7541 * Return -EAGAIN to inform caller that we need to re-search this block group
7543 static int find_free_extent_unclustered(struct btrfs_block_group_cache *bg,
7544 struct btrfs_free_cluster *last_ptr,
7545 struct find_free_extent_ctl *ffe_ctl)
7550 * We are doing an unclustered allocation, set the fragmented flag so
7551 * we don't bother trying to setup a cluster again until we get more
7554 if (unlikely(last_ptr)) {
7555 spin_lock(&last_ptr->lock);
7556 last_ptr->fragmented = 1;
7557 spin_unlock(&last_ptr->lock);
7559 if (ffe_ctl->cached) {
7560 struct btrfs_free_space_ctl *free_space_ctl;
7562 free_space_ctl = bg->free_space_ctl;
7563 spin_lock(&free_space_ctl->tree_lock);
7564 if (free_space_ctl->free_space <
7565 ffe_ctl->num_bytes + ffe_ctl->empty_cluster +
7566 ffe_ctl->empty_size) {
7567 ffe_ctl->total_free_space = max_t(u64,
7568 ffe_ctl->total_free_space,
7569 free_space_ctl->free_space);
7570 spin_unlock(&free_space_ctl->tree_lock);
7573 spin_unlock(&free_space_ctl->tree_lock);
7576 offset = btrfs_find_space_for_alloc(bg, ffe_ctl->search_start,
7577 ffe_ctl->num_bytes, ffe_ctl->empty_size,
7578 &ffe_ctl->max_extent_size);
7581 * If we didn't find a chunk, and we haven't failed on this block group
7582 * before, and this block group is in the middle of caching and we are
7583 * ok with waiting, then go ahead and wait for progress to be made, and
7584 * set @retry_unclustered to true.
7586 * If @retry_unclustered is true then we've already waited on this
7587 * block group once and should move on to the next block group.
7589 if (!offset && !ffe_ctl->retry_unclustered && !ffe_ctl->cached &&
7590 ffe_ctl->loop > LOOP_CACHING_NOWAIT) {
7591 wait_block_group_cache_progress(bg, ffe_ctl->num_bytes +
7592 ffe_ctl->empty_size);
7593 ffe_ctl->retry_unclustered = true;
7595 } else if (!offset) {
7598 ffe_ctl->found_offset = offset;
7603 * Return >0 means caller needs to re-search for free extent
7604 * Return 0 means we have the needed free extent.
7605 * Return <0 means we failed to locate any free extent.
7607 static int find_free_extent_update_loop(struct btrfs_fs_info *fs_info,
7608 struct btrfs_free_cluster *last_ptr,
7609 struct btrfs_key *ins,
7610 struct find_free_extent_ctl *ffe_ctl,
7611 int full_search, bool use_cluster)
7613 struct btrfs_root *root = fs_info->extent_root;
7616 if ((ffe_ctl->loop == LOOP_CACHING_NOWAIT) &&
7617 ffe_ctl->have_caching_bg && !ffe_ctl->orig_have_caching_bg)
7618 ffe_ctl->orig_have_caching_bg = true;
7620 if (!ins->objectid && ffe_ctl->loop >= LOOP_CACHING_WAIT &&
7621 ffe_ctl->have_caching_bg)
7624 if (!ins->objectid && ++(ffe_ctl->index) < BTRFS_NR_RAID_TYPES)
7627 if (ins->objectid) {
7628 if (!use_cluster && last_ptr) {
7629 spin_lock(&last_ptr->lock);
7630 last_ptr->window_start = ins->objectid;
7631 spin_unlock(&last_ptr->lock);
7637 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7638 * caching kthreads as we move along
7639 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7640 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7641 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7644 if (ffe_ctl->loop < LOOP_NO_EMPTY_SIZE) {
7646 if (ffe_ctl->loop == LOOP_CACHING_NOWAIT) {
7648 * We want to skip the LOOP_CACHING_WAIT step if we
7649 * don't have any uncached bgs and we've already done a
7650 * full search through.
7652 if (ffe_ctl->orig_have_caching_bg || !full_search)
7653 ffe_ctl->loop = LOOP_CACHING_WAIT;
7655 ffe_ctl->loop = LOOP_ALLOC_CHUNK;
7660 if (ffe_ctl->loop == LOOP_ALLOC_CHUNK) {
7661 struct btrfs_trans_handle *trans;
7664 trans = current->journal_info;
7668 trans = btrfs_join_transaction(root);
7670 if (IS_ERR(trans)) {
7671 ret = PTR_ERR(trans);
7675 ret = do_chunk_alloc(trans, ffe_ctl->flags,
7679 * If we can't allocate a new chunk we've already looped
7680 * through at least once, move on to the NO_EMPTY_SIZE
7684 ffe_ctl->loop = LOOP_NO_EMPTY_SIZE;
7686 /* Do not bail out on ENOSPC since we can do more. */
7687 if (ret < 0 && ret != -ENOSPC)
7688 btrfs_abort_transaction(trans, ret);
7692 btrfs_end_transaction(trans);
7697 if (ffe_ctl->loop == LOOP_NO_EMPTY_SIZE) {
7699 * Don't loop again if we already have no empty_size and
7702 if (ffe_ctl->empty_size == 0 &&
7703 ffe_ctl->empty_cluster == 0)
7705 ffe_ctl->empty_size = 0;
7706 ffe_ctl->empty_cluster = 0;
7714 * walks the btree of allocated extents and find a hole of a given size.
7715 * The key ins is changed to record the hole:
7716 * ins->objectid == start position
7717 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7718 * ins->offset == the size of the hole.
7719 * Any available blocks before search_start are skipped.
7721 * If there is no suitable free space, we will record the max size of
7722 * the free space extent currently.
7724 * The overall logic and call chain:
7726 * find_free_extent()
7727 * |- Iterate through all block groups
7728 * | |- Get a valid block group
7729 * | |- Try to do clustered allocation in that block group
7730 * | |- Try to do unclustered allocation in that block group
7731 * | |- Check if the result is valid
7732 * | | |- If valid, then exit
7733 * | |- Jump to next block group
7735 * |- Push harder to find free extents
7736 * |- If not found, re-iterate all block groups
7738 static noinline int find_free_extent(struct btrfs_fs_info *fs_info,
7739 u64 ram_bytes, u64 num_bytes, u64 empty_size,
7740 u64 hint_byte, struct btrfs_key *ins,
7741 u64 flags, int delalloc)
7744 struct btrfs_free_cluster *last_ptr = NULL;
7745 struct btrfs_block_group_cache *block_group = NULL;
7746 struct find_free_extent_ctl ffe_ctl = {0};
7747 struct btrfs_space_info *space_info;
7748 bool use_cluster = true;
7749 bool full_search = false;
7751 WARN_ON(num_bytes < fs_info->sectorsize);
7753 ffe_ctl.ram_bytes = ram_bytes;
7754 ffe_ctl.num_bytes = num_bytes;
7755 ffe_ctl.empty_size = empty_size;
7756 ffe_ctl.flags = flags;
7757 ffe_ctl.search_start = 0;
7758 ffe_ctl.retry_clustered = false;
7759 ffe_ctl.retry_unclustered = false;
7760 ffe_ctl.delalloc = delalloc;
7761 ffe_ctl.index = btrfs_bg_flags_to_raid_index(flags);
7762 ffe_ctl.have_caching_bg = false;
7763 ffe_ctl.orig_have_caching_bg = false;
7764 ffe_ctl.found_offset = 0;
7766 ins->type = BTRFS_EXTENT_ITEM_KEY;
7770 trace_find_free_extent(fs_info, num_bytes, empty_size, flags);
7772 space_info = __find_space_info(fs_info, flags);
7774 btrfs_err(fs_info, "No space info for %llu", flags);
7779 * If our free space is heavily fragmented we may not be able to make
7780 * big contiguous allocations, so instead of doing the expensive search
7781 * for free space, simply return ENOSPC with our max_extent_size so we
7782 * can go ahead and search for a more manageable chunk.
7784 * If our max_extent_size is large enough for our allocation simply
7785 * disable clustering since we will likely not be able to find enough
7786 * space to create a cluster and induce latency trying.
7788 if (unlikely(space_info->max_extent_size)) {
7789 spin_lock(&space_info->lock);
7790 if (space_info->max_extent_size &&
7791 num_bytes > space_info->max_extent_size) {
7792 ins->offset = space_info->max_extent_size;
7793 spin_unlock(&space_info->lock);
7795 } else if (space_info->max_extent_size) {
7796 use_cluster = false;
7798 spin_unlock(&space_info->lock);
7801 last_ptr = fetch_cluster_info(fs_info, space_info,
7802 &ffe_ctl.empty_cluster);
7804 spin_lock(&last_ptr->lock);
7805 if (last_ptr->block_group)
7806 hint_byte = last_ptr->window_start;
7807 if (last_ptr->fragmented) {
7809 * We still set window_start so we can keep track of the
7810 * last place we found an allocation to try and save
7813 hint_byte = last_ptr->window_start;
7814 use_cluster = false;
7816 spin_unlock(&last_ptr->lock);
7819 ffe_ctl.search_start = max(ffe_ctl.search_start,
7820 first_logical_byte(fs_info, 0));
7821 ffe_ctl.search_start = max(ffe_ctl.search_start, hint_byte);
7822 if (ffe_ctl.search_start == hint_byte) {
7823 block_group = btrfs_lookup_block_group(fs_info,
7824 ffe_ctl.search_start);
7826 * we don't want to use the block group if it doesn't match our
7827 * allocation bits, or if its not cached.
7829 * However if we are re-searching with an ideal block group
7830 * picked out then we don't care that the block group is cached.
7832 if (block_group && block_group_bits(block_group, flags) &&
7833 block_group->cached != BTRFS_CACHE_NO) {
7834 down_read(&space_info->groups_sem);
7835 if (list_empty(&block_group->list) ||
7838 * someone is removing this block group,
7839 * we can't jump into the have_block_group
7840 * target because our list pointers are not
7843 btrfs_put_block_group(block_group);
7844 up_read(&space_info->groups_sem);
7846 ffe_ctl.index = btrfs_bg_flags_to_raid_index(
7847 block_group->flags);
7848 btrfs_lock_block_group(block_group, delalloc);
7849 goto have_block_group;
7851 } else if (block_group) {
7852 btrfs_put_block_group(block_group);
7856 ffe_ctl.have_caching_bg = false;
7857 if (ffe_ctl.index == btrfs_bg_flags_to_raid_index(flags) ||
7860 down_read(&space_info->groups_sem);
7861 list_for_each_entry(block_group,
7862 &space_info->block_groups[ffe_ctl.index], list) {
7863 /* If the block group is read-only, we can skip it entirely. */
7864 if (unlikely(block_group->ro))
7867 btrfs_grab_block_group(block_group, delalloc);
7868 ffe_ctl.search_start = block_group->key.objectid;
7871 * this can happen if we end up cycling through all the
7872 * raid types, but we want to make sure we only allocate
7873 * for the proper type.
7875 if (!block_group_bits(block_group, flags)) {
7876 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7877 BTRFS_BLOCK_GROUP_RAID1 |
7878 BTRFS_BLOCK_GROUP_RAID5 |
7879 BTRFS_BLOCK_GROUP_RAID6 |
7880 BTRFS_BLOCK_GROUP_RAID10;
7883 * if they asked for extra copies and this block group
7884 * doesn't provide them, bail. This does allow us to
7885 * fill raid0 from raid1.
7887 if ((flags & extra) && !(block_group->flags & extra))
7892 ffe_ctl.cached = block_group_cache_done(block_group);
7893 if (unlikely(!ffe_ctl.cached)) {
7894 ffe_ctl.have_caching_bg = true;
7895 ret = cache_block_group(block_group, 0);
7900 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7904 * Ok we want to try and use the cluster allocator, so
7907 if (last_ptr && use_cluster) {
7908 struct btrfs_block_group_cache *cluster_bg = NULL;
7910 ret = find_free_extent_clustered(block_group, last_ptr,
7911 &ffe_ctl, &cluster_bg);
7914 if (cluster_bg && cluster_bg != block_group) {
7915 btrfs_release_block_group(block_group,
7917 block_group = cluster_bg;
7920 } else if (ret == -EAGAIN) {
7921 goto have_block_group;
7922 } else if (ret > 0) {
7925 /* ret == -ENOENT case falls through */
7928 ret = find_free_extent_unclustered(block_group, last_ptr,
7931 goto have_block_group;
7934 /* ret == 0 case falls through */
7936 ffe_ctl.search_start = round_up(ffe_ctl.found_offset,
7937 fs_info->stripesize);
7939 /* move on to the next group */
7940 if (ffe_ctl.search_start + num_bytes >
7941 block_group->key.objectid + block_group->key.offset) {
7942 btrfs_add_free_space(block_group, ffe_ctl.found_offset,
7947 if (ffe_ctl.found_offset < ffe_ctl.search_start)
7948 btrfs_add_free_space(block_group, ffe_ctl.found_offset,
7949 ffe_ctl.search_start - ffe_ctl.found_offset);
7951 ret = btrfs_add_reserved_bytes(block_group, ram_bytes,
7952 num_bytes, delalloc);
7953 if (ret == -EAGAIN) {
7954 btrfs_add_free_space(block_group, ffe_ctl.found_offset,
7958 btrfs_inc_block_group_reservations(block_group);
7960 /* we are all good, lets return */
7961 ins->objectid = ffe_ctl.search_start;
7962 ins->offset = num_bytes;
7964 trace_btrfs_reserve_extent(block_group, ffe_ctl.search_start,
7966 btrfs_release_block_group(block_group, delalloc);
7969 ffe_ctl.retry_clustered = false;
7970 ffe_ctl.retry_unclustered = false;
7971 BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) !=
7973 btrfs_release_block_group(block_group, delalloc);
7976 up_read(&space_info->groups_sem);
7978 ret = find_free_extent_update_loop(fs_info, last_ptr, ins, &ffe_ctl,
7979 full_search, use_cluster);
7983 if (ret == -ENOSPC) {
7985 * Use ffe_ctl->total_free_space as fallback if we can't find
7986 * any contiguous hole.
7988 if (!ffe_ctl.max_extent_size)
7989 ffe_ctl.max_extent_size = ffe_ctl.total_free_space;
7990 spin_lock(&space_info->lock);
7991 space_info->max_extent_size = ffe_ctl.max_extent_size;
7992 spin_unlock(&space_info->lock);
7993 ins->offset = ffe_ctl.max_extent_size;
7998 #define DUMP_BLOCK_RSV(fs_info, rsv_name) \
8000 struct btrfs_block_rsv *__rsv = &(fs_info)->rsv_name; \
8001 spin_lock(&__rsv->lock); \
8002 btrfs_info(fs_info, #rsv_name ": size %llu reserved %llu", \
8003 __rsv->size, __rsv->reserved); \
8004 spin_unlock(&__rsv->lock); \
8007 static void dump_space_info(struct btrfs_fs_info *fs_info,
8008 struct btrfs_space_info *info, u64 bytes,
8009 int dump_block_groups)
8011 struct btrfs_block_group_cache *cache;
8014 spin_lock(&info->lock);
8015 btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull",
8017 info->total_bytes - btrfs_space_info_used(info, true),
8018 info->full ? "" : "not ");
8020 "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu",
8021 info->total_bytes, info->bytes_used, info->bytes_pinned,
8022 info->bytes_reserved, info->bytes_may_use,
8023 info->bytes_readonly);
8024 spin_unlock(&info->lock);
8026 DUMP_BLOCK_RSV(fs_info, global_block_rsv);
8027 DUMP_BLOCK_RSV(fs_info, trans_block_rsv);
8028 DUMP_BLOCK_RSV(fs_info, chunk_block_rsv);
8029 DUMP_BLOCK_RSV(fs_info, delayed_block_rsv);
8030 DUMP_BLOCK_RSV(fs_info, delayed_refs_rsv);
8032 if (!dump_block_groups)
8035 down_read(&info->groups_sem);
8037 list_for_each_entry(cache, &info->block_groups[index], list) {
8038 spin_lock(&cache->lock);
8040 "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s",
8041 cache->key.objectid, cache->key.offset,
8042 btrfs_block_group_used(&cache->item), cache->pinned,
8043 cache->reserved, cache->ro ? "[readonly]" : "");
8044 btrfs_dump_free_space(cache, bytes);
8045 spin_unlock(&cache->lock);
8047 if (++index < BTRFS_NR_RAID_TYPES)
8049 up_read(&info->groups_sem);
8053 * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a
8054 * hole that is at least as big as @num_bytes.
8056 * @root - The root that will contain this extent
8058 * @ram_bytes - The amount of space in ram that @num_bytes take. This
8059 * is used for accounting purposes. This value differs
8060 * from @num_bytes only in the case of compressed extents.
8062 * @num_bytes - Number of bytes to allocate on-disk.
8064 * @min_alloc_size - Indicates the minimum amount of space that the
8065 * allocator should try to satisfy. In some cases
8066 * @num_bytes may be larger than what is required and if
8067 * the filesystem is fragmented then allocation fails.
8068 * However, the presence of @min_alloc_size gives a
8069 * chance to try and satisfy the smaller allocation.
8071 * @empty_size - A hint that you plan on doing more COW. This is the
8072 * size in bytes the allocator should try to find free
8073 * next to the block it returns. This is just a hint and
8074 * may be ignored by the allocator.
8076 * @hint_byte - Hint to the allocator to start searching above the byte
8077 * address passed. It might be ignored.
8079 * @ins - This key is modified to record the found hole. It will
8080 * have the following values:
8081 * ins->objectid == start position
8082 * ins->flags = BTRFS_EXTENT_ITEM_KEY
8083 * ins->offset == the size of the hole.
8085 * @is_data - Boolean flag indicating whether an extent is
8086 * allocated for data (true) or metadata (false)
8088 * @delalloc - Boolean flag indicating whether this allocation is for
8089 * delalloc or not. If 'true' data_rwsem of block groups
8090 * is going to be acquired.
8093 * Returns 0 when an allocation succeeded or < 0 when an error occurred. In
8094 * case -ENOSPC is returned then @ins->offset will contain the size of the
8095 * largest available hole the allocator managed to find.
8097 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
8098 u64 num_bytes, u64 min_alloc_size,
8099 u64 empty_size, u64 hint_byte,
8100 struct btrfs_key *ins, int is_data, int delalloc)
8102 struct btrfs_fs_info *fs_info = root->fs_info;
8103 bool final_tried = num_bytes == min_alloc_size;
8107 flags = get_alloc_profile_by_root(root, is_data);
8109 WARN_ON(num_bytes < fs_info->sectorsize);
8110 ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size,
8111 hint_byte, ins, flags, delalloc);
8112 if (!ret && !is_data) {
8113 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
8114 } else if (ret == -ENOSPC) {
8115 if (!final_tried && ins->offset) {
8116 num_bytes = min(num_bytes >> 1, ins->offset);
8117 num_bytes = round_down(num_bytes,
8118 fs_info->sectorsize);
8119 num_bytes = max(num_bytes, min_alloc_size);
8120 ram_bytes = num_bytes;
8121 if (num_bytes == min_alloc_size)
8124 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8125 struct btrfs_space_info *sinfo;
8127 sinfo = __find_space_info(fs_info, flags);
8129 "allocation failed flags %llu, wanted %llu",
8132 dump_space_info(fs_info, sinfo, num_bytes, 1);
8139 static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8141 int pin, int delalloc)
8143 struct btrfs_block_group_cache *cache;
8146 cache = btrfs_lookup_block_group(fs_info, start);
8148 btrfs_err(fs_info, "Unable to find block group for %llu",
8154 pin_down_extent(cache, start, len, 1);
8156 if (btrfs_test_opt(fs_info, DISCARD))
8157 ret = btrfs_discard_extent(fs_info, start, len, NULL);
8158 btrfs_add_free_space(cache, start, len);
8159 btrfs_free_reserved_bytes(cache, len, delalloc);
8160 trace_btrfs_reserved_extent_free(fs_info, start, len);
8163 btrfs_put_block_group(cache);
8167 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8168 u64 start, u64 len, int delalloc)
8170 return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc);
8173 int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info,
8176 return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0);
8179 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8180 u64 parent, u64 root_objectid,
8181 u64 flags, u64 owner, u64 offset,
8182 struct btrfs_key *ins, int ref_mod)
8184 struct btrfs_fs_info *fs_info = trans->fs_info;
8186 struct btrfs_extent_item *extent_item;
8187 struct btrfs_extent_inline_ref *iref;
8188 struct btrfs_path *path;
8189 struct extent_buffer *leaf;
8194 type = BTRFS_SHARED_DATA_REF_KEY;
8196 type = BTRFS_EXTENT_DATA_REF_KEY;
8198 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
8200 path = btrfs_alloc_path();
8204 path->leave_spinning = 1;
8205 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8208 btrfs_free_path(path);
8212 leaf = path->nodes[0];
8213 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8214 struct btrfs_extent_item);
8215 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
8216 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8217 btrfs_set_extent_flags(leaf, extent_item,
8218 flags | BTRFS_EXTENT_FLAG_DATA);
8220 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8221 btrfs_set_extent_inline_ref_type(leaf, iref, type);
8223 struct btrfs_shared_data_ref *ref;
8224 ref = (struct btrfs_shared_data_ref *)(iref + 1);
8225 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8226 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
8228 struct btrfs_extent_data_ref *ref;
8229 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
8230 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
8231 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
8232 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
8233 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
8236 btrfs_mark_buffer_dirty(path->nodes[0]);
8237 btrfs_free_path(path);
8239 ret = remove_from_free_space_tree(trans, ins->objectid, ins->offset);
8243 ret = update_block_group(trans, ins->objectid, ins->offset, 1);
8244 if (ret) { /* -ENOENT, logic error */
8245 btrfs_err(fs_info, "update block group failed for %llu %llu",
8246 ins->objectid, ins->offset);
8249 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset);
8253 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
8254 struct btrfs_delayed_ref_node *node,
8255 struct btrfs_delayed_extent_op *extent_op)
8257 struct btrfs_fs_info *fs_info = trans->fs_info;
8259 struct btrfs_extent_item *extent_item;
8260 struct btrfs_key extent_key;
8261 struct btrfs_tree_block_info *block_info;
8262 struct btrfs_extent_inline_ref *iref;
8263 struct btrfs_path *path;
8264 struct extent_buffer *leaf;
8265 struct btrfs_delayed_tree_ref *ref;
8266 u32 size = sizeof(*extent_item) + sizeof(*iref);
8268 u64 flags = extent_op->flags_to_set;
8269 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8271 ref = btrfs_delayed_node_to_tree_ref(node);
8273 extent_key.objectid = node->bytenr;
8274 if (skinny_metadata) {
8275 extent_key.offset = ref->level;
8276 extent_key.type = BTRFS_METADATA_ITEM_KEY;
8277 num_bytes = fs_info->nodesize;
8279 extent_key.offset = node->num_bytes;
8280 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
8281 size += sizeof(*block_info);
8282 num_bytes = node->num_bytes;
8285 path = btrfs_alloc_path();
8289 path->leave_spinning = 1;
8290 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8293 btrfs_free_path(path);
8297 leaf = path->nodes[0];
8298 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8299 struct btrfs_extent_item);
8300 btrfs_set_extent_refs(leaf, extent_item, 1);
8301 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8302 btrfs_set_extent_flags(leaf, extent_item,
8303 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
8305 if (skinny_metadata) {
8306 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8308 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
8309 btrfs_set_tree_block_key(leaf, block_info, &extent_op->key);
8310 btrfs_set_tree_block_level(leaf, block_info, ref->level);
8311 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
8314 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) {
8315 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
8316 btrfs_set_extent_inline_ref_type(leaf, iref,
8317 BTRFS_SHARED_BLOCK_REF_KEY);
8318 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent);
8320 btrfs_set_extent_inline_ref_type(leaf, iref,
8321 BTRFS_TREE_BLOCK_REF_KEY);
8322 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root);
8325 btrfs_mark_buffer_dirty(leaf);
8326 btrfs_free_path(path);
8328 ret = remove_from_free_space_tree(trans, extent_key.objectid,
8333 ret = update_block_group(trans, extent_key.objectid,
8334 fs_info->nodesize, 1);
8335 if (ret) { /* -ENOENT, logic error */
8336 btrfs_err(fs_info, "update block group failed for %llu %llu",
8337 extent_key.objectid, extent_key.offset);
8341 trace_btrfs_reserved_extent_alloc(fs_info, extent_key.objectid,
8346 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8347 struct btrfs_root *root, u64 owner,
8348 u64 offset, u64 ram_bytes,
8349 struct btrfs_key *ins)
8351 struct btrfs_ref generic_ref = { 0 };
8354 BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
8356 btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT,
8357 ins->objectid, ins->offset, 0);
8358 btrfs_init_data_ref(&generic_ref, root->root_key.objectid, owner, offset);
8359 btrfs_ref_tree_mod(root->fs_info, &generic_ref);
8360 ret = btrfs_add_delayed_data_ref(trans, &generic_ref,
8361 ram_bytes, NULL, NULL);
8366 * this is used by the tree logging recovery code. It records that
8367 * an extent has been allocated and makes sure to clear the free
8368 * space cache bits as well
8370 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
8371 u64 root_objectid, u64 owner, u64 offset,
8372 struct btrfs_key *ins)
8374 struct btrfs_fs_info *fs_info = trans->fs_info;
8376 struct btrfs_block_group_cache *block_group;
8377 struct btrfs_space_info *space_info;
8380 * Mixed block groups will exclude before processing the log so we only
8381 * need to do the exclude dance if this fs isn't mixed.
8383 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
8384 ret = __exclude_logged_extent(fs_info, ins->objectid,
8390 block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
8394 space_info = block_group->space_info;
8395 spin_lock(&space_info->lock);
8396 spin_lock(&block_group->lock);
8397 space_info->bytes_reserved += ins->offset;
8398 block_group->reserved += ins->offset;
8399 spin_unlock(&block_group->lock);
8400 spin_unlock(&space_info->lock);
8402 ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner,
8404 btrfs_put_block_group(block_group);
8408 static struct extent_buffer *
8409 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
8410 u64 bytenr, int level, u64 owner)
8412 struct btrfs_fs_info *fs_info = root->fs_info;
8413 struct extent_buffer *buf;
8415 buf = btrfs_find_create_tree_block(fs_info, bytenr);
8420 * Extra safety check in case the extent tree is corrupted and extent
8421 * allocator chooses to use a tree block which is already used and
8424 if (buf->lock_owner == current->pid) {
8425 btrfs_err_rl(fs_info,
8426 "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected",
8427 buf->start, btrfs_header_owner(buf), current->pid);
8428 free_extent_buffer(buf);
8429 return ERR_PTR(-EUCLEAN);
8432 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
8433 btrfs_tree_lock(buf);
8434 btrfs_clean_tree_block(buf);
8435 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
8437 btrfs_set_lock_blocking_write(buf);
8438 set_extent_buffer_uptodate(buf);
8440 memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header));
8441 btrfs_set_header_level(buf, level);
8442 btrfs_set_header_bytenr(buf, buf->start);
8443 btrfs_set_header_generation(buf, trans->transid);
8444 btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV);
8445 btrfs_set_header_owner(buf, owner);
8446 write_extent_buffer_fsid(buf, fs_info->fs_devices->metadata_uuid);
8447 write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid);
8448 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
8449 buf->log_index = root->log_transid % 2;
8451 * we allow two log transactions at a time, use different
8452 * EXTENT bit to differentiate dirty pages.
8454 if (buf->log_index == 0)
8455 set_extent_dirty(&root->dirty_log_pages, buf->start,
8456 buf->start + buf->len - 1, GFP_NOFS);
8458 set_extent_new(&root->dirty_log_pages, buf->start,
8459 buf->start + buf->len - 1);
8461 buf->log_index = -1;
8462 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
8463 buf->start + buf->len - 1, GFP_NOFS);
8465 trans->dirty = true;
8466 /* this returns a buffer locked for blocking */
8470 static struct btrfs_block_rsv *
8471 use_block_rsv(struct btrfs_trans_handle *trans,
8472 struct btrfs_root *root, u32 blocksize)
8474 struct btrfs_fs_info *fs_info = root->fs_info;
8475 struct btrfs_block_rsv *block_rsv;
8476 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
8478 bool global_updated = false;
8480 block_rsv = get_block_rsv(trans, root);
8482 if (unlikely(block_rsv->size == 0))
8485 ret = block_rsv_use_bytes(block_rsv, blocksize);
8489 if (block_rsv->failfast)
8490 return ERR_PTR(ret);
8492 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
8493 global_updated = true;
8494 update_global_block_rsv(fs_info);
8499 * The global reserve still exists to save us from ourselves, so don't
8500 * warn_on if we are short on our delayed refs reserve.
8502 if (block_rsv->type != BTRFS_BLOCK_RSV_DELREFS &&
8503 btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8504 static DEFINE_RATELIMIT_STATE(_rs,
8505 DEFAULT_RATELIMIT_INTERVAL * 10,
8506 /*DEFAULT_RATELIMIT_BURST*/ 1);
8507 if (__ratelimit(&_rs))
8509 "BTRFS: block rsv returned %d\n", ret);
8512 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
8513 BTRFS_RESERVE_NO_FLUSH);
8517 * If we couldn't reserve metadata bytes try and use some from
8518 * the global reserve if its space type is the same as the global
8521 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
8522 block_rsv->space_info == global_rsv->space_info) {
8523 ret = block_rsv_use_bytes(global_rsv, blocksize);
8527 return ERR_PTR(ret);
8530 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
8531 struct btrfs_block_rsv *block_rsv, u32 blocksize)
8533 block_rsv_add_bytes(block_rsv, blocksize, false);
8534 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0, NULL);
8538 * finds a free extent and does all the dirty work required for allocation
8539 * returns the tree buffer or an ERR_PTR on error.
8541 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
8542 struct btrfs_root *root,
8543 u64 parent, u64 root_objectid,
8544 const struct btrfs_disk_key *key,
8545 int level, u64 hint,
8548 struct btrfs_fs_info *fs_info = root->fs_info;
8549 struct btrfs_key ins;
8550 struct btrfs_block_rsv *block_rsv;
8551 struct extent_buffer *buf;
8552 struct btrfs_delayed_extent_op *extent_op;
8553 struct btrfs_ref generic_ref = { 0 };
8556 u32 blocksize = fs_info->nodesize;
8557 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8559 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8560 if (btrfs_is_testing(fs_info)) {
8561 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
8562 level, root_objectid);
8564 root->alloc_bytenr += blocksize;
8569 block_rsv = use_block_rsv(trans, root, blocksize);
8570 if (IS_ERR(block_rsv))
8571 return ERR_CAST(block_rsv);
8573 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
8574 empty_size, hint, &ins, 0, 0);
8578 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level,
8582 goto out_free_reserved;
8585 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
8587 parent = ins.objectid;
8588 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
8592 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
8593 extent_op = btrfs_alloc_delayed_extent_op();
8599 memcpy(&extent_op->key, key, sizeof(extent_op->key));
8601 memset(&extent_op->key, 0, sizeof(extent_op->key));
8602 extent_op->flags_to_set = flags;
8603 extent_op->update_key = skinny_metadata ? false : true;
8604 extent_op->update_flags = true;
8605 extent_op->is_data = false;
8606 extent_op->level = level;
8608 btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT,
8609 ins.objectid, ins.offset, parent);
8610 generic_ref.real_root = root->root_key.objectid;
8611 btrfs_init_tree_ref(&generic_ref, level, root_objectid);
8612 btrfs_ref_tree_mod(fs_info, &generic_ref);
8613 ret = btrfs_add_delayed_tree_ref(trans, &generic_ref,
8614 extent_op, NULL, NULL);
8616 goto out_free_delayed;
8621 btrfs_free_delayed_extent_op(extent_op);
8623 free_extent_buffer(buf);
8625 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
8627 unuse_block_rsv(fs_info, block_rsv, blocksize);
8628 return ERR_PTR(ret);
8631 struct walk_control {
8632 u64 refs[BTRFS_MAX_LEVEL];
8633 u64 flags[BTRFS_MAX_LEVEL];
8634 struct btrfs_key update_progress;
8635 struct btrfs_key drop_progress;
8647 #define DROP_REFERENCE 1
8648 #define UPDATE_BACKREF 2
8650 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
8651 struct btrfs_root *root,
8652 struct walk_control *wc,
8653 struct btrfs_path *path)
8655 struct btrfs_fs_info *fs_info = root->fs_info;
8661 struct btrfs_key key;
8662 struct extent_buffer *eb;
8667 if (path->slots[wc->level] < wc->reada_slot) {
8668 wc->reada_count = wc->reada_count * 2 / 3;
8669 wc->reada_count = max(wc->reada_count, 2);
8671 wc->reada_count = wc->reada_count * 3 / 2;
8672 wc->reada_count = min_t(int, wc->reada_count,
8673 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
8676 eb = path->nodes[wc->level];
8677 nritems = btrfs_header_nritems(eb);
8679 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
8680 if (nread >= wc->reada_count)
8684 bytenr = btrfs_node_blockptr(eb, slot);
8685 generation = btrfs_node_ptr_generation(eb, slot);
8687 if (slot == path->slots[wc->level])
8690 if (wc->stage == UPDATE_BACKREF &&
8691 generation <= root->root_key.offset)
8694 /* We don't lock the tree block, it's OK to be racy here */
8695 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
8696 wc->level - 1, 1, &refs,
8698 /* We don't care about errors in readahead. */
8703 if (wc->stage == DROP_REFERENCE) {
8707 if (wc->level == 1 &&
8708 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8710 if (!wc->update_ref ||
8711 generation <= root->root_key.offset)
8713 btrfs_node_key_to_cpu(eb, &key, slot);
8714 ret = btrfs_comp_cpu_keys(&key,
8715 &wc->update_progress);
8719 if (wc->level == 1 &&
8720 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8724 readahead_tree_block(fs_info, bytenr);
8727 wc->reada_slot = slot;
8731 * helper to process tree block while walking down the tree.
8733 * when wc->stage == UPDATE_BACKREF, this function updates
8734 * back refs for pointers in the block.
8736 * NOTE: return value 1 means we should stop walking down.
8738 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8739 struct btrfs_root *root,
8740 struct btrfs_path *path,
8741 struct walk_control *wc, int lookup_info)
8743 struct btrfs_fs_info *fs_info = root->fs_info;
8744 int level = wc->level;
8745 struct extent_buffer *eb = path->nodes[level];
8746 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8749 if (wc->stage == UPDATE_BACKREF &&
8750 btrfs_header_owner(eb) != root->root_key.objectid)
8754 * when reference count of tree block is 1, it won't increase
8755 * again. once full backref flag is set, we never clear it.
8758 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8759 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8760 BUG_ON(!path->locks[level]);
8761 ret = btrfs_lookup_extent_info(trans, fs_info,
8762 eb->start, level, 1,
8765 BUG_ON(ret == -ENOMEM);
8768 BUG_ON(wc->refs[level] == 0);
8771 if (wc->stage == DROP_REFERENCE) {
8772 if (wc->refs[level] > 1)
8775 if (path->locks[level] && !wc->keep_locks) {
8776 btrfs_tree_unlock_rw(eb, path->locks[level]);
8777 path->locks[level] = 0;
8782 /* wc->stage == UPDATE_BACKREF */
8783 if (!(wc->flags[level] & flag)) {
8784 BUG_ON(!path->locks[level]);
8785 ret = btrfs_inc_ref(trans, root, eb, 1);
8786 BUG_ON(ret); /* -ENOMEM */
8787 ret = btrfs_dec_ref(trans, root, eb, 0);
8788 BUG_ON(ret); /* -ENOMEM */
8789 ret = btrfs_set_disk_extent_flags(trans, eb->start,
8791 btrfs_header_level(eb), 0);
8792 BUG_ON(ret); /* -ENOMEM */
8793 wc->flags[level] |= flag;
8797 * the block is shared by multiple trees, so it's not good to
8798 * keep the tree lock
8800 if (path->locks[level] && level > 0) {
8801 btrfs_tree_unlock_rw(eb, path->locks[level]);
8802 path->locks[level] = 0;
8808 * This is used to verify a ref exists for this root to deal with a bug where we
8809 * would have a drop_progress key that hadn't been updated properly.
8811 static int check_ref_exists(struct btrfs_trans_handle *trans,
8812 struct btrfs_root *root, u64 bytenr, u64 parent,
8815 struct btrfs_path *path;
8816 struct btrfs_extent_inline_ref *iref;
8819 path = btrfs_alloc_path();
8823 ret = lookup_extent_backref(trans, path, &iref, bytenr,
8824 root->fs_info->nodesize, parent,
8825 root->root_key.objectid, level, 0);
8826 btrfs_free_path(path);
8835 * helper to process tree block pointer.
8837 * when wc->stage == DROP_REFERENCE, this function checks
8838 * reference count of the block pointed to. if the block
8839 * is shared and we need update back refs for the subtree
8840 * rooted at the block, this function changes wc->stage to
8841 * UPDATE_BACKREF. if the block is shared and there is no
8842 * need to update back, this function drops the reference
8845 * NOTE: return value 1 means we should stop walking down.
8847 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8848 struct btrfs_root *root,
8849 struct btrfs_path *path,
8850 struct walk_control *wc, int *lookup_info)
8852 struct btrfs_fs_info *fs_info = root->fs_info;
8856 struct btrfs_key key;
8857 struct btrfs_key first_key;
8858 struct btrfs_ref ref = { 0 };
8859 struct extent_buffer *next;
8860 int level = wc->level;
8863 bool need_account = false;
8865 generation = btrfs_node_ptr_generation(path->nodes[level],
8866 path->slots[level]);
8868 * if the lower level block was created before the snapshot
8869 * was created, we know there is no need to update back refs
8872 if (wc->stage == UPDATE_BACKREF &&
8873 generation <= root->root_key.offset) {
8878 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8879 btrfs_node_key_to_cpu(path->nodes[level], &first_key,
8880 path->slots[level]);
8882 next = find_extent_buffer(fs_info, bytenr);
8884 next = btrfs_find_create_tree_block(fs_info, bytenr);
8886 return PTR_ERR(next);
8888 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8892 btrfs_tree_lock(next);
8893 btrfs_set_lock_blocking_write(next);
8895 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
8896 &wc->refs[level - 1],
8897 &wc->flags[level - 1]);
8901 if (unlikely(wc->refs[level - 1] == 0)) {
8902 btrfs_err(fs_info, "Missing references.");
8908 if (wc->stage == DROP_REFERENCE) {
8909 if (wc->refs[level - 1] > 1) {
8910 need_account = true;
8912 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8915 if (!wc->update_ref ||
8916 generation <= root->root_key.offset)
8919 btrfs_node_key_to_cpu(path->nodes[level], &key,
8920 path->slots[level]);
8921 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8925 wc->stage = UPDATE_BACKREF;
8926 wc->shared_level = level - 1;
8930 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8934 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8935 btrfs_tree_unlock(next);
8936 free_extent_buffer(next);
8942 if (reada && level == 1)
8943 reada_walk_down(trans, root, wc, path);
8944 next = read_tree_block(fs_info, bytenr, generation, level - 1,
8947 return PTR_ERR(next);
8948 } else if (!extent_buffer_uptodate(next)) {
8949 free_extent_buffer(next);
8952 btrfs_tree_lock(next);
8953 btrfs_set_lock_blocking_write(next);
8957 ASSERT(level == btrfs_header_level(next));
8958 if (level != btrfs_header_level(next)) {
8959 btrfs_err(root->fs_info, "mismatched level");
8963 path->nodes[level] = next;
8964 path->slots[level] = 0;
8965 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8971 wc->refs[level - 1] = 0;
8972 wc->flags[level - 1] = 0;
8973 if (wc->stage == DROP_REFERENCE) {
8974 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8975 parent = path->nodes[level]->start;
8977 ASSERT(root->root_key.objectid ==
8978 btrfs_header_owner(path->nodes[level]));
8979 if (root->root_key.objectid !=
8980 btrfs_header_owner(path->nodes[level])) {
8981 btrfs_err(root->fs_info,
8982 "mismatched block owner");
8990 * If we had a drop_progress we need to verify the refs are set
8991 * as expected. If we find our ref then we know that from here
8992 * on out everything should be correct, and we can clear the
8995 if (wc->restarted) {
8996 ret = check_ref_exists(trans, root, bytenr, parent,
9007 * Reloc tree doesn't contribute to qgroup numbers, and we have
9008 * already accounted them at merge time (replace_path),
9009 * thus we could skip expensive subtree trace here.
9011 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
9013 ret = btrfs_qgroup_trace_subtree(trans, next,
9014 generation, level - 1);
9016 btrfs_err_rl(fs_info,
9017 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
9023 * We need to update the next key in our walk control so we can
9024 * update the drop_progress key accordingly. We don't care if
9025 * find_next_key doesn't find a key because that means we're at
9026 * the end and are going to clean up now.
9028 wc->drop_level = level;
9029 find_next_key(path, level, &wc->drop_progress);
9031 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
9032 fs_info->nodesize, parent);
9033 btrfs_init_tree_ref(&ref, level - 1, root->root_key.objectid);
9034 ret = btrfs_free_extent(trans, &ref);
9043 btrfs_tree_unlock(next);
9044 free_extent_buffer(next);
9050 * helper to process tree block while walking up the tree.
9052 * when wc->stage == DROP_REFERENCE, this function drops
9053 * reference count on the block.
9055 * when wc->stage == UPDATE_BACKREF, this function changes
9056 * wc->stage back to DROP_REFERENCE if we changed wc->stage
9057 * to UPDATE_BACKREF previously while processing the block.
9059 * NOTE: return value 1 means we should stop walking up.
9061 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
9062 struct btrfs_root *root,
9063 struct btrfs_path *path,
9064 struct walk_control *wc)
9066 struct btrfs_fs_info *fs_info = root->fs_info;
9068 int level = wc->level;
9069 struct extent_buffer *eb = path->nodes[level];
9072 if (wc->stage == UPDATE_BACKREF) {
9073 BUG_ON(wc->shared_level < level);
9074 if (level < wc->shared_level)
9077 ret = find_next_key(path, level + 1, &wc->update_progress);
9081 wc->stage = DROP_REFERENCE;
9082 wc->shared_level = -1;
9083 path->slots[level] = 0;
9086 * check reference count again if the block isn't locked.
9087 * we should start walking down the tree again if reference
9090 if (!path->locks[level]) {
9092 btrfs_tree_lock(eb);
9093 btrfs_set_lock_blocking_write(eb);
9094 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9096 ret = btrfs_lookup_extent_info(trans, fs_info,
9097 eb->start, level, 1,
9101 btrfs_tree_unlock_rw(eb, path->locks[level]);
9102 path->locks[level] = 0;
9105 BUG_ON(wc->refs[level] == 0);
9106 if (wc->refs[level] == 1) {
9107 btrfs_tree_unlock_rw(eb, path->locks[level]);
9108 path->locks[level] = 0;
9114 /* wc->stage == DROP_REFERENCE */
9115 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
9117 if (wc->refs[level] == 1) {
9119 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9120 ret = btrfs_dec_ref(trans, root, eb, 1);
9122 ret = btrfs_dec_ref(trans, root, eb, 0);
9123 BUG_ON(ret); /* -ENOMEM */
9124 if (is_fstree(root->root_key.objectid)) {
9125 ret = btrfs_qgroup_trace_leaf_items(trans, eb);
9127 btrfs_err_rl(fs_info,
9128 "error %d accounting leaf items, quota is out of sync, rescan required",
9133 /* make block locked assertion in btrfs_clean_tree_block happy */
9134 if (!path->locks[level] &&
9135 btrfs_header_generation(eb) == trans->transid) {
9136 btrfs_tree_lock(eb);
9137 btrfs_set_lock_blocking_write(eb);
9138 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9140 btrfs_clean_tree_block(eb);
9143 if (eb == root->node) {
9144 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9146 else if (root->root_key.objectid != btrfs_header_owner(eb))
9147 goto owner_mismatch;
9149 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9150 parent = path->nodes[level + 1]->start;
9151 else if (root->root_key.objectid !=
9152 btrfs_header_owner(path->nodes[level + 1]))
9153 goto owner_mismatch;
9156 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
9158 wc->refs[level] = 0;
9159 wc->flags[level] = 0;
9163 btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu",
9164 btrfs_header_owner(eb), root->root_key.objectid);
9168 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
9169 struct btrfs_root *root,
9170 struct btrfs_path *path,
9171 struct walk_control *wc)
9173 int level = wc->level;
9174 int lookup_info = 1;
9177 while (level >= 0) {
9178 ret = walk_down_proc(trans, root, path, wc, lookup_info);
9185 if (path->slots[level] >=
9186 btrfs_header_nritems(path->nodes[level]))
9189 ret = do_walk_down(trans, root, path, wc, &lookup_info);
9191 path->slots[level]++;
9200 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
9201 struct btrfs_root *root,
9202 struct btrfs_path *path,
9203 struct walk_control *wc, int max_level)
9205 int level = wc->level;
9208 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
9209 while (level < max_level && path->nodes[level]) {
9211 if (path->slots[level] + 1 <
9212 btrfs_header_nritems(path->nodes[level])) {
9213 path->slots[level]++;
9216 ret = walk_up_proc(trans, root, path, wc);
9222 if (path->locks[level]) {
9223 btrfs_tree_unlock_rw(path->nodes[level],
9224 path->locks[level]);
9225 path->locks[level] = 0;
9227 free_extent_buffer(path->nodes[level]);
9228 path->nodes[level] = NULL;
9236 * drop a subvolume tree.
9238 * this function traverses the tree freeing any blocks that only
9239 * referenced by the tree.
9241 * when a shared tree block is found. this function decreases its
9242 * reference count by one. if update_ref is true, this function
9243 * also make sure backrefs for the shared block and all lower level
9244 * blocks are properly updated.
9246 * If called with for_reloc == 0, may exit early with -EAGAIN
9248 int btrfs_drop_snapshot(struct btrfs_root *root,
9249 struct btrfs_block_rsv *block_rsv, int update_ref,
9252 struct btrfs_fs_info *fs_info = root->fs_info;
9253 struct btrfs_path *path;
9254 struct btrfs_trans_handle *trans;
9255 struct btrfs_root *tree_root = fs_info->tree_root;
9256 struct btrfs_root_item *root_item = &root->root_item;
9257 struct walk_control *wc;
9258 struct btrfs_key key;
9262 bool root_dropped = false;
9264 btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid);
9266 path = btrfs_alloc_path();
9272 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9274 btrfs_free_path(path);
9279 trans = btrfs_start_transaction(tree_root, 0);
9280 if (IS_ERR(trans)) {
9281 err = PTR_ERR(trans);
9285 err = btrfs_run_delayed_items(trans);
9290 trans->block_rsv = block_rsv;
9293 * This will help us catch people modifying the fs tree while we're
9294 * dropping it. It is unsafe to mess with the fs tree while it's being
9295 * dropped as we unlock the root node and parent nodes as we walk down
9296 * the tree, assuming nothing will change. If something does change
9297 * then we'll have stale information and drop references to blocks we've
9300 set_bit(BTRFS_ROOT_DELETING, &root->state);
9301 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
9302 level = btrfs_header_level(root->node);
9303 path->nodes[level] = btrfs_lock_root_node(root);
9304 btrfs_set_lock_blocking_write(path->nodes[level]);
9305 path->slots[level] = 0;
9306 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9307 memset(&wc->update_progress, 0,
9308 sizeof(wc->update_progress));
9310 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
9311 memcpy(&wc->update_progress, &key,
9312 sizeof(wc->update_progress));
9314 level = root_item->drop_level;
9316 path->lowest_level = level;
9317 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
9318 path->lowest_level = 0;
9326 * unlock our path, this is safe because only this
9327 * function is allowed to delete this snapshot
9329 btrfs_unlock_up_safe(path, 0);
9331 level = btrfs_header_level(root->node);
9333 btrfs_tree_lock(path->nodes[level]);
9334 btrfs_set_lock_blocking_write(path->nodes[level]);
9335 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9337 ret = btrfs_lookup_extent_info(trans, fs_info,
9338 path->nodes[level]->start,
9339 level, 1, &wc->refs[level],
9345 BUG_ON(wc->refs[level] == 0);
9347 if (level == root_item->drop_level)
9350 btrfs_tree_unlock(path->nodes[level]);
9351 path->locks[level] = 0;
9352 WARN_ON(wc->refs[level] != 1);
9357 wc->restarted = test_bit(BTRFS_ROOT_DEAD_TREE, &root->state);
9359 wc->shared_level = -1;
9360 wc->stage = DROP_REFERENCE;
9361 wc->update_ref = update_ref;
9363 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9367 ret = walk_down_tree(trans, root, path, wc);
9373 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
9380 BUG_ON(wc->stage != DROP_REFERENCE);
9384 if (wc->stage == DROP_REFERENCE) {
9385 wc->drop_level = wc->level;
9386 btrfs_node_key_to_cpu(path->nodes[wc->drop_level],
9388 path->slots[wc->drop_level]);
9390 btrfs_cpu_key_to_disk(&root_item->drop_progress,
9391 &wc->drop_progress);
9392 root_item->drop_level = wc->drop_level;
9394 BUG_ON(wc->level == 0);
9395 if (btrfs_should_end_transaction(trans) ||
9396 (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
9397 ret = btrfs_update_root(trans, tree_root,
9401 btrfs_abort_transaction(trans, ret);
9406 btrfs_end_transaction_throttle(trans);
9407 if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
9408 btrfs_debug(fs_info,
9409 "drop snapshot early exit");
9414 trans = btrfs_start_transaction(tree_root, 0);
9415 if (IS_ERR(trans)) {
9416 err = PTR_ERR(trans);
9420 trans->block_rsv = block_rsv;
9423 btrfs_release_path(path);
9427 ret = btrfs_del_root(trans, &root->root_key);
9429 btrfs_abort_transaction(trans, ret);
9434 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
9435 ret = btrfs_find_root(tree_root, &root->root_key, path,
9438 btrfs_abort_transaction(trans, ret);
9441 } else if (ret > 0) {
9442 /* if we fail to delete the orphan item this time
9443 * around, it'll get picked up the next time.
9445 * The most common failure here is just -ENOENT.
9447 btrfs_del_orphan_item(trans, tree_root,
9448 root->root_key.objectid);
9452 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
9453 btrfs_add_dropped_root(trans, root);
9455 free_extent_buffer(root->node);
9456 free_extent_buffer(root->commit_root);
9457 btrfs_put_fs_root(root);
9459 root_dropped = true;
9461 btrfs_end_transaction_throttle(trans);
9464 btrfs_free_path(path);
9467 * So if we need to stop dropping the snapshot for whatever reason we
9468 * need to make sure to add it back to the dead root list so that we
9469 * keep trying to do the work later. This also cleans up roots if we
9470 * don't have it in the radix (like when we recover after a power fail
9471 * or unmount) so we don't leak memory.
9473 if (!for_reloc && !root_dropped)
9474 btrfs_add_dead_root(root);
9475 if (err && err != -EAGAIN)
9476 btrfs_handle_fs_error(fs_info, err, NULL);
9481 * drop subtree rooted at tree block 'node'.
9483 * NOTE: this function will unlock and release tree block 'node'
9484 * only used by relocation code
9486 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
9487 struct btrfs_root *root,
9488 struct extent_buffer *node,
9489 struct extent_buffer *parent)
9491 struct btrfs_fs_info *fs_info = root->fs_info;
9492 struct btrfs_path *path;
9493 struct walk_control *wc;
9499 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
9501 path = btrfs_alloc_path();
9505 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9507 btrfs_free_path(path);
9511 btrfs_assert_tree_locked(parent);
9512 parent_level = btrfs_header_level(parent);
9513 extent_buffer_get(parent);
9514 path->nodes[parent_level] = parent;
9515 path->slots[parent_level] = btrfs_header_nritems(parent);
9517 btrfs_assert_tree_locked(node);
9518 level = btrfs_header_level(node);
9519 path->nodes[level] = node;
9520 path->slots[level] = 0;
9521 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9523 wc->refs[parent_level] = 1;
9524 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
9526 wc->shared_level = -1;
9527 wc->stage = DROP_REFERENCE;
9530 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9533 wret = walk_down_tree(trans, root, path, wc);
9539 wret = walk_up_tree(trans, root, path, wc, parent_level);
9547 btrfs_free_path(path);
9551 static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags)
9557 * if restripe for this chunk_type is on pick target profile and
9558 * return, otherwise do the usual balance
9560 stripped = get_restripe_target(fs_info, flags);
9562 return extended_to_chunk(stripped);
9564 num_devices = fs_info->fs_devices->rw_devices;
9566 stripped = BTRFS_BLOCK_GROUP_RAID0 |
9567 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
9568 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
9570 if (num_devices == 1) {
9571 stripped |= BTRFS_BLOCK_GROUP_DUP;
9572 stripped = flags & ~stripped;
9574 /* turn raid0 into single device chunks */
9575 if (flags & BTRFS_BLOCK_GROUP_RAID0)
9578 /* turn mirroring into duplication */
9579 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
9580 BTRFS_BLOCK_GROUP_RAID10))
9581 return stripped | BTRFS_BLOCK_GROUP_DUP;
9583 /* they already had raid on here, just return */
9584 if (flags & stripped)
9587 stripped |= BTRFS_BLOCK_GROUP_DUP;
9588 stripped = flags & ~stripped;
9590 /* switch duplicated blocks with raid1 */
9591 if (flags & BTRFS_BLOCK_GROUP_DUP)
9592 return stripped | BTRFS_BLOCK_GROUP_RAID1;
9594 /* this is drive concat, leave it alone */
9600 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
9602 struct btrfs_space_info *sinfo = cache->space_info;
9605 u64 min_allocable_bytes;
9609 * We need some metadata space and system metadata space for
9610 * allocating chunks in some corner cases until we force to set
9611 * it to be readonly.
9614 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
9616 min_allocable_bytes = SZ_1M;
9618 min_allocable_bytes = 0;
9620 spin_lock(&sinfo->lock);
9621 spin_lock(&cache->lock);
9629 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
9630 cache->bytes_super - btrfs_block_group_used(&cache->item);
9631 sinfo_used = btrfs_space_info_used(sinfo, true);
9633 if (sinfo_used + num_bytes + min_allocable_bytes <=
9634 sinfo->total_bytes) {
9635 sinfo->bytes_readonly += num_bytes;
9637 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
9641 spin_unlock(&cache->lock);
9642 spin_unlock(&sinfo->lock);
9643 if (ret == -ENOSPC && btrfs_test_opt(cache->fs_info, ENOSPC_DEBUG)) {
9644 btrfs_info(cache->fs_info,
9645 "unable to make block group %llu ro",
9646 cache->key.objectid);
9647 btrfs_info(cache->fs_info,
9648 "sinfo_used=%llu bg_num_bytes=%llu min_allocable=%llu",
9649 sinfo_used, num_bytes, min_allocable_bytes);
9650 dump_space_info(cache->fs_info, cache->space_info, 0, 0);
9655 int btrfs_inc_block_group_ro(struct btrfs_block_group_cache *cache)
9658 struct btrfs_fs_info *fs_info = cache->fs_info;
9659 struct btrfs_trans_handle *trans;
9664 trans = btrfs_join_transaction(fs_info->extent_root);
9666 return PTR_ERR(trans);
9669 * we're not allowed to set block groups readonly after the dirty
9670 * block groups cache has started writing. If it already started,
9671 * back off and let this transaction commit
9673 mutex_lock(&fs_info->ro_block_group_mutex);
9674 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9675 u64 transid = trans->transid;
9677 mutex_unlock(&fs_info->ro_block_group_mutex);
9678 btrfs_end_transaction(trans);
9680 ret = btrfs_wait_for_commit(fs_info, transid);
9687 * if we are changing raid levels, try to allocate a corresponding
9688 * block group with the new raid level.
9690 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9691 if (alloc_flags != cache->flags) {
9692 ret = do_chunk_alloc(trans, alloc_flags,
9695 * ENOSPC is allowed here, we may have enough space
9696 * already allocated at the new raid level to
9705 ret = inc_block_group_ro(cache, 0);
9708 alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags);
9709 ret = do_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE);
9712 ret = inc_block_group_ro(cache, 0);
9714 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9715 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9716 mutex_lock(&fs_info->chunk_mutex);
9717 check_system_chunk(trans, alloc_flags);
9718 mutex_unlock(&fs_info->chunk_mutex);
9720 mutex_unlock(&fs_info->ro_block_group_mutex);
9722 btrfs_end_transaction(trans);
9726 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, u64 type)
9728 u64 alloc_flags = get_alloc_profile(trans->fs_info, type);
9730 return do_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE);
9734 * helper to account the unused space of all the readonly block group in the
9735 * space_info. takes mirrors into account.
9737 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9739 struct btrfs_block_group_cache *block_group;
9743 /* It's df, we don't care if it's racy */
9744 if (list_empty(&sinfo->ro_bgs))
9747 spin_lock(&sinfo->lock);
9748 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9749 spin_lock(&block_group->lock);
9751 if (!block_group->ro) {
9752 spin_unlock(&block_group->lock);
9756 factor = btrfs_bg_type_to_factor(block_group->flags);
9757 free_bytes += (block_group->key.offset -
9758 btrfs_block_group_used(&block_group->item)) *
9761 spin_unlock(&block_group->lock);
9763 spin_unlock(&sinfo->lock);
9768 void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache)
9770 struct btrfs_space_info *sinfo = cache->space_info;
9775 spin_lock(&sinfo->lock);
9776 spin_lock(&cache->lock);
9778 num_bytes = cache->key.offset - cache->reserved -
9779 cache->pinned - cache->bytes_super -
9780 btrfs_block_group_used(&cache->item);
9781 sinfo->bytes_readonly -= num_bytes;
9782 list_del_init(&cache->ro_list);
9784 spin_unlock(&cache->lock);
9785 spin_unlock(&sinfo->lock);
9789 * Checks to see if it's even possible to relocate this block group.
9791 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9792 * ok to go ahead and try.
9794 int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr)
9796 struct btrfs_block_group_cache *block_group;
9797 struct btrfs_space_info *space_info;
9798 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
9799 struct btrfs_device *device;
9809 debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG);
9811 block_group = btrfs_lookup_block_group(fs_info, bytenr);
9813 /* odd, couldn't find the block group, leave it alone */
9817 "can't find block group for bytenr %llu",
9822 min_free = btrfs_block_group_used(&block_group->item);
9824 /* no bytes used, we're good */
9828 space_info = block_group->space_info;
9829 spin_lock(&space_info->lock);
9831 full = space_info->full;
9834 * if this is the last block group we have in this space, we can't
9835 * relocate it unless we're able to allocate a new chunk below.
9837 * Otherwise, we need to make sure we have room in the space to handle
9838 * all of the extents from this block group. If we can, we're good
9840 if ((space_info->total_bytes != block_group->key.offset) &&
9841 (btrfs_space_info_used(space_info, false) + min_free <
9842 space_info->total_bytes)) {
9843 spin_unlock(&space_info->lock);
9846 spin_unlock(&space_info->lock);
9849 * ok we don't have enough space, but maybe we have free space on our
9850 * devices to allocate new chunks for relocation, so loop through our
9851 * alloc devices and guess if we have enough space. if this block
9852 * group is going to be restriped, run checks against the target
9853 * profile instead of the current one.
9865 target = get_restripe_target(fs_info, block_group->flags);
9867 index = btrfs_bg_flags_to_raid_index(extended_to_chunk(target));
9870 * this is just a balance, so if we were marked as full
9871 * we know there is no space for a new chunk
9876 "no space to alloc new chunk for block group %llu",
9877 block_group->key.objectid);
9881 index = btrfs_bg_flags_to_raid_index(block_group->flags);
9884 if (index == BTRFS_RAID_RAID10) {
9888 } else if (index == BTRFS_RAID_RAID1) {
9890 } else if (index == BTRFS_RAID_DUP) {
9893 } else if (index == BTRFS_RAID_RAID0) {
9894 dev_min = fs_devices->rw_devices;
9895 min_free = div64_u64(min_free, dev_min);
9898 mutex_lock(&fs_info->chunk_mutex);
9899 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9903 * check to make sure we can actually find a chunk with enough
9904 * space to fit our block group in.
9906 if (device->total_bytes > device->bytes_used + min_free &&
9907 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
9908 ret = find_free_dev_extent(device, min_free,
9913 if (dev_nr >= dev_min)
9919 if (debug && ret == -1)
9921 "no space to allocate a new chunk for block group %llu",
9922 block_group->key.objectid);
9923 mutex_unlock(&fs_info->chunk_mutex);
9925 btrfs_put_block_group(block_group);
9929 static int find_first_block_group(struct btrfs_fs_info *fs_info,
9930 struct btrfs_path *path,
9931 struct btrfs_key *key)
9933 struct btrfs_root *root = fs_info->extent_root;
9935 struct btrfs_key found_key;
9936 struct extent_buffer *leaf;
9937 struct btrfs_block_group_item bg;
9941 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9946 slot = path->slots[0];
9947 leaf = path->nodes[0];
9948 if (slot >= btrfs_header_nritems(leaf)) {
9949 ret = btrfs_next_leaf(root, path);
9956 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9958 if (found_key.objectid >= key->objectid &&
9959 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9960 struct extent_map_tree *em_tree;
9961 struct extent_map *em;
9963 em_tree = &root->fs_info->mapping_tree;
9964 read_lock(&em_tree->lock);
9965 em = lookup_extent_mapping(em_tree, found_key.objectid,
9967 read_unlock(&em_tree->lock);
9970 "logical %llu len %llu found bg but no related chunk",
9971 found_key.objectid, found_key.offset);
9973 } else if (em->start != found_key.objectid ||
9974 em->len != found_key.offset) {
9976 "block group %llu len %llu mismatch with chunk %llu len %llu",
9977 found_key.objectid, found_key.offset,
9978 em->start, em->len);
9981 read_extent_buffer(leaf, &bg,
9982 btrfs_item_ptr_offset(leaf, slot),
9984 flags = btrfs_block_group_flags(&bg) &
9985 BTRFS_BLOCK_GROUP_TYPE_MASK;
9987 if (flags != (em->map_lookup->type &
9988 BTRFS_BLOCK_GROUP_TYPE_MASK)) {
9990 "block group %llu len %llu type flags 0x%llx mismatch with chunk type flags 0x%llx",
9992 found_key.offset, flags,
9993 (BTRFS_BLOCK_GROUP_TYPE_MASK &
9994 em->map_lookup->type));
10000 free_extent_map(em);
10009 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
10011 struct btrfs_block_group_cache *block_group;
10015 struct inode *inode;
10017 block_group = btrfs_lookup_first_block_group(info, last);
10018 while (block_group) {
10019 wait_block_group_cache_done(block_group);
10020 spin_lock(&block_group->lock);
10021 if (block_group->iref)
10023 spin_unlock(&block_group->lock);
10024 block_group = next_block_group(block_group);
10026 if (!block_group) {
10033 inode = block_group->inode;
10034 block_group->iref = 0;
10035 block_group->inode = NULL;
10036 spin_unlock(&block_group->lock);
10037 ASSERT(block_group->io_ctl.inode == NULL);
10039 last = block_group->key.objectid + block_group->key.offset;
10040 btrfs_put_block_group(block_group);
10045 * Must be called only after stopping all workers, since we could have block
10046 * group caching kthreads running, and therefore they could race with us if we
10047 * freed the block groups before stopping them.
10049 int btrfs_free_block_groups(struct btrfs_fs_info *info)
10051 struct btrfs_block_group_cache *block_group;
10052 struct btrfs_space_info *space_info;
10053 struct btrfs_caching_control *caching_ctl;
10056 down_write(&info->commit_root_sem);
10057 while (!list_empty(&info->caching_block_groups)) {
10058 caching_ctl = list_entry(info->caching_block_groups.next,
10059 struct btrfs_caching_control, list);
10060 list_del(&caching_ctl->list);
10061 put_caching_control(caching_ctl);
10063 up_write(&info->commit_root_sem);
10065 spin_lock(&info->unused_bgs_lock);
10066 while (!list_empty(&info->unused_bgs)) {
10067 block_group = list_first_entry(&info->unused_bgs,
10068 struct btrfs_block_group_cache,
10070 list_del_init(&block_group->bg_list);
10071 btrfs_put_block_group(block_group);
10073 spin_unlock(&info->unused_bgs_lock);
10075 spin_lock(&info->block_group_cache_lock);
10076 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
10077 block_group = rb_entry(n, struct btrfs_block_group_cache,
10079 rb_erase(&block_group->cache_node,
10080 &info->block_group_cache_tree);
10081 RB_CLEAR_NODE(&block_group->cache_node);
10082 spin_unlock(&info->block_group_cache_lock);
10084 down_write(&block_group->space_info->groups_sem);
10085 list_del(&block_group->list);
10086 up_write(&block_group->space_info->groups_sem);
10089 * We haven't cached this block group, which means we could
10090 * possibly have excluded extents on this block group.
10092 if (block_group->cached == BTRFS_CACHE_NO ||
10093 block_group->cached == BTRFS_CACHE_ERROR)
10094 free_excluded_extents(block_group);
10096 btrfs_remove_free_space_cache(block_group);
10097 ASSERT(block_group->cached != BTRFS_CACHE_STARTED);
10098 ASSERT(list_empty(&block_group->dirty_list));
10099 ASSERT(list_empty(&block_group->io_list));
10100 ASSERT(list_empty(&block_group->bg_list));
10101 ASSERT(atomic_read(&block_group->count) == 1);
10102 btrfs_put_block_group(block_group);
10104 spin_lock(&info->block_group_cache_lock);
10106 spin_unlock(&info->block_group_cache_lock);
10108 /* now that all the block groups are freed, go through and
10109 * free all the space_info structs. This is only called during
10110 * the final stages of unmount, and so we know nobody is
10111 * using them. We call synchronize_rcu() once before we start,
10112 * just to be on the safe side.
10116 release_global_block_rsv(info);
10118 while (!list_empty(&info->space_info)) {
10121 space_info = list_entry(info->space_info.next,
10122 struct btrfs_space_info,
10126 * Do not hide this behind enospc_debug, this is actually
10127 * important and indicates a real bug if this happens.
10129 if (WARN_ON(space_info->bytes_pinned > 0 ||
10130 space_info->bytes_reserved > 0 ||
10131 space_info->bytes_may_use > 0))
10132 dump_space_info(info, space_info, 0, 0);
10133 list_del(&space_info->list);
10134 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
10135 struct kobject *kobj;
10136 kobj = space_info->block_group_kobjs[i];
10137 space_info->block_group_kobjs[i] = NULL;
10143 kobject_del(&space_info->kobj);
10144 kobject_put(&space_info->kobj);
10149 /* link_block_group will queue up kobjects to add when we're reclaim-safe */
10150 void btrfs_add_raid_kobjects(struct btrfs_fs_info *fs_info)
10152 struct btrfs_space_info *space_info;
10153 struct raid_kobject *rkobj;
10158 spin_lock(&fs_info->pending_raid_kobjs_lock);
10159 list_splice_init(&fs_info->pending_raid_kobjs, &list);
10160 spin_unlock(&fs_info->pending_raid_kobjs_lock);
10162 list_for_each_entry(rkobj, &list, list) {
10163 space_info = __find_space_info(fs_info, rkobj->flags);
10164 index = btrfs_bg_flags_to_raid_index(rkobj->flags);
10166 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
10167 "%s", get_raid_name(index));
10169 kobject_put(&rkobj->kobj);
10174 btrfs_warn(fs_info,
10175 "failed to add kobject for block cache, ignoring");
10178 static void link_block_group(struct btrfs_block_group_cache *cache)
10180 struct btrfs_space_info *space_info = cache->space_info;
10181 struct btrfs_fs_info *fs_info = cache->fs_info;
10182 int index = btrfs_bg_flags_to_raid_index(cache->flags);
10183 bool first = false;
10185 down_write(&space_info->groups_sem);
10186 if (list_empty(&space_info->block_groups[index]))
10188 list_add_tail(&cache->list, &space_info->block_groups[index]);
10189 up_write(&space_info->groups_sem);
10192 struct raid_kobject *rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
10194 btrfs_warn(cache->fs_info,
10195 "couldn't alloc memory for raid level kobject");
10198 rkobj->flags = cache->flags;
10199 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
10201 spin_lock(&fs_info->pending_raid_kobjs_lock);
10202 list_add_tail(&rkobj->list, &fs_info->pending_raid_kobjs);
10203 spin_unlock(&fs_info->pending_raid_kobjs_lock);
10204 space_info->block_group_kobjs[index] = &rkobj->kobj;
10208 static struct btrfs_block_group_cache *
10209 btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info,
10210 u64 start, u64 size)
10212 struct btrfs_block_group_cache *cache;
10214 cache = kzalloc(sizeof(*cache), GFP_NOFS);
10218 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
10220 if (!cache->free_space_ctl) {
10225 cache->key.objectid = start;
10226 cache->key.offset = size;
10227 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10229 cache->fs_info = fs_info;
10230 cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start);
10231 set_free_space_tree_thresholds(cache);
10233 atomic_set(&cache->count, 1);
10234 spin_lock_init(&cache->lock);
10235 init_rwsem(&cache->data_rwsem);
10236 INIT_LIST_HEAD(&cache->list);
10237 INIT_LIST_HEAD(&cache->cluster_list);
10238 INIT_LIST_HEAD(&cache->bg_list);
10239 INIT_LIST_HEAD(&cache->ro_list);
10240 INIT_LIST_HEAD(&cache->dirty_list);
10241 INIT_LIST_HEAD(&cache->io_list);
10242 btrfs_init_free_space_ctl(cache);
10243 atomic_set(&cache->trimming, 0);
10244 mutex_init(&cache->free_space_lock);
10245 btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root);
10252 * Iterate all chunks and verify that each of them has the corresponding block
10255 static int check_chunk_block_group_mappings(struct btrfs_fs_info *fs_info)
10257 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
10258 struct extent_map *em;
10259 struct btrfs_block_group_cache *bg;
10264 read_lock(&map_tree->lock);
10266 * lookup_extent_mapping will return the first extent map
10267 * intersecting the range, so setting @len to 1 is enough to
10268 * get the first chunk.
10270 em = lookup_extent_mapping(map_tree, start, 1);
10271 read_unlock(&map_tree->lock);
10275 bg = btrfs_lookup_block_group(fs_info, em->start);
10278 "chunk start=%llu len=%llu doesn't have corresponding block group",
10279 em->start, em->len);
10281 free_extent_map(em);
10284 if (bg->key.objectid != em->start ||
10285 bg->key.offset != em->len ||
10286 (bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK) !=
10287 (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
10289 "chunk start=%llu len=%llu flags=0x%llx doesn't match block group start=%llu len=%llu flags=0x%llx",
10290 em->start, em->len,
10291 em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK,
10292 bg->key.objectid, bg->key.offset,
10293 bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK);
10295 free_extent_map(em);
10296 btrfs_put_block_group(bg);
10299 start = em->start + em->len;
10300 free_extent_map(em);
10301 btrfs_put_block_group(bg);
10306 int btrfs_read_block_groups(struct btrfs_fs_info *info)
10308 struct btrfs_path *path;
10310 struct btrfs_block_group_cache *cache;
10311 struct btrfs_space_info *space_info;
10312 struct btrfs_key key;
10313 struct btrfs_key found_key;
10314 struct extent_buffer *leaf;
10315 int need_clear = 0;
10320 feature = btrfs_super_incompat_flags(info->super_copy);
10321 mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS);
10325 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10326 path = btrfs_alloc_path();
10329 path->reada = READA_FORWARD;
10331 cache_gen = btrfs_super_cache_generation(info->super_copy);
10332 if (btrfs_test_opt(info, SPACE_CACHE) &&
10333 btrfs_super_generation(info->super_copy) != cache_gen)
10335 if (btrfs_test_opt(info, CLEAR_CACHE))
10339 ret = find_first_block_group(info, path, &key);
10345 leaf = path->nodes[0];
10346 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
10348 cache = btrfs_create_block_group_cache(info, found_key.objectid,
10357 * When we mount with old space cache, we need to
10358 * set BTRFS_DC_CLEAR and set dirty flag.
10360 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
10361 * truncate the old free space cache inode and
10363 * b) Setting 'dirty flag' makes sure that we flush
10364 * the new space cache info onto disk.
10366 if (btrfs_test_opt(info, SPACE_CACHE))
10367 cache->disk_cache_state = BTRFS_DC_CLEAR;
10370 read_extent_buffer(leaf, &cache->item,
10371 btrfs_item_ptr_offset(leaf, path->slots[0]),
10372 sizeof(cache->item));
10373 cache->flags = btrfs_block_group_flags(&cache->item);
10375 ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) &&
10376 (cache->flags & BTRFS_BLOCK_GROUP_DATA))) {
10378 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
10379 cache->key.objectid);
10384 key.objectid = found_key.objectid + found_key.offset;
10385 btrfs_release_path(path);
10388 * We need to exclude the super stripes now so that the space
10389 * info has super bytes accounted for, otherwise we'll think
10390 * we have more space than we actually do.
10392 ret = exclude_super_stripes(cache);
10395 * We may have excluded something, so call this just in
10398 free_excluded_extents(cache);
10399 btrfs_put_block_group(cache);
10404 * check for two cases, either we are full, and therefore
10405 * don't need to bother with the caching work since we won't
10406 * find any space, or we are empty, and we can just add all
10407 * the space in and be done with it. This saves us _a_lot_ of
10408 * time, particularly in the full case.
10410 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
10411 cache->last_byte_to_unpin = (u64)-1;
10412 cache->cached = BTRFS_CACHE_FINISHED;
10413 free_excluded_extents(cache);
10414 } else if (btrfs_block_group_used(&cache->item) == 0) {
10415 cache->last_byte_to_unpin = (u64)-1;
10416 cache->cached = BTRFS_CACHE_FINISHED;
10417 add_new_free_space(cache, found_key.objectid,
10418 found_key.objectid +
10420 free_excluded_extents(cache);
10423 ret = btrfs_add_block_group_cache(info, cache);
10425 btrfs_remove_free_space_cache(cache);
10426 btrfs_put_block_group(cache);
10430 trace_btrfs_add_block_group(info, cache, 0);
10431 update_space_info(info, cache->flags, found_key.offset,
10432 btrfs_block_group_used(&cache->item),
10433 cache->bytes_super, &space_info);
10435 cache->space_info = space_info;
10437 link_block_group(cache);
10439 set_avail_alloc_bits(info, cache->flags);
10440 if (btrfs_chunk_readonly(info, cache->key.objectid)) {
10441 inc_block_group_ro(cache, 1);
10442 } else if (btrfs_block_group_used(&cache->item) == 0) {
10443 ASSERT(list_empty(&cache->bg_list));
10444 btrfs_mark_bg_unused(cache);
10448 list_for_each_entry_rcu(space_info, &info->space_info, list) {
10449 if (!(get_alloc_profile(info, space_info->flags) &
10450 (BTRFS_BLOCK_GROUP_RAID10 |
10451 BTRFS_BLOCK_GROUP_RAID1 |
10452 BTRFS_BLOCK_GROUP_RAID5 |
10453 BTRFS_BLOCK_GROUP_RAID6 |
10454 BTRFS_BLOCK_GROUP_DUP)))
10457 * avoid allocating from un-mirrored block group if there are
10458 * mirrored block groups.
10460 list_for_each_entry(cache,
10461 &space_info->block_groups[BTRFS_RAID_RAID0],
10463 inc_block_group_ro(cache, 1);
10464 list_for_each_entry(cache,
10465 &space_info->block_groups[BTRFS_RAID_SINGLE],
10467 inc_block_group_ro(cache, 1);
10470 btrfs_add_raid_kobjects(info);
10471 init_global_block_rsv(info);
10472 ret = check_chunk_block_group_mappings(info);
10474 btrfs_free_path(path);
10478 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans)
10480 struct btrfs_fs_info *fs_info = trans->fs_info;
10481 struct btrfs_block_group_cache *block_group;
10482 struct btrfs_root *extent_root = fs_info->extent_root;
10483 struct btrfs_block_group_item item;
10484 struct btrfs_key key;
10487 if (!trans->can_flush_pending_bgs)
10490 while (!list_empty(&trans->new_bgs)) {
10491 block_group = list_first_entry(&trans->new_bgs,
10492 struct btrfs_block_group_cache,
10497 spin_lock(&block_group->lock);
10498 memcpy(&item, &block_group->item, sizeof(item));
10499 memcpy(&key, &block_group->key, sizeof(key));
10500 spin_unlock(&block_group->lock);
10502 ret = btrfs_insert_item(trans, extent_root, &key, &item,
10505 btrfs_abort_transaction(trans, ret);
10506 ret = btrfs_finish_chunk_alloc(trans, key.objectid, key.offset);
10508 btrfs_abort_transaction(trans, ret);
10509 add_block_group_free_space(trans, block_group);
10510 /* already aborted the transaction if it failed. */
10512 btrfs_delayed_refs_rsv_release(fs_info, 1);
10513 list_del_init(&block_group->bg_list);
10515 btrfs_trans_release_chunk_metadata(trans);
10518 int btrfs_make_block_group(struct btrfs_trans_handle *trans, u64 bytes_used,
10519 u64 type, u64 chunk_offset, u64 size)
10521 struct btrfs_fs_info *fs_info = trans->fs_info;
10522 struct btrfs_block_group_cache *cache;
10525 btrfs_set_log_full_commit(trans);
10527 cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size);
10531 btrfs_set_block_group_used(&cache->item, bytes_used);
10532 btrfs_set_block_group_chunk_objectid(&cache->item,
10533 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
10534 btrfs_set_block_group_flags(&cache->item, type);
10536 cache->flags = type;
10537 cache->last_byte_to_unpin = (u64)-1;
10538 cache->cached = BTRFS_CACHE_FINISHED;
10539 cache->needs_free_space = 1;
10540 ret = exclude_super_stripes(cache);
10543 * We may have excluded something, so call this just in
10546 free_excluded_extents(cache);
10547 btrfs_put_block_group(cache);
10551 add_new_free_space(cache, chunk_offset, chunk_offset + size);
10553 free_excluded_extents(cache);
10555 #ifdef CONFIG_BTRFS_DEBUG
10556 if (btrfs_should_fragment_free_space(cache)) {
10557 u64 new_bytes_used = size - bytes_used;
10559 bytes_used += new_bytes_used >> 1;
10560 fragment_free_space(cache);
10564 * Ensure the corresponding space_info object is created and
10565 * assigned to our block group. We want our bg to be added to the rbtree
10566 * with its ->space_info set.
10568 cache->space_info = __find_space_info(fs_info, cache->flags);
10569 ASSERT(cache->space_info);
10571 ret = btrfs_add_block_group_cache(fs_info, cache);
10573 btrfs_remove_free_space_cache(cache);
10574 btrfs_put_block_group(cache);
10579 * Now that our block group has its ->space_info set and is inserted in
10580 * the rbtree, update the space info's counters.
10582 trace_btrfs_add_block_group(fs_info, cache, 1);
10583 update_space_info(fs_info, cache->flags, size, bytes_used,
10584 cache->bytes_super, &cache->space_info);
10585 update_global_block_rsv(fs_info);
10587 link_block_group(cache);
10589 list_add_tail(&cache->bg_list, &trans->new_bgs);
10590 trans->delayed_ref_updates++;
10591 btrfs_update_delayed_refs_rsv(trans);
10593 set_avail_alloc_bits(fs_info, type);
10597 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
10599 u64 extra_flags = chunk_to_extended(flags) &
10600 BTRFS_EXTENDED_PROFILE_MASK;
10602 write_seqlock(&fs_info->profiles_lock);
10603 if (flags & BTRFS_BLOCK_GROUP_DATA)
10604 fs_info->avail_data_alloc_bits &= ~extra_flags;
10605 if (flags & BTRFS_BLOCK_GROUP_METADATA)
10606 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
10607 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
10608 fs_info->avail_system_alloc_bits &= ~extra_flags;
10609 write_sequnlock(&fs_info->profiles_lock);
10612 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
10613 u64 group_start, struct extent_map *em)
10615 struct btrfs_fs_info *fs_info = trans->fs_info;
10616 struct btrfs_root *root = fs_info->extent_root;
10617 struct btrfs_path *path;
10618 struct btrfs_block_group_cache *block_group;
10619 struct btrfs_free_cluster *cluster;
10620 struct btrfs_root *tree_root = fs_info->tree_root;
10621 struct btrfs_key key;
10622 struct inode *inode;
10623 struct kobject *kobj = NULL;
10627 struct btrfs_caching_control *caching_ctl = NULL;
10629 bool remove_rsv = false;
10631 block_group = btrfs_lookup_block_group(fs_info, group_start);
10632 BUG_ON(!block_group);
10633 BUG_ON(!block_group->ro);
10635 trace_btrfs_remove_block_group(block_group);
10637 * Free the reserved super bytes from this block group before
10640 free_excluded_extents(block_group);
10641 btrfs_free_ref_tree_range(fs_info, block_group->key.objectid,
10642 block_group->key.offset);
10644 memcpy(&key, &block_group->key, sizeof(key));
10645 index = btrfs_bg_flags_to_raid_index(block_group->flags);
10646 factor = btrfs_bg_type_to_factor(block_group->flags);
10648 /* make sure this block group isn't part of an allocation cluster */
10649 cluster = &fs_info->data_alloc_cluster;
10650 spin_lock(&cluster->refill_lock);
10651 btrfs_return_cluster_to_free_space(block_group, cluster);
10652 spin_unlock(&cluster->refill_lock);
10655 * make sure this block group isn't part of a metadata
10656 * allocation cluster
10658 cluster = &fs_info->meta_alloc_cluster;
10659 spin_lock(&cluster->refill_lock);
10660 btrfs_return_cluster_to_free_space(block_group, cluster);
10661 spin_unlock(&cluster->refill_lock);
10663 path = btrfs_alloc_path();
10670 * get the inode first so any iput calls done for the io_list
10671 * aren't the final iput (no unlinks allowed now)
10673 inode = lookup_free_space_inode(block_group, path);
10675 mutex_lock(&trans->transaction->cache_write_mutex);
10677 * Make sure our free space cache IO is done before removing the
10680 spin_lock(&trans->transaction->dirty_bgs_lock);
10681 if (!list_empty(&block_group->io_list)) {
10682 list_del_init(&block_group->io_list);
10684 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
10686 spin_unlock(&trans->transaction->dirty_bgs_lock);
10687 btrfs_wait_cache_io(trans, block_group, path);
10688 btrfs_put_block_group(block_group);
10689 spin_lock(&trans->transaction->dirty_bgs_lock);
10692 if (!list_empty(&block_group->dirty_list)) {
10693 list_del_init(&block_group->dirty_list);
10695 btrfs_put_block_group(block_group);
10697 spin_unlock(&trans->transaction->dirty_bgs_lock);
10698 mutex_unlock(&trans->transaction->cache_write_mutex);
10700 if (!IS_ERR(inode)) {
10701 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10703 btrfs_add_delayed_iput(inode);
10706 clear_nlink(inode);
10707 /* One for the block groups ref */
10708 spin_lock(&block_group->lock);
10709 if (block_group->iref) {
10710 block_group->iref = 0;
10711 block_group->inode = NULL;
10712 spin_unlock(&block_group->lock);
10715 spin_unlock(&block_group->lock);
10717 /* One for our lookup ref */
10718 btrfs_add_delayed_iput(inode);
10721 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
10722 key.offset = block_group->key.objectid;
10725 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
10729 btrfs_release_path(path);
10731 ret = btrfs_del_item(trans, tree_root, path);
10734 btrfs_release_path(path);
10737 spin_lock(&fs_info->block_group_cache_lock);
10738 rb_erase(&block_group->cache_node,
10739 &fs_info->block_group_cache_tree);
10740 RB_CLEAR_NODE(&block_group->cache_node);
10742 if (fs_info->first_logical_byte == block_group->key.objectid)
10743 fs_info->first_logical_byte = (u64)-1;
10744 spin_unlock(&fs_info->block_group_cache_lock);
10746 down_write(&block_group->space_info->groups_sem);
10748 * we must use list_del_init so people can check to see if they
10749 * are still on the list after taking the semaphore
10751 list_del_init(&block_group->list);
10752 if (list_empty(&block_group->space_info->block_groups[index])) {
10753 kobj = block_group->space_info->block_group_kobjs[index];
10754 block_group->space_info->block_group_kobjs[index] = NULL;
10755 clear_avail_alloc_bits(fs_info, block_group->flags);
10757 up_write(&block_group->space_info->groups_sem);
10763 if (block_group->has_caching_ctl)
10764 caching_ctl = get_caching_control(block_group);
10765 if (block_group->cached == BTRFS_CACHE_STARTED)
10766 wait_block_group_cache_done(block_group);
10767 if (block_group->has_caching_ctl) {
10768 down_write(&fs_info->commit_root_sem);
10769 if (!caching_ctl) {
10770 struct btrfs_caching_control *ctl;
10772 list_for_each_entry(ctl,
10773 &fs_info->caching_block_groups, list)
10774 if (ctl->block_group == block_group) {
10776 refcount_inc(&caching_ctl->count);
10781 list_del_init(&caching_ctl->list);
10782 up_write(&fs_info->commit_root_sem);
10784 /* Once for the caching bgs list and once for us. */
10785 put_caching_control(caching_ctl);
10786 put_caching_control(caching_ctl);
10790 spin_lock(&trans->transaction->dirty_bgs_lock);
10791 WARN_ON(!list_empty(&block_group->dirty_list));
10792 WARN_ON(!list_empty(&block_group->io_list));
10793 spin_unlock(&trans->transaction->dirty_bgs_lock);
10795 btrfs_remove_free_space_cache(block_group);
10797 spin_lock(&block_group->space_info->lock);
10798 list_del_init(&block_group->ro_list);
10800 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
10801 WARN_ON(block_group->space_info->total_bytes
10802 < block_group->key.offset);
10803 WARN_ON(block_group->space_info->bytes_readonly
10804 < block_group->key.offset);
10805 WARN_ON(block_group->space_info->disk_total
10806 < block_group->key.offset * factor);
10808 block_group->space_info->total_bytes -= block_group->key.offset;
10809 block_group->space_info->bytes_readonly -= block_group->key.offset;
10810 block_group->space_info->disk_total -= block_group->key.offset * factor;
10812 spin_unlock(&block_group->space_info->lock);
10814 memcpy(&key, &block_group->key, sizeof(key));
10816 mutex_lock(&fs_info->chunk_mutex);
10817 spin_lock(&block_group->lock);
10818 block_group->removed = 1;
10820 * At this point trimming can't start on this block group, because we
10821 * removed the block group from the tree fs_info->block_group_cache_tree
10822 * so no one can't find it anymore and even if someone already got this
10823 * block group before we removed it from the rbtree, they have already
10824 * incremented block_group->trimming - if they didn't, they won't find
10825 * any free space entries because we already removed them all when we
10826 * called btrfs_remove_free_space_cache().
10828 * And we must not remove the extent map from the fs_info->mapping_tree
10829 * to prevent the same logical address range and physical device space
10830 * ranges from being reused for a new block group. This is because our
10831 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10832 * completely transactionless, so while it is trimming a range the
10833 * currently running transaction might finish and a new one start,
10834 * allowing for new block groups to be created that can reuse the same
10835 * physical device locations unless we take this special care.
10837 * There may also be an implicit trim operation if the file system
10838 * is mounted with -odiscard. The same protections must remain
10839 * in place until the extents have been discarded completely when
10840 * the transaction commit has completed.
10842 remove_em = (atomic_read(&block_group->trimming) == 0);
10843 spin_unlock(&block_group->lock);
10845 mutex_unlock(&fs_info->chunk_mutex);
10847 ret = remove_block_group_free_space(trans, block_group);
10851 btrfs_put_block_group(block_group);
10852 btrfs_put_block_group(block_group);
10854 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10860 ret = btrfs_del_item(trans, root, path);
10865 struct extent_map_tree *em_tree;
10867 em_tree = &fs_info->mapping_tree;
10868 write_lock(&em_tree->lock);
10869 remove_extent_mapping(em_tree, em);
10870 write_unlock(&em_tree->lock);
10871 /* once for the tree */
10872 free_extent_map(em);
10876 btrfs_delayed_refs_rsv_release(fs_info, 1);
10877 btrfs_free_path(path);
10881 struct btrfs_trans_handle *
10882 btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info,
10883 const u64 chunk_offset)
10885 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
10886 struct extent_map *em;
10887 struct map_lookup *map;
10888 unsigned int num_items;
10890 read_lock(&em_tree->lock);
10891 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
10892 read_unlock(&em_tree->lock);
10893 ASSERT(em && em->start == chunk_offset);
10896 * We need to reserve 3 + N units from the metadata space info in order
10897 * to remove a block group (done at btrfs_remove_chunk() and at
10898 * btrfs_remove_block_group()), which are used for:
10900 * 1 unit for adding the free space inode's orphan (located in the tree
10902 * 1 unit for deleting the block group item (located in the extent
10904 * 1 unit for deleting the free space item (located in tree of tree
10906 * N units for deleting N device extent items corresponding to each
10907 * stripe (located in the device tree).
10909 * In order to remove a block group we also need to reserve units in the
10910 * system space info in order to update the chunk tree (update one or
10911 * more device items and remove one chunk item), but this is done at
10912 * btrfs_remove_chunk() through a call to check_system_chunk().
10914 map = em->map_lookup;
10915 num_items = 3 + map->num_stripes;
10916 free_extent_map(em);
10918 return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root,
10923 * Process the unused_bgs list and remove any that don't have any allocated
10924 * space inside of them.
10926 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10928 struct btrfs_block_group_cache *block_group;
10929 struct btrfs_space_info *space_info;
10930 struct btrfs_trans_handle *trans;
10933 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
10936 spin_lock(&fs_info->unused_bgs_lock);
10937 while (!list_empty(&fs_info->unused_bgs)) {
10941 block_group = list_first_entry(&fs_info->unused_bgs,
10942 struct btrfs_block_group_cache,
10944 list_del_init(&block_group->bg_list);
10946 space_info = block_group->space_info;
10948 if (ret || btrfs_mixed_space_info(space_info)) {
10949 btrfs_put_block_group(block_group);
10952 spin_unlock(&fs_info->unused_bgs_lock);
10954 mutex_lock(&fs_info->delete_unused_bgs_mutex);
10956 /* Don't want to race with allocators so take the groups_sem */
10957 down_write(&space_info->groups_sem);
10958 spin_lock(&block_group->lock);
10959 if (block_group->reserved || block_group->pinned ||
10960 btrfs_block_group_used(&block_group->item) ||
10962 list_is_singular(&block_group->list)) {
10964 * We want to bail if we made new allocations or have
10965 * outstanding allocations in this block group. We do
10966 * the ro check in case balance is currently acting on
10967 * this block group.
10969 trace_btrfs_skip_unused_block_group(block_group);
10970 spin_unlock(&block_group->lock);
10971 up_write(&space_info->groups_sem);
10974 spin_unlock(&block_group->lock);
10976 /* We don't want to force the issue, only flip if it's ok. */
10977 ret = inc_block_group_ro(block_group, 0);
10978 up_write(&space_info->groups_sem);
10985 * Want to do this before we do anything else so we can recover
10986 * properly if we fail to join the transaction.
10988 trans = btrfs_start_trans_remove_block_group(fs_info,
10989 block_group->key.objectid);
10990 if (IS_ERR(trans)) {
10991 btrfs_dec_block_group_ro(block_group);
10992 ret = PTR_ERR(trans);
10997 * We could have pending pinned extents for this block group,
10998 * just delete them, we don't care about them anymore.
11000 start = block_group->key.objectid;
11001 end = start + block_group->key.offset - 1;
11003 * Hold the unused_bg_unpin_mutex lock to avoid racing with
11004 * btrfs_finish_extent_commit(). If we are at transaction N,
11005 * another task might be running finish_extent_commit() for the
11006 * previous transaction N - 1, and have seen a range belonging
11007 * to the block group in freed_extents[] before we were able to
11008 * clear the whole block group range from freed_extents[]. This
11009 * means that task can lookup for the block group after we
11010 * unpinned it from freed_extents[] and removed it, leading to
11011 * a BUG_ON() at btrfs_unpin_extent_range().
11013 mutex_lock(&fs_info->unused_bg_unpin_mutex);
11014 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
11017 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
11018 btrfs_dec_block_group_ro(block_group);
11021 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
11024 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
11025 btrfs_dec_block_group_ro(block_group);
11028 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
11030 /* Reset pinned so btrfs_put_block_group doesn't complain */
11031 spin_lock(&space_info->lock);
11032 spin_lock(&block_group->lock);
11034 update_bytes_pinned(space_info, -block_group->pinned);
11035 space_info->bytes_readonly += block_group->pinned;
11036 percpu_counter_add_batch(&space_info->total_bytes_pinned,
11037 -block_group->pinned,
11038 BTRFS_TOTAL_BYTES_PINNED_BATCH);
11039 block_group->pinned = 0;
11041 spin_unlock(&block_group->lock);
11042 spin_unlock(&space_info->lock);
11044 /* DISCARD can flip during remount */
11045 trimming = btrfs_test_opt(fs_info, DISCARD);
11047 /* Implicit trim during transaction commit. */
11049 btrfs_get_block_group_trimming(block_group);
11052 * Btrfs_remove_chunk will abort the transaction if things go
11055 ret = btrfs_remove_chunk(trans, block_group->key.objectid);
11059 btrfs_put_block_group_trimming(block_group);
11064 * If we're not mounted with -odiscard, we can just forget
11065 * about this block group. Otherwise we'll need to wait
11066 * until transaction commit to do the actual discard.
11069 spin_lock(&fs_info->unused_bgs_lock);
11071 * A concurrent scrub might have added us to the list
11072 * fs_info->unused_bgs, so use a list_move operation
11073 * to add the block group to the deleted_bgs list.
11075 list_move(&block_group->bg_list,
11076 &trans->transaction->deleted_bgs);
11077 spin_unlock(&fs_info->unused_bgs_lock);
11078 btrfs_get_block_group(block_group);
11081 btrfs_end_transaction(trans);
11083 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
11084 btrfs_put_block_group(block_group);
11085 spin_lock(&fs_info->unused_bgs_lock);
11087 spin_unlock(&fs_info->unused_bgs_lock);
11090 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
11092 struct btrfs_super_block *disk_super;
11098 disk_super = fs_info->super_copy;
11099 if (!btrfs_super_root(disk_super))
11102 features = btrfs_super_incompat_flags(disk_super);
11103 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
11106 flags = BTRFS_BLOCK_GROUP_SYSTEM;
11107 ret = create_space_info(fs_info, flags);
11112 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
11113 ret = create_space_info(fs_info, flags);
11115 flags = BTRFS_BLOCK_GROUP_METADATA;
11116 ret = create_space_info(fs_info, flags);
11120 flags = BTRFS_BLOCK_GROUP_DATA;
11121 ret = create_space_info(fs_info, flags);
11127 int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
11128 u64 start, u64 end)
11130 return unpin_extent_range(fs_info, start, end, false);
11134 * It used to be that old block groups would be left around forever.
11135 * Iterating over them would be enough to trim unused space. Since we
11136 * now automatically remove them, we also need to iterate over unallocated
11139 * We don't want a transaction for this since the discard may take a
11140 * substantial amount of time. We don't require that a transaction be
11141 * running, but we do need to take a running transaction into account
11142 * to ensure that we're not discarding chunks that were released or
11143 * allocated in the current transaction.
11145 * Holding the chunks lock will prevent other threads from allocating
11146 * or releasing chunks, but it won't prevent a running transaction
11147 * from committing and releasing the memory that the pending chunks
11148 * list head uses. For that, we need to take a reference to the
11149 * transaction and hold the commit root sem. We only need to hold
11150 * it while performing the free space search since we have already
11151 * held back allocations.
11153 static int btrfs_trim_free_extents(struct btrfs_device *device, u64 *trimmed)
11155 u64 start = SZ_1M, len = 0, end = 0;
11160 /* Discard not supported = nothing to do. */
11161 if (!blk_queue_discard(bdev_get_queue(device->bdev)))
11164 /* Not writable = nothing to do. */
11165 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
11168 /* No free space = nothing to do. */
11169 if (device->total_bytes <= device->bytes_used)
11175 struct btrfs_fs_info *fs_info = device->fs_info;
11178 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
11182 find_first_clear_extent_bit(&device->alloc_state, start,
11184 CHUNK_TRIMMED | CHUNK_ALLOCATED);
11186 * If find_first_clear_extent_bit find a range that spans the
11187 * end of the device it will set end to -1, in this case it's up
11188 * to the caller to trim the value to the size of the device.
11190 end = min(end, device->total_bytes - 1);
11191 len = end - start + 1;
11193 /* We didn't find any extents */
11195 mutex_unlock(&fs_info->chunk_mutex);
11200 ret = btrfs_issue_discard(device->bdev, start, len,
11203 set_extent_bits(&device->alloc_state, start,
11206 mutex_unlock(&fs_info->chunk_mutex);
11214 if (fatal_signal_pending(current)) {
11215 ret = -ERESTARTSYS;
11226 * Trim the whole filesystem by:
11227 * 1) trimming the free space in each block group
11228 * 2) trimming the unallocated space on each device
11230 * This will also continue trimming even if a block group or device encounters
11231 * an error. The return value will be the last error, or 0 if nothing bad
11234 int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
11236 struct btrfs_block_group_cache *cache = NULL;
11237 struct btrfs_device *device;
11238 struct list_head *devices;
11244 u64 dev_failed = 0;
11249 cache = btrfs_lookup_first_block_group(fs_info, range->start);
11250 for (; cache; cache = next_block_group(cache)) {
11251 if (cache->key.objectid >= (range->start + range->len)) {
11252 btrfs_put_block_group(cache);
11256 start = max(range->start, cache->key.objectid);
11257 end = min(range->start + range->len,
11258 cache->key.objectid + cache->key.offset);
11260 if (end - start >= range->minlen) {
11261 if (!block_group_cache_done(cache)) {
11262 ret = cache_block_group(cache, 0);
11268 ret = wait_block_group_cache_done(cache);
11275 ret = btrfs_trim_block_group(cache,
11281 trimmed += group_trimmed;
11291 btrfs_warn(fs_info,
11292 "failed to trim %llu block group(s), last error %d",
11293 bg_failed, bg_ret);
11294 mutex_lock(&fs_info->fs_devices->device_list_mutex);
11295 devices = &fs_info->fs_devices->devices;
11296 list_for_each_entry(device, devices, dev_list) {
11297 ret = btrfs_trim_free_extents(device, &group_trimmed);
11304 trimmed += group_trimmed;
11306 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
11309 btrfs_warn(fs_info,
11310 "failed to trim %llu device(s), last error %d",
11311 dev_failed, dev_ret);
11312 range->len = trimmed;
11319 * btrfs_{start,end}_write_no_snapshotting() are similar to
11320 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
11321 * data into the page cache through nocow before the subvolume is snapshoted,
11322 * but flush the data into disk after the snapshot creation, or to prevent
11323 * operations while snapshotting is ongoing and that cause the snapshot to be
11324 * inconsistent (writes followed by expanding truncates for example).
11326 void btrfs_end_write_no_snapshotting(struct btrfs_root *root)
11328 percpu_counter_dec(&root->subv_writers->counter);
11329 cond_wake_up(&root->subv_writers->wait);
11332 int btrfs_start_write_no_snapshotting(struct btrfs_root *root)
11334 if (atomic_read(&root->will_be_snapshotted))
11337 percpu_counter_inc(&root->subv_writers->counter);
11339 * Make sure counter is updated before we check for snapshot creation.
11342 if (atomic_read(&root->will_be_snapshotted)) {
11343 btrfs_end_write_no_snapshotting(root);
11349 void btrfs_wait_for_snapshot_creation(struct btrfs_root *root)
11354 ret = btrfs_start_write_no_snapshotting(root);
11357 wait_var_event(&root->will_be_snapshotted,
11358 !atomic_read(&root->will_be_snapshotted));
11362 void btrfs_mark_bg_unused(struct btrfs_block_group_cache *bg)
11364 struct btrfs_fs_info *fs_info = bg->fs_info;
11366 spin_lock(&fs_info->unused_bgs_lock);
11367 if (list_empty(&bg->bg_list)) {
11368 btrfs_get_block_group(bg);
11369 trace_btrfs_add_unused_block_group(bg);
11370 list_add_tail(&bg->bg_list, &fs_info->unused_bgs);
11372 spin_unlock(&fs_info->unused_bgs_lock);